Engineering Structures 254 (2022) 113856

Contents lists available at ScienceDirect

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

Multi-objective optimization for probabilistic performance-based design of

buildings using FEMA P-58 methodology
Ali Ghasemof, Masoud Mirtaheri ∗, Reza Karami Mohammadi
Department of Civil Engineering, K.N. Toosi University of Technology, Tehran, Iran

ARTICLE INFO ABSTRACT

Keywords: This research presents an automated performance-based optimal design process by integrating a multi-objective
FEMA P-58 optimization algorithm and a comprehensive probabilistic performance assessment method. This process is
Probabilistic loss estimation used to provide a set of design options that comply with the minimum code requirements and are optimal in
Next-generation performance-based design
terms of both initial cost and mean annual repair cost. The FEMA P-58 methodology is utilized as a modern
Monte Carlo analysis
performance-based methodology to determine the various performance measures such as repair costs, repair
Time-based assessment
Multi-objective optimization
time, casualty rates, environmental impacts, and unsafe placarding for the design alternatives. To account
NSGA-II for uncertainties in the magnitude and intensity of future earthquakes, the time-based assessment, and to
account for inherent uncertainties affecting seismic performance, the Monte Carlo procedure is performed
in accordance with FEMA P-58 guideline. The optimization process is carried out using a well-known multi-
objective optimization algorithm called NSGA-II for three example office buildings with 3-, 6- and 9-story steel
moment frame structures. For each building, a detailed performance model including a set of fragility functions
for the structural and non-structural components as well as contents typically present in the building is
considered. Three optimal design options are selected from the Pareto front set obtained for each building based
on different optimization objectives. They are compared with each other and with the other design alternatives
in terms of various performance metrics under different ground motion intensities. The contribution of
structural components as well as drift-, acceleration- and velocity-sensitive non-structural components to the
repair costs of these optimal designs is investigated. It is shown that drift-sensitive nonstructural components
have the largest contribution to the repair costs of optimal structures. The contribution of these components in
earthquake events with a 29% probability of exceedance in 50 years (29%/50 yrs) is on average about 71%,
which decreases to 55% with increasing seismic intensity to 3%/50 yrs.

1. Introduction FEMA-227 [4], and ASCE 41 [5]). Currently, FEMA P-58 [6,7] is
recognized as a modern and comprehensive performance-based design
Earthquakes are one of the most catastrophic natural disasters that and evaluation methodology. In this method, earthquake losses are
cause great economic, social, and environmental losses in the world measured in terms of the probability of experiencing certain amounts
every year. The building codes’ requirements are usually provided to
of different performance metrics (known as decision variables (DVs))
maintain the life safety performance level and prevent the collapse of
such as casualties, repair costs, downtime, and environmental impacts.
the buildings under large earthquakes. However, a new building built to
the minimum building code criteria could still experience severe losses Accordingly, the mean annual occurrence rate of a DV (for example
(in some cases beyond the value of the building itself) that make it repair cost) is calculated through four steps of hazard analysis, struc-
unusable for a long time or in need of demolition [1]. Performance- tural analysis, damage analysis, and loss analysis using a Monte Carlo
based earthquake engineering (PBEE) procedures were introduced to procedure.
address this shortcoming. Performance-based analyses can help reduce The methodology presented in FEMA P-58 [6] has been the basis of
the likelihood of casualties, minimize uncertainty and downtime, and numerous research studies in recent years (see [8–14], for example).
reduce financial losses (direct and indirect) from future earthquakes.
In these studies, the FEMA P-58 method has been used to estimate
So far, various performance-based design and assessment methods
earthquake losses of one or a limited number of building archetypes
have been presented in the guidelines (e.g., Hazus [2], ATC-13 [3],

∗ Corresponding author.
E-mail addresses: [email protected] (A. Ghasemof), [email protected] (M. Mirtaheri), [email protected] (R. Karami Mohammadi).

https://doi.org/10.1016/j.engstruct.2022.113856
Received 2 June 2021; Received in revised form 10 December 2021; Accepted 4 January 2022
Available online 22 January 2022
0141-0296/© 2022 Elsevier Ltd. All rights reserved.
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

with predefined designs. Indeed, the time-consuming nature of assem- to compare between optimal and non-optimal designs. Moreover, in
bling a building performance model and the extensive computations terms of performance metrics, their study did not include the envi-
required in the FEMA P-58 methodology, often performed by PACT ronmental impacts and unsafe placarding caused by earthquakes, and
software [15], make it difficult to use this method to evaluate the in terms of multi-objective optimization, it included only a four-story
performance of a set of design alternatives (because the performance frame.
model data of each design must be entered separately in PACT). Al- A review of previous studies, as some of which were mentioned
though some researchers have tried to implement this approach with above, reveals that currently, the application of the FEMA P-58 in
some simplifications (e.g., [16]), the high computational effort to its multi-objective optimization problems has received less attention due
execution remains. to its complexity and time-consuming implementation. However, it is
On the other hand, decision-makers often look for an optimal de- known as a state-of-the-art loss assessment methodology, and its ap-
sign option from a range of optimal design alternatives. They usually plication in multi-objective optimization problems, which includes loss
choose the desired option by making a trade-off between the initial assessment of a large number of design options, can provide valuable
investment and the earthquake economic losses of different options. information about the various seismic consequences as well as details of
This is only possible by performing a multi-objective optimization vulnerable components. In this regard, the motivation of this research
process in combination with a seismic loss assessment method. Var- is the following objectives, which have not been considered in previous
ious optimization algorithms can be used for this purpose. These al- studies:
gorithms generally fall into two categories: deterministic (classical)
• A step-by-step and detailed description of how to implement
and stochastic (heuristic/metaheuristic) algorithms [17]. Classical al-
the FEMA P-58 performance assessment methodology within a
gorithms (e.g., optimality criteria), in general, converge faster than
multi-objective structural optimization problem.
metaheuristics. However, they need gradient information to drive the
• Comparison and evaluation of optimal and non-optimal feasi-
algorithm to the optimal solution. This limits the problems that can be
ble designs (i.e., those that meet the design constraints) ob-
solved with classical algorithms. In addition, these types of algorithms
tained during the optimization process in terms of various perfor-
may converge to local minima. In contrast, metaheuristic algorithms
mance metrics, including environmental consequences and unsafe
(e.g., genetic algorithms), which mimic stochastic natural phenomena,
placarding due to earthquakes.
usually converge to a global optimum, but they require a large number
• Investigation on the contribution of vulnerable components of
of objective function evaluations. Since they do not need gradient
optimal structures in their median repair cost at different seismic
information, they can solve complex problems by an exploration (in a
intensities.
stochastic manner) of a larger portion of the search space compared to
• Evaluation of the efficiency and accuracy of the simplified analy-
the case of gradient-based methods. Some well-known multi-objective
sis procedure presented in FEMA P-58 in predicting the values
metaheuristics are non-dominated sorting genetic algorithm II (NSGA-
of different performance metrics in comparison with nonlinear
II) [18], multi-objective particle swarm optimization (MOPSO) [19],
response history analysis.
and multi-objective evolutionary algorithm based on decomposition
(MOEA/D) [20]. Since the search space in structural optimization Accordingly, in this study, the FEMA P-58 performance assessment
problems is often nonlinear and includes several local optima, meta- process is employed within a multi-objective optimization problem,
heuristics are more suitable than classical algorithms for solving such and an automated performance-based design process is proposed. It
problems and have been used in most studies in this field, including results in a series of design options that are not only code-compliant
the present study. but are also optimal in terms of the initial cost and the mean annual
According to what was mentioned above, the high number of itera- repair cost of future earthquakes. This process is performed using
tions (i.e., design options evaluated) in the optimization process, along MATLAB parallel processing [39], which is linked to OpenSees [40]
with the complexity and time-consuming nature of the loss assessment software for structural analysis. Three office buildings with 3-, 6- and
using FEMA P-58, make it difficult to apply this methodology to opti- 9-story steel moment frame structures are considered to carry out the
mization problems despite its capabilities. Therefore, in optimization optimization process. The optimization is performed using the NSGA-
studies done to minimize earthquake losses, researchers have often II [18] optimization algorithm, which results in a Pareto optimal set for
used simple loss assessment approaches such as Hazus [2] ([21–24]) each archetype. Three design options are selected based on the different
or ATC-13 [3] ([9,25–27]). Moreover, some researchers have used decision-making strategies from the Pareto optimal set found for each
the engineering demand parameters (EDPs) related to seismic damage archetype. They are compared with each other, as well as with other
such as inter-story drift ratio (IDR) [28–32], floor velocity [31], floor feasible design alternatives, based on various performance measures
acceleration [29,32], Park-Ang damage index [33,34], or dissipated including repair cost, repair time, environmental impacts, injury rate,
energy [31,35,36]. Nowadays, thanks to the increasing speed of com- fatality rate, and unsafe placarding. Furthermore, the contribution of
puters and the possibility of parallel processing, these calculations can structural and nonstructural components in the median repair cost of
be done automatically and at high speed. In this regard, a limited archetypes are investigated for the three selected designs.
number of studies have recently used FEMA P-58 in multi-objective
optimization problems to advance some aspects of seismic design. 2. FEMA P-58 methodology overview [6]
For example, Ghasemof et al. [37] applied the FEMA P-58 procedure
to a multi-objective optimization problem to examine the effects of In the FEMA P-58 method, the building performance is measured
different demand parameters on the performance-based optimal design using several performance metrics such as repair costs, repair time,
of structures. However, their study did not include evaluating and fatalities, injuries, environmental impacts, and unsafe placarding due
comparing optimal and non-optimal feasible designs in terms of dif- to earthquake damages. To account for inherent uncertainties, these
ferent performance measures as well as examining the contribution of metrics are expressed as performance functions and in a probabilistic
vulnerable components in the amount of loss to the structure. Also, Mir- manner. Performance functions are statistical distributions that state
farhadi et al. [38] used the FEMA P-58 methodology in an optimization the probability that performance measures will occur at a specified
problem within the framework of a value-based design. They selected magnitude or lower magnitudes as a result of future earthquakes.
construction cost and seismic consequences as the value components. FEMA P-58 offers three types of performance assessments, including
However, their study focused more on comparing the cross-sectional intensity-based, scenario-based, and time-based assessments. In this
properties of optimal designs obtained from four different value-based study, time-based assessment is used to consider the uncertainty in
optimization scenarios than using the FEMA P-58 performance metrics the magnitude and location of future earthquakes, as well as the

