You are on page 1of 12

Computer-Aided Civil and Infrastructure Engineering 13 (1998) 6374

Genetic AlgorithmBased Methodology for Design


Optimization of Reinforced Concrete Frames
S. Rajeev & C. S. Krishnamoorthy
Department of Civil Engineering, Indian Institute of Technology, Madras, India

Abstract: Design optimization of reinforced concrete plane tity of one of the items affect the overall cost of the frame
frames using genetic algorithmbased methodology is pre- to a great extent. As a result, the problem becomes a com-
sented in this paper. Most of the approaches reported in the binatorial one, which ultimately ends up in the selection of
literature consider the design variables to be continuous, a combination of appropriate values of design variables for
and the optimal solution obtained requires further modifica- beam and column cross sections and quantity of reinforce-
tion to make it constructible. Since the area of reinforcement ment so that the cost of the frame is a minimum. Also, since
after detailing is different from that obtained theoretically, RC frames are cast in situ, many considerations related to
use of area of reinforcement as a continuous variable can- construction methods and practices have to be modeled in
not provide rational solutions. Aspects such as detailing and an optimal design formulation. These include keeping the
placing of reinforcement in beams and columns and other is- widths of beams at a given level constant, providing lesser
sues related to construction are brought into the optimal de- dimensions for columns at higher levels compared with those
sign model presented in this paper. Genetic algorithmbased at lower levels, etc. Factors such as keeping the breadth of
methodologies provide ideal techniques for handling such is- all members in a frame constant is advantageous from a con-
sues and generate rational optimal solutions. Examples of struction point of view. Consideration of the actual dimen-
reinforced concrete plane frames are solved, and results are sions of reinforcing steel bars in the optimal design model can
presented. Emphasis is placed on genetic modeling aspects, avoid rounding up values to the next possible size after the
which provide mechanisms for considering realistically the final solution is obtained, resulting in practical feasibility of
practical issues, resulting in a rigorous optimal design model the solution obtained. A genetic algorithmbased methodol-
providing rational solutions. ogy offers techniques for incorporating the issues enumerated
above into the optimal design model, and as a result, it can
generate solutions that need not be modified later in order
1 INTRODUCTION to make them practically feasible. This paper focuses on two
issues, namely, formulation of the design problem in the con-
Design optimization of reinforced concrete (RC) frames has text of genetic algorithms (GAs) and genetic modeling of the
been attempted by many researchers in the past, and many design optimization problem in order to obtain constructible
methods have been proposed. Most of these methods are solutions. The above-mentioned issues are presented first and
based on mathematical programming techniques. The prob- are illustrated through examples. The well-established simple
lem is more complex compared with the optimal design of genetic algorithm (SGA)5 is used to illustrate the design prob-
steel structures because in steel structures only one material is lem formulation and genetic modeling. The paper presents a
considered and the cost of the structure is assumed to be pro- brief introduction to GAs, formulation of the optimal design
portional to its weight. In the case of RC structures, however, problem of RC frames, issues in genetic modeling and solu-
three cost components due to concrete, steel reinforcement, tion of problems, and illustrative examples for demonstrating
and formwork are to be considered. Due to the difference in the applicability of the method.
unit cost values of these factors, slight variations in quan- A review of past work done in the field of optimal design
of RC structures shows that in most of the early works a lin-

To whom correspondence should be addressed. ear programming approach was adopted.12 Later, nonlinear

1998 Computer-Aided Civil and Infrastructure Engineering. Published by Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA,
and 108 Cowley Road, Oxford OX4 1JF, UK.
64 S. Rajeev and C. S. Krishnamoorthy

