Resilient Cities and Structures 1 (2022) 40–51

Contents lists available at ScienceDirect

Resilient Cities and Structures

journal homepage: www.elsevier.com/locate/rcns

Full Length Article

Time-dependent reliability and resilience of aging structures exposed to

multiple hazards in a changing environment
Cao Wang a,∗, Bilal M. Ayyub b, Aziz Ahmed a
a
School of Civil, Mining and Environmental Engineering, University of Wollongong, Wollongong, NSW 2522, Australia
b
Center for Technology and Systems Management, Department of Civil and Environmental Engineering, University of Maryland, College Park, Maryland 20742, United
States

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

Keywords: Engineering structures are often subjected to the influences of performance deterioration and multiple hazards
Time-dependent reliability during their service lives, and consequently may suffer from damage/failure as a result of external loads. Struc-
Time-dependent resilience tural reliability and resilience assessment is a powerful tool for quantifying the structural ability to withstand
Repairable structures
these environmental or operational attacks. This paper proposes new formulas for structural time-dependent re-
Climate change
liability and resilience analyses in the presence of multiple hazards, which are functions of the duration of the
Multiple hazards
Adaptive design reference period of interest. The joint impacts of nonstationarities in multiple hazards due to a changing en-
vironment, as well as the deterioration of structural performance, are explicitly incorporated. The correlation
between the structural resistances/capacities associated with different hazard types is modeled by employing a
copula function. It is observed that, under the context of multiple hazards and aging effects, the time-dependent
resilience takes a generalized form of time-dependent reliability. The proposed formulas can be used to guide
the adaptive design of structures, where adaptive strategies are identified across a range of possible future ser-
vice conditions. An example is presented to demonstrate the applicability of the proposed method for structural
reliability and resilience analyses.

1. Introduction Traditionally, one may conduct reliability analysis to determine the

safety level of structures, which is defined as the probability of not vio-
Civil structures and infrastructures are often subjected to the threats lating a particular limit state of interest. Let 𝑍(𝐗) = 𝑅(𝐗) − 𝑆(𝐗) be the
of natural and man-made hazards such as hurricanes and earthquakes. limit state function, where 𝑅(𝐗) is the structural resistance, 𝑆(𝐗) is the
These hazards may trigger significant damage/failure to the struc- load effect, and 𝐗 is a random vector representing the uncertainties as-
tures and further socio-economic losses as ripple effects [1,2]. Fur- sociated with structural properties (e.g., material strength, geometry).
thermore, the structural performance (e.g., strength, stiffness) also de- The structural reliability, 𝑅𝑙 , equals the probability of 𝑍 > 0 (or equiv-
grades during the service life, which may impair structural safety un- alently, 𝑅 > 𝑆; see Fig. 1(a)), and conversely, the failure probability,
der an acceptable level (e.g., that set for newly-built ones) [3]. For ex- 𝑃𝑓 , is determined by 𝑃𝑓 = Pr (𝑍 < 0) = Pr (𝑅 < 𝑆), in which Pr( ) denotes
ample, as reported in the America’s Infrastructure Report Card 2021 the probability of the event in the brackets. The reliability index, 𝛽, is
(https://infrastructurereportcard.org), there are more than 617,000 mapped to the failure probability as follows,
bridges in the United States, 42% of which have served for at least 50
𝛽 = Φ−1 (1 − 𝑃𝑓 ) (1)
years, and 7.5% are considered structurally deficient (i.e., in “poor” con-
dition). In practice, it is difficult or even impossible to exactly predict where Φ−1 is the inverse cumulative distribution function (CDF) of a
the influences of these threatening factors due to their randomness na- standard normal distribution.
ture. As a result, a probability-based framework is essential to evaluate When estimating structural reliability associated with a reference
the safety and serviceability of in-service structures. For instance, in [4], period, the reliability would be dependent on the duration of the pe-
it is stated that “the compounding impact of the disasters that Australia riod if considering the impact of structural resistance deterioration (i.e.,
has experienced is a warning sign for the uncertainty and risk that lies time-varying 𝑅(𝐗)) [5], known as time-dependent reliability. Among
ahead.” the early attempts was the work by Mori and Ellingwood [6], where
a closed-form solution was proposed for time-dependent reliability as-
sessment of aging structures based on a stationary Poisson process for

∗
Corresponding author.
E-mail addresses: [email protected] (C. Wang), [email protected] (B.M. Ayyub), [email protected] (A. Ahmed).

https://doi.org/10.1016/j.rcns.2022.10.001
Received 1 September 2022; Received in revised form 12 October 2022; Accepted 12 October 2022
Available online 29 October 2022
2772-7416/© 2022 The Author(s). Published by Elsevier B.V. on behalf of College of Civil Engineering, Tongji University. This is an open access article under the
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Fig. 1. Comparison between structural reliability and resilience.

the external loads. The formula was later applied in several engineering time of full recovery. Further, a normalized resilience model, 𝑅𝑠 , was
problems (e.g., [7,8]), and was adopted in the international standard introduced as follows [22,23],
Bases for design of structures – Assessment of existing structures [9]. How- 𝑡
ever, the reliability method in [6] did not consider the time-variation ∫𝑡 1 𝑄(𝑡)𝑑𝑡
𝑅𝑠 = 0
(2)
of loads in terms of occurrence rate and/or magnitude. In fact, many 𝑡1 − 𝑡0
types of hazards are associated with nonstationary characteristics in a
However, in Eq. (2), the randomness nature arising from the stochastic
changing environment [10–14], which should be projected reasonably
load process was not incorporated. One may also refer to [24] for dis-
for use in structural reliability assessment. For example, climate change
cussions on the limitations of Eq. (2). This suggests the necessity for a
was predicted in [12] to increase the frequency of high-intensity storms
resilience model that is for a planning horizon, called “time-dependent
in selected ocean basins. The sixth assessment report of the Intergov-
resilience”, where the uncertainty associated with the load occurrence
ernmental Panel on Climate Change (IPCC) [15] revealed strengthened
process is captured. The time-dependent resilience can be used to quan-
evidence for human influence (e.g., emission of greenhouse gases) on
titatively measure the structural ability to adopt to and recover from
the warming of the atmosphere, ocean and land components of the cli-
extreme conditions during its life cycle, and to guide the design of new
mate system. Another example is that, with the development of regional
structures provided a target resilience level. Ayyub [25] developed a
economics and population, the traffic loads may display a growing trend
resilience metric for a reference period of [0, 𝑡𝑙 ], employing a stationary
for bridges [13,14]. Motivated by the nonstationarity in loads, Li et al.
Poisson process (with a constant occurrence rate of 𝜆) for the external
[16] improved the reliability method in [6] so that the time-variant
loads. Wang and Ayyub [26] proposed a new measure for structural
properties of load process can be explicitly incorporated in structural
time-dependent resilience, where the nonstationarity in loads and the
time-dependent reliability analysis. However, both methods [6,16] are
aging effects were explicitly incorporated. They also showed that the
applicable for the case of a single hazard type. This motivates a gener-
time-dependent reliability is a special case of time-dependent resilience.
alization of the reliability methods for use in the presence of multiple
However, in the context of multiple hazards, the models in [25,26] need
hazards. The significance of considering multiple hazards in structural
to be generalized for use in structural resilience analysis. Similar to the
reliability analysis has been evidenced in the literature. In [17], the de-
case of reliability analysis, the resilience assessment incorporating mul-
sign of structures against multiple hazards was investigated based on
tihazard scenarios is motivated by the fact that civil structures are often
minimization of expected life-cycle costs. It was found that the design is
subjected to a variety of complex loading mechanisms related to envi-
often dominated by the hazard that is with large uncertainty and causes
ronmental conditions [27–29]. This is particularly the case in a changing
significant consequence. Yanweerasak et al. [18] evaluated the life-cycle
environment, where the occurrences of extreme events are expected to
reliability of existing reinforced concrete (RC) bridges under multiple
display nonstationary (e.g., increasing) characteristics.
hazards, including seismic and airborne chloride hazards for RC bridge
The scope of this paper is to derive closed-form formulas for time-
piers. They also identified the hazards that most threaten the structural
dependent reliability and resilience assessment of aging structures sub-
safety of existing RC bridges, and the structural components with the
jected to multiple hazards. The impacts of nonstationarity in loads and
lowest reliability subjected to multiple hazards.
structural performance deterioration are taken into account. Similar to
As illustrated in Fig. 1(a), the analysis of structural reliability does
the observations from [26], it will be shown that, the structural relia-
not take into account the post-hazard recovery process. During the past
bility is a special case of structural resilience in the presence of multiple
decades, the concept of structural resilience has emerged, which mea-
hazards, as the former does not account for the post-hazard recovery
sures the structural ability to prepare for, absorb, recover from, and
process. The applicability of the proposed formulas, including its use in
adapt to adverse events [19]. In this paper, only the resilience of a single
adaptive design and risk management, will be demonstrated through an
repairable structure will be discussed, which, by nature, can be extended
example.
to a broader scale such as a system (consisting of multiple interdepen-
dent structures), and a community with multiple infrastructure systems.
The term “repairable” means the structure, if suffering from hazard- 2. Time-variant structural resistance and load effects
induced damage/failure, can be restored to the pre-hazard state via re-
pair measures to account for adaptability. In [20], it was argued that in 2.1. Deterioration of structural resistance
the framework of resilience, there are four key components known as
“4R”: Robustness, Redundancy, Resourcefulness and Rapidity. Bruneau For an in-service structure, its performance (e.g., resistance) may de-
𝑡
et al. [21] defined a measure for resilience loss as ∫𝑡 1 [1 − 𝑄(𝑡)]𝑑𝑡, where grade with time due to external attacks such as corrosion of steel bars,
0
𝑄(𝑡) is the structural performance function, taking a value between 0 and concrete carbonation [3,30–33]. A deterioration function is com-
and 1, 𝑡0 is the occurrence time of hazard (see Fig. 1(b)), and 𝑡1 is the monly used to describe the deterioration process of structural resistance,

