You are on page 1of 15

Structural and Multidisciplinary Optimization (2019) 60:1123–1137

https://doi.org/10.1007/s00158-019-02260-4

RESEARCH PAPER

Optimum design of steel braced frames considering dynamic


soil-structure interaction
Milad Bybordiani 1 & Saeid Kazemzadeh Azad 2

Received: 10 September 2018 / Revised: 6 February 2019 / Accepted: 11 March 2019 / Published online: 3 April 2019
# Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Recent studies on design optimization of steel frames considering soil-structure interaction have focused on static loading
scenarios, and limited work has been conducted to address the design optimization under dynamic soil-structure interaction.
In the present work, first, a platform is developed to perform optimization of steel frames under seismic loading considering
dynamic soil-structure interaction (SSI) in order to quantify the effects of earthquake records on the optimum design. Next,
verification of the adopted modeling technique is conducted using comparison of the results with the reference solution coun-
terparts in frequency domain. For time history analyses, records from past events are selected and scaled to a target spectrum
using simple scaling approach as well as spectrum matching technique. For sizing of the steel frames, a recently developed
metaheuristic optimization algorithm, namely exponential big bang-big crunch optimization method, is employed. To alleviate
the computational burden of the optimization process, the metaheuristic algorithm is integrated with the so-called upper bound
strategy. Effects of factors such as the building height, presence of soil domain, and the utilized ground motion scaling technique
are investigated and discussed. The numerical results obtained based on 5- and 10-story steel braced frame dual systems reveal
that, although dynamic SSI reduced the seismic demands to some extent, given the final design pertains to different load
combinations, the optimum weight difference is not considerable.

Keywords Optimization . Steel frames . Soil-structure interaction . Massless foundation . Base shear . Drift ratio

1 Introduction commonly used for infrastructures such as dams; however,


the practical difficulties in the design procedure precluded
The phenomenon of dynamic soil-structure interaction has the wide-spread use of SSI for the optimum design of
long been studied for different types of structures. So as to ordinary buildings.
tackle this problem, variety of numerical and analytical solu- Obtaining an optimum design for a building can be consid-
tions have been proposed to idealize the conditions involved ered as assigning the best values to the design variables so that
(Seed and Idriss 1969; Trifunac 1972; Gazetas and Roesset a set of predefined objectives are achieved, while the con-
1976; Bielak 1976; Wolf 1985) on the basis of the finite ele- straints stipulated by the adopted design codes are not violat-
ment and boundary element methods owing to the dramatic ed. In the case of steel frames, the minimum structural weight
enhancement of computing technologies. Both methods are (or cost) is typically considered as the main objective of the
optimization problem for which the frame members are to be
selected from a list of available steel profiles. This results in a
Responsible Editor: Mehmet Polat Saka
discrete or combinatorial optimization problem which can be
exactly solved for a global optimum only through enumerat-
* Saeid Kazemzadeh Azad
saeid.azad@atilim.edu.tr
ing on all the allowable discrete values for each of the design
variables (i.e., an exhaustive search of the whole solution
Milad Bybordiani space). Indeed, the development of optimization algorithms
milad.bybordiani@sydney.edu.au
for handling such discrete optimization problems is basically
1
School of Civil Engineering, The University of Sydney, due to that such an exhaustive search is not reasonable, nor
Sydney, Australia practical in most cases. Thus, search techniques capable of
2
Department of Civil Engineering, Atilim University, Ankara, Turkey generating near optimum solutions without performing an
1124 M. Bybordiani, S. K. Azad

exhaustive search have become popular in practical applica- The third section presents a detailed statement of the
tions (Saka 2007). employed metaheuristic optimization algorithm. In the fourth
Undoubtedly, most of the recently developed optimization section, first, a short summary of the modal time history meth-
algorithms for the design of steel frames belong to the class of od is outlined. Afterwards, using the frequency response func-
stochastic search techniques or the so-called metaheuristics tions (FRF), stiffness and radiation criteria of the soil domain
(Kameshki and Saka 2001; Hasançebi et al. 2010; Kripakaran are studied. Then, the ground motions as well as the utilized
et al. 2011; Kaveh and Abbasgholiha 2011; Hasançebi and state-of-the-practice scaling techniques (NEHRP 2011) are
Kazemzadeh Azad 2012; Gholizadeh and Poorhoseini 2016; elaborated. Following the introduction of the numerical ap-
Gholizadeh and Milani 2016). The basic idea in metaheuristic proach, the numerical results obtained for two steel frames
search techniques is to investigate the vicinity of promising so- are thoroughly discussed in the fifth section considering issues
lutions found so far to determine the direction of the search at such as the optimum weight, maximum drift ratios, and stress
each iteration. This process is usually carried out in the form of a distribution in soil domain. The conclusions and the summary
random search so that the algorithm could perform a global of the study are presented in the last section.
search in the solution space to achieve a promising final design.
This random search methodology in conjunction with selection
of promising solutions at each iteration forms the basis for de- 2 Statement of the design optimization
veloping efficient metaheuristic algorithms that gradually con- problem
verge to an optimum or a near optimum solution. State-of-the-art
reviews of metaheuristic algorithms considering their applica- In practical applications, the frame members are typically se-
tions in structural design optimization can be found in Saka lected from a set of available steel sections which yields a
(2007), Lamberti and Pappalettere (2011), and Saka et al. (2016). discrete sizing optimization problem. For a steel frame com-
Nonetheless, the main focus of the research on optimum posed of Nm members grouped into Nd design groups, the
design of steel frames has been tailored towards developing optimum design problem, based on AISC-LRFD (American
efficient optimization algorithms, and limited work has been Institute of Steel Construction (AISC) 1994) code, can be
devoted to investigating the optimal structural design under stated as follows. The objective is to find a vector of integer
dynamic loadings (Kazemzadeh Azad et al. 2018). Even in the values I (1) representing the sequence numbers of steel sec-
more recent studies, the problem of the optimum design of tions assigned to Nd member groups
steel space frames considering SSI was merely limited to the
static loading scenarios (Daloglu et al. 2016). It was conclud- IT ¼ ½I 1 ; I 2 ; :::; I N d  ð1Þ
ed that consideration of the SSI leads to heavier structures.
to minimize the weight, W, of the structure
However, the optimization of the steel frames considering
SSI has not been addressed properly in the literature based Nd Nt

on rigorous time history analyses to incorporate the seismic W ¼ ∑ ρi A i ∑ L j ð2Þ


i¼1 j¼1
demands in the design procedure.
Consequently, the main purpose of the present study is where Ai and ρi are the length and unit weight of the steel
threefold: (1) to perform structural optimization of steel section selected for member group i, respectively, Nt is the
frames under seismic loadings, (2) to include dynamic soil- total number of members in group i, and Lj is the length of
structure interaction, and (3) to quantify the effects of inherent the member j which belongs to group i. Here, the objective of
variability of input earthquake records on the optimum design. finding the minimum weight structure is subjected to several
In this regard, typical steel frames were considered resting on design constraints, including strength and serviceability re-
rigid base as well as half-space. The unbounded soil domain quirements. According to AISC-LRFD (American Institute
was modeled using the standard massless foundation (United of Steel Construction (AISC) 1994) code of practice, the fol-
States Army Corps of Engineers 2003). Two sets of ground lowing design constraints (C iIEL and C vIEL ) must be satisfied
motions were used as the input record for the seismic time for the strength requirements.
history analysis of the models. For this purpose, records from
     