formulations were suggested, and methods such as sequen- the nature of the problem. In a GA-based optimal design
tial linear programming, method of feasible directions, etc. method, the two steps, namely, decoding and fitness evalu-
were employed to solve the problem.17 Attempts also were ation, are problem-dependent. Details of these steps for the
made to generate a realistic model by defining integer vari- problem under consideration are presented in a later section.
ables and solving the problem at two different levels.9 Almost Details of other steps such as initialization of population ma-
all the above-mentioned approaches and formulations con- trix, applying genetic operators, checking convergence, etc.
sidered the three cost factors noted previously. But the main are already available in the literature.7,11,15,16 Hence these
difficulty in most of the works reported in the literature is that steps are not described in this paper. The emphasis of this
the design variables are considered to be continuous, and as paper is mainly on the use of GA-based methodology to ob-
a result, the solution obtained requires further modification tain realistic optimal solutions for RC frames and not on the
to achieve practical feasibility. In the present approach, de- working of GAs.
tailing of reinforcement is considered as a design variable,
in place of area of steel adopted elsewhere, in addition to di-
mensions of the cross section. The scope of the work reported 3 OPTIMAL DESIGN FORMULATION FOR RC
in this paper is limited to rectilinear frames. FRAMES

In the present approach, cross-sectional dimensions of mem-


2 GENETIC ALGORITHMS bers and detailing of reinforcement are considered as design
variables. The objective function is the sum of the cost of con-
Genetic algorithms (GAs) simulate the Darwinian theory of crete, steel reinforcement, and formwork. The relative values
survival of the fittest and combine natural genetics and nat- of the cost factors play a major role in the optimal design solu-
ural selection phenomena along with randomized operators tion. Computation of cost of concrete and formwork is trivial
to provide search mechanisms to obtain improved solutions. once the cross-sectional parameters of beams and columns
Holland8 proposed GAs in the late 1960s, and Goldberg4 are known. Computation of cost of reinforcement is difficult,
for the first time applied the algorithm to solve engineer- since the exact quantity of reinforcement can be obtained only
ing optimization problems. Later, Goldberg5 demonstrated after detailing is complete. The formulation presented here
the usefulness of GAs for structural optimization by solv- considers detailed reinforcement as design variables. Shear
ing the classic ten-bar truss problem. Thereafter, a few re- reinforcement (stirrups) is not considered in the present for-
searchers applied the algorithm to investigate its use in struc- mulation.
tural optimization.1,2,6,11,15,16 All the engineering applica- Members of frames are first grouped into beams and
tions reported so far were with a small number of design vari- columns. Then different beam and column groups are iden-
ables, and the suitability of solving large, practical problems tified as having the same cross-sectional properties, member
has not yet been established. Depending on the nature of prob- loading, etc. In the present formulation, one beam refers to
lems being solved, improvements to the classic SGA have the continuous beam at one level of the frame. It may be
been suggested by many researchers in the literature.13,1416 a single-span beam (in the case of single-bay frames) or a
Since the scope of the work reported in the paper is limited to multispan beam. More of such beams can be combined to
design problem formulation and genetic modeling, the SGA form one group such that all the cross-sectional dimensions
is used to illustrate the work. Hence issues related to com- and reinforcement details are the same for all the beams in
putational efficiency of the GA are not considered, and a the group. Modeling of reinforcement patterns in the opti-
description of the SGA is not presented. mal design model is presented below. Consider the typical
The first step in a GA-based optimal design method is three-span beam shown in Fig. 1.
genetic modeling. This consists of identification of design One can detail the reinforcement in a beam in a number
variables and devising of a coding scheme to represent a typ- of ways. Some designers prefer bent-up bars, whereas many
ical design solution. Binary-coded transformations are the others avoid the use of bent-up bars. In the present formula-
most commonly adopted coding scheme. Depending on the tion, bent-up bars are avoided, which results in the detailing
nature of the variable, the number of binary characters re- pattern shown in Fig. 1. Two bars each at the top and the bot-
quired to define a string that will represent all possible val- tom are required to hold the stirrups in place, and they also
ues in the specified range is worked out for each variable. can contribute to the main reinforcement. Hence two pairs of
The string representing each variable is concatenated to form bars are provided, a pair each at the top and bottom for the
an individual string. Once the genetic model is defined, the entire length of the beam (marked A and B in Fig. 1). Then
optimal design process is carried out by following the stan- additional bars are provided at support and midspan to carry
dard procedure.7,11,15,16 Values of genetic parameters such design moments (marked 1, 2, 3, etc. in Fig. 1). If support sec-
as population size, probabilities of crossover and mutation, tions happen to be doubly reinforced and require compression
convergence parameter, etc. have to be fixed depending on reinforcement, the continuous bars provided (marked B) at
Genetic AlgorithmBased Methodology for Design Optimization of Reinforced Concrete Frames 65

