Vous êtes sur la page 1sur 12

Advances in Engineering Software 115 (2018) 149160

Contents lists available at ScienceDirect

Advances in Engineering Software


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

Research paper

Optimum design of 3D reinforced concrete frames using DMPSO algorithm T


a,* a b c
M.J. Esfandiari , G.S. Urgessa , S. Sheikholaren , S.H. Dehghan Manshadi
a
Department of Civil, Environmental and Infrastructure Engineering, George Mason University, Fairfax, VA, USA
b
Department of Civil Engineering Science, University of Johannesburg, Johannesburg, South Africa
c
Department of Civil Engineering, Yazd Branch, Islamic Azad University, Yazd, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: Evolutionary optimization algorithms are widely used for solving optimization problems. However, in the realm
Optimization of structures of real-world structures nding the optimum design would be dicult and time-consuming due to large number
Multi-criterion decision-making of design variables, enormous size of the search space, and availability of numerous constraints. As a result,
Particle swarm optimization holistic optimization approaches that consider architectural and other practical requirements in addition to
required code provisions are rather limited in real-world structures. This paper presents an innovative algorithm
combining multi-criterion decision-making (DM) and Particle Swarm Optimization (PSO), which is called
DMPSO, for accelerating convergence toward optimum solution in 3D reinforced concrete (RC) frames. All
practical requirements are considered for obtaining directly constructible designs without any further mod-
ications. The eectiveness of the proposed algorithm is illustrated in optimization of 3D RC frames subjected to
lateral seismic forces according to ASCE 7 requirements. The results conrm the ability of the proposed DMPSO
algorithm to eciently nd optimal solutions for 3D RC optimization problems.

1. Introduction a highly valuable tool for handling such problems.


Dierent computation approaches exist for evaluating multiple
Although numerous optimization techniques have been developed conicting criteria in decision making. Saka et al. [3] and Kaveh [4]
over the years, real-world RC frames are designed by using the tradi- include extensive reviews of both deterministic and stochastic optimi-
tional trial-and-error approach. The traditional approach usually does zation algorithms. Two examples of algorithms relevant to the work
not yield economical designs while satisfying safety criteria [1]. This present here include Multiple Criteria Decision Making (MCDM) [57]
can be explained by the presence of various design variables, the large and Evolutionary Optimization (EO) [8,9]. Although both areas address
size of the search space, and availability of numerous constraints [2]. similar problems as emphasized, they have dierent goals [10]. MCDM
More precisely, a reinforced concrete member can be designed with a supports a human decision maker in identifying the most preferred
semi-innite set of member dimensions and dierent arrangements of solution. In other words, a decision maker checks the results in every
reinforcing bars. Moreover, in optimization of reinforced concrete iteration and indicates what kind of changes in the design variables
structures the cost of dierent materials (i.e. steel, concrete, and labor), would lead to a more preferred solution. In addition, the decision maker
which are closely tied together, should be considered. The im- can ignore infeasible designs or designs that cannot result in better
plementation of optimum design formulations for real-world RC struc- optimal solutions before the calculation proceeds. Consequently, the
tural problems requires signicant computational eort because of the computational resources available are not wasted because such optimal
need for accurate numerical simulation and analysis of the structural solutions that are interesting to the decision maker are generated. This
system. Moreover, it should calculate response quantities, such as dis- method has a strong disadvantage in that a human decision maker must
placements and stresses, under various loading conditions. Therefore, be available and willing to actively participate in the solution process
the design procedure is needed for performing the constraints checks and direct the solution according to identied preferences.
imposed by design codes or other architectural/constructional re- On the other hand, EO works with a population of individuals and
quirements [1]. This huge computational cost makes optimization attempts to nd a set of non-dominated solutions near the optimal so-
procedures very dicult in designing real-world structures. Conse- lution. Typically, EO algorithms explore the design space thoroughly
quently, an optimization algorithm that satises design requirements, and do not involve any preference information [11]. Many researchers
economic, and manufacturability criteria in a reasonable time would be have shown that EOs perform well for global searching due to their

*
Corresponding author.
E-mail address: Mesfandi@gmu.edu (M.J. Esfandiari).

http://dx.doi.org/10.1016/j.advengsoft.2017.09.007
Received 6 July 2017; Received in revised form 31 July 2017; Accepted 17 September 2017
Available online 24 September 2017
0965-9978/ 2017 Elsevier Ltd. All rights reserved.
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

capability of exploring and nding promising regions in the search 2.1. Objective function
space, but they take a relatively long time to converge to an optimum
solution [8,9,12,13]. However, only few studies utilized EOs to solve In structural optimization problems, the objective function is gen-
structural optimization 3D problems mostly focusing on steel frames. erally described as the weight or total cost of the structure. When total
For example, Hasanebi et al. [14] utilized genetic algorithms, simu- cost is the case, the costs of concrete, steel and labor is considered.
lated annealing, evolution strategies, particle swarm optimizer, tabu However, when objective function is total weight of the structures, only
search, ant colony optimization and harmony search to develop seven the weight of concrete and steel are included in the problem.
optimum design algorithms for real size rigidly connected steel frames. Eq. (3) shows the objective function.
In addition, they studied optimum design of pin jointed structures [15].
fObj = Objc + Objs + Objf (3)
For the sake of the optimum design problem of reinforced concrete
frames according to ACI 318, its formulation has been carried out in where Objc, Objs, and Objf are the cost of concrete, the cost of reinfor-
some literature. Akin and Saka studied optimum design of special cing steel bars, and the cost of labor (includes labor and placement),
seismic moment 2D RC frames under earthquake loads [16]. Fadaee respectively, when the objective function is dened as a total cost of the
and Grierson [17] optimized the cost of 3D skeletal structures using frame. Otherwise, Objc and Objs are the weight of concrete, the weight
optimality criteria. They proposed a computer-based method for the of reinforcing steel bars, respectively. In this case, Objf is considered to
optimal design of three-dimensional reinforced concrete (RC) skeletal be equal to zero.
structures having members subjected to biaxial moments, biaxial shears Eqs. (4)(6) show the costs of each cost components when the ob-
and axial loads. Balling and Yao [18] optimized 3D frames with a multi- jective function is dened as a total cost of the frame.
level method by decomposing the problem into a system optimization
problem and a series of individual member optimization problems. Ncol Nbeam

