You are on page 1of 32

CLASS PRESENTATION IN PROBABILITY

FRAGILITY ANALYSIS PARAMETERIZED BY


FUNDAMENTAL RESPONSE QUANTITIES THAPATHALI CAMPUS
Institute of Engineering

Presented By:
Er. Girijesh Prasad Sah
M.Sc. in Earthquake Engineering 2076 Batch
Roll No:- THA076MSEQE007
  Seong-Hoon Jeong a,∗, Amr S. Elnashai b,1
a Department of Architectural Engineering, Inha University, Incheon, Republic of
Korea
b Civil and Environmental Engineering Department, University of Illinois at Urbana-
Champaign,
  Urbana, IL, USA
Received 12 September 2005; received in revised form 21 June 2006; accepted 23
June 2006
   Available online 2 October 2006
Abstract

Analytical probabilistic fragility studies require extensive computer simulations to


account for the randomness in both input motion and response characteristics. In this study,
an approach whereby a set of fragility relationships with known reliability is derived based
on the fundamental response quantities of stiffness, strength and ductility is presented. An
exact solution for a generalized single degree of freedom system is developed and employed
to construct a Response Database of coefficients describing commonly used log-normal
fragility relationships. Once the fundamental response quantities of a wide range of structural
systems are defined, the fragility relationships for various limit states can be constructed
without recourse to further simulation. The uncertainty associated with modeling
simplifications is quantified by conducting comparisons between the proposed approach
and detailed multi-degree of freedom (MDOF) systems. Application examples are given
to demonstrate the efficiency of the proposed approach.
1. Introduction
• The use of fragility curves, defined as a relationships between ground shaking intensity
and the probability of reaching or exceeding a certain response level, for the
assessment of seismic losses is in increasing demand, both for pre-earthquake disaster
planning and post-earthquake recovery and retrofitting programs.

• Existing fragility curves can be classified into the four generic groups
o Empirical
o Judgmental
o Analytical and
o Hybrid

The various methods of fragility assessment differ in the required resources and precision
of the assessment results. Therefore, the choice of a method should be made considering the
tradeoff between effort and precision which best fits the application. The aim of the
method proposed in this paper is to provide a tool for the construction of fragility curves
with a reasonable precision level using a rapid procedure that is mathematically sound.
In order to overcome the limited applicability of observational data and subjectivity in
damage estimation by expert opinion, an analytical platform is employed.
2. Overview of the proposed method
Fragility curves may be derived analytically by simulations. Even for a limited number of
random variables and for modest ranges of variation of these variables, the simulation effort is
very considerable, reaching several hundreds of thousands of analyses. Every time the
structure is replaced or modified, repetition of the simulation is required. It is hereafter proposed
to parameterize the problem in such a manner that a generic set of fragility curves is derived.
 
3. Response parameters
In earthquake engineering, inelastic static procedures (referred to in Applied Technology
Council documents as nonlinear static procedure — NSP) have been widely used for the
assessment of structures due to their simplicity and efficiency. Modern seismic design and
assessment guidance documents, such as ATC-40 [7] and FEMA 273 [11], incorporate
NSPs for estimating the peak displacements of multistory buildings. In the latter methods,
the peak displacement estimate is based on a first mode analogy of the building in conjunction
with an equivalent single degree of freedom (ESDOF) system. The ESDOF system can be
used to estimate the response of multi-degree of freedom (MDOF) systems when the response
is predominantly in a single mode.

The stiffness of a system is represented by the period T and the strength is


determined by SR. Ductility is calculated by dividing the ultimate displacement (∆u ) by
the yield displacement (∆y ). The definition of the ultimate displacement is determined by
the analyst according to the structure under consideration and does not affect the
proposed calculation framework.
Fig. 2. Bilinear force–displacement relationship and response parameters
4. Demand estimation employing Response Databasesa

In the proposed method, the fragility curve is constructed using parameterized structural
response characteristics (T , SR, α and damping) and the Response Database (RD). The
structural response parameters are defined for the SDOF system that is equivalent to the
complex structure.