Fig. 1. Typical three-span beam with detailed reinforcement.

the bottom can contribute to the compression reinforcement. the length of the kth member, d 0 is the cover provided for
Hence no separate additional bars are provided for compres- reinforcing steel, and CC, C S, and C F are the unit cost of
sion reinforcement at support and midspan. If beams at more concrete, steel, and formwork, respectively.
than one level have similar loading, then they can be grouped
such that the dimensions and reinforcement in those beams Beams
are the same. Problem size can be considerably reduced by
The area of reinforcement is not kept the same throughout the
such grouping.
beam member. Hence three typical sections, namely, one near
For columns, the main detailing requirements are that there
the left support, one at midspan, and one near the right sup-
should be a minimum of four bars and the number of bars
port, are considered in the design formulation. Seven design
provided always should be an even number. It is a common
variables are defined for each span of a beam. They are
practice to keep the dimensions and reinforcement in columns
the same for two to three storys. Such grouping of columns x1 = breadth of cross section (kept the same
serves also to reduce the number of design variables. The for all spans of a beam)
objective function and constraints are separately specified
for column and beam groups and are described below. x2 = depth of cross section (can vary from span
to span of the same beam)
3.1 Objective function x3 = area of top continuous bars
x4 = area of bottom continuous bars
Three variables are defined for each column group, namely, x5 = area of top additional bars at left support
breadth, depth, and area of reinforcement in the column. Since
reinforcement in columns is kept constant throughout the x6 = area of bottom additional bars at midspan
member, one variable alone is sufficient to represent rein- x7 = area of top additional bars at right support
forcing steel. The following design variables are defined for
a column member: Then the objective function for a beam group can be written
as
Columns X X
NB spans n
f (x) = L k x1 (x2 + d 0 ) CC
x1 = breadth of cross section k=1 j=1
x2 = depth of cross section +[x1 + 2(x2 + d 0 )] C F
x5 x6 x7 o
x3 = area of reinforcement in the column + x3 + x4 + + + CS (2)
4 2 4
Then the objective function for a column group can be written
in which NB is the number of beams in the group under
as
consideration.
X
NC
f (x) = L k {x1 (x2 + d 0 ) CC + x3 C S 3.2 Constraints
k=1
+ 2[x1 + (x2 + d 0 )] C F} (1) This section describes constraints imposed on column and
beam members. Constraints for column group are discussed
in which N C is the number of columns in the group, L k is first and then for beam groups.
66 S. Rajeev and C. S. Krishnamoorthy

