Vous êtes sur la page 1sur 13

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/222691493

Earthquake risk assessment of building


structures

Article in Reliability Engineering [?] System Safety · December 2001


DOI: 10.1016/S0951-8320(01)00105-3

CITATIONS READS

129 1,936

1 author:

Bruce Ellingwood
Colorado State University
298 PUBLICATIONS 5,998 CITATIONS

SEE PROFILE

All content following this page was uploaded by Bruce Ellingwood on 05 April 2017.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Reliability Engineering and System Safety 74 (2001) 251±262
www.elsevier.com/locate/ress

Earthquake risk assessment of building structures


Bruce R. Ellingwood*
School of Civil and Environmental Engineering, Georgia Institute of Technology Atlanta, GA 30332-0355, USA
Accepted 16 August 2001

Abstract
During the past two decades, probabilistic risk analysis tools have been applied to assess the performance of new and existing building
structural systems. Structural design and evaluation of buildings and other facilities with regard to their ability to withstand the effects of
earthquakes requires special considerations that are not normally a part of such evaluations for other occupancy, service and environmental
loads. This paper reviews some of these special considerations, speci®cally as they pertain to probability-based codi®ed design and
reliability-based condition assessment of existing buildings. Dif®culties experienced in implementing probability-based limit states design
criteria for earthquake are summarized. Comparisons of predicted and observed building damage highlight the limitations of using current
deterministic approaches for post-earthquake building condition assessment. The importance of inherent randomness and modeling
uncertainty in forecasting building performance is examined through a building fragility assessment of a steel frame with welded connections
that was damaged during the Northridge Earthquake of 1994. The prospects for future improvements in earthquake-resistant design
procedures based on a more rational probability-based treatment of uncertainty are examined. q 2001 Published by Elsevier Science Ltd.
Keywords: Design (buildings); Deformations; Earthquakes; Limit states; Loads; Probability; Reliability; Risk; Statistics; Structural engineering

1. Introduction building codes. The legislated National Earthquake Hazard


Reduction Program (NEHRP) initiated a building industry-
Probability-based limit states design is the basis of most wide activity under the auspices of the Building Seismic
new structural design standards and speci®cations world- Safety Council (BSSC) to develop the so-called NEHRP
wide. The use of probabilistic concepts in this context Recommended Provisions for the Development of Seismic
stems from the recognition that loads arising from occupant Regulations for New Buildings. These recommended provi-
usage and man-made and natural hazards and the strengths sions have been updated periodically, most recently in 1997
of construction materials are uncertain in nature. These [4]. These advances notwithstanding, more recent earth-
uncertainties give rise to risk, which is managed by the quakes have focused, the attention of the public on apparent
provisions in codes and standards. Structural reliability continuing de®ciencies in earthquake-resistant design prac-
methods provide tools for quantifying the safety levels tices. In particular, the Northridge Earthquake of January
implied by such codes, and have been utilized in setting 17, 1994 resulted in weld fractures in the beam-to-column
the nominal (or characteristic) loads, load factors and load connections in nearly 200 buildings with moment-resisting
combinations, and resistance or material factors found in steel frames. This behavior in what was viewed as the
speci®cations for building design. These code-related activ- premier structural system for high-seismic regions came
ities re¯ect developments in reliability theory, supporting as a surprise to many structural engineers. Subsequent to
databases, and advances in computational methods during this earthquake, a number of research programs were under-
the past 25 years [14]. taken to address concerns regarding performance of steel
The management of risk due to earthquakes during this frames with welded connections and to further improve
developmental period of probability-based limit states earthquake-resistant design practice.
design has been problematic. Following the San Fernando Traditional earthquake-resistant design practice has been
earthquake of 1971, there was a surge of activity in the to include the earthquake effect in load combinations, as
United States aimed at improving seismic provisions in with other loads. In the ®rst set of general probability-
based load combinations for buildings and other structures
* Tel.: 11-410-516-8443; fax: 11-410-516-7473. [13,19], the following load combinations for earthquake-
E-mail address: bruce.ellingwood@ce.gatech.edu (B.R. Ellingwood). resistant design were developed using ®rst-order (FO)
0951-8320/01/$ - see front matter q 2001 Published by Elsevier Science Ltd.
PII: S 0951-832 0(01)00105-3
252 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

reliability methods: are expressed quantitatively through structural mechanics;


