Vous êtes sur la page 1sur 15

THE STRUCTURAL DESIGN OF TALL AND SPECIAL BUILDINGS Struct. Design Tall Spec. Build.

19, 167181 (2010) Published online 21 October 2009 in Wiley Interscience (www.interscience.wiley.com). DOI: 10.1002/tal.546

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS UNDER EARTHQUAKE GROUND MOTIONS
FARZIN ZAREIAN1*, HELMUT KRAWINKLER2, LUIS IBARRA3 AND DIMITRIOS LIGNOS2
1

Department of Civil and Environmental Engineering, University of CaliforniaIrvine, Irvine, California, USA 2 Department of Civil and Environmental Engineering, Stanford University, Stanford, California, USA 3 Southwest Research Institute, CNWRA, San Antonio, Texas, USA

SUMMARY This paper summarizes collapse performance measures and the probabilistic basis for their development to assist in understanding of collapse behaviour of buildings and implementation of performance objectives in design and evaluation of buildings for collapse safety. Collapse in this context is dened as the loss of lateral load-resisting capability of a buildings structural system caused by ground shaking. Estimation of collapse performance requires the relation between a ground motion intensity measure (IM) and the probability of collapse, denoted as collapse fragility curve, and the relation between the same ground motion IM and the seismic hazard for the building, denoted as seismic hazard curve. Two methods for estimating the collapse fragility curve of a building are discussed: the EDP-based approach and the IM-based approach. In both approaches, collapse is associated with a scalar ground motion IM and is obtained by utilizing Incremental Dynamic Analysis. The collapse performance criteria presented in this paper are compared with the collapse performance criteria recommended in the SAC/ Federal Emergency Management Agency guidelines. An eight-storey moment-resisting frame case study is used to compare the estimates of collapse performance of various approaches discussed in this paper. Copyright 2009 John Wiley & Sons, Ltd.

1.

INTRODUCTION

Collapse of a building during and shortly after an earthquake is the consequence of loss of the buildings structural system integrity due to excessive deformation or force demand initiated in one, or several, component(s) of the buildings structural system. Excessive seismic demand triggers strength and stiffness deterioration in structural components and can lead to a partial or complete collapse of the building. Observations of collapsed buildings in past earthquakes show that two modes of collapse are the most common for a building: sidesway collapse and vertical collapse. Sidesway collapse is the consequence of successive reduction of load-carrying capacity of structural components that are part of the buildings lateral load-resisting system, to the extent that second-order (P-) effects accelerated by component deterioration overcome gravity load resistance. In contrast, vertical collapse is the result of direct loss of gravity load-carrying capacity in one or several structural components. Analytical tools devised by researchers to predict the collapse of a building subjected to a ground motion record range from single-degree-of-freedom representations of the evaluated building (Takizawa and Jennings, 1980; Bernal, 1987; Adam et al., 2004) to sophisticated nite-element-based

* Correspondence to: Farzin Zareian, Department of Civil and Environmental Engineering, University of CaliforniaIrvine, E 4141 Engineering Gateway, Irvine, CA 92697 USA. E-mail: zareian@uci.edu

Copyright 2009 John Wiley & Sons, Ltd.

168

F. ZAREIAN ET AL.

multi-degree-of-freedom (MDOF) models that can predict the successive failures of structural components up to collapse (Meguro and Tagel-Din, 2001; Sivaselvan and Reinhorn, 2002, 2006; Mosalam et al., 2008; Sasani and Kropelnicki, 2008; Sivaselvan et al., 2009). Paramount throughout this spectrum of analytical tools developed to predict building collapse is the ability to model strength and stiffness deterioration of structural components when subjected to cyclic loading (Rahnama and Krawinkler, 1993; Sivaselvan and Reinhorn, 2000; Song and Pincheira, 2000; Ibarra et al., 2005). In addition to deteriorating structural component models, sophisticated analytical tools for collapse prediction employ structural models that can incorporate large deformations (Sivaselvan and Reinhorn, 2002) and/or remove an element from the building model, once the component loses its load-carrying capacity (Khandelwal and El-Tawil, 2007; Mosalam et al., 2008). A somewhat different formulation is introduced by Meguro and Tagel-Din (2001), in which the building is modelled as an assembly of small elements connected by pairs of normal and shear springs. By using this model to simulate the response of a single-column pier subjected to the 1995 Kobe Earthquake, Sun et al. (2003) obtained a good correlation between results of the computer simulation and site investigation after the earthquake. The random nature of ground motions and the fact that no analytical model can mimic all building characteristics adds another dimension to the already complex problem of building collapse prediction under seismic excitations. Probabilistic approaches have been introduced that integrate possible sources of variability in the process of collapse prediction and present the buildings collapse potential in the form of probability of collapse (Cornell et al., 2002; Ibarra and Krawinkler, 2005; Haselton et al., 2007; Zareian and Krawinkler, 2007). In these approaches the fundamental assumption is that the effect of randomness in ground motion characteristics on estimating the collapse potential of a building is independent of the uncertainty in analytical tools used for earthquake hazard and building modelling. This paper evaluates current approaches for estimating the collapse potential of buildings in the context of Performance-Based Earthquake Engineering. The intention is to promote an approach that is less elaborate and easier to implement. Collapse, in the context of this paper, implies excessive sidesway motion and loss of stability of the structural system. The authors acknowledge that reliable collapse prediction in earthquakes has many obstacles to overcome and that we have a long road ahead of us before we can predict all collapse modes with condence. But we contend that we have to take small steps towards solving this problem or we will never come close to an answer. The approach advocated here is such a step, and it is fully recognized that it is an incomplete one. The goal of Performance-Based Earthquake Engineering is to provide seismic protection in accordance with specied performance objectives. A performance objective in the context of collapse can be expressed as a tolerable probability of collapse at a discrete hazard level or a tolerable Mean Annual Frequency (MAF) of collapse. Both performance objectives adhere to two dependent relations: the relation between a ground motion intensity measure (IM) and the probability of collapse (collapse fragility curve), and the relation between the seismic hazard and the same ground motion IM (seismic hazard curve). Both performance criteria and methods for estimating them are discussed in this paper. Two approaches are considered here for estimation of the probability of collapse of a structure given the level of ground motion intensity. These approaches are dissimilar as they use two different types of variables in the demand/capacity format. The approach that uses an engineering demand parameter (EDP) for estimation of probability of collapse given the level of ground motion intensity is denoted as EDP-based method. The approach that uses directly the ground motion IM for that purpose is denoted as IM-based method. Both approaches are introduced and compared in this paper. Methods for estimating collapse performance measures presented in the SAC/Federal Emergency Management Agency (FEMA) guidelines (FEMA, 2000), and those introduced by the authors Ibarra
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 169

