Beruflich Dokumente
Kultur Dokumente
To cite this article: Dan M. Frangopol & Paolo Bocchini (2012) Bridge network performance,
maintenance and optimisation under uncertainty: accomplishments and challenges, Structure and
Infrastructure Engineering, 8:4, 341-356, DOI: 10.1080/15732479.2011.563089
Download by: [IIT Indian Institute of Technology - Mumbai] Date: 23 October 2017, At: 07:33
Structure and Infrastructure Engineering
Vol. 8, No. 4, April 2012, 341–356
This article presents a critical review of the state-of-the-art in the field of bridge network performance analysis,
reliability assessment, maintenance management and optimisation. Previous accomplishments and results are
summarised while stressing the aspects of the analysis of a transportation network, which are more challenging for a
structural engineer. For instance, the bridge network is described as a spatially distributed system at a scale that is
much larger than any individual structure. This requires a different perspective in the modelling of natural extreme
events and involves models for the interaction between the individual components (i.e. bridges) and the overall
network. Then, the time domain is investigated, and the problem of structural deterioration and its effects on the
network performance are addressed. The most important differences between a transportation network and other
lifelines are considered next, and techniques for the transportation network analysis and performance assessment
are summarised. Finally, a set of selected problems and applications is considered and additional challenges are
identified and suggested for future developments.
Keywords: bridge network; transportation network; structural deterioration; bridge maintenance; maintenance
optimisation; uncertainty; life-cycle performance.
characteristics of the infrastructure; (c) the multi- that follow hurricanes. Therefore, it is necessary to
hazard nature of the analysis; (d) the network effects develop a general approach that accounts for all of
and (e) the economic constraints all along the life-cycle them.
of the infrastructure. As mentioned previously, there are critical compo-
The essence of the network (as opposed to nents in the networks, but a study that aims at
individual components or groups of independent providing a useful decision tool for structural de-
components) consists in the correlations, interactions signers, urban planners and risk managers should not
and mutual dependencies of its components. When disregard the network system and the network effects.
dealing with a transportation network, the various In fact, the condition of every component influences
components are spread over very large areas. There- the flows on the entire network and, in turn, modifies
fore, the infrastructure manager is required to model the demands applied to the other components. For
the relationship among the various components at a instance, for bridges that have already suffered an
scale (the spatially distributed system) that usually extreme event, this can lead to congestion-based
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
overwhelms the dimensions of a single structure. disruptions. A recent study (Hernández-Fajardo and
The variability in the time domain must also be Dueñas-Osorio 2009) estimated that the fragility of
considered. Although the effects of long-term structur- components of coupled lifeline systems can be almost
al deterioration on the performance of structures and doubled if the network dependencies are considered in
infrastructures have already been recognised (Chang the computation. Moreover, this is another possible
and Shinozuka 1996, PONTIS 2005, Pandey et al. source of delays in the post-event interventions.
2009), the implementation of methodologies taking Finally, every decision has to account for a limited
into account time-dependent effects is less common. In budget. For instance, according to ASCE (2009), in the
most existing procedures and guidelines for design and next 5 years, the investment shortfall for civil infra-
performance analysis (e.g. seismic) of civil structures, structure will be $1.176 trillion. In order to improve the
the structural capacities are assumed to be time performance of the network, the decision makers can
invariant when a post-event reactive retrofit is carried act in different ways. For instance, they can improve
out to partially or fully repair damaged structures. the reliability and the robustness of critical compo-
Obviously, this assumption does not reflect the realistic nents (Ghosn et al. 2010); they can change the
time-dependent structural capacity evolution due to topology of the network, making it more redundant
progressive deterioration. Therefore, these procedures and they can monitor more efficiently the structural
predict incorrect capacities and damage levels that are health of some components to be able to better plan
likely to be on the unconservative side. Safety and the proactive maintenance. All these tasks should be
economic consequences due to these failures, especially performed, but when the budget is limited, it is
when combined at the network level, are in general necessary to choose only the more effective ones and
much more costly than the correction of structural prioritise the interventions (Frangopol and Bocchini
deficiency due to deterioration. Therefore, it is 2011a). These decisions cannot be taken arbitrarily;
advantageous and economical to proactively apply they have to be supported by a numerical tool that
maintenance interventions for enhancing the structural solves this multi-criteria optimisation problem under
performance. In order to efficiently plan the activities, uncertainty with limited financial resources. In most
an accurate predictive tool that also considers the cases, such a tool provides a series of Pareto optimal
network performance and the relative importance of solutions and the final decision is taken considering
the individual components should be adopted. also socio-economic aspects and previous experience.
When dealing with low-probability high-conse- In short, scientific research has provided significant
quence events, it is always desirable to choose a advances in many areas associated with infrastructural
multi-hazard approach. In fact, the same structures networks and their constitutive elements, including
and infrastructure systems can be subject to several structural health monitoring (Kim and Frangopol
types of extreme events. Even if these events can induce 2010), life-cycle analysis (Frangopol 2011), mainte-
different mechanical stresses, their effects from many nance (Neves et al. 2006a,b), reliability analysis
points of view are very similar. For instance, they can (Liu and Frangopol 2005c), vulnerability assessment
lead to the unavailability of the infrastructure and (Murray et al. 2008, Matisziw and Murray 2009) and
determine the same socio-economic impact, they can optimisation (Bocchini and Frangopol 2011b). In the
be included in the same probabilistic analysis frame- following sections, some of these contributions are
work, and their effects can be mitigated with similar summarised and compared, emphasising accomplish-
proactive interventions. Moreover, they are often ments and challenges. Moreover, to the authors’
interdependent, for example fires that follow earth- knowledge, there are no investigations on the possible
quakes, tsunamis that follow earthquakes and floods interactions of the results that all the above-mentioned
Structure and Infrastructure Engineering 343
analyses provide. Therefore, this article is also an At this large scale, the interactions between the
occasion to promote an integrated framework that individual components and the correlation of their
considers all these aspects in a fully probabilistic in/out of service states should also be considered. Most
perspective. This would be a tool that can dramatically of the probabilistic network analysis techniques
improve the allocation of the available resources and it proposed in the literature assume total independence
will result in increased serviceability, safety and between the serviceability of the various bridges
security of infrastructure networks, better response (Augusti et al. 1998, Akgül and Frangopol 2004c,
and lower economic impact for every possible hazard. Liu and Frangopol 2005c, 2006a).This assumption is
certainly not realistic, since the bridges of a network
are very likely to have similar traffic loads, to be
2. Spatially distributed systems designed with the same codes, to have similar
The analysis of spatially distributed systems brings the environmental conditions and to have experienced
focus of the structural engineer to a new scale. In fact, the same extreme events. Moreover, Bocchini and
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
‘large-scale buildings’, such as skyscrapers and sport Frangopol (2010, 2011a) have shown that the degree
arenas, have dimensions in the order of hundreds of of correlation among the bridge serviceability states,
metres (Fairweather et al. 2004); similarly, ‘long-span all across the network, has a strong impact on the
bridges’ are up to a few kilometres long structures, network performance. Therefore, every analysis at
even including the abutments (Chen and Duan 2000). the network scale should account not only for the
On the contrary, when dealing with an infrastructure interactions between the various components but also
network or a lifeline, the engineer is handling an entity for the correlation of the structural conditions.
