Automation in Construction 123 (2021) 103547

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Digital twin-based collapse fragility assessment of a long-span cable-stayed

bridge under strong earthquakes
Kaiqi Lin a, b, You-Lin Xu a, *, Xinzheng Lu c, Zhongguo Guan d, Jianzhong Li d
a
Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, Hong Kong, China
b
College of Civil Engineering, Fuzhou University, Fuzhou, China
c
Key Laboratory of Civil Engineering Safety and Durability of China Education Ministry, Department of Civil Engineering, Tsinghua University, Beijing, China
d
Department of Bridge Engineering, Tongji University, Shanghai, China

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

Keywords: Fragility analysis is widely used in seismic performance assessment of bridges but uncertainties in seismic de­
Digital twin mand and bridge modelling affect the accuracy of assessment results. This study proposes a digital twin-based
Long-span cable-stayed bridge collapse fragility assessment method for long-span cable-stayed bridges under strong earthquakes. A scaled
Shake table tests
long-span cable-stayed bridge and its shake table tests are taken as an example. Three finite element (FE) models
Seismic collapse assessment
Fragility analysis
of the bridge, including a design document-based FE model, a linearly updated FE model, and a nonlinearly
Modelling uncertainties updated FE model, are established to demonstrate the necessity of the digital twin-based assessment. Incremental
dynamic analyses (IDA) are conducted to calculate the collapse fragility curves of the FE models. The assessment
results are compared with the test results in terms of collapse mechanisms, collapse ground motion intensities,
and collapse probabilities. It is found that the proposed method is feasible and accurate for seismic collapse
assessment of long-span cable-stayed bridges.

1. Introduction based on different intensity measures (IMs) and highlighted that a

proper IM is important for accurate evaluation of bridge seismic
Long-span cable-stayed bridges have been constructed around the performance.
world in recent decades, and some of them are built in earthquake-prone The seismic performance assessment relies on the quantification of
regions. The seismic performance assessment of these bridges thus be­ the uncertainties in both the seismic demand and the bridge modelling
comes an important issue to engineering community. Incremental dy­ in fragility analysis. The seismic demand uncertainties result from the
namic analysis (IDA)-based fragility analysis method [1] is currently the randomness of the ground motions, which stems from the ground mo­
most widely used method in evaluating the seismic performance of long- tion record variability and epistemic uncertainty [1]. The uncertainties
span bridges [2–10]. Casciati et al. [2] and Barnawi and Dyke [3] in the bridge modelling include (1) the uncertainties in modelling ge­
compared the seismic performance of cable-stayed bridges equipped ometries, material properties, connections, and boundaries; and (2)
with different seismic response control systems through the comparison modelling assumptions and omissions such as the use of simplified ele­
of their corresponding seismic fragility curves. Pang et al. [4] compared ments and elastic connection assumptions during the modelling process.
the effects of different sources of uncertainties on the seismic fragility of The uncertainties in seismic demand can be reduced by selecting several
cable-stayed bridges using the uniform design method. Li et al. [5] proper ground motions in fragility analysis [2–10], but the uncertainties
figured out the seismic fragility of a multi-span cable-stayed bridges in the bridge modelling are not fully addressed. In fact, most of the
using the IDA and compared the effects of various deck-pier connections current studies use the design document-based finite element (FE) model
on the seismic performance of the bridges. Zhong et al. [7] and Li et al. of the bridge in their fragility analyses. This will significantly affect the
[8] worked out the seismic fragility of long-span cable-stayed bridges accuracy of the seismic performance assessment of long-span bridges,
under multi-support ground motion excitations. Zhong et al. [6] and Wei which is particularly true for the seismic collapse assessment because
et al. [10] compared the fragility analysis results of cable-stayed bridges the physical bridge suffers serious damage and its structural behaviour is

* Corresponding author.
E-mail addresses: linkq@fzu.edu.cn (K. Lin), ceylxu@polyu.edu.hk (Y.-L. Xu), luxz@tsinghua.edu.cn (X. Lu), guanzhongguo@tongji.edu.cn (Z. Guan), lijianzh@
tongji.edu.cn (J. Li).

https://doi.org/10.1016/j.autcon.2020.103547
Received 15 September 2020; Received in revised form 10 December 2020; Accepted 30 December 2020
Available online 16 January 2021
0926-5805/© 2021 Elsevier B.V. All rights reserved.
K. Lin et al. Automation in Construction 123 (2021) 103547