and Krawinkler (2005) and Zareian and Krawinkler (2007) are briey reviewed and compared. In concept, both approaches combine the following three random and uncertain elements for estimating the collapse performance: ground motion intensity, seismic demand and seismic capacity. The difference between the method used in the SAC/FEMA guidelines and the one advocated in this paper comes from the sequence in which the aforementioned random and uncertain elements are combined. Furthermore, the SAC/FEMA guidelines dene demand/capacity based on an EDP, whereas the method introduced in this paper denes demand/capacity based on an IM. 2. COLLAPSE FRAGILITY CURVE

The collapse fragility curve of a building is the key component in evaluating its collapse performance. The collapse fragility curve shows the probability of collapse of a building for a given ground motion IM. Once the collapse fragility curve is tied with the buildings seismic hazard curve, which shows the relation between the ground motion IM and its return period for the location of the building, one can estimate building collapse performance in terms of probability of collapse at a given hazard level or average annual frequency of collapse of the building. Discussion about the collapse fragility curve, how it is obtained and its sources of variability is the subject of this section of the paper. Collapse performance measures are presented in subsequent sections. Conceptually, both the EDP-based approach and the IM-based approach are based on a demand/ capacity format in which the probability of collapse is obtained by computing the probability of seismic demand exceeding the buildings seismic capacity. However, the two methods are different in the sense that in the EDP-based approach a building response parameter (e.g., maximum interstorey drift ratio, maxIDR) is considered in the demand/capacity framework, whereas in the IM-based approach the ground motion IM is used. 2.1 EDP-based approach for estimation of collapse fragility curve

In the EDP-based approach for estimating the probability of collapse given IM, the building is assumed to be at the collapse state once the EDP demand surpasses the EDP capacity (i.e., EDPd EDPc). The EDP could be a deformation quantity (e.g., interstorey drift), or a force quantity (e.g., column axial force). The goal is to estimate a probabilistic representation of seismic demand (EDPd) and associated capacity (EDPc) and obtain the probability of collapse given IM, P[C|IM], as the probability that the EDPd exceeds EDPc, as shown in Equation (1). P [ C IM = imi ] = P [ EDPd EDPc IM = imi ] =
all edpc

P [ EDPd EDPc EDPc = edpci , IM = imi ] P [ EDPc = edpci ]

(1)

In this equation, P[EDPd EDPc|EDPc = edpci, IM = imi] is the probability that the demand exceeds the capacity value edpci at IM = imi and P[EDPc = edpci] is the probability that capacity is equal to edpci. The EDP-based approach for estimating the probability of collapse at a certain IM level using Equation (1) can have different implications according to the EDP considered in the demand/capacity format. If it is intended to estimate the probability of a global collapse of the structure, then the EDP used in Equation (1) should be capable of representing the global behaviour of the building. Cornell et al. (2002) suggest using the maxIDR among all stories of the building as EDP if the estimation of probability of global collapse is of interest. However, local EDPs (e.g., maximum plastic hinge rotation of columns or beams, column axial force, etc.) can be used in Equation (1) to provide estimates of probability of local collapse.
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