spread over several kilometres, sometimes even hun- When structural engineers start to work at this new
dreds of kilometres. Some of the analyses on the scale, one of the first problems they have to face is the
individual network components (e.g. individual need of a strong loss of details in the model. In fact,
bridges) are not affected by this change in scale, but computational efficiency as well as epistemic uncer-
they only require the repeated use of traditional tainties and insufficient data often require a drastic
methodologies. However, many other analyses have reduction in the level of detail of the models. This can
to be performed with a completely different perspective. be felt as an unacceptable loss of accuracy in the
A first example of this issue can be the effect of results, but this perception is usually wrong. In fact, a
extreme events, such as earthquakes. Several studies similar process of synthesis and abstraction is followed
have focused on the spatial variation of seismic ground very often in the everyday practice. When an engineer
motion and its effects on the demand for large analyses a steel connection with a finite element model,
structures (Zhang and Deodatis 1996, Shinozuka very small elements are employed, the geometry of the
et al. 1999, Jankowski and Wilde 2000, Zerva and plates is described with utmost precision and several
Zervas 2002, Kim and Feng 2003, Zerva and Beck layers of elements are connected to discretise the
2003, Jankowski 2006, Zerva 2009), while others have thickness of each plate. However, when the same
studied the spatial variation of different external loads engineer models the structure to which the connection
(e.g. wind velocity: Davenport 1968, Simiu and belongs, much larger elements are preferable and the
Scanlan 1996, Gioffré et al. 2000; rainfall: Korving joint itself might even disappear. This is universally
et al. 2009). However, when dealing with a whole accepted as a part of the art of approximation: deep
network, even these approaches are not sufficient. attention should be paid to the critical components,
When an earthquake occurs, the bridges of a network but to capture the overall behaviour of the system, a
experience different ground motions (Chang et al. coarser description of its functional parts is required.
2000). Their structural responses are different because Similarly, when dealing with a network, the individual
they have different structural characteristics and they structures, such as entire bridges, become just compo-
are subject to different external solicitations. However, nents of a more complex system. Therefore, only a
these bridges cannot be analysed only individually, synthetic description of their characteristics and of
because the traffic disruption caused by simultaneous their structural response should be considered.
damage on several bridges is not just the superposition A practical application of this process is the use of
of the detours and delays caused by the unservice- analytical models for the time-dependent reliability of
ability of each bridge. Therefore, problems such as the the individual bridges. It is well known that the
risk assessment, the social cost caused by traffic structural characteristics deteriorate over time and,
disruption, the optimal economic resources allocation, therefore, the structural reliability tends to decrease
both for preventive (i.e. pre-event) and post-event along the life-cycle. Specific analyses can be performed
intervention require to model the structural demand to assess this decay in the reliability performance
and the structural response at the network scale. indicators for different types of bridges. Akgül and
344 D.M. Frangopol and P. Bocchini
Frangopol (2004a,b, 2005a,b) have proposed such where F71 is the inverse standard Gaussian cumula-
analyses for steel girder bridges, prestressed concrete tive distribution function. A lower threshold b for the
bridges, and reinforced concrete bridges. However, reliability index allows the definition of a second event
only the most important bridges of a network are E2 as bE1 ðtÞ b. EM, by definition, is applied
usually thoroughly modelled and sometime monitored, whenever event E2 occurs. However, when an entire
so that their predicted reliability profiles can be transportation network is considered, the previously
considered realistic. For all the other bridges, this mentioned issue of incomplete information arises, and,
information is, in general, unavailable and a less- for most of the bridges, thorough studies on the time-
detailed description is required. In these cases, the dependent reliability profile bE1 could be unavailable.
above-mentioned life-cycle reliability models Therefore, if the moment in which bE1 down crosses
(Frangopol et al. 2001a,b, Frangopol 2003, van the threshold b is unknown there will certainly be no
Noortwijk and Frangopol 2004, Kong and Frangopol EM interventions applied at that unknown instant.
2004b) can be used. This kind of model, that includes Therefore, in these cases, it appears much more
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
uncertainties, can be assessed knowing some basic realistic to consider a third type of maintenance, called
characteristics of the individual bridge. Therefore, the Required Maintenance (RM). This kind of interven-
reliability profile can be assessed without the need (and tion is applied only when the structure manifests an
the cost) of thorough studies on every bridge. imminent state of distress or when the distress has
Another issue related to the analysis under poor occurred (Bocchini and Frangopol 2011b). Using the
information and loss of details in the network model classical distinction between ‘preventive’ and ‘correc-
arises when dealing with maintenance management. tive’ interventions (Kallen and van Noortwijk 2006),
In fact, most of the studies on bridge maintenance tend both PM and EM fall in the first category, while RM
to classify the interventions in Preventive Maintenance represents corrective maintenance, also including the
(PM), also called ‘time-based’ or ‘proactive’, and limit case in which the failure has not occurred yet, but
Essential Maintenance (EM), also called ‘perfor- it is so imminent that the restoration is exactly as if the
mance-based’ or ‘threshold-based’. PM consists of failure had occurred. Therefore, using again the
those interventions that are scheduled at predefined previously defined events, for a specific limit state,
time intervals in order to always keep the bridge in a RM is applied when event E1 is actually imminent, or
good service level. This kind of intervention is also when it has just occurred. With this definition,
characterised by the lowest impact on the bridge safety restoration can be seen as a special case of RM, where
and the lowest cost. EM, instead, is applied when an event E1 is assumed to be the bridge collapse.
