IKI23 Klein PDF

Bauhaus Summer School
023 ”Model Validation and Simulation”

Graduate Courses for Structural Engineering
Applications
Schriftenreihe des Instituts

für Konstruktiven Ingenieurbau
Bauhaus-Universität Weimar
INSTITUT FÜR KONSTRUKTIVEN INGENIEURBAU
“Model Validation and Simulation”

Graduate Courses for Structural
Engineering Applications
6th – 17th August 2012

Content
7 Foreword
WITT, Karl Josef
9 Course Description
WERNER, Frank; SCHWARZ, Jochen
13 Guest Lecturer
14 Numerical modelling of non linear constitutive laws

Boso, Daniela
15 Modeling of composite steel – concrete bridges

VAYAS, Ioannis
16 Evaluation Methods for Prediction Quality of Coupled Partial Models

KeiteL, Holger
17 Unsaturated soils – Fundamentals and Overview of Applications

TRIPATHY, Snehasis
18 Advanced transmitting boundary conditions for large-scale SSI

problems and applications on parallel computing platforms
GENES, M. Cemal
20 Seismic design of tall and slender structures including rotational

components of the ground motion: EN 1998-6 approach
BONEV, Zdravko, VASILEV, G.
22 Vibration serviceability of footbridges

VAN DEN BROECK, Peter, VAN NIMMEN, Katrien
23 Analysis and design of non-conventional cold-formed steel structures
DUNAI, László
25 Asynchronous ground excitation and soil-structure interaction in

the design of long bridges
SEXTOS, Anastasios G.
26 An analytical perspective for performance limits for structural walls

GÜLKAN, Polat
27 Earthquakes in developed mountainous regions

Laue, Jan
28 Participants
29 Vibration Based Post-Earthquake Damage Assessment of Structures

ANDIC, Halil Ibrahim, GUNES, Burcu, GUNES, Oguz
38 Fragility of Shear Wall Buildings with Torsional Irregularity

AKANSEL,Vesile Hatun, GÜLKAN, Polat, YAKUT, Ahmet
49 Vibration monitoring and structural identification of an historic

masonry bell tower
CABBOI, Alessandro
63 Knowledge based structural assessment of Mallorca cathedral

ELYAMANI, Ahmed, CASELLES, Jose Oriol, CLAPES, Jaim, ROCA, Pere
77 Seismic Strengthening of Masonry Infilled Reinforced Concrete Frames

GRUBIŠIĆ, Marin , SIGMUND, Vladimir
87 Time dependent analysis of steel-concrete composite beams

HEGYI, Péter
99 Multi-filament yarns testing for textile-reinforced concrete

KADĚROVÁ, Jana, VOŘECHOVSKÝ, Miroslav
105 Seismic response of short columns affected by brick / tile aggregate

concrete wall infills
KRAUS, Ivan, MILIČEVIĆ, Ivana
114 Experimental research on thermo-creep behavior of soft

sedimentary rock
KURIMOTO, Yuhei, ZHANG, F., NISHIMURA, Tomohiro, Xiong, Y.
120 Bridge-Wizard: Expert system for FE modeling and post —
processing of Bridges
Lesgidis, Nikolaos D.
132 Experimental and numerical assessment of the dynamic behaviour

of a railway viaduct with precast deck
MALVEIRO, Joel, RIBEIRO, Diogo, CALÇADA, Rui
144 Running Safety Evaluation of Trains Moving Over Bridges shaken

by Earthquakes
MONTENEGRO, Pedro Aires, Calçada, Rui, Vila Pouca, Nelson
158 Numerical and experimental study of the vertical interfaces

behaviour of interconnected structural masonry walls
OLIVEIRA, Luciane Marcela Filizola, CORRÊA, Márcio Roberto Silva
168 Low cycle performance of T-stub components of bolted moment

beam-to-column connections
POP, Ana-Maria, GRECEA, Daniel, CIUTINA, Adrian
174 Dynamic safety assessment of a small span high-speed

railway bridge
ROCHA, João Miguel, Henriques, António Abel, Calçada, Rui
187 Comparative study of steel frame modelling levels and Eurocode

based design methods
TÓTH, Adrienn, JOÓ, Attila
197 Assessment of structure-tuned mass damper assemblies in an

earthquake environment
TRIBUTSCH, Alexander, ADAM, Christoph
206 Performance-based evaluation of seismic demands for moment

resisting steel frames using N2-method
TSONEV, Aleksandar, BONEV, Todor, KOLEVA, Tatyana
216 Post-Earthquake Damage Screening of Steel Buildings using

Model-Based Damage Pattern Classification
YAMAGUCHI, Mayako, KURATA, Masahiro, NAKASHIMA, Masayoshi
227 Structural Solutions for the Construction of National Beverages

Company Warehouse (NBC) Project
ZIMMO, Iyad
238 Modelling of soil-structure interaction in alpine regions

MARIN, Alexandru, LAUE, Jan, MEZGER, Florence
251 Fatigue Analysis of a Steel Bridge Included in a High-Speed
Railway Line
ROCHA, João Francisco, DELGADO, Raimundo, CALÇADA, Rui
263 Classification of timber–steel connections and the behaviour of

semi rigid joints
GEČYS, Tomas
272 Numerical simulations of pseudo-dynamic full-scale test on a three

story - one span - three bays steel frame
IOAN, Adriana Mirela
281 Micro and macro modeling of internal erosion and scouring with
fine particle dynamics
KONDO, Akihiko, MAEDA, Kenichi, WOOD, David Muir
289 Reports
290 Project 1 — System Identification, Model Updating and Simulation
303 Project 2 — Soil-Structure Interaction Simulation and Experimental

Validation
309 Project 3 — Evaluation of seismic performance of a template design

RC school building before and after strengthening
319 Project 5 — Calibration of Numerical Models with Applications in

Civil Engineering
323 Project 6 — Long-span Bridge Aeroelasticity
345 Excursion Report — City tunnel Leipzig 11th August 2012
350 Bauhaus Summer School

Foreword 7
Foreword
The past decade has seen a rapid development of numerical meth-

ods and sophisticated modelling in different areas of civil engineering,
especially in constructions for a sustainable environment. Both sci-
entific approaches and practical solutions using recent developments
of data acquisition and data management related to structural engi-
neering and earth sciences are of major interest. Advanced education
has to consider that most of the challenging future constructions will
require further theoretical and practical developments in a close col-
laboration of engineering practice and research. The Bauhaus Summer
School series provides an international forum for such an exchange of
methods and skills related to the interaction between different disci-
plines of modern engineering science.
The 2012 civil engineering course was held in August over two weeks
at Bauhaus-Universität Weimar. The overall aim was the exchange
of research and modern scientific approaches in the field of model
validation and simulation between well-known experts acting as lec-
turers and active students. Besides these educational intentions the
social and cultural component of the meeting has been in the focus.
48 graduate and doctoral students from 20 different countries and
22 lecturers from 12 countries attended this summer school. Among
other aspects, this activity can be considered successful as it raised the
sensitivity towards both the significance of research in civil engineer-
ing and the role of intercultural exchange.
This volume summarizes and publishes some of the results: abstracts

of key note papers presented by the experts and selected student
research works. The overview reflects the quality of this summer school.
Furthermore the individual contributions confirm that for active stu-
dents this event has been a research forum and a special opportunity
to learn from the experiences of the researchers in terms of methodol-
ogy and strategies for research implementation in their current work.
On behalf of the Faculty of Civil Engineering, Bauhaus-Universität
8 Foreword
Weimar, I would like to thank everyone for their contribution to the

scientific content, the social events, the organization and the realiza-
tion of the Bauhaus Summer School 2012. Particular thanks to all the
lecturers who spent part of their annual leave to come to Weimar,
to Frank Werner and Jochen Schwarz as the initiators, to Lars Abra-
hamczyk and Daniela Raddi for the excellent organization. In addition,
the Faculty of Civil Engineering is extremely grateful for the sponsor-
ship from German Academic Exchange Service (DAAD). The successful
Bauhaus Summer School 2012 and this book would not have been
possible without this generous financial support.
Weimar, April 2013

Karl J. Witt, Dean of the Faculty
Course Description 9
Course Description
Introduction WERNER, Frank and

SCHWARZ, Jochen
Bauhaus Summer School has a long tradition that goes back to a
series of Summer Academies, “Advanced Studies — Structural Engi-
neering and CAE”, offered by the Faculty of Civil Engineering. From
the early beginning of these courses, it was agreed among the con-
tributing departments, lecturers and supervisors that on the basis of
demanding engineering projects an advanced level of qualification
should be ensured. The experiences from previous summer schools
at the Bauhaus-Universität Weimar, conducted in the years from 1998
until 2008, show that there is an increasing demand in training fields
between studies and business requirements.
In 1998, twenty students participated in the first summer school. In

2008, 122 international students applied for the participation in the
summer school. In 2012, 48 students from 20 different countries
could be admitted (Figure 1). Since 2011, the Summer School has been
supported by the Lifelong Learning Programme. Its innovative charac-
ter results from the ambitious engineering tasks derived from current
research topics by the participating institutions, the intensive assis-
tance, and the project related training of simulation and modeling
techniques introduced by key lectures in civil engineering.
The acquired knowledge is applied to scientific and practically relevant pro-

jects within a compact course. Such a “hands-on approach” is commonly
not included in regular teaching programmes. Several targets harmonically
overlap. The students will be trained in the use of modern software tools,
design concepts and their practical applications. The course targets post-
graduate students (i.e. second cycle (M.Sc.) and third cycle (PhD)) from
different institutions and countries to enable knowledge transfer between
the participants. The interdisciplinary teamwork in the projects com-
bined with an intensive assistance by the academic staff shall enable the
students to prepare and present a final presentation of the project results.
Targets and Objectives
Different language skills of the participants pose high demands on the

management abilities and will train the ability to work in a team. The cur-
rent state-of-the-art in the above mentioned fields will be presented to
the participants during the lectures. Recent developments in Model Vali-
dation and Simulation will be added by additional presentations of the
partner institutions or other invited experts. Special topics from the men-
tioned fields will be dealt with during the project work in small groups.
The young scientists will be offered the possibility to exchange modern

scientific knowledge in a realistic training setting. Towards the end of
the course, students are requested to present and discuss their own
work in front of a broad scientific audience. The results from several
national and international projects of the Center for Structural Dynamics
and Earthquake Engineering of the Bauhaus-Universität Weimar are
included in the programme.
The management and presentation skills of the participants will be improved

and a training opportunity will be provided to share the work and expe-
rience in a team. The participants are invited to find the value of new part-
nerships and creative networking by the participating institutions (Fig. 2).
Figure 1 — Home countries

of the Participants of summer
course
Figure 2 — European Project

Partners
Course Description 11
Current Course Topics
At the national level, no similar workshop to the International Summer

Academy “Model Validation and Simulation” is known. The relevance
of the course topics themselves is undisputable; model validation and
simulation play a central role in the work of civil and structural engi-
neers. In practice day-by-day, engineers deal with models support-
ing the solution specific design problems. Because of the increasing
demand of sophistication, more and more structures are designed by
groups of engineers from different countries.Different substructures or
materials with different properties are often combined or interacting
(e.g. soil and structure, structures and components). Due to this, mod-
els have to be validated by measurements, experimental testing and sim-
ulations. Therefore, the course topics include hybrid methods of model
calibration using instrumental testing methods.
Experimental facilities are used or being implemented for providing the

database for the numerical reinterpretation. Excursion to building sites
and leading companies illustrate the practical link between advanced
modeling and simulation techniques and qualified structural solutions.
The transfer of real existing building data into different basic struc-
tural models and the comparison of their robustness with respect to
increasing abstractness and reduced dimensionality are major points
of discussion. The effect of model assumptions will be studied on the
basis of different software tools, including considerations of the uncer-
tainty and the scatter of building response quantities.
Referring to recent code developments, participants are requested

to present national design practice and experience from applications,
focusing on the efforts that contribute to new concepts, evaluation
strategies and the quality assurance of structural models. In general,
the Summer School combines theoretical and practical tasks. The par-
ticipants are invited to use modern analyses methods in the field of
structural engineering and dynamics.
By providing students with advanced, scientifically-based interdisci-

plinary knowledge, skills and methods, they are trained to react to
demanding engineering tasks in the areas of planning, construction
and realization of structures under specific impact and action condi-
tions. They are also able to carry out site- or structure-specific risk
analyses using modern tools for gauging the threat of extreme load-
ings. In order to structure and reflect the complexity of the chain of
reactions inherent to the impact side, various disciplines and engi-
neering-related areas of the natural sciences are explored.
The Bauhaus Summer School examines the central role that civil engi-
neering plays in lessening the impact of extreme loadings e.g. natural dis-
asters and focuses on the engineering methods that can be used to assess
and reduce the vulnerability of buildings and their structural systems.
The need of hazard and impact–resistant structural systems requires

specific and innovative solutions in cases where extreme loads from
different hazard types have to be sustained, including concepts to com-
bine conventional engineered systems (R.C., steel) with locally available
materials and construction types. By offering international projects the
programme provides insight to the challenges from a global perspective.
The modular structure of the programme allows the participants to address

current and trendsetting problems and research topics. Key aspects of the
Summer School are linked to the Research Training Group 1462 (“Model
Validation in Structural Engineering"), which is involved with the pro-
ject “Calibration of Numerical Models in Civil Engineering Applications”.
The topics of the 2012 course highlight the broad spectrum of engi-
neering tasks in the following fields of modeling and simulation:
–– Structural Engineering (incl. requirements on structural models,

model verification, validation and updating);
–– Geodynamics (incl. fundamentals in soil dynamics, analytical and
numerical strategies, concepts of soil-structure interaction, dynamic
laboratory and field testing);
–– Earthquake Hazard and Risk engineering (incl. Description of
seismic action, engineering approaches of seismic hazard assessment,
Interpretation of earthquake damage cases, Recent projects and
risk studies);
–– Simulation and Testing of Steel and Glass Structures (incl. Computa-
tional models for linear and nonlinear analysis, Practical demonstra-
tion and simulation, Finite element modeling and analysis of stresses);
–– Calibration of Numerical Models in Civil Engineering Applications
(incl. Calibration of models as optimization, parameter identification
of a dynamically driven bar and calibration of a multifield model
of a gravity dam);
–– Long-Span Bridge Aeroelasticity (incl. Dynamic analysis of long-span
cable-supported bridges, realistic simulation of the structural
behaviour and Models for Predicting Wind-induced Vibrations of
Long-span Bridges);
The Summer School is divided into two parts. The first part concen-
trates on the theoretical fundamentals and the current state of research
related to projects and special lectures from members of partner insti-
tutions. An introduction to specific research themes of all partners will
be offered through special lectures.
The second part focuses on interdisciplinary team work within the pro-
ject groups under intensive supervision. Following a pre-selection and
review of the submitted papers, participants are invited to present
the advanced state of their research in Special Theme Sessions. The
outcome — presented by the herein compiled reports — reaches an
impressive level of results and graphical elaboration.
Guest Lecturer
14 Numerical modelling of non linear constitutive laws
Numerical modelling of non

linear constitutive laws
BOSO, Daniela Abstract

University of Padova
In the past, linear constitutive laws were mostly considered for the
modelling of material behaviours. However, during last decades there
has been a definite trend toward a better understanding and conse-
quently to an enhanced exploit of material performances. Non-linear
constitutive laws are very common, one can immediately think of
yielded steel or reinforced concrete. Furthermore, the modern tech-
nology gives the opportunity to develop advanced, knowledge-based
materials, to use features and qualities of the individual components
for an optimal performance.
Composite materials are more and more applied in engineering prac-

tice. They allow to take advantage of the different properties of the
initial materials, of the geometric structure and of the interaction
between the constituents to obtain a tailored behaviour as a final result.
Therefore, it seems clear that modern engineers will have to deal with
the modeling and verification of nonlinear materials and structures. In
these cases, the constitutive law in closed form is not always obtaina-
ble. When this is not possible, numerical methods and computational
approaches allow likewise for a detailed description of the non linear
behaviour. In this lecture some of the approaches used for computa-
tional modeling will be presented, such as virtual testing, and asymp-
totic homogenization. In addition, some soft computing techniques,
based on artificial neural networks, will be illustrated as they are a
powerful and promising tool both for modeling of composite materials
and improvement of computational efficiency.
Modeling of composite steel — concrete bridges 15
Modeling of composite
steel — concrete bridges
Abstract VAYAS, Ioannis

National Technical University
of Athens
The lecture presents in eight sections modeling alternatives for com-
posite bridges. Section 1 shows the current types of composite bridges
that include plate girders, box girders, trusses, arches and cable-stayed
bridges. Section 2 shows models for global analysis appropriate for
the determination of internal forces and moments, deformations and
vibrations. It includes the presentation of beam models, grillage mod-
els and 3-D models with the relevant field of application, their ben-
efits, drawbacks and limits. Section 3 introduces cracking of concrete
slab, the extent of cracked regions and how they are modeled. Sec-
tion 4 deals with the effective width in wide flanges due to shear lag
and shows methods for its determination in concrete and steel flanges.
Section 5 deals with lateral stability of girders that is of importance
specifically during construction stages. Section 6 indicates the effects
of bracings both in transverse direction and in plan, with examples
in curved bridges and at the service stage. Section 7 introduces the
effects of the rheological behavior of concrete. It shows the determi-
nation of creep shrinkage and their introduction in analysis and design.
Finally section 8 show Finite Element models that are appropriate for
local analyses and for validation of simpler models.
16 Evaluation Methods for Prediction Quality of Coupled Partial Models
Evaluation Methods for

Prediction Quality of
Coupled Partial Models
KEITEL, Holger Abstract
The process of analysis and design in structural engineering requires

the consideration of different partial models of loading, structural
material, structural elements and analysis type, among others. The
various partial models are combined by coupling of their several com-
ponents. Due to a large number of available partial models describing
similar phenomena many different model combinations are possible
to simulate the same quantities of a structure. The challenging task of
an engineer is to select a model combination that ensures a sufficient
reliable prognosis. In order to achieve this reliable prognosis of the
overall structural behavior, on the one hand a high individual quality
of the partial models and on the other hand an adequate coupling of
the partial models is needed.
Therefore, this lecture gives an introduction into the field of model

quality evaluation. First, the basics of partial models and their coupling
are given and the major modeling techniques — inverse modeling,
stochastic modeling, adaptive modeling, and multiscale modeling —
are presented. Further, several model properties are explained that
allow for a quantitative model comparison. In order to give an exam-
ple of model quality evaluation the field of stochastic modeling is
shown more detailed and this technique is applied for the evaluation
of concrete creep models. The benefits of choosing the most appro-
priate creep model are demonstrated. Finally, the prediction quality
of several coupled partial models is exploited by means of sensitivity
analysis and graph theory using an example of a bridge structure.
Unsaturated soils — Fundamentals and Overview of Applications 17
Unsaturated soils —
Fundamentals and Overview
of Applications
Abstract TRIPATHY, Snehasis

School of Engineering, Cardiff
University, UK
The main focus of the lecture will be to create an awareness concern-
ing the role of negative pore-water pressure (suction) on the engi-
neering behaviour of soils. The presentation will cover an introduction
to saturated and unsaturated soil mechanics, examples of classical
Geotechnical and Geoenvironmental engineering problems, roles of
pore-water on the engineering behaviour of soils with aids of video
demonstration of simple laboratory-scale experiments, definitions of
suction, need for a saturated / unsaturated mechanics of soils, how to
measure and induce suction in soils via video demonstration of some
laboratory tests. The presentation will close by emphasizing the appli-
cations of suction on engineering behaviour of unsaturated soils.
Advanced transmitting boundary conditions for large-scale
18 SSI problems and applications on parallel computing platforms
Advanced transmitting boundary

conditions for large-scale SSI
problems and applications on
parallel computing platforms
GENES, M. Cemal Abstract

Mustafa Kemal University,
Civil Engineering Department,
The dynamic response of massive, stiff, and embedded structures in
31200 Iskenderun/Hatay,
Turkey relatively soft soil conditions such as nuclear reactors, tunnels, liq-
uid-storage tanks, gravity dams, and high-rise buildings are affected
by soil-structure interactions (SSI) as well as the dynamic character-
istics of the exciting loads (i.e. earthquake, wind, explosion, machin-
ery vibrations) and the structures. The effect of the SSI may alter the
dynamic characteristics of the stiff, rigid and massive structures resting
on or embedded in relatively soft and / or layered media significantly.
Soil region, which in many cases presents a complex and non-lin-

ear behaviour, can be discretized using the Finite Element Method
(FEM), the Boundary Element Method (BEM) or hybrid models (Fig-
ure 1). All of these have advantages and disadvantages depending on
the applied problem.
In this course, two finite element and boundary element based SSI
models will be presented.
The first model: A finite element (FE) based SSI model and its paral-
lelized algorithm for applications on distributed systems will be pre-
sented. The SSI model is established by combining two methods: the
Consistent Infinitesimal FE Cell Method (also referred to as ‘the scaled
boundary-finite element method’) proposed by Wolf and Song (Finite-
element modelling of unbounded media, Wiley, England, 1996) for
modelling the soil region extending to infinity (far-field), and the
standard FE for the finite region (near-field) and the structure. By using
this combined model, a computer program for harmonic and transient
analyses of soil-structure systems is coded. In order to decrease the
computation time and achieve the solution of large-scale problems,
the model is parallelized. As a result of this parallel solution, signifi-
cant time is saved for large-scale problems.
Advanced transmitting boundary conditions for large-scale
SSI problems and applications on parallel computing platforms 19
The second model: a coupled model based on finite element method

(FEM), boundary element method (BEM) and scaled boundary FEM
(SBFEM) for dynamic response of 2D structures resting on layered soil
media will be presented. The SBFEM and BEM are used for modelling
the dynamic response of the unbounded media (far-field). The stand-
ard FEM is used for modelling the finite region (near-field) and the
structure. The objective of the development of this coupled model is
to combine advantages of above-mentioned three numerical models
to solve various SSI problems efficiently and effectively. These three
methods are coupled (FE–BE–SBFEM) via substructuring method, and
a computer programme is developed for the harmonic analyses of SSI
systems. The results of the proposed model agree with the results pre-
sented in the literature for the chosen problems. The advantages of
the model are demonstrated through these comparisons.
For this model also, an algorithm for a parallelized FE–BE–SBFEM for

harmonic and transient dynamic response of large-scale 2D structures
embedded in or on layered soil media will be presented. The objec-
tive of the development of this parallelized coupled model is to use
the power of high performance computing, and to take into account
the advantages and evade the disadvantages of the above mentioned
numerical methods for modelling of the unbounded media in SSI
systems. The development of the parallel algorithm for this model is
essential for solving arbitrarily shaped large-scale SSI problems, which
cannot be solved within reasonable elapsed times by a serial algo-
rithm. The efficiency of the proposed parallel algorithm and the validity
of the coupled model are shown by means of several numerical exam-
ples, indicating the excellent accuracy and applicability of the parallel
algorithm with considerable time-savings in large-scale problems.
Figure 1 – Concept of the SSI

Problem
Seismic design of tall and slender structures including
20 rotational components of the ground motion: EN 1998-6 approach
Seismic design of tall and slender

structures including rotational
components of the ground
motion: EN 1998-6 approach
VASILEV, G. Abstract
University of Architecture,
Civil Engineering & Geodesy,
First part of the study deals with implementation of an advanced
Faculty of Civil Engineering,
Bulgaria model of seismic action — ground accelerations, which allow for
taking into account the spatial variability of seismic action. Accord-
BONEV, Zdravko ing to this approach for certain types of structures (tall and slender
University of Architecture, structures, long in plan and irregular structures) inclusion of transla-
Civil Engineering & Geodesy,
tional acceleration components only into the analysis is not sufficient
Bulgaria to reach safe design solution. For this kind of structures the use of
advanced modeling of seismic action requires inclusion of rotational
components of the ground motion (rotational accelerations). The well
known model of seismic action containing three translational accelera-
tion components, now should be upgraded by adding three rotational
components of accelerations. All ground acceleration components
are represented by their response spectra. Looking then to response
spectrum method and linear analysis, it should be generalized and
extended to be capable of evaluating the mixed spectral response
derived as a result of translational / rotational components.
As a result of the analysis it is proven, that the action effects generated

by mixed translation — rotational seismic input are greater than the cor-
responding action effects resulting from translational seismic input only.
The general idea is that the missing of rotational components contribu-
tion leads to unfavorable type of loading and unsafe design solutions.
The general purpose of the second part of the study is the evaluation
of seismic soil-foundation-structure interaction effects on overall seis-
mic behavior. Seismic input is assumed to be free field motion of the
site, being represented by ground accelerations in Fourier power series
format. The analysis is carried out in time domain, assuming linear
behavior of the structure and soil. Upper structure, on one hand, and
Seismic design of tall and slender structures including
rotational components of the ground motion: EN 1998-6 approach 21
foundation-soil system on the other hand, are considered separately

by means of substructure approach assuming steady-state response
under sine signal with specific external frequency.
Steady state response of the foundation-soil system is represented

through foundation impedances. They are input frequency depend-
ent. Foundation impedances are implemented as a system of spring
and dashpot elements (Kelvin–Vogt model). Impedances are used to
connect the structure with the foundation, forming the flexible base
boundary conditions. Impedance functions for rigid foundations and
elastic homogeneous half space are available from the literature as
typically classic case. For specific soil conditions such as layered or
discontinuous media impedances should be found in advance using
boundary element method as more general and numerically efficient
computational technology.
As a result of the research a general algorithm for approximate eval-

uation of soil-foundation-structure interaction effects is provided.
Dynamic amplification factor and amplitude magnification are studied
as being dependent on input frequency. A comparative analysis with
fixed-base supporting conditions is carried out. Numerical results con-
cerning separate steps of the computational process are graphically
illustrated and discussed.
22 Vibration serviceability of footbridges
Vibration serviceability
of footbridges
VAN DEN BROECK, Peter Abstract

KU Leuven, Belgium
Predicting the dynamic performance of civil engineering structures due
VAN NIMMEN, Katrien
KU Leuven, Belgium to crowd induced loading is an increasingly critical aspect of the vibra-
tion serviceability design process. Due to the increasing strength of
materials and the trend towards greater slenderness, modern pedestrian
bridges are very often lively structures prone to human induced vibrations.
This presentation will discuss the current codes of practice (the European
guideline HiVoSS [1–2] and the French guideline Setra [3]) that provide a
methodology to assess the dynamic behaviour at design stage.
A slender steel footbridge (Eeklo, Belgium) is applied as case study.

The operational modal analysis is performed to obtain the operational
modal characteristics of the footbridge. These results are applied for
the updating procedure of the numerical model (FE model) of the
footbridge, improving the accuracy of the numerical predictions of the
dynamic response of the bridge to pedestrian loading. The vibration
serviceability check according to the current codes of practice is per-
formed for both the initial versus the updated model of the footbridge.
In some cases this results into a different evaluation, indicating the
difficulty of making a reliable assessment of the dynamic performance
of a the footbridge in the design phase.
[1] HiVoSS, R. f. (2008). Design of footbridges.
[2] Butz, e. a. (2007). Advanced load models for synchronous pedestrian

excitation and optimised design guidelines for steel footbridges
(SYNPEX). Research Fund for Coal and Steel.
[3] Charles, P., & Hoorpah, W. (2006). Évaluation du comportement

vibratoire des passerelles piétonnes sous l'action des piétons. France:
Association Française de Génie Civil, Sétra / AFGC.
Analysis and design of non-conventional cold-formed steel structures 23
Analysis and design of non-

conventional cold-formed
steel structures
Abstract DUNAI, László

Budapest University of
Technology and Economics,
Due to the development in galvanized high strength sheet production and
Hungary
cold-rolling technology the geometrical shape and size of cold-formed
profiles significantly extended in the last two decades. The qualitative
and quantitative changes of the profiles provide with the opportunity to
use them in new functions, applying non-conventional structural solu-
tions. The new arrangements, however, result in complex structural
behavior which is not covered by standardized application rules.
Major fundamental and applied research activities have been com-

pleted on non-conventional cold-formed steel and composite struc-
tures at the Department of Structural Engineering of the Budapest
University of Technology and Economics under the supervision of the
author. The developed and investigated structures are as follows:
Cold-formed frames for smaller industrial buildings and for residential

houses; the structural elements are single and double — hollow-type —
cold-formed C-profiles, laterally not, or partially supported; the struc-
tural joints are self-drilling screwed or bolted connections of beams
and columns in web-to-web arrangement.
Composite floor beams with cold-formed C-profiles, trapezoidal

sheeting and partially drilled self-drilling screw shear connectors.
Cold-formed roof trusses; the structural elements are single and dou-
ble — hollow- and I-type — cold-formed C-profiles; the structural
joints are bolted web, flange or web / flange connections.
The above arrangements of the cold-formed structural elements and

the interaction with the eccentrically connected joints resulted in
complex stability behavior with interacting local plate buckling, dis-
tortional buckling and global flexural / flexural-torsional and / or lateral
torsional buckling, under compression and biaxial bending.
24 Analysis and design of non-conventional cold-formed steel structures
In the presentation the stability phenomena are discussed on the basis of

interacting experimental and numerical research. In the first phase of
the research fundamental experiments are completed on separated struc-
tural elements and joints. The stability behaviour modes are observed
and numerical and design models are developed and verified. These
results provided with a background for the design of complex structures.
In the next phase of the research the global and interacting stability
phenomena are investigated on full-scale specimens of cold-formed
frames, composite beams and trusses. By extending the tests using
nonlinear finite element analysis the complicated behavior modes are
characterized and applicable design methods are developed.
Acknowledgement: the research work is conducted under the finan-

cial support of the Hungarian OTKA T035147 and T049305 projects
and the Lindab Ltd.
Asynchronous ground excitation and soil-structure interaction in the design of long bridges 25
Asynchronous ground excitation

and soil-structure interaction in
the design of long bridges
Abstract SEXTOS, Anastasios G.

Aristotle University of Thes-
saloniki, Department of Civil
Although bridge structures might seem at a first sight as rather linear
Engineering
and simple structural systems, their actual performance under earth-
quake loading is more complicated than that of ordinary buildings,
because bridges have typically an order of magnitude larger overall
and cross-sectional dimensions, different energy absorption mecha-
nisms, more significant contribution of higher modes, while they are
most commonly crossing non-uniform soil profiles.
Notwithstanding the significant research progress made to date which

has already shed some light on many bridge engineering problems, the
development of a “realistic” earthquake motion scenario is still asso-
ciated with the highest relative uncertainty compared to maybe all
other design and construction aspects. This is even more pronounced
in the case of long bridges, where the variation of ground motion
among its supports in terms of arrival time, frequency content and
amplitude, can strongly affect both the pseudo-static and the dynamic
components of the system. The objective of the presentation is to dis-
cuss the recent findings on the impact of asynchronous (i.e.- spatially
variable) seismic excitation on the response of long bridges. The study
focuses on the recently constructed 780 km Egnatia highway in north-
ern Greece, and the 400 m cable-stayed Evripos bridge that connects

the Evia island to the Greek mainland.
Despite the long distance from the earthquake source, the simultane-
ous free-field and on-structure recordings reveal interesting patterns
of higher mode excitation and subsequent dynamic behavior. Based
on the above comparative studies, an effort is made to focus on the
dynamic response of the entire, interacting soil-structure system
under asynchronous excitation and to propose means for assessing in
advance the degree of its potential detrimental influence.
26 An analytical perspective for performance limits for structural walls
An analytical perspective
for performance limits for
structural walls
GÜLKAN, Polat Abstract

Çankaya University, Civil Eng.
Department, 06810, Ankara,
Proposed changes to modeling and acceptance criteria in seismic
Turkey
regulations for both flexure and shear dominated reinforced concrete
structural walls suggest that a comprehensive examination is required
for improved limit state definitions. This study uses a computational
tool to investigate the deformation measures defined in terms of plas-
tic rotations and local concrete and steel strains at the extreme fiber of
rectangular structural walls. I compare requirements in ASCE / SEI 41,
Eurocode 8 and the Turkish Seismic Code. This way, a critical evalua-
tion is possible for the requirements embedded in these documents. I
conclude that the performance limits must be refined by introducing
additional parameters. Recommendations are provided for Eurocode 8
and the Turkish Seismic Code.
Earthquakes in developed mountainous regions 27
Earthquakes in developed
mountainous regions
Abstract Jan Laue,

Institute for Geotechnical
Engineering, ETH Zurich
In geotechnical earthquake engineering, a subdivision is made into pri-
mary and secondary effects. Primary effects cover the behaviour of the
soil under cyclic loadings leading to an amplification of the incoming
waves. The term secondary effects umbrellas the effects of an earth-
quake to the surface like e.g. liquefaction, slope failures or rockfalls.
This contribution deals with the primary and secondary effects covered
by geotechnical earthquake engineering. Focus is given to moderate
events, which even though of magnitudes till 6 or 6.5 can have sig-
nificant effects in mountainous regions. Valleys consisting of deep soft
deposits and topographic effect can lead to high amplification of the
event and lengthen the earthquake in time. Soft deposits are prone to
liquefaction and slopes have in a static case already factors or safety of
around one. In addition to the physical description of these risks for
infrastructure and inhabitants, political issues like microzonation and
their effects are discussed.
Participants
Vibration Based Post-Earthquake Damage Assessment of Structures 29
Vibration Based Post-

Earthquake Damage
Assessment of Structures
Abstract ANDIC, Halil Ibrahim

Middle East Technical
University, Turkey
A large portion of Turkey’s geography and population is prone to
earthquake induced hazards and it is a priority need to take measures GUNES, Burcu
towards mitigating such hazards. In this manner, rapid damage assess- Atılım University, Turkey
ment of structures after earthquakes has a vital importance for initiat-
ing effective emergency response. However, post-earthquake damage GUNES, Oguz
Çankaya University, Turkey
assessments of structures are required a detailed investigation that
should be made by a professional engineer. These kinds of evaluations
take very long time after especially severe earthquakes, and while this
evaluation process delays, economic and social losses arising from
earthquake increase. Therefore, safe and rapid post-earthquake dam-
age assessment of structures is required. In parallel with this purpose,
as a non-destructive in-situ alternative technology, vibration based
assessment method provides an opportunity of condition assessment
in a manner of structural integrity with the help of dynamic response
which is recorded from structure. Especially, as it is possible to get
environmental vibration data from structure easily before and after
earthquake, researches on damage assessment methods which use
these data is getting more popular. In this study, methods that are
used to determine the structural damage assessment with using envi-
ronmental vibration data which is belong to before and after earth-
quake will be summarized, and the developing method for RC frame
structure will be introduced. An experimental study will be prepared
in order to load RC frame gradually up to its ultimate load bearing
capacity and simulate various damage levels. At each damage level,
damage condition will be determined with the help of acceleration
data which is obtained by both environmental vibration and impact
testing. As a pre-stage of the study, vibration properties of RC frame
for before and after damage scenario is determined with a compara-
tive numerical study, and damage is tried to determine by using Dam-
age Locating Vector ( DLV ) method.
Introduction
Structural health monitoring ( SHM) is the process of implementing a

damage identification strategy for civil infrastructures. Damage iden-
tification problem involves detection, localization and assessment of
the extent of damage in a structure so that its remaining life can be
predicted and possibly extended. SHM encompasses both local and
global methods of damage identification. The local methods include
visual inspections and non-destructive evaluation tools such as acous-
tic emission, ultrasonic, magnetic particle inspection, radiography and
eddy current. All these techniques, however, require apriori locali-
zation of the damaged zone and easy access to the portion of the
structure under inspection. As an alternative that overcomes these
limitations, global vibration based methods have been widely devel-
oped over the years ( [1], [2], [3], [4], [5], [6] ).
SHM based on vibration measurements involves temporal observa-

tion of a structure using periodically sampled vibration measurements,
extraction of damage sensitive features from these measurements and
assessment of the current health state / integrity of the system. The
basic premise of the vibration-based techniques is that the vibration
characteristics or the so-called modal parameters (frequencies, mode
shapes and modal damping) are functions of the physical properties
of the structure (mass, energy dissipation mechanisms and stiffness)
and changes in these physical properties cause changes in the modal
properties. This postulation, however, is compromised by the fact that
temperature changes, moisture and other environmental factors also
produce changes in dynamic characteristics. If the causes of changes
in dynamic characteristics other than damage are considered to be
noise in the measurement, then the changes due to damage must be
significantly larger than the noise in order for the techniques to work.
Usually four different levels of damage identification are studied [7]:
damage detection (Level 1), damage localization (Level 2), damage
quantification (Level 3), and prediction of the remaining service life
of the damaged structure, or the acceptable load level to reach the
intended service life (Level 4).
Damage Detection Methods
Most of the existing damage identification methods can be classi-

fied into two groups: model-based and non-model or feature-based
methods. The model-based methods are essentially model updating
procedures in which the mathematical model or the physical param-
eters of a structure is calibrated or updated using vibration measure-
ments from the physical structure ([8], [9]). Analytical sensitivities of
response parameters to changes in physical properties are used to
update modeling assumptions, physical sizing, elastic moduli, etc. The
feature-based approaches detect structural changes by detecting dam-

age features in the measured data without the need for an analytical
model of the structure. The main task here is the extraction of dam-
age features sensitive to structural changes so that damage can be
identified from the measured vibration response of civil engineering
structures. The following methods were proposed in the literature
for feature-based damage detection in civil engineering structures
([10], [4], [11], [12]):
–– Natural frequency based metrics

–– Mode shape based metrics
–– Structural damping based metrics
–– Modal strain energy based metrics
–– Flexibility based methods and other
matrix perturbation approaches
–– Pattern Recognition, neural networks
and other statistical approaches
–– Non-linear methods based on advanced
time-variant transforms
–– Other methods
The civil engineering community has been studying the vibration-

based damage assessment of bridge and building structures since the
early 1980s. Both model-based and non-model based approaches
utilizing the measured data in time-domain, frequency domain or
modal domain were investigated. While measurements were always
performed in the time domain, data could be analyzed in any of the
three domains. Although conversion between domains involves some
data compression, Friswell and Penny [13] argued that loss of informa-
tion during conversion was minimal for linear systems and that the
frequency domain may be more advantageous in reducing the effects
of random noise. Modal domain introduces further reduction of the
measured data since only the modes within a frequency band are con-
sidered. Friswell and Penny find this acceptable unless the out of band
modes are very close, i.e. the response is dominated by the in-band
modes. Lee and Shin [14] disagree with this argument pointing out
the fact that the modal data can be contaminated by modal extraction
error which the frequency response function data does not possess.
Vibration Based Modal Parameter Determination
Eigensystem Realization Algorithm with Observer Kalman Filter,

ERA-OKID, [15] which is used for defining the vibration based linear
system in the time-space is the most commonly used methods in
the literature. With this algorithm system is carried out and vibration
inputs (u) system outputs (y) are correlated with system matrices
(A, B, C, D) as seen in equation 1.
(1)
x Ax Bu
y Cx Du
The state matrix [A] obtained in discrete time is first transformed into
continuous time and then eigenvalues () and eigenvectors () of
this matrix in continuous time are calculated to obtain system’s natu-
ral vibration frequencies (ƒID), energy dissipation ratios ( ID) and ran-
domly scaled mode shapes ().
(2) i i i
2 2
ID i i
(3) f i
2
i
(4) i 2 2
i i
p
(5) C
Variable “p” in the equation 5 must be taken as 0,1 or 2 depends on

whether output data (sensor measurements) are velocity or accelera-
tion respectively. For the later evaluation steps, mass normalized mode
shapes of equation (5) may be required. It is possible to reach this type
of mass normalized mode shapes with using B which is obtained dur-
ing system realization [16].
Altuzay Algorithm [17] can be used in a condition of vibration inputs

are not known but system is accepted as stochastic. In this case, sys-
tem is represented with these equations.
x Ax w
(6)
y Cx v
In here, w and v are noise vectors, A and C are obtained system matri-
ces. System’s natural vibration frequencies, energy dissipation ratios
and randomly scaled mode shapes can be calculated by using equa-
tions (2–6). The point is, for a stochastic system that vibration inputs
are not known and matrix B has not been generated, a different method
is required to make the calculated mode shapes mass normalized [18].
Damage Localization
After the detection of damage in the structure, Damage Locating Vec-

tor — DLV method which is based on the flexibility difference due to
damage is used for damage localization. This method (see reference
[19] and [20] for further information) can be summarized like this:
Flexibility matrix can be calculated in the sensor locations with the

help of dynamic response data that obtained from the structure:
1
F n n
(7)
In the formulation, n is the matrix of mass normalized mode shapes

and  is the diagonal matrix that includes the square of corresponded
natural vibration frequencies. FU and FD matrices can be obtained
when the before and after damage data are used with equation (7)
respectively. Load distributions that create the exact deformations for
non-damaged and damaged systems create L matrix. Then, this equa-
tion can be written:
FD FU L DF L 0 (8)
The only condition that equation (8) is valid, except the condition
of there is no damage in the system which means flexibility matrix
difference is zero (DF = 0), occurs when DF which is obtained from
the L matrix’s flexibility matrices difference includes all vectors of the
null-space. If “Singular Value Decomposition (SVD)” is applied to the
flexibility matrix difference (DF ), load distribution which is called as
DLV can be obtained. When each DLV is applied to the non-damaged
system’s analytical model as a load distribution, deformations and
stresses corresponded to this statical loading are calculated. Theoreti-
cally, stress on a damaged member must be zero [19]. In practical, due
to both the noise during evaluation and model assumptions, this stress
is not zero and it comprises smaller values than the stated limit [20].
A Case Study: One-Story One-Bay Reinforced

Concrete Frame
Reinforced concrete frame type structures take an important place in

Turkey’s structural inventory. However, it is too hard to find a research
about this type of structure in the literature. Therefore, one-story
one-bay reinforced concrete frame (Figure 1) is selected to apply vibra-
tion tests. Accelerometers on beams and columns are indicated as B
and C respectively. Before the experimental study, in order to under-
stand the behavior, the frame is modeled in 3-D with MARC / Mentat
which is finite element software like in Figure 2a. In Figure 2b, dam-
aged area is indicated with dark color and thereby reducing the elas-
ticity modulus of this region with 25% and 50%, two different damage
scenarios are simulated to represent the post-earthquake situation.
Frame model is distributed with vibration input (u) as indicated in Fig-

ure 2 for pre-damage and two different post-damage scenarios. Accel-
eration responses are gathered from sensors locations on the frame.
Structure’s modal properties are determined thereby processing these
inputs with ERA-OKID algorithm. Natural vibration frequencies from
algorithm (ƒ id) and model (ƒ model) are presented in Table 1 with
comparison. First five modes are clearly determined, the frequency
difference between the analytical calculated and the obtained from

vibration input is limited in 1%, for further modes the difference reaches
at most 7%. As the contribution of these modes to the response is min-
imal, increasement in the observed error level is an expected result.
Figure 1 — RC frame test

mechanism: Geometry and
accelerometer locations
Figure 2 — 3-D MARC / Mentat

model of the frame:
(a) Pre-damage
(b) Post- damage
(a) (b)
Mode No Damage 25 % Damage 50 % Damage
ƒmodel ƒid Diff. ƒmodel ƒid Diff. ƒmodel ƒid Diff.

(Hz) (Hz) % (Hz) (Hz) % (Hz) (Hz) %
1 34.9 34.9 0.0 33.4 33.3 0.1 30.9 30.9 0.1

2 62.8 62.6 0.4 59.3 59.1 0.3 54.2 54.1 0.2
3 64.4 64.3 0.1 61.3 61.2 0.2 56.5 56.4 0.1
4 138.6 136.8 1.3 133.7 132.1 1.2 127.0 125.6 1.1
5 142.9 141.0 1.4 140.0 138.2 1.3 135.4 133.7 1.3
6 292.4 281.4 3.8 280.0 273.0 2.5 262.0 260.0 0.8
7 297.5 298.3 0.3 287.8 290.7 1.0 272.9 278.9 2.2
Table 1 — Comparison of
8 316.6 338.0 6.7 307.6 324.1 5.4 293.6 303.6 3.4
frequencies obtained from
finite element model and 9 364.5 354.2 2.8 347.8 346.0 0.5 323.7 333.0 2.9
vibration input 10 380.2 410.1 7.9 370.4 395.7 6.8 354.9 377.2 6.3
After detection of damage by use of frequency differences, next step

would be damage localization with the help of DLV method. At the
accelerometer locations, system’s pre-damage (FU) and post-damage
(FD) flexibility matrices are determined by using vibration inputs that
are obtained from the structure (Figure 3a). Singular Value Decom-
position (SVD) is applied to the difference matrix (DF) and Damage
Locating Vector (DLV) is determined. This vector (Figure 3b) is applied
on structural modal as statical load and stresses on the structure are
calculated, thereby normalizing these stresses, an index that is defined
as nsi is formed. In this project, as mentioned structure is a moment
resisting frame, bending stresses are used and for 50% damage sce-
nario, nsi values are obtained as shown in Figure 4 throughout beams
and columns. Regions that are under the limit which is pointed by
nsi may be potentially damaged and for these regions, more detailed
investigations should be accomplished. Therefore, if nsi < 0,2 condi-
tion is chosen as the damage localization limit, it is possible to say true
damaged regions on both columns and beams are under this limit and
potentially problem regions that are pointed by DLV method includes
the damaged regions.
Conclusion
Development of rapid and safe post-earthquake damage evaluation

methods has a vital importance in earthquake-prone countries like
Turkey where structural inventory is consisted of mostly reinforced con-
crete buildings. In this manner, vibration based evaluation methods are
Figure 3 — (a) Sample accel-

eration output at location C3
on column (b) DLV applying
statical load to sensor locations
on modal
(a) (b)
Figure 4 — Damage detection

on frame with DLV method:
nsi values (a) along beam
(b) along column
(a) (b)
presented. They provide an opportunity of in-place non-destructive

damage detection by using dynamic response data obtained from the
structure. If the algorithms which enable the usage of these meth-
ods into civil engineering structures work, both normally functional
usage and rapid, safe and inexpensive post-earthquake evaluation of
primarily strategical structures would have been provided. Besides,
contribution of this method to evaluation and strengthening of price-
less historical buildings cannot be ignored. Simulation based damage
detection studies presented in this article show that success rate of pre
and post-dam age dynamical properties determination for frame that
would be evaluated with experiments is high. However, it may have dif-
ficulties about damage localization and magnitude estimation. Addi-
tional simulation studies and experiments will be held in order to obtain
more accurate damage localization and magnitude estimation with
optimum measurement parameters and damage detection methods.
References
[1] Farrar, C.R., Baker, W.E., Bell, T.M., Cone, K.M.,Darling, T.W., Duf-
fey, T.A., Eklund, A. ve Miglori, A. (1994). Dynamic characterization
and damage detection in the I-40 bridge over the Rio Grande. Los
Alamos National Laboratory Report LA-12767-MS, Los Alamos Na-
tional Laboratory, P.O. Box 1193, Los Alamos, NM, 87544, USA.
[2] Salawu, O.S. (1997). Detection of structural damage through changes

in frequency: a review. Engineering Structures, 19(9), 718–723.
[3] Doebling SW, Farrar CR, Prime MB (1998). A summary review of

vibration-based damage identification methods. Shock and Vibration
Digest; 30, 91–105.
[4] Sohn H, Farrar CR, Hemez FM, Shunk DD, Stinemates DW, ve Nadler
BR (2003). A review of structural health monitoring literature:
1996–2001, Los Alamos National Laboratory, USA.
[5] Chang, P. C., Flatau, A. ve Liu, S. C. (2003). Review Paper: Health

Monitoring of Civil Infrastructure. Structural Health Monitoring
2003; 2(3); 257–267.
[6] Farrar, C. R. ve Worden, K. (2007). An Introduction to Structural

Health Monitoring. Philosophical Transactions of the Royal Society A:
Mathematical, Physical and Engineering Sciences, 365, pp. 303–315.
[7] Rytter, A. (1993). Vibration based inspection of civil engineering

structures. Ph.D. Dissertation, Department of Building Technology
and Structural Engineering, Aalborg University, Denmark
[8] Zimmerman DC, Kaouk M. (1992). Eigenstructure assignment

approach for structural damage detection. AIAA Journal,
30(12):1848–55.
[9] Fritzen CP, Jennewein D, Kiefer T. (1998). Damage detection based

on model updating methods. Mech Syst Signal Process ;12(1), 163–86.
[10] Doebling, S.W., Farrar, C.R., Prime, M.B. ve Shevitz, D.W. (1996).
Damage identification and health monitoring of structural and
mechanical systems from changes in their vibration characteristics: a
literature review. Technical Report LA-13070-MS, Los Alamos Nati-
onal Laboratory, Los Alamos, NM
[11] Randall, B. R. 2004a State of the art in monitoring rotating

machinery — part 1. Journal of Sound and Vibration, 38, pp. 14–21.
[12] Randall, B. R. 2004b State of the art in monitoring rotating

machinery — part 2. Journal of Sound and Vibration, 38, pp. 10–17.
[13] Friswell, M.I. and Penny, J.E.T. (1997). Is damage location using
vibration measurements practical, euromech 365 international
workshop: damas 97, Structural Damage Assessment using Advanced
Signal Processing Procedures, Sheffield, UK, June / July.
[14] Lee, U. and Shin, J. (2002). A frequency response function-based

structural damage identification method. Computers and Structures,
80, 117–132.
[15] Juang, J. (1994). Applied System Identification, Prentice-Hall,

Englewood Cliffs, N.J.
[16] Gunes, B (2002). Identification and localization of damage , PhD

Thesis, Northeastern University., Boston, MA.
[17] Vanoverschee, P. ve Moor, B. L. R. (1996). Subspace Identification

for Linear Systems, Kluwer Academic Publishers, Boston.
[18] Bernal, D. (2006). Flexibility-based damage localization from

stochastic realization results, Journal of Engineering Mechanics,
132(6), 651–658.
[19] Bernal, D. (2002). Load vectors for damage localization. Journal of

Engineering Mechanics, 128(1), 7–14.
[20] Bernal, D. ve Gunes, B (2004). A flexibility based approach for the

localization and quantification of damage: A benchmark application.
Journal of Engineering Mechanics, 130, 61–70.
Fragility of Shear Wall Buildings

with Torsional Irregularity
AKANSEL,Vesile Hatun Abstract

Middle East Technical Univer-
sity, Civil Engineering Depart-
Buildings with torsional irregularity represent the main focus of many
ment, Turkey
current investigations. However, despite this volume of research, there
GÜLKAN, Polat is no established framework that describes adequately the seismic vul-
Çankaya University, Civil Engi- nerability of reinforced concrete shear wall systems. In this study, the
neering Department, Turkey three dimensional behavior of a ¼ scaled 3 story shear wall structure
YAKUT, Ahmet which is a part of a nuclear plant was taken into account under differ-
Middle East Technical Univer- ent seismic excitations. This structure was subjected to AZALEE shak-
sity, Civil Engineering Depart- ing table tests in Saclay, France under the project of “SMART 2008”
ment, Turkey which was led by CEA (Atomic Energy Agency).
The main purpose of this study is that to capture the experimental

results with the numerical model and to calculate the fragility curves
of the structure. In experimental phase, three real and ten synthetic
accelerograms was used for shaking table tests. For the calculation of
the fragility curves, maximum inter-story drift values have been used
as the damage indicator to obtain the fragility curves and different
seismic intensity measures have been used. Thirty bi-directional syn-
tetic horizontal ground motions in the ranges of 0.1 g and 1.0 g, have
been applied for the time history analyses. Fragility curves of shear
wall building have been calculated according to pre-established
damage indicators.
Introduction
Calculation of three dimensional seismic effects on buildings involving

torsion is a challenge for structural engineering, especially for non–linear
behavior under earthquake effects. Modeling these types of buildings
Fragility of Shear Wall Buildings with Torsional Irregularity 39
needs much more care to generate acceptable results. In spite of the

developing computer technology and existence of many numerical
models, there are still deficiencies in modeling because of the assump-
tions made in the numerical models for material and seismic excitation
estimation. One way to test the versatility of a model is to do para-
metric studies which may be helpful for identifying the importance of
the variables in the models.
The other crucial concept is the fragility curves for different structural
categories. These statistically-evaluated or empirically-derived curves
provide a basis for the assessment of the performance of buildings under
different ground motion intensities so that loss estimates can be made.
Literature Survey
For walls two main modeling approaches are used as macro and
micro modeling depending on the chosen finite element technology.
Micro modeling is a continuum mechanics based approach and uses
two or three dimensional solid or shell finite elements. Non–linear
behavior of concrete and steel can be applied in the model on the
basis of material constitutive relationships from experimental results
([5], [6], [10], [7] and [4]). Micro modeling is suitable for represent-
ing the local behavior in the structure. ANSYS, ABAQUS, ADINA
and DIANA are sample software packages that include a variety of
element and material models in their libraries for micro modeling.
Many researchers have used micro modeling approach to simulate
the experimental measurements ([11], [14]).
In performing a seismic risk analysis of a structural system, the vul-

nerability information in the form of fragility curves is a widely prac-
ticed approach. Performance–based design is a powerful tool for the
assessment of buildings under earthquake effects. In recent decades,
the probabilistic approaches have become popular than deterministic
approaches for the determination of fragility curves of structures.
Shinozuka et al. [17], developed fragility curves associated with differ-
ent states of damage of bridges from observations following the 1995
Kobe earthquake event. They introduced the uncertainty and statisti-
cal interpretation of randomness through the notation of combined
and composite fragility curves.
Shear wall building behavior under earthquake effects and its perfor-
mance are observed from past earthquakes and experimental results. It
has been noted that for shear wall buildings there is no collapse under
earthquake effects. The inadequate number of research on the perfor-
mance limits of shear wall buildings indicates conflicting results ([12], [18]).
Jeong and Elnashai ([8], [9]) proposed a new three dimensional dam-

age index which takes into account the bidirectional and torsional
response effects. The main purpose in their study is to estimate three
dimensional damage capacity indexes, namely the global response of
the structure under earthquake effects by way of simple frame sys-
tems. Aziminejad and Moghadam [3], investigate the different con-
figurations of centers of stiffness and strength to generate the fragility
curves. In this study, only the results of the non linear time history
analysis and damage states specified "by SMART 2008" were used to
obtain the fragility curves.
Experimental Program and Model Structure
SMART 2008 Experimental Program

A reduced scaled model (scale of ¼th) of a nuclear reinforced concrete
building was tested on the AZALEE shaking table at Commissariat à
l’Energie Atomique (CEA Saclay, France). The loadings on the model
ranged from very low seismic motions to five times the design level.
For validation of the model, three real and 10 synthetic accelograms

were generated at AZALEE shaking table [15]. The details of the
accelerogram sets are given in Table 1 and Figure 1. The earthquake
motions were applied in both orthogonal directions. For vulnerability
analysis 30 set of synthetic derived ground accelerograms were used as
bi-directionally.
Model Structure
The plan and the mock-up pictures of the SMART 2008 specimen is
displayed in Figure 2. Density of the concrete is used as 2460 kg / m³
and Young Modulus of concrete is 32000 MPa as (a reference case)
according to the SMART 2008 Phase 2 report given by CEA [16].
MKIN and CONCRETE are used for the concrete in the model. MKIN
(Multi linear kinematic hardening), rate-depended plasticity is used.
CONCRETE is a defined material model in ANSYS for Willam–Warnke
material model. For this material type open shear transfer coeffi-
cient, 0.2 and closed shear transfer coefficient, 0.8, are used. Uni-
axial cracking stress is 2.4 MPa.
The model developed for this study in Figure 3 consists of 28740

SOLID65 (3-D Reinforced concrete elements) and 5282 MASS21
(Structural mass) element types. Also, the model has 43179 nodes for
calculations. Shaking table and foundation is not modeled and base-
ment is assumed as fixed supported as proposed in [15]. Seismic exci-
tations applied at basement level in the analytical model.
No Real Earthquakes M Dist. Acc.(g)
1 REA 1 UMBRO-MARCH(AS) 5.2 23 0.05

2 REA 2 MANJIL(AS) 4.4 14 0.05
3 REA 3 UMBRO-MARCHIGIANO 5.9 81.4 0.05
Synthetic Earthquakes
Table 1 — Real and synthetic
4–13 Derived according to the response spectrum and scaled from 0.1 g to 1.0 g accelerogram sets
Figure 1 — Ground motion

data used in the experiments
Figure 2 — The Identification

of locations where results
have to be computed, SMART
2008 specimen before and
after tests
Figure 3 — Numerical model

of SMART 2008 mock-up and
SOLID 65 element type
Model Verification and Fragility Analysis
Model Verification
The verification of the model was done by comparing the experimental
results with the calculated ones. In Table 2, these results were com-
pared in terms of relative percentage errors.
Ax Ay Bx By Cx Cy Dx Dy
Accsyn 1 –0.84 –0.87 –0.88 –0.91 –0.87 –0.89 -0.85 -0.88

Accsyn 2 –0.40 –0.39 –0.45 –0.03 –0.46 0.14 -0.26 -0.42
Accsyn 3 –0.65 –0.75 –0.61 –0.69 –0.62 –0.67 -0.67 -0.74
Accsyn 4 –0.05 0.33 –0.05 0.05 –0.10 0.10 -0.20 0.36
Accsyn 5 0.18 0.26 0.36 0.07 0.07 0.14 -0.15 0.34
Accsyn 6 0.24 0.32 0.26 0.18 0.23 0.32 -0.04 0.33
Table 2 — Relative Error Per- Accsyn 7 0.22 0.37 0.28 0.31 0.05 0.36 -0.03 0.37
centages between calculated
Accsyn 8 0.39 0.38 0.42 0.55 0.23 0.69 0.14 0.44
and measured data accord-
ing to absolute maximum Accsyn 9 0.11 0.33 0.15 0.35 0.19 0.42 0.02 0.35
displacements Accsyn 10 0.17 0.12 0.08 0.35 0.12 0.43 0.07 0.20
Negative values in Table 2 mean that calculated values are smaller than
the experimental results. Increase in the level of the seismic excitation,
results in decrease in error percentages. Detailed analysis results for
verification of the model can be obtained from [1] and [13].
Figure 4 — Displacement
comparisons of the experi-
mental results and analytical
results at the 3rd floor level
for Run 10 (Accsyn–0.7 g)
Fragility Analysis
One of the main objectives of this study was to obtain the fragility curves
for this structural type to develop the behavior of shear wall buildings
under different seismic excitations with torsion effects. The thresholds
which were defined by the project team are given in Table 3, [16].
Maximum inter-story drifts were used as a damage indicator (Table 3).

To investigate the local effects of the damage, the fragility curves were
calculated at specified points shown in Figure 2. These damage levels
are used as the criteria for the fragility analysis. H is the story height
and equals to 1.2 m.
Damage Levels Drift Ratio
Light Damage H /
400 = 3 Table 3 — SMART 2008 Dam-
Controlled Damage H
/
200 = 6 age levels defined for maxi-
Extended Damage H /
100 = 12 mum inter-story drifts
Thirty bi-directionally applied time-history analyses were performed

for the fragility analysis. According to the results of the fragility analy-
sis, the log-normal distribution was assumed for the distribution of
the structural response indicators and then the fragility curves were
obtained according to median capacity, Am and standard deviation, β
of this distribution.
The probability of failure Pf of a structure or component conditioned

on seismic ground motion level “a” is expressed by fragility curves as
given in Equation (1).
Pf P Failure a P A a (1)
Failure occurs if the actual capacity of the structure is inferior to the

seismic demand, that is the given ground motion level “a”. The fail-
ure probability conditioned on ground motion parameter “a” given
by the cumulative distribution function of capacity Am is calculated
from Equation (2).
ln a Am
Pf (2)
To obtain the fragility curves from the probability density functions, we

need to define the acceptable median capacity and standard devia-
tions for the limit states defined in Table 3 under different seismic exci-
tations. One of the methods used to determine the median capacity
and the standard deviation is the regression analysis. In Figure 5, the
result of the regression analysis result for PGA versus maximum inter-
story drift at point E in X direction is displayed.
Figure 5 — Regression analyses

for maximum inter-story drift
at point E for PGA
To obtain the least error, the method of least squares is applied to the
data in this study. Following regression analyses the needed median
seismic capacity, Am and log-standard deviation,can be evaluated. For
the evaluation of Am, Ycrit value can be used as shown in Equation (3).
ln Ycrit a
(3) ln Am
b
In Equation (3), Ycrit values were defined in the SMART 2008 Phase-2
report [16] as damage levels which were given in Table 3.
In this study, Point E is taken into account so as to represent the struc-

tural behavior. This point is close to the mass center of the system
and, has the high correlation coefficients when the time history data is
compared with the fitted curves as a result of the regression analysis.
The correlation coefficients of fitted curves and time history analyses
are given in Table 4. The correlation coefficient for PGA parameter for
X and Y directions are respectively 0.89 and 0.91. These results were
thought to have reasonably enough accuracy for the fragility curves.
Table 4 — Correlation coef- PGA x PGA y PGV x PGV y PGD x PGD y CAV x CAV y
ficients for time history data
versus fitted curves Point E 0.89 0.91 0.92 0.87 0.86 0.82 0.94 0.97
According to these regression analyses, the log-standart deviation

coefficients and seismic median capacity coefficients for data are dis-
played in Table 5 and Table 6, respectively. The scatter of the fragility
curves for the given damage levels changes under different seismic
ground motion indicators as shown in Fig. 6. PGA and PGV have simi-
lar trends in shape; however the CAV and PGD differ.
The probabilistic scatters between the damage levels were slightly

wider and the structure behaves well even under relatively higher seis-
mic motions. In Figure 6, the probability of failure at the damage lev-
els under PGA seismic motion indicator were nominal till 0.5 g. Our
extensive calculations have shown a very good agreement between
experiment and theory, supporting the power of the computational
approach in obtaining far-reaching generalizations. The model was
designed to experience significant coupled translation-torsion during
its dynamic response, and the computations captured that well.
PGA x PGA y PGV x PGV y PGD x PGD y CAV x CAV y

Table 5 —  (Log-standard
Point E 0.46 0.41 0.45 0.49 0.73 0.80 0.73 0.62 deviation) coefficients for data
LD 0.45 LD 0.46
CD 0.80 CD 0.82
PGA x ED 1.42 PGA y ED 1.48
LD 0.22 LD 0.23
CD 0.39 CD 0.40
PGVx ED 0.66 PGV y ED 0.70
LD 0.08 LD 0.09
CD 0.14 CD 0.15
PGD x ED 0.25 PGD y ED 0.26
LD 7.09 LD 7.32 Table 6 — Am Seismic median

CD 14.91 CD 15.37 capacity coefficients for data
CAV x ED 31.36 CAV y ED 32.27 for Point E
Figure 6 — Fragility Curves

Comparisons of Point E for
various seismic motion
indicators at X direction
References
[1] Akansel V. H., Kazaz İ., Yakut A. and Gülkan P. (2010). Modeling and
simulation of the shear wall buildings’ behavior under earth quake
forces. Dicle Üniversitesi Mühendislik Dergisi. Vol.1, No.1, pp. 31–40.
[2] ANSYS R 12.1, Swanson Analysis System, 2011.
[3] Aziminejad A. and Moghadam A. S. (2009). Performance of asym-

metric multistory shear buildings with different strength distribu-
tions. Journal of Applien Science. Vol.9, No.6, pp. 1080–1089
[4] Fischinger M. and Isakovic T. (2000). Benchmark analysis of a struc-

tural wall. 12th World Conference on Earthquake Engineering. 0416
[5] Ile N. and Reynouard J.M. (2003). Lightly reinforced walls subjected
to multi – directional seismic excitations: Interpretation of CAMUS
2000-1 dynamic tests. Vol. 40, No. 2–4, pp. 117–135.
[6] Ile N. and Reynouard J.M. (2005). Behaviour of U-shaped walls

subjected to uniaxial and biaxial cyclic lateral loading. Journal of
Earthquake Engineering. Vol. 9, No. 1, pp. 67–94.
[7] Ile N., Nguyen X. H., Kotronis P., Mazars J. M. (2008). Shaking
table tests of lightly RC walls. Journal of Earthquake Engineering.
12:6, pp. 849–878.
[8] Jeong S.–H. and Elnashai A. S. (2006a). New three-dimensional

damage index for RC buildings with planar irregularities. Journal of
Structural Engineering. Vol. 132, No.9, pp. 1482–1490.
[9] Jeong S.–H. and Elnashai A. S. (2006b). Fragility analysis of build-

ings with plan irregularities. 4th International Conference on
Earthquake Engineering, Taipei, Taiwan
[10] Kazaz İ., Yakut A. and Gülkan P. (2006). Numerical simulation of

dynamic shear wall tests: A benchmark study. Computers and Struc-
tures. Vol. 84, pp. 549-562.
[11] Kwak H.–G. and Kim D.–Y. (2004). FE analysis of RC shear walls
subjected to monotonic loading. Magazine of Concrete Research.
56, No. 7, pp. 387-403.
[12] Moehle J. P. (1996). Displacement–based seismic design criteria.

