Engineering Structures 185 (2019) 15–25

Contents lists available at ScienceDirect

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

Simplified performance-based optimal seismic design of reinforced concrete T

frame buildings
⁎
Chunyu Zhang, Ying Tian
Department of Civil and Environmental Engineering and Construction, University of Nevada, Las Vegas, 4505 S. Maryland Parkway, Las Vegas, NV 89154, USA

A R T I C LE I N FO A B S T R A C T

Keywords: This paper presents a simplified performance-based optimal seismic design approach for multi-story reinforced
Optimal design concrete moment frames. The proposed approach minimizes construction cost and takes member plastic rotation
Cost and optionally inter-story drift as optimization constraints. Other seismic design requirements reflecting suc-
Performance-based design cessful design practice are also incorporated. Simplification is made by reducing design variables into two, one
Reinforced concrete
for overall system stiffness and the other for overall system strength. The optimization contains two stages, the
Frame
Inelastic spectrum
determination of feasible region boundary in strength and stiffness domain and optimization in material con-
Nonlinear static analysis sumption domain. Capacity spectrum method, which jointly considers nonlinear static analysis and inelastic
design spectrum, is used to estimate the global and local deformation demands at peak dynamic response. The
proposed optimization approach is applied to the design of a six-story reinforced concrete frame. The design
results indicate that 30% of needed flexural strength and 26% of cross-sectional area can be reduced from the
initial strength-based design. Nonlinear time-history analyses are conducted on the optimized structure using ten
historical ground motions scaled to represent three levels of seismic hazard.

1. Introduction the conventional strength-based design, considering only one level of

As a new generation of structural design methodology, perfor-
mance-based seismic design (PBSD) has been a major focus of earth-
quake engineering community. PBSD requires that a structure meet all
target structural performance under various considered seismic ha-
zards. The structural performance is normally measured by local and/or
global deformations associated with the extent of damage to structural
and nonstructural components as well as economical loss. PBSD permits
designing strength, stiffness, and ductility of a structural system and its
components in a coordinated manner, which is difficult to achieve
through the conventional strength-based design.
PBSD is codified for the U.S. design practice in ASCE 41-13 [1],
which evolved from FEMA 273 [2] and FEMA 356 [3]. With the aid of
these guidelines, PBSD has been widely employed to evaluate the
seismic performance of existing buildings, for which design details such
as section size and reinforcement arrangement are already known.
However, the strength-based design approach is still prevailing for
designing new buildings. One challenge hindering the adoption of PBSD
to new building designs is the lack of general design procedures [4] that
can systematically handle the numerous design variables affecting the
seismic performance of a structure. Even though PBSD has been occa-
sionally used for new building designs, the typical practice starts from
seismic hazard, and then check the structural performance under all
considered levels of hazard by means of structural analysis. If needed,
the strength-based design results are modified and examined by struc-
tural analysis again. Because this trial-and-error design process involves
high design costs, PBSD has been applied mainly to critical facilities
such as hospital buildings.
The need of considering multiple levels of seismic hazard and the
numerous design variables naturally calls for developing optimal PBSD
procedures. Past studies focusing on this topic varied dramatically in
choosing objective function, performance measurements and con-
straints, type of structural analysis, and optimization algorithm [5].
Minimizing the initial construction cost was the most commonly
adopted design objective [6–17]; however, some studies [18–21] at-
tempted to minimize life-cycle cost so that other costs, such as eco-
nomic loss and expense in post-earthquake repair, were also in-
corporated. Other types of objective functions were intended to
minimize damage at plastic hinge regions [22], minimize floor accel-
eration [23,24], or maximize eigen frequency [25]. To measure struc-
tural performance and define constraints, majority of the existing op-
timization methods employed inter-story drift, while only a few studies
[6,8,18] considered the plastic hinge rotation of beams and columns.
In general, a considerably large amount of nonlinear time-history

⁎
Corresponding author.
E-mail address: ying.tian@unlv.edu (Y. Tian).

https://doi.org/10.1016/j.engstruct.2019.01.108
Received 3 October 2018; Received in revised form 21 January 2019; Accepted 23 January 2019
Available online 01 February 2019
0141-0296/ Published by Elsevier Ltd.
C. Zhang, Y. Tian
analyses or nonlinear static analyses are needed during optimization to
identify the seismic responses of a structure with different values of
design variables. Time-history analysis better identifies seismic re-
sponse if the hysteretic behavior of structural components is properly
defined. However, the structural response predicted by time-history
analysis cannot be mathematically formulated as a function of system
stiffness and strength, especially when material nonlinearity and geo-
metric nonlinearity present. It is thus difficult to explicitly correlate the
decision variables in the objective function of an optimization problem
with the optimization constraints defined by structural performance
such as inter-story drift and plastic hinge rotation. As a result, various
metaheuristic algorithms, including genetic [12,13], neural network
[14,19], discrete gravitational search [20], and generalized pattern
search [24] algorithms, were used to perform optimizations. This ca-
tegory of algorithm, such as genetic [9], ant colony [16], and particle
swarm [17] algorithms, was also used together with nonlinear static
analysis to create an optimization procedure. On the other hand, non-
linear static analysis is better suited for mathematically formulating a
performance-based optimal design. The joint use of optimality criteria
method and nonlinear static analyses enabled explicitly formulating an
optimization algorithm in some studies [7,8,10,11,15,18] after a series
of idealization of structural behaviors. The methods suggested in these
studies optimize section size first based on inter-story drift limit under
minor earthquakes and then optimize the flexural design of components
based on the target structural performance under major earthquakes.
Most existing optimal PBSDs involve complex algorithms requiring
a fully automated design process. However, the needed computer pro-
grams are often not accessible for practicing engineers. Moreover, the
optimization methods developed specifically for steel structures are not
necessarily applicable to RC structures. This paper presents a simplified
version of optimal PBSD for multistory RC frame buildings using a new
optimization algorithm. The proposed method reduces the design
variables into only two, one for overall stiffness and the other for
overall strength, so that the optimization process can be implemented
without the aid of special optimization software. Moreover, this study
considers not only inter-story drift but also plastic hinge rotation, which
better describes the local behavior of a RC frame, to define target
structural performance. The study also incorporates the latest for-
mulations of plastic deformation capacity of RC beams and columns
recommended in ASCE 41-13 [1] based on extensive experimental data.
The proposed approach is applied to the design of a six-story RC frame
building to illustrate the optimization procedure and explore the degree
of construction cost saving. Finally, because the optimization method
uses nonlinear static analyses to estimate peak dynamic deformation
demand, the performance of the optimized structure is examined by a
suite of nonlinear dynamic analyses.
2. Statement of optimal PBSD
According to the capacity design philosophy for ductile RC frame
structures [26], the behavior of beams and columns shall be governed
by flexure with sufficient inelastic deformation capacity. The flexural
strength of a RC frame element is the design basis for force-controlled
actions such as shear. For instance, the shear demand of a beam needs
to be determined using the probable moments evaluated based on the
actual flexural design results [27]. Accordingly, this study focuses on
the optimal flexural design of RC frames with an aim to minimize the
overall cost associated with concrete and flexural reinforcement con-
sumed by the beams and columns. Moreover, the proposed optimal
design considers multiple levels of seismic hazard, including occasional,
rare, and very rare earthquakes defined as the events with an ex-
ceedance rate of 50%, 10%, and 2% in 50 years, respectively [28]. Note
that, as the basic requirements, ASCE 41-13 [1] requires a structure to
meet the performance objectives under both rare and very rare earth-
quakes. Following this PBSD standard, the plastic hinge rotation of the
beams and the columns is employed as the major performance
16
Engineering Structures 185 (2019) 15–25
measurement and used to define optimization constraints. However, a
user can optionally choose inter-story drift ratio as a supplemental
performance index to restrain damage to nonstructural components.
The overall optimal PBSD problem can then be stated as
Minimize CT = cc Ωc + cs Ws
Subject to ∀ θ ≤ [θ]
∀ γ ≤ [γ ](optional) (1)
where CT is total cost; Ωc is the total volume of concrete; Ws is the total
weight of flexural reinforcement; cc and cs are the unit costs for concrete
and steel reinforcement, taking into account not only material but also
other parameters such as labor and transportation; θ is plastic hinge
rotation angle; γ is inter-story drift ratio; and [θ] and [γ] are allowable
values for θ and γ.
The [θ] and [γ] in Eq. (1) are defined based on the target perfor-
mance of the RC moment frame of concern. ASCE 41-13 [1] measures
structural performance only using plastic hinge rotation. However, if
inter-story drift is also employed as a performance index, the drift ratio
limit can be defined according to ASCE 41-06 [29] as [γ] = 1%, 2% and
4% for target performance levels of Immediate Occupancy (IO), Life
Safety (LS) and Collapse Prevention (CP), respectively. The plastic
hinge rotation capacity [θ] for beams and columns controlled by flexure
is given in Tables 1 and 2 [1], where ρ is tensile reinforcement ratio, ρ'
is compressive reinforcement ratio, ρbal is tensile reinforcement ratio
producing a balanced strain condition, b is beam width, d is effective
depth, f c' is concrete compressive strength, Vb is beam shear force, P is
column axial force, and Ag is column gross sectional area. The plastic
rotation capacity [θ] of flexure-dominated columns increases with de-
creased axial load P or increased transverse reinforcement ratio ρv.
However, if ρv ≥ 0.006, generally the case for a column designed fol-
lowing the modern seismic design codes, [θ] becomes independent to
ρv. Thus, [θ] for the columns is considered only as a function of axial
force level, as shown in Table 2.
3. Optimization strategy
Fig. 1 outlines the optimization procedure developed in this study.
The objective function in Eq. (1) incorporates the three major char-
acteristics (i.e., strength, stiffness and ductility) of an inelastic system
resisting seismic loads because Ωc affects elastic stiffness, Ws is asso-
ciated with member flexural strength, and γ and θ are related to duc-
tility. The constraints related to acceptable structural performance
normally define a convex feasible region for the two decision variables
Ωc and Ws. Eq. (1) is a nonlinear programming problem. The objective
function represents a group of lines with a slop of –cc/cs, one of which is
tangent to the feasible region boundary. The Ωc and Ws at the tangent
point (Point A in Fig. 1(a)) leads to the minimum total cost CT.
Despite the simple format of the objective function, there are several
challenges in obtaining the optimal solution. The first challenge is the
determination of peak deformation demand caused by the different
levels of seismic hazard. This study intends to formulate the peak de-
formation demand of a nonlinear RC frame system as a function of its
lateral stiffness and strength. This, however, cannot be achieved by
nonlinear time-history analyses. Accordingly, capacity spectrum
method [30,31], which employs nonlinear static analysis, is adopted to
Table 1
Beam plastic hinge rotation capacity [θ] (unit: rad.).
ρ − ρ' Vb Performance Level
ρbal
bd f c'
IO LS CP
≤0.0 ≤3 0.010 0.025 0.05
≥6 0.005 0.020 0.04
≥0.5 ≤3 0.005 0.020 0.03
≥6 0.005 0.015 0.02
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