2
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

intensity of motion resulting from these earthquakes [6]. Time-based structure, defined by a median value 𝑆̂𝑎 (𝑇 ), and a dispersion value 𝛽.
assessments evaluate the probable performance of a building over a In this study, SPO2IDA tool [41] is utilized to determine the collapse
given time period (for example, one year or 50 years), taking into fragility of the structure. This tool uses the pushover curve obtained
account all the earthquakes that can occur during that period and the from the nonlinear static analysis to estimate the results of incremental
occurrence probability associated with each one. In fact, it includes nonlinear dynamic analysis, which subsequently results in deriving the
several intensity-based assessments that are weighted based on the collapse fragility of the structure.
frequency of occurrence. In this study, evaluation is done based on a
one-year time period. Performance evaluation by FEMA P-58 method 2.5. Calculate performance
generally consists of 5 steps, which are mentioned below.
In the FEMA P-58 methodology, the performance of the build-
2.1. Assemble building performance model ing is calculated many times to investigate the effect of uncertainty
on the predicted outcome. Each repetition is called one realization
The first step in evaluating performance using the FEMA P-58 (the number of realizations denoted by 𝑛𝑅) that represents a possible
methodology is to create a building performance model. In the perfor- performance outcome for the specific magnitude or scenario of the
mance model, the vulnerable building components, including structural earthquake being analyzed. Using a Monte Carlo procedure, the results
and nonstructural components as well as contents, are first classified of the structural analysis are used to generate 𝑛𝑅 simulated demands.
into fragility groups, and a fragility function is defined for each group. The simulated demands represent the correlation between the different
Fragility groups are a set of fragile components that have the same response values estimated in the analysis as well as the inherent uncer-
fragility functions (e.g., partitions). The fragility function expresses the tainties in predicting the structural response. The simulated demand
types of component damage states, the structural demands that cause sets form a matrix [𝑚 × 𝑛𝑅], where each column in the matrix is one set
each damage state, and the consequences of each damage state in of simulated demands, and 𝑚 is the number of demand parameters. For
terms of risk to human life, repair methods, repair costs, repair time, each of the 𝑛𝑅 realizations, a unique set of possible performance out-
environmental impacts, and post-earthquake building occupancy status puts is determined. Accordingly, it is first examined for each realization
due to unsafe placarding. Fragility groups are then categorized into whether the building collapses or not. This is specified by comparing a
performance groups and quantities are defined for each performance randomly generated number (between 0 and 1) and the collapse proba-
group. Performance groups are subsets of a fragility group that will bility for the earthquake intensity of the realization (obtained from the
experience the same demands in response to an earthquake excitation collapse fragility function). If collapse occurs, the seismic consequences
(for example, all partitions on the same story in the same direction). (e.g., repair cost) are assumed equal to the replacement-related values,
The quantities of the performance groups are determined based on the building is assigned an unsafe placard, and the collapse casualties
normative quantities found in buildings of similar occupancy and size. are determined based on the building population model and collapse
Finally, a population model is defined for the building, which includes modes. If the building does not collapse, it is checked that whether
the distribution of people inside the building and the variability of this the building repair is practical or not. This is determined based on
distribution during the day as well as different days of the year. the maximum residual drift ratio and using the building repair fragility
function. If the building is irreparable, the seismic consequences are set
2.2. Define earthquake hazards equal to the replacement values, and the building is assigned an unsafe
placard. If the building does not collapse and is considered repairable
The second step in performance evaluation based on FEMA P- in the given realization, simulated demands are utilized to calculate
58 method is to define earthquake hazards. Earthquake hazards are the damage state for each performance group. Accordingly, the damage
defined in different ways depending on the assessment type and the states are determined based on the generation of random numbers, and
analysis type employed to determine the structural response. For time- finally, the corresponding consequences are calculated based on the
based assessments, earthquake hazards are described by a set of site- determined damage states.
specific seismic hazard curves. These hazard curves are utilized to In the time-based assessment, which is employed in this research,
obtain a set of acceleration response spectra representing a range steps 3 and 5 (Sections 2.3 and 2.5) should be iterated for a set of
of ground motion intensities. The analyses are performed using the ground motion intensities that may be experienced by the building
discrete spectral response accelerations obtained from the spectra at over a time period. For this purpose, the hazard curve obtained for the
each intensity, or earthquake records that are scaled for compatibility site from Section 2.2, which provides the mean annual frequency of
with these spectra at each intensity. exceedance of different earthquake intensities, is first divided into a
series of intervals (e.g., 8 intervals). Then, for each intensity interval
2.3. Analyze building response on the hazard curve, an intensity-based assessment is performed at the
mid-point of the interval.
Structural analysis is performed to estimate the building response to
an earthquake in the form of response values (demand vectors), which 3. Modeling and assumptions
can then be related to structural and nonstructural damage. Demands
usually include maximum values for story drift ratios in both directions, 3.1. Structural models
floor velocities and accelerations, and residual drift ratios. FEMA P-58
presents two analysis procedures for predicting structural responses, Three building archetypes with the moment-resisting frame (MRF)
including nonlinear response history and simplified analyses. system are considered as a case study to implement the proposed pro-
cess for the multi-objective optimal design based on the new generation
2.4. Develop collapse fragility of performance-based earthquake engineering (PBEE). Fig. 1 illustrates
the typical floor plans and elevation views of the archetypes. Bold lines
To evaluate casualties, it is necessary to define the probability show the position of the moment frames in the structure. According to
of collapse of the structure as a function of earthquake intensity as this figure, perimeter moment frames act as the lateral load resisting
well as possible collapse modes (e.g., partial or complete collapse) for system in all structures. The specifications of moment frames are sup-
the structure. These parameters are presented in the form of collapse posed to be the same in two perpendicular directions. Thus, only one
fragility functions. The probability of collapse is expressed as a lognor- moment frame of each archetype is studied, and corresponding analysis
mal distribution of spectral acceleration at the first mode period of the (design) results are expanded for the other frames. These frames are

3
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 1. Typical floor plans and elevation views of building archetypes.

intended as steel special moment frames (SMFs) and are designed in data are adjusted to 2020 dollars using the Historical Cost Index
accordance with the ASCE 7 [42] and AISC [43,44] standards. The provided by RSMeans [49]. According to the Project Costs presented
structural member groups for each archetype are also shown in Fig. 1. in RSMeans [49] for office buildings in the U.S., and by applying the
As shown, for 3- and 6-story buildings, the frame members are classified building size and time adjustment factors, the total building replace-
into nine groups, and for the 9-story building, they are classified into ment cost for all archetypes is estimated at $180/SF (in 2020 dollars).
15 groups. The building replacement time depends on several factors, including
The ASTM A992 steel is used for the beams and columns. Both the available financial resources, the time required for the complete
beams and columns are chosen from the seismically compact W- evacuation of the building, the efficiency of the contractors, the time
sections [45] which includes a list of 130 W-sections for beams (W16, required to obtain permits, and so on. Therefore, an accurate estimate
W18, W21, W24, W27, W30, W33, W36, W40, W44) and 54 W- of the building replacement time is not possible. In this study, a time-
sections (W8, W10, W12, W14) for columns. The seismic masses and cost model [50] is used to estimate the replacement time of buildings
gravity loads of archetypes are taken from Ref. [30]. To model 𝑃 − 𝛥
according to their replacement cost. Accordingly, the replacement time
effects originated from the interior gravity frames, a leaning column is
of 3-, 6- and 9-story buildings is estimated at 650, 750 and 810 days,
connected to the moment frame by rigid truss elements pinned at both
respectively.
ends. This column bears half of the total gravity loads carried by the
The maximum number of workers to perform building repair activ-
interior frames.
In this study, structural analysis is performed using OpenSees soft- ities is set to 1 worker per 1000 square feet of floor area (the default
ware [40]. The concentrated plasticity approach has been used to value provided by FEMA P-58 [7]). In the current study, the total loss
model the nonlinear behavior of structural members. In this approach, threshold is defined as 50% of the building replacement cost. Therefore,
beams and columns are modeled as lumped plasticity elements, con- when the building repair costs exceed 50% of the replacement cost,
sisting of elastic elements with concentrated plastic hinges at each the building is considered irreparable and would be demolished and
end. The elastic spring stiffness is considered 10 times larger than the replaced instead of repaired. The environmental impacts of replacing
rotational stiffness of the beam–column element [46]. The moment– the building, including carbon emissions and embodied energy, are
rotation behavior of rotational springs (plastic hinges) is modeled calculated by multiplying the total building replacement cost by the
using the modified Ibarrai–Medinai–Krawinkler (IMK) deterioration appropriate construction sector impacts from the United States Environ-
model with bilinear hysteretic response (Bilin Material in OpenSees). mentally Extended Input-Output (USEEIO) database [51]. According to
This model reliably captures the deterioration in stiffness and strength the data provided in this database for office buildings, the amount of
of structural elements. The backbone curve of this model includes carbon and energy impacts are 0.414 kg Co2e/$ and 5.729 MJ/$ (in
several parameters of strength and deformation as well as several 2013 dollars), respectively. As the inflation rate from 2013 to 2020 is
cyclic deterioration parameters, which are considered here according equal to 1.19 [49], the amount of carbon and energy impacts based
to the relationships recommended in Ref. [47] for beams and [48] for on 2020 dollars is estimated at 0.348 kg Co2e/$ and 4.81 MJ/$,
columns. In addition to the mentioned nonlinear model, a linear elastic respectively.
model is also made to perform linear and simplified analyses, where all
structural members are modeled by elastic beam–column elements.
3.2.2. Population model
3.2. Performance models The population model deemed for the buildings is presented in
Fig. 2. As shown in this figure, the expected peak population present
3.2.1. Basic information of buildings in the building archetypes is four persons per 1000 square feet (as
The buildings presented in Section 3.1 are considered to be of- recommended by FEMA [6] for office occupancy), which varies for
fice buildings located in Los Angeles. The basic information used in weekdays and weekends during the day. A dispersion of 0.2 is also
the building performance models is presented in Table 1. All cost considered for this value.

4
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 1
Basic information for building archetypes.
Number of stories 3 6 9
Floor area (sq. ft.) 64,800 135,000 202,500
Total replacement cost, RC ($) 11,664,000 24,300,000 36,450,000
Replacement time, RT (days) 650 750 810
Max workers per sq. ft. 0.001 0.001 0.001
Total loss threshold (% RC) 50 50 50
Carbon emissions replacement (kg) 4,063,462 8,465,547 12,698,320
Embodied energy replacement (MJ) 56,230,861 117,147,627 175,721,441

Fig. 3. Los Angeles, CA seismic hazard curves, amplified for soil type C.

In these cases, the fragility function depends on the cross-sectional

characteristics of the structural member, and therefore during the
optimization process, the assigned fragility function to the structural
member changes as its cross-section changes. Table 3 also provides the
Fig. 2. Variation in population present in 1000 sq. ft. of floor area by time of day for quantities of the performance groups in each floor and each direction.
commercial office occupancy. For groups whose fragility function is not directional, only one value is
entered in the quantity column. The quantities of structural components
are determined according to the floor plan of structures and group-
3.2.3. Fragility and performance groups ing of members. While the quantities of nonstructural components
The fragility functions used in the building performance models and contents are estimated using the Normative Quantity Estimation
are presented in Table 2. These are taken from the fragility database Tool [41]. This tool is an Excel spreadsheet that estimates the quantity
prepared by the PACT tool [15]. For some fragility functions that of nonstructural components as well as contents typically found in
require additional data from the user to complete (e.g., E2022.023, buildings of a given occupancy and size.
E2022.001), this information is inserted into the related functions. Ta-
ble 2 also indicates the associated demand parameter for each fragility 3.3. Site seismic hazard
function as well as the dependence of the fragility and the demand pa-
rameter on the direction. According to Table 2, the considered fragility It is assumed that the buildings are located in Los Angeles and
functions include a set of drift-sensitive structural components, drift- situated on soil type C. Seismic hazard curves for this site is shown in
and acceleration-sensitive nonstructural components, and acceleration- Fig. 3. These curves are derived from the U.S. Geological Survey (USGS)
and velocity-sensitive contents. The repair costs supplied in fragility tool [52]. Hazard curves include explicit consideration of ground mo-
functions have been adjusted for Northern California in 2011. The tion uncertainty, which are used to derive the hazard spectra (here,
region and date multipliers are used to modify the cost data for the uniform hazard spectra, UHS) at the selected shaking intensities for the
selected region and date. According to the City Cost Index provided by building. According to FEMA P-58 [6], the appropriate seismic intensity
RSMeans [49], the region cost multiplier for Los Angeles is defined as range for new buildings designed by the code is between 𝑆𝑎𝑚𝑖𝑛 and 𝑆𝑎𝑚𝑎𝑥 .
0.85, which is equal to the ratio of the City Cost Index for this city to the Where 𝑆𝑎𝑚𝑖𝑛 = 0.05𝑔 if 𝑇 ≤ 1, and 𝑆𝑎𝑚𝑖𝑛 = 0.05𝑔∕𝑇 if 𝑇 > 1; and 𝑆𝑎𝑚𝑎𝑥 is
average of the three most populous cities in Northern California. Also, defined as 𝑆𝑎 (𝑇 ) at a mean annual frequency of exceedance of 0.0002.
according to the Historical Cost Index [49], the date cost multiplier for In the time-based assessment, the derived hazard curve for the building
modifying cost data in 2020 instead of 2011 is defined as 1.25. There- is divided into eight intervals within this intensity range.
fore, all cost data presented in the fragility functions are multiplied by Structural analysis is conducted under the spectral response accel-
a factor of 1.25 × 0.85. erations obtained from the midpoint of each interval (in simplified
The fragility and performance groups of the archetypes and their analysis) or under a suite of earthquake records scaled to match the
corresponding quantities are provided in Table 3. Each row of the table spectra obtained from the midpoint of each interval (in nonlinear
identifies a fragility group that contains a specific fragility function, and response history analysis). According to FEMA P-58 [6], at least 7 pairs
can represent several performance groups defined in specific floors and of records should be used in the nonlinear response history analysis to
directions with specified quantities. As mentioned earlier, performance reasonably estimate the median response of the structure. It should be
groups are a subset of fragility groups that are organized by floor level noted that, since the MRFs specifications are supposed to be the same
and direction (Dir-1, Dir-2, Non-Dir). Table 3 defines more than one in both perpendicular directions, and the analysis is performed in two-
fragility function for some fragility groups (e.g., B1031.021a (b, c)). dimensional (2D), the structural response is evaluated only under one

5
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 2
Fragilities used in performance assessment.
Fragility ID Description Dir EDP
B1031.001 Bolted shear tab gravity connections Yes IDR
B1031.011a (b, c)a Steel column base plates Yes IDR
B1031.021a (b, c)a Welded column splices Yes IDR
B1035.021 (22)b Welded steel moment connection other than RBS, beam one side Yes IDR
B1035.031 (32)b Welded steel moment connection other than RBS, beams both sides Yes IDR
B2022.002 Curtain walls Yes IDR
B3011.011 Concrete tile roof No PFA
C1011.001a Interior wall partition Yes IDR
C2011.031b Stairs Yes IDR
C3011.002a Ceramic tile walls Yes IDR
C3021.001a Ceramic tile floors No PFA
C3027.002 Raised access floor No PFA
C3032.004a Suspended ceiling, Area < 250 No PFA
C3032.004b Suspended ceiling, 250 < Area < 1000 No PFA
C3032.004c Suspended ceiling, 1000 < Area < 2500 No PFA
C3034.002 Independent pendant lighting No PFA
D1014.011 Traction elevator No PFA
D1014.021 Hydraulic elevator No PFA
D2021.013a Cold water piping, dia ≤ 2.5 inches, Piping Fragility No PFA
D2021.013b Cold water piping, dia ≤ 2.5 inches, Bracing Fragility No PFA
D2021.023a Cold water piping, dia > 2.5 inches, Piping Fragility No PFA
D2021.023b Cold water piping, dia > 2.5 inches, Bracing Fragility No PFA
D2022.013a Hot water piping, dia ≤ 2.5 inches, Piping Fragility No PFA
D2022.013b Hot water piping, dia ≤ 2.5 inches, Bracing Fragility No PFA
D2022.023a Hot water piping, dia > 2.5 inches, Piping Fragility No PFA
D2022.023b Hot water piping, dia > 2.5 inches, Bracing Fragility No PFA
D2031.013b Sanitary waste piping No PFA
D3031.013c Chiller No PFA
D3031.023c Cooling tower No PFA
D3041.011c HVAC ducting - less than 6 sq. ft. in cross sectional area No PFA
D3041.012c HVAC ducting - 6 sq. ft. cross sectional area or greater No PFA
D3041.032c HVAC drops/diffusers No PFA
D3041.041b Variable air volume (VAV) box with in-line coil No PFA
D3052.013f Air handling unit No PFA
D4011.023a Fire sprinkler water piping No PFA
D4011.053a Fire sprinkler drop No PFA
D5011.013c Transformer/primary service No PFA
D5012.013d Motor control center No PFA
D5012.023c Low voltage switchgear No PFA
D5012.033c Distribution panel No PFA
E2022.001 Modular office work stations No PFA
E2022.023 Desktop computers No PFA
E2022.106a Bookcase No PFV
E2022.124a Lateral filing cabinet No PFV
a The fragility group is adopted based on the weight of the column.
b
The fragility group is adopted based on the depth of the beam.

