Structures 45 (2022) 1516–1529

Contents lists available at ScienceDirect

Structures
journal homepage: www.elsevier.com/locate/structures

A machine learning approach to appraise and enhance the structural

resilience of buildings to seismic hazards
Giulia Cerè a, *, Yacine Rezgui b, Wanqing Zhao c, Ioan Petri b
a
Stantec, Bristol BS1 4NT, United Kingdom
b
BRE Trust Centre of Sustainable Engineering, School of Engineering, Cardiff University, Cardiff CF24 3AB, United Kingdom
c
School of Computing Sciences, University of East Anglia, Norwich NR4 7TJ, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: Earthquakes often affect buildings that did comply with regulations in force at the time of design, prompting the
Building resilience need for new approaches addressing the complex structural dynamics of seismic design. In this paper, we
Artificial neural networks demonstrate how strucural resilience can be appraised to inform optimization pathways by utilising artificial
Optimisation
neural networks, augmented with evolutionary computation. This involves efficient multi-layer computational
Performance-based analysis
Seismic hazards
models, to learn complex multi-aspects structural dynamics, through several levels of abstraction. By means of
single and multi-objective optimization, an existing structural system is modelled with an accuracy in excess of
98% to simulate its structural loading behaviour, while a performance-based approach is used to determine the
optimum parameter settings to maximize its earthquake resilience. We have used the 2008 Wenchuan Earth­
quake as a case study. Our results demonstrate that an estimated structural design cost increase of 20% can lead
to a damage reduction of up to 75%, which drastically reduces the risk of fatality.

1. Introduction implemented PBD and the proposed distinction is not necessarily neat,
given that not all countries may update their regulatory framework in a
Earthquakes constitute a major threat for many countries in the similar timeframe.
world, putting at stake our buildings and infrastructures, and conse­ Nonetheless, the key message evidenced by Fig. 1 is the insufficiency
quently human lives. Building regulations are constantly informed by of the the past and current regulatory provision to ensure – especially for
lessons learnt from earthquake events and technological developments the existing building stock – sufficient safeguard to human lives in face
in engineering, often enforcing stringent regulatory compliance as a of seismic hazards. On a general basis, the regulatory framework of a
standard requirement for design and post-disaster structural retrofitting. nation would undergo improvements and resvisions if proven ineffective
However, despite an abundant research in resilience [1] and a stringent after a significant seismic event. However, this is not the case for all
seismic regulatory landscape, earthquakes are still causing significant countries.
human and economic losses in different regions of the world, as illus­ Another finding that can be gathered from Fig. 1 is that the impact of
trated in Fig. 1. the analysed earthquakes over time did not reduce with the expected
Fig. 1 provides a comparison in terms of Richter magnitude, death improvements of building regulatory frameworks. Nowadays, secondary
toll and financial losses across the most impacting seismic events since to the human safeguard, it is of paramount importance to also account
the beginning of the 20th century. However, a distinction is made be­ for the environmental impact of such seismic disasters. It is crucial to
tween seismic events before and after 1980, as the implementation of develop a strategy for building design which allows to tailor the struc­
performance-based regulatory frameworks was initiated at the begin­ ture to the hazard features and existing (or intended, for those buildings
ning of that decade [12]. Fig. 1 evidences how losses endured in the undergoing seismic restoration) geometry in order to optimize the usage
aftermath of major seismic events are still significant and comparable to of construction material and consequently reducing the waste as a result
events that took place prior to the introduction of performance-based of the minimized damage.
structural design (PBD). A meaningful observation is how some coun­ The shift from prescriptive to performance-based regulations [13]
tries featured in the “post-1980” seismic events may not have involved some substantial changes in structural design. Whilst

* Corresponding author.
E-mail address: giulia.cere@stantec.com (G. Cerè).

https://doi.org/10.1016/j.istruc.2022.09.113
Received 15 June 2022; Received in revised form 13 September 2022; Accepted 28 September 2022
Available online 6 October 2022
2352-0124/© 2022 The Author(s). Published by Elsevier Ltd on behalf of Institution of Structural Engineers. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
G. Cerè et al. Structures 45 (2022) 1516–1529

prescriptive design represents a determinist approach with defined regulatory frameworks specifications or structural analysis algorithms.
provisions, performance-based design (PBD) entails the achiement of a The proposed research is such to allow integration of different structural
local or global level of performance which is independent from the analysis methods or alternative regulatory frameworks with minimum
features or solicitations affecting a single structural member [14]. A modifications within the core algorithm.
noteworthy advantage of performance-based approaches over pre­ The paper is structured into eight sections. Following this introduc­
scriptive regulations lies in the ease of integration with Machine tion, Section 2 summarizes related literature, while Section 3 outlines
Learning (ML), including optimization techniques [13,15]. As exten­ the overall methodology adopted for the proposed research. Section 4
sively highlighted in the literature [15,16], in a performance-based provides a concise overview of the selected case-study building, while
approach the design is conditional on the attainment of a specific the detailed workflow and algorithms used in the research are detailed
target and therefore trial-and-error processes are generally adopted. in Section 5. Section 6 elaborates on the results obtained from the per­
Through a series of consecutive approximations, design variables and formed experiments and relevant obversations are presented in Section
parameters are refined at each iterations, eventually leading to the so­ 7. Finally, Chapter 8 provides concluding remarks, including limitations
lution which best complies with both regulatory requirements and and future research.
project-related constraints (e.g., time, costs, resources). The integration
of machine learning techniques finds its natural application in this 2. Related work
domain since certain tasks can be easily automated [15], especially
given the availability of engineering analysis tools that publish their This section summarizes the related work that underpins the
own API (Application Programming Interface), thus promoting inter­ research with a focus on (a) the selection of Engineering Demand Pa­
operability. Nonethelss, current performance-based regulatory frame­ rameters (EDPs), (b) machine learning techniques and (c) dimension­
works still retain some prescriptive features, that don’t fully exploit ality reduction strategies applied to structural engineering.
design optioneering [17,18]. This is a gap where machine learning can
provide informed and actionable solutions. 2.1. Selection of engineering Demand parameters
The paper aims to demonstrate that a performance-based approach
to earthquake structural design supported by machine learning, which Performance-based analysis is adopted through the use of Engi­
involves optimizing the governing variables of a structure, yields an neering Demand Parameters [20]. These are generally identified as
enhanced earthquake resilience than current prescriptive approaches to values resulting from structural analysis which are (i) representative of
earthquake design. The research focusses on reinforced-concrete (RC) the building performance and (ii) used for building regulation compli­
framed buildings and uses an evidence-based methodology to under­ ance [20]. It is therefore through the benchmark of the EDP that
stand the causes that lead to a given state-of-damage through an structural performance is evaluated. However, in order to identify which
analytical and neural network-based approach. The research is based on EDP to consider, a thorough literature review has been conducted.
the 2008 Wenchuan Earthquake in China, with a registered magnitude The literature broadly groups them into two categories, namely local
of Mw 7.9 [19]. or global EDPs according to the structural element to which they refer.
The understanding of the causes of damage can help (a) explore if Examples of local EDPs can be chord rotations or node displacements,
these would have been avoided/attenuated if the structural dimen­ while the inter-storey drift ratio (IDR) is classified as global as involving
sioning was optimized while complying with existing earthquake regu­ the structure in its entirety. Specifically, the IDR is defined as the relative
lations, and (b) inform future practice, including design of new buildings displacement between two consecutive floors divided by the storey
or structural retrofitting of existing ones. height. Research has further shown how the seismic performance of a
The main contributions of this research is a systems engineering building can be represented and benchmarked by adopting
methodology that (a) leverages high precision point cloud data along­ displacement-based indicators [15].
side existing documentation (including drawings), (b) promotes inte­ Grounding on the above considerations, displacement-based EDP are
gration and interoperability with structural software, and thus enables deemed as the most authoritative and reliable method to assess the
multi-aspects engineering analysis, (c) exploits neural networks and seismic performance of a building. However, in order to identify which
genetic algorithm techniques for building features’ investigation, of the EDP to consider for the current research work, an extensive
structural performance prediction and optimization, (d) harnesses on literature review has been conducted to understand their effectiveness
real-time data to promote a more reliable and tailored performance and use across different research. In this context, a series of authoritative
monitoring; and (e) proposes a scalable framework independent from design codes [21]–[24] have been analysed, as well as two relevant

Fig. 1. Comparison between pre-1980 and post-1980 seismic events in terms of (a) Richter magnitude MW, (b) human, and (c) economic losses. The pre-1980 events
included in the review are enumerated as follows: 1-Valdivia, 1960 (Chile)[data source: USGS]; 2- Prince William Sound, 1964 (Alaska) [data source: USGS]; 3-Kam­
chatka, 1952 (Russia) [data source: USGS]; 4- Ecuador-Colombia, 1906 (Ecuador) [data source: USGS]; 5- Rat Islands, 1964 (Alaska) [data source: USGS]; 6-Tang­
shan, 1976 (China) [NGDC] [2]. The post-1980 seismic events considered consist in the following: 1- Rudbar, 1990 (Iran); 2-Izmit (Kocaeli), 1999 (Turkey) [3]; 3-
Kashmir, 2005 (Pakistan) [4]; 4-Sumatra, 2004 (Indonesia) [5,6]; 5-Sichuan (Wenchuan), 2008 (China)[7,8]; 6-Port-Au-Prince, 2010 (Haiti) [9,10]; 7-Tohoku, 2011
(Japan) [11] [additional data source: USGS].

