You are on page 1of 11

Advances in Engineering Software 31 (2000) 863±873

www.elsevier.com/locate/advengsoft

Optimum spacing design of grillage systems using a genetic algorithm


M.P. Saka a,*, A. Daloglu b, F. Malhas c
a
Civil Engineering Department, University of Bahrain, Isa Town, Bahrain
b
Civil Engineering Department, Blacksea Technical University, Trabzon, Turkey
c
Civil Engineering Department, Bradley University, Peroira, IL, USA
Received 7 February 1998; accepted 17 July 1999

Abstract
In this study a genetic algorithm based method is developed for the optimum design of grillage systems. The algorithm not only selects the
optimum sections for the grillage elements from a set of standard universal beam sections, but also ®nds the optimum spacing required for the
grillage system. De¯ection limitations and allowable stress constraints are considered in the formulation of the design problem. Due to the
fact that grillage elements are thin walled sections, warping plays an important role in their design, particularly, when they are subjected to
signi®cant torsional loading. The algorithm developed has the ¯exibility of including or excluding the effect of warping in the design process.
Grillage structures are designed for uniformly distributed loading. The optimum spacings are determined both considering and not taking into
account the effect of warping in the design. The comparison of the results shows that inclusion of warping in the design process has a
signi®cant effect on the optimum spacing. q 2000 Civil-Comp Ltd and Elsevier Science Ltd. All rights reserved.
Keywords: Grillage systems; Genetic algorithm; Warping; Optimum design; Thin-walled sections; Optimum spacing

1. Introduction up the ®nal results to standard sizes. This approach unfortu-


nately ruins the optimality and often causes the loss of what
Structural optimization, since its emergence, has attracted was gained through optimization. A number of procedures
a wide spread attention among designers. It has provided a have been developed which can work with discrete
systematic solution to age-old structural design problems variables, some of which make use of branch and bound
which were previously handled by trial±error methods or algorithm [2,3,5] while the other uses the generalized opti-
engineering intuition, or both. Applications of mathematical mality criteria method [4]. However, these techniques have
programming methods into structural design problems has not found wide spread application due to their complexity
paved the way in obtaining a design procedure which is and inef®ciency in dealing with large-size structures.
capable of producing structures with cross-sectional dimen- The recent developments in the application of biological
sions having minimum weight or cost [1]. In the last three principles into computational algorithms have produced a
decades a vast amount of research work has been conducted different class of numerical optimization methods. Among
in structural optimization, from optimum design of indi- these, genetic algorithms have attracted great attention due
vidual elements to complete frame works and continuum to its ability of providing a solution to discrete optimum
structures. However, due to the fact that most mathematical design problems [6,7]. In a recent application, a genetic
programming techniques usually deal with continuous algorithm was integrated with the penalty function method
design variables, the algorithms developed have provided [8]. In this work, different types of genetic operations are
designs where cross-sectional dimensions that are neither used and their effect in obtaining the optimum solution was
standard nor practical. Consequently, structural optimiza- investigated. Later, this method was employed in the opti-
tion has not enjoyed the same popularity among practicing mum design of realistic and large space structures [9,10].
engineers as it has enjoyed among researchers. As a result, The genetic algorithm was also found quite promising in
efforts have concentrated on the area of modifying structural obtaining the optimum topology of structures. In some of
optimization algorithms to be able to work with discrete sets the works nodal locations are treated as design variables in
of variables. One way to deal with this problem is to round addition to cross-sectional areas [11,12], in some others the
concept of ground structure is used [13±16].
* Corresponding author. Tel.: 1 973-782-194; fax: 1 973-684-844. In this article, a genetic algorithm is applied to obtain
E-mail address: mpsaka@batelco.com.bh (M.P. Saka). the optimum spacing and section designations of grillage
0965-9978/00/$ - see front matter q 2000 Civil-Comp Ltd and Elsevier Science Ltd. All rights reserved.
PII: S 0965-997 8(00)00048-X
864 M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873