41
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

denoted by 𝑔(𝑡), which is defined by [6,16,34],

𝑅(𝑡) = 𝑅0 ⋅ 𝑔(𝑡) (3)
where 𝑅(𝑡) is the resistance at time 𝑡, and 𝑅0 = 𝑅(0) is the initial resis-
tance. In Eq. (3), 𝑔(𝑡) may be a simple deterministic function, reflect-
ing the overall trend of resistance deterioration, or a stochastic process
(rewritten as 𝐺(𝑡)) that incorporates the uncertainty associated with the
deterioration process. In terms of the former, for example, the deterio-
ration function can be modeled as follows [6],
𝑔(𝑡) = 1 − 𝑎 ⋅ 𝑡𝜂 (4)
in which 𝑎 is a parameter accounting for the deterioration rate, while 𝜂
is representative of the deterioration type (dominate mechanism), tak-
ing a value of 1 for corrosion, 2 for sulfate attack, or 0.5 for diffusion-
controlled aging.
When fragility curves are used to represent structural ability to with-
stand hazardous events, one may employ the concept of “generalized Fig. 2. Joint distribution of correlated variables 𝑅1 and 𝑅2 (after [42]).
capacity” [35], whose CDF is exactly the same as the fragility curve as
a function of the load intensity. It is commonly assumed that a fragility
coefficient, 𝜏𝑅1 ,𝑅2 , is defined as the probability of concordance minus
curve has a shape of the CDF of a lognormal distribution, with which the
the probability of discordance for a random pair of samples (𝑟1,𝑖 , 𝑟2,𝑖 )
generalized capacity is a lognormal random variable. Mathematically, a
and (𝑟1,𝑗 , 𝑟2,𝑗 ) of (𝑅1 , 𝑅2 ), i.e.,
fragility curve with respect to a particular damage state, DS, is described
as follows, 𝜏𝑅1 ,𝑅2 = Pr [(𝑅1,𝑖 − 𝑅1,𝑗 )(𝑅2,𝑖 − 𝑅2,𝑗 ) > 0] − Pr [(𝑅1,𝑖 − 𝑅1,𝑗 )(𝑅2,𝑖 − 𝑅2,𝑗 ) < 0]
[ ]
ln(𝐼 𝑀 ) − ln(𝑚𝑑 ) (9)
Pr (𝐷𝑆|𝐼 𝑀 ) = Φ (5)
𝜉𝑑
Many types of copula function have been established in the litera-
where IM is the intensity measure of hazard, 𝑚𝑑 and 𝜉𝑑 are the median
ture. Illustratively, Table 1 shows four commonly-used copulas, namely
value and logarithmic standard deviation, respectively. Based on Eq. (5),
Gaussian, Clayton, Frank, and Gumbel. The parameter 𝜃 is determined
the probability density function (PDF) of the generalized capacity takes
by the correlation between the two described variables. For visualiza-
a form of
[ ( tion, Fig. 3 shows sampled pairs of correlated uniform variables (𝑈 , 𝑉 )
)2 ]
1 1 ln 𝑟 − ln(𝑚𝑑 ) with 𝜏𝑈 ,𝑉 = 0.8 for the four copulas (note that (𝑈 , 𝑉 ) can be mapped to
𝑓 (𝑟) = √ exp − , 𝑟≥0 (6)
2𝜋𝑟𝜉𝑑 2 𝜉𝑑 (𝑅1 , 𝑅2 ) by using 𝑈 = 𝐹𝑅1 (𝑅1 ), 𝑉 = 𝐹𝑅2 (𝑅2 )).
In the presence of 𝑛 mutually correlated variables (𝑛 > 2), one can
Similar to the deterioration scenario in Eq. (3), fragility curves are further generalize the bivariate copula functions in Table 1 to the 𝑛-
also associated with time-variant characteristics due to aging effects dimensional case [45]. With this regards, the Sklar’s Theorem also holds,
[36–38]. Under this context, one may also utilize a deterioration func- and Eq. (7) becomes the following for correlated variables 𝑅1 , 𝑅2 , … , 𝑅𝑛 ,
tion as in Eq. (3) to model the time-variation of the generalized capacity
[39,40]. In the following, both the structural resistance in Eq. (3) and
𝐹𝑅1 ,𝑅2 ,…,𝑅𝑛 (𝑟1 , 𝑟2 , … , 𝑟𝑛 ) = 𝐶𝑛 (𝐹𝑅1 (𝑟1 ), 𝐹𝑅2 (𝑟2 ), … 𝐹𝑅𝑛 (𝑟𝑛 )) (10)
the generalized capacity will be referred to as “resistance” for simplicity.
in which 𝐹𝑅1 ,𝑅2 ,…,𝑅𝑛 (𝑟1 , 𝑟2 , … , 𝑟𝑛 ) is the joint CDF of 𝑅1 , 𝑅2 , … , 𝑅𝑛 , and
2.2. Correlated resistances 𝐶𝑛 is an 𝑛-dimensional copula.

In the presence of multiple hazards (e.g., considering hurricane

winds and traffic loads simultaneously for a bridge), one would need 2.3. Load effects in a changing environment
to consider multiple resistances in structural reliability/resilience anal-
ysis. These resistances are by nature mutually correlated, since they are Many types of load effect are associated with nonstationary charac-
functions of the structural properties 𝐗. With this regard, copula func- teristics on the temporal scale in a changing environment. In this section,
tions can be used to describe the correlation between these resistances two selected load types will be discussed, namely hurricane winds1 and
(random variables). traffic loads.
The bivariate case is discussed first. For two correlated variables 𝑅1
and 𝑅2 , let 𝐹𝑅1 (𝑟1 ) and 𝐹𝑅2 (𝑟2 ) denote their marginal CDFs, respectively. 2.3.1. Hurricane winds
With the help of the Sklar’s Theorem [41], the joint CDF of 𝑅1 and 𝑅2 , A discrete stochastic process (e.g., Poisson process) can be used to
𝐹𝑅1 ,𝑅2 (𝑟1 , 𝑟2 ) (see Fig. 2), can be constructed as follows, describe the randomness associated with hurricane occurrence [46–48].
Due to the potential impacts of climate change, both the occurrence rate
𝐹𝑅1 ,𝑅2 (𝑟1 , 𝑟2 ) = 𝐶(𝐹𝑅1 (𝑟1 ), 𝐹𝑅2 (𝑟2 )) (7)
and the magnitude (e.g., maximum wind speed) conditional on occur-
where 𝐶(⋅, ⋅) is a copula function. With Eq. (7), the CDF of 𝑅1 conditional rence may vary with time. In terms of the former, the occurrence rate
on 𝑅2 = 𝑟2 , 𝐹𝑅1 |𝑟2 (𝑟1 ) is, is modeled as a function of time (denoted by 𝜆(𝑡)) in the Poisson model.
For the latter, the CDF of the maximum wind speed associated with one
1 𝜕𝐹𝑅1 ,𝑅2 (𝑟1 , 𝑟2 )
𝐹𝑅1 |𝑟2 (𝑟1 ) = ⋅ (8)
𝑓𝑅2 (𝑟2 ) 𝜕𝑟1
1
Hurricane are known as different names such as hurricanes, cyclones and
In order to describe the correlation between 𝑅1 and 𝑅2 , there are
typhoons. The only difference between these names is the location where the
several measures such as linear correlation (also known as Pearson’s storm occurs (see, e.g., http://oceanservice.noaa.gov/facts/cyclone.html). For
correlation) and rank correlation (e.g., Kendall’s tau) [43]. In particular, example, in the Atlantic and Northeast Pacific, the term “hurricane” is used; the
the rank correlation is more suitable to model the joint behaviour of same type of disturbance is known as a typhoon in the Northwest Pacific and a
random variables with high non-normality [44], and will be adopted cyclone in the South Pacific and Indian Ocean. In this paper, the three names
in the following. For correlated 𝑅1 and 𝑅2 , the Kendall’s correlation are not distinguished and will all be referred to as “hurricane”.

