Expert Systems With Applications 125 (2019) 425–441

Contents lists available at ScienceDirect

Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa

Structural optimization using multi-objective modified adaptive

symbiotic organisms search
Ghanshyam G. Tejani a,∗, Nantiwat Pholdee b, Sujin Bureerat b, Doddy Prayogo c,
Amir H. Gandomi d
a
Department of Mechanical Engineering, School of Technology, GSFC University, Vadodara, Gujarat, India
b
Department of Mechanical Engineering, Faculty of Engineering, Khon Kaen University, Thailand
c
Department of Civil Engineering, Petra Christian University, Jalan Siwalankerto 121-131, Surabaya 60236, Indonesia
d
School of Business, Stevens Institute of Technology, Hoboken, NJ 07030, USA

a r t i c l e i n f o a b s t r a c t

Article history: Multiple objective structural optimization is a challenging problem in which suitable optimization meth-
Received 10 August 2018 ods are needed to find optimal solutions. Therefore, to answer such problems effectively, a multi-objective
Revised 17 January 2019
modified adaptive symbiotic organisms search (MOMASOS) with two modified phases is planned along
Accepted 18 January 2019
with a normal line method as an archiving technique for designing of structures. The proposed algorithm
Available online 30 January 2019
consists of two separate improved phases including adaptive mutualism and modified parasitism phases.
Keywords: The probabilistic nature of mutualism phase of MOSOS lets design variables to have higher exploration
Adaptive mechanism and higher exploitation simultaneously. As search advances, a stability between the global search and a
Structural optimization local search has a significant effect on the solutions. Therefore, an adaptive mutualism phase is added
Meta-heuristics to the offer MOASOS. Also, the parasitism phase of MOSOS offers over exploration which is a major is-
Discrete variables sue of this phase. The over exploration results in higher computational cost since the majority of the
Constrained problems
new solutions gets rejected due to inferior objective functional values. In consideration of this issue, the
parasitism phase is upgraded to a modified parasitism phase to increase the possibility of getting im-
proved solutions. In addition, the proposed changes are comparatively simple and do not need an extra
parameter setting for MOSOS.
For the truss problems, mass minimization and maximization of nodal deflection are considered as
objective functions, elemental stresses are considered as behavior constraints and (discrete) elemental
sections are considered as side constraints. Five truss optimization problems validate the applicability of
the considered meta-heuristics to solve complex engineering. Also, four constrained benchmark engineer-
ing design problems are solved to demonstrate the effectiveness of MOMASOS. The results confirmed that
the proposed adaptive mutualism phase and modified parasitism phase with a normal line method as an
archiving technique provide superior and competitive results than the former obtained results.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction (Pholdee & Bureerat, 2013b). The applications include a bridge, a

tower, a transmission tower, and a billboard structure. In designing
Optimal truss design is among the hottest research challenge a truss structure, engineering will define its topology, shape, and
of structural engineering. Recently during each year, hundreds of elements’ sizes. Usually, a trial-and-error approach can be applied.
papers related to the topic were published. Truss structures can be Nevertheless, for large trusses, such an approach is not efficient
viewed as a set of 2-node links interconnected by spherical joints. and effective.
They have been used in several engineering applications with the Therefore, the application of an optimization technique is a bet-
advantages in that they are simple to construct, low cost, easy to ter choice. The tradition truss optimization is used to get the best
design and less difficult to construct in difficult-to-access regions suitable topology, shape, and sizes to minimize weight or cost
subject to structural safety constraints. Gradient-based optimizers
can be used in cases of continuous design variables (Allwood &
∗
Corresponding author. Chung, 1984; Fleury, 1980). However, over the last few decades,
E-mail addresses: [email protected] (G.G. Tejani), [email protected] the use of meta-heuristics (MHs) are the main focus due to
(N. Pholdee), [email protected] (S. Bureerat), [email protected] (D. Prayogo), their simplicity to use, code, and implement. Unlike its gradient
[email protected] (A.H. Gandomi).

https://doi.org/10.1016/j.eswa.2019.01.068
0957-4174/© 2019 Elsevier Ltd. All rights reserved.
426 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

counterpart, MHs can be applied to answer almost any type of de- ent as a local search to enhance the performance of MOEAs was
sign variables. The combination of several types of design variables presented in Pholdee and Bureerat (2012, 2013a). Other work with
for one optimization run is possible. This aid makes MHs more MO of trusses can be found in Angelo, Barbosa, and Bernardino
popular than gradient-based optimization methods for truss opti- (2012, 2015), Greiner and Hajela (2012), Hosseini, Hamidi, Mansuri,
mization. Moreover, some MHs can explore a Pareto front, in cases and Ghoddosian (2015), Kaveh and Laknejadi (2013), Kaveh and
of multi-objective optimization (MO), within on simulation run. Mahdavi (2018), Mousa, El-Shorbagy, and Abd-El-Wahed (2012),
The use of MHs for single-objective optimization has been com- Richardson et al. (2012), Su, Wang, Gui, and Fan (2011) and
monplace. Over the years, there have been numerous MHs newly Tejani, Bureerat, Pholdee, and Prayogo (2018c). It has been shown
invented. Some of the popular techniques for truss design include a from the literature that a study on using MOMHs for truss design
genetic algorithm (GA) (Lingyun, Mei, Guangming, & Guang, 2005; is much more advantageous.
Wei et al., 2011; Zuo, Xu, Zhang, & Xu, 2011), particle swarm op- As a result, this paper deals with modification and improve-
timization (PSO) (Gomes, 2011), cuckoo search (Gandomi, Talata- ment of symbiotic organisms search for truss MO. Since it was
hari, Yang, & Deb, 2012), krill herd algorithm (Gandomi, Talata- first invented by Cheng and Prayogo (2014), the optimizer has
hari, Tadbiri, & Alavi, 2013), differential evolution (DE), teaching- been implemented on a number of applications while many mod-
learning based optimization (Camp & Farshchin, 2014; Degertekin ified versions have been additionally proposed (Ayala, Klein, Mar-
& Hayalioglu, 2013; Savsani, Tejani, & Patel, 2016; Tejani, Savsani, iani, & Coelho, 2017; Çelik & Öztürk, 2017; Ezugwu & Adewumi,
& Patel, 2016b), Ray optimization (Kaveh & Khayatazad, 2013), col- 2017; Ezugwu, Adewumi, & Frîncu, 2017; Guha, Roy, & Banerjee,
liding body algorithm (Kaveh & Mahdavi, 2014), Parameter-less 2017; Prayogo, Cheng, Wong, Tjandra, & Tran, 2018; Secui, 2016;
population pyramid (Gandomi & Goldman, 2018), and grey wolf Zhang, Sun, Yuan, Lv, & Ma, 2016; Çelik & Durgut, 2018). SOS was
optimizer (GWO) (Kaveh & Zakian, 2017; Panagant & Bureerat, then being upgraded for MO (Tran, Cheng, & Prayogo, 2016; Duc-
2018). Later, some of those baseline algorithms have been mod- Hoc, 2017) leading to multi-objective symbiotic organisms search
ified or improved leading to more advanced versions e.g. adap- (MOSOS). Investigation on improving the performance of MOSOS
tive DE (Bureerat & Pholdee, 2015), modified symbiotic organisms for truss design is interesting since it is a new method that should
search (Kumar, Tejani, & Mirjalili, 2018; Tejani, Savsani, Patel, & be tested with this popular research topic. In this work, the main
Mirjalili, 2017; 2018c). The performance enhancement can also be contribution is an incorporation of the random migration based
achieved by means of hybridization such as hybridized passing ve- search along with adaptive benefit factors (BFs) into MOSOS. These
hicle search & simulated annealing (Tejani, Savsani, Bureerat, Patel, techniques are used to set better stability between exploration and
& Savsani, 2018b), and hybrid GWO & self-adaptive DE (Panagant to improve exploitation during mutualism phase, and to improve
& Bureerat, 2018). Recently, the performance test of a number of exploration during parasitism phase of MOSOS. A number of multi-
self-adaptive MHs on solving truss optimization has been investi- objective truss design are used to validate the new algorithms
gated. It is found that most of CEC (Congress on Evolutionary Com- while several state-of-the-art MOMHs are used to compare with
putation) competition winners are some of the top MHs for truss the new MOSOS. The results show that our proposed method is
optimization (Pholdee & Bureerat, 2017). powerful for truss optimization.
Once more than one design objectives are posed, the optimiza- Because the optimizer has just been proposed, there is room
tion problem is called MO. It is furthermore called many-objective for further development and investigation, as a consequence, this
optimization in cases of a problem having more than three objec- study is proposed to enhance the effectiveness of the MOSOS by
tive functions in order to state its difficulty to explore the entire incorporating a modified parasitism system. MOASOS and MOMA-
Pareto front. The use of multi-objective meta-heuristics (MOMHs) SOS are employed to answer multi-objective truss design problems
for truss optimization has been studied for a decade. It is well rec- while the objectives comprise truss mass minimization and nodal
ognized that a designer always needs for optimizing many objec- deflection maximization. The solutions received from various opti-
tive functions at the same time and those objectives will always be mizers are examined and presented.
conflicting with each other. The solutions for such a design prob-
lem are countless, and its solution set is termed a Pareto opti- 2. The symbiotic organisms search (SOS) algorithm
mal set (or a Pareto front) if viewed as per the objective func-
tion domain. The main reason for MOMHs popularity in MO is that Cheng and Prayogo (2014) developed SOS to serve as a
MOMHs is capable to get a Pareto front in a single run. The pio- continuous-based MH algorithm and a population-oriented search-
neering MOMHs were a multi-objective genetic algorithm (MOGA) ing technique; the technique finds global optimum solutions by
(Fonseca & Fleming, 1993), a SPEA2 (Zitzler, Laumanns, & Thiele, retaining a set of possible ones called a population. SOS is fo-
2001), and a NSGA-II (Deb, Pratap, Agarwal, & Meyarivan, 2002). cused upon symbiosis, which is the process by which organisms
Later there have been a great variety of improved versions of ex- in an ecosystem possess biological interdependence with one an-
isting algorithms (Bureerat & Srisomporn, 2010; Kaveh & Laknejadi, other which allows them to grow and survive. Due to its excellent
2011; Pholdee & Bureerat, 2012, 2013a, 2013b; Zitzler, Laumanns, performance over the benchmark algorithms, SOS has been applied
& Thiele, 2002) and newly invented methods such as DE for MO to numerous research fields since its introduction (Cheng, Prayogo,
(Robič & Filipič, 2005). Some of them were upgraded for solv- & Tran, 2015; Tran et al., 2016; Abdullahi, Ngadi, & Abdulhamid,
ing many-objective optimization such as a non-dominated sort- 2016; Çelik & Öztürk, 2017; Guha et al., 2017; Panda & Pani, 2016;
ing genetic algorithm (NSGAIII) (Deb & Jain, 2013; Jain & Deb, Prayogo & Susanto, 2018; Cheng et al., 2018; Prayogo et al., 2018;
2013), Two-arch (Wang, Jiao, & Yao, 2015), and knee-point opti- Tejani, Savsani, & Patel, 2016a, 2017, 2018a; Yu, Perwira Redi, Yang,
mizer (Zhang, Tian, & Jin, 2015). Ruskartina, & Santosa, 2017; Çelik & Durgut, 2018).
The use of MOMHs for truss optimization will provide bene- The first step performed by the SOS algorithm is the initializa-
fit in that a designer can have many solutions for decision making tion of the specific population in an ecosystem. Following this is a
(Kaveh & Mahdavi, 2018; Noilublao & Bureerat, 2011, 2013; Pholdee process by which the algorithm evaluates organisms’ locations by
& Bureerat, 2012, 2013a, 2013b). Moreover, they can be used for computing the particular objectives’, such that the organisms with
reliability optimization of trusses (Ho-Huu, Duong-Gia, Vo-Duy, Le- the best solution is elected as ‘Xbest ’. This action takes place in it-
Duc, & Nguyen-Thoi, 2018; Techasen et al., 2018). Several MOMHs erations, finding the global best solution by updating to the most
were used to tackle multi-objective truss design in Noilublao and recently available solution until the solution is found. For this in-
Bureerat (2011, 2013). The use of the so-called approximate gradi- stance, three fundamentals of symbiosis, mutualism, commensal-
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 427

