You are on page 1of 11

Design of Retaining Walls Using Big

Bang–Big Crunch Optimization


Charles V. Camp1 and Alper Akin2

Abstract: A procedure is developed for designing low-cost or low-weight cantilever reinforced concrete retaining walls, with base shear
keys, using big bang–big crunch (BB-BC) optimization. The objective of the optimization is to minimize the total cost or total weight per unit
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

length of the retaining structure subjected to constraints on the basis of stability, bending moment, and shear force capacities and the require-
ments of the American Concrete Institute (ACI 318-05). An iterative population-based heuristic search method, BB-BC optimization has a
numerically simple algorithm with relatively few control parameters as compared with other evolutionary methods. Low-cost and low-weight
designs for two retaining walls are presented. In addition, results are presented on the effects of surcharge load, backfill slope, and internal
friction angle of the retained soil on the values of low-cost and low-weight designs with and without a base shear key. DOI: 10.1061/(ASCE)
ST.1943-541X.0000461. © 2012 American Society of Civil Engineers.
CE Database subject headings: Optimization; Retaining structures; Structural design.
Author keywords: Optimization; Retaining walls; Big bang–big crunch optimization; Structural design.

Introduction tional efficiency of the method, as compared with GA and ant


colony optimization (ACO), and demonstrated its applicability
Big bang–big crunch (BB-BC) is an innovative computational heu- to solve structural engineering optimization problems.
ristic method to solve optimization problems. At its core, BB-BC One of the most common and used types of geotechnical retain-
optimization takes advantage of a relatively simple concept first ing structures is the reinforced concrete cantilever retaining wall. In
proposed by Galton (1907) that states the average or weighted aver- the analysis and design of retaining structures, the interaction be-
age of a group of estimates can be remarkably accurate. From this tween the soil and the structures poses many challenges to the de-
simple, yet powerful, Galtonian principle, Erol and Eksin (2006) signer; the structure must safely and reliably support the backfill
conceptualized an abstract model of the evolution of the universe soil, provide stability against the possibility of overturning and slid-
and applied it to develop the original BB-BC algorithm. In the “big ing, limit stresses in both the soil and the structure, and provide
bang” stage of their algorithm, a random set of solutions is gener- acceptable safety factors for all failure modes. In addition to these
ated within the search space; the “big crunch” stage effectively design objectives, there are many requirements that a reinforced
averages the solutions on the basis of their feasibility and quality concrete wall must satisfy: It must have sufficient shear and mo-
to produce a new center for the next big bang. Over a series of ment capacities in the stem, toe, heel, and base shear key sections
sequential big bang and big crunch cycles, the size and shape of of wall; the bearing capacity of the foundation cannot be exceeded
the random distribution of newly generated solutions grow smaller or allowed to be in tensile stress; and the configuration of the steel
and smaller about the moving average solution computed during reinforcement must meet all building code requirements.
the big crunch. Over multiple cycles, as some measure of the aver- Methods for developing low-cost and low-weight designs of
aged solution and/or the best solution ceases to improve, the opti- reinforced concrete retaining structures have been the subject of
mization is assumed to have converged. research for many years (Fang et al. 1980; Rhomberg and Street
Although the BB-BC algorithm is relatively new, it has been 1981; Alshawi et al. 1988; Keskar and Adidam 1989; Saribaş
shown to outperform many other evolutionary methods in structural and Erbatur 1996; Low et al. 2001; Chau and Albermani 2003;
optimization. Erol and Eksin (2006) demonstrated in their original Bhatti 2006; Babu and Basha 2008). However, the application
paper that BB-BC outperformed enhanced and classic genetic al- of heuristic and evolutionary methods to the design of retaining
gorithms (GAs) for many benchmark optimization functions. Camp structures is relatively new: Ceranic et al. (2001) and Yepes et al.
(2007) and Kaveh and Talatahari (2009, 2010) proposed hybrid (2008) applied simulated annealing (SA); Ahmadi-Nedushan and
forms of the BB-BC algorithm that greatly improved the computa- Varaee (2009) used particle swarm optimization (PSO); and Kaveh
and Abadi (2010) applied harmony search. Although the research
1
Professor, Dept. of Civil Engineering, Univ. of Memphis, 3815 Central into the design of retaining structures using evolutionary methods is
Ave., Memphis, TN 38152 (corresponding author). E-mail: cvcamp@ limited, there are numerous studies on their application to rein-
memphis.edu forced concrete structures. Coello Coello et al. (1997), Rafiqa
2
Research Associate, Dept. of Engineering Sciences, Middle East Tech- and Southcombea (1998), Rajeev and Krishnamoorthy (1998),
nical Univ., Ankara, Turkey. Camp et al. (2003), Lee and Ahn (2003), Lepš and Šejnoha
Note. This manuscript was submitted on October 19, 2010; approved on
(2003), Sahab et al. (2004), Govindaraj and Ramasamy (2005),
June 15, 2011; published online on June 17, 2011. Discussion period open
until August 1, 2012; separate discussions must be submitted for individual and Kwak and Kim (2008, 2009) all applied various forms of
papers. This paper is part of the Journal of Structural Engineering, GAs to the cost-optimization problem. Paya et al. (2008), Perea
Vol. 138, No. 3, March 1, 2012. ©ASCE, ISSN 0733-9445/2012/3- et al. (2008), and Paya-Zaforteza et al. (2009) optimized reinforced
438–448/$25.00. concrete structures using simple and hybrid SA algorithms.

438 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.