42
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Table 1
Four commonly-used types of copula function with parameter 𝜃.

Copula Type 𝐶(𝑢, 𝑣) Generator function Range of 𝜃

Gaussian ∗
𝐶(𝑢, 𝑣) = Φ2 (Φ (𝑢), Φ (𝑣), 𝜃)
−1 −1
/ [−1, 1]
[ ( )]−1∕𝜃 1 ( −𝜃 )
Clayton 𝐶(𝑢, 𝑣) = max 𝑢−𝜃 + 𝑣−𝜃 − 1, 0 𝑡 −1 [−1, 0) ∪ (0, ∞)
[ ] 𝜃
1 (exp(−𝜃𝑢)−1)(exp(−𝜃𝑣)−1) exp(−𝜃𝑡)−1
Frank 𝐶(𝑢, 𝑣) = − 𝜃 ln 1 + − ln exp(−𝜃)−1 𝜃≠0
[ ( exp(−𝜃)−1
)1∕𝜃 ]
Gumbel 𝐶(𝑢, 𝑣) = exp − (− ln 𝑢)𝜃 + (− ln 𝑣)𝜃 𝜃
(− ln 𝑡) 𝜃 ∈ [1, +∞)
∗Note. Φ2 is the CDF of a bivariate normal distribution with zero-mean, unit-variance, and
correlation coefficient 𝜃.

Fig. 3. Sampled pairs of (𝑈 , 𝑉 ) with 𝜏𝑈 ,𝑉 = 0.8 and different copula functions.

hurricane event can be described by a Weibull distribution [48] as fol- Consider a reference period of [0, 𝑡𝑙 ]. The deterioration of structural
lows, resistance is modeled by Eq. (3) with a deterministic deterioration func-
[ ( )𝛼(𝑡) ] tion 𝑔(𝑡). A nonhomogeneous Poisson process with a time-varying oc-
𝑣 currence rate of 𝜆(𝑡) is used to describe the occurrence of loads. Con-
𝐹𝑉 (𝑣, 𝑡) = 1 − exp − (11)
𝑢(𝑡)
ditional on occurrence, the CDF of load effect is 𝐹𝑆 (𝑠, 𝑡), which is also
in which 𝑣 is the 1-minute average wind speed, 𝑢(𝑡) and 𝛼(𝑡) are the time-variant. The hazard function, ℎ(𝑡), is defined as follows,
time-variant scale and shape parameters of Weibull distribution. Pr (𝑇𝑓 ≤ 𝑡 + 𝑑𝑡|𝑇𝑓 > 𝑡) Pr (𝑡 < 𝑇𝑓 ≤ 𝑡 + 𝑑𝑡)
ℎ(𝑡) = = (13)
𝑑𝑡 Pr (𝑇𝑓 > 𝑡)𝑑𝑡
2.3.2. Traffic loads
A traffic load model for use in reliability and resilience assessment of where 𝑇𝑓 is the failure time (a random variable). Conditional on 𝑅0 = 𝑟,
structures (e.g., bridges) should capture the extreme load events within a it follows that
reference period of interest. For example, let 𝐿𝑎 be the annual maximum ℎ(𝑡) = 𝜆(𝑡) ⋅ [1 − 𝐹𝑆 [𝑟 ⋅ 𝑔(𝑡), 𝑡]] (14)
traffic load, and 𝐿1 , 𝐿2 , … , 𝐿12 the monthly maximum traffic loads. By
definition, 𝐿𝑎 = max12 𝐿 , which formulates an “extreme value” prob- with which the time-dependent reliability, 𝑅𝑙 (0, 𝑡𝑙 ), is [16],
𝑖=1 𝑖
lem. With this regard, the generalized extreme value (GEV) distribution [ 𝑡𝑙 ] { 𝑡𝑙 }
can be used to reasonably model the maximum traffic load, based on 𝑅𝑙 (0, 𝑡𝑙 ) = exp − ℎ(𝑡)𝑑𝑡 = exp − 𝜆(𝑡)[1 − 𝐹𝑆 [𝑟 ⋅ 𝑔(𝑡), 𝑡]]𝑑𝑡
∫0 ∫0
the extreme value theory [5,14,49]. The CDF of GEV takes a form of,
(15)
( ( ( ))−1∕𝜉 )
⎧ 𝑥−𝜇
⎪exp − 1 + 𝜉 𝜎 , if 𝜉 ≠ 0 Based on Eq. (15), taking into account the uncertainty associated with
𝐹 (𝑥) = ⎨ ( ( )) (12) the initial resistance 𝑅0 , it follows that,
⎪exp − exp − 𝑥 − 𝜇
, if 𝜉 = 0 { }
⎩ 𝜎 ∞ 𝑡𝑙
𝑅𝑙 (0, 𝑡𝑙 ) = exp − 𝜆(𝑡)[1 − 𝐹𝑆 [𝑟 ⋅ 𝑔(𝑡), 𝑡]]𝑑𝑡 ⋅ 𝑓𝑅0 (𝑟)𝑑𝑟 (16)
where 𝜇, 𝜎, 𝜉 are the location, scale and shape parameters, respectively, ∫0 ∫0
satisfying 𝜉(𝑥 − 𝜇)∕𝜎 > −1. In Eq. (12), if the time-variation of traffic
where 𝑓𝑅0 (𝑟) is the PDF of 𝑅0 .
loads is considered, the three parameters become functions of time 𝑡,
Next, the time-dependent reliability with multiple hazards will be
denoted by 𝜇(𝑡), 𝜎(𝑡), 𝜉(𝑡). In particular, if 𝜉 = 0, Eq. (12) reduces to the
discussed. Consider the case of two hazard types first, denoted by 𝑆1
CDF of a Gumbel distribution (also known as Extreme Type I distribu-
and 𝑆2 . As illustrated in Fig. 4, both 𝑆1 and 𝑆2 are modeled by Poisson
tion).
processes with occurrence rates of 𝜆1 (𝑡) and 𝜆2 (𝑡), respectively. Corre-
sponding to 𝑆1 and 𝑆2 , the two resistance processes are 𝑅1 (𝑡) and 𝑅2 (𝑡),
3. Time-dependent reliability considering multiple hazards
whose deteriorations are described by Eq. (3). Within a reference period
of [0, 𝑡𝑙 ], the number of load events, 𝑁1 for 𝑆1 and 𝑁2 for 𝑆2 , follows a
In [16], a solution for time-dependent reliability assessment of aging
Poisson distribution. The probability mass function (PMF) of 𝑁𝑖 (𝑖 = 1, 2)
structures was proposed, which takes into consideration the impact of
is
a single load type. In this section, the method in [16] will be reviewed ( )𝑘 ( )
𝑡 𝑡
first, and generalized so that it can handle the reliability problem with ∫0 𝑙 𝜆𝑖 (𝑡)𝑑𝑡 exp − ∫0 𝑙 𝜆𝑖 (𝑡)𝑑𝑡
multiple hazards. Pr(𝑁𝑖 = 𝑘) = , 𝑘 = 0, 1, 2, … (17)
𝑘!

43
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Fig. 4. Time-dependent reliability in the presence of two load processes 𝑆1 and 𝑆2 .

