2018 - Wang Et Al

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

KSCE Journal of Civil Engineering (2018) 22(9):3443-3451 Structural Engineering

Copyright ⓒ2018 Korean Society of Civil Engineers


DOI 10.1007/s12205-018-0414-y pISSN 1226-7988, eISSN 1976-3808
www.springer.com/12205
TECHNICAL NOTE

Multivariate Probabilistic Seismic Demand Model for the


Bridge Multidimensional Fragility Analysis
Qi’ang Wang*, Ziyan Wu**, and Shukui Liu***
Received January 12, 2017/Accepted November 28, 2017/Published Online January 12, 2018

··································································································································································································································

Abstract
Seismic fragility analysis for bridges is an essential issue for risk assessment of transportation networks exposed to seismic hazards.
Considering multiple Performance Limit States (PLSs) and seismic demand parameters, the study proposes a multidimensional fragility
evaluation methodology for engineering structures, and the objective of the paper is to show that the uncertainty and dependence
between seismic demand parameters should be considered for fragility analysis. Thus, a new Probabilistic Seismic Demand Model
(PSDM) following multivariate logarithmic normal distribution is addressed. Taking PLS correlation into consideration, multidimensional
PLS formula is constructed to identify the structural failure domain. A RC bridge is studied to show the proposed theory. To consider
bridge column plastic deformation and bearing nonlinear characteristic, nonlinear dynamic analyses are carried out. The bridge
multidimensional fragility curves are derived and compared with fragility curves for an individual component. Results indicate that
uncertainty and dependence of demand parameters can be properly dealt with by the multivariate PSDM. The multidimensional
fragility is higher than fragility of any individual component, and the bridge as a system is more fragile. The ignorance of multiple
components contribution to the system will generate an overestimation for the whole structural performance, which is adverse to
engineering structural safety.
Keywords: multidimensional fragility evaluation, performance limit state, probabilistic seismic demand model, multivariate logarithmic
normal, nonlinear dynamic analysis
··································································································································································································································

1. Introduction earthquake response plans.


Fragility is defined as a conditional probability that a structure
Bridges are known to be one of the most vital and vulnerable will meet or exceed a certain damage level for a given earthquake
components of transportation network exposed to seismic intensity level. Fragility curves find their modern origin in the
hazards. Past experience, such as Northridge earthquake (1994), seismic probabilistic risk assessment of nuclear power plants
Hawaii earthquake (2006), Wenchuan earthquake in China (2008) performed in the early seventies of the twentieth century, and
and some others, showed that traffic flow can be severely they are developed either by empirical data or by analytical
disrupted because of the seismic damage in bridge, and they will methods. Empirical fragility curves (Charvet et al., 2014; Munoz
be harmful to the post-earthquake emergency response and et al., 2007; Yazgan, 2015) can be constructed based on test or
recovery activities. In order to ensure the safety of bridges and field data interpretation and engineering judgment. Analytical
minimize economic losses, vulnerability assessments of bridges fragility curves (Argyroudis and Kaynia, 2015; Bayat and
are generally carried out. The vulnerability is usually represented Daneshjoo, 2015; Ucar and Duzgun, 2013; Zentner, 2017) are
in terms of either damage probability matrices or fragility curves developed from seismic response analysis of structures based on
(Retamales, 2008; Elnashai and Di Sarno, 2015). The damage numerical simulation, including spectral analysis, nonlinear
probability matrices describe a discrete relationship between the static analysis and nonlinear time history analysis. There are
probability of damage occurrence and increasing ground motion different methods to generate structural fragility curves. Ghosh
intensity, while the fragility curves describes the same relationship in and Sood (2016) present a methodology to develop seismic
a continuous way, which is more widely used for seismic risk fragility curves for structures with deteriorating capacity limit
assessment, design verification, retrofit prioritization and post- states; Fragility curves of bridge components and system are

*Assistant Professor, State Key Laboratory for Geomechanics and Deep Underground Engineering; School of Mechanics and Civil Engineering, China
University of Mining and Technology, Xuzhou 221116, Jiangsu, China (Corresponding Author, E-mail: qawang@cumt.edu.cn)
**Professor, School of Mechanics, Civil Engineering and Architecture, Northwestern Polytechnical University, Xi’an 710129, China (E-mail:
zywu@mail.nwpu.edu.cn)
***Assistant Professor, School of Mechanics and Civil Engineering, China University of Mining and Technology, Xuzhou 221116, Jiangsu, China (E-mail:
skliu1986@cumt.edu.cn)