Design of Retaining Walls in which β = backfill slope; and θ = friction angle of the backfill
soil. The passive earth pressure coefficient kp is
Fig. 1 shows the general forces acting on the retaining wall: W W is  
the combined weight of all the sections of the reinforced concrete θ
k p ¼ tan2 45 þ ð3Þ
wall; W S is the weight of backfill acting on the heel; W T is the 2
weight of soil on the toe; Q is the surcharge load; PA is force re-
sulting from the active earth pressure; PK and PT are the forces For the sliding mode of failure, only the horizontal component
resulting from passive earth pressure on the base shear key and of the active force is considered. Horizontal resisting forces result
front part of the toe section, respectively; and PB is the force re- from the weight of wall and soil on the base, surcharge load, fric-
sulting from the bearing stress of the base soil. In general, the forces tion between soil and base of wall, and passive force owing to soil
acting on this model of a retaining structure are consistent with on the toe and base shear key sections. The factor of safety against
those presented by Saribaş and Erbatur (1996), Ceranic et al. sliding FSS is defined as
(2001), Bhatti (2006), and Yepes et al. (2008). However, this for- P
FR
mulation includes both passive forces on the front of the toe and FSS ¼ P ð4Þ
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

base shear key sections and the bearing force of the base soil. FD
Typically, three failure modes are considered in the analysis of P
in which
P F R = sum of the horizontal resisting forces; and
the retaining structure: overturning, sliding, and bearing stress. The F D = sum of the horizontal sliding forces.
overturning moment about the toe of the wall is a balance of the In this formulation, horizontal resisting forces resulting from the
force caused by the active soil pressure of the retained soil weight total vertical weight of wall, the friction of the base soil, and passive
and the self-weight of the concrete structure, the soil above the earth pressure of soil on front of wall are defined as
base, and the surcharge load. The passive forces on the front of X   
the toe and the base shear key section are not considered in the X 2ϕbase 2Bcbase
overturning moment. The factor of safety for overturning FSO FR ¼ W wall tan þ þ Pp ð5Þ
3 3
about the toe is defined as P
P in which W wall = total weight of the wall; ϕbase = internal friction
MR
FSO ¼ P ð1Þ of the base soil; B = total width of base slab; cbase = adhesion be-
MO tween the soil and the base slab; and
P
in which M PR = sum of the moments about the toe resisting over- 1 qffiffiffiffiffi
turning; and M O = sum of the moments about the toe tending to Pp ¼ γbase D21 k p þ 2cbase D1 kp ð6Þ
2
overturn the structure. The active earth pressure is computed by the
Rankine theory of active and passive earth pressure (Das 1994). in which γbase = unit weight of the base soil; and D1 = total depth of
The active earth pressure coefficient k a is the passive earth pressure block.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The horizontal component of the active force Pa that will tend to
cos β  cos2 β  cos2 θ cause the wall to slide is
k a ¼ cos β pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2Þ
cos β þ cos2 β  cos2 θ X
F D ¼ Pa cos β ð7Þ

In the bearing analysis of the structure, the base of retaining wall


is considered to be a shallow foundation. The minimum and maxi-
mum applied bearing stresses on the base of the foundation are
P  
V 6e
q min ¼ 1∓ ð8Þ
B B
max

in which qmin and qmax = bearing


P stresses on the toe and heel sec-
tions; B = width of the base; V = sum of the vertical forces (re-
sulting from the weight of wall, the soil above the base, and
surcharge load); and e = eccentricity of the resultant force system
expressed as
P P
B MR  MO
e¼  P ð9Þ
2 V

The eccentricity is determined from the ratio of the summation


of overturning moments about the toe to the sum of vertical forces.
The factor of safety for the bearing capacity FSB is
qu
FSB ¼ ð10Þ
qmax
in which qu = ultimate bearing capacity of the base soil.

Design Variables
Fig. 2 shows the design variables for the reinforced concrete
Fig. 1. Forces acting on a cantilever retaining wall
retaining wall model. The design variables are divided into two

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012 / 439

J. Struct. Eng. 2012.138:438-448.


Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Design variables for reinforced concrete cantilever retaining wall

categories: those that prescribe the geometric dimensions of the Constraints


wall cross section and those that model the steel reinforcement.
In general, there are eight geometric design variables representing The typical design philosophy for retaining structures seeks designs
the dimensions of the retaining wall: X 1 is the width of the base; X 2 that provide safety and stability against failure modes and comply
is the toe projection; X 3 is the thickness at the bottom of the stem; with concrete building code requirements. These requirements may
X 4 is the thickness at the top of the stem; X 5 is the thickness of base be classified into four general groups of design constraints: stabil-
slab; X 6 is the distance from toe to the front of the base shear key; ity, capacity, reinforcement configuration, and geometric limita-
X 7 is the width of the key; and X 8 is the depth of the key. There are tions. Each of the design constraints is posed as a penalty on
four additional design variables related to the steel reinforcement of the overall objective function of the design and is nonzero only
the various sections of the retaining wall: R1 is the vertical steel when violated. In other words, if the design is feasible, the sum
reinforcement in the stem; R2 is the horizontal steel reinforcement of the constraint penalties will be zero.
in the toe; R3 is the horizontal steel reinforcement in the heel; and
R4 is the vertical steel reinforcement in the base shear key. Table 1. Reinforcement Design Variable Pool
Whereas the geometric design variables may be either continu- Reinforcement
ous or discrete values, the steel reinforcement design variables are
modeled as a set of discrete values. In this formulation, in which the Index number (η) Bars (#) Bar size As (cm2 )
retaining structure is designed for a unit length, a discrete set of all 1 3 10 2.356
possible reinforcement configurations that meet geometric and 2 4 10 3.141
minimum steel constraints are predefined in a reinforcement data- 3 3 12 3.392
base. Table 1 shows the reinforcing steel combinations considered 4 5 10 3.926
in these designs, sorted in terms of cross-sectional area and refer- 5 4 12 4.523
enced by an index number η. A total of 223 reinforcement combi- ..
.
..
.
..
.
..
.
nations were used to represent 3–28 evenly spaced bars, sized
221 16 30 113.097
10–30 mm. The number of bars in a unit meter length of the re-
222 17 30 120.165
taining wall conforms to the minimum and maximum amount of
223 18 30 127.234
steel allowed [American Concrete Institute (ACI) 2005].

440 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.