It is assumed that 𝑆1 and 𝑆2 are two statistically independent processes. in which 𝐑 = {𝑅1 , 𝑅2 , … , 𝑅𝑛 }, 𝐫 = {𝑟1 , 𝑟2 , … , 𝑟𝑛 }, and 𝑓𝐑 (𝐫) is the joint
Conditional on 𝑅1 = 𝑟1 , 𝑅2 = 𝑟2 (note that 𝑅1 = 𝑅1 (0) and 𝑅2 = 𝑅2 (0) PDF of 𝐑, which can be determined through differentiating Eq. (10).
herein for simplicity), the hazard function is, according to Eq. (13), ob- With the proposed formula for time-dependent reliability in Eq. (22),
tained as follows, one can estimate the structural safety exposed to a range of changing
scenarios of loads. For example, in terms of adaptive design and risk
ℎ(𝑡)𝑑𝑡 = 𝜆1 (𝑡)𝑑𝑡 ⋅ [1 − 𝐹𝑆1 (𝑟1 ⋅ 𝑔1 (𝑡), 𝑡)] ⋅ (1 − 𝜆2 (𝑡)𝑑𝑡) + 𝜆2 (𝑡)𝑑𝑡 ⋅ management [28], in the presence of updated hazards (through, e.g.,
[1 − 𝐹𝑆2 (𝑟2 ⋅ 𝑔2 (𝑡), 𝑡)] ⋅ (1 − 𝜆1 (𝑡)𝑑𝑡) observational method), applying time-dependent reliability analysis can
+ 𝜆1 (𝑡)𝑑𝑡 ⋅ 𝜆2 (𝑡)𝑑𝑡 ⋅ [1 − 𝐹𝑆1 (𝑟1 ⋅ 𝑔1 (𝑡), 𝑡)] ⋅ guide the design of updated structural resistance/capacity to reflect the
variation in the hazards. Fig. 5 illustrates this concept, where the haz-
[1 − 𝐹𝑆2 (𝑟2 ⋅ 𝑔2 (𝑡), 𝑡)] (18) ard can be described by either the PDF of intensity or the probability
of exceedance, while the structural capacity is described by either a
in which 𝑔1 (𝑡) and 𝑔2 (𝑡) are the deterioration functions of 𝑅1 (𝑡) and 𝑅2 (𝑡),
PDF or fragility curve. Given a predefined threshold for structural re-
𝐹𝑆1 (𝑠, 𝑡) and 𝐹𝑆2 (𝑠, 𝑡) are the time-variant CDFs of 𝑆1 and 𝑆2 , respec-
liability (or failure probability), one would need to adjust the design
tively. As 𝑑𝑡 → 0, the hazard function is simplified as follows,
for structural capacity so as to adapt to the changes in loads. Further-
ℎ(𝑡) = 𝜆1 (𝑡) ⋅ [1 − 𝐹𝑆1 (𝑟1 ⋅ 𝑔1 (𝑡), 𝑡)] + 𝜆2 (𝑡) ⋅ [1 − 𝐹𝑆2 (𝑟2 ⋅ 𝑔2 (𝑡), 𝑡)] (19) more, adaptive risk management is usually based on a benefit-cost anal-
ysis [28,50], which assesses the cost-effectiveness of alternative risk-
with which the time-dependent reliability is obtained as follows, taking
mitigation strategies. The benefit-cost ratio is as follows,
into account the uncertainties associated with 𝑅1 and 𝑅2 ,
Benefit unmitigated risk − mitigated risk
∞ ∞ { 𝑡𝑙 = (23)
Cost cost to implement strategy
𝑅𝑙 (0, 𝑡𝑙 ) = exp − 𝜆1 (𝑡)[1 − 𝐹𝑆1 (𝑟1 ⋅ 𝑔1 (𝑡), 𝑡)]𝑑𝑡
∫0 ∫0 ∫0 A greater value of the benefit-cost ratio indicates a better risk-
𝑡𝑙 }
mitigation strategy from a view of cost-effectiveness. The probability
− 𝜆2 (𝑡)[1 − 𝐹𝑆2 (𝑟2 ⋅ 𝑔2 (𝑡), 𝑡)]𝑑𝑡 ⋅ 𝑓𝑅1 ,𝑅2 (𝑟1 , 𝑟2 )𝑑𝑟1 𝑑𝑟2 (20)
∫0 that a favorable benefit-cost ratio is achieved is estimated by
( )
where 𝑓𝑅1 ,𝑅2 (𝑟1 , 𝑟2 ) is the joint PDF of 𝑅1 and 𝑅2 . Benefit
Pr ≥ 1 = 1 − Pr (Benefit − Cost ≤ 0) (24)
In Eq. (20), if the aging effects are not considered (i.e., 𝑔1 (𝑡) = 𝑔2 (𝑡) ≡ Cost
1), the time-dependent reliability becomes, In Eq. (24), both the Benefit and Cost are random variables. The
underlying distribution of Benefit can be determined through risk and
∞ ∞
𝑅𝑙 (0, 𝑡𝑙 ) = Pr (𝑟1 > 𝑆max,1 ∩ 𝑟2 > 𝑆max,2 ) ⋅ 𝑓𝑅1 ,𝑅2 (𝑟1 , 𝑟2 )𝑑𝑟1 𝑑𝑟2 reliability analysis, where the proposed reliability method in Eq. (22) is
∫0 ∫0 applicable. For example, the Benefit could be a function of Δ𝑇𝑓 = 𝑇𝑓 ,2 −
= Pr [𝑅1 > 𝑆max,1 ∩ 𝑅2 > 𝑆max,2 ] (21) 𝑇𝑓 ,1 , where 𝑇𝑓 ,2 and 𝑇𝑓 ,1 are the structural service lives associated with
“mitigated risk” and “unmitigated risk”, respectively. With this regard,
where 𝑆max,1 = max𝑡∈[0,𝑡𝑙 ] 𝑆1 (𝑡), and 𝑆max,2 = max𝑡∈[0,𝑡𝑙 ] 𝑆2 (𝑡). Eq. (21) im- the CDF of service life equals 1 minus time-dependent reliability2 , with
plies that, the time-dependent reliability problem reduces to a classical which the PDF of Δ𝑇𝑓 is expressed in terms of 𝑅𝑙 (0, 𝑡𝑙 ) in Eq. (22). This
time-invariant one if ignoring the aging effects, since 𝑆max,1 , 𝑆max,2 are further leads to the probability distribution of Benefit as a function of
two random variables. Δ𝑇𝑓 .
The time-dependent reliability in Eq. (20) can be further extended
to the case of 𝑛 statistically independent load types, denoted by 4. Time-dependent resilience considering multiple hazards
𝑆1 , 𝑆2 , … , 𝑆𝑛 . Let 𝜆𝑖 (𝑡) be the occurrence rate of 𝑆𝑖 , 𝑔𝑖 (𝑡) the deteriora-
tion function of 𝑅𝑖 , and 𝐹𝑆𝑖 (𝑠, 𝑡) the CDF of 𝑆𝑖 conditional on occurrence. In this section, a new formula will be derived for time-dependent
With this, Eq. (20) is generalized as follows, resilience assessment of aging structures subjected to multiple hazards,
{ 𝑛 }
∞ ∞ ∑ 𝑡𝑙
𝑅𝑙 (0, 𝑡𝑙 ) = exp − 𝜆𝑖 (𝑡)[1 − 𝐹𝑆𝑖 (𝑟𝑖 ⋅ 𝑔𝑖 (𝑡), 𝑡)]𝑑𝑡 ⋅ 𝑓𝐑 (𝐫 )𝑑 𝐫 2
Mathematically, letting 𝐹𝑇𝑓 (𝑡) be the CDF of service life (or equivalently,
∫0 ∫0 ∫
𝑖=1 0 failure time as in Eq. (13)) 𝑇𝑓 , it follows that 𝐹𝑇𝑓 (𝑡) = Pr (𝑇𝑓 ≤ 𝑡) = 1 − Pr (𝑇𝑓 >
(22) 𝑡) = 1 − 𝑅𝑙 (0, 𝑡).

44
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Fig. 5. Original and updated hazard and capacity in the context of adaptive design and risk management (after [28]).

Fig. 6. Time-dependent resilience in the presence of two load processes 𝑆1 and 𝑆2 .

which is an extension of that in [26]. Let 𝑅0 be the structural resis- effects. The sequence of occurrence time associated with 𝑆1 is denoted
tance as before. The load occurrence is modeled by a Poisson process by {𝑡1,1 , 𝑡1,2 , … 𝑡1,𝑁1 }, while that for 𝑆2 is {𝑡2,1 , 𝑡2,2 , … 𝑡2,𝑁2 }, in which 𝑁1
with a rate of 𝜆(𝑡), and the probability of damage/failure at time 𝑡, and 𝑁2 are the (Poisson) numbers of load events for 𝑆1 and 𝑆2 , respec-
conditional on load occurrence, is denoted by 𝑝(𝑅0 , 𝑡) (which equals tively (see Eq. (17) for the PMF of 𝑁1 and 𝑁2 ). At time 𝑡𝑖,𝑗 (𝑖 = 1, 2; 𝑗 =
1 − 𝐹𝑆 [𝑅0 ⋅ 𝑔(𝑡), 𝑡] in Eq. (14)). The mean resilience associated with a 1, 2, … , 𝑁𝑖 ), if the load event causes structural damage/failure (i.e., the
damage/failure-causing load event is 𝑅𝑒 (𝑡) at time 𝑡. With this, the time- load effect exceeds the corresponding resistance) with a probability of
dependent resilience for a reference period of [0, 𝑡𝑙 ], 𝑅𝑠 (0, 𝑡𝑙 ), is [26], 𝑝𝑖 (𝑅𝑖 , 𝑡𝑖,𝑗 ), the functionality 𝑄(𝑡) decreases to 𝑄𝑟,𝑖,𝑗 (a random variable
within [0,1]), followed by a recovery process until time 𝑡𝑖,𝑗 + Δ𝑡𝑖,𝑗 (the
∞ { 𝑡𝑙 [ ] } duration of recovery is Δ𝑡𝑖,𝑗 ). This quantifies the resilience associated
𝑅𝑠 (0, 𝑡𝑙 ) = exp − 𝜆(𝑡)𝑝(𝑟, 𝑡) 1 − 𝑅𝑒 (𝑡) 𝑑𝑡 ⋅ 𝑓𝑅0 (𝑟)𝑑𝑟 (25)
∫0 ∫0 with a single load event, denoted by 𝑅𝑒,𝑆𝑖 (𝑡𝑖,𝑗 ) (a random variable).
By adopting the resilience models in [25,26], the time-dependent
in which 𝑓𝑅0 (𝑟) is the PDF of 𝑅0 as before. resilience for a reference period of [0, 𝑡𝑙 ], 𝑅𝑠 (0, 𝑡𝑙 ), is estimated as follows
Next, the resilience model in Eq. (25) will be extended to the case conditional on 𝑅1 = 𝑟1 , 𝑅2 = 𝑟2 ,
of multiple hazards. Starting from a relatively simple case with two (𝑁
∏ 1 [ ]
types of loads, denoted by 𝑆1 and 𝑆2 , the resilience problem is illus-
𝑅𝑠 (0, 𝑡𝑙 ) = 𝔼 𝑝1 (𝑟1 , 𝑡1,𝑗 )𝑅𝑒,𝑆1 (𝑡1,𝑗 ) + (1 − 𝑝1 (𝑟1 , 𝑡1,𝑗 )) ⋅
trated in Fig. 6, which is to be interpreted in conjunction with Fig. 4.
𝑗=1
The occurrences of both loads are modeled by Poisson processes, with
𝑁2 [
)
∏ ]
occurrence rates of 𝜆1 (𝑡) and 𝜆2 (𝑡), respectively. The structural perfor-
𝑝2 (𝑟2 , 𝑡2,𝑗 )𝑅𝑒,𝑆2 (𝑡2,𝑗 ) + (1 − 𝑝2 (𝑟2 , 𝑡2,𝑗 ))
mance function 𝑄(𝑡) is time-variant due to the consideration of aging 𝑗=1

