Flexural Design of Reinforced Concrete Frames

by Genetic Algorithm
C. Lee, P.E., M.ASCE,1 and J. Ahn2

Abstract: A genetic algorithm is used to perform the discrete optimization of reinforced concrete plane frames subject to combinations
of gravity loads and lateral loads. Difficulties in finding optimum sections from a semi-infinite set of member sizes and reinforcement
arrangements are alleviated by constructing data sets, which contain a finite number of sectional properties of beams and columns in a
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

practical range. Construction practice is also implemented by linking columns and beams by group and by considering ‘‘connectivity’’
between columns located in the same column line. It is shown that the developed genetic algorithm obtained an optimal design for
reinforced concrete plane frames.
DOI: 10.1061/共ASCE兲0733-9445共2003兲129:6共762兲
CE Database subject headings: Algorithms; Building frames; Concrete, reinforced; Design; Concrete structures; Optimization.

Introduction tion identification numbers and the resistant capacities of sections,

Unlike steel structures, reinforced concrete buildings can be which is established by regression. The regression line correlates
framed with a semi-infinite set of member sizes and amount of the section number to a discrete sectional property of a beam or
reinforcement. The main difference comes from the fact that more column. The sections in the database are selected from widely
combinatorial characteristics exist in determining the sectional used ones in practical design.
dimensions and the number of reinforcing bars for reinforced con- Balling and Yao 共1997兲 optimized three-dimensional rein-
crete members than steel members, which are usually prefabri- forced concrete frames using simulated annealing with a multi-
cated with a finite number of sections. In addition to the discrete level method. Discrete variables as well as limits on the number
and combinatorial nature of the sectional dimensions and the of reinforcing bars and their topological arrangements are consid-
number of reinforcing bars, topological reinforcement details ered in their study.
specified in the design code make optimization of reinforced con- Recently, the discrete optimization of structures has been per-
crete structures even more complicated. Optimization techniques formed using genetic algorithms 共GAs兲 共Jenkins 1991, 1992; Ra-
for the element level of reinforced concrete structures have been jeev and Krishnamoorthy 1992, 1998; Koumousis and Arsenis
presented by different researchers 共Adamu and Karihaloo 1994; 1994; Cheng and Li 1997兲. General algorithms are search proce-
Chung and Sun 1994; Zenon et al. 1995兲. These methods were dures based on the mechanics of natural genetics and natural se-
based on sequential linear programming, continuum-type optimal- lection. They combine the concept of artificial survival with ge-
ity criteria, and nonlinear programming such as Powell’s algo- netic operators, abstracted from nature, to form a robust search
rithm. Fadaee and Grier 共1996兲 suggested applying the optimality mechanism 共Goldberg 1989兲. The main advantages of a GA over
criteria method to optimize three-dimensional reinforced concrete conventional optimization techniques can be summarized as: 共1兲
structures using continuous design variables. Practical application GAs do not require gradient computations; 共2兲 GAs do not require
of these solutions, however, requires additional modifications to that the constraints be expressed explicitly in terms of design
fit the discrete nature of structural design variables. variables; 共3兲 GAs take advantage of carrying out optimization
More practical discrete optimization techniques have been processes in a stochastic framework; and 共4兲 GAs are not limited
suggested by Choi and Kwak 共1990兲 and Balling Yao 共1997兲. by restrictive assumptions about search space, such as continuity
Choi and Kwak 共1990兲 developed a simplified algorithm for dis- or the existence of derivatives.
crete optimum design of reinforced concrete frames by the con- For reinforced concrete structures, several researchers have in-
struction of a database. In their study, discrete optimum sections vestigated using GAs for discrete optimization. Koumousis and
are directly searched based on the relationship between the sec- Arsenis 共1994兲 performed the detailed design of a continuous
reinforced concrete beam on the basis of a multicriterion objec-
1 tive that represented a compromise between minimum weight de-
Professor, Dept. of Architectural Engineering, Chung-Ang Univ.,
Ansung 456-756, Korea. E-mail: [email protected] sign, maximum uniformity, and minimum number of bars for a
2
Graduate School Student, Dept. of Architectural Engineering, group of members. Rajeev and Krishnamoorthy 共1998兲 consid-
Chung-Ang Univ., Ansung 456-756, Korea. ered the detailing of reinforcement as a design variable in addi-
Note. Associate Editor: Shahram Pezeshk. Discussion open until No- tion to cross-section dimensions. Sets of allowable combinations
vember 1, 2003. Separate discussions must be submitted for individual
of reinforcement bars for columns and beams were tabulated, and
papers. To extend the closing date by one month, a written request must
be filed with the ASCE Managing Editor. The manuscript for this paper
character strings were decoded and mapped into the values for the
was submitted for review and possible publication on August 24, 2001; set to obtain appropriate reinforcement bar combinations. Width
approved on August 23, 2002. This paper is part of the Journal of Struc- and depth of columns and beams were represented by characters,
tural Engineering, Vol. 129, No. 6, June 1, 2003. ©ASCE, ISSN 0733- and they were also mapped through character strings. Their model
9445/2003/6-762–774/$18.00. showed that optimum designs of the three-bay, four-story frame

762 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

 and two-bay, six-story frame resulted in cost function value less
than those obtained from SUMT based on a direct search method.
Although many researchers have performed discrete optimum
design using GAs, a GA developed for reinforced concrete frames
is limited. It is, therefore, the main objective of this research to
develop a GA that performs the discrete optimum design of rein-
forced concrete frame structures. The present paper describes the
GA-based approach taken to optimize plane reinforced concrete
frames subject to multiple loading conditions. For the process of
GAs natural selection, databases of beams and columns are con- Fig. 1. Restriction on number of reinforcing bars in beam section:
structed. Each member, which satisfies the specifications provided 共a兲 at least four bars in section; 共b兲 layout of maximum number of
in the structural design code 共ACI 1999a兲 in each database is bars; and 共c兲 allowable arrangement of bars in two layers
assigned a unique identification number.
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