(2) identi®cation and analysis of uncertainties and support-
fRn ˆ 1:2D 1 1:5E 1 0:5L …1†
ing statistical databases; (3) target reliability measures
established through an assessment of existing and presum-
fRn ˆ 0:9D 1 1:5E …2†
ably acceptable practice; and ®nally (4) practical criteria
in which f is the resistance factor, Rn is the nominal (char- that achieve the target reliability measures for design or
acteristic) strength, D and L are dead and live loads, and E is condition assessment, in some average sense, for all struc-
the structural action (moment, shear, etc) due to earthquake. tures to which they are applied. The paper is organized
Eq. (2) governs when the dead load has a stabilizing effect around these issues.
on the structure. These combinations were incorporated in
ANSI Standard A58.1-1982, Minimum Design Loads for
Buildings and Other Structures. The basic approach has 2. Performance requirements and limit states
been retained in more recent editions of that standard
[34], the recent NEHRP code recommendations [4], and Modern approaches to earthquake engineering call for a
the new International Building Code [23]). The load factor building structure to remain undamaged during moderate
1.5 was reduced to 1.0 in 1993 to be compatible with the earthquakes and to dissipate suf®cient energy through cyclic
concurrent advances in seismic hazard analysis and changes inelastic deformations to avoid structural collapse during
in the speci®cation of the design earthquake. great earthquakes. Earthquake-resistant design procedures
The process leading to the above load combinations for take this inelastic behavior into account, implicitly through
earthquake-resistant design raised a number of issues that the speci®cation of the base shear, V, which is computed
were not encountered for the other load combinations in from a ground motion parameter (e.g. spectral acceleration
ASCE 7, and that have not been resolved in the intervening at the fundamental period of the building) and general
years. First, the earthquake is a base excitation rather than a dynamic structural characteristics. The (modal) lateral
clearly de®ned load, such as live, wind or snow. Second, the forces that are statically equivalent to this base shear are
structural response is dynamic, and must be related to distributed over the height of the building structure in a
damage that occurs under repeated (and usually inelastic) manner that is consistent with its mode shapes. In the
cycling. Third, the structural actions induced by the earth- NEHRP provisions [4], the base shear is,
quake are system-wide, whereas the effects of many other V ˆ Cs W …3†
loads are more or less localized. Finally, and perhaps most
important, the philosophy of earthquake-resistant design is in which Cs is the seismic response coef®cient and W is the
to limit the occurrence of life-threatening damage under the gravity load of the building. The coef®cient Cs is tantamount
design earthquake rather than to prevent its occurrence to a yield spectrum for an oscillator with 5% damping, and
entirely. With this philosophy, permanent inelastic lateral is a function of period and a response modi®cation factor, R,
deformations and local damage might be severe; however, which depends on the nature of the structural system and de-
the building would retain a substantial margin of safety ampli®es the elastic response. The structural action due to V
against overall collapse so that its occupants would not de®nes the load, E, in Eqs. (1) and (2). The analysis is
become trapped within the building. The limit state for performed elastically, and the members are proportioned
such a performance requirement (life safety under an earth- by the usual procedure. To account for the non-linear action
quake with a 10% probability of being exceeded in 50 years in the structure, the lateral de¯ection at story i is computed
or, equivalently, an earthquake with a 475 year mean recur- as,
rence interval) is dif®cult to de®ne. The traditional notions ui ˆ Cd uie …4†
of `strength' used in FO reliability analysis are not entirely
relevant, and the notions of code calibrations based on in which Cd is the de¯ection ampli®cation factor and uie is
member ®rst-order reliability analysis, on which the prob- the elastically computed de¯ections. Note that any non-line-
ability-based load combinations are based [19], lose some of arities in response and energy dissipation must be included
their signi®cance. More recent research following the Loma in the factors R and Cd [48].
Prieta Earthquake of 1989 and the Northridge Earthquake of Since damage rather than strength govern earthquake-
1994 and the recent move toward performance-based design resistant design and evaluation, reliability analysis in this
in a number of quarters [16] have further highlighted some case requires that a limit state be de®ned to capture the
of these dif®culties. notion of `damage' The damage done to a structure during
This paper surveys recent structural reliability methods strong earthquake ground motion depends on the energy that
for improving earthquake-resistant structural design and is transmitted to the structure as a whole and is absorbed or
condition assessment practices. The fundamental issues dissipated through cyclic inelastic deformations. The earth-
center on: (1) the underlying design methodology and quake engineering literature is replete with damage models
relation between performance requirements, which are (reviewed in Refs. [2,9,15,26]. The simplest damage vari-
expressed qualitatively, and structural limit states, which ables are those based on maximum inelastic deformation.
B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262 253

These include ductility demand, de®ned as the ratio of inter-story drift, for buildings with special moment-resisting
the maximum displacement of the system, umax, to its steel and reinforced concrete frames [16,44]. The limits for
displacement at yield, uy, and maximum inter-story drift, reinforced concrete shear wall buildings would be less by
which appears to correlate closely to damage to shear about a factor of 2.
walls, facades and non-structural partitions. Other damage Similar limits have been suggested by a number of
measures are based on damage accumulation laws similar investigators [6,49,51]. The inter-story drift at failure can
to those that are used in low-cycle fatigue analysis [38,45]. be modeled as a random variable, if such information is
For example, the widely used Park/Ang/Wen model (1985) available. For example, a review of the literature on
is: behavior of reinforced concrete frames subjected to lateral
Z forces indicated that for normal axial forces in columns,
D ˆ umax =uult 1 b‰ dEŠ=Eult …5† the mean inter-story drift at failure is 0.066 and the COV
is 0.31 Fig. 3 of Ref. [11]. The fact that P 2 D moments in
columns become signi®cant at lateral drifts in excess of
in which E is the hysteretic energy, uult is the deformation
about 0.02 lends additional support to the breakdown in
capacity under monotonic loading, Eult is the hysteretic
Table 1.
energy capacity, and b is the experimental constant.
The ®rst class of damage variables do not distinguish
cases when there is one large excursion into the non-linear 3. Sources of uncertainty in earthquake risk assessment
range from cases where there are several such excursions
during a large earthquake with long duration. On the other Earthquake risk assessment of building structures
hand, the second class of models requires a determination of requires: (1) identi®cation of the seismic hazard, described
the experimental constants in the damage accumulation by the annual probabilities of speci®c levels of earthquake
model (cf. Eq. (5)) and a time-domain analysis of the motion, P‰A ˆ aŠ; (2) analysis of response of the structural
dynamic characteristics of the excitation and structural system and its components; (3) calculation of limit state
response in the time or frequency domain. A review of the probabilities of attaining a series of (increasingly severe)
literature reveals no compelling advantage of the more limit states, LSi, through the expression:
complex damage accumulation models over the simpler X
deformation-based models. In particular, the coef®cient, P‰LSi Š ˆ P‰LSi uA ˆ aŠP‰A ˆ aŠ …6†
a
b , on the energy dissipation term in the Park/Ang/Wen
model is approximately 0.10 for steel or reinforced concrete, and, Eq. (4) relating these limit state probabilities to
making this model equivalent, for all practical purposes, to a building performance requirements. The conditional prob-
deformation-based model. Regardless of the damage ability, P‰LSi uA ˆ aŠ; is denoted as the seismic fragility.
measure selected, one must map the damage measure to Uncertainties are associated with each phase of the risk
heuristic descriptions of desired (or undesired) perfor- assessment, and must be displayed for the process to be
mance: minor damage, severe damage etc. There are transparent.
substantial epistemic uncertainties associated with this
mapping, particularly with damage accumulation models 3.1. Seismic hazard analysis
such as Eq. (5), which are normally calibrated to test data
The seismic hazard at a building site is displayed through
to yield D ˆ 1:0 at ultimate. When limit states shy of
a complimentary cumulative distribution function (CCDF),
collapse are of interest, the corresponding D must be arbi-
which displays the annual frequency of motion intensity at
trarily assigned values less than 1.0 [42].
or above a given level, x, to that intensity. The technology
Accordingly, at the current state-of-the-art, it is believed
for performing seismic hazard analysis and developing
to be suf®cient to de®ne the limit states for reliability analy-
ground motions for design probabilistically has stabilized
sis purposes in terms of the overall inelastic deformation or
[7,10,18,25,39]. Although intensity historically has been
the maximum inter-story drift of the structural system.
de®ned in terms of the maximum ground acceleration
Table 1 illustrates a possible mapping between performance
(zero-period acceleration), more recent trends have been
requirement and system limit state, expressed in terms of
toward using the spectral acceleration at the fundamental
Table 1 period of the building for this purpose. The process by
Performance requirements and limit states which the hazard curves are constructed involves uncertain-
ties that are both aleatory (or inherently random) and
Performance requirement Inter-story drift (%)
epistemic (or knowledge-based) in nature. The former
Serviceability 0.5 category would include the magnitude and location of the
Non-structural damage 1.0 next earthquake in the seismic source zone and character-
Impaired function 2.0 istics of ground motion at the site. The latter would include
Life safety 5.0
hypotheses regarding multiple seismic source zones, recur-
Onset of collapse 8.0
rence rates and Richter b-values, upper bound magnitudes,
254 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