− 3443 −
Qi’ang Wang, Ziyan Wu, and Shukui Liu

assessed using nonlinear time history analysis. Considering (Mahmoudi and Chouinard, 2016; Zhang et al., 2016). The
geometrical parameters, Zampieri et al. (2016) presented a quick mathematical form is shown as following:
P = P(R ≥ rlim IM )
procedure to develop fragility curves for different masonry
f
(1)
bridge subclass. Randomness was taken into account to define
masonry compressive strength, material density and seismic where R is the response or demand parameter that can be either a
input. The generated fragility curves can deal with general arch deformation quantity like deck displacements, or a force quantity
bridge subclasses instead of a specific bridge. Dezfuli and Alam like bending moment or shear forces in bridge members or any
(2017) discussed the effect of different types of rubber bearings other measure of damage for which adequate capacity models
on the seismic fragility for highway bridges, and analytical exist. rlim is the damage state threshold parameter. IM represents
fragility curves for isolated bridges were developed with the earthquake intensity. Fragility curves demonstrate how the
consideration of the two major vulnerable components, i.e., pier variation of earthquake intensity affects the failure probability.
and isolation. Choi et al. (2004) evaluated fragility for the individual Fragility analysis involves uncertainties associated with structural
components of bridges, and then combine them into fragility capacity, damage limit state definition and records of ground
curves that represented the entire bridge system using first-order motion accelerations (Ding et al., 2012; Mahmoudi and Chouinard,
reliability principles. Pan et al.(2007) estimated the uncertainties 2016; Zhang et al., 2016).
of capacity and demand simultaneously by using the ratios of The traditional definition of fragility can be extended to
demands to capacities at different limit states to construct seismic multidimensional fragility as the following form (Casciati et al.,
fragility curves and developed fragility curves for the complete 2008):
bridge system using multiple failure modes theory. Borekci and
P = P{R1 ≥ rlim1 ∪ R1 ≥ rlim 2  ∪ R ≥ rlim IM } (2)
Kircil (2011) calculated discrete damage probabilities for R/C f n n

buildings using statistical methods for each considered performance where Ri is response parameter and rlim i is response threshold.
level, and constructed continuous fragility curves based on the The multidimensional fragility equation describes the aggregation
lognormal distribution assumption. Research has focused primarily of the conditional probabilities for multiple response parameters.
on assessing the fragility of individual components. However, Different engineering demand parameters on major component
relatively little effort has been dedicated to examine structural are selected and evaluated. Taking bridge column curvature
systems as a whole during fragility analysis. The structural ductility and bearing displacement as an example, the bi-
system fragility is usually represented by the fragility of the most dimensional formula is written in the following form:
Pf = P{ϕ ≥ ϕlim ∪ ζ ≥ ζ lim IM }
vulnerable component, which thus may fail to consider effect of
(3)
other component damage on the system vulnerability. Take the
bridge system as an example, the pier is usually viewed as the where ϕ is a random variable representing the column curvature
most vulnerable component and fragility curves of this type of ductility, ζ represents the bearing displacement response, ϕlim and
component are developed to represent the vulnerability of the ζlim are column curvature ductility threshold and displacement
whole bridge (Ding et al., 2012; Lu et al., 2005). The component threshold, respectively.
PLS correlation is seldom investigated. There is still a lack of
ideal PSDM which can deal with multiple seismic demand 2.2 PSDM Following Multivariate Logarithmic Normal Dis-
parameters of major components in a whole bridge system. tribution
In this study, a multidimensional fragility assessment methodology The Probabilistic Seismic Demand Model, firstly put forward
is proposed, and a new PSDM following multivariate logarithmic by Shome (Shome et al., 1998), defines a relationship between a
normal distribution is proposed to discuss the dependence and seismic hazard intensity measure and a structure-specific engineering
uncertainty of different demand parameter. Taking PLS correlation demand parameter, such as maximum column curvature. The
into account, a generalized multidimensional PLS formula is PSDM is the outcome from a Probabilistic Seismic Demand
constructed to identify structural failure domain. Monte Carlo Analysis (PSDA) (Lin et al., 2011). The PSDM most widely
(MC) simulation is used to assess the exceeding probability for used is a linear one in logarithmic terms (Cornell et al., 2002).
damage states. Consequently, multidimensional fragility curves The study carried out fragility analysis considering multiple
can be developed through nonlinear curve fitting using the engineering demand parameters for structural major components,
lognormal cumulative distribution function. which are dependent of each other. A new PSDM following
multivariate logarithmic normal distribution is proposed to
2. Multidimensional Fragility Assessment Meth- discuss those dependencies. The multidimensional fragility of
odology the structural system is calulated through integrating the PSDM
over structural failure domain. The failure domain is defined by
2.1 Multidimensional Fragility performance limit state.
The seismic fragility for a structure refers to the probability When multiple demand parameters Ri are investigated, the
that an engineering demand parameter exceeds a given limit state maximum responses form a multi-dimensional random vector,
threshold when subjected to an extensive range of ground motion written as R = [ R1, R2, ..., R ] . The vector follows multivariate
n

