Ecaade2013 Vol 2 Lowres

Education and research in Computer Aided Architectural Design in Europe
eCAADe 2013
Computation and Performance
18-20 September, 2013
Delft University of Technology
Volume 2
Edited by
Rudi Stouffs and Sevil Sariyildiz
eCAADe 2013
Volume 2
Volume 2 - Computation and Performance - eCAADe 31 | 1

Editors
Rudi Stouffs*
Sevil Sariyildiz
Design Informatics, Department of Architectural Engineering + Technology, Faculty of Archi-
tecture, Delft University of Technology, Delft, The Netherlands
(* Department of Architecture, School of Design and Environment, National University of
Singapore, Singapore)
1st Edition, September 2013
Computation and Performance – Proceedings of the 31st International Conference on Educa-

tion and research in Computer Aided Architectural Design in Europe, Delft, The Nether-
lands, 18-20 September 2013, Volume 2. Edited by Rudi Stouffs and Sevil Sariyildiz. Brussels:
Education and research in Computer Aided Architectural Design in Europe; Delft: Faculty of
Architecture, Delft University of Technology.
ISBN 978-94-91207-05-1 (eCAADe)
Copyright © 2013
Publisher: eCAADe (Education and research in Computer Aided Architectural Design in

Europe) and Faculty of Architecture, Delft University of Technology
www.ecaade.org
Cover design: Bige Tunçer and Rudi Stouffs

Cover photograph: Ifigeneia Dilaveraki
All rights reserved. Nothing from this publication may be reproduced, stored in computer-
ised system or published in any form or in any manner, including electronic, mechanical,
reprographic or photographic, without prior written permission from the publisher.
2 | eCAADe 31 - Computation and Performance - Volume 2

eCAADe 2013
Volume 2
Proceedings of the 31st International Conference on Education and research in

Computer Aided Architectural Design in Europe
18-20 September 2013

Delft, The Netherlands
Faculty of Architecture, Delft University of Technology
http://ecaade2013.bk.tudelft.nl/
Edited by
Rudi Stouffs
Sevil Sariyildiz

Theme
This is the second volume of the conference proceedings of the 31st eCAADe confer-
ence, held from 18-20 September 2013 at the Faculty of Architecture of Delft University of
Technology in Delft, the Netherlands. Both volumes together contain 150 papers that were
submitted and accepted to this conference.
The theme of the 31st eCAADe conference is the role of computation in the consideration of
performance in planning and design.
Since long, a building no longer simply serves to shelter human activity from the natural en-
vironment. It must not just defy natural forces, carry its own weight, its occupants and their
possessions, it should also functionally facilitate its occupants’ activities, be aesthetically
pleasing, be economical in building and maintenance costs, provide temperature, humidity,
lighting and acoustical comfort, be sustainable with respect to material, energy and other
resources, and so forth. Considering all these performance aspects in building design is far
from straightforward and their integration into the design process further increases com-
plexity, interdisciplinarity and the need for computational support.
One of the roles of computation in planning and design is the measurement and prediction
of the performances of buildings and cities, where performance denotes the ability of build-
ings and cities to meet various technical and non-technical requirements (physical as well as
psychological) placed upon them by owners, users and society at large.
This second volume contains 75 papers grouped under eleven subthemes that vary from
Simulation, Prediction and Evaluation over Models of Computation: Human Factors to Lan-
guages of Design.

Sponsors of the eCAADe 2013 Conference
Autodesk GmbH
Bentley Systems

Acknowledgements
With the 31st eCAADe conference held in Delft, eCAADe has finally come full circle. The very first
eCAADe conference, before the actual founding of the eCAADe organization in 1983, was held in Delft
in 1982. 31 years later, we are proud to welcome the eCAADe organization back to its origins.
This Delft conference has been a while in the making. The idea was first raised by Martijn Stelling-
werff in 2006 and a preliminary proposal was presented to the eCAADe council at that time. However,
we encountered some turbulent times with the destruction by a fire of the Faculty of Architecture
building in Delft in 2008 and only in 2010 were we ready to present a definitive proposal for the
conference in Delft. From that time until the publication of these proceedings, many people helped to
make this happen and we hope to mention them all here:
First of all, we would like to thank both deans, Wytze Patijn (in 2010) and Karen Laglas (since 2011),
for their endorsement and support, and especially the director of International Affairs at that time, Ag-
nes Wijers, for her immediate support upon approaching her with the idea and for her ample support
in the early planning of the conference event.
The eCAADe council was supportive throughout the entire process and helped with many aspects
of the organisation. Both presidents, Wolfgang Dokonal (up to 2011) and José Duarte (since 2011),
were very supportive. Bob Martens, as liaison with the conference host, was particularly helpful with
many issues in the process. We received especially a lot of support from Henri Achten as previous
conference organiser. Martin Winchester made sure the OpenConf system was running smoothly and
reliably. Nele de Meyere and Maaike Waterschoot reacted promptly when approached with adminis-
trative questions. Financial support was generously provided by the sponsors Autodesk and Bentley
Systems.
The Call for Extended Abstracts yielded 287 submissions. Fortunately, we were able to count on
135 international reviewers in helping us to assess all submissions (see the List of Reviewers section).
Each submission was double-blind reviewed by three reviewers. Following the reviewers’ recom-
mendations, 150 papers were finally accepted for publication and presentation. We congratulate the
authors for their accomplishment. Next to the authors, the reviewers, who volunteered valuable time
and effort, the session chairs, who led the presentations, and the students and other volunteers, who
assisted throughout the conference and its preparations, deserve our sincere thanks and acknowl-
edgements.
As conference chairs, we had the support from the organising committee, including, Kas Ooster-
huis, Joop Paul, Bige Tunçer, Martijn Stellingwerff, Michael Bittermann, Michela Turrin, Paul de Ruiter,
Nimish Biloria and Henriette Bier. Joop Paul deserves a special note for securing Gerard Loozekoot, di-
rector of UN Studio, as keynote speaker. A special thanks goes to Irem Erbas, who, next to Bige Tunçer,
Nimish Biloria and Michela Turrin, assisted in processing part of the proceedings. The secretarial team
of the department of Architectural Engineering + Technology assisted on numerous occasions and
Françoise van Puffelen, in particular, especially assisted in all financial matters. Thijs Welman secured
the website and Martijn Stellingwerff designed the conference website. From the faculty we further-
more want to thank the FMVG (Facility Management and Real Estate) people who helped with the
planning of and preparations for the event.
We are very grateful to have as keynote speakers at the conference Sean Hanna (as prominent

academic in the area of computation and performance), Shrikant Sharma (leader of SMART Solutions
– Buro Happold’s specialist service offering advanced computational solutions to practice) and Gerard
Lozekoot (director and senior architect at leading Dutch architectural firm UN Studio) to provide their
views on computation and performance to the conference.
We wish to provide a special acknowledegment to Yunn Chii Wong, head of the Department of
Architecture at the School of Design and Environment, National University of Singapore, and Chris Ma-
gee, co-director of the SUTD-MIT International Design Centre, for offering their support to Rudi Stouffs
to chair the preparations of this conference from abroad.
Finally, we want to thank Bige Tunçer, partner and colleague, and our families for their support
and patience while we were spending late hours organising, reviewing, editing, and trouble shooting
during the past three years.
eCAADe 2013 Conference Chairs

eCAADe 2013 Organizing Committee

Rudi Stouffs
Sevil Sariyildiz
Kas Oosterhuis
Joop Paul
Bige Tunçer
Martijn Stellingwerff
Michael Bittermann
Michela Turrin
Paul de Ruiter
Nimish Biloria
Henriette Bier

List of Reviewers
Sherif Abdelmohsen, Ain Shams University, Egypt Switzerland

Henri Achten, Czech Technical University in Prague, Czech Tomohiro Fukuda, Osaka University, Japan
Republic Harald Gatermann, Bochum University of Applied Sciences,
Hasim Altan, The University of Sheffield, United Kingdom Germany
Aleksander Asanowicz, Bialystok University of Technology, Evelyn Gavrilou, University of Thessaly, Greece
Poland Sofia Georgakopoulou, Swiss Federal Institute of Technology
Gideon Aschwanden, Swiss Federal Institute of Technology Zurich, Switzerland
Zurich, Singapore David Gerber, University of Southern California, United
Shady Attia, École Polytechnique Fédérale de Lausanne, States
Switzerland Thomas Grasl, SWAP Architekten, Austria
Phil Ayres, The Royal Danish Academy of Fine Arts, Denmark Leman Figen Gül, TOBB University of Economics and Technol-
Günter Barczik, Erfurt University of Applied Sciences/HMGB ogy, Turkey
architects, Germany Ipek Gursel Dino, Middle East Technical University, Turkey
Joo Hwa Bay, University of Western Australia, Australia Jan Halatsch, Swiss Federal Institute of Technology Zurich,
A. Can Baykan, Middle East Technical University, Turkey Switzerland
Jakob Beetz, Eindhoven University of Technology, Nether- Gilles Halin, Centre de Recherche en Architecture et Ingé-
lands nierie, France
José Beirão, Technical University of Lisbon, Portugal Jeremy Ham, Deakin University, Australia
Henriette Bier, Delft University of Technology, Netherlands Małgorzata Hanzl, Lodz University of Technology, Poland
Nimish Biloria, Delft University of Technology, Netherlands Michael Ulrich Hensel, Oslo School of Architecture/OCEAN
Michael S. Bittermann, Delft University of Technology, Design Research Association, Norway
Netherlands Christiane Herr, Xi’an Jiaotong-Liverpool University, China
Stefan Boeykens, KU Leuven, Belgium Pablo C. Herrera, Universidad Peruana de Ciencias Aplicadas,
Vassilis Bourdakis, University of Thessaly, Greece Peru
Gulen Cagdas, Istanbul Technical University, Turkey Urs Hirschberg, Graz University of Technology, Austria
Gabriela Celani, Unicamp, Brazil Mustafa Emre Ilal, Izmir Institute of Technology, Turkey
Tomo Cerovsek, University of Ljubljana, Slovenia Taysheng Jeng, National Cheng Kung University, Taiwan
Ilka Cerpes, University of Ljubljana, Slovenia Iestyn Jowers, The Open University, United Kingdom
André Chaszar, Delft University of Technology/O-Design, Anja Jutraz, University of Ljubljana, Slovenia
Netherlands/USA Matevz Juvancic, University of Ljubljana, Slovenia
Sheng-Fen Chien, National Cheng Kung University, Taiwan Anetta Kepczynska-Walczak, Lodz University of Technology,
Benny Chow, Chinese University of Hong Kong, China Poland
Birgul Colakoglu, Yildiz Technical University, Turkey Sora Key, Carnegie Mellon University, United States
Claudia Czerkauer-Yamu, Vienna University of Technology, Joachim Kieferle, RheinMain University of Applied Sciences,
Austria Germany
Ruth Dalton, Northumbria University, United Kingdom Axel Kilian, Princeton University, United States
Bharat Dave, University of Melbourne, Australia Arto Kiviniemi, University of Salford, United Kingdom
Bauke de Vries, Eindhoven University of Technology, Neth- Terry Knight, Massachusetts Institute of Technology, United
erlands States
Halime Demirkan, Bilkent University, Turkey Michael Knight, University of Liverpool, United Kingdom
Wolfgang Dokonal, Graz University of Technology, Austria Tuba Kocaturk, University of Liverpool, United Kingdom
Dirk Donath, Bauhaus-Universität Weimar, Germany Volker Koch, Karlsruhe Institute of Technology, Germany
Tomás Dorta, Université de Montreal, Canada Reinhard Koenig, Swiss Federal Institute of Technology
Theodoros Dounas, Xi’an Jiaotong-Liverpool University, Zurich, Switzerland
China Branko Kolarevic, University of Calgary, Canada
José Duarte, Technical University of Lisbon, Portugal Anastasia Koltsova, Swiss Federal Institute of Technology
Antonio Fioravanti, Sapienza University of Rome, Italy Zurich, Switzerland
Pia Fricker, Swiss Federal Institute of Technology Zurich, Odysseas Kontovourkis, University of Cyprus, Cyprus

Jose Kos, Federal University of Santa Catarina, Brazil Dagmar Reinhardt, The University of Sydney, Australia
Krzysztof Koszewski, Warsaw University of Technology, Gernot Riether, Georgia Institute of Technology, United
Poland States
Alexander Koutamanis, Delft University of Technology, Luis Romao, Technical University of Lisbon, Portugal
Netherlands Peter Russell, RWTH Aachen University, Germany
Stefan Krakhofer, Krakhofer Design, Austria/China Gerhard Schmitt, Swiss Federal Institute of Technology
Sylvain Kubicki, Public Research Centre Henri Tudor, Luxem- Zurich, Switzerland
bourg Marc Aurel Schnabel, Chinese University of Hong Kong,
Antje Kunze, Swiss Federal Institute of Technology Zurich, China
Switzerland Sven Schneider, Bauhaus-Universität Weimar, Germany
Ih-Cheng Lai, Tamkang University, Taiwan A. Benjamin Spaeth, Xi’an Jiaotong-Liverpool University,
Andrew Li, Athlone Research, Japan China
Zhengwei Li, City University of Hong Kong, China Milena Stavric, Graz University of Technology, Austria
Katherine Liapi, University of Patras, Greece Martijn Stellingwerff, Delft University of Technology, Neth-
Thorsten Loemker, Zayed University, United Arab Emirates erlands
Werner Lonsing, University of Stuttgart, Germany Rudi Stouffs, National University of Singapore/Delft Univer-
Carlos L. Marcos, University of Alicante, Spain sity of Technology, Singapore/Netherlands
Earl Mark, University of Virginia, United States Georg Suter, Vienna University of Technology, Austria
Bob Martens, Vienna University of Technology, Austria Oliver Tessmann, KTH Royal Institute of Technology, Sweden
Tom Maver, Glasgow School of Art, United Kingdom Emine Mine Thompson, Northumbria University, United
Benachir Medjdoub, Nottingham Trent University, United Kingdom
Kingdom Christian Tonn, Bauhaus-Universität Weimar, Germany
Achim Menges, University of Stuttgart, Germany Bige Tunçer, Singapore University of Technology and Design/
AnnaLisa Meyboom, University of British Columbia, Canada Delft University of Technology, Singapore/Netherlands
Volker Mueller, Bentley Systems, Inc., United States Emrah Turkyilmaz, Istanbul Kültür University, Turkey
Michael Mullins, Aalborg University, Denmark Michela Turrin, Delft University of Technology/Yasar Univer-
Marc Muylle, University of Antwerp, Belgium sity, Netherlands/Turkey
Herman Neuckermans, KU Leuven, Belgium Aant van der Zee, Eindhoven University of Technology,
Yeonjoo Oh, Samsung Construction, Korea, Republic Of Netherlands
Konstantinos-Alketas Oungrinis, Technical University of Jos van Leeuwen, The Hague University of Applied Sciences,
Crete, Greece Netherlands
Rivka Oxman, Technion Israel Institute of Technology, Israel Johan Verbeke, LUCA School of Arts, Belgium
Mine Ozkar, Istanbul Technical University, Turkey Spela Verovsek, University of Ljubljana, Slovenia
Burak Pak, KU Leuven, Belgium Petra von Both, Karlsruhe Institute of Technology, Germany
Sule Tasli Pektas, Bilkent University, Turkey Maria Voyatzaki, Aristotle University of Thessaloniki, Greece
Giuseppe Pellitteri, University of Palermo, Italy Stephen Wittkopf, Lucerne University of Applied Sciences and
Chengzhi Peng, University of Sheffield, United Kingdom Arts, Switzerland
Jelena Petric, University of Strathclyde, United Kingdom Jerzy Wojtowicz, Warsaw University of Technology, Poland
Frank Petzold, Technische Universität München, Germany Stefan Wrona, Warsaw University of Technology, Poland
Dr. Ra’Ed QaQish, American University of Madaba, Jordan Gabriel Wurzer, Vienna University of Technology, Austria
Ahmad Rafi, Multimedia University, Malaysia Tadeja Zupančič, University of Ljubljana, Slovenia
Rabee M. Reffat, Assiut University, Egypt

Keynote Speakers
Sean Hanna
Sean Hanna is a Lecturer in Space and Adaptive Architectures at University College London, Director of the
MSc/MRes programs in Adaptive Architecture and Computation at the Bartlett School of Graduate Studies,
and Academic Director of UCL’s Doctoral Training Center in Virtual Environments, Imaging and Visualisation.
He is a member of the Space Group, noted as one of the UK’s highest performing research groups in the field
of architecture and the built environment.
Originally from a background of architectural practice, his application of design algorithms includes
major projects with architects Foster + Partners and sculptor Antony Gormley. His research is primarily in
developing computational methods for dealing with complexity in the built environment, including the
modelling of space and its perception, and he is on the advisory boards of two related UCL spin-out compa-
nies. His publications address the fields of spatial modelling, machine intelligence, collaborative creativity,
among others, and his work has been featured in the non-academic press, including the Architects’ Journal
and The Economist.
Shrikant Sharma
Shrikant Sharma leads SMART Solutions – Buro Happold’s specialist service that offers advanced computa-
tional solutions to support architectural design, engineering, construction and operations of buildings and
urban spaces. The team, founded by Shrikant in 2002, is renowned for delivering simple, innovative solutions
for complex engineering problems in the built environment.
Shrikant has a PhD in Engineering and over 15 years of experience in the development and application
of novel modelling and analysis techniques. A firm believer in the power of rapid design optioneering tools
that integrate architectural, functional, engineering, and environmental assessments of buildings and urban
spaces, Shrikant has been driving the development of a suite of intuitive real time software tools that work
within commercial CAD and BIM environments. He has also led the application of such technologies for
integrated modelling and optimisation on a number of projects such as Scunthorpe Sports Academy, Louvre
Abu Dhabi, Sidra Trees Qatar Convention Centre, and London City Airport.
SMART Space - Buro Happold’s crowd flow modelling and consultancy service is run by Shrikant. It uses
novel analytical and simulation techniques to help the architects, planners, developers, and regulators to
understand and optimise space layout, design and management.
Shrikant is actively engaged in the advancements in computation design and simulation through rigor-
ous ongoing research and development, and has developed innovative software tools such as SMART Form
and SMART Move.
Gerard Loozekoot
Gerard Loozekoot is Director and Senior Architect at UNStudio. He earned his Master’s degree in Architecture
from Delft University of Technology, worked as an architect at UNStudio since 2000 and became partner at
UNStudio in 2008. His great interest are innovative typological innovations, such as the projects Theater in
Lelystad, the UNStudio office tower in Amsterdam or the airport in Georgia. In addition, sustainable innova-
tions are one of the main pillars of his projects. In Dienst Uitvoering Onderwijs in Groningen and Le Toison
d’Or in Brussels, Gerard demonstrates that the added value of sustainable buildings have become the new
standard. As director and senior architect is Gerard actively involved in all phases of the construction.

Contents
5 Theme
7 Acknowledgements
9 List of Reviewers
11 Keynote Speakers
13 Contents
19 Simulation, Prediction and Evaluation

21 Using Smart Controlled AC and Ceiling Fan to Save Energy
Pin-Hung Wang, Jui-Yen Lin
29 Even ‘Clouds’ Can Burn
Antonio Fioravanti, Eolo Avincola, Gabriele Novembri
39 Inductive Aerodynamics
Samuel Wilkinson, Sean Hanna, Lars Hesselgren, Volker Mueller
49 Adaptive Fluid Lens and Sunlight Redirection System
Florian Heinzelmann, Telesilla Bristogianni, Patrick Teuffel
59 Modelling and Simulating Use Processes in Buildings
Davide Simeone, Yehuda E. Kalay, Davide Schaumann, Seung Wan Hong
69 Flexing Wind
Rafael Moya, Flora Salim, Mani Williams, Kamil Sharaidin
79 Real-time Environmental Feedback at the Early Design Stages
Greig Paterson, Sung Min Hong, Dejan Mumovic, Judit Kimpian
87 DesignScript: Scalable Tools for Design Computation
Robert Aish
97 Performative Design
99 Architectural Thermal Forms
Isak Worre Foged
107 DaylightGen: From Daylight Intentions to Architectural Solutions
Mohamed-Anis Gallas, Gilles Halin
117 Performance Driven Design and Design Information Exchange
Sina Mostafavi, Mauricio Morales Beltran, Nimish Biloria
127 Performance Based Pavilion Design
Sevil Yazici
Contents - Volume 1 - Computation and Performance - eCAADe 31 | 13

137 Engineering Performance Simulations in Architectural Design Conception
Michela Turrin, Ioannis Chatzikonstantinou, Martin Tenpierik, Sevil Sariyildiz
147 Passive Energy Devices in Ceramics
Mark Weston, Dan Greenberg
153 Generation, Exploration and Optimisation

155 Automated Simulation and Study of Spatial-Structural Design Processes
Juan Manuel Davila Delgado, Herm Hofmeyer
165 Generative Agent-Based Design Computation
Ehsan Baharlou, Achim Menges
175 Evolutionary Energy Performance Feedback for Design (EEPFD)
Shih-Hsin Eve Lin, David Gerber
185 Cloud-Based Design Analysis and Optimization Framework
Volker Mueller, Tiemen Strobbe
195 Graphical Smalltalk with My Optimization System for Urban Planning Tasks
Reinhard Koenig, Lukas Treyer, Gerhard Schmitt
205 Evo-Devo in the Sky
Patrick Janssen
215 Algorithmic Design Generation

217 The Potential of Evolutionary Methods in Architectural Design
Wassim Jabi, Barbara Grochal, Adam Richardson
227 Genetic Algorithms Applied to Urban Growth Optimization
Patricia Camporeale
237 Design Tools for Integrative Planning
Stefana Parascho, Marco Baur, Jan Knippers, Achim Menges
247 Infections
Tuğrul Yazar, Fulya Akipek
255 Algorithmic Engineering in Public Space
Jaroslav Hulin, Jiri Pavlicek
261 Integrating Computational and Building Performance Simulation Techniques
for Optimized Facade Designs
Mahmoud Gadelhak
14 | eCAADe 31 - Computation and Performance - Volume 1 - Contents

271 Models of Computation: Form Studies
273 Algorithmic Form Generation for Crochet Technique
J. Gozde Kucukoglu, Birgul Colakoglu
279 3D Regular Expressions - Searching Shapes IN Meshes
Gabriel Wurzer, Bob Martens, Katja Bühler
289 A Computational Method for Integrating Parametric Origami Design and
Acoustic Engineering
Tsukasa Takenaka, Aya Okabe
297 A Novel Method for Revolved Surface Infrastructures
Gökhan Kınayoğlu
305 Ruling Im/Material Uncertainties
Zeynep Akküçük, Mine Özkar
315 Hyperdomes
Andrea Rolando, Domenico D’Uva
325 Action Based Approach to Archaeological Reconstruction Projects: Case of the
Karnak Temple in Egypt
Anis Semlali, Temy Tidafi, Claude Parisel
333 Models of Computation: Human Factors

335 Fusion of Perceptions in Architectural Design
Ozer Ciftcioglu, Michael S. Bittermann
345 Ambient Surveillance by Probabilistic-Possibilistic Perception
Michael S. Bittermann, Ozer Ciftcioglu
355 The Jacobs´ Urban Lineage Revisited
Claudio Araneda
365 Collaborative and Human Based Performance Analysis
Mathew Schwartz
375 Visibility Analysis for 3D Urban Environments
Anastasia Koltsova, Bige Tunçer, Gerhard Schmitt
385 Human Activity Modelling Performed by Means of Use Process Ontologies
Armando Trento, Antonio Fioravanti

395 3D Model Performance
397 New Methods for the Rapid Prototyping of Architectural Models
Stefan Junk, Samantha Côté
405 Four Chairs and All the Others - Eigenchair
Miro Roman
415 Digital Design Tools versus Architectural Representation and Design
Approach
Betül Orbey, Nihan Gürel
425 Considering Physicality in Digital Models
Andrzej Zarzycki
435 Understanding and Managing the Constructive Characteristics of Vernacular
Architecture
Danilo Di Mascio
445 3D Digitization in Architecture Curriculum
Luís Mateus, Victor Ferreira
453 Performing the Past and the Present for the Knowledge of the Future
Anetta Kepczynska-Walczak
463 Building Information Modelling

465 Challenges of Integrating BIM in Architectural Education
Tuba Kocaturk, Arto Kiviniemi
475 Attaining Performance with Building Information Modelling
Eleni Papadonikolaki, Alexander Koutamanis, J. W. F. (Hans) Wamelink
485 Building Your Own Urban Tool Kit
Caner Dolas, Andreas Dieckmann, Peter Russell
495 [Architectural] Reasoning over BIM/CAD Database
Gianluigi Loffreda, Antonio Fioravanti, Luigi Avantaggiato
505 Experiencing BIM Collaboration in Education
Stefan Boeykens, Pauline De Somer, Ralf Klein, Rik Saey
515 Landscape Information Modeling
Veronika Zajíčková, Henri Achten
525 ifcModelCheck
Sebastian Ebertshäuser , Petra von Both
535 Daylight Performance Simulations and 3D Modeling in BIM and non-BIM Tools
Marina Stavrakantonaki

543 CAAD Curriculum
545 How to Teach ‘New Tools’ in Landscape Architecture in the Digital Overload
Pia Fricker, Christophe Girot, Georg Munkel
555 150 000 – Parametric Control of PET Bottle Structure
Kateřina Nováková, Lukáš Kurilla, Henri Achten
563 Identifying Cognitive Operations of Conception Implied in the Uses of
Parametric Modeling in Architectural Design: Toward Pedagogical Tools
Aurélie de Boissieu, François Guéna, Caroline Lecourtois
571 Continuous Oscillations
Günter Barczik
579 Computation/Performance
Anna Pla-Catala
587 Dances with Architects
Angelika Lückert, Volker Koch, Petra von Both
595 A Case Study in Teaching Construction of Building Design Spaces
Mahsa Nicknam, Marcelo Bernal, John Haymaker
605 Innovative Learning for Collaborative Design in Ergonomics
Viviane Folcher, Khaldoun Zreik, Samia Ben Rajeb, Pierre Leclercq
615 Shape Grammars

617 Unambiguity
Thomas Grasl, Athanassios Economou
621 Customized Cork Façade
Rui Marques, Sara Eloy
627 A Generative Approach towards Performance-Based Design
Tiemen Strobbe, Ronald De Meyer, Jan Van Campenhout
635 Tableware Shape Grammar
Eduardo Castro e Costa, José Pinto Duarte
645 Grandstand Grammar and its Computer Implementation
Yimin Sun, Lu Xiong, Ping Su
655 From Point Cloud to Shape Grammar to Grammatical Transformations
Filipe Coutinho, Luis Mateus, José P. Duarte, Victor Ferreira, Mário Kruger

665 Languages of Design
667 Combining Complexity and Harmony by the Box-Counting Method
Wolfgang E. Lorenz
677 The Rehabilitation Design Process of the Bourgeois House of Oporto: Shape
Grammar Simplification
Eugénio Coimbra, Luís Romão
687 Albertian Grammatical Transformations
Bruno Figueiredo, José Pinto Duarte, Mário Krüger
697 A Parametric Recreation of Traditional Chinese Architecture
Di Li, Michael Knight, Andre Brown
705 A Bottom-Up Social Housing System Described with Shape Grammars
Leticia Teixeira Mendes, José Nuno Beirão, José Pinto Duarte, Gabriela Celani
715 The Language of Mozambican Slums
Pedro Barros, José Beirão, José Pinto Duarte
725 Gulou Structure Grammar and its Computer Implementation
Lu Xiong, Wei Xiong, Hongxia Zhang
735 Index of Authors

Simulation, Prediction and
Evaluation
Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 19

20 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation
Using Smart Controlled AC and Ceiling Fan to Save Energy
Pin-Hung Wang1, Jui-Yen Lin2
Department of Architecture, Kao Yuan University, Kaohsiung, Taiwan
1
t60038@cc.kyu.edu.tw, 2xryan19891207@gmail.com
Abstract. This research aimed to explore the energy savings through the use of smart
control as well as ceiling fan in intelligent building. As the energy consumption of
air-conditioning (AC) accounts for about 40% of total residential energy, therefore,
applying smart control system to the use of AC to achieve the effects of comfy and energy
savings should be able to generate positive effect for the energy consumption of overall
residential. This study used the smart control system in the intelligent building lab to
transmit message to AC for its implementation of next operating step through the indoor
temperature sensor in order to achieve energy saving effect.
Keywords. Intelligent building; smart control; energy saving; ZigBee; smart living.
INTRODUCTION
This study focused on the exploration of the smart since 2006, using buildings as medium to integrate
controlled AC (Air Conditioner) and ceiling fan, using ICT and other related communication products to
the lab of intelligent building, in order to achieve the merge innovation and design application for the
objective of energy saving. Although each country construction of new living environment, there are
has different definition about intelligent building, all still few of successful intelligent buildings over
of their basic objectives are about the same. Intel- these years. The main reason is not because of the
ligent building combines structure, system, service technology problem, but of intelligent building
and operation management to create the most opti- requiring the cooperation of many different fields
mal combination and process for the construction of under cross-platforms. Without proper guidance, ar-
highly efficient, excellent function and comfortable chitects are difficult to carry out plans and designs.
buildings. Therefore, intelligent building must be In view of this, our study tried to use established
able to satisfy users’ needs, control easily, save en- intelligent building lab to conduct smart control of
ergy, improve management effectiveness and clarify energy saving on available AC and ceiling fan in the
information. space so as to explore the future development and
This study focused on the role of energy saving direction for intelligent building by means of energy
in intelligent building. Taiwan has started promot- saving efficiency.
ing the intelligent building mark since 2004; how-
ever, over the eight years, there were only ten cases METHOD AND DEVELOPMENT OF INTEL-
certificated, which is obviously lower compared LIGENT BUILDING’S ENERGY SAVING
to 359 green mark buildings in the past. Moreover, An overview of building spaces utilization shows
although intelligent building has already became that, the proportion of electricity used by AC ac-
a government policy and Taiwan’s Executive Yuan counts for about 40% of overall energy consump-
has also started promoting intelligent building tion, while lights and electric outlets takes up about

Figure 1
Framework of smart home
in lab.
40% (Taiwan Power, 2006). This study applied smart cadian rhythms which meet human body’s different
control to AC and ceiling fan as a main planning di- biological needs so that to enhance living safety and
rection to compare its energy saving effectiveness comfort (Fu et.al., 2010); setting up multiple ZigBee
to traditional model as a reference for future intel- moisture sensor around the indoor space to adjust
ligent building design. the AC operation, improve living condition and re-
For smart controlled AC and home appliance, duce energy consumption through temperature
they can be automatically adjusted by different ap- and moisture data collection (Wang et. al. , 2010);
proaches. ,For example, conducting smart control , using the position method of BBM (Best Beacon
on living environment through EEG (Electroen- Match) to smartly control living environment (Jin et.
cephalography) (Lin et. al., 2010); using BCI(Brain al., 2007), such system can control AC and lighting;
Computer Interface) as biological and electric using Smart phone as interface to monitor and con-
monitoring system to achieve the goal of active en- trol the living condition (Zhong et. al., 2011) (Li et.
vironment control; applying CPSs (Cyber Physical al., 2012) to replace remote control; using wireless of
Systems)such as Bluetooth, ZigBee RF and infrared sensor networks to establish a physical environment
ray to carry out various communication protocols so for room control to adjust the use of electrical ap-
as to convert a variety of different signals through a pliances automatically for the energy-saving effect
smart control box (Bai, 2012) using pyroelectric in- through the data monitoring (Yeh, 2009).
frared sensor-based indoor location aware system
(PILAS) as receivers (Kastner et. al., 2010) to monitor RESEARCH METHODS
residents’ activities, position, pattern, or health con- This study used ZigBee as the main transmitter to
dition to provide the best living environment; han- construct an environment and interface which are
dling complicated intelligent home equipment by able to meet the requirements of smart control (Fig-
the construction of low cost sensor and control sys- ure 1), in which the temperature and moisture in
tems based on ZigBee (Blesa et. al., 2009); using low the room were detected by sensor, and such data
budget stationary sensor to set up electronic nose were then transferred to AC through ZigBee’s sig-
for air quality monitoring (Zampolli et. al., 2004) to nals which will further advise AC about the next step
control the AC system for the best air flow; provid- to take through the intelligent control way so as to
ing different colors of light by photovoltaic lighting achieve the objective of energy-saving
systems developed in accordance with human cir- Most homes’ AC temperature control uses re-

Figure 2 Although our lab controller only has 12 keys,
AC controlle key functions. the study’s smart control model can simulate 32
controller functions, through different setups ways.
For example, despite that there were only two tem-
perature control keys (up and down), the tempera-
ture can be set from 17 °C to 31 °C; function key can
control four modes of ATC, AC, dehumidification and
fan; Fuzzy key has options of too hot, comfy, and too
cold; fan key can set speed at high, mid, low, and
auto; timing key can set time from 0.5 to12 hours;
while most keys only have one function, all of them
have on and off options. In sum, there are 48 differ-
ent options in total (Figure 2) which are more than
turned air flow temperature as a basis to judge the the 32 control functions set by this study. Owing
next operating step of AC. Ceiling fan is a good to each brand’s AC controller is different, how to
way to condition the indoor air flow, and the room choose the best control option in order to achieve
temperature can be rapidly and effectively reduced energy saving and comfort will be the central topic
through the combination of these two appliances. of this research.
However, f to adjust its frequency reduction or air This research focused on room comfort as the
conditioning running adjustment, the AC requires main objective and, coupling with the energy saving
the room temperature reaches the pre-set value objective, explored how to set up the best operation
prior to starting its next operation. As the purpose mode among the 32 simulated functions for energy
of AC is not for cooling the entire room but for us- saving and room comfort.
ers only, therefore, this study used portable sensors Most home AC (such as variable-frequency AC)
near the users which can send back detected tem- will start to lower the compressor turn rate to save
perature data around the users back to AC every energy and maintain room temperature when the
thirty seconds for AC’s next operating preparation self-perception function detects the temperature of
(such as reducing frequency, changing to air flow or return air reaching the set temperature, and start to
reducing wind flow) without waiting for the reach of run the compressor again to low the temperature
pre-set room temperature. down once the room temperature raised to a cer-
As the moisture sensor is portable and can retain degree. This study used fixed-frequency AC with
spond the temperature near the users to AC im- compressor can only be operated with on or of func-
mediately, it is with real time feedback feature and tion but cannot be operated as efficient and energy-
energy saving effect compared to the perception saving as variable AC compressor in terms of main-
mode of traditional AC. tain the indoor temperature through the change of
Most Home ACs are divided into fixed- and var- frequency.
iable-frequency of window and wall mounted mod- In this study, when the indoor temperature in
els. This study used fixed-frequency, wall mounted the intelligent building reaches a preset limit, the
AC which has been used for five years and can be controller will place to AC an order to implement
controlled through the remote control in terms next operation. Instead of by AC itself, it used the
of power switch, temperature, function, timer, fan external sensor to detect the current room tem-
speed, rhythm, sleeper, particularly the Fuzzy modes perature and then issue the next operating order.
which uses the body perception condition as feed- Owing to that the study could only use the function
back data for AC’s operating reference. key on remote controller key as simulation object

Figure 3
The measured indoor
temperatures and energy
consumptions of AC mode
under the condition of 26 °C
of temperature as well as the
shutdown of ceiling fan on
May 16.
Figure 4
The measured indoor
temperatures and energy con-
sumptions of AC automatic
mode under the condition the
shutdown of ceiling fan on
May 17.
to select the most suitable mode for the next step temperatures of spring season during the experi-
order placing, therefore, this research attempted to mental period which gradually became warming, in
explore how to use the current transmitter to let the order to obtain a more objective analysis of the data,
AC knows that the temperature has been reached as this study set the value at room temperature of 26 °C
well as what’s next step it should operate in order to and implemented four days’ records and observed
achieved the objectives of energy saving and comfy. the temperatures and energy saving effects on the
above-mentioned two air conditioning modes, then
EXPERIMENTAL PROCEDURE AND RE- explored the indoor ceiling fan impact on AC ef-
SULTS fect on the basis of data test. The gateway control-
As the AC used in this lab. was a fixed-frequency ler used in our laboratory was able to record indoor
split air-conditioning being used for five years, in temperature and total electricity consumption every
order to confirm whether the use of intelligent con- minute. The findings showed that, under the con-
trol of energy efficiency is achieved as expected, dition of indoor fan shut down as well at the tem-
this study first carried out the multi-day tests and perature of 26 °C, the room temperature was only
records for the AC mode controlled by remote con- reduced to 28.5 °C and 29.2 °C, respectively (Figures
troller and automatic mode. Owing to the unstable 3 and 4). Moreover, despite the air conditioner au-

Figure 5
The measured indoor
temperatures and energy
consumptions of AC mode
combining with ceiling fan on
May 14.
Figure 6
The measured indoor
sumptions of AC automatic
mode combining with ceiling
fan on May 14.
tomatic mode reached the energy consumption of rectly, resulting in the AC failed to carry out proper
1.995kw in 2 hours, the quality of the indoor tem- cooling effect in accordance with the actual indoor
perature cannot achieve the desired comfort. temperature value.
After opening the indoor ceiling fans to help While coupling with the ceiling fan under the
improve indoor air circulation under the same AC same set, value of the room temperature improved
and automatic AC mode at temperature of 26 °C, the significantly and the indoor temperature was fairly
room temperature value showed a significant im- satisfied -though not reaching but was closing to
provement trend (Figures 5, 6). the set temperature of 26 °C.
Test results of these few days revealed that the Based on the AC automatic mode and its achiev-
split air conditioner in our study neither could reach able target temperature values, this study set differ-
desired temperature within two hours nor reduce ent smart modes for the controller to test and record
the indoor temperature to a reasonable value, de- the AC’s energy consumption situation.
spite that indoor temperature showed a dropping • The temperature was set at 26 °C and the gate-
trend. This phenomenon showed that the air tem- way controller would transmit message to AC
perature sensor was located in a place which was to change the mode to fan mode when the
not able to detect the room temperature value cor- room temperature reached 26 °C, or restarted

Figure 7
The measured indoor
sumptions of smart control
mode on April 25.
Figure 8
Demanding the AC to cool
down rapidly by setting
temperature at 24 °C and
ordering AC to change to fan
mode when the temperature
reached the set temperature,
and to restart the AC com-
pressor when temperature
restored to 27 °C.
the AC compressor when the indoor tem- which led both operating modes of two hours
perature rose to 27 °C, so the cycle execution. test to exert the same effectiveness (Figure 7).
The initial findings showed that, although the • The temperature of AC was set at 24 °C. When
temperature was set at 26 °C, the indoor tem- room temperature reached 24 °C, the AC would
perature was unable to cool down to 26 °C. Two be ordered to change into fan mode, and
possibilities are judged for such condition: A. would restart when indoor temperature rose
Sensor mounted on the AC was too close to the to 27 °C. The results of measured data showed
outlet which led too high return cool air rate, that, indoor temperature dropped from origi-
thus causing the inconsistency between the nal 33.5 °C to 29.5 °C within ten minutes, but
temperature detected by the sensor and actual then gradually cooled to 27.5 °C in the rest
indoor temperature and resulting in automatic 110 minutes (Figure 8), failing to reach the
control mode failed to perform its function. expected set temperature. The observations
B. The variation of detected temperature be- made here for such condition were that, in or-
tween sensor of this study and sensor mounted der to avoid excess use of AC, both ACs used
on the AC. After examination, it was found that by classrooms and lab. were adjusted and set.
there was 1°C difference between both sensors Therefore, no matter how the air temperatures

were adjusted, the compressor would maintain operating modes as well as statistical analysis to ex-
a certain operation mode which was difficult to plore the relationship between the temperature of
change. AC operation and electricity consumption, in order
Experimental data showed that, as the intelli- to identify the most energy-efficient mode of opera-
gent control mode could only be operated by cool tion as a reference for future smart control studies.
and fan modes which limited its compressor opera- The AC currently used by school have been ad-
tion in terms of cooling and electricity consumption justed which are not suitable for the test of energy
adjustment, its electricity consumption was almost consumption through temperature control of smart
the same as that of automatic AC. On the other mode. However, as now the school has approved
hand, due to the operation mode of AC used by to have the lab AC restored to their original setting
the school was adjusted before installation for the condition, the related experimental modes of this
purpose of energy saving, it was incapable to reach study will be continuously conducted in order to
expected temperature through the temperature ad- obtain more objective experimental data for the ref-
justment. erences of indoor temperature quality control and
Furthermore, the data also indicated that, with- adjustment.
out the ceiling fan to adjust air flow, indoor tem-
perature was not only difficult to low down, but with REFERENCES
significant fluctuation as well. Therefore, it is obvious Taiwan power company, Taipower library, Building
that ceiling fan does help the indoor temperature to saves energy 2006, http://www.taipower.com.tw/
cool down. left_bar/45453err/2AA8An/building_saves_energy.
htm,Update: Dec. 6, 2006
CONCLUSION Lin, Chin-Teng; Lin, Fu-Chang; Chen, Shi-An; Lu, Shao-Wei;
The lab was constructed in winter time December, Chen, Te-Chi; Ko, Li-Wei 2010, ‘EEG-based brain-com-
2012, but did not start carrying out the experiment puter interface for smart living environmental auto-ad-
until the mid-April of next year when the weather justment’, Journal of Medical and Biological Engineering,
was getting warmer. During the experimental pe- v 30, n 4, pp. 237-245
riod, this study discovered that when the initial Bai, Zhi-Yong; Huang, Xin-Yuan 2012, ‘Design and imple-
indoor temperature became higher, the electric- mentation of a cyber physical system for building
ity consumption was also affected. However, the smart living spaces’, International Journal of Distributed
preliminary findings showed that the current op- Sensor Networks, v 2012, pp.1-9
erating mode of fixed-frequency AC was difficult Kim, Hyun-Hee; Lee, Suk; Lee, Kyung-Chang 2010, ‘Estima-
to achieve the objective of energy saving through tion of metabolic rate estimation for location-based
control method of smart system. Therefore, it is rec- human adaptive air-conditioner in smart home’, Jour-
ommended that the future study should focus the nal of Institute of Control, Robotics and Systems, v 16, n
application of smart control on AC with variable fre- 1, p 83-89, January, pp.182-186
quency in order to find out possible smart control Kastner, Wolfgang; Kofler, Mario J.; Reinisch, Christian 2010,
mode to meet the demand for energy saving. ‘Using AI to realize energy efficient yet comfortable
Owing to the shorter experimental period, the smart homes’, IEEE International Workshop on Factory
accumulated data of initial room temperature val- Communication Systems - Proceedings, WFCS, pp. 169-
ues in this study were rather limited and cannot be 172
regarded as objective data to present the relation- Blesa, Javier; Malagón, Pedro; Araujo, Álvaro; Moya, José
ship of electricity consumption effectiveness be- M.; Vallejo, Juan Carlos; De Goyeneche, Juan-Mariano;
tween the indoor temperature and AC operation. Romero, Elena; Villanueva, Daniel; Nieto-Taladriz, Oc-
This research will continue to apply different AC tavio 2009, ‘Modular framework for smart home ap-

plications’, Lecture Notes in Computer Science, v 5518 on Advanced Information Networking and Applications
LNCS, n PART 2, pp. 695-701 Workshops, WAINA 2010, pp. 1005-1010
Zampolli, S.; Elmi, I.; Ahmed, F.; Passini, M.; Cardinali, G.C.; Jin, Ming-Hui; Yu, Chih-Hao; Lai, Hung-Ren; Feng, Ming-
Nicoletti, S.; Dori, L. 2004, ‘An electronic nose based on Whei 2007, ‘ZigBee positioning system for smart home
solid state sensor arrays for low-cost indoor air qual- application’, Lecture Notes in Computer Science (includ-
ity monitoring applications’, Sensors and Actuators, B: ing subseries Lecture Notes in Artificial Intelligence
Chemical, v 101, n 1-2, pp. 39-46 and Lecture Notes in Bioinformatics), v 4743 LNCS, pp.
Sun, Bing; Li, Ke; Weng, Fei; Liu, Yuncai 2010, ‘Human loca- 183-192
tion and recognition for intelligent air conditioners’, Zhong, Yu; Suo, Yue; Xu, Wenchang; Yu, Chun; Guo, Xinwei;
Proceedings of SPIE - The International Society for Optical Zhao, Yuhang; Shi, Yuanchun 2011, ‘Smart home on
Engineering, v 7820 smart phone’, UbiComp’11 - Proceedings of the 2011
Fu, Shijiang; He, Yeqing; Zha, Daxin; Fang, Ye 2010, ‘Intel- ACM Conference on Ubiquitous Computing, pp. 467-468
ligent three-dimensional health heating system’, 2010 Li, Tongtong; Ren, Jian; Tang, Xiaochen 2012, ‘Secure wire-
IEEE 11th International Conference on Computer-Aided less monitoring and control systems for smart grid and
Industrial Design and Conceptual Design, CAID and smart home’, IEEE Wireless Communications, v 19, n 3,
CD’2010, v 2, pp. 1532-1536 pp. 66-73
Wang, Tzu-Ming; Lin, Po-Chun; Chan, Hung-Lin; Liao, Jen- Yeh, Lun-Wu; Wang, You-Chiun; Tseng, Tu-Chee 2009, ‘iP-
Chi; Sun, Tian-Wen; Wu, Tin-Yu 2010, ‘Energy saving of ower: An Energy Conservation System for Intelligent
air condition using fuzzy control system over ZigBee Buildings by Wireless Sensor Networks’, International
temperature sensor’, 24th IEEE International Conference Journal of Sensor Networks, v. 5, No. 1, pp. 1-10

Even ‘Clouds’ Can Burn
Fire engineering simulation for a safe, innovative and high-performance

architectural design - a case study
Antonio Fioravanti1, Eolo Avincola2, Gabriele Novembri3

Dept. DICEA - Sapienza University of Rome, Italy
1,3
http://www.dicea.uniroma1.it
1
antonio.fioravanti@uniroma1.it, 2eoloavincola@gmail.com, 3gabriele.novembri@
uniroma1.it
Abstract. Architecture, nowadays, is an even more demanding activity in which

complexity is the keyword: complex forms, complex functions and complex structures
require sophisticated facilities and components, for example, ‘The Cloud’ of D. and M.
Fuksas in Rome. These complexities can give rise to numerous risks, among which fire is
frequently a central problem.
The fire safety norms do not involve an approach integrated with other instruments or
building model (BIM), but provide a list of information and constraints. These codes are
now shifting away from a prescriptive-based towards a performance-based method due to
recent progress in fire safety engineering.
Following this approach, a case study simulation of a multi-purpose centre was carried
out in Tivoli, near Rome. This simulation allowed greater freedom in architectural
composition, a lower risk to people, a larger number of material and building components
used and higher safety standards to be achieved. The model is based on the FDS (Fire
Dynamics Simulator) language, a simulation code for low-speed flows, focused on smoke,
particle and heat transport by fire.
Keywords. Architectural design; computational fluid-dynamics; fire propagation; fire
safety; smoke propagation.
FIRE AND ARCHITECTURAL DESIGN

Architecture has always been a demanding activ- Existing fire safety regulations and codes (Bala-
ity, but in present times, has had to face intertwined ban et al., 2012) very often actually represent as
problems in which complexity is the keyword: com- many limitations imposed on architectural needs
plex forms, complex functions and complex struc- such as space layout, free form space, distribution
tures require sophisticated facilities and compo- path, space occupancy and aesthetic quality. Fire
nents. These complexities can give rise to numerous safety, more specifically smoke and heat extraction,
risks, among which fire is frequently a central prob- requires optimization and careful analysis in the
lem (Harper, 2004). early design phases. These norms are the average of

real fire safety cases, and so in some specific cases, The engineering approach makes it possible to
they are more demanding than strictly necessary exceed the limits prescribed by the codes and at the
with regard to devices or shape layout, while in oth- same time rigorously respect safety as the norms
ers they could actually be insufficient. regulate only average scenarios. This approach al-
An impressive example of the impact of fire lows designers to have a greater compositional free-
regulations on architecture is represented by the re- dom, obtaining innovative and high-performance
cent design competition for the new Rome Confer- buildings in which fire safety has become an essen-
ence Centre in 1998. It is situated near the old one, tial element of architectural conformation. In other
in the district of the Esposizione Universale Roma words, fire engineering is used to allow fire safety
- EUR (Rome Universal Exhibition) planned for 1942 to be demonstrated, despite its peculiar form and
where, according to Mussolini’s town plan, impor- dimension like the new Congress Palace in Rome
tant ministries would subsequently be transferred. designed by D. and M. Fuksas - familiarly called “The
The old Conference Centre was designed by Adal- Cloud” [1, 2]. In this case, it was possible to use fire
berto Libera in 1939 and is a reflection of late Italian engineering simulation techniques to demonstrate
rationalist style tinted with a superficial, ironical and that the danger, in the case of fire, was very low as
monumental classicism. There were numerous im- the space enclosed by the ‘glass box’ could be con-
portant responses to the 1998 competition call and sidered an open space and as the thinness of the
several high quality projects were selected. Many membrane enveloping ‘The Cloud’ renders the fire
were based on mimicking the symmetrical and con- load negligible.
strained layout resulting from a simplistic interpreta- This approach has allowed modifications to be
tion of the apparently elementary nature of the EUR avoided during the detailed building design phase
buildings. Another important motivation was the that might compromise the identity of the project.
difficulty to take into account current safety rules re-
garding structure, plant engineering and evacuation FIRE SAFETY PRESCRIPTIVE APPROACH
paths in a free form space configuration. The combustion process is a sequence of chemical
In many cases, this design logic led to the vari- reactions between a fuel and an oxidant, accom-
ous conference halls being situated as low as pos- panied by the production of heat, smoke and the
sible and for fire fighting purposes to use water conversion of chemical materials. The process can
cisterns (filled with water) at the top of the building. be specified by means of the Fire Triangle, which is
These design solutions had the following conse- composed of fuel, oxygen and a heat source. It is es-
quences: overloading at the building top, which is a sential for a correct relationship among these three
design solution to be avoided in view of the seismic elements otherwise, combustion itself cannot take
nature of Italian territory; an obstruction of the visu- place (Harper, 2004).
al permeability between inside and outside in order Inside a compartment - i.e. a homogeneous and
to respect the fire resistance of the walls; the denial limited part of the building with respect to function,
of whole roof level availability and limited panoram- use destination and fire safety class - it is possible to
ic view due to the presence of cisterns. Other exam- identify four fire phases (La Malfa, 2009):
ples are: the Twin Towers, which had cisterns on the Ignition, a heat source acts on the fuel, and if suf-
roof that unfortunately failed to cope with the com- ficient thermal capacity is released, it warms it up to
bined effect of fire and the abrasion of the intumes- its ignition point. The thermal energy needed to at-
cent paint protecting the steel structure; or the first tain the ignition temperature depends on size and
HKSBC projects in which water inside the pillars was the ratio between the mass and the surface exposed
thought to cool the structure at the price of adding to the air.
permanent loads. • Growth, the fuel materials are heated and tend

Figure 1 • Method of contrast, for the achievement of the
The HRR curve over time dur- safety objectives set (obstacles to combustion
ing burning process. product propagation, smoke devices, fire extin-
guishing systems and fume extractors);
FIRE SAFETY PERFORMANCE APPROACH

Building codes are shifting from a prescriptive-based
method towards a performance-based method due
to the progress in fire safety technologies, includ-
ing the development of an engineering approach
(McGrattan et al., 2010; Hadjisophocleous and Ben-
ichou, 1999). The traditional prescriptive method
uses a set of technical standards that are rigidly ap-
to reach their ignition temperature. The spread plied in a ‘mechanical’ way. The performance method
of the fire produces: a reduction in visibility, in- allows the actual risks for specific activities to be
creased toxic fumes, increment of the burning evaluated step-by-step by means of careful analysis
rate over time. and simulation.
• Development, all combustible materials in the In Italy, the engineering approach to fire safety
compartment are simultaneously involved is regulated by laws, the most important of which
in the burning process due to the irradiation (D.M., 2007) defines the procedural aspects and cri-
caused by the products of combustion: the teria for assessing the level of risk and consequently
“flashover” phenomenon. the mandatory design measures intended to con-
• Decay, the fire tends to slow down owing to the trast possible code violations.
progressive reduction of combustible materials
or oxygen and starts to be extinguished. Heat Release Rate Parameter
Codes do not involve an integrated approach nor do The Heat Release Rate - HRR - is the main parameter
advanced CAD tools like building information mod- governing the fire phenomenon; it influences many
elling - BIM, but they do provide a list of documenta- other fire characteristics. Literally, the HRR indicates
tions and prescriptions to be fulfilled, which are use- the heat released by the combustion of a material
ful in the early design phases (Balaban et al., 2012): over time per unit surface area (Babrauskas, 1991).
• Definition of project, detailed description of The area under the curve (Figure 1) represents
building with particular reference to ventila- the energy released during all phases, while, for the
tion openings, fire and smoke compartments, purpose of fire safety, it is essential to evaluate the
structure and distribution of furniture and phase preceding the flashovers, because after this
combustible materials; time conditions are created that are unsustainable
• Fire safety objectives and indication of perfor the human body.
formance requirements, in relation to specific For this reason, it is necessary to know the varia-
architectural goals and to requirements for tion over time of the actual fuel mass involved which,
which the analysis is applied (maximum gas in the growth phase of the fire, is expressed by the
temperature at human head height, visibility, equation (1). It displays different curves pertaining
air concentration of toxic substances); to fire growth (Figure 2) depending on the time “t”
• Determination of fire scenarios, schematizing by means of a quadratic function and on the con-
events that may occur in relation to the charac- stant “α” that takes into account different material
teristics of fire, the building and the occupants; types. The combustion speed can be: slow, α = 2.77

× 10-3 KJ/s3; medium, α = 11.11 × 10-3 KJ/s3; fast, α = Figure 2
44.44 × 10-3 KJ/s3; ultra-fast, α = 177.77 × 10-3 KJ/s3 The HRR curves over time for
(La Malfa, 2009). different materials.
HRR = α × t² (1)
The curves do not grow indefinitely, but reach a
peak and then begin to decrease, so the quadratic
part refers only to the crescent monotonic curve.
The decay phase, for common materials, accounts
for 20÷30 % of the whole combustion process.
Curve peak values change according to the ma-
terial being burnt. For example, plastic rubbish has a
peak of 80 KW, while a car can reach 6000 KW.
As the HRR increases, also the temperature and
the rate of temperature rise increase, thus acceler-
ating fire development. In addition, increased HRR the density of the -ith material component of the
results in reduced oxygen concentration and in- layer at temperature Ts, divided by the initial density
creased production of gaseous and particulate mat- of the layer.
ter; these are fundamental factors to be considered (2)
for fire safety. The model used is based on the FDS (Fire Dy-
namics Simulator) language, a simulation code for
Material Reaction Rate to Fire low-speed flows, focused on smoke, particle and
For a given material, the reaction can produce a heat transport by fire. This model provides the es-
solid, denoted as residue, plus water vapour and/or timation of the fire’s evolution, dividing the space
fuel gas; for instance, the evaporation of water from into a large number of small contiguous elements
a solid material is described by the reaction that where the thermodynamic state is calculated by
converts liquid water-to-water vapour (McGrattan et solving the conservation equations of mass, energy,
al., 2010). etc. (Ozel, 1998). methods and analysis of results, the
A pyrolyzing solid in a reaction produces a solid field
residue, water vapour and fuel gas, the sum of which This approach allows the problem to be solved
has the same weight; this means that the mass of by integrating a set of partial differential equations
the reactant is conserved. for the whole system, thus avoiding the explicit
Another important parameter to consider is the treatment of the boundary conditions. One of the
mass fraction that can be burnt at time “t” (Figure 3,
blue curve) of the normalized density of material, Figure 3
which decreases as the sample is slowly heated. The Reference Temperature. T_p=
reaction rate (Figure 3, green curve) is the rate of 300 °C ; r_p= 0.002 s^(-1); T=
change of the mass fraction at time “t”; where this 5K/min ; v_s= 0.
curve peak is referred to as the reference Temperature
“Ts”, which is not the same as the ignition tempera-
ture, but is the most important parameter for defin-
ing the reaction rate of a material.
Equation (2) defines the reaction rate - “r” - at ref-
erence Temperature Ts [°C], of the -ith material under-
going its -jth reaction; Ys,i defines the ratio between

Figure 4
Case study of a multi-purpose
exhibition centre at Tivoli - the
conference hall. The dashed
blue lines define the compart-
ment and the BEAM devices
measure the total obscuration
rate between two points.
main calculation problems is the actual choice of was applied, which simulated a fire that released
mesh: decreasing the size of the elements, a more heat but also a specific quantity of particles and
realistic simulation is obtained, although this re- gas, based on input data. The levels of HRR deve-
quires a longer computation time and greater hard- loped from the case study are based on information
ware power. derived from the experimental data and compared
The mathematical model used for solving the with the values identified in the technical literature
analysis of fire phenomenon and interface problems for those specific activities.
among contiguous space elements is the phase field In the case of a multi-purpose hall, the presence
model (Boettinger et al., 2002) that, by means of a of scene panels is assumed (Table 1) which increases
specific and infinitesimal mesh, can correct dynamic the total fire load.
interface problems. A rapid decrease of the HRR was observed due
to the action of the simulated sprinkler system
MULTI-PURPOSE EXHIBITION CENTRE AT (automatic opening), giving a value of the fire ex-
TIVOLI - CASE STUDY tinction coefficient. This was made possible only
The case study was an experimental master’s degree through the correct sizing of the sprinkler piping, by
thesis of a project for a multi-purpose exhibition cen- indicating the flow of the single sprinkler obtained
tre with a multi-storey underground parking station from the product specifications and the UNI regula-
in Tivoli, near Rome; the simulation was focused on tions (UNI EN 12845, 2005).
the compartment of the conference hall (Figure 4). Some fire scenarios of NFPA 101 (Coté and Har-
The fundamental value to be specified was, as rington, 2012) were determined in compliance with
previously mentioned, the Heat Release Rate - HRR. the law (D.M., 2007). Relevant critical scenarios rep-
This was schematized with a burner, on which a vent resentative of the actual conditions were produced

Table 1
Fire Load calculation in rela-
tion to the furniture or product
type.
using: a preliminary check of regulations; expected curve of heat release HRRPUA [KW/m²] derived from
performance levels; considering, for additional the sum of the value of the initial burner plus any
safety, the failure of the sprinkler system or devices contribution of fuel elements; visibility in the room
for automatic door opening. It was also possible to for the egress time (by rate dimming detectors -
simulate fire scenarios that are worse than the NFPA BEAM); fume, particle and heat flow trends in the
101 code fire scenarios. Those could be the most se- compartment; sprinkler operation.
vere fires ever recorded, or the average of the worst The main architectural problems were encoun-
fires having occurred with some regularity. The fire tered both:
scenarios were examined for a period of 10 minutes, • outside the building, the relationship with the
an appropriate time for a preliminary investigation square (quality of urban space)
(Figure 5 and 6). • inside the building, the layout of the confer-
The choice of particulate reaction rate is one of ence hall within the overall shape of the build-
the most critical aspects (Kittle, 1993): for example, ing (quality of architectural composition).
in the pre-flashover phase, a modest amount of As far as the first point is concerned, namely the ur-
electrical equipment could spread smoke more dan- ban environment, two aspects must be taken into
gerously than a common fire with greater reaction consideration which heavily affect the architectural
rate levels. composition and the cost of the building: the time
All the simulations verified the fire scenarios required for firemen to arrive from the nearest fire
considered by analysing variations of: gas tempera- station and the building conformation, which is
ture at human head height and at construction el- linked to the constraints imposed by the under-
ement height (by temperature detectors - THCP); ground parking safety openings. In the case of the

Figure 5
Particle diffusion showed in
a longitudinal section of the
multi-purpose conference hall
study case 5 minutes after fire
ignition.
first aspect, the most important features are the to be mainly dependent on the materials used and
non-combustible nature of the materials used and/ to exclude any reference to the ‘form’. In actual fact
or the time needed for the bearing structure to re- the shape of the building, of its interior and the ac-
sist until the fire fighters arrive. This aspect seems cesses to it constrain the possible manoeuvres of
Figure 6
Gas temperature showed in
a longitudinal section of the
multi-purpose conference hall
study case 5 minutes after fire
ignition.

the fire fighting and rescue vehicles and the extin- reduced risk to people; freedom in the use and
guishing of the fire. The second aspect, the shape of choice of a larger number of materials, furniture and
the building and of the surrounding area is directly building components.
affected by the large air wells required to make the Moreover, if the traffic regulations governing
underground parking levels comfortable and safe. the area outside the building were to change and
It was observed that by simply applying the codes cause a delay in the arrival of the fire fighting vehi-
literally, these ‘wells’ heavily affected the restaurant cles it would be possible to simulate the behaviour
area, unless the latter was shifted further north and of the fire in the building to check whether in any
thus could not be used in the winter. However, ap- case the safety of visitors and staff were guaranteed
plying the model to an alternative proposal for the or whether any further action was necessary and the
location of the ‘wells’ in such a way as to leave the extent thereof. This simulation, extended to all the
positioning of the restaurant unchanged and simu- buildings affected by the change in traffic regula-
lating the fire process, it was possible to verify the tion, would entail at the town planning level a de-
safety. tailed estimate of the costs/benefits-safety ratio.
As far as the second point is concerned, as for The main difficulty encountered in the use of
example the compartment of the multi-purpose the simulation model was due to the limited inter-
hall, according to the code (UNI 9494-1, 2012), the face with other CAD programs as it exports only
smoke and heat exhaust openings must have a 3D (AutoCAD®, Blender®). The integration of this
continuous surface as long as the corridor length analytical model with BIM would be desirable, as has
at a height of over 1.2 meters, thus conditioning already been done with structural and thermal cal-
the compositional, formal and acoustical aspects. In culation programs. An integrated design could thus
addition, the façade would have to be constituted be totally exploited in the building process, thus fa-
by solid and non-combusting materials with high cilitating collaboration and information exchange.
temperature proofing. The simulation verified safety This study, which was developed in the form of
standards also with a less invasive action on archi- an experimental master’s degree thesis, should be
tectural composition, thus allowing different mate- included in an academic course. In Italy, although
rials and shape. The simulation computed that the there are many post-graduate professional courses
required fresh air brought into the conference hall concerning the applications of fire safety, most aca-
only through emergency doors was enough for peo- demic students ignore fire safety problems, so they
ple, for structural safety and for visibility even if the should be taught at least the main principles for fire
opening (doors and windows) size of the walls did safety and, secondly, those for high-performance
not satisfy those codes. After an accurate simulation, buildings.
glass façades were allowed as they would be shat- Nowadays, the buildings are often readapted
tered by the high temperature at the beginning of through a change of use, resulting in fire load vari-
the fire process. This is dependent on the capacity of ation; besides the building codes are constantly
fire engineering simulation accurately to predict the changing, giving rise to maintenance and accom-
fire status. modation activities with additional costs not speci-
fied in the design process. Moreover, the relation-
CONCLUSIONS AND FUTURE PROSPECTS ship between the study of fire phenomena and
The simulation performed led to better use of the architectural conformation becomes fundamental in
urban square (the prescriptive method imposed historical buildings, where architectural restrictions
huge underground parking ventilation openings); and heritage preservation do not actually go hand
more freedom in architectural composition (codes in hand with fire safety needs. All these problems
constrained doors and windows position and size); can be faced by means of the fire engineering simu-

lation approach that allows performance evaluation Boettinger, WJ, et al. 2002, ‘Phase-Field Simulation of So-
during a building’s lifetime. lidification’, Annual Review of Materials, Research Vol.32,
Fire safety engineering should not be inter- pp. 163-194.
preted, in the narrow sense, as a tool allowing the Coté, R and Harrington, G 2012, NFPA 101: Life Safety Code,
prescribed regulations to be sidestepped, but as a edition 2012.
system model that can analyse real cases in depth, DM (Ministerial Decree of Italy) 2007, Direttive per
afford more solutions to problems, achieve higher l’Attuazione dell’Approccio Ingegneristico alla Sicurezza
safety standards and a more accurate evaluation Antincendio (Guidelines for Accomplishments of the En-
and analysis of the risk of fire, so as to reduce prob- gineering approach to Fire Safety), MI - Ministry of the
lems from the outset. For example, in Italy this ap- Interior of Italy, 9 May 2007, G.U. 117, 22/05/2007.
proach, when adopted, has led to a new image of Hadjisophocleous and Benichou, 1999, ‘Performance Crite-
steel structures completely free of protective coat- ria Used in Fire Safety Design’, Automation in Construc-
ings (intumescent paintings or cement or stucco or tion, 8 (4), pp. 489-501.
aluminum and insulation panels), with evident sav- Harper, CA 2004, Handbook of Building Materials for Fire Pro-
ings in construction costs. tection, McGraw-Hill Companies Inc.
The performance-based method allows a pre- Kittle, PA, 1993, Flammability of alternative daily cover mate-
liminary evaluation during the design process, rials - A summary of ASTM E1354 cone calorimeter results,
avoiding not only increased costs but also modifica- West Chester, PA, October 1993.
tions that often distort the identity of the building. La Malfa, A and La Malfa, S 2009, Approccio Ingegneristico
The use of these tools during the various phases alla Sicurezza Antincendio, Legislazione Tecnica, (Ital-
of new construction design or in rehabilitation pro- ian).
jects regarding existing buildings or during a build- McGrattan, KB, McDermott, R, Hostikka, S and Floyd, J 2010,
ing’s lifetime can allow designers and the competent Fire Dynamics Simulator - User’s Guide, National Insti-
authorities to collaboratively develop projects, en- tute of Standards and Technology.
suring enhanced security also at urban level. There Ozel, F 1998, ‘Geometric Modelling in the Simulation of Fire
are two possible future developments of these tools: - Smoke Spread in Buildings’, Proceedings of the SIGraDi
a numerical one, through integration with other Conference (Seminar), Mar del Plata, Argentina, pp.
multi-physical tools (e.g. COMSOL®), or a semantic 438-445.
one using ontology tools and AI methodologies. UNI 9494-1, June 2012 ‘Heat and smoke control devices’ - De-
sign and installation of natural evacuation systems of
REFERENCES smoke and heat.
Babrauskas, V and Peacock, RD 1991, ‘Heat Release Rate: UNI EN 12845, February 2005 ‘Fixed firefighting systems -
the single most important variable in fire hazard’, Build- Automatic Sprinkler Systems - Design, installation and
ing and Fire Research Laboratory, National Institute of maintenance’.
Standard and Technology (NIST), Gaithersburg, Mary-
land 20899, U.S.A. pp. 255-272. [1] http://www.fuksas.it/#/progetti/0002/ (last accessed on
Balaban, Ö, Kilimci ESY and Çağdas, G 2012, ‘Automated 03/06/2013).
Code Compliance Checking Model for Fire Egress [2] http://www.eurcongressiroma.it/wp-content/
Codes’, Proceedings of the 30th eCAADe Conference, uploads/2011/11/Schede-tecniche-NCC.pdf(last ac-
Prague, Czech Republic, pp. 117-125. cessed on 07/06/2013).

Inductive Aerodynamics
Samuel Wilkinson1, Sean Hanna2, Lars Hesselgren3, Volker Mueller4
1,2
University College London, UK, 3PLP/Architecture, UK, 4Bentley Systems, US
1
smlwilkinson@gmail.com, 2s.hanna@ucl.ac.uk, 3lhesselgren@plparchitecture, 4volker.
mueller@bentley.com
Abstract. A novel approach is presented to predict wind pressure on tall buildings

for early-stage generative design exploration and optimisation. The method provides
instantaneous surface pressure data, reducing performance feedback time whilst
maintaining accuracy. This is achieved through the use of a machine learning algorithm
trained on procedurally generated towers and steady-state CFD simulation to evaluate
the training set of models. Local shape features are then calculated for every vertex in
each model, and a regression function is generated as a mapping between this shape
description and wind pressure. We present a background literature review, general
approach, and results for a number of cases of increasing complexity.
Keywords. Machine learning; CFD; tall buildings; wind loads; procedural modelling.
INTRODUCTION
It is generally recognised that architects currently re- Wind engineering has traditionally been within
quire performance information to guide their deci- the remit of engineers or specialists, with numeri-
sions almost from the inception of a project. In fact, cal simulation (CFD) considered a supportive tool to
there is a mentality present of simply trying to col- physical boundary layer wind tunnel (BLWT) testing.
lect as much data as possible with the intention of For instance, in the computational wind engineering
synthesising it into a situated design response. This (CWE) literature there is substantial caution around
presents a problem, especially for computational numerical analysis, namely for Reynolds-averaged
fluid dynamic (CFD) wind simulation, whereby the Navier-Stokes (RANS) and to a lesser extent large-
time required to assess the performance is obstruc- eddy simulations (LES) (Stathopoulos, 1997; Bitsua-
tive to the fast and iterative nature of current para- mlak, 2006; Dagnew et al., 2009; Menicovich et al.,
metric design softwares. This is possibly due to the 2002). However, architects are increasingly getting
tendency for architectural software tools to origi- involved with analysis, where concerns over accu-
nate in engineering fields, without due considera- racy are less paramount since demand is typically for
tion of speed-accuracy tradeoffs to adjust for the ap- relative scenario comparison or general flow behav-
plication requirements (Chittka et al., 2009; Lu et al., iour (Lomax et al., 2001; Malkawi et al., 2005; Chronis
1991). In other words, they are typically too accurate et al., 2012).
and slow for the fast pace of modern conceptual de- The tall building typology has been identified
sign, massing or form decisions. Developing a meth- as a focal area here for a number of reasons. Firstly,
od that can give real-time performance feedback as height increases so too do the wind forces (along
about a form allows for intuitive play of the kind we with seismic and gravitational) which has conse-
are used to with physical models. quences on facade panelisation and structural effi-

ciency, amongst others. We can construct a simple 1959), but in statistics is referred to as regression
motivational argument to say that increased exter- and in engineering as function approximation or
nal wind force requires more opposing force, i.e. surrogate modelling. Once trained, this enables
more structure, more materials, larger cores, less let- us to provide a new test case and make a predic-
able floor space, less revenue etc. Therefore there is tion of the outcome. Inductive reasoning, episte-
a need to consider the aerodynamic form of these mologically, means constructing generalisations
buildings as they increase in height. Secondly, the from specific information, as opposed to deductive
trend for tall buildings is to build them as high as reasoning where small details are construed from
(contextually, economically and structurally) pos- generalisations. The fundamental outcome of this
sible, necessitating cutting-edge design and con- learning approach is therefore a continuous output
struction technologies (CTBUH, 2012). Thirdly, tall response allowing interpolation and extrapolation
building form lends itself well to parametric design between cases that have not been explicitly simu-
as there is often a high degree of vertical logic that lated. In doing so, we are essentially moving the
can be expressed neatly with mathematical expres- simulation time from the front-end to the back-end
sions (this generalisation is at least more true than of the process where more time is available for pre-
for shorter buildings). Given this, it is possible to eas- computation.
ily generate a procedural, or generic, tall building The following section provides a review of rel-
model that, with a relatively small number of param- evant literature in the generative, performative
eters, can represent a large number of potential de- design of tall buildings, wind modelling methods,
signs. This becomes useful when the objective is to speed-accuracy tradeoffs, incorporation of learn-
sample the typological space of potential buildings, ing in design, concluding with a problem-solution
which will be discussed in the methodology. hypothetical argument positioned in this state of
We present a novel approach to predict wind current literature. The subsequent structure of this
pressure on tall building models for early-stage paper will describe the methodological approach in
generative design exploration and optimisation general terms, and results are presented from a se-
(exploration as the non-discrete parametric equiva- ries of experimental case studies of increasing com-
lent of tinkering, and optimisation as the single- or plexity from trivial to practical. The conclusions, fur-
multi-objective directed design space search requir- ther work and the paper as a whole are positioned
ing iterative testing and evaluation). The method within the scope of ongoing research.
provides fast surface pressure data with the conven-
tional visualisation, reducing performance feedback LITERATURE REVIEW
time whilst maintaining verisimilitude.
This is achieved through the use of a machine Tall Buildings
learning algorithm, trained on a pre-computed set Tamura et al. (2009; 2010) and Tanaka et al. (2012)
of CFD simulation data. ANSYS CFX 13.0, a commonly acknowledge the increase in tall building complex-
used solver in engineering practice, was used for ity beyond the traditional extruded rectilinear form.
steady-state RANS with a k-E turbulence model. The We are now seeing more unconventional free-style
learning technique is grouped with artificial neural forms derived from the architect’s use of more ad-
networks (ANN), support vector machines (SVM), vanced modelling software. These new complicated
and random forest (RF) decision trees, in that there sectional shapes that may vary with height, can ac-
is a training set of cases from which generalised tually provide better aerodynamic performance by
rules are generated (Duffy, 1997). The term machine disrupting, or ‘confusing’, vortex shedding and thus
learning stems from the fields of computer science reducing crosswind response. Benefits can also be
(Mitchell, 1997) and artificial intelligence (Samuel, found in more subtle manipulations such as corner

chamfering or cutting, and by creating voids, or po- ter decision. Biological examples have been noted
rous regions, near the edges. by Chittka et al. (2009), who explains that “when it
Despite rapid advances over the past century, takes a long time to solve a difficult task, and the po-
this emerging generation of skyscrapers poses new tential costs of errors are low, the best solution from the
challenges for wind engineering. Irwin (2009) dis- perspective of an animal might be to guess the solution
cusses a number of these, such as the impact that quickly, a strategy that is likely to result in low decision
aerodynamics have on construction cost. Since the accuracy.” The two extremes can be called impulsive
structure itself is a large proportion of the cost, and and reflective. This provides a neat analogy for per-
as for tall buildings the wind is the governing lateral formance analysis in design where it is necessary to
load, there are significant benefits to be had from consider what the application of the simulation tool
reducing wind loads. This also has the effect of re- is, and the consequent risks, before deciding a suit-
ducing lateral motions that can potentially cause able accuracy.
occupant discomfort. He also suggests that shape Crucially though, and in conjunction with this
aerodynamics must be proactively considered, and reasoning, Burns (2005) demonstrates that making
iteratively optimised, early on in the design. With the more decisions with more mistakes (fast and inaccu-
new generation of super-tall towers over 600m it is rate) results in better overall performance (with bees,
simply not possible to ignore the wind performance. more nectar collected) than the more fastidious
He quotes a designer of the Burj Khalif, saying “we (slow and accurate). Defining accuracy as the propor-
practically designed the tower in the wind tunnel”, tion of choices that are correct, this highlights that
and were therefore able to produce an extremely ef- accuracy should not be confined to the immediate
ficient aerodynamic shape that enabled the height task, i.e. simulation accuracy, but to the larger one of
with reasonable structural systems and costs, and improving building performance (Figure 1).
without any damping system. Response time is critical for performance-driven
The increase in the use of parametric CAD soft- design and SATs must be considered when deve-
wares has seen a rise in the last decade namely with loping early stage tools for when large-scale deci-
the release of Bentley GenerativeComponents and sions are made. Performance information is often
Rhino Grasshopper, plus more generally with the in- scarce at this stage and iterative decisions must be
creased adoption of scripting. These allow the user made quickly, necessitating fast response times in
to create parametrically associative relationships sync with the project cycles. The development of
related to geometry. The extension of this idea is to CFD models have been focused over the past dec-
use rules to define the parameters, or where these ades on improving accuracy, and computational
rules can be related to the performance of a mod- time is optimised by specific software vendors after-
el component the geometry is directed by some the-fact, with little thought given to the accuracy re-
evaluative metric. Certain metrics can be calculated quired by the user. In contrast, recent developments
quickly without problem, but if the calculation takes in computer graphics have started with the desired
time it becomes obstructive to the modelling pro- accuracy (believable) and speed (real-time) in mind,
cess. We adopt the premise that it is better to have a with successful results.
broader range of lower resolution data rather than a In the design context, CFD can typically be used
limited amount of exact data. for a number of purposes: analysis of internal air
movement, pollution dispersion, noise propagation,
Speed-Accuracy Tradeoffs pedestrian comfort in urban environments or tall
Speed-accuracy tradeoffs (SATs) show that response building aerodynamics. As mentioned previously,
accuracy generally increases with response time, i.e. it is the last that is the focal application here, espe-
taking more time to make a decision results in a bet- cially for early design stages. There is a paradox here,

Figure 1
(Left) SAT for various task
difficulties and skills; (Right)
Notional positions of different
modelling methods on SAT.
in that the most complex flow types (bluff bodies) often difficult since the systems are typically natural,
and therefore most computationally intensive, need and therefore can have randomness, heterogeneity,
to be modelled in a scenario where fast results are multiple causes and effects, and noise. Even when
required. The numerical method must be as accu- they are successfully extracted, they may be be-
rate and fast as possible. In fact, the conclusion is yond our understanding and are held as intractable
reached that the fastest method has poor accuracy computational functions or data structures. Hanna
and the slowest the best accuracy (as would be ex- (2011) tests the hypothesis that it is unnecessary to
pected, considering the speed-accuracy tradeoffs have any understanding of this underlying system
mentioned earlier). There is general agreement be- behaviour, but rather it is possible to make predic-
tween (Lomax et al., 2001) and (Chronis et al., 2012) tions about the system simply by making observa-
that the “level of accuracy of a CFD simulation needs tions. This is demonstrated by learning the struc-
to be compromised with the turnaround time require- tural behaviour of system components and applying
ments of its application.” them to larger-scale scenarios.
Lu et al. (1991) describe the same issue in me- Graening et al. (2008) propose a method that al-
chanical engineering where slow but accurate simu- lows the extraction of comprehensible knowledge
lation makes interactive decision making impossi- from aerodynamic design data (jet-blades) repre-
ble, when only quick estimates are desired at early sented by discrete unstructured surface meshes.
stages. It is only towards the final stages of design, They use a displacement measure in order to inves-
“when the engineer has converged to a small region of tigate local differences between designs and the
decision space, more accurate simulations are needed resulting performance variation. Knowledge, or rule,
to make fine distinctions.” The problem has therefore extraction from CFD data is primarily used to guide
been present since the early 90s, but as a solution human- centred design by improving understand-
they propose integration of simulation, optimisation ing of the system’s behaviour, whether it is for jet
and machine learning. turbine blade optimisation or architectural design.
Whilst the connection between local geometric fea-
Inductive Learning in Design tures and surface pressure has been extended and
Our approach is supported by Samarasinghe (2007), changed here, and used for a different application,
who identifies the best solution to predicting sys- this work is a close precedent.
tem behaviour through observational data. This is
necessary when there is little or no understanding of Problem Hypothesis
the “underlying mechanisms because of complex and It is argued here that approximations of CFD simu-
non-linear interactions among various aspects of the lations can be made with machine learning regres-
problem.” Extracting these complex relationships is sion, using geometric shape descriptors as the learn-

Figure 2
(Left) Examples of evaluated
procedural models in the
training set on Case 4; (Right)
Mesh feature extraction.
ing features. The entire evaluation process can be The geometry for the training set was gener-
broadly split into five key work areas: i) procedural ated using a procedural tall building model with a
geometry generation; ii) batch simulation; iii) shape select number of key parameters (Figure 2). There
feature generation; iv) machine learning training; are in fact three separate topologies in the proce-
v) prediction and visualisation. Feature generation dural model with their own parameters, since it is
is essentially the core of the process since the solu- difficult to incorporate the entire design space with
tion depends heavily on geometric description so one parametric logic (Park et al., 2004; Samareh,
as to define surface pressure as a function of it. We 1999). Using the unstructured triangulated surface
hypothesise that surface pressure distribution aris- mesh from these means we are not limited by a sin-
ing from wind flow around tall buildings can be gle parametric topology in the learning phase of the
learnt and predicted with an accuracy appropriate method (Graening et al., 2008). Local surface-mesh
to early stage design (feedback from practice indi- shape characteristics are used as input features to
cates <20% error) using shape feature description. the learning algorithm instead of the design pa-
It can be shown that it is possible to combine, with rameters, avoiding reliance on any one parametric
an acceptable error, methods that have the separate model definition.
contradictory objectives of predictive accuracy and
speed. Simulation Method
An established solver, ANSYS CFX 13.0, was used
METHODOLOGY throughout to run the RANS steady-state simula-
tions, with a k-ε turbulence model as it is regarded as
Data Set Generation: Procedural Modelling the most robust. Each simulation, depending on the
The parametric model was created in Bentley Gen- complexity, requires up to 60 minutes to converge
erativeComponents. The goal was to create a gen- (on a 2.66GHz i7). Solver convergence is reached
eralised tower model, with the two properties of when residuals fall below a minimum of 1−6, typically
minimising the number of parameters used whilst at around 100 to 200 iterations. The number of cells
maximising the design representation potential, in the tetrahedral meshes varies between 0.8x106
i.e. the number of possible buildings it could cre- and 1.5x106 depending on the geometry, with pris-
ate. This is important when considering optimisa- matic expansion on surfaces 3 cells deep and a mini-
tion or exploratory design space searches to avoid mum cell size of 0.1m. The wind was applied at an
the curse of dimensionality. This means that as the upstream inlet, with a reference speed (Ur) of 1ms−1
number of variables increases, the design space in- at a reference height (Zr) of 10m. The most common-
creases exponentially by nD, where n is the number ly used distribution of mean wind speed with height
of samples taken per parameter and D is the number is the ‘power-law’ expression:
of parameters, or dimensionality. There is therefore Ux = Ur ( Zx / Zr ) α (1)
clearly a compromise to be made between model The exponent α is an empirically derived co-
efficiency and representability. efficient that is dependent on the stability of the

atmosphere. For neutral stability conditions it is training:validation. For the first two cases, separate
approximately 0.143, and is appropriate for open- sets constituting entire models were also held back
surroundings such as open water or landscape. for testing, i.e. training was at 15° and 20m inter-
Future work will include a wind profile that takes vals respectively. For the third case, there was no
surrounding surface roughness, or context, into ac- extra test set but the whole was split 70:15:15% to
count, as well as potential wind direction change training:validation:test. Validation data is to check
with height. for convergence during training. For the fourth case,
training data was from the procedural tall building
Shape Features and Learning model and test data from another set of real build-
This method creates a definition for the pressure at ings. In this case, a Random Forest (RF) algorithm
a point on the model as the function of a local geo- was used instead as it provided better results for
metric description. To describe a simple example the more complex problem. Further work is needed
of the process: there are N models of a cuboid with with both methods to understand their applicability
various orientations; each is evaluated, and the pres- to certain tasks, however it is known that the RF is
sure P is extracted at M points over each model; for better with noisy data sets than the ANN. Training
every M, a shape descriptor X is calculated, such as set sizes and summary results are given in Table 1,
the vertex height, normal components, curvature, and computation times are given in Table 2.
etc; this gives a set of geometric characteristics, and
a corresponding pressure value; these sets of P(X) RESULTS
are used as the training data. Pressure distribution
is predicted from these geometric descriptors alone Cuboid Orientation
meaning the selection is critical. A sensitivity analy- The first and most simple test is the rotation of a
sis has been conducted with a variety of descrip- cuboid, of width and depth 10m, and height 50m.
tors to determine suitable representation, details Simulations were run at 5° intervals from 0 to 85°,
of which are not included here. When a new case is and the ANN trained on 15° and tested at 5° inter-
presented, the shape descriptors are calculated and vals. The sensitivity analysis here varies the number
used to make a prediction of P. The feature definition of training samples and measures the standard de-
for point pressure in R22 vector space used through- viation, σ, of the difference between simulation and
out the following is: prediction. Figure 3 (left) shows the error σ against
P ( Z, N(x,y,z), Nσ1-5(x,y,z), U(x,y,z)) (2) orientation for various set sizes (bold vertical lines
For a specific model vertex, P is the surface pres- are training intervals of 15°), (centre) the training
sure, Z is the height, N(x,y,z) are the normal compo- regression of the entire set, and (right) the predic-
nents, Nσ1-5(x,y,z) is the standard deviation σ of normal tion error for an orientation of 25°. With less training
components of cumulative mesh neighbourhood data, it can be seen that error is highest around 45°
rings 1 through 5, and U(x,y,z) are the normalised when flow bifurcations (regime change) occur, al-
model position components. The extent of the though this is negated with sufficient data.
neighbourhood curvature can be extended beyond
5 rings, within computational resource limits. The Cuboid Height
definition in Equation 2 gives 22 inputs and 1 output Secondly, a parametric cuboid was created with
feature to train the learning algorithm for all cases width and depth 10m, and height varying from 10
described below. to 100m in 5m increments. Figure 4 (left) shows the
For the Orientation, Height and Topology variability when trained on 10, 20, 30 and 45m inter-
cases, an Artificial Neural Network (ANN) was vals, and (right) the prediction error for a height of
used, with a 70:30% split of the provided data to 25m when trained at 20m intervals.

Table 1 Case Min σ Error (%) Max σ Error (%) Training Set Size
Summary of minimum and Orientation 1.2 (55°) 1.6 (10°) 110000 (15° training intervals)
maximum error standard Height 0.7 (10m) 2.0 (50m) 44720 (20m training intervals)
deviations (% over test case Topology 1.8 (5 Edges) 3.5 (0 Edges) 50000
pressure range). Real 4.8 (Bank of China) 18.3 (Euston) 100000 (Procedural training)
Table 2 Case Train Sim. Train Feat. Gen. † Train Predict Feat. Gen. * Predict *
Summary of time (seconds) Orientation 21600 9060 2600 1540 < 0.1
required for each case, split Height 18000 2370 720 620 < 0.1
into Training (one-off back- Topology 32400 4670 1060 1750 < 0.1
end time) and Prediction Real 2160000 12000 620 720 < 0.1
(front-end time). Mean feature
generation time is 0.085s/
vertex. *Mean over all test set. Topology Tall Buildings
†
After down-sampling. Here the number of edges was varied from 3 to In the final case, training data was collected from
10, with 0 (circle), diameter 10m and height 50m. simulations of 600 procedural tall building models,
Instead of keeping a complete model separate for with a total of over 4x106 shape features extracted.
testing as in the last two cases, here all cases were This was down-sampled to 105 by removing features
used but only a fraction of the total data set was in close proximity to reduce training time. The test
used. This is varied in Figure 5 (left), with a training set contains 10 real tall buildings from around the
set ranging from 10000 to 50000. world, selected for their range of unique architectur-
Figure 3
(Left) Orientation vs. Error
σ %; (Centre) Training set
regression, R=0.99564; (Right)
Prediction error (25°).
Figure 4
(Left) dHeight vs. Error σ %;
(Centre) Training set regres-
sion, R=0.9992; (Right) Predic-
tion error (25m).

Figure 5
(Left) No. Edges vs. Error
σ %; (Centre) Training set
regression, R=0.98355; (Right)
Prediction error (n0).
al characteristics. Figure 6 shows predicted surface for less time (Table 2), with the methodology and
pressure distribution in the top row, and the error constraints described. These prediction errors are
distribution for the set in the bottom row. The pres- necessary for the compromise in avoiding consider-
sure range (-5.5 to 2.0 Pa) was taken over the entire ably intensive CFD simulation. Traditionally, for eve-
test set, as was the absolute error range (0 to 65.2%). ry individual CFD simulation the process can take a
The error distribution is shown in Figure 7 (right), minimum of 1 hour, compared to our methodology
which fits a Gaussian normal distribution. Error per- that has a total front-end prediction time of under
centiles: 99th = 35.7%, 95th = 20.0%, 90th = 13.0%, 75th 12 minutes (for feature generation and prediction)
= 6.1%. That is, 75% of the test features have an error and a back-end, one-off training set simulation time
below 6.1%. of 600 hours (for the real case). Once trained, an un-
limited number of predictions can then be made.
CONCLUSION Whilst these preliminary results are outside the
The results show that it is possible to achieve a rela- rigorous accuracy necessary for final engineering
tively small prediction error (Figure 7 and Table 1) analysis, they are within the boundaries acceptable
Figure 6
(Upper) Predicted pressure,
Pa; (Lower) Error, %. Pressure
range is the min. and max. of
the entire set for comparison,
the error range is absolute
max. error of the set (65.2%).
(Left to right) (1) Metlife
Building, NYC; (2) The Shard,
London; (3) Willis Tower
(Sears), Chicago; (4) Euston
Tower, London; (5) Taipei
101, Taiwan; (6) Shanghai
World Financial Centre; (7)
Bank of China; (8) Exchange
Place, NYC; (9) Frankfurter
Buro Centre, Frankfurt; (10)
Washington Street, NYC.

Figure 7
(Left) Error σ % for each case;
(Centre) Random Forest learn-
ing convergence; (Right) Error
probability density.
for early-stage concept design for tall buildings, sign. IEEE Expert, 12(3), pp.71–76.
where interactive response time is a significant con- Graening, L. et al., 2008. Knowledge Extraction from Aero-
sideration. The prediction accuracy and response dynamic Design Data and its Application to 3D Turbine
times achieved are promising for further work given Blade Geometries. JMMA, 7(4), pp.329–350.
the well-known complexities of fluid behaviour. Hanna, S., 2011. Addressing complex design problems
The next stages of the work are to consider time- through inductive learning.
dependent simulations to fully consider the approxi- Irwin, P.A., 2009. Wind engineering challenges of the new gen-
mation of turbulence, vortex shedding and gusts, as eration of super-tall buildings. JWEIA, 97(7-8), pp.328–
well as interference from complex urban contexts 334.
on boundary conditions, and further improvement Lomax, H., Pulliam, T.H. & Zingg, D.W., 2001. Fundamentals
to the shape feature selection and generation time. of computational fluid dynamics, Berlin: Springer.
Lu, S.C.-Y., Tcheng, D.K. & Yerramareddy, S., 1991. Integration
ACKNOWLEDGEMENTS of Simulation, Learning and Optimization to Support En-
This research was sponsored by the EPSRC, Bentley gineering Design. Annals of the CIRP, 40(1), pp.143–146.
Systems and PLP Architects. Malkawi, A.M. et al., 2005. Decision support and design evo-
lution: integrating genetic algorithms, CFD and visuali-
REFERENCES zation. AIC, 14(1), pp.33–44.
Bitsuamlak, G., 2006. Application of computational wind en- Menicovich, D. et al., 2002. Generation and Integration of an
gineering: A practical perspective. In Third National Con- Aerodynamic Performance Database within the Concept
ference in Wind Engineering. pp. 1–19. Design Phase of Tall Buildings.
Burns, J.G., 2005. Impulsive bees forage better: the advantage Mitchell, T.M., 1997. Machine Learning, McGraw-Hill.
of quick, sometimes inaccurate foraging decisions. Ani- Park, S.M. et al., 2004. Tall Building Form Generation by Para-
mal Behaviour, 70(6), pp.1–5. metric Design Process. In CTBUH. Seoul Conference, pp.
Chittka, L., Skorupski, P. & Raine, N.E., 2009. Speed-accuracy 1–7.
tradeoffs in animal decision making. Trends in ecology & Samarasinghe, S., 2007. Neural Networks for Applied Sciences
evolution, 24(7), pp.400–7. and Engineering: From Fundamentals to Complex Pat-
Chronis, A. et al., 2012. Design Systems, Ecology and Time. In tern Recognition, Auerbach Publications, NY.
ACADIA. Samareh, J.A., 1999. A Survey of Shape Parameterisation
CTBUH, 2012. Tall Buildings in Numbers : A Tall Building Re- Techniques. In CEAS/AIAA/ICASE/NASA Langley Inter-
view, 2012(1). national Forum on Aeroelasticity and Structural Dy-
Dagnew, A.K., Bitsuamlak, G. & Merrick, R., 2009. Computa- namics. pp. 333–343.
tional evaluation of wind pressures on tall buildings. In Samuel, A.L., 1959. Some Studies in Machine Learning Using
11th Americas Conference on Wind Engineering. the Game of Checkers. IBM JRD, 3(3).
Duffy, A.H.B., 1997. The “what” and “how” of learning in de- Stathopoulos, T., 1997. Computational wind engineering:

Past achievements and future challenges. JWEIA, 67-68, tions. Structures Congress 2010, pp.278–278.
pp.509–532. Tanaka, H. et al., 2012. Experimental investigation of aerody-
Tamura, Y., 2009. Wind and tall buildings. In EACWE 5. namic forces and wind pressures acting on tall buildings
Tamura, Y. et al., 2010. Aerodynamic Characteristics of Tall with various unconventional configurations. JWEIA, 107-
Building Models with Various Unconventional Configura- 108, pp.179–191.

Adaptive Fluid Lens and Sunlight Redirection System
Exploring a novel way of redirecting and altering sunlight in large span

roofs
Florian Heinzelmann1, Telesilla Bristogianni2, Patrick Teuffel3

1,3
TU Eindhoven, Faculty of Built Environment, Unit of Structural Design, Chair of Inno-
vative Structural Design 2SHAU Architects
1,3
http://www.tue.nl/universiteit/faculteiten/faculteit-bwk/de-faculteit/organisatie/units/sd/
2
http:/www.shau.nl
1
f.heinzelmann@shau.nl, 2t.bristogianni@shau.nl, 3p.m.teuffel@tue.nl
Abstract. The paper describes a novel system to alter and redirect sunlight under large
span roofs with the help of a fluid lens system. Focus lies on the computational design,
testing, measurement and evaluation of the performance of a physical prototype.
Keywords. Daylight; large span roofs; optics.
OBJECTIVE
The general aim of the presented research is to deve- comfort but also energetic aspects in terms of heat
lop a design methodology with the help of compu- gains are part of the design and research scope. The
tational tools and prototypes in order to be able to research on the adaptive fluid lens and sunlight re-
design adaptive daylight systems in large span roofs direction system described here is one of the case
under consideration of user and functional require- studies being developed within the framework of
ments and respectively daylight performance as- an ongoing PhD research at TU Eindhoven. In this
pects. Adaptive solutions for vertical facades with specific case the objective is to find a way to capture
relatively small rooms and individual requirements and utilize sunlight to be used under a large span,
of the inhabitants in form of e.g. louvers are well re- horizontal roof in order to be adaptively redirected
searched and applied. In the case of large span roofs where needed and dynamically treated or altered
where the horizontal part of the envelope exceeds in such a way that it can fulfill various functional as-
the vertical one for admitting light, the design re- pects (Figure 1). This horizontal “window” has to con-
quirements and parameters for daylighting are dif- tinuously mediate between the dynamic yet known
ferent. Here collective lighting requirements and path of sunlight in relation to the location and the
adjustability for a higher amount of inhabitants, possible change of interior use or functionality and
diversity of functions, larger spatial entities and the thus lighting requirements.
geometrical relations between sun path and general
roof alignment towards the zenith play a major role. PRINCIPLE
However not only quantities of light according to The adaptive lens and sunlight redirection system
regulations or qualitative aspects in form of visual consists of two major components (Figure 2). Firstly

Figure 1
Objective of the adaptive fluid
lens and sunlight redirection
system.
a set of mirrors which as a whole orient themselves DESIGN APPROACH

towards the general horizontal sun direction (azi- The design approach consists of five steps. (1) Physi-
muth) and individual rows of mirrors which are ro- cal principles (2) Associative 2-D and 3-D models
tated in the same angle according to the sun alti- (3) Simulation (4) Prototype (5) Prototype measure-
tude. The altitude orientation is done in such a way ments and evaluation
that the incoming sunrays are reflected downwards
into an aperture which houses the adaptive lens. Step 1: Physical principles
The lens itself consists of a transparent horizontal In order to apply physical principles like Snell’s law
lower surface, a casing and on top an elastic deform- of Refraction or Fresnel’s equation for reflection and
able and transparent membrane. By changing the transmission (Bennett, 1995) several associative
internal volume via pumping a clear and transparent files were set up in the Rhino/Grasshopper environ-
liquid, the shape of the membrane can be changed ment in order to see the effects of light refraction,
from concave to convex and continuously all the sta- transmittance, absorption or reflectance of various
ges in between, thus being able to diverge or con- geometry/material combinations and evaluate the
verge direct light according to Index of Refraction possibilities for sunlight redirection and alteration.
(Taylor, 2000) and Snell’s law (Taylor, 2000).
Figure 2
Description of the proposed
system.

Table 1
Sun altitude occurrences for
Munich.
Step 2: Associative 2-D and 3-D models riety of altitude angles. To reduce the height of the
Snell’s laws is most relevant and was translated into mirror system located on top of the lens, an array
associative 2-D sectional drawings of a light con- of mirrors was chosen instead of one larger mirror.
verging/diverging lens and a mirror system in order This approach poses several challenges in term of
to understand the capabilities and performance of overshadowing each other and being able to redi-
the proposal under changing light directions. In rect sunlight in different quantities according to the
this initial step an adaptive lens system was set up sun’s altitude angle and it turned out that there is
which is able to change the radius of an upper lens no universal system which works equally well in for
and Index of Refraction of the contained liquid ac- every sun altitude. Therefore it is necessary to take
cording to material properties of existing fluids and a closer look at several design parameters and be-
Snell’s law in order to focus or diffuse light. Here come specific about location, the respective avail-
an array of vectors is refracted within the lens and able hours with sunshine, the annual and daily sun
made visible via a bundle of lines to serve as design path and the prevalence of certain ranges of sun
and early evaluation tool. The change in altitude an- altitude angles and times of occupancy of the build-
gles of the sun leads to a change of the focal point ing. By matching these parameters it is possible to
of the refracted light and it showed that the redirec- narrow down the target range of altitude angles
tion possibilities of a lens are limited. This means where the redirection of sunlight is working a hun-
in order to keep the light e.g. diverged in terms of dred percent. By choosing a design example in Mu-
a constant area size, the membrane’s geometry has nich and as function a train station which is heavily
to be continuously changed by pumping liquid due frequented during the rush hours, lower sun altitude
to the resulting change of the focal point in relation angle ranges, which occur more frequently during
to the sun’s changing altitude. For this fact and the mornings and evenings but also during spring, au-
lens’ limited possibility of light redirection into the tumn and winter become more relevant (Table 1).
interior a secondary system for light redirection is re- By applying the Galapagos genetic algorithm solver
quired. Therefore several options like trapping light [1] to generate and validate variations of the fixed
by internal reflections (glass fiber principle), a rotat- inclination of the whole mirror array, the distance,
able prism system or plain mirrors were evaluated. sizes and amounts of the individual mirrors, a whole
The system of rotating mirrors was favored, because set of design solutions is produced which redirects
this proofed to be more “straightforward” and prom- sunlight altitudes perpendicularly to the ground by
ising in terms of light redirection under a greater va- a 100% within a Δ range of 30o (Figure 3). It is then a

Figure 3
Grasshopper study on the per-
centage of sunlight redirected
perpendicularly to the floor
according to the sun altitude.
matter of selecting the configuration which is most as VRay are able to calculate caustic effects [3] with
suitable for the design task at hand. In the example physically correct material properties and Index of
of Munich, a mirror configuration was eventually Refraction but are not able to display physical values
chosen which operates perfectly between 10-40 de- such as Illuminance, etc. It was therefore decided to
grees. After evaluating the design principles in the design and manufacture physical prototypes for the
earlier steps associative three-dimensional files were performance evaluation in accordance with the ear-
set up to further evaluate the behavior of the lens lier findings from the associative 2-D and 3-D mod-
and mirror system and also to have a geometrical in- els.
put for later daylight simulation.
Step 4: Prototype
Step 3: Simulation During the design process for the prototype, re-
The simulations were done via Diva/Radiance [2] search was done for lens diameters, change of vol-
and the VRay renderer [3] also available for Rhino ume and weight on the roof for a 1:1 case (Figure 4).
3D. The Radiance simulation was initially regarded In general it can be said that the higher a roof is situ-
as being important because it is able to show physi- ated above ground the less of a shape change in the
cal values like illumination in lux or luminance in cd/ lens has to occur in order to achieve a desired effect.
m². This would enable to check the performance for By studying theses parameters it was decided that
actual conditions and requirements as stated in e.g. a lens with a diameter of 1m would be optimal for
building codes. However the various simulations many applications in terms of weight and required
done proofed to be not accurate since Radiance volume change within the lens.
for windows is not able to calculate optical effects The final prototypes which serve as proof of
with dielectric material properties properly (Jacobs, concept and are used for daylight performance
2012). This has to be done in the Linux environment measurements were manufactured in the scale 1:10
with the help of a photon mapping module, which (Figure 5). The majority of parts including the me-
was developed by the Fraunhofer ISE [4]. This ap- chanical parts like gears and cograil for the sunlight
proach for simulating several different and adaptive redirection device are made of white ABS plastic and
geometries and the consequence to manually input 3-D printed by a Fused Deposition Modeling (FDM)
the data in Radiance for Linux defies the seamless printer. For the mirrors 3M™ Solar Mirror Film 1100 is
integration of parametric modelling and simula- applied on the rotatable ABS fins. The membrane for
tion. Curiously contemporary render engines such the lens is a self-cast and baked Polydimethylsilox-

Figure 4
Volume change required for
the convergence and diver-
gence of light in relation to the
room height.
ane (PDMS) membrane provided by Michael Debije cast sky, an Artificial Sky Simulator was employed, to
at Functional Devices research group of the Depart- achieve diffuse lighting conditions. Through all test
ment of Chemical Engineering and Chemistry at TU series, illumination measurements were done using
Eindhoven. Water with an Index of Refraction of 1,33 a Hagner Digital Luxmeter EC1 and lumination pic-
(Lide, 2009) is used as optical liquid. Other liquids tures were taken with a Canon EOS 60 D and further
like colorless and transparent oils which generally processed in Photolux 3.2. [5]
have a higher Index of Refraction are also thinkable.
Clear sky with sun, Test 1 and 2 set up
Step 5: Prototype measurements and The first two series of tests (Test 1 and 2) focused on
evaluation the performance of the adaptive fluid lens alone un-
The physical experimentations with the 1:10 scale der clear sky, supposing an ideal situation of 100%
prototype aimed at testing the performance of incoming perpendicular to the floor light, which
the system under clear sky with sun (Test 1-3) and would occur if the sun redirection systems func-
cloudy sky conditions (Test 4) (Figure 6). For the sim- tioned perfectly. To simulate the above, the altitude
ulation of the clear sunny sky, a Solar Simulator was of the solar simulator was set to a 90o angle. Two
used, providing directional light, while for the over- different in size closed boxes (Test 1: 0,5*0,5*0,35m,
Figure 5
Aspects of the adaptive fluid
lens and sunlight redirecting
system prototype in 1:10 scale.

Figure 6
Test settings: From left: Solar
Simulator at a 90o altitude
(Test 1), Solar Simulator
inclined at a 30° altitude (Test
3). Artificial Sky Simulator
(Test 4).
Test 2: 0,7*0,7*1m ) with a circular opening at the that a lens of 1m diameter is more efficient over a
center of their top surface for the 10cm diameter flu- 10m high room than a 3,5m room, as previously es-
id lens to be placed over, were used as room models. timated. To be more specific, in the case of the 1m
Water was pumped in and out of the two syringes box, the removal of 55ml of water from the lens in
connected to the lens, to reconfigure its shape from neutral state causes a circular lit area on the floor of
neutral to convex and concave. These tests approxi- 0,46m diameter (concave lens) while the addition
mate the performance of a 1m diameter fluid lens in of 12ml produce a focal point on the floor of 0,01m
a 3,5m and 10m high room respectively. diameter (convex lens) (Figure 8). In full scale the val-
ues would be 10m, 55L, 4,6m, 12L 0,1m. At the 0,35m
Test 1 and 2 observations and comparison high box r, when an almost equal water volume
The tests showed that under clear sunny sky condi- (58ml) is removed from the neutral lens, a lit area of
tions the light is indeed diverged or converged ac- 0,22m is produced. Furthermore, in order to achieve
cording to the configuration of the membrane and focused light on the floor at a point of 0,005m diam-
similarly to the predictions from the grasshopper eter, 26ml of water need to be added to the neutral
models (Figure 7). Comparing the two room scenar- lens. In full scale the values become 3,5m, 58L, 2,2m,
ios and scaling up the results to 1:1, it is confirmed 26L and 0,05m accordingly. In general, at the con-
Figure 7
Differences in the quality of
shadows from the concave to
the convex mode of the adap-
tive fluid lens.

Figure 8
Test 1: comparison between
the diverged and converged
mode.
cave mode, the lens is acting as a spotlight spread- for high contrast ratios in the room that not only
ing out the received sunrays. The light hitting the exceed the acceptable contrast thresholds for visual
floor surface and reflected by it causes the formation comfort but also surpass the 1:1000 ratio which is
of soft shadows by the scaled human figures placed the range of brightness the human eye can perceive
at the periphery inside the box. (Green et al, 2008). The most extreme converging
Considering the focused mode, the flux density mode might not be applicable for daylighting. How-
at the center of the floor surface (63.000lux for Test ever other uses are possible as described in the out-
1 settings) is excessively high in comparison to the look section of this paper.
density measured at the periphery (36lux for Test 1
settings). Given the fact that in Test 1 the sun simu- Test 3 set up
lator produces a value of 908lux at the floor center Test series no. 3 examine the effectiveness of the
and scaling up the findings, we can assume that on a sunlight redirection system on a clear sunny day
clear sunny day in summer where 100.000lux reach (Figure 9). For these tests, the 0,5*0,5*0,35m box was
the ground (Flesch, 2006), the flux density will be used as a room model and the solar simulator was
6.940.000lux at the center and 4.000lux at the pe- set at 30° sun altitude, where the sunlight redirec-
riphery. Such concentration of light is responsible tion system is expected to be 100% efficient accord-
Figure 9
Test 3: different lens configura-
tions.

Figure 10
Adaptive daylight in the
interior.
ing to the Grasshopper/Galapagos models. The sys- the sun’s azimuth alignment until reaching the wall.
tem was placed over the fluid lens at the top of the In the other direction it was possible to reach 90%
box and the mirrors were rotated as such, to direct of the space (0,5m*0,5m) with the spot. The Test 2
the received light perpendicularly to the ground. configuration in combination with the redirection
device did not fit under the sun simulator. However
Test 1 and 3 observations and comparison it should be noted that, the higher the ceiling is lo-
Although the sunlight redirection system manages cated above ground, the more the range and perfor-
to redirect the light perpendicularly to the ground, mance is increased in terms of light redirection. Due
the system in combination with the lens do not suc- to the height of the redirection device itself the area
ceed in bundling all the rays in one focal point but in of illumination is more reduced the further the light
fact a linear series of focal points is noticed. This de- beam is astray from the vertical redirection configu-
viation is caused by imperfections of the mechanical ration. Furthermore it will also be interesting to see
system controlling the rotation of the mirrors. It can the interaction and lighting design possibilities of
be derived that even minor deviations of the mir- several devices together.
rors from the correct inclination can direct the light
in a non-desired direction as well as light scattering General findings regarding the fluid lens
is further enhanced if the film is not evenly applied under clear sky conditions
over the ABS lamellas. The imperfections at the ro- The light and heat absorption by the water volume
tation mechanism are also responsible for the pres- is another issue worth to be discussed as during
ence of more accentuated shadows on the floor cast clear sky conditions, the lux value on the floor sur-
by the row of mirrors. Minor shadows are of course face under the opening is reduced in both Test se-
expected at the neutral and diverging modes of the ries 1 and 2 by 31% when the lens at its neutral state
lens due to the thickness of the mirrors but not to is placed over the opening. The Beer-Lambert Law
the observed extent. explains the logarithmic relationship between the
transmission of light through a substance, the thick-
Light redirecting possibilities ness of the medium and the wavelength of the light,
The prototype showed that initial expectations in proving that the intensity of light decreases expo-
terms of light redirection capabilities of the systems nentially with the increase of the water depth (Ryer,
were by far exceeded (Figure 10). In the Test 3 con- 1997). Taking into account that the light absorption
figuration it was possible to redirect light within coefficient of water for violet light (380nm wave-

length) is 0,00011cm-1 and for red light (725,5nm a typical flux density at ground level on a cloudy
wavelength) is 0,01678 cm-1 (Pope et al, 1997) and winter day of 4000lux (Flesch, 2006) and relate it to
by applying the Beer-Lambert law for a water depth the 962lux measured outside next to the box. This
of 1,4cm (water depth at the neutral state of the results in 220lux at the center of the floor area and
prototype), it can be concluded that 0,035% of vio- 191lux at the periphery when only the fluid lens is
let light and 5,26% of red light will be absorbed by present, and in 116lux and 79lux respectively when
the water volume. Considering the full scale lens, the sunlight redirection system is added.
the occurring light absorption will increase due to
the 10 times higher water depth. Calculations show CONCLUSION
that 0,35% of violet light and 41,77% of red light will
be absorbed by the water volume. Moreover, due to Assessment of the combined system
the selective color absorption property of water, the The physical experimentations prove the ability of
light exerting the lens will have a slight blue hue. the system to quickly adapt not only in order to con-
In addition, the light absorption coefficient of verge/diverge the sunrays but also to redirect the
water for wavelengths of 1.000nm to 1.000.000nm incident light according to the needs of the interior
ranges from 0,339 cm-1 to 128,2 cm-1 meaning that space over almost the whole floor surface. The sys-
water is strongly absorbing infrared light (Zolotarev tem successfully proved to be able to function as a
et al, 1969). Of the radiant energy emitted from the daylight spotlight which can adaptively react to the
Sun, approximately 50% lies in the infrared region moving sun position as well as interior lighting re-
(Fu, 2003). This energy is to be perceived as thermal quirements. The changes between different modes
energy, absorption of the infrared light leads to the occur gradually rather than abruptly and thus are
reduction of the amount of heat entering the room not disturbing to the user.
from the lens. Due to the high specific heat index of It is however concerning the high contrast ratio
water, the lens is expected to act as a thermal mass observed in some of the tests and the probability of
that absorbs, stores and releases heat where the glare needs to be evaluated. Also the illumination of
penetration of heat is delayed. the interior when more than one adaptive sunlight
redirecting modules are installed in proximity has to
Cloudy sky be determined. The performance under worst case,
Test series no.4, conducted in the Artificial Sky Sim- cloudy sky conditions appears to be sub optimal
ulator, study the performance of the system in the and this would either determine a certain amount of
worst case scenario, that of a cloudy winter day. For lenses required on the roof or would suggest an ap-
these tests the 0,5*0,5*0,35m box was used first with plication at a location with a high amount of sunny
the fluid lens alone at its top opening (Test 4A) and hours. By applying the system at the more southern
then with the sunlight redirection system placed hemisphere location it would not only be more ef-
over (Test 4B). According to the findings, the incom- fective in terms of direct sunlight but the building
ing light is evenly distributed rather than diverged would also benefit in terms of heat absorption and
or converged. reduced heat gain. In general the system is useful
When comparing Test 4A and 4B, we conclude for interiors which require directed light and are
that the sunlight redirection system is reducing the not affected by contrasts. This could be suitable for
amount of incoming light by 47,8% at the area un- atria, large markets, shopping malls or restaurants
der the opening (flux density is reduced from 53lux to place dynamic daylight accents and highlight
to 28lux) and by 57,5% at the periphery of the space certain spots. If a more even and diffuse light dis-
(from 46lux it becomes 19lux). To scale up the meas- tribution like in train stations, etc. is required an ad-
urements to real overcast conditions, we consider ditional adaptive light diffusor has to be thought of.

Simulation versus prototypes REFERENCES
With the current possibility to easily and cheaply Bennett, JM 1995, ‘Polarization’ in M Bass, EWV Stryland, DR
manufacture functional prototypes and the ongo- Williams, WL Wolfe (eds), Handbook of Optics, 2nd edi-
ing tendency of more, relatively cheap, user friendly tion, McGraw-Hill Inc., Volume I, Part 2, Chapter 5.
but none the less fairly accurate 3-D printing tech- Flesch, P 2006, Light and Light Sources High-Intensity Dis-
nologies like FDM printers being released on the charge Lamps, Springer-Verlag Berlin Heidelberg, The
market there is a great chance that we might face a Netherlands.
renaissance of physical testing rather than simula- Fu, Q 2003, Radiation (Solar), Elsevier Science Ltd., Univer-
tion only approaches. sity of Washington, Seattle, WA, USA.
Green, M, Allen MJ, Weintraub, L and Abrams, BS 2008, Fo-
OUTLOOK rensic Vision With Application to Highway Safety, Law-
To finalize the research the measured results will be yers & Judges Publishing Company, Inc. Tuscon.
fed back into the digital environment to close the Jacobs, A 2012, Radiance Cookbook, Internet Version, p. 76.
circle but also to have a more complete design tool. Lide, RD (ed) 2009, ‘Index of Refraction of Inorganic Liquids’
In more practical terms sensing and actuation and in CRC Handbook of Chemistry and Physics, 89th Edition
the digital interface should be thought of as well (Internet version), CRC Press/Taylor and Francis, Boca
as more material research for an actual application Raton, FL, p. 764.
has to be done. The requirements for a high degree Pope, RM and Fry, ES 1997, “Absorption spectrum (380-
in precision in redirecting sunlight needs to be fur- 700nm) of pure water. II. Integrating cavity measure-
ther considered. The extreme converging mode of ments”, Applied Optics, 36, pp. 8710-8723.
the lens and bundling the light into one spot might Ryer, AD 1997, The Light Measurement Handbook, Interna-
not be applicable for lighting the interior but could tional Right, Newburyport, MA.
be interesting in combination with photovoltaic as Taylor, AEF 2000, ‘Basic concepts in Optics’ in Illumination
a concentrator system while the space underneath Fundamentals, Lighting Research Center, Rensselaer
is less frequented or occupied by the inhabitants. Polytechnic Institute, pp. 10-12.
The photovoltaic elements could be mounted as a Zolotarev, VM, Mikhilov, BA, Alperovich, LL and Popov, SI
relatively small device close to the ceiling and would 1969, “Dispersion and absorption of liquid water in the
therefore not disturb the flow of light otherwise. It infrared and radio regions of the spectrum”, Optics and
would be also possible to employ a more expensive Spectroscopy, 27, pp. 430-432.
but highly efficient photovoltaic cell since the sur-
face area is small while being more protected from [1] http://www.grasshopper3d.com/group/galapagos
any outdoor influences like dust and rain. Next steps [2] http://www.solemma.net/
will also include research in smart material applica- [3] http://www.vray.com/
tion together with the Functional Devices research [4] http://www.ise.fraunhofer.de
group of the Department of Chemical Engineering [5] http://www.photolux-luminance.com/
and Chemistry at the TU Eindhoven in order to in-
vestigate the possibility to develop a similar func-
tioning system without any moving parts involved.

Modelling and Simulating Use Processes in Buildings
Davide Simeone1, Yehuda E. Kalay2, Davide Schaumann3, Seung Wan Hong4
1
Sapienza University of Rome, Italy, 2,3,4Technion Israel Institute of Technology, Israel
1
http://www.dicea.uniroma1.it, 2,3,4https://architecture.technion.ac.il
1
davide.simeone@uniroma1.it, 2kalay@technion.ac.il, 3deiv@technion.ac.il, 4seungwan@
berkeley.edu
Abstract. In this paper, we propose a new approach to simulating users’ behavior in

yet un-built buildings. For this purpose, we have developed a model that connects two
different representations: a formal representation of the building use, by means of a
method borrowed from Business Process Modeling and Notation (BPMN) approach; and
a game-engine based 3D virtual environment, where this process is effectively simulated
and integrated with some autonomous behaviour of users/agents. The model has been
applied to two test cases, where the activities of doctors, nurses, patients, and visitors in
different hospitals were tracked, simulated, and reviewed by medical professionals for
validation.
Keywords. Building use simulation; human behaviour modelling; BPMN; activity-based
modelling; building performances prediction.
INTRODUCTION/RATIONALE
During the design process, architects are asked to product. Specifically, basing design decisions on a
predict and evaluate future building performances set of averaged parameters, in the assumption that
related to a large number of functional, typology- the building will satisfy future users’ needs (Zimmer-
based and organization-based requirements. To man, 2003) much like “similar” buildings have done
support their design decisions, architects usually so in the past, often fails when real users, who may
rely on functional programs which are interpreta- differ from the “average” user in many ways, finally
tions of the requirements of the organization that meet the building.
will occupy and use the building, namely: the main Architects’ ability to predict in which manner
activities of future building’s occupants. their design will be used, and whether it will match
In the past, this interpretation process has been the activities of its intended users, is currently only
mainly supported by normative methods, regula- supported by the architects’ own expertise and
tions and general design rules. Nevertheless, the imagination. Sadly, the consequences are clearly
domination of normative approaches has shown its recognizable in reality: too often buildings do not
limits in light of increasing complexity of building perform as expected after their construction, and
design and typology, and the intrinsic complexity sometimes they completely fail to support the ac-
of human - building interaction (Koutamanis and tivities of the organizations that will occupy them.
Mitossi, 1996). As matter of fact, its high level of ab- The observation and analysis of human behav-
straction is not well-suited to the intrinsic unique- ior in built environments is usually considered the
ness and context-dependence of an architectural best way to understand and evaluate how a building

fits the needs and the activities of its intended users. model (Tabak et al., 2008; Goldstein et al., 2011). This
On this basis, the Post Occupancy Evaluation (POE) approach has been worthily tested in buildings such
paradigm has proposed several approaches and as hospitals, offices and airports, where the organi-
techniques to assess if the project brief has been zational workflows and the related interactions with
met (Preiser, 1988). POE approaches have, of course, the built environment actually drive and heavily in-
one major limitation: they can be applied only after fluence users’ behavior.
the building has been realized and occupied, and at Still, some criticisms has been raised of this ap-
that point it is usually too late or too costly to inter- proach in terms of its ability to realistically predict
vene in order to solve errors, critical failures, and in- human behavior in architectural design, since it
consistencies with the needs of users. relies on a rigid, ‘functionalistic’ representation of
In order to overcome this deficiency in the de- operational processes of the organization, usually
sign process, we chose to investigate how to use completely computed before the actual simulation
“virtuality” to actually integrate building occupancy and not adaptable to single users’ behaviors and to
evaluation into the design process, allowing design- the overall status of the built environment (what we
ers to test their decisions before actually entering call ‘serendipitous’ or ‘emergent’ activities). This dis-
into the construction phase. In particular, the pro- tortion inevitably reduces the ability of these simu-
posed model focuses on simulation of activities in lative approaches to predict building response to
the built environment, in order to predict how the users’ needs and activities, and rely instead on archi-
building will match the functional needs of the or- tects’ imagination and expertise to actually guess in
ganization that will occupy it. what ways their design will perform after the build-
ing will have been occupied.
RELATED WORK In the last few years, Agent-Based Modeling
Since the inception of Computer Aided Architec- approaches have been introduced in this research
tural Design, several attempts have been made to field, aiming at simulating users’ behavior in built en-
introduce the expected users’ activities in building vironments by developing a series of autonomous
representation models (Eastman and Siabiris, 1995; entities - the agents - each of whom interacting in
Carrara, Kalay and Novembri 1986; Ekholm and an autonomous way with the other users and with
Fridqvist 1996). In such models, however, activities the environment surrounding it (Macal and North,
have been explicitly represented in terms of their 2007). Although Agent-Based Modeling has been
spatial features - usually relying on the concept of successfully applied to simulation of some behav-
“functional unit” - or implicitly inferred by using sets ioral phenomena generated by individual actors/
of functional requirements as criteria for the evalua- agents (such fire-egress and pedestrian movement),
tion of the capabilities of a space (Archer, 1966). it has shown its limitations in simulating agents’ co-
Gradually, research attention in this field has operation and collaborative activities performing.
turned from a “space-based” representation of us-
ers’ activities to a “process-based” representation, AIM OF THE PROJECT
considering activities as entities on their own that The work presented in this paper aims at deve-
are clearly distinct from (but connected with) spa- loping a different model to simulate users’ behavior
tial entities (Wurzer, 2010). This new approaches is in buildings, in which the building use representa-
based on the idea of modelling processes depend- tion is still based on a process-driven system, but it is
ing on the operational workflows of the organiza- more adaptable both in terms of its activities struc-
tion - or of an organization typology if the specific ture and of users’ individual decisions and actions. In
data are not available - that will occupy the building, order to provide these capabilities, we intervene at
and then simulate their execution in the building two different levels of the model: in the formaliza-

tion of the building use process, which we define as by architects, planners and clients, the outcome is a
use scenario, and in the simulation system of the us- simulation/visualization in a 3D virtual environment
ers’ behaviour derived by the scenario. of how the use process is actually carried out by the
A building use process has a direct correspond- building users in the building spaces prefigured by
ence to the way the occupant organization works in the architect. In this way, it is possible to predict and
terms of operational workflows, procedures and sys- evaluate the correspondence and the mutual influ-
tems of activities (Ekholm, 2001). Based on this as- ence between the building and its intended users,
sumption, we chose to rely on a modeling approach and rapidly compare the simulation outcomes of dif-
-the Building Process Modeling and Notation- al- ferent design solutions and spatial configurations.
ready developed to represent how an organization To test and calibrate the model, it has been imple-
operates and to extend it to representing the use mented to simulate the functioning of two different
process of a building. hospital wards, comparing its output with the real
The BPMN level, where the use process is formal- users’ behavior.
ized, is connected to a 3D simulation environment (a
game engine in our case), where the same process REPRESENTING BUILDING USE PRO-
is effectively computed, simulated and visualized at CESSES BY MEANS OF BPMN
the same time. In this environment, Users/Agents As asserted by Ekholm (2001), we can look at an or-
are provided with the abilities to autonomously ganization as a system with relations among its parts
adapt their behavior within a predefined range, de- whose functioning is actually a process (a sequence
pending on the status of the environment model of events) derived from performing a series of activi-
and on the reference process model. In turn, the ties. In accordance with the purpose of our research,
simulated users’ serendipitous actions are fed back we chose to extend the concept of ‘system’ from the
into the process model, and can influence it. For ex- simple organization to the sum of building spaces,
ample, in the case of a hospital, when a doctor and activities and actors. The Business Process Modeling
a nurse are scheduled to check on a patient, but the and Notation (BPMN) approach allows representa-
patient has chosen this particular moment to visit tion of operational processes of an organization in
the bathroom, the absences of the patient is fed- order to orchestrate the activities and the decisions
back to the process model, which defines a different of the different actors involved (White, 2006; Lam,
flow of activities for the doctor and the nurse. Like- 2012). It provides a representational system that,
wise, if the paths taken by two agents brings them different from previous approaches, provides at the
into geometric proximity, due to the geometry of same time a formalization schema for processes
the building, they may choose to stop and chat, or and explicit semantics for its execution/simulation.
ignore each other and continue on their predefined In particular, it is able to describe different aspects
missions. of actors’ interaction in an operational process (or-
The novelty of this approach lies in making chestration, collaboration, choreography, decision
the process execution more flexible and partially points), a feature we consider relevant for the pur-
adaptable to serendipitous “emergent” activities pose of our research, since it allows formalization of
in real time during the simulation, while in current cooperation among different building users during
approaches the activities flow is usually compiled the performing of an activity.
and fixed before stepping into the effective simula- The BPMN formalization is based on a set of ele-
tion (Tabak 2008). In addition, the proposed model mentary entities that can be used to decompose and
can represent and simulate collaborative, planned represent an operational process, the main ones are:
activities, such as cooperation among various users • Activities: representation of tasks, works, or
when performing their tasks. In terms of usability operations that have to be carried out or ex-

Figure 1
A building use process (in this
case a visiting routine in a
hospital ward) represented
by means of BPMN approach.
Actors and spaces entities are
connected to the activities
sequence, providing a for-
malization of What (which
activity) is performed, Where
and by Whom.
ecuted during the process; of our modeling approach. The formalization of

• Connectors: links to connect an activity to an- spaces effectively provides the conceptual connec-
other activity in order to define an operational tion between the building use process based on the
sequence flow. Other classes of connectors al- organization’s operational dynamics, and the build-
low to associate other kinds of entities to activi- ing design solution provided by the architect. In
ties; this way, the activities, considered elementary units
• Events: occurrences that “happen” during the of the organization’s operational workflow, are not
process, starting, delaying, interrupting or end- abstract anymore, but explicitly represented in the
ing a flow of activities; building model, providing a representation of what
• Gateways: modeling elements that control the is going to happen, where, and by whom. Although
pathways of the process, its diversions and its the BPMN representation already takes into account
convergences, allowing parallel or exclusive actors’ declaration by means of “swimlanes” (White,
paths. 2006), we chose to develop a specific artifact for the
For the purpose of our research, these classes actors, since in a building use process formalization
of entities (with their subclasses) are relevant, but each actor has to be associated to several activities
not sufficient. So, in order to make the BPMN system and this is hard to represent with the swimlanes sys-
able to represent building use processes, we chose tem (Figure1).
to rely on the ability to extend the BPMN represen- The BPMN approach allows us to represent
tational approach by creating two new classes of not only complex sequences of activities, but also
artifacts: their articulation in (and relation to) time: specific
• Actors: entities representing each actor in- time-triggers or event-triggers can represent con-
volved in the building use process, in order to ditions for an activity to be activated, interrupted
connect it to the performing of the activities or deactivated, influencing the performing of the
he/she has to carry out; building use scenario. Gateways are used to for-
• Spaces: entities representing the spaces of the malize and control parallel or exclusive executions
built environment, and necessary in the BPMN of multiple activities; they can be considered deci-
environment in order to effectively connect the sion points in the flow of activities, since they allow
use process to the building, and allow its simu- testing the model status for specific conditions and
lation in the 3D environment. choosing which sequence of activities to perform.
We consider the addition of spaces and actors For instance, if we imagine a scenario where a doc-
entities to the process representation a key point tor is visiting a series of patients (as the one shown

Figure 2
An encapsulated set of activi-
ties (on the left) and an ad-hoc
process (on the right) used
in the building use process
representation. Their role is
to make more manageable
the representation of complex
processes and not-structured
activities.
in Figure 1), a gateway formalizes the necessity of to effectively represent a building use process. In
checking the patient presence and, in case of his/her fact, it is not able to take into account and conse-
absence, it adapts the use scenario by directing the quently simulate how the use process (meant as a
doctor to the next patient. set of activities and actors involved) is influenced by
The BPMN ability to encapsulate activities in the built environment, and how it will be actually be
sub-processes also allows us to manage complex carried out in it. In order to provide architects and
processes and to reuse the same activities struc- clients with a reliable prediction of how the building
ture several times. At the same time, non-structured users carry out the defined activities, we chose to
or intermediate activities (such as “using the re- integrate the BPMN representation with a 3D simu-
strooms”, or “having a walk”) (Tabak, 2008), are rep- lation environment, where the formalized use pro-
resented by means of ad-hoc sub-processes that can cess is effectively simulated within the built environ-
be invoked during the actual simulation according ment provided by the architect. In this environment
to probabilistic curves (Figure 2). (developed by means of the game engine Virtools
In order to make the process representation [2]), the building use process, previously formalized
more flexible and adaptable to different systems in an abstract way in the BPMN system, is connected
(meant as building + activities + users), we also used to the virtual model of the built environment where
BPMN messages and signals to stop and restart dif- its activities are supposed to be performed.
ferent sub-processes depending on specific condi- To compute and simulate the use scenario deve-
tions or events. The BPMN system allows us to actu- loped in the BPMN model, a specific script has been
ally export (via XML) and execute the represented developed in the Virtools game engine by means of
building use process in external simulation environ- behavioral blocks -visual programming blocks that
ments and to use it as input for such systems. For correspond to the different activities represented by
the development of the building use scenario in the them. In Virtools’ scripting environment, we chose
BPMN environment, we chose to use Bizagi, a free- to develop a specific programming level for the
ware business process modelling software [1]. formalization and computation of the use scenario;
its role being to guide and control the execution of
SIMULATING BUILDING USE PROCESSES the sequences of activities, adapting their perform-
IN 3D VIRTUAL ENVIRONMENT ing to the environment and to the status of the us-
The BPMN approach is a valid way to represent and ers’ involved. It also enables control and simulation
simulate processes composed of one or more flows of serendipitous events, triggered by the physical
of activities involving some actors. Nevertheless, the (actually, geometrical) proximity and location of the
BPMN representational system alone is not sufficient actors within the simulated built environment (Fig-

Figure 3
A part of the building use
scenario script developed in
the Virtools scripting environ-
ment.
ures 3 and 4). Such chance encounters may trigger agent-based components, intended to control some
different performance paths. In addition, we chose autonomous aspects of virtual users’ behavior (for
to equip activity entities with specific scripts to sim- instance, path decision, walking actions, obstacles
ulate their performing in order to coordinate actors’ avoidance, local interactions with other entities,
actions and cooperation. This is a fundamental dif- such as doors or other agents). This choice has two
ference from previous agent-based models, where main advantages: resemblance to the visual reality
the activities simulation is generated by the sum of of the resulting simulated phenomenon, and im-
autonomous actions and decisions of the users, with proving the manageability of the computation sys-
several limits in terms of manageability and coher- tem. The first consists of the possibility to reduce the
ence of the output. rigidity of a process-driven simulation by including
To improve the adaptation of the scenario sim- variations related to single actors’ behaviors, actions
ulation to the built environment and its status, we and decisions. In that way, we can simulate seren-
chose to integrate the scenario script with some dipitous events generated by the interactions of
Figure 4
The simulation of the build-
ing use process in the 3D
visualization environment of
Virtools.

the agents with the built environment that are not activities related to this operation in the BPNM
predictable in the scenario development. The ability environment and simulating it in the virtual en-
to provide actors with some degrees of autonomy vironment was not a difficult task, because of the
allows us to represent some aspects of users’ be- structured nature of the procedures involved in
havior that would be difficult and time-consuming scheduled surgeries. Simulating the activities per-
to represent and compute at the process level. For formed in the intensive care unit required extensive
instance, the abilities of a user to compute a path in observations, long meetings with the medical staff,
the built environment and perform the movement and still we were only able to reproduce them com-
actions can be easily developed and controlled di- putationally within a high degree of abstraction.
rectly in the agent entity, while their representation During the experiment, two different configurations
and computation at the process level would be very of the physical environment provided by the archi-
difficult, and if iterated for each agent and activity tect have been tested in relation with the same use
would make the process representation too com- scenario, in order to support the design team and
plex and difficult to manage. the hospital managers in the evaluation of their
functional quality (Figure 5).
A CASE STUDY In order to validate the simulation output, some
Operational efficiency in hospitals is strongly influ- medical specialists have been interviewed to verify
enced by the physical design of the built environ- the validity of the formalized use process and the
ment. Although hospitals are relatively complex reliability of the simulation results. The use process
buildings, their use-pattern is relatively straight-for- model proposed in this paper proved to be highly re-
ward, which is advantageous for our research since liable in situations when a clear sequence of observ-
it provides a comprehensive and agreed-upon data able activities can be recognized by lightly trained
set against which the model can be tested. For this observers. When the inherent complexity of the situ-
purpose, as a first implementation of this model, we ation produces phenomena hard to decompose in
chose to simulate the functioning of different hospi- activity chunks, the model shows some limitations.
tal wards, in routine and in discrete emergency cas- Future developments will involve exploring
es. The large quantity of money that is being invest- different way of decomposing inherently complex
ed in healthcare facilities suggests that enhancing, situations in sequence of chunks computationally
even by little, the efficiency of routine procedures manageable.
might lead to substantial savings in time and costs.
As a first implementation of our approach we CONCLUSIONS
chose as target the cardiology department of the By integrating a building use process formalization
Bnei Zion medical center in Haifa, Israel. The size of with its visual simulation in a virtual environment,
the department and the complexity of the activities the proposed model offers architects and clients
are appropriate to test our simulation model. We the opportunity to test the functionality of a design
chose to observe and simulate both a routine proce- solution and to foresee its consequence on users’
dure - a coronary catheterization - and a more com- behaviour, before actually being constructed and
plicated scenario, such as the intensive care unit, occupied. The building use process simulation ap-
whose emergent phenomena are harder to predict. proach allows architects, clients and process plan-
Several coronarography events have been ob- ners to easily formalize a use scenario in terms of
served during a period of a week, after a series of the activities performed, the actors involved and
meetings held with the principal physician, who the spaces where such activities will take place. The
explained to the observer the list of procedures visual/geometric simulation provides the necessary
performed during the operation. Reproducing the connection of the abstract scenario to its perform-

Figure 5
A screenshot from the simula-
tion of one of the proposed
configurations for the Cardiol-
ogy department of the Bnei
Zion Medical Center [3].
ing in a defined physical environment, thereby in- tional procedures, and to test different configu-
troducing the environmental constraints that affect ration of human resources such as number of
the process and contribute the important element workers, their profile and specialization, their
of serendipity. The 3D visualization of how such use scheduling.
process is effectively performed by future building So far, the research shown in this paper has
users, is helpful in making the results accessible to mainly focused on simulation of users’ behavior
the experts who must judge the outcomes of the in terms of activities performing and operational
simulation. management. It would be interesting in follow-up
Differently from previous activity-based models research to introduce social and environmental psy-
where the use process is entirely computed before chology data in the simulation model, in order to
and then merely visualized, in the proposed model provide a more comprehensive and reliable predic-
the use scenario is computed in real time during the tion of users’ life and activities in buildings.
simulation, providing a better adaptation of the se-
quence of activities to the built environment and its REFERENCES
occupants and, consequently, a more coherent and Archer, B 1966, Activity Data Method: a method for record-
reliable simulation output. ing user requirements, Ministry of Public Buildings and
By providing a real time simulation of users’ Works, London.
behavior within a defined physical environment, al- Carrara, G, Kalay, YE and Novembri, G 1986, ‘KAAD - Knowl-
though limited to specific use cases and processes, edge-Based Assistance for Architectural Design’ ,
this simulation model would support: Teaching and Research Experience with CAAD - Proceed-
• Architects and clients in evaluating the func- ings of4th eCAADe Conference, Rome, Italy, pp. 202-212.
tional performances of a design solution be- Cohen, U, Allison, D, and Witte, J 2010, Critical Issues in
fore it is actually built, leaving them the possi- Healthcare Environments, Research Report for the Cent-
bility to correct errors and solve critical points; er for Heath Design, Concord CA.
• Process planners, analysts and building manag- Eastman, CM and Siabiris, A 1995, ‘A generic building prod-
ers in testing different workflows and opera- uct model incorporating building type information’,

Automation in Construction, 4(4), pp. 283–304. ter Simulation Conference.
Ekholm, A 2001, ‘Modelling of User Activities in Building Preiser, WFE, Rabinowitz, HZ and White, ET 1988, Post Oc-
Design’, Architectural Information Management - 19th cupancy Evaluation, New York, Van Nostrand Reinhold.
eCAADe Conference Proceedings, Helsinki, Finland, pp. Simeone, D and Kalay, YE 2012, ‘An Event-Based Model to
67-72. simulate human behaviour in built environments’, Digi-
Ekholm, A and Fridqvist, S 1996, ‘Modelling of user organi- tal Physicality: Proceedings of the 30th eCAADe Confer-
zations, buildings and spaces for the design process’, ence, Prague, Czech Republic, pp. 525-532.
Construction on the information highway. CIB Proceed- Tabak, V 2008, User Simulation of Space Utilisation – System
ings Publ. 198, CIB W78 workshop, Bled, Slovenia. for Office Building Usage Simulation, PhD Thesis, Eind-
Goldstein, R, Tessier, A and Khan, A 2011, ‘Space Layout in hoven University of Technology, Eindhoven, Nether-
Occupant Behavior Simulation’, Conference Proceedings lands.
of the IBPSA-AIRAH Building Simulation Conference, Syd- White, SA 2006, Introduction to BPMN, IBM Corporation.
ney, Australia, pp. 1073-1080. Wurzer, G 2010, ‘Schematic Systems – Constraining Func-
Koutamanis A, Mitossi, V 1996, ‘Simulation for Analysis: Re- tions Through Processes (and Vice Versa)’, International
quirements from Architectural Design’, Proceedings 6th Journal of Architectural Computing 08 (02), pp. 197-214.
EFA - European Full-scale modeling Association - Confer- Zimmermann, G 2003, ‘Modeling the building as a system’,
ence, Vienna, Austria, pp. 96-101. Eighth International IBPSA Conference Proceedings, Ein-
Lam, VSW 2012, ‘A Precise Execution Semantics for BPMN’, dhoven, Netherlands, pp. 1483,1490.
IAENG International Journal of Computer Science, 39(1)
pp. 20–33. [1] http://www.bizagi.com.
Macal, C and North, M 2007, ‘Agent-based modelling and [2] http://www.3ds.com/products/3dvia/3dvia-virtools.
simulation: desktop ABMS’, Proceedings of the 2007 win- [3] http://www.youtube.com/watch?v=3OKgipQ6Pbs.

Flexing Wind
Aerodynamic study of architectural windbreak

Rafael Moya1, Flora Salim2, Mani Williams3, Kamil Sharaidin4
RMIT University, Melbourne, Australia
1
http://www.bitacoravirtual.cl, 2http://www.florasalim.com
1
s3296513@student.rmit.edu.au, 2flora.salim@rmit.edu.au, 3mani.williams@rmit.edu.au
4
kamilsharaidin@gmail.com
Abstract. The aims of the Flexing Wind project, investigated in an intensive

cross-disciplinary course, were twofold. First was to learn about aerodynamic phenomena
around buildings. Second was to explore ways to observe, measure, and control the
negative effects of wind around specific pedestrian areas, tram stops, and public sites in
Melbourne City. Using tools such as a weather station to collect data and CFD software
to simulate aerodynamic phenomena students could study the wind conditions in one
of the windiest areas in the Melbourne downtown. Various do-it-yourself tools such as
mini wind tunnels, handheld probes and sensors were used to evaluate the performance
of potential design options, which lead to prototyping full scale adaptive architectural
windbreaks.
Keywords. Urban aerodynamics; windbreak; wind tunnel simulation; Computational
Fluid Dynamics; architectural prototype.
INTRODUCTION
A better understanding of urban aerodynamics will sues (Gandemer, 1981).
positively influence design decisions in architectural This project was conducted as an intensive
and urban projects. The wind flow and dispersion three-week cross-disciplinary elective in the School
through a city determine environmental air quality, of Architecture and Design, RMIT University, offered
wind pressures on buildings, urban heat islands, pe- to architecture, landscape architecture, and engi-
destrian comfort, and ambient noise level in the sur- neering students. The outcomes of the explorations
rounding environment (Boris, 2005; Zaki et al., 2010). include:
However, only a few existing techniques have been • Wind maps of the sites (a major intersection
developed to deal with the habitability and comfort located at the northern axis of Melbourne
issues due to strong wind conditions on pedestrian and the alleyway at the rear-entry of the RMIT
areas (Cochran, 2004). These are mainly done on the University Design Hub building), derived from
urban planning level or by introducing trees and data captured using handheld probes and a
shrubs as vernacular shelterbelts. Studies have been weather station
done on how aerodynamic characteristics of wind- • Analysis and evaluation of the performance
breaks can be used to resolve pedestrian comfort is- of a series of windbreak options designed for

each particular site performed using a small- and the mini wind tunnel (Salim and Moya, 2012)
scale wind tunnel test and Computational Fluid and methods of designing kinetic façade prototypes
Dynamics simulation. (Sharaidin et al., 2012). Digital technologies such
• Three prototypes as a system of mitigation for as Computational Fluid Dynamics (CFD) were also
conflictive wind environments (deflector and used to simulate the aerodynamics of the external
diffuser devices). Two of these prototypes are environment. Additionally, using a mini wind tun-
reported in this paper. nel, physical experiments such as erosion test (wind
The main task was to investigate the design and flow visualization using particles on a dark-coloured
performance of architectural windbreaks as design background) and smoke test (Salim and Moya, 2012)
interventions on the prevailing wind conditions (di- were applied to evaluate the different designs of the
rections, pressure, speed) by controlling eddy areas windbreak systems. Through these methodology
around a building. Eddy is a turbulent wind condi- and tools the students could run different experi-
tion caused by the changes of wind pressures [1]. ments focused on the analysis of wind phenomena.
The key outcomes were design prototypes of adap- The proposed methodological experiments
tive architectural windbreaks, which were installed were divided into several tasks:
on public footpaths. Feedbacks were gathered 1. Build a system to quantify and visualise wind
around issues such as the performance of the pro- data (wind speed). The tools made available
posed windbreaks and the impact of installing the to the students were: a low-cost commercial
windbreak on the windy sites. weather station, scientific indoor condition
measuring probes, low-cost electronic sensors
METHODOLOGY and development platform (Arduino). Through
The research and design aspects of the elective pro- a selection of the above tools, the students will
jects was conducted by a group of students (11 mas- develop a system to collect and visualise the
ters and undergraduate students) and highly guided data from the public space. The intention is
by investigators (the teaching staff in the elective, to develop a wind map of the urban zone us-
authors of the paper). At the beginning, the stu- ing instruments to measure the movement of
dents had access to a wide range of literature on the wind.
theory of windbreak design. Topics included studies 2. Construct both a physical and digital module
about aerodynamics in urban contexts and wind of the selected site. Conduct simulations to
comfort using Beaufort scale criteria where wind is understand the effect of wind in an interactive
considered as a function of speed and sensations fashion.
felt (Gandemer, 1978); and systems to wind control 3. Design an artificial windbreak prototype for
such as windbreaks (Cochran. 2004). The students public space (passive or kinetic structures).
were also presented with introductory lectures on Students were encouraged to test their design
the technical aspects of the project by the teaching iteratively through both physical and digital
staff as well as an external industrial expert. During simulations.
this stage the students learned about causes and 4. Build and install a full scale windbreak proto-
effects of more common aerodynamic phenomena type. Students built a representative part of
in cities produced by the wind flow interacting with their design in a scale one by one. This proto-
buildings and affecting public areas. type was installed in a windy site in the city to
In the second stage, students were introduced to evaluate impact on public space and perfor-
the methodology and tools on windbreak design. mance to mitigate wind problems.
The literature included the authors’ earlier research 5. Evaluate the wind mitigation achieve after of
projects, such as the parallel wind analysis method the prototype installation.

Figure 1
From the left to right: the wind
rose from Vasari and wind
map derived from the site
observation.
PROJECTS: ANALYSIS, DESIGN AND sage leading to the entrance changes its direction
OUTCOMES throughout the year. The effect on the entrance is
The two main approaches to develop a wind control the same but the design of the windbreak should
structure were the exploration of porous patterns as deal with both wind directions.
a wind filter and the concept of a shell as a wind de- Using Vasari as CFD software it was possible to
flector. The challenge was to not only design a struc- visualise how the gust of wind had a curved move-
ture to control the negative effect of wind detected ment producing two separation zones with low
in the site, but also produce low impacts in the sur- pressure areas. One of these areas coincided with
rounding space. This meant that the aesthetics is an the entrance of the building, producing an input
important element alongside the functional aspect of wind when the gate was opened. These simula-
of the windbreak. tions were validated with measurements of the
wind speed using the anemometer on the low-cost
Milk.Crate.Break Project (by Tamara Cher, weather station. On the day of data collection it was
Xuan Son Nguyen, Romy Peterfreund) found that the wind coming closed to the façade
The site for prototype 1 Milk.Crate.Break is at the had an average speed of 3.7m/s at 2m height, but
rear entry of the RMIT Design Hub (Swanston Street this velocity increased up to 4.4m/s when was blow-
entrance). It is a narrow alleyway close to an alcove ing through the passage (Figure 2).
with an operable door opening. Through a study of The first approach was to use a shell or canopy
the historic data of the site it was found that there to deflect the wind to maintain the low pressure
is a prevailing wind blowing through the site, direc- area in front of the entrance. Later the design pro-
tion alternating depending on the season (Figure cess progressed to the design of a skin with some
1). When the door is opened the sudden pressure kind of porosity to control the wind speed. This po-
difference produce a noticeable wind gust into the rosity concept evolved from a surface with simple
building. patterns of holes to different patterns with a variable
The study of the site (Figure 1) was focused on density of porosity (Figure 3), following the indica-
the investigation of the differences of wind conditions of the studies by Gandemer (1981) on wind-
tions between summer and winter seasons, as well breaks.
as analysing the impact in this area. As part of the project the students were required
The first finding of this study showed that wind to construct a section of their design in full size and
has two predominant directions (north and south). install it on site. The Milk.Crate.Break team decided
This meant that wind passing through the pas- on the front section of the design to be constructed

Figure 2
From the left to right: CFD
visualisation and windmap of
the site done in Vasari.
at the full scale. This had the implication that a new more efficiently to produce an upward air through
structural system is required for the design. Through the wall. The experiments compared these differ-
further exploration the students decided on using ent patterns with 20%, 40% and 60% of porosity.
milk crates as the main building material, factors in- For each case, the pattern moves depending on the
fluenced their decisions included: structural (modu- pressure of the wind, where used to deflect the wind
lar self-supporting, easy to assemble and disassem- upwards as well as absorbs a fraction of the wind en-
ble), sourcing (free and readily available), a cultural ergy (Figure 4 and 5).
significance (Melbourne’s laneway). The inherent The final concept was a structure with a dou-
structure and porosity of the milk crates also offered ble layer of porous skin. The first porosity layer was
the students additional design possibilities: The a regular graph design which ameliorated the wind
design was adapted to take advantage of the crate speed. The second shark-skin pattern layer reduced
volume to produce a design with a double skin. The the wind flow close to zero. Between both layers, the
students explored several options of patterns for an internal chamber in the structure was designed to
adaptable second skin through CFD analysis. The deflect the wind vertically. In this way the pressure
first option had a pattern of Venturi funnels work- on the structure was reduced to maintain the struc-
ing as a diffuser to decrease the wind speed in the tural stability.
outcome side of the wall. The second version was a The students’ final design was a windbreak that
pattern of triangular petals with a more simple sys- spans the full width of the alleyway, 2.5 meters in
tem of petals aperture for the density control. The height, offering protection for a large area in the
final design was a pattern of triangle flaps based on proximity of the door. The design form is symmetri-
the shark skin. This was the options chosen because cal to work with both wind directions. There is one
the parallel triangular surfaces deflected the wind opening at each end to allow access through the
Figure 3
From the left to right: the first
version of shell, the second
version of a porous shell, the
study of porosity density.

Figure 4
From the left to right: Venturi
funnels, petals and shark skin.
site and into the building. The porosity system is de- Lyrebird Project (by Mikhail Kochev, Sara
signed for the surface of the windbreak, with lower Metanios, Daisy Leung, Rico Shuyuan
density at the lower part the design to offer more Zhang)
protection (Figure 6). The site of prototype 2 Lyrebird is located close to a
Further testing (both simulation and in the street level entrance to RMIT University Building 14
physical mini wind tunnel) were conducted to con- (Swanston Street). The onsite data measurements
firm the functionality of the adapted design. Meas- revealed a predominant wind direction from the
urements were taken to evaluate the design. south to north. The students noticed a strong and
Figure 5
From the left to right: porosity
variations and effect of wind
passing through the structure,
final installation.

Figure 6
From the left to right: design
process from the concept to
the installation and measure-
ments in the site.
turbulent wind close to the building facades side tralian Lyrebird. The idea was to mimic the natural
of the pedestrian sidewalk. Although a number of curvature of the feather to form a curved shell for
trees are present on the street near the site, it was the entrance (Figure 8). Design iteration of the form
evident that the wind conditions were not improved and dimension of the windbreak was mainly con-
for pedestrians. Erosion test of the site model was ducted through physical wind tunnel tests and CFD
conducted in the mini wind tunnel to reproduce experiments. These digital tests were focused to find
the more relevant phenomena. The phenomena ob- the more efficient curvature for this roof to deflect
served were: the wind blowing along the street, the the wind. This exploration found that a double cur-
wind effect around the corner on the building, and vature performed a very good protection rather a
the effect that could be produced by placing struc- single curve (Figure 9).
tures over the entrance (Figure 7). These simulations In the wind tunnel test, a simple curved cano-
demonstrated a channel effect occurring in this area py shows a good performance to deflect the wind
with a high level of turbulence from the friction with without significant loan on the structure: the arched
the buildings’ walls. This was identified as an issue shape was able to deflect and guide the wind over
for the building entrances and other points around the building opening. This final version was chosen
the pedestrian circulation routes. The design objec- as the structure shown to be a more effective wind-
tive was to protect the street level entrance from the break. These tests demonstrated that the structure
prevailing and local wind conditions. does not produce lateral strong gusts and the pro-
The student drew their design inspiration from tection area was large enough to provide shelter
bird feathers, in particular the fan-like tail of the Aus- against strong winds around the entrance of the
Figure 7
From the left to right: mini
wind tunnel and erosion test
of the proposed windbreak
design, wind map of the site.

Figure 8
From the left to right: concept
based on tail of the Australian
Lyrebird and hypothesis of
wind deflection.
building (Figure 10). flap with functions similar to prototype 1. As these

Detailed porosity exploration was also carried panels are more visually prominent compare with
out using smoke test to study the best sequence of prototype 1, the students worked to adapted them
gaps and thickness of the barrier with different wind as a visual wind indicator to increase public inter-
velocities (Figure 11). action with the windbreak. This was done through
The final design was a frame and infill system installing an Arduino controlled LED display, where
with triangular “slot-in” panels on vertical structural the display was driving by wind data from an elec-
frames. Each triangular panel contains a movable tronic sensor (anemometer). This was not installed
Figure 9
From the left to right: CFD
tests for different curvatures,
first project design and wind
test of curve profiles.
Figure 10
From the left to right: CFD test
and wind tunnel test for the
final design.

Figure 11
From the left to right: smoke
test, axonometric of panels
and final render.
on the final installation due to time constraints. tion of this project: to teach students to work with
Many factors such as site permit restriction and fab- methods and technological tools to study complex
rication constraints dictated that only approximate- phenomena.
ly 2 meters by 2 meters by 2 meters of the design This first part of the objective was fully complete
was constructed and evaluated on site. The students for the students. In the short three weeks, through
applied bright paint to observe how colours have the study of the literature available in the field of
a relevance to intensify the visual aesthetic of the wind engineering and the tools such as Vasari CFD
structure in the public area (Figure 12). and a low-tech mini wind tunnel, the students were
able to gain a good understanding of the basic
GENERAL EVALUATION aerodynamic effects to begin their own design ex-
One positive outcome of this academic project was ploration in this field. The potential of these tools for
the opportunity to share knowledge from different pedagogic purposes was evident. The visual interac-
fields that conduct work on the city and its current tive feedback helped the students to gasp with the
problems. The use of technology helps us to under- comprehension of these complex phenomena, and
stand the dynamic phenomena like wind in cities, as a platform for discussion the design performance.
through collecting data (both in the physical and The visual documentation of the design testing pro-
virtual realm, on site and through simulations) and cess (both still images and videos) formed a large
make sense of that information. This was the inten- part of the presentation material for the students
Figure 12
From the left to right: proto-
type, Arduino platform and
installation.

to communicate their design intent to the teaching The task of constructing a physical prototype
staff and invited critiques. and installing it on site helped students to under-
The ready-to-go low-cost commercially avail- stand how an urban intervention can also have a
able weather station, with its simple interface, was visual impact in the space. For instance, the colour in
quick to be adapted by the students to be used for the structure of Lyrebird project was considered as a
site analysis and to prototype performance evalua- parameter of communication. Additionally, the pro-
tion tool. ject also considered to use an anemometer driven
Arduino Platform and electronic sensors have LED display which could be activated with the wind,
proven to require a learning curve too steep for most sending a live visual signal to the passers-by when
of the students to be able to utilise them in their the wind speed increases over the comfortable lev-
design. The Arduino and sensors induction were els. Thus, the performative aesthetics of a windbreak
provided as part of the course material, but whether as an urban element can also have a functional as-
to incorporate it within the design was a choice left pect that informs people of the surrounding envi-
for the individual student team. All of the students ronmental conditions.
welcomed the half day hands-on session explor-
ing Arduino sensors and motors. The students were CONCLUSION
guided through a series of selected examples that Wind around buildings and in public spaces can pro-
introduced the concept of Arduino microcontroller duce negative effects that are necessary to mitigate.
and a collection of electronics. With each example This paper reports on a design-construct cycle of
its possible application to the windbreak design was a site specific architectural windbreak, conducted
discussed. Following the demonstration, a propor- in the form of a design studio taken place over an
tion of students actively sought additional equip- intensive three-week period at the RMIT University,
ment and assistance to experiment it further. This Melbourne Australia. After a detailed study of the
demonstrated that new technology was easily taken site, the students successfully incorporated aerody-
up students when it is presented as a useful resource namic theory into their design thinking and demon-
to extend the possibilities for their design proposal. strated the use of CFD simulation tools and physical
Apart of the design aspect of the windbreak, wind testing to assist their design process. The im-
the challenge of a built prototype was a very posi- pact of the windbreak on the site was quantitatively
tive learning process because students were encour- measured and evaluated. This is done by assimilat-
aged to deal with physical problems and technical ing the studies of local wind conditions and vernac-
solutions. Even if the prototype was not a fully func- ular systems and testing the design in wind tunnel
tional model, many issues concerning scale, materi- simulations.
als, and cost were considered and evaluated for each The results and prototype designs are prelimi-
project. This dialog of material and constructability nary, however they demonstrated the possibilities
also opened up new design exploration. Take the in- of designing windbreaks that have aesthetics fea-
tention of the Milk.Crate.Break project to use recycle tures as well as the functional capacities to provide
plastic boxes as an interesting example. The decision comfortable pedestrian areas.
to use milk crates as a building material led the team
to explore the features of a double-layer windbreak. ACKNOWLEDGMENTS
This project studied the performance of different The authors acknowledge the support from SIAL
porous patterns when a simple plastic mesh may be team to acquire the equipment used in this course.
functionally sufficient. This demonstrated that per- The authors also wish to thank Jon Morgan (Arup)
formance and constructability should inform each for his technical contributions and critiques of the
other in design. projects. Lastly the authors would like to thank the

participants of the “Flexing Wind” elective course for Eng Ind Aerod, 7(1), pp. 15-36.
the fruitful works over the intensive 3 weeks: Tamara Salim, F and Castro, M 2012, ‘Parallel Analysis of Urban
Cher, Xuan Son Nguyen, Romy Peterfreund, Mikhail Aerodynamic Phenomena Using High and Low-tech
Kochev, Sara Metanios, Daisy Leung, Rico Shuyuan tools’ in 30th International Conference on Education
Zhang, Marta Sophianti Singgih, Lazuardy Laisuhan- and research in Computer Aided Architectural Design in
ta, Vera Raquel dos Santos, Zin Mee Zin Win. Europe (eCAADe 2012), vol. 1, Prague, Czech Republic,
pp. 621-629.
REFERENCES Sharaidin, K, Burry, J and Salim, F 2012, ‘Integration of Digi-
Boris, J. P. 2005, ‘Dust in the wind: Challenges for urban tal Simulation Tools With Parametric Designs to Evalu-
aerodynamics’, 35th AIAA Fluid Dynamics Conference ate Kinetic Façades for Daylight Performance’, Digital
and Exhibit, Toronto, Ontario Canada, June 6 - 9, AIAA Physicality - Proceedings of the 30th eCAADe Conference
Paper 2005-5953. - Vol 2, Czech Technical University in Prague, Faculty of
Cochran, L 2004, ‘Design Features to Change and/or Ame- Architecture (Czech Republic) 12-14 September 2012,
liorate Pedestrian Wind Conditions’ in Structures Con- pp. 701-709.
gress 2004, American Society of Civil Engineers, Nash- Zaki, S. A., Hagishima, A. and Tanimoto, J. 2010, ‘Estimation
ville, Tennessee, United States, pp. 1-8. of Aerodynamic Parameters of Urban Building Arrays
Gandemer, J., et al. (1978). Discomfort due to wind near with Random Geometries using Wind Tunnel Experi-
buildings: aerodynamic concepts. Washington, D.C., U.S. ment’, Technical Papers of Annual Meeting of IBPSA-Ja-
Dept. of Commerce, National Bureau of Standards. pan.
Gandemer, J 1981, ‘The aerodynamic characteristics of
windbreaks, resulting in empirical design rules’, J Wind [1] en.wikipedia.org/wiki/Eddy_(fluid_dynamics)

Real-time Environmental Feedback at the Early Design
Stages
Creating a conceptual analysis tool by teaching artificial neural networks

with design inputs and monitored energy consumption data
Greig Paterson1, Sung Min Hong2, Dejan Mumovic3, Judit Kimpian4

1,3
Bartlett School of Graduate Studies, UCL, UK, 2UCL Energy Institute, UK, 4Aedas
R&D, UK
1,3
http://www.bartlett.ucl.ac.uk/graduate, 2http://www.bartlett.ucl.ac.uk/energy, 4http://
www.aedas.com/
1
greig.paterson.10@ucl.ac.uk, 2s.hong@ucl.ac.uk, 3d.mumovic@ucl.ac.uk, 4judit.kim-
pian@aedas.com
Abstract. It has been argued that traditional building simulation methods can be a slow
process, which often fails to integrate into the decision making process of non-technical
designers, such as architects, at the early design stages. Furthermore, studies have
shown that predicted energy consumption of buildings during design is often lower than
monitored energy consumption during operation.
In view of this, this paper outlines research to create a user friendly design tool that
predicts energy consumption in real-time as early design and briefing parameters are
altered interactively. As a test case, the research focuses on school design in England.
Artificial neural networks (ANNs) were trained to predict the energy consumption of
school designs by linking actual heating and electrical energy consumption data from the
existing building stock to a range of design and briefing parameters.
Keywords. Environmental design tool; energy prediction; artificial neural networks;
building operational performance; schools.
INTRODUCTION
There are many environmental ‘design aids’ avail- • Physical modelling
able, with the objective of helping designers make • Building simulation
sustainable design decisions. These design aids can Given that environmental design problems tend
largely be grouped into the following categories to be ‘wicked’ (Rittel and Webber, 1973), and thus
(Morbitzer, 2003): distinctly novel and unique, rules of thumb, basic
• Design guidelines / rules of thumb calculations and correlation methods are often inad-
• Steady state calculation methods equate techniques (Morbitzer, 2003; Pratt and Bos-
• Correlation based methods worth, 2011) and physical modelling has the disad-

vantage of being very costly (Morbitzer, 2003). When In light of the above, two questions emerge:
used correctly, the most powerful design aid avail- 1. Can an ANN based method for a design tool be
able for the analysis of environmental performance developed that offers non-technical users the
is building simulation (Morbitzer, 2003). Building ability to predict energy consumption in real-
simulation is, however, rarely used by architects at time as they explore the design space?
the early design stages (Pratt and Bosworth, 2011). 2. Can such a tool be based on actual energy con-
Architect and psychologist Lawson (2004; 2006) sumption, rather than simulated data, in an ac-
states that simulation tools are not ‘design’ tools but curate manner?
‘evaluation’ tools which are used to assess designs As a test case, the research focuses on school
after they have been designed. A major barrier is design in England. The purpose of this paper is
the time taken to input all the required information, firstly to summarise the data collection process and
such that the designer can only afford to do it after describe the ANN method. Finally, the tool user in-
the major design decisions have been made (Law- terface development and preliminary results will be
son, 2004). Also, the design space is constrained by presented.
the fact that commonly used building simulation
tools produce static design proposals - it is there- DATA COLLECTION
fore difficult, given time and economic constraints, The data collection process was a desktop study
to produce a wide range of design options (Pratt with the aim of collecting as much design and brief-
and Bosworth, 2011). In this way, the design space ing data as is freely available on hundreds of schools
is sparingly populated because the models are dis- across England. Table 1 and Table 2 outline the input
crete rather than continuous, thus omitting ‘in- be- and output parameters for the ANN models. The en-
tween’ solutions (Pratt and Bosworth, 2011). ergy data used to train the ANNs were sourced from
Furthermore, research, such as that carried out the Display Energy Certificate (DEC) database, which
by CarbonBuzz [1], highlight the fact that the actual are stored in the non-domestic energy performance
energy consumption of buildings regularly exceeds register maintained by Landmark [2]. The annual
the design estimates, often by more than double. electricity and heating fuel use (kWh/m²/annum)
Real world problems have complex and non- figures were used as the output in this study. The fol-
linear interactions, therefore system behavior is of- lowing criteria were used to select the school build-
ten best learned through observations rather than ings for analysis, ensuring the buildings are compa-
modelling (Samarasinghe, 2007). In view of this, an rable with each other:
alternative approach at predicting energy consump- • The school has a valid DEC
tion in buildings is to collect large amounts of actual • The school has one main building
energy and design data and analyse the patterns be- • Age of construction and material use are con-
tween the two. One such source of actual ‘observed’ sistent
energy data in the UK are Display Energy Certificates Data on 465 schools have thus far been collected.
(DECs) (CIBSE, 2009). One method of learning the In addition to energy consumption, other data
relationships between energy consumption and collected from the DEC database were total useful
design inputs are artificial neural networks (ANNs). floor area (m²) and building environmental condi-
ANNs are machine learning techniques inspired tioning type. The number of pupils in each school
by the structure and processes of biological neural was gathered from the Department for Education’s
networks that take place within the brain (Haykin, (UK) EduBase public portal [3] and heating and cool-
1999). ANNs were found to be suitable for assessing degree days were acquired from the Central
ing determinants of energy use in higher education Information Point [4]. The geometric and site data
buildings in London, UK (Hawkins et al., 2012). were gathered by measurement or visual inspec-

Table 1 Input Input Data Range / Description
ANN inputs. Parameter Neuron Type Activation Criteria
Construction Continuous 1860-2010 Year the school was built
Year
Phase of (-1) Primary/elementary, Primary schools or secondary schools/sixth
Education Binary (1) secondary/high form colleges
school
Number of Continuous 44-2013 Part-time pupils divided by 2, plus the
Pupils number of full-time pupils
Internal (-1) Nat. vent, (0) mixed Primary internal environmental
Environmental Categorical mode, (1) mech. vent conditioning strategy
Conditioning
‘Exposed’: no obstructions present (4 x the
Site Exposure Categorical (-1) Exposed, (0) semi- height of the school away); ‘semi-exposed’:
sheltered, (1) sheltered obsts. lower than the school; ‘sheltered’:
obsts. taller than the school.
Angle at which the external walls differ
Orientation Continuous -45° - +45° from absolute north, south, east and west.
Positive angle for clockwise orientations.
North Façade Obstructed if a building or tree is within
Adjacency Binary (-1) Open, (1) obstructed 1 x the height of the building from the
majority of the façade orientation
South Façade Binary (-1) Open, (1) obstructed See North Façade Adjacency
Adjacency
East Façade Binary (-1) Open, (1) obstructed See North Façade Adjacency
Adjacency
West Façade Binary (-1) Open, (1) obstructed See North Façade Adjacency
Adjacency
Floor Area Continuous 861m²-15396m² Total usable floor area
Building Continuous 2.50-16.60 Building volume / exposed external wall
Depth Ratio area
Compactness Perimeter of the building footprint /
Ratio Continuous 1.01-4.59 perimeter of a circle with the same area as
the building footprint
Surface Continuous 1.71-5.67 Building volume / exposed surface area
Exposure Ratio
North Glazing Continuous 0.00-0.13 Glazed area on the north façade / total
Ratio floor area
South Glazing Continuous 0.00-0.15 Glazed area on the south façade / total
Ratio floor area
East Glazing Continuous 0.00-0.11 Glazed area on the east façade / total floor
Ratio area
West Glazing Continuous 0.00-0.14 Glazed area on the west façade / total floor
Ratio area

Input Input Data Range / Description Table 1 continued
Parameter Neuron Type Activation Criteria ANN inputs.
Glazing Type Binary (-1) Single, (1) double Single or double/secondary glazing
Roof Shape Binary (-1) Pitched, (1) flat Pitched or flat roof
Roof Glazing Binary (-1) None, (1) glazing Existence of any roof glazing
Heating Continuous 1635.6-2843.3 Heating degree days during the DEC
Degree Days monitoring period
Cooling Continuous 73.9.7-425.2 Cooling degree days during the DEC
Degree Days monitoring period
Output Output Data Range Description Table 2

Neuron Type ANN outputs.
Heating Energy Continuous 7-272kWh/m²/annum Annual heating fuel use
Consumption
Electricity Energy Continuous 7-95kWh/m²/annum Annual electricity fuel use
Consumption
tion from the online map software Digimap [5], Bing tion outputs (Table 2) based on a set of inputs (Table
Maps [6] and Google Earth [7]. 1). A multilayer perceptron network was used for the
The building height was derived by multiplying study - Figure 1 shows the conceptual structure of
the average number of storeys by 3.62m - the aver- this ANN. The hidden layer enables the system to
age floor-floor height of schools in the UK (Stead- generate nonlinear and complex relationships by
man et al., 2000). The building volume was then intervening between the input and output neurons
derived by multiplying the building height with the (Haykin, 1999). Each neuron in the input and output
building footprint area, measured from Digimap [5]. layer took continuous, categorical or binary values
Glazing percentages were measured from Bing Map as outlined in Table 1 and Table 2. Prior to the train-
[6] images using bespoke code developed in the ing of the network, all continuous inputs were nor-
Processing programming environment [8]. malised to values between -1 and 1 to generalise
The construction year of the buildings were col- the calculation process. Two ANN models were con-
lected from each school’s website where available structed, one with heating energy consumption as
otherwise they were derived from historical digital an output and one with electrical energy consump-
map software [5]. Data on schools of varying ages tion as an output - both ANN models included all of
were collected to increase the size of the database, the input parameters (Table 1).
giving the neural network more data to learn from. A Levenberg-Marquardt backpropagation su-
A proportion of the differences in, for example, fab- pervised training technique was used to train the
ric quality and building systems between newer feedforward network to recognise the patterns that
schools and older schools are likely to be picked up exist in the dataset. The prediction performance of
in the construction year neuron. Therefore, this neu- the ANN was assessed by validating the ANN with
ron will exist within the trained network in the final 10% of the gathered database on which the ANN
design tool but fixed to the most recent date. had not been trained - the testing dataset. 10% of
the gathered database was used to stop the train-
ANN ARCHITECTURE ing process before overlearning occurred (Demuth
All ANNs were constructed in Matlab [9]. The aim of et al., 2008) and the remaining 80% of the database
the ANN method is to predict the energy consump- was used to train the network. The number of neu-

Figure 1
Conceptual Structure of the
ANN.
rons in the hidden layer were altered between 2, 4, algorithms are integrated into this environment
8, 16 and 32 neurons. Each network configuration with MATLAB Builder JA [10].
was trained five hundred times and the ANN with The tool allows the user to sketch the footprint
the lowest mean squared error (1) was selected for of the building by clicking and dropping vertices
further analysis. Further analysis consisted of calcu- in an input window - these vertices can later be
lating the coefficient of determination (R²) and the dragged or deleted. All other inputs are entered via
below performance indicators, (2) and (3): sliders (continuous inputs) and tick boxes (categori-
cal/binary inputs) thereby encouraging the user to
Mean squared error (MSE) = (same units ‘play’ and test different options, encouraging explo-
as output) (1) ration of ‘in-between’ solutions in the design space.
The ability to gain feedback in real-time results in
Root-mean squared error (RMSE) = the user being able to ‘animate’ the results and learn
(same unit as output) (2) the relationships between the design inputs and
energy outputs by the acceleration of change in the
results as the design space is explored.
Mean absolute percentage error (MAPE) =
(%) (3) RESULTS AND DISCUSSION
Where Yi and Ŷi are the target and predicted outputs ANN configurations with two and eight neurons in
respectively for the training, testing or stopping the hidden layer were found to produce the least
configuration i and n is the total number of configu- prediction errors for heating and electricity energy
rations in the training, testing or stopping datasets. consumption respectively. Table 3 summarise the
results of the errors for the best performing ANN
USER INTERFACE configurations. The electricity output was predicted
Figure 2 shows a representation of the tool user in- with a mean absolute percentage error (MAPE) of
terface. The tool is currently being developed in the 19.3%, while the heating output was predicted with
Processing programming environment [8]. The ANN a MAPE of 20.5%. These errors are an improvement

Figure 2
Representation of the user
interface.
of 10.0% and 6.7% for heating and electricity energy research in this paper better these errors by 4.6%
consumption respectively, when compared to using and 15.5% for heating and electricity respectively.
the Chartered Institution of Building Services Engi- Figure 3 show scatter plots of the ANN predic-
neers (CIBSE) Technical Memorandum 46 (TM46) En- tions vs actual annual heating and electricity energy
ergy Benchmarks as energy performance indicators consumption from the testing dataset. The coeffi-
(Table 4). As mentioned in the introduction, Hawkins cient of determination (R²) shows that the 23 design
et al. (2012) used an ANN method to assess the en- and briefing parameters (ANN inputs) explain 39%
ergy determinants in higher education buildings in and 41% of the variation in annual heating and elec-
London, UK. The ANN method by Hawkins et al. pro- tricity energy consumption of the schools respec-
duced MAPEs of 25.1% and 34.8% for heating and tively.
electricity fuel use respectively - the results from the From this initial study it appears that the ANN Table 3
Prediction errors of the ANNs
ANN Output RMSE (kWh/m²/annum) MAPE (%) - calculated from the ANN
Heating Energy Consumption 30.5 20.5 testing dataset.
Electricity Energy Consumption 10.8 19.3
Table 4
TM46 Benchmark RMSE (kWh/m²/annum) MAPE (%) Prediction errors of the CIBSE
Heating Energy Consumption 41.3 30.5 TM46 Benchmarks - calculated
Electricity Energy Consumption 16.7 26.0 from the ANN testing dataset.

Figure 3
Scatter plots of predicted
ANN vs actual heating (left)
and electricity (right) energy
consumption.
method is viable for predicting energy consumption errors, as outlined in the previous section. Neverthe-
in existing school buildings. Nevertheless, further less, it is desirable to reduce these errors further and
research is planned to improve the performance of improve the R² values. In order to improve both the
this method and ensure it is viable for new school performance of the ANN method and increase the
designs as outlined in the further work section. relevance of the tool, further design inputs are likely
to be required. The nature of this desktop study was
CONCLUSION to collect as many design and briefing inputs as are
This paper outlines research to create a user friendly freely available. Acquiring further inputs, such as
design tool that predicts energy consumption in building services and fabric data, may require direct
real-time as early design and briefing parameters communication with individual schools or local au-
are altered interactively. As a test case, the research thorities. This process is likely to be time consuming
focused on school design in England. Artificial neu- however is being pursued. Further actions to im-
ral networks (ANNs) were trained to predict the en- prove the ANN performance, as well as ensuring the
ergy consumption of school designs by linking ac- tool is relevant to the design process and applicable
tual heating and electrical energy consumption data to new school designs, are outlined in the following
from the existing building stock to a range of design section.
and briefing parameters. The initial design of the It should be noted that the development of this
user interface was introduced in this paper. tool does not have the objective of replacing tradi-
For the energy consumption predictions, the tional building simulation - instead it aims to act as a
ANN mean absolute percentage error (MAPE) was user friendly sanity check for non-technical design-
20.5% for heating and 19.3% for electricity. The co- ers, such as architects, at the early design stages.
efficient of determination (R²) was 39% and 41% for
heating and electricity energy consumption respec- FURTHER WORK
tively. The aforementioned errors were compared There are a number of developments underway in
with another method and study and produced lower order to make the method of prediction in this re-

search more accurate and the use of the design tool Haykin, S, 1999, Neural Networks: A Comprehensive Founda-
more relevant. These developments include collect- tion, Prentice Hall, NJ, USA.
ing data on more schools across England; the pur- Lawson, B, 2004, What Designers Know, Architectural Press,
suit of additional input parameters, as mentioned in Oxford, UK.
the previous section; refining the input parameters Lawson, B, 2006, How Designers Think: The Design Process
for heating and electricity energy predictions sepa- Demystified, 4th Edition, Architectural Press, Oxford,
rately; and the exploration of alternative ANN archi- UK.
tectures. The final tool will go through a validation Morbitzer, C, 2003, Towards the Integration of Simulation
process using a number of new schools as case stud- into the Design Process, PhD Thesis, University of Strath-
ies to ensure the method is applicable to new school clyde, Glasgow, UK.
designs. Finally, as the user interface develops, it will Pratt, KB, Bosworth, DE, 2011, ‘A Method for the Design and
be tested by focus groups within industry. Analysis of Parametric Building Energy Models’, Pro-
ceedings of Building Simulation 2011: 12th Conference of
ACKNOWLEDGEMENTS International Building Performance Simulation Associa-
This research was undertaken as part of an Engi- tion, Sydney, Australia, pp. 2499-2506.
neering Doctorate (EngD) project at the UCL EngD Rittel, HW, Webber, MM, 1973, ‘Dilemmas in a General Theo-
Centre in Virtual Environments, Imaging and Visu- ry of Planning’, Policy Sciences, 4, pp. 155-69.
alisation (VEIV). The work is sponsored by the En- Samarasinghe, S, 2007, Neural Networks for Applied Sciences
gineering and Physical Sciences Research Council and Engineering: From Fundamentals to Complex Pat-
(EPSRC) and Aedas, with the R&D team within Aedas tern Recognition, Auerbach Publications, NY, USA.
providing much support and collaboration. Steadman, P, Bruhns, HR, Rickaby, PA, 2000, ‘An Introduction
to the national Non-Domestic Building Stock Data-
REFERENCES base’, Environment and Planning B: Planning and Design,
CIBSE, 2008, TM46 Energy Benchmarks, Chartered Institution 27(1), pp. 3-10.
of Building Services Engineers Publications, London,
UK. [1] http://www.carbonbuzz.org/
CIBSE, 2009, CIBSE TM47 Operational Ratings and Display En- [2] https://www.ndepcregister.com/
ergy Certificates, Chartered Institution of Building Ser- [3] http://www.education.gov.uk/edubase/home.xhtml
vices Engineers Publications, London, UK. [4] http://www.landmark.co.uk/solutions/registers/nondo-
Demuth, H, Beale, M, Hagan, M, 2009, Neural Network Tool- mestic/cip
box: User’s Guide, Mathworks, MA, USA. [5] http://digimap.edina.ac.uk/
Hawkins, D, Hong, SM, Raslan, R, Mumovic, D, Hanna, S, [6] http://www.bing.com/maps/
2012, ‘Determinants of Energy Use in UK Higher Edu- [7] http://www.google.com/earth/index.html/
cation Buildings Using Statistical and Artificial Neural [8] http://processing.org/
Network Methods’, International Journal of Sustainable [9] http://www.mathworks.co.uk/
Built Environment, 1(1), pp. 50-63. [10] http://www.mathworks.co.uk/products/javabuilder/

DesignScript: Scalable Tools for Design Computation
Robert Aish
Autodesk, UK
http://labs.autodesk.com/utilities/designscript/, http://www.designscript.org/
Robert.Aish@Autodesk.com
Abstract. Design computation based on data flow graph diagramming is a

well-established technique. The intention of DesignScript is to recognise this type of
data flow modeling as a form of ‘associative’ programming and to combine this with the
more conventional ‘imperative’ form of programming into a single unified computational
design application. The use of this application is intended to range from very simple
graph based exploratory ‘proto-programming’ as used by novice end-user programmers
to multi-disciplinary design optimisation as used by more experienced computational
designers.
Keywords. Graph; scripting; associative; imperative.
INTRODUCTION
The development of DesignScript is intended to be computational design tools easier to use by non-
a response to the following trends in contemporary experts. In addition, there is an assumption or indeed
design practice (Figure 1). These trends can be de- an expectation that once designers have successfully
scribed in terms of the following dimensions. accomplished some initial tasks, they will be interest-
ed in progressing from an exploratory approach to
Scalable to different Computational skills more formal programming methods. Therefore, we
It is generally recognised that there are advantages need to step back from the immediate requirement
in making design computation more accessible to a (to make computational design tools more accessi-
wider audience of designers. As software develop- ble) and consider a more general requirement which
ers, we can interpret this not just as the need to make is to support a progression in the development of
Figure 1
The three dimensions describ-
ing critical aspects of contem-
porary design practice.

computational skills, from ‘no skills’ to novice skills tion that supports different levels of skills is that it
and hence to more proficient and expert skills. can encourage a more effective collaboration be-
This idea of a ‘progression in computational tween team members with different skill levels.
skills’ is based on the generally recognised critique
of graph diagramming methods. While graph dia- Scalable from abstract to domain specific,
gramming is an extremely powerful technique to including the support for multi-discipli-
enable novice programmers to create their first nary design integration
computational models with the minimum of expe- While much of existing design practice is based on
rience and skill, it is generally recognised that the different disciplines (architecture, structural engi-
graph node representation does not scale to more neering and environmental performance, etc.) there
complex logic. Indeed the visual complexity of the are generally recognised advantages if a more ho-
graph may become overwhelming and counter pro- listic approach is adopted that integrates different
ductive. It is exactly at this point where the applica- ways of design thinking. This approach to design
tion should encourage the novice programmer to is often exercised through the formation of multi-
make the transition from graph node diagramming disciplinary design teams and can be encouraged
to scripting: literally from ‘node to code’ (Figures 2-4). by software that combines multiple analytical and
Initially such scripts will reflect the graph node simulation methods (Figure 5).
style of exploratory programing but as the designer Another characteristic of innovative architecture
continues to enhance and extend these scripts, the practice is a ‘return to first principles’ often demon-
applications should help him to progress to a more strated through architectural form or engineering
formal style of software engineering by providing innovation. Often these principles may not have
such facilities as a ‘type’ system and tools to support been previously associated with architecture or
the refactoring of scripts into functions and classes. supported by existing design applications. This sug-
An interesting by-product of a design applica- gests a two way progression:
Figure 2
Starting with a conventional
graph node diagram…

Figure 3
The designer select a region of
the graph and use the mark-
ing menu…
Figure 4
To turn the “nodes into
code”…

Figure 5
An example of a multi-disci-
plinary design model created
with DesignScript, using a mix
of simulation and analytical
techniques provided by the
different available plug-in’s,
including: form finding [with
SmartForm], structural analy-
sis [with Robot] and insolation
on the shading devices [using
Performative Design]. The col-
our coding on the structural
members indicates structural
utilization, while the colour
coding on the shading devices
represents insolation. Note:
the structural nodes on the
periphery are fixed, the ones in
the centre are free to find their
correct position.
On the one hand there are advantages in having In some cases the ‘concept’ is the overall form
design applications support more domain specific of the building, but increasingly the design concept
functionality: not in isolation, but rather integrated may be an engineering principle or an objective to
into a single common application framework and achieve a particular combination of performance
capable of acting as a computational intermediary metrics, or the use of a particular generative algo-
between the different members of a multi-discipli- rithm distributed within the individual components.
nary design team. In these latter approaches to the design, the archi-
On the other hand there are advantages if de- tectural ‘form’ may emerge as a consequence of the
sign applications could step back from being too design process rather than imposed ‘top-down’. In
domain specific and support a ‘return to first princi- addition, the resulting design representation may
ples’ by exposing both computational and geomet- not be a traditional ‘building model’ primarily in-
ric abstractions directly to the designer. tended to support conventional drawing extraction.
Rather it may be a series of ‘geometric normalisa-
Scalable to projects of different size and tions’ intended to be the minimum information re-
complexity quired for a direct digital fabrication process or ro-
Design concepts often start as disarmingly elegant botic construction. Indeed conventional workflows
and simple ideas which can easily be explored with (or data flows) are being supplemented by innova-
lightweight models or scripts. But to be realised as tive project specific processes. Because of the essen-
a physical building, these ideas necessarily have to tial ‘open-endedness’ of this new form of design, it is
be developed into complex building models with increasingly important that design applications are
hundreds of thousands of individually detailed com- similarly ‘open-ended’.
ponents.