attenuation relations, variation of ground response with system is described by the dynamic equation of equilibrium,
depth etc. In the nuclear industry, the practice has been
Mx 1 C_x 1 F…x† ˆ 2MIag …8†
to display the epistemic uncertainty through a family of
hazard curves. In most other ®elds, including building in which x is the relative displacement, M and C are mass
code development ([4]; ASCE 7-98), the different sources and damping matrices, F(x) is the restoring force vector, I is
of uncertainty are combined to yield one (mean or median) the identity vector, and ag is the ground acceleration. The
seismic hazard curve. non-linear response of the structural system during large
Elementary seismic hazard analysis [7] shows that at earthquakes is taken into account through the restoring
moderate to large values of ground acceleration, there is a force, F(x), modeled by an appropriate hysteresis rule.
logarithmic linear relation between annual maximum With the advances in computational power and ef®ciency
earthquake ground or spectral acceleration, and the prob- during the past decade, most recent reliability studies of
ability, GA(a), that speci®c values of acceleration are complex building systems have utilized non-linear ®nite
exceeded. This relationship implies that A is described by element computational platforms and time-domain analysis,
a Type II distribution of largest values; with Monte Carlo simulation to manage the uncertainties
in ground motion and response. There appears to be little
GA …x† ˆ 1 2 exp‰2…x=u†2k Š …7†
need for approximate closed-form random vibration analy-
in which u and k are parameters of the distribution. Para- ses nowadays, other than as an independent check on the
meter k de®nes the slope of the hazard curve which, in turn, ®nite element analysis, as the choice of restoring force and
is related to the coef®cient of variation (COV) in annual ground motion model is restricted if the non-linear response
maximum peak acceleration. Earthquake hazards in high- is computed by classical random vibration methods. One
seismic areas (e.g. California) generally can be associated model that can be implemented conveniently in non-linear
with a series of capable faults. Such faults are not apparent random vibration analysis with equivalent linearization is
in areas of low seismicity (e.g. eastern United States), and the Bouc model [1,24,46]. In contrast to many of the hyster-
the larger uncertainty due to the absence of large events esis models, alluded to above, the restoring force from this
during the historical period of observation causes the hazard model is smooth.
curve for such areas to be relatively ¯at in comparison. The In general, current structural response analysis capabil-
seismic hazard analyses on which the NEHRP Recom- ities are suf®cient for risk analysis purposes. One possible
mended Provisions [4] are based indicate that k is of the exception is in the area of modeling the change in hysteretic
order of 3±3.5 (COV of 50±70%) in the western US, and forces, F(x), in structures that have suffered damage as a
decreases to approximately 2±2.5 (COV in excess of 100%) result, of strong ground motion. Experimental data to de®ne
in the eastern US. The basic seismic hazard is by far such forces are limited [12] and integrating such models into
the largest single source of uncertainty in earthquake risk ®nite element analysis can be problematic [30].
analysis.
3.4. Structural system capacity±fragility models
3.2. Characteristics of ground motion
A structural fragility represents (probabilistically) the
Earthquake ground motion can be modeled in either time capability of ground motion to cause structural damage,
or frequency domains. A vast number of ground motion and is an essential ingredient of the fully coupled risk
records are maintained by the California strong motion assessment that is summarized by Eq. (6). It can also be
network, and these have been used in numerous studies used to determine probabilistic safety margins against
[2,32,35,36,41]. The ability to model earthquake ground speci®cally identi®ed earthquake events for design, evalua-
motion stochastically in time and frequency domains has tion and regulatory purposes. This latter use of fragility
matured considerably in recent years. In the SAC project, analysis has a number of advantages: (1) the structural
[41], and in other recent studies [44], ensembles of actual response and limit state determinations are uncoupled
ground motion records, appropriately scaled to the spectral from the basic seismic hazard analysis. Thus, while knowl-
acceleration at the fundamental period of the building, have edge of the hazard is useful in identifying appropriate
been used in probabilistic studies of building response. This review-level events (e.g. an earthquake with a 2% probabil-
approach accounts for non-stationarity in amplitude and ity of being exceeded in 50 years, or 2475 years MRI earth-
frequency content in acceleration in a natural way, and is quake), such knowledge is not essential. One might, for
compatible with recent risk analysis approaches that involve example, simply inquire as to the fragility were the maxi-
a non-linear structural response analysis performed in the mum considered earthquake to be exceeded by 50%. (2) The
time domain. need to interpret and defend very small limit state probabil-
ities (in the order of 10 23 or less/year), such as might be
3.3. Structural response computed from the fully coupled risk analysis in Eq. (6), is
avoided. (3) A properly conducted fragility analysis is less
The structural response of a multi-degree of freedom of complex, less costly, and involves fewer disciplines than a
B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262 255

