Sunteți pe pagina 1din 19

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS Earthquake Engng Struct. Dyn.

2006; 35:115133 Published online 17 October 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.536

Decision support for conceptual performance-based design


Helmut Krawinkler1;; , Farzin Zareian1 , Ricardo A. Medina2 and Luis F. Ibarra3
1 Department 2 Department

of Civil and Environmental Engineering; Stanford University; Palo Alto; CA 94305; U.S.A. of Civil and Environmental Engineering; University of Maryland; College Park; MD 20742; U.S.A. 3 CNWRA; Southwest Research Institute; San Antonio; TX 78238; U.S.A.

SUMMARY Performance assessment implies that the structural, non-structural, and content systems are given and that decision variables, DVs, (e.g. expected annual loss, mean annual frequency of collapse) are computed and compared to specied performance targets. Performance-based design (PBD) is di erent by virtue of the fact that the building and its components and systems rst have to be created. Good designs are based on concepts that incorporate performance targets up front in the conceptual design process, so that subsequent performance assessment becomes more of a verication process of an e cient design rather than a design improvement process that may require radical changes of the initial design concept. In short, the design approach could consist of (a) specifying performance targets (e.g. tolerable probability of collapse, acceptable dollar losses) and associated seismic hazards, and (b) deriving engineering parameters for system selection, or perhaps better, using the relatively simple design decision support tools discussed in this paper. Copyright ? 2005 John Wiley & Sons, Ltd.
KEY WORDS:

performance-based design; conceptual design; performance assessment; collapse; losses; engineering demand parameters

1. INTRODUCTION Performance assessment, as developed in recent Pacic Earthquake Engineering Research (PEER) Center studies, implies that for a given system so-called decision variables, DVs, (dollar loss, length of downtime, or number of casualties) are determined whose values should fulll specied performance targets [14]. For instance, for life safety=collapse performance, the process of determining DVs is as summarized here: intensity measures, IMs, (e.g. spectral acceleration at the rst mode period of the structure, Sa (T1 )), are determined from seismic
Correspondence E-mail:

to: Helmut Krawinkler, Department of Civil and Environmental Engineering, Stanford University, Stanford, CA 94305-4020, U.S.A. krawinkler@stanford.edu

Contract=grant sponsor: Pacic Earthquake Engineering Research Center

Copyright ? 2005 John Wiley & Sons, Ltd.

Received 17 February 2005 Revised 13 May 2005 Accepted 13 May 2005

116

H. KRAWINKLER ET AL.

hazard analysis; relevant engineering demand parameters, EDPs, (e.g. storey drifts) are predicted from structural analysis for given values of IMs (and representative ground motions); local collapse fragility curves (e.g. for oor slabs that may drop because of shear failure at column-to-slab connection) and global collapse fragility curves of the type shown in Figure 6 are developed to predict local and global collapse probabilities; and as a last (and not yet resolved) step, predictions are made of the DVs, i.e. the number of lives lost and the number of injuries. A similar process is followed for the assessment of direct ($)loss, with collapse fragility curves replaced by fragility curves for specic damage measures (DMs) and associated loss functions. The mathematical formulation for evaluating decision variables and providing decision support to the owner=user, considering uncertainties inherent in all parts of the process, is provided by the PEER framework equation expressed as follows: (DV) = G DV|DM d G DM|EDP d G EDP|IM d (IM) (1)

where (DV) is a desired realization of the DV, such as its mean annual frequency of exceedance, and the G s represent complementary cumulative distribution functions. Design is di erent from performance assessment, by virtue of the fact that the building and its structural components and system rst have to be created. One can view design as an iterative assessment process that starts with a judgmental conceptual design for which performance assessment is carried out, and the design is improved (tuned) in successive iterations until the performance targets are met. This design process is an option, but not a very attractive one. A poor initial conceptual design may be tuned to an extent that it fullls the performance targets, but it likely will never become a good design. Good designs are based on concepts that incorporate performance targets up front in the design decision process, so that subsequent performance assessment becomes more of a verication process of an e cient design rather than a design improvement process that may require radical changes. Conceptual design is greatly facilitated by focusing on discrete performance targets associated with discrete hazard levelssimilar to the way it is being practiced in most of the performance-based guidelines presently in use. In the conceptual design phase, engineers are used (and likely will be so for many years to come) to select and rough-proportion structural systems for strength, sti ness (drift limitations), ductility, and perhaps energy dissipation capacity and oor accelerations. The art of engineering, which should be practiced in this phase, is to use global information on important performance targets in order to come up with a structural system that fullls specied performance objectives in the most e ective manner. This implies exploration of design alternatives, which may be utilizing di erent structural materials and systems or advanced technologies such as base isolation or internal energy dissipation devices. The challenge is to provide the designer with a small set of most relevant criteria for important EDPs on which good conceptual design can be based. In concept, this means reversing the information ow discussed before for performance assessment, and working towards quantication of limits for relevant EDPs, given that desired performance can be expressed in terms of targeted DV values at discrete performance levels. These relevant EDPs, such as storey drifts, oor accelerations (or velocities), and storey (or component) ductility
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