Table 2 objective function is transformed as a function of θ and γ. The former

Column plastic hinge rotation capacity [θ] (unit: rad.). approach is considered. However, the deformation demands in terms of
P Performance level γ and θ for a multistory RC frame are difficult to be explicitly for-
Ag fc ' mulated as a function of Ωc and Ws. Moreover, as indicated in Tables 1
IO LS CP and 2, the plastic rotation capacity [θ] of a member is a function of
shear or axial force, which varies during lateral loading, implying that
≤0.1 0.005 0.045 0.06
≥0.6 0.003 0.009 0.010 [θ] cannot be predefined. The coupling effect between plastic hinge
rotation capacity and seismic demand complicates the development of
an optimization algorithm and was not considered in the past studies,
which assumed a fixed value of [θ]. A two-stage optimization approach
is considered herein to address these issues. In the first stage, strength-
based design procedure is followed to obtain an initial design for the RC
frame. Optimization, as shown in Fig. 1(b), is then performed to de-
termine a feasible region boundary defined by discretized pairs of a
stiffness parameter λ and a flexural strength parameter α normalized
based on the initial design. In the second stage, the feasible region
boundary determined in the λ–α domain is converted into that in the
Ωc–Ws domain, from which the optimization problem defined by Eq. (1)
is solved.

4. Additional design requirements

In addition to restraining local and global deformations, other de-

sign requirements for achieving desired seismic performance of a RC
frame are incorporated. These requirements are enforced throughout
the proposed optimal design and beneficial for reducing computational
cost.

(1) Beam-sway mechanism (Fig. 2) is preferred under large lateral de-

formations. For this purpose, the needed flexural strength of a
column is determined based on the maximum bending moment
demand on this column when the RC frame is laterally loaded to a
target roof displacement under a very rare earthquake. To further
reduce the likelihood of column yielding at a beam-column joint,
which may lead to a soft-story mechanism, the flexural design of
columns follows ΣMnc ≥ 1.2ΣMnb [27], where ΣMnc and ΣMnb are
the summations of the flexural strength of columns and beams
framing into the joint, respectively.
Fig. 1. Outline of optimization: (a) optimization in material consumption (2) Yielding in the first-floor columns is permitted at their bottom ends.
(Ωc–Ws) domain, and (b) determination of feasible region boundary in the re- However, prior to the exhaustion of column plastic deformation
lative stiffness and relative strength (λ–α) domain. capacity, sufficient number of beams shall have been engaged in
developing yielding and dissipating energy through plastic de-
predict the peak seismic response. The details of applying this method formations. As indicated in Tables 1 and 2, if a column carries a
are given later in Section 6.3. large axial force P, its plastic hinge rotation capacity [θ] is less than
The second challenge is that the peak nonlinear deformation de- that of a beam. Moreover, the reduction in column size due to op-
mand of a RC frame is affected by not only elastic stiffness but also timization decreases [θ]. It is therefore required that column hin-
flexural strength of the beams and columns. As indicated by Krawinkler ging at the supports occur later than hinging in the first-floor
and Seneviratna [32], the peak lateral displacement of an elastic-per- beams. Following the yielding of these beams, the inflection point
fectly plastic SDOF system due to ground excitation differs from that of of a first-floor column shifts up and is assumed to be located at 3/4
a purely elastic system. The difference is a function of the fundamental of column height measured from the support, as shown in Fig. 2.
period T, the strength reduction factor R, and the ductility ratio μ. These
properties are correlated through a R–μ–T relationship [33,34]. It fol-
lows that the optimal member size, which controls system elastic
stiffness, cannot be solely determined from a target lateral displacement
without considering the flexural design of members. Accordingly, dif-
ferent from most of the past studies [7,8,10,11,15,18] that utilized
nonlinear static analysis, Ωc and Ws are not separately optimized in this
study.
The third challenge in obtaining an optimal solution is defining the
feasible region boundary. As indicated in Eq. (1), the objective function
is expressed by material consumptions Ωc and Ws, whereas the con-
straints are defined using plastic hinge rotation θ and optionally by
inter-story drift ratio γ. These two sets of variables are correlated by
structural performance predicted by structural analysis. To solve Eq.
(1), either the feasible region boundary is defined by Ωc and Ws or the
Fig. 2. Beam-sway mechanism and column flexural strength in the first floor.