Kaveh and Behnam [19] tried a charged system search (CSS) algorithm Objc = CC
bi . di . Lncolumn, i + bwj . hj . Lbeamj

for the optimization of 3D RC frames. In all of the aforementioned i=1 j=1 (4)
studies, the design variables were taken from databases, which are
Ncol Nbar , i Ncol Ntie, i
usually populated with a relatively small number of cross-section types
and sizes.
Objs = Cs. s. ASt j . Lbar j + AShk . Ltiek
In this article, we present an algorithm called DMPSO combining i=1 j=1 i=1 k=1

Nbeam Nbar , m Ncol Ntie, m


the Particle Swarm Optimization (PSO) algorithm with the basic con-
+ AStl . Lbarl + AShn . Lbarn

cept of multi-criterion decision-making (MCDM) for optimizing 3D re-
inforced structures. Esfandiari et al. utilized it for 2D frames and the
m=1 l=1 m=1 n=1
(5)
result was satisfying [20]. This algorithm has a signicant acceleration Ncol
towards convergence rate. The minimum restriction for design vari- Objs = Cf
[2(bi + hi ) . Lncolumn, i ]
ables, the modular sizes of members, standard reinforcement bar dia-
i=1
meters, spacing requirements of reinforcing bars, architectural re- Nbeam Ncol
quirements and other practical requirements, in addition to relevant + [ (bwj + 2hj) . Lbeamj ] bk + dk

provisions, are considered to obtain directly constructible designs that j=1 k=1
(6)
will need no further modications. Furthermore, 3D RC benchmark
design examples were tested using DMPSO and the results were com- where Ncol, Nbeam, b, d, bw, h, L, and Ln are the number of column
pared with those presented in the literature. The numerical results members, the number of beam members, the width of column section,
showed that the proposed DMPSO algorithm performed better than the depth of column section, the width of beam section, the height of
other established EO and hybrid EO algorithms in terms of convergence beam section, the length of the members, and the length of clear span
speed without sacricing quality of results. measured face-to-face to the supports, respectively; CC, Cf and Cs are
unit cost of the concrete, the labor and the steel, respectively; Ast, Lbar,
2. Formulation of the structural optimization problem for and Nbar are the Area, the length and the number of longitudinal re-
reinforced concrete frames inforcement bars placed in the member while Ash, Ltie, and Ntie are the
area, the length and the number of shear reinforcement bars (ties) used
Eq. (1) shows a general structural optimization problem. in the member respectively. s is the density of steel reinforcements
(kg/m3).
minn f (x ) x = [x1, , x n]T
x
gk (x ) 0, k = 1, , m 2.2. Design variables
xL x xU (1)
A semi-innite set of member sizes and steel reinforcement ar-
n
where x is a vector of n design variables, f(x):  is the objective rangements can be considered for RC structure elements. In that case, as
function which returns a scalar value to be minimized (usually the cost the dimensions of the design space are very large, the computational
or the weight of the frame), the vector function g(x): n m returns burden of the optimization process increases [21]. Consequently, in the
a vector of length m containing the values of the inequality constraints studies available in the literature, a countable number of cross-sections
evaluated at x, and xL, xU are two vectors of length n containing the have been employed in order to reduce the dimensions of the design
lower and upper bounds of the design variables, respectively. The above space and the computational cost. While the DMPSO performs well in
mathematical formulation contains only inequality constraints, as terms of convergence speed and does not waste the computational re-
equality constraints are usually not found in the case of structural op- sources, no restriction is needed to be applied to the size of sections and
timizations. arrangement of steel reinforcements. The algorithm can use all of the
Eq. (2) shows a typical constraint k in a structural optimization possible sections in the common practical range to nd the best op-
problem. timum solution for the problem. The only constraints considered for the
gk (x ) = qk (x ) qallow, k sections were those derived from the design code provisions of ACI 318
(2)
[22]. Eighteen design variables were dened for each span of a beam
where qk(x) is a response measure for design x and qallow,k is its max- which can enumerate the width and height of a cross section, number
imum allowable absolute value. and diameter of top and bottom continuous bars, number and diameter

150
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Fig. 1. Typical beam and column details.

of top and bottom additional bars at left support, number and diameter 2.3. Constraints
of top and bottom additional bars at midspan, number and diameter of
top and bottom additional bars at right support. The design variables Constraints derived from the design provisions of ACI 318 [22] for
for the column section were selected as the dimensions of columns in x intermediate moment reinforced concrete frames are strength, servi-
and y directions, the diameter of reinforcement bars at the cross-section ceability, ductility and other supplementary constraints. Specically,
of column and number of reinforcement bars in both sides of the the constraints considered for the RC members can be categorized into
column. two main types. The rst type comprises constraints on the load-car-
It is important to emphasize that all the rebars were not placed rying capacities of the sections, clear spacing limits between reinforcing
continuously across the entire length of the beams. Our DMPSO algo- bars, and the minimum and the maximum reinforcement areas of the
rithm calculated the needed rebars for each section and placed it in the members. The second type consists of constraints dening architectural
right position. The development length of reinforcement, location and requirements, constructible designs, and detailing practices. These in-
length of lap splices, and length for bars were calculated exactly ac- clude the minimum and the maximum dimensions of the member's
cording to ACI 318 [22]. According to the code splices should, if pos- section, the maximum aspect ratio of the section, the maximum number
sible, be located away from points of maximum tensile stress; so, in the of reinforcing bars and miscellaneous reinforcement requirements. The
top of the beams the splices are located at the middle of the beam's constraints, which can be imposed to column groups, beam groups or
length and the bottom bars are lapped at a third of the beam's lengths. A joint regions, are expressed in a normalized form in the following sec-
typical beam and a typical column with the considered splices details tions.
are shown in Fig. 1.
Considering the ACI 318 [22] code and other construction re-
2.3.1. Constraints for beam groups
quirements, The lower bound, upper bound, and increments of cross
At every section of a exural member where tensile reinforcement is
sectional dimensions were considered as 300, 1000, and 50 mm, re-
required by analysis, the tension area of longitudinal steel reinforce-
spectively. Moreover, at least 1 bar was used in each corner of the cross
ment, As, should satisfy the minimum and the maximum requirements
sections of the members. The cover of concrete was taken as
required by the ACI design code. It should be noted that these minimum
tc = 40 mm. For column sections, a symmetrical pattern for bars was
and maximum tensile reinforcement requirements are required for both
considered. In this study, minimum diameter of transversal steel is
positive and negative moment regions. Eqs. (7)(8) show these con-
considered as 10.
straints.