− 3444 − KSCE Journal of Civil Engineering


Multivariate Probabilistic Seismic Demand Model for the Bridge Multidimensional Fragility Analysis

logarithmic normal distribution. Thus, Y = [ lnR1, lnR2, …, InR ] n

follows multivariate normal distribution. The mean vector is v


and covariance matrix is denoted as E. The probability density
function of R can be obtained:
−n 2 −1 2 −1
f r ( r1 , r2 ,..., rn ) = ( 2π ) E ( r1 ⋅ r2 ⋅ ... ⋅ rn )
{ }
T (4)
exp − ( lnr − ν ) E −1 ( lnr − ν ) 2

where lnr represents [ lnr1, lnr2, …, lnr ] . The covariance matrix n

E indicates dependence between demand parameters. The


demand parameters are considered as independent, when the
covariance matrix is diagonal. If not, demand parameters are
dependent.
R(1), R(2), ..., R(m) represent the random sample of the response
vector. Y(1), Y(2), …, Yn represent the corresponding sample of
the vector Y. Y ′ , Sii and Sij denote maximum likelihood estimators
i

of νi (the ith element of mean vector v), eii and eij (the element of
Fig. 1. Performance Limit State Shapes
covariance matrix E), respectively. Through the probability
estimation method, mathematical forms for those parameters are
derived: PLSs, which controls multidimensional PLS shape. For a specific
structure, Ni is evaluated through nonlinear regression method
Y =
i
'

(∑ Y ) m
ij
(5)
based on field or experimental data. The structure is identified
with a damage state, when L ( R1, R2, …, R ) > 0 . Otherwise, the
∑ (Y ) n

S = −Y '
2

(6)
ii ij i
structure is in the safe state. The formula allows to consider PLS

∑ (Y )( dependence for the structure system, and the dependence is


S = ij ik
−Y Y −Y i
'

jk j
)n (7)
simulated by the constant parameter Ni.
For the case study bridge in the paper, two engineering parameter In the framework, PLS can be dependent or independent,
are considered. The two parameters are curvature ductility ϕ of random or deterministic. The shape of PLS can be linear or
the bent column and bridge bearing longitudinal displacement ζ. nonlinear. All these options can be formulated as particular cases
Thus, the bivariate logarithmic normal distribution is employed as of the main general one with a different choice of the parameters
the PSDM for the bridge. The mathematical formula of the involved. The bi-dimensional and three-dimensional limit state
probability density function is written in the following equation: shapes are given as the following figure.

f (ϕ ,ζ ) =
(
exp− α 2 − 2ραβ + β 2 2 1 − ρ 2 ) ( ) (8)
For the bi-dimensional situation, performance limit state formula
is given as:
2
2πϕζσϕσ ζ 1 − ρ
Nφ Nζ
α = [ ln( ( ϕ ) – μ ; β = [ln(ζ ) − μ (ζ ) ] σ (ζ ) and ρ
( ϕ ) ) ] ⁄ σ ln ( ϕ )
⎛ ϕLS ⎞ ⎛ ζ LS ⎞
⎜ ⎟ +⎜ ⎟ −1 = 0
ln ln ln
(10)
represents the correlation coefficient between ln(ϕ) and ln(ζ). ϕ
⎝ LSO ⎠ ⎝ ζ LSO ⎠
ϕLS and ζLS denote maximum column ductility and bearing
2.3 Multidimensional PLS Formula
displacement, respectively. ϕLSO and ζLSO represent limit state
It is vital to construct the PLS formula to identify the structrual
thresholds. To simplify the calculation, the following expression
failure domain during the fragility analysis. PLS formula is
is obtained:
usually defined through engineering structural design criteria. In
N
this study, a generalized multidimensional PLS formula, given as ϕ LS ⎛ ζ LS ⎞
+⎜ ⎟⎟ − 1 = 0 (11)
Eq. (9), is employed (Cimellaro and Reinhorn, 2010). The advantage ϕ LSO ⎜⎝ ζ LSO ⎠
is that PLSs are viewed as random and dependent random
parameters. where N describes the PLS interdependence. The dependence
N1 N2 N decreases with the increase of N. Bi-dimensional PLS shape is
⎛ R ⎞ ⎛ R ⎞ ⎛ R ⎞
n

L ( R1 , R2 ,..., Rn ) = ⎜ 1 ⎟ +⎜ 2 ⎟ + ... + ⎜ n ⎟ affected by N. The details are given in Fig. 2. PLS is a single line
⎝ rlim1 ⎠ ⎝ rlim 2 ⎠ ⎝ rlim n ⎠ when N becomes equal 1. When two PLSs are independent with
(9)
n
⎛ R ⎞
N i each other, N tends to infinite, and PLS are two lines parallel
= ∑ ⎜ i ⎟ −1 = 0 with coordinate axes. For other cases, PLS is a curve. It can be
i =1 ⎝ rlim i ⎠
derived through Fig. 2 that the area for structural failure domain
where Ri is the response parameter, and rlimi represents the becomes smaller with the increase of N, i.e., with the decrease of
response threshold. Ni is the correlation coefficient for different PLS dependence, so the fragility will also become smaller. In

Vol. 22, No. 9 / September 2018 − 3445 −


Qi’ang Wang, Ziyan Wu, and Shukui Liu

Fig. 2. Sensitivity of Bi-dimensional PLS for N

other words, if the PLS dependence is not considered ( N → ∞ ),


the fragility is smaller than the real value and the structural
seismic performance is overestimated.