17
C. Zhang, Y. Tian
The needed column flexural strength at the support, Mnc,2, can then
be derived as 1.5ΣMnb.
(3) The code-specified maximum reinforcement ratio of a member
section needs to be satisfied. Moreover, based on ACI 318-14 [27],
the positive moment strength of a beam section in the plastic hinge
regions shall be at least 50% of the negative moment resistance.
5. Simplifications
The optimal design of a structural system is ultimately represented
by the stiffness and strength properties of each component. For con-
venience of formulation, two dimensionless design variables, a relative
flexural stiffness factor denoted as λ and a relative flexural strength
factor denoted as α, are defined for each member. The relative stiffness
factor λ is defined as the ratio of section moment of inertia varying
during optimization to that determined from the initial design. The
relative strength factor α is defined as
Mn − MG
α=
Mn,0 − MG (2)
where Mn is member flexural strength varying during optimization, Mn,0
is the flexural strength determined from the initial design without
considering minimum reinforcement ratio requirement, and MG is the
bending moment caused only by the gravity loads considered in the
seismic design. If ASCE 7-10 [35] is followed, MG is determined based
on a gravity load combination of 1.2D + L (or 0.5L), where D = dead
load and L = live load.
Because many beams and columns exist in a RC frame, the total
number of design variables in terms of λ and α for the entire system far
exceeds the two decision variables Ωc and Ws in Eq. (1). To limit the
computational cost associated with optimization and enable expressing
the feasible region eventually using the two decision variables, this
study intends to reduce the number of design variables to two as well.
For this purpose, it is required that both the ratio of cross-sectional
dimension and the ratio of flexural strength among different members
determined from the initial design be maintained during optimization.
In other words, the relative stiffness factor λ and the relative strength
factor α are uniformly applied to all the RC frame members at any step
of optimization and function as the overall indices measuring stiffness
and strength of the entire system. With such a simplification, the ad-
ditional design requirements described previously, regarding strong-
column/weak-beam and positive flexural strength of a beam section,
are not necessary to be considered as optimization constraints, because
they are automatically satisfied once they have been enforced in the
initial design.
A series of values for α are selected and λ is then optimized for each
α using the algorithm described in Section 6.4 to determine the feasible
region boundary in the λ–α domain shown in Fig. 1(b). λ = 1 and
α = 1 correspond to the sectional stiffness and strength from the initial
design, respectively. The parameter λ correlates concrete volume Ωc;
the parameter α can be translated into the consumption of flexural
reinforcement Ws and indicates the degree of reducing needed member
flexural strength from that determined from the initial design. A RC
frame shall be designed also based on the load combinations purely for
gravity loading. A relative strength factor αG corresponding to gravity
design is defined accordingly and taken as the lower bound of α. αG is
evaluated using Eq. (2) by replacing Mn with MG,0 (MG,0 > MG), the
bending moment demand due to the load combinations for gravity
design (i.e., 1.2D + 1.6L per ASCE 7-10 [35]).
Due to the above simplifications, the design result based on the
proposed approach would not be a global optimal solution. However, in
a seismic design of RC frame, the member sizes are typically the same in
every two to three stories. Moreover, the beam flexural reinforcement is
normally identical on each side of a beam-column joint. All these fea-
tures can be incorporated in the initial design. Thus, even if the factors
α and λ are used to uniformly modify the initial design for
18
Engineering Structures 185 (2019) 15–25
simplification purpose, the optimal design result would be close to the
optimal solution that can reflect design practice.
6. Determination of feasible region boundary
6.1. Overview
The key to solving Eq. (1) is the determination of feasible region
boundary, which is obtained first in the λ–α domain. A point situated
on the feasible region boundary shown in Fig. 1(b) can be interpreted as
either the minimal λ satisfying performance criteria at a given lateral
strength of the system or the minimal α at a given stiffness. The former
definition is considered to formulate an iterative procedure used to
determine the feasible region boundary. A set of discrete α values
ranging from αG to 1 are selected. For each α, the flexural strength of a
member, Mn, is modified from the initial design based on Eq. (2). Note
that because optimizing λ reduces section size and thus tends to in-
crease flexural reinforcement ratio, the code-specified minimum re-
inforcement requirement is not implemented when evaluating Mn,0 in
Eq. (2) for the initial design. For the RC frame with a specific value of α,
λ is minimized for each level of earthquake hazard using a procedure
described in Section 6.4; the controlling value of λ gives the optimal λ
for this α. This procedure is repeated for all selected α values so that a
piecewise linear feasible region boundary is defined in the λ–α domain.
The height and effective depth of a beam are denoted in the fol-
lowing discussions as h and d, respectively. Their values from the initial
design are designated as h0 and d0. When minimizing λ at a given value
of the relative strength factor α, the cross-sectional area of frame
members is reduced. Accordingly, in addition to limiting plastic hinge
rotation and inter-story drift ratio, the maximum reinforcement ratio
permitted by design codes, ρmax, needs to be considered as a constraint.
The flexural capacity of a beam can be expressed as
Mn = ωbd 2f c' (1 − 0.59ω) (3)
where is concrete compressive strength; b is beam width; and ω is a
f c'
tensile reinforcement index calculated as ω = ρf y / f c' with fy being the
yield strength of reinforcement. Approximating d/d0 as h/h0, the
minimum λ corresponding to ρmax can be derived as
4
3
⎧ ⎛ MG ⎞ MG ⎤ ⎛ ω0 ⎞ ⎛ 1 − 0.59ω0 ⎞ ⎫
λmin,1 = ⎡α ⎜1 − ⎟ +
⎜ ⎟⎜ ⎟
⎨⎢ ⎝ Mn,0 ⎠ M ⎥
n,0 ⎦ ⎝ ωmax ⎠ ⎝ 1 − 0.59ωmax ⎠ ⎬ (4)
⎩⎣ ⎭
where ω0 and ωmax are beam tensile reinforcement index according the
initial design and ρmax, respectively. The flexural stiffness of a beam also
needs to satisfy the deflection serviceability requirement under gravity
loading. Denoting the minimum relative stiffness needed for this pur-
pose as λmin,2, the lower bound of λ is then defined as [λ] = max
{λmin,1, λmin,2}.
The first-stage optimization problem that minimizes the relative
stiffness factor λ at a given relative strength factor α can then be stated
as Eq. (5). Section 6.4 presents the detailed algorithm for solving this
optimization problem. Because the flexural strength of a column in-
teracts with its axial force, which varies during lateral loading, [λ]
corresponding to ρmax for columns cannot be predefined and needs to be
checked during optimization. However, the optimal design for λ may
not be affected by ρmax for the columns due to its high permitted value;
according to ACI 318-14 [27], the maximum column reinforcement
ratio evaluated using all longitudinal bars and the entire cross section
area is 4% and 8% in the regions with and without splicing, respec-
tively. Once the optimal result of λ becomes available, the section size
of each frame member and thus the total volume of concrete Ωc can be
determined based on the definition of λ. The optimization problem Eq.
(5) is implemented to all selected values of α (αG ≤ α ≤ 1) for each
level of seismic hazard. The data sets of α and optimized λ constitute
the feasible region boundary in the λ–α domain.
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

Minimize λ
Subjectto ∀ θ ≤ [θ]
∀ γ ≤ [γ ](optional)
λ ≥ [λ] (5)

Each pair of α and the optimized λ provides information regarding

the needed flexural strength and section size, from which flexural de-
signs are conducted for all the beams and columns. The flexural design
may be governed by the minimum reinforcement ratio specified in
design codes especially due to the use of relatively large section size to
obtain needed stiffness for RC frame buildings located in moderate-to-
high seismic regions. Based on the flexural design of each component,
the total weight of reinforcing steel Ws is evaluated. Because each pair
of α and the optimized λ corresponds to a pair of Ws and Ωc, the feasible
region boundary in the λ–α domain shown in Fig. 1(b) is transformed
into that in the Ωc–Ws domain shown in Fig. 1(a), where the second
stage of optimization defined by Eq. (1) can be carried out. Fig. 3. Constructing based shear vs. roof displacement response based on
nonlinear static analysis result of the structure without stiffness modification
6.2. Load-deformation response due to modified flexural stiffness (λ = 1).

