Professional Documents
Culture Documents
discussions, stats, and author profiles for this publication at: http://www.researchgate.net/publication/269986640
CITATION READS
1 133
4 AUTHORS, INCLUDING:
SEE PROFILE
Water cycle, mine blast and improved mine blast algorithms for discrete
sizing optimization of truss structures
Ali Sadollah a, Hadi Eskandar b, Ardeshir Bahreininejad c,⇑, Joong Hoon Kim a
a
School of Civil, Environmental and Architectural Engineering, Korea University, 136-713 Seoul, South Korea
b
Faculty of Engineering, Semnan University, Semnan, Iran
c
Faculty of Engineering, Institut Teknologi Brunei, Bandar Seri Begawan, Brunei
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents the applications of the mine blast algorithm (MBA) and the water cycle algorithm
Received 18 August 2013 (WCA), in addition to an improved version of MBA for weight minimization of truss structures including
Accepted 1 December 2014 discrete sizing variables. The MBA mimics the explosion of landmines, while the WCA is inspired by the
observation of water cycle process. An improved version of MBA (IMBA), is also presented. The efficiency
of the three optimization algorithms is tested using classical benchmark discrete truss design problems.
Keywords: Optimization results show that MBA, IMBA, and WCA offer a good degree of competitiveness against
Mine blast algorithm
other state-of-the-art metaheuristic techniques.
Water cycle algorithm
Truss structures
Ó 2014 Elsevier Ltd. All rights reserved.
Discrete variables
Sizing optimization
1. Introduction (HPSO) for truss structures which outperformed hybrid PSO with
passive congregation (PSOPC) [13] and standard PSO.
Over the last decades, various algorithms have been used for The PSOPC was also combined with ant colony optimization
truss optimization problems which are very popular in the field (ACO) and HS by Kaveh and Talatahari [14] to form an efficient
of structural optimization. Metaheuristic methods such as genetic algorithm for truss optimization, called discrete heuristic particle
algorithms (GAs), harmony search (HS), and particle swarm opti- swarm ant colony optimization (DHPSACO). Comprehensive
mization (PSO) can efficiently be used in truss design optimization reviews for applications of metaheuristic algorithms on skeletal
problems including discrete variables. GAs [1] mimic the processes structures have been presented in the literature [15,16].
of natural selection leading to the survival of the fittest. Sadollah et al. [17] recently developed the mine blast algorithm
For instance, Goldberg and Samtani [2] and Rajeev and Krishna- (MBA) which mimics the explosion of landmines. The MBA was
moorthy [3] performed sizing optimization of truss structures. successfully applied to discrete sizing optimization of truss struc-
Krishnamoorthy et al. [4] used GAs to optimize space truss struc- tures [17]. Furthermore, Eskandar et al. [18] proposed another
tures in the context of an object-oriented framework. Sivakumar metaheuristic algorithm, reproducing the water cycle process.
et al. [5] optimized steel lattice towers. Gero et al. [6] used GAs The water cycle algorithm (WCA) was tested in mathematical
for design optimization of 3D steel structures. and engineering problems [18]. The MBA and WCA algorithms
Geem et al. [7] developed the HS that reproduces the musical were found to be superior over other optimization methods in
process of searching for a perfect state of harmony. The harmony terms of convergence rate and quality of optimized designs
in music is analogous to the optimum design, and the musicians’ [17,18].
improvisation is analogous to local/global search schemes [8]. In this study, MBA is improved and its operators are enhanced
The HS was successfully applied to truss optimization problems in terms of efficiency so called improved MBA (IMBA). The relative
using discrete and continuous variables [9,10]. performance of the MBA, IMBA and WCA algorithms in discrete
The PSO is a population-based algorithm developed by Kennedy optimization problems of truss structures are investigated in this
and Eberhart [11]. Li et al. [12] developed an efficient heuristic PSO research. Furthermore, the efficiency of three algorithms is com-
pared with the results extracted in the literature.
The paper is organized as follows: the formulation of the dis-
⇑ Corresponding author. Tel.: +673 897 5723. crete optimization problem is presented in Section 2. The IMBA
E-mail address: bahreininejad@gmail.com (A. Bahreininejad). and WCA algorithms and their constraint handling strategies are
http://dx.doi.org/10.1016/j.compstruc.2014.12.003
0045-7949/Ó 2014 Elsevier Ltd. All rights reserved.
2 A. Sadollah et al. / Computers and Structures 149 (2015) 1–16
described in detail in Section 3. Section 4 discusses the optimiza- generated from a first shot explosion that produces a number of
tion results comparing the developed algorithms with the litera- individuals (shrapnel pieces). The size of initial population (Npop)
ture. Section 5 presents a sensitivity analysis on the effect of is taken as the number of shrapnel pieces (Ns). The MBA algorithm
algorithms internal parameters set by the user on the overall con- initially uses the lower and upper bounds of design variables and,
vergence behavior; the analysis is carried out for some of the test then, randomly creates the first shot point as follows:
problems considered in this study. Finally, Section 6 summarizes
~
X 0 ¼ L~
B þ frandg fU~
B L~
Bg: ð4Þ
the findings of this study.
Vector quantities are denoted by over sign. Assume X is the cur-
2. Discrete structural optimization problems rent location of a landmine; that is,
~
X ¼ ½x1 ; x2 ; x3 ; . . . ; xm : ð5Þ
In discrete sizing optimization problems of truss structures, the
objective usually is to minimize the weight of the structure yet Design variables (x1, x2, . . . , xm) can take real values in continu-
satisfying nonlinear constraints on element stresses, nodal dis- ous optimization problems or they can be selected from a prede-
placements, critical loads, etc. The optimization problem can be fined set of discrete values. We assume that the first shot point
formulated as follows: (X0) is the best solution (XBest = X0). For performing any optimiza-
tion method, exploration and exploitation are considered as two
min WðXÞ; X ¼ ½x1 ; x2 ; x3 ; . . . ; xN ; ð1Þ
critical steps.
subject to: The difference between the exploration and exploitation phases
is how they influence the whole search process in finding the opti-
g j ðx1 ; x2 ; . . . ; xN Þ 6 0 j ¼ 1; 2; . . . ; k; ð2Þ mal solution. Similar to other metaheuristic algorithms, MBA algo-
rithm starts with the exploration phase, which is responsible for
xd 2 Sd ¼ fX1 ; X 2 ; . . . ; X p g; ð3Þ comprehensively exploring the search space.