As, min, i
Gb1 (x ) = 1 0 i = 1, , Nbeam
As, i (7)

151
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

As, i Mu (i, j, k )
Gb2 (x ) = 1 0 i = 1, , Nbeam Gb9 = 1 0 i = 1, , Nbeam j = 1, , Nload combination k
As, max , i (8) Mn, i

where i is an index representing the number of the beam and Nbeam is = 1, 2


the total number of beams. The minimum area of reinforcement is a (15)
function of the width, bw, and the distance from extreme compression 1
M
ber to centroid of longitudinal tension reinforcement of the beam Gb10 = 3 n, end, i
1 0 i = 1, , Nbeam
Mn, i +
section, d. The maximum amount of reinforcement is calculated ac- (16)
cording to the fact that the net tensile strain in the extreme tension
1
steel, t, should be equal to or greater than 0.005 when the concrete in M
5 n, end, i
Gb11 = 1 0 i = 1, , Nbeam
compression reaches its assumed strain limit of 0.003. Mn, i+ (17)
A minimum area of shear reinforcement, Av, min, shall be provided in
all reinforced concrete exural members where Vu exceeds 0.5Vc. 1
M
5 n, end, i
Eq. (9) shows this constraint. Note than Av,min is a function of the Gb12 = 1 0 i = 1, , Nbeam
Mn, i (18)
spacing between stirrups, yield strength of reinforcing steel and the
width of the section. where k represents the negative moment and positive moment situa-
tions. Nload combination is total number of the combination loads.
Av, min, i
Gb3 (x ) = 1 0 i = 1, , Nbeam For each section of the beams, the shear force capacity, Vn,
Av, i (9) should be greater than the factored shear force at the section, Vu. In
The minimum clear spacing between parallel bars in a layer should addition, the factored shear force at the section, Vu, should be less than
be greater than the maximum of the nominal diameter of bars l and allowed maximum shear force capacity, Vmax. Furthermore, Vn of
25 mm. However, for constructional requirements, the minimum free beams resisting earthquake eect, E, should be greater than the allowed
distance between the longitudinal bars was considered as Sb = 40 mm. minimum shear force capacity, Vmin. The corresponding constraints are
In the case that parallel reinforcement was placed in two layers, bars in given in Eqs. (19)(21).
the upper layers were placed directly above bars in the bottom layers Vu (i, j)
with a clear distance of 25 mm. Eq. (10) shows this constraint. Gb13 = 1 0 i = 1, , Nbeam j = 1, , Nload combination
Vn, i (19)
Sb, min
Gb4 (x ) = 1 0 i = 1, , Nbeam Vu, i, j
Sb, i (10) Gb14 = 1 0 i = 1, , Nbeam j = 1, , Nload combination
Vmax (20)
The spacing between stirrups, Sv, should satisfy the maximum re-
Vmin
quirements permitted by the design code as shown in Eq. (11). Gb15 = 1 0 i = 1, , Nbeam
Vn, i (21)
S v, i
Gb5 (x ) = 1 0 i = 1, , Nbeam It should be noted that constraints Gb1, Gb2,, Gb8 do not need nite
Sv, max , i (11)
element analysis for their calculation; while, the remaining constraints
Also in this study, the minimum spacing between stirrups, Sv, is of the beams need nite element analysis.
limited to 50 mm because of constructional requirements. The corre-
sponding constraint is given in Eq. (12). 2.3.2. Constraints for column groups
The width, b, and the height, h, of a column section should not be
Sv, min less than the minimum dimensions limit value given for columns.
Gb6 (x ) = 1 0 i = 1, , Nbeam
S v, i (12) Eqs. (22) and (23) show these constraints.

The height of the beams, h, should be greater than the allowable bmin, i
Gc1 (x ) = 1 0 i = 1, , Ncolumn
minimum height for beams, hmin, and less than the maximum height bi (22)
limit value given for beams, hmax. These constraints are shown in
Eqs. (13) and (14). hmin,i
Gc 2 (x ) = 1 0 i = 1, , Ncolumn
hi (23)
hmin, i
Gb7 (x ) = 1 0 i = 1, , Nbeam where i is an index representing the number of the columns and Ncolumn
hi (13)
is the total number of columns. The minimum width, bmin, and height,
hi hmin, should be dened for each problem separately according to its
Gb8 = 1 0 i = 1, , Nbeam architectural requirements.
hmax , i (14)
The area of longitudinal reinforcement, Ast, in a column section
The minimum height limit is a function of the length of clear span of should be between the minimum and maximum limits, Ast, min, per-
the beam measured face-to-face to the supports and its ends connection mitted by design specication which are 1% and 8% of the gross area of
conditions. The maximum height limit value should be dened for each concrete section, respectively. The corresponding constraints are given
problem separately according to its architectural requirements. in Eqs. (24) and (25).
For each section of the beams, the negative and the positive reduced Ast, min, i
moment strength of the section Mn, should be greater than the fac- Gc3 (x ) = 1 0 i = 1, , NColumn
Ast, i (24)
tored moment force at the section, Mu. Besides, the positive moment
strength at any end (support) of the beams, Mn,end + , should be not less Ast, i
than one-third the negative moment strength, Mn,end . Furthermore, Gc 4 (x ) =
Ast, max , i
1 0 i = 1, , NColumn
(25)
neither the negative nor the positive moment strength at any section
along the length of the beam, Mn + , Mn , should be less than one-fth The spacing between longitude steel reinforcements, Sc, should
the maximum moment strength at the ends of the beams, Mn, end. satisfy the minimum and maximum requirements permitted by design
Eqs. (15)(18) show all constraints related to the moment strength of specication, Sc, min and Sc, max, as well as constructional limits
the sections. (Sc, min = 40 mm);

152
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Sc, min
Gc5 (x ) = 1 0 i = 1, , Ncolumn
Sc, i (26)

Sc, i
Gc6 (x ) = 1 0 i = 1, , Ncolumn
Sc, max (27)
A minimum area of shear reinforcement, Ash, min, should be provided
in all reinforced concrete exural members where factored shear force
at section, Vu, exceeds 0.5Vc as shown in Eq. (28).
Ash, min, i
Gc 7 (x ) = 1 0 i = 1, , Ncolumn
Ash, i (28)
The spacing between stirrups, Scv, should satisfy the maximum re-
quirements permitted by design specication, Scv, max. It should be
noted that Scv, max, which is a function of the cross section diameters and
diameter of longitudinal or transverse steel reinforcements, dier along
the beam. Also, the minimum spacing between stirrups of the columns,
Scv, min, is limited to 50 mm because of constructional requirements.
Eqs. (29) and (30) show the corresponding constraints.
Scv, i
Gc8 (x ) = 1 0 i = 1, , Ncolumn
Scv, max , i (29)

