Sie sind auf Seite 1von 47

Bulletin of Earthquake Engineering 2: 173219, 2004.

2004 Kluwer Academic Publishers. Printed in the Netherlands.


A Probabilistic Displacement-based
Vulnerability Assessment Procedure
for Earthquake Loss Estimation
HELEN CROWLEY
1
, RUI PINHO
1,
and JULIAN J. BOMMER
2
1
European School for Advanced Studies in Reduction of Seismic Risk (ROSE School),
University of Pavia, Via Ferrata, 27100 Pavia, Italy,
2
Department of Civil and Environmental
Engineering, Imperial College London, South Kensington campus, London SW7 2AZ, UK
*Corresponding author. Tel: +39-0382-505859, Fax: +39-0382-528422, E-mail:
rui.pinho@eucentre.it
Abstract. Earthquake loss estimation studies require predictions to be made of the propor-
tion of a building class falling within discrete damage bands from a specied earthquake
demand. These predictions should be made using methods that incorporate both computa-
tional efciency and accuracy such that studies on regional or national levels can be effec-
tively carried out, even when the triggering of multiple earthquake scenarios, as opposed
to the use of probabilistic hazard maps and uniform hazard spectra, is employed to real-
istically assess seismic demand and its consequences on the built environment. Earthquake
actions should be represented by a parameter that shows good correlation to damage and
that accounts for the relationship between the frequency content of the ground motion and
the fundamental period of the building; hence recent proposals to use displacement response
spectra. A rational method is proposed herein that denes the capacity of a building class
by relating its deformation potential to its fundamental period of vibration at different limit
states and comparing this with a displacement response spectrum. The uncertainty in the
geometrical, material and limit state properties of a building class is considered and the rst-
order reliability method, FORM, is used to produce an approximate joint probability density
function (JPDF) of displacement capacity and period. The JPDF of capacity may be used
in conjunction with the lognormal cumulative distribution function of demand in the classi-
cal reliability formula to calculate the probability of failing a given limit state. Vulnerability
curves may be produced which, although not directly used in the methodology, serve to illus-
trate the conceptual soundness of the method and make comparisons with other methods.
Key words: displacement-based assessment, earthquake loss estimation, RC structures,
reliability, vulnerability curves
1. Introduction
The principal requirement of an earthquake loss model is an estimation of
the proportion of buildings in an urban environment which will fall within
discrete damage bands, both structural and non-structural, when subject
to a specied earthquake demand. Currently available methods include a
174 H. CROWLEY ET AL.
number of features which may limit their accuracy and computational ef-
ciency, as described in what follows. A method that attempts to meet, in
a harmonised fashion, the two fundamental requirements of accuracy and
computational efciency for reliable loss assessments, is proposed herein.
1.1. Limitations of the current methods for earthquake loss
estimation
Traditionally, the assessment of damage for loss estimation studies has been
based on macroseismic intensity or peak ground acceleration (PGA). Both
parameters, however, have their shortcomings: intensity, although directly
related to building damage (Musson, 2000), is erroneously treated as a con-
tinuous variable in predictive relationships when in fact it is a discrete
index with non-uniform intervals, whilst PGA shows almost no correla-
tion with the damage potential of the ground motion. In addition, neither
parameter accounts for the relationship between the frequency content of
the ground motion and the dominant period of the buildings. Nonetheless,
these parameters are typically applied in damage matrix methods such as
that developed by the Applied Technology Council (ATC, 1985) wherein
damage ratios or factors, dened as the ratio between the cost of repair
and the replacement value of the structure, are related to the intensity of
shaking through the post-processing of eld data collected following dam-
aging earthquakes. The development of the damage matrices is subjective,
however, since the determination of the intensity of shaking, as well as the
level of observed damage in a structure, are based on expert opinion and
thus cannot be judged as exact procedures. Another pitfall in this approach
is that changing practices in construction may make observations of past
events of little relevance to the prediction of damage in future earthquakes.
Furthermore, the validity of applying statistics gathered from events that
may be fundamentally distinct from the area under assessment, both in
terms of seismic demand and supply, is debatable.
In order to compensate for the aforementioned shortcomings in tra-
ditional loss estimation procedures, recent proposals (e.g., Calvi, 1999;
FEMA, 1999) have made use of response spectra, in particular the
displacement (or accelerationdisplacement) spectrum, to represent the
destructive capacity of the ground motion. The rationale for using dis-
placement spectra in assessment arises from the movement towards defor-
mation-based philosophy in seismic design, which reects the much closer
correlation of displacements, as opposed to transient forces, with structural
damage.
In the HAZUS methodology (Kircher et al., 1997; FEMA, 1999), the
performance point of a building type under a particular ground shaking
scenario is found from the intersection of an accelerationdisplacement
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 175
spectrum, representing the ground motion, and a capacity spectrum
(pushover curve), representing the horizontal displacement of the structure
under increasing lateral load. This performance point provides the displace-
ment input into limit state vulnerability curves to give the probability of
exceeding the given damage band. A potential weakness in the approach is
the difculty in obtaining a physically realistic representation of the inelas-
tic response of the structure using pushover analysis. Although this aspect
can be somewhat improved using displacement-based adaptive pushover
techniques (Antoniou and Pinho, 2004), a faithful representation of the real
structural behaviour requires a great deal of information about the struc-
ture, including reinforcement details, which are unlikely to be well known
for a large building stock. Another feature of the method is that the capac-
ity curves published in the HAZUS manual are only available for buildings
in the USA having a limited range of storey heights, thus the application
of this method to other parts of the world requires additional research to
be carried out, although, of course, any method requires the gathering of
local data (e.g. Bommer et al., 2002).
Loss estimation methods are generally demanding in terms of time,
computing power and required input data. The HAZUS methodology was
originally derived not for probabilistic loss estimation but as a tool for
estimating the impact of individual earthquake scenarios. The method has
been adapted to use with models of earthquakes derived from probabilistic
seismic hazard assessment (PSHA), as in FEMA 366 (FEMA, 2001), but
it is preferable, as discussed by Bommer et al. (2002), to represent the seis-
mic demand by triggering a large number of earthquake scenarios that are
compatible in magnitude, location and associated frequency of occurrence
with the regional seismicity. However, Bommer et al. (2002) also demon-
strated that this approach becomes extremely demanding in terms of com-
putational effort: the earthquake loss model developed for Turkey using an
adaptation of the HAZUS approach had to be limited to just over 1000
scenarios for the entire country in order to reduce computer run times to
acceptable levels.
Following the long-established tradition in earthquake loss modelling for
insurance purposes initiated 30 years ago at UNAM, in Mexico City, by
Emilio Rosenbleuth and Luis Esteva, Ordaz et al. (2000) present a prob-
abilistic method for earthquake loss estimation that uses both accelera-
tion response spectra and a drift-based damage parameter. The method
uses both analytical and empirical relationships to dene the vulnerabil-
ity of realistic structural models and can account for the exibility of
foundations. In addition, the authors have extended the method to full loss,
rather than just damage, calculations. In the method of Ordaz et al. (2000)
the seismic demand is obtained using hazard maps derived from PSHA as
opposed to the use of scenario earthquakes.
176 H. CROWLEY ET AL.
In this paper a proposal for a displacement-based vulnerability assess-
ment procedure is presented, which is particularly suitable for an earth-
quake loss model owing to its computational efciency, without loss of
accuracy. The more physical model underlying the new approach is also
likely to represent an additional improvement with respect to existing
methodologies.
1.2. Proposed methodology
The most up-to-date version of a displacement-based method for seismic
vulnerability assessment, rst proposed by Pinho et al. (2002) and subse-
quently developed by Glaister and Pinho (2003), is presented herein. Fur-
thermore, an implementation strategy, as well as further developments, are
also provided, thus bringing the method one step closer to practical appli-
cation.
The procedure uses mechanically derived formulae to describe the dis-
placement capacity of classes of buildings at three different limit states.
These equations are given in terms of material and geometrical proper-
ties, including the average height of buildings in the class. By substitu-
tion of this height through a formula relating height to the limit state
period, displacement capacity functions in terms of period are attained; the
advantage being that a direct comparison can now be made at any period
between the displacement capacity of a building class and the displacement
demand predicted from a response spectrum. The original concept is illus-
trated in Figure 1, whereby the range of periods with displacement capacity
below the displacement demand is obtained and transformed into a range
of heights using the aforementioned relationship between limit state period
and height. This range of heights is then superimposed onto the cumula-
tive distribution function (CDF) of building stock to nd the proportion
of buildings failing the given limit state.
The inclusion of a probabilistic framework into the method that was
lacking in the original proposal (Pinho et al., 2002) has allowed for a con-
sideration of the uncertainty in the displacement demand spectrum and the
uncertainty in the displacement capacity that arises when a group of build-
ings, which may have different geometrical and material properties, is con-
sidered together. The addition of this probabilistic framework, however, has
meant that the simple graphical procedure outlined in Figure 1 that treated
the beam- or column-sway RC building stock as single building classes can
no longer be directly implemented, but instead, separate building classes
based on the number of storeys need to be dened; this issue is addressed
further in Section 3.4.
The aleatory variability in the demand is modelled using the widely
accepted assumption of a lognormal distribution of residuals (e.g.,
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 177
Height
cumulative
frequency
HLS1 HLS2
HLS3
PLS3
0
PLS2
PLS1
PLSi percentage of
buildings failing LSi
effective
period
d
i
s
p
l
a
c
e
m
e
n
t
LS1
LS2
LS3
Demand
Spectra
TLS1 TLS2 TLS3
HLSi =f (TLsi , LSi)
LS1
LS2
LS3
Figure 1. A deformation-based seismic vulnerability assessment procedure (Glaister
and Pinho, 2003). LS stands for limit state.
Restrepo-Velez and Bommer, 2003), whilst modelling of the displacement
capacity uncertainty requires a slightly more sophisticated approach: the
use of a rst-order reliability method (FORM). FORM can be used to cal-
culate the approximate CDF of a non-linear function of correlated random
variables. Once the CDF of the demand and the capacity have been found,
the calculation of the probability of exceedance of a specied limit state
can be obtained using the standard time-invariant reliability formulation
(e.g. Pinto et al., 2004). The probability of being in a particular damage
band may then be found from the difference between the bordering limit
state exceedance probabilities.
The authors believe that the use of the method described in this paper
leads not only to a more computationally efcient process of earthquake
loss estimation, with the possibility to calculate the losses from multiple-
scenario earthquakes, but also to a method that can be easily adapted to
suit the varied construction and design practices around the world, owing
to its transparent means of building class vulnerability assessment.
2. Deterministic Implementation of Proposed Methodology
2.1. Classication of buildings
The initial step required in this method is the division of the building pop-
ulation into separate building classes. A building class is to be considered
as a group of buildings which share the same construction material, failure
mechanism and number of storeys. The building classes currently consid-
ered within this methodology comprise the following structural types:
(1) reinforced concrete beam-sway moment resisting frames,
178 H. CROWLEY ET AL.
(2) reinforced concrete column-sway moment resisting frames,
(3) reinforced concrete structural wall buildings,
(4) un-reinforced masonry buildings exhibiting an out-of-plane failure
mechanism,
(5) un-reinforced masonry buildings exhibiting an in-plane failure mechanism.
Within each structural type, further building classes may be dened to
separate, for example, buildings with different number of storeys, buildings
designed with distinct steel grades or those built without adequate conn-
ing reinforcement. A decision regarding whether a moment resisting frame
will exhibit a beam-sway (class 1) or a column-sway (class 2) mechanism
may be made considering the construction type, construction year and
evidence of a weak ground oor storey. Many buildings built before the
inclusion of sound seismic design philosophy (i.e. capacity design) into a
countrys seismic design code and those with a weak ground oor storey
will generally adopt a soft-storey (column-sway) mechanism. The treatment
of classes 4 and 5, relating to un-reinforced masonry structures, have been
dealt with by Restrepo-Velez and Magenes (2004) in an independent effort
and will not be considered further in this study.
2.2. Structural and non-structural limit states
Damage to the structural (load-bearing) system of the building class is esti-
mated using three limit states of the displacement capacity. The building
class may thus fall within one of four discrete bands of structural damage:
none to slight, moderate, extensive or complete. A qualitative description of
each damage band for reinforced concrete frames is given in Table I along
with quantitative suggestions for the denition of the mechanical material
properties for each limit state taken from the work of Priestley (1997) and
Calvi (1999). The rst structural limit state is dened as the yield point of
the structure and the second and third structural limit states are attained
when the sectional steel and concrete strains reach the limits suggested in
Table I. Two alternative pairs of limit state 3 sectional strains have been
reported because the ultimate sectional strains that can be reached depend
on the level of connement of the structural members. Nevertheless, it
should be noted that one is not constrained to employ these limit state steel
and concrete strains and has the ability to control these, and other, param-
eters used in the building class capacity calculations.
Damage to non-structural components within a building can be con-
sidered to be either drift- or acceleration-sensitive (Freeman et al., 1985;
Kircher et al., 1997). Drift-sensitive non-structural components such as
partition walls can become hazardous through tiles and plaster spalling
off the walls, doors becoming jammed and windows breaking. Acceler-
ation-sensitive non-structural components include suspended ceilings and
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 179
Table I. Description of RC frame structural discrete damage bands
Structural damage band Description
None to slight Linear elastic response, exural or shear type hairline cracks
(<1.0 mm) in some members, no yielding in any critical sec-
tion; hence limit state to damage band is structural yield
point
Moderate Member exural strengths achieved, limited ductility devel-
oped, crack widths reach 1.0 mm, initiation of concrete spall-
ing, limits to strains may be assumed as:
c
=0.0040.005

