Development of seismic fragility surfaces for reinforced

concrete buildings by means of nonlinear time-history

analysis
Darius Seyedi, Pierre Gehl, John Douglas, L. Davenne, N. Mezher, S.
Ghavamian

To cite this version:

Darius Seyedi, Pierre Gehl, John Douglas, L. Davenne, N. Mezher, et al.. Development of seis-
mic fragility surfaces for reinforced concrete buildings by means of nonlinear time-history analysis.
Earthquake Engineering and Structural Dynamics, Wiley, 2010, 39, p. 91-108. �10.1002/eqe.939�.
�hal-00558781�

HAL Id: hal-00558781

https://hal-brgm.archives-ouvertes.fr/hal-00558781
Submitted on 2 Feb 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Article accepted in Earthquake Engineering & Structural Dynamics
Final version, February 2009

Development of seismic fragility surfaces for reinforced concrete

buildings by means of nonlinear time-history analysis
D.M. Seyedi1∗, P. Gehl1, J. Douglas1, L. Davenne2, N. Mezher2, S. Ghavamian2
1
BRGM – Development Planning and Natural Risks Division, 3 avenue Claude Guillemin,
BP 36009, F-45060 Orléans, France
2
NECS, 196 rue Houdan, F-92330 Sceaux, France

Abstract:
Earthquake shaking applies complex loading to a structure. It cannot be completely
characterized by a single parameter, e.g. peak ground acceleration. However, fragility curves
are generally developed using a single parameter to relate the level of shaking to the expected
structural damage. The main goal of this work is to use several parameters to characterize the
earthquake ground motion and relate these to the expected damage. The fragility curves will,
therefore, become surfaces when the ground motion is represented by two parameters. To this
end, the roles of various strong-motion parameters on the induced damage in the structure are
compared through nonlinear time-history numerical calculations. A robust structural model
that can be used to perform numerous nonlinear dynamic calculations, with an acceptable cost,
is adopted. The developed model is based on the use of structural elements with concentrated
nonlinear damage mechanics and plasticity-type behavior. The relations between numerous
ground-motion parameters, characterizing different aspects of the shaking, and the computed
damage are analyzed and discussed. Natural and synthetic accelerograms were
chosen/computed based on a consideration of the magnitude-distance ranges of design
earthquakes. A complete methodology for building fragility surfaces based on damage
calculation through nonlinear numerical analysis of multi-degree-of-freedom (MDF) systems
is proposed. The fragility surfaces are built to represent the probability that a given damage
level is reached (or exceeded) for any given level of ground motion characterized by the two
chosen parameters. The results show that an increase from one to two ground-motion
parameters leads to a significant reduction in the scatter in the fragility analysis and allows the
uncertainties related to the effect of the second ground-motion parameter to be accounted for
within risk assessments.

Keywords: seismic vulnerability, fragility surfaces, numerical structural modeling, dynamic

analysis, structural damage, earthquake risk assessment

∗
Corresponding author: D. Seyedi, [email protected], Tel. +33 (0)2 38 64 34 08, Fax: +33 (0)2 38 64 36 89
1. Introduction
Accurate earthquake risk evaluation requires a correct estimation of the seismic hazard and a
good evaluation of the seismic vulnerability of structures through an appropriate earthquake
damage model. The induced aggression (e.g. caused by an earthquake) should be expressed in
terms of a number of accessible and pertinent parameters for use in vulnerability analysis.

Extensive studies in the United States (e.g. HAZUS [1]) and in Europe (e.g. RISK-UE [2; 3])
have produced efficient approaches for vulnerability assessment. Such procedures consist of
finding the expected damage level induced in each kind of element at risk for a given hazard
level. Depending on the size of the area considered and the importance of the element at risk
the method may be: statistical (semi-empirical vulnerability functions), based on simplified
structural models or methods of analysis/assessment, or through direct time-history nonlinear
computations.

Existing vulnerability assessment methods can be classified into these four groups [4]:
empirical, judgmental, analytical and hybrid. The damage data used within these methods are,
respectively: observed post-seismic surveys, expert judgment, analytical simulations or
combinations of these. Each data source has associated advantages and disadvantages. A
critical review of existing vulnerability functions can be found in Rossetto and Elnashai [4].

Analytical vulnerability curves adopt damage distributions simulated from the analysis of
structural models under increasing earthquake loads (e.g. [5; 6]). A variety of analysis
procedures have been followed, ranging from the elastic analysis of equivalent single-degree-
of-freedom systems [7], adaptive pushover analysis of multistory models [6] to nonlinear
time-history analyses of 3D models of RC structures [8]. The choices made for the analysis
method, structural idealization, seismic hazard characterization and damage models strongly
influence the derived curves and have been seen to cause significant discrepancies in seismic
risk assessments made by different groups for the same location, structure type and seismicity
[9].

The classical procedures are based on the study of simplified substitute structures that are not
capable of accounting for the load redistribution inside structures due to local nonlinearities.
The objective of damage assessment is to evaluate expected losses based on sufficiently
detailed analysis and evaluation of the vulnerability characteristics at a given level of
earthquake ground motions. The conditional probability that a particular building will reach a
certain damage state should be determined using a fragility model. Note that in this article we
reserve the use of the expression ‘fragility’ for this specific purpose whereas ‘vulnerability’ is
used in a more general sense and when only the average damage level given a ground-motion
level without any notion of probability.