Summary design computation. The ‘node to code’ transition
In summary, DesignScript is a computational system process is an extremely powerful way of using the
which is intended: to support the progressive acqui- result of an initial ‘graph modeling’ as the starting
sition of computational skills, to encourage the in- point for scripting, but there are still some chal-
tegration of different design disciplines and to sup- lenges facing the designer to acquire the underly-
port projects of different size and complexity. ing programing concepts if he is to harness more of
Whatever changes occur it is important that the the potential of scripting. The following discussion
computational application can accommodate these might serve as an introduction to these issues and
changes and maintain a common underlying repre- extends the previous work (Aish, 2011; 2013).
sentation of design logic with no loss of fidelity, ca- Within a domain specific application, such as
pability or performance. DesignScript, we need to distinguish between the
functionality related to particular ‘domain-specific’
DESIGNSCRIPT AS A DOMAIN-SPECIFIC objects such as walls, windows, columns and beams
LANGUAGE which typically would be implemented in specialised
Graph node diagraming is now a well established application libraries, and more general functionality
technique in design computation. The intention of that would be implemented at the language level. In
a graph node diagramming user interface is to pro- this context a domain-specific language implements
vide the designer with an intuitive and easily used many of the facilities found in general purpose pro-
‘proto-programming’ tool that requires little or no gramming languages. In addition, a domaim specific
previous understanding of programming concepts. language takes more general concepts from the ap-
Many of the existing applications which support plication domain and promotes these concepts to
a graph node style of computational design have be ‘first class’ features of the language (Table 1).
made a separation between this type of interactive DesignScript could be described as a hybrid
dependency modeling and regular scripting using language which implements familiar concepts (and
conventional imperative languages such as Python, syntax) found in imperative, functional and object
C# and Java. oriented languages and combines these with a
Effectively, graph node diagramming is a form number of new innovations in the form of an asso-
of ‘data flow’ programming which has developed ciative language.
independently of conventional programming lan- The essential hybrid ‘associative-imperative’ na-
guages. The intention of DesignScript is to explicitly ture of DesignScript is illustrated in the way it com-
recognise graph-based dependency modelling as bines two domain specific ideas. First, the idea of a
a form of ‘associative’ programming. We can define building being composed of a series of dependent
an associative programming language as one which collections of components where some members
represents data flow in a human readable text no- have special conditions’ (typically supported by
tation. DesignScript supports the display and inter- graph node diagramming and its representation as
action with the underlying dependencies via both an associative language). Second, the idea of ‘design-
graph node diagramming and a corresponding text ing’ as an exploratory activity which involves iterative
based associative language. refinement (as supported by iteration and condition-
Indeed, while the designer is using graph node al logic found in conventional imperative languages).
diagramming he may not be aware of that the graph
is being recorded in DesignScript. But as the design- The differences between Associative and
er progresses and wishes to explore more complex Imperative programming
modeling tasks he will necessarily need to acquire Both Associative and Imperative languages have ex-
a more formal understanding of programming and ecutable statements, for example:

Table 1
DesignScript as a ‘domain-
specific’ languages imple-
ments many of the facilities
found in general purpose
programming languages
and adapts these for a
domain-specific purpose and
in addition takes more general
concepts from the application
domain and promotes these
a = 10; all combined. These are discussed in the following concepts to be ‘first class’
b = a * 2; sections. features of the language.
.
. Replication
a = 20; In many existing programming languages a distinc-
In the case of an imperative language, the final tion is made between a single variable of a particular
statement a = 20; does not cause the previous state- type and an array or collection of that same underly-
ment b = a * 2; to be re-executed. ing type. This distinction restricts the interchange-
In the case of an associative language, the state- able use of a collection or a single value of the same
ment b = a * 2; is not just an executable statement, it underlying type and forces the programmer to write
is also records a persistent dependency relationship different code for a single variable and for arrays or
between variables b and a, such that any change to collections.
a will automatically force a re-execution of all state- One of the domain-specific aspects of Design-
ments where a is referenced on the right hand side, Script is to relax this restriction and make the lan-
as in b = a * 2;. guage more flexible and more tuned to its use by
Imperative programming uses special ‘flow con- novice programmers. In this context, DesignScript
trol’ statements, such as for and while loops and if .. introduces the concept of ‘replication’. With replica-
else statements. These flow control statements are tion, anywhere a single value is expected a collec-
independent of the executable statements. In the tion may be used instead and the execution of the
absence of any flow control statements imperative statement containing the collection is automatically
programming executes statements in the sequence executed for each member in that collection.
of the source code (i.e. in the lexical order). Asso- The combined result of dependencies and rep-
ciative programming does not have separate flow lication is that it is easy to program complex data
control statements but instead uses the concept of flows (including geometric operations) involving
dependencies inherent within each statement (such collections. An upstream variable may change from
as to b = a * 2; ) to create a topological ordering of all being a single value to a collection or from a collec-
the statements. The statements are then executed in tion to another collection of different dimensions or
this topological order. size. As a consequence, the downstream depend-
In addition, associative programming within ent variables will automatically follow suit and also
DesignScript also introduces two further concepts: become a collection of the appropriate dimension
‘replication’ (in various forms) and ‘modifiers’. The and size. This makes associative programming in-
full power of DesignScript emerges when the graph credibly powerful, particularly in the context of gen-
based dependencies, replication and modifiers are erating and controlling design geometry. It avoids

the designer (as a novice programmer) from being product’ of the input collections.
pre-occupied with the size or dimensionality of vari- DesignScript introduces a special notation
ables. called ‘replication guides’ to control the order in
which the cartesian product is created and takes the
Zipped replication form <n>, where n defines the sequence of the repli-
When there are multiple collections within the same cation operations. This sequence is equivalent to the
expression we need to control how these are com- order of the nested for loops that would have had to
bined. With ‘zipped’ replication, when there are mul- be written in an imperative script.
tiple collections, the corresponding member of each a; b; c; d; // define the variables at the top or outer sco-
input collection is used for each evaluation of the pe
expression. This works well when all collections are [Associative]
the same dimension and length. If collections are of {
different lengths, then the shortest collections de- a = {1, 5, 9};
termines the number of times the expression is eval- b = {2, 4 };
uated, and hence the size of the resulting collection. c = a<1> + b<2>; // cartesian replication c = { { 3, 5
a; b; c; // define the variables to be output at the top }, { 7, 9 }, { 11, 13 } }
or outer scope // changing the sequence of replication guides chan-
[Associative] ges the resulting collection
{ d = a<2> + b<1>; // d = { { 3, 7, 11 }, { 5, 9, 13 } }
a = {1, 5, 9}; }
b = {2, 4, 6}; The equivalent Imperative code would require dou-
c = a + b; // zipped replication operation .. c = {3, bly nested for loops. To compute c the outer loop
9, 15} would iterate over a and the inner loop would iter-
} ate over b.
The equivalent imperative code would be: a = {1, 5, 9}; b = {2, 4, 6}; c; d; // define the variables at
a = {1, 5, 9}; // define the variables to be output at the the top or outer scope
top or outer scope [Imperative]
b = {2, 4, 6}; {
c; m = Count(a); n = Count(b);
[Imperative] for (i in 0..m)
{ {
n = Math.Min(Count(a), Count(b)); for (j in 0..n)
for (i in 0..n) {
{ c[i][j] = a[i] + b[j];
c[i] = a[i] + b[i]; d[j][i] = a[i] + b[j];
} }
} }
// c = {3, 9, 15} // c = { { 3, 5 }, { 7, 9 }, { 11, 13 } }
// d = { { 3, 7, 11 }, { 5, 9, 13 } }
Cartesian replication
When there are multiple input collections we need Modifiers
to control how these are combined. With ‘cartesian’ With modifiers each variable can have multiple
replication, all members of all collections are evalu- states which might be used to create a geomet-
ated so that resulting collection is the ‘cartesian ric modelling sequence. For example a geometric

Figure 6
Façade design and analysis
with DesignScript, showing
the development of a concep-
tual building model as a series
of polygons. Those polygon
which are significantly ‘out of
plane’ can be visually identi-
fied [indicated in blue in the
model, second from the right]
and this condition can be used
in a downstream fabrica-
tion process, for example by
switching to different types of
facade panels.
variable might be created (say as a curve) and then a sub collection of members. Modifiers enable these
‘modified’ by being trimmed, projected, extended, special conditions to be identified and the addition-
transformed or translated. Without the concept of al modelling operation applied.
modifiers each state or modelling operation would To give an example, imagine that a building
require to be a separate variable and this would façade is based on a set of polygons. The polygons
force the user to have to make up the names of all will be the ‘support’ geometry for the façade panels.
these intermediate variables. Modifiers avoid impos- However, in this example those polygons which are
ing this naming process on the user. ‘out of plane’ by some critical dimension require a
special modification before being used as the sup-
Combining dependencies, replication and port for the corresponding facade panels (Figure 6).
modifiers The designer may want to apply a special opera-
Dependencies, replication and modifiers can all be tion but only to the ‘out of plane’ polygons and the
combined to represent the typical modeling opera- application of this operation should not alter the
tions found in architecture and constructions. Build- particular polygon’s membership of the collection
ings are composed of collections of components. of polygons. In this context all the polygons have
Typically these collections are often the product of a a common defining operation, but some polygons
series of standard operations across all members. On have an additional ‘modifier’ operation applied.
the other hand, within such collections there may Having applied the special modifier operation
be special conditions where different or additional to the specific polygons, the designer can use the
modeling operations are required to be applied to whole collection of polygons to generate the collec-

tion of façade panels. The collection of façade pan- prerequisites. Yet behind this intuitive user interface
els is dependent on the heterogeneous collection is a highly innovative programming language that
of polygons. Additionally, the designer may want introduces a number of domain-specific program-
to embed the whole polygon collection and façade ming ideas including associativity, replication and
panel generation process in an iterative loop to opti- modifiers. These innovative ideas are combined with
mise the design for some criteria, such as solar gain, well established conventions drawn from impera-
structural efficiency. tive, functional and object oriented programming
While this example is reasonable ‘domain-spe- languages and unified into a single scalable compu-
cific’ (dealing with polygons and facade panels), the tational design application. A special characteristic
fundamental ideas (of collections, modifiers and de- of the user interface is that it supports the progres-
pendencies) on which the DesignScript language is sion from novice user to more accomplished pro-
based are necessarily quite abstract. grammer by progressively revealing or unmasking
these underlying computational concepts.
CONCLUSION
DesignScript is intended to address many of the crit- REFERENCES
ical issues in contemporary design practice, includ- Aish, R 2011, ‘DesignScript: origins, explanation, illustra-
ing the progressive acquisition of computational tion’, Proceedings of the Design Modelling Symposium,
skills, the shift to multi-disciplinary integration and Springer, Berlin.
scalability of projects. Aish, R 2013, ‘DesignScript: a Learning enviroment for De-
At one level DesignScript is an intuitive appli- sign Computation’, Proceedings of the Design Modelling
cation which can be used by designers (as novice Symposium, Springer, Berlin.
programmers) with the minimum programming

Performative Design
Performative Design - Volume 2 - Computation and Performance - eCAADe 31 | 97

98 | eCAADe 31 - Computation and Performance - Volume 2 - Performative Design
Architectural Thermal Forms
Isak Worre Foged
Department of Architecture and Media Technology, Aalborg University
www.create.aau.dk
iwfo@create.aau.dk
Abstract. The paper presents a developed method and algorithm to create environmental
sustainable optimised forms based on the solar energy received in relation to receiving,
containing and distributing energy. Different studies are created based upon this
approach, to which forms are evaluated against conventional building geometries. The
work shows a significant improvement on several aspects of environmental performance.
Lastly the work presents an idea of maximum structures, rather than minimum structures
as a path in future research work.
Keywords. Sustainable environmental architecture; performative generative algorithms;
simulation; material distribution.
INTRODUCTION
The overall form of a building describes not only its of sustainable environmental architectural forms
architectural language, but also to a large degree its and their mass properties. The argument that form
capacity to become environmental sustainable in and distribution of material has the ability to lower
relation to its local climatic environment. Movement the energy used in a building with up to 80% (Pe-
of the sun in relation to the Earth, its surface orienta- tersen, 2012), combined with the urgent need for
tion and its mass constitute the weather condition environmental adapted architecture motivates the
locally and globally as it regulates air flow, heat ac- research.
cumulation, heat transfer etc. (Oke, 1987). These re-
lations, solar geometry and mass, form the environ- Previous work
ment for life and its rhythms. The solar environment A recent trend in sustainable buildings in northern
is thus the singular most important factor in relation climates has largely been designed to minimise sur-
to climatic environmental architecture. face to volume relations, while increasing energy
The creation of architectural building forms as uptake, as seen in the Lighthouse project from 2009
derivatives of the solar environment has thus the by Christensen & Co Architects. It uses its circular
potential of improving the context specific recep- form as an optimum sustainable geometry [1]. While
tion and containment of energy towards environ- currently being an advanced example of sustainable
mental sustainable architecture. If the position of architectural form in the built environment, design
the sun in relation to Earth surface orientation and of architectural forms remains largely based on ‘sim-
its mass properties can determine local environ- ple’ design principles.
ments as briefly described above, then the same Digital techniques in simulation and generation
could be activated as a strategy for the generation of architectural form have evolved rapidly over the

last decade enabling the handling of complex cli- METHODS
matic, geometrical and manufacturing aspects. The The research applies a series of methods. Compu-
ability to construct advanced generative formal or- tation is used to simulate solar energy and thermal
ganisations have been shown in works by architects storage based on established mathematical mod-
Marc Fornes [2] and Roland Snooks (2011) through els. This is combined with a developed generative
growth systems and swarm systems. Other com- method presented here, which distributes matter in
plex arrangements of forms are created in the work space, progressively resulting in potential architec-
of Alisa Andrasak (2006) and Jenny Sabin (2008) to tural forms. Lastly the generated forms are evaluat-
just name a few. The methods and techniques from ed in a comparative analysis to conventional build-
above exhibit individual potentials with different ing forms.
objectives and at various scales. Common, is the
descriptions of local and relatively simple rules that Solar simulation
govern larger assemblies into elaborate structures. The encoded solar irradiance analysis is based upon
Today, a plethora of projects and a growing group of geometrical relations described in Lambert’s Law
people develop generative systems that construct a (Oke, 1987). Through the direct integration of the
formal output. solar geometry, azimuth and zenith angles, and
The work presented here continuous in the path thus solar energy, the variables of location can be
of generative architectural methods and systems, dynamically changed within the generative model
but develop and connects these efforts explicitly described below. The simulated surface of the dis-
with the generation of form in relation to the climat- tributed elements is always facing ‘outward’ as the
ic environment based on solar-earth-mass relations. model intents to form closed structures.
Presented work Constructor: Matter distribution, creating

This work applies digital computation as an instru- form algorithm
ment for simulation of environment and material With the intention of improving the holistic per-
behaviour and generation of architectural form from formance by solar energy, thermal forms are based
this information. With an application of generative upon the combined aspects of receiving, containing
digital models with an integrated solar analysis that and transferring energy. Receiving energy requires
progressively add material in relation to receiving an extended surface towards the energy source, fol-
and containing energy, architectural forms are cre- lowing Lambert’s Law. Containment in architecture
ated as a derivative of the solar parameters. This is based upon minimum transfer of energy from the
intents to construct both an improved platform for inside to the outside, therefore minimising the sur-
sustainable environmental architecture and a con- face to volume ratio. Transfer of energy is related to
text related architectural expression. The presented the ability to move the energy received at the sur-
research investigates the capacity to create sustain- face to the core of the form.
able environmental architectural forms that are a Based on these factors, an algorithm can be de-
response to a process of encoding a 6-dimensional scribed which is based on ‘the more energy received
factor space, constructed from x,y,z coordinates, on a surface of an element, the longer from the core
time, material properties and solar climate. It does the element can be positioned’, Figure 1. The follow-
so by application of material properties into digital ing element created undergoes the same evaluation
construction elements (points, geometrical entities), but rotated around a center point to position it next
thereby linking physical parameters to iterative digi- to the previous element. The longer the elements
tal generative studies. are positioned away from the centre point (cp) the

Figure 1
Distribution of solid matter
from centre point according
to energy.
longer distance is between the elements, effectively The factors are plotted into a scheme, Figure
extending the combined surface oriented towards 2, in relation to solar irradiance from above. The
the energy source. The progressive formation cre- scheme is related to the northern hemisphere, in
ates elements (a). Following a similar method but which southern orientations are effected by higher
with a fixed distance between a centre point and a solar gain. In case of high irradiance materials that
created element form a circular form with optimum allow high solar gain can be low etc. The scheme
surface volume relation. The progressive formation could be reconfigured for other locations and paired
creates elements (b). The formation located be- with the above form creation algorithm generating
tween elements (a) and (b), denoted elements (c) other results than presented below.
serve as an equilibrated formation. In the studies Merging the above into one model, we have the
performed in this paper, (c) is always half the dis- following algorithmic procedure:
tance between (a) and (b). From (c) solid matter is 1. Distribute test elements around center (small
distributed according to the description below. inner circle around center point)
2. Calculate element angle vector from center to
Constructor: Matter distribution, creating distributed element
mass algorithm 3. Calculate sun vector
Following the creation of form, material properties 4. Calculate angle between element (a) vector
are applied to the generative model through three and sun vector
different aspects, 1) u-value, the heat transition co- 5. Calculate radiation energy on each element
efficient, 2) g-value, the solar gain coefficient and based on Lambert’s Law
3) thermal mass. These are selected based on their 6. Distribute elements from center based from
direct reference to established architectural termi- quantity of solar energy at each test
nology, and from sensitivity analysis of the most in- 7. Distribute element (b) from center with same
fluential passive factors for sustainable architecture radii creating circle
(Petersen, 2012). 8. Calculate distance between elements (a) and

Figure 2
Distribution of mass at
elements according u-value,
g-value and thermal mass
factors.
elements (b) erties a given material should have when located at

9. Distribute elements (c) between elements (a) the given position computed by above method. This
and elements (b) as equilibrium has the intention of suggesting a distribution of ma-
10. Calculate angle between element (c) and sun terial and its properties not restricted to accessibility
vector of contemporary materials, Figure 3a.
11. Calculate radiation energy on each element Study B, thermal properties of bricks (solid mass
based on Lambert’s Law in geometric entity) is applied in the calculations of
12. Calculate optimum g- and u- value materials thermal storage in the organised form and causes
according to g-u-value scheme a scaling of the brick. This has effect on the subse-
13. Distribute elements (d) and scale/colour ele- quent distribution of thermal material to locally
ments from calculated g- and u- scale reception and further distribution of heated air
14. values (fluid matter). This is done with the intention of only
15. Calculate stored energy in elements (d) based applying material where it is needed, Figure 3b.
on material property and Study C, evaluates the solar energy on a given
16. received energy surface and suggest from two materials, high and
17. Calculate radiant energy emitted from energy low energy absorption, which material to apply
stored in elements (d) where. This has the intention of making the envi-
18. Distribute elements (e) towards center based ronment choose a material from a potential (small)
radiant energy catalogue of available materials, Figure 3c.
19. If, elements(a,b,c,d,e) angle > TWO_PI move
elements(a,b,c,d,e) up with Comparative analysis
20. distance (z) In order to understand the performance of the
21. Return to (1) forms generated, a comparative analysis is done be-
tween conventional building forms and the forms
Studies derived from the method and algorithm. The com-
Following studies are done from above developed parison is provided on below eight aspects measur-
algorithm. ing the relations between receiving, using, storing
Study A, mass is distributed based upon calcu- and transferring energy in simple measures.
lated optimum u-and g-values indicating what prop-

Figure 3
A,B,C. Distribution of solid
matter from elements (c) as
thermal storage according to
energy.
RESULTS conventional methods of applying material homo-

The formations created by the developed method geneously on all outward facing surfaces.
and algorithm illustrates clearly a circular form with
a low application of thermal mass when the sun is CONCLUSION
high on the sky. In climates with a lower sun angle Based upon the presented studies we can suggest
the form equilibrates into advanced forms with ex- that energy optimum forms cannot be generalised
tended surface towards the source and decreased to e.g. circular forms, but are much more a result of
surface in the opposing direction. This is particularly localised factors constructing the environment. If an
evident when observing the formations ‘live’ or as urban environment were included this environment
one can see in figure 3 a,b,c where the sun was set would perform different due to the change of casted
in first 1/3 of the formation high on the sky, 85 de- shadows from neighbouring buildings.
grees. The following 1/3 part of the formation the If considering a tall and dense urban environ-
sun gradually lowered to a resulting angle of 6 de- ment, where little to no direct radiation would oc-
grees creating the transitional form to the last 1/3 of cur, the hierarchy of factors could change, mean-
the formation where the sun is kept at 6 degrees. ing the equilibrated form (elements c) could have
In lower solar angles the most non-circular re- a stronger ‘weight’ towards the energy containing
sults are seen and therefore creates the largest dif- form (elements b), rather than energy receiving
ferences to conventional building forms. Within form (elements a), as the energy for reception is
the comparative analysis it is clear to see the per- not accessible. This, as a consequence, would create
formance in relation to the combined factors of forms in dense urban areas that are more uniform
receiving, containing and transferring energy, with circular than curvilinear oval forms as shown above.
significant improvements achieved based on the As the form rise above neighbouring buildings
presented method. the ‘weights’ could alter again towards a hierarchy
Based upon the thermal matter volume distrib- where reception of energy was favoured, resulting
uted on the generated surface a heterogeneous in very modified geometry compared to the perfect
surface structure density was created resulting in circle. This is what we see in figure 3abc.
an organisation of solid matter only where needed. The model, beyond the overall form, also intro-
In figure 4, this has a large effect when compared to duces the calculation of a floor plan depth by how

Figure 4
Comparative analysis be-
tween conventional building
forms and the computed
forms.

long the energy potentially could be transferred changes through urban reconstruction over time,
into the building after reception at the envelope. If the local climatic environments alter with it in a re-
further developed, this approach could suggest in- ciprocal manner. This suggests that the approach
ternal spatial organisation and divisions according could be expanded into a more responsive struc-
to light levels needed, operating temperature, and ture, which by above method could suggest new
spatial programme when considering the occupier forms, but with an underlying structural organisa-
as an environmental source. tion that would allow for a dynamic adaptation re-
Introducing more sources of energy, such as air sulting in changing the overall form of the building.
flow cooling the surface or internal loads of occu- A less drastic approach in relation to moving
pancy and equipment could in a reciprocal relation structure adaptation could be performed locally at
further alter the articulation of the generated archi- the material scale, directly utilising the presented
tectural form as energy is created both on the inside method of material property distribution. This
and outside of the envelope. would require that the time-based adaptive proper-
ties, such as phase-change-materials where imple-
DISCUSSION mented in the initial generative process of distribut-
The work follows the biological notion by Darcy W. ing mass.
Thompson (1992) that a ‘form of an object is a dia-
gramme of forces’. Additionally, we might add that REFERENCES
the local forces are derivatives of the local form it Andrasek, A., 2006, Continuum – A Self-Engineering Creature-
interacts with. Culture, AD Wiley.
Architecture has, for good reasons of structural Oke, T.R., 1987, Boundary Layer Climates, 2nd Edition, Rout-
optimisation, had a singular focus of ‘minimum ledge.
structures’ by a tremendously rich legacy from Frei Oxman, N., 2010, ‘Material Based Design Computation’, PhD
Otto, Erik Reitzel and others. However, this work also Thesis, MIT, Boston
suggest to embark on an approach that can be re- Petersen, D.M., 2011, ‘Informing the Design Process: A Study
ferred to as ‘maximum structures’, which not only fo- of Architect’s Approach to Environmental Architecture’,
cuses on minimisation of used material, but equally PhD Thesis, Aalborg University
how it maximise its performative organisation in Sabin, J. and Jones, P.L., 2008, Nonlinear Systems Biology
relation to environmental sustainable architectures. and Design – Surface Design, International Conference
Such an approach could be initiated by the pre- ACADIA Proceedings.
sented method of distributing material properties, Snooks, R., 2011, ‘Digital Crafting Symposium. Generative
such as thermal mass, transparency, even tactility, to Logics’, Lecture, 18. January, Aarhus School of Archi-
expand the performative and aesthetical instrumen- tecture
tality of the generative models. This again, would Thompson, D.W., 1992, On Growth and Form, Dover Publica-
expand the factors involved, and suggest an added tions, New York.
hierarchy of factors to the one used within this work
by the mentioned sensitivity analysis for sustainable [1] http://www.christensenco.dk/projekter&p=22 (Retrie-
architectures. ved: 29 November 2012)
The work is presented as environmental per- [2] http://theverymany.com/ (Retrieved: 25 January 2012)
formative forms, but as the built environment

DaylightGen: From Daylight Intentions to Architectural
Solutions
Implementation and experimentation of a generative and parametric

design tool
Mohamed-Anis Gallas1, Gilles Halin2

Centre de Recherche en Architecture et Ingénierie (MAP-CRAI/UMR 3495 CNRS/MCC),
Nancy, France
1
http://meurthe.crai.archi.fr/wordpressFr/?page_id=31&num_pers=247, 2http://meurthe.
crai.archi.fr/wordpressFr/?page_id=31&num_pers=4
1
gallas@crai.archi.fr, 2halin@crai.archi.fr
Abstract. This paper addresses the integration of the daylight effect during the early
stages of the architectural design process. The first part presents a design assistance
method that helps designers to characterize their daylight intentions and materialize them
in architectural solutions. In this part, we describe the implementation of this method in
a design tool, denoted DaylightGen, the implemented process and the different system
components. The second part of this paper focuses on the investigation of the potential of
the proposed method in design process. It was evaluated in educational design case study.
This part integrates the experimentation process and his results.
Keywords. Computer aided architectural design; intentions oriented design; generative
and parametric design tool; daylight simulation tool; design tool experimentation.
THE DAYLIGHTGEN METHOD

This method proposes to help young designers to scribes his daylight intentions using a visual sup-
integrate their daylight effects intentions during port. The designer uses pictures representing day-
the first step of the design process. The designer light effects to describe his own ideas.
describes the daylight effects that he would like to He selects whose are corresponding to his inten-
integrate in his project. The described intentions tions. The second step “characterize intentions (2)” of
are characterized and translated into potential solu- the method process aims to identify and character-
tions. The generated solutions will be used as a base ize the designer’s intentions. They are translated in
line models to start the project design (Gallas et al., a physicals and geometrical information that will be
2011). used at the “solution generation (3)” step. During this
The DaylightGen method is organized in five step the method will propose architectural solutions
steps (Figure 1). The method process starts by the that create the described daylight effect and verify
“declare intentions (1)” step where the designer de- by the way the designer intentions. The solution

Figure 1
DaylightGen method process
and activities.
Figure 2
DaylightGen implementation
tools.
generation step is organized as an iterative process materialize the “characterize intentions” and the
of a generation, evaluation, mutation and selection “generate solutions” activities and finally the Day-
activities. The fourth step of the DaylightGen meth- lightViewer integrates the “assessing” and the “per-
od process is the “assessing (4)” step where the gen- sonalize solutions” step’s activities (Figure 2).
erated solutions are visualized and presented to the
designer as a result of the “generate solutions” step. Day@mbiance
The “assessing” step integrates five activities where Day@mbiance is a navigation tool in a references im-
the designer visualizes and navigates in the collec- ages base proposed by Salma Chaabouni (Chaabou-
tion of generated solutions, evaluates and compares ni et al, 2008). The images base is structured as a
them and finally selects the best ones. The method MySQL® database and managed by Mamp®. The
process ends by the “personalize solution (5)” step navigation in the images base is realized by a PHP®
where designer could modify the generated solu- application with a Flex® interface. A web browser
tion features and transform them to integrate new (Firefox®) is used to visualize the Day@mbiance func-
ideas. The modified solutions will be visualized and tions and results (Figure 3).
evaluated with an iterative manner. The “personalize Day@mbiance is used to identify the designer
solution” step accompanies the designer and takes daylight intention. Its process starts with a first
end when his is satisfied.
Figure 3
THE DAYLIGHTGEN TOOL Day@mbiance implementa-
This method is implemented in a design assistance tion modules and environ-
tool denoted DaylightGen. This prototype is comment.
posed of three tools: Day@mbiance, DaylightBox and
DaylightViewer. Day@mbiance is used to implement
the “declare intentions” step activities, DaylightBox

Figure 4
Day@mbiance interface.
mosaic of images representing daylight effects. De- image, the pertinence weight of the keywords used
signer chooses images that represent his intentions, to index it increases, when the image is refused the
refuses that are at the opposite and leaves others weight decreases and finally the weight stay the
neutral to finally generate a new mosaic that takes same if the image is neutral. The pertinence Weight
into account his choices (Figure 4). All images are of the keywords is used when Day@mbiance gener-
indexed by keywords that describe the visualized arate a new mosaic to take into account the designer
chitectural configurations and daylight effects. This preferences.
process will be repeated until the designer finds a
collection of relevant images that corresponds to his DaylightBox
intentions. The DaylightBox tool is implemented in Rhinoceros®
All images used by Day@mbiance are indexed modeler environment and his graphical algorithm
using a keyword collection structured in a thesau- editor Grasshopper® (Tedeschi, 2011). Daylight-
rus. The thesaurus is divided in five facets that de- Box is a Grasshopper® definition that integrates six
scribe all images features: the daylight effect type, modules: a referenced images base (Day@mbiance
the quality, the quantity of daylight, the space sur- images base), a daylight effects knowledge base
faces aspects and the space function. The indexation (knowledge base), a parametric model (geometry), a
process is realized by Image (software developed by generative algorithm (Galapagos®), a daylight simu-
Pascal Humbert form MAP-CRAI) (Figure 5). lation tool (simulation) and a solutions database (so-
The user’s choice is then characterized a set of lutions storage) (Figure 6).
relevant keywords. The Keywords used to index the The first module “Day@mbiance imagse base” is
images has a pertinence weight that varies between a cluster that integrates a plug-in to connect Grass-
-1 (not relevant) and 1 (relevant) (Halin, Créhange hopper® to the pictures base used by Day@mbiance.
and Kerekes, 1990). When the designer choose an This module selects the most significant keywords

Figure 5
Picture indexation using
“Image”.
that characterize represented daylight effects in or- The third module “geometry” is a parametric
der to highlight the designer intentions. After that, model of parallelepiped shape defined by thirteen
the designer selects one of the identified daylight parameters (Table 1). These parameters define all
effects to start a solutions generation process. The the spatial features that influence the daylight be-
second module “knowledge base” is used to iden- havior. The model parameters are implemented in
tify and characterize the designer’s intentions. This sliders that determinate their data types and their
knowledge base contains the quantitative and variation ranges.
qualitative features of different and recognized day- The fourth module is a generative algorithm
light effects. These features are integrated in a fitness (Galapagos®) that controls the parametric model fea-
function attached to each known daylight effect. tures to generate solutions verifying the fitness val-
It is composed of variables that characterize solar ue. The genetic algorithm uses the fitness function
gains and their spatial distributions. and his objective value (fitness value) to optimize
Figure 6
DaylightBox modules and
software environment.

Table 1 Model parameters Data type Variation range
Parametric model features. Dimension Integer 250..750
Aperture rate Floating 0..1
Aperture proportion Integer 1..5
Aperture orientation Integer 0..7
Aperture direction Integer 0..1
Aperture number Integer 1..10
Aperture position Integer 1..11
Aperture face Integer 0..1
Surface type Integer 0..3
Aperture surface type Integer 0..2
the solution generation process. storage”. This module create a link between Grass-
The fifth module “simulation” integrates the hopper® and MySQL® using the Slingshot® [1] plugin.
plug-in Diva-for-Rhino® (Jakubiec and Reinhart,
2011) to connect Grasshopper® to Radiance® simula- DaylightViewer
tion software. The system process iterate on a cycle The DaylightViewer tool is implemented in Rhinoc-
composed of three main steps: eros® modeler environment and his graphical algo-
1. The genetic algorithm Galapagos® finds archi- rithm editor Grasshopper®. DaylightViwer is a Grass-
tectural parameters values using the selection, hopper® definition that integrates three modules: a
crossover and mutation operators. It optimizes visualization interface defined by the “visualization”
the generated solution behavior and tries to module, a simulation module “simulation_eva” and
reach the fitness value. the parametric model “geometry_per” (Figure 7).
2. The parametric model “geometry” generates ar- The first module “visualization” imports the best
chitectural models defined by the parameters solutions according to their fitness value (solution
values provided by the genetic algorithm. with the lower value of fitness). The selected solu-
3. The simulation module analyzes the daylight tion are visualized and organized in a colored grid
features of the geometry generated by the par- from the best to the worst one. The user selects the
ametric model. The simulation results are used number of solution to visualize and navigate under
to compute the fitness value. the visualized solutions (using Rhinoceros® visualiza-
The generating process ends after a fixed num- tion windows) to select those corresponding to his
ber of generations. All the generated solutions fea- intentions. The second module integrates a simula-
tures (parameters and fitness values) are stored in a tion tool that makes realistic and quantitative simu-
MySQL® database using the sixth module “solution lations in order to verify that chosen solutions pro-
duce the described daylight effect. The third module
Figure 7 is composed of geometrical operators that could be
DaylightViewer modules and used by designer to transform the proposed solu-
software environment. tions. The module “geometry” presents the features
of the selected solution and the list of sliders to
modify the parameters values. The transformed so-
lution could be evaluated (realistic and quantitative
simulations) and exported it in 3D geometrical ob-
jects (bake them from Rhinoceros® to Grasshopper®)
(Figure 8).

Figure 8
Solutions grid and evaluation
visualization.
THE DAYLIGHTGEN EXPERIMENTATION to the experimentation activities. We organize a

The third part of this paper assesses the DaylightGen training sessions to present the experimental frame-
tool contribution during the early design steps. We work, the design assistance method targets and the
try to determinate: utilization mode of the different components of
• the adaptation level of the method activities DaylightGen tool (Day@mbiance, DaylightViewer and
and functions to the early design steps fea- DaylightBox). The second step is organized in design
tures, sessions for duration of three hours where a pair of
• the capacity of the assistant method to take student tries to design a project sketch that express
into account the uncertainly that characterize particular and significant interest to daylight atmos-
early design steps, pheres. The experimentation participants are au-
• the creativity level and the capacity of the thorized to use different sketch and modeling tools
method to help designer to explore new ideas, (paper format, Photoshop®, AutoCad®, Sketchup® and
• the capacity to satisfy and materialize designer Rhinoceros®). All design sessions are captured on
intentions. video to keep the student’s discussions and the ex-
The potential of the method has been evaluated changes with the different tools (Figure 9).
in an educational context. We proposed to 16 Mas- The third step of the experimentation process
ter degree students in architecture to design a pro- aims to evaluate the participation of DaylightGen
ject by using the DaylightGen tool. They worked in tool during the conceptual design phases. The
pairs during three hours to design a temporary ex- evaluation step starts by analyzing the captured
hibition space dedicated to a designer (Jean Prouvé). sessions. Analyzing the experimentation feedback
The project program is composed of two temporary questionnaire realized with all participants con-
exhibition rooms. The first integrates a graphics me- cludes the evaluation step. This online questionnaire
dia describing the designer life. The second room is [2] integrates questions about the different func-
dedicated to the exhibition of models and some ar- tions and the results obtained by DaylightGen tool.
chitectural components.
Experimentation results
Experimentation process 1. Video capture analyze
The experimentation process is organized on three We determinate the design process used by the
steps. The first step aims to prepare the participants experimentation participants by analyzing the de-

Figure 9
Experimental device.
sign sessions videos. We identify the major design identifying and selecting daylight effect, generating
activities that participate to the design process and solution and proposing a spatial configuration for
their chaining. All design activities and the design the project. The second approach (used by group
supports used are transcribed in time line diagrams 4-5-6-7) starts by formulating design problem, im-
(Figure 10). The diagrams analyze reveals three de- planting the project, proposing a spatial configura-
sign approaches (Figure 11). The first one (used by tion, formulating and declaring daylight effect inten-
group 1 and 2) starts by formulating design prob- tions using Day@mbiance, identifying and selecting
lem, implanting the project, formulating and declar- daylight effect, generating solution. The third one
ing daylight effect intentions using Day@mbiance, (used by group 3) starts by formulating design prob-
Figure 10
Example of design activities
diagram.
Figure 11
Design approaches.

Figure 12
Identify, select and evaluate
generated solutions.
lem, formulating and declaring daylight effect inten- aperture in the corner or “this is a jail effect” to de-
tions using Day@mbiance, identifying and selecting scribe a solution with two long and fine vertical ap-
daylight effect, generating solution, implanting the ertures in the right and left side of the aperture face.
project and proposing a spatial configuration for The students use the evaluation function to visualize
the project. All design approaches ends by the same the daylight effect generated by the selected solu-
way: the designers evaluate and personalize the tions and verify if they corresponds to the described
generated solutions, reshape the proposed project, intentions. They operate different simulations at dif-
finalize the project and debrief the experimentation ferent times and for different sky conditions.
session. The navigation on the solution grid and the
The participants on the experimentation use evaluation of the selected solutions gives to par-
different design supports to formulate design inten- ticipant’s new ideas that were integrated using the
tions and materialize them in architectural solutions. modification and personalization functions pro-
They associate mosaic navigation activity, manual posed by the DaylightViewer tool. They used these
sketches and oral expression to precise the daylight functions to combine different configurations and
effect intentions that correspond to the project exceed the parametric model limits. The group 3
constraints. The solutions grid generated helps stu- used the modification, the evaluation functions and
dents to explore and define new design issues. The Photoshop® to create a new architectural solution
navigation in solutions grid helps students to locate with apertures on three faces that could not be real-
and identify interesting solutions that could be im- ized by the generative model (aperture only on one
plemented in their projects (Figure 12). Students ex- face) (Figure 13).
press their interest by manual gesture and some oral The best-generated and personalized solutions
expressions like “this is small, it concentrate daylight” was integrated and implemented in the project by
to identify and describe a solution with one small analogy. The final daylight effect generated was
Figure 13
Combine generated and
personalized solutions.

Figure 14
Generated and personalized
solutions implementation.
evaluated using shadow visualization functions of mentation process used to validate the DaylightGen
Sketchup® and daylight simulation plug-in (Diva-For- method and tools targets. The experimentation re-
Rhino®) of Rhinoceros®(Figure 14). sults validate the capacity of the proposed method
2. Questionnaire answers analyze and tool to assist the daylight integration during
The questionnaire answers was used to identify early design steps. These results confirm the possi-
what users think about the design assistance meth- bility to use the design intentions as basic informa-
od and about the functions proposed by the dif- tion in design assistance tools. The experimentation
ferent tool participating on the process. The main results reveal some limits that concerns the number
part of participants declares that the use of images of identified daylight effects, the fitness function
to identify the daylight effect intentions is really precision and parametric model possibilities that
adapted to the conceptual design steps. They say could be developed in future work. These results
that images constitute a fist level of the implemen- show that the proposed method could be amelio-
tation process of design intentions. The students are rated and adapted to a professional design context.
satisfied by the generated solutions that verify at
different levels of accuracy the described intentions. ACKNOWLEDGEMENT
Student’s answers reveals that the parametric model The authors are grateful to Alstan Jabubiec and Jeff
used for the generation activities needs to integrate Niemasz to their help in the Diva-For-Rhino plug-in
more apertures types and more precise functions integration. This research work is funded by La Ré-
(multiple aperture with different sizes and shapes, gion Lorraine, France.
aperture on different surfaces, integrate personal ar-
chitectural configuration in the generation process). REFERENCES
The majority of participants consider evalua- Chaabouni, S, Bignon, JC and Halin, G 2008, ‘Supporting
tion and personalization functions as very useful ambience design with visual references’, in Architecture
because they allow users to reshape proposed solu- in Computro, presented at the Education and research
tions and to integrate new ideas. Students consider in Computer Aided Architectural Design in Europe,
that the evaluation of these new solutions helps de- Antwerpen (Belgium).
signer to create an iterative process that makes the Gallas, M, Halin, G and Bur, D 2011, ‘A “green design” method
project design progress. to integrate daylight in the early phase of the design
process: The use of intentions knowledge base to gen-
CONCLUSION erate solutions’, in Respecting Fragile Places, presented
This paper presents the implementation process of at the Education and research in Computer Aided Ar-
the DaylightGen method, the choice of the software chitectural Design in Europe, University of Ljubljana,
environment, the modules and the component used Faculty of Architecture (Slovenia).
to create the design assistance tool. It presents also Halin, G, Créhange, M, and Kerekes, P 1990, ‘Machine learn-
the different steps, devices and results of the experi- ing and vectorial matching for an image retrieval mod-

el’, presented at the 13th International Conference on 2209.
Research and Development in Information Retrieval, Tedeschi, A 2011, Parametric architecture with Grasshopper®:
Bruxelles. primer, Brienza, Italy: Le Penseur.
Jakubiec, JA and Reinhart, CF 2011, ‘DIVA 2.0: Integrating
daylight and thermal simulations using Rhinoceros [1] http://slingshot-dev.wikidot.com/grasshopper-mysql
3D, Daysim and EnergyPlus’, Proceedings of the Build- [2] https://docs.google.com/spreadsheet/viewform?formk
ing Simulation Conference, Sydney, Australia, pp. 2202- ey=dGtLWXdsTHFvS0h5RFAzLWVwaEM0N3c6MQ

Performance Driven Design and Design Information
Exchange
Establishing a computational design methodology for parametric and

performance-driven design of structures via topology optimization for
rough structurally informed design models
Sina Mostafavi1, Mauricio Morales Beltran2, Nimish Biloria3

1,3
TU Delft, Hyperbody, AE+T, Faculty of Architecture, The Netherlands, 2TU Delft, Struc-
tural Design, AE+T, Faculty of Architecture, The Netherlands
1,3
www.hyperbody.nl, 2http://www.bk.tudelft.nl/en/about-faculty/departments/architectural-
engineering-and-technology/organisation/chairs-ae-t/structural-design/
1
s.mostafavi@tudelft.nl, 2m.g.moralesbeltran@tudelft.nl, 3n.m.biloria@tudelft.nl
Abstract. This paper presents a performance driven computational design methodology

through introducing a case on parametric structural design. The paper describes the
process of design technology development and frames a design methodology through
which engineering, -in this case structural- aspects of architectural design could become
more understandable, traceable and implementable by designers for dynamic and
valid performance measurements and estimations. The research further embeds and
customizes the process of topology optimization for specific design problems, in this
case applied to the design of truss structures, for testing how the discretized results of
Finite Elements Analysis in topology optimization can become the inputs for designing
optimal trussed beams or cantilevers alternatives. The procedures of design information
exchange between generative, simulative and evaluative modules for approaching the
abovementioned engineering and design deliverables are developed and discussed in this
paper.
Keywords. Performance driven design; design information; design technology; topology
optimization; parametric design.
INTRODUCTION
One of the challenges in performance driven de- fective knowledge integration in Computer Aided
sign methodologies is the way that designers can Design (CAD) techniques and methods (Cavieres et
effectively integrate simulation and optimization al., 2011). In design practice, theoretically this gap
techniques with parametric design and generative is bridged via simultaneous consultations with en-
procedures (Oxman, 2008). This challenge can also gineers and specialists. However, for many design
be attributed to as the lack of tools to support ef- problems this concurrency might not be achievable

and applicable. In this paper, as one of the directions Figure 1
towards achieving this concurrency we specifically a: Topology optimization
focus on the implementation of optimization tech- method for designing of
niques in structural design, to see how they can be bicycle [1], b: COLNAGO
integrated with parametric design techniques. To be monocoque bike frame
more explicit from a computational design point of designed using FEA, c: FEA
view, and to the design methodology itself, the fo- applied in designing of a
cus of the article is on design information modeling, tower [1], d: Casted free form
exchange and interoperability. The paper structure concrete column by Arata
here onwards addresses questions and objectives, Isosaki in Shanghai, e: Close-
the process, the tool and the methodology. Subse- up of Eifel tower column with
quently, the results from the examples and a case similar morphology with steel
study are briefly reported and eventually the discus- elements.
sion focuses on performative design methodology, for the scale of a building, while we usually have
its supporting design technology, rough Building building elements like bars, beams, columns and
Information Modeling (BIM) systems (Eastman et al., joints, directly using the discretized result of topol-
2011) and future directions. ogy optimization might not be much applicable and
might impose choosing in-site concrete casting (Fig-
Question and Objectives ures1c and 1d). While in this research, we build on
Two major questions are the subjects of exploration the assumption that it would be more relevant for
in this research. The first one, which is more from a designing a truss or frame structures, if we translate
computational design perspective, questions the the finite geometry to a proper geometric system
possibility to appropriately integrate optimization with nodes and bars (Figure 1e).
algorithms and procedures, -in our case, topology While to a certain extent the process is defined
optimization, within a parametric design system. and developed as a generic design technology, the
Pertaining to this question, the objective is to design type of structural system is intentionally and precise-
a system with connected sub-procedures and feed- ly defined as a trussed beam or cantilever. Besides
backs with appropriate methods for design informa- the developed algorithms, from a technical point
tion exchange and translation from different CAD of view, testing and developing of various methods
and programming platforms (operating as design for information exchange between software and
decision support) for performance driven design platforms like Matlab, Rhinoceros, Grasshopper plus
(in this case is a truss structural system). The concur- some of its add-on plug-ins and the needed struc-
rency and consistency in extracting, generating and tural analysis software shall also be elaborated.
structuring of geometric design information such as
size, resolution, etc.- and non-geometric design in- DESIGN PROCESS AND METHODOLOGY
formation such as load conditions, Degree of Free- The process, as illustrated in the flowchart (Figure 2),
dom (DOF), etc. will be discussed. is a set of sub-procedures -A to D- such that the out-
The second question is how to make the process put, input and procedures are systematically corre-
of topology optimization more suitable for the scale lated. In each sub-procedure there are four kinds of
of architectural design and what are the benefits of modules, which are decisions or inputs (), processes
doing so? This method and in general Finite Element (), outputs () and visualizations (). To make this tech-
Analysis (FEA) have been widely used at the scale of nology-based design process an interactive, cyclic
industrial design for uni-body or monocoques struc- and performative one, the following aspects have
tures like bike frames (Figures 1a and 1b). However been taken into consideration:

Figure 2
The flowchart of the process,
illustrating the correlated sub
procedures.
• The decision(s) or input(s) of each sub-pro- matically be processed as the input of the next
cedure are used as common inputs for more sub-procedure(s).
than one of the sub-procedures, whenever and • After each single measurement or evaluation
wherever needed. module there is either a visualization for alert-
• The final translated output in each of the sub- ing or a feedback loop to the previous stages.
procedures would automatically or semi-auto-

Figure 3
a: Some of the input param-
eters, b: Variations in design
domain definition plus load
and support conditions, c: A
case with its corresponding
output set.
The detailed descriptions of each of the sub-proce- tion. This step is done through visual programming
dures are as follows: using Grasshopper in Rhinoceros. The generated
geometry attributes and alert messages (if either
Definition, design domain, discretization the geometry or resolution is not within some pre-
and load condition [A] defined range) are simultaneously visualized (Figure
In this phase, the designer defines the geometrical 3b and 3c).
properties on which the supports and loads can be
parametrically added and modified. These proper- Material distribution (MD); topology opti-
ties are, so far, the span and the height of a canti- mization [B]
lever or a beam with either upper or lower distrib- In this stage, the goal is to find the optimal mate-
uted or point loads on sides. However, the process rial distribution of the discretized generated design.
in this stage and other stages is designed in a way This step is in Matlab and is based on the implemen-
that more irregular initial shapes are also possible to tation and development of a topology optimization
implement, by just removing some portions of the code, originally written by Sigmund (2001) with the
initial planar design domain. The main inputs in this purpose of solving linear compliance minimization
sub-procedure are the dimensions, the magnitude using an optimizer and finite element subroutine.
and coordination of loads, supports and the mesh Modifications in the code are set up, with the objec-
resolution (Figure 3a). Since this mesh resolution is tive of making it compatible with the input data files
indeed the discretization of the design domain for and supports interoperability of the output for the
the following FEA, the acceptable resolution is a next sub-procedures.
variable depending on the available computation The geometrical properties, DOFs and loads will
time, power and the desired refinement. The output be automatically called in the code and what has
is a two-dimensional matrix or data list in .txt format to be defined by the designer is the percentage of
that contains the relative dimensions of the geom- total remaining material. Consequently, two paral-
etry based on the discretization resolution, magni- lel results are the outputs of this phase, one a set of
tude, the relative coordinates and calculated DOFs images that in real-time illustrates the results of ma-
of each load positions based on the defined resolu- terial distribution simulation and the other, a set of

Figure 4
a: Tests for finding proper
resolution ranges, b: Close-up
of a result spread sheet with
digits from 0 to 1, c: An input
set example and four different
topologies of the same design
domain and load and support
conditions.
excel spreadsheets, in which numerical values rang- (Figures 5a and 5b). Although in the initial visualized
ing from zero to one are stored. In the tested cases, topology the lines are detectable with the eyes of
four spreadsheets, respectively, with 30, 40, 50 and the designer, they are not automatically distinguish-
60 percentage of remaining material have been the able for the CAD platform. So one of the main crucial
final outputs. In order to make this process more challenges here was to extract the nodes and define
semi-automatic, further modifications can also be the bars by using and developing appropriate algo-
done in the code to pre-define the range for remain- rithms in a way that the topologies do not change.
ing material in previous sub-procedures (Figure 4). This implies that if in a resultant image we see nine
white polygons in the resultant vector geometry we
Typology, defining the type of structure [C] should have also the same condition. Finally, the
The goal in this sub-procedure is to translate dis- output is a matrix as .txt file with the required infor-
crete or pixelated geometry, which is the result in- mation of nodes, bars and load conditions in the de-
formation from the topology optimization to a vec- sired format (Figure 5c).
tor-based geometric system with nodes and lines Figure 6 illustrates the applied and developed
Figure 5
a: the converted spreadsheet
to discrete geometry, b:
extracted nodes and bars
of same design domain, c:
output set containing infor-
mation on nodes, connectivity
and load condition.

Figure 6
The methods implemented for
translating discrete to vector-
based geometry with nodes
and bars.
methods for extracting the nodes from the result- gos (evolutionary solver) for finding the shapes with
ant discrete geometry. After reading the float values optimum areas. By having the straight lines of the
on the spreadsheets and re-visualizing the results positive shapes (Figure 6f ), it would also be possible
through using visual programming in Grasshopper, to develop and apply a skeletonization technique
and tagging each cells with its corresponding zero based on Voronoi algorithms (Aurenhammer and
to one value, a filter separates the cells into two lists Klein, 2000) to get axial curves with similar original
of data. The reason for having this buffer is to let the topology (Figure 6g). Then by means of a Boolean
designer find the appropriate continuous topology gate the generated points through skeletonization
similar to the image result but this time composed algorithm can be achievable in a separate point
of surfaces with the size of defined resolution. For cloud list (Figure 6h). After connecting the points
instance, in the Figure 6 this filter value is 0.3, which to their neighboring, the nodes are those which has
means that all values less than this would be within three or more connections. Therefore, another algo-
a list to create the negative shape and those cells rithm is developed to automatically detect nodes
with values equal or more than this threshold will based on the numbers of connected neighbors
create the positive shape (Figures 6a-c). (Figure 6i and 6j). Subsequent to this step another
In the next step, after joining the negative optional procedure is also developed in which the
shapes and retrieving the outer boundary curves, detected nodes would be anchor points of physical
the goal is to transform the jagged edges of these spring systems and other points will be stretched
shapes into straight lines extract polygons. This is while having the fixed nodes as their supports.
done through minimizing the difference between Therefore, with this method the poly-lines, which
the areas of shapes with straight lines from the are not geometrically straight lines, will be stretched
original one with jagged edges. (Figure 6d and 6e). to form the bars.
This part is mainly done through visual and script Using this sub-procedure for all cases would al-
based programming in Grasshopper, and Galapa- low us to have a persistent method to retrieve four

Figure 7
An overview of analysis for
a beam case for one of the
translated vector-based
topologies.
set of nodes and bards for each of the volume frac- further visualization and profile assignation in 3D
tions for any parametrically defined design domain design environment. Further information for evalua-
with distinct load and support conditions in the first tion and comparisons for different input parameters
sub-procedure. After having the nodes and bars the and topologies like total volume, maximum and
structural determinacy of the each vector-based to- minimum length of the elements can be extracted
pologies will also be measured in advance through from the optimum result depending on the design
putting the numbers of the bars nodes and supports requirements.
conditions in static equilibrium. The fitness criteria in the search process are al-
lowable stress of the bars and global displacement.
Analysis, structural behavior and search The search process finds the minimum required
for optimal solution [D] cross sectional area from the defined input sets for
This stage starts with reading the input file in Mat- each of the bar elements and simultaneously check-
lab with the information on nodes, bars and load ing the allowable global displacement. This part of
conditions from the previous step coming from the the process is mainly done implementing a code in
Rhino/Grasshopper. By having this information set Matlab for cross sectional optimization. Moreover,
for each of the four topologies, a static structural in order to check the reliability of the process, some
analysis will be run for obtaining local stresses and results have been compared with the results in the
global displacement of the truss with the initial load GSA suite. Figure 8 represents an overview for a can-
and support conditions. Other variables such as ma- tilever that has started from the discrete geometry
terial properties and available profiles can also be to vector geometry with nodes and lines in which
parametrically defined or extracted form a data set the cross sectional optimization results are directly
in this stage. Figure 7 presents an overview of this used as input data for tubular profile assignment, re-
sub-procedure for a beam case. Here, the gener- sults in differentiation in the size of the each profile.
ated data list store the results that will be used for

Figure 8
An overview of the last sub
procedure with heterogeneous
point load.
TESTS AND CASES similar design problems but with different sizes and
In addition to separate examinations inside each of proportions. This means that parametrically defin-
the sub-procedures to improve and test the func- ing the initial design domain while having concur-
tionality and generalizability of the applied methods rent performance measurements would add to the
are conducted, two A-to-Z cases have been tested efficiency of the design process itself. Additionally,
which will be briefly reported and shortly discussed as it is illustrated in Figure 10, for each design do-
here. First one is a cantilever case with one point main with different load conditions we have four
load at its end (Figure 9). As it is illustrated here the optimized topologies in vector format with nodes
results of optimization based on the initial design and bars that can be translated to steel, wood or
domain and load conditions are translated to a set any other profiles. Moreover, based on cross sec-
of optimized truss structure. In this case and for any tional optimization we will have a differentiation in
of its variation, besides the topological difference profile properties which might be a source of new
between the final topologies, the corresponding performance driven design idea for designers. In
information sets pertaining to the structural per- other words, in addition to automatic evaluations
formance and geometric properties of elements are and comparisons based on the generated and
also available for further evaluation and comparison. stored quantitative information, the developed de-
The second case is a beam but in this case with- sign system might also suggest some implicit hints
in a real world background design scenario for fur- based on the visualized information and rough per-
ther validation of developed methods. This exercise formance estimation. For instance in this case the
builds upon a featured connecting bridges based architects might decide just to have one support for
project by Steven Hall Architects (Figure 10). the roof of the bridge at a specific coordinate and
One of the benefits in this case is that there are have lateral beams to support the walking deck at
Figure 9
An overview of a tests on a
cantilever case.

Figure 10
An overview of a case study
to facilitate a performance
driven design methodology
for bridges in a project like
linked hybrid by Steven Hall
architects. Photo by [2].
every six meters. With these presumptions, based DISCUSSION

on what the designer perceived from the way the In terms of design methodology, this research ad-
algorithms lead to optimum solutions, he or she dress the integration of performance measurement
could alter the input parameters, go back to the very and evaluation modules in a parametric design
beginning stages, and find the optimum result with system, opening the black box of topology opti-
required conditions and acceptable proportions si- mization, making it more traceable, specific and
multaneously (Figure 11). applicable by designers, particularly at the scale of
Figure 11 architectural design. Proposed algorithmic-design-
A feed-back attained through information-exchange scenario between steps of
parametrically re-defining the design procedure parallel to CAD and program-
the initial conditions and the ming platforms have been considered and tested as
design domain for a similar an appropriate approach for this goal. Behind the
case to retrieve another opti- benefits that can be implicitly and explicitly enu-
mized topology plus needed merated for this specific case, a conclusion is that
geometric and non-geometric knowledge integration in parametric design needs
information for next stage. customized scenarios for integration and structur-

ing of geometric and non-geometric information. ACKNOWLEDGEMENTS
The extent till which extraction and visualization The authors highly appreciate the permission given
of this information is needed, is dependent on one by professor Ole Sigmund on the use of the 99 lines
hand on the knowledge of the user as designer, and topology optimization code for the purposes de-
on the other hand on the design requirements and scribed in this paper.
goals for a specific design problem.
From a design technology point of view, in addi- REFERENCES
tion to developed algorithms for solving the issues Aurenhammer, F. and Klein, R. 2000. Voronoi diagrams.
on interoperability or data exchange, what was pe- Handbook of computational geometry, 5, 201-290.
culiarly challenging in this research was developing Cavieres, A., Gentry, R. and Al-Haddad, T. 2011. Knowledge-
and customizing a method for translating finite or based parametric tools for concrete masonry walls:
discrete geometry to vector based and continuous Conceptual design and preliminary structural analysis.
topology. It is possible to deduce that the translation Automation in Construction, 20, 716-728.
procedure can be considered as a more generic issue Eastman, C., Teicholz, P., Sacks, R. and Liston, K. 2011. BIM
in parametric and performative driven design strate- handbook: A guide to building information modeling for
gies. In addition to the process of FEA-based topol- owners, managers, designers, engineers and contractors,
ogy optimization methods in the realm of structural Wiley.
design, FEA methods are omnipresent in the basis Oxman, R. 2008. Performance-based design: current prac-
of many simulation techniques. Considering this tices and research issues. International journal of archi-
fact, it might be beneficial to facilitate performance tectural computing, 6, 1-17.
driven design methodologies with methods, tools Sigmund, O. 2001. A 99 line topology optimization code
and strategies for such translational procedures. The written in Matlab. Structural and Multidisciplinary Opti-
implementation of skeletonization and node finding mization, 21, 120-127.
methods can be considered as some of these cases
for such translational algorithms. Future cases need [1] http://paulino.cee.illinois.edu/
to be defined with similar methodological schemes [2] http://robertacucchiaro.files.wordpress.com/2011/12/
to test their generalizability and functionality. 146946-050-c6899f9f.jpg

Performance Based Pavilion Design
A dialogue between environmental and structural performance

Sevil Yazici
Ozyegin University, Faculty of Architecture and Design, Turkey
www.sevilyazici.com
sevil.yazici@sevilyazici.com
Abstract. This paper investigates the design process of a performance based pavilion
from concept towards construction phases, by challenging conventional form and
fabrication techniques. The proposed project is considered as a temporary structure,
located in Antalya, Turkey. A free-form structure and a parametrically defined cladding
are designed to serve as an installation unit, a shading element and urban furniture.
The pavilion geometry, performance assessments and proposed fabrication schemes are
clearly described in the paper. The method integrates form, performance, material and
fabrication constraints and exposes how environmental and structural performances,
including Solar Access Analysis and Static Structural Analysis, may inform the design
project.
Keywords. Parametric design; performance; architectural geometry; material;
fabrication.
INTRODUCTION
Design process consists of various phases from con- chitectural design process, which investigate form,
ceptualization to construction, including structuring performance and material aspects of design. The
of the problem, preliminary design, refinement and rationalization process of free-form surfaces towards
detailing (Goel 1992). Towards manufacturing of ar- fabrication is widely investigated along with pan-
chitectural form, different parties are involved in the elization tools, in which architectural geometry is
design process, including the design and consultant subdivided into smaller components. The number
teams. Architectural geometry needs to incorporate of unsolved tasks is enlarged by the number of dif-
many requirements of aesthetic, programmatic, ferent materials being used, because their perfor-
functional, technical and environmental aspects mance and manufacturing technology have to enter
(Holzer and Downing, 2008). However, performance the panel layout computation (Pottman et. al. 2008).
simulation of buildings is mostly undertaken in a The statics-aware initialization procedure for the lay-
later stage and cannot be integrated into design- out of planar quadrilateral meshes is approximated
decision making (Schlueter and Thesseling, 2009). a given free-form surface, by obtaining the me-
This issue reduces the efficiency in the design pro- chanical properties of the initial mesh (Schiftner and
cess radically. Balzer, 2010). A recent study aims to explore fun-
There are methods and tools developed for ar- damental principles of a system, in which a perfor-

mance based architectural geometry is generated. fabrication schemes are investigated specified as
Parametric modeling, panelization tools and series steel and wooden structural systems.
of analysis tools are used with the intent of assigning
different materials to the geometry within the same The Description of the Pavilion Geometry
boundary conditions and comparing their structural The pavilion is created by a combination of con-
performance results (Yazici and Tanacan, 2012). An straints related to the geometry, performance, ma-
integral computational model promotes an under- terial and fabrication techniques. A free-form NURBS
standing of material, form and performance not surface is generated, which defines a 3D space spec-
as separate elements, but rather as complex inter- ified by two major construction curves.
relations (Hensel and Menges, 2008). As a similar One of the construction curves is the cross sec-
approach, a research project investigated the pos- tion rail of the geometry. This curve is generated as a
sibilities and limitations of informing a robotically free-form profile. The idea is to generate both urban
manufactured system with principles derived from furniture as a sitting element and a semi-enclosed
biological structures (Krieg et al., 2011). In parallel space, which may be organized as an installation
to the projects fabricated, a recent research offered unit and shading element. In order to accommodate
a software model, in which material, form, structural these requirements, the profile defines two major
performance and manufacturing constraints are in- curvatures, which define the section of the sitting
tegrated into one system (Yazici, 2013). element with h=0.58 m and section of the semi-en-
Although the methods and tools mentioned are closed space with h=3.32 m. The profile is designed
important in terms of incorporating various design to solve both of these requirements in a seamless
issues, a holistic approach is necessary in order to and continuous way.
integrate multi-performance requirements to the The other construction curve is a piece of a cir-
early stage of the design process. Different perfor- cle which works as a rail. The rail and cross section
mance types, such as structural or environmental rail are used to generate the NURBS surface by using
performance, require different issues to be consid- Sweep 1 tool in Rhinoceros software. The rail can be
ered in the design process. selected as a portion of a global circle with r=14.70
In this paper, a performance based pavilion m, based on the needs and space available in the en-
design is investigated as a case study. The intent vironment to construct the pavilion. Pavilion space
is to create an integrated approach of which form, can be increased or decreased by using same prin-
performance, material and fabrication constraints ciples of the design and construction. This way of
inform each other. By assessing structural and envi- construction is particularly preferred compared to
ronmental performances of the pavilion, a free-form an already completed geometrical composition, in
surface structure, along with a parametrically de- order to maintain flexibility in use of this temporary
fined cladding is designed. structure. Various scenarios can be implemented for
the pavilion, which may enrich spatial experiences
METHODOLOGY of the users.
The applied methods and constraints which influ- The pavilion is considered as a small to mid-
ence the design process are identified. First of all, scale installation unit and the suitable portion of
the pavilion geometry is clearly described in terms the global circle used for the pavilion is 50 deg. The
of its Non-Uniform Rational Basis Spline (NURBS) geometry is generated free-form style by using 20
properties. Following this, performance assess- control points. The approximate dimensions of the
ments, including Solar Access Analysis and Static unit are 9.00 m*8.20 m*4.02 m, which is considered
Structural Analysis, are explored by observing their sufficient enough to be used for an art installation
influence to the overall geometry. Finally, possible (Figure 1).

Figure 1
(a) Free-form geometry of
the pavilion. (b) The overall
geometry is a piece of a global
circle which can be increased
or decreased based on spatial
requirements. (c) The selected
piece which is considered
sufficient enough for an art
installation. (d) Geometrical
description of the pavilion.
Performance Assessments through Static the time period from September to October. The ge-
Structural Analysis and Solar Access Analy- ometry is converted into a mesh, which consists of
sis 1500 objects, in order to undertake the calculations.
The proposed pavilion is located in Antalya Turkey. The orientation and tilt angles of individual objects
Antalya has a typical Mediterranean climate, which are identified. Based on the relationship between
is characterized by warm to hot dry summers and the positioning of one piece and the angle of the
mild to cold wet winters. Because of this reason, sun light, the radiation value of that particular piece
the pavilion has the task of being used as a shading is calculated. Thus, the relationship of the objects
element. Although the pavilion does not obtain a and the sunlight can be established. The total radia-
specific site in the city, it is considered be to be in- tion values ranges from 102992.422 to 871177.938
stalled in north-south or northwest-southeast axes Wh / m2 (Watt hour/ per square meter) (Table 1).
to eliminate the undesirable effect of the sun radia- Through the solar access analysis, it is identified that
tion, through the design of its cladding. the surface pieces which are almost vertical, which
Following generation of the initial form, Solar works as a wall, obtain higher radiation values com-
Access Analysis is operated by Autodesk Ecotect, in pared to the other pieces. Reducing the area of sur-
order to evaluate the total radiation values affect- faces closer to verticality is an important parameter
ing the geometry. Solar access analysis indicates in- in design of the pavilion. The cross section rail of the
cident solar radiation on the surface. The radiation geometry is slightly adjusted through its control
calculations use direct or diffuse radiation data from points to accommodate a better solution for the
the weather file of the city of Antalya, specified for solar access analysis. Therefore, free-form geometry

Solar Access Object Orient. Tilt Total Radiation Total Direct Total Diffuse Table 1
Analysis Radiation Radiation Solar access analysis
ID Type (°) (°) Wh/m2 Wh/m2 Wh/m2 representing attributes of the
0000 Wall -53.32 22.96 750727.000 402232.625 348494.375 objects. The minimum and
0001 Wall -54.11 25.96 791037.375 429239.906 361797.500 maximum total radiations are
0002 Wall -55.05 25.01 804416.250 447105.438 357310.781 identified.
0003 Wall -51.04 20.83 703201.875 363437.031 339764.844
0004 Wall -52.09 23.02 742622.312 394182.375 348439.938
0005 Wall -52.53 22.57 755208.000 406380.375 348827.594
0006 Wall -48.36 19.40 679358.250 343629.219 335729.062
…
0739 Floor 84.37 -62.79 110386.078 230.860 110155.219
0740 Floor -171.70 -69.03 107279.023 52.300 107226.727
0741 Floor -174.91 -67.07 106750.359 200.100 106550.258
0742 Floor -174.07 -72.46 102992.422 0.000 102992.422
0743 Floor -176.76 -70.85 107836.070 34.500 107801.570
0744 Floor -179.47 -75.72 103816.391 0.000 103816.391
...
1481 Wall -53.18 28.27 824098.125 453647.031 370451.094
1482 Wall -54.10 28.76 832552.625 457229.000 375323.656
1483 Wall -54.97 28.07 818652.500 448040.781 370611.719
1484 Wall -55.73 31.58 871177.938 487550.969 383626.969
1485 Wall -27.17 33.95 747919.875 350308.875 397611.031
1486 Wall -54.55 28.96 842044.625 466881.312 375163.344
...
obtain the advantage of responding to the perfor- Comprehensive boundary conditions would
mance issues related to the solar access analysis bet- have a significant impact on the simulation, such as
ter, as well as to the design issues by working both considering the impact of earthquake in the region.
as an urban furniture and a semi-enclosed space. However, the fact that the pavilion is designed to be
In order to assess the structural performance of a temporary and preferably a lightweight structure,
the geometry, Static Structural Analysis by FEM is only the self-weight of the structure is taken into
undertaken with the Scan & Solve plug-in for Rhino. consideration for the analysis, in order to reduce the
The plug-in works with NURBS surfaces without the time of computation and simplify the simulation. By
need for converting the geometry into a mesh, un- the FEM analysis, the problem areas on the geom-
like many other software used for the FEM analysis. etry are identified. Although using different materi-
Following the assignment of material to the geome- als such as stainless steel or aluminum would influ-
try; boundary conditions and loadings are imposed. ence the numerical stress values of the simulation,
Numerical stress values and total deformations on because the boundary conditions are less compli-
the geometry can be identified, by using different cated, in which the geometry is rigidly fixed in three
materials and altering the geometry. By running the positions to the ground, the geometrical properties
simulation for various scenarios; the geometry can obtain the most important role to accommodate the
be to be adjusted based on performance require- structural performance requirements. Especially, the
ments. larger curvature which defines the semi-enclosed

Figure 2
(a) Solar access analysis un-
dertaken for the initial study
of the free-from surface, by
specifying the weather data
of Antalya. (b) FEM analysis is
undertaken for different ge-
ometry options by specifying
material, boundary conditions
and loadings.
space plays the critical role for the overall form with One of the proposed structural elements ob-
its height exceeding 3.00 m and span of approxi- tains varied thicknesses in profile ranges from 0.07
mately 5.40 m. m to 0.40 m in order to increase the strength of the
In order to generate the semi-enclosed space, structure on the necessary parts, such as the larger
a symmetrical geometry, which can be constructed curvature which defines the semi-enclosed space.
by two rail curves, is selected for this study. If there Additionally, in order to obtain a lightweight struc-
is no symmetry; the structure would be unstable, ture, the sections of these structural elements are
require comprehensive structural solutions and ad- enhanced by introducing holes in them (Figure 2).
ditional structural issues may need to be addressed
in the design process, which are considered against Fabrication Scheme of the Pavilion
the design intent. If the section of the semi-enclosed Because the pavilion is a temporary structure, pavil-
space is closer to a circle, then the stresses on the ion pieces are designed to be easily demountable
geometry are equally distributed. By modifying the and light. Geometrical description of form is clearly
cross section rail of the geometry through its control identified for the fabrication purposes. In terms of
points, the numerical stress values can be adjusted. constraints related to the transportation and as-
Altering the geometry has also a significant impact sembly, structural elements are considered to be
on the panel layouts of the cladding, in terms of the fabricated in pieces. Because of this reason, two
panel sizes and numbers material systems are tested for the structure; as steel

and wood. These structural systems are compared The number of major structural elements is 15, used
in terms of their performances, as well as their light- with holes in the sections to reduce weight of the
ness. elements. The spacing between these elements is
One of the options is using steel profiles for the dense, ranging from 40 cm to 92 cm, because their
structure, which is feasible in terms of achieving a thicknesses are considered very slim. There are also
lightweight structural system. Steel tubes are pro- horizontal wooden binding elements, being im-
posed to be bended according to curvature speci- plemented with the same principles as in the steel
fied in the NURBS geometry. The diameter of the structure. Because of the constraints related to the
circular cross section of major structural elements is machine, structural elements are proposed to be
d=10 cm. There are 9 steel tubes which work as ma- produced in pieces. The cladding of the wooden
jor structural elements, positioned according to the structure is 2D CNC cut plywood sheets with thick-
V curves of the NURBS surface. The spacing between ness of 3 mm, which can be used both flat or curved,
these tubes ranges from 70 cm to 176 cm based on because they are flexible to be bended to accommo-
the V curves. Additionally, 21 horizontal steel tubes date curvature on the surface. The maximum sheet
with d=5 cm are used to bind the major structural size of the machine, which can handle sensitive
elements by following the U curves of the NURBS tasks such as highly curvilinear surfaces, is 180 cm
surface. The spacing between these tubes ranges * 300 cm.
from 59 cm to 65 cm based on the U curves. Mate- Although the structural elements made out of
rial proposed for the cladding of steel structure op- steel obtain the advantage of being lightweight,
tion is opaque, acrylic glass with thickness of 3 mm. the fact that bending procedure of steel tubes
The idea is to use flat panels for cladding and mount reflects the quality manual operation, the preci-
them to the curvilinear surfaces. The acrylic pieces sion to the 3D geometry may be problematic. The
are considered to be fabricated by laser cutting ma- wooden structure has the advantage of fabricated
chine, of which size is 80 cm * 60 cm. It is critical how with a small tolerance by the use of CNC machine.
to attach the flat panels to the structure. Therefore, However, the weight of the wooden structure can
a particular detail is developed in which the compo- cause problems. Additionally, because the structural
nents are assembled in groups and then mounted elements made out of wood are considered to be
to the structural elements which define the surface fabricated in pieces, the assembly of the pieces can
curvature. be challenging as well, by maintaining its structural
The other option is using wood as the struc- strength. As a result, both options obtain advan-
ture, which is considered to be manufactured by tages and disadvantages to be considered for the
the Computer Numerical Controlled (CNC) machine. fabrication.
The fact that a complex geometry can be generated In order to create cladding, the NURBS surface
with a number of differentiated and highly articu- is converted into panels by using a parametric defi-
lated components by digital fabrication tools, the nition at the Rhino Grasshopper. The opacity of the
geometry is developed further to accommodate cladding is driven through series of design concerns
the fabrication constraints. The use of digital fabri- and through the intent of controlling the sunlight. A
cation is preferably selected, because of reducing gradient of surface opacity is generated to control
the waste material and obtaining advantages for a sun radiation. It is possible to identify specific ratios
more sustainable design solution. The CNC machine between the opaque and/or transparent areas of
available is able to handle maximum sizes of 200 cm the cladding and total surface area, towards finding
* 400 cm wood sheets, with standard of 5 cm ma- the optimal solution based on climatic conditions of
terial thickness. The micro laminated wood is pro- Antalya. First of all, the surface is subdivided based
posed because of its strength for a structural system. on UV curves on the surface. Following the genera-

Figure 3
The panel layout of the
pavilion is generated para-
metrically by considering the
performance requirements.
tion of individual pieces by UV parameter, the surfac- parts, they are designed to be mounted on feasible
es are exploded. The extracted points are evaluated positions of the actual pavilion structure (Figure 3).
and used for regeneration of the panels. The panels
should be unrolled to be fabricated in 2D by laser RESULTS AND EVALUATION
cutting or CNC machine. The panel sizes range from The proposed method of the performance based
23 cm to 52 cm and the total number of panels are pavilion design exposes how different performance
2000, based on the design intent. A relatively denser requirements can influence a design project. The so-
pattern increases the number of panels by alteration lar access analysis has proven that there is a direct
of the U-V curves of the surface. Although, various relationship between the geometry and the solar ra-
panel shapes such as quadrilateral, triangular or cur- diation. For the given free-form surface, the vertical
vilinear panels can be created via the definition, cur- surfaces gain the most of the sun radiation for the
vilinear panels are selected and developed for the time period from September to October for Antalya.
pavilion, due to its possibility for creating better de- Therefore, the geometry needs to be slightly al-
sign solutions for the opaque and transparent areas tered to reduce those affected areas. Working with
of the cladding. The panels should obtain flexible a free-form NURBS surface enables to modify the
connections to allow movements. The connection geometry through its control points, by maintaining
elements are introduced in four sides of each panel. its coherence and continuity. In addition to the so-
Because the panels are considered as non-structural lar access analysis, static structural analysis by FEM

has indicated the importance of developing the ge- a GH definition, by designing opaque and transpar-
ometry though the requirements of structural per- ent areas on the surface. The proposed method in
formance. The curvature, which defines the semi- this paper integrates form, performance, material
enclosed space, plays the most important role due and fabrication constraints. It exposes how struc-
to the proportions of its height and span. In order tural and environmental performances may inform
to work as an installation unit, shading element and a design project.
urban furniture, the surface profile needs to be con- The intent for the future work is to study the
tinuous and fixed from at least three points to the integration of various performance requirements
ground for being stable. with architectural form further. A model may inte-
The pavilion geometry can be enlarged or grate different types of performance requirements,
shrank, by adjusting the rail curve which is a part form, material and fabrication techniques into one
of a larger circle. This possibility brings also a sys- parametric system, where they are equally impor-
tematic approach towards the construction of the tant and inform each other. This would increase ef-
pavilion. Additionally, the panels of the cladding ficiency in the architectural design process radically.
can be informed by the climatic requirements, thus
their opacity and transparency can be increased or REFERENCES
decreased though the environmental data and solar Goel, V 1992, ‘Ill-structured representations for ill-struc-
radiation analysis. Therefore, the proposed pavilion tured problems’, Proceedings of the Fourteenth Annual
accommodates requirements related to the both Conference of the Cognitive Science Society, Hillsdale,
formal and performance issues. New Jersey. (1992): 844-849.
Although, the proposed method integrates is- Hensel, M and Menges, A 2008, ‘Versatility and Vicissitude:
sues of the free-form pavilion from design towards Performance in Morpho-Ecological Design’, Architec-
construction, additional limitations may be con- tural Design, vol: 78, pp. 54-63.
fronted during its implementation. Holzer, D and Downing, S 2008, ‘The Role of Architectural
Geometry in Performance-orientated Design’, Proceed-
SUMMARY AND CONCLUSIONS ings of Advances in Architectural Geometry, Vienna, pp.
In general practice, performance assessments of a 99-102.
design are usually undertaken in a later stage of the Krieg, OD, Dierichs, K, Reichert, S, Schwinn, T, Menges, A
design process, as an engineering input which influ- 2011, ‘Performative Architectural Morphology: Ro-
ences design radically and reduces the efficiency in botically manufactured biomimetic finger-joined plate
the design process. In order to integrate formal and structures’, Proceedings of the eCAADe Conference Con-
performance issues in an early stage, the design pro- ference, pp. 573-580.
cess of a temporary pavilion is identified. The aim is Pottmann, H, Schiftner A and Wallner, J 2008, ‘Geometry
challenging the conventional form and fabrication of Architectural Freeform Structures’, Internationale
techniques. A free-form NURBS surface is created Mathematische Nachrichten, 209, pp.15 -28.
as an installation unit, shading element and urban Schiftner, A and Balzer, J 2010, ‘Statics-Sensitive Layout of
furniture, located in Antalya, Turkey. Two structural Planar Quadrilateral Meshes’, Proceedings of Advances
systems, as steel and wood, are investigated in or- in Architectural Geometry.
der to accommodate the design requirements of Schlueter, A and Thesseling, F 2009, ‘Building information
the pavilion project. Structural performance and model based energy/exergy performance assessment
environmental performance assessments, including in earlydesign stages’, Automation in Construction, 18,
Solar Access Analysis and Static Structural Analysis pp. 153–163.
undertaken by FEM, are operated in order to inform Yazici, S 2013, A Material-Based Integrated Computational
the geometry. The cladding scheme is generated via Design Model in Architecture, PhD Thesis, Istanbul Tech-

nical University.
Yazici, S and Tanacan L 2012, ‘Material in Performance-
Driven Architectural Geometry’, Proceedings of the As-
sociation of Collegiate Schools of Architecture (ACSA)
100th Meeting, Digital Aptitudes Conference, Massa-
chusetts Institute of Technology, Boston-MA, pp. 266-
273.

Engineering Performance Simulations in Architectural
Design Conception
Atrium in Shenyang: a case study on thermal mass

Michela Turrin1, Ioannis Chatzikonstantinou2, Martin Tenpierik3, Sevil Sariyildiz4
Yasar University, Turkey; Delft University of Technology, The Netherlands, 3Delft Uni-
1,2,4
versity of Technology, The Netherlands

1
michela.turrin@yasar.edu.tr, 2i.chatzikonstantinou@yasar.edu.tr, 3M.J.Tenpierik@tudelft.
nl; 4sevil.sariyildiz@yasar.edu.tr
Abstract. The paper tackles the integration of engineering performance simulations

in the conceptual phase of architectural design, with specific focus on parametric
design processes. A general framework is exemplified, in which the use of performance
simulations and the learning process of the designer are discussed in relation to the
parameterization process. A specific case study is presented more in details regarding the
design of an atrium for the reuse of an existing building in Shenyang-China. Performance
simulations concerning the thermal comfort in the atrium are presented and discussed in
relation to the general framework.
Keywords. Conceptual design; building simulation tools.
INTRODUCTION
Since the requirements on the actual performance grounds and exemplifies the framework, by discuss-
of buildings are becoming ever tighter, accurate ing one specific case study on numerically assessed
data regarding the performance of the buildings is design alternatives for achieving indoor thermal
becoming increasingly important in the early phas- comfort. The analysis of alternative design solutions
es of design. This paper tackles the role of digital is presented by showing the learning process of the
modelling and engineering performance simula- designer through a comparative study. One chosen
tions in the conceptual phase of architectural de- alternative is then presented in details, by under-
sign. The first part of the paper focuses on a theoret- taking the integration of parametric modeling and
ical framework for performance oriented parametric performance simulations during the design process.
design, in which the design process is decomposed The parameterization process of the design concept
into and related to the design knowledge available is discussed based on the analysis previously illus-
during the design conception and its parameteri- trated; focusing on design innovation, emphasis is
zation process; moreover, this part describes some given to the importance of extracting knowledge
general case studies. The second part of the paper from the numeric analysis.

DESIGN PROCESSES TOWARDS INNOVA- theorems, respectively intended as design knowl-
TIVE DESIGN SOLUTIONS edge and properties of other design solutions. Spe-
cifically, following Roozenberg’s (1993) distinction,
Background theories ‘abduction’ in design theory and knowledge-based
Design processes towards innovative design solu- design systems is explanatory abduction while the
tions have been tackled and theorized from a num- reasoning towards new solutions for design prob-
ber of different perspectives. Geoffrey Broadbent lems follows the pattern of innovative abduction.
(1969) refers to four types of design methods, which
he calls pragmatic, iconic, analogical and canonic. Integration of engineering disciplines and
Pragmatic design makes use of available techniques Performance Simulation Tools
without relevant innovation; iconic design recalls ex- Within the broadly theorized field described above,
istent solutions and tends to replicate them; canonic focus is given here on the integration of engineering
design relies on rules and regulations as guidelines; disciplines in the conceptual phase of architectural
analogical design makes use of analogies with other design. Their use to trigger the design creativity is
fields to define new ways for structuring the prob- approached in opposition to post-engineering pro-
lems and their solutions. While all these four meth- cesses. In traditional post-engineering processes,
ods can be used to generate design alternatives by technical performances are mostly considered and
exploring various concepts, it is especially the last verified in late stages; the design variations eventu-
one that allows for major innovation. It is widely ac- ally necessary to satisfy the technical requirements
knowledged that looking for innovative solutions are tailored upon preconceived and constraining
for new design concepts deeply relies not only on architectural designs. In contrast with this attitude,
the previous experience of the designer, but also on the use of engineering performances is proposed
his/her real time learning process. The importance in order to inspire or even drive the concept im-
of prestructures, presuppositions or protomodels provements or the generation of new alternative
as the origins of solution concepts (Roozenburg and concepts; this implies that engineering feedbacks
Cross, 1991) is recognized, but leads to an evolving are an integral part of the analogical method and
design path in which the learning process is an inte- a support for innovative abduction. Aiming at this,
gral part of the exploratory design activity. In a puz- building Performance Simulation Tools and their use
zle-making approach (Alexander et al., 1977), de- in the early phase of the design play a crucial role.
signers begin with a kit of forms, including materials This perspective is in line with a number of previ-
and shape, subject to modification according to cer- ous and well-known studies, such as the ones of
tain rules until they achieve some desired functional Mahdavi and Lam (1991), according to whom sys-
qualities; inductive reasoning is used with the aid of tematic “front-end” studies based on digital simula-
metaphors, symbols, and case studies (Kalay, 1999). tions to aid preliminary design decisions should be
Analogical reasoning implies learning from previous preferred over the traditional approach, in which
or other problems similar to the actual problem by the role of building simulation is relegated to the
retrieving and transferring chains of reasoning and “back-end” of the design process. The use of feed-
knowledge to the actual problem (Veloso, 1994); it backs from analysis software to re-evaluate design
is quite beneficial to problem solving processes in- decisions is also emphasized by Caldas and Norford
cluding design (Goldschmidt, 2001; Goldschmidt (2003), who point out that ‘by using simulation tools,
and Smolkov, 2006). A number of design methods it is possible to engage in a design practice based on
are based on abduction (Tomiyama et al., 2003), us- feedback loops between making design decisions
ing logic and abductive reasoning; according to this, and evaluating their environmental impact, as a way
a design solution is defined by means of axioms and to inform the on-going process of design. However,

the view proposed here differs from previous works lution-assessment phases. The third includes design
due to its focus on the use of numeric design assess- processes in which a clear (mostly bijective) relation
ments as part of the learning process of the design- between geometry and performance can be set dur-
er, to achieve innovative design solutions. ing the parameterization process; this allows con-
sistently relating different geometric solutions with
Design knowledge acquisition in paramet- different performance requirements, which leads to
ric design bijectively deterministic parametric solution spaces;
In a previous publication, the authors proposed and, unless knowledge is already available, it implies
a parametric design framework for performance- intense use of numeric assessments in the strategy-
oriented design, in which the use of numeric design definition phase. A substantial difference across the
assessments are related to the learning process and three cases consists of the way in which the initial
knowledge available or generated during the de- design concept (here named primary generator ac-
sign process. Three phases are distinguished in the cording to Darke, 1979) is conceived in relation to
parametric design development. During the first the considered performances. Numeric design as-
phase, strategy-definition, the parameterization is sessments are considered a means for extracting
addressed based on the analysis of design challeng- knowledge to be used (or re-used) in the conception
es; during the second phase, model-building, the of (new) primary generators.
parametric model is constructed; during the third Four examples are mentioned here following;
phase, solution-assessment, the design alternatives additional details can be found in previous publica-
embedded into the parametric model are explored tions. The first example (Turrin et al., 2013) concerns
based on performance evaluations (Turrin et al., the design of an envelope controlling effects of di-
2013). Numeric design assessments play a crucial rect and indirect daylight in the interior space; it was
role both in the first and in the third phase. Accord- developed by a student (Friedhoff Calvo, 2010). The
ing to this framework, three general types of pro- primary generator was developed based on Escher’s
cesses are distinguished, in which the solution space tessellations, with intuitively defined modular varia-
of the parametric models is differently set. This usu- tions from permeable to impermeable to daylight.
ally occurs according to the knowledge the designer In order to explore the daylight effects in alternative
has or gains before or when defining the parameter- designs, the geometry of the primary generator was
ization strategy, in respect to a set of selected perfor- parameterized. Considering the intuitive nature of
mances. The first includes design processes in which the design, the parameterization aimed at a broad
little knowledge is available during the parameteri- solution space, to reduce the risk of excluding mean-
zation process, with consequent need of enlarging ingful design alternatives. As a consequence, further
the design solution space for broad performance computational support (i.e. search algorithms) was
explorations. This leads to large parametric solutions needed in combination with performance simula-
spaces; and usually implies intense use of numeric tion tools during the solution-assessment. Focus-
assessments in the solution-assessment phase. The ing on the learning process of the designer, the use
second includes design processes in which relevant of search algorithms is addressed in the following
knowledge is available during the parameterization example. The second example (Turrin et al., 2011)
process, with consequent chance of bounding the concerns the design of an envelope that reduces
solution space into a more confined collection of the solar gain but allows a high daylight level. The
alternative design solutions. This leads to narrowed primary generator was developed based on well-
parametric solution spaces; and, unless knowledge known principles of shading and orientation, but
is already available, it implies some use of numeric applied on complex geometry. The geometry of
assessment both in the strategy-definition and so- the primary generator was parameterized in order

to lead to a large solutions space; and a search for With reference to this framework, the follow-
well performing solutions was performed based on ing section presents in detail one case study from
a genetic algorithm optimization, in combination a practice-based design process for an atrium. The
with performance simulation tools. The generated performances considered in this project focus main-
solutions were stored in a database and analyzed in ly on passive climatic control.
order to extract information from badly performing,
sub-optimal, and well performing solutions, aiming STUDIES FOR AN ATRIUM IN SHENYANG
at an explicit understanding of trends between geo- The atrium is part of a larger project developed by
metric design variables and resulting performances, GWS, a company located in Beijing. The project con-
toward design knowledge generation. The third sists of the conversion of a tobacco factory into of-
example (Turrin et al., 2013) concerns the design fice buildings, organized in three blocks around a
of an envelope to control the daylight effect on the courtyard. The atrium is located in one of the three
enclosed spaces; it was developed by a student for buildings, developed along an East-West axis, on the
his M.Sc. graduation project (Van Kersbergen, 2011). northern side of the plot. The building is organized
The primary generator was developed only after in five floors and has a total volume of approximate-
an extensive number of preliminary performance ly 130.000 cubic meters. GWS developed a number
simulations on different basic primary generators. of design alternatives, in most of which the atrium is
The geometry of the chosen primary generator located on the top two floors and occupies a volume
was parameterized based on the results of the pre- of approximately 8.000 cubic meters. The spaces
vious analyses, in order to lead to a narrowed solu- around the atrium are mainly offices or flexible ar-
tions space; and, during the solution-assessment, eas, for which the atrium acts as a distribution space.
performance simulations were run only on chosen The work presented in this paper is a part of the out-
design alternatives. Based on the increased corre- put of a collaboration between GWS in China and an
spondence between the actual and desired solu- interdisciplinary team at TUDelft, in the Netherlands.
tions space, as it was expected, the chosen options The collaboration assumed the general setting of
showed performances quite close to the desired the overall project as given, while focusing on the
requirements. This attitude towards amplifying the atrium and related roof. A number of design options
learning process (by means of numeric assessments) were developed, by considering performances for
before parameterizing and even conceiving the pri- passive climatic comfort and, in general, reduction
mary generator is shown in its extreme consequenc- of energy consumption during use.
es in the fourth example. The fourth example con- The following sections present the preliminary
cerns the design of an acoustic absorber which was numeric analyses run on the building, based on
developed by a student for her M.Sc. thesis (Setaki, which challenges and potentials to reach the design
2012). Intensive work was invested in performance goals were identified. Based on these results, spe-
measurements of samples, which not only increased cific sub-goals were established, which decompose
the design knowledge, but also formalized it. Only the design requirements into more specific tasks.
when a clear relation was formalized, the primary
generator was conceived. A parametric model was Strategy-definition phase: preliminary
made based on the formalized relation, in order to performance simulations
bijectively relate specific acoustic requirements with Shenyang is located in the fifth level of Chinese cli-
correspondent geometric design alternatives. So far, mate zones, defined as “coolest level”; within this
this case showed mostly full coincidence between level, the area belongs to the class B, which corre-
the actual and the desired solution space of the par- sponds to the most moderate class of the “coolest
ametric models. zone”. According to the Shenyang IWEC weather sta-

tistics, the winter peak happens between December and virtual thermal zones, 4 of which regarded the
and February, with the coldest hour at 5am, typically atrium; these latter are named 4a and 4b for bottom
below -10 degrees Celsius; the summer peak hap- and top part of the atrium on the fourth floor; and
pens between June and August with temperatures 5a, 5b respectively on the fifth floor.
generally above 25 degrees Celsius at 2pm.
The work presented in this paper focuses on Results
thermal comfort, and specifically on passive meas- The results showed that higher insulation results in
ures for achieving thermal comfort; while consid- higher indoor temperatures both in winter and in
erations on daylight are taken into account as side summer; and higher leakage implies lower indoor
criterion only. A number of preliminary numeric summer and winter temperatures. The effects of
analyses were run on the given building, in order occupancy, internal heat loads, increased insula-
to identify expected problems and potentials for tion and increased air tightness were expected to
passive climate comfort. Simulations of thermal be beneficial in winter, and unfavorable in summer.
comfort based on Predicted Mean Vote (ASHRAE, Summer thermal comfort increased when including
2010), and of air, mean radiant and operative tem- ventilation. Both for July and January (daily values),
peratures were performed on the whole building a comparison was also made in case of additional
for both a whole year, and with focus on periods in thermal mass distributed on the floors surrounding
which worst conditions occur for risk of overheat- the atrium. The thermal zone corresponding to the
ing (July) and coldness (January). Simulations were bottom area of the atrium was obviously the one af-
run both in free-running-mode (without mechani- fected the most by the effect of thermal mass, since
cal heating and cooling) with no occupancy and no it lies on a floor, differently than the other three
internal heat loads (in order to measure the effect thermal zones. Some of the results are summarized
of the building only, for passive thermal comfort); in Table 1. In addition to the comparison between
and by including HVAC systems, occupancy and operative temperatures, relevant information was
internal heat loads. Design Builder (DesignBuilder extracted also from the analyses of air and radiant
Software Ltd) was selected as building performance temperatures; and from the behavior of the PMV,
simulation tool. Moreover, daylight conditions were especially on the bottom thermal zone. In this zone,
studied for the floors where the atrium is included, in case of little thermal mass, the PMV varied from
at equinoxes and solstices, using Radiance via Diva 2.8 (2nd July) to 4.9 (18th July); in case of additional
for Rhino. Regarding the passive thermal comfort, thermal mass, the PMV varied from 1.7 (4th July) to
digital simulations were systematically run on a 3.9 (24th July), which clearly showed the delaying
set of different variations concerning several mate- and peak-shaving effect of the thermal mass. Finally,
rial properties of the external walls, roof and glazing a simulation was run adding the effect of thermal
(different levels of insulation); air tightness of the mass and natural ventilation (5 ac/h), showing ad-
building; and thermal mass. Insulation levels (U-val- ditional benefits. As an example, in the bottom ther-
ue) varied from 0.35 to 0.25 W/(m2∙K) for the exter- mal zone the PMV varied from 0.3 (4th July) to 2.2
nal walls, from 0.25 to 0.15 W/(m2∙K) for the flat roof (18th July). Finally, a series of shadow analyses were
and from 1.978 to 1.415 W/(m2∙K) for external glaz- made, which pointed out correspondence between
ing; air tightness varied from 0.7 to 0.2 ac/h; differ- solar gains and temperatures.
ent thicknesses of the floor determined the thermal
mass, in heavy concrete; some options were tested Conclusions and specific sub-goals
also with natural ventilation. The building was mod- According to the preliminary analyses, the whole
eled based on its external envelope, subdivision into building and the atrium especially had critical ther-
floors and atrium. The model consisted of 34 real mal discomfort both in winter and in summer. The

U-value Air t. Vent. Th.M. Min. Win. temp. Max. Sum. temp. Table 1
(W/(m2∙K)) (ac/h) (ac/h) (Cº) (Cº) Minimum Winter and
Wall Roof Glazing 4a 5b 4a 5b Maximum Summer operative
0.35 0.25 1.978 0.7 0 No -14.9 -24.1 36.4 50.3 temperatures (temp.) in vari-
0.25 0.15 1.415 0.7 0 No -8.3 -21.4 41.9 53.0 ants for insulation (U-value),
0.35 0.25 1.978 0.2 0 No -8.2 -21.1 42.0 53.3 air tightness (Air t.), natural
0.25 0.15 1.415 0.2 0 No -8.2 -21.0 42.0 53.3 ventilation (Vent.), and addi-
0.25 0.15 1.415 0.2 5 No 33.8 46.8 tional thermal mass (Th.M.).
0.25 0.15 1.415 0.2 10 No 33.7 45.7
0.25 0.15 1.415 0.2 0 Floors -4.5 -20.4 38.8 53.0
analyses showed also that it is possible to reduce evant part of the strategy definition phase focused
thermal discomfort by means of passive strategies, on thermal mass. The following sections summarize
both in summer and in winter. Specific sub-goals its main aspects.
were identified. Considering the local climate, cali-
brating the design first based on the cold winter Additional analysis on thermal mass
period was recommended. This clearly included in- A set of additional analyses were carried out re-
creasing the insulation, air tightness and solar gain garding the effects of quantity and distribution of
of the building as much as possible. However, this thermal mass within the atrium. The effect of dif-
challenged summer thermal comfort. As also con- ferent distributions of additional thermal mass was
firmed in the preliminary analyses, thermal mass analyzed for four vertical (virtual) thermal zones of
and summer ventilation positively impacted sum- the atrium, with and without natural ventilation and
mer comfort. Among these factors, the work illus- shading. Among the analyzed options, the one with
trated in the following sections focuses on the dis- external shading, diurnal and nocturnal ventilation
tribution of thermal mass, natural ventilation and (10ac/h), and higher concentration of thermal mass
shading, since these factors highly depend (also) on the top part of the atrium showed the best per-
on the geometry of the overall spatial configuration formance for summer thermal comfort. The results
of the atrium. Specifically, investigations on ther- are visible in table 1 and clearly show the accumu-
mal mass were taken as starting point for the next lation of heat in the thermal mass and the cooling
phase of the strategy-definition phase, in which the effect of ventilation, as well as the reduction of over-
parameterization strategy was more specifically ad- heating through the addition of external shading on
dressed. the glazed roof. Additional tests were run accentuat-
ing the uneven distribution of thermal mass across
THERMAL MASS AS DESIGN DRIVER the levels. These analyses showed that additional
The principles described above were investigated as thermal mass on the top level leads to beneficial ef-
design drivers, by making use of digital simulations fects, while changes in the bottom level had minor
to study their thermal behavior in conjunction with effects on the thermal performances. Since mini-
the design exploration of a large range of design mizing the use of additional material and structural
possibilities. Especially when considering the di- load is generally desirable, the option of reducing
mensions of the atrium and its value as representa- the additional thermal mass on the bottom level
tive space for the new office building, conceiving and distributing it more on the top level was used
such a thermal system with emphasis on its iconic for further investigations. External shading further
value (in addition to its technical thermal function) reduced the maximum temperatures as can be seen
was proposed as beneficial for the project. A rel- from Table 2.

Table 2 ventilation shading thermal mass Max. operative temperature (deg. C.)
Maximum summer operative 4a 4b 5a 5b
temperatures of simulated no no no 42.0 43.5 48.5 53.4
variants for ventilation, no no floor 5b 41.8 43.2 47.8 48.1
shading and thermal mass no no floor 5a, b 41.6 45.4 57.2 49.7
distribution in the atrium. no no floor 4b, 5a, b 41.6 49.2 57.4 49.9
no no floor 4a, b, 5a, b 41.6 50.3 59.6 51.0
10 ac/h no floor 4a, b, 5a, b 34.8 34.8 39.6 41.1
10 ac/h yes floor 4a, b, 5a, b 33.8 32.8 35.3 41.1
10 ac/h yes floor 4b, 5a, bx2, 34.3 32.5 34.8 38.3
Based on the preliminary analyses, geometric the areas irradiated in summer were distributed
properties were extracted for the aspects having along all the levels of the atrium on its north, east
positive impact on the design goals; for different pri- and west sides; while the areas of the atrium irradi-
mary generators, the attributes of these geometric ated in winter were located on the north side of the
properties were parameterized in order to investi- top level of the atrium only. These latter areas were
gate geometric alternatives. Examples are provided therefore chosen for distributing the thermal mass.
in the following section. The other criteria were addressed within the subdo-
mains of this design space (detailed arrangement,
Primary generator and parameterization form, material and construction of the system),
process based on the absence of significant degrees of con-
Focusing on the satisfaction of the primary goal of flict with the main objective (thermal performance).
the design at hand (namely the improvement of the Among the explored directions, one is exemplified
thermal performance of the atrium), the numeric here following, in which a set of sliding panels was
analyses described above enabled the quantifica- proposed for the atrium; this resulted in a set of ver-
tion of a suitable distribution of thermal mass across tical panels in concrete, anchored along the north
the vertical levels of the atrium. This information al- side and the top part of the south side. In this design
lowed to identify a first numeric rule based on which option, the effect of thermal mass was focused on
geometric options were to be designed. Various pri- the diurnal fluctuations, leading to an active thick-
mary generators and related parameterization pro- ness of 10 to 15 cm for concrete. Considering that at
cesses were developed to explore different design the back of a 5 cm thick concrete panel the fluctua-
directions responding to this rule. Within the bound- tion is 72% of the fluctuation at the front and at the
aries of this rule, additional aspects were considered back of a 10 cm panel it is 51%, the need of releasing
in order to enhance the thermal benefits and to in- heat toward the back areas was to be addressed. In-
clude other criteria, such as structural performance stead of rotating the heavy panels, fixed panels were
and daylight. The primary generators were deve- combined with sliding thermal insulation to prevent
loped considering the thermal benefits of exposing nocturnal release of accumulated heat toward the
the mass to winter solar radiation and protecting it atrium; and to favor the thermal behavior at the
from the summer one. Additionally, they were deve- back of the panels. Figure 1 illustrates the principle.
loped considering that the heat accumulated during Given the suitable distribution of thermal mass
the winter days from the atrium should be released across the vertical levels, the general layout of the
toward the surrounding areas (back areas), which is panels was treated as a layout problem, in which the
where the thermal benefits are especially required. requirements for mass distribution may correspond
Based on a shadow analysis in Ecotect (Autodesk), to several panel layout solutions. A parametric

model was established in order to investigate layout Figure 1
alternatives, both by decreasing the dimensions of Combination of heavy panels
the panels from the topmost to the lowermost part and sliding insulation to
of the atrium, and by reducing the amount of pan- enhance the back diurnal fluc-
els in the same fashion. The latter scheme was pre- tuation in the thermal mass.
ferred since it allowed for mostly uninhibited access
to the atrium floor. In this option, the parameteriza-
tion included the number and the dimensions of the
panels (and therefore also affected their intervals),
by generating a narrowed solution space. Figure
2 schematizes this option (and suggests possible
alignment of the vertical panels with the structure
of the roof ).
A separate parametric process regarded the
form of the panels. In this case, the parameteriza-
tion aimed at a large solution space, later explored
with the support of genetic algorithms during the
solution-assessment phase. A number of require-
ments were specified with regard to thermal, func-
tional and structural performance. Specifically, the
total thermal mass should approximate the distri-
bution that resulted from the thermal calculations
and the panels of the topmost floors should be
exposed as much as possible to the south, so as to solution-assessment, the parametric model embed-
receive adequate sun radiation during wintertime. ded finite element calculations (via Karamba3d), in
Moreover, the form of the panels should be such as order to obtain data about the stresses and displace-
to allow for the sliding insulation panels to slide in ments; simple geometrical operations were used to
front and behind the thermal masses. Finally, given evaluate functional adequacy and exposure to solar
that the panels covered the full height of the atrium radiation; and, a multi-objective genetic algorithm
and were anchored to the building structure in lim- was used to search for non-dominated solutions.
ited locations, efficient distribution of loads should Given that the objective functions are conflicting, a
be achieved, so as to minimize deflections resulting set of non-dominated solutions was obtained. Se-
from their own weight as well as from occasional lection among them was performed so that the se-
horizontal loads. A number of geometric properties lected one would perform adequately with respect
affecting these requirements were parameterized;
and a multi-objective optimization problem with Figure 2
three objectives and two constraints was formu- Layout of the vertical concrete
lated. The objectives were: the approximation of panels (thermal mass).
the calculated thermal mass distribution; the maxi-
mization of surface exposure to the south and the
minimization of deflection under several load-cases.
The constraints regarded the suitability of the shape
for sliding panels and their curvature (for fabrica-
tion considerations). In the preliminary stage of the

Figure 3 dents (Marina Stavrakantonaki and Itai Cohen) are
Example of curvature resulting acknowledged, especially regarding the design for
from parametric studies of the daylight; as well as the advice and guidance of Prof.
concrete panel. Arjan van Timmeren.
REFERENCES
Alexander, C., Ishikawa, S., Silverstein, M., Jacobson, M.,
Fiksdahl-King, I., Angel, S., 1977, A Pattern Language,
Oxford Univ. Press.
to all goals, as well as according to aesthetic prefer- ASHRAE, 2010, ASHRAE Standard 55 Thermal Environmental
ence. Figure 3 exemplifies the panels. Conditions for Human Occupancy, ASHRAE Atlanta.
Broadbent, G., 1969, Design Method in Architecture, in De-
CONCLUSIONS sign Methods in Architecture.
The paper presented the studies for an atrium in Caldas, L.G., Norford, L.K., 2003 Genetic algorithms for op-
Shenyang, for which a number of design proposals timization of building envelopes and the design and
were developed based on performance-oriented control of HVAC systems. Journal of Solar Energy Engi-
parametric investigations. The process was exempli- neering, 125(3), pp. 343-351.
fied according to a parametric framework in which Darke, J., 1979, The Primary Generator and the Design Pro-
aspects affecting the thermal behavior of the atrium cess. Design Studies, 1(1), pp. 36-44.
were discussed as design drivers. The process in- Friedhoff Calvo M.A., 2010, ‘Investigations on a Parametric
cluded an extensive number of performance simu- Double Component Structure and its design oppor-
lations, whose role regarded both the strategy-def- tunities in the field of natural illumination’, Report for
inition phase and the solution-assessment phase. Design Informatics Study, MSc Course 2010, TUDelft.
Larger emphasis was given to the strategy-defini- Goldschmidt, G., 2001, Visual analogy – a strategy for de-
tion phase, in order to highlight the relevance of sign reasoning and learning. In: C. Eastman, W. News-
preliminary knowledge. Additionally to this aspect, letter and M. McCracken, Editors, Design knowing and
a conclusive remark is proposed on the crucial role learning: cognition in design education, Elsevier, New
played by performance simulations in enhancing York, pp. 199–219.
the interdisciplinarity of the process, also by height- Goldschmidt, G., Smolkov, M., 2006, Variances in the impact
ening the brainstorming across the various disci- of visual stimuli on design problem solving perfor-
plines involved in the design process. mance, Design Studies, 27(5), pp. 549-569.
Kalay Y.E., 1999, Performance-based design. Automation in
ACKNOWLEDGEMENTS Construction, 8(4), pp. 395-409.
The project was initiated as part of the Urban Knowl- Mahdavi, A., Lam, KP., 1991. Performance Simulation as
edge Network Asia (UKNA); as such, it benefitted of a Front-end Tool for Integrative Conceptual Design
a Marie Curie Actions International Research Staff Evaluation. In: Proceedings of the 2nd World Congress on
Exchange Scheme (IRSES) grant, as part of the Euro- Technology Improving the Energy Use, Comfort and Eco-
pean Union’s Seventh Framework Program. The hos- nomics of Buildings.
pitality given by the Beijing University of Technol- Roozenberg, 1993, On the Pattern of Reasoning in Innova-
ogy and the collaboration with the staff members of tive Design, Design Studies, 14(1).
Green World Solutions in Beijing were crucial. Spe- Roozenburg, N. F. M., Cross, N. G., 1991, Models of the de-
cial acknowledgments are given to Xiao Zhongfa, sign process: integrating across the disciplines, Design
GWS responsible for the re-design of the tobacco Studies, 12(4), pp. 215-220.
factory. Contributions by former TUDelft MSc stu- Setaki, F. 2012, ‘Acoustics by additive manufacturing’, MSc

Thesis, TUDelft. Turrin, M., Stouffs, R., Sariyildiz, S., 2013, Performance-Based
Tomiyama,T, Takeda, H., Yoshioka, M., Shimomura, Y, Abduc- Parameterization Strategies – A Theoretic Framework
tion for creative design, ASME 2003 — Design Engineer- and Case Studies, Proceedings of CAADRIA 2013, pp.
ing Technical Conference and Computers and Informa- 519-528
tion in Engineering Conference, EEUU, Chicago. Van Kersbergen, D.J. 2011, ‘Schiphol Interchange Station -
Turrin, M., von Buelow, P., Stouffs, R., 2011, Design explora- Integrated design research for the wind and daylight
tions of performance driven geometry in architectural performance of the building envelope’, MSc Thesis,
design using parametric modelling and genetic algo- TUDelft.
rithms. Advanced Engineering Informatics, 5 (24). Veloso, M., 1994, Planning and learning by analogical reason-
ing. Springer-Verlag.

Passive Energy Devices in Ceramics
A study in slip casting toward sweaty, scaly buildings

Mark Weston1, Dan Greenberg2
University of South Florida School of Architecture and Community Design, College of
The Arts, Tampa Florida, USA
www.arch.usf.edu
1
MW@usf.edu, 2 dgreenberg@mail.usf.edu
Abstract. Buildings with scales, buildings that sweat: this paper proposes two strategies
for a materially grounded, performance-based architecture which leverages the strengths
of computation and CNC fabrication against the basic properties of traditional ceramics.
Keywords. Building performance, CNC tooling, computer aided manufacture, ceramics,
passive energy design
TWO APPROACHES
Two prototypes for passive, energy saving devices warm and dry; rainwater is typically displaced from
involve the use of ceramics to create high perfor- the building footprint and channeled away with gut-
mance building envelopes. One strategy uses the ters, swales, and ultimately retention or detention
tendency of porous ceramic materials such as clay to strategies. Recent popular attention to sustainable
wick moisture to the exterior of a building causing building has increased the use of cisterns to collect
it to “sweat” and thus to cool itself passively in dry and reuse water, but these uses remain relegated
climates. A second strategy uses bi-colored ceramic to watering landscapes and flushing toilets. Biomi-
“scales” to create an array of solar collector/diffus- metic strategies, however, have been proposed to
ers which can be used to shuttle heat energy either mimic beneficial cooling effects caused by sweating
into our out of a building. Both strategies take a cue in mammals (Lilley, et al., 2012). Prototypes are un-
from systems in nature to leverage a material-based derway for an envisioned “Sweaty Façade” which will
strategy for thermoregulation in buildings. These use the natural osmotic characteristics of clay to al-
devices are the result of computer modelling and low buildings to sweat, thus taking advantage of the
CNC fabrication to mill positive forms for two types heat of evaporation of water to passively cool build-
of plaster mold making for ceramic slip casting. One ings in warm, dry climates. Water collected at the
technique is industrial, using pressurized multi-part roof can be used to fill façade components which
forms, while the other method uses traditional run- will sweat to save energy.
through slip casting molds. These intentionally wet façade modules will
take forms characterized by highly textured or fold-
The Sweaty Facade ing surfaces to create large surface areas for the
The built environment as we know it is characterized evaporation of water. Using simple parametric rep-
by constructions which function to keep occupants etition in Rhino/Grasshopper, along with Kangaroo

Figure 1
Multi Part Mold.
Figure 2
Sweaty Façade Components
showing large surface area.
for shape optimization, prototype forms were then cade based on repetitive ceramic elements. These
carved from polystyrene on a three-axis CNC router. elements are spaced to allow for air flow necessary
Such complex objects present a distinct difficulty for for evaporative cooling between individual units,
traditional casting techniques. The more complex a while the size and shape of the elements creates
desired form becomes, the more complex the form- maximal surface area. The spacing and shape of the
work. Multi part formwork in ceramics is used tradi- units is designed to minimize „slow“ spots in the air
tionally to cast parts with large undercuts, or with flow, but also to provide for diffused daylight to pass
surface area large enough to create impossible sce- through the screen to the building interior beyond.
narios for demolding due to friction (Reijnders and
EKWC, 2005). Scaly Exteriors
The creation of these highly articulated façade The “bubble tile is envisioned as a ceramic heat
prototype units was approached through a tradi- exchange component which has a rough, darkly
tional mold making technique. A twelve part plaster colored surface on one side, and a smooth reflec-
mold was made by hand, including captured interior tive surface on the other side (Figure 4). Tiles can
pieces held by ties piercing larger parts. The size of be used alone, or in conjunction with a liquid heat
the pieces required a so-called “run-through mold” transfer system to move heat energy through an
due to the sheer weight of the plaster. In order to array of tiles. In an architectural building façade,
empty the mold of liquid clay, a drain is placed at the heat exchange tiles can be placed in a mechanism
bottom, which when opened allows remaining clay which will allow the tiles to reverse front to back.
to drain from the form (Figure 1). This can allow infrared energy to be either reflected,
The resultant parts were deliberately designed absorbed, or diffused as necessary for environmen-
to maintain the lines left behind by the seams of the tal conditions in order to reduce heating and cool-
mold in order to give clues to the making of the ob- ing loads on a building. By motorizing each tile, an
ject (Figure 2). The large surface areas exhibited by array of heat exchange tiles can be used as a solar
these complex forms are impossible to accomplish collector, heat exchanger, as a light shelf, or as sig-
using non-sculptorly techniques. Non-developable nage. The technology takes advantage of the ther-
surfaces, in other words, must be carved or cast (Du- mal properties of ceramics to modulate the heating
arte, 2004). and cooling loads on buildings in various climates.
A rendering of an architectural corner condi- The function of the façade system is inspired by the
tion against the sky (Figure 3), shows a sensual fa- Namaqua Chameleon (Benyus, 1997) which changes

Figure 3
Sweaty Façade Architectural
Corner Condition. Rendering:
Dan Greenberg.
color, depending on its needs, to either reflect or casting. The scale of the surface articulation is such
absorb the heat of the sun in the harsh and widely that traditional mold making techniques become
varied temperatures of the desert. Following this impractical; a multipart mold for this tile would con-
biomimetic strategy, the façade system will be pro- tain many thousands of parts. The bumps on the
grammable to alter its orientation to the sun based surface are pyramidal in shape, so designed as to
on material color, climate, and the needs of building avoid micro shading of the component surface. In
occupants. terms of the intricacies of mold making, this is the
The creation of the positive forms for the bub- perfect shape for demolding, as it offers no under-
ble tiles relied on a simple parametric box-morph cuts to impede mold removal. Unfortunately, the
repetition of pyramidal forms over a simple shape bumps create such a massively increased surface
using Rhino/ Grasshopper in order to achieve a de- area that the force of friction between the part and
vice with maximal surface area on one side without the mold makes removal of the part impossible. Ad-
creating micro shading conditions. The shape of the ditionally, large flat, hollow pieces such as this tend
Bubble Tile poses an interesting problem for slip to collapse in the mold from the weight of unsup-
ported wet clay. To solve these issues, an industrial
Figure 4 slip casting technique was adopted.
Scaly Façade Component In the creation of the four part mold, a manifold
showing highly textured of perforated air tubing was embedded in side of
surface. the mold corresponding to the highly textured sur-
face of the part to aid in demolding (Figure 5). The
mold is also pierced at the end by a plastic tube for
pressurizing the interior of the part to stop it from
collapsing.
The mold is filled with clay, and after sufficient
thickness has developed in the interior of the mold,
the remaining liquid clay is emptied. Immediately,
the interior of the mold is pressurized through a
short tube in the cap of the mold. Air pressure forces

Figure 5
Perforated tubing laid prior
to casting.
Figure 6
The highly textured mold
interior.
the clay against the sides of the mold allowing the CONCLUSION
clay to harden without collapsing. As water is slowly The incredible complexities offered by the possibili-
absorbed from the clay into the plaster, the part be- ties of advanced computation create opportunities
comes “leather hard” and is able to support itself. At for new advances in building performance. These
this point, the interior pressure is released. In order complex forms, however, present unique challeng-
to now demold the part, three bars of pressure is es to traditional forms of manufacture for materi-
pumped into the perforated tube in the plaster. This als such as clay. A hybrid approach to the creation
air is forced through the pores in the plaster mold, of complex ceramic parts leverages traditional and
causing water absorbed by the plaster to diffuse industrial techniques to produce manufacturing
outward, ultimately creating a molecular mist of wa- possibilities suitable for industry. This approach can
ter at the exterior surfaces of the mold. This water, allow designers to maximize material performance
extruded at the interior surface of the mold (Figure previously inaccessible in traditional materials, while
6), creates sufficient lubrication against the rough simultaneously tapping into otherwise economi-
surface of the part to allow smooth part removal cally unfeasible material palette which, while firmly
without breakage. rooted in a progressive digital materiality, neverthe-
On a mockup of an architectural facade, (Figure less recollects a hand-made past. Furthermore, pro-
7), tiles are arranged horrizontally on the southern gressive strategies for performance based design
side of a building, and vertically on the east/west need no longer be the static fixed elements of our
sides of the building. This redering places the con- design past; instead, using biological models as a
cept in a generic glass box facade folly representa- platform, architects have the opportunity to create
tive of a default retail or office condition suggestive buildings which sweat, change color, or otherwise
of a energy sensible retrofit to an existing building. adapt to their immediate environment with biologi-
Tiles would be operable, allowing for maximization cal precision.
of absorbtion, reflection, or diffusion of heat energy,
and to admit daylight and allow views as desired. ACKNOWLEDGEMENTS
The ceramic character of such a facade would allow Special thanks to The University of South Florida
for the creation of an architectural space reliant on College of The Arts and the European Ceramics
rich materiality, while simultaneously providing a re- Workcentre for generous funding and support of
gionally adaptable high performance building. this project.

Figure 7
“Scaly” Façade over a generic
glass retail box. Rendering:
Dan Greenberg.
REFERENCES
Benyus, J. (1997). Biomimicry: Innovation Inspired by Na- tion: Multi-Scalar Development of Ceramic Material.
ture. New York: Morrow. San Francisco: ACADIA 12: Synthetic Digital Ecologies
Duarte, J. P. (2004). Free-form Ceramics - Design and Pro- [Proceedings of the 32nd Annual Conference of the As-
duction of Complex Architectural Forms with Ceramic sociation for Computer Aided Design in Architecture
Elements. Copenhagen: Architecture in the Network (ACADIA).
Society [22nd eCAADe Conference Proceedings]. Reijnders, A., and EKWC. (2005). The Ceramic Process: A
Lilley, B., Hudson, R., Plucknett, K., Macdonald, R., Cheng, Manual and Source of Inspiration for Ceramic Art and
N. Y.-W., Nielsen, S. A., et al. (2012). Ceramic Perspira- Design. London: AC Black.

Generation, Exploration and
Optimisation
Generation, Exploration and Optimisation - Volume 2 - Computation and Performance - eCAADe 31 | 153
154 | eCAADe 31 - Computation and Performance - Volume 2 - Generation, Exploration and Optimisation
Automated Simulation and Study of Spatial-Structural
Design Processes
Juan Manuel Davila Delgado1, Herm Hofmeyer2
Eindhoven University of Technology, Netherlands
1
http://www.tue.nl/en/university/departments/built-environment/the-department-of-the-
built-environment/staff/detail/ep/e/d/ep-uid/20104850/
2
http://www.tue.nl/en/employee/ep/e/d/ep-uid/19941803/ep-tab/3/
1
j.m.davila.delgado@tue.nl, 2h.hofmeyer@tue.nl
Abstract. A so-called “Design Process Investigation toolbox” (DPI toolbox), has been
developed. It is a set of computational tools that simulate spatial-structural design
processes. Its objectives are to study spatial-structural design processes and to support
the involved actors. Two case-studies are presented which demonstrate how to: (1) study
the influence of transformation methods on design instances and (2) study the influence
of transformation methods on the behavior of other transformation methods. It was
found that in design instances with the same type of structural elements the influence
of a specifically varied transformation method is more explicit; while, when different
types are present this influence is more undetermined. It was also found that the use of
two specifically different structural modification methods have little influence on the
sub-sequential spatial transformation method.
Keywords. Design process research; design process simulation; spatial design; structural
design.
INTRODUCTION
In the Architecture, Engineering and Construction organize the process of designing and (2) the gen-
(AEC) field, design processes are complex and mul- eration of support methods or tools to aid in the
tidisciplinary undertakings in which designers and design process. In the last category computational
engineers work together on the same problem to tools have been developed to increase productivity
come up with feasible solutions. The final solution is (Grobman et al., 2010), to ease the communication
usually the result of a cyclic process, in which the ini- and the exchange of information between parties
tial solution undergoes several changes and adapta- within the design process (Haymaker et al., 2004)
tions to meet pre-defined and arising requirements. and to take an active role on the design process and
It is assumed that by improving the design generate design solutions (Shea et al., 2005). How-
process, the design outcomes will improve as well ever, little research has been carried out in which the
(Cross, 2008; Brooks, 2010; Kalay, 2004). Consequent- computer is used to study the design process itself
ly, efforts have been carried out on the research of (Kalay, 2004; Coates, 2010).
design processes, roughly subdivided in two cat- The objective of the project presented in this
egories: (1) the development and study of design paper is to increase the knowledge on spatial-struc-
models, which is the formulation of frameworks to tural design processes and consequently to support
the involved actors. To that end a computational Figure 1
toolbox, a so-called “Design Process Investigation” Design Process Investigation
(DPI toolbox) has been developed. More concretely, toolbox framework.
the DPI toolbox as presented in this paper, seeks to
fulfill the following two aims: (1) to study the influ-
ence of a selected transformation method on design
instance evolution; and (2) to study the influence of
a selected transformation method on the behavior
of the other transformation methods. The next sec-
tion will briefly explain the DPI toolbox. Then a dem-
onstration of the types of investigations which can
be performed is shown, and lastly a short discussion
and an outlook on further work are presented.
DESIGN PROCESS INVESTIGATION tered into a Modified Structural Design (MStD). Af-
TOOLBOX ter that, the MStD is transformed into a New Spatial
The DPI toolbox framework (Figure 1) prescribes Design (NSpD) that finally is altered into a Modified
specific and identifiable steps to reach a design New Spatial Design (MNSpD), completing one full
solution. In that sense it could be categorized as a cycle. This cycle can be repeated causing the spa-
prescriptive design model (Cross, 2006). However, tial and structural design instances to co-evolve.
the objective of prescriptive design models is to en- For co-evolutionary designs, no classical conver-
sure successful and consistent results; whereas the gence criteria can be used to stop the process; but,
objective of the DPI toolbox is to simulate design if requirements (spatial design instances) and solu-
processes so its outcomes and more importantly the tions (structural design instances) do not change
process itself can be studied. anymore a (local) optimum is believed to be found
Design processes are cyclic and multidiscipli- (Maher et al., 1996).
nary tasks where both design solutions and design Two other relevant characteristics of the DPI
requirements undergo changes and adaptations toolbox framework are: the “transformation and
before a definitive solution is achieved (Maher et al., modification selection switches” and the “gauges”
1996; Haymaker et al., 2004). Also, design require- (Figure 1). These components have the objective of
ments are usually “ill-defined” and the design pro- facilitating the study of the simulated design pro-
cess is often not recorded properly, so it is difficult cesses. The idea is to use the DPI toolbox to simulate
to trace back or investigate the process later on. different design processes, each with different trans-
The DPI toolbox framework is developed to address formation procedures, and to measure the resulting
those characteristics and problems of a design pro- design instances, by the gauges, through the cycles
cess. for later comparison. In this way, it is possible to
The DPI toolbox framework defines the process study the influence of transformation procedures on
to be followed. During this process a design instance design instances and on sub-sequential transforma-
is subject to four different transformation phases tion procedures.
acting within or between the spatial and structural Note that the DPI toolbox framework only pre-
domains. It works as follows (Figure 1): First, a Spatial scribes the existence of a set of transformations,
Design (SpD), in the spatial domain, is transformed relations, and measurements (by the gauges) be-
into a Structural Design (StD) in the structural do- tween two different domains within a cyclic design
main. Then, within the same domain, the StD is al- process; it does not define specific transformation or
Figure 2
(a) Example of the zoning
algorithm; (b) Two structural
grammars used in the DPI
toolbox.
measurement procedures. Thus, the selected trans- at a right angle. Furthermore, the right cuboids or
formation and measurement procedures used in the spaces should be aligned with the global coordinate
DPI toolbox are not unique in any way, and these system. The Spatial Design undergoes several trans-
used in this paper were chosen primarily for their formations by procedures that are grouped in the
availability. The procedures could be changed in the following categories: (a) proposal of the structural
future to further study spatial-structural design pro- design, (b) preprocessing, and (c) structural calcula-
cesses. tions.
As mentioned before, the DPI toolbox consists The proposal of the structural design consists
of four transformation phases and these phases will of two procedures: first structural zones are created
now be shown to consist of several stand-alone pro- and then, based on them, structural elements are
cedures, put together in a seamless process. Some of generated. For the first procedure, the DPI toolbox
the used procedures have been widely studied and uses an in-house developed automated 3D zoning
utilized in the AEC field, e.g. shape grammars, pat- algorithm (Hofmeyer and Bakker, 2008) (Figure 2a).
tern recognition, and FEM simulations; others have It defines structural zones (elementary structural en-
been developed specifically for the DPI toolbox, e.g. tities) based on sets of spaces. This procedure sub-
geometrical redefinition and kinematic stabilization. divides the Spatial Design into a number of zones,
Next, the four phases of the DPI toolbox, as imple- (grouped spaces) and these are used as a basis to
mented, will be briefly described. generate structural elements. For the next proce-
dure, structural grammars (Shea and Cagan, 1999)
Spatial to Structural Design Transforma- are used to generate structural elements. Struc-
tion (SPT) tural grammars resemble shape grammars used
The first phase generates a structural design in- in the AEC area (Stiny, 1980). They prescribe which
stance and performs a FEM simulation with it, all structural elements can be used depending on the
based on the spatial design instance as used for in- geometrical properties of the previously generated
put. The generated structural design only intends to zones (Figure 2b).
formulate a starting point for the design cycles, and Regarding (b) the preprocessing category, once
it does not intend to be an immediate optimal so- a structural design has been generated, it has to un-
lution for the inputted spatial design. Likewise, the dergo several procedures to be able to be simulated
FEM simulation is not meant for stress engineering, by a Finite Element Method (FEM). First, the geom-
but is merely used to give an indication of the struc- etry of the structure has to be redefined to ensure
tural behavior of the proposed structural design. that all the finite element nodes will be coincident
The Spatial Design consists of a set of volumes or and to determine the wind loaded surfaces. Then
“spaces”. So far, the DPI toolbox is restricted to work the structure should be made kinematically deter-
with right cuboids, parallelepipeds bounded by six mined, loads and constraints should be applied, and
rectangular faces, so that each adjacent face meets a meshing algorithm has to be performed.
Lastly, regarding (c) the structural calculations, Structural to Spatial Design Transforma-
the following procedures should be mentioned: A tion (STT)
first-order linear elastic FEM simulation is carried out In this phase the MStD, an arrangement of structural
to predict nodal displacements in the structural de- elements, is transformed into the NSD, an arrange-
sign; then, the strain energy of each finite element is ment of spaces. This is currently implemented as
calculated. A clustering algorithm groups the finite follows: First, it is indentified which finite elements
elements into clusters based on their strain energy have been deleted in the previous phase and to
and a color-coded visualization is generated. The which space from the inputted Spatial Design they
data obtained during this step will be the basis belong to (i.e. which deleted elements are contained
for the next phase’s procedures, presented below. within which space). Based on that information the
More information on this procedure can be found in spaces that contain many deleted (under-utilized)
(Hofmeyer and Davila Delgado, 2013). finite elements are removed. In other words, spaces
that contain less elements contributing to withstand
Structural Design Modification (STM) the applied loads, are in a structurally-seen less im-
Having generated a Structural Design, the next step portant zone, and are thus deleted (Hofmeyer and
is to improve its structural behavior. Even though Davila Delgado, 2013).
the procedures implemented in this phase follow For the current implementation, the first 30%
closely those of traditional structural optimization, of spaces with most deleted elements are removed,
their objectives are slightly different. The objective and then the remaining spaces are investigated. If
of this phase of the DPI toolbox is not to obtain an spaces with the same number of deleted elements
optimal structural design per se, as in the traditional as the already removed spaces exist, they are re-
way, but to modify the structure only into the direc- moved as well. Note that in almost virtual case that
tion of an optimal design. Thus this phase is called all spaces have the same number of deleted ele-
Structural Design Modification rather than optimiza- ments, then only the first listed 30% of the spaces
tion. are deleted. This implementation is referred in this
This structural modification is based on mini- paper as “Delete Spaces”.
mizing strain energy. A structure that deforms under
a given case of loads and constraints shows strains Spatial Design Modification (SPM)
in its finite elements. The amount of strain energy In this process, the NSpD will now be modified into
in a finite element is a measure of its participation a MNSpD that will serve as the input for a next cycle
in bearing the applied loads. So, finite elements of the DPI toolbox. The main objective of this phase
showing low strain energy can be regarded as be- is to modify the NSpD for the next cycle such that at
ing under-utilized and thus may be deleted. Two least some of the properties of the SpD, which may
versions of existing structural optimization methods have disappeared during the transformations of the
have been implemented in the DPI toolbox namely: cycle, are restored. For example, in the end of the
Evolutionary Structural Optimization (ESO) and To- previous phase, spaces were deleted from the SpD
pology Optimization (TO). A detailed explanation and thus the NSpD has less volume and fewer spac-
of this phase can be found in (Hofmeyer and Davila es. Therefore, in this phase, the NSD could be scaled
Delgado, 2013). Note that the version of ESO used up to the same volume as the SpD and then some
has been modified so that only a single iteration spaces within the NSD could be subdivided in order
is run in the optimization procedure (in this paper to restore the initial number of spaces. This phase
referred to as 1ESO). This is done because accurate is explained in more detail in (Davila Delgado and
enough results can be obtained and computation Hofmeyer, 2013) and it is referred to in this paper as
time is reduced. “Re-scale and Subdivide”.
Figure3
A typical DPI toolbox run.
DPI toolbox example run studies will be performed.

Figure 3 shows a typical run of the DPI toolbox. Two case-studies are presented to demonstrate
Starting from left to right: the Initial Spatial Design; aims (1) and (2) as presented at the end of the In-
the Structural Design, here displaying its strain en- troduction section. In Case-study I, it is investigated
ergy distribution; the Modified Structural Design how a change of the transformation method (a dif-
where the under-utilized elements have been delet- ferent structural grammar in this case) influences
ed; the New Spatial Design (green part only) where the evolution of the design instances (in this case
the spaces with more under-utilized elements (red) structural designs) through the cycles. In Case-study
have been removed; and the Modified New Spatial II, it is investigated how a change of a transformation
Design, which has the same volume and number of method (in this case 1ESO vs. TO for STM) influences
spaces as the initial spatial design. the behavior of the sub-sequential transformation
method (STT), again with respect to an observation
DEMONSTRATION through the cycles.
The main purpose of this section is to exemplify the Figure 4 shows the initial Spatial Design used
types of investigations that can be performed with for both case-studies and the defined settings of the
the DPI toolbox. Note that the cases presented here DPI toolbox respectively. For each case-study, two
serve as a proof of concept and that in further sta- simulations (runs) have been performed, consisting
ges of the research real-life and more complex case- of four cycles each.
Figure 4
Initial Spatial Design used for
case-studies I and II; Tables
list the respective DPI toolbox
settings.
Figure 5
Resulting design instances for
run I-A (which are the same
as for II-A).
Case-study I presented in Figure 7.

For Case-study I, two different runs have been per- For run I-A, it can be seen that dmax and Ut de-
formed: I-A and I-B, with structural grammars 1 and crease at approximately the same rate at every cy-
2 (Figure 2b) respectively. All other settings were cle (Figure 7). This is probably due to the decrease
kept the same (Figure 4, table: Case-study I). Figure of the number of building levels through the cycles.
5 and 6 show the resulting design instances of both In Figure 7, on the top right corner of graph b, the
runs (Note that the resulting design instances of decrease of levels through the complete run is plot-
case-study I-A are the same as those of II-A, so both ted. That rate is similar to the rate of dmax and Ut. So it
are presented in Figure 5). In each figure, each row is not unlikely that there is a link between the num-
presents the results of one cycle, while the columns ber of levels in a StD and its dmax and Ut. Note that
represent a phase within each particular cycle. Fig- the spatial design instances in each cycle have ap-
ure 7 presents the two measurements taken in each proximately the same volume, number of spaces,
cycle for the design instance under investigation: and structural elements and that only dead load has
the maximum nodal displacement (dmax) and the been used as a load case. So, even though the struc-
total strain energy (Ut). These measurements have tural mass of all the design instances is quite similar
been selected because they give an indication of the (Figure 7, table: Case-study I-A) -meaning that the
structural behavior of the resulting StD. A graph of total amount of load is quite similar as well- Ut is dif-
each measurement through the four cycles is also ferent. An explanation for the behavior above is that
Figure 6
Resulting design instances for
run I-B.
in design instances with several levels the structural sequent cycle. The initial increase can be explained
elements at the lower part of the structure have to by two reasons: (1) after the first cycle the design
withstand their own weight plus the weight of the is divided into four fragments. In these fragments
structural elements on top of them and thus show fewer columns have to support more roof-slab area
higher strains. and (2) the roof-slabs in StD.2 are rectangular, which
In the last cycle the design instances have the deform more than square types. In both runs, I-A
same number of levels as in the previous cycle. Con- and I-B, dmax is always located at middle of the high-
sequently Ut does not reduce significantly, but dmax est roof-slab so their dimensions (ratios) have a high
does. This is because the horizontal structural plate influence on dmax and Ut. The second cycle’s decrease
elements that form StD.4 are rectangular, instead of could be explained due to the decrease in the num-
square, and such elements tend to deform more. ber of levels, as observed in the previous run. Finally,
For run I-B, using a different structural grammar, the last decrease is due to the square shape of the
the evolutions of dmax and Ut follow the same pattern; resulting roof-slabs which deform less and thus yield
but they do not correspond so clearly to the evolu- less Ut.
tion of the number of levels, as in run I-A. In Figure 7 In summary, during the evolution of run I-A a
it can be seen that dmax and Ut increase seriously af- continuous decrease for dmax and (partly) for Ut can
ter the first cycle, even though the number of levels be observed. This decrease is directly linked to the
remains the same, and then decrease in each sub- number of levels of the design. Conversely, in the
Figure 7
Result tables and graphs of
runs I-A and I-B.
evolution of run I-B, no pattern can be recognized. penalty) which cannot be compared directly with
This might be explained because in run I-A all the the physically realistic Ut from 1ESO. For that reason
structural elements are the same; whereas for run I-B the Ut values from run II-A were adjusted. This was
it is a mixture of flat-shells and columns. done by (a) matching the density of the structural
elements in the 1ESO calculations to the density of
Case-study II the first iteration of the TO procedure, and (b) by cal-
Also for case-study II, two different runs have been culating the energy of the 1ESO calculations taking
performed: II-A and II-B, using 1ESO and TO for STM, into account the power of the penalty. In this way,
respectively. All the other settings were kept the even though the Ut values are not “physically accu-
same (see also Figure 4, table: Case-study II). Figure rate” comparisons between the two procedures can
8 shows the resulting design instances of run II-B. be made.
Figure 5 presents the resulting design instances of The results tables of Figure 9 present the strain
run II-A, as they are the same as for run I-A. Figure 9 energy of the structural design before and after the
presents the two measurements taken in each cycle: STM procedure is performed, Ut and Ut-final respec-
the reduction of Ut and the difference between the tively.
number of spaces of the SpD and the NSpD. They Note that the Ut values of both runs are very
were chosen because they are indicators of the per- similar. This is because they both have a similar StD
formances of STM and STT respectively. Note that (Figure 5 & 9) with the exception of the last cycle in
the TO procedure optimizes the structural design which the StD -and thus the Ut- differs. Even though
by decreasing the density of the less strained finite for both runs Ut-final decreases at approximate the
elements and increasing the density of the most same rate, Ut-final in run II-B is always lower. This is
strained ones. During this process, a “pseudo-Ut” is because TO minimizes Ut, while 1ESO minimizes
used (in fact a strain energy to the power of a certain structural mass, by deleting the structural elements
Figure 8
Resulting design instances
from run II-b.
with less Ut. So in 1ESO, Ut is hardly reduced. It is also influence on the behavior of STT.
noticeable that the reduction of Ut-final between two
runs diminishes for every cycle. This is because in DISCUSSION AND FURTHER WORK
design instances with more levels U values among The DPI toolbox framework and its current im-
finite elements differ more, because due to grav- plementation were briefly presented. It simulates
ity loads, finite elements at the bottom part of the spatial-structural design processes to: (1) study the
structure yield more strain than the ones at the top influence of a selected transformation method on
part. So there is more opportunity for optimization design instance evolution; and (2) study the influ-
in a structure with very dissimilar U values among its ence of a selected transformation method on the
elements. behavior of the other transformation methods. Two
However, it can be seen as well that this differ- case-studies were presented, which illustrate the
ence in performance has little effect on the behavior DPI toolbox’s potential to aid in the study of design
of the subsequent transformation method (STT). For processes.
both runs, the specific spaces and the total number The first case-study investigated the influence
of spaces deleted by STT are the same during the of using a different structural grammar (a differ-
first three cycles and it only slightly changes in the ent transformation method) in the evolution of the
last cycle. Thus it can be said that within the current structural design, via the maximum nodal displace-
implementation, a different STM seems to have little ment (dmax) and the total strain energy (Ut). It was
Figure 9
Result tables and graphs of
runs II-A and II-B.
found that that in design instances with the same Linear Architectural Design Process’, International Jour-
type of structural elements the influence of trans- nal of Architectural Computing, 8(1), pp. 41–54.
formation methods is observed to be more explicit Haymaker, J, et al. 2004, ‘Engineering Test Cases to Motivate
while, when different types are present, the influ- the Formalization of an AEC Project Model as a Direct-
ence is more undetermined. ed Acyclic Graphs of Views and Dependencies’, Journal
The second case-study investigated the influ- of Information Technology in Construction, 9(January),
ence of using different Structural Modification pp. 419–441.
Methods (i.e. 1ESO vs. TO) on the behavior of the Hofmeyer, H and Bakker, MCM 2008 ‘Spatial to kinemati-
subsequent Structural Transformation Method cally determined structural transformations’, Advanced
(STT). It was found that even though TO generates Engineering Informatics, 22(3), pp. 393–409.
better structural designs than 1ESO, this has little ef- Hofmeyer, H and Davila Delgado, JM 2013, ‘Automated de-
fect on the behavior of the sub-sequential STT. sign studies: Topology versus One-Step Evolutionary
In the future, a further set of rigorous academic Structural Optimisation’, Advanced Engineering Infor-
and real-life case-studies will be devised to bench- matics, Available online 28 April 2013.
mark the DPI toolbox. New transformation methods Kalay, YE 2004, Architecture’s New Media: Principles, Theories
and amendments to the existing ones will also be and Methods of Computer-Aided Design, MIT Press.
implemented to further study the design processes. Maher, M L, Poon, J and Boulanger, S 1996, ‘Formalising De-
sign Exploration as Co-evolution: A Combined Gene
REFERENCES Approach’, Advances in Formal Design Methods for CAD:
Brooks, FP 2010, The Design of Design, Addison-Wesley Pro- Proceedings of the IFIP WG5.2 Workshop on Formal De-
fessional, Boston, MA. sign Methods for Computer-Aided Design, Springer, pp.
Coates, P 2010, Programming.Architecture, Routledge. 3–30.
Cross, N 2006, ‘Designerly Ways of Knowing’, Design Issues, Shea, K, Aish, R and Gourtovaia, M 2005, ‘Towards integrat-
17(3), pp. 49–55. ed performance-driven generative design tools’, Auto-
Cross, N 2008, Engineering Design Methods 4th ed., John mation in Construction, 14(2), pp. 253–264.
Wiley & Sons, Chichester, UK. Shea, K and Cagan, J 1999, ‘Languages and semantics of
Davila Delgado, JM and Hofmeyer, H 2013, ‘Research En- grammatical discrete structures’, Artificial Intelligence
gine: A tool to simulate and study spatial-structural de- for Engineering Design Analysis and Manufacturing,
sign processes’, Proceedings of CAAD Futures 2013, July 13(04), pp.241–251.
3-5, Shanghai, China. Stiny, G 1980, ‘Introduction to shape and shape grammars’,
Grobman, YJ, Yezioro, A and Capeluto Guedi, I 2010, ‘Non- Environment and Planning, 7(3), pp.343–351.
Generative Agent-Based Design Computation
Integrating material formation and construction constraints

Ehsan Baharlou1, Achim Menges2
Institute for Computational Design, University of Stuttgart, Germany
http://icd.uni-stuttgart.de
1
ehsan.baharlou@icd.uni-stuttgart.de, 2achim.menges@icd.uni-stuttgart.de
Abstract. Agent-based systems have been widely investigated in simulation and modeling.
In this paper, it is proposed that agent-based systems can also be developed as generative
systems, in which different aspects of performative design can be defined as separate
drivers in a proper computational framework. In this manner constrained generating
procedures (CGP’s) are studied to integrate the discrete design processes into one system.
Subsequently, this generative agent-based design tool is accompanied with generating and
constraining mechanism which are informed by material characteristics and fabrication
constraints, bringing to the forefront emergent complexity.
Keywords. Computational design; agent-based system; robotic fabrication; constrained
generating procedures (CGP’s).
INTRODUCTION
Performative design, as a design process, can be ment, and exchange that ultimately increase the
described along with several principles. Integrating complexity of the system as a whole.
such principles into a cumulative system is to in- One procedural approach, is to organize such
volve different key aspects of performance in a pro- complexity through a computational framework
cess of formation. The integration process of these that incorporates its own elements, rules and inter-
aspects requires designing a convenient generative actions (Holland, 2000). In some circumstances, this
system to explore performative approaches of form computational framework can exhibit emergent
generation. In terms of computation, form can be phenomena. In fact, the proper generative compu-
defined as an interaction between internal compo- tational framework includes both mechanisms to
nents and external forces (Kwinter, 2008). Similar to generate possibilities and constraints to limit the
natural morphogenesis, in computational design range of possibilities (Holland, 2000). Moreover, this
modeling the development of form can be informed computational framework requires to be further
by the process of materialization, production and specified during the problem solving design pro-
construction (Menges, 2008). Each one of these in- cess; developing such computational framework
ternal components can be described as a separate involves three key aspects: generation mechanisms,
driver, which in turn, can be synthesized into an in- test mechanisms, and a control strategy (Mitchell,
tegral computational design tool. These integrated 1990). Furthermore, based on constrained generat-
drivers interact with each other within an environ- ing procedures (CGP’s), the computational frame-
work should have mechanisms to progressively the topological space is described by a surface with
adapt, or learn, as its components interact (Holland, positive Gaussian curvature and by the fabrication
2006). A particularly promising method of modeling tools, which consist of a KUKA KR 125/2 (6-Axis), and
and simulating such complex adaptive systems a KUKA KPF1-V500V1 turntable (1-axis). The fabrica-
(CAS) is agent-based system (Holland, 1995). tion configuration also includes a HSD ES 350 spin-
dle unit as an effector.
RESEARCH OBJECTIVES: GENERATIVE
AGENT-BASED COMPUTATIONAL DE- GENERATIVE AGENT-BASED SYSTEM
SIGN Agent-based systems as a computational method,
Recent advances in computational capacity open facilitates for researchers the study of various fields
new perspectives into the implementation of agent- of science. An agent-based system consists of large
based systems as generative tools within compu- number of agents that follows simple local rules
tational design in architecture. The purpose of this and interacts within an environment (Gilbert, 2008).
paper is to investigate the possibility of integrating Agent-based modeling consists of defining both the
generative systems properties and constraint pro- agents and the relationships between them (Bona-
cedures into real-time computational form finding, beau, 2002); this can collectively exhibit a complex
which are coupled together to exhibit complex behavior pattern which leads to a global emergent
emergent behavior. behavior as a result. The individual autonomous
In this paper the development of this generative agent, as a self-contained learning unit, perceives its
system is investigated through constraints generat- environment and takes actions (Mellouli et al., 2004).
ing procedures (CGP’s). This approach gives the pos- Accordingly, the agent can learn from its surround-
sibility to link simultaneously different mechanisms ings by permanently repositioning itself within the
to generate and constraint possibilities, which allow overall agent-system and its environment - while ad-
for the exploration of emergent architectural solu- hering to a set of flexible behavioral rules. A system
tions. These mechanisms contain discrete design of agents thus has the ability to learn and adjust its
elements and behaviors wherein bottom up meth- behavior over time (Figure 1).
odology of behavior-based systems can be useful In social science, Gilbert (2008) illustrated that
to organize emergent complexity. This integration agent-based system can be classified into urban
is followed by a generative approach of material models, opinion dynamics, consumer behavior, indus-
properties to explore performative formation in ar- trial networks, supply chain management, electricity
chitectural practices, allowing form to emerge from markets, and participative and companion modeling
the interaction between agent systems and their (Gilbert, 2008). On the other hand, Bonabeau (2002)
surrounding environment. In this investigation form categorized the agent-based system in a business
generation is affected by different attributes, which context into flows (evacuation, traffic), markets (stock
are implemented inside the agents’ data structure. market, shopbots and software agents), organizations
For this investigation, the agents’ data structure (operational risk and organizational design), and dif-
is described by the specific geometrical behavior fusion (diffusion of innovation and adaptation dy-
of bio-inspired plate structures based on the sea namics) (Bonabeau, 2002). These two classifications
urchin. To achieve this, the agents are distributed represent the application of agent-based system for
on the topological space of UV map parameters; simulation and modeling in any behavioral systems.
the relations between agent-agent and agent-en- In the field of sociology, a generative agent-
vironment are derived from this topological space based approach has been regulated in two steps:
e.g., it describes the conceptual neighborhoods Situating agents in a relevant spatial environment
along with its topological relations. In this context, and after that utilizing agents’ interaction based on
Figure 1
A: A Complex Adaptive System
similar to that presented by
Holland (1995); B: Agent
distributions on the topologi-
cal space; C: An Agent-Based
System, topological interac-
tions between agent-agent
and agent-environment.
specific rules to generate another level of bottom- to exert the implemented constraints. This real-time
up organized regularities (Epstein, 2006). In this interaction is relied on the agents’ data structure;
generative method, the systems’ behavior cannot be the agents perceive the environment as well the
deduced to behavior of their components, whereby other agents, and based on their defined ontology
it disregards some of the interactions between the compute the proper response to any stimuli (Pfeifer
elements (Squazzoni, 2012). Accordingly, the gen- and Scheier, 2001). However, the ontology level also
erative method in agent-based system is a bottom- depends on the circumstances that will apply to
up approach to take advantage of low-level features the generative system. This knowledge distribution
e.g., material properties, in a manner that enables among agents could be specified locally in order to
emergent phenomena. avoid unnecessary computation.
In relation to architectural design, developing Consequently, the bottom-up knowledge distri-
such generative computational framework is easily bution provides agent-based system with behavior-
associated to the different methods for establishing based computation rather than knowledge-based
effective organized complexity. One of the features computation. In behavior-based computation, the
of such adaptation in complex system is emergent topological space is explored with agents along
properties, which can be obtained through Con- with their specific behaviors to behave in this prob-
strained Generating Procedures (CGP’s) (Holland, lem domain, rather than with a specific system that
2000). The advantage of CGP’s in agent-based sys- know about the problem domain (Maes, 1993).
tem provides the ability to define agents-based sys- However based on emergence properties, this
tems on mechanisms and constrains - in one specific tool has difficulty approaching a precise behavior.
system. This local generative system as a building Therefore, the underlying elements of this tool need
block has been implemented in the computational enough flexibility to emerge an approximate behav-
framework as an overall generative system which ior, as a cloud (Miller, 2007).
can be identified as a system property. However,
each one of these building blocks or agents has a METHODOLOGY
data structure, in which the mechanisms and con-
straints have a great role to find an optimal solution. Agent-Based system: Defining Mechanisms
Accordingly, the definition of mechanisms and and procedures
constraints are critical in defining real time interac- In order to investigate a generative approach for an
tions within agent-based systems, whereby this defi- agent-based system, a CGP framework is developed
nition must prepare the possibility for a system to with both generative mechanisms and constraints.
become both generative and also have the capacity This method maintains a generative computational
framework to generate all the future possibilities, chronized to the agents’ internal rules; in the steering
while maintaining specific constraints or limitations. level, the goal is decomposed into the sub goals that
The mechanisms of these generative agent-based can be represented by the steering behaviors. This in
systems are bound to material properties, fabrica- turn can become steering signals, which are intelligi-
tion and construction constraints. ble for the locomotion layer; in the locomotion layer,
Material properties in particular have the ability these signals will be converted into motion param-
to characterize geometrical behavior mechanisms. eter of the agent’s locomotion (Reynolds, 1999).
In addition, motion behavior mechanisms have the The agent-based system with Motion behavior
ability to perform as a sensory motor for each one of mechanisms can be influenced by the other steer-
these agents, where if the desired conditions are not ing behaviors, at any moment; this is, to change
being met, then the responsible mechanism will re- the agent’s location and orientation. The behaviors,
lease an appropriate response to change the agents’ which relate to the agent’s motion, have to be trans-
behavior. This reaction can be differentiated by the lated to the steering behavior parameters. The steer-
agents’ situation, which can be varied from splitting, ing behavior gives the possibility to accumulate
eliminating, or re-orientation and relocation of the different type of control behavior procedures and
agents’ situation. Predicting a proper mechanism based on weight of parameters, they can change the
for each situation or problem is not possible in a agent’s motion behaviors. Therefore, the locomotion
behavior-based bottom-up system, due to the low- mechanism must be completely independent from
level ontology that is used in it. For this purpose an steering behaviors (Reynolds, 1999), in which the
agent-based system has to deal with only primitive steering behaviors convert control signals into mo-
ontology to solve the problems, wherein it has been tion of agents (Figure 2).
situated. In the following sections some mecha- Geometrical Behavior Mechanisms: Geometrical
nisms related to this generative agent-based com- behaviors are directly affiliated to material proper-
putational design tool will be investigated. ties which are used in the process of design, fabri-
Motion Behavior Mechanisms: According to cation and construction. Therefore, the geometrical
Reynolds (1999), the motion behavior mechanisms behavior mechanism is reflection of material proper-
can be defined in three layers: action selection, steer- ties. In fact, this mechanism defines interaction ef-
ing and locomotion (Reynolds, 1999). These three fects between geometrical characteristic of agents.
behavior layers are applicable for a wide range of Since, this investigation is about plate-like structures;
autonomous motion behaviors, however, it is neces- therefore this mechanism is limited to the planes
sary to mention here, that this behavioral hierarchy geometry. Hence, geometrical interactions between
is not accessible to all range of autonomous agents agents are related to geometrical planes intersec-
e.g., it is not appropriate for chatterbot (Reynolds, tion; wherein the intersection between a selected
1999). In the other hand, the motion behavior mech- agents with surrounding agents, generates a cell
anism is specialized in specific behaviors, which is with a polygonal structure. The distribution of agents
imitated and modeled from certain behaviors of on the topological surface, defines the final shape
natural entities to relocate autonomous charac- of agents’ cell. The polygonal shape of this cell (e.g.,
ters. Therefore, this mechanism is suitable only for convex or concave polygon) is closely related to the
changing the motion behavior of the system. curvature of the surface (Troche, 2008), which the
In action selection, agents observe the state agents occupy tangentially. Due to the surface syn-
of their environment and that of the other agents clastic definition, the result will be a convex polygon.
in order to perceive their changes. After this initial The geometrical interaction between agents
perception, agents set appropriate goals, which are has been related to the tangent plane intersection.
proportion to the change of system state and syn- However the tangent plane intersection algorithm
Figure 2
Motion Behavior mechanisms
(the attraction and repulsion
steering behaviors).
(TPI) (Troche, 2008) is not appropriate for defin- agents. In that case, the generating agent need to
ing geometrical behavior mechanisms, due to the send a steering signal to change its state in relation
knowledge-based structure which has been used to the neighboring agents and environment.
inside the TPI algorithm. Instead of operating locally,
the TPI algorithm works globally. Therefore an algo- Agent-Based system: Defining Constraints
rithm that is based on a bottom-up approach has As it is mentioned in CGP’S, the generative mecha-
been developed in order to calculate the real-time nism is coupled with constraints. In terms of archi-
intersection between the plate-like structures of the tectural design, constraints can be associated with
agents’ geometry. geometrical and fabrication requirements, which
Accordingly, the intersection mechanism has lead the generated outputs from interactions be-
been developed to find the intersection vertices of tween mechanisms toward desired possibilities. It
a generating agent with other neighboring agents is critical to find a method to relate these intercon-
(Figure 3); these vertices lay on the tangent plane, nected design parameters as a part of the genera-
which is approximately located on the surface. Fur- tive tool. In term of mathematical biology, the con-
thermore, if the agent cell edges (with adjacent straints can be described by morphological spaces,
agents) are naked and not connected to them, then or morphospaces as mathematical spaces (Mitteroe-
it indicates that the agent cell relations are interrupt- cker and Huttegger, 2009).
ed with self-intersection or interpenetration of other
Figure 3
left: The generate agents’ cell
right: The intersection mecha-
nism by slicing algorithm.
The term morphospace, describes the mor- Figure 4
phological features of generative variations within The geometrical and fabrica-
a solution space or landscape of possible outcomes tion constraints: polygon
(Menges and Schwinn, 2012). In this generative tool, radius, connection angle, and
the constraints are considered from the geometric polygon angle; similar to that
limitations and possibilities of the material charac- presented by Menges (2013).
teristics. In the fabrication phase, these constraints
can be described by the morphospace of the fabrica-
tion tool. In general, the agents’ geometric attributes
dictate the need for various procedures to utilize the
generative interaction among the agents. However,
the morphospace’s definition overlaps with the dif-
ferentiation between the geometrical possibilities
and also being producible by the fabrication tool
(Menges, 2013). Therefore, the constraints of this by the specification of the effector and the length of
investigation are derived from the morphological the tool; 3) the polygon angle is indirectly influenced
space, which is categorized in geometrical, fabrica- by the fabrication tool, in which the constraints are
tion and construction constraints. related to the depth of joints who, in itself is deter-
Geometrical Constraints: Since this genera- mined by the connection angle (Menges, 2013).
tive tool is designed for plate-like structures, its
geometrical parameters are applicable to the most RESULT: COMPUTATIONAL DESIGNING
probable range of plate structures. According to TOOL
Menges (2013), the plate morphology is identified
in three major features (Figure 4): 1) the polygon ra- Agent-Based system: Agent-based Pro-
dius, which is defined as the area of the plate that gramming
is calculated based on the polygon vertices, and the A generative agent-based computational frame-
perimeter circle which is bounding these vertices; work is established by identifying the agent types
2) the connection angle, is defined as the angle be- along with their attributes (Macel and North, 2009).
tween connected plates, which is calculated based This identification will be followed by defining the
on the angle between the normal of each connect- boundaries within the surrounding environment
ed plates; 3)the polygon angle, which is defined as that the agent will explore as a topological and
the angle between the polygon edges and is related morphological solution space. After the agents and
to the shape of the polygon (in the polygon convex environment are defined, this framework will simul-
segment(0° to 180°) and in the polygon concave taneously compute all parallel interactions between
segment(-180°,0°)) (Menges, 2013). agent-agent and agent-environment. These parallel
Fabrication and construction constraints: The interactions will be associated by sending and receiv-
morphospace of the fabrication tool, in relation to ing through a feedback loop (Holland, 2006). Accord-
morphological geometry, represents the producible ingly, in the complex system behavior, convergence
parameters of fabrication. As Menges (2013) men- to the desired performance criteria is dependent on
tioned, with the fabrication tools for this investiga- the positive and negative feedback loops.
tion, the morphospace region determines the pro- This generative agent-based tool is initialized
ducible of geometrical parameters: 1) the polygon with agents (plate-like structures) and specific en-
radius depends on the distance between the robot vironment (synclastic surface). After initiation, the
and the turntable; 2) the connection angle is limited motion behavior mechanism is added to identify
Figure 5
left: Edge adhesion (the attrac-
tion to the edges); right: Cell
adhesion (the attraction and
repulsion between agents).
the attraction and repulsion between agent-agent intersections between connected cells. This prob-
and agent-environment. Additionally, this mecha- lem occurs when the edge intersection lies outside
nism is coupled to other control mechanisms for the overlapping boundary area between the two
the rotation and repositioning of agents. However, cells. In this case, an algorithm controls that the right
after the agent distributions on the surface and the intersection between cells exists, it does this by ro-
process of finding geometry interaction between tating the cell or by relocating it on parallel to its
agents takes place (generating cell for each one of normal (Zimmer et al., 2013); through this process
agent), attraction and repulsion algorithms define the intersection point will gradually change its loca-
coherency between the agents’ cell and its environ- tion until it fits inside the defined area (Figure 6).
ment. This coherency is defined by cell adhesion and The main functional component of any gen-
surface edges adhesion. By increasing the value of erative system is it capacity to constrain the pos-
the cell adhesion, agents begin to present flocking sibilities, which are emerged from the generation
behavior and by decreasing it, agents start to avoid mechanism. According to the defined constraints
each other within the bounded surface. It should for this investigation, the generated cells need to
be noted that the agent-to-agent interaction is ex- be limited by two aspects: size of the agent’s cell or
pressed between one agent and its closest neighbor polygon radius, and the angle between agents’ cell
or one agent and a range of its closest neighbors, or connection angle. The cell size can be deduced by
in which each one of these can represent different a regular polygon area formula for convex polygons,
behaviors. In the edge adhesion, by increasing the after which the radius polygon can be obtained; this
adhesion value, agents will be attracted to the edg- radius will be stored in the agents’ data structure to
es and by decreasing it they will gather in a central be accessible by the agents during the computation.
position - away from all edges (Figure 5). However, the polygon radius must be in the specific
Consequently, in geometrical behavior mecha- range imposed by the fabrication constraints, in or-
nism, it would be necessary to avoid inappropriate der to change the size of the cells, cell division and
Figure 6
left: Rotating the cell to find
the right intersection; right:
Relocating the cell on parallel
to its normal.
Figure 7
Finding the right polygon ra-
dius through the cell division
and cell growth mechanism.
cell growth mechanisms are developed to maintain DISCUSSION AND CONCLUSION

this size. Additionally, if the polygon radius becomes Current research has proven effective in implement-
lower than a predefined size due to the fabrication ing such workflow in the presented case, where
constraints (as a result of the robotic morphospace) robotic fabrication principles of the plate structure
an elimination mechanism will remove the cell from morphology have been transferred into the agent’s
the investigation (Figure 7). attributes. For this transfer to be effective, it is nec-
The connection angle is obtained through the essary to precisely investigate and analyze the bio-
normal of the connected cells. Through this arrange- logical plate structure. Although in modeling of the
ment, the angle controller mechanism finds the an- complex system it is not possible to reach a perfect
gle and checks it for conformance. If it is necessary, abstraction; it is possible to find the general behav-
the controller generates a steering signal to rotate ior of the plate structure. This behavior will form the
the cells- this is executed recursively until it reaches basis of the bottom-up mechanism. This bottom up
the range required for connection angle. This mecha- approach, provides the generative system with the
nism is developed to calculate the steering signal for possibility to exhibit emergent plate-like structure
a generating cell and all its adjacent cells (Figure 8). arrangements and patterns. For example, a prelimi-
Figure 8
controlling the connection
angle by generating a steering
signal to rotate the cell.
Figure 9
the final result of the genera-
tive agent-based computation
tool.
narily result of such generative system presented tool can be accommodated within computational
joint conditions that were similar to the Sea urchin, design-, it is imperative to differentiate the aspect of
where three plates meet each other at one corner the generative agent-based computation that con-
point – rather than four. However, although this be- tribute to integrate material system as mechanisms
havior was anticipated, it is also discernible that the with robotic fabrication constraints (Figure 9).
lack of construction mechanisms (which naturally Some of the consequences of this implementa-
has been used in the plate structure), along with in- tion might steer in a different direction expanding
sufficient construction constraints caused the initial further our understanding into the Morphospace
result to be far from what was expected. The initial of robotic fabrication (Menges, 2013). For example,
results might be enhanced by further analysis of angle and plate control mechanisms empower the
biologic model, the fabrication space and the agents design construct in a way that facilitates access for
emergent behavior so that additional mechanisms the designer to methodologies that allow him to
and constraints can be subsequently implemented achieve an optimized plate formation; they also re-
into the tool. duce the need to recourse to design process during
It is also possible to speculate that the results construction phase.
are indicative of the specific means in which agent
based tools process the input data. Unlike “Motion REFERENCES
behavior” (Reynolds, 1999), the generative agent- Bonabeau, E 2002, ‘Agent-based modeling: Methods and
based deals with the implementation of material techniques for simulating human systems’, Proceedings
characteristics, geometrical behavior and construc- of the National Academy of Sciences, vol. 99, no. 90003,
tion constraints; this implementation affects agents’ pp. 7280–7287.
behavior locally and globally. In this manner, agents Epstein, JM 2006, Generative social science: Studies in agent-
become a complex adaptive system of systems. It based computational modeling, Princeton University
is speculated in this paper that although the pre- Press, Princeton.
sented case studies in the generative agent-based Gilbert, GN 2008, Agent-based models, SAGE Publications,
Los Angeles. Foundations of an Agent Modelling Methodology for
Holland, JH 1995, Hidden order: How adaptation builds com- Fault-Tolerant Multi-agent Systems’ in D Hutchison, T
plexity, Addison-Wesley, Reading, Mass. Kanade, J Kittler, JM Kleinberg, F Mattern, JC Mitchell,
Holland, JH 2000, Emergence: From chaos to order, Perseus M Naor, O Nierstrasz, C Pandu Rangan, B Steffen, M
Books, Cambridge, Mass. Sudan, D Terzopoulos, D Tygar, MY Vardi, G Weikum, A
Holland, JH 2006, ‘Studying Complex Adaptive Systems’, Omicini, P Petta and J Pitt (eds), Engineering Societies in
Journal of Systems Science and Complexity, vol. 19, no. the Agents World IV, Springer Berlin Heidelberg, Berlin,
1, pp. 1–8. Heidelberg, pp. 275–293.
Kwinter, S 2008, ‘Who’s Afraid of Formalism’ in C Davidson Miller, JH and Page, SE op 2007, Complex adaptive systems.
(ed), Far from equilibrium: Essays on technology and de- An introduction to computational models of social life,
sign culture, Actar, New York, pp. 144-149. Princeton University Press, Princeton, New Jersey.
Macal, C and North, M 2009, ‘Agent-based modeling and Mitchell, WJ, 1990. The Logic of Architecture: Design, Compu-
simulation’, Proceedings of the 2009 Winter Simulation tation, and Cognition, Cambridge, Mass: MIT Press.
Conference (WSC), pp. 86–98. Mitteroecker, P and Huttegger, SM 2009, ‘The Concept of
Maes, P 1993, ‘Behavior-Based Artificial Intelligence’ in J Morphospaces in Evolutionary and Developmental Bi-
Meyer, SW Wilson and HL Roiblat (eds), From Animals to ology: Mathematics and Metaphors’, Biological Theory,
Animats 2. Proceedings of the Second International Con- vol. 4, no. 1, pp. 54–67.
ference on Simulation of Adaptive Behavior, MIT Press, Pfeifer, R and Scheier, C 2001, Understanding intelligence,
Cambridge, Mass. [u.a.], pp. 2–10. MIT Press, Cambridge, Mass.
Menges, A 2008, ‘Integral Formation and Materializa- Reynolds, CW 1999, ‘Steering behaviors for autonomous
tion: Computational Form and Material Gestalt’ in B. characters’, Game Developers Conference, vol. 13, pp.
Kolarevic and K. Klinger, (eds), Manufacturing Material 763–782.
Effects: Rethinking Design and Making in Architecture, Squazzoni, F 2012, Agent-based computational sociology,
Routledge, New York, pp. 195-210. Wiley and Sons, Hoboken, N.J.
Menges, A 2013, ‘Morphospaces of Robotic Fabrication. Troche, C 2008, ‘Planar hexagonal meshes by tangent plane
From Theoretical Morphology to Design Computation intersection’, Advances in Architectural Geometry, Vien-
and Digital Fabrication in Architecture’ in S Brell-Cok- na, pp. 57–60.
can and J Braumann (eds), Rob/Arch 2012. Robotic fab- Zimmer, H, Campen, M, Herkrath, R and Kobbelt, L 2013,
rication in architecture, art and design, Springer, Wien, ‘Variational Tangent Plane Intersection for Planar Po-
pp. 28–47. lygonal Meshing’ in Advances in Architectural Geom-
Menges, A and Schwinn, T 2012, ‘Manufacturing Reciproci- etry 2012, eds L Hesselgren, S Sharma, J Wallner, N
ties’, Architectural Design, vol. 82, no. 2, pp. 118–125. Baldassini, P Bompas and J Raynaud, Springer Vienna,
Mellouli, S, Moulin, B and Mineau, G 2004, ‘Laying Down the Vienna, pp. 319–332.
Evolutionary Energy Performance Feedback for Design
(EEPFD)
Interaction and automation for a design exploration process framework

Shih-Hsin Eve Lin1, David Gerber2
School of Architecture, University of Southern California, USA, 2Sonny Astani Depart-
ment of Civil and Environmental Engineering, Viterbi School of Engineering, University
of Southern California, USA
2
http://arch.usc.edu/faculty, 2http://cee.usc.edu/faculty-staff/faculty-directory
1
shihhsil@usc.edu, 2dgerber@usc.edu
Abstract. In order to understand the applicability of multidisciplinary design optimization

(MDO) to the building design process, a MDO framework, titled Evolutionary Energy
Performance Feedback for Design (EEPFD), along with the prototype tool, H.D.S.
Beagle, were developed to support designers with the incorporation of partially
automated performance feedback during the early stages of design. This paper presents
2 experimental case studies, one from the design profession and the other from a design
studio, that evaluate the applicability and impact of EEPFD on the early stage design
process. Through these two case studies two different interaction and automation
approaches for applying EEPFD are explored as part of the framework validation.
Observed benefits, challenges and suggestions of EEPFD’s implementation are then
presented and discussed.
Keywords. Conceptual energy performance feedback; design decision support;
performance-based design; multidisciplinary design optimization; genetic algorithm.
INTRODUCTION AND MOTIVATIONS

In the interest of promoting sustainable design, intensive analysis time requirements, design cycle
energy consumption has become increasingly latency amongst a diverse set of design expertise,
significant to the overall design process for archi- and limitations of design cognition and complexity
tecture and building engineering. However, there as previously researched in numerous precedents
is currently little direct or validated feedback be- (Augenbroe, 2002; Oxman, 2008; Attia et al., 2012).
tween the domains of energy simulation and the Consequently, performance assessments are typi-
early stages of the design process where such deci- cally made after the initial design has been finalized
sion making has the greatest impact on the overall with a limited set of explored design alternatives, as
design’s performance and lifecycle. Acknowledged opposed to earlier design stages where a broader
obstacles include: disconnection between and the range of potentially more optimal solutions may
lacking of domain knowledge, tool interoperability, exist (Radford and Gero, 1980). In addition, design-
ers must balance the needs of multiple competing mance” which is defined in this research as the idea
objectives, often through inefficient and imprecise of utilizing performance feedback to influence de-
means, to identify the best fit design through an sign exploration and subsequent decision making
understanding of trade-offs between energy perfor- under the assumption of pursuing higher perform-
mance and other design objectives. ing designs much earlier in the design process and
The motivation of this research stems from the arguably intrinsically coupled, not the norm in con-
potential of multidisciplinary design optimization temporary practice.
(MDO) methods to alleviate issues between the
design and energy simulation domains. MDO is a PROBLEM STATEMENT AND RESEARCH
general term used in reference to the method of OBJECTIVE
coupling parametric design and optimization algo- While current precedents in the building design
rithms in an automated or semi-automated design industry demonstrate the potential of MDO as a
process framework or workflow with the intent of means of solving performance feedback issues,
identifying “best fit” solutions to complex problems there are several inherent and unique challenges for
with competing criteria. MDO methodologies have MDO to be more robustly and pervasively applied
been successfully adopted in the aerospace industry in architectural practice. For example, when MDO is
and other engineering fields and have been gradu- applied to the aerospace industry an identified “best
ally explored in the building industry as a means fit” solution can be mass-produced once it has been
of potentially mitigating existing issues between fully optimized. In comparison, to apply MDO to find
building design and other performance analysis a best fit for building design problem always with
domains. Current research into applied MDO has a unique set of requirements, circumstances, and
initially demonstrated a capability to overcome preferences appears less cost effective by nature.
interoperability issues between domain specific In addition, the objective nature of evaluating de-
platforms. Optimization algorithms automated by sign in other engineering industries provides more
MDO have also been identified as being capable of suitability towards MDO application than the more
increasing feedback results and designer interac- subjective nature of building design, where archi-
tion. By virtue of the automation and optimization tecture is inclusive of aesthetic motivations as well.
more efficient access to performance evaluations Furthermore, a deep rooted disconnection between
of design alternatives inclusive of trade-off studies design and energy simulation domains, enumerated
between competing design criteria in support of previously adds to the factors impeding the applica-
design decision-making is also indentified (Flager et tion of MDO to be fully explored and implemented
al., 2009; Yang and Bouchlaghem, 2010). Given the within the design and energy simulation domains.
trend of computing availability, e.g. cloud comput- Another of our research observations is that the ma-
ing our research into MDO is becoming more obvi- jority of the MDO applications in architecture relat-
ously suitable to the particularities of the architec- ed to building energy performance are conducted
tural practice. We hypothesize this computing trend by researchers predominantly engineering based
results in an exponentially expanding potential of with a focus on optimizing mechanical systems or
MDO applicability. When observed in the context façade configurations, typically much later in the
of this expanded computing capability, the plausi- design process after the building envelope has been
ble bridging of the observed gap between energy finalized (Wright et al., 2002; Adamski, 2007). The im-
performance and design through MDO serves as portance of form exploration during the early stages
another driving force behind this research. MDO is of the design process is to date seldom addressed
therefore understood as a key component to achiev- and typically through overly simplified geometry for
ing the research motivation of “designing-in perfor- proof of concepts observed to be due to the limited
flexibility of existing frameworks (Tuhus-Dubrow ceptual design stage where overall building form
and Krarti, 2010; Janssen et al., 2011). Furthermore, has not been finalized. EEPFD utilizes an automated
there is only a limited number of published MDO evolutionary searching method and a custom ge-
frameworks for building energy performance that netic algorithm (GA) based multi-objective optimi-
have been fashioned and explored through a de- zation (MOO) algorithms, to provide energy per-
signer’s perspective (Caldas, 2008; Yi and Malkawi, formance feedback (i.e. energy use intensity (EUI))
2009; Janssen et al., 2011). Yet, within these applica- to assist in design decision making. Also included
tions, emphasis on the applicability and designer in- are spatial programming compliance (SPC) and a
teraction of MDO frameworks for the early stage de- schematic net present value (NPV) calculations for
sign process have not been adequately researched. consideration in performance trade-off studies. The
In response to this existing gap –emphasizing automation engine of EEPFD was developed as a
the early stage of design and design exploration sta- prototype plug-in for Autodesk® Revit® (Revit), ti-
ges- and the potential of technological affordances tled H.D.S. Beagle, to integrate design, energy, and
and trends, a design centric MDO framework, titled financial domains. The integrated platforms are Re-
Evolutionary Energy Performance Feedback for De- vit, Autodesk® Green Building Studio® (GBS) and Mi-
sign (EEPFD) was developed and has been initially crosoft® Excel® (Excel) respectively. The three com-
tested and benchmarked against conventional de- peting objectives in the algorithm are to maximize
sign processes to understand applicability to the spatial programming compliance (SPC), minimize
early stage of design (Gerber et al., 2012). The ob- energy use intensity (EUI), and maximize net present
jective of this research step presents a focus on the value (NPV). The detailed functionality of each plat-
issue of designer interaction within EEPFD through form, objective functions, and GA-encoding method
observation of two case studies: 1) a practice based can be found in previously published work (Gerber
case study involving a K-12 facility; and 2) a design et al., 2012).
studio based case study involving a single fam- The process of applying EEPFD to obtain perfor-
ily residence. To provide a consistent point of com- mance feedback for design decisions is illustrated
parison a series of measurements regarding design in Figure 1. The first step has two subcategories: the
alternative performance, process efficiency, as well generation of the initial design and the generation
as designers’ interaction and communication with of design alternatives. In EEPFD, the initial design is
EEPFD are established, collected, then discussed. generated by the user through a parametric model
Through a comparative study of these two process- in Revit and a constraints and parameter range
es adopted by these designers, the applicability and file in Excel. At this point the initial geometry, pa-
impact of EEPFD during the early stage of the design rameters and ranges, site information, program re-
process is then presented. quirements, and available financial information are
provided manually by the user. As a result, in order
THE FRAMEWORK: EVOLUTIONARY for designers to use EEPFD, it is essential for them
ENERGY PERFORMANCE FEEDBACK FOR to have the ability to formulate their design prob-
DESIGN lems in the form of a parametric model in Revit with
Evolutionary Energy Performance Feedback for De- their exploration interests translated into a series of
sign (EEPFD), a design centric MDO framework, is parameters and ranges. An understanding of and
developed to incorporate conceptual energy analy- capability with parametric practices, solution space
sis and design exploration of simple to complex thinking, and design exploration is an essential pre-
geometry for the purpose of providing early stage requisite in the implementation of EEPFD (Gerber,
design performance feedback (Gerber et al., 2012). It 2007). The generation of design alternatives is part
is intended to be used by designers during the con- of the automated process driven by the customized
Figure 1
EEPFD’s illustrated simulation
process in accordance to the
identified six step convention-
al energy simulation process.
Highlights are the observation
foci of this paper, emphasizing
the interfaces inclusive of the
interaction between designers
and EEPFD.
GA-based MOO in EEPFD. Once the initial design is has been engaged, as highlighted in Figure 1.
modeled and entered into the automated system,
the following steps are then cycled through until the RESEARCH METHODS AND EXPERIMENT
automation loop is terminated either by the user or DESCRIPTIONS
by the meeting of the system’s termination criteria. To explore the applicability of EEPFD to the design
Once the automation loop is terminated, there are process this research provides an environment in
two ways of proceeding: 1) a design alternative is se- which the interaction between designers and EE-
lected based on the multi-objective trade off analy- PFD during the early stages of design is observed.
sis provided by EEPFD and the design proceeds to This research presents two case studies observed
the next stage of development or; 2) the user manu- in this manner; Case Study I as a practiced based
ally implements changes in the initial design or con- study involving a K-12 school design, Case Study II
straints file based on the provided feedback before as a design studio based study involving a single-
reengaging the automation loop. A detailed descrip- family residential design. In both cases the general
tion of each step and the process of applying EEPFD program layout and overall building envelope de-
implemented by users can be found in previously sign concept has be decided upon, as illustrated in
published work (Gerber and Lin, 2013). Currently, Figure 2.
EEPFD has demonstrated the ability to automatically Through these case studies, two methods of
breed, select, evaluate and identify better fit design implementing EEPFD were explored with a diver-
alternatives for varying degrees of building typolo- gence occurring during the two steps of EEPFD that
gies and geometric complexity. EEPFD has also been require human interaction. While both case studies
validated against the human decision making pro- followed the previously described six step process,
cess and is able to provide a solution space with an Case Study I requires a consultant to provide techni-
improved performance over a manual exploration cal expertise while Case Study II requires only minor
process (Gerber and Lin, 2013). This paper further technical support. In both cases the authors served
validates EEPFD with a focus on understanding the as the technical process experts, thereby bypassing
usability of the framework by designers, which is any technical complications encountered through
described and measured through their interaction the prototype’s use, and were available throughout
with EEPFD prior to and after the automation system the process to provide necessary technical support
Figure 2
Case Study I and Case
Study II conceptual design
development before initial
engagement of EEPFD. Image
courtesy of Xinyue Ma (Case II)
and Swift Lee Office (Case I).
and to enable direct observation of EEPFD on the functions when compared with the initial design.
early stages of this design process and design team. This represents the affordance of the current tech-
The specific focus of our research observation nology and the built in evolutionary search method
is the interaction between the designer and EEPFD of EEPFD. The second performance definition is
in the initial problem formulation and utilization of overall quantity and quality of feedback generated
generated data, steps 1 and 6 as shown in Figure 1. through EEPFD. In this research the qualitative and
During this study three aspects of performance are quantitative analysis data regarding the design
considered and discussed. The first performance problem, process, and product was collected and
definition is that of the generated design alterna- compiled into the metrics defined in Table 1, which
tives as measured through the set of three objective summarizes the recorded data during the explora-
Table 1 Recorded Data Data Type