The exploration factor (l) serves to explore different regions of
where W(X) is the cost function corresponding to the structural
design space. This parameter, used in the early iterations of MBA, is
weight; N and k are the number of design variables and inequality
compared with an iteration number index (t): exploration takes
constraint functions, respectively. Each design variable can be cho-
place if l is greater than t. The exploration phase of MBA is gov-
sen from a discrete set Sd (X1, X2, . . . , Xp) of P available cross-sections
erned by the following equations [17]:
according to production standards. n o
2
Xe ¼ ~
~ dt1 frandn g cos h t ¼ 1; 2; . . . ; l; ð6Þ
3. Applied metaheuristic algorithms
~
Xe ¼ ~
X Best þ ~
Xe t 6 l; ð7Þ
3.1. Improved mine blast algorithm
where the dt1 vector includes the shrapnel distance for exploded
MBA algorithm is inspired by the process of landmines explo- mines with respect to each coordinate direction. Fig. 1 demon-
sion; shrapnel pieces are thrown away and collide with other strates the concept and performance of Eq. (6) from a schematic
mines in the vicinity of the explosion area causing further explo- point of view.
sions. Consider a landmine field where the goal is to clear land- By taking the square of a normally distributed random number,
mines. To clear all the mines, the position of the most explosive better exploration is achieved at the beginning of the optimization
mine must be located. This position corresponds to the optimal process (see Eq. (6)). The value of l determines the intensity of the
design. exploration. For example, increasing l makes it possible to explore
Landmines of different sizes and explosive power are planted more remote regions of design space. The shrapnel angle of inci-
under the ground. Landmine explosions cause many pieces of dence, denoted by h in Eq. (6), is given by:
shrapnel to be propelled in the air. The casualties of each piece
h ¼ k D k ¼ 0; 1; 2; . . . ; Ns 1; ð8Þ
of shrapnel are evaluated using a cost function (fitness function)
and, then, related to the presence of other landmines with different where D = 360/Ns. The value of h ranges from 0 to 360; the resulting
explosive power [17]. value of cos(h) ranges between 1 and 1. The initial distance of each
Often times, pieces of shrapnel collide with other mines and piece of shrapnel is d0 = (UB–LB); thus, the best solution is in the
trigger more mine explosions. This behavior is helpful for finding range [LB, UB]. For example, the LB and UB of a four design variable
the most explosive landmine. MBA algorithm was developed to problem are [30–20 10–5] and [30 20 10 5], respectively. Then,
find the most explosive landmine (i.e., the landmine with the most the initial distance, d0 (dt1 when t = 1), is the vector of shrapnel
casualties). Table 1 lists nomenclature of MBA parameters. distances [60 40 20 10].
The MBA algorithm requires an initial population of individuals, Improved MBA (IMBA) modifies the exploitation phase in MBA
similar to several other metaheuristic methods. The population is and distance reduction of each shrapnel pieces. For the exploitation
Table 1
Nomenclature of MBA (IMBA) parameters.
~dt1
~
dt ¼ ðt=aÞ t ¼ 1; 2; 3; . . . ; Max Iteration; ð14Þ
e
Fig. 2. Schematic view of IMBA including exploration (dashed area) and exploitation (solid area) processes in two dimensional space.
in Section 3.1, the distance of shrapnel pieces decreases adaptively precipitation phenomena. An initial population of designs variables
according to Eq. (14). (streams) is randomly generated. The best individual (i.e., the best
The value of a should be chosen so that at the final iteration, the stream), classified in terms of having the minimum cost function, is
distance of shrapnel pieces is approximately zero. For instance, if chosen as the sea [18].
the maximum number of iterations is 350 and the initial distance Then, a number of good streams (i.e., cost function values close
is 1 (i.e., UB = 1 and LB = 0), setting a = 10,000 results in a final dis- to the current best record) are chosen as rivers, while all other
tance of 0.002, which is an acceptable value close enough to zero. streams are considered as streams flowing to rivers and sea. In
It is worth mentioning that being close enough to zero varies an N dimensional optimization problem, a stream is an array of
from one problem to another. Therefore, a is a user-defined param- 1 N. This array is defined as follows:
eter in IMBA algorithm. The following formula computes a sug-
gested value for a used in the IMBA (MBA):
A Stream Candidate ¼ ½x1 ; x2 ; x3 ; . . . ; xN ð16Þ
Table 2
Pseudo-code of IMBA.
of rivers (which is a user-defined parameter) and a single sea as Fig. 3. (a) Schematic description of the stream’s flow to a specific river and (b)
given in Eq. (19). The rest of the population (streams which flow schematic of the WCA optimization process.
to the rivers or may directly flow to the sea) is calculated using
Eq. (20) as follows:
where NSn is the number of streams which flow to the specific rivers
or sea. As it happens in nature, streams are created from raining and ~
X iþ1 ~i ~i ~i
Riv er ¼ X Riv er þ rand C X Sea X Riv er ; ð25Þ
join each other to generate new rivers. Some streams may even flow
directly to the sea. All rivers and streams end up in the sea that cor- where rand is an uniformly distributed random number between 0
responds to the current best record. and 1. Vector quantities are indicated with the over signed notation.
Let us assume that there are Npop streams of which Nsr 1 are Eq. (24) describes the movement of streams who directly flow to
selected as rivers and one is selected as the sea. Fig. 3a shows sea. If the solution given by a stream is better than its connecting
the schematic view of a stream flowing toward a specific river river, the positions of river and stream are exchanged (i.e., the
along their connecting line. stream becomes a river and the river becomes a stream). A similar
The distance X between the stream and the river may be ran- exchange can be done for a river and the sea.
domly updated as the following relation: The evaporation process operator also is introduced to avoid
X 2 ð0; C dÞ; C>1 ð22Þ premature convergence to local optima. Basically, evaporation
causes sea water to evaporate as rivers/streams flow to the sea.
where 1 < C < 2 and the best value for C may be chosen as 2; d is This leads to new precipitations. Therefore, we have to check if
the current distance between stream and river. The value of X in the river/stream is close enough to the sea to make the evaporation
6 A. Sadollah et al. / Computers and Structures 149 (2015) 1–16
process occur. For that purpose, the following criterion is utilized Step 8: Exchange positions of river with a stream which gives
[18]: the best solution.