past events were selected and scaled to a target spectrum using Pu J 8 Mu J Pu J
C iIEL ¼ þ −1 ≤ 0 for ≥ 0:2 ð3Þ
simple scaling approach as well as spectrum matching tech- ϕPn IEL 9 ϕb M n IEL ϕPn IEL
nique. Effects of factors such as the building height, presence
     
of soil domain, and the utilized ground motion scaling tech- Pu J Mu J Pu J
C iIEL ¼ þ −1 ≤0 for < 0:2 ð4Þ
nique were investigated. 2 ϕPn IEL ϕb M n IEL ϕPn IEL
The study is arranged as follows. The second section de-
scribes the optimum design problem of steel frames as well as C vIEL ¼ ðV u J ÞIEL −ðφv V n ÞIEL ≤0 ð5Þ
the imposed strength and serviceability design constraints.
Optimum design of steel braced frames considering dynamic soil-structure interaction 1125

In (3) to (5), IEL = 1, 2, …, NEL is the element number, members with compact and/or non-compact elements, the
NEL is the total number of elements, J = 1, 2, …, N is the load nominal compressive strength of the member for the limit state
combination number and N is the total number of design load of flexural buckling is as follows:
combinations. Pu J is the required axial (tensile or compres-
Pn ¼ F cr Ag ð9Þ
sive) strength, under Jth design load combination. Mu J is the
required flexural strength for bending, under the Jth design where Fcr is the critical stress based on flexural buckling of the
load combination. On the other hand, Pn and Mn are the nom- member, calculated as:
inal axial (tensile or compressive) and flexural strengths of the
IEL-th member under consideration. φis the resistance factor rffiffiffiffiffiffiffi  
Kl Fy
≤ 1:5; F cr ¼ 0:658λc F y
2
for axial strength, which is 0.85 for compression and 0.9 for for λc ¼ ð10Þ
tension (based on yielding in the gross cross section), and φb is rπ E
the resistance factor for flexure, which is equal to 0.9. Here, rffiffiffiffiffiffiffi " #
(5) is used for checking members’ shear capacity wherein Vu J Kl Fy 0:877
is the required shear strength under Jth load combination and for λc ¼ > 1:5; F cr ¼ Fy ð11Þ
rπ E λ2c
Vn is the nominal shear strength of the IEL-th member under
consideration. In order to calculate the design shear strength, In the above equations, l is the laterally unbraced length of
the nominal shear strength is multiplied by a resistance factor the member, K is the effective length factor, r is the governing
φv of 0.9. radius of gyration about the axis of buckling, and E is the
In addition to the strength requirements, the serviceability modulus of elasticity.
criteria should be considered in the design process. The ser- The AISC-LRFD (American Institute of Steel Construction
viceability constraints (C tD andC dF ) considered in this research (AISC) 1994) code addresses the nominal compressive
are formulated as follows: strength based on the limit state of torsional and flexural-
torsional buckling, for doubly symmetric members with com-
C tD ¼ ΔMax J −ΔaMax ≤ 0 ð6Þ
pact and/or non-compact elements. For this limit state, (9) is
C dF ¼ ½δ J  F −½ δa  F ≤ 0 ð7Þ still applicable with the following modifications:

Equation (6) compares the maximum lateral displacement  


F cr ¼ 0:658λe F y
2

of the structure in the Dth direction (D = 1, …, ND) under Jth for λe ≤ 1:5; ð12Þ
load combination ΔMax J with the maximum allowable lateral " #
displacement ΔaMax . Similarly, (7) checks the interstory drift of 0:877
the Fth story (F = 1, 2, …, NF) under the Jth load combination for λe > 1:5; F cr ¼ Fy ð13Þ
λ2e
[δJ]F against the related permitted value [δa]F; here NF is the
total number of stories. [δJ]F is the interstory drift ratio of the where
Fth story, while [δa]F is the allowable counterpart, specified as
2% of the story height (ASCE 2010) when the seismic load is qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
applied statically. Given the aforementioned quantities are λe ¼ F y= F e ð14Þ
calculated using time history analysis results, [δa]F was in- " #
creased by 25% (ASCE 2010). π2 EC w 1
Fe ¼ þ GJ ð15Þ
ðK z l z Þ 2 Ix þ Iy
2.1 Nominal strengths
In (15), Cw is the warping constant, G is the shear modulus,
Based on AISC-LRFD (American Institute of Steel J is the torsional constant, Ix and Iy are the moments of inertia
Construction (AISC) 1994) specification, the nominal tensile about principal axes, lz is the unbraced length for torsional
strength of a member, based on yielding in the gross cross buckling, and Kz is the effective length factor for torsional
section, is equal to: buckling. In this study, Kz is conservatively taken as unity.
The nominal flexural strength of a member is the minimum
Pn ¼ F y Ag ð8Þ value obtained according to the limit states of yielding, lateral-
torsional buckling, flange local buckling, and web local buck-
where Fy is the member’s specified yield stress and Ag is the ling. The flexural capacity based on the limit state of yielding
gross cross section of the member. is as follows:
The nominal compressive strength of a member is the
smallest value obtained from the limit states of flexural buck- M n ¼ M p ¼ Z F y ≤ 1:5 S F y ð16Þ
ling, torsional buckling, and flexural-torsional buckling. For
1126 M. Bybordiani, S. K. Azad

8 qffiffiffiffiffiffiffiffiffiffiffi
>
> if h=tw ≤2:45 E= F y
where Z is the plastic modulus and S is the section modulus of >
>
>
0:6F y Aw
p ffiffiffiffiffiffiffiffiffiffiffi !
>
> qffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffi
the member for the axis of bending. For doubly symmetric < 2:45 E= F y
if 2:45 E= F y < h=tw ≤ 3:07 E= F y
Vn ¼
0:6F y Aw
> "
h=tw ð21Þ
sections, the flexural capacity considering the limit state of >
> #
>
> qffiffiffiffiffiffiffiffiffiffiffi
> 4:52E
>
: Aw if 3:07 E= F y < h=tw ≤260
lateral-torsional buckling is as follows: ðh=tw Þ 2