Fig. 1. Hazard±fragility interface, identifying dominant contributors to risk.

fully coupled risk analysis, and the results are more easily It should be noted that a fragility assessment describes the
explained to a non-specialist decision-maker or regulatory performance of the system as a whole with respect to some
authority. limit state of performance (here, measured in terms of defor-
The fragility is de®ned as the probability of attaining a mation, as described earlier). A ®nite element analysis that
limit state, conditioned on a particular value of a random includes both second-order geometric and material non-
demand, X [5,28]. It has been customary to model compo- linearities is required to support this assessment.
nent or system fragility by a lognormal cumulative distribu-
tion function (CDF): 3.5. Structural system reliability
FR …x† ˆ F‰ln…x=m R †=bR Š …9† A point estimate of the limit state probability for state i
in which F [ ] is the standard normal probability integral, mR can be obtained by convolving the fragility FR(x) with the
is the median (50th percentile) fragility, and the logarithmic derivative of the seismic hazard curve, GA(x), thus removing
standard deviation, b R, describes the inherent variability in the conditioning on acceleration (cf. Eq. (6)):
the capacity of the component subjected to an earthquake Z1 dG
with peak (ground or spectral) acceleration, x. For COV in P‰LSi Š ˆ FR …x† A dx …11†
0 dx
capacity, VR, less than about 0.3, b R ˆ VR : The lognormal
model of fragility is relatively convenient to use, and has The parameters at the fragility±hazard interface must be
some theoretical justi®cation as well [12]. Simulation dimensionally consistent for the probability estimate to be
studies involving steel moment frames [44] and reinforced meaningful.
concrete frames [11,22,42] show that the lognormal CDF Additional insight into the sources of earthquake risk can
provides an acceptable fragility model well into the inelastic be gleaned from a close examination of the convolution in
range of response. Eq. (11) at the hazard±fragility interface. This behavior is
As with the seismic hazard analysis, various modeling illustrated conceptually in Fig. 1 The parameters xmin and
assumptions and limitations in the supporting databases xmax in Fig. 1 are de®ned such that 95% of P‰LSi Š is deter-
give rise to epistemic uncertainties that affect the fragility mined by accelerations within those limits. In words, most
of a structure or component and must be taken into account. of the limit state probability in Eq. (11) is derived from
If it is assumed, to ®rst order, that these epistemic uncer- earthquakes giving rise to accelerations in this range. One
tainties, b U, affect the estimate of the median capacity, mR, might say that such earthquakes are `dominant contributors'
in Eq. (9), then mR can be thought of as a (Bayesian) random to seismic risk. Previous studies of seismic risk have
variable, MR, with median mR and logarithmic standard suggested that such accelerations correspond to 1±2 times
deviation, b U. Removing the conditioning on mR in Eq. (9) the 475 year mean return period earthquake, which has
leads to, been the basis for earthquake-resistant design. It follows
that investments in efforts to better de®ne the nature of
FR …x† ˆ F‰ln…x=m R †=bC Š …10a†
the seismic hazard outside this range will have little return
in which, insofar as risk reduction is concerned.
q If FR(x) in Eq. (11) is lognormal and GA(x) has a Type II
bC ˆ b2R 1 b2U …10b† distribution of largest values, Eq. (11) can be simpli®ed
256 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

using the observation above that the major contribution to (13) involved structural system behavior indirectly, in the
risk comes from a relatively narrow range [8]. Approxi- determination of both V (through response modi®cation
mating GA(x) as log-linear over that range, we ®nd that, coef®cient, R) and the lateral de¯ections (through Cd).
Moreover, the limit states differ as well (life safety vs. loca-
P‰LSi Š ˆ GA …mR †exp‰…kVR †2 =2Š …12† lized structural damage from yielding or formation of
In words, this says that the limit state probability is equal plastic hinges). Without suf®cient information to resolve
to the seismic hazard, evaluated at the median fragility, this issue, however, the new probability-based load combi-
multiplied by a correction factor. If VR ˆ 0:15 (a typical nations (e.g. Eqs. (1) and (2)) were based on these lower
value) and k ˆ 3 (as noted before), this correction factor b 's.
equals 1.11; if k ˆ 2; it equals 1.05. For all intents and More recent studies have utilized non-linear dynamic
purposes, then, the limit state probability can be estimated analysis tools to include inelastic deformation and energy
by performing a deterministic structural response analysis dissipation in the system reliability analysis directly, rather
with all parameters at their median values, and evaluating than through coupling R and Cd to a static elastic analysis.
the seismic hazard at the median response value thus Such studies have been performed on steel portal frames
determined. The uncertainty associated with the prediction designed by US or Japanese building codes [31,36,50], rein-
of the basic seismic hazard (measured by k) is very large in forced concrete frame and shear wall structures designed by
comparison with other sources of uncertainty in the relia- the Uniform Building Code [22] and multistory reinforced
bility evaluation, especially in areas of low seismicity. In concrete frames designed by the New Zealand and Mexican
such cases, replacing the fragility in Eq. (11) with a median codes [17,37]. These studies, without exception, all lead to a
value and neglecting VR entirely will have little impact on similar conclusion: that the apparent limit state probability
calculated risk. It would be more bene®cial to ensure that when design is governed by seismic load provisions is less
the median dynamic structural response is estimated as than when design is governed by one of the gravity load
accurately as possible. combinations. Moreover, the b 's obtained from more
complex, dynamic analyses involving limit states that
re¯ect inelastic behavior explicitly are essentially the
4. Risk assessment of building structures same as the b 's obtained from the earlier simple static-
for-dynamic analysis. Thus, the evolution of codes used
4.1. Assessment of traditional design criteria for existing earthquake-resistant structural engineering
practice apparently has led to a situation in which the
The ®rst generation of probability-based limit states measures of reliability associated with different controlling
design codes are based on a code calibration process, in design load combinations are not the same.
which (1) the reliabilities of members designed by existing
conventional earthquake-resistant design practice were 4.2. Recent developments in fragility modeling of steel
determined, and (2) these reliabilities set the benchmarks frames
for new criteria. The basic assumption in this process is that
traditional design procedures have provided acceptable The Northridge Earthquake of 1994 (Richter Scale ML ˆ
structures (in terms of reliability) for different combination 6:4†; resulted in weld fractures in beam-to-column connec-
of loads. For load combinations involving earthquake tions in nearly 200 buildings with special moment-resisting
forces, this assumption is arguable. steel frames. Many of the failures observed in connections
In the ®rst attempt to perform this calibration, [19], the that were inspected following the earthquake involved a
limit state was de®ned as, weld fracture that initiated at the root of the groove weld
connecting the bottom ¯ange of the beam to the column
M 2 cV ˆ 0 …13† ¯ange [21]. To investigate building performance issues
in which M is the ¯exural capacity of a beam or column, V is and to support improvements in probability-based limit
the base shear determined from Eq. (3), in which all para- states design for steel structures, four welded steel moment
meters are treated as random variables, and c is the struc- steel frames that suffered connection damage in the
tural analysis factor required to transform base shear to Northridge Earthquake were evaluated using a stochastic
structural action. Both steel and reinforced concrete beams approach [43,44]. The evaluation of one of these building
and columns were considered. Using advanced FO reliabil- frames is summarized below. The comparison of predicted
ity analysis [33], it was found that the reliability index, b behavior to that observed during inspections following
(50 year basis), fell within the range of 1.75±2.25 for struc- the earthquake identi®es limitations in current modeling
tures with periods between 0.5 and 1.0 s. These values of b procedures and the role of uncertainty in forecasting
seemed low in comparison with those for structural building performance.
members in which the gravity load combinations governed
design, which fell in the range of 2.5±4.0. In a sense, the 4.2.1. Description of the building frame analyzed
comparisons were misleading. The analysis based on Eq. The non-linear action in steel building frames that are
B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262 257

