Engineering Structures 305 (2024) 117759

Contents lists available at ScienceDirect

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

Time-variant seismic resilience of reinforced concrete buildings subjected to

spatiotemporal random deterioration
Zefan Chen a,b , De-Cheng Feng a,b ,∗, Xu-Yang Cao c , Gang Wu a,b
a
Key Laboratory of Concrete and Prestressed Concrete Structures of the Ministry of Education, Southeast University, Nanjing 211189, China
b
School of Civil Engineering, Southeast University, Nanjing 211189, China
c
College of Civil and Transportation Engineering, Hohai University, Nanjing 210024, China

ARTICLE INFO ABSTRACT

Keywords: Corrosion-induced deterioration weakens the seismic capacity of buildings in coastal areas. This paper presents
Time-variant resilience an investigation into the seismic resilience of RC frame structures experiencing spatiotemporal random
Corrosion effect deterioration. Two probabilistic distributions of corrosion were simulated using two modeling approaches:
Spatiotemporal random deterioration
uniform and spatial corrosion assumptions. The incremental dynamic analysis (IDA) approach was used to
Functionality quantification
obtain fragility curves for a 5-story RC frame designed according to Chinese codes. The impact of corrosion
Seismic loss
over a service life ranging from 0 to 50 years was assessed. The loss evaluation was subsequently conducted
using the method proposed by Federal Emergency Management Agency (FEMA). The functionality curves and
normalized resilience indices for different seismic hazard levels were estimated. The results indicate that the
seismic probabilistic capacity and resilience of the 5-story RC frame studied here decrease as service time
increases. The damage to structural members is more sensitive than that to nonstructural components. The
total losses of the RC frame investigated in this study mainly result from nonstructural repair losses at first,
then the collapse losses will become the main contributor to total loss as the seismic hazard level increases.
Both uniform and spatial corrosion modeling approaches are valuable for representing the impact of corrosion
on the restoration capabilities of RC frames.

1. Introduction state. According to Cimellaro et al. [2], seismic resilience can be de-
fined as a parameter linked to the probabilistic damage and functional
Over the past few decades, researches on the seismic capacity losses of structures. The seismic resilience index can be computed using
of structures have advanced significantly in various areas, including a mathematical formula they proposed [2]. A structural system with
fundamental theories, numerical simulations, and structural systems. 100% resilience exhibits an undamaged response or instant recovery ca-
Benefiting from this, the buildings designed according to current codes
pability following a seismic event, while 0% indicates complete collapse
offer a high level of seismic performance, ensuring the safety of occu-
and irreparable situations. Evaluating losses after a disaster is crucial in
pants during earthquake events. However, there are many evidences
quantifying the seismic resilience of structures. Nevertheless, the total
suggesting that the emphasis should shift from structural performance
to evaluating the influences of social consequences on the seismic loss of structures after disasters depends on multiple factors, including
behavior of structures. For instance, during the 2011 Christchurch structural characteristics, materials, societal response, and more. To
earthquake in New Zealand, although the number of collapsed build- address this issue, Federal Emergency Management Agency (FEMA) [3]
ings in the central business district was relatively low, the functionality developed a multi-hazard loss estimation methodology for various engi-
of community was compromised because essential services like water neering applications, such as general buildings, transportation systems,
and electricity were disrupted by structural damage [1]. Hence, the utility systems, and so on. The relevant repair and replacement costs
concept of seismic resilience has emerged as a new metric for describing are incorporated in a parameter called the damage ratio to quantify
the seismic capacity of structures from the perspective of social factors. the losses from both structural and nonstructural components under
Seismic resilience refers to the capacity of a system or community to
various limited states.
either maintain pre-earthquake functionality or recover to an improved

∗ Corresponding author at: Key Laboratory of Concrete and Prestressed Concrete Structures of the Ministry of Education, Southeast University, Nanjing 211189,
China.
E-mail address: [email protected] (D.-C. Feng).

https://doi.org/10.1016/j.engstruct.2024.117759
Received 20 September 2023; Received in revised form 22 February 2024; Accepted 25 February 2024
Available online 27 February 2024
0141-0296/© 2024 Elsevier Ltd. All rights reserved.
Z. Chen et al. Engineering Structures 305 (2024) 117759

Based on this loss evaluation approach, numerous researchers have

conducted resilience-based seismic assessments of both individual in-
frastructures and communities [4–12]. Dong and Frangopol [4] cal-
culated the resilience and residual functionality of both original and
base-isolated steel buildings, taking into account uncertainties in conse-
quence evaluation. Cao et al. [7] performed the seismic risk analysis for
RC frames retrofitted with external precast substructures and compared
the resilience indices of various retrofit strategies. Additionally, the
resilience assessment of infrastructure family has garnered the interest
of many researchers [13–17]. The interactive influence of various
infrastructures on functionality was taken into account [13,16]. One
of the significant challenges in resilience analysis is how to physi-
cally validate the accuracy of the results. This is because measurable
data on post-disaster structural losses are not available. To tackle this
challenge, a geographic information system-based model of a city in
Fig. 1. Schematic functionality curve of RC frame structures.
China was developed by Shang et al. [15]. This model integrates data
from previous earthquake events and previous literature, which can
be used to validate resilience analysis. Existing literature has made
outstanding endeavors in developing seismic resilience evaluation of were taken into account. The seismic fragility curves for the 5-story RC
infrastructures, however, these studies mainly quantify structural re- frame at various service times were obtained using the incremental dy-
silience based on deterministic scenarios. Although some investiga- namic analysis (IDA) method, which is preliminary for loss estimation.
tions consider uncertainties in ground motion records, they generally The loss assessment method introduced by FEMA 2003 was adopted
overlook the structural inherent variability. It is reported that uncer- to establish loss ratio fragility curves for the structures. The time-
tainties in structural geometry, material properties, and load effects dependent resilience of structures at different seismic hazard levels
have a substantial impact on structural load-bearing capacity [18]. was calculated to compare the restoration capabilities of structures.
The investigations on the seismic resilience of structures considering The key distinction between this study and previous literature lies in
various sources of uncertainty is crucial for the recovery ability and the quantification of the impact of the probabilistic behavior of the
rehabilitation of buildings after seismic disasters. corrosion effect on the seismic resilience of the RC buildings with
The functionality of structures is reduced by the degradation of different service times.
material properties, a phenomenon that is highly prevalent in coastal
regions. Buildings in aggressive environments are prone to the infil- 2. Seismic resilience assessment methodology
tration of chloride ions and carbonation of concrete. Both of these
occurrences lead to the degradation of material properties of both con- 2.1. Definition of seismic resilience
crete and reinforcement steel over time. Degradation encompasses not
only the individual properties of concrete and steel but also their inter- The concept of resilience was initially introduced by Holling in
action. Such as the compressive strength of concrete, yield and ultimate 1973 [31], which refers to the amount of disturbance that a system
strengths of reinforcement, ultimate strain of steel, and bond–slip effect can endure before undergoing a change in system control or structure.
between concrete and steel, etc [19,20]. The time-varying resilience Holling defined resilience as the ability to recover, particularly in eco-
assessments of structures have been extensively conducted [20–29]. logical systems. In civil engineering, resilience pertains to the capacity
Capacci and Biondini [23] quantified the annual failure probability of a structural system or family to recover from severe disasters such as
of aging bridges with a specific resilience target, taking into account earthquakes, floods, and tsunamis, etc. Seismic resilience of a structural
the deterioration worsened by climate change. Vishwanath and Baner- system describes its ability to regain functionality and performance
jee [21] quantified the seismic resilience of an aging bridge while following extreme hazards within a designated period known as the
accounting for the corrosion effect. Additionally, they incorporated control time 𝑇LC . In this study, seismic resilience was utilized as an
certain traffic-related random variables to estimate uncertainties in metric index to assess the impact of corrosion-induced deterioration
indirect losses. A modeling methodology to account for the coupled on the seismic performance of RC frames. The schematic functionality
interaction between corrosion and fatigue was proposed by Devendiran curve of a structural system is illustrated in Fig. 1, with 𝑡0E representing
and Banerjee [25]. The seismic fragility of a 3-story RC frame deteri- the time of earthquake occurrence. Once an earthquake event occurs,
orated by chloride-induced corrosion was quantified over the lifetime the functionality of structural systems will immediately decrease from
by Titi et al., considering the uncertainty of exposure condition [30]. the desired level (i.e., 100%) to 𝑄(𝑡0𝐸 ) due to external disturbances.
Zhou et al. developed a framework to gather seismic fragility data As repair measures are implemented, the functionality of systems will
and assess the recovery time for functionality in corroded multi-story gradually return to its original state [𝑄(𝑡) = 100%] or even exceed it
reinforced concrete (RC) buildings subjected to mainshock-aftershock [𝑄(𝑡) > 100%]. This recovery period is defined as the control time 𝑇LC .
sequences [26]. Although there have been some attempts to assess the According to Cimellaro et al. [2], the seismic resilience of a structure
seismic resilience of structures affected by the corrosion effect, the can be calculated using the following equation:
complexity and impact of the probabilistic nature of corrosion initiation 𝑡0E +𝑇LC
𝑄(𝑡)
and propagation remain inadequately elucidated. 𝑅= 𝑑𝑡 (1)
∫𝑡0E 𝑇LC
With an objective of quantifying the seismic resilience of RC frames
deteriorated by the corrosion effect from the uncertainty perspec- { }
tive, the present study investigates a 5-story RC frame structure with 𝑄(𝑡) = 1 − 𝐿(𝐼, 𝑇RE ) × 𝐻(𝑡 − 𝑡0E ) − 𝐻[𝑡 − (𝑡0E + 𝑇RE )] × 𝑓Rec (𝑡, 𝑡0E , 𝑇RE )
corrosion-induced deterioration. Service durations ranging from 0 to (2)
50 years, with intervals of 10 years, were chosen. The uniform cor-
rosion was employed to simulate the initiation and progression of
𝑓Rec = 𝑒𝑥𝑝[− ln(200)(𝑡 − 𝑡0E )∕𝑇RE ] (3)
corrosion. The spatial variability of corrosion was characterized using
the stochastic harmonic function (SHF) representation method. The where 𝐿(𝐼, 𝑇RE ) is the loss function, 𝐻(⋅) is the Heaviside step function,
uncertainties related to structural properties and the corrosion process 𝑇LC stands for the control time of the system, 𝑇RE signifies the recovery

