You are on page 1of 11

Engineering Structures 29 (2007) 1312–1322

www.elsevier.com/locate/engstruct

PDF interpolation technique for seismic fragility analysis of bridges


Jin-Hak Yi a,∗ , Sang-Hoon Kim b , Shigeru Kushiyama c
a Coastal Engineering Research Department, Korea Ocean Research and Development Institute (KORDI), Ansan, Gyeonggi 426-744, Republic of Korea
b Department of Civil and Environmental Engineering, University of California, Irvine, CA 92697, USA
c Department of Architectural Engineering, Hokkai Gakuen University, Sapporo, Japan

Received 19 January 2004; received in revised form 7 August 2006; accepted 7 August 2006
Available online 9 October 2006

Abstract

This study presents a probability density function (PDF) interpolation technique for the evaluation of seismic fragility curves as a function
of the return period. Seismic fragility curves have been developed in terms of various seismic intensities, such as the peak ground acceleration
(PGA), peak ground velocity and pseudovelocity spectrum. However, of these seismic intensity measures, the return period of design earthquakes
is more useful, as seismic hazard curves are generally represented with the return period of design earthquakes; seismic design codes also require
consideration of the return period of the design earthquake spectrum for a specific site. Therefore, this paper focuses on the evaluation of seismic
fragility curves as a function of the return period. Seismic fragility curves based on the return period are compared with those based on the PGA
as an example bridge. The seismic fragility curves developed in this study make more intuitive sense for the design, retrofit and performance
evaluation of bridges.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Probability density function; Seismic fragility analysis; Return period; Seismic damage

1. Introduction of the seismic performance of civil infrastructures, such as


bridges and buildings. In the early seismic fragility analysis, the
Performance-based seismic design methodologies have seismic fragility curves were usually represented as functions
been improved for the design of new bridges and the of the peak ground acceleration (PGA) [3]. Hirata et al. [4,
retrofit of existing ones. There are two generally accepted 5] tried to express the seismic fragility as a function of
recent performance-based design concepts: (1) the multilevel the peak ground velocity (PGV) as well as of the spectral
performance design concept introduced from SEAOC’s Vision pseudovelocity for base-isolated reactor structures. Dimova and
2000 projects [1] and BSSC’s NEHRP Guidelines for Seismic Hirata [6] investigated several seismic intensity parameters,
Rehabilitation of Buildings [2], and (2) the probabilistic including PGA, PGV and the pseudovelocity spectrum (PVS)
concept for treating seismic events as random due to the for a seismic fragility analysis of building structures. They
uncertainties of earthquake events. Based on the above- demonstrated that the mean value of the PVS provided the
mentioned recent trends, the seismic fragility analysis, which best estimate of response for a given seismic intensity in a
evaluates damage exceeding the probability of a given least-square sense. More recently, seismic fragility analyses
structure’s response to varying intensities of seismic excitation, were carried out for evaluation of the performance of network
is most popular for the evaluation of the seismic performance systems, including a transportation system [7–10]. Even though
of civil infrastructures. the PGA value has been most widely used as a measure of
Seismic fragility analysis was originally conducted to the magnitude of earthquakes in the seismic fragility analyses
evaluate the seismic safety of nuclear power plants, which has of bridges, the return period of a design earthquake, of the
recently been accepted as a reliable method for the evaluation seismic intensity measures, including PGA and PGV, can
also be very useful because seismic hazard curves or maps
∗ Corresponding author. Tel.: +82 31 400 7811; fax: +82 31 408 5823. are generally represented using the return periods of design
E-mail address: yijh@kordi.re.kr (J.-H. Yi). earthquakes; seismic design codes also require consideration of

c 2006 Elsevier Ltd. All rights reserved.


0141-0296/$ - see front matter
doi:10.1016/j.engstruct.2006.08.019
J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322 1313