of the two orthogonal horizontal components of ground motion. A suite 𝑖th element group; 𝐶𝐼𝑛 (𝑋) is the initial construction cost; 𝐶𝐸𝐴𝑅 (𝑋) is
of 22 far-field earthquake records presented in [53] is considered here. the expected (mean) annual repair cost of the building; 𝑔𝑗 (𝑋) represents
These records are taken from 14 events with magnitudes ranging from the design constraints, and 𝑝 is the number of constraints. In this study,
M6.5 to M7.6 and for site classes C and D (according to the NEHRP 𝐷𝑖 contains the seismically compact W-sections provided in the AISC
classification). The basic characteristics of these records are given in steel construction manual [45].
Table 4. It should be noted that only 15 records from this set that are
most compatible with the target spectra derived for each archetype over 4.2. Design constraints
the period range 𝑇𝑚𝑖𝑛 (0.2𝑇1 ) to 𝑇𝑚𝑎𝑥 (2𝑇1 ) are used here.
Due to constructional considerations in structures, the cross-section
4. Multi-objective structural optimization of column in the upper story should not be larger than the lower story.
This rule is defined as the column-to-column geometric constraint that
4.1. Defining the problem must be controlled in the joint of columns from two adjacent stories. In
addition to the geometric constraints, the design strengths of structural
The bi-objective problem presented for structural design optimiza- elements (i.e., columns and beams) must meet the AISC 360-16 [43]
tion is defined as below: requirements, which are defined as strength constraints.
Find 𝑋 = [𝑥1 , 𝑥2 , … , 𝑥𝑛𝑔 ]𝑇 , 𝑥𝑖 ∈ 𝐷𝑖 Since the structural system intended for the archetypes is the steel
to minimize 𝐹 (𝑋) = [𝐶𝐼𝑛 (𝑋), 𝐶𝐸𝐴𝑅 (𝑋)]𝑇 (1) special moment-resisting frame (SMF), the strong column–weak beam
(SCWB) criterion is defined as another constraint. Therefore, the flex-
subject to 𝑔𝑗 (𝑋) ≤ 0, 𝑗 = 1, 2, … , 𝑝
ural strength ratio of columns to beams connected to each joint must
where 𝑋 is the design variables vector, which includes the particular meet AISC 341-16 [44] criteria. Finally, a drift constraint is defined
cross-sectional properties of element groups of the structure; 𝑛𝑔 denotes to control the story drifts according to ASCE 7-16 [42] specifications.
the number of element groups at which all structural elements gathered Accordingly, the story drift ratios are limited to 0.025 for the 3-story
in the 𝑛𝑔 design groups; 𝐷𝑖 is the list of available cross-sections for the building and 0.02 for the 6- and 9-story structures [42].

6
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 3
Components quantities for archetypes.
Floors Fragility ID Unit Quantitya
3-story 6-story 9-story
Allb B1031.001 1 EA 0–0 0–4 0-4
All B1035.021 (22)c 1 EA 4–4 4–0 4-0
All B1035.031 (32)c 1 EA 6–6 8–8 8-8
All B2022.002 30 SF 86.4–129.6 112.5–112.5 112.5-112.5
All C1011.001a 100 LF 8.64–12.96 11.25–11.25 11.25-11.25
All C2011.031b 1 EA 0–2 2–0 2-0
All C3011.002a 100 LF 0.66–0.98 0.86–0.86 0.86-0.86
All C3021.001a 1 SF 907.2 945 945
All C3027.002 100 SF 162 168.75 168.75
All C3032.004a 250 SF 25.49 26.55 26.55
All C3032.004b 600 SF 8.1 8.44 8.44
All C3032.004c 1800 SF 5.4 5.63 5.63
All C3034.002 1 EA 324 338 338
All D2021.013a 1000 LF 0.91 0.95 0.95
All D2021.013b 1000 LF 0.91 0.95 0.95
All D2021.023a 1000 LF 0.32 0.34 0.34
All D2021.023b 1000 LF 0.32 0.34 0.34
All D2022.013a 1000 LF 1.92 2 2
All D2022.013b 1000 LF 1.92 2 2
All D2022.023a 1000 LF 0.76 0.79 0.79
All D2022.023b 1000 LF 0.76 0.79 0.79
All D2031.013b 1000 LF 1.23 1.28 1.28
All D3041.011c 1000 LF 1.62 1.69 1.69
All D3041.012c 1000 LF 0.43 0.45 0.45
All D3041.032c 10 EA 19 20 20
All D3041.041b 10 EA 15 16 16
All D4011.023a 1000 LF 4.32 4.5 4.5
All D4011.053a 100 EA 2 2 2
All D5012.023c 225 AP 3 3 3
All E2022.001 1 EA 151 158 158
All E2022.023 1 EA 48 50 50
All E2022.106a 1 EA 43 45 45
All E2022.124a 1 EA 17 18 18
1st B1031.011a (b, c)d 1 EA 10–10 12–12 12-12
1st D1014.011 1 EA 0 4 6
1st D1014.021 1 EA 2 0 0
1st D5011.013c 225 AP 6 14 20
1st D5012.013d 1 EA 3 5 8
1st D5012.033c 225 AP 3 5 8
–e B1031.021a (b, c)d 1 EA 10–10 12–12 12-12
Roof B3011.011 100 SF 58.32 60.75 60.75
Roof D3031.013c 75 TN 3 5 8
Roof D3031.023c 75 TN 3 5 8
Roof D3052.013f 8000 CFM 6 12 18
Roof D5012.023c 225 AP 3 3 3
a
For directional fragilities, quantities are given as Direction 1-Direction 2.
b
All floors except roof.
c
The fragility group is adopted based on the depth of the beam.
d The fragility group is adopted based on the weight of the column.

e 2nd, 3rd (3-story archetype); 3rd, 5th (6-story archetype); 3rd, 5th, 7th, and 9th (9-story archetype).

4.3. Analysis procedures 4.3.2. Supportive analysis (pushover)

The pushover analysis is performed according to ASCE 41-17 [5]
for obtaining the nonlinear response parameters of the structure. The
In this paper, four analysis procedures are performed to obtain
results obtained from the idealized pushover curve will be used to
the demand parameters of the structures depending on the different
determine the yield strength, the drift corresponding to the yield point,
stages of the work. These include linear static analysis according to
and also the collapse fragility of the structure, as described in the
the prescriptive code, nonlinear static analysis (pushover), simplified following. The gravity loads are considered here as 𝑄𝐺 = 𝐷 + 𝐿 to
analysis (linear static analysis according to FEMA P-58), and nonlinear combine with the seismic load actions [5]. While gravity loads are ap-
response history analysis. These procedures are presented briefly in the plied, the nonlinear model of the structure is subjected to incremental
following: lateral loads until the roof displacement reaches the target value. The
lateral load pattern is according to the fundamental mode shape of the
structure.
4.3.1. Code-based analysis (static linear)
As one purpose of performing pushover analysis here is to obtain
A linear static analysis is performed according to ASCE 7-16 [42] the collapse capacity of the structure, to reach the collapse region,
specifications to control the design constraints. To combine the effect the structure is pushed as much as possible. Therefore, the target
of gravity and earthquake loads, the load combination of 1.2𝐷 + 𝐿 + 𝐸 displacement here is equal to the maximum displacement that the
is utilized, in which 𝐷, 𝐿, and 𝐸 denote dead, live, and earthquake structure can experience before it collapses. The collapse is assumed
loads, respectively. Here, 𝐸 is determined based on the ASCE-7 design to occur when numerical instability arises in the analytical model, or
spectrum. when the maximum drift ratio of the structure reaches 10%, or at a

7
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 4
Characteristics of earthquake records used in nonlinear response history analysis [53].
No. M Year Event Station PGA(g)
1 6.7 1994 Northridge Beverly Hills-Mulhol 0.52
2 6.7 1994 Northridge Canyon Country-WLC 0.48
3 7.1 1999 Duzce, Turkey Bolu 0.82
4 7.1 1999 Hector Mine Hector 0.34
5 6.5 1979 Imperial Valley Delta 0.35
6 6.5 1979 Imperial Valley El Centro Array #11 0.38
7 6.9 1995 Kobe, Japan Nishi-Akashi 0.51
8 6.9 1995 Kobe, Japan Shin-Osaka 0.24
9 7.5 1999 Kocaeli, Turkey Duzce 0.36
10 7.5 1999 Kocaeli, Turkey Arcelik 0.22
11 7.3 1992 Landers Yermo Fire Station 0.24
12 7.3 1992 Landers Coolwater 0.42
13 6.9 1989 Loma Prieta Capitola 0.53
14 6.9 1989 Loma Prieta Gilroy Array #3 0.56
15 7.4 1990 Manjil, Iran Abbar 0.51
16 6.5 1987 Superstition Hills El Centro Imp. Co. 0.36
17 6.5 1987 Superstition Hills Poe Road (temp) 0.45
18 7 1992 Cape Mendocino Rio Dell Overpass 0.55
19 7.6 1999 Chi-Chi, Taiwan CHY101 0.44
20 7.6 1999 Chi-Chi, Taiwan TCU045 0.51
21 6.6 1971 San Fernando LA - Hollywood Stor 0.21
22 6.5 1976 Friuli, Italy Tolmezzo 0.35