2
Z. Chen et al. Engineering Structures 305 (2024) 117759

Table 1 yield/ultimate capacity, making them repairable. The typical recovery

Dimensionless damage ratios for different damage states of multi family dwelling
strategy for the structures experiencing a collapse damage state is
buildings [3].
the construction of a new building, leading to a higher percentage
Damage state Structural Drift-sensitive Acc.-sensitive
nonstructural nonstructural
of replacement cost in total loss. The loss estimation method for dif-
ferent states is identical, wherein the damage state probabilities are
Slight (𝜂𝑗=1 ) 0.3% 0.9% 0.8%
Moderate (𝜂𝑗=2 ) 1.4% 4.3% 4.3% converted into monetary loss equivalent using dimensionless loss ratios.
Extensive (𝜂𝑗=3 ) 6.9% 21.3% 13.1% Fragility analysis results for the collapse state are used to quantify
Collapse (𝜂𝑗=4 ) 13.8% 42.5% 43.7% 𝐿C|𝐼 . The other states represent the loss estimates for structures that
are not collapsed but repairable (𝐿R|𝐼 ). Three damage types of direct
losses were taken into account, including (1) structural losses, which
imply the costs of repairing or replacing of the damaged components
time from the extreme event, 𝑡0E denotes the time of occurrence of
of structures, (2) drift-sensitive non-structural losses, which encom-
extreme event, and 𝑓Rec (𝑡, 𝑡0E , 𝑇RE ) represents the recovery function.
pass the costs of repairing or replacing drift-sensitive nonstructural
The repair schemes are characterized by the recovery function in the
components in structures (like partitions, ornamentation, and glass),
resilience assessment. The repair rapidity depends on the societal re-
sponse and the available resources in the community. The exponential and (3) acceleration-sensitive nonstructural losses, which account for
recovery function proposed by Kafali and Grigoriu [32] was adopted the costs of repairing or replacing acceleration-sensitive nonstructural
here (i.e., Eq. (3)). This recovery function assumes that the repair components in structures (like ceilings, piping, and elevators) [3]. The
process occurs in a community with effective social response and ample dimensionless damage ratios for various damage types corresponding
resources. Seismic resilience index 𝑅, as per Eq. (1), is determined to different limit states are provided in Table 1.
graphically as the normalized shaded area under the functionality
curve. The shaded region enclosed by the functionality curve and the 2.3. Fragility function
desired functionality represents the loss of resilience, as depicted in
Fig. 1. The fragility function is essential for calculating the seismic re-
silience of structures, as indicated in Eq. (4). In this paper, the fragility
2.2. Loss function curves of structural damage, drift-sensitive nonstructural damage, and
acceleration-sensitive nonstructural damage were derived to evaluate
The loss function plays a pivotal role in the assessment of seismic the corresponding losses in different damage states. The spectral accel-
resilience. As depicted in Eq. (2), the loss function is defined as a func- eration of the first mode [𝑆𝑎 (𝑇1 )] was selected as the intensity measure
tion that is dependent on both earthquake intensity (𝐼) and recovery (IM) because it offers excellent predictions with reduced dispersion
time (𝑇RE ). Total losses comprise two categories: (1) direct losses (𝐿D ) and better relevance [33]. Various engineering demand parameters
and (2) indirect losses (𝐿I ) [2]. 𝐿D represents the losses that occur (EDPs) were employed for the three selected limit states to characterize
immediately upon the onset of earthquakes. 𝐿I denotes losses resulting the seismic response of structures. The interstory drift ratio (IDR)
from secondary disasters or business interruptions, and these losses was used as the EDP for structural and drift-sensitive nonstructural
exhibit significant time dependencies. In this study, the seismic loss damages, while the peak floor acceleration (PFA) was chosen for the
evaluation model provided by HAZUS [3] was utilized to estimate the acceleration-sensitive nonstructural damage.
total losses for a given earthquake intensity (𝐼): The widely used incremental dynamic analysis (IDA) approach was
𝑚 ∑
∑ 𝑛 utilized to establish the fragility function of structures. The fragility
𝐿T|𝐼 = 𝜂𝑖𝑗|𝐼 𝑃𝑖𝑗|𝐼 (4) function can be determined by fitting the IDA results using a lognormal
𝑖 𝑗
distribution, as shown below:
where 𝐿T|𝐼 is the dimensionless total loss ratio that is defined as
the ratio of total repair cost to replacement cost, 𝑚 and 𝑛 represent ⎡ ⎤
⎢ ln(𝑆𝑑 ) − ln(𝐿𝑆𝑖 ) ⎥
the numbers of loss types and damage states considered in seismic 𝑃 [𝐷 > 𝐿𝑆𝑖 |𝐼𝑀] = 𝛷 ⎢ √ ⎥ (8)
assessment, respectively, 𝜂𝑖𝑗|𝐼 stands for the dimensionless damage ⎢ 𝛽2 + 𝛽𝑐2 + 𝛽𝑚2 ⎥
⎣ 𝑑|𝐼𝑀 ⎦
ratios corresponding to different damage states, and 𝑃𝑖𝑗|𝐼 is the failure
probability of different damage states calculated by fragility analysis. where 𝐿𝑆𝑖 represents the limit states considered, 𝑆𝑑 and 𝛽𝑑|𝐼𝑀 are the
Note that 𝐿I is excluded from Eq. (4) due to its complexity and median value and variation of structural demand under a given IM,
uncertainty arising from social factors and data limitations. respectively, 𝛽𝑐 stands for the dispersion of the considered limit states
According to Xu et al. [5], 𝐿T can be further broken down into two and is set as 0.3 [34], 𝛽𝑚 is the uncertainty of modeling and is taken to
portions as follows: be 0.2 [35], and 𝛷 represents the cumulative normal distribution.
The probabilistic seismic demand model depicts the structural de-
𝐿T|𝐼 = 𝐿R|𝐼 + 𝐿C|𝐼 (5) mand characterized by the selected EDPs for a given IM. The corre-
sponding median values 𝑆𝑑 of EDPs can be calculated by the following
∑
𝑚 ∑
3
equation:
𝐿R|𝐼 = 𝜂𝑖𝑗|𝐼 𝑃𝑖𝑗|𝐼 (6)
𝑖 𝑗
𝑆𝑑 = 𝑎 (𝐼𝑀)𝑏 (9)
∑
𝑚
𝐿C|𝐼 = 𝜂𝑖,𝑗=4|𝐼 𝑃𝑖,𝑗=4|𝐼 (7) Then the logarithmic transformation is applied in Eq. (9), namely:
𝑖

in which, 𝐿R|𝐼 means the repair loss ratio due to the repair cost of ln(𝑆𝑑 ) = ln(𝑎) + 𝑏 ln(𝐼𝑀) (10)
the buildings, and 𝐿C|𝐼 is the collapse loss ratio which represents the
in which, 𝑎 and 𝑏 are the regression coefficients, respectively. Conse-
replacement cost of building collapse.
quently, the dispersion 𝛽𝑑|𝐼𝑀 can be calculated, namely:
A total of four damage states suggested by FEMA 2003 [3], i.e.,
√
slight, moderate, extensive, and collapse, were considered in this study. ∑𝑁 [ 𝑏 2
]
The collapse state describes the structures that have collapsed or is on 𝑖=1 ln(EDP𝑖 ) − ln(𝑎IM𝑖 )
𝛽𝑑|𝐼𝑀 = (11)
the verge of collapsing due to failure of critical components, making 𝑁 −2
them irreparable. The other three states refer to structures that cer- where 𝑁 is the number of nonlinear dynamic simulations and 𝑖 repre-
tain frame elements exhibit cracks of varying extents or reach their sents the 𝑖th simulation.