Utilized evaluation metrics, Design Problem Measurement
categories, recorded data Project Complexity 1. Project type descriptive
and units of measure in this 2. Project size sqft
research step. 3. Space type number number
Design Complexity 1. Energy model surface count number
2. Explored parameter numbers number/
descriptive
Design Process Measurement
Speed 1. Time spent to create design geometry minutes
2. Performance feedback time per result minutes
3. GA settings
Design Product Measurement
Feedback method 1. Feedback number per 8 hours numbers
2. Solution space quantity - feedback design descriptive
alternative number
3. Solution space quality - solution space range in
NPV, EUI and SPC. Pareto solution number
Actor
Actor 1. Main actors role descriptive
Experience 1. Parametric model experience descriptive
2. Energy simulation domain experience descriptive
tion processes. Overall, quantity is defined as the feedback was made available through the prior two
number of design alternatives analyzed and time approaches, the ability of these approaches to pro-
required for each analysis, while overall quality is de- vide relevant information at the speed necessary
fined by the number of Pareto solutions generated for supporting the designers’ rapid determination
by EEPFD. The final performance metric is that of the of optimal configurations for different site condi-
observed design process itself when compared and tions was still in question. As a result, the imple-
contrasted with the six step simulation format of the mentation of EEPFD was explored and researched
experimental design scenarios. Particular emphasis by the designers and research team to understand
is placed on the observed interaction between the whether EEPFD could provide a suitable alternative
designers and EEPFD and their ability to 1) identify approach.
and translate their design objectives and intentions In Case Study I the design problem itself was
into a functional parametric model for the system, limited to optimizing one standard classroom unit
and 2) the perceived relevance of the overall results using the defined kit-of-parts through manipula-
by the designers to assist in their early stage deci- tion of varying façade elements. As parametric
sion-making. design had not been a part of the designers’ prac-
tice prior to this experiment, the authors served as
DESIGN PROCESSES AND RESULTS consultants to assist in the translation of the design
into a parametric model. Due to unfamiliarity with
Case Study I: Practice Based Project parametric modeling, the Revit design platform, and
Case Study I focuses on a K-12 school design with the inherent limitations of both, a week and four it-
approximately 30,000 square feet of usable program erations were needed before the parametric model
space using a method allowing for easy adaptability could be finalized. The parameters explored for fa-
to multiple sites throughout the greater Los Ange- çade configurations included customized opening
les area. Due to flexibility requirements by the cli- sizes, solar screen depth, density, and mounting
ent, a kit-of-parts design concept was developed to distance from the building, as illustrated in Figure 3.
allow for multiple site adaptability and to allow for Following the completion of the parametric model,
future reconfiguration for various educational uses. necessary supplemental information regarding fi-
In addition the pursuit of a net Zero Energy Building nancial estimates, material properties, etc. was com-
configuration for each site was added to the design piled by the authors. In order to closely emulate the
goals by the designers. future implementation process, the financial model
For Case Study I, the designer role was filled by of this experiment was calibrated according to the
the two principal architects whose design philoso- cost estimation of the project. Also the material as-
phy of doing “the most with the least” focuses on signment and HVAC assignment were based on pri-
economical and sustainable qualities as prerequi- or guidelines provided by the MEP consultant.
sites to design. While the designers for Case Study Figure 3 illustrates the collected data and result-
I demonstrated an interest in utilizing innovative ant solution space in a quantified format. Through
technology and methods, neither designer had any the GA run by EEPFD a total of 384 design alterna-
experience with parametric modeling or the Revit tives were generated over a period of 4 hours with
platform prior to this case study. Prior to this case an average speed of less than a minute per result.
study, however, the designers did have experience The solution space improved from the initial EUI =
with attempts to integrating performance feedback 70.08 to 69.30 kBtu/sqft/yr and NPV from -0.51 to
as part of the design process with both in-house -0.48 million dollars. Since the program explored
performance analysis and collaboration with an was fixed in value, the SPC score remained consist-
outside MEP consultant. While energy performance ent throughout the generated solution pool. After
Figure 3
(Top Left) parametric model
of Case Study I. (Top Right)
parametric model of Case
Study II. (Bottom) the collected
quantitative measurements of
Case Studies I + II according to
the established metric.
the completion of the runs, the authors provided or through the MEP consultant. As a result, the de-
to the designers the final trade-off analysis along signers were able to include aesthetic preference as
with 3D design visualizations for their final deci- part of their trade-off analysis when examining the
sion making purposes. After the generated data had generated results.
been provided more guidance was requested from
the designers to discern desirable results from the Case Study II: Design Studio Based Project
abundantly populated solution pool provided by In Case Study II an architectural design student was
the Beagle. However, the designers indicated a posi- provided a single family residential design problem
tive response to inclusion of 3D imaging of all the located along Wonderland Park Avenue in Los An-
design alternatives along with the energy perfor- geles, CA. The program requirements for the single
mance feedback, which was not available through family residences are designated as including: 4
their prior experience with either in-house analysis bedrooms, 3 full bathrooms, 2 car garage, and liv-
ing, dining, and kitchen areas not to exceed a total tor can be identified in the trial and error period nec-
of 3,000 sqft. All room areas are subject to designer essary to define the design’s constraint file so as to
preference, with a maximum being placed on bed- maintain both design intent and model robustness
room dimensions as not to exceed 20’x20’. A 10’ set during the optimization process as the current ver-
back from all site boundaries is also specified. The sion of H.D.S. Beagle will terminate if the geometry
overall design goals are defined to include a meet- breaks.
ing all design requirements combined with con- Figure 3 illustrates the collected data and result-
sideration for a maximum decrease in energy con- ant solution space in a quantified format for Case
sumption. The designer for Case Study II is a master study II. A total of 1,010 design alternatives were
architectural candidate with 6 months instructional generated over the course of 17.8 hours. After all
experience in use of Revit but no prior experience data had been generated, the designer did not limit
of actual application of Revit to a design project or their analysis to the design alternatives receiving the
parametric design in general. The designer’s prior highest ranking from the provided data set. Instead,
environmental design experience is limited to the the designer proceeded with their own design deci-
building physics context within the typical architec- sion making strategy, taking into consideration the
tural education curriculum with no environmental context of the generated solution pool. Overall the
simulation tool use experience or as part of the de- generated solution pool through EEPFD provided
sign requirements typical to her studio design briefs. an improvement in EUI from 59 to 44 kBtu/sqft/yr ,
For Case Study II the EEPFD development and in NPV from -2.92 to -1.86 million dollars, and in SPC
research team acts as both owner and consultant, from 92 to 99. From the full data set the designer
providing all necessary project requirements and first narrowed the solution pool according to EUI
technical support as needed. After the determina- performance. The solution pool was then further
tion of her design intent to explore shading, open- narrowed to only include design alternatives with
ing and each space’s spatial compositions through an SPC score greater than 95. From this narrowed
the parametric model, the designer then proceeded solution pool the final design was selected based on
to define the parametric model in Revit according to aesthetic properties through the designer’s analysis
the proposed parameterization logic and initial de- of the provided 3D images of each design alterna-
sign concept. The final parametric model is illustrat- tive. The objective scores of the final selected design
ed in Figure 3 and was generated over the course of were: NPV = -2.38 million dollars; EUI = 52.04 kBtu/
2.5 months. This recorded time includes the design- sqft/yr; and SPC= 99.29. Once this final selection was
er’s required time to familiarize herself with the use made, the designer proceeded to the next stage in
of Revit for conceptual design through a trial and design development with the generated Revit mass-
error period. As one of the goals of this case study ing model. In this case study, despite the dominance
is to observe the ability of a designer to translate of aesthetic preference as the determining factor for
their intended design concepts into a parametrically the final design, an improvement in all three objec-
oriented mathematically defined form, the designer tive scores over the initial design was observed.
was asked to avoid any geometric simplifications
from their original design geometry for the purpos- CONCLUSION AND DISCUSSION
es of expediency. As such the complications of the EEPFD is a framework that provides a new method
original design geometry and the designer’s unfa- of applying MDO techniques through a custom-
miliarity with parametric design and use of Revit in ized GA to integrate previously inaccessible per-
application to parametric design can be considered formance feedback into the early stage building
as contributing factors to the extended experienced design process. While EEPFD has been validated
parameterization process. Another contributing fac- through tests of accuracy and efficiency, the devel-
opment of best practices through the key metrics Case Study I this led to undesirable window sizes, in
of interaction, communication, and designer ease Case Study II this led to undesirable ceiling heights.
of use is the focus of continued research. Bridging Since EEPFD possesses neither aesthetic preference
the gap between the energy domain and geomet- nor prejudice when generating design alternatives,
ric exploration remains the motivating challenge of consideration must be made when formulating the
the research that begins to address the previously parametric model for maintaining of design intent
established gaps. Secondary contributions of this or an exhaustive exploration of design alternatives
research include the demonstrated usability of EE- is desired. It can be noted that EEPFD is adaptable
PFD by designers through direct interaction during to either scenario, broad or specific, dependent on
the early stages of design. This addresses in part the designer preference. Overall, in both case studies
disconnection of domain expertise as an issue for the final result was observed to be a broader based
the integration of energy simulation for early stage design solution pool with an overall improved multi-
design. Through a comparative study of the two objective performance to enable more informed de-
processes implemented in the case studies, with sign decision making inclusive of a more expansive
specific focus on observing the interaction between simulated aesthetic and formal range. While these
designers an EEPFD, several general observations case studies provide initial observations regarding
can be made. First, in both cases, designers were the impact and interaction of EEPFD on the early
observed to have difficulty with translating their de- stage design process when implemented through
sign intent into a viable parametric model. This may the designer, a subject for future research is the en-
in part be due to unfamiliarity with both the design gagement of a more extensive experimental user
platform, and parametric modeling and parametric group so as to further observe the impact of EEPFD
design methods. While these issues remain, they on the design process. Another subject for future
may be mitigated through increased experience research is the inclusion of additional performance
and industry trends indicating an increased used considerations, such as structural and daylighting,
of parametric design. Secondly, while the design- so as to meet the complexity demands of design
ers in both case studies acknowledged the potential problems through applied MDO.
applicability of the EEPFD generated results, Case
Study II utilized the results in both steps 1 and 6 REFERENCES
more completely. In Case Study I, a net zero energy Adamski, M 2007, ‘Optimization of the form of a building
building objective was desired, and therefore the on an oval base’, Building and Environment, 42(4), pp.
scope was an over extension of the capabilities of 1632-1643.
the current form of the prototype used by EEPFD. Of Attia, S, Hensen, JLM, Beltrán, L and De Herde, A 2012, ‘Se-
particular note, there is a need to include daylight- lection criteria for building performance simulation
ing strategies as part of their analysis. In the current tools: contrasting architects’ and engineers’ needs’,
implementation of the EPFD daylighing is aggregat- Journal of Building Performance Simulation, 5(3), pp.
ed within the more generic EUI calculation handler 155-169.
GBS. As a result Case Study I was not able to fully uti- Augenbroe, G 2002, ‘Trends in building simulation’, Building
lize the generated solution pool, however the frame- and Environment, 37(8-9), pp. 891-902.
work as it is intended is extensible and conceived to Caldas, LG 2008, ‘Generation of energy-efficient architec-
include other tools and design objectives. Finally, in ture solutions applying GENE_ARCH: An evolution-
both case studies the generation of unexpected re- based generative design system’, Advanced Engineering
sults occurs in part based on the designer provided Informatics, 22(1), pp. 59-70.
parametric ranges and there lack of expertise in de- Flager, F, Welle, B, Bansal, P, Soremekun, G and Haymaker, J
sign intent to parametric modeling transcription. In 2009, ‘Multidisciplinary process integration and design
optimization of a classroom building’, Information Tech- tectural Computing, 6(1), pp. 1-17.
nology in Construction, 14(38), pp. 595-612. Radford, AD and Gero, JS 1980, ‘Tradeoff diagrams for the
Gerber, DJ 2007, ‘Parametric practices: Models for design integrated design of the physical environment in
exploration in architecture’, Dissertation, Architecture, buildings’, Building and Environment, 15(1), pp. 3-15.
Harvard Graduate School of Design, Cambridge, MA. Tuhus-Dubrow, D and Krarti, M 2010, ‘Genetic-algorithm
Gerber, DJ and Lin, S-HE 2013, ‘Designing in complexity: based approach to optimize building envelope design
Simulation, integration, and multidisciplinary design for residential buildings’, Building and Environment,
optimization for architecture’, Simulation. 45(7), pp. 1574-1581.
Gerber, DJ, Lin, S-HE, Pan, BP and Solmaz, AS 2012, ‘Design Wright, JA, Loosemore, HA and Farmani, R 2002, ‘Optimiza-
optioneering: Multi-disciplinary design optimization tion of building thermal design and control by multi-
through parameterization, domain integration and au- criterion genetic algorithm’, Energy and Buildings, 34(9),
tomation of a genetic algorithm’ in L Nikolovska and R pp. 959-972.
Attar (eds), SimAUD 2012, Orlando, FL, USA, pp. 23-30. Yang, F and Bouchlaghem, D 2010, ‘Genetic algorithm-
Janssen, P, Chen, KW and Basol, C 2011, ‘Iterative virtual based multiobjective optimization for building design’,
prototyping: Performance based design exploration’ in Architectural Engineering and Design Management, 6(1),
29th eCAADe Conference: Respecting Fragile Places, Uni- pp. 68-82.
versity of Ljubljana, Faculty of Architecture (Slovenia), Yi, YK and Malkawi, AM 2009, ‘Optimizing building form for
pp. 253-260. energy performance based on hierarchical geometry
Oxman, R 2008, ‘Performance-based design: Current prac- relation’, Automation in Construction, 18(6), pp. 825-
tices and research issues’, International Journal of Archi- 833.
Cloud-Based Design Analysis and Optimization Framework
Volker Mueller1, Tiemen Strobbe2
1
Bentley Systems, Incorporated, USA, 2University of Gent, Belgium
1
http://www.bentley.com
1
volker.mueller@bentley.com, 2Tiemen.Strobbe@ugent.be
Abstract. Integration of analysis into early design phases in support of improved building
performance has become increasingly important. It is considered a required response to
demands on contemporary building design to meet environmental concerns. The goal is
to assist designers in their decision making throughout the design of a building but with
growing focus on the earlier phases in design during which design changes consume
less effort than similar changes would in later design phases or during construction and
occupation.
Multi-disciplinary optimization has the potential of providing design teams with
information about the potential trade-offs between various goals, some of which may be
in conflict with each other. A commonly used class of optimization algorithms is the class
of genetic algorithms which mimic the evolutionary process. For effective paralleliza-
tion of the cascading processes occurring in the application of genetic algorithms in
multi-disciplinary optimization we propose a cloud implementation and describe its
architecture designed to handle the cascading tasks as efficiently as possible.
Keywords. Cloud computing; design analysis; optimization; generative design; building
performance.
INTRODUCTION
During the last decades an increased emphasis on have most impact on performance of the building
parametric design approaches is noticeable in early and least impact on implementation effort. Achiev-
architectural design phases. One of the opportuni- ing the best possible overall performance of a pro-
ties of parametric design is that it is possible to gen- ject will allow a response to the challenges posed
erate many instances of a model described in a para- by climate change, resource depletion, and unequal
metric model system, allowing the exploration of distribution of opportunities across the globe.
a large number of design variations. The challenge
is that it is not possible to examine all these design CHALLENGES
variations thoroughly enough to determine which This paper presents work towards an implementa-
ones to develop further. Therefore, integration of tion of a design performance optimization frame-
performance evaluation into the design process work that over time attempts to respond to as many
during early stages could help supporting the selec- of the following challenges as possible (Mueller et
tion process of high-performing design variations. al., 2013).
Obviously, the aim is to enable designers to make 1. Interoperability: the building design software
important decisions about their designs when they industry is similarly fragmented as the building
industry at large. There are many incompatible able, including a mix of automated iterations
software programs and data formats. Various and iterations performed by the design team
approaches have been proposed to overcome (Geyer and Beucke, 2010).
or bypass this obstacle to seamless collabora-
tion between design team members (Flager et FIRST PROTOTYPE
al., 2008; Janssen et al., 2012; Toth et al., 2012). The proposed system is composed of an analytic
2. Data equivalency: design tools may not have framework which connects tools used to generate
sufficient capabilities to create all data required design or analysis models (authoring tools) on one
by analysis tools. side and analysis or simulation tools (analysis tools)
3. Data discrepancy: conceptual design is less on the other side and of an optimization framework
concerned with the detailed information re- which connects the design and analysis system to
quired by analysis software (Bleil de Souza, optimization engines. Initially the data flow uses a
2012). Therefore, analysis opportunities in early mix of proprietary and publicized file formats. The
design may be limited by the information avail- specific components in the prototype implementa-
able to the design team or made available by tion are:
the design team. • GenerativeComponents (GC) [1] as parametric
4. Speed of feedback: design is an iterative pro- design authoring tool with add-ins extending
cess, with fast and frequent iterations. Analysis GC’s classes with structural and energy model
feedback into these design iterations has to be components;
fast enough to remain relevant for the current • STAAD structural analysis engine [2];
iteration (Hetherington et al., 2011). • EnergyPlus analysis engine [3];
5. Performance proxies: there is only insufficient • DARWIN optimization framework including
research to permit use of performance proxies two genetic algorithms [4]; and
to bypass lengthy execution times of estab- • Bentley Analytical Services Framework (ana-
lished analysis methods. Performance proxies lytic framework).
could use either simplified analysis methods, or The utilized file formats are:
simple analyses of a different type indicative of • EnergyPlus’s IDF file format for energy model
future performance. information sent from GC to the analytic frame-
6. Results display: visualization of analysis results work;
is not visually related to the geometric model • STAAD.Pro’s STD file format for structural model
(Dondeti and Reinhart, 2011). This prevents information;
designers from quickly gaining insight into • GC’s GCT file format for parametric model infor-
where in the design deficiencies are located mation;
and thus delays or prevents design improve- • XML file format for extraneous process infor-
ments through human intervention in reaction mation;
to analysis results. • TXT file format for extracted results.
7. In-context results: analysis results are not avail- This solution was introduced and tested at the
able in the digital model for access that would SmartGeometry event in 2012 in Troy, NY [5] in a
enable automation of refinement iterations or prototypical implementation (Mueller et al., 2013). It
multi-objective optimization routines. included energy analysis and structural analysis plus
8. Human-machine balance: not all design goals a genetic algorithm (GA) (Figure 1). All of the system
are measurable. How are “hard” computed per- architecture was implemented on a client system
formance metrics balanced with “soft” qualita- (desktop computer). The analysis engines were also
tive aspects. Several approaches are conceiv- implemented as analysis services on the cloud with
Figure 1
Software architecture for the
prototype implementation
at SmartGeometry 2012 with
only the analysis engines in
the cloud.
the analytic framework establishing the connection software standards, including replacement of the
between the client and cloud applications based on GAs with ones that have been in commercial use
user selection of “local” or “cloud” execution of analysis. for several years, extending them to multi-objective
Limitations of this test implementation were: optimization while still complying with production-
the implementation did not progress beyond pro- level software standards. This includes use of inter-
totype stages causing several deficiencies; analysis application communications that are robust, secure,
models were kept at the minimal implementation and prevent the problems encountered with the
necessary to allow the analysis algorithms to ex- prototype. Most importantly in the context of this
ecute while possibly achieving sufficient complete- paper, this second version of the design analysis and
ness of the models for conceptual design; there was optimization system includes all necessary compo-
only a partial deployment of the system in the cloud, nents running as services on the cloud (Figure 2).
particularly of the analysis engines, leading to accu-
mulated latency issues; and lack of robustness. The User Workflow and Software Components
conclusions of this prototype implementation were: On the surface, the user workflow and involved com-
• Increase robustness of the software compo- ponents are the same as in the previous implemen-
nents and their communications. tation: the user designs a parametric model includ-
• Increase “completeness” of analysis and simula- ing analytical model components in GC. The model
tion models without increasing required model needs to be driven by parameters so that changes
complexity. to the model can be applied in response to analy-
• Develop the system architecture in a way to sis results or any other computations evaluating
minimize negative side-effects of a deploy- the performance of the current instantiation of the
ment in the cloud while maximizing the de- parametric model (Mueller et al., 2013). The analyti-
sired positive effects. cal model components provide input to analytical
nodes which connect to external analysis engines
SECOND VERSION OF ANALYSIS AND (STAAD and EnergyPlus) via the analytic framework.
OPTIMIZATION FRAMEWORK Analysis results are returned to the parametric model
The improved second version of the analysis and for any subsequent computations to extract or de-
optimization framework responds to the limitations termine characteristic performance values. These are
by replacing the prototype optimization framework passed as fitness values to an optimization node in
with a version rewritten to meet production-level GC.
Figure 2
Cloud components of the
software architecture for the
improved implementation.
The optimization node in GC is the interface to performs any subsequent computations, resulting
the optimization framework which in turn interfaces in fitness values for the specific phenotype that are
with the optimization engine. The user also identi- fed back into the optimization node and from there
fies those parameters or design variables that the to the optimization framework. The fitness values
optimization process may manipulate. The optimi- are associated with the specific individual’s chromo-
zation framework converts parameter ranges and some and communicated back to the GA. Once the
their discrete increments (resolution or granularity) GA has received all fitness values for an entire popu-
into binary chromosomes (or the genome) for the lation it evaluates that population in order to deter-
GA in order to generate individual design solutions mine those sets of chromosomes that it will use to
or phenotypes in analogy to evolutionary processes generate the genome for the next generation, i.e.
in nature. the next set of genotypes.
When the optimization is processed locally on
MDO Process the client computer, the optimization framework
When the optimization process starts, the GA gen- pushes each phenotype’s chromosome into the
erates a first set of chromosomes to create a first parametric modeling engine and then waits for the
generation of individuals (a generation’s population) corresponding fitness values, which means that all
based on the GA’s implementation of stochastic processes are executed sequentially, starting with
principles applied to an evaluation of the nature of the parametric model update, including the analysis
the genome. The optimization framework interprets requests and subsequent evaluations, and ending
the chromosome into the set of design variables or with the return of fitness values to the optimization
parameters and pushes those into GC as engine for node. This repeats for each individual in a genera-
the parametric modeling service with the applicable tion until the entire population is processed. After
parametric model active. The parameter changes the GA evaluates the results for the generation it
propagate through the model to create the corre- creates a new set of genotypes for the optimization
sponding model instance (individual or phenotype). framework to start the process for the next genera-
Any analyses included in the model are request- tion.
ed as cascading processes via the analytic frame- When the user requests execution in the cloud,
work while the parametric modeling engine waits in contrast to the prototype this implementation
for the results to return, fully dependent on suc- processes an optimization request entirely in the
cessful termination of the analysis processes. It then cloud. The user designs and develops the paramet-
Figure 3
Software architecture of client
and cloud components of
improved implementationes.
ric model, the connection of salient parts of the CLOUD IMPLEMENTATION

model to various types of analyses, further evalua- The cloud implementation uses Microsoft’s Azure
tion computations, and the workflow connection to [6]. “The Cloud” as a concept assumes as primary
the genetic algorithm in the authoring tool GC on a benefit virtually unlimited processing resources in
desktop computer (the client) (Figure 3). This avoids form of virtually identical work horse CPUs (com-
internet latency while modeling and permits local pute nodes) with some amount of RAM and high
testing and debugging of the parametric model, its connectivity to the respective cloud infrastructure in
connection to the analysis engines, and validation of the physical server farm in which they are housed.
additional computations performed on analysis re- Of course, there are other benefits to using cloud
sults for their use as fitness values in the GA process. resources, e.g. ubiquity given internet connectiv-
When the user commits the optimization pro- ity with sufficient bandwidth, or availability without
cess to the cloud the request is packaged with all first cost of ownership or maintenance costs even
required parameters for the optimization process it- when not used. These are secondary to the issues
self, like population size and termination criteria; the addressed in this paper. Relevant is another prem-
required data, especially the parametric model as a ise of the cloud, which is that these resources are
snapshot GCT file; the composition of the parameter state-free, which correlates to the assumption that
set (names, ranges, resolution); and other required any user of these resources needs to be prepared
information not relevant in the context of this pa- that they could fail, go off-line, or be replaced at any
per. This package is sent from the optimization point in time. Cloud processes (cloud services) must
framework client to the optimization framework’s not rely on preservation of state over long time. They
web service in the cloud which then creates an op- should be set up to receive processing requests, per-
timization job based on the information received in form the necessary computations, potentially store
the package. Most important for the cloud imple- any results in a pre-determined location, potentially
mentation, the optimization framework permits to indicate successful termination in addition to depos-
dispatch all phenotype requests for a population iting the successful results, and terminate. Request-
in parallel, while generations have to be processed ors of cloud processes should be prepared that any
sequentially. This allows parallelized evaluation of all started process may fail and that requests may have
individuals in a generation and potentially acceler- to be re-issued until successfully resolved.
ates processing by a factor close to the population Compared to the implementation on a desktop
size specified by the user for the optimization pro- computer, a cloud implementation therefore intro-
cess. duces additional elements: first and foremost is a
scheduler which receives requests to process com-
Figure 4
Simple cloud scheduling
service schematic.
putational tasks and dispatches them to available er, including failure of the respective compute node.
compute nodes [7] (Figure 4). This is a very straightforward implementation
for massively parallel processing tasks (“embarrass-
“Simple” Cloud Task Scheduling ingly parallel”) which are the ideal use case for the
The sequence of cloud processing is that a client re- cloud. However, the most important implication
quests a processing job from the client-facing web of this type of cloud scheduling regime is that task
service of the cloud (1). The web service starts the sequencing as required for the MDO process can-
requested job (2). The job generates as many tasks not be guaranteed, so that process dependencies
into the task queue as needed (3). The scheduler need implementation of specific measures to ensure
polls the task queue regularly (4) and pulls the first proper sequence to avoid extended wait times in
queued task and distributes it to an available com- the best case and deadlock in the worst.
pute node (5). The task processes on the compute
node (6), retrieves any data it needs from storage MDO Cloud Process
usually via look-up in some table or database, posts For the multi-disciplinary optimization case there
or updates any task states to a table (7), and returns are various levels of dependencies that might suf-
any process results to storage (8). It indicates its or- fer substantially from the “state-free” and fail-any-
derly termination to the scheduler. The job process time premise of cloud resources. If the task is the
polls the table (9) to assess progress of individual optimization process itself, then any failure during
tasks or the overall job status. The web services polls the process will void the entire process and requires
the job for job progress or completion (10). On re- the entire optimization to restart. This suggests
quest from the client (11), the web service pulls any that the optimization framework would need to be
results from storage (12) and displays them to the state-aware. Similarly, the optimization engine (de-
client (13) or makes them available for download to pending on its architecture) needs to be state-aware
the client. Compute node fail-over is implemented or needs to store its intermediate results in such a
by the scheduler hiding a task that has been distrib- way that it could pick up the process at any point,
uted to a node. A time-out limit makes visible any perhaps from conclusion of a generation. Any para-
hidden task that has not indicated orderly termina- metric model (e.g. individual in a population) is of
tion to the scheduler within its time-out, so that the course one of the parallel processes in a generation
scheduler will distribute it again on the assumption that benefits from the virtually unlimited resource
that the computation failed for one reason or anoth- concept of the cloud. However, during several as-
Figure 5
Cloud scheduling service sche-
matic for MDO services.
pects of the MDO process various states are reached MDO web service on the cloud and sends the re-
in conflict with the state-free concept. This requires quired data package (1). The MDO web service
some additional steps in the process sequence in or- starts the MDO job (2). The MDO job comprises the
der to be handled by a standard scheduler. optimization framework and the optimization en-
gine (the GA). It extracts and applies or distributes
MDO Cloud Task Scheduling the relevant data from the data package, handles
An improved system of MDO cloud task sched- the parameter set to chromosome conversion, etc.,
uling is used to overcome the issues described generates all parameter sets for a generation, and
above (Figure 5). It uses tables to preserve states in generates phenotype tasks PN into the task queue.
an otherwise state-free system that can fail at the The scheduler polls the task queue and pulls task
parametric model and analysis level. Model gen- PN (4) for distribution to compute nodes (5). The PN
eration and analysis tasks are executed “in parallel”, task processes the appropriate parameter set (phe-
with the analytical framework as part of the para- notype) in an instance of the parametric modeling
metric model task (i.e. on the same node, because engine (GC) on the compute node (6) and posts or
these processes are sequential anyway). However, updates its execution state to the table (7).
the analytical framework can start one or multiple When the parametric model includes analysis
analysis tasks that will be queued and handled by nodes, these request analysis tasks from the analyti-
the scheduler. Possible approaches are single-queue cal framework instance AFN which runs on the same
or dual-queue, separating modeling and analysis node as the parametric model engine (8). This does
tasks. The advantage of a dual-queue system is that not impact processing speed because the paramet-
it could be designed to handle cascading depend- ric model engine needs to wait for the cascading
encies without any danger of resource deadlock; analysis processes to terminate and for the analyti-
however, its implementation is more complex. The cal framework instance to return the analysis results.
current implementation uses a single task queue The analytical framework instance AFN posts and up-
for both, modeling and analysis, tasks and is based dates any processing states in the table (9) and adds
on the premise that adaptive scaling (marshalling of analytic tasks AN1 and AN2 (etc.) to the task queue
additional resources when tasks are waiting in the (10). The scheduler pulls the analytic tasks AN1 and
queue) will prevent resource deadlock. AN2 (etc.) from the queue (11) and distributes them
The sequence of the MDO cloud processing is to available compute nodes (12) where they start
that the GC client requests an MDO job from the processing, pulling any data from storage, updating
Figure 6
Cloud software architecture
including the additional task
management elements.
their processing state to the table (13), and deposit- display them in the client context and/or instantiate
ing any results back to storage (14). the corresponding solutions.
Meanwhile, analytic framework instance AFN
polls the table for the analysis tasks’ states (15), and CONCLUSION
when they have successfully terminated it pulls the Cloud computing provides access to ubiquitous
results from storage (16). The analytic nodes in the and virtually unlimited resources. It permits accel-
phenotype task PN in the parametric modeling en- eration of processes that include tasks that can be
gine poll process AFN for analysis results and post- performed in parallel but are predominantly per-
process them to convert them into fitness values formed sequentially in conventional desktop imple-
(17). PN also computes any other fitness values and mentations. As demonstrated, even more complex
passes them to the optimization node in the para- processes like multi-dimensional optimization can
metric model. The optimization node passes the fit- be successfully handled with basic task scheduling
ness values to an optimization framework instance if any cascading and dependent tasks are set up
on the compute node which posts the results and in ways that allow the proper management of se-
task completion to storage and table, respectively quencing (Figure 6).
(18). Even though cloud resources need to be ac-
The optimization framework instance in the cessed through internet connections and the com-
MDO job polls the table for completion of all tasks putational resources available in the cloud are con-
PN in a generation (19), and pulls the fitness values sumer grade rather than high-end, cloud computing
from storage (20). The optimization framework in is advantageous whenever massive parallelization
the MDO job prepares the generation data for the of tasks can be utilized. In the case of MDO using
GA which then generates the next generation’s pa- GAs, it is obvious that the individuals in a generation
rameter set and starts scheduling a new set of phe- can be processed in parallel leading to acceleration
notype tasks PN (21). The MDO service polls the MDO of the process by approximately the population size.
job for completion of the entire optimization run In addition, any analysis processes included in the
(22) and notes completion, if applicable. The opti- MDO job could be processed in parallel, leading to
mization framework client polls the MDO service for additional time savings depending on the number
job completion (23). Upon request, the MDO service and length of the analysis processes.
pulls the optimization results from cloud storage The cascading nature of the processes and their
(24), and the client downloads them (25) in order to dependency pose a difficult challenge if in contra-
diction to the base premise of virtually unlimited and T Williamson (eds.), Proceedings of Building Simula-
resources, the computing resources are artificially tion 2011: 12th Conference of International Building Sim-
limited, e.g. if for an MDO job that includes analysis ulation Association, Sydney, Australia, pp. 1250–1257.
tasks fewer compute nodes are allocated than the Flager, F, Soremekun, G, Welle, B, Haymaker, J and Bansal, P
population size in a generation. This would cause 2008, ‘Multidisciplinary process integration and design
an irresolvable deadlock of resources because the optimization of a classroom building’, CIFE Technical Re-
parametric modeling or phenotype processes port TR175, Stanford University.
would occupy all the available compute nodes and Geyer, P and Beucke, K 2010, ‘An integrative approach for
any remaining phenotype tasks as well as the cas- using multidisciplinary design optimization in AEC’, in
cading analysis tasks would be queued up without W Tizani (ed.), Proceedings of the International Confer-
any chance of additional nodes becoming available. ence on Computing in Civil and Building Engineering,
If the set up on the cloud permits limitation of the Nottingham University Press.
number of compute nodes additional precautions Hetherington, R, Laney, R, Peake, S and Oldham, D 2011, ‘In-
need to be put in place to reserve compute nodes tegrated building design, information and simulation
for cascading processes. modelling: the need for a new hierarchy’, in Proceedings
of Building Simulation 2011, Sydney, pp. 2241–2248.
FUTURE WORK Mueller, V, Crawley, DB and Zhou X 2013, ‘Prototype imple-
With completion of the improved implementation mentation of a loosely coupled design performance
imminent, use of the system in user case studies is optimisation framework’, in R Stouffs, PHT Janssen, S
next. This will also allow benchmarking of desktop Roudavski and B Tunçer (eds.), Open Systems: Proceed-
implementation and cloud implementation to as- ings of the 18th International Conference of the Associa-
sess the impact of parallelization using “virtually tion of Computer-Aided Architectural Design Research in
unlimited” resources. Additional work will be docu- Asia CAADRIA 2013, CAADRIA, Hong Kong, pp. 675–684.
mentation and publication of APIs to allow third Toth, B, Boeykens, S, Chaszar, A, Janssen, P and Stouffs, R
party development to add analysis engines and op- 2012, ‘Custom digital workflows: A new framework for
timization engines, as well as, add-ins for design au- design analysis integration’, in T Fischer, K De Biswas, JJ
thoring tools to connect to the optimization frame- Ham, R Naka, and WX Huang (eds.), Beyond Codes and
work. Pixels: Proceedings of the 17th International Conference
on Computer-Aided Architectural Design Research in
ACKNOWLEDGEMENTS Asia, CAADRIA, Hong Kong, pp. 163–172.
This work was supported by contributions from vari-
ous groups within Bentley Systems, Incorporated, [1] GenerativeComponents from Bentley Systems: http://
especially Applied Research, Special Technology www.bentley.com/en-US/Products/GenerativeCom-
Projects, Design and Simulation’s Structural Analysis ponents/, accessed May 24, 2013.
team, and the BentleyCONNECT team. [2] STAAD.Pro user interface to the STAAD analysis engine
from Bentley Systems: http://www.bentley.com/en-
REFERENCES US/Products/STAAD.Pro/, accessed May 23, 2013.
Bleil de Souza, C 2012, ‘Contrasting paradigms of design [3] EnergyPlus from the U.S. Department of Energy: http://
thinking: The building thermal simulation tool user vs. apps1.eere.energy.gov/buildings/energyplus/, ac-
the building designer’, Automation in Construction, 22, cessed May 23, 2013.
pp. 112–122. [4] Darwin Optimization (version 0.91) by Dr. Zheng Yi
Dondeti, K and Reinhart, CF 2011, ‘A “Picasa” for BPS – An Wu, http://communities.bentley.com/communities/
interactive data organization and visualization system other_communities/bentley_applied_research/w/
for building performance simulations’, in V Soebarto bentley_applied_research__wiki/6584.aspx, accessed
Dec 7, 2012.
[5] http://www.smartgeometry.org/, accessed April 15,
2013.
[6] http://www.windowsazure.com/, accessed June 20,
2013.
[7] http://msdn.microsoft.com/en-us/library/
hh560251(v=vs.85).aspx, accessed June 20, 2013.
Graphical Smalltalk with My Optimization System for
Urban Planning Tasks
Reinhard Koenig1, Lukas Treyer2, Gerhard Schmitt
ETH Zurich, Chair of Information Architecture
http://www.ia.arch.ethz.ch/
1
reinhard.koenig@arch.ethz.ch, 2lukastreyer@student.ethz.ch
Abstract. Based on the description of a conceptual framework for the representation of

planning problems on various scales, we introduce an evolutionary design optimization
system. This system is exemplified by means of the generation of street networks with
locally defined properties for centrality. We show three different scenarios for planning
requirements and evaluate the resulting structures with respect to the requirements of our
framework. Finally the potentials and challenges of the presented approach are discussed
in detail.
Keywords. Design optimization; interactive planning support system; generative system
integration; evolutionary multi-criteria optimization.
MOTIVATION
For many computer scientists the programming lan- integration of generative methods in planning pro-
guage Smalltalk was the most pioneering human- cesses is their complicated handling. Typically they
computer interaction language of the 1970s. It was require extensive input of abstract technical rules
designed to be so simple that even children could and parameters that are unfamiliar and daunting for
program. It is one of the first totally object oriented planners.
languages – everything is an object. While today The situation is further complicated by the fact
many ideas from Smalltalk have since been adopted that planning projects typically consist of a mixture
by other languages, the visionary thinking of the of contradicting and non-contradicting criteria as
time when it was developed can still inspire us today well as of directly measurable criteria and only in-
to strive for flawless human-computer interaction in directly interpretable measures. The lack of suitable
the development of design optimization systems for optimization methods hinders a systematic evalua-
architecture and urban planning. tion of possible compromises between contradict-
ing planning requirements.
PROBLEM STATEMENT
A number of promising generative algorithms are STATE OF THE ART
available today, but none are currently employed to In their seminal book, Radford and Gero (1988) show
enhance and simplify the day-to-day work of urban various examples of how optimization strategies can
planners. Computer support for urban planning pro- be used to solve design problems. Although today
jects is usually restricted to basic CAD drawing tools. we can use more flexible evolutionary optimization
In the authors’ opinion, one reason for the lack of methods (Deb, 2001), the concept for their applica-
Figure 1
Planning scenario divided into
three levels of abstraction.
Left: Topological relations
between elements and basic
properties. Centre: Geometric
distribution of the elements.
tion and the role of pareto-optimal fronts has not tion as a basis for evolutionary optimization strate- Right: Geometric representa-
changed a lot over the past few decades (Bentley gies is as an issue that is yet to be resolved. tion of a possible planning
and Corne, 2002). A good example for state of the solution [6].
art interactive generative planning systems is the CONCEPTUAL FRAMEWORK
work of Derix (2009). Dillenburger et al. (2009) have To address the aforementioned problems, our first
also presented an interesting system for creating task is to develop a conceptual framework that in-
building designs using a weighted-sum optimiza- cludes a combination of various interaction strate-
tion algorithm. gies for the user interface, different generative tech-
Current commercial solutions for generative or niques, and some optimization methods. We have
procedural modeling, for example Grasshopper [1], approached this concept from two perspectives:
GenerativeComponents [2], or CityEngine [3] exem- from that of a planner and from that of a software
plify the problems with such systems: they require developer.
intensive training before they can be used efficiently To meet the planner’s requirements we sepa-
and though sometimes attractively designed, their rate the problem representation and the definition
user interfaces are not intuitive for urban planners. of requirements by at least two levels of abstraction
Furthermore it is not efficient to couple them with (Figure 1): The first holds the topological relations
optimization tools, because of the increased com- between various elements and basic properties (Fig-
puting time and restricted possibilities offered by ure 1 left). The graphical objects on this level can en-
their corresponding APIs. Although Galapagos [4] code parameter values e.g. by their size, position or
provides an optimization method for Grasshopper, colour, etc. The second abstraction level comprises
Rutton (2010) notice that it is only useful for simple the geometric representation of possible planning
problems. solutions (Figure 1 right). One can interact with all
the graphical objects of a current planning proposal
AIMS on each level to test different options and to refine
Our main goal is to use graphical objects to repre- a planning iteratively. From the software developers
sent a planning problem and to control an optimiza- point of view we develop a framework for combin-
tion algorithm using primarily these objects. A fur- ing evolutionary optimization techniques. These in-
ther challenge is to translate the planner’s partially clude generative algorithms and evaluation mecha-
vague qualitative requirements into a precise quan- nisms to analyze the generated variations. As a basis
tifiable problem representation for an algorithm. for this framework, we use state-of-the-art evolu-
Translation problems are one reason why planners tionary multi-criterion optimization methods. For
rarely embrace computer support. To improve this a comprehensive and easily understandable intro-
situation we aim to develop an interactive system duction to evolutionary algorithms, see Bentley and
for supporting the urban planning process with a Corne (2002). In the following description, we focus
more constructive and intuitive interface for plan- only on the essential aspects that are necessary for
ners. The combination of well-designed interaction our purposes. We take the AForge.Net Framework
strategies and planner-friendly problem representa- [7] as the starting point for the implementation of
Figure 2
Mapping process from an
instruction tree to a street
network. The grey dotted
street segments on the right
side illustrate the adaption of
instructions how to add a new
segment to a existing network.
our evolutionary algorithm (EA). The main argument below.

for EA is their flexibility in dealing with the problem
representation, which is crucial for the problems de- Generation Mechanism
scribed above. The basic idea for the generative mechanism is to
From the user’s perspective we reverse the logic use an instruction tree, which holds the instructions
of generative planning systems: instead of exploring on how to grow a street network (Figure 2). This
the results of different parameter settings or pro- growth process can be denoted as embryogeny and
cedural rule sets, we allow a planner to graphically is responsible for unfolding the abstract information
define what performance a solution shall have and stored in a genotype to the concrete structure of a
the optimization system automatically generates a phenotype, which is why this process can also be
set of best compromise solutions. This constellation described as mapping from genotype to phenotype.
of deducing a solution from its desired properties is For the genotype representation we use a chromo-
called an inverse problem [5], which was used with some which in turn is represented in our case by an
a different intention by Koltsova et al. (2012). Based instruction tree as shown in Figure 2. The rules to
on this concept we develop a method that can be create the street network (Figure 2 right) are adapt-
called bi- or multi-directional planning, since one ed from the concept of self-sensitive L-Systems by
can control a computer-based planning system from Parish and Müller (2001).
any of the abstraction levels as shown in Figure 1. As a basic component of the instruction tree we
implement a class for an individual instruction node
IMPLEMENTATION which stores the information on how to add a new
In this paper we present just one part of the afore- street segment to an existing node of the network
mentioned framework: the optimization of a street (Figure 3). The instructions for a node are reduced
network inside a planning area with specific local to three which is the minimum necessary for our
properties. The sub-areas can be defined by a user basic system: the range li for the length of a street
with the help of graphical objects in a similar way as segment, the angle αi that indicates the angular
shown in Figure 1. deviation from a regular division, and the range κi
Before we consider some example applications of possible arms at a crossroad. Figure 3 illustrates
of the optimization system, we first describe the im- the mapping of the instructions of three instruction
plementation of its basic generative and evaluation nodes to a phenotype representation. We use the
mechanisms. Taking the AforgeNet Framework [7] as additional parameter tree depth to restrict the size
our starting point, we extend it by a class for a chro- of an instruction tree to a certain limit. The individu-
mosome with mutation and crossover methods and als of an initial generation of a population start with
a class for a customized fitness function as described randomly assigned instruction values.
Figure 3
On the left, the tree shows
the main instructions of an
instruction node. In the box on
the right one can see how the
instructions are assigned to a
street network.
The instruction tree can be mutated and used betweenness centrality (choice) of a network. For
for crossover operations as illustrated in Figure 4. this we need to calculate the all-pair shortest paths
Since the instructions of a node are always relative for the network, and compute this following the
to the existing street network, new combinations af- concept elaborated by Hochberg [8] using a paral-
ter the crossover always work. The main reason why lel GPU implementation of the Floyd-Warshall algo-
we use a tree structure for the chromosome is that rithm to calculate shortest paths. For the weight-
it ensures that after the crossover and mutation op- ings in the corresponding graph we use angular
erations, the corresponding street network remains distances instead of metric distances as introduced
connected (if the initial network was connected). by Turner (2001; 2007). The choice value for a spe-
The mutation operator simply takes (e.g. 1-10%) in- cific street segment equals the number of shortest
dividual nodes of an instruction tree and assigns a paths from all street segments to all others that pass
randomly generated value to one of its parameters. through that segment. For the sake of simplicity, we
The frequency of the execution of these operators at use only the choice value in the following examples
one iteration (or generation) is defined by the cross- to characterize street networks, but other centrality
over and mutation rate. measures would be useful too.
One of the most important properties of a gen-
erative mechanism, as part of an optimization pro- EXAMPLE SCENARIO
cess, is its ability to generate very different network As a starting situation for the following examples we
topologies. It is this property that allows an optimi- use an area with the dimensions 3000m × 2000m
zation system to find interesting and surprising so- that needs to be filled with streets (Figure 5). The po-
lutions for a given set of restrictions and goal func- sitions of the existing street connections are marked
tions. by nodes with underlined numbers. The areas de-
fined in the right-hand image in Figure 5 will be
Evaluation Mechanism used in later examples to define a central sub-area
As a goal function (or fitness function) for the evalu- (red, dashed) and a quiet sub-area (blue, checked).
ation of the generated street network we use the The red center is placed near the coordinates
Figure 4
Creation of new child variants
(C and D) by a crossover op-
eration applied to two parent
instruction trees (A and B).
Figure 5
Initial planning situation: The
border defines the planning
area, and connections to the
existing street network are
represented by the nodes with
underlined numbers. The col-
oured areas in the right-hand
image denote areas where the
new street network will have
defined properties.
750m/1500m (coordinate origin in the left bottom 6 and Figure 7 show three resulting street networks
corner) while the blue quiet area fills the bottom with the corresponding diagram of the develop-
right-hand quarter of our planning area. For the fol- ment of the fitness values.
lowing examples we use the following initial param- To achieve very high choice values, the most ob-
eters: generations = 50, population size = 50, muta- vious strategy is to design a network, that is separat-
tion rate = 0.25, crossover rate = 0.75, tree depth = 8. ed into two parts which are connected by the most
It is import to select a tree depth high enough to en- used street, the one with the highest choice value. In
sure that the complete area can be filled with streets cities we find this situations, for example, in places
and that there is no indirect restriction for the opti- where a bridge crosses a river or a narrow valley di-
mization algorithm. The values for the nodes of the vides a settlement. If we consider the three street
initial instruction trees were initialized with random networks in Figure 6 we can see this structure in the
values with the intervals: αi =[-10, 10], li =[10, 40], κi network in the right-hand image. Of the three net-
=[1, 4] works in Figure 6, however, the network in the mid-
First we consider a simple basic example scenar- dle has the best fitness value, although there are no
io, where we use the aforementioned optimization two separate parts. This results from the fact that for
method to maximize the maximum choice value the calculation of the trips we use the shortest an-
(Figure 6) and the sum of all choice values of the gular distance and not the shortest metric distance.
generated street network (Figure 7). We start with an Because of this, there is one street segment at the
empty area as shown in Figure 5 on the left. Figure top-center which is used very often. If we look at the
Figure 6
First example scenario. For
each of the street networks
shown, the maximum choice
value is maximized. Red
represents street segments
with high choice values and
blue low choice values. The
diagrams in the bottom row
show the development of
the fitness values over 50
generations.
Figure 7
Second example scenario: For
shown, the sum of all choice
values is maximized. Red
represents street segments
with high choice values and
blue low choice values. The
diagrams in the bottom row
show the development of
the fitness values over 50
generations.
left-hand network in Figure 6 we find that there is no the diagrams in the bottom row in Figure 7 is similar
single bottleneck, and at least one other populated to that of the corresponding diagrams in Figure 6.
route. As a result this network has the worst fitness This indicates that both fitness functions direct the
of the three. search process in a similarly efficient way.
The fact that the three best networks have differ- Our third example is based on the initial sce-
ent maximal fitness values (diagrams in the bottom nario with two defined sub-areas as shown in the
row of Figures 6 and 7), indicates that the evolution- right-hand image of Figure 5. In Figure 8 the central
ary optimization process explores different parts of sub-area is shown as a dotted ellipse and the quiet
the search space each time it is run. But despite the sub-area as a dotted rectangle. To include the spatial
small differences in the maximum fitness values of aspect in the fitness function, we have to define how
the variants, they all fulfill the requirements relative- to represent the graphical objects that represent the
ly well. When we consider the random points (repre- sub-areas with the corresponding specified proper-
senting randomly generated variants) we can clearly ties.
see the advantage of using the evolutionary search First we consider the central sub-area. To
process compared to randomly generated solutions. achieve a highly-populated center in the defined
The best variants are improved continuously over sub-area we want to locate the street segment
the 50 generations and reach a level, which cannot with the maximized choice value in it. Therefore we
be achieved by a random generation process. measure the distance dcmax of the street segment
In our second example we use the same initial with the maximum choice value cmax to the center.
scenario as in the first one, but we adapt the fitness This distance can be used as a weight so that we can
function to maximize the sum of all choice values of decrease the fitness of a network according to the
the street network. The topologies of the resulting distance dcmax:
networks in Figure 7 are clearly different to those
in Figure 6. Here we cannot see separate network , (1)
parts and the streets segments with the highest where D is a constant which denotes the maxi-
choice values are not concentrated at one location mal possible distance. In our examples, this is the
but distributed across the network. This difference diagonal of the border rectangle D = 3606m.
proves that our optimization system is working as Secondly, we consider the quiet sub-area. To
expected. The development of the fitness shown in achieve an area with as little traffic as possible, e.g.
Figure 8
Third example scenario. For
shown, we use a combined fit-
ness function: one factor is the
maximized maximum choice
value that is weighted with the
distance to the central area
(dotted ellipse), while the sec-
ond factor is the average sum
of all choice values which are
assigned to street segments
inside the quiet area (dotted
rectangle). Red represents
street segments with high
choice values and blue low for residential usage, we want to have only streets solution in the left-hand image of Figure 8. Here the
choice values. The diagrams with low choice values in it. Therefore we sum all maximum choice value is at the edge of the central
in the bottom row show the differences of the choice values from the street seg- area and there is a relatively populated road in the
development of the fitness ment ci that are located inside the quiet area A and quiet sub-area. Maybe the optimization algorithm
values over 50 generations. the maximum choice value of the network cmax. could have found a better solution with more gen-
The average of this sum is used as the second part of erations. But the combined fitness function may be
our fitness function: hindering the improvement of this variant. We will
discuss this problem in the next section.
. (2)
From the two fitness values F1 and F2 we calculate DISCUSSION
the final fitness value as the sum of both: As outlined in the description of our framework, we
Fitness = F1 + F2. (3) have demonstrated a method of representing plan-
The street networks resulting from this optimi- ning requirements using graphical objects that can
zation process are shown in the top row of Figure 8. be used by an optimization system (Figure 5). The
The results fulfill both of our requirements relatively main challenge of the system is interacting with de-
well: the red coloured street segments with maxi- sign variants, not because of the complicated user
mum choice values are located close to or inside the interface – it is, for example, possible to change the
central sub-area, while we find primarily only blue genotype and thus the later optimization process by
coloured street segments with low choice values in manipulating the graphical objects of the pheno-
the quiet sub-area. To evaluate the effect of the de- type (street segments and crossroads). This makes
fined sub-areas on the resulting street networks we it possible to realise a multi-directional planning
can compare the variants from Figure 6 and Figure 7 method as described above.
with the ones in Figure 8. We can observe very dif- The main problem of our system is that the op-
ferent structures in comparison to the second ex- timization process is much too slow for use in an
ample scenario in Figure 7 and similar structures to interactive process. The computation of the above
our first example scenario in Figure 6 showing the examples needed 2030 minutes on an average mod-
two separated network parts (Figure 8 in the mid- ern notebook. One generation therefore needed
dle). In general the results seems self-evident, but half a minute: half a second would be a more accept-
nevertheless we can see some problems, e.g. at the able timespan. Of course these times depend a lot
on the size of the street network, the population size significant effect on the optimization process und
and other aspects. But the main time-critical aspect thus on the quality of the results. For example the
is the computation of the all-pair shortest path. This optimization process can get stuck in local optima,
needs to be improved in future using an optimized since one criterion is already very good, but the
algorithm and more powerful hardware. other not. The improvement of poor criteria may be
In addition we use a very inefficient method to hindered because it may negatively affect other very
generate instruction trees. We use a random initial good criteria, so that the resulting fitness value can-
process which produces a very huge tree from which not be improved. To avoid these kind of problems
only a small fraction of nodes are needed to grow we need to use evolutionary multi-criteria optimiza-
the street network. In the examples above, in the tion (EMO) methods (Deb, 2001).
case of κi = 4, we have 3^depth instruction nodes for
each tree. For a tree depth of 8 this results in max. CONCLUSION AND OUTLOOK
3^8 = 6561 nodes, but we only have approximately In this paper we have demonstrated the potentials
300 street segments. Alternatively one could create of using an optimization system for urban planning
random but meaningful street networks in the be- tasks using a test scenario. In this scenario we have
ginning and encode them to make much more ef- generated street networks with defined local prop-
ficient instruction trees. erties. The presented system is a first component of
Variations of the angles and placement of the an framework with basic functionality to efficiently
initial street segments as shown in Figure 5 have a search compromise solutions for complex planning
relatively significant impact on the further growth problems. A first software prototype has been im-
and thus on the final phenotype of the network. plemented with an intuitive user interface to repre-
Therefore, to search for optimal solutions it may also sent planning problems, to present various compro-
be useful to vary the initial segment. mise solutions, and to improve them interactively.
Another interesting aspect of the presented The differences in the examples presented in
examples is a product of the property of EAs to cre- Figures 6-8 show clearly that our system doesn’t
ate their own biotope for the artificial life forms – in generate globally optimal solutions – e.g. one can
our case the street network. We can observe special delete connections that enable ring trips around the
strategies for the EA to maximize their fitness (the centre to increase the traffic through the centre (and
choice values): the first is to maximize the num- to increase the corresponding choice value). This is
ber of street segments to produce more trips and an inherent aspect of EAs: they cannot guarantee
thus higher maximal choice values. This could be finding the globally best solutions, but they can
overcome by averaging the values i.e. dividing the always offer good ones. This disadvantage can be
choice values by the number of streets. The second improved by running more generations or by using
is to generate street segments at strategically ben- separate populations in parallel and migrating the
eficial places (e.g. top left corner in the left and right best variants between them. In our context, this isn’t
networks in Figure 8). These segments can produce a problem because planners are not usually looking
more trips via certain segments with high choice for global optima as goal functions represent only a
values to increase them further. part of a planning problem. Thus the interactive and
In our last example (Figure 8) we have used two adaptable search for variants is the main support for
goal functions: one to achieve a center and one to the planning process.
create a quiet area. Both are combined into one The next step for the development of our frame-
fitness function. Here we run into the problem of work is to implement a more complex EMO sys-
weighting both criteria against each other in a more tem which integrates algorithms for parceling and
or less arbitrary way. This weighting, however, has a building placement (Aliaga et al., 2008; Knecht and
Koenig, 2012). With this development we can achieve Parish, Y. I. H., and Müller, P. (2001). Procedural Modeling of
the multi-level approach illustrated in Figure 1. Cities. Paper presented at the SIGGRAPH, Los Angeles,
CA.
ACKNOWLEDGEMENT Radford, A. D., and Gero, J. S. (1988). Design by optimization
Special thanks go to our colleague Christian Tonn in architecture, building, and construction. New York:
from the Bauhaus-Universität Weimar, who imple- Van Nostrand Reinhold.
mented the fast graph calculations for GPU. Rutton, D. (2010). Evolutionary Principles applied to Prob-
lem Solving. Retrieved 18.06.2011, from http://www.
REFERENCES grasshopper3d.com/profiles/blogs/evolutionary-prin-
Aliaga, D. G., Vanegas, C. A., and Beneš, B. (2008). Interactive ciples
example-based urban layout synthesis. Acm Transac- Turner, A. (2001). Angular Analysis. Paper presented at the
tions on Graphics, 27(5), 1-10. 3rd International Space Syntax Symposium, Atlanta.
Bentley, P. J., and Corne, D. W. (2002). An Introduction to Turner, A. (2007). From axial to road-centre lines: a new rep-
Creative Evolutionary Systems. In P. J. Bentley and D. W. resentation for space syntax and a new model of route
Corne (Eds.), Creative Evolutionary Systems (pp. 1-76). choice for transport network analysis. Environment and
San Francisco: Morgan Kaufmann. Planning B: Planning and Design, 34(3), 539 – 555.
Deb, K. (2001). Multi-objective optimization using evolution-
ary algorithms: John Wiley and Sons. [1] http://www.grasshopper3d.com/ (Retrieved 01.02.2013)
Derix, C. (2009). In-Between Architecture Computation. In- [2] http://www.bentley.com/en-US/Promo/Generative%20
ternational Journal of Architectural Computing, 7(4). Components/default.htm
Dillenburger, B., Braach, M., and Hovestadt, L. (2009). Build- [3] http://www.esri.com/software/cityengine (Retrieved
ing design as an individual compromise between quali- 01.02.2013)
ties and costs: A general approach for automated build- [4] Galapagos is a plugin for evolutionary optimization for
ing generation under permanent cost and quality control. Grasshopper/Rhino3D: http://www.grasshopper3d.
Paper presented at the CAADFutures 2009. com/group/galapagos (Retrieved 01.02.2013)
Knecht, K., and Koenig, R. (2012). Automatische Grundstück- [5] http://en.wikipedia.org/wiki/Inverse_problem (Re-
sumlegung mithilfe von Unterteilungsalgorithmen und trieved 01.02.2013)
typenbasierte Generierung von Stadtstrukturen. Weimar: [6] Bundesamt für Landestopografie, swisstopo (Art. 30
Bauhaus-Universität Weimar. GeoIV), © 2011 swisstopo
Koltsova, A., Tuncer, B., Georgakopoulou, S., and Schmitt, G. [7] http://www.aforgenet.com/ (Retrieved 21.05.2013)
(2012). Parametric tools for conceptual design support [8] http://www.shodor.org/petascale/materials/UPMod-
at the pedestrian urban scale. Paper presented at the ules/dynamicProgrammingPartI
eCAADe.
Evo-Devo in the Sky
Patrick Janssen
National University of Singapore, Singapore
patrick@janssen.name
Abstract. Designers interested in applying evo-devo-design methods for performance

based multi-objective design exploration have typically faced two main hurdles: it’s too
hard and too slow. An evo-devo-design method is proposed that effectively overcomes the
hurdles of skill and speed by leveraging two key technologies: computational workflows
and cloud computing. In order to tackle the skills hurdle, Workflow Systems are used that
allow users to define computational workflows using visual programming techniques.
In order to tackle the speed hurdle, cloud computing infrastructures are used in order
to allow the evolutionary process to be parallelized. We refer to the proposed method as
Evo-Devo In The Sky (EDITS). This paper gives an overview of both the EDITS method
and the implementation of a software environment supporting the EDITS method. Finally,
a case-study is presented of the application of the EDITS method.
Keywords. Evolutionary algorithms; multi-objective optimisation; workflow system;
cloud computing; parametric modelling.
INTRODUCTION
Evolutionary design is loosely based on the neo-Dar- may be computationally expensive and slow to ex-
winian model of evolution through natural selection ecute.
(Frazer, 1995). A population of individuals is main- Designers interested in applying evo-devo-
tained and an iterative process applies a number of design methods for performance based multi-ob-
evolutionary procedures that create, transform, and jective design exploration have typically faced two
delete individuals in the population. main hurdles: skill and speed (i.e. “it’s too hard and
Evo-devo-design differs from other types of evo- too slow!”). From a skills perspective, the require-
lutionary approaches with regards to the complexity ment for advanced interoperability engineering and
of both the developmental procedure and the eval- software programming skills is often too demand-
uation procedures. The developmental procedure ing for designers. From the speed perspective, the
generates design variants using the genes in the requirement for processing large numbers of design
genotype (Kumar and Bentley, 1999). The evaluation variants can lead to excessively long execution times
procedures evaluate design variants with respect to (often taking weeks to complete).
certain performance metrics. These procedures will Previous research has demonstrated how these
typically rely on existing stand-alone programs, in- hurdles can be overcome using a VDM procedural
cluding Visual Dataflow Modelling (VDM) systems modelling software called Sidefx Houdini (Janssen
and simulation programs (Janssen and Chen, 2011; and Chen, 2011). Firstly, a number of simulation
Janssen et al., 2011). In many cases, these systems programs were embedded within this VDM system,
thereby allowing designers to define development initialisation task also creates the initial population
and evaluation procedures without requiring any of design variants.
programming. Secondly, the evolutionary algorithm The growth and feedback tasks are used to pro-
was executed using a distributed environment, cess design variants in the population. The growth
thereby allowing the computational execution to be task will take in just a single individual with a geno-
parallelized. type and will generate a phenotype and a set of per-
Although the research demonstrated how the formance scores for that individual. (In the proposed
challenges of skill and speed could be overcome, method, the processes of development and evalua-
the solution was specific to the software tools being tion are thus defined as a single growth workflow.)
used, in particular Sidefx Houdini. Furthermore, for The feedback task will take in a pool of fully-eval-
most designers, the proposed approach remained uated individuals and based on a ranking of those
problematic due to the fact that they do not have individuals will kill some and will select some for
access to computing grids. This paper will propose a generating new children. With just these two tasks,
generalized method for evo-devo-design that over- a huge variety of evolutionary algorithms can eas-
comes these limitations. The method uses two key ily be specified. For example, if the pool size for the
technologies: computational workflows and cloud feedback is equal to the population size, then a gen-
computing. In order to tackle the skill hurdle, com- erational evolutionary algorithm will result, while if
putational workflow management systems are used, pool size is much smaller than the population size, a
called Scientific Workflow Systems (Altıntaş, 2011; steady-state evolutionary algorithm will result.
Deelman et al., 2008). In order to tackle the speed The first hurdle that EDITS must address is the
hurdle, readily available cloud computing infrastruc- skills hurdle. The initialisation, feedback, and termi-
ture is used. We refer to the proposed method as nation tasks are highly standardized and can there-
Evo-Devo In The Sky (EDITS). fore be generated automatically based on a set of
The next section will focus on the proposed ED- user-defined parameters. The growth task on the
ITS method, followed by a section describing the other hand is highly problem-specific and there-
implementation of a prototype EDITS environment. fore needs to be defined by the user. In order to
The final section will briefly present a demonstra- overcome the skill hurdle, the EDITS method uses
tion of how the method and environment can be a Workflow System for defining these tasks. Work-
applied. flow Systems allow users to create computational
procedures using a visual dataflow programming.
EDITS METHOD Users are presented with a canvas for diagramming
An EDITS design method is proposed that over- workflows as nodes and wires, where tools are rep-
comes the hurdles of skill and speed in a generalized resented as a nodes, and data links as wires.
way that is not linked to specific proprietary soft- Furthermore, this approach can also be used
ware applications. to flexibly link together existing design tools such
The EDITS method enables users to run a pop- as CAD and simulation programs. Interoperabil-
ulation-based evo-devo design exploration process. ity issues can be overcome by using data mappers,
This requires four computational tasks to be gener- whereby the output data from one tool may be
ated that will automatically be executed when the linked to the input data of another tool via a set of
evolutionary process is run: initialisation, growth, data transform, aggregation, and compensation
feedback, and termination. The initialisation and ter- procedures. This approach therefore allows paramet-
mination tasks are executed at the start and end of ric modelling tools to be linked to simulation tools
the evolutionary process respectively, and perform through an external coupling, which affords the user
various ‘housekeeping’ procedures. In addition, the greater flexibility in tool choice and linking options.
The second hurdle to be overcome is the speed building workflows; in batch mode, VisTrails
hurdle. The evolutionary process consists of a con- can be used to execute previously defined
tinuous process of extracting individuals from the workflows without requiring any user interac-
population, processing them with the growth and tion.
feedback tasks, and inserting the updated and new • A set of design tools, including CAD tools (such
individuals back into the population. Since the tasks as Houdini or Blender) and simulation tools
are independent from one another, they can easily (such as Radiance, EnergyPlus, and Calculix).
be parallelized. Cloud computing infrastructures al- (Other popular commercial CAD tools could
low users to have access to computing grids on an also be integrated with this environment. How-
on-demand basis at a low cost and can therefore be ever, due to inflexible licensing policies, it is
used to enable such parallelization. In the proposed currently difficult to deploy such tools in the
EDITS method, cloud computing is used for distrib- cloud.) The CAD tools can typically run either
uting the execution of both the growth and feed- in interactive mode or in batch mode while the
back tasks. simulation programs run only in batch mode,
with all interaction being restricted to text
EDITS ENVIRONMENT based input and output files.
In order to demonstrate the EDITS method, a pro- The EDITS environment is delivered as a cloud
totype EDITS environment has been implemented. based service. Cloud computing can deliver services
Three key type of software are used: a distributed to the user at a number of different levels, ranging
execution environment called Dexen, a workflow from delivering computing infrastructure to deliv-
system called VisTrails, and a set of design tools, ering fully functional software (Rimal et al., 2009).
such as CAD and simulation programs. These levels are typically divided into three catego-
• Dexen is a highly generic Distributed Execution ries: Infrastructure as a Service (IaaS), Platform as
Environment for running complex computa- a Service (PaaS) and Software as a Service (SaaS).
tional jobs on grid computing infrastructures, These levels can also build on one another.
previously developed by the author (Janssen et The EDITS environment is divided into three lay-
al., 2011). Dexen uses a data-driven execution ers, corresponding to IaaS, PaaS and SaaS, as shown
model, where tasks are automatically execut- in Figure 1. For the base IaaS layer, the EDITS envi-
ed whenever the right type of data becomes ronment uses Amazon EC2, which is a commercial
available. Dexen consists of three main com- web service that allows users to rent virtual ma-
ponents: the Dexen Client provides a graphi- chines on which they run their own software. Ama-
cal user interface for managing jobs and tasks; zon provides a web-application where users can
the Dexen Server manages the population and manage their virtual machines, including starting
orchestrates the execution of jobs; and Dexen and stopping machines. The SaaS and PaaS layers
Workers execute the tasks. will be described in more detail below.
• VisTrails is an open-source workflow system
that allows users to visually define computa- The SaaS layer
tional workflows (Callahan et al., 2006). VisTrails The SaaS layer consists of a number of graphical
uses a dataflow execution model that is well- tools for running EDITS jobs. Overall, there are four
suited to the types of procedures that need to main steps for the user: 1) starting the server, 2) cre-
be defined. It also provides good support for ating the growth task, 3) executing the evolutionary
integrating existing programs. VisTrails can be job, and 4) reviewing progress of the job.
used in one of two modes: in interactive mode, Step 1 involves using the Amazon EC2 web ap-
VisTrails provides a graphical user interface for plication to start an EDITS server. This simply con-
Figure 1
The three layers of the EDITS
environment.
sists of logging onto the Amazon EC2 website with start the EDITS job in the cloud and then reconnect
a standard browser, and then starting an Amazon in- with the running EDITS job intermittently in order to
stance. The operating system and software installed download the latest results. A plugin has therefore
on a virtual machine is packaged as an Amazon Ma- been implemented for VisTrails that adds an EDITS
chine Image (AMI), and for EDITS a customized AMI menu to the menu bar for starting EDITS jobs. When
has been created. This AMI is saved on the Amazon a new job is started, the user can select the growth
server, so it can simply be selected by the user from workflow, and can specify a number of parameters,
a list of options. The same server can be used for including population size, mutation and crossover
running multiple jobs. probabilities, selection pool size and the ranking al-
In step 2, the user defines the growth task by gorithm. Once these parameters are set, a number
creating a workflow with the VisTrails workflow sys- of Python scripts required to run the job are auto-
tem using a set of specially developed EDITS nodes. matically generated and uploaded to the server to-
Figure 2 shows an example of such a workflow, con- gether with the growth workflow. The job will then
sisting of a development procedure followed by start running automatically.
three parallel evaluation procedures. The develop- In step 4, the user connects to the EDITS jobs to
ment procedure uses SideFX Houdini to generate review progress and analyse the data that is gener-
the phenotype. The evaluation procedures use the ated. Dexen has its own client application with a
Radiance, Calculix, and EnergyPlus simulation pro- graphical user interface that allows users to get an
grams to generate performance scores. These proce- overview of all the jobs that are running and to in-
dures will be explained in more detail in the section terrogate the execution of individual tasks in detail,
describing the demonstration. providing information on execution time, crashes,
Step 3 involves executing the EDITS job. For the error messages, and so forth. Data related to in-
user, it is good if this execution could be orches- dividual design variants can also be downloaded.
trated from within the same VisTrails environment. However, downloading and viewing design variants
However, since the EDITS job may take several hours one at the time can be tedious and error prone. In
to execute, it is preferable to interact with it in an order to streamline this process, a set of VisTrails ED-
asynchronous manner. The user should be able to ITS nodes have been created for downloading data
Figure 2
The EDITS growth workflow in
the VisTrails environment.
and design variants directly from the server running customised AMI was created for EDITS with all nec-
in the cloud. These nodes can for example be used essary software preinstalled and all settings precon-
to create a workflow that first downloads the per- figured. The EDITS AMI includes the base operating
formance scores of all design variants and then se- system, together with Dexen, VisTrails, and a set of
lects a subset of these design variants for display to commonly used CAD and simulation programs.
the user. VisTrails provides a visual spreadsheet that The software used for orchestrating distributed
can be used to simultaneously display 3D models of execution of the EDITS job is Dexen. When the EDITS
multiple design variants (Figure 5). server is started on EC2, Dexen will be automatically
started and all the other required software will be
The PaaS Layer configured and available. The two main tasks that
The PaaS layer builds on top of the Amazon EC2 IaaS need to be executed are the growth and feedback
layer, by defining an AMI for the EDITS Platform. A tasks. Dexen maintains the population of individu-
Figure 3
The software layers involved in
executing the growth task. The
workflow, highlighted in grey,
is the only layer that needs
input from the end-user.
als in a centralized database and will automatically CAD and simulation programs, VisTrails will then in-
schedule the execution of growth and feedback voke these design tools. For the end-user, the com-
tasks. For the growth task, individuals are processed plexity of the growth task is hidden, since they are
one at a time. For the feedback task, individuals are only required to create VisTrails workflow.
processed in pools of individuals, selected randomly
from all fully evaluated individuals in the population. EDITS DEMONSTRATION
Each time either a growth or feedback task needs to As a demonstration of the EDITS approach, the de-
be executed, Dexen will extract the individuals from sign for a complex residential apartment building is
the database, and send them to an available Dexen evolved. The case study experiment is based on the
worker for processing. Once the worker has com- design of the Interlace by OMA. The design consists
pleted the task, the updated and/or new individuals of thirty-one apartment blocks, each six stories tall.
will be retrieved and reinserted back into the popu- The blocks are stacked in an interlocking brick pat-
lation database. tern, with voids between the blocks. Each stack of
The Python scripts for the initialisation, growth, blocks is rotated around a set of vertical axes, there-
feedback, and termination tasks are automatically by creating a complex interlocking configuration.
generated by EDITS. The growth task is the most Each block is approximately 70 meters long by
complex due to the various layers that are involved. 16.5 meters wide, with two vertical axes of rotation
The task has a nested ‘Russian Doll’ structure, con- spaces 45 meters apart. The axes of rotation coin-
sisting of a cascade of invocations three layers deep, cide with the location of the vertical cores of the
as shown in Figure 3. The outer layer consists of the building, thereby allowing for a single vertical core
Python script. When this script is executed, it will to connect blocks at different levels. The blocks are
invoke VisTrails Batch Mode in order to execute the almost totally glazed, with large windows on all four
workflow. Since this workflow may contain numer- facades. In addition, blocks also have a series of bal-
ous nodes that link to other design tools such as conies, both projecting out from the facade and in-
Figure 4
The initial configuration based
on the original design, consist-
ing of 31 blocks in 22 stacks of
varying heights.
set into the facade. The initial configuration, shown generated in skeletal form with a minimum amount
in Figure 4, is based on the original design by OMA. of detail. The full detailed model is then generated
The blocks are arranged into 22 stacks of varying only at the end, once the decision chain has finished
height, and the stacks are then rotated into a hex- completing.
agonal pattern constrained within the site bounda- In the decision chain encoding process, the
ries. At the highest point, the blocks are stacked four placement of each of the 31 blocks is defined as
high. a decision point. The process places one block at
For the case study, new configurations of these the time, starting with the first block on the empty
31 blocks were sought that optimise certain perfor- site. At each decision point, a set of rules is used to
mance measures. For the new configurations, the generate, filter, and select possible positions for the
size and number of blocks will remain the same, but next block. Each genotype has 32 genes, and all are
the way that they are stacked and rotated can differ. real values in the range {0,1}. In the generation step,
A VisTrails growth workflow was defined that per- possible positions for the next block will be created
formed both development and three evaluations. using a few simple rules. First, locations are identi-
The workflow shown in Figure 2 was developed for fied, and second orientations for each location are
this demonstration. identified. The locations are always defined relative
to the existing blocks already placed, and could be
Growth workflow: design development either on top of or underneath those blocks. The ori-
For the procedural modelling of phenotypes, SideFX entations are then generated in 15° increments in a
Houdini was used. For the genotype to phenotype 180° sweep perpendicular to either end of the exist-
mapping, an encoding technique was developed ing block. In the filtering step, constraints relating to
called decision chain encoding (Janssen and Kaushik, proximity between blocks and proximity to the site
2013). At each decision point in the modelling pro- boundary are applied, thereby ensuring that only
cess, a set of rules is used to generate, filter, and se- the valid positions remain. In the selection step, the
lect valid options for the next stage of the modelling decision gene in the genotype chooses one of the
process. The generate step uses the rules to create a valid block positions.
set of options. The filter step discards invalid options The resulting phenotypes consist of simple po-
that contravene constraints. The select step chooses lygonal models. Three separate files are generated,
one of the valid options. In order to minimise the one for each of the simulations. These models rep-
complexity of the modelling process, options are resent different sub-sets of information relating to
the same design variant. These sub-sets of infor- ous loading conditions. In order to reduce the
mation are selected in order to match the data re- computational complexity, the building con-
quirements of the simulation programs. In order to figuration is modelled in a simplified way, by
facilitate the data mapping, custom attributes are grouping individual structural elements into
defined for geometric elements in the model. For larger wholes called super-elements (Guyan,
example, polygons may have attributes that de- 1965). The performance criterion is defined as
fine key characteristics, such as block (e.g. block1, the minimisation of the maximum strain within
block2), type (e.g. wall, floor, ceiling), and parent (e.g. the structure.
the parent of the shade is the window; the parent of • Minimisation of cooling load: An evaluation is
the window is the wall). These attributes are used by defined that executes EnergyPlus in order to
the mapping nodes in order to generate appropri- calculate the cooling load required in order to
ate input files for the simulations. The geometry to- maintain interior temperatures below a cer-
gether with the attributes are saved as JSON files (i.e. tain threshold for a typical schedule. In order
simple text files). to reduce the computational complexity, an
ideal-load air system together with a simplified
Growth workflow: design evaluations zoning model is used, and the simulation is run
For the multi-objective evaluation, three perfor- for a periods of one week at the solstices and
mance criteria were defined: maximisation of day- equinoxes. The performance criterion is de-
light, minimisation of structural strain, and minimi- fined as the minimisation of the average daily
sation of cooling load. These performance criteria cooling load.
have been selected in order to explore possible con- In Figure 2, the three workflow branches defin-
flicts. For example, if the blocks are clustered close ing the evaluation procedures are shown. Each eval-
together the cooling load will decrease due to inter- uation procedure includes two mapper nodes: an
block shading but the daylight levels will also re- input mapper for generating the required input files,
duce. If the blocks are stacked higher, then they are and an output mapper for generating the final per-
likely to get better daylight but they may become formance score. These mapper nodes are currently
less structurally stable. The three performance crite- implemented as Python scripts, but part of this re-
ria are calculated as follows: search is the development of a graphical application
• Maximisation of daylight: An evaluation is de- for defining mapper nodes. See Janssen at al. (2013)
fined that executes Radiance in order to cal- for more details.
culate daylight levels on all windows under a The input mappers transform the JSON files
cloudy overcast sky. The amount of light enter- from the developmental procedure to the appropri-
ing each window is then adjusted according to ate input files for the simulations. As well as the ge-
the visual transmittance of the glazing system ometry information from these JSON files, the map-
for that window. The performance criterion is pers also require other material information. The
defined as the maximization of the total num- output mappers transform the raw simulation data
ber of windows where the light entering the into performance scores: for the Radiance data, the
window is above a certain threshold level for mapper calculates the number of windows below
reasonable visual comfort. the daylight threshold; for Calculix, the mapper cal-
• Minimisation of structural strain: An evalua- culates the maximum strain in the structure; and, for
tion is defined that executes Calculix in order EnergyPlus, the mapper calculates the average daily
to calculate the global structural behaviour cooling load. These three evaluation scores are then
using Finite Element Analysis (FEA) under vari- provided as the final output of the growth task.
Figure 5
A set of design variants shown
in the visual spreadsheet tool
within VisTrails.
Results CONCLUSIONS
When running the job, the population size was set For designers, the EDITS approach allows two key
to 200 and a simple asynchronous steady-state evo- hurdles of skills and speed to be overcome. First, it
lutionary algorithm was used. Each generation, 50 overcomes the skills hurdle by allowing designer to
individuals were randomly selected from the popu- define growth tasks as workflows using visual pro-
lation and ranked using multi-objective Pareto rank- gramming techniques. Second, it overcomes the
ing. The two design variants with the lowest rank speed hurdle by using cloud computing infrastruc-
were killed, and the two design variants with the tures to parallelize the evolutionary process. The
highest rank (rank 1) were used as parents for repro- demonstration case-study shows how the EDITS ap-
duction. Standard crossover and mutation operators proach can be applied to a complex design scenario.
for real-valued genotypes were used, with a muta- Future research will focus on the development
tion probability being set to 0.01. Reproduction be- of VisTrails data analytics nodes. This would allow
tween pairs of parents results in two new children, users to create workflows to perform various types
thereby ensuring that the population size remains of analysis on the data generated by the evolution-
constant. ary process, including hypervolume and clustering
The evolutionary algorithm was run for a total analysis.
of 10,000 births, taking approximately 8 hours to
execute. The final non-dominated Pareto set for the REFERENCES
whole population contained a range of design vari- Altıntaş, İ 2011, Collaborative Provenance for Workflow-driv-
ants with differing performance tradeoffs. en Science and Engineering, PhD Thesis, University of
A workflow was created in order to retrieve and Amsterdam.
display designs from the Pareto front. A selection of Callahan, S, Freire, J, Santos, E, Scheidegger, C, Silva, C and
design variants are shown in Figure 5. Vo, H 2006, ‘Vistrails: Visualization Meets Data Man-
agement’, Proceedings of the SIGMOD, Chicago, pp. Janssen, PHT and Kaushik, V 2012, ‘Iterative Refinement
745–747. Through Simulation: Exploring Trade-off s Between
Deelman, E, Gannon, D, Shields, M and Taylor, I 2008, ‘Work- Speed and Accuracy’, Proceedings of the 30th eCAADe
flows and e-Science: An Overview of Workflow System Conference, pp. 555–563.
Features and Capabilities’, Future Generation Computer Janssen, PHT and Kaushik, V 2013, ‘Decision Chain Encod-
Systems, pp. 528–540. ing: Evolutionary Design Optimization with Complex
Frazer, JH 1995, An Evolutionary Architecture, AA Publica- Constraints’, Proceedings of the 2nd EvoMUSART Confer-
tions, London, UK. ence, pp. 157–167.
Guyan, RJ 1965, ‘Reduction of Stiffness and Mass Matrices’, Janssen, PHT, Stouffs, R, Chaszar, A, Boeykens S and Toth
AIAA Journal, 3(2), pp. 380–380. B, ‘Data Transformations in Custom Digital Workflows:
Janssen, PHT, Basol, C and Chen, KW 2011, ‘Evolutionary De- Property Graphs as a Data Model for User‐Defined
velopmental Design for Non-Programmers’, Proceed- Mappings’, Intelligent Computing in Engineering Confer-
ings of the eCAADe Conference, Ljubljana, Slovenia, pp. ence - ICE2012, pp. 1–10.
886–894. Kumar, S and Bentley, PJ 1999, ‘The ABCs of Evolutionary
Janssen, PHT, Chen, KW and Basol, C 2011, ‘Iterative Virtual Design: Investigating the Evolvability of Embryog-
Prototyping: Performance Based Design Exploration’, enies for Morphogenesis’, Proceedings of the GECCO, pp.
Proceedings of the eCAADe Conference, Ljubljana, Slove- 164–170.
nia, pp. 253–260. Rimal, BP, Eunmi, C and Lumb, I 2009, ‘A Taxonomy and Sur-
Janssen, PHT and Chen, KW 2011, ‘Visual Dataflow Model- vey of Cloud Computing Systems’, Proceedings of the
ling: A Comparison of Three Systems’, Proceedings of the INC, IMS and IDC Joint Conference, pp. 25–27.
CAAD Futures Conference, Liege, Belgium, pp. 801–816.
Algorithmic Design Generation
Algorithmic Design Generation - Volume 2 - Computation and Performance - eCAADe 31 | 215

216 | eCAADe 31 - Computation and Performance - Volume 2 - Algorithmic Design Generation
The Potential of Evolutionary Methods in Architectural
Design
Wassim Jabi1, Barbara Grochal2, Adam Richardson3
Welsh School of Architecture, Cardiff University, United Kingdom
http://www.cardiff.ac.uk/archi/
1
jabiw@cardiff.ac.uk, 2sar8bag@cardiff.ac.uk, 3sar8ar@cardiff.ac.uk
Abstract. In this paper we examine the potential of combining 2D shape packing

algorithms and evolutionary methods in the design process. We investigate the ways
such algorithms can be used in architectural design and how they may influence it. In
the first part of this paper we introduce the theoretical framework of packing algorithms
and genetic algorithms as well as the traditional design process and the nature of design
problems. In the second part of the paper we introduce a software prototype that tests
these algorithms in two contexts: the preliminary design of a shading façade pattern and
the design of commercial housing layouts. The aim for both experiments was to generate
optimal configurations based on user-defined criteria without resorting to exhaustive
search. Several lessons were learned that point to the potential of evolutionary methods
in architecture as well as the limitations of such methods. We conclude the paper with
recommendations for further developing this research project.
Keywords. Evolutionary design; genetic algorithm; packing algorithm; scripting.
INTRODUCTION
In general terms, a genetic algorithm (GA) can be within a set boundary. In one variation of the prob-
characterised as a highly parallel and adaptive evo- lem, a shape-packing algorithm is designed to pack
lutionary search method. GAs are described as par- as many shapes as possible, without overlapping
allel searching methods because they search for them, and attempts to achieve a required minimum
solutions using the whole population of possible coverage area to minimize waste (Lodi et al., 2002).
options as opposed to altering a single potential so- In mathematics, circle packing focuses on the ge-
lution (Frazer, 1995). Since the most favourable solu- ometry and combinatorial character of packing of
tions are obtained by progressive alterations within circles of either equal or arbitrary size (Stephenson,
the same population over time, Frazer also refers to 2005). For circles of equal size, it has been math-
them as adaptive. Due to the mentioned character- ematically proved that a hexagonal honeycomb ar-
istics, GAs are becoming more popular and are be- rangement of circles produces the highest density
ing researched and increasingly applied to practical (Hsiang, 1992). In architecture, shape packing can be
problems. used in many pattern-based problems where densi-
Shape packing algorithms are optimization ty, number of packed elements and spatial relation-
methods that attempt to pack shapes together ships between elements is important.

The aim of this paper is to study the potential of Figure 1
combining 2D shape packing algorithms and genet- Mapping genotypes in the
ic algorithms (GA) in the design process. It investi- search space to phenotypes in
gates the ways such algorithms can be used as tools the solution space.
for aiding architectural design and how these meth-
ods may influence the architectural design process
itself. This will be done by conducting two experi-
ments based on the constructive design method-
ology where the two ‘constructs’ tested would be sign process is not an easy task mainly due to the
a software prototype that combines a 2D shape- fact that design problems are classified as ‘wicked’
packing algorithm and a genetic algorithm tested in (Rittel and Webber, 1973). This term refers specifical-
two experiments. This paper discusses some of the ly to the disciplines of social planning, politics and
advantages as well as limitations of such tools as de- design. Firstly, in most of the cases the design prob-
sign aids. lems cannot be comprehensively formulated. This is
due to the fact that nowadays the design process of
BASIC STRUCTURE AND SEQUENCE OF a specific building involves collaboration between
GENETIC ALGORITHMS different parties, which hinders arriving at specific
Genetic algorithms consist of two separate spaces: requirements early in the project development. Usu-
the search space, containing genotypes, and the so- ally, the design problems appear and become clear-
lution space, containing phenotypes (Bentley, 1999). er as the process proceeds. Secondly, since design
The genotypes, which are the coded solutions to the is a collaborative effort between different parties, it
problem, have to be mapped onto the actual solu- has to unite what are sometimes radically contradic-
tions i.e. the phenotypes, which are in the solution tory interests. It has to take into account a number
space (Figure 1). Mapping refers to the process of as-
signing the genotypes from the search space to cor- Figure 2
responding phenotypes in the solution space. This Basic genetic algorithm
has to happen before the fitness of each solution can sequence.
be evaluated. The fitness of a solution is assessed
according to the fitness function that assigns scores
to all solutions. The more suitable the solution to
solve the problem at hand, the higher is the fitness
score (Mitchell, 1995). Effectively, the solutions with
higher scores will have a greater probability of be-
ing selected and reproduced in the next generation
(Figure 2).
THE TRADITIONAL DESIGN APPROACH

AND ‘WICKED PROBLEMS’
In the second half of the 20th century researchers
brought the design process to the focus of scientific
study (Cross, 2007). They investigated it and outlined
the basic sequence of actions involved in it, mainly
in order to introduce new aiding tools and regulate
it. This research has proven that systemising the de-

of other factors such as moral and social aspects, first experiment explains the basic logic and func-
aesthetic impact and sustainability. Thus, the design tioning of the GA based on façade pattern design.
solutions cannot be rationalised since design is not This supplied valuable data for the discussion of the
merely a pragmatic problem-solving or simple opti- advantages and limitations of this tool. The second
misation leading to one right solution. Even though experiment concerns a more realistic case of hous-
usually one solution is sought, the possibilities for ing layout design based on a real-life master plan.
arriving at it are limitless. Thirdly, because there is no It has to be pointed out here that both of the cases
linear sequence for a design process (Lawson, 2005), present an integrated approach towards evolution-
there is also no apparent beginning or end to it. The ary design put into practice. The experiments will
information needed to make decisions is never fully focus on first designing a pattern or layout and then
complete and thus the state of the design problem optimising it based on the design criteria set by the
is constantly evolving. Furthermore, in modern de- designer.
sign thinking, problems and solutions are deemed
to emerge together during the design process. That EXPERIMENT 1: FAÇADE PATTERN DE-
is, finding design solutions may cause other, “high- SIGN
er-level” problems somewhere else. Therefore the The design of the first experiment involves the crea-
design process involves finding a balance between tion of a panel façade of size 10,000x10,000 mm. The
solving some problems in one place and causing un- main design goal is to achieve 40-50% of the area
desirable effects somewhere else. Rittel and Webber coverage of the designed pattern in order to provide
emphasize that these ‘wicked’ design issues, unlike the required shading. The second design aim is to
science, depend heavily on the designer’s subjective design the pattern with 2000 circles of various radii.
value judgements. These main characteristics are Both of the design goals have to be achieved follow-
obstacles when working with algorithms and coming basic criteria set up at the beginning of the pro-
puter programs that need specific requirements and cess using the custom-designed software (Figure 3).
clearly defined rules in order to perform their tasks. These are as follows:
• Min. Radius: 50 mm
THE CONSTRUCTIVE METHODOLOGY • Max. Radius: 8000 mm
This part of the paper will describe two algorith- • Buffer: 20 mm
mic experiments employing custom software cre- The range for the minimum and maximum radii was
ated by the authors using the MAXSCRIPT scripting decided bearing in mind that the wider the range
language for Autodesk 3ds Max. The software inte- the more variety will be sustained in the population.
grates a genetic algorithm with a shape-packing These two values could also represent the radii of
algorithm that operates on any 2D boundary. The the smallest and biggest drills used for making the
pattern. The reasons for choosing the radii range can
Figure 3 be very different. The ‘buffer’ parameter refers to the
The user interface of the area between circles where no other elements are
custom-developed genetic allowed.
software in Autodesk 3ds Max.
Genotype and phenotype
Since the design aims at creating a façade pattern
made out of circles of various sizes the radius is the
only information contained in the genotype (Figure
4). As shown, the genotype is the number falling
within the specified range, where the phenotype

is the assigned representation of the genotype – in Figure 4
this case it is a circle of that particular radius. Exemplary genotypes (radii)
and phenotypes (circles).
Fitness function
In order to encourage variety within the panel the
fitness function favours the circles with radii as close
to the minimum or maximum radius values as pos-
sible. This will secure more diversity within the pop-
ulation and will create more interesting patterns.
Thus, the fitness function is defined as follows:
(1)
where F is the fitness of the individual, Ri is the ra-
dius of the individual circle, and Rav is the average of
the specified minimum and maximum radii.
(50,000). The section below describes the results of
The GA sequence the four tests created based on the rules described
The initial population is created randomly, covering above.
the entire range of possible solutions (search space).
In case of the script used for this experiment the Results
new individuals were created using a circle-packing We conducted four tests in order to meet the design
algorithm until the maximum number of attempts requirements and solve the stated design problem:
for fitting more individuals has been reached (in this Achieving area coverage of 40-50% with 2000 cir-
experiment it was set at 50,000 attempts). In such cles. In each test, we iterated through four genera-
a case usually the initial population does not reach tions (Table 1).
the maximum number of individuals (in this case
2000 individuals). After the initial population has Discussion and Comments
been generated the fitness of each individual is cal- As the results show, meeting both of the design
culated. The obtained fitness scores are then used goals where the fitness function is awarding the
for selecting the fittest individuals and placing them radii from the extremes of the range of 5-800 cm, is
in the mating pool. We specified a constant 50% sur- rather unlikely to be achieved in the span of 4 gener-
vival rate and a 1% mutation rate throughout the ations even if the number of attempts is 50,000. The
experiment and implemented a “roulette wheel” se- outcomes might have been different if the number
lection method to select the fittest candidates while of attempts was increased to 100,000 or more. This
maintaining a similar diversity to the one found in is, however, an area for further research that lies out-
natural selection. After the individuals for the mat- side of the scope of this experiment.
ing pool have been selected the process of repro- Compared to the non-optimised first genera-
duction begins using crossover and mutation of tion of packed circles, it is evident from conducting
their genotypes. The process of generating popula- only four tests, that applying the GA dramatically
tions continues until a termination condition is met. increases the number of circles to meet the goal to
Termination takes place either when the population pack 2000 individuals within the prescribed bound-
target is met or when the algorithm reaches the ary (Figure 5), but that has two side-effects: 1) The
maximum number of attempts to fit the individuals average radius of circles decreases, and 2) the over-

Table 1 Test Generation Packed Area Coverage Avg. Fitness Avg. Radius
Results of running four tests No. No. individuals (%) (mm)
each with four generations. 1 1 195 60.26 0.98 193.42
1 2 1489 44.28 0.98 86.55
1 3 2000 34.73 0.98 70.47
1 4 2000 27.82 0.98 64.84
2 1 212 55.78 0.95 190.03
2 2 1401 45.42 0.98 90.24
2 3 2000 34.62 0.98 70.15
2 4 2000 26.72 0.98 63.20
3 1 217 50.66 0.95 186.06
3 2 1449 49.36 0.98 83.92
3 3 2000 31.34 0.98 66.57
3 4 2000 25.09 0.98 61.68
4 1 204 57.93 0.95 192.08
4 2 1479 45.90 0.98 86.58
4 3 2000 34.20 0.98 69.44
4 4 2000 27.52 0.98 64.19
all coverage area of these circles decreases as well smaller circles. This occurred due to the fact that
(Figure 6). Also, it can be concluded that the bigger the fitness function was awarding both extremes
the coverage area, the smaller the number of ele- – the smallest and the biggest circles. The mating
ments. In all of the tests the maximum number of pool quickly biased itself towards smaller circles af-
circles was achieved when the coverage area was ter the first generation because at the point when
consistently below 40%. Based on that, the main de- the maximum number of elements was reached
sign goals had to be revised. Because the GA proved there were far more circles with radii closer to the
that both design goals couldn’t be achieved simul- minimum than those closer to the maximum radius.
taneously, the designer has to decide which is a pri- That is, since there were a larger number of smaller
ority – the coverage area or the number of packed circles and because they were considered just as
elements. Since the main aim of the experiment fit for breeding as large circles, there was a higher
was based on creating the required shading pat- probability of choosing them for breeding the next
tern, the coverage area took precedence. Therefore population. This strength in numbers phenomenon
the façade pattern with the coverage area within initiated a vicious cycle of breeding smaller and
the range and achieved with the biggest number smaller circles while larger circles quickly became
of circles was chosen as the proposed design solu- extinct. The solution seems to approach a plateau
tion. In the four conducted tests, this was achieved after the third generation. An interesting contradic-
in the second generation of the third test with 1449 tion is that the overall results did not improve with
packed individuals and 49.36% coverage area (Fig- the subsequent generations even though the indi-
ure 7). viduals’ fitness was increasing. From an interesting
It is clearly visible from both the data and the perspective, this result supports a case for diversity
visual graphs (Figure 6) that even after the maximum where even if individual fitness is high, the overall
number of packed elements has been achieved the performance of the population is unsatisfactory due
GA was still breeding a population of increasingly to a lack of diversity.

Figure 5
A visual representation of
the test results. The blue
circles represent the parents
chosen for breeding the next
generation.

Figure 6
Left: Graph representing the
decreasing average radius
over four generations in all
four tests.
Right: Graph representing
the decreasing coverage area
over four generations in all
four tests.
Figure 7 houses around the perimeter of these areas and ori-

The proposed façade pattern ents them to the nearest edge (representing a road).
based on overall results. The fitness of each individual is calculated based
upon their proximity to their nearest neighbour. This
factor of ‘remoteness’ ensures that the density of
houses in subsequent generations increases.
The program was tested on a case study housing
development that has recently been granted plan-
ning permission. For the purpose of this research,
this provided a realistic framework onto which the
program could be applied. The program is able to
vary the size of the houses within what has been de-
fined as a realistic range, based upon the house sizes
found in the case study. The intention is to achieve
a more realistic configuration of houses. The input
parameters were as follows:
Phenotype Parameters:
• Min. Radius: The minimum separation distance
EXPERIMENT 2: HOUSING LAYOUT DE- between houses.
SIGN • Min./Max. Length: The range of lengths of the
The second experiment explores the application of a houses being generated.
genetic algorithm to the design of a housing layout. • Min./Max. Width: The range of widths of the
As an early stage test, it ignores the vast number of houses being generated.
variable factors that influence the design of such • Dist. To Road: The minimum distance from the
schemes and focuses on achieving the greatest den- road at which houses can be placed.
sity of housing units. This ‘House Packing’ script ena- Population Parameters:
bles the user to define a series of areas (representing • Maximum: Sets the maximum number of hous-
buildable plots of land), the program then places es to be generated.

• Survival Rate (%): Controls the percentage of on the other hand will usually remain relatively low,
the population that will survive and be allowed since the proposed solution still contains a number
to breed the next generation. of poor performing individuals. The results recorded
• Mutation Rate (%): Controls the chances of an in this experiment appear contradictory, as the high-
individual within the population to mutate. est average fitness score is achieved by the 100%
• Max. Attempts: The maximum number of at- survival rate. The problem with such a high survival
tempts that the computer is given to place the rate is that the algorithm is relying entirely on the
houses correctly. mutation of individuals to increase the average fit-
ness. It is more of a brute force trial and error ap-
Results proach rather than systematically breeding a better
The two variables that control the effectiveness of solution. This method appears to work for this par-
the genetic algorithm are the survival rate and the ticular experiment, as the fitness score is very closely
mutation rate. To find optimal values for these, we linked to the number of individuals placed (more
carried out a series of evaluations that first tested individuals = lower factor of remoteness = higher fit-
the system at varying survival rates and then vary- ness score). However the purpose of this experiment
ing mutation rates. We tested the program over five is to test the potential of the genetic algorithm, and
generations on the case study layout. For each set- thereby following a method that nullifies part of the
ting, we recorded the three read-outs: ‘Packed’, ‘Cov- breeding process would be contrary to that goal.
erage Area’ and ‘Average Fitness’. This gave an indica- We found that the optimal survival rate was 60%
tion of how well each rate was performing, though and mutation rate was 40%, this coincides with ex-
the most telling result was the average fitness score periments by other researchers that suggest a mu-
as this is more directly linked to the overall efficacy tation rate of approximately 50% (Elezkurtaj and
of the algorithm. It would seem that the rate show- Franck, 1999). Using these settings, we ran the algo-
ing the greatest increase in average fitness over con- rithm for 15 generations to discover the effective-
secutive generations should be selected as the opti- ness of the optimisation process. The results show
mal setting. Over five generations this would appear a general positive trend in the average fitness score
to be the 100% survival rate. However, a survival rate and number of packed houses, indicating that the
of this magnitude stifles the genetic algorithm by optimisation process is functioning correctly (Figure
preventing it from removing poor performing indi- 8). What is interesting is the amount of fluctuation
viduals. As a result the values for ‘Packed’, and ‘Cov- between generations. These results indicate a pat-
erage Area’ tend to peak very early, and more often tern of 3-4 successive increases followed by a signifi-
than not exceed those achieved by lower survival cant decrease, the magnitude of which reduces with
rates, as the program attempts to squeeze more and each repetition. This is an indicator of how the ge-
more houses onto the site. The average fitness score netic algorithm works. Over successive generations,
Figure 8
Left: Generation 1. Number
Packed: 364, Area Coverage:
25.61%, Average Fitness: 2.56.
Middle: Generation 7. Number
Right: Generation 15. Number

the range of fitness scores will decrease as they all marginalising their role; they feel that “If anything
become fitter (and the average fitness increases). the development of IT in design has given more
The result of which, in terms of the roulette wheel control back to the designers”. This further reinforc-
selection method, is that each individual has a much es the role of the program as a design aid, not as a
more equal chance of being in the percentage of the complete design solution.
population which survives to the next generation
(the survival rate). Conversely, they also have a more CONCLUDING REMARKS
equal chance of being removed from the popula- The two experiments we have conducted reinforce
tion. This is demonstrated in the results where after the notion that evolutionary methods have many
3 generations at a 60% survival rate, new individuals advantages. Most importantly, the experiments il-
with lower fitness scores replace 40% of the relative- lustrated that evolutionary methods are indispen-
ly high scoring population. sable when dealing with a large potential solution
space. Rather than conducting a manual and ex-
Discussion and Comments haustive search for the best solution from a large
It is clear that a genetic algorithm based design aid data set, evolutionary methods allow the designer
holds great potential in increasing the efficiency of to set target goals and input parameters and rules
the commercial housing design process. The abil- that act together to search the population for the
ity of this kind of software to act as a catalyst for best possible candidates and use that pool of can-
design ideas whilst simultaneously conforming to didates as an input to breed an even better solution.
a plethora of constraints is something that, as the Genetic algorithms allow designers a more precise
need for greater efficiency and precision within the method to achieve the desired goal as it faithfully
design process grows, is going to prove invaluable. applies the states rules and precisely measures the
The most pressing question raised by this research is performance of individuals and the overall popu-
the way in which the software should be integrated lation with each generation. Given their parallel
into the design process. search nature, genetic algorithms can help us speed
For this experiment we shared the results with the whole design process when the pool of options
a group of architectural practitioners to gauge their is large. Additionally, evolutionary methods are ca-
reaction. Unsurprisingly, this expert consultation pable of supplying very surprising outcomes aiding
exposed a desire amongst the designers to have the designer’s creativity and suggesting new solu-
a greater amount of input in the generation of a tions. Finally, due to their high efficiency, evolution-
solution. This gave an interesting insight into the ary methods allow the design more time to focus on
way that they feel about the software. One expert the quality of the design, omitting menial tasks such
questioned the ability of software to replicate the as ensuring that the proposed design is compliant
“human ability to … make a subjective judgement”, with the stated goals and constraints.
demonstrating a lack of trust in the system to gener- The experiments also exposed some important
ate a complete design solution. The designers want limitations of these methods. Due to the ‘wicked’ na-
the software to carry out the time consuming, me- ture of design problems, it is not always feasible to
nial tasks, enabling them to focus their time on the state and code clear design rules and objectives. The
more skilled areas of design, but do not wish the design process is highly dynamic and often chang-
software to shift all decision making from human to es course. Although possible in future iterations of
machine. The lack of trust also answers the question the software, the genetic algorithm code we have
of the potential marginalisation of the architect’s developed was not designed to handle changing
role through the advancement of digital design design goals between generations. The addition of
tools. The designers do not see the technology as new parameters and fitness score methods between

generations could prove even more difficult to im- Elezkurtaj, T and Franck, G 1999, ‘Genetic Algorithms in
plement. As most computer-based systems, genetic Support of Creative Architectural Design’, Proceedings
algorithms cannot replace the tacit knowledge, of the eCAADe conference, Liverpool, United Kingdom,
common sense, and intuition of human designers. pp. 645-651.
In particular, these methods cannot replace hu- Frazer, J 1995, An Evolutionary Architecture, Architectural As-
man judgment since it is difficult to encode in the sociation, Boston.
algorithm, as it is not based on clearly definable Hsiang, WY 1992, ‘A simple proof of a theorem of Thue
rules. Yet, we believe these methods fundamentally on the maximal density of circle packings’, E2,
change the design process and the role of the de- L’Enseignement Mathématique, 38, pp. 125-131.
signer. We envisage the designer mutating from the Lawson, B 2005. How Designers Think: The Design Process De-
role of the supreme creator of the final outcome to mystified, Elsevier/Architectural Press, Amsterdam.
the role of the maker of rules. In partnership with Lodi, A, Martello, S, and Monaci, M 2002, ‘Two-dimensional
sophisticated evolutionary systems, the designer packing problems: a survey’, European Journal of Op-
can then explore the plethora of solutions offered erational Research, 141, pp. 241–252.
by these methods fluidly shifting its rules and input Mitchell, M 1996, An Introduction to Genetic Algorithms, MIT
parameters that in turn alter its path of evolution. Press, Cambridge, Massachusetts.
Rittel, HWJ and Webber MM 1973, ‘Dilemmas in a general
REFERENCES theory of planning’, Policy Sciences, 4, pp. 155-169.
Bentley, PJ 1999, Evolutionary Design by Computers, Morgan Stephenson, K 2005, Introduction to Circle Packing: The The-
Kaufmann Publishers, San Francisco. ory of Discrete Analytic Functions, Cambridge University
Cross, N 2007, Designerly Ways of Knowing, Birkhäuser Press, Cambridge.
GmbH, Boston.

Genetic Algorithms Applied to Urban Growth Optimization
Solar envelope and solar fan

Patricia Camporeale
Laboratory of Sustainable Architecture and Habitat, Faculty of Architecture and Urbanism,
National University of La Plata, Buenos Aires Argentina
http://www.arquinstal.com.ar/layhs.html, http://www.disenobioambiental.com.ar
patricia@disenobioambiental.com.ar
Abstract. This work is a research on the application of genetic algorithms (GA) to urban
growth taking into account the optimization of solar envelope and sunlight in open spaces.
It was considered a typical block of a Spanish grid, which is the most common subdivision
of the urban land in towns situated in Argentina. Two models are compared, one in which
the growth has no more limitations than building codes. The other one, in which the
growth incorporates the solar radiation as a desirable parameter.
This way of parameterizing configures a bottom-up method of urban growth. No top-down
decisions intervenes in the growth process.
This tool proves to be useful at early stages of urban planning when decisions—which will
influence along the development of the city for a long time—are taken.
Keywords. Genetic algorithms; solar envelope.
INTRODUCTION
This work analyses the application of genetic algo- distributed in the blocks around the town core.
rithms (GAs) (Mitchell, 1998) to parametric urban Our case study , Lincoln, is a town situated in Ar-
design, considering access to solar radiation. Pa- gentine Pampa. Towns founded after Independence
rameterization allows the continuous control of the period (1816) like Lincoln (1865) (Tauber, 2000),
design process and evolutionary algorithms (Russel, followed the Spanish pattern. This town, bases its
2010) optimize the solar access. We develop a simu- economy on soya crops and the cash flows that this
lation of urban growth according to current restric- activity produces every year are reflected in the eco-
tions and building codes (Leach, 2004). nomic growth (Forrester, 1970).
Spanish towns in Argentina share a same urban With this data, the author built an urban model
pattern: a 45º-rotated grid (Randle, 1977). This grid is which reaches maximum growth. The study is ap-
suitable for a plain land with no mayor geographic plied to two blocks in the town centre area (Figure
accidents like our grassy prairies. A main square, 1). There is a wide range of uses situated in this area:
which derives from the Spanish “plaza”, determines commercial, housing and office buildings. The high-
the centre of the grid and the main public buildings est density is concentrated in this zone. The inten-
are situated around: the church, the government of- tion is to show how an urban block can grow differ-
fices, the school, and the police station. Housing is ent if environmental criteria are applied.

Figure 1
Lincoln zoning: high density
in centre town area: offices,
housing, trade, lower densities
concentrically situated: hous-
ing, trade, ,small industries,
outside: rural areas.
These towns have experienced a fast growth solar radiation in open spaces in the block core. It
during last years as soya prices began to rise. Urban considers the growing season until the average first
development has occurred without any planning frost date which is 1st April in the Southern Hemi-
except for the current law that regulates the use of sphere. The intersection between these two shapes
land, which was promulgated in the ´70s. This is the defines a buildable volume.
Decree- Law 8912/77 [1] and it does not reflect envi- Another simulation with no solar restrictions is
ronmental issues, as the solar envelope concept [2]. developed. The comparison between the two mod-
In this work, solar envelope allows urban dwell- els permits the urban designer to take into account
ings to satisfy a two-hour period of solar radiation solar access from the very beginnings of the codifi-
(11AM to 1PM) all year round (Niemasz, J et als, cation process and to reflect this issue in the build-
2011). Solar fan [3] is another tool which defines ing code, making the necessary corrections to the
a volume of a four-hour period (10 AM to 2 PM) of current one.

Figure 2 the plot is placed. In our case, as the block is in the
Types of blocks in Lincoln centre town, TBF is 2.75. In order to simplify the pro-
urban grid. cess, each buildable volume is divided in modules
which occupy 0.6 plot surface and rise up to com-
plete the maximum allowed. Maximum height is
24m or approximately eight-storey high.
TBF can be increased if the floor plan surface
is smaller than the allowed by LOF. This coefficient
also increases if the building is receding, leaving an
open space in the front. These regulations promote
to build higher receding buildings that shade into
the lower neighbour ones and their backyards. It can
THE CURRENT BUILDING CODE also allow appearing isolated buildings separated
The urban land is private except for public buildings. 4m from each borderline, in which case, environ-
The owner must follow the restrictions and recom- mental conditions in open spaces tend to worsen.
mendations of the Building Code which refers to the
plots as the units that configure the urban block. THE URBAN MODEL
The law referred to the use of land lets the build- Lincoln is divided into a rectangular grid oriented
ing occupy at maximum 60% of the plot surface. A NE-SW. Unlike other Pampa towns, the grid is not
40% must be free for correct ventilation and day- square; but it is rotated 45° NE as most Spanish ur-
light. This coefficient is called Land Occupation Fac- ban grids in Argentina. This rectangular shape pro-
tor (LOF). No lateral free space less than 4m wide vokes that NE façades have best orientation as we
can be left as it is considered that the building can are in the Southern Hemisphere but the remaining
only receive daylight and ventilation from the front ones have not good orientation, especially the SW
urban space, backyards or lateral urban spaces. Win- and SE ones (Figure 2). Different kinds of blocks were
dows must ventilate and receive light from those analyzed from the town plan.
spaces exclusively. Dividing walls between different Blocks are 121m length and 81m width. Type I
plots must not have openings for ventilation and has sixteen plots of 618m2 each. Half the block looks
daylight. Therefore, they are blind. NE and the other half looks SW. In the case of mid-
Backyard dimensions are determined by an dle plots, a space in the backyard is left free to con-
imaginary line that crosses the plot 25m far front figure what is called the block-core. This open space
borderline, which constitutes the building limit and ensures daylight and ventilation to the rear façades.
configures the block core together with the other The buildings situated in the corner plots are not
plot backyards. This restriction makes the buildings compelled to leave this open space as they have two
tend to occupy the plot from one side borderline to façades looking to streets where to ventilate and get
the other one. daylight.
All these regulations make the individual build- Type II has twelve plots of 412 m2 and eight
ings to configure a solid volume like a ring with an plots of 618 m2. Four blocks look NE, another four
open space in the centre of the block called the look SW, six look SE and six look SW. Type III has
block core. twelve plots of 618 m2, six looking NE and six, look-
The maximum built surface allowed can be from ing SW. It has also six plots of 412 m2, looking NW.
1.2 times the plot´s surface up to 2.75 times, in our Type IV is a mirrored image respect to Type III.
case study. This coefficient is called Total Building Street width is 18m for NO and SE streets and
Factor (TBF). It depends on the urban zone where 23m, for NE and SW streets. The relation between

Figure 3
Solar envelope for 20th June
11AM to 1PM showing the
buildings according to current
building Code.
street width and building height is 3:2 for SE streets After that, we proceed to apply solar restric-
(18m) and 5:7.5 for SW streets (23m) for this specific tions. These solar tools determine a solar buildable
urban grid (121m x 81m). volume. It is designed for a two-hour period from 11
AM to 1PM.
METHODOLOGY The first tool solar envelope is applied to the
The objective of this work is to compare the two model. As it can be seen in Figure 3, the Code does
models: the built block according to the current not make any difference between different orienta-
Code and another one with solar restrictions and tions in order to regulate maximum height.
maximum growth for a Type I and a Type II blocks. As we observe the shades projected onto the
The whole drawing is parameterized in Rhinoc- adjacent streets —especially SE and SW— we can
eros © [4] by means of Grashopper [5]. In the first easily infer that they reach the façades of the block
case, an urban block is modeled following the cur- in front of ours (Figure 4). This image has been ex-
rent building Code (Figure 3). ported to Ecotect© by means of GECO [6], which is
Each plot reaches its maximum buildable vol- a Grasshopper plug-in. This image clearly shows
ume according to height limit (24m), LOF (0.6 plot´s shades and levels of direct radiation of buildings at
surface) and TBF (2.75 plot´s surface). No more limi- noon for June, 20th.
tations are taken into account. A wide variety of The other tool, the solar fan, is the void that has
buildings which exceed the bounding box appeared. to be left in order to assure a four-hour period of so-

Figure 4
Shades for 20th June (winter
solstice) at noon in the non-
optimized model.
lar radiation in open inner spaces of the urban block In Figure 7, the projected shade of the optimized
during growing season. This open space is called volume is analyzed in Ecotect. Shade on the SW and
‘block core’ in the Building Code and it is composed SE streets does not reach the façades of the blocks in
by all the backyards of the plots. Only corner plots front as they are limited by the solar envelope. The
are exempt from integrating this space. As it is ob- core block receives enough solar radiation during a
served in Figure 5, some volumes exceed the core four-hour period during growing season.
block when only considering current regulations. Type II block was also analyzed in the same con-
After this, a model considering solar envelope ditions as Type I and the results are shown in Figure
and solar fan was obtained by means of GAs. These 8. As the core block varies its dimensions, buildings
algorithms are used in optimization processes on NW and SE sides are not so restricted in their rear
(Goldberg, 1997). In this case, one called Galapagos facades.
has been run to maximize built volumes in each
plot. The buildable shape is obtained by the Boolean RESULTS
intersection between the solar envelope and fan. After modeling the normal Type I and II blocks and
Then, the algorithm develops the maximum allowed the optimized ones, we proceed to compare the re-
quantity of modules and their arrangements, consults between the experiences shown above.
sidering building regulations inside this bounding In Type I, while in the normal model, the maxi-
shape. The Boolean difference between the build- mum height is 24m or eight-storey in the whole
ings and this bounding box is minimized, reaching block, in the optimized one, it has to be reduced
in nearly every case, a value of zero. The results are to 15m or five-storey on SW façades to avoid shad-
shown in Figure 6. ing over the neighbour block. In SE façades, this is-

Figure 5
Solar fan showing volumes
that exceed the block core
(green areas) for non-opti-
mized model.
Figure 6
Optimized built volume inside
the solar bounding box.

Figure 7
Optimized model in Ecotect
showing the projected shad-
ows for 20th June, at noon}.
Figure 8
Type II block Boolean differ-
ence between solar envelope
and fan with optimized built
volumes.

Figure 9
Alternative buildings obtained
with GA for SE plot in block
Type II showing bounding box
in lines.
sue is even worse; the height is lowered to 12m to Two different types of blocks, I and II, are shown
maintain the built volume inside the bounding box. in Figure 10. Both types differ in the dimensions of
These buildings can be terraced up to eight-storey the block cores as the plots have different number of
height only on the rear part. plots and arrangements (Figure 2). As the core block
When analyzing Type II block, the portion of the increases in NW-SE direction, the buildings on these
buildings that exceeds the bounding box is shown façades have to diminish their height.
in red. It sums approximately 79 m2 (Figure 8). Even
when, in the plot 17, the GA improved the module DISCUSSION
arrangement, this plot cannot be completely built The use of these tools applied to urban design per-
up to 2.75 TBF. Maximum height should be reduced mits the improvement of environmental conditions
to seven-storey (21m) in plot 18 and six-storey in in buildings and open spaces. It can be applied to
plot 17 in the rear façades of the buildings. determine urban grid characteristics like:
The GA produced several alternatives that are • block dimensions
comprehended into the solar bounding box. Some • street widths
of these possible arrangements for SE plot are • maximum heights differentiated by orientation
shown in detail in Figure 9. The same procedure was • core block dimensions
followed with each plot. The solar envelope and fan as prescriptive tools
NE corner blocks are the only ones which can in urban design have to be studied together with
reach maximum height in their whole surface. The energy savings, developable density and infrastruc-
other ones have to be terraced. ture costs as well as local climate conditions. Our

Figure 10
Urban tissue with different
types of blocks: Boolean differ-
ence between solar envelope
and fan.
typical urban grid promotes adjacent buildings struction is brick masonry with reinforced concrete
which reduce heat losses through dividing walls. roofs.
This is the reason why this work considers the block
as an individual at urban level and as a whole at plot CONCLUSIONS
level. In this bottom-up process [7], the individuals Genetic algorithms provide a useful tool to test dif-
can perform certain actions regulated by the current ferent alternatives of buildable volumes inside a plot
legislation. The whole block acquires characteristics as modules were dimensioned as real architectural
that are the result of these individual behaviours. elements. They were built considering heights and
Emergent properties (Hensel et al., 2010) arise when depths as well as daylight and natural ventilation
this happens and consequences affect the urban conditions, regulated by current codes.
tissue performance, e.g. wind direction and inten- In order to widen the scope of this work, GAs
sity provoke specific microclimatic conditions that can also be used to optimize solar radiation in roofs
affects building ventilation, inner temperature and and façades (Camporeale, 2013) to install PV or solar
comfort in urban spaces. water collectors. Fenestration can also be optimized
Towns like Lincoln are benefited with this urban by orientation (Camporeale, 2012).
tissue as the blocks tend to be compact. Local cli- Current legislation on urban design and build-
mate is temperate dry with quite thermal amplitude. ing restrictions on urban plots deserves a deeper
This benefits savings in energy consumption as the study than fixing height limits, total buildable vol-
thermal mass accumulates heating in winter and ume and land occupation. Rules should be imple-
prevents overheating in summer. Traditional con- mented according to local conditions as climate,

technologies, transportation, energy prices and car- Niemasz, J., Sargent, J., Reinhart D.F., 2011, ‘Solar Zoning and
bon emissions. They must not be left aside when im- Energy in Detached Residential Dwellings’, Proceedings
plementing prescriptive tools as the studied above of SIMAUD 2011, Boston, USA.
in order to arrive to undesired results. In our case Randle, Patricio H., 1977, ‘La ciudad pampeana. Geografía
study, developable density has no influence on the histórica. Geografía urbana’, Oikos, Buenos Aires, Argen-
augment of cost transportation as it would have in tina.
a suburban neighbourhood. Concentration could Russell, S. et al, 2010, ‘Beyond classical search’ in Artificial
be a good tool in a metropolis but not in towns like Intelligence, A Modern Approach, 3rd ed., Prentice-Hall
Lincoln. Tools, as those described in this work, could Pearson Education Inc., NJ, USA, pp. 125-128.
improve comfort conditions, saving energy to reach Tauber, Fernando et als, 2000, ‘Lincoln, Reflexiones y datos
a better environment. para una estrategia de desarrollo’, National University of
La Plata, Buenos Aires, Argentina.
REFERENCES Tedeschi, A., 2011, ‘Parametric Architecture with Grasshop-
Camporeale, P., 2012, ‘El uso de algoritmos genéticos en el per’, Primer. Edizioni Le Penseur. Brienza. Italy.
diseño paramétrico de edificios energéticamente efi-
cientes: el coeficiente G y el consumo anual de energía’, [1] www.gob.gba.gov.ar/legislacion/legislacion/l-8912.htm
Proceedings of CIES, Vigo, Spain [2] www-bcf.usc.edu/~rknowles/sol_env/sol_env.html
Camporeale, P., 2013, ‘Genetic algorithms applied to urban Knowles, R , 1999, The Solar Envelope,
growth optimizing solar radiation’, Proceedings of PLEA [3] diva4rhino.com/user-guide/grasshopper/solar/,
2013, Munich, Germany. Jakubiec, A et als, (2011), DIVA Design Iterate Validate
Forrester, J., 1970, ‘Structure of an urban model’ in ‘Urban Adapt, Harvard University, Graduate School of Design,
Dynamics’, The MIT Press, MA, USA, pp.17-18. Department of Architecture, Solemma LLC.
Goldberg, D., 1997, ‘Some applications of Genetic Algo- [4] www.rhino3d.com, McNeel, R. and Ass., 2010, Rhinocer-
rithms’ in ‘Genetic Algorithms in Search, Optimization os NURBS Modelling for Windows Version 4.0, Service
and Machine Learning’, Addison-Wesley, MA, USA, pp. Release 8. Seattle USA.
88-145. [5] www.grasshopper3d.com/ McNeel, R. and Ass., 2010,
Hensel, M. et als, 2010, ‘Emergent Technologies and Design’, Grasshopper, Generative Modelling within Rhino Ver-
Routledge, NY, US. sion 0.9.0014, Seattle USA.
Johnson, S., 2001, ‘Emergence, The Connected Lives of Ants, [6] www.food4rhino.com/project/geco, Grabner, T et al,
Brains, Cities and Software’, Scribner- NY- US. GECO (2013)
Leach, N et als, (2004), ‘Design by Algorithm’ in ‘Digital tec- [7] www.re-ad.dk/files/39550564/Niels_M_Larsen_Thesis_
tonics’, Wiley-Academy, UK, pp. 79-87. screen.pdf, Larsen, N. (2012), Generative Algorithmic
Mitchell, M. 1998, ‘ Implementing a genetic algorithm’ in ‘An Techniques for Architectural Design, Aarhus School of
Introduction to Genetic Algorithms’, The MIT Press, MA, Architecture, Denmark,
USA, pp. 154-179

Design Tools for Integrative Planning
Stefana Parascho1, Marco Baur2, Jan Knippers3, Achim Menges4
1,2
University of Stuttgart, 3ITKE, University of Stuttgart, 4ICD University of Stuttgart
1,2
http://www.uni-stuttgart.de, 3http://itke.uni-stuttgart.de, 4http://icd.uni-stuttgart.de
1
s.parascho@itke.uni-stuttgart.de, 2marco@forcett.net, 3info@itke.uni-stuttgart.de,
4
mail@icd.uni-stuttgart.de
Abstract. The performance of an architectural object is highly difficult to both define

and measure in its complexity since it is integrating a constantly increasing amount of
information, from concrete measurable characteristics to the subjective perception of
individual users. The question arising though is how to predict the performance of a
building and influence the design in order to increase it according to a significantly high
number of criteria.
The presented paper proposes two design tools, both developed and programmed in rhino
python for the generation of freeform geometries. The tools are generated for specific
tasks, but may be interpreted as exemplary for a way of defining and structuring a design
program in order to increase its efficiency. Both tools rely on a computational core that is
generally defined and may be fed with as many and different constraints and criteria as
considered suitable for the defined task.
Keywords. Integrative design; evolutionary algorithm; agent-based system.
INTRODUCTION
Both measuring and changing a design’s performance, for an architectural object is a process differ-
mance is defined by such a high amount of informa- ent for each project, since context- and user-specific
tion that computational methods are of crucial im- factors vary constantly. Therefore generic programs
portance in this process. Computational tools may and tools can only cover a very general and unspe-
help process a much bigger amount of information cific area of the planning process and may only help
and variables than one may be capable to capture in the representation and simplification of a design,
intuitively and through classical design methods. therefore being insufficient for capturing the com-
The incorporation of digital tools should thus hap- plexity of an architectural object.
pen as early as possible in the design phase, even at Numerous established optimisation methods
the point of analysing the given task. Still, using the have been incorporated into architectural design
computational power of digital tools in architectural opening up a new dimension of solution options
design may prove itself more difficult than in other and a new freedom degree for designers, but at the
industries, since a building is usually an individual same time creating a new extremely complex prob-
and very context-dependent object. In architecture lem as to how such tools are to be implemented and
time and resources are often insufficient to allow further developed for the use in design tasks. Many
the development of highly performative designs questions arise, such as which method is best suit-
and design methods. Defining criteria of this perfor- able for architectural use, how this method is to be

translated and implemented, what the limitations tems, evolutionary algorithms have been of major in-
are and how these are to be defined in order to interest for architectural design from their first appear-
crease the creativity of a design and not to define a ance in computer science. Their main strength, which
too strict solution space limit. designers are trying to make use of, is the capacity of
Finally the paper proposes a comparison of two multi-criteria optimisation, therefore searching for a
acknowledged optimisation tools and their imple- solution that may fulfil more than one chosen crite-
mentation, pros and cons, for architectural design. rion. At the same time following the natural system
The purpose is to modify and further develop the as a model, such optimisation methods show the
known tools and optimisation methods as to find an immense potential that the natural systems have if
optimum implementation for specific design tasks. used as examples in design processes.
The use of evolutionary algorithms has in-
STATE OF THE ART creased in architecture in order to achieve an opti-
misation of desired criteria, starting with Frazer J.H.
Bionics in 1995. The final purpose was that their high poten-
The traditional top-down design process has already tial for the built environment should be made use of,
been questioned and a number of bottom-up strat- since they describe a much more complex system,
egies have been developed and implemented in with similarities to the natural one (Frazer, 1995)
architectural design having as purpose to include a
higher complexity into built design and create more Agent-based systems
performative and properties-specific results. Bionic A further development of evolutionary theory is the
processes are one of the main research areas in this theory of self-organization, both meant to describe
development since they draw a parallel between and explain complex and chaotic natural systems
the complexity of nature and architectural building (Frazer, 1995). The main principle of self-organiza-
systems. Looking at natural models as examples for tion implies that an organized system evolves out
specific characteristics and assets of these systems of a chaotic one only through the interaction of its
and implementing these into architecture through parts and subsystems, without any higher control-
an abstraction of its principles leads to a differently ling entity (Camazine et al., 2001). Agent models rely
hierarchized design process and a bottom-up meth- on the definition of a global system through its sim-
od that includes more information into the early sta- ple part, a so-called agent. These agents act accord-
ges of design (Knippers and Speck, 2012). In these ing to a set of given simple rules, interacting with
processes a specific natural phenomenon (e.g. self- the other agents and developing complexity and
organization), a characteristic of a biological system emergent behaviour in the system (Reynolds , 1999)
or a whole biological process (e.g. Evolution), is ana- [1]. Examples of such systems in nature are swarms
lysed and translated into design principles. (flocks of birds, schools of fish or ant colonies) where
all participating agents follow simple behavioural
Evolutionary Algorithms rules according to their neighbouring agents.
Evolutionary algorithms are widely used metaheuris- Although in computer science and mathematics
tic optimization algorithms developed in computer agent-based systems are a widely used search algo-
science and mathematics for problem solving (Ash- rithm for solving complex problems, in other fields it
lock, 2006). They are, similar to hill-climbing or simu- has been mostly used for simulation. In architecture
lated annealing, search algorithms meant to look agent-based systems have mainly offered solutions
through a solution space for the result of a complex for crowd-simulation and circulation systems, ensur-
optimisation problem. Developed also by John H. ing an improvement of circulations and spatial ar-
Holland (1995; 1998), as a reference to natural sys- rangement. or particle simulation in animations. Few

research projects have used agent-based systems be seen as an optimum to all chosen criteria but
for the generation of form/geometry [2] while these more as a compromise between the different crite-
systems often struggle with developing an autono- ria that depends a lot on the value of each criterion
mous non pre-defined system since one of the major that is predefined by the user. It is very important to
difficulties in using agent-based systems is ensuring regard such optimisation results not like in math-
the convergence of the system in a working solution. ematics as the singular and universal solution to a
complex problem, but as the specific, task adapted
DESIGN TOOLS and user dependent solution to one type of defini-
tion of the problem.
Evolutionary Design Strategies The process involved in using a genetic algo-
The first presented tool was designed for the gener- rithm includes three steps: firstly the definition of
ation of a high-rise project. Since the requirements a geometry generation algorithm, followed by the
for high-rise buildings present one of the most com- analysis after certain chosen criteria and the result-
plex systems in building design, a specific design ing fitness value, and lastly the recombination of the
tool enabling an evaluation and improvement of the fittest individuals resulting in a new geometry gen-
tower’s global performance was developed. eration. Repeating these steps until a desired fitness
As intended, the design strategy is based on a value is achieved determines the final geometry of
computational core that allows multiple use and the most successful individual. The geometry gener-
adaptation to a specific task. In this case a genetic ation algorithm describes the first and maybe most
algorithm was chosen for the core as a result of the influencing step of the genetic algorithm. Determin-
numerous and complex requirements. A tower’s di- ing the freedom degree of the geometry generation
mensions especially in height have as a result a high algorithm implies setting up a definition that allows
number of continuously changing criteria that need enough freedom for the algorithm to utilize as much
to be taken into account when designing a high-rise of the solution space as possible but at the same
building. Thus the form of the building has often lit- time ensure that the solutions are fully functioning
tle flexibility and is very difficult to influence. While and don’t escape the desired solution space.
the presented project regards the outer form as the For the chosen example the outer shape was
result of its inner constraints, the intention is to still modified with each iteration since the purpose was
have the possibility to choose the degree at which to ensure more flexibility in the form design of sky-
this form is influenced by the designer. At the same scrapers and to exploit as many options as possible.
time numerous criteria, sometimes contradictory, One of the major optimisation criteria is minimizing
make it very difficult to even only control these in- wind loads on the facade of the tower so a more
ner constraints so that the outer form often remains dynamical, organic base geometry, resulting out of
the pure functional result of these requirements. lofted curves was chosen. The geometry genera-
While evolutionary algorithms show great re- tion definition includes a various number of flexible
sults in optimisation problems with one criterion, parameters, such as the number of curves used, the
their power for architectural design lies in handling number of control points and the type of curves
more than one requirement and even working used. These are randomly set in the geometry defini-
through contradictory optimisation criteria. The tion so that one generation of individuals includes
computational capacity of handling an incredibly a set number of completely different geometries
high amount of information, comparing and chang- (Figure 1).
ing this information, makes it very useful for such a The second step includes the desired analysis
complex task. While genetic algorithms can instru- and determination of the fitness value of each indi-
mentalise contradictory criteria, the result is not to vidual. This procedure also represents a critical point

Figure 1
Geometry generation process.
since contradictory criteria cannot be 100% fulfilled and wind power in order to achieve as little wind
but need to be weighted as to how much one criteri- loads as possible on a facade pane but also to use
on shall be fulfilled in comparison to the other ones. the generated wind power with specifically located
Except for choosing and implementing appropriate wind turbines. This is one example of clearly con-
criteria, weighting these represents another step tradicting criteria in which less wind loads lead to a
that strongly influences the outcome of the GA and more stable structure but more wind power results
lies in the hands of the designer. Each specific task in more energy win. Weighting these criteria against
requires different analysis criteria and weighting each other could only be achieved after a number
and moreover a strong interdependency between of tests in order to understand the algorithms be-
the analysis criteria and the geometry generation haviour. In the end a minimal fitness value for wind
process. In the presented case the chosen criteria pressure and suction was chosen to be mandatory
include a wind load analysis, wind power analysis, so that the wind power generation was weighted
solar analysis, area and volume analysis and a num- less than the load analysis. Still using a parametric
ber of excluding absolute criteria, such as minimal definition wind turbines were located in the areas
radii in the facade and the gravitational centre for a with most wind power so that high energy efficien-
basic structural functioning of the building (Figures cy could be achieved.
2 and 3). The individual weighting of the criteria was For the recombination and regeneration of new
performed after numerous tests according to the individual generations a stochastic selection meth-
chosen purpose, not only to create a building as ef- od was chosen, such methods being acknowledged
ficient as possible but also focusing on wind loads and used for an optimal reach of a solution and con-

Figure 2
Fitness values of the fittest
Individual.
Figure 3
Resulting tower geometry.

Figure 4
Increment of fitness factors
over generations: wind loads,
individual, generation.
stant increment of the generations’ fitness values that while the presented genetic algorithm showed
(Pinsky and Karlin 2011) (Figure 4). effective results, it is still a linear process which fol-
The presented methods have proven to be ef- lows a direction (form generation – analysis – im-
fective for such a complex task as a high-rise project, provement) that happens on one hierarchic level and
but there were still numerous difficulties encoun- is furthermore based on creating a very high num-
tered along the way. One of the greatest challenges ber of random variants that are then compared and
resulted to be the black box character and random- analysed. The wish for further research was to break
ness of the written algorithm. While you can track the linearity of the algorithm and create a process in
the development of the algorithm and its success or which different hierarchical levels could interact and
failure, there are no means of intervening through- lead to a result without the need of many variants
out the running time or even logically following the but slowly adapting to the given requirements.
process of the genetic algorithm. It proved to be a
rather random process in which you could expect Agent-Based Process
that the fitness value of each individual will increase Based on the knowledge of the evolutionary algo-
but without any means of fast tracing why or how rithm developed in the precedent project, a more
much it increases. Many tests had to be run in order general and flexible tool was searched in this second
to manually try out the variable parameters, such as approach. A number of critical points discovered
the weighting of the fitness criteria or the geometry while using optimisation algorithms were defined
generation since it couldn’t be easily understood as crucial and created the basis for the second ap-
how one value influences the algorithm. While this proach. The exact purpose was to create a more flex-
is part of the power of computational means, of ible tool, with a computational core that could be
generating designs that cannot be intuitively traced extended and adapted to a given task through the
down, it remains a time consuming factor to set up addition of adaptation criteria and through chang-
all variable parameters so that a successful result will ing the input constraints. As a major point the wish
be achieved. It is a run and result process in which to destroy the linearity of such an approach and
the designer has no capability of interacting with create a process that allows input parameters from
the computational tool he created, it is meanly a tool various hierarchical levels and the communication
that needs to run through from start to end and can between all subsystems of a general system, served
only then be evaluated through its result. Further- as the starting point for this design method. While
more, a genetic algorithm needs a lot of adjustments the evolutionary algorithm allows numerous criteria
in order to simply provide a result that constantly in- to be included and considered, it has a clear differ-
creases its individuals’ fitness and does not converge entiation between the generating parameters, in
to a not satisfactory early result. One major point is this case the geometry generation, and the optimi-

Figure 5
Interaction of subsystems.
zation criteria. The generation is the one adapting to according to its own constrains and requirements,
all criteria so the information flow is unidirectional such as the smoothness of the surface resulting from
and does not allow other parameters to adapt to the angles of the different panels or the structural
the requirements of the generation method. The stability of the global force, curvature and height.
purpose of the agent-based tool is to allow this flow The second subsystem represents the panelling
of information from all input parameters into all di- elements, such as triangles or quads through which
rections and create a communication between all the global geometry is realized. These have flexible
participating subsystems, even located in different parameters such as shape, or planarity. These re-
hierarchical levels. quirements set by the designer are meant to inform
The chosen task is more general – also in order the other systems while the covering panel itself
to exemplarily represent the possibilities of such a shall change dimensions and orientation or loca-
method – and is intended to create a roof like grid- tion according to the requirements coming from the
shell structure over a given fictional site. The struc- other systems.
ture is divided into three representative subsystems The third subsystem describes a shading panel,
that are intended to show different hierarchical lev- meant to be representative for any type of facade
els of the general structure and their interdepend- panel reaching from a simple planar glass pane to a
encies intended to allow a continuous communica- complex shading element. This panel again defines
tion between these subsystems (Figure 5). a set of requirements such as planarity, dimensions
The first chosen subsystem is the global geom- or orientation. As mentioned before, these systems
etry, representing the freeform surface connected are simply exemplary and do not cover all complex-
to the predefined support locations. It defines the ity of a gridshell structure, but are meant to show
global shape and appearance of the final built result the possibilities of such a process.
and is not intended to be only a result of all other For the presented example a smooth surface
requirements and subsystems but set and adapt connecting three support locations with certain spa-

Figure 6
Generated surface without
and with geometric structural
adaption.
tial limitations, triangular covering panels and a sim- much double curvature as possible and through this
ple shading component were chosen to be imple- adaptation it defines the position and dimensions of
mented. The agent-based system was selected after the panels (Parascho et al., 2012) (Figure 6).
an extensive research for its capabilities of abstract- One of the high advantages of this method is
ing and simplifying complex behaviour into simple the flexibility of the tool and possibility to adapt it
basic rules. An agent is defined as the smallest part to a given task. Since the desire is to create a gen-
of the system (the division panel) and fed with nu- eral tool that may be changed and fed with numer-
merous rules representing all criteria of the partici- ous inputs and criteria, the agent-based system was
pating subsystems. These criteria were all translated extremely efficient in allowing such fast adaptions.
into geometric behaviour so that the agent con- Each behavioural rule may be added at any time
stantly reacts and adapts to the set of requirements during the process and may influence all defined
enabling a constant increment of these criteria. systems. It has also proven to be very powerful since
Chosen criteria include the smoothness of the new criteria can be implemented and tested very
surface, dimensions of the beams, the number of fast and the development can be traced simply by
beams coming together at a knot, geometric struc- watching the agents perform (Figure 7).
tural behaviour, static behaviour and lighting con- Still a number of points have proven to be dif-
ditions. The criteria were distributed to represent a ficult when implementing such a system. The first
specific subsystem or be external criteria in order to question arising is how to abstract such a complex
show the interdependency and connections of all model as a swarm system into a working algorithm
systems. For example, lighting conditions influence for a design purpose. It is of extreme importance
as well the shading pane that changes in size and how the singular agent is defined, what part of the
orientation as the global geometry that is created global system it represents and how flexible it is. De-
through the individual triangular panels that focus fining the basic agent has the highest effect on the
on achieving an orientation as parallel as possible to output, since too little flexibility may not ensure any
the light source. Similarly the static behaviour influ- result at all and too much will result in extreme so-
ences the global geometry that tries to achieve as lutions that may not be functioning as built objects

Figure 7
Resulting surfaces after differ-
ent adaption criteria.
CONCLUSION / COMPARISON the interaction with this system making use of the
When comparing the two methods the most impor- strength of agent-based systems to react and adapt
tant fact is to differentiate between the task types to any exterior influence at any time. The genetic
that each algorithm can be addressed with. Both al- algorithm is rather a model that strongly depends
gorithms proved to be functioning systems for gen- on the definition of the input parameters and offers
erating architectural design, but they were focusing one final solution to these options. For tasks where
on two distinct points. While the genetic algorithm the focus lies on the optimization process and where
is extremely good in handling a great number of cri- certain criteria need to be fulfilled as strongly as
teria by using the high computing capacities of the possible, evolutionary algorithms and their capac-
computer, the agent-based model is developed to ity of working with a high number of variants lead
work with less information but create constant con- to satisfying results. On the other hands, tasks that
nections between this information. The agent-based require more adaption and fast changes in the input
model’s greatest strength is abstracting any type of would rather benefit from the agent-based tool.
criteria of any hierarchical system into one equally While both systems led to successful results, the
hierarchized level at which all parts can exchange in- main difficulty encountered during the processes
formation. It does not work through numerous vari- was the correct definition of the input parameters.
ants, created with a random factor, as the genetic Whereas working through a complex solution space
algorithm does, but intends to constantly change opens up a lot more possibilities than traditional in-
and adapt in order to improve the characteristics de- tuitive design methods, this freedom of covering all
fined in the behaviour of the agent. possibilities is strongly limited by the definition of
One big difference between the presented each constraint, optimization criterion or behaviour
methods is the option provided by the agent-based definition. Most time and energy flows into defining
model of interacting with the system. The black- each parameter influencing the final output and its
box character of the genetic algorithm is broken as importance for the global tool. It is often a precon-
the designer can constantly follow the process of ception that making use of the complexity of an
the agent-based tool. Future research will focus on architectural object through computational meth-

ods automatically opens up an unlimited space of Frazer, J 1995, An Evolutionary Architecture, Architectural As-
solutions and options. The main issue which further sociation Publications, Themes VII.
research will address is how to define each compo- Holland J H 1995, Hidden Order - how adaption builds com-
nent of such a tool as to receive the optimum bal- plexity, Perseus Books.
ance between the freedom degree and the limita- Holland J H 1998, Emergence. From chaos to Order, Oxford
tions that allow it to be a functional built object. University.
Knippers, J, and Speck T 2012, ‘Design and construction
ACKNOWLEDGEMENTS principles in nature and architecture’, Bioinspiration &
The implementation of the agent-based model Biomimetics, 7.
would not have been possible without the help of Parascho, S, Baur, M, Knippers, J, Menges, A 2011, ‘Integra-
Ehsan Baharlou who strongly influenced the pro- tive Design Methods in Architecture’, in Proceedings of
gramming of the system, ICD University of Stuttgart the Scaleless Seamless Conference, Münster.
and the tool connecting Rhino to Sofistik developed Pinsky, M, Karlin, S, 2011, An Introduction to Stochastic Mod-
by Riccardo La Magna, ITKE University of Stuttgart. eling, Academic Press.
Reynolds, C W 1999, ‘Steering Behaviors For Autonomous
REFERENCES Characters,’ in the proceedings of Game Developers
Ashlock, D 2006, Evolutionary Computation for Modeling Conference 1999, Miller Freeman Game Group.
and Optimization, Springer, Vienna, New York.
Camazine, S, Deneubourg, J, Franks, N, Sneyd, J, Theraulaz, [1] http://www.red3d.com/cwr/steer/gdc99/.
G, Bonabeau, E 2001, Self Organization in Biological Sys- [2] http://cita.karch.dk/Menu/Projects/Digital+Formations/
tems, Princeton University. Lamella+Flock+(2010).

Infections
Parametric patterning and material behavior

Tuğrul Yazar1, Fulya Akipek2
İstanbul Bilgi University Faculty of Architecture, Turkey
1
www.designcoding.net, www.infections2.blogspot.com
1
tugrul.yazar@bilgi.edu.tr, 2fulya.akipek@bilgi.edu.tr
Abstract. This paper covers two workshops that are instances of a research on the
feedbacks between parametric patterning and material behavior. Infection sets the
conceptual background of these workshops utilizing pattern deformations as a generative
technique. Gridal Infection workshop focus on real-time dynamic patterns while Reflex
Patterning workshop integrates material performances to this exploration.
Keywords. Parametric patterning; material behavior; prototypes; fabrication; dataflow.
INFECTIONS: GENERATIVE DEFORMATIONS VIA PATTERNING

Patterns have been instruments of analysis and re- text. Infections is a series of workshops in which we
search in various disciplines, from social sciences study methods of interaction in-between digital and
to computer science, mathematics and biology. material to reveal the potentials of the context. The
Gleiniger and Vrachliotis (2009) state that the pat- conceptual background of infection -even if sounds
tern concept previously defined as a structural sys- like an invasion- is a challenge and springboard
tem of order began to gain a new complexity and for students to explore potentials via deformation
momentum in the light of cybernetics and system processes. This concept provokes them with sub-
theory. Reflections of these studies with the trigger- themes such as immune system, recovery, metamor-
ing of computational tools have shifted the notion phosis and becoming. The process begins with pat-
of pattern in architectural design realm. The classi- tern recognition, an attempt to observe and perceive
cal notion of pattern as, formal, ornamental, decora- existing orders of the context called host body. Next,
tive and geometric orders of repeating shapes has students are encouraged to manipulate the host
turned into the contemporary conception of pat- body in a creative way utilizing pattern deformation
tern as structural, sequential, distributed, or progres- techniques. Therefore, infection is both the meta-
sive systems of repeating units or processes (Garcia, phor and the method of these deformations, while
2009). Meanwhile parametric design tools have the host body represents a pre-defined system and
become essential to think and act on this broader the physical context to be infected.
sense of patterning by accelerating and expanding In the first workshop named Gridal Infection,
space of possibilities through variation and diversity. this manipulation is studied by projecting real-time
This paper covers two instances of a research dynamic patterns on the glass-brick walls of the fac-
that focus on the feedbacks between parametric ulty building at YTU.
patterning and material performances within a con- The second workshop, Re-flex Patterning is

Figure 1
FiberGrid; Screenshots.
based on the correlation of the digital with the during the design process.
physical via parametric patterning techniques and a In the final application, visual outputs are pro-
composite material system. jected on the host body, aligned to its existing grid.
The third and last step in the ongoing research Below are three student projects that are the prod-
will be pushing the limits through the fabrication ucts of this three day introductory workshop.
processes. Parametric patterning and CNC molded
tiles will be explored as a case study for the future FiberGrid
workshop. Students considered the host body as a dead tissue
In following sections, details of the first two of an organism, resembling the wall as a standing
workshops are explained, concluding with a discus- idle and reckless element to its environment. In order
sion on outcomes. to revitalize it surrounding sound is considered as an
injection that changes the inner structure of the or-
WORKSHOP 1: GRIDAL INFECTION ganism and transforms the grid lines into curvilinear
This initial workshop focuses on the abstract no- fibers.
tion of grid, sampled from an existing 16x11 unit Above concept of FiberGrid is realized by con-
glass brick wall, the host body. Students are asked structing a grid out of interpolated curves (Figure
to articulate its formal (grid / pattern / tessellation / 1). Surrounding sound is captured and used as a
reference), performative (transparency / light / struc- real-time input that bends the curves. The change
ture / function) and tectonic (ambient / kinetic / au- in sound level affects the process, creating tempo-
ral) properties. On the early phase, three groups of ral variations. Finally, a history enabled algorithm
students hunted concepts “lesion, plasma and fiber- captures sequences of this process, creating waves
grid”. Then, they are asked to develop their projects of fibers (Figure 2). As it is a recursive algorithm, it
by creating parametric deformations, utilizing real- responds concurrently, getting faster / slower and
time interactions with the context. Students with no more / less fibrous while the surrounding sound
previous skill on parametric design are introduced level rises / lowers (Figure 3).
with Grasshopper for dataflow parametric mod-
eling and Firefly add-on for interaction design. They Lesion
ended up with three dynamic patterns, superim- In this project, the grid is considered as cellular
posed to the existing wall. The semi-opaque mate- forms packed together. The wall represents an abso-
rial of the wall created a surreal-animate vision and lute body, in which an infection causes various chal-
an apopohenia kind of feeling for the viewers. As an lenges, and activates an immune system as well. The
educational goal, the attempt was not to create an struggle between infection and the immune system
eye catching media-wall but to introduce students creates lesions eventually (Figure 4). This concept
with digital toolsets necessary to make them think resembles infection as a distortion on the regularity
of feedbacks in-between the digital and the physical of the wall. Irregularities of the surrounding factors,

Figure 2
FiberGrid; Dataflow diagram
composed in Grasshopper.
Cluster 1 captures surrounding
sounds; Cluster 2 develops a
square grid out of curves; and
Cluster 3 generates force field
deformations to the curves,
based on the sound level.
Figure 3
FiberGrid; Application photos.
‘The Host Body’ is on the left.
such as the movements of people around causes This concept is realized by implementing a his-
pattern deformations. The host body gets infected tory based truncation process on a regular grid (Fig-
when someone gets closer to it, but eventually a ure 5). The truncation is associated with the vectors
time-based recovery process begins. of surrounding motions, captured by a webcam in
Figure 4
Lesion; Student sketches.
Figure 5
Lesion; Screenshots.

Figure 6
Lesion; Dataflow diagram
Cluster 1 creates a regular
grid to be infected; Cluster 2
captures the webcam input;
and Cluster 3 processes this
data according to the design
intentions, creating polygonal
shapes on the grid.
Figure 7
Lesion; Application photos.
real-time. There are parameters such as recovery and its surroundings. The real-time deformation input
immune system in the dataflow diagram (as seen in was a similar one with Lesion, including a webcam
Figure 6) that function as a temporal deformation capture. Distinctively, this project aims to capture
returning to its initial state progressively. In the final not all of the small details of the surrounding, but
installation, various regular grids (square and hexag- the average motion, searching for focal points of
onal) are tested with an infection caused by people movement. Students argued that this transforma-
around (Figure 7). tion of the glass brick wall to plasmatic body makes
it more interactive with other bodies around it.
Plasma In this project, students’ conception (Figure 8)
In this project, the host body is considered to be is extended into a geometric solution based on me-
infected by high fever and pressure, changing its taballs (Figure 9). After various experiments on the
solid phase into plasma. The solid molecules repre- reactions of metaballs, a grid-based deformation is
sent the strict order of the grid on the wall, while the chosen (Figures 10). When a person comes closer
plasma represents a more flexible order, sensitive to to the wall, its motion creates focal points. Eventu-
Figure 8
Plasma; Student sketches.

Figure 9
Plasma; Screenshots.
Figure 10
Plasma; Dataflow diagram
Cluster 1 collects all necessary
data including the webcam;
Cluster 2 calculates a vector
deformation on a regular grid;
and Cluster 3 creates a serie of
metaballs.
Figure 11
Plasma; Application photos.
ally these points become blob centers that react and isting patterns of the hall and transform that inert
combine into larger blobs (Figure 11). A time-based void to a reacting body.
algorithm captures sequences of this process, creat- Within this three-day workshop, we worked
ing superimposed metaball variations. with 30 students and introduced them with digital
techniques of pattern-making and pattern deforma-
WORKSHOP 2: RE-FLEX PATTERNING tion using Grasshopper. We discussed on how they
In the second Infections workshop, the host body could use parametric modelling to deform a grid
was the gallery hall of the faculty building at AİBU, a based pattern.
passive void waiting to be activated (Figure 12). The The composite material system proposed for
regular pattern dominating that body was the struc- the workshop was a combination of flexible and soft
tural grid of columns and beams that is reference to materials (textile or bubble wrap) with a stiffer but
all the details around it such as floor coverings, light- lightweight plate material (5mm. foam boards). Soft
ings etc. In this workshop students are encouraged material is to be covered with foam boards in both
to think on sub concepts of infection, recognize ex- sides with nuts and bolts to explore its composite

Figure 12
Re_Flex Patterning; ‘The Host
Body’.
material behavior. Students were required to pro- reflexes of the material and reactions of the host
pose a patterning that controls the behavior of this body. Each project was unique to explore various
composite material with the help of the re-flexing material behaviors using regular, irregular and as-
performances. sociative patterning (Figure 13). Students chose the
project that proposed a canopy formed by patterns
Prototypes of circulation .This project was able to control the
On the first day, we discussed on concepts of infec- macro-form as a self-regulating surface.
tion and the context. 5 groups of students present-
ed their proposals via diagrams and drawings. They Final
focused on changing parameters and dynamics The last step was working on patterning of the cho-
such as daylight, circulations, gatherings, vistas and sen project, and is developed with the guidance of
proposed concepts as molecules, fluid flows, coloriz- instructors. At the application phase, 1600 individu-
ing etc. We wanted them to construct their first ma- al polygonal elements are coded and laser-cut from
terial prototypes by 2 m X 2 m via various methods foam boards, attached to the textile with nuts and
of patterning. The next morning students installed bolts (Figure 14). The product of the workshop was
their physical prototypes to the hall to observe the two canopies of 1,5m. by 5m. in size. The emergent
Figure 13
Re_Flex Patterning; Initial
prototypes, testing the com-
posite material with various
tessellations.

Figure 14
Re_Flex Patterning; Third and
final day, fabrication.
performances of this product could only be experi- such as fabrication technologies, material studies,
enced when these surfaces were installed in the gal- and generative techniques. This requires not only an
lery hall via flexing them with the help of steel cables intuitive handling on digital tools and methods, but
(Figures 15, 16 and 17). Students were excited with a also an experience on material and production con-
feeling of both familiarity and alienage of this prod- straints simultaneously.
uct, mentioning that the passive void is becoming Patterning emphasizes a material shift in the
an-other living body. generative side of the digital paradigm, and a geo-
metric shift in the material side, as well. The study
CONCLUSION presented in this paper is an example of the inte-
Contemporary trend of the computational design gration between digital tools and material practices
education is grounded on an integration of domains by implementing pattern deformation as a synthe-
Figure 15
Re_Flex Patterning; Third and
final day, installation.
Figure 16
Re_Flex Patterning; Final
project.

Figure 17
Re_Flex Patterning; Final
project.
sizer. Such integration liberates students from pas- ment of Architecture for their collaboration and kind
sive and formal search of an on-screen parametric support. We would like to thank all workshop par-
modeling, familiarizing them to a more practical and ticipants for their invaluable efforts and feedbacks.
sophisticated body of knowledge about the physical More information about the workshops and full list
becoming itself. Nevertheless, the articulation and of students can be found in the blog [1].
reconstruction of patterns help pedagogical objec-
tives as they promote temporal but instant, explicit REFERENCES
but unstable nature of design exploration. Garcia, M 2009, ‘Prologue for a History, Theory and Future
of Patterns of Architecture and Spatial Design’, Archi-
ACKNOWLEDGEMENTS tectural Design: Patterns of Architecture, vol 79, no 6, M
Authors would like to thank to İstanbul Bilgi Univer- Garcia (ed), Wiley, London, pp.6-18.
sity Faculty of Architecture, bi’sürü student organi- Gleiniger, A. and Vrachliotis, G. (2009), Pattern: Ornament,
zation at Yıldız Technical University, Department of Structure and Behavior’, Birkhauser, Berlin, pp.7.
Architecture, and Bolu Abant İzzet Baysal Univer-
sity Faculty of Engineering and Architecture, Depart- [1] www.infections2.blogspot.com

Algorithmic Engineering in Public Space
Patterning strategies for a plaza paving

Jaroslav Hulin1, Jiri Pavlicek2
Czech Technical University in Prague, Faculty of Architecture
http:/www.echorost.com
1
hulinjar@fa.cvut.cz, 2jiri.pavlicek@echorost.com
Abstract. The paper reflects on a relationship between an algorithmic and a standard

(intuitive) approach to design of public space. A realized project of a plaza renovation in
Czech town Vsetin is described as a study case. The paper offers an overview of benefits
and drawbacks of the algorithmic approach in the described study case and it outlines
more general conclusions.
Keywords. Algorithm; public space; circle packing; optimization; pavement.
PUBLIC SPACE
Instead of the embodiment of a static order more French inventions. In both of these cases monumen-
and more a city is considered to be an ever chang- tal and vast public spaces allowed the gathering of
ing organism. Over few decades, architects have to the crowd and assembly of a national collective.
cope with new concepts of space imposed by global
markets, the Internet, ballooning population figures, Foam City
social isolation, and environmental crisis. Philoso- The current nature of the human environment is de-
pher Peter Sloterdijk argues in his article “Foam City” fined by the fact that nature and human action can
that architectural designs have been always integral no longer be separated. Technology and nature are
to establishing the society. considered to be all part of a network; a whole that
cannot be managed by simple urban planning strat-
City vs. Society egies.
The article focuses on the Fête de la Fédération of Sloterdijk describes the city as a Foam City: ‘The
July 14, 1790, celebrated on the first anniversary of co-isolated foam of a society conditioned to individu-
the storming of the Bastille. The author argues that alism is not simply an agglomeration of neighboring
the architectural staging of this spectacle served to (partition-sharing) inert and massive bodies but rather
generate an embodiment of the nation, enhanced multiplicities of loosely touching cells of life-worlds’
by affective and acoustic measures. While the article (Sloterdijk 2006).
is mainly concerned with the architectural technolo- In other words the idea of the collective society
gies of politics related to the French Revolution, it has disappeared and was replaced by the society
also points beyond this specific historical case and that resembles the foam, where the individuals are
briefly indicates how 20th-century fascisms used clustered in co-isolated groups. In these co-isolated
techniques that were prefigured by 18th-century groups individuals share their interests and opin-

Figure 1
The original setting out of
the circles within the plaza
boundary.
ions. Therefore the city and public space can no and the entire space was used mainly as a commu-
longer be designed for a massive collective but rath- nication corridor. By developing a right patterning
er for an ever changing multiplicity. strategy we aimed to invite as many social groups to
interact in the plaza.
PROJECT GOALS The second step was to develop the algorithm
Several years ago, we were approached by Moba ar- that would guarantee that all technical require-
chitectural studio to collaborate on a refurbishment ments are met. There were seven sizes of the con-
of a medium sized town in our country. The original crete circles with diameter ranging from 1.2 to 4 me-
design was based on Sloterdijk’s Foam City meta- ters. No two circles could intersect with each other
phor. We were commissioned to develop a method and also with multiple other objects/obstacles.
(algorithm) that would set out a layout of hundreds Furthermore the minimum continual asphalt area
of concrete circles in the surface of the refurbished among the circles could not be less than 0.5 m2due
plaza while reflecting Sloterdijk’s observation of the to given construction limitations.
society in his article “Foam City”.
The assignment consisted of two rather inde- METHODS
pendent steps. The first step was the functional anal-
ysis of the square and development of a patterning Input, brief
strategy that would initiate various activities to hap- Early in the process we realized there is a strong re-
pen in the public space. There was hardly any social lationship between any algorithmic method and the
interaction happening in the square beforehand designed pattern. The original design (an outcome

Figure 2
Technical requirements for
optimization.
of an architectural competition) was based on a ran- 3. If the concave residual space defined by any
dom distribution of circles with the highest density three circles was smaller than 0,5 m2 (gap
in the centre of the plaza (Figure 1). The paving pat- smaller than 50 millimeters was considered as
tern was created intuitively by the architects. Our closed) all three circles were moved apart.
task was to optimize the design with as little inter- The process usually took about one hundred itera-
vention as possible. tions to optimize the circles. Visual aids to mark any
possible collisions were scripted in to help to re-
Technical requirements move possible dead end suboptimal results.
We used a simple circle packing algorithm based on
collision detection and an iterative approach. The al- Functional analysis of the square
gorithm ran through the randomly generated circles Having met the technical part of the brief early in
and in every round checked for several conditions the work process, we started to question the func-
derived from the brief (Figure 2). tional quality of random distribution. The architects
1. If any of the circles collided with the boundary did not have any means of designing layout of al-
of solved space it was moved away in the op- most five hundreds circles other than random dis-
posite direction. tribution with intuitive gradient density. Our moti-
2. If any two circles were closer than 6 millimeters vation was to propose a better way of working with
to each other both of them were moved apart such a high amount of design elements and still be
in the opposite direction. coherent with the original design brief.

Table 1
Blacktop continuity
Tiles - medium size
Pattern x Purpose
Relation of the public space to
Tiles - small size
Tiles - large size

the pavement quality.
Tiles density
Communication X X
Public space X X X X
Commerce X X
Auditorium X X X X
Restaurant X X X
Relax X X X
We analyzed the public space and defined sev- become quiet rest areas were defined by a high den-
eral qualities of the pavement that we wanted to sity of the whole range circles, spots to slow down
address in a new generator (Table 1). For example, on (such as entranced to public buildings) were de-
spots with more and faster traffic would be defined fined by a high density of smaller circles.
by bigger and less dense circles (the asphalt is easier With such an approach we were able to com-
to walk and cycle on), spots that were supposed to pose a colored gradient map that served as a layout
Figure 3
Gradient map of different
qualities of the public space.

Figure 4
Realized first stage of the
project.
generator (Figure 3). However, this algorithmic driv- described by Carpo (2011)). It is disputable whether
en design method was perceived by the architects there was any control (other than intuitive visual) at
as something uneasy to control and was not deve- the first place.
loped further. In a general, yet similar, case (urban paving pat-
tern), early design stage algorithmic tools capable
CONCLUSION of gathering and manipulating vast range of design
Without an algorithmic approach it would not be information would help the team do devise a better
possible to handle a project of this size within the and more functional design. In that case, an “infor-
short amount of time given to the project. The al- mation engineer” should play a substantial role in
gorithmic approach not only helped to optimize the the design team similar probably to a role construc-
setting out of the circles but was essential for pro- tion, civil or technology engineers play during a
duction of final project documentation and for lay- standard building design process.
ing out the concrete circles on site (Figure 4).
The failure of the approach was inability to REFERENCES
change the rather simple definition of random dis- Sloterdijk, P 2006, ‘Foam City’, Distinktion: Scandinavian
tribution and gradient density of the circles. The Journal of Social Theory, 9(1), pp. 47–59.
designers were not comfortable with passing any Carpo, M 2011, ‘The Alphabet and the Algorithm’, pp. 126
control to an algorithm and with dual authorship (as

Integrating Computational and Building Performance
Simulation Techniques for Optimized Facade Designs
Mahmoud Gadelhak
Ain Shams University, Egypt
m.islam@eng.asu.edu.eg
Abstract. This paper investigates the integration of Building Performance Simulation

(BPS) and optimization tools to provide high performance solutions. An office room in
Cairo, Egypt was chosen as a base testing case, where a Genetic Algorithm (GA) was
used for optimizing the annual daylighting performance of two parametrically modeled
daylighting systems. In the first case, a combination of a redirecting system (light shelf)
and shading system (solar screen) was studied. While in the second, a free-form “gills
surface” was also optimized to provide acceptable daylighting performance. Results
highlight the promising future of using computational techniques along with simulation
tools, and provide a methodology for integrating optimization and performance
simulation techniques at early design stages.
Keywords. High performance facade; daylighting simulation; optimization; form
finding; genetic algorithm.
INTRODUCTION
The building’s skin plays the main role in deliver- mal solution without the need of testing all possible
ing natural daylight to indoor spaces. Performative ones. This paper investigates the ability of integrat-
façade design can significantly improve the indoors ing computational and simulation tools for design
visual and thermal conditions, which in turn, im- problems with different levels of complexity. The
proves the quality of life and work environment by methodology proposed in this research employed
creating productive and appropriately lit spaces. a simple Genetic Algorithm for optimizing the day-
Building skins, therefore, shouldn’t be just designed lighting performance of parametrically modeled of-
for its aesthetic aspects but also as a functioning ele- fice building facades.
ment in the building.
Building Performance Simulation (BPS) tools are Genetic Algorithm and Daylighting Perfor-
broadly used for achieving designs that have bet- mance
ter impact on the users and the environment. While A traditional optimization scheme is an algorithm
simulation tools are effective in testing and evalu- which finds the minima or the maxima of a given
ating different designs, it becomes harder when function, typically known as the objective function.
evaluating numerous solutions. Simulation engines The objective function may depend on any number
usually take a considerable amount of time for each of parameters and any combination of parameter
solution. Therefore, it is more practical to consider values within the defined search space is considered
using optimization tools that can arrive to an opti- a feasible solution. The optimal solution will be the

feasible set of parameters which minimizes (or maxi- that are predicted to offer good results are being op-
mizes) the objective function. A problem will not timized. The aim in this case is to find a better solu-
necessarily have one unique solution. It may have tion from numerous good solutions, in other words
no optimal solutions at all, a finite number of solu- an optimal solution. Conversely, the second case in-
tions, or an infinite number of solutions, which can vestigates an exploratory search that has no guides
be defined as a more specific subset of the search to start with. A free form façade design was opti-
space (Papalambros and Wild, 2000). For problems mized for better daylighting performance. A larger
which involve simulation engines heuristic search number of choices and parameters were also intro-
algorithms are usually used and considered an ap- duced in order to investigate the methodology’s
propriate choice. These algorithms are considered ability in problems with a larger pool of solutions.
heuristic as they depend on trial-and-error approach The aim of the second case is not mainly reaching an
and as such, they are not guaranteed to converge to optimal design but instead to discover the potential
true optimal solutions. However, most of these algo- of using the proposed methodology in more com-
rithms do find solutions which are very close to opti- plex contexts.
mal (Gagne and Anderson, 2010).
The Genetic Algorithm (GA) is an algorithm METHODOLOGY
which works by mimicking the process of natural A typical side-lit office room space was selected for
evolution and was first introduced by Holland and investigation. The case study was chosen to be locat-
Reitman (1977). It is one of the most commonly used ed in the city of Cairo, Egypt (30° N- 31° E). The office
heuristic search techniques, and has been applied is room is a 4.00m wide by 6.00m deep rectangular
to many types of architectural problems. Genetic space, with 3.00 m clear height. The office space was
algorithms had been successfully used in several assumed to have a 6 mm double glazed window that
research works for enhancing daylighting perfor- is 3.60m wide and 1.80m high (Figure 1 and Table
mance. Tsangrassoulis et al. (2003) presents a tech- 1). The space was considered to be on ground level
nique for the design of slat-type blinds based on the with a free horizon and no obstructions. Ground
relative light intensity distribution under a uniform reflectance of 20% was assumed. At first, the base
light source. The technique used a genetic algorithm case was modeled and its annual daylighting perfor-
in order to evolve the design according to a set of mance was observed. Afterwards, Genetic Algorithm
parameters. Another research investigated altering (GA) was used for optimizing the annual daylighting
free-form ceiling geometry design to optimize in- performance of two parametrically modeled cases.
door daylight uniformity ratios (Rakha and Nassar, In the first case, a combination of a redirecting sys-
2011). Other research works focused on optimizing tem (light shelf ) and shading system (solar screen)
the facade design and openings to achieve better was added to the base case. While in the second, a
daylighting levels and comfort (Torres and Sakamo- free-form “gills surface” façade was also optimized to
to, 2007; Gagne and Andersen, 2010; Portugal and provide acceptable daylighting performance.
Guedes 2012).
Although several investigations had been pre- Daylighting Simulation Methodology
viously carried out on using genetic algorithms for Simulation was conducted using the Diva-for-Rhino
performance optimization, most of the previous re- V 2.0, a plug-in for Rhinoceros modeling software. It
searches were made on simplified problems, such was used to interface Radiance and Daysim for an-
as window positions or shading device parameters. nual simulation and illuminance computation (Re-
This paper aims to investigate the proposed meth- inhart et al., 2011). Simulation was conducted an-
odology under two conditions. A simplified guided- nually for weekdays from 8:00 AM to 4:00 PM which
search case, where several cases and parameters represents a typical Egyptian eight-hour working

Figure 1 Space Dimensions and Materials
Isometric view of the studied Floor level Zero level
office room. Space dimensions (m) 4.00 * 6.00 * 3.00
Walls Reflectance 50%
Table 1 Material Medium Colored
Dimensions and properties of Ceiling Reflectance 80%
the tested office space. Material White Colored
Floor Reflectance 20%
Material Wooden Floor
Window Dimensions and Material

time. The reference plane on which daylight per- Width (m) 3.60
formance was simulated contained 117 measuring Sill (m) 0.90
points in a grid of 0.5m* 0.5m, at a working plane of Lintel (m) 2.70
height 0.8 m. Measurements that were found equal Glazing Double clear glass
or higher than the recommended minimum illumi- 6 mm (VT = 0.647)
nance value for an office space, 500 Lx, were consid-
ered “adequate” (IESNA, 2000). Daylight Availability a process, a set of initial solutions (a generation/popu-
Dynamic Daylight Performance Metric (DDPM) was lation) is generated at random. Each solution is sent
used for evaluation. It presents three evaluation cri- to DIVA for conducting Daylight Availability simula-
teria: “daylit” areas (the adequate areas), for spaces tions. The results are then processed to calculate the
that receive at least half the time sufficient daylight daylit area percentage using a fitness function that
compared to an outside point, “partially daylit” aris expressed as:
eas, which are below useful illuminance and “over lit” F(x) = N‘ / N (1)
areas that provide warning when an oversupply of where N is the total number of measuring nodes,
daylight (10 times target illuminance) is reached for and N‘ is the number of nodes which receive at least
at least 5% of the working year. Analysis criteria for half the time sufficient daylight.
Daylight Availability adopted in this paper assumed Solutions that result in good performance are
that the designs that achieved equal or more than used as “parents” for a new generation. Parent mem-
50% “daylit” areas and at the same time minimum bers are combined using a genetic operator called
values for “over lit” and “partially daylit” areas were crossover to create a new generation of “child” mem-
considered efficient. bers which have characteristics of the parents. Since
this new generation is based on the best performing
Parametric Modeling and Optimization solutions in the previous solutions, it is assumed that
Methodology some members of the new generation will perform
Generative designs and parametric models were better. Once evaluated, again the good performers
modeled using Grasshopper a plug-in for Rhinoc- are used as parents while the poor performers are
eros modeling software. Grasshopper allows chang- discarded. The cycle continues until a number of
ing the model parametrically where each design generations have been completed. In this research
parameter is directly linked to a floating-point slider a simple GA with 20 genome/generation was used.
that determines its value. Galapagos, an evolution- Optimization continues until an optimal solution is
ary solver, is a GA optimization tool used within the obtained and remained unsurpassed for continuing
platform of grasshopper and was used to control twenty generations (Stagnant Generations) (Figure
and modify the different parameters. During the GA 2).

Figure 2
Optimization methodology
diagram.
BASE CASE SIMULATION RESULTS nating the overlit areas. However, in many cases that
Daylight availability was analyzed for the base case came with a drawback in the overall performance
facing the South and the East orientations. Both due to the increase in partially daylit areas. Combin-
South and East facing spaces were subject to the ing the solar screen with light shelves was found to
penetration of the direct sun. Overlit areas reached achieve better results. In this case combining solar
43% in the South and 42% in the East. However, no screens and light shelves was examined. The design
partially daylit areas were anticipated in South ori- parameters of both systems change according to
ented office space where daylight area reached 53%. the results obtained from previous research works
In the East faced room, 13% of the space were found (Sabry et al., 2012). The parametrically modeled cas-
to be partially daylit (Table 2). The main challenge, es had six different changeable parameters shown
therefore, is to eliminate the overlit areas without a in Table 3. Overall, the number of resulting possible
significant increase in partially daylit areas. designs exceeds two thousand possible solutions.
FIRST CASE STUDY: LIGHT SHELF AND Daylighting performance optimization

SOLAR SCREEN COMBINATION results
In previous research work by Sherif et al. (2012), so- An optimal design was obtained and the perfor-
lar screens were found to be highly effective in elimi- mance reached 64% in the sixth generation, where

Table 2 South East
Daylight availability distribu-
tion for the base case in South -45 -59 -67 -70 -70 -71 -67 -61 -47 -18 -31 -36 -38 -40 -39 -35 -39 -37
and East orientations . -38 -45 -54 -60 -60 -59 -56 -47 -38 -11 -22 -27 -31 -34 -36 -33 -32 -30
-25 -28 -36 -39 -38 -40 -37 -30 -27 -9 -14 -17 -21 -23 -24 -22 -25 -18
-15 -19 -19 -23 -22 -23 -21 -20 -17 98 -7 -9 -11 -16 -17 -18 -17 -15
-13 -13 -16 -15 -18 -15 -17 -15 -14 96 -5 -7 -5 -9 -13 -14 -15 -9
99 99 100 99 -7 99 99 -7 -7 94 95 96 -6 -6 -7 95 -7 -9
98 98 99 99 99 99 99 99 99 86 90 91 94 95 95 95 95 -5
98 98 98 98 98 98 98 98 97 72 76 83 82 88 88 88 78 77
96 97 97 97 97 97 97 96 95 65 68 69 71 75 72 73 69 65
94 95 95 96 96 96 95 94 92 51 57 57 55 57 58 57 59 54
81 84 89 93 90 90 89 86 79 39 44 51 53 52 53 50 50 43
71 78 81 81 82 82 79 74 71 33 36 40 40 45 45 43 41 40
59 66 70 77 77 69 73 71 65 31 34 34 36 39 40 41 37 38
a 1:1 (H: V) screen with 90% perforation and 50° VSA matched. Table 4 illustrates the best results obtained
was combined with a 120 cm, 10° rotated light shelf. from the optimization process.
It remained the best solution for the next twenty
generations. However, several designs also went SECOND CASE STUDY: FORM FINDING
far beyond the performance of the base case (45% Form-finding can be described as a process of dis-
daylit area). Moreover, all the proposed solutions covery and editing (form emerges from analysis). Ex-
had minimal overlit area percentages which didn’t treme form-finding is not fully architecture but more
exceed 7% of the whole space area and several applied engineering as form exclusively determined
cases succeeded in entirely eliminating the overlit by function. In this case study a free form daylight-
area. Figure 3 shows the simulation results after the ing system was proposed. Similar to the previous
26 generations. It’s noticed that the optimal solu- studied case, this system combines a redirecting
tion was reached at an early stage within the first six and shading techniques, however the form is more
generations. The process afterwards wasn’t success- organic and fixable. A “gills surface” was modified to
ful as the coupling with new solutions resulted in be used as a shading devise in the lower part of the
worse solutions while the optimal solutions kept un- window and as a light shelf in the upper part. Gills
Table 3 Solar Screens

Parameters considered for the Parameter Possible values
first case. Vertical Shading Angle (VSA) 70°, 60°, 50°, and 40°
Perforation 90%, 80% and 70%
Aspect ratio (Horizontal: Vertical) 1:1, 2:1, and 4:1
Light Shelves
Parameter Possible values
External light shelf depth 60 cm, 80 cm, 100 cm, 120 cm
Internal light shelf depth 30 cm, 60 cm, 80 cm
External light shelf rotation angle 0°, 10°, 20°, and 30°

Figure 3
First case study after 26
generations, fittest design had
64% daylit area.
81
83
93
92
92
-7
89
95
95
93
86
92
91
95
94
94
85
81
93
92
95
95
95
96
96
95
96
96
95
95
96
97
95
95
91
93
Table 4
82 90 91 95 94 93 91 88 77 84 88 94 96 93 92 94 94 82 Cases with highest perfor-
75 79 83 92 86 86 80 85 78 83 80 93 93 94 93 88 93 83
65 75 75 79 77 78 79 73 71 71 80 87 89 -7 -7 -7 -7 76 mance for the first case study.

60 66 66 75 75 73 66 71 60 72 78 81 84 83 88 82 77 75
It’s noticeable the diversity in
57 64 64 66 67 67 67 62 59 60 68 68 76 69 72 68 75 63
44 56 57 61 61 64 56 54 52 52 57 60 59 63 65 57 55 52 the solutions and the relatively

37 42 52 50 54 51 48 54 43 46 48 58 58 53 54 53 49 45
32 42 40 42 40 45 46 41 30 29 42 38 41 47 44 40 48 33
small overlit area percentages.
24 24 26 34 34 27 33 34 34 20 30 29 27 34 35 37 34 35
19 22 32 26 26 32 28 29 28 17 18 26 31 32 29 21 28 25
14 20 26 31 27 31 21 19 20 17 17 14 19 24 27 26 25 21
Screen configurations: 90%, 1:1, 40° Screen configurations: 90%, 2:1, 50°
Light shelf configurations: Ext. 100 cm. 10°, 60 cm Light shelf configurations: Ext. 100 cm. 0°, 60 cm
Int. Int.
Daylit 64% Daylit 62%
Overlit 0% Overlit 3%
Partially Daylit 36% Partially Daylit 35%
94 97 97 97 97 -5 96 96 92 85 94 93 96 94 93 94 91 88
93 97 97 -5 97 -5 97 -5 95 85 95 -10 -11 -10 -11 -10 -5 -10
92 92 92 97 95 95 96 93 89 82 87 91 95 96 94 94 93 86
86 90 94 93 95 92 94 85 86 78 86 88 92 91 93 93 77 77
84 83 90 92 -7 -7 -6 -6 86 62 70 75 77 84 89 80 83 69
76 80 83 83 82 82 82 75 74 62 74 74 72 77 74 75 70 64
63 71 68 78 75 74 74 75 64 65 67 59 72 70 60 71 64 57
56 60 63 68 69 66 60 63 59 46 56 57 61 59 60 54 59 55
41 55 56 57 58 57 56 53 51 46 51 55 58 57 56 56 59 44
32 40 39 42 46 50 48 48 38 38 44 42 42 47 43 48 42 43
29 27 31 32 32 36 41 34 35 28 24 35 38 36 39 38 42 31
19 17 27 30 33 32 32 31 28 20 27 31 34 29 38 33 28 23
8 21 20 24 25 30 28 24 22 17 23 26 26 28 28 27 20 19
Screen configurations: 90%, 4:1, 40° Screen configurations: 80%, 1:1, 40°
Light shelf configurations: Ext. 100 cm. 0°, No Int. Light shelf configurations: Ext. 120 cm. 20°, 60 cm
Int.

Figure 4
ifferent shapes and settings
for the façade of the second
case study.
surface is a free form inspired from nature and has designs are obtainable. Such a huge pool of design
been recently used in several architecture works. choices highlights the necessity of using tools such
The proposed system was applied to the South as genetic algorithms for finding designs that can
facade and was parametrically controlled to provide provide suitable performance (exploratory analysis).
a wide range of options. Every louver had a median Figure 4 shows different shapes and settings for the
control point which represents the curve peak. This façade.
point has the ability to move in the vertical and hori-
zontal direction to control the openness and close- Daylighting performance optimization
ness of that part as well as the amount of shading results
it provides. The transition of the rest curve points is, The algorithm succeeded in providing several ac-
however, not unique; Instead it follows a Gaussian ceptable cases considering the fact that the designs
curve were transition is defined by a symmetrical were found to have a wide range of performance
sequence of values, with null extremes. Similarly, (oscillated from 56% high to as low as only 5%). It
the curved light shelf in the upper part has a simi- might be useful to use such a tool to limit the op-
lar point that controls its extension and curvature. tions in the beginning of the design phase. An op-
Sixteen positions are optional for each single part timal solution was obtained in the second genera-
of the system and more than two millions different tion and continues to be the fittest for the remaining

Figure 5
Second case study after 21
generations, fittest design had
56% daylit area.
twenty generations. The optimal solution had 56% the first case, The GA reached a near optimum solu-
daylit area. Results show that the algorithm wasn’t tion and succeeded in reaching solutions that have
able to conduct a real optimization, most likely be- a significantly better performance compared to the
cause of the extremely wide solution space and the base case. It’s, therefore, recommended to use opti-
simple characteristic of the algorithm. Larger gen- mization tools and evolutionary solving methods in
erations and more computing time (more genera- guided searches for optimal solution from various
tions) would have possibly reached better results. possible options.
However, it’s hard to judge the success of the algo- In the second case, and because of the vast
rithm without further optimizations. Figure 5 shows number of solutions, the algorithm seemed to set-
the optimization progress and Table 5 shows the tle with a local optimal. Although, it may be hard
best performing cases. to judge the results unless more optimization trials
with different setting are made, the algorithm was
CONCLUSION found to be a suitable exploratory method to limit
The two studied cases demonstrated the ability of the options when no previous experience is avail-
the Genetic Algorithm in producing designs with able. This is an exceptionally useful feature that ena-
acceptable daylight performance. However, the bles the integration of performance analysis in ear-
performance of the optimization tool was found to lier stages of design.
differ based on the complexity of the problem. In The proposed methodology can be adjusted to
68 73 76 81 83 80 76 72 66 69 77 74 82 78 82 76 74 65
Table 5
74 78 78 78 79 79 80 79 73 70 74 77 81 80 80 79 75 69
71 77 77 77 81 76 80 74 67 68 76 74 77 76 77 75 70 65 Cases with highest perfor-

73 69 76 75 77 74 74 68 69 72 74 72 74 74 76 72 71 72
64 70 76 -7 -7 -7 73 69 67 66 70 75 -7 -7 -7 71 70 67
mance for the second case
63 68 65 76 71 68 65 61 54 56 61 66 74 67 63 68 67 60
study.
48 55 59 58 57 58 57 54 43 47 52 61 60 58 56 54 62 46
24 41 47 44 49 48 49 51 32 25 41 49 44 48 43 39 51 56
3 16 22 23 21 39 15 24 13 14 10 14 26 39 32 23 17 11
10 13 5 8 9 10 12 4 1 3 2 6 5 9 14 9 8 7
7 1 2 2 9 4 0 0 0 0 0 0 0 3 3 0 0 0
0 0 0 0 2 0 0 0 0 0 0 0 0 1 1 0 0 0
0 0 0 0 5 0 0 0 0 0 0 0 0 1 0 0 0 0


diverse contexts and objectives. Form generation Algorithm’, PLEA 2012, Lima, Peru.
and form finding using evolutionary solvers can Rakha, T. and Nassar K. 2011, ‘Genetic algorithms for ceiling
open the door for more performative designs with form optimization in response to daylight levels’, Re-
unlimited creativity and minimum restrains. The pro- newable Energy, 36, pp. 2348–2356
posed methodology can also aid architects in taking Reinhart, C.F., Lagios, K., Niemasz, J., and Jakubiec, A. 2011.
design decisions in the early design stages. DIVA for Rhino Version 2.0. http://www.diva-for-rhino.
com/.
REFERENCES Sabry, H., Sherif, A. H., and Gadelhak, M. 2012, ‘Utilization of
Gagne J. and Andersen M. 2010, ‘Multi-Objective Facade Combined Daylighting Techniques for Enhancement
Optimization for Daylighting Design Using a Genetic of Natural Lighting Distribution in Clear-Sky Residen-
Algorithm’, Fourth National Conference of IBPSA-USA, tial Desert Buildings’, PLEA 2012, Lima, Peru.
New York. Sherif, A., Sabry, H., Gadelhak, M. 2012, ‘The impact of
Holland, J. and Reitman, J. 1977, ‘Cognitive systems based changing solar screen rotation angle and its opening
on adaptive algorithms’, ACM SIGART Bulletin, 63, pp. aspect ratios on Daylight Availability in residential de-
49-49 sert buildings’, Solar Energy, 86(11), pp 3353-3363
IESNA 2000, ‘IESNA Lighting Handbook’ 9th Edition, Illumi- Torres S., and Sakamoto Y. 2007, ‘Facade Design Optimiza-
nating Engineering Society of North America. tion for Daylight With a Simple Genetic Algorithm’,
Papalambros P. and Wild D. 2000, Principles of Optimal De- Building Simulation 2007, Beijing, China.
sign: Modeling and Computation, 2nd ed, Cambridge Tsangrassoulis, A., Bourdakis, V., Geros, V. and Santamouris,
University. M. 2003, ‘A genetic algorithm solution to the design of
Portugal, V., and Guedes, M. 2012, ‘Multi-Objective Facade slat-type shading system’, Renewable Energy, 31, pp.
Optimization For Daylighting Design Using a Genetic 2321–2328

Models of Computation:
Form Studies
Models of Computation: Form Studies - Volume 2 - Computation and Performance - eCAADe 31 | 271
272 | eCAADe 31 - Computation and Performance - Volume 2 - Models of Computation: Form Studies
Algorithmic Form Generation for Crochet Technique
A study for decoding crocheted surface behaviour to explore variations

J. Gozde Kucukoglu1, Birgul Colakoglu2
Yildiz Technical University, Turkey
1
http://www.automaticlogic.net, 2http://www.bot.yildiz.edu.tr/
1
jgkucukoglu@hotmail.com, 2colak@yildiz.edu.tr
Abstract. In architecture use of generative computation suggests a possibility of

rethinking the form finding process. In order to generate form, one method could be
predefining first the production technique and constraining the form by the rules of it.
In this study crochet-knitting technique is chosen as a production technique. To explore
various forms developed through this technique; a computational model, which the
behavior of crochet-knitted surface is embedded into, is developed. This paper explains
the process of decoding the behavior of a crochet-knitted surface for a computational
model in order to extract the crochet-knitting patterns of complex geometrical surfaces.
Keywords. Form generation; crochet technique; hyperbolic geometry; decoding rule
INTRODUCTION
This research aims to understand the behavior of technique. Once a form is developed through this
a crochet-knitted surface and decode its rule for computational model, the rule that is extracted from
a computational model so that it can be utilized this computational model is generic and also used
in architectural design process. Use of generative for physically knitting of it (Figure 1). This paper il-
computation suggests a possibility of rethinking the lustrates the first stage, which is concerned with the
architectural design process. Such rethinking could development of crochet-knitting computational
lead to slightly different comprehension of the de- model. Further research of this study will follow au-
signers’ decision making. In this study, the produc- tomatizing the physical production.
tion technique is predefined (crochet-knitting) and
the form generation is constrained by the rules of ALGORITHMIC THINKING IN KNITTING
production technique. Through algorithmic pro- Algorithm is a precise specification of a sequence
cess of crochet-knitting technique, various surfaces of instructions to be carried out in order to solve a
-which imply spatial and structural features- can be given problem (Rajaraman, 2003). Each instruction
generated and automatized for physical produc- tells what task is to be performed. Algorithm serves
tion through a computational model. This compu- as a codification of the problem through a series of
tational model, which contains the knowledge of finite, consistent, and rational steps. Although the
knitting surface behavior, facilities exploration of sequence of an algorithm is simple, the outcome
various forms developed through crochet-knitting could still be very complex and unpredictable. Fol-
Figure 1
Process of the research aim
lowing some specific rules -or a sequence of instruc- controlled by computers, but the mechanism of the
tions to do a job- is used in many fields. Knitting is knitting machines is not developed. If the mechan-
one of the main algorithmic procedures utilized by ics of these knitting machines can be modified for
humans till their early existence. The example below certain behaviours, it would be possible to knit as a
highlights the relation between knitting instructions whole, those complex forms that explored through
and algorithmic thinking. the computational model.
Example: Instruction to knit a sweater
Step1: Cast on 133 stiches KNITTING TECHNIQUE
Step2: Repeat steps 3 and 4, 11 times Knitting is a technique where one continuous line/
Step3: Knit 2, *Purl 1, Knit 1, Repeat from * to last thread composes not only Euclidian but also non-
stitch, Knit 1 Euclidian surfaces with a very simple operation. As
Step4: Knit 1, *Purl 1, Knit 1, Repeat from * to stitches are added, they push and pull on each other
End…Similar steps (Rajaraman, 2003) and create an emergent surface. The size of those
By proper permutation and combination of this stitches, and the number of their neighbours in the
elementary set of actions (knitting, purling, casting rows above and below, determines the shape of the
stitches on or of needles), an infinite number of knit- work [2]. For knitting a desired surface, the mathe-
ting patterns can be created. The algorithm, which matical concept of the surface needs to be convert-
is the rules for generating knitting patterns, can be ed into a pattern. This conversion requires a greater
also seen as a translator between human mind and understanding of the behaviour of the knitted sur-
computer. The power of computation, which in- face because one has to figure out where exactly to
volves vast quantities of calculations and recursions, increase and decrease stitches so that the resulting
can detect abilities that may have not ever occurred surface as a whole is as close as possible to the de-
to the human mind. The computational model, sired surface. Each work can be quantified as its own
which is based on knitting algorithm, expands the pattern. Once knitting pattern is derived than it is
limits of human imagination. generic and the same surface is created aside from
The algorithm instruction given above example the tension of the working yarn. Crochet as one of
illustrates that knitting technique can be automated the knitting techniques is chosen for handmade ex-
due to its algorithmic process. Knitting machine, periments, since it is easier to compose complex sur-
which is invented by William Lee in 1589, uses al- face. However the surface that is composed through
most the same principle with hand knitting [1]. It knitting technique is looser than the crocheted sur-
knits patterns with algorithmically defined needle face, both techniques essentially create the same
movement. Today, the needle movements can be geometry.
Figure 2
Hyperbolic crochet variations
of Daina Taimina and its
crochet patterns.
CROCHET GENERATES HYPERBOLIC DESCRIBING THE CROCHET SURFACE

GEOMETRIES BEHAVIOUR
Hyperbolic geometry is a surface that has infinite Every single decision of increasing or decreasing
number of lines that go through a point that is spec- stitches affects the shape of a whole crocheted sur-
ified on the surface but never meet the line that is face. While making decisions for each stitch, it is al-
described on the surface previously [3]. When hy- most impossible to predict how the entire surface
perbolic geometry was first discovered, mathemati- would be affected from this decision. It is because
cians did not understand how it looks like. In 1997, crochet a desired surface is mostly an intuitional
Mathematician Daina Taimina did explain how hy- act and it is not so easy to describe what makes a
perbolic geometries look like by using her crocheted linear thread to create a 3D surface. This makes the
and knitted surfaces. She found out that just by re- behaviour of the crocheted surface emergent. An
peating a very simple operation, it is possible to knit emergent behaviour or emergent property can ap-
variations of these complex mathematical geom- pear when a number of simple entities operate in a
etries (Figure 2). system to form more complex behaviours together.
Daina Taimina has crocheted variations of hy- The reason of emergent behaviour is usually the
perbolic geometries by starting from a row of fixed relation across different scales and there is often a
stitch number and then adding rows. The principle top-down feedback in entire system. Such emergent
of generating hyperbolic surface is adding one extra behaviour cannot be modelled through a standard
stitch in every n stitches (David and Taimina 2001). modelling software, since it should be embedded
The number of stitches increases per row and this into the modelling tool. This requires the necessity
arises negative local curvature. If n is smaller, more to understand the behaviour of the stitches on the
crochet stitches are added so that the concluded knitting surface. In order to find out the geometric
surface has a larger negative curvature. This curva- logic of this emergent behaviour, another technique
ture is constant if the process is repeated the same of constructing hyperbolic geometry is studied. This
everywhere (Osinga and Krauskopf 2004). This fact technique is called polyhedral model. Polyhedral
demonstrates that the crochet technique promises model consists of equilateral triangles and creates a
to create complex surfaces even with a very simple, hyperbolic surface if each vertex on the surface be-
repeated operation. longs to seven equilateral triangles (David, Taimina
2001). Since these two techniques can compose the Figure 3
same hyperbolic surface, is there any connection be- Crocheted, polyhedral models
tween them? consist of triangles and
As a result of exploring these two techniques, in polygons.
both of them there is the same physical attraction
that forces the whole surface into a hyperbolic form.
In other words, the polyhedral model behaves al-
most the same as the crocheted surface. Each stitch
in the crocheted surface behaves like one equilateral
triangle in the polyhedral model. While in crocheted
surface, each stitch pushes and pulls on each other COMPUTER MODEL OF GENERATING
and whole system creates an emergent 3D surface, SURFACE
in polyhedral model each edge of the equilateral The computational model (Figure 5) that simulates
triangles tries to stabilize itself in the same distance. crochet technique is created through the polyhe-
This knowledge is the key argument that enables to dral logic given above and is coded in processing
perceive and geometrically describe the behaviour programming language. Each pentagon, hexagon
of the crocheted surface. and heptagon is added one by one and attached
In order to establish the relationship between to each other with at least two vertices of the previ-
crocheted surface and polyhedral model, the cro- ous polygon. The code defines vertices as well as the
chet patterns that are extracted from polyhedral centre point of each polygon as a node and forces
models are tested. In this research: instead of equi- each node in order to be in the same distance with
lateral triangles; pentagons, hexagons and hepta- its neighbours. The form of the surface is governed
gons are chosen for computational and physical by the position of these nodes, which provide an
model generation. Because using these polygons easy process for calculating the overall geometry. In
generates similar smoothness that the crocheted processing code, each node as a particle is connect-
surfaces have (Figure 3). ed with its neighbouring nodes through springs.
As shown in Figure 4, hexagonal pattern gener- Therefore, the position of each node needs to be
ates a flat surface. If pentagons and heptagons are calculated until an equilibrium state was reached for
also inserted, the system deforms itself into 3D sur- the entire model while any node is added. During
face. modelling, each node determines the emergent be-
Figure 4
Extracting crochet rules from
polyhedral model.
Figure 5 EXTRACTING THE KNITTING PATTERN
Print screen of the code To generate the prototype, the code, which is writ-
interface written in processing ten in the processing programming language, is used
programming language. to create computational polyhedral model of the
desired surface. The creation of the surface starts
with the first polygon definition then the user con-
trols the surface generation by deciding the position
and the number of the edges of the next polygon.
Once computational model is generated then it is
printed as a flattened surface to build its physical
paper model. The pattern of crochet prototype (the
number and the order of the stitches for each row)
is extracted by counting the number of triangles in
each vertex on the paper model. This pattern that is
haviour by affecting on the overall shape. The pro- the output of the computational model is used for
cess is iterative and it has a different approach than crocheting the replica of desired (Figure 6).
a standard computational modelling. It does not
start with a pre-defined geometric surface, besides REALIZATION OF A FULL SCALE PROTO-
the generated geometric surface is unpredictable. TYPE
The process of form generation is nonlinear and it To scale the production of crocheted surface, the
enables negotiation between several nodes simulta- polyhedral model is also used. In polyhedral model
neously. This negotiation between nodes generates if the number of equilateral triangles is increased,
the global form from the local conditions and deci- the scale of the whole crocheted surface becomes
sions. The computational model does not have any bigger. In order to increase the number of trian-
material properties such as elasticity, etc. because gles, loop subdivision method (Figure 7), which is
the geometry of the concluded knitted surface is developed by Charles Loop, is applied. This method
not affected by the property of the material. On the multiplies the triangles by adding new vertices in
other hand the material affects the rigidity of the the middle of each edge [4]. This polyhedral model
concluded form. with more triangles –created through loop subdivi-
Figure 6
The whole process from digital
model to analog crochet.
sion method- is used to extract the crochet pattern, Figure 7
which will make the crocheted geometry bigger. Loop subdivision methods.
CONCLUSION
This research demonstrates that extracting the
crochet rules for each surface is possible through
computational polyhedral model. The crochet tech-
nique is more promising than Taimina’s crocheted
models that are shown in Figure 2 in order to gen-
erate different variations of hyperbolic geometries. REFERENCES
Using hyperbolic geometries in architecture have David, WH. Taimina, D. 2001, ‘Crocheting Hyperbolic Plane’,
an important potential since they present an op- The Mathematical Intelligencer, 23(2), pp. 17–28.
portunity to achieve self-contained structures. But Osinga, HM. Krauskopf, B. 2004, ‘Crocheting the Lorenz
the conventional way of building them is expensive Manifold’, The Mathematical Intelligencer, 26(4), pp.
and it results in material wastage because of com- 25–37.
plex custom-made casting. In this context, crochet Rajaraman, V. (ed) 2003, ‘Computer Algorithm’, Computer
technique could provide building hyperbolic struc- Programming in Pascal, Prentice Hall of India, New
tures by eliminating the need for complex casting. Delhi, pp. 1-3.
Moreover, the crochet rules that are extracted from
computational polyhedral model can also be used [1] www.supernaturale.com/articles.html?id=277
as generator code during the further research on dig- [2] www.en.wikipedia.org/wiki/Knitting_machine
ital fabrication of these complex crocheted surfaces [3] www.ted.com/talks/margaret_wertheim_crochets_the_
(Figure 8). coral_reef.html
[4] www.en.wikipedia.org/wiki/Loop_subdivision_surface
Figure 8
The polyhedral models that
are chosen for prototyping.
3D Regular Expressions - Searching Shapes IN Meshes
The development of an algorithm to identify recurring geometries

Gabriel Wurzer1, Bob Martens2, Katja Bühler3
1,2
Vienna University of Technology, Austria, 3VRVis Forschungs GmbH, Austria
1,2
www.tuwien.ac.at, 3www.vrvis.at
1
gabriel.wurzer@tuwien.ac.at, 2bob.martens@tuwien.ac.at, 3buehler@vrvis.at
Abstract. We have grown accustomed to performing elaborate queries on textual data,

e.g. via online search engines, file system managers and word processors. In the past
decade, retrieval methods that also work on non-textual data have become mainstream
(e.g. face recognition software). Sadly, these developments have so far not caught on
for data mining within geometrical data, e.g. 3D meshes generated in the course of
architectural work. Specifically during data exchange, such a search functionality would
be handy, as it often happens that geometry is exported but object identity is lost - think,
for example, of generative geometry or exported BIM data. In this paper, we present an
example of such a search functionality, based on angular search. Our method is inspired
by regular expressions, a string matching technique commonly employed for matching
substrings.
Keywords. Shape retrieval; angular search; sub-mesh; regular expressions.
NAÏVE IDEA
Instead of an introduction, let us jump directly to environment has to load identical geometry
the idea behind the search algorithm and see how multiple times into memory, which may pro-
it can be applied within the architectural workflow: hibit working smoothly with the mesh for lack
Assume that we have imported a large mesh into a of performance. What is needed is an approach
modeling environment, in which all information but that can replace instances of the same geom-
the list of vertices and faces is lost. Such a situation etry by a reference to a single one.
can occur during data exchange, entailing two ma- Our contribution concentrates on solving the
jor problems: mentioned problems and additionally brings for-
1. there is no object identity, i.e. we have to manu- ward a “search and replace” functionality for 3D
ally select vertices and faces belonging to an meshes. In more detail, we present an algorithm that
object in order to work with it. This can be a • can find shapes IN meshes (i.e. sub-meshes), giv-
challenging task, though, as geometry might en a search pattern in the way of a set of paths
overlap (see Figure 1a). (which we interpret as succession of angles);
2. In cases where there is more than one instance • can restore object identity, thus making it possi-
of the same geometry, a manual approach is ble to work with a possibly inaccessible collec-
highly tedious. Furthermore, the modeling tion of vertices and faces;
• can replace found geometry by a reference to a Figure 1
single geometry container; Searching and replacing in
• can replace found geometry by a different ge- meshes. (a) Restoring object
ometry. identity from a polygon soup.
Figure 1 gives two results obtained with the ap- (b) Searching and replacing
proach: In Figure 1a, object identity was restored geometry within a mesh.
from a previously inaccessible polygon soup. In Fig-
ure 2b, we have searched for the given outline and
replaced each occurrence by a different geometry.
The latter takes 19s on a 2.4GHz single-core proces-
sor (C++ implementation, mesh containing 12064
vertices), which is moderately fast. The pattern is
given as own mesh, which is automatically compiled
into a search description which our algorithm needs.
In the coming sections, we will describe exactly
how the search is done and how the mentioned
compilation proceeds (see “Background” and “Elabo-
ration”). We further provide details on the studies
conducted (see “Studies”), which have served as
test-bed and ground for discussion concerning the
future development of the tool. Before concluding,
we also deliver details on the two implementations
existing so far (see “Implementation”).
BACKGROUND quence of angles between each pair of edges

The approach is based on two underlying methods, on that path as criterium.
regular expressions and angular search. Therefore, we • Meshes can be tesselated, meaning that each
first take a quick look at both, before elaborating the edge can be subdivided. The intermediate
details of the presented method. points do not contain significant information
when we consider only angles as matching cri-
Regular expressions terion. We thus adopt the regex repetition, in
A regular expression (regex) is a textual search tech- order to “jump over” points that lie in the same
nique that specifies a searched string by supply- direction as previously encountered ones.
ing a grammar of characters to match. We do not • The specified angular paths can self-inter-
intend to give an elaborate introduction into this sect. In order to check that the same point is
topic here, as this information is widely available reached, we adopt the notion of back-referenc-
and considered a standard technique in computer es, i.e. the storage of a point that was reached
science. We instead forward the interested reader to so far for later equality comparison.
(Forta 2004) and focus further explanations on the Throughout the paper, we will use a couple
constructs that the search algorithm uses (also see of termini found in regular expressions. To begin
Table 1): with, we use the word automaton to signify a list
• A string is a sequence of characters, a mesh a of matching criteria (transitions) that are evaluated
set of vertices connected by edges. We search sequentially. As example, take the following regular
for paths within that mesh, taking the se- expression “ab”. This specifies two transitions “a” and
Table 1 Regex construct matches e.g. becomes in 3D regex algorithm
Regular expressions versus 3D character sequence, e.g. abc abc angle sequence, e.g. 90° 10°
regular expression. repetition (one or more times), e.g. a+ a, aa, aaa match vertices in same direction
backreferences, e.g. (a|b)\1 aa,bb match previously encountered point
Table 2 Transition long name meaning

Transition types used in the FAIL Failure Transition ends execution, reporting failure
automaton. MATCH Match Transition ends execution, reporting a submesh
ANG Angle Transition at current point, find edge pairs having angle
CLO Closure find points in the same direction
BOR Begin of Regex Group begins a new regex group (for backreferencing)
EOR End of Regex Group ends a regex group, storing the curring point
REF Backreference references a previously matched point or point at a certain
percentage of a visited edge
BAT Begin-At begins matching at a previously matched point
“b” which are put into a list (“a” “b”). Implicitly, two trast, our algorithm searches in 2D or 3D and addi-
more transitions are added to the head and the tail tionally takes proportions into account (i.e. surplus
of the list, namely MATCH and FAIL. Both signify a to the angular search). Examples of other shape re-
stop condition - in the first case, the algorithm has trieval techniques, which use statistical data instead
found the supplied string, in the second case, the of angles, are the ShapeSifter tool that is based upon
algorithm has failed. The list of transitions thus be- features such as surface area, volume etc. (Sung, Rea,
comes (FAIL, “a”, “b”, MATCH). An automaton has a Corney, Clark and Pritchard 2002) and the Princeton
transition pointer, placed initially on the second item Shape Search Engine which can compare sketches
of the list (“a”). Each transition type has its own way to sections (Funkhouser, Min and Kazhdan 2003).
of matching. In the simple case mentioned, we have
a character transition, which compares the current ELABORATION
character in the string to the one specified. If both Our search technique describes an angular path the
are equal, the transition pointer is advanced (“b”). transition types given in Table 2. The most important
This process is repeated until we finally encounter one is the angle transition (ANG), which tries to find
MATCH. In the case that the criteria specified in the an edge pairs having a given angle, extending from
current transition are not satisfied, an error flag is the current point. This is usually followed by a clo-
raised and the transition pointer is set to the preced- sure transition (CLO), which jumps over any interme-
ing transition. diate points lying at the same angle (as mentioned,
these do not contain significant information). As
Angular search further transition types, we have begin and end of a
Angular search is concerned with finding a se- regular expression group (BOR, EOR), backreference
quence of angles in a given geometry. Examples of (REF) and begin-at (BAT). These are described in due
such algorithms are to be found in the automotive course, using examples to help understanding. As in
industry, in the form of a search tool for mechanical regular expressions, we also have FAIL and MATCH;
parts in a large CAD database (Berchtold and Kriegel the latter reports the points encountered during the
2004). However, in the concrete case mentioned, whole matching process, i.e. the sub-mesh found.
only section plans were taken into account, and the We will now walk through the different pos-
whole algorithm was limited to 2D retrieval. In con- sibilities for implementing a regular expression au-
Figure 2
Transitions. (a) Angle and clo-
sure (b) backreference to point
and (c) to edge, (d) branching
at a point using the Begin-At
transition, (e) branching 50%
of an edge, counting from its
start. (f) Relative edge lengths
used to introduce proportions.
tomaton, starting with the 2D case and extending In Figure 2a, we search for an angle of 125 degrees
this to 3D. We also give details on the used compiler, formed by the current point (shown in the center),
which converts a search pattern (a mesh consisting the previous point (shown as a tiny rectangle) and
of paths) into an automaton. a possible next point (shown as circle). Two cases
are distinguished: (Case 1) If there is yet no previ-
The 2D regex automaton ous point (because we have just started), we try the
Two different modes are considered in the 2D case: combination of all neighbor points twice (neighbor
If this is the first ANG to be matched, then an inner 1 - current point - neighbor 2; neighbor 2 - current
angle (0...360°) between edge pairs situated at the point - neighbor 1), since that establishes the march-
current point is found. In all other cases, the edge ing order of the algorithm. (Case 2) In case that there
last taken is already fixed: The automaton then is a previous point, as shown, the algorithm tries to
searches for a next edge at the correct angle with continue along a non-visited neighbor which has
reference to the previous one. Depending on re- the correct inner angle. Regardless of which of both
quired strictness, angles are compared using a toler- cases the algorithm has dealt with, the marching
ance interval. The same applies to the matching of direction has been fixed (shown by an arrow). The
intermediate points. In the case no suitable angle is next transition, a closure, consumes all points of the
found, the algorithm gets to the previous transition mesh lying in that direction, which allows us to skip
and tries the next edge pair. past points that contain no significant angular infor-
Figure 2 brings examples of such a 2D search: mation.
Figure 3
Tolerance values. (a) Angular
tolerance at points, (b) closure
tolerance for marching for-
ward, (c) percentual tolerance
for matching edges.
In Figure 2b, a backreference (REF) is shown Introducing proportions

where the path self-intersects. Two steps are needed So far, the regex algorithm is length-invariant, as it
to make such checks for self-intersection possible: considers only angles. However, relative length does
The preceding transitions are enclosed in a begin- make a difference when looking at proportions: A
of-regex (BOR) and end-of-regex (EOR) group, in this square is not the same as a rectangle, for example.
case: the first regex group (#1). Internally, the au- Thus, we introduce length as shown in Figure 2f:
tomaton stores all points encountered when match- Initially, we memorize edge lengths relative to the
ing that group. The backreference REF 1 checks length first edge within the search pattern (reference
whether the current point corresponds to the last length). During matching, we can reject points if
point encountered in regex group #1. they do not lie at a certain distance of an edge start,
A variation of a backreference is given in Figure in the following manner: The edge start is given as
2c: Here, REF 50% checks whether the current point an ANG transition, the edge itself is matched via a
corresponds to one of the matched points of a regex following CLO. The next ANG represents the edge
group situated within a certain percentage of the end (and, at the same time, the start of a new edge).
edge length. Thus, this specific flavor of REF checks When trying to match the latter angle, one would
for self-intersections with an edge. usually start at the last point matched by CLO and
Figure 2d shows a fork: The regex has so far then go back point by point until we have either
matched regex group #1, arriving at the point found the angle or there are none left. However,
shown in the center. It is now possible to begin at because the relative edge length is known, we can
that point when matching, using the begin-at transi- consider only points established by CLO that are
tion at group 1 (BAT 1). In the example shown, one situated at a certain distance of the edge length
might BAT 1 for matching the right path, before ex- (expressed in percent of the reference length). This
ecuting BAT 1 to find the left path. minor modification is all that is needed to include
One may also begin matching at a certain per- proportions.
centage of a previously encountered edge, as shown
in Figure 2e: From the matched points of the given Tolerance intervals
regex group, the algorithm selects the ones lying A point yet unaccounted for are the tolerance values
within a tolerance interval around the given per- which govern the strictness of the search algorithm.
centage (BAT 50%) and tries to match onward from We have three such intervals, as given in Figure 3:
these. • Angular tolerance (Figure 3a) defines what de-
viations from a prescribed angle is acceptable.
Figure 4
3D regex. We need an ad-
ditional angle at each next
point, (a) the angle between
the previous normal and
the next leg. (b) There is an
ambiguity for successive 90°
angles. Through comparison
of lengths, we can find out
whether we have an (c) inside
• Closure tolerance (Figure 3b) specifies what de- An intermediate point neighbor’ is a point situated or (d) outside winding.
viations from the marching direction are to be at length p on the edge (testpoint, neighbor). The
accepted. distance (prev. ANG’s b, neighbor’) is defined to be q.
• Percentual tolerance (Figure 3c) states the in- r is the length (prev. ANG’s b, testpoint). If q is small-
terval around a percentage of an edge’s length. er than or equal to r, we can conclude that we have
These three values are specified globally, for the an “inside” winding (Figure 4c). In all other cases, we
time being. However, results obtained with the help have an “outside” winding (Figure 4d). The winding
of some basic test cases (see “Studies”) show that criterion is added to the transition specification and
this is a possible weakness of our algorithm, as we compared at runtime with the mesh, for cases in
cannot easily adapt to sub-meshes that contain which successions of 90° angles are present.
parts which require a more fine-grained, localized
notion of tolerance. Thus, this part is likely to be ex- The regex compiler
tended in future implementations. The regular expression description is translated
automatically from a search pattern made of paths
The 3D regex automaton (ordered edges) into a regex automaton. Briefly
For the 3D case, the 2D regex algorithm is extended. outlining the algorithm, we sequentially translate
In order to fix a next edge ei+1, we need to take the each edge pair into an ANG CLO (first pass). In that
last two edges ei and ei-1 into account (refer to Figure process, we ignore co-linear edges. However, these
4a): The cross product ei x ei-1 gives the normal vec- are needed later for checking intersections (second
tor n of the plane in which the last two edges lie. A pass): (a) In case the end vertex of the forward edge
suitable next edge is one that (1.) has the correct an- was already visited, we generate a REF after ANG
gle between ei and ei+1 (same as in the 2D case) and CLO. There are two distinct cases: If the visited ver-
additionally (2.) has the correct angle between n tex was co-linear (i.e. it was ignored during the first
and ei+1. Because of this, we need at least four points pass), we generate an edge reference (REF %). In
in the mesh to be searched. all other cases, we surround the edge that leads to
An ambiguity arises for cases in which there is the point with BOR..EOR and generate a backrefer-
an ANG 90° following an ANG 90° (see Figure 4b), ence to that regex group (REF #). (b) In case the start
since it is not clear whether to march left (“outside”) vertex of the forward edge was visited, we generate
or right (“inside”). This case can be resolved through a BAT before ANG CLO in the previous fashion (co-
projection: Let (prev. ANG’s b, prevpoint, testpoint, linear: BAT %, else BOR…EOR BAT #).
neighbor) be successive points, neighbor being the
candidate for marching onward. Then, we have to fix STUDIES
three lengths p, q and r, as follows (refer to Figures Under this section, we examine the studies con-
4c and d): p is the length (prev. ANG’s b, prevpoint). ducted with the regex algorithm in some detail. In
Figure 5
Basic Test Cases. (a) Sphere
Matching (b) Proportion
Check. Replacement of (c)
Openings (d) Pyramid steps
(e) Leafs.
all cases, we have applied a single regex pattern bling a room, simplified as cubes of different
(acting as input) to a scene (also an input), produc- sizes (Figure 5b). We were able to find different
ing an output in the form of a set of selected ver- geometries based on their proportions, how-
tices and edges of the found sub-meshes. During ever, it must be mentioned that we also had
evaluation, the number and type of matched sub- counterintuitive cases where geometry is so
meshes (not all were “correct” in a visual sense, even proportionally close that one regex intended
though the angles and proportions matched) were for a specific type of furniture also returned a
compared to the type of regex used (ranging from different geometry (not a false positive in the
“closely resembling” the searched sub-mesh to more classical sense, though). We have furthermore
perturbed versions). In a post-step, the replacement perturbed the regex pattern, and checked that
algorithm has been used on the selected point- and angular precision can comfortably cope with
edge-sets or their connected components, typically such errors.
placing and orienting an object such that it fit into • We tested the 3D “search and replace” algo-
the resulting bounding box. The latter is quite trivial rithm using the colosseum mesh shown in
to extend to arbitrary geometry that would be ill- Figure 5c. For every instance of the matched
suited for bounding-box placement, using specific search pattern (shown red in the lower part of
rules stated in a scripted program of the modeling Figure 5c), we used the bounding box to locate
platform. the replacement. The orientation (heading,
pitch, roll) of the replacement was concluded
Basic test cases from the found points in comparison with the
During the development of the approach, we have search pattern. Further tests for “search and
used a set of basic test cases for assessing the algo- replace” were also conducted with a pyramid
rithm: (Figure 5d) and a gerbera, which was turned
• In the simplest case, we have looked at a set of into a lily by replacing each leaf (Figure 5e).
spheres (Figure 5a) with increasing tesselation
(4, 8, 16, 24 vertices as base). As result, we could Reconstruction of destroyed synagogues
show that a regex of k ANG transitions can only We are currently testing the 3D regex algorithm
find objects with tesselation greater or equal on large-scale models in the context of virtual re-
to k (a 8-ANG regex will find the 8-, 16- and construction of destroyed synagogues, mainly
24-sphere but not the 4-sphere). stemming of the period 1890-1910 (see Figure 6).
• Proportions were tested on a scene resem- Though hundreds of synagogues were in this era
Figure 6
Matching “windows” in the
synagogue case study. (a)
Overview of matches, (b)
arc - real size, (c) proportional
size, (d) non-proportional size,
(e) connected arc - real size,
(f) proportional size, (g) non-
proportional size.
built all over Europe, a significant portion has been geometries.

destroyed in 1938. The option of re-erecting these To which kind of geometrical representations
sacral buildings is not on the agenda, particularly would this predominantly apply? In the past years
due to a missing usership locally. By way of a virtual the following sets of entities are for the building
reconstruction a certain degree of commemoration type “synagogue” of repetitive nature: furniture,
is facilitated. However, issues of incompleteness and bima and torah ark, ornaments, doors and windows,
missing bits and pieces of information play a central banisters, columns and ceiling beams, tower and
role in the process of reconstruction. dome elements. For our study, we have taken a first
The collection of already realized 3D models step in matching repetitive geometry, in the form
serves as a knowledge base for ongoing reconstruc- of doors and windows (which we model as arcs, see
tion activities. However, the secure detection of Figure 6 for an overview), ornaments (Figure 7a), col-
already existing modeled elements - aiming at “re- umns (Figure 7b) and ornamented windows (Figure
use” - is cumbersome. First of all a subset of suitable 7c). We are far from finished with that undertaking,
entities has to be identified and eventual adapta- but the first results are already quite promising with
tion of the existing modeling properties might be respect to insights that would occur in a “real” pro-
considered. Instead of a laborious manual search ject situation where geometry is to be searched.
through individual models, the goal is to be pointed To begin with, the mesh we are trying to search
in a straight-forward and structured way to (similar) in can be considered a pathological case: Albeit be-
building elements in the whole model collection. ing seemingly well-structured (see Figure 8a), a clos-
The 3D-model itself can be regarded as a structured er look reveals that it is composed of a multitude of
database. Extracting information from a large set overlapping components with little or no semantic
of models would enhance the orderly subsequent interconnectedness. In the example in Figure 8b, we
re-use of already modeled geometries. Any further can see a door that is formed by a wall in the back-
building models added to the collection would pre- ground (which is shared with another door to the
sumably donate to the range of so far not recorded left), a single arc in the foreground, and several ar-
Figure 7
(a) Ornaments, (b) columns,
(c) ornamented arcs.
Figure 8
(a) Pathological door made of
(b) arbitrary sub-meshes.
bitrarily connected arcs in the middle. Matching has

proceeded only in the foreground arc, since that is eling environment (written in C++), and a more
at least connected. The truth is that, however, a de- academical implementation utilizing the NetLogo
signer would not know why the regex cannot match programming language. The first one was written in
the whole door shown. What would be needed is 2002 as part of the diploma thesis of the first author;
an algorithm that can combine overlapping geom- the second one is a vanilla implementation that
etry in order to facilitate a “what you see is what you seeks to faithfully implement what was written in
match”-sort of approach, which we have still not the thesis, in order to have some degree of quality
produced (future work). control for this paper. NetLogo is rather slow with re-
Furthermore, there is a subtle influence by the gards to performance, but its visualization capabili-
regex chosen for arcs. Figures 6b and 6e show the ties and functional programming language make it
real sizes of arcs that are matchable in the mesh. an ideal test-bed for exploring further extensions of
However, as the regex is size invariant, it actually the approach.
“sees” only relative proportions as depicted in Fig- Coming to performance, we disregard the Net-
ures 6c and 6f. Even if we increase percentual toler- Logo implementation (which is rather slow because
ance, we still get false positives, such as those above of running on a Java Virtual Machine with added
the main entrance (see Figure 6a). Optionally com- NetLogo stack on top). The C++ implementation is
paring the actual size of the pattern to the found in- better - at average, 650 vertices per second per core
stance may solve the problem, but this is again to be (2.4 GHz 32-bit processor), although this largely
done as future work. depends on the mesh structure (good tesselation,
Choosing a regex without proportions reveals connectivity/density), precision settings (more toler-
that the arcs all have the same radius (see Figures ance means more possibilities are tried, which slows
6d and 6g). However, this is a bad choice, since that down the algorithm) and the use of proportions
would also match every half-circle (height of the (not using them generates more possibilities, again
lower part close to zero). Without proportions, the slowing down the program). For example, in the
situation is even worse in connected search patterns synagogue use case described earlier, the algorithm
(Figure 6g), since we could get a deformed match would find occurrences of the simple arc show un-
which is hardly what we wanted in the first place. der Figure 6c in either 6min 30s when the precision
There is no way to circumvent this problem - match- values were very strict (regex exactly resembling the
ing without proportions just produces bad results sub-mesh, all precisions set to 1) or 1h 40min for a
(or, more precisely, results which are correct from an very loose setting (regex approximating the sub-
angular view but counterintuitive to us). mesh, angular and closure precision 30, percentual
tolerance 10). This boils down to a performance of
IMPLEMENTATION either 1500 (strict case) or 100 vertices (loose case)
At the moment, we have two different implemen- per second per core, with raises a variety of ques-
tations: A plugin for the Maxon™ Cinema4D® mod- tions and analysis tasks for future work.
In that context, we also wish to note also that library of patterns used for matching as well as a
the overall efficiency is highly coupled to the search taxonomy that connects these would seem a useful
pattern (and thus: the compiler) used; a regex that extension that is yet too early to undertake, as we
discriminates early (i.e. sharp angles first, before have to fix the foundations first. Other tasks that we
coming to rounded forms) has a far better perfor- would like to look into in the future are: data extrac-
mance than one that considers discriminating fac- tion from laser scan data and 3D fractal analysis of
tors as last step. Devising a better regex compiler architecture based on “finding sub-meshes within
and searching in an optimized mesh (overlapping sub-meshes”.
edges and points merged) are clearly on our agen-
da. Also, future versions of the approach might lead ACKNOWLEDGEMENTS
away from the idea matching linearly in an automa- This work is based on a diploma thesis (Wurzer,
ton but recursively in the supplied search pattern 2004) supervised by Katja Bühler (Vienna UT and
(i.e. without compilation, but still utilizing the pre- VRVis Forschungs GmbH), Peter Ferschin and M.
sented concepts). Eduard Gröller (Vienna UT). The synagogue model
base is a results of a continuing effort in virtual re-
CONCLUSION AND OUTLOOK construction by Bob Martens (Vienna UT), Herbert
We have presented an algorithm that can search in Peter (Academy of Fine Arts Vienna), among many
meshes that have lost all information but their ver- others participating in that effort.
tices and faces, based on regular expressions and
angular search. The benefits of this are threefold: (1.) REFERENCES
We can restore object identity, (2.) we can replace Forta, B 2004, Sams Teach Yourself Regular Expressions in 10
multiple instances of the found geometry by a refer- Minutes, Sams, Indianapolis.
ence to a single geometry container and (3.) we can Berchtold, S and Kriegel, H-P 1997, ‘S3: Similarity Search
replace found geometry by an alternative one. in CAD Database Systems’, Proceedings of the SIGMOD
Two case studies frame the presented approach: Conference, May 13-15, Tucson, USA, pp. 564-567.
The “basic test cases”, which we applied during de- Sung, R, Rea, H, Corney, JR, Clark, DER and Pritchard, J 2002,
velopment, and the ongoing “synagogue” test cases, ‘Shapesifter: A retrieval system for databases of 3D en-
which use a collection of models exported from gineering data’, New Review of Information Networking,
CAD. As discussed, the first results with the latter 8 (1), pp. 33-53.
domain have shown that the complexities associ- Funkhouser, T, Min, P and Kazhdan, M 2003, ‘A Search En-
ated with “real” data are not to be underestimated: gine for 3D Models’, ACM Transactions on Graphics,
The data is both huge (typically 350K vertices, 450K 22(1), pp. 83-105.
polygons) and of bad quality (overlapping geom- Peter, H 2001, Die Entwicklung einer Systematik zur virtuellen
etry, unintelligible polygon groups forming con- Rekonstruktion von Wiener Synagogen, Diploma Thesis
nected components, bad tesselation). On top of this, (Vienna University of Techology).
the expectation regarding the growth of the model Sung, R, Rea, H, Corney, JR, Clark, DER and Pritchard, J 2002,
collection in the next coming years is expected to be ‘Shapesifter: A retrieval system for databases of 3D en-
substantial. gineering data’, New Review of Information Network-
A meta-search will therefore become of even ing, 8 (1), pp. 33-53.
more importance, connected with a pre-step for Wurzer, G 2004, 3D Regular Expressions: Searching IN Meshes,
automated simplification and cleaning of the mesh Diploma Thesis (Vienna University of Techology).
which lies on our future agenda. Also, building a
A Computational Method for Integrating Parametric
Origami Design and Acoustic Engineering
An application to a concert hall design

Tsukasa Takenaka1, Aya Okabe2
1
Toyohashi university of technology, Japan, 2Toyohashi university of technology, Japan
1
http://ans-studio.com, 2http://ans-studio.com
1
takenaka@ans-studio.com, 2okabe@ans-studio.com
Abstract. This paper proposes a computational form-finding method for integrating

parametric origami design and acoustic engineering to find the best geometric form of a
concert hall. The paper describes an application of this method to a concert hall design
project in Japan. The method consists of three interactive subprograms: a parametric
origami program, an acoustic simulation program, and an optimization program. The
advantages of the proposed method are as follows. First, it is easy to visualize engineering
results obtained from the acoustic simulation program. Second, it can deal with acoustic
parameters as one of the primary design materials as well as origami parameters and
design intentions. Third, it provides a final optimized geometric form satisfying both
architectural design and acoustic conditions. The method is valuable for generating new
possibilities of architectural form by shifting from a traditional form-making process to a
form-finding process.
Keywords. Interactive design method; parametric origami; acoustic simulation;
optimization; quadrat count method.
INTRODUCTION
Design for a concert hall includes acoustic engineer- acoustic engineering to find the best geometric
ing and architectural design (i.e. aesthetically pleas- form of a concert hall. First, we discuss the limita-
ing design that satisfies complex architectural conditions of conventional collaborations between archi-
tions, such as concert activities, building regulations, tects and acoustical engineers. Second, to overcome
structure, construction processes, budgets and so these limitations, we develop an interactive design
forth). Usually, computation is often used for gener- method and show its application to a concert hall
ating all possible geometries fulfilling those various design project in Japan (the hall will be completed
architectural constraints. However, the most difficult in 2014). The design method consists of three inter-
part of design processes is to choose the best geo- active subprograms: a parametric origami program,
metric form among the resulting various alternatives. an acoustic simulation program, and an optimiza-
This paper proposes a computational method tion program. Finally, we describe the advantages
for integrating parametric origami design and of the proposed method, including the ease with
Figure 1
A diagram of the interactive
design method.
which it visualizes engineering results obtained engineers. To bridge this gap, we propose a compu-
from the acoustic simulation program, and a final tational design method for integrating architectural
optimized geometric forms it provides to satisfy design factors and acoustical engineering factors.
both architectural design and acoustic conditions. In addition, we want to develop an objective
Because the method efficiently manages funda- method in which acoustic data derived from a simu-
mental factors underlying architectural forms, it can lation process are efficiently utilized. In this connec-
provide a design framework in which architectural tion, Leach (2009) mentioned as follows: “Within
design and acoustic engineering are integrated. contemporary architectural design, a significant
shift in emphasis can be detected – move away from
THEORETICAL BACKGROUND an architecture based on purely visual concerns to-
In the design process for a concert hall, architects wards an architecture justified by its performance.
collaborate with acoustical engineers. First, archi- Structural, constructional, economic, environmen-
tects develop a geometric form and then acoustical tal and other parameters that were once secondary
engineers analyze the acoustic efficiency of the pro- concerns have become primary – are now being em-
posed form using their simulation program. How- braced as positive inputs within the design process
ever, there are relatively few exchanges between from the outset”. Our proposed interactive design
them. As a result, in the conventional method archi- method can deal with acoustic parameters as one of
tectural optimization and acoustical optimization the primary design materials as well as origami pa-
tend to be rather independent operations, and they rameters and design intentions (Figure 1).
are not always coordinated. For instance, acous- With the recent improvement of computer per-
tic optimization does not always take into account formance, simulation technology has improved sig-
complex architectural conditions or the architects’ nificantly. As a result, it has become easy to visualize
design intentions, whereas architects do not always the state of the acoustic parameters. What makes
utilize informative data provided by the acoustical our method intriguing is that those parameters can
Figure 2
The parametric origami
program can transform a
flat sheet of paper into a
geometric form through
various folding techniques:
basic techniques (Yama-
ori, Tani-ori, Nakawari-ori,
Kabuse-ori), advanced tech-
niques (Jyabara-ori, Miura-ori,
Hira-ori and so on).
find unpredictable forms which meet both acoustic Nagata Acoustics (an acoustical consulting firm). The
conditions and design intentions. design method consists of three interactive subpro-
grams: a geometric form-generating program, an
EXISTING RESEARCH acoustic simulation program, and an optimization
In the existing studies, the use of computational program.
methods for designing concert halls is limited to
performing two tasks: acoustic simulations and Geometric form-generating program: the
generation of all possible geometries satisfying parametric origami program
various architectural constraints. However, there are The first subprogram, the parametric origami pro-
few methods for choosing the best geometric form gram, adopts the idea (proposed by the SUEP archi-
among the resulting numerous alternatives. tects) that a form is generated by folding a sheet of
In this paper, we apply a computational method paper—the traditional Japanese art called ‘origami.’
not only to acoustic simulation and generation of The program can transform any surface into a geo-
various possible geometries but also determination metric form using the basic folding parameters of
of the best geometric form satisfying both the archi- the origami folding system: folding lines, folding
tectural design and acoustic requirements. depth, folding width, folding angles and so on (Fig-
ure 2). These are mutually constraining (i.e., not in-
CONCERT HALL DESIGN PROJECT dependent) parameters.
In this paper, we apply the interactive design meth- The objective of this parametric origami pro-
od to a concert hall design project in collaboration gram is to develop a method for finding combina-
with SUEP architects (an architectural office) and tions of origami parameters which generates ge-
ometries fulfilling complex architectural constraints
Figure 3 (Figure 3). The computer technology enables us to
A diagram of generating test every combination of parameters in order to
architectural form fulfilled find out possible designs that meet certain require-
complex architectural con- ments. In this process, designers are no longer mak-
straints. ing single geometry but finding design parameters
which determines a final form (Figure 4).
Another feature of this program is: if there is no
combination of parameters that meet every require-
ment, then the program provides an alternative
Figure 4
Possible form variations.
combination. In the literature, a few studies deal sound propagation in terms of straight rays.
with computational origami methods for architec- There are some existing software packages
tural design. Most of them follow a strict origami which can simulate sound propagation and geo-
rule such that a single sheet of paper is fold into a metric forms interactively. However, it can simulate
given polyhedral surface without any cut. However, only the distribution of direct sound, which is not
in architectural design processes, this rule some- enough for sound optimization of a concert hall.
times disturbs design intention or other architec- To overcome these limitations, we developed an
tural performances. To overcome this limitation, the acoustic simulation program which visualizes sound
method enables us to balance parameter weights in propagation in a three-dimensional space over time
an optimization process. That is, the method allows in three ways: by arrows originating from a sound
us to cut a sheet of paper or loosen architectural source at an arbitrary point in a hall; the distribution
constraints, in the process of balancing between ori- of reflected sound; and the distribution of reverber-
gami rules, acoustic performances and design. ating sound (Figure 5).
These two subprograms, the acoustic simulation
Acoustic simulation program program and the parametric origami program, run
The second subprogram is an acoustic simulation interactively in the following manner.
program, which deals with geometric acoustics, i.e.
Figure 5
Left: reflected sound simula-
tion, Right: reverberating
sound simulation.
Figure 6
Distribution of sound density.
Optimization program by the quadrat

count method by the Nagata Acoustics) include: first, the sound is
First, the parametric origami program generates all distributed evenly over the hall and audience seats
possible geometries of a hall according to the pa- in 30 to 90 ms; second, there is no echo or flatter
rameters derived from the architectural conditions echo; and third, there is no sound focus. To examine
(resulting from building regulations, structure, con- whether or not sound is uniformly distributed over
struction processes, budgets and so forth), the fold- the hall, we applied the quadrat count method with
ing parameters determined by the architect’s aes- the Poisson distribution. The program automatically
thetic sense and allowable parameter values derived calculates the chi-square value for testing uniformity
from the origami folding system. Then, for each pos- in each form, given by:
sible geometric form, the acoustic simulation pro- χ2 =∑(O-E) 2/E,
gram visualizes sound propagation and the distribu- where O denotes observed values and E denote ex-
tion of sound reached at each audience member’s pected values (Figure 8).
seat (Figures 6 and 7). Given the outcomes of each Another notable aspect of the optimization
possible geometry, the optimization program judg- program is that it can easily change the conditions
es which is the best combination of parameters that of acoustic requirements which vary according to
satisfies both the architectural design and acoustic collaborators. Such flexibility produces various ar-
requirements. Acoustic requirements (proposed chitectural designs in accordance with collaborators.
Figure 7
The part of various possible
geometries of a hall according
to the parameters derived
from the architectural condi-
tions, and the results obtained
by the acoustic simulation
program.
The results of the project chitectural constraints.