117

demands relate to structural parameters (strength, sti ness, ductility, etc.) that drive design decisions on structural materials and systems. Given EDP limits and associated IM hazards for various performance levels, conceptual design for multiple performance objectives can be carried out. In general, performance should be concerned with structural and non-structural systems as well as building contents. There is no single design parameter that will control all performance goals at all performance levels. For instance, damage to non-structural components is controlled often by interstorey drift limitations, which demand large sti ness. Damage to building contents, on the other hand, is mostly proportional to oor accelerations, which can be limited by reducing the sti ness and=or strength of the structure. At the other extreme, life safety and collapse prevention are controlled by the inelastic deformation and energy dissipation capacities of ductile elements and the strength capacity of brittle ones. This discussion indicates that di erent performance objectives may impose con icting demands on strength and sti ness, and that seismic design is likely to become an iterative process in which di erent performance criteria may lead to trade-o s between strength and sti ness requirements, but in which no compromise should be made on issues of life safety and collapse prevention. This iterative process can be accomplished in two phases; a conceptual design phase in which various e ective structural systems are explored and roughsized, and a performance assessment phase in which performance of the structural, nonstructural, and content systems is evaluated and nal design decisions and modications are madewith due consideration given to all important sources of uncertainty. This paper is concerned with the conceptual design phase. Two challenges need to be addressed in the context of performance-based conceptual design. One is to develop data on EDP limits associated with specic performance targets. Once such EDP limits have been established, together with IMs that represent discrete hazard levels for which the EDP limits apply, the challenge is to create structural systems that e ciently accommodate these EDP limits. This paper is concerned with these two challenges.

2. DOMAINS THAT CONTROL PERFORMANCE-BASED DESIGN From a macro-perspective, PBD comprises three domains; the hazard domain, the structural system domain, and the loss domain, as illustrated in Figure 1. The hazard and loss domains provide design constraints, and the structural system domain contains all design alternatives worthy of exploring. In performance assessment, all three domains contain random variables to be described by a central value and a measure of dispersion to account for epistemic and aleatory uncertainties. In conceptual PBD the objective is to search for e ective solutions at a time at which the details of the structural system are yet to be determined. Therefore, it appears to be fair to focus on expected (mean) values of all random variables, and delay the uncertainty evaluation and propagation till the performance assessment phase. This facilitates up-front design decision making by permitting a focus on the most important global behaviour aspects without having to deal with mathematical formulations that are important but obscure behaviour-based decision making. In the following discussion the content of the three domains, seen from the perspective of conceptual PBD, is summarized.
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

118

H. KRAWINKLER ET AL.

Hazard Domain
Mean Hazard Curve(s) for Design Alternatives
(IM)

Structural System Domain


No Collapse Collapse

Mean IM-EDP Curves for Design Alternatives


E ( EDP | IM & NC ) |

Collapse Fragility Curves for Design Alternatives P ( C | IM )

Loss Domain
No Collapse Collapse

Mean Loss Curves


E ( Loss | EDP & NC ) |
-Direct Loss =

Mean Loss due to Collapse

E ( Loss || C ) Los
Loss -Life Loss (Death) = D Loss

$Loss

-Down Time Loss =

Figure 1. Domains controlling performance-based design.

2.1. Hazard domain The hazard domain contains the return period dependent description of the ground motion intensity plus associated time-history records. The intensity measure could be a scalar or a vector quantity [5], with the latter being of particular importance in the case of near-fault ground motions. In the numerical example presented later the elastic spectral acceleration at the rst mode period of the structural system (Sa (T1 )) is used as IM. For a scalar IM the description suitable in the context of conceptual PBD is the mean hazard curve (IM). It is important to note that the mean hazard curve changes with the rst mode period of the explored structural system. The records selected to represent the seismic input for specic IM values a ect the structural demand curves (mean IM-EDP curves) contained in the structural system domain discussed in Section 2.3. The associated issues of ground motion scaling and near-fault e ects have been and still are the subject of research [58] and are not discussed further in this paper. 2.2. Loss domain It is widely accepted that earthquake design decisions should be basedcolloquiallyon the three Ds, i.e. dollars, downtime, and deaths. Unfortunately, there are many ambiguities in this simple phrase. For one, we have not yet succeeded in quantifying deaths (and maybe we do not want to quantify it). Thus, in many cases collapse is used as a surrogate for deaths or casualties (life safety). Furthermore, we still have only vague insight into downtime and the quantication of associated losses, which may a ect not only the owner but may have more global consequences [911]. Therefore, downtime is not discussed explicitly here, even though it can be represented by the same concepts illustrated for direct ($)loss. Up front it is necessary to divide losses into two sub-domains, one containing losses if no collapse occurs (NC domain), and the other containing losses if collapse occurs (C domain). Both sub-domains contribute to the three Ds, as illustrated conceptually in Figure 1.
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

119

Drywall Partitions with Metal Frame 1.0 0.8 P(DM> dm | EDP) 0.6 0.4 0.2 0.0 0.000
Tape, Paste & Paint Replacement gypsum boards but not frame Replacement of the whole partition

Drywall Partitions with Metal Frame


1.4 1.2 E[Loss |EDP] 1.0 0.8 0.6 0.4 0.2 0.0 0.000

0.005

0.010

0.015 EDP (IDR)

0.020

0.025

0.030

0.005

0.010

0.015 EDP (IDR)

0.020

0.025

0.030

(a)

(b)

Figure 2. Damage and loss in a drywall partition: (a) damage state fragility curves [12]; and (b) expected ($)loss-EDP curve (modied from Figure 5.9 in Reference [13]).