s
= 0.0100.015
Extensive Signicant repair required to building, wide exural or shear
cracks, buckling of longitudinal reinforcement may occur,
limits to strains may be assumed as:
Inadequately conned
members:
Adequately conned
members:

c
= 0.0050.010
c
= 0.0100.020

s
= 0.0150.030
s
= 0.0400.060
Complete Repair of building not feasible either physically or economi-
cally, demolition after earthquake required, could be due to
shear failure of vertical elements or excess displacement
building contents. At present only drift-sensitive non-structural damage is
considered within this methodology, using three limit states of drift capac-
ity. The non-structural components will again fall within one of four bands
of damage: undamaged, moderate, extensive or complete. These damage
bands differ slightly from their structural counterparts in that no damage
at all occurs in the rst band to non-structural damage.
Inter-storey drift can be used to predict drift-sensitive non-structural dam-
age. Freeman et al. (1985) report that studies on dry wall partitions indi-
cate an initial damage threshold at a drift ratio of 0.25%, and a thresh-
old for signicant damage at drift ratios between 0.5% and 1.0%. How-
ever, to ensure three non-structural limit states, the suggestions given by
Calvi (1999) may be followed, as given in Table II. It should be appreciated that
these drift ratios depend on the type and quality of the non-structural compo-
nents and so, again, one may chose alternative values for these parameters.
2.3. Structural and non-structural displacement capacity as a
function of height
As stated earlier, a simple denition of the displacement capacity of
building classes using mechanical material properties and concepts under
180 H. CROWLEY ET AL.
Table II. Description of non-structural discrete damage bands
Non-structural damage band Description
Undamaged No damage to any non-structural element, damage
assumed to initiate at drift ratios between 0.1% and 0.3%,
but may depend on quality of partitions
Moderate To maintain moderate, easily repairable damage to non-
structural elements, drift ratios should not exceed 0.3%
0.5%
Extensive Extensive damage to non-structural elements, to ensure
damage is reasonably repairable, drift ratios should not
exceed the range of 0.51.0%
Complete Repair of non-structural elements not feasible, exceedance
of extensive damage drift ratio limits
different limit states, is the basis of this methodology. Structural displace-
ment capacity formulae for all of the building classes described in Section
2.1 have been, or are in the process of being, derived, but only the beam-
sway and column-sway failure mechanisms of reinforced concrete frames
(classes 1 and 2) shall be presented herein. The derivation of displacement
capacity formulae for structural wall buildings (class 3) is currently under-
way. Whilst a more thorough description of the origin of the structural dis-
placement capacity formulae for classes 1 and 2 can be found in Glaister
and Pinho (2003), important developments have been carried out since the
original derivation of these equations, such as the inclusion of a robust for-
mula to relate the yield period of a RC frame to its height, and the deriva-
tion of non-structural displacement capacity formulae, as will be discussed
presently.
2.3.1. Displacement capacity at the centre of seismic force
(i) Beam-sway frames
As stated previously, the demand in this methodology is represented by a
displacement spectrum which can be described as providing the expected
displacement induced by an earthquake on a single degree of freedom
(SDOF) oscillator of given period and damping. Therefore, the displace-
ment capacity equations that are derived must describe the capacity of a
SDOF substitute structure and hence must give the displacement capac-
ity, both structural and non-structural, at the centre of seismic force of the
original structure.
The displacement capacity at the centre of seismic force is dealt with
in two different ways in this method depending on whether it is the limit
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 181
state base rotation/drift or the roof deformation of the original structure
that needs to be predicted.
In the structural displacement capacity equations, presented in Section
2.3.2, a base rotation can be mechanically derived for both beam- and
column-sway frames and the displacement at the centre of seismic force
is given by multiplying this rotation by an effective height. The effective
height is calculated by multiplying the total height of the structure by an
effective height coefcient (ef
h
), dened as the ratio of the height to the
centre of mass of a SDOF substitute structure (H
SDOF
), that has the same
displacement capacity as the original structure at its centre of seismic force
(H
CSF
), and the total height of the original structure (H
T
), as schematically
shown in Figure 2.
For beam-sway frames, the ratio of H
CSF
to H
T
varies with the height,
independently of ductility, from 0.67 for frames less than 4 storeys high to
0.61 for frames with more than 20 storeys; however, it has been suggested
by Priestley (1997) that, for regular structures, an average ratio of 0.64 may
be taken, irrespective of building height. The effective height coefcient can
then in turn be dened as a function of the number of storeys n using the
following equations, as suggested by Priestley (1997):
ef
h
=0.64 n4 (1)
ef
h
=0.640.0125(n4) 4<n<20 (2)
ef
h
=0.44 n20 (3)

H
T

H
CSF

H
SDOF

T
SDOF
h
H
H
ef =

Deformation profile of
actual structure
Deformation profile of
equivalent SDOF
Figure 2. Denition of effective height coefcient (Glaister and Pinho, 2003).
182 H. CROWLEY ET AL.
In the derivation of the non-structural displacement capacity equa-
tions for beam-sway frames, the effective height coefcient cannot be used
directly because, rather than mechanically deriving a base rotation capacity,
as in the structural displacement capacity formulation, it is the roof defor-
mation capacity that is directly obtained, as will be described in Section
2.3.3.
Hence a relationship between the deformation at the roof and the defor-
mation at the centre of seismic force is required. The factor relating these
two displacements shall be named a shape factor (S) and it can be found
from the displacement proles suggested by Priestley (2003) for beam-sway
frames of various heights (Figure 3), where, as above, the elastic and inelas-
tic proles are assumed to be equivalent.
The shape factor at the centre of seismic force can be found directly
from Figure 3 using an assumed ratio of the height to the centre of seis-
mic force (H
CSF
) to the total height (H
T
) of 0.64, as suggested previously.
Thus it can be seen in Figure 3 that the displacement at H
CSF
varies from
around 0.64 to 0.85 times the roof displacement depending on the number
of storeys.
(ii) Column-sway frames
As stated previously, the structural displacement capacity formulae are
derived by multiplying a base rotation by an effective height coefcient.
For column-sway frames, the elastic and inelastic deformed shapes vary
from a linear prole for elastic (pre-yield) limit states to a non-linear
prole at inelastic (post-yield) limit states (Figure 4). As suggested by
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1
Shape Factor
H
e
i
g
h
t

r
a
t
i
o

(
H
i
/
H
n
)
n < 4
n = 8
n = 12
n = 16
n > 20
efh = 0.64
Figure 3. Displacement proles for beam-sway frames for varying number of
storeys, n.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 183
0 0.2 0.4 0.6 0.8 1
Displacement ratio
elastic
inelastic
Height
of
ground
floor
Height to
centre of
seismic
force

sy1st

p

p

sy

i
H
e
i
g
h
t

Figure 4. Elastic and inelastic deformed shapes of column-sway frames with ground
oor drift capacity
1
.
Priestley (1997), the linear prole at pre-yield limit states means that the
ratio of H
CSF
to H
T
can be assumed to be 0.67 and so this is to be taken
as the effective height coefcient.
At post-yield limit states, the height to the centre of seismic force of a
column-sway frame is dependent on the ductility (
Lsi
) and decreases from
a low ductility value of 0.67 to a high ductility value of 0.5, as inferred
from Figure 4 and captured in the following equation, rst proposed by
Priestley (1997) and then adapted by Glaister and Pinho (2003):
ef
h
=0.0670.17