170

F. ZAREIAN ET AL.

Estimating the probability of collapse of a building given IM using Equation (1) requires the availability of P[EDPd edpd | IM = imi] and P[EDPc = edpc]. Among many methods for estimating P[EDPd edpd | IM = imi] and P[EDPc = edpc] detailed in Jalayer (2003) and Baker (2007), we have utilized the method that incorporates Incremental Dynamic Analysis (IDA) (Vamvatsikos and Cornell, 2002). In this method it is assumed that one set of ground motion records can represent different hazard levels, once they are scaled to the representative IM level. Values of EDPd|IM for the building at IM = imi are obtained by performing nonlinear response history analysis using a single set of ground motion records, each of which is scaled such that its IM value is equal to imi. This process is repeated for several values of IM starting with a small value and increasing it in small increments until the structure loses its lateral stability (the lowest IM value for which the nonlinear response history analysis does not converge). The P[EDPd edpd|IM = imi] can be found using the IDA information as a cumulative histogram or by tting a log-normal distribution to the EDPd values at the same IM level. The process for obtaining the P[EDPd edpd|IM = imi] for an eight-storey frame structure with rst mode period of T1 = 12 s. and base shear yield strength equal to 017 times the buildings weight, using a set of 40 ground motions denoted as LMSR-N (Medina and Krawinkler, 2003), is illustrated in Figure 1. In this gure each grey line is a single IDA curve that shows the relation between the IM and EDP for the evaluated structural and a ground motion record, obtained by using nonlinear response analysis with a modied version of the Drain-2DX software program (Prakash et al., 1993). The IM is the 5% damped spectral acceleration at the rst mode period of the building, Sa, and the EDP is the maxIDR. The intersection of the IDA curves with a horizontal line representing a level of ground motion intensity measure IM = imi provides values of P[EDPd edpd|IM = imi]. In Figure 1, the intersections for two horizontal lines at IM = 075 g, and 125 g, representing two arbitrary hazard levels, are shown with grey (purple coloured) dots, and corresponding cumulative histograms for (EDPD|IM = 075 g) and (EDPD|IM = 125 g) are shown with purple-coloured lines, respectively. The probabilistic representation of EDPc is also obtained from the IDA curves. We are using the SAC/FEMA denition for estimating EDPc for global collapse, i.e. the EDP value at which the slope of the IDA curve exceeds 20% of the median of the initial (elastic) slope of all IDA curves for the last time (FEMA, 2000). In Figure 1, the so-obtained EDP capacity points are shown with empty black

IDA Curves

P [EDPd< edpd |IM=imi] Onset of collapse (0.2Se) Cap. Data Points P[EDPc< edpc |IM=imi] P(C|IM)

IM[Sa/g]

0 0 0.05 0.1

0.0

1.0 0.15

P(C|IM)

EDP[max.IDR]

Figure 1. Obtaining probability of collapse given IM data points using EDP-based approach by utilizing incremental dynamic analysis
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 171

triangles. The projection of these points on the horizontal axis and the cumulative histogram for these points are shown in Figure 1 with solid black triangles and a solid black line, respectively. The SAC/FEMA denition for estimating EDPc is approximate and was devised for practicality. As the ground motion IM is increased for a specic ground motion, the storey in which the maxIDR occurs likely migrates from the upper stories to the lower stories of the structure (Medina and Krawinkler, 2003). This translates into EDPc being a function of IM despite the assumption we made here. If one wants to exercise the EDP-based approach for estimating the collapse fragility curve without this approximation, one needs to rewrite the probabilistic representation of EDPc on the right-hand side of Equation (1) as a function of IM and the storey number at which the demand/capacity format is exercised. Using the aforementioned cumulative histograms for demand and capacity in the EDP-based approach, Equation (1) was evaluated numerically to nd the probability of collapse given IM at various IM levels. The resulting data points are plotted with dark solid grey (red coloured) diamonds on the right-hand side of Figure 1. The collapse fragility curve, using the EDP-based approach, is obtained by tting a log-normal distribution to the probability of collapse given IM data points. 2.2 IM-based approach for estimation of collapse fragility curve

The IM-based approach, in which the ground motion IM is directly used to estimate the probability of collapse of the building, was rst introduced in Ibarra et al. (2002). In this approach we dene a random variable denoted as collapse capacity (denoted as IMC) as the ground motion intensity at which the building experiences dynamic instability. In this study we select the spectral acceleration at the rst mode period of the building, Sa, as the IM, but in general the IM can be any representative single value or vector ground motion IM. IDA is employed to estimate the collapse capacity for a given ground motion. A suit of ground motions that are representative of the seismicity at the location of the building are used to obtain realizations of the collapse capacity of the building. Using these realizations, we obtain the collapse fragility as the probability distribution function associated with collapse capacity. Given a value of IM, this probability distribution function shows the probability of IMC being less than IM. In other words, the collapse fragility curve shows the probability of collapse given the value of ground motion intensity. Equation (2) describes this statement in mathematical notation. P [ C IM = imi ] = P [ IMC < IM = imi ] (2)

The process for obtaining the collapse capacity of the eight-storey moment-resisting frame subjected to the set of 40 LMSR-N records is illustrated in Figure 2. In this gure, each grey line shows an individual IMEDP relation (i.e., Sa/gmaxIDR). The solid black circles at the end of each IDA curve shows the projection on the vertical axis of the last point on the IDA curve at which the nonlinear response history analysis has converged (i.e., dynamic stability is attained) within a certain tolerance. Instability is due to P- effects in one or a combination of stories, which may be greatly accelerated by strength and stiffness deterioration in structural components. The collapse capacity of each ground motion is the IM ordinate of each solid black circle. Since Sa is used as the IM, we denote the collapse capacity with SaC. Using the set of 40 SaC values, we can estimate the probability of collapse at IM = SaCi from the ratio of the number of SaC data points that are smaller than SaCi to the total number of SaC values (in this case 40). P[C|IM] data points are plotted with solid grey (red coloured) circles on the right hand side of Figure 2. Ibarra and Krawinkler (2005) have shown that a log-normal distribution can be used to represent SaC, i.e., Ln(Sac) N(C,RC). In this log-normal distribution C is the median of the
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

172
IDA Curves

F. ZAREIAN ET AL.

Cap. Data Points P[C|IM]

IM[Sa/g]

0 0 0.05 0.1

0.0

1.0 0.15

P(C|IM)

EDP[max.IDR]

Figure 2. Obtaining collapse capacity data points using IM-based approach by utilizing incremental dynamic analysis