! ! Step 9: Exchange positions of sea with a stream which gives the
ifkX Sea X iRiv er k < dmax or rand < 0:1 i ¼ 1; 2; 3; . . . ; Nsr 1 best solution.
Raining process by Eq: ð26Þ Step 10: The rivers flow to the sea which is the most downhill
end place using Eq. (25).
Step 11: Evaluate optimization constraints for created rivers
where dmax is a small number close to 0. After evaporation, the rain- according to Section 3.3.
ing process is applied and new streams in the different locations Step 12: Similar to Steps 8 and 9, if a river finds better solution
(similar to mutation in the GAs). Indeed, the evaporation operator than the sea, the position of river is exchanged with the sea.
is responsible for the exploration phase in the WCA. For specifying Step 13: Check the evaporation condition with the pseudo-code
the new locations of the newly formed streams, the following equa- reported above.
tion is used: Step 14: If the evaporation condition is satisfied, the raining pro-
cess will occur using Eqs. (26) and (28).
~
X new ~ ~ ~
Stream ¼ LB þ rand ðU B LBÞ ð26Þ Step 15: Reduce the user-defined parameter dmax with Eq. (27).
Step 16: If the stopping criterion is not satisfied return to Step 5.
where LB and UB are lower and upper bounds of design variables,
respectively. Similarly, the best newly formed stream is considered The development of the WCA optimization process is illustrated
as a river flowing to the sea. The rest of new streams are assumed to by Fig. 3b (which contains Fig. 3a) where circles, stars, and the dia-
flow to the rivers or may directly flow to the sea. mond correspond to streams, rivers, and sea, respectively. The
By setting a large value for dmax the number of searches is lim- white (empty) shapes denote the new positions taken by streams
ited, while small values leads to search near the sea. Therefore, and rivers.
dmax controls the search intensity near the sea. The value of dmax In the WCA algorithm, rivers (a number of best selected points
adaptively decreases as follows: except the best one (sea)) act as guidance points for guiding other
i
individuals in the population toward better positions (as shown in
iþ1 i dmax Fig. 3b) in addition to minimize or prevent searching in regions
dmax ¼ dmax ð27Þ
Max Iteration containing local optima (see Eq. (23)).
Rivers are not fixed points and move toward the sea (the best
As it happens in nature, amount and location of precipitations
solution). This procedure (moving streams to the rivers and, then,
are also affected by winds and convective currents. Further
moving rivers to the sea) leads to indirectly moving toward the
research must be carried out to understand if alternative defini-
best solution.
tions of the evaporation operator may improve convergence
behavior. 3.3. Constraint handling strategy
In order to enhance the convergence rate and performance of
the WCA for constrained problems, Eq. (28) is used for streams In the search space, created streams and rivers in the WCA algo-
only which directly flow to the sea. This equation is considered rithm and generated shrapnel pieces in the MBA algorithm, may
to encourage the generation of streams which directly flow to violate either the problem specific constraints or the limits of the
the sea in order to improve the exploration near sea in the feasible decision variables. In this study, a modified feasible-based mecha-
region for constrained problems as follows [18]: nism is applied to tackle the constrained problems based on the
~ ~ pffiffiffiffi following four rules [19]:
X new
stream ¼ X sea þ l randnð1; NÞ ð28Þ
where l is a coefficient which shows the range of searching region Rule 1: Any feasible solution is preferred to any infeasible
near the sea. Randn is a normally distributed random number. Set- solution.
ting large values of l increases the possibility to exit from feasible Rule 2: Infeasible solutions containing slight violation of the
region. On the other hand, setting very small values of l leads to constraints (from 0.01 in the first iteration to 0.001 in the last
limit the size of search domain near the sea. A suitable value for iteration) are considered as feasible solutions.
l is set to 0.1. Rule 3: Between two feasible solutions, the one having the bet-
pffiffiffiffi
In mathematical point of view, term l in Eq. (28) represents ter objective function value is preferred.
the standard deviation and, accordingly, l in Eq. (28) defines the Rule 4: Between two infeasible solutions, the one having the
concept of variance. Using these concepts and Eq. (28), generated smaller constraint violation is preferred.
streams (individuals) having variance l are distributed around
the sea (best obtained optimum point). Using the first and fourth rules, the search is oriented to the fea-
The steps of the WCA are summarized as follows: sible region rather than the infeasible region. Applying the third
rule guides the search to the feasible region with good solutions
Step 1: Set the initial parameters of the WCA: Nsr, dmax, Npop, and [19]. For most structural optimization problems, the global mini-
Max_Iteration. mum locates on or close to the boundary of the feasible design
Step 2: Generate a random initial population and form the initial space.
streams, rivers, and sea using Eqs. (17), (19) and (20), Rule 2 increases the probability that trial design approach con-
respectively. straint domain boundaries and reach the global optimum [14]. The
Step 3: Calculate the value (cost) of each stream using Eq. (18). same constraint handling strategy was utilized in this research for
Step 4: Determine the intensity of flow for rivers and sea using WCA, MBA, and IMBA algorithms.
Eq. (21).
Step 5: The streams flow to the rivers by Eq. (23). 4. Test problems and optimization results
Step 6: The streams flow directly to sea by Eq. (24).