Feasible retaining wall designs should provide minimum factor in which As;min = minimum area of steel reinforcement required by
of safety coefficients for overturning, sliding, and bearing capacity the code; and As = design value of area of steel reinforcement is
failure modes. These stability constraints g½13 are defined as section st. The minimum reinforcement ratio ρmin is defined as
FSO design 1:4
g1 ¼ 1≥0 ð11Þ ρmin ¼ ½ACI 318-89ðACI 1989Þ
FSO fy
pffiffiffiffi ð20Þ
FSS design f c 1:4
g2 ¼ 1≥0 ð12Þ ρmin ¼ 0:25 ≥ ½ACI 318-05ðACI 2005Þ
FSS fy fy

FSB design
g3 ¼ 1≥0 ð13Þ Maximum reinforcement area constraints g½1720 are defined as
FSB
 
As
in which FSO design , FSS design , and FSB design = prescribed safety fac- g½1720 ¼ 1≥0 ð21Þ
As;max st
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

tors for overturning, sliding, and bearing capacity, respectively. In


general, for any constraint gi < 0, then gi ¼ 0. in which As;max = maximum area of steel reinforcement required by
For practical and functional purposes, tensile stresses in the the code. The maximum reinforcement ratio ρmax is defined as
bearing capacity of the foundation soil are not allowed. The com-  
pressive bearing stress constraint g4 is defined as f 600
ρmax ¼ 0:85β 1 c ð22Þ
qmin f y 600 þ f y
g4 ¼ ≥0 ð14Þ
100
Additional geometric constraints are imposed by limiting the
in which 100 is used to scale the stress constraint (Saribaş and members of the design variable pool for the steel reinforcement
Erbatur 1996). (Table 1) to combinations that satisfy requirements for the number
The moment capacity of any reinforced concrete wall section of rebar and per unit length, rebar size, and bar spacing.
(stem, toe, heel, or key) should be greater than the design moment There are several additional geometric constraints g½2122 that
of the structure. In the same way, shear capacities of wall sections are applied to combinations design variables to prevent infeasible
should be greater than the design shear forces. retaining wall dimensions
The flexural strength M n is calculated as
X2 þ X3
  g21 ¼ 1≥0 ð23Þ
a X1
M n ¼ ϕAs f y d  ð15Þ
2
X6 þ X7
g22 ¼ 1≥0 ð24Þ
in which ϕ is the nominal strength coefficient (ϕ ¼ 0:9 for both X1
ACI 1989 and ACI 2005); As = cross-sectional area of steel
reinforcement; f y = yield strength of steel; d = distance from com- All retaining wall sections must satisfy minimum requirements
pression surface to the centroid of tension steel; and a = depth of for the development length of steel reinforcement bars within the
stress block. Moment capacity constraints g½58 may be summa- dimensions of the structure. The development length constraints are
rized as applied in two steps. First, the minimum basic development length
  ldb is checked against the allowable space in the appropriate wall
Md section (accounting for rebar size and concrete cover). If the allow-
g½58 ¼ 1≥0 ð16Þ able space is insufficient, a hook is used to achieve the additional
M n st
development length. If hooked reinforcement is used, bars should
in which M d = design moment in section st of the retaining wall (in satisfy the minimum hook development length ldh and minimum
which st refers to the stem, toe, heel, and key of the retaining wall). hook length of 12d bh (in which d bh is the diameter of the hooked
The shear strength V n is calculated as bar). All appropriate wall dimensions should accommodate the re-
pffiffiffiffi quired development lengths for the reinforcement.
V n ¼ ϕ0:17 f c bd ð17Þ In ACI 318-89 (ACI 1989), the development length ld is given as
 
in which ϕ = nominal strength coefficient (ϕ ¼ 0:85 for ACI 1989 f y πd 2b 0:36f y d b
and ϕ ¼ 0:75 for ACI 2005); f c = compression strength of con- ld ¼ 0:019 pffiffiffiffi ≥ pffiffiffiffi ð25Þ
fc 4 fc
crete; and b = width of the section. Shear capacity constraints
g½912 are summarized as in which d b = diameter of the reinforcing steel bar. In ACI 318-05
  (ACI 2005), the development length is given as
Vd 8
g½912 ¼ 1≥0 ð18Þ
> 12f y ψt ψe λ
< 25 pffiffiffi
V n st > ≥ 300 mm for d b < 19 mm
ld fc
in which V d = design shear strength in section st. All shear forces ¼ 12f ψ ψ λ ð26Þ
db >
: pffiffiffi ≥ 300 mm for d b ≥ 19 mm
> y t e
and bending moments are calculated at the face of the support. 20 fc
The amount of steel reinforcement in each section of the retain-
ing structure must satisfy minimum and maximum reinforcement in which ψt is the traditional reinforcement location factor;
area limits required by building codes (ACI 1989, 2005). Minimum ψe = coating factor reflecting the effects of epoxy coating; and λ
reinforcement area constraints g½1316 are defined as = factor reflecting the lower tensile strength of lightweight concrete
  (in this study, all of these factors are 1.0).
As;min
g½1316 ¼ 1≥0 ð19Þ The reinforcement development length constraints g½2326 are
As st summarized for each section of the retaining structure.

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012 / 441

J. Struct. Eng. 2012.138:438-448.


For stem Big Bang–Big Crunch Optimization
ldb stem ldh stem Erol and Eksin (2006) originally proposed the BB-BC method as a
g23 ¼ 1≥0 or g23 ¼ 1≥0
X 5  cover X 5  cover two-stage procedure: an initial big bang stage in which candidate
ð27Þ solutions are randomly distributed within the search space and a big
crunch stage, in which the center of the next big bang is located by
For toe considering the weighted average of the candidate solutions. The
initial big bang stage is very similar to other evolutionary methods,
ldb toe 12d b toe
g24 ¼  1 ≥ 0 or g24 ¼ 1≥0 in that an initial population of candidate solutions is generated ran-
X 1  X 2  cover X 5  cover domly over the search space. Erol and Eksin (2006) modeled the
ð28Þ random nature of the big bang stage as the dissipation of energy or
the transformation from an ordered state to a disordered or chaotic
For heel state. Next, a contraction operation is applied during the big crunch
stage, which computes the center of mass for the next big bang
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

ldb heel 12d b heel


g25 ¼  1 ≥ 0 or g25 ¼ 1≥0 from the penalized objective function values of the candidate sol-
X 2 þ X 3  cover X 5  cover ution population as
ð29Þ PNC ~xi
i¼1 F i
For key xcm ¼ PNC
~ 1
ð35Þ
i¼1 F i
ldb key ldh key in which ~ xcm = position of the center of mass; ~
xi = position of can-
g26 ¼ 1≥0 or g26 ¼ 1≥0
X 5  cover X 5  cover didate i in an n-dimensional search space; F i = penalized objective
ð30Þ function value of candidate i; and NC = candidate population size.
Positions of candidate solutions for the next iteration of the big
If both the basic and hook forms of steel reinforcement satisfy bang are normally distributed around the center of mass ~ xcm by
development length conditions, then the smaller amount of steel
reinforcement is used in the wall design. xinew ¼ ~
~ xcm þ ~
σ ð36Þ