A single ground-motion parameter (e.g. peak ground acceleration, PGA, or macroseismic

intensity) cannot accurately characterize the loading applied to a given structure from an
earthquake. However, in current vulnerability assessment methods earthquake hazard is often
characterized by a single parameter, such as PGA or spectral displacement, SD, at the natural
period of the structure, which is perhaps modified by a factor to account for earthquake
duration such as in the HAZUS methodology [1]. Fragility curves are then developed by using
this single parameter to relate the level of shaking to the expected damage.

2
This standard method to develop fragility curves neglects the variability in the estimated
damage caused by the use of a single ground-motion parameter. Current fragility curves often
neglect this scatter, which means that this uncertainty cannot properly be propagated to
subsequent parts of the risk analysis nor can the importance of this variability be assessed.

Recently, some efforts have been made to introduce the effect of several ground-motion
parameters on the structural damage (e.g. [10; 11; 12]). Kafali and Grigoriu [11] used an
alternative intensity measure expressed by two parameters: earthquake magnitude and source-
to-site distance. On this basis, they proposed fragility surfaces for single degree-of-freedom
systems using simple synthetic ground motions. A vector-valued ground-motion intensity
measure (IM) consisting of pseudo-spectral acceleration and some additional parameters
related to shape of the response spectrum has also been investigated (e.g. [10]). Rajeev and
co-workers used spectral acceleration at two different periods as the IMs [12]: the spectral
acceleration (SA) at the first-mode structural period T1 is the first component of the IM vector
and the second period is selected to increase the efficiency of the IM (i.e. to minimize the
dispersion). With reference to an example RC-frame structure, the accuracy of prediction of
the seismic risk using the considered vector IM vs. a conventional scalar IM is presented.
They showed that an effective choice of the second period T2 leads to an estimate of the
seismic risk to that obtained employing a scalar IM consisting of SA(T1) only, while reducing
the associated dispersion in the estimate. However, for the studied example structure, the
reduction is negligible in light of the effort required for switching from a scalar to a vector IM.

Although previous researchers have shown that using more than one strong-motion intensity
parameter can lead to a better prediction of damage, very few have gone the extra step to
develop fragility functions for various damage states explicitly involving more than one
parameter and those studies that have are limited due to their use of simplistic structural
models (e.g. single-degree-of-freedom systems), ground motions (e.g. stochastic time-
histories) and/or damage levels (e.g. collapse or non-collapse). The main goal of the present
article is to improve the representation of the strong ground motion by introducing the
fragility surface concept in risk assessments for current reinforced concrete structures through
nonlinear time-history analysis of MDF systems. In this approach, the shaking is
characterized by two IMs. The selected IMs should be poorly correlated for efficient
characterization of the shaking. On the contrary, the structural damage must be correlated to
the selected parameters.

To this end, the damage level of a typical reinforced concrete structure is evaluated by the use
of nonlinear numerical calculations. By considering the parts of the structure that would suffer
significant damage during strong ground motions (plastic hinges), an adequate 3D nonlinear
robust-yet-simplified finite element model is created to allow numerous computations. The
maximum inter-story drift ratio is used to define the damage level of the studied structure.
The relationships between various IMs and the computed damage are compared. Such a study
can help find a small number of ground-motion parameters that lead to, when used together to
characterize the shaking, the smallest scatter in the estimated damage. Fragility surfaces are
then proposed for the studied structure. Note, however, that we do not propose these surfaces
for use in practice since some geometrical simplifications were made to facilitate numerical
calculations. In addition, only the scatter in the estimated damage level due to ground-motion
variability is investigated and we assume that there is no variation in the material or geometric
properties of our structure nor is there any uncertainty in the estimation of the damage level
that the structure will sustain given a measure of the structural response (in our case
maximum inter-storey drift). These limitations of our study would need to be considered

3
when developing fragility functions to be used in practice for a set of structures. The main aim
of this study is to show the advantage of using fragility surfaces developed by time-history
calculations rather than curves and to suggest a methodology for their development.

2. Models and analysis

This section discusses the structural and material models used and the analysis conducted.
2.1. Structural model
The main goal of vulnerability assessment methods is to estimate the seismic response of
existing buildings. For this purpose we define the structural model based on an existing
building constructed in the 1970s. This is an eight-storey regular frame reinforced concrete
(RC) structure as shown in Figure 1. The structural system of the building mainly consists of
parallel shear walls in the Y direction, and RC beam-columns frame in the X direction.

The main purpose of the structural model developed for this study was not to account in detail
for all mechanisms of degradation that the structure could suffer during severe ground
motions but rather to develop a realistic model to be able to conduct many nonlinear
calculations to demonstrate the fragility surface concept. Therefore, the modeled structure was
simplified based on several considerations: the induced damage is assumed to be concentrated
at the extremities of beams and columns. The failure mode is of a plastic-hinge type; shear
failure may not occur. The used model does not account for phenomena such as concrete
crushing or steel rebar buckling. Once the concept of fragility surfaces is demonstrated and
methods for their construction are refined, our structural model may be enriched to account
for degradation mechanisms such as: shear failure, rebar buckling, concrete cover spalling,
axial force-moment interaction, cyclic fatigue and steel rebar debonding.