the return period of an earthquake for a specific site. In this where, xi = 1 or 0, depending on whether the bridge sustains
study, seismic fragility curves are developed as a function of the jth damage state under ground motion, with PGA = ai ,
the return period by utilizing the probability density function and N is the number of sample input ground motions. Two
(PDF) interpolation technique. A similar approach can be found parameters c j and ζ j (or ζcommon for F jII ) in Eqs. (1) and
in Wen and Wu’s study [11]. They interpolated the median (2), respectively, are computed by maximizing the likelihood
values of the structural responses to the seismic events of many function, as follows:
samples for comparing their proposed method with the results
of the conventional method. This study can be considered as ∂ ln L Ij ∂ ln L Ij
an extended version of one part of their research. In addition, = = 0, j = 1, 2, . . . , Nstate (4a)
∂c j ∂ζ j
a conventional seismic fragility analysis, based on PGA index
and maximum likelihood estimation, is also briefly introduced. ∂ ln L IIj ∂ ln L IIj
= = 0, j = 1, 2, . . . , Nstate . (4b)
Finally, the results of the seismic fragility curves based on the ∂c j ∂ζcommon
return period are compared with those based on PGA using the
same example bridge.
3. PDF interpolation technique for seismic fragility analysis
2. Seismic fragility analysis

2.1. Revisit to seismic fragility analysis in terms of peak ground 3.1. Seismic hazard curve
acceleration
Seismic fragility curves can be evaluated by calculating
Shinozuka et al. expressed the seismic fragility curves using the probability of exceeding predetermined performance limit
two-parameter lognormal distribution functions, and estimated states for several return periods, if the PDF of the maximum
the two parameters (median and log-standard deviation) with response of a given structure is known under the ground
the aid of the maximum likelihood estimation method [12]. A
motions with an arbitrary return period. However, it is
brief description is introduced, as follows:
impractical to obtain the PDF of maximum structural responses
The fragility curve for the jth damage state, F jI (a), takes the
under an earthquake with an arbitrary return period. One
following form
solution can be achieved by interpolating the characteristics
ln(a/c j ) of the PDF of maximum structural responses at existing
 
F jI (a; c j , ζ j ) = Φ (1) probability levels (e.g. 50, 10 and 2% in 50 years), and a reliable
ζj
basis function is essential to interpolate the PDFs. For this
where a is a PGA value of the ground motion, Φ(·) the standard purpose, the mathematical form of a seismic hazard curve is
normal distribution function, and c j and ζ j the median and utilized as a basic function for interpolating the characteristics
log-standard deviation values of the fragility curve for jth of PDFs of maximum structural responses, because earthquake
damage state, respectively. Even though the fragility curves related parameters, such as seismic damages, are reasonably
can be expressed as Eq. (1), the fragility curves for different represented by a seismic hazard curve. In addition, the seismic
damage states can be observed to intersect due to the different
hazard curve can be obtained by estimating two parameters
log-standard deviations. In such a case, the damage exceeding
from a few sampling points at the available return periods. A
probability for more severely damage states are estimated as
similar idea for obtaining the structural responses with respect
being higher than for less severe damage. To avoid such an
to the return periods can be found in the study of Wen and
intersection, Kim and Shinozuka [13] proposed a common log-
Wu [11]. They interpolated the probability of exceedance of the
standard deviation, using all sample data for every damage
column drift ratio in a 50 year period, and obtained the column
states, for the maximum likelihood estimation. In this study, the
drift demand at various return periods.
following seismic fragility curves, with a common log-standard
deviation (noted as F jII (a)), are also investigated. In this study, the following two assumptions were considered
to establish the seismic fragility curves based on return period:
ln(a/c j )
 
(1) structural responses under the earthquake ground motions
F j (a; c j , ζcommon ) = Φ
II
,
ζcommon at the same exceedance probability level follow a lognormal
j = 1, 2, . . . , Nstate (2) distribution (see Eq. (6)); and (2) the characteristics of the PDF
of seismic responses follow the seismic hazard curve proposed
where ζcommon denotes the common log-standard deviation. The by Cornell [14] as:
likelihood function for the jth damage state, L Ij and L IIj , can
then be introduced as: P E 1 = P[θmax > θ] = H (θ ) = K 0 θ −K 1 (5)
N h
Y ixi h i1−xi where P E 1 is an annual exceedance probability, θmax the
L Ij = F jI (ai ) · 1 − F jI (ai ) for F jI (a) (3a)
maximum structural response, and K 0 and K 1 are the two
i=1
parameters representing the seismic hazard curve. Graphically,
N h ixi h i1−xi
Y K 0 and K 1 are given as the intercept and slope of a regression
L IIj = F jII (ai ) · 1 − F jII (ai ) for F jII (a) (3b)
line of the logarithmic plot of a seismic hazard curve.
i=1
1314 J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322

3.2. Seismic fragility curve


When the maximum structural responses are distributed
following a lognormal distribution function, the PDF of the
maximum structural responses, θmax , is expressed as:
"  #
1 ln θ − λ 2

