You are on page 1of 10

FRAGILITY SURFACES AS MEASURE OF SEISMIC PERFORMANCE

M. Grigoriu1 and E. M. Mostafa 2

ABSTRACT
Fragility surfaces give the probability of failure for structural or non-structural
systems as a function of two parameters, the earthquake magnitude and epicentral
distance. These parameters provide a unique characterization of the ground
motion. In contrast, the current characterization of the ground motion by a single
parameter, e.g., the Peak Ground Acceleration (PGA), can be unsatisfactory since
events with the same PGA can cause very different damages. In order to
demonstrate the methodology, fragility surfaces are calculated for a 22,000 gallon
water tank located at the 20th floor of a hospital in New York City. The analysis is
based on the linear random vibration theory, Monte Carlo simulation and
nonlinear dynamic analysis.

Introduction
Earthquakes are one of the most catastrophic events on earth. They are capable of
causing loss of lives, damage to buildings and systems, and interruption of essential services
such as electricity, gas, and water supply. Failure and movement of non-structural systems and
components can cause much more damage than the failure of the structural systems themselves.
It is crucial to evaluate the vulnerability of structural and non-structural systems to
earthquakes, as this helps in predicting realistic economic losses. It also helps with planning
emergency and recovery efforts. Most importantly, it helps in identifying the most vulnerable
systems and components and determining the need for retrofitting and strengthening of such
systems.
It has been a common trend to express the fragility of structures to earthquakes as a
relationship between the ground motion severity and structural damage. Fragility curves and
Potential Damage Matrices (PDM) are the most common format of this relationship. Both
methods describe the conditional probability of exceeding different levels of damage at given
levels of ground motion severity. As the names could suggest, fragility curves express the data in
a graphical format while PDM express it numerically.

Professor, School of Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853
Graduate Research Assistant, School of Civil and Environmental Engineering, Cornell University, Ithaca, NY
14853

Properties of both system and soil are embedded in the fragility curves and PDM.
However the seismic hazard at the site is not reflected. Fragility curves and/or PDM can be
integrated with seismic hazard information to give idea about the potential damage to the
system.
Fragility Curves
Fragility curves give the conditional probability of exceeding different levels of damage usually referred to as limit states - at different intensities of ground motion. Every curve gives
this conditional probability for a specific limit state over a range of ground motion intensity.
It has been proven that the uncertainty in the ground motion and soil properties at the site
can be much more than the uncertainty in the structural performance of the system. Accordingly,
many researchers have assumed deterministic parameters of the system. Some others considered
variation in the system parameters in addition to the ground motion and soil properties. Uniform
and lognormal distributions are among the most common models used with system parameters.
Types of Fragility Curves
Fragility curves can be developed for one specific system or for a class of systems. Based
on the method used to generate them, fragility curves are classified as either analytical or
empirical. Analytical fragility curves are generated using results of numerical simulations of the
system under artificial or historical earthquake records. Empirical fragility curves are based on
experimental results or damage data collected from the field after earthquakes. In some cases,
opinions of experts and personal judges can be the basis for empirical fragility curves.
Fragility curves have been developed for a great variety of structural systems over the last
few years. Examples include RC frame/wall systems (Hwang 1990 and 1994 and Singhal 1996),
RC structural walls (Sasani 2001), steel frames (Dimova 2000), and RC bridges. Unrestrained
equipment and piping systems are examples of non-structural systems for which fragility curves
have been developed (Chong 2000 and Mostafa 1999).
The main challenges for analytical methods are generating artificial ground motions that
are consistent with the site and relating numerical results of the simulation to predefined levels
of damage. Scarcity of available data is the main challenge for the empirical methods.
Characterization of Ground Motion Intensity
It is of great importance to select a suitable indicator that describes the intensity or
strength of ground motion against which fragility curves are plotted. The main properties of
ground motion, including amplitude, frequency content, and duration of strong motion, are
supposed to be reflected in this parameter. Any two or more ground motions having the same
value of this parameter are supposed to have the same potential ability to cause damage.
Different parameters have been considered by researchers to characterize ground motion.
Peak Ground Acceleration (PGA) is the most common parameter used (Hwang 1990 and