The proposed methodology has conceptual analogy with earthquake response


spectra, because it
(i) utilizes simplified structural models (SDOF system),
(ii) obtains the maximum value of the response history and
(iii) constructs curves which replace dynamic response history analyses

The Response Database is obtained from pre-run dynamic analysis results for a range of
structural response parameters. Simulation is therefore no longer needed for a newly
defined structural system.
4.1. Response Database

Fig. 3. Structure of the Response Database.


Fig. 4. Comparison of the computation time.

The structure of the Response Database is represented in Fig. 3. The database is designed
to store information on maximum responses of a wide range of structures as statistical
parameters. This enables the analyst to construct fragility curves by dealing with only two
statistical parameters (mean and standard deviation) instead of massive data from a large set
of dynamic response history analyses.
4.2. Response estimation

As an illustrative example, a generic structure with an elastic-perfectly-plastic force–


displacement relationship (Fig. 5) is studied. The period and strength ratio of the structure are 0.8
s
and 0.13, respectively.

Fig. 5. Force–displacement relationship


of the example structure.
As indicated in Fig. 3, the means and standard
deviations of maximum displacement responses from a
series of inelastic dynamic response history analyses
are collected and organized in the Response Database.
Each cell of the response database contains six
coefficients of a sixth order polynomial regression
function with intercept at zero (0, 0) and represents the
mean or standard deviation of the maximum
displacements as a function of earthquake intensity, as
shown in Eq. (1).
(a) Mean of maximum displacement.
y = a1 · x 6 + a2 · x 5 + a3 · x 4 + a4 · x 3 + a5 · x 2 + a6 · x .
(1)
Here x is the earthquake intensity and y is the mean or
standard deviation of the response quantity.
The response estimation results of the example
structure (Fig. 5) with 5% damping under Record Set 1
(Section 5.1) are represented in Fig. 6. The mean (µ) and
standard deviation (SD, σ ) of maximum responses are
expressed as sixth order polynomial functions of
earthquake intensities (Eq. (1)).
(b) Standard deviation (SD) of maximum displacement.
Fig. 6. Mean and standard deviation of maximum displacement of the
example structure.
4.3. Comparison with CSM
Fig. 7(a) represents
the mean and mean ±
σ (standard deviation)
elastic demand
spectra With 5%
damping ratio (ξ ) for
0.3g peak ground
acceleration (PGA)
level (Record Set 1). (c) Mean − σ elastic and inelastic demand.
(a) Mean and mean ± σ elastic demand.
Estimation of mean and mean ±
σ of maximum displacements
using the CSM for the example
structure in Section 4.2 above (T
= 0.8 s and SR = 0.13) is
represented in Fig. 7(b), (c) and
(d). In the latter figures, the
capacity curve overlaps with
mean (or mean ± σ ) elastic and
inelastic demand spectra for
0.3g PGA (d) Mean + σ elastic and inelastic demand.
(b) Mean elastic and inelastic demand.
Fig. 7. Capacity–demand diagrams for the estimation of maximum responses (PGA = 0.3g).
The estimates based on the Response Database are compared with those by the Capacity
Spectrum Method (CSM), one of the most widely used NSPs in earthquake engineering. In the
CSM, the maximum displacement is estimated by locating the intersection point of the capacity
curve and demand spectra. The latter are inelastic spectra in the acceleration–displacement (or
composite) response spectra (ADRS) format. Inelastic spectra represent the modified
response of an SDOF system according to the level of inelasticity. In this study, constant
ductility spectra by inelastic dynamic analysis are employed for the comparison because the
latter method gives exact results of maximum inelastic response of an SDOF system.
In order to obtain the mean and standard deviation of maximum displacements by the CSM,
the approach proposed by Shinozuka [16] is employed. In the latter study, the mean and mean ±
σ (standard deviation) of maximum displacements are obtained from the CSM with mean and
mean ± σ demand spectra, respectively. The latter spectra are developed by plotting the mean
and mean ± σ of the spectral acceleration and displacement in the ADRS. The latter procedure
requires the calculation of the mean and standard deviation of spectral accelerations and
displacements for a set of records with a given earthquake intensity.
Fig. 8. Comparison of response estimation results by using the Response Database (RD) and the CSM.