8
>
>  Mp  
<
Lb −Lp if Lb ≤Lp where h is the web height, tw is the web thickness, and Aw is
M n ¼ C b M p − M p −M r ≤Mp if Lp < Lb ≤Lr ð17Þ the web area as per AISC-LRFD (American Institute of Steel
>
> Lr −Lp
: if Lb > Lr
M cr ≤M P Construction (AISC) 1994). For members subjected to shear
perpendicular to the plane of the web, the nominal shear
where Lb is the laterally unbraced length of the member, Lp is
strength is calculated using (21) as well.
the limiting laterally unbraced length for full plastic bending
capacity, Lr is the limiting laterally unbraced length for inelas-
tic lateral-torsional buckling, Mr is the limiting buckling mo-
3 Optimization algorithm
ment, and Mcr is the critical elastic moment for the lateral-
torsional buckling. The modification factor for non-uniform
The big bang-big crunch optimization method first appeared
moment diagram, Cb, is defined by (18):
in Erol and Eksin (2006) and emerged from the big bang and
12:5 M max big crunch theories of the universe evolution. As its name
Cb ¼ ð18Þ implies, the method is based on the continuous application
2:5 M max þ 3 M A þ 4 M B þ 3 M C
of two successive stages, namely the big bang and big crunch
where Mmax, MA, MB, and MC are the absolute values of max- phases. During the big bang phase, new solution candidates
imum moment, moment at quarter point, centerline, and three- are randomly generated around a point called the center of
quarter point of the unbraced segment, respectively. mass. This point is recalculated and updated every time in
The nominal flexural strength of members with doubly the big crunch phase with respect to the generated solution
symmetric sections and non-compact flanges, considering candidates. The algorithm is composed of the following steps:
the limit state of flange local buckling, is given below:
Step 1. Initial population: Form an initial population
8 through randomly spreading individuals (candidate solu-
>
> Mp !
>
< if λ f ≤ λpf tions) over all the search space in a uniform manner (the

λ f −λpf
M n ¼ M p − M p −M rf
if λpf < λ f ≤λrf ð19Þ first big bang). This step is applied once.
>
> λrf −λpf
>
: if λ f > λrf Step 2. Evaluation: The initial population is evaluated,
M cr ≤ M P
f
where structural analyses of all the individuals are carried
where λf is the flange slenderness parameter, λpf is the limiting out with the set of steel sections adopted for design var-
iables, and force and deformation responses are obtained
value of λf for full plastic bending capacity, λrf is the limiting
under the loads. The objective function values of the fea-
value of λf for inelastic flange local buckling, M rf is the lim-
sible individuals that satisfy all the problem constraints
iting moment for flange buckling, and M crf is the critical elas- are directly computed from (2). However, infeasible indi-
tic moment for flange local buckling. viduals that violate some of the problem constraints are
The nominal flexural strength of members with doubly penalized using an external penalty function approach,
symmetric sections and non-compact webs, considering the and their objective function values are computed accord-
limit state of flange web buckling, is given below: ing to (22) (Kazemzadeh Azad et al. 2013).
8   
>
< Mp ! NEL NEL ND NF

λw −λwp if λw ≤ λwp f ¼ W 1 þ p ∑ C iIEL þ ∑ C vIEL þ ∑ C tD þ ∑ C dF ð22Þ
Mn ¼
if λwp < λw ≤ λwr
ð20Þ
: M p − M p −M r
w
> IEL¼1 IEL¼1 D¼1 F¼1
λwr −λwp

In (22), f is the constrained objective function value, and Ci,


where λw is the web slenderness parameter, λwp is the limiting C , Ct, and Cd are the strength and serviceability constraint
v

value of λw for full plastic bending capacity, λwr is the limiting violations. Here, p is the penalty coefficient used to
value of λf for inelastic web local buckling, M wr is the limiting tune the intensity of penalization as a whole. The fitness
moment for web buckling, and M wcr is the critical elastic mo- scores of the individuals are then calculated by taking
ment for web local buckling. the inverse of their objective function values (i.e., fit-
The nominal shear strength of unstiffened webs of doubly ness = 1/W or 1/f for feasible and infeasible solutions,
symmetric members, subjected to shear in the plane of the respectively). The fitness scores are assigned as the mass
web, is determined as follows: values for the individuals.
Optimum design of steel braced frames considering dynamic soil-structure interaction 1127

It is, however, important to note that the final optimum Recently, in order to enhance the performance of the big
design shall have zero constraint violation. In (22), the penalty bang-big crunch (BB-BC) algorithm in discrete sizing optimi-
coefficient (p) plays a major role in this regard. Using a very zation, (25) is proposed in (Hasançebi and Kazemzadeh Azad
small value would lead the search to infeasible regions, where- 2014) as a new formulation in lieu of (24). In the new formu-
as using a large value might prevent the algorithm from lation, the use of the nth power (n ≥ 2) of a random number ri
searching the vicinity of the constraint boundaries, where is motivated considering an appropriate statistical distribution,
promising solutions typically exist. In the present study, a which may not necessarily be a normal distribution.
value of p = 1.0 is used which is in line with previous studies  max min

(Hasançebi and Kazemzadeh Azad 2012) ensuring that the n I i −I i
I new
i ¼ I c
i þ round α⋅r i ð25Þ
algorithm would approach the optimum from both feasible k
and infeasible regions.
The rationale behind (25) is to achieve a satisfactory trade-
Step 3. Big crunch phase: Calculate the center of mass by off between the exploration and exploitation characteristics of
taking the weighted average using the coordinates (de- the BB-BC algorithm. Two instances of (25) provided in
sign variables) and the mass values of every single indi- Hasançebi and Kazemzadeh Azad (2012) are given in (26)
vidual or choose the fittest individual among all as their and (27) wherein the power of the random number (i.e., n) is
center of mass (the latter approach is used in this study). set to 3.
Step 4. Big bang phase: Generate new individuals by  max min

3 I i −I i
using normal distribution (the big bang phase). For a I i ¼ I i þ round α⋅N ð0; 1Þi
new c
ð26Þ
continuous variable optimization problem, the following k
 max min

3 I i −I i
relation can be used to generate new solutions around the
center of mass. I new
i ¼ I c
i  round α⋅E ð λ ¼ 1 Þ i ð27Þ
k
max min