Scv, min
Gc 9 (x ) = 1 0 i = 1, , Ncolumn
Scv, i (30)
For each section of the columns, the strength capacity of the sec-
tion, n, should be greater than the stress of the applied force, u, as
shown in Eq. (31).
Fig. 2. A typical column interaction surface (P0 is axial tension and +P0 is axial
u (i, j) compression force).
Gc10 (x ) = 1 0 i = 1, , Ncolumn j = 1, , Nload combination
n, i
(31) element analysis for their calculation; while, the remaining constraints
of columns need nite element analysis.
The strength of the columns should be obtained by the interaction
surface, which is numerically described by a series of discrete points
generated on the three-dimensional interaction failure surface. For each 2.3.3. Constraints for joints
section of the column, the point from the applied force and moments At frame joints, the width of beams, bw, should be smaller than the
should lies within the interaction volume. In this study, to approximate corresponding dimensions of the intersecting columns, dc. Furthermore,
the curves with acceptable accuracy, 11 interaction points for each at the joints with two columns, the cross-sectional dimension of bottom
curve was considered. Furthermore, 24 interaction curves were applied column, bb and db, should be equal or greater than the corresponding
to obtain the whole surface. A typical interaction diagram is shown cross-sectional dimension of the top column, bt and dt. Eqs. (35)(37)
Fig. 2. Furthermore, according to the ACI code [22], for compression show the corresponding constraints.
members not braced sideways, the slenderness eects can be neglected, b w, i
when klu/r < 22. And when slenderness eects are not neglected as Gj1 (x ) = 1 0 i = 1, , Njoint
dc, i (35)
permitted above, the design of compression members, restraining
beams, and other supporting members shall be based on the factored bt , i
Gj2 (x ) = 1 0 i = 1, , Njoint
forces and moments from a second-order analysis. To satisfy this ob- bb, i (36)
ligation, the second-order eects are considered along the length of
compression members in this study. dt , i
Gj3 (x ) = 1 0 i = 1, , Njoint
For each section of the columns, the shear force capacity, Vn, db, i (37)
should be greater than the factored shear force at the section, Vu. In
addition, the factored shear force at the section, Vu, should be less than where Njoint is the total number of joints of the frame.
allowed maximum shear force capacity, Vmax. Furthermore, Vn of The design story drift, , should not exceed the allowable story
columns resisting earthquake eect, E, should be greater than the al- drift, a, as shown in Eq. (38)
lowed minimum shear force capacity, Vmin. Eqs. (32)(34) show the i
corresponding constraints. Gj 4 (x ) = 1 0 i = 1, , Nstory
a, i (38)
Vu (i, j)
Gc11 = 1 0 i = 1, , Ncolumns j = 1, , Nload combination where Nstory is the total number of stories in the frame.
Vn, i (32) In this study, the elastic drifts were determined using seismic design
forces that are based on the exact computed fundamental period of the
Vu, i, j
Gc12 = 1 0 i = 1, , Ncolumn j = 1, , Nload combination structure without any upper limit according to ASCE 7 [23]. It should
Vmax be noted that constraints Gj1, Gj2, and Gj3 do not need nite element
(33) analysis for their calculation; while, Gj4 needs nite element analysis.
Vmin
Gc13 = 1 0 i = 1, , Ncolumn 3. A brief description the proposed DMPSO algorithm
Vn, i (34)
It should be noted that constraints Gc1, Gc2,, Gc9 do not need nite In a particle swarm optimization (PSO) formulation, multiple

153
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Fig. 3. Flowchart of the optimization by DMPSO algorithm.

candidate solutions, called particles, y through the problem search according to its own experience, the experience of neighboring parti-
space looking for the optimal position. Each particle has a position and cles, and the preference of the decision maker. Furthermore, the DM
a velocity in the multidimensional design space. Additionally, it has a algorithm can ignore a solution at any point of the calculation process
tness value which is evaluated by the objective function at its current where it perceives that a better tness cannot be produced.
location. A particle by itself has almost no power to solve any problem; Consequently, the computational resources available are not wasted
progress occurs only when the particles interact. Consequently, each since only such optimal solutions, which are interesting to the DM, are
particle communicates with other particles to determine its movement generated. Finally, the basic PSO is easily trapped into a local
through the search space and adjusts its velocity and position according minimum. Considering that, the decision maker imposes its preference
to the best solution it has achieved so far as well as the best point found to intelligently escape from the local minimum.
by any member of its neighborhood. The next iteration takes place after The owchart of the optimization by the DMPSO algorithm is shown
all particles have been moved. This is expected to move the swarm schematically in Fig. 3.
toward the best solutions. However, metaheuristics such as PSO do not
guarantee the existence of an optimal solution. In addition, nding
3.1. Mathematical formulation of the DMPSO algorithm
optimum solutions for RC frames is time-consuming due to the presence
of many design variables, large size of the search space, and various
The velocity and position of the particles are updated in a stochastic
constraints which must be satised simultaneously. Consequently, we
way as shown in Eq. (39):
propose an innovative algorithm based on the concept of multi-criterion
decision-making (DM) in combination with the PSO algorithm to ac- v j (t + 1) = wv j (t ) + c1 r1 . (x Pb, j x j (t )) + c2 r2 . (x Gb, j x j (t ))
celerate convergence toward the optimum solution in structural opti-
+ c3 r3 . (x DM , j x j (t ))
mization problems.
The DM algorithm plays three major roles within the DMPSO al- x j (t + 1) = x j (t ) + v j (t + 1)
gorithm. First, it checks the results in each iteration and identies what (39)
kind of changes in the design variables would lead to a more preferred j j
where v (t) and x (t) represent the velocity and the position vectors of
solution. Therefore, the DMPSO adjusts its velocity and position
particle j at time t, respectively. The term w is the inertia weight, a

154
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

5 5 5

F G H

D D
5

E
C C
4

B B

A A
6

E F G H T5 T5

T30 T30 T30 T30


T2 T2 T2 T4 T4 T4 T7 T7 T7 T9 T9

T20 T20 T20 T20 T23 T23 T23 T23 T26 T26 T26 T26 T29 T29 T29
T1 T1 T1 T3 T3 T3 T6 T6 T6 T8 T8

T19 T19 T19 T19 T22 T22 T22 T22 T25 T25 T25 T25 T28 T28 T28
T1 T1 T1 T3 T3 T3 T6 T6 T6 T8 T8

T18 T18 T18 T18 T21 T21 T21 T21 T24 T24 T24 T24 T27 T27 T27

Secon A Secon B Secon C Secon D


T5 T5

T11 T11 T13 T13 T13 T15 T15 T15 T17 T17 T17

T10 T10 T12 T12 T12 T14 T14 T14 T16 T16 T16

T10 T10 T12 T12 T12 T14 T14 T14 T16 T16 T16

Secon E Secon F Secon G Secon H


Fig. 4. Structural geometry and type classication of beams and columns in the typical three-story RC structure with 3 spans in each horizontal direction.