indicator of the bridge performance crosses a pre- However, the definition of RM is more general than
defined (acceptable) threshold. The most used indica- restoration, since it includes minor interventions if E1
tor is certainly the bridge reliability index, even if is a different limit state (e.g. excessive corrosion,
several other indicators can be considered, such as excessive deformation and serviceability limit). Since
redundancy and availability (Okasha and Frangopol the occurrence of event E2 (that triggers EM) is based
2010). In order to analytically define the EM applica- on the definition of event E1 (that triggers RM), it is
tion, a bridge limit state has to be chosen, for instance evident that these two types of maintenance are
it can be the excessive deformation of the main girders strongly interconnected. However, they are not the
or the collapse. Event E1 is defined as the bridge that same. Actually, on one hand, E2 can occur even if E1
reaches the investigated limit state. Given a specific never happened. In fact, the threshold b is usually
event E1, it is possible to define the time to failure assumed high, therefore when bE1 ðtÞ down crosses it
associated with that limit state TFE1 as the time (event E2 occurs), the reliability RELE1 ðtÞ ¼ F½bE1 ðtÞ
between a reference instant t ¼ 0 and the moment at is still very close to one (i.e. event E1 is still very
which E1 occurs (i.e. the limit state is reached). Then, unlikely to occur). On the other hand, E1 can happen
by definition, the reliability at time t is the probability at any time, because the bridge reliability is never equal
that E1 does not occur in the interval [0, t]: to unity, which means that there is always a chance of
failure. When the focus of a study is on an individual
RELE1 ðtÞ ¼ Pðt < TFE1 Þ ð1Þ bridge, it makes perfect sense to consider EM. In fact,
if a single bridge is studied, bE1 ðtÞ profiles will likely be
where P() denotes the probability of occurrence of the available, and not only the distress caused by the
event in brackets. Under the usual Gaussian assump- occurrence of event E1 should be avoided but even the
tion, the reliability index bE1 is computed as: probability of having a low reliability can be avoided.
However, in the case of a bridge network, when the
detailed information on many individual bridge
bE1 ðtÞ ¼ F1 ½RELE1 ðtÞ ð2Þ
performance indexes is unavailable, it is not possible
Structure and Infrastructure Engineering 345
to know when the bE1 ðtÞ of individual bridges cross an sets of residual capacity are presented to perform a
acceptable threshold. Therefore, it is not realistic to sensitivity analysis. For instance, the median set
apply EM at that instant. If EM is included in a model assumes that the residual flow capacity is 100% in
and then actually not applied, the final result would be the cases of no damage or minor damage; 50% in the
an unconservative maintenance schedule. case of moderate damage; and 25% in the case of
While the large scale of bridge networks yields the major damage or collapse (considering local detour
above-mentioned issues, it can also be seen as an routes). As expected, the results of the study (that
opportunity to develop very attractive unified models. consists in an exhaustive benefit/cost analysis of
For instance, Figure 1 shows a schematic representa- seismic retrofit) show a strong dependence on the
tion of a possible approach in which the hazard chosen set of residual capacity values. This further
induced by several natural extreme events is modelled enlightens the importance of an accurate and a well-
as a random field and the time-space superposition is calibrated model for the interaction between the bridge
used as basis for a multi-hazard approach. scale and the network scale.
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
In conclusion, the network analyses require a solid Song and his co-workers have developed the
background in the most traditional techniques for ‘matrix-based system reliability method’ and have
solution of structural problems, but also a novel set of proficiently applied it to the solution of several
tools and expertise that provide a better overall problems involving different networks and lifelines
description of the system and a deeper insight on the (Kang et al. 2008, Kim et al. 2009b, Lee et al. 2009,
various interactions and dynamics that happen at the Song and Kang 2009). These authors have also used
network scale. the bottleneck assumption. In the studies in which only
two bridge states are considered (failure/non-failure),
the collapse of a bridge implies the unserviceability
3. From individual structures to the network of the entire highway segment to which it belongs. In
Every time an analysis involves different scales, it is the studies in which multiple damage states are
necessary to have models for the interactions between considered, arbitrary percentage of the residual flow
the various scales. In the case of bridge networks, the capacity are assumed for every damage level (i.e. 100%
most important of these models is the one that links for no damage, 70% for moderate damage, 30% for
the structural condition and the damage level of an heavy damage and 0% for collapse).
individual bridge (structure scale) to its traffic flow A similar approach, based again on the bottleneck
capacity (relevant at the network scale). assumption has been presented by Bocchini and
Very different models have been introduced in the Frangopol (2011a). In this case, the individual bridge
literature. For instance, Shinozuka et al. (2006) and damage state was described with a real-valued variable
Zhou et al. (2010) have proposed the use of a in the interval [0, 4], where 0 and 4 represent the ‘no
‘bottleneck assumption’, according to which the damage’ and ‘collapse’ states, respectively. Therefore,
practical flow capacity of a highway segment depends the model for the interaction between the two scales
only on the residual practical flow capacity of its most had to be described by a continuous function:
damaged bridge. Given the damage state of the bridge
RC
in the worse condition, they provide the percentage fcij ¼ f c0 RC
ðmax lÞb
ij exp½a ð3Þ
flow capacity of the highway segment. Three possible
Figure 1. Space and time interconnection among the various extreme events.
346 D.M. Frangopol and P. Bocchini
where fcij is the residual practical flow capacity of should be increased to 0.2, or even up to 0.5. It should
highway segment ij; fc0 ij is the original practical flow be noted that when only two bridge states are
capacity of highway segment ij; max l is the maximum considered (i.e. in/out of service), this model can be
damage level among the bridges on highway segment seen as a generalisation of the bottleneck assumption.
ij; aRC and bRC are parameters, assumed equal to 0.03 In fact, the bottleneck assumption can be obtained as a
and 4, respectively. Also in this case, the values special case, by setting aUB ¼ 1. An even more
assumed for the parameters affect the final results, so sophisticate model has been proposed by Bocchini
they need to be thoroughly calibrated. and Frangopol (2011d). In this case, the delay caused
In a series of studies on resilience of networks (see by each bridge out of service is explicitly computed as a
Section 5), Bruneau and Reinhorn (2007) have function of the associated detour length, the flow
proposed another continuous relationship between capacity of secondary roads, the bridge damage
the damage state of a network component and its condition and the presence of ongoing restoration
serviceability. Even though they have studied different interventions on the bridge. Data required to calibrate
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
kinds of networks, and the relationship between such model for bridges in the United States are
damage level and serviceability (better called ‘func- provided by the National Bridge Inventory (FHWA
tionality’ in this case) is implicitly included in a 2009) and by the database distributed with the
sophisticated probabilistic framework, the continuous software HAZUS-MH MR4 (DHS 2009).
