Engineering Structures 150 (2017) 746–758

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

A performance-based design framework for the integrated collapse and

non-collapse assessment of wind excited buildings
Wei-Chu Chuang, Seymour M.J. Spence ⇑
Department of Civil Engineering, University of Michigan, Ann Arbor, MI 48109, United States

a r t i c l e i n f o a b s t r a c t

Article history: Current prescriptive design provisions are moving towards performance-based design approaches in
Received 9 February 2017 which system-level probabilistic measures are used to explicitly describe performance. While earthquake
Revised 12 June 2017 engineering has embraced these changes over the last few decades, the same cannot be said for wind
Accepted 11 July 2017
engineering where design provisions have remained predominantly prescriptive. The significant wind
related economic losses incurred each year around the world has spurred strong interest in developing
general performance-based wind engineering frameworks. To this end, this paper presents a
Keywords:
simulation-based framework for multistory wind excited buildings that rigorously integrates system-
Performance-based wind engineering
Aerodynamic loads
level estimates of both collapse and non-collapse losses. In particular, it is proposed to use the theory
Monte Carlo simulation of dynamic shakedown as an efficient means for describing the collapse probability of the main wind
Wind engineering force resisting system. The practicality and potential of the proposed framework is illustrated on a full
Performance-based design scale case study.
Dynamic shakedown Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction collapse performance assessment of specialized wind excited

The advantages of performance-based design over traditional
prescriptive provisions is well known in the field of seismic engi-
neering. In particular, intense research over the past 30 plus years
has led to the development of what is commonly referred to as sec-
ond generation Performance-Based Seismic Design (PBSD) [1,2]
and, in particular, the publication by the Federal Emergency Man-
agement Agency (FEMA) of the P-58 volumes [3–5]. These docu-
ments outline a general methodology that not only accounts for
the inevitable uncertainty in accurately predicting the response
of building systems subject to severe earthquakes, but also
communicates performance through system-level measures that
are easily understood by decision-makers and/or stakeholders,
e.g. expected repair time and cost. While the framework outlined
in the P-58 volumes – and more in general the model proposed
by the Pacific Earthquake Engineering Research (PEER) center
[1,2] – was developed for buildings subject to earthquake excita-
tion, it is a relatively general procedure that can in theory be
extended to other natural hazards such as wind. In this respect,
several research efforts have been conducted over the past few
years [6–15]. These works have mainly focused on the non-
⇑ Corresponding author.
E-mail addresses: wechuang@umich.edu (W.-C. Chuang), smjs@umich.edu
(S.M.J. Spence).
structures such as tall buildings and long span bridges. However,
as outlined in [15], the next steps in applying performance-based
design to wind engineering would require the development of
general methodologies that can model the behavior of systems
subject to a full range of moderate to severe wind events.
To develop such a framework, models that estimate both the
non-collapse as well as the collapse performance in terms of prob-
abilistic system-level losses are necessary. Therefore, the possibil-
ity of allowing the system to enter an inelastic response regime
needs to be taken into consideration in order to avoid leaving the
system exposed to undesirable post-yield behavior and possible
collapse. In this respect, various contributions have been made
towards better understanding how systems respond inelastically
to wind [16–24,15]. In particular, the computational challenge of
estimating the nonlinear response has been identified as a major
issue as, in a nonlinear response regime, the complete duration
of the event must be considered if meaningful results are to be
obtained. Because of the long duration of typical wind events (in
the order of several hours), the possibility of applying robust
numerical methods that require direct integration of the nonlinear
dynamic equations of motion over the entire duration of the storm,
such as incremental dynamic analysis (IDA) [25], are lost. This is
especially true if the probability of inelastic response, and
ultimately collapse, is desired (which is necessary if frameworks
such as the P-58 are to be developed for wind excited systems).

http://dx.doi.org/10.1016/j.engstruct.2017.07.030
0141-0296/Ó 2017 Elsevier Ltd. All rights reserved.
 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758
An alternative approach to directly integrating the nonlinear equa-
tions of motion, which has recently been applied to the analysis of
wind excited systems [26], is that based on applying dynamic
shakedown theory [27–32]. These methods have the potential to
rapidly provide a complete picture of the post-yield behavior of
the structure, indicating, for instance, whether the structure is in
shakedown or low cycle fatigue.
In this paper, a simulation-centered Performance-Based Wind
Engineering (PBWE) framework for multi-story wind excited
buildings is proposed that rigorously integrates both non-
collapse as well as collapse performance in terms of system-level
metrics such as probable repair costs and downtime. In particular,
in estimating the probable repair times, the recent methods out-
lined in the Resilience-based Earthquake Design Initiative (REDi)
guidelines are integrated into the proposed framework, while the
theory of probabilistic dynamic shakedown is proposed as a means
to fully describe the post-yield behavior of the main wind force
resisting system (MWFRS).
2. Performance-based wind engineering: problem setting
The PBWE framework that will be developed in this work will
be based on the recently proposed extension to wind engineering
of the PEER framework [6,7]. In particular, instead of considering
the mean annual rates of exceedance of the thresholds dv, struc-
tural performances will here be measured in terms of the annual
exceedance probability of the thresholds, Pf , as this measure is
more commonly used in wind engineering. Thus, performance is
assessed by solving the following probabilistic integral:
ZZZZ
Pf ðdv Þ ¼ Gðdv jdmÞ  jdGðdmjedpÞj
jdGðedpjipÞj  jdGðipjimÞj  pðimÞ  dim
where GðajbÞ is the Complementary Cumulative Distribution Func-
tion (CCDF) of A conditional on B (where the common convention
of using capital letters to indicate random variables and lower case
letters for their realizations has been used); DM is the damage mea-
sure indicating the state of damage of structural and/or non-
structural components; EDP is the engineering demand parameter,
i.e. the structural response responsible for causing damage; IP rep-
resents a set of interaction parameters (i.e. the aerodynamic loads
acting on the structure); and IM is the measure of the intensity of
the event with pðimÞ the probability density function of the annual
largest values of IM.
As in the case of the FEMA P-58 seismic framework, in this work
Eq. (1) is used for estimating the performances of buildings that are
repairable, i.e. for buildings that do not collapse during the wind
event. To separate collapse from non-collapse, a model based on
dynamic shakedown theory will be developed. In particular, since
collapse and non-collapse are mutually exclusively events, the
probability of the DV exceeding a threshold, dv, considering both
scenarios can be expressed through the total probability theorem
as:
PðDV > dv Þ ¼ PðDV > dv jNCÞPðNCÞ þ PðDV > dv jCÞPðCÞ
where PðCÞ is the probability of collapse, PðNCÞ the complement of
PðCÞ (i.e. the probability non-collapse), PðDV > dv jNCÞ is the annual
exceedance probability of dv given that the building does not col-
lapse, while PðDV > dv jCÞ is the annual exceedance probability of
dv given that the building collapses during the event.
In order to solve Eq. (2), models need to be defined for estimat-
ing PðNCÞ; PðDV > dv jCÞ, as well as for the sub-analysis tasks
involved in solving Eq. (1). The following sections will outline
possible models to this end.
747
3. The proposed PBWE framework
3.1. Non-collapse assessment
This section focuses on defining appropriate models for carrying
out the sub-analyses for solving the non-collapse problem of Eq.
(1).
3.1.1. Wind hazard analysis
A common measure of the intensity of an extreme wind event is
the maximum wind speed, v
z , to occur at a height of interest z (e.g.
building or eave height) averaged over a fixed time interval T (e.g.
an hour). As discussed in [13], this parameter is a logical choice for
a site specific intensity measure, IM. In particular, IM can be related
to meteorological data, v, collected at nearby weather stations
through a probabilistic transformation, P, that accounts for aspects
such as differences in the site roughnesses, averaging times, as well
as observational and modeling uncertainties [33]. In this work, the
transformation outlined in [13] is adopted, which has the following
general form:
im ¼ v
z ¼ P½T; z; z0 ; uIM ; v ðs; Hmet ; z01 Þ ð3Þ
where s is the averaging time associated with the meteorological
data, z0 and z01 are the roughnesses at the site of interest and the
meteorological station of height Hmet respectively, while U IM is a
vector collecting the uncertain parameters associated with the
transformation of Eq. (3). More details on the transformation of
Eq. (3) can be found in [13].
3.1.2. Aerodynamic analysis
After identifying the intensity of the wind event, aerodynamic
loads, F, acting on the structure need to be determined before
ð1Þ
the EDPs can be estimated through structural analysis. These loads,
defined as the interaction parameters (IP), can be obtained from
specific wind tunnel tests or associated databases, computational
fluid dynamics or quasi-steady models. In general, the interaction
parameters will depend on a vector of uncertain model parameters,
U IP , as well as the intensity of the wind event v