structures. The algorithm developed selects the beams of the value of either 1 or 0. The binary code for each design
grillage system from the Universal Beam Sections of BS4 variable represents the sequence number of this variable
(British Standards) [18]. Furthermore, it determines the in the discrete set.
optimum longitudinal and transverse spacing between the A genetic algorithm initiates the search for ®nding the
beams when the grillage is subjected to uniformly distri- optimum in a discrete space by ®rst selecting the number
buted loading. The algorithm also has the ¯exibility of of individuals randomly and collecting them together to
including or excluding the effect of warping as well as constitute the initial population. It then makes use of four
shear in ®nding the optimum design of a grillage structure. operators to generate a new population. These operators are
selection, mating, crossover and mutation. The detailed
explanation of these operators is given in Refs. [6,21,22].
2. Mathematical model Among these, the crossover operator is probably the one
which plays an important role in the production of the
The optimum spacing design problem of a grillage new generation. There are several types of crossover opera-
structure can be written as follows: tors such as single point, two-point, multi-point, uniform
X
ng X
rk and variable crossover. It is shown in Ref. [23] that two-
min Wˆ mk `i point crossover performs much better among the multi-point
kˆ1 iˆ1 crossover techniques. The detailed study carried out on the
( …1†
dj 2 dju # 0; j ˆ 1; ¼; p; evaluation of crossover techniques has shown that direct
s:t: design variable exchange produced the best solutions in
s nmax 2 s nu # 0; n ˆ 1; ¼; nm: the test problems considered [24]. Further, operators such
where mk is the unit weight of the grillage element selected as niche and cloning are suggested in Ref. [25].
from Universal Beam Sections of BS4 belonging to group k,
rk the total number of elements in group k and ng the total
number of groups in the system; `i is the length of the 4. Optimum design algorithm
element i; d j is the displacement of joint j, d ju its upper
The genetic algorithm summarized in the previous
bound and p the total number of restricted displacements;
section is used to develop a technique for the optimum
s nmax is the maximum normal stress in element n and s nu is
spacing of grillage systems. In order to establish a ®tness
the allowable stress; and nm is the total number of elements
criterion as required by the genetic algorithm, it is necessary
in the structure.
to transform the constrained design problem of (1) into an
It is usual practice in grillage systems to make all the
unconstrained one. This is achieved by using a penalty func-
longitudinal beams out of same section and the transverse
tion. This function is associated with the objective function
beams out of another. Furthermore, it is also practical to
and whenever the constraints are violated, it generates a
have equal spacing for longitudinal beams and equal
penalty to the objective function. There are different types
spacing for transverse beams. This introduces four variables
of penalty functions. While the quadratic penalty function is
into the design problem. The ®rst two of these variables are
associated with genetic algorithm in Refs. [7,17], an
the designation of the universal beam sections to be adopted
augmented lagrangian type penalty function was found
for the longitudinal and transverse beams. The last two of
quite effective in solving the optimum design problem of
these variables are the spacing variables which are selected
large structures [8±10]. In this paper, the transformation
as the number of beams in longitudinal and transverse direc-
based on the violation of normalized constraints as
tions. It is apparent that all the design variables are discrete
suggested in Ref. [17] is employed in the genetic algorithm.
variables and the genetic algorithm is employed to ®nd the
The normalized form of design constraints is expressed as
solution of the design problem.
follows:
dj
3. Genetic algorithms g j …x† ˆ 2 1 # 0; j ˆ 1; ¼; p;
dju
…2†
Genetic algorithms are developed by applying the princi- s max
pal of survival of the ®ttest into a numerical search method gn …x† ˆ n 2 1 # 0; n ˆ 1; ¼; nm:
s nu
[6,7]. They are used as function optimizers particularly
when the variables have discrete values. They achieve this The unconstrained function P(x) is then constructed as
by ®rst selecting an initial population where each individual !
is constructed by bringing together the total number of Xm
P…x† ˆ W…x† 1 1 C ni …3†
variables in a binary or other coded form. These individuals iˆ1
are called arti®cial chromosomes and they have a ®nite
length string. Every character in the string is an where W(x) is the objective function given in (1), C is a
arti®cial gene, which in the case of binary code has a constant to be selected depending on the problem and n i is
M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873 865