functions that describe the recovery in ‘quality of All the models that have been presented are
infrastructure’ during the repair activities suggest an characterised by advantages and drawbacks. Each of
underlying assumption conceptually similar to the one them is suitable for certain networks and inappropriate
in Equation (3). for others. Depending on the specific network and
For rural highway systems, with very limited the type of analysis that is performed, the choice of the
detour routes, the bottleneck assumption appears bridge scale–network scale interaction model and the
reasonable. However, when more detours are avail- careful calibration of its parameters are critical to
able, such as for urban highway systems, it is too obtain accurate results. These choices can benefit very
conservative. In fact, if a 10-km highway segment with much from empirical studies, data collection and traffic
five bridges and secondary roads is considered as an flow monitoring. The acquisition of this kind of data
example, the model has to yield very different results can be very challenging, especially after an extreme
when considering the collapse of an individual bridge event, but novel techniques, such as the use of mobile
or the collapse of all five bridges. The bottleneck phones geolocation (Herrera et al. 2010), appear very
assumption treats these two cases in the same way. On promising.
the contrary, if just one bridge has collapsed, the
vehicles can bypass that single bridge and use the rest
of the highway segment with only a limited delay. 4. Time-dependent problems
Instead, if all the bridges are collapsed, the entire In Section 2, some of the issues related to the large
highway segment is practically unserviceable. These scale of bridge networks in the space domain have been
considerations led Bocchini and Frangopol (2011c) discussed. Similarly, an aging infrastructure poses
to propose a new model for the bridge scale – network several problems due to the variability in time of its
scale interaction. Only two possible states for every characteristics, during its usually long life-cycle.
bridge are considered (in and out of service) and the Therefore, a bridge network should always be analysed
residual practical capacity of a highway segment is as a spatially distributed system with time-dependent
given by: properties.
This topic has already been widely investigated for
fcij ¼ fc0 UB individual bridges. For instance, Mori and Ellingwood
ij max 0; ð1 NUBij a Þ ð4Þ
(1993) pioneered the study of service-life assessment of
deteriorating concrete structures. Czarnecki and
where NUBij is the number of unserviceable bridges on Nowak (2008) developed life-cycle profiles of bridge
highway segment ij and aUB is a parameter that was reliability based on the model of the deterioration
proposed to be set equal to 0.1. This means that every process and on a probabilistic load description.
unserviceable bridge causes a reduction in the practical Frangopol et al. (1997) and later Estes and Frangopol
traffic flow capacity of 10%. Also in this case, the (2001) have proposed a life-cycle analysis of deterior-
value of the parameter aUB should be carefully ating structures based on system reliability and aimed
calibrated for the specific highway system, depending at assessing the risk of failure (for various limit states),
on the length of the segments and the possible ranking the components depending on their impor-
alternative routes. For rural highways, for instance, it tance and minimising the total maintenance and
Structure and Infrastructure Engineering 347
inspection cost. Frangopol et al. (2004) presented a only on physical models, but have to take into account
comprehensive state-of-the-art review of probabilistic also sociological aspects (e.g. drivers tend to adapt
models for the time-dependent degradation of bridge their routes and sometimes even their destinations
performance. Akgül and Frangopol (2004a,b, 2005a,b) depending on the traffic conditions).
have analysed the time-dependent reliability of several A consequence of the fact that all the nodes can be
structural types of bridges. Kong and Frangopol source and destination is the impossibility (in general)
(2003, 2004a, 2005) worked on the management to use the traditional series-parallel system reliability
optimisation of aging structures. Orcesi and Frangopol models. Figure 2 shows a simple bridge network, where
(2010, 2011) and Okasha and Frangopol (2011) studied the numbers in square boxes represent four cities, the
the integration of monitoring into life-cycle perfor- lines represent highway segments and the letters
mance optimisation techniques. Frangopol and represent two bridges. For a driver that goes from 1
Messervey (2007a,b, 2008, 2009) and Messervey et al. to 4, the two bridges A and B are in parallel. However,
(2011) focused on the use of structural health monitor- for a driver that goes from 2 to 3, the two bridges are in
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
ing to reduce the uncertainties, update the data, and series configuration. This trivial example shows that
thus optimise the management of bridges. Indeed, the reliability of a transportation network cannot be
the Structural Engineering Institute of the American computed as the reliability of a system built by
Society of Civil Engineers recently founded the components in series-parallel configuration, whose
Technical Council on Life-Cycle Performance, Safety, individual reliabilities are known.
Reliability and Risk of Structural Systems to study In most of the studies on the structural reliability of
all these topics (Frangopol and Ellingwood 2010). bridge networks, this problem is not addressed. On the
Fewer studies have addressed the effects of the contrary, several approaches have selected a single
individual component deterioration or damage in- node as origin and another one as destination of the
duced by extreme events on the network reliability. traffic flows or, at least, they have fixed a very limited
Akgül and Frangopol (2003, 2004c,d) have studied the number of origins and destinations (Liu and Frango-
effects of the time-dependent reliability of individual pol 2005c, 2006a, Kang et al. 2008, Golroo et al. 2010).
bridges on the network connectivity and the relative These approaches are computationally very efficient
importance of bridges. Shinozuka et al. (2006) and and accurate. They can give reliable information on the
Zhou et al. (2010) analysed the effectiveness of retrofit important ranking of the bridges for a few specific
interventions with a cost-benefit analysis. Lee et al. routes and useful indications for the maintenance
(2009) investigated the issue of the network flow planning, if the analysed routes are far more important
capacity estimation considering both extreme events than any other origin–destination pair in the network.
and deterioration. Kim et al. (2009a) proposed optimal However, they cannot describe the real dynamics of the
maintenance management techniques for railroad traffic flows and, therefore, are not suitable for a truly
bridge networks. Liu and Frangopol (2005c) and complete model of the bridge network.