1517
G. Cerè et al. Structures 45 (2022) 1516–1529

research approaches, namely by Ghobarah [25] and Calvi [26] discussed the IDR. With respect to ULS, the ductile capacity and the overall sta­
below. bility must be further investigated with more in-depth techniques which
In order to benchmark the damage on buildings, Limit States are are outside the scope of this research.
defined and associated to specific drift values [15]. Table 1 summarizes Finally, in light of the above analysis it is worth highlighting that the
different limit values regarding the IDR in accordance with a series of regulatory framework chosen for this research is the Eurocodes. This is
authoritative design codes but also research by Ghobarah [25] and Calvi due to a series of factors, such as: (i) integration within the framework of
[26]. Both approaches include IDR as the primary indicator for the IDR; (ii) differentiation of IDR ranges according to limit states; and (iii)
plasticity of the structure. Ghobarah proposes different limit states in more restrictive limitations compared to other regulatory frameworks.
relation to a series of ductility levels of infilled RC frames using the IDR
as an indicator. The advantage of Ghobarah’s approach lies in providing
2.2. Machine learning techniques applied to structural engineering
a coefficient that can be adopted for a quick estimation of damage based
on a formerly performed simulation. Therefore, this approach can be
Research evidences that the adoption of Artificial Neural Networks
useful for instance to provide an overview across a series of buildings,
(ANN) and Genetic Algorithms (GA) is well established in the domain of
but it does not provide a clear quantification or qualification of the
structural engineering [28] for design optimization and damage fore­
damage undergone by the structure.
casting both in static and dynamic conditions.
Calvi devises classes of vulnerability for both RC (i.e., frame) and
Artificial neural networks consist of biology-inspired mathematical
masonry (i.e., load bearing wall) structures developing an approach to
abstractions which rely on the exchange of information across simple
estimate the damage to a building introducing a series of limit states
entities, the neurons [29], which once trained are able to perform high-
then adopted to discretize seismic spectra combined with probability
accuracy data forecasting with a dramatic drop in simulation times.
distribution. Calvi further accounts for the variation of the elastic
Genetic algorithms (GA) represent a class of evolutionary computing
seismic spectra due to energy dissipation through the introduction of a
techniques that use metaheuristic and natural selection for complex
bespoke corrective factor.
optimisation problems. The advantages of GA over other evolutionary
With respect to regulatory frameworks, Eurocode 8 adopts IDR to
computing techniques are manifold. Firstly, the overall simplicity of the
quantify the damage in the context of Serviceability Limit Status (SLS),
optimization process [30], which mainly relies on an iterative selection
while the EDP for ultimate conditions (i.e., ULS) consists in chord ro­
of individuals across different generations based on the best fitness
tations. With respect to IDR limitations, Eurocode 8 distinguishes be­
provided in relation to the minimization objective. Furthermore, GAs
tween elements having different levels of ductility, increasing
provide more accurate results when dealing with multi-modal convex
accordingly from 0.5 % to 1.5 %. Italian building codes differentiate the
problems, enabling a smoother search for the global optimum and
requested verifications according to the building function and the LS.
therefore outperforming traditional approaches [30].
Ministerial Decree 17 January 2018 [24] shows that global rigidity
The numerous applications of intelligent ensembles in the engi­
verifications are requested for specific building functions and limited to
neering domain [28] stretch from a single element to a whole structure
serviceability limit states (SLE). It also distinguishes between the level of
[31,32]. Some research employ GA to optimize the shear capacity and
anchoring of the infill and the frame, providing different limit values for
location of dampers in a structure under seismic actions and adopting

Table 1
Inter-storey drifts limits according to different building regulations and literature.
FEMA 356 [27] DS Immediate Life Safety Near collapse
occupancy
EDP IDR IDR IDR
δ 1 % transient 2 % transient 4 % transient or permanent
or negligible 1 % permanent

Eurocode 8 [22] DS SLS (serviceability) ULS (ultimate)

EDP IDR IDR Chord rotations
δ 0.5 %, 0.75 %, 1 % 0.5 %, 0.75 %, 1
%

SEAOC Vision 2000 [23] DS Fully operational Operational Life Safe Near Collapse Collapse
EDP IDR IDR IDR IDR IDR
δ <0.2 % transient <0.5 % transient <1.5 % <2.5 % transient or >2.5 % transient or
Permanent Permanent transient permanent permanent
negligible negligible <0.5 %
permanent

DM 2018 (Ministerial Decree) – Technical DS SLE (serviceability) SLU (ultimate)

Building Regulations [24]
SLO (operational) SLD (damage) SLV (life safeguard) SLC (collapse)
EDP IDR IDR Ductility, resistance and stability verifications
δ 2/3 IDRSLD 0.5 %, 0.75 %, 1 %

Calvi [26] DS LS1 LS2 LS3 LS4 (collapse)

EDP IDR IDR IDR IDR
δ 0.1 %≤ δ ≤ 0.3 % 0.3 %< δ ≤ 0.5 % 0.5 %< δ ≤ 1.5 % >1.5 %

Ghobarah [25] DS Repairable Irreparable damage Severe damage/Life safe Collapse

damage
EDP IDR IDR IDR IDR
δ 0.2 ≤ δ ≤ 0.4 % 0.4 %< δ ≤ 0.7 % 0.7 %< δ ≤ 0.8 % >0.8 %