strongly nonlinear and plastic. intensities, and collapse probabilities. These results are further
An effective way to reduce the modelling uncertainties is to update compared with the collapse test results of the physical model of the
the design document-based FE model with real-time measurement data. bridge to demonstrate the feasibility and accuracy of the proposed
Modern long-span cable-stayed bridges have been often equipped with digital twin-based collapse fragility assessment method for long-span
advanced structural health monitoring (SHM) systems that provide cable-stayed bridges.
extensive real-time measurement data of the bridge under various
loadings [11,12]. There have been many studies on the model updating 2. A digital twin of a long-span cable-stayed bridge
of long-span bridges based on the SHM data in the literature. Benedettini
and Gentile [13] identified the model parameters of a cable-stayed 2.1. A physical model of the bridge
bridge by using the measurement data from servo-accelerometers and
radar vibro-meters and produced the updated 3D FE model of the bridge. In this study, the collapse fragility curves of a physical scaled model
Wang et al. [14] updated the deck modulus of a multi-scale FE model of of the Sutong Bridge are produced in terms of the digital twin technol­
a long-span suspension bridge by using the gradient-based optimization ogy. The Sutong Bridge is currently the longest cable-stayed bridge in
algorithm and fuzzy theory. Xiao et al. [15] and Wang et al. [16] China. The prototype bridge has a main span of 1088 m and two towers
updated various parameters of a multi-scale FE model of a long-span of 306 m high, as shown in Fig. 1a. A series of shake table tests of a
cable-stayed bridge using influence lines and dynamic characteristics physical scaled model of the bridge were conducted to investigate
in terms of the response surface method and the kriging meta-model, damage initiation, damage propagation and collapse of the bridge under
respectively. The above-mentioned model updating is, however, linear earthquakes [25,26]. The physical model has a scale ratio of 1:35, and its
model updating. Recently, a cluster computing-aided FE model updating overall dimensions are shown in Fig. 1b. The scaled tower has a height of
method was proposed by Lin et al. [17–19] for both linear and nonlinear 8.58 m and the total span is scaled down to 59.66 m with a main span of
updating of large civil structures. The measurement data from a shake 31.09 m. Both the towers and piers are fixed at the concrete foundations,
table test of a scaled long-span cable-stayed bridge was used in the which are in turn fixed on the shake table. Because the geometric scale
model updating and the validation of the proposed method. The factor is 1/35 and the acceleration scale factor equals 1.0, the time scale
development of model updating techniques facilitates the applications factor is 1/350.5 according to the similarity rules.
of digital twins in performance assessments of large civil structures.
A digital twin is a digital replica of a living or non-living physical 2.2. Shake table tests
entity, characterized by cyber-physical interaction. The concept of dig­
ital twins was proposed by Grieves and Vickers [20] and then revisited The shake table tests of the scaled bridge were conducted in the
by the National Aeronautics and Space Administration (NASA) of the US Multi-Functional Shaking Table Laboratory of Tongji University
in 2012. A digital twin contains four essential elements: (1) modelling or [25,26]. The arrangement of sensory system is displayed in Fig. 1c, and
simulation of a physical entity to have a virtual entity; (2) data collection the typical sectional details of the main tower and the steel girder are
and fusion from the physical entity; (3) interaction and collaboration provided in Fig. 1d. During the collapse tests, the ground motion in­
between the physical and virtual entities; and (4) service provided by tensities, as listed in Table 1, were gradually increased to investigate
the virtual entity [21]. Clearly, the most privileged advance of a digital damage initiation, damage propagation, and collapse of the bridge [25].
twin is to closely and timely reflect the actual service state of a physical The scaled acceleration and velocity time histories of the artificial site
entity, thereby reinforcing information-based services during the enti­ ground motion used in the test case E15 are shown in Fig. 2a. The cor­
ty’s lifecycle. Boje et al. [22] reviewed the development of building responding 5%-damped response spectrum is calculated and shown in
information modelling (BIM) and digital twins and concluded that dig­ Fig. 2b. For the test cases listed in Table 1, the girder movement was
ital twins require a more increased level of details than BIM to help constrained in the transverse direction at both pier-girder and tower-
monitor and control the physical entities throughout their lifecycle. girder connections.
Shim et al. [23] presented a design concept for a digital twin-based
maintenance system for bridges. Lu and Brilakis [24] proposed a 2.3. Digital twin development
slicing-based object fitting method for generating the geometry of a
digital twin for existing reinforced concrete bridges from labelled point To investigate the effects of uncertainties in bridge modelling and to
cluster. Some attempts have been also made to implement the updated demonstrate the importance of digital twin technology in seismic
FE model in the seismic fragility analysis of long-span bridges. Cheng collapse assessment of the scaled bridge, three FE models of the scaled
et al. [9] linearly updated the FE model of a self-anchored suspension bridge are established with reference to the test arrangement, as shown
bridge based on the SHM data and compared the fragility curves ob­ in Fig. 3. The first FE model (Model-O) of the scaled bridge is built ac­
tained from the initial FE model and the updated model. The compar­ cording to the archive data from the design document of the scaled
ative results indicated that the updated model is more vulnerable than bridge. The second FE model (Model-L) is achieved through the model
the initial model to the same earthquake excitation. However, only updating of Model-O using the archive (measurement) data from the test
linear model updating was considered in Cheng et al. [9] while the case E6. Since the scaled bridge is in linear condition in the test case E6,
uncertainties in the nonlinearity of the bridge for predicting the bridge’s only linear modelling parameters of Model-O are considered and Model-
seismic performance were not considered. L is a linearly updated model. The third FE model (Model-NL) of the
In this study, a digital twin-based collapse assessment method is scaled bridge is obtained through the model updating of Model-O using
proposed for assessing the seismic collapse performance of a long-span the archive (measurement) data from the test case E13. Since the scaled
cable-stayed bridge subject to strong earthquakes. To demonstrate and bridge is in nonlinear condition in the test case E13, both linear and
verify the proposed method, a scaled long-span cable-stayed bridge and nonlinear modelling parameters of Model-O are considered and Model-
its shake table tests are taken as an example. The test data of the physical NL is a nonlinearly updated model. The test case E15, in which the
scaled bridge are used for producing digital twins of the bridge through physical model of the bridge collapsed, is taken as a case to validate the
the linear and nonlinear model updating of the design document-based proposed collapse fragility assessment method. The above-mentioned
FE model. Seismic fragility analyses of the three different FE models are modelling and updating of the bridge are further illustrated below.
then conducted in terms of the IDA method to generate their collapse
fragility curves. The seismic collapse performance assessments based on 2.3.1. FE modelling of the bridge
the three different models are then conducted, and the results are The design document-based FE model of the scaled bridge is estab­
compared in terms of collapse mechanisms, collapse ground motion lished using the FE software package MSC.Marc [27]. Various types of

2
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 1. Shake table tests of the scaled bridge.

3
K. Lin et al. Automation in Construction 123 (2021) 103547

Table 1 Model-O.
Test arrangement of the scaled bridge.
Test Input ground PGA Test Input ground PGA 2.3.2. Model updating program
case motion (g) case motion (g) Based on Model-O, the cluster computing-aided FE model updating
E1 White noise 0.1 E9 Artificial site 0.7 program developed by Lin et al. [17–19] is adopted for both linear and
ground motion nonlinear updating of the physical model of the bridge. The updating
E2 Artificial site 0.1 E10 White noise 0.1 program is developed based on three software packages, MATLAB, MSC.
ground motion Marc and Python. The main program is developed in MATLAB to inter-
E3 Artificial site 0.2 E11 Artificial site 0.9
ground motion ground motion
connect different software and make use of their respective function
E4 Artificial site 0.3 E12 White noise 0.1 strength. During the model updating process, MSC.Marc is responsible
ground motion for either the modal analysis in linear updating or the time history
E5 Artificial site 0.4 E13 Artificial site 1.1 analysis in nonlinear updating; Python is used to collect the FE analysis
ground motion ground motion
results from the output data files generated after the FE analysis in MSC.
E6 White noise 0.1 E14 White noise 0.1
E7 Artificial site 0.5 E15 Artificial site 1.3 Marc; and MATLAB is responsible for evaluating the objective functions
ground motion ground motion based on the FE results, updating the modelling parameters, and
E8 White noise 0.1 distributing the subsequent FE analysis tasks to the computing cluster. A
PGA: peak ground acceleration. 115-core computing cluster, which comprises 5 high-performance
servers equipped with dual Intel Xeon E5–2670 v3 @2.30GHz pro­
cessors, is used to parallelly run the FE analysis tasks and accelerate the
elements are used to simulate different bridge components in consid­
updating process. Note that an individual modal analysis or time history
eration of their compatibility, as shown in Fig. 4 [17]. A sensitivity
analysis of the FE model in a single core of the cluster requires a memory
analysis is conducted to determine a proper mesh size for various types
usage of about 1.0 GB. Hence, a total of 115 GB memory is required for
of elements. After the meshing, the FE model has an element number of
the 115-core computing cluster when performing the model updating.
more than 16,000 and a node number of more than 11,500. This model
More details of the model updating program, including part of the source
contains all the attribute information of the bridge and it is denoted as
codes, can be found in Lin et al. [17–19].

Fig. 2. Details of the artificial site ground motion in the test case E15.

Fig. 3. Test arrangement and the FE models.

4
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 4. FE modelling of the scaled bridge.