x −xi Equation (26) refers to the third power reformulation of the


xi ¼ xi þ α⋅N ð0; 1Þi i
new c
ð23Þ
k big crunch phase according to a normally distributed random
where xci is the value of ith continuous design variable in the number. In a recent study in Hasançebi and Kazemzadeh Azad
(2014), a very satisfactory performance of this reformulation
fittest individual, xmin
i and xmax
i are the lower and upper bounds
was demonstrated for discrete design optimization of truss
on the value of ith design variable, respectively, N(0, 1)i is a
structures. This reformulation is referred to here as the modi-
random number generated according to a standard normal
fied BB-BC (MBB-BC) algorithm.
distribution with mean (μ) zero and standard deviation (σ)
The second reformulation (27), referred to as the exponen-
equal to one, k is the iteration number, and α is a constant
tial BB-BC (EBB-BC) algorithm, is an alternative approach to
(taken as 0.25 in this study).
both (24) and (26), where the use of an exponential distribu-
However, in case a discrete set of available sections is used
tion, E(λ = 1), in conjunction with the third power of the ran-
for sizing the frame members, (24) can be employed to round
dom number is favored. In the present study, the EBB-BC
off the real values to the nearest integers representing the se-
algorithm is employed for the structural design optimization
quence number of available sections in a given section list.
of steel frames. The approach is particularly effective when
 max min
 tackling discrete sizing optimization problems of frame struc-
I −I i tures (Hasançebi and Kazemzadeh Azad 2012). The perfor-
I new
i ¼ I ci þ round α⋅N ð0; 1Þi i ð24Þ
k mance evaluation of the EBB-BC versus other metaheuristic
optimization algorithms, such as simulated annealing, tabu
where I ci is the value of ith discrete design variable in the fittest search, and harmony search, can be found in ref. Hasançebi
individual, and I min
i and I max
i are its lower and upper bounds, and Kazemzadeh Azad (2012).
respectively. Furthermore, the employed optimization algorithm can be
readily integrated with the so-called upper bound strategy
Step 5. Elitism: Keep the fittest individual found so far (UBS) (Kazemzadeh Azad et al. 2013) to improve the com-
separately or as a member of the population. putational efficiency. The key factor in this approach is to set
Step 6. Termination: Go to step 2 until a stopping criterion the penalized weight of the current best design found during
is satisfied. The algorithm can be terminated when the the previous iterations as an upper bound for the net weight of
number of iterations reaches a predefined maximum val- the newly generated candidate designs. Hence, any new can-
ue or when no improvement is recorded for the best de- didate design with a net weight greater than this upper bound
sign over a certain number of iterations. Alternative stop- will not be analyzed, and this will alleviate the computational
ping criteria can also be imposed. burden of the optimization process.
1128 M. Bybordiani, S. K. Azad

4 Dynamic analysis (ODE) for loads approximated by linear piece-wise forcing


history within small time increments prevails against alterna-
4.1 Modal time history approach tive procedures for practical design purposes (Wilson 2002).
Taking advantage of the modal expansion as the behavior
The forced vibration of a MDoF system can be shown by (28): of the structure is considered to be in the linear range, the
relative displacement vector can be written as:
::
mu þku ¼ peff where peff ¼ − mlug
:: ð28Þ
N
u ¼ ∑ Φn qn ð29Þ
In (28), the matrices m and k represent the mass and stiff- n¼1
ness matrices of the whole soil-structure domain. The right-
hand side of the above equation shows the effective earth- Consistent with the nomenclature given in Chopra (2012),
quake force as the result of free field ground motion Φn and qn represent natural vibration mode shapes and corre-
(Fig. 1a) where l is the influence vector (Chopra 2012) taking sponding modal coordinates, respectively, with N being the
into account the horizontal component of ground motion (Fig. maximum number of modes considered. Here, Φn and qn de-
1b). Here, (€ug ) represents the free-field ground motion record- pend only on the coordinate of a DoF and time, respectively.
N
ed in all directions. Given that the soil domain is modeled with Substituting u with ∑ Φn qn in (28), pre-multiplying with
massless elements, Peff is only non-zero at structural DoFs in n¼1

the horizontal direction. It is worth noting that the distribution ΦTn , and using orthogonality property of natural modes
of loads follows the dynamic characteristics of the structure (Chopra 2012):
and frequency content of the input motion. In other words, no
ΦTn mΦn €qn þ ΦTn kΦn qn ¼ ΦTn peff ð30Þ
presumed load distribution is required when employing (28).
Solution of an initial-boundary value problem (28) for a mn €qn þ k n qn ¼ pn ð31Þ
multi-degree-of-freedom system can be extremely challenging
when numerical time integration schemes are employed. The large set of global equilibrium equations (28) reduces
Although in the range of linear behavior the use of such to a relatively small number of uncoupled second-order dif-
methods (explicit/implicit approaches) would not seem to im- ferential equations of motion resembling that of a SDoF sys-
pose any challenges, an optimization process indeed necessi- tem (31). Terms mn ¼ ΦTn mΦn and k n ¼ ΦTn kΦn in (31) are
tates implementing the fastest analysis methods. In this regard, called modal mass and modal stiffness, respectively. These
one may exploit the superposition properties of a linear MDoF quantities are scalar values corresponding to the natural fre-
structure and solve the modal dynamic equations which math- quency of the nth mode.
ematically represent single degree-of-freedom (SDoF) sys- In order to consider the damping characteristics of the
tems (Chopra 2012). On the other hand, the use of closed- structure, the modal equation of motion (32) was modified
form solution precludes numerical instability without using with modal damping ratios (33). Calculations of the damping
any compensatory numerical damping (Wilson 2002). ratios are given in the next section.
Considering the above, equation of motion is solved analyti-
mn €qn þ cn ˙qn þ k n qn ¼ pneff ð32Þ
cally using the modal equations to compute the responses,
assuming linear variation of the forcing functions (input cn ¼ ζ n ccr ¼ 2ζ n mn ωn ð33Þ
ground motion) between sampling points. Given that the anal-
yses were in the linear range, modal time history approach where ζn, mn, and ωn denote the nth damping ratio, modal
decreases the computational effort drastically utilizing the nat- mass and natural frequency of the soil-structure system, re-
ural modes in lieu of large number of degrees of freedom. spectively. Equation (32) represents the viscously damped
Thus, analytical solution of the ordinary differential equations forced vibration of the nth mode of the structure. This is

-mn

H -m2
..u -m1
g=
W
H Massless Foundation
3W

a) Free-Field Motion b) Soil-Structure Model c) Structure on Half-Space


Fig. 1 Soil-structure model
Optimum design of steel braced frames considering dynamic soil-structure interaction 1129

solved for enough number of modes so that the ratio of the acceptable range of approximation. In this regard, the depth
cumulative effective modal mass to the total structural coun- and width of the soil domain were chosen as H and 3W,
terpart would exceed 0.99. This enables the modal time histo- respectively (Fig. 1b). The chosen quantities are the lower
ry approach to fully capture the dynamic response even at high bounds for which natural frequencies match those of the ref-
frequency ranges. It is worth noting that such a choice has erence solution.
almost no effect on the computational cost yet already satis- Given the two-dimensional modeling of the soil-structure
fying the ASCE 7–10 requirements considering the minimum system, the foundation of the structures was modeled using
required number of modes (ASCE 2010). Finally, following beam elements: Rigid foundation mat approximation was not
computation of qn for N number of modes and utilizing modal necessary. In order to eliminate numerical issue associated
superposition (29), the total dynamic response can be obtained with the linear rectangular plane strain elements (i.e., shear
and further used in the post-processing. In this regard, two locking), incompatible finite elements were utilized in lieu
main sets of responses are required in the optimization pro- of their conventional counterparts (Wilson 2002). The stiff-
cess: internal forces and floor displacement histories. While ness matrix of the soil domain was obtained for an out-of-
the former is required in design of members based on strength plane length equal to the out-of-plane tributary length of the
requirements, the latter is needed to check whether the lateral 2D frame (i.e., 5 m for the models in this study) (Takewaki
drift is within the allowable bounds. et al. 1998). Poisson’s ratio and shear wave velocity of the soil
domain in all models were 0.3 and 200 m/s, respectively. A
4.2 Soil-structure interaction material damping of 5% was considered for the soil domain as
well as superstructure.
Presence of a soil domain has been found to increase two Given the stiffness properties of the soil domain was tack-
major dynamic properties of structures: (1) period and (2) led using the bounded soil domain, its radiation and material
damping (Wolf 1985); where the additional flexibility of the damping were emulated using the effective modal damping
half-space underneath the structure leads to increased natural ratios. For the calculation of the modal damping ratios to be
periods, its inertial property brings about the radiation used in the massless foundation approach, structures with a
damping. The foundation model is assumed to be massless wide range of natural periods were first modeled using a ref-
so as to avoid the reflection of outward waves (Fig. 1c) as well erence solution. Herein, the term reference solution refers to
as the use of large foundation models (United States Army modeling of the unbounded soil domain without any simpli-
Corps of Engineers 2003). Dimensions of the foundation is fication. In this regard, the state-of-the-art perfectly matched
chosen such that the bounded domain would be able to mimic layer (PML) was utilized so as to consider the soil stiffness
the stiffness characteristics of the half-space within an and inertial properties along with its material and radiation