45
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

(𝑁 )
∏ 1 [ ] in which 𝜆𝑖 (𝑡), 𝐑, 𝐫 and 𝑓𝐑 (𝐫 ) are as in Eq. (22), 𝑝𝑖 (𝑟𝑖 , 𝑡) is the proba-
=𝔼 𝑝1 (𝑟1 , 𝑡1,𝑗 )𝑅𝑒,𝑆1 (𝑡1,𝑗 ) + (1 − 𝑝1 (𝑟1 , 𝑡1,𝑗 )) ⋅ bility of structural damage/failure conditional on the occurrence of a
𝑗=1
load event (of the 𝑖th type) at time 𝑡, and 𝑅𝑒,𝑆𝑖 (𝑡) is the mean resilience
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
Part 1 associated with a single damage/failure-causing load event (of the 𝑖th
(𝑁 ) type) at time 𝑡. Similar to the concept of failure probability, it is in some
∏ 2 [ ]
𝔼 𝑝2 (𝑟2 , 𝑡2,𝑗 )𝑅𝑒,𝑆2 (𝑡2,𝑗 ) + (1 − 𝑝2 (𝑟2 , 𝑡2,𝑗 )) (26) occasions more convenient to use the term nonresilience, which equals
𝑗=1 1−resilience.
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ The proposed resilience model in Eq. (32) can also be applied in
Part 2
structural adaptive design and risk management, as is the case for struc-
in which 𝔼( ) denotes the mean value of the random variable in the tural reliability. On one hand, employing Eq. (32) can estimate the struc-
brackets. Modeling {𝑅𝑒,𝑆𝑖 (𝑡𝑖,1 ), 𝑅𝑒,𝑆𝑖 (𝑡𝑖,2 ), … 𝑅𝑒,𝑆𝑖 (𝑡𝑖,𝑁𝑖 )} as a statistically tural resilience level in the presence of changing load conditions, and
independent sequence for 𝑖 = 1, 2, using the law of total expectation, one guide the updated structural capacity accordingly. On the other hand,
has the resilience analysis based on Eq. (32) can be applied in the benefit-
{ [ 𝑁1 | ]} cost analysis in Eq. (24), if the Benefit of a particular strategy is modeled
∏ [ ]|
Part 1 in Eq. (26) = 𝔼 𝔼𝑝1 (𝑟1 , 𝑡1,𝑗 )𝑅𝑒,𝑆1 (𝑡1,𝑗 ) + (1 − 𝑝1 (𝑟1 , 𝑡1,𝑗 )) ||𝑁1 as a function of the structural resilience.
𝑗=1 |
|
⎧( )𝑁1 ⎫
5. Example
⎪ 𝑡𝑙
𝜆1 (𝑡) [ ] ⎪
= 𝔼⎨ ⋅ 𝑝 ( 𝑟 , 𝑡 ) 𝑅 𝑒,𝑆 ( 𝑡 ) + (1 − 𝑝 ( 𝑟 , 𝑡 )) 𝑑𝑡 ⎬
⎪ ∫0 ∫0 𝜆1 (𝑡)𝑑𝑡
𝑡𝑙 1 1 1 1 1
⎪
⎩ ⎭ In this section, an example will be presented to demonstrate the ap-
(27) plicability of the proposed formulas for structural time-dependent reli-
ability and resilience (Eqs. (22) and (32)).
in which 𝑅𝑒,𝑆1 (𝑡) = 𝔼(𝑅𝑒,𝑆1 (𝑡)). Substituting the PMF of 𝑁1 in
Eq. (17) into Eq. (27) gives, 5.1. Configuration
( )𝑘 ( )
𝑡 𝑡
∑∞ ∫0 𝑙 𝜆1 (𝑡)𝑑𝑡 exp − ∫0 𝑙 𝜆1 (𝑡)𝑑𝑡
Consider a bridge subjected to the joint impacts of live (traffic)
Part 1 in Eq. (26) = ⋅
𝑘=0
𝑘! load, dead load and hurricane winds, whose configuration is adopted
( )𝑘 from [51] with modifications. The probabilistic models of resistance and
𝑡𝑙
𝜆1 (𝑡) [ ]
⋅ 𝑝 ( 𝑟 , 𝑡 ) 𝑅𝑒,𝑆1 ( 𝑡 ) + (1 − 𝑝 ( 𝑟 , 𝑡 )) 𝑑𝑡 loads are summarized in Table 2, where 𝑅𝑛 , 𝐿𝑛 and 𝐷𝑛 are the nomi-
∫0 ∫ 𝑡𝑙 𝜆 (𝑡)𝑑𝑡 1 1 1 1
0 1 nal (or code-specified) resistance, live load and dead load respectively.
( )
𝑡𝑙 The combination of 𝐿𝑛 and 𝐷𝑛 satisfies the following design criterion
= exp − 𝜆 (𝑡)𝑑𝑡
∫0 1 for concrete structures in flexure [52],
( [ ] )𝑘
𝑡 0.9𝑅𝑛 = 1.25𝐷𝑛 + 1.75𝐿𝑛
∑∞ ∫0 𝑙 𝜆1 (𝑡) ⋅ 𝑝1 (𝑟1 , 𝑡)𝑅𝑒,𝑆1 (𝑡) + (1 − 𝑝1 (𝑟1 , 𝑡)) 𝑑𝑡 (33)