3. Case Study
Fig. 3. Multidimensional Fragility Development
A multi-span continuous RC bridge is used as a case study to
illustrate the proposed methodology. Column curvature ductility
and bearing displacement are selected to evaluate multidimensional The specific process is shown in Fig. 3.
fragility. Firstly, a detailed three-dimensional nonlinear analytical
model, which accounts for the nonlinear behaviour of the column, 3.1 Bridge Characteristics and Analytical Model
beam, and abutments, is developed in the finite element platform Based on the typical design details, a multi-span continuous
based on the bridge information. The bridge model is used in RC bridge was selected for the case study. The overall geometry
conjunction with a suite of 125 ground motion records for is consistent for the example and is depicted in the Fig. 4. The
nonlinear dynamic analysis. Through maximum likelihood bridge is a three span structure with two 21 m end spans and one
estimation, the PSDM following bivariate lognormal distribution 27 m interior span, thus completing a total length of the bridge is
is estimated based on the maximum responses. Taking the 60 m. The superstructure of the bridge consists of a 11.4 m wide,
dependence between column curvature ductility and bearing 0.21 m thick, continuous cast-in-place concrete deck supported
displacement PLSs into account, bi-dimensional PLS formula is on 5 beams spaced at 2.44 m. The beams are supported on RC
obtained. Finally, the probability of exceeding a certain damage three-column bents. The outermost spans of the bridge are
limit state can be calculated through Monte Carlo simulation. supported by seat-type abutments. The substructure consists of
Multidimensional fragility curves can be developed by nonlinear two bent caps, each supported by three 6.9 m tall circular
curve fitting using the lognormal cumulative distribution function. columns of 0.9 m diameter. The column reinforcement includes

Fig. 4. The Case Study Bridge

− 3446 − KSCE Journal of Civil Engineering


Multivariate Probabilistic Seismic Demand Model for the Bridge Multidimensional Fragility Analysis

Fig. 5. FEM for the Bridge and the Plastic Hinge Location

thirteen 25 mm longitudinal bars and spiral transverse reinforcement to excessive rotation of the plastic hinges. It has been observed
with a pitch of 0.75 m. The cap beam width is 1.2 m and has a that RC bridge columns subjected to lateral earthquake loading
depth of 1.0 m. Each column is supported by a square foundation usually experiences progressive reduction in strength and stiffness
with 2.4 m wide. The yield strengths for the main bar in column due to inelastic actions. The ductility that is derived from the
and cap beam are 335 Mpa and 210 Mpa, respectively. The rotation in the plastic hinge formed at the end of a column is
concrete compression strength is 35 Mpa. The pile cap is placed considered to be a good measure of the damage. Therefore, in
on nine precast concrete piles of 0.3 m diameter and a length of this study, one seismic demand parameter is defined in the form
10 m. The high type steel fixed bearings are used in the bridge of column ductility demand. According to the definition, the
and located beneath each beam. The expansion bearings are column curvature ductility is calculated as:
installed at the abutments.
ϕ =θ θy (12)
A three-dimensional Finite Element Model (FEM) of the
bridge using finite element platform SAP2000 is shown in Fig. 5. where θ in the equation represents the rotation angle for the
The deck is modelled with 4-node plane shell elements. Beams plastic hinge and θy represents the yield rotation angle, which can
are simulated by beam elements. Spring elements are employed be calculated by:
to model pile foundation. Non-integral abutments are simulated θ y = φ y Lp (13)
through beam element supported by the foundation springs. As is
known, plastic hinges are expected to occur at the ends of each where φy represents the curvature at the first yielding point, and
column. Caltrans Flexural Hinge is employed to model those Lp is the plastic hinge length. The equivalent plastic hinge length
plastic hinges, whose moment-curvature curves are bilinear. can be calculated based on integration of the curvature distribution.
There are there columns in each bridge bent. Each bent column The plastic hinge length Lp has been studied widely based on the
is modelled by using 3 beam elements and two flexural hinges static tests. In this study, Lp is calculated through the Eq. (14)
located at the two ends. Fig. 5 describes the location of plastic given by Priestley et al. (1992):
hinges. Beam elements are used to simulate abutment back-wall
L p = 0.08L + 0.022 f y d bl ≥ 0.044 f y d bl (14)
and wing-wall. The spring elements are placed under the lower
part of the bridge abutment to simulate the passive soil pressure where L represents the length from the contraflexture point to the
around the abutment. They are viewed as elastic. The stiffness of peak moment section, fy is the yield strength and dbl is the
vertical spring is infinite and the horizontal spring stiffness is longitudinal bar diameter. Therefore, the length of the plastic
calculated according to the ATC-32 regulation. hinge (Lp) in this paper was taken as 0.3908 m. Elastic-perfectly
plastic idealization for the moment-curvature relationship was
3.2 Bridge Demand Parameters and PLS Thresholds assumed. The column moment-curvature curve can be obtained
Bridge damages due to earthquakes have been observed to through SAP2000, being shown in Fig. 6 for which the
occur in various degrees from slight damage to collapse damage
state. There exist 5 most likely failure mechanisms, i.e., formation of
column plastic hinge; restrainer failure at expansion joints or
bridge abutments; pounding between bridge decks; column
premature shear failure and surrounding soil surrounding soil
liquefaction. For the case study RC bridge, two explicit performance
criteria are selected, which are column curvature ductility and
bearing longitudinal displacement.
Column failure is often the primary cause of bridge collapse.
Failure of a column can result in loss of vertical load carrying
capacity. In most cases, plastic hinges form at the column ends
for bridges. Bridge damage due to earthquakes can be attributed Fig. 6. Column Moment-curvature Relationship