Lsi
1

Lsi
(4)
The ductility cannot be calculated, however, unless the yield displace-
ment at the effective height is known, thus leading to an iterative proce-
dure to nd the effective height. Glaister and Pinho (2003) proposed that,
for the sake of simplicity, a formula similar to Eq. (4) could be used where,
instead of ductility, the steel strain
s(Lsi)
corresponding to a given limit
state is used, as presented in Eq. (5).
ef
h
=0.0670.17

s(Lsi)

s(Lsi)
(5)
For the derivation of the non-structural capacity, the inter-storey drift
capacity of the ground oor,
i
, is equated to a base rotation, as will be
described in Section 2.3.3, and so the effective height coefcient is required
to nd the displacement capacity at the centre of seismic force. For pre-
yield limit states, this coefcient will be equivalent to that used in the
184 H. CROWLEY ET AL.
structural displacement capacity formulae described above (i.e. 0.67H
T
). At
post-yield limit states, (that is, when the non-structural limit state exceeds
the structural yield limit state), it is proposed that an initial effective height
of 0.6H
T
is assumed in order to estimate the structural yield displace-
ment and corresponding ductility. This resulting ductility is then input into
Eq. (4) to obtain a better estimate of the effective height coefcient; only
one iteration is required to arrive at a stable converged solution.
2.3.2. Structural displacement capacity vs height
By considering the yield strain of the reinforcing steel and the geometry of
the beam and column sections used in a building class, yield section curva-
tures can be dened using the relationships suggested by Priestley (2003).
These beam and column yield curvatures are then multiplied by empirical
coefcients to account for shear and joint deformation to obtain a formula
for the yield chord rotation. This chord rotation is equated to base rotation
and multiplied by the total building height and an effective height coef-
cient, as introduced in Section 2.3.1, to produce the yield displacement
capacity of a SDOF substitute structure. Sound, rational and deformation-
based equations of displacement capacity can thus be derived through rst
principles and mechanical considerations.
The yield displacement capacity formulae for beam- and column-sway
frames are presented in Eqs. (6) and (7), respectively; these are used to
dene the rst structural limit state.

Sy
=0.5ef
h
H
T

y
l
b
h
b
(6)

Sy
=0.43ef
h
H
T

y
h
s
h
c
(7)
The parameters employed in these and subsequent equations are
described below:

Sy
structural yield (limit state 1) displacement capacity,
ef
h
effective height coefcient, as dened in Section 2.3.1,
H
T
total height of the original structure,

y
yield strain of the reinforcement,
l
b
length of beam,
h
b
depth of beam section,
h
s
height of storey,
h
c
depth of column section,
Post-yield displacement capacity formulae are obtained by adding a
plastic displacement component to the yield displacement, calculated by
multiplying together the limit state plastic section curvature, the plas-
tic hinge length, and the height or length of the yielding member. The
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 185
post-yield displacement capacity formulae for RC beam- and column-sway
frames are presented here in Eqs. (8) and (9), respectively. In this formu-
lation, the soft-storey of the column-sway mechanism is assumed to form
at the ground oor. Straightforward adaptation of the equations could eas-
ily be introduced in the cases where the soft-storey is expected to form at
storeys other than the ground oor, but this is not dealt with herein.

SLsi
= 0.5ef
h
H
T

y
l
b
h
b
+0.5
_

C(Lsi)
+
S(Lsi)
1.7
y
_
ef
h
H
T
(8)

SLsi
= 0.43ef
h
H
T

y
h
s
h
c
+0.5
_

C(Lsi)
+
S(Lsi)
2.14
y
_
h
s
(9)
where,
SLsi
is the structural limit state i (2 or 3) displacement capacity,

C(Lsi)
, maximum allowable concrete strain for limit state i,
S(Lsi)
, maxi-
mum allowable steel strain for limit state i.
Formulae for the ductility (
SLsi
) of beam- and column-sway frames are
shown in Eqs. (10) and (11), respectively. A detailed account of the deriva-
tion of Eqs. (6)(11) can be obtained from the work of Glaister and Pinho
(2003).

SLsi
= 1+
_

C(Lsi)
+
S(Lsi)
1.7
y
_
h
b

y
l
b
(10)

SLsi
= 1+
_

C(Lsi)
+
S(Lsi)
2.14
y
_
h
c
0.86ef
h
H
T

y
(11)
An important development that will need to be included in the meth-
odology is the calculation of the shear capacity of the structure, to ensure
that shear failure does not occur before the exural displacement capacity
is reached. Within the purely displacement-based framework of the method
it would be most convenient for such a shear capacity check to be car-
ried out through comparison between the displacement demand and shear
capacity of reinforced concrete members. The recent work of Miranda
(2004), where formulae for the shear displacement capacity of members
have been derived by relating their shear force capacity to a displace-
ment capacity, will be used in the future developments of this proposed
vulnerability assessment method.
2.3.3. Non-structural displacement capacity vs height
Non-structural displacement capacity is found from the inter-storey drift
capacity of the non-structural components, such as partition walls. Exam-
ples of the limit state drift ratios have been described previously in Table II.
For beam-sway frames, the non-linear displaced shape leads to a var-
iation in inter-storey drift from the ground oor to the roof. However,
186 H. CROWLEY ET AL.
by multiplying the drift ratio capacity by the total height of the build-
ing, a roof displacement capacity corresponding to the average inter-sto-
rey drift capacity is attained. The non-structural displacement capacity of
the SDOF substitute structure, as introduced in Section 2.3.1, can thus be
found by multiplying the roof displacement by the shape factor to give the
displacement at the centre of seismic force of the structure, as presented in
Eq. (12).

NSLsi
=S
i
H
T
(12)
where
NSLsi
is the non-structural limit state i displacement capacity, S is
the shape factor giving the ratio of the deformation at the effective height
to the roof deformation, described in Section 2.3.1,
i
the limit state i drift
ratio capacity.
For column-sway frames, the potential for concentration of non-
structural damage at the ground oor should be considered, as illustrated
previously in Figure 4. Thus it is assumed that once the rst oor reaches
the limit state inter-storey drift capacity then the non-structural damage
limit state has been attained. Therefore it should be ascertained whether
the displacement at the rst oor (
NS1st
), given in Eq. (13) by multiplying
the inter-storey drift with the storey height, is greater than the rst oor
structural yield displacement (
Sy1st
), found by multiplying the yield base
rotation by the height of the rst storey.

NS1st
=
i
h
s
(13)
If
NS1st
is lower than
Sy1st
, the non-structural displacement capa-
city at the centre of seismic force at this pre-yield limit state can sim-
ply be given by Eq. (12) with S =0.67 due to the linear deformed shape,
dened in Figure 4. If
NS1st
is higher than
Sy1st
, then the post-yield
non-structural displacement capacity of the SDOF substitute structure can
be found by the following steps. The plastic component of the displacement
(
p
) may be calculated by subtracting the yield displacement at the rst
storey (
Sy1st
) from
NS1st
.

p
=
NS1st

Sy1st
=
i
h
s
0.43h
s

y
h
s
h
c
(14)
This plastic component (
p
) may then be added to the yield displace-
ment at the centre of seismic force (
Sy
) to obtain the non-structural limit
state displacement capacity of the SDOF substitute structure (
NSLsi
), as
illustrated in Figure 4. As has been discussed in Section 2.3.1, it is sug-
gested that an effective height coefcient be calculated using Eq. (4), where
the ductility may be rst estimated for an initial guess of the yield dis-
placement at the centre of seismic force found with an effective height of
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 187
0.6H
T
, and then iterated once for a nal solution. The formula for the
non-structural limit state displacement capacity of the SDOF substitute
structure for a column-sway frame, failing in the rst storey and having
entered the non-linear range, is thus presented in Eq. (15).

NSLsi
=
p
+
Sy
=
i
h
s
+0.43(ef
h
H
T
h
s
)
y
h
s
h
c
(15)
To summarise, the non-structural displacement capacity of the SDOF
substitute structure may be calculated for beam-sway frames using Eq. (16)
where S can be found from Figure 2, assuming a H
CSF
to H
T
ratio of 0.64.
The non-structural displacement capacity of column-sway frames for limit
states before structural yielding, ascertained at the rst oor, may be found
using Eq. (17) and for limit states after structural yielding at the rst oor,
using Eq. (18).

NSLsi
=S
i
H
T
(16)

NSLsi
=0.67
i
H
T
(17)

NSLsi
=
i
h
s
+0.43(ef
h
H
T
h
s
)
y
h
s
h
c
(18)
2.4. Period of vibration of buildings as a function of height
Simple empirical relationships are available in many design codes to relate
the fundamental period of vibration of a building to its height. However,
these relationships have been realised for force-based design and so produce
lower bound estimates of period such that the base shear force will be con-
servatively predicted. Hence the displacement demand on a structure needs
to be accurately estimated; however with a conservative periodheight rela-
tionship the displacement demand would generally be under-predicted. The
use of a reliable relationship between period and height is a fundamental
requirement in this methodology, so that the displacement capacity formu-
lae can be accurately dened in terms of period and directly compared with
the displacement demand.
Glaister and Pinho (2003) recognised the need for a sound relation-
ship between period and height that would be valid throughout the entire
displacement range. However, in the absence of such a relationship, they
used a modied version of the suggested formula given in EC8 (CEN,
2003). A suitable relationship between yield period and height has since
been derived by Crowley and Pinho (2004), which can be easily related
to inelastic period as will be shown in the subsequent sections. The
pre- and post-yield structural displacement capacity formulae given in
188 H. CROWLEY ET AL.
Glaister and Pinho (2003) in terms of period have thus been updated, as
will be presented in Section 2.5.
2.4.1. Yield period
Crowley and Pinho (2004) describe how analytical procedures have been
used to obtain the yield period of European RC buildings designed before
the inclusion of capacity design in the design codes. Eigenvalue, pushover
and dynamic analyses have all been employed in the yield period determi-
nation for many buildings of various heights. Regression analysis of the
data has led to a group of best-t yield periodheight curves that are in
general agreement despite having been derived from different theoretical
backgrounds. Hence there is a high degree of condence in the results
obtained which then lead to a straightforward choice of a linear yield
period vs. height (H
T
in metres) formula for European RC moment resist-
ing frames, presented in Eq. (19):
T
y
=0.1H
T
(19)
2.4.2. Post-yield period
For post-yield limit states, the limit state period of the substitute structure,
as introduced in Section 2.3.2, can be obtained from the secant stiff-
ness to the point of maximum deection on an idealised bi-linear force
displacement curve as described already in Glaister and Pinho (2003) but
repeated here for the sake of clarity. Assuming an elasto-plastic force
displacement relationship, the secant stiffness to the point of maximum
deection (k
Lsi
) can be shown to be a geometric function of the elastic
stiffness (k
y
) and ductility (
Lsi
) only. Since the elastic period (T
y
) is also
a function of elastic stiffness, it can be assumed that the effective period
(T
LSi
) of the inelastic structure is a function of elastic period and ductil-
ity alone. Eqs. (20)(23) show the working through of these premises and
the resulting equation relating effective period at a limit state i with the
corresponding ductility level and the elastic period, independent of the fail-
ure mechanism:
f
y
=k
y

y
=k
Lsi

Lsi
(20)
k
Lsi
=
k
y

Lsi
=
k
y

Lsi
(21)
T k
1/2

1/2
(22)
T
Lsi
=T
y

Lsi
(23)
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 189
2.5. Structural and non-structural displacement capacity as a
function of period
2.5.1. Structural displacement capacity vs period
(i) Pre-yield
The derivation of a relationship between period and height that is valid
for all limit states allows the previous displacement capacity formulae pre-
sented in Glaister and Pinho (2003) to be developed into conceptually
sound functions of period. For the rst (yield) limit state, the building
height may be simply dened in terms of the yield period by rearranging
Eq. (19) as follows:
H
T
=10T
y
(24)
In the case of beam-sway RC frames, the yield capacity equations can be
obtained by substituting the height in Eq. (6) (the formula for the yield dis-
placement capacity in terms of height) with the formula in Eq. (24) above:

Sy
=5ef
h
T
y

y
l
b
h
b
(25)
For column-sway RC frames, the yield displacement equation is also
simply transformed into a function of period by substituting Eq. (24) into
Eq. (7).