Each section in the database is generated by following code

Review on Genetic Algorithm provisions regarding structural and constructional constraints and
topological arrangements for reinforcement. Depending on the
General algorithms use three basic randomized operators in place given problem, the user may adjust the size of the database. De-
of the usual deterministic ones: reproduction, crossover, and mu- tailed descriptions concerning the construction of the databases
tation. The method of implementing each operator is briefly re- for beams and columns are given in the following sections.
viewed, including how it is applied to the optimization of the
reinforced concrete frames in this study.
1. Reproduction: In reproduction, individuals with higher fit- Beams
ness values have a higher probability of being selected for Predefined beams are generated and entered in the database on the
mating in subsequent genetic action. Proportional selection basis of two conditions: 共1兲 limitations on beam width and depth
共Holland 1975兲 is adopted in this study. Selected pairs in the and 共2兲 provisions in code regarding reinforcements.
mating pool will be matched before crossover is performed;
2. Crossover: Crossover sites are selected at random along the Beam Width and Depth
string length. Position values are swapped between the ran- Minimum beam width (b min), maximum beam width (b max), mini-
domly selected two strings at the cross sites. The probability mum ratio of beam depth to beam width (␤ min), and maximum
of crossover (p c ) is defined as the ratio of the number of ratio of beam depth to beam width (␤ max) are defined prior to the
offspring produced in each generation to the population size generation of the database. Discrete increments for beam width
(n p ). If a random number r i generated in 关0,1兴 for the ith (⌬b) and beam depth (⌬h) are also given. Cross-sectional di-
chromosome is less than p c , then the ith chromosome is mensions for beams in the database are then automatically gener-
chosen for mating and put into the mating pool. Otherwise, ated with these values. The following procedure is adopted to
the ith chromosome succeeds to the child generation; and generate sections in the beam database:
3. Mutation: Mutation rate ( p m ) is defined as the ratio of the
number of mutated genes to the total number of genes in the Generate b i ⫽b min⫹ 共 i⫺1 兲 •⌬b until b i becomes b max
population. Given a mutation rate of p m , the number of Calculate h min,i ⫽␤ min⫻b i and h max i ⫽␤ max⫻b i
mutations (n m ) becomes
Let h 1,i ⫽larger of 共 h min ,h min,i 兲 (2)
n m ⫽round 共 p m ⫻N alle兲 (1)
Generate h j,i ⫽h 1,i ⫹ 共 j⫺1 兲 •⌬h until h j,i is equal to
Each mutation is performed by swapping the randomly se-
lected position value 共or value of alleles兲 from ‘‘0’’ to ‘‘1’’ or vice smaller of 共 h max ,h max,i 兲
versa of a randomly selected chromosome in the population. As
the number of generations increased, the p m was allowed to de- where h min and h max⫽minimum and maximum
crease.
beam depth, respectively

Construction of Database Beam Reinforcements

For a given beam section, the number of reinforcing bars is de-
A semi-infinite set of member sizes and reinforcement arrange- termined by the following criteria. For all cases, at least two
ments for beams and columns is reduced to a countable number of reinforcing bars are placed either at the top or bottom of a section.
arrangement by constructing data sets in a practical range. Each Reinforcement bars need to be carefully placed in order to satisfy
chromosome can be regarded as a replica of a particular frame code specified structural constraints and topological arrange-
composed of groups of beams and columns. Two databases are ments. For structural constraints, minimum and maximum rein-
constructed for beams and columns. The database for beams is a forcement areas are considered. For topological arrangements,
collection of sectional properties of candidate beams. Information horizontal and vertical spacings between longitudinal bars and
on sectional dimensions, number of top and bottom reinforcing cover thickness are taken into account. The minimum number of
bars, and corresponding moment capacities can be obtained by bars (n s,min) and maximum number of bars (n s,max) are deter-
referring to the identification numbers in the database. The data- mined by:
base for columns consists of the sectional dimensions of columns, 1. Minimum steel area: larger of 关minimum steel area (A s,min),
number of reinforcing bars, and characteristic points representing 2 reinforcing bars⫻reinforcing area per bar (a s )] 关see Fig.
simplified P-M interaction curves. 1共a兲兴; and

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 763

J. Struct. Eng. 2003.129:762-774.

 Table 1. Database for Beams 共Three-Bay, Three-Story Reinforced Concrete Frame兲
Number of Bars (D22a兲 Factored Moment Resistance 共kN m兲
Beam Width Depth Area Moment of Inertia
number 共mm兲 共mm兲 (⫻103 mm2兲 (⫻107 mm4兲 Center End Center End
0 300 450 135 227.8 2 2 99.0 99.0
1 300 450 135 227.8 2 3 99.0 143.1
2 300 450 135 227.8 2 4 99.0 184.2
— — — — — — — — —
1,021 450 900 405 2,733.8 4 8 439.0 840.8
1,022 450 900 405 2,733.8 4 10 439.0 1,029.0
1,023 450 900 405 2,733.8 4 12 439.0 1,205.4
a
D22 indicates deformed bar of diameter 22 mm.
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

2. Maximum steel area: smaller of 关maximum steel area umn depth (⌬h c ) are given in advance in order to generate can-
(A s,max), maximum number of reinforcing bars in two layers didates for the column sections. For a particular generated column
(n s )⫻reinforcing area per bar (a s )] 关see Fig. 1共b兲兴. section, a number of subsections are generated with possible re-
From the following equations, n s,min and n s,max are determined inforcement arrangements based on code provisions on the mini-

再 冉 冊冎
mum and maximum area of reinforcement:
A s,min
n s,min⫽max round ⫹0.5 ,2 (3a) 1. Minimum steel area: larger of 关minimum steel area (A sc,min),
as four reinforcing bars⫻one reinforcing bar area (a s )]; and

再 冉
n s,max⫽min round
A s,max
as 冊 冎
⫺0.5 ,n s (3b)
2. Maximum steel area: smaller of 关maximum steel area
(A sc,max), maximum number of reinforcing bars⫻one rein-
forcing bar area (n c max⫻as)] 共see Fig. 2兲.
where Using the following equations, minimum and maximum number
of bars (n c,min ,nc,max) can be determined
0.25冑 f ⬘c

再 冉 冊冎
1.4
A s,min⫽ b•d⭓ b•d 关 mm2 兴 A sc,min
fy fy n c,min⫽max round ⫹0.5 ,4 (4a)
as
f ⬘c
再 冉 冊 冎
600
A s,max⫽0.75共 0.85␤ 1 兲 • •b•d 关 mm2 兴 A sc,max
f y 600⫹ f y n c,max⫽min round ⫺0.5 ,n c (4b)
as
n s ⫽2⫻round 再 b⫺2⫻ 共 t c ⫹d s 兲 ⫹s b
d b ⫹s b
⫺0.5 冎 where