Columns The depth-to-breadth ratio of a column section should be less


than a specified limit dr in order to avoid very deep sections:
Six constraints are imposed for column groups. The first one
specifies that the design moment Md should be less than the dr x1
moment-carrying capacity Mu of the section, that is, Md 1 0 (8)
x2
Mu . In normalized form this can be represented as
Also, breadth of the column section should be less than depth,
Mu
1 0 (3) as
Md x2
1 0 (9)
Due to the interaction between axial load and moment, the x1
ultimate moment of resistance Mu cannot be explicitly ex- Constraints on minimum breadth required to avoid thin cross
pressed in terms of the design variables. In an RC section, sections are satisfied in an implicit manner.
the maximum strain is a function of the depth of the neutral
axis. Hence force and moment equilibrium equations can be Beams
written in terms of stress block parameters K 1 , K 2 , and K 3 ,
depth of neutral axis, and the design variables. The stress Since the behavior of beams is different from that of columns,
block parameters are defined as shown in Fig. 2. The force separate constraints are specified for beam groups. Altogether
equilibrium equation can be written as eight constraints are specified. As in the case of columns, the
first constraint is on the moment-carrying capacity of the
X
NR
beam section. The design ultimate moment Md should be
Pu = K 1 K 3 f ck xu x1 + Asi ( f si f ci ) (4)
i=1
less than the ultimate moment of resistance Mu of the section
under consideration:
in which f ck is the characteristic strength of concrete, and
f si and f ci are, respectively, the stress in steel and concrete Mu
1 0 (10)
at the level of the ith row of reinforcement. The moment Md
equilibrium equation is written as
For each beam group, the design constraints are evaluated for
(x2 + d 0 ) maximum moment at the end sections and midspan section
Mu = K 1 K 3 f ck xu x1 K 2 xu
2 for each span of the beam. The constraints on design ultimate
moment are expressed as
X
NR
+ Asi ( f si f ci )yi (5)
K 2 0.87 f y x j j
i=1 1 0.87 f y x j x2 /Md 0 (11)
K 1 K 3 f ck x1
in which Mu is the moment about the centroid of the section,
yi is the distance of the ith row of reinforcement from the in which x j = x3 +x7 for the left support ( j = 1), x j = x4 +x6
centroid of the section, and NR is the number of rows of for the midspan ( j = 2), and x j = x3 +x7 for the right support
reinforcement. For the given load, the neutral axis depth xu is ( j = 3) for each span.
fixed by an iterative process satisfying the force equilibrium In order to ensure ductile failure, the beam must be de-
equation given in Eq. (4). This design constraint is evaluated signed as an underreinforced beam, and the constraint can
at the two end sections of each member in a column group. In be expressed in normalized form as below, where, pt,lim is
the present work, values for stress block parameters K 1 , K 2 , a limiting percentage of steel corresponding to the balanced
and K 3 as specified in the Indian Standard Code of Practice section:
for Reinforced Concrete10 are used. For other codes such as x1 x2 pt,lim
1 0 (12)
ACI, DIN, and CP-110, appropriate values for stress block 100x j
parameters can be adopted.
which should be satisfied at all the three cross sections of a
A minimum percentage of reinforcement pmin of the area
beam, i.e., j = 1, 2, and 3.
of the cross section is prescribed in the following form:
The next constraint specified is on the minimum percentage
x3 of reinforcement to limit excessive cracking:
1 0 (6)
pmin x1 (x2 + d 0 )
xj
To avoid difficulty in placing and compacting concrete in 1 0 (13)
pmin x1 x2
formwork, the area, of reinforcement is limited to pmax spec-
ified as a percentage of the cross-sectional area, prescribed This constraint also has to be satisfied for all three beam
as sections, i.e., j = 1, 2, and 3.
pmax x1 (x2 + d 0 ) To avoid difficulties in placing and compacting concrete
1 0 (7) in formwork, the maximum area of reinforcement is limited
x3
Genetic AlgorithmBased Methodology for Design Optimization of Reinforced Concrete Frames 67

Fig. 2. Stress block parameters for beam and column sections.

to pmax as a percentage of the cross-sectional area. This is in which f p is a function of the percentage of reinforcement
expressed as provided and is computed as specified in the Indian Standard
Code of Practice for Reinforced Concrete.10 This constraint
pmax x0 (x2 + d 0 )
1 0 (14) can be expressed in normalized form as
xj
26 f p x2
Very deep beam sections have to be avoided, and the follow- 1 0 (18)
ing constraint ensures that the depth-to-breadth ratio of beam span
cross section is limited to a specified value dr :
Shear reinforcement and stirrups are not considered in the
dr x1 present formulation. Genetic modeling and transformation
1 0 (15) of the design space from parameter space to genetic space
x2
are discussed in the next section.
Depth of the section should be greater than or equal to the
breadth. This is specified as
x1 4 GENETIC MODELING
1 0 (16)
x0
The formulation discussed above takes into account as de-
The serviceability requirements for deflections are imposed sign variables, the cross-sectional dimensions of beams and
by many codes of practice in the form of span-to-effective- columns, and the area of reinforcement for columns and at
depth ratios. For continuous beams, the span-to-effective- three sections in a beam span. These design variables can be
depth ratio should satisfy the following relation: classified into two types. In the first type, i.e., cross-sectional
span dimensions, values that can be assigned to them vary in fixed-
26 f p (17) step size. In the second type of variable, i.e., area of reinforce-
x2
68 S. Rajeev and C. S. Krishnamoorthy