z:
ip ¼ Fðt; v
z ; uIP Þ ð4Þ
In this work, a quasi-steady model is used to simulate the wind
time histories as outlined in Appendix A.
3.1.3. Structural analysis
The damage models for assessing the system-level building per-
formance developed in this paper are driven by a vector of EDPs
defined by structural responses such as displacements u, velocities
u_ and accelerations u€ . In particular, the EDPs are taken as the max-
imum responses to occur over the duration, T, of the wind event:
edp ¼ max Rðt; F; uEDP Þ ð5Þ
06t6T
where Rðt; F; uEDP Þ is the vector collecting the response processes of
interest while U EDP is a vector containing the uncertain parameters
associated with estimating the structural responses (e.g. the modal
damping ratios or Young’s modulus). In particular, the vector R can
ð2Þ
be formally defined as:

Rðt; F; uEDP Þ ¼ K u ð6Þ
where K is an indicator matrix extracting the responses, including
combinations (e.g. inter-story drifts), from the augmented response

vector u ¼ fu€ ; u;
_ ugT . In particular, in order to determine the
dynamic response of the structure, a modal framework can in gen-
eral be used [34].
To account for the uncertainties involved in the structural
model and material properties, the modal damping ratios, ni , and
748 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758
Young’s modulus, ey , can be taken as random variables, as can any
other parameter used in defining the structural model. For
instance, in this work, in addition to the above mentioned param-
eters, a random variable DM is introduced to model the uncertainty
in the mass matrix. The aforementioned random variables are col-
lected in the vector U EDP ¼ fn1 ; . . . ; nm ; DM ; Ey gT .
3.1.4. Damage analysis
In general, a building system will have a large number of
repeated components (e.g. windows, partition walls, ceiling ele-
ments, etc.). These groups of common elements have the same sus-
ceptibility to damage and consequences, and can therefore be
grouped together to form what are commonly refereed to as Fragi-
lity Groups (FGs). For each FG, a series of discrete damage states
(DSi for i ¼ 1; . . .) are generally identified and associated with
appropriate fragility functions that formally represent the proba-
bility of exceeding each damage state DSi given a demand level
edp, or:
FragilityDSi ðedpÞ ¼ ProbðDSi jedpÞ
Because in general, more than one component of a FG will be sub-
ject to the same demand, a number of FG subsets can be defined.
These subsets are termed performance groups (PGs) and are associ-
ated with one demand parameter or EDP. In particular, the possibil-
ity of correlation (or lack of) between damage occurrences can be
assessed for the components of any given PG. In practice, once a
demand (i.e. an EDP) value is obtained from the structural analysis,
e.g. EDP ¼ EDP  , damage analysis is performed by entering into a
suite of fragility functions at EDP (as illustrated in Fig. 1) and con-
sidering as an ordinate the value assumed by a sample of a uni-
formly distributed random number in ½0; 1. As discussed in [2,3],
the value assumed by the sample can be used to identify which
damage state has occurred.
3.1.5. Non-collapse loss assessment
The approach adopted in this work for assessing losses is based
on the framework proposed by FEMA [3]. Accordingly, in order to
translate the damage states identified through the fragility analysis
outlined in Section 3.1.4 to losses, a unique set of consequence
functions are defined. In particular, each fragility function is asso-
ciated with a consequence function that translates the damage
state described by the fragility function into a numerical value of
the consequence, e.g. dollars for repair cost or number of days
for repair time.
For the framework developed in this work, losses are restricted
to estimated repair cost and time. However, the procedure is gen-
eral and can be extended to other relevant loss measures, such as
condition tagging or casualties.
Fig. 1. Example of fragility functions for a component with three possible damage
states, DS1 ; DS2 and DS3 (adapted from FEMA [3]).
3.1.5.1. Repair cost. Following the framework outlined above, for
repair cost estimation, the consequence functions take the form
of unit repair cost functions (i.e. functions that estimate, for exam-
ple, the cost to repair a given damage state per square meter of par-
tition wall, or the cost to repair a single window subject to a given
damage state [3]). In particular, the unit repair cost functions are
described by the following parameters: maximum and minimum
unit costs, associated maximum and minimum quantities, as well
as dispersion parameters describing the uncertainty in translating
damage into repair costs. An example of a unit repair cost function
is shown in Fig. 2. In particular, the reduction in unit repair costs as
the quantity goes up (i.e. number of elements that require repair)
models the economies of scale. In general, together with the dis-
persion parameter, a possible distribution is also indicated, for
example log-normal.
In this work, once the number, N DSi , of elements in a given dam-
age state (DSi ) are identified for a given PG, the repair cost for those
elements can be estimated by sampling from the probability den-
sity function of the corresponding consequence function (at the
ordinate N DSi ) and then multiplying by N DSi . Following this scheme,
ð7Þ the total repair cost can be evaluated by aggregating over all dam-
age states and performance groups.
3.1.5.2. Downtime. In addition to the direct financial losses, build-
ings are generally also vulnerable to indirect economic losses due
to downtime, defined as the time required to achieve a recovery
state such as re-occupancy, functional recovery, or full recovery,
after a wind event [35]. In this work, a downtime assessment
model based on integrating the recently proposed methodology
outlined in the REDi guidelines [36] with that proposed in [3] will
be developed.
In assessing downtime, impeding factors that delay the initiation
of repairs required to achieve a certain recovery state must be iden-
tified and included in the downtime estimate. As illustrated in Fig. 3,
some impeding factors occur in series while others occur simultane-
ously. In particular, after inspection the sequence producing the
greatest delay among: (1) engineering mobilization and permitting;
(2) contractor mobilizations; and (3) financing, must be used in cal-
culating downtime. Furthermore, long-lead times due to the
unavailability of certain building components will in general occur,
therefore hindering the repair of the corresponding component. As
illustrated in Fig. 3, this source of potential delay should be included
at the end of the contractor mobilization phase.
In order to estimate total repair time, delays due to impeding
factors must then be added to repair work times (i.e. the time that
workers require to carry out the repairs). To estimate these last,
two extreme cases are considered in this work. In the first case,
repair work is assumed to be carried out in series, i.e. repair for
one component has to be finished before work on another can
start. This will provide an upper bound on the repair time as in
general some work can be carried out in parallel. In the second
case, all repair work is assumed to be carried out simultaneously.
Fig. 2. Example of a possible consequence function (adapted from FEMA [3]).
 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758
Fig. 3. Sequences of delays due to impeding factors.
This will provide a lower bound on the repair work time as, in gen-
eral, some repair work will inevitably require the repair of certain
elements before it can commence. The actual time required for the
repair work will be somewhere between the upper and lower
bounds. To estimate these bounds, a model similar that proposed
in [3] is considered. Therefore, the time necessary for a single worker
to repair the damage states described by the fragility functions is
estimated through unit consequence functions. In particular, each
consequence function is described by the following parameters: a
maximum and minimum repair time; associated upper and lower
quantities; a dispersion parameter modeling the uncertainty in the
repair time. The maximum and minimum repair times mentioned
above are associated with the increased efficiency that is likely to
occur if multiple components in the same damage state are repaired.
It should be observed that, because it is assumed that only a single
worker is available, the repair work time can be reduced proportion-
ally if multiple workers are considered.
Utility disruption is also considered in estimating downtime to
full or functional recovery. In particular, for wind storms, only dis-
ruption to the electrical systems is considered. Finally, the overall
downtime is estimated as:
Downtime ¼ Maxfutility disruption; impeding factorsg
ð8Þ
þ repair work
3.2. Collapse assessment
Section 3.1 outlined a procedure for assessing the performance
of building systems that do not collapse during the wind event. In
order to determine whether collapse has occurred or not, a defini-
tion of collapse must be given. This can range from the instability
of a numerical model of the structural system, to the exceedance of
a predetermined deformation limit. As already mentioned, the dif-
ficulty of estimating collapse in the case of wind excited systems
lies in treating the long duration of typical wind events (order of
hours), which creates both computational as well as theoretical
challenges (i.e. collapse could occur due not only to incremental
plasticity but also phenomena such as low cycle fatigue). To over-
come this, in this work it is proposed to define collapse of the
MWFRS as the failure of the system to achieve ‘‘dynamic shake-
down”, where ‘‘dynamic shakedown” is defined as a state in which
plastic deformation is produced only during a first phase of finite
duration whilst the whole subsequent phase is purely elastic
[27,37,32,26]. This state precludes the occurrence not only of incre-
mental plastic collapse but also of low cycle fatigue, and would
therefore seem well adapt for defining the separation between a
749
safe inelastic state and collapse in the case of wind excited
systems.
3.2.1. Dynamic shakedown and collapse of wind excited systems
The state of dynamic shakedown is not in general easily esti-
mated. A case of particular interest is that concerning external
periodic loads of infinite duration [27,37,32,26]. Indeed, as shown
in [26] and reported in Appendix B for convenience, under this con-
dition the identification of the state of dynamic shakedown
reduces to the resolution of a Linear Programming (LP) problem.
Because LP problems can be very efficiently solved (even in high
dimensions), the state of dynamic shakedown is easily identified
in the case of infinite duration periodic loads.
Unfortunately, naturally occurring wind storms are of finite
length and non-periodic. However, by considering the wind storm
of duration T as a finite realization of a stationary stochastic process
of period T, then an artificial wind storm is defined that is periodic
and of infinite duration. For this artificial wind storm, the LP prob-
lems of Eqs. (27) and (28) of Appendix B can be directly applied for
estimating not only the load multiplier sp that separates the state of
dynamic shakedown from a state of potential incremental plastic
collapse/low cycle fatigue, but also se , which separates a purely
elastic response from a state of dynamic shakedown. The signifi-
cance of the results obtained for the artificial wind storm outlined
above lies in the fact that, if the structure shakes down for the
artificial infinite duration and periodic wind storm, then it cannot
undergo plastic fatigue failure or incremental plastic collapse dur-
ing the actual wind storm of duration T. This result is extremely
useful as it allows the state of shakedown or non-shakedown (i.e.
non-collapse or collapse) to be identified by solving a LP problem
of the type shown in Eq. (27) of Appendix B, which, as already
mentioned, does not present any particular computational issues
even for events of long duration (T in the order of hours). In
particular, a value of sp P 1 implies that, under the loads FðtÞ, the
structure will shakedown and is therefore safe against fatigue
failure or incremental plastic collapse. This implies that, within a
simulation framework, if for a given wind load realization it is
found that sp P 1, it can be stated that the system has not collapsed,
where collapse is defined as the failure to reach a state of dynamic
shakedown. It is this definition of collapse/non-collapse that is used
in the proposed framework.
Before closing this section, it should be observed that local fail-
ure mechanisms (e.g. local buckling deformations of the web and
flange of a connection) are included in this framework through
the damage states and associated fragility functions considered
in the non-collapse damage analysis of Section 3.1.4.
750 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758