1
pθmax (θ) = √ exp − (6)
2π ηθ 2 η Fig. 1. Elevation of sample bridge.

where λ and η are log-mean and log-standard deviation values,


respectively, for the PDF of θmax . In order to determine the
PDF at an arbitrary exceeding probability level, two seismic
hazard curves for median (θmedian ) and deviated values (defined
as θdeviated ≡ eλ−η ) have to be drawn. From Eq. (5), the above
two seismic hazard curves are expressed as follows:
θ
P E 1median = K o,median θ −K 1,median (7)
θ
P E 1deviated = K 0,deviated θ −K 1,deviated (8)
where K 0,median and K 1,median are two constants of a seismic
hazard curve for the median value (θmedian ), and K 0,deviated
and K 1,deviated are those for the deviated value (θdeviated ). The
median and deviated values can be obtained as follows: Fig. 2. Nonlinearity in bridge model.
θmedian = eλ (9)
superstructure consists of a longitudinally reinforced concrete
θdeviated = eλ−η . (10) deck slab, and is supported by two pairs of columns. Each
After the two constants (K 0 and K 1 ) have been obtained pair has three columns with a circular cross-section diameter
using linear regression of the logarithmic plot of Eqs. (7) and of 0.8 m.
(8), the log-mean value, λi and log-standard deviation, ηi at The bridge is modelled to experience nonlinear behaviour at
θ the columns as a form of plastic hinge. A column is modelled
an arbitrary return period, Ti (Ti = 1/P E 1,i , P E 1,imedian =
θ as an elastic zone, with a pair of plastic zones at each column
P E 1,ideviated = P E 1,i ), can be calculated as follows:
end, considering double bending behaviour. Each plastic zone
1

P E 1,i
 is then modeled to consist of a nonlinear rotational spring and a
λi = − log (11) rigid element, as depicted in Fig. 2. The plastic hinge formed
K 1,median K 0,median
  in the bridge column is assumed to have bilinear hysteretic
1 P E 1,i characteristics.
ηi = λi + log . (12)
K 1,deviated K 0,deviated The column ductility program, developed by the authors
The PDF for the maximum structural responses under an based on the work by Priestley et al. [15], is used to
earthquake with an annual exceedance probability, P E 1,i (or obtain the moment–curvature relationship of the plastic
return period Ti ), can be expressed by a lognormal distribution hinges for the columns. From the moment–curvature curve,
function, as follows: the moment–rotation relationship is calculated using the
"  # plastic hinge length (which is calculated from the sectional
1 ln θ − λi 2

1 analysis [15]). The critical parameter used to describe the
pθmax ,i (θ ) = √ exp − . (13)
2π ηi θ 2 ηi nonlinear structural response in this study is the rotational
ductility demand at the plastic hinges. The ductility demand is
Finally, the seismic fragility curves at an arbitrary return defined as θ/θ y , where θ is the rotation of a bridge column in
period, Ti , are evaluated by calculating the probability of its plastic hinge and θ y the corresponding rotation at the yield
exceeding the predefined performance limit state (θ j ’s), as point.
follows:
Z θj
4.2. Response analysis of bridge
F j (Ti ) = P[θmax > θ j | T = Ti ] = 1 − pθmax ,i (θ )dθ.
−∞
(14) The SAP2000/Nonlinear finite element computer code [16]
is utilized for extensive two-dimensional response analyses of
4. Example study the bridge for sixty (60) Los Angeles earthquake time histories.
The earthquake data were developed for the FEMA SAC
4.1. Description of a sample bridge (SEAOC-ATC-CUREe) project, as listed in Table 1 [17]. These
Fig. 1 shows a sample bridge for an example analysis. This acceleration time histories were derived from historical records
bridge has an overall length of 34.0 m, with three spans. The and physical simulations, with some linear adjustments, and
J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322 1315