Sy
=4.3ef
h
T
y

y
h
s
h
c
(26)
(ii) Post-yield
For the post-yield structural limit states (2 and 3), the height of the build-
ing needs to be written in terms of the post-yield period. For beam-sway
frames the height is simply given by rearranging Eq. (23) to give the
formula shown in Eq. (27):
H
T
=
10T
Lsi

SLsi
(27)
The post-yield displacement capacity in terms of post-yield period is
then found by replacing the height in Eq. (8) (the formula for the post-
yield displacement capacity in terms of height) with Eq. (27), to give the
following formula:

SLsi
=5ef
h
T
Lsi

y
l
b
h
b

Lsi
(28)
For the post-yield limit states of column-sway frames, the resulting for-
mula for the height has a slightly more complicated form as compared to
190 H. CROWLEY ET AL.
beam-sway frames due to the dependence of the ductility on the height,
(see Eq. (11)):
H
T
=
1
2
_
C
l
+(C
2
l
+400T
2
Lsi
)
1/2
_
(29)
where
C
l
=
_

c
+
s
2.14
y
0.86
y
_
h
c
ef
h
The post-yield displacement capacity in terms of post-yield period, pre-
sented in Eq. (30), is thus obtained by replacing the height in Eq. (9) with
Eq. (29).

SLsi
=0.215ef
h

y
h
s
h
c
(C
2
l
+400T
2
Lsi
)
1/2
+0.25(
c
+
s
2.14
y
)h
s
(30)
2.5.2. Non-structural displacement capacity vs period
(i) Pre-yield
The initiation of non-structural damage can be condently assumed to
occur before structural yielding, at a drift ratio
1
. The relationship
between the height and yield period of Eq. (24) is also used in the substitu-
tion of height for period in the non-structural displacement capacity equa-
tions. The rst limit state non-structural displacement capacity in terms
of period is thus presented in Eq. (31), where S can be obtained from
Figure 3 for beam-sway frames and may be taken as 0.67 for column-sway
frames, as has been described in Section 2.3.1.

NS
=S
1
(10T
y
) (31)
(ii) Post-yield
The moderate and signicant non-structural damage drift limit ratios,

2
and
3
respectively, may or may not occur before structural yield-
ing and so this check needs to be carried out. For beam-sway frames,
if the moderate or signicant non-structural damage displacement capac-
ity is less than the structural yield displacement capacity at the centre of
seismic force, then Eq. (31) above can be used. However, if these displace-
ments are higher than the yield displacement, then the yield period can no
longer be applied. Instead, the height should be substituted using Eq. (27),
where the ductility (
Lsi
) of the beam-sway frames can be calculated from
the ratio between the moderate/signicant non-structural damage displace-
ment capacity and the structural yield displacement:

NSLsi
=

NSLsi

Sy
=
S
i
H
T

Sy
=
S
i
0.5ef
h

y
l
b
/h
b
i =2, 3 (32)
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 191
The nal equation for the non-structural displacement capacity of
beam-sway frames in terms of the inelastic period is found by replacing H
T
,
as dened in Eq. (27), in Eq. (12) to give:

NSLsi
=S
i
_
10T
Lsi

NSLsi
_
=S
i
(10T
Lsi
)
_
0.5
y
ef
h
l
b

i
h
b
S
i =2, 3 (33)
For column-sway frames, if the non-structural displacement at the rst
storey is greater than the yield displacement, then the height of the struc-
ture should be calculated using the post-yield period, as presented previ-
ously in Eq. (27), where the ductility can be found using Eq. (34), using

p
computed from Eq. (14). The effective height coefcient in Eq. (34) is
initially taken as 0.6 to nd the ductility and then Eq. (4) is used to nd
a better estimate of the effective height coefcient for further calculations.

NSLsi
=

NSLsi

Sy
=

p
+
Sy

Sy
=1+

p

Sy
=1+

p
0.43(ef
h
H
T
)
y
h
s
/h
c
(34)
The height can then be represented in terms of inelastic period, using
the formula shown below, which again is slightly more complicated
than the formula for beam-sway frames due to the dependence of the
ductility on the height:
H
T
=
1
2C
2
_

p
+(
2
p
+400C
2
2
T
2
Lsi
)
1/2
_
(35)
where,
C
2
=0.43ef
h

y
h
s
/h
c
The resulting formula for the non-structural displacement capacity in
terms of inelastic period is then found by substituting Eq. (35) into Eq.
(18):

NSLsi
=
1
2
_

p
+(
2
p
+400C
2
2
T
2
Lsi
)
1/2
_
(36)
2.6. Displacement demand
Displacement response spectra are used in this method to represent the
input from the earthquake to the building class under consideration. The
relationship between equivalent viscous damping () and ductility (), used
to account for the energy dissipated through hysteretic action at a given
level of ductility demand is presented in the following equation:
192 H. CROWLEY ET AL.
=a
_
1
1

b
i
_
+
E
(37)
where a and b are calibrating parameters which vary according to the char-
acteristics of the energy dissipation mechanisms, whilst
E
represents the
equivalent viscous damping when the structure is within the elastic, or pre-
yield, response range. It is recognised, however, that the level of energy dis-
sipation of a given structural system may depend on the characteristics of
the input such as duration and phase content, for which reason research is
currently underway to assess the manner in which Eq. (37) can be adjusted
or improved to include this inuence. In the meantime, values of a =25 and
b=0.5, as suggested by Calvi (1999), are adopted in Eq. (37), together with
an
E
=5%.
The equivalent viscous damping values obtained through Eq. (37), for
different ductility levels, can then be combined with Eq. (38), proposed by
Bommer et al. (2000) and currently implemented in EC8 (CEN, 2003), to
compute a reduction factor to be applied to the 5% damped spectra at
periods from the beginning of the acceleration plateau to the end of the
displacement plateau:
=
_
10
5+
(38)
Bommer and Mendis (2004) have investigated the dependence of the
ratio of displacement spectral ordinates for higher damping levels to the
ordinates at 5% of critical damping on features of the earthquake motion.
The ratios are shown to decrease with increasing magnitude and with
increasing distance, both observations being consistent with the ratios
decreasing as the duration of the ground shaking increases. In the proposed
procedure of using earthquake scenarios rather than probabilistic hazard
maps to model the demand, this renement of the prediction of the spec-
tral ordinates at higher damping levels can be easily incorporated.
2.7. Illustrative example of deterministic implementation
Many of the existing buildings in Europe have not been designed with
sound seismic design philosophy, hence, as has been discussed in Section
2.1, a large proportion may be assumed to behave with a column-sway fail-
ure mechanism. A deterministic example is provided herein to show how
the yield displacement capacity of column-sway frames varies with period
and how the failure of this building class can be ascertained through com-
parison with a displacement demand spectrum. The aim of this example is
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 193
Table III. Values used for the parameters
in the limit state 1 (yield) displacement and
period capacity equations for column-sway
frames
Parameter Value
Column depth, h
c
0.38 m
Storey height, h
s
3.22 m
Steel grade 275 yield strain,
y
0.165 %
merely to illustrate the workings of the deterministic method described thus
far.
Table III shows the values that have been assigned to the parameters
required to dene the yield displacement capacity of column-sway frames,
presented previously in Eq. (26). The geometrical data has been taken from
the mean values obtained from a study of European building stock data;
this is discussed further in Section 3.3.1. The reinforcing steel in this exam-
ple has a 5% characteristic strength of 275 MPa; the calculation of the
mean yield strain shown in Table III is described in Section 3.3.2. The dis-
placement demand spectrum used in this example is based on the 1992
Erzincan (Turkey) earthquake record, but the ordinates have been scaled to
20% of their original value, for the convenience of providing a clearer dem-
onstration of the intersection between the demand and capacity curves.
In Figure 5, the yield displacement capacity/demand curves for a
column-sway mechanism are given; the circles correspond to the
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
Period (s)
D
i
s
p
l
a
c
e
m
e
n
t
(
m
)

T
= 0.90 to 3.15 seconds

H
= 9.0 to 31.5 metres

T
0 1 2 3 4 5 6 7
Figure 5. Column-sway yield capacity and demand curves.
194 H. CROWLEY ET AL.
displacement capacity at a distinct number of storeys. As has been intro-
duced in Section 1.2, failure of the limit state is assumed to occur when the
displacement capacity curve falls below the displacement demand curve;
hence a probability of failure of unity when the capacity is below the
demand and zero when the capacity is above the demand. Thus it is appar-
ent from Figure 5 that with a deterministic approach, all column-sway
buildings responding at a yield period between 0.9 and 3.15 s would be pre-
dicted to fail the rst limit state. By using the relationship between yield
period and height described in Section 2.4.1, the height range of the build-
ings failing the limit state can be found to be between 9.0 and 31.5 m,
which corresponds to buildings between 3 and 10 storeys.
3. Probabilistic Framework
3.1. Overview
A large number of geometrical and material parameters can vary among
buildings within a given class. A fully probabilistic framework is thus
necessary, and has been applied to this method to account for the fol-
lowing sources of epistemic (knowledge-based) and aleatory (random)
uncertainty:
(1) The uncertainty concerning the geometrical and material properties of
a building class.
(2) The uncertainty regarding the denition of steel and concrete strains
reached at each limit state of structural damage.
(3) The uncertainty as to the drift rotations required to dene each limit
state of non-structural damage.
(4) The model uncertainty caused by the dispersion of the empirical coef-
cients used in the derivation of the displacement capacity formulae,
such as those used to dene the yield curvature, plastic hinge length
and yield period.
(5) The aleatory uncertainty in the estimation of the 5% damped response
spectrum. (It should be noted that the mean ductility is used to reduce
the 5% damped demand spectrum for higher limit states, using the
reduction factor that has been discussed previously. This assumption
has been made to simplify the method as otherwise the variability in
the demand would be dependent on the variability in the capacity).
The probability that the earthquake demand is greater than the capac-
ity of a building, and thus failure occurs, is given by the classical time-
invariant reliability formula (e.g. Pinto et al., 2004):
P
f
=
_