To solve Eq. (5) at each selected α (αG ≤ α ≤ 1), the peak seismic
response of RC frame in terms of plastic hinge rotation θ and inter-story
drift γ need to be determined for each level of seismic hazard. This is
achieved using capacity spectrum method [30,31]. For the RC frame
with a specific value of α but without stiffness modification from the
initial design (i.e., λ = 1), nonlinear static analysis is conducted. The
analysis provides information regarding base shear, roof displacement,
inter-story drift ratio, and plastic hinge rotation. The beams and col-
umns are modeled by line elements having concentrated plasticity at
ends (plastic hinges). The sections outside the plastic hinge regions are
linear elastic with a flexural stiffness taking into account the effects of
concrete cracking. Following gravity loading, lateral loads corre-
sponding to the first vibration mode are applied. However, if the var-
iation of inertia force distribution due to inelastic response is con-
sidered, an adaptive lateral load pattern [32,36] accounting for the
effects of higher modes and member yielding can be used to more ac-
curately capture the structural response by means of nonlinear static
analyses.
The lateral loading response of the structure with λ = 1 evaluated
by a nonlinear static analysis consists of the generation of a series of
plastic hinges. Even if the occurrence of each plastic hinging modifies
the system stiffness matrix, it remains identical between two sub-
sequent hinging. Lateral loading causing the jth plastic hinge is taken as
loading step j and designated by a subscript within parenthesis in the
following discussions. Denoting the system stiffness matrix during
loading step j as K(j), the increase in horizontal displacement of the
system Δu(j) and the increase in lateral load ΔF(j) satisfies
ΔF(j) = K(j)Δu(j). Thus, the total displacement u(j) at the completion of
the jth loading step under load F(j) is
j system (Fig. 4(b)) is converted into an acceleration-displacement (A–D)
{u}(j) = ∑ K (−k1) ΔF(k)
k=1 (6)
When the relative stiffness factor λ is applied uniformly to all frame
members to reduce their flexural stiffness, the stiffness matrix of the
new system during the jth loading step can be expressed as K (λj) = λK (j) .
Additionally, at a fixed value of α, modifying λ has no impact on load
increase ΔF(λj) needed to generate a new plastic hinge, that is,
ΔF(λj) = ΔF(j) . Thus, the lateral displacement of the system with modified
stiffness can be derived as
j
1 under a specific level of seismic hazard is represented by an inelastic
u (λj) = ∑ [K (λk) ]−1ΔF(λk) = u (j)
k=1
λ (7)
Accordingly, under the same lateral load F(j) before causing a col-
lapse mechanism, the roof displacement δ, inter-story drift ratio γ, and
plastic hinge rotation angle θ of the system modified by λ must satisfy
δ γ θ 1
= = =
δ0 γ0 θ0 λ (8)
where δ0, γ0, and θ0 are the roof displacement, the inter-story drift
ratio, and the plastic rotation angle of the structure without stiffness
modification (i.e., λ = 1).
Eq. (8) is of significant convenience for solving the first-stage op-
timization problem defined in Eq. (5). As shown in Fig. 3, once the base
shear vs. roof displacement (V-δ) response of a frame system at a given
value of α but without stiffness reduction (λ = 1) becomes available, it
can be used to directly construct a nonlinear V-δ response for the system
with modified flexural stiffness. Accordingly, there is no need to per-
form extra structural analyses during the process of searching for op-
timal λ for the structure at the selected α. Thus, the total number of
pushover analyses needed to complete the proposed optimal seismic
design is identical to the number of discretized α values (αG ≤ α ≤ 1)
chosen to define a feasible region boundary.
6.3. Estimation of peak dynamic response at different seismic hazards
For each selected relative strength factor α, the peak lateral de-
formation response of the structure without stiffness optimization
(λ = 1) due to each level of seismic hazard is determined using capacity
spectrum method [30,31]. In this method, as shown in Fig. 4(a), the
lumped masses at different stories, roof displacement δ and base shear V
in the MDOF system are transformed into mass M, displacement D and
lateral force F in an equivalent inelastic SDOF system where spectral
concept applies. The ratio of F to M defines a pseudo-acceleration A.
The V–δ curve obtained from the nonlinear static analysis for the MDOF
curve in the equivalent SDOF system (Fig. 4(c)) and then idealized into
an elastic-perfectly plastic response based on energy absorption
equivalence. The idealized bilinear response is taken as a capacity curve
with a yield acceleration of Ay at a yield displacement of Dy, as shown
by the dashed lines in Fig. 4(c). According to the slop of elastic branch
of the capacity curve and the equivalent mass M, an effective period T
for the equivalent SDOF system can be determined.
Each level of seismic hazard is represented by the elastic spectral
acceleration Sae at 5% critical damping ratio evaluated using the ef-
fective period T. The seismic demand for an inelastic SDOF system
spectrum, also termed as demand curve. This curve, expressed by Eq.
(9), formulates the spectral displacement Sd as a function of Ay = Sae/R,
ductility ratio μ, and the effective period T. Among the three entities of
R, μ, and T, two are independent and various formulations for the
R–μ–T relationship [33,34] are available. The intersecting point of the

19
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

Fig. 5. Effects of modifying relative stiffness factor λ on capacity and demand

Fig. 4. Structural performance prediction using capacity spectrum method: (a)
MDOF–SDOF system transformation, (b) nonlinear static analysis and peak
displacement estimation in MDOF system, and (c) peak displacement estima-
tion in equivalent SDOF system.
demand and capacity curves, shown in Fig. 4(c), estimates the peak
displacement Dmax of the equivalent SDOF system under the considered
seismic hazard. If Sae < Ay, the structure in general behaves elastically
and μ = R = 1, which is the case for Hazard 1 shown in Fig. 4(c). Dmax
is then transformed back into the MDOF system as the peak roof dis-
placement δmax. The nonlinear analysis results at δmax provide peak
deformation demands in terms of plastic hinge rotation θ and inter-
story drift ratio γ.
μT 2 μT 2 Sae
Sd = Dmax = Ay = ⎛ ⎞
4π 2 4π 2 ⎝ R ⎠ (9)
curves.
different rates. If neither θ nor γ (if considered as a constraint) reaches
the allowable value during the previous iteration step i-1, a parameter
η(i), defined in Eq. (10) based on the deformation properties given in Eq.
(8), is considered to determine the relative stiffness factor λ(i) for the
next iteration step i. The term related to γ in Eq. (10) shall be removed if
inter-story drift ratio is not considered to define the optimization con-
straints in Eqs. (1) and (5). The parameter η(i) is used to ensure that the
deformation limit will not be exceeded at any location in the structure
during step i. The lower bound stiffness parameter [λ] also needs to be
considered. λ(i) defined by Eq. (11) is used in step i to uniformly reduce
the section stiffness of the frame members.
[θ](i − 1) [γ ]
η(i) = Min ⎧∀ (i − 1) ⋃ ∀ (i − 1) ⎫
⎨
⎩ θ γ ⎬
⎭ (10)