Table 1
Allowable combination of reinforcement bars for columns

Area Area
Di Bar combinations (mm2 ) Di Bar combinations (mm2 )
1 4 12 452 2 6 12 678
3 4 16 804 4 4 16 + 2 12 904
5 6 16 1206 6 4 20 1256
7 4 20 + 2 12 1482 8 4 20 + 2 16 1658
9 6 20 1884 10 4 25 1964
11 4 25 + 2 12 2190 12 4 25 + 2 16 2366
13 4 25 + 2 20 2592 14 6 25 2964

ment, values that can be assigned to them vary in variable-step and depth. Thus a string with 10 characters can represent a
sizes. column group.
The information available at the time of designing is the Genetic modeling for beams is also made simple based
bar diameters that can be used and allowable combinations on the reinforcement detailing adopted, as described in the
of bars. The issues that have to be taken into consideration preceding section. Detailed reinforcement is modeled in order
during genetic modeling are to obtain constructible solutions.
Four different types of reinforcements marked A, B, 1, 2,
Discrete sizes and combinations of reinforcement bars 3, etc. in Fig. 1 are provided in a beam at one level. A and
Appropriate grouping from connectivity aspects B are the bars provided for the entire length of the beam. In
Different reinforcements for members of the same group addition to serving as holder for stirrups, they can contribute
Different depths for beams in the same group to the main reinforcement in tension or compression. Any
Reinforcement detailing from available bars with accept- additional bars required either at support or at midspan to
able combinations carry moments are provided, which are marked 1, 2, 3, etc. in
Detailing at supports based on reinforcements in adjacent Fig. 1. The top reinforcement on either side of an intermediate
sections support need not be the same. In such cases, separate design
As described in the section on formulation, frame members variables are assumed for additional bars for the left and right
are divided into beam groups and column groups. Regard- spans. In the problems solved in this paper, both of these are
ing the physical dimensions, namely, breadths and depths of assumed to be the same. Thus the design variables considered
beams and columns, constant step size is assumed. For the for a beam are
problems solved in this paper to illustrate the concepts, the Breadth of beam
breadth or depth of beams or columns is assumed to vary Depth of beam
from 200 to 550 mm. Depending on the type of frame and Top continuous bar diameter (A)
loading, these limits can be appropriately fixed. The step size Bottom continuous bar diameter (B)
is decided based on size of standard formwork material avail- Additional top reinforcement at support
able. From practical considerations, a step size of 50 mm is Additional bottom reinforcement at midspan
assumed in the present work. Reinforcement variation cannot
be assumed like this. The discrete variation depends on the A typical character string to represent a beam at a given
number, sizes, and combination of bars used. For columns, level is formed as described below. Three characters each
the minimum number of bars should be four, and it should are used to represent breadth and depth of a beam. It is as-
always be an even number. To reduce the number of design sumed that breadth and depth of beams are kept constant at
variables, columns of two or three levels can be combined in a a level. To model the detailing practice described above, a
group, which is also quite common in construction practice. number of different sets of reinforcement bar combinations
A set of allowable combinations of reinforcement bars for are assumed. Five such sets are shown in Tables 2 through 6.
column members is specified as shown in Table 1. Character Di represents the decoded value for the ith variable. These
strings were decoded and mapped into the values of the set sets specify combinations having two bars of one diameter
to obtain appropriate reinforcement bar combinations. It can with other additional bars. For instance, the combinations
be seen that there are 14 bar combinations in Table 1. Four shown in Table 3 have 2 bars of 12-mm diameter plus ad-
characters are therefore required for a binary coding scheme ditional bars that are generally provided in conjunction with
to represent column reinforcement if any one of the 14 values two 12-mm bars. Seven different combinations are speci-
is to be assigned. Another 6 characters can represent breadth fied in each set. Depending on the nature of loadinghigh
Genetic AlgorithmBased Methodology for Design Optimization of Reinforced Concrete Frames 69