Objective Function in which ~


xinew = position of the new candidate solution i; and ~
σ=
standard deviations of a normal distribution computed as
The forms of the two objective functions for this optimization are
consistent with those presented by Saribaş and Erbatur (1996). The xmax ~
rαð~ xmin Þ
~
σ¼ ð37Þ
cost-minimization objective function includes the cost of concrete s
and reinforcing steel (both include the cost of the material per unit
in which r = random numbers from a standard normal distribution;
volume and costs associated with labor and installation). The cost
α = parameter limiting the size of the search space; ~ xmax and
function f cost is
~
xmin = upper and lower limits on the values of the design variables;
f cost ¼ C s W st þ Cc V c ð31Þ and s = number of big bang iteration. The range of values available
to the new candidate population~xnew
i , and thus the size of the search
in which Cs = unit cost of steel; C c = unit cost of concrete; space, decreases inversely with each succeeding big bang iteration.
W st = weight of steel per unit length of the wall; and V c = volume For optimization with discrete variables, the continuous values
of concrete per unit length of the wall. ~
xnew
i are rounded to the nearest integer value:
The second objective function relies solely on the weight of the  
materials. The weight function f weight is rαð~xmax ~ xmin Þ
~
xi ¼ ROUND ~
new
xcm þ ð38Þ
s
f weight ¼ W st þ 100V c γc ð32Þ
Values within ~ xnew
i may fall outside the limits of the search
in which γc = unit weight of concrete; and a factor of 100 is used for space. These unprescribed values are reset to the appropriate mini-
consistency of units (Saribaş and Erbatur 1996). mum/maximum value and lead to the accumulation of candidate
A penalty function is used to enforce the constraints gi on the solutions at the search space boundaries, especially in during initial
objective function. The total objective function penalty Φi for a can- iterations of the algorithm (Erol and Eksin 2006).
didate low-cost or low-weight design i is a function of the summa- A hybrid BB-BC algorithm developed by Camp (2007) employs
tion of the stability, capacity, reinforcement configuration, and an alternative method for generating new candidate populations. In
geometric constraints defined as this algorithm, the positions of candidate solutions at the beginning
 X26 ε of each big bang are normally distributed around a new point lo-
Φi ¼ 1 þ gk ð33Þ cated between the center of mass ~ xcm and the best global solution
k¼1 ~
xbest as

in which ε = positive penalty exponent (typically > 1). The penal- xmax ~
rαð~ xmin Þ
~
xinew ¼ β~
xcm þ ð1  βÞ~
xbest þ ð39Þ
ized objective function F i is a product of either the cost or the s
weight objective function of candidate design i and its total penalty
in which β is a parameter controlling the influence of the ~
xbest on
F i ¼ Φi f i ð34Þ the location of new candidate solutions. For truss structures,
Camp (2007) and Kaveh and Talatahari (2009, 2010) have shown
The penalty function imposes a numerical penalty on the value that there is a significant improvement in the quality of solution
of the objective function that tends to reflect the degree at which the and the computational efficiency of BB-BC algorithms using
constraints are violated by a candidate set of design variables. formulations similar to Eq. (39) over the original model developed

442 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.


1. Camp (2007) showed that relativity high values of α and rela-
tively low values of β tended to provide the best solutions. The
hybrid formulation given in Eq. (39) reduces to the formulation
developed by Erol and Eksin (2006) when β is set to 1.
As the value of the penalty function exponent ε increases, the
penalty for a given candidate design increases. In Phase 1 of the
BB-BC algorithm, if ε > 2, the search tends to be more exploitive
and less explorative, generating solutions that, although feasible,
are too costly or heavy to be considered good designs. However,
Fig. 3. Big bang–big crunch algorithm in pseudocode
during Phase 2, in a reduced search space, a larger penalty should
be applied to control the unfavorable tendency of convergence to
by Erol and Eksin (2006). In some sense, the weighted average of light, but slightly infeasible designs. For all the retaining wall de-
xbest and ~
~ xcm , controlled by β, may be viewed as a method for bal- sign examples, ε ¼ 2 in Phase 1 and ε ¼ 4 in Phase 2.
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

ancing the characteristics of exploration of the search space and


exploitation of cooperative information within each new candidate
population. In another view, the effects of including~ xbest in the gen- Design Examples
eration of a new candidate population is a form of an elitist strategy,
All the retaining wall design examples were originally developed
wherein the best solution is allowed to influence the direction of the
by Saribaş and Erbatur (1996). The objectives in presenting Exam-
search over many iterations and is analogous to techniques used
ple 1 are threefold: (1) to compare the results of the BB-BC algo-
in PSO.
rithm to those of Saribaş and Erbatur (1996); (2) to examine the
Fig. 3 shows a pseudocode summary of the BB-BC algorithm.
effects of applying a more modern version of the ACI requirements
The initial population is generated randomly within the entire
for structural concrete (ACI 2005); and (3) to quantify the effects of
search space using a uniform random number distribution. The
using a continuous variable or discrete variable formulation on low-
penalized objective function for each candidate solution, defined
cost and low-weight designs. In Example 2, although the material
by a set of design variables, is calculated as defined in Eq. (34).
properties and general dimensional requirements are consistent
The next set of candidate solutions is normally distributed about
with those presented by Saribaş and Erbatur (1996), the objective
the center of mass ~ xcm and the best global solution ~ xbest according
is to investigate the effects of including a base shear key.
to Eq. (39). New candidate populations are generated iteratively
going through sequences of big bangs and big crunches until
Example 1
the global best solution ~xbest has not changed for a number of con-
secutive iterations; with this condition reached, the BB-BC algo- In this set of retaining wall designs, a base shear key is not included
rithm is considered to have converged to a solution. At this in the design variables. For a specified set of design parameters, the
point, Phase 1 of the BB-BC search is complete, and Phase 2 is BB-BC retaining wall design procedure is applied twice, conform-
initiated in the region surrounding ~ xbest . In Phase 2, a local search ing to the ACI 318-89 (ACI 1989) and ACI 318-05 (ACI 2005)
space is defined around~ xbest from Phase 1 and the immediate neigh- requirements for reinforced concrete, respectively.
borhood of each design. In Phase 2, a new set of candidate solutions Table 2 shows the specified retaining wall, soil, and de-
~
x is randomly generated within the local search space, with ~ xbest for sign parameters (Saribaş and Erbatur 1996). Because no key is
Phase 1 being retained or reset. Convergence for each phase is de-
termined when the value of ~ xbest has not improved for a specified Table 2. Input Parameters for Example 1
number of analyses. Selection of~ xbest is limited to solutions that are Input parameter Unit Symbol Value
feasible, in other words, designs that have no penalty applied to
their objective function values. Height of stem m H 3.0
Yield strength of reinforcing steel MPa fy 400
Big Bang–Big Crunch Retaining Wall Design Compressive strength of concrete MPa fc 21
Parameters
Concrete cover cm dc 7
In this study, all the retaining wall design examples use the same Shrinkage and temperature — ρst 0.002
values for the BB-BC optimization parameters, such as the size of reinforcement percent
candidate solution population, values of α and β required for Surcharge load kPa q 20
Eq. (39), the penalty function exponent, the search space reduction Backfill slope ° β 10
factor used for a multiphase search, and the algorithm stopping
Internal friction angle of retained soil ° ϕ 36
criteria.
Internal friction angle of base soil ° ϕ0 0
Numerical results indicate that a population of 100 candidate
solutions is adequate to balance computational efficiency and over- Unit weight of retained soil kN=m3 γs 17.5
all algorithm performance. In addition, results from the BB-BC al- Unit weight of base soil kN=m3 γ0s 18.5
gorithm are compared with other evolutionary methods that use Unit weight of concrete kN=m3 γc 23.5
similar population sizes. A general stopping criterion of 2,000 Cohesion of base soil kPa c 125
analyses without any change in ~ xbest (overall best feasible design) Design load factor — LF 1.7
has been shown to be sufficient (Camp 2007). The size of the Phase Depth of soil in front of wall m D 0.5
2 search space reduction around ~ xbest varies with the size of the Cost of steel $=kg Cs 0.40
problem; however, approximately 10% of the original size has been Cost of concrete $=m3 Cc 40
shown to be sufficient to obtain improved designs while reducing Factor of safety for overturning stability — SFO design 1.5
overall computational time (Camp and Bichon 2004; Camp 2007).
Factor of safety against sliding — SFS design 1.5
Appropriate values for α and β required for Eq. (39) for retain-
Factor of safety for bearing capacity — SFB design 3.0
ing wall design are set from numerical results presented in Example