Comparison of calculations by the Response Database (RD) and the CSM in Fig. 8
shows that the difference between the two methods is small. Therefore, the
parameterized fragility method (PFM) instantly provides response estimates required for
fragility analysis with neither iterations nor repetitions. It gives response estimates that
are as accurate as (theoretically, more accurate than) the use of inelastic composite
spectra and the capacity spectrum assessment approach.
 
5. Effect of model simplification
 
In order to quantify the uncertainty of the response estimation based on the PFM, probabilistic
relationships between analysis results based on the ESDOF and the detailed MDOF systems are
investigated.
 
5.1. Reference simulation
 
Dynamic response history analyses were performed for detailed inelastic MDOF models of the reference
structures. The latter structures are selected to have limited irregularities and their configurations are
shown in Fig. 9. The bridge ‘CU’ is a typical MSSS (multi-span simply supported) concrete girder bridge
with multi-column bents in the Central United States. The bridge has three spans with a length of 15.01
m each, constructed of eight AASHTO Type I girders. The bents consist of a 1066.8 mm wide by 1219.2
mm deep reinforced concrete bent beam supported by three circular RC columns of 914 mm diameter.

(a) Bridge CU.


(b) Bridge SM.

(c) Bridge HS.

(d) 12-story RC frame. (e) 8-story RC frame.

Fig. 9. Layout of the reference structures.


Mid-rise RC frame buildings are designed in accordance with a modern seismic code [21], split into two groups: sets
of four 12-story RC frames (Fig. 9(d)) and four 8-story RC frame (Fig. 9(e)). Within each group, combination of two
design ground accelerations (0.15g and 0.3g) and three design ductility classes (High, Medium and Low) lead to the
four cases of each set. In the analytical models of the buildings, the ends of horizontal elements within the beam
–column joints are considered rigid. Shear spring connection elements are used to represent the shear stiffness of the
beam–column connection. The effective slab width participating in beam deformation is taken as the mean width plus
7% of the clear span of the structural member on either side of the web. In addition to the dead loads, 30% of the
live loads are accounted for to calculate the masses, while 100% of the live loads are used for the initial gravity
loads. Lumped masses are distributed at three locations on all individual beams in addition to all beam–column
connections. Viscous damping is not included in the models while hysteretic damping is considered by nonlinear
material modeling. Further details regarding the modeling are given in Mwafy [22].
The reference simulations of the detailed MDOF models of the three reference bridges were carried out with six sets
of ground motions. Record Sets 1–3 are records simulating earthquake events for the Lowlands soil profile
(Holocene deposit [28]) in the Memphis area, TN, USA and each of them consists of ten synthetic records. The three
sets of ground motions employed are based on three scenario earthquakes; large magnitude–short distance, medium
magnitude–medium distance and small magnitude–long distance. Details of the generation of the latter sets of
artificial records are given in Drosos [29]. Record Sets 4 and 5 are medium-to-large- magnitude small-distance (4.5 <
Mw < 7, 10 km < R < 30 km) natural records in Japan and Western USA, respectively. Record Set 6 consists of 35
major earthquakes around the world. The list of the natural record sets (Record Sets 4–6) is given in Appendix. The
reference simulations of the eight RC frames were performed with the three artificial record sets for the Memphis area
(Record Sets 1–3).In order to obtain a wider spectrum of response data of the reference structures, the ground motions
are scaled to various PGA levels, from 0.1 to 0.9g with a step of 0.1g. The numbers of dynamic response history
analyses using the detailed MDOF models are 2295 (3 bridges × 85 records × 9 intensity levels) for the reference
bridges and 2160 (8 buildings × 30 records ×9 intensity levels) for the mid-rise RC frame buildings.
 
5.2. Structural simplification
 
For transformation of the MDOF structure to the ESDOF system, it is assumed that the response of the
structure is predominantly in the fundamental vibration mode. The capacity curve of the MDOF system is
obtained from pushover analysis with lateral forces in proportion to the product of masses and
fundamental mode shape and it is transformed to the force–displacement relationship of the ESDOF
system. The base shear (V ∗) and displacement (u∗) of the ESDOF system are the corresponding
quantities (base shear, V ; displacement, u) of the MDOF system divided by the modal participation
  factor as
V ∗ = V /Γ and u∗ = u/Γ . (2)
The modal participation factor (Γ ) is defined as
The capacity curve of the ESDOF system is idealized to the bilinear force–displacement relationship(Fig. 10)
to obtain the fundamental response parameters. While the ESDOF simplification is an established
procedure, the bilinear representation needs engineering judgment. In this study, the effective lateral stiffness
is taken as the secant stiffness (K ∗) calculated at a base shear force equal to 60% of the yield strength of
y