3
Z. Chen et al. Engineering Structures 305 (2024) 117759

3. Modeling of spatiotemporal random field of deterioration 3.2. Spatial variability of corrosion

The seismic resistance of structures in coastal areas is diminished To realistically simulate the spatial distribution of corrosion, Dizaj
due to corrosion, caused by the permeation of chloride ions and car- et al. [41] proposed a method that reflects the spatial corrosion effect
bonation of the concrete cover. The decrease in structural durability, on RC frame structures by generating numerous independent samples at
in turn, hampers the seismic resilience of the structures. In this section, all integration points within the numerical model. Despite this method
the deterioration modeling was carried out to quantify the probabilistic being simpler, it was not utilized in this study due to the lack of a
corrosion impact on the seismic resilience of RC structures. There solid theoretical foundation. Conversely, the random field approach
are two sorts of corrosion mechanisms, uniform corrosion and pitting developed by Stewart et al. [42–44] was employed to capture the
corrosion. Pitting corrosion typically occurs in environments dominated spatial variability in corrosion distribution.
by chloride. Uniform corrosion is mainly induced by a combination of The stochastic harmonic function (SHF) representation was em-
carbonation and low to moderate chloride levels. In this study, uniform ployed to determine the spatial random field of corrosion variabil-
corrosion was adopted since the due to the challenges in the elaborate ity [45]. This method is assumed to apply to stationary stochastic pro-
simulation of pitting corrosion. To overcome this limitation, the loss cesses, i.e., one-dimensional homogeneous random field. Consequently,
of steel ductility is calculated using an exponential reduction function the random field of 𝑌 (𝑥) can be determined as follows:
to partially simulate localized corrosion. This alternative approach has
∑
𝑁
( ) ( )
found widespread use in published literature [36–39]. A probabilistic 𝑌 (𝑥) = 𝐴 𝐾𝑗 𝑐𝑜𝑠 𝐾𝑗 𝑥 + 𝜙𝑗 (16)
model is necessary for the stochastic corrosion process, including ini- 𝑗=1
tiation and propagation, to account for variability in both time and in which, 𝐾𝑗 and 𝜙𝑗 represent the random wave number and random
space domains. Therefore, this paper employed a probabilistic model to phase angle, respectively, both obey the uniform distribution, 𝑥 is the
determine corrosion time and rate, and the random field method was position coordinate, 𝑁 to the number of components, 𝐴(𝐾𝑗 ) stands for
used to describe the spatial distribution of the corrosion impact. the amplitude that expressed as a function of random wave number,
i.e.,
3.1. Initiation and propagation of corrosion √ ( )( )
( ) 2𝑆𝑌 𝐾𝑗 𝐾𝑗 − 𝐾𝑗−1
The process of corrosion consists of three stages. Corrosion begins 𝐴 𝐾𝑗 = (17)
𝜋
with the ingression of chloride ions through the concrete cover, re-
sulting in the rusting of steel bars. Subsequently, the corrosion effect where 𝑆𝑌 (𝐾) stands for the power spectral density function of the
causes concrete to crack as the reinforcement steel expands. Eventually, variable 𝑌 , which can be calculated via the Fourier transformation,
the corrosion effect progresses over time. The time at which chloride- namely:
+∞
induced corrosion begins (𝑇0 ) is determined by many factors related
𝑆𝑌 (𝐾) = 𝑅𝑌 (𝑟) 𝑒𝑖𝐾𝑟 𝑑𝑟 (18)
to material properties and environmental conditions. DuraCrete [40] ∫−∞
offers a mathematical model that includes the uncertainty effect for √
calculating 𝑇0 , and this model is utilized in this study as follows: where 𝑅𝑌 (𝐾) is the autocorrelation function and 𝑖 = −1 is the unit of
imaginary number.
{ [ ( )]−2 } 1−𝑛
1
𝑑c2 𝐶s − 𝐶cr It is essential to accurately generate the random field for each
−1
𝑇0 = 𝑋 I erf (12) beam/column component. The SHF method offers the possibility of
4𝑘e 𝑘t 𝑘c 𝐷0 𝑡𝑛0 𝐶s
constructing a random field with a small number of variables. Chen
where 𝑋I is the model uncertainty factor, 𝑑c is the concrete cover thick- et al. [45] recommended that 8–10 components are appropriate for
ness, 𝑘t reflects the influence of test methods determining the empirical obtaining an accurate random field, requiring only 16–20 random
diffusion coefficient 𝐷0 , 𝑘e and 𝑘c are the parameters that account for variables. The Gaussian correlation function was employed, and a
the influences of environment and curing time, respectively, 𝑡0 = 28 day correlation length of 2.0 m was selected since the corrosion distribution
is the reference period for 𝐷0 , which is treated as deterministic, 𝑛 stands exhibited reduced sensitivity to correlation length within the range of
for aging factor, 𝐶s and 𝐶cr represent the surface and critical chloride 1.0 m to 3.0 m. The local average method was used to determine spatial
concentration, respectively, and erf −1 (⋅) is the Gaussian function error. parameters by assigning diverse material properties to each integration
Additionally, the actual rust extent of the reinforcement steels is re- point of the force-based element. The assigned values represent the av-
flected by the corrosion rate expressed by predicting the erosion depth. erage of the random field values over the integration weights associated
Due to the uniform corrosion was adopted, a time-dependent corrosion with the respective integration points.
rate was employed to evaluate the erosion depth of reinforcement bars,
specifically: 3.3. Corrosion-induced deterioration
( )0.29
𝑟cr (𝑡) = 0.85𝑟rc,0 𝑡 − 𝑇0 (13)
Corrosion gradually leads to the deterioration of material properties
where 𝑟cr,0 is the corrosion current density at the initiation of corrosion over time, resulting in a reduction in the load-bearing capacity of
propagation, which is calculated by the following equation: structures. Material deterioration is indicated by mass loss, which is
dependent on the erosion depth as expressed in Eq. (15). This section
37.5 (1 − 𝑤∕𝑐)−1.64 ( )
𝑟cr,0 = unit ∶ μA∕cm2 (14) presents the methods for considering corrosion-induced deterioration
𝑑c
of various materials.
where 𝑤∕𝑐 is the water to cement ratio.
Eventually, the erosion depth 𝑒cr can be calculated as an integral of 3.3.1. Deterioration of reinforcement steel
the corrosion rate as follows: The corrosion process of reinforcement steels includes three inter-
𝑇0
0.5254 (1 − 𝑤∕𝑐)−1.64 ( )0.71 related aspects: the loss of reinforcement sectional area, the reduction
𝑒cr (𝑡) = 0.0116 𝑟cr (𝑡) 𝑑𝑡 = 𝑡 − 𝑇0 (15) of reinforcement strength and ductility. When the corrosion of rein-
∫𝑡 𝑑c
forcement occurs (𝑡 > 𝑇0 ), the loss of reinforcement sectional area
where 0.0116 is the unit conversion factor from μA∕cm2 to can be calculated under a specific erosion depth 𝑒cr , the corresponding
mm/year. Hence, it is straightforward to calculate the stochastic loss time-dependent area function is expressed as follows:
of the reinforcement area and corrosion level at any given time 𝑡 using
𝜋[ ]2
Eqs. (12)–(15). 𝐴s (𝑡) = 𝑑 − 2𝑒cr (𝑡) (19)
4 b0

