Knowledge-Based Systems 253 (2022) 109591

Contents lists available at ScienceDirect

Knowledge-Based Systems
journal homepage: www.elsevier.com/locate/knosys

A Decomposition based Multi-Objective Heat Transfer Search

algorithm for structure optimization
∗
Sumit Kumar a , Pradeep Jangir b,c , Ghanshyam G. Tejani d , , Manoharan Premkumar e
a
Australian Maritime College, College of Sciences and Engineering, University of Tasmania, Launceston, 7248, Australia
b
Department of Electrical Engineering, L.E. College, Morbi, Gujarat, India
c
Rajasthan Rajya Vidyut Prasaran Nigam Limited, Sikar, Rajasthan 332025, India
d
Department of Mechanical Engineering, School of Technology, GSFC University, Vadodara, Gujarat, India
e
Department of Electrical and Electronics Engineering, Dayananda Sagar College of Engineering, Bangalore 560078, Karnataka, India

article info a b s t r a c t

Article history: Multi-objective (MO) problems are complex and challenging to solve due to multiple conflicting
Received 27 March 2021 objectives and incommensurate constraints. Heat Transfer Search is a recently introduced thermo-
Received in revised form 3 June 2022 dynamic laws-based algorithm for solving many single objective problems. But it faces criticism due
Accepted 30 July 2022
to its shortcomings including local optima trap, computational complexity, and reliability issues while
Available online 5 August 2022
solving MO problems. Decompositions is a simple yet efficient framework that aids in resolving fitness
Keywords: assignments and diversity maintenance issues of MO problems. It is also popular for reducing compu-
Multi-objective optimization tational complexity and improving solution quality. Therefore, this work proposes and investigates a
Decomposition novel, simple, and robust Decomposition-based Multi-Objective Heat Transfer Search (MOHTS/D) for
Structural optimization solving real-world structural problems. To achieve the Pareto optimal solutions and to confirm their
Metaheuristics coverage behaviour, the evenly generated weight vectors sorting and Euclidean distance strategies
Constraints problems were employed in the proposed posteriori method. The performance of MOHTS/D is investigated
through eight constrained widely accepted benchmarks. The results contrast with the MOHTS, MO
evolutionary algorithm based on decomposition, MO passing vehicle search, MO slime mould, MO
symbiotic organisms search, and MO multi-verse optimization. The efficacy of MOHTS/D is evaluated
based on hyper-volume, coverage, inverted generational distance, pure diversity, spacing, spread,
coverage Pareto front, diversity maintenance, generational distance, and runtime metrics. The results
reveal that MOHTS/D is a robust optimization approach compared to others for optimizing real-
life structural problems. MOHTS/D was able to find the optimal solution with minor computational
complexity, and the obtained solutions demonstrate a better convergence, coverage, diversity, and
spread behaviour over Pareto fronts
© 2022 Elsevier B.V. All rights reserved.

1. Introduction superior to other obtained solutions in a given search space

when all objectives were optimized contemporaneously [3,4].
1.1. Literature review Nevertheless, they may be inferior to other solutions for one or
more goals, and thus remaining solutions are called dominated
Multi-objective (MO) design optimization problems are es- solutions. Owing to the non-dominance nature of solutions of
sentially significantly different from single-objective problems Pareto set in MO issues, all solutions are competent enough, and
as MO decision-making problems involve multiple diverse ob- accordingly, different designers can choose other solutions as per
jectives that require simultaneous optimization [1,2]. Moreover, their particular design requirements or environments. Thus, it
unlike single objective problems with one best solution (often
is essential to know alternative Pareto-optimal solutions for MO
called global maximum or global minimum), a set of solutions
optimization problems [5,6].
exists in MO design issues. These solutions are known mainly
Because of its widespread application in civil, aerospace, and
as Pareto-optimal solutions or non-dominated solutions. They are
mechanical enterprises, optimal structural design has been a hot
∗ Corresponding author. topic of research in recent decades [7]. The primary goals of
structural design optimization are to have lightweight and thin
E-mail addresses: [email protected] (S. Kumar),
[email protected] (P. Jangir), [email protected] (G.G. Tejani), systems to get the minimum investment cost and the maximum
[email protected] (M. Premkumar). safety of the final design. This challenge involves not only a

https://doi.org/10.1016/j.knosys.2022.109591
0950-7051/© 2022 Elsevier B.V. All rights reserved.
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