By placing reinforcing bars incrementally from n s,min to n s,max

for both the top and bottom of a section, subsections are gener-
n c ⫽2⫻round 再 w⫺2⫻ 共 t c ⫹d t 兲 ⫹s c
d b ⫹s c
⫺0.5 冎
ated for a fixed beam dimension. In cases when two layers of If n c,min is an odd number, add one additional bar to n c,min .
reinforcements are needed, all reinforcing bars in the upper layer Likewise, if n c,max is an odd number, reduce one bar from n c,max .
are symmetrically positioned with respect to the vertical axis of a Even numbers of reinforcing bars are placed incrementally from
section and are placed directly above the reinforcing bars in the n c,min to n c,max for a given column section. As sectional dimen-
lower layer 关see Fig. 1共c兲兴. sions and possible numbers of reinforcing bars are determined,
Unique identification numbers are assigned to all candidate characteristic points on the P-M interaction curve for each candi-
beam sections selected according to the above criteria. Factored date column are evaluated and stored in the database. The P-M
moment capacities are calculated by ␾•M n ⫽␾•A s • f y (d⫺a/2) interaction curve can be approximately constructed by linearly
and saved in the database for the beam sections, where ␾ is the connecting characteristic points.
strength reduction factor (␾⫽0.9), A s is the area of tensile rein- In Fig. 3, axial forces of P 0 and P 1 ⫽0.8P 0 are obtained ac-
forcement, f y is the yield strength of tensile reinforcement, d is cording to ACI 318-99 code specification as P 0 ⫽␾ 关 0.85f c⬘ (A g
the effective depth of a beam, a⫽A s • f y /(0.85f ⬘c b), and b is the ⫺A st )⫹ f y •A s 兴 . The point (M 2 , P 2 ) is the intersection point of
beam width. Table 1 shows the database for beams generated for the line P⫽ P 1 and the line connecting (M 0 , P 0 ) and (M 3 , P 3 ).
the three-bay, three-story reinforced concrete frame to be opti-
mized in the section ‘‘Three-Bay, Three-Story Reinforced Con-
crete Frame.’’ The table contains information on width, depth,
number of reinforcing bars for positive and negative moments,
and corresponding moment resisting capacities for candidate
beams.

Columns
Candidates for column sections are similarly generated as those of
the beam candidates in Eq. 共2兲. Minimum column width (w min),
Fig. 2. Restriction on number of reinforcing bars in column section:
maximum column width (w max), minimum ratio of column depth
共a兲 at least four bars in section; and 共b兲 maximum number of bars
to column width (␻ min), maximum ratio of column depth to col-
satisfying required spacing between bars and cover thickness
umn width (␻ max), increments for column width (⌬w), and col-

764 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Linearized P-M interaction curve by connecting characteristic points

The point (M 3 , P 3 ) corresponds to the balanced condition of a The objective and constraint functions can be stated as
column. The axial force P 4 is defined as the smaller of 0.1f ⬘c
•A g and P 3 , and consequently the moment capacity M 4 is Minimize: C⫽C b ⫹C f ⫹C c (5)
obtained from reading the intersection of the line P⫽ P 4 and the Subject to
line connecting (M 3 , P 3 ) and (M 6 , P 6 ). In finding
M 5 关 ⫽(0.9/0.7)M 6 for rectangular tied columns兴 or M 6 , pure M⫹ ⫹
u,i jk ⭐␾M n,i
moment resistance of a section without axial forces is considered.
兩M ⫺ ⫺
ul,i jk 兩 ⭐ 兩 ␾M n,i 兩
The axial resistance under pure tension is taken to be P 7 ⫽ f y (6)
•A s . 兩M ⫺ ⫺
ur,i jk 兩 ⭐ 兩 ␾M n,i 兩 for beams
Table 2 shows the database for columns in three-bay, three-
story reinforced concrete frame considered in the section, ‘‘Three- 共 P u ,M u 兲 ⭐ 共 ␾ P n ,␾M n 兲 for columns
Bay, Three-Story Reinforced Concrete Frame.’’
where
NGB NM BG 关 i 兴

Formulation of Optimization C b⫽ 兺 j⫽1

兺
i⫽1
⫹
兵 b i •h i •L Bi j • 共 m c ⫹w c 兲 ⫹A s,i •L ⫹
f •L i j •␥ s
B

⫺
Objective and Constraint Functions • 共 m s ⫹w s 兲 ⫹2⫻A s,i •L ⫺
f •L i j •␥ s • 共 m s ⫹w s 兲
B

Material costs and labor costs must be included in the objective ⫹

共 L Bi j ⫺L Bi j •L ⫹ ⫺
⫹A se,i f 兲 •␥ s • 共 m s ⫹w s 兲 ⫹A se,i
function for the economic design of a reinforced concrete frame.
Three constraints on a beam member are considered with respect • 共 L Bi j ⫺2•L Bi j •L ⫺
f 兲 •␥ s • 共 m s ⫹w s 兲 其
to negative moments at both ends and positive moments at mid-
span under different loading cases. NGB NM BG 关 i 兴

Due considerations are given for the column so that the com- Cf⫽ 兺 j⫽1
兺
i⫽1
兵 共 b i ⫹2h i 兲 •L Bi j • 共 m f ⫹w f 兲 其
bination of axial and flexural load exists inside the P-M envelope
of the column. In the following, the subscripts i, j, and k stand for NGC NM CG 关 i 兴
group number, member number in group, and load case, respec-
tively.
⫹ 兺 j⫽1
兺
i⫽1
w i •h ci •L ci j • 共 m f ⫹w f 兲

Table 2. Database for Columns 共Three-Bay, Three-Story Reinforced Concrete Frame兲

Number of
Column Width bars P0 P1 P3 P4 P7 M2 M3 M4 M5 M6
number 共mm兲 (D25a兲 共kN兲 共kN兲 共kN兲 共kN兲 共kN兲 共kN m兲 共kN m兲 共kN m兲 共kN m兲 共kN m兲
0 300 4 1,783.6 5,576.2 428.3 302.8 794.8 21.6 83.3 75.5 69.6 57.8
1 300 6 2,048.2 5,782 394 302.8 1,225 25.5 101.9 92.1 97.0 57.8
] ] ] ] ] ] ] ] ] ] ] ] ]
62 850 20 12,740 10,192 4,811.8 2,430.4 3,978.8 658.6 2,048.2 1,734.6 1,303.4 1,421
63 850 22 13,034 10,388 4,811.8 2,430.4 4,370.8 682.1 2,146.2 1,793.4 1,430.8 1,421
a
D25 indicates deformed bar of diameter 25 mm.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 765

J. Struct. Eng. 2003.129:762-774.

Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Flow chart of developed general algorithm

NGC NM CG 关 i 兴
兩M ⫹
u,i jk 兩
C c⫽ 兺 j⫽1
兺
i⫽1
兵 w i •h ci •L Ci j 共 m c ⫹w c 兲 ⫹A s,i •L Ci j • 共 m s ⫹w s 兲 其 具g⫹
M ,i jk 典 ⫽
␾M ⫹
n, j
⫺1⭓0