1518
G. Cerè et al. Structures 45 (2022) 1516–1529

inter-storey drift ratio (IDR) as a benchmark for performance [33]. The above research questions translates into the following three
Research also evidences the application of machine learning to damage main stages, as illustrated in Fig. 2:
prediction in RC frames though the adoption of a multi-layer ANN [32]
MATLAB and Opensees are combined to perform reliability assessment • Stage 1: Identification of unknown design and as-built parameters.
targeting IDR as a performance indicator; however the rigid ANN ar­ • Stage 2: Structural sensitivity analysis for dimensionality reduction.
chitecture limits the applicability to only one type of building configu­ • Stage 3: Structural behavior optimization using Neural Networks and
ration [34]. Genetic Algorithms.
From an industry perspective, the application of machine learning
tends to be limited in day-to-day engineering practices. Buro Happold The data sources used for the research include:
used Autodesk Robot for topology optimization [35,36]. However, their • Displacement time-series of the Wenchuan seismic event (source:
proposed tool doesn’t have the capacity to perform predictions on IRIS, Wilber 3 database).
structural performance and is limited to the optimization of the building • Acquired 3D point-cloud data of the entire site.
topology. Also, existing commercial solutions such as Galapagos [37] • Photographic materials and on-site observations gathered during
and Karamba [38] suffer from significant limitations, including opti­ two field trips.
mization running time [39], as a consequence of the computational • Satellite imagery (source: Landsat/Copernicus and DigitalGlobe).
demand of their underpinning evolutionary algorithms. As illustrated in Fig. 2, the preliminary knowledge investigation
Overall, it is observed how existing approaches have limited appli­ involved data collection, through a 3D laser scanning campaign, and two
cations in terms of material or analysis strategies, therefore reducing the field investigations in Wenchuan (China), which informed the devel­
scalability of the methodology. Similarly, there is often few imple­ opment of a digital model of the site and a selected subset of buildings, as
mentations of existing simulation software for structural analysis or well as their structural appraisal. Point cloud data has been registered
when those are considered, a consistent lack of integration between the and imported in an architectural/BIM modelling tool to deliver an ac­
two is observed hence resulting in fragmented applications. curate parametric building model. In parallel to these tasks, veloc­
ity–time series (from the IRIS database) and the pseudo-acceleration
2.3. Dimensionality reduction strategies (PSA) spectra in three dimensions (i.e., x,y,z) were collected to simulate
the seismic action once implemented into the structural simulation tool.
Dimensionality-reduction strategies represent a key process when In order to automate the structural analysis and perform the neces­
dealing with relatively large datasets as they allow to identify a smaller sary optimization tasks, the analysis involved reliance on the structural
number of parameters that can account for the variability of the whole simulation tool API, as well as manipulation of the BIM objects and their
dataset [40]. These algorithms are generally adopted in combination properties [45]. The main computational work in all stages (1 to 3)
with other techniques, oftentimes for ANN training in order to avert involved the adoption of artificial intelligence (AI) techniques, namely
overfitting. genetic algorithms (GA) and Artificial Neural Networks (ANN).
In the domain of structural engineering, Principal Component
Analysis (PCA) represents the most established dimensionality reduction 4. Case study: Building investigation
algorithm. It is widely implemented in the domain of vibration analysis
for building seismic performance [41,42]. Here, frequency–response This section provides a description of the Beichuan Hotel located in
functions are generated and then reduced through PCA to be adopted for Old Beichuan in China (Fig. 3a) and a diagnostic of the failure mecha­
ANN training in order to perform data prediction. PCA is also employed nisms triggered by the seismic hazard. This section is provided in the aim
in combination with sensitivity analysis to investigate relevant factors to fulfil the ensuing objectives: (i) to benchmark the proposed building
for local infills seismic failure [43], as well as when adopting modal model with the real case-study structure; (ii) to present the relevant
analysis techniques [44]. building features used to characterize the digital model of the case-study
Linear approaches such as PCA have been proven sometimes to fail to building; (iii) to describe the rationale of the analysis mode employed to
capture the real variability of the dataset when the interrelations be­ obtain the deformed configuration for the modelled building used for
tween the variables are too complex. To resolve these shortcomings, the current research work.
nonlinear approaches can be adopted, including Kernel PCA (KPCA), The Hotel is a masonry-infilled RC frame structure composed of two
Laplacian Eigenmaps, and Maximum Variance Unfolding. Nonlinear main blocks. The Hotel is a masonry-infilled reinforced concrete (RC)
analyses are generally more accurate when dealing with real-world frame structure composed by two main blocks featured by different el­
problems. PCA is widely employed in engineering-related applications evations. The lower block serves as a garage, while the other hosts the
given the ease of use and the reliability of results. reception services. The building does not exhibit lateral-force resisting
Similarly to §2.2, it is worth observing how existing approaches in technologies such as shear walls. The only lateral-force resisting system
structural engineering tend to perform the structural analysis separately which appears to have been implemented at design stage is façade
from the machine learning tasks, leading to time-consuming and re­ masonry infills. However, having these only been provided for part of
petitive calculations that also increase the likelihood for inaccuracies. the façade height, their impact is mostly detrimental to the overall
building behaviour leading in fact to reduce the effective length of
3. Methodology columns and hence increasing their stiffness (and reducing their
ductility as a result).
The methodology involves a case-study based approach using the The building features significant geometric irregularities, both in-
2008 Wenchuan Earthquake. The following research questions (RQ) are plan and in-height which have contributed to the disrupted configura­
posited: tion evidenced in Fig. 3a. The latter exhibits a classic soft-storey failure
RQ1. How structural design parameters can be inferred to accurately due to torsional action resulting from the torque generated by the lever
characterise and model a seismically compromised building? arm created by the centre of mass and the centre of stiffness which –
RQ2. How dimensionality-reduction can be achieved to identify a given the irregular geometry – may be significantly apart from each
subset of parameters that account for the complex behaviour of a other.
structure under earthquake loading? This consideration was then validated by the ensuing simulation
RQ3. Can these sensitive parameters inform the development of illustrated in Fig. 3b. This led to a maximum displacement in the junc­
machine learning techniques to assess and enhance the structural per­ tion between the two blocks, i.e. in the corner column highlighted in
formance of a building? Fig. 3. Fig. 4 further illustrates the geometry of building in its as-built

1519
G. Cerè et al. Structures 45 (2022) 1516–1529

Fig. 2. Overall methodology. GA = Genetic Algorithm, ANN = Artificial Neural Network, PCA = Principal Component Analysis, GUI = Graphic User Interface.

Fig. 3. Beichuan Hotel, following the earthquake (a) and top view of building model with torsional mechanism as per first mode (b).

Fig. 4. Beichuan Hotel, overlapping of point-could and digital model.

configuration prior to the deformation due to 2008 seismic event. storey.

The research focusses on the superstructure of the selected building, Based on the analysed Chinese Building Code it was gathered that for
which based on in-situ surveying doesn’t present any differential- the functionality and typology of the structure, the most likely concrete
settlement failure. This suggests the adoption of a deep foundation class used at design stage would be C30. In terms of the frame members
system. Satellite imagery analysis evidences that the date of construc­ features, the columns have been surveyed to be 45x45 cm of section,
tion was between 2001 and 2010. Consequently, the regulatory frame­ whereas the beams featured a 45x70 cm section. Differently, the roof
work used at design stage was identified as the Chinese Code for Seismic beams dimensions been gathered drawing upon the photographic ma­
Design GB 50011–2001. A first visual assessment helped identify a terial and point cloud data, inferring therefore a member section of
torsional mechanism initiating a soft-storey phenomenon in the first 45x30 cm.

1520
G. Cerè et al. Structures 45 (2022) 1516–1529

The masonry infills are emulated with by-directional struts disposed narrowing down the number of independent variables needed to char­
as a St. Andrew’s cross [46]–[48]. This technique entails the charac­ acterize the masonry properties to one, namely fk.
terization of two diagonally disposed struts and it is widely established
Em = 1000 • fk (3)
as able to reliably account for the masonry stiffening capabilities [27,49]
as presented in Fig. 5.The struts’ features are calculated as presented in
G = 0.4 • Em (4)
Equations (1) and (2), adopting the approach proposed by Stafford-
Smith and Carter [50,51]. Equation (1) shows how the relative infill As a result, the properties used to characterize the masonry infills
to frame stiffness, represented by the product λH, is calculated based on used for this research are the following:
Stafford Smith and Carter formulation [50], where Em and Ec respec­
tively represent the Young modules of masonry and concrete for the • Load: 11 KN/m3;
infill and the frame. I stands for the second area moment of the column • Shear Modulus, G: 680 MPa;
while t and ϑ represent the thickness of the infill and the angle of the • Young Modulus, Em: 1700 MPa.
diagonal to horizontal. The final strut section “a” (where a represents the
section side dimension) is calculated using the Mainstone [52] approach The present research involved the application of linear elastic (i.e.,
formulated by Equation (2) and as shown in Fig. 5, where D represents modal) analysis. This is due to the unavailability of a detailed enough set
the diagonal length between two opposite nodes of the masonry infills. of data to characterize the properties necessary to perform a nonlinear
Equations 1–4 are factored into the optimization algorithm both for simulation. Deploying such algorithm with data inferred featuring an
investigation and resilience enhancement strategies in Stage 3. excessive level of uncertainty would have resulted in inaccurate and
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ unreliable findings.
[ ]
Em • t • sin2ϑ
(1)
4
λH =
4 • Ec • Icol • h 5. Proposed approach for seismic resilience enhancement

a = 0.175 • D(λH)− 0.4

(2) The proposed methodology as described earlier in the paper involves
3 stages: (a) investigation of unknown parameters, (b) sensitivity anal­
Considering that Chinese regulations do not provide a methodology
ysis, and (c) optimum calculation of the unknown parameters. Each of
to estimate the value the masonry characteristic compressive resistance
these stages is elaborated below.
fk and given the low incidence of masonry in the building’s facades, it is
Stage 1: Investigation on unknown design parameters.
herein chosen to adopt the formulation proposed by Eurocode 6 [53].
Table 3 summarizes the scenarios devised to validate the proposed
Therefore, in the absence of site-specific data regarding fk [53,54], the
methodology by simultaneously answering the first research question.
Young and shear moduli are calculated respectively according to
This is attained by performing single (i.e., SO) and multi (i.e., MO)
Equations (3) and (4). Equation (3) shows how to attain the value for
optimization where the objective is the minimization of the discrepancy
masonry elastic modulus Em according to regulations stated in Eurocode
between the EDP value derived from the optimization EDPGA and the
6 [53], and fk was inferred based on experimental trials on blocks
known one EDPREAL, as per Equation (5).Table 4.
equivalent to the ones identified in the site [54].
First, the debris from other buildings located in the same site have Minimize|EDPGA − EDPREAL | (5)
been analysed to identify equivalent blocks in order to define the ty­
Any object within the model is univocally identified by means of a
pology (e.g., hollow, semi-hollow) and the dimensions. A comparative
label. As a result, when manipulating such objects as part of the research
analysis across five different manufacturers producing infill hollow
work, multiple actions are performed which can result in the alteration
blocks of similar dimensions have been performed. The resulting loading
of such labels. As an example, if a new object (e.g., a column) is created
for the Beichuan Hotel masonry blocks was calculated via interpolation
yielding to a value of 11 KN/m3.
Table 3
Grounding on these data, a review of existing literature has been
Scenario description.
conducted in order to determine masonry properties based on the
defined geometry, density and weight [54]. Eventually, given the lack of Scenario Features Description
a mathematical formulation to estimate the shear modulus G in Chinese Name S1_SO_PL
Building Codes, its value has been calculated according to Equation (4), Description The size of main frame elements (i.e., beams and
columns) is investigated without operating on the
material features
Number of variables 4
Design variables Beams and columns’ height (i.e., BH, CH) and width
(i.e., BB, CB)
Constraints Assumption of square columns’ section given field
investigation (CB = CH).
BH > 1.2 CB
30 ≤ BH ≤ 80 cm
30 ≤ CB ≤ 50 cm
CB, BH ∈ ℕ
Variables following to 2: CB (=CH), BH
constraints application
Research question RQ1, RQ2, RQ3
applicability

GA settings
Objective Minimize |EDPGA − EDPREAL |
Number of generations (GA) 30
Population number 80
Average simulation time 45 s
Average optimum number of 10÷15
generations
Stopping criterion |EDPGA − EDPREAL | ≤ 0.001 cm
Fig. 5. Illustration of the equivalent struts method [48].