single solution, but a number of them. Such solutions repre- optimization [49], discrete HTS to solve travelling salesman prob-
sent the possible trade-offs among the different objectives to be lem [50], simultaneous HTS [51], quadratic interpolation-based
optimized. Structure optimization problems with many compet- HTS to solve dynamic optimization problems [52], improved HTS
ing objectives and conflicting constraints including stress, side for unconstrained design issues [53]. Considering the importance
and cross-sectional area make them complex and challenging of MO optimization in practical design issues, few studies in-
in nature which necessitates effective and robust optimization vestigated the HTS performance for MO problems also. Savsani
methods to identify optimal solutions [8]. et al. [54] investigated a novel MO version of HTS called MOHTS
Classical techniques for solving these challenging, intricate, to decode the dynamic half car passive ride optimization prob-
multi-modal, multi-constraints MO optimization problems often lem and found it superior to the well-distinguished algorithm
collapse and take considerable computational time [6–8]. Thus, NSGA-II in terms of non-dominated Pareto fronts and their diver-
in many real-world challenging problems with diverse frame- sity. Tawhid and Savsani [55] investigated a ∈-constraint MOHTS
works, metaheuristics (MHs) have become increasingly fascinat- algorithm for optimization of numerous engineering design prob-
ing as a powerful tool for optimization. Harris hawks algorithm lems and found its improved efficiency in comparison to MOGA,
[9] based on cooperative behaviour and chasing style of Har- MOPSO, MODE, NSGA-II, PAES, and MOWCA. Raja et al. [56]
ris’ hawks, Equilibrium optimizer [10] inspired by mass bal- investigate the MOHTS for parameter optimization of a plate
ance models used to calculate equilibrium state, Seagull algo- heat exchanger design problem with conflicting objectives. The
rithm [11] based on seagull sea bird natural migration and at- computational investigations exhibit a fall of nearly 93% in heat
tacking behaviour, Aptenodytes forsteri algorithm [12] that imi- transfer coefficient and present a good agreement with experi-
tates emperor penguin’s warm-hugging natural behaviour, Snake mental outcomes. Similarly, Tejani et al. [57] proposed MOHTS for
algorithm [13] inspired from snakes special mating behaviour, solving structure optimization problems and found it better than
White shark algorithm [14] based on navigating and foraging MOAS, MOACS, and MOSOS algorithms. Moreover, to enhance the
actions of great white sharks, Flow direction algorithm [15], performance of MOHTS, a few modifications have been reported
Arithmetic optimizer [16] that is based on main arithmetic oper- in the literature, such as modified MOHTS [58] and improved
ators distribution behaviour, Horse herd algorithm [17] inspired MOHTS [59].
from the social performances of horses at different ages, Slime
mould algorithm [18] based on the natural oscillation mode of 1.2. Research gap and motivation of the study
slime mould, Keshtel Algorithm [19–21] inspired by Keshtel’s
feeding, social engineering optimizer [22–24] depends on strate-
One of the biggest disadvantages of many of MHs, such as
gies of an attacker attacks to a defender, Red deer algorithm
GAs and SA, is their sluggish convergence rate, which leads to
[25,26] that imitate the Scottish red deer unusual mating be-
high computational costs. Another shortcoming is the likelihood
haviour, are a few of the examples of recently developed MHs
of the solution being stuck in a local optimum like in PSO, TS, ABC,
and their applications in wide engineering, medical and supply
and ACO. To overcome these limitations, the emergence of novel
chain domain.
and improved MHs is thus rising drastically for incorporating
Due to ease, versatility, derivation-free framework, and local
their more beneficial attributes [60,61]. A few of the prominent
optimum escape, MHs procure a substantial amount of attention