Vol. 22, No. 9 / September 2018 − 3447 −


Qi’ang Wang, Ziyan Wu, and Shukui Liu

Table 1. Column Curvature Ductility Threshold dynamic analysis. Peak Ground Acceleration (PGA) is defined
Damage state Description Ductility demand as the maximum ground acceleration of the ground motion, and
No damage First yield 1.00 it is a critical intensity measure for the earthquake. In addition,
Slight Cracking, spalling 2.01 the design basis of earthquake ground motion is usually defined
Moderate Loss of anchorage 6.03 by PGA. Hence, PGA was chosen to characterize the seismic
Extensive Incipient column collapse 11.07 intensity level in the paper (Bojórquez et al., 2012). The suite
Complete Column collapse 23.65 consists of 20 real earthquake records and 5 synthetic ground
motions. The real earthquakes are selected from the Pacific
Table 2. Bridge Bearing Displacement Threshold Earthquake Engineering Research Center database. The synthetic
Damage Slight Moderate Extensive Collapse earthquake is generated through defining the earthquake intensity,
state damage damage damage damage epicentral distance and the geological conditions. All the earthquake
Displacement Sc γc Sc γc Sc γc Sc γc records are scaled to be 5 levels according to China earthquake
(mm) 28.6 0.6 104 0.55 136 0.59 187 0.65 intensity scale. 125 ground motions are obtained for nonlinear
time history analysis.

corresponding φy can be derived as 3.502 × 10−3, which in turn 3.4 Multidimensional Fragility Analysis for the Bridge System
gives θy as 1.3685 × 10−3 calculated through Eq. (13). According The maximum column ductility and bearing displacement can
to previous researches (Dutta and Mander, 2001; Kim and Feng, be calculated through nonlinear dynamic analysis. Through
2003), curvature ductility thresholds can be determined through maximum likelihood estimation, the logarithmic mean μ and
Table 1. standard deviation σ can be estimated (Table 3). Further, the
Bearings are another major vulnerable component for this kind correlation coefficient ρ is obtained (Table 4). Then, the probability
of bridge. They transfer forces between the bridge superstructure density of bivariate lognormal distribution PSDM is calculated.
(beam or deck) and the substructures (pier, abutment or foundation). The multidimensional fragility for the bridge sytem is estimated
Failure of these bearings in an earthquake can cause redistribution of through integrating the PSDM over failure domains. These
internal forces, which may overload either the superstructure or failure domains are defined by bi-dimensional PLS formula of
substructure, or both. Collapse is also possible when bearing the bridge. Monte Carlo simulation is performed as a numerical
support is lost. Therefore, the bearing longitudinal displacement method for solving the integral.
is selected as the other demand parameter, with a lognormal For instance, the PSDM probabilistic densities are shown in
distribution. The median Sc and dispersion γc are evaluated the Fig. 7(a) and (b) when PGA = 0.55 g and 0.75 g. A Monte
through Bayesian analyses based on experimental results and Carlo sampling method is employed to generate random
field survey data. The independent displacement thresholds ζLSO response data Ntotal = 106. The data located in the failure domain
in Eq. (11) are represented by the medians. area are counted Nfailure, shown in the Fig. 7(c), and then the
structural failure probability is estimated. The procedure is
3.3 Bi-dimensional PLS Formula and Ground Motion Suit repeated for all PGA levels to get probability of exceeding four
The PLSs for column ductility and bearing longitudinal damage states. Details are shown in the Table 5.
displacement are viewed to be dependent. In fact, the value of N In the end, multidimensional fragility curves (Fig. 8) with a
for a specific structure should be evaluated based on field data or lognormal cumulative distribution can be developed through
engineering judgment. The N in the paper is assumed to be 2. For nonlinear fitting method. Through the curves, the bridge failure
a specific structure, the parameter N should be estimated through probability can be read under different PGA levels for 4 damage
nonlinear fitting based on structural damage history data or lab states. For instance, the fragility is 68% when PGA = 0.6 g.
test data. In addition, the sensitivity of fragility curves to N is
investigated in the Section 3.5. The bi-dimensional performance Table 3. Logarithmic Mean and Standard Deviation
limit state formula can be determined. Taking extensive damage PGA (g) μφ σφ μζ (mm) σζ
limit state as an example, the equation is listed as: 0.15 0.8332 0.2761 3.0098 0.3303
2 0.35 1.7830 0.5166 3.8544 0.3476
ϕE ⎛ζ ⎞
+ ⎜ E ⎟ =1 (15) 0.55 2.1593 0.2980 4.3216 0.3386
11 .07 ⎝ 136 ⎠
0.75 2.4241 0.3093 4.5997 0.3593
To perform the fragility analysis, different levels of ground 0.95 2.6827 0.2791 4.8648 0.3343
motions have to be considered to evaluate the probability that the *μ and σ are respectively the logarithmic mean and logarithmic standard
bridge will exceed a certain damage state. Multiple ground deviation of response data
motions should be selected to capture the uncertainty arising
from earthquakes. These ground motion records must be Table 4. Correlation Coefficient between ln(φ) and ln(ζ)
representative of the seismic characteristics of the bridge site. In PGA (g) 0.15 0.35 0.55 0.75 0.95
the study, a suite of 25 ground motions is used for nonlinear ρ 0.7693 0.9147 0.9319 0.7894 0.7978