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012 / 443

J. Struct. Eng. 2012.138:438-448.


Table 3. Design Variables for Example 1 (Continuous Variables) Fig. 4 shows the effects of various values of α and β, as defined
Design variable Unit Lower limit Upper limit in Eq. (39), on the average cost of wall designs (calculated from
100 runs). Detailed computational results show that α ¼ 1 and
X1 M 1.3090 2.3333 β ¼ 0:2 routinely provide the best retaining wall designs. Using
X2 M 0.4363 0.7777 a value of α ¼ 1 enables the initial search to sample the full range
X3 M 0.2000 0.3333 of values for each design variable. The relatively small value
X4 M 0.2000 0.3333 β indicates that better designs are obtained when the best design
X5 M 0.2722 0.3333 of each generation is weighted more heavily than the center of
R1 — 1 223 mass of the entire candidate population. In other words, the center
R2 — 1 223 of the new population of normally distributed candidates is shifted
R3 — 1 223 more toward the local best design than the population center of
mass. These values for α and β are used for all the retaining wall
designs.
In a series of 100 runs, the BB-BC procedure using the cost
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

considered, there are five design variables associated with the


objective function and ACI 318-05 (ACI 2005) requirements
geometry of the retaining wall and three additional design variables
had an average cost of $71:32=m with an SD of $0:57=m. The
representing the steel reinforcement in the stem, heel, and toe sec- multiphase BB-BC design procedure required an average of
tions of the wall. 24,831 analyses (but as few as 9,500) to converge with an average
In the first design, continuous variables are used to model the of 11,668 analyses in Phase 1. On average, 47% of the computa-
dimensions of the wall (X 1  X 5 ), and discrete variables are used to tional effort is completed in Phase 1, and the average difference
represent the steel reinforcement (R1  R3 ). Table 3 shows the between the Phase 1 best solution and the overall best solution
upper and lower limits of the design variables. The upper and lower is 0.3%. When considering weight objective function and ACI
values of the dimensional design variables of the wall are identical 318-05 (ACI 2005) requirements, the average low-weight design
to those presented by Saribaş and Erbatur (1996); however, the steel is 2;616:21 kg=m with an SD of 5:89 kg=m. The multi-
reinforcement is modeled as discrete combinations, as shown in phase BB-BC design procedure required an average of 17,785
Table 1. analyses (but as few as 8,500) to converge with an average of
Table 4 list the designs developed by the BB-BC procedure. 8,582 analyses in Phase 1. On average, 48% of the computational
The best BB-BC designs, when considering the cost objective
function under both the ACI 318-89 (ACI 1989) and ACI 318-
05 (ACI 2005) codes, are approximately $70:89=m, which is ap-
proximately 14% less than the design presented by Saribaş and
Erbatur (1996). However, the best BB-BC designs, when consid-
ering the weight objective function under both the ACI 318-89
(ACI 1989) and ACI 318-05 (ACI 2005) codes, are approximately
2;608:4 kg=m, which is approximately 4% heavier than the design
presented by Saribaş and Erbatur (1996). There are several possible
reasons for these dissimilar results: (1) the steel reinforcement
combinations in the BB-BC procedure are discrete and probably
lead to discontinuous jumps in both the cost and weight objective
functions; and (2) the penalty constraint functions presented by
Saribaş and Erbatur (1996) are difficult to reconstruct and verify.
Acknowledging these differences, the general nature of BB-BC
designs compares well with those presented by Saribaş and
Erbatur (1996) using a penalty function approach and nonlinear
programming. Fig. 4. Effects of α and β on average low-cost design

Table 4. Designs for Example 1 (Continuous Variables)


American Concrete Institute 318-89 American Concrete Institute 318-05
Design variable Cost Weight Cost Weight
X1 1.7428 1.7450 1.7427 1.7450
X2 0.6197 0.6424 0.6017 0.6557
X3 0.2678 0.2000 0.2678 0.2000
X4 0.2000 0.2000 0.2000 0.2000
X5 0.2722 0.2722 0.2722 0.2722
R1 33 (15–10 mm) 77 (27–10 mm) 30 (14–10 mm) 77 (27–10 mm)
R2 17 (10–10 mm) 17 (10–10 mm) 17 (10–10 mm) 17 (10–10 mm)
R3 17 (10–10 mm) 17 (10–10 mm) 17 (10–10 mm) 17 (10–10 mm)
Steel (kg=m) 59.6182 82.1380 59.6173 82.1383
Concrete (m3 =m) 1.1761 1.0750 1.1760 1.0750
Design objective $70:89=m 2;608:37 kg=m $70:96=m 2;608:38 kg=m