1521
G. Cerè et al. Structures 45 (2022) 1516–1529

Table 4 Table 3 details the scenario in relation to the genetic algorithm

Scenario 1, cases. SO: single-objective optimization, MO: multi-objective opti­ specifications where the established constraints are evidence-based. It is
mization, PL: provided labels, AL: automated labels’ detection, IDR: inter-storey- therefore assumed that the post-seismic displacement is measurable by
drift ratio; CB,CH = columns’ section sizes; BB,BH = beams’ section width and means, for instance, of in-place sensors or 3D laser scanning. In this
height. specific case, the number of variables has been reduced to two as the
Scenario ID Number of Investigated User Benchmark columns’ section was observed to be square and physically corre­
investigated unknown interaction (EDP) sponding to the beam width. The GA objective is the minimization of the
unknown parameters
discrepancy between the EDP value derived from the optimization
parameters
EDPGA and the known one EDPREAL.
S1_SO_PL 2 CB, BH X Node
Stage 2: Sensitivity analysis and neural network design.
displacement
S1_MO_1P_PL 2 CB, BH X Node This section summarizes the dimensionality-reduction process,
displacement which aids the identification of the key governing variables for struc­
S1_MO_1P_AL variable ALL FRAME Node tural performance. This stage answers the second research question and
SECTIONS displacement it is structured into two related phases: (a) sensitivity analysis and (b)
S1_MO_IDR_PL 2 CB, BH X IDR
ANN engine creation.
The structural building model in Robot is iteratively invoked via
and in order to feature it with all necessary properties such as e.g., COM-based APIs in MATLAB [45] for a total of 1000 simulations, and
material or section, it needs to be associated to a label. Two methodol­ relevant outputs are collected in a matrix. The matrix is then elaborated
ogies are considered in this research for the creation of new labels, one in MATLAB using Principal Component Analysis (PCA) in order to
automated (i.e., automated label, “-AL”) and the other one requiring perform the dimensionality reduction process. The relevant outputs are
user input (i.e., provided labels “-PL”). the variance and coefficient matrixes [55]. However, the number of PCs
Considering the example of the aforementioned label pertaining a to be considered is determined by the total variance represented by the
new section for a column, in the case of “-AL” the algorithm would fetch cumulative sum of each PC’s variance. The number of PCs representing
the section dimensions from the new element and automatically at least 99 % of the entire data set consists in the final PC number that is
construct the string and the resulting label object. On the contrary, the going to be adopted as input for the neural network training.
“-PL” scenario would require the algorithm to prompt the user to insert It is worth noting that we have used 1000 samples of simulations to
the section details as input for the label string and object. benchmark the ANN and the node displacement data is based on the
In terms of optimization strategies instead, another distinction is to results of the individual ANN. The accuracy of the ANN is presented in
be highlighted. Whereas single-objective (SO) optimization aggregates Table 6.
the displacement (or IDR) in its three spatial directions (i.e., x,y,z) in one The output of the neural network is adopted in one case as IDR and in
value consisting in their sum. Conversely, the multi-objective optimi­ the other in terms of node displacement. Both have been used as ob­
zation strategy adopts the displacement (i.e., absolute node displace­ jectives in the first stage of the research (i.e., investigation of unknown
ment or IDR) in each direction as an individual objective. parameters) while in this context, they consist in the target for the ANN
With respect to the automated detection of section labels (i.e., -AL), training. In order to investigate the benefit of adopting multi-layer
case-specific constraints were introduced in the detection algorithm. neural networks, a set of simulations was conducted adopting a trial-
The number of design variables involved in the GA is known based on and-error approach. The selected artificial net consists in a back­
the number of unknown parameters to be investigated, which in this propagation feedforward neural network featured by 4 hidden layers
specific scenario coincides with all the section-related dimensions of the
frame. Based on field investigation, the columns’ section is square (CB =
CH), and its side equals the beams’ section base (CB = CH = BB). Table 5
Therefore, the investigated variables consist of 2 different columns’ ty­ Damage scale addressing earthquake disruption level to RC structures.
pologies (i.e., CB1 and CB2) and 3 beams’ heights (i.e., BH1, BH2, BH3) Damage FEMA 356 SEAOC Vision Eurocode 8 Calvi
based on the building’s topology and 3D laser scanning data. The array V Index 2000 (1999)
providing the input values for each iteration of the GA is represented in D0 – – –
Equation (6); however for the case study, it is considered n = 2 and m = D1 Immediate IDR < 0.2 % IDR ≤ 0.5 Damage ≤
3. occupancy transient % LS1
IDR ≤ 1 % or Permanent IDR ≤ 0.1
V = [CB1 CB2 ⋯CBn BH 1 BH 2 ⋯BH m ] (6) negligible, negligible %
D2 transient or IDR < 0.5 % LS1 <
With each constraint condition representing an inequality, Equation permanent transient Damage ≤
7 can be then formulated in a matrix form. To overcome the case- Permanent LS2
specificity of the constraint definition, Equation 7 shows how the negligible 0.1 % <
IDR ≤ 0.3
different coefficients Ri were calculated based on the model values of the %
investigated variables in order to establish a proportional relationship D3 Life safety IDR < 1.5 % Chord LS2 <
across the dimension of the elements. IDR ≤ 2 % transient rotations Damage ≤
CBi BH transient IDR < 0.5 % LS3
CB1 ≤ Rk , CB1j ≤ Rk i = 2, … n; j = 1…m; k = 1…(i + j-1) (7).
IDR ≤ 1 % permanent 0.3 % <
The final matrix containing the inequalities, hence constraints re­ permanent IDR ≤ 0.5
lationships, is then formulated in Equation (8). This is needed to %
establish a cap for the variation of geometrical features. The values of D4 Near collapse IDR < 2.5 % LS3 <
IDR < 4 % transient or Damage ≤
the individual variables range according to their pre-defined lower and transient or permanent LS4
upper bounds. permanent 0.5 % <
⎡ ⎤ ⎡ ⎤ IDR ≤ 1.5
− R1 1 0 ⋯ 0 0 %
⎢ − R2 0 1 ⋯ 0 ⎥ ⎢0⎥ D5 IDR > 2.5 % Damage >
⎢
⎣ ⋮
⎥V ≤ ⎢ ⎥ (8)
⋮ ⋮ ⋯ 0⎦ ⎣⋮⎦ transient or LS4
− Rk 0 0 ⋯ 1 0 permanent IDR > 1.5
%

1522
G. Cerè et al. Structures 45 (2022) 1516–1529

Table 6
Overview of ANNs adopted to perform the trial-and-error process to identify the optimum structure of the final neural network.
ID Layers Neurons, Layers {: } R2 Training [%] R2 Test [%] R2 overall [%] Results accuracy [%]