− 3448 − KSCE Journal of Civil Engineering


Multivariate Probabilistic Seismic Demand Model for the Bridge Multidimensional Fragility Analysis

Fig. 7. (a) PSDM with PGA = 0.55G, (b) PSDM with PGA = 0.75G, (c) Monte Carlo for Failure Probability Estimation

Table 5. Bridge Failure Probabilities


Probability of exceedance
PGA (g)
Slight Moderate Extensive Collapse
0.15 0.9245 0.0029 0 0
0.35 0.9947 0.5422 0.1473 0.0071
0.55 0.9995 0.9637 0.5813 0.0593
0.75 1.0000 0.9881 0.8277 0.2437
0.95 1.0000 0.9999 0.9891 0.5599

According to Chinese Seismic Intensity Scale information,


failure probability for 8 and 9 degree are marked in the figure. Fig. 8. Bridge Multidimensional Fragility
Although the failure probability is high at both 8 and 9 degree for
the slight damage, the bridge may be put back into normal
function with minor repair work. The extensive damage probability becomes a straight line. While for N = ∞ , the two PLSs are
reaches 70% for 9 degree, which may cause serious economic unrelated, and the shape of multidimensional performance limit
losses. The administrative department may need to do some state are represented by two lines parallel to coordinate axes. The
disaster prevention measures ahead of time. result of multidimensional fragility evaluation will be definitely
affected due to the change of multidimensional PLS shape. Thus,
3.5 Sensitivity of Multidimensional Fragility to Depen- the sensitivity of fragility curves to N is investigated.
dence Between Limit States The bridge multidimensional fragility for extensive and collapse
The correlation coefficient N describes PLSs dependence. As damage states are developed when N = 1 , N = 2 and N = 4 .
is shown before, the dependence reduces with the increase of The curves are plotted in Fig. 9. N describes the dependence
correlation coefficient. For different values of N, it is possible to between the column PLS and bearing PLS. It is concluded that
change the shape of the performance limit state (Fig. 2). If N = 1 multidimensional fragility fluctuates significantly for different N
when column ductility and longitudinal displacement PLSs are values. Multidimensional fragility decreases with the increase of
linearly dependent, the limit state for the whole bridge structure N. In other word, fragility becomes smaller when the PLS

Vol. 22, No. 9 / September 2018 − 3449 −


Qi’ang Wang, Ziyan Wu, and Shukui Liu

Fig. 9. Sensitivity of the Multidimensional Fragility

dependence reduces. When the two PLSs for the bridge system
are considered to be independent, i.e., N = ∞ , fragility assessment
results will be smaller than the real value. In this situation, the
structural seismic performance will be overestimated. The actual
value of N should be estimated using probabilistic analysis and
engineering judgment.