ism, and parasitism inspired the principle rules used by the algo- 2.3. Parasitism phase
rithm. These rules were used to update the positions of new organ-
isms. Once the algorithm reaches the maximum number of func- This parasitism phase requires one organism to be nega-
tion evaluations, termination of the loops is implemented. Below is tively affected to the benefit of a second organism. Humans and
an explanation of how mutualism, commensalism, and parasitism mosquitoes demonstrate this symbiotic relationship, in which the
come into play in the MH. mosquito’s bite releases a parasite into the human. Growing inside
the body, the parasite can cause harm or kill the host if the situa-
tion becomes severe enough. One can identify the parasitic nature
2.1. Mutualism phase
of this relationship in the fact that an organism is helped while the
second is hurt.
This phase involves an association by which both parties pos-
The solution ‘Xi ’ takes queues from the Anopheles mosquito,
itively benefit. In the relationship between a flower and a polli-
constructing Parasite Vector (PV) which mimics the behaviors of
nator, the pollinator benefits from the food it can take from the
the parasite. The formation of PV requires regeneration of parts of
flower, while the flower can turn into fruit from its contact with
the solution ‘i’ which are chosen via partially randomized selection
the pollinator. Due to this twofold positive benefit, the relationship
using specific boundaries (LB and UB) as shown in Eq. (7). Mean-
can be deemed as a mutually beneficial symbiosis.
while, ‘Xk ’ is derived from a solution chosen via randomized selec-
For mutualism phase, organism ‘i’ is assigned as a solution (Xi )
tion (note that k = i) and serves as the host of the parasite. Should
to interact with the secondary solution chosen via randomized se-
the fitness value of solution ‘k’ be surpassed by PV, the host will
lection (Xk ) (in this instance, k = i). This symbiotic relationship
die, and the PV will take its place.
positively impacts both solutions. The BFs and a mutual vector 
(MV) dictate new solutions. BF1 and BF2 are determined via ran- Xij i f rand (0, 1 ) ≤ rand (0, 1 )
PV = (7)
domized selection between 1 or 2 (see Eqs. (4) and (5)). Because of LB + rand (0, 1 ) ∗ (UB −LB ) Otherwise
this, either BF demonstrates an example of a solution experiencing
positive benefits somewhat or entirely through symbiosis. The best where, j ∈ (1, 2, …, m); k ∈ (1, 2, …, n); j signifies design variables.
solution (Xbest ) is an additional variable which solutions can be im-
pacted by; it is selected through a random search from the Pareto 3. Modifications in multi-objective symbiotic organisms search
set of non-dominated sorting. Meanwhile, a greedy selection is uti- (MOSOS)
lized in order to determine the fitter solutions. The following is the
mathematical formulation behind the mutualism phase. Performance of MHs largely depends on the stability in the
exploration & the exploitation. The exploration characterizes the

Xi = Xi + rand (0, 1 ) ∗ (Xbest − MV ∗ BF1 ) (1) global search capacity of the MHs and decides the accuracy of ob-
tained solutions. The exploitation characterizes the local search ca-
 pacity of the MHs and plays a significant part in the rapid con-
Xk = Xk + rand (0, 1 ) ∗ (Xbest − MV ∗ BF2 ) (2)
vergence. As discussed earlier, the application of an adaptive con-
trolling mechanism on the various MHs set a stability between the
Xi + Xk global search and a local search. Thus, adaptive BFs are proposed
MV = (3)
2 in the mutualism phase of MOSOS. Also, the parasitism phase of
MOSOS is upgraded leading to a modified parasitism phase to ad-
BF1 = round[rand (0, 1 )] + 1 (4) dress the issue regarding population diversity. The detailed discus-
sion of the proposed improvements on the MOSOS algorithm is
BF2 = round[rand (0, 1 )] + 1 (5) presented in the subsequent sections.

where, i, k ∈ (1, 2, …, n); i = k 3.1. Multi-objective adaptive symbiotic organisms search (MOASOS)

2.2. Commensalism phase In the mutualism phase of MOSOS, the two organisms of dif-
ferent species result from interactive learners into personal bene-
Commensalism involves a single organism receiving benefit fit of the symbiotic collaboration. Thus, the BFs (BF1 and BF2 ) are
from a symbiotic relationship while another is completely unaf- main components which defines the effect of MV. BFs are definite
fected positively or negatively. One example of this includes the by a heuristically, and their values are one or two. This step out-
shark and remora fish, in which the fish suctions under the shark comes in the state where populations/organisms ‘Xi ’ and ‘Xk ’ ben-
and gains access to the nutrients the shark does not eat. Mean- efit partly or completely from MV. Therefore, in the mutualism
while, the shark is not impacted by the exchange in any way. Com- phase the populations progress only with two possibilities. How-
mensalism is mimicked by the algorithm based on this fundamen- ever, in the original mutualism, BF should not be at end positions
tal. only, but it can be in-between these limits also. Given this fact,
For the commensalism phase, two solutions interact with one Tejani et al. (2018c) upgraded this phase to adaptive mutualism
another (Xi and Xk ) (in this instance, k = i). While solution ‘i’ is phase by incorporating adaptive benefit factors (ABF1 and ABF2 )
positively benefited from the other solution, solution ‘k’ experi- to advance search capacity of the MHs, defined by the following
equations:
ences no impact. The best solution (Xbest ) is an additional variable 
which solutions can be impacted by; it is selected through a ran- fa (Xi )/ fa (Xbest ), i f fa (Xbest ) = 0
dom search from the Pareto set of non-dominated sorting. Mean-
ABFa = (8)
1 + round[rand (0, 1 )], i f fa (Xbest ) = 0
while, the greedy selection is utilized in order to determine the ⎧
fitter solutions. The following is the mathematical formulation be- ⎨1, i f ABFa < 1
hind the commensalism phase. BFa = 2, i f ABFa > 2 (9)
 ⎩
Xi = Xi + rand (−1, 1 ) ∗ (Xbest − Xk ) (6) ABFa , otherwise

where, i, k ∈ (1, 2, …, n); i = k where, a = 1 & 2.

428 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

The design variables (Xi ) may get small and large displacement
from their positions as various factors govern it during mutualism
phase. These displacements of the design variables influence the
exploration and the exploration. Hence, smaller value of BF lets the
fine/local search in tiny moves but then results in faster conver-
gence and bigger value of BF lets global search but then results in
slower convergence. The ‘ABF1 ’ and ‘ABF2 ’ affects the exploration
capability of the optimizer when a solution (‘Xi ’ or ‘Xk ’) is away
from the best solution (‘Xbest ’). The adaptive mutualism phase sets
good exploitation when a solution is the neighbor of the result-
ing solution. Multi-objective adaptive SOS (MOASOS) purposes to
efficiently incorporate the local and global search characteristic by
using an adaptive mutualism phase.