89 3
x10 experiences and the decision maker's preference on the trajectory of
each particle, respectively. r1, r2, r3 are three random vectors with
84 numbers uniformly distributed in the interval [0, 1]. The symbol . is
the element-wise product of two vectors.
79 Particles velocities in each dimension i (i = 1, . . ., n) are re-
Cost ($)

stricted to a maximum velocity vimax . The vector vimax determines the


74 maximum change each dimension can undergo in its positional co-
ordinates during an iteration. It is more appropriate to use a vector
69 rather than a scalar, as in the general case dierent velocity restrictions
can be applied for dierent dimensions of the particle [24]. However,
64 providing that the particle moves outside the bounds for a dimension i
0.0 1.0 2.0 3.0 4.0 5.0 6.0 after the position update, xi xL or xU xi, the design variable xi
x103
Structural analyses limits the closest bound, xi = xL or xi = xU.
Fig. 5. Convergence history of DMPSO the typical three-story RC structure.
3.2. Constraint handling
scaling factor employed to control the exploration abilities of the
The PSO algorithm mainly focuses on nding an optimum solution
swarm. Vector xPb,j denotes the personal best position which is re-
in unconstrained problems. While most real-world applications have
corded by particle j, vector xGb,j is the global best position obtained by
constraints, there has been relatively little work related to the in-
the entire swarm up to the current iteration, and vector xDM,j indicates
corporation of constraints into the PSO algorithm [20]. Methods that
the position of preference of the decision maker in the search space. The
have been previously proposed for handling constraints can be grouped
acceleration coecients c1 and c2, and c3 are coecients which control
into four categories: (i) methods based on preserving feasibility of the
the impact of the particle's own experiences, the other particles'
solutions; (ii) methods based on penalty functions; (iii) methods that

155
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

1 search for feasibility; (iv) other hybrid methods [25].


While dierent constraint-handling techniques are implemented in
Demand capacity rao(DCR)

0.8 literature [26], we propose a nonlinear penalty approach in which some


levels of violation for each constraint are dened. If any of the con-
0.6 straints are violated, a penalty, whose value is related to the degree of
the violated constraint, is applied to the objective function as shown in
0.4 Eqs. (40)(43).

Ntotal
0.2
= k (x )
k=1 (40)
0
1 5 9 13 17 21 25 29
Number of beam and column type classicaons k (x ) = c (Gk (x ))2 (41)
Fig. 6. Maximum DCR of members for each type classication.

Table 1
The results of optimum design obtained by DMPSO.

Optimization results

Sectional dimensions Reinforcements

Member type Type classication Width (mm) Depth (mm) T-rebar left T-rebar middle T-rebar right C-rebar left C-rebar middle C-rebar right

Beam T1 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12


3-D18 3-D18 1-D18
Beam T2 300 400 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
1-D18 1-D18 1-D12
Beam T3 300 300 3-D12 3-D12 3-D12 3-D16 3-D12 3-D16
3-D18 3-D18 2-D22
Beam T4 350 350 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
2-D24 2-D24 2-D18
Beam T5 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D14 2-D14 2-D14
Beam T6 300 300 3-D12 3-D12 3-D12 2-D18 2-D18 2-D18
3-D18 3-D18 3-D16
Beam T7 300 300 3-D12 3-D12 3-D12 2-D18 2-D18 2-D18
3-D18 3-D18 3-D16
Beam T8 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D20 2-D20 1-D16
Beam T9 350 350 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
3-D12 3-D12 1-D12
Beam T10 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D16 3-D16 1-D18
Beam T11 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D14 2-D14 2-D14
Beam T12 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D18 3-D18 1-D16 2-D24 1-D16
Beam T13 300 550 3-D16 3-D16 3-D16 3-D16 3-D16 3-D16
1-D14 1-D14 1-D12
Beam T14 300 300 3-D12 3-D12 3-D12 2-D22 2-D20 2-D22
3-D18 3-D24 2-D16
Beam T15 300 550 4-D16 3-D16 4-D16 3-D16 3-D16 3-D16
1-D12
Beam T16 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D16 3-D16 1-D18
Beam T17 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D16 2-D16 1-D18
Column T18 350 300 3-D16a 2-D16a
Column T19 350 300 2-D25 2-D25
Column T20 350 300 3-D25 3-D25
Column T21 450 350 3-D16 3-D16
Column T22 450 350 3-D16 3-D16
Column T23 350 350 5-D14 5-D14
Column T24 550 550 4-D14 8-D14
Column T25 300 350 4-D14 2-D14
Column T26 300 350 4-D20 2-D20
Column T27 550 550 4-D14 8-D14
Column T28 400 450 5-D14 3-D14
Column T29 350 300 5-D14 2-D14
Column T30 300 300 3-D14 4-D14

a
For columns the rst number shows the number of rows of rebars in X or Y directions.

156
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Fig. 7. Structural geometry and type classication of beams and columns in the typical three-story RC frame with 3 spans in each horizontal direction.