Fig. 2 FRFs for base shear:


reference solution, massless
model, and rigid base

a) 5-Story Frame b) 10-Story Frame


1130 M. Bybordiani, S. K. Azad

Table 1 Selected ground motions

ID Event Date PGA (g) Magnitude Rjb (km) Rrup (km) Fault mechanism Vs30 (m/s) Lowest usable frequency

1 Chuetsu-Japan 2007 0.150 6.8 47.35 47.45 Reverse 187 0.02


2 Iwate-Japan 2008 0.231 6.9 46.77 48.36 Reverse 158 0.08
3 Loma Prieta-US 1989 0.234 6.9 24.27 24.58 Reverse Oblique 239 0.10
4 Northridge-US 1994 0.536 6.7 0 5.92 Reverse 629 0.15

damping (Bybordiani and Arici 2019). The effective damping 4.3 Ground motion selection and scaling
for each mode was obtained using modified half-band width
method (Bybordiani and Arıcı 2017; Wang et al. 2013). Then, The study of FRFs is effective only to a comparative extent
using the above-mentioned damping quantities, effective (Sect. 4.2); that is, engineering demand parameters used for the
damping ratios associated with the natural frequencies of design and assessment of buildings are defined in the time do-
massless models were obtained using linear interpolation. In main. Moreover, in the study of the resonant peak values (or the
order to verify the adopted method, frequency response func- equivalent damping ratios), the location of the frequency and
tions (FRF) of models resting on the massless foundation were corresponding interaction with the ground motion is inadvertent-
compared to the reference solution counterparts. Given the full ly ignored. Thus, the use of time domain analysis to determine
frequency response matrix is hard to obtain and impossible to the seismic response becomes inevitable. This approach ensures
present, the FRF for the base shear was used as a representa- that the variability due to the uncertainty in the ground motions
tive tool. It is worth noting that many responses such as dis- is introduced into the analysis. For the computation of dynamic
placement of different DoFs, internal forces, etc. can be com- response, a suite of ground motions comprised of three records
pared for the purpose of verification of the implemented ap- chosen from the Pacific Earthquake Engineering Research
proach. However, while the above-mentioned quantities rep- (PEER) strong motion database was used (Motion 1 to 3 in
resent a very localized response of a building, the base shear Table 1). In addition to this suite, another record was chosen
provides an overall behavior. In this respect, unit acceleration and spectrally matched, capable of representing the seismic
over the frequency range of the interest was applied to a 5- hazard defined per the target spectrum of (ASCE 2010). The
story building resting on a half-space and the corresponding time histories of these records are presented in Fig. 3.
base shear FRF was obtained for the massless model (Fig. 2a). The ground motions chosen are required to be scaled to
For the sake of comparison, reference solution and rigid base conform to the target demand levels traditionally defined by
counterpart are also given in the figure. Investigation of results a response spectrum for the seismic hazard at the site (NEHRP
shows that the adopted method renders resonant amplitude/ 2011; ASCE 2010; FEMA450-1 2003). Due to the fact that
frequency in ± 2% of the reference counterpart (Fig. 2a). For a specific characteristics of earthquakes reflected on their time
10-story building (Fig. 2b), comparison of the results shows histories (e.g., the nature of the pulse, frequency content and
again an accurate prediction of the above-mentioned quanti- the duration) are not present on a response spectrum, different
ties similar to those of the 5-story frame. scaling techniques may be employed to match the target

Fig. 3 Earthquake record time history


Optimum design of steel braced frames considering dynamic soil-structure interaction 1131

Fig. 4 Ground motion definition

a) Response Spectra b) Fourier Amplitude

spectrum (ASCE 2010; Al Atik and Abrahamson 2010; Baker a minimum of 1 ground motion for the spectrum matching
2010). In this regard, two of state-of-the-practice scaling tech- technique can be used given that the analyses are in the elastic
niques were used in this study. When the prediction of both the
mean value of a demand as well as its dispersion is required,
simple scaling procedure should be used as it retains the record-
to-record variability inherent into the ground motions. On the
other hand, if an average engineering demand parameter (such
as story drift or base shear) is of interest, spectrum matching is
preferred since it reduces the variability (jaggedness) of the
acceleration spectrum for individual ground motions and leads
to a higher confidence in the predictions of average responses
(NEHRP 2011). The response spectra and Fourier amplitude of
the scaled selected records are presented in Fig. 4.

4.3.1 Simple scaling

In this approach, the ground motions shall be scaled such that


the average value of the 5% damped response spectra for the
suite of motions is not less than the design response spectrum
of the site for periods ranging from 0.2T1 to 1.5T1 where T1 is
the fundamental period of the structure in the direction of
response. Engineering demand parameters used in the design
of structural members shall be taken as the maximum quantity
obtained from the time history analyses when less than seven
motions are used in the analysis (ASCE 2010). Figure 4a
shows an example of simple scaling procedure for a 5-story
building considering the SSI effects (T1 = 0.42 s). A minimum
of three ground motions for linear analysis is required in sim-
ple scaling procedure (ASCE 2010).