algorithms are MO plasma generation optimizer (MOPGO) [62],
from many researchers. This eventually engender the develop-
MO thermal exchange optimizer (MOTEO) [63], MO Keshtel al-
ment of renowned methodologies for MO design problems as
gorithm [64], hybrid MO cuckoo search algorithm [65], MO hy-
well such as MO Genetic Algorithm (MOGA) [27], MO Simulated
bridized PVS algorithm (MOHTS-PVS) [66], and many more. For
Annealing (MOSA) [28], MO Differential Evolution (MODE) [29],
MHs, the dynamic balance between global diversification and lo-
MO Passing Vehicle Search (MOPVS) [30], MO Particle Swarm Op-
timization (MOPSO) [31], Non-dominated Sorting Genetic Algo- cal intensification is of great importance [67,68]. In principle, the
rithms (NSGA-II) [32], MO Ant Colony Optimization (MOACO) [33], terminology diversification corresponds to search space explo-
MO Symbiotic Organisms Search (MOSOS) [34,35], strength Pareto ration, while the expression intensification leads to the utilization
Evolutionary Algorithms (SPEA2) [36], similarity-based coopera- of cumulative search knowledge. As mentioned, the harmony
tive strategy [37] and multidirectional prediction approach [38] between diversification and intensification is crucial because the
for dynamic MO optimization problems. While the outcomes gen- former helps in promptly identifying the high-quality solutions
erated by MHs are not necessarily the optimal results, they can regions in the search arena, whereas the latter leads to minimal
be accomplished in a reasonable time. Moreover, the potential to time in search areas that are either already being explored or
reach the Pareto front in a single run is the most striking feature that do not offer high-quality solutions [69,70]. A quite burning
of these MHs [39–41]. question today is the quest for even more potent methods. The
The Heat Transfer Search (HTS) algorithm [42] is a recently emergence of novel hybridized MHs is thus rising drastically,
proposed MH inspired by the system behaviour to reach a ther- however, this field has not been properly addressed until now.
mal equilibrium state under the heat transfer and thermodynamic Moreover, as per the prominent ‘No Free Lunch’ hypothe-
laws. It is a population-based global optimization algorithm ini- sis [71], an MH cannot solve every problem effectively and ef-
tially explored for numerous unimodal, multimodal, composite, ficiently. In a specific design issue, an MH may yield a good
and hybrid test functions along with numerous real-world bench- result, but still, the same strategy might generate a feeble result
marks. HTS was found superior in performance over many al- in another challenge. To put it another way, there is no MH that
gorithms available in the literature such as PSO, GA, DE, ABC, provides an optimal response for every problem [62,63,68]. Hertz
and TLBO because of its advantages like straightforwardness, and De Werra [72], for example, argued that Tabu search (TS) in
few tuning parameters, and better divergence rate [42]. It was graph colouring problems is far nicer than SA. In contrast, SA is
examined for truss structure optimization [43], non-linear eco- better than TS in the lot-sizing problem, as per Kuik et al. [73].
nomic dispatch design issue [44], heat exchanger design and However, Lee and Kim [74] described that TS and SA were equally
optimization [45], process parameter optimization of machining efficient in a project scheduling problem. Furthermore, Yang [75]
process [46], combined heat and power economic dispatch prob- argued that there is no accepted method for contrasting the
lems [47], economic power generation design issue [48]. Few performance of different MHs. Consequently, discovering new,
scholars examined its performance with improvement and mod- more powerful MHs is an active subject [57–59,76]. Notable,
ification like modified sub-population based HTS for structure Mernik et al. [77] figured out a couple of misconceptions in MH
2
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