4
Z. Chen et al. Engineering Structures 305 (2024) 117759

in which, 𝑑b0 is the diameter of the uncorroded reinforcement bars. cracked concrete, which can be calculated by:
Easily, the mass loss 𝑄cr of reinforcement at time 𝑡 can be obtained as,
𝜀1 = (𝑏f − 𝑏0 )∕𝑏0 = 𝑛c 𝑤cr ∕𝑏0 (24)
𝐴s (𝑡 = 0) − 𝐴s (𝑡) where 𝑏0 stands for the section width in the state with no corrosion
𝑄cr = (20)
𝐴s (𝑡 = 0) cracks, 𝑏f is the beam width expanded by corrosion cracking. The
In terms of the stiffness and strength of reinforcement, the loss can increased beam width can be approximately estimated by the product
be determined as [46,47]: of the number of compressed bars 𝑛c and the average crack opening for
each bar 𝑤cr . The calculation model proposed by Molina et al. [55] was
( ) employed to evaluate 𝑤cr as follows:
𝐸s,cor (𝑡) = 1 − 100 × 𝛼E 𝑄cr 𝐸s0 (21a)
( ) ( )
𝑓y,cor (𝑡) = 1 − 100 × 𝛼y 𝑄cr 𝑓y0 (21b) 𝑤cr = 2𝜋 𝑣rs − 1 𝑒cr (𝑡) (25)
( )
𝑓u,cor (𝑡) = 1 − 100 × 𝛼u 𝑄cr 𝑓u0 (21c) where 𝑣vs is the ratio of volumetric expansion because of the corrosion
product, which is set as 2.0 [55]. Based on Eqs. (23)–(25), the extent of
where 𝐸s,cor (𝑡), 𝑓y,cor (𝑡), and 𝑓u,cor (𝑡) represent the elastic stiffness, yield
the reduction in compressive strength of the concrete cover depends on
and ultimate strengths of reinforcement steel at corrosion time 𝑡, re-
the corrosion depth. In other words, concrete cover spalling will hap-
spectively, 𝐸s0 , 𝑓y0 , and 𝑓u0 are the initial elastic stiffness, yielding, and
pens if the corrosion effect reaches a certain level. However, this paper
ultimate strengths of reinforcement, respectively, Note that 𝛼E , 𝛼y , and
𝛼u are taken to 0.0075, 0.005, and 0.005, respectively, with reference does not consider the acceleration of rebar corrosion after cracking or
to Du et al. [48]. spalling of the concrete cover.
Then, the loss of steel ductility can be described as an exponential Regarding confined concrete, the confined effect from the stirrups
reduction function related to the ultimate strain [39,49,50], namely: reduces with the increasing corrosion of the transverse bars, leading
{ to a decrease in strength and ductility. The Kent-Park model [56] is
𝜀u0 , 0 < 𝑄cr < 0.016
𝜀u,cor (𝑡) = (22) employed to calculate the properties of confined concrete, requiring in-
0.1521𝑄−0.4583
cr 𝜀u0 , 0.016 ⩽ 𝑄cr < 1.0
formation on the area, yield and ultimate strengths, and elastic modulus
where 𝜀u,cor (𝑡) and 𝜀u0 stand for the ultimate strain of corroded and of transverse steel. When modeling the confined concrete deteriorated
uncorroded reinforcement, respectively. by corrosion effect, during the calculation, the required parameters
Given the complexity and uncertainty of the corrosion effect, the are substituted with those of the corroded transverse bars, specifically
degradation of corroded steel is mainly determined by functions de- determined using Eqs. (19), (20), (21a), (21b), (21c), and (22).
rived from the regression analysis of experimental results. Interpreting
The constitutive relationship of corroded and uncorroded concrete
the corrosion mechanisms of both transverse and longitudinal rein-
is shown in Figs. 2b and c.
forcement bars simultaneously proves to be quite challenging. Conse-
quently, the impact of the corrosion sequence on both transverse and
longitudinal reinforcement is not considered for simplicity [22,30,51]. 3.3.3. Deterioration of bond–slip effect
Some existing studies imply that the buckling capacity of longitudinal Zhao and Sritharan [57] introduced a material model to simulate
reinforcement diminishes with the corrosion and the aging of the stir- the behavior of reinforcement fiber. In this model, the yield slip of
rups and the size effect has a great influence on the buckling capacity of steel (𝑠y ) is estimated by a linear regression analysis, which derives
reinforcement [52,53]. The reduction in the buckling capacity depends a function of the diameter of reinforcement bars (𝑑b ) and material
on the reinforcement details of the structures, i.e., the ratio of stirrup
properties of steel and concrete, i.e.:
spacing and bar diameter (𝐿∕𝐷) [52]. Nevertheless, the impact of the [ ]
corrosion effect on the buckling capability of reinforcement can be 𝑑b 𝑓y 1∕𝛼
disregarded when 𝐿∕𝐷 is below a certain value, which is discussed in
𝑠y = 2.54 √ (2𝛼 + 1) + 0.34 (26)
8437 𝑓c
Section 5.2.
The constitutive relationship of corroded and uncorroded steel is where 𝛼 is the parameter used in the local bond–slip relation, and is
shown in Fig. 2a. taken as 0.4 [57].
According to Zhao and Sritharan [57], the slip 𝑠u at the ultimate
3.3.2. Deterioration of concrete strength equal to 30 to 40 times 𝑠y would be appropriate, thus we adopt
The concrete corrosion typically manifests in both the concrete 𝑠u = 35𝑠y here.
cover and core concrete. The corrosion effects on the concrete cover are The constitutive relationship of corroded and uncorroded bond is
associated with the deterioration of reinforcement bar. Subsequently, shown in Fig. 2d.
additional splitting stresses will develop in the concrete surrounding
the corroded rebar, leading to cracking and, in extreme cases, spalling
of the concrete cover. The occurrence of cracking and spalling is 4. Uncertainty consideration
characterized by a reduction in the compressive strength of the concrete
cover fiber. The reduced compressive strength of concrete cover, taking 4.1. Selection of ground motions
the corrosion effect into account, can be calculated using the following
equation:
The quantity of ground motion records employed in IDA signifi-
𝑓c0 cantly influences the variability in structural response. The suitable
𝑓c,cor = ( ) (23)
1 + 𝐾 𝜀1 ∕𝜀c0 number of IDA, in order to thoroughly reflect the uncertainty charac-
where 𝑓c,cor and 𝑓c0 stands for the corroded and uncorroded com- teristics of earthquake, ranges from 10 to 20 [58]. Hence, a total of 22
pressive strength of concrete cover, respectively, 𝐾 represents the far-field records were selected from the PEER-NGA database to calcu-
coefficient determined by the diameter and roughness of reinforcement, late the fragility curves used in the resilience analysis of deteriorating
which is recommended to be set as 0.1 for medium-diameter ribbed RC frames. Additionally, these ground motions were adjusted to align
bars, 𝜀c0 is the strain corresponding to the uncorroded compressive with the design spectrum as specified by Chinese code [59]. The seismic
strength of concrete cover, and 𝜀1 is the average tensile strain in details of the selected ground motions are presented in Table 2.

5
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 2. Constitutive relationship of corroded and uncorroded material properties [54].

Table 2
Selected seismic ground motion records.
Number Name Year Magnitude Recording station PGA (g) PGV (cm/s)
1 Northridge 1994 6.7 Beverly Hills-Mulhol 0.52 63
2 Northridge 1994 6.7 Canyon Country-WLC 0.48 45
3 Duzce, Turkey 1999 7.1 Bolu 0.82 62
4 Hector Mine 1999 7.1 Hector 0.34 42
5 Imperial Valley 1979 6.5 Delta 0.35 33
6 Imperial Valley 1979 6.5 El Centro Array #11 0.38 42
7 Kobe, Japan 1995 6.9 Nishi-Akashi 0.51 37
8 Kobe, Japan 1995 6.9 Shin-Osaka 0.24 38
9 Kocaeli, Turkey 1999 7.5 Duzce 0.36 59
10 Kocaeli, Turkey 1999 7.5 Arcelik 0.22 40
11 Landers 1992 7.3 Yermo Fire Station 0.24 52
12 Landers 1992 7.3 Coolwater 0.42 42
13 Loma Prieta 1989 6.9 Capitola 0.53 35
14 Loma Prieta 1989 6.9 Gilroy Array #3 0.56 45
15 Manjil, Iran 1990 7.4 Abbar 0.51 54
16 Superstition Hills 1987 6.5 El Centro Imp. Co. 0.36 46
17 Superstition Hills 1987 6.5 Poe Road (temp) 0.45 36
18 Cape Mendocino 1992 7.0 Rio Dell Overpass 0.55 44
19 Chi-Chi, Taiwan 1999 7.6 CHY101 0.44 115
20 Chi-Chi, Taiwan 1999 7.6 TCU045 0.51 39
21 San Fernando 1971 6.6 LA - Hollywood Stor 0.21 19
22 Friuli, Italy 1976 6.5 Tolmezzo 0.35 31

4.2. Sampling of spatiotemporal random corrosion are (0.9, 17%) and (1.283, 45%), respectively. Then, eighteen parame-
ters are obtained by disaggregating random fields via the SHF method
The uncertainties affecting the structures located in coastal and ma- to consider the spatial corrosion uncertainty. Eventually, a total of 23
rine areas throughout their lifetime stem from various factors, including parameters were selected to reflect the variability from the corrosion
(1) inherent structural variability, such as structural geometry, material effect, namely, 5 and 18 parameters for corrosion processes and distri-
properties, and gravity loads, (2) the corrosion initiation and propaga- bution, respectively. The mean values and COVs of those parameters
tion processes, and (3) the spatial distribution of corrosion. Two types are listed in Tables 3 and 4. Moreover, a total of sixteen variables
of variables are defined in this paper, say, random variables and spatial that may affect the seismic performance of RC frame structures were
variables. The structural design parameters and the variables used to used to characterize the inherent structural variability [60,61]. Note
establish the probabilistic corrosion model for the corrosion rate at that the mean values of the structural random variables serve as the
different times (see Eq. (12)) are referred to random variables. As for building design parameters. The corresponding statistical information
the spatial variables, the surface and critical chloride concentrations will be presented in Section 5.1.1 (refer to Table 5). The optimized
(𝐶s and 𝐶cr ) are included and represented by a random field. The mean point selection approach based on GF-discrepancy was adopted to
values and coefficients of variation (COVs) corresponding to 𝐶s and 𝐶cr generate samples. This method has been proven to achieve effective