3.2.2. Collapse loss assessment
As mentioned in Section 2, the estimation of losses given that
collapse occurs requires the estimation of PðDV > dv jCÞ. In order
to estimate PðDV > dv jCÞ, it is here proposed – in accordance with
the FEMA P-58 guidelines – to take the repair cost/time equal to
the replacement time/cost in the case of building collapse. There-
fore, PðDV > dv jCÞ is a probabilistic distribution representing the
building performances if replacement is required, that is:
PðDV > dv jCÞ ¼ 1  F DVjC ðdv jCÞ
where F DVjC is the conditional cumulative distribution function of the
total losses if the MWFRS collapses, e.g. the conditional distribution of
the total replacement cost if the decision variable is financial loss, or
the total replacement time if downtime is the decision variable.
4. Simulation strategy
Section 3.1 presented models for assessing the repair costs and
ð9Þ downtime associated with the occurrence of a severe wind storm

Fig. 4. Flowchart of the overall simulation strategy.

 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758 751

conditioned on the MWFRS not collapsing during the extreme where dv in Eq. (11) is evaluated following the models of Section 3.1
wind event. Section 3.2, on the other hand, presented a model for if the system is deemed not to have collapsed (i.e. the building
determining the integrity of the MWFRS with respect to collapse. shakedowns as defined in Section 3.2) while, if the system is
These models can be used to define a Monte Carlo simulation deemed to have collapsed, a sample from F DVjC (Eq. (9)) is
framework for estimating Eq. (2). To this end, consider generating generated.
N s samples of the basic random vector U ¼ fV; U IM ; U IP ; U EDP ; U sh gT A flowchart of the proposed simulation-based framework is
that collects all the uncertain parameters of the problem (with Ush shown in Fig. 4. In particular, the steps defining the Monte Carlo
a vector of uncertain parameters associated with the shakedown algorithm are as follows:
analysis – see Appendix B). By estimating a decision variable value
(i.e. a repair cost and repair time value) for each sample, the fol- 1. Generate a realization v ðiÞ of V and of the random vector
ðiÞ
lowing expression can be used to directly estimate Eq. (2): U IM ; uIM , and use Eq. (3) to estimate the intensity measure,
1 X Nnc
Nnc 1 X Nc
Nc v
z ðT; z; z0 Þ, at all heights of interest.
PðDV > dv Þ  IðuðjÞ
nc Þ  þ IðucðkÞ Þ 
2. Generate a realization uIP of FðtÞ to use together with v
ðiÞ
Nnc j¼1 Nnc þ N c Nc k¼1 Nnc þ Nc
z ðT; z; z0 Þ of
" # the previous step in order to estimate a realization of the interac-
1 X
Nnc XNc
tion parameters, i.e. the aerodynamic loads FðtÞ of Eq. (4).
ðjÞ ðkÞ
¼ Iðu Þ þ Iðuc Þ ð10Þ
Nnc þ Nc j¼1 k¼1
ðiÞ
3. Generate a realization uEDP of U EDP to use together with the
aerodynamic loads of the previous step, for estimating the
where N c and Nnc are the total number of samples for which collapse/ 
ðkÞ ðjÞ response vector u.
non-collapse occurred with uc for k ¼ 1; . . . ; N c and unc for ðiÞ
j ¼ 1; . . . ; N nc the corresponding samples of U, while I is an indicator 4. Generate a realization ush of U sh and use it together with the
function defined – irrespective of whether collapse occurs or not – as: responses from step 3 to determine the shakedown safety fac-
( tor, sp , by solving the linear programming problem of Eq. (27).
1 if dv ðuðiÞ Þ > dv Check if the structure has collapsed during the wind event, i.e.
IðuðiÞ Þ ¼ ð11Þ
0 if dv ðuðiÞ Þ 6 dv if sp < 1.