Probability of Collapse

0.75

P(C|IM) Data Points


0.5

EDP-Based IM-Based FC(IM) Distribution EDP-Based IM-Based

0.25

0 0 1 2 3 4

IM[Sa/g]
Figure 3. Collapse fragility curves obtained by tting lognormal distributions to the data points obtained using the IM-based and EDP-based approaches

collapse capacity and RC is the dispersion due to record-to-record variability (Randomness in collapse Capacity). The collapse fragility curves for the eight-storey frame structure, obtained from both the EDP-based and IM-based approaches, are shown in Figure 3. It is seen that the fragility curves are comparable but that the probability of collapse given IM is larger if the EDP-based approach is used. This is due to the method used to dene and incorporate EDPc in the demand/capacity format of the EDP-based approach. In every IDA curve the structure has some capacity left after it passes the collapse criterion of 20% of the median of initial slope of all IDA curves, which results in a consistent underestimation of EDPc. Furthermore, the fact that the EDPc is a function of IM adds another approximation to the EDP-based method.
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 173

2.3

Variability in estimation of collapse fragility curve

In the previous section we have assumed that the variability in estimation of IMc in the IM-based approach, and in estimation of EDPd and EDPc in the EDP-based approach, is due to the random nature of the ground motion. The effect of this variability (denoted as aleatory variability) is incorporated in the dispersion of the collapse fragility curve that is estimated through these approaches. However, the mathematical models and tools for obtaining the collapse fragility curve along with the variables used in these mathematical models are uncertain due to our lack of knowledge (epistemic variability) about the true structural system. These models can only provide estimates of the collapse fragility curve of the building. A brief review of up-to-date analytical tools for collapse prediction was presented in the introduction of this paper. Furthermore, the uncertainty in the construction phase and human errors are additional sources of variability that were not included in the process for estimation of collapse fragility curve discussed in the previous sections. Simultaneous consideration of effects of both types of variability in the estimation of the collapse fragility curve can be attempted by an elaborate Monte Carlo simulationpresuming that probabilistic representations of all uncertain variables can be acquired. Approximate methods for combining the effects of the two types of variability on the estimation of collapse fragility assume that the effects of aleatory and epistemic sources are independent. To apply this assumption, a model of the building with component properties set to their median values, and a set of representative ground motion records are used to develop a median estimate of the collapse fragility curve. The dispersion in the median estimate fragility curve, RC, is assumed to be solely aleatory. Then, it is assumed that the median of the median estimate collapse fragility curve is by itself a random variable with a log-normal distribuC and whose dispersion, UC, is solely epistemic and independent of the aleatory tion whose median is dispersion, RC. For each source of variability (e.g., buildings period, base shear yield strength, cyclic deterioration parameters, etc.; and indexed with variable i) the associated value for UC (denoted as UC,i) can be estimated by nding the variability in the collapse capacity using a model of the building whose properties are set to their median values except for the parameter whose effect on epistemic variation of probability of collapse is investigated. The total UC can be estimated by using the Square Root of Sum of Squares method for combining values of UC,i. Based on a study on MDOF systems and using the First-Order Second-Moment reliability method, Ibarra and Krawinkler (2005) suggest that a reasonable estimate forUC is on the order of 04. Two approximate methods for combining the effects of the two types of variability in estimating of probability of collapse given IM of a building can be utilized: condence level method and mean method (Zareian and Krawinkler, 2007). In the condence level method we are interested to nd the collapse fragility curve with a certain condence level, say Y, which is the probability that the true C. The collapse fragility curve median value (denoted as Y C) of collapse capacity is greater than associated with Y condence level is a log-normal distribution whose median is equal to Y C and dispersion is equal to RC. In the mean method for combining the effect of epistemic and aleatory variability in estimating the probability of collapse it is assumed that the effect of two sources of variability can be concentrated in the dispersion of the collapse fragility curve (denoted as TC). Therefore, in the mean method the collapse fragility curve is a log-normal distribution whose median is C and dispersion is equal to TC = RC 2 + UC 2 . The reader is referred to Zareian and equal to Krawinkler (2007) for a detailed discussion of this approach. This approach is utilized also in the collapse assessment procedure described in FEMAP695 (FEMA, 2009).

Copyright 2009 John Wiley & Sons, Ltd.

Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

174 3.

F. ZAREIAN ET AL.

SEISMIC HAZARD CURVE

The seismic hazard curve provides the link between the collapse fragility curve and the seismic hazard associated with the ground motion IM considered in the collapse fragility curve. By denition, a seismic hazard curve is the return period description of the ground motion intensity and is obtained through Probabilistic Seismic Hazard Analysis. Figure 4 shows the data points representing the seismic hazard curve for the location of the eight-storey moment-resisting frame case study from the USGS seismic hazard calculation toolkit (http://earthquake.usgs.gov/research/hazmaps/design/). The lines in Figure 4 show the lines tted to seismic hazard data points in the log-log domain (i.e., curves of the form k0Sak in arithmetic domain). The grey (purple coloured) line is obtained by using all data points and the dark grey (red coloured) line is obtained by only using the two points in the vicinity of 00004 hazard level (i.e., 2% in 50 years probability of exceedance). This closed-form equation for the seismic hazard curve is used in subsequent sections for obtaining a closed-form equation for calculating a representation of the buildings collapse potential. For a specic building location, various seismic hazard curves can be obtained by utilizing different models for the ground motion attenuation relationship. This variability in estimation of seismic hazard is of epistemic nature. Cornell et al. (2002) suggested that by using the mean hazard curve, Sa(Sa) one can assume that the epistemic variability in estimation of seismic hazard is denoted as incorporated. The ground IM used in this study to address the seismic hazard for a building is a scalar parameter, Sa. Recently, Baker and Cornell (2005) shown that a vector-valued ground motion IM in the form of (Sa, ) can substantially reduce the variability in estimation of seismic response. In this vector, is a measure of the difference between Sa of a specic ground motion and the median Sa predicted by the attenuation equation for a specied hazard level. Furthermore, it is shown by Zareian and Krawinkler (2007), Haselton (2006) and FEMAP695 (FEMA, 2009) that the median of the collapse fragility curve obtained using the vector-valued ground motion IM (Sa, ) can be clearly different from the median obtained using the scalar Sa.