2.2.1. Direct ($)loss due to damage (no collapse). Losses may occur in any of the subsystems of a building, and in general need to be aggregated over components and subsystems. A building could be divided into subsystems according to functional use (e.g. structural, non-structural, content) and according to the sensitivity of the subsystem components to engineering demand parameters, EDPs (e.g. interstorey drift, oor acceleration). The objective is to establish, for each subsystem, a relationship between loss (repair or replacement cost) and a most relevant engineering demand parameter (EDP), so that the latter can be used by the engineer to guide design decisions. The EDP has to be well correlated with losses in all components of the subsystem, and it has to be well correlated with global structural response in order to permit deduction of global design decisions. The process of computing loss-EDP relationships requires the following ingredients and steps: 1. The availability of a set of fragility curves for each component of the subsystem, which dene, as a function of the EDP, the probability of being in, or exceeding, specic damage states requiring specic repair actions. Figure 2(a) shows a typical set of fragility curves for drywall partitions with metal frames [12], using interstorey drift ratio (IDR) as an EDP. 2. The availability of cost-of-repair functions for each damage state. 3. The ability to integrate fragility and cost-of-repair functions for each component. In the context of conceptual design it appears quite acceptable to use expected cost of repair for each damage state (E[Lj |DM = d mi ]) and compute only expected loss as a function of EDP, i.e. E[Lj | EDPj = edp] =
m i=1

E[Lj |DM = d mi ]P(DM = d mi |EDPj = edp)

(2)

The result is an expected (mean) loss-EDP relationship for the specic component, as shown in Figure 2(b). 4. The summation of expected losses (as a function of EDP) over all the components of the subsystem. The result will be a single mean loss-EDP curve for this subsystem (mean
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

120

H. KRAWINKLER ET AL.

loss curve in Figure 1, i.e. E(Loss|EDP&NC). The same process needs to be carried out for all other subsystems, which will provide information on the contribution of the individual subsystems to the total loss, given a value of EDP. This process is straightforward, in concept, but at this time it is most di cult or impossible to implement, simply because of the lack of data. Mean loss-EDP curves for most individual components and for subsystems are not available at this time. But work has started in the PEER Center on the development of mean loss curves, and it is hoped that the development of such curves will become a focus of future research. It is di cult to see how PBD can be implemented meaningfully without the availability of such curves. 2.2.2. Direct ($)loss due to collapse. Collapse also is a contributor to direct ($)loss. It may be a major contributor for non-conforming structures (e.g. older RC frame structures [14]). For conforming structures (i.e. structures designed according to modern standards) most of the direct ($)loss comes from damage in relatively moderate but frequent events, rather than from complete or partial collapse. Thus, for the purpose of ($)loss estimation it is assumed that collapse causes total monetary loss, and its impact on expected loss is estimated as this total loss times the probability of collapse. 2.3. Structural system domain The structural system domain contains information that relates, for selected design alternatives, engineering parameters with the hazard and loss domains. 2.3.1. No collapse sub-domain. As long as the structure does not collapse, the structural system domain contains mean IM-EDP curves, with IM being the intensity measure employed in the hazard domain, and EDP being an engineering demand parameter that correlates well with the loss in one of the subsystems dened in the loss domain. The EDP could be a storey parameter (if losses are evaluated on a storey-by-storey basis) or a global parameter such as the average of the maximum storey drifts (or of maximum oor accelerations) over the height of the structure. In the conceptual design phase, mean values of EDPs are appropriate measures. Mean IM-EDP curves (E(EDP|IM&NC)) are obtained by subjecting structural systems of specic properties to sets of ground motions representative of specic IM values. If it can be assumed that the frequency content of the ground motions is insensitive to magnitude and distance within the IM range of primary interest, then incremental dynamic analyses (IDAs) of the type illustrated in Figure 3 can be utilized to obtain mean IM-EDP curves. Such curves have been developed and stored in a database for many EDPs for a wide range of moment resisting frames [7], and additional curves are under development for reinforced concrete wall structures. In order to implement the proposed conceptual PBD process, it will be necessary to have available mean IM-EDP curves for the range of design alternatives to be evaluated. Development of such curves is a relatively straightforward process for regular structures, and a more elaborate but also more critical e ort for irregular structures. 2.3.2. Collapse sub-domain. This domain contains collapse fragility curves, which portray the probability of collapse as a function of the intensity measure (P(C|IM)). Such curves can be obtained by subjecting deteriorating structural systems of specic properties to sets of ground
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

121

Average Interstory Drift Ratio IDA curves


N=9, T1=0.9, =0.3, =0.05, Peak-Oriented model, BH, LMSR-N

1.5 Sa(T1)/g

Individual responses Mean

0.5

0 0 0.01 0.02 0.03 0.04 0.05 Average Storey Drift, IDRavg.

Figure 3. Mean IM-EDP curve for a nine-storey frame with T = 0:9 s obtained from incremental dynamic analyses (IDAs).

Fc Fy

F
Ke

Ks=sKe

Capping (Peak) Point Kc=cKe

Residual Strength

Fr = Fy y c r

Elastic Stiffness

Post-Capping Stiffness

Hardening Stiffness

Figure 4. Backbone curve for deteriorating component hysteresis models [16].

motions representative of the range of IMs in which collapse is expected. If all component deterioration modes are adequately presented in the analytical model, it should be feasible to analytically trace structures till collapse by incrementing the IM of the ground motion until a minute increment in IM leads to a very large increment in a global EDP, indicating dynamic instability. Research has been performed recently on the collapse capacity of moment resisting frames, utilizing component hysteresis models that account for strength deterioration in the backbone curve (see Figure 4) and for cyclic deterioration in strength and sti ness [15]. The collapse capacity is dened as that value of the relative intensity, dened here as [Sa (T1 )=g]= ( = base shear strength coe cient, i.e. the base shear strength of the system normalized by
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

122

H. KRAWINKLER ET AL.

MAX. STOREY DUCTILITY vs. NORM.STRENGTH


N=9, T1=0.9, =0.05, K1, S1, BH, =0.015, Peak-Oriented Model,

10 8 [Sa (T 1)/g ]/ 6 4 2

s=0.05, c /y=4, c=-0.10, s,c,k,a =inf, LMSR-N

Individual responses

0 0 10 20 Maximum Storey Ductility Over the Height, s,max 30

Figure 5. IDAs till collapse and distribution of collapse capacity [16].