Table 1
FEMA-SAC LA earthquakes

SAC name Record Earthquake magnitude Scale factor PGA (cm/s2 ) Probability of exceedance
LA01 Imperial Valley, 1940, El Centro 6.9 2.01 452.03
LA02 Imperial Valley, 1940, El Centro 6.9 2.01 662.88
LA03 Imperial Valley, 1979, Array #05 6.5 1.01 386.04
LA04 Imperial Valley, 1979, Array #05 6.5 1.01 478.65
LA05 Imperial Valley, 1979, Array #06 6.5 0.84 295.69
LA06 Imperial Valley, 1979, Array #06 6.5 0.84 230.08
LA07 Landers, 1992, Barstow 7.3 3.2 412.98
LA08 Landers, 1992, Barstow 7.3 3.2 417.49
LA09 Landers, 1992, Yermo 7.3 2.17 509.70
LA10 Landers, 1992, Yermo 7.3 2.17 353.35 10% in 50 years
LA11 Loma Prieta, 1989, Gilroy 7 1.79 652.49
LA12 Loma Prieta, 1989, Gilroy 7 1.79 950.93
LA13 Northridge, 1994, Newhall 6.7 1.03 664.93
LA14 Northridge, 1994, Newhall 6.7 1.03 644.49
LA15 Northridge, 1994, Rinaldi RS 6.7 0.79 523.30
LA16 Northridge, 1994, Rinaldi RS 6.7 0.79 568.58
LA17 Northridge, 1994, Sylmar 6.7 0.99 558.43
LA18 Northridge, 1994, Sylmar 6.7 0.99 801.44
LA19 North Palm Springs, 1986 6 2.97 999.43
LA20 North Palm Springs, 1986 6 2.97 967.61

LA21 1995 Kobe 6.9 1.15 1258.00


LA22 1995 Kobe 6.9 1.15 902.75
LA23 1989 Loma Prieta 7 0.82 409.95
LA24 1989 Loma Prieta 7 0.82 463.76
LA25 1994 Northridge 6.7 1.29 851.62
LA26 1994 Northridge 6.7 1.29 925.29
LA27 1994 Northridge 6.7 1.61 908.70
LA28 1994 Northridge 6.7 1.61 1304.10
LA29 1974 Tabas 7.4 1.08 793.45
LA30 1974 Tabas 7.4 1.08 972.58 2% in 50 years
LA31 Elysian Park (simulated) 7.1 1.43 1271.20
LA32 Elysian Park (simulated) 7.1 1.43 1163.50
LA33 Elysian Park (simulated) 7.1 0.97 767.26
LA34 Elysian Park (simulated) 7.1 0.97 667.59
LA35 Elysian Park (simulated) 7.1 1.1 973.16
LA36 Elysian Park (simulated) 7.1 1.1 1079.30
LA37 Palos Verdes (simulated) 7.1 0.9 697.84
LA38 Palos Verdes (simulated) 7.1 0.9 761.31
LA39 Palos Verdes (simulated) 7.1 0.88 490.58
LA40 Palos Verdes (simulated) 7.1 0.88 613.28

LA41 Coyote Lake, 1979 5.7 2.28 578.34


LA42 Coyote Lake, 1979 5.7 2.28 326.81
LA43 Imperial Valley, 1979 6.5 0.4 140.67
LA44 Imperial Valley, 1979 6.5 0.4 109.45
LA45 Kern, 1952 7.7 2.92 141.49
LA46 Kern, 1952 7.7 2.92 156.02
LA47 Landers, 1992 7.3 2.63 331.22
LA48 Landers, 1992 7.3 2.63 301.74
LA49 Morgan Hill, 1984 6.2 2.35 312.41
LA50 Morgan Hill, 1984 6.2 2.35 535.88 50% in 50 years
LA51 Parkfield, 1966, Cholame 5W 6.1 1.81 765.65
LA52 Parkfield, 1966, Cholame 5W 6.1 1.81 619.36
LA53 Parkfield, 1966, Cholame 8W 6.1 2.92 680.01
LA54 Parkfield, 1966, Cholame 8W 6.1 2.92 775.05
LA55 North Palm Springs, 1986 6 2.75 507.58
LA56 North Palm Springs, 1986 6 2.75 371.66
LA57 San Fernando, 1971 6.5 1.3 248.14
LA58 San Fernando, 1971 6.5 1.3 226.54
LA59 Whittier, 1987 6 3.62 753.70
LA60 Whittier, 1987 6 3.62 469.07
1316 J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322

(a) Example of 50 in 50 Set (LA56).

Fig. 4. Moment–rotation curve for plastic hinge of column 1.