where R = Sae/Ay and μ = Sd/Dy. λ(i − 1) 1

λ(i) = = i
≥ [λ]
η(i) ∏m = 1 η(m) (11)
6.4. Determination of optimal stiffness at given flexural strength
Once λ(i) is determined, a new bilinear capacity curve in the
For each relative strength factor α (αG ≤ α ≤ 1), the application of equivalent SDOF system is created. Because λ(i) uniformly modifies
capacity spectrum method described previously estimates the peak member flexural stiffness in the MDOF system, it can be approximated
deformation demands. If the limiting values ([θ] and [γ]) are not ex- that λ(i) equally affects the lateral stiffness of the equivalent SDOF
ceeded under any seismic hazard, the relative stiffness factor λ is op- system. Its effective period is then determined as T(i) = T(0)/(λ(i))0.5. As
timized by reducing member section size until one of the constraints in shown in Fig. 5, due to decreased system stiffness, the demand curve
Eq. (5) controls. When the section size decreases, both reinforcement shifts to the right side, resulting in an increased spectral displacement.
ratio and beam shear stress increase, resulting in decreased beam Eq. (12) can be derived from Eqs. (8) and (9) and used to determine the
plastic hinge rotation capacity [θ], as indicated in Table 2. Due to the spectral displacement Dmax (i)
, story drift γ(i), and plastic hinge rotation θ(i)
in iteration step i.
interaction between λ and [θ], the optimal λ cannot be directly solved
and an iterative approach is therefore employed. (i)
Dmax γ (i) θ (i) 1 μ(i)
A superscript is used for the variables to denote iteration step. The (0)
= (0)
= (0) = (i) ⎜⎛ (0) ⎞⎟
Dmax γ θ λ ⎝μ ⎠ (12)
iterations, schematically illustrated in Fig. 5, start from λ(0) = 1. The
capacity spectrum method is applied to evaluate effective period T(0) where μ(i) is determined from the R-μ-T relationship. The strength re-
(0)
and peak displacement Dmax of the equivalent SDOF system as well as duction factor R(i) needed to apply the R-μ-T relationship at step i is
story drift ratio γ and plastic rotation angle θ(0) at the peak roof
(0)
(0) (i)
displacement δmax of the MDOF system. As stated previously and shown Sae
in Fig. 3, at the same value of α, the based shear vs. roof displacement R(i) =
Ay (13)
(V–δ) response of the structures having different values of λ can be
(i)
determined directly from that of the structure without stiffness mod- where is elastic spectral acceleration for the equivalent SDOF
Sae
ification (λ = 1). It follows that the idealized bilinear capacity curve for structure having an effective period of T(i).
λ(0) = 1 can be used to construct new bilinear capacity curves for Iterations with different values for λ(i) are performed until one of
λ(i) < 1 (i ≥ 1) in the subsequent iterations. Because member flexural the optimization variables in Eq. (5) reaches a limit. Reducing λ in-
strengths are not changed, the yield acceleration Ay for the capacity creases both γ and θ; meanwhile, [γ] is unchanged and [θ] is reduced
curves associated with different λ values can be approximated to be due to the increased reinforcement ratio and increased shear stress
identical, as shown in Fig. 5. caused by decreased section size. Accordingly, the peak deformation
Between two subsequent plastic hinging in the structure under in- demands in terms of θ(i) and γ(i) gradually approach the limiting values
creasing lateral deformation, the inter-story drift ratio γ and beam ([θ] and [γ]) after each iteration step and a converged result can be
plastic hinge rotation θ at different locations increase linearly but at obtained.

20
C. Zhang, Y. Tian
Fig. 6. Flow chart of the proposed optimal PBSD method.
7. Summary of optimization procedure
The procedure of the proposed optimal PBSD approach is outlined
in Fig. 6 and summarized as follows:
Step 1: A strength-based seismic design satisfying the requirements
given in Section 4 is conducted to determine the needed flexural
strengths of all members. The initial design results without con-
sidering minimum reinforcement ratio requirement (λ = 1 and
α = 1) are subjected to optimization.
Step 2: A series of relative strength factors (αG ≤ α < 1) are se-
lected. Each α is used to uniformly reduce the member flexural
strengths in the initial design to generate a new structure. Steps 3
and 4 are then implemented for each new structure. The ratio of
section dimension and the ratio of flexural strength among different
members are maintained throughout optimization.
Step 3: The relative stiffness factor is set as λ = 1 and a nonlinear
static analysis is conducted on the structure with a specific α. The
considered seismic hazards are represented by elastic spectral ac-
celerations. Capacity spectrum method is then applied to determine
the peak seismic deformation demands at different seismic hazards
for the structure with λ = 1.
Step 4: For the structure with the α value considered in Step 3, the
relative stiffness factor λ is optimized by solving Eq. (5) using the
numerical approach presented in Section 6.4.
Step 5: Based on the pairs of α and optimized λ determined from the
previous steps, a feasible region is defined in the λ–α domain. For
each pair of α and optimized λ, section flexural design is conducted
and the design code requirement for minimum flexural reinforce-
ment ratio is implemented. The total material consumptions Ωc and
Ws are evaluated based on the design results for each pair of α and
optimized λ. The feasible region in the λ–α domain is then ac-
cordingly converted into that in the Ωc–Ws domain.
Step 6: The Ωc and Ws that minimize total cost are determined by
solving Eq. (1). The values of λ and α corresponding to the optimal
Ωc and Ws provide information regarding the optimal section size
and flexural reinforcement design of each member.
21
Engineering Structures 185 (2019) 15–25
Fig. 7. A six-story RC frame for illustrating the proposed optimal PBSD.
8. Application of proposed optimization approach
8.1. Prototype RC frame building and initial design
The suggested optimal PBSD approach is applied to a four-bay, six-
story RC frame building shown in Fig. 7. The building, with a story
height of 3.66 m (12 ft.) and a span of 9.14 m (30 ft.) in both directions,
is located on a soft rock site in southern California, where the mapped
short-period and 1-sec spectral accelerations for very rare earthquake
are Ss = 1.50 g and S1 = 0.60 g, respectively. Concrete compressive
strength is assumed as 34.5 MPa (5000 psi) and reinforcement yield
strength as 414 MPa (60 ksi). The floors consist of 203 mm (8 in.) thick
two-way slabs. A dead load of D = 5.60 kN/m2 (117 psf) and a live load
of L = 0.958 kN/m2 (20 psf) act on the roof, whereas these values are
D = 5.75 kN/m2 (120 psf) and L = 2.39 kN/m2 (50 psf) on other floors.
The building is expected to meet a target performance of IO, LS, and CP
under occasional, rare, and very rare earthquakes, respectively.
Modal response spectrum analysis per ASCE 7-10 [35] is used to
perform a strength-based design. To meet inter-story drift limit, the
section size of beams is chosen as 559 mm (22 in.) × 813 mm (32 in.) in
the 1st to the 3rd floors and 508 mm (20 in.) × 711 mm (28 in.) in the
4th to the 6th floors; the columns are 737 mm (29 in.) square in the 1st
to the 3rd floors and 610 mm (24 in.) square in the 4th to the 6th floors.
The structure has a fundamental period of 0.81 sec. The flexural design
follows the seismic design provisions in ACI 318-14 [27]. Table 3 gives
the flexural resistances as well as reinforcement ratios (after enforcing
the minimum reinforcement ratio ρmin) for the beams and columns of an
interior frame based on the initial design. The column flexural re-
inforcement designs are all controlled by ρmin = 1%.
8.2. Structural optimization
In addition to restraining plastic hinge rotation, γ ≤ [γ] is also taken
as a design constrain to examine the effects of limiting inter-story drift
ratio on the optimal design. Open System for Earthquake Engineering
Simulation (OpenSees) [37] is used to conduct nonlinear static analyses
for optimization and nonlinear time-history analyses to examine the
dynamic response of the optimized structure. The beams and columns
are modeled using displacement-based line elements with zero-length
plastic hinges at the element ends. Backbone moment-rotation curves
are defined based on ASCE 41-13 [1] for the plastic hinges. The elastic
flexural stiffness of the beams and columns is defined as 30% and 50%
of uncracked sectional stiffness, respectively. The model suggested by
Ibarra et al. [38] is used to define the hysteretic behavior of plastic
hinges. The gravity loads applied on the beams include dead load plus
50% of design live load.
To apply Eq. (9) during optimization, Sae needs to be determined for
different levels of seismic hazard. Based on ASCE 7-05 [39], Sae is
formulated as a function of mapped spectral accelerations at short
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