the idealized structure, as shown in Fig. 10 [11]. The equal energy assumption is then employed to complete
the bilinear curve: the areas under the actual and the idealized curves until the displacement at plastic
mechanism (dp ) are approximately equal [21]. For a simpler procedure of idealization, the plastic
mechanism point is assumed to be at 2% drift ratio (u∗ = 0.02H/Γ ) which is the Damage Control
displacement limit: the threshold between moderate and major damage according to ATC-40 [7] and Zhong [30].
Based on the bilinear idealization of the force–displacement relationship (Fig. 10), response parameters of
the ESDOF system can be obtained as follows: the period is T =2π√m∗/K ∗, the strength ratio is
SR = V ∗/(m∗ · g) and the post-yield stiffness is a K ∗.

Fig. 10. Bilinear idealization of the capacity curve.


5.3. Maximum response evaluation by the ESDOF systems
The maximum displacement (umax) of the MDOF system can be estimated by the maximum response of
the ESDOF system as follows:

umax = Γ · u*max (5)


Since the ESDOF approach is based on the assumption that the deflected shape of the MDOF system is
proportional to a fixed shape throughout its response history, the maximum displacement at DOF i (∆i ) is
  expressed as Eq. (6).
∆i = φi umax (6)
For the building structure, the maximum interstory drift ratio, δi , between story i and story i − 1 may be
obtained by
  iδ = (∆ − ∆ )/ h
i i −1 i (7)
where hi is the story height between story i and i − 1. The maximum interstory drift ratio (δmax) of the whole structure is
 

 
δmax = Max (δ1, δ2, . . . , δn ) (8)
The reference structures are simplified to ESDOF systems using the procedure in Section 5.2 and their response under the
given record sets (Section 5.1) are obtained from the Response Database. The maximum displacement of the reference
bridges (Fig. 9(a), (b), (c)) and the maximum interstory drift ratio of the RC frame buildings (Fig. 9(d), (e)) are estimated by
Eqs. (5) and (8), respectively. Each of the latter responses is compared to the corresponding analysis result of the detailed
MDOF system to quantify the uncertainty of the response estimation by the ESDOF system in the following section.
5.4. Probabilistic relationship between response estimations by the ESDOF and MDOF systems
Since fragility analysis requires the predictive mean of response quantities, the uncertainty due to the
ESDOF simplification is quantified by comparing the mean responses of the reference structures
(Fig. 9) estimated by the two methods: ESDOF and detailed MDOF systems, as represented in Fig. 11
Based on the comparison in Fig. 11, to predict mean responses of the detailed MDOF system based
on the mean of the ESDOF system responses, the following relationships are suggested for structures with
limited irregularities. For bridges, the mean of maximum ductility demands can be estimated by

and for buildings, the mean of maximum interstory drift ratio may be obtained as follows.
a) Reference bridges. (b) Reference buildings.

Fig. 11. Relationships between the means of the maximum responses


estimated by the ESDOF and MDOF systems.
6. Derivation of fragility curves
6.1. Calculation of limit state probabilities

Three bridges (CU, SM, HS) and one building (12H30) in Fig. 9 are selected for the reference derivation of fragility
curves. For each structure, two limit states corresponding to (i)‘at least minor’ and (ii) ‘complete’ damage levels are
defined, as presented in Table 1. For the bridges, the limit state probability of the pier damage is calculated for fragility
analysis purposes. The limit states of the ‘complete’ damage are determined by previous research for the corresponding
example structure.