11th World Conference on Earthquake Engineering. 2125, ISBN:
0 08 042822 3.
[13] Nazirzadeh S., Yakut A. Akansel V. H. and Kazaz İ. (2011). Numerical

modeling of an unsymmetrical building tested on shaking table.
Advances in Structural Engineering and Mechanics (ASEM 11)
Conference.
[14] Palermo D. and Vecchio F. J. (2007). Simulation of cyclically loaded

concrete structures based on the finite element method. Journal of
Structural Engineering. Vol. 33, No. 5, pp. 728–738
[15] RAPPORT DM2S, (2007). Presentation of the Benchmark Contest–

Phase 1 Project SMART 2008.
[16] RAPPORT DM2S, (2009). Presentation of the Benchmark Contest–

Phase 2 Project SMART 2008. COMMISSARIAT A L’ÉNERGIE
ATOMIQUE, SEMT / EMSI/ PT / 09-011 / A
[17] Shinozuka M., Feng M. Q. Lee J. and Naganuma T. (2000). Statis-

tical analysis of fragility curves. Journal of Structural Engineering.
Vol. 126, No.12, pp. 1224–1231.
[18] Wallace J. W. and Moehle J. P. (1992). Ductility and detailing

requirements of bearing wall buildings. Journal of Structural Engi-
neering. Vol. 118, No.6, ASCE, ISSN 0733-9445 / 92 / 0006-1625.
Vibration monitoring and structural identification of an historic masonry bell tower 49
Vibration monitoring and

structural identification of an
historic masonry bell tower
Abstract CABBOI, Alessandro

University of Cagliari, Italy
In this study, partial results of a long-term dynamic monitoring per-
formed on an historic masonry bell-tower were carried out. The tower
has significant structural problems, such as wide cracks on several
façades. After a brief description of two previous ambient vibration
tests, the paper focuses on the simple permanent monitoring system
installed inside the tower (for several months) in order to investigate
the variation of modal parameters. The experimental investigation is
complemented by a 3D F.E. model and its subsequent vibration-based
updating to get a first structural identification. The aim of this study
is to assess how damage and environmental factors can influence the
dynamic parameters.
Introduction
The preservation of historic masonry towers has gained an important

role in Architectural Heritage conservation. In fact, these structures
are affected by several issues due to the ageing of materials, damage
and uncertain boundary conditions related to adjacent structures or
to geotechnical problems. Another fundamental feature is their slen-
derness that confers a high sensitivity to dynamic actions such traffic-
induced microtremors, bell swinging, wind and earthquakes. In this
sense, ambient vibration testing and overall continuous dynamic moni-
toring seem to be ideal tools to evaluate the dynamic behavior under
operational condition [1] and to validate F.E. models [2–3], crucial for
successive safety assessment.
A further important application of dynamic monitoring is the evaluation

of the influence of environmental factors on the modal parameters
and its use for dynamics-based damage assessment. In the literature,
damage identification (DI) methods can be categorised as non model-
50 Vibration monitoring and structural identification of an historic masonry bell tower
based and model-based techniques [4, 5]. The first methods are based
on the change of eigenfrequencies, mode shapes and its derivatives.
Therefore, it becomes important to analyse the effects of environmental
factors — since could mask the changes due to structural damages — on
the modal parameter [6–7]. On the other side, the second DI meth-
ods need accurate numerical models. For these last methods, model
updating is a fundamental step to enhance the quality of the model
before it is used for damage assessment [8].
The paper details several main steps of the vibration-based damage

assessment of historic bell towers, and involves both, experimental
and theoretical modal analysis. The developed procedure starts by the
research of the available information regarding: a) historical studies;
b) the accurate survey of the tower geometry and of the crack pattern;
c) previous ambient vibration testing and identification of the dynamic
characteristics of the tower (June 2007–2008); d) the development of
a first numerical tuned model [3]. Since the identified modal param-
eters of the tower highlight a clear non-linear behaviour caused by
the bell-swinging excitation, a simple dynamic monitoring system was
installed inside the tower for several months (2009–2010).
The frequencies tracked on the acquired data exhibit a significant

dependence on temperature. Subsequently, in the theoretical part of
the study, the identified natural frequencies were used to estimate
some uncertain structural parameter of the F.E. model by using the
simple Douglas-Reid method [9]. In addition, based on the infor-
mation carried out from the dynamic monitoring and the numerical
model, it is possible to assess how environmental factors and damage
condition the dynamic behaviour of the tower.
The bell tower of San Vittore church in Arcisate
The investigated tower (Figure 1a), about 37.0 m high, is built in irreg-
ular stonework masonry on a square base and connected on the East
side and partly on the South side to the San Vittore church (XI century)
in Arcisate (Varese). The first historic document of the bell tower dates
back to XVI century, even though it was probably built on a previous
roman building and modified along the centuries.
The masonry tower exhibits 6 orders of floors, with 5 of them being

defined by masonry offsets at the corners and by corresponding
sequences of small hanging arches marking the floor levels. The two
upper orders of floors were probably added in the XVIII century to
host the bell trusses.
The accurate survey of tower geometry and crack pattern clearly reveals
that the masonry texture appears often highly disordered, with local
prevalence of continuous or little staggered vertical joints; in addi-
tion, it is difficult to distinguish between vertical cracks and insuffi-

cient stone interlocking because of erosion of the mortar joints. Along
all sides, the tower exhibits long vertical cracks, most of them cutting
the entire wall thickness and crossing through the keystones of the
arched window openings. These cracks are mainly detected between
the second / third order of the tower and the base of the belfry, show-
ing a maximum aperture at the upper end (Figure 1b). Many superficial
cracks are also diffused, particularly on the North and West fronts
(Figure 1b), which are not adjacent to the church.
The masonry has been characterized through sonic tests; the results
generally indicate a relatively compact masonry of fairly good execution.
(a) (b)
Figure 1 — (a) Axonometric

view of the bell-tower;
(b) crack pattern on the differ-
ent fronts (dimensions in m)
Dynamic characteristics of the tower
Preliminary ambient vibration tests

Two ambient vibration tests were conducted on the tower: on June
2007 and June 2008. In both tests the response of the tower was
measured in 15 selected points, according to the sensor layout shown
in Figure 2a. For each test two different series of ambient vibration
data were recorded: in the first series (AV1), the ambient excitation
was only provided by the wind and the micro-tremors; in the second
series (AV2), the excitation was provided by the swinging of bells. In
both tests, the sampling frequency was 200 Hz and the time of acqui-
sition was larger than 2000 times [10] the period of the structure’s
fundamental mode.
An example of the acceleration time-histories recorded in the test of

June 2007 at the same channel for both test series is given in Fig-
ures 2b and 2c respectively. It should be noticed that the bell swinging
(Fig. 2c) produced a significant increase of the vibration level associ-
ated with micro-tremors and other ambient excitations (Fig. 2b); it can
be observed that the maximum amplitude of acceleration responses
is increased of 10-20 times by the bell swinging. Similar results have
been obtained in the test performed on June 2008.
(a) (b)
(c)
Figure 2 — (a) Sensor layout
adopted in AVT of the tower
(dimensions in m); Accelera-
tion induced at the same point
by: (b) micro-tremors and
wind; (c) swinging of bells
The extraction of modal parameters was carried out by using two

well-known and complementary techniques: the Frequency Domain
Decomposition [11] and the data-driven Stochastic Subspace Iden-
tification [12]. In this paper, only the results obtained using the SSI
technique will be reported, since this technique was used in the
operational modal analysis (OMA) of monitoring's collected data. The
application of SSI technique to the AV1 data recorded in June 2007
allowed the identification of 5 vibration modes in the frequency range
of 0–6 Hz (Fig. 3a). Figure 3b shows the identified mode shapes: dom-
inant bending (B) modes were identified at 1.21 (B1), 1.29 (B2), 3.98
(B3) and 4.14 Hz (B4) while only one torsion mode (T1) was identified
at 3.56 Hz. The analysis of AV1 time series collected in June 2008
provided very similar results.
Since each test was conducted under different level of ambient exci-
tation (Figure 2b and 2c), additional investigation was performed on
the time invariance of resonant frequencies and mode shapes. This
investigation was carried out by evaluating the corresponding aver-
age normalized auto-spectral (ANPSD) estimates [13] of the accelera-
tions acquired on June 2007. The ANPSDs obtained from each test
are plotted in Figure 4 and reveals that all natural frequencies tend
to decrease at the higher level of ambient excitation; the decreases
approximately range from 1.7% (mode B1) to 2.7% (mode B3). It has
to be further noticed that differences were detected between the
mode shapes as well. Those difference generally tend to increase as
the mode order increases.
Figure 3 — (a) Stabilization

diagram (SSI technique);
(b) Vibration modes
(a) (b)
Figure 4 — ANPSD of the

accelerations acquired on
June 2007 under different
levels of ambient excitation
Vibration monitoring
From June 2009 until March 2010 a continuous dynamic monitoring
system operated inside the tower in order to better investigate the
time variance of resonant frequencies (Figure 4). The instrumentation
consisted of 3 uni-axial Dytran 3191A1 piezoelectric accelerometers
(10 V/g), a 4-channel data acquisition system (NI 9234) and a local
PC for the management of the continuous acquisition and the data
storage (Figure 5). The sampling frequency was 200 Hz. The choice of
instrumented level was based on the information collected in the pre-
vious ambient vibration tests, so that all previously identified modes
are observable at that level.
In addition, averaged temperature data were recorded by a static

monitoring system previously installed on the tower and managed by
Politecnico di Milano [14], as well: 8 thermocouples measured internal
and external temperature at different levels of the tower.
The OMA of collected accelerations was performed in two main steps.

The first step, performed by a powerful tool developed in LabView
[15], involved the following tasks: a) the creation of a database with
the original data for subsequent developments; b) detrending, auto-
matic recognition and extraction of the time series associated to
swinging of bells (the responses induced by the swinging of bells do
not comply with some basic hypothesis of OMA, such as white noise
and stationarity of input); c) statistical analysis; d) low-pass filtering
and decimation of the data containing the response to micro-tremors
and wind only; e) creation of a second database with essential data
Figure 6 — General arrange-

ment of the dynamic monitor-
ing system
Figure 7 — Identified natural

frequencies of the first five
modes and internal tempera-
ture at the first level of the
south side masonry between
Sept 09 and Oct 07, 2009
records. The second step consisted in the modal identification using

the data driven automatic SSI method [12] available in the commer-
cial software ARTeMIS.
The inspection of Figure 7 reveals that the bending modes are much
more frequently identified than the torsion mode, probably as a con-
sequence of the low level of ambient actions that generally excite the
tower. Since only the response to low level excitation (micro-tremors
and wind) has been considered, the cyclic frequency variation is likely
to be related to temperature. In order to clearly point out this aspect
the frequency variation and the correlation with temperature are bet-
ter illustrated in Figure 8.
(a) (b)
(c) (d)
Figure 8 — Identified frequen-

cies of bending modes f 1, f 2,
f 4 and f 5 versus: (a,c) time
and (b–d) temperature re-
corded between September
09 and October 07, 2009
A linear trend is clearly detected between frequency and temperature

data: more specifically, the inspection of Figures 8b–d reveal that the
natural frequencies increase as temperature increases. This behaviour
can be explained through the closure of superficial and deep cracks
induced by the thermal expansion of materials; furthermore, the
stone interlocking of irregular stonework masonry conceivably tends
to increase as the temperature increases. The natural frequencies of
the lower bending modes exhibit higher percent variations: the mode
frequencies f 1–f 2 daily oscillates of about 0.05 Hz (around 4% of the

average values), whereas the mode frequencies B3–B4 exhibit daily
variation of about 0.1 Hz (around 2.4% of the average frequency). It is
further noticed that the internal south side masonry temperature were
ranging between 22.8–35.5 °C during the investigated period, with a
mean value of 29.8 °C.
Monitoring Monitoring
(26 / 06 / 09 - 31 / 07 / 09) (09 / 09 / 09 - 07 / 10 / 09)
Mode fm δf fmin Fmax fm δf fmin Fmax

Id. [Hz] [Hz] [Hz] [Hz] [Hz] [Hz]
B1 – 1 1.249 0.0138 1.198 1.282 1.229 0.0107 1.203 1.265

B2 – 2 1.319 0.0190 1.258 1.367 1.297 0.0134 1.270 1.343
T1 – 3 3.524 0.0574 3.365 3.688 3.551 0.0394 3.457 3.655
Table 1 — Natural frequencies
B3 – 4 4.056 0.0329 3.941 4.163 4.020 0.0209 3.941 4.095 identified during the investi-
B4 – 5 4.247 0.0442 4.094 4.383 4.193 0.0249 4.104 4.261 gated period of monitoring
Table 1 presents the mean values, the standard deviations and the
range of variation of natural frequencies estimated during the investi-
gated period of monitoring and compares these values with the ones
obtained around July 2009 (studied in a previous work [16]). The
mean values of natural frequencies of the current period are lower
than ones obtained during July. It can be also observed a decrease
of the standard deviations from July to September. These differences
can be explained by the temperature effect, since the weather in July
was warmer (T1-Sint, mean = 33.5 °C). Therefore, higher temperature
induced higher thermal expansion of materials that closed superficial
and deep cracks, increasing the global stiffness of the tower.
Hence, the dynamic characteristics of the tower are clearly depend-

ent on the environmental factors, due to the very low amplitude
of ambient excitation in the time series collected during the moni-
toring. After all, to better understand how damage influences the
modal parameters, it becomes necessary to develop a finite element
model and update the latter in agreement with the experimental
response of the structure.
F.E. modeling and structural identification
Finite element model and sensitivity analysis

The F.E. program Straus 7 was used to create the numerical model.
The tower was modelled by using 8-node brick elements. A relatively
large number of finite elements have been used in the model, so that
regular distribution of masses could be obtained and all the openings
in the load-bearing walls could be reasonably represented. The model
consist of 3475 solid elements with 17052 active degrees of freedom.

The geometry of the tower was accurately checked; the main uncer-
tainties are related to the boundary conditions and the characteris-
tics of the material. The following assumptions were introduced in
order to reduce the number of uncertainties in the model calibration:
(a) the weight per unit volume of the masonry equal to 17.0 kN / m³;
(b) constant Poisson’s ratio of the masonry equal to 0.15; (c) fixed
tower footing to account for the fact that the soil-structure interaction
is hardly involved at the low level of ambient vibrations that existed
during the tests. The F.E. model was firstly refined in 4 steps of system-
atic manual tuning. The correlation between the dynamic characteris-
tics of the F.E. model and the experimental results was evaluated via
the maximum absolute frequency discrepancy DF,max:
(1) DF, max max DF, i

exp
fFEM , i fmean , i
(2) DF, i 100 exp
fmean ,i
and the average frequency discrepancy J:
M
1
(3) J D F,i
M i l
The mean values of the identified natural frequencies regarding the

period around September are assumed as reference experimental val-
ues inasmuch show a smaller variation in time.
In the first dynamic analysis (FEM1) the Young’s modulus of stone

masonry was assumed equal to 3.00 GPa (value suggested by engi-
neering judgement and the results of sonic tests). The comparison
between theoretical and experimental modal parameters shows highly
imperfect correlation: a) the model is much stiffer than the tower;
b) the torsion mode T1 of the model does not follow the experimen-
tal sequence; c) the mode shapes of FEM1 bending modes involve
motion along N-S and E-W directions while the experimental tower
modes involve bending along the diagonals (Fig. 4b). The poor quality
of correlation clearly indicates that the assumptions on the isotropic
behaviour of stone masonry and on the connection with neighbouring
building need to be revised. Hence, an orthotropic elastic behaviour
(FEM2) was assumed for the stone masonry; the average character-
istics of the material were E = 3.00 GPa, G13 = G23 = 0.3 GPa. The
orthotropic model improved the correlation with the experimental
results: a) the stiffness of the model significantly decreased: b) the
torsion mode T1 of FEM2 correctly follows the experimental sequence.
Subsequently, a third model (FEM3) was developed, by accounting for
the connection between the tower and the church through a series
of linear (nodal) springs of constant stiffness k. An initial stiffness of
Figure 9 — Mode shapes

of FEM4
B1 = 1.198 Hz B2 = 1.234 Hz T1 = 3.375 Hz B3 = 4.280 Hz B4 = 4.359 Hz
springs was assumed equal to k = 4* 104 kN / m. After the correction,

the bending modes are fully consistent with experimental data. Finally,
based on a preliminary sensitivity analysis, the uncertain parameter
were assumed as follows (FEM4): E = 3.00 GPa; G13 = G23 = 0.4 GPa;
k = 3* 104 kN / m. The latter model improve the correlation with the
experimental setup (Fig. 9). In Tab. 2 are reported the evaluated fre-
quency discrepancies of the main steps carried out.
In order to assess the most sensitive parameters of the F.E. model and
to select the minimum number of uncertain parameters for the inverse
problem of structural identification, the base F.E. model (FEM4) was
checked through a sensitivity analysis. The sensitivity coefficients sik
(Fig. 9) are computed — by a perturbation technique — as the rate
of change of a particular response of the model fi,FEM with respect
to a change in a structural parameter Xk (1). If sensitivity coefficients
for different types of parameters are to be compared and normalized
with respect to the base response value, then sik can be calculated as
percentage like (2). The kth parameter X vary into a range defined by
an upper and lower bound (3); the responses fi,FEM are respectively
computed at these limit values (4).
Figure 10 — Sensitivity
coefficients
FEM
fi
sik (1)
Xk
FEM
Xb, k fi
s ik 100 FEM
(2)
fb,i Xk
U L
(3) Xk Xk Xk
FEM
(4) fi fi FEM , U fi FEM , L
The inspection of the sensitivity coefficients clearly reveals that: a) the

bending modes, above all lower order modes, are significantly affected
by E; b) the shear modulus G mostly influence the torsion mode and
affected considerably the higher order bending modes; c) the stiffness
constant of springs k weakly conditions the dynamic characteristics.
Therefore, E and G are good candidates to be selected as updating
parameters in a structural identification approach, since they do not
affect the inverse problem by ill-conditioned states.
On the other hand the influence of constant k deserves more care-

ful regards. Comparing the sensitivity coefficients between the first —
second and third — fourth bending mode, it seems that this parameter
conditions them differently. K is the only parameter able to regular-
ize the correct distance between the bending modes. From Tab. 2, it
appears that the computed bending modes from FEM3 and FEM4 are
spaced between them in a different way compared with the experi-
mental results SSImean. Therefore, a further variation of the constant k
was necessary in order to tune the correct spacing between the bend-
ing modes. It was assumed a new value of k equal to 8* 104  kN / m.
After that point, the model FEM5 can be chosen as “base” model for a
structural identification process, containing only two uncertain param-
eters: the elastic modulus E and the shear modulus G.
Structural identification procedures and damage assessment

The uncertain parameters E and G were estimated by minimizing the objec-
tive function (3). According to the Douglas-Reid approach [9], the rela-
tionship between any response (natural frequencies and mode shapes)
of the model and the structural parameters Xk of the model is approxi-
mated around the current values of Xk by the following expression:
N
2
(5) f i* X 1 , X 2 , , XN Aik X k B ik X k Ci
k l
To satisfy the expression and solve the problem, (2N + 1) constant (Aik,

Bik and Ci) must be calculated before to compare each f i * to its experi-
mental counterpart. In order to calculate the coefficients, structural
base parameters XBk, upper XUk and lower XLk values were estimated
with engineering judgement. When those (2N + 1) parameters are
defined, the constants on the right hand side of (5) can be computed
by satisfying the equation with frequencies obtained for defined
parameters. Thus, the equations stated above will be obtained:
B B B FEM B B B
fi * X 1 , X2 , , XN fi X1 , X 2 , , XN
fi * X1L , X2B , , X BN fi FEM
X1L , X 2B , , X BN
U B B FEM U B B
fi * X 1 , X2 , , XN fi X1 ,X 2 , , XN (6)
fi * X 1B , X2B , , X LN fi FEM
X1B , X 2B , , X LN
B B U FEM B B U
fi * X 1 , X2 , , XN fi X1 , X 2 , , XN
The first choice of the structural parameters corresponds to the base

values; then, each of them is varied one at time from the base value
to the upper and lower limit respectively. Once the coefficients Aik ,
Bik and Ci have been computed, the approximation (5) is completely
defined and it can be used to update the structural parameters. In
order to obtain the optimal parameters, the average frequency dis-
crepancy (3) was minimized through an unconstrained optimization
algorithm, well known as Nelder-Mead simplex method. The opti-
mizing iterative procedure identified the uncertain values concern-
ing the material properties used in the updated model as follows:
E = 3.00 GPa; G = 0.43 GPa. Table 2 summarizes the main results
obtained for the updated process.
B1 B2 T1 B3 B4 J Df, MAX
SSImean 1.229 1.297 3.551 4.020 4.193 – –

FEM3 1.171 1.218 2.992 4.031 4.144 5.600 15.742
FEM4 1.198 1.234 3.375 4.280 4.359 4.553 6.468
Table 2 — Correlation between
FEM5 1.218 1.285 3.426 4.394 4.544 4.603 9.303 the experimental results and
FEMott 1.226 1.292 3.534 4.468 4.610 4.440 11.144 computed frequencies in [Hz]
The correlation between theoretical (FEMott) and experimental behav-

iour (SSImean) is satisfactory for the first three modes. In addition, phys-
ical constraint like spacing between the modes shapes was achieved.
On the other hand the Df, max for the higher bending modes remains
quite high (11.14% for the fourth mode). Probably this discrepancy
could be due to the following reasons: a) as a consequence of the
simplified distribution of the model elastic properties; b) damage has
a considerably influence on the higher modes that are characterized
by more complex mode shapes, suffering much more the physical and
mechanical properties alteration of the material.
A future development of this work is to improve the numerical model,

simulating the most damaged areas of the tower and the parts that
were built or modified along the centuries by a non-homogeneous
distribution of the Young’s modulus. To limit the number of structural
parameters in the structural identification procedure, the tower could
be divided in three regions assuming different masonry Young’s modu-
lus as constant within each zone. Once the model will be satisfactory
tuned on the experimental data, it will form a helpful tool for damage
identification and numerical analysis for successive safety assessment.
As a final remark, if we add the information carried out from long-term
dynamic monitoring, the dynamic-based damage assessment meth-
odology outlined in the paper seems a powerful tool for diagnosis of
historic structures. It is worth underlining that long term monitoring
is mandatory to distinguish between the cyclic (temperature) and the
irreversible (damage) effects on the global dynamic characteristics [7];
in this sense, it will become important to apply a pattern recognition
technique on the experimental data.
Conclusions
An experimental and theoretical dynamic investigation of a historic

masonry tower is described in the paper. The following conclusions
can be drawn from the study: a) notwithstanding the very low level of
vibration during the dynamic monitoring, operational modal analysis
has proved to be effective tools for modal identification of masonry
towers; b) within the frequency range 0–6 Hz, 4 dominant bending
and 1 torsion modes were clearly identified from ambient vibration
data; c) the dynamic behaviour of the investigated tower is signifi-
cantly affected by the temperature variation (the natural frequencies
identified increases as the temperature increases) and by the ampli-
tude of bell swinging excitation; d) it becomes important to investigate
the time-history of identified modal parameters by pattern recogni-
tion techniques in order to distinguish between cyclic and irreversible
effects on the global dynamic characteristics.
Furthermore, the comparison between measured and predicted modal

parameters was used to verify the assumptions adopted in formulating
a F.E. model of the structure. The correlation between theoretical and
experimental modal parameters provides partially accurate verification
of the model main assumptions. One of them regards the hypothesis
of orthotropic elasticity, that seems more suitable when modelling
stone masonry structures.
Although this part of the study cannot be considered concluded due to

excessive lack of information on the mechanical characteristics of materi-
als and to the too simplified assumptions adopted for the numerical
model, dynamic-based damage assessment methodology seems a prom-
ising approach to monitor and evaluate damage in historic structures.
References
[1] Schmidt, T. (2009). FE Comparison of the dynamic behavior of 16

historical twin bell towers. Proceedings of 3th Int. Operational
Modal Analysis Conference (IOMAC-I), Portonovo / Italy, S. 483–490.
[2] Ivorra, S., Pallares F. J. (2006). Dynamic investigation on a masonry

bell tower. Engineering Structures 25(5), S. 660-667.
[3] Gentile, C., Saisi, A. (2010). FE modeling of an historic masonry

tower and vibration-based systematic model tuning. Proceedings of
7th Int. Conf. on Structural Analysis of Historic Constructions
(SAHC), Shanghai / China, S. 435-430.
[4] Fritzen, C.P. (2005). Vibration-based structural health monitoring

concepts and application. Proceedings of the DAMAS 2005, Key Engi
neering Materials, Trans Tech Publications, Zürich / Switzerland, S. 3-20.
[5] Maeck, J. (2003). Damage assessment of civil engineering struc-

tures by vibration monitoring. PhD Thesis, K.U. Leuven / Belgium.
[6] Ramos, L. F., Marques, L., Lourenço, P. B., De Roeck, G., Campos-
Cost,a A., Roque, J. (2010). Monitoring historical masonry struc-
tures with operational modal analysis: two case studies. Mechanical
Systems and Signal Processing 24, S. 1291-1305.
[7] Magalhães, F., Cunha, A., Caetano, E. (2012). Vibration based struc-
tural health monitoring of an arch bridge: From automated OMA to
damage detection. Mechanical Systems and Signal Processing 28,
S. 212-228.
[8] Gentile, C., Saisi, A. (2007). Ambient vibration testing of historic

masonry towers for structural identification and damage assess-
ment. Construction and Building Materials 14(4), S. 1311-1321.
[9] Douglas, B. M., Reid, W. H. (1982). Dynamic test and system iden-
tification of bridges. Journal of the Structural Division 108(10),
S. 2295-2312.
[10] Cantieni, R. (2005). Experimental methods used in system identifi-

cation of civil engineering structures. Proceedings of 1st Int. Oper-
ational Modal Analysis Conf. (IOMAC’05), Copenaghen / Denmark,
S. 249-260.
[11] Brincker, R., Zhang, L. M., Andersen, P. (2000). Modal identifica-

tion from ambient responses using Frequency Domain Decomposition.
Proceedings of 18th Int. Modal Analysis Conference (IMAC-XVIII),
San Antonio / Texas / U.S.A., S. 625-630.
[12] Van Overschee, P., De Moor, B. (1996). Subspace identification for

linear systems: Theory, implementation, applications. Kluwer Aca-
demic Publishers.
[13] Felber, A.J. (1993). Development of a hybrid bridge evaluation system.

Ph.D. Thesis, University of British Columbia / Canada.
[14] Binda, L., Condoleo, P., Tiraboschi, C. (2012). On-site investigation

and crack monitoring of an ancient bell-tower. Proceedings of Struc-
tural Faults & Repair-2012 (in print).
[15] Gentile, C., Saisi, A., Busatta, F. (2011). Dynamic testing and perma-
nent monitoring of an historic iron arch bridge. Proceedings of 8th
Int. Conference on Structural Dynamics (EURODYN), Leuven / Belgium,
S. 1186-1193.
[16] Gentile, S., Saisi, A., Cabboi, A. (2012). Dynamic monitoring of a

masonry tower. Proceedings of 8th Int. Conf. on Structural Analysis
of Historic Constructions (SAHC), Wroclaw / Poland, (in print).
Knowledge based structural assessment of Mallorca cathedral 63
Knowledge based structural

assessment of Mallorca cathedral
Abstract ELYAMANI, Ahmed

Technical university of
Catalonia, Spain
The paper presents the current state of an on-going research on Mal-
lorca cathedral. The cathedral is an audacious Gothic structure built in CASELLES, Jose Oriol
the island of Mallorca in Spain during XIV–XVI centuries, character- Technical university of
ized for its large dimensions and slender structural members. So far, Catalonia, Spain
experimental and numerical modal analysis, in addition to tentative
CLAPES, Jaime
model updating and seismic analysis were performed. The dynamic Technical university of
identification were carried out by ambient vibration testing, and then Catalonia, Spain
the frequency domain decomposition (FDD) technique was used to
obtain the modal parameters. A 3D Finite Element (FE) model was ROCA, Pere
Technical university of
used to determine the vibration modes. The model was updated to
Catalonia, Spain
improve the matching between experimental and numerical modal
parameters. Once updated, the model was utilized to study the seis-
mic response of the cathedral using non-linear static pushover analy-
sis. The results of a continuous dynamic monitoring system currently
installed in the building are also presented.
Introduction
Even though Mallorca Cathedral is located in a low-to-medium seis-

mic region, its seismic behaviour is of concern due to the vulnerability
which may result from its audacious structural features, as in particu-
lar the long-span and very high central nave vaults and very slender
nave piers. Moreover, the structure shows significant deformation and
some cracking can be observed in piers, vaults and walls [1], [2].
The island of Mallorca has been struck by three moderate earth-

quakes (with intensity higher than VI on the Mercalli scale) in the
last 400 years with epicentre in Campos-Palma (1660), Selva (1721)
and Palma-Marratxí (1851). This last event, with estimated maximum
intensity of VIII in the MSK scale, is considered as the major earth-

quake having occurred in Mallorca in the last four centuries [3].
In the current research the dynamic identification tests were carried

out as one of the tasks planned to assess the seismic performance of
Mallorca cathedral. Those tests are useful to characterize the main
modal parameters (natural frequencies, mode shapes and damping
ratios) which represent the structure’s dynamic behaviour as a result
of physical or mechanical properties (as the elastic modulus of the
masonry composite) that may be difficult to obtain [4]. The extracted
modal parameters were then used to tune a FE model. The tuned
model was used for the seismic assessment by means of nonlinear
static pushover analysis.
Another aspect of the research is the dynamic monitoring of the

cathedral. It was instrumented with a network of critically located
high-sensitivity accelerometers. This system has allowed the cap-
ture of the dynamic response for ambient vibration and also in the
occasion of some seismic events with different epicentre distance
to the building location.
Mallorca cathedral
The construction of Mallorca Cathedral (Figures 1 and 2) started around

the year 1300. The Trinity Chapel (part A in Figure 2) was completed
in year 1311 while the Royal Chapel (part B in Figure 2) was finished
around the year 1370.
It was then decided to modify the design from that of a single nave
building to a three-nave one. Unfortunately, the reason behind this dra-
matic change is not known with certainty. The imposing main large nave
(part C in Fig. 2) and the west facade were completed by the year 1601.
Figure 1 — General view of

Mallorca cathedral showing
south facade and apse
Figure 2 — Plan of the cathe-

dral indicating the main parts
and the measured points in the
dynamic identifications tests
The main nave is composed of a central nave and two lateral naves
surrounded by a series of lateral chapels built between the but-
tresses. The central nave spans 19.9 m and reaches 43.9 m at the
vaults’ keystone. The two lateral naves span 8.72 m each and reach
a height of 29.4 m. The naves are sustained on octagonal piers with
a circumscribed diameter of 1.6 or 1.7 m and a height of 22.7 m to
the springing of arches. The bell tower is 45 m high and has a square
cross section with side length of 12.5 m.
The building has been subjected to important repairs throughout its

history. During the XVIII and XIX centuries a significant number of
vaults of the central nave were repaired or even reconstructed. The west
facade was totally demolished and reconstructed in the XIX century [5].
Dynamic identification tests
Description of tests
Three tri-axial force balance accelerometers were used to carry out
the ambient vibration dynamic tests. Two of them correspond to the
CMG-5T model with dynamic range of 140 dB for 0.005 to 0.05 Hz
and 127 dB for 3 to 30 Hz, bandwidth ranging from DC to 100 Hz,
full scale from 0.1 to 4 g, and weight of 2.7 kg. The other one is Titan
model with dynamic range of 166 dB for 1 Hz and 155 dB for 3 to
30 Hz, bandwidth ranging from DC to 430 Hz, full scale from 5 V/g to
80 V/g, and weight of 960 gm.
Acceleration records were measured in 25 points in 15 different set-

ups; see Table 1 and Figure 2. The test time was chosen to be approx-
imately 1000 times the fundamental period of the cathedral [6]. The
fundamental period was found analytically and experimentally by [3].
The recording time for each setup was 15 minutes with 100 samples
per second. A 9-channel Digital to Analogue Converter was used for
data acquiring.
Setup 1 2 3 4 5 6 7 8
Points P3 P6 P20 P9, P1 P2, P5, P8,

P12 P4 P7 P10
Setup 9 10 11 12 13 14 15
Table 1 — measured points in
each setup (p16 is measured Points P11, P15, P19, P22, P24, P14, P23,
in all setups) P13 P17 P21 P23 P25 P18 P25
Results of tests
The Frequency Domain Decomposition (FDD) technique [7] was used
for determining the modal parameters (natural frequencies and mode
shapes). In this technique the Power Spectrum Density (PSD) matrix is
firstly evaluated then decomposed with the Singular Value Decompo-
sition (SVD) method.
The recorded signals were processed by a decimation of 1 (Nyquist fre-

quency of 50 Hz), the number of lines between 0Hz and the Nyquist
frequency is 1024, and the Hanning window overlap is 66,67%. The
peaks related to resonant frequencies were picked (Figure 3), and then
the corresponding mode shapes were defined.
It was not possible to identify all the modes. The identified modes in
each individual setup and in all setups processed together were inves-
tigated. Based on that, it was concluded that the fourth, fifth, sixth,
seventh and eighth modes were reasonably identified. Those modes
were considered for the modal matching and the updating process of
the FE model. It is noteworthy that the identified frequencies were
confirmed by the long term monitoring of the cathedral, performed for
nine continuous months just after the dynamic identification tests, [8].
Figure 3 — FDD of all set-

ups processed together and
picked peaks
Model updating
The FE model previously used by [9] was used. The same properties of
materials were also used as initial values in the updating process. The
model was built in Diana code [10].Vaults were modelled using T15SH
elements, three-node triangular isoparametric curved shells. The rest
of the cathedral was modelled using TE12L elements, four-node three-
side isoparametric solid pyramid. The model includes 149248 nodes
and 491851 elements with 490789 degrees of freedom (Figure 4).
In the model updating process, experimental and numerical frequen-

cies and mode shapes were compared. The mode shapes were com-
pared using the Modal Assurance Criterion (MAC) [11], defined as:
n 2
e e
i i
i l
MAC e ,n (1)
n n
e 2 n 2
i i
i l i l
In which e and n are the experimental and numerical mode shape vec-
tors, respectively. A MAC value less than 0.40 is considered a poor match
while MAC value greater than 0.80 is considered a good match [12].
So far, two updating steps have been carried out. In the first step,
we introduced springs to compensate for un-modelled parts. Those
parts include mainly the vaults of lateral chapels, the adjacent building
to the tower and the trinity chapel. In the second step, the Young’s
moduli (E) of different materials were modified. Figure 5 shows the
comparison between numerical and experimental mode shape with
Figure 4 — FE model of
Mallorca cathedral
the highest MAC value. Table 2 summarizes the updating process

results. In terms of frequency difference, the updating is sufficient, but
in terms of MAC values more updating steps are required.
Seismic analysis
The seismic analysis was carried out by the static nonlinear pushover
method. In this method, a monotonically increasing horizontal load is
applied under constant gravity load. The horizontal load distribution
adopted was a uniform load proportional to the structural elements’
masses. The cathedral was subjected to seismic loads in the longitudi-
nal (X-direction) and transversal (Y-direction) considering both posi-
tive and negative sings (see Figure 4 for axes directions).
Figure 5 — Comparison
between second num. mode
(left) and fourth exp. mode
(right)
The expected spectral acceleration is about 0.1 g according to [9]. The

procedure is well known and is both proposed by Eurocode 8 and the
Spanish seismic code (NCSE-02); it has also been applied previously to
historical structures, e.g. [13] and [14].
Num. Exp. mode Initial model Last Step

mode no. F F Error MAC F (Hz) Error MAC
no. (Hz) (Hz) % %
2 4 1,563 1,592 1,9 0,697 1,624 3,9 0,713

3 5 1,904 1,694 11,0 0,363 1,868 1,9 0,248
7 6 2,246 2,116 5,8 0,137 2,263 0,8 0,308
9 7 2,393 2,446 2,2 0,326 2,372 0,9 0,378
10 8 2,686 2,638 1,8 0,467 2,576 4,1 0,498
Table 2 — Model updating Aver- 4,54 0,398 Aver- 2,32 0,429

results age age
To simulate nonlinear behaviour of masonry, both cracking (tensile

regime) and crashing (compressive regime) were considered in the
material model. Tensile regime was modelled using tension cut-off
with multi-directional fixed crack model (smeared cracking). The pre-
vious model was accompanied by isotropic plastic Drucker-Prager
model in compressive regime. The nonlinear properties of materials
were guided by the values previously used by [9] and [15].
The updated values of Young’s moduli (E) were also used. The tensile
strength (f t ) was assumed as 5% of compressive strength (f c ). The cohe-
sion (C) was calculated from angle of internal friction (ø) and angle of
diltancy () assuming associative plasticity. The materials parameters
are specified in Table 3.
Structural E  fc u C =ø
elements (GPa) (Density) (MPa) Ultimate (MPa)
kg / m³ strain
Columns
and flying 13,6 2400 8 0,1% 3,36 10º
arches
The rest
Table 3 — Properties of differ-
of the 3,4 2100 2 0,4% 0,84 10º ent materials used in nonlinear
cathedral seismic analysis.
Seismic response of the cathedral

The structure shows different seismic capacity depending on the direc-
tion of the applied seismic forces. When the seismic forces are in the
transversal direction, the building can resist it with eight frame-like
structures including imposing buttresses. These frames show large
capacity because the forces are applied in its more resistant (in plane)
direction. On the other hand, when seismic forces are applying in the
longitudinal direction the structure shows a lower seismic capacity
because of the loading of buttresses in a direction perpendicular to
their plane and the large windows in the clerestory walls.
Figure 6 shows the capacity curves obtained. In both X and Y direc-

tions the structure behaves linearly up to seismic load about 0.05 g
with clear higher stiffness for an earthquake in the Y-direction, which
is the stronger direction of the structure. The collapse occurs at seismic
load of 0.08 g in X-direction and 0.12 g in Y-direction. The curves are
saw-tooth like. This can be related to the number of insufficient inte-
gration points through the thickness (1 point for solids and 3 points
for shells were utilized), the usage of relatively coarse mesh in the
damaged regions, and the opening of several cracks. This type of saw-
tooth curves have also been obtained in other studies [16] and [17] on
historical structures.
The damage experienced by the structure under seismic forces applied

in the transverse +Y direction, at the last step of analysis, is described
includes the following aspects:
–– The pillars are showing cracks at their bases, at both the com-
pressed the tensioned sides and the springing of arches (Fig. 7).
Figure 6 — Capacity curves

under seismic loads in differ-
ent directions
–– Diagonal cracking in southern buttresses, starting from most ten-

sioned zones and passing through windows openings. Compres-
sion damage also appears on the other side of the buttresses in
the most compressed zone. Northern buttresses are showing also
diagonal cracks around windows (Figure 7).
–– Cracking along the full span of flying arches. The cracks are more
intensive at the connections with buttresses and clerestory walls
(Figure 7).
–– Intensive cracks in vaults of lateral and central naves. This finding

is matched with historical documentation of several collapses and
reconstruction of vaults.
For the case of seismic load in +X-direction, and at the last step of
analysis, the damage is characterized by:
–– Diagonal cracks passing around windows’ openings in upper clere-

story walls connected to pillars, and lower clerestory walls conected
to buttress (Fig. 8). Diagonal cracking in apse walls (Fig. 8).
–– Cracks at pillars bases and at its connection with clerestory wall.
–– There are intensive cracks in vaults like the earthquake in +Y

direction.
Figure 7 — Crack patterns

(in blue) in buttresses, pillars
and flying arches (case of +Y
earthquake)
Figure 8 — Crack patterns

(inblue) at apse walls and
frame of buttresses and lower
clerestory wall (case of +X
earthquake)
Continuous dynamic monitoring
The monitoring system was chosen to continuously measure, record,

and wirelessly transfer the records of the accelerations without hav-
ing to set up an activating threshold. It was decided to implement
this type of monitoring because the amplitude of the seismic motion
expected in the island of Mallorca is low to moderate and may be sim-
ilar in magnitude to frequent wind effects. The system is composed of
digitizer, data acquisition system, GPS antenna, an internet router and
the three tri-axial accelerometers previously used in dynamic identifi-
cation. Two accelerometers were kept above two arches of the main
nave: the first (called S1 Station) was kept at point P6, and the other
at point P16 (called 145 Station), see Figure 2. The third accelerometer
(called Soil Station) was kept as shown in Figure 2. The system worked
from 17th December 2010 to 13th September 2011.
Response during recorded earthquakes

The website of European-Mediterranean Seismological Centre
(http://www.emsc-csem.org/) has been checked periodically for the
occurrence of any earthquake with significant magnitude that might
produce a recordable effect to the cathedral. From following some
seismic events, it has been noticed that regional earthquakes with
magnitude higher than 4, and teleseismic earthquakes with magnitude
higher than 8 are of interest. For this purpose, the time-frequency dis-
tribution of each channel has been computed (Figures 9 and 10) to
determine whether the earthquake arrived with enough energy.
In the post-processing of the information for any earthquake, the

time-frequency distributions are calculated by 100 seconds Hanning
windows, which is considered enough to obtain appropriate fre-
quency resolution and to avoid side lobes. Power spectral densities,
coherence and two transfer function estimators (H1 and H2) for differ-
ent combinations of pairs of channels of (S1 Station-Soil Station) and
(145 Station-Soil Station) are computed before, during, and after the
considered seismic event. Frequencies with coherence higher than 0.8
are considered for calculations of transfer function.
During the monitored period it has been possible to capture one

local seismic event, corresponding to the recent Menorca earthquake
(31 / 7 / 2011), and six regional earthquakes with epicentre in Lorca
(11 / 5 / 2011), Alagüeña (10 / 7 / 2011), Gulf of Lion (2 / 7 / 2011 and
7 / 7 / 2011) and Northern Italy (17 / 7 / 2011 and 25 / 7 / 2011). Also, it
has been possible to register the teleseismic earthquake of Honshu
occurred in Japan (11 / 3 / 2011).
Considering the range of interest of natural frequencies of the cathe-

dral, it was found that the effective duration of regional earthquakes is
less than 250 seconds and for the local ones is less than 100 seconds.
Also, It was noticed that only the teleseismic earthquake of Honshu
arrived with very low range of frequency contents (less than 1,5 Hz),
whereas all the other captured regional and local earthquakes had
frequency contents able to excite the range of interest of natural fre-
quencies of the structure (Figure 9).
Some effects observed in the dynamic response, even for very low
amplitude motion, are attributed to the non-linear response of the
building. Hence, the appearance of multiple close peaks for the same
mode in the spectral diagrams is believed to be due to a breathing
behaviour caused by existing cracks [18]. This phenomenon can be
seen, for instance, in the 4th mode during the occurrence of the
Menorca earthquake (Figures 10 and 11). Due to the short effective
duration of the captured earthquakes, the uncertainty in the power
spectral densities is greater than the shift experienced by the peaks.
The change of frequencies produced by earthquakes is less than 1%.
Figure 9 — time-frequency
distribution of the NS com-
ponent at Soil Station for the
Gulf of Lion earthquake
Figure 10 — Time-frequency
distribution of the EW com-
ponent at 145 Station for
Menorca earthquake
Figure 11 — Transfer function,

coherence and power spectral
densities of EW component
at 145 Station for Menorca
earthquake, before earthquake
with calm wind and windy,
and during earthquake
Conclusions
The paper presents an application of ambient vibration dynamic iden-

tification tests on a complex structure. Only well identified mode
shapes and frequencies were used for the purpose of updating the
global FE. The modal shape of each associated pair of experimental
and numerical modes was visually compared to assure an adequate
correspondence. In terms of frequencies, the updating process was
sufficient and satisfactory. In terms of MAC values, it can be consid-
ered only acceptable. Many effects may be affecting the obtained MAC
values, among which the soil-structure interaction, not considered so
far in the modal, and the existing cracks. In addition to include the
influence of the soil, it is intended to improve the model by simulating
the main existing cracks as discontinuities in the FE mesh.
The tuned model was subjected to seismic loads in both the transversal
and longitudinal directions using nonlinear static pushover analysis.
The curves showed that the cathedral has higher resistance in trans-
versal direction than in longitudinal one. The resisted acceleration in
transversal direction is a little higher than the expected spectral one,
whereas in the longitudinal direction it is slightly lower.
The use of a long term dynamic monitoring in Mallorca Cathedral has

provided a better understanding of the dynamic behaviour in spite
that only very low motion earthquakes registered since its implemen-
tation in December 2010. For the very low motion earthquakes reg-
istered, the only non-linear effect so far observed is the doubling of
some frequency peaks. This peak doubling is probably due to a breath-
ing crack effect. It is expected that the capture of higher motions in
the near future may allow a better characterization of the dynamic
response including such non-linear effects.
Acknowledgements
This research has been carried out within the project “New Integrated
Knowledge based approaches to the protection of cultural heritage
from Earthquake-induced Risk-NIKER” funded by the European Com-
mission (Grant Agreement n° 244123), whose assistance is gratefully
acknowledged.
References
[1] Roca P. (2001) Studies on the structure of Gothic Cathedrals. In:

Proc. 3rd Int. Seminar on Historical Constructions Guimarães, Portugal,
University of Minho, 71–90.
[2] González R., Caballé F., Domenge J., Vendrell M., Giráldez P., Roca P.
and González J.L.(2008) Construction process, damage and struc-
tural analysis. Two case studies. In D‘Ayala D. and Fodde E. (ed)
Structural Analysis of Historical Construction. CRC Press Balkema,
643–651.
[3] Martinez G., Roca P., Caselles O. and Clapés J. (2006) Characteriza-
tion of the dynamic response for the structure of Mallorca cathedral.
In Lourenço P., Roca P., Modena C. and Agrawal S. (ed) Structural
Analysis of Historical Constructions. Macmillan Advanced Research
Series.
[4] Ramos, L.F., Alaboz, M., Aguilar, R. and Lourenco, P.B. (2011).
Dyanmic identification and FE updating of S.Torcato Church, Portugal.
Dynamics of Civil Sturctures. Volume 4, 71–80.
[5] Domenge J., (1999). L’obra de la Seu. El process de construcció de la

Catedral de Mallorca en el tres-cents. Institu d’Estudis Balears,
Palma de Mallorca (in Catalan).
[6] Ramos, L. F. (2007). Damage identification on masonry structures

based on vibration signatures. PhD thesis, University of Minho,
Portugal.
[7] Brincker, R., Zhang, L. and Andersen, P. (2000). Output-only

Modal Analysis by Frequency Domain Decomposition. 25th Interna-
tional Conference on Noise and Vibration Engineering. Vol II.
[8] Elyamani, A, Caselles,J,O, Clapes,J, & Roca,P, Assessment of

Dynamic Behavior of Mallorca Cathedral, 8th internation conference
of structural analysis of historical constructions, 15–17 Oct, 2012,
Wroclaw, Poland,
[9] Martínez, G. (2007) Seismic vulnerability for middle and long span
masonry historical buildings (in Spanish). PhD thesis, Technical Uni-
versity of Catalonia, Barcelona,Spain.
[10] TNO DIANA BV 2005. DIANA-Finite Element Analysis, The Neth-

erlands.
[11] Allemang, J. R. (2003). The modal assurance criterion — twenty

years of use and abuse. Sound and Vibration 37:8,14–21.
[12] Gentile, C. and Saisi A. (2004) Dynamic-based F.E. model updating

to evaluate damage in masonry towers. Proc. of 4th Int. Seminar on
Structural Analysis of Historical Constructions pp. 439–449.
[13] Betti, M. and Vignoli, A. (2008). Modelling and analysis of a

Romanesque church under earthquake loading: assessment of seismic
resistance. Engineering Structures 30:2, 352–67.
[14] Elyamani, A. (2009). Wind and earthquake analysis of spire of cim-

borio of Barcelona cathedral. MSc thesis, Technical university of
Catalonia, Barclona.Spain.
[15] Clemente,R. (2006). Structural analysis of historical buildings using

localized crack models (in Spanish). PhD thesis, Technical University
of Catalonia, Barcelona,Spain.
[16] Rots, J.G. (2001). The role of structural modelling in preserving

Amsterdam architectural city heritage. 3rd International Seminar
on Historical Constructions. 685–696.
[17] Trujillo,A. (2009) Stability analysis of Famagusta Churches: St.

George of the Latins. MSc thesis, University of Minho, Guimarães.
Portugal.
[18] Chondros T. G., Dimarogonas A. D., and Yao J. (2001) Vibration of

a beam with a breathing crack. Journal of Sound and Vibration,
239(1): 57–67.
Seismic Strengthening of Masonry Infilled Reinforced Concrete Frames 77
Seismic Strengthening of
Masonry Infilled Reinforced
Concrete Frames
Abstract GRUBIŠIĆ, Marin

J. J. Strossmayer University
Structural frames are often filled with masonry panels for architec-
of Osijek, Croatia
tural purposes as divisional walls or cladding. There has been much
work conducted into the seismic behaviour of infilled frame build- SIGMUND, Vladimir
ings (e.g., Abrams, 1994–1996; Chrysostomou et al, 1992; Žarnić and Faculty of Civil Engineering,
Tomažević, 1984). However, the most significant outcome is perhaps J. J. Strossmayer University
of Osijek, Croatia
the general consensus that brickwork infill can have a beneficial effect
on the overall seismic performance of the building if it is properly tied
into the rest of the building.
The purpose of the present study was to investigate possible seismic

strengthening options for use in the seismic upgrade of a infilled
frame structures with an emphasis on strengthening brick masonry
infill walls. In order to test the numerical models and to define a sim-
ple method for analysing the behaviour of reinforced concrete frames
with infill in earthquake regions, we have done a series of tests with
the aim to investigate the behaviour of strengthened infills of r /c
frames under seismic loads.
Model frames represent part of a model structure, designed according

to the EN 1998. Here we consider six one story — one bay specimens,
built in a scale 1:2.5; infilled with two types of masonry (hollow and
solid clay masonry units) and three types of strengthening methods,
tested under constant vertical and cyclic in–plane lateral loads. Pre-
sented is the relationship between drift capacity and properties of the
frame–wall system controlling drift capacity.
Keywords: reinforced concrete (r /c) frame, masonry infill, seismic

strengthening, infilled r /c frame, experimental testing
Introduction
Infilled frame structures

As mentioned, in many countries situated in seismic regions, rein-
forced concrete frames are usually infilled by brick masonry panels.
For seismic design and evaluation of existing buildings, the contribu-
tion of infill walls is usually neglected although it can be — and usually
is beneficial. Its contribution is of special importance also for seismic
strengthening of r /c frame buildings. Until today, a large number of
seismic strengthening projects have been documented and provide an
excellent database for practising engineers. The strengthening meth-
ods used naturally reflect the state-of-the art at the time. A selection
of case studies encompassing a wide range of strengthening options,
of which are most commonly referred and discussed, are beam–col-
umn joint and connection strengthening, column strengthening and
many types of frame bracing. Therefore, we can see that strengthening
methods of masonry infill are not sufficiently investigated, especially
experimentally, and the investigation of these is necessary.
The complex nature of the interaction between concrete frames and

masonry infill wall panels is reflected perhaps in the large number of
experimental studies conducted on this topic. The results of previous
work suggest that the brickwork can be accurately modelled using
“equivalent struts”, but still is a common principle of structural anal-
ysis that often regards the masonry infill as a non-structural (archi-
tectural) element. So, moment frame members are designed for the
full lateral loads while the contribution of infill masonry panels is
usually neglected, and this approach may represent a major threat
considering different mechanisms of the response and seismic loads
transmission for moment resisting bare frame and fully masonry
infilled frame, shown in Figure 1.
Since the nonlinear behaviour is in close relation with the connection

between the frame and infill, it is very difficult to predict it by analytical
methods if the analytical models are not supported and verified by
the experimental data. Due to the complex behaviour of such com-
posite structures, experimental research is of great importance to
determine the strength, stiffness and dynamic characteristics at each
stage of loading.
Figure 1 — Mechanisms of
response and seismic loads
transmission for moment
resisting bare frame and fully
masonry infilled frame.
Strengthening of Masonry Infills

In multi-story construction, the most important attribute of the struc-
ture is its capability to retain its integrity at interstorey drifts ranging
from 1.5% to 2.0%. The results of neural-network analysis (Sigmund,
V., Kalman, T., 2009) based on approximately 100 tests of one bay —
one story infilled frames have demonstrated that interstorey drifts of
that magnitude can be achieved by a reinforced concrete frame with
solid infill walls provided the columns have the ability to sustain the
required shear force under reversals of shear and axial forces.
There has been a substantial increase in the topic of seismic retrofit

or strengthening of existing buildings in recent years as evidenced by
the growing number of research papers published in this area. The
overall aim of many projects was to identify the optimal combina-
tion of strengthening options that would enable the building to meet
the present–day “life-safety” performance criteria for buildings subject
to a design magnitude earthquake. Today we have growing need for
strengthening of existing structures, and infills contribution is of spe-
cial importance for seismic strengthening of r /c frame buildings.
Often mentioned out-of-plane strength of unreinforced masonry infill

walls is well-recognised as being one of the major weaknesses of these
types of infills with regard to seismic actions. However, masonry infill
walls must not fail in the out-of-plane direction if they are to maintain
their in-plane load capacity. Aničić (1995) reports that horizontal ties,
tie beams and / or columns and “rigid” floor diaphragms are normally
effective at preventing out-of-plane collapses of masonry walls pro-
vided the span-height ratio and height-thickness ratio are kept within
normal design limits. Therefore, we have observed only in-plane
strengthening methods.
Where the masonry infill is susceptible, it can be strengthened in a

wide variety of ways. While recent research trends are towards the
use of advanced fibre composites, energy dissipation devices and seis-
mic isolation schemes for the seismic strengthening of buildings, the
more traditional methods should not be neglected when considering
which system(s) to employ. Within the scope of our research, we have
planned to investigate possible strengthening methods, of strong and
weak frames, by application of the masonry infill. Up to now, follow-
ing strengthening methods of the strong r /c frame, were investigated
and shown in Figure 2:
a) Masonry infill anchored to the surrounding r /c frame

(Type 1 / I – hollow, Type 1 / II – solid),
b) Reinforced masonry infill anchored to the surrounding r /c
frame (Type 2 / I – hollow, Type 2 / II – solid), and
c) Masonry infill wall with additional confining element
(Type 3 / I – hollow, Type 3 / II – solid).
Figure 2 — Seismic strength-

ening methods applied on six
test specimens
Tests were performed at the Laboratory for experimental mechanics

at the Faculty of Civil Engineering in Osijek, Croatia. The results of
strengthened specimens are compared to the previous experiments
when the r /c frame was without infill (Model TYPE 0) and with infill
with no connection to the frames (Model TYPE 1). The main goal of
this experiment was to determine the contribution of each method
of strengthened infill to the carrying capacity of infilled frames under
horizontal actions and the actual behaviour of the infill panel itself.
Description of the prototype
In order to separate the experimental infilled frame model, a rein-

forced concrete frame structure with ground plan dimensions of
16.00 m x 17.50 m was used, as shown in Figure 3.
When calculating and designing the prototype, weight of infills are

taken into account using the appropriate line load, but infills alone
was intentionally not modelled. Dead weight of all the structural ele-
ments was taken into account by Radimpex Tower 6 in the physical
spatial 3D design model, so we obtain the internal forces and design
of frame members. Modal analysis, seismic analysis and design of r /c
elements were made according to latest EN 1992 and EN 1998 stand-
ards. Selected sections and reinforcement amounts of frame columns
and beams are given in Figure 4.
Figure 3 – Model of the pro-

totype frame building
Figure 4 — Selected dimen-

sions, cross sections and rein-
forcement of the frame build-
ing members, columns (1–1)
and beams (2–2 and 3–3)
Test specimens
Within the research project a total of six r /c frame models with
strengthened infill, one bare r /c frame model (without infill) and one
frame with infill who has no connection to the frames — were con-
structed and tested through in-plane cyclic loading. The specimens
are a scaled part of a model 7-story building (middle frame of the
“Tsukuba” building) located in the ground floor and constructed in a
scale 1:2.5 as shown on Figure 5. Two types of infill walls with different
strength properties were used: the high strength solid bricks (Opeka
Osijek), medium strength hollow brick blocks (Sladojevci). Mechanical
properties of the infill blocks were tested on three samples each. The
compressive strength of concrete was obtained by testing the cube
size 15 x 15 x 15 cm, and the compressive strength of the infill ele-
ments can be found in [3]. Comparison of the results obtained by test-
ing the r /c infilled frame models with two different infill types and
those obtained by testing the r /c bare frame model was made. The
test models were 2,2 m wide and 1,5 m high, with columns and beam
cross section of 20 x 20 cm and 12 x 20 cm, respectively. Materials
used were concrete of class C 30 / 37 and reinforcement of type B500B.
Figure 5 — Selected dimen-

sions and the position of the
instruments on the test model
R /C columns rebar percentage is 2,36%, while the rebar percentage

of r /c beams are 1,31% in the middle of the beam span and 3,27%
at its ends. Masonry panels were built subsequently, i.e. frame ele-
ments were made first and then the masonry infill walls were added
with frame and infill in close connection without spacing. They after
the r /c frame was done and hardened, and they were made by the
cement-lime mortar (cement-lime-sand in volume proportion 1:1:5
and nominal strength of 3 MPa) with thicknesses of horizontal joints of
1 cm and completely filled vertical joints in a manner typically used in
Croatia. After 28 days infilled frames have been tested under vertical and
reversed in-plane cyclic loading. Test models are designed according to
EN 1992 and EN 1998 standards and according to the common practice
while their experimental results were analysed and compared as follows.
Test setup and procedure
Before testing, we set the measuring devices and hydraulic presses as

shown in Figure 5. We continue to set four load converters, eight LVDT
to measure drift, and four deformeters (extensometers) to measure
deformations of columns and beams, and of course — all the measur-
ing instruments are connected to the computers and monitored by
software Dewetron DeweSoft 6.6.7. Columns of the test specimens
were loaded with constant vertical load that replaced the stories
above. Scaled load applied at the column tops was 365 kN in each col-
umn. Horizontal loading has been applied in-plane as reversed cyclic
load, and it steadily increased by increment of 10 kN as shown on Fig.
6. Vertical load are slightly oscillated during the test due to the addi-
tional moment introduced to the frame by horizontal load. During the
test measured were: loads at each point, vertical and horizontal dis-
placements at the top of the frame, sliding of foundation and exten-
sion / shortening of diagonals, deformations in the columns and beams.
Figure 6 — Example of applied

lateral cyclic loading and
changes of vertical forces due
to specimen rotation for
one frame
Figure 7 — View of the exper-

iment in progress and asso-
ciated samples of masonry
and mortat
Model TYPE 1 Model TYPE 1-II
Model TYPE 2-I Model TYPE 2-II
Model TYPE 3-I Model TYPE 3-II
Model TYPE 0 (Bare frame) Envelopes of all models
Figure 8 — Review of hyster-

etic curves and their enve-
lopes for all models
Results
Part of the measured results is present in a form of base shear vs.

horizontal displacement of the frame for three types of strengthened
frames with two types of infills. The hysteretic curves and envelopes
are presented in the following Figure 8.
Conclusions
The behaviour of the infilled frame under seismic loading is difficult

to predict by analytical methods unless the analytical models are sup-
ported and revised by the experimental data. Within the scope of
a larger project we have tested six r /c frames infilled with two types
of masonry infill, solid and hollow brick units. The frames were
made with standard materials and procedures as used in Croatia.
From the measured results could be observed that any strengthen-
ing method enhances the initial stiffness of a composite structure
and allows larger drifts with less loss of stiffness compared with
models without any strengthening.
Thus strengthening models can withstand much greater interstorey

drifts at almost maximum lateral load level, while infill panels will
reach their deformation capacity and lose their load carrying ability
at much larger displacements, where we can see model TYPE 3-I that
can withstand up to 2.6% IDR. Increase in the load carrying capacity
depends on the strengthening method and brings from 5 to 15% in the
increase. It is important to notice that much larger energy dissipation
is prominent in the models which infills contains solid brick elements,
and thus we observe much better effect of mortar sliding with regard
to out-of-plane face dropping of hallow clay blocks.
Acknowledgements
The research presented in this full paper is a part of the research project
“Seismic design of infilled frames”, No. 149-1492966-1536, supported
by the Ministry of Science, Education and Sports of the Republic of
Croatia and its support is gratefully acknowledged.
References
[1] Griffith M.: Seismic Retrofit of RC Frame Buildings with Masonry

Infill Walls: Literature Review and Preliminary Case Study, JRC
European Commission Report, JRC 44166. 2008.
[2] Kalman, T., Sigmund, V.: Seismic capacity of infilled frames using
neural networks, 6th ICCSM Proceedings, Croatian Society of
Mechanics, Zagreb, 2009.
[3] Matošević, Đ., Sigmund, V., Zovkić, J.: Experimental Testing of

Masonry and Masonry Piers, 6th ICCSM Proceedings, Croatian Society
of Mechanics, Zagreb, 2009.
[4] Sigmund, V., Zovkić, J., Sigmund, Z.: Experimental Tests of R /C
frames with Masonry infill, 14ECEE, Ohrid, Macedonia, 2010.
[5] El Gawady, M., Lestuzzi, P., Badoux, M.: A Review of Conventional

Seismic Retrofitting Techniques for URM, 13th International Brick
and Block Masonry Conference, Amsterdam, July 4–7, 2004.
Time dependent analysis of steel-concrete composite beams 87
Time dependent analysis of

steel-concrete composite beams
Abstract HEGYI, Péter

University of Technology and
Economics, Budapest
In this paper the phenomenon of the creep of steel-concrete compos-
ite beams is investigated. Some of the available material models are
briefly introduced, including the Dischinger-theorem and the ageing
linear viscoelasticity theorem. Based on these material models differ-
ent structural models were developed of which five are involved in this
paper: the Fritz-method, a finite element approach, numerical inte-
gration methods for both material models and the codified formulae
of the Eurocode. Analyses were carried out considering a three-step
load sequence of a simply supported beam with the assumption of the
Bernoulli-Navier hypothesis. Calculations were performed for different
widths of concrete slab to get information about the behaviour of each
model. The results were compared to the values based on the Eurocode.
Introduction
A steel-concrete composite structure consists of a steel girder and a

composite slab, connected with shear connectors along the girder, so
both of the components contribute to the load-bearing capacity of
the composite girder. These structures became frequently used after
the Second World War based on the several advantages (e.g. higher
stiffness than steel structures, less self-weight as concrete structures,
unpropped construction) which make them very competitive and
economic in different parts of the building industry. One of the dis-
advantages is that the modelling and analysis of a composite girder
is more complex than in the case of a homogenous structure. The
cases of some bridges show that the time dependent behaviour of the
concrete should be included in the analysis accurately. In Palau [1] a
prestressed box girder bridge (main span 240 m) was built, which pro-
duced 1.2 m deflection after 13 years of service, and it was predicted
in a detailed analysis that an additional 0.9 m deflection will develop

over the next 100 years. Another example is in Turku (Finland) [2],
where a slender 90 m long beam leg bridge was built in 1975, which
produced twice the deflection that was estimated during the design.
The concrete cracked in the support zone so the strengthening of the
bridge became necessary. The aforementioned and also some further
examples show the importance of the time dependent behaviour of
concrete on the structural response, therefore, it is necessary to take
it into consideration during the analysis. This makes the modelling of
the composite structures even more difficult.
In this paper different methods of analysis are presented which

take into consideration the time dependent behaviour of the con-
crete slab. Some of these methods were developed with significant
amount of simplification to reduce the computational requirements
which made them suitable in the practical calculations. Apart from
these methods, more accurate solutions are also introduced. Calcu-
lations are carried out on the case of a simply supported beam to
compare these methods to each other.
Time dependent behaviour
The time dependent behaviour causes two problems, which are inter-
related but frequently considered independent during the analysis [3]:
–– material behaviour: the prediction of creep and shrinkage strains;

–– structural behaviour: the determination of the structural response.
To treat with these problems different models have been developed.