3.6 Comparison of Multidimensional Fragility and Compo-


nent Fragility
Multidimensional fragility is further compared with bridge
component fragility, and the column fragility is used as an
example to show the details. Column limit state thresholds are
shown in Table 1, and the maximum ductility responses are
calculated through nonlinear dynamic analysis. Thus, the column
fragility curves are developed. For all damage states, column
Fig. 10. Comparison of Multidimensional Fragility with the Column
fragility is compared with the multidimensional fragility for the
Fragility
whole bridge system (Fig. 10).
As shown in the figure, the multidimensional fragility is higher
than column fragility, and the situation is quite obvious for the approach.
extensive and collapse damage states. Similar scenarios can be Results show that the probability of failure becomes smaller
found for other bridge components. For the extensive damage when the dependence between component PLSs is ignored, and
state, multidimensional fragility is nearly the same with column the structural seismic performance may be overestimated. The
fragility when PGA is less than 0.25 g. As the PGA increases, the proposed multivariate logarithmic normal distribution PSDM allows
numerical difference between those two fragility probabilities considering the dependences among different seismic demand
becomes bigger. The probability of multidimensional fragility parameters to obtain the joint probability density function of
equals to 70%, when PGA = 0.6 g, for which the column fragility multiple responses. Multidimensional fragility for the bridge system
reaches 30%, setting a difference close to 40%. In this situation, is higher than any component fragility. Thus, the bridge system is
a non-conservative assessment results will occur by substituting more vulnerable, and multidimensional fragility may avoid
column fragility for the whole bridge fragility, and the seismic overestimation for the structural seismic performance. The
performance for the bridge system will be overestimated. multidimensional fragility analysis methodology is general, and can
be applied to other engineering structures under different natural
4. Conclusions hazards. In this paper, the randomness of structural responses is fully
considered. In the future study, the PLS randomness will be
Multidimensional fragility estimation methodology is presented. investigated. At the same time, for a certain structural system, the
In order to investigate dependence between demand parameters, PSDM will be extended to three dimensions to consider the
a new multivariate logarithmic normal PSDM is addressed. structural fragility analysis under more complicated circumstances.
Multidimensional PLS formula is constructed to identify
structural failure domain. Then, the multidimensional fragility Acknowledgements
can be calculated through multiple integral of the PSDM over the
failure domain. A RC bridge is used as an example to illustrate The study presented in the paper was conducted with financial

− 3450 − KSCE Journal of Civil Engineering


Multivariate Probabilistic Seismic Demand Model for the Bridge Multidimensional Fragility Analysis