displacement corresponding to a base shear of 0.4𝑉𝑦 , whichever occurs in the desired direction, for the selected level of ground shaking; 𝑊1 is
earlier. the first modal effective weight in the desired direction (not less than
80% of the total weight, 𝑊 ); 𝐹𝑥 represents the pseudo lateral force at
4.3.3. Analysis to performance assessment floor level of 𝑥; 𝑤𝑥 (or 𝑤𝑖 ) is the lumped weight at floor level 𝑥 (or 𝑖); ℎ𝑥
Two analysis approaches are presented in FEMA P-58 [6] for es- (or ℎ𝑖 ) stands for the height above the base level of the building to floor
timating structural responses and evaluating building performance. level 𝑥 (or 𝑖); 𝑁 is the number of floors in the building above the base;
These include nonlinear response history analysis and simplified analy- and 𝑘 is equal to 2 for the buildings with a first mode period greater
sis. The simplified procedure is employed to expedite the high number than 2.5 s, or equal to 1 for buildings with a first mode period less
of performance evaluations required to obtain structural response pa- than or equal to 0.5 s (linear interpolation can be used for intermediate
rameters (drifts, velocities, and accelerations) during the optimization periods).
process. However, nonlinear response history analyses are conducted to The uncorrected story drift ratios at each level, 𝛥𝑖 , obtained from
the performance assessment of selected design alternatives at the end the analysis are then converted into a median estimate of the nonlinear
of optimization process. story drift ratios, 𝛥∗𝑖 , using the following equation:
Nonlinear response history analysis determines the response of a ( )
𝛥∗𝑖 = 𝐻𝛥𝑖 𝑆, 𝑇1 , ℎ𝑖 , 𝐻 × 𝛥𝑖 , for 𝑖 = 1 to 𝑁 (4)
nonlinear structural model to a set of earthquake records that have been
selected and scaled to be consistent with the target spectrum. According where 𝐻𝛥𝑖 is the drift correction factor for story 𝑖, which is a function of
to FEMA P-58 [6], at least 7 pairs of ground motion records should be parameters 𝑆, 𝑇1 , ℎ𝑖 , and 𝐻; 𝑆 is the strength ratio, which is calculated
used to a reasonable estimate of the median response of the building. using Eq. (5), and the other parameters are already defined.
Records are selected to match reasonably with the target spectrum over 𝑆𝑎 (𝑇1 )𝑊
the period range of 𝑇𝑚𝑖𝑛 (0.2𝑇1 ) to 𝑇𝑚𝑎𝑥 (2𝑇1 ). Where 𝑇1 is the funda- 𝑆= (5)
𝑉𝑦1
mental period of the building in the direction under consideration. In
nonlinear response history analysis, similar to pushover analysis, the in which 𝑉𝑦1 is the estimated yield strength of the building in the first
gravity load effects are considered as 𝑄𝐺 = 𝐷 + 𝐿 to combine with the mode response, which is determined by the pushover analysis.
seismic load effects. To compute median estimates of peak floor accelerations and peak
In the simplified procedure, the median values of the demand floor velocities in the simplified analysis, their values at the base of the
parameters are determined using the estimated lateral yield strength building are taken as equal to the peak ground acceleration (PGA) and
of the structure and performing static analysis on the linear model. In the peak ground velocity (PGV), respectively. At other floor levels, 𝑖, the
this method, similar to the code-based analysis, the earthquake forces median estimates of peak floor acceleration (𝑎∗𝑖 ) and peak floor velocity
are applied to the structure in the form of an equivalent lateral load, (𝑣∗𝑖 ) relative to a fixed point in space (total values, not relatives), are
and a static linear analysis is carried out. The same load combination calculated as follows:
considered in the code-based analysis (1.2𝐷 + 𝐿 + 𝐸) is used here. ( )
𝑎∗𝑖 = 𝐻𝑎𝑖 𝑆, 𝑇1 , ℎ𝑖 , 𝐻 × 𝑃 𝐺𝐴, for 𝑖 = 2 to 𝑁 + 1 (6)
However, in the simplified analysis, the seismic load (𝐸) is determined
as a pseudo lateral force, 𝑉 , from Eq. (2) and then distributed using ( )
𝑣∗𝑖 = 𝐻𝑣𝑖 𝑆, 𝑇1 , ℎ𝑖 , 𝐻 × 𝑣𝑠𝑖 , for 𝑖 = 2 to 𝑁 + 1 (7)
Eq. (3) at the height of the building.
where 𝐻𝑎𝑖 and 𝐻𝑣𝑖 are the acceleration and velocity correction factors
𝑉 = 𝐶1 𝐶2 𝑆𝑎 (𝑇1 )𝑊1 (2)
for floor 𝑖, respectively; and 𝑣𝑠𝑖 is the reference floor velocity, which is
computed as follows:
𝑤𝑥 ℎ𝑘𝑥
𝐹𝑥 = 𝑉 (3) 𝑇1 𝑉𝑦1 𝛿
∑
𝑁+1 𝑣𝑠𝑖 = 𝑃 𝐺𝑉 + 0.3 ( 𝛤 ) 𝑖 (8)
𝑤𝑖 ℎ𝑘𝑖 2𝜋 𝑊1 ∕𝑔 1 𝛿𝑟
𝑖=2 in which 𝛤1 is the first mode participation factor; 𝛿𝑖 stands for the
where 𝐶1 is an modification factor for inelastic displacements; 𝐶2 is an uncorrected displacement of floor 𝑖; 𝛿𝑟 is the uncorrected roof dis-
modification factor for cyclic deterioration; 𝑆𝑎 (𝑇1 ) stands for the 5% placement with respect to the base, and all other terms are defined
damped spectral acceleration at the fundamental period of the building previously.

8
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

In both analysis procedures presented in FEMA P-58, the following Table 5

equation is used to estimate the median residual drift ratio (𝛥𝑟 ) as a Parameter settings for NSGA-II.

function of the peak transient response of the structure: Parameter Value

Selection type Binary tournament
⎧0 if 𝛥 ≤ 𝛥𝑦 Crossover type Intermediate
⎪ Crossover fraction 2/nVara
𝛥𝑟 = ⎨0.3(𝛥 − 𝛥𝑦 ) if 𝛥𝑦 < 𝛥 < 4𝛥𝑦 (9)
⎪ Mutation type Gaussian
⎩𝛥 − 3𝛥𝑦 if 𝛥 ≥ 4𝛥𝑦 Mutation fraction 2/nVar
Population size 200
where 𝛥 is the median story drift ratio calculated by analysis, which
Number of generations 200
is set to 𝛥∗𝑖 in the simplified analysis; and 𝛥𝑦 is the median story drift
ratio at the yield point. In this study, 𝛥𝑦 is assumed to correspond to a
nVar is the number of design variables.
the roof drift ratio (roof displacement divided by the building height)
at the yield point and is determined from the pushover analysis [7].
The structural demands obtained from nonlinear response history functions. The first characteristic is the nondomination rank (known
or simplified analysis are considered as estimated median values of as front), which is obtained by applying a nondominated sorting pro-
response parameters. In this approach, in order to develop demand cedure. In this process, the first front belongs to individuals that are
distributions, the effective uncertainties in calculating the response not dominated by any others. Those that are dominated only by the
quantities must also be considered. Three sources of demand param- individuals of the first front are inserted in the second front, and so on.
eter uncertainty include modeling uncertainty (𝛽𝑚 ), record-to-record The second characteristic is the crowding distance, which is applied
variability (𝛽𝑎 ), and ground motion variability (𝛽𝑔𝑚 ). However, in the to specify the priority of the individuals on each front. This measure
intensity-based and time-based assessments, as adopted in this study, prefers the individuals located in a lesser crowded region of the search
uncertainty in ground motion intensity will be zero (𝛽𝑔𝑚 = 0). More-
space. A simple approach is employed for the constraint-handling in the
over, if response history analyses are used to determine the structural
NSGA-II algorithm. In this approach, to find a better solution between
response, the uncertainty corresponding to the record-to-record vari-
two individuals, they are first examined in terms of the constraint
ability is considered as zero [6]. Therefore, in this case, the only source
violation. If the violation is zero for both (i.e., both are feasible), the
of uncertainty in the intensity-based and time-based assessments is the
dominance principle is directly utilized to specify the winner. If only
modeling uncertainty, which is calculated as follows:
√ one of them is feasible, the feasible individual dominates. Whenever
𝛽𝑚 = 𝛽𝑐2 + 𝛽𝑞2 ≤ 0.5 (10) both individuals are infeasible, the individual with a smaller constraint
violation dominates.
where 𝛽𝑐 and 𝛽𝑞 are the uncertainties corresponding to the construction
quality, and the quality of the nonlinear analytical model, respec- 4.4.1. Optimization parameters
tively. In this study, these two parameters are set to 0.1 and 0.4, In this study, the control parameters of NSGA-II are considered
respectively [6]. according to Table 5. It is noteworthy that the operators of crossover
In simplified analysis, unlike response history analysis, uncertainty and mutation apply to all individuals in the population, but only to a
associated with record-to-record response variability must be consid- fraction of design variables indicated by parameters of mutation frac-
ered. Consequently, the total dispersion values in the intensity-based tion and crossover fraction in the table [59]. Due to the random nature
and time-based assessments are computed using the following equa- of NSGA-II algorithm, the optimization process has been performed
tions for each of the demand parameters: several times and the best results have been presented in this paper.
√
𝛽𝑆𝐷 = 𝛽𝑎𝛥 2 + 𝛽2 (11)
𝑚
√ 5. Automated design and evaluation procedure
𝛽𝐹 𝐴 = 2 + 𝛽2
𝛽𝑎𝑎 (12)
𝑚
√ In this section, by integrating a multi-objective optimization algo-
𝛽𝐹 𝑉 = 2 + 𝛽2
𝛽𝑎𝑣 (13)
𝑚 rithm (e.g., NSGA-II) and the FEMA P-58 comprehensive performance
where 𝛽𝑆𝐷 , 𝛽𝐹 𝐴 , 𝛽𝐹 𝑉 are the total dispersion and 𝛽𝑎𝛥 , 𝛽𝑎𝑎 , 𝛽𝑎𝑣 are assessment methodology, an automated process for the bi-objective
record-to-record dispersion values, respectively, for drift, floor accel- optimum design of buildings is presented. The design alternatives
eration, and floor velocity; and 𝛽𝑚 is the modeling uncertainty. The obtained as the Pareto-optimal front are not only code-compliant but
values of these parameters are obtained based on 𝑇1 and 𝑆 from Table are also optimal for the defined objectives of initial cost (𝑓1 ) and
5-6 of FEMA P-58 [6]. Notably, since the 𝑇1 and 𝑆 values are different average annual repair cost (𝑓2 ). The evaluation of objective functions
for the design alternatives obtained in the optimization process, their (determination of 𝑓1 and 𝑓2 values) during the optimization process
total dispersion values are also different. Also, it is noteworthy that the for each of the design alternatives (individuals) in the population is
total dispersion in the residual drift is considered equal to 0.8 in the performed according to the flowchart presented in Fig. 4 and as follows:
simplified analysis. While in the response history analysis, similar to Step 1. At the first step, the initial construction cost (𝑓1 ) is deter-
other demand parameters, it is determined according to the modeling mined for the adopted design. Since the aim of optimization in this
uncertainty. study is to compare design alternatives with each other, regarding the
fixed costs of nonstructural components, contents, and gravity frames
4.4. Optimization algorithm for all design alternatives, only the cost of moment frames is defined
here as 𝑓1 . Accordingly, 𝑓1 is calculated by multiplying the unit cost
As mentioned in the Introduction, metaheuristic optimization algo- of structural steel by the total weight of the moment frames. The
rithms are robust in solving a wide range of problems and covering the present study assumed the unit cost of structural steel as $2.14/lb in
true Pareto front. Therefore, in this work a well-known multi-objective Los Angeles (in 2020 dollars).
metaheuristic algorithm, NSGA-II [18], is used to solve Eq. (1). This Step 2. Design constraints are examined. Except for the geomet-
algorithm has been applied in numerous studies for multi-objective ric constraints, controlling other constraints defined for the problem
structural optimization purposes (see [29,54–58], for example). In the requires linear analysis according to the code requirements.
NSGA-II algorithm, two characteristics are calculated for each individ- Step 3. If the design constraints are not satisfied, the average annual
ual of the population according to the values of corresponding objective repair cost (𝑓2 ) is considered as the building replacement cost. By doing

9
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Step 4. At this step, the current design alternative is deemed as a

feasible solution to the problem, and a building performance model is
developed for it. Also, the possible building collapse modes are defined
together with developing the performance model. It should be noted
that except for the fragility functions associated with the structural
members, other performance model data, as well as collapse modes,
will remain constant during the optimization process for different
design alternatives.
Step 5. According to the fundamental period of the structure, the
corresponding hazard curve is derived from the series of site-specific
seismic hazard curves. The hazard curve is then split into several
equal spectral acceleration intervals (8 intervals are considered here
as suggested by FEMA P-58 [6]). The mid-point intensity (spectral
acceleration) of each interval as well as its corresponding probability
is then determined. The probability of each interval (the mean annual
frequency of occurrence of ground motion having an intensity that
falls within each interval) is equal to the difference between the mean
annual frequencies of exceedance at the end-points of the interval. Also,
a target spectrum is developed for each interval, using the mean annual
frequency of exceedance of each mid-point spectral acceleration.
Step 6. The nonlinear static analysis (pushover) is performed. The
obtained pushover curve is then idealized, and its results are utilized
to determine the lateral yield strength as well as the yield drift of the
structure. Also, using the SPO2IDA tool [41], the results of the idealized
curve are used to estimate the collapse capacity of the structure (𝑆̂𝑎 (𝑇 )).
In this study, the total dispersion for deriving the collapse fragility
curve is considered equal to 0.6 [6].
Step 7. The structure is analyzed for the mid-point intensity of each
interval, which is characterized as the spectral response acceleration
obtained from the corresponding spectrum in the simplified procedure,
or a suite of earthquake records that are selected and scaled for con-
sistency with the corresponding spectrum in the nonlinear response
history analysis. This includes 𝑛𝐼 × 𝑛𝑅𝑒𝑐 structural analyses, in which
𝑛𝐼 represents the number of ground motion intensities (here 𝑛𝐼 = 8).
The parameter 𝑛𝑅𝑒𝑐 is equal to 1 in the simplified analysis, while in
the nonlinear response history analysis, it is equal to the number of
selected earthquake records. As the structural analysis is performed for
each shaking intensity, the demand parameters, including story drift
ratios, floor accelerations, floor velocities, and the maximum residual
story drift ratio, as well as the total dispersions corresponding to each
demand parameter are determined.
Step 8. The results obtained from step 7 for the current earthquake
intensity (𝐼𝑀𝑖 ) form the demand ([𝐸𝐷𝑃 𝑠]𝑚×𝑛𝑅𝑒𝑐 ) and total dispersion
([𝐵]𝑚×1 ) matrices. A Monte Carlo procedure is then utilized to generate
simulated demand sets ([𝑆𝑖𝑚𝐸𝐷𝑃 𝑠]𝑚×𝑛𝑅 ) for this intensity. The number
of demand parameters, 𝑚, is equal to the sum of 2 × (𝑛𝐹 − 1) story drift
ratio parameters, 𝑛𝐹 floor acceleration parameters, 𝑛𝐹 floor velocity
parameters, and one parameter of the maximum residual drift ratio (𝑛𝐹
is the number of floors in the building). The number of realizations, 𝑛𝑅,
is taken equal to 500 in this work based on [7]. Note that the algorithm
of generating the simulated demands using the results of structural
analysis is described in Appendix G of FEMA P-58 [6].
Step 9. The 𝑛𝑅 simulated times are generated that include a ran-
dom definition of the time of day and day of the year at which the
earthquake is assumed to occur.
Step 10. According to the earthquake occurrence time resulting
from step 9, along with the time-dependent population model, the
number of persons present on each floor of the building at that time
Fig. 4. Flowchart to determine the initial and mean annual repair costs of each
building design option. is determined for the current realization.
Step 11. By comparing the collapse fragility function obtained
from step 6 with the current earthquake intensity, the probability of
structural collapse for the current intensity is determined. Then, by
generating a random number between 0 and 1, it is determined whether
so, evaluating the objective functions for the current design alternative the collapse has occurred for the current realization or not. If the
is completed. Otherwise, the process is followed from step 4. random number is less than or equal to the conditional probability