This paper deals with mainly the structural problem.
Dischinger-theorem
The first basic models describing the creep behaviour were developed
in the 1930s, including the Dischinger-theorem [4]. In this model one
single creep coefficient-curve is determined taking into consideration
the age of concrete at the first loading time ((t)). Then the creep
strains are calculated based on this one curve. (see Fig. 1 and Eq. (1)):
(1) cr t , t0 t , t0 el t0 t t0 el t0
where: t0 is the time of the loading, and t is the actual time;

el (t0) = c (t0) / Ec is the elastic strain caused by the imposed
c stress at t0;
(t, t0) is the creep coefficient at time t for a sustained cons-
tant load imposed at time t0;
cr (t, t0) is the creep strain at time t from the el elastic cons-
tant strain imposed at time t0.
Figure 1 — The Dischinger-

theorem
With the use of the Dischinger-theorem the creep strain is considered

permanent after unloading. This behaviour does not coincide with the
experimental results, which shows a creep recovery after unloading.
On the contrary, this formulation allows the usage of simple methods
for calculating the creep response of structures.
Ageing linear viscoelasticity (ALV)

As the computational capacity increased, the ageing linear viscoelas-
ticity (ALV) theorem was adopted to describe the behaviour of the
concrete. This theorem already considers the ageing of the material.
With the application of this more realistic theorem new creep pre-
diction methods were developed, which differs from one another in
the parameters which they take into account to determine the creep
coefficient. These methods include parameters such as the relative
humidity, notional size, cement type, cross sectional shape, aggregate
content, compressive strength of concrete, etc. Different methods can
lead to rather different results, as shown in Figure 2.
Figure 2 – Different compli-

ance functions
The main equations of the ALV material behaviour are shown in Eq. (2)
(for creep) and Eq. (3) (for relaxation):
t
t J t, d c (2)
0
t
c t R t, d (3)
0
where: c is the total stress;

 is the total stress dependent strain (including elastic and
time-dependent strain);
J(t, ) is the compliance function (strain response at time t to a
unit imposed stress applied at time );
R(t, ) is the relaxation function (stress response at time t to a
unit imposed strain applied at time ).
This material model fits to the experimental results better than the
previous models. To solve these integral equations a variety of meth-
ods were developed, which will be described in chapter 3.2.
Analysis methods
Dischinger-theorem
Traditional methods
As the steel-concrete composite structures came into general use, sev-
eral methods were developed to consider the long term behaviour of
the structural type based on the Dischinger-theorem. The most accu-
rate among these methods is the Fröhlich-method, which is an ana-
lytical approach and uses exponential-type equations [5]. Since this
method was found too complicated to use in the practical calculations,
Sontag applied a simplification. He stated that the bending stiffness
of the concrete slab is negligible, so the bending moment of the con-
crete slab can be eliminated from the equilibrium equations. By this
simplification a more simple equation could be obtained, however, it
was still deemed rather complex.
Fritz described the effective modulus of the concrete with the aid of
a correction factor (), which was defined based on Sontag’s results:
Ec
(4) E c,
1
where the reduction factor for permanent load is
e v
1
(5)
v
and
1
v
Ec A c Ec A c
1 a2
Ea Aa Ea Ia
where the subscription ‘a’ stands for steel and ‘c’ for concrete, whilst a
is the distance between the centroid of the steel and concrete.
To simplify the calculations the stiffness-dependent correction fac-

tor was replaced with a constant coefficient. This was determined
for permanent load as 1.1 considering average stiffness properties
(for shrinkage it was defined as 0.52). These values can be found
in the corresponding Eurocode (EC4 1–1). This is the most generally
used method in the manual calculations and was widely used before
the FEA softwares became capable of taking into consideration time
dependent behaviour.
Numerical integration
Since the values of the Fritz-method are average values they do not
describe the real behaviour of the cross sections. To identify the actual
behaviour a numerical approach was established for the case of a sim-
ply supported beam. It is based on the Dischinger-theorem to ensure
the comparability with the Fritz method.
The basic differential equation of Dischinger is shown in Eq. (6):
c
' t c t
' t t , t0 (6)
Ec Ec t
where:  id the total stress dependent strain at the centroid of the slab;
σc is the stress at the same fibre;
Ec is the elastic modulus of the concrete at the age of 28 days;
(t, t0) is the creep coefficient.
The time interval [t0; t] can be represented with sufficient amount of

time steps (ti) and so  (tn) can be expressed from Eq. (6) by replacing
the integral with a summation using Eq. (1):
c tn n
c ti 1
tn ti ti 1 (7)
Ec i 1 Ec
Supposing that the history of the strains and stresses is known until
time step tn-1, this equation contains two unknowns. Thus another
equation has to be formed, based on the compatibility and equilib-
rium assumptions (built upon the Bernoulli-Navier hypothesis):
A2 2 2
D t t B t
A 4 C
c t t (8)
2 C C
A, B, C, and D are constants, which contain the stiffness properties of

the steel (subscript ‘a’) and concrete (subscript ‘c’) part and also the
bending moment (MEd) acting on the cross section:
Ac Ic Ac
A Ac a 2 Ia Ic C
Aa Ea A a
B Ea Ia D Med a
The calculations can be performed based on Eq. (7) and (8).
ALV-theorem
Age Adjusted Effective Modulus method

The ALV methods are more complex than those based on the Dis-
chinger theorem so their usage is more elaborate. Different solution
methods were developed for this theorem, including Age Adjusted
Effective Modulus (AAEM) method which can well be used in the con-
ceptual and preliminary design stages [3]. The phenomenon of ageing
is taken into account through an ageing coefficient (i.e. (t, t0)), which
can be calculated based on the J and R functions and the elastic mod-
ulus at time t0. Using this ageing coefficient the age adjusted effective
modulus can be defined as follows:
Ec t 0
(9) Ec , adj t , t 0
1 t , t0 t , t0
where Ec(t0) is the elastic modulus of the concrete at the loading time
t0. With the aid of this modulus a quasi-elastic analysis can be performed,
which makes the AAEM method efficient for preliminary calculations:
c t0 c t c t0
(10) t
Ec , t , t0 Ec , adj t , t 0
where
Ec t 0
Ec , t , t0
1 t , t0
Its formulation is quite similar to the Fritz method but its approach
is different: the AAEM method describes the behaviour of an ageing
material under varying loads, whilst the Fritz method describes the
behaviour of a composite structure with an imposed permanent load.
Apart from AAEM, numerical solutions are also feasible for solving the
equations. One possible method is the time-history integral, where
the basic equations of ALV are solved by means of numerical integra-
tion. This method uses great computational efforts, since the whole
history of stresses and strains have to be stored during the calculation
(see 3.2.2.). A more efficient numerical method is based on rate-type
creep laws. Using this method only the results of the last time step
should be stored, so this means a more efficient computation. This

method is also efficient to adapt to FEM softwares, which makes this
method the most powerful tool in the design process because of its
adaptability. Some of the modern finite element softwares can deal with
the time dependent behaviour of concrete such as MIDAS, SOFiSTiK,
etc. With these softwares even the most complex structures can be ana-
lyzed in a more accurate and detailed way than with the Fritz method.
Time-history integral
To solve Eq. (2) a similar numerical procedure was established as in the
case of the Dischinger-theorem. From Eq. (2) the incremental stress
dependent strain can be expressed as follows:
t
c
' t t0 c
' t
'
c t t , t0 t, d (11)
Ec Ec t t0 Ec t
It can be observed that this equation shows some similarity to Eq. (6).

It is also clear that the creep strain increment depends on the actual
stress in Eq. (6), however, in the ALV-theorem it depends on the
whole history of the stress. Likewise in the case of Eq. (6), c(t) can
be expressed from Eq. (11), and the time interval can be replaced with
sufficent amount of time steps:
c tn c t0 n 1
c ti c ti 1
tn tn , t0 tn , ti (12)
Ec Ec i 1 Ec
Since the compatibility and equilibrium assumptions are no different than

in 3.1.2., the same expression (Eq. (8)) can be adopted to this analysis.
Analysing a simply supported beam
Analysis strategy
The purpose of the analysis was to compare the results of the different
models. A simply supported beam was used with a length of 30 metres.
The cross section can be seen in Figure 3. To get information about the
behaviour depending on various stiffness ratios, the analysis was per-
formed using several slab widths namely: 0.5 m, 1.0 m, 2.5 m, 5.0 m
and 10.0 m. To take into account the effect of loading, a loading pro-
cess was considered with three load steps at age 14 days, 60 days and
3,000 days, with loads 22.222 kN / m, 11.111 kN / m and 11.111 kN / m
respectively. This means a total bending moment of 5,000 kN / m at
the centre of the span. The steel is S235 type according to Eurocode,
the concrete is C 20 / 25. The Eurocode was also used to describe the
creep phenomenon. The basic parameters for the calculation of the
creep coefficient were determined as relative humidity: 80%; and
cement type: normal.
Figure 3 — The investigated

cross section (dimensions
in mm)
The calculations were carried out by obtaining the procedure of the

Eurocode (a Fritz method using  = 1.1), and also with the original
 factor of Eq. (5), determined by Fritz. These results were then com-
pared to those obtained by the numerical integration described in
chapter 3.1.2. The Dischinger-based models were compared to the
ALV-methods too, using AAEM method and a time-history integral (as
per chapter 3.2.2.). The numerical integration was carried out assum-
ing that t1 = 10–4 days after the first loading, as the end of the first time
step. Every following step’s length equals the previous multiplied by
1.005. The last day of the integral was over 200,000 days (540 years)
in order that the creep coefficients would be within 0,1% range of
their final values. The results were also compared to a finite element
model of MIDAS / Civil software, which is fully capable to consider steel-
concrete composite sections and time dependent material behaviour.
Results
 factors
At first, analyses were carried out on single load steps to compare the
 factors of the models. Based on the results of the numerical integra-
tion the  factor could be easily expressed for every considered beam.
These values are summarized in Figure 4 and 5.
Figure 4 —  factors for load-

ing at age 14 days
Figure 5 —  factors for load-

ing at age 3,000 days
It can be observed that the numerical integration results of the

Dischinger-theorem agree well with those obtained by Eq. (5), in
fact the difference is negligible. In the case of small concrete slabs
the difference between the Eurocode factor and the Fritz value may
become significant. It is also shown that the increasing of loading time
decreases the  factor. In the case of t0 = 3,000 days, the actual value
of  remains mostly under the Eurocode value.
It is clear that the  factors obtained by the ALV time-history inte-

gral are less than the ones obtained by the Dischinger-theorem. In
the case of t0 = 3,000 days it can be considered as 1.0 without any
notable loss of accuracy. Based on these figures the  factor can be
decreased not only in the case of high age loading but also when
stiff concrete slabs are used.
Altogether the value of the reduction factor  depends on the adopted

material theorem, the age of concrete when imposing the load and the
stiffness ratio of the cross section. These dependences are neglected
by the Eurocode in favour of a simple calculation method.
Inner forces
During the design process the inner forces have greater importance
than the  factors. The bending moment which acts on the composite
cross section can be divided into three components (see Fig. 6): bend-
ing moment of the slab (Mc), bending moment of the girder (Ma) and
axial force (N) (an equal force in the girder and the slab).
Figure 6 — Inner forces

Although the difference in the values of Mc was significant, their

contribution to the global bending moment is negligible. So in this
study only the bending moment of the girder and the axial force
are discussed. The results of these inner forces are summarized in
Figure 7 and 8, respectively. The values were compared with the results
obtained by the Eurocode.
Figure 7 — Bending moment

of the girder
Based on both figures it is clear that the tendency of inner forces

is analogous to the  factors (Figure 4). In the case of the bending
moment it can be observed that each model produces less than 5%
difference from the Eurocode. Moreover, the difference is around 1%
in the case of normal-width concrete slabs (between 2 and 8 metres).
The Dischinger-based numerical integral produces greater differences
with small slab width than the ALV-based ones. It can be also observed
that the time-history integral shows the same behaviour as the FEM
results (calculated with MIDAS / Civil), their difference is about 0.5%.
This similarity is more evident in the case of the normal forces.
Figure 8 — Axial force

Based on Figure 7 and 8 it can be said that the time-history inte-

gral and the finite element model yield to the same result. The third
ALV-method (i.e. AAEM) shows different behaviour than the previous
ones but its results are slightly closer to the values of the Eurocode. It
can be observed that all the models produce greater differences in the
case of axial force than of the bending moment; the maximum differ-
ence is 13%, in the case of normal slab width it reduces to 2–3%.
Both figures show that the Dischinger-theorem provides results with

different behaviour than the other models. The ALV-methods are
more complex and in the case of small slab widths they can produce
closer results to the Eurocode procedure. However, above 2 metres of
slab width the Dischinger-theorem is nearer to the results according
to EU Standards. Still, in the case of normal concrete slab width all
methods produce maximum 3% difference, which is quite small, par-
ticularly according to the differences in the material models predicting
the creep coefficient. Although, in the case of a statically determinate
structure the difference is small for each model, the usage of the finite
element software is recommended because of its greater adaptability.
Conclusion
In the design phase of a steel-concrete composite girder the long-

term behaviour of concrete plays an important role. In this paper
some available material models were discussed in order to take into
account the creep behaviour of the concrete. Structural models were
also introduced including the method suggested by the Eurocode, the
Fritz-method, the Age Adjusted Effective Modulus method and cal-
culations based on numerical integration and finite element analysis.
These were compared to each other.
In the case of the reduction factor () it was found that the value of 
reduces if the structure is loaded at high age. It was shown that the
 factor also depends on the stiffness ratio of the cross section. The
difference between the actual value of  and that is to be found in
the Eurocode may become significant. The trend of the inner forces is
influenced by the  factor, although, the magnitude of differences are
significantly smaller. The time-history integral and the finite element
model produced almost equal results. The AAEM method showed dif-
ferent behaviour and it also provides a slightly smaller difference com-
pared to the results of the Eurocode. In the case of normal slab width
each model produces around 1% difference at the bending moment
of the girder and 3% at the axial force. This difference is quite small,
particularly according to the differences in the material models pre-
dicting the creep coefficient.
Acknowledgement
The previous researches which laid the foundation of this paper were
carried out along with Balázs Somodi and with the aid of Prof. László
Dunai and Dr. Tamás Kovács as consultants. Their part in the work is
gratefully acknowledged.
References
[1] Burgoyne, C.J. and Scantlebury, R.C.: Why did Palau Bridge Collapse?,
The Structural Engineer, 84 / 11, 30–37, June 2006.
[2] Aarne Jutila: Vasbeton hidak Finnországban, Vasbetonépítés, XIII / 1,

p. 14–25, 2011 / 1 (in Hungarian)
[3] Analisys of Creep and Shrinkage Effects in Concrete Structures,

reported by ACI Committee 209 (ACI 209.3R-XX, March 2th, 2011.)
[4] Chiorino, Mario: Analysis of creep and shrinkage effects in concrete

structures, Advanced Professional Training, CISM Training, May,
23–27., Udine, 2011.
[5] Platthy Pál: Vasbeton lemezzel együttdolgozó acéltartók, Műegyetem

Kiadó, Budapest, 1983. (in Hungarian)
Multi-filament yarns testing for textile-reinforced concrete 99
Multi-filament yarns testing for

textile-reinforced concrete
Abstract KADĚROVÁ, Jana

Institute of Structural
Mechanics, Faculty of Civil
The present paper copes with an experimental study of multi-filament
Engineering, University of
yarns made of AR-glass which are used for textile-reinforced concrete. Technology, Brno
The behaviour under tensile loading was investigated by laboratory
tests. A high number of yarn specimens (over 300) of six different VOŘECHOVSKÝ, Miroslav
Institute of Structural
lengths ranging from 1 cm to 74 cm was tested to obtain statistically
Mechanics, Faculty of Civil
significant data which were subsequently corrected and statistically Engineering, University of
processed. The behaviour of the yarn was predicted by the studied Technology, Brno
numerical model of the multi-filament bundle that was applied also
for the later results interpretation. The model of n parallel filaments
describes the behaviour of a bundle with varying parameters repre-
senting different sources of disorder of the response. The aim of the
carried experiment was to validate the model presumptions and to
identify the model parameters to fit the real load-displacement curves.
Due to a distortion of the measured displacements caused by the
unstiff experiment device, only the maximal load values were statisti-
cally evaluated and the identification of the model parameters was
not successful. The effect of the specimen length on its strength was
demonstrated and the size effect curve did not exclude the existence
of spatial correlation of material mechanical properties modifying the
classical statistical Weibull theory.
Introduction
Textile-reinforced concrete is a developing composite material with a

high potential of application in civil engineering structures and also
in other industrial branches. The material combines a cementitious
matrix providing the compressive strength and a tensional reinforce-
ment made by multi-axial fabrics.
The advantages compared to usual steel-reinforced concrete are the

reduction of thickness and subsequently of weight of the structural
elements and a wide shape variability.
The fibers used for the textile reinforcement have to meet several
criteria: high fibre tenancy, breaking elongation and modulus of elas-
ticity much higher than the modulus of the concrete matrix, so that
the stiffness of a building component is not drastically reduced by
occurring cracks. The most common material of fibers is alkali resistant
glass (AR-glass). Filaments are combined to yarns. One yarn composes
of several hundreds up to thousands of single filaments with diameter
measured in tens of micrometers. The fineness of the yarn is defined
by the unit “tex” (gram per 1,000 meters) and depends on the aver-
age filament diameter, the fibre material density and the number of
filaments. Yarns are subsequently combined into textiles. Historically,
the textile-reinforced concrete developed from fiber-reinforced con-
crete with short filaments of random orientation by aligning the fila-
ments in the direction of the tensile stresses similarly to classical steel
reinforcement, which led to better effectiveness of the reinforcement,
increased load-bearing capacity and cost reduction. The material is a
subject of an intensive research in many institutions [1].
Numerical model of a bundle
The yarn structure is a system made by many parallel filaments with

random properties. These properties vary randomly over the length
of the yarn as well as within each cross-section due to imperfections
from the production process. To describe the complex behaviour of
the bundle, definition and study of each individual property and the
influence of its randomization on the overall performance is essential.
Statistical approach is applied as the most convenient way to capture
the yarn behaviour under tension. The presented project studied the
computational model introduced in [2] and [3].
This model consists of a set of parallel fibers with no interaction

among them. Each filament is considered to act independently and
the response of the whole bundle during displacement-controlled ten-
sile loading can be evaluated in an analytical and numerical approach.
Parameters used for the bundle model can be divided into two groups:
those describing the separate filament and those describing the yarn
Figure 1 — Left: Filaments in

the bundle and their elemen-
tary characteristics varying
over the length. Right: Fila-
ment lengths.
(a set of filaments). Parameters appointed to i-th filament can be used

regardless of the composition of the bundle. For AR-glass, where lin-
ear brittle fracture behaviour is considered, these parameters are the
Young’s modulus of elasticity Ei, the cross-section area Ai and the fila-
ment strength i. For cases where the filament parameters are ran-
domized within the cross-section, no variability in these parameters
over the filament length is considered.
Parameters appointed to the bundle describe the variability of fila-

ments within the bundle. Each of the filament parameters is rand-
omized and described by a probability distribution function — GE(Ei),
GA(Ai), G(i). Differences in filament lengths from the nominal length
l of the bundle are captured by two extra parameters: parameter  for
the different distance of fixing points of each filament and parameter 
for the different global activation strain of each filament due to wavi-
ness of filaments in the bundle (slack). The total length of i-th filament
is then li,, — see Figure 1 Right.
Figure 2 — Left: Load-strain

diagrams of one filament.
Right: the whole yarn.
The above mentioned parameters are featured in the load–strain

diagram equation of the i-th filament made of linear–elastic brittle
material: q,i (i) = Ei Ai i H(i) H(i−i) using the Heaviside (unit step)
function: H(x) = 1 for x ≥ 0 and H(x) = 0 elsewhere, which zeros the
filament stress out of the filament’s possible strain interval.
The overall response in form of the load–strain diagram of the bundle

during tensile loading can be obtained numerically simply by sum-
ming up the contributions of all filaments at each level of global strain.
In practical applications the number of the bundle filaments is very
high (several hundreds to thousands). For such a high number of fila-
ments n the bundle mean response M(e) can be solved analytically as
n-multiple of the mean filament response (e): M (e) = n  (e):
qe e ; dG where i Ai , Ei , i , i , i
Experiment
The material selected for the tensile tests was the AR-glass yarn pro-
duced by Saint Gobain Vetrotex with brand name Cem-FIL Direktroving
LTR 5325, 2400 tex. The experiment was focused on the observation
of effect of size (resp. length) on the yarn strength [4]. Consequently, a
wide range of yarn lengths was desired with emphasis on production
of the longest possible specimen length, so that the behavior in this
region can be mapped. Six length groups were suggested with equal
distribution of their logarithms, as the size effect curve is visualized in
the double-logarithmic scale: 1, 2.5, 6, 13, 31 and 74 cm (the longest
possible specimen length with regards to the testing equipment). The
most problematic part of tensile testing was to deal with the anchor-
ing of glass yarns into the machine. Basically, there are two ways how
to create bundle supports: endings can be either directly coiled up
on a cylinder or poured into anchoring blocks and clamped. Anyway,
both techniques show certain deficiencies. As the testing machine is
equipped with self–locking holders, yarn endings were poured into
75 mm long anchoring blocks made of epoxy resin. The total number
of tested samples was 317 pieces.
Tensile tests were performed using the machine Z100 Zwick / Roell

equipped by two load cells measuring the force (20 and 2.5 kN) and
mechanical tensile clamps (jaws) of combined type (self-locking with
pre-stressing screws). The displacement was measured at the top edge
of the upper jaw by deflection extensometer. Samples were loaded by
displacement-increments of the cross-head with the constant rate and
the reaction force was measured by the load–cell. The test speed was
chosen to correspond to 1.1% elongation of the nominal length per
a minute. Force–displacement curves obtained from experiment are
plotted in Fig. 5. Different colours represent different length groups of
samples. “Raw” curves plotted in the figure are obtained directly from
experimental device without any modification.
Before it could be proceeded to the evaluation of experimentally obtained

results, some modifications of the data set had to be made. Firstly,
the number of samples was reduced by elimination of samples either
with strength value extremely differing from the corresponding length
group’s average, or due to serious imperfections caused in the produc-
tion (such as the sample damage, epoxide penetration into the yarn, etc.).
The number of these outliers was 37. These samples were excluded from
further evaluation; they are marked with a cross in the graphs (Figure 6).
Also the displacements measured during the tensile loading embodied

some harmful patterns. While the force reaction induced by displacement
loading can be measured without any errors, the objective measurement
of sample deformation is much more challenging. In the presented

experiment, the deflection extensometer was placed on the upper side of
the top jaw, which caused an inaccuracy of the measured displacements.
By placing the extensometer on the mentioned position the read

deformation did not belong only to the yarn elongation, it was
a sum of deformations of all the device parts under the exten-
someter. Furthermore, the unstiff behaviour of jaws developed
other additional more complicated displacement distortion. After
several unsuccessful attempts to subtract these parasite deformations
by finding the proper calibration curve, the statistical evaluation of the
experiment had to be reduced to interpretation of the maximal force
with respect to the sample’s length.
Figure 3 — Specimens’ length

groups.
Figure 4 — Unmodified
force–displacement curves
(all samples).
Figure 5 — Yarn strengths vs.

yarn lengths of tested sample
groups and the size-effect
curve as an average ± std of
modified (red) and original
(grey) data set.
Conclusion
By knowing the true load–displacement diagrams from the experiment,

the parameters of the computational model and their distribution
could be identified so that the model would fit the reality. Unfortu-
nately, this intention was not fulfilled as the test-curves’ adjustment
was not successful. Consequently, the only objective information
obtained from the experiment were the bundle strengths (maximal
load value). The strength of each yarn was plotted versus its length in
a double-logarithmic scale – Figure 6. The obtained size-effect curve
was intuitively fitted by the equation of modified Weibull size effect
with the spatial strength autocorrelation. Although those two curves
show a good agreement, it should be stated that the sample strength
could be influenced also by other effects, e.g., effects of unequal
length of the filaments, individual activation strain of each filament
in the bundle, friction among filaments of the longer lengths, or with
other unpredictable factors (as the local conditions or the human fac-
tor). Without knowing the correct load–deformation curves, these
statements are only hypothetical. More detailed information about
the experiment can be found in [4].
Acknowledgment
The work has been supported by the Czech Science Foundation

under project no. P 105 / 10 / J 028 and also by the Junior project no.
FAST-J-12-21.
References
[1] RILEM (2006). Textile Reinforced Concrete. State-of-the-Art Report 36,

RILEM Technical Commitee, Aachen, Germany
[2] Chudoba, R.; Vořechovský, M.; Konrad, M. (2006). Stochastic

modeling of multi-filament yarns I: Random properties within the
cross-section and size effect. International Journal of Solids and
Structures (Elsevier), volume 43, nr. 3–4, pages 413–434
[3] Vořechovský, M.; Chudoba, R. (2006). Stochastic modeling of

multi-filament yarns II: Random properties over the length and
size effect. International Journal of Solids and Structures (Elsevier),
volume 43, nr. 3–4, pages 435–458
[4] Kaděrová, J., Vořechovský, M. (2012). Multi-filament yarns testing

for textile-reinforced concrete. Master thesis, Institute of Structural
Mechanics, Faculty of Civil Engineering, Brno University of Technology
Seismic response of short columns affected by brick / tile aggregate concrete wall infills 105
Seismic response of short

columns affected by brick / tile
aggregate concrete wall infills
Abstract KRAUS, Ivan

J. J. Strossmayer University
Various post-earthquake reports indicate that short columns are vul-
of Osijek,Croatia
nerable structural members and significant source of serious earth-
quake damage. Observation that non-structural partial brick infills, MILIČEVIĆ, Ivana
comparatively to partial concrete infills, have less harmful effects on Faculty of Civil Engineering,
short columns put a new perspective on the use of recycled brick / tile J. J. Strossmayer University
of Osijek, Croatia,
as concrete aggregate with the additionally benefit of providing a sus-
tainable management of such material. This paper discusses a numeri-
cal study on the influence of brick / tile aggregate concrete infills on the
response of short concrete columns. In addition, this paper presents
an approach for the analysis and design of short columns, similar to
that of tie beams. Linear elastic analysis was carried out on four- and
seven-story three-bay concrete frames using modal response spectrum
analysis. Mechanical properties of brick and tile aggregate concretes
determined by laboratory testing at Faculty of Civil Engineering Osijek
are described and used in this study.
Introduction
Various post-earthquake reports indicate that short columns are vul-

nerable structural members and significant source of serious earth-
quake damage (see Figure 1). Nevertheless, it is not uncommon to
find short columns even in new buildings, for instance in buildings on
sloping ground; buildings with partially buried basements; or build-
ings with frames having partial infills, [1].
As described in [4], short columns are those that fail in shear having
height-to-depth ratio s ≤ 3. In addition, columns of a depth equal
to 75 cm with a clear height of 230 cm, which is not uncommon for
industrial buildings, can also be designated as short, [4].
106 Seismic response of short columns affected by brick / tile aggregate concrete wall infills
Figure 1 — Examples of short

column failure: a) 5-story build-
ing, built in 1997 according to
the new Greek Seismic Code,
[2]; b) Ø10 mm at 10 cm trans-
verse reinforcement did not
prevent heavy damage, [3]
Post-earthquake field investigations have shown that non-structural,

or better to say “non-intentional structural”, partial brick infills, com-
paratively to partial concrete infills, have less harmful effects on short
columns, [1]. This observation put a new perspective on the use of
recycled brick / tile as concrete aggregate with the additionally benefit
of providing a sustainable management of such material. A numerical
study on the influence of brick / tile aggregate concrete infills on the
response of short concrete columns and an approach for the analysis
and design of short columns, similar to that of tie beams are presented
and discussed in this paper.
Bi-diagonal reinforcing of short columns
An extensive theoretical and experimental research conducted pro-

vided efficient principles for analysis and design of tie beams for walls
with openings. Due to similarities in behaviour of short columns and
tie beams when subjected to earthquake action ([4]) the same princi-
ples were used in this study.
The procedure for calculating the amount of bi-diagonal reinforce-

ment used here is illustrated by Figure 2 and given by the following
expressions, [4]:
VEd
(1) Fs Fc As fyd
2 sin
or
VEd
(2) As
2 sin f yd
where
hc d1 d2
(3) tg
Ls
and
fy
(4) fyd
s
where Fs is tensile force, Fc is compressive force, VEd is design shear

force, α isangle between longitudinal and diagonal reinforcement, As
is cross sectional area of reinforcement of one diagonal steel-bar cage,
fyd is design yield strength of reinforcement, d1 is the distance of the
centre of the compression reinforcement from the extreme compres-
sion fibres, d2 is the distance of the centre of the tension reinforcement
from the extreme tension fibres and Ls is the clear length of column.
Figure 2 — Bi-diagonal rein-

forcing of short column, [4]
(edited by the authors)
The transversal reinforcement of the bi-diagonal steel-bar cages should

be calculated and placed in the same manner as recommended for
ordinary columns, [4].
Horizontal and vertical reinforcement of the short column (see Fig. 2)

has the role to keep the crushed concrete in its place during strong
earthquakes, [4].
Numerical models
2D linear elastic analysis was carried out on four- and seven-story three-
bay facade frames using modal response spectrum analysis. In order to
concentrate on the effects induced from both column slenderness and
the type of material used in infill walls, the frames were taken from
buildings regular in plan and elevation. It is assumed that the buildings
are located in seismically active area on ground type B defined accord-
ing to Eurocode 8 ([5]). Also, it is assumed that the buildings are fixed at
the ground level. The basement is not included in the analytical model.
The numerical analysis was performed using structural analysis pro-

gram SAP2000, [6]. Beams and columns were modelled using frame
elements while infill walls were modelled using four-noded quadri-
lateral shell elements.
Materials
The frame elements and floor and roof slabs were made from normal
conventional C 30 / 37 concrete with characteristic compressive cyl-
inder strength at 28 days fc = 30 N / mm², secant modulus of elastic-
ity Ecm = 330,00 N / mm² and density ρc = 2,400 kg / m³. Only infill
walls were made from different types of concrete described in Table 1.
Mechanical properties of brick and tile aggregate concrete (BTAC)
were determined by laboratory testing conducted by the authors.
Concrete fc (N / mm²) Ecm (N / mm²) ρc (kg / m³)
C 12 / 15 Eurocode 2 12 27,000 2,400

C 16 / 20 Eurocode 2 16 29,000 2,400
Table 1 — Types of concrete BTAC 9 Laboratory test 9 10,500 1,800
used in infill walls BTAC 15 Laboratory test 15 21,500 2,000
Table 2 shows percentage of replacement of fine and coarse dolomite

aggregate with crushed brick and tile in BTAC mixtures.
Concrete Brick Tile Dolomite

mixture
0–4 mm 4–16 0–4 mm 4–16 0–4 mm 4–16

Table 2 — Brick and tile aggre- BTAC 9 50% 50% 25% 25% 25% 25%
gate concrete mixtures BTAC 15 50% 0% 25% 25% 25% 75%
Poisson’s ratio is taken equal to 0,2 for all types of concrete.
Structural elements were reinforced with B500B steel bars with char-
acteristic yield strength fy = 500 N / mm² and modulus of elasticity
Es =  200,000 N / mm². It is assumed that all of the reinforced con-
crete structural elements were reinforced with normal percentage of
reinforcement which adds 100 kg / m³ to concrete density values, as
recommended in [7].
The partial safety factors for materials for ultimate limit state were adopted
according to [8]: c = 1.5 for concrete; and s = 1.15 for reinforcing steel.
Geometry
Geometry of structural elements is provided in Table 3. Floor and roof

solid reinforced concrete slabs were 18 cm thick thus satisfying the
provision for concrete diaphragms, [5]. Wall thickness is equal to one-
half of the column cross sectional height. As noted in [9], the minimal
thickness of the partial infill wall should not be less than 12 cm. The
concrete cover is taken equal to 3 cm.
Element Geometry (cm)
Beam bb / hb 40 / 50

Beam length (CL to CL) 600
Column bb / hb 50 / 50
Column height at the 1st level (CL to CL) 375
Column height at higher levels (CL to CL) 300
Slab thickness 18 Table 3 — Geometry of struc-
Wall thickness 25 tural elements
Height of the first level infill walls is a function of the captive col-
umn height-to-depth ratio which varies from 1 to 3 with step size of
0.5. Infill walls at higher levels have fixed height equal to one-half of
the story height.
Loading on frames
The vertical load includes both dead and live load and was added to
the frames as described in [10]. The structural analysis program ([6])
calculates the self weight of the structural frames and weight of infill
walls. The weight that includes wall partitions and ceilings was taken
into account by uniformly distributed load equal to 3.35 kN / m² as
suggested in [11]. The weight of perpendicular beams were added to
frames as described in [10].
The live load for floors was taken equal to 3 kN / m², which is the nor-
mative value for school buildings ([7]), while the live load for roof is
taken equal to 2 kN / m² as suggested in [12].
Seismic loading is applied by using modal response spectrum method.

Four horizontal design response spectrums of type 1 were made by
combination of two earthquake intensities and two ductility classes
in accordance with Eurocodes (see Figure 3). Behaviour factors were
calculated in accordance with [5] for multi-storey, multi-bay frames or
frame equivalent dual systems. It is assumed that the stories have cor-
related occupancy. Load combinations were made as described in [10]
and in accordance with Eurocodes.
Figure 3 — Design response

spectrums used in analysis
Results and discussion
Results provided here concerns the inner columns of the first story of
the frames defined in this paper.
Figure 4 shows reduction in the design shear force VEd to design shear
resistance of the member without shear reinforcement VRd,c ratio which
goes up to 5% for the four-story frames and up to 15% for the seven-
story frames when the conventional concrete is replaced with BTAC.
Figure 4 — VEd / VRd, c-slen-

derness of column relation-
ships for different types of
concrete
By inspecting Figure 4 and Table 1 it was observed, and as expected,

that the density of the concrete influences the magnitude of the shear
forces the most. Furthermore, it is observed that compressive strength
of concrete and modulus of elasticity of concrete significantly influ-
ences the magnitude of shear force of columns having height-to-depth
ratio equal to 1. Significant influence of compressive strength of con-
crete and modulus of elasticity of concrete on the magnitude of the
shear force is also observed for columns having height-to-depth ratio
equal to 3 and relatively low normalised axial force (average values of
normalised axial forces νd,av are provided in Figure 5).
Figure 5 shows reduction in the amount of required bi-diagonal rein-

forcement up to 3.80% for the four-story frames and up to 4.60% for
the seven-story frames when the conventional concrete is replaced by
BTAC. This reduction is also mostly influenced by the relatively low
density of BTAC, while the influence of compressive strength of con-
crete and modulus of elasticity of concrete are negligible.
Small influence of compressive strength of concrete and modulus of

elasticity of concrete on reduction of required bi-diagonal reinforce-
ment is observed for columns having height-to-depth ratio equal to 1
and relatively high normalised axial force (see Figure 5).
Figure 5 — As-slenderness of
column relationships for
different types of concrete
Conclusion
Numerical study on the behaviour of short columns subjected to seis-

mic action was conducted on four- and seven-story frames having
partial infills made of different types of concrete. Two types of conven-
tional concrete and two types of non-conventional brick / tile aggre-
gate concrete, all having relatively small compressive strength, were
used in this study. The use of brick / tile aggregate concrete in partial
infills reduces the magnitude of shear forces in the short columns thus
reducing the amount of required reinforcement. Among others, due
to relatively small density of brick and tile, when compared to the
density of natural stone, brick / tile aggregate concrete has relatively
lower specific weight. This observation gives additionally benefit on
the use of brick / tile aggregate concrete in buildings located in seismi-
cally active zones thus providing a sustainable management of such
material. Further research on brick / tile aggregate concrete should
be conducted to inspect its mechanical and physical properties and
to prove its usage as building material. Moreover, an experimental
research should be performed to verify the results of this study.
References
[1] Guevara, L. T., Garcia, L. E. (2005). The captive- and short-column

effects. Earthquake Spectra 21 (1), pp. 141–160, [online] Available at:
http://15wceess.blogspot.com / 2012 / 05 /captive-and-short-column-
effects.html [Accessed 11 May 2012]
[2] Multidisciplinary Center for Earthquake Engineering Research

(MCEER) at University of Buffalo, 1999. Failure of column, due to
short column effect, of a 5-storey building in Ano Liosia. [image online]
Available at: http://mceer.buffalo.edu/research/Reconnaissance/
greece9-7-99/fig_5.asp [Accessed 26 July 2012]
[3] Fierro, E., 2001. Short column at another school in Camana, [image
online] Available at: http://www.eeri.org/2001/06/southern-peru/
[Accessed 26 July 2012]
[4] Tomičić, I. (2009). Analysis and design of short reinforced-concrete

columns. Građevinar 62 (8), 717–722, [online] Available at: http://
www.casopis-gradjevinar.hr/arhiva/article/279 [Accessed 11 May 2012]
[5] CEN — European Committee for Standardization (2004). Eurocode 8

— Design of structures for earthquake resistance — Part 1: General
rules, seismic actions and rules for buildings
[6] SAP2000, version 14.1.0. Structural analysis program. Pacific Earth-

quake Engineering Research Center, University of California,
Berkeley, California
[7] CEN — European Committee for Standardization (2002). Eurocode 1 –

Actions on structures — Part 1–1: General actions — Densities, self-
weight, imposed loads for buildings
[8] CEN — European Committee for Standardization (2004). Eurocode 2

— Design of concrete structures — Part 1–1: General rules and rules
for buildings
[9] Tomičić, I. (1996). Betonske konstrukcije (3rd edition), Društvo

hrvatskih građevinskih konstruktora, Zagreb
[10] Kraus I., Morić, D. Netinger, I. (2011). Comparison in seismic

response of reinforced concrete buildings made of non conventional
concrete. The Electronic Journal e-GFOS, Number 3 — December,
[online] Available at: http: //e-gfos.gfos.hr / en / index.php / archive /
number-3 [Accessed 23 May 2012]
[11] Konstantinidis, A. (2008). Earthquake resistant buildings from

reinforced concrete, Volume A — The art of construction and the
detailing. Athens, [online] Available at: http: //www.pi.gr / CMSnew /
index.php [Accessed 23 May 2012]
[12] Sigmund, V., et al (2000). Comparison in the application of Croatian

regulations and Eurocode 8. Građevinar 52 (7), 379–388, [online]
Available at: http: //www.casopis-gradjevinar.hr /arhiva / article / 77
[Accessed 17 July 2012]
114 Experimental research on thermo-creep behavior of soft sedimentary rock
Experimental research on
thermo-creep behavior of
soft sedimentary rock
KURIMOTO, Yuhei Abstract

Department of Civil Engi-
neering, Nagoya Institute
In Japan, nuclear electricity generation has occupied about 30% of
of Technology, Japan
total power generation. On the other hand, we are faced with a seri-
ZHANG, F ous problem of how to manage the nuclear waste disposal, especially
Department of Civil Engi- the disposal of high-level radioactive waste (HRW). Until now, deep
neering, Nagoya Institute geological disposal is regarded as one of the most viable and the saf-
est ways of permanent disposal of HRW. The soft sedimentary rock is
XIONG, Y one of the host rocks that are suitable for the geological disposal of
Department of Civil Engi- the nuclear waste. The half-decay period of many nuclear elements,
neering, Nagoya Institute however, is very long and the amount of the thermal energy gener-
ated during the half-decay period is huge enough to make troubles. It
NISHIMURA, Tomohiro is known that temperature and its change may affect the mechanical
Department of Civil Engi- behaviors of the rock, especially the long-term stability. Therefore, it
neering, Nagoya Institute is necessary to study carefully the interacting factors related to the
of Technology, Japan mechanical properties of host rock, especially the thermo-mechanical
properties. Author conducted the thermal triaxial compression tests
and thermal triaxial creep tests under different constant tempera-
tures. From these tests, it is known that: (1) peak strength of soft rock
decreases as the temperature increases; (2) creep failure occurs more
quickly as temperature increases.
Introduction
It is commonly known that deep geological disposal is regarded as

one of the most viable and the safest ways of permanent disposal
of HRW [1]. On the other hand, there are so many problems to be
resolved. The half-decaying span of many nuclear elements is very
long and the amount of the thermal energy generated during the
radioactive decaying period is so huge. Therefore, it is natural that
Experimental research on thermo-creep behavior of soft sedimentary rock 115
shielding material like rock mass for deep geological disposal should
be stable in a long time under high temperature conditions. So, it
is necessary to study carefully the interacting factors related to the
mechanical properties of host rock, especially the thermo-mechanical
properties. In this paper, in order to grasp the mechanical properties
of host rock under high temperature conditions, author conducted
the thermal triaxial compression tests and thermal triaxial creep tests
under 4 different constant temperatures.
Test device and specimen
Test device
Figure 1 shows the temperature-controlled triaxial compression and
creep test device. It consists of a load controlling unit for cell pressure
and axial load, and a temperature controlling unit, as shown in Figure 2.
Performances of test device are listed in Table 1.
Figure 1 — Temperature-con-
trolled triaxial compression
and creep test device.
Figure 2 — Temperature con-

trolling unit.
Control method PC
Maximum temperature [°C] 90

Maximum confining pressure [MPa] 10
Maximum axial pressure [MPa] 25 Table 1 — Performances of
Maximum back pressure [MPa] 1 test device.
Heating system
The heating system includes a heater, a mixer and a temperature sen-
sor that is installed within the pressure cell chamber and connected to
the temperature controlling unit. To reduce the difference of tempera-
ture of the water inside the triaxial cell chamber, an inclined propeller
(45°) was installed within the cell to keep the temperature of water
distributed uniformly and quickly, as shown in Figure 3. Besides, the
rotating speed of the inclined propeller is adjustable. During the test,
for safety, the highest temperature within the chamber was set at 90°C.
Figure 3 — Schematic illustra-

tion of pressure cell chamber.
Specimen
It is known that the homogeneity of the test samples is a very impor-
tant factor to acquire reliable test results, especially for studying
thermo-mechanical properties of soft sedimentary rocks. Ohya stone,
a typical green tuff, is distributed widely in the northeastern part of
Japan and used as the test specimen in this experiment. The stone is a
typical soft volcanic tuff formed in Miocene age, and is widely used in
laboratory tests for their mechanical properties in Japan. Ohya stone
usually has a uniaxial compressive strength of around 10 MPa, and
some of its basic physical properties are given in Table 2.
The cylindrical test specimen of Ohya stone is 100 mm in height and

50 mm in diameter, as shown in Figure 4. To obtain a satisfactory
degree of saturation (more than 95%) for the specimens in the test,
saturation using vacuum pump was conducted.
Soil particle density s [g / cm³] 2.56

Moisture content  [%] 19.54
Wet density t [g / cm³] 2.05
Table 2 — Some physical Dry density d [g / cm³] 1.71
properties of Ohya stone. Initial void ratio e0 0.49
Figure 4 — Soft sedimentary

rock specimen.
Test procedure and results
Thermal triaxial compression test

Based on the test specification [2] of the Japanese Geotechnical Soci-
ety, the thermal triaxial compression tests were conducted under the
room temperature (20°C), 40°C and 60°C, 80°C respectively. The pro-
cedures include three stages:
(1) Heating stage. The temperature was increased at a rate of

0.5 °C / min to the target temperature under drained condition and
then was kept constant.
(2) Consolidation stage. Isotropic consolidation was conducted for

24 hours under the prescribed confining pressure and the target
temperature.
(3) Compression stage. Shearing was conducted under drained condi-

tion and the constant target temperature.
The detailed thermo-mechanical conditions are listed in Table 3.
Temperature [°C] 20, 40, 60, 80

Effective confining pressure [MPa] 0.49
Strain rate [% / min] 0.002 Table 3 — Conditions of com-
Drainage condition Drainage pression tests.
Figure 5 shows the stress difference-axial strain relations of the tested

soft rock subjected to compression under different constant tempera-
tures. From Figure 5, it is known that Ohya stone shows strain sof-
tening and peak strength of soft rock decreases as the temperature
increases. Figure 6 shows the relations between the volumetric strain
and the axial strain. It can be seen that the volumetric strain in all tests
first contracts and then expands (dilatancy characteristics).
Moreover, in spite of residual stress, volumetric strain is not constant.

This cause is that failure surface is very rough (roughness is indexes of
discontinuity plane), as shown in Figure 7.
Figure 5 — Stress difference

vs. Axial strain.
Figure 6 — Volumetric strain

vs. Axial strain.
Figure 7 — Specimen after

thermal triaxial compres-
sion tests.
Thermal triaxial creep test

The thermal triaxial creep tests were conducted under the room tem-
perature (20°C), 40°C and 60°C, 80°C respectively with constant
creep load. In the creep tests, instead of abrupt loading, the axial
creep load was applied at a constant rate until the specified creep
stress was reached. The specified creep stress, which is defined as the
percentage of the peak strength obtained from the drained compres-
sion tests of 80°C, is listed in Table 4.
Temperature [°C] 20, 40, 60, 80

Creep stress [MPa] 7.37 (95%)
Effective confining pressure [MPa] 0.49
Table 4 — Conditions of Rate of loading [kN / min] 0.05
creep tests. Drainage condition Drainage
The axial strain was plotted against time on a logarithmic scale, as

shown in Figure 8. From Figure 8, the creep failure time is largely
dependent on the temperature. Figure 9 shows the relations between
the axial strain rate and time on a logarithmic scale. The general char-
acteristics of creep behavior, such as the transient creep, the steady
creep and the accelerating creep can be clearly observed. Moreover,
the higher the temperature is, the faster the creep rupture will occur.
Figure 8 — Axial strain vs.

Time.
Figure 9 — Axial strain rate

vs. Time.
Conclusions
In this paper, the thermal triaxial compression tests and thermal triax-
ial creep tests were conducted under different constant temperatures.
From these tests, the following conclusions can be drawn:
(1) The temperature does influence the peak strength of the soft rock.
In the triaxial compression tests, peak strength of soft rock
decreases as the temperature increases and the volumetric strain
in all tests first contracts and then expands.
(2) In the triaxial creep tests, the creep failure time is largely depend-
ent on the temperature. In other words, the creep failure occurs
more quickly as temperature increases.
References
[1] Okada T. Mechanical properties of sedimentary soft rock at high

temperatures (part 1) — evaluation of temperature dependency based on
triaxial compression test. Chiba, Japan: Central Research Institute of
Electric Power Industry, 2005 (in Japanese).
[2] Japanese Geotechnical Society. (2000). Field study and testing me

thod for rock. [S.l.]: JGS-Soft, in Japan.
120 Bridge-Wizard: Expert system for FE modeling and post — processing of Bridges
Bridge-Wizard: Expert system

for FE modeling and post —
processing of Bridges
LESGIDIS, Nikolaos D. Abstract

saloniki, Department of Civil
Experience is essential at the process of structural modeling, espe-
Engineering, Greece
cially in case of large scale structures. Modern Finite Element software
provides advanced capabilities related to complex constitutive laws,
geometry and loading patterns; however, many pre- and post-process-
ing tasks still require a high level of engineering judgment while others
are significantly time consuming. To this end, a bridge analysis expert
system called Bridge-Wizard has been developed. Bridge-Wizard is an
interactive front-end software that applies a step by step logic in the
process of modeling and post-processing and consults the user in two
different ways, i.e., by providing (a) conceptual assistance in develop-
ing the FE model in an open source program environment (OpenSees)
through the user’s description of the structure and (b) expert advice
with respect to the simulation of the components expected to exhibit
nonlinear response (particularly, deck, piers, bearings and abutments).
The software also provides interactive modeling of the boundary con-
ditions considering the effect of soil-structure interaction at both the
abutments and the middle piers. Bridge-Wizard also offers in depth
visualization of the results of the non linear response history analysis,
through a set of graphs, thus providing a spherical overview of the
analysis results. It is deemed that the development of Bridge Wizard
improves the efficiency and robustness of the finite element of com-
plex structures with the use of OpenSees.
Introduction
Similarly to every other scientific area, the technological breakthrough

of the last two decades has greatly affected structural engineering
as well. The use of advanced finite element software is increasing in
popularity day by day making it a vital part in the structural engineer’s
every day work. Nonetheless there is still room for improvement by
Bridge-Wizard: Expert system for FE modeling and post — processing of Bridges 121
eliminating difficulties the user may encounter. Such difficulties are

the large amount of time consumed in the construction of an analysis
or the error possibility due to the user’s misjudgement. The develop-
ment of an expert system named Bridge wizard is an improvement
attempt in the field of bridge computer analysis. Bridge wizard copes
with the difficulties mentioned above by counseling the user in his
modeling decisions and offering him automated solutions.
Bridge wizard is capable of performing time history (linear or non linear),

modal and static analysis of a bridge structure that belongs in the
category of beam bridges. Since Bridge Wizard does not include an
analysis algorithm the open source software OpenSees was selected to
complete that task. Thus a cooperation takes place were bridge wizard
is exporting the structure’s data to Opensees and importing the analysis
results from it. In the following pages the structure, the applied methods
and the general image of Bridge wizard will be discussed and in the con-
clusion the results and the objectives accomplished will be commented.
Structure of Bridge Wizard
Bridge Wizard is a windows form application created in Microsoft

Visual Studio 2008. The main program is written in Visual Basic but
knowledge of tcl was needed for the exchange of data between Open-
Sees and Bridge Wizard.
Since Bridge Wizard is a front to end wizard software, there is a spe-

cific path the user can follow as depicted in the main algorithm figure
(Figure 1). In order for the user to start and complete the modeling of
a structure, a seven step process must be followed. In each step the
software obtains information from the user in order to complete a suf-
ficient description of the structure. This information is carefully stored
into a large number of Matrixes and waited to be used in the final
step of the process. In the first step the user is asked to supply general
information on the geometry of the Bridge. During the four following
steps the software asks for more specific information corresponding on
the four parts the structure is divided (Deck, Piers, Abutments, Pier-
Deck connection). At the 2 final steps the users supplies the software
with the details of the static, modal and time history analysis.
By the end of the 7th step a tcl data file is created and all the data
obtained in these 7 steps, combined with other preset information,
are organized and printed in that file according to an Opensees input
file template. The Opensees software initializes and runs an analysis
according to the directions of the tcl data file. The tcl data file consists
of a repeated three part process. The first part directs the creation of
the model of the structure while the second part declares the selected
methods of analysis. Finally the third part organizes the location and
format of the post processing output data. The process repeats itself
Figure 1 — Main Algorithm of

Bridge Wizard
in the data file for each loading combination respectively. By the end
of the analysis OpenSees terminates automatically and all the data
results of the analysis are stored and organized by Bridge Wizard.
By the end of the seven step procedure the user has the opportunity to
check the results of the analysis either in a graphical manner through
the software Bridge Wizard or by the output files produced from
OpenSees. For the graphical depiction of the results Bridge wizard has
in its possession a small number of tools. These tools are a compo-
nent named Ms Chart specialized in the creation of visual charts and
a 2D graphics algorithm created for Bridge Wizard. The component is
used in the visual display of stress or displacement to time charts of a
specific node of the structure where the graphics algorithm is used in
the depiction of the stress or displacement of the overall structure for
a specific moment of time. The user has the ability to leap from one
method of post processing to the other. Finally, if the user wishes to
analyze a different structure or change details on the current structure,
there is the option of repeating the whole procedure from the first step.
Modeling and Analyzing Methods
As expected from an expert system, Bridge Wizard has the ability to

input information from the user and based on that information lean on
the correct decisions regarding the modeling and analysis of a bridge
structure. For these decisions Bridge wizard follows a number of meth-
ods stored in its memory. There are modeling methods selected for every
structural part of the bridge (Deck, Piers, Abutments, Bearings, Gaps)
and analyzing methods for every analysis available in Bridge Wizard.
Deck: The finite elements used for the simulation of the deck are stick
elements. The use of shell or solid elements is avoided due to 2 main
reasons. Firstly with a different selection from the stick model, the
number of finite elements are greatly increased resulting to a more
complicated differential equation system for OpenSees to solve, while
the accuracy of the results remain the same. Secondly, due to the com-
plexity and lack of regularity of the section’s geometry, modeling the
deck with shell or solid elements could be considered as a significant
geometrical challenge. Bridge wizard’s main objective is to be able
to solve the majority of the structural cases addressed by the user, an
objective reached only by the use of stick model’s simple geometry.
Regarding the stress — strain curve simulated for the deck material, Bridge
wizard considers the curve linear. The reason of the selection of an elastic
material for the simulation of the deck is the capacity design of the deck
itself in an earthquake load combination. Since the deck is not considered
as a dissipative mechanism of the structure, the design is calculated in way
of not exciting the yielding stress of the deck and thus making unnecessary
the use of a nonlinear simulation of the stress-strain curve. Considering the
meshing technique used, every span of the bridge is divided in 10 finite
elements. The meshing technique used may cause small inaccuracies if
the bridge analyzed consists of highly unequal span lengths.
Piers: For the simulation of the piers, the finite elements used are
also stick elements. The reasons of the stick element selection are the
ones mentioned on the deck section with the addition of the neces-
sary compatibility between deck and piers. Foundations of the piers
are modeled with either a hinged or fixed connection depending on
the user’s preferences. The height of the deck’s section is simulated
through rigid finite elements extending from the piers. Considering
the meshing technique used, every Pier of the bridge is divided in
10 finite elements excluding any rigid finite elements.
In contrast to the deck, piers are frequently used as dissipative mecha-

nisms of a bridge due to their ductile formation of plastic hinges in
their end, making the non linear simulation of the piers an ideal solu-
tion. However, due to simplicity the user’s only option is the linear
stress — strain curve simulation for the piers material.
Bearings: Bridge wizard bearings simulation can be either linear or

non linear (bilinear). The option of non linearity applies only to the
horizontal displacements (Ux, Uy) of the bearing, leaving every other
(Uz, Rx, Ry, Rz) with a default option of elastic simulation. For the
modeling process of the bearings, OpenSees offers a finite element
object called “elastomeric bearing element”, an object capable to sup-
ply all the perquisites named by Bridge Wizard.
Bridge wizard offers the option to add any number of bearings in one
pier with the only request of being identical. However, adding dif-
ferent bearings from pier to pier is a possible action. In contrast to
the common way of modeling a bearing connection, Bridge wizard
encounters the problem with a more simplified method. Instead of
using one object for each bearing connected to the pier, only one
individual object is used with characteristics defined in a way to create
a modeling equal. In order to calculate these characteristics, Bridge
wizard follows the equations below.
ind i ind i
(2.1, 2.2) K Ux , el K Ux , el , K Ux , pl K Ux , pl
n n
ind i ind i
(2.3, 2.4) K Uy , el K Uy , el , K Uy , pl K Uy , pl
n n
ind i
(2.5) K Uz , el K Uz , el
n
ind i
i 2 K Uz , el l 2
i
(2.6) K Rx , el K Rx , el
n n
2
ind i
(2.7) K Ry , el K Ry , el
n
ind i
i 2 K Ux , el l 2
i
(2.8) K Rz , el K Rz , el
n n
2
Like every other simplified method the bearing modeling method has
its disadvantages. As long as the bearings are linear there is no differ-
ence in the results from the analytical method. However, in the case of
multiple non linear bearings at one pier, the overall rotational stiffness
KRzind becomes also non linear due to geometrical affection of the non
linear stiffness KUx (the geometrical affection is obvious in the equa-
tion 2.8). This affection may result in a miscalculation of the overall
rotational stiffness KRzind making the model unrealistic.
Abutments: Bridge Wizard offers the user two possible options for the
modeling of the abutments. On the first method the node correspond-
ing to the location of the abutment can have its degrees of freedom
fixed according to the user’s selections.
The second method refers to abutments using bearings to connect

with the deck and optionally gaps in the horizontal directions. The
modeling of abutment bearings follows the exact same method
explained in the previous “bearings” subchapter. For the simulation of
a possible gap in the horizontal direction Bridge Wizard uses a uniaxial
Figure 2.1 — Stress and strain

curve of OpenSees gap object
material object from OpenSees with the name “Elastic perfectly Plastic
Gap Material”. A clearer picture can be obtained from the stress-strain
curve of the gap element (Figure 2.1). The material is used through a
zero-length element.
Time history analysis: The time history analysis can be either linear or
non linear depending on the user’s preferences. The method used for
the solution of the differential equation system in both linear and non
linear case is the numerical integration method “Newmark-Beta”. For
the solution algorithm object which determines the sequence of steps
taken to solve the non-linear equation, Bridge wizard has a default
selection of the Newton algorithm.
Presentation of Bridge Wizard
In order to achieve a decent presentation of a modern software, it

is essential for the reader to take a tour in the software mechanics
from the perspective of the user. The presentation can be divided in 2
subchapters. The topic of the first subchapter is the seven step process
the user follows from the initiation of the software to the last point of
modeling and analysis of the structure. In the second subchapter Bridge
Wizard’s main tools of post-processing are introduced to the user.
Initiation of the software: The user’s first contact attempt with the soft-
ware is through its main window as shown in figure 4.1. By the startup
of the main window almost every possible action is locked leaving the
user with the only options of creating a new project, loading an existing
project or starting the help guide. By clicking the “New project” button
the user initiates the modeling and analysis process of a bridge structure.
Figure 4.1 — Main Window

after initiation
First step — General geometry: On the first step of the modeling

process, the software collects preliminary information describing the
number of spans and piers of the structure as depicted in figure 4.2.
Second step — modeling of the Deck: The software navigates the

user to the deck configuration window as depicted in figure 4.3.a giv-
ing him the option to select a span and edit its properties. The user
imports material, geometry and section properties of the selected
span through a section window of his choice (an example of a section
window is depicted in figure 4.3.b).
Third step — modeling of the Piers: The software navigates the user
to the Piers configuration window as depicted in figure 4.4.(a) giving
him the option to select a Pier and edit its properties. The user imports
material, geometry and section properties of the selected Pier through
a section window of his choice (an example of a section window is
depicted in figure 4.4.(b)).
Figure 4.2 — Bridge geometry