(4)@6.1 m = 24.4 m
Level 40
154.7 m

24.4 m

Level 30
116.6 m

36.6 m

Level 20
3.8 m
Typical 78.5 m

Frame
Considered
Level 10
Building Plan Column and
40.4 m
beam sizes vary,
see Table 2

Ground
Surface Level 0
0m

Fig. 5. Schematic of the 40 story steel frame of the case study.

752 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758

5. If the structure does not shakedown (i.e. sp < 1), it is deemed to

have collapsed and repair cost and downtime for the ith realiza- 1
tion, dv ðuðiÞ Þ, are set equal to the building replacement cost and
time.
6. If collapse has not occurred, i.e. sp P 1, a realization of the -1 1
demand vector edpðiÞ (Eq. (5)) is generated. The vector edpðiÞ is -0.15 0.15
then used to estimate damage to the components of the perfor-
mance groups through their respective fragility functions.
Values of downtime and repair cost are then estimated for each -1
performance group through the consequence functions associ-
ated with each damage state. By aggregating over all perfor- Fig. 6. Piecewise linear failure domain of the plastic hinges in the columns.
mance groups, non-collapse values for dv ðuðiÞ Þ are obtained.
7. Use Eq. (11) to determine the value of IðuðiÞ Þ.
ear domain shown in Fig. 6, where N y is the axial generalized yield
By repeating steps 1–7 for N s samples of the random vector U, stress given by N y ¼ ry A while M y is the yield moment given by
estimates of the probability of the decision variable exceeding a
M y ¼ ry S with A and S the area and section modulus of the cross
threshold value, PðDV > dv Þ, can be obtained from Eq. (10).
section respectively. For plastic hinges located in the beams, the
yield domain is simply defined by the yield moments due to the
action of the floor diaphragms.
5. Case study
For the application described in this section, the distribution
characterizing the largest yearly meteorological wind speeds v
5.1. Description
was taken as a Type II distribution with mean value of 30 m/s
and standard deviation 3.5 m/s. In calibrating the transformation
As a case study, the 40-story four-span plane steel frame of
of Eq. (3), s was assumed as 1 h while a roughness length of
Fig. 5 is considered. The geometry of the frame is described by
z01 ¼ 0:05 m and a meteorological height of Hmet ¼ 10 m were con-
beam span lengths of 6.1 m (20 ft) and by the inter-story heights
sidered. The roughness length at the site of the structure, z0 , was
of 6.1 m (20 ft) at ground level and 3.8 m (12.5 ft) for all other
taken as a random variable with uniform distribution between
floors. The overall height of the structure is 154.7 m (507.5 ft). Col-
0.01 m and 0.03 m. For the parameters of U IM , the distributions
umn splices occur every three floors. The columns are built-up box
indicated in Table 2 of [13] were used. It should be observed that
members while the beams are wide flange standard W sections. A
in this work a ‘‘time-based” P-58 style assessment of the building
summary of the section sizes is reported in Table 1. The steel com-
system is carried out, i.e. losses for all possible wind intensities
posing the frame is assumed to be elastic perfectly plastic and is
over a duration of 1 year are estimated. To perform an ‘‘intensity
therefore completely described by the yield stress ry and Young’s
based” assessment, it would simply be necessary to consider a
modulus Ey , which are considered as uncertain parameters with
fixed wind speed value with desired mean recurrence interval.
log-normal distributions. In particular, the yield stress is assigned
In order to model the aerodynamic loads acting on the frame,
a mean 355 MPa and a standard deviation of 15 MPa while the
the quasi-steady model outlined in Appendix A was adopted while
Young’s modulus has a mean of 210 GPa and standard deviation
considering an influence width for each floor of 12.2 m (40 ft), a
of 15 GPa. The mass of the structure is lumped at each floor and
pressure coefficient of C j ¼ 1:3, and an air density of q ¼ 1:25 kg/
is calculated as the sum of the element mass and dead load mass
m3. In calibrating the model, N was chosen to be 2048 therefore
M ¼ qo L=g, where g is the gravitational acceleration while
yielding wind storms with a stationary duration of T = 16,085 s
qo ¼ 11:96 [kN/m2 ]  h is the dead load with h and L the height
with a sampling frequency of 1.27 Hz. In order to generate realiza-
and influence width of each floor respectively. To model uncer-
tions of the forcing functions FðtÞ, a total of 81,920 independent
tainty, the mass at each floor is multiplied by a log-normal random
and uniformly distributed random numbers in ½0; 2p are required.
variable, DM , with unit mean and standard deviation 0.05. To com-
In the proposed framework, these random numbers are collected in
plete the calibration of U EDP , the modal damping ratios are also
the random vector U IP . A typical realization of the 40th floor forc-
taken as log-normal random variables with mean 0.05 and stan-
ing function, F 40 ðtÞ, and the corresponding interstory drift ratio
dard deviation 0.005 [38,39].
response are shown in Fig. 7, while Fig. 8 shows the target and sim-
For analyzing the dynamic shakedown multipliers, rigid-
ulated power spectral density (PSD) functions.
perfectly plastic hinges are assumed at the extremes of all beams
In defining the fragility and consequence functions, building
and columns. The yield domain associated with the plastic hinges
components were grouped into five main categories, as follows:
of the columns (defined by the interaction between the axial stress
(1) structure; (2) façade; (3) egress; (4) mechanical, electrical
N and bending moment M) is modeled through the piece-wise lin-
and plumbing (MEP); and (5) office fitouts. The FGs included in
the case study together with the characteristics of the fragility
Table 1 functions – chosen from the FEMA fragility database [5] – and unit
Section sizes defining the steel frame. repair cost functions for the first damage state are summarized in
Level range Wide-flange beams Box columns Table 2. In practice, more than one damage state was considered
for each FG (the definition of the fragility and consequence func-
Interior (in.) Exterior (in.)
tions for each additional damage state can be found in [5] using
1–10 W36  282 22  22 20  20 the FEMA designation codes reported in Table 2). For failure modes
t¼3 t ¼ 2:5
11–20 W36  194 20  20 20  20
occurring at the column splices and beam-column connections, six
t¼2 t¼2 different FGs (FGs 2–7) were defined based on the total weight
21–30 W33  169 18  18 18  18 above the component and whether one or two beams were con-
t¼1 t¼1 nected to the beam column joint. Both the fragility functions and
31–40 W27  84 18  18 18  18
the URCFs (unit repair cost functions) were assumed to be lognor-
t ¼ 0:75 t ¼ 0:75
 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758 753

60
1.2
50
1
40
0.8
30
0.6
20 0.4

10 0.2

0 0
0 2000 4000 6000 8000 10000 12000 14000 16000

10-3
1.6

1.4

1.2

0.8

0.6

0.4

0.2

0
0 2000 4000 6000 8000 10000 12000 14000 16000

Fig. 7. A realization of the top floor stochastic forcing function and corresponding interstory drift ratio response: (a) Net force and corresponding effective pressure, i.e. ratio
between story force and building width times story height; (b) interstory drift ratio response.

for FG 32, 40 PGs are defined). Analogously, for FGs 11–26, which
are sensitive to floor acceleration, the components were divided
into 40 PGs, depending on the floor to which they belong while
the peak floor acceleration was assigned as the EDP. FGs 27–31
were MEP facilities assumed to be installed at the top floor, so
(kN2/Hz)

one PG was assigned to each FG. As a result, a total of 899 PGs were
considered.