444 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.


effort is completed in Phase 1, and the average difference between and discrete increments of the design variables. The size of
the Phase 1 best solution and the overall best solution is 0.1%. The the resulting search space is approximately 6:26ð1013 Þ possible
performance of the search in Phase 2 for retaining wall designs is designs.
consistent with applications of BB-BC to other structural optimi- Table 6 shows the designs developed by the BB-BC procedure.
zation problems (Camp 2007). The best BB-BC designs, when considering the cost objective
In general, there was little improvement between the Phase 1 function under both the ACI 318-89 (ACI 1989) and ACI
best solution and the overall best solution. Fig. 5 shows a typical 318-05 (ACI 2005) codes, are approximately $70:38=m, which
multiphase convergence history for both low-cost and low- is approximately 0.7% less than the design using continuous
weight designs. The spike in average cost and weight between design variables. The best BB-BC designs, when considering
50 to 75 iterations is the onset of Phase 2 and is the result of the weight objective function under both the ACI 318-89 (ACI
the best solution from Phase 1 being reset and not carried over into 1989) and ACI 318-05 (ACI 2005) codes, are approximately
the next phase. 2;601:3 kg=m, which is approximately 0.3% less than designs us-
In a second version of Example 1, discrete variables are used to
ing continuous variables. The difference in the performance when
model both the dimensions of the wall (X 1  X 5 ) and the steel
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

using discrete design variables is probably related to the smaller


reinforcement (R1  R3 ). Table 5 shows the upper and lower limits
search spaces associated with the discrete variable model. The
multiphase BB-BC design procedure required an average of
12,880 analyses (but as few as 7,900) to converge with an average
of 8,980 analyses in Phase 1. On average, 70% of the computa-
tional effort is completed in Phase 1, and there is no improvement
of the best solution from Phase 1 to Phase 2. In the smaller discrete
search space assumed for this design, a multiphase BB-BC is not
required.
As a measure of the sensitivity of the BB-BC retaining wall de-
sign procedure, a series of 100 designs are generated, and the aver-
age low-cost and low-weight designs are computed for a range of
typical surcharge loads, backfill slopes, and retained soils. Fig. 6
shows the average low-cost and low-weight designs for various val-
ues of the surcharge load. In this case, as surcharge load increases
from 0 to 50 kPa, the low-cost designs increase by 51% as com-
pared with an increase of only 28% for the low-weight designs,
Fig. 5. Typical multiphase convergence history for Example 1 indicating that the cost formulation is more sensitive to changes
in surcharge load than the weight model. Fig. 7 shows the average
low-cost and low-weight designs as the backfill slope increases
Table 5. Design Variables for Example 1 (Discrete Variables)
from 0 to 30°. The low-cost designs increase by 6.0% as compared
Design variable Unit Lower limit Upper limit Increment with an increase of only 4.8% for the low-weight designs. Fig. 8
X1 M 1.31 2.33 0.01 shows the average low-cost and low-weight designs as the internal
X2 M 0.44 0.78 0.01 angle of friction of the retained soil varies from 28 to 36°. Over this
X3 M 0.20 0.33 0.01 range, the low-cost designs decrease by 16% as the friction angle
X4 M 0.20 0.33 0.01
increases, whereas the low-weight designs decrease by 13%. In
X5 M 0.27 0.33 0.01
general, this analysis indicates that the cost formulation is more
sensitive to changes in surcharge load, backfill slope, and the in-
R1 — 1 223 1
ternal angle of friction of the retained soil than the weight model. In
R2 — 1 223 1
particular, the cost model is especially sensitive to changes in the
R3 — 1 223 1
surcharge load.

Table 6. Designs for Example 1 (Discrete Variables)


American Concrete Institute 318-89 American Concrete Institute 318-05
Design variable Cost Weight Cost Weight
X1 1.75 1.75 1.75 1.75
X2 0.70 0.71 0.68 0.70
X3 0.28 0.20 0.28 0.20
X4 0.20 0.20 0.20 0.20
X5 0.27 0.27 0.27 0.27
R1 30 (14–10 mm) 77 (27–10 mm) 30 (14–10 mm) 77 (27–10 mm)
R2 14 (9–10 mm) 14 (9–10 mm) 14 (9–10 mm) 14 (9–10 mm)
R3 14 (9–10 mm) 14 (9–10 mm) 14 (9–10 mm) 14 (9–10 mm)
Steel (kg=m) 56.7195 80.9275 56.7195 80.9275
Concrete (m3 =m) 1.1925 1.0725 1.1925 1.0725
Design objective $70:3878=m 2;601:3025 kg=m $70:3878=m 2;601:3025 kg=m

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012 / 445

J. Struct. Eng. 2012.138:438-448.


Table 7. Input Parameters for Example 2
Input parameter Unit Symbol Value
Height of stem m H 4.5
Yield strength of reinforcing steel MPa fy 400
Compressive strength of concrete MPa fc 21
Concrete cover cm dc 7
Shrinkage and temperature — ρst 0.002
reinforcement percent
Surcharge load kPa q 30
Backfill slope ° β 0
Internal friction angle of retained soil ° ϕ 28
Internal friction angle of base soil ° ϕ0 34
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

Unit weight of retained soil kN=m3 γs 18.5


Fig. 6. Effects of surcharge load on low-cost and low-weight designs Unit weight of base soil kN=m3 γ0s 17.0
Unit weight of concrete kN=m3 γc 23.5
Cohesion of base soil kPa c 0
Design load factor — LF 1.7
Depth of soil in front of wall m D 0.30
Cost of steel $=kg Cs 0.40
Cost of concrete $=m3 Cc 40
Factor of safety for overturning stability — SFO design 1.5
Factor of safety against sliding — SFS design 1.5
Factor of safety for bearing capacity — SFB design 1.5