The structure is represented by a three-dimensional finite element mesh, using shells to

represent slabs and walls, beams for beams and columns, and discrete rotational elements for
beam/slab junctions. In the present study, slabs and shear walls are considered to remain
elastic linear. As mentioned above, the only nonlinearity was represented by concentrated
plastic hinges placed at the extremities of beams and columns. These were represented by
discrete elements, and their characteristics were derived from the design detailing (concrete
and longitudinal rebars). These elements may be used to provide rigid links, linear elastic
stiffness and nonlinear connections between two nodes in any of the six degrees of freedom.
In our case, only the flexural bending rotations were allowed to behave nonlinearly. The other
degrees of freedom were set as rigid links. Details on the material model for this element are
provided in the next section.

The building is anchored at its base, without considering any flexibility from the soil.
Permanent loads were taken into account. Finally, Rayleigh damping is included in the
calculation algorithm to assure a certain level of energy dissipation. The damping ratios were
set at 5% for the first two eigenfrequencies; this produces stronger damping for higher
frequencies of vibration of the model.

The earthquake is applied to the structure in the X direction as an acceleration time-history.

Only one horizontal component was used for each analysis. Calculations were carried out
with Code_Aster® [13] finite element program, using the Newton-Raphson implicit algorithm.

4
 Figure 1 - Finite element mesh of the studied building

2.2. Nonlinear material model

Observations made of buildings that have suffered damage during past earthquakes show that
well designed frame structures exhibit nonlinearities mainly through the creation of plastic
hinges at the ends of beams and columns (see Figure 2). Assuming that the building chosen
for the purpose of this study would behave similarly, and seeking an efficient model (see
Section 2.1), discrete finite elements were placed at the ends of all beams and columns to
allow the creation of plastic hinges.

Figure 2 – Examples of damage to RC frame buildings. Photographs taken by S. Ghavamian in Dujiangyan

following the 2008 Sichuan earthquake

The main features of the material model used to describe the behavior of these plastic hinges
are: maximum moment capacity, isotropic strain hardening and inelastic rotation (energy
dissipation).

5
The formulation of the model is given below:

Yield function f = M − M p .
If f <0 the behavior remains elastic: M& = K eθ& .
Otherwise the evolution follows the following conditions:
∂f
f = 0 and f& = 0 , with θ&p = λ& (1)
∂M
where Ke denotes elastic stiffness, Fe yield stress, Mp yield moment, Mu ultimate moment
capacity, n is a constant and:

K eθcp
M p(θcp) = 1
(3)
  K eθcp  
n n

1 +   
  M u  

where ke denotes the yield stiffness, θcp is the cumulated plastic rotation and Mu the ultimate
yield moment.

The behavior of the model for a cyclic loading is shown in Figure 3.

Figure 3 - Nonlinear behavior of the discrete models used to represent the formation of plastic hinges

To identify the parameters of this model, the reinforcement detailing of the cross section of
each column is introduced in a RC design software package to compute the global moment-
rotation evolution of the element, based on the material parameters of the concrete and steel
rebars.

6
3. Selection of input time-histories
As shown by Douglas [14] input time-histories for structural analyses are often selected
without considering the seismic hazard of the region of interest. In addition, it has been
common to employ a small set of ‘standard’ time-histories (e.g. El Centro 1940) when
conducting analyses even though this will underestimate the variation in possible structural
response and, in addition, some of these ‘standard’ strong-motion records will not be
appropriate for all sites due to, for example, the magnitude of the earthquake being larger than
is possible in the study region [14]. For the study summarized in this article we have paid
attention to the selection of time-histories that are appropriate for the seismic hazard of the
region of interest and also provide an appropriate range of motions so that the possible
variation in structural response is not underestimated – more details are given in [15].

To better understand the behavior of the modeled structure under earthquake loading and to
investigate which characteristics (e.g. amplitude, duration or frequency content) of the ground
motions are most important five sets of eight accelerograms were selected based on ideas
from the theory of Design of Experiments. The procedure followed is summarized in Douglas
[14] and the results obtained using these accelerograms are presented in Seyedi et al. [16].
These analyses showed that the effect of the duration of strong shaking (characterized by the
interval between 5 and 95% of total Arias intensity) does not have a significant impact on the
damage sustained by the studied structure. To explain this observation the following
simplifications in the structural model should be noted. With respect to local structural
behavior the nonlinear constitutive model does not exhibit any fatigue under reverse loading.
Therefore, discrete lumped damage elements placed at the extremities of beams and columns
would provide a steady capacity under constant loading. However, from a global structural
point of view the evolution of damage is more or less progressive with the duration of
earthquake shaking, which then induces varying patterns of load evolution at different parts of
the structure. This may be interpreted as fatigue behavior, depending on the degree of
regularity of the structure, possibilities of load distribution and the intensity of the ground
motion (sudden appearance of damage or their smooth progression). In addition, the analysis
showed that response spectral displacements at periods close to the natural period of the
modeled structure are strongly correlated to the damage level. These preliminary findings
were used as a basis for the more comprehensive analysis summarized here.
3.1. Selection based on seismic hazard of the French Antilles
In this article it is assumed that the analyzed structure is located in the French Antilles
(Guadeloupe and Martinique), which is a region affected by both crustal and subduction
(interface: shallow-dipping thrust events and intraslab: deep, generally normal-faulting,
events) earthquakes. Therefore, a set of strong-motion records were compiled that is
consistent with the seismicity of this region.