⫺
兩 M ul,i jk 兩
Transformation to Unconstrained Optimization 具g⫺
M 1,i jk 典 ⫽ ⫺ ⫺1⭓0 (7)
兩 ␾M n,i 兩
For the application of a GA to the optimization of a reinforced
concrete frame, the above constrained optimization problem is 兩M ⫺
ur,i jk 兩
transformed to an unconstrained optimization one. Each internal 具g⫺
M r,i jk 典 ⫽ ⫺ ⫺1⭓0
兩 ␾M n,i 兩
moment is normalized with respect to corresponding member
strength. Beams are usually subject to flexural moments and shear The safety of a column is evaluated by taking the ratio of
forces. Flexural strengths are considered only in this study. All distances from the origin to the loading point in the P-M interac-
beams are assumed to be uniformly loaded. Lateral loads due to tion plane 共Fig. 3兲. As a loading point 共say, Point B in Fig. 3兲 gets
earthquake loads are applied as joint loads. farther from Point A which is an intersecting point of the line
The previously defined constraints are normalized and then from the origin to Point B with an angle ␪ on P-M interaction
used later for constituting the unconstrained objective function curve the level of violation is increased.

766 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. Representative chromosome and generation of children by

crossover: 共a兲 groups of genes in chromosome; 共b兲 cross sites in
parents’ chromosome; and 共c兲 generated children after crossover

w T,l
具 g cw,l 典 ⫽ w ⫺1⭓0
B,l
(10)
h cT,l
具 g ch,l 典 ⫽ h ⫺1⭓0
Fig. 5. Reinforced concrete frame made of one beam group and two cB,l
column groups: 共a兲 three-bay, three-story reinforced concrete frame; Similarly, if n T and n B represent the number of reinforcing
共b兲 genes of first beam and first column in corresponding groups; and bars of columns in Group T and Group B, respectively, then con-
共c兲 beam and column sections in each database nectivity for reinforcing bars in columns is evaluated by
n T,l
具 g s,l 典 ⫽ n ⫺1⭓0 (11)
B,l
Let L m,i jk and L u,i jk be the distances between origin O and The final form of unconstrained objective function can be ex-
Point A and the distance between the origin and point B on P-M pressed as follows:
interaction envelope, respectively
Minimize:
L m,i jk ⫽ 冑共 ␾ P n,i jk 兲 2 ⫹ 共 ␾M n,i jk 兲 2 G C
(8)
F⫽W g • ⫹W c •
M • 共 W M ⫹W PM⫹W CON兲 C initial (12)
L u,i jk ⫽ 冑共 P u,i jk 兲 2 ⫹ 共 M u,i jk 兲 2
where
Given M u,i jk and P u,i jk , the line OB can be expressed by
using ␪⫽tan⫺1(Pu,ijk /M u,ijk). Knowing ␪ and the given point
B(M u,i jk , P u,i jk ), the point A(␾M n,i jk ,␾ P n,i jk ) on the P-M en-
velope can be obtained by finding the intersection point of the
P-M envelope and the straight line OB. The penalty function for
column strength is then expressed in normalized form as

L u,i jk
具 g PM ,i jk 典 ⫽ L ⫺1⭓0 (9)
m,i jk

In addition to strength requirements, column sizes and the

number of reinforcing bars need to be properly selected for those
columns located in the same column line so that columns below
would have larger or equal column sizes, and number of reinforc-
ing bars, than those in columns above. This is termed ‘‘connec-
tivity’’ in this study. Let T and B represent two different column
group numbers in the lth connectivity condition. If all the col-
umns in Group B with width w B 共depth h cB ) are located lower
Fig. 7. Convergence trend for three-bay, three-story reinforced con-
than those columns in Group T with w T 共depth h cT ) connectivity
crete frame
for these column sizes is evaluated by

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 767

J. Struct. Eng. 2003.129:762-774.

Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. Searching process for beam groups and column groups as generations proceed: 共a兲 distribution of initial population; 共b兲 distribution of
population at iteration 15; and 共c兲 distribution of population at convergence

NDLC NGB NMBG关 i 兴 process 共Gen and Cheng 1997兲. In order to minimize genetic drift,
G⫽ 兺 兺 j⫽1
k⫽1 i⫽1
兺 共 WM•具g⫹ ⫺
M ,i jk 典 ⫹W M • 具 g M ,i jk 典 兲 a preselection concept 共Cavicchio 1972兲 is introduced to the pre-
vious three basic GA operations: reproduction, crossover, and mu-
NDLC NGC NMCG关 i 兴 tation. According to the preselection concept, an offspring re-
⫹ 兺 i⫽1
兺 j⫽1
k⫽1
兺 共 W PM• 具 g PM,i jk 典 兲 places its parent if the offspring’s fitness exceeds that of an
inferior parent. This procedure is illustrated in Fig. 4. Given this
condition, new offspring always have genetic traits superior to or
NCON
at least similar to their parents. Comparing fitness values of F
⫹ 兺
l⫽1
共 W CON•NC l 共 具 g cw,l 典 ⫹ 具 g ch,l 典 ⫹ 具 g s,l 典 兲兲 between parents and their children, superior ones among these are
succeeded to the next generation. The GA adopting the preselec-
Constant weights of 100 are assigned to all penalty parameters tion concept for optimizing reinforced concrete frame is briefly
except that the value of W g is allowed to increase at each genera- summarized:
tion by a factor greater than 1.0. 1. Determine string length S L by

Application of Genetic Algorithm SL⫽NGB⫻N B,alle⫹NGC⫻N C,alle (13)

In general, the fitness function may be multimodal, and thus many Generate n p chromosomes having string length S L , and ran-
peaks exist. This might cause a GA to drift during the optimizing domly assign ‘‘0’’ or ‘‘1’’ to each allele in each chromosome;

768 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

 Fig. 9. Subspace contour map of F values near local minimum in
prescribed search space for three-bay, three-story frame
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