5.2. Results

The analyses were carried out for N s = 26,000 simulation points.

f plastic collapse was estimated to be Pðsp < 1Þ ¼ 0:0004 while the
Fig. 8. PSD functions for the target and simulated forcing function of floor 40.
mally distributed with dispersion bf for the fragility functions and
bc ¼ 0:25 for all URCFs. The median of the fragility functions is
indicated with mf in Table 2, while the median of the URCF is
defined by the four values C max ; C min ; Q max and Q min and is fixed
for a given wind storm (i.e. simulation point) by the quantity of
damaged components.
The PGs were identified as follows. PGs belonging to FGs 1–10
and 32 are defined by interstory drift sensitive components and
were therefore assigned peak interstory drifts as EDPs. The PGs
were therefore defined by grouping elements at each floor (e.g.
Fig. 9 reports the distributions of sp and se . In particular, for a
threshold value sp ¼ 1, i.e. no amplification of the forcing function
FðtÞ, the annual system-level probability of the structure suffering
annual probability of the structure exiting the elastic regime was
estimated to be Pðse < 1Þ ¼ 0:039. The difference between the
two curves indicates the significant plastic reserves of the frame
under investigation. In addition to the information above, the dis-
tributions of sp and se also provide useful information on the sen-
sitivity of the probabilities associated with undergoing plastic
collapse or exiting the elastic regime as a result of amplifications
in the external excitation.
Fig. 10 shows annual probability of exceedance associated with
total repair costs. As can be seen, the total repair cost with a return
period of 5 years (Pf ¼ 0:2) is about $2,267,000, indicating that the
structure was poorly designed and underperformed, while the cost
with a 50 year return period (P f ¼ 0:02) is approximately
754 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758

Table 2
Summary of the FGs and associated fragility functions and URCFs for the first Damage State.

FG Category FEMA Description EDP Damage state 1

number designation type
Fragility Unit repair cost functions
functions
mf bf C max [$] C min [$] Q max Q min

1 Structure B1031.011c Steel column base plates Drift 0.04 0.4 25,812.00 17,208.00 15 5
2 Structure B1031.021b Column splice (150 < W < 300 plf) Drift 0.04 0.4 12,168.00 8112.00 15 5
3 Structure B1031.021c Column splice (W > 300 plf) Drift 0.04 0.4 13,608.00 9072.00 15 5
4 Structure B1035.041 Pre-Northridge Welded Unreinforced Flange-Bolted (WUF-B) Drift 0.017 0.4 14,472.00 9648.00 30 10
beam-column joint, beam one side
5 Structure B1035.042 Pre-Northridge WUF-B beam-column joint, beam one side of Drift 0.017 0.4 15,072.00 10,048.00 30 10
column, beam depth >= W30
6 Structure B1035.051 Pre-Northridge WUF-B beam-column joint, beam both sides Drift 0.017 0.4 21,096.00 14,064.00 30 10
7 Structure B1035.052 Pre-Northridge WUF-B beam-column joint, beam both sides of Drift 0.017 0.4 21,096.00 14,064.00 30 10
column, beam depth >= W30
8 Façade B2011.201a Precast concrete panels with user-specified in-plane Drift 0.005 0.5 17,160.00
connection capacity
9 Fitout C1011.001a Wall Partition, Type: Gypsum with metal studs, Full Height, Drift 0.0021 0.6 2.50 2.04 13,000 1300
Fixed Below, Fixed Above (based upon 130  1000 Panels)
10 Fitout C3011.001a Wall Partition, Type: Gypsum + Wallpaper, Full Height, Fixed Drift 0.0021 0.6 3.90 2.40 5000 500
Below, Fixed Above (based upon 90  1000 Panels)
11 Fitout C3027.001 Raised Access Floor, non seismically rated. Acc. 0.5 0.5 1.27 1.04 10,000 1000
12 Fitout C3032.001b Suspended Ceiling, SDC A, B, Area (A): 250 < A < 1000, Vert Acc. 0.55 0.4 1.89 1.31 2500 250
support only
13 Fitout C3034.001 Independent Pendant Lighting – non seismic Acc. 0.6 0.4 643.50 396.00 10 5
14 Fitout E2022.023 Desktop electronics including computers, monitors, stereos, Acc. 0.4 0.5 1000.00
etc., smooth surface
15 MEP D2021.011a Cold Water Piping (dia > 2.5 in.), SDC A or B, PIPING FRAGILITY Acc. 1.5 0.4 700.00 210.00 10 5
16 MEP D2022.011a Hot Water Piping – Small Diameter Threaded Steel – (2.5 in. in Acc. 0.55 0.5 279.00
diameter or less), SDC A or B, PIPING FRAGILITY
17 MEP D2022.011b Hot Water Piping – Small Diameter Threaded Steel – (2.5 in. in Acc. 1.2 0.5 383.00
diameter or less), SDC A or B, BRACING FRAGILITY
18 MEP D2022.021a Hot Water Piping – Large Diameter Welded Steel – (greater Acc. 1.5 0.5 348.00
than 2.5 in. in diameter), SDC A or B, PIPING FRAGILITY
19 MEP D2031.021a Sanitary Waste Piping – Cast Iron w/bell and spigot couplings, Acc. 2.25 0.5 3167.00
SDC A, B, PIPING FRAGILITY
20 MEP D2031.021b Sanitary Waste Piping – Cast Iron w/bell and spigot couplings, Acc. 1.2 0.5 423.00
SDC A, B, BRACING FRAGILITY
21 MEP D3041.011a HVAC Stainless Steel Ducting less than 6 sq. ft in cross Acc. 1.5 0.4 1300.00 390.00 10 5
sectional area, SDC A or B
22 MEP D3041.012a HVAC Stainless Steel Ducting less than 6 sq. ft in cross Acc. 1.5 0.4 1900.00 570.00 10 5
sectional area, SDC C
23 MEP D3041.031a HVAC Drops/ Diffusers in suspended ceilings – No Acc. 1.3 0.4 360.00 240.00 50 10
independent safety wires, SDC A or B
24 MEP D3041.041a Variable Air Volume (VAV) box with in-line coil, SDC A or B Acc. 1.9 0.4 14,796.00
25 MEP D4011.021a Fire Sprinkler Water Piping – Horizontal Mains and Branches – Acc. 1.1 0.4 348.00
Old Style Victaulic – Thin Wall Steel – No bracing, SDC A or B,
PIPING FRAGILITY
26 MEP D4011.031a Fire Sprinkler Drop Standard Threaded Steel – Dropping into Acc. 0.75 0.4 526.00
unbraced lay-in tile SOFT ceiling – 6 ft. long drop maximum,
SDC A or B
27 MEP D5012.021a Low Voltage Switchgear – Capacity: 100 to <350 Amp – Acc. 1.28 0.4 550.00 450.00 3 1
Unanchored equipment that is not vibration isolated –
Equipment fragility only
28 MEP D3031.011c Chiller – Capacity: 350 to <750 Ton – Unanchored equipment Acc. 0.2 0.4 2420.00 1980.00 4 1
that is not vibration isolated – Equipment fragility only
29 MEP D3031.021c Cooling Tower – Capacity: 350 to <750 Ton – Unanchored Acc. 0.5 0.4 2420.00 1980.00 4 1
equipment that is not vibration isolated – Equipment fragility
only
30 MEP D3052.011d Air Handling Unit – Capacity: 25,000 to <40,000 CFM – Acc. 0.25 0.4 2066.00
Unanchored equipment that is not vibration isolated –
Equipment fragility only
31 MEP D5012.013a Transformer/primary service – Capacity: <100 kVA – Acc. 0.73 0.45 4167.00
Equipment that is either hard anchored or is vibration isolated
with seismic snubbers/restraints – Anchorage fragility only
32 Egress C2011.001b Prefabricated steel stair with steel treads and landings with no Drift 0.005 0.6 520.00 320.00 5 1
seismic joint.