2.3.3. Digital twins girder are used as the archive data to update the modelling parameters
Based on Model-O, the archive data recorded by the data acquisition of Model-O.
system during the shake table tests are used to update Model-O to pro­ Compared to linear updating, in addition to the linear modelling
duce two different digital twins of the bridge. A linearly updated digital parameters, a few nonlinear modelling parameters are updated to derive
twin, denoted as Model-L, is derived based on the measurement data Model-NL, including the reloading stiffness and residual strength of the
obtained from the test case E6 (white noise, peak ground acceleration concrete for the towers and piers, the yield strength of the reinforcement
(PGA)=0.1 g), while a nonlinearly updated digital twin, denoted as steel, structural damping ratio and others. Furthermore, to consider the
Model-NL, is derived based on the test case E13 (artificial site ground unevenly distributed seismic damage on the concrete bridge tower, the
motion, PGA=1.1 g). elements of two main towers are divided into several subsets and the
The modal analysis-based linear model updating method is adopted material parameters in different subsets are updated independently.
to update the linear modelling parameters of Model-O to have a linear Hence, compared to the linear updating, a total of 50 modelling pa­
digital twin Model-L. Note that in many existing studies, hammer test is rameters are considered in the nonlinear model updating of Model-NL.
adopted to identify the modal properties of the bridge [28–31]. How­ More details of the nonlinear updating process can be found in Lin
ever, in the shake table tests of Guan et al. [25], only ground motion et al. [18]. The comparisons of the structural responses at some key
tests were conducted so that the measurement data under white noise measurement points between the digital twin and the physical model of
ground motion case (test case E6) are used to identify the modal pa­ the bridge are depicted in Fig. 6, in which the measurement point IDs
rameters for the linear model updating. can be found in Fig. 1c, and the comparative results are found satis­
Many different methods can be used to identify the modal parame­ factory. Moreover, the updated model Model-NL is further used in the
ters from the measurement data, such as the eigensystem realization collapse prognosis of the bridge under the subsequent ground motion
algorithm [32], the power spectral density function method [29], (the test case E15), as shown in Fig. 7 [18].
discrete analytical mode decomposed method [33], the stochastic sub­
space identification (SSI) method [34] and so on. In this study, the SSI 3. Collapse fragility curve development
method is used to process the measurement data and identify the natural
frequencies and modal shapes of the physical model of the bridge The above-mentioned three FE models (Model-O, Model-L and
because the SSI method has been widely used for the modal identifica­ Model-NL) are used to calculate the collapse fragility curves of the
tion of civil structures [35–39] and many studies have proved the bridge using IDA. Several ground motions are selected based on the
feasibility and accuracy of the SSI method. The identified modal infor­ design seismic response spectrum of the bridge. The intensities of the
mation is further used to update Model-O and obtain Model-L. During ground motions are gradually increased until the collapse of each of the
the updating, 11 linear modelling parameters are considered, which three FE models. The collapse ground motion intensities are then taken
include the initial stiffness of the materials for the steel deck steel and to form the fragility curves by using the lognormal distribution. The
the concrete tower, the thickness of the deck, the mass and the stiffness collapse damage locations of the bridge under different ground motions
of the girder-tower and girder-pier connections. The details of the linear are also recorded to investigate the potential collapse mechanisms of the
updating process can be found in Lin et al. [17]. After updating, the three different FE models. To keep consistent with the shake table test,
natural frequencies of the digital twin and the physical model for the test the ground motions in the collapse fragility analyses are input uniformly
case E6 are compared in Table 2 and the mode shapes of the digital twin to all the piers and towers along the transverse direction of the bridge,
are shown in Fig. 5. while the site-structure interaction and the spatially varying character­
The digital twin Model-NL is derived by the time history analysis- istics of ground motion are not considered.
based nonlinear model updating. The measured displacement time his­
tories of the bridge from the test case E13 at the two main towers and the 3.1. Definition of collapse

Table 2 There are two commonly-used definitions of collapse of a structure.

Modal analysis-based updating results of Model-L. One definition is that the values of one or multiple engineering demand
Modal shape Identified Updated Relative parameters (EDPs) exceed the predefined thresholds [4–10,40,41].
frequency frequency error Nevertheless, because the EDP-based collapse definition often considers
1st symmetric transverse 0.5932 Hz 0.5944 Hz +0.20% a fixed deformation threshold for a specified type of structural compo­
bending of the girder nents, it is thus possible that one component is defined as “failed” while
1st transverse bending of the 2.1756 Hz 2.1673 Hz − 0.38% the entire structure does not fail and can still withstand further seismic
south tower
loads. In the second definition of collapse, the structural deformation
1st transverse bending of the 2.3399 Hz 2.3500 Hz +0.43%
north tower keeps increasing along one direction until the point of dynamic insta­
2nd symmetric transverse 2.5082 Hz 2.4938 Hz − 0.57% bility of the structure occurs [42–44]. The second definition of collapse
bending of the girder is used in this study, and the collapse simulation of the bridge is ach­
3rd symmetric transverse 3.5389 Hz 3.5334 Hz − 0.16% ieved by using the elemental deactivation technique [33]. The material
bending of the girder
level failure criteria are adopted to determine the deactivation of the

5
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 5. Identified modal shapes of the bridge.

Fig. 6. Comparison of the response time histories [18].

elements. The strain values at all integration points of the elements are connections and pier-girder connections, respectively [25], and these
checked against the predefined failure criteria at each increment of the components remain elastic during the tests. Hence, they are modelled
time history analysis. According to the experimental observations dur­ with elastic spring elements with updated stiffness and considered not to
ing the shake table tests, the steel girder remains elastic throughout the fail in the collapse simulation. Apart from the above-mentioned ele­
tests. Moreover, the girder movement was constrained in the transverse ments, the failure of other reinforced concrete components and cables is
direction with steel retainers and fixed bearings at the tower-girder considered. For the reinforced concrete, the concrete crushing is

6
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 7. Collapse prognosis of the bridge [18].

considered by the softening branch of the material constitutive model damping ratio equalling to 5% are compared with the design seismic
approaching zero. The ultimate tensile strain and compressive strain of response spectrum of the prototype bridge in Fig. 8a. Note that the
the reinforcing steel are 15% and 10%, respectively. For the stayed ca­ physical bridge is a scaled physical bridge model, and therefore the time
bles, a failure tensile strain of 10% is adopted [18]. Typical displacement of the selected ground motions is compressed to 1/350.5 before they are
time history of the main tower entering dynamic instability or collapse is input to the FE models. After scaling, the response spectra of the selected
shown in Fig. 7. ground motions are depicted in Fig. 8b.