0.1

Data Points from USGS Fitted ko(Sa)-k using all datapoints: 0.00021(Sa)-2.721

0.01

(Sa)

Fitted ko(Sa)-k using datapoints in the vicinity of (Sa) = 0.0004: 0.00024(Sa)-3.544

0.001

0.0001

0.00001 0 0.5 1 1.5 2

Sa (g)
Figure 4. Seismic hazard curve for the location and period (T1 = 12 s.) of the eight-storey moment-resisting frame case study
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 175

4.

COLLAPSE PERFORMANCE MEASURES

We suggest two performance measures for assessment of collapse potential of a building: a tolerable probability of collapse at a discrete hazard level, and a tolerable MAF of collapse. Both randomness (aleatory variability) and uncertainty (epistemic variability) should be incorporated in estimating the probability of collapse. As a rst step, we assume in the next section that the only source of variability in estimating the collapse potential of the building is due to randomness in ground motion records. The epistemic variability in estimating the collapse fragility curve and the seismic hazard curve of the building can be incorporated in the collapse performance criteria in two ways. One is by implementing condence statements on the performance objective statements described above. By doing so, we are using the condence level method for combining the effects of the two types of variability in estimating of probability of collapse given IM. The revised performance criteria that incorporate both aleatory and epistemic variability in estimation of seismic hazard and probability of collapse given the level of ground motion intensity will read as: tolerable probability of collapse at a discrete hazard level with a certain condence, and a tolerable MAF of collapse with a certain condence. Alternatively, the epistemic variability in estimating the collapse fragility curve and the seismic hazard curve of the building can be incorporated in the collapse performance measure without any change in the statements of the performance objectives described before. In this way, we are using the mean method for combining the effects of the two types of variability in estimating of probability of collapse given IM. 4.1 Collapse probability without consideration of the effects of epistemic variability

The collapse fragility curve disregarding the effects of epistemic variability along with the buildings seismic hazard curve is used here to obtain the probability of collapse at a given hazard level. For given hazard level, say PR, one can obtain the associated spectral acceleration, SaPR, using the seismic hazard curve, and read the probability of collapse from the buildings collapse fragility curve at the ground motion intensity level SaPR. This process is explained in Equation (3) in mathematical notation where () is the standard normal distribution function. Ln( Sa PR ) Ln ( C ) P ( C Sa PR ) = RC (3)

For instance, for an IM = Sa = 087 g, which is representative of a ground motion intensity with a 2475-year return period for the eight-storey structure in Los Angeles (T1 = 12 s.), one nds the probability of collapse equal to 11% and 4% using the collapse fragility curves developed using the EDPbased and IM-based approaches, respectively. The reason for the difference originates from the judgmental denition for estimating EDPc in the EDP-based approach. The MAF of collapse is obtained by integrating the collapse fragility curve of the building over the seismic hazard curve as shown in Equation (4).

C =

all IM

P ( C IM ) d Sa ( IM )

(4)

Cornell (1996) and Jalayer (2003) obtained a closed-form solution for estimating the MAF of collapse, Equation (5), by assuming that the seismic hazard curve is linear in the log-log domain with a slope equal to k, and that the collapse fragility curve has a log-normal distribution with median of c and dispersion of RC.
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

176

F. ZAREIAN ET AL.

Table 1. Comparison between mean annual frequency of collapse obtained from different methods for obtaining the collapse fragility curve and integration of the fragility curve over the hazard curve Fragility curve obtained using EDP-based approach Numerical integration Equation (4) Closed-form solution Equation (5) 149 105 145 105 Fragility curve obtained using IM-based approach 705 105 836 105

1 C = [ Sa ( C )] exp k 2 RC 2 2

(5)

Table 1 shows the values of the MAF of collapse obtained by numerically integrating Equation (4), and by solving the closed-form solution of Equation (5). The collapse fragility curves obtained using the EDP-based and IM-based methods for the case study eight-storey moment-resisting frame are used in the calculations. As seen in this table, using the EDP-based approach overestimates the MAF collapse compared to using the IM-based approach for estimating the buildings collapse fragility curve. The relatively small difference between the results due to difference in utilizing Equation (4) or Equation (5) is interpreted as the relative accuracy of the closed-form solution developed for exercising Equation (4) by Equation (5). 4.2 Collapse probability including the effects of epistemic variability

In this section we describe two methods (condence level and mean methods) for combining the effect of epistemic variability in the probability of collapse. In the condence level method the probability of collapse at hazard level PR with Y condence can be obtained using Equation (6), where Y C is the median of collapse capacity with Y condence, and KY is the standardized Gaussian variate associated with the probability Y of not being exceeded (e.g., KY = 10 for 84% condence level and KY = 128 for 90% condence level).
Y Ln ( Sa PR ) Ln ( C ) PY ( C Sa PR ) = RC Y C e UC KY C =

(6) (7)

The MAF of collapse with Y condence is obtained by integrating the collapse fragility curve associated with Y condence over the mean hazard curve as shown in Equation (8). A close-form solution to Equation (8) has been proposed by Jalayer (2003) as shown in Equation (9).
Y C = all IM