Step 7: Evaluate optimization constraints for created streams The WCA, MBA, and IMBA algorithms were applied to four clas-
according to Section 3.3. sical weight minimization problems of truss structures with,
A. Sadollah et al. / Computers and Structures 149 (2015) 1–16 7
respectively, 52, 72, 200, and 582 elements, including discrete siz- Table 3
ing variables. Internal parameters of WCA were set as follows: pop- Discrete values available for cross-sectional areas from AISC norms.
ulation size (Ntotal), number of rivers (Nsr 1), and dmax were No. in2 mm2 No. in2 mm2
chosen as 50, 8, and 1e5, respectively. Initial parameters of 1 0.111 71.613 33 3.84 2477.414
MBA and IMBA were set as follows: 50, maximum number of iter- 2 0.141 90.968 34 3.87 2496.769
ation divided by five, and 50,000 for population size, exploration 3 0.196 126.451 35 3.88 2503.221
factor (l), and reduction factor (a), respectively. 4 0.25 161.29 36 4.18 2696.769
5 0.307 198.064 37 4.22 2722.575
The limit number of iterations was increased with the problem 6 0.391 252.258 38 4.49 2896.768
size. Structural analyses were performed with the finite element 7 0.442 285.161 39 4.59 2961.284
method. The number of design variables for the 52, 72, 200, and 8 0.563 363.225 40 4.8 3096.768
582-bar structures were, respectively, 12, 16, 96, and 32. The cor- 9 0.602 388.386 41 4.97 3206.445
10 0.766 494.193 42 5.12 3303.219
responding numbers of optimization constraints were 80, 198, 700,
11 0.785 506.451 43 5.74 3703.218
and 1573, respectively. Fifty independent optimization runs were 12 0.994 641.289 44 7.22 4658.055
carried out for each test problem in order to have statistically sig- 13 1 645.16 45 7.97 5141.925
nificant results. 14 1.228 792.256 46 8.53 5503.215
Except for the 200-bar and 582-bar tower truss problem, statis- 15 1.266 816.773 47 9.3 5999.988
16 1.457 939.998 48 10.85 6999.986
tical data for the MBA were extracted from Ref. [17]. In this paper,
17 1.563 1008.385 49 11.5 7419.34
all considered truss structure are investigated using IMBA and 18 1.62 1045.159 50 13.5 8709.66
WCA algorithms, while MBA is implemented for optimal solving 19 1.8 1161.288 51 13.9 8967.724
of 200-bar and 582-bar tower truss structures. The computations 20 1.99 1283.868 52 14.2 9161.272
21 2.13 1374.191 53 15.5 9999.98
entailed by this optimization study were performed in the MATLAB
22 2.38 1535.481 54 16 10322.56
23 2.62 1690.319 55 16.9 10903.204
24 2.63 1696.771 56 18.8 12129.008
25 2.88 1858.061 57 19.9 12838.684
26 2.93 1890.319 58 22 14193.52
27 3.09 1993.544 59 22.9 14774.164
28 3.13 2019.351 60 24.5 15806.42
29 3.38 2180.641 61 26.5 17096.74
30 3.47 2238.705 62 28 18064.48
31 3.55 2290.318 63 30 19354.8
32 3.63 2341.931 64 33.5 21612.86
The second test case deals with the weight minimization of the
Fig. 4. Schematic of the planar 52-bar truss. spatial 72-bar truss shown in Fig. 6. This benchmark problem was
8 A. Sadollah et al. / Computers and Structures 149 (2015) 1–16
Table 4
52-bar truss problem: comparison of WCA, MBA and IMBA optimization results with the literature.
Variables (mm2) SGA [20] HS [12] HPSO [12] DHPSACO [14] MBA [17] WCA IMBA
A1–A4 4658.055 4658.055 4658.055 4658.055 4658.055 4658.055 4658.055
A5–A10 1161.288 1161.288 1161.288 1161.288 1161.288 1161.288 1161.288
A11–A13 645.160 506.451 363.225 494.193 494.193 494.193 494.193
A14–A17 3303.219 3303.219 3303.219 3303.219 3303.219 3303.219 3303.219
A18–A23 1045.159 940.000 940.000 1008.385 940.000 940.000 940.000
A24–A26 494.193 494.193 494.193 285.161 494.193 494.193 494.193
A27–A30 2477.414 2290.318 2238.705 2290.318 2283.705 2283.705 2283.705
A31–A36 1045.159 1008.385 1008.385 1008.385 1008.385 1008.385 1008.385
A37–A39 285.161 2290.318 388.386 388.386 494.193 494.193 494.193
A40–A43 1696.771 1535.481 1283.868 1283.868 1283.868 1283.868 1283.868
A44–A49 1045.159 1045.159 1161.288 1161.288 1161.288 1161.288 1161.288
A50–A52 641.289 506.451 729.256 506.451 494.193 494.193 494.193
Weight (kg) 1970.142 1906.76 1905.495 1904.83 1902.605 1902.605 1902.605
Number of structural analyses N/Aa N/A 105,000 11,100 5450 7100 4750
a
Not available.
Table 5
Statistical optimization results obtained from WCA, MBA, and IMBA for the 52-bar truss problem.
Methods Best optimized weight (kg) Average optimized weight (kg) Worst optimized weight (kg) Standard deviation on weight (kg)
WCA 1902.605 1909.856 1912.646 7.09
MBA 1902.605 1906.076 1912.646 4.09
IMBA 1902.605 1903.076 1904.83 1.13
Fig. 5. Comparison of convergence curves recorded for the 52-bar truss problem.
previously studied by Lee et al. [10], Li et al. [12], Kaveh and Tala- Two variants were considered for this optimization problem: (i)
tahari [14], Sadollah et al. [17], and Wu and Chow [20]. The mate- Case 1, where values of cross-sectional areas could be selected
rial density and the modulus of elasticity of the truss structure are from the discrete set D = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9,
0.1 lb/in3 (2767.99 kg/m3) and E = 10 Msi (68.95 GPa), respectively. 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 2.1, 2.2, 2.3, 2.4,
The maximum stress developed in the elements must not 2.5, 2.6, 2.7, 2.8, 2.9, 3.0, 3.1,3.2] (in2); (ii) Case 2, where values
exceed 25,000 psi (172.4 MPa). The top nodes are subjected to dis- of cross-sectional areas could be selected from the discrete set of
placement limits of 0.25 in (6.35 mm) in coordinate directions x Table 3. The structure is subject to two independent load cases
and y. Because of the symmetry of the structure, the 72 elements described in Table 6.