The concert hall design project is part of the design • To find out a geometric form satisfying not only
project for a cultural-arts complex in Ureshino-shi, acoustic performance but also designer's in-
Saga-ken (southern part of Japan). SUEP architects tention. Because following the advice of acous-
designed this whole project under the design con- tic engineers often results in a cave-like form
cept of a folding roof. In collaboration with them, (which is acoustically effective) but such a form
we considered that the following three conditions does not always meet architectural design in-
should be satisfied. tentions.
• To shorten the processing time for finding vari- • To discover unexpected geometries which op-
ations of origami folding patterns fulfilling ar- timize for both acoustic and design conditions.
Figure 8
The results obtained by the
quadrat method
Figure 9
Left: the final model of the
concert hall generated by
modified Miura-ori taking
account of various folding
parameters. Right: Details of
the final model of the concert
hall.
With the proposed computational method, to take account of acoustic parameters as one of the
about two hundred possible geometries were gener- primary design materials as well as designer’s sense.
ated, among which the final geometry of the concert Computational technology is not only useful for
fall was chosen through the optimization process. improving or automating design processes, but it is
The folding pattern of the final geometry is also valuable for generating new possibilities of ar-
based on Miura-ori, consisting of concave polyhe- chitectural form by shifting from a form-making pro-
dral surfaces. At first glance, the final geometry looks cess to a form-finding process.
simple but it is complex in the sense that the folding
depth and angle are delicately controlled (Figure 9). REFERENCES
Diggle, PJ 2003, Statistical analysis of point patterns, Arnold
CONCLUSION publishers, New York.
This interactive relationship enables us to choose Egan, MD 1941, Architectural acoustics, J. Ross publishing,
the best combination of parameters satisfying both New York.
architectural design and acoustic requirements Leach, N 2009, ‘Digital Morphogenesis’, Architectural Design,
among numerous possible forms. Parametric design V 79, I 1, pp. 34-37.
is often used to explore complex geometries, but in Reas, C and McWilliams, C 2010, Form+Code in design, art,
this method it is used to promote complex interac- and architecture, Princeton Architectural, New York.
tions of collaborators. Tachi, T 2010, ‘Freeform variations of origami’, Journal for ge-
Terzidis (2006) mentioned about a form-mak- ometry and graphics, 14(2), pp. 203–215.
ing process as follows: “architects and designers Takenaka, T and Okabe, A 2010, ‘Networked coding method
believed that the mental process of design is con- for digital timber fabrication’, Proceedings of the ACADIA
ceived, envisioned and processed entirely in the hu- Conference, Calgary/Banff, Canada, pp. 390-395.
man mind and that the computer is merely a tool for Terzidis, K 2006, Algorithmic Architecture, Architectural
organization, productivity, or presentation”. Howev- Press, Oxford.
er, the computational form-finding process allows us
A Novel Method for Revolved Surface Infrastructures
Gökhan Kınayoğlu
Bilkent University, Turkey
http://www.tectonicon.com
gokhan.kinayoglu@bilkent.edu.tr
Abstract. This paper presents an algorithm for the formation of single or double curved
revolved surface’s infrastructures through standardized parts. Any revolved surface
can be generated with only two types of parts, interconnected by a ribbed structure
technique. The proposed method differs from the accustomed orthogonal rib structures by
the varying angle in-between coupling parts. The algorithm can be customized through
several parameters like the number, width of parts and thickness of the material used
for the infrastructure. The algorithm also offers an advantageous nesting pattern with
minimum loss of material regardless of the revolved surface cross-section.
Keywords. Revolved surface; standardization; ribbed structure; contouring; nesting
pattern.
INTRODUCTION
Following the design process, the transformation of number of unique variations, each requiring an in-
a complex shape from the digital medium of com- creased degree of computation and a heavy process
puter software into the physical reality through of production. Therefore, rationalization is required
material existence requires a further computation to transform the non-standard parts into standard-
and rationalization (Griffith et al., 2006). The ration- ized ones, in terms of their geometry, variability and
alization process, depending on the geometrical economy. Besides rationalizing the surface parts, the
complexity of the intended shape, aims for the infrastructure that builds up the surface also needs
standardization of parts for possible ways of manu- a process of computation and rationalization (Figure
facturing, while lowering the manufacturing costs 1). It is the latter one, the infrastructure of a surface
at the same time. Therefore, the process of ration- that this study is going to focus on specifically.
alization becomes an inevitable part of computa- This study is an attempt to devise a methodol-
tional design, as long as the target shape is a non- ogy for producing single or double curved surfaces
standard surface and requires non-standard parts through standardized parts. This kind of approach
for its constructability. By non-standard surface, any can be considered as the primary step for enhanc-
surface - single or double curved - that cannot be ing the current computational processes and bring-
built through conventional manufacturing and con- ing forth a novel method for the fabrication of sin-
struction techniques, and requires a computational gle or double curved surfaces. As Branko Kolarevic
design process for both rationalization and manu- states, the production of a surface, whether single
facturing is indicated. In the case of non-standard or double curved, can be realized through “contour-
surfaces, the number of units may end up in a large ing, triangulation, use of ruled, developable surfaces
Figure 1
An example of the infrastruc-
ture for a spherical revolved
surface.
and unfolding” (Kolarevic, 2003, p. 43). The devised tectural projects, especially in skyscraper structures.
method can be regarded as an innovative contribu- The environmental and structural factors make it
tion to the contouring technique in which the slices suitable for the diagrid system to be used in high-
are not ran parallel to each other in two orthogonal rise buildings. Therefore, another objective of this
directions. paper can be considered as the investigation of
As the starting point of this study, revolved sur- possible extensions of the diagrid system through a
faces have been chosen, for their constant curvature smaller scale implementation.
along the rotation axis. The constancy of curvature
among the rotation axis enables the standardiza- RELATED WORK
tion of parts and apart from the cross-section of the Apart from the orthogonal ribbed structures, which
revolved surface, the algorithm generates only two have been studied and implemented extensively,
types of parts which form the final infrastructure. studies made on non-orthogonal ribbed structures
The method introduced in this paper focuses mainly have been analyzed. Agnieszka Sowa’s study, at ETH
on small scale infrastructures, which are possible to Zurich, explores the possibility of separating parts
be manufactured from sheet materials, like card- in relation to each other, for going beyond the sca-
board, acrylic, fiberboard or metal. lar limitations of the material and manufacturing
The final infrastructure can be denoted as a techniques. (Sowa, 2004) Sowa’s method focuses on
diagrid system, which is being widely used in archi- generating and optimizing a cubic structure as an
instance of many possible forms, through a number a predefined maximum amount of variation in the
of cross-sections integrated through a ribbed struc- intersection angles. Therefore the firmness of the
ture, by using an algorithm. The parts are manufac- structure is attained when the assembly of the parts
tured from planar timber elements with a constant is completed.
thickness. Similar to this study’s approach, the varia-
tion of angles between the parts has not been con- ALGORITHM
sidered in Sowa’s study, but instead all parts were The algorithm was devised through a script writ-
manufactured by a two-axis CNC milling machine, ten in Autodesk Maya’s Maya Embedded Language
hence the perpendicular slits, apart from the angle (MEL) and later re-implemented in Rhinoceros and
of intersection. The main aim of the study is to de- Grasshopper plug-in. It produces the production
vise an algorithm that is capable of generating the drawings of the parts of the infrastructure and they
intersecting ribs, optimizing the structure through can be manufactured by a two-axis router, laser-cut-
some parameters, separating each rib where nec- ter or water jet, for the planar quality of the parts at
essary and nesting the manufacturing drawings for the current stage of the study.
generating the cutting scheme. The algorithm starts with a cross-section curve on
In Kenfield Griffith, Larry Sass and Dennis the XY plane for the revolved surface, where Y-axis
Michaud’s study of generating a strategy for irration- is the axis of revolution; hence farther the curve to
al building design, contouring technique is adopted Y-axis, larger the revolved surface will be. The cross-
for irrational surfaces, generating horizontal and section curve should not intersect the Y-axis for
vertical ribs, with every part different from each oth- the preservation of tubular quality of the revolved
er (Griffith et al., 2006, p. 467). While horizontal ribs surface. Additionally, because of the limitations of
are located at differing levels with equal intervals the infrastructure, the cross-section curve should
and remain parallel to each other, the vertical ribs pass the horizontal line test, i.e. a line in X direction
are generated perpendicular to the surface, end- should intersect the cross-section only once at any
ing up with a more complex arrangement. The final point, however a line in Y direction can intersect the
structure resembles a waffle slab system used in cross-section any number of times. By revolving the
concrete constructions, projected onto a curvilinear cross-section around the Y-axis for 180˚, a revolved
irrational surface. The perpendicularity of horizontal surface is formed.
and vertical ribs solves the problem of slit angle vari- The surface is then intersected by a plane ly-
ation, thus enabling the manufacturing of the coming in the YZ plane and inclined through Z-axis. The
ponents on a two-axes milling machine. However, degree of inclination is the first parameter of the
it should be noted that the variance among shapes infrastructure. To guarantee the infrastructure cover-
creates a highly irregular nesting pattern. age of the whole revolved surface, the intersection
An ongoing study by Yuliy Schwartzburg and points of the plane and the revolved surface at the
Mark Puly, at Ecole Polytechnique Federale Laus- top and bottom points should be checked, and the
anne, Switzerland, explores the possibility of con- angle of inclination may be lessened in order to fully
structing any shape through intersecting planar intersect the revolved surface.
shapes (2013). A devised algorithm searches for The resulting curve of the intersection is cop-
solutions taking into consideration the cases of in- ied towards the Y-axis with a certain distance, the
tersection and optimizing the intersection angles parameter for the width of all infrastructure parts,
and positioning of every rib through assembly limi- and two intersecting curves form the main construc-
tations, slit constraints and material qualities. The tive element by lofting to generate a surface. The
study also takes into account the varying angle in- resulting surface is arrayed radially N times around
between the parts and comes up with a solution of Y-axis with a total angle of 360˚, where N is the third
Figure 2
Three examples of the infra-
structure.
parameter of the infrastructure. By mirroring the re- in-between the parts, slit width increases and a 90˚
sulting surfaces across XY plane, the infrastructure is intersection produces slit width equal to the mate-
formed by planar elements with a total number of rial thickness.
elements of 2*N.
Beyond the computation of the slit dimensions, ADVANTAGES, LIMITATIONS AND FU-
thickness of the material is also useful for visualiza- TURE WORK
tion purposes (Figure 2). By extruding the parts with When the same revolved surface is produced
thickness of the material, the infrastructure is basi- through orthogonal ribbed structures, in which par-
cally formed, except for the slits. allel horizontal and axial vertical sections are used to
Each part consists of a number of intersections generate the infrastructure, the method introduced
depending on the formal characteristics and size of displays some advantages. First of all, the devised al-
the cross-section, degree of the inclination angle gorithm guarantees that there will be only two types
and the number of elements. The inclined nature of parts to generate the infrastructure. However in
of each part results varying intersection degrees, orthogonal ribbed structures, while the vertical sec-
but they are limited to the number of intersecting tions will be identical due to the revolving quality
parts, i.e. if each part has 7 intersections, the model of the surface, the horizontal sections differentiate
will have a maximum 7 varying angles at total. To according to the cross-section, all being circular. Ad-
determine the position and size of each slit, angle ditionally, for all of the slits are parallel to each other
between the parts and thickness of material are on every part, stability of the infrastructure depends
used. The diagram shows the mathematical relation on the frictional forces or the infrastructure requires
between the intersection angle and the slit width additional elements to fixate the parts; whereas the
(Figure 2). With the intersection angle decreasing introduced method has the advantage of interlock-
ing itself due to the varying directions of each slit. and around the implementation of 70% of the parts,
Moreover, the standardization of parts also allows the interlocking becomes completed. It should be
the perfect nesting in the cutting scheme regardless noted that the elasticity of the material used plays
of the shape of the final part or the cross-section of an important role for the different directions of the
the surface (Figure 3). slits. While assembling the infrastructure, the parts
For the structural quality of the infrastructure is need to be bended to a certain degree for joining
not crucial in small-scale implementations and ma- them. This property also ensures the interlocking
terial efficiency has a higher priority over structure, of the infrastructure. Bendable elastic materials like
in the algorithm the intersection curve is not offset- acrylic, medium-dense fiberboard (MDF), spring
ed to attain a constant width among the curve but steel or cardboard suit well to the infrastructure.
instead copied towards the Y axis. As a direct out- If two-axis manufacturing techniques and thick
come, the parts have varying width throughout, but materials are used for the infrastructure, the surface
on the contrary they can be nested in the produc- quality will be highly coarse. This can be better visu-
tion drawings regardless of the initial cross-section alized by increasing the thickness of the material in
of the revolved surface. Additionally, as the nesting the algorithm (Figure 4). Additionally, ribbed struc-
pattern also allows a lossless configuration of the ture technique has a problematic slit connection in
parts, routing time is also decreased by the shared cases other than the parts are intersected perpen-
edges in between the parts. As long as the manu- dicularly. Both problems may be overcome through
facturing techniques allow, instead of cutting each the use of thin materials and further manufacturing
part separately and producing left-over material in- techniques. The algorithm is capable of generat-
between the parts, the parts are arranged perfectly ing the exact three dimensional model for precise
without any loss of material. However, it should also interconnection between parts, resulting from the
be noted that this kind of approach may bring forth varying degrees of intersection. Therefore, when the
structural inadequacies, for the uncontrolled vari- parts require a higher degree of precision, a five-axis
ance in the material widths. Nevertheless, the nest- milling machine becomes more adept. Instead of
ing algorithm may be updated depending on the planar pieces, five-axis milling machine will be able
structural necessities for each case, still protecting to incorporate the surface curvature of the revolved
the advantages of the lossless nesting pattern. surface into the parts through thicker materials, be-
The perfect nesting quality of the parts results sides the angle variations in the slit connections.
in high efficiency in terms of material use when Another possible manufacturing method, and
the number of parts being manufactured increase. maybe the most suitable and optimized one for
Therefore, an infinite number of parts manufactured the infrastructure is the use of injection molding
from a roll of steel leads to a 100% material effi- technique. As the number of parts required for the
ciency, which is a distinctively advantageous feature implementation of the infrastructure is limited only
when mass fabrication is considered. to two, in cases of mass-production of a specific
The assembly process follows a relatively sim- cross-section, the manufacturing of two molds will
ple system of formation. There are only two types of be enough. Together with the decreased manufac-
parts, inner and outer. Parts are interconnected to turing times and costs, the precision of the final in-
each other through corresponding slits, i.e. first slit frastructure is achieved through the use of molding.
to first, second to second and so on. For the varying Since in its current formation the model does
direction of slits, assembly process is a bit problem- not offer any structural properties, there should be
atic at the beginning when only a few number of further studies for the optimization of parameters
parts are assembled. With the increasing number of in terms of structural criteria. Number of elements,
parts, the infrastructure becomes more interlocked width and thickness of parts, degree of inclination
Figure 3
Visual representation, param-
eters and nesting patterns of
the previous three examples.
Figure 4
Slit width calculation formula.
should be tested and analyzed structurally to find inquiry into the relation between individual part
out the structural advantages and deficiencies of and overall form, together with their integral rela-
the method. Alterations to the algorithm may be in- tion in between, which cannot be separated in any
troduced according to the findings about the struc- phase of the design process. This attempt reflects an
tural behavior of the infrastructure. Additionally, understanding of an approach, which prioritizes the
the setbacks of the varying width may be studied potentials of rationalization from the initial steps of
as an outcome of the structural findings, together the design process, while also taking material con-
with the nesting possibilities apart from the algo- siderations into account.
rithm’s current potentials. Consequently, material
limitations may be overcome and larger prototypes REFERENCES
and implementations may be further achieved. For Griffith, K., Sass, L., Michaud, D 2006 ‘A strategy for complex-
attaining larger scale infrastructures, Sowa’s tech- curved building design: Design structure with Bi-lat-
niques may be adopted (2004). By dividing each part eral contouring as integrally connected ribs‘, SIGraDi
into sub-parts and connecting them with additional 2006 [Proceedings of the 10th Iberoamerican Congress of
elements, dimensional restraints may be extended. Digital Graphics] Santiago de Chile, Chile pp. 465-469.
Kolarevic, B (ed) 2003, Architecture in the Digital Age: Design
CONCLUSION and Manufacturing, Spon Press, New York.
Apart from the formal qualities of the form, stand- Schwartzburg, Y. and Puly, M. 2013 ‘Fabrication-aware de-
ard or non-standard, through designed methods sign with intersecting planar pieces‘, Eurographics, 32
and techniques, the part can be rationalized. The (2), pp.317-326.
proposed algorithm shows an example of rationali- Sowa, A. 2004 Generation and Optimization of Complex
zation of a non-standard surface through computa- and Irregular Construction/surface: On the Example of
tional processes and attaining standardized parts. NDS2004 Final Project. Postgraduate studies final the-
In a larger scope of context, this study proposes an sis in ETH Zurich. ETH Press: ETH Zurich.
Ruling Im/Material Uncertainties
Visual representations for material-based transformations

Zeynep Akküçük1, Mine Özkar2
Technical University of Istanbul, Turkey
1
zeynepakkucuk@gmail.com, 2ozkar@itu.edu.tr
Abstract. Visual rules are powerful in loosely capturing the impact of material behavior
on form in designer’s hands-on experimentation. They present a first step to translate
the causal relations between material and form to computation without sacrificing the
uncertainties in the designer’s interaction with the materials. This study investigates
how to model the relation between material and form with visual rules so that the
model embodies some of the phenomenological aspects of reality, rather than merely
reproducing it.
Keywords. Digital materiality; physics-based modeling; abstractions; visual schemas;
shape studies.
INTRODUCTION
Recent developments in programming and digital interacts with the materials on an immediate level or
production technologies create a new conscious- builds a system of different materials and lets them
ness within the architectural profession, yielding interact with each other while s/he acts as the ob-
to new design methodologies. The high level of server and the controller of this process. The altera-
product precision in digitally calibrated fabrication tion of the designed form based on these interac-
requires a high level of precision in design represen- tions is phenomenological, in that it involves the
tation. This numerical certainty finds its expression interpretation of various instances of the materials
in mechanistic design approaches that make use of that are “transcomputable” (Glanville, 1998). In this
quantifiable, solid data for performance and opti- paper we focus on incorporating the founding rela-
mization. However these approaches mostly adopt tions (Rota, 1997) of form and material to address
the limitations of existing computational techniques the interpreted in design representations through
instead of exploring design beyond the limits of the visual schemas (Stiny, 2011).
quantifiable phenomena. In the following paragraphs we discuss digital
Digital representations that are constructed and physical experiments. We review preliminary
with mathematical descriptions of the physical ob- studies firstly in the physical modeling of plaster
ject are only capable of reproducing some part of in elastic formwork and secondly in visual abstrac-
the reality. The description of a reality limited to its tions of this process in different digital modeling
known finite qualities is insufficient for the designer approaches. We then develop and present a set of
who alters this reality in direct and indirect ways visual schemas to illustrate the physical processes in
throughout the design process. The designer either material based transformations.
Figure 1
SANAA, House, New York, USA,
2008. Image source: [1].
Figure 2
Pipeline Model of a tree:
Diagrammatic representation
showing the progress of tree
growth. Image source: (Shino-
zaki, 1964).
ABSTRACTION IS CONTEXTUAL exploration and discovery of new realities possible.

Abstractions employed in representations are im- The representations we seek in our investigation are
portant tools when communicating contextual as- similarly selective, being particularly based on the
pects of represented objects. Building an abstrac- material context, instead of being quantitatively ac-
tion requires being selective for features that suit curate models.
the purpose of a representation. For example there
are many ways to represent a tree in the physical PLASTER SHAPED BY FORCES ACTING
model of an architectural project. Sometimes it does ON ITS CONTAINER
not even have to look like a tree in order serve the The research presented in this paper started as part
purpose of being a representation of a tree. In the of a graduate design studio where students con-
model shown in Figure 1, the house is surrounded ducted experiments to trace the emergent proper-
by a blurry transparent nature. That nature is rep- ties of various materials. Drawing from one of these
resented with individual trees made of a transpar- experiments, observing the behavior of plaster in
ent material. The house is visible through the trees. elastic molds of party balloons, authors investigate
Transparency is an instance of features one could at- novel ways to represent form’s material causality.
tribute to a tree. It may not be an absolute property
of a tree but in a particular setting how we experi- Hand as a Mold
ence it. Through such an illustration of the tree, its In the preliminary modeling exercise, plaster filled
relevance to the context and the designer’s intent is balloons are individually shaped by hand, to be later
communicated. modeled in the digital environment. This way a di-
Differently, a plant ecologist’s diagrammatic rect interaction with the composite material system
pipe model of the tree form reveals the relations is provided in the physical experiments. Different
between the plant’s growing patterns and the envi- shape transformations are observed as different ac-
ronmental factors (Shinozaki, 1964). This metameric tions of the hands are tried (Figure 3). Digital mod-
conception of the tree form divides it into its sup- els are constructed based on the examination of the
posed longitudinal parts “unit pipes” that have simi- end products of the physical experiments.
lar growth behavior and illustrate mechanical prop- The transformations of the elastic surface are
erties of the plant (Figure 2). digitally modeled (in Rhino with the Grasshopper
In each example key features of the tree are de- plug-in) with a series of straight lines and attractor
fined in relevance to the context and these features points that control the geometry of each line. Lines
are used to build abstract schemas. Both types of define the surface to illustrate topological trans-
models, the scientist’s and the architect’s are en- formations of the elastic material in cross-section.
riched by these types of schemas, in that they make As lines morph due to the position of the attractor
Figure 3
Squeeze, pull, twist, push.
Figure 4
Grasshopper model represent-
ing liquid transformations.
points, the density of the lines locally changes (Fig- forward as it might appear. In order to change the
ure 4). The actions of the hands are to a certain de- shape of the model by moving the points around
gree abstracted in attractor points. The smoothness the scene, first a mathematical description of the
of the movement caused by the pressure of the liq- change needs to be made and then the attractor
uid on the elastic surface is acquired with a cosine points’ relation to the change needs to be defined.
function. The shape of the bump on the surface Furthermore, if the user of the plug-in is not very
could be determined by changing the parameters familiar with the analytical descriptions of shapes
of the function. In this case the key features of the s/he might have difficulty in controlling the shape
transformation process that are used to construct changes. Throughout the design process every
abstract models are smoothness of the liquid move- time the designer changes the model, the model
ment and flexibility of the elastic surface. is reevaluated based on the design objectives. This
Rather than being the exact reproduction of kind of an evaluation comes mostly from intuitive
the physical models, the digital model is meant to aspects of seeing, and as Stiny (2006) suggests “see-
simulate the interaction with the modeled object. In ing and drawing work perfectly without rational (ana-
order to achieve similarity, the manipulation of the lytic) thought”. In computation, analysis is valuable
digital model must to some degree correspond to when coupled with seeing. Hence, our analysis aims
the physical material transformations. In the physi- to sustain the phenomenal aspects in the designer’s
cal environment the designer is able to touch the interaction with the material.
materials that s/he is working with; this is a direct
way to interact with the materials. Commonly used Cellular Interaction
method of attractor points in the digital models is a In subsequent investigations, conducted as group
way to “touch” the models in the digital medium. Still work in the graduate studio, plaster-filled balloons
it happens on a symbolic level, and is not as straight- are put in a rigid mold and their interactions with
Figure 5
Physical models.
each other and the surrounding rigid mold are ob- geometry of the rigid mold and division rules of Vo-
served. Different experiments are held changing the ronoi tool. In Softbody each module is treated as a
parameters, particularly the number of units in the consistent whole with predefined properties. Their
rigid mold and the amount of fluid in one balloon. interaction with each other and the rigid mold is
The circumstances that caused the shape change simulated through the behavior of each module.
are examined in connection with the morphology of Softbody plug-in of 3dsmax is a physics-based
the end products (Figure 5). modeling environment and its interface allows the
Two digital models for the project were gener- user to control the material behavior of the mod-
ated with Grasshopper and Softbody plug-ins of eled objects by changing the parameters like stiff-
Rhino and of 3dsmax respectively. The end results ness, damping, friction and the gravity (Figure 7).
were very similar as shown in Figure 6. Grasshop- Physics-based modeling approaches like these have
per model was generated in a top-down manner by proven to be useful when building lifelike represen-
dividing a whole into its part. In this case the mod- tations of the materials with in the design process.
ules are handled as parts of a whole defined by the Principally a physics-based modeling environment
Figure 6
Digital models: from left
Grasshopper Voronoi model,
3dsmax Softbody, negative
space model.
Figure 7
Interface of the Softbody
plug-in.
Figure 8
Softbody model.
operate on the basis of a simulation algorithm deve- shaped differently. This unpredictable variability in
loped for the physical process it represents and the material transformation is a challenge for represen-
user interacts with the model through visual out- tations that are expected to support it. We propose
puts. Visual schemas play an important role in phys- visual rules to achieve this. Stiny’s (2011) definition
ics-based modeling approaches. For example the of general transformation rules and the unrestricted
Softbody plug-in of 3dsmax simulates the elasticity rules suit the variability in question here. An initial
of objects through the principles of particle phys- shape schema is crucial with parts that could be al-
ics in that the surface of a “softbody” is defined with tered to generate different products of the transfor-
points which are interconnected with hypothetical mation process. Stiny’s (2011) examples of Goethe’s
springs. With the help of this surface abstraction it Urpflanze and Semper’s Urhutte are both archetypal
becomes possible to model the elastic deformation schemas for a class of objects, that are varied and
of materials (Figure 8). each with definite parts. Our question has been how
we can formalize visual schemas for objects with-
VISUAL SCHEMAS OF PHYSICAL PRO- out definite parts such as the plaster filled balloons.
CESSES It is insufficient to observe just the products of the
The models above are attempts at representing transformation for formalization of such a schema.
material properties that impact form. They are pur- An examination of the conditions that bring about
posefully incomplete exercises that serve to analyze the transformation is also necessary (Figure 9, 10).
material properties and to see where digital models The rules are derived by looking at the transfor-
may fall short. As seen above, each plaster unit is mation process, and the relations between proper-
Figure 9
Multiple products of the ‘cel-
lular interaction model’.
Figure 10
Cellular interaction model:
neighboring components.
Figure 11
Different processes leading
to different morphologies: a)
cellular interaction, b) hand as
a mold, c) plaster in a balloon:
fully filled - no deformation,
d) plaster in a balloon: half
filled, e) plaster in a balloon:
half filled.
ties of the components. Key features among the end as shown in Figure 13. Parts are recognizable at the
products of the transformation process are identi- contact areas with the other components and they
fied to construct abstract schemas. These common are either concave or flat (Figure 14).
features, when varied, are what make the products The geometry of the contact areas are deter-
unique incidents. Figure 11 shows the results of the mined by the material properties of the components
different transformation processes for plaster in a in the system. When two plaster-filled balloons come
balloon. It is clearly seen how both materials circum- into contact with one another, the more rigid one
stantially take the shape of each other. For example imposes its shape on the other. The rigidity in this
in the case of half filled balloons shown in figures 11 case is determined by the two factors: the amount of
d and e plaster take the shape of the creases of the liquid in the balloon and the physical state (liquidity)
balloon, however in the ‘hand as the mold’ model of the plaster at the moment of contact. Another fac-
in 11 b with the balloon squeezed liquid plaster tor that specifies the geometry of one object is the
stretches the balloon rushing away from the pres- number of objects that it is in contact with. When we
sure of the hand. mark the differentiating surface parts on the contact
To find out which of their parts make them dis- areas with surface partitioning rule, the polyhedron
tinguishable as the products of different processes, like structure of the remaining parts of the surface
first these parts need to be determined. It can be is revealed (Figure14). The more tightly the plaster-
simply done with Hoffman and Richards’ (1983) filled balloons are packed in a rigid mold, the more
smooth surface partitioning rule (Figure 12). Ac- angular is the appearance of this polyhedron-like
cording to this rule human vision enables recogni- structure. The polyhedron-like structure and surface Figure 12
tion of objects by dividing them into their parts. differentiations at the contact areas are recurring Smooth surface partitioning,
The minima rule states that this partitioning process features in each component, whereas the angular- “Minima Rule: Divide a surface
takes place based on the discontinuities on a sur- ity of this polyhedron-like structure and concavity of into its parts at loci of nega-
face (Hoffman, Richards, 1983). With this method the differentiating surfaces are varied. tive minima of each principal
we divide the surface of the model in Figure 11-a A visual rule illustrates the relation between curvature along its associated
the shape transformations of each plaster object family of curvature.” (Hoffman
and Richards, 1983).
Figure 13
Smooth surface partitioning
of the surface of a “cellular
interaction” model.
Figure 14
Marking the convex and flat
surfaces with smooth surface
partitioning method.
and the way they are packed in a rigid mold (Figure We vary this rule to capture emergent proper-
15). In this rule, the initial shape of the plaster-filled ties of the plaster-formwork interaction. The rules
balloon is represented with a circle plane. The right in Figure 16 display the condition where the outer
side of it shows the transformed shape while the in- rigid mold gets smaller while the number of the
dicator above the arrow gives us information about units in the mold stays the same. The increase in the
the context used in the action. The area of the circle angularity of the resulting shape is visible as the sur-
plane corresponding to the volume of a component rounding units get closer to one another. The rules
stays the same during the transformation process. in Figure 17 show the formation of polygon-like
Figure 15 Colored lines represent the neighboring units. shape of a unit with the increasing number of sur-
Visual Rule 1: Gray circle plane rounding units. It also reveals the relation between
represent the initial shape the number of surrounding units and the number
of the plaster-filled balloon of sides of the polygon. By changing the position
and. Red lines stand for the and the number of the surrounding units, different
surrounding units of a compo- shape computations can show the gradual transfor-
nent in transformation. mation of a unit (Figure 18).
Figure 16
Visual Rule 1 elaborated: An-
gularity of the plaster object
increases as the volume of the
outer rigid mold gets smaller.
Figure 17
Visual Rule 1 elaborated:
Angularity and the number of
the sides of the plaster object
increase as more balloons are
put in the rigid mold.
Figure 18
Shape computations showing
the gradual transformation of
a unit as the surrounding units
The rules presented in Figures 15-18 give a clue 20 the thickness of the line signifies the rigidity of get closer to each other.
about how shapes come about. Nevertheless, it is the elastic mold while the grey tone stands for the
not possible to fully comprehend the process by just hardness of the plaster. These are depictive rules. In
looking at these. The mold of the surrounding piec- search for alternatives that can be more generaliza-
es shapes the plaster-filled balloon. The rules pre- ble, we also develop the visual rules in Figure 21 that
sent the surrounding units as solid shapes, however serve the same purpose but more generally to work
this is not always the case. Rules still need to reveal even for singular objects. They exhibit two different
the interaction of the neighboring objects. cases of being in a mold. Based on the rigidity of the
Further examining the transformation, it is pos- components, which is represented with line thick-
sible to improve the visual rule to contain more in- nesses, their potential to transform one another is
formation on the process. That leads us to a less gen- displayed. Different weights (color and thickness)
eral rule. In Figure 19 the visual rule for the schema x signify properties that undergo transformations.
à x – prt(x) + prt(x)’ is presented. Here the transfor- Shapes are generic and can be interpreted to sub-
mation of the subtracted part of the initial shape is sume others.
displayed with parametric variation rule under gen-
eral transformation rules (Stiny, 2011), for the areas CONCLUSION
of the subtracted and added parts are equal. Current digital modeling environments have the ca-
To further enhance the rules, properties could pacity to provide the designer some form of inter-
be assigned as weights (Stiny, 1992). As the shape action with the model but phenomenal aspects of
transformations are mainly regulated by the rigidity the physical environment often get lost in symbolic
of each component in the system, it is the first mate- reductions. In most cases the designer interacts with
rial aspect to be included in the visual rules. In figure the digital models on a symbolic level and forgoes
Figure 19
Visual rule of the schema xà
x- prt(x)+prt(x)’ .
Figure 20
Weights as tones of gray and
line thicknesses. The darker
the gray tone the harder the
plaster it represents
Figure 21
Weighted shapes representing
the rigidity of the bound-
ary of an object. The rigidity
increases with line thickness.
the causality shaping the design. In the digital mod- based shape transformations can be visualized to
el, the designer is able to perform transformations be compared with one another, to be manipulated
on the model by changing some numeric values if necessary, and to be understood within a broader
within set ranges. In addition to this capacity, there picture of how shapes come about. Differently than
is a need for case-specific visual rules. This is to em- rules, schemas, as defined and categorized by Stiny
body the designer’s unique reasoning which feeds (2011), aid in understanding the rules within formal
from the interaction with the material. The variation categories that might prove helpful in setting up the
of plaster-in-balloon morphologies in Figure 11 il- support system in the digital platforms. Visual rules,
lustrates differences between instances. We study and visual schemas as their more general versions,
a particular hands-on experimentation in order to not only document transformations but also sum-
showcase how visual rules may document the form- marize and help systematize the designer’s percep-
material relation with the aim of supporting the in- tion of founding relations of actions. Visual rules
teraction of the designer in the digital form-finding presented in this paper also utilize weights that can
processes. We have developed exemplary rules and be used to represent magnitudes of certain material
schemas as general and visual as possible based on properties.
parameters derived from hands-on experimenta- Further research requires applying these kinds
tion. There are many parameters that determine the of rules for synthesis, as opposed to for analysis, and
composite behavior of the materials. In this study, in parallel to a design exercise as opposed to a ma-
they add up to two main features: geometry (cur- terial exploration exercise as the one referred to in
vature) and rigidity. The values indicating material this paper. This would help us see how the results
properties of components are employed in the com- correspond to the rich interactions the designer has
putations of shape transformations. in the material world. Additionally, since visual rules
The rules given in this paper are in no way a are specific to case and designer but can be catego-
complete grammar but are directives for phrases rized using more general schemas, it is meaningful
that can belong to a grammar if a designer wishes. to pursue a system to support various visual rules in
These rules are mere instances of how material the digital platforms.
ACKNOWLEDGEMENTS 56-62.
The graduate studio mentioned in the text is Digital Hoffman, D.D., Richards, Whitman 1983, ‘Parts of Recogni-
Architectural Design Studio, a required course in Ar- tion’, Cognition, 18, pp. 65-96.
chitectural Design Computing Graduate Program in Rota, G-C 1997, Indiscrete Thoughts, Birkhauser, Boston, MA.
Istanbul Technical University. The studio was super- Stiny, G 1992, ‘Weights’, Environment and Planning: Planning
vised by Mine Özkar and teaching assistant Ethem and Design, 19, pp. 413-430
Gürer in the academic term of Spring 2012. The Stiny, G 2006, Shape: Talking about Seeing and Doing, MIT
group work that serves as the object of this inves- Press, Cambridge, MA.
tigation was conducted by students Aslı Aydın, Halil Stiny, G 2011, ‘What Rule(s) Should I Use?’, Nexus Network
Sevim, Ersin Özdamar, and Zeynep Akküçük. The Journal, 13, pp. 15-47.
analysis of the experiments with visual schemas are Shinozaki, K, Yoda, K, Hozumi K and Kira T 1964, ‘A quantita-
entirely done subsequent to the studio. tive analysis of plant form—the pipe model theory, I.
Basic analysis’, Japanese Journal of Ecology, 14, 97–105
REFERENCES
Glanville, Ranulph 1998, ‘A (Cybernetic) Musing: Variety and [1] http://archimodels.info
Creativity’, Cybernetics And Human Knowing, 5-3, pp.
Hyperdomes
Non-standard roofing structures, technological evolution and

distinctiveness in urban environment
Andrea Rolando1, Domenico D’Uva2

Politecnico di Milano, Dipartimento di Architettura e Studi Urbani, Milan
1
andrea.rolando@polimi.it, 2doduva@gmail.com
Abstract. The development of new shapes in architecture has deeply influenced the
current perception of the built environment. The analysis of the processes behind this
evolution is, therefore, of great interest. At least two well known factors, influencing this
development, may be pointed out: the great improvement of digital tools and the tendency
toward building distinctiveness.
In particular, the innovation of digital tools such as parametric modeling is resulting
in an overall diffusion of complex shapes, and the phenomenon is also evident in a
clear expressionistic search for architectural singularity, that some might consider as a
negative effect of globalization trends.
Though, if we can consider as a positive result the fact that parameterization allows
a deeper control over design factors in terms of reference to cultural, historical and
physical context, at the same time such control possibilities are sometimes so stark to
be even auto-referential, stepping over site-specific parameterization, to create unusual
shapes just for the sake of complexity.
The ever-growing diffusion of generative design processes is in fact going to transform
niche procedures, frequently limited to temporary decontextualized structures, into an
architectural complexification as an end in itself.
The hypothesis of this paper is to demonstrate that site-specific parametrization can be
considered as a tool able to translate intentions into shape; it is necessary, for this aim,
the widening of the meaning of the word singularity.
Keywords. Urban environment; distinctiveness; non-standard roofing structures.
INTRODUCTION
The need for new shapes in architecture has brought tendency toward the building distinctiveness. The
a great development in techniques and processes aim of this work is to define the which digital tools
able to control and manage the building construc- produce distinctive shapes, by analyzing a set of sig-
tion. It is worth to focus on two factors of this evo- nificant case-studies through times.
lution, the improvement of digital tools and the
Figure 1
Digital tool scheme. Boxes in
blue are the tool families of
interest for this work.
DIGITAL TOOLS
We refer to digital tools as a generic expression to instructions, routines and by running an algorithm
define a giant umbrella of different software, which until the shape which performs better is reached.
are very different in aims and efficiency. It is there- The final shapes are so produced only by a sequence
fore useful framing the parametric tools into a nar- of instructions that produces a result. The designer
rower family of design instruments. The use of para- isn’t the only actor in the shape creation process,
metric tools for design complex shapes creates new because it is paired with the machine results, which
methods, often unexplored, to describe a compre- might go beyond the starting idea. In fact the initial
hensive notion of building performance. shape design, might even be developed into some-
The meaning given to parametric tools is worth thing unpredictable at the start of process. So it is
to be deepened because of relative youngness of essential to focus the attention on the component
this discipline, which lacks of acknowledged notion. that directly modifies the production design, which
There are at least two families of parametric tools is the code.
that are radically different in methodologies and fi- The code writing, as an act of creation, corre-
nalities. The framing applied to this work has been sponding to the designer’s intention, gives complete
schematize in Figure 1. freedom to choose the road to the shape definition.
The first is the Building information Modeling This freedom is partially constrained in control-
(BIM), widely used to optimize building perfor- ling the resulting shape, which may go beyond the
mance with a certain degree of constraint. The limit choice of the preferred shapes. It is so introduced a
of BIM is the creation of new shapes, which are not disruptive innovation in the design process, which
pre-build inside the software. changed deeply the ordinary design method.
The second family, which is of higher interest for The ordinary design process is made of a circu-
this work, is that corresponding to the so called gen- lar correspondence between the mental knowledge
erative design tools. These digital tools that works of the shape and its final representation. In genera-
in strict connection with coding, which embraces tive design, instead, the effort is focused in thinking
an area of knowledge quite far from traditional ar- about the code that will produce the shape, until the
chitectural design procedures. In this case the ar- desired shape is reached. The resulting shape, there-
chitectural form is defined through code, made fore, is generated with an indirect procedure, not by
by declaring variables and constants, by writing direct modeling and editing of shape. In this sense,
two classes of design process drivers may be out- The singularity is intended as a recognizability of
lined, external and internal: site-specific parameters a building in an urban environment. It is considered
and building related parameters. at the same time as internal and external character
of architecture that has to relate to imageability of
USE OF TOOLS – SITE SPECIFIC AND the shape, considering its connection with urban
BUILDING RELATED PARAMETERS environment.
Site-specific parameters are made by the elements Internal singularity is related to the distinctive-
of the urban environment that influence the build- ness of structural and technological performance
ing in its components. The effect of these external of architecture which makes exceptional a building
constraints is evident in some aspect, as the external in itself. External singularity, instead, is the recog-
skin of buildings, but it may influence the structure nizability character of the architecture on a larger
and the functions of the generated spaces. The ap- scale, making it a relevant element of the urban
plication of these specific parameters is important environment. A parallel can be set with the rela-
to provide the building with the correct contextual- tion between internal and external singularity and
ization within the neighboring spaces. It is therefore the aforementioned connection between building-
important to understand the rules that define the related and site-specific parameters, as pointed out
urban environment to better set up parameters that in Figure 2. The strict relation between the tools for
will characterize the building, giving it the character form-finding and the pursued aim creates a disrup-
of distinctiveness. The use of these elements points tion in the process of singularity creation. The lin-
out the importance toward building located in ur- ear process where tools creates the singularity is
ban environments, which are endowed of their own transformed into a design loop where tools create
characters, which cannot be ignored. complexity, and the singularity generates new pa-
In parallel with these elements, collected from rameters to drive the software.
external environment, it seems important to under- Despite the tools limitless shapes creation, their
line the importance a second class of factors, the complex approach and the steep learning curve,
building related parameters. These may be defined keeps away from the use outside academia and top-
as the set of relationships established within the notch designer.
geometric elements of the building skin. This ap- So that most of the buildings created with the
proach works perfectly with art installation, which is generative process are endowed with internal sin-
needful by itself. The aim of this design method is to gularity because they are small-scale architectural
give a complex and appealing perception to build- manufactures, pavilions and temporary installations
ings because for some kind of aesthetical need, a that are designed intentionally ignoring the connec-
lack of intricacy in shape is perceived as a lack by a tion with urban environment.
large part of designers. This need is largely fulfilled This tendency toward singularity wasn’t so defi-
by the use of generative design tools, which easily nite through the times. It is to be underlined, in this
generates an auto-referential complexity. With these sense, the denial of monumentality, in Le Corbusier’s
specifics, it is easily understandable how the gen- architecture.
erative-design tools have been pointed as the next Therefore it has seemed uncompleted conduct-
-generation step in the evolution of design process. ing an analysis of this phenomena, limiting the
analysis to contemporary buildings endowed with
SHAPE DISTINCTIVENESS external and internal singularity, so it was chosen to
The innovation of digital tools, is one of the two driv- consider the domes, which have always been distin-
ers in new shape generation, the other is the ten- guishing elements of verticality emerging in hori-
dency toward distinctiveness. zontally dominated urban environment.
Figure 2
Shape distinctiveness: the case
of the Kunsthaus emerging in
the roofscape of Graz.
FROM DOMES TO HYPERDOMES gies to achieve such aim. The dome issue is shown
The meaning of dome, intended as a “large hemi- in Figure 3 in which Andrea di Bonaiuto, painted the
spherical roof or ceiling” (Merriam Webster diction- Church before Brunelleschi’s design. The depicted
ary) has a deeper significance connected with the its dome is a fake because in 1350 there was no built
function in the past. In fact the spaces too wide to dome, just designs. because of the complexity of the
be covered with normal ceilings, were closed with aim. Therefore Santa Maria del Fiore dome may be
hemispherical roofing structures. One renowned considered as a reference example of non-standard
example of these issues is the cathedral dome of roofing structure clearly emerging in an urban land-
Santa Maria del Fiore in Florence. The base of the scape. Further Case studies for these past domes are
dome was built in 1315 and it remained unfinished the XVII century Sindone dome by Guarino Guarini
until 1436. It took more than 100 year to be finished in Turin (Figure 4 left) and the XIX century San Gaud-
because at that time nobody was able to design a enzio Church dome by Alessandro Antonelli, in No-
cover for such a span of space, until Brunelleschi, vara (Figure 4 right).
in 1418 conceived a series of structural and strate-
Figure 3
Detail in Santa Maria Novella
from “ Spanish chapel” 1350.
Santa Maria del Fiore is
depicted with a fake dome
because it wasn’t possible at
that age to build a real one.
HYPERDOMES creates singularities in urban skylines.

The domes as a symbol of this phenomenon have As traditional domes produced a break in the
been transformed through times into a more com- skyline of cities in the past, so hyperdomes make a
plex form of architecture, rather similar to a singular rupture in actual urban context. The singularity that
roofing structure than to a classical structural ele- characterized domes of the past is not limited to ur-
ment. In this work the non-standard roofing struc- ban environment, because they represented both
tures (as architectural elements to be considered technological excellence and structural innovation
in the broader sense) are acknowledged as key ele- at their time. We could also refer to more examples,
ments to select specific study cases, where the high- such as the Pantheon in Rome, S. Sofia in Istanbul,
er level of complexity of non-standard structures the Antonelli Mole in Turin, the works of Boullée or
can create points of singularity within the context. Speer, Nervi and Buckminster Fueller in recent times,
When it came the need to define these special but we try to limit our attention to some cases of spe-
non-standard domes, it was necessary to specify a cific relationship between the digital innovation pro-
word for structural elements which were a compro- duced by design tools and its translation into distinct
mise between geometrically defined domes and shapes. Similarly, new generations of contemporary
mesh structures. Therefore the term “hyperdome” “domes”, hyperdomes are landmarks in the skyline,
will be used in this work to widen the meaning of because of their shape and structural singularity,
domes including all the roofing structures which so they get imageability (in the meaning attributed
Figure 4
Churches of XVII century
Sindone dome by Guarino
Guarini in Turin and XIX cen-
tury San Gaudenzio Church
dome by Alessandro Antonelli,
in Novara, relevant prototypes
of hyperdomes clearly mark-
ing the urban and, for the lat-
ter, even regional landscapes.
to this term by Kevin Lynch (1960), as a “quality in a hall in Kamigahara (Figure 11), by Toyo Ito and the
physical object which gives it a high probability of Lingotto dome by Renzo Piano (Figure 12).
evoking a strong image in any given observer”).
In this sense, that might seem not only chal- CONCLUSIONS
lenging but even provocatory, some case studies This study has analyzed the aforesaid series of case
for contemporary structures are the Future Systems’ studies, pointing out how the new relationships be-
Selfridges building in Birmingham (Figure 5), the tween design tools, structural conception, shape in-
Kunsthaus in Graz by Peter Cook and Colin Fournier novation, contextual references and symbolic values
(Figure 2), the Opera House in Lyon by Jean Nouvel become key factors to understand the evolution of
(Figure 6), the Greater London Authority building hyperdomes.
(Figure 7), British Museum Great Court in London Starting from the given hypothesis this paper
(Figure 8) and Reichstag in Berlin (Figure 9), all by has shown a possible interpretation of the current
Norman Foster, the recent roofing structures in Gent interpretation of domes and how both the internal
and Taiwan (Figure 10) and Meiso no Mori funeral and external singularity may be considered for giv-
Figure 5
Selfridges building in Birming-
ham - Future Systems. The
hyperdome creates a singular-
ity by integrating itself in the
urban environment though
being a complex shape.
ing the building a shape distinctiveness in the urban the phenomenon of the complex shapes in terms of
context. A positive or negative assessment of the relationship to the urban context, without involving
role of hyperdomes goes beyond the aim of this pa- aesthetic and historical issues that deserve further
per that mainly aims at recognizing and interpretate and specific disciplinary attention. Nevertheless, it
Figure 6
Opera Nationale de Lyon –
Jean Nouvel. A contemporary
dome, which creates a singu-
larity in urban environment.
Figure 7
Greater London Authorith
builing - Norman Foster.
Geometric singularity by the
discovery of the only rotation
angle that creates circular sec-
tion from a elliptical ellipsoid.
Figure 8
Queen Elizabeth II Great Court
at British Museum, Norman
Foster. Structural singularity.
Effects of compression and
bending must pass through
the nodes in all directions,
decreasing bear loading
of central building. Green
performance is achieved
through the glass perceived as
clear, which is shielding 75%
of ultraviolet rays.
Figure 9
Reichstag dome - Norman
Foster. Internal geometric
singularity. Ramp as a spiral
inscribed in the circumference
( Loxodrome).
seems possible to anticipate that the sake for search- through a further long term process of historical, cul-
ing the shape singularity as an end in itself, that tural and even social interpretation and acceptance.
many recognize as a common issue in contemporary
architectonic structures, it seems to be necessary, REFERENCES
but not sufficient, to mark the urban environment Lynch, K., 1960, The Image of the City, MIT Press, Cambridge
with significant permanent signs that need to go MA
Figure 10
Taichung Metropolitan Opera
House - Toyo Ito. Singularity
in flux allowed by the walls
which bends to merge with
floors and ceilings.
Figure 11
Meiso no Mori Crematorium
- Toyo ito. A generative design
applies the mechanichal thory
that minimizes strain energy
in a structure to create a ra-
tional free-cureved surface.
Figure 12
The organic shape of the
“Bolla” (Bubble) designed by
Renzo Piano on the roof of the
Fiat Lingotto Factor in Turin.
Action Based Approach to Archaeological Reconstruction
Projects: Case of the Karnak Temple in Egypt
Anis Semlali1, Temy Tidafi2, Claude Parisel3
1
École Supérieure des Sciences et Technologies du Design, Tunisia, 2,3University of Mon-
treal, Canada.
2,3
http://www.grcao.umontreal.ca
1
anis.semlali@gmail.com, 2temy.tidafi@umontreal.ca, 3cparisel@videotron.ca
Abstract. The proposed paper deals with a numerical approach that could better assist
the archaeologist in the archaeological reconstruction projects. The goal of our research
is to explore and study the use of computerized tools in archaeological reconstruction
projects of monumental architecture in order to propose new ways in which such
technology can be used.
Keywords. Architectural heritage; archaeological reconstruction; action-based
modeling; architecture and complexity.
INTRODUCTION
The definition and development of new modeling The case study: Karnak Temple
methods is the objective of a research project in To do this, our project uses as laboratory the Kar-
progress at the CAD research group (GRCAO) of Uni- nak temples in Egypt: certain information is already
versité de Montréal. These methods aim for a better available in the form of plans, surveys, elevations
integration of the varying types of knowledge impli- and sections of existing monuments (with or with-
cated in the reconstitution of ancient architectural out proposed restitutions), and excavation reports,
structures, as well as greater flexibility in the manip- while other information is still to be surveyed on
ulation and utilization of this knowledge. To reach site, or catalogued. Beyond the technical aspects
this objective, technology will not suffice. It is nec- that allow for the precise encoding of the basic com-
essary to integrate methods, knowledge and goals ponents of constructions and structures, the meth-
of a collection of scientific disciplines that are not od allows for the elaboration of a reconstitution
used to working together (without forgetting the that notes the different proposed reconstitutions of
inherent incoherencies): to social sciences as with parts that are either currently missing or have been
archeology (in the classical and not the anthropo- modified several times over a millennium of history.
logical sense of the term), history, art history, epigra- Also the method takes into account the degree of
phy and chronology, architecture, geometry, optics, probability of the proposed reconstitutions.
and information technology must be joined. This To experiment our general approach of restitu-
requires that each discipline define itself in terms tion, we choose the case study of the VIIth pylon in
of what it can bring to the reconstitution of physical the Karnak Temple. The choice of this case study fol-
objects in an environment, and thus, the reconstitu- lows directly from the data availability. Indeed, the
tion of architectural heritage. seventh pylon was a pretext for testing survey func-
because the inscriptions are deteriorating at great Figure 1
speed and there is a real risk of losing completely The VIIth pylon divided into as
some impotents scenes. many blocks as it contains.
The main problem at the present time is that
the traditional methods carried out to survey the
inscriptions are very time-consuming. For example
the most common of these methods consists in
making facsimiles of the wall to be surveyed, with
photographs as background or simply with trans-
parent sheets placed against the surface of the wall.
This method involves numerous checks during the
drawing process and is therefore rather tedious, be-
cause it requires the collaboration between different
drawers.
Research carried out by the GRCAO leads to
present method of computerized epigraphic survey
tions in the project “Karnak-1” in GRCAO. We made a that can be used for drawing and recording the hier-
division of the complete structure of our case study oglyphic signs for all planar, but also conical and cy-
(the VIIth pylon in the Karnak Temple) in as many lindrical, architectural elements of Egyptian temples.
blocks as it contains (Figure 1). Our goal is to assign This method is user-friendly for archaeologists and
to each of the blocks from corpus, its place in the epigraphists alike, thanks to the very detailed menus
general scheme of the studied structure. It is a set of created in the AutoCAD© software. Numerous choic-
epigraphied blocks with variable dimension. es are constantly available during the surveying
When the corpus of blocks to be processed is process, and every operation can be undone if nec-
very important, it is necessary to find the necessary essary. Each surveyed sign is recorded in a database,
resources to divide the whole into “manipulable” in the form of a text file, which can later be used for
units through a multitude of actions which identify other research purposes: studies on the shapes of hi-
the blocks that have one or more common charac- eroglyphs, automatic translation of the texts, search
teristics. The data can be used to identify the relative for missing elements, etc. This method considers the
position of a unit with respect to another. Based on needs of the epigraphists and offers them the pos-
the geometric attributes, iconographic or other, the sibility of controlling various operations during the
goal is to identify, manipulate and / or to connect computerized survey process. Particular empha-
and recreate these attributes in order to find indices sis has been put on the fact that the decoration of
that will help us to argue one or more assumptions a monument is indissociable from its architectural
about the hypothetic position of a block in relation support. The drawings must be recorded with all
to the adjacent ones, respecting of course the over- the information necessary to understand their real
all assembling of the general unit. meaning (i.e. the architectural and archaeological
context). The recording format has been normalized
The survey and description of blocks so as to be exploitable for research purposes (statis-
Carrying out epigraphic surveys is a very important tics, restoration of structures, etc.) (Figure 2).
task in archaeology, particularly in Egyptology be- Moreover, various exploitations (reconstitution,
cause all the monuments contain numerous texts paleography, etc.) are possible, thanks to the fact
and scenes engraved on their architectural ele- that all the signs drawn are recorded in a universal
ments. It is a matter of urgency to do such surveys, format. The publication of the texts can still be made
Figure 2 archaeologists an effective utilization of the pos-
Survey of the blocks in the sibilities offered by computer-assisted tools. This
VIIth pylon using the GRCAO study allowed us to demonstrate the complex and
method. systemic nature of using ICT in the field of archaeo-
logical reconstruction. The multiple actors, condi-
tions, means and goals considered in archaeological
reconstruction projects have led us to explore a new
approach that reflects this complexity.
in paper form, but can now be in numerical form Study of the publications in the archaeo-
too, which in turn leads to other possibilities such logical projects
as data exchange. This approach is of course adapt- In order to achieve the goal of our research, it was
able to the survey of other types of temples (Greek necessary to further study the nature of the archae-
for example). ological process. This involved understanding the
links and interrelations between the various compo-
DEVELOPMENT AND VALIDATION OF nents that defines the archaeological approach and
OUR ARCHAEOLOGICAL RECONSTITU- the various thought processes involved in archaeo-
TION MODEL logical reconstruction projects.
This part presents an exploratory prototype deve- In summary, archaeologists perceive and de-
loped to assist (but not to control) the reasoning scribe their approaches through filters determined
and decision-making in the formulation and com- by their use of these descriptions. Any scientific de-
puter simulation of architectural reconstruction hy- scription is both the result of past constructions, and
potheses. This “assistance” will be taking advantage the source of present and future constructs to enrich
of the knowledge and data available or extrapolated them or replace them. These filters can, in many cas-
by the production of computer models as well as es, push the archaeologists to become very attached
alphanumeric documents resulting from targeted to their hypothesis and persist in not recognizing
questions of the databases. their weaknesses.
From this perspective, archaeological publica-
The Use of ICT in archaeological recon- tions all look the same slightly “there can not describe
struction projects a monument without referring implicitly to the state
In our quest to answer this question, we begin with of knowledge and research objectives that determine
a study of the different restitution approaches used proper method the substance and form of the de-
in various phases of archaeological reconstruction scription, so that a catalog, especially when the terms
projects. This involves understanding how the differ- “rational” is a theoretical construct in the same way if
ent methods of approach have evolved (epistemo- not to the same extent that any historical essay “ (Gar-
logically), how those involved in such projects have din, 1979). This study showed a direct relationship
put information and communication technologies between the subjective nature of the process and
(ICT) to use in the field of built heritage. This study the diversity of approaches and thought processes
has identified two main avenues: one whose aim is which can be implemented.
the “representation” of project results and another This exploratory and propositional research
whose aim is to model this process in order to as- reinforces the systemic and complex nature of our
sist the archaeologist through various phases of a approach and prompts us to explore, in practice
project. It is the second approach that can better and through published literature, the elements of
respond to our goals and that can guarantee to the known reality. The study of archaeological reason-
ing through academic publications has allowed us Figure 3
to propose an initial typology of arguments studied. Operating principle to search
Each of these typologies reflects a methodological for blocks assembly using
approach based on organized actions that can be the elliptical mathematical
recorded in a set of reasoning modules. The clas- method.
sification of the various arguments by type of rea-
soning in order to determine the configuration of a
building has enabled us to establish a model of the
various components of the archaeological process
as well as validation rules that have been used by ar-
chaeologist in real reconstruction projects.
This research has allowed us to highlight phe-
nomena and observed processes, leading to a
model representing interrelationships and interac-
tions as well as the specific results of these complex
interconnections. This pattern reflects a cyclical pro-
cess of trial and error, in which the actors consecu-
tively ‘experience’ (according to the project’s goals
and through reasoning modules), several answers
to the questions exposed to him under the corpus
definition, description, structure, interpretation and
validation of the results until the latter would appear
to meet the original targets. Three examples of rea-
soning modules have been developed and tested
through a case study of the VIIth pylon of the Karnak up of lines and surfaces defining the boundaries of
temple in Egypt. those faces of points and defining the ends of these
lines. These data are essential for encoding neigh-
Geometric approach to restitution: Ex- borly relations between blocks because they are the
ample of a module using the geometric main reliable parameters of adjacencies. Each block
reconstruction of 2D objects is individually registered using the survey method
Considering the large number of blocks that archae- adopted (GRCAO method), in two ways:
ologists handle in archaeological reconstruction • The outline of the blocks: this corresponds to
project, it will be extremely difficult for them to visu- the detailed record of the actual boundaries of
ally identify formal and geometric complementari- the block witch will be saved as control points.
ties among the studied blocks. The main goal of this • The min-max block: this corresponds to the
module is to present a reasoning tool to search for polygon including the useful surface of the
possible complementarities among the geometric studied block. This contour is stored as control
characteristics of the identified blocks. It can assist points coordinates.
archaeologists to identify, among the huge mass of Although this reasoning module is based on
available data, a manipulable subsets based on their complex mathematical models, the user will not,
geometric characteristics (Figure 3). This reasoning in any case to manipulate them. All calculations
module may bring, in this case, a considerable as- will be back plans and the user will have to han-
sistance. dle only ‘objects’, which he is used to deal with.
Stone, the basic component of a wall, is made We have demonstrated, through the explora-
tion of a 2D topological help reclaim the blocks, This will be our first approach.
the relevance of such an approach and the rela- Another very important consideration in the
tive facility on the computer translation of the overall structure and composition of a scene ele-
actions that may include the reasoning module. ment: the “continuity of the theme.” Indeed, each
This module, and meeting our original goals, open to of the blocks that make up the structure includes a
the implementation of other actions (respecting the portion of the overall scene. Our second approach
same logic). Depending on the initial objectives and will be to consider a scene as an assemblage of icon-
methods that archaeologists adopt, a specialized ographic elements. These elements, taken together,
team that will be responsible for translating actions could possibly give a meaning to the whole (regis-
that he wishes to undertake and thus to optimize the ter, wall, room, etc.). It is therefore to study the type
contribution that can bring computer tools in the of continuity and propose a reasoning module that
success of the architectural reconstruction projects. can assist archaeologists in this type of work.
Iconographic approach to restitution First approach: Metrological aspects in the

More than any other civilization, the ancient Egyp- Egyptian iconography
tians have associated iconography and epigraphy Our introduction to this metric aspects of Egyp-
to architecture. Indeed, the temple walls are cov- tian iconography is based primarily on the work of
ered with inscriptions and bas-reliefs whose theme, Robins et al. (1994), Donovan (1986) and Carlotti
which meets to a large extent to known conven- (1995). Egyptian architects built their monuments
tions, is the basis for the restitution of significant ar- by following rules that respected the standards of
chitectural parts of the temple. proportion generally defined by type of building
Schwaller de Lubicz (1999) introduced a tech- space, whether courses, pylons or columns (Car-
nique that allows the study of the key of reading lotti, 1995). Restoring a monument according to
scenes through the disposal of their constitutive length units commonly used in Pharaonic Egypt,
iconographic elements. It is to identify the different using sometimes the digital measurement system
modes of expression of what he defined as “an icon- (for architecture) uncial (the decor on the walls),
ographic language”, that when combined together, proves a worthy track looking to be explored.
can give an early sense of the studied scenes. The Iconographic representations on the walls of tem-
three modes of expression used by this language are: ples do not escape this rule and follow a so-called
• Figurative language: it is a language that can “The Last Tile” technique based on the canons of
cover all the iconographic elements that be- proportions. Indeed, measurements of different
long to different known elements and their as- parts of an Egyptian artwork, especially the wall dec-
sociated scenes. orations, were set one with respect to the other in
• The processes of arrangement: Addition and a precise and rigorous way. This “Canon” was based
supervision of iconographic elements. on the proportions of the human body. Modern re-
• Figures of style: It is to identify the language construction of this canon of proportions is mainly
of the image through reading and decrypting based on the study of the guidelines partially pre-
formal and subtle arrangements that the Egyp- served on the walls and statues. The data, the units
tians invented. of measurement as well as the rules of their use by
These complex arrangements are governed by Egyptian artists are still controversial.
mathematical principles that affect the structure of For our case study, we have based our approach
the scene. One approach will be to take advantage on the work of Carlotti (1995) to establish the pro-
of these rules of composition through the imple- portion module used in the VIIth pylon at Karnak.
mentation of the correspondent reasoning module. Depending on the construction period, we may
Figure 4
Restituation of the VIIth
pylone scene using the icono-
graphic module.
have an approximate value of the module used in goal is to study their type and their variants. Our
the scene that appears on the “Medinet Habu” py- analysis allowed us to identify eight types of conti-
lon. The iconographic analysis of the pylon reveals nuities between the studied blocks :
a traditional theme that represents the scene of the • linear continuity
“massacre of the enemies”. This scene was repeated • Iconographic continuity: human body, hiero-
in several Pharaonic structures. In an approach for a glyphic sign, other
restitution by completion, and based on the com- • relief continuity: Level Difference, Surface Un-
plete scene on the temple pylon “Medinet Habu”, coupling
our goal is to determine the missing elements, and • Continuity of the type of engraving: texture
so complete the scene studied (Figure 4). This treat- continuity, etc,
ment was made in four steps: • Text continuity
• Step 1: Determination of the modulus value • Geometric continuity: same Min-max, etc.
proportion of the VIIth pylon at the Karnak • Theme continuity: Text, cartouche, human, etc
temple, • Zone continuity: horizontal or vertical text, etc.
• Step 2: survey of the representation of the
Pharaon on “Medinet Habu temple” and on the
VIIth pylon with the GRCAO method,
• Step 3: Making the superposition of the 2 sur-
veys after practicing the technique of “tiles
making”,
• Step 4: Completing the missing part of the
scene (Figure 5).
Second approach: Typology and icono- Figure 5

graphic connections Restitution of the toponimic
To study the iconographic continuity, we conducted list of the VIIth pylon according
an analysis of vicinity of each of the connections be- to the similar text engraved on
tween the blocks that make up the VIIth pylon. Our the VIth pylon.
Figure 6
Blocks assembled using the
typology and iconographic
connections.
Other elements of description can be added CONCLUSION

according to the competence of the actor and the This experiment allowed us to confirm the overall
degree of precision considered. Actors must not to appearance and total interdependence between the
describe the block as an overall unit, but exploded different components and activities of the archaeo-
into as many constituent elements. That is to say, it logical approach in the archaeological restoration
must offer the user the opportunity to indicate areas projects. On one hand, the procedure used to iden-
of vertical text, horizontal text areas, friezes, human tify and describe the artifacts has a direct impact on
body, etc. A module based reasoning approach can the possibilities of data operations implemented.
provide the archaeologist additional tools to search, On the other hand, the goals of the actors and the
for example, all the text boxes with a width x or an nature of reasoning implemented determine the
additional sign truncated, etc.. This module is based types of data to be recorded and the degree of inter-
on features to combine iconographic themes geo- pretation to be made.
metric aspects of signs and symbols. It is a tool to The results show that the reasoning modules
assist players in the field in studies of possible con- offer an interesting solution to assist archaeologists
nections between the blocks to reconstruct. in theses projects. The multiple action combina-
The general topology of the block is defined in tions offered by theses modules are an advantage
a data structure based on the control points where to many approaches and thought processes which
we can find the points that define the edges of the could be useful to such projects while maintaining
block which themselves define the surveyed sur- the progressive nature of the overall system.
face. The direction of the view can also be defined The type of goals of our project has greatly in-
(based on control points, too) in order to give mean- fluenced the course of all operations related to the
ing to the “left” expression, “right”, “up” and “down” choice of corpus, to their description, to their struc-
(eg the scene). All objects must be observed on the ture, their interpretation and the validation of the
same side. results we had. Indeed, the survey methods and
Attributes faces statements therefore represent data structures have been chosen according to the
qualitative information such as their orientation in nature of our corpus (pictures recovered) as well as
the scene, exposure direction (smooth and treated “inputs” and “outputs” on the reasoning modules
when it is visible to spectators and untreated when that we developed . The choice of data, the way to
it is not visible). It can also be determined by the see them, how to classify them and how to com-
inscriptions and it carries this on two levels: iconog- ment them were determined with reference to the
raphy (the figures, ornaments and inscriptions) and purpose of our construction: restoring the blocks of
semantics (the interpretation and understanding of the VIIth pylon (Figure 7).
the iconographic elements) (Figure 6). Our model expresses not the way has been built
construction, but its logical architecture, once com- Figure 7
pleted. Indeed, the results presented, through our Model of archaeological
case study, have shown that the approach is often reconstruction using the
an iterative process that is constantly progressing reasoning modules.
(by trial and error) through the manipulation of
data by actions encapsulated in various modules of
reasoning (order of epigraphic (text and phonetic),
constructive order, physical or geometric, etc..). In
this progression, we have “experienced” various hy-
potheses through the application or implementa-
tion of new reasoning modules reasoning. The find-
ing of inadequacy determined each iteration and
pushed us back to the data and the means available
to a new definition of corpus, description, structure
or interpretation. It was mainly to combine several
reasonings to reduce the number of available pos-
sibilities and progress until the results can meet the
objectives of the study. The evolutionary aspect of
the system allows us to add other modules of rea- Gardin JC 1979, Une archéologie théorique, Hachette, Paris.
soning if the resources available cannot enable the French adaptation of the original edition: Archaeologi-
objectives of the actors. cal Constructs: an Aspect of Archaeological Theory, Cam-
bridge University, Cambridge, UK.
REFERENCES Robins, G, Gardin, JC, Guillome, O, Herman, PO, Hesnard, La-
Carlotti, JF 1995, Contribution à l’étude métrologique de grange, MS, et al. 1994, Proportion and style in Ancient
quelques monuments du temple d’Amon-Rê à Karnak, Egyptian art, Austin-University of Texas, pp. 62-85.
Cahier de Karnak, Paris 10: 65-125. Schwaller de Lubicz, RA 1999, Le temple de l’homme : Apet
du Sud à Louqsor, Dervy, Paris.
Models of Computation:
Human Factors
Models of Computation: Human Factors - Volume 2 - Computation and Performance - eCAADe 31 | 333
334 | eCAADe 31 - Computation and Performance - Volume 2 - Models of Computation: Human Factors
Fusion of Perceptions in Architectural Design
Ozer Ciftcioglu1, Michael S. Bittermann2
Delft University of Technology, The Netherlands
http://bk.tudelft.nl/en/research/research-projects/computational-intelligent-design
1
o.ciftcioglu@tudelft.nl, 2m.s.bittermann@tudelft.nl
Abstract. A method for fusion of perceptions is presented. It is based on probabilistic

treatment of perception, where perception quantifies the chance an unbiased observer
sees an environmental object, and the associated probability can be interpreted as
degree of awareness for the object. The approach uniquely accounts for the fact that final
realization or remembrance of a scene in the brain may be absent or elusive, so that it is
subject to probabilistic considerations. For objects that are to be perceived from multiple
viewpoints, such as a sculpture in a museum, or a building in its urban context, the
probabilistic approach uniquely defines the fusion of perceptions. This is accomplished by
carrying out the probabilistic union of events. The computation is presented together with
its geometric implications, which become rather intricate for multiple observers, whereas
the computation is straight forward. The method is exemplified for two applications in
architectural design at different scales, namely interior and urban design, indicating the
generic nature as well as the large application potential of the method.
Keywords. Perception; vision modeling; architectural design; evolutionary search.
INTRODUCTION
Perception, and in particular visual perception, is Gelade, 1980; O’Regan et al., 2000; Treisman, 2006),
an interdisciplinary concept taking an important so that gaining insight into the nature of human
place in many diverse applications. These range perception from the experiments remains minimal.
from design of objects and spaces, for which per- However, considering that the perception phe-
ceptual qualities are aimed (Bittermann and Ciftcio- nomenon is due to brain processing of retinal pho-
glu, 2008), to robotics where a robot moves based ton-reception, it should be clearly noted that the
on perception (Ciftcioglu et al., 2006a; Bülthoff et phenomenon is highly complex. That is, the same
al., 2007). However, although visual perception has experimenter may have different perceptions of the
been subject to scientific study for over a century, same environment at different times, depending on
e.g. see Wertheim (1894), it is interesting to note that the complexity of the environment, psychological
it remained mysterious what perception precisely is state, personal preferences and so on, not to men-
about, while it eluded mathematical modeling until tion different vantage points. Due to the complexity
very recently. Many approaches to perception, in of the brain processes and diversity of environments
particular in the domain of psychology and neuro- subject to visual perception, the empiric approaches
science, are based on experiment, while underlying to perception yielded merely rudimentary under-
theoretical models or hypotheses are either sim- standing of what perception is. Although some ver-
plistic, ambiguous or even absent (Treisman and bal definitions of the concept are presented in the
literature, e.g. (Gibson, 1986; Palmer, 1999; Foster, to probabilistic considerations. This approach has
2000; Smith, 2001) due to excessive ambiguity of the been described and its validity demonstrated (Ciftci-
linguistic expressions, these are not to be converted oglu et al., 2006b; Bittermann and Ciftcioglu, 2008).
to precise or even more unambiguous mathematical This probabilistic approach is unique in the
expressions. sense that the perception refers to human percep-
Computational approaches addressing some tion. In the field of computer vision perception is
perception aspects have been proposed by Marr considered to be a mere image processing and en-
(Marr, 1982) whose prescription is to build computa- suing pattern recognition process, where Bayesian
tional theories for perceptual problems before mod- methods are appropriate (Knill et al., 2008; Knill and
eling the processes which implement the theories. Richards, 2008; Yuille and Bulthoff, 2008). Bayesian
Explicitly, different visual cues are computed in sep- approach is to characterize the information about
arate modules and thereafter only weakly interact the world contained in an image as a probability dis-
with each other, where each module separately es- tribution which characterizes the relative likelihoods
timates scene properties, such as depth and surface of a viewed scene being in different states, given
orientation, and then the results are combined in the available image data. The conditional probabil-
some way. These works can be termed as image pro- ity distribution is determined in part by the image
cessing based approaches, and they are determinis- formation process, including the nature of the noise
tic in nature, starting from simulation of retinal data added in the image coding process, and in part by
acquisition. The retinal photon-reception certainly is the statistical structure of the world. The Bayes’s rule
the first stage in the time sequence of the process- provides the mechanism for combining these two
ing in the visual system, and it might be dealt with factors into a final calculation of the posterior distri-
by means of an image specified as a two-dimension- bution. This approach is based on Bayes formula
al matrix. However, the ensuing neural processes p (i | s ) p ( s )
p( s | i) =
are highly complex, so that retinal image does not p (i ) (1)
imply that all the information in the scene is regis- Here s represents the visual scene, the shape and
tered in the human brain and remembered shortly location of the viewed objects, and i represents the
afterwards. Only part of the visual information is re- retinal image. p(i|s) is the likelihood function for
membered. For instance, it is a common experience the scene and it specifies the probability of obtain-
that when we look at a scene, we are not aware of ing image i from a given scene s. p(s) is the prior
the existence of all objects the scene comprises. This distribution which specifies the relative probability
is easily verified for scenes where the number of ob- of different scenes occurring in the world, and for-
jects exceeds about seven objects. mally expresses the prior assumptions about the
In this work a probabilistic approach is adopted scene structure including the geometry, the light-
for perception, where perception is considered a ing and the material properties. p(i) can be derived
whole process from the stimulus coming from the from p(i|s) and p(s) by elementary probability theory.
scene to mental realization in the brain. In other Namely _ _
words, all complex processes, e.g. image formation p (i ) = p (i | s ) p ( s ) + p (i | s ) p ( s ) (2)
on the retina, processes in the visual cortex in the so that (1) becomes
brain, and final realization of ‘seeing’ is modeled as p (i | s ) p ( s )
a single probabilistic event, where ‘seeing’ in that p( s | i) = _ _
probabilistic description is considered to be percep- p (i | s ) p ( s ) + p (i | s ) p ( s ) (3)
tion, where remembrance is a matter of probability. The posterior distribution p(s|i) is a function giving
The final realization or remembrance of the scene in the probability of the scene being s if the observed
the brain may be absent or elusive, which is subject image is i. Bayesian approach is appropriate for
Figure 1
Plan view of the basic geo-
metric situation of perception;
P represents an observer’s
point, viewing an object (a);
probability density function
characterizing perception
along y direction for lo=2 (b).
(a) (b)
computer vision, because for human p(i|s) is almost described elsewhere (Ciftcioglu et al., 2006b;
clearly known, that is p(i|s)=1. Consequently, p(i|s)=0 Bittermann and Ciftcioglu, 2008) and briefly men-
and from equation (3) tioned as follows. We consider a basic geometric
situation as shown in Figure 1a. For a visual scope
-p/4£q£p/4 the probability density characterizing
(4) perception along the y-direction is shown in Figure
which is independent of the probabilistic uncer- 1b for lo=2 and given by
tainties about the scene. This means, as the p(i|s)
is definitive for human recognizing a scene, p(s|i) (5)
is also definitive, being independent of p(s) which The probability density with respect to q is given by
is the prior assumptions about the scene structure fq(q)=1/qS , where qS=π/2. The one-dimensional per-
including the geometry, the lighting and the mate- ception of an object spanning from arbitrary object
rial properties. The effectiveness of Bayes for ma- boundaries a and b on the y-axis is obtained by
chine vision is due to its recursive form, providing
improved estimation as the incoming information is (6)
sustained. yielding perception as an event being subject to
The organization of the paper is as follows. In probabilistic computation. For the case of percep-
the modeling human perception section a vision tion of an object by a single human observer the
model is established. In the perception from multiple computation is accomplished always by (6) when
viewing positions section, the fusion of perceptions the projection of the object is considered as one-
from multiple viewpoints is derived. In the section dimensional along a line. The same computation
experiments, two experiments demonstrating the can be valid for three-dimensional objects, provided
fusion of perceptions in architectural design are pre- we consider the projection of the object on a plane.
sented, and this section is followed by conclusions. In this case, the same formulation can be used twice
for each respective orthogonal dimension of the
MODELLING HUMAN PERCEPTION plane in the form of product of the two probability
In the human perception an object is visually densities integrated over the projected area on the
seen, but its remembrance is subject to some plane.
degree via probabilistic considerations. This is
Figure 2
Perception events E1, E2, and
E3 respectively denoting
perception of an object from
three viewpoints VP1, VP2,
VP3; the union of the events is
indicated by the white dashed
line (a); Venn diagram cor-
responding to the perception
events in Figure 2a (b).
(a) (b)
PERCEPTION FROM MULTIPLE VIEWING conspicuous, in order to determine for instance the
POSITIONS building entrance that should preferably be posi-
In many occasions an object is subject to perception tioned. Consequently it will be easily noticed.
from multiple viewing positions, either by the same A scene subject to investigation as exemplary
observer or by multiple observers. That is, percep- case is shown in Figure 2a, with the three perception
tions from different viewing positions are subject to events E1, E2, and E3. The figure shows a plan view
fusion. As the perception is expressed in probabilis- of the space and the location of an object subject
tic terms, the union of different perception events is to perception assessment and optimal positioning.
subject to probabilistic computation. Requirements The object is subject to perception from the three
with respect to perception from multiple viewing viewing positions VP1, VP2, and VP3, where it re-
positions can occur in many practical applications. spectively subtends the angle domains θ1, θ2, and
To demonstrate fusion of perceptions we restrict the θ3 as seen in the figure. The dashed lines in the fig-
study to two basic examples. It is noted that they ure indicate the boundaries of the observer’s visual
may not be important depending on the particular scope at the respective viewing positions spanning
design problem; however the examples are simple the angles θS1, θS2, and θS3. Figure 2b shows a Venn
in order to clearly explain the method. The same diagram corresponding to the perception situa-
method can be applied in more complex tasks, such tion in Figure 2a. In the case of perceiving an object
as courtroom design (Bhatt et al., 2011), auditorium from several viewing positions this corresponds to
design, office design, as well as urban design. In the the probabilistic union of the perceptions, which
first case study we consider an exhibition gallery en- is obtained by P(E1ÈE2ÈE3)=P(E1)+P(E2)+P(E3)-
vironment, where there are several entrances to a P(E1ÇE2)-P(E1ÇE3)-P(E2ÇE3)+P(E1ÇE2ÇE3), as this
gallery space, and we are wondering what the best is seen from Figure 2b. It is noted that the events
position to place an object is, so that perception of P(E1), P(E2), and P(E3) are independent. In the three
the object is maximized. In the second application dimensional perception case θ1, θ2, and θ3 become
we are considering an urban environment, where solid angles Ω1, Ω2, and Ω3 and the scopes θS1, θS2, and
a building will be erected that will be seen from a θS3 become solid angles ΩS1, ΩS2, and ΩS3.
number of prominent viewing positions. We are in-
terested to obtain the perception of the different EXPERIMENTS
parts of the future building as fusion of perceptions Computer experiments are carried out, where P(E1),
from these viewing positions. In the latter case study P(E2), and P(E3) are obtained by probabilistic ray
this is to identify which part of the building is most tracing, so that a three-dimensional object is sub-
ject to perception measurement without need for which is termed fitness. In the algorithm population
projection to a plane as shown in Figure 1a. That is, members with a comparatively high fitness will be
the solid perception angle Ω subtended by the ob- favored over solutions with low fitness, by giving
ject, as well as the solid angle ΩS, which defines the the former a higher chance to remain in the popu-
observer’s visual scope, are simulated by vision rays lation and to produce new solutions by combining
that are sent in random directions within the three fit solutions. The combination among solutions is re-
dimensional visual scope. The randomness in terms ferred to as crossover operation, and it is carried out
of unit Ω is characterized by fΩ(Ω)=1/ΩS, conform- among pairs of population members referred to as
ing to the uniform pdf fq(q)=1/qS that models the parents. Crossover entails that the parameters con-
unbiased observer in the case of perception of an stituting a parent are treated as binary strings, and
object that is contained in the scope of vision plane, portions of the strings are exchanged among the
as seen in Figure 1a. In the experiments the number two solutions to create new solutions with features
of vision rays is denoted by nv. An object within the from both parents. This process is repeated for sev-
visual scope will be hit by a number of vision rays np, eral iterations, and due to the probabilistic favoring
and these rays are termed perception rays. The per- of fit solutions, eventually optimal solutions appear
ception of the object is given by P=np/nv. in the population (Goldberg, 1989; Zalzala and Flem-
ing, 1997).
Experiment Nr. 1 The resulting best solution after 40 generations
The first experiment concerns a basic issue in an is shown in Figure 3a in a plan view and in Figure 3b
architectural design, namely positioning an object, in perspective view, where the perception rays are
so that its perception from several viewing posi- seen. The circles in the figures mark the boundaries
tions is maximized in the sense that the object will at 3.0 m distance from the doors. In Figure 3c-e the
be perceived well at least from one of the relevant space is shown from the respective viewing position
viewpoints. This issue is exemplified by means of VP1, VP2, and VP3. The best position of the sculp-
positioning a sculpture in a museum space having ture is at the edge of the circle in front of viewing
several entrances; namely the space has three doors, position VP1. This position has the highest union of
where the relevant viewing positions are located de- perceptions in the feasible region, namely PU=.307.
noted by VP1, VP2, and VP3. The problem is to posi- This is composed of the perceptions P1=.157 at VP1,
tion the sculpture in the space, so that the visitors P2=.063 at VP2, and P3=.048 at VP3.
entering the space from either door will notice the For comparison the second best position is
object. The problem is to maximize the union of the shown in Figure 4, namely the perceptive plan view
perceptions from the three viewpoints P(E1ÈE2ÈE3), in Figure 4a, perspective perceptive view in Figure
while at the same time the sculpture positioned at 4b, and the perceptive views from VP1, VP2, and VP3
point x should not obstruct entrance to the room in Figure 4c-e respectively. The union of the percep-
from either door. The latter constraint is formulated tions PU=.255, that is 17% lower compared to the
by the condition ‖x-xo‖≥3, where xo is the position of best solution in Figure 3. The union is composed of
each viewing position. The maximization is carried the perceptions P1=.072 at VP1, P2=.152 at VP2, and
out by the method of random search, accomplished P3=.033 at VP3. The results demonstrate a common
through the method of genetic algorithm. Genetic design knowledge, namely when one aims to maxi-
algorithm is a stochastic optimization method from mize the perception of an object in a space with
the domain of computational intelligence. The al- several possible viewing positions, it is preferable to
gorithm starts from a number of random solutions position the object to have a high perception for at
referred to as members of a population. Each mem- least one of the possible positions, for that matter
ber satisfies the objective function to some degree, VP1, rather than having several moderate percep-
Figure 3
Best position for the sculpture
in a plan view (a); in a percep-
tive view (b); for VP1, P1=.157
(c); for VP2, P2=.063 (d); for VP3,
P3=.048 (e).
(a) (b)
(c) (d) (e)
Figure 4
Second best position for the
sculpture in a plan view (a); in
a perceptive view (b); for VP1,
P1=.072 (c); forVP2, P2=.152 (d);
for VP3, P3=.033 (e).
(a) (b)
(c) (d) (e)
Figure 5
Location in an urban scene,
where a new building is
subject to perception consid-
erations.
Figure 6
Zoomed out rendering of the
urban scene in Figure 5, where
a new building is subject to
perception considerations
from three viewpoints.
tions, i.e. without any outstandingly high one. The tions, is seen in Figure 5. The zoomed out rendering
lower perceptions in Figure 4c demonstrate the im- of the scene in Figure 5 is shown in Figure 6, where
plications of the Cauchy function in Figure 1b, where the three viewing positions VP1, VP2, and VP3 are
deviation from the frontal direction for an object, in indicated. Figure 7 schematically shows the floor
particular at a near distance from the observer, yield plan of the urban situation, as well as the percep-
reduction in probability density, i.e. visual attention tion cones and vision scopes belonging to the view-
is diminished in this case. ing positions, which are the endpoints of streets
entering to a square where the building is located.
Experiment Nr. 2 Figure 7b shows random vision rays having uniform
A second experiment concerns the perception of a pdf with respect to the vision angle modeling visual
building in an urban context from three viewpoints scopes for three viewing positions. Figure 7c shows
that are prominent locations in the surrounding of those rays among the vision rays that hit the build-
the building. The location in an urban scene, where ing subject to perception, for perception computa-
a new building is subject to perception consideration. The results from the perception fusion for the
respective building envelope portions are shown Figure 7
Figure 8, where the numbers display the fused per- Scheme of an urban situation,
ception associated with the respective portion. From where a building is subject
the analysis it is seen that the part of the envelope to perception analysis from
that is most intensely perceived from the three view- three viewpoints in plan view
points, is the area in front of VP2, while the second (a); random vision rays with
most intense part is the part of the building corner uniform pdf w.r.t. the vision
oriented towards VP1, which is expected consider- angle modeling visual scopes
ing the influence of the distance lo in the perception for three viewing positions
computations in (5). The information obtained from VP1, VP2, and VP3 (b); the rays
perception fusion is of relevance for a designer de- among the vision rays that
termining formal and functional details of the en- hit the building subject to
velope, for instance determining the position of en- (a) perception (c).
trance during conceptual design. Figure 9 shows the
fused perceptions of the building envelope from the
three viewpoints per envelope element with a vision
scope that is 20% narrower compared to Figure 8.
CONCLUSIONS
A method for fusion of perceptions is presented and
demonstrated with two examples from architectural
design. The probabilistic treatment, where percep-
tion quantifies the chance that an unbiased observ-
er notices an environmental object, is accomplished
through fusion of perceptions. The method of quan-
tified union of perceptions has been an unresolved
issue up till now, that is resolved in this presentation.
The fusion by probabilistic union yields significant (b)
information for designers. With the presented ap-
proach an object is to be perceived from several
viewpoints at the same time. Such abstraction is
necessary, since the precise analysis of the percep-
tions is a formidable issue due to abundant visual
scene information. The use of perception fusion
as constrained design objective has been demon-
strated by coupling the method with a probabilistic
evolutionary algorithm performing the constraint
optimization. The combination of the two proba-
bilistic methods is a powerful tool for designers as
it permits treatment of architectural design to be
highly constrained and involving many perception
related demands. Although the examples presented
are rather basic, the method is generic and yields
highly appreciable scoring executions in diverse ap- (c)
Figure 8
Fused perceptions of the
building envelope from the
three viewpoints per envelope
element.
Figure 9
Fused perceptions of the build-
ing envelope from the three
viewpoints per envelope ele-
ment with a vision scope that
is 20% narrower compared to
Figure 8.
plications in the areas where perception plays a role, ies on visual perception for perceptual robotics’, ICIN-
such as architecture, urbanism, interior and indus- CO 2006 - 3rd Int. Conf. on Informatics in Control, Auto-
trial design, as well as robotics. mation and Robotics, Setubal, Portugal, pp. 468-477.
Ciftcioglu, Ö, Bittermann, MS and Sariyildiz, IS 2006b, ‘To-
ACKNOWLEDGEMENT wards computer-based perception by modeling visual
Technical design assistance by Architect Paul de perception: a probabilistic theory’, 2006 IEEE Int. Conf.
Ruiter, providing us with the scene presented in the on Systems, Man, and Cybernetics, Taipei, Taiwan, pp.
second experiment is gratefully acknowledged. 5152-5159.
Foster, J 2000, The Nature of Perception, Oxford University,
REFERENCES Oxford.
Bhatt, M, Hois, J and Kutz, O 2011, ‘Ontological Modelling Gibson, JJ 1986, The Ecological Approach to Visual Percep-
of Form and Function for Architectural Design’, Applied tion, Lawrence Erlbaum Associates, Hillsdale, New Jer-
Ontology, pp. 1-32. sey.
Bittermann, MS and Ciftcioglu, Ö 2008, ‘Visual perception Goldberg, DE 1989, Genetic Algorithms, Addison Wesley,
model for architectural design’, Journal of Design Re- Reading, MA.
search, 7(1), pp. 35-60. Knill, DC, Kersten, D and Mamassian, P 2008, ‘Implications of
Bülthoff, H, Wallraven, C and Giese, M 2007, ‘Perceptual ro- a Bayesian formulation for visual information for pro-
botics’, in B Siciliano and O Khatib (eds), The Springer cessing for psychophysics’, Perception as Bayesian Infer-
Handbook of Robotics, Springer, pp. 1481-1495. ence, Cambridge, Cambridge, pp. 239-286.
Ciftcioglu, Ö, Bittermann, MS and Sariyildiz, IS 2006a, ‘Stud- Knill, DC and Richards, W 2008, Perception as Bayesian Infer-
ence, Cambridge University, Cambridge, UK. Treisman, AM and Gelade, G 1980, ‘A feature-integration
Marr, D 1982, Vision, Freeman, San Francisco. theory of attention’, Cognitive Psychology, 12(1), pp. 97-
O’Regan, JK, Deubel, H, Clark, JJ and Rensink, RA 2000, ‘Pic- 136.
ture changes during blinks: looking without seeing Wertheim, T 1894, ‘Ueber die indirekte Sehschaerfe’, Z Psy-
and seeing without looking’, Visual Cognition, 7(1-3), chol Physiol Sinnesorg, 7, pp. 172-189.
pp. 191-211. Yuille, AL and Bulthoff, HH 2008, ‘Bayesian decision theory
Palmer, SE 1999, Vision Science, MIT, Cambridge, MA. and psychophysics’ in DC Knill and W Richards (eds),
Smith, D 2001, The Problem of Perception, Harvard Univer- Perception as Bayesian Inference, Cambridge University,
sity, Cambridge, MA. Cambridge, UK, pp. 123-161.
Treisman, AM 2006, ‘How the deployment of attention de- Zalzala, AMS and Fleming, PJ 1997, Genetic Algorithms in
termines what we see’, Visual Cognition, 14(4), pp. 411- Engineering Systems, IEE Control Eng., Series 55, Cam-
443. bridge University, New York.
Ambient Surveillance by Probabilistic-Possibilistic
Perception
Michael S. Bittermann1, Ozer Ciftcioglu2
Delft University of Technology, The Netherlands
http://bk.tudelft.nl/en/research/research-projects/computational-intelligent-design
1
m.s.bittermann@tudelft.nl, 2o.ciftcioglu@tudelft.nl
Abstract. A method for quantifying ambient surveillance is presented, which is based

on probabilistic-possibilistic perception. The human surveillance of a scene through
observing camera sensed images on a monitor is modeled in three steps. First immersion
of the observer is simulated by modeling perception of the scene from the camera
locations using probabilistic perception approach. The perceptions are thereafter
combined by means of probabilistic union, simulating simultaneous watching of the scene
from multiple viewing positions. As third step the combined perceptions are converted
to a possibility using triangular possibility density function. The latter step accounts for
the fact that surveillance takes place via monitor depiction and not directly as perception
of the actual physical scene. The method is described and demonstrated by means of an
ambient surveillance application involving three cameras. The resulting possibility of
perception is compared to the case of using two cameras, quantifying the added value of
additional camera as to surveillance.
Keywords. Perception; possibility; ambient intelligence; surveillance.
INTRODUCTION
Ambient Intelligence refers to electronic environ- support for human interactions is aimed for. In this
ments that are sensitive and responsive to the pres- vision people are surrounded by intelligent intui-
ence of people (Aarts and Encarnacao, 2006). Such tive interfaces that are embedded in different kinds
electronic environments are called as ambient envi- of objects yielding an environment that is capable
ronment, referring to the surveillance of a physical of recognizing and responding to the presence of
ambience in the computer screen environment. Am- different individuals in a seamless, unobtrusive or
bient Intelligence involves different fields including invisible way (Ducatel et al., 2001). The European
electrical engineering, computer science, industrial Commission’s Information Society Technologies Ad-
design, human machine interaction, and cognitive visory Group (ISTAG) considers Ambient Intelligence
sciences. It stems from the combination of the three an important concept, as they predict that the con-
concepts ubiquitous computing, ubiquitous com- cept will be applied to everyday objects such as fur-
munication, and intelligent user friendly interfaces. niture, clothes, vehicles, roads and smart materials.
It is considered to provide a vision of the informa- According to ISTAG, Ambient Intelligence implies
tion society, where greater user-friendliness, more machine awareness of the specific characteristics
efficient services support, user-empowerment, and of human presence and personalities, taking care of
Figure 1
A door’s functional space is
not fully encompassed by the
field of view of two cameras
(a) (Bhatt et al., 2009); the
functional space is fully
encompassed by the field of
(a) (b) (c) view of two cameras (b); the
space is fully encompassed
needs and being capable of responding intelligently situation where the door and its functional space are by the field of view of three
to spoken or gestured indications of desire (Weyrich, entirely within the fields of view of the two cameras, cameras (c).
1999). Benefits in some practical applications have thereby complying with the requirement. In Figure
been reported, see e.g. Augusto and Shapiro (2007), 1c three cameras are used, and the consistency re-
Streiz et al. (2007), Ramos et al. (2008), Augusto and quirement is also fulfilled.
Nugent (2006). Examples of application areas are In an ambient intelligent system, human su-
personal assistance by mobile devices (Richard and pervision may be important in case continuous in-
Yamada, 2007), clothing (Boronowsky et al., 2006), situ monitoring of scenes is demanded for instant
entertainment (Saini et al., 2005; Dornbush et al., human intervention. In such a case, the functional
2007), office and meetings rooms (Waibel et al., space shown in Figure 1 is to be supervised by hu-
2010), and home environments (Aarts and Died- man through monitor watching. Here the human
eriks, 2007; Nakashima, 2007). The benefits in the perception plays an important role. The actual scene
applications concern enhanced security, and utility. is surveyed by the cameras, and at this stage human
Concerning security, an issue of common relevance perception is not in the play. However, the image of
is surveillance of objects in buildings, e.g. see (Take- the functional space is propagated to a screen, and
mura and Ishiguro, 2010). The objects may concern then the human perception via the screen becomes
building elements such as doors, hallways, etc., as an issue of assessment. Such assessments should be
well as valuable articles. For instance, in an environ- quantified to understand the difference among the
ment the monitoring of people passing through the probable camera positions, or among cases where
doors may be of relevance for security purposes, so different number of cameras are used. It is empha-
that the locations where surveillance cameras are sized that two, three, or more cameras may be used
suitably placed, and the number of cameras used to cover the functional space entirely, as exempli-
to supervise the environment, are important issues fied in Figure 1b and 1c, so that compliance with
to consider. This may be relevant both during the the consistency condition described above can be
design of an ambient environment, as well as dur- achieved in several ways that are not equivalent
ing the assessment of the surveillance provided for with respect to surveillance. As the human should
an existing environment. In an existing work this realize the presence of objects and events in his
issue is addressed by verifying if a functional space mind, which is a complex brain process involving
of a door is fully covered by supervision cameras uncertainty, quantitative assessment of the human
(Bhatt et al., 2009), which is a requirement to guard perception in the ambient environment surveil-
the traffic between the rooms. This is seen in a plan lance case becomes desirable and is challenging to
view in Figure 1a, where the door and its functional accomplish. Comparing the situations in Figure 1b
space, which is shown by a rectangle, are not fully and 1c, qualitatively three cameras in Figure 1c are
covered by the fields of view of two cameras. This favorable with respect to the human perception of
yields requirement inconsistency. Figure 1b shows a the functional space, providing more visual infor-
mation about the object to the human. Following analysis by human, probability theoretic computa-
the approach of existing works, such as Bhatt et al tions are used to simulate perception of objects by a
(2009), surveillance in Figure 1b and 1c is considered human, who is immersed in the scene at the camera
to be the same, as requirement consistency is treat- viewpoints.
ed as a binary statement. Binary verification of the
requirement compliance is giving some indication Probabilistic Perception Revisited
about the effectiveness of the camera surveillance. Due to the complexity of brain processes underlying
However, this may be not enough for the case of hu- perception, perception is to be modeled as a proba-
man supervision, which is based on human percep- bilistic event. That is, there is a chance to see an ob-
tion. Based on this view, the present work intends to ject, meaning the presence of the object is realized
make some steps forward along this line, providing in mind, which implies a chance of overlooking the
measured assessment about the quality of surveil- object, too. We can term this as the uncertainty of
lance of an ambient environment based on percep- human vision (Rensink et al., 1997; Bittermann and
tion modeling. Measured assessment is desirable in Ciftcioglu, 2008). For a single unbiased observer this
particular when optimal solutions are sought dur- uncertainty is quantified as described in Ciftcioglu
ing design of an environment, for instance with re- et al (2006b), Bittermann and Ciftcioglu (2008). Con-
spect to maximizing surveillance by optimal place- sider the basic geometry as shown in Figure 2a. P
ment and orientation of sensors, or minimizing the represents an observer’s point, where he is viewing
number of cameras while sufficient surveillance is an object. We consider a perception plane located at
provided. We note that in this work we assume that distance lo from the observer, and a scope of vision
there is no automated camera system for object rec- plane orthogonal to the perception plane, having
ognition involved, although even in that case, differ- the observer’s point and the object in it. The inter-
entiation among alternative camera utilizations, in section of the perception plane and the scope of vi-
order to determine the effectiveness of the machine sion plane is the y-axis. A line perpendicular to the
recognition, still remains an issue. perception plane, passing from the point P, is the x-
The organization of the paper is as follows. The axis. The observer has a visual scope in the scope of
methodology section describes the treatment of the vision plane, defined by the angle θS=π/2, which is
probabilistic and possibilistic aspects of the surveil- termed as vision angle. He is viewing the object that
lance. The computer experiment section describes an subtends the angle θb-θa. An unbiased observer is
example application of the method for an ambient modeled, i.e. he has no preference for any direction
environment, and the section is followed by conclu- within the visual scope. This means the probability
sions. density function (pdf ) with respect to θ is given by
fθ(θ)=1/θS, as seen in Figure 2b upper. As the object
METHODOLOGY subtends the perception angle θb-θa, it has an as-
This research aims to make assessment about the
quality of human surveillance of an object based on sociated perception
camera sensed information. When a human views a , shown by the gray shaded area in Figure 2b upper.
camera sensed scene on a screen, in order to give P quantifies the probability the object is mentally re-
meaningful interpretation to the scene he infers the alized by the observer. The perception can be com-
information about the camera position and orienta- puted along the y-axis in Figure 2a by radially pro-
tion from the scene, without having been explicitly jecting the object from P on the y-axis. It yields a line
informed about these. This process of assuming of segment, spanning ya and yb, as seen in the figure.
a camera position by human is called immersion. To The uniform pdf with respect to the vision angle θ is
model this early stage of the ambient environment given by fθ(θ)=1/(π/2) and corresponds to the follow-
Figure 2
An object projected on
the perception plane and
perceived from P (a); sketch
of the probability density
function (pdf) characterizing
perception with respect to θ
(b upper); pdf characterizing
perception with respect to the
y direction for lo=2 (b lower).
(a) (b)
ing probability density with respect to y (Bittermann Figure 3. The scene subject to investigation is shown
and Ciftcioglu, 2008) in Figure 3a, presenting a plan view of two rooms
connected by a door and an associated functional
(1) space shown by a rectangular box around the door.
The plot of (1) for lo=2 is seen in Figure 2b lower. The The functional space is subject to surveillance via
perception is computed by the three cameras, where the visible portions of this
space respectively subtend the angles θ1, θ2, and θ3
(2) as indicated by the dark shaded areas in the figure.
and the result is shown by the gray shaded area in The dashed lines in the figure indicate the bound-
the figure. It is emphasized that the sizes of the gray aries of the cameras’ fields of view, where their as-
shaded areas in Figure 2b upper and 2b lower are sociated angles θS1, θS2, θS3 are taken to be the same
the same. We note that for the perception of a three in this example. The intersection among the three
dimensional object both vision angle and percep- camera scopes form a universe of discourse for the
tion angle become respective solid angles. surveillance events as shown in Figure 3b by means
of bold dashed lines. We define the following three
Union of Perception Events perception events within this universe as seen in
We emphasize that for the surveillance of the ambi- Figure 3c. The event a human observer, who is im-
ent environment being considered, the consistency mersed at camera 1, becomes aware of the function-
requirement mentioned above stipulates that the al space that is at the same time within the scopes of
functional space should be entirely encompassed camera 2 and camera 3, is denoted by event E1. Con-
by multiple cameras’ fields of view. This means a versely, the perception event from camera 2 that is
human observing the scene will obtain the infor- at the same time within the fields of view of camera
mation from multiple cameras at the same time. In 1 and camera 3 is denoted by E2. In the same way,
this respect we consider the case shown in Figure the perception event from camera 3 that is at the
1, where a single camera is not sufficient to comply same time within the fields of view of camera 1 and
with the consistency requirement, and in this study camera 2 is denoted by E3. The regions in the scene
we consider the perceptions by means of three cam- corresponding to the events are shown in Figure
eras, denoted camera 1, camera 2 and camera 3 in 3c, where the space belonging to E1 is delimited by
Figure 3
Functional space of a door
subject to surveillance by
means of three camera sen-
sors (a); universe of discourse
for the surveillance, where
θS1, θS2, and θS3 denote the
respective fields of view of the
cameras; perception events
E1, E2, and E3 , their union, and
intersection (c); Venn diagram (a) (b) (c) (d)
corresponding to the events in
figure 3c. means of red dashed lines, for E2 by means of blue responding to E1ÇE2ÇE3 from plan view, and Figure
dashed lines, and for E3 by means of orange dashed 4f shows the same region from a perspective view.
lines. The probability of the perception events is ob- The probabilities P(E1), P(E2), and P(E3) are obtained
tained by P(E1)=θ1/θS1, P(E2)=θ2/θS2, and P(E3)=θ3/θS3. It by similar computations as given by (2) but for three
is to note that E1, E2, and E3 are independent events. dimensional space, where θ becomes solid angle Ω.
With respect to ambient surveillance assessment
being aimed for in this work, the event subject to Converting the Probability into Possibility
computation is the union of the perception events It is emphasized that the computations above mod-
PU=E1ÈE2ÈE3. The union refers the event that the ob- el the perception of observers, who are viewing the
server becomes aware of the functional space either functional space being present at all three camera
via immersion at camera 1, camera 2, camera 3 or via positions. However, the scene is actually viewed on
combinations among them at the same time, while a monitor screen and not directly from locations in
the consistency condition, namely that the event the physical environment. That is, no actual object is
is to take place within all cameras’ fields of view, is being perceived in the ambient environment case,
fulfilled at the same time as boundary condition. but a visual representation of the scene on a screen
The region of space in the scene that corresponds is being perceived. This yields the immersion phe-
to E1ÈE2ÈE3 is delimited by the white dashed line in nomenon, which we can also term as virtual percep-
Figure 3c. The region of space in the scene that cor- tion. In the ambient environment case, instead of
responds to E1ÇE2ÇE3 is visualized in the same figure perception alone an assessment of the perception
by means of a yellow dashed line. Figure 3d shows is to be carried out, and this assessment should be
a Venn diagram corresponding to the perception expressed in possibilistic terms, namely as possibil-
events in Figure 3c. ity of perception. This means the probability quan-
The regions corresponding to the universe of tifying the perception of the object by the observer
discourse and encompassing the perception events should be converted to a possibility of perception.
are shown in 3D renderings in Figure 4. Figure 4a This is shown in Figure 5. Figure 5a shows the per-
shows the fields of view of the cameras from top ceptions of the functional space from the three
view in red color, as well as the cones encompass- cameras. The probability density functions fθ(θ) are
ing the respective perception events E1, E2, and E3 in integrated along angle dimension θ, yielding the
yellow color. The same regions are shown in Figure perceptions P(E1), P(E2), and P(E3). It is to note that
4b from a perspective view. Figure 4c shows the uni- each of the three integrals have their center points
verse of discourse from top view and Figure 4d from at θ=0 as seen in the figures. This is due to the sur-
a perspective view. Figure 4e shows the region cor- veillance purpose, where the cameras are oriented
Figure 4
Fields of view of the cameras
denoted by C1, C2, C3 and
the cones in which perception
events takes place from top
view (a); from a perspective
view (b); universe of discourse
from a top view (c); from a per-
(a) (b) (c) spective view (d); The region
corresponding to E1ÇE2ÇE3
from top view (e), from a
perspective view.
(d) (e) (f )
in such a way that the object subject to perception ulating the immersion, the point θ=0 also represents
is located at the center of the respective fields of a reference point for perception possibility compu-
view of the cameras. The probability of the union of tation on the monitor, as zero refers to the center
the perception events P(E1ÈE2ÈE3) is shown by the of the fields of views of the cameras, i.e. center of
hatched area in Figure 5b. Being an integral of the monitoring screen. For the possibility assessment,
uniform pdf fθ(θ)=1/θS, P(E1ÈE2ÈE3) corresponds to the possibility density is subject to integration over
an angle domain θ’, as seen in the figure. It is noted the angle domain θ’, where the integration starts
that P(E1ÈE2ÈE3) is also centered at θ=0 being the ref- from θ=0, yielding the dark gray shaded area in Fig-
erence point of the perception computation in the ure 5b, the size of which quantifies the possibility
scene as result of the immersion phenomenon. The of perception. It is emphasized that the integration
pdf has a possibilistic density counterpart, namely a starts from zero, i.e. in the middle of the screen, as
triangular possibility density function as seen in the to human perception, the possibility of perception
figure. It is noted that the possibility density is maxi- is assessed starting from the middle of the screen. θ’
mum at the place that corresponds to the expected starts from zero and maximally extends covering the
value of the uniform probabilistic density with re- interval -θS/2 and +θS/2, so that its maximum value
spect to θ, namely θ=0. Therefore, next to being the becomes θS. Figure 5c shows a sketch of the relation-
reference point for the perception computation sim- ship between possibility of perception versus the Figure 5
Perception of the functional
space from one of the cameras
(a); conversion of the union of
the perceptions to possibility
of perception (b); possibility of
perception versus perception
(a) (b) (c) (d) as sketch (c); as plot (d).
Figure 6
Camera picture taken from
camera 1, where P(E1)=.246
(a); from camera 2, where
P(E2)=.207 (b); from camera 3
where P(E3)=.310 (c).
(a) (b) (c)
corresponding union of perceptions, and Figure 5d COMPUTER EXPERIMENT

shows a plot of the same relationship. From Figures Based on the considerations above a computer
5c and 5d it is seen that for a certain perception P, experiment is carried out, where the possibility of
there is always a perception possibility having a perception is obtained for the scene shown in Fig-
greater value than P. As the perception is increas- ure 4 with the camera positions as indicated in the
ing, the associated possibility is also increasing in a Figure. It is noted that the cameras are located at the
non-linear way. In this treatment, obviously there is ceilings of the rooms at the same height, and they
no possibility consideration if perception is not oc- are oriented in such a way that the central line of
curring. This means a triangular possibility density the cameras’ fields of view are directed towards the
cannot be constructed without having referred to a center points of the respective visible portion of the
probability density associated with perception. Such functional space. The camera pictures of the scene
probability density is known to be attention (Ciftcio- taken from the three positions are shown respec-
glu et al., 2006b). It is noted that shape of the func- tively in Figures 6a, 6b, and 6c.
tion shown in Figure 5c is independent of the size of In the experiment, the unbiased visual attention
the scope θS. given by the probability density per unit solid vision
The possibility density function defined as a angle Ω given by fΩ(Ω)=1/ΩS and ΩS=π sr is approxi-
triangular fuzzy set shown in Figure 5b is the coun- mated by means of probabilistic ray tracing, in order
terpart of the probability density function with re- to deal with geometric complexity of environment.
spect to perception along the y-axis shown in Fig- In this treatment rays are sent in random directions
ure 2b lower. The form is precisely represented by from camera position, and the intersections with en-
the Cauchy function given by (1) that simulates the vironmental objects are analyzed. The ray directions
human perception in the scene as result of the im- are generated in such a way that fΩ(Ω)=1/ΩS is ap-
mersion process. Both functions, namely triangular proximately fulfilled, which is accomplished by using
possibility density function and Cauchy probability multiple Gaussian pdf as described in Ciftcioglu et al
density function, have a maximum at the respec- (2006a). Figure 7 shows the rays sent to simulate the
tive reference starting points. This is confirmed by perceptions via the three cameras. Figure 7a shows
the common vision experience, that an observer is the rays that simulate the unbiased vision within the
more aware of an object positioned in front of him, scope defined by the cameras’ fields of view, from
compared to a similar object that is located at some a plan view. These are termed as vision rays. The
lateral distance from the former object. This is be- same rays are shown in Figure 7b from a perspective
cause the observer will remember more details of view. It is noted that in order to display individual
the former compared to the latter. It is noted that rays, in the figure merely 200 rays per camera posi-
the shape of the monitor screen is not relevant to tion are shown, although in the experiment 2000
this computation. rays are used for accuracy of the results. Figure 7c
shows those rays among the vision rays that inter-
Figure 7
Vision rays from a plan view
(a); from a perspective view
(b); perception rays from a
plan view (c); from a perspec-
tive view (d).
(a) (b) (c) (d)
sect the functional space in a plan view, and these ments, and in particular provides information on the
are termed as perception rays as they simulate the remaining surveillance in the case of a camera fail-
perception events E1, E2, and E3. The same perception ure, which provides an indication of the robustness
rays are shown in Figure 7d in a perspective view. of a surveillance situation.
The perception event P(E) is obtained by P(E)=np/nv
, where np denotes the number of perception rays, CONCLUSIONS
and nv the number of vision rays. A probabilistic-possibilistic approach that models
The results from the experiment are P(E1)=.246; surveillance of a scene by human via three cameras
P(E2)=.207; P(E3)=.310, so that P(E1ÈE2ÈE3)=.588, yield- is described. The first stage in camera based human
ing possibility of perception as pp=.830. This quantifies surveillance is the immersion phenomenon, and
the possibility of perceiving an event at the func- this is modeled in the presented work by means
tional space of the door based on the camera posi- of perception computations that are probabilistic
tions considered. It is interesting to investigate what in nature. These computations reflect the fact that
the difference in perception possibility is in case two remembrance of visual information processed by
cameras are used instead of three. Considering the human vision system is not certain, i.e. it is subject
case camera 1 is not used, then P(E2ÈE3)=.453, yield- to probabilistic considerations. The second stage of
ing the perception possibility as pp=.701. In case the surveillance is conversion of the perception into
camera 2 is not used, then P(E1ÈE3)=.480 yielding possibility. The possibilistic treatment accounts for
perception possibility as pp=.729; and for camera 3 the fact that the observation event does not con-
being not used P(E1ÈE2)=.402, so that the possibil- cern perception of an object from an actual location
ity becomes pp=.642. Thus, compared to using two in space, but perception of a camera sensed image
cameras, use of three cameras increases the pos- of the object on a monitor. This way perception is as-
sibility of perception by 18.4%, 13.9%, and 29.3% sessed in the form of a fuzzy statement. In the same
respectively for the three cases. It is also interesting way as probability is due to integration of a prob-
to consider using only one camera compared to usability density over some physical domain, so that
ing three cameras. Using camera 1 exclusively, the it is associated to an event, possibility is computed
perception possibility is pp=.431 so that the three by means of integration of an associated possibil-
cameras entail an increase of 93%; using camera 2 ity density function belonging to the same domain.
exclusively the possibility is pp=0.371 implying an in- The domain in the present case is vision angle. The
crease for the three cameras of 124%; and in case ex- computer experiments presented in this paper con-
clusively camera 3 is used the perception possibility firm the qualitative statement, that the number of
is pp=.524 implying an increase of 58% for the case of cameras influences the possibility of perception. The
using the three cameras. This information is essen- probabilistic-possibilistic treatment described in this
tial in determining the surveillance level of environ- paper uniquely quantifies this possibility, providing
precision assessment of surveillance of ambient en- perception: a probabilistic theory’ in 2006 IEEE Int. Conf.
vironments. This implies that through the novel ap- on Systems, Man, and Cybernetics, Taipei, Taiwan, pp.
proach, subtle differences among surveillance situa- 5152-5159.
tions are distinguished, allowing for more conscious Dornbush, S, Joshi, A, Segall, Z and Oates, T 2007, ‘A human
decision making. This may have important place in activity aware learning mobile music player’ in 2007
diverse applications, such as domestic healthcare, Conf. on Advances in Ambient Intelligence, IOS Press, pp.
safety and security of buildings and cities, applying 107-122.
to both, existing situations, as well as during design Ducatel, K, Bogdanowicz, M, Scapolo, F, Leijten, J and
of new environments. It is interesting to note that Burgelman, J-C 2001, ‘Scenarios for ambient intelli-
different stakeholders may use the method for dif- gence in 2010’ in IPTS, Seville, pp. 1-55.
ferent purposes, such as verifying if surveillance is Nakashima, H 2007, ‘Cyber assist project for ambient intel-
sufficient, or verifying that it is not excessive, for in- ligence’ in JC Augusto and D Shapiro (eds), Advances in
stance for the sake of privacy of users. Ambient Intelligence, IOS Press, pp. 1-20.
Ramos, C, Augusto, JC and Shapiro, D 2008, ‘Ambient intel-
REFERENCES ligence: The next step for artificial intelligence’, IEEE In-
Aarts, E and Encarnacao, J 2006, ‘Into Ambient Intelligence’ telligent Systems, 23(2), pp. 15-18.
in E Aarts and J Encarnacao (eds), True Visions - The Rensink, RA, O’Regan, JK and Clark, JJ 1997, ‘To see or not
Emergence of Ambient Intelligence, Springer, pp. 1-16. to see: The need for attention to perceive changes in
Aarts, EHL and Diederiks, E 2007, Ambient Lifestyle : from scenes’, Psychological Science, 8(5), pp. 368-373.
Concept to Experience, Book Industry Services. Richard, N and Yamada, S 2007, ‘Two issues for an ambient
Augusto, J and Shapiro, D 2007, Advances in Ambient Intelli- reminding system: context awareness and user feed-
gence. Frontiers in Artificial Intelligence and Applications, back’ in JC Augusto and D Shapiro (eds), Advances in
IOS Press Amsterdam. Ambient Intelligence, IOS Press, pp. 123 - 142.
Augusto, JC and Nugent, CD 2006, Designing Smart Homes, Saini, P, de Ruyter, B, Markopoulos, P and van Breemen, A
LNCS, Springer, Heidelberg. 2005, ‘Assessing the effects of building social intelli-
Bhatt, M, Dylla, F and Hois, J 2009, ‘Spatio-terminological gence in a robotic interface for the home’, Interacting
inference for the design of ambient environments’ in with Computers, 17, pp. 522-541.
Spatial Information Theory, Springer, pp. 371-391. Streiz, NA, Kameas, AD and Mavrommati, I 2007, The Disap-
Bittermann, MS and Ciftcioglu, Ö 2008, ‘Visual perception pearing Computer, LNCS, Springer, Heidelberg.
model for architectural design’, Journal of Design Re- Takemura, N and Ishiguro, H 2010, ‘Multi-camera vision for
search, 7(1), pp. 35-60. surveillance’ in HNe al. (ed), Handbook of Ambient Intel-
Boronowsky, M, Herzog, O, Knackfuß, P and Lawo, M 2006, ligence and Smart Environments, Springer, pp. 149-168.
‘Empowering the mobile worker by wearable comput- Waibel, A, Stiefelhagen, R, Carlson, R, Casas, J, Kleindienst,
ing - wearIT@work’, J. of Telecommunications and Infor- J, Lamel, L, Lanz, O, Mostefa, D, Omologo, M, Pianesi,
mation Technology, (2), pp. 9-14. F, Polymenakos, L, Potamianos, G, Soldatos, J, Sutschet,
Ciftcioglu, Ö, Bittermann, MS and Sariyildiz, IS 2006a, ‘Au- G and Terken, J 2010, ‘Computers in the human inter-
tonomous robotics by perception’ in SCIS & ISIS 2006, action loop’ in H Nakashima (ed), Handbook of Ambi-
Joint 3rd Int. Conf. on Soft Computing and Intelligent Sys- ent Intelligence and Smart Environments, Springer, pp.
tems and 7th Int. Symp. on advanced Intelligent Systems, 1071-1109.
Tokyo, Japan, pp. 1963-1970. Weyrich, C 1999, ‘Orientations for workprogramme 2000
Ciftcioglu, Ö, Bittermann, MS and Sariyildiz, IS 2006b, ‘To- and beyond ‘ in Information Society Technologies Ad-
wards computer-based perception by modeling visual visory Group (ISTAG), Luxemburg.
The Jacobs´ Urban Lineage Revisited
Analytical rudiments for the further development of the

phenomenological approach to the study of the perception of people in
urban space implicit in Jane Jacobs´ work
Claudio Araneda
Universidad del Bío-Bío, Chile
claraneda73@gmail.com
Abstract. Since the almost simultaneous publication of Kevin Lynch and Jane Jacobs´
seminal and pioneer urban manifestos, the discipline has been increasingly permeated by
what could be rightly called the phenomenological impulse. While sharing methodologi-
cal principles, however, they represent two very distinct approaches to the study of
urban matters, a distinction rooted on their chosen object of study. The drawing of
this distinction constitutes this research´s point of departure. Its fundamental aim is to
help further the development of what we characterize as the Jacobs´s lineage of urban
thought. To this end, the paper outlines methodological rudiments for the development of
a methodological tool that would allow the beginning of a systematic study of the patterns
of people´s presence and absence in urban space (streets). We call it Urban Polaroid. This
work is part of a government funded (fondecyt 11110450) project.
Keywords. Urban phenomenon; phenomenology; Urban Polaroid; space syntax; Jane
Jacobs.
INTRODUCTION
Lewis Mumford famously branded Jane Jacobs` of urban thought. This said, we argue that there is
work as “home remedies for the urban cancer” (Mill- another equally distinct lineage. One that springing
er, 1986). It was more than the derisive characteriza- from Jacobs’ seminal work- for reasons that will pres-
tion of an opinionated and well-read urban scholar. ently be discussed- has remained markedly under-
It reflected the mood of a whole generation of urban developed. This lineage seeks urban knowledge not
planners that have systematically sought the source in the study of the perception of urban space (3D
of urban knowledge in the study of the already built and 2D) but in the perception of that other highly
cities and how we perceive it. This archaeological differentiated spatial manifestation in the city: peo-
kind of approach to urban studies has evolved, via ple. This paper offers methodological rudiments to
Space Syntax, into a highly sophisticated and suc- further the development of this lineage. Its central
cessful methodological corpus for the understand- argument connects with Hillier´s fundamental cri-
ing of urban space. We call this the Lynch lineage tique regarding urbanism´s historical and atavic
tendency to dogmatically prescribe as well as to Lynch´s urban approach (1960), on the other
the sheer lack of analytical tools for urban analysis. hand, was also rooted in the pedestrian´s percep-
It also connects with Ratti´s critique of Hillier´s work tion, but this time, of urban space. That is to say,
regarding the leap of faith implicit in Space Syntax Lynch´s object of study was the perception of con-
predictions based on the axial map. It differs with stant spatial patters through our daily navigation of
both, however, in the object of study to which we the city streets, the current validity of his approach
apply ourselves. becoming manifest in freshly opened avenues of ur-
ban research (Morello and Ratti, 2008). Indeed, it has
DRAWING A DISTINCTION BETWEEN been this later lineage of urban studies the one that
THE LYNCH AND THE JACOBS` LINEAGE. has seen the most dramatic developments in the
Written from the point of view of a pedestrian sen- last decades. This approach, characterized by its in-
sible to the unique and ever understudied phenom- trinsically archaeological nature, concerns itself with
enon of perceiving another human being, Jacobs the perception of urban space- inhabited or in ruin
manifesto Death and Life of Great American Cities like state- from a geometrical or topological point of
(1961) was an open and frontal attack against the view, depending on the placed emphasis.
planning tradition advocated by leading figures A representative and consistent offspring of
such as Le Corbusier and Ebenezer Howard and their this lineage of urban studies is the ground breaking
anti-autopoietic impulse towards the ruralization of body of work developed by the Space Syntax Lab at
the urban universe and the urbanization of the rural the Barttlet School of Architecture, UCL in London
universe respectively. Both of them united by their and all that has sprouted from it. Hillier- its founder
reliability on the new means of transport as a solu- father- succeeded in developing a precise tool for
tion to urban ailments, a stance widely trumpeted the study of architectural and urban layouts, discov-
by Soria Matta and already implemented in two ering in the process a close relationship between
highly praised precedents: Barcelona´s eixample and their topological configuration and the patterns of
the rebuilding of Paris. A school of urban thought pedestrian flow they describe (Hillier, 1996). It has
that at least in the United States had by then be- been its intrinsically non-discursive, phenomeno-
come the school of choice of both, planners of aca- logical stance that has rendered most of its findings
demic pedigree and business speculators alike. It irrefutable, setting a new standard not only in urban
was the already proved pernicious consequences analysis methodological consistency but also in ur-
of this school of thought that Jacobs famously per- ban data representation.
ceived in the economically informed interventions Another prolific offspring of this lineage has
of Robert Moses in Manhattan. been the work developed at the Senseable Lab in
Against this tradition, one of her prevailing concerns MIT directed by Carlo Ratti. He and his team have
was, as she called it, “the social behavior of people mainly focused on the analysis of urban data in the
in the cities”, meaning by cities the streets we walk form of electromagnetic pulses emitted by elec-
every day. Whereas it is a fact that her observations tronic devices (chiefly mobile telephones) carried
lacked rigorous analytical backup, her general meth- by people every day during their daily urban navi-
odological framework and object of study were une- gations. Interestingly enough, it has been precisely
quivocal: the experience of walking through the city Carlo Ratti, one of Space Syntax´s techniques most
focusing on the patters of people presence. Not for effective critics, who has brought Space Syntax prin-
nothing she is credited with having introduced the ciples to its last logical consequences by developing
notion of “eyes on the street”, by which she meant Digital Elevation Models (DEMs) with a view to com-
not “private eyes” but presence and co-presence in plement Space Syntax reductive two-dimensional
the Hillerian sense, particularly, that of residents. approach (Ratti, 2005). That is to say: a three dimen-
sional version of Space Syntax that aims to incorpo- in need of urban knowledge. Once this distinction is
rate sophisticated simulations of pedestrian move- made, Space Syntax is revealed for what it is, namely,
ment and view sheds, among other factors. a very specific kind of urban phenomenology: a phe-
This said, we argue that despite the great pro- nomenology of the city´s topology. One that gives
gress made by these representative techniques of us no direct knowledge about pedestrians patters of
urban analysis, they remain fundamentally specu- behaviour. Ratti has already pointed out that space
lative with regard to perception of people in space syntax´s way of proving the connection between
in the sense that none of them approaches it from these two variables is by means of surveys. This is, by
an experiential point of view. That is, from the means of “a posteriori” correlations between the ax-
point of view of an embodied, walking subject. In- ial map results and observed movement data (Ratti,
deed, whereas lines of research derived from Space 2005). In the case of Ratti´s DEMs, we see a similar
Syntax´s developments have led to the develop- procedure applied this time to the study of view
ment of agent-based models of pedestrian flow sheds. His strategy of electromagnetic signal track-
(Batty and Jiang, 1998), Ratti´s work has given rise ing on the other hand brings us closer to the pedes-
to the “wiki city” notion, approach whose object of trian who, nonetheless, remains an electromagnetic
study is made up of electromagnetic signals emitted mobile signal. As for Batty´s agent based models, we
by electromagnetic devices (Calabrese et al., 2007a; already fall into a thoroughly speculative stance re-
2007b; 2007c; 2007d; 2007e; Calabrese, 2008). In garding the study of pedestrian behaviour. In sum,
both cases, real people, understood as living hu- whereas space syntax´s approach to people study
man bodies, are nowhere to be seen. As a result of is post analysis (and at any rate not sophisticated as
their eminently speculative nature, the predictions an analytical tool), Ratti´s and Batty´s are downright
related to people´s presence on the street have re- non-experiential.
mained potentially flawed in that they do not pro- Jacobs´s approach, although still in a rudimen-
ceed from direct observation of people but from a tary stage, was a phenomenology of embodied peo-
priori speculations derived from computer models. ple perception, an object of study that from a meth-
odological point of view proved to be very difficult
PEOPLE AS OBJECT OF STUDY to map due to it being a moving target, so to speak.
As it has already been amply discussed elsewhere, Thus although the impulse latent in Jacob´s work
Jacobs and Lynch´s approaches were indeed tac- can be traced back, through Hall´s proxemics (Hall,
itly grounded on a phenomenological standpoint 1969; 1973; 1976), down to the rather unknown
(Seamon, 2012). This said there is a distinction that work of the German architect Herman Maertens
has not yet been clearly made. In phenomenol- (1884), it has ultimately remained analytically weak.
ogy- at least in the case of the proto phenomenol- The fundamental aim if this work is to help to further
ogy of Goethean extraction- the fundamental law of its development by means of introducing meth-
knowledge generation is that this should be derived odological rudiments that would allow a systematic
from a direct relationship with the chosen object of mapping of the human universe, so to speak, thus
study. Thus the unequivocal Goethean admonition: complementing the successful efforts made by the
“seek nothing beyond the object, they themselves, phenomenologists of urban space.
well contemplated, are the theory” (Seamon and Za-
jonc, 1998, p. 4) URBAN POLAROID (A METHODOLOGI-
Seen from this point of view, it becomes clear CAL OUTLINE)
that the Jacobs and the Lynch´s lineages differ not in Acknowledging from the outset that all record of an
method but in their chosen object of study. Put dif- experience is a reduction of it, the basic methodo-
ferently, they differ on the source they turn to when logical principle is the following. If what we want is
to know which are the patters of people presence CASE STUDY (CONCEPCIÓN, CHILE)
in the streets of any given urban area at any given All the streets belonging to the historical layout of
time, then what we need is a simultaneous photo- the city of Concepción are journeyed along (for a
graphic record of all the streets within the defined plan, see Figure 5). To this end, we created a patrol
perimeter. In principle, this could be done in two of photographic record composed of 26 students.
ways. One of them is by means of satellite or drone´s The format established that every street should be
aerial pictures. In this kind of record, people become walked in straight, from one end of the perimeter
dots on the street. Another way of achieving this to the other, and that all journeys should start at
without having to fly away from the streets is to re- the same time, in this case, midday. Unless impos-
sort to a photographic scanning of the streets at ob- sible, the itinerary must be made through pedes-
server level. While we consider this later path to be trian areas only. While doing so, a video record
a properly experiential one, its implementation pre- from a constant observer level and with constant
sents considerable and at the same time, interesting lens aperture is done. Depending on the amount of
practical problems. frames exported from the video record, we obtain
Just like space syntax´s axial map calculus de- a photographic record of variable intensity. That is,
pends upon the distance from all to all lines or an “n” number of frames per street segment, under-
streets, in order to obtain an accurate estimate re- standing for segment, the length of the street de-
garding actual presence of people in the street, we fined between intersections with other streets or, if
need to capture the state of all the streets within the preferred, between corners. In this case, we used a
chosen area of study at the same time. So for exam- low intensity: 2 pictures per segment. Arranged in
ple, if the intensity needed to validate the reliability filmstrip format, this raw, unedited record shows as a
of the study is, say, 3 pictures by segment (or block) result a reduced general state of the behavior of our
and the total numbers of segments (or blocks) that visual field during the journeys (Figure 1).
make up the street is, say 10, the total amount of Applying simple raster graphics, we then pro-
pictures needed for this particular street would be ceed to transform, frame-by-frame, all visual infor-
of 30. Seen from an ideal point of view, this means mation in the shape of human beings into colored
that what we really need in order to obtain a true “in- surfaces, in this case, a red surface (Figures 2 and 3).
stant” or “Polaroid” of this street are 30 different cam- This first step has the peculiar characteristic of
eras (people) taking a snapshot (in the same format: being quantitatively and qualitatively very eloquent
height, level, lens aperture, etc.) at exactly the same in that it already reveals a great deal of information
time. This process in turn should then be repeated regarding the patterns of behavior of our visual field
in all the streets contained within the chosen area. with regard to the presence of people in it. (We have
If the total number of streets within this chosen area defined three kinds of archetypal visual information
is, say, 20, the amount of cameras (people) needed in to be found within the urban universe: information
order to get the Polaroid is 600. Since logistically this in the shape of people, information in the shape of
is extremely difficult and probably counterproduc- urban space and information in the shape of nature.
tive- though not impossible-, we resorted to urban This paper only deals with the first kind.) That is, how
journeys or navigations. That is, video/photographic much surface of our photographically reduced visu-
journeys along all the streets contained within the al field is populated by information in the shape of
chosen area of study. The consistency and reliabil- people. Comparing the colored area to the total of
ity of this approach will depend exclusively on the the frame, we then obtain the percentage of visual
amount of journeys undertaken in a day, month and information in the shape of people for that particular
year per each street under observation. frame. Doing the same operation with every frame
Figure 1
Photographic filmstrips of
both, longitudinal (bottom)
and transversal (top) journeys
through the analyzed area.
of a particular street, we then obtain the average percentage of information in the shape of people for
percentage of information in the shape of people for this particular city at the particular time in which the
that particular street. Finally, repeating the same op- video/photographic journeys were done (Figure 4).
eration in all streets gives us as a result the average
Figure 2
Rasterisation and calculus
procedure for visual density
of information in the shape of
people of each frame.
Figure 3
Filmstrips of both, longitudi-
nal (bottom) and transversal
(top) journeys with informa-
tion in the shape of people in
raster form.
PRACTICAL CONTRIBUTIONS space syntax, Batty´s agent based computer models

We believe the most important contribution that and Ratti´s DEM´s and signal tracking techniques
the implementation of this tool brings about is that (to name a few) tells us in which streets people are
of effectively complementing the abstract archaeo- more likely to be found, the urban Polaroid reveals
logical approach developed so far by the Lynch ipso facto where people effectively are. Put different-
lineage of urban thought. Indeed, whereas Hillier´s ly, by departing from actual experience, the urban
Figure 4
Table of quantitative levels
of visual information in the
shape of people per frame
(columns) with averages per
street (files) in the last column.
Longitudinal journeys bottom,
transversal journeys on top.
Figure 5
Two dimensional syntax of
the levels of people presence
per street.
Polaroid reduces to nil the speculation implicit in tively akin- as it can be readily gathered from the di-
most of the urban tools of analysis developed by the agrams they yield- they differ dramatically in quality.
Lynch lineage representatives. Indeed, transform- One showing potential, the other showing actuality;
ing the table numbers into graphs, the result shows one being of interest only to specialists (particularly
a remarkable similarity with space syntax findings transport engineers and property developers), the
(Figure 5). other to urbanists and citizens in general; one re-
Red, our chosen color for high visual people vealing information about the already built city, the
density, tends in this case to coincide with the axial other about the city yet to be built.
map analysis result, which, if applied locally, would
show the central streets of the chose area as the THEORETICAL CONTRIBUTIONS
most integrated one. Hence, although the knowl- One of Jacob´s central declared concerns was to
edge obtained via the two approaches is quantita- get to know “how a city works”. What she called “the
underlying order of cities” (Jacobs, 1961, p. 25). In abstract computer aided analysis. This might prove
line with Luhmann (1995), today we might say that crucial in the cases where the axial map analysis
Jacobs quest was for the discovery of the laws that does not conform to actual reality of a determined
secure the autopoiesis of the urban universe and, as street or area of the city. Moreover, it achieves this
a consequence of this, its perpetuation in time. If, as without the need of people carrying mobile phones
the Goethean maxim goes, theory building derives or any kind of microchips. This makes it less invasive
from direct object contemplation, then, in order to and more citizen friendly.
obtain urban knowledge, all we need to do is to find Future complementary applications include a
out which is the urbanist´s object of study. Jacobs Polaroid of the visually perceived built universe and
was neither explicit nor sure about the answer to another of the natural universe, aspects that might
this question. Yet her main object of study always throw light upon the other two archetypal kinds of
remained people on the streets. visual information in the city and the relationship
Very few shared with her this interest. One of between them. In sum, the Urban Polaroid approach
them was the urbanist Jaime Garretón, author of the offers a portrait, and as such, a reduced view of a
first truly general urban theory, for whom “nothing complex that we have called the “archetypal citizen”,
is definite in a city, except its laws” (Garretón, 1975, according to previous research, the urbanist true ob-
p. 273), laws that, according to him, are essentially ject of study (Araneda, 2008; 2010; 2011).
communicative laws. To be sure, the laws of commu-
nication between people. This said, neither Jacobs ACKNOWLEDGMENTS
nor Garretón developed analytical tools for the This work is part of an ongoing fondecyt project
study of their chosen object of study. To be more (11110450) and was made possible by the enthusi-
precise, neither of them built a systematic corpus of astic collaboration of the students of second year of
study cases and as a result of this, as Hillier would architecture at the Universidad del Bío-Bío in Con-
put it, they remained prescriptively strong but ana- cepción (Unit 4, 2012) and by the expert assistance
lytically weak. of architect and renowned Chilean authorial pho-
By focusing on the study of people in space rath- tographer Nicolás Sáez.
er than on space itself, this paper represents a pri-
meval impulse towards the development of general REFERENCES
analytical rudiments for the further development of Araneda, C 2009, Dis-Information in the Information Age Cit-
the Jacobs´ lineage. Whether the urbanist´s own ob- ies. The Size of the American Block as an Urban Anachro-
ject of study is urban space or people remains too nism, VDM Verlag, Germany/UK/USA.
big a question to be answered in these pages. Araneda, C 2010, Urban Protophenomenon. Introducing
the Notion of Primordial Phenomenon in Urbanism,
CONCLUSIONS Proceedings of the eCAADe Conference, Zurich, Switzer-
Even at these early rudimentary stages, the Urban land, pp. 207-215.
Polaroid technique of urban analysis has demon- Calabrese, F., Kloeckl, K., and Ratti, C. 2007a, ‘WikiCity:
strated to be a most useful as well as didactical com- Real-Time Location- Sensitive Tools for The City’. Pro-
plement to the abstract techniques championed by ceedings of the 10th International Conference on Com-
the Lynch lineage advocates. It does not only allow puters in Urban Planning and Urban Management (CU-
the systematic exploration of a thoroughly under- PUM07). Available at: http://senseable.mit.edu/papers/
studied, parallel universe, to the one by them ex- pdf/2007_Calabrese_et_al_WikiCity_CUPUM.pdf [Ac-
plored. More important still, by being grounded in cessed: 02/05/2011].
experience, it renders unnecessary all speculation Calabrese, F., Kloeckl, K., and Ratti, C. 2007b, ‘WikiCity: Con-
regarding patters of people presence inherent in necting the Tangible And the Virtual Realm of a City’.
GeoInformatics, 10(8), pp. 42--45. Jacobs, J 1961, The Death and Life of Great American Cities.
Calabrese, F., Kloeckl, K., and Ratti, C. 2007c, ‘WikiCity: real- The Failure of Town Planning, Penguin Books, UK.
time location- sensitive tools for the city’. Conference on Luhmann, N: `Social Systems´, Stanford University Press,
Communities and Technologies - Digital cities. Available California, USA, 1995.
at: http://senseable.mit.edu/papers/pdf/2007_Cala- Lynch, K 1960, The Image of the City, MIT Press, Cambridge
brese_et_al_WikiCity_Digital_Cities.p df. [Accessed: MA.
02/05/2011]. Maertens, H 1884, Der Optsiche Masstab in den Bildenden
Calabrese, F., Kloeckl, K., and Ratti, C. 2007d, WikiCity. ‘Real- Kuensten, Wassmuth, 2nd edition, Berlin.
time urban environments’. IEEE Pervasive Computing, Morello, E & Ratti, C 2008, ‘A Digital Image of the City: 3-D
6(3), 52-53. IEEE Computer Society. doi: http://doi. Isovists and a Tribute to Kevin Lynch’. Available at:
ieeecomputersociety.org/10.1109/MPRV.2007.69 http://senseable.mit.edu/papers/pdf/2008_Morello_
Calabrese, F., Colonna, M., Lovisolo, P., Parata, D. and Ratti, Ratti_Environment%20and%20Planning%20B.pdf [Ac-
C. 2007e, ‘Real- Time Urban Monitoring Using Cellular cessed: 02/05/2011].
Phones: a Case-Study in Rome’. Available at: http:// Mumford L & Miller D (ed.), 1986, The Lewis Mumford Reader.
senseable.mit.edu/papers/pdf/2007_Calabrese_et_al_ Pantheon Books, USA.
Rome_Unpub.pdf [Accessed: 02/05/2011]. Ratti C, 2005, ‘The lineage of the line: space syntax parame-
Calabrese, F., Kloeckl, K. and Ratti, C 2008, ‘Wikicity: Real- ters from the analysis of urban DEMs’, Environment and
Time Location-Sensitive Tools for the City’. In Foth, M. Planning B: Planning and Design 32(4) 547 – 566.
(Ed.) Handbook of Research on Urban Informatics: The Batty M, Jiang B, Thurstain-Goodwin M, 1998, ‘Local move-
Practice and Promise of the Real-Time City. Hershey PA, ment: agent-based models of pedestrian flow’’, WP 4,
IGI Global. Centre for Advanced Spatial Analysis, University Col-
Garretón, J 1975. Una Teoría Cibernética de la Ciudad y su Sis- lege London, http://www.casa.ucl.ac.uk
tema. Ediciones Nueva Visión, Buenos Aires, Argentina. Seamon, D and Zajonc, A. (eds.) 1998, Goethe’s Way of Sci-
Hall, E 1969, The Hidden Dimension, Anchor Books, New ence. A Phenomenology of Nature, SUNY, USA.
York. Seamon, D 2012, “Jumping, Joyous Urban Jumble:Jane
Hall, E 1973, The Silent Language, Anchor Books, New York. Jacobs’s Death and Life of Great American Cities as a
Hall, E 1976, Beyond Culture, Anchor Books, New York. Phenomenology of Urban Place”, Space Syntax Journal,
Hillier, B 1996, Space is the Machine, Cambridge University, 3(1), pp.139-149.
UK.
Collaborative and Human Based Performance Analysis
Mathew Schwartz
Advanced Institutes of Convergence Technology, South Korea
www.smart-art.org
mat@cadop.info
Abstract. This research presents methods for simulation and visualization of human
factors. This allows for a performance based analysis of buildings from the local human
scale to the larger building scale. Technical issues such as computational time and
mathematically describing a buildings geometry are discussed. The algorithms presented
are integrated in a 3D modeling software commonly used in design and architecture
through a plugin.
Keywords. Universal design; human analysis; collaboration; education; disability.
INTRODUCTION
In general, the field of Building Information Mod- are not usually considered necessary to include. For
eling (BIM) is related to structural or environmental one, many important aspects of a building in re-
analysis. This is likely due to the main users of BIM: gards to humans are regulated in laws, for example,
Architecture, Engineering, and Construction (AEC). the American Disabilities Act (ADA) (Dept. of Justice,
By utilizing BIM approaches these fields are able to 2010). Similarly, many of the components related to
reduce human error and cost from design to con- humans are standardized, which has been needed
struction (Azhar et al., 2008). While BIM can include as the AEC community is not focused on researching
many forms of analysis, human factors have not ergonomics or biomechanics, large research fields in
traditionally been included. Additionally, the defi- themselves. Additionally, excluding emergency situ-
nition of BIM is still being debated and developed ations, the cost of an error in construction is much
(Eastman, 2008). For this reason, this paper concen- greater than a door knob being difficult to use.
trates on the term building performance to discuss While these reasons are valid, they only represent
the relationship of human factors to overall building the minimum of what design can be.
analysis. However, this term is only for clarity, and The philosophy of Universal Design brings to
the work presented is wells suited to be integrated question the lack of human factor analysis for build-
as a feature of BIM software. ing performance. Exclusion of human factors from
As with BIM, building performance is largely fo- the design and analysis stage while relying on stand-
cused on structural and environmental factors that ards and prescribed law can be stigmatizing and
can be quantified. When the human is included in may need to be fixed later, demonstrating the failure
an analysis of building performance it is usually re- of design (Story, 1998). While the ADA is specific to
lated to the environmental effects of a human, or disabilities, Universal Design is meant to benefit all
multiple humans, such as heat transfer or acoustical people. Universal Design has become part of many
properties (Mahdavi, 2011). There are a few reasons curriculums in architecture schools (Vance, 2012),
the ergonomic and physiological aspect of humans however, understanding the problems with designs
related to human factors can require costly physi- ronment by comparing the reach ability to desired
cal experiments or communication with experts in reach locations. The inverse kinematic (IK) system
other fields. Schubert et al. (2011) present a method used to control the manikin arm was developed in
for integrating physical tools with the design work- MAYA, however, additional research into disability
flow, and Maver et al. (2001) demonstrate the ways reach would greatly improve the simulations. Cur-
in which universal design concepts can be explored rently one of the most used and advanced simula-
through virtual reality. Although both of these pre- tion tools for ergonomics uses a reach envelope, vis-
sent alternative methods to real world experimenta- ualizing the extent of a persons reach (Blanchonette,
tion, they still require physical space and money. 2010). This is generally okay when a user is able to
This paper is meant to function as a reference rotate to one side or another, however, if the user
on the types of quantifiable human factors, an ap- has a specific type of reach ability a more complete
proach to education, and a method for collaboration map needs to be simulated.
between the complicated fields of ergonomics and A voxel volume was used to create a complete
biomechanics with architecture and design. A 3D 3D map of the reach ability. Each voxel is given a
Manikin plugin is presented with a graphical user value based on the ability for the IK solver to find
interface allowing access to the underlying algo- a solution for the voxels around the initial voxel as
rithms that simulate and visualize human factors of seen in equation (1).
the manikin with the space at different scales. Algo-
rithms written in python are presented for five main
areas: Reach, Vision, Zone, Search, and Movement.
Aspects of biomechanics software are integrated
with the plugin to create a platform for collabora-
tion. To cohesively demonstrate the application of
each component a manikin in a wheelchair is used.
METHODOLOGY OF COMPONENTS (1)

There are many components to building design that Physically, this means if a persons hand is in an
effect human performance. The methods described initial position, that position is ranked based on how
in this section describe both human factors that many ways the hand can leave that position. The
should be addressed as well as methods in which to summation over 1 can be modified for a more elabo-
perform the simulation, however, this list is not ex- rate or specific analysis of the reach ability volume.
haustive. The methods describe the algorithms used For instance, by changing 1 to the distance from the
in the plugin written for MAYA, but can be applied in shoulder multiplied by gravity, areas further away
almost any 3D modeling tool. from the body will be given lower scores than those
closer. In the case of spinal cord injuries, this value
Reach can reflect the joint angles required to reach a posi-
While every person is slightly different in their ability tion. The visualization corresponds to the simulation
to reach, the effects of a disease or spinal cord injury results such that opaque green voxels are the high-
on reach ability (Jacquier-Bret et al., 2008) is impor- est ranking and transparent red voxels are the low-
tant to know. A large problem with the ADA stand- est ranking. The method for coloring voxels, without
ards is the assumption that a person in a wheelchair transparency, is similar to robotics work in which the
is only limited in leg functionality. As this is not true, workspace must be defined and visualized (Zacha-
the ability to simulate reach ability for different users rias et al., 2007). The visualization can be seen in the
allows for a performance based analysis of the envi- results section.
Vision if the light is within the vertical 45 degree angle is
Designing with a persons vision in mind requires seen in (3).
two aspects, items that should be seen, and items
that should not. The former case includes items (3)
such as signs and navigational cues, while the latter Second, the location of the light relative to the
includes direct lighting. An important aspect of vi- head must be found. For this, a reference point is
sion is neck mobility. If for instance a person is in a placed at any distance in front of the head in the di-
wheelchair, the range of movement in the neck may rection it is facing. A triangle is constructed from the
be limited. In this case, as well as in many elderly head position, reference position, and light position
people, it cannot be assumed that a building occu- with the light projected onto the head up axis, as-
pant will turn their head to view an item. In the case suming y is the up axis (4).
of direct lighting, many times a person will be un-
comfortable, or worse, unable to see in the distance
resulting in a dangerous situation.
The importance of vision is well known in archi- (4)
tecture and design. In Tilley et al. (2001), a diagram where H is the Head position, R is the reference po-
showing many aspects of the human vision is used sition, and L is the light position. Using the law of
as a reference for designers. The problem with the cosines, the side angle is calculated and checked
diagram is the difficulty in understanding and trans- against the 94 degree limit (5).
lating the content to ones own design within both
practice and education. The diagram can be repre- (5)
sented in 3D space by referencing both top and side The lights found to be in the disabling glare
maximum angles. For displaying the vision regions zone can be marked in the model and the total
each one of the four points, referred to as Limit, that number can be displayed for the designer.
describe the region are input at both an initial dis-
tance and the distance to where the visualization Zone
should end. Equation (2) shows the formula to calcu- The space around a person can be classified in mul-
late where each point should be placed. tiple ways. There is the space someone needs to
be physically present, space that makes someone
comfortable, and space in order to complete a task.
(2) The latter two are most often related to psychology
In order to analyze the building for direct light- as well as physiology. The importance of the space
ing at a specific point the location of the manikin around a person can be seen in design reference
head as well as direction is calculated in regards to books (Zelnik and Panero, 1979) as well as in research
each light. According to the diagram in Tilley, the papers (Lantrip, 1993). When designing the spatial
disability glare zone is 45 degrees above the persons needs of a wheelchair the designer must understand
eye level and the extent a person can see to the side the many situations that exist. For example, if a hall-
is 94 degrees. The domain is found with two calcu- way is designed to be wide enough for a wheelchair
lations. First, the angle above the eye level is found to pass through, it may not be wide enough for the
creating a right triangle between the head position wheelchair to turn around, or more commonly, for a
and light position with the hypotenuse being the person in the opposite direction to pass by.
distance between the two points. The adjacent side The implementation of a spatial zone is relative-
is the horizontal distance and the opposite side is ly easy in a CAD program. The spatial zone needed
the vertical distance. The algorithm for determining can be visualized by a transparent cylinder around
the manikin as seen in the results section. In the Figure1
case of multiple people more than one cylinder can Left: Key locations exist in Car-
be displayed. The coordinates in the mesh are gen- tesian space. Edge values are
erated the same as in (2) where Limit is defined as the distance between them.
the distances from the center of the manikin. As the Right: If the building walls
generated mesh contains a Cartesian domain, it is are included it is easy to see
possible to analyze environmental interference in the problem with calculating
regards to the generated geometry. distance directly.
“floor”. Anything following that name is irrelevant
Search and can be used for the designers own organization.
Out of many human factors that can be included The second assumption is that the manikin is placed
in building analysis, the distance and way in which over any valid ground. If these conditions are satis-
someone moves throughout is one of the most dif- fied, the algorithm will send a ray from the top of the
ficult to integrate, yet is extremely important. While manikin head to the floor and return the Cartesian
laws exist for wheelchair ramps and elevators, know- position. Each valid ray is checked against the mini-
ing how a building layout or size affects the oc- mum required space, in this case, a wheelchair turn-
cupants is required to design above the minimum. ing radius. The algorithm then expands outward and
There are many forms of analysis of human factors continues to store valid node locations as seen in (6).
that are useful at the human scale, including egress
and accessibility. Two components needed to ana-
lyze movement throughout a building are: math-
ematically describing the building and creating the
path of movement. From this it is possible to run nu-
merous simulations of human factors to analyze the
path created. The current implementation refers to
the path as the energy required.
The easiest way to describe a building using a
node graph is to make landmark areas nodes and
create connections, known as edges, between them. (6)
When the fastest way between two nodes is desired, The edge values are the key to analyzing the
a search algorithm calculates the edge values from building. As each node is created, so are the edge
one point to the other (Dijkstra, 1976). In general, costs to each of the available nodes surrounding it
the distance between each landmark is the value (7).
given to the connection. However, the distance
between each landmark must be the length of the
route from start to end, not the Cartesian distance.
This creates a problem when attempting to translate
a 3D model of a building into a searchable graph
without individually measuring each path (Figure 1).
Using the internal raytracing function of MAYA,
a technique was developed for mathematically de-
scribing a 3D model of a building. The first assump-
tion is that any geometry that should be considered
as ground should begin with the naming convention (7)
Figure 2 tegrating human factors in design workflows is the
The GUI inside of PYQT. The extra knowledge of human biomechanics required
tabs can be seen on the top. of designers. As both a teaching and collaboration
tool, this research demonstrates a method for inte-
grating the two fields.
Many types of analysis in biomechanics use joint
angles. To integrate this workflow a python library
for creating matlab style graphs (Hunter, 2007) was
implemented within the plugin. As some design
modeling tools do not have animation features, all
time related elements are stored in comma separat-
ed value (CSV) files. The file structure implemented
is the same as those from the exported joint angles
of biomechanics software.
Graphical User Interface

Combining the algorithms together and linking
them to the 3D modeling software is done through
callbacks and signals within a graphical user inter-
face (GUI) (Summerfield, 2007). The GUI is designed
to give the designer access to all of the variables
Altering the edge cost can be useful to analyze within the algorithms. Each component is given a
different situations such as the path of a wheelchair separate tab in the GUI with a full range of settings
or a person with crutches. For example, implement- and variables (Figure 2).
ing this algorithm for a wheelchair would be done
by making any edge that is at a slope, larger than RESULT
the maximum required by law, an infinite value. By The implementation of the algorithms resulted in a
returning a path through the nodes, the physical cohesive tool that integrates with a 3D model of a
distance can also be calculated. As each node has a building. A variety of visualizations, simulations, and
total value assigned to it by the connecting neigh- analysis are available for the designer to use. As the
bors a visual representation can also be created. building is being designed the only requirement
of the designer is to place the manikin in a desired
Movement location. Once the manikin is placed, the designer
As previously mentioned, one problem with fully in- can use any of the default settings to run the simu-
lations. As shown by Eriksson et al. (2000), there is
Figure 3 great value in participatory planning. This plugin
The MAYA window with a new allows for participatory planning by creating a sec-
camera perspective of the ond window in which the view of the manikin is dis-
manikin in the lower right. played (Figure 3).
As it is not realistic to expect all architecture
schools to have time for real world experiments, or
access to wheelchairs, it is important to have a vir-
tual method of exploring the same topics. During
the early stages of this research a Universal Design
Figure 4
Left: Student points out
objects in the way of naviga-
tion. Top-right: the path
from manikin to a location is
drawn. Bottom-right: objects
placed over the ground are
accounted for, creating a path
that avoids the object.
class led by Sean Vance at the University of Michigan lems needed to be addressed. Some of these issues
was documented (Vance, 2012). The students were include obstacles in the way of a wheelchair (Figures
not aware or directly influenced by this research. 4 and 5), the line of sight from a person in a wheel-
Multiple physical experiments were conducted by chair (Figure 6), and items out of reach (Figures 7
the students to simulate the relationship of disabili- and 8).
ties to the built environment around them. While While it is possible to quantify all of the present-
it is very useful for students to have a hands-on ed human factors relative to a building model, the
knowledge it may not always be possible. The tool largest scale quantifiable factor is in navigation. The
presented here is capable of simulating and visual- graph search algorithm is very flexible and can be
izing the same ways in which students felt the prob- used for a variety of situations, one of which is seen
Figure 5
Left: Student points out
problems with the distance
between toilet and wall
for turning radius. Right:
Wheelchair dimensions are
visualized in a 3D model of
a room.
Figure 6
Left: Student draws line of
sight to point out the difficulty
in communication due to the
high partition. Right: Two
different vision cones are visu-
alized. Two perspectives are
shown, one from the building
participant and a normal 3D
model view.
Figure 7
Left: Student shows problems
with the placement of differ-
ent items in a room. Right: The
reach ability map is simulated
and visualized with transpar-
ency and color.
in Figure 4. Another aspect of the graph search is someone to move from one point to another. In the
the ability to analyze the amount of time it takes for case of a wheelchair, depending on the type of dis-
Figure 8
Left: Student drawing angles
of human body to analyze
limits. Right: Motion capture
data of reach is displayed
through a graph.
ability and wheelchair, a person will move between steeper incline, the path generated will reflect the
~50 and ~80 meters per minute (Beekman et al., best method. This situation is ideal for bringing in
1999). Taking data from a subject with paraplegia in a biomechanics expert, or in the case of education,
a standard wheelchair a rate of 75 meters per minute biomechanics students. As the biomechanics side
can be used for analysis. When a designer runs the is able to analyze the movement of a person during
search algorithm the GUI will display the amount of wheelchair propulsion and quantify the results, the
time it takes to move along that path (Figure 9). design of the ground floor can be changed. Using
In addition to the speed of movement it is this alongside the colored visualization shown in
possible to calculate the energy expenditure of a Figure 8 creates an opportunity for informed itera-
movement. While there is a lack of human subject tive design.
testing that can give an accurate simulation for all
terrain, some basic estimations can be made. This CONCLUSION
approach can be used when designing a ramp and This paper presents a variety of functions, algo-
the designer is deciding if the ramp should be short, rithms, and systems that can better integrate human
with the maximum allowed slope, or longer with a factors with the design workflow. Each algorithm
lower slope. As the search algorithm will find the has potential to be both improved with speed and Figure 9
lowest cost method to get to the end point, if the expanded on for functionality. The significance of Left: Student demonstrates
algorithm creates a higher cost of movement for a the work is not limited to BIM and Building Analy- the narrow hallway does not
allow for a wheelchair to turn
around. Student notes the
length of the hallway with no
outlets can cause arm fatigue.
Right: GUI Displays estimated
time to travel using a wheel-
chair. Color coded ground
shows results of simulated
values. Red areas show either
narrow hallways or a steep
slope.
sis, but can bring human factors to the forethought gineering, IEEE Transactions on 8(1), 94-106.
of designers and architects, greatly influencing the Hunter, J 2007, ‘Matplotlib: A 2D Graphics Environment’,
style in which designs are created and simultane- Computing in Science Engineering 9(3), 90-95.
ously benefiting the building occupant. Recognizing Jacquier-Bret, J, Rezzoug, N and Gorce, P 2008, Synergies
the impossibility of designers to be a master of eve- during reach-to-grasp: A comparative study between
ry field related to human ability, and the advantages healthy and C6-C7 quadriplegic subjects, in ‘Engineer-
of expert collaboration, a system has been presenting in Medicine and Biology Society, 2008. EMBS 2008.
ed in which biomechanics, architecture, and design 30th Annual International Conference of the IEEE’, pp.
can collaborate. Although some schools have been 5366-5369.
able to implement physical experiments to teach Dept. of Justice 2010, 2010 ADA Standards for Accessible De-
students Universal Design, many schools do not sign, Dept. of Justice, USA. http://www.ada.gov/regs20
have the resources for these lengthy and costly ex- 10/2010ADAStandards/2010ADAstandards.htm
periments. Integrating human factor simulations in Lantrip, D B 1993, ‘Environmental Constraint of Human
design programs would allow for students to learn Movement: A New Computer-Aided Approach’, Pro-
the same basic principles. ceedings of the Human Factors and Ergonomics Society
Annual Meeting 37(15), 1043.
ACKNOWLEDGEMENTS Mahdavi, A 2011, The Human Dimension Of Building Per-
I thank Sean Vance for allowing me to document his formance Simulation, Proceedings of Building Simula-
Universal Design Class, as well as Karl Daubmann tion 2011, Sydney, Australia, pp. K16-K33.
and David Lantrip for their valuable feedback. I Maver, T, Harrison, C, Grant, P, de Vries, B, van Leeuwen,
would also like to acknowledge the contribution of J and Achten, H (eds), 2001, Virtual environments for
both Forest Darling and Robert Van Wesep for their special needs - changing the VR paradigm, Springer, pp.
help with programming and algorithms. 151-159.
Schubert, G, Artinger, E, Petzold, F and Klinker, G 2011,
REFERENCES Tangible tools for architectural design : seamless inte-
Azhar, S, Hein, M. and Sketo, B 2008, Building information gration into the architectural workflow, Proceedings of
modeling (BIM): Benefits, risks and challenges, Proceed- ACADIA 2011.
ings of the 44th ASC National Conference. Story, M F 1998, ‘Maximizing Usability: The Principles of Uni-
Beekman, C E, Miller-Porter, L and Schoneberger, M 1999, versal Design’, Assistive Technology 10(1), pp. 4-12.
‘Energy Cost of Propulsion in Standard and Ultralight Summerfield, M 2007, Rapid GUI Programming with Python
Wheelchairs in People With Spinal Cord Injuries’, Physi- and Qt: The Definitive Guide to PyQt Programming, Pear-
cal Therapy, 79(2), 146-158. son Education.
Blanchonette, P 2010, ‘Jack Human Modelling Tool: A Re- Tilley, A and Henry Dreyfuss Associates 2001, The Measure
view’, DEFENCE SCIENCE AND TECHNOLOGY ORGANI- of Man and Woman: Human Factors in Design, Wiley.
SATION VICTORIA (AUSTRALIA) AIR OPERATIONS DIV. Vance, U S (ed) 2012, Equilibrium: Universal Design Primer,
Dijkstra, E 1976, A discipline of programming, Prentice-Hall, MPublishing.
Incorporated. Zacharias, F, Borst, C and Hirzinger, G 2007, Capturing ro-
Eastman, C 2008, BIM Handbook: A Guide to Building Infor- bot workspace structure: representing robot capa-
mation Modeling for Owners, Managers, Designers, Engi- bilities, Intelligent Robots and Systems, 2007. IROS 2007.
neers and Contractors, Wiley. IEEE/RSJ International Conference on, pp. 3229-3236.
Eriksson, J, Ek, A and Johansson, G 2000, ‘Design and evalu- Zelnik, M and Panero, J 1979, Human Dimension and Inte-
ation of a software prototype for participatory plan- rior Space: A Source Book of Design Reference Standards,
ning of environmental adaptations’, Rehabilitation En- Crown Publishing Group.
Visibility Analysis for 3D Urban Environments
Research development and practical application