Table 2 Table 5
Allowable bar combinations for beams with 2 10 mm Allowable bar combinations for beams with 2 20 mm
diameter bars diameter bars

Di Bar combinations Di Bar combinations


0 2 10 0 2 20
1 2 10 + 1 10 1 2 20 + 2 12
2 2 10 + 1 12 2 2 20 + 1 12
3 2 10 + 2 12 3 2 20 + 2 16
4 2 10 + 1 16 4 2 20 + 1 16
5 2 10 + 2 16 5 2 20 + 1 20
6 2 10 + 1 20 6 2 20 + 2 20

Table 3 Table 6
Allowable bar combinations for beams with 2 12 mm Allowable bar combinations for beams with 2 25 mm
diameter bars diameter bars

Di Bar combinations Di Bar combinations


0 2 12 0 2 25
1 2 12 + 1 12 1 2 25 + 2 16
2 2 12 + 1 16 2 2 25 + 1 16
3 2 12 + 2 16 3 2 25 + 2 20
4 2 12 + 1 20 4 2 25 + 1 20
5 2 12 + 2 12 5 2 25 + 2 25
6 2 12 + 2 20 6 2 25 + 1 25

Table 4
Allowable bar combinations for beams with 2 16 mm regardless of the number of member groups, only three char-
diameter bars acters are needed to represent the breadth of all members,
which results in reduced individual string length. In a simi-
Di Bar combinations lar manner, any kind of practical issues can be modeled so
0 2 16 that the solution obtained is rational and does not require
1 2 16 + 2 12 subsequent modification for constructibility.
2 2 16 + 1 12 The two problem-dependent computations in standard GA-
3 2 16 + 2 16 based optimization procedures are decoding and evaluation of
4 2 16 + 1 16 fitnesses of individual strings. The decoding scheme adopted
5 2 16 + 2 20 is used to obtain decimal (base 10) values for design vari-
6 2 16 + 1 20
ables. For instance, let there be three characters to represent
the depth of a beam, and suppose the decoded decimal value
is 5. Then the value of the design variable is computed as
or mediumany four of the preceding sets are considered 200 + 5 50, resulting in 450, in which 200 is the lower limit
at a time, which requires two characters to be represented for depth and 50 is the step size assumed. Let the decoded
in a string. Another three characters are required to repre- decimal value of the string representing continuous bars at the
sent additional bars at each critical section. For instance, in bottom of a beam be 3. This implies that the reinforcement
a three-span beam, 7 sections are considered, which results values are to be taken from the fourth table, i.e., Table 5.
in 21 characters to represent additional reinforcement bars. Similarly, let the decoded decimal values corresponding to
Thus 31 characters represent a three-span beam, as shown reinforcement at that section be 4. Then the fifth combina-
in Table 7. More such beams with similar span and loading tion, i.e., 2 20 + 1 16, is selected for reinforcement at
can be grouped together in order to reduce the problem size that section. Once the dimensions of beams and columns and
in genetic space. Depending on the number of beam groups details of reinforcements are known, the objective function
and column groups, corresponding strings are concatenated and constraints are evaluated. Fitness of a design solution
to form an individual string in the population. represented by a string in the population is computed as
In many cases, from the construction point of view, it would
be desirable to have all the members the same breadth. Then, Fi = f i (1 + C) (19)
70 S. Rajeev and C. S. Krishnamoorthy

Table 7
Composition of typical individual string for a three-span beam

Additional bars at sections


Top Bottom
bar bars
Breadth Depth diameter diameter 1 2 3 4 5 6 7
No. of characters 3 3 2 2 3 3 3 3 3 3 3

Fig. 3. Geometry and loading: three-bay four-story frame.