IDR-ANN ND-ANN IDR-ANN ND-ANN IDR-ANN ND-ANN IDR-ANN ND-ANN

1 1 20 99.44 97.08 94.62 90.42 98.72 96.04 96 94

2 2 10 98.99 97.65 93.36 89.98 98.13 96.31 91 96
3 2 2 97.65 97.48 95.44 97.36 97.31 97.47 93 85
4 2 8 98.63 94.51 95.18 94.6 98.1 94.52 88 65
5 3 3 97.67 94.9 93.62 94.56 96.99 94.85 83 91
6 3 8 99.06 97.65 90.91 91.03 97.76 96.47 93 92
7 3 10 99.52 99.54 91.23 77.09 98.16 94.34 94 93
8 4 8 99.42 99.5 89.77 84.69 97.8 96.91 92 97
9 4 6 98.8 99.04 93.46 92.5 97.82 97.95 95 95
10 4 20{1}, 1 {2, 4} 98.93 99.04 90.96 89.32 97.65 97.31 96 85
11 4 5{1,2}, 2 {3,4} 98.31 96.23 95.07 95.76 97.82 96.16 98 65
Mean accuracy 92.49 86.90

and trained using a Bayesian regression algorithm, given its more ac­ networks techniques. However, it is observed how a single layer ANN for
curate performance with a potentially noisy data set [31]. The ‘’logsig’’ the IDR provides better performance than a 4-layer ANN even though
activation function is adopted for information exchange between the the results attained by the latter are 2 % more accurate. Fig. 6 shows to
neurons, given its more reliable performance in the case of multilayer this regard the plots for the ANN trained with the IDR target devised for
ANNs [56]. experiment 11. It is observed how a good fit is achieved as the dataset
As part of this stage, it is also noteworthy illustrating the rationale appears well distributed along the 45◦ degrees line. Grounding on these
behind the ANNs devised for the proposed methodology. As presented in observations, the neural network devised for experiment 11 is then
Table 6, 11 different neural network structures were considered in this adopted.
stage and calibrated based on the variables (i.e., EDPs) proposed in stage Stage 3: Optimum calculation of the unknown parameters.
1. This section describes the methodology adopted to replace the
The combination of ANN-GA is then adopted to perform the pro­ structural simulation engine with a mulit-layer intelligent surrogate.
posed experiments. The devised neural networks were trained using the The process starts from the identification of design variables and their
IDR and the node displacement (ND) as targets and calibrated on the initialization. Next, the GA iteratively optimizes the unknown parame­
variables proposed in stage 1. The number of hidden layers and neurons ters thanks to the implemented neural network which replaces the
were adjusted to maximise the performance and reduce the likelihood of structural simulation tool. Analogously to Stage 1, the benchmark pro­
overfitting. This was attained by reducing the number of neurons for cess for the GA entails the minimization of the discrepancy between the
each hidden layer when increasing their number. target IDR and the one attained by the ANN model. A dedicated neural
A comparison between the performance obtained during test, network, the Reinforcement ANN, is then trained in order to provide
training and the overall one is presented in Table 6. Table 6 also outlines reinforcement percentages data based on frame sections and infill fea­
the accuracy of the proposed neural networks in terms of precision tures. The optimum design variables resulting from the GA-ANN utili­
during the calibration stage adopting the variables investigated in stage zation correspond to the input array for the reinforcement ANN. Both
1. Globally, the IDR outperforms the ND in terms of overall performance the Design ANN and the Reinforcement one are trained as part of Stage 2
and accuracy although in some instances, the ND exhibits a lower and the resulting multi-layer ANNs are adopted across this stage to
discrepancy between training and test phases. Despite that, the highest identify the optimum design variables (e.g., frame-related features,
accuracy is attained by experiment 11 adopting IDR as the target, where longitudinal and shear reinforcement percentages).
an accuracy of 98 % is achieved, and an approximate discrepancy be­ The integration of the reinforcement ANN benefits both post-disaster
tween training and test stages consists in about 3 %. structural assessment and investigation phases, but it can also be
The results also evidence how the ND performs better for single or adopted as a predictive performance-based tool for a specific event by
double-layer neural networks, whereas the IDR exhibits a consistently adjusting the performance indicator (i.e., IDR). The proposed strategy
stable performance and an increasing accuracy when introducing neural does not require user input, apart from the Target IDR needed to

Fig. 6. Training (a), Test (b) and overall (c) performance of the Design neural network devised for the 11th experiment.

1523
G. Cerè et al. Structures 45 (2022) 1516–1529

establish building regulatory compliance. As previously outlined in yields to a reinforcement provision which allows to provide the opti­
Section 2, the regulatory framework chosen to validate this research mum performance (i.e., IDR) also taking into account potential geo­
work is the Eurocodes and specifically for shear reinforcement design a metric irregularities of the structure.
capacity design approach has been adopted. In order to factor the complexity of a buildings’ structural system,
The process is iterated until the optimum values are achieved in three different datasets were generated as presented in Fig. 7 by grad­
compliance with existing regulations. In this case, it is considered the ually expanding the set of variables from the geometry to the loads, but
maximum 4 % of reinforcement areas prescribed by Eurocode 2 [57]. also considering infills-related parameters. When the variables are not
Different benchmarks for the IDR were adopted in accordance with the changed, their initial value in the model is preserved and considered for
reviewed authoritative building regulations and the correlation between calculation. The masonry infill load Wi on the frame at each new iter­
building damage, and IDR is presented in Table 5. ation results from the product between the factor R and the load from
In order to fully characterize the frame performance, reinforcement the previous iteration, as shown in Equation (13). This stems from the
data were collected as follows: assumption of constant masonry density, which is an approximation
• Identification of the storey with the highest IDR; adopted for this research. The multiplying factor R is instead defined as
• Within the storey, selection of the most stressed beam Bn and the ratio between the infill thicknesses of two subsequent iterations, as
column Cn based on the maximum value between shear and moment in Equation (13).
achieved for a specific load combination Lc, as in Equations (9) and (10). tti− 1
With respect to beams, the maximum moment and shear is picked be­ R= , Wi = R • Wi− 1 (13)
tt
tween the middle span and supports given the inversion of the diagram
in those positions; 6. Results
Cn = C(argmax(S(LC ), My (LC )) (9)
This section presents the results following the application of the
Bn = B(argmax(S(LC ), My (LC )) (10) proposed methodology. The results reported in this section will address
the following objectives:
• The reinforcement is then calculated in 5 points for the previously
• Investigation of the specific building features through the inte­
selected elements and specifically to the most demanding load combi­
gration of commercial structural behaviour simulation tools and a pro­
nation along the beam/column;
posed machine learning module.
• The maximum value for both shear and moment reinforcement is
• Calculation of optimum values for a specific set of variables which
then selected out of the minimum required values, as outlined in Eqs.
promote a more risk-based seismic design of RC structures.
(11) and (12). Longitudinal reinforcement is represented in percentage
• Damage forecasting and assessment at the building level and
while stirrups are calculated in mm2/m of beam’s length.
consequent cost implications.
( ))
Acol,R = max(min As,R (Cn ) (11)
( )) 6.1. Investigated geometrical frame features
Abeam,R = max(min As,R (Bn ) (12)
Stage 1 entails the investigation of the variables hypothesised as
A consideration worth mentioning is related to the reinforcement
unknown adopting both node displacement and IDR as objective for the
calculation pertaining to each face of the section. On one side – based on
GA. These results are plotted in Fig. 8. Overall, the best-fit results from
the performed site investigations – the original design does not appear to
the scenarios where a specific set of labels are investigated (i.e., -PL),
be reflected in the frame design (e.g., section sizing and reinforcement
rather than where the whole set of building section typologies is
provision) the geometric irregularities as well as the impact on such
involved (i.e., S1_MO_1P_AL). This is perhaps due to the highest vari­
aspect of the seismic spectrum. On the contrary, the proposed algorithm
ables number entailed in the latter scenario and consequently resulting

Fig. 7. Variables and approaches involved in the generation of each data set produced throughout stage 3.