2. Convert binary numbers encoded in each chromosome into

integer numbers and identify appropriate beam group num-
bers and column group numbers by mapping these numbers
to a member identification number in the beam or column Fig. 10. Three-bay, nine-story reinforced concrete frame subject to
databases. Generate n p frames from the n p chromosomes; gravity and lateral loads (span⫽9,000 mm, story height
3. Evaluate the objective function value F for each frame by ⫽3,600 mm)
Eq. 共14兲. Let F i be the value of F for frame i. Evaluate the
fitness value for ith frame ( f i ) by the following rule:
f i⫽共Fmax⫹Fmin兲⫺Fi , i⫽1,2, . . . ,n p (14) as joint loads. Uniform gravity loads are assumed for a dead load
4. Using f i values obtained in Step 3, operate the reproduction 共D兲 and a live load 共L兲. Different loading cases are considered as
by the method presented in the section, ‘‘Review on Genetic suggested in ACI 318-99 code 共1999a兲 for strength design.
Algorithm.’’ U⫽1.4D⫹1.7L
5. Perform the crossover as described in the above-mentioned
section. For a randomly chosen pair of chromosomes in the U⫽0.75共 1.4D⫹1.7L⫾1.87E 兲 (15)
mating pool, four positions 共two in strings for beam group
and two for column group兲 are selected uniformly at random. U⫽0.9D⫾1.43E
A pair of children are generated by exchanging mapping Uniform service dead load of 16.5 kN/m and uniform service live
strings between parent chromosomes; load of 7.2 kN/m are assumed in all examples. Live load reduc-
6. Apply preselection concept; and tion based on the alternate equation in the Uniform Building Code
7. Check the convergence criterion and stop if it is met. Other- 共1997兲 is applied. Earthquake loads are evaluated according to the
wise mutate randomly and proceed to the next generation. static force procedure given in the Uniform Building Code
Repeat from Step 2 until the convergence is obtained. The 共1997兲. Columns in each story are checked for their slenderness
convergence is assumed to be attained when the percentage and their end moments are magnified if necessary according to
of the population ( P c v ) possessing the best fitness at a cer- the ACI code and notes 共ACI 318-99 code 1999a; Notes on ACI
tain generation reaches 20% at that generation. 318-99 1999b兲. In order to find the effect of pattern loading, the
Fig. 4 summarizes the above procedure in a flow chart. subframe analyses on each story are performed and positive and
negative moments are increased as needed.
The assumed concrete strength and yield strength of reinforc-
Illustrative Examples ing bars in these examples are f ⬘c ⫽23.5 MPa and f y ⫽392 MPa,
The performance of the developed algorithm is investigated for respectively, for all frames. Rebar sizes are fixed for beams and
reinforced concrete plane frames having a different number of columns throughout optimization process. The costs of concrete,
stories. Lateral equivalent static earthquake loads 共E兲 are applied forming, and reinforcing steels are given as 54, 50.5, and 5.5

Table 3. Results of Optimum Design for Three-Bay, Three-Story Reinforced Concrete Frame
OPTIMIZATION RESULTS
Sectional Dimensions Reinforcements
Member Group Story Type of Section Width Depth Positive Negative
type number level member number 共mm兲 共mm兲 moment moment
Beam 1 1–3 Beam 40 300 550 3-D22a 5-D22
Column 1 1–3 Exterior column 1 300 300 6-D25
2 1–3 Interior column 0 300 300 4-D25
Initial design Maximum F cost 1,644.2 18,233 US dollars
Minimum F cost 54.7 12,892 US dollars
Optimum design F cost 44.9 10,566 US dollars
a
D22 indicates deformed bar of diameter 22 mm.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 769

J. Struct. Eng. 2003.129:762-774.

 Table 4. Results of Optimum Design of Three-Bay, Nine-Story Reinforced Concrete Frame with Different Parameter Values
PENALTY WEIGHT
Convergence
Mutation Wg
Population criterion Crossover
Case size percentage probability Rate Initialb Final Number of
Parameters number np P c v (%) pc pm Factora (⫻102 ) (⫻102 ) F State iteration
1 50 20 0.8 N.A.c N.A. 100 100 48.0 d 67
np f
2 500 20 0.8 N.A. N.A. 100 100 43.4 f 87
3 1,000 20 0.8 N.A. N.A. 100 100 45.8 inf d 107
4 2,000 20 0.8 N.A. N.A. 100 100 44.1 f 163
P cv 2 500 50 0.8 N.A. N.A. 100 100 43.4 f 88
6 500 80 0.8 N.A. N.A. 100 100 43.4 f 89
pc 7 500 20 0.6 N.A. N.A. 100 100 47.8 inf 200
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

8 500 20 1.0 N.A. N.A. 100 100 43.2 inf 87

pm 9 500 20 0.8 0.01 0.98 100 100 45.3 f 186
10 500 20 0.8 0.05 0.98 100 100 41.7 f 316
11 500 20 0.8 0.01 1.0 100 100 N.C.e N.A. ⬎1,000
12 500 20 0.8 0.05 1.0 100 100 N.C. N.A. ⬎1,000
Wg 13 500 20 0.8 N.A. N.A. 1.0 100 40.6 inf 134
14 500 20 0.8 N.A. N.A. 100 Var.f 45.0 f 119
p m ,W g 15 500 20 0.8 0.01 0.98 1.0 Var. 40.8 f 227
a
Factor⫽factor multiplied to the p m in previous generation to reduce its value in current generation.
b
Initial⫽initially assigned penalty value.
c
N.A.⫽not applicable.
d
f, inf⫽feasible and infeasible solution, respectively.
e
N.C.⫽not converged.
f
Var.⫽initial value is increased by a factor of 1.05 at each iteration.

⫻10⫺1 US dollars/kg, respectively 共Information on Consumers’ For the beam database, b min , b max , ⌬b, h min , h max , ⌬h, ␤ min ,
Prices 2002兲. All examples were run on a 1.8 GHz Pentium per- and ␤ max are given as 300, 450, 50, 450, 900, 50 mm, 1.5, and
sonal computer. 2.5, respectively. Following the rule presented in the section,
‘‘Construction of Database,’’ 1024 beam candidates are generated
Three-Bay, Three-Story Reinforced Concrete Frame and stored in the beam database. For the column database, only
square columns are considered. w min , w max , ⌬w, h c min , h c max ,
The frame shown in Fig. 5 has nine beams and 12 columns. ⌬h c , ␻ min , and ␻ max are given to 300, 850, 50, 300, 850, 50 mm,
Beams and columns are designed in groups: one group for all 1.0 and 1.0, respectively, and 64 candidate columns are generated.
beams and two groups for columns.

Table 5. Results of Optimum Design for Three-Bay, Nine-Story Reinforced Concrete Frame
OPTIMIZATION RESULTS
Sectional
Dimensions Reinforcements
Member Group Story Type of Section Width Depth Positive Negative
type number level member number 共mm兲 共mm兲 moment moment
Beam 1 1–3 Beam 70 300 500 3-D22a 6-D22
2 4–6 Beam 70 300 500 3-D22 5-D22
3 7–9 Beam 75 300 500 4-D22 6-D22

Column 1 1– 4 Exterior column 6 450 450 6-D25

2 1– 4 Interior column 10 650 650 10-D25
3 5–9 Exterior column 2 400 400 6-D25
4 5–9 Interior column 5 500 500 4-D25

Initial design Maximum F 202.7

cost 53,750 US dollars
Minimum F 54.9
cost 44,082 US dollars

Optimum design F 40.8

cost 32,936 US dollars
a
D22 indicates deformed bar of diameter 22 mm.