Table 8. Design Variables for Example 2


Design variable Unit Lower limit Upper limit Increment
X1 m 1.96 5.50 0.01
X2 m 0.65 1.16 0.01
Fig. 7. Effects of backfill slope load on low-cost and low-weight X3 m 0.25 0.50 0.01
designs X4 m 0.25 0.50 0.01
X5 m 0.40 0.50 0.01
X6 m 1.96 5.50 0.01
X7 m 0.20 0.50 0.01
X8 m 0.20 0.50 0.01
R1 — 1 223 1
R2 — 1 223 1
R3 — 1 223 1
R4 — 1 223 1

long-term effects on the design of the retaining wall, the cohesion


of the base soil is assumed to zero.
In the first retaining wall design of Example 2, the structure
does not have a base shear key. In this model formulation, there
are five design variables associated with the geometry of the retain-
ing wall (X 1  X 5 ) and three additional design variables represent-
Fig. 8. Effects of internal friction angle of retained soil on low-cost and ing the steel reinforcement in the stem, heel, and toe sections of
low-weight designs the wall (R1  R3 ). The size of the resulting search space is
approximately 1:16ð1023 Þ possible designs. Table 9 shows the
designs developed by the BB-BC procedure. From more than
Example 2 100 designs, the best low-cost design is $233:81=m, with an
average cost of $246:23=m. The best low-weight design is
In this set of retaining wall designs, discrete values will be used for 7;821:55 kg=m, with an average of 8;087:75 kg=m.
both geometry and steel reinforcement variables, and the design In the second retaining wall design using Example 2 data, three
will satisfy the ACI 318-05 (ACI 2005) requirements for reinforced additional geometric variables (X 6  X 8 ) and one additional steel
concrete. Table 7 shows the specified retaining wall, soil, and de- reinforcement variable (R4 ) are considered to model the location,
sign parameters, and Table 8 shows the upper and lower limits of size, and steel reinforcement of a base shear key. Table 9 shows
the design variables (Saribaş and Erbatur 1996). To consider the designs developed by the BB-BC procedure. From more than

446 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.


Table 9. Designs for Example 2
Without key With key
Design variable Cost Weight Cost Weight
X1 4.40 4.31 3.97 3.76
X2 0.70 0.65 0.84 0.68
X3 0.49 0.41 0.49 0.41
X4 0.25 0.25 0.25 0.25
X5 0.40 0.40 0.40 0.40
X6 — — 2.90 3.22
X7 — — 0.20 0.20
X8 — — 0.48 0.49
R1 97 (23–12 mm) 124 (22–14 mm) 97 (23–12 mm) 124 (22–14 mm)
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

R2 33 (15–10 mm) 33 (15–10 mm) 33 (15–10 mm) 45 (18–10 mm)


R3 114 (27–12 mm) 124 (22–14 mm) 102 (24–12 mm) 116 (20–14 mm)
R4 — — 7 (6–10 mm) 7 (6–10 mm)
Steel (kg=m) 242.0282 280.4037 233.7902 270.9573
Concrete (m3 =m) 3.425 3.209 3.349 3.087
Design objective $233:8113=m 7;821:554 kg=m $227:4761=m 7;525:407 kg=m

Fig. 11. Effects of internal friction angle of retained soil on low-cost


Fig. 9. Effects of surcharge load on low-cost designs with and without designs with and without base shear key
base shear key

100 designs, the best low-cost design is $227:48=m, with an aver-


age cost of $250:66=m. The best low-weight design with a base
shear key is 7;525:41 kg=m, with an average of 7;999:81 kg=m.
When applying a design strategy that considers long-term
strength of the base soil (no base shear cohesion) and includes a
base shear key, the best design is approximately 2.7% less
expensive when using the cost as the objective function and
approximately 3.8% lighter when using the weight as the objective
function.
As in Example 1, a series of 100 retaining wall designs with and
without a base shear key are developed, and the average low-cost
design is computed for a range of typical surcharge loads, backfill
slopes, and retained soils. Fig. 9 shows that low-cost designs for
various values of the surcharge load (0–40 kPa) are not signifi-
cantly affected by using a base shear key. Fig. 10 shows average
low-cost designs as a function of the backfill slope (0–25°); low-
cost designs with a base shear key are on average approximately
4% lighter than designs without a key. Fig. 11 shows average low-
cost designs as a function of the internal angle of friction of the
Fig. 10. Effects of backfill slope load on low-cost designs with and retained soil (28–36°); low-cost designs with a base shear key
without base shear key
are on average approximately 3% lighter that design without a

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012 / 447

J. Struct. Eng. 2012.138:438-448.


key. In general, this analysis indicates that retaining structures with Coello Coello, C., Christiansen, A. D., and Santos, F. (1997). “A simple
base shear keys tend to be slightly more cost-effective than those genetic algorithm for the design of reinforced concrete beams.” Eng.
without keys for a wide range of surcharge loads, backfill slopes, Comput., 13(4), 185–196.
and retained soils. Das, B. M. (1994). Principles of geotechnical engineering, PWS
Publishing, Boston.
Erol, O. K., and Eksin, I. (2006). “A new optimization method: Big bang-
Summary big crunch.” Adv. Eng. Software, 37(2), 106–111.
Fang, H. Y., Atsuta, T., and Chen, W. F. (1980). “Optimum design of re-
A hybrid multiphase BB-BC algorithm, on the basis of concepts taining walls on sloping hillside.” Proc., 6th Southeast Asian Conf. on
proposed by Erol and Eksin (2006) and Camp (2007), is applied Soil Engineering, Southeast Asian Society of Soil Engineering
to the low-cost and low-weight design of reinforced concrete retain- (SEASSE), 391–406.
Galton, F. (1907). “Vox populi.” Nature, 75(1949), 450–451.
ing structures. Through a series of example design problems, the
Govindaraj, V., and Ramasamy, J. V. (2005). “Optimum detailed design of
hybrid BB-BC algorithm demonstrated that it was both computa-
reinforced concrete continuous beams using genetic algorithms.”
tionally efficient and capable of generating low-cost retaining wall Comput. Struct., 84(1–2), 34–48.
Downloaded from ascelibrary.org by University of Brighton on 07/09/14. Copyright ASCE. For personal use only; all rights reserved.

