A Probabilistic Approach For The Prediction of Seismic Resilience of Bridges

EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS
Earthquake Engng Struct. Dyn. 2013; 42:1469–1487

Published online 25 January 2013 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.2282
A probabilistic approach for the prediction of seismic resilience of

bridges
Alberto Decò, Paolo Bocchini and Dan M. Frangopol*,†

Department of Civil and Environmental Engineering, Engineering Research Center for Advanced Technology for Large
Structural Systems (ATLSS), Lehigh University, Bethlehem, PA 18015-4729, USA
ABSTRACT
This paper proposes a probabilistic approach for the pre-event assessment of seismic resilience of bridges,
including uncertainties associated with expected damage, restoration process, and rebuilding/rehabilitation
costs. A fragility analysis performs the probabilistic evaluation of the level of damage (none, slight, moderate,
extensive, and complete) induced on bridges by a seismic event. Then, a probabilistic six-parameter sinusoidal-
based function describes the bridge functionality over time. Depending on the level of regional seismic hazard,
the level of performance that decision makers plan to achieve, the allowable economic impact, and the available
budget for post-event rehabilitation activities, a wide spectrum of scenarios are provided. Possible restoration
strategies accounting for the desired level of resilience and direct and indirect costs are investigated by
performing a Monte Carlo simulation based on Latin hypercube sampling. Sensitivity analyses show how the
recovery parameters affect the resilience assessment and seismic impact. Finally, the proposed approach is
applied to an existing highway bridge located along a segment of I-15, between the cities of Corona and
Murrieta, in California. Copyright © 2013 John Wiley & Sons, Ltd.
Received 22 March 2012; Revised 18 December 2012; Accepted 19 December 2012
KEY WORDS: probabilistic resilience; seismic hazard; bridge; recovery; cost; decision analysis
1. INTRODUCTION
Transportation systems must guarantee the reliable and efficient flow of people, goods, and services. The
condition of bridge infrastructures plays a key role with respect to the functionality of transportation
networks. The effects induced by natural and/or man-made extreme events may lead to sudden
deterioration of infrastructure conditions, causing potential traffic disruptions and economic losses. In
this context, a prompt emergency response may limit consequences and restore functionality.
In studies dealing with seismic hazard, resilience can be considered as a performance indicator that
quantifies the residual functionality along with the effort of the society in responding to a seismic
event. In the last decade, some progress has been achieved by the introduction of analytical
deterministic definitions of resilience [1–3], many of them focusing on post-event recovery analysis.
Further developments [4–6] opened the path for the probabilistic treatment of resilience, in order to
develop models oriented to its prediction.
The scope of this paper is to provide a new methodology to evaluate the probabilistic seismic
resilience (PSR) of bridges and assess the impact in terms of direct and indirect costs. The proposed
approach does not aim at providing a tool for assisting decision making for the post-event disaster
recovery, but a tool that performs pre-event probabilistic comparison of the outcomes of various
*Correspondence to: Dan M. Frangopol, Department of Civil and Environmental Engineering, Engineering Research Center
for Advanced Technology for Large Structural Systems (ATLSS), Lehigh University, Bethlehem, PA 18015-4729, USA.
†
E-mail: dan.frangopol@lehigh.edu
Copyright © 2013 John Wiley & Sons, Ltd.

1470 A. DECÒ, P. BOCCHINI AND D. M. FRANGOPOL
scenarios in terms of resilience. This pre-event evaluation tool can be used for decisions concerning
proactive maintenance, retrofit, or life-cycle management for key components of the infrastructure
systems. The inclusion of uncertainty extends resilience quantification towards a fully probabilistic
approach that can be embedded in a comprehensive structure/infrastructure management tool, based on
expected and characteristic values. The significant difference between this paper and other previous
studies on seismic resilience is that the pre-event probabilistic assessment of resilience is herein
considered as a support tool for decisions making within the bridge life-cycle. The estimation of
the residual functionality after an exceptional event (called ‘robustness’ [6]) is a process affected by a
large amount of uncertainty, and the subsequent response phase adds further uncertainty to the
prediction of resilience. The probabilistic quantification of the seismic-induced damage on bridges is
performed by means of fragility analyses. On the basis of the assessed damage level (probability of the
bridge being in a particular damage state), a recovery function is proposed. Bridge functionality is
directly influenced by restoration activities. Different damage-functionality relationships are adopted on
the basis of the data provided in [7] and the expert opinions reported in [8]. A six-parameter sinusoidal-
based recovery model proposed in [9] is used in this paper. The ability of such model to represent
different types of rehabilitation options is exploited.
In this paper, consequence analysis is performed in a probabilistic context with respect to direct and
indirect costs. Direct costs are those associated with rebuilding-rehabilitation expenses and directly
related to the level of bridge damage. Indirect costs account for the losses caused by additional
travel time and distance when a bridge is partially or fully closed.
The final outcome of an extreme event in terms of consequences is strongly affected also by the
different choices of the decision makers. For this reason, the proposed model investigates multiple
potential strategies, and tries to identify the most likely (i.e., those which appear most convenient)
for different values of available monetary resources and target recovery performance (which is often
a function of the relative importance of a bridge within a transportation network). Uncertainties are
included in the six-parameter recovery model as well as in the cost analysis. A Monte Carlo
simulation based on Latin hypercube sampling [10] is performed, and expected values and standard
deviations of PSR are obtained for a bridge carrying a segment of the northbound I-15 crossing the
Temescal Wash, located between the cities of Corona and Murrieta in California. For this case
study, a sensitivity analysis with respect to different recovery parameters is performed and
discussed. Although the proposed approach can potentially be used for the pre-event assessment of
resilience of entire transportation networks, the study on a single bridge performed in this paper is
necessary in order to explain and investigate all the introduced parameters in detail.
2. RESILIENCE
Various definitions of resilience can be found in several disciplinary fields [5, 11–13]. According to
[1], seismic resilience is ‘the ability of social units (e.g., organizations and communities) to mitigate
hazards, contain the effects of disasters when they occur, and carry out recovery activities in ways
that minimize social disruption and mitigate the effects of future earthquakes’. However, the
qualitative definitions of resilience have not converged yet to a single generally accepted definition,
because resilience covers social and technical aspects too broadly. Therefore, the quantification of
resilience R can lead to several slightly different analytical definitions [1, 6, 14]. In this paper, the
following definition is adopted [2]:
Zth
Qðt Þdt
t0
R¼ (1)
th t0
where Q(t) is the time-dependent functionality, t is the time, t0 is the time at which the seismic event
occurs, and th is the investigated time horizon (Figure 1). The functionality of a structure or
infrastructure can generally be defined as its ability to provide adequate service to the users.
Copyright © 2013 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2013; 42:1469–1487
DOI: 10.1002/eqe
PROBABILISTIC EVALUATION OF SEISMIC RESILIENCE OF BRIDGES 1471
100%
80%
Functionality Q(t)
60%
40%
20%
0%
Time t
Figure 1. Schematic representation of resilience and rapidity including uncertainties that are represented by
qualitative probability density function (PDFs). Uncertainties represented by hatched qualitative PDFs are
accounted for in the proposed probabilistic model.
Functionality is usually expressed as a percentage of the service associated with the intact structure
(which is set to be 100%), thus it is dimensionless. Therefore, resilience as defined by Equation (1)
is dimensionless as well.
The effort of a community to recover from damage and activity disruptions caused by an extreme
event can also be evaluated by means of ‘rapidity’. As introduced in [1] and subsequently
quantitatively defined in [6], rapidity r is herein considered as:

Q tf Q½t0
r ¼ arctan t0 ≤t≤tf ¼ minðtr ; th Þ (2)
tf t0
where tf is the time at which the recovery process ends.