6
Z. Chen et al. Engineering Structures 305 (2024) 117759

Table 3 Table 5
Corrosion process random variables. Random variables for the 5-story RC frame.
Random variable Mean Coefficient Distribution Random variable Mean COV Distribution
of variation
Beam span 6000 mm 0.3% Normal
Modeling uncertainty factor 𝑋𝐼 1.0 5.0% Lognormal 1st story height 4500 mm 0.3% Normal
Environment factor 𝑘𝑔 0.676 17.0% Gamma Standard story height 3600 mm 0.3% Normal
Correction factor for test 𝑘𝑡 0.832 3.0% Normal Column depth 500 mm 1.0% Normal
Reference diffusion coefficient 𝐷0 473 10.0% Normal Beam depth 500 mm 1.0% Normal
Aging factor 𝑛 0.362 68.0% Beta Beam width 250 mm 1.0% Normal
Concrete cover 30 mm 1.0% Normal
D20 diameter 20 mm 4.0% Normal
Table 4 D18 diameter 18 mm 4.0% Normal
Corrosion distribution random variables. D12 diameter 12 mm 4.0% Normal
Random variable Lower bound Upper bound Distribution Concrete compressive strength 26.8 MPa 18.0% Lognormal
Concrete peak strain 0.002 15.0% Normal
𝑗th random wave number 𝑘𝑗 𝑘𝑗+1 Uniform Concrete tensile strength 2.68 MPa 18.0% Normal
𝑗th random phase angle 0 2𝜋 Uniform Rebar elastic modulus 200 000 MPa 3.3% Lognormal
Note: 𝑗 = 1, 2, … 9; 𝑘𝑗 = (𝑗 − 1) × 100 × 𝜋/72. Rebar yielding strength 450 MPa 9.3% Beta
Rebar ultimate strength 700 MPa 8.0% Lognormal
Rebar fracture strain 0.12 15.0% Normal
Dead load on the floor 5.0 kN/m2 10.0% Normal
Dead load on the roof 7.0 kN/m2 10.0% Beta
results with a small sample size for practical applications [62]. In this Live load on the floor and roof 2.0 kN/m2 40.0% Beta
paper, a sample size of 500 was selected. All variables are assumed to
be independent of each other because there is currently no available
method to accurately represent the relationships between different
random variables. The COVs of selected parameters and the distribution from transverse bars is considered using the Mander model [68]. To
they obeyed are referred to previous literature [18,63–66]. capture the bond–slip effect between concrete and reinforcement bars,
the Joint2D element is used in the beam-to-column connections. In the
Since the corrosion effect triggers the deterioration of material prop-
Joint2D element, the shear behavior of the joint panel is simulated
erties, the structural design parameters need to be updated at specific
by the shear spring assigned with the Pinching4 material. The shear
service time 𝑡. For a given service time 𝑡, the erosion depth 𝑒cr should be
spring of the Joint2D element is assigned with the Pinching4 material,
calculated firstly by substituting the parameters in Table 3 into Eq. (12),
whose backbone is calculated according to the modified compression
then the mass loss is determined through Eqs. (19) and (20). Finally,
field theory (MCFT). The bond–slip behavior between the beam and
the material properties are modified based on the calculated mass loss
column interfaces is modeled by the interface spring of the Joint2D
to reflect the corrosion-induced deterioration.
element. The unit length fiber section analysis is conducted to obtain
the tri-linear moment–rotation relation of the interface spring of the
5. Numerical example
Joint2D element. Note that the stress–strain relation for the steel fiber
must be replaced by the stress-slip relation. The numerical model
5.1. Case study
of the 5-story RC structure is shown in Fig. 4. The deterioration of
both concrete and reinforcement caused by the corrosion effect was
5.1.1. Profile of case study building attained by the method provided in Section 3.3. The feasibility and
A two-dimension 5-story RC frame structure designed following accuracy of the modeling strategy described above have been validated
Chinese codes [59,67] was used as the illustrative example in this by comparing the simulation results with experimental data. For further
study, as shown in Fig. 3. The first story has a height of 4.2 m, while details, please refer to Ref. [51].
the remaining stories are 3.6 m in height. The sectional dimension
of the beams and columns are 250 mm × 500 mm and 500 mm × 5.2. Time-variant fragility analysis
500 mm, respectively, the reinforcement details are depicted in Fig. 3.
The material properties for the concrete and steel are provided in The fragility analysis of the 5-story frame subjected to spatiotempo-
Table 5. The design dead loads in the floor and roof are 5.0 kN/m2 and ral random deterioration was performed based on the widely used IDA
7.0 kN/m2 , respectively, whereas the live loads are both 2.0 kN/m2 for method. During the IDA, the selected IM was incrementally increased
the floor and roof. This RC frame has a seismic design intensity of VII from 0, at intervals of 0.5, up to the value where structural collapse is
and a class of II. This implies that the peak ground acceleration with reached. This was done to estimate the probabilistic seismic demand of
a 10% probability of exceedance in 50 years is 0.10 g, where g ≈ 9.8 structures. It is important to note that time-variant fragility analysis was
m/s2 and the soil shear wave velocity 𝑣𝑠 ranges between 250 m/s and performed every 10 years throughout the 50-year designed service life
500 m/s. recommended by Chinese code [67]. The IDA curves, 16%, 50%, and
84% fractile curves for various damage types subjected to corrosion
5.1.2. Methodology of numerical modeling effect are shown in Fig. 5, where different colors indicate the curves
The open-source finite element software OpenSEES was used to concerning different service times.
perform the IDA of the 5-story RC frame structure. The seismic response Prior to the fragility analysis, logarithmic regression was performed
of RC frame structure can be effectively modeled by the fiber beam– for both uniform and spatial corrosion scenarios to establish the rep-
column elements. In this study, the force-based element was adopted to resentative relationship between 𝑆𝑎 (𝑇1 ) and the EDPs of IDR and PFA,
simulate the structural response of the beam and column components. as shown in Fig. 6. The 16%, 50%, and 84% fractile curves are also
Each beam/column member was modeled by one force-based element. given in Fig. 6. The scatters are the results obtained from the IDA,
A total of four integration points are assigned to each element. The the regression curves with various colors represent the different service
fiber section in each integration point was discretized into concrete and times. The regression outcomes imply that the dispersion between
steel fibers using different material models. The bilinear stress–strain 𝑆𝑎 (𝑇1 ) and IDR is similar to that between 𝑆𝑎 (𝑇1 ) and 𝑃 𝐹 𝐴. Hence, the
material model, Steel01, was used to model the behavior of reinforce- use of 𝑆𝑎 (𝑇1 ) as an IM to build a probabilistic seismic demand model
ment steel. The ConcreteD material model was adopted for concrete, for quantifying the influence of the corrosion effect on structures is well
developed based on plastic-damage mechanics. The confinement effect suited.

7
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 3. Prototype of the 5-story structure (unit: mm).

Fig. 4. Finite element model of the 5-story RC structure.

The corrosion effect diminishes the compressive buckling capacity The corrosion effect reduces the seismic fragility of structures for
of reinforcement bars [53,69,70]. Afsar Dizaj et al. proposed that this both structural and nonstructural damage types. In the case of uni-
detrimental effect be integrated into the limit states utilized in the form corrosion, the 𝑆𝑎,𝑃 =50% for structural damage is 0.8398, esti-
seismic fragility analysis of corroded structures [71,72]. However, the mated for a 50-year service life, representing a 20.25% reduction
original limit states recommended by FEMA 2003 [3] are still used in compared to the original structure. For nonstructural damages, the
this paper, as shown in Table 6. One of reasons is that reinforcement 𝑆𝑎,𝑃 =50% for drift-sensitive and acceleration-sensitive damage types at
bars with different effective lengths demonstrate varying degrees of 50-year are reduced to 1.0970 and 0.6028 from their original values
reduction in deformation capability. The impact of corrosion on the of 1.3258 and 0.7050, respectively. The largest reduction is observed
buckling capacity of bars is negligible when 𝐿∕𝐷 is less than equal to in structural damage (20.25%), followed by drift-sensitive nonstruc-
5 [52]. The column reinforcement details of the 5-story RC structure tural damage (17.56%), and acceleration-sensitive nonstructural dam-
used in this work satisfy this requirement (𝐿∕𝐷 = 5, see Fig. 3). Another age (14.50%). This pattern of variation is also evident in the dis-
reason is that structural total loss needs to be evaluated separately in tribution of the fragility curves in Fig. 7. Specifically, the fragility
terms of structural and nonstructural damage (see Table 1) to account
curves for acceleration-sensitive nonstructural damage exhibit a tighter
for monetary and social functional losses in resilience analysis [3].
distribution, whereas the curves for structural and drift-sensitive non-
The existence of multiple damage types poses challenges in precisely
structural damages are more dispersed. This occurs because structural
determining threshold values of limit states for corroded RC buildings.
damage stems from the deterioration of structural components where
Time-dependent fragility curves for both structural and nonstruc-
the corrosion effect is most pronounced.
tural damages are presented in Fig. 7, where the solid and dash lines
represent the uniform and spatial corrosion, respectively. The 𝑆𝑎 (𝑇1 ) The reduction in structural seismic capacity is minimal from 0 to
corresponding to failure probability 𝑃 = 50% (denoted as 𝑆𝑎,𝑃 =50% ) 10 years, then typically reaches its peak from 10 to 30 years, and
based on uniform and spatial corrosion are summarized in Tables 7 subsequently decreases again from 30 to 50 years. This phenomenon
and 8. The results demonstrate that the corrosion effect aggravates may be attributed to the use of an exponential reduction model for rein-
the seismic fragility of RC structures with the service time increases. forcement ductility. Compared to two corrosion modeling assumptions,
One possible reason for this observation is that the actual thresholds it can be found that the 𝑆𝑎,𝑃 =50% obtained from the spatial corrosion
of the limit states for the seismic fragility of corroded RC structures modeling are larger than those calculated based on the uniform cor-
are smaller than those of pristine structures. The use of original limit rosion assumption. However, the differences are so minor that they
states in the fragility analysis of corroded RC structures may lead to an can be considered negligible. This indicates that employing the uniform
overestimation of the structural resistance. Nevertheless, the general corrosion assumption in time-variant fragility analysis of structures is
trend of the fragility curves corresponding to the different corrosion slightly more conservative than using spatial corrosion modeling (see
levels in this paper is in line with Ref. [72]. Table 7).