1994a). Elastic response spectral ordinates Sa (spectral acceleration) and Sv (spectral velocity) at
various frequencies have been used (Dimova 2000, Singhal 1996, and Sasani 2001), as have
qualitative parameters such as the Modified Mercali Intensity (MMI) (Dumova-Jovanoska
2000). The main advantage of PGA and response spectral ordinates is the ease of estimation.
MMI has the advantage over other parameters in that it has been recorded for old events for
which other parameters cannot be calculated. However, for artificial records, it is not visible
how to estimate such quantitative parameter for the generated motions.
It has long been recognized, however, that PGA has poor correlation with both actual
observed and theoretically computed damage in structures. In addition, elastic response-based
measures, Sa or Sv are incapable of describing the ability of ground motion to cause damage to
the structure through the various sequences and magnitudes of nonlinear response cycles.
In order to explore the visibility of using PGA as a ground motion parameter, a group of
120 artificial records have been generated using the stochastic model described below. The
epicentral distance was kept constant for all records and equal to 50 km. Six values of the
magnitude were considered, ranging from 5 to 6 at increments equal to 0.2. Twenty samples
have been generated for each magnitude value.
The response of a linear SDOF system under each
record was calculated and the maximum displacement,
velocity, and acceleration were obtained for each case.
The system has a frequency of 14 rad./sec and a modal
damping ratio of 5%. Figure 1 shows the maximum
response of the system against PGA for 20 samples with
the same magnitude and epicentral distance. The figure
and the values of the correlation coefficient reflect the
lack of correlation between PGA and the maximum
response of the system. For such simple system, PGA
failed to represent the ability of the motion to affect the
system. It is very unlikely for it to do the job with more
complicated nonlinear systems.
PGA has been shown to be a misleading and
inadequate parameter to characterize the ground motion
(Sewell 1989). It seems to be almost impossible to fully
characterize the ground motion with one parameter.

Figure 1.

Maximum response
vs. PGA

Fragility Surfaces
Similar to fragility curves, fragility surfaces give the conditional probability of exceeding
different limit states given the occurrence of earthquake of certain intensity. The only difference
is that the ground motion intensity is characterized with two parameters, magnitude, M, and
epicentral distance, D, instead of one parameter as the case is with fragility curves. Fragility
surfaces are plotted against these two parameters.

Limit states represent different levels of damage and are characterized using one or more
damage indices, DIi. The fragility surface corresponding to the limit state i is defined as follows:
Fi (x, y) = P [DI > DIi | Mw = x, D = y]

(1)

where DI is the damage index and DIi is its value corresponding to the ith limit state.
Probabilistic Model for Ground Motion
The input parameters for this model are magnitude, epicentral distance, and soil
properties. The model has been developed at the university at Buffalo and is based on the
Specific Barrier model (Papagiorgio 1983a and 1983b).
The ground motion acceleration, A(t), is modeled by a nonstationary stochastic process
~
A(t ) = w(t ) A(t )

(2)

~
where t is the time, w(t) is a deterministic amplitude modulator envelope function and A(t ) is a
stationary Gaussian process with a one-sided power spectral density g().

Numerical Example
An example of a steel water tank located at the 20th floor of a hospital in New York City
is considered to demonstrate the methodology of developing fragility surfaces for non-structural
systems. The effect of adding bracing elements to the system on its seismic performance is
explored by comparing fragility surfaces of the system with and without the bracing elements.
Problem Definition
The hospital consists of three sections: the plaza, the middle tower (core), and the main
tower. The plaza is seven stories high while the two towers are twenty stories high. Figure 2
shows a photo for the plaza and the middle tower of the hospital and Figure 3 shows the location
of the water tank on a schematic plan of the hospital. The hospital is a steel frame building with
full and partial moment resistance connections. Most floors in the building are made of two-way
concrete slabs. The building is supported by pile foundations.
The water tank has dimensions of 20.3x16.8x4.8 ft. It is supported at its corners by four
vertical legs each 6.83 feet high and a cross-section W 8 X 28. This representation is postulated
because of lack of detailed information on the secondary systems of the hospital.
The ground excitation is assumed to be a non-stationary stochastic process based on the
Specific Barrier model. It is also assumed that the ground acceleration has only one component
in the x direction (Fig. 3). The ground excitation depends on two parameters: magnitude and the
epicentral distance of the earthquake event. The fragility surfaces are plotted against these two
parameters.

Fragility surfaces are generated for the water tank with three limit states. The limit states
considered are DI > DIi, where DI is drift of the water tank legs in the x direction and DIi takes
the values of 0.5, 2 and 10% of the height of the legs.
X
Y

Figure 2.

Photo of the plaza and the middle


tower

Figure 3.