4.3.2 Spectrum matching

The time domain spectral matching approach tries to achieve a


good fit between the individual ground motion’s spectrum and
the target counterpart. The method, implemented in a software
for response spectrum matching (RSPMatch), utilizes wavelet
transforms to modify the motion while preserving the non-
stationary nature of the originals (Al Atik and Abrahamson
2010). It is worth noting that the target and RSMatched mo-
tion’s spectra are almost identical to the point that no clear
difference can be seen between the two (Fig. 4a). Moreover, Fig. 5 Flowchart of the design optimization procedure
1132 M. Bybordiani, S. K. Azad

range (United States Army Corps of Engineers 2003; NEHRP Table 2 Optimized designs obtained for 5-story steel frame
2011). Detailed definition of the aforementioned scaling pro- Groups SSI Rigid SSI_RSPM Rigid_RSPM
cedures along with their effect on the dynamic response are
presented in Bybordiani and Arıcı (2018), Reyes et al. (2014), COL1 W10X39 W12X45 W10X39 W10X39
and Kurama and Farrow (2003). For the sake of clarity, flow- COL2 W6X15 W6X15 W6X15 W8X18
chart of the dynamic interaction process and the employed BM1 W12X16 W12X16 W12X16 W12X16
optimization algorithm is presented in Fig. 5. BM2 W12X16 W12X16 W12X16 W12X16
BM3 W12X16 W12X16 W12X16 W12X16
BM4 W12X16 W12X19 W12X16 W12X16
5 Numerical examples BM5 W12X14 W12X14 W12X14 W12X14
BR1 W6X20 W8X24 W6X15 W6X15
This section presents the sizing optimization examples of steel BR2 W6X15 W6X15 W4X13 W4X13
frame structures subjected to earthquake excitations. The in- BR3 W6X15 W6X15 W6X12 W4X13
vestigated examples consist of a 125-member 5-story and a BR4 W4X13 W6X15 W6X12 W6X12
250-member 10-story steel braced frame dual system. Here, BR5 W6X9 W6X12 W6X9 W6X9
the optimization runs are performed using a regular PC with Weight (ton) 15.7921 17.3953 15.1755 15.4875
Intel Core i7-4720HQ, 2.6 GHz CPU and 8 GB RAM. Due to
the stochastic nature of the employed evolutionary optimiza- COL column group with respect to Fig. 6, BM beams, BR bracings
tion algorithm, the design instances are independently solved
ten times and the best solution obtained is reported herein as Each structure was first considered to rest on a rigid base.
the optimum design. For all the investigated examples, the Then, the effects of unbounded soil domain were simu-
maximum number of iterations and population size of the lated using the massless technique (Sect. 4.2). In addi-
optimization algorithm are set to 500 and 50, respectively. tion, two different scaling procedures (i.e., simple scal-
The wide-flange (W) profile list composed of 268 ready sec- ing and spectrum matching) were used in the time history
tions is used to size the structural members, and the material analysis. Consequently, for the instances considered, four op-
properties of the steel are taken as follows: modulus of elas- timum solutions were obtained: (1) rigid base with scaled
ticity (E) = 200 GPa, yield stress (Fy) = 248.2 MPa, and motions, (2) rigid base with RSPMatched motion, (3) soil-
weight per unit volume (ρ) = 7.85 ton/m3. structure system with scaled motions, and (4) soil-structure
Given the two-dimensional modeling of the soil-structure system with RSPMatched motions.
system, the foundation of the structures was modeled using
beam elements: Rigid foundation mat approximation was not 5.1 Design example 1: 5-story building
necessary. Euler-Bernoulli beam was used for the idealization
of structural elements. In order to eliminate numerical issue The soil-structure system shown in Fig. 6 is comprised of 125
associated with the linear rectangular plane strain elements frame members and 480 rectangular plane strain elements.
(i.e., shear locking), incompatible finite elements were utilized The numbers of nodes for the soil domain, superstructure,
in lieu of their conventional counterparts (Wilson 2002). The and the soil-structure interface are 539, 65 and 9, respectively.
stiffness matrix of the soil domain was obtained for an out-of- For the beam and brace members, individual groups were
plane length equal to the out-of-plane tributary length of the considered at each story level. Columns, on the other hand,
2D frame, i.e., 5 m for the models in this study (Takewaki et al. were assigned to two groups: first three stories and the last two
1998). stories. Dirichlet boundary conditions are applied on the

Fig. 6 Soil-structure model BR5 BM5


of 5-story steel braced frame BR4 BM4
COL2
4 3.5 m BM3
BR3
BR2 BM2 COL1
BM1
5m BR1
Optimum design of steel braced frames considering dynamic soil-structure interaction 1133

Iwate RSPMatched Northridge

-120. -111. -102. -92. -83. -74. -65. -55. -46. -37. -28. -18. -9. 0.

a) Compressive Vertical Stress (KPa)

b) Maximum Drift Ratio (1st Floor)


Fig. 7 Envelope responses of 5-story building

truncated soil domain. The design example includes 12 base assumption when subjected to the suite of ground
sizing design variables representing the steel profiles motions. For this case, the maximum seismic demands
selected for the design groups (Fig. 6). The strength were obtained from the load combination which included
and drift ratio constraints are imposed in accordance the Iwate ground motion. In other words, the strength re-
with the regulations presented in Sect. 2. Design opti- quirements in the design process were either due to gravity
mization of the 125-member steel braced frame is car- loading (majority of beams and columns) or the load com-
ried out, and the optimized designs are presented in bination which included gravity as well as Iwate record
Table 2. seismic loading (bracings). However, the simple scaling
Investigation of the results shows that a design weight procedure, although does not alter the nature of the earth-
of 17.4 ton is obtained for the 5-story structure with rigid quake in any sense (except the amplitude), almost always

Fig. 8 Typical convergence


history for 5-story building
(Case: SSI_RSPM)
1134 M. Bybordiani, S. K. Azad

overestimates the seismic demands given jaggedness in- Fig. 9. Similar to the first instance, columns were assigned at
herent to real motions especially in the lower period range every three-story level (the last floor is assigned to a different
(Bybordiani and Arıcı 2018). Thus, in order to obtain an group therefore). The design example includes 14 sizing de-
optimum solution with a motion exactly resembling the sign variables based on the 14 member groups shown in Fig.
target spectrum defined per (ASCE 2010), RSMatched 9. The same procedure (as described in the previous example)
Northridge earthquake record was also used in the time was implemented in order to impose strength and displace-
history analysis. The design weight of the 5-story building ment constraints. Optimum design of the 250-member steel
in this case was obtained as 15.5 ton. braced frame is carried out, and the final design weights are
In addition, the above-mentioned models were also ana- presented in Table 4.
lyzed considering SSI effects. Accordingly, the design weight Investigation of the results shows that a design weight of
of the structure subjected to the suite of three motions was 49.2 ton is obtained for the 10-story structure with rigid base
obtained as 15.8 ton. Similar to the case with rigid base, the assumption when subjected to the suite of ground motions.
Iwate record led to highest seismic demands among the suite For this case, the maximum seismic demands were obtained
comprised of three motions. When the same model was sub- from the load combination which included Loma Prieta
jected to spectrally matched motion, a design weight of ground motion. The change in the ground motion leading to
15.2 ton was obtained. Here, the difference was not in the higher seismic demands from design example 1 to 2 stems
order of the rigid counterparts. This partly could be the result from two facts: (1) the relative spectral acceleration of the
of high spiky shape of response spectra for the individual ground motion at different periods (Fig. 4a) and (2) more
motions which makes it almost impossible to follow a simple contribution of higher modes to the dynamic response in the
trend. Envelope of the compressive vertical stresses for the case of taller buildings. In other words, the frequency content
Iwate and RSPMatched motions is given in Fig. 7a. As shown of the ground motions combined with the complete FRF de-
in Fig. 7b, the maximum drift ratio was observed in the first termines the seismic response, further underlying the effect of
floor for both records. Yet, the drift ratio demands were almost inter-set variability on obtaining engineering demand param-
17% of the allowable limits (ASCE 2010): For all the models eters of interest. In order to obtain an optimum solution with a
considered, strength requirements were the governing factor, motion exactly matching the target spectrum, RSMatched
drift ratio constraints never worked. Figure 8 depicts a typical Northridge motion was again used in the time history analysis.
convergence history of the optimization algorithm for 5-story With a slight decrease, the design weight of the 10-story build-
building instance. The statistical results of ten optimization ing was obtained as 48.3 ton.
runs as well as CPU time for this test example are presented In addition, the above-mentioned models were also ana-
in Table 3. In terms of the computational efficiency, the results lyzed considering SSI effects. Accordingly, the design weight
clearly indicate the usefulness of the UBS in reducing the total of the structure subjected to the suite of three motions was
computational time of the optimization process. obtained as 48.5 ton. Similar to the case with rigid base, the
Loma Prieta record led to highest seismic demands within the
5.2 Design example 2: 10-story building suite comprised of three motions. When the same model was
subjected to spectrally matched motion, a design weight of
In the second example, a 10-story steel frame was considered 47.9 ton was obtained. Given the pronounced contribution
(Fig. 9). The soil-structure model is comprised of 250 frame of the higher modes for 10-story frames, and the jaggedness
and 768 plane strain elements, respectively. The number of of the simply scaled motions, identifying a clear trend for
nodes for the soil domain, superstructure, and the interface different cases was not possible: Similar design weights in a
are 824, 90, and 9, respectively. Beam and brace members range of ± 3% were obtained. Envelope of the compressive
were assigned to groups at every two-story level as shown in vertical stresses for the Loma Prieta and RSPMatched motions