Anastasia Koltsova1, Bige Tunçer2, Gerhard Schmitt3
1,3
Swiss Federal Institute of Technology, Switzerland, 2Singapore University of Technology
and Design, Singapore
1,3
http://www.ia.arch.ethz.ch, 2http://asd.sutd.edu.sg
1
koltsova@arch.ethz.ch, 2bige_tuncer@sutd.edu.sg, 3gerhard.schmitt@sl.ethz.ch
Abstract. This paper presents a visibility analysis tool for 3D urban environments and its
possible applications for urban design practice. Literature exists for performing visibility
analysis using various methods and techniques, however, tools that result from such
research are generally not suitable for use by designers in practice. Our visibility analysis
tool resides in Grasshopper, Rhino. It uses a ray casting method to analyze the visibility
of façade surfaces from a given vantage point, and of a given urban setting, in particular,
buildings and roads. The latter analysis provides information on the best visible buildings/
building facades from segments of roads. We established a collaboration with a practicing
architect to work on a design competition together, using this tool. The paper elaborates
on the visibility analysis methods, presents the tool in detail, discusses the results of our
joint work on the competition, and briefly reflects on the evaluation of the use of the tool
by design practitioners.
Keywords. Visibility analysis; pedestrian design; urban space quality; design practice.
INTRODUCTION
This paper presents a visibility analysis tool for 3D methods of analysis using terms such as “visual ab-
urban environments and its possible application for sorption”, “visual corridor” or “visual intrusion” (Lynch,
design practice. Visual perception of space is one of 1976). A view analysis example is an ‘isovist’ analysis
the factors that defines spatial experience and cog- which measures a volume of space that is visible
nition of architectural/urban space. Analyzing the from a single point in space. The term was introduced
impact of design decisions on perception of space by Tandy in 1967 (Tandy, 1967). This research gave
may help to significantly improve the quality of ur- raise to the development of a multitude of methods
ban developments (Bittermann et al., 2008). for quantitative analysis of space perception. Ben-
Many design and architectural researchers inves- edikt was the first who introduced a set of analytic
tigated the relation between urban space morphol- measurements of isovist properties (Benedikt, 1979).
ogy and its experiential qualities as perceived by us- In the field of landscape architecture and planning
ers. Among them are Appleyard et al. (1964), Lynch there is a similar concept called “viewshed” (Turner et
(1960), Benedikt (1979), and Thiel (1961). Kevin Lynch al., 2001), which analyzes the visibility of an environ-
stipulated on the importance of view analysis and mental element from a fixed vantage point.
Quantitative methods for visibility analysis can analysis performed using different computational
be roughly divided into the following categories: a) implementations. The research underlines that dif-
scientific landscape evaluation (LE) provides meth- ferent computational methods tackle different as-
ods for ‘quantitative description of natural landscape pects of spatial analysis and provoke different ways
visual quality or impact prediction’ (these approach- of thinking about a problem. Therefore, a computa-
es do not consider human perception); b) methods tional tool can become a flexible element that sup-
such as ‘isovist’ concentrate on the visibility of an en- ports creative thinking during design process.
vironmental element from a fixed vantage point and Turner et al. (2001) uses visibility graph method,
neglect the landscape resources (He et al., 2005). first introduced in De Floriani et al. (1994), for spa-
The most common examples of utilizing visibil- tial analysis of architectural space. This research in-
ity analysis methods and tools in the field of urban vestigates how visual characteristics of a location
design are analysis of visibility from important (stra- are related and how this can have a potential so-
tegic) points (e.g., large transportation hubs, major cial interpretation. The graph representation that is
public spaces, etc.) to dominants (e.g. tall buildings, used incorporates isovists to derive a visibility graph
monuments, etc.), which can help to improve navi- of mutually visible spots in a given spatial layout
gation of pedestrians in the city. Another case is the (Turner, 2001). This leads to the definition of some
preservation and/or strategic use of views to natu- measures that describe both local and global spatial
ral landscape elements such as a river or park. This properties that may relate to the perception of the
is especially relevant to high-density urban areas built environment.
that are still undergoing an extensive development The literature discussed above presents research
process, such as Moscow, Hong Kong or Singapore. for performing visibility analysis using various meth-
Uncontrolled development in such big cities leads ods and techniques. An issue that arises concerning
to fragmentation or complete blockage of views to the tools that result from such research is that the
valuable landscape resources, which are more de- tools are not suitable for use by designers in prac-
sirable for people than man-made structures (He et tice. Most designers do not have knowledge and
al. 2005). This results in a drop of real estate values skills of programming, or using specialized software.
and deterioration of city fabric. In this context, He et This has several reasons, e.g., time pressure in a de-
al. (2005) presents an approach to visual analysis of sign project. Designers also don’t tend to use spe-
high density urban environments, which quantita- cialized analysis software during the early design
tively integrates human visual perception (analysis phase, because these are difficult to use, and the
from a fixed vantage point) with the visible land- model usually needs to be exported and imported
scape resources (LE), using GIS as database and tech- back and forth between the analysis and modeling
nical platform. This approach can help architects software. Performing analysis on the model in the
to take more informed decisions at an early design familiar modeling environment would increase the
stage regarding the preservation of valuable land- usability of these tools. Furthermore, developing
scape resources and view corridors. Another exam- the tools with their use by designers in mind would
ple is the work described in Fisher-Gewirtzman et al. increase their usability. Our research development
(2005), which compares various coastal urban mor- aims to introduce visibility analysis tools in the ur-
phologies with the variation of density levels and ban design practice.
their influence on the visibility of the water front. The most recent visibility analysis methods that
The assumption is that the morphological results designers and architects use today rely heavily on
can be used as criteria for future urban planning. computing power. Some of the well-known analysis
Do and Gross (1997) present a set of tools for software such as, Ecotect, Space Syntax and ArcGIS
spatial analysis among which are tools for visibility offer methods for visibility analysis. However, these
Figure 1
Analysis of visual pollution by
advertisement billboards.
offer very limited methods for visibility analysis of boards and other large signs create a view pollution
building facades, or as we call it in this paper, anal- of building façades on this street. The definition of
ysis of 3D urban environments. In addition to that, view pollution may be interpreted differently in dif-
all this software are standalone applications that do ferent contexts. For instance, billboards and signs
not support 3D modeling. Every new design ver- characterize Times Square in New York, as these
sion must be imported and analyzed in a modeling form the identity of place in this context. However,
software. This approach does not support dynamic on this pedestrian street in Moscow, uncontrolled
manipulation of the design model and slows down placement of advertisement billboards results in a
the design process. We developed a tool for visibility complete blocking of 18th century historic heritage
analysis in Grasshopper, parametric plug-in for the buildings. Furthermore, the scene created by the
Rhinoceros modeling platform. Rhino is widely used signs do not contribute positively to the identity of
among architects and designers today. Our tool can the place, on the contrary, it diminishes the overall
be used to analyze models directly in Rhino, and quality of public space.
dynamic changes can be made and revised models In our current work we aim to investigate poten-
analyzed by the tool in real time. Our tool uses a ray tial uses of our tool for design practice. Therefore, we
casting method to analyze the visibility of façade established a collaboration with a practicing archi-
surfaces. tect to work on a design competition together, us-
Our tool combines two possibilities, referring to ing the 3D urban settings visibility analysis tool.
the two quantitative methods for visibility analysis This paper elaborates on the visibility analysis
described earlier in this section: a) analysis of visibil- methods, presents the tool in detail, and discusses
ity from a given vantage point and; b) visibility anal- the results of our joint work on the competition. We
ysis of a given urban setting (in particular, buildings end the paper with a brief evaluation on the use of
and roads). The latter analysis provides information the tool by design practitioners, and directions for
on the best visible buildings/building facades and future work.
segments of roads that ‘see’ most of the buildings.
The view pollution analysis became a first case THE VISIBILITY ANALYSIS TOOL
study for the tool (Koltsova et al., 2012). An example This section elaborates on the functionality of the
that we analyzed is one of the pedestrian streets in visibility analysis tool and its development process.
the historic center of Moscow, Russia (Figure 1). Bill- We used Grasshopper, the parametric environment
for Rhinoceros, as the development platform. In Figure 2
Grasshopper it is possible to write your own code Custom Grasshopper
in C# .NET or VB .NET and create a custom tool (or component for visibility
component) that performs the specific function. analysis. Inputs: road network
Such custom components require potential users (N), building geometry (B),
(architects and urban designers) only to know what mesh tessellation (M), terrain
to feed in as an input (curve, points, geometry, etc.) analysis (optional, (T)), max
and what the output would be. We developed two viewing distance (D), max
custom tools that perform the following functions: view angle (A).
visibility analysis of building geometry, and visibil-
ity analysis of the road network (Figure 2). Visibility ceeds to the analysis of the whole mesh. Intersec-
analysis uses a ray casting method. The algorithm tion calculation of the ray and bounding box takes
requires the following inputs: less time then ray-mesh intersection, which helps to
• building geometry as Breps considerably reduce calculation time.
• terrain as a mesh surface The main parameters that the tool uses are:
• road network as curves or polylines • the view distance from a view point to a façade
The algorithm converts the building geometry surface,
(Breps) into a mesh. The possibility to define mesh • maximum visual angle (vertical and horizon-
tessellation for building and terrain surface geome- tal), and,
try individually is embedded in the tool. This is done • angle from the view point to a façade surface.
due to the difference in scales and analysis preci- For different design tasks specific parameters
sions for the two geometry types. are retrieved by the tool. For example, for the analy-
The road curves are selected automatically by a sis of city dominants (tall buildings or city monu-
“Pipeline” component (Figure 2). This is the in-built ments), the tool solely checks if the object is visible
Grasshopper component that allows for automatic or not from a certain point or path (Figure 4a). Con-
selection of a specified type of geometry by object sidering factors such as the visibility of city domi-
layer. The road network is split into segments and nants during the design of new public spaces can
at intersection points. The length of every segment improve navigation within a city. For pedestrians it
can be defined according to the design scale. The is easier to choose the direction of movement if they
smaller the segment the more precise the analysis see a dominant and know the location of it. Visual
is. The mid points of segments become visibility connections in the city also help to create better
nodes. The algorithm generates rays between mid
points of the curves and mid points of mesh faces of Figure 3
building/terrain geometry. Then, the algorithm re- Analysis results (best visible –
turns intersection points between vectors and each yellow; non-visible – white),
face’s mid points and checks if there is any obstruc- viewing points are distributed
tion between the viewing point and façade surface. along the pedestrian walks
Depending on the result it assigns each face a color: with a span of 20 meters.
gradient between yellow (best visible and blue –
worst visible; white – non-visible) (Figure 3).
In order to save calculation time we use bound-
ing box of building meshes at first iteration step to
check for possible intersections. If generated ray
intersects a bounding box then the algorithm pro-
Figure 4
In red – viewing point,
gradient shows the best/worst
(yellow/blue) visible building
facades:
a) Tool checks for visible/non
visible buildings – true or false
b) Distance to façade surface
is added c) Distance and angle
to façade surface are added
d) Direction of pedestrian
movement and its view angle connected public spaces (network instead of iso- bility tool it is possible to set a starting point and an-
is added lated spots). alyze how far one can get by walk/car or bus within
For the analysis of how detailed pedestrians can a certain time period. In this case destination points
see the facades and which are the most exposed are the mid point of previously generated segments
surfaces, the maximum distance and angle from of the road network (refer to the visibility tool de-
a view point to façade mesh faces is added. An an- scription before). The input parameters for this com-
gle closer to 90 degrees and less distance to façade ponent are:
means better visibility. Gradient illustrates the best/ • max walking distance, or;
average/worst visible façade surfaces (Figure 4b, c). • time and speed by car/walk/public transport
For the moment the influence of distance and angle (in which case max walking distance is calcu-
on the analysis result is 50/50. Naturally, the impor- lated based on these two parameters).
tance of each of the parameters can vary depend- We use the graph component to analyze structure
ing on the design task. Therefore, we plan to further and create topology of the road network (Figure
evaluate the tool with architects and revise it based 5). This information in turn is used by the Dijkstra’s
on their feedback. We have already added additional algorithm to calculate the shortest path between
constraints such as the horizontal and vertical view starting and destination points.
angles to be able to analyze what a person can see Combining the two types of analysis methods
while walking in a specific direction (Figure 4d). It is provides the possibility to analyze how far one can
possible to activate or deactivate the functions de- go within a certain time span and what one can
scribed above by right-clicking the title of the com- see while walking this route (Figure 6). Figure 7(a)
ponent and checking/unchecking them (angle to shows the accessibility analysis results and (b) what
surface, distance to surface, one direction). This is a one can see while walking this path. The resulting
feature that can be programmed by a tool developer path is used for the visibility analysis of best visible
in Grasshopper. façade surfaces from the path. Rays are created be-
In our work we combined two types of urban tween the road segment and building mesh faces. If
Figure 5 analysis: visibility and accessibility. With the accessi- a mesh face is visible from the road segment then
Custom component for the algorithm assigns a segment ID to the mid point
accessibility analysis. Input of the mesh face. The more segments “see” a certain
parameters: network topology mesh face the higher the mesh face’s visibility value
(G), starting point of move- becomes (in terms of color: yellow – best visible,
ment (P), speed (V), duration/ blue – worst visible, white – non-visible).
time of movement (T), max Using our tool it is also possible to analyze best
walking distance (D). visible buildings. In this case the algorithm stores
building IDs instead of individual mesh faces and Figure 6
analyzes what are the buildings that most of the Combination of accessibility
road segments can “see”. The same logic applies to and visibility analysis custom
road segments. The more buildings/mesh faces a components.
road segment can “see” the higher visibility value
(closer to yellow color) is assigned to it (Figure 7b, c).
Using the tool it is possible to analyze the visibil-
ity of a single building and the road segments that
can “see it” (Figure 7d). The algorithm principle is the
same, with the exception that the information of the
road segment is stored as a boolean (True/False).
PRACTICAL APPLICATION OF THE TOOL EVALUATION

We worked with a practicing architect and applied During this collaborative work it was important for
our tools for a design competition. The brief was to us to understand what the challenges are that pre-
develop a design proposal for the transformation of vent architects from using parametric tools and
a former industrial area into a techno park. This new what should be changed (in the design process/tool
development is supposed to become a new local functionality) to integrate these better into the de-
economic center and attraction point. Therefore, its sign practice. We have conducted an interview with
visual perception from the main access points, such our partner where we obtained his opinion about
as bus stops, train station and highway, is an impor- the general situation and about using our tool in
tant aspect for analysis as it directly influences the particular. In general, the use of tools depends on
accessibility and integration of the new develop- the size of the office and the scale of the projects
ment within the local context. in this office. In Switzerland, rapid urban expansion
Figure 8 presents the design proposal. Accord- was not such a dominant issue until recently. Peo-
ing to the task set by the architect the tool checked ple do not yet feel the influence of it on their lives,
for visibility from important points around the pro- therefore, there are not that many design offices
ject site (points in orange), such as bus stop, city that deal with such challenges, and consequently,
public space and tram stop, to objects on the project have a need to upgrade their processes or tools. An-
site (i.e. design dominants such as conference cent- other, quite a straightforward reason, is that people
er, old factory chimney etc.). The analysis process is are used to certain software and associated process-
shown in Figure 8, right side. The idea of the archi- es that they establish in their offices and as there is
tect was to have a so called “target” matrix where no immediate need, they don’t want to change any-
he documents which elements should be seen from thing (or have time to change the routine). “As long
important view points according to his design con- as it works, its fine”.
cept (Table 1). The tool analyzes each new design The architect that we have been working with is
scenario and creates a new matrix (Table 2). This ma- one of the few whose office deals mostly with urban
trix is compared to the target matrix and if there are design projects. He works mainly using traditional
discrepancies, building shapes are adjusted to pro- approaches when designing, which are usually suf-
vide better visibility. For the moment this process of ficient. However, he remarks that the parametric ap-
changing the design based on the target matrix is proach is sophisticated, because it helps to resolve
manual. We are convinced this method is more in- many different challenges by allowing the architect
tuitive for an architect and provides more control on to systematically explore on a few issues at a time.
the design process. His feedback on our tool was that this tool becomes
Figure 7
a) Accessibility by walk
within 15 minutes, b) Vis-
ibility – what one can see
walking 15 minutes, and most
visible buildings (from all the
analyzed visibility points)
and road segments that “see”
most of the buildings, c) Most
exposed façade surfaces and
road segments that “see”
the most of the surfaces, d)
Building with index number
56 is analyzed, in gray road
segments that can “see’ the
building, black – not.
really useful as soon as the 3rd dimension comes project, then it becomes quite hard to estimate the
into play. Architects are trained and can estimate visual impact of the new design and its perception
what a person can see on the plan. However, when from different city locations. In his opinion, our tool
elements of context such as a complicated terrain can be used for the projects with, as he called it,
with high-density developments are a part of design “multiple levels and dimensions”. Based on the feed-
Figure 8
Top: project site and design
proposal; right: visibility
analysis from strategic points
(street view, tram stop, bus
stop, point in the city)
back we introduced additional function that allows the two geometry types (Breps – buildings, and sur-
for terrain surface visibility analysis. The meshing face - terrain). The parametric nature of the model
of the terrain surface can be controlled individually allows for an interactive change of the design form
due to the scale difference and analysis precision of in order to improve the visibility.
1 chimney 2 research lab 3 conference c. 4 admin offices 5 entrance N Table 1

A street view Reference matrix.
B tram stop
C bus stop
D point in city
1 chimney 2 research lab 3 conference c. 4 admin offices 5 entrance N Table 2

A street view Matrix for one design scenario.
B tram stop
C bus stop
D point in city
CONCLUSIONS Benedikt, ML 1979, ‘To take hold of space: isovist and isovist
This paper demonstrates the working process be- fields’, Environment and Planning B, 6(1), pp. 47-65.
tween a research group and a design practitioner. Bittermann, MS and Ciftcioglu, O 2008, ‘Visual Perception
The application of parametric tools for design prac- Model for Architectural Design, Journal of Design Re-
tice has the potential to establish a better commu- search, Vol. 7, pp. 35-60.
nication between design theory and practice, and Do, E Y L and Gross M D, 1997, Tools for visual and spatial
improve the quality of future urban spaces through analysis of CAD models, Computer Assisted Architectur-
better informed design processes. We will proceed al Design Futures, R Junge (ed), pp. 189-202.
with collaborative work with architects in order to Fisher-Gewirtzman, D Shach Pinsly, D Wagner, IA and Burt,
enhance our methods and adapt them to the needs M 2005, ‘View-oriented three-dimensional visual analy-
of the design practice. sis models for the urban environment’, Urban Design
In our future work we also plan to enhance the International, 10, pp 23-37
functionality of the presented tool by introducing He, J Tsou, JY Xue, Y and Chow, B 2005, ‘A Visual Landscape
additional inputs based on architects’ feedback. For Assessment Approach for High-density Urban Devel-
example, it is important to consider in the analysis opment’ Proceedings of the 11th International CAAD
the type of urban space and the type of movement Futures Conference, Austria, pp 125-134
it implies. In more specific terms, square/piazza or a Koltsova, A, Tunçer, B Georgakopoulou, S and Schmitt, G
shopping street implies lingering. The road between 2012, Parametric Tools for Conceptual Design Support
the transportation hub and business district would at the Pedestrian Urban Scale: Towards inverse urban
most probably have linear/directional type of move- design, Achten, H Pavlicek, J Hulin, J Matejdan, D (eds.),
ment. The perception of space by pedestrians large- Digital Physicality - Proceedings of the 30th eCAADe
ly depends on these factors and we will work on the Conference - Volume 1, pp. 279-287
ways to introduce this information into our paramet- Lynch, K 1976, Managing the sense of a region, MIT Press,
ric tools which would result in more accurate results. Cambridge.
Lynch, K 1960, The Image of the City, Cambridge, MIT Press,
ACKNOWLEDGEMENTS Cambridge.
The authors would like to thank architect Michael Tandy, CRV 1967, ‘The isovist method of landscape survey’,
Gueller for his valuable input during our collabora- in Symposium: Methods of Landscape Analysis, HC Mur-
tion and Lukas Kurilla for his support in the tool de- ray (ed), Landscape Research Group.
velopment. Thiel, P 1961, ‘A Squence Experience Notation for Architec-
tural and Urban Space’, Town Planning Review, V 32,
REFERENCES pp. 33-52.
Appleyard, D K Lynch, K and Myer, J 1964, View from the Turner, A Doxa, M O’Sullivan, D and Penn, A 2001, ‘From iso-
Road, MIT Press, Cambridge. vists to visibility graphs: a methodology for the analy-
Batty, M 2001, ‘Exploring isovist fields: space and shape in sis of architectural space’, Environment and Planning B:
architectural and urban morphology’, Environment and Planning and Design, 28 (1), pp. 103-121.
Planning B: Planning and Design, V 28, pp. 123-150.
Human Activity Modelling Performed by Means of Use
Process Ontologies
Armando Trento1, Antonio Fioravanti2
1
Sapienza, University of Rome
1
www.armandotrento.it, 2http://www.dicea.uniroma1.it
1
armando.trento@uniroma1.it, 2antonio.fioravanti@uniroma1.it
Abstract. Quality, according to Pirsig’s universal statements, does not belong to the
object itself, nor to the subject itself, but to both and to their interactions. In architecture
it is terribly true as we have a Building Object and Users that interact with it.
The problem we approach here, renouncing at the impossible task of modelling the actor’s
“libero arbitrio”, focuses on defining a set of occurrences, which dynamically happen
in the built environment. If organized in a proper way, use process knowledge allows
planners/designers to represent usage scenario, predicting activity inconsistencies and
evaluating the building performance in terms of user experience.
With the aim of improving both, the quality of buildings and the user experience, this
research explores a method for linking process and product ontologies, formalized to
support logic synchronization between software for planning functional activities and
software for authoring design of infrastructures.
Keywords. Design knowledge modelling; process ontology; knowledge management.
AEC INDUSTRY AND INFORMATION MODELS

The approach / methodology called Product Infor- through the re-modulation of its structural general
mation Modelling (PIM), historically consolidated “core” adjusting it with little variations relevant to
throughout the industrial world, is typical of serial specific production requirements.
systems. In the last decades, the PIM has proven its Through a slow process of technological trans-
robustness and effectiveness even in the most com- fer, still in early stages, PIM was introduced about
plex areas, characterized by the uniqueness of the ten years ago in the AEC industry, starting the so-
product, such as automotive, aero-spatial, etc.. called Building Information Modelling (BIM).
Its final object, towards which this PIM pro- Well known peculiarities of AEC industry con-
duction system, is realised by means of a unique stitutes an important challenge for BIM approach,
process, characterized by low seriality. In the last which in turn is a method / tool able to innovate this
decades, this production system revealed their ex- sector - central to the European economy - especial-
perimentations to be successful with repeatable, op- ly in a time of crisis like the current one.
timizable and personal solution, the so-called mass In many European countries, governments
customization. pushed an industrial policy based on BIM, because
The PIM operates by manufacturing prototype the crisis in the sector has not only a financial nature,
products and then contextualizing its instances but, especially, an industrial nature. Moreover inef-
ficiencies of property and infrastructure investments In order to get this overall performance, build-
affect the public finances, even if current spending is ings and cities behaviour has to meet various techni-
much more relevant. cal and non-technical requirements (physical as well
Through BIM which is accompanied by a more as psychological) placed upon them by owners, us-
efficient information management, the sector may ers and society at large.
acquire a production quality typical of more mature Research in this field will be seeking to reduce
industries. the gap between technology and society, to in-
The efforts of the community identified as In- crease the quality of building production, by means
ternational Alliance for Interoperability (IAI), estab- of open and participatory approach.
lished by scientific communities in partnership with In terms of technological solutions, the product
key players in the commercial sector, in the last 10 knowledge has been fairly studied and a number of
years aimed at establishing BIM standards for the modelling techniques have been developed. Most
use of object technology in construction and facili- of them are tailored to specific products or specific
ties management. aspects of the design activities.
These standards, known as Industry Founda- Current research on AEC product modelling can
tion Classes (IFC) are now contained within the most be classified in two main categories:
comprehensive model of design, construction and • geometric modelling, used mainly for support-
Facility Management information yet created. All the ing detailed design, and
main software developers in this industry segment • knowledge modelling, aimed at supporting
worldwide are committed to producing IFC-compli- conceptual aspects of designs.
ant software. Specifically, on the need to govern the symbio-
Studying the IFCs structure, we can observe that sis between building and its functions, so that com-
they have been developed by means of a space- puters can support every phase of construction (e.g.
components product approach, successful in terms Solibri program), it is necessary to have information
of data exchange and information interoperability models based on an adequate knowledge represen-
between programs, not intended for human under- tation, formally computable.
standing. This lack of semantics is reflected in the This kind of knowledge, oriented to solve com-
modelled buildings, once it is required to simulate plex technical problems, cannot avoid to qualify the
its behaviour in terms of usage, safety and comfort. product building through its relationship with the
More specifically, to predict human behaviour in context and with the actors.
a building during its usage, by means of the actual In terms of social contributions, on the other
standards, tools and technologies is an urgent open side, we need to clarify roles and identify responsi-
problem which challenges knowledge engineers bilities of actors involved along the building life cy-
and building designers since long time. As well it cle, starting from the client, through designers, pro-
involves a lot of resources in terms of industrial re- viding for the participation of users from the early
search and developments in the fields of army and stages of design concepts.
videogames. The BIM methodology assumes that there is a
client able to schedule formally a process of briefing,
FUNCTIONAL PROGRAM VS. BUILDING design, production and management, for example
PRODUCT DESIGN using “template” for the programming of functions
A shared goal, typical of all AEC industry products, is and activities, and thus reducing the level of ambi-
to functionally facilitate its direct and indirect users’ guity in the requirements definition.
activities, being aesthetically pleasing (Fioravanti et Client, especially if they must also manage the
al., 2011a, p. 185). constructed facility, are the largest beneficiary of the
process-product models development, because of There are at least two problems with the way all
their risk-based reasoning approach drives the opti- applications typically represent process information:
mization of contract management. • They use their own internal representations,
Designers, challenged to become more aware therefore communication between them, a
of product and process models, are the key to the growing need for industry, is nearly impossible
spread and development of the most advanced in- without some kind of translator.
formation systems. An open area of research works • The meaning of the representation is captured
on the interface between designer and tool, to en- informally, in documentation and example, so
able the first to clearly face pre-defined patterns and little automated assistance can be given to the
then customize them while using the software they process designer.
are familiar with. In terms of Process Knowledge Modelling, at the
Users, generally, as well known, play the central state of the art, it is important to refer to some on-
role in Architecture. The problem we approach here, going researches at the international level.
renouncing at the impossible task of modelling the
actor’s “libero arbitrio”, free unpredictable will, focus- NIST CPM
es on defining a set of occurrences that dynamically A design repository project at NIST attempts to
happen in the built environment. model three fundamental facets of an artifact repre-
Planners’ traditional approach consists in enter- sentation: the physical layout of the artifact (form),
ing planned processes (expertise, technical regula- an indication of the overall effect that the artifact
tions, best practices, etc.), in an architectural schema creates (function), and a causal account of the op-
(Wurzer, 2009; Wurzer et al., 2010). However, those eration of the artifact (behaviour).
processes are correct only if the planner can correct- The NIST Core Product Model (CPM) has been
ly anticipate and inform the usage of the building by developed to unify and integrate product or as-
different building user groups. sembly information [1]. The CPM provides a base-
If organized in a proper way, it is possible to rep- level product model that is: not tied to any vendor
resent usage scenario, predicting activity inconsist- software; open; non-proprietary; expandable; inde-
encies and evaluating the performance of the build- pendent of any one product development process;
ing in terms of user experience. capable of capturing the engineering context that
At the same time it is possible to design a build- is most commonly shared in product development
ing use programme if it can be re-modelled during activities. The entity-relationship data model influ-
the building design process. ences the model heavily; accordingly, it consists of
With the aim of improving the quality of user two sets of classes, called object and relationship,
experience, this paper explores a method based on equivalent to the UML class and association class,
process-product knowledge, formalized to support respectively.
logic synchronization between the planning of ac-
tivities and design of infrastructures The buildingSMART
Standard for processes (formerly known as the Infor-
STATE OF THE ART IN META-PROCESS mation Delivery Manual or IDM [2]) specifies when
MODELLING RESEARCH certain types of information are required during the
Many applications use process information, includ- construction of a project or the operation of a built
ing production scheduling, process planning, work- asset. It also provides detailed specifications of the
flow, business process reengineering, simulation, information that a particular user (architect, build-
process realization, process modelling, and project ing service engineer, etc.) needs to have at a point in
management. time and groups together information that is need-
ed in associated activities: cost assessment, volume levels of need, which is its serviceability. It rates fa-
of materials and job scheduling are natural partners. cilities—supply—in performance language as a first
Thus the buildingSMART standard for proces offers a step toward an outline performance specification.
common understanding for all the parties: when to A set of tools was designed to bridge between
exchange information and exactly what is needed. “functional programs” written in user language on
The linked Model View Definition (MVD) turns the one side and “outline specifications and evalua-
the prerequisites and outcomes of the processes for tions” written in technical performance language on
information exchange into a formal statement. Soft- the other. Although it is a standardized approach, it
ware developers can take the standard and specific can easily be adapted and tailored to reflect the par-
Model View Definitions that derive from it and incor- ticular needs of a specific organization.
porate them into their applications [3]. The detailed
information for this is described in the ISO standard: Limits
ISO 29481-1:2010 Building information modelling -- Building Modelling is not an objective process, but
Information delivery manual -- Part 1: Methodology rather subjective, aimed at very specific purposes
and format. that depend, first and foremost, on contractual ty-
ISO 29481-1:2010 specifies a methodology and pology. On process models there are a lot of mis-
format for the development of an Information De- leading quarrels, in the sense that many models
livery Manual (IDM). ISO 29481-1:2010 specifies a have always appeared very reductionist and sim-
methodology that unites the flow of construction plistic in relation to the complexity of the real and
processes with the specification of the information the articulation of the reasons of the different actors
required by this flow, a form in which the informa- involved.
tion should be specified, and an appropriate way to Typically, in architecture, when a product design
map and describe the information processes within falls, analysts want to insert a design process to fix
a construction life cycle. the bad design. However, a one-size-fits-all design
process does not exist. Experience teaches that it is
ASTM Standard Scales quite hard to force a fixed process on a design team
The ASTM standard scales provide a broad-brush, that every actor must follow. Every designer has
macro level method, appropriate for strategic, over- their own unique way of solving design problems.
all decision-making [4]. The scales deal with both de- Design domain experts, usually, argue that bad
mand (occupant requirements) and supply (service- product design is fixed by hiring good designers not
ability of buildings) (McGregor and Then, 1999). They by adopting a better design process.
can be used at any time, not just at the start point There is a need to produce not more models,
of a project. In particular, they can be used as part but environments where it is more easily possible
of portfolio management to provide a unit of infor- to reformulate the existing process-product models.
mation for the asset management plan, on the one Specifically, process models influence the Infor-
hand, and for the roll-up of requirements of the busi- mation Modelling much more than drafting based
ness unit, on the other. The ASTM standard scales methods. Each actor instinctively wants to rearrange
include two matched, multiple-choice question- the software built-in model, because a single infor-
naires and levels. One questionnaire is used for set- mation model cannot meet all the Requirement.
ting workplace requirements for functionality and To set up an information modelling process
quality. It describes customer needs—demand—in since briefing phase, implies reasoning primarily on
everyday language, as the core of front-end plan- the building functions and on the physical environ-
ning. The other, matching questionnaire is used for mental solutions, such as energy modelling or usage
assessing the capability of a building to meet those planning.
USE PROCESS KNOWLEDGE MODELLING pens during the AEC design. Each ‘pole’ is constitut-
To provide a reliable, comprehensive and up-to-date ed by knowledge-based system in its respective do-
knowledge base on use process, we thought of rely- main. In particular on the knowledge of the product
ing on a general structure for knowledge represen- (building - with its components and its multidiscipli-
tation already presented and discussed among the nary aspects), context (site - with reference to physi-
scientific community by this research group (Car- cal, legal, planning, ecological and climatological as-
rara et al., 2009; Fioravanti et al., 2011a; 2011b), and pects), the actors involved (humans - professionals,
working to extend its application field to this spe- contractors, customers and non-humans - agents,
cific purpose. intelligent assistants) and procedures that regulate
This general process representation model is this process (such as commitment, design phases,
linked to a specific Building Knowledge Model (BKM) economic and financial aspects, administrative and
structure, oriented to formalization and description organizational rules). All these ‘poles’ evolve in time.
of each entities composing design product (spaces, This Research Group (RG) has structured and
building components, furniture, equipments, etc.). formalized product knowledge, through a logic
Each entity is represented in its main features decomposition of the building organism. “Product
and in its relations with other entities by means of ontologies” were implemented, starting from IFC
the ‘knowledge template (Carrara et al., 2009) based standards and developing a method for explicitly
on the already discussed “Meaning-Properties-Rules” modelling the rules that qualify the intrinsic mean-
structure. ing at different levels of aggregation.
Starting from this representation model, already The RG approached has structured and formal-
applied to represent building design products, the ized context knowledge, both physical-environmen-
new challenge is to extend it to the representation tal and jurisdictive, implementing with the same
and evaluation of spatial and technological require- method the “Context ontology “, allowing for ad hoc
ments defined according to user needs. support during decision-making processes of archi-
Specifically, the interdisciplinary processes tectural product design-programming.
which BKM aims to support include the following: In the last few years RG has been studying the
• Design of Use Functional Program to be per- “Actors ontology”, approaching the problems related
formed in an existing infrastructure; both to modelling specialist profiles involved in the
• Design of an infrastructure in accordance with design-programming process, and profiles involved
a defined Use Functional Program; in the process of use. Some rules governing the ob-
• Design of an infrastructural renovation in ac- jective part of user behaviour have been identified.
cordance with a defined Use Functional Pro- This paper reports on early results of a study
gram and / or rescheduling of activities defined which explores a method for structuring “Process
by Use Functional Program on the basis of the ontology”. The backbone lemma of this tetrahedron
existing infrastructure. “knowledge realm” is the recognition of the dynamic
dimension that characterizes every process model.
Tetrahedron Of Knowledge “Tetrahedron of knowledge” finds its most com-
Scenario in which a building project is delineated plete application in real AEC problems because, un-
by means of the outlines and guidelines is marked like the existing knowledge structures, allows actors
by four ‘poles’ of a Knowledge symbolic Tetrahedron to dynamically model process-product structures,
that represent the different kinds of knowledge: with explicit semantics.
product, context, actors and procedures (Fioravanti The BKM system based on the tetrahedral
et al., 2011b). knowledge structure, enables actors to intervene in
The four ‘poles’ of knowledge shape what hap- the course of work on the definition of process enti-
ties and rules. The system supports the re-modula- (activities) and some related research efforts have
tion of the constraints and objectives of the process been conducted. For example, a web-based proto-
that are bi-univocally related to functional and be- type system for modelling the product development
havioural properties of the product. process using a multi-tiered DSM is developed at
“Situatedness” of development processes is MIT. However, few research endeavours have been
a key issue in both the software engineering and found on design rationale (Peña-Mora et al., 1993)
the method engineering communities, as there is Events: particular process entities, “milestones”
a strong felt need for process prescriptions to be that occur in the dynamics of the activities. Emer-
adapted to the situation at hand. gencies necessary to structure the causal and de-
Specifically, the formalization of Use Process On- pendency relationship between Use Process entities.
tology, qualifies and is qualified through rigorously
structured relationship with the product-context- Use Process Requirements, Performance,
user ontologies. Behaviour
To model use process entities and rules means From a computational point of view, use process
governing the integration between product form, requirements can be defined as variables, because
function and behaviour and vice versa. they establish a mapping between a set of process
entities and a set of values which express some of
Use Process Design Knowledge their qualitative (and quantitative) aspects.
Use Process Knowledge is represented by means The specific values that satisfy a particular use
of Use Process Ontology, a structure based on Use process requirement in a particular situation (con-
Process Entities, qualified by a system of Use Process text and objective dependent) can be defined as use
Rules. On one hand these process rules govern ac- process performance.
tivities planning and on the other hand they control The set of all use requirements and performanc-
relationship with the rest of knowledge realms: who es can be defined as the behaviour of the represent-
does what, where, when and how. ed process entity/class in terms of use.
Use Process Knowledge can be described by
means of process classes, at different levels of ag- Design Goals Knowledge Structure
gregation: Design process goals can be stated as desirables
Use Process Actions: elementary class entities performance measures of the sought solution. Alter-
structuring the Use Process Ontology. They repre- natively they can be stated as set of constraints that
sent the process based on user’s minimum ergo- the proposed solution must satisfy.
nomic function. Each constraint indicates the specific level of
Use Process Activities: a set of Use Process Ac- performance a design solution should achieve in a
tions structured in time and space, oriented by the particular category or an acceptable range of perfor-
functional programme. They qualify the relation mance values.
between users and building (spaces, components, It can be represented formally using this general
facilities, equipment, etc.) annotation:
Use Process Rationale: aggregation of Use Pro- constraint ( value | range )
cess Design Activities. The importance of represen- where the vertical bar stands for ‘or’. A constraint can
tation for use rationale has been recognized but it is be stated in terms of a specific value it must satisfy
a more complex issue that extends beyond artifact or a range of values.
function. It is function of social-economical-environ- The function of the goals is thus to group a
mental sustainability. (The Design Structure Matrix number of related constraints that should all be sat-
(DSM) has been used for modelling design process isfied together (Carrara and Fioravanti, 2003). More
formally, goals can be represented by this general and rule formalizations. In order to interrogate De-
notation: sign Solutions, Ontology Rules have been imple-
goal ( { goal } | { constraint } ) mented in SWRL and tested on prototype instances
This definition is recursive: a goal can be stated of developed Ontology Classes to check use process
in terms of constraints, or in terms of goals. There - product constraints:
is no inherent difference between goals and con- • Space configuration and topological relation-
straints. Rather, they form a hierarchical structure ships among spaces;
where terminal nodes represent constraints and in- • Furniture and equipment dotation for each
termediate nodes represent goals. building unit;
The conditions under which a constraint is con- • MEP system, Structural elements and Space
sidered satisfied must be established and eventually configuration compatibility.
modified during the design process by the actors, In this specific case of study, the process repre-
according to the internal and external requirements. sentation is oriented to the use programming and
designing, so as to match the Activity Program, de-
LOGICAL IMPLEMENTATION PATH fined by means of traditional project management
The implementation pipeline, is oriented to predict software, together with the design solution of space
and evaluate the performance of a building based configuration.
on (planned or to be re-planned) usage scenarios By means of BKM general knowledge structure,
and vice versa modelling scenarios of use in a (exist- it is possible to connect a labelled graph of inten-
ing or to be renewed) building. tions, called strategy map, as well as its associated
This work focuses on a multi-model view of pro- flowchart guidelines to layout solutions.
cess modelling which supports this dynamicity. The It has been implemented a critical path diagram
approach builds beside the BKM product represen- of Hospital operating room renovation, and now
tation (geometric and non-geometric), a BKM pro- we are working on the actual link to Process Activi-
cess representation. ties Gantt chart. This map is a navigational structure
Since BKM provides a semantic structure and a which supports the dynamic selection of the inten-
standard language (XML, OWL) what we are working tion to be achieved next and the appropriate strat-
on is the implementation of a bidirectional synchro- egy to achieve it.
nization between software for Programming and A set of task guidelines, intended to help in the
software for Authoring space solutions. operationalisation of the selected intention, repre-
The assumption of this process modelling ap- sents some basic ergonomic rules about flow of pa-
proach is that process prescriptions should be se- tients, staff, equipment and material.
lected according to the actual situation at hand, i.e. Once accomplished the task of formally repre-
dynamically in the course of the process. senting Use Process and Product Knowledge accord-
To implement this process, the proposed Build- ing to the BKM Knowledge Structure, the implemen-
ing Knowledge Model, a formalized extension of tation steps are namely (Figure 1):
actual Building Information Models, includes repre- 1. Connect Product Design Ontologies and Use
sentation of both the characteristic of the ontology Process Ontologies (e.g. expressed in OWL lan-
entity and the constraints. By means of Protégé, an guage by means of ontology editors, e.g. Pro-
ontology editor, we implemented some representa- tégé);
tive use process design requirements on top of 2. Connect Use Process Ontologies with actual
some building ontology entities. BIM, or IFC (by means of API, or using Beetz et
Knowledge Representation allows queries and al. (2006; 2010) transcription of IFC in OWL lan-
constraint-verifications by means of proper reasoner guage);
Figure 1
Building Entities and Goals
Knowledge Modelling.
3. Connect BIM + BKM with a Process Manage- software,

ment environment (e.g. OpenProject, etc.). • by means of Revit DB Link (an Autodesk Revit
In order to connect the modeled knowledge add-in) it has been possible to export the BIM
with graphics in CAD / BIM, technologies related to model to a database (also edit-able and query-
databases definition have been used. Specifically: able).
• BKM ontological structure has been exported A proper database has been defined in order to
to a query-able database, ensure consistency check of the unique identifiers
• Autodesk Revit has been selected for CAD/BIM assigned by Revit to represented graphical entities
and instances of the implemented classes in the pro- etc.) and to evaluate the building quality before its
tégé Knowledge Structure. construction will increase the chances for the client
Similarly, this approach has been used to real- to be satisfied and will provide more guarantees
ize the link between OpenProject software, used to to success in terms of future efficiency and perfor-
manage the XML-OWL process instances, to the Use mance.
Process Knowledge Base, in Protégé.
Revit and OpenProject represented entities REFERENCES
are associated to instances of the BKM Knowledge Beetz, J, van Leeuwen, JP, de Vries, B 2006, ‘Towards a topo-
Structure; data associated to entity Properties can logical reasoning service for IFC-based building infor-
be “extracted” from the BIM model while other fea- mation models in a semantic web context’, Proceedings
tures can be manually specified in Protégé accord- of the Joint International Conference on Computing and
ing to the implemented Knowledge Representation Decision Making in Civil and Building Engineering, Mon-
Structure. tréal, Canada, pp. 3426-3435.
Linking the database allows keeping consisten- Beetz, J, de Laat, R, van Berlo, R and van den Helm, P 2010,
cy between IDs from the two different environments ‘Towards an Open Building Information Model Server
referring to the same represented concept. - Report on the progress of an open IFC framework’,
Proceedings of DDSS, P-18, pp. 1-8.
CONCLUSIONS Carrara, G, Fioravanti, A, Loffreda, G and Trento, A 2009, ‘An
There is an urgent need for tools able to link and Ontology based Knowledge Representation Model
translate business rules and programme-project for Cross Disciplinary Building Design. A general Tem-
processes to check where business processes are plate’, in G Çağdaş, and B Colakoglu (eds), Computa-
not following policies and rules. tion: the new Realm of Architectural Design, Istanbul, pp.
A benefit of the proposed knowledge represen- 367-373.
tation is to provide automated assistance for process Carrara, G, Fioravanti, A and Nanni, U 2004, ‘Knowledge
development by defining the semantics of process Sharing, not MetaKnowledge. How to Join a Collabora-
entities in a computer-manipulable way. For exam- tive Design Process and Safely Share One’s Knowledge’,
ple, many businesses have rules, policies, space- Proceedings of InterSymp-2004 Special Focus Symposium
activity requirements, that their processes are sup- on Intelligent Software Systems for the New Infostructure,
posed to follow. Baden-Baden, Germany, pp. 105-118.
However, the representation of these, typically Carrara, G and Fioravanti, A 2003, ‘Needs Requirements Per-
do not enable tools to check whether they are con- formances Vs Goals Constraints Values in Collaborative
sistent. BKM represents rules about processes in the Architectural Design’, Proceedings of SIGraDi Confer-
same way as the processes themselves, and uses a ence, Rosario, Argentina, pp. 253-255.
formalism that supports automated reasoning. Fioravanti, A, Loffreda, G and Trento, A 2011a, ‘Computing
Introducing and enhancing reasoning mecha- Ontologies to Support AEC Collaborative Design - To-
nisms it will go beyond the potential of existing wards a Building Organism delicate concept’, Proceed-
commercial tools for supporting decision making ings of eCAADe Conference, Ljubljana, Slovenia, pp.
activities. 177-186.
The proposed knowledge-based system sup- Fioravanti, A, Loffreda, G and Trento, A 2011b, ‘An innova-
ports process traceability and, consequently, allows tive comprehensive knowledge model of architectural
responsibilities recognition and re-usable experi- design process’, International Journal of Design Sciences
ences collection. & Technology, 18(1), pp. 1-16.
The possibility to coordinate design process be- McGregor, W, and Then, D 1999, Facilities Management and
tween different actors (including clients, final users, the Business of Space, Arnold, London.
Peña-Mora, F, Sriram, RD and Logcher, R 1993, ‘SHARED- game-oriented environment’, Proceedings of eCAADe
DRIMS: SHARED Design Recommendation-Intent Man- Conference, Zürich, pp. 389-394.
agement System’, Proceedings of Enabling Technologies:
Infrastructure for Collaborative Enterprises, Morgatown, [1] http://www.mel.nist.gov/msid/conferences/talks/
WV, IEEE Press, pp. 213–221. rsriram.pdf (last access 30-05-2013)
Wurzer, G 2009, ‘Systems: Constraining Functions Through [2] http://www.buildingsmart.org/standards/idm (last ac-
Processes (and Vice Versa)’, in G Çağdaş, and B Colako- cess 30-05-2013)
glu (eds), Computation: the new Realm of Architectural [3] http://www.iso.org/iso/home/store/catalogue_tc/
Design, Istanbul, pp. 659-664. catalogue_detail.htm?csnumber=45501 (last access
Wurzer, G, Fioravanti, A, Loffreda, G, Trento, A 2010, ‘Func- 30-05-2013)
tion & Action - Verifying a functional program in a [4] http://www.wbdg.org/design/func_oper.php (last ac-
cess 30-05-2013)
3D Model Performance
3D Model Performance - Volume 2 - Computation and Performance - eCAADe 31 | 395

396 | eCAADe 31 - Computation and Performance - Volume 2 - 3D Model Performance
New Methods for the Rapid Prototyping of Architectural
Models
Production of detailed models with 3D printing

Stefan Junk1, Samantha Côté2
University of Applied Sciences Offenburg, Campus Gengenbach, Faculty of Business Ad-
ministration and Engineering, Germany
http://www.hs-offenburg.de
1
stefan.junk@hs-offenburg.de, 2samantha.côté@hs-offenburg.de
Abstract. Various Rapid Prototyping methods have been available for the production
of physical architectural models for a few years. This paper highlights in particular the
advantages of 3D printing for the production of detailed architectural models. In addition,
the current challenges for the creation and transfer of data are explained. Furthermore,
new methods are being developed in order to improve both the technical and economic
boundary conditions for the application of 3DP. This makes the production of models with
very detailed interior rooms possible. The internal details are made visible by dividing
the complex overall model into individual models connected by means of an innovative
plug-in system. Finally, two case studies are shown in which the developed methods are
applied in order to implement detailed architectural models. Additional information about
manufacturing time and costs of the architectural models in the two case studies is given.
Keywords. Architectural model, CAAD, Rapid Prototyping, 3D printing, architectural
detail.
INTRODUCTION
Various Rapid Prototyping (RP) respectively Additive facturing). The application of these Rapid Prototyp-
Manufacturing (AM) technologies, which enable the ing technologies for the production of architectural
direct implementation of 3D drafts in models, have models provides a number of advantages over the
already been available for a few years. Today the conventional model production. For example, it al-
most popular technologies among these are 3D- lows models to be created in minimum time with
Printing 3DP with plaster powder and Fused Layer a greater degree of details. Furthermore, the repro-
Modelling FLM with plastic filament. A common fea- duction and variation of drafts and models are also
ture of these technologies is that the models are cre- simplified considerably.
ated directly from the 3D-CAAD-data. Another advantage in addition to this imple-
The physical 3D models are manufactured gen- mentation speed is the low costs for the systems
eratively, i.e. the models are created layer by layer by and materials used, resulting in a considerable re-
adding material (hence the name Additive Manu- duction of the model costs. However, there are cur-

Virtual Model Physical Model Figure 1
Data transfer from CAAD-Soft-
CAAD or ware or 3D-Scanning System
Native STL or 3D-Printer
BIM Interface 3D-Printer to 3D-Printer.
Format VRML
Software Software: Device:
Sliced
Slicing Model
and Additive
Reverse Printing manufacturing
3D-Scanning Points
cloud
Engineering STL parameters of the model
System
Software
rently still problems with regard to the data transfer terms, this means that the 3D-CAAD data have to
and the preparation of the models for 3D printing, be converted from the original file formats into a
which stand in the way of further expansion of this format that can be read in by the RP systems. Data
technology (Sullivan, 2012). These problems are based on CAAD are usually complete and consist-
highlighted and dealt with in this paper. ent. However, there are still some problems with re-
gard to the interfaces from the CAAD or BIM system
CURRENT CHALLENGES FOR THE PRO- to the RP software. RP devices only accept a neutral
CESSING OF 3D DATA TRANSFER FOR format, notably STL or VRML, but no native formats
RAPID PROTOTYPING from individual commercial CAAD system manufac-
With all Rapid Prototyping respectively Additive turers.
Manufacturing technologies, the 3D-CAAD data are The simple data format STL only reproduces the
imported first and the Rapid Prototyping prepared surfaces of 3D objects. In doing so, the 3D object is
as part of pre-processing. System-specific software approached with triangles, allowing the degree of
is available for this purpose. The actual construction detail and hence the data volume usually to be set.
of the model in layers then takes place in a Rapid However, this format does not provide information
Prototyping device. Finally, the model has to be on the colour or texture of surfaces, with the effect
post-processed, e.g. in order to remove supporting that monochrome models are created. The advan-
structures or improve the stability of the model. tage of VRML format is the opportunity to reproduce
The 3D-CAAD data may come from different surfaces but also coloured textures.
sources. On the one hand, 3D data created by means With data based on 3D scanners, there are usu-
of a commercial 3D-CAAD or BIM system are usually no problems with regard to the data format,
ally already available for new projects. But, on the since Reverse Engineering Software often uses the
other hand, only 2D drawings are often available for STL format themselves. However, the same prob-
existing buildings. No plans are often available for lem occurs time and again that the data records of
historical or even archaeological buildings. 3D scan- point clouds by the 3D-Scanning systems are incom-
ners are often used nowadays in these cases in order plete, since the scanners, which use optical sensors,
to register the contours of the exterior façades and find it difficult to register areas in which no light is
interior rooms. reflected. These “shaded” areas, such as grooves and
The different data sources must be prepared in recesses, re-appear as “holes” in the data record and
such a manner to allow them to be processed by the have to be removed with complex software opera-
RP technologies as shown in Figure 1. In concrete tions by the use of a Reverse Engineering Software.

SPECIAL DEMANDS ON DATA FOR RAPID • Interior structures can be easily exposed with
PROTOTYPING compressed air.
Besides the data format already described, it still The economic advantages of 3D printing include
has to be checked whether the data are suitable for the following:
Rapid Prototyping in preparation for the construc- • In contrast to numerous RP technologies,
tion process when processing the CAAD data. This with this technology no additional support-
includes, in particular, checking whether the model ing structures which have to be subsequently
has a sufficient wall thickness. With a model at a removed are required. In other words, only the
scale of 1:100 or 1:200 or even smaller, it may be the material required to create the model is con-
case that the wall thickness of the masonry in the sumed. As a result, the technology goes easy
model does not meet the minimum requirements on resources and is therefore sustainable (Junk
of the Rapid Prototyping system. It should also be and Côté, 2012).
checked which details are not able to be reproduced • Overall the costs for 3D printing of models with
in the model. this technology are nowadays about € 0.40/
For example, details, such as windows or ban- ccm. Furthermore, at approx. 23 mm/h, the
isters, are often so delicate that they are unable to construction time for a model is also relatively
be reproduced in sufficient quality by the Rapid Pro- short (ZCorporation, 2009).
totyping system, since they could break off due to 3D printing is particularly suitable for applica-
their weight and fineness. Especially openings in the tion in the field of architecture, since it is the only
outer shell, such as window surfaces, skylights and Rapid Prototyping technology that can be used to
doors, must be observed in particular in this respect. create coloured models. Furthermore, the model
They are often displayed in closed condition in order can be transferred in STL format. Since no colour in-
to show only the exterior facade of a building. Any formation is transferred in this way, the components
insights in the building or lines of sight through the have to be “dyed” in the software of the 3D printer.
building are lost in this process. The second available option is the transfer in
VRML format. In this case the colour information is
TECHNICAL AND ECONOMIC BOUND- transferred in addition to the geometry. Further-
ARY CONDITIONS OF 3D PRINTING FOR more, additional textures, such as logos and writ-
THE PRODUCTION OF ARCHITECTURAL ings, can be read in and applied in the 3D printer
MODELS software.
3D printing with polymer-plaster is used and further
developed in this contribution in order to overcome NEW METHODS FOR THE PRODUCTION
the previously described disadvantages for the data OF DETAILED ARCHITECTURAL MODELS
creation and detailing of the models. This RP tech- In order to extend the application options of 3D
nology provides several process-related advantages: printing to the production of architectural models, a
• Due to the simple printing technology, which method has been developed to divide the buildings
can be compared to an ink-jet printer, the ac- into individual areas. This allows very detailed mod-
quisition of the printer incurs only low costs, els to be created, which also enable interior insights.
resulting, in turn, only in low hourly machine To assemble the individual models easily without
rates. mixing them up, the Poka Joke method is also ap-
• The materials (polymer plaster, binding agent, plied.
ink and infiltrate) used is relatively inexpensive. At first the building is divided into individual
• Coloured models with textures and lettering areas in the CAAD system. In the event of a single-
are easy to produce with ink cartridges. family house, these sections are preferably the cel-

Figure 2
CAAD-Model of Single fam-
ily house (left) and physical
model (right), Scale 1:100.
lar, floors, the roof and adjoining buildings. Indus- the individual models (Santos et al., 2006). This pre-
trial buildings can be divided according to their vents the individual models from being assembled
functions, e.g. office and workshop areas, adjoining in the wrong combination. It also reduces the risk of
buildings, supply facilities. This gives the model a damage due to incorrect assembly considerably.
structure and allows the individual models to be
designed in detail. The design usually comprises the CASE STUDY 1: SINGLE-FAMILY HOUSE
reproduction of flooring and exterior walls. Further- A single-family house as shown in Figure 2 was 3D-
more, supporting interior walls and non-supporting printed in this case study. The building was divided
lightweight walls can be reproduced. The function into the individual models: cellar, two living floors,
of the individual walls can be depicted by different roof and adjoining building.
wall thicknesses in the models. Interior staircases as Each floor was created with a floor plate and
well as pillars and supports are also reproduced. side walls. Furthermore, the interior walls and all
Further details, such as windows and door open- openings (windows and doors) were reproduced. As
ings as well as gates and skylights of industrial build- needed, the functions of the rooms could be applied
ings can be integrated in the model. This allows the in the form of writings to the floor plate to provide
room layout to be recognised. Since the models are the constructor with a better understanding. Also
set up floor by floor and are open at the top, it is pos- details like the grey painting of the oriel at the fa-
sible for the architect and customer to discuss and çade in the front could be demonstrated.
assess the design. Since the model was created at a scale of 1:100,
The individual models are equipped with con- the interior and also exterior walls could also be re-
nection elements to make it easier to handle them. duced to the scale without falling short of the mini-
They are simple plug-in connections which allow mum requirements of the 3D printing system (see
the overall model to be quickly assembled or disas- Figure 3). The individual model parts are joined by
sembled. The Poka Yoke method is used for this pur- means of plug-in connections. This allows the roof
pose, i.e. the plug-in connections are positioned in and the individual floors to be raised in order to ob-
such a way that there is only one way of connecting serve and assess the underlying areas. Lines of sight

Figure 3
View on internal structures
in CAAD (left) and in physical
model (right), Scale 1:100.
can be seen in the building due to the open design essary details to the constructor.
of the building. The scale 1:200 was applied to this model to al-
low even the largest individual model (cellar) to fit
CASE STUDY 2: INDUSTRIAL BUILDING into the construction space of the 3D printer (204
The 3D printing of an industrial building is imple- mm x 253 mm x 204 mm, LxWxH). In this case, the
mented in the second case study as demonstrated wall thickness of the interior and exterior walls had
in Figure 4. It is used as a test centre for persons and to be adjusted (i.e. enlarged) in order to adhere to
motor vehicles. It consists of a cellar, several test the minimum wall thickness of the 3D printer and,
halls and an office building with common rooms. in this way, create a stable, durable model. As shown
The building is initially divided into storage, test, of- by the view onto the internal structure of the build-
fice and social function areas. However, these areas ing in Figure 5 this distorted the scale to a certain
are so large that some of them need to be divided degree, since the lengths and heights of the build-
even further into floors in order to illustrate all nec- ing are true-to-scale, but not the wall thickness.
Figure 4
CAAD-Model (left) and
physical model of complete
industrial building (right),
Scale 1:200.

Figure 5
CAAD-Model (left) and
physical model of industrial
building without roof for view
onto internal details (right),
Scale 1:200.
The cellar, which covers all areas, serves as the ing days, although, in this case too, printing can be
basis for the overall model. The individual models performed overnight to accelerate the availability of
are plugged onto the cellar either entirely or floor by the model. In addition it takes in both case studies
floor. The office and common areas are particularly some hours to maintain the preparation of the data
challenging, since they comprise numerous details. because only 2D-drawings are available.
When comparing the manufacturing costs, a
COMPARISON OF THE MANUFACTURING distinction must be made between material and
TIME AND COSTS machine costs. The material costs consist of the
The manufacturing times and also the costs for the costs for the polymer plaster powder and binding
two case studies are specified in Table 1. The manu- agent used during the production phase. The costs
facturing time is based on the actual construction for ink are negligible in these examples. The material
time for the model and the time required for post- costs also include the costs for the resin used dur-
processing, which consists of cleaning the model to ing post-processing for the infiltration and hence
remove residual powder and the subsequent infil- the increase of the strength of the models. The ma-
tration with resin. The total manufacturing time for chine costs are based on different boundary condi-
the single-family house is considerably shorter than tions (e.g. acquisition costs, service life, depreciation,
that for the industrial building due to the lower con- interest) used to calculate the hourly machine rate.
struction volume, the lower number of individual The personnel costs are not included in this calcu-
parts and the lower complexity of the geometry. lation, since, by experience, they vary considerably.
The single-family house is completed within In the both case studies the material costs of the
a working day and can be printed, for example, industrial building are more than the double of the
overnight. In contrast, printing and reworking the costs of the single-family-house. The machine costs
industrial building is expected to take 1.5 work- of the industrial building are almost the triple of the
Manufacturing time Single-family Industrial Building, Table 1

House, Scale 1:200 Comparison of manufacturing
Scale 1:100 time of single-family house
Manufacturing (3D-Printing) 3 h, 2min 8h, 30 min and industrial building.
Post-processing (Cleaning, Infiltration) 2h 2h, 30min
Total manufacturing time 5h, 2 min 11h

costs of the single-family house because the geo- weak points, such as insufficient wall thicknesses,
metric complexity of the model. The total manufac- and enable the weak points to be eliminated quick-
turing costs are currently significantly higher than ly. Assistance in the division of the building into ap-
the literature value (ZCorporation, 2009) due to the propriate individual models would also be helpful.
annual price increases for materials since the publi- For further examinations, there are plans to in-
cation date. vestigate the usage of other Additive Manufacturing
technologies, such as Fused Layer Modelling FLM.
SUMMARY This technology usually enables models of higher
The current challenges for the implementation of ar- strength to be produced in comparison to 3DP,
chitectural models by means of 3D printing are illus- since plastic filament (ABS) is used as building ma-
trated in this paper. They include, in particular, the terial. However, a restriction here is that only single-
data transfer and the adaptation of data to meet the coloured models can be produced. The application
requirements of the 3D printing system. It has been of multi-material printing (e.g. Multi Jet-Modelling
demonstrated that 3D printing has a number of MJM) could also be examined in order to visualise
technical and also economic advantages in compari- further details. This could result in a more realistic
son to other additive manufacturing technologies. reproduction in particular of transparent details. In
The methods introduced for the division of the addition a higher accuracy of the modes is expected
building into individual models allow numerous due to the smaller layer thickness that as used by
details to be reproduced in the exterior as well as this technology.
interior. These individual models can be joined to
form complex overall models using the Poka-Yoke REFERENCES
method. Both case studies were able to show how Junk S and Côté S 2012, ‘A practical approach to comparing
to implement these methods successfully for a sin- energy effectiveness of rapid prototyping technolo-
gle-family house and also for a complex industrial gies’, Proceedings of AEPR’12, 17th European Forum on
building. The comparison of the manufacturing time Rapid Prototyping and Manufacturing, Paris, France
and costs gives a reliable basis for the calculation of Santos J, Wysk RA and Torres JM 2006, Improving Production
future projects. with Lean Thinking boken, Wiley, Hoboken
Sullivan, C 2012, ‘In architecture, is 3D printing the new nor-
OUTLOOK mal?’, www.smartplanet.com
To simplify the data exchange in future, interfaces Z Corporation 2009, Funktionsweise des 3D-Drucks: Die Vi-
are to be developed which simply the creation of sion, die Innovation und die Technologien hinter dem
STL data considerably. Furthermore, the printer soft- Tintenstrahl-3D-Druckverfahren, White paper, Z Corpo-
ware should be extended to make it easier to detect ration.

Four Chairs and All the Others - Eigenchair
Data driven design

Miro Roman
Future Cities Laboratory - Chair for CAAD - Institute for Technology in Architecture –
Swiss Federal Institute of Technology (ETH)
http://www.caad.arch.ethz.ch
roman@arch.ethz.ch
Abstract. By contemplating on the Eigenchair project, we ponder upon strategies and

concepts of designing by using information technologies. What are the potentials of data
driven design? What happens with objects when they are abstracted and reduced to a set
of data? The emphasis is no longer on the creation of physical objects, but on conceiving
meta-objects in the possibility space. Furthermore, this enables us to manipulate with a
whole population of objects, instead of a single object. How do we get this abstract system
to relate to the real world? Information technologies have opened up a number of new
ways of thinking about the world and the object and they, by far, surpassed the formally
simplified expression in design and architecture. Based on intellectual heritage of history
and culture, information technologies can, by utilizing and recycling various elements and
information, explore the 21st century object.
Keywords. Eigenchair; eigenvector; Principal Component Analysis; data; indexing.
EIGENCHAIR
The project Four Chairs and all the others opens the Eigenchair is a concept that results from the ef-
possibility of an alternative definition of design. fort to design a chair that refers to the genealogy
Rather than offering yet another thesis in support of of chairs, yet carries the potential of all chairs that
linear design development, it emphasises its polyse- might ever be created in future (Figure 1). This is the
mantic nature by understanding design process as central subject of the project Four Chairs and all the
an open field of possibilities, which not only explore others - Eigenchair which is observed and explored
physical limitations of space, but also react to con- as a sum of ideas. Prefix Eigen is commonly used in
temporary social and cultural phenomena. In order linear algebra in compounds such as eigenfunction,
to explain the idea, specific techniques were used to eigenstate, eigenvector. It comes from the German
replace simple design concepts with a series of par- word eigen which means “own”. The basic tool for the
allel narratives, thus provoking new and unexpected design of the population of chairs - i.e. “all the oth-
situations. The primary field of interest of this pro- ers” - is the Principal Component Analysis algorithm
ject becomes the intersection of different domains (Abdi and Williams, 2010). It is a standard tool for
of human knowledge, especially architecture, cul- contemporary data analysis that has been adapted
ture and information sciences. to various needs, from the neuroscience to comput-

Figure 1
Different renderings of an
EigenChair.
er graphics, and which is now being applied in the creation and modification of rules and systems,
field of design (Sirovich and Kirby, 1987; Turk and which then generate an abstract machine - or a
Pentland, 1991). Principal Component Analysis reduc- population of objects. The designer therefore does
es a given data set to a set of principal components, not manipulate the “artifact” itself, but rather the
i.e. eigenvectors. The key feature of this algorithm rules and systems which generate it. The emphasis
is the intersection and interconnection of all data, is no longer on the creation of physical objects, but
whose result adapts and changes according to the on conceiving meta-objects in the possibility space.
required point of view, i.e. subjective interpretation.
The objective of this project is to show strate- Recycling Information
gies and concepts of designing by using information The postmodern condition defines a set of critical,
technologies. What happens with objects when they strategic and discursive practices which, as their
are abstracted and reduced to a set of data? What main tools, use concepts such as difference, repeti-
are the potentials of data driven design? tion, simulacrum, hyperreality, in order to destabilize
modernist concepts such as identity, linear pro-
ALTERNATIVE DEFINITION OF DESIGN - gress of history, or unambiguity (Aylesworth, 2013).
DESIGN APPROACH The supermodern condition, on the other hand,
Radical view of the world and society is today me- is not focused on the creation or identification of
diated through advanced technological systems. the existing “truths“, but on the filtration of useful
Thanks to – or perhaps due to – such circumstances, information among the plenitude of new media
the design seeks new ways of thinking and concep- cultural practices. In order to avoid postmodernist
tualizing, as well as producing objects and ideas. The tautological nihilism, the supermodernist paradigm
informatization of the society and the computer- approaches the deafening cacophony of sings in an
aided design are opening a whole range of new ide- active manner. This paradigm also operates within
as about the perception of time and space we live in. the field of design, in which it is no longer the object
The algorithmic design is now based on new param- that is in the focus of research, but its characteristics,
eters: design of ideas, narratives, procedures, popu- features, relations, ratios, structures, indexes. The in-
lations, digital production, and new understanding formation age enables a redefinition of postmodern
of the materiality. Generative design methods mean techniques such as collage, assemblage or bricolage,

Figure 2
Initial data set of 12 chairs.
all of which define an object by collecting and reas- design process becomes an abstract definition of al-
sembling various information and elements. The gorithms. Instead of focusing on a “perfect” chair a
newly created object is now a fusion of different ob- whole population of chairs was designed (Figure 3).
jects’ data but it is also completely unique and inde-
pendent in form. This project is an example of digital Imposed Materiality
recycling which, recycles information and data of In generative object design, materiality of an object
chairs (Figure 2). is not a precondition for its final manifestation. The
choice of material has so far served as the basis for
Elitism And Exceptionalism Of Singular determining the design process, defining the ex-
Object Vs. Individual Populism Of Generic pected execution of details, connections and tex-
Objects tures. Today, the generative system design enables
So far, the field of design understood practices the imposition of materiality to the object. The form,
which dealt with singular objects, that is, the crea- uncomplimentary to certain material, can now be
tion of unique and specific “ideal” objects. Such an attached to it by mere use of intellectual control.
approach was closely related to the modernist para- Therefore, the objects, previously described by fixed
digm. Today, however, the emphasis is moving from geometry, can now gain relative geometry that can
the design of an object to the design of an idea. The be rendered into reality via 3D printing. Its materi-
new paradigm changes designer’s relation to a static ality is the last, almost arbitrary decision done by a
object by putting an emphasis on conceptualiza- designer (Figure 4).
tion, interaction of the components, systems, and
processes. What was once the design of a perfect, Designing Narratives
unique object featuring specific materiality is today By rethinking the notion of “good design”, one
the design of a population of objects featuring any comes to the conclusion that design is just a tan-
materiality. Instead of a specific object, the designer gible fragment of reality, which narrates one of the
creates an algorithm. Elitism and exceptionalism many stories that surround us. Design never appears
contained in the idea of a singular object is replaced in silence. What we call “good design” nowadays is
by “individual populism” of generic objects. The key imbued with a series of narratives constructed by
role is taken over by generative systems that of- different discourses: formal, ideological, psycho-
fer methodology and theoretical world view in the logical, and theoretical. Only one part of the design
framework that go beyond dynamic processes. The process is constituted by its material and formal as-

Figure 3
Population of chairs defined
by maps and Eigenweights.
pects, while most of it is built upon stories which de- Wire Chair by Charles and Ray Eames, Panton Chair,
scribe it and the individuals who transfer the stories and Ghery’s Wiggle Side Chair (Vegesack et al., 1996).
or identify with them. Therefore, besides designing Their main mutual link is specificity and uniqueness
an object, it is also necessary to design a narrative of the material and their respective technological in-
which defines its meaning. novation, depending on the context in which they
The research focus of the project Four Chairs and were designed. It is the richness of meaning and
all the others is the design of a chair which does not historical references of these examples that are re-
carry on the heritage of iconic or functional pieces sponsible for enabling us further creation of analo-
of furniture, but a one which contains information gies, stories and narratives, which, in turn, fertilizes
about “all chairs ever created“, for which the term viewer’s active participation in the process of visual
Eigenchair is used - to describe a sum of ideas. The representation.
algorithm database contains a number of “other
chairs”. Their fusions enable an infinite variety of MULTIDIMENSIONAL VECTOR - TECHNI-
possible results. In order to achieve a certain con- CAL APPROACH
trol over the results, out of “all other chairs” we have The project Four Chairs and all the others deals with
chosen four chairs as a precondition for creating manipulating data, thereby generating new objects.
identity and narrative. Fusions of characteristic parts A whole library of chairs, that is, their geometric and
of those four chairs with all the others are defined spatial characteristics, along with their historical im-
by user made maps that define the transformations, portance and their narratives, is taken as the start-
upgrade the performance of the Principal Compo- ing point of the project. By using open source 3D
nent Analysis tool, and enable the control of the re- models of chairs from Google warehouse, their ge-
sult (Figure 5). The project Four Chairs and all the oth- ometry is appropriated through a set of algorithms,
ers refers to four iconic chairs: Thonet’s Chair No.14, after which the Principal Component Analysis algo-

Figure 4 were used as testing data set due to the compu-
3D printed models of chairs tational limitations. All data had to fit in the same
generated by EigenChair bounding box, and mesh vertices were equally dis-
application. tributed throughout the mesh.
The whole application consists of three main
parts. The first part is the Algorithm for Voxelizing
Polygon Meshes. This algorithm transforms each
mesh into a voxel based object defined by a one-di-
mensional numerical array list, i.e. multidimensional
rithm is used to fuse, merge and manipulate input vector. In case of the highest resolution, each chair
information and create new objects. The result is a is represented with 2,788,875 values. Each value
population of objects that are over coding cultural marks the distance between the given voxel and
and historical space-time relations through logistic the closest mesh vertex. Values for each chair are ex-
networks. Final design is entirely a product of math- ported as separate txt files, in order to reduce com-
ematical and logical thinking. The object now be- puting time of the main application (Figure 6).
comes a product of pure intellect, grounded in his- The second part is the Algorithm for Morphing
tory and culture. The main algorithm, which defines Chairs. The base of this algorithm is the Principal
the whole project, is Principal Component Analysis Component Analysis. The goals of Principal Compo-
algorithm. nent Analysis are to extract the most important infor-
mation from the data set; compress the size of the
Logical steps data set by keeping only the important information;
The initial step was to normalize and prepare the simplify the description of the data set; and analyse
data of all the chairs. In this case study, 12 chairs the structure of the observations and the variables.
Figure 5
New chair as a fusion of
Panton Chair and Wiggle Side
Chair.

Figure 6
Chair abstracted to a multidi-
mensional vector.
In order to achieve these goals, Principal Component also played by a series of open source libraries, espe-
Analysis computes new variables, called principal cially the Marching Cubes Algorithm (Lorensen and
components or Eigenvectors, which are obtained as Cline, 1987), responsible for generating watertight
linear combinations of the original variables. The mesh objects ready for 3D printing. All codes were
first principal component is required to have the written in Java programming language.
largest possible variance. The second component is Having in mind referential and recycling dis-
computed under the constraint of being orthogonal course, it is important to note that the algorithms
to the first component and thus having the second used in the project, e.g. Principal Component Analy-
largest possible variance. The other components are sis algorithm and Marching Cubes Algorithm, are al-
computed likewise (Figure 7). ready in practice. They are thoroughly adapted and
According to the size of the initial bounding box, functionally redirected, recycled to fit the needs of
a voxel-based space is created. Each voxel receives design.
values from txt files exported in the first step. With
the use of Principal Component Analysis we can rep- ARTICULATING INDEXES - THEORETICAL
resent each chair by using only a set of Eigenweights, APPROACH
e.g. (-5673, -85184, 50, -25533, 31594). By changing
the values of principle components, i.e. Eigenweights, Information
we are able to achieve linear transformations be- The key term which best describes and corresponds
tween all the chairs (Figure 8). to contemporary society and science is information.
The third part is the Algorithm for Mapped Mor- Information technologies are entering all spheres
phing. It is an upgrade from linear Principal Compo- of society: from the ways in which we organise our
nent Analysis transformation to a nonlinear mapped everyday life, to the ways in which we think about
transformation. An RGB map, in which each color natural sciences and humanities. This leads to the
represents a particular chair, is projected to the vox- conclusion that is impossible to understand human
el-based space. This enables us to define and control environment only in material terms of energy and
the nonlinear transformations and fusions of three matter; in order to create a comprehensive world
different chairs into a new one. Thus created, chairs view, the analysis must take into consideration the
can be used again as input chairs for the second step, category of information. At the same time, being
and achieve a new nonlinear variability (Figure 9). surrounded by excessive amount of information, the
The rest of the algorithms served to prepare the analysis requires a stable environment, which ena-
data for Principal Component Analysis and to help bles their observation and use.
with their final visualization. An important role was

Figure 7
Workflow of the Algorithm for
Morphing Chairs.
Reflection On The Real between the real and the virtual, with Deluze’s and
It is impossible to comprehend or examine what is Guattari’s negation of the linear approach to the
“real”, because it depends on quantisation and for- real. Such understanding of reality is supported by
malisation of ideas. The hierarchy and the relation the vanishing of boundaries and the influence of the
between the original and its copies, which was the virtual on the real. Simulation is a process that pro-
key concern of the materially oriented society, have duces the real, and vice versa (Massumi, 1987).
become completely irrelevant in an age in which vir-
tual reality dominates human lives. Depending on Abstraction
the ways of our understanding and accepting of the The Internet age is exactly such condition, in which
“unfamiliar”, we legitimatise and comprehend the immaterial information is part of what we call real-
real. Brian Massumi perceives this in a multifaceted ity. In this context, the only way of manipulating
way, by comparing Baudrillard’s interpretation of with information is abstraction, and it can be ad-
the reality-simulation, in which there is no division equately used only by those who are, in a mass of
information, able to define their context as a flexible,
Figure 8 adjustable field of possibilities with multiple mean-
Main EigenChair application ings. The project Four Chairs and all the others con-
interface – left Voxel based siders the abstraction of objects to the degree that
space – right – control board. enables their manipulation and the creation of new
meanings (Figure 10). If objects - chairs, or whole
populations of objects, are abstracted to the level
of multidimensional vectors, i.e. to a series of num-
bers in a line - indexes, they become very potent and
manipulative (Figure 11). Such abstract objects, i.e.

Figure 9
Diagram of an Algorithm for
Mapped Morphing.
indexes, are placed in a meta-space, which contains expected. This project shows that design is able to
potentials of all objects present there (Figure 12). manipulate predetermined potentials, while filling
Governed by the Principal Component Analysis al- them, at time same time, with narratives. Design is
gorithm, meta-space is able to correlate indexes of not a part of the endless evolutionary process aimed
all objects, creating thus an open logistic network, at creating the next new ideal object, but a part of
a possibility space. This marks the level of articula- a defined context and chosen references with their
tion of different elements and the creation of whole respective genealogies.
populations of objects of the same “kind”. By looking
at objects through the level of their abstraction, we EIGENCHAIR - DATA DRIVEN DESIGN
realise the potency of information (in meaning and By using information manipulation and various spa-
shapes, with which we can manipulate), but at the tial conceptions, algorithmic design approaches an
same time its complete emptiness when perceived object in a completely abstract manner, separating
on the index level alone. it thus from the “reality”. While making it extremely
flexible for different interpretations and contextu-
Meaning, Context And Narrative alisation, it also contributes to the instability of the
Post-traditional society offers new perspectives on process as a whole. The object can easily be reduced
old concepts, to which we give new meanings or to a geometry exercise. Therefore, the key feature of
judge them by creating our own context. The mass design is not only the definition of algorithms, but
of information shapes our world: text, visual rep- also the construction of parallel narratives around
resentation, music, money. However, the idea that the object. It is therefore necessary to re/turn to the
“information does not carry meaning” offered by the postulates of the pre-Socratic philosopher Empedo-
information theory pioneer Claude Shannon, has cles who claimed that “nothing comes out of noth-
become rather liberating in the academic discourse; ing and nothing disappears into nothing”. Such phil-
information carries unlimited freedom of manipu- osophical re/turn marks an effort to observe context
lation. It is important to emphasise that contextu- and processes as more important factors for defin-
alisation and the successive creation of narratives ing the object, than those implicit in the Objectiv-
“fill” the systems of information. They gain power by
careful selection of data implanted in them, taking Figure 10
care at the same time that the contextualisation and Levels of abstraction.
the creation of stories which surround them rely on
culture and history (Figure 13). It is also important to
note that in the process of contextualising the ge-
neric before the generation itself, there is a whole
scale of possibilities which had been predetermined,
but which are also opening the potential for the un-

Figure 11
Different resolutions and visu-
alization of an EigenChair
ism (Terzidis, 2012). The advantage of the processual shift of design’s limits. Finally, the algorithmic design
design in contemporary world is its ability to refer to should adopt strategies and dynamics which deal
the sum of global knowledge and to use it effective- with the creation of narrative and contextualisation.
ly. The result of such turn/over is the creation of new This project tries to show – by conceiving and shap-
perspectives in defining objects, as well as a gradual ing the idea of a chair for the 21st century – the ne-
Figure 12
Meta space - possibility of
interconnection and interrela-
tion of all active data.

Figure 13
Contextualisation and the
creation of stories around
objects.
cessity of perceiving design through three equally GRAPH Comput. Graph., 21(4), pp. 163-169.
important, interdependent positions: design, theory Massumi, B 1987, ‘Realer than real’, Copyright no.1, pp. 90-
and technology. Design is now data driven. 97.
Sirovich, L and Kirby, M 1987, ‘Low-dimensional procedure
REFERENCES for the characterization of human faces’, JOSA A, 4(3),
Abdi, H and Williams, LJ 2010, ‘Principal component analy- pp. 519-524.
sis’, Wiley Interdisciplinary Reviews: Computational Sta- Terzidis, K 2012, Algorithmic Architecture, Taylor & Francis.
tistics, 2(4), pp. 433-459. Turk, M and Pentland, A 1991, ‘Eigenfaces for recognition’,
Aylesworth, G 2013, ‘Postmodernism’ in EN Zalta (ed), The Journal of cognitive neuroscience, 3(1), pp. 71-86.
Stanford Encyclopedia of Philosophy. Vegesack, A, Dunas, P and Schwartz-Clauss, M 1996, 100
Lorensen, WE and Cline, HE 1987, ‘Marching cubes: A high Masterpieces from the Vitra Design Museum Collection,
resolution 3D surface construction algorithm’, SIG- The Museum.

Digital Design Tools versus Architectural Representation
and Design Approach
A reading off architectural press

Betül Orbey1, Nihan Gürel2
Doğuş University, Turkey
1
betul.orbey@gmail.com, 2nihangurel@gmail.com
Abstract. This study aims to investigate the relation between evolving graphic
representations and due to new digital tools and how they affect architects’ approach to
design process. In order to do this, Yapı Magazine being published since 1973 in Turkey
will be retrieved and data related to types of architectural design representation used will
be recorded. The study will conclude with an evaluation of new representation means such
as 3D render, other 3D digital products and diagrams and how they have influenced new
approach to design.
Keywords. Digital design tool; architectural representation, architectural design
thinking.
INTRODUCTION
This study is interested in transformative reflections read this shift through a collection of architectural
of digital design developments at two levels: archi- representations.
tectural graphic representation and architectural
stand. METHOD
The notion of “generic design” proposes that This study will attempt to demonstrate a relation be-
there are great similarities between design acts tween the shift in graphic representation and design
(Gero and Purcell. 1998) independent of the domain stand of architects in respect to architectural press,
(Zimrig and Caine, 1994). On the contrary, there are Yapı Magazine. Yapı, the oldest established magazine
also opinions supporting the presence of significant still in press today in Turkey, will be utilized as a tool in
differences depending on design situations (Visser, order to evaluate the chronological period between
2009). Visser (2009) enhances the notion of generic 1973, the year Yapı had first been published, and
design and states that there are different forms of 2012 to understand how digital tools have affected
design. He defines three dimensions as sources of architectural representation and approach. Yapı Mag-
differences in design consisting of the process, the azine, as a tool to navigate through time and variety
designer and the artefact of projects, will provide evidence for claimed mutual
Here it is hypothesized that as new tools of de- evolution between digital design tools and architec-
sign are adopted, such as digital tools, a relevant tural design approach that is proposed to be read off
shift in design stand takes place. This study aims to architectural graphic representation.

READING OFF ARCHITECTURAL PRESS tent is to convince reviewers that the design solu-
The reason for studying evolution of architectural tion is the preferable one to the design problem in
representation due to digital technologies off of an hand. To achieve this, the project has to speak for it-
architectural press is that these magazines publish, self through its graphics. The core of the design idea
at least today, a variety of projects with consistency should be able to reveal itself through the way it is
and that they reach a large amount of people. Signif- represented. Sole image of the design proposed is
icance of its audience is that although it refers to ar- not always enough. The reviewer needs additional
chitects, professionals and architecture students by graphic representations that reveal intangible in-
content, anyone can reach it. It makes information formation about the design to fully comprehend
on architectural design affordable and it constructs it. Architects produce plans, sections, elevations,
a communication line with the people in the same scale model, diagrams, flow charts, exploded axons
field. to make the project as explicit as possible. The differ-
An architectural publishing may reflect personal ence between these representation types is defined
views of the author or a collective view of the edi- as either yielding the “receiver o be an active partici-
tors. Certain types of style may be or may not be of pant in the communication process, or pass the idea
preference to the magazine or which aspects of a directly to the receiver” (Kalay, 2004).
design to focus on such as detail, process, construc- As printing technology evolved and digital de-
tion etc. may depend on the principles set by the sign tools became ubiquitously used, they made
editorial board. Although a magazine may be re- indispensible impact on graphic representation of
produced from a subjective point of view, this is a a design which we have previously related to re-
consequence of communication through any kind flection of design stance. In 1970’s, a project was
of medium (Sert, 2006). What is more valued here represented solely with its technical drawings such
is that architectural magazines provide a record of as plans, sections and elevations. However, due to
change in terms of preferences, culture, tool and technological restrictions of the time, they were far
representation. from being explanatory and acted more as techni-
cal reports. These drawings are referred to as “static”
REPRESENTATION AS A REFLECTION OF by Kalay (2004). The drawings were made by hand
THOUGHT: TOOLS HAVE SHAPED THE and there was no way of altering them so that re-
WAY DESIGNERS REPRESENT AND AP- viewers other than the client and the contractor may
PROACH TO DESIGN be involved in the process. These representation
Representation is a mean in which information is norms have also been necessary as they are today.
embedded. Abstraction and use of symbols are its Today, these technical drawings are enhanced with
core features through which the aimed information sketches, perspectives and other graphical materi-
is communicated. It is a process in which an idea is als. Today, we should review the gradual change of
presented in another form. In this process, informa- graphic representation norms and attitudes. The
tion is put into related forms of conveyance. hypothesis is that, graphic representation of an ar-
Architectural representation acts as means of com- chitectural design have become more analytical and
munication of those ideas embedded in design revealing in terms of how the building will behave
through a visual forum. In order to represent an once it starts functioning as well as the process lead-
architectural design idea, scale models, sketches, ing to the particular solution.
renderings, perspective drawings and photographs In order to support this argument, a collection
are used most commonly (Kalay, 2004). When com- of examples relevant to a 40 years span of timeline
municating a design idea, representation methods will be utilized and make this tool dependent shift
that explain the project best are preferred. The in- in graphic representation in architectural design

comprehensive. A paradigm shift from sketches to tecture group in charge in addition to information
orthogonal drafted representations of design pro- regarding year, issue and page references. This ad-
posals are expected to evolve into photo-realistic ditional information may help for a deeper under-
renders, exploded axons and more descriptive dia- standing of representational means utilized as they
grams regarding performative analysis, function and may vary according to function or the design team
expected behaviour of the building and algorithms. being local or international.
As Archim Menges (2010) points out, design- And in the third phase, an assessment of the
ing with digital software requires to think of design nature of drawing belonging to one category during
with an algebra of collections, in a more abstract the span of 39 years were compared qualitatively to
way. Also architectural representation is not static as investigate how that certain types of representation
it had been; it also has transformed with regard to have evolved in time.
new ways of design approach.
In this study, architectural graphic representa- RESULTS
tion is taken as a reflection of architectural design Yapı Magazine was retrieved between 1973 and
stand. Results regarding the design approach adopt- 2012. Relevant issues in 1973, 1977 to 1982, 1984
ed will be driven through analyzing its graphic rep- and 1986 to1987 were either could not be found or
resentation; and the impact of digital design tools has been observed not to contain any design pro-
on architectural design thinking is expected to be- jects. It is almost with 1990 that design projects are
come more rationalized as navigated towards pre- published with consistency and it is after 2001 that
sent day on the chronologic timeline. the magazine publishes multiple design projects in
each issue. According to this, a total of 155 design
THE STUDY projects were recorded in terms of which represen-
Yapı Magazine was examined from 1973 to 2012 to tation types they have used.
investigate the change in architectural graphic rep- The first phase of the study shows that plan, sec-
resentation. Three issues for each year were exam- tion, elevation, detail (construction drawings) 3D
ined. These issues are January, May and September renders, 3D digital wire-frame/perspective/axono-
– so that the publishing is followed with 4 months metric views, sketches, hand drawn perspectives,
interval for the last 39 years. diagrams and physical model photographs (explan-
The retrieval process consists of three phases. atory drawings) and photographs are used collec-
Graphic representation means were extracted tively with varying percentages in time.
through the first phase. They are as follows: plans, In the second phase, as already expected, the
section, elevations, partial details, 3D renders, 3D results show that plans, sections and elevations as
digital perspectives – wireframe and axonometric conventional architectural representation types are
views, sketches, hand drawn perspectives, diagrams, commonly used. It should be noted that actual pho-
physical model photographs and photographs. tographs of the buildings are the most frequently
In the second phase, these graphics were quan- used representation type. However, this was omit-
tified to see how intensely they are utilized. In addi- ted in the results so that the evaluations can be
tion, 3D digital perspectives – wireframe and axono- made within those types of representation that are
metric views as also referred to as other 3D digital involved during the design process. Beginning with
products, and diagrams were assessed in terms of 1994, 3D renders as well as diagrams and other 3D
for what purposes they were used so that they may digital products start to enhance architectural rep-
be related to possible tool developments. During he resentation medium. In Figure 1, a bar is dedicated
second phase of the study, the records also included for each year. Each graphic representation type is as-
project name, function of the building and the archi- signed a colour and bars are segmented according

Figure 1
Distribution of Representation
Types Among Years.
to their percentage of use. The increase in the varie- of representation among these three (Figure 2).
ty of colours between 1973 and 2012 represents the These data have also been re-interpreted in
variation in representation medium used in present- terms of by the design teams they have been pro-
ing architectural design projects (Figure 1). duced. Expected frequency of 3D renders, other
As it may be read on Figure 2, first 3D render has 3D digital products and diagrams have been much
been encountered in 1994, first 3D digital product higher in contemporary design projects. However,
consisting of wireframe, perspective or axonomet- since the data is acquired from a local magazine a
ric views with no intention regarding photo-realistic question of whether the results may come up as
images in 1997 and first diagram is encountered in expected had this study been conducted through
1999. It may also be observed that use of diagrams a magazine of another nation or an international
are usually aligned with use of 3D digital products magazine. Results have shown that the number of
and 3D renders are always the most preferred type international design teams using diagrams are triple
Figure 2
Beginnings of representing
with 3D digital media and
diagrams

Figure 3
Use of diagrams, 3d Render
Images and other 3D Digital
Products by Local and Interna-
tional Design Teams.
the number of local design teams. The number of in- Previously, architectural representation was once a
ternational design team using 3D renders are almost language that can be understood merely by archi-
equal to local design teams. The number of interna- tects, planners and related disciplines but now it is
tional design teams using other 3D digital products transformed into a language that can be understood
are almost double the number of local design teams by everyone. Even traditional representations such
utilizing these representation mediums to present as plan, section and elevation have transformed into
their project (Figure 3). a simpler and schematic form with reduced level of
The third phase of this study was an evaluation detail and high level of abstraction (Figure 4 and 5).
of qualitative features of the architectural represen- As Kalay (2004) mentioned, main mechanism
tations. In the last years digital technology has influ- that transforms an idea into a communicable mes-
enced architectural representation and transmission sage is abstraction. Abstraction, extracts and distills
of design ideas with new methods and tools. With the meaning of the message, focusing attention on
new possibilities of expression in architecture, trans- its salient characteristics. Higher degree of abstrac-
mission of ideas has differentiated from traditional tion makes communication more efficient and it
architectural representations. helps to focus the receiver’s attention on the parts
In this direction, the presence of multi-discipli- of the message that the sender considers most im-
nary approach such as graphic techniques based portant. According to the results of the third phase,
on diagrams and schematic drawing, the use of ab- simple graphical expressions, schematic drawings
stract representations, more simple drawings even and diagrams become a more efficient way of repre-
cartoonish, the presence of simple mathematical senting design ideas, an ideal method of communi-
expressions can be found in architecture milieu. cating ideas to others (Figure 6).

Figure 4
Evolving Representation,
Plan in 1990 and 2010 (Yapi
Magazine).
Figure 5
Evolving Representation,
Elevation, Section (Yapi Maga-
zine, 2010).
EVALUATION OF RESULTS work across linguistic and cultural boundaries. (Ka-

lay, 2004) Another important feature of diagrams is
What Diagrams Tell that they tell stories regarding the evolution of an
Diagrams provide a visual medium of communica- architectural idea. They may be referred to as inclu-
tion for the sake of architectural representation. The sive and dynamic. They include any viewer into the
most important feature of a diagram is that it has its process of form in formation. Contrary to traditional
own language that everyone can comprehend. They ways of architectural representation, they do not

Figure 6
New Representations,
Diagram, Battery Project,
JDS+BIG (Yapi Magazine,
2010).
act as final reports of a process but represent the sign proposal is presented as a finished product. This
process itself. They “explore, explain, demonstrate or representation type is specifically chosen for pres-
clarify relationships among parts of a whole”. (Kalay, entation purposes rather than aiding design devel-
2004) Similarly, according to Rowe (1987) diagrams opment phase. These images are used to aid those
are used to explore, analyze and synthesize ideas. who are not architects or professionals in familiar
Diagrams may be utilized to establish design prin- fields but individuals who can not read construction
ciples that help the designer reflect on and prepare documents.
for subsequent exploration (Rowe 1987). Although this representation medium needs to
Architectural diagrams do not only represent be evaluated differently than diagrams and other
physical elements, but also forces and flows. In the 3D digital products, it also serves for the same pur-
early phases of designing, architects draw diagrams poses: inclusion and exposition. Similar to diagrams,
and sketches to develop, explore, and communicate 3D render images also tell stories. They are used to
ideas and solutions. Design drawing is an iterative reveal how the space created during different times
process. It involves externalizing ideas to store them of the day or different days of the year. Through the
and recognizing functions as well as finding new photo-montages made, they give clues regarding
forms and integrating them into the proposal. Thus how the spaces may be used and what kind of life
drawing is not only a vehicle for communication will take place once its inhabited. These images are
with others. It also helps designers understand the used for revealing a certain experience provisioned
forms they work with (Edwards 1979; Do and Gross, for designed space.
2001). According to Bares-Brkljac (2009), these im-
With a more thorough approach, Oxman (2000) ages inherit accuracy, realism and abstraction. It is
states that diagrams play a role in visual reasoning. through these features that non-professionals be-
And through what this representation medium pro- lieve in what they see. According to this, accuracy
vides, what Schon (1992) refers to as “reflection in aids non-professionals to be acquaintance with the
action”, what Lawson (1980) describes as having a space. It is related to scale, distances and relations
“conversation with the drawing” takes place and aids of volumes and spaces. (Bares-Brkljac 2009) It is also
the design process. related to chosen view points regarding angle and
height. Human eye angles are preferred on pur-
What 3D Products Tell pose so that the viewer can imagine himself in the
According to Lopes (1996) due to techno cultural picture. Realism helps the viewers understand and
changes, pictures are re-emerging. They now play a evaluate the proposal the same way they perceive
role in terms of storage, manipulation and commu- the environment. (Bares-Brkljac 2009) Abstraction
nication of information. refers to reduced information about design. (Bares-
Beginning with 1994, 3D render images have Brkljac 2009) A high level of abstraction may not
evolved into photo-realistic images where the de- sufficiently present the proposal where a low level

of abstraction may overload an image and draw the has enriched the way architects represent their
viewer away with the information he does not need work. And how they represent their work is here
and comprehend. According to Bares-Brkljac, the associated with their stand and the way they think
collective effect of accuracy, realism and abstraction about architecture through what these new repre-
in relevance to the form, influence observer’s per- sentation types offer. According to this and through
ceptual responses. the data acquired from Yapı Magazine regarding
On the other hand, Koutamanis (2000) draws the use of representation types, this study may con-
attention to a cognitive property of this representa- clude that architects have become more process ori-
tion medium by stating that they act as a reinforce- ented, expository, transparent in terms of reflecting
ment of internal representation by external ones. the design process, inclusive rather than exclusive or
isolating, abstract as well as more precise in reveal-
Why have we not encountered more? ing experience where on the contrary it had been all
Contrary to expectations, research areas studied in about communicating the information to the con-
the computational design such as parametric de- tractor to build the project.
sign, scripting and building information modelling The authors had expected to encounter traces
(BIM) has not taken place in the design magazine of contemporary design methods such as algorith-
to represent design proposals yet. This may be due mic design, building information modelling and
to similar reasons for why use of CAD tools have parametric design. However, due to reasons ex-
been slow in the 1980’s. In 1980’s, Jon Pittman explained above the expectations have not been met.
plains the reason for slowness in computerization The belief is that these methods still need time to
as the expenses of owning the machines. Koutama- penetrate into more design offices and find place in
nis’ (2000) approach also supports Pittman’s. After design magazines ubiquitously.
the democratization of computer technologies do- If another magazine was chosen as source, the
main specific systems such as drafting, modelling data may come up differently due to publishing
and computer generated images were expected to principles of the magazine. However, this approach
flourish. However, this process took time as well. Ac- is still seen valid since it establishes a controlled ex-
cording to Koutamanis, “The main reason has been periment by stabilizing the source and searching for
the understandable caution with which we approach evolution of new representation types among years.
systems that purport to improve not only efficiency but Again, if another magazine was chosen with an
also design quality and performance.” international identity, some results may have dou-
Although today, every design office holds suf- bled since the results of this study has shown that
ficient amount of hardware and a more advanced the number of international design teams using
software along with its know how, these offices can diagrams, 3D renders and other 3D digital products
not afford to spend time to master even more recent are double the number of local design teams utiliz-
software offering new methods and possible incoming these representation mediums to present their
patibilities due to new design methods between co- project
workers and other professional teams. However, the A future method to test this idea is to look at
authors expect to see these contemporary methods design competition entries both locally and inter-
of design and their graphic representatives in archi- nationally. This way, more ambitious sets of repre-
tectural design magazines in the following years. sentation is expected to encounter as well as more
contemporary methods of representation where ar-
CONCLUSION chitects feel encouraged to try new methods rather
Evolution of new digital tools and new representa- than to follow conventional methods of a design of-
tion medium for the architectural design process fice.

REFERENCES
Bares-Brkljac, N.:2009. Assessing perceived credibility of Manufacturing. New York: Taylor and Francis. pp.254-8.
traditional and computer generated architectural rep- Purcell, A.T. and Gero, J.S.:1998. Drawings and the design
resentations, Design Studies, 30, 415-437 process, Design Studies, 19, 389-430.
Do, E. and Gross M, 2001, Thinking with diagrams in archi- Rowe, Peter. Design Thinking. Cambridge: MIT Press, 1987.
tectural design, Artificial Intelligence Review, Kluwer Schon, D.A.,Wiggins, G.:1992. Kinds of seeing and their
Academic Publishers, Netherlands, 135-149. function in designing, Design Studies 13, pp. 135 – 156.
Edwards, B. (1979). Drawing on the Right Side of the Brain: Sert, G. B., 2006, A Survey On Photographic Representation In
a course in enhancing creativity and artistic confidence. Architectural Magazine Covers: Covers Of Arredamento
Los Angles, J. P. Tarcher. Mimarlik, Master Thesis, METU, Ankara.
Kalay, Y., 2004, Architeture’s New Media, MIT Press. Visser, W 2006. Both generic design and different forms
Koutamanis, A. 2000. Digital architectural visualization, Au- of designing, Paper presented at Wonderground, the
tomation in Construction, 9, 347–360 2006 DRS(Design research society) International Con-
Lawson, B. 1980. How Designers Think, The Architectural ference, Lisbon, Portugual, Novemver 1-4. http://hal.
Press, London. inria.fr/inria-00118256/en
Lopes, D. 1996. Understanding Pictures, Clarendon, Oxford. Willis, D. and Woodward, T. : 2010, Diminishing Difficulty:
Menges, A.2010, Instrumental Geometry, in Corser, R. (eds), Mass Customization and the Digital Production of Ar-
Fabricating Architecture, Princeton Architectural Press, chitecture, in Corser, R. (eds), Fabricating Architecture,
New York, pp. 22-41. Princeton Architectural Press, New York, pp. 178-213.
Oxman, R. 2000. Design media for the cognitive designer. Zimrig, C.M., Craig, D.L. 1994. Defining design between
Automation in Construction 9, 337–346. domains:an argument for design research a la carte in
Pittman, J. 2003. Chapter 18: Building Information Mode- C Eastman, M McCracken and Wnewstetter (eds) De-
ling: Current Challenges and Future Directions, in Kola- sign Knowing and Learning: cognition in design edu-
revic B. (ed), Architecture in the Digital Age: Design and cation, Elsevier Science, Amsterdam pp 125-146.

Considering Physicality in Digital Models
Andrzej Zarzycki
New Jersey Institute of Technology, USA
andrzej.zarzycki@njit.edu
Abstract. This paper discusses the integration of physical and digital models in the
context of building technology teaching. It showcases projects that explore the design
possibilities of a chosen structural system with the use of parametric and behaviour-based
computational modelling. It uses detailed mock-ups as vehicles to study, optimize, and
evaluate the design as well as to provide feedback for student learning and the direction
in which future designers may engage computational design. Finally, it investigates
digital-to-physical design translations, the importance of which becomes more and more
critical in the context of the current, computer-intensive architectural education and
professional practice.
Keywords. BIM; building information modelling; parametric construction details;
construction assemblies.
INTRODUCTION
With digital tools firmly established in professional ceptualization framework; and the lack of physical
practice and academia, the question of the contin- considerations are just some of the issues waiting to
ued relevance of physical and traditional methods be addressed by the computational creative frame-
is often overlooked or unexamined. Certainly, there work.
are passionate statements being formulated on This paper specifically looks at materiality em-
both sides, with analog thinking more and more bedded in architectural models, their physically
on the defensive. However there is a need for closer based behaviour, and the haptic feedback designer
investigation of the analog-to-digital and digital-to- and makers receive when interacting with their
analog phase changes to further improve the devel- products. The emerging question is what forms of
opment of computational tools and digitally driven digital software and interface would provide a com-
creative processes. parative level of interactivity: what software features
This paper looks at the close integration of and design interface would facilitate full virtualiza-
physical and digital models in design practice and tion of the design process.
investigates the ways both design environments in-
form each other. The goal of this paper, however, is PHYSICAL-TO-DIGITAL TRANSLATIONS
not to justify why we need physical and traditional To research the topic, students investigated struc-
modes of thinking, but rather to point to needs in tural systems that actively informed architectural
the further development of computational design tectonics (form-active structures) and explored their
thinking, which in many aspects is still not up to par design possibilities with the use of parametric and/
with the traditional (not digital) design process. The or behaviour-based computational modelling. Once
intuitive, even haptic, use of tools; a natural con- the research phase was completed, students deve-

Figure 1
Deployable structural frame-
work, conceptual model.
loped a number of physical mock-ups of the final uisites started to point to the solutions of using light
designs to compare their behaviour with computer metal framing with a possible fabric enclosure.
simulations they developed earlier for the same de- In the second stage, students developed a
sign. number of conceptual studies that allowed them
This allowed students to reflect on the material- to apply researched systems into new spatial con-
ity of digitally designed architecture, to understand figurations and test their appropriateness. After
the opportunities and limitations various design developing a number of designs, both physically
tools provide, and to visualize structural behaviour and digitally, students focused on the solution that
in more intuitive and direct ways that available with followed the logic of the Hoberman Sphere. Simi-
digital tools alone. The following examples illustrate larly to Huberman’s design, the student structure
the process and discoveries students made. was capable of folding down to a fraction of its
fully deployed size. It also used a version of a scissor
Adaptive Forms mechanism. Instead of a sphere-like configuration,
A number of projects looked at scissor-like mecha- students experimented with a cylindrical form with
nisms to develop shading and spatially adaptable the ability to expand both vertically and horizontally
systems. The project in Figure 1 investigated a de- by increasing the cylinder radius.
ployable assembly that can be a temporary struc- After completing the chipboard model and
ture or an adaptive space. interacting with it, students realized that the pro-
Initially, students researched various relevant posed structure did not have the desired rigidity
precedents that dealt with temporary, portable, and and durability. Components had difficulty support-
deployable structures. This research gave students ing themselves, resulting in sizable deflections.
insight into different kinetic systems, assemblies, When expanding and contracting the structure,
and material applications. Students were particu- individual components were subject to twisting in
larly interested in the ability for the design form to the joints, resulting in kinetic friction and deforma-
be contracted into a relatively small volume and to tions. While this is rather obvious observation with
have a low total weight. Immediately, these prereq- a model made of chipboard, students also noticed

Figure 2
Deployable structural frame-
work, conceptual model.
possible issues with the actuation of the kinetic as- corporate three-dimensional composition (Figure
sembly. While displacing only some, not all joints, 3). Students similarly started by analysing various
at the same time, the softness of structural com- expandable designs that used scissor-like mecha-
ponents was causing the entire system to deform, nisms. Their focus was on using scissors both as a
putting additional stress on connections and caus- structural element and as an adaptable enclosure/
ing material fatigue. This was important feedback shading. By testing various scissor joint geometries,
for students, since it suggested that the scaled-up they looked at possible shapes and the resulting
structure, even when made with higher-grade mate- planar tiling to provide a variety of expressions of a
rial, may still have similar rigidity and stability issues. façade shading system.
The subsequent study model introduced more The physical and digital explorations revealed a
rigid material (acrylic glass), sized up the cross-sec- number of intricacies, both technical and geometri-
tion of individual components, and doubled vertical cal, that were not immediately evident at the begin-
structural members. The locking mechanism was ning of the project. What seemed like a straight-
added to further stabilize the structure by introduc- forward design quickly became a complex project,
ing triangulation in the vertical supports (Figure 2). particularly when multiple instances of a scissor
The second project started as purely two- mechanism were interconnected into larger as-
dimensional shading system and evolved to in- semblies (Figure 4). The attachment details became
Figure 3
Façade screen mock-ups.

Figure 4
Aggregation of individual
assembly components.
more involved, with diverse rotating and sliding themselves in a fixed position, physical prototypes,
motions occurring within the component connec- due to their relative imprecision and material flex-
tion. The connection had to account for competing ibility, gave a better indication of the overall as-
movements between various sub-elements. One sembly behaviour. They were also more informa-
of the studies employed a three-dimensional ver- tive because they provided a tactile feedback that
sion of the scissors mechanism to form a dome-like helped to advance design. While laser-cut mock-ups
structure (Figure 5). To accommodate the three-di- allowed for a high level of precision, the initial pro-
mensional rotation of scissor plates, students deve- totypes were developed in the more forgiving me-
loped wedge-like adapters to control the curvature dium of chipboard, as compared to later prototypes
of the resultant form. Unlike other groups working made of acrylic glass. This helped to track kinetic
on kinetic designs, this team relied heavily on physi- movements, particularly registering material fatigue
cal models to complement their digital simulations. and failures for further design refinements.
Students felt that the tactile qualities of physical While physical and tactile feedback was im-
models gave them valuable feedback about the lev- portant to the team, there were also limitations in-
els of friction within joints and material resistance. volved in deferring exclusively to physical mock-ups.
Particularly in the situations when digital models It was often difficult to distinguish between minus-
were getting easily over-constrained and locking cule kinetic transformation and fabrication toleranc-
Figure 5
Applying scissors mechanism
to a dome-like structure.

Figure 6
Testing parametrically geo-
metric relationships between
various components.
es as well as material’s ability to hide stress through to parametric models. They were able to implement
deformations (strain). Whereas material deforma- complex assemblies in more informed and intuitive
tions may seem a desirable quality, these deforma- ways.
tions may ultimately lead to material fatigue and
assembly failure. To address these concerns, the de- Kinetic Movements
sign team used parametric digital models to validate Inspired by Theo Jansen‘s kinetic sculptures, stu-
their findings and fine-tune the final set of physical dents investigated parametrically defined adaptive
mock-ups. These parametric models allowed for ef- structures that mimics skeletal systems. They started
fective tracking of numeric values and maintaining with the exact replica, both physical and digital, of
geometrical relationships between various compo- Jansen’s Strandbeest kinetic mechanism. Then, with
nents (Figure 6). parametric models, students investigated how spe-
Once students established a general under- cific component dimensions and radii impact the
standing of kinetic system behaviours, they became kinetic behaviour of the entire system (Figure 7).
significantly more efficient in developing variations Parametric definitions allowed for fluid changes to a
Figure 7
A study of the kinetic behav-
iour of the entire assembly.
Original Jansen’s design (left),
and student design explora-
tions (centre and right).

Figure 8
A kit with multiple parts
worked effectively as an
analog parametric study.
digital model and for immediate feedback on its ki- ing of kinetic mechanisms developed over time
netic behaviour. This helped to understand the role with multiple prototype reiterations. To shortcut the
individual elements played within the entire assem- discovery process, students started with an already
bly and the types of motions these elements were resolved design and investigated ways the logic for
capable to produce. this particular mechanism can be extended to other
While these parametric studies became effec- forms of movement. While a physical working pro-
tive tools in understanding how Jansen’s kinetic totype was an ultimate goal for the project (Figure
sculptures worked, it became difficult to extrapolate 9), it was easier to experiment with variations of the
these findings into new meaningful movements. To base mechanism using digital modelling.
overcome this issue, students started with changing However, conventional three-dimensional mod-
element proportions, folding ratios, and adding ad- elling software was not effective for this type of
ditional components (Figure 8). These speculative prototyping. The design team turned to parametric
explorations led students to propose and develop software, such as Revit (parametric BIM) and Grass-
an adaptable vertically climbing mechanism that hopper (graphical algorithm editor for Rhino), that
used core principles of Jansen’s models with chang- was capable of dealing with constraints and pass-
es to the types of constraints and possible motions. ing these constraints between various assembly
Kinetic designs such as Jansen’s sculptures that components. In addition to these two software ap-
mimic walking structures, or Hoberman’s expand- proaches—parametric and prescriptive—the team
ing dome, require close and detailed understand- briefly looked into VFX packages such as 3DMax and
Figure 9
Kinetic movements, final
assembly.

Figure 10
Rigid origami as an adaptive
envelope.
Maya with inverse kinematics (IK) capabilities. While can be seen as the limitation of the system, not only
IK provided ready-made functionalities that could from the visual but also from the occupancy view-
be applied to walking structures, the inability to di- point, since the origami-generated forms are hard
rectly “hack” into the algorithms behind IK functions to reconcile with horizontal surfaces such as floors,
became a significant deterrent in using them as ex- both because of these forms’ flatness and their
ploration tools. Dealing with actual parameters— changing height. However, they can still be effec-
angle values and component dimensions—allowed tively employed in other enclosure surfaces.
students to get more direct confirmation of their
initial design propositions and develop a stronger LEARNING FROM PHYSICALITY
intuitive feel for the entire mechanism. In discussed cases, students worked with additional
constraints defined by a number of component and
CONTINUOUS ENCLOSURES connection types to simplify manufacturing and as-
With adaptive designs, the issue of the continuous sembly. These became important design bounda-
weather tight exterior enclosure resistant to mate- ries, focusing students on pursuing optimal solu-
rial fatigue is a major challenge. When elements tions and driving questions of component assembly
move or stretch, they wear off connection seals and and functionality. While in some cases students did
may cause material failures. To address these issues, not produce an actual one-to-one mock-up, the
students looked at form-active designs, particularly scaled-down models became effective in setting the
those that deploy tensile (fabric), pneumatic and stage for understanding the overall kinetic system
foldable strategies in conjunction with kinetic as- behaviour and speculating on further development
semblies. of design by giving students direct feedback. The
One of the approaches looked into rigid origami haptic feedback included not only the component
as continuous yet spatially reconfigurable forms that movements but, more importantly, the levels of ma-
do not rely on material deformations (Figure 10). terial resistance to deformations, joint frictions, and
While hinged joints provide opportunities for mate- material fatigue. Additionally, physical mock-ups be-
rial fatigue, the rigid plates are durable, with all the came a lesson in understanding issues of manufac-
performance qualities of traditional wall systems, turing precision and design and construction toler-
including thermal and structural. Since rigid origami ances. These mock-ups allowed students to feel the
solutions carry a particular design signature, the behaviour of the material and the entire assembly in
underlying structural framework would naturally addition to visually understanding its movements.
follow the same geometry, both from performance Furthermore, the discrete numericals used in
and aesthetic considerations. To some extent this defining computational models do not help to un-

derstand the need for and the role of design toler- process of building and rebuilding mock-ups, dis-
ances. This is particularly evident in the kinetic and covering imprecision in produced work, facilitates
adaptive structures, where the movements of indi- the discussion on types of loads (concentric ver-
vidual components can compound the amount of sus eccentric) and moments associated with them.
displacement and rely on material elastic deforma- Students experience first-hand the need for design
tions. The ideal computational models, BIM or not, tolerances and the ways they can be incorporated
should be able to consider these factors, and use in their designs. Overall, students moved away from
them as design constraints and validators. Prefer- idealized computer-based reasoning toward more
ably, these models would provide easily understood, holistic thinking about a building as a probabilistic
intuitive, and perhaps even tactile feedback that structure—a result of compounding imperfections
could not only evaluate but also stimulate design. and tolerances. Also, the use of advanced digital
With today’s generation of designers, who rou- modelling tools such as BIM helped shift the design
tinely have a better grasp of digital than of physical focus from the model itself toward broader and in-
tools, the requirement to manually construct de- terdependent modelling of an assembly or a build-
signs is probably even more important than in the ing (Smith and Tardiff, 2009).
past. Since the architectural profession ultimately
deals with physically constructed buildings, there is SOFTWARE LIMITATIONS
a need for designers to understand the translation Many of the physical-to-digital and digital-to-phys-
process of their ideas from the digital to the physi- ical translations discussed above were partial and
cal. punctuated. This is particularly evident with fold-
There seems to be a perception among many able structures that utilize rigid origami. Origami
students that once a design is modelled in a three- designs are not easily to conceptualize and extrapo-
dimensional virtual environment, it is fully resolved. late. While there are plenty of examples of various
While this may be true from the geometrical point origami designs, they tend to be difficult to extract
of view, as compared to traditional two-dimensional and modify as sources for new designs (Stavric and
representation of buildings where different draw- Wiltsche, 2013). Their design requires a significant
ings did not have to be reconciled spatially, it is not level of involvement and experience. Also, there is
true in other aspects of design. a limited number of software applications that can
The present computational tools solve some of be used to explore origami design, particularly in in-
these issues but still leave many of them unresolved. teractive and kinetic ways. Furthermore, many of the
Specifically, material properties, physical behaviour, software packages are stand-alone applications that
and contractibility continue to remain unaccounted do not port models into other applications.
for in most software packages. While the approach For those projects, the physical modelling ap-
discussed above points to ways of addressing the is- proach was more effective than its digital counter-
sues of material properties and physical behaviour, part. However, the physical models did not stimulate
physical mock-ups prove to be an effective learning tectonic explorations and versioning in the same
environment. By constructing kinetic and adaptive ways as other projects that were realized with com-
designs, students experience the intricacies of me- putational tools.
chanical assemblies and material limitations. While many of the examples discussed here
The geometric precision taken for granted with show only structural elements of the larger assem-
software packages becomes a major issue when bly, the question of how kinetic structures can be
manually constructing kinetic designs. Centre of combined with an adaptive façade or building skin
gravity and points of rotation are important factors system is critical. For this reason, a number of stu-
in the effective operation of kinetic assemblies. The dents looked into form-active systems such as ten-

sile, pneumatic, and foldable structures to see how construction assemblies becomes critical. Material
they can be combined with kinetic assemblies to intelligence (MI) that is largely missing from com-
provide both adaptability and continuity of the en- puter applications, including BIM, sets unnecessary
closure. Consequently, a number of student groups limits to the digital design processes. A file-to-fabri-
adopted a form-active approach towards a façade or cation approach works effectively when the file cre-
a building envelope to work with kinetic structural ator has an explicit, or at least a tacit, understanding
systems, while other students focused exclusively of the materiality of the final design medium (Perez,
on the form-active design for both structure and 2010). In addition to materiality, the physics-based
building envelope. behaviour functionalities would allow for greater
relevance and integration of digital tools in the mak-
DISCUSSION ing of architecture (Zarzycki, 2009; 2011). The meth-
Adoption of digital tools in design serves as an op- odology discussed in this paper provides opportuni-
portunity to redefine existing teaching and prac- ties for addressing material knowledge learning and
tice paradigms. This position is held by a number of the increased convergence between digital tools
practitioners, researchers, and educators who ad- and building construction.
vocate the necessity of the design process change The digital-physical design dialogue is intricate
as a consequence of new computational tools, par- and bidirectional, involving simulations, perfor-
ticularly BIM software (Mayne et al., 2006; Clayton mance analyses, and component optimization. By
et al., 2010). Individual authors propose various de- connecting digital prototyping with physical mock-
sign frameworks. Some of the approaches focus on ups, material becomes an important variable—an-
component-based design thinking, where an overall other consideration in the otherwise parametrically
building is depicted as a combination of construc- driven design process. Material acts as yet another
tion details (Wallick, and Zaretsky, 2009). Others ad- feedback loop that informs design and provides
vocate the role of BIM software in capitalizing on a a set of constraints to guide designers (Cabrinha,
tacit knowledge associated with any creating-mak- 2008).
ing work (Clayton et al., 2010). This study aligns itself Future work will focus on material translations
with the latter approach by emphasizing digital-to- (Decker, 2012) from physical to digital environments
physical translations and using materiality as a feed- and closer interaction between digital and physi-
back mechanism to inform digital tools. cal modes of thinking. Specifically, I am interested
While tacit knowledge is generally acknowl- in simulations of material properties and physics-
edged as critical aspect of the design process based behaviour to develop seamless digital-to-
(AIA,1969), it is also evident that the formation of physical translations.
tacit knowledge (Polanyi, 1983) is associated with
experiential learning and learning-by-doing. While FINAL CONSIDERATIONS
the process of learning-by-doing can be informed The examples discussed in this paper provide a
by both digital and physical making, there is a par- starting point for outlining the software functionali-
ticular benefit from bridging both modes of creativ- ties that are missing from digital tools and digitally
ity. Since the end goal of the design process in ar- enabled design processes. The ultimate aspiration
chitecture is a building or a structure, the ability to is that BIM, or computational models in general, as-
understand the connection between the digital de- sume the role of the virtualized final construction-
sign process and its physical actualization is crucial. ally real designs, with the only difference between
With the emerging robotic applications in archi- digital BIM models (mock-ups) and physical models
tecture, the ability of BIM software to increasingly being that computational models are actualized be-
reflect the reality of the physical world and actual fore a physical structure are built. While this goal is

ambitious, it is also necessary in order to bring digi- pp. 118-125.
tal design process to the level where it can be de- Clayton, MJ, Ozener O, Haliburton J and Farias, F 2010, ‘To-
ployed universally and holistically, independent of wards Studio 21: Experiments in Design Education Us-
the designers’ location and individual capabilities. ing BIM’, in Goosens., K and Agudelo, L (eds), SIGraDi
The present limitations lie both in computation- 2010 Proceedings of the 14th Congress of the Iberoameri-
al software that does not address a number of criti- can Society of Digital Graphics, SIGraDi, Bogota, Colum-
cal design considerations, such as materiality and bia, pp. 43-46.
physically based behaviour, and also in the interface Decker, M 2012, ‘Material Dynamics: Emergent Materials in
designers use to interact with virtual models. Per- Architecture’, Lecture presented at the Inaugural Sym-
haps the latter is more challenging, since it would posium at NJIT, Newark, New Jersey. http://youtube.
require incorporating more sensory and intuitive com/watch?v=h3ffl2kBNKA (accessed, May 21st, 2013)
inputs and breaking away from two-dimensional Kloft, H 2005, ‘Non-standard Structural Design for Non-
displays. Ultimately this would require shifting the standard architecture’, in Branko Kolarevic and Alim M.
computational interface from a mostly visual to Malkawi (eds), Performative Architecture: beyond instru-
amore dimensional feedback system that would ad- mentality, Spon Press, New York and London.
dress multiple-sense inputs. Mayne, T 2006, ‘Change or Perish’, in Broshar, M, Strong, N
Purely geometrically driven digital models miss and Friedman, DS, American Institute of Architects: Re-
many design opportunities. What may seem a fail- port on Integrated Practice, AIA, Washington DC, Sec-
ure of resolving constraints within parametric BIM tion 1, pp. 3-5.
systems with physical models can be a close enough Perez, S 2010, ‘M.I. Material Intelligence’, Proceedings of the
(good enough) solution that material elasticity and Creating-Making Forum, University of Oklahoma, Nor-
tolerance allow to function, giving a designer an im- man, Oklahoma.
portant clue of being in the proximity of a solution. Polanyi, M 1983, The Tacit Dimension, Peter Smith, Glouces-
The binary quality of computational feedback may ter, MA, p. 20
often be misleading, particularly in the boundary Smith, D. and Tardiff, M. 2009, Building information mod-
conditions when investigated designs lie immedi- eling: A strategic implementation guide for architects,
ately outside the zone of computationally correct engineers, constructors and real estate asset managers.
solutions. A more forgiving and probabilistic ap- John Wiley & Sons, Hoboken.
proach associated with physical models and materi- Stavric, M and Wiltsche A 2013, ‘Investigations on Quadri-
als may often be more informative and effective as lateral Patterns for Rigid Folding Structures’, Open Sys-
a design tool. Feeling material behaviour provides tems: Proceedings of the 18th International Conference
a broader design feedback than simple “works” or on Computer-Aided Architectural Design Research in Asia
“does not work” (over-constrained) statements. (CAADRIA 2013), pp. 893–902.
Wallick, K and Zaretsky, M 2009, ‘Fragmentation and Inter-
REFERENCES rogation as an Approach to Integration’, ACSA Proceed-
AIA. 1969, Architect’s handbook of professional practice, ings, Value of Design.
chapter 11: project procedures. The American Institute Zarzycki, A 2009, ‘Dynamics-based tools: an unusual path to
of Architects, Washington. design integration’, ACM SIGGRAPH 2009 Asia Proceed-
Cabrinha, M. 2008, Gridshell Tectonics: Material Values Digi- ings, Yokohama, Japan, (DVD ROM).
tal Parameters, Silicon + Skin: Biological Processes and Zarzycki, A 2011, ‘Form-making Without Form Making’, Pro-
Computation, in Proceedings of the 28th Annual Confer- ceedings of the Parametricism (SPC) ACADIA Regional
ence of the Association for Computer Aided Design in Ar- 2011 Conference, Lincoln, NE.
chitecture (ACADIA), Minneapolis 16-19 October 2008,

Understanding and Managing the Constructive
Characteristics of Vernacular Architecture
Two raw earth dwellings

Danilo Di Mascio
Department of Architecture. G. d’Annunzio University, Chieti-Pescara, Italy
http://www.danarchitect.com
ddimascio@danarchitect.com
Abstract. In this paper a methodology will be presented to investigate and document

the constructive characteristics of two raw earth houses: artifacts that belong to the
vernacular architecture. The comprehension, analysis and documentation of these
architectures presents several problems mainly linked to the impossibility of using a
predefined method, because the difficulties relating to each artifact and its characteristics,
to particular geographic, cultural and social situations, are unique. To understand and
document the constructive features it was decided to realize a three-dimensional digital
reconstruction of the two artifacts, using 3D modeling software. Subsequently several
graphic works have been elaborated (technological breakdown, sheets with detailed
information about the materials, used constructive techniques, etc), useful in managing a
recovery or maintenance project.
Keywords. Vernacular architecture; raw earth dwelling; 3D modeling; digital
reconstruction; knowledge management.
VALUES OF VERNACULAR ARCHITECTURE

In this paper a methodology will be presented to a remarkable value because it is an expression of
investigate and document the constructive charac- the environmental and cultural characteristics of a
teristics of two raw earth houses, artifacts that be- place and of the way of life of its population (May
long to the vernacular architecture. The use of the and Reid, 2010). It defines the identity of a site, its
adjective “vernacular” next to the word architecture cultural diversity, hence it should be preserved for
was first used by Rudofsky (1987) and it means an present and future generations. Furthermore, it
artifact that has been built by common people represents a precious deposit of meanings, materi-
(hence not by professional figures such as archi- als and constructive systems that can be used as a
tects) to satisfy particular needs, such as to shelter source of inspiration during the design and build of
or inhabit. This spontaneous architecture is widely more sustainable buildings.
diffused worldwide: it constitutes about the 90% of Unjustifiably the interest in the application of
the world’s buildings (Oliver, 2003) and it possesses the digital technologies in this architectural sector

is not well diffused, although these constructions OBJECTIVE OF THE RESEARCH
have inspired (and still inspire) many (even famous) The main objective of the research is to define and
architects (such as Renzo Piano for the Jean-marie test a methodology to analyze and document the
Tjibaou Cultural Centre in Noumea, New Caledonia, constructive characteristics of two raw earth houses.
but there are other famous examples, such as Frank Raw earth is one of the most used building materi-
Lloyd Wright or Le Corbusier). They have particular als. The concept of constructive characteristics indi-
spatial and material qualities, which make them cates the constructive system of an artifact, referring
more fascinating than many anonymous modern to the number and type of technical elements and
and contemporary buildings designed by professio- the materials that comprise it, to which require-
nals. ments it corresponds and how they are connected/
assembled.
ISSUES STUDYING VERNACULAR ARCHI- The analysis and documentation of the con-
TECTURE structive characteristics of vernacular architecture is
The artifacts belonging to vernacular architecture interesting for various aspects:
can be broadly classified into various categories: do- • to produce a knowledge (understanding of
mestic, agricultural, industrial, religious, etc. the artifacts) and documentation aimed to the
Each of these categories can be further divided into definition of recovery, maintenance and reno-
subcategories: for instance, domestic architectures vation projects;
can be further divided according to the place (city, • to learn from the past sustainable constructive
countryside, etc.) and to the dimension, but other solutions that can be updated and used to cre-
differences can be taken into consideration (Brun- ate a healthier and more human built environ-
skill, 1988). The technical-constructive solutions ment. These new solutions could be exported
and the materials used in vernacular architectures in other contexts having the same characteris-
strongly depend on the geographic features of the tics, and in this case this should be considered
place where they are located: materials can be found as “technological transfer”.
on site and in surrounding areas, but climatic condi- As time passed by, most of the know-how related
tions, together with social and cultural aspects also to these constructions got lost, including the tech-
play an important role. It is often possible to have nical-constructive knowledge and the capacity of
very different constructive techniques within the organizing a recovery site. It is not always possible
same geographic area. The aesthetic, morphologi- to understand some technical-constructive solu-
cal and technical-constructive variety of vernacular tions by only consulting the existing documents or
houses worldwide is impressive (Oliver, 2003). observing the real artifacts. The digital technologies,
The comprehension, analysis and documenta- and in particular the application of 3D digital mode-
tion of these architectures present several problems ling packages, allow us to investigate and compute
mainly linked to the impossibility of using a prede- many characteristics of buildings that have been
fined method, because the difficulties related to built spontaneously, without a predefined design.
each artifact and its characteristics, to particular ge- Hence we can find out details such as the number
ographic, cultural and social situations, are unique. and position of technical elements, the quantity of
This uniqueness characterizes the use of unique materials used and other technical/constructive
terms that indicate specific technical elements of a characteristics.
particular construction, and for this reason it is nec-
essary to deepen the knowledge of the object that THE CASE STUDIES
has to be studied through documents and the com- For this research two dwellings located in two
munication with local people. countries have been chosen (both belonging to

Figure 1
(Left) - Picture of the dwelling
in Loreto, abandoned and in
ruin (Source: picture provided
by Prof. M.C. Forlani); (right) -
Picture of the inner courtyard
of the house in raw earth
(Source: picture provided by
the working group of Africa
‘70 project).
the Mediterranean area). In this way the similarities village) there are collective structures such as
and the differences between the two constructive barns, shops, religious buildings and private
systems have been compared. The two artifacts be- homes. The houses have courts and they are
long to the same category, domestic building and all juxtaposed to form a compact urban tissue
have a similar size. Both case studies analyzed by which facilitates defensive actions. The main
the author belong to wider researches coordinated constructive material for all the structures is
by Prof. M.C. Forlani (G.d’Annunzio University, Chieti- adobe (the composition is similar to the cob,
Pescara, Italy). the difference is that the dough is shaped into
bricks (using frames) and dried in the sun; in
General description some cases it is possible also to find cut stone.
• Loreto (Abruzzo, Italy): The house (Figure 1
left) is located near Loreto, a small village in the Issues to be faced
countryside of the Abruzzo region, in central • Loreto (Abruzzo, Italy): The artifact is aban-
Italy. In this region there is a long tradition of doned and in ruin. Due to different climatic
raw earth buildings and to date a number of conditions (mainly snowfall and rainfall), the
over 800 artifacts have been surveyed (Forlani, artifact is badly damaged, the roof and the
2011). The typology of the artifact is a tower floor are partially collapsed and also the pe-
type, like many other buildings in raw earth in rimetral walls present fissures. Moreover, the
the region, and it is located in a rural area with dwelling is surrounded by vegetation, in par-
no other buildings nearby. The constructive ticular blackberry bushes. With this situation it
technique is the cob (made by clay, sand, straw, was dangerous to enter inside the building or
water and earth), but other dwellings in the re- to move around easily and take more informa-
gion are also made by rammed earth. tion and pictures.
• Figuig (Marocco): The analyzed building (Fig- • Figuig (Marocco): In this second case the main
ure 1 right) is located in the city of Figuig in difficulty was due to the impossibility of visit-
the eastern area of Marocco, at the border with ing the place personally, together with the ne-
Algeria. The city, built around an oasis, consists cessity of analyzing a non-standard building
of seven ksour which are typical fortified vil- system.
lages in north Africa. In a ksar (that is a single

METHODOLOGY of the artifact. During a process of digital recon-
Both situations required the use of digital technolo- struction there are many choices to be taken.
gies of 3D modeling. To study and carry out the two During his Ph.D. research, the author examined
digital reconstructions and create a series of graphic and underlined the importance of some aspects
works to document the constructive peculiarities of that have to be considered when undertaking a
the two raw earth houses, it has been defined a simi- three-dimensional digital reconstruction process,
lar methodological process that can be synthesized namely: abstraction, geometry and organization of
in four main phases: the model. The model that results from the digital
• Collection of documents; reconstruction is, like any other model, a simplified
• Analysis of the whole available documentation; description of a system [1].
• Digital reconstruction (interpretation, mod- In the analysis of the artifacts and in the digital
eling and organization of the elements); reconstruction as reference a classification scheme,
• Representation and organization of the infor- namely the Italian UNI norms (UNI Norm 8289/2,
mation in worksheets. 1981 has been used. In particular, these norms have
been a useful support to the understanding and
The digital reconstruction phase classification of a complex system made up of a hi-
In order to understand and document the con- erarchical structure of the technical elements. The
structive features of an artifact, the most appropri- norms have been properly readjusted to describe a
ate choice is the realization of a 3D digital model. non standard building and only the classes of tech-
The traditional two-dimensional drawings (plans, nological units available in the artifact have been
elevations and sections) do not allow an analysis, a selected.
representation and a proper communication of the
constructive features. These graphic works are an The case study of Loreto
ineffective theoretical description of the informa- The information gathered about the dwelling in
tion concerning the technical elements and their Loreto are the results of some previous researches,
relations in the three-dimensional space: this in- including a photographic documentation, a two-
formation can instead be provided by a 3D model. dimensional survey in Autocad, that includes plans,
Moreover, the two-dimensional drawings have elevations and sections, and worksheets with the
some inconsistencies which document and commu- information about technical elements (floors and
nicate the information in an unclear and wrong way, walls) belonging to other raw earth dwellings of the
even because of difficulties caused by the irregular- region.
ity of the technical elements that come from a non The dwelling, made of just two rooms is located
industrial production (hence not standardized). The on a slight slope and it is partially recessed in the
two-dimensional drawings are not adequate for ground. This characteristic affects the access to the
studying irregular morphological-constructive arti- two rooms positioned on opposite walls and at a
facts, because in these situations they appear to be different layout level. The room on the ground floor
excessively approximate. was used as stable/warehouse, while the room on
The expression digital reconstruction means a the first floor represents the real dwelling and it was
process that foresees the action of building again as used as kitchen and bedroom.
an existing artifact (or an artifact that no longer ex- Even in this case, like in the digital reconstruc-
ists), in a virtual environment. This process possesses tion of the Figuig dwelling, the two-dimensional
an autonomous value, independent from further drawings (properly cleaned of irrelevant informa-
analysis, because the same reconstruction process tion and reorganized into layers) have been really
contributes to deepen and broaden the knowledge important within 3D Studio Max. The first elements

Figure 2
(Left) - the constructive system
of the floor (internal view of
the room on the first floor);
(right) - rendered view of the
constructive system of the
pitch roof.
to be modeled have been the four perimeter walls, ters) in wood. The sills on the outside of the open-
built with the technique of the cob and tapered up- ings over the windows are in brick tiles (“pianelle”).
ward. The wall surfaces, because of the taper effect, The pitch roof has a bearing structure with a double
are therefore inclined both inside and outside of the wooden grid. On the second structural grid, simi-
artifact. Three sides of the building are mutually or- larly to the floor, is positioned a lattice of rods that
thogonal, while one side has a different inclination. constitute the base of a mantle in earth-straw. The
The influence of this detail in the technical-con- external layer is made up by clay roof tiles (Figure 2
structive solutions clearly emerged during the study right).
of the floor. The floor consists of a double frame of In order to provide some quantitative data on
wooden beams: a main structure (beams of square the number of elements for this house two exam-
section, 15 cm x 15 cm) and a secondary one per- ples have been chosen: the tiles, about 310 for the
pendicular to the first (rafters of rectangular section, pavement of the ground floor and about 366 for the
8 cm x 3 cm). The beams are partially embedded in flooring of the first floor (where have also been used
the masonry load-bearing and probably the rafters on the input threshold of the house) and roof tiles,
at both ends are embedded in the walls. The space about 1007. Obviously, this numerical and quanti-
between the beams is always constant, the only tative information is indicative because there is the
exception is the resulting surface between the last awareness of being in the presence of elements
rafter parallel with the rest of the secondary and not which do not result from manufacturing, and that
the one built into the wall at right angles to the oth- can also significantly vary in size from one another.
ers. On the secondary structural grid is positioned a For both projects a set of textures to communicate
lattice of rods which constitutes the basis of a layer the materiality of the technical elements has been
of clay (about 8 -10 cm thick) on which it is resting created.
in the flooring tile brick (called “pianelle”) of rectan-
gular shape (Figure 2 left). The structure in raw earth The case study of Figuig
did not allow large openings in the walls, which is The collected basic information include a photo-
why the number and size of windows are limited. graphic documentation, technical data on slabs and
On the first floor there are three windows, two on roofs, a two-dimensional survey in Autocad, which
the ground floor. The openings have frames (lintels, includes plans, elevations, and a section with the
jambs and sills) and fixtures (including the shut- description of the materials. The artifact of study is

Figure 3
(left)- rendered image of two
karnef; (right) - closure of the
slab in karnef on palm wood
beams.
inserted within a compact urban settlement, char- This wide margin of difference comes both from the
acterized by buildings with courtyard. The walls bor- variable dimensions of the environments, and from
der with other houses - therefore they are common the irregular thickness of the wall (Figure 3).
walls - or with paths, some of which are covered. In order to cover all the spatial units in a con-
The building has a courtyard and three levels, two structively rational way and to understand the tech-
of which are practicable: the ground floor, the first nical problems that could arise from this construc-
floor and the roof/terrace. tive technique, the beams and the karnef have been
The floor plans in Autocad have been cleaned by manually positioned, as if they were actually build-
dimensions, crosshatchings and other non essential ing the construction in a traditional way (Figure 4).
details, and imported in the 3D modeling software, Each beam has been rotated, spaced from the pre-
3D Studio Max. The use of a 3D modeler, instead of vious and adjusted in its length to correctly adapt
CAD software used for precision drawing like Au- itself to the spaces that have to be covered and to
toCad, is the proper tool to model and manage a adjust the upper layer of the karnef. The position of
non standard artifact composed by a high number the karnef on the same row is alternated and each
of objects on screen. Almost all the basic elements row of karnef is mirrored with respect to the previ-
have been created through extrusions, and the most ous row.
relevant exception is made of karnef, semi-triangular The digital model also allows to calculate/ as-
wooden elements that constitute the base of the sume the quantity of some technical elements
palm trees. necessary to build intermediate floors between the
This digital reconstruction investigated and ground floor and the first floor (both floors of the
documented the constructive aspects of the artifact, court and those of the individual rooms): about 286
therefore for this reason special attention was paid palm timbers on which are placed about 4209 kar-
to the analysis of the grid of the slabs. The grid is nef (1951 for the ceiling of the court and 2258 for
made of palm wood beams and the lower closure of indoors).
the roof in contact with the beams is in karnef, which
are elements that also play a structural function. REPRESENTATION AND ORGANIZATION
The span between two beams where the karnef are OF THE INFORMATION IN WORKSHEETS
placed is of about 34 cm; the karnef have an average In both digital reconstructions all the technical ele-
size of 30 cm and they are between 3 and 5 cm high. ments have been grouped and divided per layer,
The length of the palm wood beams varies between according to categories of homogeneity, belonging
a minimum of 140 cm and a maximum of 270 cm. to a floor plan or a specific field. The first prepared

Figure 4
Rendered image of the grid
and closure of the roof of first
level.
graphic work is the representation of the techno- to combine all the information gathered and elabo-
logical breakdown: an axonometric exploded view rated during various inspections and researches.
of different levels of the artifact, from the ground
floor to the slab of the roof. The levels are related to DISCUSSIONS OF THE RESULTS AND
re-adapted categories, from the UNI Norm 8290 of FINAL REMARKS
the technological breakdown. This scheme clearly The realization of the digital models of the two
communicates the affiliation of each layer to one dwellings in raw earth located in Marocco and Italy
or more of these categories, and therefore its func- have allowed an in-depth study of their construc-
tion. Other graphic works are made of analysis work- tive systems and to hypothesize technical solutions
sheets of the main technical elements. For Figuig adopted in some critical points which could be dif-
the masonry (Figure 5 left) and the slab (Figure 5 ficult to analyze and document in other ways. The
right) have been documented. The worksheets have produced information is useful to undertake tech-
detailed information about the materials, the con- nically efficient interventions on the built environ-
structive techniques, the constructive phases and ment, without compromising the local architectural
the performances of materials and technical ele- and constructive culture. As for other researches un-
ments, referred to specific classes of requirements. dertaken by the author in the field of digital media
The methods used to document the different con- and cultural heritage (Di Mascio, 2009), the three-
struction phases, called evolutionary characteristics, dimensional model was recreated in 3D Studio Max,
have been described and analyzed by the author in because the digital reconstruction has required
other publications (Di Mascio, 2012a; 2012b). The tools able to guarantee a better control of the fre-
obtained information has been reorganized taking quent modifications and to quickly visualize various
into account the possibilities of realizing a database hypothesis on the adopted constructive solutions.

Figure 5
(Left) - Technological
breakdown of the raw earth
dwelling of Figuig, with the
indication of the elements
according to the classification
scheme proposed by the UNI
Norms;(Right) - Analysis card
of the technical elements:
Slab.
In these cases a certain level of creative interpreta- Most of the literature available on vernacular ar-
tion is also necessary, because there are still many chitecture consists of descriptive texts, pictures and
information gaps to fill. This methodology is particu- drawings (sketches and two-dimensional reliefs).
larly suitable in the study and analysis of non-stand- The use of 3D digital reconstructions is another
ard artifacts pertaining to vernacular architectures step forward in observation, analysis, documenta-
as for example the Turchinio’s trabocco (Di Mascio, tion and management of these artifacts, because it
2009). During the digital reconstruction work a ma- allows the investigator to assess and monitor addi-
jor technical problem due to the high number of 3D tional parameters, such as spatial and quantitative,
elements used for the roof tiles, the karnef and the that methods and tools so far available did not al-
tiles of the floor has also been tackled. low to analyze with the same precision and effec-
The two case studies, although they have ma- tiveness. Obviously the study and testing of these
terial and technical-constructive characteristics in digital technologies must be accompanied by a
common, clearly present evident differences, mainly continuous processing and verification of theoreti-
due to matters linked to their geographic localiza- cal and methodological apparatus. The introduction
tion, but also to different life styles: different roofing of new methods and tools always leads to a criti-
systems, the presence or absence of windows, differ- cal re-evaluation of what has been done so far and
ent organization of the spaces / environments, dif- the opening of new avenues of study and research.
ferent materials used, etc.. In the study of vernacular architecture (as in many

other fields of knowledge) it will be increasingly im- ing the digital reconstruction. A continuous testing
portant to become aware of the importance of mul- of additional case studies will be very useful to im-
tidisciplinary approaches in order to bring the stud- prove and expand the theoretical and methodologi-
ies in this area to a higher level of detail and quality. cal approach, and of course the use of various digital
Effective collaboration between specialists from dif- tools.
ferent disciplines (architects, engineers, historians, This approach can be used in an efficient way to
anthropologists, archeologists, geographers, and rebuild and analyze many vernacular architectures
many others) will largely depend on the ability of in a digital environment; this refers in particular to
everyone to communicate to others in a clear and huts, villages on stilts, buildings in stone, raw earth,
effective way information related to their area of ex- and many other pertaining to different cultural and
pertise. Communicating the technical-constructive geographical environments worldwide.
information only through two-dimensional draw-
ings and texts could limit the comprehension and ACKNOWLEDGEMENTS
hence the capacity of other specialists to contribute I would like to thank Prof. Maria Cristina Forlani for
to the research, influencing negatively the final re- having involved me in these research projects.
sult. The methods and analyzing instruments in the
field of vernacular architecture are still too compart- REFERENCES
mentalized in the single disciplines and this could Brunskill, R W 1988, Illustrated Handbook of Vernacular Ar-
be limiting (Oliver, 2006). chitecture, Faber and Faber Limited, London, United
Kingdom.
CONCLUSIONS Di Mascio, D 2009, Digital Reconstruction and Analysis of
Within this research methods and tools to improve Turchinio’s Trabocco: A method of digital reconstruc-
the comprehension and documentation of two raw tion of a complex structure as a way to improve our
earth houses have been investigated, elaborated knowledge of a cultural heritage artifact, Digitizing Ar-
and proposed. During the digital reconstruction chitecture: Formalization and Content [4th International
phase it has been possible to interpret, suppose and Conference Proceedings of the Arab Society for Computer
document technical-constructive solution, that, in Aided Architectural Design (ASCAAD 2009)], Manama,
the best situations, could be understood only after Kingdom of Bahrain, 11-12 May 2009, pp. 177-189.
several inspections, which could not be performed Di Mascio, D 2012a, ICT in the knowledge and in the documen-
in both cases. tation of the peculiarities of the historical and contempo-
The reconstruction of a three-dimensional digi- rary built environment. Methodological paths and case
tal model can improve and expand the knowledge studies, PhD thesis, Scuola Superiore “G. d’Annunzio”
of vernacular architecture in order to improve the School of Advanced Studies, G..d’Annunzio University,
understanding, documentation, management, and Chieti-Pescara, Italy.
conservation and simultaneously stimulate and Di Mascio, D 2012b, Digital mediated techniques for the
evaluate the use of technical solutions and new con- knowledge of the evolutionary peculiarities of the
structions. Digital technologies and new theories built environment: Three case studies, CONVR2012
and methodologies resulting from them are not in- (Proceedings of the 12th International Conference on
tended to replace traditional methods and tools, but Construction Applications of Virtual Reality), Taipei, Tai-
to complement and help to achieve new levels of wan, 1 -2 November 2012, pp. 37-46.
accuracy and in-depth information and reflections. Forlani, M 2011, Conservation strategies of Abruzzo’s histor-
Textual descriptions and two-dimensional graphic ical and cultural heritage/Un patrimonio storico-cul-
works alone are not able to trigger some technical- turale in Abruzzo: le strategie di recupero, in S Mecca,
constructive problems that come to light only dur- S Briccoli Bati, M C Forlani, M L Germanà (eds), Earth/

Lands (Terra/Terre). Earthen Architecture in Southern Ita- troduction to Non-Pedigreed Architecture, University of
lia / Architetture in terra nell’Italia del Sud, Edizioni ETS, New Mexico Press, Albuquerque, New Mexico, United
Pisa, Italy, pp. 226-245. States.
May J, Reid A 2010, Buildings Without Architects: A Global Schmitt, G 1998, Information Architecture, basi e futuro del
Guide to Everyday Architecture, Rizzoli International CAAD, testo&immagine, Torino, Italy.
Publications, New York, USA. UNI Norm 8289/2 1981, Edilizia residenziale. Sistema tecno-
Oliver, P 2003, Dwellings, Phaidon Press Limited, London, logico. Analisi dei requisiti / Residential building. Building
United Kingdom. elements. Analysis of requirements, June 1981.
Oliver, P 2006, Built to Meet Needs. Cultural Issues in Vernacu-
lar Architecture, Architectural Press (Elsevier), Oxford, [1] Model, in Oxford Dictionaries, The world’s most trusted
United Kingdom. dictionaries, <http://www.oxforddictionaries.com/>
Rudofsky, B 1987, Architecture Without Architects: A Short In- (last accessed 29 May 2013).

3D Digitization in Architecture Curriculum
Luís Mateus1, Victor Ferreira2
CIAUD, Faculdade de Arquitectura, Universidade Técnica de Lisboa
1
http://www.fa.utl.pt/~lmmateus, 2http://www.fa.utl.pt/~victor
1
lmmateus@fa.utl.pt, 2victor@fa.utl.pt
Abstract. In this paper we describe an experience undertaken in the Faculty of

Architecture of Technical University of Lisbon, concerning the introduction of a 3D
Digitization course in the frame of the PhD doctorate program in Architecture and in the
frame of the Master’s programs in Architecture, Urbanism and Design.
We start by describing the theoretical and instrumental frameworks proposed. Then
we describe and discuss a set of two exercises developed during one semester, giving
examples of the work produced by the students. Finally we end with some considerations
to be taken into account in future editions of the course.
Keywords. 3D digitization; architectural recording; laser scanning; digital
photogrammetry; teaching and learning.
INTRODUCTION
In the book Digital Design Media, Mitchell and Mc- for the possibility of doing themselves the work, to
Cullough (1995) describe a design studio fully inte- use freeware and open source software as much as
grating traditional and digital media. Basically the possible in all the workflow from data processing to
Building is at the centre and several paths are dis- the final models scaling and orientation, and to pro-
played between the building and its possible forms pose a theoretical framework for the teaching and
of representation, being Digital Models one of those learning process of this subject among architecture
forms. According to the presented and well known students.
scheme, the shortest path between the building and 3D Digitization was offered as an optional
the Digital Model is through electronic surveying. course at our University. It could be attended by Ar-
This electronic surveying encompasses what can be chitecture, Urbanism and Design master students
designated as 3D Digitization. during their fourth or fifth year of their studies, and
There are multiple techniques of 3D Digitiza- by PhD Architecture students during their first year.
tion, passive and active (Lillesand et al., 2004), range This paper is structured in six sections: a) a short
based and image based (Remondino, 2006), using history and related work, b) the theoretical frame-
different kinds of light, etc. work, c) the instrumental framework, d) the practical
This paper describes an experience of imple- exercises proposed, e) Results of the exercises and
menting a 3D Digitization course in Architecture discussion, and f ) conclusions and further work.
curriculum.
The goals of this experience were to make ar- A SHORT HISTORY AND RELATED RE-
chitecture students familiar with 3D digitization SEARCH
techniques and tools, to develop their awareness Traditionally the survey of the built environment

was driven to the production of 2D deliverables. This of some kind of merging between the fields of pho-
was particularly truth before digital technologies togrammetry and computer graphics.
become a common place. The techniques used were In the academic field, in the context of architec-
the traditional manual and topographic survey, ana- ture, the insertion of digital technologies opened a
logic photography and analogic photogrammetry, Pandora’s Box and an endless, and sometimes mean-
that relies in the human faculty of perceiving in ob- ingless, discussion about the harms and benefits of
jects 3D, from pairs of images (anaglyphs), with the the digital (Mitchell and McCullough, 1995).
aid of some specific instrumental apparatus. Some authors (Duarte, 2007; Duarte et al., 2010)
Photogrammetry was then a very expensive and have presented successful experiences about the
specialized technique, and very time consuming, insertion of digital technologies in the architectural
since all the work was done manually. One of the ob- curricula and even mention 3D scanning as part of
vious fields of application was the survey of monu- that insertion (Pupo et al., 2008) in the context of
ments surfaces. digital fabrication.
During the 80ies and 90ies, digital technolo- On the other hand there is a tradition that fol-
gies become into broader use. Photogrammetry lows from the survey field, often linked to architec-
meets another stage, the analytic, meaning that a tural heritage documentation or archaeology, that
computer was used to translate into numerical for- nowadays, fully use 3D digitization techniques, such
mat the analog inputs of the operator. In this way, as terrestrial laser scanning (TLS) or digital photo-
although the restitution process was still manual, it grammetry (DP). In this context it is common to see
became possible to store 3D data in digital format. courses of Architectural Photogrammetry often sup-
This was one initial form of 3D digitization as some- ported by Photogrammetry or Survey Labs [1] [2].
thing more direct between the object and its 3D We understand 3D digitization as a fusion of
digital model representation. By the same time the these two views, the tradition of survey and photo-
topographic survey with electronic distance meas- grammetry and the field of Computer graphics (Re-
urements devices was also in use, but it could only mondino, 2006). As being part of a school where we
provide a much more discrete digitization that was have a diversified range of academic offers, going
generally used to set ground control for the photo- from Fashion Design and Design to Architecture and
grammetric survey or for the traditional manual sur- Urbanism, we think that 3D digitization techniques
vey (Mikhail et al., 2001). can be used with benefit in all these scales, not just
In the beginning of the nineties optical triangu- in a mere instrumental way, but adding new reflex-
lation scanners time-of-flight laser scanners were ive possibilities to the praxis. That was the motto for
introduced. And at the beginning of this century a the creation of the 3D Digitization course.
new kind of scanners, phase-based, much faster, be-
come also into use. Dense point clouds changed the THE THEORETICAL FRAMEWORK
way 3D digitization was done. The theoretical framework focuses on acquaintance
Simultaneously, advanced image processing with the basic knowledge of concepts by the stu-
techniques were developed that enabled the use of dents that allow them to operate in a proper way in
2D images to produce 3D dense point cloud models an architectural context, mostly from a user stand-
in an almost unattended way. point. This is done in a three step approach going
With the increasing power of computation ca- from the principles level, to the guidance level and
pabilities, and the sophistication of software, it be- finally to the specification level (Wu and Di, 2009).
come more feasible, easier, accessible and cheaper These three levels work as a metaphor both for the
to deal with bigger amounts of data, and in general, teaching and learning and for the practical framing
to access to 3D digitization. This was also the result of the work.

First we start by discussing the 3D digitization hardware capabilities. The third option implies that
techniques and methods in the broader context of all the processing is done locally what obliges to
recording techniques (Boehler and Heinz, 1999), have better hardware capabilities but also allows a
and these as a tool for the understanding of the more effective control of the process.
built structures (Letellier, 2007). At this level is also Both for TLS and SFM/MVS point clouds process-
important to notice what kind of 3D models can be ing, it was recommended MeshLab [7] and Cloud-
produced and to understand in what contexts dif- Compare [8] as well as Notepad++ [9] and Libre Of-
ferent models are used (Chader, 2008). This is in the fice Spreadsheet [10] for some manual manipulation
principles level. of matrices.
Then we selected a subset of the 3D digitization For the estimation of transformation param-
techniques as a subject to address in more detail. eters, namely for scaling, rotation and translation of
Namely we selected terrestrial laser scanning (TLS), the point cloud and mesh models, we recommend-
as an active range based technique, and photo- ed JAG3D software [11].
grammetry as a passive image based technique, in This is a collection of freeware software, some
particular with the structure-from-motion/multi- open source, totally available in Internet, what
view-stereo – SFM/MVS (Snavely, 2008; Furukawa means that all the steps of the workflow, from the
and Ponce, 2009) approach. We discussed the typi- point clouds processing from acquisition to the final
cal workflows and its usual steps and concerns, such textured mesh models, can be done at no cost.
as planning data acquisition, data processing, model This means a bigger democratization of the ac-
refinement and model orientation. With respect to cess to digital recording media. The exception to this
model orientation we gave an overview of matrix rule is the parsing of terrestrial laser scanning point
representation of the transformations of scale, rota- clouds from the proprietary format of the scanner
tion and translation and how to estimate the trans- supplier to a more common format such as PTX or
formation parameters from homologous sets of data PLY, which was done with proprietary software.
points. This is in the guidance level. Since we own a FARO Focus 3D laser scanner, we
Finally, the specification level corresponded to used Faro Scene to solve for this step.
the statement of terms that students had to follow
to accomplish two practical exercises. THE PRACTICAL EXERCISES PROPOSED
Two practical exercises were proposed to the stu-
THE INSTRUMENTAL FRAMEWORK dents.
The instrumental framework comprises a hardware In the first exercise, the students choose an
part and a software part. Both parts are intercon- architectural or sculptural detail from which they
nected and influence each other. So, it was neces- should be able to produce a textured mesh model,
sary to make sure that all students could accomplish without holes. This was supposed to be done solely
their exercises even if some kind of limitation arises. from photographic imagery. To scale and orient
For the SFM/MVS approach we recommended, the produced model, two homologous data sets
but didn’t limit to, three operative alternatives: a) of points are to be used. One is obtained from the
My3DScanner [3], b) Photosynth [4] + Photosynth point clouds and the other should be the result of
toolkit 11 [5], and c) Visual SFM [6]. The first option independent measurements done directly in the se-
implies that all the processing of images is done on- lected object or scene. In addition to the models, in
line, so it suited to students that have severe hard- the end of the exercise, the students should submit
ware limitations. The second option allows doing a report describing and justifying their options dur-
some steps of the processing online whilst other ing the work.
steps are done locally, so it is suited to medium In the second exercise, a common set of point

Figure 1
Left: illustration of a reference
frame. Center: Cameras rela-
tive orientation. Right: Dense
point cloud.
clouds, from a terrestrial laser scanning survey of a The statement of the exercise consisted of 10
building, was given to all students. They were asked steps.
to follow a described procedure to align the point • Defining a reference frame in the object. In the
clouds, to dissipate accumulated errors, and to pro- simpler form it could be set out as a couple of
duce a final merged and sub-sampled point cloud measurements with a measuring tape in a flat
model. This contributed to a broader investigation rectangular surface from which one could re-
on the development of an expeditious method for trieve the coordinates of at least four control
dissipating what we designated as a closure error points (CP) as it can be seen in Figure 1 (left
that arises from the alignment of a closed ring of and center). Although control points should be
point clouds. widely spread about the object, for the exercise
it was allowed to consider them more locally,
RESULTS AND DISCUSSION whilst noting that the first procedure is better
In this section we describe the workflow that was since it diminishes the scaling and orientation
adopted to do the two aforementioned exercises. error.
While describing the steps of the exercise we will • Image acquisition. This step involves the un-
also describe the competences that were supposed derstanding of the SFM principle. Images need
to be acquired by the students as well as the difficul- to be taken with small base distances, what
ties that were felt. means that the camera view point in space
must always be changing and large amount
First exercise - 3D digitization small or of images of images with a high level of re-
medium scale object using image based dundancy is to be taken. It is noticed here that
techniques there is no need to use high resolution imag-
The objective of the first exercise was to produce a es. In fact it is better to use more images with
textured mesh model of a small or medium scale ob- less resolution and adopt a hierarchical strat-
ject in an architectural context using automatic im- egy while taking the pictures. This means that
age based techniques following the structure-from- several rings of images at different camera/
motion (SFM) principle. First of all it was declared object distances should be considered. This im-
that this kind of techniques have severe limitations plies another constraint to have in mind when
in their use to record poor textured surfaces or very choosing the object to digitize; the surround-
reflexive ones, since they rely on the texture of im- ings of the object have to be accessible.
ages to recover 3D information. This fact should be • Image processing is the step where colored
understood as constraint about the kind of object dense point clouds are generated. Although
that should be chosen; rich texture objects were this is done automatically we believe that the
more adequate. correct approach is to provide to students

some insight into the phases of this step so this least squares criterion.
doesn’t look like a black box. To understand • After the final point cloud is oriented, it follows
what happens during this process, even if it is a decimation operation that results with the
at a superficial level, helps the students to ac- purpose to have an even spatial resolution.
quire the needed vocabulary to discuss with • Then a mesh model is created.
experts, in in the future specialized services are • If there are any holes in the mesh they should
required, and helps bridging the gap with the be filled.
service providers. This phase includes the fol- • And finally the color of the points is transferred
lowing sequence controlled by a set of param- to the mesh to produce more appealing and re-
eters that determine the quality of the results: alistic models, as it can be seen in Figure 2.
a) SIFT, b) image matching, c) cameras rela- One of the outcomes to achieve with this exer-
tive orientation, and subsequent sparse point cise was to understand that, for many purposes in
cloud generation (Figure 1 right), and d) dense the architecture framework, one can obtain reliable
point cloud reconstruction. 3D data without the use of very expensive hardware
• If the outcome of the previous step is more and without having to hire specialized services. At
than one point cloud independently placed, the same time, by taking into account the restraints
then it is necessary to proceed to the relative that were mentioned, students are capable to un-
orientation. Usually, according to the charac- derstand the potential and limitations of the SFM
teristics of the objects and procedures, it is ex- approach.
pected that the generated point clouds share
the same coordinate frame, that is, the rela- Second exercise - relative orientation of
tive orientation comes as a result of the cam- terrestrial point
eras relative orientation. Relative orientation of The objective of the second exercise was to ori-
point clouds will be discussed in more detail in ent a given set of 16 terrestrial laser scanning (TLS)
the next exercise. point clouds forming a closed loop around an exist-
• Following from the previous step, if one has ing building. The point clouds were acquired with a
multiple point clouds, they should be merged FARO Focus 3D. They were parsed and decimated to
to produce a single model. This is important if 1cm spatial resolution a delivered to the students in
we want to produce a final “water proof” model the PLY format. This allowed to have smaller files and
that can be used, for example, in digital fabrica- made the process more feasible. The basic concerns
tion. Before merging, spurious data should be with this exercise were to compare the quality of TLS
removed from the point clouds. and SFM/MVS point clouds, to provide a way to eval-
• External orientation is the operation that re- uate the quality of the orientation process through
covers scale and orientation of the model in the analysis of a closure error that arises when one
the reference coordinate frame. In fact the con- tries to close the loop, and to provide a method for
trol points generated in the first step suffice to the acceptance and correction of that error by dis-
solve for the seven parameters of the Helmert tributing its translation component proportionally
transformation, that are one scale factor, three to the distances between the origins of the frames
rotations and three translations. This is the of the point clouds.
step where the students with less Math’s back- The statement of the exercise consisted of 5
ground feel more uncomfortable. So it is need- steps.
ed to do some extra exercises to explain how • First the point clouds have to be cleaned. Data
homologous data sets can be used to estimate that presumably didn’t remain stationary be-
transformation parameters, and to explain the tween scans had to be removed. Otherwise the

Figure 2
Some of the models that
resulted from the first exercise.
registration (orientation) errors would increase. dent should visually verify if the quality of the
This task was divided by all the students. registration. This can be easily done by inspect-
• Then, the set of cleaned point clouds was ori- ing the point clouds with at least two mobile
ented by a specific order and criteria; each plans with different orientations and analyzing
point cloud should only be registered with the the section that they produce in the model.
previous one and a particular point cloud was • When closing the loop, that is, when orienting
set as the reference frame, what means that its the last point cloud (that is simultaneously the
position is given by an identity matrix. This way, first) with the previous one (the last of the set)
the final results could be compared. After each its matrix position represents the accumulated
registration step, after optimization with the error. We refer to this as the closure error as in
iterative closest point (ICP) algorithm, the stu- topography (Casaca et al., 2000).

Figure 3
One of the point cloud models
that resulted from the second
exercise.
• If the closure error is found to be under the ac- CONCLUSIONS AND FURTHER WORK
ceptable tolerance, then we can consider that The practical results obtained showed that, with a
no gross errors occurred and we can distribute minimum of theoretical framing, satisfactory results
the error through the poses (matrices) of the can be obtained by agents that are not usually from
point clouds as mentioned above. the field of specialized surveying, and almost with
• Finally the point clouds were merged and deci- no costs. We notice that, at the PhD level some of
mated to produce the final model. the students were architects that didn’t even knew
In Figure 3 we present an example that resulted about the existence of some of the methods and
from the second exercise. tools discussed.
With this exercise it was possible to verify the It was also interesting to notice that that fact
need to be careful when dealing with multiple point didn’t prevent them from accomplishing all the ex-
clouds because if the work is not well controlled, it ercises with high quality results. This means a shift
is easy to accumulate large errors or even to make in the paradigm for architectural recording and
blunders. Students are then told that there are com- means bridging a gap between fields of knowl-
plementary methods, such as topography, to con- edge that traditionally were separated. At the end,
trol the overall quality of the process. But it is also it was achieved the idea that new possibilities arise
noticed that for medium scale objects, such as the by adding these methods and tools to the architect
church surveyed, it is possible to use TLS as a stand- toolbox. It was possible to understand that the tech-
alone method. As with SFM/MVS it was also under- niques presented can be used alone or together or
lined that there are surfaces that aren’t good can- complemented with other techniques, such as to-
didates for laser scanning recording. Among those pography or manual survey.
there are glassy surfaces or low reflectance surfaces. We also noticed that, in some cases, the lack of

mathematical background of some students caused Furukawa, Y and Ponce, J 2009 ‘Accurate, Dense, and Ro-
some difficulties when dealing with matrix notation bust Multi-View Stereopsis’, IEEE Transactions on Pattern
and when operating with geometrical transforma- Analysis ans Machine Intelligence (PAMI 2009), volume:
tions via matrices. The issue was easily overcome 32, issue: 8. pp. 1362-1376.
with complementary materials and exercises where Letellier, R 2007, Recording, Documentation, and Information
those difficulties were addressed. This is unavoid- Management for the Conservation of Heritage Places -
able since there are some degrees that don’t of- Guiding Principles, The Getty Conservation Institute,
fer mathematics on their curricula and the course Los Angeles, USA.
doesn’t have any prerequisite. Snavely, N 2007, Scene Reconstruction from Internet Photo
The contents of this course focused in the point Collections, PhD thesis, University of Washington, USA.
cloud processing towards the production of scaled Mikhail, E, Bethel, J and McGlone, J 2001, Introduction to
textured mesh models. It is necessary to understand modern photogrammetry, John Willey & Sons, New
that, usually, this is not an end in itself, but rather a York, USA.
starting point for further modeling. The discussion Mitchell, W and McCullough, M 1995, Digital Design Media
with the students at the end of the semester point- (2nd Edition), Van Nostrand Reinhold, New York, USA.
ed out they felt to have acquired a very important Lillessand, T, Kiefer, R and Chipman, J 2004, Remote Sensing
tools for the future, but they also noted that it would and image interpretation (5th edition), Wiley, New Jer-
be interesting to give a further step in the applica- sey, USA.
tion of the produced models. Pupo, R, Duarte, J and Celani, G 2010 ‘Introducing digital
As a consequence, in future editions of this fabrication into the architectural curriculum - Two simi-
course, broader concerns will be considered, name- lar experiences in different contexts’, Proceedings of the
ly, practical uses of these models and how to use eCAADe conference, Antwerp, Belgium, pp. 517-524.
them as positional, geometric and radiometric con- Remondino, F 2006, Image-based modelling for object and
straints for the production of other kinds of 3D mod- human reconstruction, PhD thesis, Swiss Federal Insti-
els such as Nurbs models or CAD/BIM models. tute of Technology (ETH), Zurich.
Wu, C and Di, Y 2009 ‘Approach to standardization system
REFERENCES of Architectural Heritage Documentation in China’, Pro-
Boehler, W and Heinz, G 1999 ‘Documentation, surveying, ceedings of the XXII CIPA Symposium, Kyoto, Japan.
photogrammetry’, Proceedings of the XVII CIPA Sympo-
sium, Olinda, Brazil. [1] http://157.88.193.21/~lfa-davap/
Casaca, J, Matos, J and Baio, M 2000, Topografia Geral, Lidel, [2] http://www.unife.it/centri/diaprem/Settori%20di%20
Lisbon. ricerca/rilievo-3d
Chader, M 2008 ‘The value of third generation, Parametric [3] http://www.my3dscanner.com/
Modeling from 3D Scan Data’, Proceedings of the SME [4] http://photosynth.net
conference Rapid 2008, Florida, USA. [5] http://www.visual-experiments.com/demos/photosyn-
Duarte, J 2007 ‘Inserting new technologies in undergradu- thtoolkit/
ate architectural curricula - a case Study’, Proceedings [6] http://www.cs.washington.edu/homes/ccwu/vsfm/
of the eCAADe conference, Frankfurt, Germany, pp.423- [7] http://meshlab.sourceforge.net/
430. [8] http://www.danielgm.net/cc/
Duarte, J, Celani, G and Pupo, R 2010 ‘Inserting computa- [9] http://notepad-plus-plus.org/
tional technologies in architectural curricula’, Proceed- [10] http://www.libreoffice.org/
ings of the eCAADe conference, Zurich, Switzerland. [11] http://javagraticule3d.sourceforge.net/

Performing the Past and the Present for the Knowledge
of the Future
Anetta Kepczynska-Walczak
Lodz University of Technology, Institute of Architecture and Urban Planning, Poland
http://www.p.lodz.pl
anetta.kepczynska-walczak@p.lodz.pl
Abstract. The aim of this paper is to discuss the role three-dimensional models play
in addressing performance issues in virtual reconstructions of the heritage buildings.
Heritage visualisation is considered here as a process of representing knowledge about
space, time, behaviour, light, and other elements that constitute cultural environments.
The author aims to analyse the process of digital reconstruction of heritage buildings
and the impact of the decisions taken during its development on the final performance.
Based on the examples drawn from practice, various stages of development are discussed,
confronted with the principles of London Charter.
Keywords. Virtual reconstructions; cultural heritage; 3D modelling; London Charter.
BACKGROUND
Information technologies support a number of do- (Kepczynska-Walczak, 2003). However, the use of 3D
mains, including - among the others - virtual model- modelling in the virtual reconstructing of heritage
ling of built heritage. One of the earliest examples buildings is no longer a subject of research itself.
of such projects was a reconstruction of ancient It rather opens new fields of research and applica-
buildings in Bath, which was done as early as 1983 tion. First, it is necessary to indicate approaches to
(Dave, 2005). Another example might be Winches- considering a heritage building reconstruction as
ter cathedral which was modelled in 1984-1986. A a data container. For example Boeykens and Neu-
decade later the Urban Simulation Team from the ckermans (2009) studied the possibility to improve
University College in Los Angeles was commisioned and increase information by adding supplementary
a real-time visual simulation model of the Forum of metadata to the 3D model through “metadata en-
Trajan, the largest of the Imperial Fora in the Forum richment”. According to the authors “this structured
Romanum for the exhibition at the Getty Center. The information can, in turn, facilitate the retrieval and
project aimed at exploring the historical, cultural, recovery of such models, when searching or brows-
and technological information contained within an- ing for design information through online architec-
cient works of art as well as examining new ideas in tural repositories”. Another interesting project was
archaeology, conservation, scholarship, education, the use of BIM deployed in the historical reconstruc-
and digital technology (Jepson and Friedman, 1998). tion of the Vinohrady synagogue in Prague (Bo-
For many years, digital reconstructions have eykens et al., 2012). It is worth mentioning here the
been presented and discussed at the eCAADe con- book devoted to the former Viennese synagogues
ferences. The author also contributed to this subject that were destroyed and disappeared from the city

space. Rebuilt virtually, accompanied with the his- In such case it is only a supposed, hypothetical im-
torical photographs recreate and perform the past age of a building based on the archival documents.
in the context of the present day city (Martens and However, the resources are often either incomplete
Peter, 2011). or represent only architectural drawings, while
Other contributions of particular interest in- implementation records are usually not available.
clude ornament modelling and deployment of rapid This means that it is often impossible to confront
prototyping technology in making physical models the above-mentioned archival documents with an
(Breen and Stellingwerff, 2008) and augmented re- executed object due to the lack of photographic
ality (AR) allowing better understanding of original images or other reasons. The problem of trustwor-
appearance of a heritage interior and instant com- thiness emerged already with the first pioneering re-
parison with an extant state (Tonn et al., 2009). construction drawings made in the 18th century by
Concurrently, at the beginning of 21st century Giambatista Piranesi, who filled them with a number
researchers started to express their interest in the of imaginary elements. The foundations of scientific
computation and the performance in architecture approach to the subject were laid in the mid 19th
(Kolarevic, 2003). The subject is complex due to the century by Austin Henry Layard and Luigi Canina,
multiplicity of associated meanings, including sus- who paid a particular attention to the evidence and
tainable, technical, social and semantic issues. The veracity of performed reconstructions (Dave, 2005).
performative approach was also implemented in the The subject of reliability was recently discussed
case of built heritage objects (Albayrak and Tunçer, also in the context of 3D modelling of heritage
2011). Authors stressed the importance of “the shift buildings. A good example of problems emerging
in the orientation of architectural theory and prac- in this field might be a question of light analysed
tice from what the building is to what it does. There- by Hauck (2009) in the Hagia Sophia in Istanbul. In
fore, it defines the architectural object, not by how this case, the researcher dealt with the problems of
it appears, but rather by its capability of affecting, reflection of materials. Although a number of vari-
transforming and doing; in other words, by how it ous ready-made rendering software were deployed,
performs”. Their research suggested that this meth- the results remained unsatisfactory, especially when
od might be useful in the heritage conservation - in compared with the existing object. The solution was
this case in transforming fortifications of Amster- the use of an open source program, which allowed
dam, listed as the UNESCO World Heritage. to write appropriate scripts to expand capabilities of
the software. What is more, the modelled building
CREDIBILITY OF VIRTUAL RECONSTRUC- was visualised with the use of ‘sky models’, depend-
TIONS ing on the location, date and time. The ‘sky models’
In the light of the above, it is clear that contempo- were provided by the International Commission on
rary digital technology offers a vast arsenal of tech- Illumination - also known as the CIE after its French
niques of modelling, representation and analysis. title, the Commission Internationale de l’Eclairage -
Objects of any chosen time period can be recon- an organisation devoted to worldwide cooperation
structed and placed inside their original context. Es- and exchange of information on all matters relating
pecially in cases where the building is not existing, to the science and art of light and lighting, colour
is demolished or largely renovated or altered, the and vision, photobiology and image technology.
reconstructed model can be used to provide insight An issue of great importance was also raised
into the evolution of the building or the site. by Earl (2011) who dealt with the problem of insuf-
In this context the interpretation issues seem ficient data and, in consequence, tried to answer a
extremely important, especially when the recon- question how to visualise the hypothesis proposed
struction of not existing object is being considered. by researchers - in this case by archaeologists. In

other words, computer based visualisation tools hypotheses and differentiation degree of probability
have the capacity to create convincing reconstruc- of arguments.
tions of historical structures that appear to be au- What is interesting, similar assumptions were
thentic and complete. The challenge is how to make made some 80 years ago, when the Athens Char-
the process of reasoning drawn from relatively lim- ter for the Restoration of Historic Monuments was
ited evidence, more self-evident in the model and adopted in 1931 during the First International Con-
also make known the alternative options that were gress of Architects and Technicians of Historic Mon-
possible but less probable. uments. For example, it was stressed that in case of a
Therefore, virtual reconstructions of heritage heritage building reconstruction new materials used
buildings might be considered as a process of rep- for this purpose should in all cases be recognisable.
resenting knowledge about space, time, behaviour, Such an approach to the heritage reconstruction
light, and other elements that constitute cultural was developed in the Venice Charter in 1964: “the
environments. What is more, data credibility is of process of restoration is a highly specialized opera-
particular importance in the development of the so- tion. Its aim is to preserve and reveal the aesthetic
ciety of knowledge. and historic value of the monument and is based
on respect for original material and authentic docu-
LONDON CHARTER ments. It must stop at the point where conjecture
Taking into account the current state of research begins, and in this case moreover any extra work
already presented and the issues of the virtual re- which is indispensable must be distinct from the ar-
construction reliability, it is of crucial importance to chitectural composition and must bear a contempo-
present the London Charter for the Computer-based rary stamp” [2].
Visualisation of Cultural Heritage [1], which was con- Therefore, the London Charter is of great value
ceived in 2006 to ensure the methodological strict- and importance, since the availability of power-
ness of visualisation as a means of researching and ful hardware and software allows to perform de-
communicating cultural heritage. The Charter was lusively realistic reconstructions. What is more, at
officially approved by several national and interna- present nearly everything can be straightforwardly
tional bodies, including the Italian Ministry of Cul- published on-line and, in consequence, easily avail-
ture, which adopted it as an official guideline. able to unlimited number of the Internet users. This
In the Chapter preamble it is stated that the doc- might be regarded as a great advantage but, on
ument “aims to enhance the rigour with which com- the other hand, there is a risk that laymen lacking
puter-based visualisation methods and outcomes analytical capacity may consider those visualisa-
are used and evaluated in heritage contexts, thereby tions as representing the truth - according to no-
promoting understanding and recognition of such tion “seeing is believing”. Such situation in case of
methods and outcomes”. What is more, authors in- virtual modelling may cause erroneous interpreta-
dicate a number of earlier documents and initiatives tions of a history. This issue was further developed
(including AHDS Guides to Good Practice for CAD by some authors, who indicated that although for
and Virtual Reality, Virtual Archaeology Special Inter- certain purposes visualisations can exceed text in
est Group and Cultural Virtual Reality Organisation), an expressive power, their explanatory value may
which stressed necessity for scholarly reliability of be poor. Therefore Denard (2012) stressed that “for
virtual visualisation methods, as well as the care a heritage visualisation to match the rigour of con-
for the choice of an appropriate form of presenta- ventional research, its rigour must be visible. That is
tion of research results, reflecting the current state why, at the heart of The London Charter is the prin-
of historical knowledge. The central issue is clear ciple that heritage visualisations should accurately
distinction between facts confirmed by sources and convey to users the status of the knowledge that

Figure 1
Virtual reconstructions of
the Scheiblers chapel at the
Lutheran cemetery (left), the
Richters villa (top right) and
the Mutual Credit Society (bot-
tom right).
they represent, such as distinctions between evi-

dence and hypothesis, and between different levels buildings in Lodz, including the Richters villa, the
of probability.” Scheiblers funeral chapel and the Mutual Credit So-
ciety premises (Figure1). All the buildings represent
CASE STUDIES ANALYSIS various architectural types and forms from the late
The purpose of this section is an analysis of a digi- 19th century. The Mutual Credit Society, built in the
tal imaging process of heritage buildings and the 1870s, is an excellent example of neo-renaissance
impact of decisions on the final output. The analysis public edifice. The Scheiblers chapel is an impressive
is based on examples taken from the author’s di- mausoleum of one of the most prominent textile
dactic experiences in virtual reconstruction. Parallel manufacturers in Europe and his family. This build-
references to the London Charter (LC) allow better ing erected in 1888 is one of the best exemplars of
understanding of its principles and practical appli- 19th-century European gothic revival. While the
cation. Richters villa illustrates the living conditions in the
The issues related to credibility of virtual re- industrialist residence at the turn of the 19th and
constructions are based on the cases of historic 20th centuries.

Figure 2
The rose window of the front
façade of the Scheiblers
chapel.
It is necessary to stress that all the above-men- uments Fund as one of 100 most endangered sites
tioned objects exist, so the process of digital recon- in the world since 2006.
struction required a high quality realistic represen- The inventorial measured drawings and pho-
tation in accordance with Rule 6 of LC: “the creation tographic documentation were used as the initial
and dissemination of computer-based visualisation material for digital reconstruction. The inventory
should be planned in such a way as to ensure that was made using a hybrid method that combines a
maximum possible benefits are achieved for the traditional analogue and digital techniques of doc-
study, understanding, interpretation, preservation umenting heritage buildings. The range of meas-
and management of cultural heritage (…) The aims, urement drawings included not only the shells of
methods and dissemination plans of computer- buildings, but also their interiors. The high level of
based visualisation should reflect consideration of accuracy was obtained, which can be seen on some
how such work can enhance access to cultural herit- of details drawings. Therefore, such comprehensive
age that is otherwise inaccessible due to health and data enabled to create very detailed digital models
safety, disability, economic, political, or environmen- (Figure 2 and 3). Unfortunately, it was impossible
tal reasons, or because the object of the visualisa- to use 3D scanning due to high costs. Despite the
tion is lost, endangered, dispersed, or has been de- growing knowledge on this technology among the
stroyed, restored or reconstructed.” conservators, the financial barrier makes 3D scan-
Among the principal goals of analysed cases ning in Poland not widely used in heritage docu-
was an education, including the dissemination of mentation practice.
Lodz cultural heritage, allowing access to these Pursuant to Rule 4 of LC the goal was clearly
magnificent buildings which are not open to public defined - to reflect the existing state: “4.4. It should
due to their current state and use. What is more, the be made clear to users what a computer-based visu-
Scheiblers chapel has been listed by the World Mon- alisation seeks to represent, for example the existing

Figure 3
The lantern at the main en-
trance of the Richters villa.
state, an evidence-based restoration or an hypo- including, in particular, the problem of texture map-
thetical reconstruction of a cultural heritage object ping and performance of the same texture in differ-
or site, and the extent and nature of any factual un- ent lighting conditions (Figure 5). Texturing turned
certainty”. out to be a very difficult task, many attempts have
In the context of the above the question arises been done to achieve an effect similar to reality. It
whether - referring to the London Charter principles was impossible to use textures from photographic
- a model and subsequent visualisation, made on pictures since in different lighting conditions the
the basis of the inventory, are sufficiently reliable for same material performed different appearance.
“study, understanding, interpretation, preservation Another interesting observation was a selection
and management of cultural heritage”? On the oth- of lighting - mimicking the actual lighting condi-
er hand, however, one of the principles of LC is that tions in a virtual environment, the virtual textures
“the costs of implementing such a strategy should changed their characteristics unlike to what could
be considered in relation to the added intellectual, be observed in reality. What is more, a colour palette
explanatory and/or economic value of producing of the interior successfully reproduced in one visu-
outputs that demonstrate a high level of intellectual alisation, turned up different from the actual interior
integrity”. appearance in another visualisation.
Discussed reconstructions present high level of It is worth to confront the observations with
details - not only exteriors but also interiors were one of the objectives of the London Charter, which
modelled carefully (Figure 4). Special regard was “seeks to establish principles for the use of comput-
paid to the issues of lighting and texturing objects, er-based visualisation methods and outcomes in the

Figure 4
The three-dimensional cross-
section showing variety of
volumes and spaces within
the Mutual Credit Society
building.
research and communication of cultural heritage in The main obstacle of such tasks lies in limited source
order to (…) ensure that computer-based visualisa- materials. On the other hand, it is relatively easy to
tion processes and outcomes can be properly un- accept the achieved results, since it is impossible to
derstood and evaluated by users”. compare them with the actual building. On the con-
To sum up this section, it is necessary to stress trary, when the existing object is a subject of model-
that the ability to confront the results achieved in ling, it is perfectly possible to achieve its geometry
the process of creating the virtual model with the through the measuring or scanning. However, there
actual state allowed the ongoing verification of the is much stronger pressure on reliable representation
decisions and to introduce necessary adjustments. of real appearance. It is not easy if not just a general
It might be argued that the situation was comfort- impression but the knowledge about the object is
able since modelled objects existed. Nonetheless, it to be represented. What is more, the problems as-
was impossible to avoid the compromises because, sociated with modelling of existing structures make
as the experience has shown, a reliable digital repre- clear that reconstructions of non-existent objects
sentation depends not only on the input data. may occur extremely imperfect.
Similar problems apply to other fields of art -
SUMMARY AND CONCLUDING REMARKS such as sculpture. For example, replicas made in a
The considerations put forward in the first part of different material, although keep shapes of originals,
this paper relate to the reconstruction of non-ex- trigger different aesthetic experience. The topicality
istent, destroyed objects and to existing structures. of the above-mentioned issues can be proved by

Figure 5
The entrance hall and the
main stairs of the Mutual
Credit Society edifice.
the solution adopted in the Tate Gallery on-line cata- Taking above issues into account, the author be-
logue, in which objects could be seen in a different lieves the paper will contribute to the discussion on
light exposure, allowing their better understanding, performative values of virtual reconstructions in the
including texture and other features (Stanicka-Brze- cultural heritage domain.
zicka, 2012).
To summarise, the author aimed to analyse the REFERENCES
process of digital reconstruction of heritage build- Albayrak, C, Tunçer, B 2011, ‘Performative architecture as a
ings and the impact of the decisions taken during its guideline for transformation: Defense Line of Amster-
development on the final performance. dam’ in T Strojan Zupancic, M Juvancic, S Verovsek and
Assuming that the imaging is treated as a visu- A Jutraz (eds.) Respecting Fragile Places: Proceedings
alisation of knowledge, these issues are of particular of the 29th Conference on Education and Research
importance, since contemporary culture is based on in Computer Aided Architectural Design in Europe,
the visual perception, in which not intellect, but the eCAADe / University of Ljubljana, Ljubljana, pp.501-
senses are activated to experience the past [figure 510.
6.]. What is more, the image acts as the dominant Alkhoven, P 1992, ‘The Reconstruction of the Past: The
form of memory. According to Szpocinski (2009) Application of New Techniques for Visualization and
memory visualisation is a phenomenon which es- Research in Architectural History’, in G N Schmitt (ed.)
sence is the dominance of visual events in the pro- Computer Aided Architectural Design Futures: Educa-
cesses of transmission and perception of the past. tion, Research, Applications, Friedrich Vieweg & Sohn

Figure 6
The pulpit and the presbytery
of the Scheiblers chapel.
Verlagsgesellschaft mbH, Braunschweig/Wiesbaden, (eds.), Paradata and Transparency in Virtual Heritage,

pp. 507-522. Ashgate, Farnham, pp. 57-71.
Boeykens, S, Himpe, C, Martens, B 2012, ‘A Case Study of Earl, M 2011, ‘Visualizing the Unknown in Historical Vernac-
Using BIM in Historical Reconstruction: The Vinohrady ular Architecture’ in T Strojan Zupancic, M Juvancic, S
synagogue in Prague’ in H Achten, J Pavlicek, J Hulin, Verovsek and A Jutraz (eds.) Respecting Fragile Places:
D Matejovska (eds.), Digital Physicality - Proceedings of Proceedings of the 29th Conference on Education and
the 30th eCAADe Conference - Volume 1, Czech Tech- Research in Computer Aided Architectural Design in
nical University in Prague, Prague, pp. 729-737. Europe, eCAADe / University of Ljubljana, Ljubljana,
Boeykens, S, Neuckermans, H 2009, ‘Architectural design pp. 868-874.
analysis, historical reconstruction and structured ar- Hauck, O 2009, ‘Computing the „Holy Wisdom” – FormZ and
chival using 3D models. Techniques, methodology and Radiance as analytic tools for historic building research’
long term preservation of digital models’ in T Tidafi, T in C I Yessios (ed.) FormZ Joint Study Program Report
Dorta (eds.) Joining Languages, Cultures and Visions: 2007/2008, Columbus, Ohio.
CAAD Futures, PUM, Montreal, pp. 119-132. Jepson, W, Friedman, S 1998, A real-time visualization sys-
Breen, J, Stellingwerff, M 2008, ‘Capital A to Z’ in M Muylle tem for large scale urban environments, UCLA Urban
(ed.) Architecture ‘in computro’. Integrating methods Simulation Team.
and techniques: Proceedings of the 26th Conference Kepczynska-Walczak, A, Walczak, BM 2003, ‘The Integration
on Education and Research in Computer Aided Archi- of IT within the Early Stage of Architectural Conserva-
tectural Design in Europe, eCAADe / Artesis University tion Design’ in W Dokonal, U Hirschberg (eds), Digital
College of Antwerp, Antwerp, pp. 759-766. Design: Proceedings of the 21st Conference on Edu-
Dave, B 2005, ‘Labyrinthine Digital Histories’ in B Martens, cation and Research in Computer Aided Architectural
A Brown (eds.), Computer Aided Architectural Design Design in Europe, eCAADe / Graz University of Technol-
Futures 2005, Springer, Dordrecht, pp. 53-62. ogy, Graz, pp. 493-499.
Denard, H 2012, ‘A New Introduction to The London Char- Kolarevic, B 2003, ‘Computing the Performative in Architec-
ter’ in A Bentkowska-Kafel, H Denard and D Baker ture’ in W Dokonal, U Hirschberg (eds), Digital Design:

Proceedings of the 21st Conference on Education and Szpocinski, A 2009, ‘Wizualizacja pamieci spolecznej’ in A
Research in Computer Aided Architectural Design in Szpocinski (ed.) Pamiec zbiorowa jako czynnik inte-
Europe, eCAADe / Graz University of Technology, Graz, gracji i zrodlo konfliktow, vol. 4, Instytut Studiow Poli-
pp. 457-464. tycznych PAN, Wydawnictwo Naukowe Scholar, Colle-
Martens, B, Peter, H 2009, ‘Potemkin Village, reloaded Visu- gium Civitas, Warsaw, pp. 227-236.
alization of Destroyed Synagogues within the City- Tonn, C, Petzold, F and Donath, D 2009, ‘See-through Histo-
scape of Vienna’, Projecting Spaces: Proceedings of ry. 3D Augmented Reality for the Reconstruction of the
the 9th European Architectural Endoscopy Association Bauhaus Director’s Office Reception Area from 1923’
Conference, Brandenburg University of Technology, in W A Abdelhameed, N Hamza and A Bennadji (eds.)
Cottbus, pp. 52-57. Digitizing Architecture: Formalization and Content: 4th
Martens, B, Peter, H 2011, The Destroyed Synagogues of Vi- International Conference Proceedings of the Arab Soci-
enna - Virtual city walks, LIT-Verlag, Wien. ety for Computer Aided Architectural Design, ASCAAD,
Stanicka-Brzezicka, K 2012, paper presented at the confer- Manama, pp. 259-267
ence Cyfrowe Spotkania z Zabytkami (Digital Encoun-
ters with Cultural Heritage) in Wroclaw. [1] http://www.londoncharter.org/
[2] http://www.icomos.org/

Building Information Modelling
Building Information Modelling - Volume 2 - Computation and Performance - eCAADe 31 | 463

464 | eCAADe 31 - Computation and Performance - Volume 2 - Building Information Modelling
Challenges of Integrating BIM in Architectural Education
Tuba Kocaturk1, Arto Kiviniemi2
University of Liverpool, School of Architecture, United Kingdom
1
T.Kocaturk@liverpool.ac.uk, 2A.Kiviniemi@liverpool.ac.uk
Abstract. This paper provides a critical overview of some of the fundamental issues
regarding the adoption and integration of BIM – both as a method and as a technology –
in Architectural education. It aims to establish a common ground for the rationale behind
such integration and reflects on the past and present state of the cultural, intellectual,
professional and technological context of Architecture. The paper will introduce the core
issues to be considered in order to succeed in this challenging and transformational
process. It will also introduce a framework for a gradual and progressive adoption of BIM
and integrated design in the architectural curriculum.
Keywords. Architectural education; BIM and integrated design; distributed cognition;
integrated design studio.
INTRODUCTION
The emerging visions for an “Integrated Practice” in as a stepping stone in order to be more efficient and
building industry, through BIM (Building Informa- effective. So how do these ambitions affect archi-
tion Modelling), carry potential to fundamentally tects and architectural education at large? The RIBA
transform the way in which architectural education believes that architects have a central role to play in
engages with issues of design knowledge, technol- ensuring that the construction industry responds
ogy, representations and collaboration (Ambrose et to the opportunities offered by BIM in both public
al., 2008). In this article we aim to develop a frame- and private sectors and has developed a new Plan of
work for the integration of BIM into architectural Work (launched in May 2013) as an important piece
education. We also aim to identify the core issues to of new guidance for architects and co-professionals
be considered in order to succeed in this challeng- [1]. However, there is yet no guidance or a roadmap
ing transformational process. for architectural schools/institutions as to how they
In UK, the government has set out an ambitious could adapt to the forthcoming challenges in the in-
plan to have fully collaborative BIM, with all project dustry and to educate the future architects accord-
and asset information, documentation and data be- ingly.
ing electronic, on all public sector projects by 2016. There are both complementary and contradic-
The UK programme based on this new BIM strategy tory views as to “if” and “how” BIM – either as a soft-
is seen as one of the most ambitious and advanced ware, or as a process or in any combination – should
government led programs to embed the use of BIM be integrated into the academia’s curriculum struc-
across all centrally procured public construction pro- ture. Some of the resistance stem from a shared set
jects. Through this Government-led incentive, the of concerns which have been outlined by some of
construction industry is getting ready to utilize BIM the contributors of a recently edited book by Deam-

er and Bernstein (2011); 1) architectural curricula is educational approaches will be followed with the
already overloaded and there is no room for any more introduction of a new framework for BIM integra-
content, 2) The inherent practice-driven approach of tion into the architectural education, a discussion
BIM methodology is not compatible with the explora- regarding some of the pedagogical and cognitive is-
tive character of design thinking, 3) the structure of the sues, as well as the future of the profession.
architectural curriculum is not suitable to adopt BIM.
Other factors have also been reported as impeding DEFINING A COMMON GROUND AND
the successful adoption of BIM in the design curricu- RATIONALE
lum, such as; varying definitions and interpretations BIM is widely used as the acronym for ‘Building In-
of BIM by different professions; issues regarding formation Modelling’ which is commonly defined
accreditation, and disproportionate emphasis on using the Construction Project Information Commit-
“technical skills” (Kiviniemi, 2013). tee definition as: ‘...digital representation of physical
There are two major areas where BIM will have and functional characteristics of a facility creating a
direct impact on the architectural curriculum. First shared knowledge resource for information about it
is its implicit proposition as to how design and pro- forming a reliable basis for decisions during its life
ject partners should collaborate, and the second is cycle, from earliest conception to demolition.’ [2].
regarding how information (geometric and non- Before posing the question of “how to”, it’s es-
geometric) can be modelled, embedded and shared sential to revisit an important question: What is
during the entire project life cycle. However, using the rationale behind using BIM in the first place?
a BIM software doesn’t automatically guarantee a Obviously, one would easily argue that the ways in
superior level of collaboration, unless conditions for which architectural education will embed BIM into
a successful collaboration are met which is not only its curriculum - with supporting pedagogies - must
through software. Similarly, the ability to virtually be compatible to the rationale behind it and some-
model both geometric and non-geometric project how guide this transformational process. The an-
information doesn’t immediately bring maximum swer to this question is explained in a most recent
efficiency unless the representations are modelled government documentation as 1) promote greater
and shared properly, the information needs in the transparency and collaboration between suppliers
process are correctly understood, and a robust tech- and thereby reduce waste (procurement, process
nical infrastructure and a proper business model to and material) through all levels of the supply chain
support this process are present. Therefore a view of 2) enable intelligent decisions about construction
BIM solely as a software would be a rather superfi- methodology, safer working arrangements, greater
cial and unsustainable approach. The focus should energy efficiency leading to carbon reductions and
instead be on the principles that the concept of “in- a critical focus on the whole life performance of fa-
tegrated design and project delivery” was founded cilities [3]. In other words, without any specific men-
on in the first place, so the changes in the curricu- tion to any specific technology or software, the mes-
lum wouldn’t become obsolete each time a new BIM sage is: how we used to collaborate, make decisions,
technology is developed and introduced. exchange information, use and organize our time and
The paper will initially identify the rationale be- resources in the past in design and construction sector
hind BIM integration into the Architectural curricula have been full of inefficiencies causing a lot of waste of
(both as a concept and as a technology) and some time, money, and resources. And this needs to change.
of the common misunderstandings which impede This is a statement which each and every person in
its successful adoption. It will then try to explicate our sector, including architectural educators and
some of the fundamental reasons for the resistance even the biggest BIM sceptics would probably agree
against BIM. A critical review of some of the existing with. BIM, in this present time, is “a proposition” as “a

possible solution” to tackle some of the major prob- capacity (Allen, 2012), and thereby created a sub-
lems of our industry. And just like every proposition, stantial divide between architectural theory and
it comes with its own methodology, supported by architectural practice. 1990s marked a big shift from
its own technical infrastructure, implemented as a cultural theory to building practice in the academic
“technological solution” by various software provid- circles. Theory dominance began to subside as new
ers, albeit with variations in their focus and in the architecture practices emerged which better suited
support they provide, and with certain bottlenecks. to meet the challenges issued by globalisation. To-
One of the biggest misunderstandings about day, globalisation, digital technology, environmental
BIM and the changes associated with its integration change, and increasingly market-driven education
into our work practices is to compare it to the shift economy are already reshaping academia (Ockman
from drafting (on tracing paper) to CAD. Firstly, the and Williamson, 2012). Consequently, there have
shift from tracing paper (as 2D drawings) to CAD (as been changes in the curriculum to address a broad-
2D computer files) did not change the outputs that er understanding of social, economic, technological
were issued to the industry (Kiviniemi and Fischer and cultural variables in order to design buildings
2009). Consequently, it didn’t have a major impact which perform to higher environmental and energy
on the structure and the hierarchies in the sector, standards. However, still a majority of architectural
nor required new working methods. However, al- schools are following the traditional educational
ready the shift from level 0 BIM to level 2 BIM (do- models with less engagement with technology and
main specific, federated models) does indeed pose with the larger community of the built-environment
fundamental changes such as; Handling and creating that make up the building industry. Having made
information rich models, new ways of working with this distinction, it’s also important to note that con-
other stakeholders, re-aligning the disciplinary roles cerns over BIM integration into the architectural cur-
and responsibilities, opportunities for new additional ricula are not only limited to the more conventional
roles for Architects. Although even the definition of schools of architecture. A certain level of scepticism
level 3 BIM (fully integrated models) is at the mo- and hesitation exists even among the most technol-
ment somewhat vague, it will introduce new chal- ogy savvy schools, although they have since long
lenges in the future. These changes become great embraced new design methods and technologies in
challenges that will relate to training and education their curriculum and research programmes.
of Architects that cannot be solved simply by add- Although there are a few innovative and suc-
ing new content and skills to the existing curricu- cessful implementations of BIM in current practice,
lum, but will also necessitate the modification and architectural education has been slow to respond.
deletion of some of the existing content. There are two issues to understand about the ex-
istent resistance. One is associated with some of
THE STATUS ANXIETY AND THE ROOTS the established values embedded deeply within its
OF RESISTANCE professional culture. The other is due to the nature
The discussions surrounding the “BIM integration of architectural education as an institution, and the
into Architectural education” should not be under- cultural and intellectual capital it entails. Profession-
stood in isolation, but in connection with the general education in Architecture doesn’t only provide the
al changes of the relationship between architectural necessary epistemological and cultural context for
schools and professional practice since the begin- architects, but also helps define the social and pro-
ning of the 1990s. For the most part of the 20th cen- fessional context within which architects work and
tury, Architectural education has been looking to operate within the construction industry. Although
cultural studies and literary criticism for its theoreti- there are distinct variations in the positions different
cal models, minimizing its operative and technical schools of Architecture take, there have always been

an unspoken but almost anonymously accepted cul- A GENERAL REVIEW OF CURRENT EDU-
tural norms and codes as to how architects associate CATIONAL EFFORTS OF BIM INTEGRA-
themselves with the rest of the construction indus- TION
try and within the society at large (Stevens, 1998). A recent study suggest that universities are lagging
BIM is not just a new technology. If it was just behind the AEC industry in terms of adopting BIM
another CAD, or another piece of technology, ar- technologies and improved collaborative working
chitects would have already been the first to adopt practices, and that universities are not currently
and advocate, as we have already been witnessing meeting the needs of industry in terms of collabora-
through the highly creative and innovative use of tive building design and BIM education (Macdonald
the recent parametric and computational “digital and Mills, 2011). A majority of developments in UK
design software” both in practice and in various relate to the emergence of new master courses in
Schools of Architecture. BIM has an implicit proposi- BIM and Integrated Design as well as CPD courses
tion as to how the sector should/could be realigned, addressing to different disciplinary groups. This
restructured and work together. In other words, it makes sense as they are fast track options and pro-
will have wider social and professional implications vide concentrated content. They usually address to
within the sector, which makes its potential future a multi-disciplinary audience and therefore entail
users more hesitant. And this is probably at the core rather generic content. They do not always address
of one of the least pronounced reasons for resist- the individual disciplinary challenges and often aim
ance against BIM, especially by the educators in ar- to provide “an introduction” to the subject. CPDs
chitecture, that is; by entering into an unchartered and software-vendor led trainings usually provide
(BIM) territory architects can become a mere player, more discipline specific teaching, and can range
one of “the others”, instead of “the creator, the in- from more theoretical to more technical. However,
novator”. BIM is not a fixed or a finished concept or when we look at UG level education, the situation
methodology, and technology is continually being is quite different. There are only very few adventur-
updated and developed to meet the industry needs, ous institutions where BIM is already a part of their
giving way to the emergence of new concepts and curricula, albeit quite disintegrated from the rest of
insights on a continuous basis. Another justified the more conventional content and methods of de-
reason for the existing anxiety is about how to in- livery. BIM integrated design studios have also be-
tegrate something that is not yet theoretically nor come a rather experimental and safe option in the
practically complete into an “educational system” introduction of BIM into the curricula, which are im-
which is historically based on established theoretical plemented in both UG and PG level studios. In most
and discursive models? There are various concerns cases, architecture students work in “collaborative
that naturally follow this discussion, such as: design teams” pretending to be another disciplinary
• How will the changes affect our accreditation member of the design team. In few cases, students
status? from different disciplinary backgrounds are brought
• How will the existing staff adapt to the new together to collaborate on design/engineering chal-
skills and knowledge required by this fast mov- lenges. In both cases, the studio is quite isolated
ing and industry-led approach? from the rest of the curriculum and the chosen ap-
• Can creative artistic expression co-exist with proach doesn’t follow any particular institutional
collaborative practice? and pedagogical agenda. Determining the success
• Can we retain and protect our professional criteria for these experiments are also quite chal-
values in a new-found collaborative and demo- lenging due to differences between the maturity
cratic pluralism? levels of students, their varying familiarity with the
software used, and the focus of the studio challenge.