770 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

 A chromosome representing this frame is assumed to contain
two different strings—strings for beam and column members.
Strings for each member are further divided by substrings, each of
which represents a particular beam 共or column兲 in the group 共see
Fig. 6兲. The chromosome in Fig. 6 has one substring with a length
of ten for one beam group and two substrings having a length of
six for two column groups. This is so because the number of
beam candidates in the database is 2 10⫽1,024, while that in the
column is 2 6 ⫽64. The lengths of substrings, however, may be
variable depending on the assigned number of sectional candi-
dates.
Each position 共or allele兲 in each substring takes on a value of
1 or 0 and, thus, a substring is in the form of a binary number.
This binary number represents a member identification number in
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

the database. Figs. 5 and 6 show that the string representing the
first beam group is 关0010110111兴. This binary number represents
the decimal number 183 and all the beams in the frame possess
sectional properties of the beam with the identification number
183 out of 1,024 beam candidates. Similarly, substrings for col-
umn groups in the chromosome are interconnected with the iden-
tification number in the database. The string for the first column
group represented by 关110011兴 assigns all the columns in the first
group with the column section with identification number 51 in
the column database. Tables 1 and 2 present part of these data-
bases. In Fig. 6, mapping strings in the crossover operator are
those substrings in the middle, separated by two cut-off positions,
i.e., substrings between ib 1 and ib 2 for beams and between ic 1
and ic 2 for a column. To define the initial population, 300 indi-
viduals were randomly generated. Crossover probability ( p c ) is
0.8, but no mutation is allowed.
For the purpose of tracking, the 300 individuals were ranked
according to their F values at each iteration. Fig. 7 shows that the
fittest F value in each generation is decreasing as generations
proceed. Fig. 8 illustrates the converging process as the number of Fig. 11. Three-bay, 20-story reinforced concrete frame subject to
iterations increases. Only the results for the first column group is gravity and lateral loads (span⫽9,000 mm, story height
shown in Fig. 8 since the second column group showed similar ⫽3,600 mm)
trends. After 59 iterations, initially randomly distributed chromo-
somes converged to the optimum ones for beams and columns.
The size of the domain space and the sampling space with n p
sented in the section, ‘‘Construction of Database,’’ 2,048 beam
⫽300 are found to be: (2 N Balle) NGB⫻(2 N C,alle) NGC⫽(2 10) 1
candidates are generated and stored in the beam database. For the
⫻(2 6 ) 2 ⫽4.2⫻106 for the domain space and n p ⫻iteration num-
column database, only square columns are considered. w min ,
ber at convergence⫽300⫻59⫽1.77⫻104 , respectively. The
w max , ⌬w, h c min , h c max , ⌬h c , ␻ min , and ␻ max are given to 300,
order of sampling space relative to domain space is 1.77
900, 25, 300, 900, 25 mm, 1.0 and 1.0, respectively, and 128
⫻104 /(4.2⫻106 )⫽4.2⫻10⫺3 , which is in the order of O(10⫺3 ).
candidate columns are generated.
An exhaustive search was performed by evaluating all the can-
didate sections stored in the database to find a local optimum
design point. As a result, a subspace contour map of F values is
generated near the local minimum in the prescribed search space
as shown in Fig. 9. Numbers shown on abscissa represent two
different column group numbers: the first digit for one group
number and the second two digits right of the decimal point for
the second group number. Numbers shown on ordinate represent
beam group numbers. Comparing Table 3 and Fig. 9, it can be
seen that the developed GA could successfully find the local mini-
mum value in the prescribed search space for this problem in spite
of many valleys and peaks near the optimum point.

Three-Bay, Nine-Story Reinforced Concrete Frame

The frame shown in Fig. 10 is composed of three beam groups
and four column groups. For the beam database, b min , b max , ⌬b,
Fig. 12. Convergence trend for three-bay, 20-story reinforced con-
h min , h max , ⌬h, ␤ min , and ␤ max are given as 300, 400, 25, 450,
crete frame
1,000, 25 mm, 1.5 and 2.5, respectively. Following the rule pre-

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 771

J. Struct. Eng. 2003.129:762-774.

 Table 6. Results of Optimum Design for Three-Bay, 20-Story Reinforced Concrete Frame
OPTIMIZATION RESULTS
Sectional
Dimensions Reinforcements
Member Group Story Type of Section Width Depth Positive Negative
type number level member number 共mm兲 共mm兲 moment moment
Beam 1 1– 4 Beam 681 350 575 2-D25a 5-D25
2 5– 8 Beam 717 350 600 4-D25 6-D25
3 9–12 Beam 69 300 575 2-D25 5-D25
4 13–16 Beam 57 300 550 3-D25 5-D25
5 17–20 Beam 284 325 525 2-D25 4-D25

Column 1 1–5 Exterior column 1,029 725 800 12-D25

Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

2 1–5 Interior column 1,041 775 1,025 16-D25

3 6 –10 Exterior column 726 650 650 12-D25
4 6 –10 Interior column 530 575 975 14-D25
5 11–15 Exterior column 72 425 550 6-D25
6 11–15 Interior column 130 450 600 6-D25
7 16 –20 Exterior column 54 425 425 4-D25
8 16 –20 Interior column 54 425 425 4-D25

Initial design Maximum F 140.8

cost 155,020 US dollars

Minimum F 59.6
cost 137,083 US dollars

Optimum design F 43.2

cost 100,833 US dollars
a
D25 indicates deformed bar of diameter 25 mm.