can be grouped into 16 groups: (1) A1–A4, (2) A5–A12, (3) A13– The maximum number of optimization iterations was set as
A16, (4) A17–A18, (5) A19–A22, (6) A23–A30, (7) A31–A34, (8) A35– 1000 for all problem variants. Statistical results obtained by the
A36, (9) A37–A40, (10) A41–A48, (11) A49–A52, (12) A53–A54, (13) WCA, MBA, and IMBA algorithms are compared in Table 7. All algo-
A55–A58, (14) A59–A66 (15) A67–A70, and (16) A71–A72. rithms converged to the best weight of 385.54 lb in the case of
A. Sadollah et al. / Computers and Structures 149 (2015) 1–16 9
Table 6 problem variant 1, while the WCA and IMBA algorithms converged
Loading conditions for the spatial 72-bar truss structure problem. to a lighter design in the case of problem variant 2. The NSAs
Nodes Case 1 Case 2 required by the WCA ranged between 3200 and 19,750 for problem
Px (kips) Py (kips) Pz (kips) Px (kips) Py (kips) Pz (kips)
variant 1 and between 4600 and 26,050 for problem variant 2.
Tables 8 and 9 compare optimization results of WCA and IMBA
17 5 5 5 0 0 5
18 0 0 0 0 0 5
algorithms with the literature for Cases 1 and 2, respectively. In the
19 0 0 0 0 0 5 case of problem variant 1, the WCA, IMBA, MBA, and DHPSACO
20 0 0 0 0 0 5 converged to the lowest weight overall; however, the MBA, IMBA,
Table 7
Statistical optimization results obtained from MBA, IMBA, and WCA for the 72-bar problem.
Methods Best optimized weight (lb) Average optimized weight (lb) Worst optimized weight (lb) Standard deviation on weight (lb)
Case 1 Case 2 Case 1 Case 2 Case 1 Case 2 Case 1 Case 2
WCA 385.542 389.334 385.842 389.941 386.800 393.778 0.55 1.43
MBA 385.542 390.739 387.665 395.432 390.615 399.49 1.62 3.04
IMBA 385.542 389.334 385.765 389.823 387.942 389.457 0.41 0.84
Table 8
72-bar truss problem variant 1: comparison of WCA, MBA and IMBA optimization results with the literature.
Variables (in2) SGA [20] HS [12] HPSO [12] DHPSACO [14] MBA [17] WCA IMBA
A1–A4 1.5 1.9 2.1 1.9 1.9 1.9 1.9
A5–A12 0.7 0.5 0.6 0.5 0.5 0.5 0.5
A13–A16 0.1 0.1 0.1 0.1 0.1 0.1 0.1
A17–A18 0.1 0.1 0.1 0.1 0.1 0.1 0.1
A19–A22 1.3 1.4 1.4 1.3 1.4 1.4 1.4
A23–A30 0.5 0.6 0.5 0.5 0.5 0.5 0.5
A31–A34 0.2 0.1 0.1 0.1 0.1 0.1 0.1
A35–A36 0.1 0.1 0.1 0.1 0.1 0.1 0.1
A37–A40 0.5 0.6 0.5 0.6 0.5 0.5 0.5
A41–A48 0.5 0.5 0.5 0.5 0.5 0.5 0.5
A49–A52 0.1 0.1 0.1 0.1 0.1 0.1 0.1
A53–A54 0.2 0.1 0.1 0.1 0.1 0.1 0.1
A55–A58 0.2 0.2 0.2 0.2 0.2 0.2 0.2
A59–A66 0.5 0.5 0.5 0.6 0.6 0.6 0.6
A67–A70 0.5 0.4 0.3 0.4 0.4 0.4 0.4
A71–A72 0.7 0.6 0.7 0.6 0.6 0.6 0.6
Weight (lb) 400.66 387.94 388.94 385.54 385.54 385.54 385.54
Number of structural analyses N/A N/A 12,500 10,650 9450 3200 5750
10 A. Sadollah et al. / Computers and Structures 149 (2015) 1–16
Table 9
72-bar truss problem variant 2: comparison of the WCA, MBA and IMBA optimization results with the literature.
Variables (in2) SGA [20] HPSO [12] DHPSACO [14] MBA [17] WCA IMBA
A1–A4 0.196 4.97 1.800 1.800 1.99 1.99
A5–A12 0.602 1.228 0.442 0.602 0.442 0.442
A13–A16 0.307 0.111 0.141 0.111 0.111 0.111
A17–A18 0.766 0.111 0.111 0.111 0.111 0.111
A19–A22 0.391 2.88 1.228 1.266 1.228 1.228
A23–A30 0.391 1.457 0.563 0.563 0.563 0.563
A31–A34 0.141 0.141 0.111 0.111 0.111 0.111
A35–A36 0.111 0.111 0.111 0.111 0.111 0.111
A37–A40 1.800 1.563 0.563 0.442 0.563 0.563
A41–A48 0.602 1.228 0.563 0.442 0.563 0.563
A49–A52 0.141 0.111 0.111 0.111 0.111 0.111
A53–A54 0.307 0.196 0.250 0.111 0.111 0.111
A55–A58 1.563 0.391 0.196 0.196 0.196 0.196
A59–A66 0.766 1.457 0.563 0.563 0.563 0.563
A67–A70 0.141 0.766 0.442 0.442 0.391 0.391
A71–A72 0.111 1.563 0.563 0.602 0.563 0.563
Weight (lb) 427.203 933.09 393.380 390.73 389.334 389.334
Number of structural analyses N/A 50,000 12,500 11,600 4600 6250
Fig. 7. Comparison of convergence curves recorded for 72-bar truss problem for Variant 1.
and WCA found a slightly different design from that found by the studied in literature considering different number of design vari-
DHPSACO (see Table 8). In the case of problem variant 2, the ables and constraint types [21–23].
WCA and IMBA found the best design overall (see Table 9) and out- In this research, elements were grouped into 96 groups accord-
performed others in terms of number of structural analyses. ing to [21]; hence, this test problem included 96 sizing variables.