comparison. Eventually, Crepinsek et al. [78] cautioned that a 2. Heat transfer search: The fundamental HTS algorithm
meaningful comparison between the different MHs is extremely
difficult. The basic principle of thermal equilibrium between system
Although HTS demonstrated reliability in solving complex en- and surroundings inspires the development of the HTS algo-
gineering optimization design problems, it still has a high likeli- rithm [42]. In thermodynamics, a system always tries to achieve a
hood of slipping into the optimum local [49,50]. HTS also lags in thermal energy balance from its surroundings when its molecules
the iterative process of selection of population size. This is due to
are at a high or low energy state. Three types of heat transfer
the fact that at some events considering a small population size
mechanism viz. Conduction, Convection, and Radiation are used
generates a better response rather than a large size population
to reach a stable condition. The algorithm was developed with
and vice-versa. Results obtained from the modification in the
a proposition of equal probability of execution of these three
basic HTS algorithm may depict the rapid convergence speed,
heat transfer phases. Moreover, to create harmony between local
greater precision, and superior robustness [54–59]. HTS has a
relatively limited number of research investigations thus, it es- intensification and global diversification, the search process of
sentially demands further expedition to substantiate its vigour in the algorithm is improvised in such a way that during the early
the arena of optimization techniques. Moreover, as HTS is a newly generation (during the first half) all three heat transfer phases
developed optimizer, it is always fascinating to uncover various execute the global diversification in search space and after a
improvements that may increase its efficiency and performance. predefined number of generations (during the second half) local
These skills and prospects provide an impetus for formulating intensification is performed by the all the three phases.
a novel optimizer called the decomposition-based multi-objective Analogous to other nature-inspired population-based opti-
heat transfer search (MOHTS/D) algorithm and examining its ef- mization techniques, HTS initializes its search process through
fect on numerous real-life structure optimization problems. Based random generation of a population of size ‘n’ and ‘m’ design
on the flexibility and suitability in the physical world MO de- variables that are equivalent to molecules of different temper-
sign problems, the decomposition-based framework has attracted ature levels. Objective fitness values are analogous to molecular
the attention of many researchers so far [79,80]. In contrast to energy levels, and the surrounding is treated as the best solution.
most of the MO evolutionary algorithms including NSGA-II and Based on the conduction, convection, and radiation search phases
MOGLS that handle MO problems as a whole, the decomposition in each iteration Iter (Iter = 1, 2, 3. . . ., Iter max ) after population
technique disintegrates an MO problem into several subproblems initialization, the population is updated continuously. Greedy
which are optimized by applying the algorithm individually. In selection is employed in HTS as the solution which is superior
this study widely used Tchebycheff Approach [81–83] is applied to the previous one is acknowledged and continues in the next
for the scalarization of the MO problem. Decomposition is a iteration, where the worst solution is superseded with the elite
simple, efficient and computationally less complex method, that solution and against the existence of a duplicate solution, it
provides better solution quality and facilitates the generation of
is replaced by a randomly generated solution. The three heat
an even distribution of solutions along the Pareto Front (PF).
transfer phase of the HTS is explained in the following sections.
Thus, the present study integrates the advantages of decomposing
methodology with HTS to achieve a robust global optimization
technique that creates a proper trade-off between local intensi- 2.1. Conduction phase
fication and global diversification of search. Hence, the proposed
study contributes as follows: This phase is driven by the Fourier law of conduction which
• A novel decomposition-based MO global optimization al- states that the molecules with a higher energy level will trans-
gorithm is proposed and investigated based on the evenly fer the heat to the lower energy level molecules to achieve
generated weight vectors sorting and Euclidean distance the equilibrium state. With the physical interaction between the
approach. molecules of the system and surrounding the conduction, heat
• Depending on the Tchebycheff decomposition strategy, the exchange is equally possible. In this phase for population, modi-
suggested MOHTS/D algorithm correlates each solution with fication is governed by Eqs. (1) and (2) as follows:
a scalar subproblem, making the optimization operations {
,i 1 − R , if f (Sp ) > f (Sq )
( )
Sqold 2
more effective. Spne,iw = ; if Iter ≤ Itermax /Fcond
, if f (Sp ) < f (Sq )
old
( 2
)
• Equivalent to the fundamental HTS algorithm, the suggested Sp,i 1 − R
algorithm exploits the conduction, convection, and radiation (1)
phase to realize harmony between search intensification {
and diversification. Sqold (1 − ri ) , if f (Sp ) > f (Sq )
Spne,iw = ,i
; if Iter > Itermax /Fcond
• The MOHTS/D algorithm is tested with eight challenging Spold
,i (1 − ri ) , if f (Sp ) < f (Sq )
real-world MO structural optimization design problems.
(2)
• Performance comparison has been made qualitatively and
quantitatively with MO heat transfers search, MO Evolu- Here p and q are random solutions where p ̸ = q; porq ∈
tionary Algorithm Based on Decomposition and four other (1, 2, . . . ., n); i is a random design variable index and
state-of-the-art MHs. i ∈ (1, 2, . . . ., m); R ∈ [0, 0.3333] and ri ∈ [0, 1] are the prob-
This subsequent report is structured as follows: Section 2 gives ability value and random number respectively for the conduction
details of the fundamental HTS algorithm; Section 3 elaborates phase; Fcond is a factor of conduction whose value is taken as
on the suggested decomposition-based MOHTS methodology and 2 [50,53,54] for the total generation first half. During the first
its working; Section 4 provides the details of the considered half, the two functions contrasted, and the inferior solution is
MO design optimization mathematical model followed by the substituted based on Eq. (1), while Eq. (2) is used to replace
outcomes and detailed discussion for all test examples considered the inferior solution found in the latter half of the search. This
in Section 5; Section 6 illustrates general conclusions of this switching of Eqs. (1) to Eq. (2) provides both local intensification
study according to the performance metric, and PFs gained with and global diversification of the algorithm to find an optimal
prospects. solution.
3
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