Bocchini and Frangopol (2011c) have proposed More appropriate analysis techniques have been
techniques for the computation of the time-dependent developed by transportation engineers. One of the
network reliability (in terms of the ability to maintain a most comprehensive, advanced and popular methodol-
good level of service) as a function of the time- ogies is the one presented by Evans (1976). This
dependent reliability of the individual bridges. methodology has the advantage to be able to solve the
Further studies in this field are certainly desirable, combined traffic assignment and distribution problem
since the first investigations have proved that the without the need of an entire enumeration of the
impact of the degradation process on the network possible paths, which is computationally impractical
reliability can be very significant. for large networks (Rubin 1974). The transportation
network has to be described according to graph theory
(Gibbons 1985), while the travels of the network users a strongly non-linear function of the traffic flow, TTT
are distributed among the various cities by means of a and TTD are not proportional to each other. Other
‘gravitational model’ (Levinson and Kumar 1994). performance indicators, mostly based on the
This computational procedure imposes the ‘user total travel time, have been reviewed by Lomax et al.
equilibrium’ (Frank and Wolfe 1956, Lee and Ma- (2003).
chemehl 2005) by satisfying the following two princi- Another important indicator of the network
ples (Wardrop 1952): (i) travel times on all routes performance is the measure of the possibility to reach
actually used are equal, and less than those which every node, from every other node. Several analytical
would be experienced by a single vehicle on any unused measures of this ability have been proposed in the
route and (ii) the total network travel time is minimum. literature. For instance, Liu and Frangopol (2006a)
Moreover, this technique is particularly suitable for computed ‘connectivity’ as the probability to be able to
the analysis of bridge networks that also take into reach a specific destination from a fixed origin. Ng and
account extreme events. In fact, it does not assume that Efstathiou (2006) defined the ‘network disconnected-
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
the travel origin and destination are fixed a priori, ness’ as:
instead they are adaptive to the conditions of the
network (i.e. the in/out of service state of the bridges e
NetDis ¼ ð8Þ
and/or their damage level). Therefore, combining this 1max
network analysis technique with one of the methodol-
ogies for the components–network interaction pre- where e is the number of unreachable pairs of nodes
sented in Section 3, it is possible to assess the network and 1max is the maximum possible number of links.
performance after the occurrence of extreme events, Bocchini and Frangopol (2011a) have introduced the
bridge failures and all along the life-cycle (Bocchini ‘Fully Connected Ratio’, computed through Monte
and Frangopol 2010, 2011a,c). Carlo Simulation as:
The most common transportation network perfor- Scott et al. (2006) presented a critical review of the
mance indicator is the total travel time (TTT), defined literature about the use of the ratio
as the time spent to reach the destination by all the
users that depart in a fixed time window, such as one V traffic volume on a high way segment
hour (Bocchini and Frangopol 2010): ¼ ð10Þ
C high way segment capacity
XXZ fij
as an index of the important ranking for highway
TTT1 ¼ tij ðfÞdf ð5Þ
i2I j2J 0 network segments.
In the same study, the network robustness index
or sometimes computed as (Scott et al. 2006): (NRI) is proposed as a better indicator, since it takes
XX into account the benefits for the overall network due to
TTT2 ¼ fij tij ð6Þ the improvement of an individual segment capacity:
i2I j2J
X
NRIa ¼ fa ta da TTT2 ð11Þ
In both Equations (5) and (6), i and j are nodes of a
the network; I is the entire set of network nodes; J is
the subset of nodes that can be reached from node i where index a runs over the highway segments; fa is the
using a single highway segment ij; fij is the traffic flow traffic flow on segment a as computed by the network
on segment ij as computed by the network analysis and analysis; ta is the time required to cover segment a, as
tij is the time required to cover segment ij, as computed computed by the network analysis; da is equal to 1 if
by the network analysis. Similarly, the total travel segment a is not the one considered (removed) and 0
distance (TTD) can be computed as: otherwise; and TTT2 is given by Equation (6), with all
XX the bridges intact. NRIa is computed for every highway
TTD ¼ fij lij ð7Þ segment and it quantifies the performance loss due to
i2I j2J
the removal of a highway segment (caused for instance,
where lij is the length of segment ij. Since the time by the failure of its bridges). The segments with the
required to cover a segment and its length are linked by highest NRIa are those that should be maintained first.
Structure and Infrastructure Engineering 349
Figure 4. Pareto front of optimal solutions for the network level maintenance.
application for each bridge of the network. The grey performance indicator is resilience. Several definitions
area is the domain of solutions that are compliant with of resilience in civil engineering can be found in the
the constraints but are non-optimal. Three dots are literature, but the most widely accepted is the one
highlighted and labelled S1, S2 and S3. The Pareto provided by Bruneau et al. (2003): ‘resilience is defined
front presents a rapid increase after the region where as the ability of social units (e.g. organisations,
solution S2 is located. This means that all the optimal communities) to mitigate hazards, contain the effects
solutions on the right hand side of the group to which of disasters when they occur, and carry out recovery
S2 belongs, such as S3, determine a small increase in activities in ways that minimise social disruption and
the network performance associated, with a large mitigate the effects of future earthquakes’. In the same
increase in the expected total cost. On the contrary, article and in subsequent ones (Cimellaro et al. 2006,
the optimal solutions on the left hand side, such as S1, 2010, Bruneau and Reinhorn 2007), an exhaustive
yield a limited reduction in the total cost associated description of the properties of resilience and an
with a large loss in the network performance. There- analytical formulation are provided,
fore, in this specific example, the overall most Z t0 þth
convenient solutions are those belonging to the cluster r1 ¼ QðtÞdt ð13Þ
of S2. t0
This kind of study is a very valuable tool for
decision makers. It provides an overall view of the where r1 is the resilience associated to a certain extreme
effects of the maintenance interventions and drives event, t0 is the time of occurrence of the extreme event,
the choice of the optimal one. However, it should th is the investigated time horizon and Q is the
always be combined with other studies on the optimal functionality. Alternatively, the value of resilience
maintenance of individual bridges (Liu and Frangopol can be normalised with respect to the investi-
2005d,e, Bucher and Frangopol 2006, Furuta et al. gated time horizon in order to have a non-dimensional
2006, 2011, Neves et al. 2006a,b, Yang et al. 2006a,b, index:
Frangopol and Liu 2007, Petcherdchoo et al. 2008, Z t0 þth
Orcesi et al. 2010) and on studies on the bridge 1
r2 ¼ QðtÞd t ð14Þ
importance ranking (see, for instance, Akgül and th t0
Frangopol 2003, 2004e, Liu and Frangopol 2005b).