Table 2
Description of damaged states

Damage state Description Drift limits

(b) Example of 10 in 50 Set (LA17). Almost no damage First yield 0.005


Slight damage Cracking and spalling 0.007
Moderate damage Loss of anchorage 0.015
Extensive damage Incipient column collapse 0.025
Complete damage Column collapse 0.050

Table 3
Peak rotational ductility demand of columns of a sample bridge

Damage state Column 1 Column 2

(c) Example of 2 in 50 Set (LA22). Almost no 1.0000 1.0000


Slight 1.3263 1.3263
Moderate 2.6315 2.6316
Fig. 3. Acceleration time histories generated for Los Angeles area.

consist of three groups (each consisting of 20 time histories) responses at the column bottom end of column 1 are plotted in
with probabilities of exceedance of 50%, 10% and 2% in Fig. 5.
50 years at the Los Angeles site. These sets of ground motions
are referred to as 50 in 50 Set, 10 in 50 Set, and 2 in 50 Set. 4.3. Damage states
It should be noted that these three sets should only be used
as a set, and not individually or as small subsets, in order While the seismic damage states can be classified in several
to correctly represent the probability levels specified, which ways, Dutta’s recommendation is now being widely used as an
means that the median spectral acceleration of the set might indication of reasonable damage states [18–20] as the guideline
reasonably match the target value at any particular period, and was conducted using real-scaled bridge column tests. A set
the spectral acceleration for an individual record might be quite of five different damage states was introduced, as shown in
different from the expected target spectral acceleration. Typical Table 2, which provides a description of the five damage states
ground acceleration time histories for each group are plotted and the corresponding drift limits for a tested column. In this
on the same scale in Fig. 3 for comparison of the intensities of study, the drift limit was transformed to the rotational ductility
the acceleration. Typical ground acceleration data are selected demand of the plastic hinge at the columns, as the structural
taking into consideration the average PGA values for each data analysis programme used in this study utilized the nonlinear
set. The average PGA values were measured as 0.426, 0.588 rotational spring element; therefore, it is easier to read the
and 0.881g for the 50 in 50 Set, 10 in 50 Set and 2 in 50 Set. rotational displacements than the drift [15]. A similar approach
The nonlinear response characteristics associated with the can be found in the other paper by Choi et al. [20], where
bridge, based on a moment–curvature curve analysis, are the curvature ductility demand was used instead of drift limits.
incorporated taking into consideration the axial loads and Table 3 shows the rotational ductility demand limits for the
confinement effects. The moment–rotation relationship is sample bridge used in this study. It is noticeable that two
obtained using the plastic hinge length (L p = 0.58 m), and is peak rotational ductility demands for two columns are virtually
modelled using a bilinear rotational spring element without any identical because the values were directly obtained from the
stiffness degradation; its parameters are established according sectional analyses of the same sections of the two columns.
to the equations in Priestley et al. [15]. The moment–rotation Small discrepancies occurred due to the difference in the axial
curves were obtained, as shown in Fig. 4, and typical nonlinear forces on the sections.
J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322 1317

(a) Example of 50 in 50 Set (under LA56).

(b) Example of 10 in 50 Set (under LA17).

(c) Example of 2 in 50 Set (under LA22).

Fig. 5. Responses at column 1.

The corresponding return periods were easily obtained, as


72, 475 and 2475 years, by taking the reciprocals of the annual
exceedance probabilities, as follows:
1
T = . (16)
P E1

Table 4 lists the two parameters for the lognormal


distributions (λ and η), and the median and deviated values
(eλ and eλ−η ) for the three prespecified return periods. Table 5
lists the constants for two seismic hazard curves obtained by
interpolating the characteristics (i.e. the median value θmedian
Fig. 6. Responses at column end of column 1.
and the deviated value θdeviated in Eqs. (7) and (8)) for the PDF
4.4. Seismic fragility curve by PDF interpolation technique of maximum structural responses for the three different return
periods, those being, 72, 475 and 2475 years. Fig. 7 shows
Fig. 6 shows the results of nonlinear time history analyses in the seismic hazard curves obtained for maximum structural
terms of the ductility factors for the sample bridge employing responses (rotational ductility factors) for columns 1 and 2,
the three data sets. where the circle and square marks depict the median and
The annual exceedance probability, P E 1 , corresponding to deviated values obtained at the sampling points, respectively.
50 for 50 Set, 10 for 50 Set and 2 for 50 Set were calculated to From Tables 4 and 5, the probabilistic characteristics are
be 0.0138, 0.0021, and 0.0004, respectively, using the following observed to be quite different for columns 1 and 2, even though
equation: the ductility demands and sectional properties are almost the
same, as shown in Table 3. This is due to the difference
P E 1 = 1 − (1 − P E 50 )1/50 . (15) in the column lengths. As shown in Fig. 1, the length of
1318 J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322