123
x10 constraints which do not need nite element analysis for their calcu-
lation and simultaneously computes the penalty of the unsatised
10 constraints. In each step, the DM checks the possibility of obtaining
better tness. Whenever it perceives that the penalized objective
ECSS
function cannot obtain a better value compared to the global
8 CSS
optimum, Gb, found by the entire swarm until iteration t, it will ignore
Weight (ton)

Proposed Algorithm the rest of the calculation and will assign a value which is greater than
6 the local best, Pb, and the global best, Gb, to the tness. This assures
that the tness of this particle will not be considered in the calculation
of Pb or Gb. Therefore, a plethora of calculations for particles which do
4
not satisfy the code's requirements or are not likely to have a better
tness are avoided. Otherwise, the corresponding objective function
2 value is computed and a nite element analysis is performed for the
constraints check. As long as no violation is detected, no penalty will be
0 imposed on the objective function f (x).
0 1 2 3 4
x1053
Structural analyses 4. Case studies
Fig. 8. Convergence rate of DMPSO, CSS, and ECSS for seven-story RC structure.
To evaluate the performance of the proposed DMPSO algorithm for
optimization of large real frames, two benchmark case studies were
0 if Gi (x ) 0 examined. These are typical three story and seven-story RC framess
1 if 0 < Gi (x ) 0.01
c with 3 spans in each horizontal direction [19]. The three-story frame
2 if 0.01 < Gi (x ) 0.1 was optimized according to its construction cost. It should be noted that

10 if Gi (x ) > 0.1 (42) the present approach for calculating cost is simplistic and careful at-
tention should be taken to extrapolate it to real-world structures with
fPenalizedObj = fObj *( + 1) complex factors aecting cost. The objective function of the seven-story
(43)
frame was its construction weight. The frames were solved three times
where is the total penalty value, Ntotal denotes the total number of and among the optimum solutions obtained for each set, the best one
constraints, x represents the penalty function, Gk(x) is the typical was taken as the optimum design. Moreover, SAP2000 [27] is used for
constraint k and fPenalizedObj is the penalized objective function. the nite element analysis.
Considering the fact that much of the computing time belongs to the The structural characteristics of the frames include a service dead
structural calculations, the DMPSO algorithm rst checks the load of D = 5.9 kN/m2, uniform service live load of L = 2 kN/m2,

157
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Table 2
The results of optimum design form DMPSO and those from the literature.

Optimization results

CSS ECSS DMPSO

Sectional dimensions Reinforcements Sectional dimensions Reinforcements Sectional dimensions Reinforcements

Type Height of Width of T-rebar C-rebar Height of Width of T-rebar C-rebar Height of Width of T-rebar C-rebar
classication the the the the the the
optimal optimal optimal optimal optimal optimal
section section section section section section

T1 300 250 7-D10 4D10 300 250 6-D10 0-D10 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D14 3-D14 3-D14 1-D16 3-D14
T2 300 250 7-D10 0-D10 300 250 7-D10 0-D10 350 350 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
3-D14 3-D14
T3 300 250 7-D10 0-D10 300 250 7-D15 0-D15 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
4-D16 4-D16 1-D16
T4 300 250 7-D10 0-D10 300 250 8-D15 0-D15 350 350 3-D14 3-D14 3-D14 4-D14 4-D14 4-D14
2-D22 2-D22
T5 400 350 6-D15 0-D15 400 350 9-D15 0-D15 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D20 3-D20 1-D14 1-D16 1-D14
T6 500 350 9-D15 7-D15 500 350 6-D20 0-D20 350 350 3-D14 3-D14 3-D14 4-D14 4-D14 4-D14
2-D22 2-D22
T7 700 450 9-D25 6-D25 700 450 7-D20 4-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D22 2-D22 1-D16
T8 600 400 6-D20 0-D20 600 400 8-D20 5-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
1-D16 1-D16 1-D10 4-D16 1-D10
T9 600 400 7-D20 0-D20 600 400 7-D20 0-D20 300 300 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
3-D22 3-D22 1-D12
T10 600 400 8-D20 5-D20 600 400 7-D20 0-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D24 3-D24 3-D12 3-D18 3-D12
T11 650 400 7-D20 0-D20 650 400 9-D20 7-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D25 3-D25 2-D18 3-D18 2-D18
T12 700 450 9-D25 6-D25 700 450 9-D20 7-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
4-D22 4-D22 3-D16 3-D18 3-D16
T13 750 450 8-D25 0-D25 750 450 9-D20 7-D20 350 350 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
3-D25 3-D20 2-D12 3-D16 2-D12
T14 750 450 9-D25 0-D25 750 450 9-D25 7-D25 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D24 3-D24 2-D12 3-D18 2-D12
T15 300 250 7-D10 4D10 300 250 6-D10 0-D10 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D16 2-D16 1-D16
T16 300 250 7-D10 0-D10 300 250 7-D10 0-D10 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
-D16 3-D16 1-D16
T17 300 250 7-D10 0-D10 300 250 7-D15 0-D15 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
4-D16 4-D16 1-D12
T18 300 250 7-D10 0-D10 300 250 8-D15 0-D15 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D20 3-D20 1-D16 2-D16 1-D16
T19 400 350 6-D15 0-D15 400 350 9-D15 0-D15 400 300 3-D14 3-D14 3-D14 3-D14 3-D14 3-D14
3-D20 3-D20 2-D16 1-D12 2-D16
T20 500 350 9-D15 7-D15 500 350 6-D20 0-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
2-D25 2-D25 1-D16 1-D16 1-D16
T21 700 450 9-D25 6-D25 700 400 7-D20 4-D20 350 300 2-D16 2-D16 2-D16 2-D16 2-D16 2-D16
2-D22 -D22 3-D14 3-D14 3-D14
T22 600 400 8-D20 5-D20 600 400 6-D20 0-D20 600 300 2-D20 2-D20 2-D20 2-D20 2-D20 2-D20

T23 650 400 7-D20 0-D20 650 400 7-D20 0-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
4-D20 4-D20 2-D24
T24 650 400 7-D20 0-D20 650 400 8-D20 5-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D25 3-D25 3-D14 2-D24 3-D14
T25 650 400 9-D20 7-D20 650 400 7-D20 0-D20 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
5-D20 5-D20 2-D20 2-D24 2-D20
T26 650 400 9-D20 7-D20 650 400 9-D25 6-D25 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
5-D20 4-D20 2-D20 2-D24 2-D20
T27 650 400 9-D20 7-D20 650 400 8-D25 0-D25 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
5-D20 5-D20 2-D20 2-D24 2-D20
T28 750 450 9-D25 7-D25 750 450 9-D25 0-D25 300 300 3-D12 3-D12 3-D12 3-D12 3-D12 3-D12
3-D20 3-D20 2-D16 2-D20 2-D16