Window
Figure 4.3 — (a) Deck con-

figuration window (b) General
section Window (deck)
Figure 4.4 — (a) Pier configu-

ration window (b) Rectangular
Hollow section Window (Pier)
Forth step — modeling of the Abutments: The software navigates the

user to the abutment configuration window as depicted in figure 4.5.
If the simplified option is selected, the user can assign the restrains on
the abutment node DOFs from the abutment configuration window. If
the “abutment with bearings” option is selected the user imports the
bearing connection properties through a bearing configuration window
of his choice (example given in fig. 4.6.(a). Then the user may proceed
to the optional gap configuration window as depicted in figure 4.6.(b),
were gaps at the horizontal directions of the abutment can be assigned.
Figure 4.5 — Abutment con-

figuration window
Figure 4.6 — a) Bilinear bear-

ing configuration window
b) Gap configuration Window
Fifth step — modeling of the pier-deck connections: The software

navigates the user to the “pier-deck connection window” were the
user proceeds to import the bearing properties to every non mono-
lithic pier-deck connection. For the bearing connection the process
followed is identical to the one described in the “abutments modeling”
subchapter with the exception of the gap configuration step.
Sixth step — Time history analysis preferences: The step is initiated

with the display of the “time history analysis” window depicted in
figure 4.7.a, were the user can define the directions affected by the
earthquake and import the analysis properties. For every affected
direction the user must import its ground acceleration record through
the “ground acceleration window” depicted in figure 4.7.b. The record
must be in NGA format. With the record imported the software pre-
sents the graphs of the acceleration — time curve and its response
spectrum to the user.
Seventh step — Static / Modal analysis preferences: on the final step

of the procedure the software navigates the user to the “modal / Static
analysis window” where he can provide to the software the modal and
static analysis details. Opensees initiates and completes the analysis
and the user returns to the main window with the post processing
options unlocked.
Figure 4.7 — a) Time History

Analysis Window b) Ground
Acceleration Window
Figure 4.8 — a) modal / Static

analysis window b) Opensees
starting Window
Post-Processing analysis results: Bridge wizard can offer to the user

two different post processing tools in order to evaluate the analysis
results. The first tool consists of two plane screens located in the main
window representing a side view and an overview of the bridge as
depicted in figure 4.9. With that tool the user has the ability to view
the deformation, the stresses and the modes of the overall structure
in both planes for a given moment of time for any load combination.
Figure 4.9 — a) Main window

— Moment y from loading
combination G + 0.3Q + E
(t = 4.0sec)
The second tool is located in the graph display window as depicted in

figure 4.10. The user can process the deformation or stress results of a
specific node of the structure through time. The selected node can be
located as a red dot through a screen depicting the structures sketch.
Conclusions
The basic scope of the development of bridge wizard was the creation
of an expert system capable of eliminating the error possibility due
to user’s misjudgement and simultaneously greatly decrease the time
needed to complete the modeling process of the structure. In order
to verify a possible success in the first objective a number of com-
parison tests have taken place. The tests compared various analysis
results (stress, deformation, modes) of a bridge analyzed once manu-
ally with SAP2000 and afterwards with Bridge Wizard. The results
Figure 4.10 — Graph display

window
of the comparison were quite pleasing as the deviation between the

results didn’t exceeded the 5% while the average deviation was esti-
mated around 2%. In reference to the second objective an important
increase in the speed of modeling process is expected since a great
number of user’s decisions are now made by the software itself, how-
ever no concrete proof can be provided what so ever.
References
[1] Chopra, A. K. (2001) “Dynamics of structures: Theory and applications

to earthquake engineering 2nd Edition” Prentice Hall, Englewood
Cliffs, N.J.
[2] Sextos, A. and Katsanos, E. (2009) ‘Programming Techniques and

use of specialized software for structural engineering’, Aivazis Pub-
lications, Thessaloniki
[3] Fardis, M., Kolias V., Panagiotakos T., Katsaras C., Psychogios T.,
(2012). ‘Guide for bridge design with emphasis on seismic aspects’,
university of Patras, Patra
[4] Opensees Ver 2.3.2 ,User’s documentation, Berkeley, California, U.S.
[5] SAP 2000 Ver. 14.0, User’s Reference Manual, Berkeley, California, U.S.
132 Experimental and numerical assessment of the dynamic behaviour of a railway viaduct with precast deck
Experimental and numerical

assessment of the dynamic
behaviour of a railway viaduct
with precast deck
MALVEIRO, Joel Abstract

Department of Civil Engineer-
ing, Faculty of Engineering,
This paper describes the experimental and numerical work developed
University of Porto, Portugal
in order to evaluate the dynamic behaviour of a railway viaduct with
RIBEIRO, Diogo precast deck. The experimental work consists on an ambient vibration
Department of Civil Engineer- test and a dynamic test under railway traffic. The ambient vibration
ing , Polytechnic Institute of test allows the identification of the modal parameters (natural fre-
Porto, Portugal
quencies, modal configurations and damping coefficients) related to
CALÇADA, Rui global and local modes of vibration. These modal parameters were
Department of Civil Engineer- used to calibrate the numerical model using an iterative methodology
ing, Faculty of Engineering, based on a genetic algorithm. The dynamic test under railway traf-
fic allows obtaining the dynamic response in terms of displacements,
accelerations and deformations at different deck locations for the pas-
sage of Alfa Pendular train. The calibrated model was validated based
on a comparison between numerical and experimental responses and
a very good correlation was obtained.
Introduction
Railway bridges and viaducts are structures subjected to high intensity

moving loads, where the dynamic effects can reach significant values.
These effects have assumed greater importance due to the increase
of the circulation speed, especially in high speed lines. In these lines
the dynamic effects tend to increase considerably as a result of the
resonance phenomena that occur due to the passage of trains com-
posed by regularly spaced groups of axles. The dynamic analyses are
usually based on finite element numerical models of the structure
that involves assumptions and simplifications that may cause errors.
These errors are basically related to the inaccuracy in the FE model
discretisation, uncertainties in geometry and boundary conditions and
variation in the material properties. Therefore, the accuracy of the FE
model strongly depends on the experimental validation of the numeri-
Experimental and numerical assessment of the dynamic behaviour of a railway viaduct with precast deck 133
cal results that is usually performed by means of dynamic tests. Thus,

this paper describes the experimental and numerical work developed
in order to evaluate the dynamic behaviour of a railway viaduct with
precast deck. The experimental work consists on an ambient vibration
test and a dynamic test under railway traffic. The numerical model
was calibrated using an iterative methodology based on a genetic
algorithm. The calibrated model was validated based on a comparison
between numerical and experimental responses and a very good cor-
relation was obtained.
Alverca railway viaduct
Alverca railway viaduct is located in the Northern line of the Portu-

guese railway network that establishes the connection between Lis-
bon and Porto (Figure 1). The viaduct supports only one railway track
and consists in 47 successive simply supported spans, with lengths of
16.5 m, 17.5 m and 21 m.
Figure 1 — Alverca railway

viaduct: (a) side view and
(a) (b) (b) bottom view
Each span is composed by a prefabricated “U” beam connected by an

upper slab cast in situ, forming a single-cell box-girder deck. The deck
is directly supported in the piers by elastomeric reinforced bearings.
These bearings are fixed at one of the extremities of each span and
guided in the transverse direction at the other extremity. The track is
continuous between successive spans and is composed by 30 cm of
ballast, monoblock sleepers and UIC60 rails.
Figure 2 — Numerical model

of the Alverca railway viaduct:
overview and a detail
Numerical model
For the dynamic analysis of the viaduct, a three-dimensional numeri-

cal model was developed with the software ANSYS [1]. Figure 2 shows
an overview of the model and also a detailed cross section [2]. In this
model three simply supported spans were considered, adjacent to the
north abutment, one of them with 16.5 m (span 1) and two spans with
21 m (spans 2 and 3). It was also modelled 6 m of track beyond the abut-
ment to simulate the effect of the track on the adjacent embankment.
The density adopted for concrete and ballast was respectively 25 and
20 kN / m³. The modulus of elasticity of concrete corresponds to the
average correction to 28 days for the average value at the time of
experimental test. As the bridge was built to 13 years ago, the values
considered for the modulus of elasticity of concrete C 30 / 37 (upper
slab) and C 45 / 55 (prefabricated beam) were respectively 35.4 GPa
and 40.9 GPa. The structural bearings were considered flexible in the
vertical direction and with stiffness equal to 5,200 x 106 N / m.
Ambient vibration test
The ambient vibration test allowed the identification of the modal

parameters of the structure, namely its natural frequencies, mode
shapes and damping coefficients. This test was performed in two dif-
ferent phases: the first was focused on the identification of the global
modal parameters and the second allowed the characterization of the
local dynamic behaviour of the upper slab of the box-girder section.
Both tests were performed using a technique with fixed reference

points and using 12 piezoelectric accelerometers (PCB model 393A03).
The dynamic response was evaluated in several setups, in terms of accel-
erations in the vertical and longitudinal directions. The acceleration
series were measured during 4 minutes with a sampling frequency of
2,000 Hz and decimated to frequencies of 100 Hz for global modes and
200 Hz for local modes of vibration. The data acquisition system was
the NI cDAQ-9172®, including IEPE analog input modules (NI 9233®).
An external excitation was applied to the structure by a group of peo-

ple jumping randomly in time and space. This excitation technique
increased the signal-to-noise ratio and consequently increased the
coherence of the measured signals. Figure 3 shows an overview of the
instrumented spans as well as details of the accelerometers, the exci-
tation technique and the acquisition system.
The identification of the modal parameters was performed using the

stochastic subspace identification method (SSI-DATA), variant UPC,
available in the commercial software ARTeMIS® [3].
Figure 3 — Ambient vibration

test: (a) instrumented spans;
(b) external excitation;
(c) acquisition system
(a) (b) (c)
Phase 1
In this phase of the ambient vibration test 57 measurement points
distributed along the three spans in analysis, with particular emphasis
in the intermediate span of 21 m (Figure 4a), were considered. The
accelerometers were placed on the ballast retaining walls, on the
vertical direction, through metal plates bonded to the concrete sur-
face (Figure 4b). Attending that these walls are monolithically con-
nected to the upper slab and deviate from the symmetry axis of the
deck, the measured responses at these locations allowed the iden-
tification of the modes of vibration associated to the bending and
torsion movements of the deck.
Additionally, there were also some accelerometers positioned near the

structural joints between decks of 21 m span, at two distinct levels: in
the webs of the prefabricated beam nearby the bearings (Figure 4c)
and at the same vertical alignment but at the level of the ballast retain-
ing walls (Figure 4d). The data obtained from these accelerometers
allowed the characterization of the relative rotation between these
two successive spans in order to characterize the structural continuity
provided by the track, namely by the rails and the ballast layer.
(a)

test — phase 1: a) location of
measurement points; b) accel-
erometer in the ballast retaining
wall; c) and d) accelerometers
nearby the joint between the
(b) (c) (d) successive spans
Mode Type of mode Frequency Damping

(Hz) coefficient (%)
1G Vertical bending 6.76 1.63

2G Vertical bending 6.95 3.56
Table 1 — Natural frequencies 3G Vertical bending 9.65 2.39
and modal damping coeffi- 4G Torsion 13.04 4.99
cients of global modes 5G Vertical bending 20.94 3.46
Table 1 shows the values of natural frequencies and damping coef-

ficients for the global vibration modes identified by the application of
SSI-DATA method. Figure 5 shows the mode shapes in correspond-
ence with the identified natural frequencies.
Mode 3G: f = 9.65 Hz Mode 4G: f = 13.04 Hz
Figure 5 — Global experimen-

tal modes of vibration
Mode 5G: f = 20.94 Hz
Phase 2
In this phase of the ambient vibration test 84 measurement points

located in the intermediate span of 21 m, with particular incidence in
half of that span, were considered (Fig. 6a). The accelerometers were
placed on the sleepers, on the ballast retaining walls, on the precast
girders located in the extremity of the footway cantilever and on lower
slab of the deck, through metal plates bonded to the concrete surface.
The accelerometers placed on the sleepers allowed the identification
of the local modes of vibration of the upper slab, once it was impos-
sible to access directly to the slab either internally, trough the box-
girder deck or externally, by removing the ballast layer. Figure 6b and
c shows the installation process of accelerometers on the sleepers.

test – phase 2: (a) location
of measurement points; (b)
installation process of the ac-
celerometers on the sleepers;
(d) accelerometer on the
sleep-er edge
(a)
belongs to Figure 6
(b) (c)
Table 2 shows the values of natural frequencies and damping coeffi-

cients for the local modes of vibration identified by the application of
SSI-DATA method.
Mode Frequency Damping

(Hz) coefficient (%)
1L 25.48 2.16
2L 27.41 2.13
3L 53.18 4.47
Table 2 — Natural frequencies
4L 60.18 3.35 and modal damping coeffi-
5L 74.60 4.81 cients of local modes
As shown in the table above, the values of the damping coefficients are
higher than those obtained by applying the EN1991-2 (2003) standard
for this type of structures and spans. Figure 7 shows the mode shapes
in correspondence with the identified natural frequencies.
Dynamic test under railway traffic
The dynamic test under railway traffic allowed obtaining records of

displacements, accelerations and deformations on the middle span
section of the second span (Figure 8).
Figure 9a shows an acceleration record obtained for the passage of the

Alfa Pendular train at 185 km / h. Figure 9b shows the frequency con-
tent related to the temporal response. The dynamic response is domi-
nated by frequencies associated to the train action, especially to the
passage of regularly spaced groups of axles with a spacing of 25.9 m.
Mode 1L: f = 25.48 Hz Mode 2L: f = 27.41 Hz Mode 3L: f = 53.18 Hz
Figure 7 — Local experimental

modes of vibration
Mode 4L: f = 60.18 Hz Mode 5L: f = 74.60 Hz
Calibration of the numerical model
The calibration of the numerical model of the viaduct was performed

taking account the modal parameters obtained in the ambient vibra-
tion tests. This process involved two phases: the sensitivity analysis
and the optimization technique. The technique adopted for the mode
pairing between experimental and numerical modes is also discussed.
Phase 1: Mode pairing

The mode pairing criteria establishes the correspondence between
each experimental and numerical mode of vibration. In this work, that
correspondence is done by the application of the modal assurance
criterion MAC [41] for global modes and of the energy-based modal
assurance criterion EMAC [42] for local modes of vibration. This last
criterion enhances the traditional modal assurance criterion MAC by
the relative modal energy of a certain cluster of numerical degrees-of-
freedom related to a numerical mode.
Figure 8 — Dynamic test:

(a) accelerometer, (b) strain
gage and (c) LVDT
(a) (b) (c)
Figure 9 — Dynamic response:

(a) acceleration record and
(b) frequency content (a) (b)
The clusters reflect the information obtained from the distribution

of experimental measurement points and their directions in ambi-
ent vibration test. These clusters were related to the translational
degrees-of-freedom according to Figure 10. Figure 11 shows the MAC
and EMAC correlation matrices between experimental and numerical
modes of the initial finite element model.
Figure 10 — Clusters used in

the mode pairing
(a) (b)
Figure 11 — Mode pairing:

(a) MAC and (b) EMAC cor-
relation matrices
Phase 2: Sensitivity analysis

The sensitivity analysis allows the selection of the parameters that
most influence the frequencies and MAC values of the global and
local modes, and consequently should be included in the subsequent
optimization phase.
The sensitivity analysis showed that the modulus of deformability of

the prefabricated beam, the density of the concrete and the horizontal
stiffness of the supports have a significant influence on frequencies
and MAC values of the global modes. On the other hand, the modulus
of deformability of the upper slab influences the frequencies and MAC
values of the local modes of vibration.
Phase 3: Optimization
The optimization phase allowed obtaining the parameters values that

minimize the differences between the numerical and experimental
modal responses, and involved the definition of an objective func-
tion and the application of an optimization technique based on a
genetic algorithm. The objective function (f ) comprises two terms,
one related to the natural frequencies of global and local modes, and
other related to the MAC values of global and local modes:
exp num
n mod es fi fi n mod es
exp num
f a b MAC i , i 1 (3)
exp
i 1 fi i 1
where fiexp and finum are the experimental and numerical frequencies
for mode i, iexp and iexp are the vectors containing the experimental
and numerical modal information regarding the mode i, a and b are
weighing factors of the terms of the objective function, assumed in
this case equal to 1.0, and nmodes is the total number of modes.
Figure 12 shows the flowchart [6] with the iterative process of calibra-
tion of the numerical model, using three commercial software packages.
In Figure 13 are presented the optimization results in terms of ratio

between the experimental and numerical frequencies (Figure 13a)
and the MAC values (Figure 13b), for the three independent optimi-
zation runs (GA1 to GA3).
Figure 12 — Flowchart of the

iterative optimization process
Figure 13 — Response values

for runs GA1 to GA3: (a) fre-
quencies and (b) MAC
(a) (b)
Figure 14 — MAC values,

before and after the model
calibration
Mode 1G Mode 1L
Mode 2G Mode 2L
Figure 15 — Comparison be-

tween the experimental and
numerical mode shapes after
the calibration process
Mode 3G Mode 3L
Figure 14 shows the MAC values of global and local modes of vibra-
tion, before and after the calibration of the numerical model.
Figure 15 presents a comparison between numerical and experimental

mode shapes after the model calibration, for the first three global and
local modes of vibration.
Figure 16 — Experimental and

numerical dynamic responses
of the deck: (a) vertical dis-
placement and (b) vertical
acceleration
(a) (b)
Validation of the numerical model
The validation of the numerical model was performed based on a

dynamic test under railway traffic that consisted in the measurement
of the dynamic response in terms of displacements and accelerations of
the deck. The dynamic analyses were performed by the modal super-
position method, using a moving loads methodology, considering the
modes of vibration with frequencies up to 30 Hz and an integration
time increment equal to 0.002 s. The modal damping coefficients
adopted were those obtained experimentally.
Figure 16 shows the comparison between the experimental and

numerical dynamic responses for the passage of Alfa Pendular train at
185 km / h. The records were filtered based on a Chebyshev (type II)
low-pass digital filter with a cut-off frequency equal to 30 Hz.
Conclusions
This paper aimed to describe the experimental and numerical work

developed in order to evaluate the dynamic behaviour of a railway
viaduct with precast deck.
The ambient vibration test allowed the identification of the modal

parameters of the structure, not only associated to the global modes
of vibration but also to the local modes associated to the vibration
of the upper slab of the deck. This test also allowed verifying that
certain modes involve the vibration of adjacent spans, although all
the spans of the viaduct are simply supported, due to the continuity
established by the track.
The optimization of the numerical model was performed using a

genetic algorithm and involved 11 numerical parameters and 12 modal
responses. The results of 3 optimizations runs, based on different ini-
tial populations, led to very similar values of frequencies and MAC
values, for example, above 95% for global modes and between 85%
and 90% for local modes of vibration.
The dynamic test under railway traffic allowed obtaining a database of

displacements, accelerations and deformations of the structure. The
comparison between numerical and experimental dynamic responses
due to the passage of Alfa Pendular train revealed an excellent agree-
ment after the model calibration.
References
[1] ANSYS. Structural Analysis Guide - Release 11.0. In: ANSYS, editor. 2007.
[2] Horas, C. (2011). Dynamic behavior of bridges with precast deck on

high-speed rail lines (in Portuguese), MSc Thesis. Porto: Faculdade
de Engenharia da Universidade do Porto.
[3] ARTeMIS. (2009). ARTeMIS Extractor Pro — Academic Licence.

User’s Manual. In: SVS A, editor. Aalborg, Denmark.
[4] Allemang, R. J. (2003). The modal assurance criterion — twenty years

of use and abuse. Journal of Sound and Vibration, 37: 14–21.
[5] Brehm, M., Zabel, V., Bucher, C. (2010). An automatic mode pairing
strategy using an enhanced modal assurance criterion based on modal
strain energies. Journal of Sound and Vibration, 329: 5375–92.
[6] Ribeiro, D., Calçada, R., Delgado, R., Brehm, M., Zabel, V. (2012).
Finite element model updating of a bowstring-arch railway bridge
based on experimental modal parameters. Engineering Structures,
40: 413–435.
144 Running Safety Evaluation of Trains Moving Over Bridges shaken by Earthquakes
Running Safety Evaluation of

Trains Moving Over Bridges
shaken by Earthquakes
MONTENEGRO, Pedro Aires Abstract

University of Porto, Faculty of
Engineering, Portugal
This paper shows the development of a wheel‑rail contact model for
CALÇADA, Rui running safety evaluation of trains moving over bridges subjected to
University of Porto, Faculty of earthquakes. A contact point search methodology is presented and
Engineering, Portugal validated with results from commercial softwares. The algorithm is pre-
pared for three dimensional situations where the wheelset yaw angle
VILA POUCA, Nelson
University of Porto, Faculty of
can have a significant influence in the position of the contact point.
Engineering, Portugal This research also includes the implementation and validation of two
algorithms for computing the contact forces: FASTSIM and USETAB.
Introduction
The running safety analysis of railway vehicles during earthquakes

is one of the major concerns in railway engineering. This problem
assumes larger proportions if the train is moving over a bridge sub-
jected to such actions. Several statistics elaborated in Japan concluded
that 25% of the derailments in railway bridges were caused by trans-
verse vibrations of the deck originated by earthquakes that did not
provoke significant structural damage [1]. Such facts suggest that the
running safety can be compromised not only for intense shakings, but
also for service earthquakes with a higher probability of occurrence.
Hence, in order to take into account the effects of the transverse vibra-
tions in the running safety of railway vehicles, it is necessary to develop
a suitable train‑structure interaction model capable of describing the
wheel‑rail contact behavior. The present paper describes a research
focused on the relative motion between the wheelset and rails. With
such methodology, it is possible to store all the characteristics of the
contact points (relative position between the gravity center of the
wheelset and the rails, position of the contact points, contact angles,
body curvatures in each contact point, among others, necessary for
Running Safety Evaluation of Trains Moving Over Bridges shaken by Earthquakes 145
computing the normal forces by the Hertz theory) in a lookup table

(offline calculation) to be later used and interpolated in the dynamic anal-
ysis, resulting in a considerable reduction in computation time. Although
it is possible to make an interaction model with an online contact point
search, i.e., determining the contact point in each time step during the
dynamic analysis [2, 3], many authors proved the efficiency of an offline
calculation supported by pre‑calculated values stored in a lookup table
[4, 5]. Santamaria et al [5], showed that an offline calculation could be
up to 10 times faster than an online calculation. In this paper, a meth-
odology for three dimensional wheel‑rail geometrical problems is shown
and validated with results obtained with different commercial software
packages. Such results were obtained in the Wheel‑Rail Contact (WRC)
Manchester Benchmark [6] and published in [7]. This benchmark had the
objective of comparing the efficiency of different multi‑body dynamic
codes in the analysis of train‑track interaction.
The contact information presented in the lookup table is used to com-

pute the contact forces associated with wheels and rails. These forces
are subdivided into normal and tangential forces, being the former
computed according to the Hertz Theory [8]. The tangential wheel‑rail
creep forces, which are dependent on the normal contact tensions,
are computed with Kalker’s Simplified Theory [9], throughout FAST-
SIM algorithm, and through USETAB algorithm, based on results given
by Kalker’s Exact Theory [10]. FASTSIM algorithm is a compromise
between efficiency and accuracy (the deviation in the results is not
larger than 10% compared to those obtained with the Exact Theory).
USETAB consists on a table‑based program, using pre-calculated val-
ues obtained with the Exact Theory, for the creep‑spin force / moment
law and it is 8 times faster than FASTSIM. For this reason, FASTSIM
and USETAB, with some modifications to improve its accuracy, were
implemented in this work.
Methodology for evaluation of the contact geometry
Assumptions and coordinate systems

In a wheel‑rail contact model, the geometrical problem is a crucial
point in order to achieve good results during the dynamic analysis. The
geometrical problem consists of searching the contact point through
the wheel and rail surface to get all the geometrical information in
that same point (rolling radius, curvature radius in both longitudinal
and lateral directions, among others) to be subsequently used in the
computation of the contact forces. For achieving this, the following
assumptions were made:
–– There is only one contact point for each wheel‑rail pair;

–– There is always contact between the wheel and the rail;
–– Both the wheels and rails are considered to be rigid;
–– At a contact point, the vertical distance between the wheel and

the rail is null (no penetration allowed), and is greater than zero
at non‑contact points.
The searching contact point algorithm depends on the relative dis-

placements of the wheelset with respect to the centerline of the track.
Unlike other simpler where the assumption of a total compatibility
of movements between structure and vehicle is made, in a complete
wheel‑rail interaction model, relative movements between these two
bodies are permitted. However, considering the previous assumptions,
the six degrees‑of‑freedom (DOF) of the wheelset can be reduced to two
independent DOFs: the relative lateral displacement, yg, and the yaw
angle, . In these cases, the problem is considered three dimensional.
Some simplification can be made if the influence of the yaw angle is

neglected, which is allowable in certain situations, for instance for
trains running on straight tracks or curves with large curvature radius.
In such cases, all the DOFs of the wheelsets are dependent only on the
lateral displacement and the contact point search is restricted to a two
dimensional section called main cross‑section. Wheel‑rail coordinate
systems are the basis for wheel‑rail relationships. Figure 1 summarizes
all the coordinate systems used in the present work that could be
described as follows:
–– G-XYZ: Global inertial coordinate system, which has origin at the

center of gravity of the wheelset and moves with it at constant
forward speed. The X axis is longitudinal, the Y axis is directed to
the right side and the Z axis is vertical, pointing downwards;
–– G-XgYg Zg: Wheelset coordinate system, which follows the wheel-
set motion and has origin in its gravity center. The Yg axis is
positioned along the wheelset axle directed to the right side, the
Xg axis is longitudinal and Zg forms a Cartesian coordinate system,
pointing downwards;
–– CPr-XcrYcr Zcr: Right contact patch coordinate system, with its
origin in the right contact point. The Xcr axis is longitudinal,
while Ycr and Zcr are located in the right tangent contact plan
defined by the right contact angle, r;
–– CPl-XclYclZcl: Left contact patch coordinate system, which has its
origin in the left contact point. Its definition is analogous to the
right contact patch coordinate system.
Figure 1 — Definition of the

coordinate systems a) Wheelset coordinate systems b) Contact patch coordinate systems
Wheel-rail geometrical contact problem
Algorithm implementation
Considering that the wheelset can displace perpendicularly to the
track, its relative movement with respect to the track can be totally
described by four components: the lateral displacement, yg, the ver-
tical one, zg, the roll rotation, x and the yaw rotation, . However,
these four DOFs can be reduced to two independent DOF, yg and .
Hence, when the yaw angle is take into account, mandatory when
the train is passing in a tight curve, the contact point can shift in the
Xg direction, which implies the wheel and rail meshing and. The rail
surface is obtained by means of an extrusion of the rail profile along the
X axis, while the wheel surface is generated through a revolution of its
main cross-section around the Yg axis.
In order to build an efficient algorithm for this purpose, the geometry

of the surfaces from both wheel and rail have to be known. Since these
surfaces are not provided by simple analytical functions, a series of
control points were considered to define both sections with a spline
parameterization scheme. All the following algorithms were pro-
grammed with the commercial software MATLAB.
The search for the contact points is accomplished by the calculation

of the minimal vertical distances between wheels and rails in the
left,  z lmin, and right, min  z rmin , sides of the wheelset. The two
points are considered to be the left and right contact points if the
following condition is met:
l r
Z min Z min (1)
Otherwise, the condition is violated and the roll angle has to be

adjusted. The number of iterations needed to find the contact points
is substantially reduced if the roll angle adjustment, , takes the fol-
lowing value [11]: l r
Z min Z min
x r l
(2)
y min y min
l r
where y min and y min is the lateral location of both left and right points
with minimum vertical separation between wheel and rail. The search
ends when condition (1) is fulfilled, with a tolerance value of 10¯³ mm
(enough from the engineering point of view). When this point is
reached, it is possible to know the values of the dependent DOFs of
the wheelset, such as the roll angle, x, and the relative vertical dis-
placement, zg (see Figure 1):
k k
0 0 i i
x xx x x x
i 1
i 1 (3)
l r
zg zmean
g meanz l z, r
z, z
min min min min
where x0 is the initial roll angle (considered equal to zero in this
work) and k is the number of iterations needed to fulfill condition (1).
The longitudinal displacement and wheelset rotation are constraints of
the problem. The search algorithm can be resumed by the flow chart
presented in Figure 2.
Results
Figure 3 presents the relation between the lateral displacement of the
wheelset and its dependent DOFs. The shape profile of the sections of
the wheels and rails are S1002 and UIC60, respectively, with an inter-
nal gauge of 1.360 m, a wheel nominal radius, rw 0, of 0.460 m and
a rail inclination of 1:40 (these profiles and geometry were the same
as those used in the WRC Manchester benchmark [6]). In order to
clarify the relative movement between the wheelset and the track, an
example of the wheel‑rail contact for a lateral displacement of 10 mm
and a yaw rotation of –5º is also presented. Note that, for a lateral
displacement larger than 6 mm, the contact in one of the wheels takes
place in the flange, which leads to a sudden rise of the roll angle of the
wheelset and the position of its gravity center.
Validation with WRC Manchester Benchmark results
In 2006, the Rail Technology Unit at Manchester Metropolitan Uni-

versity coordinated a benchmark for assessing the impact of modelling
assumptions for the wheel-rail contact on the simulation of railway
vehicle dynamics [6]. A significant number of multi‑body dynamic
codes were tested in order to investigate the effects of different
wheel‑rail contact models. One of the tests consisted on evaluating
Figure 2 — Flow chart for

the wheel‑rail contact point
search procedure
a) Relation between {yg, } and zg b) Relation between {yg, } and x
Figure 3 — 3D wheel-rail con-

tact case results
c) Left contact (yg = 10 mm;  = –5º) d) Right contact (yg = 10 mm;  = –5º)
the contact conditions of a single wheelset subjected to a prescribed

motion. Based on this test, it was possible to validate the contact
point search algorithm with results obtained by various commercial
software packages. Such results were published by Shackleton and
Iwnicki [7]. In order to compare the results from the present work
with the ones obtained in the benchmark, the same local coordinate
systems were adopted:
–– XwYw Zw: Local wheel coordinate system with origin in the point

with nominal radius. The Xw axis is located along the rolling
direction and the Zw vertically down. The Yw axis makes a right-
hand Cartesian for the right wheel and a left‑hand Cartesian for
the left wheel;
–– XrYr Zr: Local rail coordinate system with origin in the highest
point of the inclined rail profile. The axis orientation follows the
same principle as the local wheel coordinate system.
One of the main variables needed for solving the normal and tangen-
tial problem described in the next section is the contact point posi-
tion. Figure 4 compares the contact point position, referred to the
wheel coordinate system, obtained in the present work and with the
commercial software packages tested in [7]. The wheelset was dis-
placed laterally from 0 mm (central position) to 10 mm at 0.5 mm
increments. The results obtained in the present work are consistent
with the ones shown in [7].
When the wheelset displaces laterally in respect to the track, the

wheel radius and contact angle changes due to the characteristics of
the wheel profile. Figure 5 and Figure 6 compare the differences of the
rolling radius of the wheel, between the right and left wheel, and con-
tact angles obtained in this research and in [7]. For the computation
of the contact angles, the wheelset was subjected not only to a lateral
a) Left wheel contact position [7] b) Left wheel contact position (present)
Figure 4 — Contact positions

according to the wheel'
coordinate systems
c) Right wheel contact position [7] d) Right wheel contact position (present)
displacement, but also to a yaw rotation. The lateral displacement

sequence mentioned above was accompanied by a yaw angle increment
of 1.2 mrad from 0 mrad to 24 mrad (negative values). The results show
a good agreement, even when the yaw angle is taken into account.
a) According to [7]
Figure 5 — Rolling radius dif-

ference between right and left
wheel for a lateral displace-
ment of the wheelset
b) Present algorithm
Figure 6 — Contact angle for

the right wheel with a lateral
displacement and yaw rotati-
on of the wheelset
a) According to [7] b) Present algorithm
Contact forces computation
The contact forces are divided in two groups: the normal forces and
the tangential forces. Such forces are projected in the contact patch
coordinate systems presented in Figure 1, in which the normal force
lies on Zci axis while the tangential ones are in the Xci (longitudinal)
and Yci (lateral) axis. In this work, Hertz theory is used to solve the nor-
mal problem, while the tangential problem is solved by FASTSIM [9]
and USETAB [12] routines.
Normal problem
The normal contact problem consists on determining the contact sur-

face dimensions and the normal stresses produced on it when two
bodies are compressed to each other. In this research, the shape of the
contact surface is elliptical and it is calculated by means of Hertz theory.
This model is only valid if the following assumptions are made [8, 13]:
–– The contact area is small when compared to the curvature radius

of the bodies;
–– The curvatures inside the contact patch are constant;
–– The deformations arising from the normal contact are elastic;
–– Both bodies in contact have smooth surfaces.
–– The semi-axes a and b of the ellipse can be computed using the
following expressions proposed by Iwnicki [13]:
3 3 1 v2 1
a m FN (4)
2 E A B
3 3 1 v2 1
b n FN (5)
2 E A B
where E and v are the combined elasticity modulus and Poisson’s ratio
from both materials, Fn the normal contact force, A and B the com-
bined curvatures from both bodies (wheel and rail) in direction Xg and
Yg, respectively, and m an n two non-dimensional coefficients depend-
ent of the curvatures (detailed in [13]).
Tangential problem
The wheel‑rail contact is a rolling friction contact that differs from

the Coulomb's sliding friction theory because, unlike the last one, the
contact patch is divided into an area of adhesion and an area of slip
which appears progressively as the slip speed increases. In the pre-
sent research, two methods for solving the tangential contact problem
were implemented: FASTSIM algorithm [9] and USETAB routine [12].
FASTSIM algorithm
FASTSIM uses the simplified theory of elasticity for relating the defor-
mations in the contact patch with the tangential stresses. In this algo-
rithm, the contact ellipse is discretized in strips parallel to the rolling
direction, each one divided into the same number of elements. The
stresses are computed in the center of each element through a numer-
ical integration method, where the tangential stress in one element
depends on the tension of the previous element in the strip. The main
inputs in a rolling contact theory are the relative velocities between
the two bodies (wheel and rail in this specific case), which are called
creepages. The creepages are divided in three components: the longitudi-
nal creepage, , the lateral creepage, , and the spin creepage, ϕ. Generi-
cally, these can be computed by the following expressions:
xc,1 xc,2
(6)
V
yc,1 yc,2
(7)
V
zc ,1 zc ,2
(8)
V
where vxc,i, vyc,i and ωzc,i are the longitudinal, lateral and spin veloci-
ties of body i projected in the contact patch coordinate system and V
the imposed translation speed of the train.
Thus, taking advantage of the simplified theory of elasticity (the rela-

tion between displacements and stresses is linear), the tangential
stress in both directions, x and y, in a point of the contact patch with
coordinates (xc, yc) can be related with the stress on a previous point in
the same strip (xc +  xc, yc)through the following expression:
x xc , yc x xc xc , yc xc yc xc sxc
(9)
y xc , yc y xc xc , yc L xc L syc
where sxc and syc are the relative displacements between the two bod-
ies in direction xc and yc, respectively, and L is a flexibility parameter
described, for example, in [9].
As previously mentioned, the algorithm is nonlinear, since the tangen-

tial tension cannot overcome the Coulomb’s friction limit given by:
x c , yc . x c , yc (10)
where  is the friction coefficient and  (xc, yc) is the normal tension.
Therefore, in a first iteration, the current point is considered to be in
adhesion, which implies that the last term in equation 9 is equal to
zero. The computed stresses are called hypothetical tensions, H, and
are compared to the Coulomb’s friction limit. If H overcomes such
limit, the point is in slip condition, otherwise the assumption is cor-
rect and the hypothetical tension coincides with the real tension.
USETAB algorithm
The USETAB routine consists of a table with four entries (creepages and
ellipse semi‑axis ratio) and three outputs (longitudinal and lateral creep
forces and spin moment). This table, proposed by Kalker [12], contains
pre‑calculated values from creep forces for a large number of cases,
computed with the software CONTACT [14] which solves rolling con-
tact problems with Kalker’s Exact Theory. Such routine cannot be used
directly in a railway dynamic code due to its high computational cost;
instead, it is possible to build a table and then linearly interpolate its val-
ues during the dynamic analysis. For solving each rolling contact problem,
USETAB is 8 times faster than FASTSIM and gives more accurate results.
The table uses a smart layout, taking advantage of all creepage sym-
metries in order to reduce the number of cases and to normalize it. So,
the table is made exclusively for railroad dynamics, with a Poisson’s
ratio of 0.28 (typical for steel) and can be used only with elliptical
contact areas. According to Kalker’s work exposed in [12], the USETAB
routine is constructed with the following normalized quantities:
–– Shear modulus of rigidity, G, equal to 1;

–– Coulomb's friction limit, .FN, equal to 1;
–– Square root of the ellipse semi‑axis product, a.b, equal to 1;
Finally, in order to obtain the correct values of creep forces, Fx and Fy,
and spin moment, Mz, the outcome is multiplied by:
Fx Fx G.a.b. . FN (11)
Fy Fy G.a.b. .FN (12)
3
Mz Mz G. a.b 2 . .FN (13)
Kalker proposed a table with 115,000 entries, transforming the range

0 < q < into 0 < q < 1 and 0 < 1/q <, where q is one of the table entry,
with 7 interpolation intervals each. However, for high creepages or
high ellipse semi‑axis ratios, the interpolation intervals become large

(the interpolation is made with the inverse value of the entry for q > 1),
which implies a loss of accuracy. Therefore, in order to overcome
such limitation, the original table was replaced by a larger one, with
288,000 entries, where the range q > 1 is no longer interpolated through
inverse values. Instead, the interval 0 < 1/q < is replaced by 0 < q < 15
with 10 interpolation intervals (the value 15 was chosen in order to
double the accuracy limit provided by the original table, which is 7).
Hence, although the USETAB routine is valid for almost all cases, when
the input value is beyond the table boundaries (rarely attendant), FASTIM
is called during the dynamic analyses for solving these exceptions.
Validation of the tangential algorithms

In the present sections, the two implemented algorithms for solving
the rolling contact problem are validated through a comparison with
results given by the Exact Theory implemented in the commercial
software CONTACT. The following graphics show the evolution of the
contact forces for different creepage combinations. The creepages
were scaled through the following expressions:
abG
(14) c 11
3 FN
abG
(15) c
3 FN 22
32
ab G
(16) c23
FN
where c11, c22 and c23 are the Kalker’s coefficients that can be found in
[10]. For this validation, a Poisson’s ratio of 0.28, a Young’s modulus
of 210 GPa and a normal force of 100 kN were considered.
Figure 7 compares the evolution of the tangential contact forces,

normalized by Coulomb's friction limit, μ.FN, with μ = 0.3, among
the exact solution, the original USETAB, the implemented FAST-
SIM and the modified USETAB with a larger number of entries, for
equal semi‑axes of the ellipse and two combinations of creepages.
The original USETAB provides good results for creepages until 1.0, but
after that the accuracy starts to decreases. This happens because, for
creepages above 1.0, the interpolation intervals start to be too large,
resulting in a loss of accuracy for such cases. Analyzing the modified
USETAB results, the accuracy achieved by this routine is better, since
the number of entries in the table is much higher, resulting in smaller
interpolation intervals. As for FASTSIM, as previously mentioned, the
results follow the tendency of the exact theory but with errors that can
reach 10% for very high creepages.
a) Lateral contact force for ϕ = –
Figure 7 — Comparison of the

contact force obtained by dif-
b) Tangential contact forces for f ϕ =  ferent algorithms
Conclusions
A wheel‑rail contact model for evaluating the contact position in 2D

and 3D applications is presented and properly validated with results
from commercial software packages obtained in the WRC Manchester
Benchmark. When several contact characteristics, namely the position of
the contact point, the instantaneous rolling radius of the wheel, the
wheel‑rail contact angle, among others, obtained with the proposed
contact search algorithm are compared with the ones obtained with
commercial programs, the agreement among them appears to be suffi-
ciently accurate. Therefore, the pre‑processor is suitable to make a lookup
table for the wheel‑rail interaction in a railroad dynamic code. The pro-
posed three dimensional version of the algorithm is appropriate, not only
for vehicles running in straight tracks, but also for cases where the yaw
angle of the wheelset can have a significant influence in the final results.
For the computation of the tangential contact force, two algorithms

were implemented with successful results when compared to the
exact solution given by the commercial software CONTACT developed
by Kalker. A more accurate version of Kalker’s USETAB routine is also
proposed. The contact search algorithm, as well as the contact force
computation routines, will be used in a railroad dynamic code with
the intention of evaluating the running safety of trains passing over
bridges subjected to vibrations imposed by service state earthquakes.
Acknowledgments
The present work has been funded by the Portuguese Foundation for
Science and Technology (FCT), in the context of the doctoral grant
number SFRH / BD / 48320 / 2008. The authors would also like to
thank Prof. Simon Iwnicki and Prof. Philip Shackleton (Manchester
Metropolitan University) for the permission to use figures from their
paper referenced in [7].
References
[1] Matsumoto, N., M. Tanabe, M. Sogabe and H. Wakui — “Running

safety analysis of vehicles on structures subjected to earthquake
motion”, Quarterly Report of Railway Technical Research Institute
(Japan), 45 (3), 116–122, 2004.
[2] Shabana, A., K. Zaazaa, M. Tobaa and H. Sugiyama — “On the com-
puter formulations of the wheel / rail contact problem”, Nonlinear
Dynamics, 40 (2), 169–193, 2005.
[3] Pombo, J., J. Ambrosio and M. Silva — “A new wheel-rail contact

model for railway dynamics”, Vehicle System Dynamics, 45 (2),
165–189, 2007.
[4] Antolín, P., A. Alonso, J.M. Goicolea and M.A. Astiz, A methodology
for analysing lateral coupled behavior of high speed railway vehicles
and structures, in 9th World Congress on Computational Mechanics
and 4th Asian Pacific Congress on Computational Mechanics 2010:
Sydney, Australia.
[5] Santamaria, J., E.G. Vadillo and J. Gómez — “A comprehensive method

for the elastic calculation of the two-point wheel-rail contact”, Vehicle
System Dynamics, 44 (1), 240–250, 2006.
[6] Shackleton, P. and S. Iwnicki — “Wheel-Rail Contact Benchmark

Report”, Version 3.0, Manchester Metropolitan University, Manchester,
U.K., 2006.
[7] Shackleton, P. and S. Iwnicki — “Comparison of wheel-rail contact

codes for railway vehicle simulation: an introduction to the Man-
chester Contact Benchmark and initial results”, Vehicle System
Dynamics, 46 (1), 129–149, 2008.
[8] Wriggers, P. and T. Laursen — “Computational Contact Mechanics”,

SpringerWienNewYork, Udine, Italy, 2007.
[9] Kalker, J. J. — “A fast algorithm for the simplified theory of rolling

contact”, Vehicle System Dynamics, 11 (1), 1–13, 1982.
[10] Kalker, J. J. — “The computation of three-dimensional rolling contact

with dry friction”, International Journal for Numerical Methods in
Engineering, 14 (9), 1293–1307, 1979.
[11] Li, Z. — “Wheel-Rail rolling contact and its application to wear

simulation”, Ph.D. Thesis, Delft Technical University, Delft, The Nether-
lands, 2002.
[12] Kalker, J. J. – “Book of tables for the Hertzian creep-force law”,

in I. Zobory (Ed.) Proceedings of the 2nd Mini Conference on Con-
tact Mechanics and Wear of Wheel/Rail Systems, Budapest,
Hungary, 1996.
[13] Iwnicki, S., J.–B. Ayasse and H. Chollet — “Chapter 4: Wheel-Rail

Contact”, in Handbook of Railway Vehicle Dynamics, CRC Press,
Boca Raton, FL, USA, 2006.
[14] CONTACT 11.1 — “User guide for CONTACT, J.J. Kalker’s variational
contact model”, VORtech Computing, 2011.
Numerical and experimental study of the vertical interfaces
158 behaviour of interconnected structural masonry walls
Numerical and experimental

study of the vertical interfaces
behaviour of interconnected
structural masonry walls
OLIVEIRA, Luciane Marcela Abstract

Filizola
School of Engineering of São
The shear strength of the vertical interfaces is a key parameter to
Carlos, University of São Paulo
guarantee the flange contribution to shear walls and the possibility
CORRÊA, Márcio Roberto of shear force transference between interconnected walls. In this
Silva regard, this research presents the results of an experimental and
School of Engineering of São numerical investigation to study the behaviour of vertical interfaces
Carlos, University of São Paulo
of interconnected concrete blockwork masonry walls. H-shaped test
specimens were used, constructed in running bond. A uniformly dis-
tributed pressure was vertically applied on the web of the specimens.
The experimentally obtained shear strengths are compared with the
Brazilian code. Numerical analysis was also carried out in DIANA®
finite element program to simulate wall behaviour focusing on the
shear at the vertical interfaces and their ability to transfer shear forces
from web to flanges. The shear strength of the vertical interface was
obtained with a good correlation to the theoretical analysis and Bra-
zilian code can be very strict.
Introduction
When the walls are connected and subjected to different loads, there
is interaction between them. The interaction is dependent on the shear
capacity of the interface. In practical situations, many walls are stiff-
ened due to flanged sections which increase their shear strength. Tests
on H-shaped walls have revealed that more studies and improvement
are needed in shear strength in interconnected masonry walls area.
The type of connection between the walls influences the way inter-
action occurs. The supporting flanges are usually linked to the web
through the running bond. In that case, interaction might be higher
because the interface plane crosses units which may be consid-
behaviour of interconnected structural masonry walls 159
ered common to the connected walls. Alternatively, the connection

between the walls is through metal shear connectors embedded in the
bed joints and extended across the shear plane with the vertical joint
filled with mortar at the interface. The present study used the running
bond connection type.
Masonry walls with flanges have already been studied by some

researchers (Lissel et al. [1]); Camacho et al. [2]; Yoshimura et al. [3];
Modena et al. [4]; Maurício [5]; Capuzzo Neto et al. [6]; Bosiljkov et al.
[7]). However, these types of specimens need special testing apparatus.
Lissel et al. [1] tested shear walls with flanges and observed the influ-
ence of the type of connection on their behaviour and strength. The
tests indicated that bonding pattern of the web-flange connection
affect the shear strength of masonry. Bonding the web-flange connec-
tion with brickwork clearly increases the shear strength of this joint.
Hence, a series of specimens with H-shaped cross sections was tested
in an attempt to determine the effect of bond pattern on the strength
of the web-flange intersection. A nominal compressive force approxi-
mately equivalent to a normal floor load was applied to the flanges
to stabilize the specimen during and after the test. According to Lis-
sel et al. [1], the results of these tests indicate that the mechanical
interlock of a bonded web-flange connection provides a significant
structural advantage over a tied connection. The forces applied on
the specimens with bonded wall are three times the forces applied on
specimens with tied connection.
Capuzzo Neto et al. [6] conducted new studies at the School of Engi-
neering of São Carlos (University of São Paulo), contributing towards
the understanding of the wall interaction phenomena. The author
strived for a better representation of possible stress trajectories along
the building structure, including proposing an H-shaped specimen to
evaluate the shear strength of the vertical interface. These specimens
were used in this work to obtain the shear strength through experi-
mental tests. The computational modelling of masonry structures can
focus on micro-modelling of individual components or macro-model-
ling of masonry as a composite (Lourenço and Rots [8]). According to
Lourenço [9], it is possible to use one of these strategies, depending
on the level of accuracy and the simplicity desired.
In the detailed micro-modelling units and mortar in the joints are rep-
resented by continuum elements whereas the unit-mortar interface is
represented by discontinuous elements. In simplified micro-modelling
expanded units are represented by continuum elements whereas the
behaviour of the mortar joints and unit-mortar interface is lumped
in discontinuous elements. Macro-modelling represents units, mortar
and unit-mortar interface smeared out in the continuum. (Figure 1)
Figure 1 — Modelling strat-

egies for masonry structures
(Lourenço, 1996) [9]:
(a) detailed micro-modelling;
(b) simplified micro-modelling;
(c) macro-modelling.
It is well known that both approaches provide satisfactory results.

Macro-models are applicable when the dimensions of the analysed
structure are sufficiently large so that the stresses across or along a
macro length will be essentially uniform. Besides, the low compu-
tational cost supports its use on the analysis of large structures. On
the other hand, micro-models are applicable in very specific prob-
lems where local failures should be analysed. Even though micro-
modelling includes all the basic failure mechanisms that characterize
masonry and enables the detailed representation of resisting mecha-
nisms of the walls, the macro-modelling approach was chosen since
the research is in the calibration phase, and this approach is enough
for preliminary analyses.
Experimental Details
For the physical and mechanical characterization tests, concrete hol-

low blocks were used for the construction of the masonry specimens.
They had a percentage of holes equal to 55. The mortar utilized for bed
joints was a mixture of cement, hydrated lime and sand. The thickness
of the bed joints was approximately 10 mm. Face shell bedding was
used on the construction of the specimens.
Uniaxial compression tests were carried out on 12 blocks, 12 mortar

specimens, 12 prisms and 6 wallets (about 790 x 1000 x 140 mm
of dimension), according to EN 772-1 [10], NBR 13279 [11], NBR
15812-2 [12] and EN 1052-1 [13], respectively. Tests in 12 units were
carried out to determine the splitting tensile strength of masonry units
according to ASTM C1006 [14]. Plus, triplet specimens were built
with the same materials to characterize the shear behaviour of con-
crete unit-mortar through initial shear tests carried out according to
EN1052-3 [15] (Figure 2).
However, due to the difficulty of obtaining mode I fracture energy and

the shear stiffness of the unit-mortar interface, this mechanical prop-
erty could not be used in the numerical model. Table 1 summarises the
material characterization tests results.
Figure 3 shows the compressive strength tests and splitting tensile

strength tests of blocks, and the compressive strength tests of wallets.
Figure 4 shows the stress-strain diagram results of all compressive tests.
Material Physical and mechanical characteristics
Blocks Dimension Weight Mean comp. Elastic Splitting

(mm) (kg) strength modulus tensile
(N / mm²) (N / mm²) strength
(N / mm²)
390 x 190 x 140 12.70 10.21 9,910 0.91
Mortar Composition Composition Mean comp. Elastic

(C:L:S) in (C:L:S) in strength modulus
volume weight (N / mm²) (N / mm²)
1:1:6 1:0.66:8.21 3.52 6,800
Wallets Compressive Elastic

strength, modulus
(N / mm²) (N / mm²) Table 1 — Characteristics of
4.88 7,900 the materials
H-shaped specimen was used for the shear tests. All the tests were
carried out in the Structures Engineering Laboratory at the School of
Engineering of São Carlos, University of São Paulo. The specimen was
prepared by a professional mason and were cured in laboratory condi-
Figure 2 — Initial shear tests
(a)
Figure 3 — (a) Units in com-

pressive strength and (b) split-
ting tensile strength tests.
(c) Wallets in compressive
strength tests.
(b) (c)
tions for 28 days before testing. The type of bonding used was run-
ning bond. Face shell bedding was used on the construction of the
specimens. (Figure 5)
As shown in Figure 6, the relative vertical displacements between the

web and the flanges (due to shear deformations along the flange-web
interface) were measured using LVDTs located in the sample and con-
nected to the actuator at a rate of 0.001 mm / s (Figure 6).
The web in H-shaped test specimens was 990 mm in height and

810 mm in length. The length of the flanges was 940 mm. Prior to
applying a shear load to the H sections, a small pre compressive stress
of 0.5 MPa was applied to each flange to stabilize the specimen and
simulate a typical level of preload. According to Bosiljkov [7], the level
of pre-compression in the flanges also influenced the shear resistance,
Figure 4 — Stress-strain dia-

grams of the specimens in
compressive strength tests.
Figure 5 — Construction of
the Specimens.
Figure 6 — Instrumentation of
the specimens.
Figure 7 — Geometric specifi-

cation of the specimens.
but only up to a certain limit (of approximately 0.5 MPa). A compres-

sive load has been applied on the top web of the sections under dis-
placement control to produce shear in the flange-web interface. The
load was applied monotonically to failure (Figure 7).
Numerical Modelling
The numerical model applied to study masonry walls was defined using
the DIANA® finite element program. The macro-modelling approach
was chosen due to the reasons previously described. Some mechanical
properties used in the description of the material models were obtained
through experimental tests carried out on materials and masonry assem-
blages. Others were adopted from current research, due to the difficulty
of obtaining them. The mesh was composed of continuum elements
representing the masonry units and the masonry joints as a composite.
It is common to model potential cracks in units in order to avoid

an overestimation of the collapse load and stiffness. Thus, potential
cracks placed at the web-flange intersection were considered. Inter-
face elements with a discrete cracking model were used to represent
the interlocking of units located in the connection of the flanges with
the web wall. The stiffness of these interfaces was defined by fitting
the numerical to the experimental results obtained in the masonry
walls. An exponential softening behaviour was adopted for the tensile
of these interfaces. The tensile strength (ft = 0.91 MPa) was obtained
through tensile tests performed on the concrete units. The mode I
fracture energy (Gf I = 0.047 N / mm) was obtained from the experi-
mental results in Mata (2011) in concrete blocks with a similar com-
position of raw materials. The constitutive law for discrete cracking
Figure 8 — Elements used

in 3-D numerical modelling
(DIANA®)
in DIANA® is based on a total deformation theory, which expresses

the tractions as a function of the total relative displacements. Eight-
node isoparametric solid brick elements with Gauss integration scheme
were adopted to represent the walls (HX24L-DIANA®). An interface
element between two planes in a three-dimensional configuration
was adopted to represent the interface at the flange-web intersection
(Q24IF-DIANA®) (Fig. 8).
Monotonic loading was applied to the specimens in the numerical

analysis. Similarly to the experimental tests, the flanges have firstly
been subjected to a vertical uniformly distributed load p = 0.5 MPa
which was maintained constant, in addition to the self weight. After
that, vertical load were imposed on the web until failure. The bound-
ary conditions considered were fixed bottom walls to simulate the
experimental tests. The bottom concrete slabs were not included since
their consideration does not influence the masonry wall behaviour.
Results and Discussion
The ultimate capacity of interlocked specimens was 169 kN. The shear

strength of interlocked wall intersection was 0.61 MPa. The test
exhibited little deformation and some cracks before failure and almost
pure shear failure could be achieved at the flange-web intersection. It
can be seen that the vertical head joint at the flange-web intersection
broke clearly away from the flanges at the same time as shearing of the
intersecting block occurred (Figure 9).
The numerical load versus displacement curve is plotted in Figure 10

and compared with the experimental one. A reasonably good agree-
ment has been found. However, the results presented linear behaviour
until the experimentally obtained ultimate load. This indicates that it
is necessary to calibrate the properties of the interface for the model
represents its behaviour correctly.
Conclusions
In the experimental test, masonry walls failed under shear and the shear
strength of interlocked wall intersection was 0.61 MPa. From previ-
ous research on the same materials (Mauricio [5]), the shear strength
of interlocked wall intersection found was 0.88 MPa. Both results are
larger than the Brazilian code indicates, 0.15 MPa. It indicates that
Brazilian code can be very strict. Results pointed out some aspects and
indicators in the behaviour of flanged walls that should be numerically
and experimentally confirmed in order to clarify the behaviour of these
structures and to allow the development of accurate design models.
Figure 9 — Failure mode of

specimen.
Figure 10 — Vertical web

displacement.
References
[1] Lissel, S. L., Shrive, N. G., & Page, A. W. (2000). Shear in plain, bed
joint reinforced, and posttensioned masonry, In : Canadian Journal
of Civil Engineering, Vol. 27 , No. 5, pp. 1021–1030.
[2] Camacho, J. S., Ramalho, M. A.; & Andolfato, R. P. (2001). An

experimental study of the interaction among walls submitted to
vertical loads, In: Amc — 6th Australian Masonry Conference, Ade-
laide: Griffith, Vol. 1, pp. 95–104.
[3] Yoshimura, K., Kikuchi, K., Kuroki, M., Nonaka, H., Kim, K.T., Mat-
sumoto, Y. Itai, T., Reeznag, W., Ma, L.. Experimental study on re-
inforcing methods for confined masonry walls subjected to seismic
forces, Proceedings of 9th North American Masonry Conference,
Clemson, South Carolina, USA, 2003.
[4] Modena, C., Porto, F., Valluzzi, M.R., Reinforced and rectified clay
block masonry, Proceedings of 6th National Congress of Seismology
and Seismic engineering, Guimarães, Portugal, 2004, pp. 155–177
[5] Maurício, R. M. (2005). Theoretical and experimental study of direct

stretcher bond connections between walls of concrete blocks in real
and small-scale 1:4, Msc Thesis, Engineering College of Ilha Solteira,
Ilha Solteira, São Paulo, Brazil. (in Portuguese).
[6] Capuzzo Neto, V., Corrêa, M. R. S., & Ramalho, M. A. (2007). Shear
strength of vertical interfaces of intersecting walls, In: 10th North
American Masonry Conference, Saint Louis, Boulder, The Masonry
Society, Vol. 1. pp. 872–883
[7] Bosiljkov, V., Page, A. W., Simundic, M. S. G., & Zarnic, Roko (2010).
Shear capacity of the Flange-Web Intersections of Brick Masonry
Nonrectangular Sections, Journal of Structural Engineering, Vol. 136,
No. 5, May, pp. 574–585.
[8] Lourenço, P.B.; Rots, J.G. (1997). Multisurface interface model for ana-
lysis of masonry structures, Journal of engineering mechanics,
V. 123(7), 660–668
[9] Lourenço, P.B. (1996). Computational strategies for masonry struc-

tures, PhD Thesis, Delft University of technology, Delft, The Nether-
lands. ISBN 90-407-1221-2.
[10] EUROPEAN STANDARD. EN 772-16, Methods of tests for masonry

units — Part16: Determination of dimensions, 2000.
[11] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS (ABNT).

NBR 13279, Mortars applied on walls and ceilings — Determination
of the flexural and the compressive strength in the hardened stage.
Rio de Janeiro, 2005. (in Portuguese)
[12] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS (ABNT).