Several studies have used resilience as restoration
performance index or as prioritisation criterion for the
6.3. Post-event recovery and resilience restoration of bridge networks (Bocchini and Frango-
A particular case of intervention is the restoration of pol 2011d, Frangopol and Bocchini 2011b) and other
the network after an extreme event that has damaged lifelines (Çagnan et al. 2006, Xu et al. 2007) subject to
the bridges. In this case, the most interesting network extreme events.
Structure and Infrastructure Engineering 351
The properties of resilience and the technical, be disregarded. Therefore, the network analysis cannot
organisational, social and economic factors by which consist entirely of a repeated application of analyses on
it is affected are thoroughly defined qualitatively in the the individual components. A clear example has been
mentioned references. However, to the best of the given in Section 6.2; the bridge maintenance schedule
authors’ knowledge, the quantification of these proper- that yields the optimal network performances is not
ties is strongly dependent on the application and a obtained as the collection of the optimal schedules for
complete framework for the resilience analysis of individual bridges.
bridge networks is still missing. When all the components of a networked system
have to be analysed together, computational issues
arise very frequently and often they become the main
6.4. Networks of networks challenge in the solution of the problem. In particular,
Another interesting problem is the one of the interac- when uncertainties are involved, or when a numerical
tion between several networks of the same type or of optimisation is performed, the solution has to be
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
and decision making are just a few of the skill sets that Alexoudi, M.N., Kakderi, K.G., and Pitilakis, K.D., 2009.
are usually required. For all these topics, analytical Seismic risk and hierarchy importance of interdependent
lifeline systems using fuzzy reasoning. In: H. Furuta,
and computational proficiency, as well as experimental D.M. Frangopol, and M. Shinozuka, eds. Safety,
facilities are required. Therefore, it appears evident reliability and risk of structures, infrastructures and
that these studies can only be performed by strong engineering systems. Osaka, Japan: CRC Press, 2881–
research groups that involve people with different 2888.
backgrounds. Moreover, the scientific community American Society of Civil Engineers (ASCE), 2009. Report
card for America’s infrastructure. American Society of
plays a role of utmost importance, providing knowl- Civil Engineers.
edge and advancements in the fields that cannot be Augusti, G., Ciampoli, M., and Frangopol, D.M., 1998.
covered by an individual research group. The future of Optimal planning of retrofitting interventions on bridges
scientific research on lifelines and networks strongly in a highway network. Engineering Structures, 20 (11),
relies on the ability of researchers around the world to 933–939.
Bocchini, P., Frangopol, D.M., and Deodatis, G., 2011.
operate as a strongly interconnected, redundant and
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
Chang, S.E., Shinozuka, M., and Moore, J.E., II, 2000. Frangopol, D.M. and Bocchini, P., 2011a. Integrated
Probabilistic earthquake scenarios: extending risk analy- maintenance-monitoring-management framework for
sis methodologies to spatially distributed systems. Earth- optimal decision making in bridge life-cycle performance:
quake Spectra, 16 (3), 557–572. emphasis on SHM and bridge networks. Proceedings of
Chen, W.-F. and Duan, L., 2000. Bridge engineering hand- the 2011 NSF Engineering Research and Innovation
book. Boca Raton: CRC Press. Conference, 4–7 January 2011, Atlanta, GA. In: En-
Cimellaro, G.P., Reinhorn, A.M., and Bruneau, M., 2006. gineering for Sustainability and Prosperity Disk 1 (Data
Quantification of seismic resilience. In: Proceedings of the DVD): Papers, 9.
8th national conference of earthquake engineering, 18–22 Frangopol, D.M. and Bocchini, P., 2011b. Resilience as
April 2006, San Francisco, California. optimization criterion for the rehabilitation of bridges
Cimellaro, G.P., Reinhorn, A.M., and Bruneau, M., 2010. belonging to a transportation network subject to earth-
Seismic resilience of a hospital system. Structure and quake. In: SEI-ASCE 2011 Structures Congress. 14–16
Infrastructure Engineering, 6 (1), 127–144. April 2011, Las Vegas, NV.
Czarnecki, A.A. and Nowak, A.S., 2008. Time-variant Frangopol, D.M. and Ellingwood, B.R., 2010. Life-cycle
reliability profiles for steel girder bridges. Structural performance, safety, reliability and risk of structural
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
Furuta, H., et al., 2006. Optimal bridge maintenance Kim, S.-H. and Feng, M.Q., 2003. Fragility analysis of
planning using improved multi-objective genetic algo- bridges under ground motion with spatial variation.
rithm. Structure and Infrastructure Engineering, 2 (1), 33– International Journal of Non-Linear Mechanics, 38 (5),
41. 705–721.
Furuta, H., Frangopol, D.M., and Nakatsu, K., 2011. Life- Kim, Y., et al., 2009b. Seismic risk assessment of complex
cycle cost of civil infrastructure with emphasis on interacting infrastructures using matrix-based system
balancing structural performance and seismic risk of reliability method. In: H. Furuta, D.M. Frangopol, and
road network. Structure and Infrastructure Engineering, 7 M. Shinozuka, eds. Safety, reliability and risk of
(1–2), 65–74. structures, infrastructures and engineering systems. Osaka,
Gao, L., Xie, C., and Zhang, Z., 2010. Network-level multi- Japan: CRC Press, 2889–2893.
objective optimal maintenance and rehabilitation sche- Kong, J.S. and Frangopol, D.M., 2003. Evaluation of
duling. In: Proceedings of the 89th annual meeting of the expected life-cycle maintenance cost of deteriorating
Transportation Research Board of the national academies, structures. Journal of Structural Engineering, 129 (5),
10–14 January 2010, Washington, DC, USA. 682–691.
Ghosn, M., Moses, F., and Frangopol, D.M., 2010. Kong, J.S. and Frangopol, D.M., 2004a. Cost-reliability
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
Redundancy and robustness of highway bridge super- interaction in life-cycle cost optimization of deteriorating
structures and substructures. Structure and Infrastructure structures. Journal of Structural Engineering, 130 (11),
Engineering, 6 (1–2), 257–278. 1704–1712.