a violation coef®cient which is computed as where zi is the displacement along the z-axis, u xi and u yi are
the rotations about the x- and y-axis, respectively. The corre-
if gi …x† . 0 then ni ˆ gi …x†; if gi …x† # 0 then ni ˆ 0
sponding joint load vector becomes
…4†
where i varies from 1 to m, which is the total number of {Pi } ˆ {Qzi Mxi Myi Mwi }: …7†
constraints. The function P(x) is an unconstrained function,
a minimum of which will be obtained by the genetic algo- Member end forces and corresponding displacements for
rithm. The algorithm requires a criterion to carry out the a grillage element are shown in Fig. 2. It is apparent that if
selection among the individuals. This is carried out in the effect of warping is ignored in the analysis of the grillage
such a way that the best individual has the maximum ®tness. systems, u wi will disappear from the joint displacement
The expression for ®tness is selected as vector of (6). The degree of freedom for a grillage joint
then becomes three, namely zi, u xi, and u yi. The correspond-
Fi ˆ …P…x†max 1 P…x†min † 2 Pi …x† …5†
ing joint load vector consists of Qzi, Mxi and Myi as Mwi drops
where Fi is the ®tness of the ith individual, P(x)max and out from the load vector. The stiffness matrix of a grillage
P(x)min are the maximum and minimum values of the uncon- element with three degrees of freedom at each joint is avail-
strained function of (3) among the all population. Pi(x) is the able in standard textbooks and is not repeated here. Instead
value of (3) computed for the ith individual. The ®tness the stiffness matrix of grillage element where the effect of
factor P for each individual is calculated by Fi =F av , where warping is taken into account is considered and given with
Fav ˆ Fi =n; in which n is the total number of individuals relative detail.
in the population. It is this ratio that decides whether the The stiffness matrix of a grillage element is obtained by
particular individual should die or should be copied into the combining its ¯exural and torsional stiffness matrices [19].
mating pool. If the factor is less than 0.5 the individual dies, Such matrix for a typical element rigidly connected to joint i
if it is more than 1.5 it sends two copies to the mating pool. and j, as shown in Fig. 1, has the following form:
After the mating pool is constructed, individuals are coupled 2 3
randomly and crossover is applied as explained in the ..
6 d e 2h 0 . d e 2h 0 7
previous section. 2 3 6
6 ..
72
7 d 3
Qzi 6 e
It is apparent that computation of the ®tness of an indi- 6 7 6 f g 2s . 2e v z 2s 7
76 i 7
6M 7 6 76
76 uxi 7
6 xi 7 6 .. 7
vidual requires the values of displacements and stresses in 6 7 6 76 7
6 7 6 2h g u 2t . h z w 2t 7
76 7
the grillage system with the values of design variables that 6 Myi 7 6 6
76 yi 7
u
6 7 6 7
6 7 6 .. 76 7
are coded in the individual. This is achieved by carrying out 6 Mwi 7 6 0 2s 2t T3 . 0 s t T4 7
76 uwi 7
6 7 6 6 7
6 7 6 76 7
the analysis of the system under the external loads. 6 ¼ 7 6 6 .. 76
7 ¼ 7
6
6
7ˆ6 ¼
7 6 ¼ ¼ ¼ . ¼ ¼ ¼ ¼ 76 7 …8†
6 7 76 7
6 Qzj 7 6 76 7
6 7 6 .. 766 d j 7
7
4.1. Analysis of grids including warping and shear effects 6 7 6 2d 2e h 0 . d 2e h 7
0 76 7
6M 7 6 76 7
6 xj 7 6 6 u 7
7 6 76
.. xj
6 7
6 7 6 e v z s . 2e f g s 7 6
76 uyj 7
The steel grillage system shown in Fig. 1 is made out of 6 Myj 7 6 7
4 5 6
6
74
7 5
6 .. 7
universal beam sections and is subjected to transverse out- Mwj 6 2h z w t . h g u t 7 uwj
6 7
of-plane loading. When the members are rigidly connected 4 5
..
to each other, bending and torsional moments develop at 0 2s 2t T4 . 0 s t T3
their ends due to external loading. The universal beam
sections are thin walled sections and they posses very where
small torsional rigidity. When such sections are subjected
to torsional moments, they warp and if the warping is d ˆ k1 ; e ˆ k2 sin g; h ˆ k2 cos g; s ˆ T2 cos g;
restrained, it causes large values of normal stresses in the
section. Hence, it becomes necessary to consider the effect t ˆ sin g; f ˆ k3 sin 2 g 2 T1 cos2 g;
of warping in the analysis of grillage systems [19,20].
g ˆ 2…k3 1 T1 †sin g cos g; u ˆ k3 cos2 g 2 T1 sin2 g;
The warping of a thin walled section is caused by bi-
moment Mw which results from the effect of non-uniform z ˆ …T1 2 K4 †sin g cos g; v ˆ k4 sin2 g 1 T1 cos2 g;
torsion of a grillage element. Due to its simplicity in the
formulation, it is common practice to consider the twisting w ˆ k4 cos2 g 1 T1 sin2 g: (9)
of unit length uw ˆ du=dy as a corresponding deformation
for Mw. Hence, the degree of freedom of a joint in a
grillage system becomes four. The displacement vector for The ¯exural stiffness coef®cients are
joint i of a grillage system shown in Fig. 1 has the following
12EIc1 6EIc1 4EIc2 2EIc3
form: k1 ˆ 3
; k2 ˆ 2
; k3 ˆ ; k4 ˆ
` ` ` `
{Xi } ˆ {zi uyi uyi uwi } …6† …10†
866 M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873