8
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 5. IDA and fractile curves for different damage types.

Fig. 6. Relationship between EDP and IM for different damage types.

5.3. Time-variant resilience analysis including the frequent earthquake (FE), fortification earthquake (FoE),
and rare earthquake (RE). It should be noted that seismic hazard levels
5.3.1. Loss assessment are typically characterized by the peak ground acceleration (PGA).
Based on Eqs. (5)–(7), the seismic total loss 𝐿T of the 5-story RC However, 𝑆𝑎 (𝑇1 ) was adopted as IM in the fragility analysis herein. As a
frame is calculated by incorporating the corresponding fragility analysis result, 𝑆𝑎 (𝑇1 ) values for different seismic hazard levels were calculated
results, as depicted in Fig. 8. To discuss the influence of seismic levels based on site-specific data and the fundamental period of structures
on the expected losses of structures, a total of 3 seismic hazard levels (𝑇1 = 0.90 s for the 5-story RC frame in this study). Namely, 𝑆𝑎,FE =
suggested by the Chinese code [59] were considered in this study, 0.04 g, 𝑆𝑎,FoE = 0.11 g, and 𝑆𝑎,RE = 0.27 g, respectively. In Fig. 8, the

9
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 7. Fragility curves for different damage types.

Table 6 Table 8
Limit states for various damage types. 𝑆𝑎,𝑃 =50% at different service times for spatial corrosion scenario.
Damage type EDP Limit state Service time Slight Moderate Extensive Collapse
Slight Moderate Extensive Collapse (years) Structural damage
Structural IDR (%) 0.33 0.58 1.56 4.00 0 0.0653 0.1218 0.3645 1.0353
Drift-sensitive nonstructural IDR (%) 0.40 0.80 2.50 5.00 10 0.0648 0.1208 0.3603 1.0193
Acc.-sensitive nonstructural PFA (g) 0.25 0.50 1.00 2.00 20 0.0598 0.1120 0.3378 0.9658
30 0.0535 0.1015 0.3118 0.9068
40 0.0495 0.0948 0.2948 0.8693
Table 7 50 0.0463 0.0895 0.2838 0.8520
𝑆𝑎,𝑃 =50% at different service time for uniform corrosion scenario.
Drift-sensitive nonstructural damage
Service time Slight Moderate Extensive Collapse
0 0.0808 0.1740 0.6150 1.3258
(years) Structural damage 10 0.0803 0.1723 0.6065 1.3043
0 0.0653 0.1218 0.3645 1.0353 20 0.0740 0.1605 0.5718 1.2388
10 0.0613 0.1153 0.3500 1.0058 30 0.0668 0.1463 0.5323 1.1678
20 0.0553 0.1048 0.3225 0.9403 40 0.0618 0.1370 0.5068 1.1233
30 0.0498 0.0958 0.3005 0.8928 50 0.0580 0.1303 0.4923 1.1053
40 0.0460 0.0893 0.2853 0.8613 Acc.-sensitive nonstructural damage
50 0.0423 0.0830 0.2718 0.8398
0 0.0648 0.1438 0.3183 0.7050
Drift-sensitive nonstructural damage 10 0.0643 0.1433 0.3113 0.6760
0 0.0808 0.1740 0.6150 1.3258 20 0.0638 0.1385 0.3008 0.6530
10 0.0760 0.1655 0.5938 1.2918 30 0.0605 0.1328 0.2915 0.6403
20 0.0688 0.1510 0.5513 1.2118 40 0.0573 0.1268 0.2803 0.6198
30 0.0623 0.1388 0.5183 1.1555 50 0.0565 0.1250 0.2758 0.6085
40 0.0578 0.1303 0.4960 1.1193
50 0.0533 0.1220 0.4783 1.0970
Table 9
Acc.-sensitive nonstructural damage 𝑆𝑎,𝐿T =50% at different service times for different corrosion scenarios.
0 0.0648 0.1438 0.3183 0.7050
Service time (years) 0 10 20 30 40 50
10 0.0578 0.1308 0.2960 0.6698
20 0.0553 0.1258 0.2853 0.6475 Uniform corrosion 0.4978 0.4733 0.4455 0.4288 0.4128 0.3998
30 0.0550 0.1245 0.2820 0.6390 Spatial corrosion 0.4978 0.4863 0.4598 0.4373 0.4180 0.4083
40 0.0540 0.1220 0.2748 0.6193
50 0.0530 0.1190 0.2680 0.6028

results of fragility analysis, the maximum decreasing of 𝑆𝑎,𝐿T =50% also

occurs in the time range from 10 to 30 years, which is interpreted by
vertical black, blue, and red lines represent the seismic hazard levels the same reason.
of FE, FoE, and RE, respectively. The 50% fractile values of the total To determine the primary source of 𝐿T , the total loss ratio fragility
loss ratio fragility curves (𝑆𝑎,𝐿T =50% ) are highlighted by the horizontal curves are further decomposed into curves for structural and nonstruc-
grey line and summarized in Table 9. The loss ratio curves for service tural repair losses as well as collapse losses (see Fig. 8). As shown in
times ranging from 0 to 50 years are depicted in various colors. Fig. 9 Figs. 8 and 9, it is observed that, up to the RE hazard level, the primary
illustrates the specific total losses of the 5-story RC frame at different source of expected loss is the nonstructural repair losses, irrespective
seismic hazard levels. For each seismic hazard level, the total losses of service time and corrosion modeling approaches. When the hazard
from 0 to 50 years are disaggregated according to the damage types level reaches RE, nonstructural repair losses continue to constitute
considered here. the majority of 𝐿T until 𝑆𝑎 (𝑇1 ) reaches approximately 0.55 g, how-
Referring to the results given in Fig. 8, the total loss ratio fragility ever, the corresponding percentage gradually decreases. Concurrently,
curve tends to shift to the left over time for both uniform and spatial the collapse losses replace nonstructural repair losses as the primary
corrosion scenarios. This indicates that the corrosion effect leads to an contributor to 𝐿T and subsequently become the dominant factor. The
increase in the cost resulting from the earthquake events. The 𝑆𝑎,𝐿T =50% increasing trend in collapse losses from FE to RE hazard levels further
at 50 years of the 5-story RC frame are 0.3998 and 0.4083 based on uni- supports this observation (see Fig. 9). Interestingly, the impact of struc-
form and spatial corrosion assumptions, respectively, which decrease tural losses on 𝐿T is minimal across all hazard levels. This observation
19.69% and 17.98% compared to the original RC frame. Similar to the can be attributed to the fact that the damage ratios to structural repair

10
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 8. Loss ratio fragility curves for different corrosion scenarios.

Fig. 9. Loss disaggregation at different hazard levels.

costs are significantly smaller than those for nonstructural damages (see spatiotemporal random deterioration on structural seismic resilience is
Table 1). Furthermore, Fig. 9 indicates that the total loss of structures quantified in the following section. The functionality curves of the 5-
increases as the service time progresses. story RC frame considering corrosion-induced deterioration based on
FE, FoE, and RE hazard levels are presented in Fig. 10. When deriving
the functionality curves, the recovery time 𝑇RE is taken to 240 days
5.3.2. Resilience functionality assessment without considering delays in decision-making, financing, inspection,
Seismic resilience describes the recovery ability of a structure etc [3]. The calculated seismic resilience indices 𝑅 for each corrosion
to a desired functionality after an earthquake event. The impact of scenario are given in Tables 10 and 11.

11
Z. Chen et al. Engineering Structures 305 (2024) 117759

Fig. 10. Functionality curves for different hazard levels.