Figure 1 shows the classic schematic representation of resilience and rapidity, considering a single
important disruptive event. The functionality pattern is first determined by the pre-event level of
functionality and the potential occurrence of an event, which induces damage to the structure (point
A). If such event occurs, the functionality suddenly drops to a residual value (point B).
Subsequently, the post-event recovery is actuated in two phases. The first phase is represented by
the idle time interval di, during which the functionality remains constant until time ti, and the
restoration has not yet started (B–C). The second phase includes the restoration leading to point D
in Figure 1. The potential occurrence of aftershocks [15] may cause further losses of functionality,
usually more difficult to be quantified, which could lead to a different resilience. Because this paper
provides a tool for pre-event decision making, the inclusion of aftershocks in the proposed model is
not likely to improve the quality of the decision process. In fact, including aftershocks in this
approach would significantly increase the complexity of the model and will require additional input
data, obtaining a higher uncertainty of the results against a marginal increase in the accuracy of the
prediction and a likely negligible impact on the pre-event decision making process.
The main uncertainties are those represented in Figure 1 by qualitative probability density
functions (PDFs) affecting the time-dependent functionality. The proposed PSR model includes the
uncertainties represented by the hatched PDFs (Figure 1), treating the overall recovery process as a
function of a set of random variables. This is, by definition, the ‘parametric representation’ of a
stochastic process. Representations based on analytic expressions that depend on finite collections of
random variables are the foundation of the most popular methodologies for the simulation of
random processes, such as the discrete spectral representation, the truncated Karhunen–Loeve series
expansion, and the truncated polynomial chaos series expansion [16]. The reason that led to the
DOI: 10.1002/eqe
choice of a small number of random parameters is that the data available for the probabilistic
calibration of this stochastic process is scarce, and trying to estimate more parameters or a complete
autocorrelation function was deemed unrealistic. The ‘shapes’ that the recovery profile can assume
in the presented numerical application are taken from results available in the literature. Similarly, the
marginal distributions of the parameters were extrapolated from historical data and expert surveys.
This is found to be a reasonable compromise between the need of a probabilistic description of the
phenomenon and the scarcity of data for its statistical characterization.
3. PROBABILISTIC RECOVERY MODEL
A community affected by the consequences of an earthquake needs to recover its former status in a
time efficient way. For a given earthquake, the exact predictions of the damage level and loss of
functionality are impossible, making the overall recovery process even more uncertain. The pre-
event deterministic assessment of seismic resilience may lead to misjudgment that could head
towards an underestimation of the consequences.
The assessment of PSR and of the economic loss induced by potential traffic disruption is based
upon the knowledge of the relationship between bridge damage level and associated loss of
functionality. Although recent studies have tried to provide general approaches for the determination
of such relationship (e.g., [17]), the decision to totally or partially close a bridge is usually made by
inspectors and officials who analyze case by case after a seismic event strikes. Therefore, the a
priori determination of bridge functionality, which is most likely to depend on human judgment, can
be carried out only probabilistically.
Empirical approaches, such as the one proposed in [18], can estimate the recovery of functionality
based on the importance of the route carried by the bridge under investigation. Analytical methods
have been proposed, such as the one developed in [19], in which the functionality is measured in
terms of the traffic load-carrying capacity. Additionally, expert opinions were collected by surveys
for the establishment of recovery curves for given specific damage states [8, 17].
3.1. Recovery pattern

The evolution of the functionality over time is at first affected by the post-event loss of functionality
(segment A–B of Figure 1), depending only on the assessment of the post-event damage. A fragility
analysis can be used for the evaluation of the probability of the bridge being in a certain damage
state (e.g., no damage, slight, moderate, extensive, and complete damage). The software HAZUS [20],
which provides loss estimations, is used in this paper to perform the fragility analysis. Upon the
selection of a seismic source, HAZUS considers both ground shaking and ground failure induced
damage, including attenuation functions accounting for the propagation of the seismic waves. In this
study, the residual functionality Qr (Figure 2(a)) is associated with the bridge damage level based on
the values provided in [8], adopted by HAZUS [21].
The recovery pattern mainly depends on the physical rehabilitation/reconstruction process involving
construction techniques and scheduling. As mentioned in Section 2, the first phase of the recovery
process (segment B–C of Figure 1) occurs during the idle time interval di (Figure 2(a)), that has a
twofold purpose. It can be used to plan the initiation of rehabilitation works, depending on the
availability of construction companies able to rehabilitate the bridges. In this situation, the inclusion
of an idle time interval is crucial, because it is improbable that, within a bridge network, all bridges
are simultaneously repaired [22], and that the repair process will begin immediately after the
occurrence of an extreme event. Actually, the proper idle time interval can be evaluated only if an
analysis at the network level is performed, and by investigating network performance indicators
(e.g., total travel time, total travel distance) the importance of the bridges is assessed. The idle
time can also be due to specific types of restoration activities that do not improve the bridge
functionality (e.g., the phase of demolition of a damaged bridge or the removal of the debris).
For the second phase (path C–D of Figure 1), as proposed in [9], the recovery model is a generalized
six-parameter sinusoidal-based continuous function, as shown in Figure 2(a). The choice of describing
DOI: 10.1002/eqe
(a)
PDF of the random variables
Qualitative recovery function
Qr,m
fQr(Qr)
f i( i)
1.0
Functionality Q(t)
Qr,min Qr,max i,min i,max
0.8 Residual Idle time interval
A functionality Qr i (months)
0.6
Qt
r,m Qt,mode
fQt(Qt)
f r( r)
0.4
0.2 r,min r,max Qt,min Qt,max

i r
Qr
Recovery duration Target

0.0 r (months) functionality Qt
0 2 4 6 8
Time from extreme event t (months)
(b)
Qualitative recovery functions

No damage Slight damage
Moderate damage
1.0
Functionality Q(t)
0.8
Complete damage
0.6 (with bypass)
0.4
Extensive damage
0.2 Complete damage