The ultimate curvature ductility of the bridge ‘CU’ is 5.24 according to Neilson [19]. The latter value is determined by
Bayesian update to incorporate the information from the physics-based approach [31] with the information from survey
results. The displacement ductility (µ∆) can be obtained by the following equations [32] using the given curvature
 ductility (µφ = 5.24).
µ∆ = 1 + 3(µφ − 1) · (L p /L) · [1 − 0.5(L p /L)] = 2.6 (11)
L p = 0.08l + 9db = 0.08 · 4.6 + 9 · 0.0254 = 0.6 (m). (12)
The probability of reaching or exceeding a limit state (LS) at a given earthquake intensity (s) can be
  expressed as follows:
  P (LS/s) = P [(dLS ≤ dmax)/s] = 1 − Φ (r) (13)
where dLS and dmax are limit state capacity and maximum demand, respectively. Assuming that the
responses follow a log-normal distribution, the standard normal variate r can be expressed as
6.2. Effect of randomness in material properties

(a) Brige CU. (b) Bridge SM.

(c) Bridge HS.


Fig. 12. Comparison of statistical parameters of maximum responses estimated by simulations with and without
considering material randomness.
6.3. Fragility curves

(a) Bridge CU. (b) Bridge SM.

(d) Building 12H30.


(c) Bridge HS.
Fig. 13. Comparison of fragility curves.
6.4. Application to retrofitting options

Fig. 14. Fragility contour.

If the point is to the right of or above the contour line, the corresponding perfor- mance target is satisfied. . Fig. 14
shows that all performance targets in Table 2 can be met by increasing the stiffness to 68 kN/mm (Opt #3), or by
increasing the stiffness to 59 kN/mm and the strength ratio to 0.82. In order to satisfy performance target T2, a
ductility-only intervention can be used and increasing the ductility to 4 (Opt #1) satisfies the retrofit objective
7. Advantages of the proposed method

  The parameterized fragility method (PFM) proposed in this paper is an efficient


tool for rapidly deriving probabilistic fragility relationships with quantifiable
levels of uncertainty. By virtue of its instantaneous nature, the proposed method
is especially useful for practical application of analytical fragility curves to the
planning of seismic rehabilitation, and regional earthquake mitigation where
fast estimation of probabilities of reaching damage states for a large number of
structural con- figurations and different mitigation measures is required. For
instance, seismic performance assessment of highway bridges requires instant
fragility results for a large inventory of different structures as an essential
component of transportation network analysis. The PFM responds to this
identified need.
8. Conclusions
Derivation of probabilistic fragility curves for a class of structures requires many analyses (tens or even
hundreds of thousands), especially when a large number of random variables is considered. These curves will
then have to be re-derived for different structural configurations as well as for different repair and/or
strengthening schemes. In this paper, a new analytical fragility assessment framework based on characterizing
a response database, parameterized by the fundamental quantities of stiffness, strength and ductility, is
proposed.The Response Database contains polynomial coefficients for structures with a wide range of stiffness
and strength. Therefore, with pre-determined stiffness, strength and a set of earthquake records, the database
provides mean values and their associated standard deviation of inelastic response quantities of the
corresponding structure without the need for further analysis. The effect of ductility variation is included in the
limit state definition in the calculation of cumulative conditional probabilities. Thus, once the user defines the
stiffness, strength and ductility of a structural system, a set of probabilistic fragility curves is instantly available,
the accuracy of which is the same as closed-form ESDOF-based inelastic simulation approaches and
acceptably close to fragilities from detailed, fiber-based finite element analysis. Verification examples are given
for both buildings and bridges. The implications of the success of the developed approach are wide-ranging.
For cases of selection between different retrofitting options, the parameterized fragility curves approach gives
rapid estimates of probabilities of various damage levels being inflicted onto the structures under consideration,
given only the stiffness, strength and ductility for each alternative retrofitting scheme. Additionally, the
presented fragility assessment method enables the analyst to practically investigate the vulnerability of large
numbers of different structural configurations instantly without simulation. Therefore, this method blends very
well with the Consequence-Based Risk Management (CRM) approach of the Mid-America Earthquake Center,
where the fragility assessment for generic structural systems in a large region is sought.
Acknowledgements
 
The work presented above was undertaken as part of the Mid-America
Earthquake (MAE) Center research project EE-1: Vulnerability Functions,
which is under the Engineering Engines Thrust Area. The MAE Center is a
National Science Foundation Engineering Research Center (ERC), funded
through contract reference NSF Award No. EEC-9701785. Finalization of the
paper was supported by the Inha University.
Thank you

You might also like