3.2. Multi-objective modified adaptive symbiotic organisms search

(MOMASOS)

Furthermore, a parasitism phase is upgraded to a modified par-

asitism phase which leads MOASOS (Tejani et al., 2018c) to a new
algorithm called multi-objective modified adaptive symbiotic or-
ganisms search (MOMASOS). The parasitism phase of MOSOS per-
formances a significant role in upgrading the exploration ability
of MOSOS. However, it is also observed that over exploration re-
sults in higher computational cost as a majority of the new solu-
tions generated by the parasitism phase gets rejected due to in-
ferior objective functional values compared to previous one (Do &
Lee, 2017). Therefore, parasitism phase is improved with a modi-
fied parasitism phase of MOSOS.
In the original parasitism phase, a parasitism vector (PV), Xi ,
is generated by mutating/altering values of few heuristically cho-
sen design variables of the population ‘Xj ’, the Anopheles mosquito.
Thus, the PV is a blend of ‘Xi ’ and random values within its bounds.
The graphical representation of the parasitism phase is presented
in Fig. 1. Let, ‘Xi ’ is the current solution with two design variables
(x1 , y1 ) as shown in Fig. 1(a). Therefore, the updated solution (Xi )
or PV can either get a position within dotted lines (if single vari-
able changes) as shown in Fig. 1(c and d) or it holds its position Fig. 1. The parasitism vectors.
(see Fig. 1(b)) or it may move any random point within its bounds
(if both variables change) with an equal probability. Hence, the
original parasitism phase offers too explorative search which gen-
erates a large number of inferior solutions and consumes higher ‘Xj ’ are to be changed directly by the newer solutions ‘Xi ’ and ‘X j ’,
unnecessary computational cost. respectively. Else, the ‘Xi ’ and ‘X j ’ will be incorporated to the ad-
In the modified parasitism phase, a modified parasitism vec- vanced solution for choosing the next iteration ecosystem. Thus,
tor (MPV), Xi , is generated by migrating values of few heuristically these MHs are able to converge better by keeping good diversity
chosen design variables of the population, ‘Xj ’, to the heuristically among solutions. Since MHs may advance few significant data from
selected solution ‘Xk ’ (where k = i; selected randomly from non- dominated solutions in future update.
dominated archive), or the Anopheles mosquito, to the current so- The original version of MOSOS exploited the elitism strategy
lution ‘Xi ’, a human host. Thus, MPV is a blend of design variables in combination with the crowd comparison for selection of the
‘Xi ’ and randomly selected solution ‘Xk ’. The graphical representa- next generation population. This numerical strategy was success-
tion of the modified parasitism phase is presented in Fig. 2. As dis- fully employed in NSGAII and some other MOEAs. The method
cussed earlier randomly selected design variables of the solution works by using the dominance level of solutions being selected.
(Xk ) migrates to a current solution (Xi ). Let, ‘Xi ’ is the current so- Given a set of design solutions, non-dominated solutions are those
lution and ‘Xk ’ is randomly selected solution with two design vari- who have dominance level being 1. If the non-dominated solu-
ables (x1 , y1 ) and (x2 , y2 ). Thus, the updated solution (Xi ) or MPV tions are removed from the set, solutions having dominance level
can acquire a corner position of a dotted rectangle as shown in as 2 will be non-dominated solutions and so on. The idea is to
Fig. 2(b–e) with equal probability. Thus, this modification advances choose solutions with lowest dominance levels for the next gener-
the exploration of search and also provides better the exploitation ation population. In cases that the number of solutions with low-
which offers a large number of acceptable solutions, and it also est dominance levels exceeds the predefined population size, some
reduces computational cost. The following is the mathematical for- of the solutions with the highest dominance level in the set who
mulation of MPV behind the parasitism phase. have lower cuboids are removed from the set. In this version of
 MOMASOS, a similar strategy is used but the normal line tech-
Xij i f rand (0, 1 ) ≤ rand (0, 1 )
MPV, Xi = (10) nique (Bureerat & Srisomporn, 2010) is used instead of the crowd
Xkj otherwise
comparison when some solutions are to be removed. The normal
where, i, k ∈ (1, 2, …, n); j ∈ (1, 2, …, m); i = k line method was originally proposed as an archiving technique for
Solutions are growing to a better form only if newer fitness is multiobjective population-based incremental learning. The method
better than the previous one. Thus, the current solutions ‘Xi ’ and is illustrated in Fig. 3 where there are 5 solutions (circle markers)
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 429

normal lines are those who are perpendicular to the Utopia line
and equally placed along the line. The number of the normal lines
is equal to the number of solutions require from those with the
dominance level being 3, which for this example is 6 lines. The 6
selected solutions whose dominance level is 3 are those who are
the closet solutions to their corresponding lines. In Fig. 3, the 6
selected solutions are inside the dashed circles.
The proposed MHs simulates initialization, mutualism phase
(or adaptive mutualism phase), commensalism phase, parasitism
phase (or modified parasitism phase), and stopping criteria. The
combined flowchart of the proposed MHs is presented in Fig. 4.

4. Problem definition

A multi-objective truss design problem is defined to find dis-

crete elemental cross-sections (design variables) to minimize truss
mass and maximize deflection of nodes subject to elemental stress
constraints. The truss optimum design problem is stated as:

F ind, A = {A 1 , A 2 , . . . , A m } (11)
to minimize mass and maximize nodal deflection of truss

m  
f 1 (A ) = Ai ρi Li and f2 (A ) = max δ j 
i=1

Subject to:
Behavior constraints:
g(A ) : St ress const raint s, |σi | − σimax ≤ 0
Side constraints:
Discrete cross − sectional areas, Amin i
≤ Ai ≤ Amax
i
where, i = 1, 2, . . . , m; j = 1, 2, . . . , n
where, Ai, ρ i, Li , Ei , and σ i represent design variables (elements’
cross-sections), density, elemental length, young’s modules, and el-
emental stress of the ‘ith’ the element respectively. ‘δ j ’ is a deflec-
tion of the ‘jth’ node. The superscripts ‘max’ and ‘min’ stands for
upper and lower allowable bounds respectively.
Fig. 2. The modified parasitism vectors.
4.1. Dynamic penalty function

Considering both objective functions differently and it is to

minimize objectives subject to ‘p’ limitations, the dynamic penalty
function is stated as:
 

p  qi 
f ( X ) ∗ ( 1 + ε1 ∗ C ) , C =ε2 
Ci , Ci = 1 − ∗  (12)
q i
i=1

where, qi signifies constraint violation with respect to the limit ‘q∗i ’.

The parameter p signifies a count of live constraints. The variables
‘ε 1 ’ and ‘ε 2 ’ are can be assumed by considering the problem char-
acteristics. In this investigation, both ‘ε 1 ’ and ‘ε 2 ’ are assumed as 3,
as per investigation of their effect on exploitation and exploration
equilibrium (Tejani et al., 2016a, 2017, 2018a, 2018b).

5. Truss design problems and discussions

Fig. 3. Normal line method.
Five truss problems from Angelo et al. (2012, 2015) and
Tejani et al. (2018c) are considered to test the effectiveness of
having dominance level as 1, 4 solutions (diamond markers) hav- the proposed MHs. For fair comparison, the similar parameters
ing dominance level as 2, and 10 solutions (square markers) hav- (Angelo, Bernardino, & Barbosa, 2015; Tejani et al., 2018c) are fol-
ing dominance level as 3. If the predefined population size is 15, lowed in this study. Thus, all the problems were performed with
it means the optimizer will keep all solutions with dominance lev- the population (organism) size of 100 and 50,000 functional eval-
els 1 and 2; while, for the dominance level as 3, 4 solutions will uations. The proposed MHs are tested for 100 discrete runs. The
be deleted from the population. The normal line method works, in front-hypervolume (HV) & front spacing-to-extent (STE) tests are
cases of two objective functions, by identifying the anchor points considered for the assessment. The mean value of the HV of each
who currently give the minimum values for f1 and f2 . Then, the MH is chosen to quantify the convergence rate of the MH and the
so-called Utopia line is drawn connecting the two anchors. The standard deviation (STD) of HV is considered to quantify the search
430 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

Fig. 4. Flowchart of the proposed algorithms.

 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 431

Table 1
Design considerations of the truss problems.

The 10-bar truss The 25-bar truss The 60-bar truss The 72-bar truss The 942-bar truss

Design variables Ai, i = 1, 2, …, 10 Ai, i = 1, 2, …, 8 Ai, i = 1, 2, …, 25 Ai, i = 1, 2, …, 16 Ai, i = 1, 2, …, 59

Constraints σ max = 25 ksi σ max = 40 ksi σ max = 40 ksi σ max = 25 ksi σ max = 25 ksi
Density ρ = 0.1 lb/in3 ρ = 0.1 lb/in3 ρ = 0.1 lb/in3 ρ = 0.1 lb/in3 ρ = 0.1 lb/in3
Young modules E = 104 ksi E = 104 ksi E = 104 ksi E = 104 ksi E = 104 ksi

Table 2
The hypervolume values of results obtained for the 10-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 47,302.93 53,090.89 50,902.21 1294.12 100 5

MOACS 52,662.60 54,395.00 53,639.77 307.79 200 4
MOSOS 55,646.49 56,543.02 56,055.02 191.00 335 3
MOASOS 55,528.07 56,642.93 56,220.00 195.67 404 2
MOMASOS 55,890.60 56,873.05 56,389.83 166.10 461 1

reliability. Also, a front spacing (S) measure (Schott, 1995) is con-

sidered to test comparative distance between successive popula-
tions in the non-dominated set. The spacing of the front can be
calculated as:
|P | 

1 2
Spacing = di − d¯ (13)
|P| − 1 i=1

where |P| is count of associates. di is the Euclidian distance of ob-

jective ‘i’ to its adjacent solution. d¯ is the mean result of di ’
The front extension is considered as:
M 
 