Fig. 2. Hysteresis model for damaged welded connection.

subjected to a number of reversals of inelastic deformation of the steel members in the structural system differed
during a large earthquake is concentrated in the beam- from the values that are assumed for design purposes.
to-column connections. The analysis must replicate this The steel in the beams satis®ed the requirements of
moment-rotation behavior in order to model the deteriora- ASTM A36 steel, with speci®ed yield point 248 MPa
tion in connection rigidity and the resulting redistribution of (36 ksi), while the columns met the requirements of
forces and increases in lateral drift within the frame. Fig. 2 ASTM A572 Grade 50 steel, with speci®ed yield point
illustrates a hysteretic moment-rotation model that was 345 MPa. The mean beam and column yield strengths,
developed to model the effect of this particular type of based on tests conducted at Lehigh University [27] were
connection damage [21]. This model is based on tests of 291 and 393 MPa, respectively.
structural connections that were conducted as part of the A 2D non-linear dynamic model of the building frame
SAC Joint Venture [40], a large research program initiated was constructed. The fundamental period of the building
following the Northridge Earthquake to deal with perfor- was calculated as 1.6 s, and the structural damping was
mance issues raised. In the undamaged state, the hysteresis assumed to be 2%. Ground motions from the Northridge
is characterized by a bilinear envelope with a yield capacity Earthquake were unavailable at the site of this building.
speci®ed as My. The moment at fracture of the groove weld Thus, an ensemble of nine ground motions developed in
is denoted Mcr, which is speci®ed as a fraction of My. Phase 1 of the SAC Joint Venture [20] for a 1 km grid
Following weld fracture, the primary envelope is replaced surrounding this particular building site were used in its
by a degraded bilinear representation with reduced stiff- analysis. The ®rst mode dominated the response of the
ness, reduced capacity, and a modi®ed post-yield slope. frame in the elastic range. The roof displacement angle
Typical values of the hysteresis parameters are k2 ˆ (RDA), de®ned as the maximum roof displacement normal-
0:05k1 ; b1 ˆ 0:4; b2 ˆ 0:2; and b3 ˆ 0:4; with both b 4 ized by the building height, often is used as a relative
and b 5 equal to 1.0 [43]. This moment-rotation model was measure of building response when de®ning deformation-
incorporated in a program that performs inelastic dynamic related limit states in earthquake engineering. The non-
analyses of planar steel frames [30]. linear response of a frame can be characterized by a
The of®ce building analyzed is relatively modern, and its `pushover' analysis, in which the lateral forces (proportional
framing system was designed using the seismic provisions to the ®rst-mode forces) are applied to the system [29]. The
in Los Angeles in about 1985. The building is located pushover analysis of this building frame indicated that
approximately 10 km from the epicenter on an alluvium non-linear action initiated at an RDA of approximately
site. Its overall plan dimensions are 45 £ 33 m 2, and the 1%, which is typical of such structural systems. A detailed
story heights range from 4 to 4.7 m. The moment frames comparison of the predicted and observed dynamic
in this building are single-bent, and are approximately responses of this building to a suite of ground motion
12.5 m in span [43]. A typical moment frame from this records is presented elsewhere [43]. Fig. 3 compares the
building is illustrated in Fig. 3. An explanation of the predicted and observed connection damage patterns in
symbols on Fig. 3 will be provided as its behavior is the frame obtained using one of the nine ground motions
discussed below. The actual strength and stiffness properties in the ensemble. If connection damage of the type modeled
258 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

Fig. 3. Surveyed and predicted damage for typical moment-resisting frame.

by Fig. 2 was observed or predicted, that portion of the bility compared to the uncertainty in earthquake ground
circle is darkened. The agreement is reasonable, but motion [36], and are set equal to their mean values.
differences are clearly apparent. These differences stem The stochastic analysis of non-linear building response in
from the large uncertainties in the ground motion model the time domain requires an ensemble of ground motions.
and structural response modeling process. The uncertainty
in structural response can be captured by the building frame Table 2
fragility. Random strength parameters

4.2.2. Structural fragility of steel building frame Parameter Mean COV CDF
The structural parameters that are treated as random Fy (column) (MPa) 393 0.12 Lognormal
variables, and their distributions are listed in Table 2. An Fy (beam) (MPa) 290 0.12 Lognormal
analysis of variance [44] indicated that the most statistically b1 0.40 0.29 Uniform
signi®cant hysteresis parameters were b 1 and b 5; other b5 0.95 0.09 Uniform
E (GPa) 200 0.06 Uniform
hysteresis parameters are treated as deterministic. Gravity
Damping 0.02 0.62 Lognormal
loads and building mass are known to have a small varia-
B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262 259

Fig. 4. Building fragilities for minor and severe damage states.