2.2. Convection phase λm

2 = 1. Suppose z
∗
= (z1∗ , z2∗ )T are the minimum objective
value vector that will be treated as a reference point, and the
This phase is governed by the law of convection (Newton’s law elements of this vector are calculated as z1∗ = min{f1 (x)|x ∈ Ω }
of cooling) in which the system molecules (at the mean temper- and z2∗ = min{f2 (x)|x ∈ Ω }. After the decomposition of MOP into
ature Smean ) interact with surroundings (at the temperature Ssurr ) N subproblems, the objective function of mth subproblem is:
to attain thermal equilibrium where the latter one is treated as
λm ∗
1 |f1 (x) − z1 | }
the best solution. The solutions are updated as follows: g te (x|λm , z ∗ ) = max{ (7)
λm
2 |f 2 (x) − z ∗
2|
Spne,iw = Spold
,i + R × [Ssurr − Smean × abs (R − ri )] if Iter ≤ Itermax /Fconv
z ∗ is all likely to be unknown beforehand. During the search pro-
(3) cess, the algorithm uses the lowest values of f1 (x) and f2 (x) found
Spne,iw = Spold
,i + R × [Ssurr − Smean × round (1 + ri )] if Iter > Itermax /Fconv
thus far to substitute z1∗ and z2∗ , respectively in the objectives of
the subproblems. MOHTS/D performs simultaneous minimization
(4)
of all N objective functions (for N subproblems) in a single run.
It must be emphasized that each design variable is also mod- Each subproblem is optimized by utilizing information mainly
ified throughout this phase along with the solution Sp . Here R from its neighbouring subproblems. Moreover, in MOHTS/D, all
∈ [0.6666, 1] and factor of convection(Fconv ) is taken as 10 [57– subproblems are given equal computational effort.
59]. Based on Eqs. (3) and (4) both local intensification and global The objectives of weight minimization and high compliance
diversification is being regulated. in the truss bar problem are highly disparate. Normalization of
the objectives is necessary for MOHTS/D, else the algorithm may
2.3. Thermal radiation phase evolve towards the objective with a higher numerical value. The
normalization process brings the values of both objectives within
Radiation mode of heat transfer mechanism is used in this the range 0–1. It is implemented by modifying the decomposed
phase to neutralize the thermal imbalance between the system objective function in Eq. (8) as:
and its surrounding. It should be noted that communication can f (x)−z ∗
occur between system molecules themselves or amidst system λm 1 1
1 | z nad −z ∗ |
molecules and surrounding molecules. As per Eqs. (5) and (6), the g (x|λ , z , z
te m ∗ nad
) = max{ 1 1
∗ } (8)
m f2 (x)−z2
solutions are updated as follows: λ |
2 z nad −z ∗ |
2 2
{
,i (1 − R) + R Sq,i , if f (Sp ) > f (Sq )
( )
Spold old
where, z nad = (z1nad , z2nad )T is the vector of maximum values of
Spne,iw = ; if Iter ≤ Itermax /Frad
,i (1 + R) − R Sq,i , if f (Sp ) < f (Sq )
( )
Spold old
f1 (x) and f2 (x) and similar to z ∗ , z nad is also evaluated by HTS
(5) during the search process using maximum values of f1 (x) and
{ f2 (x).
(1 − ri ) + ri Sqold , if f (Sp ) > f (Sq )
( )
Spold During the search process, MOHTS/D with Tchebycheff ap-
Spne,iw = ,i
( ,i ) ; if Iter > Itermax /Frad
Spold
,i (1 + ri ) − ri Sqold
,i , if f (Sp ) < f (Sq ) proach [56] maintains:
(6) • A population of N vectors x1 , x2 , . . . ., xN And in this specific
Here p and q are random solutions where p ̸ = q; porq ∈ problem, each vector contains 100 elements.
(1, 2, . . . ., n) and their corresponding fitness values are compared • Function values of FV 1 , FV 2 , . . . ., FV N where FV m = {f1 (xm ),
based on which the modification of solutions are performed fol- f2 (xm )} for m = 1, 2, . . . , N.
lowing Eqs. (5) and (6). The probability variable R ∈ [0.3333, • z ∗ = (z1∗ , z2∗ )T , where z1∗ and z2∗ are the minimum values
0.6666] and Frad is the radiation factor which is set as 2 [57–59]. found far evaluating f1 (x) and f2 (x).
The details of all the three-phase of HTS working are depicted in • z nad = (z1nad , z2nad )T where z1nad and z2nad are the maximum
Fig. 1. values found till now evaluating functions f1 (x) and f2 (x).
• Utility of the sub-problems, π m For m = 1, 2, . . . , N.
3. Decomposition-based multi-objective heat transfer search • Current generation number gen.
(MOHTS/D) algorithm The general framework of MOHTS/D is presented in Algo-
rithm 1.
The MO optimization algorithm based on decomposition was The flowchart of the MOHTS/D algorithm is presented in Fig. 2.
firstly developed by Zhang and Li [79]. Decomposition is a
straightforward, effective, and computationally less demanding
strategy for improving solution quality and generating an eq- 3.1. Best compromise solution (BCS) based on fuzzy decision
uitable distribution of solutions besides the PF. In this study, a
decomposition-based strategy is integrated with HTS to develop To find the best and most suitable fit solution from the achieved
a simple but efficient technique to solve structural optimiza- Pareto-optimal package a fuzzy-based compromised solution ap-
tion problems. The proposed MOHTS/D algorithm does not have proach [84] is considered in this investigation. A solution that
algorithm-specific parameters that make it robust and suitable provides the best degree of satisfaction to each objective is
for a comprehensive application. In MOHTS/D, MO problems pursued out of feasible Pareto-optimal solutions and governed
are decomposed into several scalar optimization sub-problems by fuzzy membership functions. Each optimization problem of
by applying decomposition approaches, and these sub-problems the jth solution is expressed by a membership function µi , given
j
are optimized concurrently by using HTS. MOHTS/D utilized the the imprecise nature of the decision-makers. Assume that µi is a
Tchebycheff approach method to decompose the problem of ap- monotonic function, and it is defined as follows:
proximation of PF into several scalar optimization sub-problems
j j
⎧
with N evenly distributed weight (λ1 , λ2 . . . .λN ) vectors. In this ⎪
⎪ 1, if fi ⩽ fmin
⎨ j j
specific problem of truss bar design optimization, N is considered fmax −fi
µji = j j
j
, if fmin j j
⩽ fi ⩽ fmax (9)
with each weight vector being of 2 dimensions for the two ⎪ fmax −fmin
objectives i.e. λm = (λm 1 , λ2 ) for m = 1, 2, . . . , N and λ1 +
m T m
0, if fi ⩾
j j
⎪
fmax
⎩