Extent =  fimax − fimin  (14)
i=1

The smaller value of Spacing shows the superior Pareto front

while, in contrast to the higher values of Extent is the superior. Fig. 5. The 10-bar truss.
The simultaneous consideration of spacing and extent excels a new
evaluation metric which simultaneously exams spacing and extent
together, which is presented as the ratio of spacing to the extent, 191.00, 195.67, and 166.10 respectively. It is found from the re-
sults that MOMASOS performs the best followed by MOASOS and
ST E = Spacing/Extent (15) MOSOS as per the measure of search consistency. The Friedman’s
rank test is used to compare different MHs based on the ranks.
where the smaller value of STE shows the superior non-dominated According to the Friedman’s rank test at 95% significant level, MO-
front. MASOS performs the best among the implemented MHs followed
In addition, Friedman’s rank test, a statistical measure, is em- by MOASOS and MOSOS. Also, the results show that both versions
ployed to rank the MHs based on the solutions found by the var- of MOSOS are better than its basic version and previous studies.
ious optimizers. The five structural problems are addressed in the The results from the Friedman’s rank test also indicates the signif-
following units. icant difference among the considered MHs.
The front STE metric is considered, and the solutions are shown
5.1. A 10-bar truss in Table 3. According to the Friedman’s rank at 95% significant
level, MOAS beats other MHs followed by MOMASOS and MOA-
Fig. 5 presents the 10-bar truss which is a simplest and widely SOS, and similar results expressed as per mean of front STE. Also,
used truss problem compare to others. The truss properties and MOMASOS and MOASOS perform superior than MOSOS.
constraints are presented in Table 1. Fig. 5 also presents the length Fig. 6 illustrates the best Pareto fronts of the considered MHs.
of each element, loading conditions, and support conditions of this It should be noted that best Pareto fronts obtained using MOAS
truss. The discrete design variables (i.e. elemental cross-sectional and MOACS are discontinuous. On the contrary, Pareto fronts ob-
areas) are assumed from forty-two discrete sections as 1.62, 1.8, tained using the proposed MHs are continuous, smooth, and have
1.99, 2.13, 2.38, 2.62, 2.63, 2.88, 2.93, 3.09, 3.13, 3.38, 3.47, 3.55, a wide range of diverse results, and the results are well distributed.
3.63, 3.84, 3.87, 3.88, 4.18, 4.22, 4.49, 4.59, 4.8, 4.97, 5.12, 5.74, 7.22, Overall, these tests validate that MOMASOS is better performer fol-
7.97, 11.5, 13.5, 13.9, 14.2, 15.5, 16, 16.9, 18.8, 19.9, 22, 22.9, 26.5, lowed by MOASOS and considered improvements upgrade the effi-
30, and 33.5 in2 as per the previous studies (Angelo et al., 2012, ciency of MOSOS.
2015; Tejani et al., 2018c).
Table 2 presents the HV values for 100 optimization runs of
this truss. The best, mean, and STD values of HV are considered 5.2. A 25-bar space truss
to measure the effectiveness of the considered MHs statistically.
The best mean solutions obtained by MOAS, MOACS, MOSOS, MOA- The 25-bar truss is illustrated in Fig. 7. The truss properties
SOS, and MOMASOS are 50,902.21, 53,639.77, 56,055.02, 56,220.00, and constraints are presented in Table 1. Loading is assumed
and 56,389.83 respectively. Also, the STD obtained using MOAS, as Px1 = 1 Klb, Py1 = Pz1 = Py2 = Pz2 = − 10 Klb, Px3 = 0.5 Klb,
MOACS, MOSOS, MOASOS, and MOMASOS are 1294.12, 307.79, Px6 = 0.6 Klb. Twenty-five elements are clubbed into eight groups
432 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

Table 3
The front Spacing-to-Extent values of results obtained for the 10-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 0.005387 0.024711 0.010590 0.003781 231 1

MOACS 0.007219 0.029625 0.014219 0.004558 385 5
MOSOS 0.007835 0.035081 0.011886 0.003791 318 4
MOASOS 0.007705 0.021266 0.011380 0.002189 305 3
MOMASOS 0.008818 0.018366 0.010721 0.001259 261 2

Table 4
The hypervolume values of results obtained for the 25-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 1848.04 1902.35 1878.74 9.77 123 5

MOACS 1850.64 1918.92 1890.61 14.39 177 4
MOSOS 1937.75 1940.43 1939.42 0.54 329 3
MOASOS 1938.75 1940.75 1939.84 0.51 371 2
MOMASOS 1944.13 1946.51 1945.61 0.45 500 1

as per symmetry about x–z and y–z planes (Angelo et al., 2012,
2015; Tejani et al., 2018c). The discrete design variables (i.e. ele-
mental cross-sections) are taken from thirty discrete sections as
0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6,
1.7, 1.8, 1.9, 2, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.8, 3, 3.2, and 3.4 in2 .
Table 4 compares the HV values for 100 independent optimiza-
tion runs found from this work. The best, mean, and STD values of
HV are given and will be considered to measure the performance
of the considered MHs statistically. The best mean results reported
by MOAS, MOACS, MOSOS, MOASOS, and MOMASOS are 1878.74,
1890.61, 1939.42, 1939.84, and 1945.61 respectively. Also, the STD
obtained using MOAS, MOACS, MOSOS, MOASOS, and MOMASOS
are 9.77, 14.39, 0.54, 0.51, and 0.45 respectively. It is found from
the results that MOMASOS gives the finest convergence and con-
sistency followed by MOASOS. Based on the Friedman’s rank test
at 95% significant measure, MOMASOS & MOASOS are the best &
second-best players. Here also results show that both the versions
of MOSOS are better than its basic version and previously used
algorithms such as MOAS and MOACS; and MOMASOS variant is
better than all the implemented MHs. The results from the Fried-
man’s rank test also indicates the significant difference among the
considered algorithms.
Fig. 6. Best Pareto fronts of the 10-bar truss. The front STE is tested for the truss and the findings are pre-
sented in Table 5. According to the Friedman’s rank at 95% signifi-
cant level, MOMASOS, MOSOS, and MOASOS rank first, second, and
third respectively and mean of front STE values obtain similar re-
sults. Also, MOMASOS outperforms its basic version.
Fig. 8 presents the best Pareto fronts of the considered MHs. It
should be noted that best Pareto fronts obtained using MOAS and
MOACS are slightly discontinuous. On the contrary, Pareto fronts
obtained using the proposed MHs are continuous, smooth, and
have a wide range of diverse results, and the results are well dis-
tributed. Overall, these tests validate that MOMASOS is a fairly su-
perior performer compare to others like the10-bar truss and these
improvements elevate the efficacy of MOSOS.

5.3. A 60-bar ring truss

The 60-bar ring truss is illustrated in Fig. 9. The mechanical

properties and limits are presented in Table 1. Sixty elements are
grouped into twenty-five in view of symmetry similar to previous
studies (Angelo et al., 2012, 2015; Tejani et al., 2018c). Multiple
loading is assumed as load case 1: Px1 = − 10 Klb and Px7 = 9 Klb,
load case 2: Px15 = Px18 = − 8 Klb and Py15 = Py18 = 3 Klb, and load
case 3: Px22 = − 20 Klb and Py22 = 10 Klb. The discrete design vari-
Fig. 7. The 25-bar space truss. ables (i.e. elemental cross-sections) are chosen from forty-five dis-
crete sections as [0.5, 0.6, 0.7, …,4.9] in2 .
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 433

Table 5
The front Spacing-to-Extent values of results obtained for the 25-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 0.007937 0.058983 0.022595 0.008424 445 5

MOACS 0.005254 0.044937 0.017026 0.008361 354 4
MOSOS 0.011763 0.013790 0.013255 0.0 0 0364 294 2
MOASOS 0.012350 0.014755 0.013326 0.0 0 0332 306 3
MOMASOS 0.005622 0.006911 0.006569 0.0 0 0213 101 1

Table 6
The hypervolume values of results obtained for the 60-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 2465.08 3397.56 3179.88 166.65 173 4

MOACS 2905.27 3276.04 3106.68 74.18 127 5
MOSOS 4271.94 4304.66 4293.25 5.92 327 3
MOASOS 4290.28 4302.54 4297.03 2.81 373 2
MOMASOS 4303.67 4316.33 4311.69 2.30 500 1

Fig. 8. Best Pareto fronts of the 25-bar truss.

Table 6 presents the HV obtained for the truss. The best, mean,
and STD values of HV are specified and will be considered to mea- Fig. 9. The 60-bar ring truss.
sure the performance of the considered MHs statistically. The best
mean results reported by MOAS, MOACS, MOSOS, MOASOS, and
MOMASOS are 3179.88, 3106.68, 4293.25, 4297.03, and 4311.69 re- and MOASOS and similar outcomes are obtained as per mean of
spectively. Also, the STD obtained using MOAS, MOACS, MOSOS, front STE. Also, MOMASOS performs the better compared to its ba-
MOASOS, and MOMASOS are 166.65, 74.18, 5.92, 2.81, and 2.30 sic version.
respectively. This is observed from the assessment that the best Fig. 10 shows the best Pareto fronts for all the proposed MHs.
convergence and search consistency are obtained for MOMASOS It is observed that the best Pareto fronts obtained using MOAS and
while the second-best is MOASOS. The Friedman’s rank test is used MOACS are discontinuous and the results are distributed in a small
to compare different algorithms based on the ranks. According to region. On the contrary, Pareto fronts obtained using the proposed
the Friedman’s rank test at 95% significant level, MOMASOS per- MHs are continuous, stable, and have a wide range of diverse re-
forms the best among the considered MHs followed by MOASOS sults, and the results are well distributed. Overall, it is determined
and MOSOS. Moreover, the results show that both the versions of that MOMASOS is slightly better performer compare to the other
MOSOS are better than its basic version and previous studies such MHs and considered improvements upsurges efficacy of MOSOS.
as MOAS and MOACS. MOMASOS still obtains the maximum HV for
this truss. The results Friedman’s rank test also indicates the signif- 5.4. A 72-bar space truss
icant difference among the considered algorithms. The conclusion
based on the Friedman’s rank test is that MOMASOS and MOASOS The 72-bar truss is illustrated in Fig. 11. The truss proper-
are again the top two performers. ties and constraints are presented in Table 1. Multiple loading is
The front STE is considered for the truss and the findings are supposed as load case 1: F1x = F1y = 5 kips and F1z = − 5 kips and
illustrated in Table 7. According to the Friedman’s rank test at 95% load case 2: F1z = F2z = F3z = F4z = −5 kips. Seventy-two elements
significant level, MOMASOS beats other MHs followed by MOSOS are grouped into sixteen in view of symmetry similar to previous
434 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