3.2. Ground motion database 3.3. Intensity measures

In the seismic fragility analysis, the uncertainties in seismic loads are In performance-based earthquake engineering, the prediction of
reduced by selecting an adequate number of ground motions. There seismic performances is dependent on the IMs of the ground motions
have been many studies on the ground motion selections for seismic [6,10,48]. Various IMs have been used to investigate the seismic
fragility analysis of cable-stayed bridges, as summarized in Table 3. The fragility of cable-stayed bridges as summarized in Table 3. Wei et al.
number of selected ground motions for fragility analysis is generally [10] evaluated 16 commonly-used IMs in terms of practicality, effi­
between 8 and 100. It is often believed that 10 to 20 ground motion ciency, proficiency, sufficiency, and hazard computability. They
records are good enough to estimate the seismic demands of a structure concluded that peak gound velocity (PGV) is most suitable in the seismic
[1,40]. In this study, a total of 30 ground motion records are selected fragility analysis of cable-stayed bridges with super-high piers. How­
from the NGA-West2 ground motion database of the Pacific Earthquake ever, different cable-stayed bridges have different structural configura­
Engineering Research Center (PEER) according to the design seismic tions so that their dynamic characteristics and collapse mechanisms vary
response spectrum of the prototype bridge. In consideration of the site significantly. As a result, the optimal IMs also diverge for different
conditions of the prototype bridge, the selected ground motions have a bridges. In this study, four types of IM are selected in the collapse
moment magnitude between 6.0 and 8.0, and a V30 s ranging from 100 m/ fragility analyses of the three FE models, which include PGA, PGV,
s to 200 m/s, where V30 s is the average shear wave velocity of the top 30 Sa(T1) and Sa(T2), in which Sa(T1) and Sa(T2) are the 5%-damped spec­
m of soil. More details of these ground motions are summarized in tral accelerations corresponding to the first transverse vibration period
Table 4. The response spectra of the selected ground motions with a of the girder and the main tower, respectively. During the collapse
fragility analysis, the intensities of selected ground motions are gradu­
Table 3 ally increased with an interval of 0.05 g in terms of PGA up to the
Recent studies on ground motion selections for seismic fragility analyses of collapse of the bridge. Furthermore, the collapse intensity values cor­
cable-stayed bridges. responding to other IMs are calculated based on the ground motion in­
No. Author Number of ground Database IM puts with the collapse-related PGAs.
motions

1 Casciati et al. [2] 24 Synthesized Sa 3.4. Collapse fragility curves

2 Barnawi and Dyke 20 Synthesized Sa
[3]
3 Noghabi et al. [40] 15 PEER Sa The extensive collapse fragility analyses of the bridge are performed
4 Pang et al. [4] 56 PEER Sa based on the three FE models, the selected ground motions and IMs, the
5 Zhong et al. 100 PEER and SAC PGA definition of collapse, and the collapse analysis method. The collapse
[6,7,45] project probabilities of the bridge under the 30 selected ground motions are
6 Wu et al. [41] 100 PEER Sa
7 Li et al. [5] 80 PEER Sa
plotted in Fig. 9 against the increasing IMs. Fig. 9a to d are the results
8 Pang [46] 8 / PGA corresponding to PGA, PGV, Sa(T1) and Sa(T2), respectively. Further­
9 Li et al. [8] 80 Synthesized PGA more, the commonly-used lognormal distribution [2,4,10] is used to
10 Li et al. [47] 20 PEER PGV regress the data points for different IMs.
11 Wei et al. [10] 80 PEER PGV
It can be found from Fig. 9 that whichever IM is used, the fragility
PGV: peak ground velocity. curves of Model-L fall on the right side to those of Model-NL. It is hence

7
K. Lin et al. Automation in Construction 123 (2021) 103547

Table 4 selecting a proper IM in the collapse fragility analyses of a long-span

Details of the selected ground motions in this study. cable-stayed bridge. These two issues will be discussed in the
No. Event Year Station Ma Rrupb PGA following sections.
(km) (g)

EQ1 Imperial Valley, 1979 EC County Center 6.5 7.31 0.212 4. Digital twin-based collapse fragility assessment
USA FF
EQ2 Imperial Valley, 1979 El Centro Array 6.5 12.56 0.367 The performance-based earthquake engineering aims to provide an
USA #11 accurate estimation of the structural performance of the bridge under
EQ3 Imperial Valley, 1979 El Centro Array 6.5 17.94 0.145
USA #12
future earthquakes and thereby help stakeholders make their decisions.
EQ4 Imperial Valley, 1979 El Centro Array 6.5 12.85 0.267 In the shake table tests of the physical model of the bridge [25], the
USA #3 ground motion intensities are gradually increased until the collapse of
EQ5 Superstition 1987 El Centro Imp. 6.5 18.20 0.357 the bridge. In this study, the ground motion used in the collapse test case
Hills, USA Co. Cent.
(test case E15) is considered as a future earthquake, while the ground
EQ6 Superstition 1987 Westmorland 6.5 13.03 0.173
Hills, USA Fire Sta. motions used in other cases are considered as the historical earthquakes
EQ7 Superstition 1987 Wildlife 6.5 23.85 0.179 of the bridge. The test data collected from the other test cases are used to
Hills, USA Liquefaction calibrate the digital twins for assessing the seismic performance of the
Array bridge under the future earthquake. The collapse fragility assessment
EQ8 Loma Prieta, 1989 Foster City - 6.9 45.58 0.110
USA Menhaden Court
results of the three different models are compared in terms of collapse
EQ9 Loma Prieta, 1989 Hollister City 6.9 27.60 0.246 damage locations, collapse ground motion intensities, and collapse
USA Hall probabilities. The results are also compared with the experimental ob­
EQ10 Kobe, Japan 1995 Port Island (0 m) 6.9 3.31 0.348 servations on the physical model of the bridge from the test case E15 to
EQ11 Kocaeli, Turkey 1999 Ambarli 7.5 69.62 0.253
illustrate the feasibility and accuracy of the digital twin-based collapse
EQ12 Chi-Chi, Taiwan 1999 CHY047 7.6 24.13 0.181
EQ13 Chi-Chi, Taiwan 1999 CHY054 7.6 48.49 0.095 fragility assessment method. Moreover, the suitability and accuracy of
EQ14 Chi-Chi, Taiwan 1999 CHY107 7.6 50.61 0.094 different IMs (PGA, PGV, Sa(T1) and Sa(T2)) for estimating the seismic
EQ15 Chi-Chi, Taiwan 1999 ILA013 7.6 84.12 0.149 performance of the bridge are also investigated.
EQ16 Tottori, Japan 2000 SMN002 6.6 16.61 0.154
EQ17 Niigata, Japan 2004 NIG014 6.6 28.42 0.120
EQ18 Chuetsu-oki, 2007 NIG013 6.8 29.80 0.142
4.1. Collapse damage locations
Japan
EQ19 Chuetsu-oki, 2007 NIG014 6.8 27.09 0.111 From the results of the collapse fragility analyses, three typical
Japan damage locations of the bridge at collapse are observed as shown in
EQ20 Chuetsu-oki, 2007 NIG018 6.8 10.78 0.681
Figs. 10 to 12. The three damage locations refer to (1) bottom of the
Japan
EQ21 Iwate, Japan 2008 MYG006 6.9 30.39 0.243 tower column, (2) beam-column conjunction, and (3) top of the middle
EQ22 El Mayor- 2010 Westmorland 7.2 42.61 0.157 tower column. The damage locations of the three FE models at collapse
Cucapah, Fire Sta. under the 30 selected ground motions are also summarized in Table 5.
Mexico Note that in the shake table tests, strong transverse constraints, which
EQ23 El Mayor- 2010 El Centro Array 7.2 41.29 0.183
Cucapah, #3
remain elastic during the test [25], are used to provide fixed girder-pier
Mexico and girder-tower connections. In the FE models, these connections are
EQ24 Darfield, New 2010 Christchurch 7.0 18.05 0.155 modelled by elastic spring elements and hence the girder-pier and
Zealand Botanical girder-tower connection failure is not considered in this study.
Gardens
Fig. 10 shows two types of bottom column failure-initiated collapse
EQ25 Darfield, New 2010 Christchurch 7.0 19.48 0.263
Zealand Resthaven mode observed from the fragility analyses of the FE models. Both modes
EQ26 Christchurch, 2011 Christchurch 6.2 5.55 0.553 are initiated from the compression-induced elemental failure at the
New Zealand Botanical bottom of the main tower. In the failure mode 1 (Fig. 10a), the damage is
Gardens firstly spotted at the two ends of the bottom tower columns. Then, the
EQ27 Christchurch, 2011 Christchurch 6.2 3.26 0.384
New Zealand Cathedral
elements at the bottom of the main tower are deactivated due to
College excessive compression. Subsequently, the elemental failure is also found
EQ28 Christchurch, 2011 Christchurch 6.2 4.85 0.346 at the top end of the bottom tower column, leading to the side-sway
New Zealand Hospital collapse of the main tower. In the failure mode 2 (Fig. 10b), the
EQ29 Christchurch, 2011 Christchurch 6.2 5.13 0.371
seismic damage concentrates at the bottom of the main tower. The main
New Zealand Resthaven
EQ30 El Mayor- 2010 El Centro Array 7.2 11.26 0.331 tower tilts and then collapses because of the compression failure at the
Cucapah, #12 column bottom.
Mexico Fig. 11 shows the collapse mode due to the failure at the beam-
a
Moment magnitude. column conjunction. The failure of the beam-column conjunction
b
The closest distance to rupture plane. leads to the inclination of the tower columns above the beam and
eventually leads to the collapse of the bridge. Fig. 12 shows the collapse
concluded that Model-L has a higher seismic collapse resistance than mode due to the failure at the top of the middle tower column. The
Model-NL. The degradation of seismic performance mainly results from compression failure is first spotted at the top of the middle tower column
the seismic damage accumulation during the test cases between E6 and below the anchorage area. Then, the upper anchorage part falls off from
E13. However, the comparative results between Model-O and Model-L the main tower and leads to the collapse of the bridge.
vary significantly when different IMs are used. For the fragility curves From the shake table test case E15, the collapse of the bridge is
depicted by PGA and PGV, the order of seismic collapse resistance is observed to occur at the beam-column conjunction of the north tower as
Model-L>Model-O>Model-NL. For the results described by Sa(T1), the shown in Fig. 6 [18,25]. However, for the design document-based
order becomes Model-L≈Model-O>Model-NL. However, when Sa(T2) is Model-O, the collapse initiates from the beam-column conjunction of
used, the order changes to Model-O>Model-L>Model-NL. These results the main tower only 4 times out of 30 selected ground motions, in which
highlight the necessity of developing a high-fidelity digital twin and only 3 IDAs exhibit the same collapse mode as the experimental obser­
vation from the test case E15. For Model-L, which is the linearly updated