T29 200 200 7D-10a 7D-10 200 200 4D-15 4D-15 300 350 2D-14 4D-14
T30 250 250 4D-15 4D-15 250 250 4D-15 4D-15 300 350 2D-20 2D-20
T31 250 250 4D-15 4D-15 250 250 4D-15 4D-15 300 350 2D-25 2D-25
T32 250 250 4D-15 4D-15 250 250 4D-15 4D-15 300 350 3D-25 2D-25
T33 250 250 4D-15 4D-15 250 250 4D-15 4D-15 300 350 2D-25 2D-25
T34 400 400 5D-20 5D-20 400 400 7D-15 7D-15 350 400 3D-16 3D-16
T35 400 400 7D-20 7D-20 400 400 4D-20 4D-20 400 400 2D-16 6D-16
T36 450 450 10D-25 10D-25 450 450 8D-25 8D-25 300 300 4D-20 2D-20
(continued on next page)

158
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

Table 2 (continued)

Optimization results

CSS ECSS DMPSO

Sectional dimensions Reinforcements Sectional dimensions Reinforcements Sectional dimensions Reinforcements

Type Height of Width of T-rebar C-rebar Height of Width of T-rebar C-rebar Height of Width of T-rebar C-rebar
classication the the the the the the
optimal optimal optimal optimal optimal optimal
section section section section section section

T37 450 450 10D-25 10D-25 450 450 8D-25 8D-25 300 300 2D-20 2D-20
T38 500 500 7D-25 7D-25 500 500 8D-25 8D-25 400 300 2D-16 6D-16
T39 500 500 7D-25 7D-25 500 500 8D-25 8D-25 450 300 2D-16 6D-16
T40 500 500 7D-25 7D-25 500 500 8D-25 8D-25 450 350 2D-25 2D-25
T41 500 500 10D-25 10D-25 500 500 7D-30 7D-30 450 400 2D-25 2D-25
T42 600 600 7D-27 7D-27 600 600 7D-30 7D-30 450 400 2D-25 6D-25

a
For columns the rst number shows the number of rows of rebars in X or Y directions.

IColumn = 0.7 Ig (48)


1
where Ig is the gross moment of inertia of the section of the beam or
column.
Demand capacity rao(DCR)

0.8
4.1. Case study 1: a typical three-story RC frame with three spans in each
horizontal direction
0.6
The rst case study was a typical three-story RC frame with three
spans in each horizontal direction comprised of 119 elements, 70 beams
0.4
and 49 columns. These elements were divided into 17 beam groups and
13 column groups. Consequently, 371 design variables exist in the
0.2
problem, 306 for beams (18 for each beam design group) and 65 for
DCR for proposed algorithm
columns (ve for each column design group). The frame geometry and
DCR for ECSS algorithm grouping details is depicted in Fig. 4.
0 For this case study, DMPSO needed 5000 nite element structural
1 5 9 13 17 21 25 29 33 37 41 analyses for converging to an optimum cost of $64,800. Fig. 5 depicts
Number of beam and column type classicaons the convergence history for the DMPSO. It can be seen that DMPSO did
Fig. 9. Maximum DCR of members for each type classication. not trap to local values and continued to converge until it reached the
lowest value for the objective function. Fig. 6 illustrates the maximum
values of demand/capacity ratio under the critical loading case for
compressive strength of concrete fc = 25 MPa, yield strength of steel
member groups in the optimum solutions obtained by the DMPSO al-
reinforcements fy = 400 MPa, Cc = 105 $/m3, Cs = 0.9 $/kg,
gorithm. Table 1 demonstrates the results of optimum design from the
Cf = 92 $/m2, C = 2450 kg/m3, and s = 7850 kg/m3. To be more
DMPSO algorithm.
exact, because when steel is placed in the section, the same volume of con-
crete is removed, it is considered s = 7850 kg/m32450 kg/
4.2. Case study 2: a typical seven-story RC frame with 3 spans in each
m3 = 5400 kg/m3. Moreover, lateral loads, are calculated based on the
horizontal direction
ASCE7 code, considering the updated mass of the structure in every
iteration based on the dimensions of the columns and beams.
The second case study was a typical seven-story RC frame with three
Furthermore, six dierent factored load combinations are considered
spans in each horizontal direction reinforced concrete frame whose
per ACI 318 [22]. Eqs. (44)(46) show the load combinations.
geometry and grouping details are shown in Fig. 7. This frame had been
U1 = 1.2D + 1.0L (44) designed by the charged system search (CSS) and enhanced charged
system search (ECSS) [19]. This frame has a total of 180 members, 84
U2 = 1.2D + 1.0L 1.4E (45) beams and 96 columns, which are arranged into 42 groups; 28 groups
for beams and 14 groups for columns. This case study contains 574
U3 = 0.9D 1.4E (46) design variables, 504 of which are for beams and the remaining 70 are
for columns. Beams and columns are grouped to satisfy the uniformity
where D, L and E are the assumed dead, live and lateral loads, re- of members subjected to close design forces and have similar behaviors
spectively. according to their place in the frame and loading conditions. A popu-
The frames in both directions were moment resisting and all joints lation size of 150 is used to assure the best results for stochastic decline.
were rigid. The ooring system was considered to be a two-way slab. The minimum weight of the optimum design was 338 Ton after
Lateral forces aecting the frames were applied at the center of the 4000 structural analyses, which is lower than those achieved by other
mass of each story. Furthermore, in order to consider the eect of algorithms taken for comparison (CSS and ECCS). The convergence rate
cracking, the moment of inertia of the cross section for each member is of DMPSO was impressive. Fig. 8 shows that although DMPSO found
calculated according to ACI 318 code [22] using the equations shown in the best optimum solution after 4000 structural analyses, it achieved
Eqs. (47) and (48). optimum results faster in the rst 200 structural analyses. The best
design obtained by DMPSO together with the results achieved from the
IBeam = 0.35 Ig (47) mentioned studies is presented in Table 2. Because no considerable