Table 7
The front Spacing-to-Extent values of results obtained for the 60-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 0.009977 0.133920 0.034915 0.019500 451 5

MOACS 0.007890 0.074504 0.029912 0.013732 436 4
MOSOS 0.010137 0.012876 0.012025 0.0 0 0522 248 2
MOASOS 0.010783 0.012879 0.012147 0.0 0 0419 265 3
MOMASOS 0.005460 0.007324 0.006247 0.0 0 040 0 100 1

studies (Angelo et al., 2012, 2015; Tejani et al., 2018c). The discrete
design variables (i.e. elemental cross-sectional areas) are assumed
from twenty-five discrete sections as [0.1, 0.2, 0.3,…, 2.5] in2 .
Table 8 shows the HV values obtained from this work. The best,
mean, and STD values of HV are specified and will be consid-
ered to measure the effectiveness of the various MHs statistically.
The best mean results reported by MOAS, MOACS, MOSOS, MOA-
SOS, and MOMASOS are 2094.40, 2097.08, 2223.81, 2227.73, and
2233.05 respectively. Also, the STD obtained using MOAS, MOACS,
MOSOS, MOASOS, and MOMASOS are 10.01, 18.78, 1.81, 1.38, and
1.05 respectively. It is found from the results that MOMASOS per-
forms the best followed by MOASOS and MOSOS as per the mea-
sure of search consistency. The Friedman’s rank test is used to
compare different MHs based on the ranks. According to the Fried-
man’s rank test at 95% significant level, MOMASOS performs the
best among the considered MHs followed by MOASOS and MOSOS.
Also, the results show that both the versions of MOSOS are better
than its basic version and the previously used MHs such as MOAS,
and MOACS; and MOMASOS variant is better than all the consid-
ered MHs. The results obtained from using the Friedman’s rank test
also indicates the significant difference among the considered MHs.
Fig. 10. Best Pareto fronts of the 60-bar truss.

Fig. 11. The 72-bar space truss.

 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 435

Table 8
The hypervolume values of results obtained for the 72-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 2066.98 2116.65 2094.40 10.01 142 5

MOACS 2039.80 2129.58 2097.08 18.78 158 4
MOSOS 2220.29 2227.53 2223.81 1.81 303 3
MOASOS 2223.28 2231.00 2227.73 1.38 397 2
MOMASOS 2229.91 2235.22 2233.05 1.05 500 1

Table 9
The front Spacing-to-Extent values of results obtained for the 72-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 0.010919 0.043568 0.022728 0.007183 422 4

MOACS 0.007918 0.076088 0.026837 0.013808 429 5
MOSOS 0.013350 0.015643 0.014393 0.0 0 0435 289 3
MOASOS 0.013026 0.015822 0.014224 0.0 0 0481 260 2
MOMASOS 0.006386 0.007519 0.006910 0.0 0 0202 100 1

Table 10
The hypervolume values of results obtained for the 942-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 44,994,042.34 56,957,732.05 51,655,929.40 3,380,287.34 150 4

MOACS 49,107,320.86 54,406,191.49 52,288,426.39 939,066.55 150 4
MOSOS 60,523,090.07 63,407,711.68 62,031,040.61 575,341.58 330 3
MOASOS 60,620,317.17 63,708,998.98 62,540,452.51 655,217.19 370 2
MOMASOS 64,805,943.66 112,438,733.19 80,006,236.39 7,903,607.97 500 1

Table 11
The front Spacing-to-Extent values of results obtained for the 942-bar truss.

Algorithms Min Max Mean STD Friedman test Friedman rank

MOAS 0.014273 0.120749 0.042659 0.020981 469 5

MOACS 0.010403 0.079907 0.029013 0.014678 401 4
MOSOS 0.012192 0.016924 0.014856 0.0 0 0761 257 2
MOASOS 0.013294 0.016218 0.015050 0.0 0 0637 273 3
MOMASOS 0.004880 0.008056 0.006423 0.0 0 0566 100 1

Fig. 12 shows the best Pareto fronts obtained all MHs. It is

observed that the best Pareto fronts obtained using MOAS and
MOACS are discontinuous and the results are distributed in a small
region. On the contrary, Pareto fronts obtained using the proposed
MHs are continuous, stable, and have a wide range of diverse re-
sults, and the results are well distributed. Overall, it is noticed that
MOMASOS is slightly better performer compare to the other MHs
and considered improvements upsurges efficacy of MOSOS.

5.5. A 942-bar tower truss

The 942-bar truss is illustrated in Fig. 13. The mechanical prop-

erties and limits are presented in Table 1. Vertical loading along
Z-axis is –3, –6, and –9 kips at each node in the 1st, 2nd, and
3rd sections, respectively; lateral loading along X-axis is 1.5 and
1.0 kips at each node on the right and left sides of the tower
truss; and Lateral loading along Y-axis is 1 kips at each node, re-
spectively. The discrete design variables (i.e. cross-sections) are se-
lected from two-hundred discrete sections as [1, 2, 3,…, 200] in2 .
The bars are clustered into fifty-nine clusters to see structural sim-
Fig. 12. Best Pareto fronts of the 72-bar truss. ilarity (Angelo et al., 2012, 2015; Tejani et al., 2018c).
Table 10 presents the HV values calculated for the truss. The
best, mean, and STD values of HV are shown and will be con-
The front STE is considered for the truss and the findings are sidered to measure the effectiveness of the considered MHs. The
shown in Table 9. It is noticed as per the Friedman’s rank at best mean results obtained by MOAS, MOACS, MOSOS, MOASOS,
95% significant level, the MOMASOS, MOASOS, and MOSOS are top and MOMASOS are 51,655,929.40, 52,288,426.39, 62,031,040.61,
three performer in this order and mean values of front STE also 62,540,452.51, and 80,006,236.39 respectively. Also, STD obtained
obtain similar results. Also, the MOMASOS and MOASOS perform using MOAS, MOACS, MOSOS, MOASOS, and MOMASOS are
superior compare to MOSOS. 3,380,287.34, 939,066.55, 575,341.58, 655,217.19, and 7,903,607.97
436 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

Fig. 14. Best Pareto fronts of the 942-bar truss.

respectively. It is found from the results that MOMASOS performs

the best followed by MOASOS and MOSOS as per the measure of
search consistency. In this large-scale problem case, it is shown
that, with the use of the normal line method, MOMASOS is con-
siderably superior to those using the crowd comparison as with
NSGAII. The Friedman’s rank test is used to compare different MHs
based on the ranks. According to the Friedman’s rank test at 95%
significant level, MOMASOS performs the best among the con-
sidered MHs followed by MOASOS and MOSOS. Here also results
demonstrate that both the versions of MOSOS are better than its
basic version and previous studies such as MOAS, and MOACS; and
MOMASOS variant is better optimizer compare to other MHs. The
results obtained from the Friedman’s rank test also indicates the
significant difference among the considered MHs.
The front STE are tested for the truss and the findings are pre-
sented in Table 11. According to the Friedman’s rank at 95% sig-
nificant level, MOMASOS, MOSOS, and MOASOS are top three per-
former in this order and mean values of front STE also obtain sim-
ilar results. Also, MOMASOS performs better compared to its basic
version.
Fig. 14 shows the best Pareto fronts found for considered MHs.
It is observed that the best Pareto fronts obtained using MOAS and
MOACS are discontinuous and the results are distributed in a small
region. On the contrary, Pareto fronts obtained using the proposed
MOMASOS, MOASOS, and MOSOS are continuous, stable, and have
a wide range of diverse results, and the results are well distributed.
The Pareto fronts found using the proposed MHs are superior com-
pare to MOAS and MOACS. Fig. 15 presents the convergence history
of the truss using MOSOS and MOMASOS. The higher hypervolume
value the better the Pareto front. It has been found that MOMA-
SOS has very good convergence rate compared to MOSOS and thus
owns better convergence characteristics for the truss problems.
Overall, it can be observed that MOMASOS performs superior
than the other considered MHs and proposed improvements lead
to better efficacy of MOSOS.

6. Engineering benchmark problems

Fig. 13. The 942-bar tower truss. In this section, four well-known engineering design optimiza-
tion problems (Mirjalili, Jangir, & Saremi, 2017) are employed to
validate the efficiency of MOMASOS. The first design problem is
speed reducer design to minimize weight and stress of the speed
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 437

Table 12
The results obtained for the speed reducer design optimization problem (the previous results are adopted from Mirjalili et al.,
2017).

GD MS S IGD Hypervolume

Algorithms Mean STD Mean STD Mean STD Mean STD Mean STD

MOPSO 0.9883 0.1789 — — 16.6850 2.6960 — — — —

NSGA-II 9.8437 7.0810 — — 2.7654 3.5340 — — — —
MOALO 1.1767 0.2327 0.8390 0.1267 1.7706 2.7690 0.8672 0.1490 — —
MOSOS 1.6794 0.1358 0.4977 0.0082 39.3635 4.2992 0.0 0 05 3.63E-05 1,864,894 19,141
MOMASOS 0.8501 0.1152 0.4953 0.0111 14.8755 1.9790 0.0 0 03 3.08E-05 1,893,295 18,865

Table 13
The results obtained for the disk brake design optimization problem (the previous results are adopted from
Mirjalili et al., 2017).

GD MS S IGD Hypervolume

Algorithms Mean STD Mean STD Mean STD Mean STD Mean STD

MOPSO 0.0244 0.1231 0.4604 0.1096 — — — — — —

NSGA-II 3.0771 0.1078 0.7972 0.0661 — — — — — —
MOALO 0.0011 0.0025 0.4496 0.0543 0.0421 0.0058 0.0194 0.0 0 08 — —
MOSOS 0.0055 0.0 0 05 0.5649 0.0066 0.1687 0.0121 0.0 0 06 0.0 0 01 5.75 0.04
MOMASOS 0.0029 0.0 0 03 0.5121 0.0041 0.0403 0.0032 0.0 0 03 3.65E-05 5.94 0.05

Table 14
The results obtained for the welded beam design optimization problem (the previous results are adopted from
Mirjalili et al., 2017).