(without bypass)
0.0
0 2 4 6 8 10 12
Figure 2. Qualitative representation of (a) the adopted six-parameter bridge recovery function including the
distribution types of specific random variables, and (b) the recovery patterns associated with different types
of damage and recovery options.
the recovery pattern with this continuous function is justified by the fact that this general formulation is
able to capture shapes approaching stepwise, linear, and positive and negative exponential functions in
addition to sinusoidal curves. This flexibility makes the proposed function the good candidate to cover
a large group of recovery patterns and potentially fit the results obtained by other researchers [17, 23].
The shape and the position of this curve is based upon setting the residual functionality Qr, the idle
time interval di (months), the recovery duration dr (months), the target functionality Qt, and two more
parameters called s and A, which control the position of the inflection point and the amplitude of the
sinusoidal curve, respectively. In principle, all these parameters can be modeled as independent random
variables. However, assessing the probability distributions of parameters s and A is very difficult for
practical applications. Therefore, it is recommended to treat Qr, di, dr, and Qt as independent random
variables, and s and A as functions of the random initial damage. Figure 2(a) shows a qualitative
recovery curve and the types of distribution for the random variables of the probabilistic model.
As mentioned earlier, parameters s and A, governing the shape of the recovery pattern, depend on
the random post-event damage level. Figure 2(b) qualitatively shows the recovery patterns
associated with the four damage levels. For instance, for slight damage, the functionality recovers at
a faster pace at the beginning of the restoration works (negative exponential-type), whereas for
moderate damage, a stepwise-type curve is used because of partial reopening during the
rehabilitation period. For extensive damage and complete damage (without bypass), a positive
exponential-type pattern is adopted assuming that the bridge is fully operable only at the end of the
rehabilitation process. Moreover, for the case of complete damage, an additional option is available.
During the phase of reconstruction of a completely damaged bridge, decision makers can opt for
DOI: 10.1002/eqe
constructing a temporary bypass that carries a portion of the original traffic flow, leading to a stepwise-
type recovery curve (case with bypass). Although other parameters may be introduced, making the
recovery pattern even more flexible (e.g., accounting for the effects of aftershocks), a six-parameter
curve is retained capable of adapting to the most common recovery patterns. A detailed explanation
of the proposed recovery model can be found in [9].
A comprehensive calibration of recovery curves covering different bridge classes and different rural
and metropolitan areas is still missing in the existing literature. Only an extensive focus of future
research can fill this lack of information leading to more accurate evaluations of PSR that is also
affected by the decisions made in emergency situations involving several political and social factors,
among others. Although the proposed probabilistic approach is illustrated for seismic hazard
affecting bridges, this approach can be extended to different types of hazards and structures upon
the calibration of the relevant recovery parameters.
4. RESTORATION STRATEGY
Rather than focusing on the decision making process for the prioritization of the after-shock recovery
activities [22], this paper deals with the pre-event assessment of resilience, useful for assisting decision
makers for the purpose of planning retrofit, maintenance, and/or monitoring. As already mentioned, the
final PSR and economic losses will depend also on the recovery strategy adopted by decision makers
and the priority given to the restoration of the investigated bridge. These decisions are extremely
difficult to predict because they may be based on factors affected by high uncertainty. This paper
investigates potential resilience and functionality outcomes associated with a spectrum of feasible
options considering specific level of losses (risk) and the awareness and attitude of decision makers. In
particular, it is assumed that rehabilitation and construction activities can be performed with three
different velocities: fast, average, and slow pace (denoted F, A, and S, respectively). Because these
velocities affect duration and costs of the recovery phase, different velocity coefficients for duration Vt
and costs Vc are introduced, assuming that their values are positive and negative for fast and slow
recovery, respectively. These coefficients can be evaluated by analyzing historical records of bridge
rehabilitation and reconstruction regarding early completion bonuses or late completion penalties (e.g.,
[24] among others). From the review of this document and similar ones, it is found that Vt is usually
about 20%, whereas Vc is about 15%. Thus, the updated recovery duration dr0 = dr(1 Vt) depends
on the effort associated with the intervention.
Therefore, considering the previously mentioned velocity coefficients and including the option that
does not consider any intervention (called DN in Figure 3), the investigated individual recovery
patterns are represented by the 20 outcomes (Nout = 20), as reported in the decision tree of Figure 3.
Accordingly, the fragility analysis provides the probabilities associated with the first five branches
(chance node A), then depending on the damage level, different decisions regarding the recovery
velocity can be made (decision nodes B to G).
A wide spectrum of recovery strategies is investigated depending on the effort associated with the
recovery process (with velocities F, A, and S) and on the bridge damage level. Restoration strategies are
represented by combinations of the possible outcomes of Figure 3, considering the selection of a single
recovery pattern for each damage level. The outcomes included within a single considered strategy are
collectively exhaustive, and the sum of their associated probabilities leads to one. For instance, in the
case study, among all the possible mutually exclusive combinations, 24 restoration strategies (Nres = 24)
are investigated, as summarized in Table I. Strategy no. 1 indicates that all the recovery activities are
performed at the maximum velocity (i.e., selecting outcomes 2, 6, 10, and 14 of Figure 3 for slight,
moderate, extensive, and complete damage, respectively), thus disregarding potential budget constraints
and achieving full recovery in the shortest time after the event occurrence. On the other hand, decision
makers can decide to fit potential budget constraints by selecting strategy no. 18, which provides fast
recovery for slight and moderate damage (outcome 2 and 6 of Figure 3, respectively), whereas no
actions are taken for the cases of extensive and complete damage (outcome 13 and 20 of Figure 3,
respectively), in order to recover in the shortest time for small damages and do not consider recovery in
case of large damage (extensive and complete).
DOI: 10.1002/eqe
Post-event bridge
Recovery Options Outcome ID
damage level
Strategy 1
No damage
DN 1
F 2
Slight
B A 3
damage
S 4
DN 5
F 6
Moderate
damage C A 7
S 8
DN 9
A F 10
Extensive
D A 11
damage
S 12
DN 13
Rebuilding F 14
with bypass F
A 15
Complete
damage E S 16
Rebuilding
without F 17
bypass G
A 18
Chance node
S 19
Decision node
DN 20
Strategy 18
Figure 3. Decision tree indicating the 20 considered outcomes that include the velocity of the recovery actions
and the options without repair. Note: DN, do nothing; F, fast recovery; A, average recovery; S, slow recovery.
Table I. Definition of the restoration strategies indicating the recovery velocities for specific damage levels.
Recovery velocity for damage level Restoration Recovery velocity for damage level
Restoration strategy no.
strategy no. SD MD ED CD (cont.) SD MD ED CD
1 F F F F(wb) 13 DN DN S S(w/ob)
2 A A A A(wb) 14 S S F F(wb)
3 S S S S(wb) 15 S S A A(wb)
4 DN DN DN DN 16 S S F F(w/ob)
5 F F F F(w/ob) 17 S S A A(w/ob)
6 A A A A(w/ob) 18 F F DN DN
7 S S S S(w/ob) 19 A A DN DN
8 DN DN F F(wb) 20 S S DN DN
9 DN DN A A(wb) 21 F F S S(wb)
10 DN DN S S(wb) 22 A A S S(wb)
11 DN DN F F(w/ob) 23 F F S S(w/ob)
12 DN DN A A(w/ob) 24 A A S S(w/ob)
SD, slight damage; MD, moderate damage; ED, extensive damage; CD, complete damage; F, fast recovery;
A, average recovery; S, slow recovery; DN, do nothing; wb, with bypass; w/ob, without bypass.
For each considered strategy, PSR, rapidity, direct, and indirect costs are computed by performing a
Monte Carlo simulation based on Latin hypercube sampling [10]. This provides a better representation
of the sample space, compared with crude Monte Carlo simulation. Figure 4 shows the flowchart of the
proposed procedure.
5. COST ANALYSIS
The losses induced by an earthquake are usually quantified by their associated monetary values. This
paper accounts for direct and indirect costs. In this paper, the discount rate of money is neglected, given
that PSR is evaluated over a short period of time (i.e., usually less than two years after the occurrence
of an extreme event).
DOI: 10.1002/eqe
Generate the statistical samples by using

Assign statistical descriptors and
Latin hypercube
deterministic parameters to the recovery
functions
Simulation
Evaluate functionality pattern, resilience,
Assign inputs for cost analysis rapidity, and direct and indirect costs for
each considered strategy (see Table I)
Input
Define a spectrum of recovery strategies
considering rational combinations of the NO Is the total number of
outcomes of Figure 3 samples reached?
YES
Perform fragility analysis of the selected
Obtain probabilistic distributions and
Results
bridge given a seismic scenario and obtain
statistics of functionality, resilience, rapidity,
probabilities of the bridge being in specific
and direct and indirect costs for each
damage states
considered recovery strategy
Figure 4. Flowchart of the proposed procedure for the evaluation of probabilistic seismic resilience, rapidity,
and socio-economic impact.
5.1. Direct costs