Gibbons, A.M., 1985. Algorithmic graph theory. Cambridge: Kong, J.S. and Frangopol, D.M., 2004b. Prediction of
Cambridge University Press. reliability and cost profiles of deteriorating structures
Gioffré, M., Gusella, V., and Grigoriu, M., 2000. Simulation under time- and performance-controlled maintenance.
of non-Gaussian field applied to wind pressure fluctua- Journal of Structural Engineering, 130 (12), 1865–1874.
tions. Probabilistic Engineering Mechanics, 15, 339–345. Kong, J.S. and Frangopol, D.M., 2005. Probabilistic
Goldberg, D.E., 1989. Genetic algorithms in search, optimiza- optimization of aging structures considering maintenance
tion, and machine learning. Reading, MA: Addison- and failure costs. Journal of Structural Engineering, 131
Wesley Professional. (4), 600–616.
Golroo, A., Mohaymany, A.S., and Mesbah, M., 2010. Korving, H., et al., 2009. Risk-based design of sewer system
Reliability based investment prioritization in transporta- rehabilitation. Structure and Infrastructure Engineering, 5
tion networks. In: Proceedings of the 89th annual meeting (3), 215–227.
of the Transportation Research Board of the national Lee, C. and Machemehl, R.B., 2005. Combined traffic signal
academies, 10–14 January 2010, Washington, DC, USA. control and traffic assignment: algorithms, implementation
Hernández-Fajardo, I. and Dueñas-Osorio, L., 2009. Time and numerical results. Southwest Region University
sequential evolution of interdependent lifeline systems. Transportation Center, Center for Transportation Re-
In: H. Furuta, D.M. Frangopol, and M. Shinozuka, eds. search University of Texas at Austin, Research Report
Safety, reliability and risk of structures, infrastructures No. SWUTC/05/472840-00074-1.
and engineering systems. Osaka, Japan: CRC Press, 2864– Lee, Y.-J., Song, J., and Gardoni, P., 2009. Post-hazard flow
2871. capacity of bridge transportation network considering
Herrera, J.C., et al., 2010. Evaluation of traffic data obtained structural deterioration. In: H. Furuta, D.M. Frangopol,
via GPS-enabled mobile phones: The Mobile Century and M. Shinozuka, eds. Safety, reliability and risk of
field experiment. Transportation Research Part C: Emer- structures, infrastructures and engineering systems. Osaka,
ging Technologies, 18 (4), 568–583. Japan: CRC Press, 2894–2901.
Jankowski, R., 2006. Numerical simulations of space-time Levinson, D.M. and Kumar, A., 1994. Multimodal trip
conditional random fields of ground motions. Computa- distribution: structure and application. Transportation
tional Science – ICCS 2006, 56–59. Research Record, 1466, 124–131.
Jankowski, R. and Wilde, K., 2000. A simple method of Liu, M. and Frangopol, D.M., 2005a. Balancing connectivity
conditional random field simulation of ground motions of deteriorating bridge networks and long-term main-
for long structures. Engineering Structures, 22 (5), 552– tenance cost through optimization. Journal of Bridge
561. Engineering, 10 (4), 468–481.
Kallen, M.-J. and van Noortwijk, J., 2006. Optimal periodic Liu, M. and Frangopol, D.M., 2005b. Bridge annual
inspection of a deterioration process with sequential maintenance prioritization under uncertainty by multi-
condition states. International Journal of Pressure Vessels objective combinatorial optimization. Computer Aided
and Piping, 83, 249–255. Civil and Infrastructure Engineering, 20 (5), 343–353.
Kang, W.-H., Song, J., and Gardoni, P., 2008. Matrix-based Liu, M. and Frangopol, D.M., 2005c. Time-dependent
system reliability method and application to bridge bridge network reliability: novel approach. Journal of
networks. Reliability Engineering and System Safety, 93 Structural Engineering, 131 (2), 329–337.
(11), 1584–1593. Liu, M. and Frangopol, D.M., 2005d. Multiobjective
Kim, L., Cho, H.-N., and Cho, C., 2009a. Life cycle maintenance planning optimization for deteriorating
performance-based optimal allocation methodology for bridges considering condition, safety and life-cycle cost.
railroad bridge networks. In: H. Furuta, D.M. Frango- Journal of Structural Engineering, 131 (5), 833–842.
pol, and M. Shinozuka, eds. Safety, reliability and risk of Liu, M. and Frangopol, D.M., 2005e. Maintenance planning
structures, infrastructures and engineering systems. Osaka, of deteriorating bridges by using multiobjective optimi-
Japan: CRC Press, 2033–2040. zation. Transportation Research Record: Journal of the
Kim, S. and Frangopol, D.M., 2010. Optimal planning of Transportation Research Board, 11 (S), 491–500.
structural performance monitoring based on reliability Liu, M. and Frangopol, D.M., 2006a. Probability-based
importance assessment. Probabilistic Engineering Me- bridge network performance evaluation. Journal of
chanics, 25 (1), 86–98. Bridge Engineering, 11 (5), 633–641.
Structure and Infrastructure Engineering 355
Liu, M. and Frangopol, D.M., 2006b. Optimizing bridge Peeta, S. and Zhang, P., 2009. Modeling infrastructure
network maintenance management under uncertainty with interdependencies: theory and practice. In: H. Furuta,
conflicting criteria: life-cycle maintenance, failure, and D.M. Frangopol, and M. Shinozuka, eds. Safety, relia-
user costs. Journal of Structural Engineering, 132 (11), bility and risk of structures, infrastructures and engineering
1835–1845. systems. Osaka, Japan: CRC Press, 2872–2880.
Lomax, T., et al., 2003. Selecting travel reliability measures. Peeta, S., et al., 2010. Pre-disaster investment decisions for
Texas Transportation Institute Monograph, College strengthening a highway network. Computers and Opera-
Station, Texas. tions Research, 37 (10), 1708–1719.
Matisziw, T.C. and Murray, A.T., 2009. Modeling s-t path Petcherdchoo, A., Neves, L.A.C., and Frangopol, D.M.,
availability to support disaster vulnerability assessment 2008. Optimizing lifetime condition and reliability of
of network infrastructure. Computers and Operations deteriorating structures with emphasis on bridges.
Research, 36 (1), 16–26. Journal of Structural Engineering, 134 (4), 544–552.