[Sa(T1)/g]/ vs PROBABILITY OF COLLAPSE


N=9, T1=0.9, BH, Peak Oriented Model, LMSR-N, = 5%, s = 0.03, c /y =Var, c=Var, s,c,k,a =Inf, = 0

1 0.8 0.6
c/y=6, c=-0.10

Probability of Collapse

0.4 0.2 0 0 5 [Sa(T1)/g]/ 10

c/y=4, c=-0.10 c/y=2, c=-0.10 c/y=4, c=-0.30 c/y=2, c=-0.30

15

Figure 6. Collapse fragility curves for nine-storey frame structures with T1 = 0:9 s [16].

its seismic weight, Vy =W ) at which dynamic instability occurs due to deterioration and P e ects. It is noted that [Sa (T1 )=g]= is equivalent to the ductility dependent strength reduction factor R . The probability distribution function, PDF (assuming a lognormal distribution) of the collapse capacity is obtained as illustrated in Figure 5, and the corresponding cumulative distribution function, CDF, represents the collapse fragility curve, P(C|IM). Collapse fragility curves of the type shown in Figure 6 have been derived for regular frames subjected to a set
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

123

Hazard Domain
Mean Hazard Curve(s) for Design Alternatives

Structural System Domain


Mean IM-EDP Curves for Design Alternatives Collapse Fragility Curves for Design Alternatives

IM

IM

(IM)
E($Loss | EDP & NC)

EDP

P (C | IM)

ss C E ($L oss |C )
EDP
Mean $Loss Curve(s) (No Collapse) Mean $Loss Value (Collapse)

Loss Domain

Figure 7. Conceptual PBD for acceptable ($)loss at discrete hazard levels.

of 40 ground motions [16]. In Reference [16] it has been concluded that the collapse fragility depends primarily on the component ductility capacity c = y (which is assumed to be the same for all components in the structure), the post-capping sti ness ratio c (see Figure 4), and the cyclic deterioration parameter s;c;k;a . These parameters, together with the fundamental period T1 and the base shear strength parameter = Vy =W , control the design for collapse safety. 3. DECISION SUPPORT FOR CONCEPTUAL DESIGN In the up-front conceptual design process, decisions have to be made on the type of structural system and its global strength, sti ness, and ductility properties. The information contained in the three domains discussed before can be utilized to provide much needed support for this design decision process. 3.1. Design decisions derived from performance targets for direct ($)loss The conceptual process of making design decisions based on acceptable direct ($)loss is illustrated in Figure 7. One way of expressing desired performance is to specify acceptable losses at specic hazard levels, such as the 50% probability of exceedance in 50 years (50=50) hazard. As discussed previously, losses may occur in several subsystems that are sensitive to di erent EDPs. Losses in di erent subsystems could be assessed simultaneously, or the focus could be placed on the subsystem that contributes most to the value of the building, with other subsystems being evaluated subsequently. The latter approach is illustrated in Figure 7. The lower left portion of the gure illustrates the mean ($)loss-EDP curve for the dominant subsystem if no collapse occurs, E($Loss|EDP&NC), and the lower right portion shows
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

124

H. KRAWINKLER ET AL.

the expected loss if collapse occurs, E($Loss|C). The upper left portion shows the mean IM hazard curve for the specic site (usually the elastic spectral acceleration at the estimated rst mode period of the structure is used as IM). The upper central portion shows mean IM-EDP relationships for several design alternatives, and the upper right portion shows collapse fragility curves for the same design alternatives. The process is to enter the lower graph with a value of acceptable loss and obtain the associated EDP, and to enter the left upper portion with the hazard level at which the loss is acceptable and obtain the associated IM. The intersection of the IM value and the EDP value in the graph presenting mean IM-EDP curves for design alternatives can be viewed as a design target. All design alternatives that intersect the IM line to the left of the design target are feasible solutions, i.e. the associated expected losses are smaller than the target acceptable loss. The best solution will depend on many considerations, some of them being discussed in the example given in Section 4.1. The feasibility of a design alternative depends also on the potential ($)loss caused by collapse of the structure. If the IM line associated with the hazard level of interest is extended to the right, its intersection with the collapse fragility curves will provide insight into the collapse probability of the design alternatives, P(C |IM). The total expected loss can then be expressed as E($Loss|IM) = E($Loss|IM&NC) P(NC|IM) + E($Loss|C) P(C |IM) (3) with P(NC|IM) = 1 P(C |IM) being the probability that collapse will not occur. Since E($Loss|C) is very large, it is prudent, from an economic perspective, to eliminate design alternatives that have a noticeable probability of collapse at the hazard levels of primary interest for ($)loss control. The approach here illustrated can be extended, if so desired, to estimate expected annual loss rather than focus on a specic hazard level. For each design alternative the mean IM-EDP curves can be computed for the full range of hazards signicantly a ecting the direct ($)loss. Thus, the expected ($)loss can be evaluated for several discrete hazard levels, and the expected annual loss can be computed by numerical integration over the hazard curve, i.e. E[$Loss] =
IM

E[$Loss|im ]|d

IM (im)|

(4)

This is a rather powerful extension as it permits, for selected design alternatives, direct comparison of expected annual loss for the full range of hazards, rather than committing to specic hazard levels. This approach is more in line with the presently employed PEER performance assessment methodology, which considers decision variables (losses, etc.) as continuous variables rather than parameters associated with discrete hazard levels. This approach is particularly attractive for a building system that has competing subsystems, i.e. subsystems of signicant value but with di erent relevant EDPs, such as drift and acceleration sensitive subsystems. In this case an integrated loss measure appears to be much more objective than a measure associated with one specic hazard level. There are many questions to be answered, with a few of them addressed in general next, and a few more addressed specically in the example illustrated later. What is an appropriate EDP? The choice of the EDP is driven by the sensitivity of the loss to variation in a basic design parameter. For a subsystem that is sensitive to interstorey
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