Table 3
Flexural capacity and flexural reinforcement ratio of members in initial design (unit: kN-m for moment).
Floor Beam Column

Exterior bays Interior bays Exterior Interior

M− ρ− M+ ρ+ M− ρ− M+ ρ+ M ρ M ρ

Roof 416 0.49% 336 0.39% 509 0.61% 336 0.39% 475 1.0% 565 1.0%
5 592 0.72% 336 0.39% 625 0.75% 336 0.39% 565 1.0% 723 1.0%
4 638 0.77% 336 0.39% 684 0.83% 342 0.40% 622 1.0% 881 1.0%
3 764 0.62% 487 0.35% 803 0.66% 487 0.35% 1085 1.0% 1514 1.0%
2 790 0.64% 487 0.35% 816 0.66% 487 0.35% 1175 1.0% 1695 1.0%
1 736 0.59% 487 0.35% 776 0.62% 487 0.35% 1254 1.0% 1921 1.0%

period (Ss) and 1-sec periods (S1) as well as the site condition. Ss and S1
are 1.50 g and 0.60 g for very rare earthquakes (2%/50-yrs events) and
1.00 g and 0.377 g for rare earthquakes (10%/50-yrs events), respec-
tively. Ss and S1 for occasional earthquakes (50%/50-yrs events) are not
available from the seismic maps. Based on ASCE 41–06 [29] and a mean
return period of 72 years, Ss and S1 are evaluated as 0.436 g and 0.164 g
for occasional earthquakes. The R-μ-T relationship suggested by Vidic
et al. [34], which considers the effects of hysteretic behavior, is em-
ployed for Eq. (9). The peak ground motion acceleration and velocity at
a specified seismic hazard, needed for applying the R-μ-T relationship,
are determined based on the mapped data provided by U.S. Geological
Survey.
The optimization procedure described previously is followed. Eight
values for the relative strength factor, including α = αG = 0.28 and
α = 0.4–1 at an interval of 0.1, are selected to determine the feasible
region boundary in the λ–α domain. For each α, the optimal relative
stiffness factor λ is evaluated for the three considered levels of seismic
hazard and shown in Fig. 8. Because eight α values are preselected,
totally eight nonlinear static analyses are conducted. Moreover, for
each α, no more than seven iteration steps are used to obtain a con-
verged result of the optimal λ. For each selected α value, the numerical
procedures involved in performing a nonlinear pushover analysis,
constructing an idealized bilinear capacity (A-D) curve, and searching
the optimal stiffness parameter λ take 30–50 min for a personal com-
puter with a 16 GB RAM and a dual core 2.6 GHz CPU. The total
computer time used to complete the optimal PBSD design of the pro-
totype building is about 5 h. Note that this amount of time does not
include the time required to perform flexural designs of the critical
sections and evaluate material consumptions Ωc and Ws for each pair of
α and the corresponding optimal λ. These procedures can be compu-
terized but not coded into a computer program in this study.
For comparison purpose, Fig. 8 shows the optimized λ values
evaluated based on the individual types of constraint (i.e., plastic hinge
ration limit, drift limit, and minimum relative stiffness factor) for each
seismic hazard level. It is seen that the optimized λ is approximately
linear with respect to α except for inter-story drift due to occasional
earthquakes. In general, the optimized λ decreases with increased α for
each individual type of constraint. In other words, the flexural strength
and stiffness of the structural members are not independent to each
other in terms of the optimization results. As seen in Fig. 8, if no
strength reduction is considered (i.e., α = 1), inter-story drift limit for
LS under rare earthquakes governs the optimal relative stiffness factor
λ. In this case, λ is determined as 0.51, leading to a 29% reduction in
cross-sectional area. If inter-story drift is not taken as a constraint, the
optimal relative stiffness factor becomes λ = 0.40, controlled almost
identically by plastic hinge rotation limit for LS under rare earthquakes
or inter-story drift limit for IO under occasional earthquakes.
The pairs of selected α and the governing optimal λ based on all
constraints constitute the feasible region boundary in the λ–α domain
shown in Fig. 8. The optimal λ is controlled by either inter-story drift
ratio limit for Life Safety (LS) under rare earthquakes if 0.5 ≤ α ≤ 1 or
plastic-hinge rotation limit for Immediate Occupancy (IO) under occa-
sional earthquakes if 0.28 ≤ α < 0.5. Note that the inter-story drift
limit under occasional earthquake is far from controlling the feasible
region boundary. This result is remarkably different from some existing
optimal PBSD approaches for RC frames [7,8,15,18], which optimized
section size based only on the inter-story drift limit under minor
earthquakes. Moreover, deformation limits defined for CP under very
rare earthquakes also have no impact on the feasible region boundary.
For each pair of α and λ on the feasible region boundary shown in
Fig. 8, the needed section size is determined, and the flexural designs of
beams and columns are conducted. Then, the total weight of steel re-
inforcement Ws and total concrete volume Ωc for all the beams (in-
cluding those in the transverse direction) and the columns located be-
tween the slab center lines enclosing an interior frame are evaluated.

Fig. 8. Determination of feasible region boundary in the λ–α domain. Fig. 9. Feasible region boundary and optimal solution.

22
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

Design Spectrum
2.5
Individual Response Spectrum

Spectral Acceleration (g)

Mean Response Spectrum
2

1.5

0.5

0
0 0.4 0.8 1.2 1.6 2
Period (sec)

Fig. 10. Application of capacity spectrum method to the optimized structure. Fig. 11. Acceleration response spectra for scaled ground motions for very rare
earthquakes.