GD MS S IGD Hypervolume

Algorithms Mean STD Mean STD Mean STD Mean STD Mean STD

MOPSO 0.3741 0.0422 — — 2.5303 0.2270 — — — —

NSGA-II 0.3601 0.0470 — — 2.3635 0.2550 — — — —
MOALO 0.1264 0.0327 0.3700 0.0025 1.1805 0.1440 0.1062 0.0152 — —
MOALO 0.0240 0.0060 0.4694 0.0506 0.3469 0.0554 0.0012 0.0012 1.31 0.23
MOMASOS 0.0123 0.0036 0.4474 0.0625 0.1424 0.0156 0.0010 0.0012 1.54 0.32

are tested for 30 independent runs. The Generational Distance (GD)

(Veldhuizen & Lamont, 20 0 0), Maximum Spread (MS), and metric
of spacing (S) (shown in Eq. (13); Schott, 1995), Inverted Gener-
ational Distance (IGD) (Sierra & Coello, 2005), and Hypervolume.
Smaller values of these measure show the superior Pareto front ex-
cept for hypervolume.

|P|
i=1 (di )2
GD = (16)
|P|
where ‘|P|’ is count of obtained Pareto optimal solutions and di is
the Euclidean distance between the ith Pareto optimal solution and
the adjacent neighbor true Pareto optimal solution.

Fig. 15. Convergence graphs of the 942-bar truss. 

f
MS = max(d (ai , bi ) ) (17)
i=1

reducer subject to eleven behavior constraints and seven side con- where f is count of objective functions, and d counts the Euclidean
straints. The second design problem is disk brake design to mini- distance, ai and bi are the highest and lowest values of ith objec-
mize stopping time and mass of the brake subject to five behav- tive function respectively.
ior constraints and four side constraints. The third design problem
|P  |  2
is welded beam design to minimize fabrication cost and deflec- i=1
di
tion of the beam subject to four behavior constraints and four side IGD = (18)
|P|
constraints. And the last design problem is cantilever beam design
to minimize weight and end deflection of a cantilever beam sub- where ‘|P |’ is the true Pareto optimal solutions and di is the Eu-
ject to two behavior constraints and two side constraints. The de- clidean distance between the i-th true Pareto optimal obtained so-
tails of the considered problems can be studies from Ray and Liew lution and the adjacent neighbor true Pareto optimal solution. The
(2002) and Deb, Zhu, and Kulkarni (2015). For fair comparison, the engineering design problems (i.e. speed reducer design, disk brake
similar parameters (Mirjalili et al., 2017) and constrained handling design, welded beam design, and cantilever beam design) used in
techniques are followed in this study. Thus, all the problems were this study are stated Ray & Liew, 2002 and Deb et al., 2015.
performed with the population size of 100, functional evaluations Tables 12–15 compare results obtained for the considered engi-
size of 10,0 0 0, and an archive size of 100. The proposed algorithms neering design problems using multiple objective particle swarm
438 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

Table 15
The results obtained for the cantilever beam design optimization problem (the previous results are adopted from Mirjalili et al.,
2017).

GD MS S IGD Hypervolume

Algorithms Mean STD Mean STD Mean STD Mean STD Mean STD

MOPSO — — — — — — — — — —
NSGA-II — — — — — — — — — —
MOALO 0.0 0 02 1.62E-05 0.7673 0.1685 0.0083 0.0029 0.0 0 02 0.0 0 01 — —
MOSOS 0.0 0 03 1.71E-05 0.3755 0.0 0 09 0.0359 0.0036 4.80E-05 6.50E-06 0.5595 0.0 0 05
MOMASOS 0.0 0 02 3.60E-06 0.3748 0.0 0 02 0.0170 0.0010 3.32E-05 2.82E-06 0.5597 0.0 0 01

Fig. 16. Best Pareto optimal front obtained by the MOSOS and MOMASOS on the speed reduced design problem.

Fig. 17. Best Pareto optimal front obtained by the MOSOS and MOMASOS on the disk brake design problem.

optimization (MOPSO) (Coello Coello & Lechuga, 2002), NSGA-II is noted that Pareto optimal front obtained using MOMASOS is
(Deb et al., 2002), Multi-Objective Ant Lion Optimizer (MOALO) uniformly distributed and near to true Pareto front compare to
(Mirjalili et al., 2017), MOSOS, and MOMASOS with the various MOSOS.
measures such as GD, MS, S, IGD, and hypervolume obtained. As The best Pareto optimal fronts attained by MOSOS and MOMA-
per the results shown in result tables, it can be concluded that SOS for disk brake design problem is shown in Fig. 17. It can be
MOMASOS gives the best results compare to the true Pareto front observed that Pareto optimal front obtained using MOMASOS is
and the minimum values of D, MS, S, and IGD. Also, the hypervol- widely spread compare to MOSOS and also have majority part on
ume of MOMASOS seems better compare to MOSOS. the true Pareto front.
The best Pareto optimal fronts obtained by MOSOS and MO- The best Pareto optimal fronts obtained by MOSOS and MOMA-
MASOS for speed reducer design problem is shown in Fig. 16. It SOS for disk brake design problem is shown in Fig. 18. It is noted
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 439

Fig. 18. Best Pareto optimal front obtained by the MOSOS and MOMASOS on the welded beam design problem.

Fig. 19. Best Pareto optimal front obtained by the MOSOS and MOMASOS on the cantilever beam design problem.

that Pareto optimal front obtained using MOMASOS is broadly dis- weight, displacements, stresses, and geometrical configurations. Al-
tributed compare to MOSOS and also have majority part on the though multiple improvements have been reported in past litera-
true Pareto front. ture, the problems are becoming more complex and challenging to
The best Pareto optimal fronts obtained by MOSOS and MOMA- be solved using the existing meta-heuristic algorithms. Thus, the
SOS for cantilever beam design problem is shown in Fig. 19. It is meta-heuristic algorithms need to be improved in terms of effi-
noted that both Pareto optimal fronts are nearly identical and on ciency and fitting specific problems.
the true Pareto front. A modified version of MOSOS is proposed in this paper. The
Overall, it can be observed that MOMASOS performs superior new algorithm is modified in such a way to improve both the
than the methods that available in the previous literature. It can exploration and exploitation of MOSOS in reproducing design so-
be seen from the results that the proposed improvements lead to lutions for multi-objective optimization. The comparative perfor-
better efficacy of MOSOS. mance studies, based on the hypervolume and spacing-to-extent
indicators, reveal that the proposed MOMASOS outperforms its
7. Conclusions original MOSOS and other multi-objective meta-heuristics imple-
mented in multi-objective truss optimization and engineering de-
Due to the high performance and reasonable quality of obtained sign optimization problems. The proposed adaptive mutualism and
results in complex problems, nature-inspired meta-heuristics be- modified parasitism phases significantly improve the performance
come an important field in expert and intelligent systems. Many of MOSOS.
decision-making problems in engineering are highly nonlinear The contribution of this paper is twofold. First, we proposed
and challenging to be solved using traditional methods. Truss two new modifications along with a normal line method as
design optimization problems constitute a large number of de- an archiving technique on an existing MOSOS to improve the
sign variables and complex objective and constraints including quality of obtained results for multi-objective optimization in
440 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441