125

drift, and if loss-based design is performed for the full building, the best choice is believed to be the average of the maximum drifts over the height, which is a measure of the expected maximum drift. It is larger than the global drift, because maximum storey drifts occur at di erent times [7]. If loss-based design is performed on a storey-by-storey basis, then the maximum interstorey drift for the storey under consideration is an appropriate EDP. What are the design alternatives? In concept, design alternatives incorporate all coste ective structural systems and materials. They also may include systems utilizing innovative technologies. The e ort is in developing mean IM-EDP relationships for all systems. The design variables could be base shear strength Vy , sti ness or strength variation over the height, or rst mode period (if T1 is varied, the hazard curve changes accordingly). For regular frame structures many such IM-EDP curves are documented in Reference [7], and more are under development. Similar curves are being developed for wall structures. Why bother with MDOF IM-EDP curves rather than use approximate SDOF representations? In the writers opinion the usefulness of SDOF representations in seismic design is overestimated. Their use invites approximations that are justied in some cases but not in others. Why not avoid their use if appropriate MDOF information is available? This, in fact, is a necessity when the EDP is oor acceleration, which is an EDP that correlates poorly with any SDOF parameter. 3.2. Design decisions derived from performance targets for life safety In most codes and guidelines, it is assumed that adequate collapse safety (and life safety) is provided by limiting the maximum storey drift at the design earthquake level to a specic value (e.g. a drift limit of 0.02 at the 10=50 hazard level). The drift at this hazard level is estimated from either an elastic analysis or an inelastic time history analysis. But the latter usually is executed with the use of component hysteresis models that do not account for strength and sti ness deterioration. Thus, these EDP predictions provide no insight into the probability of collapse. With the advent of deterioration models that do account for important aspects of deterioration it is becoming possible to trace the response of structures to collapse (e.g. [1518]) and to be specic about a collapse performance target. Such a target could be expressed as a tolerable probability of collapse (say, 10% at the 2=50 hazard level), or more general, as a tolerable mean annual frequency of collapse. Both options can be pursued in the proposed approach. But in both options it is understood that collapse itself is not the most relevant performance target. Of primary societal interest is to provide adequate life safety, which could be translated into a tolerable loss of life (or casualties). Figure 8 illustrates the PBD process for life safety, congruent with that for $loss, but with many curves dashed. The dashed curves imply that parts of the process still have to be developed. There is little doubt that casualties may occur even if the structure does not collapse (e.g. falling objects hazards), but the associated mean life loss curves and IM-EDP curves have yet to be developed. Equally, even if collapse occurs, the relationship between collapse (in all its various forms) and the expected loss of lives (or casualties) still has to be determined. Some information is available on this issue [12], but much more research needs to be done to establish such relationships. Thus, from the components of the process illustrated in Figure 8, and expressed by the following equation: E( D Loss|IM) = E( D Loss|IM&NC) P(NC|IM) + E( D Loss|C) P(C |IM)
Copyright ? 2005 John Wiley & Sons, Ltd.

(5)

Earthquake Engng Struct. Dyn. 2006; 35:115133

126

H. KRAWINKLER ET AL.

Hazard Domain
Mean Hazard Curve(s) for Design Alternatives

Structural System Domain


Mean IM-EDP Curves for Design Alternatives Collapse Fragility Curves for Design Alternatives

IM

IM

(IM)
E( D Loss | EDP & NC)

EDP

P (C | IM)

E ( D Loss | C ) |
EDP
Mean
D

Loss Curve(s) (No Collapse)

Mean

Loss (Collapse)

Loss Domain

Figure 8. Conceptual PBD for life safety at discrete hazard levels.

only the probability of collapse given IM, P(C |IM), can be evaluated with some condence [16]. Even in that context, much more research needs to be performed to assess various local collapse modes, partial collapses, and progressive collapse. Thus, the desirable process is as shown in Figure 8, but at this time it only appears to be feasible to express a life safety performance target in terms of a tolerable probability of collapse, or more general, as a tolerable mean annual frequency of collapse. For the former, the centre portion of Figure 8 is to be skipped, and the intersection of the line denoting the IM value at the specied hazard level with the line denoting the tolerable probability of collapse divides the design alternatives into a feasible and an unfeasible solution space. As an additional feature, the mean annual frequency of collapse can be computed by integrating the collapse fragility curve with the hazard curve, i.e.
c

=
IM

P(C |im)|d

IM (im)|

(6)

Both approaches are illustrated in the following PBD examples. 4. EXAMPLE IMPLEMENTATIONS OF DECISION SUPPORT FOR PBD The following examples attempt to illustrate the implementation of the concepts discussed so far. It soon will become evident that parts of the implementation are hypothetical, simply because some or much of the data needed for full implementation is missing. We hope that this does not discourage the potential utilization of the process, but to the contrary, encourages engineers and researchers to give the process some thought and a chance for implementation, which will lead to the creation of the needed information through research and development.
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

127

| |

Figure 9. Example of conceptual PBD for acceptable ($)loss at discrete hazard levels.