Schematic plan of the hospital

Assumptions
The primary structure remains linearly elastic during ground motion. Neither material
nor geometrical nonlinearity is considered. The first and the fourth modes of the primary
structure are dominant and are translation modes in the x direction. Therefore the four points
supporting the water tank can be assumed to have equal displacements in x direction and zero
rotations and translation in the y and z directions.
The body of the tank is modeled as a rigid body. All the deformations are assumed to be
concentrated in its four legs. The tank is assumed to be full of water and closed, so that the
problem of water sloshing is not considered.
The interaction between the secondary system and the primary structure is ignored. Only
the effects of the primary system are considered while the possible feedback from the secondary
system on the primary structure is ignored. The uncertainty in the primary-secondary system
parameters is not considered. The only source of randomness is the seismic input. Only the first
20 modes of the primary structure are considered in the modal analysis.
Solution Technique
The primary structure is analyzed separately neglecting the interaction with the
secondary system. The ground motion is filtered through the primary structure and the response
at the attachment points is calculated and used as the input to the secondary system. The
secondary system is analyzed under this input. Many samples of the ground excitation are used
to generate the fragility surfaces of the system using Monte Carlo simulation. Figure 4
summarizes the main steps of the methodology.

Compute Eigenvalues, Eigenvectors, and modal


participation factors of the primary structure

Choose the limit state, DIi

Choose a pair of magnitude and epicentral distance M-D of the


excitation on the primary structure

Compute the power spectral density function of the


excitation using the Specific Barrier model

Generate N(M,D) realizations for the input excitation to the


secondary system using Eq. 11

Analyze the secondary system under each of the N(M,D)


realizations and calculate Nex(DIi, M, D), the number of samples
exceeding the limit state DIi

Calculate fragility of the secondary system at the given


level of excitation M-D
Fi (M, D) = Nex(DIi , M, D) / N(M, D)

Repeat the analysis to obtain another pair ( M-D, Fi(M, D)) for the
given limit state DIi.

Fit a surface to ( M-D, Fi(M, D)) to get the fragility surface for the
limit state DIi

Repeat to get the fragility surfaces at different limit states DIi

Figure 4.

Main steps of the methodology

Response at Attachment Points


As mentioned above, the input to the primary system is a nonstationary stochastic
process. The synthetic acceleration time history a(t) can be generated as follows (Grigoriu
1995):
n

a (t ) = w(t ) k ( Ak cos( k t ) + Bk sin( k t ) )

(3)

k =1

where n is number of frequency intervals, k = g ( k ) , = c / n , c is cut-off


frequency, k = (k 0.5) , and Ak and Bk are independent Gaussian random variables of zero
mean and unit variance.
The response at degree of freedom i of the primary system is
m

ui (t ) =

j =1

ij q j (t )

(4)

where m is number of modes considered in the analysis, ij is the displacement at degree of


freedom i in mode j and qj(t) is the jth modal coordinate that can be calculated as follows
(Clough 1993):
j

q j (t ) =

dj

j j ( t s )

sin( dj (t s )) a ( s ) ds

(5)

dj = j 1 2
j

(6)

j, j and j are the modal participation factor, frequency and damping ratio for mode j,
respectively. Using Eq. 2, 3, and 4, the response ui(t) can be expressed as follows:
m

ui (t ) = k Ak ij

where
t

I1 = e

k =1

j =1

dj

I1 (t ) + Bk ij
j =1

dj

I 2 (t )

(7)

j j ( t s )

sin( dj (t s )) w( s ) cos( k s ) ds

(8)

j j (t s )

sin( dj (t s )) w( s ) sin( k s ) ds

(9)

I2 = e
0

For some forms of the function w(t), closed form solutions can be found for the integrals
I1 and I2. The following is a suitable choice for w(t) that closed form solutions were found.
w(t ) = c1 e c2t sin(c3t )

(10)

where c1, c2 and c3 are constants. Finally the acceleration time history at the attachment points
can be obtained as follows:

k =1

Ak ij
j =1

dj

..

I 1 (t ) + Bk ij
j =1

dj

..

I 2 (t )

..

u i (t ) =

(11)

The mode shapes and natural frequencies of the primary structure were calculated at the
University at Buffalo using SAP2000. The stiffness effect of in-fill walls was considered in the
Finite Element model using X-type brace elements. Rigid diaphragms were assumed in all the
floors. The modal participation factors due to ground excitation in x direction have been
calculated as well.
Eq. 11 is used to generate random excitations at the attachment points. 25 pairs of
magnitude-epicentral distance are considered with M ranging form 4 to 8 with increment 1 and
D ranging from 25 to 225 with increment 50 km.
Nonlinear Dynamic Analysis
The secondary system was studied using
DIANA under each of the excitations that were
generated in the previous step. The body of the
tank was assumed to be rigid so that deformations
are concentrated in the four legs. The frequencies
of the first three vibration modes of the tank are
9.34, 15.19, and 17.95 rad./sec., respectively.
Both geometrical and material nonlinearities were
considered in the analysis. The material model
used for the four legs was von Mises plasticity
with strain hardening as shown in Fig. 5. For each
sample, the maximum drift of the legs, DI, was
calculated.