complex truss structure and engineering design problems. Two Deb, K., & Jain, H. (2013). An evolutionary many-objective optimization algorithm
modifications are introduced including the addition of adaptive using reference-point based non-dominated sorting approach, Part I: Solving
Problems with Box Constraints. IEEE Transactions on Evolutionary Computation,
parameters in mutualism and an improved parasitism phase in 18(4), 577–601 doi:oRg/10.1109/TEVC.2013.2281534.
order to significantly improve both the exploration and ex- Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. (2002). A fast and elitist multiobjec-
ploitation of MOSOS. Second, the successful improvement and tive genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation,
6(2), 182–197 doi:oRg/10.1109/4235.996017.
applications of meta-heuristic algorithms are in greater need in Deb, K., Zhu, L., & Kulkarni, S. (2015). Multi-scenario, multi-objective optimization
the background of expert and intelligent systems and it provides using evolutionary algorithms: Initial Results. In 2015 IEEE congress on evolu-
a potential alternative for solving more complex and challenging tionary computation (CEC’2015) (pp. 1877–1884).
Degertekin, S. O., & Hayalioglu, M. S. (2013). Sizing truss structures using teaching-
problems that cannot be solved using the existing meta-heuristic
learning-based optimization. Computers and Structures, 119, 177–188 doi:oRg/10.
algorithms. 1016/j.compstruc.2012.12.011.
Our future work is to extend the proposed MOMASOS for solv- Do, D. T. T., & Lee, J. (2017). A modified symbiotic organisms search (mSOS) al-
gorithm for optimization of pin-jointed structures. Applied soft computing: 61.
ing reliability optimization of trusses. Once uncertainties or ran-
Elsevier B.V. doi:10.1016/j.asoc.2017.08.002.
dom variables are taken into consideration, the truss design prob- Ezugwu, A. E. S., & Adewumi, A. O. (2017). Discrete symbiotic organisms search al-
lem is considered robust or reliability optimization which is more gorithm for travelling salesman problem. Expert Systems with Applications, 87,
complex than deterministic optimization. The powerful MOMASOS 70–78 doi:oRg/10.1016/j.eswa.2017.06.007.
Ezugwu, A. E. S., Adewumi, A. O., & Frîncu, M. E. (2017). Simulated annealing
should positively respond to these difficulties. based symbiotic organisms search optimization algorithm for traveling sales-
man problem. Expert Systems with Applications, 77, 189–210. https://doi.org/10.
1016/j.eswa.2017.01.053.
Credit authorship contribution statement Fleury, C. (1980). An efficient optimality criteria approach to the minimum weight
design of elastic structures. Computers and Structures, 11(3), 163–173 doi:oRg/10.
Nantiwat Pholdee: Data curation. Sujin Bureerat: Data cura- 1016/0045-7949(80)90155-8.
Fonseca, C. M., & Fleming, P. J. (1993). Multiobjective genetic algorithms. IEE Collo-
tion. Doddy Prayogo: Formal analysis. Amir H. Gandomi: Concep-
quium on Genetic Algorithms for Control Systems Engineering, 1–6.
tualization, Data curation, Formal analysis. Gomes, H. M. (2011). Truss optimization with dynamic constraints using a particle
swarm algorithm. Expert Systems with Applications, 38(1), 957–968 doi:oRg/10.
1016/j.eswa.2010.07.086.
Acknowledgment Greiner, D., & Hajela, P. (2012). Truss topology optimization for mass and relia-
bility considerations — Co-evolutionary multiobjective formulations. Structural
The authors are grateful for the support from the Thailand Re- and Multidisciplinary Optimization, 45, 589–613 doi:oRg/10.10 07/s0 0158-011-
0709-9.
search Fund (TRF), grant number RTA6180010. Guha, D., Roy, P., & Banerjee, S. (2017). Quasi-oppositional symbiotic organism
search algorithm applied to load frequency control. Swarm and Evolutionary
References Computation, 33, 46–67. https://doi.org/10.1016/j.swevo.2016.10.001.
Gandomi, A. H., Talatahari, S., Yang, X.-S., & Deb, S. (2012). Design optimization of
Abdullahi, M., Ngadi, M. A., & Abdulhamid, S. M. (2016). Symbiotic organism search truss structures using cuckoo search algorithm. The Structural Design of Tall and
optimization based task scheduling in cloud computing environment. Future Special Buildings, 22, 1330–1349. https://doi.org/10.1002/tal.1033.
Generation Computer Systems, 56, 640–650. doi:10.1016/j.future.2015.08.006. Gandomi, A. H., Talatahari, S., Tadbiri, F., & Alavi, A. H. (2013). Krill herd algorithm
Allwood, R. J., & Chung, Y. S. (1984). Minimum-weight design of trusses by an opti- for optimum design of truss structures. International Journal of Bio-Inspired Com-
mality criteria method. International Journal for Numerical Methods in Engineer- putation, 5(5), 281. https://doi.org/10.1504/IJBIC.2013.057191.
ing, 20(4), 697–713 doi:oRg/10.10 02/nme.162020 0409. Gandomi, A. H., & Goldman, B. W. (2018). Parameter-less population pyramid for
Angelo, J. S., Barbosa, H. J. C., & Bernardino, H. S. (2012). Multi-objective ant colony large-scale tower optimization. Expert Systems with Applications, 96, 175–184.
approaches for structural optimization problems. In Proc. of the eleventh interna- https://doi.org/10.1016/j.eswa.2017.11.047.
tional conference on computational structures technology. doi:10.4203/ccp.99.66. Ho-Huu, V., Duong-Gia, D., Vo-Duy, T., Le-Duc, T., & Nguyen-Thoi, T. (2018). An ef-
Angelo, J. S., Bernardino, H. S., & Barbosa, H. J. C. (2015). Ant colony approaches for ficient combination of multi-objective evolutionary optimization and reliability
multiobjective structural optimization problems with a cardinality constraint. analysis for reliability-based design optimization of truss structures. Expert Sys-
Advances in Engineering Software, 80(C), 101–115. doi:10.1016/j.advengsoft.2014. tems with Applications, 102, 262–272 doi:oRg/10.1016/j.eswa.2018.02.040.
09.015. Hosseini, S. S., Hamidi, S. A., Mansuri, M., & Ghoddosian, A. (2015). Multi objective
Ayala, H. V. H., Klein, C. E., Mariani, V. C., & Coelho, L. D. S. (2017). Multiobjec- particle swarm optimization (MOPSO) for size and shape optimization of 2D
tive symbiotic search algorithm approaches for electromagnetic optimization. truss structures. Periodica Polytechnica Civil Engineering, 59(1), 9–14 doi:oRg/10.
IEEE Transactions on Magnetics, 53(6), 2015–2018 doi:oRg/10.1109/TMAG.2017. 3311/PPci.7341.
2665350. Jain, H., & Deb, K. (2013). An evolutionary many-objective optimization algorithm
Bureerat, S., & Pholdee, N. (2015). Optimal truss sizing using an adaptive differential using reference-point based non-dominated sorting approach, Part II: Handling
evolution algorithm. Journal of Computing in Civil Engineering, 4015019 doi:oRg/ constraints and extending to an adaptive approach. IEEE Transactions on Evolu-
10.1061/(ASCE)CP.1943-5487.0 0 0 0487. tionary Computation, 18(4), 602–622.
Bureerat, S., & Srisomporn, S. (2010). Optimum plate-fin heat sinks by using a Kaveh, A., & Khayatazad, M. (2013). Ray optimization for size and shape optimiza-
multi-objective evolutionary algorithm. Engineering Optimization, 42(4), 305–323 tion of truss structures. Computers and Structures, 117, 82–94 doi:oRg/10.1016/j.
doi:oRg/10.1080/03052150903143935. compstruc.2012.12.010.
Camp, C. V., & Farshchin, M. (2014). Design of space trusses using modified Kaveh, A., & Laknejadi, K. (2011). A novel hybrid charge system search and parti-
teaching–learning based optimization. Engineering Structures, 62–63, 87–97 cle swarm optimization method for multi-objective optimization. Expert Systems
doi:oRg/10.1016/j.engstruct.2014.01.020. with Applications, 38(12), 15475–15488 doi:oRg/10.1016/j.eswa.2011.06.012.
Cheng, M.-Y., Prayogo, D., & Wu, Y.-W. (2018). Prediction of permanent deformation Kaveh, A., & Laknejadi, K. (2013). A hybrid evolutionary graph-based multi-objective
in asphalt pavements using a novel symbiotic organisms search–least squares algorithm for layout optimization of truss structures. Acta Mechanica, 224, 343–
support vector regression. Neural Computing and Applications. 364 doi:oRg/10.10 07/s0 0707- 012- 0754- 5.
Cheng, M.-Y., & Prayogo, D. (2014). Symbiotic organisms search: A new metaheuris- Kaveh, a., & Mahdavi, V. R. (2014). Colliding Bodies Optimization method for opti-
tic optimization algorithm. In Computers and structures: 139 (pp. 98–112). Else- mum design of truss structures with continuous variables. Advances in Engineer-
vier Ltd. doi:10.1016/j.compstruc.2014.03.007. ing Software, 70, 1–12 doi:oRg/10.1016/j.advengsoft.2014.01.002.
Cheng, M.-Y., Prayogo, D., & Tran, D.-H. (2015). Optimizing multiple-resources lev- Kaveh, A., & Mahdavi, V. R. (2018). Multi-objective colliding bodies optimization al-
eling in multiple projects using discrete symbiotic organisms search. Journal of gorithm for design of trusses. Journal of Computational Design and Engineering,
Computing in Civil Engineering. doi:10.1061/(ASCE)CP.1943-5487.0 0 0 0512. 1–11 doi:oRg/10.1016/j.jcde.2018.04.001.
Çelik, E., & Durgut, R. (2018). Performance enhancement of automatic voltage reg- Kaveh, A., & Zakian, P. (2017). Improved GWO algorithm for optimal de-
ulator by modified cost function and symbiotic organisms search algorithm. sign of truss structures. Engineering with Computers, 1–23 doi:oRg/10.1007/
Engineering Science and Technology, an International Journal, 21(5), 1104–1111. s00366- 017- 0567- 1.
https://doi.org/10.1016/j.jestch.2018.08.006. Kumar, S., Tejani, G. G., & Mirjalili, S. (2018). Modified symbiotic organisms search
Çelik, E., & Öztürk, N. (2017). First application of symbiotic organisms search algo- for structural optimization. Engineering with Computers.. https://doi.org/10.1007/
rithm to off-line optimization of PI parameters for DSP-based DC motor drives. s00366- 018- 0662- y.
Neural Computing and Applications, 30(5), 1689–1699. https://doi.org/10.1007/ Lingyun, W., Mei, Z., Guangming, W., & Guang, M. (2005). Truss optimization on
s00521- 017- 3256- 5. shape and sizing with frequency constraints based on genetic algorithm. Com-
Coello Coello, C. A., & Lechuga, M. S. (2002). MOPSO: a proposal for multiple objec- putational Mechanics, 35, 361–368 doi:oRg/10.10 07/s0 0466-0 04- 0623- 8.
tive particle swarm optimization. In Proceedings of the 2002 congress on evolu- Mirjalili, S., Jangir, P., & Saremi, S. (2017). Multi-objective ant lion optimizer: A
tionary computation, CEC 2002, 2 (pp. 1051–1056). https://doi.org/10.1109/CEC. multi-objective optimization algorithm for solving engineering problems. Ap-
20 02.10 04388. plied Intelligence, 46(1), 79–95. https://doi.org/10.1007/s10489- 016- 0825- 8.
 G.G. Tejani, N. Pholdee and S. Bureerat et al. / Expert Systems With Applications 125 (2019) 425–441 441