By observing Table 9, IMBA was successful to find cheapest Details of element grouping are given in Table 10. Material proper-
design compared with MBA for the second variant. Talking about ties were set as follows: modulus of elasticity of 30 Msi (206.9 GPa)
NSAs, Tables 8 and 9 (last row) give required NSAs used by consid- and material density of 0.283 lb/in3 (7833.4 kg/m3). The allowable
ered algorithms. WCA has performed all optimizer finding the opti- displacement was limited to 0.5 in (1.27 cm), while the allowable
mal solution in a fewer NSAs. Figs. 7 and 8 compare the stress was set equal to 30,000 Psi (206.9 MPa).
convergence curves recorded for the different optimization algo- Cross sectional areas could be selected from the following dis-
rithms for Cases 1 and 2, respectively. As shown in Figs. 7 and 8, crete set including 30 values: A = [0.100, 0.347, 0.440, 0.539,
the WCA is superior over the MBA and IMBA as it designed a lighter 0.954, 1.081, 1.174, 1.333, 1.488, 1.764, 2.142, 2.697, 2.800,
structure requiring less structural analyses for the 72-bar truss. 3.131, 3.565, 3.813, 4.805, 5.952, 6.572, 7.192, 8.525, 9.300,
However, in terms of convergence rate, IMBA shows superiority 10.850, 13.330, 14.290, 17.170, 19.180, 23.680, 28.080, 33.700]
against MBA for two considered cases. (in2).
The structure may be subject to three independent loading con-
4.3. Planar 200-bar truss problem ditions: (i) 1000 lbf acting in the positive x direction at nodes 1, 6,
15, 20, 29, 34, 43, 48, 57, 62, and 71; (ii) 10,000 lbf acting in the
The third test problem is the weight minimization of the planar negative y direction at nodes 1, 2, 3, 4, 5, 6, 8, 10, 12, 14, 15, 16,
200-bar truss structure schematized in Fig. 9. This structure was 17, 18, 19, 20, 22, 24, 26, 28, 29, 30, 31, 32, 33, 34, 36, 38, 40, 42,
A. Sadollah et al. / Computers and Structures 149 (2015) 1–16 11
Fig. 8. Comparison of convergence curves recorded for 72-bar truss problem for Variant 2.
Table 10
Details of element grouping adopted in the 200-bar truss problem.
Table 11 Table 12
Details of the best design obtained by the WCA for the 200-bar truss problem. Details of the best design obtained by the MBA for the 200-bar truss problem.
No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2)
1 0.347 25 4.805 49 10.85 73 7.192 1 0.347 25 2.697 49 7.192 73 9.3
2 0.347 26 0.1 50 0.1 74 0.1 2 0.1 26 0.44 50 0.1 74 1.764
3 5.952 27 0.347 51 0.539 75 0.539 3 5.952 27 0.347 51 2.697 75 1.333
4 0.1 28 7.192 52 5.952 76 13.33 4 0.347 28 3.813 52 7.192 76 4.805
5 0.1 29 0.347 53 0.539 77 1.174 5 0.1 29 0.1 53 0.347 77 4.805
6 2.697 30 0.1 54 0.539 78 0.44 6 2.697 30 0.1 54 0.1 78 0.1
7 0.539 31 13.33 55 7.192 79 10.85 7 0.347 31 6.572 55 7.192 79 13.33
8 0.954 32 0.1 56 0.1 80 1.333 8 0.347 32 0.1 56 0.1 80 1.764
9 2.142 33 0.44 57 0.1 81 0.1 9 2.697 33 2.142 57 0.1 81 0.539
10 0.1 34 8.525 58 14.29 82 9.30 10 0.1 34 5.952 58 7.192 82 8.525
11 0.1 35 0.539 59 0.1 83 0.347 11 0.1 35 0.1 59 0.1 83 0.1
12 0.1 36 0.347 60 0.1 84 0.954 12 0.44 36 0.347 60 1.764 84 0.1
13 3.813 37 6.572 61 10.85 85 10.85 13 4.805 37 7.192 61 10.85 85 3.813
14 0.44 38 0.1 62 0.1 86 0.347 14 0.1 38 1.333 62 1.333 86 0.1
15 0.1 39 0.1 63 1.174 87 1.174 15 0.539 39 0.347 63 0.1 87 5.952
16 4.805 40 8.525 64 7.192 88 7.192 16 3.813 40 6.572 64 6.572 88 14.29
17 0.1 41 0.347 65 0.44 89 0.1 17 0.1 41 2.142 65 0.539 89 0.954
18 0.539 42 0.44 66 0.954 90 2.697 18 0.347 42 0.1 66 0.347 90 2.142
19 3.813 43 5.952 67 13.33 91 8.525 19 8.525 43 7.192 67 8.525 91 10.85
20 0.347 44 1.488 68 0.954 92 7.192 20 1.081 44 0.1 68 2.142 92 3.565
21 0.347 45 0.1 69 0.539 93 8.525 21 1.174 45 0.44 69 0.347 93 1.488
22 13.33 46 10.85 70 9.3 94 10.85 22 7.192 46 4.805 70 14.29 94 5.952
23 0.1 47 0.347 71 0.954 95 9.30 23 1.488 47 0.1 71 0.44 95 19.18
24 0.1 48 0.347 72 0.347 96 9.30 24 0.1 48 0.1 72 1.333 96 6.572
Weight = 29,304.76 lb (13,292.41 kg) Weight = 27,532.95 lb (12,488.73 kg)
Maximum constraint violation = 3.1556e4 Maximum constraint violation = 2.9048e5
Number of structural analyses = 29,550 Number of structural analyses = 26,750
size of 200, while the population size of IMBA, MBA, and WCA algo- loads in the Z-direction, respectively. Cross-sectional areas of ele-
rithms was only 50. ments can be selected from the discrete list of 140 W-shape steel
profiles (see Table 15) based on cross-sectional areas and radius
4.4. Spatial 582-bar tower of gyration properties [24]. Cross-sectional areas of all elements
can range between 39.74 and 1378.09 cm2 (i.e., between 6.16
The last test problem solved in this study is the weight minimi- and 215 in2).
zation of the 582-bar tower structure schematized in Fig. 11. This Modulus of elasticity (E) and yield stress (Fy), respectively, are
structure was first proposed by Hasançebi et al. [24] and then 200 GPa (29,000 ksi) and 248 MPa (36 ksi), according to ASD-AISC
investigated in detail by Kaveh and Talatahari [14]. The tower specifications [25]. Stress limits also are set according to [25] as
includes 582 elements grouped in 32 independent groups. follows:
The structure is subject to a single loading condition: concen-
rþi ¼ 0:6F y ri P 0
trated forces of 5 kN (1.12 kips) and 30 kN (6.74 kips) applied ; ð29Þ
to all nodes as lateral loads in both X- and Y-directions, and vertical ri ri < 0
A. Sadollah et al. / Computers and Structures 149 (2015) 1–16 13
Table 13
Details of the best design obtained by the IMBA for the 200-bar truss problem.