in which Fi is the fitness, f i is the value of objective function form. All other steps in the optimization process follow the
of the ith individual, and C is the sum of all violated con- standard SGA.
straints for the solution under consideration. Since the prob-
lem is a minimization problem, the fitness is transformed to
facilitate carrying out the reproduction operation using the 5 ILLUSTRATIVE EXAMPLES
transformation function
Two problems are solved to illustrate the genetic modeling
issues as well as design optimization using the proposed
Fi0 = (Fmax + Fmin ) Fi (20) method. The genetic parameters assumed in the present work
are
in which Fmax and Fmin , respectively, are the maximum and Probability of crossover = 0.8
minimum fitness values in the current generation, and Fi0 is Probability of mutation = 0.001
the transformed fitness used for the reproduction operation. Convergence parameter = 0.85
The preceding equations for computing fitness can be used
only when all the constraints are represented in normalized The minimum value for breadth of beam and column mem-
Genetic AlgorithmBased Methodology for Design Optimization of Reinforced Concrete Frames 71

Fig. 4. Details of optimal solution: three-bay four-story frame.

bers is limited to 200 mm. A maximum value of 3 is assumed solution was obtained for a population of 90. The optimal so-
for depth-to-breath ratio in order to avoid very thin sections. lution obtained had a cost function value of 20,737.9, whereas
For all problems, the following values are assumed for unit that obtained using SUMT based on a direct search method13
cost parameters. had a cost function value of 22,211.3, which is 7.1% higher.
The real cost advantage will be much more when comparison
Unit cost of concrete (CC) = 735/m3
of the present result is made with that obtained from SUMT
Unit cost of steel reinforcement (C S) = 7100/ton
after the detailing for constructibility. Details of the optimal
Unit cost of formwork (C F) = 54/m2
solution obtained are given in Fig. 4.
These values are assumed only for comparing the solution ob-
tained with those values obtained with a classic optimization 5.2 Two-bay six-story frame
method SUMT using a direct search technique.13
The frame has 30 members, as shown in Fig. 5. Members are
5.1 Three-bay four-story frame divided into three column groups and two beam groups. The
full column from the ground to the roof level in one column
The geometry and loading details of the frame are shown line is considered to be one group. The roof beam is in the
in Fig. 3. The frame consists of 16 columns and 12 beams. first beam group, and all the other beams are in the second
Design member linking is adopted by combining columns group. In contrast to the preceding problem, the depths of the
of two successive storys to form eight column groups. Sim- beam at the same level are allowed to vary in different spans.
ilarly, four beam groups are formed by assuming beams at Two variables are assumed for depth of the two spans. This
each story level to be in the same group. The genetic model genetic model resulted in an individual string length of 100.
for this problem consists of an individual string of length 204. The solution obtained in the present work had a cost function
The problem size can be reduced if all the floor beams are value of 17,804.1 and that obtained from SUMT was 19,480,
combined together in one group. A number of runs were made the difference being 8.6%, which is considerable. The details
with population sizes ranging from 80 to 120. The optimal of the solution obtained are presented in Fig. 6. Progress of
72 S. Rajeev and C. S. Krishnamoorthy

Fig. 5. Geometry and loading: two-bay six-story frame.

the optimization in the form-generation history is shown in placing of reinforcement. The issues of computational effi-
Fig. 7. ciency of the genetic algorithm used are not addressed, since
the focus of the paper is on design problem formulation in
the context of GA, genetic modeling, and its illustration. Two
6 SUMMARY AND CONCLUSIONS example problems are solved, and the results are presented
to illustrate the proposed method. Based on the present work,
Design optimization of RC plane frames using genetic the following conclusions are drawn:
algorithmbased methodology is presented in this paper.
From the design optimization point of view, these structures
are difficult to handle, due to the discreteness of member sizes 1. Genetic algorithmbased methodology provides an
and reinforcement, including consideration of detailing to en- ideal technique to model practical design considera-
sure appropriate placement of reinforcement in the members. tions such as predefined discrete variations in depth
These issues are addressed in the present optimal design ap- and breadth of concrete frame members and detailing
proach, and rational design solutions are obtained that are the combinations and placing of reinforcement bars.
constructible. In addition to demonstrating the usefulness of 2. The proposed genetic modeling concepts allow the de-
the proposed optimal design methodology, a realistic optimal signer to specify available sizes and allowable combi-
design formulation and genetic modeling are presented that nations of reinforcement bars.
take into account a number of factors related to detailing and 3. The methodology arrives at rational and realistic design
solutions that are directly constructible.
Genetic AlgorithmBased Methodology for Design Optimization of Reinforced Concrete Frames 73