Table 3 Statistical results of ten


runs and CPU time for 5-story 5-story steel frame SSI Rigid SSI_RSPM Rigid_RSPM
steel frame
Best weight (ton) 15.7921 17.3953 15.1755 15.4875
Worst weight 18.02 20.78 18.59 17.67
Mean weight 17.11 18.48 16.76 16.32
Standard deviation 0.94 1.04 1.26 0.80
CPU time with UBS (h) 21.3 12.8 11.8 7.8
Expected CPU time without UBS 133.3 106.9 78.5 59.7
Saving in structural analyses by UBS (%) 84 88 85 87
Optimum design of steel braced frames considering dynamic soil-structure interaction 1135

BR5 BM5 COL4

BR4 BM4
COL3
BR3 BM3
BM2 COL2
BR2
9 3.5 m BM1
BR1 COL1

4.5 m

Fig. 9 Soil-structure model of 10-story steel braced frame

is given in Fig. 10a. As shown in Fig. 10b, the maximum drift was due to Iwate record. Given that the relative difference in
ratio was observed in the ninth floor for both of the records. the spectral accelerations of records in the suite comprised of
This further stresses the considerable contribution of specifi- three records, Loma Prieta ground motion led to the highest
cally second mode to the seismic response. However, the drift demands for the 10-story building. In the design process, the
ratio demands were around 28% of the allowable limits strength requirements were the governing constraints, regard-
(ASCE 2010): For all the models considered, strength require- less of the building height, presence of soil, and the input
ments were the governing factor, and drift ratio constraints ground motion. In other words, the maximum drift ratios for
never worked. The statistical results of ten optimization runs the 5- and 10-story frames were 17 and 28% of the allowable
as well as CPU time for 10-story steel frame instance are given
in Table 5. Table 4 Optimized designs obtained for 10-story steel frame

Groups SSI Rigid SSI_RSPM Rigid_RSPM

6 Summary and conclusions COL1 W10X60 W12X65 W10X60 W10X60


COL2 W14X43 W8X40 W14X43 W8X40
In this work, optimization of the steel braced frames under COL3 W8X28 W8X28 W8X28 W8X28
seismic loading in a dynamic framework was studied. The test COL4 W10X15 W6X12 W10X15 W10X15
examples including 5 and 10-story buildings were first ana- BM1 W12X19 W12X19 W12X19 W12X19
lyzed with the rigid base assumption. Soil domain was also BM2 W12X19 W12X19 W12X19 W12X19
considered and the stiffness and radiation damping character- BM3 W12X22 W12X19 W12X22 W12X19
istics of the foundation were simulated using the standard BM4 W12X22 W12X19 W12X22 W12X22
massless foundation technique and verified with refer- BM5 W12X22 W12X22 W12X22 W12X22
ence solutions. In the time history analysis of the afore- BR1 W8X31 W8X31 W8X31 W8X31
mentioned models, a suite comprised of three motions BR2 W8X24 W8X28 W6X20 W8X24
was chosen and scaled in order to conform to the seis- BR3 W6X20 W8X24 W6X20 W8X24
mic hazard defined using a target spectrum. Seismic demand BR4 W6X20 W6X20 W6X20 W6X20
predictions in an average sense were also obtained using BR5 W6X15 W6X15 W6X15 W6X15
RSPMatched Northridge record. Weight (ton) 48.5421 49.2001 47.9437 48.3411
Investigation of the design optimization results indicated
that the highest seismic demands for the 5-story structure COL column group with respect to Fig. 9, BM beams, BR bracings
1136 M. Bybordiani, S. K. Azad

Loma Prieta RSPMatched Northridge

Z Z

X X

-180. -166. -152. -138. -125. -111. -97. -83. -69. -55. -42. -28. -14. 0.

a) Compressive Vertical Stress (KPa)

b) Maximum Drift Ratio (9th Floor)


Fig. 10 Envelope responses of 10-story building

limit. The highest level of the drift ratio was observed at first counterpart. However, the difference in the optimum design
and ninth floor for 5- and 10-story structures, respectively. It weight is not considerable as the seismic demands are not the
was also concluded that although consideration of dynamic only governing parameter. In fact, one of the main findings of
SSI reduced the seismic demands to some extent, given the the present work is that although there can be a considerable
final design pertains to different load combinations, the opti- decrease in the seismic demands because of SSI, a complete
mum weight difference between models resting on a rigid base design where all types of loads and associated combinations
and half-space did not exceed 10 and 2% for 5- and 10-story are considered is not too different from a rigid base assumption
steel frames, respectively. counterpart.
Decrease in the seismic demands cannot directly be associ- Finally, in terms of the computational efficiency, the nu-
ated with the decreased fundamental frequency given jagged- merical results indicated the usefulness of the UBS in reduc-
ness of the ground motions’ spectra. However, considering ing the total computational effort of the optimization process.
decrease in the resonant amplitude because of radiation effects, It is also worthwhile to note that the above-mentioned conclu-
the dynamic response is expected to decrease. Interaction of sions regarding dynamic SSI are based on the performance of
the aforementioned effects may lead to a high range of seismic the employed metaheuristic algorithm in the investigated test
demand decrease. In the present study, as much as 35% de- examples of the present study and should not be considered as
crease was observed in base shear demand for 5-story building general conclusions valid for the cases not investigated here.
subjected to Iwate record when compared to rigid base Nevertheless, the results obtained may provide some general