NBR 15812-2, Structural masonry — Clay blocks Part 2: Execution
and site quality control. Rio de Janeiro, 2010. (in Portuguese).
[13] EUROPEAN STANDARD. EN 1052-1, Methods of test for masonry:

Part 1 — Determination of compressive strength, 1999.
[14] ASTM C1006, Standard Test Method for Splitting Tensile Strength
of Masonry Units (1996).
[15] EUROPEAN STANDARD. EN 1052-3, Methods of test for masonry:

Part 3 — Determination of initial shear strength, 2002.
168 Low cycle performance of t-stub components of bolted moment beam-to-column connections
Low cycle performance of T-stub

components of bolted moment
beam-to-column connections
POP, Ana-Maria Abstract

Politehnica University of
Timisoara, Romania
The present research represents the numerical investigation used to
GRECEA, Daniel characterize the behaviour of bolted steel beam-to-column joints in
Politehnica University of low-cycle fatigue. The study is based on existing experimental investi-
Timisoara, Romania gation on T-stub components performed at the Politehnica University
of Timisoara, using mild and high strength steel. First a calibration of
CIUTINA, Adrian
Politehnica University of numerical results is performed based on the monotonic experimental
Timisoara, Romania response. The challenge of the numerical investigation is represented
by the calibration of cyclic curves, using adequate FEM techniques.
Introduction
In the seismic design of steel structures, the Moment Resisting Frames

(MRF) are recognized as highly dissipative structures. In consequence,
the seismic input energy is dissipated through plastic deformations
concentrated in specific locations, whose behaviour has to be pre-
dicted by proper design. According to the Eurocode 8 rules, the dissi-
pative zones could be located either in elements or in beam-to-column
joints [1]. As shown by previous studies, the end-plate connections
could prove adequate rotation capacity if special measures are taken,
e.g. use of relatively thin end-plates, avoiding brittle failure of welds
and bolts etc. From this point of view, the T-stub element (see Fig. 1)
represents the key-macro-component of the connection behaviour.
Figure 1 — End-plate connec-

tion: T-stub element
Low cycle performance of t-stub components of bolted moment beam-to-column connections 169
Globally the T-stub macro-component can lead to 3 types of failure

modes, depending of the connection component characteristics (see
Figure 1 and Table 1). Mode 1 is represents generally a very ductile behav-
iour and is characteristic to thin end-plates, but cannot be considered a
real solution to strength demands in case of seismic resistant structures.
On contrary, the Mode 3 leads usually to a good T-stub resistance, while
the failure mode is fragile in nature by bolt rupture. In consequence,
the Mode 2 could answer well to both strength and ductility demands.
From this point of view, the Dual Steel (DS) concept can be extended
to connections, based on the same philosophy used in structures:
using High Strength Steel (HSS) in non-ductile components that should
poses over-strength and Mild Carbon Steel in ductile components in
order to achieve both ductility and robustness criteria.
Figure 2 — Types of T-stub

failure modes
Based on this principle, a large experimental research program was car-

ried out at the CEMSIG Research Centre of the “Politehnica” University of
Timisoara, in order to study the performance of dual-steel configura-
tion for beam-to-column joints under monotonic and cyclic loading. The
experimental study was considering full joint specimens, T- stub and weld
detail specimens [2], [3], [4]. The present research focuses on the numeri-
cal investigation (by FEM) of the T-stub elements, through parametric
studies and low-cycle fatigue interpretations of the results. The para-
metric study also shows the borders for T-stub macro-component failure
mode 2 –> 1 and 2 –> 3 with careful classification of element behaviour.
Results of experimental investigation
Summary of testing program

The main objective of the experimental program was to study the
performance of welded and bolted end-plate beam to column joints
realized from two different steel grades. For this purpose, the experi-
mental program integrated experimental investigation on materials,
welded components, T-stub components, and beam to column joints.
In this way the main sources of ductility were investigated, in local
(material and components) and global (joint tests) manner. Although
the entire research is much larger, this paragraph describes only the
investigations performed on T-stub components, chosen for numerical
study. Previous papers by the same authors already summarized the
results on materials, welded components, weld details and beam-to-
column joints (Dubina et al. 2008 a, b, c). Both monotonic and alter-
nating cyclic tests were performed on T-stub components obtained by
welding S 235 web plates to S 235, S 460 and S 690 end-plates, using
K bevel full-penetration welds. MAG welding was used, with G3Si1

(EN 440) electrodes for S235 to S235 welds, and ER 100S-G / AWS
A5.28 (LNM Moniva) for S235 to S460 and S690 welds. T-stubs were
connected using M20 gr. 8.8 bolts. The EN 1993-1.8 was used to obtain
the design strength of T-stubs and failure modes. From the experimental
program, for the numerical modelling was set a T-stub type A (Table 1).
Design
T-stub type Label Web End-plate failure
mode
TST-12A-S235 S235 t = 12 mm 2

TST-20A-S235 S235 t = 20 mm 2 –> 3
TST-10A-S460 S235 S460 t = 10 mm 2
TST-16A-S460 t = 15 mm S460 t = 16 mm 2 –> 3
TST-8A-S690 S690 t = 8 mm 2
Table 1 — T-stub characteristics TST-12A-S690 S690 t = 12 mm 2 –> 3
T-stub performance on monotonic an cyclic loading

Table 2 shows the measured average values of yield stress fy, tensile
strength fu and elongation at rupture A. It has to be recognized that the
value of elongation for S460 is surprisingly large. Bolts were tested in
tension as well, showing an average ultimate strength of 862.6 N / mm².
Nominal steel grade fy, N / mm² fu, N / mm² A, % Actual steel grade
S235 266 414 38 S235

S460 458 545 25 S460
Table 2 — Material properties S690 831 859 13 S690
Loading was applied in displacement control under tension and force

control under compression. Compressive force was chosen so as to
prevent buckling of the specimen. In general a good ductility was
observed for all specimens. However, thicker end-plate specimens,
even for S235 show a smaller ductility. The conclusions of the experi-
mental study proves the fact that the choice of thickness associated
with steel grade is important in the conception of a proper connection,
for obtaining a good balance between strength, stiffness and ductility
of components. Figure 3 shows as example the experimental results
under the form of force-displacement relationships of T-stub speci-
mens, while the photos in the same figure show two types of failure
modes (mode 2 and 1 respectively). No significant differences in force
values between failure modes of monotonic and cyclic specimens
were recorded. Also, both results agreed with the analytical predic-
tions computed according to EN 1993-1.8. Cyclic loading reduced the
maximum resistance of the T-stub specimens, though the reduction
was not significant. The ductility of the T-stub specimens was quanti-
fied through the ultimate displacement Du. Under monotonic loading,
ultimate displacement was smaller for specimens with thicker end-

plates that failed in modes 2 and 3 involving bolt failure. Cyclic loading
reduced significantly ultimate displacement of specimens with thinner
end-plates that failed in mode 1. This behaviour is attributed to low-
cycle fatigue that generated cracks in the HAZ near the welds, along
yield lines. On the other hand, cyclic loading did not affect much ulti-
mate displacement for specimens with thicker end-plates that failed
in modes 2 and 3, governed by bolt response. It is to be emphasized
that specimens realized from high-strength end plates (S460 and 690,
with lower elongation at rupture), had a ductility comparable with the
one of specimens realized from mild carbon steel (S235). The param-
eters governing the ductility of T-stubs were type of loading (monot-
onic / cyclic) and failure mode (end-plate or bolts) [5], [7], [8].
Numerical analysis
Calibration of T-stub FEM response

The first step in the FEM analyses was the calibration of T-stub responses
presented in Figure 4, using ABAQUS computer code [6]. T-stubs mod-
elling passed the following steps: choose of the FEM, mashing, define of
the material properties. The used FEM was continue solid element (brick
element) C3D8R, with reduced integration. For material, it was used
a stress-displacement real curve. Between all the elements a normal
“hard contact” law was defined, with the surfaces separation possibility.
To calibrate the numerical model developed in ABAQUS, an analysis
was made in order to compare behaviour curves of some T-stubs of the
experimental program with curves obtained by numerical simulation
with real characteristics of steel obtained by tests. The results are pre-
sented in the following figures. Reference numbering should be placed
between brackets and left aligned. The text is to be 7.5 mm indented
from the left margin, as presented in the following example.
Figure 3 — Examples of failure

TST-12A-S690 TST-16A-S460 TST-20A-S235 modes of T-stub specimens
Failure mode 2 Failure mode 2 Failure mode 1 ® 2
Conclusions
The strength and ductility of the bolted end-plate beam-to-column joint

is highly influenced by the T-stub behaviour. The FE model of some pre-
viously tested T-stubs show good agreement response under the form
of Force-Displacement curves, both in monotonic and cyclic response.
Figure 4 — T-stub behaviour

and failure mode according to TST-20A-S235-C1
TST-12A-S690-C1 TST-16A-S460-C1
numerical analysis
Failure mode 2 Failure mode 2 Failure mode 1 –> 2
Monotonic response Cyclic response
Figure 5 — T-stub behaviour

according to numerical analysis
Acknowledgments
This work was partially supported by the strategic grant POSDRU

107 / 1.5 /S /˘77265, inside POSDRU Romania 2007–2013 co-financed
by the European Social Fund — Investing in People
References
[1] EN 1993-1.8. 2003. Design of steel structures. Part 1–8: Design of

joints, European standard.
[2] Dubina, D, Stratan, A, Muntean, N, Grecea, D, “Dual-steel T-stub be-

haviour under monotonic and cyclic loading”, ECCS / AISC Work-shop:
Connections in Steel Structures VI, Chicago, Illinois, USA, 23–55, 2008.
[3] Dubina, D, Stratan, A, Muntean, N, Dinu, F, “Experimental program

for evaluation of Moment Beam-to-Column Joints of High Strength
Steel Components”, ECCS / AISC Work-shop: Connections in Steel
Structures VI, Chicago, Illinois, USA, June 23–55, 2008.
[4] Dubina, D, Muntean, N, Stratan, A, Grecea, D, Zaharia, R, “Testing-

program to evaluate behaviour of dual steel con-nections under
monotonic and cyclic loading”, Proc. of 5th European Conference on
Steel and Composite Structures – Eurosteel 2008, 3–5 September,
Graz, Austria, 609–614, 2008.
[5] Dubina, D, Grecea, D, Stratan, A, Muntean, A., “Performance of

dual-steel connections of high strength components under mono-
tonic and cyclic loading”, STESSA 2009, Behaviour of Steel Structures
in Seismic Areas, Taylor & Francis Group, London, 16–20 Aug. 2009,
Philadelphia, USA, 437–442, 2009.
[6] ABAQUS Inc. “ABAQUS analysis user manual”, version 6.10, 2010.
[7] Dubina, D, “Dual-steel frames for multistory buildings in seis-mic

areas”, Keynote lecture, Proceedings of SDSS’Rio 2010 International
Colloquium Stability and Ductility of Steel Structures, 8–10 Septem-
ber, Rio de Janeiro, Brazil, 59–80, 2010.
[8] Muntean, N, Grecea, D, Dogariu, A, Dubina, D, “Strength and ductility

of bolted T-Stub macro-components under mono-tonic and cyclic loading”,
Proceedings of SDSS’Rio 2010 International Colloquium Stability and
DuctilityofSteelStructures,8–10Sept,RiodeJaneiro,Brazil,223–230,2010.
174 Dynamic safety assessment of a small span high-speed railway bridge
Dynamic safety assessment

of a small span high-speed
railway bridge
ROCHA, João Miguel Abstract

University of Porto, Faculty
of Engineering, Portugal
The main purpose of this work is to evaluate the sensitivity of small
HENRIQUES, António Abel span bridges in high speed railway lines to the resonance phenomena
University of Porto, Faculty bearing in mind the real variability of the parameters that influence
of Engineering, Portugal the bridge dynamic response. As a case study, the railway bridge of
Canelas, located in the Northern line of the Portuguese railway, was
CALÇADA, Rui
University of Porto, Faculty
selected. Simulation techniques, namely the Monte Carlo and Latin
of Engineering, Portugal Hypercube methods, were applied as they allow an accurate consid-
eration of the randomness of structural parameters that influence the
bridge dynamic response. The influence of the train-bridge interac-
tion effects on the dynamic response was also studied. Furthermore, a
safety assessment of the bridge was done based on the results obtained
in the simulations in order to define the train speed limit on the bridge.
Introduction
The dynamic effects caused by high speed trains crossing a bridge is

one of the most relevant facts to take into account during the design
stage and for the selection of the proper structural solution. Exces-
sive vibrations have a higher tendency to occur for speeds higher than
200 km /h as a consequence of resonant effects [1].
During the design stage it is common that several aspects are not
entirely known, leaving the designer to deal with these uncertain-
ties. Unfortunately, these uncertainties may be responsible for unex-
pected dynamic response of railway bridges. In short span bridges
this variability can be due to structural-related parameters as well as
train-related parameters [2]. The current work intended to take these
aspects into account and to analyze how the dynamic response of
bridges is affected by the variability of these parameters. To do so, two
different simulation techniques were applied, Monte Carlo and Latin
Dynamic safety assessment of a small span high-speed railway bridge 175
Hypercube methods, and scenarios with different sample sizes and

different train speeds were analysed. When the dynamic properties of
the train are taken into account, train-bridge interaction needs to be
included in the analysis. The interaction effects take particular impor-
tance when analysing the dynamic response of short span bridges [2, 3].
The European standards allow not taking into consideration the train-
bridge interaction, suggesting, however, that an additional damping is
added to the bridge damping coefficient [1].
As a case study, a bridge with six simply supported spans was selected.
The goal of the present work is to perform a safety assessment of the
bridge and compare the differences in the dynamic response when
different methods, namely the moving loads method and meth-
ods which include train-bridge interaction, are applied, allowing the
assessment of the influence that the train-bridge interaction has in the
dynamics response.
Case Study – Canelas bridge
The case study used in this paper is Canelas Bridge, located in the
Northern line of the Portuguese railway. The bridge has six simply
supported spans of 12 m each, leading to a total length of 72 m. The
bridge deck is a composite structure consisting of two half concrete
slab decks with nine embedded rolled steel profiles HEB 500. This
kind of structural system is called filler beam and is a very common
structural solution for small span bridges in the European high-speed
railway lines, especially in France and Germany [4, 5]. A general view
of the bridge used as case study, as well as the typical cross section of
the bridge deck is shown in Figure 1.
As it can be seen in Figure 1b, due to the existence of a longitudinal

expansion joint between the decks throughout the entire structure,
the bridge has two separate and independent decks, each one sup-
porting a single railway. However, the columns and the abutments
are unique, supporting both decks. Each deck has a total width of
6.2 m, where 4.5 m correspond to the width of the concrete slab with
a height of 0.7 m and 1.7 m is the width of the cantilever with a vari-
able height from 0.3 to 0.5 m. Embedded on the concrete slab with a
spacing of 0.475 m are the referred nine rolled steel profiles HEB 500.
Figure 1 — Canelas Bridge

(a) General view
belongs to Figure 1
(b) Typical cross section
Cement plates were placed underneath the concrete slab, between the
steel profiles, to be used as formwork during the concreting of the slab.
It should also be pointed out that the deck has small ballast retention wall
placed on the top of the slab and right before the beginning of the canti-
lever, corresponding to a beam with 0.6 m of height and 0.3 m of width.
Numerical models
Dynamic model of the bridge

A 2D beam model of the bridge was developed using a FEM program
[6]. The numerical model was defined according to the design drawings.
Along with the deck, the model included the ballast, the rail and the
bearings. The deck and the rail were modelled as beams, positioned at
the corresponding centre of gravity, whereas the bearings were mod-
elled as springs. The connection to the bearings is positioned at the
corresponding centre of rotation. The laminated neoprene elastomeric
bearings are placed under each rolled steel profile HEB 500. Due to
the existence of steel plates between each neoprene layer, each layer
acts as an individual spring so the bearing, as a whole, works as a sys-
tem of series-connected layers. Horizontal springs were used to repro-
duce the frictional behaviour of the ballast. According to [2] the ballast
has a linear elastic range until a relative horizontal displacement, u0,
of 0.002 m is reached, followed by a plastic phase. In this paper a
load, k, of 20 kN / m, corresponding to an unloaded track, was con-
sidered. Since the relative displacements are very small (smaller than
the u0 limit), the ballast stiffness was computed in accordance with
the elastic range. The connection between the track elements and the
deck elements, and the connection between the deck elements and
the bearings was accomplished by means of rigid beams. A schematic
view of the bridge model used can be seen in Figure 2. Despite the
(a) General scheme
Figure 2 — Bridge model

(b) Detail A
structural system of the bridge consisting of simply supported beams,

the rail is continuous and this continuity affects the dynamic response
of the bridge. This was included in the numerical model by extending
the rail 10.5 m in both directions over the length of the bridge.
Dynamic model of the train

In this paper two models were used to reproduce the train, corre-
sponding to two different applied methods. For the moving loads
method, the simplest used method, the train consists in a series of
concentrated loads, representing each axle of the train. For this model,
only the value of the load per axle is required. Prior to the analysis,
the time history of the loads must be defined for each axle. For each
time step the axle loads are converted into equivalent nodal forces.
The other used method took into consideration the train-bridge inter-
action. For this analysis a dynamic model of the train is required. The
axle positioning as well as the load per axle are the same in both mod-
els. The car bodies and bogies are simulated by rigid bodies with mass,
Mc and Mb, and rotational inertia, Ic and Ib, respectively. The primary
and secondary suspensions are simulated by a spring-damper set with
stiffness, Kp and Ks, and damping coefficient, cp and cs, respectively. The
wheelset is simulated by a concentrated mass, Me, whereas the wheel-
rail contact stiffness is simulated by a spring with stiffness Kh. It should
be mentioned that the dynamic properties of the train were defined
according to [2]. A scheme of the model used can be observed in Fig. 3.
Random variables
In an early stage of this work several variables that might have non-
deterministic properties and whose variation might lead to a relevant
variability on the structural response were selected. This required the
definition of variables related to the structure, the train, the track and
also the wheel-rail contact.
The random variables used in the current paper can be divided into
two separate groups. One group related with the structural and track
variables and the other related with the train variables. The bridge-
related and track-related variables selected, as well as their corre-
spondent distribution and variability [7, 8], are presented in Table 1.
The train-related variables are presented in Table 2. The variability of

these parameters was defined according to information provided by
the manufacturers. A uniform distribution was assumed for all the
parameters. It should also be mentioned that the variation of the
car body mass was defined according to the occupancy of the train.
The values of load per axle referred in [2] were assumed for a train
operating at full capacity.
(a) Lateral view
(b) Dynamic model
Figure 3 — Train model

(c) Beam model
Mean
Distri- (Gaussian) Standard deviation
Variable [simulation] bution or Min. (Gaussian) or
(Uniform) Max. (Uniform)
Concrete density weight (1, 2) Gaussian 2,5 t / m³ 0.1 (CV = 4%)
Ballast denisty weight (3, 4) Uniform 17 kN / m³ 21 kN / m³
Ballast area (5, 6) Uniform 1.48659 m² 2.76081 m²
HEB 500 area (7, 8) Gaussian nominal area 0.04 x nominal area
Elasticity modulus concrete (9, 10) Gaussian 36.1 GPa 2.888 (CV = 8 %)
Concrete weight
Uniform minimum area maximum area
(geometrical variation) (11, 12)
Concrete height (13, 14) Gaussian nominal value 10 mm
Concrete width (18, 19) Gaussian nominal value 5 mm
Shear modulus neoprene (16, 17) Uniform 0.75 MPa 1.18 MPa
Table 1 — Bridge random vari-
ables, distribution functions Elasticity modulus neoprene
Uniform 420 MPa 600 MPa
and variability (15, 20)
Variable Variation (%)
Primary suspension stiffness ± 10

Primary suspension damping ± 15
Secondary suspension stiffness ± 10
Secondary suspension damping ± 15
Carbody mass ± 15 Table 2 — Train-related vari-
Wheel-rail contact stiffness ± 15 ables and variability
Dynamic properties of the numerical models

In this section the properties of the dynamic models developed, both
the bridge and the train, are presented.
In both cases the obtained results correspond to an analysis where

the average values were assigned to all the random variables previ-
ously defined.
In Figure 4 the first two bending modes of the bridge are illustrated.
It can be seen that the first natural frequency of the bridge has a
particularly high value, but is within the typical range of values for this
type of structure.
In Fig 5 the dynamic properties of the train are presented with the illus-
tration of the global modes of the train. It can be observed that the val-
ues obtained in the modal analysis are in agreement with the literature.
Variable screening procedure
Due to the large number of variables used in the analysis, a variable

screening procedure was performed. The goal of this procedure is
to assess the importance of each variable on the dynamic response,
allowing to check if any of the used variables can be considered as
(a)
Figure 4 — Schematic view

(b) of the bending modes of
the bridge
(a)
Figure 5 — Train modal analy-

sis (only one half of the train
is presented)
(b)
deterministic in the following analysis. The variable screening proce-

dure was also divided into two separate analysis. The method used for
the screening procedure is similar to one presented in [9].
Procedure
The first stage of the procedure corresponds to the selection of the
random variables that allow the definition of the problem. Afterwards, a
structural analysis is performed adopting the mean values for all the
variables. At next step, the bridge dynamic response is computed keep-
ing all the variables with their mean values except for the “tested” var-
iable which is changed from its mean value (about two times the standard
deviation). This step is repeated for all variables allowing the evaluation of
the sensitivity coefficients and the importance indicators for each variable.
The sensitivity coefficients and the importance indicators are obtained

by comparing the difference between the reference results (corre-
spondent to the analysis where all the variables have their mean val-
ues) and the results obtained for each “tested” variable, as expressed
on Equations 1 and 2:
yk ym yk ym
(1) bk
xik xim h xk xim
(2) b k bk CV
where yk is the difference between the structural response reference

results and the results obtained for each “tested” variable, xik is the
difference between the mean value of the “tested” variable and the
value used in the analysis and ym and xim are the structural response
reference value and the “tested” variable mean value, respectively.
After analysing these coefficients for all variables, the maximum impor-
tance indicator is determined and the relative importance for each
variable is established by comparing to the maximum value obtained.
Variables which relevance is smaller than a pre-established value
(10%) are considered irrelevant for the bridge dynamic response and
in further analysis are considered as deterministic. In these cases the
variables values used on the analyses will be their mean value.
Figure 6 — Results of the

sensitivity analysis for the
accelerations
Results
Bridge-related and track-related variables

Some of the results obtained from the variable screening procedure for
the bridge-related and track-related variables can be seen on Figure 6.
It should also be pointed out that in this work four distinct response
parameters were analysed: natural frequencies, displacements, accel-
erations and reactions.
The variables with higher influence on the dynamic response were the
inertia variation due to the variation of the section height, the ballast
area, the elasticity modulus of the concrete and the stiffness of the bridge
bearings. On the other hand, the geometrical variation of the rolled
steel profiles and the shear modulus of the neoprene had a very small
influence on the response and could be taken as deterministic variables.
Train-related variables
In order to assess the influence of the train-related variables on the
dynamic response, the acceleration on the bridge and inside the
coaches was analysed. Consequently, the influence of the train-related
variables can be split into two different analysis: one concerning the
bridge response and the other related with the train response. How-
ever, this work is focused on the variations of the bridge response. In
Figure 7 some of the obtained results are presented.
It could be observed that the bridge response is practically unaffected

by the changes on the train properties. The only exception occurs
when the carbody mass and the wheelset mass are changed. All the
other parameters can be kept as deterministic on dynamic analysis
focused on the bridge dynamic response.
Simulations
Initially, the structure and track variables are analysed, keeping the
train properties constant. On this stage the dynamic response is ana-
lysed by the moving loads method.
Figure 7 — Influence of the

carbody mass on the dynamic
response
Two different simulation methods were used: the Monte Carlo method,
which is the basis of most of the simulation methods, and the Latin
Hypercube method, which is a variation of the previous method. The
bridge dynamic behaviour was analysed for the crossing of a train with
the same characteristics of the type B train presented in [1] with a vari-
able speed ranging from 200 km / h to 450 km / h in steps of 5 km / h.
For the Monte Carlo method four different sample size simulation sce-
narios were analyzed: 5,000, 2,500, 2,000 and 1,000 simulations.
For the Latin Hypercube two different scenarios were analyzed with
500 and 250 simulations, respectively.
Results
In this section some of the obtained results will be presented. In
Fig. 8 histograms of the acceleration values for different train speeds
obtained for the 5,000 Monte Carlo simulation are shown.
From the analysis of the histograms it can be seen that the structural
response does not follow a Gaussian distribution. In Figure 8b a skew-
ness in the distribution can be observed and in Figure 8a shows a com-
bination of distributions. This is observed for several speeds, namely
the speed range from 285 km / h to 305 km / h, and can be explained
by the existence of two distinct structural responses (resonant and
non-resonant response).Figure 9 shows the evolution of the mean val-
ues and the maximum values for the simulation scenarios analyzed
when the speed of the train increases.
The obtained results in terms of mean values and standard deviation

values are consistent for the different applied simulation methods and
for the different sample sizes analyzed. However, when analysing the
maximum response values obtained for the different simulation sce-
narios a higher scatter on the results can be observed. This turns out
to be a key issue for the analysis since these are the values that have a
higher importance in structural reliability problems.
Train-bridge interaction
Prior to the safety assessment, the influence on the train-bridge inter-

action effects on the dynamic response was analysed. A deterministic
analysis was performed assigning the average value to all the random
variables. A comparison between the results obtained from the mov-
ing loads method and the interaction method is shown in Figure 10.
From the observation of Figure 10 some significant differences can

be noticed. The differences occur particularly in the resonant peaks,
where a reduction in the acceleration values (~ 10%) is registered, con-
firming what was observed in the time history response. Non-resonant
speeds do not show significant differences between both methods.
a) v = 285 km / h
Figure 8 — Histograms for dif-

ferent train speeds
b) v = 320 km / h
Figure 9 — Histograms for dif-

ferent train speeds
Figure 10 — Maximum accel-

eration comparison for differ-
ent methods
Safety assessment
At this stage the safety of the bridge was only assessed taking the
bridge and track variables into account. This analysis was based on the
acceleration values registered at mid span of the bridge which turned
out to be the most restrictive aspect of the response. The acceleration
limit was considered to be 7 m / s² which is the value that some experi-
mental tests confirmed to be the limit to the beginning of the ballast
instability [10]. In the European standards, however, this limit is lower
and the maximum acceleration allowed for ballasted track is 3,5 m / s²
which comes from the use of a safety factor equal to 2 [1].
A curve was fitted to the upper extremity of the cumulative prob-

ability function in order to perform a safety assessment of the bridge.
For this safety assessment several types of regression functions were
used, namely sigmoidal and exponential functions. Taking into account
the type of studied problem probability values up to 10–4 are consid-
ered to be acceptable to measure admissible failure due to the instabil-
ity of the ballast layer. Analysing Figure 11 it can be seen that this value
is not exceeded for speeds lower than 290 km / h, making this the
train speed limit to be imposed on the bridge. It should also be pointed
out that in the cases where less than 2,000 simulations were used, the
obtained probabilities of failure are not accurate. This comes as a result
of the misrepresentation of the interest zone in these cases due to the
small number of points on the upper extremity of the structural response
distribution, which had already been mentioned before, leading to an
inadequate fit of the data.
Conclusions
This work intended to assess the influence that the variability of the
train-track-bridge parameters have on the dynamic response. The
Canelas railway bridge was selected as the case study and the iden-
tification of the parameters with higher importance on the bridge
dynamic behaviour was done through a sensitivity analysis. This sen-
sitivity analysis showed that the inertia, the area of the ballast layer,
Figure 11 — Safety assessment

for different sample size
the concrete elasticity modulus, the stiffness of the bridge bearings,

the carbody mass and the wheelset mass are the parameters that most
affect the bridge response.
By comparing the results from the two applied simulation methods it

can be confirmed that the Latin Hypercube method is more efficient
than the Monte Carlo method. However, the analysis of the maximum
response values obtained for the two simulation methods shows a sig-
nificant difference between them. Since these are the values that have
a higher importance in structural reliability problems this turns out to
be a relevant aspect and does not allow the applicability of the Latin
Hypercube method to this problem due to the inadequate representa-
tion of the interest area when this method is applied.
The train-bridge interaction effects were analysed for a deterministic

scenario where all the bridge variables were assigned their mean values.
The comparison between the obtained results from the moving loads
method and the interaction method showed important differences.
These differences were, however, limited to the speeds that induced
resonant effects on the bridge. The safety assessment of the bridge was
based on the acceleration values registered at mid span of the bridge
which were the most restrictive aspect of the dynamic response. The
safety assessment concluded that the train speed limit on the bridge is
290 km / h in order to guarantee the running safety of the trains.
Acknowledgments
The present work has been funded by the Portuguese Foundation for
Science and Technology (FCT), in the context of the Research Project
FCOMP-01-0124-FEDER-007195 — Advanced methodologies for the
assessment of the dynamic behaviour of high speed railway bridges
(Ref. FCT PTDC/ECM/69697/2006). The authors also wish to acknowl-
edge the information provided by REFER about Canelas Bridge.
References
[1] Commité European de Normalisation (CEN) (2003) EN 1991 2 —

Eurocode 1 — Actions on Structures — Part 2 — Traffic loads on Bridges.
Brussels.CEN.
[2] European Rail Research Institute (ERRI) (1999) ERRI D-214 / RP 9.

Rail bridges for speeds > 200 km / h. Final report. Utrecht.
[3] Museros, P., Romero, M.L., Poy, A. & Alarcón, E. (2002) Advances
in the analysis of short apsn railway bridges for high-speed lines.
Computers and Structures 80: 2121–2132.
[4] Hoorpah, W. (2005) Dynamics calculations of high-speed railway

bridges in France — some case studies. Dynamic of High-Speed
Railway Bridges: 167–183. University of Porto — Faculty of Engi-
neering: Porto.
[5] Martínez-Rodrigo, M., Lavabo, J. & Museros, P. (2010) Dynamic

performance of existing high-speed railway bridges under resonant
conditions retrofitted with fluid viscous dampers. Journal of Engi-
neering Structures 32: 808–828.
[6] Azevedo, A. (2012) — “FEMIX 4.0 — Finite Element Analysis”,

http://www.alvaroazevedo.com/femix/.
[7] Joint Committee on Structural Safety (JCSS) (2001) Probabilistic

Model Code — 12th draft. http://www.jcss.ethz.ch/.
[8] Wiśniewski, D.F. (2007) Safety formats for the assessment of con-
crete bridges — with special focus on precast concrete. Ph.D. Thesis.
University of Minho. Guimarães.
[9] Henriques, A.A. (1998) Use of new safety concepts in the design of
concrete structures (in Portuguese). Ph.D. Thesis. University of Porto —
Faculty of Engineering. Porto.
[10] Zacher, M. & Baeßler, M. (2005) Dynamic behaviour of ballast

on railway bridges. Dynamic of High-Speed Railway Bridges:
125-142. Porto.
Comparative study of steel frame modelling levels and Eurocode based design methods 187
Comparative study of steel

frame modelling levels and
Eurocode based design methods
Abstract TÓTH, Adrienn

BME, Department of Struc-
tural Engineering, Hungary
In this paper a numerical study is presented which examines a steel
frame with two different finite element programs. Stability failure is JOÓ, Attila
more frequent in a lot of cases than the strength failure hence it is BME, Department of Struc-
important to focus on these failure modes: global, in-plane-, out-of tural Engineering, Hungary
plane-, lateral-torsional- and local buckling.
Three models were used with different elements such as shell ele-
ments and 7 DOF beam elements. 7 DOF beam elements were used
in the first model, shell elements were used in the other two. The first
of the shell models gave too much local buckling shapes therefore it
was improved with local constraints and that is the third model where
global buckling shapes can be examined.
There are three different procedures to calculate the resistance: (i) the
general method, (ii) the method of the reduction factors and (iii) the
simulation. The analysis results of the different programs and design
methods were compared to each other and to the manual calculation
based on the Eurocode 3 standards.
Introduction
The idea of the investigation of a frame came from a home work of

an MSc class at BME Department of Structural Engineering. The con-
nection of the column to the ground was hinged. The purlins and wall
girts are not included in the finite element models, its effect is consid-
ered by lateral supports (see Fig. 1.).
By the design of steel frames it is very important to examine the stabil-

ity failure modes. There are different types: the global, in-plane-, out-
188 Comparative study of steel frame modelling levels and Eurocode based design methods
of plane-, lateraltorsional- and local buckling. All of them have to be

considered in the investigation of the frame.
The resistance of the whole frame can be calculated with the general
method according to Eurocode 3 [2] standards. It is necessary to cre-
ate a finite element model to obtain the collapse load factor related
and the first buckling shape together with the critical load factor.
The resistance of the frame members can be also calculated according

to the Eurocode 3, it is named manual calculation method. This is an
approximation which is necessary because of the various height of the
beam at the frame corner. The frame was divided into four members
and reduction factors were used to calculate the resistances. These
factors depend on the type of buckling. They were calculated to each
member and to each buckling mode.
The resistance of the frame members can be obtained with the reduc-
tion factors and the help of the buckling shape of each member. The
critical load factor of the first buckling shape related to the given
member used for these calculations. The procedure is the same as
the manual calculation method apart from the calculation of the criti-
cal force. The critical forces were determined by the finite element
programs. The shell model provided only local buckling shapes which
cannot be used to define the reduction factors. So this model was
developed, the cross sections were constrained hence the local buck-
ling cannot appear.
With the help of advanced finite element analysis we can have the
resistance of the whole frame by numerical simulation. It has to be
applied equivalent geometrical imperfections according to the buck-
ling shapes of the members. On the imperfect finite element model
the structural behaviour is simulated by Geometrical and Material
Nonlinear Imperfect Analysis. [1]
The resistance and the utilization of the members were derived from
each design method.
Models
7 DOF beam model (#1)

For the 7 DOF beam model the ConSteel [7] program was used. Each
element with the 7 degrees-of-freedom can follow the behaviour of
the lateral-torsional buckling, but the beam element is cannot provide
local buckling which is considered as effective cross-section.
The internal flange bracings of the column and the beam were given as
a lateral support on the internal flange of the members by the help of
eccentric support as illustrated in Fig. 1.
C1 B1
B2
C2 Figure 1 — 7 DOF beam model

(ConSteel)
Shell model (#2)

Ansys [6] was used for the shell model. The applied shell element was
SHELL181, what is a 4-node-element with six degrees of freedom at
each node. The internal flange bracings of members were given as fix
lateral support on one node of the internal flange. The purlins and wall
girts are modelled as support on the outer flange.
The mesh was generated by free meshing method of Ansys with the
maximum element length of 50 mm to be able to model local buckling
at the frame corner. The full model and the details are illustrated in Fig. 2.
(a) (b)
Figure 2 — a) Shell model

(Ansys), b) Details of the
shell model
The results of the linear instability analysis were local buckling shapes
and this is a big disadvantage, because the global buckling shapes can-
not be founded. To solve this problem the next model is developed.
Locally constrained shell model (#3)

The model #2 was modified to avoid the local buckling shapes. The
column and the beam were divided into thin slices with the length
of 50 mm which are perpendicular to the axle. This model will be
the third one.
The nodes of the former mentioned divided sections are linked in the
centre of gravity of each cross section and the section deformations
are constrained. Because of these constraints the local buckling cannot
come up. The constraints are displayed in Fig. 3. by pink colour.
(a) (b)
Figure 3 — a) Locally con-

strained shell model (Ansys),
b) Constraints
By this modification of the model the critical forces of global buckling

modes can be found to compare to the manual calculation and to the
beam model. And later these modes can be used as the shapes of geo-
metrical imperfections for the simulation.
Loading
Only one load step was applied which contains almost all loads accord-
ing to the EC. The self-weight was applied as a steel density material
property and gravity to the structure. Besides it the weight of the roof
as permanent load and snow load as meteorology load was applied.
Each of these was multiplied with the appropriate partial factors.
Analysis levels
Three different analysis levels were investigated. The first is the linear
analysis to check the models. Chapter 3 shows more details. The next
analysis is the linear instability analysis in Chapter 4. In Chapter 5 the
nonlinear analysis is summarized which was necessary to perform a
numerical simulation based design.
Linear analysis
The deflection of the centre point of the beam and the vertical reac-
tions at the bottom of the columns were compared on the three mod-
els. There is a bit difference between the two different programs in the
reaction force and in the deflections as well. The difference is about
5% in each case, which was accepted for further studies.
Model #1 Model #2 Model #3
Table 1 — Results of the lin- Deflection (mm) 40,60 38,44 38,38

ear analysis Summa reactions (kN) 359,8 381,1 381,1
Linear instability analysis
Buckling shapes for manual calculation with reduction factors

The aim of searching the buckling shapes is to apply the critical load
factors for manual calculation in case of model #1 and model #3. The
column and the beam were divided into two parts so four members
were investigated separately as former shown in Figure 1.
–– C1: the part of the column, above the internal flange bracing
–– C2: the part of the column, below the internal flange bracing
–– B1: the part of the beam, right from the internal flange bracing
(permanent cross-section)
–– B2: the part of the beam, left from the internal flange bracing
(variable height cross-section)
Different buckling shapes were searched. There were shapes which

included more members and in a few cases and there were ones which
were belonged to only one member. The in-plane, out-of plane and
the lateral-torsional buckling were considered.
The critical internal forces (Mcr, Ncr) were calculated by the help of
critical load factor related to the concerning buckling shape.
Buckling shapes for imperfections

For the nonlinear analysis it was necessary to know certain buckling
shapes to apply them as geometrical imperfections according to the
Eurocode 3 standards.
The shape of imperfections correspond to the shape of the related

buckling modes of the model #3, and the size was chosen to reach the
maximum amplitude defined by the Eurocode.
Model #1
The buckling shapes of the first model (the 7 DOF beam model) can
be seen in Figure 4–5.
Model #3
Figure 6–8. illustrate the buckling shapes of the model #3 (the locally
constrained shell model).T
he out-of-plane buckling shapes differ significantly from model #1. It is

difficult to choose which shape should be compare to a beam model
because each member has different shapes where the out-of-plane
Figure 4 — (a) In plane B1, B2,

(b) In plane C1, C2, (c) Out of
plane C1, C2, (d) Lateral-tor-
sional buckling C1, C2, B1, B2
(a) (b) (c) (d)
(a)
Figure 5 — (a) Out-of-plane

(b) B1, (b) Out-of-plane B2
Figure 6 — a) In plane B1, B2,

b) In plane C1, C2
(a) (b) (c) (d)
deformation is the maximum. This can be a usual problem for practic-

ing engineers as well if they use shell model. The critical load factors
are compared between the two models, and later the utilization of the
members as well to see the effect on design resistances.
Results of the linear instability analysis

The results of the critical load factors were summarized in Table 2. The
values of in plane buckling were similar. The other types of buckling
gave different results. The biggest difference was 129.97%. In case of
out-of-plane buckling the shell model resulted higher values by each
member than the 7 DOF beam model and were inversely in the case of
lateral-torsional buckling. The results at the lateral-torsional buckling
are the double by the beam model, because the internal nodes of the
frame corner are stiffer in that case. The out-ofplane-buckling gave
also big differences because these shapes were not clear modes and
difference explained above.
Nonlinear analysis
Simulation
The Eurocode standards [3] allow perform calculation with advanced
finite element analysis. The effect of non-perfect geometry is consid-
ered by the “equivalent geometry imperfection”. The shape and the
magnitude of these can be applied according to EC. In this case the
assumed shapes of imperfections were derived from the elastic buck-
ling mode of the whole frame and the members.
(a)

Figure 7 — (a) Out-of-plane B1,
(b)
(b) Out-of-plane B2
Figure 8 — (a) Lateral-torsional

buckling B, (b) Lateral-torsional
buckling C
(a) (b)
In In Out- Out- Out- Out- Lateral- Lateral-

plane: plane: of- of- of- of- torsional torsional
B1, B2 C1, C2 plane: plane: plane: plane: buckling: buckling:
B1 B2 C1 C2 B1, B2 C1, C2
Model #1 35.32 166.52 57.26 52.66 82.93 82.93 12.51 12.51

Model #3 38.07 164.11 109.55 92.11 177.06 190.71 5.68 7.55
Table 2 — Critical load factors Diff. [%] 7.79 1.47 91.32 74.91 113.51 129.97 120.25 65.70
The result of this analysis is a force-displacement diagram, which

shows the resistance of the frame. The material model was a bi-lin-
ear material model with nominal material yield stress 235 N / mm².
The double of the design load was applied to the FE model for the
simulation.The largest load step was the 1 / 10 of the design load; the
minimum load step was 1 / 100 of the design load. If the analysis can-
not find solution of the GMNI analysis by the smallest load step, the
calculations stopped.
Results of the nonlinear analysis

The failure is shown in Fig. 9. This result came from the last converged
load step. The red areas mean that the stress of the plate reaches the
nominal yield point.
Figure 9 — Result of the simu-

lation (detail: frame corner)
Evaluation of design resistance
The following calculations were compared:
–– general method with model #1

–– general method with model #3
–– method of the reduction factors with model #1
–– method of the reduction factors with model #3
–– method of the reduction factors with manual calculation
Table 3–6 show that the discrepancies of reduction factors between

the two finite element models are not so big as the discrepancies of
the critical load factors. The maximum difference is between ± 10 %.
General General
Model #1 Model #3 Manual method: method:
Model #1 Model #3
y 0.947 0.947 0.998 – –

z 0.849 0.932 0.943 – –
LT 1.000 0.934 0.925 – –
Utilization 0.586 0.618 0.620 0.633 0.638
% 94.8 100.0 100.3 102.4 103.2 Table 3 — C1 column
General General
Model #1 Model #3
y 0.949 0.948 0.998 – –

z 0.852 0.940 0.943 – –
LT 0.929 0.818 0.816 – –
Utilization 0.395 0.434 0.433 0.416 0.481
Table 4 — C2 column % 91.0 100.0 99.8 95.9 110.8
General General
Model #1 Model #3
y 0.669 0.690 0.699 – –

z 0.726 0.841 0.849 – –
LT 0.974 0.806 0.834 – –
Utilization 0.463 0.547 0.529 0.477 0.574
Table 5 — B1 beam % 84.6 100.0 96.7 87.2 104.9
General General
Model #1 Model #3
y 0.584 0.608 0.843 – –

z 0.640 0.766 0.847 – –
LT 1.000 0.859 0.796 – –
Utilization 0.538 0.626 0.639 0.561 0.617
Table 6 — B2 beam % 85.9 100.0 102.1 89.6 98.6
The simulation provides a force-displacement diagram (Fig. 10). The

displacement is the deflection of middle node of the beam. The end of
the curve shows the resistance of the frame. The horizontal line in the
diagram is the 100% resistance according to the manual calculation.
The partial factors of model uncertainty and scatter of loading and
resistance are chosen to 1.0 [1]. The result of the simulation shows that
it satisfies the requirements of EC. The resistance is 3% higher than the
load level correspond to 100% utilization.
Figure 10 — Force-displace-
ment diagram
Summary and conclusion
As it can be seen in Chapter 6 the magnitude of utilization is almost

the same in model #3 and in the manual calculation. The values of
model #1 are smaller in each case if only the method of the reduction
factors is considered. We cannot declare that general method gives
higher utilization value.
The non-linear analysis gave similar result to the manual calculation. As

it was shown we can use also the GMNI analysis to design a frame but
there were some uncertainties first of all at the out-of-plane buckling
shapes. The linear instability analysis did not give clear shapes there-
fore the giving of the imperfections was also uncertain. The resistance
is sensitive to the imperfection amplitudes. It was difficult to find the
suitable imperfection shapes, because a small difference can create
bigger discrepancy.
References
[1] Joó, A, Dunai, L (2011). EUROSTEEL 2011 6th European Conference

on Steel and Composite Structures (Volume C): Finite element
simulation based design of steel frames
[2] EN 1993-1-1 (2005); Eurocode 3: Design of steel structures — Part 1–1:

General rules and rules for buildings

Plated structural elements

Strength and stability of shell structures
[5] EN 1990 (2000): Basis of structural design
[6] ANSYS Finite Element Program; Documentation, Ansys Inc. (2003)
[7] ConSteel Finite Element Program; Documentation, (2012)

Assessment of structure-tuned mass damper assemblies in an earthquake environment 197
Assessment of structure-tuned
mass damper assemblies in an
earthquake environment
Abstract TRIBUTSCH, Alexander

University of Innsbruck, Unit
of Applied Mechanics, Austria
In this paper the seismic performance of tuned mass dampers (TMDs)
attached to multi-degree-of-freedom (MDOF) structures is assessed. ADAM, Christoph
In previous publications of the authors a wide range of single-degree- University of Innsbruck, Unit
of-freedom (SDOF) systems was investigated, in particular focussing of Applied Mechanics, Austria
on the effect of detuned TMD-parameters. Design relations were
derived to provide the engineering practice with response reduction
coefficients for structures equipped with TMDs compared to struc-
tures without these devices.
Here, the consistency of these design equations is evaluated when

applied to generic MDOF structures, represented by generic shear
frames. If the effective TMD-structure mass ratio is kept constant, the
seismic response of MDOF systems, which are dominated by the first
vibration mode, and the TMD performance is almost independent
from the position of the TMD. Nevertheless, a non-optimal attach-
ment position results in an increase of the actual TMD mass. Addi-
tional results show that the application of a further TMD tuned to the
second structural natural frequency hardly improves or even impairs
the overall seismic response.
Finally, the paper presents a practical application of recently proposed

“modified” response spectrum method considering the effect of a
TMD on the seismic response. Planar frame structures with a varying
number of stories, and both uniformly and non-uniformly distributed
stiffness are considered. Estimated values provided by the proposed
methodology are in good agreement with exact outcomes. It can be
concluded that this methodology yields in an early design phase rea-
sonable estimates of both the TMD space requirements and structural
response quantities without performing costly time history analyses.
198 Assessment of structure-tuned mass damper assemblies in an earthquake environment
Introduction
Recent developments in material science, construction methods, and

numerical simulations allow for the construction of tall and / or slender
structures, which consequently may be prone to structural vibrations.
Application of tuned mass dampers (TMDs) is one of the measures to
reduce these structural vibrations. In the last decades various optimi-
zation strategies for optimal applications of TMDs have been proposed
(see e.g. [1], [2], [3]). The positive effect of TMDs to mitigate periodic
small-band induced vibrations (such as wind) is generally accepted,
however still controversially discussed for seismic induced vibrations.
Nevertheless, when located in an earthquake environment the seismic
performance of a TMD (e.g. designed to protect the structure against
wind) must be assessed, although mitigation of earthquake induced
vibrations is not the primary objective of this TMD. This may result in
larger space requirements or measures to limit the maximum TMD stroke.
Until recently, the seismic performance of TMDs could not be assessed

reliably without performing computationally expensive time history
analyses. The authors addressed the effect of TMDs on earthquake
excited single-degree-of freedom (SDOF) structures in previous stud-
ies [4], [5]. Based on recorded ground motion sets, numerous time
history analyses, statistical response evaluation, and subsequent
multi-regression analyses analytical reduction coefficients have been
derived to quantify the effect of a TMD. These coefficients allow for
the estimation of important key parameters in an early design phase,
such as the peak displacement reduction of the main structure and the
TMD stroke (relative peak displacement of the TMD with respect to
the attachment location of the main structure). Furthermore, an effi-
cient methodology for the practical engineer is proposed to assess the
seismic response reduction and TMD stroke of regular planar multi-
degree-of-freedom (MDOF) high-rise structures [6].
Seismic response coefficients
Fundamentals
Response coefficients are appropriate to quantify the benefits of a TMD
[7]. They are defined as the ratio of a response quantity from a structure
with attached TMD to the corresponding quantity of the same structure
without TMD. To account for the record-to-record variability the assess-
ment is not based on the outcomes induced by single earthquake events,
but on statistical response quantities such as quantiles and median values
considering the ground motions of a complete set. The peak displace-
ment reduction of the main structure is expressed by the coefficient.
i
max x
(1) P i with TMD
i
max x
without TMD
In Eq. (1) x(i) is the characteristic displacement of the considered struc-

ture with respect to the base (with and without TMD) subjected to a
single earthquake record denoted by i. Corresponding statistical quan-
tities are denoted as P (med), P (16Q) and P (84Q).
The limited deflection (stroke) of a mechanical TMD is considered by

the response coefficient
max x TMD , i i
x
with TMD
Di (2)
i
max x
without TMD
which is defined as the maximum relative displacement of the TMD

with respect to the attachment location, scaled by the maximum
characteristic displacement of the stand-alone SDOF system. D(med),
D(16Q) and D(84Q) are the corresponding statistical response quantities.
Seismic response coefficients for SDOF structures with attached TMD

have been derived in previous studies of the authors [4], [5]. Three
sets of real recorded ground motions from FEMA P-695 (“ATC63 pro-
ject”) [7] and [9] were utilized as seismic input. The first set, denoted
as ATC63-FF set, contains 44 far-field records with a fault to rupture
distance larger than 10 km, while the second set (ATC63-NFwP set)
includes 28 near-field records with a strong pulse characteristic. All
earthquake events show magnitudes ≥ 6.5 and have been recorded
on NEHRP Site Class B (rock), C (soft rock / very dense soil), or D (stiff
soil). A third set, denoted as LMSR-N, comprises 40 far-field records
of moment magnitude between 6.5 and 7 and closest distance to the
fault rupture between 13 km and 40 km on NEHRP site class D.
Analytically derived optimal TMD parameters for stationary white noise

base excitation [2] are sufficiently accurate to provide the optimal per-
formance of TMDs subjected to earthquake induced vibrations, as shown
in [4], [5]. Accordingly, the optimal natural frequency ratio (= TMD /S)
and the optimal viscous damping coefficient TMD of the TMD is [2]:
1 2 1 4
opt , TMD , opt (3)
1 4 1 1 2
Here, (= mTMD /mS) denotes the mass ratio between TMD mass

mTMD and structural mass mS. S and TMD is the natural circular
frequency of the stand-alone SDOF structure and of the stand-alone
TMD, respectively.
Exemplarily, in Figure 1 the median peak displacement reduction

coefficient P(med) and the median TMD stroke coefficient D(med) are
depicted for structural periods TS between 0.05 s to 5.0 s, and mass
ratios  between 0.02 to 0.08.
To provide the evaluated results for the engineer in practice, analytical

expressions of seismic response (reduction) coefficients were derived
by multivariate regression analyses. They are listed in [10]. These ana-
lytical expressions are valid for optimally tuned TMD parameters, but
also detuning is considered [5]. The range of admissible structural
parameters, which enter the analytical expressions, is listed in Table 1.
Figure 1 — (a) Median reduc-

tion coefficient P(med): struc-
tural peak displacement.
(b) Median response coeffi-
cient D(med): TMD stroke. Op-
timal TMD parameters, struc-
tural damping S = 0.005,
ATC63-FF earthquake set.
Assessment of multi-degree-of-freedom structures
Considered structural models

A shear frame model is utilized for the investigation of multi-degree-
of-freedom (MDOF) structures, as depicted in Figure 2. Mass and
stiffness is distributed uniformly along the building height. Stiffness
coefficient knj denotes the stiffness in the jth floor of a building with
a number of n storeys. Here, stiffness coefficient is independent from
the floor and tuned to fit a given first period of the main structure.
Damping coefficients cnj are determined by the assumption of Rayleigh
damping based on a selected damping ratio for the first and second
structural mode.
Effective mass ratio

For the jth mode of the n-story stand-alone structure, the modal mass
ratio j is defined [11] as:
Parameter Variable Min. Max.
Period of the stand-alone SDOF structure TS 0.20s 5.00 s

Mass ratio (= mTMD / mS) 0.005 0.08
Natural frequency ratio (= ωTMD / ωS) 0.86 1.04
Damping coeff. of the decoupled TMD TMD 0.04 0.24
Damping coeff. of stand-alone structure S 0.005 0.05
Table 1 — Range of structural
parameters for analytical Ground motion sets ATC63-FF, ATC63-NFwP,
response coefficients LMSR-N
Figure 2 — MDOF shear frame

structure with attached TMD
at the roof.
2
jr mTMD
j T
, j 1, ,n (4)
j M j
Here, M is the diagonal mass matrix (size n x n), and j denotes the
jth mode shape of the structure without TMD. jr is the rth element
(denoting the rth storey, where the TMD is attached) of j.
According to Eq. (4) for a given TMD mass mass ratio μj increases
with increasing jr. In other words, in order to keep the fundamental
mass ratio 1 constant, the TMD mass mTMD must to be increased
when the TMD is not located at its optimal position on top of the
building. This effect is shown in Figure 3, where the seismic response
(ATC-63FF record #1) of a five-storey building with inherent damping
of 0.5% and a structural period of 1.0 s is plotted. It can be seen that
the structural response of the third and the fifth floor is considerably
reduced by the TMD, compare the red and blue dotted lines (stand-
alone frame) with red and blue solid lines (coupled frame-TMD con-
figuration). Note that in this example the TMD parameters are tuned
optimally to the fundamental structural mode. Furthermore, it is read-
ily observed that the structural response of structure-TMD assemblies
with constant fundamental TMD mass ratios of 5%, but different TMD
positions is almost identical, compare the solid lines (TMD at the 5th
floor) and the dashed lines (TMD at the 3rd floor). The actual TMD
mass is about 71% larger for the TMD attached to floor 3, compared
to the mass of the TMD attached to floor 5. However, the response of
the TMD itself is different. A TMD with a larger mass at a lower floor
level results in a smaller TMD stroke.
The effect of an additional TMD tuned to the second mode is studied

for medium height buildings (5 to 15 storeys). From the outcomes it
can be concluded that their effect on the structural response is almost
negligible. Exemplarily, Figure 4 shows a comparison of the structural
response of configurations with one respectively two TMDs, attached
Figure 4 — Structural response.

Frame equipped with one
TMD tuned to the first mode
versus frame equipped with
two TMDs tuned to the first
and second mode, respec-
tively. Optimal TMD positions.
Figure 3 — Structural response

without and with TMD
attached at different storeys
to a five-story frame as described above. For all TMDs the mass ratio is
5%. The structural response is almost unaffected by the second TMD,
compare e.g. the red solid line and with red dotted line. The optimal
attachment location for the second mode TMD is at the second floor.
These findings presented for a single earthquake were evaluated and
confirmed to hold true also in a statistical sense based on the responses
to complete sets of earthquake records.
Simplified assessment of multi-degree-of-freedom

structures
Field of application
The proposed simplified assessment method applies to structures, whose
seismic response is dominated by the fundamental mode. Consequently,
the MDOF structure can be transformed into an equivalent SDOF system
[12] without involving large errors by neglecting higher mode effects.
For the equivalent SDOF system the response spectrum method [12] can
be utilized to obtain the estimated seismic peak response.
Modified response spectrum methodology

A simple and time-saving procedure for estimating the performance
and benefits of a TMD is proposed in [6]. It is based on the equiv-
alent SDOF system approach, the derived analytical expressions for
response coefficients (capturing the effect of a TMD), and displace-
ment response spectra. Thereby, the engineer in practice obtains
expected maximum seismic displacements, the TMD induced reduc-
tion of these displacements, and the TMD stroke (relative displace-
ment of the TMD with respect to the attachment location) without
time-consuming analyses appropriate for an early design phase.
The general approach is presented in [6] and can be summarized as

follows:
–– Transform the considered stand-alone MDOF structure into its

equivalent SDOF system, and determine its period, damping
coefficient, and modal participation coefficient.
–– The maximum displacement of the equivalent SDOF system with-
out TMD can be read from a displacement response spectrum
(e.g. from design codes).
–– Select the TMD mass and determine the modal mass ratio and
optimum TMD parameters according to Eqs (4) and (3).
–– Apply the derived analytical relations for the response coefficients to
obtain the expected maximum structural displacement and stroke
based on the equivalent SDOF system equipped with a TMD.
Median values, as well as 16% and 84% quantiles can be determined.
–– The maximum structural and relative TMD displacements of the
MDOF structure are back-calculated from these outcomes.
Validation
The proposed methodology is based on several assumptions and simpli-
fications, each of them leading to a deviation from the “exact” outcome:
1. Higher mode effects are not captured by the equivalent SDOF system.
2. The response spectrum method itself is a simplified approach for
estimating maximum displacements of SDOF systems.
3. Analytical (smooth) response coefficients are an approximation of
the actual coefficients.
Separating errors by these simplifications is a challenging task, and there-

fore, the results of the methodology are evaluated in a global approach.
10-storey shear frame structures with fundamental periods varying

from 0.5 s to 5.0 s, fundamental modal mass ratios of 2%, 4%, and
5%, inherent damping of 0.5%, subjected to the 44 records of the
ATC63-FF set, are subsequently utilized to assess the outcomes of the
proposed methodology. For each earthquake record the correspond-
ing response spectrum is calculated in advance. In Figure 5 maximum
roof displacements ( i = 1,…,44) from time history analyses are plotted
against the corresponding quantity based on the proposed simplified
method. Red circles identify outcomes for stand-alone structures with-
out TMD, while blue dots represent results for frames with attached
TMD. Three distinct mass ratios are considered, thus, each red circle is
connected (by green lines) to three blue dots.
An offset of the red circles from the diagonal dashed line results from
the equivalent SDOF system approach. A green line pointing directly
to the coordinate origin indicates a perfect approximation of the seis-
mic response coefficient. For the utilized shear frame model in most
of the considered cases the equivalent SDOF system with an assumed
Figure 5 — Comparison of
approximate and time history
analysis results; individual out-
comes of the peak displace-
ment at top floor level for a
10-storey structure. (a) Full
range. (b) Zoomed view.
linear first mode shape overestimates the structural peak displace-

ment. For a small number of configurations the TMD even impairs
the structural response, i.e. the green lines exhibit a negative slope.
However, in a statistical sense the proposed approach provides appro-
priate response predictions.
Conclusions
After introducing seismic response coefficients the impact of a non-opti-

mal TMD location in an MDOF structure was studied. A constant effec-
tive mass ratio requires a larger actual mass to provide the same efficiency.
A time-saving and easy-to-apply methodology for the engineer in

practice was described to assess in an early design phase the seismic
TMD performance for high-rise structures. It is based on the equiva-
lent SDOF model of the structure, a displacement response spectrum,
and analytical seismic response coefficients capturing the effect of a
TMD. A global validation of the proposed approach shows that the
results are in good agreement with the “exact” outcomes. The meth-
odology provides absolute values of structural peak displacements and
their reduction by a TMD. Moreover, the maximum relative TMD dis-
placement is estimated to decide on space requirements or mechani-
cal measures to limit the TMD stroke.
References
[1] Den Hartog, J. P. (1956). Mechanical Vibrations. McGraw-Hill,

New York.
[2] Warburton, G.B. (1982). Optimum absorber parameters for various

combinations of response and excitation parameters. Earthquake
Engineering and Structural Dynamics, 10 (3), 381–401.
[3] Hoang, N., Fujino, Y., Warnitchai, P. (2008). Optimal tuned mass
damper for seismic application and practical design formulas. Engi-
neering Structures, 30 (3), 707–715.
[4] Tributsch, A., Adam, C., Furtmüller, T. (2011). Mitigation of earth-

quake induced vibrations by Tuned Mass Dampers. Proceedings of
EURODYN Conference, Leuven / Belgium.
[5] Tributsch, A., Adam, C. (2012). Evaluation and analytical appro-

ximation of Tuned Mass Damper performance in an earthquake en-
vironment. Smart Structures and Systems, accepted for publication.
[6] Tributsch, A., Adam, C. (2012). Simplified assessment of structure —

Tuned Mass Damper assemblies under earthquakes. Proc. 5th Euro-
pean Conference on Structural Control (EACS 2012), Genoa / Italy.
[7] Sadek, F., Mohraz, B., Taylor, A. W., Chung, R. M. (1997). A Me-
thod of Estimating the Parameters of Tuned Mass Dampers for Seis-
mic Applications. Earthquake Engineering and Structural Dynamics,
26 (6), 617–635.
[8] FEMA P-695 (2009). Quantification of Building Seismic Performance

Factors. Federal Emergency Management Agency, Washington D.C.
[9] Medina, R.A., Krawinkler, H. (2003). Seismic demands for non-

deteriorating frame structures and their dependence on ground
motions. Report no. 144, The John A. Blume Earthquake Engineering
Center, Stanford University.
[10] Tributsch, A., Adam, C. (2011). Application of Tuned Mass Dampers in

an earthquake environment. Internal report, University of Innsbruck.
[11] Hochrainer, M. (2001). Control of vibrations of civil engineering

structures with special emphasis on tall buildings. PhD thesis,
Vienna University of Technology.
[12] Chopra, A. (2007). Dynamics of structures: theory and application

to earthquake engineering. Third edition. Pearson Prentice Hall,
New York.
206 Performance-based evaluation of seismic demands for moment resisting steel frames using N2-method
Performance-based evaluation
of seismic demands for moment
resisting steel frames using
N2-method
TSONEV, Aleksandar Abstract

UACEG, Bulgaria
Performance-based design approach is currently recommended in the
BONEV, Todor
UACEG, Bulgaria modern seismic design and its regulations and provisions. The biggest
advantages of N2 method in comparison to capacity spectrum method
KOLEVA, Tatyana are summarized mainly in two items — there is no need to identify the
UACEG, Bulgaria elastic response and to define elastic-perfectly plastic capacity spec-
trum curve. Also, there is a good opportunity to control the solution
through displacement ductility ratio.
Transformation of the original multi-degree of freedom system into

equivalent single degree of freedom system is carried out on the basis
of plastic mechanism reached in the process of loading. Performance
point is found as intersection point considering two curves — capacity
curve, being associated to inelastic response of the structure subjected
to monotonically increasing lateral load and design spectrum curve,
defined through design spectra provisions of Eurocode 8.
As a result on output inelastic displacement demand, internal force

demand and interstorey drift demand are obtained and graphically
presented and discussed.
Introduction to nonlinear methods
Nonlinear methods are usually used to estimate the structural capac-

ity of buildings, designed on the basis of linear methods. Results of
nonlinear analysis give us useful information, such as:
–– Maximum reactions, which can be taken by specially designed

structures.
–– Maximum deformation of elements, which dissipate energy by
inelastic behavior.
Performance-based evaluation of seismic demands for moment resisting steel frames using N2-method 207
–– The influence of the decreased resistance of some elements over

the whole structural behavior.
–– Plastic hinges formation and their location.
–– Actual displacement between adjacent storeys and P-∆ effects.
–– The most appropriate way for leading the load forces to the struc-
ture foundations.
Main methods in modern seismic design are dynamic nonlinear analy-

sis and static nonlinear analysis. The first one is more precise but also
difficult and time consuming. Therefore it is used for special studies.
Some of static nonlinear methods are capacity spectrum method
and N2 method.
N2 method
In recent years, simplified non-linear methods based on pushover

analysis, equivalent single degree of freedom (SDOF) system and
response spectrum approach have been implemented in guidelines,
standards and codes for seismic resistant design of new buildings and
evaluation of existing buildings. The methods have been developed
for fixed base structures. One of simplified non-linear methods is the
N2 method, which is based on extensive studies of inelastic response
building structures performed by the Prof. P. Fajfar and his research
team at the University of Ljubljana.
The method combines pushover analysis of a multi degree of freedom

(MDOF) model with the response spectrum analysis of an equivalent
SDOF model. Two main assumptions of the method are: the response
of a structure is governed by one mode and this fundamental mode
does not change significantly when the structure is subjected to differ-
ent seismic intensities.
Pushover analysis is a nonlinear analysis carried out under conditions

of constant gravity loads and monotonically increasing horizontal
loads. It may be applied to verify the structural performance of newly
designed and the existing buildings for the following purposes:
–– To verify the overstrength ratio values.