Fig. 1. Grillage system.

where c1, c2 and c3 are the terms associated with shear effect. where I is the second of area, ` the length, As the
They have the following form: reduced cross-sectional area of the grillage member, E
the modulus of elasticity, G the shear modulus and l
  the shape factor. The angle g in Eq. (9) is the angle
1 c
c1 ˆ ; c2 ˆ 1 1 ; between member axis y and the global axis X as shown
…1 1 2c† 2
…11† in Fig. 1. The shear effects become particularly impor-
…1 2 c† lk2 tant in grillage systems with elements having deep
c3 ˆ ; cˆ
…1 1 2c† …GAs † webs.
M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873 867

Fig. 2. Types of grillage systems.

The torsional stiffness terms in Eqs. (8) and (9) are of the cross-section of the element. If the element has a
simple connection in torsion as well as in bending at one
GJ a sinh a` GJ…cosh a` 2 1† of its ends, warping can take place at that end and it becomes
T1 ˆ ; T2 ˆ ;
m m necessary to employ this new boundary condition and
re-derive the stiffness terms [19].
GJ…sinh a` 2 a` cosh a`†
T3 ˆ ;
…am† 4.1.1. Automatic data generation
GJ…sinh a` 2 a` cosh a`† The genetic algorithm generates populations during
T4 ˆ ; each cycle of the search procedure by making use of four
…am†
operators mentioned earlier. Each individual in the popula-
s tion has four strings each corresponding to one of the design
GJ
m ˆ 2cosh a` 2 a` sinh a` 2 2; a ˆ (12) variables. When decoded, the ®rst two give the sequence
EIw number of the universal beam sections from the standard set
which are to be adopted to the longitudinal and transverse
where J is the torsional and Iw the warping moment of inertia beams of the grillage system. Decoding of the last two
868 M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873