A magnitude-distance-earthquake-type filter is applied to a large database of strong-motion

records from various regions of the world (mainly the Mediterranean region, the Middle East,
western North America, central America, Japan and Taiwan) to exclude records from
magnitudes and distances that are not possible for the French Antilles. This selection has been
made by considering the seismogenic sources in the region in terms of their location and the
magnitude range of the earthquake they produce. Also records from small earthquakes and
great distances were excluded since such weak motions will not lead to significant damage for
the structure analyzed here. For this study, the seismogenic sources for the region of the
French Antilles proposed by Geoter [17] were used to select the records. Based on this study
these selection criteria were adopted:

7
Crustal earthquakes: 5.5 ≤ Mw ≤ 7.5, 0 ≤RJB ≤ 30km and 0 ≤ h ≤ 30km
Interface earthquakes: 5.5 ≤ Mw ≤ 8.0, 0 ≤ RJB ≤ 50km and 0 ≤ h ≤ 50km
Intraslab earthquakes: 5.5 ≤ Mw ≤ 7.5, 0 ≤ RJB ≤ 50km and 50 ≤ h ≤ 150km

where Mw is the moment magnitude, RJB is the distance to the surface projection of the
rupture (Joyner-Boore distance) and h is the focal depth.

3.2. Further selection to limit number of records required

The first step leads to a preliminary selection of many hundreds of strong-motion records.
Due to the long run times of the finite element modeling of the structural response it was
necessary to optimize the number of accelerograms selected. On the other hand a minimum
number of accelerograms covering the range of ground motions that could be experienced in
the region of interest must be selected. From the pre-selected dataset (Section 3.1), we retain
only those natural accelerograms with ground-motion characteristics that cover useful ranges
of possible motions (i.e. we discard the accelerograms that have the same characteristics as
previously selected records). In a later stage we introduce some synthetic accelerograms to
complete this dataset for poorly sampled ground-motion ranges.

To select a limited set of records that still cover the range of possible ground motions two
strong-motion intensity parameters that are poorly correlated were chosen and then records
are selected from intervals defined by ranges of these parameters. The two parameters
selected are: elastic response spectral displacement at 1s for 5% damping, SD(1s), and relative
significant duration (defined by the interval between 5 and 95% of total Arias intensity), RSD.
The ranges of observed SD(1s) and RSD were divided into 20 logarithmical spaced intervals
(i.e. 20 x 20 = 400 bins in total) and one record was randomly selected from each bin. The
aim of this selection is to choose the minimum number of records that covers the range of
possible ground motions. Due to a non-uniform distribution with respect to SD(1s) and RSD
records were only found in 129 out of the 400 bins: few strong-motion records were found
that had both long durations and high SD(1s) values. In addition to these 129 records analyses
had already been run using the 40 records selected during the preliminary analyses discussed
above. Therefore, in total 169 natural accelerograms were used as input to the structural
modeling to construct the fragility surfaces.

3.3. Use of simulated strong-motion records

Since the whole range of possible motions is not covered by the selected natural
accelerograms it was decided to augment the input time-history dataset with records simulated
using the non-stationary stochastic method of Pousse et al. [18], which is an extension of the
procedure of Sabetta & Pugliese [19]. In the technique of Pousse et al. empirical ground-
motion prediction equations (GMPEs) for a handful of characteristics of shaking (peak ground
acceleration, duration, Arias intensity and central frequency) are used to shape and scale a
Fourier spectrum defined from a seismological-based source model and an envelope function
consistent with seismological considerations (e.g. P and S phase arrival times). This envelope
and the estimated Fourier spectrum are then used to shape white noise in the frequency and
time domains to produce realistic strong-motion records. Advantages of this method are [20]:
its rapidity (hundreds of records can be generated in a few minutes); its simplicity (since it
does not require complex source, path and site modeling); and that the time-histories
produced include ground-motion variability (an important consideration when conducting
structural modeling).

8
The computer code of Pousse et al. was used to simulate time-histories with Mw between 4.5
and 7.5 in steps of 0.5, epicentral distances between 10 and 100km in steps of 10km and for
all five site classes of Eurocode 8. The GMPEs developed by Pousse et al. were used and their
distributions were truncated at three standard deviations in order not to simulate non-physical
ground motions. Since the GMPEs of Pousse et al. were developed from K-Net data from
Japan, like the French Antilles an area close to subduction zones, these equations are suitable
for ground-motion simulations in our study.

We confirmed that the ground motions simulated using the technique of Pousse et al. led to
damage levels consistent with those found using natural accelerograms (Figure 4). In total,
571 synthetic accelerograms were used in addition to the 169 natural records previously
selected. An advantage of using these simulated ground motions over the natural
accelerograms is that the structural modeling was faster due to the generally shorter length of
the records.

4. Selection of strong-motion parameters

The goal of this section is to find a small number of ground-motion parameters that lead to,
when used together to characterize the shaking, the smallest scatter in the estimated damage,
computed using the structural model presented in Section 2. Orthogonal parameters are sought
that characterize different aspects of the shaking, e.g. the amplitude, frequency content and
duration, so that the fewest number of parameters is needed. A balance needs to be struck
between introducing more parameters, thereby leading to a reduction in the scatter in the
fragility curve, and the uncertainty in the estimation of these parameters for given earthquake
scenarios.