A recent article examined 3 integrated studios that tials of BIM for our own profession and the design
variously explored designs (design collaboration, team, and influence the direction of BIM as such. We
formal possibilities, engineering integration into need to understand the viewpoint of the “others” in-
design) and how they adjusted to the protocols of stead of acting like them. Therefore we propose two
BIM, each providing interesting and contrasting ex- core modules that need to be delivered already at
amples (Pihlak et al., 2011). Some of the key findings the undergraduate education (Part 1 in UK Schools)
of this study can be summarized as follows: starting from year 1, progressing in complexity and
1. BIM teams that strived to minimize the conflict content as the student matures, and with an in-
produced the least innovative designs, creasing degree of integration with the design stu-
2. The collaboration across different disciplines dio; Modelling and Representation and Collaborative
seemed to be productive when designers were Working.
strong and confident and when the engineers
were flexible enough to go along with the non- Modelling and Representation
linear creative process, The intelligent modelling approach, advocated by
3. Too much compromise led to less than optimal recent digital design media and BIM technology are
design solutions, fundamentally changing the way architects used
4. The design went into the direction of the disci- to produce and communicate design information.
pline where there was more confidence, In architectural education, the reproduction of the
5. Design emphasis could easily get lost in an ex- “drawing” has a special focus. The fundamental
panded field where numbers, time and money change BIM introduces is the separation of the rep-
are so present. resentation and content; information in models can
be viewed using different representations for differ-
A FRAMEWORK FOR BIM INTEGRATION ent purposes and audiences. Thus, there is an urgent
INTO ARCHITECTURAL CURRICULUM need to shift the focus from “drawings” to creating
BIM is not just a new topic to be added to the exist- “intelligent” models of the design (including the
ing curriculum, as it currently is being implemented possibility to generate drawings from the model). As
by many schools, being introduced to the students Hugh Whitehead eloquently put it, design requires
either as “a new technology” in the studio, and/or as a “federation of models’ (Whitehead et al., 2011) at
a “new topic” in the professional practice modules, different levels of abstractions at different phases
mainly towards the end of the Bachelor level educa- of the design. Thus, a sketch is a model, but with a
tion with not much real connection with the rest of high degree of abstraction. A physical model, is yet
the curricula. If the concept of “BIM and integrated another.
design” is to be embedded into Architectural edu- Heavy emphasis on “drawing” has also brought
cation, this needs to be a gradual and progressive about “layered thinking” in terms of scales. So the
integration, instead of an “add and stir” approach. level and detail of thinking has almost become re-
It needs to be connected with the rest of the cur- stricted to the scale. Although scaled drawings can
riculum, and we must be able to make sense of this be produced from BIM models, certain information
new method and technology in a continuum, and has to be thought through quite early in the process
by identifying our frames of references in relation to and embedded into the model. A core module on
how things were in the past, how they are now and “modelling and representation” should convey the
how they are changing with new tools and working fundamental understanding of how various tools
methods. In order to collaborate, we have to be even and techniques help designers model different
more confident and competent in our ability as ar- types of information, on different level of detail in
chitects and understand the capabilities and poten- different stages, and the degree to which they in-

fluence and affect the design thinking and process. er community of built environment professionals
There are other motivations to develop representa- and introduces the various tools, issues and meth-
tions as well, for example, “how to sell your ideas”, ods that are geared toward varying goals and prac-
“how to effectively communicate the value in your tices of collaborative working. As the students get
design”, etc. There are different ways of embedding more mature, they should be exposed to diverse set
value and information into a design through dif- of tools, methods and techniques with which they
ferent levels of abstractions and different types of can experiment, compare and appreciate the differ-
representations according to “who” you are dealing ences between individual and collaborative working
with and “for what purpose”. However this variety in different design exercises.
has not been sufficiently present (or encouraged for A common tendency in BIM integration in ar-
that matter) in Architectural Education. A majority of chitectural education today is the introduction of
schools still require a pre-determined set of scaled “multidisciplinary” design studios, as discussed in
drawings for the final studio presentations which the previous sections. Although it is useful to mimic
are prepared usually for the eyes of other design- the actual design practice by bringing students from
ers, whereas in real practice, architects communi- different disciplines together (each drawing on their
cate and negotiate information across a much wider disciplinary knowledge) at certain point in their for-
community of professionals and clients, who seldom mal education, the timing of such an interaction is of
can fully understand traditional drawings. What we vital importance and could only be useful if the stu-
are proposing is a shift of emphasis from the final dents have already gained a certain degree of ma-
product representation towards the process of de- turity in their own specialization. In order to make
sign creation, development, coordination, commu- this point clear, we refer to Marilyn Stember’s paper
nication and negotiation through representations. (1991) where she offers the following overview of
In such a discussion, we implicitly define the role we different levels of disciplinarity (descriptions are
think the architect should play in the future. Does summarized to fit into our context).
the architect’s role end by preparing the right type • Intradisciplinary: working within the profes-
and format of information, or do we want to coordi- sional boundaries of a single discipline.
nate this multi-layer information web? • Cross-disciplinary: a design team (or an indi-
The student learning in this module should pro- vidual designer) from one discipline is viewing
gress from understanding the fundamentals, then and using concepts of another discipline from
developing intelligent models, and then selectively the perspective of their (his/her) own.
sharing and exchanging information in data-rich • Multidisciplinary: people from different disci-
models. Students should also be able to understand plines are present and working on the same
the underlying concepts of creative and operational project, however each one is operating on their
modelling and the degree of abstraction, clarity and own disciplinary knowledge
precision required in both. • Interdisciplinary: a design team (or an indi-
vidual designer) is integrating knowledge and
Collaborative Working: Tools, Issues and methods from different disciplines, using a syn-
Methods thesis of approaches.
The appreciation of the differences in professional, • Trans-disciplinary: a unified and commonly
disciplinary and cultural values during collaborative accepted framework (understanding) beyond
working has become an important concern with an any individual disciplinary perspectives of a
increase in global mobility and global practices. We community of practice
propose a core module in “Collaborative Working” Architectural education needs to address all of
that explicitly clarifies the role of architects in a larg- these levels of disciplinarity. Design is intrinsically an

interdisciplinary activity as the design process entails of the conditions of a successful “integrated design”
a continuous synthesis and negotiation of different is that “each discipline should become more skilled
knowledge from different disciplines. However, this at what they do and, most importantly, respect and
requires an awareness of our own disciplinary roles value the contribution of each other as a first step
and responsibilities (intradisciplinarity). Only with towards new working processes” (as cited in Pihlak
this awareness, we can utilize the knowledge of oth- et al., 2011).
er disciplines, appreciate the role they play in rela-
tion to ours (cross-disciplinarity) and be able to draw COGNITIVE IMPLICATIONS
on our own disciplinary knowledge in a confident These recent trends in “digital media and BIM inte-
way and can make an effective and creative con- gration” in collaborative design studios point out to
tribution in a multidisciplinary context. When these a common tendency across many schools of archi-
concepts are used as pedagogical approaches, they tecture; aiding the learner development through
are all valid and necessary at different stages of ar- both social and technological scaffolds. In this re-
chitectural education (and training). However each spect, we identify the emergence of a dominant
requires a different level of maturity (of students), ‘tool-aided’, ‘socially shared’, contextual and highly
and a brief (on an appropriate scale) to support the situated forms of cognition commonly referred to
intended outcome of the relevant studio or taught in literature by developmental psychologists and
module it’s applied in. Similarly, one should also pay learning theorists as “distributed cognition” (Hutch-
attention to the suitability of each method in dif- ins et al., 1986) and “distributed intelligence” (Pea,
ferent context. For example, interdisciplinarity and 1993). The central idea is that the resources that
cross-disciplinarity are more appropriate for creative shape and enable activity are distributed in con-
design, and multidisciplinarity is more appropriate figuration across people, environments, situations
for relatively more mature students. For example and artefacts (tools). One of the main pedagogical
pairing architectural students with engineering and dilemmas today can be grounded on the gap be-
construction management students in “integrated tween the distributed and the individual levels of
BIM studios” in the early years of UG design educa- intelligence that students are building through di-
tion would be “pedagogically incorrect” if the stu- verse methods of knowledge acquisition and meth-
dents are not yet at a maturity to recognize each ods of delivery without any explicit recipes of how
discipline’s own values, procedures and protocols. to build the link between the two. This separation
Students would still learn useful skills but this would has become even more distinct with the integration
most probably be at the expense of the quality of of technology and collaborative working methods
the design outcomes as design emphasis would in the design studio.
most probably be diminished. It would probably Salomon (1993) introduces two kinds of cogni-
make more sense to introduce “intradisciplinary” tive effects of technologies on intelligence:
and “interdisciplinary” studios during early and later • Effects with technology is obtained during in-
years of UG education, respectively, and “multidis- tellectual partnership with it, and
ciplinary” studios at the PG level, when the learners • Effects of technology in terms of the transfer-
have the necessary maturity. Trans-disciplinarity is able cognitive residue that this partnership
still an undefined territory in design and construc- leaves behind in the form of better mastery of
tion. Some of the concepts advocated by “integrated skills and strategies.
design” are currently being recognized as trans-dis- While effects with refers to the development of
ciplinary. However, this should not be seen as a cas- Distributed Cognition, effects of is attributed to the
ual blurring of disciplinary boundaries. On the con- development of individual cognition and solo intel-
trary, as emphasized eloquently by Hanif Kara, one ligence which are essential for the learner to deve-

lop an autonomous response as a residue to inter- distributed cognition.
action with the social and technological scaffolds. It is also important to note the changes in the
Today, the special emphasis on the use of a variety professional services provided and required from
of BIM software and skill building workshops offered the architects, on a global scale. We are experiencing
by many tool builders provide mainly technical scaf- the emergence of additional profiles, new speciali-
folds to the learner. This disproportionate emphasis zations and consultancy services high in demand in
placed on the “tools” present a risk of promoting building industry. In such expanded modes of prac-
design and collaboration as solely a tool-driven ac- tice, one size doesn’t fit all. Is it sustainable or even
tivity, especially for the novice learner, displacing possible to reproduce “architects” with exactly the
the innermost values of architecture, and as a con- same profile? New specializations (such as BIM Man-
sequence, weakening and changing the role of de- ager) should be introduced to students as possible
signer in the society (Kocaturk et al., 2012). In sum, (alternative) pathways already during their UG edu-
we propose the development of individual and dis- cation, which they could later choose to specialize
tributed competencies within the same pedagogical on during their PG studies. This might be influential
framework. in harnessing the most out of individual talent and
interest (e.g. design, technical skills, business).
SUMMARY AND DISCUSSION Will architectural education just follow BIM as a
The article gave a critical overview of some of the trend, solely as a beneficiary, or become one of the
fundamental issues regarding the adoption and in- driving forces in this industry transformation? Do
tegration of BIM – both as a method and as a tech- architects really have the chance to renegotiate their
nology – in Architectural education. It also tried to status and maybe even regain their master-builder
explicate some of the fundamental reasons for the status with BIM – as implied by many – or does BIM
current resistance while reflecting on the past and further emphasize and legitimise the hybridity of
present state of the cultural, intellectual, profes- architectural profession? The answer to these ques-
sional and technological context of Architecture. A tions, for each and every architectural institute in
critical review of some of the existing educational particular, will be the main guide in setting up a
approaches to BIM integration, such as BIM inte- plan for their interpretation and integration of BIM
grated studios, revealed the fact that the current in their educational agenda with relevant and nec-
implementations are quite opportunistic, disinte- essary technical infrastructure and pedagogical ap-
grated from the rest of the curriculum and lack any proaches. And there is a good indication that there
clear strategic and/or pedagogical agenda. Two will be parallel and contrasting approaches across
major areas have been identified where BIM will institutions, which ultimately will determine a new
have direct impact on the architectural curriculum; plural agenda for the profession. Therefore the un-
1) modelling and representation, and 2) collabora- derlying challenge is about renegotiating architects’
tive working. These areas have been proposed to be multiple identities and redefining the problematic
added to the architectural curricula as the two new relationship that has long existed between academ-
core modules, starting from year 1, and progressing ia and practice; techniques and aesthetics; science
in complexity and content as the students mature, and humanities.
with an increasing degree of integration with the
design studio. Some of the critical pedagogical and
cognitive issues have been identified according to REFERENCES
the extent to which the new technology and work- Allen, S 2012, ‘The Future is Now’ in J Ockman and R Wil-
ing methods will have an impact on the process of liamson (eds), Architecture School: Three Centuries of
learning and development of both individual and Educating Architects in North America, MIT Press, Cam-

bridge, pp. 204-229. ica, MIT Press; Association of Collegiate Schools of Ar-
Ambrose, MA, Lostritto, C and Wilson, L 2008, ‘Animate Edu- chitecture, Cambridge, Mass.Washington, D.C.
cation: Early Design Education Pedagogy’ in CAADRIA Pea, RD 1993, ‘Practices of Distributed Intelligence and De-
2008 Computer Aided Architectural Design Research in signs for Education’, in G Salomon (eds), Distributed
Asia, Thailand, pp. 29-35. Cognitions, Psychological and Educational Considera-
Deamer, P and Bernstein, P (eds) 2011, BIM in Academia, tions, Cambridge University Press, UK.
Yale School of Architecture, CT. Pihlak, M, Deamer, P, Holland, R, Poerscheke, U, Messner, J
Hutchins, EL, Hollan, JD, and Norman, DA 1986, ‘Direct Ma- and Parfitt, K 2011, ‘Building Information Modeling
nipulation Interfaces’, in DA Norman and SW Draper (BIM) and the Impact on Design Quality’, Architec-
(eds), User Centred System Design: New Perspectives on tural Engineering Technology, 1(1), URL: http/;//dx.doi.
Human-Computer Interaction, Hillsdale, NJ. org/10.4172/jaet.1000101.
Kiviniemi, A and Fischer, M 2009, ‘Potential Obstacles to Use Salomon, G 1993, ‘No distribution without Individuals’ Cog-
BIM in Architectural Design’, in G Shen, P Brandon, and nition: a dynamic interactional view’, in Distributed
A Baldwin (eds.), Collaborative Construction Information Cognitions, Psychological and Educational Considera-
Management, Taylor & Francis, PA, pp. 36-54.Kiviniemi, tions, G. Salomon (ed), Cambridge University Press, UK
A 2013, Challenges and opportunities in the BIM edu- Stember, M 1991, ‘Advancing the Social-Sciences through
cation – How to include BIM in the future curricula of the Interdisciplinary Enterprise’, Soc Sci J, 28(1), pp.
AEC professionals? Luncheon keynote in Building Inno- 1-14.
vation 2013 Conference, Washington DC. Stevens, G 1998, The favored circle : the social foundations of
Kocaturk, T, Balbo, R, Medjdoub, B and Veliz, A 2012, ‘An In- architectural distinction, MIT Press, Cambridge, USA.
novative Approach to Technology Mediated Architec- Whitehead, H, Castellier, X, Gallou, I and Kocaturk, T 2011,
tural Design Education’, in Achten, H., Pavlicek, J., Hulin, “Interview with the Specialist Modelling Group (SMG)
J., Matejdan, D. (eds.), Digital Physicality - Proceedings of - The Dynamic Coordination of Distributed Intelligence
the 30th eCAADe Conference - Volume 1, Czech Techni- at Foster and Partners”, in: Tuba Kocaturk, Benachir
cal University, Czech Republic, pp. 55-65. Medjdoub ed(s). Distributed Intelligence in Design, Ox-
Macdonald, JA and Mills, JE 2011, ‘The Potential of BIM to ford, Wiley-Blackwell, pp.232-246.
Facilitate Collaborative AEC Education’, American So-
ciety for Engineering Education Annual Conference, [1] http://www.ribaplanofwork.com/
Vancouver, June 2011 in Proceedings of the 118th ASEE [2] http://www.cpic.org.uk/en/bim/building-information-
Annual Conference, American Society of Engineering modelling.cfm
Education, Vancouver, Canada, pp. 1-7. [3] https://www.gov.uk/government/uploads/system/up-
Ockman, J and Williamson, R 2012, Architecture school: loads/attachment_data/file/34710/12-1327-building-
three centuries of educating architects in North Amer- information-modelling.pdf

Attaining Performance with Building Information
Modelling
A systematic literature review of product and process modelling in AEC

Eleni Papadonikolaki1, Alexander Koutamanis2, J. W. F. (Hans) Wamelink3
Delft University of Technology, Faculty of Architecture, Department of Real Estate and
Housing, Section Design and Construction Management, Netherlands.
1
E.Papadonikolaki@tudelft.nl, 2A.Koutamanis@tudelft.nl, 3J.W.F.Wamelink@tudelft.nl
Abstract. The paper presents the findings of a systematic literature review of

approximately 200 scientific sources. It is designed with the aim to identify the current
benefits and factors of high performance in Architecture, Engineering, Construction
(AEC) since the introduction of Building Information Modelling (BIM). We formed and
confirmed two main propositions associating the performance of the AEC to the use
of BIM. The mapping of the current impact and benefits of BIM showed that the role
of the managers, suppliers, owners and authorities is underestimated, as well as the
initiation and use stage of project development. At the same time, the performance in
the AEC industry can be improved by an array of possibilities where IT research and
policy-making authorities contribute – from establishing new collaboration protocols until
improving existing or creating new BIM tools.
Keywords. Building Information Modelling (BIM); Architecture, Engineering and
Construction (AEC); supply chain management; life-cycle phases; stakeholders.
INTRODUCTION
Performance in architecture
The idea of performance in architecture has been and phenomena in architectural artefacts.
extensively debated during the last years, in exam- Performance is an important consideration in
ple in the “Performative Architecture” symposium many other industries, ranging from education to
organized in 2003 by Kolarevic and Malkawi (2005). commerce. For example, terms such as Performance
Discussion has focused on the “apparent disconnect Indicators or Key Performance Indicators – although
between geometry and analysis” despite the variety still a jargon from industry that lacks clear defini-
of the available digital tools (Kolarevic and Malkawi, tion – are “items of information collected at regular
2005) and on performance perceived as a qualitative intervals to track the performance of a system” (Fitz-
criterion in architecture. For the recipient of the built Gibbon and Tymms, 2002). From this perspective,
environment and the critical thinker, performance is performance is also the success factor in design. The
an objective quality measure, which offers rationale design process and the design object are the two
and clarifies the multiplicity of current approaches sides of the same coin. Yet, in architecture, arguably

due to the subjective input of the designer, perfor- struction and design management, i.e. the process
mance is neglected and the emphasis is on the aes- side of AEC. In addition to being a design tool, BIM
thetic qualities of the architectural object. But yet is also a powerful management tool (Hardin, 2009).
again, when performance becomes a serious con- Currently, there is still a lot of room for theory
sideration in architecture, it is generally restricted to building on how exactly BIM and management can
the performance of the design product, the artefact, collaborate towards the achievement of an integrat-
the building. Little is being researched over the per- ed approach in architecture. This statement coin-
formance of the design process and its significance cides with the position of the authors. Thus, the ap-
for the quality of the product. proach of BIM from our part is nor technical, nor in
Until recently Computer Aided Design (CAD) terms of design, but from a design-and-construction
software was the basis of computerization in archi- management point of view and particularly from a
tectural practice and its use portrayed the contem- supply chain (SC) perspective. AEC has a “highly frag-
porary architectural process (Aouad, 2012). The in- mented” structure (O’ Brien et al, 2009). Due to the
troduction of Building Information Modelling (BIM), lack of collaboration and coordination between the
approximately ten years ago (Eastman et al, 2008) different organizations that participate in the indus-
and its broad settlement as an integrated interdis- try, its contemporary image is of low performance.
ciplinary design environment (Deutsch, 2011); sug- We define performance as the maximum proportion
gests significant changes in not only the representa- of output to input. The four theoretical perspectives
tion of the design product but also in the structure of approaching the building SC are economy, or-
of the design and construction processes. ganisation, social and production perspectives (Vri-
jhoef, 2011). Originating from these four SC perspec-
Authors’ approach tives, we categorised ten focal points – interrelated
The advancements in technology, in both software but loosely clustered – from which to research BIM
and hardware engineering (programming, comput- (Figure 1) through a “construction SC management”
ing and networks) have resulted in a variety of so- conceptual lens.
lutions that gradually ameliorate the status of the
field. BIM technology – approach or process – is an Research question
object-oriented modelling tool that contains 3D The main inquiry of the paper is the impact of BIM in
data with “parametric intelligence” (Eastman et al, the performance of AEC industry from a supply chain
2008). In design theory, BIM is seen as an evolution perspective. First (Q1), which exact phases and ac-
of pre-existing technologies and approaches includ- tors of the AEC currently receive more benefits from
ing CAD and product modelling. This attribute ad- the application of BIM? This investigation will identi-
equately covers the product aspect of architecture. fy the positive performance of BIM in certain phases
On the other hand, BIM is a recent arrival in con- and stakeholders. The reverse argument identifies
Figure 1
Foci for researching BIM
(adapted from Vrijhoef, 2003).

the neglected – or less researched – phases and ac- types of research design for a supply chain study:
tors during the development and employment of substantive justification for theory building, surveys
BIM applications. Second (Q2), which specific fea- in SC management, case study research in SCs, ac-
tures of BIM improve the performance of AEC indus- tion research in SCs and modelling SCs (Seuring et
try throughout the whole supply chain and how? al, 2005). However, there is no right or wrong as to
The study addresses both the professional and which method to choose – descriptive or empirical
academic side of the field, given that knowledge – as long as the scope is clear and its application ad-
form practice and theory is interchangeable in the equately employed.
engineering domain. Our goal is to reconsider the Literature review has proven to be a very use-
ways we can achieve the performance required by ful tool for both qualitative and quantitative stud-
contemporary architecture and the built environ- ies. Among others, a literature review provides a
ment by illuminating the benefits of applying BIM in framework for establishing the importance of the
architectural projects and identifying gaps in litera- study and relates it to the larger on-going discus-
ture and theory. At the same time, the authors con- sion. In this case, the proposed methodological tool
sider performance in suggesting a system of actions is a “systematic quantitative literature review”, a tool
for the management of architectural projects. originating from ecology and environment sciences.
It offers an overview of existing approaches and “by
METHODOLOGY mapping the literature it is possible to highlight the
This paper uses a qualitative approach and per- boundaries around generalisations derived from the
forms a quantitative or quantifiable analysis on a literature” (Pickering and Byrne, (In Press)). The lit-
merely qualitative material, in this respect the bib- erature review described here acts as a big variance
liographic material. Traditionally case studies are the data entry method. This paper is the report of only
norm in conducting research into the performance one component – due to paper length limitations
of a process. Indeed, case studies offer a variety of – of the larger systematic literature review that was
qualitative results with inductive character and conducted between February and June 2013. The
deep understanding of the research problem. Unfor- findings that are presented here are content-related
tunately, a case study produces a local, specific and preliminary findings of the whole study.
project-driven knowledge output that is difficult to
be generalised into the performance of other sys- DATA COLLECTION AND DISPLAY
tems. The output is often limited to the interviewees’
or the researchers’ point of view. On the other hand, Collection of the material
a wide bibliographic research offers a spherical cov- The primary material comes from scientific sources
erage in actors, phases and processes, which is after and has no commercial origin. Periodical scientific
all the focus of the managerial perspective. These texts or other sources that require a short time inter-
two methods (case study and literature research) are val between the development and the publication
better to be considered as complementary rather of the research are the main material of this study.
than rivals in designing a research. Comparing such material to books, they capture a
At this point, it is important to clearly position more genuine and dynamic – although sometimes
this paper in its scientific context. As mentioned be- raw or incomplete – state of the research in time.
fore, it follows a managerial perspective concerning Consequently the selection of the material is limited
the impact of BIM in the performance of AEC. Since to peer reviewed material such as journal articles
the managerial perspective in this case is focused and conference papers.
on SC management, it is crucial to categorise it ac- Throughout the literature, the term Building In-
cording to this research approach. There are five formation Modelling or BIM appears from 2002 on-

wards mostly in commercial publications but it is tion method uses computer-assisted survey informa-
only around 2006 that the subject starts to become tion collection software, i.e. an online survey tool. An
a research object in scientific publications too. The evaluation form designed as a questionnaire collects
total number of scientific source on Building Infor- and displays the data. The questionnaire has a three-
mation Modelling or BIM during 2006-2013 accord- part skeleton composed of a descriptive part (basic
ing to Google Scholar (assessed on May 6, 2013) is information), an analytic part (focus on the content
5010, where the key terms appear in the body of the and the quality) and a conclusive part (epilogue and
text, and 344, where the key terms appear only in recap). The first two parts of the evaluation form are
the title. In order to emphasise on the material that designed in a quantitative manner with closed ques-
is indeed relevant to BIM, the appearances through- tions in order to categorise the sources and commu-
out the body of the text are considered of minor nicate the findings as much objectively as possible.
importance, since the term BIM may be only men- The third part contains open questions and accumu-
tioned in the “discussion” or the “reference” sections lates qualitative data. From these two types of data
of the source. At the same time in Scopus database only the quantitative – first two parts – are presented
(assessed on May 6, 2013), which does not hold a and discussed in this paper.
global coverage in journals and conference proceed-
ings, there are 272 sources and our topic appears on DATA ANALYSIS AND RESULTS
the title, the abstract or the keywords. All the afore-
described steps are quantifiable and repeatable. Analysis of data
The data collected from the evaluation forms of the
Display of data selected material underwent a two-level analysis.
The collected material complies with the main re- The first level involved the semantics of the scientific
search context, which is BIM in the AEC industry texts. It included data in relation to the characteris-
from an academic and an industrial point of view. Of tics of the publication and the categorisation of the
course a sort of bias may be applied in the selection primary author (Table 1). The second level concerns
of the material, since selecting a sample of 272 or the pragmatic content of the sources focusing on
344 sources allows the subjective character of the re- the overview level as well as quality assessment of
search designer to interfere. The inability to include the sources. The main findings are presented in the
material for reasons such as copyright and acces- second and third table (Tables 2 and 3). Descriptive
sibility was the only limitation in this process. Since statistics have been used to summarise the data in a
the resulted collected sample of 198 sources was not shorter form. Since most of the variables are qualita-
too large, no special sampling strategy was used. The tive or categorical, the mode – or else the frequency
failure to include the rest of the material was consid- measure – of the sample is the most important and
ered random. Aiming to explore the potential of BIM usable measure that can be applied to all variables
(as a process or a tool) from a managerial perspec- regardless their type.
tive, the research focuses on ten sub-themes: time
& cost, facility management, design process, engi- Findings
neering & consultancy, construction management, The raw findings are presented in the three tables.
construction field, sustainability, building product, Table 1 presents certain attributes of the scientific
human resources & roles and technology & data. sources in a condensed form. Both the number of
The selected material is analysed and catego- publications and percentages are stated here in or-
rised in order to answer the questions posed in the der to present an overview of the origins of the ex-
introduction. After the analysis, each scientific source isting research material on BIM. Table 2 gathers the
provides a data set with information. The data collec- mentions of BIM benefits. We categorise and corre-

Table 1 Characteristics of scientific sources on BIM Sources Frequency (%)
Condensed results from first Primary author with background from academic research 161 81.31
level of analysing the scientific Primary author with background on Construction Management 60 30.30
sources on BIM in absolute Primary author with senior expertise in industry or academia 136 68.69
and relative units. Research based on case studies 126 63.64
Global applicability of the research 159 80.30
Research published in scientific journals 105 53.03
late these benefits according to the project phases of Results

AEC and the actors participating to it. Table 3 collects Most of the publications on BIM (81.30%) have
the benefits from using certain BIM features catego- been authored by researchers originating from the
rised under the relevant SC research perspectives. academia (Table 1). A number of these researchers is
also active in industrial organisations. Construction
Table 2
BIM benefits
Owners & FM
Benefits from using BIM per
Consultants
Contractors
Authorities
Regulatory
(Number of
Architects
actor of the AEC industry and
Engineers
Managers
Suppliers
project phase in absolute references in
numbers of the sources. the sources)
Initiative 53 43 35 22 29 12 26 21
Design 124 78 107 52 64 20 42 30
Construction 54 57 74 29 65 27 27 20
Use 16 27 23 17 15 13 48 16
Table 3 BIM features
Direct fabrication control

(Number of
Construction scheduling
BIM features and SC research
Environmental analysis
Facilities management
perspectives correlation references in
Preliminary massing
derived from the literature in the sources) Collaboration tools

Specification tools
Contracting tools
Quantity take-off
Mechanical tools
absolute numbers.
Feasibility tools
Cost estimation
Clash detection
Visualisation
Time & cost 32 14 22 5 16 22 28 12 8 33 10 35 11 6

Facility mgmt 6 2 16 5 11 15 12 10 13 5 5 5 0 18
Design process 31 24 46 27 87 58 44 18 28 24 16 18 11 15
Engineering 22 18 31 12 38 37 34 23 15 16 14 23 10 6
Constr. mgmt 20 24 27 7 23 43 23 8 4 21 14 44 11 5
Constr. field/site 11 17 23 7 20 20 19 13 3 15 12 30 10 5
Sustainability 13 0 5 1 9 9 9 1 20 6 2 2 0 5
Bldg product 13 6 17 3 18 22 9 7 18 8 3 6 2 13
HR & roles 8 9 7 4 12 31 12 0 3 4 3 6 1 7
Technology & data 13 12 49 8 40 40 30 20 18 15 14 24 8 14

BIM benefits Table 4
Owners & FM
(%) Summary of references on the
Consultants
Contractors
Authorities
Regulatory
Architects
Engineers
Managers
actors and phases of the AEC
Suppliers
industry benefited from BIM.
Initiative 26.77 21.72 17.68 11.11 14.65 6.06 13.13 10.61

Design 62.63 39.39 54.04 26.26 32.32 10.10 21.21 15.15
Construction 27.27 28.79 37.37 14.65 32.83 13.64 13.64 10.10
Use 8.08 13.64 11.62 8.59 7.58 6.57 24.24 8.08
managers have authored 30.30% of the research on Table 5 emphasises with bold the mentions of
BIM. This element strengthens our initial proposition the most prominent BIM features and with italics
that BIM is considered more as a tool to achieve an of the most underused. The aim here is to indicate
occasional high performance, rather than as a per- the features of BIM that improve the performance
manent project management tool or a process to of AEC industry throughout the whole supply chain
be used towards the integration of the construction (A2). According to the literature review, BIM features
supply chain. At the same time, the fact that not such as visualisation, clash detection and collabo-
only junior but also a lot of senior researchers are ration tools are the most researched by far, which
keeping busy with BIM reveals that they are already on the one hand increases the performance of the
convinced about its potential and the impact and building product but on the other hand contributes
are committed to put their expertise in action. The to the performance of AEC only incidentally. For in-
majority of the publications (63.64%) use case stud- stance, quantity take-off and facility management
ies and experiments to validate their hypotheses. tools – employed mostly for facilitating the suppliers
Table 4 contains the data of Table 2 in percent- and the owners respectively – are either neglected
ages and indicates with bold the number of sources for certain research perspectives or only appear
where the actors and the phases experience more in the 6 to 9% of research into BIM. Likewise, while
benefits from the employment of BIM and with ital- there are tools for the construction field, such as di-
ics where the actors and the phases profit less. Ap- rect fabrication tools (Table 5), they are seemingly
parently (A1) the architects and the engineers are not widely applied or reported. Other BIM features
the actors who are either the participants more in- mentioned but not included here are laser scanning
volved in the research and adoption of BIM – or are and tools for safety on the building site.
simply considered the primary actors – in BIM litera-
ture (Table 4). Surprisingly, construction managers DISCUSSION AND CONCLUSION
are not equally prominent to these primary actors,
as one might have expected but they are more in- Discussion
volved in all phases of a project, while contractors The research design answered sufficiently the re-
are referred to mostly in the construction phase. search questions. Comparing this research to other
Suppliers are also referred in the construction phase studies, the most apparent difference is to be found
but are limited to peripheral roles in the rest of the in the methodology. Identifying benefits and quan-
building life-cycle. On the other hand, owners and tizing performance via literature review is not the
regulatory agencies seek immediate involvement norm in this domain. The present research shares
but achieve only fragments of presence mostly dur- common concerns and limitations as publications
ing the initiative and use stages. based on case study research. Comparing it with

Table 5 BIM features
Direct fabrication control
Construction scheduling
Summary of correlations (%)
Environmental analysis
Facilities management
between BIM features and SC
Preliminary massing
Collaboration tools
Specification tools
research perspectives.
Contracting tools
Quantity take-off
Mechanical tools
Feasibility tools
Cost estimation
Clash detection
Visualisation
Time & cost 16 7 11 3 8 11 14 6 4 17 5 18 6 3
Facility mgmt 3 1 8 3 6 8 6 5 7 3 3 3 0 9
Design process 16 12 23 14 44 29 22 9 14 12 8 9 6 8
Engineering 11 9 16 6 19 19 17 12 8 8 7 12 5 3
Constr. mgmt 10 12 14 4 12 22 12 4 2 11 7 22 6 3
Constr. field/site 6 9 12 4 10 10 10 7 2 8 6 15 5 3
Sustainability 7 0 3 1 5 5 5 1 10 3 1 1 0 3
Bldg product 7 3 9 2 9 11 5 4 9 4 2 3 1 7
HR & roles 4 5 4 2 6 16 6 0 2 2 2 3 1 4
Technology & data 7 6 25 4 20 20 15 10 9 8 7 12 4 7
previous studies, it focused on the performance of ure 2). Undoubtedly, with the still rapidly evolving
the AEC process via the use of BIM rather than “dis- state of the information age, research directions
cussing how information systems can further con- in the field may change day by day. However, this
tribute to this research domain” (Merschbrock and study has showed that there are certain neglected
Munkvold, 2012). There are again limitations over research areas and correlations that arguably ex-
how exactly to measure performance, a problem plain the low performance of the AEC industry. The
already mentioned in other studies (Barlish and Sul- content of this paper offers a guide to improving
livan, 2012). A solution to this problem is be the clas- the behaviour of the neglected project phases and
sification of benefits as having a positive or a nega- actors by integrating the construction supply chain.
tive impact, as suggested in research on case studies Concerning the methodology, the research adds on
(Bryde et al, (In Press)). Apart from sharing common how to conduct literature research with an eye not
concerns and limitations with existing researches, only in the semantics and external characteristics of
the present study has the dual advantage of includ- the scientific material but also in the overview level
ing all the involved participants in the AEC industry of the scientific material. The present method could
and referring to all the stages of the AEC. also be employed in the future by either focusing on
Although the findings presented here do not a narrower research field or including certain types
cover the full extent of the research conducted – of publications. Lastly, anticipating the criticism over
due to paper length limitations – the main results the credibility of a literature review, we defend the
already suggest concrete directions for further use. selection of this research design by restating the
From the summarising tables in the results section quality assessment that was incorporated during
(Tables 4 and 5) we indicate certain directions that the employment of the experiment.
require further attention and investigation (Fig-

Figure 2
Proposed action framework
for BIM research (Horizontal
stripes: Collaboration proto-
cols, dark grey: IT tools).
Conclusion take-off are the most underused tools.

The study described in this paper identified the Combining the research results from Tables 4
life-cycle phases and stakeholders who experience and 5, we propose a framework towards a highly
more benefits from the current application of BIM performative AEC industry (Figure 2). New and
and thus are considered high performance param- stronger collaboration protocols between the man-
eters for the industry. On the other hand, the phases agers and the owners should be implemented in the
and stakeholders who are either neglected or simply initiative phase, and likewise appropriate collabora-
left behind in the adoption of BIM are of low perfor- tion protocols should be applied in the use stage of
mance and may be subjected to future research. The AEC (dashed horizontal stripes). The employment of
research also revealed the BIM features that are cur- supply chain integration in construction aims to reg-
rently used more extensively and the BIM features ularise and enhance the involvement of all actors in
with low applicability of popularity in a sense. this stage and at the same time improve the perfor-
Using the most and the least benefited from mance of the AEC (light grey horizontal stripe). We
BIM actors and AEC phases (as indicated in Table 4), also argue that the underused BIM features should
we identify gaps not only in the performance of the be extended or improved in order to serve the in-
participants, but also in the performance of the vari- volvement of all the stakeholders in the AEC process
ous processes that take place during a project. For (dark grey stripes).
example, in the initiative phase, only the architects Finally, we observe that while the role of the ar-
are adequately involved, while they are less involved chitect is being given adequate attention in BIM re-
– along with all other actors apart from the owners search and adoption, the role of the manager is not
– in the stage of end-use. On the other hand, in the equally emphasised. The existence of many and vari-
design phase almost all the participants – with the ous BIM features arguably makes as the manager an
exception of the suppliers – are equally benefited integrator of the whole process rather than merely
from the implementation of BIM, whereas in the another BIM-user. To conclude, apart from reach-
construction phase there is a continuous fluctuation ing our own research objectives, the present study
in regard to the involvement of the actors (Table 4). forms a roadmap for fellow researchers interested in
At the same time, Table 5 reveals that preliminary the domain of BIM by revealing subjects for further
massing, direct fabrication control and quantity exploration.

REFERENCES apolis, Indiana, USA.
Aouad, G 2012, Computer Aided Design Guide for Architec- Kolarevic, B and Malkawi, A 2005, Performative Architecture:
ture, Engineering and Construction, Routledge, London, Beyond Instrumentality, Spoon Press, New York, USA.
UK. Merschbrock, C and Munkvold, BE 2012, ‘A Research Review
Barlish, K and Sullivan, K 2012, ‘How to measure the ben- on Building Information Modeling in Construction–An
efits of BIM — A case study approach’, Automation in Area Ripe for IS Research’, Communications of the Asso-
Construction, 24(0), pp. 149-159. ciation for Information Systems, 31, pp. 207-228.
Bryde, D, Broquetas, M and Volm, JM (In Press), ‘The project O’ Brien, WJ, Formoso, CT, Vrijhoef, R and London, KA 2009,
benefits of Building Information Modelling (BIM)’, Inter- Construction Supply Chain Management Handbook, CRC
national Journal of Project Management. Press Taylor & Francis Group, Boca Raton, Florida, USA.
Deutsch, R 2011, BIM and Integrated Design: Strategies for Ar- Pickering, CM and Byrne, J (In Press), ‘The benefits of pub-
chitectural Practice, John Wiley & Sons, Hoboken, New lishing systematic quantitative literature reviews for
Jersey, USA. PhD candidates and other early career researchers.’,
Eastman, C, Teicholz, P, Sacks, R and Liston, K 2008, BIM Higher Education Research and Development (HERD).
Handbook: A Guide to Building Information Modeling for Seuring, S, Müller, M, Reiner, G and Kotzab, H 2005, ‘Is There
Owners, Managers, Designers, Engineers, and Contrac- a Right Research Design for Your Supply Chain Study?’
tors, Second ed., John Wiley & Sons, Hoboken, New in H Kotzab, S Seuring, M Müller and G Reiner (eds),
Jersey, USA. Research Methodologies in Supply Chain Management,
Fitz-Gibbon, CT and Tymms, P 2002, ‘Technical and ethical Physica-Verlag HD, pp. 1-12.
issues in indicator systems: Doing things right and do- Vrijhoef, R 2011, Supply chain integration in the building in-
ing wrong things’, Education Policy Analysis Archives, 10. dustry: The emergence of integrated and repetitive strat-
Hardin, B 2009, BIM and Construction Management: Proven egies in a fragmented and project-driven industry, IOS
tools, Methods and Workflows, Wiley Publishing, Indian- Press, Amsterdam, The Netherlands.

Building Your Own Urban Tool Kit
Utilizing parametric BIM components as smart early design tools for large-
scale urban planning
Caner Dolas1, Andreas Dieckmann2, Peter Russell3

Chair for Computer Aided Architectural Design, RWTH Aachen University, Germany
http://caad.arch.rwth-aachen.de
1
dolas@caad.arch.rwth-aachen.de, 2dieckmann@caad.arch.rwth-aachen.de, 3russell@
caad.arch.rwth-aachen.de
Abstract. The paper describes the development of a set of smart BIM components to
facilitate and accelerate the creation of large-scale urban models in the early design
phase in a BIM software environment. The components leverage the analytical,
parametric and modelling capabilities of the BIM environment to support adaptive
parameter-driven building geometry, patterning of different building types, early
numerical and graphical design evaluation, various simulation methods and the
exploration of design alternatives. The toolset consists of the most common building
shapes, but can be extended with additional shapes and their respective area and
volumetric calculations when necessary. The rapid large-scale deployment of the
components has been achieved by diverting existing tools from their intended use.
Keywords. BIM; urban planning; early design; rule-based design; parametric design.
PROJECT CONTEXT: BIM VS. GIS.

Building Information Modelling (BIM) is quickly et al., 2005), it is supported by a growing number
becoming the de-facto standard in the computer of commercial software products [1]. At present,
aided design and documentation of buildings, albeit however, none of these are BIM applications (and
with varying adoption rates in different world re- are thus located outside of the “comfort zone” of ar-
gions (McGraw-Hill Construction, 2010 & 2012). Data chitects), but the issue of interoperability between
structures in BIM applications can be described as building and city models has become a hot topic
semantic, parametric and component-centric (East- of research in recent years (Nagel and Häfele, 2007;
man et al., 2008). BIM applications utilize the Indus- Isikdag and Zlatanova, 2009; El-Mekawy, 2010; De
try Foundation Classes (IFC) file format that was first Laat and Van Berlo, 2011). Therefore it stands to rea-
released in 1996 (Bazjanac and Crawley, 1997) for son that in the foreseeable future BIM applications
the exchange of semantic data models. will gain the capability to author semantic urban
For the urban scale, similar efforts to create a se- models.
mantic data model have been made with CityGML The parametric and analytical capabilities of
(Gröger and Plümer, 2012). Although, compared to BIM applications have proven to be somewhat use-
IFC, CityGML is a relatively young data format (Kolbe ful for urban design in the past (Miller at al., 2009)

and research has been conducted towards the impact (e.g. CO2 footprint).
plementation of zoning requirements in building in- In turn, the requirements for the host application it-
formation models (Donath and Lobos, 2006; Kim et self were defined as follows:
al., 2011). However, with increasing project scale, the 1. Allow for the creation, distribution and mainte-
mere process of placing a large number of diverse nance of components according to the require-
elements, altering their attributes and exploring ments listed above.
design alternatives has left much to be desired with 2. Generate tabular reports of component attrib-
regard to speed and usability. utes.
3. Allow for visual filtering based on component
REQUIREMENTS FOR A SMART TOOLBOX attributes.
The focus of the project was to create a toolbox of 4. Facilitate the creation and comparison of de-
smart components that could be used as generic sign alternatives.
building masses inside a BIM application for large- 5. Possibly even provide additional analysis tools.
scale urban planning projects. Building models in Based on the above requirements, Autodesk’s Re-
urban design projects are usually made up of a lim- vit platform [2] was chosen as host application.
ited number of building archetypes, yet each build- Not only does it meet all the requirements, but its
ing instance has to accommodate the geometric conceptual modelling application Vasari [3] also in-
conditions and zoning requirements of its respec- cludes analysis tools for environmental factors like
tive parcel as well as the overall design intent. Addi- sun and wind.
tionally, there is a strong need for evaluation, espe-
cially in the early design phase, in order to facilitate DESIGNING FOR RAPID DISTRIBUTION
informed decision-making. Hence, the following re- Repetition and variation are common concepts in ar-
quirements were set for the components: chitecture. They can be easily identified in building
1. A large number of components can be dis- elements such as curtain walls, staircases, railings,
tributed rapidly in a given spatial framework, structural systems etc. BIM applications generally
making it possible to create a large-scale urban provide dedicated tools for these types of build-
model in a fairly limited amount of time. ing elements. The same concepts of repetition and
2. The components can adapt to varying lot sizes variation apply to urban planning as well, perhaps
and geometries. with a special emphasis on the adaptability of build-
3. The components allow for the rule-based para- ings to the geometric conditions of their respective
metric generation of building forms according parcels. However, there are no dedicated tools for
to zoning requirements. distributing a large number of building masses in a
4. The components can accommodate different typical BIM application. Therefore, the approach was
terrain conditions, i.e. they work on level and to divert tools readily available in the chosen appli-
sloped terrain surfaces. cation from their intended use.
5. It is possible to automatically or at least semi- With the 2010 version of Revit, Autodesk intro-
automatically place components in patterns duced a new conceptual modelling environment
(such as ABAB etc.) to allow for variations in the that was intended for the modelling of building
design. masses [4]. The potentials of this modelling envi-
6. The components can instantly report most if ronment were described by Miller et al. (2009), but
not all essential numerical information relevant the workflow outlined by them involved the man-
in a typical urban planning scenario such as ual modelling of each building (or at least manual
areas, volumes etc. but also, if possible, some changes to placed building instances). The 2010 ver-
statistical data on costing or environmental im- sion did, however, come with another functionality

Figure 1
From left to right: a) Mass
surface, b) Automatic subdivi-
sion, c) Manual subdivision.
with a lot of potential regarding the adaptability of a ure 1c) by drawing a number of lines to gener-
large number of objects to varying geometric condi- ate the subdivisions.
tions: Mass surfaces could be rationalized by using The actual toolbox consists of several types of build-
the “divide surface” functionality and subsequently ing masses created as pattern-based elements and
be populated with “pattern-based curtain panels”. adaptive components that can be hosted on and
Revit 2011 saw the introduction of the “adaptive rapidly distributed across divided surfaces. Depend-
components” functionality: placement point based ing on the desired outcome, two separate modelling
components that can adapt to varying spatial condi- strategies can be applied for populating the grid
tions [5]. Lastly, with the 2013 version came the “re- with the building masses:
peat and divide” workflow that can be used to cre- 1. For a simple pattern, the divided surface can
ate more complex arrays of objects (Dieckmann and be assigned a pattern-based component (Fig-
Kron, 2012) and facilitate the large-scale distribution ure 2a), essentially distributing instances of the
of reactive components (“reactors”) as described by same building block across the entire grid of a
Woodbury (2010). block. Exceptions can be defined by selecting
Surely none of these functionalities were de- individual instances and manually switching
signed with large-scale urban planning in mind – their type or altering their instance properties
most of them are typically used for the creation of (Figure 2b).
curtain wall systems and other building elements 2. More complex patterns of several alternating
– but they can be “abused”. In the context of the pro- building types can be created as one or two di-
ject, the aforementioned tools are used as follows: mensional arrays by employing the repeat and
1. The footprints of city blocks are created as divide workflow (Figure 2c). In addition, this
mass surfaces (Figure 1a). workflow allows for the rapid deployment of
2. These mass surfaces can then be subdivided context-aware adaptive components that can,
into lots using the divide surface functionality, for instance, react to the proximity of other ob-
creating a grid within the city block. The grid jects in the model (Figure 2d). A common ap-
can either be generated automatically (Figure plication for this method would be the increase
1b) using a layout algorithm (e.g. number of of density towards certain zones in the urban
Figure 2 subdivisions in U/V direction) or manually (Fig- model (see below).
From left to right: a) Di-
vided surface populated with
pattern-based components, b)
Manual exceptions,
c) Patterning with divide & re-
peat functionality, d) Reactor
pattern with context-aware
adaptive components.

Figure 3
Nested component structure.
COMPONENT OVERVIEW ronment. They mainly consist of “dumb” geometry

For the purpose of surpassing a mere proof-of-con- and only perform the following tasks:
cept stage, component types were developed for 1. Evaluation of the input received from the par-
most commonly found building shapes: I-shaped, ent component.
L-shaped, U-shaped, O-shaped and solitaire. The lot 2. Generation and positioning of building geom-
and building block components are organized in a etry in the context of the lot geometry based
nested object structure (Figure 3). The lot compo- on the received inputs.
nent, a pattern-based element, is intended for: In the case of more complex design intent such as
1. Placement on and distribution across the city gradually increasing the building density towards a
block’s grid. subregion within the planning area, lot components
2. User input. Depending on the component de- can be nested in another context-aware adaptive
sign, the input can consist of different types of component that is able to track its proximity to said
rules and constraints such as building dimen- subregion and drive these parametric constraints
sions, setback, plot area ratio (PAR), floor-to- in the building block components, as described by
floor height, usage type, building orientation Dieckmann and Kron (2012) for curtain wall panels.
etc.
3. Evaluation of lot geometry (dimensions and Lot Component Anatomy
angles, where applicable). The lot component is created as a pattern-based
4. Communication of user input and lot geometry element, a component that is based on a number
to the nested building block component. of placement points. As the lot component has to
5. Calculation of the required numerical data adapt to varying geometric conditions set by the
needed for design evaluation (e.g. building geometry of the city blocks, it needs to be aware of
footprint, building volume, cubic index etc.). its own shape and size, i.e. the lengths of its edges
Nested inside the lot component are one or sev- and the angles between those edges. In Revit, such
eral instances of building block components. These properties can be measured by using so-called re-
adaptive components are linked to their parent porting parameters that report the varying dimen-
component by parametric relationships. As they are sions for each placed instance of a pattern-based
created as what is referred to in Revit as “non-shared” component. While the components may be placed
components, they are completely absorbed by their on sloped surfaces, the dimensions need to be
parent component and can neither be selected nor measured in top projection in order to be used for
scheduled as separate elements in the project envi- the calculation of areas and lengths later on (Figure

Figure 4
From left to right: a) Dimen-
sion measurements on
horizontal plane, b) Horizontal
datum between vertical rays,
c) Definition of building zone,
d) Placement of building
component.
4a). This is done by hosting all the dimensions on the controlled by a parameter, making it easy to change
horizontal work plane of the first placement point. the orientation of the component (front, right, back
The geometry of a pattern-based component by and left side of the lot) as well as the building shape
default inherits the orientation of its host, i.e. the di- (I, L, U, O). This also allows for the subsequent crea-
vided surface of the city block. That means that ver- tion and substitution of other building shapes es-
tical elements created in the lot component would sentially making it a modular system. Additionally,
rather orient themselves according to the surface all the parameters that control the building shape
normals of the city block than vertically at their point (building depth for all sides of the lot and building
of placement. By changing the orientation mode of height) are also passed to the subcomponent. As
the placement points the lot component geometry stated above, the building subcomponents merely
can however be forced into a strictly vertical orienta- consist of the building geometry driven by the lot
tion. The placement point location can then be pro- component parameters and thus warrant no further
jected upwards by means of vertical rays. On sloped description.
lots, the building may have to be moved up or down For the purpose of calculating the building foot-
so as not to be fully or partly immersed in the ter- print and related data like floor space and building
rain. This can be achieved by creating a horizontal volume, the central zone is again subdivided into
datum between the aforementioned rays (Figure nine zones, this time by using the building depths
4b) that can be moved by manipulating a parameter for the four sides. Again, the depth for each side is
that controls the vertical offset of the datum. user-controlled with a safeguard against overlaps as
The horizontal datum serves as the placement described above for the street offsets. The footprint
plane for the building component itself. It is subdi- of each building type can now be calculated as the
vided into nine zones by projecting the street off- sum of some of the zone areas (Figure 5), depend-
set for all four sides of the lot onto the datum (Fig- ing on the selected building type, e.g. the footprint
ure 4c). These offsets can be controlled by the user of the O-shaped building would be the sum of all
through four parameters. In case the street offsets zones except for the central zone. The zone areas
of opposing sides of the lot overlap, the user inputs themselves are calculated on the basis of the report-
will be substituted by a “safe” value that is automati- ing parameters (see above) using Heron’s formula
cally calculated. and the law of cosines. Subsequently, all other data
The four intersection points of the street offsets necessary for evaluation such as cubic index, floor
form the location for the placement points of the area ratio or site occupancy index can be derived
building component (Figure 4d) and also mark the from the building footprint, the number of floors,
vertices of the central zone that forms the basis for the floor height and the site area. In Revit, custom
the building footprint calculations (see below). Once component parameters can not be scheduled or
a building component is placed here, its type can be annotated in the project environment by default.

Figure 5
From left to right: a) Known
(red, dotted), auxiliary (green,
dashed) and calculated (blue,
continuous) dimensions
for area calculations, b)
Zone combinations for area
calculations.
Thus, in order to have the data readily available in ric relationships that aid with the area calculations
the project for evaluation, they need to be declared for the standard building types (I, L, U, O). Instead, it
as so-called “shared” parameters making them avail- contains a center point for the free-standing build-
able globally (in the component itself as well as in ing geometry that can be moved parametrically in
the project). U and V direction on the lot surface. The building
geometry that is hosted on the point in turn has a
Component Variations rotation parameter to allow for flexible alignment of
The lot component can be used as a template to cre- the building mass.
ate further variations. They can either be different A reactor component (Figure 6b) as described
building types than the four types described above, above can use either the solitaire component or the
more complex parametric components that utilize standard lot component as nested subcomponent.
the lot component as a subcomponent or a combi- It is basically an adaptive component that sets up
nation of both. rules for the behaviour of its subcomponent. It has
The solitaire component (Figure 6a), for instance, one or several additional placement points that
makes use of the spatial and parametric framework act as sensing devices. By hosting these additional
of the lot component. However, it needs neither placement points on certain fixed points in the pro-
the street offset grid nor the majority of paramet- ject and measuring their distance from each placed
Figure 6
From left to right: a) Solitaire
component, b) Reactor
component.

Figure 7
From left to right: a) Example
of a lot schedule with
conditional formatting, b)
Example of a filtered model
view colour-coded by cubic
index value ranges.
instance of the reactor component, the components able in the model. The schedules can utilize condi-
gain spatial awareness. This information can then tional formatting to highlight lots that do not meet
be used to control the geometric properties of each certain requirements like, for instance, a cubic index
placed subcomponent, e.g. the number of storeys. that exceeds a certain limit (Figure 7a).
A schedule is, however, just one way of looking
WORKING WITH THE TOOLKIT at information. The same information can also be
The typical workflow has been, at least in part, de- visualized in isometric, perspective or plan views,
scribed above already: The city blocks are created displaying the information in a spatial context. In Re-
as mass surfaces and subsequently subdivided into vit, model views can be reformatted with so-called
lots. Depending on design intent, several distribu- view filters. By means of a few view filters a perspec-
tion methods (uniform, uniform with exceptions, tive view of the project can be colour-coded accord-
patterned and reactive/parametric) are available ing to value ranges of any given parameter like, for
(Figure 1). The component type(s) assigned to a instance, the cubic index of each lot, with different
block, a lot or a pattern can be changed and their colours for different value ranges (Figure 7b).
instance properties can be modified. The shapes of Often, the building type has a significant influ-
the mass surfaces themselves and the number of ence on the measurable characteristics of a building.
their respective subdivisions can also be modified For instance, the energy use of a building depends
at any time. Moreover, several out-of-the-box func- quite heavily on the activity within that building.
tionalities like design options (managing different There are some statistical resources available for
design alternatives) and phasing (managing the that kind of information, like the Buildings Energy
temporal properties of elements, i.e. differentiating Data Book by the U.S. Department of Energy [6].
between existing and new building blocks) can be However, for the purpose of this paper, the authors
utilized to structure and control the design. have focussed on costing. In a lot of countries, there
The main reason for using a BIM environment are statistical data available on the building costs for
for urban design, however, is the ability to create various building types. For the german market, this
information-rich content and leverage that informa- data is made available by the BKI Baukosteninfor-
tion to evaluate the design. All the numerical data mationszentrum (2013). In Revit, external data can
produced by the placed components can be easily be inserted in the form of so-called key schedules,
scheduled. Each lot component contains a flag pa- either by inputting it manually or by using third-
rameter that facilitates the creation of a schedule party applications [7] to import it from Excel. A row
that only displays the lot components placed in the of values from a key schedule can be assigned to a
project and ignores all other site components avail- placed component by means of a key parameter.

After that, a costing schedule can easily be created ence. Prague (Czech Republic).
that contains parameters that, for instance, calculate BKI Baukosteninformationszentrum (Ed.), 2013. Baukosten
the building cost on the basis of the building vol- Gebäude 2013 Teil 1: Statistische Kostenkennwerte.
ume and the cost per cubic meter specified in the Stuttgart (Germany).
key schedule for a particular building type. De Laat, R and Van Berlo, L 2011, Integration of BIM and GIS:
Once the schedules and filtered views have The Development of the CityGML GeoBIM Extension,
been set up properly, the project file can be used to in: Kolbe, T.H., König, G., Nagel, C. (Eds.), Advances in
create a project template for future urban planning 3D Geo-Information Sciences. Springer Berlin Heidel-
projects. This way, the information will be readily berg, pp. 211–225.
available as soon as the designers start placing the Dieckmann, A and Kron, Z 2012, Echo Chamber: Complex
first lot components – they could even model the Parametric Arrays in Autodesk Revit, in: Autodesk Uni-
project in a filtered view for direct visual feedback. versity 2012. Las Vegas, NV (USA).
Donath, D and Lobos, D 2006, Massing Study Support: A
DISCUSSION New Tool for Early Stages of Architectural Design, in:
The presented method facilitates a quick, albeit Architecture in Computro [26th eCAADe Conference
makeshift, workflow to create early design models Proceedings]. Antwerpen (Belgium), pp. 101–108.
for large-scale urban planning projects in a BIM ap- Eastman, CM, Teicholz, P, Sacks, R and Liston, K 2008, BIM
plication. All relevant numerical data is generated Handbook: A Guide to Building Information Modeling
on-the-fly by the components themselves as they for Owners, Managers, Designers, and Contractors.
are placed in the context of the site. The design can Wiley, Hoboken, NJ (USA).
therefore be immediately evaluated – either numeri- El-Mekawy, M 2010, Integrating BIM and GIS for 3D City
cally or graphically – making it easy to explore dif- Modelling – The Case of IFC and CityGML, Licentiate
ferent design alternatives. Additionally, the chosen Thesis.
host application has the capability of performing Gröger, G and Plümer, L 2012, CityGML – Interoperable Se-
environmental analyses for the impact of sun and mantic 3D City Models. ISPRS Journal of Photogram-
wind on the design. The components, of course, metry and Remote Sensing 71, 12–33.
still have some limitations, e.g. a useful functionality Isikdag, U and Zlatanova, S 2009, Towards Defining a
would be to be able to assign more than one usage Framework for Automatic Generation of Buildings in
type to a building, perhaps per storey. CityGML Using Building Information Models, in: Lee,
For the sake of interoperability, a sensible next J., Zlatanova, S. (Eds.), 3D Geo-Information Sciences.
step would be to reach the capability to export the Springer Berlin Heidelberg, pp. 79–96.
model to CityGML format. Previous research on Kim, JB, Clayton, MJ and Yan, W 2011, Parametric Form-
the subject of marrying IFC and CityGML quoted in Based Codes: Incorporation of Land-use Regulations
this paper has focussed on the conversion of entire into Building Information Models, in: Parametricism
building models to several levels of details (LOD) in (SPC) ACADIA Regional 2011 Conference Proceedings.
CityGML. In this specific case, a conversion of single Lincoln, NE (USA), pp. 217–223.
elements (generic models) in the building model to Kolbe, TH, Gröger, G and Plümer, L 2005, CityGML – Inter-
LOD 1 or 2 CityGML building entities would do the operable Access to 3D City Models, in: Proceedings of
trick. the International Symposium on Geo-Information for
Disaster Management. Delft (Netherlands).
REFERENCES Miller, L, Ho, HH and Light, D 2009, Autodesk Revit for Ur-
Bazjanac, V and Crawley, DB 1997, The Implementation of ban Design, in: Autodesk University 2009. Las Vegas,
Industry Foundation Classes in Simulation Tools for the NV (USA).
Building Industry, in: Building ’97 Simulation Confer- Nagel, C and Häfele, KH 2007, Generierung von 3D-Stadt-

modellen auf Basis des IFC-Gebäudemodells, in: En- [1] http://www.citygmlwiki.org/index.php/Commer-
twicklerforum Geoinformationstechnik 2007. Berlin cial_Software
(Germany). [2] http://usa.autodesk.com/revit
McGraw-Hill Construction, 2010. The Business Value of BIM [3] http://autodeskvasari.com
in Europe. [4] http://www.aecbytes.com/review/2009/RevitArch2010.
McGraw-Hill Construction, 2012. The Business Value of BIM html
in North America: Multi-Year Trend Analysis and User [5] http://www.aecbytes.com/review/2010/RevitArch2011.
Ratings (2007-2012). html
Woodbury, R 2010, Elements of Parametric Design, Rout- [6] http://buildingsdatabook.eren.doe.gov/default.aspx.
ledge, NY. [7] http://www.whitefeet.com/Tools

[Architectural] Reasoning over BIM/CAD Database
How to combine reasoning powered ontologies with BIM/CAD tools (and

vice-versa?)
Gianluigi Loffreda1, Antonio Fioravanti2, Luigi Avantaggiato3

Sapienza University of Rome
http://www.dicea.uniroma1.it
1
gianluigi.loffreda@uniroma1.it, 2antonio.fioravanti@uniroma1.it, 3lavanta@tiscali.it
Abstract. Design activity is pervasive as it is increasingly expanding into all sectors and
every day it is increasingly difficult to anticipate the often unpredictable changes resulting
from new inventions and changes in technology, tools, methods and social customs using
current design systems, and at the same time we need to preserve and store knowledge
and experiences that can help facing aforementioned problems. The present paper
illustrates an innovative Rule Layer overlying existing commercial software in order to
model Reasoning and Performance verification Rules to be applied to design instances.
The authors developed two different prototypes, one on BIM and one on CAD commercial
software in order to validate the proposed approach. Results demonstrate the general
system potentials opened up to further research development and deepening.
Keywords. Building ontologies; building design reasoning; BIM/CAD; collaborative
design.
INTERLEAVED WORLD
Comparing our “era” with the past, people in deve- in the world completely absorbed what was elabo-
loped countries obviously live in better conditions rated from most advanced scientific and philosophy
than before due to the organization of society and researches: the importance of science (also social)
technological evolution. On the other hand, sociality facts, of measurable quantities, referred to phenom-
has been replaced by competitiveness, mainly as a ena expressed in mathematical-analytical formulas.
result of increasing complexity and changing needs Hence fundamental science courses were
that require new approaches in all human activi- taught, on which ‘objective’ base the following dis-
ties in order to meet increasing demands (Einstein ciplines were set up. This ‘functional’ logic character-
2006). ized scientific as well as humanistic Schools.
The problems are mainly related to the ‘idea’ of In short: avant-gardes inquired into “first” prin-
science and science law we have. In the past, our ciples until the First World War; afterwards, the re-
general conceptual elaborations were based on sults of these researches were applied to well lim-
Thirties period. That time all academic institutions ited scopes. It was usual to describe phenomena by

means of first order theories (superposition of ef- ponent, space, 4D and 5D design process.
fects) in application domains precisely defined (limit In the contemporary world, which is increas-
conditions). ingly linked to reduced distances in terms of space,
However, in the middle of last century new language and, sometimes, even culture, humans are
studies and ever-growing specializations have led evolving (or counter-evolving) into users connected
scientific communities to verify that many phenom- by emails, social networks, news and any other web-
ena could not be explained by means of these two based sources.
assumptions. We discovered that also apparently All these social links among different users con-
simple phenomena are interrelated with context trast with the exponential increase in specialization
so belong to the category of Complexity, the right of professionals in their own domain: as stated by
approach to these studies is a multi- and cross-dis- Simon: “Once a profession reaches the point where
ciplinary ones and very often only Chaos theory can it takes 10 years to master, it tends to break up into
explain them. We passed from mathematical formu- specializations”.
lae expressed by means of multiplications and loga- In Architectural Design processes, many differ-
rithms to system of integrals with not exact solution, ent specialist domains are involved as well as several
but partial derivative differential equation solutions; specialist designers in their respective fields. These
lastly concepts can be computed thanks to perform- are changing the twentieth century-related design
ative computer systems. approaches made up of functionality verification,
Referring to Building Design Tools, the evolution client requirement checking, cost control and time
from paper and pencil to CAD systems, to Object scheduling, cross disciplinary expertise into increas-
Oriented Systems and ultimately to BIM platforms ingly narrow specializations and knowledge secto-
has led, step by step, to new design methodology ralisation, under the effect of increasing technology
and consequently to different design results. As a complexity and discipline multiplication.
matter of facts, these methodologies and tools to- Nowadays buildings are evolving into “smart”
gether with new social and architectural sensibilities buildings, control panels substitute light switches,
influenced and in turn were influenced by current sensors will ultimately allow program building ser-
contemporary buildings and also by high perform- vices to adjust performance to suit changing envi-
ing ones as shown in architectural masterpieces of ronmental conditions, forms will change according
archistars like Eisenman, Gehry, Toyo Ito, Fuksas, etc. to weather, climate or use functions.
CAD systems and their three-orthogonal coordi- However, resources are limited, costs are out of
nates systems have certainly supported the design control in a blurred economic, technical and social
process in building representation but have also context, and sustainability becomes a necessity, not
indirectly influenced generations of architects and just a possibility.
designers as far as space configuration and overall In order to allow different Specialists to collabo-
building design are concerned. These systems were rate in a Design Process in an effective and produc-
used to represent only geometry, 2D or 3D and tive way, the present abstract presents a prototype
sometimes tagged entities to specify space destina- structure for an innovative design tool, a System that
tion or specific entity meaning. adds a Reasoning and Performance Layer to existing
Object-oriented Systems and actual BIM plat- BIM and CAD software.
forms allowed another step to be taken towards
Design and Process Support Tools: each represented BUILDING DESIGN SUPPORT TOOL
entity has a recognizable tag linking it to a general Existing BIM software systems are evolving into
concept and a set of properties that contribute to Collaborative BIM environments: different domains
meaningfully defining the designed building com- are combined into a single software or several con-

nected software families; user interfaces are chang- PROPOSED SYSTEM STRUCTURE: A
ing into domain-oriented GUIs that adapt to suit the MIXED MODEL
target user. Referring to previous research results (Fioravanti
The actual design activity is increasingly being et al., 2011a; 2011b) the proposed Design Support
extended into all sectors and every day it is more Tool combines existing software tools with a system
and more difficult to try and foresee the often un- composed of developed routines and Knowledge
predictable changes resulting from new inventions Structures. The research analyzed and tested several
and changes in technology, tools, methods and so- software(s), although the proposed system uses the
cial customs using existing design support tools. following components:
In addition, client needs, requirements and con- • Autodesk Revit 2013: BIM software for Private
straints are becoming more specific day by day (for and Shared Design Workspace Interface;
instance, real estate societies) and designers have to • Protégé OWL 3.4.8 (Frames): Ontology Editor
continuously check out their own design solutions for Knowledge Modelling in terms of Hierarchy,
in order to fulfill certain domain expectations. Properties and Entity Relationships;
Each Specialist Designer with her/his own ex- • Semantic Web Rules Language Plug-in: Protégé
pertise uses ‘personal reasoning rules’ in order to plug-in for Rule Modelling on Ontology Entity
develop her/his own design solution. Representation based on predicate logic for-
In order to verify specialist domain constraints and malization;
general overall design consistency, coherence and • Jess Rules: an ontology reasoner for rule check-
congruence, “on-the-fly” performance verification ing and verification [1];
systems are needed. • Revit DB: a tool for Revit projects exported into
The proposed Layer is complementary to ex- a Database;
isting BIM and CAD software in order to support • Oracle MySQL: a relational database manage-
on-the-fly designers, allowing them to model their ment system.
respective constraints, verification algorithms and In order to validate the proposed research, au-
checking rules at different levels (Beetz et al. 2006; thors developed ad hoc plug-ins, software add-ons
Fioravanti et al., 2012): and tools to connect Knowledge Structures imple-
• Internal Domain Private Rule Verification: Ac- mented in OWL with Autodesk commercial software.
tors model their own rules in order to check the This link with CAD software expresses proposed
on-going design solution in their own personal platform potentials and also opens up fresh discus-
specialist domain; sion on future developments:
• Internal Domain Shared Rule Verification: Ac- • Link with existing commercial software with
tors check rules shared by other actors involved its limits and constraints due to its proprietary
their own private design solution in order to nature;
check for possible conflicts with a different spe- • Enterprise system development including “ad-
cialist domains; hoc” and/or open-source graphical representa-
• Collaborative Rule Verification: in the Shared tion systems.
Design Workspace (Loffreda and Fioravanti, Reasoning results could affect the geometri-
2009) performance/verification and checking cal aspects of the modeled entities, but due to the
rules shared by all the actors involved in the “proprietary” nature of Autodesk Revit 2013, it is not
specific design process are performed in order possible to interact with geometrical properties on
to evaluate and check out the proposed design built-in Families from external software, not even
solution (with input needs and requirements). from the Database via the Revit DB Link.

Graphical Entities Design Entitiy Class Definition Design Solution with
Univocal ID assignment
and standard property (for each designed assigned Entity Class
[Interface custom made
Figure 1
definition entity) (Meaning) and
Database] Design process Flowchart.
[Revit 2013] univocal ID
Database Translation
Revit Database Import BIM model export to Custom-made
into Protégé
into Protégé Ontology BIM Database Interface Linking
compatible Database
Editor [Revit DB] Database
format
Coherence, Congruence,
Reasoning over Reasoning inferred
Compilation of Property Consistency and
Building Knowledge axioms
defined fields Performance checked
Ontology (Reasoning results)
Ontology
Revit Model update with

Protégé Ontology export to
blinking violated Revit DB update with Entity DB link among
Database
constraints and missing Reasoning Results Protégé and Revit DB
[Protégé DB export Plug-in]
performances
PROPOSED SYSTEM IMPLEMENTATION Phase 2: Building Design Process workflow

WITH BIM The proposed prototype concept has been con-
ceived of as a sequence of necessary steps (Figure 1)
Phase 1: Knowledge and Rule Modeling that are transparent to final users (designers).
In order to represent Building Knowledge Entities, 1. Actors develop their own design solution by
a specific Design Ontology has been developed. It means of Autodesk Revit 2013;
has been structured with reference to a Meanings- 2. Specific Revit shared parameters have been de-
Properties-Rules Template devised by the authors fined in order to specify Ontology Class and IDs
and implemented by means of a Protégé Ontology for each designed Revit entity (Figure 2);
Editor. 3. BIM Design Solutions include only BIM entities
Entities description, properties, relationships and properties; that implies: no space semantic
and hierarchical structure have been modeled by definitions, no specialist domain properties, no
means of predicate logics (Hofstadter, 1979) and on- rules;
tology formalization [2]; the Knowledge Represen- 4. BIM model is exported to a Database by means
tation allows queries and constraint verification by of Revit DBLink;
means of specific reasoner and rules formalization in 5. An open-source database MySQL is created in
Semantic Web Rules Language (SWRL). order to interrogate the exported database;
In order to interrogate Design Solutions, Ontol- 6. An ad hocLinking Database has been created
ogy Rules have been implemented and tested on in order to connect exported Revit DB (Revit
prototype instances of developed Ontology Classes: data) and Protégé Ontology Instances (Knowl-
a hospital ward has been modeled both in terms of edge Entities);
general entity Classes and testing instances (spaces 7. Respecting Protégé Database exporting for-
and components) (Calvanese et al., 2008); moreover mat, Knowledge entities are instantiated and
SWRL rules have been formalized to check specialist property fields filled in with available values
domain constraints: from Revit designed entities;
• Space configuration and topological relation- 8. The DB obtained represents a combination of
ships among spaces; Knowledge and Graphical Entities in an Ontol-
• Furniture and equipment provided for each ogy query-able format;
building unit; 9. By means of modelled SWRL rules, constraints,
• MEP system, Structural elements and Space performance, consistency, coherence and con-
configuration compatibility. gruence verification can be performed;

Figure 2
Autodesk Revit Shared Parameter Protégé Building
Autodesk Revit 2013 and
Definiition for Ontology Mapping Ontology definition
Protégé Frames 3.4.8 customi-
zation.
Entity Class Building Design Building Design Reasoning

Entity ID
Name Classes Properties SWRL Rules
10. SWRL inferred axioms can be checked and veri- The CAD prototype refers to AutoCAD® for
fied by each Specialist Designer; graphical representation and is limited to lines and
11. By means of Protégé “Export to Database” com- 8-vertex polylines representation to suit the mod-
mand, an inferred Entity Database is created; eled knowledge structure.
12. The Revit Database is then updated with new
values and definition from the inferred Entity Phase 2: Building Design Process workflow
Protégé Database by means of the developed The following workflow shows the step-by-step im-
Linking Database. plemented prototype:
Due to the proprietary nature of Autodesk Revit, 1. Launch Protégé Ontology Editor with classes,
even if Reasoning Rules in Protégé may possibly af- properties and rules definition;
fect geometrical properties modifying and/or add- 2. Query Tab launch: classes list (Figure 3);
ing values, Revit does not allow them to be changed 3. Class List Export in a TXT file;
because they are System Parameters and it is not 4. Autodesk AutoCad Launch;
possible to edit them out of Revit itself. 5. AutoLisp implemented application launch
Due to this limitation, in order to validate the for automatic Layer creation with layer name
proposed system, the authors implemented plug- equivalent to class name (Figure 4);
ins and add-ons for AutoCAD® which allows interac- 6. Design solution representation by means of 2D
tion with the DXF drawing format. lines and/or (at most) 8 vertex polylines;
7. Design solution saved as DXF format file;
PROPOSED SYSTEM IMPLEMENTATION 8. A specific software has been implemented in
WITH CAD order to parse the saved DXF file and then to
create as many CSV files as the layers used.
Phase 1: Knowledge and Rules Modeling Each CSV file will contain as many rows as el-
Knowledge Ontology modeled by means of Protégé ements are present in corresponding layer in
for the above-mentioned test has been modified in DXF file, separating the element features with
order to allow further prototype tests on the pro- a semicolon, for example sake:
posed system. • Instance type: Line or LwPolyline;
AreaXY property has been linked to Product • Handle: unique AutoCAD® ID;
class and its sub-classes and several has_xn (with n • numVertex: number of instance vertex
from 1 to 8) have been linked to classes in order to (only for polyline definition)
specify 2D geometrical instance definition. • has_xn-has_yn: (with n from 1 to 8) x and

Figure 3
Protégé OWL 3.4.8 Ontology
Editor. Query Tab Class List
result.
y vertex coordinates; means of SWRL in order to validate proposed

9. In order to facilitate DB content management, system potentials. The testing rule checks all
the CSV files obtained are merged into a single the Room instances and verifies whether the
Microsoft Access® DB, importing each CSV into bounding windows area is greater than room
a different table of the database; area/8 (Figure 5);
10. ODBC DNS system link with the created DB; 14. Reasoning rules are applied by means of the
11. Protégé and DataMaster v.1.3.2 Launch: allows Jess Rules [1] reasoner and, according to rules
the linked DB to be connected and the existing definition, unverified instances will have the
tables can then be imported into the existing Boolean modified property set to true.
ontology; 15. Query Tab launch: at this stage it is possible to
12. By setting Datamaster import under Thing, search for all instances with modified Boolean
the system will automatically create as many property set to true;
instances as there are rows on each table as 16. It is then possible to export the modified In-
a subclass of related Class with name equal stances List as a txt file with handle property
to the table name. As a result of the previous associated values;
steps, the table name will correspond to the 17. By means of some other developed software,
ontology class name so that instances will in- the system will check the previous created
herit knowledge properties and rules definition DXF file, compare it with exported txt file and
including also geometrical values obtained by modify the entities colors to Red if the handle
AutoCAD® representation; in dxf is present also in the exported txt modi-
13. A testing Design Rule was implemented by fied entities list.

Figure 4
Autodesk Visual LISP Editor.
Autodesk AutoCad Layer crea-
tion program.
CONCLUSIONS Due to the proprietary nature of Revit®, first

The system prototype illustrates an innovative ap- prototype implementation did not allow the Revit
proach to Building Design Support Tools by means designed entity to be modified even though the
of a mixed model using commercial application pro- authors used specific Revit add-ons and extensions.
grams, ontology management systems and custom- According to these limitations and in order to vali-
made reasoning rules, database and interface tools. date theories and design process logic, the authors
A set of existing BIM software, Knowledge Rep- developed a second prototype based on CAD soft-
resentation systems and Database improves existing ware.
commercial software, enhancing definition and the This approach allowed a different design pro-
modeling of building design. cess workflow definition; plug-ins identification and
In order to develop an innovative, powerful, scal- specific program implementation help check con-
able and useful design support system, the authors straints and design reasoning rule results affecting
implemented an ad-hoc Linking Database interfac- 2D CAD developed design solutions and demon-
ing previous modeled Design Knowledge Structures strating the overall system potentials.
(ontology classes and properties) (Fioravanti et al., Tests showed that it is possible to enhance ex-
2011) with Revit Entity Database. Afterwards SWRL isting commercial software by applying on top of
Rules can allow the combined (Knowledge and them a Reasoning Layer which includes Specialist
Graphic) Database Ontology to be queried in order and/or Common Rules, expertise and Building Per-
to perform consistency, coherence, congruence and formance verification.
performance verification on design solutions. Each involved Actor will then be able to model

Figure 5
Protégé OWL 3.4.8 Ontology
Editor. SWRL Testing Rule
definition.
as many rules as needed in the specific “on-the-fly” ACKNOWLEDGEMENTS

design process checking: The presented work is a result of an Italy-Israel
• Personal Specialist Domain consistency and in- joined research program “BKM - Building Knowledge
ternal coherence; Model: structuring building knowledge for a new gen-
• Other Domain Rules congruence; eration of design tools” (2011-2013). The Italian side
• Overall Design Verification and Client Needs has been funded by MAE (Ministry of Foreign Office).
fulfillment.
This system represents an on-the-fly tool for REFERENCES
Specialist Designers designed to suit Client needs Beetz, J, Leeuwen, JP and van de Vries, B 2006, ‘Towards a
and to correct the on-going design process accord- topological reasoning service for IFC-based building
ing to performance and requirements goals. information models in a semantic web context’, in Pro-
The implementation of the system shows its po- ceedings of Joint International Conference on Computing
tentiality in proposing a new generation of Design and Decision Making in Civil and Building Engineering,
Tools that allows further research development and Montreal, Canada.
deepening. Calvanese, D, De Giacomo, G and Lenzerini, M 2008, ‘Con-
As far as its scalability is concerned, the pro- junctive query containment and answering under de-
posed tool is easily applicable to other Knowledge scription logic constraints’, ACM Transactions on Com-
“Realms” aimed at improving different Design and putational Logic.
Collaborative Processes in order to enhance knowl- Einstein, A 2006, Come io vedo il mondo. La teoria della
edge sharing, innovation spreading and collabora- relatività, Giachini, Bologna, (Italian translation of Die
tive problem solving. Grundlagen der allgemeinen Relativitatstheorie, Hamil-

tonisches Prinzip und allgemeine Relativitatstheorie and sign Process’, International Journal of Design Sciences &
Kosmologische Betrachtungen zur allgemeinen Relativi- Technology, 18(1), pp. 1-18.
tatstheorie, Newton Compton eds., 1976). Hofstadter, DR 1984, Gödel, Escher, Bach: un’Eterna ghirlanda
Fioravanti, A, Loffreda, G, Simeone, D and Trento A 2012, brillante, Adelphi Edizioni (Italian translation of Gödel,
‘”Divide et Impera” to dramatically and consciously Escher, Bach: an Eternal Golden Braid, Basic Books Inc.,
simplify design. The mental/instance path - How rea- NY, 1979).
soning among spaces, components and goals’. Digital Loffreda, G, Fioravanti, A 2009, ‘Formalizing and computing
Physicality. vol. 1, Proceedings of eCAADe Conference, Ontologies to Speed Up the Construction of Knowl-
Opus V.D.I., Prague, pp. 269-278. edge-based Collaborative Systems. Three different ap-
Fioravanti, A, Loffreda, G, Trento, A 2011a, ‘Computing On- proaches’, Computation: the New Real of Architectural
tologies to Support AEC Collaborative Design: Towards Design, eCAADe Conference Proceedings, Cenkler Print-
a Building Organism delicate concept’, Respecting Frag- ing, Istanbul, pp. 341-348.
ile Places, Proceedings of the eCAADe Conference, Lju-
bljana, Slovenia, pp.177-186. [1] www.jessrules.com/ [Last accessed on January 29th
Fioravanti A, Loffreda G, Trento A 2011b, ‘An innovative 2013]
comprehensive knowledge model of Architectural De- [2] http://protege.stanford.edu/ [Last accessed on February
7th 2013]

Experiencing BIM Collaboration in Education
Stefan Boeykens1, Pauline De Somer2, Ralf Klein3, Rik Saey4
1
KU Leuven, Department of Architecture, 2,3KU Leuven, Department of Civil Engineering,
3,4
KU Leuven, Faculty of Engineering Technology (Belgium)
1
http://www.asro.kuleuven.be, 2,3http://www.bwk.kuleuven.be, 3,4http://www.iiw.kuleuven.be
1
stefan.boeykens@asro.kuleuven.be, 2pauline.desomer@bwk.kuleuven.be, 3ralf.klein@
bwk.kuleuven.be, 3ralf.klein@kahosl.be, 4rik.saey@kahosl.be
Abstract. In a context of a slow uptake of the Building Information Modeling (BIM)

methodology in the Flemish region, we present the results of an educational research
project, carried out over 4 semesters, in a multi-disciplinary, cross-campus collaboration.
This project fosters an improved application of BIM, information management and
communication, by organizing building teams involving students from different schools.
The project partners collaborated on a shared framework of supportive learning material,
collaboration scenarios between teams of students and the integration of digital tools for
communication, information management and collaboration in the curriculum.
This article, in particular, will elaborate on one of the collaborative exercises, involving
architecture and engineering students, using BIM for modeling, information exchange and
model evaluation.
Keywords. BIM; education; collaboration; interoperability; IFC.
INTRODUCTION AND CONTEXT

While the use of the Building Information Modeling ners. Local software vendors seem to claim no more
(BIM) methodology is increasing worldwide, we no- than 20 or 30% of their clients have licensed BIM
tice a slow uptake in the Flemish region (Belgium). software, while the effective use in projects is proba-
In contrast with surrounding countries, such as The bly even lower. While the BIM concept is slowly gain-
Netherlands or the UK, there is no governmental ing more attention, amongst professionals and edu-
incentive that promotes the use of BIM for building cators, it is sometimes regarded as merely a software
projects. We are far from the adoption rate as noted tool or an alternative method to make 3D models.
in the McGraw-Gill SmartMarket reports (Young Jr. et Too often, its usefulness is misunderstood, as it
al., 2009; Bernstein and Jones, 2012), which mostly is not seen as part of the design process. As such, the
reflect the North-American industry BIM uptake. added value and the potential productivity gains
Eastman et al. (2011, chap.4) describe the value of that are promised by BIM use are not experienced
BIM for building owners, but apart from a single ini- and might hinder BIM adoption. This is in line with
tiative, the demand for BIM from building owners in the experiences mentioned by Holzer (2012): the
our region is non-existing. Even architects, who are resistance to process changes and the “monodisci-
using BIM software more often, apply it only in their plinarity” of BIM tools. Hartmann and Fischer (2008)
own, local context and still rely on traditional, draw- note the lack of knowledgeable practitioners as one
ing based collaboration with other building part- of the major bottlenecks for widespread BIM adop-

tion. Lockley mentioned in the NBS Building Informa- pitfalls and how BIM could be applied in a design
tion Report (Anon, 2011, p.20+21) the important role studio context. He mentions an underestimating of
for educational institutes in the unavoidable transi- the way digital tools and methods change the “way
tion to BIM. Livingston (2008) argues that BIM mani- we make architecture”. By defining BIM as a design
fests itself both as a 3D-modeling tool (the software) methodology instead of a design tool, it “acts as pro-
and an information management system, which aids vocateur to design education and the how and what
students for “a greater understanding of building of the academic design studio”. Many designers see
systems integration”. BIM as a hindrance to the creativity during the de-
The authors of this article participate in the COM. sign process. Morton (2012) expresses this concern
BI project: “Communication and management of digi- in schools, where teachers often fail to see beyond
tal building information in a multidisciplinary team the tool and how BIM can be “interfaced” with the
during the construction process” [1]. This two-year design process. Even students who are being intro-
educational research project implements collabo- duced to BIM, often fail to look beyond the 3D mod-
ration in building teams and introduces innovative eling aspects as mentioned by Weber and Hedges
software tools. Since all building-related disciplines (2008). In that context, prior CAD knowledge proved
are represented within the KU Leuven Association [2], both beneficial and discouraging at the same time.
there is an opportunity to form multi-disciplinary Beneficial, since drawing functionality is still provid-
student teams and to share knowledge. Also the tu- ed to finish the drawings where the model is incom-
tors can share their different fields of research and plete or lacks detail; discouraging because drawing
professional experience. The didactical approaches a floor plan directly is usually quicker in 2D. “The
and methods have been elaborated in (Boeykens et simplest tasks proved to be a challenge.” Students
al., 2013), whereas this article will focus on how the and design studio teachers need to look beyond the
BIM methods and technologies are being applied in effort inherent with BIM.
an educational context. To avoid too much friction of introducing a BIM-
We deliberately want to foster the application of based working method and enforcing it in the de-
the BIM methodology in the architecture and engi- sign studio, the project collaboration was taken out
neering curriculum, by allowing students to expe- of the design studio assignment and hosted as part
rience the benefits of using BIM in a collaborative of a course where the authors were directly involved.
context, rather than as a mere modeling and docu- There is the Architectural Computing course on the
mentation tool, as it happened before. Students will side of the architectural-engineering Bachelor pro-
learn to work together, to exchange information gram, during the fourth semester and the course on
using BIM models and to apply BIM as part of the 3D-Design on the side of the construction engineer-
design process. We want them to migrate from “lit- ing students, which is an elective in the Master track.
tle” BIM to “BIG” BIM (Jernigan, 2008), which includes However, this directly implies a second challenge,
model exchange and model evaluation. as students with different maturity levels will need
to work together. They have different background
CHALLENGES knowledge, do not possess the same technical ex-
The first challenge is mostly a mentality problem. pertise and speak a partially different language.
As witnessed by the authors, the majority of archi- There are also multiple technical challenges. It
tectural design studio work still relies on traditional was decided for this exercise to enforce the usage
drawing based methods, where 2D CAD software, of BIM software and to apply it intensively as part
such as AutoCAD and VectorWorks, is used mostly of the collaboration. The different BIM applications
for representational purposes or construction docu- that are used in the two schools are inherently in-
mentation. Ambrose (2012) discusses some of the compatible, as they apply a proprietary data model

and file format. In the architecture course, Graphisoft METHODOLOGY AND ORGANIZATION
ArchiCAD was used, whereas the engineering course OF PROJECT TEAMS
relied on Autodesk Revit. It was a deliberate choice Since design collaboration is an important driver for
not to avoid this challenge, as it is a good illustration BIM adoption, we will focus on the collaborative as-
of a real context of various players and tools. Even pects of BIM, by simulating a Project Team, contain-
if everyone would use the same software, interop- ing different roles and disciplines, with students.
erability is not guaranteed. Version differences can Because it was not feasible to assign an external en-
make file exchange impossible: e.g. it is not possible gineer to each project team, we had to make the dis-
to export Revit models to older releases. In addition, tinction between internal, local collaboration (e.g.
even when people would use the same software architects from the same office) and collaboration
version, there are still many differences in model with an external office. The two types of collabora-
structure or applied model templates. tion provided a challenge of trying to balance the
The evident solution was to rely on ISO/ amount of work, as not to place an advantage on
PAS 16739 IFC (Industry Foundation Classes) [3] to one or the other.
exchange building models between ArchiCAD and Figure 1 gives a schematic overview of the col-
Revit. However, 100% complete information transfer laborations. We indicate different roles, the used
cannot reasonably be expected due to implementa- software systems and document types and the in-
tion limitations. Rather than to complain about this, formation flow.
we decided that this represents a realistic context As can be seen on the previous diagram, in the
for collaboration, so the challenge was included as collaboration with an external engineer, the archi-
part of the exercise. This was motivated by the ob- tectural team collaborates internally on the design
tained successes in several case studies and reports model (e.g. using ArchiCAD TeamWork [4]), whereas
(Bos, 2012; Mitchell et al., 2007; Hitchcock and Wong, the engineering student fulfilled a consultant role,
2011), which testified of the usability of IFC and how focusing on heat-loss calculation and designing
it can be implemented in construction projects, de- the ventilation system. In the internal collaboration,
spite most authors acknowledging that the informa- there is an additional role indicated, namely that of
tion transfer was seldom perfect. As long as project the model evaluator. It was decided to have the pro-
team members have means to assess the limitations ject team assign one group member with this role,
and are willing to communicate constructively, this as an equivalent consultant role. These students per-
should not prevent collaboration. Moreover, deeper form some qualitative and quantitative analyses of
understanding of the usage of IFC in practical sce- the design model, using model evaluation software
narios can help software developers to improve (Solibri Model Checker [5]). This necessitates the us-
their BIM implementations. age of IFC as an exchange format.
In addition, there were several logistic chal-
lenges. Students had different timetables and were PROCESS AND WORKFLOW ORGANIZA-
located in different cities, about 90 km apart. Moreo- TION
ver, the architectural students were more than 100, The design team prepares an architectural model
whereas the engineering students are part of an that is transferred to the other disciplines, to provide
elective course, for which only six students regis- feedback on how to improve building performance.
tered this semester. Only some of the groups with To support this process, students are provided with
architecture students could be linked to an engi- basic guidelines and learning material, but they
neering student. The assignments for all groups, have to organize the actual collaboration indepen-
however, had to be similar and present equal work. dently, by specifying tasks and appointing responsi-

Figure 1
Schematic overview of the
collaboration.
bilities. The guidelines document is derived from a tion between design team and external consultant,
regional, localized IPDP (Integrated Project Delivery the engineering students set up ER, including room
Protocol) document [6], in which the students need areas, volumes or the façade area. This enforces
to formulate the different roles in their team, respon- them to specify required information precisely and
sibilities, the information flow and the necessary col- unambiguously, including the correct way to meas-
laboration activities (deadlines and meetings). We ure certain quantities, e.g. gross or net values and
reformulated this document, which was oriented to inner or outer dimensions. Based on the ER, the ar-
professional practices, for an educational context, chitectural model is filtered to contain only the re-
but kept a similar objective. Instead of formulating quested information, rather than transferring a full
the document as a static checklist, it was written as model. This is in line with the MVD concept. Before
a series of open questions, for which the students the filtered model is exported to IFC, it can be fur-
needed to define an answer within their group. ther optimized to improve model exchange, as will
buildingSMART [7], an organization focusing be described in the next section.
on interoperability and building process improve- To accompany the exported model with quan-
ments, defines Exchange Requirements (ER) and Mod- titative results, the architects create additional table
el View Definitions (MVD). To prepare the collabora- schedules. Engineering students who receive the

IFC model follow a reference model-based approach, students of construction management, which was
indicating that they will use the exported models for also undertaken during the COM.BI project.
information exchange only and continue working in
their native BIM environment. This is in line with cur- Technical description of the collaboration
rent best practices for BIM application (Bos, 2012). The collaboration in the architectural team used the
Reference models help to clarify who is responsible ArchiCAD TeamWork functionality, which allowed all
for which part of the model. The imported model, af- group members to work synchronously on the same
ter being further optimized, is then linked to a sepa- model hosted inside Graphisoft BIM Server [4].
rate project file for the modeling of an HVAC instal- To be able to access the Solibri License Server and
lation, using a series of custom Revit Family objects, to sign in to the BIM Server from outside the uni-
kindly provided by a regional engineering office. versity computer room, a Virtual Private Networking
Ideally, the HVAC model is assembled again with the (VPN) connection is required. While this is provided
design model of the architects, but this was left as by the university ICT services, it posed certain re-
optional. strictions on the computer systems. The connec-
For the local collaborations, the exported IFC tion would only succeed if the operating system on
model is used to create a qualitative and quantita- the student laptops were up to date and equipped
tive assessment of the model, which is returned to with a correctly configured virus scanner and Java-
inform the design team about possible model im- runtime. Unfortunately, some students were not
provements. This includes clash detections (element able to establish the necessary VPN connection, due
overlaps), verification of model completeness, pro- to technical problems. As a backup plan, the main
ject validity and checking against a prepared build- computer classroom was fully configured and some
ing regulations rule set. teams needed to rely on this.
In both types of collaboration, we completed a We encountered limitations when trying to keep
full feedback-loop, in a model-based workflow. While the exported IFC model fully usable in the receiving
such a workflow might be a bit simplified, compared BIM software. Since we cannot expect students to
to expected future professional exchange, students be aware of all peculiarities of particular IFC imple-
are learning several important characteristics of de- mentations in BIM software, possible problems are
sign collaboration: description of exchange require- documented and, where possible, solutions are giv-
ments, preparing models for a particular goal, cross- en that present better model exchange. The actual
checking the exported information and learning export requires additional configuration of the IFC
about means to improve the model preparation. By writer settings, with direct impact on the usability of
introducing students to such processes, they are be- the model. This is also documented for students, to
ing prepared for a future construction industry and provide them with a usable starting point. It is very
might be able to introduce this knowledge in the of- important that they experience the implication of
fices they will join in a few years. the used configuration. In a Coordination MVD, the
Experiences about the didactic and digital com- focus is on accurate geometry, enforcing the mod-
munication aspects of the collaborative work are el to be exported using Boundary Representation
elaborated by the authors in another conference pa- (BRep) geometry. This has the disadvantage that the
per (Boeykens et al., 2013). There we describe more model will be completely static, preventing it from
in detail the importance of the different modes of being further edited or corrected. Using extruded
communication, the different roles that members in geometry on the other hand generates better edit-
a project team take on and the experiences learned able objects, containing more IFC attribute values,
from a related collaborative assignment on “Zero Im- but at the cost of a loss of certain model details. To
pact Building” between students of architecture and assist this last step, students are obliged to perform

Problem Situation in Revit Recommended Action in ArchiCAD Table 1
Objects lack properties Enforce attribute export in IFC Translator Overview of ArchiCAD model
Walls become “in-place” families, Window/Door Simplify geometry, e.g. remove Solid Element adjustments to improve IFC
vertically offset from opening Operations, reveals, casing. import into Revit.
Orientation of walls, doors and windows not Add custom IFC tag or agree on ID naming
available. convention.
Spaces do not contain category Add category name manually as added attribute
using PSet_SpaceCommon
Walls do not connect to Sloped Roofs Replace with legacy “crop” operation and convert
multi-plane roofs to single-plane
Problem Situation in Revit Recommended Action in Revit Table 2

Door and Window voids do not protrude through Edit family object and extend void to wall façade. Overview of some Revit model
the wall volume. adjustments to re-create the
Rooms get wrong height (= default value from List ArchiCAD Zone Height alongside Revit Room ArchiCAD design model.
template file). height and manually correct.
Identical windows and doors become separate Select and replace with appropriate family
Family objects. object (if desired).
Door & Window height attribute not set Check extended IFC attributes for exported value
from ArchiCAD
a visual check using an IFC Model viewer. low the provided lessons at their own pace. While
During the preparation of the learning material, the recording of this material was a serious effort,
a series of recommendations and known problems student reactions were mostly positive. They ap-
were documented, giving the students pointers to preciated the freedom to watch the material, rewind
problems they might encounter. At first, before the where necessary or even skip parts that seem un-
IFC model is exported, the design model in Archi- needed to them. As the computer skills vary con-
CAD can be altered to generate more usable output, siderably between students, this was a valuable
as summarized in Table 1. alternative to classroom-based tuition, which has
Some aspects were not solvable within the de- become more difficult over the last few years. After
sign model, so in those cases, manual changes in Re- students were given time to assimilate the tutorials,
vit are required, as described in Table 2. guided consulting sessions were organized, where
Even then, certain information was inconsistent students were able to pose individual questions,
and some workarounds were still required. In those based on their own work. This was accompanied
cases, students were free to use other means to with an online forum for questions. An added ad-
communicate the missing information. vantage of video-tutorials is that they can be easily
shared. They not only serve our own classes, but also
Didactical learning material and facilities attract many other users worldwide, since they were
As part of the didactical vision on the Architectural openly hosted as YouTube playlists [8].
Computing course, which also comprised other
themes (visualization, freeform modeling, digital Is it worth the effort?
documentation), it was decided that as much classes While some of these aspects might seem rather
as possible would be replaced with video-tutorials. trivial, they do pose a serious technical challenge to
After a general introduction seminar, students fol- properly set up and we have been lucky to have full

collaboration of the system administrators through- facilities is essential.
out the whole project. It also required a serious ad- Some people would argue that learning a specif-
ditional management effort of the teaching staff to ic BIM software tool is not the task of the university,
e.g. set up user accounts on the TeamWork server but we have the experience that it is very valuable
and to assist with the ICT aspects of the exercise. to at least learn the proper usage of one BIM soft-
It should also be noted that, even with more than ware application in a practical context, which can be
100 users, the system appeared quite stable and transferred to other software later on, if required.
reliable and only some minor interventions were
required. We are aware that such additional efforts EVALUATION AND FEEDBACK
and support are more difficult to impose on design Students are asked at several points in the collabo-
studio teachers, certainly if they are not accustomed ration exercise to reflect on the process and the
to BIM software and all required network functional- applied techniques, using questionnaires and a log-
ity. book. In addition, their own role in the simulated de-
We are convinced about the value of BIM in sign team is also evaluated.
construction and the necessity of learning BIM as The project results in two types of guiding docu-
a methodology in education. However, to appreci- ments. The process and project based documents
ate these values, students should not only learn the are oriented to be used by students. The process-
theoretical basis. They also need to experience the based documents are formulated as open questions
BIM methodology in a realistic collaboration scenar- instead of a fixed checklist, as to foster reflection
io and learn to use the software tools in an efficient and discussion among team members. For a more
way. detailed description of the different guiding docu-
As part of the first BIM assignment of the archi- ments that were set up, we refer back to (Boeykens
tecture students, the focus was on understanding et al., 2013). In addition, we summarize the didactic
the concept and applying it in a practical context. guidelines towards educators. They are categorized
Most of the students testified that they understood into BIM, construction-related and domain-neutral
the idea of BIM, but not all of them were convinced aspects, e.g. organizing student teams.
about the effectiveness. Many students still fall back Student feedback is also necessary to gradually
on their approach of using AutoCAD for 2D drafting improve the collaborations. As it is the intention that
and SketchUp for quick 3D modeling. The underlying the collaborations continue after the COM.BI pro-
theme was that BIM was often regarded as a soft- ject-funding timeframe, we take their remarks and
ware tool for 3D modeling instead of a collaboration considerations seriously. We list some comments in
method through the building process. They specifi- Table 3 and how they are improved.
cally mentioned limitations in modeling freedom, Overall, students noted that the collaboration
due to the indirect nature of describing geometry exercise did not take an enormous amount of time
with parametric entities in BIM software. However, and many groups finished their assignment before
teachers should also not overestimate student ma- the deadline. The main intention of the exercise – a
turity and insight into the design process, especially first-hand experience of how BIM could be applied
at the Bachelor level. Instead of enforcing the use of as part of the design process – was clearly specified,
BIM in the design studio, for which there is no gen- avoiding students to spent excessive time on solv-
eral consensus, at least in our schools, our processing detail-problems.
based introduction will give students a more hands- It should be noted that the building model was
on insight in how the BIM methodology could not a new design, but rather a retake of one of their
work, with obtainable objectives. The support with earlier design studio projects, to avoid the overload
adequate learning material and sufficient technical of the full design process, with conceptual devel-

Student comment Afterthought Table 3
Assignments need to be clear and Refine the assignment, limit requested documents/models/ Comments and recommenda-
compact reports, be specific tions from student feedback.
Improve team coherence Initial group meeting, esp. with external engineering students.
Have them clarify their tasks to each other.
What should the model checker role Add explanations about the rulesets.
do?
Graphics in model checker look Clarify software limitations. Model checker is not a visualisation
weird environment. Focus on relevance in process.
Model geometry is not exact in Clarify that the purpose of a viewer/checker is different from a
model viewer/checker documentation tool.
opment and extensive teacher consults. It was ex- (model checker). The formulation of the assign-
plained that the architectural quality of the design ments and the final expectations need further re-
would not be part of the grading criteria, because finement and valuable lessons were learned to im-
this was already done as part of the original design prove the assignment for the next semester.
studio exercise. The quality of the model however, Several adjustments to the curriculum are initi-
taking the received feedback into account, and the ated, to ensure a durable implementation of the
reporting were effectively graded. collaboration beyond the period covered by project
funding. All involved partners are enthusiastic to
CONCLUSIONS AND FUTURE OUTLOOK continue further, precisely as we seemingly are, in
All project partners are convinced that this is most our Flemish region, at a turning point, where several
valuable exercise, providing a huge experience for professional, academic and commercial parties are
both students and educators. This approach illus- increasingly moving towards BIM, which will form
trates BIM as a process, rather than as a tool. Stu- the basis for a regional knowledge network.
dents are stimulated to reflect on both the product
(the design) and the process (the design collabora- ACKNOWLEDGMENTS
tion in a Project Team). However, do not underesti- The project funding from the “Education Develop-
mate the need for continuous evaluation and the ment Fund” of the Association KU Leuven, with refer-
increased technical complexity to facilitate the col- ence OOF 2011/24 is gratefully acknowledged. The
laboration. Luckily, the software tools, while not project also builds on experience gathered during
perfect, proved to be adequate and fairly stable. IFC a previous related project from the same funding
support, while not perfect, due to implementation body, with reference OOF 2007/24, focusing on mul-
differences between software vendors, is gradually ti-disciplinary collaboration in building teams.
improving and is, indeed, “good enough” to be used
at the core of the collaboration process. REFERENCES
It is very important to be clear about the ex- Ambrose, MA 2012, ‘Agent Provocateur – BIM In The Aca-
pectations towards students, as the exercise should demic Design Studio.’ International Journal of Architec-
present them a positive experience with BIM and tural Computing. 10 (01), pp.53–66.
not blur this with an extensive assignment. The ex- Anon 2011, NBS Building Information Modelling Report
ercise attempts to capitalize on the virtues of BIM: March 2011
synchronization of representations (all documents), Bernstein, HM and Jones, SA (eds) 2012, SmartMarket
extraction of information (schedules), model-based Report: The Business Value of BIM in North America.
information exchange (IFC) and model evaluation McGraw-Gill Construction, New York, USA

Boeykens, S, De Somer, P, Klein, R and Saelens, D 2013, ‘Digi- Ettouney (ed) AEI 2008. American Society of Civil Engi-
tal communication in multi-disciplinary teams. Prepar- neers, Reston, VA. pp.1–9.
ing students for a future building project context.’ in Mitchell, J, Wong, J and Plume, JIM 2007, ‘Design collabo-
Educating the Future: Architectural Education in Interna- ration using IFC - A case study of thermal analysis.’ in
tional Perspective. EAAE, Istanbul (Turkey). pp.1–13. A Dong, Andrew Vande Moere, and John S Gero (eds)
Bos, P 2012, ‘Collaborative engineering with IFC : new in- CAAD Futures. Springer. pp.317–329.
sights and technology.’ in Gudnason and Schrerer Morton, DE 2012, ‘Bim: A Transformative Technology within
(eds) eWork and eBusiness in Architecture, Engineering the Architectural Curriculum in Schools of Architecture
and Construction, 9th ECPPM Conference Proceedings. (Pedagogic Stages of Architectural Education and the
Taylor & Francis Group, London, Reykjavik (Iceland). Transformative Effect of BIM).’ International Journal of
pp.811–818. 3-D Information Modeling. 1 (4), pp.50–68.
Eastman, C, Teichholz, P, Sacks, R and Liston, K 2011, BIM Weber, D and Hedges, KE 2008, ‘From CAD to BIM: The Engi-
Handbook - A Guide to Building Information Modeling for neering Student Perspective.’ in Mohammed Ettouney
Owners, Managers, Designers, Engineers, and Contrac- (ed) AEI 2008. American Society of Civil Engineers, Res-
tors (2nd edition). John Wiley and Sons ton, VA. pp.1–8.
Hartmann, T and Fischer, M 2008, Applications of BIM and Young Jr., NW, Jones, SA, Bernstein, HM and Gudgel, JE eds.
Hurdles for Widespread Adoption of BIM - 2007 AISC- 2009, SmartMarket Report: The business value of BIM:
ACCL eConstruction Roundtable. Center for Integrated Getting Building Information Modeling to the Bottom
Facility Engineering, Stanford University, New Orleans, Line. McGraw-Gill Construction, New York, USA
LA (USA)
Hitchcock, RJ and Wong, J 2011, ‘Transforming IFC Archi- [1] http://associatie.kuleuven.be/onderwijs/oof-in-
tectural View BIMS for Energy Simulation: 2011.’ in Pro- fo/2011/24
ceedings of Building Simulation 2011: 12th Conference of [2] http://associatie.kuleuven.be
International Building Performance Simulation Associa- [3] http://www.buildingsmart.org/standards/ifc
tion. Sidney (Australia). pp.1089–1095. [4] http://www.graphisoft.com/bim_server
Holzer, D 2012, ‘BIM ’ s Seven Deadly Sins.’ International Jour- [5] http://www.solibri.com
nal of Architectural Computing. 09 (04), pp.463–480. [6] http://kubusinfo.nl/OpenBIM-solutions/OpenBIM/IPDP-
Jernigan, FE 2008, BIG BIM, Little BIM (2nd edition). 4Site protocol
Press [7] http://www.buildingsmart.org
Livingston, C 2008, ‘From CAD to BIM: Constructing Op- [8] http://www.youtube.com/user/stefkeB/
portunities in Architectural Education.’ in Mohammed videos?view=1&tag_id=&sort=dd

Landscape Information Modeling
Plants as the components for information modelling

Veronika Zajíčková1, Henri Achten2
Czech Technical University in Prague, Czech Republic
1
http://cz.linkedin.com/pub/veronika-zajíčková/67/741/b20, 2http://www.molab.eu
1
verca.zajickova@centrum.cz, 2achten@fa.cvut.cz
Abstract. In this paper we report on a recently started PhD project in which we

investigate the extension of the concept of “Building Information Model” (BIM) to the
domain of landscape design. The potential benefits of BIM in the field of architecture have
been reported many times (e.g., Ibrahim et al., 2004; Eastman et al., 2008; Abdelmohsen
et al., 2011). However, in landscape design an information model in the way of BIM seems
to be missing. Benefits of a Landscape Information Model would be (a) formalisation of
knowledge in landscape design; (b) information model to support multiple participants
in landscape design; (c) improved information exchange between landscape design,
architecture, and urban design. In this paper we set out the basic outline of the research.
Keywords. BIM; landscape design; LIM.
LANDSCAPE INFORMATION MODEL

In this paper we report on a recently started PhD • Integration of landscape design with urban
project in which we investigate the extension of the and building designs.
concept of “Building Information Model” (BIM) to the In our research project, creation of an information
domain of landscape design. The potential benefits model for landscape design proceeds in two phases:
of BIM in the field of architecture have been report- first, the definition of an ontology of relevant con-
ed many times (e.g., Ibrahim et al., 2004; Eastman et cepts and objects for landscape; and second, the
al., 2008; Abdelmohsen et al., 2011). As a technology, formalisation of this ontology into an information
various instances of BIM can be found throughout model. This process allows knowledge capture and
many applications in the Building & Construction representation which right now is not available.
Industry. To the best of our knowledge, there is no Similarly as in architectural and urban design,
similar application of the BIM principle for landscape in landscape design multiple parties are involved
design. In our view, there are three major benefits in (landscape designers, municipality, urban planners,
the application of BIM principles to landscape de- and so on). Reliable exchange of information during
sign: the design, realisation, and management phases will
• Provision of an information model for the do- increase effectiveness and reduce errors and the risk
main of landscape design. of (costly) corrections.
• An information model that would support multi- Finally, when an information model for land-
ple participants in the landscape design process. scape would be available, it would allow easier

integration of landscape design with urban and and threatened species.
architectural design through exchange of infor- • Management of water regime.
mation models. This is important to note, because • Hygiene environment (in particular attention
landscape design is naturally involved in urban to air pollution, noise and dust fallout) - trans-
structures: parks, green areas, public spaces, and port solution.
the immediate surroundings of buildings are often • Care for uniqueness, aesthetic and cultural val-
subject of landscape design. Therefore a more flu- ues of landscape identity.
ent link through a Landscape Information Model
between architecture, urban design, and landscape The functions of greenery
design would be very beneficial. The “soft” materials (greenery) have a distinct role in
the design of landscapes (Sklenička, 2003):
CHARACTERISTICS OF LANDSCAPE • Environmental (system of wildlife corridors,
DESIGN insulation - emissions and partly noise, regula-
In our view, a landscape information model consists tion of temperature and air humidity, air flow
of two major components: first, the information and - acceleration due to temperature differences
knowledge about sites (terrain, ground conditions, between greenery and buildings, sanitation -
weather, micro- and macro-climate, and so forth), clean air, produce oxygen, ...)
and second the information and knowledge about • Recreational, aesthetic, social.
landscape objects (“soft” materials such as vegeta- • Water managing.
tion, and “hard” materials of built objects). • Soil-protecting, meliorating.
Landscape design is a multi-disciplinary field in • Architectural -creation of space - part of the
which areas such as ecology, geography, geology, urban image (counterweight to artificial urban-
horticulture, and botany are used. This means that ized environment).
knowledge from quite a lot of areas has to be cap- • Economic - production.
tured in an information model. Because of the bio- • Microclimatic.
logical nature of landscape design, it is inherently • Fire-preventing.
dynamic in time: changes in the terrain, weather, We can distinguish between greenery in main func-
and most of all the “soft” materials have to be taken tion, and in additional function. In the main func-
into account. Therefore the design of a landscape tion, greenery is applied in parks, natural non-forest
does not just feature a “final” or “fixed” state, but areas, cemeteries, gardeners, green insulation, bo-
more resembles the design of a process in which the tanical gardens, and zoological gardens. In the ad-
individual development of components and interac- ditional function, greenery is applied in residential
tions between components have to be balanced. areas and private green spaces. Our major focus will
be on greenery in main function.
LANDSCAPE AND NATURAL ELEMENTS
IN URBAN CONCEPTS Compositional elements of garden and
Landscape design in urban structures serves the fol- landscape architecture
lowing purposes: The compositional elements of garden and land-
• Maintaining and enhancing ecological stability. scape architecture consist of technical elements
• Maintaining and enhancing ecological biodi- (“hard”) and vegetation elements (“soft”).
versity. The technical elements are:
• Protection of soil (environment for the realiza- • Terrain.
tion of natural processes). • Water (water features, irrigation, drainage).
• Protection of natural monuments and valuable • Solid surfaces, paving.

• Exterior furnishings. Vegetation is distributed according to their ability
The vegetation elements are: of growing in the similar climatic conditions. This is
• High vegetation - tree layer. called zoning.
• Medium vegetation - shrub layer.
• Low vegetation - ground cover plants, lawn, ASPECTS OF TREE VEGETATION CLAS-
herbs (perennials, annuals). SIFICATION
Vegetation is a live, volatile and organic component In the following section we provide an overview of
of the composition and it is dependent on the con- generally conceived major properties of tree veg-
ditions of surrounding environment. When design- etation that may form a starting point for a general
ing the garden/landscape we have to take into ac- information structure (Michálková, 2011) (Table 1).
count all the conditions that could possibly affect
the behaviour of the vegetation elements. The dynamic character of the vegetation
elements
Conditions for inclusion of vegetation ele- As we mentioned previously the vegetation is very
ments in the architectural design dynamic element of the design. We have to predict
Here we have to begin with the introduction and its changes during the whole life-cycle (size, shape)
acclimatization of species. In the region of central as well as its seasonal cycle.
Europe there are about 80 native species but more • Development/growth: Youth, Period of fertility,
than two thousand non-native species can actually Period of death.
grow here. • Reshaping: compact design changes in the last
When some foreign species are introduced we stage to the irregular shapes.
have to let them adapt to local conditions. Accord- • Seasonal cycle - behaviour, requirements, and
ing to E. Quitt (1971), our region has several climatic appearance throughout the aspects of spring,
regions: summer, autumn, winter.
1. VT - very warm, dry (average temperature
9-10°C, annual rainfall 500-600) OUR APPROACH
2. T1 - warm, dry (a.t. 8-9°C, annual rainfall <500) The strategy of the research project consists of two
3. T2 - warm, light dry (a.t. 8-9°C, annual rainfall parallel tracks: one track is breadth-first and top-
500-600) down, in which we aim to comprehensively describe
4. T3 - warm, light humid (a.t. 7-9°C, annual rain- the domain of landscape design; the second track is
fall 550-700) depth-first and bottom-up, in which we aim to deve-
5. MT1 - moderate warm, dry (a.t.7-8,5°C, annual lop specific information models for representative
rainfall 450-550) components of landscape.
6. MT2 - moderate warm, light dry (a.t.7-8°C, an- The top-down approach provides the general
nual rainfall 550-700) framework for the landscape information model,
7. MT3 - moderate warm, humid, lowlands (a.t.7- and ensures that we do not oversee relevant as-
8°C, annual rainfall 700-900) pects. The bottom-up approach ensures that we can
8. MT4 - moderate warm, humid (a.t. 6-7°C, an- define realistic and applicable information models.
nual rainfall 650-750)
9. MCh - moderate cold, humid (a.t. 5-6°C, annual BUILDING INFORMATION MODEL
rainfall 700-800) Although Building Information Modeling (BIM) has
10. Ch - cold, humid (a.t. below 5°C, annual rainfall its roots in the mid 1980s, its popularity within the
over 800) Architectural, Engineering and Construction indus-
tries has risen only recently- that is reason why in-

Composition type Table 1
• Basic, skeletal trees - long-living they stay on the spot throughout the parks lifetime - oak, linden, Aspects of tree vegetation
maple, elm. classification.
• Additional trees.
• Filling, temporary - fast-growing species.
• Under-layer (middle and lower floors).
• Ground covering species.
Visual characteristics
• Size.
• Habitus (shape variable within species).
• Structure.
• Texture - the period of dormancy.
• Color - foliage, flowers, fruits.
Size of tree
• Small 5-10 m fruit trees, ornamental cultivars.
• Medium 10 (15)-20M hornbeam, birch, rowan, birch, spruce, pine, fir.
• Large 20-30 (40) m to 40m: Buba, lime, ash, poplar, elm, maple.
Branching structure
• Branching dense, sparse.
• Direct branches, gnarled.
formation modeling is still being developed for the companies Graphisoft (ArchiCAD) and Autodesk
buildings mainly - the construction management (Revit) that bring their products to our market. Ad-
and facility operations. ditionally there are also Nemetschek (Allpan Archi-
That may explain that in our investigation we tecture) and Bentley (Microstation V8i).
did not find any indication of existence of a survey
or a research in our direction. Firstly we needed to LANDSCAPE MODELLING SOFTWARE
get a deeper knowledge of the term “BIM” and also We have found several smart “CAD based” programs,
the software that supports this system of work. but all of them are used for 2D design (the tradition-
What we found about BIM could be generalised al way). One case is for example ArborCAD, which is
to this: Building Information Modeling covers more special purpose CAD software for the needs of ar-
than geometry- it extends the traditional approach borists. It is based on the landscape software LAND-
to the building design (two-dimensional drawings WorksCAD. It has a lot of features specially aimed at
as plans, elevations, sections, etc.) beyond 3-D to information about plants, trees, and their properties.
time as the fourth and cost as the fifth dimension, Additionally there are also some applications
and sixth dimension is the life cycle management. It that work with plants as dynamic objects:
also covers spatial relationships, light analysis, geo- • NatFX is a plug-in (for Alias|Wavefront’s Maya)
graphic information, quantities and properties of designed for modeling and animating 3D
building components (for example manufacturers’ plants (age, season, and scale). Animation is
details). possible either as a realistic effect (with bo-
There are two major BIM software developing tanical constraints) or as a cartoon effect (by-

Table 1 continued Foliage texture
Aspects of tree vegetation • Texture fine (small leaves)
classification. • Coarse texture (large-leaved trees - maples, horse chestnut, sycamore)
• Airy texture (platane).
Growth and growing characteristics

• Age.
• Growth rate.
• Environmental requirements.
Lifetime
• Short-living (about 80 years ,willows, poplars, alder, birch), fast-growing.
• Average, (200-300 years, maple, hornbeam, rowan, spruce, pine), survivor type.
• Long-living (600-800 years, oak, linden, beech, yew).
Environmental requirements
• Temperature: thermophilic (oak, beech, maple), cryophilous (spruce), no particular requirements-
pine, birch, sycamore.
• Moisture: Xerophilic (rowan tree), Mezophilic (ash tree), Hydrophilic (willow).
• Light: Heliophile (larch, pine, birch), Average, Heliophobic (yew, hornbeam).
• Nutrients: most of the plants have average demand for nutrients, but some may demand alkaline
soils - Addition of Ca or acidic that requires addition of peat. Some species tolerates salinity - that
ones are suitable for street alleys (acacia).
passing the constraints), using natFX’s built-in • EASYnat by Bionatics is an add-in that can be
plant skeletons. All natFX trees and plants are used within AutoCAD or 3Dmax. Currently the
fully textured, based on scans of actual leaves, trees are parametrically generated from the in-
stems, and bark. NatFX is currently available as put data (height, width, age and season) (Fig-
a plug-in. ures 1 and 2).
Figure 1
EASYnat examples of render-
ings of the Horsechestnut tree
models.

Figure 2
EASYnat simulation of a
growth of a Horsechestnut
tree.
3rd 5th 7th 10th 15th 20th 50th year
EASYnat could be suitable for our further re- ple for trees height, canopy perimeter, trunk perime-
search - for implementation into object orientated ter), surrounding reaction (for example for trees soil
software like Revit or ArchiCAD and development requirements, geographic location, moisture), and
of their attributes and actions. The plants should interaction (for example for trees spread, ecologi-
have more attributes and while simulating the grow cally close trees, and antagonistic plants).
process, they should react and interact with its sur- In real life coding examples, the difference be-
roundings. tween inheritance and aggregation can be confus-
ing. If you have an aggregation relationship, the
UNIFIED MODELLING LANGUAGE aggregate (the whole) can access only the PUBLIC
BIM involves representing a design as combina- functions of the part class. On the other hand, inher-
tions of ‘objects’ that carry their geometry, relations itance allows the inheriting class to access both the
and attributes. Such objects and relations can be PUBLIC and PROTECTED functions of the superclass.
described by UML (Unified Modelling Language).
UML is a standardized general-purpose modeling DESCRIPTION OF POTENTIAL APPLICA-
language in the field of object-oriented software TION OF UML DIAGRAMS
engineering. It is a tool for defining the structure We have applied the class diagrams to our study
of a system through several types of diagrams. UML case: we needed to state the classes and all the
enables to model an application specifically and in- atributes that our classes should have. Then we have
dependently of a target platform (Fowler and Scott, created one instance of that class and the relations
2000; Blaha and Rumbaugh, 2004). (Figure 4).
UML diagrams represent two different views of These diagrams we use to map the classes that
a system model: static (structural) and dynamic (be- appear in our application study. While describing
havioral) diagrams (Schmuller, 2001). There are more the plants as object for BIM you have to take in ac-
than ten different types of UML diagrams. In our in- count all their properties and attributes -such as
quiry we are interested in class diagrams, object dia- height, perimeter of canopy, perimeter of a trunk,
grams and activity diagrams (Figure 3). classification in the meaning of family- genus- spe-
In our research we aim to advance the informa- cies- cultivar, to state if the plant is decidious or ev-
tion model of the vegetation to be comprehensive. ergreen, the aspect of season (mainly to state the ex-
This means that we not only model the general at- istence and color of the leaves), the age of the plant
tributes of vegetation (for example for trees family, and its habitus an finally for most efficient work with
genus, species, cultivar, decicious/evergreen, sea- the objects the detailing and the way of graphic ap-
son), but also the time dynamic attributes (for exam- pearance.

Figure 3
Activity diagram.

Figure 4
Class and instance diagrams.
Figure 5
Diagram of relations between
classes.
The same (properties and attributes) we should clists-parking places) and infrastructure objects.
state for the other objects that we would like to in-
volve in our application. DISCUSSION
In this preliminary diagram (Figure 5) you can In the paper we have presented our first findings to-
see buildings (can cover the minor architectural wards a Landscape Information Model. The work is
works in the countryside), site equipment (such as still in a very preliminary phase, but we hope to have
benches, trash cans, lightning, equipment for cy- laid some foundations. As trees are dominant visual

elements in landscape design, we have started with Sklenička, P 2003, Základy krajinného plánování, Naděžda
more concrete knowledge capture of trees. In the Skleničková, Praha
next steps we aim to implement a parametric data
model of trees in Revit. [1] http://www.bionatics.com/Site/product/index.php?lang
ue=en&UrlSuite=%2FSite%2Fproduct%2Fshop%2Fpla
REFERENCES nt%2Fdetail.php%3FCommunity%3D2%26ProductLin
Abdelmohsen, S, Lee, J and Eastman, C 2011, ‘Automated eId%3D10%26Category%3D11%26SubCategory%3D1
Cost Analysis of Concept Design BIM Models’, Comput- 12%26PlantId%3Daescu_b
er Aided Architectural Design Futures 2011 [Proceedings [2] http://en.wikipedia.org/wiki/Class_diagram
of the 14th International Conference on Computer [3] http://www.smartdraw.com/resources/tutorials/uml-
Aided Architectural Design Futures], Liege, Belgium, class-diagrams/#/resources/tutorials/Introduction-
4-8 July 2011, pp. 403-418. to-UML
Blaha, M, Rumbaugh, J 2004. Object-oriented modeling and [4] http://bim.wikispaces.com/file/view/Landscape+Inform
design with UML, Prentice Hall, Upper Saddle River. ation+Model+Building.pdf [5] http://www.cadinterna-
Eastman, C, Teicholz, P, Sacks, R and Liston, K 2008, BIM tional.com/cadinternational/product.php?prod=AB&p
Handbook: A Guide to Building Information Modeling for rodtype=PR#overview
Owners, Managers, Designers, Engineers and Contrac- [6] http://www.uml-diagrams.org/class-diagrams.html;
tors, Wiley, Chichester. http://www.visual-paradigm.com/product/vpuml/
Fowler, M and Scott, K 2000, UML Distilled: A Brief Guide to tutorials/
the Standard Object Modelling Language, Addison-Wes- [7] http://www.bimmanager.com/wp-content/up-
ley, Amsterdam. loads/2009/11/BIM-Process-Exchange-IDM.pdf
Ibrahim, M, Krawczyk, R and Schipporeit, G 2004, ‘Two Ap- [8] http://www.bimthinkspace.com/2008/11/effects-of-
proaches to BIM: A Comparative Study’, Architecture bim-on-project-lifecycle-phases.html
in the Network Society [22nd eCAADe Conference Pro- [9] http://www.cad-addict.com/2010/03/from-cad-to-bim-
ceedings], Copenhagen, Denmark, 15-18 September part-ii-from-cad-2d-to.html
2004, pp. 610-616 [10] http://databaze.dendrologie.cz/index.
Michálková, R 2011, Tvorba zeleně, VOŠZa a SzaŠ, Mělník. php?menu=5&id=67
Quitt, E 1971, Climate Regions of Czechoslovakia, Studia [11] http://www.cgw.com/Publications/CGW/2001/
Geographica 16:1-74, Geografický ústav ČSAV, Brno. Volume-24-Issue-8-August-2001-/Bionatics-Grows-
Schmuller, J 2001, Myslíme v jazyku UML: knihovna pro- Virtual-Plants.aspx
gramátora, Grada, Praha. [12] http://forums.cgarchitect.com/6326-trying-out-plant-
plugins.html

ifcModelCheck
A tool for configurable rule-based model checking

Sebastian Ebertshäuser 1, Petra von Both2
Karlsruhe Institute of Technology (KIT) Germany
http://blm.ieb.kit.edu
1
sebastian.ebertshaeuser@kit.edu, 2petra.vonboth@kit.edu
Abstract. On behalf of the BBR (German Federal Office for Building and Regional
Planning) the development of an Industry Foundation Classes (IFC) based inspection
tool was accomplished as application on an underlying work-in-progress development
framework. By providing a machine-based checking process the tool ModelCheck
was rolled out to meet demands emerged during pilot projecting. Thus it is capable of
processing automated compliance checks on quality criteria for the authorities, e.g.
documentation guidelines of BBR regarding building and real estate documentation or
building information modeling (BIM) quality criteria formed for the Humboldt-Forum
project – a BIM pilot-project managed by BBR. ModelCheck supports checks on IFC
models - formal against schemes and logical inspection with regards to alpha-numeric
content by using xml-based configurable rules.
Keywords. BIM; quality assurance; rule-based model checking; collaboration
INTRODUCTION
Besides being a promising concept from a general Thus, the participants of the survey highly
point of view, building information modeling (BIM) in agreed (65%) to the statement that the quality of
real world is still confronted with problems in terms digital building models in form and content is not
of overarching business and process related co-oper- adequately standardized yet (Figure 1).
ation on a base of its models. The results of a market The specification of such defined process in-
analysis regarding potentials and hindrances of BIM terfaces can be mentioned here as an important
application in Germany identify great prejudice and precondition for a simplified and secure contract-
reticence coming to business overarching transmis- ing and cooperation: By referring to normative
sion and cooperative usage of BIM models (von Both, descriptions contracts can be concluded very effi-
2012). In the market the benefit of BIM in terms of co- ciently and securely between client and planner re-
operation with project partners is worse than other spectively contractor as well as among the planners
more operative aspects by far. In such co-operation themselves. This becomes very important when the
activities nowadays one reason, besides inadequate contract partners – like in Germany – are composed
technical interfaces, seems to lay in insufficient newly for each project.
specification of exchange conditions and qualities of On one side Germany’s Architecture, Engineer-
model data respectively building information. ing and Construction (AEC) sector addresses this

problematic situation on a normative level by dis- Figure 1
cussing a development of a set of standard specifi- Agreement to the statement
cations with regard to contents and process related that the quality of digital
quality criteria describing building models. On the building models in form
other side an increasing number of larger construc- and content is not adequate
tion companies and general contractors among standardized yet.
practitioners have already started developing inter-
nal (process related) standards of quality. These are
used as mandatory basis for co-operation towards
project partners. Though the German authorities
have also developed generalized quality guidelines,
for instance the ‘Dokumentationsrichtlinien für
den deutschen Gebäudebestand’ (documentation
guidelines for German building stock (Figure 2)).
These are capable of describing - in a BIM context
- the execution of existing quality requirements in e.g. part submission warrant and material coding.
form and content. A main part of this directive is the Toward this quality standard a first step was tak-
alphanumerical building description that includes en by developing appropriate checking reports that
the constructional as well as the technical room data enabled a process accompanying inspection and
sheets. Thereby ‘article’ (countable types of furnish- evaluation of model data received by project part-
ing respectively equipment that can be specified ners. Taking exchange scenarios into account, prior
with further properties) and ‘feature’ (abstract defini- defined in the contract agreement, a process-ac-
tions of objects’ properties) are fundamental struc- companying quality control process was able to be
tures for description. Equal characteristics of plots, executed. However, in the context of the HUF project
buildings, spaces, equipment etc. are described with manual (with eyes) quality control of the acquired
a uniform code of property. alphanumeric criteria proved to be very extensive
Uniqueness of quantity-on-hand data has to be and ineffective. Thus a demand clearly emerged for
permanently assured by all involved stakeholder suitable tools regarding model management and
in order to guarantee consistency towards forward alphanumerical model analysis respectively model
projection with the streamlining exchange of digital checking. The special need was to enable process-
building information data (coding of plots, build- accompanying capabilities of generically rule-based
ings, floors, spaces and if necessary identification semantic inspection of BIM models. After a market
numbers) [1]. Thus corresponding concepts for pro- analysis it became obvious that so far none of the
cess-accompanying quality assurance in the context available tools had been able to satisfy the special
of BIM were developed and exemplarily evaluated in demands of the authorities like a high level of flex-
pilot projects. ibility and dynamic references. The business models
In the pilot project Humboldt-Forum (HUF) initi-
ated by the Federal Office for Building and Regional Figure 2
Planning (BBR) quality criteria referring to this direc- Detail from documentation
tive were specified. These criteria do not only regard guidelines for German build-
to a future use of model data in facility manage- ing stock.
ment, e.g. space coding, topological allocations etc.,
furthermore they also cover important information
for materialization and cost-estimation processes,

of the available tools did also not allow a flexible requirements side. This model checking workflow
extension of the rule sets - they mostly only sup- meant to be carried out by users without specific
port a configuration of already existing basic rules. knowledge required - therefore developing aimed
The implementation of totally new rules requires a at a lightweight tool as a so called ‘standard’ ver-
charged development order. Thus, the BBR decided sion. In four steps it should be possible to load a
to develop own adapted BIM model checking soft- model file, check its contents with chosen rules from
ware in cooperation with the KIT to extend their IT- a given repository and finalize a result report that
infrastructure and meet their specific requirements. can be exported as Excel-XML. While focused on the
implementation in the HUF project it was meant to
REQUIREMENTS AND SOFTWARE CON- streamline first practical experience on BIM applica-
CEPTS tion by BBR.
Prefacing the description of the tool development, The deduction of the development steps from
crucial requirements will be subsumed on the tech- requirements for the proper checking tool was
nical side and crucial aspects of the software con- confronted by an opposed development task – a
ception are pointed out. Besides a brief overview on rule configuration conception for BBR. A complex
the technical implementation, the concept of rule requirement level with emergent outcomes to be
configuration is subsequently presented more de- considered stood at the core of this task. For in-
tailed. stance, technical implementation and maintenance
On the technical level, the above derived logical of machine-readable logical contents did require
requirements were further divided in those concern- designing totally new workflows (developing rule,
ing the HUF context and the ones with regard to maintaining repository etc.) accompanied by the
content. In terms of BBR – aspects related to actively need to define the new user roles (model checker,
running processes of the overarching BIM context in rule administrator). Setting up the overall concep-
the ongoing HUF pilot project on the one side and tion had to be considered as a fuzzy context of fu-
on the other side the content- and organization- ture usage (after implementation in pilot project).
related approaches needed for implementation Beside these ‘invisible‘ requirements, future user/
of automated compliance checks on documenta- user’s structures is unknown at the time of defin-
tion guidelines contents in process-accompanying ing specific tool functionalities – all requirements
BIM-model use cases with a long-term perspective for managing and manipulation of rule logic share
on deriving robust internal IT standards results of common grounds e.g. need at different granularity
the pilot projects. For BIM this meant the concep- of different previous knowledge on certain sets of
tion of a well reachable and compact workflow in circumstances (knowledge of model, knowledge of
order to organize the discipline-overarching data rules, knowledge of use case workflows etc.).These
exchanges. Contrary, on the side of developing and identified kinds of overarching concerns made it
managing generic rule repositories for documenta- necessary to consider in the overall conception, the
tion guidelines, flexible ‘organizing’- functionalities very diffuse yet universal requirements for the user
stood alongside with requirements regarding the interface logic of the software.
user’s autarkic manipulation of rule logic in the fo- At last are the specific demands condensed in
cus of conception. Both levels of requirements were seemingly arbitrary requirements, for instance the
thematically separated and developed into two final reduction of inspection to only a part of the subject
versions of the software. matter (IFC model instance), because of BBR’s over
At first a clear conception guideline for devel- time evolved structures. So while all rules concern
opment was given through the clearly described the alpha-numeric part of IFC-model, this relates
operative focus on one central workflow on the in terms of content to the requirements regard-

ing the rules to be checked with the software, as tensive administration version. In this way, the rule
they deflect documentary guidelines with regards development environment always has the model
to content that were first adapted for BIM context checking functionality at hand as needed for testing
and then transferred into generic rules for model and verifying rule logic. Again placing the manage-
checking. Regarding IT infrastructure, BBR already ment of rules in the separate version generally pro-
had software for checking model geometry (clash vides a base for robust user’s structure – e.g. version-
detection), which was not capable of providing suffi- ing control separated from usage of rules.
cient user-specific configurable inspection of model In order to present the logical integration of the
semantics. In contrast to heavyweight/monolithic management environment rule library on a macro
proprietary checking software that fail to support level e.g. the user-regarded convention-based frame
users with full autarkic rule development capabili- settings, with the manipulation environment rule
ties on behalf of large functionality range (geometry configuration on the micro level e.g. code, notations,
and semantics), causes a high level of complexity in standards, derived conventional settings, a bottom
terms of rule layout e.g. Solibri [2]. In a long-term up approach will illustrate the latter level on behalf
perspective for instance, a central requirement was of the constitution of a rule respectively crucial as-
to keep this generic rule repository up to date by pects into turning logic to automated compliance
own means - in case of further developments in checks e.g. as a cross cutting concern of the conven-
federal guidelines regarding buildings. Thus devel- tion on how one rule logic expands over several rule
opment aimed at a freely configurable and manage- files.
able user’s rule base. On the bottom level, the logical counterparts
By especially enabling a separation of concerns to the main XML tags in a rule file refer to loosely
between the knowledge of model users put into coupled constitutes of one logical clause – a so
rule developments, and developers taking care of called rule envelope. All assigned rules with regard
software related maintenance regarding the inspec- to contents were therefore decomposed in rule
tion shell, it/this seems to be the better business components that serve as elements in a model kit
model as compared to mixing these concerns. Final- – embodied in the rule configuration. An overriding
ly, the aim was integrated in the software concep- ‚orchestrating‘ algorithm, that is distinct accord-
tion by physically decoupling of the rule logic from ing to the issue of the rule logic and sets the logical
the inspection-shell tool (business logic), and plac- parts of the specified rule clause in relation to each,
ing it in open described xml schemes describing the other respectively condenses the parts to the above
user’s domain-specific rule logic. introduced serial process-able check workflow. As in
every rule file components are linked likewise to a
IMPLEMENTATION AND USAGE CON- checking workflow on the level of the beneath ‘code
CEPTS skeleton’ (cloze with wildcards for the return values
Besides serving BBR as a tool (standard version) in of the rule components), a set of core templates were
practice with ad hoc revenues within the HUF pro- introduced as basic rule envelopes. Every variegated
ject, having all rule logic at hand (admin version) rule is derived from one kind of query at the core - in
stays warranted if enhancements in rules take place other words this origin (basic rule) connects all other
because of acquired experience in practice over sibling rules in the tool to each other. Allocating one
time that the user can successively put all together – basic approach to each general central questionings
self-tailored for internal workflows – into a common gives the user a decomposition principal at hand for
rule library and manage it from there. formation respectively further development of the
Strategically it had been considered to imple- own rule repository and eases editorial aiding of a
ment the standard version as a subset of the ex- current state through elimination of redundancies.

Figure 3
Software conception: logical
contents of rule XML.
Along the basic structure of an example rule Regardless of the user’s knowledge, a verbally
(Figure 3), the rule components will be introduced. represented questioning (in the example: “Does
The decomposition of the components, hence the every window have a U-value less than 1.2?”) can
basic structure was thereby adopted from the Ob- function as an initial point of the rule development.
ject Constraints Language’s (OCL) main statements In a first step, the rule administrator dismembers
[3] in order to promote a convention-driven limita- this concrete question according to the four com-
tion to the source code used in the user’s rule files by ponents of the check workflow. Depending on a ba-
only implementing common templates. sic population of all components embedded in the

rules of the repository, it can at this stage already be At the structural base of the hereinafter de-
possible to suit the logical clause of the rule, only by scribed concept of the knowledge-oriented usage
choosing existing instances of components and if strategy for dealing with complexity of involved
applicable alter the specific values. matters, stands again the decomposition into the
Specific quantities of element instances are al- above mentioned rule components. Three access
ready gathered during loading the model in order levels (in Figure 3: dashed lined boxes) are provided
to show a brief summary on the model content. to the user in the rule configuration. In every higher
These can also be adapted in the components by level thereby – as a further abstraction of complex-
referring to their specific keyword. In cases where ity - less knowledge is required to manipulate rules
specific quantities are not set up at start nor defined with regard to contents. However, this abstraction is
in another rule’s context, the rule administrator alter again streamlined by limited amount of editing pos-
a similar query or create a new one. Thus specified sibilities. Only elements that have been previously
in the so called Select Type component, the related defined in the level below can therefore be referred
congregation of instances is being hooked to a main to in the higher level. At the top level, keywords en-
iteration mechanism/slope in order to check a cir- able the cascading/coverings of complex logic to
cumstance on each of the selected IFC Type instance. the unseen background. On this level it is possible
Further specification of a property to the prior to create variants of existing rules only with combi-
chosen IFC type is defined in the so called Property nation of given components and knowledge of the
component (in the example: the pointer to the U- rule constitution. When additional object quantities
Value in the property set of the window type). So far are needed, domain knowledge of the constitution
the specified components will allow the main itera- of the model to be checked is already crucially re-
tion through specific actual values of the selected quired – the only further knowledge required is on
congregation - in order to check these values, a cor- how to handle the statements, as described by the
responding desired output of the value is assigned OCL standard.
in the so called Desired Out/ Target Value compo- In a middle level queries are therefore described
nent. By defining an appropriate condition for the in OCL-conform syntax and made referable by a
relationship between actual and targeted value in self-chosen synonym (keyword for the top level),
the so called Check Criterion/ Criteria, the last part together they are persisted in the XML rule file as
of describing the rule clause is accomplished. By key-value-pairs. Knowledge to its full extent – rule
standard, this now fully specified checking work- constitution, domain-related as well as application
flow will return all elements of the congregation of the code skeleton being used (basic rule tem-
that do not meet the specified condition – a sub- plate) – are only crucial on the ground level in order
set with error prone instances. In order to achieve to create or extend the basic rules. With this concept
a comprehensive output of the checking results, at hand, the BBR staff is on the one hand in the posi-
the so called Result component can be adjusted by tion to ad hoc address different questionings within
specifying a suitable error description as a cloze with the HUF project on a base of manipulating the rolled
wildcards for crucial single values to clearly outline out basic rules, and they are on the other hand in a
the returned error prone circumstance. With access long-term perspective by gathering necessary ex-
to functions of standard version for model checking pertise over time, so that the user can then be ena-
and functionality for verifying of developed rule, the bled to maintain and further develop rule repository
rule configuration is set up with all necessary means on own means.
of editing. This is streamlined by a concept of usage Through inclusion of the configuration envi-
(UI) which is decomposed in different levels aligning ronment with the rule repository in the hereinafter
with required stages of user’s knowledge. described rule library environment, an overarching

management of the development work on base of prone because of the fact that it is non existent, it
the rules is ensured. Bound by convention - all rules is not covered by the questioning concerning its
are obligatory referred to by at least one rule set – value check. In this case, the user has to warranty
in the rule library environment the user is provided that inside the set of the specific rule, there is a prior
with manipulation functionalities of rule sets only. rule that checks the existence of covered elements
By the ability of individual assignment of rules to before the more specific check on the element val-
respectively different sets, a user-specific structur- ue takes place. Since there are “chains” of existence
ing is also enabled e.g. project-oriented. The struc- checks for every specific rule in order to ensure the
turing of the own rule repository is on the technical overall correctness of covered contents - all involved
side supported by a technology known from com- elements - there seems to be great potential for in-
mon Integrated development environments (IDE) ternally standardizing this issue and structuring it as
for dynamically compiling and executing rule code a basic subset in the repository. Although turning
during runtime of the tool. Thus this enabled physi- the exhaustive responsibility for the domain-specific
cally decoupling the user-specific logic concerning logic in the user’s hand is accompanied by complex-
knowledge on model domain, rule repository, work- ity in terms of managing the rule repository in the
flows etc. from the checking shell, since the current case of BBR application this seemed reasonable.
repository is loaded from the XML database at runt- Beginnings in the field of BIM model check research
ime only. Therefore, the own rules can also be struc- can be seen especially in countries were BIM-models
tured and managed independently of the tool, if for are put to practice, for instance Balaban (2012) sup-
instance there is available infrastructure like Content porting Turkish authorities with automated compli-
Management System (CMS) etc. ance checks for fire guidelines. However segmented
A basic structuring along the domains of the chains of requirements due to complex application
checked models was loosely integrated by strictly contexts seem a common ground thereby. In or-
keeping XML parsing separated from checking der to achieve the goals mainly focuses get carried
mechanism in business logic and user-domain logic. on very specific context-dependent solutions and
Since the tool initially construed for checking the IFC therefor often show contradictive mixing of user
XML standard already at core of basic rule develop- and developer concerns. Hence they seem to be
ment, a latter decoupling of basic XML-operations more vulnerable in a long term perspective; it is dif-
from the domain-specific aspects was focused. The ficult to find a suitable allocation of developer and
basic rules that were therefore rolled out with the user respectively a suitable organization of the fur-
software, give the user a solid base to start an own ther development of the rule repository after solu-
robust repository from. Thus in terms of different fu- tions are implemented.
ture demands, it could also be further developed in Because the pilot character of the HUF project in
a model-overarching manner. the usage of BBR, it was necessary to create a totally
As pointed out in different aspects covered new rule repository. The separation of concerns was
above, making the software more flexible on the suitable since the user first has to build up expertise
technical side is crucially accompanied by transfer- regarding different domain-specific disciplines. It
ring of logical and structural responsibility to the enables thereby also successively gaining grip on
user as well. This means for instance that before the full functionality range of the software.
checking specific details, all model-instances in- Other than these concerns regarding logical and
volved have to be checked whether their instances structural responsibilities, all overarching concerns
are completely implemented in the way the specific regarding plainly information technology were
rule refers to them. Since a non existing instance considered in the tool shell. Thus throughout all
of a model element will not be recognized as error developing processes, a strict internal library-wide

Figure 4
Concept of collaboration
framework.
versioning keeps every applied rule with regards TECHNICAL BASE OF IMPLEMENTATION
to content referable to every produced inspection As an application-independent central service, the
result. This seems important e.g. it guarantees es- framework provides base functionalities for the
sential consistency for archiving purposes – older model based data analysis. The base functionalities
inspection reports are always mapped to the valid serve collaborative data handling, e.g. type- and
rule at time of inspection. Import and export func- attribute-oriented selections of partial and aspect
tionality enables an administrating user to deploy models, integration of these partial models as well
newly finalized rules to users of a standard version as supplying mechanisms for versioning, change
that only use production ready rule sets. This allows management and transaction control (Figure 4).
team-internal organization of roles within a user Thus a kind of “Meta-Model-Server” is provided
group. Together, versioning and a project specific for further research and development projects that
structuring enables parallel usage-oriented organi- in different application scenarios can be imple-
zation of similar rules in different contexts. mented for different kinds of model standards like
Putting ModelCheck into practice (HUF project) ifc STEP, CityGML or GAEB. Furthermore it supports
enables iterative evaluation and optimization regard- model overarching model-operations (Hartmann
ing rule logic for the user – whereon further develop- and von Both, 2011).
ment of inspection shell functionality is then based
upon. ModelChecks further development also takes OUTLOOK
advantage of extensions to the underlying frame- Further development on the analysis and visualiza-
work, which will be briefly introduced hereinafter. tion components in the context of energy efficiency
will take place within the science project “EneffBIM”

Figure 5
Conception draft EneffBIM
project.
(funded by the German funding program “EnTools” towards the layer of decision making. ModelCheck
released by the German Ministry of Economy) start- then serves as a checking and analysis tool for evalu-
ing summer 2013. As seen in Figure 5, especially for ation of variations with their different simulation re-
the usage of dynamic energetic simulation, the logi- sults.
cal content-regarding model analysis shall be the In this case a great meaning is beard to seman-
quality management vehicle for securing the inter- tic visualization of simulation results (specific con-
face from BIM to simulation. straints of property values), representation of the
With involvement of different Frauenhofer insti- range in values that exists in the comprehensive
tutes (ISE and IBP) as well as the universities RWTH model and also checks on the characteristics of val-
Aachen, UDK Berlin, KIT and buildingSMART on one ues.
side, the IFC model will be extended with energy
relevant base types (input parameter) and suitable REFERENCES
geometric representation forms. On the other side von Both, P. 2012: Potentials and Barriers for Implementing
regarding energetic simulation, tools for co-sim- BIM in the German AEC Market - Results of a current
ulation in the context of Modelica will be further Market Analysis, contribution to the30th eCAADe con-
developed and a synchronization of existing model ference, Prague, 09/2012
libraries is been focused on. Hartmann, U.; von Both, P. 2011: Ein Framework zur
Concerning model checking aggregated simu- Definition und Durchführung interdisziplinärer,
lation results shall be led back in the BIM model in modellübergreifender Analysen am Beispiel so-
order to ensure better re-transition and evaluation larer Einstrahlpotentiale im urbanen Kontext.
of simulation results into the planning process and BauSIM2012 Conference - Gebäudesimulation auf den

Größenskalen Bauteil, Raum, Gebäude, Stadt, Stadt- [1] http://www.bfr-gbestand.de/inhalt/inhalt.html.
quartier. Berlin 09/2012. [2] http://www.solibri.com/
Balaban, Ö. 2012: Automated Code Compliance Checking [3] https://en.wikipedia.org/wiki/Object_Constraint_Lan-
Model for Fire Egress, contribution to the30th eCAADe guage
conference, Prague, 09/2012

Daylight Performance Simulations and 3D Modeling in
BIM and non-BIM Tools
Interoperability and accuracy – an experience aiming to a more integrated

and interdisciplinary approach
Marina Stavrakantonaki
Brussels, Belgium
marina.stavrak@gmail.com
Abstract. The fusion between building assessment and design can lead to better informed
design decisions. Performance oriented design is better supported through the use of
interoperable file formats for data exchange between BIM and non-BIM tools. At the same
time, the parameters that influence the calculation during a performative assessment are
no longer a purely engineering problem, since 3D modeling is of primary importance
in defining the numerical output. The role of the designer along with the selection of
the tools becomes all more relevant in this direction. A framework is presented hereby,
which can be used for the selection between different BIM tools for daylight assessment.
An insight is also given on the major parameters that can affect the outcome and on the
obstacles that were experienced in four case-studies in relation to data exchange and
information flow.
Keywords. Performance simulations; parameters; interoperability; daylight.
INTRODUCTION
During the last years, there is an increasing demand ance and DIVA 2.0 as plug-in for Rhinoceros NURBS
for the integration of BPS (Building Performance modeler, and to provide suggestions for future use.
Simulation) tools in the early design phase (Attia et All of the examined tools can provide dynamic day-
al., 2012). The interoperability of BIM (Building Infor- light simulations under given conditions. The prob-
mation Modeling) and non-BIM tools influences the lems that consulting with the use of this software
workflow within the design team, while the building faces on a daily base, are related to incompatibility
practice is progressively oriented to a more interdis- between the architectural 3D model and the simula-
ciplinary approach (Augenbroe, 1992). The hereby tion software, the long 3D modeling times and the
presented study initiated as an internal research error probability when complex geometries are in-
for the consultants of the company DGMR in the volved. The aim is to acquire semantic information
Netherlands, with the task to evaluate the three fol- on the performance of the building over time, in a
lowing daylight performance simulation packages; way that it can be integrated in the design process.
Design Builder v.3.0.0.105, Ecotect 2011 v.5.60/Radi- The evaluation is based on the following criteria:

• Ability to simulate detailed and complex build- bases. The process can be analyzed through the fol-
ing forms lowing steps;
• Accuracy 1. setting of the model
• Interoperability of Building Modeling (IBM). 2. link of the model to the relevant for each case-
The above-mentioned criteria refer to two of the five study semantic information
major tools’ selection criteria as defined by Attia et 3. calculation and presentation of the output.
al. (2012) and incorporate the three factors that are Each of these steps presents a different level of
given in the ASHRAE handbook: accessibility and control potential for the user, the
• Capability of the tool to deal with the project later becoming considerably limited as we proceed
requirements. from 1 to 3. A fully controlled process would opti-
• Complexity of input. mally lead to more accurate results, since param-
• Quality of output. eters such as the geometry, material definitions or
Their selection for this study, reflects the problems sky conditions would be fully editable, provided that
with which engineering is confronted the most dur- the user is conscious of the influence of each param-
ing collaboration with the architectural team for eter on the calculation output. Radiance provides
daylight assessments. At the present moment, it is such a possibility through a number of file formats
common practice that the analytical model is de- (i.e., .rad, .dat), without posing restrictions regard-
rived from the architectural model after the extrac- ing the computed geometry. An example on how a
tion of a significant amount of geometrical data. variation of the parameters affects the results is pre-
Effort and long working hours are dedicated to the sented during this paper.
restructuring of the architectural model before sim- Ideally, data exchange would happen automati-
ulating; and it is common that decorative elements cally in both directions, so that every alteration in
need to be extracted, or the layering structure of the 3D model can change the simulation output and
the 3D model has to be redefined in order to pro- vice-versa (Kensek and Sumedha, 2008). In such a
vide appropriately defined layers where material case better building performances can be achieved
properties can be accurately assigned. At the same through a seamless back and forth process. Interop-
time, the ability of the software to simulate detailed erability can be defined as the possibility of informa-
or complex building forms is closely related to issues tion exchange through interoperable file formats
of processing power and computing robustness and that allow for the use of the exchanged information
becomes clear that the building geometry that is Its importance lies on the possibility to diminish the
used for performance evaluations has a direct im- time lost due to the exchange of data between the
pact on the calculation output. BIM or non-BIM modeler and the BIM tool.
A better collaboration between the design and In building practice, performance assessment is
performance assessment team is therefore neces- often carried out separately from the architectural
sary; in such a case the architects need to be in- design. The engineering team, is in many cases de-
formed on the effect of their model to the calcula- tached from the design of the building, while the
tion process, so that they can structure it efficiently architect uses assessments only as external informa-
in order to facilitate, not only an accurate three-dition when compliance with regulations is strictly re-
mensional representation of the building, but also quired. A better merging of the groups can be facili-
a fast performative assessment that can provide tated with a better collaboration of the software, so
feedback for better informed design decisions. Re- that both teams can refer to the same core models
garding accuracy, in each of these tools, the output through interoperable file formats.
is the result of the connection of the imported or This study traces the flexibility of the examined
designed geometrical entities to a number of data- software in importing and modeling 3D geometry

Table 1 Basic input, model 1 to 3
Basic input data models 1 to 3. Design NLD_Amsterdam. double concrete ceiling, Grid : 0.1m-0.2m
Builder 062400_IWEC weather file glazing with brick walls, Calc. plane: 0.7m
v.3.0.0. CIE overcast sky airgap wooden floor
LTA 0.7
Ecotect NLD_Amsterdam. LTA 0.7 ceiling ref. 80%, Grid size:
2011 v.5.60 062400_IWEC weather file brick walls ref. 50%, 40x32x32cm
CIE overcast sky wooden floor ref. 20% Calc. plane: 0.7m
DIVA 2.0 NLD_Amsterdam. double generic ceiling ref. 80%, Grid-nodes’
062400_IWEC weather file glazing generic wall ref. 50%, density 500x800
CIE overcast sky low-e generic floor ref. 20% Calc. plane: 0.7m
and the range of deviations that can be expected projects and to a simplified setting:
during the calculation of the analytical model, and • Case study 1: a complex geometry (Figure 1).
specifies the information that is being lost during • Case study 2: a purely orthogonal geometry
the process. The Drawing Exchange Format (DXF) is (Figure 2).
hereby used as the basic means of design informa- • Case study 3: a simplified setting of a typical of-
tion transfer. fice space (Figure 3).
The common feature between the geometries is the
METHODOLOGY linear form. They refer to two on-going projects and
For the needs of the present research three case- one simplified setting that is often met in everyday
studies were used on the grounds of the following practice.
methodology; a base model was prepared and sim- The simulation was oriented to one-variable
ulated in Design Builder v.3.0.0.105. This model was approach in order to facilitate the comparison be-
exported in .dxf format and recalculated in DIVA 2.0 tween the tools. With regard to precision, the fol-
and Ecotect 2011 v.5.60. Rhinoceros 3D-CAD mod- lowing settings were used: ambient bounces (ab) =
eler was used as a complimentary tool in order to 2, ambient accuracy (aa) = 0.1, ambient resolution
model the missing export data. All the three pack- (ar) = 300, ambient divisions (ad) = 1000, ambient
ages were linked to Radiance and provided output super-samples (as) as default. For the needs of this
based on the climate data of Energy- Plus _IWEC study, the Daylight Factor was chosen as the main
weather data files. The following input data were calculation measurement; the prediction of the
given for each one of the packages. The settings re- Daylight Factor under a CIE overcast sky condition
flect the effort to use equal input data. Identical in- is at the moment the dominant approach in evalu-
put is not possible at the moment due to differences ating daylight, despite the fact that it provides only
in the software settings (Table 1). a rough estimation of the yearly indoor conditions
The three case-studies refer to two on-going (Tregenza, 1980). Yet, it is in broad use by the Euro-
Figure 1
Case study 1 as modeled and
imported in the three tools;
from left to right, in Design
Builder, in Ecotect and DIVA.

Figure 2
Case study 2 as modeled and
imported in the three tools;
from left to right, in Design
Builder, Ecotect and DIVA.
pean building regulations and most assessment rat- only in Ecotect and DIVA (Figure 5). This last model
ing systems including BREEAM, in order to provide was further used to monitor the effect of the input
benchmark values for indoor daylight quality. The parameters, regarding the architectural form as one
specification of daylight quality for the presented of them. The tests were performed in DIVA 2.0 and
case-studies lies beyond the interest of this study. provided an insight on the deviations that should be
The aim is to evaluate the three packages on the se- expected with the change of specific variables. The
lected criteria and to provide a general framework most important of the variables that were tested
that can optimally facilitate the selection between and the resulting output under CIE overcast sky for
the numerous daylight performance calculation the same IWEC weather file are listed in Table 2.
tools that are at disposal as open-source or commer- The above listed results are some of the tests
cial software packages. that were carried out in order to specify the influ-
Further on, a fourth model was chosen as a ence of the precision settings, the grid density and
separate case study. Its geometry combines circular material properties on the output. Hereby we set as
openings on a circular wall and is part of a project Low precision: ab (ambient bounces) = 2, ad (ambi-
currently under development (Figure 4). The model ent divisions) = 1000, as (ambient super-samples) =
could not be created in Design Builder and the ge- 20, ar (ambient resolution) = 300, aa (ambient accu-
ometry was imported in Rhinoceros and Ecotect as racy) = 0.1, geometric density =70. High precision:
an .obj file format, which was provided by the archi- ab (ambient bounces) = 3, ad (ambient divisions) =
tectural team. Importing an appropriate model for 2048, as (ambient super-samples) = 20, ar (ambient
daylight calculation via gbXML or connection with resolution) = 512, aa (ambient accuracy) = 0.2, geo-
SketchUp in Design Builder proved also problemat- metric density =70.
ic. As a result, calculation output could be obtained
Figure 3
Case study 3 as modeled in the
three tools; from left to right,
in Design Builder, Ecotect,
and DIVA.

Figure 4
Model 4 in Rhinoceros inter-
face: model setup.
Figure 5
Case study4 as calculated in
Ecotect (left) and DIVA (right).
RESULTS number of ambient bounces that could give pre-

The tests prove that the default precision settings cise results was 3, whereas a very high precision (4
should be considered with skepticism; the minimum to 6 ambient bounces) did not considerably change
Table 2 Tests model 4

Tests on model 4. Test nr. Material ref. (%) Precision Mean D.F(%)
Test 1 Ceiling 0.9, int.wall 0.9, floor 0.5, int/ext. Low, Grid point 1.08
glass tr.0.9 dist. = 0.1m
Test 2 Ceiling 0.9, int.wall 0.9, floor 0.5, int/ext. High, Grid point 1.80
Test 3 Ceiling 0.9, int.wall0.9, floor 0.5, int/ext. High, Grid point 1.78
Test 4 Ceiling 0.9, int.wall 0.9, floor 0.5, int. glass High, Grid point 0.70
tr.0.65, ext. glass tr. 0.9 dist. = 0.1m
Test 5 Ceiling 0.9, int.wall 0.8, floor 0.2, int/ext. High, Grid point 1.59
Test 6 Ceiling 0.9, int.wall 0.8, floor 0.2, int. glass High, Grid point 0.98
tr.0.8, ext. glass tr.0.9 dist.= 0.1m
Test 7 Ceiling 0.8, int.wall 0.8, floor0.4, int.glass High, Grid point 1.07
tr.0.8, ext. glass tr.0.9 dist.= 0.1m

Detailed output model 1 Floor area Average Min. Min./Max. Table 3
Above threshold(%) DF (%) DF (%) Illuminance (lux) Detailed simulation output,
Design Builder v.3.0.0.105 model 1.
Ground Floor
Zone1 61.350 3.916 0.016 1.57/2291.96
Zone2 83.245 2.86 0.5 50.35/964.46
Second Floor
Zone1 58.898 4.339 0.116 11.62/1970.44
Zone2 48.911 3.248 0.2 20.29/1825.88
- Ecotect 2011 / Radiance Mean Illum.
Ground Floor graphically 5.385 graphically 309.6258
Second Floor presented 4.366 presented 199.1255
- DIVA 2.0
Ground Floor 62.8 5.5 graphically 217.5
Second Floor 50.7 3.91 presented 158.13
Output av. DF(%) Design Builder Ecotect 2011/ DIVA 2.0/ Table 4
v.3.0.0.105 Radiance Rhinoceros Simulation output, average
Model 1 (15.993 meshes) DF, model 1 to 4.
Ground Floor Zone 1/2 3.916/2.86 5.385 5. 5
Second Floor Zone 1/2 4.339/3.248 4.366 3.91
Model 2 (23.580 meshes)
First Floor 2.194 2.2 3.25
Second Floor 2.112 2.34 3.35
Model 3 (136 meshes) 6.378 7.66 7.94
Model 4 (74 meshes) Not obtained 0.95 1.07
the output. The grid density is important, yet a me- building forms as well as the differences in preci-
dium density with a point-to-point distance = 0.1m sion, even when the input settings appear identical.
is enough to provide a reliable output. For double Moreover, the information that we can obtain with
density (point-to-point distance = 0.05m) the effect one and only simulation from each program, varies
on the results was at the range of 1.11%, meaning significantly (Table 3). The analysis of the results of
that for models consisted of a large number of sur- test-model 4 have already provided a ranked list on
faces, an extremely dense grid can be safely avoid- the deviations that we should expect when altera-
ed. The most important material settings are ranked tions in the input parameters occur.
in the following line from the most to the least im- As seen in the results from model 1 (15.993
portant: visible transmittance of glazing, reflectivity meshes), the setting on the principle of zones does
of the walls, reflectivity of the floor, reflectivity of not facilitate the acquisition of direct information
ceiling. on a specific floor. The averages obtained from the
Tables 3 and 4 present the output from the four zones hereby do not provide clear input for the
case studies as simulated in the three software pack- design team. At the same time, the differences be-
ages. The differences in the output are indicative of tween Ecotect/Radiance and DIVA range between
the deviations that can result from the different abil- 2 and 10%. In the second model (23.580 meshes)
ity of the software to simulate detailed and complex the deviations are bigger; The difference between

Design Builder and Ecotect are below 0,5% for the We attempted to show that the architectural
first floor, yet they are around 10% for the second model is of major importance in defining the choice
floor. DIVA gave a daylight factor, by approx. 50% between the available software packages and fa-
higher if compared to the other two programs. In cilitating the flow of data from the design model to
order to explain the differences, a third simple cubi- the calculation/analytical model and vice versa. The
cal space was prepared through the same process following criteria can be used as a guideline for the
(136 meshes). The deviation in the output between selection of the appropriate BIM tool for the calcula-
Ecotect and DIVA was approx. 3,6%, while Design tion of daylight:
Builder presented by approx. 20% lower values. In • The type of geometry that the tool is expected
the fourth case study we observe that Ecotect and to model and calculate; failure in modeling
DIVA show a deviation around 10%, whereas no re- does not necessarily mean failure in perform-
sults were obtained in Design Builder due to mod- ing the simulation. In such a case, the data
eling difficulties as already explained. exchange formats that will be used to import
The deviations prove, that accuracy is relevant the model are of primary importance. The user
and highly related to program settings, while they has to be informed on the amount of data that
are expected to rise when a significant amount of can be transferred through the supported file
surfaces is computed. Identical settings for the pro- extensions.
grams are not possible, whereas the 3D form should • The expected precision in input/output; the
not be considered identical once imported in differ- choice of the tools should depend on the de-
ent packages. manded precision. In any case knowledge of
In relation to interoperability, the workflow can the influence of the input variables on the re-
be supported between BIM and non-BIM tools, yet sult is necessary.
it is important to control the information transfer • The tolerance of the software in handling
while exporting a model; 3D geometry can be easily different amounts of data; regarding geom-
prepared within Design Builder for example, yet the etry, the total number of surfaces that will be
exported .dxf files do not transfer material database computed has to be taken into consideration.
information and are normally deprived of the glazed Abstraction is always necessary in order to fa-
surfaces when imported in Rhinoceros or Ecotect. cilitate the time and precision of a daylight as-
Additional modeling is then necessary, in order to sessment.
correct the missing surfaces. An export in multiple
files is sometimes advised, in order to facilitate a CONCLUSIONS
multi-layered imported model. A number of major drawbacks were experienced
The main reason for which Design Builder was during the process, especially in relation to compu-
used as a base modeler was the fact that most of tational effort when a significant number of surfaces
the energy calculations demand an apart modeling was involved, regarding a higher probability for er-
based on energy zones. In such a case, the daylight rors and longer calculation times. Early stage assess-
performance simulation of a building could be per- ment tools, such as DIVA, can also prove less flexible
formed either on the architectural or the energy 3D during later stages of the design. The reason is the
model, since both are prepared by the team during lack of detailed simulation output and the inability
the study. In many cases, such as the three first case to set input values directly on the modeled surfaces.
studies, daylight simulations are possible in both The three software packages should be evaluat-
models. Yet, there are cases that the demands of the ed differently with regard to the three criteria as set
3D model exclude one of the two tools, as seen in at the beginning of this study. In this frame, Design
case study 4. Builder and DIVA proved that we are not far from the

moment when one tool will be used both for the in facilitating the collaboration of the team and pro-
setting of the 3D model and its performative evalua- viding an efficient base model that will be bidirec-
tion. Nonetheless, accuracy in the calculation output tionally used, therefore resulting to a better fusion
remains highly dependent on the model setting. For between the disciplines. In this sense, an answer to
this reason, further examination of the consecutive the current interoperability problems can be given
calculation steps and the relevant files which are by a better coordinated design and assessment
produced during simulation (i.e .rad, .tmp), are of team, both having a global understanding of the
great importance, and can explain some of the de- process. Such a team will facilitate the flow of in-
viations in the output. On the other hand, Ecotect formation through the setting of the 3D models, in
presented lower flexibility with regard to geometry both BIM and non-BIM tools, when the complexity
manipulation, even though it supports detailed ma- of the project or the limitations of file formats do not
terial input. allow for one core model both for design and day-
Interoperability between BIM and non-BIM tools light assessment.
remains an issue especially for Design Builder with
regard to 3D geometry input and processing, while ACKNOWLEDGEMENTS
DIVA/Rhinoceros was experienced as the most flex- The study took place during a sixth month training
ible during the process. Ecotect facilitated a rather period in the consultancy firm DGMR, The Hague.
one-sided approach (capable of importing 3D ge- The office permitted for the use of the designs as
ometries), and proved to be an analytical BIM tool were presented in case studies 1 and 2. Thanks are
which supports 3D input even when large amounts also due to Studiebureau Boydens in Brussels, that
of surfaces are involved, without serious problems. allowed for the use of design model 4, and inspired
In conclusion, interoperability as a seamless flow the study with comments and remarks regarding in-
of information is partially supported, while the us- teroperability and simulation accuracy issues in the
ability of the architectural model for performance way that the office experiences them in everyday
evaluations is heavily dependent on a conscious practice. The colleagues of both firms provided sup-
modeling by the side of the designer, long before port during this work, while the administration in
the geometry is exported for the calculation, in any both cases provided consensus for the publication
interoperable file format. of the models.
The hereby presented experiences regarding
modeling and simulating geometries of various sizes REFERENCES
and complexity, prove that further steps need to be Attia, S, Hensen, J, Beltránc, L and De Herde, A 2012, ‘Selec-
taken towards a better integration of 3D modeling tion criteria for building performance simulation tools:
capabilities in the engineering simulation software contrasting architects’ and engineers’ needs’, Journal of
and vice-versa, especially with regard to the preci- Building Performance Simulation, 5(3), pp. 155–169.
sion of the input values and the operability of the 3D Augenbroe, G 1992, ‘Integrated Building Performance Eval-
geometry in the BIM environment. The integration uation in the Early Design Stages’, Building and Environ-
of Building Performance Simulation Techniques in ment, 27(2), pp. 149–161.
the design process, can be facilitated by the selec- Kensek, KM and Sumedha, K 2008, ‘Sustainable Design
tion of a tool that will provide reliable feedback dur- Through Interoperability: BIM and Energy Analysis Pro-
ing the early but also the later design phase. Basic grams, a Case Study’, Cadernos de Pós-Graduação em
guidelines towards such a choice were given during Arquitetura e Urbanismo, 8(1), pp. 42–58.
this study, while addressing the issue of accuracy Tregenza, P 1980, ‘The daylight factor and actual illumi-
through four different case studies. nance ratios’, Lighting Research and Technology , 12(2),
Furthermore, the role of the designer is central pp. 64–68.

CAAD Curriculum
CAAD Curriculum - Volume 2 - Computation and Performance - eCAADe 31 | 543

544 | eCAADe 31 - Computation and Performance - Volume 2 - CAAD Curriculum
How to Teach ‘New Tools’ in Landscape Architecture in the
Digital Overload
Developing emergent design methodologies

Pia Fricker1, Christophe Girot2, Georg Munkel3
ETH Zurich, Departement of Architecture, Chair for Landscape Architecture, Professor
Christophe Girot, Switzerland
http:// www.girot.arch.ethz.ch
1
fricker@arch.ethz.ch, 3munkel@arch.ethz.ch
Abstract. The central theme of the paper is the introduction of hands-on tools showing
the integration of information technology within a postgraduate study program (MAS
LA) for landscape architects. What has already become a part of the discourse in the field
of architecture – generic design – is now also finding more resonance in the context of
large-scale landscape architectural design. If one studies the educational backgrounds
of landscape architects, however, they often do not match the same standard as those of
architects. A solid background in the area of innovative use of information technology,
especially computer-assisted design and CAD/CAM construction is only at a preliminary
state at most universities. The critical arguments in the choice of the selected medium and
the building up of a continuous digital chain stand here in the forefront. The aim is not to
improve the quality of the landscape design based on the variety of the applied tools, but
rather through the sensible use of the said. Reflections as well as questions of method and
theory stand at the forefront of our efforts.
Keywords. Design tool development; computational design research and teaching; new
design concepts and strategies; parametric and evolutionary design.
BACKGROUND
At the Department of Architecture of ETH Zurich, Chair, Planning of Landscape and Urban Systems of
both students and researchers have the newest Professor Grêt-Regamey (IRL). The LVML is equipped
technical equipment and software at their disposal. with special devices for large-scale data collection,
Optimal networking with professionals in the area of for example a 3D landscape scanner with 2 km
construction as well as CAD/CAM production allow range and a drone. To this end, various software and
us to offer courses in the curriculum that allow ex- hardware solutions are combined experimentally in
perimentation with the newest techniques and ma- order to investigate new boundaries of perception
terials. This excellent infrastructure is supplemented and illustration of the built environment (Figure 1).
with the advanced resources of the Landscape Visu- Professional partnerships to the developers of soft-
alization and Modeling Lab (LVML) founded by the ware and hardware solutions as well as experts in
Chair of Professor Girot in cooperation with the PLUS the areas of landscape and urban planning allow for

Figure 1
Generated Data: Combina-
tion of point cloud data from
airborne and terrestrial laser
scanner.
a hands-on examination and implementation in the more and more incapable of understanding and vis-
various research areas. The difficulty in teaching lies ualizing complex landscape designs. Moreover, the
not in the lack of equipment, but rather it is seeing level of expertise in the daily use of digital tools is at
this digital overload as a new challenge in a positive a level that is no longer acceptable for an efficiently
sense. organized office. In order to close this gap, the MAS
LA (Master of Advanced Studies in Landscape Archi-
INTRODUCTION tecture) Program of Professor Christophe Girot was
At present, information technologies are an essen- newly conceived in 2009 in terms of content, both
tial component of design and building construction. methodically as well as didactically. To this end the
Contemporary architecture and large-scale land- former design-specific focus was transferred to the
scape architecture as designed by top offices would learning and use of up-to-date tools.
not be thinkable without them. Without computer-
assisted manufacturing and logistics, modern form The MAS LA Set-Up
language and structural solutions would hardly be The course of studies is divided into themed mod-
realizable. Meanwhile, the software and hardware ules, workshops and one concluding synthesis
involved has become so sophisticated that the stu- module. The modular structure allows a concentra-
dents’ generally increased computer skill levels suf- tion on individual themes, which can be combined
fice for an architectural program at a technical uni- within the framework of an individual project as the
versity. At the present, a heated discussion on the concluding thesis module. The main focus of the
level of education is taking place especially in the program is not the acquisition of new software skills
field of landscape architecture within Switzerland. but rather the integration of cutting-edge modeling,
In contrary to architecture offices, practicing land- visualization and presentation technologies as de-
scape architecture offices, especially those in Conti- sign tools within the field of landscape architecture
nental Europe, criticize that university graduates are (Figure 2).

Figure 2
Testing different ways of
presenting large scale
topographical information by
projecting animated informa-
tion to a CNC milled model.
Each module begins with a phase where new should be applied and recombined to explore new
techniques are learned. In this phase, individual ex- design methodologies in their final project (Hagan,
ercises are connected to current issues in landscape 2008). The concluding module of the postgradu-
architecture. The students are encouraged to rec- ate program acts as a test case for the questions or
ognize global as well as local economic and socio- agenda, which have been defined throughout the
logical demands and integrate them in their designs teaching year.
using and connecting the learned tools. The achieve-
ment lies in the diligent selection and connection of Parametrism – the Solution for All Prob-
the technology with the environment we live in. The lems in Architecture?
critical debate regarding issues of sustainability in Ever since the 11th Architecture Biennale in Venice
conjunction with large-scale design work in urban 2008, is Patrick Schumacher’s postulated concept
landscapes plays an essential role next to the techni- “Parametricism as Style - Parametricist Manifesto”
cal aspects. In the second part of the module, par- all the rage. For many students, complexity equals
ticipants grapple with complex problems, which will quality. The use of parametric tools, i.e. Grasshop-
be discussed during a concluding presentation. The per is often seen as the solution to conceptual prob-
sequence of modules start with modeling and CNC lems. Our desire is to show students solutions and
production followed by visualization, programming, approaches how they can choose the right tools for
GIS, applied progamming and ends with media and the design problem at hand in order to test their
photography. ideas efficiently and unconventionally, realize them,
We challenge the students to go beyond the as well as also later generate suitable formats for the
boundaries of conventional domains and test the construction process. Here the learning of specific
tools in analysis, design, and visualization. The pro- software does not stand in the forefront but rather
grams and different CAAD/CAAM techniques, which the learning of a new way of thinking that under-
the students have become acquainted with in the stands the tools as integrative design tools (Mitchell,
different modules, complement each other and 1990). The sequential structure of the MAS LA pro-

gram allows the students to deepen their individual How to Influence Topographical Data
interests and test the newly learned work processes within a Dynamic Planning Process?
on their applicability to professional practice. In An application influences to a great extent the work
terms of content, classical, traditional themes like processes and therefore the design process. If, for ex-
modeling, visualization, GIS, data acquisition are ample, at any point the designer is aware of statistics
complemented and leading to a new perspective like areas and details regarding the execution of a
by working with experimental themes like percep- building, he or she will be able to orient themselves
tion through film photography and programming to these numbers and take them into account dur-
(Greenberg, 2007; Reas and Fry, 2010) . The goal is ing the development of a design. If these numbers
to create an overview that shows spatially relevant can only be gained through tedious procedures,
aspects in landscape structures enabling the stu- they will only be collected at a few points in time
dents to communicate their future projects within and will therefore have a limited influence on the
real practice in a professional way. Together with ar- design. Even more so than numbers, visualizations
chitects, urban planners and city authorities, gradu- of data can give the designer a feedback as long as
ates will be able to illustrate and communicate their these are created dynamically and without much
design ideas professionally. effort. If these presentations contain data on which
the design should orient itself, a loop is created ide-
DEVELOPMENT OF SPECIALISED TOOLS ally from the designer’s actions and the presentation
What are the tools that adequately serve current of their ramifications.
landscape architectural trends and how can they If the presentation of data influences the de-
be conveyed? The past years have shown us that signs, in that they are permanently in front of the
programming within architecture has become as designer’s eyes, the designer must determine how
commonplace as CAD drafting (Wanner, 2010). these illustrations look like. The students have to
Within landscape architecture, urban planning but learn how to program in order to maintain a dy-
also building construction, for example, parametric namic planning process and in order to control his
designs are often the only solution to dealing with design.
complex form language. To this end we would like A further development of digital tools is re-
to use the computer to realize projects that would quired, which allows the interaction between design
not be possible using conventional methods. This and data. This has to come from the designers them-
requires the further development of digital tools, selves, because they have to determine to which
which allow for the subsequent design and work- presentations they choose to react during the de-
ing with the extracted information (Kolarevic, 2003). sign process.
Programming has the advantage of solving complex Taking these problematics into account, the final
tasks, accompanied by the risk of sacrificing the in- project of the MAS LA Program shows one example
tuitive abilities of a designer. Our goal within the of implementing this thinking within the process.
MAS LA program is to reduce the technical hurdles The task focused on the excavation material of a
and apprehension towards programming in order to highway tunnel, which had to be ‘concealed’ within
first reach an understanding of its necessity within the context of a landscape design. When designing
landscape architecture. (Bohnacker et al. 2009). To in the conventional sense, our students could not
reach this goal we have developed specific tools, like make any concrete statements on the actual cubic
the “Sandbox Tool” for the special need of landscape meters involved. Only with the integration of an
architecture. own script they were able to comment at any time
on the already built amount of material and thereby

Figure 3
Depthmaps for Terrain(1),
Cones(2) and Top View on the
Result(3) by student: Chrisine
Baumgartner.
maintained absolute control over the impact of de- ied processing possibilities. The user can fall back on
sign changes. the entire toolset of image manipulation software
The Processing script of MAS LA student Chris- in order to influence the height information. Here
tine Baumgartner allows one to manipulate the comes in the Processing script itself tools like brush-
height of the cones and the terrain lying underneath es for localized work as well as global ones, such as
independently from one another (Bader-Natal, the blur filter. The application allows a view over the
2010). The heights were saved in two separate depth area with an orbit camera. Any current condition can
maps that represent the relative heights through be saved with the pressing of a button as an image
gradations of gray. The height of the cones is always and the model as a DXF to be used later on for gen-
taken relative to the terrain (Figure 3). The sum of erating a visualization (Figure 4).
the cone volumes can be calculated at any time and
in conclusion adapted so that it corresponds with Theoretical Programming
the volume of the excavation material. The simple Going one step further, we introduced in spring
data modeling of the heights in the form of gray- 2013 the first time an intensive 3-day workshop for
scale images allows for simple data saving and var- the MAS LA students called “Theoretical Program-
Figure 4
Using the generated data to
automate the visualization
process (student: Christine
Baumgartner).