Direct costs include those associated with the rehabilitation–reconstruction of the bridge, removal of debris,
and the construction of a temporary bypass. A simplified method for their estimation is adopted, as
customary. Rehabilitation–reconstruction costs are assumed to be proportional to the bridge replacement
values. The proportionality factor is called ‘damage ratio’, and it can be related to the bridge damage
level [7, 21]. Similarly, the cost of the construction of a temporary bypass is assumed to be the ratio of
the bridge replacement value and the costs of debris removal proportional to the bridge deck area. Direct
costs associated with the ith restoration strategy are evaluated as follows (based on [7, 18]):
8 2 39
>
>
rehabilitation and debris removal temporary bypass >
>
>
> 6 reconstruction costs 7>>
Nout > >
< 6︷ cost costs
X 6 ︷︷ 7 =
7
Cdir;i ¼ WL P 6 a c d 1 þ Vc;ij þ bj crem þ gj creb br 1 þ Vc;ij 7 (3)
j¼1 >
> ij 6 j reb r;j 7>
>
> 4 5>>
>
>
: >
;
where W and L are the bridge width (m) and length (m), respectively, index j runs over the outcomes,
creb is the rebuilding cost per square meter ($/m2), dr,j is the damage ratio immediately after the
occurrence of the earthquake (i.e., the proportionality factor between bridge repair costs and the
bridge replacement value), Vc,ij is the velocity coefficient for costs, crem is the debris removal cost
per square meter ($/m2), br is the bypass cost ratio (i.e., the proportionality factor between the cost
for the construction of a temporary bypass and the bridge replacement value), and aj, bj, and gj are
coefficients such that aj = 1 if outcome j includes rehabilitation/reconstruction (i.e., outcomes
included in set A = {2,3,4,6,7,8,10,11,12,14,15,16,17,18,19} according to Figure 3), bj = 1 if
outcome j includes debris removal (i.e., outcomes included in set B = {14,15,16,17,18,19} according to
Figure 3), gj = 1 if outcome j includes the construction of a temporary bypass (i.e., outcomes included in
set C = {14,15,16} according to Figure 3), and aj = 0, bj = 0 and gj = 0 otherwise. Each jth outcome is
associated with an initial damage state with probability PDS,j and its inclusion into the ith strategy is
determined by coefficient dij leading to Pij = dijPDS,j. If outcome j is included in strategy i, then
dij = 1, otherwise dij = 0. For instance, Figure 3 shows that for strategy 1 (i.e., i = 1) the coefficients
are d1 1 = d1 2 = d1 6 = d1 10 = d1 14 = 1 and d1 3 = d1 4 = d1 5 = d1 7 = d1 8 = d1 9 = d1 11 = d1 12 = d1
13 = d1 15 = d1 16 = d1 17 = d1 18 = d1 19 = d1 20 = 0. In this paper, according to Figure 3, the cases in
which no damage occurs or no actions are taken (i.e., outcomes 1, 5, 9, 13, and 20 of Figure 3) do
not generate any direct costs. Because uncertainty highly affects the estimation of costs, unitary
costs creb and crem, damage ratio dr,j, and bypass cost ratio br are treated as random variables.
DOI: 10.1002/eqe
5.2. Traffic flow analysis

The bridge functionality is directly related to the number of operable lanes within each traffic direction,
and according to the level of induced damage, the bridge can be fully serviceable (Q(t) = 100%),
partially serviceable (0% < Q(t) < 100%), or temporarily closed (Q(t) = 0%). The traffic flow will
accordingly be redistributed between the route segment containing the bridge, called ‘link’ in the
reminder and the detour.
The traffic flow carried by a route can be evaluated on the basis of the average daily traffic (ADT)
traveling on the link. In the USA, the ADT for every bridge is provided by the National Bridge
Inventory (NBI) database [25]. On the basis of [26] and accounting for peak and off-peak hours
[27], the total traffic flow Ft (vehicles/h/lane) on a given link direction is given by

K1 DAt peak hours
Ft ¼ (4)
K2 DAt off peak hours
where K1 and K2 are the percentages of traffic associated with the portion of ADT during the peak and
off-peaks hours, respectively, D is the directional factor, and At is the ADT.
Depending on the capacity and traffic, the traffic speed on the link and on the detour is provided by [28]
S0
St ¼
4 (5)
1 þ 0:15 FFC tnl
where S0 is the free-flow traffic speed (km/h), FC is the traffic flow capacity (vehicles/h), and nl is the
number of lanes.
Because a comprehensive traffic analysis is out of the scope of this paper, a simplified method for
the traffic redistribution between link and detour is adopted herein. The traffic flows carried by the
link and the detour are considered in equilibrium before the strike of an earthquake. It is assumed
that their ADTs depend only on the respective traffic flow capacities, called FC,l (vehicles/h/lane)
and FC,d (vehicles/h/lane) for the link and detour, respectively. Because the ADT of the link and the
capacities FC,l and FC,d are known, the total ADT associated with both link and detour can be
obtained as follows:

Al FC;l nl;l þ FC;d nl;d
Atot ¼ (6)
FC;l nl;l
where Al and Atot are the link and the total ADTs, respectively, and nl,l and nl,d are the number of lanes
of the link and detour, respectively. In a similar way, the ADT of the detour Ad can be obtained
Atot FC;d nl;d
Ad ¼ (7)
FC;l nl;l þ FC;d nl;d
When an earthquake strikes and the rehabilitation phase begins, the bridge functionality Qij(t),
associated with the ith strategy (Table I) and the jth outcome (Figure 3), varies over the restoration
time, affecting the traffic flow of the link and of the detour. Simplified expressions for the estimation
of their associated ADTs are as follows:
0 Atot FC;l nl;l Qij ðt Þ
A l;ij ðt Þ ¼ (8)
FC;l nl;l Qij ðt Þ þ FC;d nl;d
0 Atot FC;d nl;d

A d;ij ðt Þ ¼ (9)
FC;l nl;l Qij ðt Þ þ FC;d nl;d
Where A0 l,ij(t) and A0 d,ij(t) are the time-dependent post-event ADTs of link and detour, respectively.
Therefore, the time-dependent traffic speed accounting for peak and off-peaks hours (Equation (4))
can be assessed for both link and detour (Equation (5)).
DOI: 10.1002/eqe
5.3. Indirect costs

Indirect costs are mainly caused by traffic disruption. Marginal operational costs occur when
users travel through detours, which are usually longer than the original route. The proposed
general formulation accounts for cars and trucks, and for traffic at peak and off-peak hours
(based on [18]):
8 9
Nout <
X Zth h i =
0
Cop;i ¼ Pij Dl cop;car ð1 tÞ þ cop;truck t A d;ij ðt Þ Ad dt (10)
j¼1
: ;
t0
where cop,car and cop,truck are the operating costs of cars and trucks per kilometer ($/km),
respectively, t represents the average daily truck traffic (ADTT) as percentage of ADT, Dl is
the additional travel distance (km), A0 d,ij(t) is the time-dependent ADT of the detour after the
seismic event, Ad is the ADT of the detour before the seismic event, t0 is the time at which
the seismic event occurs, th is the investigated time horizon, t is time, and dt is the variable of
integration (time).
After the seismic event, if traffic disruption occurs, users are forced to travel through the link and
detour at a lower speed than usual, thus marginal travel time generates monetary losses. Time loss
costs associated with the ith strategy are (based on [18]) as follows:
Ctl;i ¼ ½cw Ocar ð1 tÞ þ ðctc Otruck þ cinv Þt