Problems associated with the choice of parameters for ground-motion and damage
characterization can be identified in almost all existing vulnerability relationships [4]. The
parameter chosen to represent ground motion in the construction of vulnerability curves must
be both representative of the damage potential of earthquakes and easily quantifiable from
knowledge of the earthquake characteristics. Many existing vulnerability assessment methods
use macroseismic intensities to characterize the seismic aggression (e.g. ATC-13 [21], RISK-
UE [2; 3] and [22; 23]) but intensity is a subjective and discrete scale and, in addition, it
cannot be accurately predicted using GMPEs. In this study, we selected several strong-motion
intensity parameters that can be well estimated through GMPEs and that are likely to have a
strong influence on the building response.

As noted in Section 3, the plastic hinge model used in this study does not take into account
any phenomena like fatigue that depend on the number of cycles. Hence, we do not examine
the damage level with indicators such as the Park & Ang index [24], which relies on
hysteretic energy and accumulated plasticity. Therefore, it is more suitable to evaluate the
damage level by measuring the peak displacements of the structural elements during the
simulation. The inter-storey drift ratio (ISDR), i.e. the relative peak displacement between
two consecutive floors, is a widely-used indicator to measure the behavior of the structure.
We evaluate the ISDR at each storey and retain the highest value to give an overall idea of the
building damage level. The 740 selected accelerograms discussed in Section 3 were used to
run the nonlinear structural analysis. Table 1 presents the linear best-fit equations connecting
the inter-storey drift ratio and the IMs. The choice of the periods at which the spectral
displacements were examined was guided by eigenvalue analysis of the modeled building.
The two main eigenperiods along the X direction (the direction where the seismic input is

9
applied) were found to be 1.26 and 0.41s and, hence, SDs at these two periods were
considered as IMs.

Table 1 - Correlation between various IMs (x) and the building inter-stroy drift ratio (ISDR). The equation
reported is derived through linear regression.

Parameter Description Equation Correlation

(ISDR=mx+c) coefficient R²
SD(1.26s) Spectral displacement at the period 5.6025 x – 0.025 0.837
of the first mode
SD(0.41s) Spectral displacement at the period 20.99 x – 0.0084 0.650
of the second mode
PGA Peak ground acceleration 0.1239 x + 0.0213 0.578
AI Arias intensity 0.1506 x + 0.1479 0.565
AUD Absolute uniform duration (based 0.0834 x + 0.0454 0.532
on a threshold on 0.03g)
RSD Relative significant duration (based -0.0025 x + 0.3045 0.305
on the interval between 5 and 95%
Arias intensity)
RUD Uniform relative duration (based on -0.0027 x + 0.2972 0.003
a threshold of 10% PGA)
NC Equivalent number of effective 0.0191 x + 17.784 0.000
cycles (Rainflow method)

As shown in Table 1, the parameters that are related to the relative duration of the signal or
the number of cycles do not have any influence on the building response. This is unsurprising
since the plastic hinge model does not take phenomena such as fatigue into account. On the
other hand, spectral displacements at the natural structural periods are strongly correlated with
the ISDR (Figure 4).

Figure 4 - Linear regression between the spectral displacements at the first two natural structural periods and
the inter-storey drift ratio. Black squares (and solid lines) refer to natural accelerograms whereas gray
triangles (and dashed lines) are from synthetic records.

5. Fragility surfaces
This section presents the fragility surfaces developed for the prediction of the damage level.

10
5.1. Choice of parameters
The previous parametric study led to the choice of the SDs at the periods of the first and
second modes as useful variables for constructing the fragility surfaces (see Table 1). Many
input time-histories are required to construct reliable fragility surfaces. Considering a graph
with the two selected strong-motion IMs on the two axes, the chosen time-histories should
cover in a homogeneous manner the dataspace of possible motions. This is not really true for
the two selected parameters (Figure 5). Nevertheless these two parameters are sufficiently
uncorrelated (correlation coefficient of 0.83) to be used to build a fragility surface, e.g. for a
SD(1.26s) of 0.01m Figure 5 shows that SD(0.41s) can range between roughly
6 × 10-4m and 0.02m. Only some extreme situations (i.e. low SD(1.26s) and very high
SD(0.41s) and vice versa) are not represented; these situations are not, in fact, realistic for
earthquake shaking.

One potential drawback in using SD(T2) as an IM for the construction of fragility surfaces is
that the period of the second mode (T2) is not commonly measured (or calculated) or reported
for most RC buildings. Therefore, it could be difficult to construct or select appropriate
fragility surfaces requiring an estimate of T2 for a population of buildings, which in general
will have unknown T2s. However, since T2 can be easily computed by eigenvalue analysis
from numerical structural models (or measured in the field using ambient- or forced-vibration
measurements) it should be possible to develop guidance on the expected range of T2 for
typical RC buildings, which could be used if and when fragility functions are developed using
IMs based on T2.

Figure 5 - Distribution of selected time-histories in SD(0.41s)- SD(1.26s) space. The crosses (and the gray
triangles) represent the synthetic (and the natural) accelerograms respectively. The stars denote the nodes of
the grid (i.e. the points in the neighborhood of which the damage probability is evaluated).