10
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

of collapse, collapse is deemed to occur in this realization. If collapse in all performance groups is the output of the overall performance for
occurs, the process enters step 12, otherwise step 13 is followed. the current realization. Also, at this stage, according to the number
Step 12. The building is assumed to be a total loss due to the of damaged components and their damage state, the condition of the
collapse. In this case, the repair cost, repair time, and environmental building is examined for whether or not it is unsafe for occupancy. If
impacts are assumed equal to the values associated with the build- for a fragility group, the ratio of the number of components damaged
ing replacement, and the building is labeled as unsafe. The collapse to a given damage state (including all performance groups with this
casualties, including serious injuries and fatalities, are also calculated fragility that are in this damage state) to the total number of building
in this case. To do this, first, it is determined which of the collapse components for which this fragility is assigned, exceeds the triggering
modes defined for the building will occur in the current realization. The percentage which is defined in its fragility function, the building is
collapse mode probabilities are supposed to be an array that varies from assigned an unsafe placard.
1 to 100, with 𝑛𝑖 sequential numbers assigned to each mode, where 𝑛𝑖 is Step 16. In this step, it is determined based on two criteria whether
the conditional probability of occurrence for mode 𝑖, multiplied by 100. the building is repairable or not. If the ratio of the repair cost to the
A random number between 1 and 100 is generated. The range of the building replacement cost exceeds a threshold value (e.g., 50%), the
array in which this number falls indicates the collapse mode that has repair of the building is not practical, and therefore the building is
occurred. Then, by using the fraction of the floor area that is exposed to assumed to be irreparable. Also, the maximum residual drift ratio is
collapse in the given mode and the total number of people present on taken from the vector of demand parameters (step 13) for the current
the floor resulting from step 10, the collapse casualties are computed realization. By comparing this value with the building repair fragility,
for the current realization. The process then continues from step 17. the probability of labeling the building as irreparable is calculated. The
Step 13. The simulated demand vector for the current realization is repair fragility is supposed as a lognormal distribution with a median
taken from step 8. value of 1% residual drift ratio and a dispersion of 0.3. [6]. A random
Step 14. Based on the fragility functions assigned to each perfor- number is then generated between 0 and 1. If this number is less than or
mance group (step 4) and using random number generation, a damage equal to the conditional probability of the building being irreparable, it
state is determined for each component of each group. In this study, is considered that the building is irreparable for this realization. If the
the damage states of all components within a performance group building is considered to be irreparable, the repair cost, repair time, and
are considered correlated. This means that all the components in a environmental impacts calculated from step 15 will be replaced with
performance group are expected to have the same damage states, and the building replacement-related values (just like the collapse case).
determination of damage state is performed only once for each per- Moreover, in this case, the building is assigned an unsafe placard.
formance group. It should be noted that defining performance groups Step 17. Steps 10 to 16 are repeated until all realizations are
as uncorrelated can significantly increase the data handling time and evaluated (i.e., 𝑛𝑅 is reached).
required computer disc space because the process of determining the Step 18. At the end of the performance assessment for all real-
damage state must be repeated n times, where n is the number of izations, the median and dispersion values are determined for each
components in the group. Depending on the fragility function defined performance metric, and then a lognormal distribution function is fitted
for each performance group, the expected damage states may be se- to each one (a loss curve is developed for each performance metric).
quential, mutually exclusive, simultaneous, or a combination of these. Regarding the unsafe placarding, the probability of the building is
The process of determining the damage state for each of these is unsafe for occupancy in the current earthquake intensity is calculated
different. as the number of realizations in which an unsafe placard is assigned to
For sequential damage states, first, by comparing the value of the building divided by the total number of realizations.
demand parameter and fragility curve, the probability of not incurring Step 19. Steps 8 to 18 are repeated until the maximum number of
each damage state and higher is determined as 𝑃𝑖 = 1 − 𝑃 (𝐷𝑆 ≥ 𝐷𝑆𝑖 ). intensities (𝑛𝐼) is reached.
Then, a range of sequential numbers is assigned to each damage state. Step 20. At the end of the performance assessment for all earth-
The range of numbers assigned to the no damage state would be 1 quake intensities, the following process is performed for each per-
to (𝑃1 × 100), the range assigned to the first damage state would be formance metric to derive its associated annual loss (performance)
(𝑃1 × 100 + 1) to (𝑃2 × 100) and so on. Finally, a range of numbers from curve. First, the loss curve obtained for each earthquake intensity
(𝑃𝑛 ×100+1) to 100 would be assigned to the last damage state, where 𝑛 (resulting from step 18) is multiplied by its corresponding probability
is the total number of damage states. Subsequently, a random number of occurrence (resulting from step 5). Then, for a given value of the
between 1 and 100 is generated. The range in which this number falls loss (e.g., a given value of repair cost), the annual frequencies of all
indicates the damage state that has occurred. intensities are added together. This calculation is defined as:
In the case of mutually exclusive damage states, the probabilities of
the damaged states are assumed to be an array that varies from 1 to 𝑃 (𝐿 > 𝑙) = 𝑃 (𝐿 > 𝑙|𝐼𝑀 = 𝐼𝑀𝑖 )𝑑𝜆(𝐼𝑀) (14)
∫𝜆
100, with 𝑛𝑖 sequential numbers assigned to each damage state where
𝑛𝑖 is the conditional probability of occurrence for the damage state 𝑖 in which 𝑃 (𝐿 > 𝑙) denotes the annual probability that the loss, 𝐿, will
multiplied by 100. A random number between 1 and 100 is generated. exceed a value, 𝑙; the term 𝑃 (𝐿 > 𝑙|𝐼𝑀 = 𝐼𝑀𝑖 ) is the loss curve
The range of the array in which this number falls shows the type of derived from step 18 for intensity 𝐼𝑀𝑖 ; and 𝑑𝜆(𝐼𝑀) represents the
damage that occurs. annual occurrence rate of 𝐼𝑀, which is determined from step 5.
For simultaneous damage states, an initial random number is first Step 21. The area below the annual loss distribution curve is
generated to determine whether the damage has occurred or not. If the calculated by numerical integration and considered as the average
random number is less than the probability of damage occurrence, it (expected) annual loss. Although only the expected amount of annual
is considered that damage of some type will occur in the realization. repair cost is defined as the second objective of the optimization
Under the conditions in which the damage has occurred, a second ran- problem (𝑓2 ) in the present study, in order to compare different design
dom number is generated for each possible damage state to determine alternatives, the average annual amount of other performance measures
whether one or more of the simultaneous damages states have occurred are also calculated and stored during the optimization process.
or not.
Step 15. Considering the damage state obtained for each of the 6. Results and discussion
components in each performance group, the losses are calculated based
on the damage sustained by the component and the consequence Fig. 5 illustrates the results of multi-objective optimization for
functions assigned to each performance group. The sum of the losses archetypes. The horizontal and vertical axes represent the initial cost

11
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 5. Optimal and non-optimal feasible designs for (a) three-story, (b) six-story, and (c) nine-story archetypes.

of moment-resisting frames (MRFs) and the average annual repair and consequently the total cost. The level of uncertainty in the men-
cost, respectively. In this figure, in addition to the Pareto front, all tioned parameters relative to the base values (i.e., the interest rate
the feasible design alternatives found for the structures during the of 7% and downtime cost of $0.31/ft2 /day) was considered between
optimization process are presented. It is noteworthy that the time 0% to 40%. Accordingly, interest rates range from 4% to 10%, and
spent to achieve these results for 3-, 6- and 9-story structures is 47, downtime costs range from $0.18 to $0.43 per square foot per day.
59 and 89 h, respectively. These times, which were recorded using It should be noted that the effect of uncertainty in other components
an Intel(R) Core(TM) i7-4790 @ 3.6 GHz processor and based on involved in the annual loss, such as annual repair cost and repair time,
parallel processing with 6 threads, include the total time spent in the is considered through Monte Carlo simulation during the time-based
optimization algorithm search process, linear and non-linear analyses, assessment process. Also, the unit cost of structural steel (the initial cost
and loss assessment. As can be seen, an increase in the initial cost of of MRFs) is assumed as a deterministic component here. Fig. 6 shows
moment frames results in a reduction of the average annual repair cost the uncertainty and sensitivity analysis on the total cost of selected
of the building. For example, in the 3-story building, increasing the designs of the building archetypes with respect to the interest rate and
initial cost of moment frames up to 2.5 times decreases the average the downtime cost. Total costs are normalized to the floor area of the
annual repair cost of the building up to 2 times. As is clear, the amount buildings.
of increase in the initial cost and the corresponding decrease in the As shown in Fig. 6(a), the total cost decreases with an increase in
building repair cost (i.e., relative to the left extreme point of the Pareto the interest rate. However, according to the slope of the graphs, it can
front) reduces with enlarging the dimensions of the structure. be concluded that the effect of interest rate uncertainty on the total
In order to compare the design specifications and seismic per- cost decreases with increasing the value of this parameter. For instance,
formance of optimal design alternatives, optimal designs with the in design B of the 6-story building, increasing the interest rate from
minimum values for the initial cost and the average annual repair cost 4% to 7% reduces the total cost from $24.26/ft2 to $19.53/ft2 (−20%),
are labeled A and C in Fig. 5, respectively. Depending on the project and increasing it from 7% to 10% reduces the total cost to $17.63/ft2
aims and personalized risk, the decision-makers can choose the desired (−10%). Also, as can be seen, the effect of the interest rate decreases
option from a set of Pareto front design alternatives located between with increasing the initial cost of the structure (e.g., from design A to
designs A and C. The cost–benefit studies can here help the decision- C). For example, raising the interest rate from 4% to 10% in design A of
makers to choose the preferred investment option. The cost–benefit the 6-story building reduces the total cost from $24.9/ft2 to $17.55/ft2
studies include a comparison between the net present value (NPV) of (−30%), while in design C, it reduces the total cost from $37.45/ft2 to
average annual costs avoided by improving seismic resistance, versus $32.67/ft2 (−13%).
the costs associated with improving seismic resistance. The net present As shown in Fig. 6(b), contrary to the interest rate, the relationship
value, 𝑁𝑃 𝑉 , of the avoided costs for a long-term investment (with an between the total cost and downtime cost is a direct and linear one.
expected life of more than 40 years) is determined from the following Moreover, the effect of this sub-component on the total cost is less
equation [6]: than the interest rate. Assuming the maximum uncertainty level for the
𝐴𝐶𝑜𝑠𝑡 downtime cost (±40%), the total cost changes by a maximum of ±10%
𝑁𝑃 𝑉 = (15)
𝑖 (average of ±6%) relative to the base value. While for the maximum
where 𝐴𝐶𝑜𝑠𝑡 stands for the avoided annual costs, and 𝑖 is the interest uncertainty level in the interest rate, the total cost varies on average by
rate (internal rate of return on investment). Therefore, to find the 23% and 9% relative to the base value, respectively, for interest rates
preferred design option, given the average annual repair cost and repair lower and higher than the base value. As shown in Fig. 6(b), similar
time resulting from the time-based assessment, the NPV of the average to the interest rate, the slope of changes in the total cost decreases
annual loss for all Pareto front design alternatives is first determined. with increasing initial cost, indicating a lesser effect of downtime cost
For this purpose, the downtime cost and the interest rate are assumed uncertainty on the total cost of design C compared to design A.
to be $0.31∕ft2 ∕day and 7%, respectively. Subsequently, the design The specifications of designs A, B, and C for the 3-, 6-, and 9-
with the minimum sum of the initial construction cost and the NPV of story archetypes are presented in Table 6, Table 7, and Table 8,
the average annual loss is selected as the alternative for a reasonable respectively. According to the values given in Table 6, the average
investment. This design is labeled B in Fig. 5. Option B indicates a annual repair cost for design option B of the 3-story building is $18432,
design in which any further investment in construction costs (cost of the and the average annual loss due to business interruption is 0.6 days
structural moment-resisting frames) will not result in any reduction in ×($0.31/ft2 /day × 64800 ft2 ) = $12,053, which leads to the total
long-term total costs. In other words, the advantage of reducing future average annual loss of $30,485. The NPV of the annual loss over a
costs will be less than any additional initial investment. long period for this design is $435,500. Considering the initial cost
The sensitivity analysis was performed for designs A, B, and C of of the design, its total cost will be $1,032,175, which is less than
the building archetypes to investigate the effect of sub-components the corresponding values for design options A ($1,170,777) and C
of interest rate and downtime cost on the NPV of the annual loss ($1,499,161), as well as other Pareto front design options obtained for

12
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 6. Uncertainty and sensitivity analysis of total cost on designs A, B, and C of the archetypes, with respect to (a) interest rate, (b) downtime cost.