Table 10 following Chinese codes was investigated. Different sources of uncer-

𝑅 of different seismic hazard levels for different service times considering uniform tainties were incorporated, including structural properties, corrosion
corrosion.
process, and corrosion distribution. Both seismic fragility and resilience
Seismic hazard level Service time (years)
analyses of RC frames subjected to spatiotemporal random deteriora-
0 10 20 30 40 50 tion were conducted. The fragility curves at different service times
Frequent earthquake 0.9991 0.9988 0.9985 0.9983 0.9980 0.9977 were developed by IDA method. In resilience assessment, the loss
Fortification earthquake 0.9905 0.9890 0.9876 0.9866 0.9858 0.9847 evaluation method proposed by FEMA was adopted to quantify the total
Rare earthquake 0.9606 0.9569 0.9530 0.9506 0.9479 0.9456
loss of structures under varying limit states. The functionality curves
and normalized resilience index were finally calculated to discuss the
Table 11 impact of the corrosion effect on the restoration abilities of structures,
𝑅 of different seismic hazard levels for different service times considering spatial from which the following conclusions can be drawn:
corrosion.
Seismic hazard level Service time (years) 1. The most significant reduction in collapsed 𝑆𝑎,𝑃 =50% over 50
0 10 20 30 40 50 years, assuming uniform corrosion, is observed in the fragility
Frequent earthquake 0.9991 0.9990 0.9989 0.9986 0.9983 0.9981
analysis for structural damage (20.25%), followed by drift-
Fortification earthquake 0.9905 0.9904 0.9893 0.9879 0.9865 0.9859 sensitive nonstructural damage (17.56%) and acceleration-
Rare earthquake 0.9606 0.9596 0.9562 0.9527 0.9493 0.9475 sensitive nonstructural damage (14.50%). This demonstrates
that structural damage is more sensitive to the corrosion effect
due to the material deterioration is mainly concentrated on
structural components.
It is observed that 𝑅 of original RC frame (i.e., 𝑡 = 0 years) for 2. Both the seismic fragility and resilience of the 5-story RC frame
FE, FoE, and RE are 0.9991, 0.9905, and 0.9606, respectively. Once studied here drop with the service time going away, indicating
corrosion occurs, 𝑅 starts to decrease due to the deterioration of the corrosion effect has a great impact on the seismic capacity
material properties, however, the rate of reduction varies. Take the case and restoration abilities of structures. The seismic resilience of
of corrosion modeling for example, the computed 𝑅 corresponding to structures will decrease as the seismic hazard level increases, and
FE, FoE, and RE drop to 0.9977, 0.9847, and 0.9456, respectively. The the differences in seismic resilience of various service times will
reduced seismic resilience of the RC frame at a service time of 50 years be amplified as the seismic hazard level is enhanced.
demonstrates that corrosion-induced deterioration has diminished the 3. The primary contributor to the total loss of the 5-story RC frame
structural restoration capabilities. Evidently, structural restoration abil- studied in this research will change as the seismic hazard level
ities reduce with the service time growing up since the corrosion effect. rises. The majority of the structural total loss is nonstructural re-
However, the dropping rate will slowly decrease over an extended pair losses when the seismic intensity is lower than RE. However,
service period because the reduction in material properties may have when the seismic hazard level reaches RE, the relative share of
reached its full extent. collapse-related losses within the total loss gradually rises and
Referring to Tables 10 and 11, from 20 to 40 years, 𝑅 calcu- eventually becomes the predominant factor at 𝑆𝑎 (𝑇1 ) ≈ 0.55 g.
lated based on uniform corrosion modeling are slightly smaller than 4. The 𝑆𝑎,𝑃 =50% of different service times obtained from uniform
those values obtained from spatial corrosion assumption, however, corrosion are consistently smaller than those values obtained
the opposite trend is observed in the remaining periods. Nevertheless, from spatial corrosion. In seismic resilience assessment, there
the disparities between the results obtained from the two corrosion are minor differences in the resilience index 𝑅 calculated based
modeling assumptions are negligible. It is thus concluded that both the on uniform and spatial corrosion assumptions. This phenomenon
uniform and spatial corrosion are appropriate for estimating the seismic illustrates that both uniform and spatial corrosion modeling
resilience of RC frames based on corrosion scenarios. Moreover, the are viable methods to accurately reflect the influence of the
distinctions in resilience results will become more pronounced as the corrosion effect on the seismic response and recovery capacity
seismic hazard levels rise. of structures.

CRediT authorship contribution statement

6. Conclusions
Zefan Chen: Writing – original draft, Visualization, Investigation,
This paper explores the influence of the corrosion effect on the Formal analysis, Data curation. De-Cheng Feng: Writing – review &
time-variant seismic resilience of RC frame structures. The initiation, editing, Supervision, Methodology, Funding acquisition, Data curation,
propagation, and spatial random distribution of the corrosion were Conceptualization. Xu-Yang Cao: Writing – review & editing, Visual-
considered during the numerical simulation to capture the deteriorating ization, Supervision, Resources. Gang Wu: Writing – review & editing,
behavior of material properties. Then a 5-story RC frame designed Supervision, Methodology.

12
Z. Chen et al. Engineering Structures 305 (2024) 117759

Declaration of competing interest [22] Pang Y, Wei K, Yuan W. Life-cycle seismic resilience assessment of highway
bridges with fiber-reinforced concrete piers in the corrosive environment. Eng
Struct 2020;222:111120.
The authors declare that they have no known competing finan-
[23] Capacci L, Biondini F, et al. Resilience-based seismic risk assessment of ag-
cial interests or personal relationships that could have appeared to ing bridge networks under climate change. In: Bridge maintenance, safety,
influence the work reported in this paper. management, life-cycle sustainability and innovations-proceedings of the 10th in-
ternational conference on bridge maintenaince, safety and management, IABMAS
2020. CRC Press/Balkema; 2021, p. 2085–93.
Data availability
[24] Qian J, Zheng Y, Dong Y, Wu H, Guo H, Zhang J. Sustainability and re-
silience of steel–shape memory alloy reinforced concrete bridge under compound
Data will be made available on request. earthquakes and functional deterioration within entire life-cycle. Eng Struct
2022;271:114937.
[25] Devendiran DK, Banerjee S. Influence of combined corrosion–fatigue deteriora-
Acknowledgments tion on life-cycle resilience of RC bridges. J Bridge Eng 2023;28(5):04023014.
[26] Zhou Z, Han M, Dong Y, Yu X. Seismic resilience of corroded mid-rise re-
The authors would like to appreciate the financial supports from the inforced concrete structures under mainshock-aftershock sequences. Eng Struct
Project of National Key Research and Development Program of China 2023;288:116192.
[27] Yu X-H, Dai K-Y, Li Y-S. Variability in corrosion damage models and its effect
(Grant No. 2022YFC3803004), National Natural Science Foundation of on seismic collapse fragility of aging reinforced concrete frames. Constr Build
China (Grant No. 52311540017), and the Natural Science Foundation Mater 2021;295:123654.
of Jiangsu Province, China (Grant No. BK20211564). [28] Bojórquez J, Ponce S, Ruiz SE, Bojórquez E, Reyes-Salazar A, Barraza M,
Chavez R, Valenzuela F, Leyva H, Baca V. Structural reliability of rein-
forced concrete buildings under earthquakes and corrosion effects. Eng Struct
References 2021;237:112161.
[29] Zheng X-W, Li H-N, Gardoni P. Life-cycle probabilistic seismic risk assessment
[1] Kam WY, Pampanin S, Elwood K. Seismic performance of reinforced concrete of high-rise buildings considering carbonation induced deterioration. Eng Struct
buildings in the 22 february christchurch (lyttleton) earthquake. Bull N Z Soc 2021;231:111752.
Earthq Eng 2011;44(4):239–78. [30] Titi A, Bianchi S, Biondini F, Frangopol DM. Influence of the exposure scenario
[2] Cimellaro GP, Reinhorn AM, Bruneau M. Framework for analytical quantification and spatial correlation on the probabilistic life-cycle seismic performance of
of disaster resilience. Eng Struct 2010;32(11):3639–49. deteriorating RC frames. Struct Infrastruct Eng 2018;14(7):986–96.
[3] Multi–hazard loss estimation methodology: Earthquake model. Hazus–MH 2.1, [31] Holling CS. Resilience and stability of ecological systems. Annu Rev Ecol Evol
Technical Manual, Washington, DC: FEMA; 2003. Syst 1973;4(1):1–23.
[4] Dong Y, Frangopol DM. Performance-based seismic assessment of conventional [32] Kafali C, Grigoriu M. Rehabilitation decision analysis. In: ICOSSAr’05: proceed-
and base-isolated steel buildings including environmental impact and resilience. ings of the ninth international conference on structural safety and reliability. IOS
Earthq Eng Struct Dyn 2016;45(5):739–56. Press Amsterdam, The Netherlands; 2005.
[5] Xu J-G, Wu G, Feng D-C. Near fault ground motion effects on seismic resilience [33] Pang Y, Sun Y, Zhong J. Resilience-based performance and design of SMA/sliding
of frame structures damaged in wenchuan earthquake. Struct Infrastruct Eng bearing isolation system for highway bridges. Bull Earthq Eng 2021;19:6187–211.
2020;16(10):1347–63. [34] Wen Y, Ellingwood BR, Bracci JM. Vulnerability function framework for
[6] Caprili S, Panzera I, Salvatore W. Resilience-based methodologies for design of consequence-based engineering. MAE Center Report 04-04, University of Illinois;
steel structures equipped with dissipative devices. Eng Struct 2021;228:111539. 2004.
[7] Cao X-Y, Feng D-C, Wu G, Xu J-G. Probabilistic seismic performance assessment [35] Celik OC, Ellingwood BR. Seismic fragilities for non-ductile reinforced concrete
of RC frames retrofitted with external SC–PBSPC BRBF sub-structures. J Earthq frames–role of aleatoric and epistemic uncertainties. Struct Saf 2010;32(1):1–12.
Eng 2022;26(11):5775–98. [36] Choe D-E, Gardoni P, Rosowsky D, Haukaas T. Seismic fragility estimates for
[8] Shang Q, Wang T, Li J. A quantitative framework to evaluate the seismic reinforced concrete bridges subject to corrosion. Struct Saf 2009;31(4):275–83.
resilience of hospital systems. J Earthq Eng 2022;26(7):3364–88. [37] Vu NS, Yu B, Li B. Prediction of strength and drift capacity of corroded reinforced
[9] Khaghanpour-Shahrezaee R, Khanmohammadi M. A new methodology for esti- concrete columns. Constr Build Mater 2016;115:304–18.
mating seismic resilience of buildings under successive damage-retrofit processes [38] Deng P, Zhang C, Pei S, Jin Z. Modeling the impact of corrosion on seis-
during the recovery time. Struct Infrastruct Eng 2022;1–25. mic performance of multi-span simply-supported bridges. Constr Build Mater
[10] Castillo JGS, Bruneau M, Elhami-Khorasani N. Functionality measures for 2018;185:193–205.
[39] Biondini F, Vergani M. Deteriorating beam finite element for nonlinear analysis
quantification of building seismic resilience index. Eng Struct 2022;253:113800.
of concrete structures under corrosion. In: Design, assessment, monitoring and
[11] Gutiérrez J, Ayala AG, López-Ríos SE. Multidimensional functionality limit states
maintenance of bridges and infrastructure networks. Routledge; 2020, p. 107–20.
for seismic resilience analysis of urban buildings. Bull Earthq Eng 2023;1–24.
[40] DuraCrete. Statistical quantification of the variables in the limit state func-
[12] Wei B, Jia J, Bai Y, Du X, Guo B, Guo H. Seismic resilience assessment
tions. The European Union-Brite EuRam III-Contract BRPR-CT95-0132-Project
of bridges considering both maximum and residual displacements. Eng Struct
BE95-1347/R9, 2000.
2023;291:116420.
[41] Dizaj EA, Madandoust R, Kashani MM. Probabilistic seismic vulnerability analysis
[13] Yang Y, Ng ST, Zhou S, Xu FJ, Li H. A physics-based framework for analyzing
of corroded reinforced concrete frames including spatial variability of pitting
the resilience of interdependent civil infrastructure systems: A climatic extreme
corrosion. Soil Dyn Earthq Eng 2018;114:97–112.
event case in Hong Kong. Sustain Cities Soc 2019;47:101485.
[42] Stewart MG. Spatial variability of pitting corrosion and its influence on structural
[14] Xiong C, Huang J, Lu X. Framework for city-scale building seismic resilience
fragility and reliability of RC beams in flexure. Struct Saf 2004;26(4):453–70.
simulation and repair scheduling with labor constraints driven by time–history
[43] Stewart MG, Al-Harthy A. Pitting corrosion and structural reliability of corroding
analysis. Comput -Aided Civ Infrastruct Eng 2020;35(4):322–41.
RC structures: Experimental data and probabilistic analysis. Reliab Eng Syst Saf
[15] Shang Q, Guo X, Li Q, Xu Z, Xie L, Liu C, Li J, Wang T. A benchmark city for 2008;93(3):373–82.
seismic resilience assessment. Earthq Eng Eng Vib 2020;19:811–26. [44] Stewart MG. Mechanical behaviour of pitting corrosion of flexural and shear
[16] Xiao Y, Zhao X, Wu Y, Chen Z, Gong H, Zhu L, Liu Y. Seismic re- reinforcement and its effect on structural reliability of corroding RC beams. Struct
silience assessment of urban interdependent lifeline networks. Bull Earthq Eng Saf 2009;31(1):19–30.
2022;218:108164. [45] Chen J, He J, Ren X, Li J. Stochastic harmonic function representation of random
[17] Chen M, Mangalathu S, Jeon J-S. Seismic reliability assessment of bridge fields for material properties of structures. J Eng Mech 2018;144(7):04018049.
networks considering travel time and connectivity reliabilities. Earthq Eng Struct [46] Du Y, Clark L, Chan A. Effect of corrosion on ductility of reinforcing bars. Mag
Dyn 2022;51(13):3097–110. Concr Res 2005;57(7):407–19.
[18] Lin K, Chen Z, Li Y, Lu X. Uncertainty analysis on progressive collapse of [47] Cairns J, Plizzari GA, Du Y, Law DW, Franzoni C. Mechanical properties of
RC frame structures under dynamic column removal scenarios. J Build Eng corrosion-damaged reinforcement. ACI Mater J 2005;102(4):256.
2022;46:103811. [48] Du Y, Clark L, Chan A. Residual capacity of corroded reinforcing bars. Mag Concr
[19] Franceschini L, Vecchi F, Belletti B. The parc_Cl 2.1 crack model for NLFEA of Res 2005;57(3):135–47.
reinforced concrete elements subjected to corrosion deterioration. Corros Mater [49] Almusallam AA. Effect of degree of corrosion on the properties of reinforcing
Degrad 2021;2(3):474–92. steel bars. Constr Build Mater 2001;15(8):361–8.
[20] Michelini E, Belletti B, Franceschini L, Martinelli E. Time-dependent seismic [50] Apostolopoulos CA, Papadakis V. Consequences of steel corrosion on the ductility
fragility curves for existing RC core-wall buildings exposed to corrosion. Struct properties of reinforcement bar. Constr Build Mater 2008;22(12):2316–24.
Concr 2023;24(1):170–88. [51] Feng D-C, Xie S-C, Li Y, Jin L. Time–dependent reliability-based redundancy
[21] Vishwanath BS, Banerjee S. Life-cycle resilience of aging bridges under assessment of deteriorated RC structures against progressive collapse considering
earthquakes. J Bridge Eng 2019;24(11):04019106. corrosion effect. Struct Saf 2021;89:102061.