11
5.2. Damage analysis
Based on thousands of observations Rossetto and Elnashai [4] developed empirical functions
that correlate the ISDR and the damage level for different types of European RC buildings.
Depending on the type of RC structure, relations convert the ISDR into a homogenized
reinforced concrete damage index (DIHRC), which can then be linked to the EMS-98 [25]
damage scale. Table 2 represents the relationships proposed by Rossetto and Elnashai [4]. The
correlation between DIHRC and EMS-98 damage level is shown in Table 3 [4]. Given the
model of the studied structure we use the relation for non-ductile moment-resisting frames to
estimate the damage level of the structure based on the ISDR.

Table 2 – Relationships between inter-storey drift ratio (ISDR) and DIHRC [4]

Structural system Proposed relationship R²

Non-ductile moment-resisting frames DIHRC = 34.89 ln(ISDR) + 39.39 0.991
In-filled frames DIHRC = 22.49 ln(ISDR) + 66.88 0.822
Shear-wall systems DIHRC = 39.31 ln(ISDR) + 52.98 0.985
General structures DIHRC = 27.89 ln(ISDR) + 56.36 0.760

Table 3 - Correlation between the damage index DI and the EMS 98 damage level [4]

DIHRC value EMS 98 damage level

DI < 0 D0
0 <DI <30 D1
30 < DI < 50 D2
50 < DI < 70 D3
70 < DI <100 D4
DI > 100 D5

5.3. Development of fragility surfaces

The following procedure has been developed to construct the fragility surfaces. For the two
chosen strong-motion intensity parameters (here, SD(T1) and SD(T2), T1 and T2 being the first
(1.26s) and second (0.41s) natural periods of the building) 25 couples (nodes) of coordinates
(xi, yi) have been selected so as to cover the space defined by the parameters x and y, where x
and y are SD(0.41s) and SD(1.26s), respectively (see Figure 5).

For each of the 740 accelerograms used in the simulations, we seek the closest nodes (xi, yi)
on the grid. To avoid bias due to differently-scaled parameters the distances between the
recorded SDs and the nodes are first normalized. For each node, the probability of exceeding
the damage level Dk is evaluated using the following equation:

P(d > Dk ( xi , y i )) = N ( Dk , ( xi , y i )) / N tot ( xi , y i ) (4)

where Ntot(xi, yi) is the number of points which are closest to the node (xi, yi).

This approach should enable us to obtain 25 probability values for each damage level Dk and
then through linear interpolation we could obtain the fragility surfaces. However, as shown in
Figure 5 not all the 25 nodes have sufficient points close enough to produce an accurate
surface. Hence, it was decided to consider only nodes that are in the neighborhood of at least
five points. Applying this constraint we only obtain a dozen acceptable probabilities, which is

12
not enough to draw a smooth surface through simple interpolation. Therefore we chose to
apply the following approach:
- a linear interpolation is performed between the dozen nodes;
- “slices” [by considering SD(0.41s) as constant] are extracted from the surface, for
many values of SD(0.41s);
- “slices” are extracted along the other axis [i.e. SD(1.26s)];
- a regression is performed, to find a, b, c and d, for each slice, using equation 5 (where
erf is the error function). This equation is adapted from the shape of commonly used
fragility curves, which usually correspond to a normal distribution cumulative
function:

  ln ( x ) − ln (b ) 
y ( x ) = a ⋅ 1 + erf   + d (5)
  c 

- the analytical equations are plotted to obtain a smooth surface [each point of the
surface is a mean between a curve along SD(0.41s) and a curve along SD(1.26s)].

Figure 6 shows the calculated fragility surfaces for four different damage levels: D1 to D4.
Each surface represents the conditional probability of reaching or exceeding Dk damage state “k”
for each given value of parameters SD(1.26s) and SD(0.41s).

Figure 6 - Fragility surfaces of the studied building for four damage levels (based on EMS 98). The
parameters are spectral displacements at T1 and T2, the periods corresponding to the main eigenmodes of
the building in the X direction.

13
These surfaces can be visualized as fragility curves by projecting the surfaces onto the
SD(1.26s) or SD(0.41s) planes (Figure 7, Figure 8). These figures demonstrate the wide
variation that can exist between fragility curves based on only one strong-motion intensity
parameter.

Figure 7 - Fragility curves with spectral displacement at period T1. The curves are plotted as “slices” of
the fragility surface for different values of SD(T2). Note that the slices have been slightly smoothed and
they are only plotted where they are sufficiently constrained by the analysis.

Figure 8 - Fragility curves with spectral displacement at period T2. The curves are plotted as “slices” of
the fragility surface for different values of SD(T1). Note that the slices have been slightly smoothed and
they are only plotted where they are sufficiently constrained by the analysis.

14
Figure 7 and Figure 8 demonstrate the prominent role that could be played by fragility
surfaces in risk analysis. Considering only one intensity measure parameter, e.g. SD(T1), we
could build an average curve that does not incorporate the variability in ground motions as
measured by another parameter, e.g. SD(T2). It can be seen (Figure 7) that there can be a
discrepancy of up to 50% between two curves (probability of 30% for SD(T2)=0.015 m and
80% for SD(T2)=0.05 m for an SD(T1) of 0.05m).