159
M.J. Esfandiari et al. Advances in Engineering Software 115 (2018) 149160

limits were imposed to the design variables, the algorithm used the best References
section for the optimum solution. Fig. 9 provides a comparison between
the maximum values of demand/capacity ratio (DCR) under the critical [1] Lagaros ND. A general purpose real-world structural design optimization computing
loading case for member groups in the optimum solutions obtained by platform. Struct Multidiscip Optim 2014;49(6):104766.
[2] Kaveh A, Talatahari S. Optimal design of skeletal structures via the charged system
DMPSO and the ECCS algorithm used for comparison. search algorithm. Struct Multidiscip Optim 2010;41(6):893911.
The above gure shows that the best optimum designs are achieved [3] Saka MP, Dogan E, Aydogdu I. Review and analysis of swarm-intelligence based
by the DMPSO algorithm. Moreover, DMPSO's convergence rate is algorithms. In: Yang XS, Cui Z, Xiao R, Gandomi AH, Karamanoglu M, editors.
Swarm intelligence and bio-inspired computation. Elsevier; 2013.
signicantly better over the iterations when compared to the other [4] Kaveh A. Applications of metaheuristic optimization algorithms in civil engineering.
existing algorithms. The superiority of the DMPSO algorithm becomes Springer; 2017.
even more evident considering the fact that shear design of concrete [5] Vira C, Haimes YY. Multiobjective decision making: theory and methodology. 1983.
North Holl. Ser. Syst. Sci. Eng., no. 8.
members, development lengths and hook lengths of reinforcement steel [6] Miettinen K. Nonlinear multiobjective optimization vol. 12. Springer Science &
bars in the concrete members, and the other detailing issues were not Business Media; 1999.
taken into the account in the comparative algorithms. [7] Sawaragi Y, Nakayam H, Tanino T. Theory of multiobjective optimization. Elsevier;
1985.
This excellent performance can be explained by the feature of the
[8] Coello CC, Lamont GB, Van Veldhuizen DA. Evolutionary algorithms for solving
DMPSO algorithm that considers the preference of the decision making. multi-objective problems. Springer Science & Business Media; 2007.
As a consequence, the algorithm tries to escape from the local minimum [9] Justesen PD. Multi-objective optimization using evolutionary algorithms. Denmark:
so that it would not be easily trapped into a local value. Because the Univ. Aarhus, Dep. Comput. Sci.; 2009.
[10] Branke J, Deb K, Miettinen K, Slowinski R. Multiobjective optimization: interactive
DMPSO algorithm has almost no restrictions for selecting sections, it and evolutionary approaches vol. 5252. Springer Science & Business Media; 2008.
selected the most useful sections ensuring that the constraints were [11] Sindhya K, Miettinen K, Deb K. A hybrid framework for evolutionary multi-objec-
achieved near the threshold values. Especially in beams, where the tive optimization. IEEE Trans. Evol. Comput. 2013;17(4):495511.
[12] Michalewicz Z. Genetic algorithms+ data structures= evolution programs.
rebar was not continuous the entire length, the algorithm only utilized Springer Science & Business Media; 1996.
the adequate rebar needed. [13] Coello CAC, Lamont GB. Applications of multi-objective evolutionary algorithms
vol. 1. World Scientic; 2004.
[14] Hasanebi O, arba S, Doan E, Erdal F, Saka MP. Comparison of non-determi-
5. Conclusions nistic search techniques in the optimum design of real size steel frames. Comput.
Struct. Sep. 2010;88(17):103348.
[15] Hasanebi O, arba S, Doan E, Erdal F, Saka MP. Performance evaluation of
This article introduces optimization of real-world 3D reinforced metaheuristic search techniques in the optimum design of real size pin jointed
concrete frames using the multi-criterion Decision Making and Particle structures. Comput. Struct. Mar. 2009;87(5):284302.
[16] Akin A, Saka MP. Harmony search algorithm based optimum detailed design of
Swarm Optimization (DMPSO) algorithm. The main objective was to
reinforced concrete plane frames subject to ACI 318-05 provisions. Comput. Struct.
minimize the frames weight or construction cost of reinforced concrete Jan. 2015;147:7995.
frames while satisfying the limitations and specications of the ACI 318 [17] Fadaee MJ, Grierson DE. Design optimization of 3D reinforced concrete structures.
design code [22]. The limitations and specications were formulated as Struct. Optim. Oct. 1996;12(23):12734.
[18] Balling RJ, Yao X. Optimization of Reinforced Concrete Frames. J. Struct. Eng.
a series of constraints to the optimization problem and applied as pe- 1997;123(2):193202.
nalties on the tness function of the algorithm. Moreover, an ecient [19] Kaveh A, Behnam AF. Design optimization of reinforced concrete 3D structures
constraint handling technique, which demonstrated excellent perfor- considering frequency constraints via a charged system search. Sci. Iran.
2013;20(3):38796.
mance, was proposed. The algorithm was shown to lead to feasible [20] Esfandiary MJ, Sheikholaren S, Bondarabadi R, A H. A combination of particle
optimal designs, while also taking advantage of infeasible designs swarm optimization and multi-criterion decision-making for optimum design of
during the optimization procedure. Furthermore, the DMPSO algorithm reinforced concrete frames. Iran Univ. Sci. Technol. Jun. 2016;6(2):24568.
[21] Kaveh A, Sabzi O. A comparative study of two meta-heuristic algorithms for op-
not only considers the design code requirements but also construc- timum design of reinforced concrete frames. Int. J. Civ. Eng. 2011;9(3):193206.
tional, architectural, and reinforcement detailing constraints. As a re- [22] A. A. C. I. Standard, Building code requirements for structural concrete (ACI 318-
sult, the optimum design obtained by the DMPSO algorithm is directly 11), 2011.
[23] Minimum design loads for buildings and other structures (ASCE). [Online].
ready for practical application without the need of any further mod- Available: http://ascelibrary.org/doi/book/10.1061/9780784412916. [accessed
ication or processing. The numerical results demonstrated the sound 22.04.15].
performance of the DMPSO algorithm for nding optimum cost (or [24] Plevris V, Papadrakakis M. A hybrid particle swarm-gradient algorithm for global
structural optimization. Comput.-Aided Civ. Infrastruct. Eng. 2011;26(1):4868.
weight) of RC structures in terms of eciency and the convergence
[25] Koziel S, Michalewicz Z. Evolutionary algorithms, homomorphous mappings, and
rate. Based on the ndings of this research, we concluded that the constrained parameter optimization. Evol. Comput. 1999;7(1):1944.
DMPSO algorithm is suitable to model practical design problems that [26] Coello Coello CA. Theoretical and numerical constraint-handling techniques used
consider the variations in cross-sectional dimensions of concrete with evolutionary algorithms: a survey of the state of the art. Comput. Methods
Appl. Mech. Eng. Jan. 2002;191(11):124587.
frames, the variation in detailing and the variations in placement of [27] CSI (Computers and Structures Inc.). SAP2000 v10 analysis reference manual.
reinforcement bars. Berkeley: CSI; 2004.

160

Vous aimerez peut-être aussi