Messervey, T.B., Frangopol, D.M., and Casciati, S., 2011. PONTIS, 2005. User’s manual, Release 4.4. Cambridge, MA:
Application of the statistics of extremes to the reliability Cambridge Systematics, Inc.
assessment and performance prediction of monitored Rose, A., 2004. Defining and measuring economic resilience
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017
highway bridges. Structure and Infrastructure Engineer- to disasters. Disaster Prevention and Management, 13 (4),
ing, 7 (1–2), 87–99. 307–314.
Moghtaderi-Zadeh, M. and Der Kiureghian, A., 1983. Rose, A. and Liao, S., 2005. Modelling regional economic
Reliability upgrading of lifeline networks for post-earth- resilience to disasters: acomputable general equilibrium
quake serviceability. Earthquake Engineering & Structur- analysis of water service disruptions. Journal of Regional
al Dynamics, 11 (4), 557–566. Science, 45 (1), 75–112.
Mori, Y. and Ellingwood, B.R., 1993. Reliability-based Rubin, F., 1974. A search procedure for Hamilton paths and
service-life assessment of aging concrete structures. circuits. Journal of the Association for Computing
Journal of Structural Engineering, 119 (5), 1600–1621. Machinery, 21 (4), 576–580.
Murray, A.T., Matisziw, T.C., and Grubesic, T.H., 2008. A Scott, D.M., et al., 2006. Network Robustness Index: a new
methodological overview of network vulnerability ana- method for identifying critical links and evaluating the
lysis. Growth and Change, 39 (4), 573–592. performance of transportation networks. Journal of
Neves, L.A.C., Frangopol, D.M., and Cruz, P.J., 2006a. Transport Geography, 14 (3), 215–227.
Probabilistic lifetime-oriented multiobjective optimiza- Shinozuka, M., et al., 1999. Modeling, synthetics and
tion of bridge maintenance: single maintenance type. engineering applications of strong earthquake wave
Journal of Structural Engineering, 132 (6), 991–1005. motion. Soil Dynamics and Earthquake Engineering, 18
Neves, L.A.C., Frangopol, D.M., and Petcherdchoo, A., (3), 209–228.
2006b. Probabilistic lifetime-oriented multi-objective Shinozuka, M., et al., 2006. Cost-effectiveness of seismic
optimization of bridge maintenance: combination of bridge retrofit. In: P.J.S. Cruz, D.M. Frangopol, and
maintenance types. Journal of Structural Engineering, L.A.C. Neves, eds. Bridge maintenance, safety, man-
132 (11), 1821–1834. agement, life-cycle performance and cost. London: Taylor
Ng, A.K.S. and Efstathiou, J., 2006. Structural robustness of & Francis.
complex networks. In: Proceedings of the international Simiu, E. and Scanlan, R., 1996. Wind effects on structures.
workshop and conference on network science –NetSci2006, 3rd ed. New York: John Wiley and Sons.
Bloomington, IN. Song, J. and Kang, W.-H., 2009. System reliability and
Okasha, N.M. and Frangopol, D.M., 2010. Novel approach sensitivity under statistical dependence by matrix-based
for multi-criteria optimization of Life-Cycle preventive system reliability method. Structural Safety, 31 (2), 148–
and essential maintenance of deteriorating structures. 156.
Journal of Structural Engineering, 136 (8), 1009–1022. Tomassini, M., 1995. A survey of genetic algorithms. Annual
Okasha, N.M. and Frangopol, D.M., 2011. Integration of Reviews of Computational Physics, 3 (1), 87–118.
structural health monitoring in a system performance van Noortwijk, J.M. and Frangopol, D.M., 2004. Two
based life-cycle bridge management framework. probabilistic life-cycle maintenance models for deterior-
Structure and Infrastructure Engineering, doi: 10.1080/ ating civil infrastructures, Probabilistic Engineering Me-
15732479.2010.485726. chanics, 19 (4), 345–359.
Orcesi, A.D. and Frangopol, D.M., 2010. Optimization of Wardrop, J.G., 1952. Some theoretical aspects of road traffic
bridge management under budget constraints: Role of research. ICE Proceedings, Engineering Divisions, 1, 325–
structural health monitoring. Transportation Research 362.
Record: Journal of the Transportation Research Board, 1– Xu, N., et al., 2007. Optimizing scheduling of post-earth-
3 (2202), 148–158. quake electric power restoration tasks. Earthquake
Orcesi, A.D. and Frangopol, D.M., 2011. Optimization of Engineering & Structural Dynamics, 36 (2), 265–284.
bridge maintenance strategies based on structural Yang, S.I., et al., 2006a. The use of lifetime functions in the
health monitoring information. Structural Safety, 33 optimization of interventions on existing bridges con-
(1), 26–41. sidering maintenance and failure costs. Reliability En-
Orcesi, A.D., Frangopol, D.M., and Kim, S., 2010. gineering and System Safety, 91 (6), 698–705.
Optimization of bridge maintenance strategies based on Yang, S.I., Frangopol, D.M., and Neves, L.C., 2006b.
multiple limit states and monitoring. Engineering Struc- Optimum maintenance strategy for deteriorating bridge
tures, 32 (3), 627–640. structures based on lifetime functions. Engineering
Pandey, M.D., Yuan, X.-X., and van Noortwijk, J.M., 2009. Structures, 28 (2), 196–206.
The influence of temporal uncertainty of deterioration on Zerva, A., 2009. Spatial Variation of Seismic Ground
life-cycle management of structures. Structure and Infra- Motions. Boca Raton, USA: CRC Press, Taylor &
structure Engineering, 5 (2), 145–156. Francis Group.
356 D.M. Frangopol and P. Bocchini
Zerva, A. and Beck, J.L., 2003. Identification of parametric Zhang, R. and Deodatis, G., 1996. Seismic ground motion
ground motion random fields from spatially recorded synthetics of the 1989 Loma Prieta earthquake. Earth-
seismic data. Earthquake Engineering and Structural quake Engineering and Structural Dynamics, 25, 465–481.
Dynamics, 32, 771–791. Zhou, Y., Banerjee, S., and Shinozuka, M., 2010. Socio-
Zerva, A. and Zervas, V., 2002. Spatial variation of seismic economic effect of seismic retrofit of bridges for highway
ground motions: an overview. Applied Mechanics Re- transportation networks: a pilot study. Structure and
views, 55 (3), 271–297. Infrastructure Engineering, 6 (1), 145–157.
Downloaded by [IIT Indian Institute of Technology - Mumbai] at 07:33 23 October 2017