–– To estimate the expected plastic mechanism and the distribution
of damage.
–– To assess the structural performance of existing or retrofitted
buildings.
–– As an alternative to the design based on linear analysis which uses
the behavior factor q.
In that case the target displacement should be used as the basis for
the design.
Performance-based evaluation of seismic demands for

moment resisting steel frames using N2-method
Structure modelling
The purpose of this paper is to examine the seismic response of a 3D
steel frame structure in transverse direction. In this direction seismic
loading is taken almost entirely by the steel frames. The whole con-
struction fulfills the requirements of Eurocode 8, part 1, point 4.2.3
criteria for structural regularity. Therefore a planar model is used. It
represents one typical steel frame.
Material definition
Figure 1 — Material definition

for columns and beams
Transversal sections definition
Figure 2 — Column and beam

section definition
Geometry definition and frame supports

Two spans l1 = 7.2 m and between them one span l2 = 6.0 m; floor
heights h1 = h2 = 3.9 m; in transverse direction the frames are situated
in intervals of b = 6 m; support conditions — moment resisting columns.
Figure 3 — Beam section

definition
Seismic action parameters

I = 1.0 — importance factor
аgR = 0.25 — reference peak ground acceleration for Sofia, Bulgaria
according to map of seismic districts of Bulgaria
soil conditions — soil type “С”;
behavior factor q = 3.0 which transforms elastic acceleration spectrum
into design spectrum
Defining and entering the loads in the design model

Loads in constructions are categorized in 2 groups: permanent and
temporary. Temporary loads are 4 types — long term, short term, spe-
cial and seismic. According to the way they change in time — dynamic
and static. In this example every load is defined in its own load case
with base normative values. To calculate their design values load coef-
ficients G и Q are used.
Defining vertical static loads

–– Self weight (DEAD) — includes main frame construction self weight
–– Permanent weight (G) — includes weight of floor and roof con-
struction, surroundings, coverings, isolations and instalations -
gk, 1, gk, 2, gk, 3;
–– Serviceability loads on the floor construction — vk;
–– Short term loads of snow — st.
Figure 4 — Static normative

load cases definition
Figure 5 — Applied self-weight

loads
Figure 6 — Applied service-

ability loads
Entering static loads in design model

To find linear distributed loads on the frame elements we have to mul-
tiply the given area load by the structure center distance.
Mass source definition

Concentrated masses in joints of construction are calculated automat-
ically after defining mass sources. They have influence on dynamic
behavior of the system — eigenvalues and forms.
Figure 7 — Applied snow loads
Figure 8 — Mass sources

Analysis case definition
Figure 9 — Analysis cases
Results
Eigenvalues
TABLE: Modal Participating Mass Ratios
Output StepNum Period UK UZ SumUX SumUZ

Case
Text Untiless Sec Untiless Untiless Untiless Untiless
MODAL 1 1.726939 0.91604 0 0.91604 0

MODAL 2 0.644128 0.08184 0 0.99788 0
MODAL 3 0.512976 0 0.12615 0.99788 0.12615
MODAL 4 0.482115 0.0008 0 0.99868 0.12615
MODAL 5 0.401951 0 0.18409 0.99868 0.31024
MODAL 6 0.371829 0.00016 4.718E-20 0.99884 0.31024
MODAL 7 0.325506 0 0.23317 0.99884 0.54341
MODAL 8 0.244561 0 0.16149 0.99884 0.70489
MODAL 9 0.178429 0.00112 8.306E-18 0.99995 0.70489
MODAL 10 0.173827 0 0.00279 0.99995 0.70769
MODAL 11 0.135365 0.00003618 0 0.99999 0.70769 Table 1 — Eigenvalues from
MODAL 12 0.131621 0.000005989 0 0.99999 0.70769 modal analysis
Definition of plastic hinges

N2 method considers the inelastic work of the material due to seismic
loading. It is necessary to be defined specific zones, where material
plastification is expected. Moment — curvature dependence should
be defined as well.
We assume that plastification zones have size which tends to 0. That

is why we use dependence between moment and rotation at specific
section. Plastic moment is characteristic of transverse section and it is
calculated from geometric parameters of the section.
Masses distribution to frame joints
TABLE: Assembled Joint Masses
Joint U1 U2 U3 Height
Text KN-s2 / m KN-s2 / m KN-s2 / m m
1 0.38 0.38 0.38 0

2 5.44 5.44 5.44 3.9
3 2.75 2.75 2.75 7.8
4 0.14 0.14 0.14 0
5 10.17 10.17 10.17 3.9
6 5.14 5.14 5.14 7.8
7 0.14 0.14 0.14 0
8 10.17 10.17 10.17 3.9
9 5.44 5.44 5.44 7.8
10 0.38 0.38 0.38 0
11 5.44 5.44 5.44 3.9
12 2.75 2.75 2.75 7.8
13 9.36 9.36 9.36 3.9
14 9.36 9.36 9.36 3.9
15 9.36 9.36 9.36 3.9
16 9.36 9.36 9.36 3.9
17 9.36 9.36 9.36 3.9
18 9.36 9.36 9.36 3.9
19 4.74 4.74 4.74 7.8
20 4.74 4.74 4.74 7.8
21 4.74 4.74 4.74 7.8
22 4.74 4.74 4.74 7.8
23 4.74 4.74 4.74 7.8
24 4.74 4.74 4.74 7.8
25 10.4 10.4 10.4 3.9
26 10.4 10.4 10.4 3.9
Table 2 — Assembled joint 27 5.26 5.26 5.26 7.8
masses 28 5.26 5.26 5.26 7.8
Figure 10 — Location of pos-

sible places for plastic hinges
in columns and beams
Placement of plastic hinges

Plastic hinges are formed where lateral moment gets highest values.
It’s accepted that hinges will come near joints.
Loads for horizontal static nonlinear analysis (pushover)

We use obtained distribution of masses (equation (7), (8)) to define
proportion of horizontal forces. We accept triangle distribution (Fig.
11) and force on second storey will have value of 1. The forces are
proportional to the mass on this storey and the storey height. They are
located on storey levels. The force on first storey (F1) is calculated with
equation (9). We save these loads in load case named “push”.
Figure 11 — Distribution of
horizontal forces
Setting up pushover analysis

Horizontal monotonically increasing forces are applied after vertical
loads (i.e. self-weight, serviceability loads, snow loads). So two steps
of loading are used:
–– Step 1: vertical loading — nonlinear analysis is used because step 2

can’t start from linear analysis (fig.12). The different load types are
combined according to Eurocode 1 with appropriate coefficients.
–– Step 2: horizontal pushover start after vertical nonlinear analysis
(Figure 13)
In step 2 horizontal increasing forces are proportional to F1, F2. We

define maximum displacement which is monitored.
Figure 12 — Setting up step 1

Figure 13 — Setting up step 2
Results of pushover analysis

The analysis finishes after the construction transforms into mechanism
(free movement). The steel frame fulfils the requirements of capacity
design — weak beams, strong columns. Therefore the choice of col-
umn and beam sections is appropriate.
Figure 14 — Hinge plasti-

fication
Figure 15 — Plastic hinges

destruction (in orange)
Capacity curve-shearing base force Fb – crucial displacement d
Figure 16 — Capacity curve

Equivalent Single Degree Of Freedom System
Figure 17 — ESDOFS curve
Figure 18 — Test results
Demand Spectra for constant ductility = 2 in AD format normalized to

1.0 g peak ground acceleration. From the test results we conclude —
the system works in plastic condition.
References
[1] проф. д-р инж. Росица Ганчева, (2005) Проектиране

стоманобетонни конструкции за сеизмични въздействия
[2] CEN (2003) Eurocode 8: Design of Structures for Earthquake Resis-

tance — Part 1: General Rules, Seismic Actions and Rules for Buildings.
[3] Peter Fajfar, M. EERI, A Nonlinear Analysis Method for Performance

Based Seismic Design, Earthquake Spectra, Vol.16, No.3, pp. 573–592,
August 2000
[4] проф. д-р инж. Здравко Бонев, гл.ас. д-р инж. Дилян Благов
(2012) Ръководство за упражнения по Проектиране на конструкции
на сеизмични въздействия
216 Post-Earthquake Damage Screening of Steel Buildings using Model-Based Damage Pattern Classification
Post-Earthquake Damage
Screening of Steel Buildings
using Model-Based Damage
Pattern Classification
YAMAGUCHI, Mayako Introduction

Department of Architecture
and Architectural Engineering,
After a large earthquake, the ability to quickly assess the structural
Kyoto University
integrity of buildings greatly contributes to the safety of building users
KURATA, Masahiro and the smooth recovery from the disaster. In Japan, professional
Disaster Prevention Research engineers registered as inspectors visit each building to inspect the
Institute, Kyoto University damage. This process takes a considerable amount of time. Moreover,
NAKASHIMA, Masayoshi
damage of structural members covered with architectural finishing is
Disaster Prevention Research difficult to detect by visual inspection in many cases.
Institute, Kyoto University
Sensors installed to measure structural responses of buildings can pro-
vide objective information on the structural integrity to building own-
ers and decision makers in a quick manner. Such a sensing system,
referred to as a structural health monitoring (SHM) system, has been
explored by many researchers and practitioners in recent decades.
However, SHM systems have yet to be widely implemented in build-
ings primarily due to initial and management costs from installing and
maintaining the sensors.
This paper describes the development of a damage detection system

for steel buildings using a damage pattern which enhances informa-
tion acquired from a limited number of sensors. The developed system
is intended to provide accurate and objective information on the dam-
age state of the monitored building with reduced cost and labor in
installing a sensing system. In the presented framework, information
extracted from collected sensor data and information estimated from
an antecedently prepared dataset from the structural analysis of the
monitored building are assimilated using a pattern classification algo-
rithm (i.e. decision tree in this paper) to estimate the damage intensity
and location from the changes in building dynamic characteristics.
Post-Earthquake Damage Screening of Steel Buildings using Model-Based Damage Pattern Classification 217
Damage detection system
The specifically designed system to assimilate limited information from

acceleration or strain sensors takes the following steps: 1) analyzes the
structure to obtain modal properties for a large number of damage
patterns, including damage of beams, columns, and/or bases and cre-
ates a dataset, 2) constructs the decision tree for damage detection
by applying a classification algorithm to the dataset, 3) identifies the
properties of the building using sensors before and after the earth-
quake, and 4) estimates the state of damage by applying the property
change rate data to the decision tree. This outlines is shown in Fig. 1.
Analyze preliminarily
P Modal analysis of intended building on various damage patterns.
Build damage detection classifier

P Prepare learning data for the change of dynamic characteristics of
the building with damage.
P Build a classifier by applying a machine learning algorithm to the
learning data.
Build target data

P
Estimate dynamic property of the building from sensors after an
earthquake. The changes in dynamic properties are the target data.
Detect damage
Figure 1 – Damage detection
P
Apply the target data to classifier and estimate the damage pre-
system with pre-analysis and
sumption. classifier
Application to low-rise steel building
What follows presents the application of the proposed damage detec-

tion system. The building used in numerical simulations is a five story
steel frame with concrete slabs located on Uji campus of Kyoto Uni-
versity. A damage type considered in the simulation is fracture of the
bottom flange of beams near beam-to-column connections. Target
data of vibration properties is obtained from system identification,
which is applied to the results of time history analysis of the five story
model, instead of the results from the sensor on the real structure.
Five story steel frame model

The building was modelled in a general purpose structural analysis
program, SAP2000 [1], based on the measured dimensions of the
five story steel frame building. Figure 2 shows the overall dimen-
sions of the building. The beams of the building are Japanese rolled
Figure 2 – 5 story steel frame

building
H-section, H-500 × 200 × 8 × 11 for B1 and B3 in 1st and 2nd floor,

and H-500 × 250 × 9 × 13 for 3rd to roof floor and B2 part of 1st and
2nd floor. The columns of the building are H-400 × 400 × 12 × 19 for
1st and 2nd floor and H-400 × 400 × 9 × 12 for 3rd to roof floor. The
transverse direction of the building is braced with L-65 × 65 × 5.5.
Steel-reinforced concrete slabs are used to construct floors and ceilings.
Fixed
Cutout
3750 mm
Displacement control
Figure 3 – FEM model
of beam
Initial stiffness falling rate Bearing force falling rate

Table 1 – Decrease in stiffness
of member property with bot- Positive bending 24% 41%
tom flange fracture Negative bending 23% 39%
In the numerical simulations the effect of fractures at the end of beams

induced by cyclic loading are considered, as this is a major damage
mechanism that has been historically observed under strong earth-
quake ground motions,. For simplicity, the analysis model simulates
beam fracture by reducing the cross section of the beam as shown in
Fig 3.The length of the reduced section is determined to achieve the
amounts of bending stiffness reduction equivalent to what observed
in a high-profile beam analysis using the finite element (FE) analysis
method. A beam with a fracture near the fixed end is analysed using a
general purpose FE program, ABAQUS 6.10 [2]. Considering the center
part of the beam as a point of contraflexure, half of beam is modeled

as a cantilever beam whose length is the half span of the structure. A
cutout is made near the edge of fixed side. The bending stiffness of
the beam is estimated from the pushover curve of the beam obtained
by the displacement-control loading in the direction of the strong axis.
Table 1 shows the results of FE analysis. When the bottom flange has a
cutout of 10 mm, the bending stiffness drops by 24%.
The analysis model accounts for the contribution of the concrete slab
on the bending stiffness of the beams by adding a concrete section with
an equivalent width calculated following the Design Recommenda-
tions for Composite Constructions [3]. The mass of the members is dis-
tributed uniformly at each joint. The foundation is assumed to be fixed.
The steel members and concrete are assumed to be elastic with Young’s
moduli of 2.0 × 105 N / mm² and 2.6 × 104 N / mm² respectively.
Modal analysis for various damage patterns

The modal properties of the building with various damage patterns are
obtained by modal analysis with Lanczos method in SAP2000. The
parameters in the damage patterns are the number and location of
damage. In total, 96 different patterns are analysed. Figure 4 shows
the possible locations of damage. For example, Figure 5 and Table 2
show the mode shapes and natural frequencies of the undamaged
model in the X-direction.
x: Damage location
Equivalent concrete slab
Beam section
Beam Column Figure 4 – Building model

showing locations of possible
damage at beam ends
Damaged beam
Figure 5 – Mode shapes of the

undamaged building model in
the X-direction
Mode 1st 2nd 3rd 4th 5th

Table 2 – Natural frequencies
of the undamaged building
model Natural frequency 1.73 4.96 8.27 11.6 13.9
Classifier for damage detection

The modal properties and the associated damage patterns serve as a
learning dataset for the damage detection classifier. As mentioned in
the previous section, the dataset contains information for 96 damage
patterns. A data-mining software Weka [4] is used to build the classifier.
Using a machine learning algorithm, the classifier which statistically

learns the correlations between a damage pattern and the changes
in modal properties is constructed. In the classifier, the number of
damages in each story is objective variable. And the change in modal
properties as explanation variable is:
U d,n
(1) C d ,n 1 100 n 1, 2 , 3 , 4 , 5
U u,n
where n is the floor number, U is modal displacement at each floor

normalized by roof floor displacement, of the damaged (d ) and
undamaged model (u).
In this study, the local additive regression of decision stumps (DS) [5]
is selected as the algorithm after the trial of several famous machine
learning algorithms (i.e, J48, regression of support vector machine).
DS is a one-level decision tree. It searches over all possible features to
split on, and for each one, search over all possible thresholds. DS does
regression based on mean-squared error where each node in DS repre-
sents a feature in an instance to be predicted, and each branch repre-
sents a value that the node can take. This ensemble learning algorithm
consists of the four steps explained in Fig 6.
Determine a suitable distance metric
Find the k nearest neighbors using the selected distance metric
Apply boosting to the decision stump algorithm using as training

instances the k instances
Figure 6 – Local additive Obtain the answer of the ensemble as the prediction for the
regression of decision stumps testing instance
Preparation of supplied dataset

Before and after an earthquake, the vibration characteristics of build-
ings are estimated by experimental modal analysis using building
responses that is obtained from sensors installed on the building. By
supplying the changes in the characteristics to the classifier, a pos-
sible damage pattern is estimated. In the numerical simulation, the
modal properties for supplied dataset are estimated from the accelera-
tion time history at each story that are computed by transient analysis
with the input ground motion as white noise, using the same model
as the aforementioned modal analysis. To obtain modal frequencies
and mode shapes, system identification methods are applied to the
acceleration responses at each story. The data contains errors from
the analysis and from the system identification process. By substitut-
ing the estimated modal properties with error into the classifier, the
number of damage locations at each floor is estimated. Provided with
the simulation results for several damage patterns, the accuracy of the
constructed classifier is evaluated.
Time history analysis

Ambient vibrations induced by wind and traffic loadings are assumed
as random white noise with broad-band frequency components. The
root mean square (RMS) of the input white noise is 10 cm / s² with a
sampling frequency of 100 Hz, a frequency range of 0.3 ~ 20 Hz, and
a time duration of 600 s.
Modal property extraction system identification technique

For extracting modal properties, the linear direct method (LD) is
applied for input and output acceleration data, and the frequency
domain decomposition method (FDD) is applied for output-only
acceleration data [6]. In general, LD offers more accurate estimates
than FDD, but accurate information on input is not always available
especially in case input of each story contains uncertain factors such
Figure 7 – Time history

analysis
as strong non-stochastic wind loading. Tables 3 and 4 show the com-

parison of the modal analysis results and the estimates by the LD and
FDD methods. Modal assurance criteria (MAC) is used to study the
similarity between mode shapes [7]. LD method gives the estimates
for all 5 modes, while FDD method has difficulty in estimating the 4th
and 5th mode. The mode shape estimated by LD method shows very
good agreements with the modal analysis results, with MAC values
higher than 0.998. The MAC values for the 2nd and 3rd modes are
slightly smaller for FDD method. Both methods estimate the natural
frequencies for the 1st to 3rd modes with similar accuracy.

Table 3 – Comparison of
modal analysis result and LD MAC value of mode shape 1.0000 1.0000 0.9999 0.9999 0.9983
estimates Error of natural frequency (%) 1.32 –0.71 –2.68 –4.18 –5.76
Table 4 – Comparison of
modal analysis result and MAC value of mode shape 1.0000 0.9990 0.9941 – –
FDD estimates Error of natural frequency (%) 0.81 –0.63 –3.73 – –
Damage detection using classifier

The rates of change in the modal properties estimated by the system
identification methods are inserted to the classifier for estimating the
number of damage locations in each story. The test dataset consists of
7 amage patterns for 0 to 2 damage locations in different stories (see
Table 5). Each pattern has 4 simulations in which output data includes
no noise or 3 noise levels of 2.5%, 5%, 15% in terms of S / N ratio.
Indices applied for evaluating the accuracy of damage detection are the
correlation coefficient (CC) and the mean absolute error (MAE) between
the estimated and the actual number of damage locations. When esti-
mation is good, the CC is close to one and MAE is close to zero. When
estimation is poor, CC is close to zero and MAE becomes a larger number.
Figure 8(a) shows the comparison of damage detection accuracy for

several different classifiers with variations in explanation variables (i.e.,
modal parameters estimated by the LD method). For instance, 3MS
means that the classifier uses 1st to 3rd mode shapes. 5MSF means
that the classifier uses 1st to 5th mode shapes and natural frequen-
cies. As a result, the classifier with 1st to 3rd mode shapes (3MS) pre-
sented the best estimation. Due to the lower accuracy in identifying
higher modes, the results with classifiers including the 4th and 5th
mode shapes were relatively poor. On the other hand, when classifiers
included only 1st and 2nd mode shapes, the available information was
not enough to classify 96 damage patterns in the learning data, and
thus 1MS and 2MS gave relatively poor results. econdly, the effect of
noise in sensors is evaluated by changing noise levels as 2.5%, 5%, 15%
in terms of S / N ratio. Figure 8(b) shows the damage detection accuracy
for different noise levels. In all cases, 1st to 3rd mode shapes (3MS) are
considered. In spite of variations in each case, the accuracy tended to
decrease when noise increased. For example, the CC dropped by 0.07
and the MAE increased by 0.06 when noise of 15 %was considered.
2nd floor 4th floor Roof floor

Pattern number Number of damage locations
1 1 1 0
2 1 0 1
3 0 1 1
4 2 0 0
5 0 2 0
6 0 0 2 Table 5 – Damage patterns
7 1 0 0 present in the dataset
Future research subjects
Improvement of classifier
In order to identify the intensity and location of local damage such as
a fracture of beam flange, a large number of sensors must be installed
to a building. On-going research is to detect the change in bending-
moment distribution using strain sensors (Figure 9). Such index is
considered to be more sensitive to local damage compared to modal
properties. Although the detection of local damage by the sensors
placed in adjacent to the damage is possible, the amount of cost and
labour to scatter the sensors in a building become are quite large.
Therefore, how to decrease the number of sensors yet increase the
accuracy becomes a challenge.
Figure 8 – Comparison of
errors in damage detection
using classifier: (a) Different
explanation variables; (b) Dif-
ferent noise level
Strain sensors Moment ratio of each location

Figure 9 – Example of poten-
tial strain sensor locations
(floor plan)
Shaking table test
A scale steel structure without concrete slabs was built in the labora-
tory at the Uji campus of Kyoto University. A series of shaking table
tests are undergoing with. a measurement system contained with
accelerometers, and strain sensors. Figure 10 shows the scaled steel
structure mounted on the shaking table. At connections, joint panels
and a beam or column are connected using five dog-born shape steel
connectors; two for connecting each flange and one for web. Beam or
column fractures are replicated by removing the connectors. A wire-
less sensing system is used to send acceleration or strain data to a
main server, which reduces labour costs in wiring sensors.
Connection Wireless sensor
Figure 10 – Scaled model of

three story steel structure
Conclusions
The development of a damage detection system using machine-learn-

ing-based classifiers for steel structures was described. Pre-analysis
result showed that the changes in mode shapes differ with the various
damage patterns. Using an additive regression of decision stumps, the
classifier for 96 damage patterns are developed and the best accuracy
was obtained when the 1st ~ 3rd mode shapes change rate was chosen
as explanatory variables.
Further research is necessary to increase the robustness of the pre-

sented damage detection system for accounting higher level of uncer-
tainties resulted from measurement noise, non-structural components
and non-stochastic wind loadings. The learning dataset of 96 patterns
should be expanded to consider all possible combination of damage
in a building. In such case, the classifier and system identification
algorithm with better performance are desirable to accommodate the
increase complexity in damage detection.
References
[1] Computers and Structures (2011). SAP2000 Documentation.
[2] Dassault Systems (2010). ABAQUS 6.10 Documentation.
[3] AIJ (2010). Design Recommendations for Composite Constructions
[4] The University of Waikato (2011). Weka Documentation.
[5] Sotiris B. Kotsiantis, Dimitris Kanellopoulos, Panayiotis E. Pintelas

(2006). Local Additive Regression of Decision Stumps. Lecture Notes
in Computer Science, Springer-Verlag, Volume 3955, S. 148–157.
[6] Rune Brincker, Lingmi Zhang, Palle Andersen (2002). Output-Only

Modal Analysis by Frequency Domain Decomposition. 25th Internati-
onal Seminar on Modal Analysis (ISMA), Leuven / Belgium, S. 717–723.
[7] Randall J. Allemang (2003). The Modal Assurance Criterion (MAC):

Twenty Years of Use and Abuse. Sound and Vibration Magazine,Vol. 37,
No. 8, S. 14–23.
Structural Solutions for the Construction of National Beverages Company Warehouse (NBC) Project 227
Structural Solutions for the

Construction of National
Beverages Company Warehouse
(NBC) Project
Abstract ZIMMO, Iyad

Civil Engineering
This paper is intended to describe the structural analysis and design
of the National Beverages Company Warehouse (NBC) to be built in
Ramallah, Palestine. This building comprises of 7 floors of which 2 are
underground, all with a total construction area of 3400 m² per floor.
This structure was analyzed through three different design flooring sys-
tem alternatives: Composite Construction, Two-way slab on beams,
and flat slab. Studying these alternatives to serve for large spans and
long cantilevered zones, steel manual (AISC) was used for composite
construction while ACI-318 Code used in both the two-way slab on
beams and flat slab systems. Software programs like SAP2000 and
ETABS were used to check the results obtained from manual analy-
ses of the structural systems. The two-way slab on beams system was
applicable for underground floors for the reason of fire considerations
and resistance to underground water, while composite construction
system was suitable for the superstructure where large spans of com-
posite beams obtained a level of economy, function, and safety.
Introduction
The main objective of this project is to present different structural

alternatives for the design of the NBC warehouses in which these
alternatives are evaluated and compared to determine the best alter-
native that is economical, functional, and safe. But in this paper, inten-
tion will be focused on the challenges faced in this project for using
different flooring systems, and the way dealing with these challenges
will contribute to the best choices. Starting with an introduction about
the project, the proposed project site is located in the Industrial Zone of
Bitunia — Palestine with a total floor area of 3400 m². Land uses in this
industrial zone are mainly commercial, small industries, and warehouses.
228 Structural Solutions for the Construction of National Beverages Company Warehouse (NBC) Project
The project serves for a 7 floors building (two underground floors) for
multi-functions like storage areas, plant rooms, car park, maintenance
workshop, and services offices.
The heights and live loads in each floor of the project are summed
as following:
–– Basement floor: Basement floor is 7.8 m below the ground level.

This floor will serve for a car parking area with a live load of
2 KN / m².
–– Ground floor: The ground floor is +0.0 m with a ground mezzanine
of +3.90 m level. Ground floor includes warehouse area of
2606 m². The total area of ground floor with ground floor mezza-
nine is 2,980 m2 with a live load of 5 KN / m²
–– Typical floors: There are three typical floors with identical heights
of 3.45 m, and they serve as an open plan offices and services with
a live load of 5 KN / m².
–– Roof floor: This floor is the last upon the floors of the building,
and will not be covered, while the live load on this floor
is 1.5 KN / m².
Structural Systems
In this section, three distinct and viable options are presented through
the design requirements and the structural solutions. The section con-
cludes with a comparison behavior between the structural systems.
Composite Construction with Steel Deck:

This system consists of structural steel deck, a concrete slab with steel
mesh against shrinkage and temperature connected to a steel beam
by shear connector studs to prevent the slipping between the con-
crete and the steel. The studs are welded through deck to the top
flange of the beam. [1]
At construction stage:
For this stage, there is no composite action, but steel section is fully
restrained against lateral torsional buckling by the deck and studs,
hence the strength in which the beam strength is checked for only
weights of slab and construction loads is: [2]
(1) bMn = ZxFy
On the other hand, the serviceability requirements for the beam should
be checked. The loads causing deflection at the construction stage are
the weight of steel beam, slab, and construction loads only. Shoring
is not to be used; the steel shape and metal deck have to support the
weight of concrete by themselves. It is recommended according to the

AISC Specifications that an incidental construction loads of a value of
1 KN / m² to be added if shoring is not applied for the composite con-
struction. This requires deflection to be checked for construction and
composite stages. [1] Deflection is computed by:
∆b = 5wDl2384EsIs (2)
At composite stage
After concrete curing, there will be a unity action between steel and
concrete. Hence a transformed section is obtained by converting the
concrete slab into an equivalent steel section, since no slipping is
permitted by the shear studs, full composite action is assumed. The
strength is then tested for the dead, live, and partition loads. All loads
must be resisted by the internal couple corresponding to the stress
distribution at failure. The figure below shows the plastic analysis of
the composite section:
Figure 1 — Plastic analysis of

composite section
Serviceability check at composite stage must comply with the deflec-

tion caused by live and partition loads. Deflection caused by live and
partition loads is computed by:
∆a = 5wll2384EIeff (3)
Strength and serviceability for full composite behavior: Strength is

now tested with a reduction of the nominal capacity of the composite
section. This serves as a factor of safety for flexure. Reduced nominal
capacity of the composite section will be the smallest of:
bMn = bAsFyy = b(0.85fc’Ac)y (4)
Since strength is checked for the composite section, serviceability

requirements should be checked for the total deflection; deflection at
construction stage, and deflection at the composite stage caused by
live and partition loads. The limitation over the deflection is:
∆b + ∆a < l240 (5)

Shear Strength: [2]

All shear forces must be resisted by the web of the steel shape.
Depending on the chosen steel section, the adequacy of shear can be
determined.
(6) ØvVn ≥ Vu
Shear Connectors:
The horizontal shear force to be transferred between the concrete and
steel (V’) is equal to the compressive force in the concrete (C). After
determining the controlling value of the horizontal shear force to be
transferred between the concrete and steel, the number of shear con-
nectors (N1) is determined.
Complying with all the requirements of strength and serviceability for

the composite design with steel deck, a summary of the dimensions of
the designed beams is made. For choosing the adequate steel section
in terms of strength and serviceability, one of these two criteria will
decide upon the choice of the required section.
Span Tributary Beam Mu bMn Final Allowable Controlling

length area section (KN.m) (KN.m) deflection deflection criteria
(m) width (m) (mm) (mm)
9.6 9.825 W27X94 1760 1760 23.1 40.1 Strength

7.5 7.5 W18X46 805 855 25.9 31.2 Strength
7.5 9.825 W21X55 1070 1115 27.2 31.2 Strength
15.0 6.4 W30X108 2051 2148 62.0 62.5 Serviceability
Table 1 — Shows the results
of the composite construction 9.6 3.75 W18X46 672 875 35.6 40.1 Serviceability
design 7.5 3.75 W16X26 408 462 29.2 31.2 Serviceability
Figure 2 — Column layout

for the suggested composite
construction of the flooring
system of the typical floor
Composite Construction in Cantilevered Zones:

The structure involves two cantilevered zones, the first with 1 m can-
tilever, and the second with 6 m. These two zones are treated sepa-
rately. Since the first cantilevered zone has small span length of 1 m,
the metal deck used in composite construction will span as a cantile-
ver parallel to the primary beams for the 1 m distance. The metal deck
along its strong axis serves for tension steel for the concrete slab and
eliminates the chance of lateral torsional buckling from occurring. So
the steel deck will be extended for the required length of the cantile-
ver to comply with the cantilevered zone.
For the second cantilevered zone, the cantilever will span over a 6 m
length over the first typical floor to the roof floor. To solve for this can-
tilever span, a truss positioned over the roof is designed to carry the
distributed load of the roof and the load carried by the cantilever from
floors below it. External forces and reactions of these forces act on the
nodes of the truss resulting in either tensile or compressive forces in the
truss members, so truss must be designed to satisfy the strength and ser-
viceability requirements of compressive and tensile members. The load
of the roof will be treated as uniformly distributed load over the truss
bottom, while the loads of the floors below the truss will be treated as
point loads acting at the end of the cantilevered section of the truss. [3]
Cantilevered zones of the typical floors are lifted through tension col-
umns where each column carries the loads from its floor in addition
to loads from lower floors. The tension columns are then hanged by
the end of the cantilever truss where the columns assign dead and live
point loads at the end of the cantilever.
Figure 3 — Cantilevered zone

section showing tension col-
umns, truss cross-section and
6 m cantilever span
The truss will extend over two spans in which the first one extends
for a length double that of the cantilever and the second which is the
cantilever span. The figure below shows the modeling of the truss:
Figure 4 — Cross-section of the

truss of the cantilevered zone
Analysis of the forces in the truss members shows that all upper mem-
bers of the truss are in tension. On the other hand, the lower members
are in compression, so the problem of lateral torsional buckling has to
be solved by connecting the truss bottom with the steel deck below it.
Through analysis and design of the truss members, the section that com-
plies with the strength and serviceability requirements is W40 x 199
used for all members of the truss. Deflection at the end of the cantile-
vered zone equals 2.8 mm which is less than the limit of 25 mm (l240).
Shown below the deformation shape of the truss:
Figure 5 — Deformed shape

of the truss due to point and
uniformly distributed loads
or the tension columns, a tension member of W14x26 is required for

the tension column in the third typical floor. This preliminary sizing
of this tension column is the most critical case in which the column
carries the largest loads in comparison with columns below it, so this
section is generalized over the tension columns of the first and second
typical floors. Serviceability requirement for the tension member is
checked where (lrmin) = 126 which is less than 300.
Reinforced Concrete Options
Two way slab on beams:

One of the options in which large spans challenge can be solved is by
using the system of two way slab on beams. According to ACI318-08
requirements [4], the thickness of the solid slab equals 25 cm. The
beams dimensions for the exterior support of 10 m span (interior beam)
are (35 x 70) cm, with 1025 for maximum negative, 825 for positive,
and 325 for other negative supports. The dimensions for the beams of
span 7.5 m (interior beam) are (30 x 50) cm, with 725 for maximum
negative, 425 for positive, and 725 for other negative supports. [5]
Two way ribbed slab on beams:

This option provides a reduction in the dead load of the slab, where
the solid slab is replaced by an equivalent ribbed slab. The moment
of inertia of the repetitive unit for the equivalent ribbed slab must be
equal or greater than the 1 m cross-section of the solid slab. The equiva-
lent ribbed slab repetitive unit to the 25 cm solid slab depth is a 10 cm
topping slab, 15 cm rib width, 25 cm rib height, and (45 x 25) cm of
hollow block. This reduces the dead load by 29% compared with the
option of using two-way solid slab on beams.
The maximum steel reinforcement was found to be 520 bars in the

10 m span at the support. For the typical floors the maximum rein-
forcement was found to be 420 bars at the middle of the span. The
beams dimensions are, for the exterior support of span 10 m (interior
beam) are (35 x 70) cm, with 1025 for maximum negative, 625 for
positive, and 425 for other negative support. The edge beam of span
10 m was found to be (40 x 50) cm, with 725 for maximum negative,
525 for positive, and 325 for other negative support. The dimen-
sions for the beams of span 7.5 m (interior beam) are (30 x 45) cm,
with 525 for maximum negative, 425 for positive, and 525 for
other negative support. The edge beam of span 7.5 m was found to
be (30 x 40) cm, with 525 for maximum negative, 325 for positive,
and 525 for other negative support. [5]
Reinforced concrete option in cantilevered zones:

The western side of the structure has a 6 m and 3 m cantilevers, and
has to be solved by a reinforced concrete option. The interior and
edge beams are extended through the cantilever and then cantilever
beams were analyzed either with two or three interior spans. Sizing for
the cantilever beams is now to be designed in terms of serviceability
and strength. [5]
The allowable thickness for cantilever beams is (l / 8) [4]. In which for
6 m span, the minimum thickness is 75 cm, and for the 3 m spans
is 38 cm. The elastic deflection was determined for the 3.0 m spans
(G2, G4) and found to be, for the live loading 3.5 mm and for both
Dead and live loads is 8.5 mm. The 6 m span (G1) cantilever is ana-
lyzed as a grillage, in which it carries an extended beam to its end, the
carried beam acts on the cantilever as a point load. In addition to the
point load, a distributed load acts on the beam in which the size and
the deflection are determined. The deflection for the 6 m span (G1) for
live loading is 5 mm, while for both dead and live load is 11 mm. The
size of the beam was determined to be (50 x 70) cm, but for service-
ability requirements (50 x 75) cm beam is used.
Figure 6 — Cantilevers of 3 m
and 6 m in the western side
of the structure
Flat Slab:
In this option, the slab directly rests on the column, and the load from
the slab is directly transferred to the columns and then to the foundation.
Figure 7 shows the column layout for the suggested flat slab in accord-
ance to the ACI code [4]. The thickness of the flat slab that is considered
adequate to resist the punching shear varies from region to region.
Typically, it ranges from 200 mm to 300 mm.
Figure 7 — Section through

slab and drop panel according
to ACI 13.2.5
For carrying out this project an interior panel of a flat slab with dimen-
sions 7.5 × 7.5 m, and super imposed load 21.93 KN / m² was designed
using the using ACI-318 code. Through the study of the suitability of
using a flat slab for spans greater than 7.5 x 7.5 m, the option of flat
slab is terminated for such case due to punching shear failure. [5]
Discussion
The flooring system for this warehouse undergoes a study through dif-
ferent alternatives considering reinforced concrete flooring systems
versus composite and steel construction systems.
The composite construction flooring system provides an economical

and effective system by taking advantage of the composite action
between steel beams and concrete slab. Such system offers a light and
fast-track construction. In addition, the steel deck serves as:
Figure 8 — Column layout

for the suggested flat slab
construction of the flooring
system of the typical floor
–– Formwork for the concrete slab, forming a permanent shatter.

–– Tension steel for the concrete slab.
–– A diaphragm resisting lateral torsional buckling of the beam during
the construction stage and allows the beams to develop their full
moment capacity.
Flat slab flooring system serves for some advantages such as:
–– Easier to construct and requires simple formwork, with shorter

construction time.
–– Flexibility with services and quicker to cast
–– Using drop panel increases shear strength of slab, increases neg-
ative moment capacity of slab, stiffens the slab and hence
reduces deflection.
Two-way solid or ribbed slab are useful when considering their fire
resistance and applicability for long spans and cantilevered zones. In
term of fire considerations, concrete is naturally fire-resistant. It can
reach high fire resistances for more than 3 hours depending on the
concrete cover. The rate of temperature rise across the concrete sec-
tion is low. Concrete sections can be repaired after fire if no severe
damages occur in the construction members. Using concrete cover
of 25 mm, 35 mm, and 50 mm protects the structure for 1 hr, 2 hrs,
and 3hrs respectively. The underground car park requires 2–3 hrs to
maintain a safe evacuation time for people inside the building. Steel
is poor in fire resistance, and the rate of temperature rise across the
section is very high with a low reparability potential after occurrence
of fire. Steel fire resistance can be improved by using fire protection
which is expensive. [6]
To sum up, concrete in underground car park provides this building

with high fire resistance and low costs in comparison with the use of
steel. This leads to the conclusion of using reinforced concrete two-
way ribbed slab for the car park since flat slab can’t be an option for
spans larger then 7.5 m. In addition, sprinklers, manual extinguishers,
and sand buckets are needed for fire extinguishing
For the maximum and typical span lengths, the composite and rein-
forced concrete flooring systems can serve for long spans, while
through the study of using the flat slab flooring system; it will not be
applicable for spans of more than 7.5 x 7.5 m. This is due to problems
that may arise for longer spans due to punching shear failure. On the
other hand, the increase in the steel beam cross-section in the com-
posite construction system is more economical than the increase in
the dimensions and steel in the cross-section of the two-way solid or
ribbed slab flooring system.
In terms of weight, the table shown downwards clarifies the dead

load weight per unit of area for each flooring system. The orienta-
tion should be towards lighter weights since it will reduce both cost
and time, but it must be noted that earthquake requirements will be
taken into consideration when launching the detailing process for
the chosen flooring system.
Type of Slabs Typical & Ground floors (KN / m²)
Table 2 — Shows the dead Flat slab 11.61

load weight of the flat slab,
Two way solid slab 10.20
two way solid slab, two-way
ribbed slab and composite Two way ribbed slab 7.26
construction framing system Composite system 6.42
This table indicates that the lightest flooring system in terms of the
dead load is the composite construction framing system since it
requires lighter weights of beams with the reduction of the amount
of concrete in comparison with the rest of the other systems. On the
other hand, the two-way ribbed slab system has considerably a light
weight flooring system, and should be used for fire resistance in com-
parison to composite steel construction framing system for the use
in underground floor.
Conclusion
This paper is accomplished for presenting different solutions for the

flooring systems that can be used in NBC warehouse project, and the
choice of the best combination of these solutions. For underground
floors, reinforced concrete system is to be used by a two-way ribbed
slab on beams, while for ground, typical and roof floors, composite
construction flooring system will be used.
Further analysis and work of this project will be the detailed design of

slabs, columns, and foundation. On the other hand, earthquake forces
will be calculated according to UBC-97 code. Bill of quantities (BOQ)
will be done using Microstation Triforma software in which the exact
quantities will be calculated with the advantage of clash detection by
modeling the structure. This leads to preparing a detailed cost esti-
mate for the whole structure.
References:
[1] Steel Construction Manual (2011). American Institute of Steel Con-

struction 325–11. 14th ed.
[2] William T. Segui. (2006). Steel Design, Composite Construction,

P. 552–610
[3] ACI 318, 2002, Building Code Requirements for Reinforced Concrete,
American Concrete Institute, Detroit, Michigan.
[4] Chu–Kia Wang, Charles G. Salmon, and José A.(2008). Reinforced

Concrete Design, Design of Two-Way Floor Systems, P. 620–709
[5] R. C. Hibbeler. (2011). Structural Analysis (8th Edition), Analysis of

Statically Indeterminate Structures, P. 395–448
[6] J. D. Hill, G. Rhodes, S. Voller, C. Whapples.(2005).Car Park Desi-

gners’ Handbook
Modelling of soil-structure
interaction in alpine regions
MARIN, Alexandru Abstract

Institute of Geotechnical
Engineering, ETH Zurich
The understanding and quantification of soil structure interaction
LAUE, Jan becomes more and more important in new projects as well as in the
Institute of Geotechnical risk assessment of existing buildings. The recognition of local site
Engineering, ETH Zurich effects in conjunction with special performance requirements imposed
by different types of structures (e.g. industrial buildings) is challenging
MEZGER, Florence
Institute of Geotechnical and appropriate evaluation of the soil-structure interaction is there-
Engineering, ETH Zurich fore neglected. The possibility of using widely available FE software to
consider soil structure interaction is being explored in this contribu-
tion. The studied case is an industrial building with different types of
foundations located in a glacier alpine valley with extensive deposits
of alluvial material, characterized by a moderate seismicity. Performed
verification procedures and calibration of input parameters are pre-
sented with the purpose of facilitating the future use of the employed
software in other similar situations.
Introduction
Changes in risk perception, as well as increasing requirements aris-

ing from the corresponding changes in codes for engineering pur-
poses, result in a need in practice for a clear, simple and efficient way
to deal with field situations incorporating soil structure interaction.
These have to be provided with the limitations of possible numeri-
cal modelling using available FE program and respecting the bound-
ary conditions of modelling procedures, different ways of analysis and
computation capabilities.
In this contribution, focus is given to a specific case situation, which

involves the task of modelling the soil, the foundation and the soil-
structure interaction for a generalised industrial building located in a
Modelling of soil-structure interaction in alpine regions 239
glacier alpine valley with alluvial deposits, in the area of Visp (Switzer-
land). Visp is characterized by a moderate seismicity (PGA = 1.6 m / s²
according to SIA 261, 2003) and experienced an estimated moment-
magnitude (Mw) of 6.4 event on July 25, 1855 (Fritsche, 2008).
The research is directly connected to COGEAR (Coupled seismogenic

Geohazards in Alpine Regions), a multidisciplinary project within the
Swiss Institute of Technology, funded by the Competence Centre for
Environment and Sustainability (CCES). The main object of study of
COGEAR is the seismic induced hazard chain and one of the main
interest areas is represented by the physics of non-linear processes,
which have been reported by eye witnesses about 150 years ago
(Fritsche, 2008), in relation to topography and soil disposition.
Idealised Modelling
The FE modelling was performed using PLAXIS 2D. Considering the

multitude of parameters and required assumptions it becomes very
important to understand their influence on the reliability of results.
Soil Profile
The soil profile (Figure 1) was obtained as the result of compiling already
existing data on the subsoil in the area of interest such as borehole
profiles (COGEAR Platform) and soil investigation reports (Resonance,
2010; Weber et al. 2007). Additionally using shear wave velocity pro-
files from ambient noise array measurements (Burjanek 2011), a final
simplification and verification of the model could be performed.
Figure 1 — Soil Model
A sensitivity analysis was performed in order to obtain the optimal

width of the model, which will minimise the unwanted effect of the
reflected waves. PLAXIS offers the possibility to define absorbent
boundaries by means of the relaxation coefficients. By varying these
coefficients from the standard values (C1 = 1 and C2 = 0.25), a change
in the response spectrum can be observed. The width at which a dou-
bling of these coefficients had an influence of less than 5% on the
response spectrum was considered the optimal width. After several
iterations, the final optimal value for the model width was determined
to be 900 m (Figure 9) for an assumed constant depth of the soft
deposits of approx. 100 m. This depth has been chosen as relevant for
the site of the industrial area (comp. Roten 2009).
Figure 2 — Spectral variation

with the increase of the relaxa-
tion coefficients
Material Parameters
For the simplicity of the calculations a Mohr-Coulomb model was
employed. The set of corresponding soil parameters is given in Table 1
based on Weber et al. (2007):
Strength
Layer Soil Type Stiffness parameters parameters
vs ν’ G E’ Eoed vp ϕ’ c’ref
[m / s] [-] [MPa] [MPa] [MPa] [m / s] [°] [kPa]
1 Silt 200 0.3 81.63 163.3 163.3 282 33.9 1

2 Gravel 390 0.3 357 713.9 713.9 551.5 35 1
3 Silt 400 0.3 326.5 653.1 653.1 565.7 33.9 1
4 Gravel 480 0.3 540.7 1,081 1,081 678.8 35 1
5 Silt 490 0.3 490 980 980 693 33.9 1
6 Moraine 720 0.3 1,217 2,433 2,433 1,018 35 1
7 Moraine 800 0.3 1,502 3,004 3,004 1,131 35 1
Table 1 — Soil Parameters 8 Moraine 850 0.3 1,696 3,391 3,391 1,202 35 1
The interface elements between structure and soil are considered to

have the same parameters as the rest of the soil elements in the data
set. Therefore the PLAXIS 2D parameter called Strength of the Inter-
face has been defined as rigid. One of the most important aspects of
the soil parameters in the dynamic analysis is the material damping,
which, in PLAXIS, is introduced by the two Rayleigh coefficients αR
and βR. According to Park and Hashash (2004) these coefficients may
be determined using the following system of equations:
1
f
(1) m 1 fm m R
4 1
n f R
fn n
where fm and fn are two significant natural frequencies and m and m

are the corresponding damping ratios.
In order to choose the most suitable pair of Rayleigh coefficients, the

proposed PLAXIS model was calibrated using a frequency domain
analysis in the elastic range, using the linear-equivalent analysis tool
EERA (Bardet et al. 2000).
The elastic range was identified as the range where the damping values
within the soil profile are under 5%. Above this threshold significant plas-
tic deformations occur and the behaviour of the soil becomes non-linear.
Four different earthquake signals (two earthquakes recorded by differ-
ent seismic stations: Aquila 2009, Mw = 6.3 (Figure 3 and Figure 4) and
Iceland 2008, Mw = 6.3; were employed by down-scaling them so
that all the values in the damping profile derived with EERA remained
under 5%. These specific earthquakes were chosen because they rep-
resent seismic motions with similar magnitudes and seismic frequency
contents with those expected in Switzerland.
Figure 3 — Iceland earthquake

signals (two different stations)
Figure 4 — L’Aquila earthquake

signals (two different stations)
Calibration of the Rayleigh Coefficients

The calibration procedure consisted of a trial-and-error method for
picking the right frequencies, which will lead to a good agreement
between the results delivered by the time domain (PLAXIS) and the
frequency domain analyses (EERA). There are already some publica-
tions (Park & Hashash, 2004; Visone & de Magistri, 2010) present-
ing different methods to choose the appropriate frequencies, which
would deliver convergence. In the present paper the best convergence
of the results was obtained for combinations between the second and
another superior frequency (third, fourth, etc). The damping ratios,
which are required in the equations system (1) were obtained for each
soil layer from the EERA analysis The results obtained with EERA and
PLAXIS for the above input signals are presented in Figures 5–8, also
indicating the natural frequencies.
Figure 5 — Calibration results

for Iceland 2008 — 013010
Signal
Figure 6 — Calibration results

for Iceland 2008 — 013011
Signal
Figure 7 — Calibration Results

for L’Aquila 2009 — ACG
Signal
Figure 8 — Calibration Results

for L’Aquila 2009 — GSG
Signal
Structural Elements, Foundation Types

The structural elements are simulated using plate elements, accord-
ing to Mindlin’s plate theory (Bathe, 1982). The superstructure was
replaced using the simplifying assumption of an equivalent distributed
load (including the self-weight and the live loads on the structure).
The analysis includes three types of foundations: a shallow foundation,

a deep basement and a piled foundation. In the case of piled founda-
tion, the spatial interaction between soil and pile rows was accounted
for by employing sheet pile walls with reduced stiffness, as suggested
by Randolph (1981) and Naylor (1982). The determined equivalent
Young’s Modulus resulted in a value of E = 4.04 GPa, which is 13%
of the E modulus for concrete, used in the other structural elements.
In order to obtain this value, the dimension of the foundation in the
direction perpendicular to the 2D problem modelled by PLAXIS was
assumed to be 30 m (assumption of a rectangular layout of the foun-
dation). The 7 circular piles, of diameter D = 75 cm, were assimilated
with a sheet pile wall of equivalent width d = 60 cm characterized by
a reduced Young Modulus of E = 4.04 GPa (Figure 10).
Input signal
The most common earthquake mechanisms in the area of Wallis, are
either strike-slip or normal fault (Deichmann & Ballarin-Dolfin 2003)
at a shallow depth and therefore all computations included in this
paper were performed using the 2008 Iceland (013010 Signal) earth-
Figure 9 — Foundation lay-

outs: a) shallow foundation;
b) deep basement; c) piled
foundation
quake (Mw = 6.3), which was characterized by a normal fault mecha-

nism and an epicentral depth of 10 km (USGS, 2008). One method
to obtain an input signal appropriate for the bottom of the FE model
from PLAXIS is to back-calculate from the given outcrop signal an
inside motion signal (which is more suitable for the non-rigid bottom
layer of the FE model). This can be easily obtained in EERA by applying
the input original signal to the model as an outcrop motion and cal-
culating the inside motion corresponding to the lower most layer. This
motion (Figure 11, left) is then used in the further analyses.
Results related to soil-structure interaction
The first parameter to be analysed is the acceleration acting at the

foundation level (Figure 10, right) using the inside motion derived by
the free field analyses.
Figure 10 — Left: Input Motion

(Iceland 2008 — 013010, in-
side motion); Right: Output
acceleration spectra at foun-
dation level
The spectral values of acceleration for the shallow and the piled foun-
dations are larger than the ones for the deep basement, but the peaks
stay at the same periods. From the point of view of the shallow foun-
dation, these results are in perfect correlation with the reality, consid-
ering that this foundation is basically placed on top of the soil layers,
where the amplification has the largest values. On the stiff side of
the spectrum (periods between 0.01–0.08 s) the shallow foundation
exhibits larger values than the other two.
As the foundation goes deeper in the soil deposits the amplifications

are smaller and therefore the accelerations decrease. On the other
hand, with the increasing depth, for a deep basement foundation, the
forces acting on the exterior walls are larger. This aspect will be con-
sidered in the following paragraphs.
The interesting fact is that, from 0.08 s towards the end of the spec-
trum, the spectral values for the piled foundation are basically in the
same range or even higher than the ones characterizing the shallow
foundation. This may be explained by the fact that the piled founda-
tions are subjected to a double action: the waves traveling through the
soil itself and the waves travelling through the concrete piles.
Figure 11.shows the horizontal displacements recorded in the centre

point of the foundation for the three foundations types.
Figure 11 — Horizontal dis-

placements for the three
types of foundation (Iceland
2008 — 013010)
The recorded displacements are ranging from 13.6 cm to 14.4 cm. It

becomes obvious that the smallest displacements are recorded at the
bottom of the deep basement foundation system. For the cases of
shallow and piled foundation the displacements are almost identical
(14.0 cm for the shallow foundation and 14.4 cm for the piled founda-
tion). This hints again to the special behaviour of the piled founda-
tion where the raft on top of the piles is subjected to superimposed
dynamic actions coming from the soil itself and from the piles, which
result in a slightly increased dynamic response.
The tilting of the foundation can be also analysed and compared for
the three chosen systems. Figure 12 shows the development over time
of the tilting, measured as an average from three points (the right
and left margins and the centre of the foundation). As expected the
largest residual, permanent inclinations are recorded for the shal-
low foundation, with an absolute value 0.052%. The piled founda-
tion follows with a residual tilting of 0.026% which means basically
half of the value from the shallow foundation. The smallest values are
recorded again for the deep basement, where an absolute average
inclination of only 0.016% was recorded. For this case, the vertical
displacements used in the tilt calculation, were measured on the top
plate of the basement, at the same level with the piled raft and
the shallow foundation mat, considering rigid box behaviour for the
deep basement structure.
Figure 12 — Average tilting

of the foundations (Iceland
2008 — 013010)
Interaction between basement walls and surrounding soil mass

Figure 13 shows the modelled behaviour using the recorded lateral
displacements of the two pairs of points on the wall and in the soil
from the close vicinity (Points A through D in Figure 13d). It becomes
obvious that the bottom points move together with the same ampli-
tude and frequency (Figure 13b).
On the other hand the points at the top of the wall show a different
behaviour: when the motion is in the positive direction (from left to
right), point A (the top of the wall) tends to move more than the
point B (top of the soil mass). In this way the interface opens creat-
ing an empty space behind the wall. Considering the fact that the
modelled soil is a granular material, this space will be quickly filled.
When the motion occurs in the negative direction (from right to left)
the observed tendency of point A (top of the wall) is to move less,
to remain behind, as point B (the top of the soil mass) advances and
leaves again an open space next to the wall.
Figure 13c shows the evolution of the relative displacement between

the points A and B, which increases incrementally in four main events
(out of which the largest one is marked with the dotted line). Impor-
tant to be mentioned is the fact that the residual relative displacement
of 5.9 cm is not the width of the open interface at the end of the
motion. In reality the material is granular and will fill any open spaces
behind the wall at least partially. Thus this value is rather another way
to observe and quantify the soil-structure interaction. With time, as
the interface repeatedly opens and is filled with granular material,
the lateral earth-pressure acting on the wall will vary (Figure 14) and
increase, as it also has been experienced with integral bridge abut-
ments (Ng et al. 1998). The observed behaviour is in perfect accord-
ance with previously observed data in the literature (Chari & Meyerhof,
1983; Tsang et al., 2002).
Figure 13 — Soil-wall interac-

tion: a) top lateral displacement
for wall and soil, b) bottom
lateral displacement for wall
and soil, c) top relative dis-
placement, d) schematic repre-
sentation of the first opening
of the interface wall-soil (not
to scale) — (Iceland 2008–
013010)
Figure 14 — Variation of the

pressure with time on the
left wall of the foundation
basement — (Iceland 2008 –
013010)
The mechanism behind this change in the pressure profile can be

explained in three steps. First, the material flows in the open interface
and accumulates behind the wall. Secondly, due to material accumu-
lation, a material densification will start to localize. Finally, as a result
of the densification, the stiffness of the system increases in the denser
area. The stresses are transferred from the lower stiffness areas (top
and bottom) to this area, causing higher pressures on the wall.
Performance based design perspective and conclusions
The results presented above matched, in all cases, expected and

already observed behaviour, confirming in this way the reliability of
the employed methods and software. The motivation behind this reit-
eration was introduced in the context of Performance Based Design
tendencies observed especially in the Swiss Norms. According to
SIA 260 (2003), paragraph 2.2 (Figure 15) the design of a structure
involves a service criteria agreement, which has to be defined in a
dialogue between the client and the designers of the project.
Figure 15 — SIA 260 Basis of

structural design (Par. 2.2)
Thus, there are two sides of the project, which have to be compared
and matched in advance: the required performance, which is given
by the service criteria requested by the client (owner) and the pre-
dicted performance, offered by the designer. Considering the required
boundary conditions the performance of the proposed solution can
only be evaluated by means of boundary criteria using simplified cal-
culations or by numerical simulations. The alternative of reliable cal-
culations and designing charts involves, in the case of seismic design,

a serious database of past events. In Switzerland, an area of moderate
seismicity, this matter is very delicate as the recorded significant events
are very few and a reliable set of results is therefore difficult to obtain.
The other alternative of using simulations in order to deliver predicted
performance provided the motivation of performing the above men-
tioned FE modelling. In this way it becomes very important to have a
clear, straight-forward, simple and in the same time efficient method
to approach design situations from the performance based perspective.
The finite element analysis for the sub-structure of an industrial build-

ing located in an alpine valley filled with fluvial sediments was con-
ducted. The material input parameters for the 2D model were obtained
from already existing data and the dynamic parameters were calibrated
using the frequency domain analysis in the linear range. Finally the
obtained results showed that the widely available PLAXIS software
can actually provide a reasonable approach to the performance based
design of structures, even though, in case of pile foundations, a 3D
analysis should be preferred.
The effect of the super-structure on the entire system cannot be

neglected, but as the target of the present paper was actually to con-
firm and present a possible approach of the performance based design,
this effect was not effectively analysed. Joining together an appropri-
ate modelling of the superstructure with the model presented in this
paper will lead to a consistent qualitative and quantitative analysis.
References
Antes, H., Jessberger, H. L., Schmid, G., Chow, N., Forchap, E., Grund-
hoff, T., Laue, J., Le, R., Siemer, T. (1996). Dynamic and cyclic soil-
structure interaction. In Krätzig et al. Eds. Dynamics of Civil Engineering
Structures, pp. 451-532. A. A. Balkema, Rotterdamm.
Bathe, K. J. (1982). Finite element procedures in engineering analysis.

Prentice-Hall, Englewood Cliffs.
Burjanek, J. (2011). Development of 3D Velocity Model for Visp area.