4
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 1. Flowchart of the fundamental HTS algorithm.

The normalized membership function can be constructed at each function. The best-compromised value is the one with a high
non-dominated solution as follows: value of µi .
∑Nobj
j=1 µij
µi = ∑ ∑N (10) 4. Multi-objective design-optimization problem mathematical
M obj
i=1 j=1 µij formulation
where M is the number of non-dominated solutions, Nobj is
j j
the number of the objective function, and fmax and fmin are Multi-objective design problems are usually formulated to
the maximum and minimum values of the respective objective find the decision vector subjected to numerous constraints while
5
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

concurrently optimizing diverse objectives. In this study, struc- The mathematical formulation for the MO truss optimization
ture weight is the first objective that needs to be minimized, problem is as follows:
whereas the maximum nodal deflection is accounted for as the
second objective. Find, A = {A1 , A2 , . . . , Am } (11)
6
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 2. Flowchart of the MOHTS/D algorithm.

∑m
to minimize the structure mass: f1 (A) = 1 Ai⏐ρ
i=(⏐ )i Li 5. Empirical evaluation
and maximize nodal deflection: f2 (A) = max ⏐δj ⏐
Subjects to: To examine the convergence, coverage, intensification, and
diversification of the proposed MOHTS/D numerous large 2-D and
Behaviour constraints: 3-D structural tests were examined and contrasted with other
Stress constraints, g (A) = |σi | − σimax ≤ 0 distinguished MO optimization strategies existing in the litera-
Side constraints: ture viz. Multi-Objective Passing Vehicle Search (MOPVS) [30],
Cross − sectional area constraints, Amin
i ≤ Ai ≤ Amax
i Multi-Objective Slime Mould Algorithm (MOSMA) [85], Multi-
Objective Symbiotic Organisms Search (MOSOS) [86], and Multi-
where, i = 1, 2, . . . , m; j = 1, 2, . . . , n
Objective Multi-Verse Optimizer (MOMVO) [87]. These algorithms
Here, Ai is a design variable vector; ρi and Li are the mass have shown their competitiveness and efficacy in solving MO
density and length of the elements, respectively; Ei and engineering benchmarks including structure optimization prob-
σi Correspond to the ‘i’ element Modulus of elasticity and lems. For instance, MOPVS is a Pareto dominance-based approach,
stress, respectively. Moreover, the allowable upper and lower that demonstrated its higher fitness and efficiency relative to
bounds are represented by superscripts ‘max’ and ‘min’, the well-known MO ant system, MO ant colony system, and
respectively. MO heat transfer search algorithms while solving several MO
convex tests functions and structure optimization problems [30].
MOSMA method shows better quality solution relatively to MO
4.1. The dynamic penalty function Evolutionary Algorithm Based on Decomposition (MOEA/D), and
MO Water-Cycle Algorithm (MOWCA) for several unconstrained,
constrained, and real-world engineering design problems by us-
For processing constrained problems like Equation (11), one of ing an elitist non-dominated sorting approach [85]. For work
the best ways is to transform these design issues into dynamic shift scheduling problem MOSOS outperform popular NSGA-II,
unconstrained optimization problems using a dynamic penalty MOPSO, MODE, and MO artificial bee colony (MOABC) [86]. Simi-
approach. Considering that in the design arena, all the objective larly, MOMVO implemented an archive with an updating mecha-
function values are positive, the multiplication-based penalty nism to improve solution quality and outperforms distinguished