13
Z. Chen et al. Engineering Structures 305 (2024) 117759

[52] Kashani MM, Crewe AJ, Alexander NA. Nonlinear stress–strain behaviour of [63] Mirza SA, MacGregor JG. Variations in dimensions of reinforced concrete
corrosion-damaged reinforcing bars including inelastic buckling. Eng Struct members. J Struct Div 1979;105(4):751–66.
2013;48:417–29. [64] Mirza SA, MacGregor JG. Variability of mechanical properties of reinforcing bars.
[53] Kashani MM, Lowes LN, Crewe AJ, Alexander NA. Phenomenological hysteretic J Struct Div 1979;105(5):921–37.
model for corroded reinforcing bars including inelastic buckling and low-cycle [65] Ellingwood B, Ellingwood BR. Development of a probability based load cri-
fatigue degradation. Comp Struct 2015;156:58–71. terion for American National Standard A58: Building code requirements for
[54] Xu J-G, Wu G, Feng D-C, Cotsovos DM, Lu Y. Seismic fragility analysis of shear- minimum design loads in buildings and other structures. vol. 577, Department
critical concrete columns considering corrosion induced deterioration effects. Soil of Commerce, National Bureau of Standards; 1980.
Dyn Earthq Eng 2020;134:106165. [66] Chen Z, Zhu Y, Lu X, Lin K. A simplified method for quantifying the pro-
[55] Molino F, Alonso C, Andrade C. Cover cracking as a function of rebar corrosion: gressive collapse fragility of multi-story RC frames in China. Eng Fail Anal
II: Numerical model. Mater Struct 1993;26(163):532–48. 2023;143:106924.
[56] Kent DC, Park R. Flexural members with confined concrete. J Struct Div [67] Code for design of concrete structures. Beijing, China: China Architecture and
1971;97(7):1969–90. Building Press, Ministry of Housing and Urban-Rural Development of the People’s
[57] Zhao J, Sritharan S. Modeling of strain penetration effects in fiber-based analysis Republic of China (MOHURD); 2010, (GB50010–2010), (in Chinese).
of reinforced concrete structures. ACI Struct J 2007;104(2):133. [68] Mander JB, Priestley MJ, Park R. Theoretical stress-strain model for confined
[58] Shome N. Probabilistic seismic demand analysis of nonlinear structures. Stanford concrete. J Struct Eng 1988;114(8):1804–26.
University; 1999. [69] Kashani MM. Size effect on inelastic buckling behavior of accelerated pitted
[59] Code for Seismic Design of Buildings. Beijing, China: China Architecture and corroded bars in porous media. J Mater Civ Eng 2017;29(7):04017022.
Building Press, Ministry of Housing and Urban-Rural Development of the People’s [70] Dizaj EA, Padgett JE, Kashani MM. A Markov chain-based model for structural
Republic of China (MOHURD); 2010, (GB50011–2010), (in Chinese). vulnerability assessmentof corrosion-damaged reinforced concrete bridges. Philos
[60] Cao X-Y, Feng D-C, Wu G. Pushover-based probabilistic seismic capacity Trans R Soc 2021;379(2203):20200290.
assessment of rcfs retrofitted with PBSPC BRBF sub-structures. Eng Struct [71] Afsar Dizaj E, Madandoust R, Kashani MM. Exploring the impact of chloride-
2021;234:111919. induced corrosion on seismic damage limit states and residual capacity of
[61] Feng D-C, Wu G. Interpretable machine learning-based modeling approach reinforced concrete structures. Struct Infrastruct Eng 2018;14(6):714–29.
for fundamental properties of concrete structures (in Chinese). J Build Struct [72] Afsar Dizaj E, Salami MR, Kashani MM. Impact of asymmetrical corrosion of
2022;43(4):228–38. piers on seismic fragility of ageing irregular concrete bridges. Struct Infrastruct
[62] Chen J, Yang J, Li J. A GF-discrepancy for point selection in stochastic Eng 2023;1–20.
seismic response analysis of structures with uncertain parameters. Struct Saf
2016;59:20–31.

14