The ensembles of ground motions simulated for the 1 km Eqs. (15a) and (15b). Fig. 4 compares the fragilities for
grid surrounding the building can be used for this purpose. two limit states: minor damage …RDA ˆ 0:02† and onset
Alternatively, historic earthquake accelerograms with of severe damage …RDA ˆ 0:05†: The variability in perfor-
comparable magnitudes and epicentral distances can be mance increases substantially as the damage state becomes
used [41]. Either way, the peak ground motion intensities more severe. In passing, it is interesting to note that the
vary from record to record. Fragility and risk analysis fragility for the 0.05 drift limit state in Fig. 4 is similar to
require a common ground motion intensity measure to the fragility for the `moderate' damage state in reinforced
which the building resistance can be keyed. The spectral concrete frames shown in Fig. 11 of Ref. [42].
acceleration, Sa, at the fundamental period of the building
can be used for this purpose (NEHRP, 1997).
The fragility, FR(x), is de®ned by the conditional prob- 5. Condition assessment and probability-based design
ability (cf. Eqs. (6), (9), (10a) and (10b)),
Current earthquake-resistant design procedures [23] do
FR …x† ˆ P‰LSuSa ˆ xŠ …14†
not relate performance goals to probability. Probability is
in which LS represents the structural limit state correspond- however, used to de®ne the return period of the design
ing to the performance requirement (cf. Table 1) and the earthquake. Thus, in the near future, one might envision a
control variable is the spectral acceleration at the fundamen- building safety evaluation process requiring, in part, that an
tal period of the building. In this study, limit states are ordinary building should not suffer `severe damage' with
associated with the RDA, to be consistent with research 90% con®dence if subjected to an earthquake with a mean
conducted elsewhere as part of the SAC Joint Venture recurrence interval of 500 years. The 500 year earthquake
[50]. The fragility is obtained from the cumulative distribu- might be characterized by a spectral acceleration of approxi-
tion function of the RDA associated with that particular mately 0.6g in many areas of the Western United States.
limit state. For example, for the RDA ˆ 0:02 associated Using the building fragility summarized in Fig. 4, the 10-
with the LS as `minor damage' (cf. Table 1), percentiles for the steel frame considered range between
0.84 and 1.02g, depending on whether simulated or recorded
FR …x† ˆ P‰minor damageuSa ˆ xŠ …15a†
ground motions are used. Thus, it is unlikely that building
would be severely damaged. On the other hand, the prob-
FR …x† ˆ 1 2 P‰RDA , 0:02uSa ˆ xŠ …15b†
ability of moderate damage …RDA ˆ 0:02† ranges between
To determine these conditional probabilities, the ground 27 and 61%. Such analyses could be used as a basis for
motion ensembles are scaled to increasing values of Sa at public policy and determining the extent of post-disaster
the fundamental period of the building, the corresponding rehabilitation likely to be required following the occurrence
dynamic responses of the frames are determined at each of such an earthquake.
value of Sa, the responses are rank-ordered on lognormal Carrying the analysis one step further to the fully coupled
probability plots, and the fragility is determined from risk analysis case, the limit state probabilities can be
260 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

estimated from Eq. (8) if the slope, k, of the seismic hazard based on analytical models, rather than on prototype
curve is known. Assuming, for example, that k ˆ 2:4; the experimental data, because of the complexity of building
annual probabilities for moderate and severe damage to this structural systems. In such circumstances, the credibility
frame are, respectively, 0.0028 and 0.00034. On a 50 year of the computational platforms and supporting databases
basis, the limit state probability for severe damage is becomes a critical issue. While the best way to validate,
approximately 0.017, which corresponds to a reliability an analytical model is through a comparison of calculated
index of approximately 2.1. These results are similar to and observed failure rates, such comparisons can be made
those obtained recently for steel frames by other investiga- in earthquake engineering only for very simple models.
tors [6,49,50], and to values reported during the past two Absent such comparisons, the risk analysis process
decades in a variety of studies and contexts. must be `transparent', in the sense that the computational
One of the goals of reliability-based design is to achieve (simulation) model and the supporting databases must be
measures of reliability across competing hazards that are well-documented, explainable to technically quali®ed non-
consistent with the nature of the hazard, its consequences, specialists, and subject to independent veri®cation and
and the expectations of the building owner and occupants. audit. If quantitative risks and numerical safety goals
The ®rst generation of probability-based design criteria (e.g. (expressed in terms of hazard or limit state probabilities)
Eqs. (1) and (2)) were developed in such a way as to be are to be used as a basis for codes and public policy, the
easily used in the customary manner by engineers who were calculated risks must be the result of an acceptable and
not familiar with the underlying probabilistic concepts. communicable process to facilitate public and professional
There is only one performance objective (life safety) and con®dence in the decision. This is particularly important
one corresponding safety checking equation for earthquake- nowadays, where such decisions are closely scrutinized in
resistant design. The aftermath of several natural disasters in the public arena.
the United States has made it apparent that this narrow view Most structural design and condition assessment pro-
of building performance is not socially or economically cedures impart some bene®t from redundancy. Conversely,
acceptable [16]. Performance-based design is a new para- a `redundancy factor' has been introduced into the earth-
digm, in which the design objectives are aimed at meeting quake provisions in ASCE 7-98 that effectively increases
the performance expectations of the building occupants and the member design forces to compensate for lack of
the public. These expectations often exceed what is implied redundancy. The impact of redundancy on performance
by the life-safety requirement in current codes. Most recent and limit state probabilities of buildings subjected to earth-
proposals for performance-based codes distinguish the quakes, where it depends on the framing system and pattern
performance requirements for different building occupan- of lateral load, remains unclear. Recent studies have sug-
cies, depending on what are the consequences of failure. gested a statistical approach for dealing with the effect of
Moreover, designing to multiple levels of hazard is a system redundancy in the member safety checks as part of
common ingredient of such proposals. Recent advances in earthquake-resistant design [3]. However, a simple system
earthquake risk analysis make such an approach feasible, reliability analysis of a realistic frame for lateral earthquake
and a number of investigators [6,47,49] have suggested forces shows that the probabilistic effects of redundancy are
design for two levels of earthquake. For example, the negligible because the uncertainty in the earthquake force is
code might require that no structural damage be suffered so much larger than the variability in structural member
during a 100 year mean recurrence interval event (a properties. Moreover, in an era in which static non-linear
`serviceability' requirement of sorts), while no collapse pushover analyses to determine inelastic force redistribution
should occur during a 2500 year event. Questions naturally on the frame are becoming feasible in earthquake-resistant
arise as to what level of hazard should be speci®ed for design, such simple redundancy factors appear to have
different performance levels, and with what con®dence questionable value and are dif®cult to rationalize [52].
should the building survive if such a hazard where to
occur. For example, is the probability 0.017 of severe
damage to a building with a steel moment frame under a 6. Concluding remarks
500 year return period event acceptable? It is clear that the
®rst-generation probability-based LRFD criteria, with their Recent development in earthquake engineering [6,4,50]
constant load and resistance factors, provide little help in indicate that probabilistic methods will be used increas-
answering such questions. It is dif®cult to see how perfor- ingly to provide the basis for improvements in earth-
mance-based design can reach its full potential without an quake-resistant structural design and condition assessment.
explicit consideration of the risk levels associated with Comparisons of predicted and observed building damage
different performance objectives for different building show that despite recent advances in non-linear dynamic
categories. analysis and structural modeling, there are limitations in
The probabilities for the life-safety and incipient building using purely deterministic approaches for post-earthquake
collapse limit states in earthquake structural engineering building condition assessment. Uncertainties can lead to
are necessarily low. The risk estimates invariably are signi®cant discrepancies between predicted and observed
B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262 261