function is among the most effective penalty functions. MOEA/D, NSGA-II, and SPEA-II algorithms for solving 80 con-
q ⏐ ⏐ strained, unconstrained and MO engineering case studies [87]. In
pi ⏐ general, these algorithms are population-based nature-inspired
f (A) ∗ (1 + ε1 ∗ ∁)ε2 , ∁ =
∑
∁i , ∁i = ⏐⏐1 − ∗ ⏐⏐
⏐
(12)
pi techniques implemented to solve MO problems from the aspects
i=1
of regulating the degree of knowledge transfer, balancing the
where pi is the dimension of requirement infringement having diversity, convergence of the population and explicit knowledge
the bound as p∗i . The parameter q is the number of constraints. transfer, respectively and share similar characteristics with MO-
The factors ε1 and ε2 are pre-specified by a user. In this investi- HTS. Therefore, it is valuable that these algorithms are used to
gation, the estimations of both ε1 and ε2 are set at 3, which were verify the effectiveness of the proposed MOHTS/D. The subse-
gotten from testing their impact on the parity of the investigation quent section elaborates on the information about the eight test
balance [57–59]. examples that were considered.
7
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

• The 60-bar ring structure is the fourth test example con-

sidered, where Fig. 12 illustrates its ground structure. In
the line of previous studies, this truss is segregated into
members of 25 out of the total 60 elements.
• The fifth benchmark accounted for is the widely used 3-D
72-bar truss problem, and Fig. 15 depicts its ground struc-
ture. Out of a total of 72 members, the whole truss is
grouped into 16 members following its symmetry.
• The 120 bar truss (3-D dome structure) problem is the sixth
spatial structural test example investigated. Fig. 18 illus-
Fig. 3. The 10-bar truss. trates its ground structure in which, based on its symmetry,
a total of 120 elements are divided into 49 groups.
• The seventh test example considered is the large 3-D 200-
bar truss problem. Its ground structure is illustrated in
5.1. Test problems
Fig. 21, which is clustered into 77 groups due to its sym-
metry.
• Following the previous studies available in the literature • The tower structure i.e. 942-bar truss is the last and eighth
[57–59,62,63] eight benchmarks were analyzed to evaluate benchmark accounted for in this study, where Fig. 24 shows
the effectiveness of all considered MO optimization algo- its ground structure. The whole 942 members of the truss
rithms. are segregated into 52 members in this investigation.
• The first test example considered is the widely used
2-D 10-bar truss design problem whose ground structure is Table 1 presents all design considerations and mechanical prop-
illustrated in Fig. 3, which encompassed all elements, nodes, erties used for the simulation of eight truss MO optimization
loads type, and dimensions. examples in one frame. Moreover, Figs. 3, 6, 9, 12, 15, 18, 21, and
• The 3-D 25-bar truss problem is the second test example in- 24 illustrates load directions, constraints, and truss dimensions.
vestigated in this study. Fig. 6 illustrates its ground structure
in which, based on its x–z and y–z planes symmetry, a total
of 25 elements are divided into 8 groups. 5.2. Evaluation method
• The third benchmark problem considered here is shown in
Fig. 9, which is a 37-bar planar truss, called a simply sup- In this study, every algorithm is performed 30 times individ-
ported bridge problem. Based on the symmetry of structure ually for all considered eight test examples with a 40 popula-
about its middle vertical plane bars are clustered into, 5 tion size, 500 maximum iteration number, and 20,000 functional
shape variables and 14 sizing variables groups. evaluations [85,87].