Table 4
Log-mean and log-standard deviation for three data sets

Return period Column 1 Column 2


λ η eλ eλ−η λ η eλ eλ−η
72 −0.2914 0.6616 0.7472 0.3856 −0.1893 0.6675 0.8275 0.4245
475 0.4123 0.5593 1.5103 0.8633 0.5312 0.5604 1.7010 0.9712
2475 1.2514 0.7019 3.4952 1.7324 1.3596 0.6924 3.8946 1.9488

Table 5
Constants for seismic hazard curves for columns 1 and 2
Column 1 Column 2
K 0,median K 1,median K 0,deviated K 1,deviated K 0,median K 1,median K 0,deviated K 1,deviated
0.0064 2.2771 0.0015 2.3481 0.0082 2.2712 0.0019 2.3140

(a) For column 1.

(b) For column 2.

Fig. 7. Seismic hazard curves for the example bridge.


J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322 1319

Table 6
Probability characteristics obtained from seismic hazard curves

Return period Column 1 Column 2


eλ eλ−η λ η eλ eλ−η λ η
100 0.8240 0.4427 −0.1936 0.6212 0.9180 0.4899 −0.0856 0.6280
1000 2.2650 1.1803 0.8176 0.6518 2.5300 1.3251 0.9282 0.6468
10000 6.2264 3.1467 1.8288 0.6824 6.9730 3.5842 1.9421 0.6655

Fig. 9. Seismic fragility curves utilized by PDF interpolation technique.

Fig. 8. Probability density function for earthquakes with 1000 year return
period. 5. Discussion

column 1 is shorter than that of column 2, which makes the Seismic fragility curves for the same sample bridge are
applied moments at the two sections different. This difference evaluated using the maximum likelihood estimation in terms
is transferred to the probabilistic characteristic values. of a PGA index. Fig. 10(a) and (b) show the analytical seismic
The PDF of the maximum structural responses (ductility fragility curves obtained using Eqs. (1) and (2), respectively.
factors) under earthquakes with an arbitrary return period can From Fig. 10(a), the fragility curves for almost no and
be obtained using the seismic hazard curves, as shown in slight damages are observed to intersect around 0.22g, and
Fig. 7, which makes it possible to evaluate the seismic fragility the damage exceeding probabilities for slight damage were
curves from the calculation, using Eq. (14), of the exceedance evaluated to be slightly larger than those for almost no damage;
probability of pre-defined damage states. Table 6 lists the in the range of 0–0.22g, due to the different log-standard
probability characteristics obtained from the seismic hazard deviation. In Fig. 10(b), the intersection of two fragility curves
curve with return periods of 100, 1000 and 10 000 years. have been removed by adopting the common log-standard
The median values of the structural responses under the deviation. However, it can also be observed that the overall
earthquakes with return periods of 100, 1000, and 10 000 years shapes are severely altered.
were calculated to be 0.8240, 2.2650 and 6.2264, respectively. The damage exceedance probabilities of the 50 in 50 set, 10
This implies, for example, that half of the maximum responses in 50 set and 2 in 50 set for the five damage states are compared
under the earthquakes with a 1,000 year return period might with those obtained by a previous approach. Since the two
have a ductility factor over 2.2650. Fig. 8 shows the PDF of the indices are different (i.e. return period and PGA indices), the
maximum structural response under earthquakes with a 1,000 equivalent measure should be considered in their comparison.
year return period and predefined damage state. In this respect, the equivalent damage exceedance probabilities
Fig. 9 shows the seismic fragility curves represented by the are calculated from the fragility curves indexed, using PGA
return periods. From the result shown in Fig. 9, the probability values, for several return periods. We recall that the damage
of exceeding the damage states for an earthquake with an exceedance probability function can be expressed as follows:
arbitrary return period can be obtained easily. The probability of
ln(a/c j )
 
exceeding a complete damage state, for example, is almost 40%
F j (a; c j , ζ j ) = Φ
I
(17a)
under earthquakes with a 10 000 year return period; whether ζj
the structure should be upgraded to resist the earthquakes for 
ln(a/c j )

a specific probability of a damage state can also be easily F jII (a; c j , ζ j ) = Φ . (17b)
determined. Furthermore, if the fragility curves are periodically ζcommon
updated, based on real measurements, the structures can be The first equivalent probability of exceeding the jth damage
more reliably maintained [21]. state ( F̃ jI,1 and F̃ jII,1 ) can be evaluated by reading the fragility
1320 J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322

(a) F jI (a). (b) F jII (a).