behavior, even when state-of-the-art modeling is employed. [8] Cornell CA. Risk-based structural design. Proceedings of the Sympo-
A probabilistic analysis of building response is essential in sium on Risk Analysis. University of Michigan, 1994. p. 37±48.
[9] Cosenza E, Manfredi G. Damage indices and damage measures. Prog
providing perspective on building behavior that is likely to Struct Engng Mater 2000;2(1):50±9.
occur. The use of fragility models and limit state probabil- [10] Der Kiureghian A. Seismic risk analysis of structural systems. J
ities can serve as a basis for improvements in structural code Engng Mech Div ASCE 1981;107(6):1133±53.
development, condition assessment and performance-based [11] Dymiotis C, Kappos AJ, Chryssanthopoulos MK. Seismic reliability
design. At the same time, there is a need for relatively of RC frames with uncertain drift and member capacity. J Struct
Engng ASCE 1999;125(9):1038±47.
simple procedures for evaluating earthquake risk for [12] Ellingwood B. Validation studies of seismic PRAs. Nucl Engng Des
purposes of practical design, codi®cation and condition 1990;123(2):189±96.
assessment. Reliability analysis tools have become quite [13] Ellingwood B, MacGregor JG, Galambos TV, Cornell CA. Probabil-
sophisticated. Advanced computational methods may be ity-based load criteria: load factors and load combinations. J Struct
warranted for the evaluation of special structures, but such Div ASCE 1982;108(5):978±97.
[14] Ellingwood B. Probability-based codi®ed design: past accomplish-
methods are not economically feasible for the design and
ments and future challenges. Struct Saf 1994;13(3):159±76.
condition assessment of the vast majority of buildings. [15] Ellingwood B. Probability-based codi®ed design for earthquakes.
Nowadays, the needs are less for even more advanced, Engng Struct 1994;16(7):498±506.
computational platforms, but rather for a better understand- [16] Ellingwood B. Reliability-based performance concept for building
ing of risk that are calculated and their implications for construction. Structural Engineering Worldwide 1998, Paper T178-
4, Elsevier (CD-ROM), 1998.
building performance. The stage now has been reached
[17] Esteva L, Ruiz SE. Seismic failure rates of multistory frames. J Struct
where it is easier to obtain answers than it is to obtain Engng ASCE 1989;115(2):268±84.
good answers. [18] Frankel A, Mueller C, Barnhard J, Perkins DM, Leyendecker EV,
The process leading to the ®rst-generation probability- Dickman N, Hanson S. National seismic hazard maps: documentation
based limit states design criteria sidestepped the important June 1996. U.S. Geological Survey Open File Report 96-532, 1996.
p. 110.
risk validation issue, and this is an important area for
[19] Galambos TV, Ellingwood B, MacGregor JG, Cornell CA. Probabil-
further research. More accurate probabilistic models of ity-based load criteria: assessment of current practice. J Struct Div
ground motion or descriptions of response may not enhance ASCE 1982;108(5):959±77.
the credibility of provisions for earthquake-resistant [20] Graves RW, Somerville PG, McLaren JP. Ground motion time
structural design or condition assessment, particularly at histories of the 1994 Northridge Earthquake at three sites. Report
prepared for NIST by Woodward-Clyde, National Institute of Stan-
the very low failure rates typical of civil construction.
dards and Technology, Gaithersburg, MD, 1995.
Failure rates of structures depend not only on stochastic [21] Gross JL. A connection model for the seismic analysis of welded steel
variability, but also on the adequacy of models of behavior moment frames. Engng Struct 1998;20(4±6):390±7.
used in analysis and design, control of design/construction [22] Hwang H HM, Jaw J-W. Probabilistic damage analysis of structures.
errors through communication, QA/QC, proper construction J Struct Engng ASCE 1990;116(7):1992±2007.
[23] International Building Code (IBC) 2000. International Code Council.
supervision and in-service inspection. Methods for integrat-
Whittier, CA.
ing these factors and communicating the credibility of the [24] Iwan W, Paparizos. Stochastic response of strongly yielding systems.
analytical models, data and expert judgement in reducing Probabil Engng Mech 1988;3(2):75±82.
earthquake risk should be on the critical path for further [25] Kanda J, Dan K. Distribution of seismic hazard in Japan based on an
development. empirical extreme value distribution. Struct Saf 1987;493:229±39.
[26] Kappos AJ. Seismic damage indices for RC buildings: evaluation of
concepts and procedures. Prog Struct Engng Mater 1997;1(1):78±87.
[27] Kaufmann EJ, Fisher JW, DiJulio RM, Gross JL. Failure analysis of
References welded steel moment frames damaged in the Northridge Earthquake.
Report NISTIR 5944, National Institute of Standards and Technology,
[1] Baber TT, Wen YK. Random vibration of hysteretic degrading Gaithersburg, MD, 1997.
systems. J Engng Mech Div ASCE 1981;107(6):1069±89. [28] Kennedy RP, Ravindra MK. Seismic fragilities for nuclear power
[2] Bazzurro P, Cornell CA, Shome N, Carballo JE. Three proposals for plant studies. Nucl Engng Des 1984;79:47±68.
characterizing MDOF nonlinear seismic response. J Struct Engng [29] Krawinkler H, Seneviratna G. Pros and cons of a pushover analysis of
ASCE 1998;124(11):1281±9. seismic performance evaluation. Engng Struct 1998;20(4):452±64.
[3] Bertero R, Bertero VV. Redundancy in earthquake-resistant design. J [30] Kunnath SK. Enhancements to program IDARC: Modeling inelastic
Struct Engng ASCE 1999;125(1):81±8. behavior of welded connections in steem moment-resisting frames.
[4] BSSC (1998). NEHRP recommended provisions for the development Report GCR 95-673, National Institute of Standards and Technology,
of seismic regulations for new buildings (1997 ed.). FEMA Report Gaithersburg, MD, 1995.
302, Federal Emergency Management Agency, Washington, DC, [31] Kuwamura H, Galambos TV. Reliability analysis of steel building
1998. structures under earthquakes. Structural Engineering Report 86-01,
[5] Casciati F, Faravelli L. Fragility analysis of complex structural Department of Civil Engineering, University of Minnesota,
systems. Somerset, UK: Research Studies Press, Wiley, 1991. Minneapolis, MN, 1986.
[6] Collins KR, Wen YK, Foutch DA. Dual level seismic design: a [32] Lai SP. Statistical characterization of strong ground motions using
reliability-based methodology. Earthq Engng Struct Dynm 1996;25: power spectral density functions. Bull Seismol Soc Am 1982;
1433±67. 72(1):259±74.
[7] Cornell CA. Engineering seismic risk analysis. Bull Seismol Soc Am [33] Melchers RE. Structural reliability: analysis and prediction. Chiche-
1968;58(5):1583±606. ster, UK: Ellis-Horwood Ltd, 1987.
262 B.R. Ellingwood / Reliability Engineering and System Safety 74 (2001) 251±262