PY ( C IM ) d Sa ( IM )

(8)
( kUC )]]

1 Y C )] exp k 2 RC 2 C = [ Sa ( 2

) exp K
[ [

(9)

Using the mean method the probability of collapse at hazard level PR is obtained by utilizing the log-normal distribution that incorporates the effects of both sources of variability as shown in Equation
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 177

(10). The MAF of collapse is computed by integrating this collapse fragility curve over the mean hazard curve as shown in Equation (11) whose close form solution is shown in Equation (12). C ) Ln ( Sa PR ) Ln ( P ( C Sa PR ) = TC (10) (11)

C =

all IM

P ( C IM ) d Sa ( IM )

1 C )] exp k 2 TC 2 C = [ Sa ( 2

(12)

5.

SAC PROCEDURE FOR ESTIMATING THE COLLAPSE PERFORMANCE OF BUILDINGS

This section is concerned with a comparison between the SAC method for estimating the collapse performance of a building and the method elaborated in the previous section. The SAC/FEMA recommendations were developed for buildings whose lateral load-resisting system consists of steel momentresisting frames. The differences between the two methods are in the denition of collapse performance objectives and the process for its estimation. The reader is referred to a wealth of documentation about the SAC/FEMA recommendations and guidelines for design and evaluation for collapse safety in FEMA-355F (FEMA, 2000), Cornell et al. (2002), Yun et al. (2002) and Hamburger et al. (2003). The discussion presented in this paper is intended to show the similarities and differences between the two approaches. The SAC/FEMA guidelines recommend that the probability of collapse for the building due to seismic action be less than 2% in a 50-year period with a recommendation that 95% condence for this objective be adopted. This objective translates into limiting the MAF of collapse to 00004 (with 95% condence) assuming a Poisson process for collapse occurrence (i.e., 2% = 1 e0000450). The SAC/FEMA guidelines do not include an objective that addresses collapse at a specic hazard level. This is in line with the SAC/FEMA design philosophy for collapse prevention in which an overall performance of the building at various hazard levels is of interest. The framework devised by the SAC/FEMA guidelines for estimation of MAF of collapse is founded on the EDP-based demand/capacity concepts. In this framework, MAF of collapse is obtained by combining probabilistic representations of: IM, EDPd|IM, and EDPc. It is assumed that EDPd|IM is Pd = a(Sa)b and dispersion of dR (which is assumed to not log-normally distributed with a median ED be a function of Sa). Values of a, b and dR are obtained by tting the functional form a(Sa)b to IMEDP data points obtained through IDAs. It is assumed that EDPc is log-normally distributed with Pc and dispersion of cR. Furthermore, it is assumed that the mean hazard curve can be median of ED Sa (Sa) = k0Sak. Combining the three elements one nds Equation represented by the functional form (13) for estimating the MAF of collapse with Y condence (Jalayer, 2003). In this equation, SaED P is c Pc and is obtained using Equation (14). The parameters the spectral acceleration corresponding to ED dU and cU are epistemic variabilities in estimation of EDPd|IM and EDPc. Details of this framework is documented in Cornell et al. (2002).
2 2 1 k ( 2 + 2 ) exp K k ( 2 + 2 ) Y C = [ Sa ( SaEDP c )] exp dR cR Y dU cU 2 b2 b2

(13)

Copyright 2009 John Wiley & Sons, Ltd.

Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

178

F. ZAREIAN ET AL.

SaEDP c

cb EDP = a

(14)

The MAF of collapse using Equations (13) and (14) is equal to 435 105 for the eight-storey case study moment-resisting system without consideration of epistemic variability in estimation of demand and capacity. The process for obtaining this value is as follows: (a) Estimate k by tting the functional form k0Sak to the mean hazard curve. Figure 4 shows the seismic hazard data obtained for the location of the building (black diamonds in Figure 4). SAC/ FEMA guidelines recommend that k be estimated by tting the functional form k0Sak to the data in the vicinity of the 2% in 50 years hazard level (i.e.,Sa (Sa) = 404 105). Figure 4 shows this curve, Sa (Sa) = 000024Sa3544, with a dark grey line (red coloured). The slope of the tted seismic hazard curve is k = 3544. (b) Estimate a and b by tting the functional form a(Sa)b to the median IDA curve. Figure 5 shows IM-EDP curves for the eight-storey case study building used in this study. This gure is similar to Figure 1 but the vertical axis is plotted with a smaller scale. SAC/FEMA guidelines recommend that a and b are estimated by tting the functional form a(Sa)b to the median of IDA curves in the vicinity of Sa associated with Sa (Sa) = 404 105, that is Sa = 086 g. The dark grey (red coloured) line in Figure 5 shows the median IM-EDP curve obtained by connecting the median values of EDPd|IM values. The solid grey (purple coloured) circles show the values of EDPd|Sa Pd = a(Sa)b to the median IDA curve in the vicinity of = 086 g. By tting a curve of the form ED Sa = 086 g we obtain b = 0879 and a = 0024. Pc, cR, and dR. In Figure 5, the empty black triangles show the values of EDPc whose (c) Estimate ED projection on the horizontal axis is shown with solid black triangles. The median and dispersion of Pc & cR. SAC/FEMA guidelines the log-normal distribution tted to the latter are estimates of ED recommend that EDPd & dR are estimated by median and dispersion of the tted log-normal distribution to the values of EDPd for a spectral acceleration associated with Sa (Sa) = 404 105,