$4,564,000. In interpreting these numbers, it should be observed
that it is generally acknowledged that the time and cost conse-
quences of the 2012 FEMA fragility database (as used in this work)
associated with interior partitions, gravity shear tabs and in plane
precast panels are somewhat conservative leading to relatively
high repair cost and time estimates. In the proposed approach,
the repair costs are calculated at an individual component-level
before being aggregated for estimating the total system-level con-
sequence. Therefore, by deaggregating over all FGs, information on
how the various FGs contribute to the total repair cost can be
found. As can be seen from Fig. 11, FG 8–14 (concrete panels and
office fitouts) and 32 are responsible for the bulk of the repair costs
 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758 755

0.8

0.6

0.4

0.2
Elastic multiplier
Shakedown multiplier

0
0 2 4 6 8 10 12 14

Fig. 9. Distributions of the elastic, se , and dynamic shakedown, sp , safety factors.

Fig. 11. Deaggregation of the total repair cost among the fragility groups.
1

0.8 1
In Series
In Parallel
0.6 0.8
Possible repair strategy

0.4 0.6

0.2 0.4

0 0.2
104 105 106 107 108
C[$(US)] 0
102 103 104
Fig. 10. Annual probability of exceedance of the total repair cost C.

Fig. 12. Annual probability of exceedance of the total downtime T D .

while FG 4–7, 23, 24, 30 and 31 contribute minimally to the total
cost. This information would be particularly useful in the redesign
or retrofit of the building and illustrates the type of decision sup-
port information that the proposed framework has been developed
to provide.
In parallel with the repair cost estimation, downtime for full
recovery was also assessed. In this respect, Fig. 12 shows the
annual probability of exceedance of total downtime thresholds.
In particular, upper and lower bounds were estimated considering
the repair work done in series or parallel respectively. The corre-
sponding total downtime with a return period of 5 years
(P f ¼ 0:2) is about 726 and 230 days, respectively. In addition, a
strategy in which repairs were carried out floor by floor in a serial
manner (i.e. all structural and non-structural components repaired
simultaneously at a certain floor before repairs can commence at
the next floor) was also considered. In this case, the estimated
downtime with a return period of 5 years (P f ¼ 0:2) is around
626 days which, consistent with how the repair strategy was
defined, is between the two extreme scenarios.
Fig. 13 shows the deaggregation of the system-level total repair
time among all fragility groups. In particular, it can be seen that
non-structural components, such as concrete panels and office fit-
outs, contribute mostly to the total repair cost, independently of
whether the repairs are assumed to be carried out in series or in
parallel. A breakdown of the different downtime contributors for
the intermediate repair strategy considered above as well as the
deaggregation of the impeding factors is illustrated in Fig. 14. It
can be observed that the delay caused by the impeding factors, lar-
gely driven by the delays associated with contractor mobilization,
contributes – in this example – almost as much as the repair work
time to the total downtime. This simple example would suggest
that, the total downtime may be greatly underestimated if delays
that prevent the initiation of the repairs are neglected, as in the
FEMA P-58 procedures. Moreover, for this case study, utility dis-
ruption does not control over impeding factors and therefore – in
this case – does not have a direct impact on the downtime
assessment.
5.3. Discussion
The primary objective of this work was the introduction of an
efficient PBWE framework for the integrated collapse and non-
collapse assessment of multistory wind excited buildings systems.
The case study presented in this section demonstrated the poten-
tial of the framework for assessing the performance of a system
given a set of input data with associated probabilistic distributions.
Obviously, the quality of the performance estimates obtained from
the proposed framework depends on the correct estimation of the
input parameters. Having said this, it should be kept in mind that
the framework presented in this work is based on Monte Carlo
simulation. Therefore, the sensitivity of the results can be studied
in a straightforward manner with respect to any of the input
parameters. This aspect was not investigated in this paper as it
756 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758

300

250

200

150

100

50

0
Utility Disruption Impeding Factors Repair Work

200

150

100

50

0
Inspection Engineering Financing Contractor Permitting

Fig. 13. Deaggregation of the total repair time: (a) repair work carried out in series; Fig. 14. (a) Downtime contributors; (b) deaggregation of impeding factors for the
(b) repair work carried out in parallel. structure.

was focused on the development of the framework itself. However,
a separate study into this very aspect would be of sure interest.
Before closing this section, it should be mentioned that, while
the case study presented here is 2D, the extension to 3D problems
would be straightforward. Indeed, due to the efficiency with which
high dimensional LP problems can be solved, an extension to 3D
should not present any particular computational difficulties.
mating not only the integrated system-level collapse and non-
collapse losses, but also the deaggregated non-collapse losses.
Due to the efficiency of the procedure and its systematic estima-
tion of both structural and non-structural damage, it is believed
that the framework outlined in this work could be of significant
and immediate interest to the designers of wind excited building
systems.

6. Summary and conclusions Acknowledgements

A probabilistic PBWE framework is proposed in this paper that
integrates probabilistic dynamic shakedown analysis with system-
level damage and loss estimation models. In particular, perfor-
mance is expressed at a system-level through metrics such as
repair costs and downtime, therefore enabling straightforward
interpretation by decision-makers of diverse technical back-
grounds. With respect to downtime estimation, a model is devel-
oped that depends not only on the estimation of repair work
time but also on the estimation of delays caused by impeding fac-
tors that prevent the start of actual repair work. By applying
dynamic shakedown theory, an efficient model is defined for
describing the post-yield behavior of structural systems subject
to long duration (order of hours) dynamic wind loads. By defining
safety against collapse as the occurrence of dynamic shakedown,
collapse scenarios for the MWFRS are efficiently estimated and
integrated into the overall loss model. To illustrate the potential
of the proposed PBWE framework, a case study consisting of a
40-story frame building was presented. This allowed for the illus-
tration of the capability of the framework for systematically esti-
This research effort was in part supported by the National
Science Foundation (NSF) under Grant No. CMMI-1562388 and
CMMI-1462084. This support is gratefully acknowledged.
Appendix A. Quasi-steady stochastic wind loading model
The quasi-steady model used in this paper for generating the
forcing function Fðt; v

z ; uIP Þ is related to the fluctuating and mean
wind speeds through:
F j ðt; v
z ; uIP Þ ¼ gj ðv
zj þ v j ðtÞÞ2 ’ gj ðv
2zj þ 2v
zj v j ðtÞÞ;
j ¼ 1; 2 . . . ; n ð12Þ
where n is the number of floors; v
zj is the mean wind velocity at
height zj determined from the wind hazard analysis; v j ðtÞ is the
corresponding fluctuating component of the wind speed; gj is a
coefficient equal to 0:5qC j Aj , where q is the air density, C j is a
quasi-steady pressure coefficient and Aj is the influence area of
the jth degree of freedom in the direction of the mean wind.
 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758 757