𝑘=0
𝑘! For simplicity, it is assumed that 𝐷𝑛 = 𝐿𝑛 in the following analyses. In
( 𝑡𝑙 𝑡𝑙 [ ] ) Table 2, both the mean value and the occurrence rate of live load are
= exp − 𝜆1 (𝑡)𝑑𝑡 + 𝜆1 (𝑡) ⋅ 𝑝1 (𝑟1 , 𝑡)𝑅𝑒,𝑆1 (𝑡) + (1 − 𝑝1 (𝑟1 , 𝑡)) 𝑑𝑡 functions of time 𝑡 to account for the nonstationarity in the load process,
∫0 ∫0
( 𝑡𝑙 [ ] ) featured by the two parameters 𝑐𝐿 and 𝑐𝜆,𝐿 , respectively. A special case
= exp − 𝜆1 (𝑡)𝑝1 (𝑟1 , 𝑡) 1 − 𝑅𝑒,𝑆1 (𝑡) 𝑑𝑡 (28) is that, if both 𝑐𝐿 and 𝑐𝜆,𝐿 equal 0, the live load process would be a
∫0
stationary one.
Similar to Eq. (28), The bridge is located in New York City, United States, and thus is
( 𝑡𝑙 [ ] ) subjected to the impact of hurricane winds. The probability model of
Part 2 in Eq. (26) = exp − 𝜆2 (𝑡)𝑝2 (𝑟2 , 𝑡) 1 − 𝑅𝑒,𝑆2 (𝑡) 𝑑𝑡 (29) wind load can be obtained through examining the historical data. Fig. 7
∫0
shows the historical tracks of hurricanes that struck New York City over
Thus, Eq. (26) becomes a period of 1902–2021 (an area with a radius of 40 km is considered),
( 𝑡𝑙 [ ] which are available from the US National Hurricane Center’s Database.
𝑅𝑠 (0, 𝑡𝑙 ) = exp − 𝜆1 (𝑡)𝑝1 (𝑟1 , 𝑡) 1 − 𝑅𝑒,𝑆1 (𝑡) 𝑑𝑡 There are totally 21 hurricane events within this period, yielding an
∫0
𝑡𝑙 [ ] ) occurrence rate of 0.175/year. The Weibull distribution (see Eq. (11))
− 𝜆2 (𝑡)𝑝2 (𝑟2 , 𝑡) 1 − 𝑅𝑒,𝑆2 (𝑡) 𝑑𝑡 (30) is employed to model the maximum wind speed associated with each
∫0 hurricane event (1-minute average sustained wind speed), with which
which further is rewritten as follows if taking into account the uncer- the scale and shape parameters are obtained as 28.67 m/s and 4.11,
tainties associated with 𝑅1 and 𝑅2 , respectively. In order to reflect the potential impacts of climate change,
∞ ∞ ( 𝑡𝑙 [ ] the scale parameter and the hurricane occurrence rate are described by
𝑅𝑠 (0, 𝑡𝑙 ) = exp − 𝜆1 (𝑡)𝑝1 (𝑟1 , 𝑡) 1 − 𝑅𝑒,𝑆1 (𝑡) 𝑑𝑡 time-variant functions as follows (the shape parameter is fixed as 4.11),
∫0 ∫0 ∫0
𝑡𝑙 [ ] )
𝑢(𝑡) = 28.67 + 𝑐𝑊 𝑡, 𝜆(𝑡) = 0.175 + 𝑐𝜆,𝑊 𝑡 (34)
− 𝜆2 (𝑡)𝑝2 (𝑟2 , 𝑡) 1 − 𝑅𝑒,𝑆2 (𝑡) 𝑑𝑡 ⋅ 𝑓𝑅1 ,𝑅2 (𝑟1 , 𝑟2 )𝑑𝑟1 𝑑𝑟2(31)
∫0
in which 𝑐𝑊 and 𝑐𝜆,𝑊 are two constants reflecting the changing rates of
where 𝑓𝑅1 ,𝑅2 (𝑟1 , 𝑟2 ) is the joint PDF of 𝑅1 and 𝑅2 as before. Compar- 𝑢(𝑡) and 𝜆(𝑡), respectively. Note that with a fixed shape parameter (𝛼(𝑡) in
ing Eqs. (20) and (31), it is observed that the time-dependent re- Eq. (11)), if 𝑢(𝑡) increases by 50% over 50 years, the mean wind speed,
silience is a generalized form of time-dependent reliability. In fact, which is proportional to 𝑢(𝑡), also increases by 50% over 50 years. It
if 𝑅𝑒,𝑆1 (𝑡) = 𝑅𝑒,𝑆2 (𝑡) ≡ 0, Eq. (31) reduces to Eq. (20), since 𝑝𝑖 (𝑟𝑖 , 𝑡) = is assumed that the fragility curve of the bridge has a median value of
1 − 𝐹𝑆𝑖 (𝑟𝑖 ⋅ 𝑔𝑖 (𝑡), 𝑡) for 𝑖 = 1, 2. 60 m/s and a logarithmic standard deviation of 0.15 (i.e., 𝑚𝑑 = 60m/s,
If there are 𝑛 types of hazards, the time-dependent resilience in and 𝜉𝑑 = 0.15 in Eqs. (5) and (6)) with respect to the 1-minute average
Eq. (31) can be further extended as follows, sustained wind speed.
( )
∞ ∞ 𝑛
∑ 𝑡𝑙 [ ] With the above configuration, if ignoring the aging effects, the struc-
𝑅𝑠 (0, 𝑡𝑙 ) = exp − 𝜆𝑖 (𝑡)𝑝𝑖 (𝑟𝑖 , 𝑡) 1 − 𝑅𝑒,𝑆𝑖 (𝑡) 𝑑𝑡 ⋅ 𝑓𝐑 (𝐫 )𝑑 𝐫 (32)
∫0 ∫0 𝑖=1
∫0 tural failure probability considering traffic (live) load only is 2.85 × 10−3

46
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Table 2
Statistical descriptions for structural resistance and loads.

Load Type Variable Mean COV∗ Distribution Occurrence rate

Dead Initial resistance 1.15𝑅𝑛 0.15 Lognormal /

& Dead load 1.05𝐷𝑛 0.05 Normal /
Live Live load (0.55 + 𝑐𝐿 𝑡) ⋅ 𝐿𝑛 0.40 Extreme Type I (1.0 + 𝑐𝜆,𝐿 𝑡)/year
Hurricane
load Initial resistance 60.68 m/s 0.151 Lognormal /
wind Maximum wind speed 0.908 ⋅ (28.67 + 𝑐𝑊 𝑡) m/s 0.274 Weibull distribution (0.175 + 𝑐𝜆,𝑊 𝑡)/year
∗
COV = Coefficient of variation.

Fig. 7. Historical tracks of hurricane from 1902 to 2021 that struck New York City (Reproduced from https://coast.noaa.gov/hurricanes).

for a service period of 50 years (corresponding to a reliability index of

2.76 in Eq. (1)), while that due to hurricane winds is 3.77 × 10−3 over
50 years (𝛽 = 2.67). In the following analyses, it is assumed that both
structural traffic load-bearing capacity and the wind-bearing general-
ized capacity (a lognormal variable with a median value of 60 m/s and
a logarithmic standard deviation of 0.15) degrade linearly by 15% over
50 years. In the presence of the two types of hazards, Eqs. (20) and
(31) are used to compute the structural reliability and resilience. The
following four cases will be considered for comparison purpose,
Case 1 The intensity and frequency of both loads (traffic and wind load)
are time-invariant (two stationary processes);
Case 2 The intensities of both loads increase by 50% over 50 years, with
fixed frequencies;
Case 3 The frequencies of both loads increase by 50% over 50 years,
with fixed intensities;
Case 4 The intensities and frequencies of both loads increase by 50%
over 50 years.
Note that these cases are representative of possible changing patterns
of loads in the future. Reliability and resilience analyses considering Fig. 8. Impact of resistance correlation on the time-dependent failure probabil-
these cases would inform the decision makers/asset owners “what are ities for reference periods up to 50 years.
the consequences” in the presence of different load scenarios. In practi-
cal engineering the prediction of future loads is often with uncertainties
(either aleatory or epistemic), rather than deterministic. In such a case,
the reliability and resilience methods proposed in this paper are to be correlation between the two resistances leads to a greater failure proba-
used in conjunction with the law of total probability to fully incorporate bility, due to the greater uncertainty associated with the “weaker” resis-
the randomness associated with the load model prediction. tance. The failure probability associated with hurricane wind is greater
than that of live load, which is consistent with the case of no aging ef-
5.2. Time-dependent reliability assessment fects. For a service period of 50 years, the failure probability (with tau
being 0.5) is underestimated by 57.41% and 36.69%, if considering a
Fig. 8 shows the time-dependent probabilities of failure for reference single load type of live load and wind load, respectively.
periods up to 50 years for case 1. The Gaussian copula is used with dif- The impact of copula type on failure probability is presented in Fig. 9
ferent values of Kendall’s tau for the two resistances. The failure proba- for case 1, where a Kendall’s tau of 0.5 is used. The COV of the two re-
bilities considering single load types are also plotted in Fig. 8. A smaller sistances is 0.15 in Fig. 9(a) and 0.25 in Fig. 9(b) (Note that in Eq. (6),

47
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

Fig. 9. Dependence of failure probability on copula function.

Fig. 10. Time-dependent failure probabilities for reference periods up to 50 Fig. 11. Dependence of time-dependent failure probability on the modification
years associated with four changing scenarios of load. factor 𝜁 .

√
with a small value of 𝜉𝑑 , the COV, which is preciously exp(𝜉𝑑2 ) − 1, can
be approximated by 𝜉𝑑 ). The failure probability associated with Frank benchmark failure probability is that associated with case 1. Assume that
copula is the largest, followed by those associated with Gumbel, Gaus- the strategy is to simultaneously increase the mean values of both resis-
sian, and Clayton copulas, respectively. The impact of copula selection tances with a factor of 𝜁 ≥ 1 (illustratively, if 𝜁 = 1.2, the mean values
is enhanced in the presence of a greater COV of the resistances. For a of both resistances become 1.2 times the original ones), the dependence
reference period of 50 years, the difference in Frank and Clayton-based of failure probability for a reference period of 50 years on 𝜁 is presented
failure probabilities is 7.6% in Fig. 9(a) and 16.4% in Fig. 9(b). in Fig. 11 for cases 2, 3 and 4. The Gaussian copula with Kendall’s tau
In Fig. 10, the difference in failure probability due to nonstationary being 0.5 is used for the two resistances. A greater value of 𝜁 indicates
load processes is examined. Assuming a Gaussian copula for the two stronger structural performance and thus a smaller failure probability.
resistances with a Kendall’s tau of 0.5, the failure probability associated In order to achieve the target (benchmark) safety level for the bridge
with case 4 is the largest, followed by those associated with cases 2, 3 (the failure probability equals 2.53 × 10−2 from case 1), 𝜁 is determined
and 1 (stationary load processes), respectively. The failure probability as 1.345, 1.02 and 1.38 for cases 2, 3 and 4 respectively. This infor-
is more sensitive to the change in load intensity than that in occurrence mation can be used in structural adaptive design, where “adaptation is
rate, as observed from the comparison between cases 2 and 3. triggered as changes occur” [28]. For example, in the presence of case
In the presence of increased failure probability due to nonstationary 4, the mean values of both structural resistances are to be increased by
loads, as revealed in Fig. 10, one would need to enhance the structural 38% so as to meet the minimum reliability requirement calibrated by
resistances against the nonstationarity (increasing trend) in loads, if the case 1 (no change).