8 9
Nout <
X Zth =
0 Ll Ll 0 Ll þ Dl Ll þ Dl (11)
P A b;ij ðt Þ 0 þ A d;ij ðt Þ 0 dt
j¼1
: ij S b;ij ðt Þ Sb S d;ij ðt Þ Sd ;
t0
where cw is the wage per hour ($/h), ctc is the total compensation per hour ($/h), cinv is the inventory
cost ($/h), Ocar and Otruck are the vehicle occupancies for cars and trucks, respectively, Ll is the link
length (km), S0 l,ij(t) and S0 d,ij(t) are the time-dependent traffic traveling speeds (km/h) of link and
detour after the seismic event, respectively, and Sl and Sd are the traffic speeds (km/h) of link and
detour before the seismic event, respectively.
In most practical cases, the majority of the involved parameters, cop,car, cop,truck, cw, ctc, cinv, Sb, and
Sd are affected by uncertainties; therefore, the proposed technique treats them as random variables.
6. CASE STUDY
The proposed approach is exemplified assuming that a bridge owner has to plan retrofit, maintenance, or
monitoring activities on a bridge located in a seismic area, based on the assessment of the expected
resilience and seismic economic impact. For instance, an existing bridge (structure no. 560680R
according to [25]) carrying a three-lane segment of the northbound I-15 highway crossing the Temescal
Wash, located between the cities of Corona and Murrieta in California, is investigated. It is assumed
that PSR is evaluated for a representative earthquake with magnitude 8 (Richter scale), having the
historical epicenter of the 15 May 1910 earthquake, with latitude 33º420 , longitude 117º240 , and depth
10 km, that is situated at a distance of about 4 km from the bridge. This bridge was selected because it
carries an important route in an active seismic area and crosses a stream, which leads to the evaluation
of indirect consequences generated only by the carried route. Bridge vulnerability is obtained by means
of a fragility analysis performed using the software HAZUS [20].
Assuming that the potential restoration strategies reported in Table I are those considered for this
assessment, the associated mean profiles of the bridge functionality over time are obtained by
performing Monte Carlo simulation setting 1/10 of a month as computational time interval and
considering 14 months as time horizon th (with t0 = 0 being the moment when the event strikes). By
using Latin hypercube simulation, sets of 50 000 samples are generated for each outcome of Figure 3,
for a total of 100 000 samples. Then, the results of each set of samples (50 000) are weighted by
DOI: 10.1002/eqe
the probability of their associated jth outcomes Pij for each selected ith strategy. The software
MATLAB [29] is used to run the simulation based on a parallel processing code, which requires
about 10 min for 100 000 samples on a DELL Precision T7400 workstation with two quad-core Intel
Xeon processors with 8 GB of RAM.
Although the assessment of the parameters involved in the recovery process is not an easy task, in
this paper, the types of distribution and their associated statistical descriptors are proposed on the basis
of literature surveys and engineering judgment. Table II summarizes the values of the random variables
and parameters assumed for the adopted six-parameter recovery function, and the associated
references. The involved random variables are assumed to be statistically independent. For slight,
moderate, and extensive damage, different triangular distributions for the residual functionality Qr
are provided (Figure 2(a)), whereas for the case of complete damage, residual functionality is null.
The modes of the triangular distributions coincide with the values provided in [8, 21]. Because
conducting estimates on the bridge importance, which is related to an analysis at the network level,
is outside the scope of this paper, the values of the idle time interval di have been assumed
uniformly distributed (Figure 2(a)) between 1 and 2 months for all the recovery patterns of
Figure 2(b). The assumed values for the recovery duration dr follow a triangular distribution
(Figure 2(a)) and are estimated in accordance with those provided in [7] for each damage state.
Although the proposed model can treat the target functionality Qt as random variable, in this case
study, it has been assumed that the recovery patterns completely restore the bridge functionality
(100%). If complete restoration is not a feasible option, this assumption can be removed, and Qt can
be treated as a random variable with possible values lower than 1. Finally, parameters A and s are
functions of the random post-event damage, determined by engineering judgment in order to obtain
the recovery pattern of Figure 2(b).
On the basis of Table I, strategy 6 prescribes an average recovery velocity for any initial damage,
without the construction of the bypass, and is selected herein as example to illustrate the results in
terms of resilience and economic impact. Figure 5(a) shows five random sample recovery patterns
for each of the estimated random post-event damage levels for the selected strategy 6. As shown
in Figure 5(a), and in accordance with Figure 2(b), five groups of curves associated with the five-
bridge damage levels may be identified for strategy 6. The complete simulation of these recovery
patterns leads to the mean recovery pattern shown in Figure 5(b). Indication on the dispersion of
the functionality pattern is also shown in Figure 5(b), which provides the frequency of the
functionality when the earthquake strikes, and 4 and 8 months after, for strategy 6. Additionally,
the frequency of the full recovery time is shown (vertical bars). It can be noticed that, because of
the results of the fragility analysis, functionality is quite dispersed. This is because HAZUS
considers two distinct failure modes (ground shaking and ground failure) that provide dispersed
probabilities of the bridge being in specific damage states. Figure 5(c) shows the mean recovery
patterns of 10 representative strategies (1, 4–8, 12, 14, 19, and 23 of Table I). The fastest
expected full recovery is about 8 months (strategies 1 and 14), whereas the slowest expected full
recovery is achieved only about 13 months after the earthquake occurrence (strategies 7 and 23).
Moreover, during the investigated time frame, it can be noticed that strategy 19 only recovers
about 40% of the expected value of functionality, whereas according to its definition (Table I),
strategy 4 does not account for any rehabilitations; hence, the residual functionality does not
increase over time. Figure 5(c) shows also that, in this case, the expected residual functionality of
the bridge is about 30%.
Probabilistic seismic resilience and rapidity are then assessed on the basis of the simulated
functionality profiles by applying Equations (1) and (2), respectively. Figure 6 shows the frequency
histogram of resilience (Figure 6a) and rapidity (Figure 6b) for strategy 6. The two frequency
histograms are clearly subdivided into two parts. One part is represented by the bars located at
values 1 and 0 for resilience and rapidity, respectively, which represent no damage, whereas the
remaining bars are associated with the occurrence of any damage level. For the representative
strategy 6, the expected resilience and rapidity are mR = 0.687 and mr = 6.050 , respectively.
Mean values and standard deviations of resilience and rapidity are shown for all the 24 considered
strategies in Figure 7(a) and (b), respectively. Bars represent intervals equivalent to one standard
deviation. The impact of the earthquake in terms of expected direct costs and external consequences
DOI: 10.1002/eqe
1480
Table II. Values of the random variables and parameters associated with the assumed six-parameter recovery model that describes the evolution of functionality over time for
each recovery pattern associated with a specific bridge damage state (Figure 2) and the assumed damage ratio dr.
Type of recovery
Complete damage
Slight Moderate Extensive Complete damage Rebuilding without
Type of recovery No damage damage damage damage Rebuilding with bypass bypass
* Residual functionality Qr Qr,min 1 0.5 0 0 0 (Deterministic) 0 (Deterministic)
Copyright © 2013 John Wiley & Sons, Ltd.

Qr,mode (Deterministic) 0.75 0.25 0.1
Qr,max 1 0.5 0.2
Idle time interval di (months) di,min Does not apply 1 1 1 1 1
di,max 2 2 2 2 2
†
Recovery duration dr (months) dr,min Does not apply 0.333 0.667 2 2.5 2.5
dr,mode 2.667 3.667 5.167 6.25 6.25
dr,max 5 6.667 8.333 10 10
* Recovery duration dr (months) dr,min Does not apply 0.033 0.167 3 8 8
dr,mode 0.067 0.25 4 12 12
dr,max 0.1 0.333 5 14 14
Target functionality Qt Qt,min Does not apply 1 1 1 1 1
Qt,mode 1 1 1 1 1
Qt,max 1 1 1 1 1
{
Fixed parameters s Does not apply 0.1 0.5 0.9 0.5 0.9
A 0 0.1 0.1 0.1 0.1
A. DECÒ, P. BOCCHINI AND D. M. FRANGOPOL
* Damage ratio dr dr,min 0 0.01 0.02 0.1 0.3 0.3

dr,mode (Deterministic) 0.03 0.08 0.25 1 1
dr,max 0.03 0.15 0.4 1 1
*Data assessed based on the ATC-13 [8] and FEMA [21]
†
Data assessed based on Shinozuka et al. [7]
{
Based on engineering judgment
DOI: 10.1002/eqe
Earthquake Engng Struct. Dyn. 2013; 42:1469–1487
(a)
No damage Strategy 6
Slight damage
Moderate damage
Functionality Q(t)
1.0
0.8
Extensive damage
0.6
0.4 Complete damage