The Power Spectral Density (PSD) function of the longitudinal Appendix B. Dynamic shakedown for infinite duration periodic
component of the fluctuating wind velocity, v j ðtÞ, was modeled loads
as proposed in [40] and therefore as:
Consider an elastoplastic structure with the vectors qðtÞ and
1 200 2 zj 1
Sj ðxÞ ¼ v  5=3 ; j ¼ 1; 2; . . . ; n ð13Þ Q ðtÞ collecting the time varying generalized strains and stresses
2 2p  v
zj xz
1 þ 50 2pv
jz induced by the external wind loads FðtÞ. Equilibrium and geomet-
j
ric compatibility can then be expressed in terms of qðtÞ and Q ðtÞ
as:
where v  is the shear velocity of the flow given by:
ka € t ðtÞ  Ctt u_ t ðtÞ
BT Q ðtÞ ¼ FðtÞ  Mtt u ð22Þ
v  ¼ v 10 b   ð14Þ
10
ln z0 qðtÞ ¼ But ðtÞ ð23Þ
where ka ¼ 0:4 is the Von Kármán’s constant while z0 is the where B is the strain-displacement (compatibility) matrix that
ground roughness height. The cross power spectral density was depends on the undeformed geometry of the system, ut ðtÞ is the
defined as: response of the system in the dynamically significant degrees of
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi freedom (DOFs) – with over-dot indicating derivation with respect
Sjk ðxÞ ¼ Sj ðxÞSk ðxÞcjk ðxÞ; j; k ¼ 1; 2; . . . ; n; j – k ð15Þ to time – Ctt is the damping matrix associated with the dynamic
DOFs, and Mtt is the mass matrix associated with the dynamic DOFs.
where cjk is the coherence function between v j ðtÞ and v k ðtÞ. The The strain vector qðtÞ is defined as the sum of an elastic, eðtÞ, and a
model suggested in [41] was used for the coherence function plastic, pðtÞ, strain. The elasticity equations for this system can be
between the fluctuating velocities at heights z1 and z2 : expressed as:
" #
x C z Dz Q ðtÞ ¼ DeðtÞ þ Q  ðtÞ ð24Þ
cjk ðDz; xÞ ¼ exp  ð16Þ
2p 12 ðv
1 þ v
2 Þ
where D is the block diagonal matrix collecting the elastic stiffness
matrices of each element while Q  is a vector of the perfectly
where v
1 and v
2 are the mean wind speeds at heights z1 and z2 ,
clamped element generalized stresses.
respectively, Dz ¼ jz1  z2 j is the difference between two heights
In order to satisfy the yield criterion, the generalized stresses,
and C z is a constant that can be set equal to 10 for design purposes
i.e. the internal forces, at each cross section cannot lie outside
[41].
the yield surface. This implies that the following must hold:
In order to simulate the n-dimensional multivariate stochastic
process v ðtÞ, the cross-spectral density matrix SðxÞ, defined with f ¼ NT Q ðtÞ  R 6 0 ð25Þ
diagonal components given by Eq. (13) and off-diagonal terms
given by Eq. (15), is decomposed into the following form: where f is a piece-wise linearized yield vector, N is the block diag-
onal matrix of unit external normals to the yield surface while R is a
SðxÞ ¼ HðxÞH ðxÞ T
ð17Þ plastic resistance vector that depends on the yield stress, ry , of the
material. To account for possible uncertainty in the material
where ðÞT is the transpose of the complex conjugate. This decom-
strength, ry can be taken as uncertain. Under these circumstances,
position can be performed using Cholesky’s method or eigenfunc-
ry would be a component of the uncertain vector U sh the collects
tion expansion.
uncertain parameters associated with the dynamic shakedown
Once the cross-spectral density matrix is decomposed, the com-
analysis.
ponents of the stochastic vector process v ðtÞ can be simulated by
A necessary and sufficient condition for dynamic shakedown
the following series as N ! 1 [42,43]:
[27,37] to occur is that there exists a finite time r P 0 such that
X
n X
N pffiffiffiffiffiffiffiffi the sum of a free motion stress field Q F ðtÞ, a time independent
v j ðtÞ ¼ 2 jHjm ðxml Þj Dx cos½xml ðtÞ  hjm ðxml Þ þ /ml ;
self-stress distribution q representing the initial plastic strains
m¼1 l¼1
along with the elastic stress response to the loads backward trun-
j ¼ 1; 2; . . . ; n ð18Þ
cated at r; Q E ðr þ sÞ, proves to be inside the yield surface at any
where /ml for m ¼ 1; 2; . . . ; n and l ¼ 1; 2; . . . ; N are sequences of time s 0:
independent random phase angles uniformly distributed in ½0; 2p
f ¼ NT ½Q E ðr þ sÞ þ Q F ðsÞ þ q  R < 0; 8s P 0 ð26Þ
and collected in the vector U IP ; xml is given by
m In order to estimate the safety factor against inadaptation (i.e.
xml ¼ ðl  1ÞDx þ Dx; l ¼ 1; 2; . . . ; N ð19Þ non-shakedown) of the structure, a scalar multiplier s > 0 of the
n
forcing function is introduced so that the following scaled loads
with Dx the sampling frequency; while hjm is a complex angle that can be defined F s ðtÞ ¼ sFðtÞ. For this time dependent loading sys-
can be written in the following form if the off-diagonal elements tem, the shakedown safety factor is defined as the threshold
Hjm ðxÞ of HðxÞ are written in the polar form: value s, indicated by sp , for which shakedown will always occur
Im½Hjm ðxÞ if s < sp .
hjm ðxÞ ¼ tan1 ð20Þ In the case of periodic external actions, sp can be estimated by
Re½Hjm ðxÞ
dropping the free vibration stress and setting r ¼ 0 as it is always
where Im and Re are imaginary and real parts of the complex num- possible to find some particular initial conditions that will make
ber Hjm ðxÞ. the purely elastic response of the system coincide with its forced
The period of the function simulated in this way is given by: vibration counterpart [27,37,32]. Accordingly, the elastic backward
truncated stress response Q E ðr þ sÞ can be represented by the
2pn 2pnN
Tb ¼ ¼ ð21Þ steady state response Q S . Therefore, the shakedown safety factor
Dx xup
can be estimated by solving the following linear programming
where xup is the cut-off frequency. problem:
758 W.-C. Chuang, S.M.J. Spence / Engineering Structures 150 (2017) 746–758

sp ¼ max s [15] Larsen R, Klemencic R, Hooper J, Aswegan K. Engineering objectives for

s;q
performance-based wind design. In: Proceedings of the geotechnical and
subject to structural engineering congress. Phoenix, AZ, USA; 2016.
[16] Vickery BJ. Wind action on simple yielding structures. J Eng Mech Div

s ¼ max NT Q s ðtÞ
Q ð27Þ 1970;96:107–20.
06t6T
[17] Wyatt TA, May HI. The ultimate load behavior of structures under wind
f ¼ sQ  s þ NT q  R 6 0 loading. In: Proceedings of the 3rd international conference on wind effects on
buildings and structures. Tokyo, Japan; 1971. p. 501–10.
BT q ¼ 0 [18] Tsujita O, Hayabe Y, Ohkuma T. A study on wind-induced response for inelastic
structure. In: Proceedings of the 7th international conference on structural