1524
G. Cerè et al. Structures 45 (2022) 1516–1529

ANN). Given the impact of frame elements’ inertia on the seismic per­
formance, the discrepancy between horizontal (i.e., beams) and vertical
(i.e., columns) dimension of structural elements is factored in the opti­
mization process. These results are plotted in Fig. 10 for the three
different datasets displaying an evident correlation between the level of
damage (i.e., IDR) and the section inertia discrepancy. While Fig. 8a and
8b exhibit a consistent trend, Fig. 10c shows an anomaly in relation to
the unconstrained dataset in correspondence to the 2 % IDR, which can
be explained by analysing Fig. 8.
Fig. 8. Comparison between values attained across the different investigation Analysing the inertia of the structural elements separately in Fig. 8 it
algorithms for stage 1. can be observed that (i) the discrepancy across the dataset its maximum
for the 2 % IDR as in Fig. 10c and (ii) beams appear to be mainly
in lower accuracy. Node displacement appears to be a more represen­ accountable for the dynamic inertia discrepancy in seismic conditions.
tative EDP compared to IDR, however, it is worth noting that the latter is In detail, regarding frame elements, a lower level of damage (e.g., IDR
mostly accepted by in-force building regulations. equal to 0.5 %, hence damage level 1/2) could have been achieved
enhancing the columns’ shear capacity in terms of inertia, as it can be
seen in Figs. 10 and 11.
6.2. Optimised values for structural building features Fig. 12 shows the stirrups’ area percentage for the most stressed
structural members for the storey registering the highest IDR as per al­
This section presents the validation and deployment of the meth­ gorithm detection. Fig. 13 presents instead the stirrups’s area in relation
odology proposed for Stage 3, which represents an evolution of Stage 1, to the corner column as highlighted is Section 3. Due to surveying lim­
and entails the integration of the Design ANN into the GA to replace the itations, only the columns reinforcement data are available for com­
structural simulation tool. parison. Fig. 11 shows that across the three datasets a negligible
discrepancy exists between the real stirrup area for columns (i.e.,
6.2.1. As-built values vs neural network results AsB_column) and the percentage resulting from the algorithm.
This section presents the validation of the methodology proposed for On the contrary, Fig. 14 reveals a significant deficiency of longitu­
Stage 3, which represents an evolution of Stage 1, and entails the inte­ dinal reinforcement for columns. Given the non-variability across the
gration of the Design ANN into the genetic algorithm to replace the three datasets in relation to this parameter, Fig. 13 presents only the
structural simulation tool. As anticipated in previous sections, two output regarding the first dataset. Fig. 14 shows a breakdown of section
different EDPs were initially considered, however, the IDR was selected sizing across the three different data sets generated using the GA-ANN
as the most appropriate both because of the results achieved and its framework and compared with the as-built condition (i.e., AsB) rela­
adoption in various building regulations. Removing the equality tively to the IDR value of 2 %, being the effective damage level under­
constraint for section geometry, the number of investigated variables gone by the building. This is limited to the constrained scenario where
grows up to 6. one of the column’s sides (i.e., CH) is imposed as equal to the beam’s
Fig. 6 represents a comparison across the values for the six consid­ base (i.e., BB) and therefore being more consistent for a comparison with
ered building features as attained adopting node displacement and IDR the as-built configuration. It is worth noting that the second data set D2
as EDPs. Fig. 9 shows that despite the increase of variable number, the appears to always provide a more convenient solution rather than one
neural network still provides a considerably reliable level of accuracy. effectively adopted in the building.
This drops though in relation to the masonry characteristic Fig. 15 also highlights the potential for a twofold risk-reduction and
compressive resistance (i.e. fk) and the infills’ thickness (i.e., tt). financial benefit. Considering the current state of damage coinciding
Coherently with the real failure mechanism of the building and its ir­ with a 2 % IDR and comparing it with the sections resulting from the
regularities in elevation, the algorithm selects the first storey as the one second data set and specifically with the 0.5 % IDR, it is evident how an
featured with the highest IDR therefore confirming the consistency be­ optimization-based technique would have entailed an overall approxi­
tween the results attained and the as-built situation. Grounding on the mate increase of 20 % of concrete volume adopted for construction.
above, the ANN adopting the IDR as a benchmark is selected to calculate Correspondingly, an IDR reduction of up to 75 % can be obtained with
the optimum values for the analysed building features. an estimated cost increase of 20 %.
Based on the above results, it is therefore possible to devise the
6.2.2. Calculation of optimum frame and optimum reinforcement optimized column section while benchmarking this with the as-built
This section presents the results of the combination of the Design one, as presented respectively in Fig. 16b and 16a. The optimized col­
ANN in the context of the GA and the subsequent reinforcement calcu­ umn section is devised regarding dataset 1, an IDR of 0.5 % (i.e., damage
lation using the pertinent neural network engine (i.e., Reinforcement levels 1/2) and an overall minimum longitudinal reinforcement area of
1.7218 %. Through the introduction of 12 φ20 bars, an overall per­
centage of longitudinal reinforcement corresponding to 1.752 % is
achieved, satisfying the minimum required area. Thanks to the intro­
duction of 4-arms stirrups, the optimized section exhibits a better
binding through the longitudinal bars compared to the as-built section.
The Chinese seismic code adopted for design, the GB 50010–2001,
prescribes a minimum longitudinal reinforcement area of 0.2 % for each
column’s side. Besides, specifically for corner columns, the requested
minimum total area is 0.8 %. The as-built columns’ longitudinal rein­
forcement for each column’s side approximately corresponds to the
percentage of 0.22 % and the total of 16φ12 in the whole section cor­
Fig. 9. Comparisons between results attained through the artificial neural responds to 0.89 %, proving the regulatory compliance.
network (ANN) engine and the effective values. fk: characteristic compressive With respect to shear reinforcement, the Chinese seismic code GB
resistance of masonry [N/ mm2]; tt: masonry infill thickness [cm]; CB,CH,BB, 50010–2001 provides indications relatively to the minimum re­
BH: frame sections [cm]. quirements for hoops. Considering a hoop diameter φ6, the most likely

1525
G. Cerè et al. Structures 45 (2022) 1516–1529

Fig. 10. Inertia discrepancy of frame section elements (i.e., beams and columns) in relation to the first (a), second (b) and third (c) set of data.

spacing corresponds to either 6 or 8 times the diameter of the hoop, as it

appears lower than 10–15 cm following to site investigations. As a
result, and given the above considerations, the sections were effectively
designed in accordance to in-effect building regulations.

7. Discussion

The previous section has evidenced how the adoption of an

optimization-based approach can effectively benefit seismic design. The
results also suggest that regulatory compliance does not necessarily
guarantee a high level resilience of a structure to seismic stressors. The
Fig. 11. Inertia section values for individual variables (i.e., beams and col­ building characteristics (i.e., frame sizes and masonry properties) and
umns) in relation to the third set of data. performance-related variables were calculated based on an imposed
EDP, namely the inter-storey drift ratio (IDR).

Fig. 12. Stirrups density of most stressed column and beam for the storey registering the highest IDR, respectively relatively to the first (a), second (b) and third (c)
data sets. MS = most stressed.

Fig. 13. Stirrups density corner column and beam for the storey registering the highest IDR, respectively relatively to the first D1 (a), second D2 (b) and third D3 (c)
data sets.

1526
G. Cerè et al. Structures 45 (2022) 1516–1529

corresponding to 2 %, it is evident that the building reinforcement has

been consistently under dimensioned. This is evident from the frame
sections representation, with a clear steel area deficiency. Conversely,
the stirrups density appears to be comparable in relation to the mini­
mum required values, as shown in Figs. 12 and 13. The columns’ sec­
tions calculated adopting the combination of GA-ANN show how the
longitudinal reinforcement percentage displayed in Figs. 12 and 13 can
be distributed across the corner column section. This is relevant when
compared to the as-built situation, reinforcing the effective section in its
current form. Overall, despite regulatory compliance, the as-built
configuration performed inefficiently under the seismic action.
Comparing Figs. 11 and 12, it can noted how both the singularity
Fig. 14. Longitudinal reinforcement percentage of most stressed beam and column and the most stressed element of the storey featuring the highest
column for the storey registering the highest IDR, respectively relatively to the IDR undergo significant shear stress. Nonetheless, columns exhibit a
first dataset. MS = most stressed. higher fluctuation in stirrups density than beams, specifically for the
corner column element. This leads to two considerations: (a) the con­
The final decision to adopt the IDR as main EDP – and to discard the sistency with a seismically stressed RC frame where columns supply the
node displacement – for structural performance assessment was vali­ majority of shear resistance, given the generally higher horizontal
dated by the data presented in this research but it could be expected seismic acceleration component compared to the vertical one; (b) the
given the nature of this particular indicator. As a matter of fact, the IDR coherence between results and the hypothesis of a layout irregularity
provides a relative representation of the structural displacement leading to a concentrated deformation in the portion corresponding to a
whereas the node displacement only accounts for the absolute change in rigidity. This coincides with the height variation in the two
displacement of a particular point (i.e., node) of the structure. Utilizing blocks as in Fig. 1a, which leads to local surge in the shear action of the
the node displacement could have led to an underestimation of the level analysed corner column and globally to torsional mechanisms triggering
of stress for the structural members, particularly relevant for the case- the soft-storey phenomenon.
study proposed in this research where the soft-storey behaviour led to The data set that best fits the frame sizing individually is the first one,
a displacement-induced failure of the structure. and specifically the constrained option, given its consistency with the
Considering the real IDR registered for the Beichuan Hotel, real frame geometry in relation to the equality between the beam’s

Fig. 15. Comparison between as-built (AsB) condition and different frame section sizing across the three data sets in constrained conditions for CB (a), CH = BB (b)
and BH (c).

Fig. 16. Comparison between current column section for a 2% IDR (a) the calculated one for a 0.5% of IDR (b).