The obtained fragility surfaces are compared with the conventional fragility curves
considering only one intensity measure parameter on Figure 9 where the dotted lines represent
the fragility curves. These curves were computed using the method developed by Shinozuka
et al. [26], which expresses the fragility curve in the form of a two-parameter lognormal
distribution function:

 ln (x ) − ln(α ) 
F ( x) = φ   (6)
 β 

where x represents the studied parameter (in this case SD(T1) or SD(T2)), and Φ the
standardized normal distribution function. The two parameters α and β represent the median
and the lognormal standard deviation respectively and are computed so as to maximize the
likelihood function [27], given by equation 7.

N
M = ∏ [F ( xk )] k [1 − F ( xk )]
y 1− y k
(7)
k =1

where N is the total number of simulations, xk the SD (either at T1 or T2) of the k-th
accelerogram, and yk is equal to 1 or 0, whether the structure has reached the given damage
state or not (realization from a Bernoulli experiment). Table 4 gives the values of the
parameters α and β for each dotted curve on Figure 9.

Table 4 – Parameters of the equation for each fragility curve

α β
SD(1.26s) (Figure 7) 0.0721 0.2753
SD(0.41s) (Figure 8) 0.0190 0.4014

15
 Figure 9 – Comparison between fragility surfaces and fragility curves (Dk = D1). The solid line
corresponds to a diagonal cut of the fragility surface. The dashed line represents the analytical function
calculated by equation 6.

On Figure 9 the solid line corresponds to a diagonal cut of the calculated fragility surface.
This cut was made along a linear equation connecting SD(T1) and SD(T2) derived from
Figure 5, i.e. it accounts for the observed correlation between SD(T1) and SD(T2). The
comparison shows that a fragility curve constructed by a standard method and considering
only one parameter at a time gives results close to those obtained from a fragility surface
when two parameters are well correlated (along a diagonal cut, SD(T1) and SD(T2) vary in the
same manner). For other combinations of SD(T1) and SD(T2), e.g. a relatively high SD(T1)
and a relatively low SD(T2) (which may occur at a soft soil site that amplifies long-period
motions), a standard fragility curve can not represent the effect of the second parameter on the
seismic behavior of the structure. The main advantage of fragility surfaces is that the
variability of structural fragility due to a second IM can be estimated in contrast to when
fragility curves are used. This enables us to correctly propagate the uncertainties due to this
variability in risk assessments. Thanks to the fragility surfaces, it should be possible to
conduct more accurate risk analyses, which we plan to test in a future study.

6. Conclusions
The current methods used to model the seismic vulnerability of structures (i.e. through
fragility curves) often represent the ground motion by a single parameter (e.g. PGA).
However, a single parameter cannot fully represent the effect of an earthquake on the seismic
response of the structure. It is expected that an increase from one to two ground-motion
parameters would lead to a significant reduction in the scatter in the fragility function.

To this end, a nonlinear structural model to calculate the induced damage in a reinforced
concrete structure has been developed. Several parameters were chosen to represent the
characteristics of the earthquake shaking. Five sets, each containing eight strong-motion
records, were used, so that each set could represent the various combinations of the tested
parameters. A strong correlation between SD(T1), SD(T2) (where T1 and T2 are the periods of
the first and second modes) and the maximum inter-story drift ratio (as the damage measure)
is observed. The obtained results indicate that the parameters related to structural dynamic

16
characteristics (e.g. SD) are more suitable for fragility analysis this structural model, which
does not include fatigue effects.

Based on the obtained results, fragility surfaces that relate the strong-motion intensity
(represented by two parameters) to the possible damage of the structure are developed. The
spectral displacements at the two first modes of structure are chosen to represent the strong
ground motion. A comparison between different profiles obtained by the obtained fragility
surfaces shows the significant role of the second parameter in calculated damage probability.
If only one parameter, e.g. SD(T1) is considered, we can build an average curve that does not
incorporate the variability due to the other characteristics of the ground motions. Fragility
surfaces provide a robust way to propagate the effects of this variability in seismic risk
assessments.

As discussed in the Introduction, the analysis presented here is simplified to enable us to

conduct many hundreds of nonlinear time-history analyses in a reasonable time. Because of a
lack of field data some of the properties used in our simulations have been estimated.
Therefore, results of this study should only be considered qualitatively since the quantitative
results are sensitive to the assumed material properties and construction details. In the future,
a more detailed damage analysis should be performed in order to properly evaluate the role of
all parameters. In this view, the effect of each parameter on several damage indicators (e.g.
energy-based indicators and combination measures) can be investigated. The strong-motion
parameters must be chosen for each structural type as a manner to incorporate the dynamic
response of the structure.

7. Acknowledgements
The work presented in this article has been supported by the French Research National
Agency (ANR) through the CATTEL-2005 program (project ‘Seismic vulnerability of
structures: A damage mechanics approach’ (VEDA) under grant number ANR-05-CATT-
017). We thank the providers of the strong-motion and the structural data used. The authors
wish also to thank Prof. Hormoz Modaressi and Mr Florent De Martin for useful discussions.
In addition, we thank Guillaume Pousse for sending us his computer program to simulate
strong-motion records using his method. Finally, we thank Dr Tiziana Rossetto and an
anonymous reviewer for their detailed reviews of an earlier version of this article, which led
to significant improvements.