Table 6
MRFs sections for the selected designs from the Pareto optimal set found for the three-story archetype.
Design A B C
Element group 1 W12X136 W14X132 W14X311
2 W14X211 W14X257 W14X808
3 W14X132 W14X132 W14X283
4 W14X211 W14X233 W14X550
5 W12X106 W14X132 W14X233
6 W14X145 W14X233 W14X398
7 W24X76 W24X84 W18X130
8 W24X84 W30X108 W27X194
9 W18X50 W21X68 W24X94
MRFs weight (kips) 228.34 278.82 561.05
Initial cost of MRFs (U.S. dollars) 488,648 596,675 1,200,647
Average annual repair cost (U.S. dollars) 27,862 18,432 13,263
Average annual repair time (days) 0.99 0.6 0.38

Table 7
MRFs sections for the selected designs from the Pareto optimal set found for the six-story archetype.
Design A B C
Element group 1 W14X233 W14X211 W14X455
2 W14X283 W14X283 W14X665
3 W14X159 W14X176 W14X426
4 W14X283 W14X283 W14X550
5 W14X132 W14X145 W14X283
6 W14X233 W14X257 W14X550
7 W27X114 W27X114 W27X281
8 W27X94 W27X94 W24X279
9 W21X62 W21X83 W27X114
MRFs weight (kips) 798.44 833.84 1860.05
Initial cost of MRFs (U.S. dollars) 1,708,662 1,784,418 3,980,507
Average annual repair cost (U.S. dollars) 30,520 27,393 19,992
Average annual repair time (days) 0.85 0.77 0.55

the 3-story archetype. Similarly, the sum of the initial construction cost the shaking intensities No. 1 to No. 8 correspond approximately to the
and the NPV of the average annual loss for design options A, B, and C occurrence probabilities of 68%, 29%, 14%, 8%, 4%, 3%, 2%, and 1%
of the 6-story building are $2,652,848, $2,636,104, and $4,594,936, in 50 years, respectively.
respectively, and of the 9-story building $4,535,596, $4,517,094, and
To compare the seismic performance of designs A, B, and C, the
$5,601,630, respectively.
median values of different performance metrics obtained from the
The loss curves for the repair cost associated with the eight se-
lected seismic intensities, as well as the annual loss curve of repair evaluation of structural performance at eight seismic intensities (500
cost considering all intensities, are shown in Figs. 7, 8 and 9 for the realizations per intensity) are considered. The values provided for these
design B of 3-, 6- and 9-story archetypes, respectively. According to the metrics at each intensity are equal to the median of 500 realizations
hazard curves derived for different design alternatives of archetypes, occurring at that intensity.

13
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 8
MRFs sections for the selected designs from the Pareto optimal set found for the nine-story archetype.
Design A B C
Element group 1 W14X283 W14X257 W14X665
2 W14X370 W14X455 W14X808
3 W12X252 W12X252 W12X252
4 W14X370 W14X426 W14X665
5 W12X230 W12X252 W12X230
6 W14X342 W14X342 W14X455
7 W14X145 W14X159 W14X211
8 W14X311 W14X233 W14X455
9 W14X82 W14X159 W14X176
10 W14X159 W14X193 W14X342
11 W33X141 W33X141 W33X141
12 W30X132 W30X124 W30X132
13 W30X124 W30X124 W30X124
14 W24X94 W27X102 W24X131
15 W21X50 W21X50 W24X76
MRFs weight (kips) 1473.85 1521.88 2136.95
Initial cost of MRFs (U.S. dollars) 3,154,039 3,256,823 4,573,073
Average annual repair cost (U.S. dollars) 36,445 32,349 26,801
Average annual repair time (days) 0.96 0.89 0.72

Fig. 7. (a) Cumulative probability distributions of total repair cost at eight earthquake intensities (b) distribution of mean annual total repair cost, for Design B of the three-story
archetype.

Fig. 8. (a) Cumulative probability distributions of total repair cost at eight earthquake intensities (b) distribution of mean annual total repair cost, for Design B of the six-story
archetype.

Figs. 10 and 11 compare the feasible design alternatives obtained for of days and as a percentage of the building replacement time. As shown
the building archetypes in terms of their median repair cost and repair in Figs. 10 and 11, for design A of the 3-story building, the repair cost
time at different ground motion intensities, respectively. The repair and repair time of the building under the shaking intensity No. 6 and
cost is presented both in dollars and as a percentage of the building higher are set to the values associated with the building replacement.
replacement cost. Similarly, the repair time is provided both in terms This means that adopting design A (i.e., an optimal design with the

14
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 9. (a) Cumulative probability distributions of total repair cost at eight earthquake intensities (b) distribution of mean annual total repair cost, for Design B of the nine-story
archetype.

Fig. 10. Intensity-based assessment results in terms of repair cost for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

Fig. 11. Intensity-based assessment results in terms of repair time for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

minimum initial cost of moment frames) for the 3-story building can 29%, and 23% of the building replacement cost, and the median repair
make the building irreparable when it is subjected to the mentioned time at this intensity for mentioned designs is 200, 197, and 163 days,
intensities. As can be seen, at low ground motion intensities (less respectively. Also, the median values for repair cost and repair time
than No. 5), there is not much difference between building repair of designs A, B, and C of the 9-story building at shaking intensity No.
times for different design alternatives of the 3-story building. However, 8 are 26%, 24%, and 17% of the building replacement cost, and 253,
at high seismic intensities especially when the building needs to be 228, and 171 days, respectively. As shown in Figs. 10(b) and 11(b),
demolished and replaced (for example, design A at seismic intensity No. the repair cost and repair time for a large number of design alternatives
6 or higher), there will be a large difference between different design obtained for the 6-story building are higher than design option A of this
alternatives in terms of the downtime and resulting indirect losses. building (this can also be seen in Fig. 5(b) by imagining a horizontal
The difference between design alternatives A, B, and C of the 6- and line passing through point A in this figure). In other words, these design
9-story archetypes in terms of repair cost and repair time at different options not only have a higher initial cost than design A but also have
seismic intensities is less than the corresponding values of the 3-story a higher repair cost and repair time than design A at different ground
archetype. The median repair cost under the shaking intensity No. 8 motion intensities. This indicates the availability of many designs with
for design A, B, and C of the 6-story building is respectively, 31%, an improper strength (stiffness) distribution for the 6-story archetype.

15
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 12. Intensity-based assessment results in terms of carbon emissions for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

Fig. 13. Intensity-based assessment results in terms of embodied energy for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

It is noteworthy that, unlike the 3- and 6-story archetypes, the repair collapse, and the fatality rate is directly related to the collapse capacity
cost of the 9-story archetype has not reached 50% of the building of the structure, while injuries can occur due to the collapse or damage
replacement cost for none of the feasible design alternatives. to the components and contents. For instance, under shaking intensity
Figs. 12 and 13 provide the environmental impacts of structural No. 8 (seismic events with 1% probability of occurrence in 50 years) the
seismic damage, including carbon emissions and embodied energy, at fatality rates for design options A, B, and C of the 3-story building are
different shaking intensities for the design alternatives. As can be seen, 4.8%, 2.6%, 0.7%, of the 6-story building are 2%, 1.9%, 0.0%, and of
the environmental impacts, similar to the repair time, are directly the 9-story building are 2.6%, 1.8%, 0.3%, respectively. While, under
related to the annual repair cost. However, in general, the differences this seismic intensity, the injury rates for designs A, B, and C of the
between the design options in terms of environmental impacts are 3-story building are 15.9%, 14.6%, 14.9%, of the 6-story building are
greater than their differences in terms of repair cost and repair time. 12.3%, 12.7%, 11%, and of the 9-story building are 10.9%, 10.1%, 8%,
Also, at each ground motion intensity, the ratio of environmental im- respectively.
pacts caused by earthquake damage to the building replacement-related As shown in Figs. 14 and 15, the casualty rate of several design
values is more than the ratio of the corresponding values for the repair options is higher than design A, and for some, it is less than design
cost (repair time) to the building replacement cost (replacement time). C. In other words, contrary to the previous performance metrics, the
For example, for design B of the 6-story building, the median values injury and fatality rates cannot be directly related to the building repair
for carbon emissions and embodied energy at the shaking intensity cost. Therefore, optimizing (minimizing) the annual repair cost of the
No. 8 are 52% and 58% of values related to building replacement, building does not necessarily result in optimizing (minimizing) the
respectively, while the median values for repair cost and repair time fatality or injury rates of the building. A design option may have a
are 29% and 26% of building replacement-related values, respectively. higher average annual repair cost than design C, but its fatality or injury
It should be noted that similar to the performance metrics of repair cost rate may be less than this design option, or an option may have a less
and repair time, the difference between design alternatives A, B, and C annual repair cost than design A, but it has a higher fatality or injury
of the 6- and 9-story archetypes in terms of environmental impacts at rate.
different seismic intensities is less than the corresponding values of the Another performance metric determined in the framework of the
3-story archetype. FEMA P-58 methodology is unsafe placarding. If the extent of damage
In order to assess the casualties, two performance metrics of fatality to structural and nonstructural components is significant (based on the
rate and injury rate are calculated for different design alternatives, and values provided in the consequence functions for each component) to
their median values are presented in Fig. 14 and Fig. 15, respectively. make the building unsafe for occupancy, the building is assigned an
The fatality (injury) rate is equal to the probability that each occupant unsafe placard. It also results in the posting of an unsafe placard, if
in the building will die (be injured) as a result of an earthquake. the building has collapsed or when it is considered irreparable due to
As can be seen, despite the large differences in fatality rates between residual drift. Therefore, the probability of assigning an unsafe placard
different design alternatives, there is not much difference between to the building can be considered as a measure for assessing the overall
them in terms of injury rates (especially between optimal designs A, performance of the building under an earthquake. The probability of
B, and C of the 3- and 6-story buildings and designs A and B of the 9- causing an unsafe placard after an earthquake at different intensities for
story building). This is because fatalities often occur due to structural the design alternatives of the building archetypes is shown in Fig. 16.

16
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 14. Intensity-based assessment results in terms of fatality rate for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

Fig. 15. Intensity-based assessment results in terms of injury rate for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

Fig. 16. Intensity-based assessment results in terms of unsafe placarding for feasible design options of (a) three-story, (b) six-story, (c) nine-story archetypes.

As shown in this figure, the probability of assigning an unsafe placard for occupancy varies between the design alternatives. For example,
to the 3-story building with designs A, B, and C under the maximum at the MCE seismic intensity, the probability of unsafe placarding for
considered earthquake (MCE) (here intensity No. 7) is higher than 90%, the 9-story building with designs A, B, and C is 60%, 54%, and 45%,
80%, and 70%, respectively. At the design basis earthquake (DBE) level respectively.
(shaking intensity between No. 3 and No. 4), the probability of causing As mentioned earlier, the simplified analysis has been utilized in
an unsafe placard for the 3-story building with designs A, B, and C is this study to obtain the structural response parameters and evaluate
about 50%, 30%, and 20%, respectively. the performance of the archetypes during the optimization process
The probability of incurring an unsafe placard for designs A, B, because of the large number of performance assessments. To evaluate
and C of the 6-story is almost equal under different shaking intensi- the reliability of the simplified analysis procedure in the probabilistic
ties. Accordingly, the probability that the 6-story building with any performance assessment of the building archetypes, the three selected
design options A, B, or C is unsafe for occupancy under MCE and design alternatives have been reanalyzed using nonlinear response
DBE intensities is about 60% and 15%, respectively. In the 9-story history analyses with a set of 15 ground motion records. The results
building, almost all the feasible designs are close to each other in of different performance metrics obtained from the time-based assess-
terms of unsafe placarding at low shaking intensities (intensities less ment of the three design alternatives using both analysis methods are
than No. 5). For example, at DBE seismic intensity, the probability that presented in Table 9, Table 10, and Table 11, for 3-, 6- and 9-story
the 9-story building with any of the three selected designs is assigned buildings, respectively.
an unsafe placard is approximately 13%. However, at high ground The results provided in Table 9 demonstrate that the simplified anal-
motion intensities, the probability of the 9-story building being unsafe ysis procedure estimates some performance metrics for designs A and B

17
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Table 9
Time-based assessment results based on simplified and nonlinear response history analysis for the three-story archetype.
Performance measure Design A Design B Design C
(annualized value)
Simplified Response History Simplified Response History Simplified Response History
Repair cost (U.S. dollars) 27,862 21,420 18,432 16,685 13,263 14,021
Repair time (days) 0.99 0.65 0.6 0.49 0.38 0.35
Fatalities (person) 0.00025 0.00026 0.00011 0.00011 0.00002 0.00003
Injuries (person) 0.013 0.014 0.0124 0.0136 0.0135 0.0153
Carbon emissions (kg) 12,101 9,602 7,444 6,621 4,631 4,826
Embodied energy (MJ) 221,222 181,027 138,100 126,042 80,526 85,149