48
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

( ) ( )
1 𝑎 1 𝑎
 (𝑎, 𝑧) = exp − −  (37)
𝑎 2𝑧 2𝑧 2𝑧

1
𝑅=1− 𝜃𝑎
(38)
2− 𝑏 𝑙 + 𝜃𝑎

( ) ( )
1 1 1
𝔼(𝓁𝑞 ) = exp  (39)
𝜇𝑋 𝜇𝑋 𝜇𝑋

and  is the exponential integral function, defined as (𝑎) =

∞
∫𝑎 exp(−𝑡)∕𝑡𝑑𝑡. Assume that 𝓁𝑞 increases from 0.4 at the initial time
to 0.8 by the end of 50 years parabolically. The lower bound of recov-
ery rate, 𝑏𝑙 , corresponds to the scenario of restoring 50% functionality
over 8 months, and the upper bound, 𝑏𝑢 , is twice 𝑏𝑙 .
Fig. 12 shows the time-dependent nonresiliences for reference pe-
riods up to 50 years for case 1 (with stationary traffic and hurricane
processes). The Gaussian copula is used for the two resistances with
a Kendall’s tau of 0, 0.5 and 0.8, respectively. Similar to the observa-
Fig. 12. Impact of resistance correlation on the time-dependent nonresiliences tions from Fig. 8, a stronger resistance correlation results in a smaller
for reference periods up to 50 years.
nonresilience. However, the impact of resistance correlation is weak-
ened on structural nonresilience compared with that on failure proba-
bility. In fact, one may alternatively interpret[ the resilience
] model in
5.3. Time-dependent resilience assessment
Eq. (31) as follows. Introducing 𝜆′1 (𝑡) = 𝜆1 (𝑡) 1 − 𝑅𝑒,𝑆1 (𝑡) and 𝜆′2 (𝑡) =
[ ]
Assume that the deterioration of structural functionality 𝑄(𝑡) is par- 𝜆2 (𝑡) 1 − 𝑅𝑒,𝑆2 (𝑡) , Eq. (31) becomes the same as Eq. (20), if replacing
allel to the deterioration of structural resistance. Conditional on the oc- 𝜆1 (𝑡) and 𝜆2 (𝑡) with 𝜆′1 (𝑡) and 𝜆′2 (𝑡) respectively in Eq. (20). The weakened
currence of a damage/failure causing-event at time 𝑡, the functionality impact of resistance correlation on nonresilience is due to the smaller
degrades by 𝓁𝑞 for both load types (i.e., 1 − 𝓁𝑞 equals the ratio of 𝑄𝑟 values of 𝜆′1 (𝑡) and 𝜆′2 (𝑡) compared with 𝜆1 (𝑡) and 𝜆2 (𝑡). Furthermore, if
in Fig. 6 to the degraded functionality 𝑄(𝑡)), and the recovery process considering a single load type only, the structural nonresilience is un-
is linear over time, with a rate varying within [𝑏𝑙 , 𝑏𝑢 ]. With this, the derestimated in the presence of multiple hazards. For example, for a
mean resilience associated with a single load event, 𝑅𝑒,𝑆𝑖 (𝑖 = 1, 2), can service period of 50 years, the nonresilience equals 9 × 10−3 with tau
be computed as follows [26], being 0.5, which is underestimated by 59.14% if considering live load
( ) [ ( )] only, or and 38.45% by a single load type of hurricane wind.
𝜃𝑎 1 1
𝑅𝑒,𝑆𝑖 =  exp ⋅ (1 − 𝑅) (1, 𝜇𝑋 (1 − 𝑅)) + 𝑅 exp − The impact of copula selection on time-dependent nonresilience is
𝑏𝑢 − 𝑏𝑙 𝜇𝑋 2𝜇𝑋 (1 − 𝑅)
shown in Fig. 13. The load scenario of case 1 is used, with a Kendall’s tau
(35) of 0.5 for the two structural resistances. The COV of resistances equals
0.15 in Fig. 13(a) and 0.25 in Fig. 13(b). Similar to the observations
in which 𝜃𝑎 is the rate of functionality deterioration (which equals
from Fig. 9, the Frank copula leads to the largest nonresilience, while the
0.15/50 = 0.003/year herein),
Clayton copula results in the smallest one. The difference between the
( ) ( )
𝜃𝑎 𝜃𝑎 nonresiliences associated with these two copulas is 4.1% in Fig. 13(a)
= ,𝜇 − ,𝜇 (36)
𝑏𝑢 + 𝜃𝑎 𝑋 𝑏𝑙 + 𝜃𝑎 𝑋 and 12.3% in Fig. 13(b).

Fig. 13. Dependence of nonresilience on copula function.

49
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

6. Concluding remarks

In this paper, new formulas have been derived for time-dependent

reliability and resilience analyses of aging structures exposed to mul-
tiple hazards. The structural resistances corresponding to the multiple
hazards are modeled as correlated variables, whose joint distribution
is constructed through copula functions. The nonstationarities in loads
– reflecting the potential impacts of changing environment – are also
incorporated in the proposed reliability and resilience models. The fol-
lowing conclusions can be drawn from this paper.
1. In the presence of multiple hazards, the structural resilience takes
a generalized form of structural reliability, as the former addi-
tionally considers the post-hazard recovery processes. If the struc-
ture is non-repairable (i.e., the resilience associated with a single
damage/failure-causing load event equals 0), the time-dependent re-
silience and reliability become mathematically identical.
2. Considering the impact of a single load type alone would unavoid-
ably underestimates the structural failure probability and nonre-
silience, if the structure is subjected to the joint impact of multi-
Fig. 14. Time-dependent nonresiliences for reference periods up to 50 years
associated with four changing scenarios of load. ple hazards. In the example considered in this paper, the failure
probability is underestimated by 57.41% and 36.69% (59.14% and
38.45% for nonresilience) for a service period of 50 years, if consid-
ering a single load type of live load and hurricane wind, respectively.
Furthermore, ignoring the correlation between structural resistances
overestimates structural failure probability and nonresilience.
3. The time-dependent reliability and resilience are dependent on the
selection of copula function for structural resistances. Of the four
copula types (Gaussian, Clayton, Frank and Gumbel), the structural
failure probability/nonresilience associated with Frank copula is the
largest, while that associated with Clayton copula is the smallest.
4. The proposed formulas for structural reliability and resilience can be
used in structural adaptive design and risk management. In order to
achieve a predefined level of structural reliability or resilience, the
structural resistances (in the presence of multiple hazards) are to be
enhanced to counter the impact of load nonstationarity in a changing
environment, where the modification factor is quantitatively deter-
mined through structural reliability/resilience assessment using the
proposed formulas.
Finally, note that the occurrences of different types of hazards have
been assumed to be statistically independent in the derivation of time-
dependent reliability and resilience in this paper; it is a promising topic
to examine the impact of correlation (interaction) between different haz-
Fig. 15. Dependence of time-dependent nonresilience on the modification fac-
ard types in future works.
tor 𝜁 .
Declaration of Competing Interest

In Fig. 14, the time-dependent nonresiliences associated with differ- The authors declare that they have no known competing financial
ent load conditions are presented for reference periods up to 50 years. interests or personal relationships that could have appeared to influence
The nonresilience associated with case 4 is the largest, since both the the work reported in this paper.
load intensities and occurrence rates increase by 50% over 50 years,
featuring the most severe changing scenario among the four cases. The Acknowledgements
nonresilience is more sensitive to the change in load intensity compared
with that in occurrence rate, with which case 2 leads to a greater non- The research described in this paper was supported by the Vice-
resilience compared with case 3. Chancellor’s Postdoctoral Research Fellowship from the University of
To account for the increased hazards due to load nonstationarity in Wollongong. This support is gratefully acknowledged.
a changing environment, as observed from Fig. 14, the structural resis-
References
tances are to be enhanced (with a modification factor of 𝜁 ≥ 1) so as to
achieve a predefined resilience level. The dependence of nonresilience [1] Ayyub BM. Systems resilience for multihazard environments: definition, metrics, and
for a reference period of 50 years on 𝜁 is shown in Fig. 15 for cases 2– valuation for decision making. Risk Anal 2014;34(2):340–55.
[2] Argyroudis SA, Mitoulis SA, Winter MG, Kaynia AM. Fragility of transport assets ex-
4. Treating the nonresilience associated with case 1 as the benchmark
posed to multiple hazards: state-of-the-art review toward infrastructural resilience.
level, 𝜁 is found to be 1.354, 1.021 and 1.392 for cases 2, 3 and 4. In Reliability Eng Syst Safety 2019;191:106567.
the context of adaptive design and risk management, if the future load [3] Wang C, Beer M, Ayyub BM. Time-dependent reliability of aging structures:
scenario is as in case 4, one would need to amplify the mean values of overview of assessment methods. ASCE-ASME J Risk Uncertain Eng Syst Part A
2021;7(4):03121003.
both resistances by 39.2% to counter the impact of load nonstationarity [4] Infrastructure Australia. Reforms to meet Australia’s future infrastructure needs:
on structural nonresilience. 2021 australian infrastructure plan; 2021.