0
[1F
D
()]f
SC
()d (39)
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 195
where F
D
() is the CDF of the demand and f
SC
() is the probability
density function of the capacity, dened in terms of a particular damage
parameter (). The adaptation of this reliability formulation initially car-
ried out by Restrepo-Velez and Magenes (2004) to suit the methodology
described herein is shown in (40):
P
f
=
_
x
_
y
[1F
D
(x/T
Lsi
=y)]f

LSi
T
LSi
(x, y)dxdy (40)
where F
D
(x/T
LSi
=y) is the CDF of the displacement demand, x, given
a period, T
Lsi
and f

LSi
T
LSi
(x, y), is the joint probability density func-
tion (JPDF) of the limit state displacement capacity,
Lsi
, and limit state
period, T
Lsi
.
The JPDF, f

LSi
T
LSi
(x, y), may be dened as the product of the probabil-
ity density function of
Lsi
, conditioned to T
Lsi
, and the probability den-
sity function of T
Lsi
:
f

Lsi
T
Lsi
(x, y) =f

Lsi
(x/y)f
T
Lsi
(y) (41)
Thus the nal formulation for the calculation of the probability that the
displacement demand is greater than the displacement capacity of a build-
ing class, for a given limit state, is given by Eq. (42).
P
f
=
_
y
_
x
[1F
D
(x/T
Lsi
=y)]f