4.1. Example of design for acceptable direct ($)loss An example of design decisions based on acceptable direct ($)loss is illustrated in Figure 9. The example addresses a nine-storey o ce building, located in Southern California at a site for which the spectral acceleration hazard curves for the periods of 0.9 and 1.8 s are as shown in the upper left portion of the gure. It was decided to use reinforced concrete as
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

128

H. KRAWINKLER ET AL.

the primary structural material. Design alternatives comprise moment frame, wall, and dual wall-frame systems. Only moment frame alternatives are evaluated here because of space constraints in the gure (and because mean IM-EDP curves for wall and wall-frame systems have not yet fully matured). Again because of space constraints, only moment frames with T1 = 0:9 and 1.8 s and with a base shear coe cient = Vy =W = 0:1, 0.2, and 0.3 are considered as alternatives. It is assumed that for ($)loss estimation purposes the building can be divided into three subsystems, a structural subsystem (SS), a non-structural drift sensitive subsystem (NSDSS), and a non-structural and content subsystem that is sensitive to oor acceleration (NSASS). (The subsystem classication may vary with the use of the structure, e.g. there also may be a oor velocity sensitive subsystem.) It is assumed that the NSDSS and NSASS subsystems are known and can be quantied before structural design decisions have to be made, i.e. their value is essentially independent of the structural system (this is not a necessary assumption). The cost of the SS subsystem is design dependent, but it usually is a relatively small contributor to total investment (in a loss estimation study on a Californian hotel building the SS contributed less than 20% to the total investment [12, 14]). The mean loss-EDP curves for the three subsystems are as shown in the lower portion. The EDP for the NSDSS subsystem is the average of the maximum interstorey drifts over the height (IDR avg ), the EDP for the NSASS subsystem is the average of maximum oor accelerations over the height (FAavg ), and the EDP for the SS subsystem is the average of the maximum storey ductility ratios over the height ( avg ). (The implication is that the loss evaluation is done for the full building and not on a storey-by-storey basis.) At this time there is little hard data behind these loss-EDP curves; they are based on judgement. Noteworthy is the jump in the mean loss- avg curve for the SS subsystem from a relatively small value to the value of total loss (usually the replacement cost in present dollars). This jump occurs when the owner decides to demolish even though the structure has not collapsed. Jumps also may be present in other loss-EDP curves if large losses are associated with the attainment of specic EDP values. Because di erent EDPs control the loss in the three subsystems, Figure 9 shows three di erent sets of mean IM-EDP curves for the same design alternatives. These are shown in the centre portion of the upper half of the gure. Based on relative monetary value of the subsystems, in this example designing for acceptable ($)loss needs to focus on the NSDSS subsystem. An owner could target an acceptable loss in the NSDSS subsystem of about 0.5 million dollars at the 50=50 hazard level. Thus, design targets are created by entering the gure with the EDP associated with the target acceptable loss and with the 50=50 mean Sa hazard (in this example Sa (T1 ) is used as IM) for appropriate periods. For illustration, periods of 0.9 and 1:8 s are selected. The upper central portion of the gure contains mean Sa (T1 )-EDP curves for various design alternatives (obtained from statistical studies reported in Reference [7]). From the left of the three IM-EDP graphs it is evident that any solution with T1 = 1:8 s would cause losses in the NSDSS subsystem that by far exceed the targeted acceptable losses at the 50=50 hazard level. In fact, the three T1 = 0:9 s design solutions barely meet the loss target, i.e. solutions with T1 0:9 s are discarded. In the three presented T1 = 0:9 solutions the base shear strength coe cient = Vy =W is varied from 0.3 to 0.2 and 0.1. (The fact that the three solutions overlap for a sizeable range is a consequence of adherence to the equal displacement rule.) Only the = 0:3 solution is attractive because for solutions with smaller values the average storey ductility avg (third IM-EDP graph) becomes clearly larger than 1.0, indicating
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

129

signicant structural damage. (Moreover, for 0:3 the right-most collapse fragility curves indicate measurable collapse probabilities at relatively small Sa (T1 )=g values.) The merits of the T1 = 0:9 and = 0:3 solution can be assessed further by inspecting the expected NSDSS, NSASS, and SS losses at other hazard levels, such as the 10=50 and 2=50 levels, as illustrated in Figure 9. For the case illustrated, these losses appear to be acceptable. Moreover, the upper right-most graph shows that for the selected design alternative the probability of collapse is negligible unless the IM becomes very large (very long return period hazard). But the picture could change radically if the building were a museum (or a hospital), in which case the NSASS mean ($)loss-EDP curve likely will be much steeper and may be the dominant contributor to total loss. In this case, longer period structures may become more attractive (note that the hazard curve changes as the rst mode period of the structure becomes longer). The versatility of the proposed approach, which is best implemented in graphical form of the kind illustrated in Figure 9, lies in the ability to evaluate various design alternatives, considering simultaneously all important subsystems and various hazard levels of interest. The visualization aspect permits a behaviour assessment that should be of much value to engineers practicing this approach. The approach would clearly point out benecial trade-o s in strength and sti ness, based on the shape of loss curves for individual subsystems. It also provides preliminary insight into the collapse issue, although that one is usually controlled by collapse (life) safety issues discussed in the next section. The data provided in Figure 9 permit also the evaluation of the expected annual loss according to Equation (4). For the T1 = 0:9 s and = 0:3 and 0.2 design alternatives, the expected annual ($)loss amounts to $44 000, and $51 000, respectively. As an additional advantage, the proposed approach provides a tool for assessing objectively the costs and benets of various innovative technologies such as base isolation or internal energy dissipation devices. The costs of implementing the technology would show up in the SS costs, and the benets would show up in the reduction of losses. 4.2. Example of design for a tolerable probability of collapse Providing collapse safety implies adherence to capacity design concepts, and it implies design for ductility. The latter is implicitly considered in present design approaches with the judgmental response modication (R) factor or behaviour (q) factor. These factors are tied to component detailing (ductility) requirements, and in the design process they are used to reduce the strength design level to a fraction of the elastic demand associated with the spectral acceleration at the rst mode period. To this date it is not known whether or not this R (or q) based design process provides a quantiable, or for that matter even remotely consistent, factor of safety against collapse. Provided one can develop condence in the collapse fragility curves of the type illustrated in Figure 6, the process illustrated conceptually in Section 3.2 can be utilized to perform designs that target a specic tolerable probability of collapse. This is illustrated next on the same example for which design for ($)loss has been illustrated. 4.2.1. Design for tolerable probability of collapse at a specic hazard level. Desired performance at the collapse prevention level could be expressed in terms of a tolerable probability of collapse at a specied hazard level, as for instance, a tolerable probability of collapse of 0.1 at the 2=50 hazard level. For the T1 = 0:9 s and = 0:3 structure, the
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