Figure 5
MAS LA students using
CRC-cards to communicate
their set up of the project.
ming”. The overall objective of this workshop was on the module with the students.
the critical reflection of the implementation possi- The set-up of the workshop was a combination
bilities of programming, within the real practice as of lectures and role play in the loneliness of a moun-
an landscape architect. Out of our experience these tain hut in the swiss alps. Within groups of 4, the stu-
projects often fail due to misunderstanding and dents were asked to communicate a complex design
wrong expectations from both sides. problematic to a programming consultant. The stu-
After the students find an initial foothold in dents have been learning different techniques and
programming with the module “Programming methods how to bridge and communicate a design
Landscape” this knowledge was deepened in the to an IT company (Figure 5).
subsequent module “Applied Programming” where The result of this experimental workshop was
first applications were searched for within known surprisingly positive. The students understood
CAD workflows. The results of the past few years, through this playful attempt the problematics and
however, have shown us that the students have no could define potential application fields of pro-
problem with creating their own programs only they grammed solutions in their design. The feed-back
often do not have the fundamental understanding of the students pointed out the importance of such
for their necessity and potential within professional reflections and training for the real practice. With
practice. the 3 days they learned different techniques, like the
In order to make a convincing appearance in usage of “UML”, “CRC-cards”, “class diagrams”, “story
a professional context, it is necessary to be able to board”... to prepare all the desires of the landscape
speak the language of the other profession. To this architects, in order to have the IT company program-
end, we supplement our teaching team with a com- ming the software.
puter specialist who spans a continuative theoreti-
cal background. Programming paradigms of greater The Sandbox Tool
scope, for example object-oriented, automata-based Another aspect that we are investigating at the mo-
and genetic programming will also be presented ment is the simplification of workflows through the
(Hight, 2008). The students become acquainted with unconventional linking of existing software and
concepts such as spring systems, shape grammar, hardware. What happens when traditional manual
Lindenmayer systems and agent systems. techniques are combined with state-of-the-art
At the same time, we take advantage of this CAAD/CAAM technologies that are adapted to the
short but intensive time together in order to reflect workflow? Are the students accepting this simple

Figure 6
In order to understand the
complex transformation
processes of water landscapes,
the student will each be using
modelling-sand, realtime
scanning stations, and a
specially made software to
give feedback.
tools as part of their design workflow? We developed our own software that allows
Within the framework of a case study, an exist- one to generate two-dimensional analyses and visu-
ing topography is created in a milled negative mod- alizations directly. Different analytical methods have
el. The students could use this as the formwork, in been programmed to supply the students analysing
order to create the same point of departure at any their design proposals:
time. The knowledge of this possibilites frees the We supply the designer with an elevation map,
students to work in a very experimental way. All including a separation of the gradients above and
manual techniques are allowed in the modeling of beneath the water table. Additional features include
the designs. If the designer wants to capture a state a visualization of the slopes and normalized lighting
of the model the student makes two photos from conditions for all the models (Figure 7).
different viewpoints, which are then transferred As a result, the students never work on a com-
through photogrammetry into a computer network puter 3D model but rather focus their work directly
model where they can be digitally stored and ana- on design statements in plan. The ‘analog’ sand
lyzed (Figure 6). model thereby maintains its significance and be-
Figure 7
LandscapeAnalysis_flodding:
3 Waterlevels composed with
Contour Curves from different
Layers of the Analysis Tool.
Elective Course Students 2012.

comes an interface for analysis and visualization of the design process already with the gathering
through the associated digital workflow. of necessary actual data. Through the use of long-
According to our observations, the students range terrestrial scanners, one is able to gather data
are very creative when working on the sand model. in the form of point-clouds that can be successively
The milled negative form allows them to return to used in combination with aerial photography taken
the original state at any time and they can save any by a drone to create specific and up-to-date con-
variation of the design with only two photographs. texts for design analysis and the planning process.
A real physical model cannot offer these qualities. This first step of the design is especially important
On the other hand, the physical model has clear ad- for the general understanding and individual ad-
vantages over a digital one: one is able to work with vancement of the various design tools. In order to
one’s own hands and with any kind of tool imagina- gain this general understanding, students must be
ble. When one removes material, one is aware of the given a broad overview of current techniques and
volume in his or her hands. One inevitably gains a research fields in information technology. Through
feeling for how much earth is being moved. In addi- simple and short exercises the manipulation and
tion, a group of students can work on the model si- use of different software modules are to be tested.
multaneously and be able to discuss from any point More in-depth courses und workshops offer the
of view. possibility to elaborate the rudimentary knowledge
based on the working out of current problems and
CRITICAL THINKING through the building of full-scale models, in order to
While most universities are extremely well- be able to test them until the very end in the form of
equipped, the students often lack reflection as to a semester design.
the application potential of these digital tools. This
often leads to an uninhibited combination of all ACKNOWLEDGEMENTS
available tools. The use of generative components The research and teaching work described in this
like Catia, Rhino, Grashoppper, etc. was only avail- paper are the result of a very committed teaching
able for experts until around 10 years ago. In the team at the Chair of Christophe Girot, special thanks
meantime, each student has access to these pro- to all of them, especially to Ilmar Hurkxkens, James
grams and is able to use the software to a certain de- Melsom and Pascal Werner.
gree without any official schooling (Mertens, 2010). The authors would like to thank the MAS LA stu-
The designs become more complex and are often dents of 2010-2011, 2011 – 12, 2012 -2013 as well as
no longer controllable from a certain stage, and the the students of the elective course Landscape De-
quality suffers under the unfiltered use of the tools. sign, “Constructing Landscape”, for their dedicated
Therefore a rethinking needs to take place within participation in the program.
design methodology so that not the learning of spe-
cific software stands in the forefront but rather the REFERENCES
learning of new ways of thinking that exploit the po- Bohnacker, H, Gross, B, Laub, J and Lazzeroni, C 2009, Gen-
tential of these tools in integrative design (Leinonen, erative Gestaltung, Verlag Hermann Schmidt, Mainz.
2010). Greenberg, I 2007, Processing, Creative Coding and Compu-
On the other hand, we see the great challenge tational Art, Friends of Ed, Springer Verlag, N.Y.
in the sensible linking of these tools. The concept Hagan, S 2008, Digitalia, Routledge, N.Y.
“Emergent Design Methodologies” spans a top cur- Hight, CH 2008, Architectural Principles in the Age of Cyber-
rent field, which has the goal of using information netics, Routledge, N.Y.
technology to support design as well as an analysis, Kolarevic, B 2003, Architecture in the Digital Age, Design and
design, and production tool. We see the beginning Manufacturing, Spon Press, N.Y.

Leinonen, T 2010, Designing Learning Tools, Aalto University, O`Reilly, Sebastopol, CA.
Finland. Wanner, A 2010, Processing. School of Interactive Arts and
Mertens, E 2010, Landschaftsarchitektur visualisieren. Technology, Canada.
Birkhäuser Verlag AG, Basel-Boston-Berlin. Bader-Natal, Ari: 2010: “Studio Sketchpad - Processing on
Mitchell, W 1990, The Logic of Architecture, MIT, USA. Etherpad”. Available from: Open Source Repository
Reas, C and Fry, B 2010, Getting Started with Processing, <http://sketchpad.cc> (accessed 1 December 2012).

150 000 – Parametric Control of PET Bottle Structure
Kateřina Nováková1, Lukáš Kurilla2, Henri Achten3
Faculty of Architecture CVUT Prague
1
http://www.katerinanovakova.cz, 2caad.kurilluk.net, 3www.molab.eu
1
bruhokat@fa.cvut.cz, 2mail@kurilluk.net, 3achten@fa.cvut.cz
Abstract. In this paper we describe the pedagogical and methodological approach to

a parametric project and workshop for the design of a tower which consists of 150000
PET bottles. The ultimate goal of the project is to actually realize the PET bottle tower;
therefore the constraints on the projects are very strict. Additionally, because of the large
number of bottles to be used in the design, the problem lends itself well to a parametric
approach.
Keywords. Pedagogy; garbage architecture; workshop; parametrics; PETower.
GENERAL FRAMEWORK
This paper is focused on the suitability of paramet- ever, they could also see extreme approaches such
ric design tools for the generation of a tower de- as ready sketch design followed by modeling of the
sign that consists of 150000 PET bottles. The tool is same in computer, the use of parametric tool to ac-
taught and used in a 6-day workshop, and is embed- tually generate the design and the switch back and
ded in the context of an experimental collaborative forth among the two tools. Our report differs in in-
design studio between two faculties of architecture tegration of physical material tests into the task and
of two different countries. the tested group regarding rather novice students.
The problem lends itself well to a parametric ap- Before the workshop we invited expert users
proach, as it concerns a composition of many similar (colleague teachers of those programs) of various
units and students need to have control over the CAD programs (3DStudio Max, Revit, AutoCAD,
total number of components. However, parametric ArchiCAD, and Rhino) to create an arbitrary three-
tools may not be sufficient and the only method of dimensional structure of 150000 similar objects. The
designing suitable in all phases of this task. We ob- choice of object and composition were left com-
served that our students naturally hand-sketched as pletely open to the experts (Table 1).
well. The precedence we could find in Sanguinetti Manipulation of the designed object failed due
and Abdelmohsen (2007), where the authors suc- to the computer capacity in three of four programs.
cessfully describe integration of sketching and para- From their feedback we selected Rhino in combina-
metric modeling in conceptual design task. How- tion with Grasshopper as the most promising tool in
Table 1 3D studio Max Revit Autocad Rhino

Criteria of choice of the design Possibility to generate yes yes design unsatisfactory yes
tool. Numeric control yes no no yes
Manipulation of model impossible impossible impossible yes

the project. Compared to the other programs, it was opment of the plans.
able to control the high amount of units and also PET bottles are not an uncommon material for
provided numeric control over the design. experimentation in architecture. Several projects
We followed the scheme of a traditional archi- exist where PET bottles are used as infill material in
tectural design process, where hand drawing comes facades. The notable EcoARK [3] building utilizes 1.5
in the beginning and manufacturing of a model is million specially fabricated PET bottles. The Ameri-
in the end of the process [1]. Because of the experi- can architect Michael Reynolds uses waste and PET
mental nature of the material and the specific na- bottles in his projects [4]. Experimental structures
ture of PET bottles it was important that the model with PET bottles are presented for example by Baer-
was in 1:1 measure. The workshop was open for stu- lecken et al. (2012).
dents at any level of expertise. During the workshop
we observed the work flow of the students. METHOD OF TEACHING
The first two to three days we were teaching
Grasshopper. The students discovered that paramet- First semester: problem exploration
ric design tools not only generate interesting and The first semester we led a collaborative design stu-
cool pictures, but they are also an inevitable part of dio between our faculty and the Faculty of Architec-
the design process to support the conceptualiza- ture at ETH Zürich. For this project we developed
tion of the design. Last but not least hand drawing and tested a new software called ColLab sketch,
played a significant role during the design process. which was implemented in both faculties’ media
facilities [6]. The advantage of such a collaboration
DESCRIPTION OF THE PROJECT was that part of the team was situated at the site, so
Both setup of the workshop as a part of the experi- that they could supply the team with maps, photo-
mental design studio are informed by the context graphs and sketches of the building place.
of the project. The project of the experimental de- The major problem the students faced was visu-
sign studio is located in Zurich, Switzerland. The city alization of such an amount of units such as 150 000.
of Zurich organizes a very large festival every three Figure 1 shows the total number of about 150 000
years, called the Zürifäscht. In the current edition, to bottles in a flat area.
take place 5-7 July 2013, the festival runs under the In the first semester a number of 3D designs
theme of recycling. The theme has three subtopics: were made with PET bottles starting by coupling
Recyclata, Recyclodge and Recyctower (PETower). single bottles into bunches wrapped in food foil,
The Recyctower (PETower) is the subject of the ex- which worked well but was not of convincing visual
perimental design studio [2, 5] in our institution and quality. Also the stability and reliability of such a
in the workshop reported in this paper. The task is structure was doubtful. Some informal tests were
quite challenging: the tower must be built from re- done to verify the strength but without reliable re-
cycled PET bottles and contain 150 000 pieces. The sults. Finally, having an overview of number of the
structure must be self-standing and lit. It must be single bottles within the designed tower was almost
built and demolished in no more than three days. impossible, together with the limited design expres-
For the construction process school children must sion. Apart of the communication software which
be involved to assemble the units of the tower. For students had to learn we had not recommended
transportation purpose the units may not be bigger any software or design method and we had left their
than 2x2x8 meters. choice open (Figure 2).
Originally we were contacted to advise on the The final designs resembled towers, but it could
PETower design. It seemed logical for us that stu- not be proven that they could actually be built.
dents of architecture would take part on the devel-

Figure 1
Visualization of 150 000 PET
bottles made by a student.
Second semester: focus on design tools dents could re-use the earlier gained knowledge
In the second semester the project was developed from their colleagues and shift the start point, so
at our faculty only, but we had consultation sessions that we could start designing earlier. We also intensi-
with advisors from the partner university. Different fied the switch between design and material experi-
students than from the previous semester took part ence according to the scheme in the table (Table 2).
in this studio. In the first semester we mainly fo- Most of the time during the semester was de-
cused on feasibility of the project and did informal voted to laboratory tests of the PET bottles, where
material tests and based designs on these findings. students got the reliable values of the load capac-
The second semester [7] was supported by a 6 day ity of bottles and pull capacity of the cap, which we
long workshop devoted to information visualization report on at CESB13 conference in Nováková et al.
tool lectures and manufacturing. The results were (2013). They also worked on analysing the possi-
regularly communicated over distance to the advi- bilities of connections using PET materials in a sheet
sors in partner faculty and other parties involved in form. We noticed that the designs of the previous
the project. semester were limited to regular shapes such as
The scheme of teaching varied in both semes- cubes, hexagons or cylinders. In the second semes-
ters. In the first semester we allowed the students ter the students obtained more freedom for their
to design only after analyses were ready: analyses PET bottle designs. Towards the end of the second
– design – fabrication. In the second semester stu- seminar (during the workshop) we encouraged stu-
Figure 2
Final projects from the first
semester.

Activity Hours Description Table 2
Lecture (film) 2 Motivation of the students, primary introduction of topic Scheme of activity switch dur-
• Waste land ing the semester.
• Wall I
• Midway [8]
Design 4 Initial ideas, concepts, hand sketches
Measurement tests 2 Tests of PET bottles in Civil Engineering (CE) material lab
Design 4 Specifying design strategies
Excursion to PET bottle 8 Learning material generation principles
production plant
Design 6 Implementing knowledge from tests and excursion into design
Cap tension tests in CE 2 Tests of the screw and cap connection, the strength of the screw.
laboratory
Pressure and behavior 2 Tests on manufactured building blocks in CE laboratory
tests in CE laboratory
Workshop - tools 24 Learning of parametric tools
Workshop - prototyping 16 Prototype fabrication
Design 4 Final design + presentation
dents with correlating ideas to group and work out maximum size 2x2x8 m3). We found that they could
one project. Half of the students formed teams of easily change their design towards the initial hand
two or three people (Table 3). sketch without loosing the awareness of the num-
During the workshop students took their initial bers. Consequently we allowed the students to play
sketches and tried to model them within Rhino. In with their newly acquainted skill. By allowing them
the first two days they learned basics of grasshop- the “play phase” within the teaching hours, the stu-
per and understood parametric thinking, on the dents were more motivated to experiments, deepen-
third day we introduced several ways of generating ing their understanding of parametric modeling and
tower-like objects (Figure 3). got quicker feedback from the teachers. Together
After the initial phase they constructed their vir- with this parametric attitude to the problem, the 4th
tual models with the data constraints (150 000 bot- and 5th day was devoted to building the actual pro-
tles, minimum 20 meter height, one component of totypes of the components. Students not only had to
Figure 3
Examples of different pos-
sible principles for tower
generation. Combination
of geometrical shapes and
mathematical formulas.

Table 3 Students' sketch Grasshopper Physical model of the Project Evaluation according
Table of project developments. model building unit to feasibility
Petr yes no PETower 10
Simon yes yes worked
Lenka yes no Wall 3
Adam yes yes failed Tripod 9
Petr yes yes failed Mobius 8
Juraj yes no Lighthouse 7
Sori no no Twisting Tower 6
Ondrej yes yes failed Atomium Not feasible
Ivana no no i Not feasible
Jiri yes yes worked
Karel yes yes worked T 2
Honza yes no
Maria yes yes failed Swiss cross
Leila yes yes failed 4
David yes no no Not feasible
Peter yes no Growing Tower 5
Jean yes yes worked Plasticienne
Pauline yes yes worked Plasticienne 1
Vera yes no Pentagon Tower Not feasible
collect the bottles, but also tried to set the connection OUTPUT OF THE SECOND SEMESTER
between them according to their previous research. The projects of the first semester did not result in
We could observe students sitting by the screens feasible designs which could be actually construct-
and sketching their technical ideas on the paper. This ed. We assume that the main reason for this was the
workflow programming-sketching was efficient for lack of expertise in sophisticated CAD tools.
them and it proved to be the fastest (Figure 4). We collected several physical prototypes of
Because the material was experimental, not PETower building modules together with connec-
all assumptions and designs were successful and tion strategies (Figure 5).
students had to change their virtual models again In contrast to the first semester we did see thir-
according to the real scale component prototypes. teen feasible projects in the second semester, which
Students grouped again in this phase of component were consequently communicated to several parties
generation. in Zurich (municipality, festival organizers, potential
Figure 4
Students programming and
sketching designs simultane-
ously.

Figure 5
Building blocks from the
left: circular type (Wall),
stick type (Swiss cross,
Mobius, Tripod),3D panel
(Plasticienne).
sponsors). Three projects were highly realistic and computer modeled in the middle phase of design-
one of them was accepted as a realization project ing, but all the time of project development they
(Figure 6). were sketching even when sitting by the computer.
Especially in the phase of moving towards construc-
CONCLUSION tion we could see simple drawings of connection
We collected all sketches, made screenshots of the details or patterns of assembled units. The paramet-
developing projects and made documentation of ric tool proved to be very important. Some students
the 1:1 models. We observed that students not only tried to develop the project with other CAD tools,
Figure 6
Plasticienne: 23 m high, 148
000 bottles.

Figure 7
Project development
documentation: two initial
sketches made by two indi-
viduals matched, therefore
they decided to cooperate and
delivered feasible project.
but failed. Thus they responded very enthusiasti- struction methods with PET bottles was on the other
cally to the parametric tool when they saw how the hand made on paper and parametric tool turned out
design remained flexible while keeping also numeri- to be unsuitable. Hand sketch helped developing
cal control. As we observed, teams of two or three initial designs, visualizing partial ideas, generating
people were able to deliver results, which fulfilled all details and clarifying technical solutions. In general,
conditions of the project, while individuals did not we feel that it is necessary to introduce similar work-
progress beyond trials and failures. For all students shops focused on parametric design in the begin-
it was interesting to see how much the parametric ning or middle of the design studio together with
tool enabled them to deal with the real problem. the same focus on hand sketch, where we believe in
Furthermore they were able to follow their initial better impact of using of these tools directly in the
hand sketch graphics giving it exact numerical and design studios by project development.
structural control (Figure 7).
For our experimental design studio and this FUTURE WORK
special task using parametric tools proved to be of In the next semester of the experimental design stu-
critical importance. Also the hand sketch technique dio we shift focus to the design of shelters and small
seems to be crucial. Parametric tools enabled stu- service structures made of PET bottles for the same
dents to experiment with the tower design, while festival event. We would like to observe the direct
also keeping control over the various constraints impact of using parametric tools on final designs
that apply to the project. Exploring various con- and the role of sketching in this process.

ACKNOWLEDGEMENT Sanguinetti, P and Abdelmohsen, S, 2007, ‘On the Strategic
We kindly thank the students of both semesters and Integration of Sketching and Parametric Modeling in
the workshop, all external consultants and the staff Conceptual Design’, Georgia institute of Technology in
of the Civil Engineering material lab. Atlanta, USA.
REFERENCES [1] http://www.fa.cvut.cz/Cz/Publikace/ZakladyArchitek-

Baerlecken, D, Reitz, J and Duncan, D 2012. ‘Junk: Reuse of tonickehoNavrhovani
Waste Materials’ in H Achten, J Pavlicek, J Hulin and D [2] http://www.facebook.com/groups/285845308192762/
Matejovska (eds), Digital Physicality - Proceedings of the [3] http://www.miniwiz.com/miniwiz/projects/ecoark
30th eCAADe Conference - Volume 2, Prague, pp. 143- [4] http://www.garbagewarrior.com
150. [5] http://195.113.84.161/~hugo/collab/
Nováková, K, Kurilla, L and Achten H, 2013, ‘PET bottle struc- [6] http://www.facebook.com/ExperimentalAtelier
tures development in collaborative design studio’, in [7] http://www.molab.eu
Central Europe for Sustainable Building, Prague, Czech [8] http:// www.midwayfilm.com
Republic.

Identifying Cognitive Operations of Conception Implied
in the Uses of Parametric Modeling in Architectural
Design: Toward Pedagogical Tools
Aurélie de Boissieu1, François Guéna2, Caroline Lecourtois3
MAP-maacc, Ecole Nationale Supérieure d’Architecture de Paris la Villette, France
http://www.maacc.archi.fr
1
aurelie.deboissieu@maacc.archi.fr, 2fguena@maacc.archi.fr, 3clecourt@maacc.archi.fr
Abstract. The research presented in this paper aims at identifying the cognitive
operations implied in the uses of parametric modeling in architectural conception. The
uses of parametric modeling in architectural design remain emergent and marginal.
How can we teach these practices? The identification of the main cognitive operations of
conception allows us to propose accurate pedagogical objectives. This paper presents: the
research methods employed, the results achieved and propositions for pedagogical tools.
Keywords. Parametric modeling; architectural conception, CAAD curriculum,
architecturology.
INTRODUCTION
Parametric modeling is part of computer aided de- to identify the characteristics of the cognitive opera-
sign process of industrial sectors, such as automo- tions of conception implied in the uses of paramet-
bile or aeronautic, for over three decades. For a few ric modeling in architectural conception. We interro-
years architectural sector has carried out parametric gate here architectural “conception” that we define
modeling. as the cognitive aspect of design activity.
Visual programming languages as Grasshopper This paper presents: the research methods em-
[1] have certainly something to do with this amazing ployed, the results achieved and propositions for
and growing adoption by architects. Popular among didactic tools.
students and professionals, this plug-in of Rhinoc-
eros 3D modeler enables them to build paramet- METHODOLOGY
ric models without any programming or scripting
knowledge. However, the uses of parametric mod- Context of the research
eling in architectural conception remain emergent Analysis of design practices in architectural con-
and marginal. texts (our as well as Kolarevic, Picon or Lindsey ones
How parametric modeling is involved in archi- (Kolarevic, 2005; Lindsey, 2001; Picon, 2010)) shows
tectural conception process? How architects can be that parametric modeling is linked to various com-
trained to parametric modeling and visual program- puter assisted tasks: complex form finding and rep-
ming language? These two issues must be clarified. resentation, evaluation, optimization, fabrication,
In order to address these questions, we search communication, collaboration, etc. We observed

that most often parametric modeling actors are not ing measurements to an artifact. These operations
the architects who conceive the projects. Parametric are described in terms of dimensions, references and
modeling requires expert skills and knowledge that relevances. Dimensions refer to the measurement
most architects have not yet mastered. How archi- supports, i.e. to what the measurement is given. Ref-
tects could be assisted for parametric modeling dur- erences refer to the viewpoint from which measure-
ing the conception process and how they could be ment is given. Relevances refer to how measurement
trained are thus crucial questions. is given. The dimensioning operation is an elemen-
Different studies interrogate the role of para- tary operation of conception that consists in linking
metric modeling in architectural design (Wood- a dimension to a reference through a relevance (Bou-
bury, 2010; Davis et al., 2011; Chein and Yeh, 2012). don et al., 2000, p.154).
The laboratory MAP-maacc interrogates the uses of From these architecturological concepts, we
parametric modeling in architectural conception have analyzed our corpus and identified several
thanks to a cognitive approach. This paper synthe- cognitive operations of conception by using the
sizes researches carried out in the context of a PhD method of Applied Architecturology developed by
in this laboratory. The purposes of these researches Lecourtois (2011).
were to describe the use of parametric modeling
in the architectural sector in order to identify the MAIN COGNITIVE OPERATION IDENTI-
cognitive operations of conception involved. This FIED
knowledge on the cognitive operations of concep-
tion allows developing methods and didactics tools. Distinction between parametric model
conception and architectural conception
Corpus Before presenting our architecturalogical research
In order to identify cognitive operations of concep- methods, we need to explain how we proposed to
tion implied in the uses of parametric modeling in interrogated the use of parametric modeling in ar-
architectural conception, observations and inter- chitectural conception. We didn’t interrogate para-
views were performed. We have analyzed concep- metric modeling as assistance for a design, but as
tion practices carried out: in professional contexts a conception activity in itself. We proposed to use
(at Foster and Partners, Hugh Dutton Architectes the architecturalogical apparel to interrogate the ar-
and Ateliers Jean Nouvel among others); and in de- chitectural conception as well as the conception of
sign studios of schools of architecture at Vienne (at parametric model. Does the conception of paramet-
the Universitat die Angewandte) and Paris (at the ric model can be distinguished of the architectural
Ecole Nationale supérieure de Paris Malaquais, Ecole conception? How these two conception process ex-
Nationale Supérieure d’Architecture de la Ville et des change or influence each other? These are the issues
Territoires among others). we interrogated in this research.
Methodology Elementary operations of conception

In the cognitive science field, very few approaches Our analysis shows that the elementary operations
interrogate the conception and especially the trans- of conception built by architecturology can describe
formation of the conceived artifacts (conception is the conception of a parametric model. “Slicing”, “ref-
mostly interrogated as activity). Architecturology is erencing” and “dimensioning” are operations allow-
a scientific field of research on Architecture that al- ing the description of the conception of parametric
lows it (Boudon et al., 2000). This research field pro- models as activity of attribution of measurements.
vides a scientific language for describing cognitive The “slicing” operation is the activity led by
operations of architectural conception by which giv- someone when he decides to conceive a specific

part of an object. For example, when a conceiver ent uses: parking, circulation, playing ground, sitting
decides to model a grid of circles and to drive the place, etc. (Figure 1).
diameter of the circles thanks to their proximity to a In the parametric model developed in this pro-
specific point, he mentally does a “slicing” that focus- ject a surface is used as a reference to create a grid
es the attribution of measurement on the diameters of points (Figure 2). Each of these points is moved in
of the circles. the Z direction in function of its distance to a set of
The “dimensioning” is the operation by which curves. The more the points of the grid are closed to
a conceiver gives a measure to an object thanks to one curve of the set, smaller is the Z translation. All
an activity linking a relevance and a dimension. For the Z translations are bounded by a maximum and a
example, in the precedent case of the grid of circles, minimum values defined by the students.
when the conceiver decides to measure the circles The surface on which the point’s grid is created
thanks to their distance to a point, he is doing a “di- corresponds to the easement of the public parking
mensioning”. We have developed more specifically on which the “topography” is supposed to lean. The
this operation in a previous article (de Boissieu et al., set of curves that controls the Z translations repre-
2011). sents the main circulations ways and some parking
The “referencing” operation is implemented by a places. The minimum Z translation value is zero to
conceiver when he chooses a context or a family of allow an easy access from the “topography” to the
relevancies to attribute measures. For example, still street. The maximum Z translation value is defined
in the case of the grid of circles, if the conceiver de- not to mask the view from the first floor apartments
cides to link the measure of the circles in an exact of the surrounding buildings.
equality with the distance of each circle to the spe- // the “dimensioning” of the original grid
cific point, then he sets up the attribution of meas- points
urement in a specific mental world (a geometric • within an architectural conception:
one) which proceeds from “referencing”. When the students have decided to conceive
If the analysis shows that these elementary op- their topography on the parking easement,
erations of conception are useful to interrogate par- they have done a “slicing” operation linked with
ametric models conception, the analysis also shows a “dimensioning” one: the measure of the “to-
where the uses of parametric modelling in architec- pography” fit with the existent public space.
tural conception overflow these elementary opera- • within the conception of a parametric model :
tions of conception. It is also observed in the conception of the par-
ametric model: to dimension the grid of point
A case of conception described thanks to in Grasshopper (Figure 2) a surface on Rhinoc-
the “elementary operations” of conception eros is used as an easement for code.
Let’s take a specific example from a student’s work, // the “dimensioning” of the Z translations
in purpose to describe the cognitive operations • within an architectural conception:
that can be observed. The project “Topographies” The students conceive the “topography” pro-
has been developed by Aurea Rodriguez, Pablo ject to allow different uses such as : circulation,
Gancedo, Samya Pelloquin and Mathias Saboya in playing ground, parking spaces: the topog-
the course “Architecture Paramétrique” led by Na- raphy of the project is dimensioned from the
dir Tazdait and Francesco Cingolani in 2012 at the wanted functionalities of the project. For that,
Ecole Nationale Supérieure d’Architecture Ville et the slopes of the topography are conceived to:
Territoire. This project is an urban installation in the allow an easy walk, separate a parking place
Parisian suburb “Cité des 3000”. A topography is cre- from a playing ground zone, etc.
ated to animate the public space and to allow differ- • within the conception of a parametric model :

Figure 1
« Topographies » picture of the
project from the students Au-
rea Rodriguez, Pablo Gancedo,
Samya Pelloquin and Mathias
Saboya (Ecole Nationale
Supérieure d’Architecture de
la Ville et des Territoires, Nadir
Tazdait, 2012).
The Z translations of the grid points are linked Rajeb had identified some operations implied in
to a set of curves. The curves are linked to the collaborative conception that not give directly
main ways observed in the existing public measurement to an object. She formalized these op-
place and to the wanted parking places. The erations as “pragmatic” one (Ben Rajeb, 2012, p.281).
Z altitude of each point is dimensioned to be The following operations that seem to be implicated
close to zero in proximity of a set of curves. in the uses of parametric modelling in architectural
In these different operations we can observe conception are kind of pragmatic operations.
that we can distinguish operations of conception Operations of collaboration: In our analysis, we
of parametric model and operations of architectural observed the occurrence of two pragmatic opera-
conception. But we can also observe that these dis- tions identified by Ben Rajeb: the operation of “pool-
tinctions show how intricate and porous they are. ing” and the operation of “interpretation”. The
The links and exchanges built between these two “pooling” operation is an operation by which col-
kinds of activity (conceiving the architecture and laborators with different point of view and different
conceiving a parametric model) are interrogated in expertises, share information in purpose to attribute
our research in terms of third operations: the “prag- measurements to an object (Ben Rajeb, 2012 p.286).
matic operations”. This operation operates in the use of parametric
modelling in architectural conception when differ-
Not an operation of architectural concep- ent collaborators (for example architects and model
tion either an operation of parametric manager or parametric design experts, etc) share
model conception … knowledge about the projects (the architectural in-
“Pragmatic operation”: In her research, Samia Ben tentions, the necessity or potentiality of a paramet-

Figure 2
« Topographies » picture of the
parametric model.
ric modeller, etc) to link the architectural conception ty, etc.). It operates in parametric modelling among
and the parametric model conception. The “pooling” other when an architect builds an understanding of
is an operation by which a conceiver translate and the potentiality of the modeller to propose a spe-
negotiate his meaning in purpose to communicate cific way to conceive a space. It can be observed for
it for a collaborator. example in the imaginary built by Frank O. Gehry
The operation of “interpretation” is an opera- about the parametric modeler CATIA (Lindsey, 2001).
tion by which a conceiver gives a personal mean- Operation of translation in parametric geometry:
ing to their collaborators discourse and information The translation in parametric geometry is a prag-
(Ben Rajeb, 2012 p.285). It operates for example matic operation specifically observed in the uses of
when an expert of parametric modelling interprets parametric modelling in architectural conception.
the discourse on architectural intention in purpose By this operation, a conceiver shift from one system
to define constraints or parameters of a parametric to another (from an architectural system to a geo-
model. The “interpretation” is an operation by which metric and parametric one and reverse). For exam-
a conceiver gives a personal meaning to information ple, in the case “Topography” previously described,
shared by a collaborator. we can observe a translation from an architectural
Pooling and interpretation are operations aim relevance “answer with different slopes to the want-
at building some “référenciel opératif commun” (de ed uses” to a relevance for the parametric model “di-
Terssac and Chabaud, 1990) that we can observe in rectly associate the Z positions of the grid points to
shared relevancies and references. a set of curves that position in the space the wanted
Elaboration of cognitive representation of the uses”.
tools: From the case analysis, an operation of elabo- Hugh Whitehead and some of the actors of the
ration of cognitive representation of the tools can be SMG and ARD teams at Foster and Partners talk
identified. These representations are based on inter- about such an activity of “interpretation” (Whitehe-
pretations of a specific tool (its potentiality, difficul- ad, 2009; Freiberger, 2010). We use the term “trans-

late” in this research because if interpretation is TOWARD PEDAGOGICAL TOOLS
needed (as giving a personal meaning [2]) it seems The results of the research presented above, led us
that this last activity implies to transpose from one to propose didactics tools for assisting architects to
system to another one [3]. parametric design. These didactic tools are: accurate
These different operations show that if ac- pedagogical objectives for parametric modeling
tivities of architectural conception and activities of training and training resources.
parametric model conception can be distinguished
for the analysis, they are also porous and intricate General training objectives
thanks to different third operations. These opera- Parametric modelers are constantly updated: new
tions are, to some extent, already proposed by archi- versions are regularly proposed as well as new plug-
tecturology such as operation of collaboration (with in. Furthermore, research is active in this field: new
the operations of “pooling” and “interpretation”) and explorations are regularly published. Practices of
the interpretation of the conceiver about the tool. parametric modeling seem to be in a demanding
But there is also specific operation to the uses of situation: fundamental knowledge is needed (as we
parametric modelling in architectural conception: formalized above), as well as a continuous update of
the transposition of relevance or reference from a the knowledge and practice of the field.
system to another one (the operation of translation To answer to this specificity, a training of para-
in parametric geometry). metric modeling in architectural conception should
focus on :
Logical operations and knowledge • fundamental knowledge,
Logical operations: Logical operations of induction • learning abilities in this specific field,
and verification can also be observed in the uses of • relevant didactic tools that can be used to im-
parametric modelling in architectural conception. prove specific issues even when the training is
Some induction operations appear when a conceiv- over.
er establishes few cases of his object to then induce General skills that training in parametric mod-
some rules or a system able to define these differ- eling should aim to develop are:
ent cases. An operation of verification appears when • abstraction (logical and geometrical);
a conceiver tests his parametric model in different • organization and anticipation;
particular cases. By induction, the conceiver thinks • participation to a community of amateurs, a
from the instances to the parametric model. In veri- network linked to the students interest, work
fication, the conceiver thinks from the parametric as a team;
model to the instance. • curiosity, self-motivated.
Knowledge for parametric modelling in archi-
tectural conception: The elementary operations of Pedagogical objectives
conception as well as the pragmatic operations Thanks to the results previously built on cognitive
described need some specific knowledge to be im- operations implied, we can also define specific re-
plemented. The analysis of the different operations quested behavior and know-how that are, for the
helps us to define this needed knowledge. As part moment, specifics to parametric modeling experi-
of this knowledge we can list among others: archi- enced experts. As Hugh Whitehead says « I think the
tectural knowledge; geometric and mathematical changes are more about attitudes than about technol-
knowledge; knowledge in computer science and ogy and that comes with experience » (Whitehead in
more precisely propagation based system (Aish and Kocaturk and Medjdoub, 2011, p.238).
Woodbury, 2005; Woodbury, 2010). For each cognitive operation previously identi-
fied in the use of parametric modeling in architec-

Table 1 To implement these operations: A learner should be able to :
Pedagogical objectives linked slicing operation Decompose an intention of parametric model
to cognitive operations. in terms of logical and geometrical chain of
dependencies
dimensioning operation Decompose an intention of parametric model with
proper and specific relevancies, especially linked to
his architectural intentions
logical operation of verification Test the viability of his parametric model in
extremum instances of his solutions domain
logical operation of induction Define general rules from particular sketches
operation of translation in parametric Interpret an intention in terms of propagation
geometry systems, hierarchy of the dependencies and
geometrical constraints
operations of collaboration Give a personal meaning to a collaborator discourses,
(interpretation and pooling) find information specific to a new problem
tural conception, we can propose some pedagogical sis shows that elementary operations of conception
objectives (Table 1). proposed by architecturology (slicing, dimensioning,
referencing) are accurate to describe the conception
Didactics resources of parametric models. We identify as well some po-
We defined then a pedagogical framework of web rosity between conception of parametric modeling
resources that architects can exploit during design and architectural conception. These porosities are
process and for training. This framework includes a allowed by third operations: operations of collabora-
general knowledge support and a specific knowl- tion (interpretation and pooling), an elaboration of a
edge support for the visual programming language cognitive representation of tools and a specific op-
Grasshopper. The general knowledge support [2] eration of parametric modeling (translation in para-
provides resources on geometry, computer graphics metric modeling).
and more broadly on applications of computer sci- Thanks to the identification of these character-
ences to architectural design. istics of the uses of parametric modeling in architec-
The Grasshopper resources are gathered into a tural conception, we proposed accurate: -general
library of samples presented with images of possible training objectives, -pedagogical objectives and -di-
produced shapes, a describing text with keywords dactics resources.
and obviously the corresponding *.ghx code [4].
This library is proposed as mediation for the use of REFERENCES
Woodbury’s patterns (Woodbury, 2010) by non-ex- Aish, R and Woodbury, R 2005. ‘Multi-level Interaction in
pert in parametric modeling. Parametric Design’. In Proceeding Smart Graphics, Lec-
ture Notes in Computer Science. Springer, Berlin.
CONCLUSION Ben Rajeb, S 2012. Modélisation de la collaboration distante
This paper presents some of the main result we dans les pratiques de conception architecturale : Caracté-
obtain on the identification of cognitive opera- risation des opérations cognitives en conception collabo-
tions implied in the uses of parametric modeling rative instrumentée. PhD Thesis, Ecole d’Architecture de
in architectural conception. We have interrogated Paris la Villette, Paris.
parametric modeling as an activity of conception in Boudon, Ph Deshayes, Ph Pousin, F and Schatz, F
itself (conception of parametric models). Our analy- 2000. Enseigner la conception architecturale, cours

d’architecturologie. Paris, Editions de la Villette. Kolarevic, B 2005. Architecture in the Digital Age: Design And
Chein, S and Yeh, Y 2012, ‘On creativity and Parametric De- Manufacturing New, Taylor & Francis Ltd, London.
sign’, Proceedings of 30th eCAADe conference, Prague. Lecourtois, C 2011. ‘Architecturological and epistemologi-
Davis, D Burry, J and Burry, M 2011. ‘Untangling Parametric cal research on collaborative design’, in International
Schemata : Enhancing Collaboration through Modu- Journal of Design Sciences and Technology, Volume 18
lar Programming’. Designing Together, CAAD Futures. Number 1, Europia productions, p. 31-46, Paris.
Lièges, Belgique. Lindsey, B 2001. Digital Gehry, Birkhäuser, Basel.
de Boissieu, A Guéna, F and Lecourtois, C 2011. ‘Operation Picon, A 2010. Culture numérique et architecture, Une Intro-
of parametric modelling and/or operation of architec- duction, Birkhauser, Paris.
tural conception ? Expressing relationships in paramet- Whitehead, H 2009. Interview with Aurélie de Boissieu.
ric modelling‘. In Proceeding 29th eCAADe. Ljubljana. Delft (not published)
De Terssac, G and Chabaud, C 1990. ‘Référentiel opératif Woodbury, R 2010, Elements of Parametric Design, Rout-
commun et fiabilité‘. In Les facteurs humains de la fia- ledge.
bilité dans les systè mes complexes. Toulouse.
Freiberger, M 2007. ‘Perfect buildings: the maths of mod- [1] www.grasshopper3d.com
ern architecture’. In plus.maths.org. available at : http:// [2] cnrtl.fr/definition/interpretation
plus.maths.org/ [3] cnrtl.fr/definition/traduction
Kocaturk, T and Medjdoub, B 2011. Distributed Intelligence [4] dnarchi.fr
in Design, Wiley-Blackwell, London. [5] www.parametric-ressources.com

Continuous Oscillations
A didactic for augmenting architectural design education with

computational design techniques via integrative feedback loops
Günter Barczik
Erfurt School of Architecture, Germany, HMGB Architects, Berlin
fh-erfurt.de/arc/ar/werkschau/master/digitales-gestalten/, hmgb.net
guenter.barczik@fh-erfurt.de, gb@hmgb.net
Abstract. We present and discuss a didactic for augmenting architectural design

education with computational design techniques via integrative feedback loops and show
examples of student projects. Our goal is to embed new technical skills into existing
design abilities as quickly as possible, in order to enable our students to exploit and
explore the extended capabilities of digital design techniques within the framework of
architectural design projects. We instigate a process of continuous mutual feedback
between different fields: on the one hand between technique-based exercises and
design-related steps, and on the other hand between the digital and the physical. Through
oscillation and feedback, the newly learned skills are directly interwoven with the existing
ones. Special emphasis is put on the illuminative effects of transitions between different
media and on issues of fabrication.
Keywords. Design curriculum; tools; shape studies.
GENERAL AIM: EXTENDED POSSIBILITIES DESIGNING WITH NEW TOOLS: THREE

Digital tools extend the scope of possible design TYPES OF OSCILLATIONS
solutions. With them, designers can formulate, con- We therefore introduce two things simultaneously:
trol and construct solutions to design problems that designs tasks that challenge the borders of non-
would otherwise be either too time-consuming or digital possibilities, and together with them digital
impossible to conceive and handle. This is mostly tools that allow students to cross those borders and
due to the difficulty of the geometry involved (such extend their own scope of abilities. Thus, the stu-
as intersections between polyhedral or curved ele- dents access new aspects of the design problem
ments), or its complexity, its number of elements. It from an understanding of new tool possibilities and
is important for us that our students can understand vice versa. Their learning oscillates between design-
and handle the new possibilities in such a way that ing and tool acquisition.
they are free to choose which solution they see as Additionally, we always task students with
the most appropriate, regardless of matters of com- the production of physical models. Building those
plexity, style or form. models becomes possible through the new digi-

Figure 1
Study models in progress ad-
jacent to Rhino Workstations.
tal design tools. The models are not just for the fi- projects, and we invite students to bring problems
nal presentations, but also sketch models that as from their more complex projects into the course so
early as possible transfer into the physical what was that they can be discussed and solutions be found.
sketched digitally (Figure 1). This has three main rea-
sons: Firstly, physical models in ‚real‘ 3D space are DESIGN COURSE STRUCTURE AND DE-
much more comprehensible and expose a design’s SIGN TASK SEQUENCE
qualities much better than digital models project- We have structured our design course in three steps.
ed onto a 2D screen. Secondly, even simple sketch In each step, a pavilion has to be designed and pre-
models already start to hint at production chal- sented in two-dimensional representations as well
lenges that become much more important when as in physical models. Before the precise design task
building in 1:1. Today, we think, it is rather easy to is set, we introduce various CAD tools. The design
be seduced by the possibilities of digital tools into tasks themselves then include certain requirements,
conceiving projects that then run into problems conditions and restrictions which invite if not re-
when they come to be realized. Early model-making quire employing the tools just introduced.
makes the students address such possible difficul- In each step, the physical models use less mate-
ties literally at first hand. Thirdly, the transition from rial, but the parts are more laborious to assemble.
digital to physical sometimes comes with mistakes, Where digital models can be made up of geom-
especially when students try out certain techniques etries that are continuous and as large as designers
for the first time, experimenting with production desire, physical models and - even more so - real
tools and materials. Such mistakes can very often buildings have to be assembled from components.
be made productive because the ‚wrong‘ or ‚failed‘ The ever faster development of large scale 3D print-
translations can unintentionally show new aspects ers only partially remedies this, because the printers
of the original that were difficult or even impossible mostly rely on very fine strata which, when viewed
to see there. So a second oscillation occurs between closely, again dissolve the continuities.
the digital and the physical. As software we use Rhino in conjunction with its
A third oscillation is attempted between the lit- Grasshopper Plug-In. Rhino is in the process of be-
tle, simple design tasks in the course and the larger coming the lingua franca for architectural 3D mode-
and longer design projects students undertake in ling, dito with Grasshopper for simple programming
parallel courses. The tasks we set are aimed to equip of such software.
students with techniques that also serve their larger

Figure 2
Stratified pavilion models.
MILIEU FOR WORKING AND STUDYING where investigations can be faster and more radical
We chose simple pavilion as topic for the design within a protected experimental realm isolated from
exercises in order to bridge the gap between exer- various restrictions.
cises dealing with the technical capabilities of the
software and the challenges of architectural de- STEP 1: CURVED FREEFORM SURFACES
sign - context, construction, spatial program, func- AND STRATIFIED MODELS
tionality. Pavilions do incorporate the latter, but to The first pavilion has to have various seating possi-
a degree that can be rather freely chosen by the bilities inside as well as outside, and its roof has to
students, so that there remains space for play and be accessible. It has to be a continuous form, not an
experimentation with the former. We strive to cre- assembly of components: all functional and circula-
ate a playground-like milieu where playfulness, ex- tion elements have to be synthesized and integrated
perimentation and risk-taking are common, so that into one coherent shape. Its physical model has to
students dare to - so to speak - flex the new-found be built from different strata cut manually or with a
muscles they have been equipped with (meaning laser-cutter (Figures 2 and 3).
the new software tools). We encourage the students We introduce free-form modeling tools in two
to attempts in which they are at first likely to fail. The steps: at first solids are manipulated through their
learning effect, and the sense of self-satisfaction on control points for quick but imprecise shape explo-
the students’ side, to us appears higher this way. For ration. Thereafter, surfaces are created from control
the students to be able to easily move conceptually curves - a more laborious but much more precise
we aim to create a ground that is both slippery and and intentional design method. Sculptural and func-
padded so that they can move swiftly and fall eas- tional aspects of the created surfaces are discussed,
ily - but soft. and the relationships between their aesthetical
Our intention is that students transfer the new qualities as objects ore public sculptures and their
possibilities explored through the new skills ac- usability as architecture. Categories like ‘furniture’,
quired onto other design projects they are or will ‘house’, ‘wall’, ‘roof’, ‘stair’, ‘ramp’ that appeared fixed
be working on; projects with more numerous and become fluent. A solution space for architectural
realistic requirements in terms of spatial program, design that was compartmentalized becomes a con-
constructability, functionality and relationship to ur- tinuum. The prevalence of purely horizontal surfaces
ban and socio-economic contexts. Our more simple in architecture is questioned and uses for inclined
pavilion designs are intended to serve as test cases, planes found and discussed.

Figure 3
Stratified pavilion models.
The digital designs are then sectioned into stra- more, students enjoy the possibilities of doubly
ta, stacked vertically or side by side. The stratification curved surfaces.
becomes a design theme in itself: how are the strata We encourage students to see occasional transi-
orientated, and how thick are they, i.e. how many of tion difficulties between digital and physical model-
them are there ? Students explore different stratifi- making as ‘happy accidents’ and exploit those as
cations, even un-parallel ones, experimenting with welcome design ideas (Figure 5).
radial and curved arrangements and strata that have
trapezoid instead of rectangular sections (Figure 4). STEP 2: INTERSECTING SPACES AND
Students experience that the stratification can DEVELOPABLE SURFACES
be seen either as an unwelcome tainting of the se- The second pavilion has to be the result of three in-
ductively perfect digital model, or as a means of tersecting shapes. The different source shapes have
structuring the endlessly pliable; making it more dis- to be recognizable in the resultant exterior shape
ciplined and taut. and create different spatial regions inside. These re-
Very often, the resulting designs play with the gions - as opposed to separate rooms - have to be
difference between outside and inside shape and associated with different functions. The hull surface
exhibit rather thick intermediate spaces. Further- has to be developable and built as a shell as thin as
Figure 4
Stratification Studies.

Figure 5
‘Happy Accidents’: Transition
Difficulties between digital
and physical exploited as
design ideas.
possible. Apart from the technical possibilities of in- ing pieces.

tersecting objects, means of dividing space into dif- The formal and sculptural freedom experienced
ferent areas other than primitive walls are explored. in the first task is restricted, but the designs more
Boundaries between different regions are discussed tightly coupled with the production possibilities.
as ambiguous. They do not stem from separating el- Much less material is used, and its fabrication does
ements which are inserted into an already existing not rely on the availability of laser printers (Figures
shell but result from the mere shape of a complex 6, 7 and 8). The interplay between (almost) unre-
space. In other words: continuous spaces are dif- stricted digital form-making and the reduced pos-
ferentiated via widening and contraction, embod- sibilities of physical production are experienced and
ied in the seams of the different intersections. Fur- discussed.
thermore, the possibilities of ‘negative’ spaces are Special emphasis is put on how additional sta-
explored; spaces that are created through boolean bility can be achieved in the physical models by hav-
difference. ing the intersected parts support one another.
The ‘Unroll‘ command is introduced together Again, we aim to let the students welcome the
with boolean commands for intersecting and split- parameters of physical production into the design as
ting shapes and separating and exploring the result- informative factors rather seeing them as obstacles.
Figure 6
Intersecting spaces and
developable surfaces pavilion
models.

Figure 7
Intersecting spaces pavilion
sectional models.
STEP 3: COMPLEX ROOF AND DIS- simple form are the surface, the geometry to be
SOLVED HULL mapped, the number of u and v separations and the
The third pavilion is more of a roof, i.e. for an archeo- height of the projections. Occasionally, we extend
logical excavation. It has to be a single surface that the definition with more parameters, varying the
changes from convex to concave at least once. In height or leaving the uniform division of the surface
order to fabricate it, using Grasshopper the surface behind more complex patterns.
is populated with a three-dimensional pattern in The de-materialization from Step 1 to Step 2 is
such a way that it is divided into multiple develop- further continued as the resulting surface is perfo-
able surfaces. The population pattern has to include rated so that its holes are larger than its solid parts.
holes so that the resulting populated surface be- The geometrical restraints that were introduced
comes porous (Figure 9). from Step 1 to Step 2 are removed again. The formal
We employ a simple and well-known grasshop- freedom from in Step 1 is synthesized with the con-
per definition that divides a surface into a rectangu- struction capabilities from Step 2.
lar grid and maps a given geometry onto the indi- Students study the effects of the geometry of
vidual cells. The parameters in the definition’s most the population modules and the population system
Figure 8
Intersecting spaces and
developable surfaces pavilion
models.

Figure 9
Three-dimensional pattern
population variations.
on the original surface and explore the difficulties are not builders. It is, though, becoming ever more
and possibilities in fabricating modules and surface important as the growing number of design tools
(Figures 10 and 11). Certain combinations of surface and fabrication methods increases the number of
curvature and mapping height easily create self-in- specialists while decreasing the percentage share of
tersections. The definition does not check for those existing skills that any individual can have - thereby
- this would have been to difficult to implement raising the number of specialists and therefore the
within our course structure. need for shared work and communication of goals,
The possibilities of customized mass-production intentions and ideas.
- already hinted at in design step 1 - are explored
and discussed. CONCLUSION AND OUTLOOK
In order to fabricate and assemble the numer- In order to extend existing design skills, we intro-
ous parts that make up the surfaces, the students, duce technical possibilities of CAD software with
after having worked individually in tasks 1 and 2, conceptual and geometrical design tasks. We at-
now form groups of 3-4. So in addition to the CAD tempt 3 oscillations: between technical tools and
techniques, design possibilities and fabrication design possibilities, between digital and physical
methods, teamwork is experimented with: who does models, and between simple architectural designs
what, in which sequence are steps undertaken, how within the design course and the larger design pro-
are communal decisions reached ? Such teamwork jects students work on in parallel. These repeated
has always been important in a discipline where, like movements between different modes of working in
with composers but unlike visual artists, designers time weave numerous conceptual strands that be-
Figure 10
Complex roof and dissolved
hull pavilion models.

Figure 11
Complex roof and dissolved
hull pavilion models.
gin to tie different conceptual regions into a whole. and the work done outside of it, so that the new ter-
In the future, we aim to intensify this weaving, ritories opened up for designing architecture can be
especially of the work done within the design course traversed more naturally.

Computation/Performance
Problematizing education_ integrating computational methods in

relationship to ‘performance’ within a new undergraduate curriculum
Anna Pla-Catala
IE University, Spain
http://www.ie.edu/university/studies/academic-programs/bachelor-architecture, http://
www.ie.edu/school-architecture-design/, http://ienudl.wordpress.com
anna.pla@ie.edu
Abstract. Setting up a completely new architectural academic curriculum for a

brand-new school of architecture. Elaborating, critical space within the new curriculum
for the teaching and learning of ‘digital-technologies’ (DT) through the integration of
them with its core subjects, rather than understanding the teaching/learning of such
technologies as an add-on set of skills that comes a posteriori. How to articulate the
potential of the ‘Computational-Architect’ as a professional capable of being a productive
agent within society; that is, capable of adding Value. The nature of such is what it’s at
stake here, if we want to avoid to become or be reduced to mere providers of services.
Keywords. Computation; performance; ecology; code; maker.
INTEGRATING COMPUTATION_PROBLEMATIZING EDUCATION

To have the opportunity of setting up a completely years of incipient existence (2008-2013). This period
new architectural academic curriculum for a brand- is equivalent to getting a Bachelor’s degree, and in
new school of architecture is an extremely exciting consequence, the level of produced work is of a:
challenge that does not appear often. Basic Level. However, and maybe precisely, it is con-
Within this task, even more exciting for the dis- sidered especially relevant as is the result of estab-
cussion at place here, is the project of intention- lishing DT’s basic-structural disciplinary seeds at the
ally dedicating, as well as elaborating, critical space very foundation of the development of a practice.
within the new curriculum for the teaching and The paper will try to evaluate the failures and
learning of ‘digital-technologies’ (DT) through the successes, and the reasons behind those. Examples
integration of them with its core subjects, rather of work accompany the text.
than understanding the teaching/learning of such
technologies as an add-on set of skills that comes a NUMERIC CULTURE
posteriori. The question of DT’s autonomy -or not- within archi-
Besides summarizing the above mentioned ex- tectural making, together with the capacity of this
perience, this paper tries to also evaluate the out- technology to form an independent corpus of work
come of the implementation of such DT after its 5 and discourse by itself, is key.
Chapter - Computation and Performance - eCAADe 31 | 579

In the light of the European higher education Considering the current socio-economic Euro-
area as prescribed by European policies, together pean current context; i.e. the complete re-organiza-
with the practice of an architect being a regulated tion of the global-economy and the new emerging
one by the Government (in its final form of profes- cultural paradigm slowly appearing in the midst of
sional associations needed to practice/perform/sign), a crisis that particularly affects spatial practices, the
it is more than legitimate to affirm that there is no ac- matter becomes:
tual spot allocated for DT (computation+fabrication) How to articulate the potential of the ‘Computa-
in an undergraduate/eec curriculum today. And tional-Architect’ as a professional capable of being a
hence it is also legitimate to question why that is. productive agent within society; that is, capable of
Since the inception of DT into architecture dur- adding Value. The nature of such is what it’s at stake
ing the 90s, DT have come into play as a radically here, if we want to avoid to become or be reduced
disruptive form of knowledge for the architectural to mere providers of services.
discipline. Over the years, there has been an increas- The underlying guiding principle in structur-
ing acceptance of them on every front. However, the ing this DT-CV, has been from its beginning, a very
myriad of courses of all sorts of formats that have simple urge to convey to students the relevance of
appeared since then, still place the teaching/learn- moving on from their initial (immediately gratifying)
ing of DT at the ‘End’ of an architectural academic use of DT as representational tools, into generative
degree (post-graduate levels) or at most, as a spe- methodologies.
cialization (end of bachelor or workshop). Our goal has been to add the necessary skills
This paper aims at posing critical questions as that allow for both the conception and control of
well as describing a critical reflection on the educa- higher geometrical orders as well as software inter-
tion of the ‘digital-native’-architect. operability. But above all, to convey: Why all this
A clear bias is placed on an Education Model as matters, i.e., to be able to fully communicate Why
a platform for research (as a focus of investment in it’s important to expand the limits of ‘conventional’
education as much as a true cultural and produc- CAD concepts and ‘user-friendly’ interfaces that,
tive innovation), over a model based on teaching as nevertheless, do nothing much more than constrain
a mere form of transfer of previously accumulated architectural language through predetermined ar-
knowledge. chitectural elements. In short, the aim of this project
Implementing DT undergraduate program has been to create the intellectual foundations for
2008-2013 implied settling down the very founda- a design culture based on algorithmic thinking and
tions of a new kind of culture. What’s important is digital fabrication (Figure 2).
the creation of a ‘not-analogue’ kind of culture, but a
numeric one (Figure 1). Epistemology
This alone, and perhaps simply said, implies none-
PART 01_COMPUTATION / EDUCATION theless an enormous and unparalleled epistemo-
‘I think everybody in this country should learn how logical change. It definitely is the most radical shift
to program a computer because it teaches you how of mentality that has occurred since Modernism and
to think’ (Steve Jobs, The Lost interview). its Post-s... i.e., the fact that within an associative de-
This academic curriculum aims at establish- sign framework, students are forced to address the
ing the proper intellectual environment for active problem of how to design and fabricate architec-
engagement in building-up a set of skills and tech- tural components that are ‘Programmed’ rather than
niques based on the embracement of Digital Tech- ‘Drawn’.
nologies as the relevant tooling apparatus for cur- Probably the most obvious of the proven conse-
rent and future generations of architects. quences of introducing DT in undergraduate archi-
580 | eCAADe 31 - Computation and Performance - Chapter

Figure 1
NuDL_Digital Technolo-
gies_First Year.
tecture school (programming- fabrication, whether resistance (even fear) towards the reformulation of
hard-coded or graphic-scripting), are the questions design Authorship and what constitutes such notion
and discussions on ‘process-driven’ design that in- today.
evitably and immediately arise. Even if ‘rule-based’ This might explain why implementing DT has
design systems have been mainstream for decades been (in our specific context) reasonably achievable
already in some design contexts even in analogue and successful (by being accepted and willing to im-
form (Eisenman) , there still exists an extremely high plement it) in almost every area of the academic CV
Figure 2
NuDL_Fellowship.

Figure 3
gies02_First Year.
but design studio, which is the area where we have initial intuitive hunch by means of the hard-core
found most resistance from. Before jumping into ob- rigor that computational tools entail is such, that
vious criticisms however, this fact might have a very the designer must be skilled first, and above all, in
simple explanation. One that lies at the core of the the ‘Logic of Design’ of highly complex systems that
problematics that emerge out of the profound shift comprise -geometric, algebraic and logical- relation-
in architecture-making due to the impact of DT (Fig- ships.
ure 3). As educators, a two-fold task presents ahead of
us; on the one side, to keep up with the fast rate de-
Instrument vs Method velopment of DT’s as intrinsic to themselves (Com-
If structure, construction and representation classes puter Science), and on the next, to focus on the
have welcome DT’s corpus of knowledge in col- relationship with the corresponding culture of ‘use’
laboration with their own, it is primarily because within Design Practices. What is key, is how to trig-
parametric modeling, programming and digital fab- ger the combination of ‘Intuition and Logic’ both of
rication are mainly valued as ‘Instrument’ and not ‘Ideas and Skills’ in one single but multidimensional
‘Method’. To be more accurate, as an instrument for dynamic ensemble.
improving: a) workflow, b) variable input/output Experience over the past 5-years has proved
and, c) delivery of precise geometric data to be taken that prejudices as to what architectural design
to digital fabrication and/or performance analysis. ‘is’ or ‘ought to be’ still exist. And the introduction
Nonetheless, this fact alone we argue, merely of programming and digital fabrication within
constitutes a slight automation device of otherwise architecture´s education has still to overcome an ex-
traditional and conventional design procedures, by- tensive set of deep-rooted classical values. Most sur-
passing the truly essential foundation of parametric prisingly is the fact however, that these prejudices
and algorithmic thought. do not always come from some of the more estab-
The degree of control necessary to develop an lished layers of the profession (as perhaps expected),

Figure 4
NuDL_Scripting_First Year_
Professsors: Carlos Barrera,
Adolfo Nadal.
but also, from the collection of ‘a-priori assumptions’ (Code.org Co-founder Claire Sutcliffe (Geere, 2012))
that young candidates arrive at architecture school Such initiatives deserve our deepest respect
with . . . not only about the discipline, but also in re- indeed. It is most admirable to have achieved for
spect to the the digital, and the radical change that computer code to surpass the ‘geek’ community in
is involved in making a highly ‘strict’ use of what they order to become a Country’s policy for children’s ed-
otherwise have known to be ‘playful’ devices. ucation; a generation, let’s not forget, that will still
At an institutional level and in contrast with the take 10 years approximately to get to Undergradu-
type of architectural education’s resistant attitude ate Schooling.
we have tried to convey, a couple of non-architectur- And this is fact alone proves, that architecture
al examples are here worthy of noting. Such projects schools should stop worrying about how to preserve
are born out of a true honest belief in the capacities traditional disciplinary knowledge modes and cease
of computer code and the new epistemological to have a conservative attitude in order to fully (and
paradigm opened-up by DT. Those are: Code.org rapidly!) embrace programming and fabrication, as
[1] in the USA, and the recent enterprise taken on well as the rest of the vast array of DT.
board by Code Club [2] in the United Kingdom (an Because, to put it very simply: These are our
afterschool voluntary initiative that aims at teaching New Standards. And as such, this is the responsibil-
computer programming to 10 year-old kids). ity of architecture education today (Figure 4).
‘At age 10-11 (on average) children have the
necessary numeracy, literacy and logic skills to learn PART 02_COMPUTATION / PERFOR-
the concepts of coding’ .... ‘Some might argue that MANCE
they have these skills even earlier than that. To be Even if the expected resulting final work to be de-
blunt, ICT lessons today mainly consist of learning livered by an ‘Architect ‘ remains being a physical
Microsoft Office. Are we raising a nation of secretar- structure (a built design), it has indeed become
ies? I sincerely hope not. It’s insulting to children to more than clear that the contemporary architec-
think they can’t handle something more creative, tural model we all participate in (every agent in the
inspiring and powerful than an Excel spreadsheet’ design-to-construction process), is an evolving one

Figure 5
NuDL_Digital Design_First
Year.
that has become as much cybernetic as material. a clear distinction from the one of the Engineer.
Computation has given the designer an unprec- Hence, computation in relation to performance is
edented degree of Control over the complete spec- evaluated here with an explicit criticism towards sta-
trum of design-build processes. As a design tool, it tistical and self-referential efficiency models as sole
is capable of dynamically defining the global coor- alibis or testing-modes of resulting prototypes.
dinates of a generic continuum, to then yield up to In biology, epigenetics studies how environ-
a specific (intentional) configuration. The criteria for mental factors affect genetic function (genotype).
evaluating which single instance is most suited for a Similarly, ‘rule-based’ design processes have at their
particular design problem, is what drives us to the starting point the definition of a robust ‘genotype’
notion of Performance. that can be subsequently refined according to feed-
As a measure of the direct output of a driven back-loops that incorporate further information ex-
process, performance is usually conceptualized as ternal to itself.
the increase-or-decrease in efficiency of such pro- Ecology is not sustainability. In an effort to re-
cess. Although computation has been incorporated consider the Holistic ‘intelligence’ formed by the
into the discipline of architecture, it has been mainly whole complex set of spatial components (digital,
used for two main tasks: a) to generate complex ge- physical, material, economic, atmospheric, etc.),
ometries that intensify the function of the Formal; computational design ought to develop a model
or, b) instrumentalised as mere optimization device capable of strategically, tactically and synergetically
without exploiting its ontological/cultural potential relate to its environment. The utilization of Code as
beyond technocracy. design method acquires full meaning only if it dy-
Our mission has been to articulate a digital ex- namicaly integrates the affects of the material con-
pertise for the 21C Architect whose practice is of text in which it develops (Figure 5).

Figure 6
gies03_First Year.
PERFORMANCE_ (NON)OPTIMIZATION: approach to sustainability. Such a reductive notion

NEW DISCIPLINARY CODES of performance responds to a linear way of thinking
Computer Code is the 21st century Architect’s Tool- that prioritizes the minimum use of material, struc-
ing Apparatus, and as such it is irreducible to any of ture and energy to fulfill single conditions. Paradoxi-
Architecture’s traditional design taxonomies. Algo- cally, it is also one that is in direct opposition to the
rithmic thinking has entered design practices to ir- non-linear dynamic nature of computational en-
remeably transform them bottom-up. Code has be- gines.
come policy, albeit not in its restrictive sense but as For the discipline of Architecture not to fully
generative protocol. incorporate and reflect the enormous impact that
Performance needs to be theorized in order to Computer Science has had on all spheres of knowl-
move beyond modernist models based on a func- edge, is not a wasted opportunity but a retrograde
tionalist paradigm of efficiency and a mechanical act. If design research can attempt at becoming

scientific method, there needs to be a complete Kieran, S and Timberlake, J 2004, Refabricating Architecture,
consistency between technology-tool-technique- McGraw-Hill, NewYork.
episteme. Kuhn, TS 2012, The Structure of Scientific Revolutions, Univer-
Although only 5 years old, this academic pro- sity of Chicago Press, USA.
gram has attempted to propose richer and more Kwinter, S 1992, Emergence: or the Artificial Life of Space,
complex approaches to the generation and evalu- Anywhere, New York.
ation of built forms to the extent possible and by Leach, N, Turnbull, D and Williams, C (eds) 2004, Digital Tec-
means of diverse methods and exercises of diverse tonics, Wiley-Academy, UK.
difficulty for an Undergraduate level (Figure 6). Sakamoto, T, Ferre, A and Kubo, M (eds) 2008, From Control
to Design: Parametric/Algorithmic Architecture, Actar,
REFERENCES Barcelona.
Anderson, C 2012, Makers, Random House Business Books, Terzidis, K 2006, Algorithmic Architecture, Elsevier, Oxford.
London.
Geere, D 2012, ‘Afterschool ‘Code Clubs’ planned to teach [1] www.code.org.
kids programming’, Wired.co.uk, 17 April 2012. http:// [2] www.codeclub.org.uk
www.wired.co.uk/news/archive/2012-04/17/code-club

Dances with Architects
Interactive performance as a new concept for architectural design studios

Angelika Lückert1, Volker Koch2, Petra von Both3
Karlsruhe Institute of Technology (KIT), Germany
1
http://blm.ieb.kit.edu/135_722.php, 2http://blm.ieb.kit.edu/135_72.php, 3http://blm.ieb.kit.
edu/135_44.php
1
angelika.lueckert@kit.edu, 2volker.koch@kit.edu, 3petra.vonboth@kit.edu
Abstract. This paper proposes a complementary approach for the architectural design
studio. By interpreting architecture by means of an interactive (dance) performance as
design task it combines architectural theoretical examination with the implementation
of new technologies and event realization. This design studio concept integrates
scenography, choreography, sound design and event management, providing workshops
carried out by external and internal experts to give insight into these disciplines and new
tools. The experimental form allows the students to define the specific form within a broad
scope, ranging from a dance performance performed by the students themselves to an
interactive installation. The focus for the students was on dealing with the diverse input
and on the decision-making process and its reflection.
Keywords. Interactive; performance; teaching; collaboration; gesture control.
INTRODUCTION
The prevailing concept of design studios at architec- the essential characteristics into a different medium.
ture schools is that of simulating a design process The concept of this design studio involves a very
in an architectural office. Consequently, a standard broad range of topics so as to incite the students’
task may consist of designing a building or other learning process with a task that represents the
objects in the context of the built environment. By complexity of the design process in architecture. The
contrast in an alternative approach several exist- project therefore combines elements of architectur-
ing works combine architecture and dance (Bronet al theory with an extensive number of efficient tools
and Schumacher, 1999; Pekol, 2011) and served as a and basic introduction in choreographic and sceno-
starting point for a new design studio concept. graphic acting. As collective project the students in-
In contrast to established concepts, this design volved have to clarify their specific art and technical
studio has an interdisciplinary setting, using new skills and assemble all competences of the group for
media, dance and scenographic elements to trans- a presentable and conclusive stage event.
late architecture into an interactive performance. A The future outcome being an interactive perfor-
central educational aim of the project described in mance – instead of (relatively) theoretic drawings of
this paper was to instruct the students to analyse and an architectural object – fundamentally transforms
abstract architecture in order to be able to transfer the design and working process. Firstly, the students

are introduced to completely new subjects, which On the basis of the aspects highlighted above
are scenography, dance and man-machine inter- and with the awareness that certain pedagogical
faces (MMI). Secondly, they have to learn the use of settings, like tutoring and reviews, limit the students’
hardware and software tools, and most importantly possibility to express themselves elements were
they need to develop the ability to decide which introduced to create a more informal learning set-
tools are best suited for their intention. ting. Therefore an essential part of this design studio
The project is based on an interdisciplinary co- project was for the students to decide which tools to
operation between a department of the faculty of use and how to convey their ideas, and to continu-
architecture and external specialists from the field ously reflect their decisions. And the output, i.e. the
of scenography, choreography, programming and final ‘product’, was purposely not exactly defined in
visual arts. Accordingly, there are different levels and the preparatory phase.
focuses in this project: the acquisition of expertise in Instead of defining how to present and repre-
handling the technical tools, the creation of a com- sent their ideas the students were asked to define
plex scenographical work, which included the work by themselves how and via which medium their pro-
with dancers, and the organization of the event it- posal should be transported. And based on Schön’s
self. (1983; 1987) assumption that reflection plays an im-
The course, which was named “Dances with Ar- portant role in the development of elaborate design
chitects”, was first offered in summer 2012 and open expertise we introduced specific tools to facilitate
to advanced students. The students could choose to this process, like the documentation book.
work on specific aspects of the project. The establishment of groups, instead of the usu-
ally promoted individual works, pursued several
THE ARCHITECTURAL DESIGN STUDIO objectives. Opening the narrowed learning environ-
Project-based learning has traditionally been an element concept, where the one-on-one tutoring is the
mentary component of the architectural curriculum central element, the group work takes into account
with the design project as its centrepiece. the fact that students can learn as much from their
This innovative concept for the design studio peers as they learn from their teachers. And not only
aims at opening the narrow concept of learning did the students work in small groups, but they had
environment (Webster, 2008) reflected in Schön’s to organise the work of the entire studio and the
influential works ‘The Reflective Practitioner’ (1984) event itself.
and ‘Educating the Reflective Practioner’ (1987). As Whereas the basic structure of the design studio
Webster justifiably assesses, Schön limits everything was defined beforehand – based on the premises
to the formal one-on-one meeting between design and aims outlined above – some elements partly
tutor and student. Additionally his understanding of evolved in the course of the project due to its exper-
learning and teaching is based on the premise that imental character. This gave the opportunity to react
teachers know the right solutions and that some- to the needs and request of the students.
thing like a generally shared aesthetic values exist.
Though Donald Schön’s decisive works turned Documentation book
the focus from the accumulation of technical knowl- To reinforce and systematize the reflexive process
edge in architectural education to “reflective prac- each group was asked to draw up a book that should
tice”, which was defined as “the capacity to reflect assemble all information, inspiration, sketches, im-
on action so as to engage in a process of continuous ages, etc. that appeared relevant to the students
learning” (Schön, 1983) the understanding of archi- in forming their decisions. Beyond this rather intui-
tectural education remains somehow limited (Web- tive collection of material this book should include
ster, 2008). explanatory texts that described the reasons for de-

Figure 1
Concept Collage: Created by
student group working on the
Barcelona Pavilion (1929) –
upper row: Concept visualisa-
tions of interactive projection;
lower row: images from the
Barcelona Pavilion.
cisions taken in the different phases of the design The next phase involved introductions to the
process. different hardware and software tools and presen-
As explained above one reason was to collect tations with related topics. While becoming familiar
as much design relevant information as possible. with the different tools and aspects of this project,
Another important aspect – which was communi- the students were given the choice to organise
cated to the students – was to enable the students themselves in three different work groups: costume
to rethink and if necessary to revoke decisions and design, music and technology, and stage design. As
go back in to that design phase. the work progressed, the students organized all nec-
essary coordination themselves and project man-
CREATING AN INTERACTIVE DANCE agement became an important and integral part of
PERFORMANCE this complex work.
The objective of this project was the realization of The main part started after the students had
an interactive dance performance using tools that been introduced to this new area and gotten ac-
would react to the dancer’s movements. The pro- quainted with the tools. The learning process still
ject was based on cooperation with a scenographer played an important role in the students’ work pro-
and a choreographer. In the first phase, the students cess, and it shaped the ideas of what their work
were presented with four different buildings from could look like. Not only were the students intro-
accredited architects. The students had to analyse duced to some examples of the enormous amount
the buildings and detect the main architectural of interactive dance performances [2,3] that are
themes. With these themes and metaphors in mind, being developed, but they also had to find their in-
they were asked to find the adequate form in trans- dividual answer in relation to their abilities and the
lating them into a performance (Figure 1). In order selected building.
to facilitate the students’ progress in this new field, The final performances – which will be ex-
we included activities like visiting a final rehearsal plained in more detail below – cover a broad range
at the theatre and a modern dance performance reflecting the different architecture, which they try
[1]. Furthermore, the students were asked to make to translate. They range from the creation of inter-
improvisation exercises in order to develop a basic active illusions (Mies) to an interactive installation
understanding of the language used in dance. based on design methodology (Haller) to a provoca-

Figure 2
Selected buildings: Barcelona
Pavilion (1929), HTL (1966),
Jewish Museum (1999), Rolex
Learning Center (2010).
tive theatre play (Libeskind) and a colourful dance benefit from dance elements, if the dance elements
performance (SANAA). Some of the works focus on would be performed by themselves or by dancers,
the choreography, others explore the effects, which or if the performance would not include any dance
can be produced on projections via Kinect sensor, elements at all. It was continually made clear to the
and still others play a virtual game creating an aug- students that the choice of tools and elements was
mented reality with the use of markers and specific theirs and that their decisions should be based on
glasses. their evaluation of the adequacy to support the in-
terpretation of architecture with the language of
DESIGN STUDIO ORGANIZATION dance and interactive media.
As the setting and organisation of the design studio The following list presents all tasks that were
is quite complex, it will be explained in detail below. part of the project (Figure 3):
Whereby the emphasis is on the first two phases. • Creation of a performance based on famous
This architectural design studio project involved buildings.
– to a greater or lesser extent – the following partici- • Production of music or music compilation (us-
pants: architecture students (20), scenographers (2), ing Ableton software).
dancers (2), choreographer (1), media artist (1), and • Event management (organization of dance
architecture design tutors (2). performance event, including opening speech,
At the beginning the students were presented catering, designing invitations, stage setting,
with a choice of buildings and were asked to work etc.)
on the “translation” of these buildings in groups of The first phase was a conventional architectural
five. analysis, consisting of research about the history of
The selected buildings where (Figure 2): the building, the architect(s), the most important
• Mies van der Rohe, Barcelona Pavilion, 1929 historic trends and other relevant aspects of that
• Fritz Haller, HTL, Brugg-Windisch, 1966 time. As a kick-off, the professor for architectural
• Daniel Libeskind, Jewish Museum, Berlin, 1999 theory made a presentation in which the buildings
• SANAA, Rolex Learning Center, Lausanne, 2010 were introduced, focusing on the atmosphere.
It should be pointed out that the students where In addition the students were asked to de-
given great freedom in creating their performance. fine the characteristics of the buildings and name
They were free to decide if their concept would three themes and metaphors for each. They were
Figure 3
Organisation of Architectural
Design Studio.

informed that these themes and metaphors would had created transferring the movements through
help them translate the architectural work into an- the respective building into dance elements.
other medium. Concurrently the second phase introduced the
As outcome of the first phase the groups pre- students to the new software and tools. This was
sented their findings. It became apparent that apart done through presentation, demonstrations and in-
from a thorough architectural analysis, the groups tense workshop sessions. For this project the main
tried to convey the atmosphere of the building by focus was on the interaction between graphics
using pictorial metaphors and the like. developed with Processing, the Kinect sensor and
The concept of the design studio now foresaw human movements.
to introduce the students to the language of dance. The third phase, Concept + Cooperation, follows
So the next phase started with a workshop carried the normal structure of a design studio, with in-
out by the choreographer and dancer Patricia Wolf. tense work by students and regular discussions and
The theoretical introduction briefly outlined the “La- coaching with the design tutors. And the fourth and
banotation”, a system that analyses and records the final phase consisted of the production of the vari-
human movement developed by Rudolf von Laban ous elements of the performance and organisation
(1991). of the event.
Based on Laban’s basic explanation of how space
is viewed and experienced in dance, the “Improvisa- APPLICATION OF SOFTWARE AND
tion Technologies” developed by William Forsythe HARDWARE TOOL
(1999/2003) were introduced. With the improvisa- In the course of the project the students had to
tion exercises presented in the related video (and develop their expertise in the use of different soft-
some other improvisation techniques) the students ware and hardware tools. In this case: Ableton Live,
had to create short choreographies within one hour. an audio/music production software, which they
They either worked with changing rhythms or they used not only to arrange tones, but also to modify
used words to create and perform an improvised music they had produced by analogue means; the
choreography. video mapping software MadMapper that enables
Simultaneously a new task was assigned to the users to define the form of the projection. The open
students, which required analysing the building source programming language Processing was se-
based on movements. The student groups conse- lected to use the gestures captured by the Microsoft
quently studied the ground plans of the buildings Kinect sensor to transform projections.
and tried to trace the movements of an imaginary The technical implementation of the perfor-
visitor. The students depicted the movements by mances is based on the integration of gesture cap-
drawing curved lines for lines of movement and ture by the Microsoft Kinect device and their con-
marked spots where the visitors’ attention was at- nection with different tools of acoustic and visual
tracted so that they came to a halt. This analysis was output. In detail, the students were basically intro-
intended to serve as a base for the design of their duced in the use of the Ableton Live sequencer as a
future choreography. As we will see later this proved tool for music production, the video mapping soft-
to be true only for some of the student groups. ware MadMapper and the integrated development
The students presented the outcome of this environment Processing for interactive graphic pres-
analysis at the end of the 2nd phase. The presenta- entation. All these tools offer a direct connection to
tion was composed of two parts: the first part was the Kinect sensor and allow immersive conversion of
a slide presentation featuring the “movement analy- the dancers’ movements by tracking specific parts of
sis”; the second part was a live performance or video the human body.
presentation of a small choreography, which they

Created Performances Dance elements (Interactive)projection Table 1
“MIES in motion” Yes Yes Categorization of (dance)
“HALLER interaktiv” No Yes performances.
“Decertatio” [Libeskind] Yes No
“Colour feelings” [SANAA] Yes (Yes)
CREATION OF PARAMETRIC DANCE sic or sounds had been created analogue by record-
COSTUMES ing sounds that interpreted the used materials.
The concept of this interdisciplinary project foresaw The work “Haller interactive” was not referring
a development of the dance costumes using para- only to the selected building (HTL), but in response
metric design methods. Therefore, a three-day work- to Fritz Haller’s universal design method focused on
shop for Rhino and Grashopper was held to provide the topics regularity, modularity and order. The stu-
a profound introduction to the parametric design. In dents decided that the best representation would
addition to that, a broad range of devices was made be to create an interactive installation resembling
available to the students in the department’s labora- a computer game. By playing this interactive game,
tory. This included a 3D printer and a laser cutter; the which was composed of augmented reality and
latter had been chosen by the students because of interactive elements, the participants intuitively
its easy handling and suitability for the task. learned the rules (Figures 4 and 5).
As the starting point for developing the costume, In conformity with the prevalent atmospheric
the students selected a central element of the cho- power the work “Decertatio” interprets Libeskind’s
reography, which consisted in the idea that the Jewish Museum as a theatrical enactment. In the
dancers would open the imaginary boundary, the centre of the production stand disorientation,
so-called “fourth wall”, between themselves and the provocation and conflict. Set in a black box the audi-
audience. The opening up was expressed through ence was placed in a square in the centre of which a
the dissolution of the costume, which consisted seemingly lost dancer danced in a spot light, while
of two layers. The first layer, a simple fabric band an actor recited a text that expresses conflicting
wrapped around the dancer, was slowly unwound. emotions.
The second layer mainly constituted the costume The performance “Colour Feelings” enacts SAN-
and was produced with the laser cutter. The pattern AA’s Rolex Learning Center as dance performance
was created with the use of Rhino and Grashopper with the students themselves as dancers. Each spa-
on the basis of a parametric design patterns. tial impression is represented by an own scene char-
acterized by a specific colour and performed by one
INTERACTIVE (DANCE) PERFORMANCES of the students. Interactive projections that react to
Due to the experimental form the four student the movements reinforce the impression but play a
works vary considerably with regard to conception minor role. In each scene one person personifies the
and execution. To expand on this variety with more visitor, and another person the built environment.
detail, Table 1 presents a categorization and the fol- The performance ends in a scene where all dancers
lowing summaries shortly describe the content and come together and leave their ‘marks’ expressed by
differences of the student works. the specific colours on the surrounding.
The “MIES in motion” sets its focus on the aspects
of materiality and space and perception. Images of CONCLUSION
the surfaces where projected onto a special screen. Related to the complex task of developing an in-
The dancer’s movement in front of the screen modi- teractive dance performance, the structure of the
fied the projected graphics. The accompanying mu- design studio was equally complex. As a first evalu-

Figure 4
Augmented reality set-up
(“Haller interactive”).
ation, we can point out that the unexpected com- public performance requires cooperation on differ-
bination of theoretical aspects of architecture with ent levels and functions. In this context, the project
the management and organization of a public per- supports students’ skills in cooperation, time man-
formance and the integration of modern soft- and agement, budget and the interaction with a group
hardware applications leads to challenging but also of specialized participants.
very stimulating tasks for students. Setting up this
project therefore activates competences in vari- REFERENCES
ous ways. On the one hand, it brings students into Bronet, F; Schumacher, J 1999, ‘Design of the Movement:
close, direct and playful contact with software and The Prospects of Interdisciplinary Design’, Journal of
techniques like rapid prototyping, gesture recogni- Architectural Education, 53(2), pp. 97-109.
tion, sound editing or visual programming. On the Forsythe, W 1999/2003, Improvisation Technologies: A Tool
other hand, it establishes a strong and intense link fo the Analytical Dance Eye, (CD-Rom), ZKM, Karlsruhe.
to architecture itself and forces students to reduce Laban, R von 1991, Choreutik: Grundlagen der Raumharmon-
the intentions of architects and their buildings to ielehre des Tanzes, Noetzel Verlag, Wilhelmshaven.
very essential statements. Combined with a high Pekol, B 2011, ‘BodyCAD: Creative Architectural Design
motivation of the participants, the management of a Through Digital Re-Embodiment’, ISEA, 2011.

Schön, DA 1983, The Reflective Practitioner: How Professio- Figure 5
nals Think In Action, Basic Books Inc, New York. Video glasses and sound
Schön, DA 1987, Educating the Reflective Practitioner, Jos- system for augmented reality
sey-Bass, San Francisco. set-up (“Haller interactive”)
Webster, HM 2008, ‘Professional Education after Schön:
Cracks, Blurs, Boundaries and Beyond’, Journal for Edu-
cation in the Built Environment, 3(2), pp. 63–74.
[1] http://www.kulturverein-tempel.de/index.php?id=355/
[2] http://anarchydancetheatre.org/en/project/seventh-
sense/
[3] http://www.wedream.co/interactive-dance-perfor-
mance-2/

A Case Study in Teaching Construction of Building Design
Spaces
Mahsa Nicknam1, Marcelo Bernal2, John Haymaker3
1
Georgia Institute of Technology
1
mnicknam@gatech.edu, 2marcelo.bernal@gatech.edu, 3john.haymaker@coa.gatech.edu
Abstract. Until recently, design teams were constrained by tools and schedule to only
be able to generate a few alternatives, and analyze these from just a few perspectives.
The rapid emergence of performance-based design, analysis, and optimization tools
gives design teams the ability to construct and analyze far larger design spaces more
quickly. This creates new opportunities and challenges in the ways we teach and design.
Students and professionals now need to learn to formulate and execute design spaces in
efficient and effective ways. This paper describes curriculum that was taught in a course
“8803 Multidisciplinary Analysis and Optimization” taught by the authors at Schools of
Architecture and Building Construction at Georgia Tech in spring 2013. We approach
design as a multidisciplinary design space formulation and search process that seeks
maximum value. To explore design spaces, student designers need to execute several
iterative processes of problem formulation, generate alternative, analyze them, visualize
trade space, and address decision-making. The paper first describes students design space
exploration experiences, and concludes with our observations of the current challenges
and opportunities.
Keywords. Design space exploration; teaching; multidisciplinary; optimization; analysis.
INTRODUCTION
In the current practice, the process of designing neering, and construction. Students and industry
buildings is rapidly becoming more collaborative professionals must learn to work together to formu-
and integrated through the use of Computer-Aided late and construct design spaces in order and under-
Design and Engineering (CAD/CAE) technologies. stand performance trends and trade-offs to solve
However the use of these technologies in the early issues central to practice.
stage of design is limited due to the time required Geordia Tech’s curriculum demonstrates an im-
to formulate and complete design cycles. A new portant issue in digital design education. Georgia
class of technology, involving automated multidis- Tech’s Schools of Architecture, Civil Engineering,
ciplinary analysis and design space exploration is and Construction, offer a variety of courses in design
increasing by the order of magnitude of the number studio, design theory and process, computer-aided
of alternatives that a design team can generate and design (CAD), building information modeling, para-
analyze (Haymaker, 2011). This creates new chal- metric design, energy analysis, structural analysis,
lenges in the ways we educate tomorrow’s design- cost analysis, decision analysis. However our Insti-
ers and managers in schools of architectural, engi- tute lacks integrated courses that help students un-

Figure 1
Design professionals gave de-
sign challenges that students
formalized into multidiscipli-
nary building design space
exploration processes.
derstand how to work together to systematically for- and exploration processes. The curriculum engages
mulate, execute, and understand multidisciplinary students in a team-based approach to problem for-
building design spaces. mulation alternative generation, alternative analy-
Several organizations and associations such as sis, design space exploration and optimization, and
the American Institute of Architecture (AIA) Technol- trade-space visualization and decision-making.
ogy in Architectural Practice [1], the National Coun-
cil of Architectural Registration Boards (NCARB) METHODOLOGY
award for the integration of practice and education The methodology in this course consists of five
[2], the American Society of Civil Engineers (ASCE) phases that are described in more detail below:
excellence in civil engineering education teaching Problem formulation, alternative generation, alter-
workshop series [3] and the Associated General Con- native analysis, design space exploration and opti-
tractors of America (AGC) BIM Education program [4] mization, and trade space visualization and decision
support the efforts of academic programs to create making. The students utilize these phases to con-
and implement effective new curriculum that bring struct design spaces for the professional challenges
together students from multiple disciplines, indus- in the semester long group project.
try professionals, and advanced design technologies
to learn to address practical design challenges. To Problem formulation
address this need, some curriculums are emerging In this first phase, we engaged professional design-
in architectural schools such as Columbia University, ers to present challenges from their own practice
Harvard University (Kara and Georgoulias, 2013) and that they felt could have benefitted from more
University of Southern California and Stanford Uni- exploration if they were given more time and bet-
versity (Gerber and Flager, 2011). ter tools. Figure 1 and the following text describe
This paper describes new curriculum under de- the challenges presented by the design teams. The
velopment in Georgia Tech’s Schools of Architecture benefits of engaging design teams in this way were
and Building Construction that engages architec- twofold. It helped students confront real world de-
ture, engineering, construction, and computer sci- sign challenges without needing to spend too much
ence students and industry professionals in collabo- time gathering information about them. It also gave
rative multidisciplinary design space construction professional designers access to new design space

Figure 2
A preliminary Wecision model
developed by a group of
students to define the project
goals and objectives, includ-
ing the units and criteria for
measurement.
exploration tools and ways of thinking about their total square footage and aesthetic attributes.
challenges. • Case 3: Mixed-use tower
• Case 1: Cancer treatment center The tower in china was conceived with the vi-
A new cancer treatment process provides an sion of a “the Breathing Tower” that uses green
opportunity to develop a new design meth- energy techniques, including passive lighting
odology. The professional design team found and ventilation. The student’s goal in analyz-
the massing phase challenging because of the ing the design for the tower involves optimiz-
very large equipment involved with the new ing the quality and comfort levels of the occu-
treatment process. Several programming and pants. They look at performance criteria such
crane access issues constrained the potential as daylighting, passive ventilation, structural
solutions somewhat, but the design team was stability and attempt to preserve the grace and
interested in more systematically exploring the symmetries of the original design aesthetic,
tradeoffs of different building massing in terms while keeping costs at a minimum.
of their visibility from highway, energy con- Students first used Wecision’s Choosing by Ad-
sumption, daylight factor, sensitivity to adja- vantages model (Abrams et al., 2013) to model the
cent neighbors, and connection to adjacent organizations involved, the goals and constraints
green space. they needed to consider, the range of alternatives
• Case 2: Children’s hospital they wanted to explore, and the preferences on out-
The hospital, located in the Middle East, was comes (Figure 2). They also enter initial estimates
conceived to emphasize western healthcare of what they believe the outcomes are likely to be
ideas such as patient comfort, equality, and based on intuition.
external views. The students were asked to Students then developed Meta Model (MM) in
evaluate the current proposal and provide in- System Modeling Language (SysML) to describe the
sight into how the geometry and solar shad- structural and behavioral aspects (Reichwein and
ing could be modified to improve solar and Paredis, 2011) of their design challenges. The MM
day lighting performance, thermal gain, and is an abstract model of the data of the actual geo-
patient views. The design team focused on the metric model. It captures the structural aspects of
trade-off between designing for solar radiation the model such as domain specific semantics, attrib-
and day lighting factor; however, other factors utes and relationships among parts through block
contributed to the final evaluation including or class definitions. From these definitions multiple

Figure 3
A SysML block instance
diagram describes the data
blocks and relationship used
to generate alternatives for
the room dimensions in a
children’s hospital.
Instance Models (IM) of design alternatives can be tive parametric design models that are driven by the
generated by changing the parameters. The behav- design variables specified in the MM. In some cases
ioral aspects of the challenges are captured though custom scripting is also included to enable topologi-
activity diagrams that represent the sequence of cal transformations that are difficult to achieve using
actions to be performed in order to generate, ana- parametric logic alone. The students tested the par-
lyze and select a design alternative that describe the ametric model and generated different alternatives
generative and analytical systems in their design by modifying the variable values (Figure 5).
spaces. Students used commercial parametric design
They used SysML Block and Instance diagrams tools such as Rhino/Grasshopper, Revit, and Digital
(Figure 3) to describe the alternative’s components Project to generate the parametric model. Output
and relationships that will be important in the analy- of these tools would be a set of architectural forms
sis, and SysML Activity diagrams (Figure 4) to de- in which their geometry and properties are easily
scribe the analysis processes they wish to perform modified by changing the parameters.
on these models. In these diagrams they explore
and communicate the detailed input parameters for Alternative analysis
analysis tools, as well as the output parameters of In this third phase, the integration of their para-
the analysis, and whether they are to be minimized metric model with analysis tools allows students to
or maximized. analyze and evaluate the performances of different
alternatives in a design space and compare them
Alternative generation based on their performance metrics. To this end,
In the second phase, to represent the design alter- students need to integrate CAD and CAE tools in a
natives geometrically, students then made associa- way that the data flows between the tools in an au-

Figure 4
A SysML activity diagram de-
scribes the high-level integra-
tion and optimization process,
as well as detailed processes
for each goal analysis.
tomated fashion to reduce design cycle latency. The lop their own workflows, for example students in the
simulation and analysis tools were selected based high-rise group developed a customized workflow
on the performance objectives, inputs, and familiar- to minimize the total structural weight. The deve-
ity from among available commercial software such loped workflow is able to calculate the wind pres-
as EnergyPlus, Green Building Studio, eQuest, DIVA, sure on the façade based on ASCE 7-10, calculate
and IES VE for energy analysis, SAP2000, GSA Oasys, tip deflection on the top of the building, and modify
STAAD, Karamba, and ETABS for structural analysis, the columns’ cross section until achieving the most
Radiance, Ecotect, DIVA, and Daysim for Daylighting efficient column sections (Figure 7). Students in the
simulation. Figure 6 shows student daylight analyses Cancer Treatment Facility developed several geo-
comparing the original design team’s design with metric scripts to analyze designs automatically for
one of the alternative’s generated from their para- visibility from the highway, sensitivities to adjacent
metric model. buildings, and access to open space (Figure 8).
Students were introduced to experimental work-
flows such as ThermalOpt (Welle et al., 2011) and Design space exploration and optimization
BiOpt (Flager et al., 2013) that build in data trans- Due to the potential size and complexity of poten-
formations and strategies that help prepare models tial building design spaces, analyzing and testing
for fully automated simulations and contain domain for every parametric variation can be impossible.
specific knowledge necessary for more efficient op- Additionally, many of the design objectives are hard
timization. Students were also encouraged to deve- to formalize, and so it is often more fruitful to en-

Figure 5
A range of tower designs
generated from the student’s
parametric model.
able the designer and tool to work iteratively visu- how to use Pareto frontiers, performance trends,
alizing and generating aspects of the design space. and sensitivity analyses in order to make informed
Hence, in this fourth phase, the students learn to decisions in guiding the optimization process. They
apply computational techniques such as design of used the built in tools provided by ModelCenter and
experiments and use optimization and sensitivity Wecision. Figure 10 shows two examples of student
algorithms to systematically guide the generation approaches to exploring the multidisciplinary de-
of alternatives. Students used commercial design sign spaces.
exploration and optimization tools such as Octopus At the end of the class, students return to We-
and Galapagos by Grasshopper, and ModelCenter. cision to identify several prominent alternatives in
the design spaces they explored, and to report on
Trade space visualization and decision the multidisciplinary performance and weigh the
making importance of the advantages of each alternative.
The visualization of performance enables students At the end each alternative is evaluated based on its
to engage in computer-based exploration and visu- total advantages.
alize tradeoffs. In this final phase, the students learn
Figure 6
Students analyse alternative
designs for structural perfor-
mance, daylight, energy, cost
and more.

Figure 7
Students developed a custom
process for analysing high-rise
structural performance.
Figure 8
Students developed a custom
process for analysing the pro-
jects relationship to adjacent
green spaces.
CONCLUSION els to iteratively search through a space of designs

This paper described an exploratory class in which and negotiate to find the best and most sustainable
students from multiple disciplines worked with designs. We discuss several challenges in teaching
industry mentors, and learned how to formulate the class, and discuss ongoing work to overcome
design space exploration problems, parametrically them.
define alternatives, integrate CAD and CAE tools to
rapidly analyze alternatives, explore design spaces Improve team formation
and trade-offs, and make and communicate deci- Students appreciated the multidisciplinary teams in
sions. Students learned to build and integrate mod- which students have individual domain knowledge
Figure 9
Students developed an
optimization process, each
team found designs that
outperformed the industry
chosen design, for the objec-
tives analysed.

Figure 10
Students experienced with
different methods of exploring
and communicating design
space exploration informa-
tion.
and skills to contribute. Each teams requires an ap- understanding of student skills and interest.
propriate mixture and level of domain knowledge in
the programs as well as general computer scripting. Separate learning of concepts from apply-
In future versions of the class we plan simple tuto- ing concepts
rial exercises early in the class, and delay choosing We taught students the concepts and tools directly
teams a few weeks until we have developed a better in the context of the industry problems. This was
Figure 11
A final Wecision model that
communicates the multidis-
ciplinary advantages of a
selection of alternatives in the
design space.

difficult because of the large number of concepts metric ranges, allowing quicker model building and
students needed to absorb at one time, and the fact computational processing time. There are very few
that there was no ground truth to determine if stu- prior cases to look to understand best practices for
dents were applying the concepts “correctly.” We are how designers best frame design exploration prob-
creating a series of short tutorials for a simple exam- lems, how to choose the right objectives, and pa-
ple that teach students the fundamental concepts rameterization of the problem. Through better met-
of design space exploration. Students will begin to rics for describing challenges, and more case studies
learn about the industry problems early in the class, that illustrate good and bad problem formulations,
but will not begin to apply these concepts until they we would be able to improve the efficiency and ef-
have completed the introductory tutorials. fectiveness with which students formulate design
problems.
Improve integration of tools
While several workflows are emerging, some work- Improve the ability adopt and adapt the
flows are very complex, and require better docu- right strategy to the right challenge.
mentation for students to be able to apply them. Given a clearly understood set of objectives, stu-
Others are overly simple, and students quickly run dents have difficulty identifying and applying the
into objectives they cannot analyze, alternatives right strategy. We teach students to explore the
they cannot generate, and spaces they cannot ex- sensitivity of each input parameter and the influ-
plore because of limitations of one workflow or an- ence of weighting the different goals in each chal-
other. Another important area to improve integra- lenge guides to the next set of decisions. However,
tion is between meta modeling tools like Wecision we need better documentation of the strategies that
and MagicDraw with CAD and analysis tools. Despite are available, and better assistance in finding the
the limitations in terms of integration, the meta, right strategy for the right challenge.
parametric, analysis and decision models co-evolve
during the design exploration process. Not all the Improve the ability to assess and compare
knowledge integration occurs at the tool level. For explorations.
example, while the students implement the para- Ultimately as designers we want to choose the strat-
metric model based on the meta model they also egies that enable the best exploration, and as in-
evolve the meta model by adding, deleting or ed- structors we want to be able to evaluate and guide
iting attributes required by the actual parametric students towards ever better exploration, Research
geometric model. Therefore, the issue of integration is ongoing to define the metrics for assessing the
it is not only related to implementation of the inter- efficiency and effectiveness of design exploration
operability among tools, it is also related to the de- (Clevenger et al., 2011; Senescu and Haymaker,
velopment of co-evolution (Dorst and Cross, 2001) 2013). Development and integration of these met-
mechanisms and methods, since the different mod- rics into design systems will enable students assess
els are abstractions that represent only aspects of and compare their own explorations to those of oth-
the design challenge interacting with other aspects. er students and professionals on similar challenges.
Improve ability to systematically frame, ACKNOWLEDGEMENTS

define and formulate the challenge We would like to thank the students of the class: Eli-
This process is beneficial to the designer when there ah Cappi, Jonathan Cook, Kyungki Kim, Mark Luffel,
is a schematic idea, with strong initial intuitions for Jessica Marquardt, David Moore, Jaesuk Park, Natalia
effective performance. For time efficiency a well Quintanilla, Keyan Rahimzadeh, Philip Richardson,
thought out schematic will cut down on the para- Daniel Snider, Di Sui, Jun Wang and Tianyao Zhang.

We also want to thank the industry mentors includ- Kara, H and Georgoulias, A (ed) 2013, Interdisciplinary De-
ing Keith Besserud at Skidmore Owings & Merill sign: New Lessons from Architecture and Engineering,
(SOM) and Diana Davis, Bruce McEvoy, Mark Tagawa, Actar, New York, USA.
Dawn Mixon Bennet, and David Green and their col- Oxman, R 2008, ‘Digital Design as a challenge for design
leagues at Perkins+Will who helped formulate and pedagogy: theory, knowledge, models and medium’,
critique the design spaces. We thank the Digital Design Studies, 9(2), pp. 99-120.
Building Laboratory and the Georgia Tech Schools Oxman, R 2008, ‘Performance-based Design: Current Prac-
of Architecture and Building Construction for their tices and Research Issues’, International Journal of Ar-
support. chitectural Computing, 06(01), pp. 1-17.
Oxman, R 2009, ‘Performative Design: A Performance-mod-
REFERENCES el of digital architectural design’, Environment and Plan-
Abraham, K, Lepech, M, and Haymaker, J 2013, ‘Selection ning B: Planning and Design, 36(6) pp. 1026-1037.
and Application of Decision Methods On a Sustainable Reffat, R 2007, ‘Revitalizing architectural design studio
Corporate Campus Project’, Proceedings of the 21st An- teaching using ICT: Reflections on practical implemen-
nual Conference of the International Group for Lean Con- tations’, International Journal of Education and Develop-
struction, IGLC, Fortaleza, Brazil. ment using Information and Communication Technology
Chok, K, and Donofrio, M 2010, ‘Abstractions for informa- (IJEDICT), 03(01), pp. 39-53.
tion based design, International Journal of Architectural Reichwein, A and Paredis, C 2011, ‘Overview of Architec-
Computing, 08(03), pp. 233-256. ture Frameworks and Modeling Languages for Model-
Clevenger, C and Haymaker, J 2011, ‘Metrics to Assess De- Based Systems Engineering’, Proceeding of ASME 2011
sign Guidance’, Design Studies, 32(5), 431-456. International Design Engineering Technical Conference
Dorst, K and Cross, N 2001, ‘Creativity in the design pro- & Computers and Information in Engineering Conference,
cess: co-evolution of problem–solution’, Design Studies Washington, DC, USA.
22(5), 425-437. Roudavski, S 2011, ‘Selective Jamming: Digital Architectural
Flager, F, Welle, B, Bansal, P, Soremekun, G and Haymaker, J Design in Foundation Courses’, International Journal of
2009, ‘Process Integration and Design Optimization of Architectural Computing, 09(04), pp. 437-462.
a Classroom Building’, Journal of Information Technol- Senescu, R, and Haymaker, J 2013, ‘Evaluating and Improv-
ogy in Construction (ITcon), Vol.14, pp. 595-612. ing the Effectiveness and Efficiency of Design Process
Flager, F, Adya, A, Fischer, M and Haymaker, J 2011, ‘BIOPT: a Communication’, Advanced Engineering Informatics, pp.
Method for Shape and Member Sizing Optimization of 293-313.
Steel Frame Structures’, CIFE Technical Report, No. 202. Shea, K, Aish R and Gourtovaia, M 2003, ‘Towards Integrat-
Gerber, D and Flager, F 2011 ‘Teaching Design Optioneer- ed Performance-based Generative Design Tools’, Pro-
ing: A Method for Multidisciplinary Design Optimiza- ceeding of eCAADe 21, Graz, Austria, pp. 553-560.
tion’, Proceedings of the ASCE Conference: Computing in Welle, B, Haymaker, J and Rogers, Z 2011, ‘ThermalOpt: A
Civil Engineering, Miami, FL, USA, pp. 883-890. Methodology for BIM-Based Passive Thermal Multi-
Geyer, P 2012, ‘Systems modelling for sustainable building disciplinary Design Optimization’, Building Simulation,
design’, Advanced Engineering Informatics, 26(04), pp. 4(4), pp. 293-313.
656-668.
Haymaker, J 2011, ‘Expanding Design Spaces’, Academy of [1] www.network.aia.org
Engineering’s 2011 US Frontiers of Engineering Sympo- [2] www.ncarb.org
sium - Engineering Sustainable Buildings, Google Head- [3] www.asce.org/exceed
quarters, Mountain View, CA, September 19 – 21. [4] www.agc.org

Innovative Learning for Collaborative Design in
Ergonomics
Viviane Folcher1, Khaldoun Zreik2, Samia Ben Rajeb3, Pierre Leclercq4
1
C3U-CiTU Paragraph - University Paris 8, France, 2CiTU Paragraph - University Paris
8, France, 3,4LUCID-ULG University of Liège, Belgium
1
viviane.folcher@univ-paris8.fr, 2khaldoun.zreik@univ-paris8.fr, 3samia.benrajeb@ulg.
ac.be, 4pierre.leclercq@ulg.ac.be
Abstract. The proposed article deals with introducing collaborative architectural design
into the training of ergonomists at the Master 2 level. The collaborative design workshop
aims to confront ergonomists with the difficulties any design project involves, and which
challenge architects, designers, engineers and so on: collaboration between people with
different skills and different expertise; powerful time constraints; need for their work to
converge; working together and/or at a distance; sharing documents; decision-making,
etc. The article will present a short review of work carried out in the domains of
architecture and design, and of the contribution of ergonomics within architectural
projects. We shall then present the workshop’s educational aims, and give details of the
way it functioned. Finally, observation results will be presented and discussed.
Keywords. Collaborative design; architecture; ergonomics; training workshop.
COLLABORATIVE DESIGN LEARNING IN AN ARCHITECTURAL PROJECT
Architecture and design

An architectural project generally depends on a different skills and tools needed to bring the project
context – temporal, geographical, political and eco- to completion (Farel, 1995; Bucciarelli, 2002).
nomic. From the earliest design phases it includes Nowadays architects no longer work alone, but
different skills areas such as engineering, ecology, collaborate with engineers, landscapers, designers,
law, ergonomics and sociology (Hubers, 2009). It has economists and so on. So the project becomes col-
to respond to a program taking account of a certain lective, uniting a variety of skills and knowledge that
number of parameters – technical, social, environ- are all involved from the earliest design phase.
mental, industrial, legislative, political, etc. It devel- These particularities lead to consider design
ops in an increasingly restrictive regulatory frame- in terms of skills and collectivity (Cross et al., 1995;
work, and since the 90s a variety of research has Jeantet et al., 1998; Larsson, 2005; Bucciarelli 1988,
shown that it is rarely led by a single individual (Vis- 2002; Minneman, 1991). This means that the activ-
ser, 2002). Faced with competition, short deadlines ity of design combines collective actions, as well
and complicated quality and regulatory require- as the individual actions of different experts (Pahl
ments, architecture firms have to innovate, using et al., 1999). It is not simply a matter of assembling
interdisciplinary design methods that combine the points of view. It is the result of interaction between

the different components of the project, subjective remains outside the design process and the main
in character (Pousin et al., 1986) and the consensus players use what s/he gives them whenever they
achieved between all the players who participate in think it opportune. Is s/he a potential actor in the
designing the architectural project. design process, able to provide clarification about
Such collective activity also obliges partici- future and possible activity and to give structure to
pants to use instruments of sharing, without which the design collective? In this case s/he is included
it would be impossible (Boujut, 2000). The aim of in the design process, within the collective of those
most of these instruments is to facilitate cooperation involved, and takes full part in debate, orientations
between those involved in a project, by making it and the choices finally made (Beguin, 2004; Daniel-
easier for them to exchange information at different lou, 2004). The latter is our point of view (Folcher,
times and at a distance. They are only used, however, 2003) and the raison d’être of the workshop offered
when the project is sufficiently advanced. in the Master 2 professional training, whose main is-
The sketch-phase, when important design choic- sues we will spell out below.
es are made, continues to make little use of such in-
struments of collaboration. Yet it is essential for the EDUCATIONAL AIMS AND HYPOTHESES
start of design activity in an architectural project The training workshop in collaborative design’s first
that participants be able to meet despite distance, aim is to create a cross-disciplinary, collaborative de-
and communicate in real time. Sharing-instruments sign milieu with its own tools. The design approach
aim to facilitate exchange. Exchange is important in adopted is that of design-for-use (Folcher, 2010),
constructing what cognitive ergonomists call “cog- which relies on methodological tools from the field
nitive synchronization”, “temporal-operational syn- of ergonomics to achieve overall understanding of
chronization”, “mutual awareness” or “common op- projects and a broader exploration of design possi-
erative references”. All these concepts have already bilities.
been defined in the field of cognitive ergonomics The second aim is to accompany students of
and constitute “the backbone of the collective reso- ergonomics and design in the gradual construction
lution of a design problem” (Darses, 2004). of shared references for the possible forms of future
activity. In another words, it means allowing and
Ergonomics and architecture equipping the construction of common operative
Various works in ergonomics has explored different references (de Terssac and Chabaud, 1990).
positions for the ergonomist within architectural The third aim is to accompany and structure rap-
projects, wishing to reinforce the dialogue between id exchange between ownership and management
project management and ownership (Martin, 2000). of the project throughout its duration, all the way
Diverse experiences of participating in design pro- from the earliest schematic models to projections in
jects have made it possible to identify the positive physical space.
contributions of each position as well as the main The design exercise proposed in the workshop,
stumbling-blocks: weak ownership which delegates through a lived experience of collaborative design,
the whole project to its management; projects is conceived with a didactic goal and has certain
mainly oriented towards technical or financial possi- characteristics which ensure its ecology: a variety of
bilities; a residual place accorded to future and pos- actors express their potentially-contradictory points
sible human activity. of view on the project; collaborative design activity
The real question about the ergonomist’s contri- unfolds in two different situations (physically pre-
bution in the design process is that of his/her posi- sent, and at a distance); the exercise is time-limited;
tion. Is s/he a specialist, expert in the human factor, a final result in the form of a sketch is required.
called on at precise moments? In this first case, s/he Two hypotheses underlie the workshop:

Figure 1
Social analysis of the project.
1. The first is that collaboration in design requires expressed; and it makes explicit relevant elements
tools to enable designers to develop a broader of the context. Figure 1 provides a schematic illus-
vision (of the diversity of stakeholders and tration.
the plurality of issues involved) and construct The second tool is an enlarged method for ex-
shared references to the future, possible hu- ploring the possibilities in design, which are struc-
man activities when taking account of all the tured according to three broad types of contribution:
constraints that arise. 1. The project management contribution: explor-
2. The second, more exploratory hypothesis seeks ing questions about the will to change and cre-
to document the way the design collective ex- ate new things. They address the way the pro-
ists in two distinct situations: when physically ject is piloted and how it develops, considering
present and when working at a distance. all the elements deemed relevant: political,
strategic, financial, temporal, human;
PROPOSED METHODOLOGICAL TOOLS 2. The ownership contribution: exploring ques-
The design-for-use approach to collaborative design tions about how the will behind the project is
articulates two methodological tools. made concrete in the form of something via-
The first is a tool for social analysis of the project, ble. They guarantee the feasibility of achieving
used by practitioners of ergonomics to structure the project on various levels: technical, legal,
their interventions. It details the preoccupations security, ecological, human;
and/or problems expressed by each stakeholder, 3. The end users point of view contribution: ex-
and their issues; it presents the people likely to be ploring questions from the point of view of
concerned by the preoccupations and/or problems future, possible activity at the heart of the

project. They address the way general human Figure 2
characteristics are taken into account, as well School design outline
as the specific, multi-determined human activi- (extract).
ties in the precise situation.
This method takes the form of a table which stu-
dents complete as their developing projects unfold.
An extract from the method used in the projects will
be presented in the results section.
COLLABORATIVE DESIGN WORKSHOP:

PRESENTATION
Two groups of 5 students (ergonomists and design-
ers) took part in the collaborative design workshop,
accompanied by three student observers (ergono-
mists) for each group, so as to record the process 3. Collaborative design working physically to-
and analyze it at the end of the experience. gether;
Thus in this workshop students in ergonomics 4. Collaborative design working at a distance us-
were either designers or observers, whereas design ing SketSha software;
students exclusively played the role of designer. 5. Finalizing ideas and presenting the project.
Each group within the collaborative workshop was In the distance-working situation, designers
asked to design a secondary school for 240 pupils. were able to work simultaneously with the help of
They had to comply with a set of constraints in the an innovative technological tool, the Studio Digital
program they were given, and were invited to im- Collaboratif developed by LUCID of Liège University.
plement an innovative approach to design enabling It enables virtual working meetings to take place, at
them to imagine and design a school which would which participants can share spoken exchanges as
be “different” from the way schools are traditionally well as any sort of document they have brought –
conceived (Figure 2). these documents can also be modified and annotat-
Four stakeholders who originated the demand ed graphically in real time (Ben Rajeb and Leclercq,
for the school design set out their points of view 2012).
on the project, and the issues involved: the Mayor
of the town where the school would be located; a RESULTS: PROJECTS PRODUCED AND
member of the regional council with responsibility DESIGN PROCESS
for questions of disability and accessibility; a repre- In the time allowed, the two groups produced pro-
sentative of the building standards department; and jects that explained the main lines of orientation
the future head teacher of the school. Throughout and justified the choices they had made, and deliv-
the design procedure students were also helped ered a final graphic sketch:
and guided by two engineering-architecture train- • Group 1 opted to design an “innovative school
ers, who advised and helped them with the presen- that makes culture more visible and the town
tation of their ideas and thoughts. more attractive”. They called their project
1. The design procedure fell into four phases: ART’CADEMIE (Alexandre et al., 2012).
2. Forming and presenting the exercise, the work- • Group 2 focused on “a school for everyone, a
ing context, listening to the four stakeholders, place where activities can meet”. They called
and learning to use the SketSha software in the their project ECHOLOGIA (Houidek at al., 2012).
distance-working situation; Analysis of the way the design process unfolded

shows the existence of ‘moments’ which gave struc- produced, very rapidly led the group to focus
ture and rhythm to the collaborative design activity on detailed feasibility calculations for certain
in the groups (see Table 1): spaces (measuring a stairwell for example)
• Moment 1 “Brainstorming”: consisting in social to the detriment of the place and function of
analysis of the project on the basis of what these elements in the project. As a result of this
the stakeholders said, and construction of an distance, the tool and the type of project that
overall vision; initial structuring of the project’s they have not yet mastered, numerous conflicts
general orientation. Students were physically came to light between the project stakeholders.
present, producing documents to illustrate They had to adapt quickly to the conflict situa-
and synthesize their ideas. During this period tion that arose, and aim for a more global vision
an overall vision of the project is constructed of the project. This moment then led to anoth-
gradually: the stakeholders, the problems and er period centered on the general dimensions
preoccupations they express, and the people which had to be respected in the plan (e.g. the
likely to be concerned by those problems, are situation and dimensions of classrooms relative
at the heart of the project. This vision is the to the library and dining-room) rather than pre-
basis for preliminary sketches on paper, result- cise calculations of isolated elements.
ing from oral exchanges between the different • Moment 4 “Abolishing m²”: consisting in elabo-
designers, who have differing references and rating plans according to dimensional con-
experience. straints, rather than strictly complying with
• Moment 2 “Sketches taking shape”: consisting in criteria about m², with return to a more overall
further exchange on the project’s orientations, vision. Here, the students work at a distance
and drawing up initial schemes. Students were and produce finalized plans for each level. This
still physically present and produced drawings moment is when they return to the heart of the
and plans on paper. Students gradually make project, and takes the form of different sketch-
their sketches more concrete, adding meas- es, more and more finalized according to level,
ures and integrating the constraints of the plan and structured by the orientations defined dur-
given in the outline. Certain structure-giving ing the design process.
choices appear in the paper sketches, such as • Moment 5 “Reorganizing and finalizing project”:
arrows marked in red that indicate the inten- consisting in plans aligned with project orien-
tion to open up the school and facilitate people tations, space constraints examined, choices
movement towards the interior of the building. made in line with orientations. Students at a
• Moment 3 “Translating drawings and sketches distance, produced presentation aids. At this
into m² on the plan”: consisting in carrying out point, groups retook possession of the project
numerous calculations of space and levels, and were able to present it and make its guid-
transforming drawings into concrete plans. ing principles explicit, illustrating them with
Very detailed calculations (e.g. the height of a freehand drawings and SketSha sketches.
step) and difficulty in achieving them. Students All through the construction of their projects,
were working at a distance, and produced un- students practiced the method of enlarged design-
finished plans within each level. Moment 3 possibility exploration, which they interpreted in
makes it possible to pursue reflection around their plans and sketches afterwards, introducing site
the project, using SketSha for the sketches and program constraints. Table 2 presents an extract
and video-conferencing for oral exchanges. of this exploration work.
But the system, by offering a grill which allows The five moments that structured the collabora-
the addition of a graphic scale to the sketches tive design process bear witness to reorganizations

Moments Situation and tools Table 1
Moment 1 – Presence Five moments structuring the
Brainstorming Paper/pencil design process.
Moment 2 – Presence
Sketches taking shape Paper/pencil
Moment 3 – Presence
Translating drawings and Paper/pencil
sketches into m2 on the plan
Moment 4 – Distance
Abolishing m2 SketSha – video
conference
Moment 5 – Distance
Reorganizing and finalizing SketSha – video
project conference

Table 2 Objectives Project Ownership Solutions explored and selected
Enlarged method for exploring Management Contribution
design possibilities for Echolo- contribution <=> End user
gia school project (extract)). <=> End user contribution
contribution
Accessibility A school for - For all disabled Classrooms: arranged in blocks of 3 on 2
everyone: persons: wide levels to create fluid people-movement
-taking account of corridors, large - Room doors 1m wide for wheelchair
diversity entrance, ease of access
-simplifying access to all levels - Creating an identical, superposed access
people-movement - Avoid multiplying area on all floors
- respecting others means of access: Blocks: stairs/elevators, toilets and security
long corridors, post at an equal distance for everyone,
accessible toilets everywhere, and close to each classroom.
Ease of access:
Facilitating people-movement towards the
inside for everyone: all corridors in line for
greater security and to avoid getting lost +
accessibility for maintenance staff (disabled
or not).
of collective activity, notably under the effect of their arguments and opens them up for debate in
teachers’ interventions which guided or even reori- the group. New ideas gradually emerge and a con-
ented the designers’ work: sensus forms. When distance working, confronta-
• Help with the initial definition of the school tions still relate to the project’s main ideas but are
project: what school? For whom? expressed by one pair towards the other situated at
• Help with more detailed definition: a school for a distance.
all; each future user’s activity; accessibility; Withdrawal. When physically present, with-
• Reorientation: from detailed calculation of ele- drawal takes the form of a less active role for one
ments towards the overall meaning of the proof the participants, which has several functions: in-
ject: placing sketches in the plan, rather than dicating disagreement, or the wish to start another
precise calculations about the size of a stair- activity, related to the activity going on (e.g. making
case. a drawing while the group progresses with produc-
ing ideas). This type of withdrawal turned out to be
COLLABORATIVE DESIGN WHEN PHYSI- productive, as it makes it possible to share the draw-
CALLY PRESENT, AND AT A DISTANCE ings which fuel the ideas produced.
The work of the observers made it possible to fol- When distance working, withdrawal took the
low the collaborative design process and identify form of disappearing from the camera angle. This
the specific aspects of each situation and the role was less comprehensible and thus less productive in
played by different artifacts (Belaitouche et al., 2012; terms of taking the collective work forward.
Mateev et al., 2012). Speaking and decision-making. When physi-
Conflict. When physically present, confronta- cally present, the flow of speech enables a certain
tions relate to the main ideas of the school project proliferation of ideas. People occasionally talking
and are expressed individually: everyone sets out over one another can be dealt with in the situation.

Decision-making happens before action: consensus sequentiallity can be observed in activities, as they
is developed orally and is then translated into action. develop step by step with the progress of exchanges
When distance working, for reasons of com- and as the designers pass the SketSha pen from one
prehension, the pairs have to alternate speaking to another.
to each other in the form of questions and an-
swers. Decision-making seems to follow the op- CONCLUSION
posite movement from that observed when physi- In the first part of the paper we pointed out char-
cally present: action takes place before agreement acteristics of design situations and the importance
is reached, which can generate disagreement and of collaboration, given multiple players: the impor-
even conflict within the collective. Actions carried tance of communication, the nature of tasks and
out before oral agreement can include erasing part how to distribute them, individual skills, negotiation
of a drawing or alternative propositions. They can procedures, conflict management and synchroniza-
serve a function of collegial decision-making, or the tion (Darses and Falzon, 1994). The ergonomic col-
imposition of an idea. laborative design workshop enabled our students
Diversity of activities. The work of design is as to encounter complexity of design activity and
much a matter of exchange and debating ideas as collaboration. This innovative learning approach
it is of drawings, sketches, tables and plans. When involves not only questions related to process and
physically present these diverse activities – exchang- the “how-to” of design, but also questions related to
ing, writing, drawing – coexist without difficulty and the different types of contribution which have been
mutually enrich one another: a quick sketch can be articulated throughout the project, namely project
produced to support an idea. At a distance, on the management contribution, the ownership contribu-
other hand, it seems necessary to sequence activi- tion and the End users point of view contribution.
ties, and this takes the form of a different way of or- The objective of the design methodology we
ganizing activity: expressing an idea, then writing proposed is to create unity of design and to allow
or drawing – or developing a drawing and then ex- diverse constraints to be expressed in an equivalent
plaining it. manner: constraints related to feasibility, and the
Using artifacts. When physically present, de- will of stakeholders, go alongside those which deal
signers make use of a wide variety of artifacts – with the planned end users. The End user point of
sheets of paper, notebooks, pens, pencils, felt-tip view contribution plays a mediating role as it fertiliz-
pens, erasers – which enables them both to express es different points of view, and makes it possible to
numerous ideas and to withdraw in order to further anchor propositions in human-scale reality. It struc-
develop an idea or drawing. It would appear that tures the whole project, from the first proliferation
these artifacts serve a function of collaborative work of ideas and drawings up to its finalization in the
just as much as individual, withdrawn work in the form of a sketch and/or plan.
service of the collective work. Through being involved in the design process,
At a distance, work done on paper has been observing and analyzing it, our students were able
scanned and designs and drawings are available in to realize that this sort of architectural design situ-
digital form. Pens, erasers and felt-tip pens take the ation has its advantages and its limits. They none-
form of the SketSha software pen. This pen has sev- theless expressed the view that it would have been
eral functions: it can write like a pen or felt-tip pen, better to include architecture students too, so that
erase, place and displace items on the plan. In ad- they could contribute different expertise and points
dition there is only one pen, so that whichever de- of view; this would have enriched their ideas, reflec-
signer happens to be holding it has considerable tions and questionings about both about the pro-
power to transform the project. Here too a certain ject itself and about the collaboration process.

REFERENCES
Alexandre D., Ali Ahmad, Adam J., Quattrociocchi A (2012) when design-for-use meets design-in-use’, Interacting
Art’Cadémie, une école pour vous, Project realized dur- With Computers, 15, pp. 647-663, Elsevier.
ing the collaborative design workshop, Master 2 Er- Folcher V 2010, ‘Développement des hommes et des tech-
gonomie et conseil psychologique, University Paris 8. niques, perspectives de conception pour et dans
Unpublished document l’usage’, HDR Ergonomie, University Paris 8.
Beguin, P 2004, ‘L’ergonome acteur de la conception’, in Er- Houidek L., Jarry A., Lantz M., Lambert A. (2012) Echologia,
gonomie, P Falzon (ed), Paris, PUF, pp. 376-390. Project realized during the collaborative design work-
Belaitouche M., Hammouchene R., Wagner C. (2012) Ob- shop, Master 2 Ergonomie et conseil psychologique,
servations, Project realized during the collaborative University Paris 8. Unpublished document.
design workshop, Master 2 Ergonomie et conseil psy- Hubers JC 2009, ‘Collaborative design in Protospace 3.0’,
chologique, University Paris 8. Unpublished document. in Wamelink H., Prins M., Geraedts R., ‘Changing roles;
Ben Rajeb, S. and Leclercq, P 2012, ‘Quelles collaborations new roles, new challenges’, TU Delft Faculty of Archi-
distantes synchrones dans les pratiques de conception tecture Real Estate & Housing, Delft, www.changing-
architecturale ? : Analyses comparées des pratiques de roles09.nl.
conception assistées par la visioconférence, le partage Jeantet A and Boujut JF 1998, ‘Approche socio-technique‘,
d’écran et le Studio Distant Collaboratif’, in Echelles, Es- in Tollenaere M (ed), Conception de produits mécan-
paces, Temps, 01Design’8, Bruxelles. iques, Méthodes, modèles et outils, Hermès.
Boujut, JF 2000, ‘Intégration produit-process en concep- Larsson A 2005, ‘Engineering Know-Who: Why social con-
tion : organisation et outils’, report, PROSPER program: nectedness matters to global design teams’, Ph.D the-
Systèmes de Production, Stratégies, Conception, Ges- sis, University of Technology of Luleå.
tion. Martin C 2000, ‘Maitrise d’ouvrage, maitrise d’œuvre, con-
Bucciarelli, LL 2002, ‘Between thought and object in engi- struire un vrai dialogue’, Toulouse: Octarès Editions.
neering design’, in Design Studies, 23 3, pp. 219-231. Mateev C., Sangare L., Tichadou C. (2012) : Art’Cadémie:
Cross N, Clayburn Cross A 1995, ‘Observations of teamwork observations from group 1, Project realized during the
and social processes in design, Design Studies, 16 (2), collaborative design workshop, Master 2 Ergonomie et
pp. 143-170. conseil psychologique, University Paris 8. Unpublished
Daniellou, F 2004, ‘L’ergonomie dans la conduite de projets document
de conception de systèmes de travail’, in Ergonomie, P Mineman S 1991, ‘The Social Construction of a Technical
Falzon (ed), Paris: PUF, pp. 359-373. Reality: empirical studies of group engineering design
Darses, F 2004, ‘Processus psychologiques de résolution practice’, Ph.D. Dissertation, Stanford University.
collective des problèmes de conception: contribution Pahl G, Frankenberger E and Badke-Schaub P 1999, ‘Histori-
de la psychologie ergonomique’, HDR Psychologie Er- cal background and aims of interdisciplinary research
gonomique, University Paris V- René Descartes. between Bamberg, Darmstadt and Munich’, in Design
Darses F and Falzon P 1994, ‘La conception collective : une Studies, 20 (5), pp. 401-406.
approche de l’ergonomie cognitive’, in Coopération et Pousin F 1986, ‘Concevoir et visualiser : la représentation en
Conception, Toulouse, pp. 1-12. question. Les nouvelles images et la CAO’, in Le Carré
de Terssac G and Chabaud C 1990, ‘Référentiel opératif Bleu, 2/3.
commun et fiabilité’, in Leplat J., de Terssac G. éds., Les Visser W 2002, ‘Conception individuelle et collective: Ap-
facteurs humains de la fiabilité dans les systèmes com- proche de l’ergonomie Cognitive’, in M Borillo, JP Gou-
plexes, Octarès, Toulouse, pp. 110-139. lette (eds), Cognition et création: Explorations cognitives
Folcher V 2003, ‘Appropriating artifacts as instruments: des processus de conception, pp. 311-327.

Shape Grammars
Shape Grammars - Volume 2 - Computation and Performance - eCAADe 31 | 615

616 | eCAADe 31 - Computation and Performance - Volume 2 - Shape Grammars
Unambiguity
Difficulties in communicating shape grammar rules to a digital interpreter

Thomas Grasl1, Athanassios Economou2
1
SWAP Architects, Austria, 2Georgia Institute of Technology, USA
1
http://plus.swap-zt.com, 2http://www.coa.gatech.edu
1
tg@swap-zt.com, 2athanassios.economou@coa.gatech.edu
Abstract. The rule editor of a parametric shape grammar interpreter is presented. The
problems that arise are discussed along with their solutions.
Keywords. Shape grammar; parametric shape grammar editor; implementation.
INTRODUCTION
Shape grammar implementations, in theory and with binary devices. In the context of a rule editor
praxis, have received increasing attention over the ambiguity will result in little more than serendip-
last years. Recent projects have been introduced by ity. Despite the successes of serendipity, here we
Yazar and Colakoglu (2007), Trescak et al. (2009), Yue will try to restrict it to a minimum. Rather it is dur-
et al. (2009), Keles et al. (2010; 2012), Jowers and Earl ing the match finding process, while decomposing
(2011) and Grasl (2012). Each project offers some a shape into its constituent parts, that ambiguity has
solutions to the general problem, but few imple- its place.
ment or describe rule editing capabilities. A com- As simple as many rules may seem to the human
plete shape grammar interpreter should support mind when drawn on paper and explained by an
emergence, parametric rules and rule editing via a accompanying text, it is quite a different matter if a
graphical editor. Here the results of such an effort digital computer is the entity trying to understand.
(Grasl and Economou, forthcoming) are described It is fantastic how many details the human mind can
with a focus on the peculiarities of the parametric overlook unbothered, how many blanks are filled in
rule editor. on the fly in order to understand a rule.
In order to implement a rule editor for a shape Of course if one were to sit down and describe a
grammar interpreter several difficulties have to be rule in computer code some ambiguities might fall
overcome. Mark Tapia (1999) already introduced away, most likely though it will take several attempts
some ideas concerning the user interface, most of until the desired result is achieved. In any case this is
which still hold today. However, the editor described not the desired solution, rather the designer should
by Tapia did not support parametric rules. be able to communicate with the computer in a
Although ambiguity is sometimes seen as a more intuitive language, that of the drawn shape.
strength of shape grammars (Stiny, 2006), it is not This is difficult enough for a non-parametric, or rig-
during rule definition that it should come into play. id, shape, but once parametric matching is allowed
Here unambiguity is essential. This is perhaps the things become all the more complicated.
dilemma of trying to implement shape grammars

EDITOR Figure 1
Rule 1: Copy and translate a
Conventions square along half its diagonal.
Like Tapia proposed, it has turned out that draw-
ing the left hand side (LHS) and the right hand side
(RHS) on top of each other, rather than on their re-
spective sides of an arrow, is advantageous.
If for example rule 1 (Figure 1) cannot be de-
scribed by absolute changes, like:
‘Move the square 2 units to the right and 2 units the sense that its verbal description could for exam-
up’ ple be:
but rather has to be captured in relative terms like: ‘Rotate a square by 45° and then scale the result by
‘Translate the square by the vector AC ⁄ 2’ 1/√2’
it is clearer if they are overlaid. Clearly all the param- or
eters used for such a description have to be derived ‘Inscribe into the square ABCD a second square
from the LHS, since the RHS does not exist at the EFGH such that AE = EB, BF = FC, CG = GD, DH = HA’
time of rule application. In a sense the RHS has to For a normal rule it does matter which descrip-
be rigged onto the LHS and overlaying makes this tion is selected. If however the rule is parametric, in
clearer. the sense that it can be applied not only to squares,
Colours are according to what is used for con- but also to rectangles and other quadrilaterals, the
struction documents of a renovation project. The difference is meaningful. If the rules
LHS is drawn in black, the colour of existing objects ‘Rotate a quadrilateral by 45° and then scale the re-
that are not altered. New objects, in our case the sult by 1/√2 ’
RHS, are drawn in red. Objects that are deleted are or
yellow. Figures in this paper follow a slightly differ- ‘Inscribe into the quadrilateral ABCD a second qu-
ent convention: Existing, unaltered lines are dark adrilateral EFGH such that AE = EB, BF = FC, CG = GD,
grey, new lines are black, deleted lines are not used, DH = HA’
but could be light grey. are applied to a rectangle the outcome will be very
different (Figure 2b).
Rule interpretation Additionally a distinction is made for rules that
Although it was attempted to keep the interface as leave the LHS unaltered
graphical as possible, it did turn out that making a a→a+b
few categorisations via drop-down lists simplifies
the interpretation of the rule. Figure 2
Not all rules can be handled in the same way. In (a) The parametric rule 2 and
this respect the schemas introduced by Stiny (2006) (b) varying results depending
have proven to nicely organise the rules. Of course on its interpretation.
the general schema
a→b
has to be covered. In addition a distinction was
made for rules that use transformations only
a → t(a)
This distinction is a necessity, to deal with rules
similar to rule 2 (Figure 2a). Rule 2 is ambiguous in

Table 1 Position Length Angle & Ratio Parallelism Cross ratio Neighbourliness
Possible mappings and their Identity • • • • • •
meaning. Isometry • • • • •
Similarity • • • •
Topology •
and triangulation ties together disjoint shapes (see rule

a → a + t(a) 3; Figure 3) and helps to avoid having to deal with
respectively. The first two schemas could also deal angles and trigonometric functions. The triangles
with these cases. It is more a matter of convenience are then constrained according to the selected map-
to prevent having to redraw the LHS. ping as shown in Table 2.
The mapping used for match finding is another
category the designer has to select. Essentially the Symmetry constraints
permissible transformations are defined. Rules are Finding a match in all its symmetric variations is a
assigned one of four mappings: Identity, Isometry, task assigned to the shape grammar engine rather
Similarity and Topology (Table 1). than the editor, and is described elsewhere (Grasl
The first, ‘Identity’ is included for completeness and Economou, forthcoming). Sometimes however
sake. Scale, rotation and position must match ex- too many variations of a match can be returned.
actly. This mapping is normally not used for shape This is the case if the rule itself exhibits some sym-
grammars, but it is not unthinkable. metry and thus cancels out some of the prospective
‘Isometry’ is a mapping where rotation and matches. Here the editor can help by detecting such
translation is allowed. Shapes must have the same symmetries of the rule and by adding appropriate
lengths and angles. constraints to the description. For this purpose an
‘Similarity’ allows for scaling in addition to rota- algorithm as described by Wolter et al. (1985) was
tion and translation. This is the default mode, as it is implemented.
also the most common method of matching shape For example, rule 1 (Figure 1) shows a rule that
rules. has mirror symmetry through [AB]. Normally the en-
‘Topology’ requires only the structure, the gine would return each square in eight variations,
neighbourliness, to match. It is the most flexible the eight isometries of the square. In case of rule 1
mapping and the basis for truly parametric rules. however each left turning variation will have one
right turning variation that will result in the same
CONSTRAINTS shape. Here a constraint on rotation is added by the
editor
Mapping constraints (AB × AC) ∙ z > 0
Rules of all mappings except for ‘Topology’ are tri- restricting the cross product of two vectors to a pos-
angulated over the set of vertices; this is for internal itive number.
use only and is not displayed to the designer. The
Manual constraints
Figure 3 In addition to these automatic constraints rules can
Rule 3 and its triangulation also be constrained manually. The web based imple-
that is generated for some of mentation does not allow for constraints to be de-
the mappings. fined graphically, instead each vertex is marked with
an alphanumeric value; these identifiers are then
used to formulate constraints mathematically.

Topology No automatic constraints. Table 2
Similarity The proportions of the edges of the triangulation are constrained. The constraints that are added
Isometry In addition to the proportions one length is constrained to an absolute value. automatically depend on the
Identity The coordinates of all vertices are constrained. chosen mapping.
Manual constraints are important to model possible. Again things are simplified by offering this
specific requirements that cannot be covered by a additional possibility.
mapping alone. If for example a rule should be ap-
plicable to rectangles of all proportions, then the REFERENCES
topology mapping hast to be used in combination Grasl, T 2012, ‘Transformational Palladians’, Environment and
with geometric constraints restricting a quadrilat- Planning B: Planning and Design, 39(1), pp. 83 – 95.
eral to a rectangle. Grasl, T and Economou, A (forthcoming) ‘From topologies
to shapes: Parametric shape grammars implemented
CONCLUSION by graphs’, Environment and Planning B: Planning and
Implementing a general parametric shape grammar Design.
interpreter is not an easy task. Once it is achieved Jowers, I and Earl, C 2011, ‘Implementation of curved shape
the question arises of how to feed the interpreter grammars’, Environment and Planning B: Planning and
with rules. Of course the rules could be formulated Design, 38(4), pp. 616 – 635.
directly in the underlying representation used by Keles, H Y, Özkar, M and Tari, S 2010, ‘Embedding shapes
the interpreter. In the case at hand this would mean without predefined parts’ Environment and Planning B:
describing the rules as graph grammar rules in the Planning and Design, 37(4), pp. 664 – 681.
description language provided by the graph gram- Keles, H Y, Özkar, M and Tari, S 2012, ‘Weighted shapes for
mar engine. While this is indeed a flexible option, embedding perceived wholes’ Environment and Plan-
and possibly the best approach for some very specif- ning B: Planning and Design, 39(2), pp. 360 – 375.
ic rules, in general it is desirable to be able to specify Stiny, G 2006 Shape: Talking about Seeing and Doing, MIT
rules graphically. Press.
The approach presented here builds on the un- Tapia, M 1999, ‘A visual implementation of a shape gram-
derlying graph representation of shapes. Neverthe- mar system’ Environment and Planning B: Planning and
less most of the findings should be application to Design, 26(1), pp. 59 – 73.
other implementations as well. Trescak, T, Esteva, M and Rodriguez, I 2009, ‘General shape
Defining the required mapping is not essential, grammar interpreter for intelligent designs gener-
since the same effect can be achieved by manu- ations’ in B Werner (ed) Proceedings of the Computer
ally adding constraints. However, this is tedious and Graphics, Imaging and Visualization 2009, pp. 235–240.
more often than not a mapping will suffice. Wolter, J D, Woo, T C and Volz, R A 1985, ‘Optimal algorithms
Manual constraints are of course an absolute ne- for symmetry detection in two and three dimensions’,
cessity for a parametric shape grammar interpreter. The Visual Computer, 1(1), pp. 37 – 48.
Visual constraint definition would be nice, but for- Yazar, T and Colakoglu, B 2007, ‘QSHAPER’ in J B Kieferle and
mula based constraints have proven to be a good K Ehlers (eds) Predicting the Future: 25th eCAADe Confer-
and flexible solution. Visual constraints can be add- ence, Frankfurt am Main, Germany, pp. 941-946.
ed fairly easily if the editor is implemented in a CAD Yue, K, Krishnamurti, R and Grobler, F 2009, ‘Compu-
environment that supports such constraints. tation-friendly shape grammars: Detailed by a
Treating rules of the schemas a → b and a → sub-framework over parametric 2D rectangular shapes’
t(a) differently could perhaps be avoided if complex in T Tidafi and T Dorta (eds) Joining Languages, Cultures
rule definitions based on constructive geometry are and Visions: CAADFutures 2009, pp. 757- 770.

Customized Cork Façade
A generative design process based on shape grammars

Rui Marques1, Sara Eloy2
1
ISCTE-IUL, Portugal, 2ISCTE-IUL / ADETTI-IUL, Portugal
1
rui20silvamarques@hotmail.com, 2sara.eloy@iscte.pt
Abstract. The propose paper presents an ongoing research which main goal is to use
cork in a customized modular façade system. Cork is used due to its ecological value,
renewable characteristic, insulation properties and aesthetic value. The modular system
design is bio-inspired in the microscopic cork pattern and the study aims at reproducing
in the façade some of the natural characteristics that enable cork to be suitable for the
function it plays in construction. Façades are design by a generative design process
based on a parametric shape grammar which encodes shape rules and an algorithm to
guide the generation. The developed cork modules are part of a back-ventilated façade
system which is assembled upon a substructure that reproduces the cork cell structure
and enables both the assemblage of the modules to the support wall and the connection
between them.
Keywords. Shape grammar; generative design; cork; façade; digital fabrication.
INTRODUCTION
The use of a generative design system enable the quirements and still maintain production costs.
generation of multiple solutions based on different However architectural quality is not absolutely
scenarios and requirements that would introduce measurable, there are some specific qualities that
different variables to the system. are well measurable. Energy efficiency is one of
Recently both the fabrication and the design those qualities. The use of materials with good insu-
process in Architecture are being questioned by lation values and the optimization of window open-
the use of digital technologies for the promotion ings according to the site insulation characteristics
of more efficient buildings. Requirements such as will improve a specific type of building quality.
good structural or thermal performance and cus- Kroes et al. (2008) state that the emphasis upon
tomization are the cause of the arising of new gen- building performance brings the architecture world
erative processes. much closer to engineering design. According to
Design assisted by generative processes such as Gruber (2011) the quality of a final project is defined
shape grammars allows the customization and op- by the quality of investigation conducted in the im-
timization of solutions by manipulating parameters. portant stages of design. The challenge proposed
Combining these processes with new digital fabrica- is to combine the expanded vision of the architect
tion techniques enable new products to be design with the fulfillment of specific variables using also
which are customized, respond to pre-defined re- quantitative criteria rather than just the qualitative

criteria typically used in the architectural work. or darker according to weather conditions. Both
Gruber (2011: 49) states that “in architecture, natural and black agglomerate cork is used in con-
problems in design often affect many levels of the struction because of its very good performance as
project, and often they are difficult to define.” As a acoustic, thermal and vibration insulation (Gil 1996).
consequence several tools are too specific to be use These materials are available in sheets or boards
in architectural problems since they are usually not with variable thickness and their manufactured is
looking to all the levels of the problem but only to mainly done with “granulates from cork stripping
some of them. However, these tools may be suitable (virgin cork) obtained from pruned branches of the
to solve specific design tasks and questions (Gruber cork oak tree” (Gil 1996).
2011: 49). The microscopic pattern of cork (Figure 1) was
used as an inspiration pattern for the develop-
DESIGN PROBLEM AND GOALS ment of the façade modular design with the aim of
The problem this research wishes to address is mimetizing Mother Nature’s harmony by using simi-
how to design a system to generate customizable lar logics and principles of organization. Through
façades with a natural material as cork without be- Hooke’s observations cork was found to be made of
coming to expensive. cavities (cells) full with air which enable the material
The main goals of this research are: to float, be firm and yet compress under force. How-
• Use pure cork agglomerate as a building coat- ever there are also structures around the air cavities
ing material and explore its thermal isolation which support the material (Humes s.d.). This metic-
and aesthetics potential; ulous natural design has inspired and lead to differ-
• Define a generative design process bio-in-
spired in cork composition based on shape Figure 1
grammars which meets a visual-performative Drawing of the structure of
language; cork as it appeared under the
• Define a modular cork façade system which microscope to Robert Hooke
meets a variety of thermal requirements con- (http://en.wikipedia.org/wiki/
sidering different contexts; Robert_Hooke).
• Define a substructure to support the cork mod-
ules in a back-ventilated façade system.
Besides those, other motives were to explore
architecture performance and mass-customization,
materials and construction technologies as well as
CAD/CAM digital technologies.
FRAMEWORK
Cork pattern, materiality and visual char-

acteristics
Cork is a natural, renewable and environmental-
friendly product that comes from the renewa-
ble bark of the Cork Oak. Natural cork agglomerate is
used in this research because of its excellent thermal
isolation properties and because it has great proper-
ties to be used at exterior façades becoming lighter

ent patterns in the proposed design. (bi and tri-dimensional) and the constraints needed
The choice of materials in architecture have an to design a façade. Shape grammar languages do
immense effect on the behavior of the people us- not look for one solution to a given problem but for
ing a space because of their functional performance multiple solutions based on the same set of rules
and their image, brand, identity feel and the atmos- or criteria (Eloy 2012). In shape grammars, rules are
phere they give to the space (Gagg 2012). Beside the used as mechanisms for generating designs.
enunciated measurable advantages of cork regard-
ing its technical specifications and performance, its DESIGN GENERATION
aesthetic may give a sense of warm and comfort
to users which we want to explore when using it in Phase 1: Cork as pattern
new ways. The process started by the use of the microscopy im-
age of cork and with the identification of its pattern
Generative design and shape grammar by visual analysis. From that stage a vector draw-
Recently architecture and biology are overlapping ing was done over the cork image and a triangular
fields of research by exploring different methodolo- module emerged as the simplest shape and the one
gies of translating knowledge gained from nature which enabled the most infilling diversity.
into technical solutions (Gruber 2011). Generative During this phase different shape approaches
processes of design in architecture are being stud- have been followed resulting in different shape vo-
ied and the arising of emergent shapes represent a cabularies and shape rules. It was considered since
current research line both in architecture and other the beginning that the modular façade construc-
areas such as computation sciences and biology. tion should consider two types and stages of work:
Emergent patterns may be observed, recognized a first stage involving industrial fabrication of the
and extracted from natural patterns and mathemati- base modular pieces and a second stage involving
cally explained through algorithms or parametric a customized design and fabrication by digital tools
shape grammars. The goal is the generation of effi- on those pieces. These criteria lead us to develop
cient architectural structures inspired by the logics the design grammar by using the simplest shape
of nature performance. Bio-inspired design can be possible to the base modular piece that will be re-
used in architecture in different ways, from the use peated. This triangle is use both for the surface pat-
of stylistic and aesthetic nature’s logic to the use of tern of cork modules and for the load bearing façade
nature’s performance potentials as the ability to use substructure inspired in the cork’s air cells and edge
in an efficient way the solar radiation. structure discovered by Hooke.
Shape grammars are generative processes deve- It was our goal to explore the visual complexity
loped since the 70s by George Stiny and James dynamic of the microscopic pattern of cork and use
Gips (1971). They are “algorithmic systems for cre- it dynamically in a building façade to stress cavities
ating and understanding designs directly through (windows), dark/light or shades (material thickness)
computations with shapes, rather than indirectly and strength (support structure).
through computations with text or symbols.” (Knight
2000) Phase 2: Encoding the rules
The generation process enabled by shape gram- The inference of shape rules was done by hand af-
mars allows for multiple designs to be generated, ter analyzing the microscopic patterns of cork. In
based on a single language but determined by dif- the first stage the geometric shapes were identified
ferent choices (Eloy and Duarte 2012). The use of a through the process of isolating possible combina-
shape grammar enables to encode both the shape tions. The goal was that the modular shapes could
rules that explore the visual properties of the pattern be easily fabricated and assembled and simulta-

neously would enable the generation of multiple support it. This metal substructure, located at the
combinations of design. This was done by using a borders of the modules, enables both the assem-
triangle as the base module. The adding rules that blage of the modules to the support wall and the
introduces a second triangle in the façade uses the connection between them.
left part of the rule and adds a new triangle by mir-
roring through one of its sides (rule 2 and rule 3, Fig- Phase 4: Prototype
ure 2). The final phase will be the test of the customized
After developing rules that fill the façade with cork façade in 1/2 scale models by using CNC digital
the base module the second stage was the defini- fabrication techniques. Two possibilities will be test-
tion of the shape geometry inside the triangle. In be- ed in triangular pieces, cork with 20cm of thickness
tween side by side triangles, these inside geometry and two layers of different cork with a total thick-
should generate the polygons that are characteristic ness of 20cm. The goal is to test the assemblage of
from the cork microscopic pattern. To both gener- the modules both between them and to the metal
ate polygons and make them dynamically different substructure and, if necessary, to go back and re-
from each others, a parametric rule was added to view unresolved situations detected during the fab-
the shape rules 4 and 5 in Figure 2. rication.
The third stage was the definition of several
shape rules which encoded the thickness possi- CONCLUSIONS
bilities as well as rule conditions (Figure 2, rules 6 to At this point of research we can preliminary con-
10e). clude that it is possible to interpret and infer pat-
These rules allow the use of Euclidean transfor- terns of nature using a shape grammar and that we
mations as symmetry, rotation and translation to the can use this grammar to generate similar patterns.
generation of different designs. The proposed design is bio-inspired by the mi-
The developed rules obey to two main criteria: i) croscopic structure of natural cork. The understand-
thickness differs with solar orientation of the facade; ing of this pattern led us to re-interpret it and design
ii) one base module has always two different thick- the proposed façade in a two layer system consider-
nesses. ing the loadbearing structure (edges) and the infill/
coating material (cells).
Phase 3: Design generation The use of a modular cork façade system which
The generation begins in a vertex of the façade and can be generated by a generative process allows a
the use of shape rules allows the all surface to be more rational design process since it enables multi-
filled of modules. Rule’s application follows criteria ple design to be developed but establishing modu-
like: different thicknesses in north/south and east/ lar standards that has to be obey.
west facades; diversity in the thickness in adjacent By addressing an experimental issue this re-
modules; non-repetition in adjacent modules; high- search helped to understand the potentials of re-
er thickness in more exposed places of the façade; lating bio-inspired design, generative process and
windows openings/voids position. digital fabrication techniques. With the use of these
Figure 3 show a possible layout of façade that technologies more sustainable structures can be
was generated through the developed shape gram- obtained which will meet nowadays and future re-
mar. quirements.
The coating cork modules are part of a back-
ventilated façade system which is assembled upon REFERENCES
a triangular substructure that disappears from the Eloy, S 2012, A transformation grammar-based methodology
exterior face of the coating to the back in order to for housing rehabilitation. PhD thesis, Universidade Téc-

Figure 2
Simplified shape rules and
sample derivation.
nica de Lisboa IST. eCAADe Conference - Volume 1, Czech Technical Uni-

Eloy, S; Duarte, J.P. 2012, ‘Transformation Grammar for versity Prague, 12-14 September 2012, 471-478
Housing Rehabilitation: from a specific to a general Gagg, R. 2012, Texture + materials. AVA Publishing. Gruber, P.
grammar’. Achten, H.; Pavlicek, J.; Hulin, J.; Matejdan, 2011, Biomimetics in Architecture: architecture of life and
D. (eds.), Digital Physicality - Proceedings of the 30th buildings. Vienna, Springer.

Figure 3
Examples of one final design
of the customized cork system
façade, with two different
types of cork or just one.
Gil, L 1996, ‘Densification of black agglomerate cork boards Sousa, J.P., 2010, From Digital to Material: Rethinking Cork in
and study of densified agglomerates’ in Wood Science Architecture through the use of CAD/CAM Technologies.
and Technology 30, pp 217-223. Springer-Verlag. UTL, Instituto Superior Técnico, PhD thesis.
Kotsopoulos, S.D.; Casalegno, F.; Carra, G.; Graybil, W.; Stiny, G.; Gips, J. 1971, ‘Shape Grammars and the Genera-
Hsiung, B. 2012, ‘A Visual-Performative Language of tive Specification of Painting and Sculpture’ in C V Frei-
Façade Patterns for the Connected Sustainable Home- man (ed) Proceedings of IFIP Congress 71 (Amsterdam:
Sotirios’ in SimAUD ‘12 Proceedings of the 2012 Sym- North-Holland) 1460-1465. Republished in O R Petro-
posium on Simulation for Architecture and Urban De- celli (ed), The Best Computer Papers of 1971 (Philadel-
sign. Pp.97-108 phia: Auerbach) 125-135.
Knight, T. W. 2000, Shape Grammars in education and prac- Velasco, R.; Robles, D. 2011, ‘Eco-envolventes: A parametric
tice: history and prospects. MIT, 14 Sep. 2000. Avail. design approach to generate and evaluate façade con-
<URL http://web.mit.edu/tknight/www/IJDC/> figurations for hot and humid climates’ in Respecting
Kroes, P.; Light, A.; Moore, S.A.; Vermaas, P. 2008, ‘Design in Fragile Places [29th eCAADe Conference Proceedings /
Engineering and Architecture: Towards ao Integrated ISBN 978-9-4912070-1-3], University of Ljubljana, Fac-
Philisophical Understanding’, in Vermaas, P.; Kroes, P.; ulty of Architecture (Slovenia) 21-24 September 2011,
Light, A.; Moore, S.A. (editors). Philosophy and Design. pp.539-548
From Enginering to Architecture. Springer. 1-17.

A Generative Approach towards Performance-Based Design
Using a shape grammar implementation

Tiemen Strobbe1, Ronald De Meyer2, Jan Van Campenhout
Ghent University, Belgium
http://smartlab.elis.ugent.be
1
Tiemen.Strobbe@UGent.be, 2Ronald.DeMeyer@UGent.be
Abstract. Due to a growing number of regulations and standards, building performance

becomes equally important as traditional design drivers. Therefore, it is necessary to
quickly explore design alternatives that meet these performance requirements. To support
this complex design task, a rule-based design system is proposed that is founded on
a shape grammar. This paper describes a graph-based implementation of this shape
grammar that allows subshape detection, parametric rules and attributed shapes.
The implementation described in this paper forms the basis to further investigate to
what extent rule-based design systems can support a generative approach towards
performance-based design.
Keywords. Shape grammar, evolutionary algorithm, performance-based design,
implementation, generative design.
INTRODUCTION
A growing number of regulations and standards More specifically, added programming possibilities
to which nowadays buildings have to comply, has allow the continuous generation and evaluation of
put design performance back on the architectural parametric variations in order to select (sub-)opti-
agenda. This has led to the emergence of a perfor- mal design solutions (Strobbe et al., 2012).
mance-oriented architectural design paradigm in Such CAD-systems are often founded on a
which building performance (regarding sustain- geometric representation of the design. However,
ability, safety, accessibility, comfort, etc.) becomes the current increased emphasis on building perfor-
equally important as traditional design drivers such mance in architectural design starts to question this
as functionality, history or aesthetics (Kalay, 1999). In central role of geometry in CAD. The designer has to
contrast to the traditional design process, in which work within the constraints of government rules and
performance issues are often dealt with in a post- regulations to accomplish a good compromise from
engineering optimization phase, the performance- a wide range of design solutions. Such constraints
based design process takes into account the per- are diverse in nature and often difficult to trans-
formance requirements in an early design phase. late into a graphical or geometric form. Therefore,
Recent technologic advances in computational de- parametric modeling so far allows the generation of
sign systems allow the integration of performance quite restricted geometric variations in the model.
requirements and have led to an increased produc- Furthermore, a geometric representation of design
tivity of the design process (Petersen et al., 2010). is inappropriate from a computational and automat-

Figure 1
The rule production system
consisting of the database
and the inference engine.
ic reasoning point of view, as it is computationally base and an inference engine (Figure 1). The data-
expensive to identify features to be computed from base contains all the design knowledge in form or
a set-theoretic representation of a geometric object. production rules (knowledge base), together with
More appropriate in the context of supporting information about the current state or knowledge
the design process, is the combination of a geomet- (working memory). A production rule is modeled as
ric representation and its rule-based representation. a transition between a ‘before’ and ‘after’ state. The
This allows the representation of the design through left-hand side of the production rule describes all
a sequence of design rules that preserve design in- the preconditions that need to be satisfied, while
formation that otherwise has to be reconstructed. the right-hand side describes the action of the rule
This representation is more suitable for computer execution. If the production rule’s precondition
implementation and can allow both (1) automated matches the current state of the working memory,
reasoning on the design and (2) the generation of a the production’s action is executed.
design grammar (i.e. family of designs) that extends The inference engine starts with the facts in the
restricted parametric variations. These functionali- working memory and uses the matched production
ties can support the designer in different types of rules to update, remove or add new knowledge to
problem-solving activities by generating alternative the working memory. In order to select and execute
or even sub-optimal design solutions within a gram- the production rules, the inference engine contains
matical paradigm. a pattern matcher, agenda and action executer.
Our research investigates to what extent rule- Firstly, the pattern matcher uses an algorithm to col-
based design systems can provide essential sup- lect production rules of which the preconditions are
port in the core design activities of architectural satisfied with the facts in the working memory. The
designers. This paper describes an implementation collection of rules resulting from the matching algo-
of such a rule-based design system, founded on a rithm is called the conflict set. Secondly, the agenda
graph-based shape grammar. One possible graph- determines the resolution strategy of the conflict
based design representation is described and it is set, for example according to priorities assigned to
discussed how a grammar of designs can be gener- the rules or according to the order in which the rules
ated through the application of design rules. were written. Thirdly, the action executer performs
the production rule’s action and removes it from the
SHAPE GRAMMAR IMPLEMENTATION agenda. This process is continuously iterated, result-
ing in different production system derivations.
Rule production system
A rule-based design system founds on the collec- Shape grammar
tion and management of design ‘knowledge’ in form The use of production systems to provide some
of rules, as is the case in rule production systems. A form of artificial intelligence is found useful in sever-
rule production system generally consists of a data- al scientific domains. In the domain of architectural

design, shape grammars (Stiny et al., 1972), a class an example of the former approach, Grasl (2012) de-
of production systems that generates geometric scribes a graph representation of a Palladian-style
shapes, are used to capture design knowledge into villa that consists of an adjacency graph for the grid
shape rules. Shape grammars define a rule-based underlying the design. While this approach is mainly
formalism to represent visual thinking and to handle useful to implement specific graph representations,
ambiguities that are characteristic of architectural the latter approach is more suitable for general im-
design. A key concept in the shape grammar formal- plementations. For example, Heisserman (1994)
ism is the embedding relation that lies at the basis introduces a boundary graph to represent three-
of handling ambiguities (Stiny, 2006). Embedding dimensional solids. The graph nodes represent
allows the recognition of emergent (sub)shapes topological elements (such as vertex, edge, face,
that are not predefined in the shape grammar, but etc.) and the arcs represent the adjacencies between
emerge from shapes generated by rule applications. the elements. Geometric information (coordinates)
Once a rule is applied, all the shape parts fuse and is associated with each vertex node as an attribute.
the new shape needs to be ‘reframed’ in order to The boundary graph is based on the generalized
proceed with the rule application. Therefore, em- split-edge data structure that is capable to store
bedding allows an open-ended way to generate adjacency information of geometric objects. Heis-
multiple design derivations and to explore a design serman’s approach has been further investigated in
solution space towards novel or unexpected areas. several research efforts, for example the GRAMATICA
One of the current challenges is the computer three-dimensional interpreter (Correia et al., 2012)
implementation of shape grammars. Gips (1999) or the web-based implementation GRAPE (Grasl et
provides an overview of previous research efforts on al., 2011).
shape grammar implementations and describes sev- The main advantage of the graph-based im-
eral issues and obstacles, of which most of them are plementation of shape grammars is that both
still valid to date. These obstacles include the lack parametric shape rules and subshape detection
of parametric shape support and the subshape de- are supported (Grasl et al., 2011). Firstly, the ability
tection problem. The support of parametric shapes to support parametric shape rules is caused by the
involves parameterized shape rules. The subshape topological nature of graphs. A graph can represent
detection problem is to determine whether a shape multiple geometric shape variations. For example,
contains a specific subshape. The main challenge of the graph representation of all quadrilateral shapes
the computer implementation is thus to find an un- (square, rectangle, trapezoid, etc.) are isomorphic or
derlying representation that is both able to support topologically identical. Secondly, the ability to sup-
the visual nature of shape grammars and amenable port subshape detection is based on the subgraph
to computer interpretation. In the remainder of this isomorphism problem (Ullmann, 1976). Subgraph
paper, an approach is described that uses a graph- isomorphism involves determining whether a graph
based representation of the design (facts) and de- contains a subgraph that is isomorphic to a given
sign rules, together with a graph transformation sys- graph. While the subgraph isomorphism problem is
tem (inference engine). proven to be NP-complete, Grasl et al. (2011) dem-
onstrate that practical solutions can nevertheless be
GRAPH-BASED REPRESENTATION created. Thirdly, attributed graphs also contain non-
The graph-based representation of a design consists topological information that is described in the form
of pairwise relations between objects. In the domain of attributes.
of architectural design, graphs are mainly used to
model topological configurations of architectural Approach
spaces or to describe geometric data structures. As Based on Heisserman’s approach (1994), the attrib-

Node Type Attributes Type Table 1
Vertex ID String Node types and their at-
x, y, z Float tributes.
Edge ID String
length Float
Face ID String
uted graph represents (geometric) objects as nodes, can be applied in different environments using an
and relations between these objects as directed inference engine. Graph rules are described as a
arcs. Several node types are distinguished that rep- transition between two graphs, following a typical
resent different objects (vertices, edges and faces) IF-THEN statement. The left-hand side of the graph
(Table 1). This approach can be extended to an un- rule describes the graph that needs to be matched
limited number of nodes that represent different to the host graph, together with additional condi-
objects (for example: “Solid”, “Wall”, “Door”, “Window”, tional statements. These conditions include attrib-
etc.). Similarly, several arc types are distinguished ute conditions and negative application conditions
that represent different relations between the geo- (NAC). Attribute conditions define restrictions on
metric objects, for example: “hasVertex”, “hasEdge”, the attributes of the graph, while NACs specify re-
etc. In addition, several attributes are associated quirements for non-existence of sub matches. The
with the graph objects in order to store non-topo- right-hand side describes the transformation of the
logical information: unique ID’s are associated with host graph. This graph transformation includes de-
all nodes, coordinate geometry is associated with leting or manipulating existing graph objects, creat-
vertex nodes, etc. ing new objects, and also performing computations
The graph representation contains only topo- on the object attributes.
logical information of the shape, which allows the As an example, the graph-based representation
support of parametric shapes. Additional attribute of a shape rule that generates a Koch curve is dis-
information (e.g. coordinate geometry) is needed played in Figure 3. The Koch curve is a mathemati-
to restrict the parametric shapes to specific geo- cal fractal curve that is constructed by recursively
metric shapes. Therefore, an attributed graph-based replacing a line segment with an equilateral triangle.
representation ensures a unique mapping between Therefore, the left-hand side of the rule represents
the shape and the graph. An example of the graph a line segment and the right-hand side represents
representation of a geometric line object is given a graph with both modified and new graph ob-
in Figure 2. The graph consists of two vertex nodes
(white), one edge node (black), and two directed Figure 2
arcs from the edge node to the vertex nodes. The Graph representation of a
line is considered parametric, if the coordinate ge- non-parametric line object.
ometry attributes contain parametric values.
GRAPH RULES
Parallel to the graph-based representation of the
design, the rule set can also be described using the
same representation. This graph-based rule repre-
sentation enables the collection and management
of design knowledge with a far greater expressive
power than pure data. Once a rule is described, it

Figure 3
Graph-based representation
of a shape rule that generates
a Koch curve.
jects. The attributes of the new nodes are defined tection of subshapes, which is an important chal-
as expressions that take into account the attribute lenge for shape grammar implementations. Also,
context of the matched host graph (X1, Y1, X2, Y2). the matching conditions are evaluated in order to
Furthermore, an additional attribute condition is recognize specific shape features. The preservation
implied in order to ensure a minimum length of the morphism describes the mappings of the objects of
generated line segments. one graph to those of another, using tags to indicate
the mapped objects. If multiple matches are found,
GRAPH TRANSFORMATION SYSTEM all possible morphisms are calculated and stored.
A graph transformation system is used to allow the The selection of the morphism can happen non-de-
stepwise application of graph rules on the original terministically or using a user-defined sequence (for
host graph. Among others, AGG is a general graph example through priorities assigned with the rules).
transformation system written in JAVA (Rudolf and The user can go back and forth in this transforma-
Taentzer, 1998). Rule application is performed by tion process, and generate multiple alternatives.
matching the left-hand side of a graph rule to the In the following example, the initial line shape
host graph and replacing it using a preservation that is described in Figure 2 and the Koch rule that
morphism. The AGG system solves the problem of is described in Figure 3 are used to generate shape
graph matching, i.e. the subgraph isomorphism grammar transformations. Figure 4 shows several
problem, as a constraint satisfaction problem (CSP). graph transformation steps in the generation pro-
As indicated previously, this feature enables the de- cess (step 0, 1, 2 and 10). Rule matching and selec-

Figure 4
Graph-based (left) and shape
(right) representation of
several graph transformation
steps (0, 1, 2, 10).
tion is performed non-deterministically, in order to approach. In the case of performance-based design,

allow a depth-first design exploration process. further research is needed to determine which do-
main knowledge can be incorporated in the rules
CONCLUSION AND FUTURE RESEARCH and whether heuristics are needed to go back and
This paper describes the implementation of a rule- forth in the transformation process. Therefore, the
based design system using a graph-based repre- work demonstrated in this paper forms the basis to
sentation of a shape grammar. An approach is pro- further investigate to what extent rule-based design
posed that allows subshape detection, parametric systems can provide support in the design process.
shape rules and attributed graph objects. Therefore,
it is possible to quickly explore design alternatives REFERENCES
that extend parametric variations. A simple case Kalay, Y 1999, ‘Performance-based design’, Automation in
study has demonstrated the feasibility of this ap- Construction, 8(4), pp. 395-409.
proach, however, a more complex design context Petersen, S and Svendsen S 2010, ‘Method and simulation
is needed in order to test the practical value of this program informed decisions in the early stages of

building design’, Energy and Buildings, 42(7), pp. 1113- Planning B, 39(1), pp. 83-95.
1119. Heisserman, J 1994, ‘Generative geometric design’, IEEE
Strobbe, T, Pauwels, P, Verstraeten, R, De Meyer, R, Van Computer Graphics and Applications, pp. 37-45.
Campenhout, J 2012, ‘Optimization in compliance Correia, R, Duarte, J, Leitao, A 2012, ‘GRAMATICA: A general
checking using heuristics: Flemish Energy Performance 3D shape grammar interpreter targeting the mass cus-
Regulations (EPR)’, eWork and eBusiness in Architecture, tomization of housing’, Proceedings of the 30th eCAADe
Engineering and Construction, pp. 477-482. Conference, pp. 489-496.
Stiny, G, Gips, J 1972, ‘Shape grammars and the generative Grasl, T and Economou, A 2011, ‘GRAPE: A parametric shape
specification of painting and sculpture’, Information grammar implementation’, SimAUD 2011 Conference
Processing, 71, pp. 1460-1465. Proceedings, pp. 45-52.
Stiny, G 2006, Shape: talking about seeing and doing, MIT Ullmann, J 1976, ‘An algorithm for subgraph isomorphism’,
Press, Cambridge. Journal of the ACM, 23(1), pp. 31-42.
Gips, J 1999, ‘Computer implementation of shape gram- Rudolf, M and Taentzer, G 1999, ‘The AGG approach: lan-
mars’, Workshop on Shape Computation. guage and environment’, Handbook of graph grammars
Grasl, T 2012, ‘Transformational Palladians’, Environment and and computing by graph transformation, pp. 551-603.

Tableware Shape Grammar
Towards mass customization of ceramic tableware

Eduardo Castro e Costa1, José Pinto Duarte2
CIAUD, Faculdade de Arquitetura, Universidade Técnica de Lisboa,
Rua Sá Nogueira, Pólo Universitário, Alto da Ajuda, 1349-055 Lisbon, Portugal
1
http://home.fa.utl.pt/~castroecosta/, 2http://home.fa.utl.pt/~jduarte/
1
castroecosta@fa.utl.pt, 2jduarte@fa.utl.pt
Abstract. Mass customization is pointed as a means of improving a company’s

competitiveness, which is an essential trait in Europe’s economic situation of today.
This paper reports a mockup implementation of the mass customization paradigm to
ceramic tableware design, through the use of shape grammars, parametric modelling
and rapid prototyping. Focus is emphasized on the initial development of a parametric
shape grammar as a design system, operating on curved surfaces and solids. Mapping
operations are suggested for dealing with the formal complexity of these shapes.
This initial academic experiment poses as a first step into the development of a mass
customization system that is expected to meet industry standards.
Keywords. Mass customization; ceramic tableware; generative design; shape grammars;
rapid prototyping.
INTRODUCTION
Research is currently being developed towards the design system and articulates it with the production
application of the mass customization paradigm to system.
the design and production of ceramic tableware. According to Pine (1993), mass customization
The experience documented in this paper repre- can improve a company’s competitiveness, allowing
sents a first mockup of the framework necessary for it to offer differentiated products to its customers.
the implementation of such paradigm. The ultimate This research is thus intended to be applied to an in-
objective is that end users can create their own, dustrial context, through collaboration with a local
highly customized, tableware set. ceramics company.
According to Duarte (2008), the implementa-
tion of a mass customization system implies devel- Methodology
opment on three fronts: a design system that en- A small scale implementation of the mass customi-
capsulates the stylistic rules of tableware elements, zation paradigm was tested as an exercise, in a se-
generating the corresponding digital models; a pro- mester long course about shape grammars. In this
duction system that allows to automatically materi- exercise, the shape grammar apparatus, invented by
alize those models into usable tableware elements; Stiny and Gips (1972), was used as a design system,
and a computational system that implements the encoding the rules for the shape generation of ta-

bleware elements into a parametric shape grammar. (2011), who have developed and implemented a
This paper will focus mainly on this design system. curved shape grammar for generating Celtic knots,
Subsequently, parametric models correspond- using quadratic Bézier curves. Their research fo-
ing to derivations of the shape grammar were im- cuses mainly on the problem of shape recognition
plemented using the visual programming interface and emergence, which is a typical research problem
Grasshopper. The implemented system allows the when implementing a shape grammar into a com-
user to combine shape rules and manipulate its pa- puter program.
rameters, generating derivations of the tableware Although an implementation was developed in
elements and the corresponding digital models. this exercise in order to test formal aspects of the
Finally, these digital models could then be auto- shape grammar rules, it is not the grammar itself
matically produced using powder-based 3D printing that has been implemented, but rather parametric
equipment. models that correspond to specific derivations of
the grammar. Different derivations can be gener-
Application of shape grammars to three- ated, but it is the user who must directly manipulate
dimensional curved shapes the rules, selecting the ones to be applied. There-
Although the original shape grammar has been fore, issues like shape recognition and emergence,
developed for straight line shapes (Stiny, 1980), which are typical of shape grammar implementa-
some authors have used this tool to work with tion, will not be addressed for now.
curved shapes, which are predominant in tableware So, although some work has been developed on
design. shape grammars of curved shapes, these shapes are
Knight (1980) developed a shape grammar on mostly linear. Research on shape grammars applied
Hepplewhite-style chair-back designs. While deriv- to surfaces is still sparse. We hope that this research
ing the chair-back shape mostly in terms of rectilin- can contribute to fill in this shortcoming.
ear elements, in the final designs they are replaced
by the typical curved shapes in a descriptive proce- DEVELOPMENT OF THE SHAPE GRAMMAR
dural fashion. One of the objectives of this research is to develop a
McCormack and Cagan developed work on shape grammar that is able to encode the styles, re-
branding of automobiles, namely a shape gram- garding both shape and decoration, of different col-
mar for generating the front-end view of a Buick lections of tableware sets. However, for this first ex-
(McCormack et al., 2004). In this two-dimensional ercise, the goal was set to develop a shape grammar
shape grammar, curved lines are used extensively. that would encapsulate the style of one collection
The curved lines are controlled by parametrically (Figure 1). The shape grammar should automatically
positioned control points. Although it is not explic- generate the different elements of the selected col-
it, these curves seem to be Bézier curves. A similar lection, taking into account that it would be subse-
strategy will be used in the definition of the profile quently extended to other collections.
curve of the base shape of the tableware elements.
Around the same time a shape grammar is Element types
developed that features curved shapes, using both Even before focusing on any specific collection, a
straight lines and circular arcs to generate the bot- reflection was made on the different types of table-
tles of Coca-Cola and Head & Shoulders (Chau et ware elements. A first observation on the dimen-
al., 2004). Despite mentioning research towards the sions of these elements led to establishing a distinc-
generalization for NURBS curves, it is not present in tion between deep and shallow types. The shallow
this research. type category includes the charger plate, the dinner
This issue was addressed by Jowers and Earl plate and the soup plate, while the deep type cat-

Figure 1
Shape grammar corpus.
egory includes the bowl, the mug and the cup. tableware elements. Let’s take the example of the
Types in the same category were considered soup plate. Three functions can be identified in its
formally similar among each other, differing only shape: the broad border is used to hold the plate;
in terms of dimensions. In the deep type category, the soup needs to be contained, and so the dish
if we disregard the handle in the mug and the cup, must have a deeper part for this purpose; and finally,
these types are formally similar to the bowl. it needs a broad and flat bottom, so it lays steady on
Different types are thus characterized by differ- the table. So the three functions are: holding, con-
ent sizes, measured both in height and radius. Table taining and laying on the table.
1 illustrates the dimensional relations between the The functional parts were analyzed in terms of
six chosen types within the selected collection, as dimensions, namely height and radius. The relations
well as the mentioned distinction between deep between dimensions of the different functional
and shallow types: in shallow types, radius is larger parts are described by a functional configuration.
than height, and so the ratio between these dimen- The functional configuration for the soup plate, sys-
sions is higher than 1, whereas in deep types, it is tematizing the example given above, is shown in
lower than 1. Table 2.
The shape similarity among types within the Different types feature different functional con-
same category is interpreted as a parametric varia- figurations. For example, to be able to contain liq-
tion of the same entity, justifying the development uids, the soup plate and the deep types feature a
of a parametric shape grammar. taller containing part than the dinner plate.
Generally, all three functions are present in each
Functional parts type. However, in some types they are assigned to
This first observation has also brought attention parts other than the main body of the tableware ele-
to the distinction between functional parts within ment. For example, in the mug or the cup, the hold-
Table 1
Classification and dimensions
of the different types in one
collection
Type Charger Dinner Soup Bowl Mug Cup
Radius (cm) 15,50 13,50 11,50 7,00 4,50 3,75
Height (cm) 3,00 2,50 4,00 8,50 10,00 5,00
Radius/Height 5,16 5,40 2,88 0,82 0,45 0,75
Category Shallow types (plates) Deep types

Function to lay on (ps*) to contain (ct*) to hold (pg*) Table 2
Functional configuration of
the soup plate (* from the
portuguese “pousar”, “conter”
e “pegar”, respectively).
radius 46% 20% 35%
height 14% 58% 29%
ing function is assigned to the handle. In its current element, as well as the upward direction (h, from
state, this shape grammar is only encoding shape for height) and the outward direction (w, from width).
the main body of the elements. Parts like the han- The general dimensions of the tableware element
dles in the cups and mugs will be addressed in the are introduced as input parameters of rule 1, which
future. creates an object in which the element is inscribed.
Further development of the shape grammar This object is called the envelope (Figure 3, en). In
will focus its extension to other collections. It is ex- the two-dimensional view of the profile, the enve-
pected that for the same types within other collec- lope is represented by a rectangle, or more generally
tions, functional configurations, despite featuring speaking, a quadrilateral - since the rectangle will be
some dimensional variation, are somewhat similar, subsequently distorted, the general term quadrilat-
and therefore characteristic of the type. Variations eral, or quad, is more appropriate.
on this configuration are to be registered as other The initial envelope is subsequently subdivided
collections are analyzed, and will be properly inte- into the element’s functional parts by rule 2, which
grated into the design system. is parameterized according to the correspondent
functional configuration (Figure 3). Rule 2a subdi-
Initial shape and first derivation steps vides the envelope into three functional parts, and
This first analysis is incorporated into the first two can be applied to the shallow types. For the deep
rules of the shape grammar, which are applied in the types, rule 2b should be applied, which disregards
beginning of each derivation of a tableware element the holding part (pg).
(Figure 2). Derivation of the tableware elements can be
Derivation begins with the initial shape, which is split into three phases: initialization - which applies
a referential determining the center of the tableware rules 1 and 2 as seen above -, base shape definition
Figure 2
First steps of derivation for the
soup plate: envelope creation
and functional partitioning.
Figure 3
Shape grammar rules for
envelope creation and func-
tional partitioning.

Figure 4
Overview of the soup plate
derivation.
Figure 5
Resumed derivation for base
shape of soup plate.
Figure 6
Subdivision rule and example
of application: subdivision al-
lows for more complex shapes.
and decoration - which applies subsequent rules scribed in them.

(Figure 4). To each phase corresponds a bi-dimen- Therefore, in the base shape definition phase,
sional view, which best illustrates the three-dimen- three types of rules can be applied: subdivision, dis-
sional geometry of tableware elements, as well as tortion and substitution.
the operations applied to them. Subdivision rules in the base shape definition
phase (rule 3) are similar to the functional subdivi-
Base shape definition sion rules seen previously, the difference being that
Since the selected collection features circular ele- the resulting envelopes keep the type of the preced-
ments, their base shape can be described as a solid ing envelope. These rules allow for more complex
of revolution. Therefore, rules and derivation steps shapes, through subsequent combination of more
regarding base shape generation are represented than one Bézier curve (Figure 6).
by a bi-dimensional profile. For this first approach, Distortion rules allow to manipulate the quads
the profile thickness is not being considered, and that correspond to the envelopes, and therefore ma-
therefore the resulting shape is more accurately de- nipulate the control points of the Bézier curves that
scribed as a surface, rather than as a solid. will substitute them. Distortion rules can be applied
The base shape is defined by formal manipula- recursively, allowing to generate additional distor-
tion of the functional parts that result from the ini- tions not coded into rules through a compound
tialization phase. In the end, the base shape profile effect. Figure 7 shows how to obtain this through
is the product of a combination of quadratic Bézier applying rule 4 recursively and with different trans-
curves, controlled by their corresponding envelopes formations - which are shown under the rule appli-
(Figure 5). The quadrilateral envelopes are used as cation arrow. Parameters are used both for the dis-
auxiliary shapes for defining the Bézier curves in- tortion effect as for the shape matching, and were
Figure 7
Distortion rule and example of
recursive application.

omitted for clarity. Figure 8
Substitution rules replace the envelopes by Replacing envelopes for
their corresponding Bézier curves. For each curve, corresponding Bézier curve:
three of the vertices of the corresponding quad cor- substitution rules and exam-
respond to the curve’s control points, whereas the ples of its application.
directions of the curve’s start and end tangents are
defined by the quad’s edges (Figure 8). Figure 9
The rules are allowed to be applied in different Different sequence order
order. For example, we can subdivide an already generates different results.
distorted envelope. This is possible because of the
parametric nature of the grammar, which allows for
perspective transformations (Stiny, 1980). However, three-dimensional surfaces, representation of rules
the result of applying the rules in a different order is and derivations in two dimensions is not straight-
not necessarily the same (Figure 9). In this way, some forward - in fact, neither is imagining them. For this
flexibility is added to the grammar. purpose, we make use of the parametric representa-
Figure 10 shows all the derivation steps for de- tion of surfaces. In this representation, the Cartesian
fining the base shape of the soup plate, including coordinates of a surface point depend on two differ-
the initialization phase. The final step corresponds ent parameters u and v, allowing for a continuous
to the revolution of the profile into a set of surfaces, mapping of a two-dimensional region into space
the actual base shape upon which decoration rules (Pottmann et al., 2007).
will be subsequently applied. A typical example of a parametric representa-
tion is a world map. The map’s horizontal lines repre-
Decoration sent coordinates on the surface of the Earth with the
Decoration is achieved through the application of same latitude, while the vertical lines correspond to
shape grammar rules to the resulting base shape points with the same longitude (Figure 11). There-
surface set. fore, latitude and longitude can be considered the
Rules for defining the base shape of the table- uv parameters of the Earth’s surface. A point with
ware element operated on its profile, which is a given parameters in the map can be easily and accu-
two-dimensional representation. In the decoration rately identified using a GPS device. Similarly, lines
phase, however, and because we are dealing with and shapes can be mapped from the map into the
Figure 10
Derivation for base shape of
the soup plate.

Figure 11 tions - transformations that map straight lines into
World map as parametric curves (Pottmann et al., 2007) - , which are not usu-
representation of the surface ally addressed in the shape grammar formalism.
of the Earth (world map from In fact, Stiny (1980) states that “a shape rule a → b
www.free-world-maps.com). applies to a labeled shape c when there is a trans-
formation t such that t(a) is a subshape of c” (p.
347), limiting these transformations to translation,
rotation, scale and reflection, which he refers to as
globe and vice-versa. Euclidean transformations (p. 344). However, the au-
These mapping operations are used in the deco- thor points out that for transformations other than
ration phase. In this phase’s derivation steps, the Euclidean, such as affine or perspective transforma-
tableware elements are represented in two-dimen- tions, a parametric grammar should be used (p. 351).
sional space as top views. In the case of the selected For the development of this grammar, we are
collection, these elements correspond to circular assuming that, similarly to affine and perspective
objects, and inherently their parts feature circu- transformations, non-linear transformations can be
lar arcs. In rule description however, such as in the also be used in parametric grammars. This hypoth-
base shape definition rules, surfaces are represented esis should be later proven in order to validate this
by squares, a generic shape which evokes the two- grammar.
dimensional nature of the uv parametric representa- In the selected collection, decoration is based
tion (Figure 12). on a) subdivision of surfaces, and b) application of
Therefore, rules are mapped into the design in a relief- and contour-based motif onto the resulting
the same way a shape in a map is mapped back and subsurfaces (Figure 13).
forth into the surface of the Earth. These mapping Subdivision rules are similar to the subdivision
operations imply the use of non-linear transforma- operations mentioned previously, except they op-
Figure 12
Example of UV mapping: rules
are mapped onto surfaces.
Figure 13
Derivation for decoration of
soup plate.

Figure 14
Decoration subdivision rules
and application in the soup
plate derivation.
erate on the surface’s parametric, or UV, space. Rule are to be replaced by more elaborate ones featur-
10a subdivides the surface into two parts with the ing relief-based motifs. For the selected collection,
same U parameter differential (Figure 14). In the der- two rules are defined, 11a and 11b, which are to be
ivation for the selected collection, rule 10a is used applied exclusively onto shallow and deep types
recursively to subdivide the plate into eight para- respectively. Both rules apply a slight depression to
metrically equal parts. the target labelled subsurface, a motif which is typi-
Rule 10b subdivides it into three parts, also cal for the selected collection. However, contrary to
along U, but in this case the first and third part have rule 11b, rule 11a also changes the subsurface’s con-
the same U parameter differential, which is different tour (Figure 15).
for the second independent part. The parametric re- Recursive application of motif replacement rules
lations among the parts are variable. In the selected to all labelled subsurfaces is the final stage of the
collection, the second part is larger (Figure 14). How- derivation, resulting in a design that belongs to the
ever, in the corresponding rule this constraint is not collection’s language (Figure 15).
set, so to allow for a wider range of variation.
Rule 10b uses a label to mark the subsurfaces to APPLICATIONS OF THE SHAPE GRAMMAR
which subsequent decoration rules can be applied.
Since the use of labels is still under development Parametric modelling
and lacking consistency, it has not been addressed A three-dimensional parametric grammar is difficult
in this paper. to test without some kind of implementation. On
Similarly to the base shape definition phase, the one hand, the combination of several param-
surfaces resulting from the subdivision operations eters corresponds to a large number of solutions.
Figure 15
Motif replacement rules and
application in the soup plate
derivation.

Figure 16
Parametric model developed
in Grasshopper: using rules as
groups of components.
On the other hand, some three-dimensional geo- two models were generated using different param-
metric operations are difficult to represent in two- eter configurations. Therefore, a total of four digital
dimensional drawing, such as the surface mapping models were generated in Grasshopper (Table 3).
operations. These models were to be later produced through
Therefore a computational model was deve- rapid prototyping.
loped in Grasshopper, a visual programming in-
terface that interacts with geometrical modelling 3D printing
software Rhinoceros, and allows implementing, and The digital models generated in the Grasshop-
therefore evaluating, parametric models. It should per program were materialized through available
be noted that the Grasshopper model is not con- 3D Printing technology. For saving purposes, four
sidered an implementation of the shape grammar. quarters of dishes were produced, instead of four
However, if we consider that the result of the deriva- complete dishes (Figure 17). This was also useful to
tion of a parametric shape grammar is a parametric evaluate the results in terms of their section.
model, than we can argue that we are implementing Prototyping the models provided for a general
a derivation. Actually, the Grasshopper model was first impression about the models being generated,
developed so that rules can be identified as groups namely in terms of scale and weight. The 3D print-
of components, in a modular fashion (Figure 16). ed models are especially useful for communication
With this tool, two derivations of the soup plate purposes, allowing to better illustrate the project
were implemented as parametric models. The two design, either to faculty members as well as to po-
derivations differ slightly, having different rules ap- tential partners in the industry.
plied in the base shape. Then, for each derivation, Further research will aim at determining if these
Table 3
Digital models of derivations
of the shape grammar.
Derivation 1 Derivation 1 Derivation 2 Derivation 2

Parameters 1A Parameters 1B Parameters 2A Parameters 2B

3D printed models are suitable for mold making, Figure 17
and if it is possible to use 3D printing for digital fab- 3D printed prototypes.
rication, i.e., for producing final products.
CONCLUSIONS
This exercise brought the attention to the many
questions that should be answered in order for the
mass customization paradigm to be applied to ce- financial support for the present publication and
ramic tableware. Focusing particularly on the de- corresponding communication.
velopment of the shape grammar, the manipulation
of three-dimensional and predominantly curved REFERENCES
shapes poses as the main challenge, which needs to Chau, H.H., Chen, X., McKay, A., Pennington, A., 2004. Evalu-
be mastered in order to serve as an effective design ation of a 3D Shape Grammar Implementation, in:
system. Gero, J.S. (Ed.), Design Computing and Cognition’04.
However, the success of this first mockup poses Kluwer Academic Publishers, Dordrecht, The Nether-
as a good indicator for further research, which is lands, pp. 357–376.
planned to develop along the three systems. Duarte, J.P., 2008. Synthesis Lesson - Mass Customization:
Models and Algorithms - Aggregation Exams (Agre-
Future developments gação). Faculdade de Arquitectura, Universidade Téc-
Some aspects of the Shape Grammar formalism nica de Lisboa, Lisboa.
are to be further addressed, stabilizing the design Jowers, I., Earl, C., 2011. Implementation of curved shape
system so it can be extended to other collections grammars. Environment and Planning B: Planning and
and element types, namely the validation of map- Design 38, 616 – 635.
ping operations, as well as control mechanisms Knight, T.W., 1980. The generation of Hepplewhite-style
such as labels and parameter intervals. Concerning chair-back designs. Environment and Planning B: Plan-
implementation, although Grasshopper was used ning and Design 7, 227 – 238.
in this first mockup, a study must be conducted on McCormack, J.P., Cagan, J., Vogel, C.M., 2004. Speaking the
the available programming technologies for imple- Buick language: capturing, understanding, and explor-
menting the design system. Concerning the pro- ing brand identity with shape grammars. Design Stud-
duction system, a thorough research on production ies 25, 1–29.
techniques is to be developed, both for handcrafted Pine, B.J., 1993. Mass Customization: The New Frontier in
and industrial ceramics manufacturing. Last but not Business Competition. Harvard Business Press.
least, establishing a partnership with a manufactur- Pottmann, H., Asperl, A., Hofer, M., Kilian, A., 2007. Architec-
er is a key factor for the success of this research. tural Geometry, 1st ed. Bentley Institute Press.
Stiny, G., 1980. Introduction to shape and shape grammars.
ACKNOWLEDGEMENTS Environment and Planning B: Planning and Design 7,
Eduardo Castro e Costa is funded by FCT (Fundação 343 – 351.
para a Ciência e a Tecnologia) with PhD grant SFRH/ Stiny, G., Gips, J., 1972. Shape Grammars and the Generative
BD/88040/2012, and by CIAUD (Centro de Investi- Specification of Painting and Sculpture, in: Freiman,
gação em Arquitectura Urbanismo e Design) with C.V. (Ed.), Information Processing 71. North Holland,
Amsterdam, pp. 1460–1465.

Grandstand Grammar and its Computer Implementation
A computational approach to facilitate decision making and encourage

efficiency in the design of sports facilities
Yimin Sun1, Lu Xiong2, Ping Su3

State Key Laboratory of Subtropical Building Science, South China University of Technol-
ogy
1
arymsun@scut.edu.cn, 2bbxlarc@gmail.com, 3suping@scut.edu.cn
Abstract. In sports facilities, a grandstand is the structure which provides good sight
quality and safety evacuation conditions for the spectators. Grandstand plays important
functional and formative roles in sports facilities, and especially in large scale stadia.
This paper argues the notion of shape grammar and its computer implementation will
solve the difficulties in grandstand design. The authors identify the specific difficulties of
grandstand design, then set the aims of the grammatical computer tool. Afterwards the
shape grammar of grandstand design is formulated, and a computer tool is developed
based on the grammar. At last, the paper discusses the application and usage of the
grammar and the computer tool both in early design phase and design development phase
with a design practice case study of a large scale stadium.
Keywords. Grandstand design; shape grammar; parametric modelling.
INTRODUCTION
In sports facilities, a grandstand is the structure Therefore the in the early phase of the design prac-
which provides good sight quality and safety evacu- tice, architects are likely to use existing grandstand
ation conditions for the spectators. Grandstand design with similar condition rather than design a
plays important functional and formative roles in new grandstand for the project. In the design devel-
sports facilities, and especially in large scale stadia. opment phase, modification of grandstand design
Apart from the function and form of the grand- will result in the large amount of remaking of docu-
stand, the designs of other parts of stadium such as mentation. Furthermore, the modification process
the facade surface and the roof are closely related of the other parts of the building would be delayed
to the grandstand, and most of the interior rooms by the grandstand. Three problems are identified in
are placed under the grandstand. In the very early the traditional grand stand approach. How to pro-
design phase of a large scale stadium, the design vide a highly customized grandstand model in early
of the grandstand must be considered to accom- design phase? How to provide rapid response to the
modate the spectators and the other basic need of modifications in the design developments phase?
the building. Traditionally, the process of a grand- How to rapidly negotiate the relationship between
stand design trends to be complicated and tedious. grandstand and the other parts of the building?