8. References
1. FEMA, HAZUS earthquake loss estimation methodology: User's manual. Federal
Emergency Management Agency, Washington DC, USA., 1999.
2. Lagomarsino, S. and Giovinazzi, S., Macroseismic and mechanical models for the
vulnerability assessment of current buildings. Bull. Earthquake Eng. 2006; 4: 415-443.
3. Mouroux, P. and Le Brun, B., Presentation of RISK-UE project. Bull. Earthquake Eng.
2006; 4: 319-321.
4. Rossetto, T. and Elnashai, A., Derivation of vulnerability functions for European-type RC
structures based on observational data. Engineering Structures 2003; 25: 1241-63.
5. Saxena, V., Deodatis, G., Shinozuka, M. and Feng, M., Development of fragility curves for
multi-span reinforced concrete bridges. In Proceedings of International Conference on Monte
Carlo Simulation, 2000
6. Rossetto, T. and Elnashai, A., A new analytical procedure for the derivation of
displacement-based vulnerability curves for populations of RC structures. Engineering
Structures 2005; 27: 397-409.

17
7. Mosalam, K., Ayala, G., White, R. and Roth, C., Seismic fragility of LRC frames with and
without masonry infill walls. J. Earthquake Engg. 1997; 1: 693-719.
8. Singhal, A. and Kiremidjian, A., A method for earthquake motion damage relationships
with application to reinforced concrete frames. Research Report NCEER-97-0008 State
University of New York at Buffalo: National Center for Earthquake Engineering, 1997.
9. Strasser, F. O., Bommer, J. J., Sesetyan, K., Erdik, M., Çagnan, Z., Irizarry, J., Goula, X.,
Lucantoni, A., Sabetta, F., Bal, I. E., Crowley, H. and Lindholm, C., A comparative study of
european earthquake loss estimation tools for a scenario in Istanbul. J. Earthquake Engg.
2008; 12(1): 246-256.
10. Baker, J. W. and Cornell, C. A., A vector-valued ground motion intensity measures
consisting of spectral acceleration and epsilon. Earthquake Engng Struct. Dyn. 2005; 34:
1193-1217.
11. Kafali, C. and Grigoriu, M., Seismic fragility analysis: Application to simple linear and
nonlinear systems. Earthquake Engng Struct. Dyn. 2007; 36: 1885-1900.
12. Rajeev, P., Franchin, P. and Pinto, P. E., Increased accuracy of vector-IM-based seismic
risk assessment? J. Earthquake Engg. 2008; 12: 11-124.
13. Code_Aster Code d'Analyses des Structures et Thermomécanique pour Etudes et
Recherches, 2008 http://www.code-aster.org, EDF R&D
14. Douglas, J., Strong-motion records selection for structural testing. 1st European
Conference on Earthquake Engineering and Seismology, ECEES, Geneva, Switzerland, 2006
15. Douglas, J., Selection of strong-motion records for use as input to the structural models of
VEDA. Rapport final BRGM/RP-54584-FR BRGM, 2006.
16. Seyedi, M., Gehl, P., Davenne, L., Ghavamian, S., Mezher, N., Douglas, J., Martin, F. D.
and Modaressi, H., Numerical modelling of the influence of earthquake strong-motion
characteristics on the damage level of a reinforced concrete structure. Colloque AFPS07, Paris,
2007
17. Martin, C., Combes, P., Secanell, R., Lignon, G., Fioravanti, A., Carbon, D., Monge, O.
and Grellet, B., Revision du zonage sismique de la France : Etude Probabiliste, Rapport de
Phase 3. Report GTR/MATE/0701-150, Affaire no. 1601, 2002.
18. Pousse, G., Bonilla, L. F., Cotton, F. and Margerin, L., Non stationary stochastic
simulation of strong ground motion time histories including natural variability: Application to
the K-net Japanese database. Bulletin of Seismological Society of America 2006; 96(6): 2103-
2117, DOI697 10.1785/0120050134.
19. Sabetta, F. and Pugliese, A., Estimation of response spectra and simulation of
nonstationary earthquake ground motions. Bulletin of Seismological Society of America 1996;
86(2): 337-352.
20. Douglas, J. and Aochi, H., A survey of techniques for predicting earthquake ground
motions for engineering purpose. Surveys in Geophysics 2008; 29(3): 187-220.
21. ATC-40, Seismic evaluation and retrofit of concrete buildings. Applied Technology
Concil Report, 1996.
22. Spence, R. J. S., Coburn, A. W. and Pomonis, A., Correlation of ground motion with
building damage: the definition of a new damage based seismic intensity scale. In:
Proceedings of the 10th World Conference Of Earthquake Engineering, Rotterdam, AA
Balkema, 1992 1, 551-556.
23. Orsini, G., A model for buildings’ vulnerability assessment using the parameterless scale
of seismic intensity (PSI). Earthquake Spectra 1999; 15: 463–83.
24. Park, Y. and Ang, A., Mechanistic Seismic Damage Model for Reinforced Concrete. J.
Structural Engg. 1985; 111: 722-739.
25. Conseil de l'Europe, European macroseismic scale 1998 (EMS-98). Chaier du Centre
Européan de Géodynamique et de Séismologie. G. Grünthal (Eds.) 1998 15.

18
26. Shinozuka, M., Feng, Q., Lee, J. and Naganuma, T., Statistical analysis of fragility curves.
J. Engineering Mech., ASCE 2000; 126(12): 1224-1231.
27. Shinozuka, M., Statistical analysis of bridge fragility curves. Proceedings of the US-Italy
Workshop on Protective Systems for Bridges, New-York, April 26-28., 1998

19