1527
G. Cerè et al. Structures 45 (2022) 1516–1529

section base and one column’s side. Nevertheless, the consideration of % of concrete volume. Similarly, referring to Fig. 12 and assuming a
the frame alone in this particular analysis would be inaccurate, given the target IDR of 0.5 % for the second data set, the main increase in mate­
significant contribution to the overall stability and rigidity of the rials, hence in cost, can be attributed to columns. In fact, Fig. 13b and
structure provided by the infills [46,47]. As a result, it would be more 15c illustrate how moving from an IDR of 2 % to 0.5 %, the beam’s
representative for this case-study to consider the third dataset as it section and one of the column’s sides remain also unchanged. However,
factors in the infills’ contribution to the overall structural stability the main increase in material and hence in cost can be attributed to the
favouring the adoption of leaner frame sections and a more cost- other column’s side, as shown in Fig. 15a. This is consistent with the
effective material allocation, but preserving the benefits in terms of above considerations that columns account for the majority of seismic
resilience enhancement. resistance. Therefore, an approximate 20 % of the additional cost could
However, an IDR of 2 % as in the as-built configuration is not a result in a reduction to up to a fourth of the damage.
desirable option. It can, therefore, be acknowledged that in order to
pursue a significantly lower level of damage such as 1 or 2 (i.e., IDR 8. Conclusion
imposed of 0.5 %), it is necessary to adopt a bespoke design that factors
in the dynamic capacity of the structure. Therefore, for an expected IDR The paper argues that earthquakes often affect buildings that did
of 0.5 %, the optimum section parameters can be fetched from Fig. 13a. comply with regulations in force at the time of design and evidences the
In a complementary way, the sections’ second area moment discrepancy potential of adopting machine learning and optimization strategies to
for a target IDR of 0.5 % is significantly low and notably, much lower enhance the seismic resilience of buildings, while promoting an
than the one registered for an IDR of 2 %. This shows the consistency informed cost / risk reduction analysis.
with the hypothesis of a direct relationship between the endured dam­ As such, the paper proposed an evidence-based methodology to
age in the form of IDR and second area moment discrepancy between assess and enhance the seismic resilience of RC buildings using neural
vertical (i.e., columns) and horizontal (i.e., beams) frame elements. networks and optimization techniques, involving three main stages: (i)
Furthermore, an increase in structural resilience is ensured where the investigation of unknown design and as-built variables, (ii) dimension­
second area moment is higher for columns than it is for beams, albeit this ality reduction and sensitivity analysis to determine the variables gov­
appears more evident in the case of the constrained scenario. erning the building dynamic performance, and (iii) calculation of the
The resulting second area moment discrepancy and the section sizing optimum value for a specific set of variables.
show how the structure’s dynamic behaviour would have benefitted The contribution of 3D laser scanning techniques has been extremely
from the adoption of the proposed approach in terms of costs and per­ beneficial to determine accurate measurements in the absence of
formance, based on two factors: (i) the simultaneous compliance to documentation. As such, the proposed approach can also inform struc­
building regulations and (ii) the selection of the optimum option across a tural retrofitting strategies.
much broader set of options compared to the ones that could be The proposed research demonstrates how an multi-layer neural
considered through manual calculations. Nonetheless, it has to be noted network approach and optimization-based technique can effectively
that an expected damage level of 1 or 2 on the proposed scale would lead to resilience enhancement with a clear reduction of the risk factor
entail higher costs given the need to increase the resilience of the to occupants. The proposed methodology can be applied at the building
structure to dynamically withstand the seismic action. level for structural assessment both through the design and post-disaster
This is confirmed by Fig. 15a, where the frame geometry only is diagnostic phases. Overall, the contributions provided by the current
optimized, preserving the existing values for masonry and infill prop­ work to the existing body of knowledge can be summarized as follows:
erties. However, Fig. 15a also exhibits a requirement of significantly
consistent columns’ sections across the three data sets, even in corre­ • Devising a generic and scalable methodology applicable both to new
spondence with the highest IDR where it would be expected for the and existing structures, hence suitable for risk-based design but also
structure to be leaner and more deformable. This is motivated by the in the event of post-disaster assessment;
significantly higher shear stress that columns undergo during seismic • Providing a viable tool for engineers to enhance building structural
hazards. Where the infills are not optimized and preserved in their surveys and thus avoiding often time-consuming bureaucratic pro­
original form, such as in the case of the first dataset, the frame has to cesses and hurdles, exacerbated by the lack of technical
supply the necessary stiffness. However, being unable to do so, a sig­ documentation;
nificant damage increase is registered, justifying the results of the first • Developing a methodology that can be easily integrated with other
dataset. It is evident with respect to the three data sets how the disciplines (e.g., architecture) while promoting a holistic approach to
consideration of infills features benefits more columns than beams, the design/post-disaster assessment of a structure;
confirming the above-mentioned considerations. • Creating a process that can exploits recent advances in both building
It is worth mentioning that it was not possible to retrieve the original engineering and computational structural analysis, augmented with
design material such as calculation reports or blue prints and therefore machine learning;
most of the information pertaining reinforcement provision have been • Presenting a fully scalable methodology which can be used featuring
acquired via site investigation. Based on this, it was possible to several analyses algorithms (e.g., linear, nonlinear) simply by
acknowledge that the reinforcement provision was symmetrical within customizing the relevant parameters through the API.
the section, conversely to what it would have been expected considering
the geometric irregularities of the structures. This yields to consider that In sum, the devised research work stands out given its comprehen­
no distinction was made at the design stage to differentiate in terms of sive approach in tackling structural analysis, as opposed to traditional
solicitation the top and bottom face of the frame members. techniques seldom involving all the above aspects simultaneously.
On the contrary, the proposed algorithm does not pose a constraint in Additionally, its transferability and potential for integration with any
terms of this aspect, allowing the opportunity for the reinforcement and architectural or structural behaviour simulation tool – as well as regu­
section geometry to be combined in the optimum way such to provide latory frameworks – makes it advantageous for practical applications.
the best performance given a target IDR. The proposed research is also suitable for applications using different
The potential for a twofold risk-reduction and financial benefit ap­ analyses algorithms (e.g., nonlinear) from the one presented herein. For
pears when analysing Fig. 15. Considering the current state of damage this scope, the user shall customise the relevant analysis setting via the
coinciding with a 2 % IDR and comparing it with the sections resulting API prior to performing the simulation.
from the second data set, it is evident how an optimization-based Finally, a limitation of this study is identified in the way costs are
technique would have entailed an overall approximate increase of 20 accounted for. Namely, in order to extensively quantify costs it would be