Table 1 member i, Mxi the bending moment about the x-axis and
Input data for grillage types Mwi the bi-moment at one of the ends of the member i.
Input data Type 1 Type 2 For each member only one stress constraint is considered.
This is achieved by comparing the values of normal
Number of joints NX p NY NX p NY stresses at both ends and then introducing the stress
Number of supports NX p NY 2 p NY
constraint into the design problem for the end with the
Number of members …NX 1 1† p NY1 …NX 1 1† p NY1
…NY 1 1† p NY …NY 2 1† p NX greater value. In universal beams, the sectional coordinate
Spacing in X-direction LX=…NY 1 1† LX=…NY 2 1† at the tip of either ¯ange is equal to 0.25BD where B is the
Spacing in Y-direction LY=…NX 1 1† LY=…NX 1 1† width of the ¯ange and D the depth of the beam. All the
cross-sectional properties are given in the standard section
table.
results in the number of beams to be provided in each direc-
tion in the structure, respectively. The spacing between the 4.2. Steps of optimum design algorithm
beams along each direction is then computed using these
values. The design algorithm developed for obtaining the opti-
It is apparent that in order to carry out the analysis of the mum beam sections and the spacing in both directions for a
grillage system, in addition to the cross-sectional properties given area and uniformly distributed load consists of the
the other geometrical properties are also required. For this following steps:
purpose two subroutines were developed, each of which
1. Initial population is constructed randomly.
depending on the type of the grillage system prepares the
2. In each individual, the binary codes for the design
input data for the analysis.
variables are decoded and the sequence number in the
In both types which are shown in Fig. 2, LX and LY are the
available universal beam section list and the number of
lengths to be spanned in the X- and Y-direction. NX is
beams in both directions are obtained.
the number of transverse beams in the X-direction and NY
3. The input data required for the analysis of the grillage
the number of longitudinal beams in the Y-direction. Using
system such as the total number of joints, members of
the values of NX and NY in each type, the total number of
joints, members and supports is generated automatically
joints, members, supports and spacing in each direction are
using the information in step 2.
computed as shown in Table 1.
4. The grillage system is analyzed and its response is
For the support conditions of the grillage system, two
determined.
different support types are considered. The supports can
5. Using the relationships (2)±(4), the value of the uncon-
be either all ®xed or all pinned. The other information
strained function P(x) is calculated for each individual.
such as member incidences, grouping and joint coordinates
The maximum and minimum values of this function are
are also computed using the values of NX and NY according
obtained for the population.
to the numbering order shown in Fig. 2(a) and (b). The
6. Using the relationship (5), the ®tness value for each
grillage system is assumed to be subjected to uniformly
individual is calculated. The average ®tness is
distributed load of W (kN/m 2). The vertical loads to be
found and the ®tness factor is calculated for each
considered at joints in both grillage types are calculated as
individual.
SX p SY p W where SX and SY are the spacings in both
7. Depending on their ®tness factor individuals are copied
directions. It should be added that this value is to be divided
into the mating pool.
by 2 for the joints on the ®rst and last longitudinal beams in
8. The individuals are coupled randomly and the reproduc-
grillage type II shown in Fig. 2.
tion operator is applied. Using two-point cross-sites two
off-springs are generated and the new population is
4.1.2. Computation of stresses
obtained. The value of 0.8 is used for the probability
When the analysis of the grillage system is carried out,
of crossover as suggested in Ref. [8].
the joint displacements and member end forces become
9. Mutation is applied to the new population with the prob-
available. The normal stresses in members are then
ability value of 0.005.
calculated at the cross-sections of both ends using the
10. The initial population is replaced by the new population
end forces. The maximum normal stress which develops
and steps 1±9 are repeated until the population is nearly
at the tip of either ¯ange of the universal beam section is
uniform.
obtained by combining the ¯exural and warping normal
stress At each generation, among the individuals which satisfy
Mxi M v all the design constraints the one with minimum weight
s imax ˆ 1 wi i …13† is stored and compared with the similar individual of
Zxi Iwi
the next generation. If the new one is heavier than the
where Zxi is the sectional modulus, v i the sectorial coordi- old one, then there is a loss of good genetic material.
nate, Iwi the warping constant of the section adopted for This situation is recti®ed by using elitist strategy which
M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873 869

guarantees the survival of the best individual in a 5. Design examples


generation [23]. This strategy is implemented by
replacing the worst individual in the next generation The algorithm developed is applied to the optimum
by the best individual from the previous individual. In spacing design of a grillage system which is to cover the
this way the loss of good individuals during the genera- area of 16 m £ 16 m as shown in Fig. 3. The modulus of
tions is prevented. elasticity and the shear modulus are taken to be 205 and

Fig. 3. Grillage system for 16 £ 16 m2 area.