LSi
/T
LSi
(x/T
Lsi
=y)f
T
LSi
dxdy (42)
The inner integral in the above equation gives the probability that the
displacement demand is greater than the displacement capacity, condi-
tioned to a period, and so may be referred to as the conditional probability
of failure. Thus it may be read that Eq. (42) is the integral of the product
of the conditional probabilities of failure by the probabilities of the con-
ditioning events, over the full range of their possible intensities (Franchin
et al., 2002).
The JPDH can be used in conjunction with the demand CDF through
the use of the reliability formulation of Eq. (42) to nd the probability
of exceeding each of the three limit states described in Section 2.2. The
probability of a building class being in each of the four structural damage
bands, outlined in Table I, can then simply be found from the difference
between the exceedance probabilities of the bordering limit states to the
damage band in question. This probability is equated to the proportion of
buildings (P) falling within each damage band:
196 H. CROWLEY ET AL.
P
none/slight
=1P
f 1
(43)
P
moderate
=P
f 1
P
f 2
(44)
P
extensive
=P
f 2
P
f 3
(45)
P
complete
=P
f 3
(46)
The same process is also applied to nd the proportion of a building class
that falls within one of the four non-structural damage bands in Table II.
3.2. Probabilistic treatment of the demand
The CDF of the displacement demand can be found using the median dis-
placement demand values and their associated logarithmic standard devia-
tion at each period. The CDF gives the probability that the displacement
demand exceeds a certain value (x), given a response period (T
Lsi
) for a
given M-D scenario.
The displacement demand spectrum that might be used in a loss estima-
tion study could take the form of a code spectrum or else a uniform hazard
spectrum derived from PSHA for one or more annual frequencies of excee-
dance. Both of these options have drawbacks in being obtained from PSHA
wherein the contributions from all relevant sources of seismicity are com-
bined into a single rate of occurrence for each level of a particular ground-
motion parameter. The consequence is that if the hazard is calculated
in terms of a range of parameters, such as spectral ordinates at several
periods, the resulting spectrum will sometimes not be compatible with any
physically feasible earthquake scenario (Bommer, 2002). Furthermore, if
additional ground-motion parameters, such as duration of shaking, are to
be incorporated as they are in HAZUS, in the denition of the inelas-
tic demand spectrum then it is more rational not to combine all sources
of seismicity into a single response spectrum but rather to treat individ-
ual earthquakes separately, notwithstanding the computational penalty that
this entails.
The approach recommended therefore is to use multiple earthquake
scenarios, each with an annual frequency of occurrence determined from
recurrence relationships. For each triggered scenario, the resulting spectra
are found from a ground-motion prediction equation. In this way, the ale-
atory uncertainty, as represented by the standard deviation of the lognor-
mal residuals, can be directly accounted for in each spectrum. The CDF
of the displacement demand can then be compared with the JPDFs of
displacement capacity, using Eq. (42), and the annual probability of failure
for a class of buildings can be found by integrating the failure probabilities
for all the earthquake scenarios.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 197
The method proposed herein for vulnerability assessment can equally be
employed in conjunction with seismic demand obtained from probabilistic
hazard maps, provided that the aleatory variability of the ground motion
is then removed from the calculation of the probability of exceeding the
limit states. The authors also acknowledge that such an approach is sig-
nicantly more efcient in terms of computational effort. However, there
are many benets in using a multiple earthquake scenario approach, not
least amongst which is the facility to obtain clear and reliable disaggre-
gations of the calculated losses. The probabilistic implementation of the
method enables scenario-based loss calculations, which take full account of
the ground-motion variability, to be performed efciently.
3.3. Probabilistic treatment of the capacity
The probability density functions of the limit state displacement capacity
and period are found using the FORM. The reader is referred for example
to Pinto et al., (2004) for a description of the theory of FORM, as well
as Restrepo-Velez (2004) for a detailed description of the application of
FORM to the displacement capacity equations for un-reinforced masonry
structures. Essentially, FORM can be used to compute the approximate
CDF of a non-linear function of correlated parameters, such as the limit
state displacement capacity function and limit state period function.
As has been presented previously, the limit state displacement capac-
ity
Lsi
) of each building class can be dened as a function of the fun-
damental period (T
Lsi
), the geometrical properties of the building, and
the mechanical properties of the construction materials. Similarly, the limit
state period (T
Lsi
) of each building class can be dened as a function of the
height (or number of storeys), the geometrical properties of the building,
and the mechanical properties of the construction materials. The uncer-
tainty in
Lsi
and in T
Lsi
is accounted for by constructing a vector of
parameters that collects their mean values and standard deviations. By
assigning probability distributions to each parameter, FORM can be used
to nd both the CDF of the limit state displacement capacity, conditioned
to a period, and the CDF of the limit state period.
In the following section, the probability distributions suggested for each
parameter in the capacity equations are discussed. In the absence of data
from which the denition of the probabilistic distributions for the param-
eters could be obtained, the work of other researchers has been con-
sulted, as indicated below. Sufcient data to fully construct the matrix of
correlation coefcients between parameters are not available at present and
so the parameters are currently assumed to be uncorrelated. Where exten-
sive data are not available, it is apparent that statistical properties are often
based on engineering judgement. This identies an area where additional
198 H. CROWLEY ET AL.
research could be focused, but the authors believe that systematic and
comprehensive sensitivity studies should rst be carried out in order to
establish a hierarchy of priorities for renement of input parameters to
earthquake loss models.
3.3.1. Probabilistic modelling of geometrical properties
A given building class within a selected urban area may comprise a large
number of structures that present the same number of storeys and failure
mode, but that feature varying geometrical properties (e.g., beam height,
beam length, column depth, column/storey height), due to the diverse
architectural and loading constraints that drove their original design and
construction. Since such uncertainty does affect in a signicant manner
the results of loss assessment studies (see Glaister and Pinho, 2003), it
is duly accounted for in the current method by means of the probabilis-
tic modelling described below. Clearly, one could argue that by carrying
out a detailed inspection of the building stock, such variability could be
signicantly reduced (in the limit, if all buildings were to be examined,
it could be wholly eliminated), however at a prohibitive cost in terms of
necessary eld surveys and modelling requirements (vulnerability would
then be effectively assessed on a case-by-case basis).
The geometrical properties of buildings present also a random variabil-
ity, due to imperfections introduced at the construction phase, which
affects nominally identical structures. This aleatory variability in the geo-
metrical properties of reinforced concrete structural members, documented
by Mirza and MacGregor (1979a), amongst others, is much smaller in
magnitude than its epistemic counterpart described above (up to 20 times
smaller), for which reason its inuence in a loss assessment outcome is of
reduced importance. In addition, the inclusion of geometrical random var-
iability in the proposed methodology, although feasible, would increase sig-
nicantly the computation efforts involved. Therefore, only the epistemic
component of the geometrical variability of reinforced concrete members
has been modelled in the present work, as described in what follows.
Preliminary studies have been carried out to aid the somewhat demon-
strative scope of this presentation. The probability distribution functions
to describe the variability of geometrical properties in an urban environ-
ment have been studied using a database of 21 European buildings from
the following countries: Portugal, Italy, Greece, Romania, and Yugoslavia
(see Crowley, 2003). The recently designed buildings have been separated
from buildings designed before 1980; it is assumed that the latter have been
designed before the advent of sound seismic design philosophy and so can
be used to describe the parameters of column-sway frames. The geometric
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 199
Figure 6. Histogram to show the proportions of beam length found in a population of
recently designed (i.e. post-1980) European buildings and a normal distribution tted
to the data.
properties that have been obtained from this population of buildings com-
prise the following: beam length, beam depth, storey height and column
depth.
Normal or lognormal probability distributions have been tted to the
histograms produced from the data; an example is given in Figure 6 where
a normal distribution can be seen to describe fairly well the distribution of
beam length in recently designed structures.
The data used in this brief study is by no means extensive and fur-
ther data will be added as it becomes available to this ongoing research.
Nevertheless, the current values and probability distributions for the
geometric properties, which have been obtained from the aforementioned
European database, are presented in Tables IV and V, respectively for old
and recent buildings.
Table IV. Mean and standard deviation values and probability distribution
for the geometrical parameters from a database of old (i.e. pre-1980) Euro-
pean RC buildings
Parameter Mean (m) Standard deviation (m) Distribution
Beam length, l
b
4.02 1.14 Normal
Beam depth, h
b
0.44 0.06 Normal
Storey height, h
s
3.22 0.59 Lognormal
Column depth, h
c
0.38 0.14 Lognormal
200 H. CROWLEY ET AL.
Table V. Mean and standard deviation values and probability distribution for
the geometrical parameters from a database of recent (i.e. post-1980) RC
European buildings
Parameter Mean (m) Standard deviation (m) Distribution
Beam length, l
b
4.57 0.62 Normal
Beam depth, h
b
0.56 0.06 Normal
Storey height, h
s
3.00 0.12 Normal
Column depth, h
c
0.51 0.09 Lognormal
The values in Tables IV and V seem rational; for example the mean
beam length of older structures is shorter than newer structures (expected
since recent years have witnessed an increase in adopted spans) which then
accounts for the higher mean beam depth found in the newer structures
category. The mean column depth of older structures is lower than that
in newer structures due to the lack of consideration of capacity design in
the former. The standard deviations of the geometric properties in older
structures are generally higher than in newer structures, which would also
be expected as structures built to more recent design codes are more likely
to comply with prevalent dimension standards.
The mean values found for the older buildings in Table IV have
been used in the deterministic example application in Section 2.7 whilst
the mean values, standard deviations and probabilistic distributions in
Table IV are used in a probabilistic example application to be presented in
Section 3.4.
3.3.2. Probabilistic modelling of reinforcing bar yield strain
It will be assumed that once a probabilistic distribution for yield strength
has been found, it can be divided by a deterministic value of the modulus
of elasticity of 200 GPa to nd the distribution of the yield strains due to
the low coefcient of variation (CV) of this property in reinforcing steel
(Mirza and MacGregor, 1979b). Mirza and MacGregor (1979b) studied the
variability of the material properties of Grade 40 and Grade 60 reinforc-
ing bars using the test data available in North America. They concluded
that for the yield strength of the bars, a normal distribution correlated well
in the vicinity of the mean whilst a beta distribution correlated well over
the whole range of data. The CV in the yield strength was found to be
between 8% and 12% when data were taken from different bar sizes from
many sources. More recently, the Probabilistic Model Code (JCSS, 2001)
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 201
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24
Yield strain (%)
Yield strain (%)
P
D
F
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.1 0.15 0.2 0.25 0.3 0.35
P
D
F
275 MPa
325 MPa
400 MPa
(a)
(b)
Figure 7. The normal distribution of yield strain for reinforcing steel with an assumed
CV of 10% for (a) a 5% characteristic strength of 275 MPa alone and (b) 5% charac-
teristic strengths of 275, 325 and 400 MPa compared together.
has suggested that a normal distribution can be adopted to model the yield
strength of steel. A normal distribution for the steel yield strength (and
subsequently yield strain) will be used in this method.
202 H. CROWLEY ET AL.
Figure 7a shows an example of the normal probability density func-
tions of yield strain for reinforcing steel with a characteristic strength
of 275 MPa, dened as the strain that has a 95% probability of being
exceeded. The CV has been assumed to be 10% using the aforementioned
suggestions by Mirza and Macgregor (1979b) to account for the variability
in the strength of bars of different sizes and from different manufacturers.
Figure 7b shows the probability density functions of yield strain for
three different characteristic yield strengths, each with an assumed CV
of 10%. The mean yield strain obviously increases with the mean yield
strength, and as the CV is assumed equal for all steel strengths, the stan-
dard deviation (equal to the CV multiplied by the mean) thus increases
with increased strength. The shape of the three functions shown in
Figure 7b can be explained by considering that the dispersion increases
with strength but the area under the probability density function must
always be equal to 1.
The main difculty in assigning a probability distribution to the yield
strength of the steel used in a group of buildings, however, is the possibility
that different grades have been used which would lead to a distribution
with multiple peaks and troughs, as illustrated in the example in Figure 7b.
One approach to solve this problem could be to calculate the probabil-
ity of failure for the building class given each possible steel grade, using
the normal distribution to model the dispersion for each grade such as in
Figure 7a, and then a weighted average of failure can be found, knowing
or judging the use of each steel grade within the building class. The validity
of such an approach would become questionable, however, if different steel
grades were often used within individual buildings.
3.3.3. Probabilistic modelling of limit states threshold parameters
Dymiotis et al. (1999) have studied the seismic reliability of RC frames
using inter-storey drift to dene the serviceability and ultimate structural
limit states. They have found that a lognormal distribution may be used to
describe the variability in inter-storey drift for both limit states; the drift
ratios found from test specimens for the serviceability limit state are plot-
ted in Figure 8a as a histogram with the corresponding lognormal distribu-
tion superimposed. Kappos et al. (1999) report the ultimate concrete strain
reached in 48 tests of very well-conned RC members. A simple statistical
analysis of this data shows that it would appear that in the case of limit
state sectional strains a lognormal distribution is also able to describe the
variability (Figure 8b).
The non-structural limit states are dened in this method using inter-
storey drift. Considering it has been found by Dymiotis et al. (1999) that
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 203
Figure 8. (a) Histogram and suggested lognormal distribution of maximum experi-
mental inter-storey drifts at the structural serviceability limit state, reproduced from
Dymiotis et al. (1999) and (b) histogram and suggested lognormal distribution for the
ultimate concrete strain of very well-conned test specimens using data taken from
Kappos et al. (1999).
a lognormal distribution denes well the variability in the limit state inter-
storey drift for test specimens, a lognormal distribution will be assumed
herein, using the mean drift ratios that have been provided previously in
204 H. CROWLEY ET AL.
Table II. For the structural limit states, it is the sectional steel and concrete
strains that are used to dene the limit states in this method and it would
appear from Figure 8b that a lognormal distribution may also be applied
to describe the variability in these limit state parameters. The CV of the
limit state parameters can be seen from Figure 8a and b to be high and so
a value of 50% will be currently assumed until further data is available to
substantiate this assumption.
3.3.4. Probabilistic modelling of scatter in empirical relationships
A number of empirical relationships have been used to derive the functions
of displacement capacity and period that have been presented in Section 4.
These include expressions for the plastic hinge length members and the
yield curvature of RC members, all of which are discussed in Glaister and
Pinho (2003). An additional empirical relationship has since been added to
the methodology and that is the formula derived by Crowley and Pinho
(2004) to relate the height of the building to its yield period of vibration
that has been discussed in Section 2.4. All of the aforementioned relation-
ships rely on a given coefcient to relate one set of structural properties
to another, as for example the coefcient of 0.1 in the yield period ver-
sus height equation, T
y
=0.1H
T
. The mean value and standard deviation
of these coefcients have been taken from the studies carried out to derive
these formulae and a normal distribution is used to model the dispersion
in the coefcient.
As has been frequently noted in the literature, the use of a normal dis-
tribution for quantities that are non-negative is inappropriate; however, it is
also claimed that when the variability in the parameter is small in relation
to its mean, the probability of obtaining a negative quantity would be vir-
tually zero (Sasani and Der Kiureghian, 2001). Thus it has been decided
that a normal distribution will be used to model the dispersion in the
empirical coefcients in the example given in the following section.
3.4. Illustrative example of probabilistic implementation
The illustrative example introduced in Section 2.7 is considered again here
in a probabilistic sense and so the uncertainty in the displacement demand
spectrum as well as the dispersion in each of the parameters used to calcu-
late the yield limit state displacement capacity are considered. Readers are
referred to the work of Iaccino (2004) for an example of application of the
current vulnerability assessment method, in its probabilistic version, to a
real case study: the province of Imperia in Liguria, Italy.
For comparative purposes, the median displacement demand spectrum is
assumed to be that used in the deterministic example in Section 2.7, which
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 205
was taken from the 1992 Erzincan record. In order to obtain a measure of
the variability, it is assumed that this spectrum represents that produced by
a specic scenario and that the aleatory variability of the spectral ordinates
would be represented by the logarithmic standard deviations of the ground-
motion prediction equations. For the purpose of this exercise, the standard
deviations associated with the predictive equations for displacement ordi-
nates derived by Bommer et al. (1998) are employed; these only cover peri-
ods up to 3.0 s, but are stable from about 0.8 s and are therefore assumed
to remain constant for longer response periods. The CDF of the demand
displacement at each period is then easily obtained assuming a lognormal
distribution, leading to the three-dimensional surface shown in Figure 9a.
At 3 s, the 50-percentile (median), 16-percentile (median minus 1 standard
deviation) and 84-percentile (median plus 1 standard deviation) values of
the displacement spectrum are indicated. The median displacement spec-
trum, that has been presented previously in Figure 5, is again shown here
in Figure 9b, along with the 16-percentile and 84-percentile spectra. The
response ordinates obtained at 3 s for each spectrum have been highlighted;
these correspond to those values indicated in the three-dimensional CDF
plot in Figure 9a.
The parameters required in the denition of the displacement capac-
ity in this example are presented in Table VI. The origin of the values of
mean and standard deviation and the chosen probabilistic distributions has
been discussed in Section 3.3, however it is recalled here that these are
merely indicative, obtained from a preliminary analysis of a limited sample
of European buildings (Crowley, 2003). These parameters are assumed to
be uncorrelated at present until extensive data is available to calculate the
correlation coefcients between pairs of parameters. For the coefcients in
Table VI. Mean and standard deviation values and assumed distributions used for
the parameters in the limit state 1 (yield) displacement and period capacity equa-
tions for column-sway frames
Parameter Mean Standard deviation Distribution
Storey height, h
s
3.22 m 0.59 m Lognormal
Column depth, h
c
0.38 m 0.14 m Lognormal
Steel grade 275 yield strain,
y
0.165% 0.0165% Normal
Coefcients
Periodheight 0.1 0.015 Normal
Column-sway yield rotation 0.43 0.09 Normal
Plastic hinge length 0.5 0.15 Normal
Column yield curvature 2.14 0.214 Normal
206 H. CROWLEY ET AL.
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0 1 2 3 4 5
Period (s)
S
p
e
c
t
r
a
l

D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
50-percentile
16-percentile
84-percentile
(a)
(b)
7 6
Figure 9. (a) CDFs of the demand displacement at each period, with median, 16-per-
centile and 84-percentile values of displacement response indicated at 3 s and (b) the
median, 16-percentile and 84-percentile displacement demand spectra.
the empirical equations, the mean values have been taken from the liter-
ature where the derivation of these relationships is presented, introduced
earlier. The standard deviation has been found from either the CV pub-
lished with the associated empirical formula or, where this was not avail-
able, by assuming a tentative CV from the degree of scatter.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 207
0.00
0.05
0.10
0.15
0.20
0.25
0 0.5 1 1.5 2
Period (s)
D
i
s
p
l
a
c
e
m
e
n
t