1528
G. Cerè et al. Structures 45 (2022) 1516–1529

necessary to consider the Expected Annual Loss function in conjunction [22] The European Union Per Regulation 305/2011, EN 1998-1:2004+A1:2003.
Eurocode 8: Design of structures for earthquake resistance. 2004.
with a quantity surveying report of the predicted costs connected to the
[23] Structural Engineers Association of California (SEAOC), “Vision 2000: Performance
optimized layout. It is in fact acknowledged that the relationship be­ Based Seismic Engineering of Buildings,” Sacramento, CA (USA), 1995.
tween building’s layout and costs is not linear as there are several [24] Ministry of Infrastructures and Transportations, Ministry of the Interior, and Head
additional cost-related variables to be considered in terms of construc­ of Department of Civil Protection, DM (Ministerial Decree) 17 January 2018. Nuove
Norme Tecniche per le Costruzioni [New technical codes for constructions]. 2018.
tion site, material provision, design and maintenance. Our approach [25] A. Ghobarah, “On drift limits associated with different damage levels,” in
estimates cost reduction based on the amount of construction material Performance-based seismic design-concepts and implementations. vol. 28. Proceedings of
for a specific layout. Future work could consider integrating the devel­ international workshop, 2004, pp. 321–332.
[26] Calvi GM. A displacement-based approach for vulnerability evaluation of classes of
opment of a cost analysis based on material quantity takeoff facilitated buildings. J Earthq Eng Jul. 1999;3(3):411–38.
by a BIM-based design. [27] Federal Emergency Management Agency and American Society of Civil Engineers,
“FEMA 356. PRESTANDARD AND COMMENTARY FOR THE SEISMIC
REHABILITATION OF BUILDINGS,” Washington D.C., 2000.
Declaration of Competing Interest [28] O. I. Abiodun, A. Jantan, A. E. Omolara, K. V. Dada, N. A. E. Mohamed, and H.
Arshad, “State-of-the-art in artificial neural network applications: A survey,”
The authors declare the following financial interests/personal re­ Heliyon, vol. 4, no. 11. Elsevier, p. e00938, 01-Nov-2018.
[29] Ponce-Espinosa H, Ponce-Cruz P, Molina A. Artificial Organic Networks. Springer
lationships which may be considered as potential competing interests: International Publishing Switzerland; 2014.
Yacine Rezgui reports financial support was provided by Cardiff Uni­ [30] Fogel DB. “The advantages of Evolutionary Computing”, in Biocomputing and
versity School of Engineering. Yacine Rezgui reports a relationship with emergent computation. Proceedings of BCEC97 1997.
[31] Tahir Z ul R, Mandal P. Artificial neural network prediction of buckling load of thin
Natural Environment Research Council that includes.
cylindrical shells under axial compression. Eng Struct, Dec 2017;152:843–55.
[32] Morfidis K, Kostinakis K. Approaches to the rapid seismic damage prediction of r/c
Acknowledgments buildings using artificial neural networks. Eng Struct 2018;165:120–41.
[33] Aydin E. A Simple Damper Optimization Algorithm For Both Target Added
Damping Ratio And Interstorey Drift Ratio. Earthquakes Struct 2013;5(1):83–109.
This research is supported by the Building Research Establishment [34] Vazirizade SM, Nozhati S, Zadeh MA. Seismic reliability assessment of structures
(BRE) and the Natural Environment Research Council (NERC) under using artificial neural network. J Build Eng 2017;11:230–5.
grant NE/N012240/1 (Resilience to earthquake-induced landslide risk [35] A. Fisher and S. Sharma, “Exploiting Autodesk Robot Structural Analysis
Professional API for Structural Optimization,” 2010.
in China). [36] Fisher A. Engineering Integration: Real-Time Approaches to Performative
Computational Design. Archit Dig 2012;82:112–7.
References [37] “Galapagos – Grasshopper (Computer software).” [Online]. Available: https://
www.grasshopper3d.com/group/galapagos. [Accessed: 29-May-2019].
[38] Preisinger C, Heimrath M. Karamba – A Toolkit for Parametric Structural Design.
[1] Cerè G, Rezgui Y, Zhao W. Critical review of existing built environment resilience
Struct Eng Int 2014;24(2):217–21.
frameworks: Directions for future research. Int J Disaster Risk Reduct 2017;25:
[39] Tomei V, Imbimbo M, Mele E. Optimization of structural patterns for tall buildings:
173–89.
The case of diagrid. Eng Struct Sep. 2018;171:280–97.
[2] USGS, “July 1976 Tangshan, China Images.” [Online]. Available: https://data.
[40] Van Der Maaten L, Postma E. and J Van Den Herik, “Dimensionality Reduction: A
nodc.noaa.gov/cgi-bin/iso?id=gov.noaa.ngdc.mgg.photos:252. [Accessed: 20-
Comparative Review” 2009.
May-2019].
[41] Bandara RP, Chan THT, Thambiratnam DP. The Three-Stage Artificial Neural
[3] Qorashi M, Jackson JA, Priestley K, Wallace T. The Rudbar-Tarom earthquake of
Network Method for Damage Assessment of Building Structures. Aust J Struct Eng
20 Juin 1990 in NW Persia: Preliminary field and seismological observation and its
2013;14(1):13–25.
tectonic significance, vol. 82, no. 4. Seismological Society of America; 1911.
[42] Li J, Dackermann U, Xu Y-L, Samali B. Damage identification in civil engineering
[4] Shafique M, van der Meijde M, Khan MA. A review of the 2005 Kashmir
structures utilizing PCA-compressed residual frequency response functions and
earthquake-induced landslides; from a remote sensing prospective. J Asian Earth
neural network ensembles. Struct Control Heal Monit Mar. 2011;18(2):207–26.
Sci 2016;118:68–80.
[43] Šipoš TK, Sigmund V, Hadzima-Nyarko M. Earthquake performance of infilled
[5] Ishii M, Shearer PM, Houston H, Vidale JE. Extent, duration and speed of the 2004
frames using neural networks and experimental database. Eng Struct 2013;51:
Sumatra-Andaman earthquake imaged by the Hi-Net array. Nature Jun. 2005;435
113–27.
(7044):933–6.
[44] Han S, Feeny BF. Enhanced Proper Orthogonal Decomposition for the Modal
[6] Klein RJT, Nicholls RJ, Thomalla F. Resilience to natural hazards: How useful is
Analysis of Homogeneous Structures. Modal Anal Jan. 2002;8(1):19–40.
this concept? Environ Hazards Jan. 2003;5(1):35–45.
[45] Cerè G, Zhao W, Rezgui Y. “Structural behavior analysis and optimization,
[7] Chan EYY. Bottom-up disaster resilience. Nat Geosci May 2013;6(5):327–8.
integrating MATLAB with Autodesk Robot”, in 35th CIB W78 Conference: IT in
[8] Guo Y. Urban resilience in post-disaster reconstruction: Towards a resilient
Design. Construction, and Management 2018:379–86.
development in Sichuan, China. Int J Disaster Risk Sci Mar. 2012;3(1):45–55.
[46] M. Dolce, “Commentary to Ministerial Decree 16.1.1996 and Circ. n.65/AA.GG. del
[9] Cavallo E, Powell A, Becerra O. Estimating the Direct Economic Damages of the
10.4.1997 of the Minister of LL. PP.,” F. Braga, Ed. pp. 9.1-9.53.
Earthquake in Haiti. Econ J Aug. 2010;120(546):F298–312.
[47] Di Trapani F, Macaluso G, Cavaleri L, Papia M. Masonry infills and RC frames
[10] Bilham R. Lessons from the Haiti earthquake. Nature Feb. 2010;463(7283):878–9.
interaction: literature overview and state of the art of macromodeling approach.
[11] Daniell JE, Vervaeck A. A timeline of the socio-economic effects of the 2011
Eur J Environ Civ Eng 2015;19(9):1059–195.
Tohoku Earthquake with emphasis on the development of a new worldwide rapid
[48] G. Al-Chaar, “Evaluating Strength and Stiffness of Unreinforced Masonry Infill
earthquake loss estimation procedure. Australian Earthquake Engineering Society
Structures,” 2002.
2011.
[49] Mohammad AF, Faggella M, Gigliotti R, Spacone E. Seismic performance of older
[12] Meacham BJ. Risk-informed performance-based approach to building regulation.
R/C frame structures accounting for infills-induced shear failure of columns. Eng
J Risk Res Oct. 2010;13(7):877–93.
Struct Sep. 2016;122:1–13.
[13] Foliente GC. Developments in performance-based building codes and standards.
[50] Stafford Smith B, Carter C. A method of analysis for infilled frames. Proc Inst Civ
For Prod J 2000;50(7/8):12–21.
Eng, Sep 1969;44(1):31–48.
[14] Aktan EA, Ellingwood BR, Kehoe B. Performance-Based Engineering of Constructed
[51] Stafford Smith B. Behavior of Square Infilled Frames. J Struct Div 1966;92(1):
Systems. J Struct Eng 2007;133(3):311–23.
381–404.
[15] Fragiadakis M, Lagaros ND. An overview to structural seismic design optimisation
[52] Mainstone RJ. On the Stiffness and strength of infilled frames. Proc Inst Civ Eng
frameworks. Comput Struct Jun. 2011;89(11–12):1155–65.
1971;no. Supplement IV:57–90.
[16] Fragiadakis M, Lagaros N, Manolis P. Performance-based earthquake engineering
[53] The European Union Per Regulation, EN 1996-1-1 (2012): Eurocode 6: Design of
using structural optimisation tools. Int J Reliab Saf 2006;1(2):59–76.
masonry structures – Part 1-1: General rules for reinforced and unreinforced masonry
[17] May PJ. Performance-Based Regulation and RegulatoryRegimes: The Saga of Leaky
structures. 2005.
Buildings. Law Policy 2003;25(4):381–401.
[54] N. Carnal, “Resistenza meccanica di blocchi forati a fori orizzontali,” 2006.
[18] Roësset JM, Yao JTP. State of the Art of Structural Engineering. J Struct Eng Aug.
[55] Jolliffe IT. Principal Component Analysis. 2nd ed. New York Inc: Springer-Verlag;
2002;128(8):965–75.
2002.
[19] Pei S, et al. Seismic velocity reduction and accelerated recovery due to earthquakes
[56] “Multilayer Shallow Neural Network Architecture -MATLAB.” [Online]. Available:
on the Longmenshan fault. Nat Geosci May 2019;12(5):387–92.
https://uk.mathworks.com/help/deeplearning/ug/multilayer-neural-network-
[20] Terrenzi M, Spacone E, Camata G. Collapse limit state definition for seismic
architecture.html. [Accessed: 26-Jun-2019].
assessment of code-conforming RC buildings. Int J Adv Struct Eng Sep. 2018;10(3):
[57] The European Union Per Regulation 305/2011, “EN 1992-1-1 (2004): Eurocode 2:
325–37.
Design of concrete structures – Part 1-1: General rules and rules for buildings,”
[21] ASCE and FEMA, “Prestandard and commentary for the seismic rehabilitation of
2010.
buildings,” 2000.

1529