(without bypass)
0.2
0.0
0 2 4 6 8 10 12 14
(b)
Mean functionality
Functionality Q(t)
1.0
Full recovery
0.8
Initial
0.6 Mean functionality
0.4
4 months 8 months
0.2
Strategy 6
0.0
0 2 4 6 8 10 12 14
(c) Time from extreme event t (months)
Strategy 6 Mean functionality

Strategy 12
Strategy 1
Functionality Q(t)
1.0
Strategy 14
Strategy 23
0.8
Strategy 7
0.6 Strategy 8
Strategy 5 Strategy 19
Strategy 4
0.4
0.2
0.0
0 2 4 6 8 10 12 14
Figure 5. (a) Random sample recovery patterns for strategy 6. (b) Expected functionality of strategy 6 and
graphical representation of the frequency histogram of functionality when the earthquake strikes, 4 and
8 months after, and frequency histogram of the full recovery time. (c) Expected functionality profiles asso-
ciated with the representative strategies 1, 4–8, 12, 14, 19, and 23 of Table I.
(a) (b)
0.3 0.3
µR = 0.687 Strategy 6 Strategy 6 µr = 6.050º
R = 0.211 r = 3.770º
Mean rapidity µr
0.2 0.2
Frequency
Frequency
Mean resilience µR
0.1 0.1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 5 10 15 20
Resilience R Rapidity r (º)
Figure 6. Frequency histograms of the obtained (a) resilience and (b) rapidity for strategy 6. m and s are the
mean and standard deviation, respectively.
DOI: 10.1002/eqe
(a)
1.0
0.8
Resilience R
0.6
0.4
0.2
0.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
(b)
12
12
Rapidityr (º)
8
6
4
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
(c)
35
Costs (USD millions)
30 Mean direct costs

25 Mean time loss costs
20 Mean operating costs
15
10
5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Strategy
Figure 7. Bars reporting on the mean and standard deviation of (a) resilience and (b) rapidity for all the 24
considered strategies. The intervals refer to one standard deviation. (c) Histogram of the expected direct
and indirect costs for each strategy.
(time loss and additional operating costs) is shown in Figure 7(c). Costs are evaluated on the basis of
the deterministic and random variables summarized in Tables II and III. Figure 7(c) shows the
importance of large investments in trying to reduce indirect losses.
Indications regarding the available options for post-event response can be based on the results
reported in Figure 8, where the expected direct costs are shown along with expected resilience and
full recovery time for each strategy. Figure 8(a) shows decreasing expected resilience and associated
direct costs provided for the 24 strategies, whereas Figure 8(b) shows decreasing direct investments
against resilience. Moreover, as illustrated in Figure 8(c), the expected full recovery time is also
computed for those strategies that imply recovery from each damage state (i.e., strategies 4, 8–13,
and 18–20 are excluded because they include the outcomes ‘DN’ of Figure 3).
6.1. Discussion
Probabilistic seismic resilience can be included in a comprehensive structural management tool [34–36]
in order to develop optimal retrofit, maintenance, or monitoring activities. As indicated previously, the
aim of this paper is to evaluate resilience for assisting decision makers in planning pre-event activities;
therefore, a discussion regarding the results obtained in terms of expected resilience and economic
impact is presented. With the data obtained for the selected bridge, it can be noticed from Figure 7(a)
and (b) that strategy 4 (which according to Table I considers no interventions) provides the lowest level
of resilience and rapidity, and together with strategies 18–20, is affected by the largest standard
deviations depending exclusively on the dispersion obtained by the fragility analysis. Smaller standard
deviations are found when functionality is approaching full recovery (100%). Hence, strategies with
higher expected resilience and rapidity (such as strategies 1 and 14, which require greater social effort)
have relatively small standard deviations (the coefficient of variation is around 15%). According to
DOI: 10.1002/eqe
Table III. Statistical descriptors and deterministic parameters used for the cost analysis. Costs refer to their
values on year 2011.
Deterministic parameter Value Reference
ADT 39 500 FHWA [25]

ADTT/ADT ratio 13% FHWA [25]
Bridge length (m) 115.5 FHWA [25]
Bridge width (m) 21 FHWA [25]
Detour additional travel distance (km) 2 FHWA [25]
Detour number of lanes 1 Google Inc. [30]
Detour traffic flow capacity (vehicle/hour/lane) 1000 Assumed
Directional factor 0.5 Assumed
Length of the link (km) 5.95 Google Inc. [30]
Link traffic flow capacity (vehicle/hour/lane) 2000 Assumed
Link number of lanes 3 Google, Inc. [30]
Parameter K1 0.085 DOT-CA [27]
Parameter K2 0.030 DOT-CA [27]
Vehicle occupancies for cars 1.50 AASHTO [31]
Vehicle occupancies for trucks 1.05 AASHTO [31]
Random variable Mean COV Distribution type Reference
Compensation for truck drivers ($/h) 29.87 0.31 * LN AASHTO [31]
Detour speed (km/h) 50 0.20 * LN Assumed
Inventory costs ($/h) 3.81 0.20 * LN AASHTO [31]
Operating costs for cars ($/km) 0.40 0.19 * LN AASHTO [31]
Operating costs for trucks ($/km) 0.57 0.19 * LN AASHTO [31]
Wage for car drivers ($/h) 11.91 0.28 * LN AASHTO [31]
Random variable Min. Max. Distribution type Reference
Link speed (km/h) 90 120 * Uniform Assumed
Debris removal cost ($/m2) 224 560 Uniform DOT-FL [32]
Rebuilding costs ($/m2) 1318 3294 Uniform DOT-CA [33]
Random variable Min. Mode Max. Distribution type
Bypass cost ratio 0.3 0.5 0.7 * Triangular
*Assumed
LN, log-normal distribution; COV, coefficient of variation.
Figure 7(c), strategies 1 and 14 limit indirect costs, whereas strategies 4, 18–20 do the opposite. Minimum
expected indirect costs (about $7M) are obtained only if about $4M are invested (strategy 1). Finally, as
observed in Figure 7(c), the construction of a temporary bypass (e.g., strategies 1–3) heavily
decreases the impact on indirect costs against an investment that is almost twice with respect to the
same strategy without the bypass (e.g., strategies 5–7).
Expected resilience has been investigated for different cases, such as full and fast restoration
disregarding budget limitations versus available options subjected to budget constraints (Figure 8).
For instance, if a fast recovery that reduces external costs was the priority, strategy 1 would be
likely to be selected (Figures 7c and 8a). Among practitioners and experts, this strategy is believed
to be preferable even if costly. In fact, reducing the damage is the primary goal for bridges with
respect to functionality of the road network, and also in terms of social acceptance of structural/
infrastructural performance. However, if the available budget is limited, for instance to $2.5M
(Figure 8b,c), the available options would remain strategies 4–7, 11–13, 16–20, 23, and 24.
Therefore, strategy 5 being the one with largest mean resilience and shortest mean full recovery time
among the remaining available options, it is the most likely to be selected. It can be further noticed
that resilience does not increase considerably when a much greater investment is provided, whereas
the reduction of indirect costs is more significant. The construction of a temporary bypass
determines a large reduction of external costs; instead, only slight resilience is gained by adopting
these strategies, as shown in Figure 8. Sensitivity analyses have been performed with respect to
parameters such as idle time interval di, recovery duration dr, and earthquake magnitude. A variation
of the idle time interval di affects both resilience and indirect costs. Assuming that the mean idle
time interval varies between 0.5 and 5.5 months, and maintaining a uniformly distributed interval
between di,min and di,max of 1 month, resilience almost linearly decreases and indirect costs almost
DOI: 10.1002/eqe
(a)
Direct costs Cdir(USD millions) Direct costs Cdir(USD millions) Direct costs Cdir(USD millions)
1.0 4
Mean resilience
0.8
Mean direct costs 3
Resilience R 0.6
2
0.4
1
0.2
0.0 0
1 14 2 15 5 16 21 22 8 3 9 6 17 11 10 23 24 7 12 13 18 19 20 4
(b)
1.0 4
Unavailable options Available options
Mean resilience
0.8
Mean direct costs 3
Resilience R
0.6 Budget constraint