130

H. KRAWINKLER ET AL.

Figure 10. Collapse spectra for frames with beam hinges: (a) for 10% probability of collapse; and (b) for 50% probability of collapse [16].

collapse fragility curve in the right upper portion of Figure 9 indicates that the actual probability of collapse is greater than 0.1, rendering this solution undesirable for collapse safety. One could plot collapse fragility curves for stronger or more ductile structures, or, perhaps better from the perspective of behaviour, take advantage of collapse spectra of the type shown in Figure 10 [16]. These spectra show the relative intensity [Sa (T1 )=g]= associated with a certain probability of collapse (10% in Figure 10(a), and 50% in Figure 10(b)) for frame structures with T1 = 0:1N (N = number of stories) and several combinations of system parameters (see Figure 4). The spectra illustrate the e ect of component ductility capacity ( c = y ) on this intensity, assuming c = 0:1 and no cyclic deterioration ( s;c;k;a = innite). For a tolerable probability of collapse of 10% in a 2=50 event, data of the type shown in Figure 10(a) provides the necessary design decision support (similar spectra are available for other combinations of system parameters). For instance, if T1 is selected as 0.9 s and the component ductility capacity is 4.0, the [Sa (T1 )=g]= value for a 10% probability of collapse is 4.6, which for the 2=50 hazard of the example problem (Sa (0:9) = 1:7g) results in a required base shear strength coe cient of = 1:7=4:6 = 0:37. Thus, collapse prevention would control the required strength, unless a larger ductility capacity (better detailing) is utilized or a more exible structure is used. For instance, for T1 = 0:9 s and c = y = 6, the [Sa (T1 )=g]= value is 5.4, which would result in = 1:7=5:4 = 0:31. Alternatively, a more exible structure could be selected (albeit this would be a poor solution based on direct ($)loss, see Figure 9). For T1 = 1:8 s and c = y = 4, the [Sa (T1 )=g]= value is 3.5, which for the 2=50 Sa value of 0:86g at 1.8 s results in a required base shear strength coe cient of = 0:86=3:5 = 0:25. These are the kind of trade-o s that can be evaluated through the use of collapse probability spectra of the type shown in Figure 10, presuming that a tolerable probability of collapse is specied at a specic hazard level. 4.2.2. Design for tolerable mean annual frequency of collapse. A di erent way to express desired collapse performance is to target a tolerable mean annual frequency (MAF) of collapse, C . This performance target is more general (it permits the estimation of the probability of
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

131

collapse over an expected life time), but it is more di cult to implement because the computation of a MAF requires integration over the Sa hazard curve. An approximate implementation is possible by means of the simplied closed form expression proposed in Reference [19], which estimates the MAF of collapse as follows:
C

=
Sa

P(C |Sa )|d

Sa (Sa )| =

1 Sa (C ) exp( 2

k 2 2 ) RC

(7)

The simplied expression on the right-hand side contains the MAF of the spectral acceleration associated with the 50% probability of collapse, Sa (C ), and a term that accounts, in an approximate manner, for the uncertainties inherent in the computation of the collapse capacity (the [Sa (T1 )=g]= value causing collapse). This term contains the slope of the hazard curve at the referenced spectral acceleration value, k, and the dispersion(s) in the collapse fragility curve, (the standard deviation of the natural logarithm of the data if a log-normal distribution is assumed for the probability of collapse given the spectral acceleration). In the example illustrated here only record-to-record (RTR) variability is considered, which is explicitly contained in the fragility curves shown in Figure 6 (these fragility curves are obtained by using a deterministic structural model subjected to 40 ground motions). Thus, the term RC in Equation (7) expresses the e ect of RTR variability only (Randomness in collapse Capacity), and is found to be on the order of 0.4 to 0.5 (except for long period structures for which it is smaller because of the dominance of P e ects). As an example, let us target a tolerable mean annual frequency of collapse of 0.0002 (i.e. a tolerable probability of collapse of approximately 0:0002 50 = 0:01 in a 50 year life span). This criterion could be used, together with [Sa (T1 )=g]= spectra for a 50% probability of collapse (see Figure 10(b)), to arrive at e ective design solutions. Again using the example of the nine-storey frame structure, the following design alternatives could be explored. If a period of 0.9 s and a component ductility capacity of c = y = 4 are targeted, then the median [Sa (T1 )=g]= value from Figure 10(b) is 7.7. For the site specic hazard curve the slope of the Sa hazard curve in the neighbourhood of a MAF of 0.0001 to 0.0004 is about 2.2, and the RC value is about 0.4 (from collapse fragility analyses [16]). Thus, from Equation (7), the MAF of the Sa associated with the 50% probability of collapse, Sa (C ), is equal to (0:0002)= exp(0:52:22 0:42 ) = 0:000136. From the Sa hazard curve for the site of the example problem, the corresponding Sa is 2:8g, and the corresponding value is 2:8=7:7 = 0:36. Again, this is a larger value than that obtained from ($)loss-based performance targets. Alternatives are to increase the component ductility capacity (if it is increased from 4 to 6, [Sa (T1 )=g]= is 8.9, and for the same Sa (C ) of 0.000136 the value becomes 2:8=8:9 = 0:31), or to increase the structure period. If, for instance, T1 is 1.8 s, the [Sa (T1 )=g]= value from Figure 10(b) is 5.4 (for c = y = 4), and, using the site specic k value of 2.4 for the T = 1:8 s hazard curve, Sa (C ) becomes 0.000126, the Sa value for this MAF is 1:4g, and the base shear strength parameter becomes 1:4=5:4 = 0:26. 5. CONCLUDING REMARKS Conceptual PBD implies a decision process that leads to the selection of one or several e ective design alternatives based on performance targets for acceptable losses and a tolerable probability of collapse. In this context, three challenges have to be addressed. One is to develop
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