Table 10
Time-based assessment results based on simplified and nonlinear response history analysis for the six-story archetype.
Performance measure Design A Design B Design C
(annualized value)
Simplified Response History Simplified Response History Simplified Response History
Repair cost (U.S. dollars) 30,520 30,028 27,393 30,017 19,992 20,241
Repair time (days) 0.85 0.84 0.77 0.83 0.55 0.52
Fatalities (person) 0.00142 0.00147 0.00144 0.00145 0.00003 0.00002
Injuries (person) 0.0237 0.0272 0.0241 0.0268 0.0258 0.0312
Carbon emissions (kg) 10,961 10,978 10,014 10,765 6,688 6,244
Embodied energy (MJ) 204,888 203,836 187,578 194,809 113,551 109,877

Table 11
Time-based assessment results based on simplified and nonlinear response history analysis for the nine-story archetype.
Performance measure Design A Design B Design C
(annualized value)
Simplified Response History Simplified Response History Simplified Response History
Repair cost (U.S. dollars) 36,445 38,394 32,349 35,964 26,801 32,589
Repair time (days) 0.96 1.00 0.89 0.95 0.72 0.83
Fatalities (person) 0.004 0.0037 0.0021 0.0022 0.0002 0.0002
Injuries (person) 0.0351 0.0417 0.032 0.0406 0.0302 0.0399
Carbon emissions (kg) 12,389 13,446 10,982 12,054 8,828 10,476
Embodied energy (MJ) 228,488 242,144 201,268 223,400 157,805 184,905

(especially A) of the 3-story building more than the nonlinear response This means that STR components have not experienced any damage at
history analysis, while its estimation for design option C is less than the this intensity.
response history analysis. However, in general, there is good agreement A comparison between Figs. 17 and 18 indicates that increasing the
between the results of different performance metrics obtained from the shaking intensity from No. 2 to No. 6 reduces the contribution of NSD
two procedures for the 3-story building, and the differences between and NSV components to the building repair cost, while the contribution
the results are not significant, especially for the design options B and of STR and NSA components increases. Nevertheless, the drift-sensitive
C. The results given in Table 10 demonstrate that for all three design nonstructural components still have the largest contribution to building
alternatives of the 6-story archetype, the simplified analysis can predict repair costs. In addition, by moving from design A to C, the contribution
the values of different performance measures with acceptable accuracy. of STR components to the building repair costs at intensity No. 6
In the 9-story archetype, the results of Table 11 show that the simplified decreases, while the contribution of NSA components increases. As
analysis estimates lower values for the performance metrics compared can be seen, for design option C of the 6- and 9-story buildings, the
to the response history analysis for the three design options. However, contribution of STR components to the building repair cost under the
this difference is not remarkable except for design option C. shaking intensity No. 6 has been reduced to close to the contribution
The contribution of the structural components (STR), drift-sensitive of NSV components.
nonstructural components (NSD), acceleration-sensitive nonstructural The results presented in Figs. 17 and 18 can be analyzed in more
components (NSA), and velocity-sensitive nonstructural components detail so that the contribution of each of the building components to the
(NSV) on the median repair cost of the archetypes at intensities No. repair costs at different seismic intensities is determined. For instance,
2 and No. 6 are shown in Fig. 17 and Fig. 18, respectively. It is Fig. 19 illustrates the 13 components that have the largest contribution
noteworthy that for the sake of brevity, contents such as bookshelves to the repair costs of the building archetypes (with design B) at shaking
are considered as nonstructural components here. intensity No. 6. As can be seen, a considerable portion of the repair
Fig. 17 demonstrates that a considerable contribution of the build- costs is caused by the curtain walls and the interior wall partitions as
ing repair cost at intensity No. 2 belongs to the drift-sensitive non- NSD components and the moment connections as STR components.
structural components (e.g., interior wall partitions). After NSD, the
acceleration-sensitive and velocity-sensitive nonstructural components, 7. Limitations and future work
have the largest contribution to the 3-story building repair costs at
shaking intensity No. 2, respectively. However, in the 6- and 9-story This paper employed FEMA P-58 within an automated multi-objective
buildings, the largest contribution after NSD first belongs to the NSV optimal design framework to determine various performance metrics of
and then to the NSA components. Furthermore, by moving from design a large number of feasible design options (optimal and non-optimal).
alternative A to C, the contribution of NSD components reduces, and in Given the wide range of options evaluated, the results of this re-
turn, the contribution of NSA and NSV components increases. As shown search can be useful in correctly identifying the vulnerable components
in Fig. 17, at ground motion intensity No. 2 and for all three design (including structural and non-structural) and contents, as well as pro-
options, the structural components do not contribute to the repair cost. viding loss mitigation approaches for use in existing design procedures.

18
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 17. Relative contributions to median repair costs of designs A, B, and C of the archetypes at shaking intensity No. 2.

Fig. 18. Relative contributions to median repair costs of designs A, B, and C of the archetypes at shaking intensity No. 6.

19
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

Fig. 19. Disaggregation of components’ contributions to median repair cost of (a) three-story, (b) six-story, and (c) nine-story archetypes (Design-B), at shaking intensity No. 6.

In addition, the obtained data can be useful for reviewing the gen- use of the building and its floor area, and therefore in the practical
eral fragility functions for structures (e.g., Hazus fragility functions). engineering design should be determined more accurately and in
However, there are limitations to the present study, which include the more detail.
following:
Despite the limitations mentioned, the design framework presented
• Research findings on different performance metrics are concluded in this article are still applicable for the multi-objective optimal design
based on a case study of three steel SMF archetypes with office of different structures (not just moment frame structures) with different
occupancy located on a site in Los Angeles with a soil type uses and located at different sites to provide a set of optimal options
C. Since the performance measures are determined through the in which the initial and seismic damage costs are in a trade-off rela-
four steps of hazard analysis, structural analysis, damage anal- tionship. The following work can be done in the continuation of this
ysis, and loss analysis, the research findings can be different research:
for applications in buildings with other structural systems, other
• Applying the proposed framework for optimizing building arche-
occupancies, and located at other sites. For example, changing
types with conditions different from those considered in this
the use of a building from an office to a hospital (healthcare) research (in terms of structural system, occupancy, and seismicity
affects the building population model, vulnerable non-structural of the site) for more general conclusions.
components and contents included in the performance model, and • Comparison of the results obtained in this study with the corre-
their quantities. Therefore, this causes different values for the sponding values obtained from other quantitative loss assessment
damage and loss analysis to be obtained under the same output methods (e.g., deterministic approaches, story-based or building-
from the hazard analysis and structural analysis steps (assuming based approaches instead of component-based).
the structural system and site are fixed). In the same way, in • Incorporating other concepts related to seismic losses, such as
terms of the change in the assumed site, it causes the results resilience and sustainability, into the optimization objectives.
derived from the hazard analysis, obtained from the site-specific • Evaluation of the performance of optimal structures in the frame-
seismic hazard maps, to change, and this affects the results of work of the ASCE 41 [5] guideline (based on qualitative perfor-
the structural analysis and, consequently, the damage and loss mance metrics) to suggest improved performance objectives.
analysis. • Applying different multi-criteria decision-making (MCDM) meth-
• There are many inherent uncertainties in assessing building per- ods to select the desired option from Pareto-optimal designs based
formance in future earthquakes. The present study employs a on the desired goals and then comparing the selected designs
state-of-the-art technique, Monte Carlo simulation as recommen- based on various performance metrics. In this regard, the most
ded by FEMA P-58, to apply the effect of these uncertainties, as widely used MCDM approaches, including Shannon entropy, Eu-
much as possible, on the probable building performance. How- lerian distance (LINMAP, TOPSIS, deviation index), fuzzy mem-
ever, the dispersion values presented in the FEMA P-58 guideline bership function, and evidential reasoning can be examined (see
for use in analysis procedures are only general values that have [60,61] for more details on MCDM methods).
been proposed based on past research studies and, of course, with
the numerous quality assurance measures. Dispersion values for 8. Conclusions
the practical engineering design of a real structure can be more
or less than the values considered in this research. This paper presents in detail the implementation of a comprehensive
• The results obtained in this research are largely dependent on performance assessment process, called the FEMA-58 methodology,
the vulnerable components considered in the building perfor- within a multi-objective structural optimization problem. A set of three
mance model as well as the quantity assumed for each of them. steel buildings with perimeter special MRF system and office occupancy
In addition, the population model is also effective in terms of is selected as a case study. For each of the building archetypes, three
consequences such as fatalities and injuries. In the present study, design alternatives including designs with the minimum: initial con-
these are determined based on two electronic tools provided as struction cost (Design-A), average annual repair cost (Design-C), and
Volume 3 of FEMA P-58 [41] (Normative Quantity Estimation the sum of the initial cost and NPV of average annual loss (Design-B)
Tool and Fragility Database) as well as population models for are selected from the set of design alternatives obtained as the Pareto
common building occupancies recommended in Appendix E of front. These are compared with each other as well as with other feasible
Volume 1 [6]. All of these are still in development. Moreover, the options obtained for the building in terms of various performance
quantities assumed for vulnerable components and the number of measures at several seismic intensities.
building occupants are just general values obtained based on the The main findings of this study can be summarized as follows:

20
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

• In the 3-story building, selecting an optimal design with the Declaration of competing interest
minimum initial cost (Design-A) for investment can render the
building irreparable when subjected to earthquakes with high The authors declare that they have no known competing finan-
intensities (here seismic events with less than 3% probability of cial interests or personal relationships that could have appeared to
exceedance (POE) in 50 years). Therefore, in case of such seismic influence the work reported in this paper.
events, significant direct and indirect financial losses would be
imposed on the project due to the need to replace the building References
with a new one. In contrast, for 6- and 9-story buildings, if design
A is selected, the building can still be repaired even when it [1] FEMA. Building the performance you need: A guide to state-of-the-art tools
is subjected to an earthquake with the highest assumed shaking for seismic design and assessment: FEMA P-58-7. Washington, D.C: Federal
intensity (1% POE in 50 years). Emergency Management Agency; 2018.
• In general, the range of variations in all performance metrics [2] FEMA. Multi-hazard loss estimation methodology-earthquake model: HAZUS-
MH-2.1, technical manual. Washington, D.C: Federal Emergency Management
(between the various design alternatives) decreases as the height
Agency; 2013.
of the building increases. [3] ATC. Earthquake damage evaluation data for california: ATC-13. Redwood City,
• The probability of causing an unsafe placard at different shaking California: Applied Technology Council; 1985.
intensities for the building design options decreases with increas- [4] FEMA. A benefit-cost model for the seismic rehabilitation of buildings-volume 1:
ing the building height. For instance, the probability of assigning a user’s manual: FEMA-227. Washington, D.C: Federal Emergency Management
Agency; 1992.
an unsafe placard to the 3-, 6-, and 9-story buildings under the
[5] ASCE/SEI 41-17. Seismic evaluation and retrofit of existing buildings. Reston,
maximum considered earthquake (MCE) (here intensity No. 7) is
Virginia: American Society of Civil Engineers; 2017.
on average about 80%, 65%, and 53%, respectively. [6] FEMA. Seismic performance assessment of buildings: volume 1 – methodology:
• Unlike the performance metrics of repair time and environmental FEMA P-58-1. Washington, D.C: Federal Emergency Management Agency; 2018.
impacts, which are directly related to the building repair cost, [7] FEMA. Seismic performance assessment of buildings: volume 2 – implementation
two metrics of the injury and fatality rates cannot necessarily be guide: FEMA P-58-2. Washington, D.C: Federal Emergency Management Agency;
2018.
considered corresponding to the building repair cost. Therefore,
[8] Jarrett JA, Judd JP, Charney FA. Comparative evaluation of innovative and
between the two design options, one that has more repair cost traditional seismic-resisting systems using the FEMA P-58 procedure. J Constr
may at the same time have a lower fatality or injury rate. Con- Steel Res 2015;105:107–18. http://dx.doi.org/10.1016/j.jcsr.2014.10.001.
sequently, optimizing (minimizing) the building repair cost does [9] Basim MC, Estekanchi HE. Application of endurance time method in
not necessarily mean minimizing the number of casualties. On the performance-based optimum design of structures. Struct Saf 2015;56:52–67.
other hand, since a significant amount of the fatality rate as well http://dx.doi.org/10.1016/j.strusafe.2015.05.005.
[10] Zeng X, Lu X, Yang TY, Xu Z. Application of the FEMA-P58 methodology
as a portion of the injury rate is due to the building collapse,
for regional earthquake loss prediction. Nat Hazards 2016;83(1):177–92. http:
and the values of these two metrics correspond to the collapse //dx.doi.org/10.1007/s11069-016-2307-z.
capacity of the building, so it can be concluded that there is not [11] Judd JP, Pakwan N. Seismic performance of steel moment frame office
necessarily a direct relationship between the building repair cost buildings with square concrete-filled steel tube gravity columns. Eng Struct
and its collapse capacity. 2018;172(10):41–54. http://dx.doi.org/10.1016/j.engstruct.2018.06.016.
• In general, there is not a considerable difference between the [12] Terzic V, Kolozvari K, Saldana D. Implications of modeling approaches on seismic
performance of low- and mid-rise office and hospital shear wall buildings. Eng
design alternatives A, B, and C in terms of injury rate for all Struct 2019;189(3):129–46. http://dx.doi.org/10.1016/j.engstruct.2019.03.033.
archetypes. However, in terms of fatality rate, there is a signif- [13] Güner T, Topkaya C. Performance comparison of BRBFs designed using different
icant difference between design C and designs A and B, so that, response modification factors. Eng Struct 2020;225(5):111281. http://dx.doi.org/
unlike designs A and B, the fatality rate for design C is close to 10.1016/j.engstruct.2020.111281.
zero. [14] Nobahar E, Asgarian B, Mercan O, Soroushian S. A post-tensioned self-centering
yielding brace system: Development and performance-based seismic analysis.
• A significant portion of the building repair costs under shaking
Struct Infrast Eng 2021;17(3):392–412. http://dx.doi.org/10.1080/15732479.
intensity No. 2 (about 29% POE in 50 years) is allocated to 2020.1752262.
drift-sensitive nonstructural components (such as interior wall [15] Applied Technology Council. Performance assessment calculation tool (PACT).
partitions and exterior curtain walls) (on average about 71%), Washington, D.C: Federal Emergency Management Agency (FEMA); 2018.
and structural components do not contribute to the building [16] Papadopoulos AN, Vamvatsikos D, Kazantzi AK. Development and application of
repair costs in this shaking intensity. By increasing the initial FEMA P-58 compatible story loss functions. Earthq Spectr 2019;35(1):95–112.
http://dx.doi.org/10.1193/102417EQS222M.
cost of MRFs under intensity No. 2, the contribution of drift-
[17] Kaveh A. Applications of metaheuristic optimization algorithms in civil engineer-
sensitive nonstructural components to the building repair costs ing. Cham: Springer International Publishing; 2017, http://dx.doi.org/10.1007/
decreases, while the contribution of acceleration- and velocity- 978-3-319-48012-1.
sensitive nonstructural components increases. [18] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective
• With increasing shaking intensity from No. 2 (about 29% POE genetic algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6(2):182–97. http:
in 50 years) to No. 6 (about 3% POE in 50 years), the con- //dx.doi.org/10.1109/4235.996017.
[19] Coello C, Pulido GT, Lechuga MS. Handling multiple objectives with particle
tribution of drift- and velocity-sensitive nonstructural compo-
swarm optimization. IEEE Trans Evol Comput 2004;8(3):256–79. http://dx.doi.
nents to the repair costs decreases (on average from 71% and org/10.1109/TEVC.2004.826067.
16% to 55% and 4%, respectively), while the contribution of [20] Zhang Q, Li H. MOEA/D: A multiobjective evolutionary algorithm based on
structural and acceleration-sensitive nonstructural components decomposition. IEEE Trans Evol Comput 2007;11(6):712–31. http://dx.doi.org/
increases (on average from 0% and 13% to 16% and 25%, respec- 10.1109/TEVC.2007.892759.
tively). Furthermore, with the increase in the initial cost of MRFs, [21] Rojas HA, Foley C, Pezeshk S. Risk-based seismic design for optimal structural
and nonstructural system performance. Earthq Spectr 2011;27(3):857–80. http:
the contribution of structural components in the building repair
//dx.doi.org/10.1193/1.3609877.
cost at intensity No. 6 reduces, and instead, the contribution of [22] Saadat S, Camp CV, Pezeshk S. Seismic performance-based design optimiza-
acceleration-sensitive nonstructural components in this intensity tion considering direct economic loss and direct social loss. Eng Struct
increases. 2014;76:193–201. http://dx.doi.org/10.1016/j.engstruct.2014.07.008.
• Comparison of performance metrics obtained from simplified [23] Saadat S, Camp CV, Pezeshk S. Probabilistic seismic loss analysis for the design
analysis and nonlinear response history analysis indicates that, of steel structures: Optimizing for multiple-objective functions. Earthq Spectr
2016;32(3):1587–605. http://dx.doi.org/10.1193/080513EQS223M.
in general, the simplified analysis has acceptable accuracy in
[24] Ghasemof A, Mirtaheri M, Karami Mohammadi R, Mashayekhi MR. Multi-
predicting different performance metrics for all three archetypes. objective optimal design of steel MRF buildings based on life-cycle cost using
However, this procedure is more accurate for the middle design a swift algorithm. Structures 2021;34(3):4041–59. http://dx.doi.org/10.1016/j.
option (Design-B) than other alternatives. istruc.2021.09.088.