2
0.4
1
0.2
0.0 0
1 14 8 2 15 9 21 22 3 10 5 16 11 6 17 12 23 24 7 13 18 19 20 4
(c)
10 4
Full recovery time (months)
Unavailable options Available options Mean full recovery time

8
Mean direct costs 3
6 Budget constraint
2
4
1
2
0 0
1 14 8 2 15 9 21 22 3 10 5 16 11 6 17 12 23 24 7 13 18 19 20 4
Strategy
Figure 8. Histograms of expected resilience versus expected direct costs in the cases where strategies are
sorted by (a) decreasing resilience, (b) decreasing direct costs, and (c) expected full recovery time versus de-
creasing expected direct costs. Note: symbol is used for the strategies where the expected full recovery
time cannot be evaluated.
linearly increase when the mean idle time interval increases. However, it has been found that the
variation of indirect costs is higher than resilience, showing that a small loss of resilience is
associated with a significant increase of indirect costs.
As for the recovery duration dr, two different cases have been considered. In the first case, the
recovery duration was based on the values provided in [7] (previously analyzed), whereas for the
second case, the recovery duration dr was based on the values adopted by HAZUS [21]) as listed in
Table II. It is found that the mean functionality profiles significantly differ from the previous ones
(first case). This demonstrates that the choice of appropriate values for the recovery duration dr is of
critical importance. Moreover, the second case provides lower resilience and overall higher standard
deviation than the first case. Expected functionality, resilience, and economic impact are affected by
earthquake magnitude. A sensitivity analysis has been conducted considering magnitudes in the
range from 6.0 to 8.4 (with steps of 0.2) and maintaining the same epicenter. When the magnitude
increases, the decrease of resilience is relatively low, whereas the needed investment exponentially
rises, approaching a plateau of about $2.2M for earthquakes of magnitude greater than 8.
In summary, a large spectrum of scenarios determined by considering different strategies, idle time
interval, recovery duration, and earthquakes magnitudes, has been analyzed. Depending on the level of
regional seismic hazard, the level of performance that decision makers plan to achieve, allowable
economic impact, and available budget for post-event rehabilitation activities, proper scenarios can
be selected. Therefore, similar analyses can be repeated changing the bridge characteristics in the
fragility analysis and accounting for a potential retrofit. Comparing in terms of expected resilience,
DOI: 10.1002/eqe
predicted restoration time, economic impact, and estimated costs for the cases with and without the
retrofit, decision makers can make a more informed decision on whether or not to apply such retrofit
on the bridge. Similarly, preventive maintenance and monitoring activities can also be prioritized by
performing this type of analysis. Moreover, these pre-event management activities can be
concentrated on those bridges of a stock that have the most critical recovery profile. The proposed
approach can be further developed to accommodate multiple hazards [35, 37, 38] and be included
into a comprehensive life-cycle management framework [39, 40].
7. CONCLUSIONS
An approach for the pre-event assessment of PSR and social impact in terms of direct and indirect costs
has been presented. A large spectrum of scenarios including different restoration strategies have been
investigated on the basis of a six-parameter model that describes the variation of functionality over
time. Although the proposed framework can be generally applied for the pre-event management of
different types of structures subjected to multiple hazards, this paper places emphasis on bridges
affected by earthquakes.
The following conclusions are obtained:
1. The assessment of PSR can be conducted in a probabilistic way by including uncertainty affect-
ing both bridge vulnerability and recovery phase. The versatility of a six-parameter model for
assessing the evolution of functionality after the occurrence of a seismic event is exploited.
The proper use of the parameters allows investigating several recovery processes associated with
different levels of damage and the required rehabilitation operations.
2. The fragility analysis provides very disperse results for the post-event damage state. Therefore,
resilience and rapidity are affected by large uncertainty. It has been found that their dispersion
decreases when bridge functionality approaches full recovery for those strategies involving
large investments.
3. The combined information given by resilience and direct and indirect costs allows investigating
different restoration strategies. It has been found that the average functionality recovery heavily
depends on the rehabilitations associated with extensive and complete damages. Strategy ranking
has been provided with respect to expected resilience and direct investments. The inclusion of
financial constraints could preclude the selection of the most expensive options and could show
which strategies would remain available for the evaluation of resilience. Contrarily, it has been
shown that the most expensive strategy greatly limits the indirect consequences and provides the
highest expected resilience and shortest expected full recovery time, resulting as the most conve-
nient for limiting the overall loss. Some communities can retain this strategy as the only acceptable
response disregarding budget considerations. Moreover, in the case of complete damage, the
construction of a temporary bypass helps in mitigating the impact of indirect costs. However, the
gain of resilience is not as high as expected. Accordingly, pre-event decisions can be made on the basis
of the proper selection of the most representative strategies depending on the level of performance
that decision makers plan to achieve, allowable economic impact, and available budget for post-event
rehabilitation activities.
4. The results provided by the sensitivity analysis show that the appropriate selection of the
parameter distributions is critically important for conducting a reliable assessment. The results
obtained from data collected from various studies have been compared and discussed showing
important differences.
The assessment of PSR and economic impact may be affected by the unavailability of appropriate
data regarding the actual recovery curves for different classes of bridges. The proposed approach
based on the six-parameter curves can simulate a variety of recovery patterns by setting appropriate
parameters and integrate additional data when available. Groups of bridges and networks of
infrastructure systems can potentially be investigated by enhancing the approach proposed herein for
the case of a single bridge.
DOI: 10.1002/eqe
REFERENCES
1. Bruneau M, Chang SE, Eguchi RT, Lee GC, O’Rourke TD, Reinhorn AM, Shinozuka M, Tierney K, Wallace AW,
von Winterfeldt D. A framework to quantitatively assess and enhance the seismic resilience of communities.
Earthquake Spectra 2003; 19(4):733–752.
2. Frangopol DM, Bocchini P. Resilience as optimization criterion for the rehabilitation of bridges belonging to a
transportation network subject to earthquake. Proceedings of the ASCE 2011 Structures Congress SEI 2011,
Ames D, Droessler TL, Hoit M (eds). Las Vegas, NV, 2011; 2044–2055.
3. Bocchini P, Frangopol DM. Optimal resilience- and cost-based post-disaster intervention prioritization for bridges
along a highway segment. Journal of Bridge Engineering 2012; 17(1):117–129.
4. Bruneau M, Reinhorn A. Exploring the concept of seismic resilience for acute care facilities. Earthquake Spectra
2007; 23(1):41–62.
5. Xu N, Guikema SD, Davidson RA, Nozick LK, Çağnan Z, Vaziri K. Optimizing scheduling of post-earthquake
electric power restoration tasks. Earthquake Engineering and Structural Dynamics 2007; 36(2):265–284.
6. Cimellaro GP, Reinhorn AM, Bruneau M. Seismic resilience of a hospital system. Structure and Infrastructure
Engineering 2010; 6(1–2):127–144.
7. Shinozuka M, Zhou Y, Kim S-H, Murachi Y, Banerjee S, Cho S, Chung H. Socio-economic effect of seismic retrofit
implemented on bridges in Los Angeles highway network. Final report to the California Department of Transportation,
University of California, Irvine, CA, 2005.
8. ATC. Earthquake damage evaluation data for California. Technical Report ATC-13, Applied Technology Council
(ATC), Redwood City, CA, 1985.
9. Bocchini P, Decò A, Frangopol DM. Probabilistic functionality recovery model for resilience analysis. In Bridge
Maintenance, Safety, Management, Resilience and Sustainability, Biondini F, Frangopol DM (eds). CRC Press,
Taylor and Francis: UK, 2012; 1920–1927.
10. McKay MD, Conover WJ, Beckman RJ. A comparison of three methods for selecting values of output variables in
the analysis of output from a computer code. Technometrics 1979; 21(2):239–245.
11. Chang SE, Shinozuka M. Measuring improvements in the disaster resilience of communities. Earthquake Spectra
2004; 20(3):739–755.
12. Rose A. Defining and measuring economic resilience to disasters. Disaster Prevention and Management 2004;
13(4):307–314.
13. Park J, Nojima N, Reed D. Nisqually earthquake electric utility analysis. Earthquake Spectra 2006; 22(2):491–509.
14. Bocchini P, Frangopol DM. Resilience-driven disaster management of civil infrastructure. In Computational
Methods in Structural Dynamics and Earthquake Engineering, Papadrakakis M, Fragiadakis M, Plevris V (eds).
COMPDYN 2011: Corfu, Greece, 2011.
15. Franchin P, Pinto PE. Allowing traffic over mainshock-damaged bridges. Journal of Earthquake Engineering 2009;
12(5):585–599.
16. Ghanem G, Spanos PD. Stochastic Finite Elements: A Spectral Approach. Springer-Verlag: New York, NY, 1991;
214.
17. Padgett JE, DesRoches R. Bridge functionality relationships for improved seismic risk assessment of transportation
networks. Earthquake Spectra 2007; 23(1):115–130.
18. Stein SM, Young GK, Trent RE, Pearson DR. Prioritizing scour vulnerable bridges using risk. Journal of Infrastructure
Systems 1999; 5(3):95–101.
19. Mackie KR, Stojadinović B. Post-earthquake functionality of highway overpass bridges. Earthquake Engineering
and Structural Dynamics 2006; 35(1):77–93.
20. FEMA. HAZUS-MH MR4 – Earthquake Model User Manual. Department of Homeland Security, Federal
Emergency Management Agency (FEMA): Washington, DC, 2009.
21. FEMA. HAZUS-MH MR4 – Earthquake Model Technical Manual. Department of Homeland Security, Federal
Emergency Management Agency (FEMA): Washington, DC, 2009.
22. Bocchini P, Frangopol DM. Restoration of bridge networks after an earthquake: multi-criteria intervention optimization.
Earthquake Spectra 2012; 28(2):426–455.
23. Miles SB, Chang SE. Modeling community recovery from earthquakes. Earthquake Spectra 2006; 22(2):439–458.
24. Engius. Case study: interstate 40 bridge reconstruction; Webbers Falls, Oklahoma. Engius, Stillwater, OK, 2002.
(Available from: http://www.engius.com.)
25. FHWA. National Bridge Inventory (NBI) database. U.S. Department of Transportation, Federal Highway
Administration (FHWA), Washington, DC, 2010. (Available from: www.fhwa.dot.gov/bridge/nbi.htm.)
26. TRB. Highway Capacity Manual, Special Report 209. Transportation Research Board (TRB), National Research
Council: Washington, DC, 2000.
27. Caltrans. California Life-Cycle Benefit/Cost Analysis Model (Cal-B/C) – Volume 3. California Department of
Transportation: Sacramento, CA, 2009.
28. Bureau of Public Roads. Traffic Assignment Manual. U.S. Department of Commerce, Urban Planning Division:
Washington DC, 1964.
29. The Mathworks Inc. Matlab Version 7.13-R2011b. Natick, MA, 2011.
30. Google Inc. Google Earth, Release 6.1.0.5001. Keyhole Inc. & Google Inc.: Menlo Park, CA, 2011.
31. AASHTO. A Manual of User Benefit Analysis for Highways (2nd edn). American Association of State Highway and
Transportation Officials (AASHTO): Washington, DC, 2003; 288.
DOI: 10.1002/eqe
32. DOT-FL. Transportation Costs Report. State of Florida - Department of Transportation: Tallahassee, FL, 2009.
33. Caltrans. Comparative Bridge Costs. California Department of Transportation: Sacramento, CA, 2010.
34. van Noortwijk JM, Frangopol DM. Two probabilistic life-cycle maintenance models for deteriorating civil
infrastructures. Probabilistic Engineering Mechanics 2004; 19(4):345–359.
35. Decò A, Frangopol DM. Risk assessment of highway bridges under multiple hazards. Journal of Risk Research 2011;
14(9):1057–1089.
36. Frangopol DM. Life-cycle performance, management, and optimization of structural systems under uncertainty:
accomplishments and challenges. Structure and Infrastructure Engineering 2011; 7(6):389–413.
37. Akiyama M, Frangopol DM, Matsuzaki H. Life-cycle reliability of RC bridge piers under seismic and airborne
chloride hazards. Earthquake Engineering and Structural Dynamics 2011; 40(15):1671–1687.
38. Decò A, Frangopol DM. Life-cycle risk assessment of spatially distributed aging bridges under seismic and traffic
hazards. Earthquake Spectra 2013; 29(1):1–27.
39. Frangopol DM, Kong JS, Gharaibeh ES. Reliability-based life-cycle management of highway bridges. Journal of
Computing in Civil Engineering 2001; 15(1):27–34.
40. Frangopol DM, Liu M. Maintenance and management of civil infrastructure based on condition, safety, optimization,
and life-cycle cost. Structure and Infrastructure Engineering 2007; 3(1):29–41.
DOI: 10.1002/eqe