870 M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873

81 kN/mm 2, respectively. The allowable de¯ection is taken the second it is included. Both optimum designs are
as span/360 which provides an upper bound of 44.4 mm. shown in Fig. 3(a) and (b). Comparison of both opti-
The allowable stress is taken as the yield stress of mild mum designs clearly shows the effect of warping causing a
steel of grade 43 which is 275 N/mm 2. The grillage system 7% increase in the weight of the grillage. It was noticed that
is subjected to the uniformly distributed load of 3.5 kN/m 2. in both designs stress constraints were dominant. Later the
Sixty-four universal beam sections starting from 838 £ supports were changed to pin supports and the system was
292 £ 176 UB down to 127 £ 76 £ 13 UB are selected once more designed twice. The optimum systems for both
from the complete set of 72 sections given in Ref. [18]. ignoring and including the effect of warping are shown in
These sections constitute the discrete set to be used for the Fig. 3(c) and (d). As expected, warping causes an increase
grillage elements. Fifteen discrete values are selected for the of 48.7% in the optimum weight of the grillage system.
number of beams in the X- and Y-directions. The number of In both optimum systems stress constraints were also
beams in either direction is assumed to vary from two dominant. It was noticed that in all four cases, the genetic
beams to 16 beams with an increment of one beam. algorithm reached the optimum designs within 70±80
These values are repeated in the discrete set to ®ll the generations.
column of 64 elements. The population size is taken as The area to be spanned in this example was square. In
20. The value of C in Eq. (3) is taken as 10. This value order to study the effect of the ratio of sides on the spacing
was determined after carrying out several designs with of beams in the grillage system, LX was changed to 14, 12,
different values of C such as 1, 10, 50 and 100. Among 10 and 8 m, while LY was kept 16 m. Each time the grillage
the values experimented 10 has produced grillage structures system was designed twice. In the ®rst the effect of warping
with the least weight. is ignored and in the second it is considered. The optimum
The grillage system with ®xed supports was designed designs obtained are given in Tables 2 and 3 for grillage
twice. In the ®rst the effect of warping is ignored and in structures with ®xed supports and in Tables 4 and 5 for

Table 2
Optimum designs for grillage structures with ®xed supports, warping ignored

Design number LX (m) LY (m) Optimum UB sections Number of beams s max (N/mm 2) Weight (kg)

X-Dir Y-Dir X-Dir Y-Dir

1 16 16 254 £ 146 £ 43 762 £ 167 £ 173 8 8 255 27 648


2 14 16 762 £ 267 £ 173 127 £ 76 £ 13 6 4 237 15 364
3 12 16 686 £ 254 £ 125 127 £ 76 £ 13 6 4 270 9846
4 10 16 610 £ 305 £ 238 305 £ 165 £ 54 2 3 217 7354
5 8 16 457 £ 191 £ 82 127 £ 76 £ 13 6 4 259 4768

Table 3
Optimum designs for grillage structures with ®xed supports, warping ignored

Design number LX (m) LY (m) Optimum UB sections Number of beams s max (N/mm 2) Weight (kg)

X-Dir Y-Dir X-Dir Y-Dir

1 16 16 762 £ 267 £ 173 610 £ 229 £ 101 6 8 268 29 562


2 14 16 762 £ 267 £ 147 610 £ 229 £ 113 6 4 275 19 572
3 12 16 762 £ 267 £ 134 457 £ 152 £ 67 6 4 257 13 942
4 10 16 838 £ 292 £ 176 457 £ 152 £ 52 2 6 255 8539
5 8 16 762 £ 267 £ 134 356 £ 127 £ 33 2 6 252 5320

Table 4
Optimum designs for grillage structures with pin supports, warping ignored

Design number LX (m) LY (m) Optimum UB sections Number of beams s max (N/mm 2) Weight (kg)