132

H. KRAWINKLER ET AL.

data on EDP limits associated with selected performance targets. For this purpose the need exists to develop mean ($)loss-EDP curves for building subsystems for various occupancy and use categories, either for individual stories or for the building as a whole. Such curves would take the sting out of elaborate probabilistic formulations that appear to make performancebased design so complex. The second challenge is to develop collapse fragility curves that permit relatively simple design for collapse safety. Such curves are already available for regular and soft storey moment resisting frame structures. The third challenge is to develop mean IM-EDP relationships that permit rapid evaluation of various structural systems. Such relationships are available for moment resisting frame structures and are under development for reinforced concrete shear wall structures. If these ingredients are available, e ective conceptual PBD can be performed with the semi-graphical process discussed in this paper. The proposed process permits design decision making based on constraints imposed by multiple performance objectives, considering tradeo s between di erent structural system choices and between strength, sti ness, and ductility characteristics. The proposed process also permits a consistent evaluation of the costs and benets derived from the use of innovative technologies, such as base isolation and internal energy dissipation devices.

ACKNOWLEDGEMENTS

This research was carried out as part of a comprehensive e ort at Stanfords John A. Blume Earthquake Engineering Center to develop basic concepts for PBEE and supporting data on seismic demands and capacities. This e ort is supported by the NSF sponsored Pacic Earthquake Engineering Research (PEER) Center.

REFERENCES 1. Cornell CA, Krawinkler H. Progress and challenges in seismic performance assessment. PEER News, April 2000. 2. Krawinkler H. A general approach to seismic performance assessment. Proceedings of the International Conference on Advances and New Challenges in Earthquake Engineering Research, ICANCEER 2002, vol. 3, Hong Kong, 2002; 173 180. 3. Deierlein G. Overview of a comprehensive framework for earthquake performance assessment. Proceedings of the International Workshop on Performance-Based Seismic DesignConcepts and Implementation, Bled, Slovenia, 2004; 15 26. 4. Krawinkler H, Miranda E. Performance-based earthquake engineering. In Earthquake Engineering: From Engineering Seismology to Performance-Based Engineering, Chapter 9, Bozorgnia Y, Bertero VV (eds). CRC Press: Boca Raton, 2004; 9-1 9-59. 5. Baker J, Cornell CA. Choice of a vector of ground motion intensity measures for seismic demand hazard analysis. Proceedings of the 13th WCEE, Vancouver, Canada, 2004; Paper no. 3384. 6. Shome N, Cornell CA. Normalization and scaling accelerograms for nonlinear structural analysis. Proceedings of the Sixth U.S. National Conference on Earthquake Engineering, Seattle, Washington, l998; Paper no. 243. 7. Medina RA, Krawinkler H. Seismic demands for nondeteriorating frame structures and their dependence on ground motions. John A. Blume Earthquake Engineering Center Report No. 144, Department of Civil and Environmental Engineering, Stanford University, 2003. 8. Alavi B, Krawinkler H. Behavior of moment-resisting frame structures subjected to near-fault ground motions. Earthquake Engineering and Structural Dynamics 2004; 33(6):687 706. 9. Comerio MC. The in uence of performance engineering on disaster recovery: Priorities for limiting downtime. International Symposium on Earthquake Engineering, Japan Association for Earthquake Engineering, Awaji Island, 13 16 January 2005. 10. National Institute of Building Sciences. Assessment of state-of-the-art earthquake loss estimation methodologies. FEMA-249, Federal Emergency Management Agency, Washington, DC, 1994.
Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

DECISION SUPPORT FOR CONCEPTUAL PERFORMANCE-BASED DESIGN

133

11. Whitman RV, Anagnos T, Kircher CA, Lagorio HJ, Lawson RS, Schneider P. Development of a national earthquake loss estimation methodology. Earthquake Spectra 1997; 13(4):643 661. 12. Krawinkler H (ed.). Van Nuys hotel building testbedexercising seismic performance assessment. Pacic Earthquake Engineering Research Center, UC Berkeley, California. 13. Taghavi S, Miranda E. Response assessment of nonstructural building elements. Pacic Earthquake Engineering Research Center Report No. 2003/05, UC Berkeley, California. 14. Miranda E, Aslani H, Taghavi S. Assessment of seismic performance in terms of economic losses. Proceedings of the International Workshop on Performance-Based Seismic DesignConcepts and Implementation, Bled, Slovenia, 2004; 149 160. 15. Ibarra LF, Medina RA, Krawinkler H. Hysteretic models that incorporate strength and sti ness deterioration. Earthquake Engineering and Structural Dynamics 2005; 34(12):14891511. 16. Ibarra LF. Global collapse of frame structures under seismic excitations. Ph.D. Dissertation, Department of Civil and Environmental Engineering, Stanford University, Stanford, CA, 2003. 17. Sivaselvan MV, Reinhorn AM. Hysteretic models for deteriorating inelastic structures. Journal of Engineering Mechanics 2000; 126:633 640. 18. Song J, Pincheira J. Spectral displacement demands of sti ness and strength degrading systems. Earthquake Spectra 2000; 16(4):817 851. 19. Cornell CA. Calculating building seismic performance reliability; a basis for multi-level design norms. Proceedings of the 11th WCEE, Acapulco, Mexico, 1996; Paper no. 2120.

Copyright ? 2005 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2006; 35:115133

S-ar putea să vă placă și