[34] Minimum design loads for buildings and other structures (ASCE [44] Song J, Ellingwood BR. Seismic reliability of special moment steel
Standard 7-98; formerly ANSI A58.1-1982). American Society of frames with welded connections II. J Struct Engng ASCE 1999;
Civil Engineers, Reston, VA, 2000. 125(4):372±84.
[35] Newmark NM, Hall WJ. Earthquake spectra and design. Earthq [45] Stephens J, Yao JTP. Damage assessment using response measure-
Engng Res Inst Monogr Ser Berkeley, CA, 1982. ments. J Struct Engng ASCE 1987;113(4):787±801.
[36] O'Connor J, Ellingwood B. Reliability of nonlinear structures with [46] Sues RH, Wen Y-K, Ang AH-S. Stochastic evaluation of seismic
seismic loading. J Struct Engng ASCE 1987;113(5):1011±28. structural performance. J Struct Engng ASCE 1985;111(6):1204±18.
[37] Ogawa S. A simpli®ed procedure for assessing failure probabilities of [47] Takeda T. New challenges for probabilistic performance-based
reinforced concrete frame buildings under earthquake loading. seismic design, Proceedings of the APSSRA'99 (Asian-Paci®c Sym-
Research Report 88-11, Department of Civil Engineering, University posium on Structural Reliability and Applications). Taipei, Taiwan:
Of Canterbury, Christchurch, New Zealand, 1988. National Center for Research in Earthquake Engineering, 1999.
[38] Park YJ, Ang AHS, Wen YK. Seismic damage analysis of reinforced p. 330±41.
concrete buildings. J Struct Engng ASCE 1985;111(4):740±57. [48] Uang C-M. Establishing R (or Rw) and Cd factors for building seismic
[39] Reiter L. Earthquake hazard analysis: issues and insights. New York: provisions. J Struct Engng ASCE 1991;117(1):19±28.
Columbia University Press, 1990. [49] Wen YK. Building reliability and code calibration. Earthq Spectra
[40] SAC Joint Venture. Experimental investigations of beam-column 1995;11(2):269±96.
subassemblages. Report No. SAC-96-01 Parts 1 and 2, SAC Joint [50] Wen YK, Foutch DA. Proposed statistical and reliability framework
Venture, Sacramento, CA, 1996. for comparing and evaluating predictive models for evaluation and
[41] Shome N, Cornell CA, Bazzurro P, Carballo JE. Earthquakes, records, design, and critical issues in developing such a framework. Report
and nonlinear responses. Earthq Spectra 1998;14(3):469±500. No. SAC/BD-97/03, SAC Joint Venture, Sacramento, CA, 1997.
[42] Singhal A, Kiremedjian AS. Method for probabilistic evaluation of [51] Whittaker A, Constantinou M, Tsopelas P. Displacement estimates
seismic structural damage. J Struct Engng ASCE 1996;122(12): for performance-based seismic design. J Struct Engng ASCE 1998;
1459±67. 124(8):905±12.
[43] Song J, Ellingwood BR. Seismic reliability of special moment steel [52] Whittaker A, Hart G, Rojahn C. Seismic response modi®cation
frames with welded connections I. J Struct Engng ASCE 1999;125(4):
factors. J Struct Engng ASCE 1999;125(4):438±44.
357±71.

View publication stats

Vous aimerez peut-être aussi