Mousa, A. A., El-Shorbagy, M. A., & Abd-El-Wahed, W. F. (2012). Local search based Techasen, T., Wansasueb, K., Panagant, N., Pholdee, N., & Bureerat, S. (2018). Simul-
hybrid particle swarm optimization algorithm for multiobjective optimization. taneous topology, shape, and size optimization of trusses, taking account of un-
Swarm and Evolutionary Computation, 3, 1–14 doi:oRg/10.1016/j.swevo.2011.11. certainties using multi-objective evolutionary algorithms. Engineering with Com-
005. puters doi:oRg/10.10 07/s0 0366- 018- 0629- z.
Noilublao, N., & Bureerat, S. (2011). Simultaneous topology, shape and sizing opti- Tejani, G. G., Savsani, V. J., & Patel, V. K. (2016a). Adaptive symbiotic organisms
misation of a three-dimensional slender truss tower using multiobjective evo- search (SOS) algorithm for structural design optimization. Journal of Computa-
lutionary algorithms. Computers and Structures, 89(23–24), 2531–2538 doi:oRg/ tional Design and Engineering, 3(3), 226–249. doi:10.1016/j.jcde.2016.02.003.
10.1016/j.compstruc.2011.08.010. Tejani, G. G., Savsani, V. J., & Patel, V. K. (2016b). Modified sub-population teaching-
Noilublao, N., & Bureerat, S. (2013). Simultaneous topology, shape, and sizing opti- learning-based optimization for design of truss structures with natural fre-
misation of plane trusses with adaptive ground finite elements using MOEAs. quency constraints. Mechanics Based Design of Structures and Machines, 44(4),
Mathematical Problems in Engineering, 2013 doi:oRg/10.1155/2013/838102. 495–513 doi:oRg/10.1080/15397734.2015.1124023.
Panagant, N., & Bureerat, S. (2018). Truss topology, shape and sizing optimization Tejani, G. G., Savsani, V. J., Bureerat, S., & Patel, V. K. (2018a). Topology and size
by fully stressed design based on hybrid grey wolf optimization and adaptive optimization of trusses with static and dynamic bounds by modified symbiotic
differential evolution. Engineering Optimization, 0273 doi:oRg/10.1080/0305215X. organisms search. Journal of Computing in Civil Engineering, 32(2), 1–11. doi:10.
2017.1417400. 1061/(ASCE)CP.1943-5487.0 0 0 0741.
Panda, Arnapurna, & Pani, Sabyasachi (2016). A symbiotic organisms search algo- Tejani, G. G., Savsani, V. J., Bureerat, S., Patel, V. K., & Savsani, P. (2018b). Topology
rithm with adaptive penalty function to solve multi-objective constrained opti- optimization of truss subjected to static and dynamic constraints by integrat-
mization problems. Applied Soft Computing, 46, 344–360. doi:10.1016/j.asoc.2016. ing simulated annealing into passing vehicle search algorithms. Engineering with
04.030. Computers. https://doi.org/10.10 07/s0 0366- 018- 0612- 8.
Pholdee, N., & Bureerat, S. (2012). Performance enhancement of multiobjec- Tejani, G. G., Bureerat, S., Pholdee, N., & Prayogo, D. (2018c). Multiobjective adaptive
tive evolutionary optimisers for truss design using an approximate gradient. symbiotic organisms search for truss optimization problems. Knowledge-based
Computers and Structures, 106–107, 115–124 doi:oRg/10.1016/j.compstruc.2012. systems. Elsevier B.V.
04.015. Tejani, G. G., Savsani, V. J., Patel, V. K., & Mirjalili, Seyedali (2017). Truss optimiza-
Pholdee, N., & Bureerat, S. (2013a). Hybrid real-code population-based incremental tion with natural frequency bounds using improved symbiotic organisms search.
learning and approximate gradients for multi-objective truss design. Engineering Knowledge-Based systems. Elsevier B.V. doi:10.1016/j.knosys.2017.12.012.
Optimization, 46(8), 1032–1051 doi:oRg/10.1080/0305215X.2013.823194. Tran, Duc-Hoc, Luong, Duc-Long, Duong, Minh-Tin, Le, Trong-Nhan, & Pham, Anh-
Pholdee, N., & Bureerat, S. (2013b). Hybridisation of real-code population- Duc (2017). Opposition multiple objective symbiotic organisms search
based incremental learning and differential evolution for multiobjective de- (OMOSOS) for Time, cost, quality and work continuity tradeoff in repetitive
sign of trusses. Information Sciences, 223, 136–152 doi:oRg/10.1016/j.ins.2012. projects. Journal of Computational Design and Engineering. Society for Computa-
10.008. tional Design and Engineering. doi:10.1016/j.jcde.2017.11.008.
Pholdee, N., & Bureerat, S. (2017). A comparative study of eighteen self-adaptive Tran, Duc-Hoc, Cheng, Min-Yuan, & Prayogo, Doddy (2016). A novel multiple objec-
metaheuristic algorithms for truss sizing optimisation. KSCE Journal of Civil En- tive symbiotic organisms search (MOSOS) for time–cost–labor utilization trade-
gineering, 22(8), 2982–2993 doi:oRg/10.1007/s12205- 017- 0095- y. off problem. Knowledge-Based Systems, 94, 132–145. doi:10.1016/j.knosys.2015.11.
Prayogo, D., & Susanto, Y. T. T. (2018). Optimizing the prediction accuracy of friction 016.
capacity of driven piles in cohesive soil using a novel self-tuning least squares Veldhuizen, D. A., & Lamont, G. B. (20 0 0). Multiobjective evolutionary algorithms:
support vector machine. Advances in Civil Engineering, 2018. Analyzing the state-of-the-art. Evolutionary Computation, 8, 125–147.
Prayogo, D., Cheng, M.-Y., Wong, F. T., Tjandra, D., & Tran, D.-H. (2018). Optimization Wang, H., Jiao, L., & Yao, X. (2015). Two Arch2: An improved two-archive algorithm
model for construction project resource leveling using a novel modified symbi- for many-objective optimization. IEEE Transactions on Evolutionary Computation,
otic organisms search. Asian Journal of Civil Engineering. 19(4), 524–541 doi:oRg/10.1109/TEVC.2014.2350987.
Ray, T., & Liew, K. M. (2002). A swarm metaphor for multiobjective design op- Wei, L., Tang, T., Xie, X., & Shen, W. (2011). Truss optimization on shape
timization. Engineering Optimization, 34(2), 141–153. https://doi.org/10.1080/ and sizing with frequency constraints based on parallel genetic algorithm.
03052150210915. Structural and Multidisciplinary Optimization, 43(5), 665–682 doi:oRg/10.1007/
Richardson, J. N., Bouillard, P., Adriaenssens, S., Bouillard, P., & Filomeno, R. (2012). s00158-010- 0600- 0.
Multiobjective topology optimization of truss structures with kinematic stabil- Yu, V. F., Perwira Redi, A. A. N., Yang, C. L., Ruskartina, E., & Santosa, B. (2017). Sym-
ity repair. Structural and Multidisciplinary Optimization, 46, 513–532 doi:oRg/10. biotic organisms search and two solution representations for solving the ca-
10 07/s0 0158-012-0777-5. pacitated vehicle routing problem. In Applied soft computing: 52 (pp. 657–672).
Robič, T., & Filipič, B. (2005). DEMO: Differential evolution for multiobjective opti- Elsevier B.V. doi:10.1016/j.asoc.2016.10.006.
mization, 520–533. doi:oRg/10.1007/978-3-540-31880-4_36 Zhang, B., Sun, L., Yuan, H., Lv, J., & Ma, Z. (2016). An improved regularized extreme
Savsani, V. J., Tejani, G. G., & Patel, V. K. (2016). Truss topology optimization learning machine based on symbiotic organisms search. In Proceedings of the
with static and dynamic constraints using modified subpopulation teaching– 2016 IEEE 11th conference on industrial electronics and applications, ICIEA 2016
learning-based optimization. Engineering Optimization, 48(11), 1–17 doi:oRg/10. (pp. 1645–1648). doi:oRg/10.1109/ICIEA.2016.7603849.
1080/0305215X.2016.1150468. Zhang, X., Tian, Y., & Jin, Y. (2015). A knee point-driven evolutionary algorithm
Schott, J. R. (1995). Fault Tolerant Design Using Single and Multicriteria Genetic Al- for many-objective optimization. IEEE Transactions on Evolutionary Computation,
gorithm Optimization Master’s Thesis. Cambridge, Massachusetts: Department of 19(6), 761–776 doi:oRg/10.1109/TEVC.2014.2378512.
Aeronautics and Astronautics, Massachusetts Institute of Technology. Zitzler, E., Laumanns, M., & Thiele, L. (2001). SPEA2: Improving the strength pareto
Secui, D. C. (2016). A modified Symbiotic Organisms Search algorithm for large evolutionary algorithm. Evolutionary Methods for Design Optimization and Control
scale economic dispatch problem with valve-point effects. Energy, 113, 366–384 with Applications to Industrial Problems, 95–100 doi:oRg/10.1.1.28.7571.
doi:oRg/10.1016/j.energy.2016.07.056. Zitzler, E., Laumanns, M., & Thiele, L. (2002). SPEA2: Improving the strength pareto
Sierra, M. R., & Coello Coello, C. A. (2005). Improving PSO-based multi-objective evolutionary algorithm for multiobjective optimization. Evolutionary Methods for
optimization using crowding, mutation and ∈-Dominance. Evolutionary Multi- Design, Optimization and Control, 1, 1–6.
Criterion Optimization, 1, 505–519. https://doi.org/10.1007/978- 3- 540- 31880- 4_ Zuo, W., Xu, T., Zhang, H., & Xu, T. (2011). Fast structural optimization with fre-
35. quency constraints by genetic algorithm using adaptive eigenvalue reanalysis
Su, R., Wang, X., Gui, L., & Fan, Z. (2011). Multi-objective topology and sizing methods. Structural and Multidisciplinary Optimization, 43(6), 799–810 doi:oRg/
optimization of truss structures based on adaptive multi-island search strat- 10.10 07/s0 0158-010- 0610- y.
egy. Structural and Multidisciplinary Optimization, 43, 275–286 doi:oRg/10.1007/
s00158-010- 0544- 4.