8
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 8. Response spectra of the selected ground motions (damping ratio: 5%).

Fig. 9. Collapse fragility curves of different FE models.

model, 10 times out of 30 selected ground motion exhibit the beam- 4.2. Collapse ground motion intensities
column conjunction failure-initiated collapse, in which 6 IDAs predict
the same collapse region as the test one. These results indicate that the The collapse ground motion intensities are determined in this study
conventional design document-based model has a large uncertainty using the two approaches: (1) use the fragility analysis results with the
when assessing structural collapse performance. Moreover, the linearly 30 selected ground motions; and (2) use the ground motion input in the
updated Model-L can improve the damage location prediction to a test case E15 and carry out the IDA analyses of the three FE models.
certain extent but the results still deviate from the experimental obser­ From the regression results of the fragility analyses, the regressed
vation due to the following reasons: (1) the test data used for updating logarithmic means and standard deviations (μ and σ) of the lognormal
(from the test case E6) are not the data recorded just before the collapse distributions of the fragility curves are obtained and summarized in
of the bridge at the test case E15; and (2) only linear parameters are Table 6. It is noted that μ is also the natural logarithm of the median
updated in Model-L. In Model-NL-based collapse fragility assessment, all value of the critical collapse ground motion intensity (IM50%). IM50% can
the IDAs exhibit the beam-column conjunction failure-initiated collapse, be used as an index for describing the structural seismic collapse resis­
in which 18 simulations predict the same collapse mode as the test tance [49]. Moreover, σ represents the degree of correlation between the
collapse mode. It is hence concluded that the nonlinearly updated digital IM and the structural collapse probability. An IM is considered more
twin, which is calibrated by the latest measurement data collected just efficient when the associated σ is smaller [10].
before the test case E15, can locate the most vulnerable area and provide It can be seen from Table 6 that Sa(T2) (the spectral acceleration
a more accurate seismic performance assessment for the cable-stayed corresponding to the first transverse vibration mode of the main tower)
bridge. is the most suitable IM for assessing the collapse resistance of the bridge
because its logarithmic standard deviations are the smallest among the 4
IMs. The primary reason behind it is that all the observed collapse modes

9
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 10. Bottom column failure-initiated collapse of the bridge.

Fig. 11. Beam-column conjunction failure-initiated collapse of the bridge (Model-NL, EQ19: Chuetsu-oki earthquake, PGA=1.00 g).

initiate from the main towers (see Table 5). It is noted that for long-span corresponding to other spectral accelerations are larger than those
cable-stayed bridges, higher-order vibration modes of the main tower corresponding to Sa(T2) especially when Model-NL is evaluated. It is
participate in the seismic responses and may affect the structural failure believed that the existing seismic damage affects the failure mode of the
modes. Hence, the spectral accelerations of other higher-order vibration tower, and the observed column conjunction failure-initiated collapse of
modes are also examined. The results indicate that the regressed σ values Model-NL during the fragility analyses is majorly influenced by the first-

10
K. Lin et al. Automation in Construction 123 (2021) 103547

Fig. 12. Top of middle tower column failure-initiated collapse of the bridge (Model-O, EQ6: Superstition Hills earthquake, PGA=2.00 g).