Presentation at: Third Annual Meeting of COGEAR, January 2011.
Chari, T. R. & Meyerhof, G. G. 1983. Ultimate capacity of rigid single

piles under inclined loads in sand. Canadian Geotechnical Journal, 20.
pp. 849–854.
COGEAR Platform. (2010). https://geodata.ethz.ch/cogear/

Deichmann, N. & Ballarin-Dolfin, D. (2003). Seismotektonische Karte
der Schweiz. Schweizerisches Erdbeben Dienst.
Bardet, J. P., Ichii, K. & Lin, C.H. (2000). EERA. A computer program for
equivalent — linear earthquake site response analyses of layered soil de-
posits. University of Southern California. Department of Civil Engineering.
Fritsche, S. (2008). Large historical earthquakes in Switzerland. Mul-

tidisciplinary studies on damage fields and site-effects. Dissertation no.
17710. ETH Zurich.
Naylor, D.J. (1982). Finite element study of embankment loading on

piles. Report for the Department of Transport (HECB). New Jersey.
Ng, C., Springman, S. M. & Norrish, A. (1998). Soil-structure interaction

of spread-base integral bridge abutments. Soils and Foundations, 38(1).
pp. 145–162.
Park, D & Hashash, Y.M.A. (2004). Soil damping formulation in non-

linear time domain site response analysis. Journal of Earthquake Engi-
neering, 8(2). pp 249–274.
Randolph, M.F. (1981). Pilot study of lateral loading of piles due to soil
movement caused by embankment loading. Report for Department of
Transport (HECB).
Résonance Ingénieurs-Conseils SA. (2010). Réévaluation du danger de

liquéfaction du sol sur le site de Lonza à Viége. Rapport technique RT
226.26/CL/MK.
Roten, D., Fäh, D., Bonilla, L. F., Alvarez-Rubio, S., Weber, T. M. &
Laue, J. (2009). Estimation of non-linear site response in a deep alpi-
ne valley. Geophysical Journal International, 178. pp. 1597–1613. doi:
10.1111/j.1365-246X.2009.04246.x.
SIA 261. (2003). Actions on structures. Registered code of the Swiss

Standards Association. Zürich.
SIA 260. (2003). Basis of structural design. Registered code of the Swiss
Standards Association. Zürich.
Tsang, N. C. M., England, G. L., Dunstan, T. (2002). Soil-Structure inter-

action of integral bridge with full height abutments. 15th ASCE Enginee-
ring Mechanics Conference. Columbia University, New York.
US Geological Survey. (2008). Iceland region earthquake 2008. Shake-

Map Archive ID: sqba2008.
Visone, C. & de Magistris, F. S. (2010). Comparative study on frequency

and time domain analyses for seismic site response. Electronical Journal
of Geotechnical Engineering, vol. 15.
Weber, T., Laue, J. & Springman, S. M. (2007). Geotechnical laboratory

tests for identification of soil parameters for the cyclic mobility model
of sandy soil from Visp (VS). IGT Internal Project no. 4698. ETH Zürich.
Fatigue Analysis of a Steel Bridge Included in a High-Speed Railway Line 251
Fatigue Analysis of a Steel

Bridge Included in a
High-Speed Railway Line
Abstract Rocha, João Francisco

The dynamic behaviour of bridges has been largely studied since the
Delgado, Raimundo
collapse of the first railway bridges. This dynamic behaviour is par- University of Porto, Portugal
ticularly important in high speed railway bridges which are usually
exposed to important cyclic loading. These cyclic loads can lead to Calçada, Rui
fatigue phenomenon in several structural elements of the bridge. Con- University of Porto, Portugal
sequently, fatigue must be properly taken into account either during
the design of new structures or during the assessment of existing ones.
The present paper focuses the fatigue analysis of an existing bowstring

steel bridge in a high-speed railway line. A Finite Element Model of
the bridge was developed and dynamic analyses of real train passage
were made in order to obtain the structure response in terms of defor-
mations, accelerations and stresses. This study is carried out according
to the Damage Accumulation methodology and is based on the Euroc-
ode procedures for fatigue analysis. Furthermore, a parametric analysis
of the classification of the constructive details is performed in order to
understand its influence regarding the structural fatigue damage.
Introduction
The dynamic behaviour of bridges has been largely studied. This

dynamic behaviour is particularly important in railway bridges, espe-
cially in high speed lines, which could give rise to important resonant
effects. Furthermore, railway bridges are usually exposed to important
cyclic loads which make them sensitive to fatigue phenomenon. This
phenomenon is especially important in high speed lines because of
the high stresses induced in the steel elements by the heavy loads.
This paper presents the fatigue analyses of an existing bowstring
bridge based on the Damage Accumulation Method. This methodol-
ogy allows the calculation of the fatigue damage induced by real traffic
252 Fatigue Analysis of a Steel Bridge Included in a High-Speed Railway Line
scenarios. The principal steps for the fatigue damage assessment are:
the definition of a traffic scenario and the selection of the suitable S–N
curves to describe the fatigue resistance of the materials, the calcula-
tion of the stress history at each critical point of the structure (struc-
tural elements or constructional joint details), the identification of the
extreme values of the stress history, the calculation of the stress range
histograms (carried out through the Rainflow Method) and the calcu-
lation of the accumulated damage through the Palmgren-Miner rule.
The traffic scenario and the S–N curves must be chosen according to
EN 1991, Part 2 [1] and EN 1993, Part 1–9 [2], respectively. Then, the
stresses time histories are determined using the FE model. The local
peaks of stresses are identified and the Rainflow Method is applied
to convert the data into stress range histograms. The S–N curves can-
not be directly applied to the stress history to calculate the fatigue
damage unless all the stress amplitudes are constant. Thus, the
Palmgren-Miner rule is applied to the stress histograms to account for
the different stress ranges. According to this rule the fatigue damage
should be the sum of all the partial damage for each stress range:
n nEi
(1) D i 1 NRi
where i is the number of different stress ranges, nEi the number of

cycles for each stress range and NRi the number of resisting cycles for
the respective stress range according to S–N curves. The fatigue limit
criterion is D = 1.
For the fatigue analyses of an existing bridge, it is necessary to classify

the structural profiles and the joint components. This step should be
done according to the detail specification of [2]. Only with a correct
classification of the constructional details can the fatigue damage be
accurately assessed. Figure 1 presents three types of constructional
details for three different types of fatigue failure.
Case Study
The present case study is a bowstring Bridge located in the Belgium

high speed line, between Brussels and Köln. The Prester Bridge has an
Figure 1 — Constructional
details for classification
(adapted from [2])
orthotropic metallic deck hanged by two lateral arches. These arches

have a rectangular hollow section linked by four tubular hollow sec-
tion girders. The deck has two main I-shape girders along the span at
each side, forty-one T-shape cross girders, and twelve U-shaped lon-
gitudinal stiffeners (Figure 2). The arches are connected to the main
girders by eleven squared hollow section hangers at each side of the
deck. The total length of the bridge span is approximately 136 m cov-
ered by an 18 mm thickness steel slab.
Figure 2 — Prester Bridge:

a) side view and b) bottom
(a) (b) view
The metallic bridges usually have complex joint details between the
structural elements. Although the bridge presents both welded and
bolted joints, the present work focuses only in the last ones. As an
example, Figure 3 illustrates two types of bolted joints adopted in the
Prester Bridge.
Figure 3 — Bolted joint details:

(a) (b) a) cross girders, b) hangers
For the dynamic analysis of the bridge, the three-dimensional model

developed in [3] was used. A few improvements were made to this
model in order to obtain the required stresses. The finite element
model of the Prester Bridge was developed with shell elements for the
deck and beam elements for the rest of the structural elements, such
as hangars, main beams and arches. The track was modelled through
beam elements for the rails and spring elements for both the pads and
the ballast stiffness. The mass of the ballast, sleepers and rails was
included on the shell elements of the deck. The FE model was devel-
oped with the software ANSYS10.0 (Figure 4).
Figure 4 — FE model of the

Prester Bridge
Fatigue Analysis
The dynamic analyses of the crossing of the real train on the bridge
were performed considering a linear-elastic model. The modal super-
position method was applied in the calculations. According to EN
1990-A2 [4], the scope of the modal analysis must be within the range
of frequencies of [0-30] Hz, resulting in 246 modes of vibration (some
examples at Figure 5).
(a) (b)
Figure 5 — Vibration modes

of the Prester Bridge:
a) and b) global modes,
c) Intermediate modes and
d) Local modes (c) (d)
Since the bridge is located in the high speed line between Belgium
and Germany, the train used in the fatigue analyses is the Thalys train.
A traffic scenario of 50 train passages / day is considered. This train is
approximately 393 m long, has a load per axle of 170 kN and a cir-
culating speed of approximately 300 km / h. The fatigue analysis was
performed for the range of speeds of [250–500] km / h, considering
a time step of 2 milliseconds. The values considered for the damping
ratios of each vibration mode are described in [5].
The fatigue damage was analysed on the rolled steel profiles and their
bolted joints. Three different parts of the bolted joints are analysed in
the present work: the rolled steel profile with holes, the cover plate and
the bolts. The classification of these constructional details was performed
according to [2] and for further details [5] should be consulted. The
fatigue damage results presented in this study are shown in percentage,
i.e., 100% of fatigue damage corresponds to the fatigue limit criterion.
Main Girders
The main girders are one of the most important structural elements of
the bridge. So, it is important to carry out an accurate analysis of the
damage over the years. According to the classification of the structural
elements and joint details specified in [5], the fatigue damage of the
main girders is presented in Figure 6.
(a)
Figure 6 — Fatigue Damage

on the main girders:
a) speed of 350 km / h,
b) speed of 450 km / h
(b)
The results presented in Figure 6 show that the fatigue damage for
100 years is significantly low for the speed of 350 km / h (only the bolts
at the joint at approximately 17 m have some damage). Similar con-
clusions can be withdrawn for the speed of 450 km / h where damage
only occurs on the bolts. In this case, the damage is close to 25% of
the fatigue damage limit. The conclusions are similar for several other
circulating speeds. However, for a speed of 490 km / h it is possible to
observe higher fatigue damage even for the rolled steel profile.
Cross Girders
For the analyses of the T-shape cross girders the same details of the joints
are considered. The results of the fatigue damage of one of the cross gird-
ers, located approximately at 19 m, can be observed in Fig. 7. Fatigue dam-
age in the cross girders is only registered for speeds higher than 450 km / h.

on a cross girder for the speed
of 490 km / h
The results presented show that the reduced flange of the girder is the
critical section. However, the cover plate has almost the same value of
damage over 100 years. It can be also noted that the rolled profile has
higher damage in the centre of the cross section of the bridge. Obvi-
ously, fatigue damage values over 100% are not admissible, i.e., the
fatigue life for these details is under 100 years.
Longitudinal Stiffeners
The U-shape longitudinal stiffeners do not have joints or any other
bolted connection details. This way, only the results of the rolled steel
profile are presented. Only the results for the longitudinal stiffener
under the centre of the track are presented because it is the section
which is exposed to higher stress range.
The longitudinal stiffeners are very sensitive to vibration caused by

rail traffic. Since the rigidity of these elements is relatively low high
fatigue damage values can be observed in some locations for a speed
of 350 km / h (Figure 8).
Hangers
The hangers, like the arches and the main beams, are one of the most
important structural elements of the bridge. These elements transfer
the loads from the main girders to the arches. They have two differ-
ent bolted joints, the first one between the square hollow section of
the hanger and the I‑shape rolled profile and the second between
this profile and the main girder. The results of the fatigue damage for
all hangers are in Figure 9. Through the analysis of the results pre-
sented in Figure 9, it is concluded that for the speed of 350 km / h
the fatigue damage on the bolts (either first or second joint detail) are
relatively high and very close to the limit criterion. For the train speed
of 450 km / h, this limit is exceeded on the bolts of five of the hangers.
However, it was not for the same hangers that registered the higher
fatigue damage for the speed of 350 km / h.

over a longitudinal stiff-
ener flange for the speed of
350 km / h

for the hangers:
a) speed of 350 km / h,
b) speed of 450 km / h
Parametric Fatigue Analysis
As a final analysis, a parametric study was performed in order to

understand the influence of the fatigue resistance classification of the
constructional details and also the influence of the frequency range on
the damage of the structure.
Detail Categories
The constructional details classification of an existing bridge can be
ambiguous. So, the calculation of the fatigue damage may not be
accurate. To understand the influence of this classification, a para-
metric analysis is carried out considering three suitable categories of
resistance for each constructional detail. For the main girders and the
longitudinal stiffeners only the rolled steel profiles can be classified
differently. In the case of the cross girders, it is possible to apply differ-
ent categories to the rolled profile and the cover plates. The hangers
are a specific case because the classification of almost every compo-
nent offers no doubt. The only exception is the cover plates that can
be classified one or two categories higher.
Main Girders
As mentioned before, the classification of the constructional details
can be ambiguous. To understand how important the classification of
resistance of the constructional details is for the fatigue damage two
categories of resistance are analysed for the main girders (one upper
and one lower) in addition to the reference category (112 MPa) [5].
As it can be seen in. Figure 10, for the category of resistance of 125 MPa
the fatigue damage is almost zero (for a train speed of 490 km / h).
However, when the lower category is considered, the fatigue damage
increases almost 60%, in the cases where damage already exists.

the fatigue damage of the
main girders for the speed of
490 km / h
Cross Girders
In the case of the cross girders three categories of resistance for the
rolled profiles and the cover plates are analysed. Figure 11 presents
the results of the comparison with the reference category (112 MPa).
Figure 12 shows the comparison between three categories of resist-
ance of the cover plate. The main concern in the cross girder must be
the holed flange of the steel profile. However, since there is no other
category of resistance for this type of detail, the parametric analysis is
considered only for the cover plate.
rolled steel profile of a cross
girders for the speed of
490 km / h
cover plate of the joint of a
cross girders for the speed of
490 km / h
The analysis of Figure 11 shows that the damage for the upper category
of resistance is practically zero. This is valid as long as this option is
correctly justified. However, the consideration of the lower category
results in an increase of the structural fatigue damage of over 50%. As
it can be seen in Figure 12, the consideration of the upper category
of resistance reduces significantly the final damage of the cover plate.
Longitudinal Stiffeners
As mentioned before, the longitudinal stiffeners are very sensitive
to vibrations caused by rail traffic. Therefore, it is observed that the
fatigue damage is already high for a train speed of 350 km / h. How-
ever, in terms of project management, the category of resistance can
be decided in order to improve the damage over the years. In this case,
applying a profile with an upper category of resistance can improve
the fatigue resistance of this element. Figure 13 presents the compari-
son between the reference category and an upper resistance category.
As it can be seen, for the speed of 350 km / h, an increase of the resist-

ance category improves the fatigue damage in an important way. It
can also be seen that the reduction of fatigue damage for speeds over
350 km / h is relevant in the case of consideration of the category of
resistance of 80 MPa.

the fatigue damage of a longi-
tudinal stiffener flange for the
speed of 350 km / h
As it can be seen, for the speed of 350 km / h, an increase of the resist-

ance category improves the fatigue damage in an important way. It
can also be seen that the reduction of fatigue damage for speeds over
350 km / h is relevant in the case of consideration of the category of
resistance of 80 MPa.
Hangers
The results presented in Figure 9 show that the fatigue damage on the
hangers occurs essentially on the bolts. The only way to reduce this
problem is applying preloaded bolts to this type of joints. In that case,
the shear force is not taken into account because the resistance is due
to the friction between the steel plates. This way, the fatigue analysis
is not performed in these elements.
Frequency Range
The dynamic analyses of the bridge considered the range of frequencies
from [0–30] Hz. For an accurate fatigue analysis this may not be suffi-
cient. Therefore, a dynamic analysis is carried out considering frequen-
cies until 42 Hz. The next sections presents the comparison between
the results obtained for the cross girders and the hangers for different
frequency limits.
Cross Girders
The cross girders are one of the elements more subjected to interme-
diate and local modes of vibrations, essentially present in frequencies
higher than 30 Hz. Figure 14 presents the results for the steel profile
and for the cover plate of one of the cross girders.

of the fatigue damage of a
cross girder for the speed
of 490 km / h
As expected, accounting for a larger number of vibration modes can

influence the final results of the fatigue damage. As shown in Figure 14
accounting for frequencies up to 42 Hz generates a higher fatigue
damage, especially at the centre of the cross girder.
Hangers
The hangers are one of the essential structural elements of the bridge.
Hence, an accurate prediction of the fatigue damage is important. Fig-
ure 15 presents the results of the fatigue damage for the hangers, con-
sidering frequencies up to 42 Hz.
of the fatigue damage on
the hangers for the speed
of 350 km / h
These results show that accounting for a larger range of frequencies

directly affects the fatigue damage of the hangers. However, it cannot
be said that frequencies up to 42 Hz have a negative impact on the
fatigue damage because not all of the hangers present a higher dam-
age comparing to the first analysis.
Conclusions
This work focuses on the fatigue analysis of the principal structural elements
of a bowstring bridge inserted in a high speed railway line. A procedure
for the fatigue analysis was presented based on the Damage Accumulation
Method. The Thalys train was considered for the fatigue analysis.
The case study used was presented and consisted on a bowstring

bridge. The structural elements analysed were the main girders, the
cross girders, the hangers and the longitudinal stiffeners. The bolted
joints details were also analysed. The classification of the resistance
category of the profiles and the joint details was done according to
the Eurocodes.
The fatigue analysis showed that the main and cross girders do not
have relevant fatigue damage for train speeds below 400 km / h. In the
case of the hangers it was observed that for a train speed of 350 km / h
some hangers already present significant fatigue damage. However,
the main fatigue damage is registered on the longitudinal stiffeners
due to their low rigidity. In these elements, high levels of damage
occur for speed of 350 km / h.
The fatigue analysis is sensible to various aspects of the dynamic analy-

sis. To reduce the uncertainty of the results a parametric analysis was
performed. The first topic analysed was the influence of the fatigue
resistance classification of the constructional details according to the
Eurocodes. It was shown that with small changes on the specificities
of the joints details, the fatigue damage can be reduced. The second
topic analysed was the frequency range that should be considered
in the dynamic analysis. It was concluded that the cross girders are
very sensitive to the vibration modes between 30 Hz and 42 Hz. The
hangers present a lower sensibility to this increase of number of vibra-
tion modes. However, in general, the increase of the frequency range
results in an increase of the fatigue damage.
References
[1] CEN (2003). EN 1991-2 — Eurocode 1: Actions on structures — Part 2:

Traffic loads on bridges. Brussels.
[2] CEN (2005). EN 1993-1-9 — Eurocode 3: Design of Steel Structures —

Part 1–9: Fatigue. Brussels.
[3] Albuquerque, C., (2008). “Dynamic behaviour of high speed railway

bridges with orthotropic deck”, MSc. Thesis (in Portuguese). Faculty
of Engineering of University of Porto. Porto.
[4] CEN (2005). EN 1990-A2 — Eurocode 0: Basis of Structural Design —

Annex A2: Applications for Bridges. Brussels.
[5] Rocha, J. F., (2009). “Fatigue Analysis of high speed railway steel
bridge with orthotropic deck”, MSc. Thesis (in Portuguese). Faculty
of Engineering of University of Porto. Porto.
Classification of timber–steel connections and the behaviour of semi rigid joints 263
Classification of timber–steel
connections and the behaviour
of semi rigid joints
Abstract GEČYS, Tomas

Department of Steel and
Timber structures, Vilnius
Semi rigid mounting joints of glued laminated timber elements are
Gediminas technical Uni-
connections which are fully assembled on construction site and capa- versity, Lithuania
ble to take bending moments, axial and shear forces. The need of
new type semi rigid timber elements’ connections is obvious. Pos-
sibilities of glued laminated timber elements’ manufacture are wide:
straight element’s length may be till 40 m or curved axis elements
with dimensions till 10 meters in height. So every day manufacturers,
erectors and others face with transportation problems of non–stand-
ard oversized elements. Semi rigid mounting joints are used for this
purpose. Non–standard oversized glued laminated timber elements
are cut into standard transportation dimensions in factory and fully
assembled into integral element on construction site using semi rigid
mounting joints. In this article the main types of timber joints are dis-
cussed and the classification according to strength and stiffness is pre-
sented. The influence of joint’s rotational stiffness on effective length
of element is discussed. Laboratory experiments of semi rigid timber
joints which are carried out by author are discussed.
Introduction
Glued laminated timber is universal structural material which is widely

used in glued laminated timber elements’ manufacture ([4]). Glued
laminated timber elements may be straight or curved axis till 40 m in
length and till 2.20 m in height.
The critical point of glued laminated timber elements is design and

manufacturing of joints ([3]). Timber is anisotropic material with dif-
ferent physical and mechanical properties along and perpendicular to
the grain. Joint can easily collapse because of exceeded stress in one
264 Classification of timber–steel connections and the behaviour of semi rigid joints
direction while in other direction there will be safe stage of stress. The
main collapse form is timber splitting along the gain which cause in
sudden loss of bearing capacity ([4]).
All timber structures’ joints generally treated as pinned (taking axial

and shear forces) or rigid (taking axial, shear forces and bending
moments). Analysing timber–steel joints in more detail, third type of
joints can be distinguished; it is semi rigid joints ([9]). An example
of semi rigid mounting timber joint with steel plates and dowels is
shown in Figure 1.
In timber structures almost all connections (between column and

foundation or column and beam) can be treated as semi rigid joints.
Timber element joint’s rotational stiffness is analysed by dividing joint
into components. All joints where large compression stresses arise in
contact zone perpendicular to the grain can be treated as semi rigid
connections because modulus of elasticity perpendicular to the grain
is about 7 times less than along the grain.
It is reasonable to take into account the real joint’s behaviour when

we need to design optimal cross sections of frame elements’. Taking
into account the real behaviour of joints there is possibility to redis-
tribute bending moments of frame elements’ ([1]). Element’s joints
rotational stiffness has great influence on element’s effective length.
Classification of timber–steel joints
As it was mentioned before, traditionally all timber–steel joints are

divided into two types: pinned (Fig. 2) and rigid (Fig. 3). In Figure 2
there is pinned support which can easily rotate around steel dowel
and such joint does not cause any bending moments in timber ele-
ment at the support.
Figure 1 — Semi rigid beam to

beam mounting joints in glued
laminated timber structures in
Moscow sport centre (joints
are shown in the circle)
Figure 2 — Glued laminated

timber element’s pinned
support
In Figure 3 there is rigid mounting joint with glued in steel rods. Based
on previous studies which experimentally and theoretically analysing
glued in rods in timber structures, it can be said that such joints cre-
ate tight contact between timber and steel because even at failure the
displacement between timber and steel rod is about 0.5–0.8 mm ([3]).
Figure 3 — Rigid mounting

joint of glued laminated
timber frame
In Figure 4 these is schematic view of semi rigid timber steel joint. Ana-
lysing timber joints’ rotational stiffness it was noticed that almost all
joints in timber structures which are assumed as rigid, can be treated
as semi rigid, except flocked connections which are shown in Fig. 3.
Figure 4 — Semi rigid timber

steel mounting joint:
1 — timber element,
2 — steel bar,
3 — steel plate,
4 — increased washer
Timber joints can be classified according to strength (Table 1) and

rotational stiffness (Table 2) ([6]). Classification according to stiffness
is based on element’s effective rigidity (EI / L) and initial joint’s stiff-
ness (Sj, ini). In normative documents there are no classifications of
timber joints according to strength and stiffness. Classification of steel
joints presented in EN 1993-1-8. Comparing timber and steel joints’
classification, it can be noticed that semi rigid timber joints’ rages are
narrower than steel.
Joint Condition
Full strength Mj,Rd ≥ 0.85 Mfull-stregth

Table 1 — Timber joints’ classi- Partly strength 0.25 Mfull-stregth < Mj,Rd < 0.85 Mfull-stregth
fication according to strength Pinned Mj,Rd ≤ 0.25 Mfull-stregth
Joint Condition
Rigid Sj,ini ≥ 12 EI / L

Table 2 — Timber joints’ clas- Semi rigid 0.5 EI / L< Sj,ini < 12 EI / L
sification according to stiffness Pinned Sj,ini ≤ 0.5 EI / L
From Table 1 it is seen that joint can be treated as full strength joint
when joint’s strength is more than 0.85 Mfull-stregth so there is no such
requirements as for steel joints. In steel joints’ classification there are
two cases when joints are treated as full strength, it is for frames with
bracing and without bracing. When frames are with bracings, there are
smaller restrictions for rigid joints and the use of them is wider ([1]).
Previous studies and types of analysis
Increasing demand for material savings and searching for rational con-
nections’ solutions, it was started to analyse the actual behaviour of
the joint. Semi rigid timber steel joints analysed by: laboratory experi-
ments ([5, 7]), computer simulation and component method ([5]). Lab-
oratory experiment is one of the most popular types of analysis. Using
laboratory experiments joints are analysed in: beam to beam connec-
tions ([7]), corner frame connections ([8]). Usually finite element mod-
elling of connections is used together with laboratory experiments to
validate results ([5, 8, 9]).
One of requirements for semi rigid as well for rigid timber joints is
elimination of the initial movement of the joint. So it means that joint’s
components need to be provided with initial pretension of pre-com-
pression. This problem is being solved by various methods: using bolts
with initial pretension and steel details ([8, 9]), using timber or steel
wedges ([5]), using filler to fill gaps between timber and steel ([3]).
Using different thickness wedges, there is possibility to obtain joints

with different rotational stiffness. Thickness of wedge influences rota-
tional stiffness, especially initial, but it does not have influence on final
bearing capacity of the joint ([5]).
Joint’s strength and stiffness are analysed by component method. Ana-

lysing joint by component method, joint is divided into separate com-
ponents and than analysed strength and stiffness of each component
([2, 6]). In Fig. 5 semi rigid timber joint with separate components is
presented. In Fig. 5 only components with timber compression are
presented. While analysing such joint’s behaviour (strength and stiff-
ness), steel details (steel detail and bar) must be taken into account.
Joint’s strength analysis usually performed using component method:

calculating each component bearing capacity and than joint’s bending
bearing capacity is determined. Bearing capacity of timber element
compression can by found from equation 1:
Fu , i Ai fc , ,d (1)
In equation 1:
Ai — area of component i,
fc,,d — strength of timber in compression.
After calculation of each component bearing capacity, bearing capac-

ity of the joint can be found from equation 2:
Mu Fu , i zi (2)
i
In equation 2:
Fu,i — bearing capacity of component i,

zi — distance from component’s i area centre to centre of joint’s rotation.
Figure 5 — Semi rigid timber

steel mounting joint:
a) joint’s scheme, b) strength
and stiffness components,
1 — timber element,
2 — steel detail,
3 — contact stress,
4 — steel bar
Figure 6 — Traditional cases

of determining the effective
length of timber element
Influence of joints’ rotational stiffness on element’s

effective length
Joint’s rotational stiffness has high influence on element’s effective

length. Traditional element’s end joints’ and length coefficients are
shown in Figure 6. These effective lengths are the same according to
EC5 and Lithuanian timber structures design codes.
When taking into account element’s joint rotational stiffness, the effec-
tive length can be increased or decreased. Some cases of determination
element’s with semi rigid joints, effective length are shown in Fig. 7 ([1]).
Generally, truss elements are treated as beams taking only axial ten-
sion and compression forces, so their effective lengths’ coefficients are
equal to 1.0. This is not correct assumption because usually there are
more than two bolts, so there is no free rotation at the joint. The same
case is with frame’s corner joint between column and beam. When
taking into account joints’ rotational stiffness there is possibility to
reduce element’s effective length.
Figure 7 — Some examples

of determination the effective
length of the element when
taking into account the sup-
port rigidity
The main design principles of semi rigid mounting

timber steel connections
The most responsible part in semi rigid timber steel joint is reliable
steel detail’s anchoring in timber element. In fact, four design princi-
ples which are used to anchor steel detail into timber element can be
identified. The simplest method is to use steel plates and bolts to fix
steel detail into timber ([14, 15]). Another way is to use glued in steel
rods to anchor mounting steel details. The third method to anchor
steel detail is to use wedges or fillers to fill the gap between timber
and steel. This method’s superiority is that using wedge the initial
rotational stiffness of the joint can be controlled or just to ensure tight
initial contact between timber and steel by using fillers. One more
method is to use large diameter circular section steel details, which
are anchored in timber by studs. Peter Bertsche developed this solu-
tion in mounting timber elements’ joints ([16]).
Laboratory experiments of new type semi rigid timber steel connections

are being done in Vilnius Gediminas technical university. Experimen-
tally and theoretically analyzed connection is made of three different
materials: the main element is timber, steel details are used to connect
these elements and the third material is filler which is used to fill the
gap between timber and steel, so that timber and steel would have
initial contact before loading. Two types of fillers are used: two com-
ponent polyurethane based filler and the other is cement based filler.
Conclusions
1. Joint’s rotational stiffness has great influence on distribution of

bending moments in frame’s elements. It is useful to take into
account the actual behaviour of joints when we need to design >
optimal cross sections of timber elements.
2. Joint’s effective length in some cases can be reduced when taking
into account joint’s actual behaviour, especially for joints which are
tend to be as pinned.
3. Semi rigid timber elements’ joints can be analysed by: laboratory
experiments, computer simulation (modelling by finite element) or
component method (analytically). Usually finite element modelling
is done with laboratory experiments to validate results.
4. All timber elements’ rigid connections where high stresses perpen-
dicular to the grain occur must be treated as semi rigid joints.
5. First laboratory experiments of semi rigid joints showed that there
is possibility to use fillers (to fill gaps between timber and steel and
to create tight initial contact between these three materials) in semi
rigid timber steel joints.
References
[1] Racher, P. (1995). Moment resisting connection. Proceedings of

Timber Engineering STEP I, C16/1-C16/10, Center Hount.
[2] Jirka, O., Mikes, K. (2010). Semi–rigid joints in timber structures.

An International Journal for Engineering and Information Sciences,
5(2): 19–26.
[3] Gečys, T., Gurkšnys, K., Rasiulis, K. (2011). Jungties įklijuotojo

plieninio sriegtojo strypo ir medienos elgsenos tyrimas. Engineering
Structures and Technologies, 3(1): 5–15.
[4] Guan, Z., Rodd, P. D. (1999). Hollow steel dowels — a new application
in semi–rigid timber connections. Engineeri ng Structures, 23 (2001):
110–119.
[5] Guan, Z. W., Kitamori, A., Komatsu, K. (2008). Experimental study

and finite elemet modelling of Japanese “Nuki” joints — Part two:
Racking resistance subjected to different wedge configurations. En-
gineering Structures, 30 (2008): 2041–2049.
[6] Chang, W. S., Hsu, M. F., Chen, C. J. (2004). Estimating rotational

stiffness of timber joints by using fractional factorial experiments
combined with computer simulation. 8th World Conference on Timber
Engineering, Lathi, Finland.
[7] Komatsu, K., Hosokawa, K. (1998). Glulam semi–rigid portal frames

composed of hardwood wedges and metal wares. Proceedings of
the 5th World Conference on Timber Engineering, 2: 246–253,
Montreux, August.
[8] Vašek, M. (2008). Semi rigid timber frame and space structure
connections by glued-in rods. WCTE 2008 Conference Proceedings,
Vol. 1, S. 207, Miyazaki, Japan.
[9] Vašek, M., Vyhnalek, R. (2006). Timber semi rigid frame with
glued–in–rods joints. WCTE 2006 Conference Proceedings, Vol. 1,
S. 275, Portland Oregon State University, Portland.
[10] Kermani, A. (1996). A study of semi–rigid and non–linear behaviour

of nailed joints in timber portal frames. International Journal of Forest
Engineering, 7(2): 17–33.
[11] EN 1995-1-1:(2004). Eurocode 5. Design of timber structures. Part 1–1:

General–Common rules and rules for buildings.
[12] EN 1993-1-8:(2003). Eurocode 3. Design of steel structures. Part 1–8:

Design of joints.
[13] STR 2.05.07:(2005). Medinių konstrukcijų projektavimas. Vilnius,

2005.
[14] Awaludin, A., Smittakorn, W., Hayashikawa, T., Hirai, T. (2007).

M–theta curve of timber connection with various bolt arrange-
ments under monotonic loading. Journal of Structural Engineering,
Vol. 53/A 853–862.
[15] Bruehl, F., Kuhlmann, U., Jorissen, A. (2011). Consideration of plas-

ticity within the design of timber structures due to connection duc-
tility. En gineering Structures, 33 (2011) 3007–3017.
[16] Steiniger, M. (2003). Tensile strength and performance of the

INDUO–heavy timber connector in combination with structural
composite lumber and Douglas fir. Master’s thesis, The University of
British Columbia, Vancouver, Kanada.
Numerical simulations of pseudo-dynamic full-scale
272 test on a three story — one span — three bays steel frame
Numerical simulations of
pseudo-dynamic full-scale test
on a three story — one span —
three bays steel frame
IOAN, Adriana Mirela Abstract

Politehnica University of Tim-
isoara, Romania
The proposed research aims at reducing the repair costs and downtime
of a structure hit by an earthquake, and consequently more rational
design approach in the context of sustainability. These objectives are
to be attained through removable dissipative members and re-cen-
tring capability of the structure, concepts that are to be implemented
in a dual structure, obtained by combining steel eccentrically braced
frames with removable bolted links with moment resisting frames.
Numerical simulations and experimental tests were done in order
to assess the performance and investigate the possibility to replace
bolted links within a realistic structure following significant inelastic
deformations and the practical feasibility of the replacement proce-
dure. Practical solutions regarding order in which bolted links need to
be replaced are proposed. Once the seismic link is removed from the
structure, this system slowly releases the accumulated forces and the
structure comes back to its initial position.
Introduction
Different strategies can be employed in order to reduce damage to

structures under moderate to strong earthquakes. The most radical
solutions are base isolation and various implementations of active and
semi-active structural control. Other strategies rely on supplemental
damping conferred to the structure through various devices based on
viscous, friction, or yielding dampers. More or less efficient for reduc-
tion of structural damage, all these solution have the disadvantage of
requiring specialised knowledge at the design stage and during erec-
tion, need for careful maintenance and high initial cost. Alternatively,
a conventional design can be employed, but with the dissipative
members realised to be removable (e.g. through bolted connections),
test on a three story — one span — three bays steel frame 273
allowing replacement of the dissipative elements damaged as a result

of a moderate to strong earthquake, and reducing the repair costs.
For the structure to be repairable, in addition to constraining inelas-
tic deformations to removable dissipative members, the permanent
(residual) drifts should be eliminated.
The proposed research aims at reducing the repair costs and downtime
of a structure hit by an earthquake, and consequently more rational
design approach in the context of sustainability. These objectives are
to be attained through removable dissipative members and re-cen-
tring capability of the structure, concepts that are to be implemented
in a dual structure, obtained by combining steel eccentrically braced
frames with removable bolted links with moment resisting frames. The
bolted links are intended to provide the energy dissipation capacity
and to be easily replaceable, while the more flexible moment resist-
ing frames would provide the necessary re-centring capability to the
structure. The columns of the structure are to be realised from high
strength steel, in order to keep these members in the elastic range
even under strong seismic input. The validation of the proposed solu-
tion is to be realised through a pseudo-dynamic test of a full-scale
model of a dual eccentrically braced structure at the European Labora-
tory for Structural Assessment (ELSA) facility at JRC in Ispra.
Removable bolted links in eccentrically braced frames
Application of the concept of removable dissipative members to

eccentrically braced frames (EBFs), where links act as dissipative zones,
is presented in Figure 2. The connection of the link to the beam is
realized by a flush end-plate and high-strength friction grip bolts. The
(a) (b)
Figure 1 — General view

(a) and plan layout (b) of the
(a) (b) test structure
main advantage over other dissipative devices is the fact that remov-
able links can be designed using methods readily available to struc-
tural engineers and can be fabricated and erected using procedures
standard to the profession.
Three series of experimental tests on removable link assemblies were

carried out at the “Politehnica” University of Timisoara in order to
determine cyclic performance of bolted links and to check the feasibil-
ity of the removable link solution (Stratan & Dubina 2004 [1], Dubina
et al. 2008 [2], Danku, 2011 [3]). The first series of tests was realised
on isolated links (see Figure 3a–b), while the other two on almost full-
scale model of a single bay and single storey eccentrically braced frame
with removable link (see Figure 3c). Tests on links showed an impor-
tant influence of the connection on the total response of the bolted
link, in terms of stiffness, strength and overall hysteretic response.
Shorter links were found to be suitable for the bolted solution, as plas-
tic deformations were constrained to the link, while the connection
response was almost elastic, allowing for an easy replacement of the
damaged link. At the frame level, the experimental tests showed that
the removable link solution is feasible. Inelastic deformations were
constrained to the removable links alone, all other frame members
and connections remaining in the elastic range. Additionally, it was
possible to replace the damaged removable links with new elements.
Figure 2 — Bolted link concept
Bolted extended end-plate connections for eccentrically braced

frames with link-column connection configuration were previously
investigated experimentally by Ghobarah & Ramadan (1994) [4]. Their
inelastic performance was found to be similar to fully-welded connec-
tions. Balut & Gioncu (2003) [5] suggested a replaceable solution for
links in eccentrically braced frames, and investigated more in detail
two replaceable “dog-bone” solutions in moment-resisting frames:
Figure 3 — Experimental test

on removable bolted link
(a), tests on almost full-scale
frame with bolted links (b)
and test on almost full-scale
model of a single bay and
a single storey eccentrically
braced frame with remov-
able link (c).
(a) (b) (c)
one using an I-beam extended end-plate bolted connection, and

another one consisting of two channels bolted using high-strength
friction-grip bolts to the beam. Mansour et al. (2006) [6] investigated
experimentally replaceable shear links composed of two bolted back-
to-back channels. The links exhibited a ductile behaviour, with stable
hysteretic response, but the connections were found to be critical in
the design of removable links.
Dual-frame re-centring systems
Most of the structures designed to modern code procedures would

experience inelastic deformations even under moderate seismic
action, with permanent (residual) displacements after the earthquake.
Repair is difficult in such cases. Solutions providing self-centring of the
structure exist, but are technically demanded (post-tensioned strands,
shape memory alloy devices, etc.). An alternative solution is the one
that provides re-centring capability (as opposed to self-centring),
through removable dissipative members and dual (rigid-flexible)
structural configuration.
Past numerical and experimental research on removable link assem-

blies and eccentrically braced frames with removable links showed that
this is a feasible solution for earthquake resistant construction. How-
ever, there are critiques to the removable link solution. One concerns
permanent (residual) deformations of the structure after a damaging
earthquake, which can lead to difficulties in replacing removable links.
This issue can be solved by realising a structure as a dual one, by

combining eccentrically braced frames (EBFs) with moment-resisting
frames (MRFs). The elastic response of the flexible subsystem (MRF)
can provide the re-centring capability to the structure, once the links
damaged during an earthquake are removed. For this principle to be
efficient, the flexible subsystem should remain in the elastic range.
A possible way to favour this is to realise some members from high-
strength steel. Additionally, the residual deformations of the link
should allow bolt removal. If these deformations are too large, it is
possible to slowly release the residual stresses and deformations in the
link by removing part of the link through flame cutting prior to remov-
ing the bolts, eliminating thus the residual stresses in the link.
An idealized dual system consisting of two inelastic springs connected

in parallel is shown in Figure 4a. In order to provide the re-centring
capability, the flexible subsystem should be kept in the elastic range
up to the displacements at which the rigid subsystem attains its ulti-
mate plastic deformation capacity. However, a conventional dual sys-
tem that satisfies this condition will not return to the initial position
following deformations into the inelastic range, even if permanent
displacements pD in the dual system are smaller than the ones that
would be obtained in a rigid system alone pr (see Figure 4b). Perma-
nent deformations can be eliminated if the rigid subsystem is realised
to be removable. Upon unloading of the dual system, there is a per-
manent displacement pD, and corresponding residual forces in the
flexible (Fpf) and rigid (Fpr) subsystems. Once removable dissipative
members are dismantled, stiffness and strength of the system is pro-
vided by the flexible subsystem alone (Fpr = 0). If the flexible subsys-
tem is still in the elastic range, it will return the system to the initial
position, implying zero permanent deformations (see Figure 4c).
Figure 4 — Simplified model

of a generalized dual system
(a), and permanent deforma-
tions in a conventional dual
system (b) and in a dual sys-
tem with removable dissipa-
tive members (c)
(a) (b) (c)
Design of full-scale frame to be tested at JRC Ispra
The test structure (Figure 1) has 3 spans of 6 meters, 1 bay of 6 meters

and 3 stories of 3.5 meters each. The main lateral load resisting sys-
tem is composed of the two eccentrically braced frames. The beams
on the other direction have pinned ends and are composite steel-con-
crete beams. The columns are fixed at the base. Columns are made of
HEA240 profiles. Beams in the moment resisting frames are made of
IPE240 profiles. Braces are made of HEB200 profiles and the beams
in the braced frames are HEA240, while links are made of welded H
sections 230 x 170 x 12 x 7 mm (for the lower two levels, where the
dimensions represent: h x b x tf x tw) and 230 x 120 x 12 x 4 mm
(for the upper level).
The columns are fabricated of high strength steel S460, while all the
other elements are fabricated of mild carbon steel S355, except links
that are of S235 steel.
A 4.9 kN / m² dead load and 3.0 kN / m² live load were considered. The
building is analysed for stiff soil conditions (ground type C according
to EN1998-1), characterised by 0.19 g peak ground acceleration and
TC = 0.6 s. A behaviour factor q = 4 (ductility class M) and inter-storey
drift limitation of 0.0075 of the storey height are used.
In order to account for the nonlinear behaviour of the structure, plas-

tic hinges were assigned to the elements in the sections where the
moments and forces are maximum, according to the different modes
of failure of the elements as follows:
–– For the links — shear force-displacement type plastic hinge assigned

at the middle of the bar
–– For the MRF beams — moment-rotation type plastic hinge at the
ends of the bar
–– For the EBF beams and for the columns — moment-rotation type
plastic hinges, accounting for axial force — bending moment inter-
action, assigned at the ends of the bar
–– For the braces — axial force-displacement type plastic hinges at
the middle of the bar
For the links, the peak force was set equal to 2 times the yield force
(Barecchia et al. [7]), while for all the other elements is to 1.25 times
the yield force. Flexibility of the link bolted connections was accounted
for by an equivalent reduced link stiffness.
All of the above plastic hinges were modelled according to FEMA356 [8].

Pushover analyses were made (Figure 5) in order to assess the perfor-
mance of the structure.
Figure 5 — Pushover curve
Numerical simulations of the test
Numerical simulations were done in order to investigate the possibility

to replace bolted links within a realistic structure following significant
inelastic deformations and the practical feasibility of the replacement
procedure. Practical solutions regarding order in which bolted links
need to be replaced were proposed. The link removal procedure uses
a tensioned brace in series with a viscous damper system in order to
allow smooth release of forces locked in the links due to plastic excur-
sions. The link removal will take place after pseudo-dynamic simulation
of seismic loading on the structure, following the steps outlined below:
–– the installation of the bracing systems in the moment resisting

frames of one span;
–– the elimination of the links by flame cutting;
–– the release of brace force through the damper;
–– the elimination of the bracing systems.
Figure 6 — Bracing systems

layout per storey
This procedure will repeat for every storey, beginning from the last
storey towards the first.
Because of the damper’s presence, the two bracing systems per storey
are to be unloaded simultaneously (see Figure 6).
Also an important factor of the link removal procedure is the height-

wise link removal order.
In order to simulate the link removal procedure following a damag-

ing earthquake, the static nonlinear staged construction analysis from
SAP2000 was used. This analysis allows sequential application and
removal of loads and / or parts of the structure. Nonlinear structural
behaviour is possible. There was established, on a 2D model of the
test structure, that the link removal order to be from up towards down,
as described in the following figures:
(a) (b)
(c) (d)
Figure 7 — Height-wise link

removal order (g) (f)
Firstly the structure is loaded with gravitational forces and afterwards

with lateral forces up to a displacement corresponding to the maxi-
mum pushover force (Figure 7a), then it is unloaded (Figure 7b). After
installing the bracing systems at the third level, the link is removed
(Figure 7c) and afterwards, the brace forces are released through braces
with dampers. The procedure is repeated for the second (Figure 7d)
and the first levels (Figure 7e) and, in the end, there are removed
the bracing systems from the first storey (Figure 7f ) and the structure
comes back to its initial position.
Conclusions
Bolted links are the dissipative components in the DUAREM tests

structure. Residual forces and deformations are present in the links
after they have experienced plastic excursions during an earthquake.
Removing a damaged link involves redistribution of residual forces to
other parts of the structure, more precisely from eccentrically braced
frame to moment resisting ones. Numerical simulation of the link
removal order showed that there is negligible redistribution of forces
among storeys. Therefore, the link replacement procedure can be per-
formed on a storey by storey basis, starting from the least loaded to
the most loaded one (from the upper storey toward the lower one).
The technically easiest way to release the forces in links is by flame

cutting the web and flanges of the link. However, there is concern that
this solution might lead to a sudden release of link shear force and
therefore might be dangerous to the operating personnel. Therefore
an improvement of the solution was analysed in the present report.
It employs some bracing systems (a tension brace in series with a
damper) that are installed in one moment resisting bay prior to link
removal. Once the links are flame cut, the forces locked in the links
are smoothly transferred to the temporary bracing system. Once all
links from a storey are removed, all structural components from that
storey are in elastic range of response. Therefore, as the brace forces
are released through braces with dampers, the structure recovers its
initial (plumb) position, becoming free of any locked-in forces.
References
[1] Stratan, A. & Dubina, D. 2004. Bolted links for eccentrically braced
steel frames. Proc. of the Fifth AISC / ECCS International Workshop
“Connections in Steel Structures V. Behaviour, Strength & Design”,
June 3–5, 2004. Ed. F.S.K. Bijlaard, A. M. Gresnigt, G. J. van der
Vegte. Delft University of Technology, The Netherlands, pp. 223–232
[2] Dubina, D., Stratan, A., Dinu, F. 2008. Dual high-strength steel
eccentrically braced frames with removable links. Earthquake Engi-
neering & Structural Dynamics, Vol. 37, No. 15, p. 1703–1720
[3] Danku, G., 2011. Study of the development of plastic hinges in

composite steel-concrete structural members subjected to shear
and / or bending. PhD thesis. Politehnica University of Timisoara
[4] Ghobarah, A. & Ramadan, T. 1994. Bolted link-column joints in eccen-

trically braced frames. Engineering Structures, Vol. 16 No. 1: 33–41
[5] Balut, N. & Gioncu, V. 2003. Suggestion for an improved ‘dogbone’

solution. STESSA 2003, Proc. of the Conf. on Behaviour of Steel
Structures in Seismic Areas, 9–12 June 2003, Naples, Italy, Mazzolani
(ed.), A. A. Balkema Publishers, p. 129–134
[6] Mansour, N., Christopoulos, C., Tremblay, R. 2006. Seismic design

of EBF steel frames using replaceable nonlinear links. STESSA 2006 —
Mazzolani & Wada (eds), Taylor & Francis Group, London, p. 745–750
[7] Barecchia, E., D’Aniello, M., Della Corte, G., Mazzolani, F. M.,
Landolfo, R., (2006). “Steel eccentric braces”.
[8] FEMA 356, 2000. Prestandard and commentary for the seismic
rehabilitation of buildings. Federal Emergency Management Agency.
Micro and macro modeling of internal erosion and scouring with fine particle dynamics 281
Micro and macro modeling of

internal erosion and scouring
with fine particle dynamics
Abstract KONDO, Akihiko

Nagoya Institute of Technology,
Nagoya, Japan
When water flows through a broadly graded soil, internal erosion has
been occurred, and it can lead to narrowing grading and sink hole. MAEDA, Kenichi
We approached this phenomenon by model test and DEM simulation, Nagoya Institute of Technology,
focusing on the local deformation and failure between its occurrence Nagoya, Japan
and internal erosion. The process of internal erosion was modelled by
WOOD, David Muir
progressively removing the finer particles while maintaining the sam- University of Dundee,
ple under constant stresses for some different gradings. Even if stresses Dundee, UK
were hold, microstructures in soil which is network of contact force
chains in soil were changed. Analysis results remind us that the first-
order effect of narrowing grading is the raising of the critical state line
in the compression plane. The performance of the model was shown
and validated. We succeed in developing constitutive modelling of
deformation of soil subjected to internal erosion, based on analysis
results by discrete element modelling of fine particle removal.
Introduction
Recently, damage to the ground caused by caving, which is a sudden

settlement of ground surface including formation of holes in ground,
has occurred not only in Japan, but also throughout many parts of the
world. The damage occurs at various locations, such as the ground
behind revetment due to water level fluctuation and circumferential
areas of deteriorated city infrastructures. From in-site investigations
of areas prone to caving, the formation of a loosened region and a
change of grading distribution to narrow grading mixture (percolation
of fine particles) were confirmed in the circumferential ground. The
particle percolation itself does not cause macroscopic changes such
as the occurrence of sinkholes, but microscopic dynamics such as the
282 Micro and macro modeling of internal erosion and scouring with fine particle dynamics
clogging of fine particles and internal erosion induce local deforma-

tions and eventual deformation. In this paper, the process of erosion
was reproduced in discrete element method analysis, and then a con-
tinuum model based on the DEM simulation results was developed.
DEM modeling of deformation and failure of soil by

internal erosion
Numerical simulation procedure and tests

Two-dimensional DEM employing PFC2D was used for the analysis
described in this chapter (Wood and Maeda, 2008; Wood, Maeda and
Nukudani, 2010). The main parameters used in two-dimensional DEM
were detailed in a previous paper (Maeda and Hirabayashi, 2006). In
the present study, only circular particles were used and four gradings
were prepared, as shown in Fig. 1(a). The grain size ratio RD was used
as a grain size index, where Dmin was changed under Dmax = 100 mm
(RD = 2, 5, 10, and 20). A biaxial compression test was performed
under a nongravity condition. By moving four wall elements (non-
friction), the stress and strain of the specimen were controlled. In
addition, all of the stress was effective stress. The direction of the
maximum principal stress was set as y direction, and normal-strain εxx
and εyy, volumetric strain εv, mean principal stress m, and maximum
shear stress m were used.
We conducted three kinds of tests according to external actions as

shown in Figure 2. While maintaining the stress state (mean principal
stress) as a constant, current particles with minimum diameter were
forcibly removed from each of the four gradings shown in Figure 1(a).
Due to the generation of the internal unbalance force following the

removal, macroscopic deformation was induced. The removal was
considered to be the simulation of the particle percolation, and the
removal was repeated until the diameter of the removed particles
reached 5% of the original particle diameter or εyy > 25%.
Figure 1 — Samples used and

DEM analysis appearance;
(a) gradings of samples used
for DEM (solid curves) and
grading reached by removal
of particles; (b) appearance
of stress chain caused by one
particle removal.
(a) (b)
Figure 2 — Test cases and con-

dition; monotonic loading,
removal test of fine particle un-
der constant stress and reload-
ing test after removal test.
Monotonic loading test — intact shearing —

To understand only the effect of worsened grading distribution, behav-
iors of deformation and deformation were investigated using the vol-
ume change of the entire specimen consisting of a material without
internal erosion, and the specific volume v (= 1 + e). Figure 3 shows
a response at the time of monotonic loading on the densest material.
This figure shows that when the compaction state (Dr) is the same, the
same behavior is observed regardless of the grading. As shown in this
figure, the same flow rule can be used.
Figure 3 — Stress-strain-dila-
tancy responses for monot-
onic loading test with different
gradings.
Figure 4 — Critical state line

(CSL): (a) CSL for RD = 20,
(b) grading effect on CSLs.
(a) (b)
Fig. 4(a) shows a change in the specific volume when the most dense
and loose materials are sheared at various values of mean normal
stress (m). As shown in this figure, when m is the same, the critical
state is reached at a certain specific volume regardless of the density.
The straight line connecting m exhibits the critical state of the specific
volume, and the specific volume at the critical state is regulated accord-
ing to the material. Figure 4(b) shows the critical state lines of samples
with different grading. The specific volume is smaller with a larger RD.
This is why under the same compaction energy, a material is com-
pacted more easily (Dr becomes higher) when the grading is superior.
Removal test — internal erosion —

Figures 5 (a, b) demonstrate axial deformation and dilatancy behav-
ior, respectively, when the removal test was performed. These figures
show only the results at RD = 10 in the dense state. In these figures, the
deformation volume generated when a particle was removed is exhib-
ited by the distance between dots. When the number of removed parti-
cles increased and the grading was worsened (a decrease in RD), axial
strain developed. When the stress ratio was high (above 0.271), a defor-
mation larger than 10% was generated that resulted in deformation.
Figure 6(a) shows the results obtained by the monotonic loading and
removal tests in the relationship between the specific volume v and RD.
In this figure, the curves to complement the relationship between the
critical specific volume vcs (m = 0.1 MPa) and RD at each grading are
interpolated. As shown in this figure, the void ratio of the specimen
increased as RD worsened. The reason for this was that although the void
equal to the volume of the removed particle increased, the compression
deformation due to the structure change following the removal was
smaller than the volume in question. A series of removal tests revealed
that deformation occurred when the current specific volume v of the
specimen approached the critical specific volume vcs, which had been
altered due to the grading change. Figure 6(b) shows a conception dia-
gram of the deformation and deformation behaviors shown in Figure 6(a).
(a) (b)
Figure 5 — Removal tests

(initial grading RD = 10; mean
stress 0.1 MPa): (a) stress
ratio; (b) strain paths (dashed
line, monotonic shearing;
solid lines, particle removal).
(a)
Figure 6 — Change in specific

volume by removal of fine
particles: (a) specific volume (b)
change of the specimen and in
the critical state following the
change in the grading under a
constant stress ratio removal
tests (initial grading RD = 10;
mean stress 0.1 MPa); (b) a
conception diagram of Fig. 6(a).
Reloading test — shearing after erosion —

Figures 9(a, b) show the axial deformation and the dilatancy behavior
of the specimen in the shear after the removal test. The specimen
after the removal reached the critical state without exerting the peak
strength and dilatancy that could be exerted. The stress ratio in the
critical state during the reloading phase was the same as that in the
critical state before the removal test. It is known that deformation
and destruction behaviors of soil depend on the levels of density and
stress, and the critical state can be systematically understood using
these behaviors. For the critical state, parameter  = ν – νcs is known
to be one of the important indices. This is expressed through the dif-
ference between the current specific volume ν and the critical specific
volume νcs. The state parameter is an index for the compaction state
of soil that is not expressed by the level of the void ratio, but instead
by the relative void level against the critical state. When  > 0 (ν > νcs),
soil is evaluated to be in the loose state. When  < 0 (ν < νcs), soil is
evaluated to be in the dense state. When  = 0 (ν = νcs), soil is evalu-
ated to be in the critical state.
(a)
(b)
Figure 7 — Reloading tests

(initial grading RD = 10; mean
stress 0.1MPa): (a) stress ratio;
(b) strain paths.
Figure 8 — Peak strength

and state parameter with and
without particle removal.
Continuum modeling of deformation and failure of soil

by internal erosion
Formulation of mechanism
We developed a continuum model to express the deformation-failure
behaviors of soil due to internal erosion (Wood and Maeda, 2008;
Wood, Maeda and Nukudani, 2010). Since a model was constructed
based on a change in the grading distribution due to the fine parti-
cle percolation, the change in the grading must be clearly examined
without depending on the initial conditions. The grading index IG is
expressed by the area ratio of the current grading (ABC in Fig. 9) to
the limiting grading (ABD in Fig. 9). In addition, IG is 0 in the grad-
ing distribution consisting of a single particle diameter, and it is 1 in
the limiting grading. Since IG is obtained based on the area ratio, it
is greatly affected by the maximum particle diameter Dmax, the mini-
mum particle diameter Dmin, and RD, which is the ratio between them.
The particle removal causes an increase in the specific volume v, volu-

metric shrinkage due to the confining pressure, and worsening of the
grading distribution ( IG) due to material defect. These changes in the
material itself interact with each other. To formulate these changes,
the change in the specific volume is examined, focusing on the specific
volume in the critical state (vCS) using the state parameter . Here, 
must be changed from A to B. The amount of change () can be
expressed as the sum of the specific volume increment (v), volumetric
shrinkage due to the confining pressure (v1), and volumetric shrinkage
due to the second plastic mechanism (v2) of the volume of the perco-
lated particle. A previous report2) described the above matter in detail.
Figure 9 — Definition of grad-

ing index IG (as ratio of areas
ABC and ABD) and the shift
operation of critical state and
change in specific volume
with particle removal.
(a) (b)
Continuum model performance and verification

Figure 11 shows the axial deformation and dilatancy behaviors when
IG was input. The generated strain was larger when a particle was
removed in the state where the stress ratio was higher. Therefore, these
figures can display an increase in the generated strain with an increase
in the number of removed particles. Figure 12 illustrates behaviors
when reloading was performed on the specimen in which the grading
distribution had been changed to a particle diameter of 5%. We can
see a reduction of the peak strength and an increase in the compres-
sion tendency in sample subjected to erosion as shown in Figure 7.
(a)
Figure 11 — Removal test

simulated by continuum
model for erosion: (a) stress
(b) ratio; (b) strain paths.
(a)
Figure 12 — Reloading test

simulated by continuum model
for shearing after erosion:
(b) (a) stress ratio; (b) strain paths.
Conclusion
We succeeded in developing constitutive modelling of deformation

of soil subjected to internal erosion, using analysis results obtained
by DEM modelling of fine particle removal. As for the modelling con-
cept, it can be reproduced by transferring of critical state stress ratio
based on the behaviour of state parameter based on the traditional
continuum scheme.
References
[1] Maeda, K. & Hirabayashi, H. (2006). Influence of grain properties

on macro mechanical behaviors of granular media by DEM, Journal
of Applied Mechanics JSCE, 9: 623–630
[2] Maeda, K., Miura, K. & Toki, S. (1995). Mechanical properties of

elliptic microstructure formed in granular materials. Soils and Foun-
dations, 35(2): 1–13.
[3] Maqsood, S. T., Schwarz, J. (2008). Analysis of building damage during

the 8th October, 2005 Earthquake in Pakistan. Seismological Research
Letters, 79(2), S. 163–177
[4] Kenichi MAEDA et al. (2010). — Stress-chain based micromechanics

of sand with grain shape effect, Granular Matter, 12(5), 499–505.
[5] Wood, D. M. & Maeda, K. (2008). — Changing grading of soil: effect

on critical states, Acta Geotechnica 3: 3–14.
[6] Wood, D. M., Maeda, K. & Nukudani, E. 2010. — Modelling mecha-

nical consequences of erosion, Geotechnique 60(6): 447–457.
Reports
Project 1 — System Identification,

Model Updating and Simulation
ZABEL, Volkmar Introduction

Bauhaus Universität, Weimar
The aim of this project was to evaluate the dynamic behaviour of the
CABBOI, Alessandro
Pöppelmannbrücke footbridge in order to assess the levels of the
GALVAN, Giulio human induced vibrations and the possible effects on the comfort for
the end users. The first step was to create FE models of the structure
GEČYS, Tomas to investigate the modal properties, which have been subsequently
MALVEIRO, Joel
verified by ambient and forced vibration tests. By processing the first,
it was possible to identify mode shapes and eigenfrequencies of the
MONTENEGRO, Pedro structure, while using the forced vibration data, the damping proper-
ties of the footbridge have been calculated. Model updating has then
ROCHA, João Miguel
been performed both with numerical methods and manual tuning of
TRIBUTSCH, Alexander the unknown boundary conditions as to achieve a higher degree of
precision of the numerical models which have finally been used to
VOUKIA, Dimitra assess the levels of human induced vibrations according to the guide-
lines SETRA and HiVoSS. Furthermore, a TMD was designed and a
YAMAGUCHI, Mayako
parametrical study on the mass ratio of the device was conducted.
Case study
The considered footbridge is Pöppelmannbrücke and is located in the

city of Grimma, Saxony. It was first constructed in 1719 and during its his-
tory was destroyed and rebuilt several times. During a flood in August
2002 the structure reported severe damage yet again therefore the
city of Grimma decided to rebuild it. The new structure, a steel arch
bridge spanning 65 meters, was attached to the historical remainings.
The construction started in 2009 and the footbridge will be opened to
the public in August 2012. Field vibration tests were performed in the
earlier stages of design and construction, however a tuned mass damper
was recently installed and further vibration testing was required.
Project 1 — System Identification, Model Updating and Simulation 291
Figure 1 — The flood in 2002
Figure 2 — The bridge in

August 2012
Numerical models
Three numerical models of the bridge were created in order to repro-

duce the geometry and the behaviour of the structure. The softwares
used were ANSYS, Straus7 / Strand7 and SLANG. The models include
the archs, the main girders, the deck and the longitudinal stiffeners as
well as the cross girders. 12 DOFs beam elements were employed for
the arches, the main girders and the cross girder. The presence of the
longitudinal stiffener was simulated with an increase of the bending
stiffness of the deck, defined by 20 DOFs shell elements.
The boundary conditions at the end of the deck are the main source of
uncertainty: the end of the structure cantilevers out of the bearings and
there is also some uncertainty in the bearing stiffnesses. Therefore, it was
decided to put translational and rotational springs at the end of the bridge
deck which in later stages of the investigation have been calibrated.
Figure 3 — The FE model

(Straus 7)
Figure 4 — Accelerometer
and measurement equipment
Experimental tests
6 bi-axial accelerometers were utilized to measure the acceleration

response of the structure in vertical and lateral direction. A sampling
frequency of 204.8 Hz was selected. An accelerometer and the meas-
urement equipment is depicted in Figure 4.
Five configurations of sensor locations were considered, see Table 1

and Figure 5. Configuration 1 to 4 were chosen to identify the mode
shapes of the bridge at 18 nodes. Therefore, two accelerometers were
placed at reference nodes 4 and 15, while the other four remaining
sensors were used as roving sensors in four configurations. Configura-
tion 5 includes only nodes at or near the center of the bridge.
Sensor Config 1 Config 2 Config 3 Config 4 Config 5
REF1 4 4 4 4 4
REF2 15 15 15 15 15
MP-1 3 5 1 2 5
Table 1 — node number of MP-2 8 9 7 6 6
sensor location for five con- MP-3 10 11 12 13 13
figurations MP-4 14 16 17 18 14
Figure 5 — sketch of node

numbers along the bridge
deck
Measurements
Measurements 1 to 4 correspond to ambient vibration test with sensor

configuration 1 to 4. Measurements 5 to 30 were conducted with sen-
sor configuration 5. In contrast to the first four measurements, where
only ambient vibrations with released TMD had been measured, the
following include bouncing, jogging and ambient tests with released
or blocked TMD, respectively.
Only a selected number of measurements were utilized for the further

investigations, which are presented in Table 2.
Measurement Configuration Duration type TMD

[s]
1 1 600 ambient released

5 5 80 bouncing / free decay released
6 5 60 bouncing / free decay released
26 5 600 ambient blocked
Table 2 — Node number of
28 5 150 bouncing / free decay blocked sensor location for five con-
30 5 60 bouncing / free decay released figurations
System identification
SSI based identification of modal properties

The data collected during the ambient vibration tests were processed
with MATLAB and the MACEC toolbox. A high pass filter was applied
to the selected measurements and the data points were decimated by a
factor of 4 in order to improve the signal quality and avoid aliasing issues.
The signal of data set from 1 to 4 and 15 was then analysed by the use
of Stochastic Subspace Identification [4] and for every data set the
eigenfrequencies and the corresponding mode shapes were identified
in the stabilization plot.
The MACEC toolbox allows for combination of subsequent identifica-

tions in order to better identify the mode shapes of the structure and to
distinguish structural modes from numerical ones. The identification was
carried out when the TMD was already installed and released, and as
the data available for blocked TMD condition is scarce and of poor qual-
ity, all of the identification process refers to the structure with the TMD.
Project 1 — System Identification, Model Updating and Simulation
294
Mode Frequency Damping Mode Shape Modal Phase

[Hz] [%] Collinearity
1 2.5993 2.3340 V bending 0.96288

2 2.6993 1.6227 V bending 0.99911
3 3.0949 1.2028 H bending and twisting 0.96029
4 3.9539 0.48701 V bending 0.74309
5 4.0526 0.26181 twisting 0.99782
6 5.7934 0.40022 twisting 0.99758
7 6.6590 0.72283 V bending 0.99832
Table 3 — Model identifica- 8 7.7364 0.54935 twisting 0.99409
tion by SSI technique 9 8.6178 0.73039 bending 0.99409
Validation of identified damping ratios by free decay tests.