Various cases were tested for this frame with varying popula- Three-Bay, 20-Story Building
tion size (n p ), percentage of population at convergence ( P c v ),
The 20-story frame 共Fig. 11兲 subject to both uniformly distributed
probability of crossover (p c ), mutation rate (p m ), and weight for
gravity loads and equivalent static lateral earthquake loads is
penalty terms (W g ). Table 4 summarizes its results. Different
composed of five beam groups and eight column groups. Com-
convergence trends were observed when different values for the
pared to the previous nine-story frame, sectional dimensions as-
parameters were used. For cases with the fixed mutation rate
sociated with b max , h max , and ␤ max for beam are increased to 500,
共Cases 11 and 12 in Table 4兲, convergence rate decreased and
1,000 mm, and 2.0, respectively. Rectangular columns are consid-
showed more variation as the mutation rate increased. It seems
ered. w min , w max , ⌬w, h c min , h c max , ⌬h c , ␻ min , and ␻ max are
that random perturbation by a fixed mutation rate might have
given to 400, 1,200, 25, 400, 2,000, 25 mm, 1.0 and 2.0, respec-
altered the superiority inherited from the parent generation and
tively. 8,192 and 2,048 candidate beams and columns are gener-
the offspring lose the ability to learn from the history of search
ated, respectively. The size of the domain space turns out to be
共Gen and Cheng 1997兲. For the given problem, Case 15, where a
(2 13) 5 ⫻(2 11) 8 ⫽1.14⫻1046. Parameter values of p c ⫽0.8, p m
gradual increase of W g and a gradual decrease of the mutation
⫽0.01, p c v ⫽20%, and W c ⫽100 are used. The p m is gradually
rate as generations proceed resulted in a better design at iteration
reduced by a factor of 0.98 at each generation. The value of W g
277 than in other cases. It is worth mentioning that the relative
was increased from the initially assigned 100 by a factor of 1.05
size of the domain space being used for the GAs’ selection pro-
at each generation until convergence is attained. Starting from
cess is 500⫻277/关 (2 11) 3 ⫻(2 7 ) 4 兴 ⫽6.01⫻10⫺14, which is in the
randomly distributed 1,000 initial designs in the domain space,
order of O(10⫺14).
the GA could successfully yield an optimized design at 360 itera-
Table 5 summarizes the results obtained for Case 15 in Table
tions, consuming 72 min. The convergence trend is shown in Fig.
4. Results from Table 5 show that interior columns have larger
12. The order of sampling space relative to domain space is
column sizes and a larger number of reinforcing bars than exterior
1,000⫻360/1.14⫻1046⫽3.16⫻10⫺41⫽O(10⫺41). Table 6 shows
columns at the same level. Also columns located below in the
the resulting design. It seems that depending on the locations of
same column line were found to need larger column sizes and a
beams and columns, their dimensions and amounts of reinforce-
larger number of reinforcing bars than columns above due to con-
ments are properly designed. The value of minimum object func-
nectivity constraints. Beams located at the lower level of the
tion 共F兲 and its cost at the initial stage are reduced from 59.6 and
frame are subject to larger moments than those in the upper level
$137,083 to 43.2 and $100,833 at final design stage, respectively.
due to lateral loads and consequently, are designed with larger
sectional dimensions and reinforcement amounts. The value of
minimum objective function 共F兲 and its cost at the initial stage are
Conclusions
reduced from 54.9 and $44,082 to 40.8 and $32,936 at final de-
sign, respectively. It took 17 min of computing time to get the Difficulties in finding optimum sections from a semi-infinite set
converged solution. of member sizes and reinforcement arrangements are alleviated

772 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.

 by constructing data sets containing a finite number of sectional Cf ⫽ cost of forming for beams and columns;
properties of beams and columns in a practical range. A particular C initial ⫽ initial cost;
chromosome representative of a particular frame consists of a ci ⫽ column size in column group i;
collection of genes, each of them representing a member group in d ⫽ effective depth of beam 共mm兲;
a frame. Databases are constructed for beams and columns, and db ⫽ flexural reinforcing bar diameter 共mm兲;
genes are mapped to the member sections in these databases. ds ⫽ stirrup diameter 共mm兲;
Given a particular sectional dimension, subsections are generated dt ⫽ diameter of tie bar 共mm兲;
according to code provisions on reinforcement amounts and their Fi ⫽ fitness value for chromosome i;
topological arrangements in a practical range. A procedure linking F max ⫽ maximum of F i ;
the GA to a database is then developed in this study for the F min ⫽ minimum of F i ;
optimization of reinforced concrete plane frames. F 1c ⫽ compressive strength of concrete 共MPa兲;
The following conclusions are made from this study: fy ⫽ yield strength of steel 共MPa兲;
1. The developed GA reinforced by a preselection operator in h ci ⫽ depth of column in column group i 共mm兲;
addition to its three basic operators 共reproduction, crossover, ⫽ depth of beam in beam group i 共mm兲;
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

hi
and mutation兲 successfully led the randomly distributed ini- L⫹ f ⫽ length factors of rebars for positive mements
tial design points in the design space to the local optimum in beam 共mm兲;
design point; L⫺
f ⫽ length factors of rebars for negative mements
2. It was found that the developed GA could successfully find in beam 共mm兲;
the local minimum for the prescribed search space for three-
L Bi j ⫽ length of jth beam in group i 共mm兲;
bay, three-story reinforced concrete frames subject to gravity
load; L Ci j ⫽ length of jth column in group i 共mm兲;
3. The developed GA reached local optimum design for the M ⫽ total number of members;
frames considered in this study, sampling small fractions of Mn ⫽ nominal flexural strength of column subject to
the domain in the order of O(10⫺3 ) for a three-story frame, axial load P u ;
O(10⫺14) for a nine-story frame, and O(10⫺41) for a 20- M⫹
n ⫽ nominal flexural strength at midspan of beam;
story frame, respectively; M⫺
n ⫽ nominal flexural strength of beam at joint;
4. Different convergence trends were observed when different Mu ⫽ externally applied moment of column subject
values for the parameters were used. For cases with the fixed to axial load P u ;
mutation rate, convergence rate decreased and showed more M⫹
u ⫽ externally applied moment at midspan of
variation as the mutation rate increased. For the given prob- beam;
lems 共three-bay, nine-story frame兲, increasing the W g and M⫺
ul ⫽ externally applied moments at left beam
decreasing the mutation rate gradually as generations pro- column joints;
⫺
ceeded resulted in a better design than in other cases; and M ur ⫽ externally applied moments at right beam
5. Although it is applied only to the optimization of reinforced column joints;
concrete plane frames, the main algorithm developed in this mc ⫽ material cost of concrete 共$/mm3兲;
study can also be applied to the discrete optimization of mf ⫽ material cost of forming 共$/mm2兲;
three-dimensional reinforced concrete frames. ms ⫽ material cost of rebar 共$/kg兲;
N alle ⫽ n p ⫻S L ;
NB ⫽ number of candidate beam sections;
Notation N B,alle ⫽ round (ln NB /ln 2);
NC ⫽ number of candidate column sections;
The following symbols are used in this paper: N C,alle ⫽ round (ln Nc /ln 2);
Ag ⫽ gross area of column 共mm2兲; NC ⫽ number of columns related to lth connectivity
A s,i ⫽ area of bar in column group i 共mm2兲; condition;
⫹ ⫽
A s,i ⫽ areas of bar for positive moments in beam NCON number of required connectivity in columns;
group i 共mm2兲; NDLC ⫽ number of different loading conditions;
⫺ ⫽
A s,i ⫽ areas of bar for negative moments in beam NGB number of beam groups;
group i 共mm2兲; NGC ⫽ number of column groups;
A sc,min ⫽ minimum reinforcement area of column, 0.01 NMBG关 i 兴 ⫽ number of beams in Beam Group i;
⫻A g (mm2 ); NMCG关 i 兴 ⫽ number of columns in Column Group i;
A sc,max ⫽ maximum reinforcement area of column, 0.08 ns ⫽ 2⫻round 兵关b⫺2⫻
⫻A g (mm2 ); (t c ⫹d s )⫹s b ]/(d b ⫹s b )⫺0.5 其 ;
⫺
A se,i ⫽ area of continuos top bar of beam group i Pn ⫽ nominal axial strength of column;
共mm2兲; Pu ⫽ externally applied axial force of column;
⫹
A se,i ⫽ area of continuous bottom bar of beam group pc ⫽ the crossover probability;
i 共mm2兲; pm ⫽ mutation rate;
A s,min ⫽ minimum reinforcement area for beam; round共•兲 ⫽ round to nearest integer;
A s,max ⫽ maximum reinforcement area for beam; SL ⫽ string length of chromosome;
b ⫽ width of beam 共mm兲; sb ⫽ larger of (d b , 4/3 maximum aggregate size,
bi ⫽ width of beam in beam group i 共mm兲; 25 mm兲;
C ⫽ total cost 共$兲; sc ⫽ spacing between logituding bars in column;
Cb ⫽ cost of rebars and concretes for beams; ⫽ larger of 共40 mm, 1.5d b , and 4/3⫻maximum
Cc ⫽ cost of rebars and concretes for columns; aggregate size兲 共mm兲;