designs that satisfy safety, stability, and material constraints. In ad- Kaveh, A., and Abadi, A. S. M. (2010). “Harmony search based algorithm
dition, a sensitivity analysis indicated that cost formulation is more for the optimum cost design of reinforced concrete cantilever retaining
sensitive to changes in surcharge load, backfill slope, and the in- walls.” Int. J. Civ. Eng., 9(1), 1–8.
ternal angle of friction of the retained soil than the weight model. In Kaveh, A., and Talatahari, S. (2009). “Size optimization of space trusses
addition, there is a slight decrease in cost associated with retaining using Big bang-big crunch algorithm.” Comput. Struct., 87(17–18),
structures that include a base shear key. 1129–1140.
In general, the hybrid BB-BC algorithm has several important Kaveh, A., and Talatahari, S. (2010). “A discrete big bang-big crunch al-
differences from other evolutionary methods. The two most signifi- gorithm for optimal design of skeletal structures.” Asian J. Civ. Eng.
cant differences are the fairly simple numerical structure of the (Build. Hous.), 11(1), 103–122.
algorithm and a relatively small of number of algorithmic param- Keskar, A. V., and Adidam, S. R. (1989). “Minimum cost design of a can-
eters. Without considering typical evolutionary algorithm parame- tilever retaining wall.” Indian Concr. J., 63(8), 401–405.
Kwak, H. G., and Kim, J. (2008). “Optimum design of reinforced concrete
ters (e.g., population size, convergence criteria, penalty function
plane frames based on predetermined section database.” Comput. Aided
structure), the performance of the hybrid BB-BC algorithm is con- Des., 40(3), 396–408.
trolled by just two parameters: the upper limit α on the search space Kwak, H. G., and Kim, J. (2009). “An integrated genetic algorithm com-
and the center of mass weighting factor β. If a multiphase appli- plemented with direct search for optimum design of RC frames.”
cation of the BB-BC algorithm is used, an additional search space Comput. Aided Des., 41(7), 490–500.
reduction parameter is required. Lee, C. L., and Ahn, J. (2003). “Flexural design of reinforced concrete
frames by genetic algorithm.” J. Struct. Eng., 129(6), 762–774.
Lepš, M., and Šejnoha, M. (2003). “New approach to optimization of re-
References inforced concrete beams.” Comput. Struct., 81(18–19), 1957–1966.
Low, B. K., Teh, C. I., and Tang, W. H. (2001). “Efficient reliability-based
Ahmadi-Nedushan, B., and Varaee, H. (2009). “Optimal design of rein- design using spreadsheet optimization.” Proc. 8th Int. Conf. on Struc-
forced concrete retaining walls using a swarm intelligence technique.” tural Safety and Reliability, ICOSSAR 2001, A.A. Balkema Publishers,
Proc., 1st Int. Conf. on Soft Computing Technology in Civil, Structural Rotterdam, The Netherlands.
and Environmental Engineering, Civil-Comp Press, Stirlingshire, U.K.
Paya, I., Yepes, V., González-Vidosa, F., and Hospitaler, A. (2008). “Multi-
Alshawi, F. A. N., Mohammed, A. I., and Farid, B. J. (1988). “Optimum de-
objective optimization of concrete building frames by simulated
sign of tied-back retaining walls.” Struct. Eng. (London), 66(6), 97–105.
annealing.” Comput. Aided Civ. Infrastruct. Eng., 23(8), 596–610.
American Concrete Institute. (1989). Building code requirements for struc-
Paya-Zaforteza, I., Yepes, V., Hospitaler, A., and González-Vidosa, F.
tural concrete and commentary (ACI 318-89), Detroit.
(2009). “CO2 -optimization of reinforced concrete frames by simulated
American Concrete Institute. (2005). Building code requirements for struc-
annealing.” Eng. Struct., 31(7), 1501–1508.
tural concrete and commentary (ACI 318-05), Detroit.
Babu, G. L. S., and Basha, B. M. (2008). “Optimal design of cantilever Perea, C., Alcala, J., Yepes, V., González-Vidosa, F., and Hospitaler, A.
retaining walls using target reliability approach.” Int. J. Geomech., 8 (2008). “Design of reinforced concrete bridge frames by heuristic opti-
(4), 240–252. mization.” Adv. Eng. Software, 39(8), 676–688.
Bhatti, M. A. (2006). “Retaining wall design optimization with MS excel Rafiqa, M. Y., and Southcombea, C. (1998). “Genetic algorithms in optimal
solver.” Proc., 17th Analysis and Computation Specialty Conf., ASCE, design and detailing of reinforced concrete biaxial columns supported
Reston, VA. by a declarative approach for capacity checking.” Comput. Struct.,
Camp, C. (2007). “Design of space trusses using big bang–big crunch op- 69(4), 443–457.
timization.” J. Struct. Eng., 133(7), 999–1007. Rajeev, S., and Krishnamoorthy, C. S. (1998). “Genetic algorithms-based
Camp, C. V., and Bichon, B. J. (2004). “Design of space trusses using ant methodology for design optimization of reinforced concrete frames.”
colony optimization.” J. Struct. Eng., 130(5), 741–751. Comput. Aided Civ. Infrastruct. Eng., 13(1), 63–74.
Camp, C. V., Pezeshk, S., and Hansson, H. (2003). “Flexural design of Rhomberg, E. J., and Street, W. M. (1981). “Optimal design of retaining
reinforced concrete frames using a genetic algorithm.” J. Struct. walls.” J. Struct. Div., 107(5), 992–1002.
Eng., 129(1), 105–115. Sahab, M. G., Ashour, A. F., and Toropov, V. V. (2004). “Cost optimisation
Ceranic, B., Fryer, C., and Baines, R. W. (2001). “An application of simu- of reinforced concrete flat slab buildings.” Eng. Struct., 27(3), 313–322.
lated annealing to the optimum design of reinforced concrete retaining Saribaş, A., and Erbatur, F. (1996). “Optimization and sensitivity of retain-
structures.” Comput. Struct., 79(17), 1569–1581. ing structures.” J. Geotech. Eng., 122(8), 649–656.
Chau, K. W., and Albermani, F. (2003). “Knowledge-based system on Yepes, V., Alcala, J., Perea, C., and González-Vidosa, F. (2008). “A para-
optimum design of liquid retaining structures with genetic algorithms.” metric study of optimum earth-retaining walls by simulated annealing.”
J. Struct. Eng., 129(10), 1312–1321. Eng. Struct., 30(3), 821–830.

448 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / MARCH 2012

J. Struct. Eng. 2012.138:438-448.

You might also like