Table 5 Statistical results of ten


runs and CPU time for 10-story 10-story steel frame SSI Rigid SSI_RSPM Rigid_RSPM
steel frame
Best weight (ton) 48.54 49.2001 47.9437 48.3411
Worst weight 57.84 59.44 57.64 58.78
Mean weight 51.29 53.13 53.77 53.65
Standard deviation 2.74 3.85 4.17 4.11
CPU time with UBS (h) 28.6 25.1 13.1 10.1
Expected CPU time without UBS 168.1 139.5 93.6 67.1
Saving in structural analyses by UBS (%) 83 82 86 85
Optimum design of steel braced frames considering dynamic soil-structure interaction 1137

guidelines for future studies on design optimization of steel Kaveh A, Abbasgholiha H (2011) Optimum design of steel sway frames
using big bang big crunch algorithm. Asian J Civil Eng 12(3):293–
frames considering dynamic SSI.
317
Kazemzadeh Azad S, Hasançebi O, Kazemzadeh Azad S (2013) Upper
Acknowledgements The authors would like to thank Mr. Sina bound strategy for metaheuristic based design optimization of steel
Kazemzadeh Azad for his helpful comments on this work. frames. Adv Eng Softw 57:19–32
Kazemzadeh Azad S, Bybordiani M, Kazemzadeh Azad S, Jawad FKJ
(2018) Simultaneous size and geometry optimization of steel trusses
References under dynamic excitations. Struct Multidiscip Optim 58(6):2545–
2563
Kripakaran P, Hall B, Gupta A (2011) A genetic algorithm for design of
Al Atik L, Abrahamson N (2010) An improved method for nonstationary
moment-resisting steel frames. Struct Multidiscip Optim 44(4):559–
spectral matching. Earthquake Spectra 26(3):601–617
574
American Institute of Steel Construction (AISC) (1994) Manual of steel
construction, load & resistance factor design, 2nd edn. Chicago Kurama YC, Farrow KT (2003) Ground motion scaling methods for
ASCE (2010) Minimum design loads for buildings and other structures. different site conditions and structure characteristics. Earthq Eng
ASCE/SEI-7-10. Structural engineering institute of the American Struct Dyn 32(15):2425–2450
society of civil engineers, Reston, p 608 Lamberti L, Pappalettere C (2011) Metaheuristic design optimization of
Baker JW (2010) Conditional mean spectrum: tool for ground-motion skeletal structures: a review. Comput Technol Rev 4(1):1–32
selection. J Struct Eng 137(3):322–331 NEHRP (2011) Selecting and scaling earthquake ground motions for
Bielak J (1976) Modal analysis for building-soil interaction. J Eng Mech performing response-history analyses
Div 102(5):771–786 Reyes JC, Riaño AC, Kalkan E, Quintero OA, Arango CM (2014)
Bybordiani M, Arıcı Y (2017) The use of 3D modeling for the prediction Assessment of spectrum matching procedure for nonlinear analysis
of the seismic demands on the gravity dams. Earthq Eng Struct Dyn of symmetric- and asymmetric-plan buildings. Eng Struct 72:171–
46(11):1769–1789 181
Bybordiani M, Arıcı Y (2018) Effectiveness of motion scaling procedures Saka M (2007) Optimum design of steel frames using stochastic search
for the seismic assessment of concrete gravity dams for near field techniques based on natural phenomena: a review. Civil Eng
motions. Struct Infrastruct Eng:1–16 Comput: Tools Tech 6:105–147
Bybordiani M, Arici Y (2019) Structure‐soil‐structure interaction of Saka MP, Hasançebi O, Geem ZW (2016) Metaheuristics in structural
adjacent buildings subjected to seismic loading. Earthq Eng optimization and discussions on harmony search algorithm. Swarm
Struct Dyn. https://doi.org/10.1002/eqe.3162 Evol Comput 28:88–97
Chopra AK (2012) Dynamics of structures: theory and applications to Seed HB, Idriss IM (1969) Soil moduli and damping factors for dynamic
earthquake engineering, 4th edn. Prentice Hall, Englewood Cliffs response analyses. Earthquake Engineering Research Center,
Daloglu AT, Artar M, Özgan K, Karakas Aİ (2016) Optimum design of University of California, Berkeley
steel space frames including soil-structure interaction. Struct Takewaki I, Nakamura T, Hirayama K (1998) Seismic frame design via
Multidiscip Optim 54(1):117–131 inverse mode design of frame-ground systems. Soil Dyn Earthq Eng
Erol OK, Eksin I (2006) A new optimization method: big bang–big 17(3):153–163
crunch. Adv Eng Softw 37(2):106–111 Trifunac M (1972) Scattering of plane SH waves by a semi-cylindrical
FEMA450-1 (2003) NEHRP recommended provisions for seismic regu- canyon. Earthquake Engineering & Structural Dynamics 1(3):267–
lations for new buildings and other structures, Part 1: Provisions. 281
Washington, United States United States Army Corps of Engineers (2003) Time-history dynamic
Gazetas GC, Roesset JM (1976) Forced vibrations of strip footings on analysis of hydraulic concrete structures. Department of the Army,
layered soils. In methods of structural analysis. ASCE Washington, DC, p 401
Gholizadeh S, Milani A (2016) Optimal performance-based design of
Wang J, Lu D, Jin F, Zhang C (2013) Accuracy of the half-power band-
steel frames using advanced metaheuristics
width method with a third-order correction for estimating damping
Gholizadeh S, Poorhoseini H (2016) Seismic layout optimization of steel
in multi-DOF systems. Earthq Eng Eng Vib 12(1):33–38
braced frames by an improved dolphin echolocation algorithm.
Struct Multidiscip Optim 54(4):1011–1029 Wilson EL. (2002) Three-dimensional static and dynamic analysis of
Hasançebi O, Kazemzadeh Azad S (2012) An exponential big bang-big structures, 3rd edn. Computers and Structures, Inc
crunch algorithm for discrete design optimization of steel frames. Wolf JP (1985) Dynamic soil-structure interaction. Prentice-Hall,
Comput Struct 110-111:167–179 Englewood Cliffs
Hasançebi O, Kazemzadeh Azad S (2014) Discrete size optimization of
steel trusses using a refined big bang–big crunch algorithm. Eng
Optim 46(1):61–83
Hasançebi O, Çarbaş S, Doğan E, Erdal F, Saka MP (2010) Comparison Publisher’s note Springer Nature remains neutral with regard to jurisdic-
of non-deterministic search techniques in the optimum design of real tional claims in published mapsand institutional affiliations.
size steel frames. Comput Struct 88(17):1033–1048
Kameshki ES, Saka MP (2001) Optimum design of nonlinear steel frames
with semi-rigid connections using a genetic algorithm. Comput
Struct 79(17):1593–1604

You might also like