(
m
)
(a)
(b)
Figure 10. (a) Probability density functions of yield displacement capacity, condi-
tioned to period and (b) two-dimensional deterministic curve of yield displacement
capacity vs period.
Figure 10 shows an example of the probability density functions, for
a range of periods, of the rst limit state (yield) displacement capac-
ity of a column-sway RC building class. If there were no uncertainty in
the calculation of displacement capacity then this graph would be the
208 H. CROWLEY ET AL.
two-dimensional curve simply relating displacement capacity to period that
has been shown previously in the deterministic example in Figure 5. This
is shown by the line on the plot in Figure 10a and is repeated, for clarity,
in two-dimensions in Figure 10b.
The conditional probability of failure, introduced in Section 3.1, can
be calculated using the inner integral of Eq. (42) now that the condi-
tional CDF of the displacement demand (Figure 9a) and the conditional
probability density function of the displacement capacity (Figure 10a) have
been found. In this probabilistic framework, the aforementioned condi-
tional probabilities of failure can be unconditioned using the probability
density function of period corresponding to a given number of storeys.
Figure 11 shows the probability density functions of yield period for
various heights of column-sway RC frames. The increased dispersion in the
probability density function with increased number of storeys is notable
and can be explained when one considers that the mean storey height and
its associated standard deviation of all frames has been assumed to be the
same; however, the dispersion in the total height will be much higher when
the building contains more storeys. Considering that the period has been
shown to be related to the total height of a building, as discussed in Sec-
tion 2.4, it is expected that there will be more dispersion in the period of
vibration of buildings as the number of storeys increases.
For a given number of storeys, at a given period, the probability den-
sity of that period (from Figure 11) may be multiplied by the probability
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
0 0.5 1 1.5 2 2.5 3
Period (s)
P
D
F

(
T
)

1 storeys
2 storeys
3 storeys
4 storeys
5 storeys
Figure 11. Example probability density functions of yield period for column-sway
frames of varying number of storeys.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 209
Figure 12. Example JPDF of capacity for a four storey column-sway RC building
class.
density function of the corresponding displacement capacity (from Fig-
ure 10) so that the JPDF of period and displacement can be obtained, as
illustrated in Figure 12 for a four storey column-sway building class. As
should be expected, the volume below this surface is 100%; if there were no
uncertainty in the period or displacement capacity then this gure would
show a single spike with a probability equal to unity, such as the single
points of displacement capacity shown for each number of storeys in the
deterministic example in Figure 5.
The JPDF can be used in conjunction with the demand CDF through
the use of the reliability formulation of Eq. (42) to nd the probability of
exceeding the yield limit state for each number of storeys. Table VII shows the
results of this probabilistic example and compares them with those obtained
in the deterministic example. It is observed that although earthquake loss
estimation studies based on a deterministic procedure of vulnerability assess-
ment would differ greatly from those based on a fully probabilistic method,
the higher vulnerability of the building stock between 3 and 10 storeys is
identied with both methods. The application of a fully probabilistic method
is recommended as this allows for a systematic and rational treatment of the
210 H. CROWLEY ET AL.
Table VII. Comparison of yield limit state exceedance probabilities (P
f
) for column-
sway frames obtained using a deterministic and a fully probabilistic procedure
Number of storeys P
f
in deterministic example P
f
in probabilistic example
0 0 0.00
1 0 0.06
2 0 0.24
3 1 0.44
4 1 0.54
5 1 0.60
6 1 0.62
7 1 0.60
8 1 0.55
9 1 0.49
10 1 0.42
11 0 0.35
12 0 0.29
uncertainties that exist when trying to predict the actions from future earth-
quakes and the resulting response of groups of buildings.
4. Brief Comparison with Existing Methodologies
4.1. Preamble
The methodology described in this paper allows the proportion of build-
ings falling within dened damage bands to be calculated for loss estima-
tion studies. The method of HAZUS (FEMA, 1999) has been discussed
in Section 1 wherein it was mentioned that the proportion of buildings
exceeding a given damage band is found in HAZUS using vulnerability
curves. A vulnerability curve gives the probability of failing a limit state,
given a value of displacement demand. In order to make a brief compar-
ison between HAZUS and the method outlined in this paper, the implied
vulnerability curves associated with this method are presented, even though
they are neither derived nor required for the application of the proposed
new approach. By making this comparison only in terms of the vulnera-
bility functions, it is possible to compare the new approach with HAZUS
without also considering the differences in the treatment of the demand
between the two methods. An extensive comparison with other vulnerabil-
ity curves has not been made in this work as vulnerability functions that
adopt spectral displacement as the demand parameter are lacking, the only
exception known to the authors being the recent work of Rossetto and
Elnashai (2004).
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 211
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8
Spectral Displacement, S
D
(m)
P
f

/

S
D
HAZUS_High-Code
HAZUS_Pre-Code
1
Figure 13. Collapse limit state vulnerability curves from HAZUS (FEMA, 1999) for
mid-rise RC buildings designed to High-Code and Pre-Code standards.
4.2. Hazus vulnerability curves
Figure 13 shows the collapse limit state curves proposed by HAZUS
(FEMA, 1999) for buildings designed to a High-Code seismic standard and
a Pre-Code standard. The HAZUS manual suggests that buildings of newer
construction built in an area of high seismicity may be modelled using the
High-Code vulnerability (or fragility) curves, whilst older buildings built
anywhere in the United States before seismic provisions were introduced
may be modelled using the Pre-Code curves. Immediately it is apparent
from Figure 13 that the shape of the two curves is very similar with only
a slight increase in the slope for the Pre-Code buildings. Hence it would
be predicted from these curves that for a given displacement demand, sim-
ilar proportions of well designed and poorly designed buildings collapse,
especially at high levels of spectral displacement.
4.3. Vulnerability curves obtained with proposed method
Vulnerability curves for each limit state may easily be generated with the
method outlined in this paper. The vulnerability is dened as the prob-
ability of exceeding a limit state, given a value of displacement demand.
Once the JPDF of limit state displacement capacity and period has been
calculated, as has been illustrated in Figure 12, the probability of fail-
ing the limit state is simply found from the volume of the JPDF below a
given level of displacement response, S
d
. Figure 14 illustrates this process of
212 H. CROWLEY ET AL.
Figure 14. (a) The JPDF of capacity below a given displacement demand level and (b)
the JPDF of capacity above a given displacement demand level.
generating vulnerability curves; Figure 14a shows the volume of the JPDF
of capacity falling below a given displacement response plane of 0.078 m,
and thus the probability of failure at this displacement, whilst Figure 14b
is a rotated view of Figure 14a to show the volume falling above this dis-
placement level.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 213
0
0.2
0.4
0.6
0.8
1
0 0.1 0.2 0.3 0.4
Spectral Displacement, S
D
(m)
Spectral Displacement, S
D
(m)
P
f

/

S
D

P
f

/

S
D

Limit State 1
Limit State 2
Limit State 3
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8
Limit State 1
Limit State 2
Limit State 3
(a)
(b)
1
Figure 15. Vulnerability curves for the three limit states for mid-rise (a) column-sway
and (b) adequately conned beam-sway RC frames.
The whole vulnerability curve can then be generated by increasing
the level of S
d
, and thus moving the plane in Figure 14 from zero
displacement until a level is reached when all of the JPDF falls below the
displacement demand level, and hence the probability of failure is unity.
Figure 15a shows the vulnerability curves obtained for the three limit
states for inadequately conned column-sway frames and Figure 15b
214 H. CROWLEY ET AL.
likewise for adequately conned beam-sway frames, implicit in the method
proposed in this work. The failure mechanism of a column-sway frame is
due to the concentration of plastic rotation in columns which have not
been designed for such demand. A failure mechanism develops abruptly
through the formation of a soft-storey; this mechanism is associated with
a rapid transition from low levels of damage to collapse, which can be
seen from the closeness of the three limit state curves and the steepness of
their slopes in Figure 15a. This indicates that for buildings susceptible to
soft-storey collapse, the susceptibility is effectively dened by a threshold
level of loading.
The beam-sway frames, on the other hand, can be seen to withstand
much higher displacement response for a given probability of failure and
the three limit state curves show a slower transition from low levels of
damage to collapse, as would be expected for this highly ductile failure
mode. Such differences between the vulnerability of buildings designed to
older seismic codes (i.e. column-sway frames) and those designed to mod-
ern seismic codes (i.e. beam-sway frames) have also been noted by Rossetto
and Elnashai (2003), through their work on the derivation of vulnerability
curves based on observational data.
Figure 16 illustrates together the collapse limit state vulnerability curves
of beam- and column-sway frames. Assuming that the displacement demand
to mid-rise beam- and column-sway frames is comparable (which may be
the case if one considers that the ductility of the latter will be less, leading
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8
Spectral Displacement, S
D
(m)
P
f
/