21
A. Ghasemof et al. Engineering Structures 254 (2022) 113856

[25] Mitropoulou CC, Lagaros ND, Papadrakakis M. Life-cycle cost assessment of op- [44] ANSI/AISC 341-16. Seismic provisions for structural steel buildings. Chicago,
timally designed reinforced concrete buildings under seismic actions. Reliab Eng Illinois, USA: American Institute of Steel Construction; 2016.
Syst Saf 2011;96(10):1311–31. http://dx.doi.org/10.1016/j.ress.2011.04.002. [45] AISC. Steel construction manual. 15th ed.. Chicago, Illinois: American Institute
[26] Mousazadeh M, Pourreza F, Basim MC, Chenaghlou MR. An efficient approach of Steel Construction; 2017.
for LCC-based optimum design of lead-rubber base isolation system via FFD [46] Ibarra LF, Krawinkler H. Global collapse of frame structures under seismic
and analysis of variance (ANOVA). Bull Earthq Eng 2020;18(4):1805–27. http: excitations. Tech. Rep. No. 152, John A. Blume Earthquake Engineering Center;
//dx.doi.org/10.1007/s10518-019-00754-6. 2005.
[27] Asadi P, Hajirasouliha I. A practical methodology for optimum seismic design of [47] Lignos DG, Krawinkler H. Deterioration modeling of steel components in support
RC frames for minimum damage and life-cycle cost. Eng Struct 2020;202:109896. of collapse prediction of steel moment frames under earthquake loading. J Struct
http://dx.doi.org/10.1016/j.engstruct.2019.109896. Eng 2011;137(11):1291–302. http://dx.doi.org/10.1061/{%}28ASCE{%}29ST.
[28] Gholizadeh S, Baghchevan A. Multi-objective seismic design optimization of steel 1943-541X.0000376.
frames by a chaotic meta-heuristic algorithm. Eng Comput 2017;33(4):1045–60. [48] Lignos DG, Hartloper AR, Elkady A, Deierlein GG, Hamburger R. Pro-
http://dx.doi.org/10.1007/s00366-017-0515-0. posed updates to the ASCE 41 nonlinear modeling parameters for wide-
[29] de Domenico D, Qiao H, Wang Q, Zhu Z, Marano G. Optimal design and flange steel columns in support of performance-based seismic engineering. J
seismic performance of multi–tuned mass damper inerter (MTMDI) applied to Struct Eng 2019;145(9):04019083. http://dx.doi.org/10.1061/(ASCE)ST.1943-
adjacent high–rise buildings. Struct Design Tall Special Build 2020;2014(12):11. 541X.0002353.
http://dx.doi.org/10.1002/tal.1781. [49] The Gordian Group, Inc. RSMeans Data. 2019, https://www.rsmeans.com.
[30] Ghasemof A, Mirtaheri M, Karami Mohammadi R. A new swift algorithm for bi- [50] Hoffman GJ, Thal AE, Webb TS, Weir JD. Estimating performance time for
objective optimum design of steel moment frames. J Buil Eng 2021;39(5):102162. construction projects. J Manage Eng 2007;23(4):193–9. http://dx.doi.org/10.
http://dx.doi.org/10.1016/j.jobe.2021.102162. 1061/(ASCE)0742-597X(2007)23:4(193).
[31] de Domenico D, Hajirasouliha I. Multi-level performance-based design opti- [51] USEnvironmental Protection Agency. USEeio v1.1 - matrices. 2020, https://
misation of steel frames with nonlinear viscous dampers. Bull Earthq Eng catalog.data.gov/dataset/useeio-v1-1-matrices [Accessed 23 May 2021].
2021;44(3):1. http://dx.doi.org/10.1007/s10518-021-01152-7. [52] USGeological Survey (USGS). Unified hazard tool. 2020, available online at
[32] Sun F-F, Wang M, Nagarajaiah S. Multi-objective optimal design and seismic https://earthquake.usgs.gov/hazards/interactive/.
performance of negative stiffness damped outrigger structures considering damp- [53] FEMA. Quantification of building seismic performance factors: FEMA P695.
ing cost. Eng Struct 2021;229(2):111615. http://dx.doi.org/10.1016/j.engstruct. Washington, D.C: Federal Emergency Management Agency; 2009.
2020.111615. [54] Park HS, Hwang JW, Oh BK. Integrated analysis model for assessing CO2 emis-
[33] Kaveh A, Fahimi-Farzam M, Kalateh-Ahani M. Optimum design of steel frame sions, seismic performance, and costs of buildings through performance-based
structures considering construction cost and seismic damage. Smart Struct Syst optimal seismic design with sustainability. Energy Build 2018;158(2):761–75.
2015;16(1):1–26. http://dx.doi.org/10.12989/sss.2015.16.1.001. http://dx.doi.org/10.1016/j.enbuild.2017.10.070.
[34] Gholizadeh S, Fattahi F. Multi-objective design optimization of steel moment [55] Sharif SA, Hammad A. Simulation-based multi-objective optimization of institu-
frames considering seismic collapse safety. Eng Comput 2019;81(3):50. http: tional building renovation considering energy consumption, life-cycle cost and
//dx.doi.org/10.1007/s00366-019-00886-y. life-cycle assessment. J Buil Eng 2019;21(1):429–45. http://dx.doi.org/10.1016/
[35] Rezazadeh F, Talatahari S. Seismic energy-based design of BRB frames j.jobe.2018.11.006.
using multi-objective vibrating particles system optimization. Structures [56] Fathizadeh SF, Dehghani S, Yang TY, Noroozinejad Farsangi E, Vosoughi AR,
2020;24:227–39. http://dx.doi.org/10.1016/j.istruc.2020.01.006. Hajirasouliha I, et al. Trade-off Pareto optimum design of an innovative curved
[36] Tu X, He Z, Huang G. Performance-based multi-objective collaborative opti- damper truss moment frame considering structural and non-structural objectives.
mization of steel frames with fuse-oriented buckling-restrained braces. Struct Structures 2020;28(5):1338–53. http://dx.doi.org/10.1016/j.istruc.2020.09.060.
Multidiscip Optim 2020;61(1):365–79. http://dx.doi.org/10.1007/s00158-019- [57] Zhou Y, Cao S, Kosonen R, Hamdy M. Multi-objective optimisation of an
02366-9. interactive buildings-vehicles energy sharing network with high energy flex-
[37] Ghasemof A, Mirtaheri M, Karami Mohammadi R. Effects of demand parameters ibility using the Pareto archive NSGA-II algorithm. Energy Convers Manage
in the performance-based multi-objective optimum design of steel moment frame 2020;218(15):113017. http://dx.doi.org/10.1016/j.enconman.2020.113017.
buildings. Soil Dyn Earthq Eng 2022;153(1):107075. http://dx.doi.org/10.1016/ [58] Liu J, Yang H, Zhou Y. Peer-to-peer trading optimizations on net-zero energy
j.soildyn.2021.107075. communities with energy storage of hydrogen and battery vehicles. Appl Energy
[38] Mirfarhadi SA, Estekanchi HE, Sarcheshmehpour M. On optimal proportions 2021;302:117578. http://dx.doi.org/10.1016/j.apenergy.2021.117578.
of structural member cross-sections to achieve best seismic performance using [59] Song Lin. NGPM – A NSGA-II program in matlab v1.4. College of Astro-
value based seismic design approach. Eng Struct 2021;231(3):111751. http: nautics, Northwestern Polytechnical University, China; 2011, available on-
//dx.doi.org/10.1016/j.engstruct.2020.111751. line at https://www.mathworks.com/matlabcentral/fileexchange/31166-ngpm-
[39] Math Works, Inc. MATLAB. Math Works Inc.; 2018. a-nsga-ii-program-in-matlab-v1-4 [Accessed 23 May 2021].
[40] OpenSees. Open system for earthquake engineering simulation (OpenSees). Uni- [60] Jing R, Wang M, Zhang Z, Liu J, Liang H, Meng C, et al. Comparative study of
versity of California, Berkeley.: Pacific Earthquake Engineering Research Centre; posteriori decision-making methods when designing building integrated energy
2016, https://opensees.berkeley.edu. systems with multi-objectives. Energy Build 2019;194(12):123–39. http://dx.doi.
[41] FEMA. Seismic performance assessment of buildings: volume 3 - supporting org/10.1016/j.enbuild.2019.04.023.
electronic materials and background documentation: FEMA P-58-3. 3rd ed.. [61] Weerasuriya AU, Zhang X, Wang J, Lu B, Tse KT, Liu C-H. Performance evalua-
Washington, D.C: Federal Emergency Management Agency; 2018. tion of population-based metaheuristic algorithms and decision-making for multi-
[42] ASCE/SEI 7-16. Minimum design loads and associated criteria for buildings and objective optimization of building design. Build Environ 2021;198(1364):107855.
other structures. Reston, Virginia: American Society of Civil Engineers; 2017. http://dx.doi.org/10.1016/j.buildenv.2021.107855.
[43] ANSI/AISC 360-16. Specification for structural steel buildings. Chicago, Illinois,
USA: American Institute of Steel Construction; 2016.

22