A Probabilistic Approach For The Prediction of Seismic Resilience of Bridges

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Probabilistic Approach For The Prediction of Seismic Resilience of Bridges

Uploaded by

Copyright:

Available Formats

EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2013; 42:1469–1487

A probabilistic approach for the prediction of seismic resilience of

Alberto Decò, Paolo Bocchini and Dan M. Frangopol*,†

Received 22 March 2012; Revised 18 December 2012; Accepted 19 December 2012

Copyright © 2013 John Wiley & Sons, Ltd.

where tf is the time at which the recovery process ends.

3. PROBABILISTIC RECOVERY MODEL

3.1. Recovery pattern

0.2 r,min r,max Qt,min Qt,max

Recovery duration Target

Qualitative recovery functions

0.2 Complete damage

Generate the statistical samples by using

5.1. Direct costs

5.2. Trafﬁc ﬂow analysis

0 Atot FC;d nl;d

5.3. Indirect costs

Ctl;i ¼ ½cw Ocar ð1 tÞ þ ðctc Otruck þ cinv Þt

Copyright © 2013 John Wiley & Sons, Ltd.

* Damage ratio dr dr,min 0 0.01 0.02 0.1 0.3 0.3

0.4 Complete damage

Strategy 6 Mean functionality

30 Mean direct costs

ADT 39 500 FHWA [25]

0.6 Budget constraint

Unavailable options Available options Mean full recovery time

You might also like

Ctl;i ¼ ½cw Ocar ð1 tÞ þ ðctc Otruck þ cinv Þt