S
D


Limit State 3 - Beam-sway
Limit State 3 - Column-sway
1
Figure 16. Comparison of limit state 3 (collapse limit state) for mid-rise column-sway
and beam-sway RC frame.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 215
to reduced period and hence displacement demand but also reduced
damping), it may be predicted from Figure 16 that when the displace-
ment demand is 0.22 m, 95% of column-sway buildings have failed but
only 10% of beam-sway buildings reach collapse. This is consistent with
eld observations following damaging earthquakes: the majority of well-
designed beam-sway frames do not collapse, whereas signicant numbers of
the poorly designed buildings suffer a soft-storey type of failure.
The structural behaviour implied by the curves in Figure 16 contrasts
with that implied by the vulnerability curves proposed by HAZUS (Fig-
ure 13), where the fundamental difference in the behaviour of well designed
and poorly designed reinforced concrete buildings is not so apparent. The
quantitative differences between the failure probabilities are not being con-
trasted here but the shapes of the curves, as it is most likely that differ-
ent assumption have been made herein to those used in the derivation of
the HAZUS curves. Using the vulnerability curves suggested by HAZUS
in Figure 13, it may be predicted that when 10% of buildings designed to
modern design philosophy collapse, only 35% of those designed with no
considerations of capacity design are found to have collapsed; this seems an
unlikely outcome, at least where European buildings are concerned. These
ndings perhaps lead to the questionable use of such vulnerability curves
for the urban assessment of European structures which are, for the most
part, poorly designed structures that are likely to suffer soft-storey collapse
at low levels of ductility. This may in part explain the differences between
predicted and observed losses due to the 1999 Kocaeli earthquake found by
Spence et al. (2003) using an adaptation of HAZUS to Turkey described
by Bommer et al. (2002).
It is reiterated that the authors do not recommend the direct use of the
vulnerability curves presented in this section in standard loss estimation
procedures; the curves are presented for illustrative purposes only. Fur-
thermore, it should be borne in mind that these curves have been con-
structed using the values and distributions described in Sections 3.3 and
3.4, which assume a steel grade of 275 MPa and geometric properties based
on a limited European building stock data set.
5. Conclusions
The rst steps to develop an innovative displacement-based method for
earthquake loss estimation of classes of buildings were made by Pinho
et al. (2002). In this present work, a probabilistic framework, as well as fur-
ther developments, have been provided, thus bringing the method one step
closer to practical application.
The limit state structural displacement formulae, given in terms of
period, rst proposed by Pinho et al. (2002) have been updated owing to
216 H. CROWLEY ET AL.
the derivation of a sound relationship between period and height (Crowley
and Pinho, 2004). Non-structural displacement capacity formulae, at differ-
ent limit states, and given in terms of period, have been derived herein for
the rst time. Further developments that will need to be applied to the
methodology include the consideration of the inuence of inll panels on
the displacement capacity of RC frames, the shear deformation capacity
of non-exure controlled members, and the inclusion of the displacement
capacity formulae for RC wall and dual-system structures.
The treatment of the uncertainty in the displacement demand and
capacity has been incorporated in the method with the inclusion of a
fully probabilistic framework. A brief preliminary study regarding the
probabilistic distributions that should be applied to the parameters used
within the displacement capacity formulae has been presented; work is
ongoing to rene this part of the method.
The procedure presented herein has been shown to be particularly suit-
able for loss estimation studies due to its transparency, theoretical accuracy
and computational efciency. The denition of the displacement capacity
is transparent as one may use any chosen number of storeys, geometri-
cal, material or limit state threshold properties in the equations and adapt
these easily for use in any part of the world. The conceptual soundness
of the methodology has been preliminarily examined through a compari-
son of vulnerability curves derived using this procedure and those provided
in HAZUS (FEMA, 1999). The curves derived using the method proposed
in the current paper are distinct for well- and poorly designed buildings,
leading to more realistic vulnerability curves which appear to be consis-
tent with eld observations following destructive earthquakes. On the other
hand, the curves proposed by HAZUS appear not to provide such sound
and rational predictions of damage between various classes of buildings.
The computational efciency of the methodology will also permit the
performance of systematic sensitivity studies to establish the relative inu-
ence of each input parameter on the results of an earthquake loss model.
Such work is needed in order to identify the areas where additional effort
on data collection and model resolution is likely to be justied in terms of
impact on the resulting losses.
Acknowledgements
The authors would like to express their gratitude to Juliet Bird whose com-
ments and observations generated many interesting discussions, the out-
come of which has signicantly improved this paper. Particular mention
is due to Luis Fernando Restrepo-V elez and Guido Mangenes for kindly
providing the Fortran code related to their original work in applying a
probabilistic framework to a comparable vulnerability assessment method
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 217
developed for masonry structures. The authors thank also Nigel Priestley
and an anonymous reviewer for their useful and encouraging comments.
The rst author would like to acknowledge the nancial support provided
by the European Community, through the Safety Assessment For Earth-
quake Risk Reduction (SAFERR) Research Training Network.
References
Antoniou, S. and Pinho, R. (2004) Development and verication of a displacement-based
adaptive pushover procedure. Journal of Earthquake Engineering 8(5), 643661.
Applied Technology Council (1985) Earthquake damage evaluation data for California,
Report ATC-13, Applied Technology Council, Redwood City, California.
Bommer, J.J. (2002) Deterministic vs. probabilistic seismic hazard assessment: an obstructive
and exaggerated dichotomy. Journal of Earthquake Engineering 6(special issue 1), 4373.
Bommer, J.J., Elnashai, A.S., Chlimintzas, G.O. and Lee, D. (1998) Review and development
of response-spectra for displacement-based seismic design. ESEE Research Report No.
98-3, Imperial College, London.
Bommer, J.J., Elnashai, A.S. and Weir, A.G. (2000) Compatible acceleration and displace-
ment spectra for seismic design codes. In Proceedings of the 12th World Conference on
Earthquake Engineering, Auckland, New Zealand, Paper no. 207.
Bommer, J.J. and Mendis, R. (2004) Scaling of displacement spectral ordinates with damping
ratios. Earthquake Engineering and Structural Dynamics, in press.
Bommer, J.J., Spence, R., Erdik, M., Tabuchi, S., Aydinoglu, N., Booth, E., del Re, D. and
Peterken, O. (2002) Development of an earthquake loss model for Turkish catastrophe
insurance. Journal of Seismology 6, 431446.
Calvi, G.M. (1999) A displacement-based approach for vulnerability evaluation of classes of
buildings. Journal of Earthquake Engineering 3(3), 411438.
Comite Europeen de Normalisation (2003) Eurocode 8, design of structures for earthquake
resistance Part 1: General rules, seismic actions and rules for buildings, Pr-EN 1998-1.
Final Draft. December 2003.
Crowley, H. (2003) Periods of vibration for displacement-based assessment of RC structures.
MSc dissertation, European School for Advanced Studies in Reduction of Seismic Risk
(ROSE School), University of Pavia, Italy.
Crowley, H. and Pinho, R. (2004) Period-height relationship for existing European reinforced
concrete buildings. Journal of Earthquake Engineering 8(special issue 1), 93119.
Dymiotis, C. Kappos, A.J. and Chryssanthopoulos, M.K. (1999) Seismic reliability of RC
frames with uncertain drift and member capacity. Journal of Structural Engineering
125(9), 10381047.
FEMA (1999) HAZUS99 Earthquake Loss Estimation Methodology: Users Manual, Fed-
eral Emergency Management Agency, Washington, DC.
FEMA (2001) HAZUS99: Estimated Annualized Earthquake Losses for the United States .
FEMA 366, Federal Emergency Management Agency, Washington, DC.
Franchin, P., Lupoi, A. and Pinto, P.E. (2002) Methods for seismic risk analysis: state of
the art versus advanced state of the practice. Journal of Earthquake Engineering 6(special
issue 1), 131155.
Freeman, S.A., Messinger, D.L., Casper, W.L., Mattis, L.W., Preece, F.R. and Tobin, R.E.
(1985) Structural moments no. 4. drift limits: are they realistic? Earthquake Spectra 1(2),
203390.
218 H. CROWLEY ET AL.
Glaister, S. and Pinho, R. (2003) Development of a simplied deformation-based method
for seismic vulnerability assessment. Journal of Earthquake Engineering 7(special issue 1),
107140.
Iaccino, R. (2004) Probabilistic implementation of a mechanics-based procedure for seis-
mic risk assessment of classes of RC buildings. MSc Dissertation, European School for
Advanced Studies in Reduction of Seismic Risk (ROSE School), University of Pavia,
Italy.
Joint Committee for Structural Safety (JCSS) (2001) Probabilistic Model Code Working
document. Last update 13/03/2001 (online). Available from URL: www.jcss.ethz.ch.
Kappos, A.J., Chryssanthopoulos, M.K. and Dymiotis, C. (1999) Uncertainty analysis of
strength and ductility of conned reinforced concrete members. Engineering Structures
21, 195208.
Kircher, C.A., Nassar, A.A., Kustu, O. and Holmes, W.T. (1997) Development of
building damage functions for earthquake loss estimation. Earthquake Spectra 13(4),
663682.
Miranda, P. (2004) Displacement capacity of RC columns with limited shear resistance.
MSc Dissertation, European School for Advanced Studies in Reduction of Seismic Risk
(ROSE School), University of Pavia, Italy.
Mirza, S.A. and MacGregor, J.G. (1979a) Variations in dimensions of reinforced concrete
members. Journal of the Structural Division, Proceedings of the American Society of Civil
Engineers 105(ST4), 751766.
Mirza, S.A. and MacGregor, J.G. (1979b) Variability of mechanical properties of reinforc-
ing bars. Journal of the Structural Division, Proceedings of the American Society of Civil
Engineers 105(ST5), 921937.
Musson, R.M.W. (2000) Intensity-based seismic risk assessment, Soil Dynamics and Earth-
quake Engineering 20, 353360.
Ordaz, M., Miranda, E. Reinoso, E. and P erez-Rocha, L.E. (2000) Seismic loss estimation
model for Mexico City. In Proceedings of the 12th World Conference on Earthquake Engi-
neering, Auckland, New Zealand, Paper no. 1902.
Pinho, R., Bommer, J.J. and Glaister, S. (2002) A simplied approach to displacement-based
earthquake loss estimation analysis. In Proceedings of the 12th European Conference on
Earthquake Engineering, London, England, Paper no. 738.
Pinto, P.E., Giannini, R. and Franchin, P. (2004) Methods for Seismic Reliability Analysis of
Structures. IUSS Press, Pavia, Italy.
Priestley, M.J.N. (1997) Displacement-based seismic assessment of reinforced concrete build-
ings. Journal of Earthquake Engineering 1(1), 157192.
Priestley, M.J.N. (2003) Myths and fallacies in earthquake engineering revisited. In The
Mallet-Milne Lecture. IUSS Press, Pavia, Italy.
Restrepo-Velez, L.F. (2004) A simplied mechanical-based procedure for the seismic risk
assessment of unreinforced Masonry buildings. Individual Study, ROSE School, Pavia,
Italy.
Restrepo-Velez, L.F. and Bommer, J.J. (2003) An exploration of the nature of the scatter in
ground-motion prediction equations and the implications for seismic hazard assessment.
Journal of Earthquake Engineering 7(special issue 1), 171199.
Restrepo-Velez, L.F. and Magenes, G. (2004) Simplied procedure for the seismic risk assess-
ment of unreinforced masonry buildings. In Proceedings of the 13th World Conference on
Earthquake Engineering, Vancouver, Canada, Paper No. 2561.
Rossetto, T. and Elnashai, A. (2003) Derivation of vulnerability functions for European-type
RC structures based on observational data. Engineering Structures 25, 12411263.
PROBABILISTIC DISPLACEMENT-BASED VULNERABILITY ASSESSMENT 219
Sasani, M. and Der Kiureghian, A. (2001) Seismic fragility of RC structural walls: displace-
ment approach. ASCE Journal of Structural Engineering 127(2), 219228.
Spence, R., Bommer, J.J., Del R, D., Bird, J., Aydinoglu, N. and Tabuchi, S. (2003) Com-
paring loss estimation with observed damage: a study of the 1999 Kocaeli earthquake in
Turkey. Bulletin of Earthquake Engineering 1(2), 83113.

Das könnte Ihnen auch gefallen