s is the vector of elastic envelope stress defined as the max-
where Q safety and reliability (ICOSSAR 97). Kyoto, Japan; 1998. p. 1359–66.
imum of the plastic demand for each yielding mode in time, Q S ðtÞ is [19] Ohkuma T, Kurita T, Ninomiya M. Response estimation based on energy
balance for elasto-plastic vibration of tall building in across-wind direction. In:
the purely elastic steady state response while T represents the per- Proceedings of the 7th international conference on structural safety and
iod of the forcing function. reliability (ICOSSAR 97). Kyoto, Japan; 1998. p. 1379–86.
It is interesting to observe that, together with optimization [20] Chen D, Davenport AG. Vulnerability of tall buildings in typhoons. In:
Advances in structural dynamics, vol. 2; 2000. p. 1455–62.
problem outlined above, the following linear programming prob- [21] Tamura Y, Yasui H, Marukawa H. Non-elastic responses of tall steel buildings
lem may also be defined for identifying the load multiplier se that subjected to across-wind forces. Wind Struct 2001;4(2):147–62.
separates a perfectly elastic response state from the initiation of [22] Gani F, Légeron F. Relationship between specified ductility and strength
demand reduction for single degree-of-freedom systems under extreme wind
plasticity: events. J Wind Eng Ind Aerod 2011;109:31–45.
se ¼ max s [23] Judd J, Charney F. Inelastic behavior and collapse risk for buildings subjected to
s wind loads. In: Ingraffea N, Libby M, editors. Structures congress 2015. April
subject to 23–25, Portland, Oregon, USA; 2015, p. 2483–96.

s ¼ max NT Q s ðtÞ ð28Þ [24] Feng C, Chen X, Liang D. Inelastic crosswind responses of tall buildings with
Q bilinear hysteretic restoring force characteristics. In: 8th International
06t6T
colloquium on bluff body aerodynamics and applications (BBAA VIII).

R60
f ¼ sQ
s
Boston, Massachusetts, USA; 2016, p. 1207–14.
[25] Vamvatsikos D, Cornell CA. Incremental dynamic analysis. Earthq Eng Struct
where now the vector of plastic strains q is considered equal to 2002;31(3):491–514.
[26] Tabbuso P, Spence SMJ, Palizzolo L, Pirrotta A, Kareem A. An efficient
zero. framework for the elasto-plastic reliability assessment of uncertain wind
excited systems. Struct Saf 2016;58:69–78.
References [27] Ceradini G. Dynamic shakedown in elastic-plastic bodies. J Eng Mech Div
1980;106:481–99.
[28] König J, Maier G. Shakedown analysis of elastoplastic structures: a review of
[1] Moehle J, Deierlein GG. A framework methodology for performance-based
recent developments. Nucl Eng Des 1981;66:81–95.
earthquake engineering. In: 13th World conference on earthquake
[29] Maier G, Munro J. Mathematical programming methods in engineering plastic
engineering. Vancouver; 2004.
analysis. Appl Mech Rev 1982;35(12):1631–43.
[2] Yang TY, Moehle J, Stojadinovic B, Der Kiureghian A. Seismic performance
[30] König J. Shakedown of elastic-plastic structures. Fundamental studies in
evaluation of facilities: methodology and implementation. J Struct Eng
engineering. New York: Elsevier; 1987. ISBN 9780444989796.
2009;135(10):1146–54.
[31] Maier G, Carvelli V, Cocchetti G. On direct methods for shakedown and limit
[3] Federal Emergency Management Agency (FEMA). Seismic performance
analysis. Eur J Mech A-Solid 2000;19:S79–S100.
assessment of buildings, Volume 1 – Methodology (FEMA Publication P-58-
[32] Polizzotto C, Borino G, Caddemi S, Fuschi P. Theorems of restricted dynamic
1). Washington, DC; 2012a.
shakedown. Int J Mech Sci 1993;35:787–801.
[4] Federal Emergency Management Agency (FEMA). Seismic performance
[33] Minciarelli F, Gioffrè M, Grigoriu M, Simiu E. Estimates of extreme wind effects
assessment of buildings, Volume 2 – Implementation (FEMA Publication P-
and wind load factors: influence of knowledge uncertainties. Prob Eng Mech
58-2). Washington, DC; 2012b.
2001;16:331–40.
[5] Federal Emergency Management Agency (FEMA). Seismic performance
[34] Spence SMJ, Kareem A. Data-enabled design and optimization (DEDOpt): tall
assessment of buildings, Volume 3 – Supporting electronic materials and
steel buildings frameworks. Comput Struct 2013;134(12):134–47.
background documentation (FEMA Publication P-58-3). Washington, DC;
[35] Bonowitz D. Resilience criteria for seismic evaluation of existing buildings. A
2012c.
2008 special projects initiative report to structural engineers association of
[6] Ciampoli M, Petrini F, Augusti G. Performance-based wind engineering:
Northern California (SEAONC); 2011.
towards a general procedure. Struct Saf 2011;33(6):367–78.
[36] Almufti I, Willford M. Resilience-based earthquake design initiative (REDi) for
[7] Petrini F, Ciampoli M. Performance-based wind design of tall buildings. Struct
the next generation of buildings; 2013.
Infrastruct E 2012;8(10):954–66.
[37] Polizzotto C. Dynamic shakedown by modal analysis. Meccanica
[8] Spence SMJ, Gioffrè M. Large scale reliability-based design optimization of
1984;19:133–44.
wind excited tall buildings. Prob Eng Mech 2012;28:206–15.
[38] Spence SMJ, Kareem A. Tall buildings and damping: a concept-based data-
[9] Bernardini E, Spence SMJ, Kareem A. A probabilistic approach for the full
driven model. J Struct Eng 2014;140(5). 04014005-1–15.
response estimation of tall buildings with 3D modes using the HFFB. Struct Saf
[39] Spence SMJ, Bernardini E, Guo Y, Kareem A, Gioffrè M. Natural frequency
2013;44:91–101.
coalescing and amplitude dependent damping in the wind-excited response of
[10] Seo DW, Caracoglia L. estimating life-cycle monetary losses due to wind
tall buildings. Prob Eng Mech 2014;35:108–17.
hazards: fragility analysis of long-span bridges. Eng Struct 2013;56:593–1606.
[40] Kaimal JC, Wyngaard JC, Izumi Y, Coté OR. Spectral characteristics of surface-
[11] Caracoglia L. A stochastic model for examining along-wind loading uncertainty
layer turbulence. Quart J Roy Meteorol Soc 1972;98(417):563–89. http://dx.
and intervention costs due to wind-induced damage on tall buildings. Eng
doi.org/10.1002/qj.49709841707.
Struct 2014;78:121–32.
[41] Davenport GA. The dependence of wind load upon meteorological parameters.
[12] Bobby S, Spence SMJ, Bernardini E, Kareem A. Performance-based topology
In: of Toronto Press U, editor. Proceedings of the international research
optimization for wind-excited tall buildings: a framework. Eng Struct
seminar on wind effects on building and structures; 1967. p. 19–82.
2014;74:242–55.
[42] Deodatis G. Simulation of ergodic multivariate stochastic processes. J Eng
[13] Spence SMJ, Kareem A. Performance-based design and optimization of
Mech 1996;122:778–87.
uncertain wind-excited dynamic building systems. Eng Struct
[43] Chen X, Kareem A. Proper orthogonal decomposition-based modeling,
2014;78:133–44.
analysis, and simulation of dynamic wind load effects on structures. J Eng
[14] Bernardini E, Spence SMJ, Kwon DK, Kareem A. Performance-based design of
Mech 2005;131(4):325–39.
high-rise buildings for occupant comfort. J Struct Eng 2015;41(10). http://dx.
doi.org/10.1061/(ASCE)ST.1943-541X.0001223.