Sa(T2) is considered to be the most appropriate IM for the collapse

Table 5
fragility analyses of the physical model of the bridge in this study.
Summary of the damage locations at collapse found by the collapse fragility
Furthermore, the comparison of IM50% with the test result indicates that
analyses.
Model-NL provides the closest estimation of structural seismic collapse
Model Collapse damage locations Counts
resistance. The error between the estimated collapse ground motion
Model-O Top of middle tower column North tower 6 10 intensity and the actual intensity of the shake table test is less than 10%,
South tower 4 demonstrating the feasibility and accuracy of the implementation of
Beam-column conjunction North tower 3 4
digital twin technology in the collapse fragility assessment of a long-
South tower 1
Bottom of tower column (Mode 1) North tower 1 16 span cable-stayed bridge.
Bottom of tower column (Mode 2) North tower 7 By considering the shake table tests, the PGA of the input ground
South tower 8 motion is increased with an interval of 0.2 g (see Table 1). Therefore, it
Model-L Top of middle tower column North tower 4 5
can only be guaranteed that the collapse ground motion intensity of the
South tower 1
Beam-column conjunction North tower 6 10
bridge under the artificial site ground motion is between 1.1 g (test case
South tower 4 E13) and 1.3 g (test case E15). The IDA analyses of the three FE models
Bottom of tower column (Mode 1) North tower 4 15 of the bridge under the artificial site ground motion are conducted to
Bottom of tower column (Mode 2) North tower 10 find the critical collapse ground motion intensities of the three FE
South tower 1
models. The acceleration time history of the artificial site ground motion
Model-NL Beam-column conjunction North tower 18 30
South tower 12 (see Fig. 2) is taken as an input and gradually increased up to the
collapse of the FE model with an interval of 0.05 g. The collapse in­
tensities obtained from the IDA analyses for the three FE models are also
order transverse vibration mode of the main tower. In addition, PGA summarized in Table 6.
performs slightly better than PGV while Sa(T1) is considered not suitable It can be seen from Table 6 that Model-NL predicts a collapse in­
for assessing the seismic collapse performance of the bridge. Therefore, tensity value within the range of 1.1 g to 1.3 g. Although the collapse

Table 6
Comparison of the collapse ground motion intensities.
Model μ σ IM50% Test resulta Error(IM50%)b IDA result Error(IDA)c

PGA based result (IM unit: g)

Model-O 0.576 0.346 1.780 1.281 38.93% 1.350 31.83%
Model-L 0.748 0.276 2.114 65.00% 1.700 24.33%
Model-NL 0.267 0.351 1.305 1.91% 1.150 13.52%

PGV based result (IM unit: m/s)

Model-O − 0.766 0.334 0.465 0.312 49.06% 0.329 41.50%
Model-L − 0.613 0.298 0.542 73.64% 0.414 30.89%
Model-NL − 1.082 0.328 0.339 8.61% 0.280 21.03%

Sa(T1) based result (IM unit: g)

Model-O − 2.009 0.722 0.134 0.133 1.24% 0.140 − 3.95%
Model-L − 2.038 0.602 0.130 0.126 3.64% 0.167 − 21.92%
Model-NL − 2.564 0.627 0.077 0.122 − 36.96% 0.110 − 29.79%

Sa(T2) based result (IM unit: g)

Model-O 0.594 0.270 1.811 1.526 18.71% 1.608 12.62%
Model-L 0.268 0.246 1.307 0.758 72.45% 1.006 29.92%
Model-NL − 0.329 0.178 0.720 0.746 − 3.47% 0.670 7.50%
a
The actual input PGA of ground motion in the test case E15 is 1.281 g;
b
Error(IM50%)=(IM50%-Test)/Test×100%;
c
Error(IDA)=(IM50%-IDA)/IDA×100%.

11
K. Lin et al. Automation in Construction 123 (2021) 103547

intensity of Model-O is the closest to the test result, it cannot give the Table 7
correct collapse mode as discussed below. Moreover, the average error Estimated collapse probabilities of the three FE models.
between the two approaches is the smallest when Sa(T2) is used as IM. IM Model-O Model-L Model-NL
Hence, Sa(T2) is the most proper IM among the 4 IMs. In terms of the
PGA 17.07% 3.47% 47.83%
damage locations at collapse, the collapse of Model-O initiates from the PGV 11.59% 3.19% 39.99%
bottom of the south tower column (failure mode 2 in Fig. 10b) while that Sa(T1) 49.32% 47.63% 76.91%
of Model-L is from the bottom of the north tower column (failure mode 1 Sa(T2) 26.25% 1.34% 57.89%
in Fig. 10a). Only the IDA results of Model-NL predict the same collapse
mode as the experimental observation during the shake table test.
Declaration of Competing Interest