Fig. 6. Details of optimal solution: two-bay six-story frame.

REFERENCES

1. Adeli, H. & Cheng, N. T., Integrated genetic algorithm for opti-


mization of space structures, Journal of Aerospace Engineering,
ASCE, 6 (4) (1993), 31528.
2. Adeli, H. & Cheng, N. T., An augmented lagrangian genetic
algorithm for structural optimization, Journal of Aerospace En-
gineering, ASCE, 7 (1) (1994), 10418.
3. Adeli, H. & Cheng, N. T., Concurrent genetic algorithms for
optimization of large structures, Journal of Aerospace Engi-
neering, ASCE, 7 (3) (1994), 27696.
4. Goldberg, D. E., Computer-aided gas pipeline operation using
genetic algorithms and rule learning, Dissertation Abstracts In-
ternational, 44 (10) (1983), 3174.
5. Goldberg, D. E. & Samtani, M. P., Engineering optimization
via genetic algorithms, Electronic Computation, ASCE, (1986),
4716.
6. Grierson, D. E. & Pak, W. H., Optimal sizing, geometrical and
topological design using a genetic algorithm, Structural Opti-
Fig. 7. Progress of optimization: two-bay six-story frame. mization, 6 (1993), 1519.
7. Hajela, P., Genetic search: An approach to nonconvex optimiza-
tion problems, AIAA Journal, 28 (7) (1990), 120510.
8. Holland, J. H., Adaptation in Natural and Artificial Systems,
4. The proposed methodology is less mathematically com- University of Michigan Press, Ann Arbor, 1975.
plex and easier than the traditionally used mathematical 9. Huanchun, S. & Zheng, C., Two-level optimum design of re-
programming techniques and gives realistic and con- inforced concrete frames with integer variables, Engineering
structible solutions.
74 S. Rajeev and C. S. Krishnamoorthy

Optimization, 9 (1985), 21932. 14. Kumar, S. & Adeli, H., Minimum weight design of large struc-
10. IS:456-1978, Code of Practice for Plain and Reinforced Con- tures on a network of workstations, Microcomputers in Civil
crete, Beurau of Indian Standards, New Delhi, India, 1978. Engineering, 10 (6) (1995), 42332.
11. Jenkins, W. M., Structural optimization using genetic algo- 15. Rajeev, S., Genetic algorithmbased methodologies for design
rithms, The Structural Engineer, 69 (24) (1991), 41822. optimization of framed structures, Ph.D. thesis, Indian Institute
12. Krishnamoorthy, C. S. & Munro, J., Linear program for optimal of Technology, Madras, India, 1993.
design of reinforced concrete frames, Proceedings of IABSE, 3 16. Rajeev, S. & Krishnamoorthy, C. S., Discrete optimization of
(1) (1973), 11941. structures using genetic algorithms, Journal of Structural En-
13. Krishnamoorthy, C. S. & Rajeev, S., Computer-aided optimal gineering, ASCE, 118 (5) (1992), 120520.
design of reinforced concrete frames. In Proceedings of the In- 17. Shamnugavel, P., Optimization of two dimensional reinforced
ternational Conference on Engineering Software, eds. C. V. Ra- concrete frame, Ph.D. thesis, University of Illinois at Urbana
makrishnan & A. Varadarajan, Narosa Publishing House, New Champaign, 1974.
Delhi, India, 1989, pp. 91724.

You might also like