X-Dir Y-Dir X-Dir Y-Dir

1 16 16 762 £ 267 £ 134 152 £ 89 £ 16 13 14 268 31 435


2 14 16 762 £ 267 £ 197 127 £ 76 £ 13 4 13 268 14 141
3 12 16 762 £ 267 £ 134 127 £ 76 £ 13 5 11 266 10 322
4 10 16 838 £ 292 £ 176 152 £ 89 £ 16 2 9 252 5822
5 8 16 762 £ 267 £ 134 127 £ 76 £ 13 2 9 233 4014
M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873 871

Table 5
Optimum designs for grillage structures with pin supports, warping considered

Design number LX (m) LY (m) Optimum UB sections Number of beams s max (N/mm 2) Weight (kg)

X-Dir Y-Dir X-Dir Y-Dir

1 16 16 762 £ 267 £ 134 610 £ 229 £ 101 12 13 227 46 759


2 14 16 838 £ 292 £ 176 406 £ 140 £ 39 5 15 248 21 673
3 12 16 838 £ 292 £ 176 254 £ 102 £ 25 3 15 258 12 380
4 10 16 610 £ 305 £ 238 305 £ 165 £ 54 2 9 252 7089
5 8 16 762 £ 267 £ 134 178 £ 102 £ 19 2 13 221 6094

Fig. 4. Grillage system for 16 £ 8 m2 area.