4.3. Collapse probabilities None.

The collapse fragility curves can also be used to estimate the collapse Acknowledgements
probability of the bridge under the earthquake of a given intensity. As
shown in Fig. 9, a dotted line corresponding to the intensity of the input The authors wish to acknowledge the financial support from the
ground motion of the test case E15 is used to intersect with the fragility Research Grants Council of Hong Kong through a competitive GRF grant
curves. The intersecting points are the estimated collapse probabilities (Grant No. 15269516). The financial support from The Hong Kong
of the three FE models and are summarized in Table 7. The results Polytechnic University through a special grant (Grant No. 1-ZVN3) and
indicate that Model-NL predicts the highest collapse probability while the National Natural Science Foundation of China (No. 51908133) is
those of Model-L and Model-O are much smaller, which further high­ also appreciated. The authors are grateful for the Information Tech­
lights the necessity of using a nonlinearly updated digital twin in the nology Services of the Hong Kong Polytechnic University for providing
seismic collapse assessment of the long-span cable-stayed bridge. the servers. The authors would also like to acknowledge Prof. Hong Hao
from Curtin University, Australia, for his help in securing the first
research grant for this work. Any opinions and conclusions presented in
5. Conclusions
this paper are entirely those of the authors.
A digital twin-based collapse assessment method has been proposed
for long-span cable-stayed bridges under strong earthquakes. A scaled References
long-span cable-stayed bridge and its shake table test results have been
[1] D. Vamvatsikos, C.A. Cornell, Applied incremental dynamic analysis, Earthq.
used to verify the proposed method. The major findings from this study Spectra 20 (2) (2004) 523–553, https://doi.org/10.1193/1.1737737.
are summarized below. [2] F. Casciati, G.P. Cimellaro, M. Domaneschi, Seismic reliability of a cable-stayed
bridge retrofitted with hysteretic devices, Comput. Struct. 86 (2008) 1769–1781,
https://doi.org/10.1016/j.compstruc.2008.01.012.
(1) With the advance of FE modelling, structural health monitoring, [3] W.T. Barnawi, S.J. Dyke, Seismic fragility relationships of a cable-stayed bridge
and model updating techniques, the digital twin technology can equipped with response modification systems, J. Bridge Eng. ASCE 19 (8) (2014),
be used for fragility analysis-based seismic performance assess­ https://doi.org/10.1061/(ASCE)BE.1943-5592.0000468. A4013003.
[4] Y.T. Pang, X. Wu, G.Y. Shen, W.C. Yuan, Seismic fragility analysis of cable-stayed
ment of long-span cable-stayed bridges. bridges considering different sources of uncertainties, J. Bridge Eng. ASCE 19 (4)
(2) To accurately assess the seismic collapse of a long-span cable- (2014), 04013015, https://doi.org/10.1061/(ASCE)BE.1943-5592.0000565.
stayed bridge, the digital twin should be the nonlinearly updated [5] L.F. Li, S.C. Hu, L.H. Wang, Seismic fragility assessment of a multi-span cable-
stayed bridge with tall piers, Bull. Earthq. Eng. 15 (9) (2017) 3727–3745, https://
FE model using the latest response data recorded by the sensory doi.org/10.1007/s10518-017-0106-x.
system installed on the bridge. [6] J. Zhong, Y.T. Pang, J.S. Jeon, R. DesRoches, W.C. Yuan, Seismic fragility
(3) The results from this study show that the digital twin Model-NL assessment of long-span cable-stayed bridges in China, Adv. Struct. Eng. 19 (11)
(2016) 1797–1812, https://doi.org/10.1177/1369433216649380.
can provide the closest prediction of the collapse ground mo­ [7] J. Zhong, J.S. Jeon, W.C. Yuan, R. DesRoches, Impact of spatial variability
tion intensity and collapse probability of the bridge, compared parameters on seismic fragilities of a cable-stayed bridge subjected to differential
with the design document-based Model-O and the linearly support motions, J. Bridge Eng. ASCE 22 (6) (2017), 04017013, https://doi.org/
10.1061/(ASCE)BE.1943-5592.0001046.
updated Model-L.
[8] C. Li, H.N. Li, H. Hao, K.M. Bi, B.K. Chen, Seismic fragility analyses of sea-crossing
(4) By comparing the collapse modes of the three FE models obtained cable-stayed bridges subjected to multi-support ground motions on offshore sites,
from the fragility analyses, only Model-NL can provide a consis­ Eng. Struct. 165 (2018) 441–456, https://doi.org/10.1016/j.
tent and accurate prediction of the collapse mode of the bridge engstruct.2018.03.066.
[9] Y.Y. Cheng, J. Zhang, J.J. Wu, Fragility analysis of a self-anchored suspension
under earthquake excitations, and Model-O and Model-L cannot bridge based on structural health monitoring data, Adv. Civil Eng. (2019),
predict the damage location of the bridge properly. 7467920, https://doi.org/10.1155/2019/7467920.
(5) It is important to select a proper IM in the digital twin-based [10] B. Wei, Z.L. Hu, X.H. He, L.Z. Jiang, Evaluation of optimal ground motion intensity
measures and seismic fragility analysis of a multi-pylon cable-stayed bridge with
collapse fragility assessment. For the physical model of the super-high piers in mountainous areas, Soil Dyn. Earthq. Eng. 129 (2020) 105945,
Sutong bridge investigated in this study, all the observed collapse https://doi.org/10.1016/j.soildyn.2019.105945.
modes initiate from the main towers. Therefore, the spectral ac­ [11] Y.L. Xu, Y. Xia, Structural Health Monitoring of Long-Span Suspension Bridges,
CRC Press, London, U.K, 2012, https://doi.org/10.1201/b13182.
celerations corresponding to the first transverse vibration mode [12] J.X. Mao, H. Wang, D.M. Feng, T.Y. Tao, W.Z. Zheng, Investigation of dynamic
of the main tower is considered as the most suitable IM for properties of long-span cable-stayed bridges based on one-year monitoring data
evaluating the collapse fragility of the bridge. However, this is under normal operating condition, Struct. Control. Health Monit. 25 (5) (2018)
e2146, https://doi.org/10.1002/stc.2146.
not a general solution, and different IM may be required for [13] F. Benedettini, C. Gentile, Operational modal testing and FE model tuning of a
evaluating the collapse fragility of other structures. The selection cable-stayed bridge, Eng. Struct. 33 (6) (2011) 2063–2073, https://doi.org/
of IM is proved to be an essential step for the seismic fragility 10.1016/j.engstruct.2011.02.046.
[14] Y. Wang, Z.X. Li, C.M. Wang, H. Wang, Concurrent multi-scale modelling and
analyses of long-span bridges.
updating of long-span bridges using a multi-objective optimisation technique,
(6) The proposed digital twin-based collapse fragility analysis can Struct. Infrastruct. Eng. 9 (12) (2013) 1251–1266, https://doi.org/10.1080/
accurately quantify the seismic collapse resistance and identify 15732479.2012.683198.
the vulnerable components of a bridge, and accordingly provide [15] X. Xiao, Y.L. Xu, Q. Zhu, Multiscale modeling and model updating of a cable-stayed
bridge. II: Model updating using modal frequencies and influence lines, J. Bridge
valuable information for the lifetime maintenance and manage­ Eng. ASCE 20 (10) (2015), https://doi.org/10.1061/(ASCE)BE.1943-
ment of the bridge in engineering practice. 5592.0000723, 04014113.