1.2

0.9 IM[Sa(T1)/g]
IDA Curves

0.6

Fd (EDPD|IM=imi) Demand Data Points Onset of collapse (0.2Se)

0.3

Cap. Data Points Fc (EDPc)

0 0 0.02 0.04 0.06 0.08 0.1 EDP[max.IDR]

Figure 5. Obtaining the required parameters for estimating the MAF of collapse of the case study eight-storey moment-resisting frame using the SAC/FEMA closed-form solution
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 179

that is Sa = 086 g. In Figure 5, the grey (purple coloured) solid circles shows the values of EDPd|Sa = 086 g for the case study moment-resisting frame used in this study. The solid grey (purple coloured) line shows the tted log-normal distributions to EDPd|Sa = 086 g. Using the tted log Pd = 0021 & cR = 028. normal distribution we obtain ED Pc in (d) Estimate MAF of collapse using Equations (13) and (14). By substituting a, b, and ED P = 235 g. Using the tted seismic hazard curve we obtain Equation (14), we obtain SaED c 3544 P ) = 000024 235 Sa(SaED = 116 105. By substituting the calculated information for c P ), k, b, cR, cR, into Equation (13) and ignoring the effect of epistemic variability in Sa(SaED c estimation of demand and capacity (dropping the third bracket in Equation 10) one obtains the MAF of collapse for the case study building is equal to 435 105. The estimate of the MAF of collapse obtained using the SAC/FEMA guidelines for the case study eight-storey moment-resisting frame is almost a third of the MAF of collapse obtained by utilizing the EDP-based approach for estimating the collapse fragility curve, and half of the MAF of collapse obtained by utilizing the IM-based approach. (i.e., MAF of collapse for the former is equal to 145 105 and for the latter is 836 105 as shown in Table 1). The difference between the estimates is mainly due to the approximations in the SAC/FEMA approach inherent from using closed-form rep Pd, whose parameters a and b are not a function of IM, and dR, which is assumed resentations for ED to be constant regardless of the value of IM.

6.

SUMMARY

Two collapse performance criteria, in the form of tolerable probability of collapse given the ground motion hazard level and tolerable mean annual frequency of collapse, are discussed in this paper. Methods for quantifying the collapse performance are presented. Two approaches for estimating the probability of collapse given the level of ground motion intensity, denoted as EDP-based approach and IM-based approach, are discussed. The mean annual frequency of collapse for a building is obtained (numerically and in functional form) by integrating the buildings collapse fragility curve over the seismic hazard curve for the location of the building. It is shown that collapse performance is estimated with better accuracy and less upfront judgemental assumption if the IM-based approach is utilized for obtaining the collapse fragility curve of a building rather than the EDP-based approach. Using a case study eight-storey moment-resisting frame, it was shown that using the EDP-based approach for estimating the collapse fragility curve of a building can result in an overestimation of probability of collapse given the level of ground motion intensity and mean annual frequency of collapse by 50%. Two approximate methods for combining the effects of aleatory and epistemic sources of variability in estimation of collapse fragility curve for a building are introduced: condence level method and mean method. In both methods it is assumed that the effects of aleatory and epistemic sources of variability in estimation of collapse capacity of the building are independent. By using the condence level method condence statements are implemented on the performance objective statements. This condence statement relates to the condence level of the collapse fragility curve used to estimate the collapse performance of the building. Alternatively, by utilizing the mean method, the collapse fragility whose dispersion is inated to incorporate effect of both sources of variability in estimation of collapse capacity is used to estimate the collapse performance of the building. A comparison between the methodology for implementing the performance criteria presented in this paper and that adopted by the SAC/FEMA guidelines is presented. It is shown that the assumptions used in SAC/FEMA guidelines for the development of closed-form solution for estimating the mean annual frequency of collapse of a building can result in underestimation of this parameter by
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

180

F. ZAREIAN ET AL.

50%. The primary source of this variation is the approximation involved in the process of obtaining the closed-form solution proposed in the SAC/FEMA guidelines for estimating the MAF of collapse.

ACKNOWLEDGEMENTS

This research was supported by the NSF-sponsored Pacic Earthquake Engineering Research (PEER) Center and was carried out at Stanford Universitys John A. Blume Earthquake Engineering Center as part of a comprehensive effort to develop basic concepts for PBEE and supporting data on seismic demands and capacities. Any opinions, ndings and conclusions or recommendations expressed in this paper are those of the authors and do not necessarily reect the views of the sponsors.