Fig. 10. Fragility curves for the same sample bridge evaluated by Shinozuka et al.
PN
value at the averaged PGA value ( i=1 ai /N ) for each data set those from the previous approaches, as well as the sample
using the following equation: means. In this Figure, F̃ jI,1 ( F̃ jII,1 ), F̃ jI,2 ( F̃ jII,2 ) and F̃ j3 are
referred to as ‘Previous 1’, ‘Previous 2’ and ‘Sample’. From
F̃ jI,1 (a1 , a2 , . . . , a N ; c j , ζ j )
these results, those from the present approach are closer to
   N    the sample means than those from the previous approaches.
1 P
 ln N ai cj  The damage probabilities are observed to be over estimated in
i=1
= Φ (18a) the case of the previous results for the 50 in 50 set, and the
 
ζj

  damage probabilities are over- and under-estimated in the case
of the 10 in 50 set. Also, no considerable discrepancies were
F̃ jII,1 (a1 , a2 , . . . , a N ; c j , ζcommon ) found between the results from the present study and those
   N    from the previous methods, especially for the case with the
1 P common log-standard deviation (Fig. 11(b)). The common log-
 ln N ai cj 
standard deviation derived for preventing intersection between
i=1
= Φ . (18b)
 
 ζcommon  the fragility curves with different damage states may have
contributed to this discrepancy. This shortcoming is shown
in Fig. 6, for the ductility factors of the sample bridge for
The second equivalent probability of exceeding the jth the 60 Los Angeles earthquakes, and Fig. 10 for the seismic
damage state ( F̃ jI,2 and F̃ jII,2 ) can also be obtained by averaging fragility curves. For the complete damage state, only one
the fragility values under each ground motion in a data set, as case is over the damage state limit, which can make the
follows: maximum likelihood estimation difficult and unreliable. From
N   ln(a /c )  Fig. 6, the damage probabilities are observed to be under-
1 i j
F̃ jI,2 (a1 , a2 , . . . , a N ; c j , ζ j ) =
X
Φ estimated in the case of almost no and slight damage states by
N ζ j
i=1 incorporating the common log-standard deviations, while the
 (19)
N ln(ai /c j ) damage probabilities are over estimated in the case of severe
 
II,2 1 X
F̃ j (a1 , a2 , . . . , a N ; c j , ζcommon ) = Φ . and complete damage states. This observation may explain
N i=1 ζcommon
why there are more discrepancies between the sample means
In Eqs. (18) and (19), a1 , a2 , . . . , a N are the PGA values for and the damage probabilities with the previous approaches.
a certain earthquake group having the same return period, for The proposed method can be concluded to give unbiased
example, 20 data in the 50 in 50 set. The sample means are information about the damage exceeding probabilities, whereas
used as the third equivalent damage exceeding probability, as the previous approaches can over- or under-estimate the damage
follows: exceeding probabilities.
N
1 X 6. Conclusions
F̃ j3 (a1 , a2 , . . . , a N ) = xi j (20)
N i=1
The probability density function (PDF) interpolation
where xi j = 1 if the jth damage state occurs in a bridge technique is proposed for the evaluation of seismic fragility
subjected to a PGA value a = ai , or xi j = 0 in all other cases. curves as a function of the return period. The seismic hazard
For example, if the structures experience almost no damage curves for maximum structural responses are determined from
for all individual earthquakes in the 10 in 50 set, the obtained the structural responses under FEMA SAC earthquake ground
equivalent damage exceedance probability will be 1.0. motions at a Los Angeles site.
Fig. 11 shows a comparison of the results of the estimated The following two assumptions were employed: (1)
damage exceeding probabilities from the present study with the maximum structural responses follow the lognormal
J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322 1321

(a) With individual logstandard deviation. (b) With common logstandard deviation.