No. Area (in2) No. Area (in2) No. Area (in2) No. Area (in2)
1 1.764 25 5.952 49 8.525 73 1.174
2 2.142 26 0.954 50 0.1 74 0.954
3 1.081 27 0.1 51 4.805 75 0.539
4 3.565 28 2.8 52 9.3 76 3.131
5 0.1 29 0.1 53 3.813 77 4.805
6 1.174 30 0.1 54 0.1 78 0.1
7 0.1 31 6.572 55 1.081 79 19.18
8 1.333 32 0.1 56 0.1 80 1.174
9 1.174 33 0.347 57 1.488 81 0.1
10 0.1 34 3.565 58 6.572 82 2.697
11 0.539 35 0.1 59 0.1 83 0.539
12 0.539 36 1.764 60 1.081 84 0.44
13 3.813 37 3.813 61 14.29 85 3.565
14 0.44 38 2.142 62 1.081 86 0.347
15 2.8 39 1.174 63 0.44 87 3.813
16 1.764 40 5.952 64 1.488 88 23.68
17 1.174 41 2.697 65 1.174 89 0.1
18 0.1 42 0.1 66 0.44 90 0.954
19 2.142 43 6.572 67 7.192 91 2.697
20 0.1 44 0.1 68 2.697 92 2.697
21 2.697 45 4.805 69 0.954 93 0.954
22 7.192 46 0.1 70 14.29 94 4.805
23 0.539 47 0.44 71 0.954 95 23.68
24 0.347 48 0.1 72 0.954 96 3.813
Weight = 27,349.72 lb (12,405.62 kg)
Maximum constraint violation = 6.9466e5
Number of structural analyses = 29,150
Fig. 10. Comparison of convergence curves recorded in the case of the 200-bar truss
problem.
where ri depends on the slenderness ratio as follows:
8 h . i
> k2i 3ki k3
< 1 F y 53 þ 8C 8Ci 3 for ki < C c
2C 2c c
r ¼
i
c
: ð30Þ
> 2
: 12p 2E for ki P C c
23ki
by 12 p2 E [25].
2
23k i
Table 14
Statistical optimization results obtained by the WCA, MBA, and IMBA for the 200-bar problem.
Methods Best optimized Average optimized Worst optimized Standard deviation Max. number of
weight (lb) weight (lb) weight (lb) on weight (lb) structural analyses
WCA 29,304.76 29,885.78 30,188.52 409.75 30,000
MBA 27,532.95 28,667.09 29,742.63 312.68 30,000
IMBA 27,349.72 28,500.25 29,338.65 579.09 30,000
14 A. Sadollah et al. / Computers and Structures 149 (2015) 1–16
Table 15
Profile list from AISC code for the spatial 582-bar tower.
Table 16
Spatial 582-bar tower problem: comparison of WCA, MBA, and IMBA optimization results with the literature.
Table 17
Statistical optimization results of WCA, MBA, and IMBA algorithms for the 582-bar tower problem.
Methods Best optimized volume (m3) Average optimized volume (m3) Worst optimized volume (m3) Standard deviation on volume (m3) Max. NSAs
WCA 20.1919 20.4253 20.7339 1.92e1 17,500
MBA 20.0737 20.1023 20.1489 4.07e2 17,500
IMBA 20.0688 20.0990 20.1027 2.01e2 17,500
A. Sadollah et al. / Computers and Structures 149 (2015) 1–16 15
Table 18
Sensitivity of WCA to internal parameters for the 72-bar truss problem.
Table 19
Sensitivity of WCA to internal parameters for the 582-bar tower problem.
In general, setting dmax as a large value reduces the exploitation The IMBA slightly outperformed the MBA and WCA algorithms
capability, while setting small values of dmax reduces the explora- in the large scale truss problems in this paper. Hybridization of the
tion capability. This is confirmed by the sensitivity analysis data WCA with MBA and/or other optimizers may lead to further
reported in Tables 18 and 19. The worst designs were obtained improve convergence behavior and is certainly a topic to be inves-
for the very large value dmax = 1e1 and for the very small value tigated in future research.
dmax = 1e7 that also entailed the largest standard deviations (SD).
The last parameter affecting performance of WCA is the size of Acknowledgments
initial population (Npop). It should be noted that Npop is a common
input parameter for most metaheuristic algorithms. In general, This work was supported by a National Research Foundation of
Npop should be considerably larger than Nsr. In the WCA case, if Korea (NRF) Grant funded by the Korean government (MSIP) (NRF-
Nsr = 8 and Npop = 10, some rivers are generated without having 2013R1A2A1A01013886).
any streams moving toward them. This has a detrimental effect
on optimization results. In particular, if Nsr = 8, the worst and best References
performance of the WCA algorithm occur for Npop = 10 and
Npop = 50, respectively. [1] Holland J. Adaptation in natural and artificial systems. Ann Arbor
(MI): University of Michigan Press; 1975.
In summary, the best value for dmax is in the range [1e5,1e3], [2] Goldberg DE, Samtani MP. Engineering optimization via genetic algorithms. In:
while the value of Nsr should be selected based on the value of Npop Proc of the ninth conference on electronic computations, ASCE. Birmingham,
(looking at sensitivity analysis data, Nsr should be about one sixth Alabama; 1986. p. 471–82.