1, ksi
60
36

0.001

Figure 5.

0.132

0.585

Material model

Fragility Surfaces
The fragility surfaces were generated for three limit states of the secondary system
corresponding to drifts of 0.5, 2 and 10%. For a given limit state, DIi, and a specific magnitudeepicentral distance pair of values x and y, respectively, the fragility is estimated as the ratio of
the samples with maximum drift, DI, exceeding DIi.

Fi (x, y) = Nex (DIi, x, y) / N (x, y)

(12)

where N is the total number of samples generated for the given x and y. Figs. 6 and 7 show the
fragility surfaces of the system with and without bracing, respectively. They show the effect of
adding bracing elements to the system in reducing the potential damage to earthquakes of
different levels.
DI1 = 0.5 %

DI2 = 2 %

D, km

D, km

Figure 6

D, km

DI2 = 2 %

Fragility surfaces without bracing

DI1 = 0.5 %

D, km

DI3 = 10 %

D, km

Figure 7

DI3 = 10 %

D, km

Fragility surfaces with bracing


Conclusions

A methodology for generating fragility surfaces for structural and non-structural systems
has been developed. It was shown that using only one parameter to characterize the ground
motion could be misleading and inaccurate. A simple example of a 22,000-gallon water tank on
the top of a hospital was considered to show the potential of the introduced methodology.
Acknowledgments
This paper is based on research supported by the Multidisciplinary Center for
Earthquake Engineering Research (MCEER) at Buffalo, N. Y.
References
Clough, R. W. and J. Penzien (1993). Dynamics of Structures, McGraw-Hill, Inc., New York.

Dimova, S. L. and K. Hirata (2000). Simplified seismic fragility analysis of structures with two types of
friction devices, Earthquake Engineering and Structural Dynamics 29, 1153-1175.
Dumova-Jovanoska, E. (2000). Fragility curves for reinforced concrete structures in Skopje (Macedonia)
region, Soil Dynamics and Earthquake Engineering 19, 455-466.
Grigoriu, M. (1995). Applied Non-Gaussian Processes: Examples, Theory, Simulation, Linear Random
Vibration, and MATLAB Solutions. Prentice Hall, Upper Saddle River, New Jersey.
Hwang, H. M., and J. Huo (1994a). Generation of hazard-consistent fragility curves, Soil Dynamics and
Earthquake Engineering 13 (5), 345-354.
Hwang, H. M., and J. Huo (1994b). Generation of hazard-consistent ground motion, Soil Dynamics and
Earthquake Engineering 13 (6), 377-386.
Hwang, H. M., and J. Jaw (1990). Probabilistic damage analysis of structures, Journal of Structural
Engineering, ASCE 116(17), 1992-2007.
Mostafa, E. M. and M. Grigoriu (1999). Fragility curves for non-structural systems, Report No. 99-2,
School of Civil and Environmental Engineering, Cornell University, Ithaca, NY.
Papageorgiou, A. S. and K. Aki (1983a). A specific barrier model for the quantitative description of
inhomogeneous faulting and the prediction of strong ground motion. Part I. Description of the
model, Bulletin of the Seismological Society of America 73 (3), 693-722.
Papageorgiou, A. S. and K. Aki (1983b). A specific barrier model for the quantitative description of
inhomogeneous faulting and the prediction of strong ground motion. Part II. Applications of the
model, Bulletin of the Seismological Society of America 73 (4), 953-978.
Sewell, R. T. (1989). Damage effectiveness of earthquake ground motion: characterizations based on the
performance of structures and equipment, Ph.D. Thesis, Stanford University, Stanford, CA.
Singhal, A. and A. S. Kiremidjian (1996). Method for probabilistic evaluation of seismic structural
damage. Journal of Structural Engineering, ASCE 122(12), 1459-1467.
Sasani, M. and A. D. Kiureghian (2001). Seismic fragility of RC structural walls: displacement approach.
Journal of Structural Engineering, ASCE 127(2), 219-228.
Chong, W. H. and T. T. Soong (2000). Sliding fragility of unrestrained equipment in critical facilities.
Report No. 00-0005, MCEER, Buffalo, NY.

You might also like