support by the National Natural Science Foundation of China Ghosh, J. and Sood, P. (2016). “Consideration of time-evolving capacity
(NSFC) under Award Number 51708545, Basic Research project distributions and improved degradation models for seismic fragility
(LTKY201701), NSFC 51278420 and 51608520. Support and assessment of aging highway bridges.” Reliability Engineering &
System Safety, Vol. 154, pp. 197-218, DOI: 10.1016/j.ress.2016.06.001.
help from the Structural Engineering Laboratory at CUMT are
Kim, S. H. and Feng, M. Q. (2003). “Fragility analysis of bridges under
highly acknowledged. ground motion with spatial variation.” International Journal of Non-
Linear Mechanics, Vol. 38, No. 5, pp. 705-721, DOI: 10.1016/
References S0020-7462(01)00128-7.
Lin, L., Naumoski, N., Saatcioglu, M., and Foo, S. (2011). “Improved
Argyroudis, S. and Kaynia, A. M. (2015). “Analytical seismic fragility intensity measures for probabilistic seismic demand analysis. Part 2:
functions for highway and railway embankments and cuts.” Earthquake application of the improved intensity measures.” Canadian Journal Of
Engineering & Structural Dynamics, Vol. 44, No. 1, pp. 1863-1879, Civil Engineering, Vol. 38, No. 1, pp. 89-99, DOI: 10.1139/L10-111.
DOI: 10.1002/eqe.2563. Lu, Y., Gu, X., and Guan, J. (2005). “Probabilistic drift limits and
Bayat, M. and Daneshjoo, F. (2015). “Seismic performance of skewed performance evaluation of reinforced concrete columns.” Journal of
highway bridges using analytical fragility function methodology.” Structural Engineering, Vol. 131, No. 6, pp. 966-978, DOI: 10.1061/
Computers and Concrete, Vol. 16, No. 5, pp. 723-740, DOI: (ASCE)0733-9445(2005)131:6(966).
10.12989/cac.2015.16.5.723. Mahmoudi, S. N. and Chouinard, L. (2016). “Seismic fragility assessment
Bojórquez, E., Iervolino, I., Reyes-Salazar, A., and Ruiz, S. E. (2012). of highway bridges using support vector machines.” Bulletin Oof
“Comparing vector-valued intensity measures for fragility analysis Earthquake Engineering, Vol. 14, No. 6, pp. 1571-1587, DOI:
of steel frames in the case of narrow-band ground motions.” 10.1007/s10518-016-9894-7.
Engineering Structures, Vol. 45, No. 1, pp. 472-480, DOI: 10.1016/ Munoz, A., Blondet, M., Aguilar, R., and Astorga, M. A. (2007).
j.engstruct.2012.07.002. “Empirical fragility curves for Peruvian school buildings.” Wit
Borekci, M. and Kircil, M. S. (2011). “Fragility analysis of R/C frame Transactions on the Built Environment, Vol. 93, pp. 269-277, DOI:
buildings based on different types of hysteretic model.” Structural 10.2495/ERES070261.
Engineering and Mechanics, Vol. 39, No. 6, pp. 795-812, DOI: Pan, Y., Agrawal, A. K., and Ghosn, M. (2007). “Seismic fragility of
10.12989/sem.2011.39.6.795. continuous steel highway bridges in New York State.” Journal of
Casciati, F., Cimellaro, G., and Domaneschi, M. (2008). “Seismic Bridge Engineering, Vol. 12, No. 6, pp. 689-699, DOI: 10.1061/
reliability of a cable-stayed bridge retrofitted with hysteretic devices.” (ASCE)1084-0702(2007)12:6(689).
Computers & Structures, Vol 86, Nos. 17-18, pp. 1769-1781, DOI: Priestley, M. J. N., Seible, F., and Chai, Y. H. (1992). Design Guidelines
10.1016/j.compstruc.2008.01.012. for Assessment Retrofit and Repair of Bridges for Seismic Performance.
Charvet, I., Suppasri, A., and Imamura, F. (2014). “Empirical fragility San Diego: Dept. of Applied Mechanics & Engineering Sciences,
analysis of building damage caused by the 2011 Great East Japan University of California.
tsunami in Ishinomaki city using ordinal regression, and influence of Retamales, R. (2008). New experimental capabilities and loading
key geographical features.” Stochastic Environmental Research and protocols for seismic fragility and qualification of nonstructural
Risk Assessment, Vol 28, No. 7, pp. 1853-1867, DOI: 10.1007/ components, Doctor of Philosophy, The State University of New
s00477-014-0850-2. York at Buffalo, Buffalo.
Choi, E., DesRoches, R., and Nielson, B. (2004). “Seismic fragility of Shinozuka, M. and Banerjee, S. (2005). Damage Modeling of Reinforced
typical bridges in moderate seismic zones.” Engineering Structures, Concrete Bridges, Tri-Center Meeting on Transportation Networks.
Vol. 26, No. 2, pp. 187-199, DOI: 10.1016/j.engstruct.2003.09.006. Shome, N., Cornell, C. A., Bazzurro, P., and Carballo, J. E. (1998).
Cimellaro, G. P. and Reinhorn, A. M. (2010). “Multidimensional “Earthquakes records and nonlinear responses.” Earthquake Spectra,
performance limit state for hazard fragility functions.” Journal of Vol. 14, No. 3, pp. 476-500, DOI: 10.1193/1.1586011.
Engineering Mechanics, Vol. 137, No. 1, pp. 47-60, DOI: 137 Ucar, T. and Duzgun, M. (2013). “Derivation of analytical fragility
10.1061/(ASCE)EM.1943-7889.0000201. curves for RC buildings based on nonlinear pushover analysis.”
Cornell, C. A., Jalayer, F., Hamburger, R. O., and Foutch, D. A. (2002). Teknik Dergi, Vol. 24, No. 3, pp. 6421-6446.
“A probabilistic basis for 2000 SAC Federal Emergency Management Yazgan, U. (2015). “Empirical seismic fragility assessment with explicit
Agency steel moment frame guidelines.” Journal of Structural modeling of spatial ground motion variability.” Engineering Structures,
Engineering-ASCE, Vol. 128, No. 4, pp. 526-533, DOI: 10.1061/ Vol. 100, pp. 479-489, DOI: 10.1016/j.engstruct.2015.06.027.
(ASCE)0733-9445(2002)128:4(526). Zampieri, P., Zanini, M. A., and Faleschini, F. (2016). “Derivation of
Dezfuli, F. H. and Alam, M. S. (2017). “Effect of different steel-reinforced analytical seismic fragility functions for common masonry bridge
elastomeric isolators on the seismic fragility of a highway bridge.” types: Methodology and application to real cases.” Engineering
Structural Control & Health Monitoring, Vol. 24, No. 2, DOI: Failure Analysis, Vol. 68, pp. 275-291, DOI: 10.1016/j.engfailanal.
10.1002/stc.1866. 2016.05.031.
Ding, Y., Wu, M., Xu, L., and Li, Z. (2012). “Vulnerability curves-based Zentner, I. (2017). “A general framework for the estimation of analytical
seismic damage assessment of RC columns.” Engineering Mechanics, fragility functions based on multivariate probability distributions.”
Vol. 29, No. 1, pp. 81-86. Structural Safety, Vol. 64, pp. 54-61, DOI: 10.1016/j.strusafe.2016.
Dutta, A. and Mander, J. B. (2001). “Rapid and detailed seismic fragility 09.003.
analysis of highway bridges.” Technical Report at Multidisciplinary Zhang, Y., Fan, J., and Fan, W. (2016). “Seismic fragility analysis of
Center for Earthquake Engineering Research, New York, USA. concrete bridge piers reinforced by steel fibers.” Advances in
Elnashai, A. S. and Di Sarno, L. (2015). Fundamentals of earthquake Structural Engineering, Vol. 19, No. 5, pp. 837-848, DOI:
engineering, Wiley and Sons, UK. 10.1177/1369433216630440.

Vol. 22, No. 9 / September 2018 − 3451 −

You might also like