[3] Rajeev S, Krishnamoorthy CS. Discrete optimization of structures using genetic
of Npop). By increasing the number of iterations, WCA as well as algorithms. J Struct Eng 1992;118(5):1233–50.
other metaheuristic algorithms will have more chances to find [4] Krishnamoorthy CS, Venkatesh PP, Sudarshan R. Object-oriented framework
the best solution and explore the design space. For example, by for genetic algorithms with application to space truss optimization. J Comput
Civ Eng 2002;16:66–75.
increasing the number of iterations to 500, WCA could reach the [5] Sivakumar P, Rajaraman A, Natajan K, Samuel KGM. Artificial intelligence
best solution for both test problems. techniques for optimization of steel lattice towers, recent developments in
Talking about MBA and IMBA internal parameters, the sug- structural engineering. In: Proc structural engineering convention; 2001. p.
435–45.
gested value of l is the maximum iteration number divided by five
[6] Gero MBP, Garcia AB, Diaz JJDC. Design optimization of 3D steel structures:
recommended in Section 3.1.1. The exploration factor (l) gives the genetic algorithms vs. classical techniques. J Constr Steel Res 2006;62:1303–9.
algorithm more freedom to search a wider range, which results in [7] Geem Z, Kim J, Loganathan GV. A new heuristic optimization algorithm:
harmony search. Simulation 2001;76:60–8.
better detection of solutions. However, increasing the value of this
[8] Lee KS, Geem ZW. A new meta-heuristic algorithm for continuous engineering
parameter more than the predefined value may diverge the results optimization: harmony search theory and practice. Comput Method Appl
and prevent further exploitation near the best obtained solutions. Mech 2005;194:3902–33.
The reduction factor (a) divides the distance of shrapnel pieces [9] Lee KS, Geem ZW. A new structural optimization method based on the
harmony search algorithm. Comput Struct 2004;82:781–98.
into a interval distances, which enables searching within the [10] Lee KS, Geem ZW, Lee SH, Bae KW. The harmony search heuristic algorithm for
reduced intervals at each iteration. Searching smaller intervals is discrete structural optimization. Eng Optim 2005;37(7):663–84.
better and more efficient than searching in large intervals. [11] Kennedy J, Eberhart R. Particle swarm optimization. In: Proc IEEE IJCNN. Perth,
Australia; 1995. p. 1942–48.
Higher values of a increase the probability of finding the global [12] Li LJ, Huang ZB, Liu F. A heuristic particle swarm optimization method for truss
optimal solution; however, increasing a more than the suitable structures with discrete variables. Comput Struct 2009;87:435–43.
value may inhibit searching regions near the optimal solution dur- [13] He S, Wu QH, Wen JY, Saunders JR, Paton RC. A particle swarm optimizer with
passive congregation. Biosystems 2004;78:135–47.
ing the final iterations. The user specified parameter sensitivity for [14] Kaveh A, Talatahari S. A particle swarm ant colony optimization for truss
the MBA for solving truss structures is given in [26] which can also structures with discrete variables. J Constr Steel Res 2009;65:1558–68.
be extend for weight/volume minimization problems of complex [15] Lamberti L, Pappalettere C. Metaheuristic design optimization of skeletal
structures: a review. Comput Technol Rev 2011;4:1–32.
truss structures.
[16] Saka MP. Optimum design of skeletal structures: a review. In: Topping BHV,
editor. Progress in civil and structural engineering computing. Stirlingshire
(UK): Saxe-Coburg Publications; 2003. p. 237–84. http://dx.doi.org/10.4203/
csets.10.10 [chapter 10].
6. Conclusions [17] Sadollah A, Bahreininejad A, Eskandar H, Hamdi M. Mine blast algorithm for
optimization of truss structures with discrete variables. Comput Struct
This paper presented some applications of two recently devel- 2012;102–103:49–63.
[18] Eskandar H, Sadollah A, Bahreininejad A, Hamdi M. Water cycle algorithm – a
oped optimization techniques, the mine blast algorithm (MBA) novel metaheuristic optimization method for solving constrained engineering
and water cycle algorithm (WCA) to discrete sizing optimization optimization problems. Comput Struct 2012;110–111:151–66.
problems of truss structures. The WCA and MBA are inspired by [19] Montes EM, Coello CAC. An empirical study about the usefulness of evolution
strategies to solve constrained optimization problems. Int J Gen Syst
observation of water cycle process and explosion of landmines, 2008;37:443–73.
respectively. The MBA was improved and applied to solve bench- [20] Wu SJ, Chow PT. Steady-state genetic algorithms for discrete optimization of
mark problems, so called improved MBA (IMBA). trusses. Comput Struct 1995;56:979–91.
[21] Ghasemi MR, Hinton E, Wood RD. Optimization of trusses using genetic
The efficiency and performance of IMBA, WCA, and MBA was
algorithms for discrete and continuous variables. Eng Comput 1999;16–
examined using four truss structures. The improvements made to 3:272–301.
MBA are as follows: the distance of each shrapnel piece was chan- [22] Dede T, Bekiroglub S, Ayvazc Y. Weight minimization of trusses with genetic
ged by the value of the objective function (i.e., it was reduced when algorithm. Appl Soft Comput 2011;11:2565–75.
[23] Cai J, Thierauf G. Discrete structural optimization using evolution strategies.
there were no improvements to the obtained solution). The per- In: Topping BHV, Khan AI, editors. Neural networks and combinatorial
ception of direction was replaced with the concept of moving to optimization in civil and structural engineering, civil-comp, Edinburg. p. 95–
the best solutions. 100. doi: http://dx.doi.org/10.4203/ccp.16.6.11993.
[24] Hasançebi O, Çarbas S, Dogan E, Erdal F, Saka MP. Performance evaluation of
Optimization results obtained for four benchmark truss prob- metaheuristic search techniques in the optimum design of real size pin jointed
lems including a number of sizing variables ranging between 12 structures. Comput Struct 2009;87(5–6):284–302.
and 96 seem to prove that the IMBA, WCA, and MBA algorithms [25] American Institute of Steel Construction (AISC). Manual of steel construction
allowable stress design. 9th ed. Chicago (IL); 1989.
are highly competitive with other metaheuristic algorithms pre- [26] Sadollah A, Bahreininejad A, Eskandar H, Hamdi M. Mine blast algorithm: a
sented in the literature in terms of optimized weight and number new population based algorithm for solving constrained engineering
of structural analyses required by the optimization process. optimization problems. Appl Soft Comput 2013;13:2592–612.