Fig. 11. Comparisons between the results from present study and previous study.

distribution; and (2) the PDF can be interpolated from the established at the Korean Advanced Institute of Science and
form of seismic hazard curve proposed by Cornell. Under these Technology (KAIST) under sponsorship from the Ministry of
assumptions, the log-mean, λ, and log-standard deviation, η, Science and Technology (MOST) and the Korean Foundation
are easily obtained from two seismic hazard curves. Using of Science and Technology (KOSEF).
the seismic hazard curves, the seismic fragility curves, as a
function of return period, are then evaluated from the numerical References
integration of PDF at each exceeding probability level. Seismic
fragility curves have been successfully constructed for a sample
[1] SEAOC. Vision 2000 — A frame work for performance based design,
bridge, utilizing the proposed method, where the damage volumes I, II, III. Technical report. Sacramento (California): Structural
exceeding probability for a certain return period can be very Engineers Association of California, Vision 2000 Committee; 1995.
easily read. The results obtained using the present method [2] BSSC. NEHRP Guidelines for the seismic regulation of existing buildings
are compared with those obtained by the conventional method and other structures. Technical report. USA: FEMA 273. Building Seismic
represented by PGA, it was found that the results indexed on Safety Council; 1997.
PGA and the maximum likelihood estimation generally over- [3] Kennedy RP, Ravindra MK. Seismic fragilities for nuclear power plant
risk studies. Nuclear Engineering and Design 1984;79:47–68.
or under-estimated the seismic fragilities.
[4] Hirata K, Kobayashi Y, Kameda H, Shiojiri H. Fragility of seismically
isolated FBR structure. Nuclear Engineering and Design 1991;128:
Acknowledgements 227–36.
[5] Hirata K, Ootori Y, Somaki T. Seismic fragility analysis for base-isolated
This study was carried out with financial support from the structure. Journal of Structural and Constructive Engineering 1993;452:
Smart Infra-Structure Technology Centre (SISTeC), which was 11–9. Architectural Institute of Japan.
1322 J.-H. Yi et al. / Engineering Structures 29 (2007) 1312–1322

[6] Dimova SL, Hirata K. Simplified seismic fragility analysis of structures retrofitted by column jacketing. Probabilistic Engineering Mechanics
with two types of friction devices. Earthquake Engineering and Structural 2004;19(1–2):105–12.
Dynamics 2000;29:1153–75. [14] Cornell CA. Calculating building seismic performance reliability: a basis
[7] Shinozuka M, Feng MQ. Nonlinear static procedure for fragility curve for multi-level design norms. In: Proceedings of the 11th world conference
development. Journal of Engineering Mechanics, ASCE 1995;126(12): on earthquake engineering. 1996.
1287–95. [15] Priestley MJN, Seible F, Calbi GM. Seismic design and retrofit of bridges.
[8] Dong X, Shinozuka M. Performance analysis and visualization of electric John Wiley & Sons, Inc. p. 270–73.
power systems. In: Proceedings of the SPIE’s 10 international symposium [16] Computer and structures, Inc. SAP2000/Non-linear users manual version
on smart systems and NDE for infrastructures. 2003. 8. Berkeley (CA, USA); 2002.
[9] Zhou Y, Murachi Y, Kim SH, Shinozuka M. Seismic risk assessment [17] FEMA-SAC. Develop suites of time histories project task: 5.4.1. Draft
of retrofitted transportation systems. In: Proceeding of the 13th world report of joint venture steel project phase 2. 1997.
conference on earthquake engineering. 2004. Paper no. 1128. [18] Dutta A, Mander JB. Seismic fragility analysis of highway bridges.
[10] Shinozuka M, Zhou YW, Banerjee S. Recent studies in fragility In: Proceedings of the center-to-center project workshop on earthquake
information on highway bridges. In: Proceedings of the 3rd US-PRC engineering in transportation systems. 1999.
workshop on seismic behavior and design of special highway bridges. [19] Dutta A. On energy based seismic analysis and design of highway bridges.
2004. Ph.D. dissertation. Buffalo (NY): Department of Civil, Structural and
[11] Wen YK, Wu CL. Uniform hazard ground motions for Mid-America Environmental Engineering, State University of New York; 1999.
cities. Earthquake Spectra 2001;17(2):359–84. [20] Choi ES, DesRoches R, Nielson B. Seismic fragility of typical bridges in
[12] Shinozuka M, Feng MQ, Kim H, Uzawa T, Ueda T. Statistical analysis of moderate seismic zones. Engineering Structures 2004;26(2):187–99.
fragility curves. Technical report. Buffalo (NY): MCEER, State Univ. of [21] Yi JH, Kim SH. Probabilistic seismic safety assessment of bridges
New York; 2000. utilizing measured response data. Journal of Korea Society of Civil
[13] Kim SH, Shinozuka M. Development of fragility curves of bridges Engineers 2004;24(5A):1093–102.

You might also like