872 M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873

grillage systems with pin supports. Comparison of optimum References


weights for the same areas to be covered reveals the
fact that effect of warping results in heavier grillage [1] Levy R, Lev OE. Recent developments in structural optimization.
structures. The increase in the weight varies between Journal of Structural Engineering, ASCE 1987;113(9):1939±62.
[2] Templeman AB, Yates DF. A linear programming approach to
7 and 27% for the systems with ®xed supports. The increase discrete optimum design of trusses. In: Eschenaver H, Olhoff N,
in the weight due to inclusion of warping even reaches to editors. Optimization methods in structural design, Mannheim: B.I.
51% in the grillage structures with pinned supports. In all Wissen-Schaftsverlag, 1983.
the optimum designs, the stress constraints were dominant. [3] Zho DM. An improved Templeman's algorithm for optimum design
Inspection of the results also reveals that when the topology of trusses with discrete member sizes. Engineering Optimization
1986;9:303±12.
is taken as a design variable, the genetic algorithm may
[4] Fluery C, Schmit LA. Dual methods and approximation concepts in
converge to a local optimum. In this case, the design structural synthesis. NASA CR-3226, 1980.
space becomes quite large and the search mechanism of [5] Ringertz UT. A branch and bound algorithm for topology optimiza-
the genetic algorithm may not be able to reach the global tion of truss structures. Engineering Optimization 1986;10:111±24.
optimum. [6] Goldberg DE. Genetic algorithm in search, optimization and machine
The optimum grillage structures for the area of 8 £ 16 m2 learning. Reading, MA: Addison-Wesley, 1989.
[7] Goldberg DE, Santani MP. Minimum weight design by genetic
with ®xed supports are shown in Fig. 4(a) and (b) where the algorithm. In: Engineering optimization via genetic algorithm,
effect of warping increases the optimum weight by 12%. Proceedings of Electr. Comp., 1986. p. 471±82.
Later, another design is carried out including the effect of [8] Adeli H, Cheng N-T. Augmented Lagrangian genetic algorithm for
warping for the same area by making use of the type II structural optimization. Journal of Aerospace Engineering, ASCE
grillage structure. The optimum system obtained for 1994;7(1):104±18.
[9] Adeli H, Cheng N-T. Integrated genetic algorithm for optimization of
this case is shown in Fig. 4(c). In this system, both
space structures. Journal of Aerospace Engineering, ASCE
de¯ection and normal stress constraints were dominant. 1993;6(4):315±28.
Naturally this structure is much heavier than the one [10] Adeli H, Cheng N-T. Concurrent genetic algorithm for optimization
shown in Fig. 4(b). of large structures. Journal of Aerospace Engineering, ASCE
1994;7(3):276±96.
[11] Rajan SD. Sizing, shape and topology design optimization of trusses
using genetic algorithm. Journal of Structural Engineering, ASCE
6. Conclusions 1995;121(10):1480±7.
[12] Wu S-J, Chow P-T. Integrated discrete and con®guration optimization
A genetic algorithm based optimum design approach is of trusses using genetic algorithms. Journal of Computers and
presented for grillage structures. The design from algorithm Structures 1995;55(4):695±702.
obtains the optimum universal beam sections the British [13] Walters GA, Smith DK. Evolutionary design algorithm for optional
layout of tree networks. Engineering Optimization 1995;24:261±
Standards for the longitudinal and transverse grillage 81.
beams and their optimum spacing in both directions. [14] Grierson DE, Pak WH. Optimal sizing, geometrical and topological
Equal spacing is considered along each direction due to design using a genetic algorithm. Journal of Structural Optimization
practical application. 1993;6:151±9.
It is shown that the genetic algorithm can successfully be [15] Shankar N, Hajela P. Heuristic-driven strategies for near-optimal
structural topology development. Structural Engineering Review
used in ®nding optimum topology in grillage structures.
1993;5(2):157±67.
However, due to the large size of the design space, the [16] Hajela P, Lee E. Genetic algorithms in truss topological optimization.
global optimum cannot be guaranteed. International Journal of Solids and Structures 1995;32(22):3341±57.
It was noticed that the selection of the initial population [17] Rajeev S, Krishnamoorthy CS. Discrete optimization of structures
plays an important role in reaching the ®nal solution. One using genetic algorithms. Journal of Structural Engineering, ASCE
way of checking the uniqueness of the ®nal solution is to 1992;118(5):1233±50.
[18] BS4, Structural Steel Sections, Part 1. Speci®cations for Hot Rolled
restart the genetic algorithm with different initial popula- Sections, British Standards Institution, London, UK, 1985.
tions using different seed values. Furthermore, it was also [19] Saka MP. Optimum design of steel grillage systems. In: Proceedings
observed that good individuals may be destroyed from one of the Third International Conference on Steel Structures, Singapore,
generation to another. This was avoided by using an elitist March 1987. p. 273±90.
strategy which reintroduces the good individual of the [20] Saka MP. Optimum design of steel grillage systems using genetic
algorithm. In: Proceedings of the Second International Conference
previous generation into the next one, if it is eliminated in
in Civil Engineering on Computer Applications, University of
that generation. Bahrain, Bahrain, Vol. 1, 1996. p. 285±95.
It was shown that warping plays a signi®cant role in the [21] Leite JPB, Topping BHV. Improved genetic operators for structural
optimum design of structures made out of thin walled steel engineering optimization. In: Topping BHV, editor. Developments in
sections and should not be ignored. In spite of the fact that neural networks and evolutionary computing for civil and structural
the grillage system was subjected to symmetrical uniformly engineering, Edinburgh, UK, 1995. p. 143±69.
[22] Camp C, Pezesk S, Cao G. Optimized design of two-dimensional
distributed load, the inclusion of the effect of warping structures using a genetic algorithm. Journal of Structural Engineer-
caused an increase of 40% in the overall weight of some ing, ASCE 1998;124(5):551±9.
of the grillage systems. [23] Chen T-Y, Chen C-J. Improvements of simple genetic algorithm in
M.P. Saka et al. / Advances in Engineering Software 31 (2000) 863±873 873

structural design. International Journal for Numerical Methods in [25] Suresh A, Mohammed A. An improved genetic algorithm for optimal
Engineering 1997;40:1323±34. design of large trusses. In: Topping BHV, editor. Advances in
[24] Hasancebi O, Erbatur F. Evaluation of crossover techniques in genetic engineering computational technology, Edinburgh, UK: CIVIL-
algorithm based optimum structural design. In: Topping BHV, editor. COMP Press, 1998. pp. 97±102.
Advances in engineering computational technology, Edinburgh, UK:
CIVIL-COMP Press, 1998. pp. 111±23.

You might also like