This paper argues the notion of shape grammar first row profile, the plan drawing is a tedious
and its computer implementation will give a solu- process. The designer spend most of the time
tion to the above three problems. Similar studies drafting the offset row profiles. If the first row
show the parametric design approaches to facilitate profile is modified, the whole drawing will have
the grandstand design (Hudson, 2010; Miller, 2009). to be remade.
However, the studies trend to focus on the grand- 3. The capacity calculation in the early design
stand as an isolated part. The relationship between phase. The capacity is related to multiple fac-
the grandstand and the other parts of the stadium tors such as the area of the grandstand, the
is not explored. The design rules of grandstand is are number of aisles, the number of vomitories,
not presented in formal ways. Grandstand design row distance and the seat width. The precise
follows strict and complicated function rules and capacity can only be obtained at the very late
patterns. The shape grammar approach provides a phase of grandstand design. If the row profiles
computational and logic device for recording design is a curve, the distribution of seats will be a lot
rules and patterns. Computer tools can be made more difficult than the linear row profile.
base on a shape grammar to solve specific design 4. The generation of the 3D profile of the grand-
problems. stand boundary. For the case that the bound-
In the following part, the authors identify the ary is not parallel with the row profile, the
specific difficulties of grandstand design, then set profile of the boundary will be a 3D curve and
the aims of the grammatical computer tool. After- define the skyline of the grandstand. It plays
wards the shape grammar of grandstand design is an important role in the façade of the build-
formulated, and a computer tool is developed based ing and acts as a key reference of the roof. The
on the grammar. At last, the paper discusses the ap- curve can only be generated in the late phase
plication and usage of the grammar and the com- of grandstand design.
puter tool both in early design phase and design de-
velopment phase with a design practice case study The aims of the grammatical grandstand
of a large scale stadium. design tool
The digital model should be served a reusable tools
THE DIFFICULTIES IN GRANDSTAND to assist the designs of varied grandstands for differ-
DESIGN AND THE AIMS OF THE GRAM- ent projects. After the identification of the difficul-
MATICAL COMPUTER TOOL ties in grandstand design, the tool should achieve
the following aims:
The difficulties in grandstand design 1. Ease for use. In this case the ease for use con-
The design of grandstand requires complicated pro- tains two aspects: the easy acquisition of de-
fessional knowledge. The difficulties in grandstand sign knowledge and the friendly user interface.
design can be identified in the following aspects: The designers who are not familiar with grand-
1. Section design. The raise of each row should stand design could quickly gain the accord-
be precisely calculated to guarantee the sight ing knowledge in a systematic way. The user-
quality of the spectators. The amount of raise friendly interface could promote the designers
will be affected by the type of game, the first to use the tool and focus on the grandstand
row profile, the elevation of the first row, C val- design regardless of their knowledge of com-
ue and the row distance. The calculation could puter programming and 3D modeling.
be time consuming and tedious if was done 2. Real-time visual feedback. The 3D model can
manually. be updated synchronously with the design
2. Plan drawing. After the configuration of the conditions and parameters.

3. Real-time performance feedback. The perfor- space beneath the grandstand and the posi-
mance indicators such as capacity, elevation tion of columns and beams, the user can also
of each level and the sight quality of each seat manually input the aisle axis. The position of
can be updated with the 3D model. vomitory will change according to the aisle axis.
4. Enhance design efficiency. Majority of the man- 3. Raise calculation. The raise of each row away
ual works can be overtaken by the computer from its low row is calculated in the task. The
tool. input shape are the first row profile and the
5. Enhance design quality. More energy could be focus point. The input parameters are number
put in the generation of alternatives, exploring of rows, row distance and C value. Calculation
of design space, refinement of design decision is made according to equation (1). In the equa-
making. The 3D model would help reduce de- tion, Y is the elevation from the focus point to
sign mistakes that are difficult to be reflected eye point, K is the number of rows, C is C value
in 2D drawings. (John and Sheard, 2000). The output are the el-
evation of each row and the sightlines.
THE FORMULATION OF THE GRAND- Yn = [(Yn-1+ (Kn-1) ×C] × Xn / (Xn-1) (1)
STAND GRAMMAR 4. Generation of the grandstand 3D model. The
task elevates the shapes on the construction
Subdivision of tasks plane to their designed height. Then the solid
Considering the complicity of the grandstand de- model are generated from the shapes. For the
sign, the design task is divided into the following consideration of quick feedback and the time
sub tasks: saving from the solid computation, the user
1. Generation of the plans of rows. The task con- can choose only to elevate the curves rather
tains the generation of the plan profile of each than generate the solid models. The input
row, the position of the vomitory, seat distribu- shapes are all the row profiles and seat guide
tion guide line and the estimated capacity of lines. The input parameters are the elevations
the grandstand. The input shapes are the first of the shapes. The output are the elevated
row profile, boundary of the grandstand, focus shapes and solids.
point and the aisle axis. The input parameters 5. Seat distribution. The task inserts seats on the
are row distance, number of rows, aisle width, seat guide lines and calculate the precise ca-
evacuation method, vomitory width, vomitory pacity of the grandstand. The input shapes are
start level, vomitory end level, seat width and the seat guide lines, seat rectangle and the 3D
seat offset. The outputs are the plan shapes of seat model. The input parameters are the di-
rows, seat guide lines and estimated capacity. mensions of the seat. The output are the insert-
2. Aisle generation. The input shapes are the first ed seat shapes and the precise capacity.
row profile and the focus point. Input param- 6. Sight quality analysis. The task analyses the
eters are seat width, maxim seat number in a sight quality of each seat. The inputs are the
row, row distance, number of rows and aisle seats model, 3D model of possible obstructive,
width. The outputs are the aisle axis and aisle seated people model and eye level height. The
region curve. The generation could be con- outputs are the sight quality indicators such as
trolled automatically or manually. In the auto- view angle and view distance. Collision test be-
matic generation, the distance between the tween the sight line and the possible obstruc-
aisles is determined by the maxim number of tive will be operated to show the blocked sight
seats and the seat width. Since the position of lines. First person perspective render can also
aisle is influenced by other factors such as the be obtained to simulate the view of spectators.

Figure 1
R1 to R12 are the rules of
row and seat guide curve
generation.
The contents of Grandstand Grammar tion (Figure 1); rules of aisle generation (Figure 2);
After the identification of the design tasks, the rules rules of seat distribution (Figure 3); rules of elevation
can be translated to a shape grammar called Grand- calculation and elements translation (Figure 4). Figu-
stand Grammar (GG). Rules in GG are organized into res 5 to 7 show the process of using GG to generate
4 groups: rules of row and seat guide curve genera- a single tier grandstand.

Figure 2
R13 and R14 are the rules of
aisle generation.
Figure 3
R15 to R19 are the rules of seat
distribution.
Figure 4
R20 and R21 are the rules of
elevation calculation and
elements translation.
THE COMPUTER IMPLEMENTATION OF GG design. Grasshopper in Rhino3D is chosen as the

The goal of the computer implementation of GG is parametric modeling platform. Scince the seat dis-
to develop a reusable tool for grandstand design. tritution task is strongly relied on the plan drawing
The stability and ease for use should be considered. task, the 2 tasks are incoporated into 1 component.
The user could design with the tool without refer- The components of aisle axis generation, elevation
ing to the detail rules. Therefore the rules should be calculation and 3D model generation are also deve-
sealed in the tool and not to be exposed to the users loped (Figure 8). İnitial sets of parameters are intro-
in order to avoid the miss-operation and the confu- dued to the components to guarentee the ease of
sion of the user. The input and output of the tool use. The user can use and connect the components
should reflect the simple need of the grandstand to solve design problems of the grandstand.

Figure 5
Step 1 to 9 show the gen-
eration of the plan curves for
rows and aisle steps.

Figure 6
Step 13 to 16 show the popu-
lation of seats.
Figure 7
Step 17 to 19 show the genera-
tion of the whole grandstand
3d model.
Figure 8
Main components of the
grandstand design tool.

THE APPLICATION OF GG AND ITS COM- grandstand process. The designer could also adjust
PUTER TOOL the grandstand design according to the evaluation
GG and its computer tool is applied to the design of the roof. The relationship between the grand-
practice of a stadium with 25000 seats. In the prepa- stand and the whole building was reflected in real-
ration phase of the project, new designers were time to enhance the design communication and fa-
trained to learn the knowledge of grandstand de- cilitate the making of design decision (Figure 9).
sign and the use of the parametric model. Their abili- İn the design develpment phase, the parametric
ties to build the 3D grandstand model varied greatly model was used to tackle the intense design modi-
in terms of the understanding of the grandstand fications. During the process, several modifications
rules and the 3D modeling software. However, the were carried out. Curvature and elevatioin of the
learning curves were benefited from the systemati- first row profile, row distance, seat width, number of
cally formulated shape grammar, friendly user inter- rows, elevation of the upper tier, distance between
face and the detailed initial set of parameters. All of aisles, position of vomitories and the boudary of the
the designers could basically operate the tool in the upper east tier are changed compare to the grand-
4 hours session of training. After the training they stand in the design competion phase. Thanks to the
could use the tool to generate simple grandstand rapid response and the flexibility of the parametric
3D models. As we should point out, the tool did not model, the grandstand designs were quickly updat-
turn them into experts of grandstand design in the ed to facilitate the modifications of the other parts
very short time span. Also the digital tool should not of the building (Figure 10).
be seen as a guarantee of good grandstand design. During the design practice, the aims of GG and
In the design process, the model should always be its computer implementation were verified. The
reviewed and evaluated by the experts to avoid de- combination of shape grammar and parametric
sign mistakes. modeling enhance the learning experience of the
In the design competition phase, the tool was grandstand design knowledge. Real-time 3D visual
used to guide the design decision making. A main feedback, performance feedback and the integra-
issue of the project is the configuration of seat num- tion of the whole building enables the architect to
ber on the tiers of the two sides. The change of con- widely explore the design space in the early design
figuration has an important impact on the height of phase. İn the design develpment phase, the parmet-
the grandstand: symmetry configuration results in ric model could cope with the multiple design modi-
the same height of both sides; uneven configuration fications and promote the efficiency of the whole
results in different height of sides. Multiple designs building.
were generated to reflect the relationship between
the height and the seat configuration. REFERENCES
In Grasshopper, model of the grandstand and Gernaint John and Rod Sheard. 2000. Stadia: A Design and
the other parts are inter-related to each other. The Development Guide. Architectural Press.
first row profile and the boundary of the grandstand Andre Shields and Michael Wright (eds.). 1989. Arenas:
are the key shapes for both grandstand design and A planning, design and management guide. Sports
roof design. Therefore different design tasks could Council.
share some key parameters. The design of the 2 Roland Hudson. 2010. Strategies for parametric design in
parts can be carried out simultaneously and sepa- architecture - An application of practice led research.
rately because of the parametric feature of the mod- PhD thesis.
el. Designers could parametrically model the roof Nathan Miller. 2009. ‘Parametric Strategies in Civic Architec-
according to its relationship with the grandstand. ture Design’. ACADIA 09: reForm( ) - Building a Better
Therefore the roof model can be updated in the Tomorrow [Proceedings of the 29th Annual Confer-

Figure 9
3 grandstand designs with
approximate 25000 seats.
Different seat population of
the west and east tiers result
in different heights and form
of the grandstands and the
associated roof structures.
Figure 10
grandstand design of the
schematic design phase (left)
and the design development
phase (right). Many param-
eters were changed during the
design process. The model can
be updated quickly according
to the adjustment of the
parameters.
ence of the Association for Computer Aided Design in

Architecture (ACADIA) ISBN 978-0-9842705-0-7] Chi-
cago (Illinois) 22-25 October, 2009, pp. 144-15

From Point Cloud to Shape Grammar to Grammatical
Transformations
Using Terrestrial Laser surveying to develop and compare shape grammars

Filipe Coutinho1, Luis Mateus2, José P. Duarte3, Victor Ferreira4, Mário Kruger5
1,5
University of Coimbra, 2,3,4 Universidade Técnica de Lisboa
1
http://www.darq.uc.pt./~filipecoutinho, 3http://home.fa.utl.pt/~jduarte, 4http://home.fa.utl.
pt/~victor, 5http://www.ces.uc.pt/whee/investigadores/
1
filipecoutinho@darq.uc.pt, 2lmmateus@fa.utl.pt, 3jduarte@fa.utl.pt, 4Victor@fa.utl.pt,
5
kruger@ci.uc.pt
Abstract. This paper describes a generative design approach integrating real building
data in the process of developing a shape grammar. The goal is to assess to which extent
it is feasible the use of a reverse engineering procedure to acquire actual building data
and what kind of impact it may have on the development of a shape grammar.
The paper describes the use of Terrestrial Laser Scanning (TLS) techniques to acquire
information on the São Vicente de Fora church, then the use of such information to
develop the corresponding shape grammar, and finally the comparison of this grammar
with the grammar of Alberti’s treatise, to determine the grammatical transformations that
occurred between the two grammars.
Keywords. Alberti, shape grammar, shape recognition, design automation,
transformation in design.
INTRODUCTION
This paper is centered on the construction of the on Portuguese architecture in the counter-reform
shape grammar of a Portuguese church called São period (Kruger et al., 2011) to determine the gram-
Vicente de Fora. For this propose a point cloud from matical transformations (Knight, 1983) that occurred
a TLS surveying was used and a part of a church ele- from the original Albertian grammar to the actual
ment (a Doric base) was then closely analyzed. buildings grammars.
This research is part of a wider project aimed at Established in 1147 by King Afonso Henriques
decoding Alberti’s treatise De Re Aedificatoria by in- both the monastery and its church of São Vicente
ferring the corresponding shape grammar using the de Fora had their reformation by King Filipe I in the
computational framework provided by description 16th century.
grammars (Stiny, 1981) and shape grammars (Stiny It is believed that these renovations followed
and Gips, 1972). The goal is to compare the grammar drawings of Juan de Herrera who was in Lisboa by
of the treatise with the grammar of actual buildings 1580-1583 and the drawings of Filipe Terzi (So-
to determine the extension of Alberti’s influence romenho, 1995). The Portuguese architect Baltazar

Álvares was in charge of directing the construction church. Some of these drawings were then used as
from 1597 to 1624. raster images for Autocad and Rhinoceros. Finally,
The church has a unique nave, with a transept after using the 2D drawings from the laser scan-
and a deep main chapel. The walls have pilasters ning survey in the construction of the grammar, an
ornamented with a Doric base, a plain shaft and an element was chosen to evaluate and compare with
almost Corinthian Capital with one level of leafs and Alberti’s rules for drawing such type of element.
in the center of the symbols of Saint Vincent and Namely, a subset of the point cloud that contained
Saint Sebastian. The entablature is Doric with friezes, the best column was chosen for the analysis. The in-
grooves and mutules. The main nave cradle vault is formation contained in the retrieved data was used
coffered and its façades have intercommunicating to develop the grammar of the column system of
chapels built from 1605 to 1629. There was a dome São Vicente de Fora. For this purpose we only con-
over the transept that collapsed during the 1755 sidered a column and the intercolumn.
earthquake. This grammar was then compared them with
In the experiment described in this paper, TLS the Column System shape grammar inferred from
techniques (Mateus, 2012) were used to acquire Albert`s treatise. The deviations were analyzed to
information from the existing church building and identify which rules transformations were needed
work was focused on the column system. A mesh to obtain the new grammar from the treatise gram-
surface of the church elements was generated from mar, and then to determine to which extent Alberti’s
the point cloud obtained using TLS. A Doric col- rules are present in the column system of São Vi-
umn base section line was then detached from that cente de Fora.
mesh. A process of analysis and evaluation of such
line will be showed in the final section of this paper. SÃO VICENTE DE FORA SHAPE GRAM-
MAR
METHODOLOGY In this article we show the rules needed to generate
There was a previous survey of the building using first a colonnade and then the church main nave in-
a phase-based laser scanner (FARO Focus 3D) as terior facade by adapting previously developed Dor-
shown in Figure 1. The scanner was placed in 24 dif- ic and Corinthian grammars, which are part of the
ferent stations for the acquisition of colored point column system grammar developed after Albert`s
clouds. After the registration process a colored point treatise.
cloud model (PCM) was obtained. The registration The Albertian grammar was developed as a
and decimation of the point clouds were done with parallel grammar encompassing four views: plan,
the open source software MeshLAB. For the pur- section, elevation, and axonometric (Coutinho et
pose of this work it was considered a local coordi- al., 2011). The first three views are developed in the
nate frame, aligned with the main directions of the Cartesian product of the algebras U12 and V12, and
church. the fourth of the algebras U13, U33 and V13. The São
The alignment intended to merge pairs of Vicente de Fora grammar has the same structure.
clouds. The final alignment of the PCM was done us- Each grammar rule has a section containing param-
ing their matrix and was the basis for the extraction eters and descriptions and, when needed, a set of
of ortho images, multiple sections and triangulated functions organized in a way similar to the one used
models. Several 2D drawings were produced using in previous grammars (Duarte et al., 2013). In this ar-
JRC software. The final project .Aln containing the ticle, we show only the elevation view due to space
alignment of the 24 point clouds was sectioned in a constraints.
dozen of vertical and horizontal planes configuring Rule 0 has on its left side an empty set and on
the multiple planes, sections and elevations of the its right side a guide line extracted from the site

Figure 1
PCM of São Vicente Fora. 2d
images from JRC. Column
base Mesh surface.
plot, which goes from point A to point A10. Rule intercolumn. M relates to a quarter of the church’s
1 generates recursively a generic structure of the main nave, measured closer to the transept, w1 is
church main compositional elements in eleva- the remaining width, which goes from the pilaster
tion. This grid contains a set of labels A, B, C and D axis to the beginning of the arch. Finally L is equal
inserted in a horizontal line from the bottom to to ½ of the church’s main nave plot minus M. In Rule
the highest central line of the barrel vault ceiling. 2 an insertion point (A2) is given to start the genera-
Each label has several sublabels from Kn to Kn-1 tion of a proto pilaster. This point is obtained from
being K ∈ {A} and n>1; n ∈ {1, 2, 3, 4, 5, 6, 7, 8, 9, the interior of the church structure both in plan and
10} and K ∈ {B, C, D} and n>1; n ∈ {1, 2, 4, 6, 7, 8, 9, in section using the previous rule. Rule 3 call previ-
10}. The lines containing the set points {A, D} and ously developed grammars to insert detailed base,
{A3, D3} are mirror axes. The equation of this set is shaft and capital. Rule 4 inserts an arch from a lateral
L=wr+4d+2ic+3/2IC+M; where the mirrored part chapel and a point B. Rule 5 takes the former arch
of the main nave is M=4d+2ic+3/2IC+w1; wr is the and inserts a Doric entablature with triglyphs and a
church high chorus width, d is the pilaster width, ic point C using point B as a reference. Rule 6 inserts
is the inter chapel’s intercolumn, and IC is the main the barrel vault ceiling on the top of the entablature

Figure 2
São Vicente de Fora church
shape grammar rules.
and a point D using point C as the insertion point. A1`. Rule 12 (e me ma) erases labels. A 3D model of
Rule 7 inserts a half chapel and a pilaster and a line the main nave was generated by applying the gram-
with A3 and C3 points. Rule 8 mirrors the half chapel mar rules. Rules are presented in Figure 2.
using points A3 and C3 to define the mirror axis.
Rule 9 inserts a half chapel, a sub inter column (ic) NEW SHAPES FROM AN OPTIMIZATION
and an axis from point A6 to D6. Rule 10 mirrors two PROTOCOL - THOUGHTS ON EMBED-
sets of pilasters (with labels A9, A8, A4, A2) using the DING
axis with points A and D as the mirror line. Rule 11 As mentioned above, the experiment described in
generates the space to accommodate the high cho- this paper relates to the extraction of data from the
rus using the pilasters with labels A9, A8, A4, A2 and 3D model generated out of the point cloud model

Figure 2 continued
São Vicente de Fora church
shape grammar rules.
and then the analysis of part of it elements. The combination of these L, C, S elements and
There are two main concepts that were taken their parametric variation generate the moulds
into consideration in the task of evaluating a line Ovule, Channel, Wave and Gulens. In turn, the com-
extracted from the point cloud. One is the notion of bination of these moulds gives different column
LCS system (Figure 3) mentioned in Alberti’s De Re system elements, such as Pedestal, Base, Column
Aedificatoria and the other is George Stiny’s notion (Shaft), Capital and Entablature. Finally, these might
of embedding. be used to obtain different combinations of Doric,
In Book VII, Chapter VII of the Re Aedificatoria Ionic, Corinthian and Composite style elements.
(Alberti, 2011), while describing the Bases and the The combinations of these column system elements
Capitals of the column system, Alberti mentions that may produce around 900 different columns. This
these can be constructed from a minimum vocabu- is the size of the language of columns that can be
lary composed by the letters L, C, S, reversed C and generated from the LCS system, and which might
reversed S. Is to be noted that Alberti`s original trea- be recognized using the system showed in the next
tise edition contains no drawings. section. In this way, Alberti was providing a proce-

Figure 3
LCS system shape grammar.
dure to generate almost all the column system with according to him quantity, proportion, and location.
a sub system that was embedded in it. Finally anoth- The rules in use are rules of the type x→x` de-
er function of the LCS system was to provide a loca- fined by Stiny, meaning that a shape x is trans-
tion for decorative elements like flowers, leafs, and formed in a similar shape with parametric variations
eggs, which are not addressed in this paper. (Stiny, 2011). If we get rid of the parameterization,
The column systematization shape grammar we may obtain rules of the type x→y, which trans-
that can be developed from the LCS system is a form a shape x into another shape y.
grammar of detail and it is identified as a bottom Both x and y are elements in the index of dimen-
up shape grammar. The results of the experiment sions i and j, where i ∈ {0d, 1d, 2d, 3d}, that is, points,
described in this paper support this hypothesis. Ap- lines, planes, solids, and j ≥i). In the LCS vocabulary,
parently the rules from the treatise are to be applied C is part of the base that is part of the column and so
in an almost straightforward fashion. But Alberti`s on. A general definition of this embedding feature
established that the designer must use them as is x→prt(x).
pleased in order to achieve “concinitas”, that means In this particular case, LCS shapes are boundary

Figure 4
Diagram specifying the
operations for transforming a
grammar.
elements of the columns. Let’s take then the defini- this paper, r2, r4, r5 and r7 ∈ I; r6, r10 and r11 ∈ A; r0,
tion rule x→b(x) to encode the transformations that r1,r8 and r9 ∈ C; and r3 ∈ S.
occurred in the design of an element from an origi- As Knight mentions, in rule change transforma-
nal grammar to a transformed grammar. tions C shapes are defined as transposed shapes,
that is, as new shapes or as resized and/or reposi-
THE GRAMMATICAL TRANSFORMA- tioned shapes. Rules 2, 4, 5 and 7 are equal to those
TIONS - EVALUATION PROCESS AND found in the original Albertian grammar. In Rule 3
FEEDBACK the changes verified are in the constituents of the
One role of the grammar is to help tracing the influ- Capital. The disposition of the capitals are similar to
ence of Alberti’s treatise on the design of the São those found on the second level of the Palacio Ru-
Vicente de Fora church by verifying whether its ele- cellai’s façade. In this case, there is a simultaneous
ments can be obtained from Alberti’s rules or some subtraction and resize and reposition transforma-
sort of transformation of such rules. tions. The Shaft, Base and Capital’s heights are equal
The Transformations in Design framework pro- to the ones described in the treatise. Rule 6 adds a
posed by Knight (1983, 1994) -- according to which new element to the grammar, a barrel vault ceiling.
the transformation of one style into another can be Is to be noted that this element don’t belong to the
explained by changes of the grammar underlying column system. Rule 10 adds a new chapel and Rule
the first style into the grammar of the second -- will 11 the high chorus. Rule 8 and 9 a mirror but chang-
be used as the theoretical background. es the axis location. Is to be noted that in the treatise
According to Knight, There are at least four is not specified the notion of mirror but of symme-
different ways of transforming a grammar as dia- try. Rule 0 and Rule 1 change the location of labels.
grammed in Figure 4, namely, rule addition, rule These rules manipulate elements that need to
subtraction and rule changing, which can be des- be more closely observed. A technique to recognize
ignated by letters A, S, and C, respectively. A fourth and analyze sets of curves was used. This technique
transformation type I can be added if we consider consists in a Grasshopper code (Figure 5) whose
that a rule can remain unchanged, This transfor- main goal is to extract and compare section lines
mation I is important for our study because each from an element of the column (a Doric Base torus),
time such a transformation is used there is strong which proceeds automatically in three different
evidence that the designer was knowledgeable of steps:
Alberti`s rules, as seen in the Loggia Rucellai shape The first step is the extraction of a line section
grammar by Alberti himself (Coutinho et al., 2013). from a mesh surface out of the point cloud.
In the São Vicente de Fora grammar presented in The second step consists in comparing this

Figure 5
GH code to identify and ana-
lyze segments of the column
system elements.
line with a curve previously embedded in the code mar; and third, curves that are sections of parts of
through the distances between the control points of the “real” building obtained from the PCM are auto-
both lines, which need to be bigger than 0. In rule matically recognized, analyzed and evaluated sug-
x→b(x), b is such that (in this particular experiment) gesting that this technique might be an efficient
b>0; and b ∈ B where B ∈ LCS. way of evaluating large data sets.
Then, the third step analyses the difference in The use of such survey data to generate the
value of linear distance and rejects the ones that grammar was of great help, particularly, considering
are not in the acceptable range. This last stage is the level of accuracy and detail that is possible to
not completely implemented yet. A similar process achieve from such a method.
using canonical representation (Keles et al., 2010) The code to automate the shape recognition
graphs are in use in order to better visually under- proved to be helpful but improvements are neces-
stand the differences and similarities between the sary, namely the generation of mesh surfaces direct-
topology of different lines. As said above, the con- ly from the PCM in a complete automated way.
trol points of the target curve and the points of sec- The process for choosing the curves from the
tion extracted from the cloud are points contained corpus (that are the models to be merged) needs to
in parallel lines. The distance measured is the seg- be optimized and is not completely defined. So far a
ments of such lines. This process is not completely linear distance is in use but the notion of neighbor-
efficient. It might work well for straight lines con- hood (Krishnamurti and Stouffs, 2004) might be of
tained in parallel planes but not in the case of curves great help in order to understand the kind of trans-
in the 3d space. It is interesting to note that the pro- formations occurred in the application of the rules
cess used in this experiment reduces the Algebras so in different buildings. This task will be the focus of a
that U33 → U12 → U02. future research article.
CLOSURE ACKNOWLEDGEMENTS
The contributions of this paper are threefold: first, This work is part of the “Digital Alberti” project fund-
it is the first Portuguese grammar obtained from ed by Fundação para a Ciência e Tecnologia (FCT),
Alberti`s treatise shape grammar rules; second, it Portugal, and hosted by CES at the University of Co-
uses a laser surveying and the resulting point cloud imbra (PTDC/ AUR/64384/2006) and by ICIST at the
model (PCM) as a way of transforming the grammar Technical University of Lisbon. The project is coordi-
and develop the São Vicente de Fora shape gram- nated by Mário Krüger. Filipe Coutinho is funded by

FCT with PhD grant SFRH/BD/81260/2011. rimento laser 3d terrestres, PhD Thesis, Faculdade de
Special thanks to Prof. Luis Mateus and to Fac- Arquitectura, Universidade Técnica de Lisboa, Portugal.
uldade de Arquitectura da Universidade Técnica de Keles, HY, Ozkar, M and Tari, S 2010, ‘Embedding shapes
Lisboa for the laser scanner FARO Focus 3D use. without predefined parts’, Environment and Planning B:
Planning and Design, 37, pp 664 - 681.
REFERENCES Knight, TW 1983, ‘Transformations of language of design’,
Alberti, LB 2011, Da arte edificatória, translated by AF Santo, Environment and Planning B: Planning and Design 10,
Calouste Gulbenkian Foundation, Lisbon, Portugal. pp. 125-177.
Duarte, J, Figueiredo, B, Costa, E, Coutinho, F and Kruger, Knight TW 1994, Transformations in Design: A Formal Ap-
M 2013, ‘Alberti Digital: investigando a influência de proach to Stylistic Change and Innovation in the Visual
Alberti na arquitetura portuguesa da contra-reforma’, Arts, Cambridge University Press, Cambridge, UK, pp
in CAL Brandão, F Furlan, P Caye and M Loureiro, Na 41-107.
gênese das racionalidades modernas: em torno de Leon Krishnamurti, R and Stouffs, R 2004, ‘The boundary of a
Batista Alberti, Editora UFMG, Belo Horizonte, Brazil, pp shape and its classification’, Journal of Design Research,
488-548. 4(1).
Coutinho, F, Castro e Costa E, Duarte, JP and Kruger, M 2011, Kruger, M, Duarte, JP and Coutinho, F 2011, ‘Decoding De
‘Interpreting De re edification - a shape grammars of Re Aedificatoria: using grammars to trace Alberti’s in-
the orders system’, Proceedings of International Confer- fluence on Portuguese classical architecture’, Nexus
ence eCAADe 2011, Ljubljana, Slovenia, September 21- Network Journal, 13(1), pp. 171 - 182.
24, pp 788-198. Soromenho, M 1995, ‘Do Escorial a São Vicente de Fora’,
Coutinho, F, Castro e Costa, E, Duarte, JP and Kruger, M Monumentos, nº 2, Lisboa, pp 24-26.
2013, ‘A shape grammar to generate Loggia Rucellai’, Stiny, G and Gips, J 1972, ‘Shape Grammars and the Gen-
in R Stouffs, PHT Janssen, S Roudavski, B Tunçer (eds), erative Specification of Painting and Sculpture’ in CV
Open Systems: Proceedings of the 18th International freiman (ed), Information Processing, vol 71, North-Hol-
Conference of the Association of Computer-Aided Archi- land, Amsterdam, pp. 1460-1465.
tectural Design Research in Asia, CAADRIA, Singapore, Stiny, G 1981, ‘A note on the description of designs’, Environ-
May 15-17, pp. 791-800. ment and Planning B: Planning and Design, 8, pp. 257-
Mateus, L 2012, ‘Contributos para o projecto de con- 267.
servação, restauro e reabilitação uma metodologia Stiny, G 2011, ‘What Rule(s) Should I Use?’ Nexus Network
documental baseada na fotogrametria digital e no var- Journal, 13(1), pp. 15-47.

Languages of Design
Languages of Design - Volume 2 - Computation and Performance - eCAADe 31 | 665

666 | eCAADe 31 - Computation and Performance - Volume 2 - Languages of Design
Combining Complexity and Harmony by the Box-
Counting Method
A comparison between entrance façades of the Pantheon in Rome and Il

Redentore by Palladio
Wolfgang E. Lorenz
Vienna University of Technology; Institute of Architectural Sciences; Digital Architecture
and Planning
http://www.iemar.tuwien.ac
lorenz@iemar.tuwien.ac.at
Abstract. When Benoît Mandelbrot raised the question about the length of Britain’s
coastline in 1967, this was a major step towards formulating the theory of fractals,
which also led to a new understanding of irregularity in nature. Since then it has become
obvious that fractal geometry is more appropriate for describing complex forms than
traditional Euclidean geometry (not only with regard to natural systems but also in
architecture). This paper provides another view on architectural composition, following
the utilization of fractal analysis. The procedure concerning the exploration of a façade
design is demonstrated step by step on the Roman temple front of the Pantheon by
Appolodorus and its re-interpretation – in the particular case the entrance front of
Il Redentore, a Renaissance church by Palladio. Their level of complexity and range
of scales that offer coherence are visualized by the specific measurement method of
box-counting.
Keywords. Fractal analysis; box-counting method; Pantheon; Il Redentore; Palladio.
INTRODUCTION
This paper has two objectives: rent study focuses on the overall viewpoint specified
1. The first one concerns the description of har- by a harmonic expression of distributions across
mony defined by the appearance of architec- different scales. The author uses for the first time a
tural elements of different sizes and scale. particular fractal analysis method as measurement
2. The second one utilizes the first one, introduc- of reminiscence, applied to the Roman temple front
ing an objective comparison method between of the Pantheon (built between 110 and 125 AD by
an architectural design (acting as origin) and its Appolodorus) and the Renaissance temple front of Il
historical followers. Redentore in Venice by Palladio (groundbreaking in
Apart from an analysis concerning the utiliza- 1577) – The Pantheon was chosen as Palladio (1984)
tion of characteristic architectural elements, the cur- emphasized the particular importance of that build-

ing. As benefit of the quantitative method, similarity category is, in terms of harmony, a consistent
between two façades can be proved with regard to whole, which is reflected in all of its parts – a
visual complexity. concept that is rather close to Fractal geometry
(Mandelbrot, 1981; 1982).
Fractal analysis
Fractals – the term was introduced by Mandelbrot Harmony and Box-Counting
in 1975 – are characterized by specific properties, Harmony fulfills the expectations of the observer for
which include development through iterations, in- a quantity of new architectural elements on smaller
finite complexity, roughness, irregularity, scale in- scales that reflect – at least in their roughness – the
variance and self-similarity. The latter is a central fea- whole (Salingaros 2006). However, parts need not be
ture – although not a guarantee that the structure exact, scaled down copies of the whole, but should
is fractal – and sometimes, if statistically, difficult to reflect the basic motif or the basic idea with varia-
describe. In mathematical terms, a self-similar com- tion (Lorenz 2011). Otherwise the result gets monot-
position exists, if parts look exactly or approximately onous or in the other extreme confusing. In short, a
like the whole. With variations, however, it is difficult continuing irregularity is the reflection of a harmo-
to detect the basic connection between the whole nious connection between the whole and its parts
and its parts, or, in other words, to decipher the un- (as it is true for a theme in music). Nevertheless, due
derlying rules. Characterization is then provided by to the process of building, the intention of the ar-
the Hausdorff dimension – Mandelbrot (1982) calls chitect and material restrictions, fractal characteris-
it fractal dimension – which in the case of a fractal tics are, in any case, restricted to a certain range of
structure exceeds its topological dimension. In ad- scales.
dition, according to Bovill (1996), visually, fractal The starting-point of our investigation is the def-
dimension is the expression of the degree of rough- inition of a harmonic whole by an appropriate bal-
ness – that is how much texture an object has. With ance between the number of architectural elements
regard to architecture, it specifies the relationship of different sizes and the respective scale of consid-
between a building unit on a higher level (larger eration. The characteristic values remain the same,
scale) and its components on a lower level (smaller irrespective of the considered detail. Box-counting –
scale). Throughout this paper, in order to measure a fractal analysis method introduced by Mandelbrot
the fractal dimension, box-counting – whose result (1982) – enables the examination of how character-
is equivalent to the fractal dimension – is used as istics of a structure (details) change with scale. If this
fractal analyzing method. method is applied to a façade, this means basically,
As is described elsewhere (Lorenz, 2012), fractal to translate its two-dimensional representation (the
analysis in architecture ostensibly leads to two dif- elevation) into a grid-based Pixel image for the pur-
ferent groups: pose of getting the number of boxes that cover the
3. The first one includes buildings with rather image. This can sufficiently be achieved by placing
smooth façades and a few well distinguishable a grid over the plan in order to count those boxes
architectural elements. Such a conception in- that contain a significant part of the elevation –
dicates closer relationship to Euclidean geom- represented by lines (Figure 1a). Subsequently, the
etry. scaling factor si, given by the reciprocal number
4. In contrast, the second group comprises build- of boxes in the bottom row of the grid, is reduced
ings with elements of many different scales and covering boxes Ni are counted again. This pro-
whose number increases while scale decreas- cedure is repeated depending on the scale of the
es and whose smaller parts reflect the whole plan, i.e. until the detail richness corresponding with
through a common idea. An object of this the distance of the observer is reached. Finally, in a

Figure 1
a) Pantheon: A grid is placed
over the front view of the Pan-
theon. Those boxes that cover
the composition are colored
gray. The reciprocal number
of boxes at the bottom row
defines the scaling factor.
b) Il Redentore: A given coeffi-
cient of determination R² leads
to a specific range of scale and
finally to the box-counting
dimension DB. DB is equivalent
to the slope of the regression double-logarithmic graph with the number of boxes rameters that influence the result in the one or other
line in the graph. Ni versus scaling factor si, the slope of the regression way have to be taken into consideration (see section
line defines the box-counting dimension DB (Figure Influences by Parameters). Some of them, such as
1b) for a certain range of scales (Foroutan-pour et al. line thickness, have been solved by the author’s im-
1999). plementation of the algorithm in a CAAD software
Concerning the box-counting method, a con- (Lorenz, 2009; 2012). Other factors are still unsolved,
sistent whole across many scales is expressed by a e.g. the definition of what is measured, concerning
continuing characteristic of complexity, with the the selection of relevant parts of a façade and its
characteristic of complexity given by the relation translation to a plan (elevation). As a consequence,
between scale and number of boxes covering the el- one part of this paper deals with the application of
evation. A small deviation signifies the continuation a fractal analysis method for the purpose of figuring
of a similar irregularity across different scales. Con- out a correct and efficient way of a grid-based repre-
sequently, it is the straight part of the data-curve in- sentation of an elevation on plan and of testing the
dicating a harmonious distribution (Figure 1b). The box-counting method implemented in AutoCAD. As
straight part is expressed by a coefficient of determi- a word of notice, plans that are used throughout this
nation R² close to one. Hence, in turn, with a certain paper have been prepared in the same manner to
single measurement, the smallest and the largest guarantee consistence.
scale act as limits of the specific range of coherence,
derived from a given minimum value for the particu- Box-Counting as Comparison Method
lar characteristic coefficient (Figure 1b). Throughout history of architecture, one is confront-
Bovill (1996) was the first who applied box- ed with buildings that refer to preceding epochs.
counting to architecture as a method for measur- Descriptions of visual complexity provide a means
ing the characteristic visual complexity of buildings. for comparison, independent of rearranged com-
Since then, it has been used by many research- ponents or of changes of the purpose that the re-
ers (Zarnowiecka, 1998; Lorenz, 2003; Ostwald et spective building is used for (church/villa), meaning
al., 2008; Vaughan et al., 2010). Advantages of the that the characteristic values of complexity detect
method are on the one hand its easy usage (hence connections between two related buildings. In the
its simple implementation) and on the other hand specific case, a Roman temple front, the Pantheon
its applicability to any object (with and without self- in Rome, serves as a starting point, while the Renais-
similar characteristics). Nevertheless, in order to use sance building Il Redentore in Venice represents its
box-counting as a comparison method, several pa- successor. Andrea Palladio, the architect of the latter,

used the motif of interlocking different combina- with a dominant middle order) when commissioned
tions and modifications of classical temple fronts de- to design the façade of San Francesco della Vigna
liberately as a harmonic transition from the entrance in Venice in 1562. Finally, both, San Giorgio Mag-
view to the dome (Wundram et al., 2004). giore and Il Redentore in Venice, act as results of his
The study is based, on the one hand, on the as- continuing development to combine the strict im-
sumption that higher complexity leads to a higher pression of classical temple fronts in a three-aisled
box-counting dimension and, on the other hand, church – with Il Redentore providing an obvious re-
that the harmony of a composition is reflected by a lation to the Pantheon (Puppi, 1994).
trend of the results, i.e. by a straight line of the data-
points in a double-logarithmic graph with grid-scale Andrea Palladio and Venice
versus number of boxes that cover the composition Palladio’s first assignment in Venice, and moreover,
(see section Harmony and Box-Counting). On this his first practical work on a church was the redesign
basis, the paper describes a further development of the façade of San Pietro in 1558. However, it was
of the concept with two aspects as indices of com- not executed before 1594 – presumably because
plexity: the box-counting dimension (Bovill, 1996) of the commissioner’s, the Patriarch Vicenzo Diedo,
and the interquartile range (Lorenz, 2012) – i.e., the death and in a modified form (Puppi, 1994). The first
robust estimate of the variability of the data under design Palladio actually executed in the city of Ven-
consideration gives a valuable description of visual ice was the Convento della Carità (convent of Santa
complexity and harmony. This suggests that if the Maria della Carità), the construction of which began
harmonic expression (given by the range of scales) in 1561. The concept is based on a Roman house
and the height of the characteristic box-counting di- transformed into monumental scale. While the atri-
mension are similar for the ancient temple and Il Re- um and a cloister beyond it consist of a Corinthian
dentore, Palladio’s interpretation follows its historic order, the inner court represents a vertical stacking
inspiration with regard to harmonic expression. of three different orders, with the Doric at the base,
the Ionic in the middle and the Corinthian at the up-
ANDREA PALLADIO per level (Society for the Diffusion of Useful Knowl-
Palladio’s (1508-1580) work is characterized by re- edge 1840).
discovering and applying classical Roman architec- In the city of Venice, Palladio, well entrusted with
ture – strongly influenced by five travels to Rome designing villas and palazzos, finally could translate
conducted in the period between 1541 and 1554, classical orders – which he regarded as the embodi-
during which he studied classical buildings cap- ment of beauty – to two churches. At first he got
tured in various drawings. The results of his studies the commission for San Giorgio Maggiore situated
were first published in “L’Antichità di Roma” (Palladio, on San Giorgio di Castello in 1564. The front façade,
2009), a list of preserved and recovered monuments which is composed by two different reminiscences
of Rome as they there stand by the mid 16 century. of classical temple fronts, was finished 30 years after
In his book, which is entirely textual, Palladio dedi- his death (in 1610). The front façade is dominated
cates more lines to the Pantheon than to any other by its middle part, the entrance, consisting of four
monument. His views are, however, based solely on three-quarter columns of Composite order on high
existing references. Influences of Palladio’s later un- pedestals, supported by a pediment. The second
derstanding of form can be deduced from drawings temple front covers the church aisles by two halves
he made of the Pantheon, in which he develops two of a pediment. Visually it continues behind the first
gables at the same façade (Puppi, 1994). Later, Pal- temple front which is supported by the use of pilas-
ladio brought forward the topic of overlapping ga- ters (of Corinthian order) instead of columns. Both
bles (establishing interlocking architectural orders sides are nevertheless held together by the hori-

zontal entablature (especially the cornice), which ued along the dominant temple front (behind the
continues along the main temple front while the up- cut off columns), it is now interrupted and only con-
per part of the tympanum is interrupted. Moreover, tinues in form of the pediment supplementing the
the pilasters of either side of the entrance belong to entrance. This time, the pilasters of the second order
the second temple front. Finally, decoration is only are protruding in the middle part in the form of two
found with columns, entablatures and niches. half columns flanking the entrance. Moreover, the
intercolumns change from broad, narrow, broad,
Il Redentore – composition and architec- narrow, broad in the case of the earlier church to a
tural elements more harmonic sequence of narrow, narrow, broad,
Towards the end of his life, Palladio was commis- narrow, narrow in the case of Il Redentore. The fron-
sioned to plan his second church in Venice, Il Re- tal view of Il Redentore provides a third temple front
dentore, situated on the island of La Guidecca. The formed by the upper part, including the backwards
erection of the (procession, monastery and) votive sloping roof as pediment and the side parts sweep-
church was decided after Venice had been visited ing the aisles.
by a plague in 1575, which killed forty thousand of
the citizens. The construction work began in 1578, FRACTAL ANALYSIS
only two years before Palladio’s death. Concerning Methodologically, the author follows the box-count-
urban planning, the task was similar to San Giorgio ing algorithm described in Lorenz (2009; 2012). As
Maggiore in setting up a connection between the noted elsewhere (Lorenz, 2012), results are either
new church and Piazza di San Marco across the wa- influenced by the transformation of the façade into
ter. The composition of Il Redentore is similar to San a plan – hence, the preparation of the plan – and by
Giorgio Maggiore in so far that the dominant middle certain factors that are coming along by the method
part of the front view is formed by a large Compos- itself (Foroutan-pour et al., 1999). In consequence of
ite order, while a broader Corinthian order supports the transformation into a plan and to ensure consist-
the flanking aisles as a transition to the high middle ency in analysis, the author considers vector-based
nave. Both façades provide reminiscence of interre- re-drawings of both façades concerned in this paper.
lating Roman temple façades and are characterized
by simplicity in the ornaments. Differences only be- Influences by Parameters
come obvious on closer view. Concerning Il Reden- One of the most crucial aspects influencing the re-
tore, the middle order is placed on a higher platform sult is the choice of significant parts of the eleva-
and consists of lower pedestals (which look more tion – i.e., translating the façade into a black and
familiar). Moreover, while the dominant temple white plan (Lorenz, 2003; 2009). In consequence, the
front of San Giorgio Maggiore consists of four three- choice of represented architectural elements has to
quarter columns of Composite order, the entrance be defined unambiguously, referring to the visual
of Il Redentore is flanked by two half-columns of perception (Bovill 1996; Lorenz, 2003; 2009), and
larger intercolumniation followed by one pilaster on justified carefully. Bovill (1996) refers to Maertens
each side (both again of Composite order). The mid- (1884) when defining the relation between distance
dle dominant front does no longer appear to stand and smallest detail (Lorenz, 2009). The smallest de-
free (as the wall behind continues above the gable). tail, for instance, is influenced by the reading field,
Another difference concerns the position of the hor- that is the minimum size of clearness of seeing with-
izontal cornice of the smaller order which is in the in an eye angle of 0°1’. In addition, Märtens distin-
latter case much higher in relation to the columns guishes between three distances of observation that
and pilasters of the middle order. While at San Gior- correspond to the scale of the façade (in meters). The
gio Maggiore this architectural element was contin- first one includes the environment (deduced from a

viewer’s position of 18-20° of building height), the 1/4th, 1/8th and so forth) and the number of
second considers the whole building (27° of build- boxes at the smaller side (from which the ini-
ing height) and the third one focuses on details (45° tial grid-size is deduced). Furthermore, for the
of building height). The present measurements cor- purpose of accuracy the number of displace-
respond to the second and third distance. From a ment in x- and y- direction can be defined. The
distant view, only main parts of the design are per- number of covered boxes of a certain box-size
ceived and consequently taken into consideration. is then given as the minimum number of all re-
Beside the silhouette, this includes columns, the placements of one and the same grid-size.
gable, main parts of the entablature (architrave and • Changing settings of layout. This includes draw-
cornice), but no detail of the capitals. The latter being a copy of the measured segment and as-
longs to a closer distance of observation. signing this segment to a corresponding layer.
The experimental set-up not only includes the A second modified algorithm does not start
selection of lines, but also the definition of the small- from a reduction factor of one half but takes into ac-
est and largest box-size. While the smallest box-size count that the difference in box-size between two
depends on the smallest detail and is reflected by successive grid-sizes is larger as the scale size in-
the point where the data curve calculates only the creases. Consequently, the user can adjust accuracy
single lines of the elevation (DB = 1), the largest box- by a value that defines the addition of boxes from
size should be one fourth of the smallest side of the one scale to the next, where the number of added
measured image. Other influences include the rela- boxes increases with smaller scales.
tive position of the grid, the orientation of the grid Finally, the data, coming along as text-file, is es-
and the reduction factor of the grid. With a reduc- timated by means of statistical methods, specifically
tion factor of one half, the number of boxes at the by linear regression. With this technology, a regres-
bottom row doubles for the next smaller grid-size. sion line is to fit the logarithmically transformed out-
put of grid-scale versus number of covered boxes. In
Implementation the particular case, the analysis of the data is done in
The author’s implementation of the grid-based box- a spreadsheet program, again supported by a spe-
counting algorithm into AutoCAD uses vector-based cial script. For evaluating the relation of the regres-
geometries in a tool architects are used to. The script sion line with regard to the measurement points, the
allows various options, which are available in form coefficient of determination R² is used. The range of
of tabs: R² reaches from 0, indicating no relation, to 1, which
• Selecting the area for measurement. If the area means highest possible correlation. As more than
contains the image completely, the bounding one measurement is used for analysis (see section
box serves for further calculations, otherwise it A Set of Measurements), the range of scales can be
is cut. adopted for the whole set, resulting in a minimum
• Defining specifications. This includes, for exam- and maximum coefficient of determination R². Only
ple, the number of iterations (how often the if both, the minimum and the maximum are close to
grid-size is reduced), the enlargement factor one, the result is called ‘consistent‘. Otherwise, the
(percentage of empty space around the selec- regression line does not fit for single results, indicat-
tion area), the number of steps between two ing higher diversity.
scales (the reduction factor – which is defined
as the ratio of how much the grid-size is re- A Set of Measurements
duced from one step to the next – is defined It could be demonstrated elsewhere (Lorenz, 2009;
by one half; by inserting a number of steps 2012) that different measurements lead to different
between two scales the factor is changed to results, due to influences of several parameters com-

ing along with the box-counting method. Therefore, four) and
a set of measurements is necessary, rather than a • the number of replacements in x- and y-direc-
single one. Accuracy is then expressed by the in- tion (one by one or three by three).
terquartile range of the box-plot (containing 50% The interquartile range and the coefficient of
of all values). The smaller this range, the smaller is determination are the basic instruments of evalua-
the fluctuation of data-points (single box-counting tion of the results: While the first value is related to
dimensions) and the more meaningful is the result. the whole set, single measurements are taken into
The characteristic values are therefore account by the second criterion. In particular, the
• the range of scale, given by the smallest and latter is specified on the one hand by the minimum
largest box in meter, R², which tells us about the most deviating result of
• the median, as a characteristic for roughness a whole set, and, on the other hand, by the average
and R², which describes the general fluctuation of data
• the interquartile range, as indication of varia- of all measurements.
tion. When discussing the results of measurements
In turn, a given coefficient of determination it is conspicuous that for any single measurement
leads to a specific range of coherence for a whole of Il Redentore, the coefficient of determination
set of measurements and, following from that, to a exceeds 0.996 (0.997), which is very close to one,
definitive value by the median of the data (Lorenz, proving that each regression line fits the data well
2012). (minor deviation). Finally, the spectrum of the result-
ing box-counting dimensions (slope of the regres-
ELABORATION sion line) is shown in a box-plot, separately for set
To ensure the required conditions of the author’s A and B (Figure 2a). The respective small interquar-
implementation of the box-counting method, a tile ranges express high accuracy of all data: For
vectorized representation of the real façade Il Re- set A it is 1.89 percent (in relation to two as possi-
dentore is considered. As statues were added only ble results in a two-dimensional space are between
in the second half of the 17th century (Wundram et 0 and two) while for set B it does not even exceed
al., 2004), they are excluded from measurement, as 1.5 percent. Finally, the median of each set – that is
well as small details including the shaping of capi- the break line where 50 percent of all values can be
tals. In general, the selection includes main design found above and below respectively – equals 1.677
elements according to a distance from where the and 1.685. From these results, it can be deduced that
building is perceived as a whole (see section Fractal the façade of Il Redentore is of high complexity, with
Analysis). a consistent use of architectural elements from the
In order to minimize potential sources of error, whole to a very small scale (Table 1).
both algorithms are used – dividing the grid by one As shown in Table 1 the results of both algo-
half (set A) and adding boxes for each step (set B) rithms are very close with slightly higher accuracy of
– with 11 different configurations in each case. The the gradual increase of boxes, i.e. set B (higher min
configurations include: R² and smaller interquartile range).
• the factor of enlargement (either one, three or
five percent of minimal side length), DISCUSSION AND COMPARISON OF
• the number of starting boxes in x-direction (ei- RESULTS
ther three of four), Because of its importance for Palladio (see section
• depending on the algorithm, either the num- Andrea Palladio), it is the Pantheon in Rome that
ber of steps between two grid-sizes (none, one, serves as a reference object. For analysis, two sets
two or three) or the factor of accuracy (three or (A and B) of 11 measurements each are carried out.

Il Redentore Table 1
Median 1.677 1.685 Il Redentore: Results of meas-
Interquartile range 0.038 (1.89%) 0.029 (1.45%) urement; Left: dividing by half;
Minimal R² 0.996 0.997 Right: adding boxes.
Average R² 0.998 0.998
Range of coherence
Maximum box-size 7.93 meters 8.46 meters
Minimum box-size 0.32 meters 0.31 meters
Range in % of the height of the front view
Maximum box-size 29.95 % 31.95 %
Minimum box-size 1.20 % 1.16 %
Despite differences of overlapping elements, the re- and the pillars flanking the entrance with own ga-
sults nevertheless display a similar range of coher- bles display another additional level.
ence in comparison to Il Redentore (Table 2 and Fig- Concerning the different algorithms, both sets
ure 2b). Moreover, the medians of the two sets are of measurement lead, as it is true for Il Redentore,
similar to Il Redentore: the median of set A equals to very similar results (Table 2). The deviation of the
1.661 (1.677) and for set B it is 1.660 (1.685) – with data is again low, although this time minimum R² is
slightly higher interquartile ranges of 2.07 and 1.32 slightly lower (0.992 and 0.994) than in the case of Il
percent. This leads to the conclusion that both fa- Redentore (0.996 and 0.997).
çades are characterized by a similar development of
architectural elements across a similarly broad range CONCLUSION
of scales (range of coherence: 1-30 percent with Il The box-counting method provides an objective
Redentore and 1.5-28 percent with the Pantheon). comparison method between design solutions
In particular, this means that details of a certain demonstrated by Il Redentore and the Pantheon.
size have their correspondence in both façades, al- It visualizes the development of roughness across
though differences in design are obvious. E.g., Il Re- multiple scales and, derived from that, the harmonic
dentore, for instance, displays not only one but two relations between the whole and its parts. Both re-
clearly interrelating Roman temple façades, while sults discussed in this paper show a similar depth of
the Pantheon consists of two vertically arranged details and a similar level of complexity. Specifically,
gables. In the case of Il Redentore, niches for statues this means that, even if Palladio changes the com-
Figure 2
Il Redentore and Pantheon:
box plot diagram of box-
counting dimensions (a) and
box size in percentage of the
height of the front view (b).

Table 2 Pantheon
Pantheon: Results of measure- Median 1.661 1.660
ment. Left: dividing by half; Interquartile range 0.041 (2.07%) 0.026 (1.32%)
Right: adding boxes. Minimal R² 0.992 0.994
Average R² 0.994 0.995
Range of coherence
Maximum box-size 9.10 meters 9.10 meters
Minimum box-size 0.55 meters 0.50 meters
Range in % of the height of the front view
Maximum box-size 28.40 % 28.40 %
Minimum box-size 1.71 % 1.57 %
position of the temple front, the harmonic distribu- HB Schmiedmayer, H Stachelberger and IC Gebeshu-
tion across all scales is similar to the Pantheon. This ber (eds), Biomimetics – Materials, Structures and Pro-
proves that, although variations in the reinterpreta- cesses: Examples, Ideas and Case Studies, Springer, Ber-
tion occur, Il Redentore nevertheless takes up the lin, pp. 179-200.
same characteristics as its origin of a Roman temple Lorenz, WE 2012, ‘Fractal Geometry of Architecture: Imple-
front. mentation of the Box-Counting Method in a CAD-soft-
Box-counting reveals similarities and differences ware’, Proceedings of the eCAADe Conference, Prague,
between styles with regard to different degrees of Czech Republic, pp. 505-514.
roughness and depth of self-similarity. Up to now, Maertens, H 1884, Der optische Maßstab: oder die Theorie
the author has analyzed façades, corresponding to und Praxis des ästhetischen Sehens in den bildenden
a larger distance of the observer. As ornaments are Künsten, Wasmuth, Berlin.
characteristic elements of a building, it would be in- Mandelbrot, BB 1981 ‘Scalebound or scaling shapes: A use-
teresting for future work to deal with a smaller dis- ful distinction in the visual arts and in the natural sci-
tance as well. ences’, Leonardo, Vol. 14, No. 1, pp. 45-47.
Mandelbrot, BB 1982, The fractal geometry of nature, W.H.
REFERENCES Freeman, San Francisco.
Bovill, C 1996, Fractal Geometry in Architecture and Design, Ostwald, MJ, Vaughan, J and Tucker, C 2008, ‘Characteristic
Birkhäuser, Boston. Visual Complexity: Fractal Dimensions in the Architec-
Foroutan-pour, K, Dutilleul, P and Smith, DL 1999 ‘Advances ture of Frank Lloyd Wright and Le Corbusier’, Nexus VII:
in the implementation of the box-counting method of Architecture and Mathematics, 7, pp. 217-232.
fractal dimension estimation’, Applied Mathematics and Palladio, A and Beyer, A 1984, Die vier Bücher zur Architektur
Computation, Volume 105, Issue 2-3, pp 195-210. (I quattro libri dell’architettura, Venice 1570), Verl. für Ar-
Lorenz, WE 2003, Fractals and Fractal Architecture, Master chitektur Artemis, Zürich.
Thesis, Department of Computer Aided Planning and Palladio, A and Davis, MD 2009, Andrea Palladio: L’Antichità
Architecture, Vienna University of Technology, Vienna. di Roma Raccolta brevemente da gli auttori antichi,
Lorenz, WE 2009, ‘Fractal Geometry of Architecture – Imple- et moderni; Nuovamente posta in luce (Rom 1554),
mentation of the Box-Counting Method in a CAD-Soft- Universitätsbibliothek der Universität Heidelberg, Hei-
ware’, Proceedings of the eCAADe Conference, Istanbul, delberg.
Turkey, pp. 697-704. Puppi, L 1994, Andrea Palladio: das Gesamtwerk, DVA, Stutt-
Lorenz, WE 2011, ‘Fractal Geometry of Architecture: Fractal gart.
Dimension as a Connection Between Fractal Geometry Salingaros, NA 2006, A theory of architecture, Umbau-Verlag,
and Architecture’ in P Gruber, D Bruckner, C Hellmich, Solingen.

Society for the Diffusion of Useful Knowledge 1840, The Wundram, M, Palladio, A, Pape, T and Marton, P 2004, An-
Penny Cyclopædia of the Society for the Diffusion of Use- drea Palladio, 1508-1580: architect between the Renais-
ful Knowledge, Organ - Pertinax. Vol. 17, Knight, London. sance and Baroque, Benedikt Taschen, Cologne.
Vaughan, J and Ostwald, MJ 2010, ‘Refining a computa- Zarnowiecka, JC 1998, ‘Chaos, databases and fractal di-
tional fractal method of analysis’, Proceedings of of the mension of regional architecture’, Proceedings of the
15th International Conference on Computer-Aided Ar- eCAADe Conference, Paris, France, pp. 267-270.
chitectural Design Research in Asia CAADRIA, pp 29-38.

The Rehabilitation Design Process of the Bourgeois
House of Oporto: Shape Grammar Simplification
Eugénio Coimbra1, Luís Romão2
1
University of Minho, Portugal, 2Technical University of Lisbon, Portugal
2
http://home.fa.utl.pt/~lromao/2012_13/index_20122013.html
1
eugeniocoimbra@iol.pt, 2lromao@fa.utl.pt
Abstract. This study was accomplished in the context of a broader research to be

developed in an ongoing PhD program in architecture. The purpose of this study is to
give a perspective of the research progress and to present a shape grammar simplification
that will be improved to assist the rehabilitation design process of the bourgeois house of
Oporto.
The typology of the bourgeois house of Oporto, built from the late sixteenth century until
the early twentieth century, is dominant in the ancient fabric of the city and in need of
rehabilitation. From the analysis of a representative sample of a moment of its evolution,
it is possible to verify patterns and to define rules.
This first approach intends to validate the use of shape grammars as a tool, able to assist
the architect in the rehabilitation design process of the bourgeois house of Oporto.
Keywords. Design process; rehabilitation; shape grammars.
INTRODUCTION
The overall goal of the ongoing PhD is the develop- According to the Management Plan for the
ment of a tool able to assist the architect in the reha- Historical Centre of Oporto, a strategic document
bilitation design process of the bourgeois house of created in 2010, a requirement of UNESCO, when it
Oporto, Portugal. The research described in this pa- revised its classification program of world heritage
per start this with the presentation of a shape gram- sites, the historical center consists of 1,796 build-
mar simplification focused only on the topology of ings, 443 in good condition, 649 in average condi-
rehabilitated or in rehabilitation buildings. tion, 575 in poor condition and 78 in ruins, with 51
The old center of Oporto should be preserved being works in progress. The dominant function is
not only for the knowledge and symbolism present housing, constituting 80% of the buildings (Loza, et
in its built historic heritage, but also for its intrinsic al. 2010). The bourgeois house is the building type
material and economic values. that predominates in this territory.
In this city, the Porto Vivo, SRU - Society for Ur- In the critical success factors in the report of the
ban Rehabilitation has recently been created and its 2010 activities of the SRU, we can identify as weak-
mission is to lead the process of urban regeneration. nesses the extent of the territory and the complexity
This institution has replaced CRUARB (commission of the task and as strengths the experience, knowl-
for the urban renewal of Ribeira/Barredo) restructur- edge and results.
ing their political action. The report of the 2011 activities emphasizes the

experience gained during its seven years of exist- old heart of the city and in the streets that radiated
ence. The report of 2012 given the economic context from him, continued the local tradition, assimilating
and the restrictions on access to bank credit esti- successive styles and techniques (Oliveira and Gal-
mates more constraints to the rehabilitation process. hano, 1992).
At present the urban regeneration in the city of In Oporto there is an apparent lack of uniform-
Oporto is seen as an objective in the medium, long ity in their houses, different shapes, sizes and colors
term. mark the first impression you may have.
Architecture is an open work and should not A closer look identifies two fundamental types.
be regarded as an achievement but as a process, it The high narrow house with three or four floors,
is never finished and is continuously used (Vieira de sometimes reaching the five floors with two or three
Almeida, 2008). openings was in origin a hybrid type. Congregating
The bourgeois house of Oporto has been the residence with commercial activity, belonged to the
subject of some interventions which resulted in the bourgeoisie, had stores, warehouses and workshops
accumulation of a niche of experience. In the vast on the ground floor and housing on upper floors (Fig-
work still to be done it is important to understand ure 2). Among them, rarer, large and low houses com-
how this experience can be used efficiently. posed mainly of ground-floor and main floor with
With the intention of supporting the rehabilita- numerous façade doors and windows was the noble
tion design process of this buildings, a shape gram- house with large spaces as a statement of prestige
mar simplification was developed based on the in- and power in the city (Oliveira and Galhano, 1992).
formation extracted from a sample of rehabilitated The two fronts bourgeois house set in a Gothic-
or in rehabilitation buildings. This study validates mercantile lot has taken origin in the duplication of
the use of shape grammars in the analysis of the de- smaller lots of only one front houses. Those houses
sign solutions used for this buildings and in the crea- of the oldest part of the city, raised in narrow and
tion of new solutions in the same language. deep lots, adherent to the existing relief, are an ur-
ban and architectural fact immediately associated to
THE BOURGEOIS HOUSE OF OPORTO Oporto (Barata Fernandes, 1996).
The house is an elementary part in the conforma- These houses persisted in local tradition and
tion of the streets and in the fabric of the city as a followed the evolution of the city from the late six-
whole. In Oporto the high narrow house (Figure 1), teenth to mid-nineteenth century and kept some of
originated from the old borough, prominent in the their basic features to the early twentieth century.
Figure 1
Street Clérigos (author’s
photograph).
Figure 2
Street 31 de Janeiro (author’s
photograph).

This type of house is not exclusive from Oporto The analytical studies are based in a set of exist-
and it is possible to establish relations with other ing designs that represent the language - the corpus
European cities such as Genoa, Bologna, Pavia and - used to infer the rules of the grammar, which is
Florence or even cities of northern Europe (Barata then tested by using the rules to generate designs in
Fernandes, 1996). the corpus, as well as new designs in the language.
The bourgeois house of Oporto for its domi- George Stiny (1981; 1992) has also shown that
nance and symbolism presents itself as a key ele- shapes, labels, and weights can be combined to
ment in the rehabilitation of the built historical form shape grammars that encode specific languag-
heritage of the city. Over time it becomes important es of designs.
to reevaluate their internal organization and uses, The shape grammar simplification that will be
materials and building construction system, in a presented follow the steps of the analytical studies
gesture to rescue the past and strengthen its basic and is based on the Malagueira simplification gram-
features. mar of José P. Duarte (2004). It differs from this one
because it not only aims to introduce a more com-
THE SHAPE GRAMMARS plex grammar but also serves to explore, in a flex-
The first publication of shape grammars goes back ible way, different possibilities of developing the
to 1972 and had as authors George Stiny and James grammar for the rehabilitated or in rehabilitation
Gips (1972). In an early stage, it was applied in the buildings of the bourgeois house of Oporto. Unlike
interpretation and evaluation of pictorial works. Lat- the Malagueira simplification grammar, the lots and
er in 1980, with the publication of the paper “Kinder- the buildings have different dimensions and uses
garten grammars: designing with Froebel’s building weights in a different representation to highlight fu-
gifts”, George Stiny (1980) presents a grammar de- ture intentions.
fined in the three dimensional space that was the
initiator of the architectural grammars that followed. THE CORPUS OF DESIGNS
Shape grammars can be defined as algorithmic The theoretical support for this research consists in
systems for creating and understanding designs the study of a moment in the evolution of the bour-
directly by computing shapes, instead of text or geois house of Oporto since the late sixteenth cen-
symbols. A shape grammar is a set of rules that are tury to the early twentieth century.
applied step-by-step to generate a language of de- The bourgeois house of Oporto has been the
signs [1]. subject of some interventions in the historic center
Shape grammars are descriptive, analytic and and in parts of downtown. The collection of informa-
generative: descriptive because they explain the tion led to the identification of five different house
formal structure of the designs that are generated, designs that constituted the corpus for the grammar.
analytical because they can be used to tell whether This collection was made taking into account
a new design is in the same language, and genera- some examples of rehabilitated or in rehabilitation
tive because they can be used to create new designs nineteen century buildings that in their origin were
in the language (Stiny & Mitchell, 1978). from the hybrid type, congregating residence with
In 1976, George Stiny demonstrated that shape commercial activity, with warehouses and work-
grammars can be original or analytical when ap- shops on the ground floor and housing on upper
plied to the creation of new design languages or floors.
in the study of existing ones (1976). According to Two fronts buildings, raised in narrow and deep
Terry Knight [1] this gesture was the basis for new lots, with four or five floors, three openings and
approaches and an enhancer of its use in education yards with some variations. The central staircase was
and practice. structuring in the internal organization, the ground

floor had a separated entering from the rest of the includes several steps, as locating functional zones,
building. The functional organization of the housing locating the staircase and dividing functional zones
part of these buildings, in their origin, was not spe- into rooms.
cialized, with exception of the living room located in This simplification is composed by rules for the
the front of the first floor and the kitchen that it was manipulation of simple geometries representing
always placed in the back of the last floor near the rooms by dissecting, connecting, extending and add-
roof (Barata Fernandes, 1996). ing new shapes to them, as well as rules for assigning
The buildings analyzed have been adapted over and changing functions associated with them.
time to the needs of its dwellers, but they kept same A very simplified set of rules will be presented
of their basic features to the present day. only with the two dimensional information, where
Over time, the bourgeois house of Oporto was lines represent walls and shaded areas and labels
able to adapt to new circumstances and techniques represent the functions associated to the different
and became prevalent in the city (Oliveira and Gal- rooms. The functions of the rooms that the shapes
hano, 1992). represent are indicated by the labels fn (n = 1, 2).
The five rehabilitation solutions studied, divide The identification of the last line placed and the in-
the buildings in small apartments that are organ- dication on which side the next dissection may oc-
ized respecting the location of the central staircase cur is made by the label dot (•): on both sides (rules
in a logic that separates the front from the back of A, C and E) or only in one side (rules B, D and F). In
the buildings. The ground floor with is own entering rules A and B, dissections are perpendicular to the
continued, in some cases, the commercial activity. bigger edge of the rectangle. In rules C and D dis-
With the study of the rehabilitation designs of sections are perpendicular to the smaller edge of
this buildings it was possible to verify patterns and the rectangle. Rules E and F add a new rectangle.
to define rules in the definition of a grammar fo- Rule G deletes the label •, preventing further dissec-
cused on the topology of the bourgeois house of tions. Rules H, I, J, K and L concatenate two adjacent
Oporto buildings in the corpus. shapes to form a larger room. Rule M assigns a func-
tion to a room. Finally, rule N subtracts two lines of
THE SHAPE GRAMMAR SIMPLIFICATION the rectangle (Figure 3). The generation of basic lay-
The bourgeois house of Oporto shape grammar sim- outs, obtaining different patterns, with these rules
plification combines shapes, labels, and weights to comprises three steps. In the first step, the lot is
encode different ways of seeing and describing de- divided into different functional zones - yard, work-
signs. ing, service, circulation, living and sleeping or living/
This grammar is defined in the Cartesian product sleeping. In the second step, the vertical circulations
of algebras U12 V02 and U22 V02 in the two-dimen- are located. In the third step the functional zones are
sional representations of the different floor plans. divided into rooms to obtain the final layout.
The decisions made in the organization of the A diagram in the form of a tree, in which it is
lower floors can condition the generation of the possible to recognize the basic patterns behind the
upper floors, this dependency is encoded into the houses in the corpus is shown in Figure 4.
grammar through the use of sequential, parallel This diagram is composed by nodes represent-
grammars, one for each floor. ing the state of the design and by arcs representing
The derivation of a design in the grammar goes the application of rules in the definition of the func-
through several successive stages defining each tional organization of the first floor.
floor. When the generation of a lower floor finishes, A case study will be presented showing the sev-
the state changes, thereby activating the genera- eral steps referred above in the first floor derivation
tion of the upper floor. Each of these stages, in turn, of one building in the corpus.

Figure 3
Shape grammar simplifica-
tion rules.

Figure 4
Tree diagram showing the
definition of the functional
organization of the first floor.
I - inside; O - outside; lv - living
zone; sl - sleeping zone; ls - liv-
ing / sleeping zone; ya - yard;
ci- circulation zone; se - service
zone.

CASE STUDY ond floor.
The example of an Oporto nineteen century build-
ing located in the Almada Street is presented (Fig- CONCLUSIONS
ures 5, 6 and 7). The author of the rehabilitation de- The purpose of this paper has been to present the
sign is the well-known Souto de Moura. first approach to the shape grammar for the rehabili-
The existent layout and the design proposed by tation of the bourgeois house of Oporto, explaining
the architect for this floor are shown with the first the reasoning behind this work in progress.
floor derivation in Figure 8. This floor is organized re- With the intent of supporting the rehabilitation
specting the location of the central staircase with an design process of the bourgeois house of Oporto,
apartment in the front and other in the back of the being aware of the actual rehabilitation context of
building. Secondary vertical circulations are added. Oporto and based on “best practice” procedures in
The back apartment also occupies a part of the sec- heritage conservation, this study not only reintro-
Figure 5
First floor existent layout.
Figure 6
First floor proposal layout.

Figure 7
Second floor proposal layout.
duce the discussion of the design process in archi- REFERENCES

tecture but also serves to explore a more efficient Barata Fernandes, F (ed) 1996, Transformação e Permanên-
way to assist the architect in the rehabilitation of cia na Habitação Portuense – As formas da casa na for-
these buildings. ma da cidade, Transformation and Remaining in Oporto
This shape grammar simplification was based housing, the shapes of the house in the shape of the
in the derivation of five rehabilitation proposals town, FAUP, Oporto.
and was tested in two new designs in the language. Duarte, JP 2004, ‘Towards the mass customization of hous-
The derivations were useful in the definition of the ing: the grammar of Siza’s houses at Malagueira’, Envi-
rules and in the viewing of the shape grammar fu- ronment and Planning B: Planning and Design, 32, pp.
ture development. The new designs served also to 347–380.
update the grammar that is constantly evolving. This Loza, R et al. 2010, Management plan, Historic center of
approach proved to be proficient in the study of the Oporto world heritage, CMP, Oporto.
buildings topology. Oliveira, VE and Galhano F (ed) 1992, Arquitectura Tradi-
A more complex shape grammar is now being cional Portuguesa, Traditional Portuguese Architecture,
developed. It will be parametric and will encode in- D. Quixote, Lisbon.
formation on dimensioning, on function and on the Stiny, G 1992, ‘Weights’, Environment and Planning B: Plan-
building system. ning and Design, 19, pp. 413–430.
This information can be useful to establish the Stiny, G 1981, ‘A note on the description of designs’, Environ-
link with BIM (Building Information Modeling). In ment and Planning B, 8, pp. 257–267.
this approach BIM represents the overall method Stiny, G 1980, ‘Introduction to shape and shape grammars’,
of handling building information, and not the com- Environment and Planning B, 7, pp. 343–351.
puter implementations. The possible combination Stiny, G 1980, ‘Kindergarten grammars: designing with
of shape grammars and BIM is being considered. Froebel’s building gifts’, Environment and Planning B, 7,
A shape grammar for the rehabilitation of the pp. 409–462.
bourgeois house of Oporto has been initiated, carv- Stiny, G 1976, ‘Two exercises in formal composition ’, Envi-
ing a path for the future development of a new ap- ronment and Planning B, 3, pp. 187–210.
proach to the rehabilitation design process of these Stiny, G and Gips, J 1972 ‘Shape Grammars and the Genera-
buildings. tive Specification of Painting and Sculpture’, Freiman

Figure 8
Derivation of the first floor
functional organization.
bl - balcony; ba - bathroom;
cl - closet; ki - kitchen; la -
laundry; lf - lift; li - living room;
ls - living sleeping zone; st
- staircase; wc - water closet,
ya - yard.
CV Information Processing 71, Amesterdam, pp. 1460– Vieira de Almeida, P (ed) 2008, Apontamentos para uma
1465. Republished in Petrocelli OR 1972 The Best Com- Teoria da Arquitetura, Notes for a Theory of Architecture,
puter Papers of 1971, Philadelphia, pp. 125–135. Horizonte, Lisbon.
Stiny, G and Mitchel, WJ 1978, ‘The Palladian grammar mod-
els’, Environment and Planning B, 5, pp. 5–18. [1] www.mit.edu/~tknight/IJDC/.

Albertian Grammatical Transformations
From the treatise to the built work in the design of sacred buildings
Bruno Figueiredo1, José Pinto Duarte2, Mário Krüger3
1
School of Architecture, University of Minho, Portugal, 2CIAUD, Faculdade de Arcqui-
tectura, Universidade Técnica de Lisboa, Portugal, 3Department of Architecture, Centre
for Social Studies, University of Coimbra, Portugal
1
http://www.arquitectura.uminho.pt/3751.page-pt, 2http://home.fa.utl.pt/~jduarte/ 3http://
woc.uc.pt/darq/person/ppgeral.do?idpessoa=1
1
bfigueiredo@arquitectura.uminho.pt, 2 jduarte@fa.utl.pt, 3 kruger@ci.uc.pt
Abstract. This paper presents a research on the use of shape grammars as an analytical
tool in the history of architecture. It evolves within a broader project called Digital
Alberti, whose goal is to determine the influence of De re aedificatoria treatise on
Portuguese Renaissance architecture, making use of a computational framework (Krüger
et al., 2011).
Previous work was concerned with the development of a shape grammar for generating
sacred buildings according to the rules textually described in the treatise. This work
describes the transformation of the treatise grammar into another grammar that can also
account for the generation of Alberti’s built work.
Keywords. Shape grammars; parametric modelling; generative design; Alberti; classical
architecture.
INTRODUCTION
The research described in this paper is part of a in deriving solutions in the same language. Howev-
larger project called Digital Alberti, whose aim is to er, certain features of Portuguese classical churches
determine the influence of Alberti’s treatise De re ae- are not identifiable in such solutions and, therefore,
dificatoria on Portuguese Renaissance architecture, its source of inspiration remains uncertain.
making use of a computational framework (Krüger Several scholars in the history of Portuguese
et al., 2011). Renaissance architecture report that Portuguese
This paper analysis the task of achieving a shape royal house contracted Italian architects and pro-
grammar that can contribute for clarifying the influ- moted the visit of Portuguese architects to Italy dur-
ence of Alberti’s work on Portuguese architecture of ing the 15th and 16th century. (Moreira 1991, 1995;
the counter-reform period. Previous work was con- Soromenho, 1995; Branco, 2008) This fact may have
cerned with the translation of De re aedificatoria’s caused architects who worked in Portugal during
descriptions of sacred buildings into a generative that period to contact with Alberti’s buildings that
shape grammar. (Duarte et al., 2011; Figueiredo et were erected in the late 15th century.
al., 2013) This grammar has shown to be successful This fact led us to consider the transformation

of the shape grammar (Knight, 1983) for generat- 1460, designed in a Greek cross plan, in consonance
ing sacred buildings according to the rules textually with Antonio Labacco drawings (Tavernor, 1996,
described in the treatise, into another grammar that p.128), to which Alberti planned the construction of
could account for the generation of his built work, a dome in the central space, instead of the existing
namely, its morphological and proportional fea- coved vaults; the external walls of the church of San
tures. Francesco, known as the Temple Malatestiano in Ri-
This paper presents the methodology used to mini, begun in 1453, unfinished, and mainly rebuilt
transform the initial treatise grammar into a gram- after being severely damaged during World War II;
mar that can unveil the origin of certain features of and finally, the facade of Santa Maria Novella in Flor-
Portuguese Renaissance architecture. ence (1458–70) which resulted from a commission
of the Rucellai family.
METHODOLOGY The sources of the drawings used in this task
A previous grammar, directly inferred from the read- were the photogrammetric surveys done by the
ing of De re aedificatoria, was considered for this re- Olivetti Group [1] for the exhibition held in Palazzo
search (Duarte et al., 2011). Their rules are mainly de- Te, Mantua in 1994. These drawings were chosen be-
scribed on Chapters IV and V of Book 7 – Ornament cause they do not include later modifications in the
to Sacred Buildings, where the treatise expresses in buildings layout and architectural details, which can
algorithmic terms the knowledge base for the de- then be considered more loyal to Alberti’s design
sign of sacred buildings – temples. intentions. The first step in accomplishing this task
In accordance with the objectives described in was to collect data identifying Alberti’s contribu-
the introduction above, the approach followed in tions for the design of each of the buildings (Borsi,
this research included four steps: (1) to analyze the 1989; Tavernor, 1998; Rykwert and Angel, 1994).
most representative sacred buildings by Alberti, Following to this, drawings of the buildings
with the aim of identifying morphological and pro- were analyzed with the aim of identifying morpho-
portional features that were not encoded by the logical and proportional features that have not been
treatise grammar; and subsequently synthesizing considered in the treatise grammar. Two comple-
that information in parametric schemas; (2) to in- mentary analysis were performed.
troduce the knowledge encoded by the parametric The first analysis was to fill in a survey in which
schemas in the grammar, by changing existing rules entries collect the buildings’ features taking in ac-
or adding new ones; (3) to determine the relation count the parts of sacred buildings described in the
among these rules, the grammar’s recursive struc- treatise grammar. This information was registered
ture, and the process of derivation solutions in the on tables gathering the parameters, conditions and
language; and (4) to translate the grammar’s prin- spatial relations translated from both the treatise
ciples into a parametric computational model that and the buildings thereby allowing to identify simili-
allowed one to evaluate the generative outcome of tudes and deviances between them.
the grammar in a different generative paradigm. The second analysis was to draw schemas that
were useful for synthesizing the buildings’ propor-
FROM ALBERTI BUILDINGS TO GRAM- tional principles, and to identify morphological fea-
MAR TRANSFORMATION tures that were absent in the treatise’s descriptions.
The first task in this research was focused on the Due to the space restrictions, this article
analysis of Alberti’s designs of sacred buildings. focuses on the analysis of Sant’Andrea’s plan. The
Namely, the church of Sant’Andrea in Mantua, re- result of this analysis is synthesized in the Table 1,
built according to Alberti’s 1470 design for Ludovico which synthesizes a survey comparing Sant’Andrea’s
Gonzaga; the church of San Sebastiano in Mantua, features with those described in the treatise and in

Table 1
Excerpt from a table sum-
marizing the analysis of the
morphological features and
proportions of Sant’Andrea’s
plan. Green rectangles mean
that the feature is contem-
plated in the grammar, while
red rectangles mean that it is
new to the grammar.
Figure 1, which diagrams Sant’Andrea’s plan propor- length (Li) is directly dependent of cell width (Wi):
tional schema. Both analyses revealed three main Li = α Wi ; α ∈ {1, 1 1/3, 1 1/2, 2}.
aspects that differentiate Sant’Andrea’s plan from Sant’Andrea’s Li dimension corresponds to 3Wi,
the treatise grammar generative outcome: (1) cell resulting in a 3:1 proportion. Although this propor-
proportions; (2) the relative proportions between tion does not comply with the descriptions in Book
the lateral chapels’ openings and the skeleton be- 7, it is foreseen in the proportions described by Al-
tween them; (3) the rooms that fill space between berti in Chapters V and VI of Book 9 - Ornament on
lateral chapels. Both the analysis and the subse- Private Buildings: “…The method of defining the
quent shape rules implications are described below. outline is best taken from those objects in which Na-
ture offers herself to our inspection and admiration
(1) cell proportions as we view and examine them. […] The very same
In Book 7, Chapter IV, paragraph two, Alberti de- numbers that cause sounds to have that concinnitas,
scribes the principles for defining the proportion of pleasing to the ears, can also fill the eyes and mind
cells in rectangular temples. The rule of the treatise with wondrous delight. From musicians therefore
grammar considered these proportions, where cell who have already examined such numbers thor-

Figure 1
Sant’Andrea’s plan summariz-
ing the analysis of cell propor-
tions; the skeleton between
lateral chapels proportions;
the rooms that fill space
between chapels.
oughly, or from those objects in which Nature as dis- generate a temple with the length Li of Sant’Andrea
played some evident and noble quality, the whole (Figure 2 left), and with the further integration of the
method of outlining is derived. […]” correspondentiae inatae in set of conditions, further
On Chapter VI, Alberti refers to and describes in solutions can be achieved by the application of Rule
detail, the use of musical consonances to determine 1:
cell proportions. In synthesis, he defines that the Li = α Wi ; α ∈ {1, 1 1/3, 1 1/2, 2, 2 1/4, 1 7/9 ,3, 2 2/3,
proportions may be either short, long, or intermedi- 4, √2/√1, √3/√2, √3/√1, √4/√3}.
ate: as short proportions he considers Square (1:1),
Sesquialtera (3:2) and Sesquitertia (4:3); as intermedi- (2) the proportion of the skeleton between
ate proportions Double (1:2), Duplicate Sesquialtera lateral chapels
(9:4) and Duplicate Sesquitertia (16:9); and finally, as The proportional relation between lateral chapels
long proportions Triple (3:1), Double Sesquitertia (8:3) openings (Wcl) and the walls separating the vari-
and Quadruplus (4:1). ous chapels (Ws) is described on Chapter IV, Book
In the same chapter, Alberti describes that con- 7, between paragraphs 4 and 7: “… the bones, that
cinnitas is reached by the use of musical consonanc- is, of the building, which separate the various open-
es, but he also considers the use of correspondentiae ings to the tribunals in the temple - be nowhere less
inatae to establish “certain natural relationships that than a fifth of the gap, nowhere more than a third,
cannot be defined as numbers, but that may be ob- or, where you want it particularly enclosed, no more
tained through roots and powers.” Further reading of than a half.”
this chapter enabled the inference of correspondenc- These parameters and conditions were synthe-
es between certain ratios – (√2:√1), (√3:√2), (√3:√1), sized in the Rule 4 of the treatise grammar by the
(√4:√3) – that can be used to define proportions. equation:
By incorporating the musical consonances in the Ws = φ’ Wcl; 1/5 ≤ φ’ ≤ 1/3 ∨ φ ‘= 1/2.
initial conditions of Rule 1, the grammar will able to The Ws dimension is also dependent on cell

Figure 2
Shape rules from the rec-
tangular temples grammar,
with the included parameters
and conditions: (left) Rule 1
defines cell proportions; (right)
Rule 4b defines the lateral
chapels’ openings andplaces
labels to design the chapel’s
outline using the set of Rules 5.
length Li, which is equal to the sum of the lateral rear facade form a room connected to the cell that-
openings, plus the width Ws between them the conforms a rectangular plan (Figure 1). This spatial
temple’s end walls, and it can be deduced by the fol- relation was not considered in the treatise shape
lowing function: grammar because it is not described in De Re Aedifi-
Ws = ( Li - Ncl Wcl ) / (Ncl + 1). catoria. While the addition of one single chapel per
Since at Sant’Andrea, the proportion Wcl:Ws cor- facade, as it happens in San Sebastiano, results in a
responds to √3:√2 (Table 1), it does not verify the relatively evident spatial relation between lateral
conditions specified for φ’ in the initial rule. In a strict chapels and the cell’s wall, when several chapels are
understanding of the principles laid out in Book added to the same facade, such a spatial relation can
VII, such a non-correspondence could have been be configured in several ways. The set of Rules 7 (Fig-
considered as an error in the Albertian canon. How- ure 3 center) show the spatial relations translated
ever, several authors (Tavernor, 1985; Kruger, 2011) from the treatise, while Rule 7a’ and Rule 7b’ (Figure 3
showed that the use of the proportion √3:√2 to de- right) show the new spatial relations introduced by
sign the chapels’ openings and the skeleton could reproducing the ones existent in Sant’Andrea.
be considered Albertian by introducing the use of
correspondentiae inatae in the definition of such a Sant’Andrea grammar add-ons
proportion. The subsequently inclusion of such cor- According to Terry Knight (1983), to transform a
respondences in the set of conditions in the original shape grammar, at least one rule addition, rule de-
Rule 4 (Figure 2 right) results in: letion or rule change has to be performed. By tak-
Ws = φ’ Wcl; 1/5 ≤ φ’ ≤ 1/3 ∨ φ‘ ∈ {√2/√1, √3/√2, ing into consideration her definition of rule change:
√3/√1, √4/√3}. “Rule change changes a rule, initial shape, or final
state by changing any of its spatial or nonspatial
(3) rooms filling space between chapels, components: spatial relations, spatial labels, or state
frontispiece and rear facade. labels.” - the operations performed to Rule 1 and
In Sant’Andrea, the spaces in between the row of lat- Rule 4 can be considered a rule change because they
eral chapels and the edges of the frontispiece and add new dimensional conditions to the initial ones.

Figure 3
Shape rules that define the
addition and arrangement
of walls in lateral rectangular
chapels. The left rules repre-
sent the rules added to the set
of Rules 7, in accordance with
the spatial relation inferred
from Sant’Andrea.
Despite the maintenance of the parametric schema, their interrelations, which have been applied to de-
new spatial relations can be achieved by resizing fine the recursive structure of the treatise grammar.
the plan. The addition of a new rule, as in the opera- Since this structure encapsulates the formal and par-
tion described above for the addition of Rule 7a’ and ametric logic of Alberti’s buildings, it was decided to
Rule 7b’, can also be considered a transformation of maintain the core of their recursive structure during
grammar. the transformation process. Although the recursive
structure of the grammar was kept, several rules
SHAPE GRAMMAR TRANSFORMATION were transformed by changing their spatial rela-
WITH A CONSTANT RECURSIVE STRUC- tions, and other rules were added.
TURE Figure 4 shows a step by step computation, il-
The treatise grammar followed mimetically the or- lustrating the different options of derivation at each
der of description of the temples’ parts in the trea- step, where only one derivation is subsequently
tise. Their morphology is mainly described on Chap- transformed by the use of the next set of shape
ters IV and V of Book 7, in which the constituent rules.
parts of the temples are treated: cell – inner space
of the temple, defined by the geometry of their area; THE CLASSIC NUMBER VERSUS THE
tribune; lateral chapels and their skeletons; portico CONTEMPORARY PARAMETRIC MODEL
informed by the column systems – shaft, base, capi-
tal and entablature – and their proportions; pedi- The ‘number’ in the algorithmic nature of
ment; walls; roof; and main openings. De re aedificatoria translated by contem-
While in Palladian Villas grammar (Stiny, 1978) porary eyes
the Villas constant partition features were useful to In classical philosophy, numbers have a specific
define the grammar recursive structure, in Alberti’s meaning before its scientific dimension. Alberti sys-
built work, the few examples of designs of sacred tematizes classic architectonic dimensions through
buildings, and the typological variety of those ex- considerations on the perfection of numbers, as well
amples do not seem to be the more appropriate as establishing relationships between music harmo-
for setting up a grammar representative of Alberti’s nies and proportional systems in architecture (Book
sacred buildings. From the reading of the treatise, it 9, V).
was possible to consider a framework for the defini- In Nexus 2002 conference, during a round table
tion of the morphological parts of the temples and discussion about the significance of both the quan-

Figure 4
Computation tree of the tem-
ples shape grammar showing
only one possible subsequent
rule application at each step
of the computation.
tity and the quality of numbers in De re aedificato- portional and morphological dependency. Thus, the
ria, Lionel March in answering to Robert Tavernor’s schema presented in Figure 1 and the shape rules il-
question [2] - “Can anyone explain exactly what lustrated in Figure 2 feature the possibility of assign-
might be meant by the “quality” of a number?” - ar- ing different values to their dimensional parameters
gued for the numbers dual nature in the treatise: and also by the interdependencies between the cell
“When Alberti was writing, the words ‘quantity’ and and chapels dimensions and location, resulting in
‘quality’ still retained their Aristotelian roots. (…) a parametric shape rule. This kind of relations is re-
Thus, from an Aristotelian perspective, in giving peated in other shape rules resulting in a parametric
shape to an architectural work, Alberti is engaged in grammar (Stiny, 1980).
qualitative decisions, but in dimensioning the work Like the initial grammar, the transformed gram-
he is acting quantitatively. (…) A pediment is quali- mar still is a parametric grammar. Therefore, each
tatively ‘triangular’, but its dimensions are quantita- derivation of the grammar can potentially generate
tively 24 feet long to 5 feet high.” a family of design solutions, rather than one single
March’s argument in the discussion of number solution. A computational parametric model was
significance follows to the idea that “a contemporary developed in Grasshopper with the aim of managing
approach would be computational with respect to the generation of multiple design solutions within
‘number’ and semiotic with respect to reference and the grammar (Grasshopper is a Visual Programming
usage. Interface that interacts with modeling software Rhi-
The treatise grammar inference and their subse- noceros. A program written in Grasshopper consists
quent transformation followed this notion of work- of a combination of interlinked components per-
ing simultaneously with a ’contemporary’ under- forming operations on primitives, usually but not
standing of ’shapes’ and ‘numbers’. ’Shapes’configure necessarily geometrical ones. This programming
the essence of the spatial relations of shape rules, paradigm allows visually developing parametric
while ‘numbers’ introduce their dynamic dimension- geometrical models, whose outputs correspond to a
al significance. family of solutions). The parametric model encodes
the knowledge gathered in the grammar inferring
De re aedificatoria a pre-digital parametric and transformation processes. The output depends
model on the variation of parameters, which correspond to
The process of inferring shape rules directly from the what Alberti prescribes for the number and dimen-
reading of De re aedificatoria exposed the algorith- sions of the elements that should, according to the
mic nature of their content. Alberti notations on the author’s theory and practice, conform the temple
sacred buildings parts are described in terms of nu- (Figure 5).
merical qualities and quantities defining their pro- In the last three decades, computational tools

Figure 5
Parametric model and one
possible design solution99.
gained an extraordinary importance in the contem- I pass from the drawings to the model, I sometimes
porary architectural discourse. Parametric design notice further mistakes in the individual parts, even
is one of the computational models that acquired over the numbers”
more relevance in these process. Despite their im- Despite this incongruity, the analysis of the
portance, little discussion has been given to the use buildings contributed for the systematization of
of parametric design in a pre-digital era. The transla- a coherent body knowledge of Albertian sacred
tion of Alberti’s work into a shape grammar revealed buildings because our focus on the buildings was
that the De re aedificatoria’s descriptions of sacred constrained by our concern for the structure of the
buildings is precursor to the use of parametric de- treatise grammar.
sign to define a set of architectonic principles. Thus, The methodology presented for the inference
it is inevitable that a research on De re aedificatoria of transformations of the treatise shape grammar
today gives rise to its implementation as a compu- contributed for encoding new knowledge into the
tational model. grammar. Although, the algorithmic nature of the
treatise descriptions eased the task of matching
CONCLUSION building proportions and morphology with the
The variety of context and the role that Aberti had grammar shape rules, this reinforces the notion
in the design of his buildings results in very specific that inferring rules from the analysis of a corpus of
knowledge that can be retrieved from them. Thus, existing buildings is an adequate tool to reinforce a
the sole analysis of the buildings were not sufficient grammar’s capability for generating solutions in ac-
to set up rules defining a consistent architectural ty- cordance to both textual and design descriptions
pology. Furthermore, they do not always verify his (Mitchell, 1990).
treatise’s principles. Regarding to this subject, Tav- Both shape grammar and parametric model
ernor (1996, p. 178) remembers that Alberti (IX, 10, implementations prove to be effective tools for
p.137) made reference to the difficulty of translating generating design solutions in the same style. The
his theoretical principles in a successful design: “I former introduces a step by step computation that
can say this of myself: I have often conceived of pro- reinforces the visual perception of formal transfor-
jects in the mind that seemed quite commendable mations. The latter, by automating the process of
at the time; but when I translated them into draw- generation, emphasizes the variation on the solu-
ings, I found several errors in the very part that de- tions by controlling their parameters. Even though
light me most, and quite serious ones; again, when their structure has different philosophies, they used
I return to drawings, and measure the dimensions, I the same knowledge on the design, resulting in the
recognize and lament my carelessness; finally when same corpus of solutions.

Given the objectives of the project Digital Al- de Alberti na arquitectura portuguesa da contrarrefor-
berti, it is supposed to expand the methodology ma’ in CAL Brandão, P Caye, F Furlan and MA Loureiro,
presented to a set of sacred buildings, representa- Na Gênese das Racionalidades Modernas: Em torno de
tive of classical Portuguese architecture. The aim of Leon Battista Alberti, Editora UFMG, Belo Horizonte,
this analysis is to identify possible deviations and Brasil, pp. 488-548.
similitude between Alberti theoretical and design Figueiredo, B, Castro e Costa, E, Duarte, JP and Krüger, M
principles and classical Portuguese architecture. The 2013, ‘Digital Temples: a Shape Grammar ro Generate
results of these investigations it will be presented in sacred buildings according to Alberti’s theory‘ in JP
future essays. Sousa and JP Xavuir (eds.), Future Traditions - Rethink-
ing Traditions and Envisioning the Future in Architecture
ACKNOWLEDGEMENTS Through the Use of Digital Technologies, Faculdade de
This work is funded by FEDER Grants through Arquitectura da Universidade do Porto, Porto, Portugal
COMPETE – Programa Operacional Factores de , pp.63–70.
Competitividade and by National Grants through Li, AI 2002 ‘Algorithmic architecture in twelfth-century
FCT – Fundação para a Ciência e a Tecnologia, China: the yingzao fashi‘, in F Rodrigues and K Williams,
as part of the “Digital Alberti” project (PTDC/ (eds.), Nexus IV: architecture and mathematics, Kim
AUR-AQI/108274/2008 – FCOMP-01-0124-FED- Williams Books, Fuccechio, Italy, pp.141–150.
ER-008842). The project is hosted by CES at the March, L 1996, ‘Renaissance mathematics and architectural
University of Coimbra and ICIST at the Technical Uni- proportion in Alberti’s De re aedificatoria’, Architectural
versity of Lisbon, and coordinated by Mário Krüger. Research Quartely, vol.2, pp.54-65.
Bruno Figueiredo is funded by FCT with PhD grant Moreira, R 1991, ‘A Arquitectura do Renascimento no sul de
DFRH -SFRH/BD/69910/2010. Portugal. A encomenda régia entre o Moderno e o Ro-
The evolution of this work is in debt with Prof. mano’, Ph.D. thesis, FCSH-UNL, Lisbon.
Terry Knight due to her immensely helpful com- Moreira, R 1995 ‘Arquitectura: Renascimento e Classicismo’
ments. in P Pereira (ed.), História da Arte Portuguesa Vol II, Cir-
culo dos Leitores, Lisbon, pp. 302-375.
REFERENCES Krüger, M and Duarte, J P and Coutinho F 2011 ‘Decoding
Alberti, LB 2011, Da arte edificatória, AME Santo (transla- De Re Aedificatoria: using grammars to trace Alberti’s
tion), MJT Krüger (introduction and notes), Calouste influence on Portuguese classical architecture’ in K Wil-
Gulbenkian Foundation, Lisbon. liams, JN Tavares and JP Xavier (eds), Nexus Network
Borsi, F 1977, Leon Battista Alberti, Phaidon, Oxford. Journal, 13, no. 1, pp.171-182.
Branco, RL 2008, Italianismo e Contra-Reforma: a obra do ar- Knight, TW 1983, ‘Transformations of language of design’, in
quitecto Baltasar Álvares em Lisboa, FCSH-UNL, master Environment and Planning B: Planning and Design 10,
thesis, Lisbon. pp.125-177.
Coutinho, F, Castro e Costa, E, Duarte, JP and Krüger, M Rykwert, J, Leach, N and Tavernor, R 1988, On the Art of
2011, ‘A computational interpretation of ”De re aed- Building. De Re Aedificatoria, MIT Press, Cambridge,
ificatoria”: Translating Alberti’s column system into a Massachusets.
shape grammar‘ in RESPECTING FRAGILE PLACES - 29th Rykwert, J and Angel, A 1994, Leon Battista Alberti: Catal-
eCAADe Conference Proceedings, University of Ljublja- logo della mostra Palazzo Te, Olivetti and Electa, Milan.
na, Faculty of Architecture (Slovenia), pp.788–798. Soromenho, M 1995, ‘Do Escorial a São Vicente de Fora:
Duarte, JP 2008, Synthesis Lesson - Mass Customization: Mod- Algumas notas sobre Filipe II e a arquitectura portu-
els and Algorithms, Aggregation Exams, FAUTL, Lisbon. guesa’ in Monumentos 2, Direcção Geral dos Edifícios e
Duarte, J, Kruger M, Coutinho, F, Figueiredo, B and Castro e Monumentos Nacionais, Lisbon, pp. 24-26.
Costa, E 2013, ‘Alberti Digital: Investigando a influência Stiny, G and Gips, J 1972: ‘Shape Grammars and the Genera-

tive Specification of Painting and Sculpture’, Computer, [1] http://www.bath.ac.uk/casa/alberti/index.html
71, 125–135. [2] http://www.emis.de/journals/NNJ/Query03-Quality.
Stiny, G. 1980, ‘Introduction to Shape and Shape Grammars’, html
in Environment and Planning B: Planning and Design 7,
pp. 343-352.

A Parametric Recreation of Traditional Chinese Architecture
A case study on the floor plan

Di Li1, Michael Knight2, Andre Brown3
University of Liverpool, UK
1
di.li2@liv.ac.uk, 2mknight@liv.ac.uk, 3andygpb@liv.ac.uk
Abstract. This paper presents the current state of progress investigating the possibility
of modelling traditional Chinese architecture using parametrics based on the two rule
books. This builds on the work of producing systematic analysis on both rule books and
contributing knowledge from extant buildings. The case study target is the floor plan
described in Ying Zao Fa Shi. Discussion and future works are suggested at the end.
Keywords. Parametric modelling, traditional Chinese architecture, Ying Zao Fa Shi,
Kung-ch’eng tso-fa tse-le, floor plan.
INTRODUCTION
When studying traditional Chinese architecture, two before the generation of the floor plan models using
references are essential—literary records and extant Grasshopper and Rhino. The models are discussed
buildings. China, a country with over 5000 years of and evaluated and additionally, the comparison
history boasts remarkable architecture from all dy- with parallel research of the Shape Grammar ap-
nasties and periods. Unfortunately, almost none of proach to the floor plan is discussed.
the buildings before Tang Dynasty (618-907) remain
and many buildings from Song Dynasty (960-1279) THE TWO RULE BOOKS
to Ching Dynasty (1616-1912) have been badly Figure1 illustrates the chronological diagram indi-
damaged or destroyed. cating a brief history of China and the two dynasties
However, two important texts survive: Ying Zao (in the square boxes) in which the two rule books
Fa Shi (Building Standards) from Song Dynasty and were compiled.
Ching Dynasty: Kung-ch’eng tso-fa tse-le (Structural Ying Zao Fa Shi (Li, 1103) was the official building
Regulations) from Ching Dynasty, which are known standard as a guidance of design and construction
as the “two text books of Chinese ancient architec- in Song Dynasty. During the period of the Song Dy-
ture” (Liang, 1985). They are the only remaining clas- nasty, an increasing number of different levels and
sical Chinese literature which deals with architecture types of buildings were constructed which led to an
and are, in essence, rule books that govern most as- urgent requirement of an official instruction. There
pects of the design. As a starting point, the analysis were three original purposes of this book. First, to
on the two rule books is a key factor in understand- set the design guidelines to articulate the social
ing architecture of this period. This paper looks at status of feudalism. Second, to establish a unified
generation of the floor plan using the Ying Zao Fa architectural form and style to guarantee a consist-
Shi. A series of rules and hypotheses are reviewed ent level of detail and artistic effects. Third, to define

Figure 1
Chinese dynasties in chrono-
logic order.
the material choices and quantities as well as the even points out that the Hui Zong Emperor was a
work load to avoid corruption and embezzlement. naive politician, but was an excellent artist. Mean-
The first edition was published in 1091 and with ex- while Li Jie was also good at drawing and music. This
tended second edition compiled by Li Jie, the court might be one reason why occasionally Li Jie omitted
architect of the Hui Zong Emperor in 1103. some important descriptive rules but paid more at-
This book consists of thirty-four volumes. Vol- tention to the architectural style and decoration.
umes one and two are the overall introduction to Together with Li’s research (Li, 2001; 2003) and the
different types and components of the architecture. on-going research in the case study on the ting tang
Volume three is about the foundations, masonry section by the authors, it has been shown that Chi-
structures and carving of handrails. Volumes four nese traditional architecture has some parametric
and five introduce the structural carpentry system. characteristics.
Volumes six to eleven introduce the finished car- As shown in Figure 2, Ying Zao Fa Shi was writ-
pentry. Volume twelve includes three timber precast ten in an ancient form of the Chinese language
methods and bamboo weave method. Volume thir- which has no punctuation. The characters, vo-
teen explains tile and cement processing. Volume cabulary, grammar and text direction were all dif-
fourteen focuses on the composition and colour ferent from contemporary written Chinese which
matching of decorative painting. Volume fifteen de- presents a big problem to modern researchers. In
scribes the precast of bricks and ceramic materials. relative terms, Ching Dynasty: Kung-ch’eng tso-fa
Volumes sixteen to twenty-five presents the work tse-le (1734) is linguistically more acceptable since
load required in the previous volumes. Volumes it is compiled in 1734, more than six hundred years
twenty-six to twenty-eight outlines the material closer to us. Meanwhile, more extant buildings from
consumption of the components mentioned above. Ching Dynasty can be studied as practical evidence.
Volumes twenty-nine to thirty-four are the selected In this book, twenty-seven types of buildings with
diagrams. accurate size and dimensions are given as examples,
The significance of Ying Zao Fa Shi is not “simply making it useful for reconstruction of buildings of
for its existing” (Li, 2001). The book is, in general, well the period.
organised, logical, systematic and rigorous which
is quite rare in ancient literature. Although some RULES FOR THE FLOOR PLAN AND PARA-
aspects such as the floor plan are relatively lacking METRIC APPROACH
in systematic description, the whole book provides In order to understand and recreate floor plans, a
readers with a “rule-based and parametric” system description of the ancient floor plan system is neces-
(Li, 2001) for the ancient style buildings. Liang (1983) sary. In Ying Zao Fa Shi, the following factors or pa-

Figure 2
Sample pages of Ying Zao Fa
Shi (left: text description [1];
right: floor plan diagram [2]).
rameters can be used to describe a building: form the area of a building. This area can be divided
• The building type (such as dian tang or ting into small units (small rectangles) ie bays (usually,
tang, here tang means hall). each bay has four columns at the four corners, al-
• The overall dimension (measured in modular though not in every case). Each bay is determined
unit): by the bay width and bay depth, as shown in Figure
• Building width (and bay width) 3. The sum of the bay width or bay depth gives the
• Building depth (and rafters) building width or depth. But in reality, the bay depth
• The grade (which is used to calculate the abso- is not described in the set of parameters above. In-
lute value of the modular unit) stead, the horizontally projected rafter is used to
In the most common and formal cases, the floor measure the depth. There are three reasons. First,
plan of a single house is rectangular and consists Ying Zao Fa Shi mentions for ting tang type building,
of two major factors: building width and building one bay depth equals to two rafters deep but it does
depth, which determines the dimension and scale not mention the relationship for the dian tang type
of the house. The building width and building depth building. Therefore in order to unify the parameters
Figure 3
Factors of the floor plan.

Size of fen (in cun) 0.60 0.55 0.50 0.48 0.44 0.40 0.35 0.30 Table 1
Grades of dian tang 1 2 3 4 5 Absolute values of fen for dif-
Grades of ting tang 3 4 5 6 ferent grades.
Grades of other types 6 7 8
ting tang dian tang Table 2

Building width (number of bays) 3,5,7,9 bays 3,5,7,9,11 bays Definition of the four param-
Building depth (number of rafters) 4,6,8,10 rafters 2,4,6,8,10,12 rafters eters.
Bay width 200-450 fen
rafter ≤ 150 fen
in the later parametric modelling, the rafter is se- could still be built up by first making hypothesis
lected as the depth measurement for both building based on the information in hand and then evaluat-
types. Second, to be consistent with the research of ing with the diagrams in the rule book and extant
case study on the section, the rafter is a key param- building measurement data. The assumptions here
eter in defining the section. The rafter is closely relat- are based on the investigation of historian Chen
ed to the disposition of the columns and beams and (1993). As shown in Table 2, the four parameters are
the total number of columns. Third, from the work- summaried. In particular, the bay width is not given
ers’ experiences, they tend to use rafters rather than directly. The calculation is as follows: a bay has two
bay depth. Apart from the rectangular forms, there sets of dou gong (the bracket joint) that sit on each
are also several non-rectangular floor plans, known side of the columns (the black dots in Figure 3) and
as non-formal architecture, which includes the use either one or two sets between the columns (inter-
of the triangle, circle, sector, octagon, polygon, and columnar dou gong). The centre-to-centre distance
the superposition of polygons and Wan shape. They of dou gong is 125 fen ± 25 fen. Thus the bay width
are widely used in pavilions and gardens which typi- with one intercolumnar dou gong is 250 fen ± 50
cally appear in Southern China. But these irregular fen, and with two intercolumnar dou gong is 375
shape floor plans are not discussed in this paper. fen ± 75 fen. Therefore the total range of bay width
At this point, it is worth describing the measure- is 200-450 fen. In addition, the centre bay is often
ment units used. Depending on the eight grades of wider and in most extant examples the two outer
the buildings, a fen can have eight different absolute bays are often slightly narrower than the others [3].
values (Liang, 1983), measured in cun (a Chinese In the table, the modular unit fen is used rather than
length unit), as shown in Table 1. Given that 1 cun = the absolute values.
32mm approximately in Song Dynasty, the final ab- After all the parameters are clarified, the para-
solute values of width and depth can be obtained. metric model now can be built. Figure 4 shows the
For example, if the building is ting tang type and in logic diagram for the parametric modelling. In this
Grade Three, 1 fen =0.5 cun x 32mm/cun=16mm. logic diagram, ting tang and dian tang types are in-
In order to build up the parametric logic, there tegrated together since Liu (1984) points out that
are four more details which need to be clarified: the “although Ying Zao Fa Shi distinct the two types
value of bay width and rafter, and number of both strictly, buildings are slickly dealt with in practice”.
(which constitutes the building width and building The rectangular floor plan grid is set first by defin-
depth). Unfortunately, at this stage, Ying Zao Fa Shi ing the x-y plane as the base plane. The next step is
does not provide a systematic definition. Instead, it to define the size and number of bays. In order to
gives information partially by defining and partially achieve this, the four parameters described above
by enumerating. Despite this, the parametric model are outlined here. As the primary parameters, the

Figure 4
Parametric logic diagram.
rafter, number of bays and number of rafters can be in Ying Zao Fa Shi (Figure 2 right), they are high-
directly controlled by the corresponding number ly consistent in form. And there is one extant
range listed in Table 2. The bay width is a multiplica- building example—Fuoguang Temple Wenshu
tion of two factors: centre-to-centre distance of dou Dian (Figure 6), which built at 1137, located at
gong and number of intervals between dou gong. Shanxi Province. It is a Grade Two dian tang type
Additionally, the number of intervals between dou building with seven bays. From the paramet-
gong is equal to the number of intercolumnar dou ric model, the minimum building width is given
gong plus one. Since the number of intercolumnar as 7x200x0.55x32=24640mm=24.64m while the
dou gong is known directly, this is the fourth pa- maximum is 7x450x0.55x32=55440mm=55.44m.
rameter. Thus, overall, there are only four simple pa- Similarly, the building depth spans from 15.84m
rameters that can be controlled depending on the to 21.12m. Wang (2011) provides its measurement
building type and grade. In addition, there is one data of 31.56m in width and 17.60m in depth. As Liu
judgment in this logic diagram: the building width (1984) argues that “there is not such an extant build-
should always be larger than the building depth. ing completed follows Ying Zao Fa Shi found so far”,
And if so, the conclusion will appear true. Under this if the measurement data is within the range of the
one set of logic diagrams the floor plan of both ting parametric model, then the two are consistent.
tang and dian tang types, all eight grades of build-
ing with different bays and rafters are involved. Fig- DISCUSSION
ure 5 shows two examples of the model. Parametric design differs from the conventional de-
Comparing the examples with the diagrams sign mode of adding and removing marks in that

Figure 5
Examples (left: 9 bays 4 rafters
floor plan; right: 9 bays 6 raft-
ers floor plan).
the relationships between the parameters are the and a set of rules, and then the design rules act on
essence of parametrics. In this case study, the re- the initial symbols repeatedly, resulting in a final de-
lationships are not based on one specific example, sign. Following each typical set of rules will result in
but a systematic description and summary of all the one corresponding final design. Thus Shape Gram-
buildings in a typical period—which are the rules. mar generates a language of design. How and in
The logic diagram of the formal rectangular floor what sequence do the rules applied makes up the
plan is then built up based on the rules. Following so called grammar? Compared with Li’s research (Li,
this, different outcomes can be generated to indi- 2001), the advantage of the parametric method is
cate the advantages of parametric method which the ease with which the process can be extended
can result in different final products without a new into three dimensional modelling. For instance, the
set of logic diagrams or the removal/addition indi- intersectant points could be the column locations
vidual components. In particular, all the floor plan when combining with the case study on the section.
formats are included in this set of logic diagram, in- Then, the two dimensional representation of archi-
cluding both the building types (dian tang and ting tecture through the plan and section will form the
tang), any dimensions and all the grades. three dimensional parametric model. And indeed,
There is parallel research in the Shape Gram- according to Wang (2011), more special propor-
mar approach to the floor plan (Li, 2001). In the re- tions (relationships) exist in the elevations, as well
search, Li derives the process with initial symbols as many other building factors. On the other hand,
Figure 6
Fuoguang Temple Wenshu
Dian [4] [5].

if all the attempts are limited in two dimensional REFERENCES
planes, it may omit rules and relationships that exist Chen, M,1993, A study of structural carpentry in the Ying Zao
between the section and the floor plan. Fa Shi, Wenwu, Beijing.
Ching Dynasty, 1734, Kung-ch’eng tso-fa tse-le (Structural
CONCLUSION Regulations).
Accordingly, three major similarities between Shape Gips, J, 1975, Shape Grammars and their Uses, Birkhauser,
Grammar and parametrics can be drawn. Both of Basel.
them are derived systematically and logically; both Li, A, 2001, A shape grammar for teaching the architectural
of them are generative and productive; both of style of the Ying Zao Fa Shi, MIT, Cambridge, MA.
them can be symbolically and graphically illustrated. Li, A, ‘The Yingzao fashi in the information age’, The Beaux-
In contrast, three key differences can be identified Arts, Paul Philippe Cret, and Twentieth Century Architec-
as well. The parameters in parametric method can ture in China, The University of Pennsylvania, 3-5 Octo-
be any variables, abstract or concrete, without the ber 2003.
limitation of just geometrical entities. The rules used Li, J, 1103. Ying Zao Fa Shi (Building Standards).
in parametrics can be any logic relationships, not Liang, S, 1983, Ying Zao Fa Shi Zhu Shi (The annotated Ying
only repetitions and the final outcome could involve Zao Fa Shi), China Architectural Industry Press, Beijing.
many different variations. Liang, S, 1985, A pictorial history of Chinese architecture: a
When combined with parallel work in the para- study of the development of its structural system and the
metric generation of the ting tang section, it is evolution of its types, MIT Cambridge, MA, pp.1-16.
found that the characters summarized from deriving Liang, S, 2006, Analysis on Ching Dynasty: Kung-ch’eng tso-
the floor plans are consistent with those from gen- fa tse-le (Structural Regulations), Tsinghua University
erating sections. Taking Li’s work (Li, 2001) of shape Press, Beijing.
grammar approach to the floor plan and ting tang Liu., D, 1984. Chinese ancient architectural history, China Ar-
section into consideration, it can be concluded that chitectural Industry Press, Beijing.
Chinese traditional architecture has parametric char- Stiny, G, 1980, Introduction to Shape and Shape Grammars,
acteristics. Since the whole structure of traditional Environment and Planning B, 7.
buildings constructed using the rule books is com- Wang, G, Liu, C and Duan, Z, 2011, The research on propor-
plex and closely interrelated, a parametric method tion of Chinese ancient timber frame architecture, China
has the advantage of illustrating and generating Architectural Industry Press, Beijing.
the principles from the rule books to complex digi-
tal reconstructions. The application is not limited to [1] http://andrew.li/yzfs/yzfs_new/b2.htm (accessed 15
the restoration of ancient building, but could also be May 2013)
used as inspiration in the generation of new designs. [2] http://dl.zhishi.sina.com.cn/upload/61/88/61/
Apart from individual buildings, based on the floor 1146618861.9708083.jpg (accessed 15 May 2013)
plan logic diagram, a similar parametric method [3] http://andrew.li/yzfs/yzfs_new/b11.htm (accessed 15
can be used in recreation of many city plans, since May 2013)
several Chinese ancient cityies also show parametric [4] http://www.izy.cn/travel_photo/e30/155133.html (ac-
characteristics. cessed 15 May 2013)
[5] http://www.panoramio.com/photo/32278856 (accessed
15 May 2013)

A Bottom-Up Social Housing System Described with
Shape Grammars
The case of Belapur low-income housing development

Leticia Teixeira Mendes1, José Nuno Beirão2, José Pinto Duarte3, Gabriela Celani4
1,4
School of Civil Engineering Architecture and Urban Planning, State University of
Campinas - Brazil, 2,3Faculty of Architecture, TU Lisbon
1
leticia@fec.unicamp.br, 2jnb@fa.utl.pt, 3jduarte@fa.utl.pt, 4celani@fec.unicamp.br
Abstract. This paper presents the analysis of a bottom-up design system using shape
grammars. This research is part of a larger study that proposes the development of a
generic grammar to improve the quality of site development in social housing plans,
including the improvement of their public spaces. We show the use of shape grammars as
an analytical method to study the design of Belapur social housing development, designed
by Charles Correa, in 1983.
Keywords. Design methodology; shape grammar; analytical grammar; low-income
housing.
INTRODUCTION
This research aims at applying shape grammars as a up housing systems able to be implemented as in-
method for generating improved social housing de- cremental urban developments based on the pro-
sign systems which may contribute to the develop- gressive addition of housing clusters and associated
ment of more diversity in external areas and public community areas. The underlying hypothesis is that
spaces, creating identity and appropriation by its we can develop a set of generic parallel grammars
dwellers. In order to achieve such goals we start by which allow (and control) the generation of such
analyzing social housing plans as case studies to housing systems.
infer design patterns to propose the development
of generic grammars, which may enable the gen- OBJECTIVES
eration and management of incremental housing This paper is part of a larger study that proposes a
systems with locally captured spatial qualities. In development of a generic grammar to improve the
the book A Pattern Language (1977), Alexander and quality of low-income housing plans, including the
his collaborators define a theory and application in- improvement of public spaces and community ar-
structions for the use of a pattern language at dif- eas. The generic grammar is developed from analy-
ferent design scales - from the scale of the city and sis of four case studies by capturing the underlying
urban design to the building scale, garden and lay- common rules that were used in their design. The
out of housing units. The main goal of this research four case studies are: Belapur plan – located in New
is to define a methodology for developing bottom- Bombay, India, designed by Charles Correa in 1986,

Jardim São Francisco, – located in São Paulo, Brazil, describing the Belapur housing system as an incre-
designed by Demetre Anastassakis in 1989, Proyecto mental system following Correa’s cardinal principles
Experimental de Vivienda (PREVI) located in Lima, of incrementality and malleability as defined in his
Peru, designed by Christopher Alexander in 1969 bill of rights for housing in the third world (Correa,
and Quinta Monroy, in Iquique – Chile, designed by 1999).
Elemental in 2004. This paper is focused on the anal- From the analysis of the case studies it is pro-
ysis and grammar development of the Belapur plan. posed to infer a set of generic design patterns
The structure for developing a generic gram- codified as shape grammars. The concept of generic
mar consists of 4 parts: (1) the inference of a specific grammar was presented by Li (2001) and consists
analytical grammar from an existing case study; (2) of a widely applicable shape grammar embedding
creating a generic grammar by generalizing the the features of a reasonably abstract language
grammars obtained from the analysis of the case which is able to produce design in several specific
studies; (3) to improve the quality of the final de- languages by adjusting the grammar accordingly.
signs the generic grammar is revised according to Current researches present the application of ge-
qualitative requirements for public and community neric grammars in different contexts - Benrós et al.
areas based on urban design and housing plan lit- (2012) proposed developing a generic grammar
erature (Pedro, 1999; Barton et al., 2010); and (4) applied to housing and Beirão (2012) developed a
applying the new revised grammar in the develop- generic grammar for urban design context. However
ment of specific designs for social housing plans. Pe- the novelty of the ongoing research lies in the use
dro (1999) developed a systematic methodology of of this approach in a different scale – between hous-
qualitative requirements to be applied in the devel- ing scale and urban design scale, and to present a
opment of housing complexes in Portugal and Bar- bottom-up design system for the development of
ton et al (2010) presents a set of guidelines to design a housing complex which includes emergent pub-
neighborhoods that are safe, friendly and attractive. lic spaces related with the traditional use of public
This literature contributes to define the qualitative space. The generic grammar obtained by such anal-
requirements to be applied as a control mechanism ysis will then be assessed by comparing its features
in the generic grammar. The generic grammar is with the quality requirements defined by (Pedro,
organized in small thematic generative sets corre- 1999; Steiner and Butler 2007; Barton et al., 2010).
sponding to design patterns following the principles The goal is to obtain an improved grammar contain-
defined in (Beirão et al., 2012). ing heuristics based on the above mentioned qual-
ity requirements. Such structure resembles that of a
DESIGN METHODOLOGY discursive grammar as defined by Duarte (2001).
The design methodology consists on the use of This research is seen as an initial step in the de-
shape grammar as an analytical method applied to velopment of a generic grammar for social housing
study the design of Correa’s Belapur plan, to capture developments, especially regarding the generation
the design method used by the architect to obtain of housing plans, public spaces and community ar-
emergent community areas as the result of the re- eas. It goal is to add to the typical mass production
cursive application of local rules involving spatial strategy, features of diversity both at housing level
relations between plots and common public access and public space level.
areas. The interesting aspect of the Belapur plan is
that it succeeds in generating simultaneously hous- CASE STUDY: BELAPUR – CHARLES COR-
ing types and an emergent hierarchized structure REA
of public space. The spatial relations and local rules The Belapur Housing was designed by Charles Cor-
can be expressed in the form of a shape grammar rea, in 1983-1986 to accommodate more than 90%

Figure 1
1. Belapur housing develop-
ment: 1.7 housing units
around a courtyard 8m x 8m;
2. bigger module of 21 houses,
surrounding a community
space of 12m x 12m; 3. Sitting
plan (Correa, 1989).
of Bombay’s low-income profile with a variation more privacy and a sense of neighborhood at the
from 45m2 to 70m2 on house typology. The pro- smaller scale. Three of these clusters combine to
ject demonstrates high densities – 500 inhabitants form a bigger module of 21 houses, surrounding a
per hectare, including external areas, schools, etc community space of 12m x 12m (Figure 1-2).
(Correa, 1999). The site is located on six hectares of The houses were designed as an evolutionary
land 1 km away from the city center of New Bom- module, where “units are packed close enough to
bay and the development had to cover almost the provide the advantages of high density, yet separate
entire range of low-income groups – from the low- enough to allow for individual identity and growth”
est to the upper-middle categories (Correa, 1989). – this strategy allows growth from “a single lean-to
This plan presents a hierarchy of community spaces roof to urban town-houses” (Correa, 1989; 1999) be-
as a fractal structure; it consists of organizing 7 cause each dwelling is freestanding and does not
housing units around an intimate courtyard with share any wall or land with its neighbors, allowing
approximately 8m x 8m (Figure 1-1). This composi- a family to extend its home according to their needs
tion is repeated at a higher scale as shown in Figure by means of self-construction. Such policy towards
1-3 creating a similar composition which can itself house extension resembles that of the Elemental
be repeated at an even higher scale, hence creating concept developed by Alejandro Aravena (Aravena
the fractal structure. The first configuration provides and Iacobelli 2010). The plan clearly expresses Cor-

rea’s principle of incrementality and order becomes same geometry and is always composed of identical
an underlying feature which emerges in the growth lots although slightly different within each block.
process. This corresponds exactly to Correa’s drawings (Fig-
ure 1). Second, Grammar 0 is ignored here for it does
BELAPUR HOUSING SHAPE GRAMMAR not play any role in the emergence of public space.
The Belapur plan was analyzed because this design However, because some of the block rules (Gram-
concept contains qualities that allow growth at a lo- mar B) are related with the fact of a lot being already
cal level through a bottom-up process and contains built or not we assume that the Grammar A simply
rules to evolve and generate hierarchical external fills a lot with a fixed house shape (Figure 3).
spaces and community areas. This shape grammar The combination of these grammars gener-
gives rise to a growth system as a fractal structure. In ates designs in an incremental way following Cor-
other words, this bottom-up approach is capable of rea’s cardinal principles by applying the mentioned
responding to growth needs and provide social and grammars in parallel. The process is mainly additive
community spaces according to populations’ needs. and based on local rules which act according to vi-
According to Correa: cinity conditions (Figure 2).
“If there ever is a Bill of Rights for housing in
Third World, it would surely have to include – en- Grammar A
shrine! – the following cardinal principles: Incremen- Grammar A consists of organizing 7 housing units
tality, Pluralism, Malleability, Participation, Income by filling lots around an intimate courtyard. In its
generation, Equity, Open-to-sky space, Disaggrega- initial state each lot contains a label (+) defined as
tion” (Correa, 1999, p.109). “RI 01”, which can be replaced with a house. Placing
The bottom-up reasoning underlying the design the first house turns off all labels (+) except those in
can be demonstrated in the shape grammar that neighboring lots (Figure 3). This rule guarantees that
we developed. The system can be described with the generation proceeds by following a neighbor-
three parallel grammars: Grammar A, which fills lots ing condition. The block is filled by replacing labels
with an initial house volume, Grammar 0 which de- + with houses and adding + labels to the empty
fines the house extension rules within the lots and neighbor lots until all lots become filled (see rules RL
Grammar B which defines spatial relations between 01 to RL 07).
blocks. The first lot to be filled along a side of a block
In this paper we focused our attention on the generates a label () at the side of the block for
emergence of public space order, in particular, in applying the Grammar B, which may be applied in
the emergence of the three hierarchical levels of any iteration since the moment the label is available
public space. To do so, and due to space constraints (Figure 4).
some assumptions were considered to simplify the The rule RL 08 defines the insertion of the label
presentation of the argument in discussion here. (A) whose function is to remove one lot, thereby
Furthermore, Correa’s drawings actually show less enabling the permeability of the block. This feature
than the real potential underlying his system, even allows the grammar to create different access paths
though his claims for incrementality and malleabil- to the housing blocks as well as greater diversity of
ity would suggest a much wider use of the rules and public spaces (Figure 5).
the exploration of other simple variations as it will
be argued in the discussion section. As such, the as- Grammar B
sumptions considered here are basically two. First, The set of rules generates urban scale design by
that the block - 7 lots organized around a common defining spatial relations between three blocks (21
public space - is always the same, i.e., has exactly the houses), around a community space. The Grammar B

Figure 2
Belapur shape grammar
derivation.
consists in 4 rules by associating two blocks: all rules Control rules

erase the label () and associate a new block with 7 Control rules deletes labels in the space of one block
lots in 4 distinct spatial relations. The rule defined as by inserting optional 4 lots with houses. Thus, the
“Start Block” (RS) replaces the new block in gray and community area allows the insertion of 3 more ar-
starts the Grammar A placing labels (+) in the exist- eas which are constituted as a courtyard. The labels
ing lots (Figure 6).  and  defines possible association with other

Figure 3
Grammar A.
blocks (Figure 7). Note that labels  and  concern of the square and block: all rules off label () and
two different spatial relations regarding the orienta- associate a new block with 7 lots in two distinct spa-
tion of the blocks. The rule RC 09 removes one of the tial relationships (Figure 9).
optional lots inside the courtyard (see Rule - RC 02)
and inserts a portico to isolate a neighborhood unit DISCUSSION
constituted by three blocks. Thus, the rule allows The paper presents the Belapur grammar as a de-
creating different levels of privacy in accordance sign system, which may contribute to the future de-
with the growth of the scale of public spaces. velopment of a generic grammar for social housing
After applying control rules that generate the plans. The development of a generic grammar for
common spaces of the housing development, it is housing intends to contribute to the improvement
possible to apply the rules of Grammar B (see rules of public spaces and community areas by insertion
RB 05 to RB 08) which allow the continuous insertion of qualitative requirements as a control mechanism
of blocks and increase the scale of public space (Fig- that allows adding public spaces and facilities in a
ure 8). hierarchical structure in accordance to needs.
The Grammar B’ defines association between The bottom-up grammar’s structure explains
blocks with corner for the passage and association the concept of incrementality, pluralism and malle-
Figure 4
Derivation of a block.
Figure 5
Derivation of a block with an
empty lot.

Figure 6
Grammar B.
ability (Correa, 1999, p.109) and can transcend the nomena where order emerges from a multiplica-
set of design solutions from a few additional rules tion of local interactions eventually represented by
that are not part of the initial urban plan proposed a local rule, Correa’s plan intensions are best cap-
by Correa. Additional rules can also be set to react tured by a bottom-up grammar where local rules
to pre-existing features such as natural barriers, riv- provide not just the incremental procedure but also
ers, topography and empty spaces, among others. the underlying order which is always a goal in plan-
Although not formally expressed such rules are al- ning. This can explain the predictability of the result,
ready present in Correa’s design. despite its spontaneous characteristic. With this in
After analyzing Correa’s project and developing mind, we could argue that, although a hypothetical
a grammar for it an important issue has emerged: al- top-down grammar may computationally generate
though the grammar allows the incremental growth the same shape and order, from the analytical view-
of the housing development, how can the overall point it fails to capture the conceptual principles
result display such an orderly character? This issue underlying the system and therefore they cannot be
leads us the question of the design being a bottom- considered equivalent.
up or top-down process. Finally, a subject that needs further attention
To achieve his concepts of incrementality, plu- involves considering how the incremental structure
ralism and malleability, Correa, resorts to a design of these grammars deal with the subject of neigh-
system which is supposed to be implemented in a borhood facilities location. Such theme is present
bottom-up fashion. Similarly to many natural phe- in Correa’s plan but rules are not formally expressed

Figure 7
Control Rules.
and as a design attitude it seems that Correa simply tional item, therefore in a top-down fashion. How-
decided their location using them as a composi- ever, considering that one of the goals involved in

Figure 8
Rules RB 05 to RB 08.
the design is to follow an incremental growth princi- port to this research.

ple, this subject needs a more careful and profound
study regarding the definition of facility require- REFERENCES
ments for community planning social housing de- Alexander, C., 1977. A Pattern Language.
velopments. This subject will be extensive enough Aravena, A., Iacobelli, A. 2012 Manual de Vivenda Incremen-
for a future paper. tal y Diseno Participativo, 1ed, German: Hatje Cantz.
Barton, H., Grant, M. & Guise, R. 2010. Shaping Neighbour-
ACKNOWLEDGEMENTS hoods: For Local Health and Global Sustainability, 2nd
The authors thank São Paulo Research Foundation edn, Routledge.
(FAPESP), grant #2011/50139-0 and National Council Beirão, JN 2012, CItyMaker – Designing Grammars for Urban
for Scientific and Technological Development – Bra- Design, Architecture and the Built environment, Delft
zil (CNPq), grant # 201693/2012-7 for financial sup- University of Technology, Faculty of Architecture, n.05.
Figure 9
Grammar B’.

Beirão J, Duarte J, Stouffs R, Bekkering H, 2012, “Designing Thames & Hudson. p.48-51.
with urban induction patterns: a methodological ap- Correa, C 1989, The New Landscape: Urbanization in the
proach” Environment and Planning B: Planning and Third World. London: Butterworth Architecture, Mimar
Design 39(4) 665 – 682 Book.
Benrós, D, Granadero, V, Duarte, JP, Knight, T 2011, ‘Auto- Duarte, JP 2001, Customizing mass housing: a discursive
mated Design and Delivery of Relief Housing : The Case grammar for Siza’s Malagueira houses. PhD Dissertation,
of post-Earthquake Haiti.’ CAAD Futures: Designing To- MIT.
gether, pp.247-263. Li, A 2001, A shape grammar for teaching the architectural
Benrós, D, Duarte, JP and Hanna, S, 2012, ‘A New Palladian style of the Yingzao fashi. PhD Dissertation, MIT.
Shape Grammar - A Subdivision Grammar as alterna- Pedro, JB 1999, Programa Habitacional – Vizinhança Próx-
tive to the Palladian Grammar’. International Journal ima. Laboratório Nacional de Engenharia Civil, Infor-
of Architectural Computing, issue 04, volume 10, pp. mação Técnica Arquitectura - ITA 7.
521–540. Steiner, F. & Butler, K., 2007, Planning and Urban Design
Correa, C 1999. Housing and Urbanization. New York: Standards, Ed. John Wiley & Sons, Inc.

The Language of Mozambican Slums
Urban integration tool for Maputo’s informal settlements

Pedro Barros1, José Beirão2, José Pinto Duarte3
CIAUD, Faculdade de Arquitetura, Universidade Técnica de Lisboa, Rua Sá Nogueira,
Pólo Universitário, Alto da Ajuda, 1349-055 Lisboa, Portugal
2
http://www.urbangrammars.com/, 3http://home.fa.utl.pt/~jduarte/
1
prbarros.arq@gmail.com, 2jnb@fa.utl.pt, 3jduarte@fa.utl.pt
Abstract. A shape grammar was developed for analyzing the evolution of Maputo´s slums
with the strategic objective of capturing the evolution of house types and understanding
the social agreements behind the spatial relations of their house elementary spaces in
order to reuse such rules for the purpose of rehabilitation. This paper shows preliminary
results of the research and aims at developing, based on the resulting grammars, a
parametric tool able to execute morphological analyses, simulations and generate
improved design solutions for the qualification of Maputo´s informal settlements.
Keywords. Shape grammars; urbanism; computation; regeneration; informal settlements.
INTRODUCTION
This paper introduces a new approach for an urban settlement a slum, or is a slum created in unplanned
simulation framework for deteriorating unplanned areas, but it is fair to say that in most cases slums
settlements in the city of Maputo (also known as happen to be informal or unplanned areas that are
Caniços), areas that often are regarded as ‘slums’. suffering from multiple physical or socio economic
According to UN‐Habitat’s report (2003), The problems (Karimi and Parham, 2012).
Challenge of Slums, in 2003, 31.6 per cent of the
world’s urban population lived in slums or squatter RESEARCH PROJECT
settlements. The 2010 report - The UN State of Afri- This paper shows preliminary results of a PhD re-
can Cities – states that Mozambique’s urban popula- search aimed at developing a parametric tool able
tion will raise from 9 million in 2010 to 15.6 million in to execute morphological analyses, simulations and
2025, confirming the country’s position as having a generation of improved design solutions for the
significant growing of urban population for the next qualification of Maputo´s informal settlements. The
few decades. main goal is the creation of an integrated model
The difference between an informal settlement, that substantiates planning decisions and presents
an unplanned settlement, a slum, or a deteriorated itself as a viable methodology in the search of more
urban area is not always easy to define (despite the sustainable solutions.
UN-Habitat (2006) definition). In reality all these ar- This model is based on Stiny’s shape grammars
eas often overlap in terms of their characteristics, (1980) as means to elaborate plans capable of adapt-
function and appearance. Not always is an informal ing to changes in premises without losing its urban,

civic and environmental integrity. It is claimed that Finally, by recurring to shape grammars and us-
such a model has the ability of maintaining the ing them in a bottom-up fashion one may be able
urban and aesthetic coherence and the harmony to deal more easily with the emergent phenomena
with the place of intervention due to the capability that typically occurs in informal settlements by add-
of shape grammars of capturing the morphologi- ing new rules which express a local emergent recur-
cal characteristics of house types and therefore the rent pattern without losing the main order intended
features which are inherent to the dynamics of use, for the planning strategy.
to the cultural identity and social dwelling protocols
(Habraken, 2000). CONTEXT
The work is structured on the hypothesis that Authors like Paul Jenkins (2012) and Isabel Raposo
there is a grammar capable of, in a common lan- and J. Oppenheimer (2008) have been revealing a
guage, capturing simultaneously the intrinsic values weak interest by Mozambique authorities in urban
of the informal the needs regarding infra-structure development and weak capacity of the programs
requirements and respond in general to local popu- of decentralization by the municipality of Maputo
lation’s needs. For the latter purpose, the research that underestimates the tendencies of expansion
proposes the use of urban design quality standards of these settlements (the foresight is of 2,5 million
to constrain the grammar within qualitatively vali- in 2010 growing up to 4 million in 2025 in the city
dated boundaries. of Maputo). “Why do strategic approaches to urban
This methodology uses the Shape Grammar development generally not recognize emerging peri-
formalism resorting to its analytical and generative urban forms as valid and work with these, rather than
capabilities. The analytical process enables the iden- assuming these need to be replaced?” (Jenkins, 2012).
tification of the rules that generate informal urban Even The Challenge of Slums (UN-Habitat, 2003) re-
fabrics, hence it permits the description through port “suggests that in-situ slum upgrading is more ef-
Shape Grammars of the emergent phenomenon of fective than resettlement (…)” and that as “(…) slums
the informal. It also identifies representative fabric are in fact the dwelling places of much of the labour
samples that reveal objective qualities which will be force in their cities, they provide a number of important
used to reference the valuable boundaries for gram- services and are interesting communities in their own
mar parameters. Secondly, the generative process right.(…)” (UN-Habitat, 2003).
will aim at the creation of a design grammar that It seems then necessary to find new ways to un-
will adapt the rules inferred in the first process to the derstand the qualitative properties of existing urban
goals defined in a development vision. It will consist solutions so that we can consistently evaluate them
in the contextualized adaptation of the rules with and propose new ones. Such processes of revitaliza-
the strategic objective of rehabilitation and infra- tion can be modeled and simulated through genera-
structuring in order to improve the cultural setting. tive systems in a bottom-up approach.
Also, it is intended to develop an interpretative con-
sistent model that allows the deduction of the syn- THE CANIÇO SHAPE GRAMMAR
tactic and semantic rules by aggregating the analy- The idea of defining a Shape Grammar applied to
sis systems - Spacematrix (Berghauser-Pont and the structure of Maputo´s slums (so called ‘Cidade de
Haupt, 2010) and Space Syntax (Hillier and Hanson, Caniço’) appears, first of all, like an analysis of a social
1984) - in a substantiated operational base for deci- space displaced from the urban process and second-
sion support. It can then check the validity of the ly, with the strategic objective of rehabilitation.
interpretative model and of the proposed digital im- Here we present the preliminary results of this
plementation in order to promote the generation of ongoing research. The shape grammar presented
more sustainable urban solutions within the context. here is based on the Built Environment Study (An-

Figure 1
‘Bairro’ Magoanine B in the
city of Maputo. Sample of
‘Unofficial Planned fabric
(Google Earth).
dersen et al., 2012) data which is part of the broader facilitate the definition of the grammar at this stage.
research program designed by Prof. Paul Jenkins – In order to focus on the fundamental compo-
Home Space Maputo. It is based on what Andersen nents that constitute a typical plot, other elements
et al. and Jenkins (2012) define as ‘Type A’ houses. In are considered such as trees (which can be pre-
Home Space Maputo – Built Environment Study (An- existent) and the toilets separated from the house
dersen et al., 2012) the “many different house plans buildings.
have been divided into five general house types. It is Designs are shown bi-dimensionally. The gram-
however important to stress that many of the house mar develops by configuring the arrangement of the
types overlap each other; some house types become plots and then placing the basic form of the house (a
transformed into other types (…). These five general 7m x 3.5m rectangle) in each one of them. Addition-
house types are classified as the most common.” (Ibid). ally, the rectangle is divided in two functional zones,
Accordingly to this study, the first phase of the as “the most simple house type and in general has
house building construction often is to start with two divisions” (Andersen et al., 2012) - private and
the most basic type. Type A house is then the most social. Without any reference of construction tim-
simple, with only two divisions. “The house is entered ing or order, it´s here established that the placing of
from the center of the long façade directly into one the outside toilets happens before any other exten-
slightly larger room and with further direct access to sion of the house is made. The same order issue is
a bedroom. The house has one private room while the presented with the trees, especially with the larger
‘sala’ is for receiving visitors and at times acts partly as types. It´s assumed that most of them existed be-
a kitchen” (Ibid). fore any kind of land division. For design purposes
The plot and the block shape and size are taken it´s used the average treetop diameter of the three
from a sample of the ‘bairro’ Magoanine B (Figure 1) most common species to determine constraints re-
described in Home Space (Jenkins, 2012) as an ‘Un- garding the placement of houses in relation to the
official planned area’. This particular area was chosen tree’s position.
because unofficial planned areas, “which had com-
munity / private planning and sub-division interven- GRAMMAR
tions at some period, but were not registered formally The view of most informal settlements suggests an
in the land cadastre and/or registry” (Ibid), show a cer- organic and almost chaotic land occupation. How-
tain level of regularity in terms of urban layout that ever, Paul Jenkins’s (2012) studies identify four de-

Figure 2
a) Officially planned; b) Unof-
ficially planned; c) Upgraded;
d) Un-planned (in Home Space
Maputo, Jenkins 2012).
velopment statuses for land occupation in Maputo P. Rectangle P is 28m x 16m (average plot size from
city, according to different criteria such as the level the sample shown in Figure 1) and it is composed by
of planning, land registry and socio-economic con- the lines l’, l’’, 4/7 l and e (which define the limits and
ditions (Figure 2). Those statuses are: the entrance side of the plot). Line e contains at its
“Officially planned – areas which had state plan- midpoint a triangle symbol for the entrance.
ning and sub-division interventions at some period Once the first plot is established, other plots are
(…); added in order to create city blocks. Rule 1.2 mirrors
Unofficially planned – areas which had commu- P by its 4/7 l line and Rule 1.3 copies the mirrored
nity / private planning and sub-division interventions plots thirteen times until there is a twenty eight
at some period (…); plots city block. From here the blocks are replicated
Upgraded - areas which had been unplanned but orthogonally nine meters away from each other (the
had state, community or private planning or sub-divi- average street width). This presentation uses three
sion (…) city blocks (eighty four plots).
Un-planned – areas which had no previous (…)
planning or sub-division, often referred to as ‘sponta- Stage 2 - House in the plot
neous’ or ‘informal’ areas (…)” (Jenkins, 2012). After the city blocks generation, Type A houses are
The research focuses on the latter three statuses placed in each plot (Figure 4). “Three general tenden-
– the ones that show some kind of self-organization cies were recognized regarding how the houses of this
despite of any level of state/private intervention or type where located on the plot. The most common
planning. The grammar presented here is then the situation was the house located in the very far corner
first approach to the most ‘regular’ status of the of the plot and with two sides of the house connected
three that reveal ‘informal’ qualities - the ‘Unofficially to the perimeter walls (situation 1). The next common
planned’. location of the house on the plot is where the short end
The grammar is divided in three general sta- of the house was connected to the plot boundary and
ges. The first one relates the plots in order to create closer to the street (situation 2). Some cases also had
blocks. The second stage places the Type A house their house centrally free standing on the plot (situ-
inside the plot and the third configures the house ation 3)” (Andersen et al., 2012). Because this is the
extensions and other components like outside toi- only quantitative information for each of the three
lets and trees. situations, it’s established the probability of occur-
rence for each of the situations: situation 1 (the most
Stage 1 - Plots and blocks common) will occur three times in every six cases;
The composition starts with a given point (0,0,0) as- situation 2 will occur two times in every six cases; sit-
sociated with the symbol * (Figure 3). To this initial uation 3 will occur once in every six cases. The house
shape is then established the location of the first is represented by the 7m x 3.5m rectangle H, com-
plot. The plot is for now represented by the rectangle posed by the lines a’, a’’, b’ and b’’. Rectangle H can

Figure 3
Stage 1 in the grammar - Plots
and blocks.
be positioned horizontally (if a = 7m then b = 3.5m) and that dh1 = 0 or dh2 = 0 depending on the house
or vertically (if a = 3.5m then b = 7m) inside the plot. being placed in the right or in the left corner of the
In no case Type A house appears in the front of the plot.
plot. One important issue raised was the probable In Situation 2 (rule 2.2), where only the short end
pre-existence of trees. Trees are represented by cir- of the house is connected with the plot boundary
cumference i and its diameter corresponds to each (anyone but the front) there are different conditions
species average treetop diameter. To ensure that depending on vertical or horizontal positioning. If
no house is placed under a tree, two conditions are it is horizontal then dh1 = 0 or dh2 = 0 depending
established. The first is that circumference i cannot on the house being placed in the right or in the left
intersect rectangle H and the second is that circum- side of the plot. Also dv1 ≥ 1 to ensure some space
ference i cannot contain rectangle H. in the back of the house and dv2 ≥ l/3 (since there is
To control the placement of rectangle H (or any no case where the house is placed in the front of the
other component) inside the plot, to each side of plot it was established that the minimum distance to
rectangle H (a’, a’’, b’ and b’’) is added a dimension the front end is one third of the plot length – about
arrow d (dh for horizontal arrows and dv for vertical 9.3 meters). If it’s vertical then dv1 = 0. Also dh1 ≥
ones) that will manage the distances between each 1 or dh2 ≥ 1 depending on the house being placed
side of the house (rectangle H) and the limits of the more to the right or more to the left side of the plot,
plot (rectangle P). ensuring some space in the back of the house in any
Situation 1 (rule 2.1) is the most common loca- of the cases.
tion of the house (three in every six cases), where its In Situation 3 (rule 2.3) the house has a centered
corner coincides with one of the far corners of the position in the plot. None of its walls touch the
plot. Whether the house in a vertical or in horizontal boundaries of the plot which means that no dis-
position, the condition is that dv1 = 0 plus dv2 = l - b tance d equals 0. So dh1, dh2, dv1 ≥ 1 and dv2 ≥ l/3,

Figure 4
House location in the plot.
maintaining the same criteria of placing the house the examples shown in Home Space have the toilet
away from the front. placed in the front end of the plot. So line e (from
rectangle P) can not contain any line p (from square
Stage 3 - Extensions and components T). To ensure a considerable distance from the house,
“The Home Space study provides evidence that the lo- it is established that the toilet must be placed in one
cation of toilets and bathrooms most commonly are of the other three plot’s limits and that the minimum
in a separate building or a screened off location as far distance to the house is six meters. This minimum
as possible from the main house. This configuration distance is assured by the placement of an auxiliary
was seen in 74% of the cases” (Andersen et al., 2012). circumference c with a six meter radius. Circumfer-
It also shows that the transition from outdoor to ence c is centered with square T. Circumference c
indoor toilets corresponds to an upgrade process cannot intersect or contain rectangle H (house).
that seems to be slow due to the lack or insufficient The next stage is to divide the house (rectangle
sewage infrastructures. Because Type A house is the H) in two labeled divisions – bedroom B (B = 3m x
most basic one (associated with the lower income 3.5m = 10.5m2) and ‘sala’ L (L = 4m x 3.5m = 14m2)
families) and corresponds to the starting stage of – and mark the door label with a triangle. As men-
the house building construction, it is settled that for tioned above, the main entrance is in the center of
this type all toilets are outside of the house. For the the long façade. Because there are two, the door
grammar the toilets are represented by the 1.5m x label is to be placed in the one that has a longer
1.5m square T, composed by the lines p1, p2, p3 and distance d (whether it is a dv or a dh). This condition
p4 (Figure 5). Square T is inside rectangle P (plot). denies any chance of having a door facing directly at
Another observation we can make is that none of the boundary wall. It´s important to stress that the

Figure 5
Toilets location; House divi-
sion.
yard is a scene for “everyday life and a space for so- Extension 1 adds one room to the front of the
cialising, household work and sometimes also a space house. Extension 2 creates a big living area in the
for economic activities” (Ibid). The toilet door follows front of the house and a new entrance way to the
similar criteria. In this case the only condition is that side of the ‘sala’ (formally labeled with L). The only
it never faces the front side of the plot (line e). In condition here is that the distance d at this side
other words, it can only be placed in its horizontal must be greater than the opposite d. This will grant
distance dhn or in its vertical dv1. a larger yard area in front of the new entrance of
“House type A can be extended in various ways” the house. The only extension that can be made to
(Ibid). The extensions used for the grammar fol- the back of the house is Extension 3. It basically mir-
low the four examples shown in Home Space. Type rors the house to the back or to the front, creating a
A houses are extended with additional one or two new inner door. If the extension is to the back, the
rooms with similar size and shape as the existing distance d in the back of the house has to be big-
ones (larger extensions would transform the Type A ger than 4.5 meters (3.5m for the new body plus 1m
house into other types). Therefore, five different lay- for the new back door passage). Extension 4 creates
outs are created for Type A house: extension 0, 1, 2, a large living room to the front plus a ‘veranda’. “The
3 and 4 (in extension 0 the house keeps its original veranda is not only a transition space between outside
configuration - Figure 6). In the absence of quanti- and inside - which can be used for practical purposes
tative information, it is established that all the five as cooking, storage, or a social space for - but also a
layouts are applied in the same number (each is ap- way of representing the house in the neighbourhood.
plied once in every five cases). The application of Many of the verandas had burglar bars and some of
these transformation rules implies the elimination these were richly ornamented” (Andersen et al., 2012).
of the labels in the house and the insertion of the This extension requires that the house have a hori-
doors. zontal position because the “veranda is always facing
Figure 6
House extensions.

Figure 7
3D Model.
the street” (Ibid). CONCLUSION

“Trees provide shade and hence create spaces for The research described in this paper constitutes the
socializing and domestic work, bear fruit or can fig- first step towards the development of a computa-
ure as decorative elements. Trees also had a spiritual tional model for Maputo informal settlements. The
function vis-à-vis the reverence for ancestors and tra- ultimate goal is to use this model to support deci-
ditional ceremonies often involve trees. The ‘quintal’ is sion in urban interventions that have similar spatial
a space for social interaction where e.g. visitors can be features and to improve them from the environmen-
received in the shade under a tree. The majority of the tal viewpoint by manipulating the rules in the mod-
cases, 73%, have a shade providing tree in their yard el. The model uses shape grammars to encode the
and in 2/3 of these cases, the tree is a big mango tree underlying syntactic rules describing the language
and in other cases there are mafureira (Natal Mahoga- of the ‘Caniços’ which capture the social features un-
ny) or canhueira (Marula) trees, both producing fruits” derlying these morphological types. It aggregates
(Andersen et al., 2012). As seen above there is a pre- two essential grammars: the grammar to generate
liminary condition about the possible pre-existance the urban fabric and the grammar to generate the
of trees. In order to control the placement of planted houses and its components. This paper describes
trees, there is one first rule that defines the contour the preliminary results of the second generative
of the house. This contour line prevents that the tree grammar.
is placed over the house. Thus the contour line can- An important point for discussion here is to
not intersect circumference i. evaluate the capacity of the grammar of really cap-
Despite the fact that Type A houses only have turing what are the reasons behind the transforma-
one floor and for that the grammar is essentially bi- tion happening in the ‘Caniços’. In many situations,
dimensional, there are a set of rules that extrude the different grammars can be used to produce the
print for visualization purposes. This third dimen- same formal arrangements, in other words, in some
sion uses basic façade composition shown in Home circumstances different sets of rules are able to pro-
Space as well as the low pitched roofs (made of cor- duce the same shapes. However, as analytical tools
rugated iron sheets). When applied to the eighty there is only one grammar which is capable of re-
four plots, the 3D visualization gives us a better view producing simultaneously shape and the generative
of the set as it represents a powerful and effective process which originated the shapes of a corpus of
way of communication (Figure 7). designs used as case studies. Therefore, if a grammar

produces the shape of our case studies in Maputo, developed having in mind three kinds of extension
this is not enough evidence that the grammar is ef- opportunities: (1) plot subdivision; (2) house exten-
ficient for our purposes. We still need to validate the sion and (3) addition of areas for commercial ac-
grammar against evidence of the real motivations tivities. Evidence of possible valid occupations may
behind one or another particular transformation to be captured from the analyses of denser and more
check if it is really replicating the social behavior un- compact areas.
derlying such transformations. This is important con- Another key issue will be the transformation of
sidering that the reuse of the grammar is the main the grammar in terms of sanitation conditions. Ac-
goal in mind and that it is supposed to deal with the cording to Andersen et al. (2012), one of the main
typical social behavior while providing some degree aspirations of these populations concern improving
of control to improve the final outcome. This valida- water and sanitation facilities, such as having run-
tion still needs to be done and is planned as the next ning water on the site, including building a bath-
step of the research. room inside the house. This is an essential aspect to
According to Andersen et al. (2012), “buildings be developed in order to improve the living condi-
are in continued process, always under construction tions.
and in various stages of being extended, appended, The current grammar is essentially a bi-dimen-
built or finished. This process sometimes spans over sional one due the fact that the case studies worked
decades (…)”. It has to be stressed that this gram- until now refer to ground floor houses, but if a den-
mar deals only with what could be considered as sification strategy is to be taken in consideration,
the most basic module unit of house buildings in evidence from other informal areas where density
peri-urban Maputo though it encompasses some is already higher and construction includes taller
evolutions of the Type A (one of the cases may be informal buildings will need to be incorporated in a
included in Type B) houses. As Home Space Built En- more complex grammar. This work is already under
vironment Study (Andersen et al., 2012) mentions, development. The only 3D approach done until now
“the size of the house was clearly linked to wealth consists of a simple extrusion of the layouts gener-
of the residing household” though “the survey does ated by the grammar (Fig.2) which is used for visuali-
not carry any clear evidence of a relation between zation purposes.
economic status, location in the city or planning cat- Finally, one of the main problems involved in
egory”. This means that Type A house evolution is in slams’ sprawl is the fact that it generates continuous-
most cases linked with socio-economic changes in ly spreading homogeneous areas where no urban
the household which need further understanding. hierarchies are usually found. A bottom-up grammar
Evidence raised here was the low density occu- will tend to simply replicate this behavior. Therefore,
pation of the plots underlining the clear importance in planning terms it becomes evident the need for
of the yard as a space for social interaction, domes- introducing evaluation and control mechanisms
tic work or for small economic activities. Another (and an additional grammar) which should react to
aspect is that this low density occupation is respon- the changes in the occupation conditions (and den-
sible for an increasingly spreading urban sprawl ef- sity) and generate the additional features that oth-
fect which creates a great dependency on car use erwise would not be defined by the grammar. This
in a city with no adequate traffic infrastructure and process will be the subject of a future paper.
a large amount of population with no motorized
means of transport. Therefore, the study of densifi- REFERENCES
cation strategies within the existing structures will Andersen, JE, et al. 2012, ‘Built Environment Study’, Home-
be a key issue in terms of the development of plan- space in African Cities, Danish Research Council for In-
ning strategies. Densification studies still need to be novation, Maputo.

Berghauser-Pont, MB, Haupt, P 2010, Spacematrix - Density puto, Edições Colibri / Centro de Estudos sobre África e
and Urban Form, NAi Publishers, Rotterdam. do Desenvolvimento, Lisboa.
Habraken, NJ 2000, The Structure of the Ordinary, MIT Press, Stiny, G 1980, ‘Introduction to shape and shape grammars’,
Cambridge. Environment and Planning B, vol. 7: p.343-351.
Hillier, B., and J. Hanson. 1984, The Social Logic of Space, Vol. UN‐HABITAT 2010, State of African Cities, Governance, In-
2. Cambridge University Press Cambridge. equalities and Urban Land Markets, United Nations Pub-
Jenkins, P 2012, Homespace in African Cities, Danish Re- lications.
search Council for Innovation, Maputo. UN‐HABITAT 2006, State of the World’s Cities Report
Karimi, K, Parham, E 2012, ‘An Evidence Informed Approach 2006‐2007: The Millennium Development Goals and
To Developing an Adaptable Regeneration Programme Urban Sustainability - 30 Years of Shaping the Habitat
For Declining Informal Settlements’, Eighth Internation- Agenda, United Nations Publications.
al Space Syntax Symposium, Santiago. UN‐HABITAT 2003, The Challenge of Slums: Global Report on
Raposo, I, Oppenheimer, J 2008, Subúrbios de Luanda e Ma- Human Settlements, Earthscan Ltd.

Gulou Structure Grammar and its Computer
Implementation
A computational approach to preserve the ethnic building technique and

to guide new designs
Lu Xiong1, Wei Xiong2, Hongxia Zhang3

1
State Key Laboratory of Subtropical Building Science, South China University of Tech-
nology, 2Guangxi University, 3Guangzhou Urban Planning and Design Survey Research
Institute
1
bbxlarc@gmail.com, 2g.bear@163.com, 3zhxlr@gmail.com
Abstract. Gulou is a type of building found in ethnic Dong people’s settlements in south
west China. It plays a significant role in the traditional Dong architecture and shows
both social and technical values. In the near future the technique as an intangible culture
heritage would face the risk of extinction because of globalization. The paper argues
that the use of formal grammar and computer tools could help the preservation and
learning of the design knowledge of Gulou Structure and develop Gulou designs which
would be adapted to modern needs. A shape grammar called Gulou Structure Grammar
(GSG) and its computer implementation are made to achieve the goals of capturing the
design knowledge of Gulou structure, generating new Gulou designs and promoting the
education of Gulou building techniques.
Keywords. Gulou structure; shape grammar; parametric model; ethnic building
technique.
INTRODUCTION
Gulou is a type of building found in ethnic Dong social activities and communications. For example,
people’s settlements in south west China. Dong the children should be given their names in Gulou;
settlement is composed of the basic family group people gather in the Gulou after work to exchange
“Dou”, every Dou should have its own Gulou as a ideas and chat with each other. All the cultural and
symbol for the family group. There trend to be 2 to social activities of Dong people are related to and
4 Gulou in a village (Figure 1). Gulou plays a signifi- influenced by Gulou. The culture identity and value
cant role in the traditional Dong architecture and system of Dong people are emerged from the activi-
shows both social and technical values. The wooden ties carried out in Gulou.
tower is the most important public building in a set- Apart from the social importance and value, Gu-
tlement as the village council and the senior states- lou is also famous for the unique uprising shape and
men’s centre. It also acts as an important place for the excellent building techniques without using any

Figure 1
Two Gulou in a Dong village.
Figure 2
The typical section of Gulou.
metal parts. As an ancient high-rise wooden struc- willing to learn the technique from the old. In the
ture which is still been applied nowadays, Gulou is near future the technique as an intangible culture
one of the most excellent traditional Chinese wood- heritage would face the risk of extinction. Unlike the
en buildings. There are 2 kinds of main columns in building and construction manuals for the official
Gulou: the inner columns and the outer columns. Chinese traditional buildings, the rules of Gulou are
The inner columns together with the connecting not originally organized in any written form or draw-
beams play the very similar structure role as the core ing. Instead, they are passed down from generations
in the modern high-rise tower. The outer columns to generations by pithy formulas. While designing
are connected to the core by beams which support and building a Gulou, the craftsman seldom produc-
the multi-level uprising structure. The flexible joint es any drawing as well. The implicit way of design
design also enhances the anti-earthquake perfor- makes the technique to be difficult to understand
mance. In terms of ecology, the Gulou technique and to learn, which also limits the wide spread of the
accumulates the long term low-tech experience of building culture. Although various studies managed
Dong people to avoid the impact of the hot and hu- to uncover the rules of Gulou building technique,
mid weather condition. For instance, the multi-level few efforts were made in the area of the formal and
roofs could promote ventilation while keep the inte- computational approach (Cai, 2004).
rior dry from the rains (Figure 2). The paper argues that the use of formal gram-
Like the other developing regions in China, the mar and computer tools could help the preserva-
old and vernacular building and technique are chal- tion and learning of the design knowledge of Gulou
lenged with globalization. Seldom young men are Structure and develop Gulou designs which would

be adapted to modern needs. Shape Grammars umn or multiple columns. Besides the single inner
were successfully applied in the research of tradi- column case, both the number of inner column and
tional Chinese wooden structures (Li, 2001) and outer column should be even number. The research
dwellings (Chiou and Ramesh, 1995). In this paper focuses on the most common type of Gulou – the 4
a shape grammar called Gulou Structure Grammar inner columns and 8 outer columns Gulou.
(GSG) and its computer implementation are made to The inner columns are placed at the corners of
achieve the following aims. a square while the outer columns are placed at the
1. Capture the design and construction knowl- corners of an octagon. The relationship between the
edge of Gulou structure via shape grammar. square and the octagon varies into 2 situations. In
2. New Gulou design could be generated with the the first situation the diagonal lines of the 2 shapes
use of digital tools which are developed based are parallel while the second situation the edges of
on GSG. the 2 shapes are parallel. Different alignment leads
3. GSG and the digital tools could be used in edu- to different placement of sections. In the first situa-
cation to facilitate the teaching of the rules of tion the sections are placed between the corners of
Gulou. 2 shapes and between the corners of the octagon
4. The digital tools could be used to explore the and the edges of the square. In the second situation
performance of Gulou. the sections are only placed between the corners of
the 2 shapes (Figure 7).
THE GENERATION OF GULOU STRUC- The section of Gulou can be divided vertically
TURE GRAMMAR into 3 parts: the base, the body and the top (Figure
The authors managed to extract the compositional 2). The structure of the base is composed of the main
rules of the Gulou wooden structure. The rules were columns and is rather simple. The body contributes
further translated to a shape grammar – Gulou most for the symbolic appearance of the building.
Structure Grammar (GSG). It is the most important part and highly reflects the
building technique of Dong people. The decorative
Analysis of the compositional rules of the top with fine structure acts as a balanced visual end-
Gulou wooden structure ing to the building. The following article focuses on
The exterior appearance and interior space of tradi- the composition of the body and the top.
tional Chinese wooden building are entirely defined The wooden frame of the body is composed
by the main wooden structure. The composition of of inner column, outer column, short column, cen-
the wooden frame is best demonstrated in sections. tre column, rafter and beam. The inner column and
The function and plan design of Gulou trend to be outer column are connected with beams which are
simple due to its symbolic meaning in the village. cantilevered to support the eaves and rafters. A
The plan shape is usually a square, a hexagon or short columns are added to a beam to support its
an octagon. Therefore the section design plays the upper level beam and rafter. After several iterations
most significant and complicated roles in the de- the body grows to its full height and the top col-
sign of Gulou. During the design process, the master umns are connected to a central column by beams.
craftsman is in charge of the section drawings rather The span of beams decreases level by level therefore
than plan drawings. The plan provides the guide each level gets smaller and smaller to form the taper
planes which the sections will be attached on. outlook of the tower. When the decreasing reaches
Gulou could be classified by the number of main to the point when the position of short column will
columns which touch the ground. There are 2 types be placed on the inner side of the inner column, a
of main columns: Inner column and Outer column. centre column must be added to the section to sup-
The inner column can be composed of a single col- port the upper level beams (Figure 3). Via the control

Figure 3
When the decreasing reaches
to the point that the position
of short column will be placed
on the inner side of the inner
column, a centre column must
be added to the section to sup-
port the upper level beams.
Figure 4
By the control of the position
of main columns, the number
of the position of main columns, number of levels, brackets. of levels and the height of
height of levels and the decreasing distance of each 3. Equally divide each edge of the profile in to N levels, the facade profile could
level, the facade profile could be adjusted (Figure segments and get the division points. In this be adjusted.
4). Equalized decreasing will result in the tilted lin- case N=10.
ear profile while the uneven decreasing will result in 4. Connect the points with the odd number index
curved profile (Figure 5). i on the odd number level to the points with
The top of the Gulou is also called “honey comb” according index on the upper level; connect
by the extinguished look (Wu Lin, 2009). It is com- the points with even number index i on the
posed of many layers of overhanging and self-sup- even number level to the points with according
porting wood pieces. Each layer is subdivided into index on the upper level. For instance, point 3
many segments so the overall structure of the top is on the 1st level will be connected to point3 on
a complicated cell-looking system. The composition the 2nd level; point 6 on the 4th level will be
of the top could be illustrated by the following steps connected to point 6 on the 5th level.
(Table 1): 5. Connect the points with the odd number index
1. Define the plan profile of the base of the top. In i on the odd number level to the points with in-
this case the profile is an octagon. dex i + 1 on the upper level; connect the points
2. Offset the profile to get the shape of each lev- with even number index i on the even number
el. In this case the top is composed of 6 level level to the points with index i + 1 on the upper
Figure 5
By the control of the decreas-
ing distance of each level, the
facade profile could be adjust-
ed. Equalized decreasing will
result in the tilted linear profile
while the uneven decreasing
will result in curved profile.
In this case the decreasing is
controlled by a Bezier curve.

Table 1
The steps to generate the
honeycomb top.
Step 1 Step 2 Step 3 Step 4
Step 5 Step 6 Step 7
level. For instance, point 3 on the 1st level will are generated from step 4 to 6. The brackets are
be connected to point 4 on the 2nd level; point attached according to the guide lines to form
6 on the 4th level will be connected to point 7 the base part of the honeycomb top.
on the 5th level.
6. Connect the points with the odd number index The content of GSG
i on the odd number level to the points with in- After the analysis of the compositional rules of Gu-
dex i - 1 on the upper level; connect the points lou, GSG was formulated. It consists of 3 initial de-
with even number index i on the even number sign and 24 rules. Rules were divided into 3 groups:
level to the points with index i - 1 on the upper plan rules (Table 2), body section rules (Table 3) and
level. For instance, point 3 on the 1st level will top rules (Table 4).
be connected to point 2 on the 2nd level; point
6 on the 4th level will be connected to point 5 THE COMPUTER IMPLEMENTATION OF
on the 5th level. GSG
7. All the guide lines for the leaf-shaped brackets A parametric model was built based on GSG. Sev-
Table 2
Rules of plan generation.
Plan initial shape R1: get the section guide 1 R2: get the section guide 2

Table 3
Rules of section generation.
initial shape: the inner column R3: get the outer column R4: get an end point of beam
R5: add an outer beam R6: get the centre column R7: get an inner beam
R10: get an overhang short

R8: get a rafter R9: get a short column column
R11: get a rafter R12: extend a column to a rafter R13: get a connecting beam
R14: get a connecting beam R15: get an inner column which

seats on a beam.
eral key parameters were identified: type of plan, no3d is chosen as the platform to develop the para-
distance between inner columns, distance between metric model. Both the axis and the solid model of
inner and outer columns, height of the base, body the wooden frame pieces can be obtained from the
level height, number of body level, a Bezier curve to model (Figure 6).
control the span decreasing of each body level, top
level height, number of top level and the increase THE APPLICATION OF GSG AND THE
span of each top level. Detailed parameters were PARAMETRIC MODEL
also defined: cantilevered distance of beams, rafter As an ancient building type oriented from the Ming
angle, and column lower extension length. Dimen- dynasty (1368–1644), Gulou is still being built and is
sion parameters were added to determine the size playing an important role in the life of Dong people
of the components such as the radius of columns nowadays. It is also widely used in public parks and
and the height of the beams. Grasshopper in Rhi- tourism sites in non-Dong areas for its distinguished

Table 4
Rules of bracket axis genera-
tion.
initial shape: the inner columns R16: get the plan profile R17: offset the profile
R18: divide the edge by n R19: connect the division points R20: connect the division points
R21: connect the division points R22: connect the division points R23: connect the division points
R24: connect the division points
symbolic form and strong landmark effect. From view of the tool. He held the view that the tool could
the modern use of Gulou, it could be identified as a rapidly generate designs according to the rules and
type of contemporary architecture. However, both parameters, therefore the communication with cli-
its design and construction are still based on the ent could be carried out efficiently. Also the model
old manual approaches. Digital technologies could provided all the dimensions of the main structure
serve the design and construction of Gulou as new pieces and a spread sheet of the use of material.
instrument, therefore Gulou could evolve and be The work used to take months to do could be com-
adapted to the information age. pressed to be finished in days.
The parametric Gulou model was used in the de- GSG is also used in the teaching of the course:
sign of a landmark structure in a resort area in San- Guangxi ethical buildings in the architecture school
jiang, Guangxi province. The famous Dong crafts- of Guangxi University. Gulou is an important topic
man Wu Shikang was invited as a design consultant of the course. GSG and the parametirc model are in-
for the project. During the design process, a series troduced to unveil the design rules and construction
of design models were generated with the help of process of Gulou. The students can learn the rules
the parametric tool (Figure 7). Wu gave a positive re- from GSG in a graphic and formal way. Then design

Figure 6
Both the axis and the solid
model of the wooden frame
pieces can be obtained from
the parametric model.
experiments could be carried out. Students use their DISCUSSION AND FUTURE WORK
own set of parmeters to generate designs with the The shape grammar of Gulou structure was formu-
parametric model. The vivid digital way of teach- lated as shown by the paper and a computer tool
ing encourges the students to ananlysis traditional was made to assist the design and education of the
chinese buildings in a computational point of view, ethnic building. However, the potential of GSG and
and to explore new designs based on the traditional its computer implementation is not fully explored.
building tecniques. The paper serves as a start point for the long term
Figure 7
4 different designs are gener-
ated for the project in a tourist
site.

study of the building techniques of Gulou. The fu- simulation software would be introduced in future
ture work based on GSG would address the issue of studies to unveil the ecological means in Gulou
building performance evaluation, the detail wooden building techniques.
structure of Gulou and its implication to contempo- The detail joint design plays a significant role
rary architecture. in the wooden structure. The sophisticated tenon-
The building performance of Gulou mainly con- and-mortise work connects all the wooden pieces
tains 2 aspects: the structural performance and the together without using any nail or metal part. The
passive solar comfort. The performance evaluation of design grammar of the detail wooden structure will
Gulou should take the considerations of the culture be carried out in the further study.
and activities of Dong people. In this paper the struc- The aesthetic, structural, ecological and detailed
tural performance is focused on material consump- design features of Gulou would have great implica-
tion. The material for building Gulou is fir. On one tion to the design of contemporary architecture. The
hand, Dong people are willing to travel hundreds of features could be further explored and applied to
kilometers to find the right and divine fir and trans- new designs. Further design experiments and prac-
port it back via man labor. On the other hand, nowa- tices will be carried out with the considerations of
days the forest resources in China are getting more the features of Gulou. We believe that the computa-
and more precious. The two factors require the size, tional design approach will help us adopt and apply
length and numbers of the fir for Gulou construction the characters of ethnic building for the emergence
to be carefully calculated in order to minimize the of the new architecture.
material usages while achieving the design purpos-
es. Future work will study the relationship between REFERENCES
the form and the structural performance of Gulou. Cai Lin, 2004, ‘The construction technique of Gulou in eth-
Passive solar comfort evaluation consists of the nic Dong group’, Huazhong Architecture, 03, pp. 137.
assessment of daylight factor, solar radiation and Li, Andrew I-kang. 2009, A shape grammar for teaching the
ventilation. They are related to the overall dimen- architectural style of the Yingzao fashi. PhD dissertation,
sions of the tower and the openings. There is one Department of Architecture, Massachusetts Institute of
unique factor of thermal comfort in Gulou: there Technology, Cambridge, USA.
is always a fire place in the centre of the ground Chiou, Shang-chia, and Ramesh Krishnamurti 1995. ‘The
floor. It is a symbol of energy in Dong culture. The grammar of Taiwanese traditional vernacular dwell-
fire place plays the role of a heat source and will af- ings’. Environment and planning B: planning & design 22,
fect the air flow in the chimney-shape of the inner pp. 689-720.
space. Multi-level roofs could encourage the use of Wu Lin, 2009, ‘The composition of the top of Gulou in eth-
daylight and natural ventilation while keep the inte- nic Dong group’, Architectural Journal, 2009(s1), pp. 20-
rior dry away from the rains. Advanced performance 23.

Index of Authors
A Bühler, Katja 2-279
Achten, Henri 1-51, 1-213, 1-477, Buš, Peter 1-283
2-515, 2-555
Aish, Robert 2-87 C
Akipek, Fulya 2-247 Çağdaş, Gülen 1-375
Akküçük, Zeynep 2-305 Camporeale, Patricia 2-227
Altan, Hasim 1-137 Canavezzi de Abreu, Sandro 1-449
Andersen, Marilyne 1-147 Carlson, Anders 1-69
Androutsopoulou, Eirini 1-275 Castro e Costa, Eduardo 2-635
Angeloni, Guilherme 1-157 Celani, Gabriela 2-705
Araneda, Claudio 2-355 Cerovic, Milutin 1-633
Araya, Sergio 1-575 Chang, Jia-Rey 1-529
Arrobas, Pedro 1-291 Charlton, James 1-205
Attia, Shady 1-21, 1-147 Chatzikonstantinou, Ioannis 2-137
Avantaggiato, Luigi 2-495 Ciftcioglu, Ozer 2-335, 2-345
Avincola, Eolo 2-29 Cohen, Itai 1-175
Coimbra, Eugénio 2-677
B Colakoglu, Birgul 2-273
Baharlou, Ehsan 2-165 Côté, Samantha 2-397
Balakrishnan, Bimal 1-101 Coutinho, Filipe 2-655
Barczik, Günter 2-571 Cutellic, Pierre 1-393
Barros, Pedro 2-715
Baur, Marco 2-237 D
Bayraktar, Mehmet Emin 1-375 Davila Delgado, Juan Manuel 2-155
Becerra Santacruz, Habid 1-137 de Boissieu, Aurélie 2-563
Becker, Mirco 1-643 De Meyer, Ronald 2-627
Beirão, José 1-291, 2-705, 2-715 De Somer, Pauline 2-505
Ben Rajeb, Samia 2-605 Díaz, César 1-117
Bereuter, Martin 1-605 Dieckmann, Andreas 2-485
Bernal, Marcelo 2-595 Di Mascio, Danilo 2-435
Bernhard, Mathias 1-597 Dolas, Caner 2-485
Bhagra, Saurabh 1-625 Donath, Dirk 1-109, 1-487
Bielik, Martin 1-109 Dritsas, Stylianos 1-655
Biloria, Nimish 1-529, 2-117 Duarte, José Pinto 2-635, 2-655, 2-687,
Bittermann, Michael S. 2-335, 2-345 2-705, 2-715
Boeykens, Stefan 2-505 Duddumpudi, Krishna 1-507
Bosi, Felipe 1-383 Duro-Royo, Jorge 1-585
Both, Katherine 1-337 D’Uva, Domenico 2-315
Bourdakis, Vassilis 1-185 E
Boyd, Jeffrey E. 1-467 Ebertshäuser, Sebastian 2-525
Boyce, Mary 1-575 Economou, Athanassios 2-617
Brakke, Aaron 1-117 El Ahmar, Salma 1-539
Bristogianni, Telesilla 2-49 Eloy, Sara 2-621
Brito, Thiago Mello 1-157 Erdine, Elif 1-559
Brown, Andre 2-697 Etman, Omar 1-459

Ezzeldin, Sherif 1-459 I
Ianni, Manuela 1-31
F Igarashi, Takeo 1-693
Falcão, Ana Paula 1-265 In, Jessica 1-665
Ferreira, Victor 1-439, 2-445, 2-655
Figueiredo, Bruno 2-687 J
Fioravanti, Antonio 1-81, 1-539, 2-29, 2-385, Jabi, Wassim 2-217
2-495 Jakubal, Vladimír 1-213
Fiorito, Stefano 1-439 Janssen, Patrick 2-205
Florián, Miloš 1-51 Janusz, Jan 1-91
Foged, Isak Worre 2-99 Junk, Stefan 2-397
Folcher, Viviane 2-605
Fraguada, Luis 1-433 K
Fricker, Pia 2-545 Kaftan, Martin 1-683
Fukuda, Tomohiro 1-219 Kalaouzis, Giorgos 1-185
Kalay, Yehuda E. 2-59
G Kalvo, Raul 1-655
Gadelhak, Mahmoud 2-261 Kanasaki, Kenji 1-711
Gallas, Mohamed-Anis 2-107 Kayser, Markus 1-585
Gargaro, Silvia 1-81 Kepczynska-Walczak, Anetta 2-453
Geddert, Florian 1-109 Kimpian, Judit 2-79
Georgakopoulou, Sofia 1-255 Kiviniemi, Arto 2-465
Gerber, David 1-69, 2-175 Kınayoğlu, Gökhan 2-297
Girot, Christophe 1-433, 2-545 Klein, Ralf 2-505
Gokmen, Sabri 1-497 Klofutar Hergeršič, Ana 1-227
Gonzalez Uribe, Carlos David 1-585 Knight, Michael 2-697
Grasl, Thomas 2-617 Knippers, Jan 1-549, 2-237
Greenberg, Dan 1-569, 2-147 Kocaturk, Tuba 2-465
Grochal, Barbara 2-217 Koch, Volker 2-587
Guéna, François 2-563 Koenig, Reinhard 2-195
Gün, Onur Yüce 1-61 Koltsova, Anastasia 2-375
Gürel, Nihan 2-415 Kos, Jose Ripper 1-157
Koutamanis, Alexander 2-475
H Kretzer, Manuel 1-615
Haeusler, M. Hank 1-233 Krüger, Mário 2-655, 2-687
Halin, Gilles 2-107 Kucukoglu, J. Gozde 2-273
Hanafi, Mohamed 1-539 Kunze, Antje 1-41
Hanna, Sean 2-39 Kurilla, Lukáš 1-51, 2-555
Hanzl, Małgorzata 1-319
Haymaker, John 2-595 L
Heinzelmann, Florian 1-175, 2-49 LaMagna, Riccardo 1-549
Heitor, Teresa 1-337, 1-403 Laucks, Jared 1-585
Hesselgren, Lars 2-39 Leclercq, Pierre 2-605
Hirschberg, Urs 1-683 Lecourtois, Caroline 2-563
Hofmeyer, Herm 2-155 Li, Di 2-697
Hong, Seung Wan 2-59 Lin, Jui-Yen 2-21
Hong, Sung Min 2-79 Lin, Shih-Hsin Eve 2-175
Hulin, Jaroslav 2-255 Loffreda, Gianluigi 2-495

Lorenz, Wolfgang E. 2-667 P
Lotte, Fabien 1-393 Pak, Burak 1-423
Lückert, Angelika 2-587 Papadonikolaki, Eleni 2-475
Papasarantou, Chrissa 1-185
M Parascho, Stefana 2-237
Makris, Michael 1-69 Parisel, Claude 2-325
Malheiros, Victor 1-383 Paterson, Greig 2-79
Margotto, Mário 1-383 Patlakas, Panagiotis 1-137
Marques, Rui 2-621 Pavlicek, Jiri 2-255
Martens, Bob 2-279 Pla-Catala, Anna 2-579
Massara, Bruno 1-383 Pungerčar, Enej 1-227
Matějovská, Dana 1-213
Mateus, Luís 2-445, 2-655 R
Medeiros, Valério 1-337 Reichert, Steffen 1-549, 1-575
Melsom, James 1-433 Rezvani, Samaneh 1-357
Mendes, Leticia Teixeira 2-705 Richardson, Adam 2-217
Menges, Achim 1-549, 2-165, 2-237 Roccasalva, Giuseppe 1-41
Miltiadis, Constantinos 1-517 Rolando, Andrea 2-315
Mitani, Jun 1-693 Roman, Miro 2-405
Mohammed-Amin, Rozhen K. 1-467 Romão, Luís 2-677
Moleta, Tane 1-507 Roxo, Ana Filipa 1-265
Moloney, Jules 1-507 Rua, Helena 1-265
Morales Beltran, Mauricio 2-117 Rumpf, Moritz 1-643
Mostafavi, Sina 2-117 Russell, Peter 2-485
Moya, Rafael 2-69
Mueller, Volker 2-39, 2-185 S
Mumovic, Dejan 2-79 Saey, Rik 2-505
Munkel, Georg 2-545 Salim, Flora 2-69
Sánchez de León, Michelle 1-31
N Sariyildiz, Sevil 1-357, 2-137
Narahara, Taro 1-673 Schaffranek, Richard 1-347
Nicknam, Mahsa 2-595 Schaumann, Davide 2-59
Nielsen, Stig Anton 1-413 Schindler, Christoph 1-605
Noble, Doug 1-69 Schloz, Manuel 1-549
Nourian, Pirouz 1-357 Schmitt, Gerhard 1-41, 1-255, 2-195, 2-375
Nováková, Kateřina 1-213, 2-555 Schneider, Sven 1-109, 1-487
Novembri, Gabriele 2-29 Schwartz, Mathew 2-365
Nunes, Mário 1-403 Schwinn, Tobias 1-549
Semlali, Anis 2-325
O Serdoura, Francisco Manuel 1-439
Ohshima, Taisuke 1-693 Sevtsuk, Andres 1-655
Okabe, Aya 2-289 Shah, Mahnaz 1-329
Okuda, Shinya 1-625 Sharaidin, Kamil 2-69
Oprean, Danielle 1-101 Simeone, Davide 2-59
Orbey, Betül 1-167, 2-415 Sokmenoglu, Ahu 1-301
Orsi, Francesco 1-439 Song, Juha 1-575
Ortiz, Christine 1-575 Sonmez, N. Onur 1-301
Oxman, Neri 1-585 Stavrakantonaki, Marina 2-535
Ozel, Filiz 1-127 Steinø, Nicolai 1-195
Özkar, Mine 1-195, 2-305 Stojanovic, Djordje 1-633

Stoutjesdijk, Pieter 1-719 Vidmar, Jernej 1-311
Strobbe, Tiemen 2-185, 2-627 von Both, Petra 2-525, 2-587
Sun, Yimin 2-645 von Mammen, Sebastian 1-467
Su, Ping 2-645 Vrouwe, Ivo 1-703
Symeonidou, Ioanna 1-683
W
T Waimer, Frédéric 1-549
Tabatabai, Ali 1-605 Wamelink, J. W. F. (Hans) 2-475
Taguchi, Masaharu 1-219 Wang, Pin-Hung 2-21
Takenaka, Tsukasa 2-289 Weigele, Jakob 1-549
Tamke, Martin 1-605 Welzner, Ivo 1-175
Tanaka, Hiroya 1-693, 1-711 Weston, Mark 1-569, 2-147
Tapias Pedraza, Estefania 1-41 Wilkinson, Samuel 2-39
Tenpierik, Martin 2-137 Williams, Mani 2-69
Teuffel, Patrick 2-49 Wurzer, Gabriel 2-279
Thompson, Emine Mine 1-205
Tidafi, Temy 2-325 X
Tolba, Osama 1-459 Xiong, Lu 2-645, 2-725
Tomé, Ana 1-403 Xiong, Wei 2-725
Tompson, Tim 1-233 Y
Trento, Armando 2-385 Yazar, Tuğrul 2-247
Treyer, Lukas 2-195 Yazici, Sevil 2-127
Tunçer, Bige 2-375 Yıldırım, Miray Baş 1-195
Turrin, Michela 1-175, 2-137 Yoshida, Hironori 1-665
V Z
Van Campenhout, Jan 2-627 Zajíčková, Veronika 2-515
van Swieten, Peter 1-703 Zarzycki, Andrzej 2-425
Vardouli, Theodora 1-243 Zhang, Hongxia 2-725
Vasku, Michael 1-367 Zolotovsky, Katia 1-575
Velasco, Rodrigo 1-117 Zreik, Khaldoun 2-605
Veliz, Felipe 1-575 Zünd, Daniel 1-255
Verbeke, Johan 1-423 Zupančič, Tadeja 1-227

Volume 2
This is the second volume of the conference proceedings of the 31st eCAADe conference, held
from 18–20 September 2013 at the Faculty of Architecture, Delft University of Technology, Delft,
The Netherlands.
The theme of this conference is the role of computation in the consideration of performance in
planning and design.
Since long, a building no longer simply serves to shelter human activity from the natural envi-
ronment. It must not just defy natural forces, carry its own weight, its occupants and their pos-
sessions, it should also functionally facilitate its occupants’ activities, be esthetically pleasing,
be economical in building and maintenance costs, provide temperature, humidity, lighting and
acoustical comfort, be sustainable with respect to material, energy and other resources, and so
forth. Considering all these performance aspects in building design is far from straightforward
and their integration into the design process further increases complexity, interdisciplinarity
and the need for computational support.
One of the roles of computation in planning and design is the measurement and prediction of
the performances of buildings and cities, where performance denotes the ability of this built
environment to meet various technical and non-technical requirements (physical as well as psy-
chological) placed upon them by owners, users and society at large.
eCAADe — the association for Education and research in Computer Aided Architectural Design
in Europe – is a non-profit making association of institutions and individuals with a common
interest in promoting good practice and sharing information in relation to the use of computers
in research and education in architecture and related professions. eCAADe was founded in 1983.
All papers of these proceedings are accessible via CumInCAD (cumincad.scix.net).
ISBN: 978-94-91207-05-1

Ecaade2013 Vol 2 Lowres

Uploaded by

Copyright:

Available Formats

Ecaade2013 Vol 2 Lowres

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ecaade2013 Vol 2 Lowres

Uploaded by

Copyright:

Available Formats

Education and research in Computer Aided Architectural Design in Europe

Volume 2 - Computation and Performance - eCAADe 31 | 1

1st Edition, September 2013

Computation and Performance – Proceedings of the 31st International Conference on Educa-

ISBN 978-94-91207-05-1 (eCAADe)

Publisher: eCAADe (Education and research in Computer Aided Architectural Design in

Cover design: Bige Tunçer and Rudi Stouffs

2 | eCAADe 31 - Computation and Performance - Volume 2

Proceedings of the 31st International Conference on Education and research in

18-20 September 2013

Volume 2 - Computation and Performance - eCAADe 31 | 3

Computation and Performance

Rudi Stouffs and Sevil Sariyildiz

Volume 2 - Computation and Performance - eCAADe 31 | 5

6 | eCAADe 31 - Computation and Performance - Volume 2

Volume 2 - Computation and Performance - eCAADe 31 | 7

eCAADe 2013 Conference Chairs

eCAADe 2013 Organizing Committee

8 | eCAADe 31 - Computation and Performance - Volume 2

Sherif Abdelmohsen, Ain Shams University, Egypt Switzerland

Volume 2 - Computation and Performance - eCAADe 31 | 9

10 | eCAADe 31 - Computation and Performance - Volume 2

Volume 2 - Computation and Performance - eCAADe 31 | 11

19 Simulation, Prediction and Evaluation

Contents - Volume 1 - Computation and Performance - eCAADe 31 | 13

153 Generation, Exploration and Optimisation

215 Algorithmic Design Generation

14 | eCAADe 31 - Computation and Performance - Volume 1 - Contents

333 Models of Computation: Human Factors

Contents - Volume 1 - Computation and Performance - eCAADe 31 | 15

463 Building Information Modelling

16 | eCAADe 31 - Computation and Performance - Volume 1 - Contents

615 Shape Grammars

Contents - Volume 1 - Computation and Performance - eCAADe 31 | 17

735 Index of Authors

18 | eCAADe 31 - Computation and Performance - Volume 1 - Contents

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 19

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 21

22 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 23

24 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 25

26 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 27

28 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Fire engineering simulation for a safe, innovative and high-performance

Antonio Fioravanti1, Eolo Avincola2, Gabriele Novembri3

Abstract. Architecture, nowadays, is an even more demanding activity in which

FIRE AND ARCHITECTURAL DESIGN

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 29

30 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

FIRE SAFETY PERFORMANCE APPROACH

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 31

32 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 33

34 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 35

36 | eCAADe 31 - Computation and Performance - Volume 2 - Simulation, Prediction and Evaluation

Simulation, Prediction and Evaluation - Volume 2 - Computation and Performance - eCAADe 31 | 37

Abstract. A novel approach is presented to predict wind pressure on tall buildings