50
C. Wang, B.M. Ayyub and A. Ahmed Resilient Cities and Structures 1 (2022) 40–51

[5] Wang C. Structural reliability and time-Dependent reliability. Switzerland AG: [28] Committee on Adaptation to a Changing Climate. Climate-Resilient infrastructure:
Springer Nature; 2021. adaptive design and risk management (MOP 140); 2018.
[6] Mori Y, Ellingwood BR. Reliability-based service-life assessment of aging concrete [29] Badroddin M, Chen Z. Lifetime resilience measurement of river-crossing bridges with
structures. J Struct Eng 1993;119(5):1600–21. scour countermeasures under multiple hazards. J Eng Mech 2021;147(9):04021058.
[7] Ellingwood BR, Mori Y. Reliability-based service life assessment of concrete struc- [30] Clifton JR, Knab LI. Service life of concrete. Tech. Rep.. Nuclear Regulatory Com-
tures in nuclear power plants: optimum inspection and repair. Nucl Eng Des mission, Washington, DC (USA) Div of Engineering; 1989.
1997;175(3):247–58. [31] Frangopol DM, Lin K-Y, Estes AC. Reliability of reinforced concrete girders under
[8] Akiyama M, Frangopol DM, Yoshida I. Time-dependent reliability analysis of exist- corrosion attack. J Struct Eng 1997;123(3):286.
ing rc structures in a marine environment using hazard associated with airborne [32] Vidal T, Castel A, François R. Analyzing crack width to predict corrosion in rein-
chlorides. Eng Struct 2010;32(11):3768–79. forced concrete. Cem Concr Res 2004;34(1):165–74.
[9] ISO. Bases for design of structures – Assessment of existing structures. ISO 13822: [33] Ellingwood BR. Risk-informed condition assessment of civil infrastructure: state of
2010(E); 2010. practice and research issues. Struct Infrastruct Eng 2005;1(1):7–18.
[10] Knutson TR, McBride JL, Chan J, Emanuel K, Holland G, Landsea C, et al. Tropical [34] Wang C. A stochastic process model for resistance deterioration of aging bridges.
cyclones and climate change. Nat Geosci 2010;3(3):157–63. Adv Bridge Eng 2020;1:3.
[11] Lin N, Emanuel K, Oppenheimer M, Vanmarcke E. Physically based assessment of [35] Baker JW. Introducing correlation among fragility functions for multiple compo-
hurricane surge threat under climate change. Nat Clim Chang 2012;2(6):462–7. nents. In: 14th World Conference on Earthquake Engineering, October 12–17, 2008,
[12] Mendelsohn R, Emanuel K, Chonabayashi S, Bakkensen L. The impact of climate Beijing, China; 2008.
change on global tropical cyclone damage. Nat Clim Chang 2012;2(3):205–9. [36] Ghosh J, Padgett JE. Aging considerations in the development of time-dependent
[13] Maljaars J, Steenbergen R, Abspoel L, Kolstein H. Safety assessment of existing high- seismic fragility curves. J Struct Eng 2010;136(12):1497–511.
way bridges and viaducts. Struct Eng Int 2012;22(1):112–20. [37] Choe D-E, Gardoni P, Rosowsky D. Fragility increment functions for deteriorating
[14] OBrien EJ, Bordallo-Ruiz A, Enright B. Lifetime maximum load effects on short-span reinforced concrete bridge columns. J Eng Mech 2010;136(8):969–78.
bridges subject to growing traffic volumes. Struct Saf 2014;50:113–22. [38] Song S, Qian Y, Liu J, Xie X, Wu G. Time-variant fragility analysis of the bridge
[15] IPCC. Climiate change 2021: the physical science basis. Contribution of Working system considering time-varying dependence among typical component seismic de-
Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate mands. Earthquake Eng Eng Vibration 2019;18(2):363–77.
Change Cambridge University Press; 2021. [39] Wang C, Zhang H. Assessing the seismic resilience of power grid systems considering
[16] Li Q, Wang C, Ellingwood BR. Time-dependent reliability of aging structures in the the component deterioration and correlation. ASCE-ASME J Risk Uncertain Eng Syst
presence of non-stationary loads and degradation. Struct Saf 2015;52:132–41. Part B Mech Eng 2020;6(2):020903.
[17] Wen Y. Minimum lifecycle cost design under multiple hazards. Reliability Eng Syst [40] Wang C, Teh LH. Time-dependent seismic reliability of aging structures. ASCE-ASME
Safety 2001;73(3):223–31. J Risk Uncertain Eng Syst Part A 2022;8(2):04022010.
[18] Yanweerasak T, Pansuk W, Akiyama M, Frangopol DM. Life-cycle reliability assess- [41] Schweizer B, Sklar A. Operations on distribution functions not derivable from oper-
ment of reinforced concrete bridges under multiple hazards. Struct Infrastruct Eng ations on random variables. Studia Mathematica 1974;52(1):43–52.
2018;14(7):1011–24. [42] Melchers RE, Beck AT. Structural reliability analysis and prediction. John Wiley &
[19] McAllister T. Developing guidelines and standards for disaster resilience of Sons; 2018.
the built environment: A Research needs assessment (NIST TN 1795). Na- [43] Spiegel MR. Correlation theory. In: Theory and Problems of Probability and Statis-
tional Institute of Standards and Technology (NIST), Gaithersburg, MD; 2013. tics. 2nd ed. New York: McGraw-Hill; 1992. p. 294–323.
https://doi.org/10.6028/NIST.TN.1795 [44] Bishara AJ, Hittner JB. Testing the significance of a correlation with nonnormal data:
[20] Bruneau M, Reinhorn A. Exploring the concept of seismic resilience for acute care comparison of Pearson, Spearman, transformation, and resampling approaches. Psy-
facilities. Earthquake Spectra 2007;23(1):41–62. chol Methods 2012;17(3):399.
[21] Bruneau M, Chang SE, Eguchi RT, Lee GC, O’Rourke TD, Reinhorn AM, Shi- [45] Nelsen RB. An introduction to copulas. Springer Science & Business Media; 2007.
nozuka M, Tierney K, Wallace WA, Von Winterfeldt D. A framework to quantita- [46] Katz RW. Stochastic modeling of hurricane damage. J Appl Meteorol
tively assess and enhance the seismic resilience of communities. Earthquake Spectra 2002;41(7):754–62.
2003;19(4):733–52. [47] Mudd L, Wang Y, Letchford C, Rosowsky D. Hurricane wind hazard assessment for
[22] Attoh-Okine NO, Cooper AT, Mensah SA. Formulation of resilience index of urban a rapidly warming climate scenario. J Wind Eng Ind Aerodyn 2014;133:242–9.
infrastructure using belief functions. IEEE Syst J 2009;3(2):147–53. [48] Li Q, Wang C, Zhang H. A probabilistic framework for hurricane damage assess-
[23] Cimellaro GP, Reinhorn AM, Bruneau M. Seismic resilience of a hospital system. ment considering non-stationarity and correlation in hurricane actions. Struct Saf
Struct Infrastruct Eng 2010;6(1–2):127–44. 2016;59:108–17.
[24] Singh RR, Bruneau M, Stavridis A, Sett K. Resilience deficit index for quantification [49] OBrien EJ, Schmidt F, Hajializadeh D, Zhou X-Y, Enright B, Caprani CC, et al. A
of resilience. Resilient Cities Struct 2022;1(2):1–9. review of probabilistic methods of assessment of load effects in bridges. Struct Saf
[25] Ayyub BM. Practical resilience metrics for planning, design, and decision making. 2015;53:44–56.
ASCE-ASME J Risk Uncertain Eng Syst Part A 2015;1(3):04015008. [50] Ayyub BM. Risk analysis in engineering and economics, second edition. Chapman
[26] Wang C, Ayyub BM. Time-dependent resilience of repairable structures subjected and Hall/CRC; 2014.
to nonstationary load and deterioration for analysis and design. ASCE-ASME J Risk [51] Wang C, Li Q, Zou A, Zhang L. A realistic resistance deterioration model for
Uncertain Eng Syst Part A 2022;8(3):04022021. time-dependent reliability analysis of aging bridges. J Zhejiang Univ Sci A
[27] Dong Y, Frangopol DM. Probabilistic time-dependent multihazard life-cycle assess- 2015;16(7):513–24.
ment and resilience of bridges considering climate change. J Perform Constr Facil [52] AASHTO. The manual for bridge evaluation. Washington DC: American Association
2016;30(5):04016034. of State highway and Transportation Officials; 2008.

51