12
K. Lin et al. Automation in Construction 123 (2021) 103547

[16] F.Y. Wang, Y.L. Xu, B. Sun, Q. Zhu, Updating multiscale model of a long-span responses, Eng. Struct. 183 (2019) 459–474, https://doi.org/10.1016/j.
cable-stayed bridge, J. Bridge Eng. ASCE 23 (3) (2018), https://doi.org/10.1061/ engstruct.2019.01.043.
(ASCE)BE.1943-5592.0001195, 04017148. [34] R. Brincker, P. Andersen, Understanding stochastic subspace identification, in:
[17] K.Q. Lin, Y.L. Xu, X.Z. Lu, Z.G. Guan, J.Z. Li, Cluster computing-aided model Proceedings of Modal Analysis Conference 2006: IMAC-XXIV: A Conference &
updating for a high-fidelity finite element model of a long-span cable-stayed Exposition on Structural Dynamics, Society for Experimental Mechanics, St. Louis,
bridge, Earthq. Eng. Struct. Dyn. 49 (9) (2020) 904–923, https://doi.org/10.1002/ U.S, 2006.
eqe.3270. [35] D.J. Yu, W.X. Ren, EMD-based stochastic subspace identification of structures from
[18] K.Q. Lin, Y.L. Xu, X.Z. Lu, Z.G. Guan, J.Z. Li, Collapse prognosis of a long-span operational vibration measurements, Eng. Struct. 27 (2005) 1741–1751, https://
cable-stayed bridge based on shake table test and nonlinear model updating, doi.org/10.1016/j.engstruct.2005.04.016.
Earthq. Eng. Struct. Dyn. (2020), https://doi.org/10.1002/eqe.3341 eqe.3341. [36] E. Reynders, R. Pintelon, G. De Roeck, Uncertainty bounds on modal parameters
[19] K.Q. Lin, Y.L. Xu, X.Z. Lu, Z.G. Guan, J.Z. Li, Time history analysis-based nonlinear obtained from stochastic subspace identification, Mech. Syst. Signal Process. 22
finite element model updating for a long-span cable-stayed bridge, Struct. Health (2008) 948–969, https://doi.org/10.1016/j.ymssp.2007.10.009.
Monit. (2020), https://doi.org/10.1177/1475921720963868. [37] V. Boonyapinyo, T. Janesupasaeree, Data-driven stochastic subspace identification
[20] M. Grieves, J. Vickers, Digital twin: mitigating unpredictable, undesirable of flutter derivatives of bridge decks, J. Wind Eng. Ind. Aerodyn. 98 (2010)
emergent behavior in complex systems, in: F.J. Kahlen, S. Flumerfelt, A. Alves 784–799, https://doi.org/10.1016/j.jweia.2010.07.003.
(Eds.), Transdisciplinary Perspectives on Complex Systems: New Findings and [38] B. Moaveni, E. Asgarieh, Deterministic-stochastic subspace identification method
Approaches, Springer International Publishing, Switzerland, 2017, pp. 85–113, for identification of nonlinear structures as time-varying linear systems, Mech.
https://doi.org/10.1007/978-3-319-38756-7_4. Syst. Signal Process. 31 (2012) 40–55, https://doi.org/10.1016/j.
[21] F. Tao, H. Zhang, A. Liu, A.Y.C. Nee, Digital twin in industry: State-of-the-art, IEEE ymssp.2012.03.004.
Trans. Ind. Inform. 15 (4) (2019) 2405–2415, https://doi.org/10.1109/ [39] C. Rainieri, G. Fabbrocino, Influence of model order and number of block rows on
TII.2018.2873186. accuracy and precision of modal parameter estimates in stochastic subspace
[22] C. Boje, A. Guerriero, S. Kubicki, Y. Rezgui, Towards a semantic construction identification, Int. J. Lifecycle Perform. Eng. 1 (4) (2014) 317–334, https://doi.
digital twin: Directions for future research, Autom. Constr. 114 (2020), 103179, org/10.1504/IJLCPE.2014.064099.
https://doi.org/10.1016/j.autcon.2020.103179. [40] V.A. Noghabi, K. Bargi, H. Heidary-Torkamani, Seismic fragility assessment of
[23] C.S. Shim, N.S. Dang, S. Lon, C.H. Jeon, Development of a bridge maintenance cable-stayed bridge using incremental dynamic analysis and uniform design
system for prestressed concrete bridges using 3D digital twin model, Struct. method, Int. J. Earthq. Eng. Hazard Mitig. 2 (3) (2014) 80–88.
Infrastruct. Eng. 15 (10) (2019) 1319–1332, https://doi.org/10.1080/ [41] W.P. Wu, L.F. Li, X.D. Shao, Seismic assessment of medium-span concrete cable-
15732479.2019.1620789. stayed bridges using the component and system fragility functions, J. Bridge Eng.
[24] R. Lu, I. Brilakis, Digital twinning of existing reinforced concrete bridges from ASCE 21 (6) (2016), https://doi.org/10.1061/(ASCE)BE.1943-5592.0000888,
labelled point clusters, Autom. Constr. (2019), 102837, https://doi.org/10.1016/j. 04016027.
autcon.2019.102837. [42] C.B. Haselton, A.B. Liel, G.G. Deierlein, B.S. Dean, J.H. Chou, Seismic collapse
[25] Z.G. Guan, J.Z. Li, W. Guo, H.Y. Qu, Design and validation of a shaking-table test safety of reinforced concrete buildings. I: Assessment of ductile moment frames,
on a long-span cable-stayed bridge with inverted-Y-shaped towers, Eng. Struct. 201 J. Struct. Eng. ASCE 137 (4) (2011) 481–491, https://doi.org/10.1061/(ASCE)
(2019), https://doi.org/10.1016/j.engstruct.2019.109823, 109823. ST.1943-541X.0000318.
[26] L.X. Zhou, X.W. Wang, A.J. Ye, Shake table test on transverse steel damper seismic [43] X. Lu, X.Z. Lu, H. Guan, L.P. Ye, Collapse simulation of reinforced concrete high-
system for long-span cable-stayed bridges, Eng. Struct. 179 (2019) 106–119, rise building induced by extreme earthquakes, Earthq. Eng. Struct. Dyn. 42 (5)
https://doi.org/10.1016/j.engstruct.2018.10.073. (2013) 705–723, https://doi.org/10.1002/eqe.2240.
[27] MSC. Marc, Volume A: Theory and User Information, MSC. Software Corporation, [44] X.W. Wang, B. Zhu, S.G. Cui, Research on collapse process of cable-stayed bridges
Newport Beach, CA, 2012. under strong seismic excitations, Shock. Vib. (2017), 7185281, https://doi.org/
[28] O. Hasançebi, T. Dumlupınar, Linear and nonlinear model updating of reinforced 10.1155/2017/7185281.
concrete T-beam bridges using artificial neural networks, Comput. Struct. 119 [45] J. Zhong, Z.L. Hu, W.C. Yuan, L. Chen, System-based probabilistic optimization of
(2013) 1–11, https://doi.org/10.1016/j.compstruc.2012.12.017. fluid viscous dampers equipped in cable-stayed bridges, Adv. Struct. Eng. 21 (12)
[29] B.K. Oh, M.S. Kim, Y. Kim, T. Cho, H.S. Park, Model updating technique based on (2018) 1815–1825, https://doi.org/10.1177/1369433218756429.
modal participation factors for beam structures, Comp. Aid. Civil Infrastruct. Eng. [46] Y.T. Pang, Seismic fragility assessment of an isolated multipylon cable-stayed
30 (2015) 733–747, https://doi.org/10.1111/mice.12139. bridge using shaking table tests, Adv. Civil Eng. (2017), 9514086, https://doi.org/
[30] Q. Mao, M. Mazzotti, J. DeVitis, J. Braley, C. Young, K. Sjoblom, E. Aktan, F. Moon, 10.1155/2017/9514086.
I. Bartoli, Structural condition assessment of a bridge pier: A case study using [47] S. Li, F.H. Dezfuli, J.Q. Wang, M.S. Alam, Seismic fragility of a long-span cable-
experimental modal analysis and finite element model updating, Struct. Control. stayed bridge isolated by SMA wire-based lead rubber bearings, in: Proceedings of
Health Monit. 26 (2019) e2273, https://doi.org/10.1002/stc.2273. Eleventh U.S. National Conference on Earthquake Engineering, Los Angeles,
[31] X.M. Yang, T.H. Yi, C.X. Qu, H.N. Li, H. Liu, Modal identification of high-speed California, 2018.
railway bridges through free-vibration detection, J. Eng. Mech. ASCE 146 (2020), [48] X. Lu, X.Z. Lu, H. Guan, L.P. Ye, Comparison and selection of ground motion
https://doi.org/10.1061/(ASCE)EM.1943-7889.0001847, 04020107. intensity measures for seismic design of super high-rise buildings, Adv. Struct. Eng.
[32] X.M. Yang, T.H. Yi, C.X. Qu, H.N. Li, H. Liu, Continuous tracking of bridge modal 16 (7) (2013) 1249–1262, https://doi.org/10.1260/1369-4332.16.7.1249.
parameters based on subspace correlations, Struct. Control. Health Monit. 27 [49] Federal Emergency Management Agency, Quantification of building seismic
(2020) e2615, https://doi.org/10.1002/stc.2615. performance factors, in: FEMA P695, Applied Technology Council, Washington, D.
[33] Y. Xin, H. Hao, J. Li, Z.C. Wang, H.P. Wan, W.X. Ren, Bayesian based nonlinear C., 2009.
model updating using instantaneous characteristics of structural dynamic

13