Table 4
Comparison of cost for the initial and optimal designs. based on the optimal design are evaluated and compared in Table 4.
The optimal design cuts 18.9%, 22.4%, and 21.1% of material cost,
Initial design Optimal design Cost reduction
($) ($) (%) labor cost, and total cost, respectively. The optimal design reduces the
needed flexural strengths of the beams and columns by 30%. However,
Concrete 37,497 27,977 25.4 because of the 26% decrease in cross sectional area, the cost of flexural
Flexural reinforcement 17,305 16,441 4.99
reinforcement in the optimal design is reduced by only 5%. Major cost
Labor 86,098 66,805 22.4
Total cost 140,900 111,223 21.1 saving is contributed by the significantly decreased concrete con-
sumption. The transverse reinforcement needed for shear and confine-
ment in the beams and columns are designed based on ACI 318–14
Fig. 9 shows the converted feasible region boundary defined in the [27]. If taking into account the costs associated with transverse re-
Ωc–Ws domain. To define the objective function expressed in Eq. (1), inforcement, the overall cost of the interior frame is reduced by 18%
the unit cost of concrete and steel reinforcement, cc and cs, are calcu- from $171,760 for the initial design to $140,394 for the optimal design.
lated based on the construction cost data published by the BNi Building
News [40] and RS Means [41]. cc and cs consider both material and 8.3. Dynamic response of optimized RC frame building
labor cost; cc also includes formwork cost defined based on four uses.
The dashed line in Fig. 9 shows the object function. The tangent point Compared with the initial strength-based design, the optimal PBSD
between the objective function and feasible region gives the optimal considerably reduces needed flexural stiffness and strength of the
solution in terms of Ws and Ωc, which correspond to α = 0.70 and beams and columns. Moreover, the proposed optimization approach
λ = 0.55. The optimal design leads to a 30% reduction in needed estimates the peak dynamic response of a nonlinear system using ca-
flexural strength and a 26% reduction in cross-sectional area from the pacity spectrum method. Nonlinear time-history analyses are conducted
initial strength-based design. to examine the dynamic response of the optimized structure. Ten
Fig. 10 shows the capacity curve for the optimized structure ground motions recorded during M6.2 to M7.3 earthquakes at soft rock
(λ = 0.55, α = 0.70) in the equivalent SDOF system and the demand sites, are selected from the database provided by the Pacific Earthquake
curves (inelastic displacement–acceleration spectra) for the three levels Engineering Research Center. Table 5 shows the record number, mag-
of considered earthquake hazard. As shown by the capacity curve de- nitude (Mw), fault type, closest distance to the surface projection of
rived from nonlinear static analysis, the structure experiences little rupture surface (Rjb), and closest distance to the rupture surface (Rrup)
strength degradation after reaching the peak load. Moreover, the de- of the chosen ground motions. To represent the three levels of earth-
mand curve for occasional earthquakes intersects the first branch of the quake hazard, the peak ground accelerations (PGAs) of the chosen re-
idealized bilinear capacity curve, indicating an overall elastic behavior cords are scaled so that the response spectrum for each seismic hazard
of the optimized structure under occasional earthquakes. matches the design spectrum. During the scaling, the mean squared
The costs of the interior frame based on the initial design and that error within 0.2 T and 1.5 T (T = fundamental period of the structure

Table 5
Ground motion details.
No. Year Earthquake Record No. in database Mw Rjb (km) Rrup (km) Fault Type Scaled PGA (g)

occasional rare very rare

1 1983 Coalinga 357 6.36 32.8 34.0 Reverse 0.301 0.692 1.086
2 1984 Morgan Hill 472 6.19 31.9 31.9 Strike-slip 0.199 0.458 0.718
3 1989 Loma Prieta 748 6.93 43.9 44.1 Reverse/Oblique 0.153 0.351 0.551
4 1994 Northridge 948 6.69 41.1 41.4 Reverse 0.242 0.557 0.874
5 1999 Chi-Chi 1259 7.62 43.5 47.9 Reverse/Oblique 0.194 0.446 0.700
6 1999 Hector Mine 1762 7.13 41.8 43.1 Strike-slip 0.157 0.361 0.566
7 1992 Landers 3752 7.28 45.3 45.3 Strike-slip 0.200 0.461 0.723
8 2004 Niigata 4230 6.63 36.8 39.4 Reverse 0.273 0.627 0.984
9 2007 Chuetsu-oki 5261 6.80 42.0 45.4 Reverse 0.144 0.331 0.519
10 2008 Iwate 5779 6.90 36.3 36.3 Reverse 0.139 0.319 0.501

23
C. Zhang, Y. Tian Engineering Structures 185 (2019) 15–25

7 7 7

Individual Individual Individual

6 6 6
Mean Mean Mean
5 Mean +/- S.D. 5 Mean +/- S.D. 5 Mean +/- S.D.
Story Number

4 4 4

3 3 3

2 2 2

1 1 1
(a) Limit = 1% (b) Limit = 2% (c) Limit = 4%

0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 0.5 1 1.5 2 2.5 3 0 1 2 3 4 5
Inter-story Drift Ratio (%) Inter-story Drift Ratio (%) Inter-story Drift Ratio (%)

Fig. 12. Peak inter-story drift for optimized building subjected to ground motions scaled for (a) occasional earthquakes, (b) rare earthquakes, and (c) very rare
earthquakes.

Fig. 13. Ratio of average peak plastic hinge rotation demand to capacity (θmax/[θ]) for optimized building (α = 0.70 and λ = 0.55) subjected to ground motions
scaled for (a) occasional earthquakes, (b) rare earthquakes, and (c) very rare earthquakes.

with the initial design) is minimized. Table 5 lists the scaled PGAs of the
ten ground motions. Fig. 11 shows the response spectra of each ground
motion, the average response spectrum, and the design spectrum for
very rare earthquakes.
Dynamic analyses are conducted on the optimized building using
the scaled ground motions. In general, the average peak dynamic re-
sponses meet the target performance requirements under the three le-
vels of seismic hazard. Fig. 12 shows the profiles of peak inter-story
drift γmax caused by individual ground motions and the average γmax.
The maximum average γmax, occurring in either the 4th or the 5th floor,
is 1.02% for occasional earthquake, 2.09% for rare earthquake, 3.13%
for very rare earthquake. Thus, the maximum average γmax only slightly
exceeds the drift limit of 1% for the occasional earthquakes and 2% for
the rare earthquakes, which are shown by the vertical dashed lines.
The rotation capacity of a plastic hinge varies during dynamic
loading due to the change in beam shear or column axial force. Thus, a
normalized plastic hinge rotation, defined as the ratio of plastic hinge
rotation demand to capacity (θ/[θ]), is considered. The peak θ/[θ]
experienced over entire loading history, denoted as θmax/[θ], is iden-
tified for each ground motion. Fig. 13 shows the distribution of average
θmax/[θ] for the three levels of seismic hazard. Under occasional
earthquakes, only one ground excitation causes yielding and, as shown
in Fig. 13(a), the yielding occurs only in a few beams. Under rare
earthquakes, all the beams experience yielding in terms of the average
response (Fig. 13(b)). The average θmax/[θ] ranges from 0.011 to 0.518
and is much lower in the columns than in most of the beams. Under
very rare earthquakes, the average θmax is less than 67% of capacity in
any member (Fig. 13(c)). As shown in Fig. 13(b) and (c), column
yielding occurs at the upper stories, which is not predicted by the
nonlinear static analyses, indicating the effects of higher vibration
modes.
Note that Fig. 13 is used to demonstrate the magnitude of θmax/[θ]
evaluated in an average sense. However, for an individual ground
motion, the actual plastic hinge distribution pattern at the time of
reaching peak response is not identical to that shown in Fig. 13. Even if
all the columns in the 5th story are labeled with plastic hinges at both
the top and bottom ends in Fig. 13(c), those plastic hinges do not ap-
pear at the same time under a specific ground motion; thus, no soft-
story mechanism is formed by any ground motion. As examples,
Fig. 14(a) and (b) show the normalized plastic hinge rotation dis-
tribution at the peak response during (1) the 2004 Niigata earthquake
scaled for very rare earthquakes, which leads to the highest inter-story