REFERENCES

Adam C, Ibarra LF, Krawinkler H. 2004. Evaluation of P-Delta effects in non-deteriorating MDOF structures from equivalent SDOF systems. Proceedings of 13th World Conference on Earthquake Engineering, Vancouver, B.C., Canada, Paper ID. 3407. Baker JW, Cornell CA. 2005. Vector-valued ground motion intensity measure consisting of spectral acceleration and epsilon. Earthquake Engineering & Structural Dynamics 34(10): 11931217. Baker JW. 2007. Probabilistic structural response assessment using vector-valued intensity measures. Earthquake Engineering & Structural Dynamics 36(13): 18611883. Bernal D. 1987. Amplication factors for inelastic P- effects in earthquake analysis. Earthquake Engineering & Structural Dynamics 15(5): 635651. Cornell CA. 1996. Calculating building seismic performance reliability, a basis for multi-level design norms. Proceedings of the 11th World Conference on Earthquake Engineering, Paper No.2122. Cornell CA, Jalayer F, Hamburger RO, Foutch DA. 2002. Probabilistic basis for 2000 SAC/FEMA steel moment frame guidelines. Journal of Structural Engineering 128(4): 526533. Federal Emergency Management Agency (FEMA). 2000. State of the art report on performance prediction and evaluation of steel moment-frame buildings. Report No. FEMA-355F, Federal Emergency Management Agency, Washington, DC. Federal Emergency Management Agency (FEMA). 2009. Quantication of building seismic performance factors. Report No. FEMA P695, Federal Emergency Management Agency, Washington, DC. Hamburger RO, Foutch DA, Cornell CA. 2003. Translating research to practice: SAC/FEMA performance-based design procedures. Earthquake Spectra 119(2): 255267. Haselton CB. 2006. Assessing seismic collapse safety of modern reinforced concrete moment frame buildings. PhD Dissertation, Department of Civil and Environmental Engineering, Stanford University. Haselton CB, Liel AB, Dean BS, Chou JH, Deierlein GG. 2007. Seismic collapse safety and behavior of modern reinforced concrete moment frame buildings. In Proceedings of ASCE Structures Congress, Long Beach, California, 1619 May, 2007. Ibarra L, Medina R, Krawinkler H. 2002. Collapse assessment of deteriorating SDOF systems. Proceedings of the 12th European Conference on Earthquake Engineering, London, September 2002, Elsevier Science Ltd. Paper #665. 913 Sept. 2002. Ibarra LF, Krawinkler H. 2005. Global collapse of frame structures under seismic excitations. Report No. PEER 2005/06, Pacic Earthquake Engineering Research Center, University of California at Berkeley, Berkeley, California. Ibarra L.F, Medina RA, Krawinkler H. 2005. Hysteretic models that incorporate strength and stiffness deterioration. Earthquake Engineering & Structural Dynamics 34(12): 14891511. Jalayer F. 2003. Direct probabilistic seismic analysis: implementing non-linear dynamic assessments. PhD dissertation, Stanford University. Khandelwal K, El-Tawil S. 2007. Collapse behavior of steel special moment resisting frame connections. ASCE Journal of Structural Engineering 133(5): 646655. Medina RA, Krawinkler H. 2003. Seismic demands for nondeteriorating frame structures and their dependence on ground motions. Report No. 144, John A. Blume Earthquake Engineering Center, Department of Civil Engineering, Stanford University, Stanford, CA.
Copyright 2009 John Wiley & Sons, Ltd. Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

BASIC CONCEPTS AND PERFORMANCE MEASURES IN PREDICTION OF COLLAPSE OF BUILDINGS 181

Meguro K, Tagel-Din H. 2001. Applied element simulation of RC structures under cyclic loading. ASCE Journal of Structural Engineering 127(11): 12951305. Mosalam KM, Talaat M, Park S. 2008. Modeling progressive collapse in reinforced concrete frame structures. Proceedings of the 14th World Conference on Earthquake engineering (14WCEE), Paper ID S15018. Prakash V, Powell G, Campbell S. 1993. DRAIN-2DX: basic program description and user guide. Report No. UCB/SEMM-93/17, University California at Berkeley, Berkeley, CA. Rahnama M, Krawinkler H. 1993. Effects of soft soil and hysteresis model on seismic demands. Report No. 108, John A. Blume Earthquake Engineering Center, Department of Civil Engineering, Stanford University, Stanford, CA. Sasani M, Kropelnicki J. 2008. Progressive collapse analysis of an RC structure. The Structural Design of Tall and Special Buildings 17(4): 757771. Sivaselvan MV, Reinhorn AM. 2000. Hysteretic models for deteriorating inelastic structures, ASCE/Journal of Engineering Mechanics 126(6): 633640. Sivaselvan MV, Reinhorn AM. 2002. Collapse analysis: large inelastic deformations analysis of planar frames. ASCE/Journal of Structural Engineering 128(12): 15751583. Sivaselvan MV, Reinhorn AM. 2006. Lagrangian approach to structural collapse simulation. ASCE/Journal of Engineering Mechanics 132(8): 795805. Sivaselvan MV, Lavan O, Dargush GF, Kurino H, Hyodo Y, Fukuda R, Sato K, Apostolakis G, Reinhorn AM. 2009. Numerical collapse simulation of large-scale structural systems using an optimized-based algorithm. Special Issue of Journal of Earthquake Engineering & Structural Dynamics 38(5): 655677. Song JK, Pincheira JA. 2000. Spectral displacement demands of stiffness-and strength-degrading systems. Earthquake Spectra 16(4): 817851. Sun L, Zhou C, Qin D, Fan L. 2003. Application of extended distinct element method with lattice model to collapse analysis of RC bridges. Earthquake Engineering & Structural Dynamics 32(8): 12171236. Takizawa H, Jennings P. 1980. Collapse of a model for ductile reinforced concrete frames under extreme earthquake motions. Earthquake Engineering & Structural Dynamics 8(2): 117144. Vamvatsikos D, Cornell CA. 2002. Incremental dynamic analysis. Earthquake Engineering & Structural Dynamics 31(3): 491514. Yun S, Hamburger RO, Cornell CA, Foutch DA. 2002. Seismic performance evaluation for steel moment frame. Journal of Structural Engineering 128(4): 534545. Zareian F, Krawinkler H. 2007. Assessment of probability of collapse and design for collapse safety. Earthquake Engineering & Structural Dynamics 36(13): 19011944.

Copyright 2009 John Wiley & Sons, Ltd.

Struct. Design Tall Spec. Build. 19, 167181 (2010) DOI: 10.1002/tal

Vous aimerez peut-être aussi