Bouncing test were performed to assess the damping properties of the
foot bridge and to validate the outcomes of the system identification
procedure. A group of 10 to 12 people was synchronizing with the
expected first natural frequency (2.6 Hz) to excite the structure to a
clearly measurable amplitude response. When an appropriate level of
vibration was accomplished, bouncing was stopped in order to meas-
ure the free decay.
Accelerations were recorded both, in vertical and lateral direction, at

6 nodes located near the centre of the bridge (4, 5, 6, 13, 14 and 15).
A Butterworth filter of order 2 with cut-off frequencies at 1.5 and 3.5 Hz
was applied to clean the data for the following identification. Damping
was estimated by the Logarithmic Decrement method considering the
first 7 to 13 peaks of the free decay and averaging of the identified values.
Four measurements were considered for the evaluation, listed in

Table 4. Only vertical records of nodes 5 and 14 were taken into
account to minimize disturbances due to the higher, anti-symmetric
modes. Mean values of the identified dimensionless damping coef-
ficient and frequency are also provided in Table 4.
Measurement No. of TMD Identified Identified

people damping frequency
bouncing coefficient [Hz]
[%]
5 11 released 2.9 2.7

6 11 released 2.7 2.7
30 12 released 2.8 2.7
Table 4 — Free decay tests 28 12 blocked 0.3 2.6
295
Figure 6 — free decay,

comparison of response with
blocked and released
The exemplary response of the discussed four measurements is

shown in Figure 6.
Model updating
Description of the models

In most cases, the modal parameters obtained from the numerical mod-
els are not in correspondence with the experimental values. Charac-
teristic properties of the structures, such as modulus of elasticity of the
materials or properties related to the supports may need to be adjusted.
In this sense, the use of calibration techniques of numerical models based
on experimental results has shown strong growth, in order to minimize
the differences between numerical and experimental results. It is an
iterative process that consists in the minimization of one objective func-
tion through the variation of various properties of the numerical model.
The objective function, shown in equation [1], depends on the numerical
and experimental natural frequencies and also the coordinates for each
mode of vibration according to the experimental measurement points.
N freq num exp 2 N freq N num exp 2
fi fi u ij u ij
J x (1)
exp 2 exp
i 1 fi i 1j 1 u ij
The calibration of the numerical models was performed using the Opti-
mTool from the MatLab calibration toolbox. The function “minsearch”
finds the minimum of an objective function, starting at an initial esti-
mate of the numerical parameters. For the calibration of the numerical
models only the stiffness of the supports were considered as variables
to change. Figure 7 shows the difference between the numerical and
experimental natural frequencies for both models.
between measured and
numerical frequencies
According to Figure 7 the frequency of the first mode of vibration

obtained in all the numerical models, before the calibration process,
are closed to the one measured in the experimental tests. After the
calibration the first frequency of all the three models converged to
the measured result.
For the others modes of vibration, there are a huge variability in the
frequencies before the calibration, especially for the third and fourth
modes (second bending mode and torsion mode). However, the cali-
bration process allowed a better approximation between the numeri-
cal and experimental results for all the other frequencies, except for the
third and fourth modes in the ANSYS model. This could be explained
by errors in the bearing and geometrical properties of the model that
couldn’t be fixed during the calibration process.
Figure 8 shows the first 4 modes of vibration of the numerical model

developed in the commercial software Straus7, which was after used
in the assessment of human induced vibration levels.
Assessment of the vibration levels
Guidelines
The deck vibration level was assessed by two different methodologies:
the SETRA guideline and the HiVoSS guideline. SETRA is the French
guideline to the assessment of pedestrian induced vibrations on foot-
bridges. HiVoSS is an European guideline for the assessment of human
induced vibration on steel structures.
Mode 1 Mode 1
Mode 4 Mode 4
Mode 3 Mode 3
Figure 8 — Modes of vibra-

tion: (a) before and (b) after Mode 4 Mode 4
the calibration process (a) (b)
Both methodologies introduce a few basic assumptions in order to

simplify the process of obtaining the deck accelerations. In both cases
the random number of pedestrians, N, is substituted by an equiva-
lent number of uniform pedestrians, Neq. These methodologies assess
the dynamic behaviour for each mode separately and only take into
account frequencies below 5 Hz.
Regarding the pedestrian loading the two methodologies have some

differences. SETRA defines three pedestrian densities: 0.5 p / m²,
0.8 p / m² and 1 p / m², while HiVoSS defines 5 different pedestrian
densities: 15 p / deck, 0.2 p / m², 0.5 p / m², 1 p / m² and 1.5 p / m².
The equivalent number of pedestrians is function of the deck’s surface

area, the pedestrian density and the damping:
N eq 10.8 j N,d 1 (2)
N eq 1.85 N , d 1 (3)
The equivalent loading is uniformly distributed and depends on the

equivalent number of pedestrians, Neq, the weight of the pedestrian,
G, the dynamic loading factor, , and the reduction factor, .
N eq
q eq eh G eh fj (4)
S
The two guidelines have also some slight differences regarding the
reduction factor which can lead to some differences in the accelera-
tion levels on the deck, which is shown in Figure 9.
Figure 9 — Reduction factor

for vertical modes
Furthermore, SETRA also requires the calculation of the natural frequen-

cies accounting for the pedestrians mass whereas HiVoSS only requires this
calculation if the pedestrians mass is higher than 5% of the modal mass.
The maximum acceleration for each mode is determined according

to Eq. (5):
qeq S i j
üj , emax max j,p i (5)
2 j
298
Simulations
The vibration assessment was performed only for the updated model.
Several damping scenarios were analysed including the theoretical and
experimental results. In the first scenario the structural damping was
considered to be constant and the damping used was 0.4%, which
corresponds to the mean value suggested by the guidelines for steel
structures. In Figure 10 the maximum accelerations obtained for each
mode and for the several pedestrian densities are presented.
The second and third scenarios took into account the damping val-
ues obtained in the experimental tests. The second scenario did not
include the additional damping provided by the TMD, whereas in the
third scenario this was taken into consideration. The maximum accel-
erations for the second scenario are presented in Figure 11.
SETRA HiVoSS
1° mode
3° mode
4° mode
Figure 10 — Vertical accelera-

tion assessment for  = 0.4%
299
SETRA HiVoSS
1° mode
3° mode
4° mode

tion assessment for experi-
mental damping
Figure 12 illustrates the maximum acceleration obtained for the third

scenario where the TMD was active.
Discussion
In general the increase of the pedestrian density leads to higher acceler-
ations of the bridge deck. However, there are some exceptions which
can be explained by the variation of the reduction factor. For the current
case study this occurs for the analysis of the first mode using the HiVoSS
guideline. This is due to the fact that for frequencies ranging from 2.3 Hz
to 2.5 Hz this guideline suggests the use of a reduction factor of 0.
The reduction factor also explains the differences obtained by the two
guidelines. Although the functions are quite similar the acceleration
level can be significantly different. This shows that the acceleration
level on the bridge deck is highly dependent on this parameter.
SETRA HiVoSS
1° mode
3° mode
4° mode

tion assessment accounting
for TMD damping
Comparing the first and second scenarios some differences can be

noticed. The identified damping ratio during the experimental tests was
lower than the mean value suggested in the guidelines (0.29% vs 0.4%).
This explains the higher accelerations obtained for the second scenario.
For the third scenario the damping was estimated for each mode sepa-
rately and took into account the additional damping due to the TMD. An
important decrease of the acceleration levels can be seen particularly for
the first mode. This confirms the effectiveness of the TMD on the con-
trol of the deck vibrations. For this scenario some unacceptable accel-
erations can be observed for the 4th mode. This can be explained by
the experimentally identified damping ratio for this mode ( = 0,26%).
This value seems to be too low and it might not be accurate since the
torsional mode was not well excited during the experimental tests.
Tuned mass damper (TMD)
Design
The main aim of a tuned mass damper design and installation to this
bridge is to reduce the 1st mode vibration response. Figure 13 illus-
trates the general steps of design of a tuned mass damper (TMD) [1].
Figure 13 — General steps for

a TMD design
The amplitude ratio x / xstat of the main mass as a function of the four
essential variables , c / cc, f, g is given by
c 2 2
2 c gf g2 f2
x c
x stat c 2 2 2
2 c gf g2 1 g2 f 2 g2 g2 1 g2 f2
c
In equation 5:
 = m / M — ratio between absorber (TMD) and main system mass,
CC — critical damping,
f — frequency ratio (natural frequencies),
g — forced frequency ratio,
x — dynamic deflection of system,
xstat — static deflection of system.
Application
The optimal attachment position corresponding to the maximum of
the 1st mode shape is located in the middle of the bridge. A non-
optimal position leads to a decrease of the effective TMD mass ratio.
Figure 14 shows the results of tuned mass damper design. While

designing the TMD, the parametric analysis with different mass of
TMD was made to see the influence of TMD mass to x / xstat ratio. A
range of TMD mass from 1 t to 5 t, including the actual design of 2.3 t,
was considered in this study.
Figure 14 — A tuned mass

damper‘s parametric analysis
results
From Figure 14 it can be seen that a higher mass results in a decreased

response. The practical engineer has to decide on the actual mass
according to the required response reduction.
References
[1] Den Hartog, J. P. (1985). Mechanical Vibrations, Dover Publications,

Mineola, New York.
[2] HiVoSS — Human Induced Vibrations of Steel Structures (2008).
[3] Setra — Guide to assessing pedestrian induced vibratory behavior of

footbridges (2006).
[4] Van Overschee, P. De Moor , B. Subspace identification for linear

systems: theory, implementation and applications, Kluwer Academic
Publishers, Boston, London, Dordrecht (1996).
Project 2 — Soil-Structure Interaction Simulation and Experimental Validation 303
Project 2 — Soil-Structure
Interaction Simulation and
Experimental Validation
WUTTKE, Frank
BUW, Weimar, Germany
Abstract
KAFLE, Binod
The project deals with the study of soil structure interaction (SSI)
through experimental study and the numerical validation by the use ANDREEV, Stoyan
of available finite element code and the Macroelement, incorporating, Risk Engineering LTD.
Sofia, Bulgaria
interaction between the structural forces, Elsto-Plastic soil behaviour
and the interface between the soil and the foundation. The response HOANG, Vu
of Macroelement consisting of few constitutive parameters incorpo- Waseda University Tokyo,
rating SSI, which is able to predict the prognosis of the displacement, Japan
serves as the realistic tool for the study of SSI.
KONDO, Akihiko
Nagoya Institute of
Technology, Japan
Introduction
KRAUS, Ivan
University of Osijek, Croatia
Soil Structure Interaction (SSI) is referred as the phenomenon in which
the motion of the soil and the motion of the structure are interde- KURIMOTO, Yuhei
pendent and both of them are influenced by each other. According Nagoya Institute of
to [1] SSI is defined as ‘The process in which the response of the soil Technology, Japan
influences the motion of the structure and the response of the structure
MAEDA, Keita
influence the motion of the soil.’ This interdependency in the response Waseda University Tokyo,
of the soil and structure and the role played by the interface, between Japan
the foundation and the soil, brings intrinsic complexity to solve the
MARIN, Alexandru
problem. This complex behavior of SSI can be conceived by the use
Swiss Federal Institute of
of elasto-plastic strain hardening Macroelement constituting coupled Technology Zurich,
system of super structure, soil half space and the boundary between Switzerland
them. The use of granular model likes Macroelement consisting of
few constitutive parameters and capable of delivering high ended NAKAYA, Hiroki
Waseda University, Tokyo
results, reduces the ambiguity of model response and the complexity
in the analysis of the system under consideration. (Fig. 1), shows the
pictorial representation of Macroelement, consisting of forces from
structure and the arbitrary elasto-plastic region of soil under consid-
eration of Macroelement.
304 Project 2 — Soil-Structure Interaction Simulation and Experimental Validation
Figure 1 — Pictorial represen-

tation of Macroelement of
shallow footing
Experimental Studies
Static bearing capacity test was carried out in a box of size

1 m x 0.5 m x 0.5 m in dimension. Oven dry Hostun sand was filled in
layers and compacted to obtain the void ratio of 0.76, with the height
of 0.42 m. The surface was then leveled, and a rough rigid square,
steel footing was placed over the sample, (Fig. 2). Displacement con-
trolled loading at the rate of 0.0025 mm / s was applied on the foot-
ing, until the failure load was reached.
Figure 2 — Footing under

centric vertical loading
Figure 3 — Load displacement

behaviour of Footing
The stiffness of the soil decreases with the increase in time of loading
at the constant rate of loading. This variation of the stiffness during
the test was recorded by the use of bender elements, embedded in
the soil at different depth (Fig. 4.), from the surface. By observing
the change of shear-wave velocity, the change in stiffness was calcu-
lated at different depths, at different time steps. (Fig. 5), shows the
transmitted and received signal from the bender element. The time of
first arrival of shear wave velocity across the footing was calculated to
obtain the instantaneous stiffness from (Eq. 1 and Eq. 2).
2
(1) G max p Vs
(2) E max 2 G max 1 v

Figure 4 — Placement of
bender-elements below the
footing at different depths
Figure 5 — Received and

transmitted signal of Bender
Element
Numerical Studies
Numerical validation of the experimental results, obtained from the

test was done, by the use of available finite element code. PLAXIS was
used to simulate the test results by the use of two different constitu-
tive models, Mohr-Coulomb and the Hardening-soil Small strain Soil
Model. 15 nodded triangular element for plain strain case was used to
model the behaviour (Fig. 6 and 7).
Figure 6 — Mesh layout for

PLAXIS Simulation, Mohr-
Coulomb Soil Model
Figure 7 — Mesh layout for

PLAXIS Simulation — Hard-
ening-Soil Small Strains Soil
Model
Constitutive Modelling
A Macroelement from [2], constituting structural component, the inter-

face between soil and foundation, the interaction between the differ-
ent forces with inclusion of single surface hardening soil model is used
herein. The parabolic failure criteria taking account of both inclined and
eccentric forces in three-dimensional space, is as follows, [2] (Fig. 8);
h2 m2 2
(3) f Q , pc 1 0
pc
Figure 8 — 3-D failure enve-

lope in V, H, M / B space
By the introduction of new function called plastic potential denoted by g,

the expression for the increment of plastic displacement for a given state
of forces(Q) and the given incremental displacement is obtained as;
g
(4) dq p
Q
Figure 9 — Change in stiffness

with the increase in displace-
ment during the monotonic
loading of footing.
Results
The numerical response of the load deformation behaviour of rigid square

footing was obtained by the use of finite element code excluding the
effect of SSI and the Macroelement which includes the effect of SSI. The
degradation of stiffness at the depth of 10 cm from the surface of footing,
during the loading was obtained by the use of bender-element (Fig. 9).
Numerical response from PLAXIS by the use of Mohr-Coulomb Soil

Model (Fig. 10), was obtained at three iteration steps with the final value
of stiffness E = 8785.3 kPa. Whereas, by the use of Hardening-Soil Small
Strain Soil Model, numerical response for load deformation behaviour,
was obtained with the final value of stiffness, E = 13400 kPa, (Fig. 11).
The numerical response of the Macroelement which includes the effect

of SSI provides the prognosis of the displacement under static load-
ing from small to large displacement, including the failure. (Fig. 12),
shows the comparative study of numerical response of load deforma-
tion behaviour, with the experimental result.
Figure 10 — Deformation of
soil obtained from PLAXIS
Simulation, Mohr-Coulomb
Soil Model
Figure 11 — Deformation of
soil obtained from PLAXIS
Simulation — Hardening-Soil
Small Strains Soil Model
Figure 12 — Numerical
response of load deformation
behaviour.
Conclusion
The effect of SSI on the numerical response was studied by the use
of single surface hardening Macroelement. The comparative study of
the model response by the use of PLAXIS for two different consti-
tutive models, Mohr-Coulomb Soil Model and Hardening-Soil Small
Strain Soil Model showed that the numerical response can only be
obtained for small deformation. For the large deformation the model
response is conservative. On the other hand Macroelement is able to
predict the prognosis of deformation in the complete range of loading
including failure. The pronounced decrease in soil stiffness was also
observed by measuring the shear wave velocity obtained by the use of
bender elements.
References
[1] Kramer, S. L. (1999), “Geotechnical Earthquake Engineering” Pearson

Education, Inc.
[2] Nova, R. and Montrasio, L. (1991). “Settlement of shallow founda-

tions on sand.” Geotechnique, 41(2), 243–256.
Project 3 — Evaluation of seismic performance of a template
design RC school building before and after strengthening 309
Project 3 — Evaluation of seismic

performance of a template design
RC school building before and
after strengthening
Abstract Abrahamczyk, Lars

The aim of the project was to introduce the participants into advanced
Genes, M.Cemal
requests in the field of earthquake engineering. The studies were con- Mustafa Kemal University
centrated on a template design RC school building (Figure 1), which
revealed low seismic resistant in the past earthquakes in Turkey. There- ANDIC, Halil Ibrahim
fore the Turkish Government started after the damaging events in 1999 METU, Ankara, Turkey
a campaign to investigate and strengthen these buildings all over Turkey. BONEV, Todor
BOSHNAKOV, Simeon
In the frame of these campaign one typical school building, with UACEG, Sofia, Bulgaria
quite similar ground plans to the school building which failed during
ELYAMANI, Ahmed
the Bingöl Earthquake in 2003 could be instrumentally investigated
Technical University of Cata-
before and after strengthening. lonia, Spain
The participants of the project assessed the performance of the struc- GRUBISIC, Marin
GFOS, Osijek Croatia
ture before and after the applied strengthening measures on the basis
of measured building response data and evaluate the earthquake HAVLIKOVÁ, Ivana
resistant design. They carried out damage prognosis for the given seis- Brno University of Technology,
mic action and compared it with the real occurred damage after the Czech Republic
Bingöl EQ by using modern software tools.
UDREA, André
Introduction UGURLU, Tolga

YELEN, Burcu
MKU, Hatay, Turkey
School buildings belong to the building class of higher priority accord-
ing to the common code requirements, because of their use as meet- STILIANOPOULOU, Lily
ing point and shelter in the immediate aftermath of a disaster as well NTUA, Athens, Greece
as their high occupancy rate.
TSONEV, Aleksander
UACEG, Sofia, Bulgaria
The major part of the school buildings (80%) were constructed before
1997, which means according to the requirements of the 1972 earth- ZIMMO, Iyad
quake code and under the consideration of obsolete seismic demands. Birzeit University, Palestine
In many cases rules of earthquake resistant design were not applied.
310 design RC school building before and after strengthening
The devastating Marmara Earthquake (17th August 1999) and Bingöl

Earthquake (1st May, 2003) caused heavily damages and collapses on a
serious number of school buildings. Therefore the Turkish Government
initiated a programme for the systematic inspection of school and hos-
pital buildings in Turkey together with the application of strengthen-
ing measures for governmental buildings with poor performance.
In the frame of a project the project partner of the summer course

Mustafa Kemal University could accompany a strengthen measure of a
school building by dynamic investigation.
Study Object
General Information
The considered school building is one of three school buildings in
Hatay (Turkey), which could be dynamically investigated before and after
strengthening. It is a 3-story RC frame structure building and was con-
structed in 1997 according to the 1972 earthquake design code (see Fig. 1).
Figure 2 show the modification of the ground plan by the replace-

ment of non-structural masonry infill walls with reinforced concrete
structural walls (red walls), which is a common strengthening tech-
nique in Turkey to increase the building capacity and to decrease the
maximum displacements. The chosen school building is thus rep-
resentative for a large number of school buildings in whole Turkey
before and after strengthening.
Evaluation of the structural system
A first evaluation of the study object before and after strengthening

was carried out on the basis of regularity and construction details
check. The general layout obeys with the regularity criteria of an
earthquake resistant designed structure which will be confirmed by
the check of the location of the centre of mass (red point) and rigidity
(blue point) in Figure 2.
The location of the centre of mass and rigidity a translational behav-

iour of the building in both directions can be expected under seismic
action due to the quite small distances between as well as the location
Figure 1 — Front view of the

study object [1]
of both points. That means in case of the unstrengthen structure only

less additional forces can be expected due to torsional behaviour. Due
to the retrofitting of the structure the centre of rigidity was shifted,
thus a larger distance occurred and the influence of torsional motion
will become higher.
The reinforcement details are in general of poor quality, which was

also the highest motivation for the retrofitting.
Observed building performance after the Bingöl EQ
The magnitude 6.6 Bingöl (Türkiye) earthquake on May 1, 2003

caused severe damages and collapse to school buildings (see Figure 3).
The comparison of the general layouts from the damage structures
and the study object indicates the representativeness of the studied
object and provides the opportunity to adopt the experiences from
the Bingöl earthquake to the study area Antakya. Thus, the results
of the damage prognosis can be compared with the observed dam-
ages on school building with a nearly identical (template) design for a
known seismic action.
a) Before strengthening
Figure 2 — Ground plan of

b) After strengthening the study object
Figure 3 — Damage of a
school building caused by the
1st May Bingöl earthquake [2]
Experimental tests
Building response measurements

Mustafa Kemal University carried out building response measurements
before and after strengthening of the school building in 2011. The
building was equipped with five triaxial velocity sensors Type MS2004+
and the corresponding recorder Type MR2002 (SYSCOM Inc.). The
sensors are oriented at the main axis of the building. In general, two
sensors were installed in two opposite corners on the roof and two
sensors in the same corners on a mid-floor story. The fifth sensor was
installed in the middle of the ground floor. Sensor number six was
installed outside of the building (see scheme in Figure 4).
A “lightweight” exciter (transportable by two men; covering a fre-

quency range between 1 and 15 Hz) was successfully applied [3].
Figure 4 — Applied instru-

mentation scheme [1]
Elaboration and interpretation of the measured data
Different programmes were applied to determine the eigenfrequen-

cies and corresponding mode shapes. The ambient and forced vibra-
tion data were conditioned with the programme MATLAB for the use
in the programme ARTeMIS, the MATLAB toolbox MACEC as well as
MATLAB routines.
The signals were than analysed by the application of FFT techniques [3]

and use of the Stochastic Subspace Identification Method [1]. Thus only
the programme ARTeMIS and the toolbox MACEC provide directly a
sketch of the mode shape.
Mode 1 Mode 2
Before Strengthening
ARTeMIS
MACEC
After Strengthening
ARTeMIS
MACEC
Figure 5 — Determined char-

acteristic modes before and
after strengthening by the use
of instrumental data
The pure analysis of the data by the application of FFT techniques

only identifies the natural frequencies of the building at each meas-
urement point.
Figure 5 show the first two mode shapes analysed with the programme
ARTeMIS and MACEC, which could be determined for the building
before and after strengthening with. It can be clearly seen,
Figure 6 show the normalized FFT response values from the sensor MR1
and MR3 in x- and y-direction before and after strengthening; the nor-
malization accounts for different amounts of weights which were used
during the measurements to cover the whole frequency band [3]. Espe-
cially in the case without strengthening clear peaks can be determined,
which indicate the highest amplification of the introduce signal due to
the resonance to the natural frequencies of the building in each direction.
Table 1 provide the summary of the determined first structural fre-

quencies. The comparisons show quite identical results for the first
two modes (f1, f2) before and after strengthening. But it has to be
noted, that the identification of the natural frequencies by the FFT
approach only might be lead to other results, because at some sensors
no clear peak can be identified and a comparison to other sensors or
approaches is recommended.
Frequency [Hz]
Software before strengthening after strengthening
f1 f2 f1 f2
MACEC 6.90 7.50 8.30 9.12

Table 1 — Determined funda-
mental frequencies from the ARTeMIS 7.08 7.52 8.25 9.03
experimental testing FFT 6.79 7.25 8.25 8.97
Analytical Studies
For the building investigated, a 3-dimensional model was created using

the software tools ETABS and Sap2000. Provided construction plans of
the building supplied the required geometrical data. Material properties
for the given quality were assumed according to the Turkish standards.
Excitation in x-direction Excitation in y-direction
a) Before strengthening
Figure 6 — Recorded build-

ing response (normalized FFT
Amplitude) at Sensor MR1 and
MR3 for excitation in x- and
y- direction
b) After strengthening
Columns were generally assumed as rigidly connected to the under-

ground. Floors were modeled as rigid diaphragms; roof constructions
were taken into account by planar loads. Masonry infill walls were
neglected to reduce the modeling and analysis effort, because of the
limited time window.
The material parameters for concrete were assumed to have character-

istic cube strength of 16 MPa (as denoted in construction plans). Rein-
forcement was assumed to be of Turkish steel grade S220a (220 MPa
yield strength and 10% strain at ultimate strength; generally smooth,
not profiled), also corresponding with specifications in plans.
Figure 7 show the 3D model before and after strengthening modeled

with beam and shell elements. The adaption of beam elements for
the added shear walls was necessary to apply nonlinear hinges for the
push over analysis.
before strengthening after strengthening
Figure 7 — 3D model before

and after strengthening (Using
shell elements for the mod-
eling of the structural walls)
The fundamental periods of the buildings, which were known due to

the dynamic recordings, were used as a measure for the calibration of
the stiffness parameters. Young’s Moduli were reduced (in comparison
to recommended values given in code specifications) for all elements
of identical material. In the frame of the project the goal was not to
achieve a perfect match for the building. Congruency of the main mode
shapes and frequencies/periods was nevertheless seen as important.
Table 2 shows the comparison of the experimental and first analytical

(without calibration) determined fundamental periods of the school
building before and after strengthening. The calibration of the model
was foreseen by the link of SAP2000 with an optimization MATLAB
program prepared by the participants of Project 5. Due to technical
problems and limited time only the communication could be realized
and the routine be rune. A final calibration could not be achieved.
Therefore the natural frequencies of the analytical model are so far

not comparable with the results from the measurement, which indi-
cates the need of model calibration. The results show, that modeling
of complex 3D structures bases on the construction plan can lead to
much different dynamic behavior and thus to un-conservative seismic
load assumption (see Figure 8).
Period [sec]
Software before strengthening after strengthening
T1 T2 T1 T2
MACEC 0,145 0,133 0,120 0,110

Table 2 — Comparison of the
ARTeMIS 0,141 0,133 0,121 0,111
fundamental periods before FFT 0,147 0,138 0,129 0,118
and after strengthening SAP2000 0,310 0,180 0,135 0,132
Damage Prognosis
Nonlinear static pushover analysis was applied to determine the build-

ing capacity in form of the capacity spectrum in x- and y-direction and
to carry out a damage prognosis. The N2-method proposed by Fajfar
[4] was applied to investigate the response of the school building to a
design spectrum excitation. A design spectrum according to the Turk-
ish seismic code and assuming different PGA level, factor of building
importance class I = 1.0, and subsoil conditions Z2 was used. For a
PGA of 0.4 g corresponding to the highest seismic zone no intersec-
tion between the capacity spectrum and the design spectrum resulted;
thus, according to the calculations performed here, the building would
not be able to withstand an earthquake corresponding to the required
specifications. Further calculations were carried out in order to deter-
mine the level of peak ground acceleration the building could resist.
The value of PGA = 0.33 g was found to be the threshold for total col-
lapse (see Figure 9).
Figure 8 — Indication of the

influence of the considered
characteristic periods on the
example of the Turkish Code
Spectrum
a) x-direction
Figure 9 — Demand spectrum

in x- and y- direction by the
b) y-direction application of N2-method
Conclusions
In the frame of the project a damage prognosis was carried out for
a 3-story RC frame school building under the use of building response
measurement data and modern software tools. Experimental testing at
the building could be done before and after strengthening of the struc-
ture by the project partner MKU, who provided the data for the project.
For the strengthening of the structure shear walls were added to increase
the building capacity and to decrease the maximum displacements.
The provided building response data were analyzed to determine

the principle fundamental periods and corresponding mode shapes,
which were used for the calibration of the structural models. The com-
parisons of the experimental and first analytical periods indicate the
necessity of model calibration. First analytical results indicate a much
weaker behavior of the structure, which can lead to un-conservative
seismic load assumptions.
Estimations of structural performance were conducted using site-spe-

cific response spectra reflecting subsoil conditions of class Z2 according
to Turkish code and the application of the N2-Method. The results for
the origin structure indicate that the vulnerability of the 3-story school
building does not fulfill the requirements by the design code. For the
site considered ground acceleration of 0.4 g will cause heavy damages
or collapse of the building. The structure is only capable to withstand

ground motion smaller than 0.33 g. There was a need to strengthen the
structure. The check of the sufficiency of the carried out strengthening
measures could not be realized in the frame of the project.
References
[1] Genes M.C. (2011). Forced vibration testing of school buildings before
and after strengthening process. Bauhaus Summer School — Model
Validation and Simulation 2011, Presentation
[2] Schwarz, J., Abrahamczyk, L., Lang, D.H., Maiwald, H. (2004).

Ingenieuranalyse von Erdbebenschäden: Das Bingöl (Türkei) Erdbeben
vom 1. Mai 2003. Bautechnik 81, 6, pp. 445–460.
[3] Abrahamczyk, L., Schwarz, J. (2009): Vibration measurements on

an 8 story office building in Bern (Switzerland). In Proceedings
Workshop on — Case studies of seismic building instrumentation
and monitoring. Bauhaus-Universität Weimar
[4] Fajfar P. (2000): “A nonlinear analysis method for performance

based seismic design.” Earthquake Spectra 2000; 16(3): 573–592.
Project 5 — Calibration of Numerical Models with Applications in Civil Engineering 319
Project 5 — Calibration of
Numerical Models with Applica-
tions in Civil Engineering
Abstract Akansel, Vesile Hatun

Middle East Technical Univer-
sity, Turkey
The course aimed in understanding and testing well-known optimiza-
tion tools for the nonlinear problem of parameter identification for Bankir, Sahin
problems occurring in civil engineering applications. The students Mustafa Kemal University,
worked on 4 different classes of optimizers and applied them to both Turkey
academic ad real world examples
OLIVEIRA, Luciane Marcela
Filizola
School of Engineering of São
Introduction Carlos, University of São Paulo
Faraonis, Periklis
It is a common practice in engineering to calibrate numerical models
against sets of measurements in order to predict the structures’ behav- saloniki, Greece
iour more precisely.
KADĚROVÁ, Jana
University of Technology, Brno
In particular in civil engineering, structures may have been built
recently, where the design was supported with the help of numerical Koilanitis, Ioannis
models. As now, after the construction phase, first loads are acting Aristotle University of Thes-
on the structure, its responses can be observed. With this, the model saloniki, Greece
used in the design process can be validated posteriorily. Additionally,
Lahmer, Tom
the model parameters can be tuned in that way that the model’s pre- Bauhaus-Universität Weimar,
diction coincide with the measured data as good as possible. Germany
This tuning, solving an inverse problem, can effectively be solved with Papadopoulos, Savvas
methods of finite dimensional nonlinear optimization tools.
saloniki, Greece
Skiada, Dafni
Gradient and Newton Methods Aristotle University of Thes-
saloniki, Greece
Classical approaches in the field of nonlinear optimization are Gradi- TÓTH, Adrienn
ent-based and Newton-type methods. These methods are motivated BME, Department of Struc-
by studying necessary and sufficient conditions of optimality. These tural Engineering, Hungary
class of optimizers are generally supported by mathematical analysis
320 Project 5 — Calibration of Numerical Models with Applications in Civil Engineering
concerning convergence and convergence rates. In this course gradi-

ent methods with line-search strategies and a so-called quasi-Newton
method, where the Hessian matrix is approximated by a rank two update
according to the BFGS approach are tested The latter approach allows
to approximate Hessians and to use Newton directions in the optimizer
instead of gradient directs, which depending on the topology of the
optimizing problem may tend to zig-zagging leading to slow convergence.
Direct Search Methods
It is not always possible to calculate gradients in a satisfying manner,

mainly due to missing differentiability of the cost functional. In these
cases it can be advisable to apply gradient free methods, so called
direct search methods. A very effective and well-proven method is the
algorithm according to Nelder-Mead, which bases on simplex. A sim-
plex is the connection of (n + 1) points in the parameter space, where
n is the number of parameters which need to be identified. Now, the
strategy is to replace always the vertice which has the largest value
in the cost function. The replacement is done by reflecting this point
to the centroid of all other points. A series of expansion and contraction
approximates a kind of line search. After finding a new point, the verti-
ces are sorted again and the vertice with the worst cost function value is
again replaced. Being close to an optimum, the simplices start to shrink.
Optimizers based on Swarm Intelligence
Another class of optimizers is motivated by swarm intelligence, where

every member of the swarm has its own flight direction and veloc-
ity, however generally for the whole swarm there is one global flight
direction. Applying such methods one can run global searches. With
a high probability, the methods will converge to a global minimum;
however, there is no guarantee for this. As the particles are spread
randomly on the set of all admissible input parameters, the probability
in finding the global optimum is increased using more particles. How-
ever, more particles mean higher numerical efforts, which can become
a challenge when the computing time of the model is high.
Bayesian Updating
In Bayes Framework the optimization corresponds in maximizing the

likelihood (the probability that the model parameters fit well to the
measurements). The advantages of the Bayesian approach are that
information of the prior knowledge of the sought-for-quantities can
be taken into account. Accordingly also any knowledge about uncer-
tainties in the measurements are evaluated, which allows at the end
for a qualitative assessment of the identified parameters in terms of
Project 5 — Calibration of Numerical Models with Applications in Civil Engineering 321
probability density functions. Comparing them, information can be

provided if sufficiently many measurements have been available in
order to detect the parameters sharply, i.e. with a low uncertainty.
Application and Results
The methods have been applied to identify characteristic data of an

Harmonic Oscillator. The model is the solution of the equation of
motion in 1 D. The second order differential equation is solved ana-
lytically including the three parameters
–– Stiffness (k)
–– Damping (c)
–– Mass (rho)
From recorded time signals, these parameters need to be identified. The

cost function measures the squared difference between computed and
simulated signals for a set of time steps. In Figure 1 the recorded data
are plotted as blue points. The model' solution with initially guessed
parameters resulted in the green curve. After the calibration, simula-
tion results, i.e. the red line, and measurements are in good agreement.
Besides the calibration of the parameters we discussed questions of

unique identifiability and tried to visualize this by the help of countor
plots. In the following figures, always one parameter was kept fixed,
the others varied and isolines of the cost function have been com-
puted and visualized.. From these plots one can graphically detect the
optimal parameter pairs leading to a good fit. In case that both the
menachical stiffness and the mass are unknown, the problem has infi-
nitely many solutions, see Figure 3. Adding a regularizing term in the
definition of the cost function, any deviation from the initial guess is
penalized. This approach helps in finding a solution from the infinitely
many combinations of mass and stiffness which provide a good fit.
Summary
In this course a series of optimization methods have been imple-

mented by the students. The methods provided the bases for calibrat-
ing some numerical models against measured data of the other groups
in the second part of the summer academy.
Students of the calibration course could help the others to get things
quickly run. It has shown, that in particular when little a priori knowl-
edge is given, the method according to Nealder-Mead showed the
most robust optimization results when beeing applied to more com-
plex models in engineering. Being close to solution, gradient-type and
Newton methods showed good performances.
322 Project 5 — Calibration of Numerical Models with Applications in Civil Engineering
Figure 1 — Measured, simu-

lated and fitted displacements
of the harmonic oscillator
Figure 2 — Contour lines of

the cost function with assumed
known mass. Stiffness and
damping are varying. The opti-
mum is in point (0.2, 4).
Figure 3 — Contor lines of the

cost function with assumed
known damping. Stiffness and
mass are varying.A clearly dis-
tinghuishable minimal point
cannot be identified.
Figure 4 — Same as previous

Figure. However, the cost
function includes a penalty
term, which allows to find a
minimum in (2.2, 4.1).
Project 6 — Long-span Bridge Aeroelasticity 323
Project 6 — Long-span Bridge

Aeroelasticity
Abstract MORGENTHAL, Guido

The project deals with the structural modelling and dynamic analysis
ABBAS, Tajammal
of long-span bridges under wind excitation. A realistic simulation of BUW, Weimar, Germany
the structural behaviour first requires the analysis of the geometric
stiffness under self-weight. Based on this, the dynamic response of DANYILDIZ, Bekir
the structure due to various critical wind excitation phenomena will MKU, Hatay, Turkey
be simulated numerically. These include turbulence-induced buffeting, HENAO PUERTA, Alejandra
vortex-induced vibrations and instabilities like flutter. UNC, Manizales, Colombia
The project uses Computational Fluid Dynamics (CFD) simulations to SÁNCHEZ CORRIOLS, Abra-
ham, UPM, Madrid, Spain
determine the aerodynamic properties of the bridge girder. These will
then be used on the full three-dimensional bridge model to assess the DE BORTOLI, Marta
aeroelastic behaviour of the structure. UNIPD, Padova, Italy
Four work teams have been constituted to carry out four different HEGYI, Peter
BUTE, Budapest, Hungary
topics related to bridge aerodynamics: influence of windshields on
bridges, improvement of aerodynamic behaviour of cross-section ABERLE, Marcus
geometry, vortex-induce vibration on bridges and flutter instability. KUAS, Karlsruhe, Germany
LESGIDIS, Nikolaos
A brief description of them and the main results achieved are
AUTH, Thessaloniki, Greece
presented below.
ROCHA, João Francisco
FEUP, Porto, Portugal
BUDAHAZY, Viktor
KOLEVA, Tatyana
UACEG, Sofia, Bulgaria
KISHAIJA, Paul
Study of the influence of windshields

on bridge sections
DANYILDIZ, Bekir Abstract

MKU, Istambul, Turkey
This paper analyses the effect of windshields on the aerodynamic
HENAO PUERTA, Alejandra
UNC, Manizales, Colombia behaviour of bridges. To that end, numerical simulations have been
carried out by modelling the interaction between the airflow and three
different cross-section geometries: a standard bared cross section
formed by a simple box girder with two cantilevers on both sides, the
standard section with vertical full windshields curved on top, and the
standard section with porous windshields. VXflow, a particle-based
CFD code, have been used to perform these numeric simulations. The
post processing have been done with MatLab.
Introduction
Wind protection barriers (or windshields) are widely used in all bridges
that are subject to constant wind actions in order to protect the traf-
fic of vehicles (and pedestrians) against sudden variations of wind
speed (gusts), and other effects that may cause a number of instabili-
ties, especially on trucks, and represent a dangerous problem. Figure 1
shows the influence of wind on vehicles over the bridge, through the
display velocity field. Blue tones correspond to low speeds, and the
red colour to higher wind speeds.
The addition of the barriers is counterbalanced by the increase of

the bridge area exposed to wind, resulting in an increase of the wind
forces acting on the structure. Bridge cross sections must be designed
also considering this action. Altering the cross-sectional geometry
should be taken into account also in the study of aeroelastic of bridge.
With the adding of windshields, fluid-structure interaction is modified,
and may lead to the emergence of larger dynamic vibration problems
motivated by vortex shedding, whose frequency is mainly influenced
by the section geometry and the wind speed.
Figure 1 — Effect of wind-

shields on traffic (velocity
field visualization obtained by
numerical simulations)
Case Study
Three different cross-sections have been considered: the standard

bared box girder section with two cantilevers on both sides, the stand-
ard section with windshields curved on top, and the standard sec-
tion with porous windshields (see Figure 2). The vertical dimension of
windshields is 4.0 m, as recommended on various standards to achieve
a adequate protection against wind effects on traffic crossing the bridge.
Figure 3 shows some examples of actual bridges windshields.
Figure 2 — Sections models

performed: standard, curved
and porous
Figure 3 — Some examples of

windshields on real bridges
Static simulation tests
Regarding to the values of the drag coefficient (CD, Figure 4) as a func-
tion of different angles of attack (), it is found that the standard sec-
tion is the one with the lowest values. The highest values correspond
to the cross section with porous windshields. The drag coefficient
for standard section with windshields curved on to is slightly lower
because of its better aerodynamic behaviour. Regarding to the time
history of the lift coefficients (CL, Figure 5), it is observed that the section
with windshields curved on top has the major values. Dynamic effects
caused by the wind will be more important in this section. The lowest
mean lift coefficient corresponds to the section with porous windshields.
Figure 4 — Drag Coefficient

of cross-sections
Figure 5 — Time history for

lift coefficient of the three
cross-sections (CL) : standard
(top left), with rounded wind-
shields (top right), and porous
windshields (bottom)
Table 1 shows a summary of force coefficients for the three cross-

sections considered
STANDARD CURVED POROUS

SECTION SECTION SECTION
Table 1 — Force coefficients CL 0,96 0,67 0,3

for the three cross-sections CD 1,25 2,60 2,96
performed CM 0,19 0,22 0,33
Conclusions
In view of the above results, it follows that the section with full wind-
shields and curved on top presents a better behaviour against wind
effects on traffic. However, this section also has the major force coeffi-
cients, indicating that static wind actions are greater, and also possible
effect due to dynamic vibration that may suffer as a result of fluid-
structure interaction phenomenon.
Since it is important to protect traffic from the destabilizing effects

of wind and, at the same time, to have reasonable values for force
coefficients, the section with porous windshields is the optimal solu-
tion to be used.
References
[1] Morgenthal, G. “VXflow Primer” version 0.994, Weimar, Germany,

May 2012
[2] SIMIU, E., SCANLAN, R. “Wind Effects on Structures” John Wiley &
Sons, Inc., New York, USA, 1996, Chaps. 6 and 14.
Improvement of aerodynamic
instability on bridge sections
DE BORTOLI, Marta Abstract

UNIPD, Padova, Italy
The aerodynamic behaviour of Lillabælt bridge cross-section is dis-
HEGYI, Peter
BUTE, Budapest, Hungary cussed in this paper. Four cross sections are investigated to compare
their aerodynamic properties (force coefficients), and analyse the
vortex-induced vibration phenomenon. Among the numerical results,
also a brief theoretical introduction is provided.
Introduction
Lillebaelt or New Little Belt Bridge (Figure 1) is a suspension bridge

situated over the Little Belt strait in Denmark. The total length of the
bridge is 1.080 m, with a central span of 600 m and two side spans
of 240 m each. The two piers have a total height of 113 m and stand
on deep pile foundations. The bridge was opened in 1970. The cor-
responding cross section can be seen in Figure 2. Due to its geometry,
the bridge was likely to be sensitive to wind-induced oscillations, and
this is the reason why it was largely analyzed. In fact, other bridges
with similar structural characteristics — suspended bridges with very
slender decks — had shown problems in the past, for example the
Tacoma Bridge in Washington, USA.
Figure 1 — The Lillebaelt

Bridge
Figure 2 — Original cross sec-

tion of Lillebaelt Bridge
Moreover, the geographical environment is characterized by steady

wind speeds, which worsen the aerodynamic phenomenon. In this
paper we will compare the original cross section with three other cross
sections to observe the different behaviour of them. VXflow, a particle-
based CFD is used to study the wind-structure interaction phenomenon.
Static analyses
Introduction
The wind velocity was set to 15 m / s to simulate a realistic case. The
same cross sectional depth, 2,85 m is used for all the cases, in order to
compare the force coefficient properly. Different wind directions from
–4° to 4° are studied to analyse the influence of the angle of attack.
Also, the boundary layer around the section to assess the comfort lev-
els for truck drivers is analysed.
Cross sections
A sketch of the four cross sections can be seen in Figure 3. The original
section has triangular fairings, while one of the proposed cross sec-
tions has round fairings and the other one is a bare rectangular section.
The fourth one is an original cross section equipped with windshields.
Figure 3 — Considered cross

sections
Figure 4 — Drag, lift and

moment coefficient in the
case of different wind direc-
tions
Results
Force coefficients
Figure 4 shows the drag and lift coefficients corresponding to the cross
sections with triangular and round fairings. It can be observed that
the triangular cross section has lower drag and lift force coefficients
than the rounded one. Since the derivative of lift coefficient of the
triangular cross section is smaller than the rounded one, the triangular
section is less sensitive to buffeting phenomenon, because it will be
less excited by the varying wind velocity directions.
In Figure 5, the force coefficients are compared in the case of hori-

zontal wind velocity. It is clear again that the triangular cross sec-
tion has optimal properties comparing to the other sections. The cross
section with the windshields shows worse behaviour in terms of lift
and moment coefficients, as a consequence of his larger depth and its
asymmetry about the horizontal axis.
Boundary layers and vortex formation

In Figure 6, lower velocities are marked with blue, while higher veloci-
ties are in red colour. The smaller the blue region around the cross
section, the more efficient the shape is. It can be observed that around
the rectangular section there are some blue areas, which indicates that
the wind flow is very whirling. The same can be observed in the case
of the cross section with windshields. However, the windshield has a
positive effect on the velocity in the area of the traffic. The average
wind speed decreased drastically because of the windshield, but based
on Figure 7 it can be observed that the direction of the velocity is vary-
ing, because vortices are formed between the shields. The rounded
and triangular sections are more efficient since they hardly influence
the wind flow around them. This is also their disadvantage, because
the wind acting over the traffic is not reduced by anything.
Figure 5 — Force coefficient

for horizontal wind velocity
Dynamic analysis
Overview
In the wind speed-displacement graphic (Figure 8) there are two clear
peaks: the first one occurs for lower velocities and identifies the critical
wind speed for vortex induced vibrations (VIV), while the second and
higher peak represents the instability caused by flutter. We focused our
dynamic analysis on the first peak only. When the critical wind speed is
reached, the displacements increase and maintain a constant pattern,
thus reaching a limited value. This is the difference with the flutter
instability, in which the wind energy is too high and produces increasing
amplitudes of displacements. In the VIV, however, the peak is reached
when the vortex frequency approaches the natural frequency associ-
ated to a considered mode shape, causing a resonance phenomenon.
Figure 6 — Boundary layers of

the cross sections
Figure 7 — Vortices between

the windshields
Figure 8 — Critical wind

velocities
Each mode shape frequency can be excited by the vortex-induced

vibrations, so at least the first 4–5 modes should be taken into account
to identify critical wind speeds. The critical wind speed is reached
when the displacements do not show higher oscillations but they are
steady while wind speed is kept.
In Figure 9 the spectral density of lift coefficient (S(CL)) is drawn against

the Strouhal number. A region should be defined to identify the upper
and lower border of the possible resonance. In this zone dynamic analy-
sis should be performed to define accurately the VIV phenomenon. The
bordering velocities can be evaluated using the following equation:
f D
(1) V
St
where f is the vortex shedding frequency, D is the total depth of cross
section, St is the Strouhal number.
Figure 9 shows the spectral density of lift coefficient against the Strou-
hal number for the cross section with windshields. Strouhal number
value is located between 0.05 and 0.07, for which the lower velocity
is 6,4 m / s and the upper one is 8,9 m / s.
Figure 9 — Strouhal-number
diagram
Figure 10 — Vertical displace-

ments
We performed analysis assuming wind speeds of 7.0, 7.5, 8.0 and

8,5 m / s. Natural vibration frequency associated to the first vertical
mode shape is f = 0.156 Hz and D = 2.85 m. A very low value for the
damping ratio is used ( = 0,1%) in order to have visible magnitudes
for vertical displacements. The modal mass is 11,100 kg / m.
The analysis was carried out assuming these numerical values. The
steadiest vertical displacements were observed in the case of 8.0 m / s,
they are shown in Figure 10.
The critical wind speed depends on the modal mass, the less mass is
excited, the more evident are the effects. Although, the analysis was
performed only considering the first vertical mode shape, which is
characterized by a frequency of 0,156 Hz, further dynamic analyses
should be carried out considering higher natural vibration frequencies.
References

May 2012
[2] SIMIU, E., SCANLAN, R. “Wind Effects on Structures” John Wiley &
Sons, Inc., New York, USA, 1996, Chaps. 6 and 14.
Study of vortex-induced vibrations

on Volgograd Bridge
ABERLE, Marcus Abstract

KUAS, Karlsruhe, Germany
This paper investigates force coefficients, Strouhal number and critical
LESGIDIS, Nikolaos
AUTH, Thessaloniki, Greece wind speed to estimate the aerodynamic performance of the Volgo-
grad Bridge. A global FEM model was created using a commercial soft-
ROCHA, João Francisco ware package to determine the relevant dynamic parameters. Vertical
FEUP, Porto, Portugal displacements were obtained using a particle-based CFD code for a
wide range of wind speeds. Numerical results show significant ampli-
tudes even for moderate wind speeds. To reduce these vortex-induced
vibrations (VIV) to an acceptable limit, tuned mass dampers (TMD)
are used and their effects are also studied.
Introduction
Long span bridges are usually sensitive to wind loads. Since the col-
lapse of the Tacoma Bridge in Washington many studies concerning
the dynamic response of bridges were initiated in order to acquire more
information on aeroelastic phenomena. These findings play an impor-
tant role in the design process regarding the structural safety of bridges.
Several aerodynamic phenomena are capable of initiating critical vibra-

tions in bridge structures such as buffeting, vortex-induced vibrations,
galloping and flutter, among others. Vortex-induced vibrations can be
an important issue for the safety of the bridge in cases when vortex
shedding frequency and natural vibration frequency associated to one
mode shape coincide, resulting in a resonance phenomenon.
An interesting case of vortex-induced vibrations was observed the

20th of May 2010 when the Volgograd Bridge (Figure 1) experienced
vertical vibrations with maximum amplitudes of up to 71 cm.
Figure 1 — Volgograd Bridge

oscillating due to wind excita-
tion
Most of the problems observed in vortex-induced vibrations phenom-

ena are directly connected to the geometry of the cross section. Regard-
ing this, the current study tries to reproduce the measured response
using various numerical methods and discussing some possible solutions.
Case Study
The Volgograd Bridge crosses the Volga River in the city of Volgograd
(Russia). It is a long continuous steel girder, distributed in 10 spans
with a total length of 1211. There are three main spans in the centre
of 155 m long. The steel bridge deck consists of a box girder section
strengthened with interior stiffeners (Figure 2). The continuous bridge
beam as well as the cross section were modelled using the commer-
cial software SOFiSTiK [1]. To simplify the dynamic calculations, the
stiffeners were replaced using an equivalent thickness for the top and
the bottom plate of the box section as shown in Figure 3. The cross-
sectional properties of the simplified model were callibrated using
experimental measurements of the Volgograd Bridge.
The internal pier to beam connenctions were modeled using pinned

supports. A similar process was used for the abutments with an addi-
tional restraint of the torsional degrees of freedom at each abutment.
The mass of the deck was modelled using 30 equally spaced lumped
masses along each span.
Figure 2 — Cross section of

the Deck
Figure 3 — Simplified deck

cross section (SOFiSTiK FEM
model)
Figure 4 — Volgograd bridge

1st bending mode
A modal analysis of the first 10 modes was performed on the global beam
model. In the present study only the first vertical bending mode, with a
frequency of 0.44 Hz (Fig. 4), was utilized for the following calculations.
Static Analysis
To model the aerodynamic properties through the VXflow code [2], it

is first necessary to discretize the perimeter of the cross-section and
to specify the number of panels running in an anti-clockwise manner
around the section as shown in Figure 5.
Figure 5 — Cross section of

the deck (VXflow Model)
The vortex particle method in which VXflow is based [3], allows the
user to obtain the force coefficients as well as the Strouhal number for
a specified cross-section. Their knowledge then allows the determina-
tion of further parameters such as critical wind speed that identifies
vortex-induced vibrations problems.
VXflow provides also the force coefficients shown in .
Figure 6 — Static Forces Coef-

ficients
The mean force coefficients are as follows:
Lift Coefficient — Cl = 0.34

Drag Coefficient — Cd = 1.18
Moment Coefficient — Cm = 0.26
The Strouhal number for the cross-section given in Figure 5 can be

identified from the spectral density plot of the lift forces given by
VXflow [2]. Hence, the Strouhal number for the cross-section of the
Volgograd Bridge is approximately St = 0.12 (Figure 7).
Figure 7 — Spectral density of

lift forces
With knowledge of the Strouhal number, the cross-sectional reference

depth and the first bending frequency of the structure it is then pos-
sible to calculate the critical wind speed for vortex-induced vibrations:
f D f D 0,44 3,8
St Ucrit 13,9 m s (1)
U St 0,12
where f = 0,44 Hz = first bending frequency of the structure

D = 3,8 m = reference depth
St = 0,12 = Strouhal number
Dynamic Analysis
The dynamic analysis performed with VXflow can then be used to deter-
mine the vertical displacements of the bridge for a wind speed range
from 10 m / s to 20 m / s with increments of 1 m / s. However, from
13,1 m / s to 15 m / s vertical displacements were calculated in steps of
0.1 m / s. The modal mass of the first bending mode is 2500 kg / m and
the modal stiffness is 19110 N / m². As the Volgograd bridge is a steel
structure, the damping ratio ( ) adopted for the present study is 0,3%.
A plot of maximum vertical displacements is shown in Figure 8. A local
peak can be observed associated to a wind speed of 14.7 m / s.
Figure 8 — Maximum vertical

displacement for different
wind speeds
In order to understand the evolution of the displacement for three

different wind speeds, the time histories for 10 m / s, 14,7 m / s and
15 m / s are presented in Figures Figure 9 to Figure 11. As can be seen,
positive displacements are higher than negative displacements. This
is due to the fact that the cross section is not symmetrical respect to
its horizontal axis. Also, vertical displacements for a wind speed of
14,7 m / s are considerably larger than the others. This is the critical
value that identifies the VIV phenomenon.

ment for a wind speed of
10 m / s

14,7 m / s

15,0m/s
A visualization using vortex-streaks generated on the airflow as a

consequence of wind wind-structure interaction for a wind speed of
14.7 m / s can be seen in Figure 12.
The increase of vertical displacements for some critical wind speeds

can generate some comfort problems or even fatigue damage of the
structural elements of the bridge. In order to reduce them, tuned mass
dampers (TMD) can be employed. For the Volgograd Bridge, these
problems were minimized through TMD for the first three principal
modes of vibration.
Figure 12 — Visualization of
vortex formation through
VXflow
Conclusions
This study presents the importance of wind-structure interaction anal-

ysis for long span bridges. Vortex-induced vibrations can influence on
the bridge safety in cases when the vortex-shedding frequency and
the natural vibration frequency of the bridge are close. As shown, for
moderate constant value of the wind speed some amplification of the
displacements of the cross section centre can be observed. This fact is
due to the wind flow through the cross section. This study shows the
feasibility of numerical methods to predict vortex-induced vibrations
in some structures. This has been achieved by using a commercial
finite element program for the determination of the dynamic proper-
ties of the structure in combination with the particle-based CFD code.
It also helps to decide which countermeasures could be used in order
to minimize such vibrations.
References
[1] SOFiSTiK TUTORIAL, Version 10.21-99

May 2012
[3] Morgenthal, G. “Aerodynamic Analysis of Structures Using High-

resolution Vortex Particle Methods” PhD Dissertation, Department
of Engineering, University of Cambridge, UK, Oct. 2002
Flutter analysis of Machang Bridge

(South Korea)
Abstract BUDAHAZY, Viktor

The aim of this paper is to investigate the flutter behaviour of the
KOLEVA, Tatyana
Machang Bridge by performing a 2-DOF-section model analysis. Aero- UACEG, Sofia, Bulgaria
dynamic studies are essential for slender bridges. Fluttering and aero-
dynamic properties can be improved by modifying the original section KISHAIJA, Paul
geometry. Three different sections were chosen and static, forced and BUW, Weimar, Germany
dynamic simulations were performed and compared.
Introduction
The Machang Bridge is 1,7 km-long, dual two-lane bridge and it joins
the cities of Masan and Changwon in the southern region of South
Korea. The project features a 740m, cable-stayed bridge with a main
span of 400m, side spans of 175m and two 164m-high, H-shaped
Figure 1 — The Machang

cable-stayed Bridge in South
Korea
Figure 2 — Analysed cross

sections
pylons. The bridge deck is composite, consisting of concrete slab on

steel plate girders. The bridge opened to traffic in July 2008, shorten-
ing the commute between the twin cities by 30 minutes.
Flutter is a self-excited oscillatory instability, which is caused when

aerodynamic forces supply energy to the oscillating structure and
amplitude of the motion progressively increases. This behaves as “neg-
ative damping” and it occurs at any wind speed higher than a critical
value, which is known as flutter limit.
Classical flutter is an aerodynamic phenomenon in which rotation and

vertical translation couple together in a flow-driven, unstable oscilla-
tion. In this case the negative aerodynamic damping occurs and defor-
mations reach such a level that failure happens [1].
Cases of analysis
Static simulations
The results show that the drag force has been reduced by more than
50% for the modified geometry of the cross-section.
Cross Section numbers
CR 1 CR 2 CR 3
CL –0.27279 –0.5102 –0.19482

Table 1 — CL CD CM results CD –0.91905 0.48555 0.49897
for the three sections CM –0.01676 0.034571 0.017325
velocity field of original cross
section (#1) with VXflow
velocity field of modified cross
section (#3) with VXflow
Forced oscillation tests

In case of forced vibrations, the structure is forced in heave and pitch
oscillation with pre-determined frequency and amplitude. The aero-
dynamic derivatives are then calculated from the measured lift and
moment and the critical wind speed is computed [1].
For the aerodynamic forces on a bridge cross-section, Scanlan pro-

posed a set of expressions. It assumes that the self excited lift FL and
moment FM may be treated as linear in displacement h and rotation
α and their first derivatives.
The equations of motion are given in linearized form:
1 h B h
FL U 2 B KH 1 KH 2 K 2H 3 K 2H 4 (1)
2 U U B
1 h B h
FM U 2 B KA1 KA2 K 2A 3 K 2A4 (2)
2 U U B
where Hi and Ai depend on the section geometry; U∞ is wind flow veloc-

ity; h is the vertical displacement amplitude;  is the rotation amplitude.
Ai and Hi are known as aerodynamic or flutter derivatives and are

measured only if the body is in oscillatory state.
In this paper the results of three different sections are compared by

defining the A2 values for each one of them, which represents tor-
sional moment induced by pitching motion velocity. In order to define
these derivatives, pitch forced oscillation cases are performed sepa-
rately from the heave forced oscillation ones.
A2 parameter can be used to predict the critical wind speed for each
section. We can assume that section is unstable when the A2 curve
becomes positive, which simplifies the flutter problem avoiding the
influence of other flutter derivatives, which are less important than
the first one (Figure 5).
Figure 5 — A2–VR graphics

Flutter instability wind speeds can be predicted from as the point in

with A2* becomes positive for each cross-section geometry. A2* values
are normally given as a function of reduced wind speed (v,red), which is
equal to v,red = v · T / B, being v = wind speed, T = oscillation period cor-
responding to the modal shaped of the bridge (rotation, in this case) and
B = section width (chosen as the reference geometry). With the improved
two other sections, we move the instability wind speed to a higher value
by getting a positive value for A2 at a higher reduced wind speed (vred)
Dynamic analysis
Table 2 — Critical wind speed

for associated to flutter insta-
bility
The dynamic analysis shows that the original section becomes unstable
when the wind speed reaches 75 km / h, while the modified cross sec-
tions are stable even at higher wind speeds. By using the triangular fair-
ings on the borders, the bridge shows a stable behaviour under a wind
speed of 95 m / s, which is the maximum expected wind speed in this
region (typhoons effects). The closed section is the most stable, but only
with the triangular fairings we are able to improve the cross-section
behaviour section against flutter (torsional galloping, in this case).
Conclusions
–– The behaviour of Machang Bridge was examined and evaluated

–– The modification of the section reduces static wind forces. Drag
and lift forces were decreased.
–– The dynamic wind forces were also decreased. The fluttering is
avoidable using only the fairings.
References
[1] Abbas, T., Flutter analysis of suspension bridges using a two-stage

numerical approach & flutter derivatives, Masters Thesis, 2011

May 2012
[3] Morgenthal, G. “Aerodynamic Analysis of Structures Using High-

resolution Vortex Particle Methods” PhD Dissertation, Departament
of Engineering, University of Cambridge, UK, Oct. 2002
Excursion Report — City tunnel Leipzig 11th August 2012 345
Excursion Report —
City tunnel Leipzig
11th August 2012
According to the schedule of the Bauhaus Summer School, we started

our excursion towards Leipzig on the August, 11th 2012 around
8:30 am at the Atrium in Weimar. It was one of the most exciting tours
out of Weimar which had been fixed for the participants of Summer
School “Model Validation and Simulation”.
The bus was full of 50 persons out of whom 46 were students from
the different universities of the different countries from Asia to Europe
and South America. The tour was quite successful on both educa-
tional and cultural aspects. We visited an inner city railway construc-
tion site at Markkleeberg, at which a lot of specialities have to be
considered and the City-Tunnel Leipzig which was one of the most
attraction of our educational visit where we had seen the combina-
tion of latest engineering technologies applied to the reality. It was a
great journey in the context of sharpening and shaping the engineer-
ing knowledge and experiences.
Figure 1 — Network of Cen-

tral German suburban railways
Figure 2 — Networking of
Leipzig city map
Brief description of Leipzig City Tunnel:
The Leipzig city tunnel is the final and central component of the rail-
way system restructuring process in Greater Leipzig. The comple-
tion of the tunnel will not only bring improvements to the suburban
train system — transport between the region and the city will also be
improved, and the tunnel will give a major impetus to the develop-
ment of the city’s transport system.
The tunnel provides a direct rail link between the area to the south of
the city and the central station in the northern part of the city centre.
Until now, trains travelling along the north-south axis have had to
circumvent the city centre — a time-consuming process. In future,
they will simply travel underneath it. In this way the Leipzig city
tunnel will make regional rail traffic faster, more efficient and more
convenient for passengers across the board. The fast north-south axis
below Leipzig city centre will reduce total travel time by as much as
40 minutes on some routes. The city tunnel is the fast transport hub
for central Germany. The S-Bahn urban railway network for the Cen-
tral German region is the new backbone of S-Bahn and regional rail
transport in the Leipzig/Halle conurbation. As of late 2013, six new
S-Bahn urban railway lines will link the business region around Halle
and Leipzig with the surrounding area faster and more efficiently,
ensuring much shorter journey times by rail. The benefits will be felt
by the entire region – economically, environmentally and in terms of
a better quality of life. [1]
The Leipzig city tunnel has two 1.438 km long tunnels, one in each
direction, and four underground stations — Bayerischer Bahnhof, Wil-
helm-Leuschner-Platz, Markt and Hauptbahnhof (Central Station). The
shells of the tunnel and the stations have already been completed;
work is currently in progress on fitting out the interiors of the four sta-
tions and preparing the tunnels for the trains. [1]
Tunnelling (brief features)
–– The tunnel boring machine (TBM) has been designed and tailored
to the specific needs of the city tunnel and the local soil condi-
tions
–– The shield diameter is 9.0 m
–– The total weight of the TBM is 1,100 Mton
–– Including the trailing section, the TBM has a length of 65 m
–– The TBM has a total connected load of 2.5 MW
–– The TBM is driven by 14 pairs of presses. Their maximum pressing
power is 65 MN
–– The integrated stone crusher is capable of crushing stones up to
80 cm edge length
–– To pre-explore the ground, the TBM has been provided with a special
seismic system capable of recognising any anomalies or boundaries
of a bed 40 m in advance
Puskar Rijal
(author at the Bauhaus Summer School – Model Validation and Simu-
lation)
[1] http://www.citytunnelleipzig.de/de/kontakt-informationen/eng-
lish-information.html (last visit 10/2012)
Impressions from the Excursion:
a) Group picture of most of the participants

b) Enjoining Lunch at restaurant Bayrischer Bahnhof
Figure 3 — General
Impressions
c) Sightseeing in Leipzig city d) Getting back to Weimar
a) Introduction to the construction site
b) Questions by participants c) Walk over railway under

construction
Figure 4 — Impressions from

the construction site Mark- d) Strenghtening of the e) Reinforcement detailings
kleeberg ambutment by anchors
a) Waiting for entry to the subway b) Introduction to the metro station

station at Leipzig main station
c) Cabrio train tour at the city tunnel d) Model of the drilling maschine at
the InfoBox
Figure 5 — Impressions from

e) Getting out from video season f) Information about the construction
the city tunnel tour
of the City tunnel at the InfoBox
In the frame of the Bauhaus Summer

School a diversified programm was
organized. The participants were invited
to participate at different activities like
sightseeing tours, slogan parties, gaming
night or a rubber rafting tour.
With the slogan “Bonvena”
we welcome all the par-
ticipants to Weimar. With
music and good food the
ice was fast brocken.
Everyone comes wearing
the colours of their home
country or those of the
country whose language
they’re going to learn at
the Summer School.
In the frame of a guided
city tour the participants
could discover Weimar and
its historical places, history
and presence. Also trips to
other cities like Berlin and
Leipzig were offered.
Further a diversified even-
ing programm were organ-
ized. The participants could
show their talents at the
legendary Bauhaus Song
Contest and could come in
contact with other people
during the international
gaming night. There was
also a big final ceremony
with the motto “Back to the
50's — Let's twist again.”
IMPRESSUM
Schriftenreihe des Instituts für Konstruktiven Ingenieurbau
der Bauhaus-Universität Weimar
HERAUSGEBER
© Bauhaus-Universität Weimar, Fakultät Bauingenieurwesen,
Institut für Konstruktiven Ingenieurbau
Alle Rechte vorbehalten
REDAKTION
Dipl.-Ing. Lars Abrahamczyk, Dr.-Ing. Jochen Schwarz,
Prof. Dr.-Ing. habil. Frank Werner
Satz
Lena Schindler
BEZUGSMÖGLICHKEIT
Verlag der Bauhaus-Universität Weimar
Fax +49 (0) 3643 58 11 56, E-Mail verlag@uni-weimar.de
DRUCK
docupoint Magdeburg GmbH
UMSCHLAGGESTALTUNG
Ulrike Mönnig
HAFTUNGSAUSSCHLUSS
This project has been funded with support from the European
Commission.
This publication [communication] reﬂects the views only of the
author, and the Commission cannot be held responsible for any
use which may be made of the information contained therein.
http://nbn-resolving.de/urn:nbn:de:gbv:wim2-20130805-19969
ISBN 978-3-86068-502-0

IKI23 Klein PDF

Uploaded by

Copyright:

Available Formats

IKI23 Klein PDF

Uploaded by

Document Information

Original Title

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

IKI23 Klein PDF

Uploaded by

Copyright:

Available Formats

Bauhaus Summer School

023 ”Model Validation and Simulation”

Schriftenreihe des Instituts

“Model Validation and Simulation”

6th – 17th August 2012

14 Numerical modelling of non linear constitutive laws

15 Modeling of composite steel – concrete bridges

16 Evaluation Methods for Prediction Quality of Coupled Partial Models

17 Unsaturated soils – Fundamentals and Overview of Applications

18 Advanced transmitting boundary conditions for large-scale SSI

20 Seismic design of tall and slender structures including rotational

22 Vibration serviceability of footbridges

25 Asynchronous ground excitation and soil-structure interaction in

26 An analytical perspective for performance limits for structural walls

27 Earthquakes in developed mountainous regions

29 Vibration Based Post-Earthquake Damage Assessment of Structures

38 Fragility of Shear Wall Buildings with Torsional Irregularity

49 Vibration monitoring and structural identification of an historic

63 Knowledge based structural assessment of Mallorca cathedral

77 Seismic Strengthening of Masonry Infilled Reinforced Concrete Frames

87 Time dependent analysis of steel-concrete composite beams

99 Multi-filament yarns testing for textile-reinforced concrete

105 Seismic response of short columns affected by brick / tile aggregate

114 Experimental research on thermo-creep behavior of soft

132 Experimental and numerical assessment of the dynamic behaviour

144 Running Safety Evaluation of Trains Moving Over Bridges shaken

158 Numerical and experimental study of the vertical interfaces

168 Low cycle performance of T-stub components of bolted moment

174 Dynamic safety assessment of a small span high-speed

187 Comparative study of steel frame modelling levels and Eurocode

197 Assessment of structure-tuned mass damper assemblies in an

206 Performance-based evaluation of seismic demands for moment

216 Post-Earthquake Damage Screening of Steel Buildings using

227 Structural Solutions for the Construction of National Beverages

238 Modelling of soil-structure interaction in alpine regions

263 Classification of timber–steel connections and the behaviour of

272 Numerical simulations of pseudo-dynamic full-scale test on a three

290 Project 1 — System Identification, Model Updating and Simulation

303 Project 2 — Soil-Structure Interaction Simulation and Experimental

309 Project 3 — Evaluation of seismic performance of a template design

319 Project 5 — Calibration of Numerical Models with Applications in

323 Project 6 — Long-span Bridge Aeroelasticity

345 Excursion Report — City tunnel Leipzig 11th August 2012

350 Bauhaus Summer School

The past decade has seen a rapid development of numerical meth-

This volume summarizes and publishes some of the results: abstracts

Weimar, I would like to thank everyone for their contribution to the

Weimar, April 2013

Introduction WERNER, Frank and

In 1998, twenty students participated in the first summer school. In

The acquired knowledge is applied to scientific and practically relevant pro-

Targets and Objectives

Different language skills of the participants pose high demands on the

The young scientists will be offered the possibility to exchange modern

The management and presentation skills of the participants will be improved

Figure 1 — Home countries

Figure 2 — European Project

Current Course Topics

At the national level, no similar workshop to the International Summer