24
C. Zhang, Y. Tian
Fig. 14. Ratio of peak plastic hinge rotation demand to capacity (θmax/[θ]) for
optimized building (α = 0.70 and λ = 0.55) subjected to (a) the 2004 Niigata
ground motion and (b) the 1999 Chi-Chi ground motion scaled for very rare
earthquakes.
drift ratio, and (2) the 1999 Chi-Chi ground motion scaled also for very
rare earthquakes, which causes the largest normalized plastic rotation
in the columns as well as significant higher vibration mode effects.
9. Conclusions
Using plastic hinge rotation and optionally inter-story drift to
measure structural performance and define constraints, this study de-
velops a new framework of optimal performance-based seismic design
(PBSD) for multi-story reinforced concrete moment frame buildings.
The proposed approach has two features. First, the design variables are
simplified into two, one for overall system stiffness and the other for
overall system strength. Second, and a two-stage optimization process is
employed, including the determination of feasible region boundary in
normalized strength and stiffness domain and the optimization in ma-
terial consumption domain. Capacity spectrum method, which jointly
considers nonlinear static analysis and inelastic design spectrum, is
used during the optimizations to estimate the peak global and local
deformations.
The proposed optimal PBSD approach is applied to a six-story re-
inforced concrete frame building. The result is controlled by inter-story
drift limit under rare earthquakes. Compared with the conventional
strength-based design, the proposed optimal PBSD leads to 30% re-
duction in the needed flexural strength for the beams and columns, 26%
reduction in the cross-sectional area, and 21% reduction in the overall
cost. If only plastic hinge rotation is used to measure structural per-
formance, higher degree of cost saving can be reached. The perfor-
mance of the optimized structure is examined by nonlinear time-history
analyses using ten scaled ground motions. It is found that drift limits are
exceeded by less than 5% for occasional and rare earthquakes, whereas
plastic hinge rotation limits are satisfied under any considered level of
seismic hazard.
References
[1] ASCE (American Society of Civil Engineers). Seismic rehabilitation of existing
buildings. ASCE/SEI 608 41-13, Reston, VA; 2014.
[2] FEMA (Federal Emergency Management Agency). NEHRP guidelines for the seismic
rehabilitation 610 of buildings. FEMA 273, Washington, D.C.; 1997.
[3] FEMA (Federal Emergency Management Agency). Prestandard and commentary for
25
Engineering Structures 185 (2019) 15–25
the seismic rehabilitation of buildings. FEMA 356, Washington, D.C.; 2000.
[4] Ghobarah A. Performance-based design in earthquake engineering: state of devel-
opment. Eng Struct 2001;23:878–84.
[5] Fragiadakis M, Lagaros ND. An overview to structural seismic design optimisation
frameworks. Comput Struct 2011;89:1155–65.
[6] Ganzerli S, Pantelides CP, Reaveley LD. Performance-based design using structural
optimization. Earthquake Eng Struct Dyn 2000;29(11):1677–90.
[7] Zou XK, Chan CM. An optimal resizing technique for seismic drift design of concrete
buildings subjected to response spectrum and time history loadings. Comput Struct
2005;83(19–20):1689–704.
[8] Zou XK, Chan CM. Optimal seismic performance-based design of reinforced concrete
buildings using nonlinear pushover analysis. Eng Struct 2005;27(8):1289–302.
[9] Liu M, Burns SA, Wen YK. Multiobjective optimization for performance-based
seismic design of steel moment frame structures. Earthquake Eng Struct Dyn
2005;34(3):289–306.
[10] Xu L, Gong Y, Grierson DE. Seismic design optimization of steel building frame-
works. J Struct Eng 2006;132(2):277–86.
[11] Grierson DE, Gong Y, Xu L. Optimal performance-based seismic design using modal
pushover analysis. J Earthquake Eng 2006;10(1):73–96.
[12] Rojas HA, Foley C, Pezeshk S. Risk-based seismic design for optimal structural and
nonstructural system performance. Earthquake Spectra 2011;27(3):857–80.
[13] Fragiadakis M, Papadrakakis M. Performance-based optimum seismic design of
reinforced concrete structures. Earthquake Eng Struct Dyn 2008;37(6):825–44.
[14] Möller O, Foschi RO, Quiroz LM, Rubinstein M. Structural optimization for per-
formance-based design in earthquake engineering: applications of neural networks.
Struct Saf 2009;31(6):490–9.
[15] Li G, Lu H, Liu X. A hybrid simulated annealing and optimality criteria method for
optimum design of RC buildings. Struct Eng Mech 2010;35(1):19–35.
[16] Kaveh A, Azar BF, Hadidi A, Sorochi FR, Talatahari S. Performance-based seismic
design of steel frames using ant colony optimization. J Constr Steel Res
2010;66(4):566–74.
[17] Gholizadeh S, Moghadas RK. Performance-based optimum design of steel frames by
an improved quantum particle swarm optimization. Adv Struct Eng
2014;17(2):143–56.
[18] Zou XK, Chan CM, Li G, Wang Q. Multiobjective optimization for performance-
based design of reinforced concrete frames. J Struct Eng 2007;133(10):1462–74.
[19] Möller O, Foschi RO, Ascheri JP, Rubinstein M, Grossman S. Optimization for
performance-based design under seismic demands, including social costs.
Earthquake Eng Eng Vibration 2015;14(2):315–28.
[20] Yazdani H, Khatibinia M, Gharehbaghi S, Hatami K. Probabilistic performance-
based optimum seismic design of RC structures considering soil–structure interac-
tion effects. J Risk Uncertainty Eng Syst Part A: Civil Eng 2016;3(2):G4016004.
[21] Mitseas IP, Kougioumtzoglou IA, Beer M. An approximate stochastic dynamics
approach for nonlinear structural system performance-based multi-objective op-
timum design. Struct Saf 2016;60:67–76.
[22] Hajirasouliha I, Asadi P, Pilakoutas K. An efficient performance-based seismic de-
sign method for reinforced concrete frames. Earthquake Eng Struct Dyn
2012;41(4):663–79.
[23] Pan P, Ohsaki M, Kinoshita T. Constraint approach to performance-based design of
steel moment-resisting frames. Eng Struct 2007;29:186–94.
[24] Xu J, Spencer Jr BF, Lu X. Performance-based optimization of nonlinear structures
subject to stochastic dynamic loading. Eng Struct 2017;134:334–45.
[25] Arroyo O, Gutiérrez S. A seismic optimization procedure for reinforced concrete
framed buildings based on eigenfrequency optimization. Eng Optim
2017;49(7):1166–82.
[26] Paulay T, Priestly MJN. Seismic design of reinforced concrete and masonry build-
ings. John Wiley & Sons; 1992.
[27] ACI (American Concrete Institute). Building code requirements for structural con-
crete and commentary. ACI 318-14, Farmington Hills, MI; 2014.
[28] SEAOC (Structural Engineers Association of California). Conceptual framework for
performance-based seismic design; 1999.
[29] ASCE (American Society of Civil Engineers). Seismic rehabilitation of existing
buildings. ASCE/SEI 41-06, Reston, VA; 2007.
[30] Fajfar P, Gaspersic P. The N2 method for the seismic damage analysis of RC
buildings. Earthquake Eng Struct Dyn 1996;25(1):31–46.
[31] Fajfar P. A nonlinear analysis method for performance based seismic design.
Earthquake Spectra 2000;16(3):573–92.
[32] Krawinkler H, Seneviratna GDPK. Pros and cons of a pushover analysis of seismic
evaluation. Eng Struct 1998;20(4–6):452–64.
[33] Miranda E, Bertero VV. Evaluation of strength reduction factors for earthquake-
resistant design. Earthquake Spectra 1994;10(2):357–79.
[34] Vidic T, Fajfar P, Fischinger M. Consistent inelastic design spectra: strength and
displacement. Earthquake Eng Struct Dyn 1994;23:507–21.
[35] ASCE (American Society of Civil Engineers). Minimum design loads for buildings
and other structures. ASCE/SEI 7-10, Reston, VA; 2010.
[36] Kalkan E, Kunnath SK. Assessment of current nonlinear static procedures for seismic
evaluation of buildings. Eng Struct 2007;29:305–16.
[37] OpenSees, Open system for earthquake engineering simulation; 2016. http://
opensees.berkeley.edu.
[38] Ibarra LF, Medina RA, Krawinkler H. Hysteretic models that incorporate strength
and stiffness deterioration. Earthquake Eng Struct Dyn 2005;34(12):1489–511.
[39] ASCE (American Society of Civil Engineers). Minimum design loads for buildings
and other structures. ASCE/SEI 7-05, Reston, VA; 2005.
[40] BNi Building News. ENR General Contracting Costbook; 2015.
[41] RS Means. Foot Costs; 2015.