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003 / 773

J. Struct. Eng. 2003.129:762-774.

 t c ⫽ cover thickness 共mm兲; Balling, R. J., and Yao, X. 共1997兲. ‘‘Optimization of reinforced concrete
W CON ⫽ weight of penalty function for column connec- frames.’’ J. Struct. Eng., 123共2兲, 193–202.
tivity of column; Cavicchio, D. J. 共1972兲. ‘‘Reproductive adaptive plans.’’ Proc., Associa-
W c ⫽ weight for C; tion for Computing Machinery 1972 Annual Conf., 1–11.
W g ⫽ weight for G; Cheng, F. Y., and Li, D. 共1997兲. ‘‘Multiobjective optimization design with
Pareto genetic algorithm.’’ J. Struct. Eng., 123共9兲, 1252–1261.
W M ⫽ weights of penalty function for beam of
Choi, C., and Kwak, H. 共1990兲. ‘‘Optimum RC member design with
column;
predetermined discrete sections.’’ J. Struct. Eng., 116共10兲, 2634 –
W PM ⫽ weights of penalty function for column of
2655.
column;
Chung, T. T., and Sun, T. C. 共1994兲. ‘‘Weight optimization for flexural
w ⫽ column width 共mm兲; reinforced concrete members with static nonlinear response.’’ Struct.
w c ⫽ labor cost of concrete 共$/mm3兲; Optim., 8, 174 –180.
w f ⫽ labor cost of forming 共$/mm2兲; Fadaee, M. J., and Grierson, D. E. 共1996兲. ‘‘Design optimization of 3D
w i ⫽ width of column in group i 共mm兲; reinforced concrete structures.’’ Struct. Optim., 12, 127–134.
w max ,wmin ⫽ maximum and minimum column widths, Gen M., and Cheng, R. 共1997兲. ‘‘Genetic algorithm and engineering de-
Downloaded from ascelibrary.org by Istanbul Universitesi on 07/22/13. Copyright ASCE. For personal use only; all rights reserved.

respectively; sign.’’ Reading, Wiley, New York.

w s ⫽ labor cost of rebar 共$/kg兲; Goldberg, D. E. 共1989兲. ‘‘Genetic algorithms in search, optimization, and
␤ max ,␤min ⫽ maximum and minimum ratio of beam depth machine learning.’’ Reading, Addison-Wesley Lonǵman.
to beam width, respectively; Holland, J. H. 共1975兲. ‘‘Adaptation in natural and artificial systems.’’
␤ 1 ⫽ factor determining depth of equivalent rectan- Reading, University of Michigan Press, Ann Arbor, Mich.
gular compressive stress block as defined in Information on Consumers’ Prices. 共2002兲. Korea Price Information,
ACI 共1999a, 1999b兲; Corp.
␥ s ⫽ density of rebar 共kg/mm3兲; Jenkins, W. M. 共1991兲. ‘‘Structural optimization with the genetic algo-
⌬b ⫽ beam width increment; rithm.’’ Struct. Eng., 69共24兲, 418 – 422.
⌬h ⫽ beam depth increment; Jenkins, W. M. 共1992兲. ‘‘Plane frame optimum design environment based
⌬h c ⫽ column depth increment; on genetic algorithm.’’ J. Struct. Eng., 118共11兲, 3103–3112.
⌬w ⫽ column width increment; and Koumousis, V. K., and Arsenis, S. J. 共1994兲. ‘‘Genetic algorithms in a
␻ max ,␻min ⫽ maximum and minimum ratio of column multi-criterion optimal detailing of reinforced concrete members.’’
depth to column width, respectively. Advances in structural optimization, CIVIL-COMP Ltd, Edinburgh,
Scotland, 233–240.
Rajeev, S., and Krishnamoorthy, C. S. 共1992兲. ‘‘Discrete optimization of
References structures using genetic algorithms.’’ J. Struct. Eng., 118共5兲, 1233–
1250.
Adamu, A., and Karihaloo, B. L. 共1994兲. ‘‘Minimum cost design of rein- Rajeev, S., and Krishnamoorthy, C. S. 共1998兲. ‘‘Genetic algorithm-based
forced concrete beams using continuum-type optimality criteria.’’ methodology for design optimization of reinforced concrete frames.’’
Struct. Optim., 7, 91–102. Computer-Aided Civil Infrastruct. Eng., 13, 63–74.
American Concrete Institute 共ACI兲. 共1999a兲. ‘‘Building code require- Uniform Building Code 共1997兲. International Conference of Building Of-
ments for reinforced concrete.’’ ACI 318–399, Detroit. ficials, Whittier, Calif.
American Concrete Institute 共ACI兲. 共1999b兲. ‘‘Notes on ACI 318-99 Zenon, A., Long, Z. W., and Troitsky, M. S. 共1995兲. ‘‘Designing rein-
building code requirements for structural concrete with design appli- forced concrete short-tied columns using the optimization technique.’’
cations.’’ PCA, Detroit. ACI Struct. J., 92共5兲, 619– 626.

774 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / JUNE 2003

J. Struct. Eng. 2003.129:762-774.