Fig. 4. Best Pareto fronts of the 10-bar truss by all algorithms.

8
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 1
Parameter settings of different optimizer.
Parameters MOHTS/D MOPVS MOSMA MOSOS MOMVO
Number of runs 30 30 30 30 30
Population size/Search agent (Npop ) 40 40 40 40 40
Maximum number of iterations (Max_t) 500 500 500 500 500
Number of function evaluations (FES max ) 20,000 20,000 20,000 20,000 20,000

Fig. 5. Box plots of the 10-bar truss by all algorithms.

• The Pareto Front-Hypervolume (PFHV) and Inverted Gener-

ational Difference (IGD) metrics are employed to concur-
rently examine the uniformity-convergence-spread of the
non-dominated set of solutions procured from the compu-
tation experiments.
• To examine the search efficiency and reliability of consid-
ered algorithms in terms of faster convergence rate Gen-
erational Distance (GD), Spread (SD), Coverage (CVG), and
Coverage over Pareto Front (CPF) metrics are used [85,87].
• To measure the computational complexity, Runtime (RT)
metric and for combined diversity-spread, Spacing (SP), Di-
versity (DM), and Pure Diversity (PD) metrics are calcu-
lated [51–53].
• The mean and standard deviation (STD) values of the met-
rics are regarded as the statistical performance yardstick
[57–59].
• Friedman’s rank test is employed as a statistical review of
all the optimizers examined [62,63].
√∑
no
i=1 d2i
GD = (13)
Fig. 6. The 25-bar space truss. n
√
∑nt ( ′ )2
i=1 di
IGD = (14)
n
9
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 7. Best Pareto fronts of the 25-bar truss by all algorithms.

Fig. 8. Box plots of the 25-bar truss by all algorithms.

10
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 9. The 37-bar truss.

Fig. 10. Best Pareto fronts of the 37-bar truss by all algorithms.

 ∑n

 1 ∑ n i=1 T
( )2 RT = (22)
SP ≜ √ d − di (15) n
n−1
i=1

 o
∑ Where no is the number of True Pareto solution (PS), nt is the
SD = √ max (d (ai , bi )) (16) number of true Pareto optimal solutions, o is the number of ob-
i=1 jectives, d is the average of all di , di , and d′i specifies the Euclidean
( ( )
∑
m (h (i, j, . . .)) distance, n is the number of obtained PS, di = minj |f1i ⃗ x −
H (i,j,...)̸ =0 j
( ) j
)
PD = ∑ (17) x)| + |f2i ⃗
f1 (⃗ x − f2 (⃗
x) for all i,j = 1,2,. . . ,n., ai and bi is the maxi-
H (i,j,...)̸ =0 m (H (i, j, . . .)) mum and minimum value in the ith objective.
( )
⋃{ Friedman’s test is a non-parametric test for finding differences
HV = Λ ′ ′ nadir
}
s |s≺s ≺s (18) in treatments across multiple attempts. Nonparametric means
s∈PF the test does not assume your data comes from a particular
∑N −1 ⏐⏐ ⏐ distribution (like the normal distribution). Basically, it is used in
df + dl + i=1 di − d⏐
DM = (19) place of the ANOVA test when you do not know the distribution
df + dl + (N − 1) d of your data [62,63].
f1 (x) Step 1: Sort your data into blocks
∑n {
ψi 1, if Pi ∈ PF and αi−1 ≤ tan f2 (x)
≤ αn
CVG = i=1
, ψi = Step 2: Rank each column separately. The smallest score should
N 0, Otherwise get a rank of 1.
(20) Step 3: Sum the ranks (find a total for each column).
∑n Step 4: Calculate the test FNRT statistic. You will need:
i=1 PFi k: the number of algorithm
CPF = (21)
N j: the number of runs
11
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 11. Box plots of the 37-bar truss by all algorithms.

Fig. 12. The 60-bar ring truss.

12
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 13. Best Pareto fronts of the 60-bar truss by all algorithms.

Fig. 14. Box plots of the 60-bar truss by all algorithms.

13
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 15. The 72-bar 3D truss.

Fig. 16. Best Pareto fronts of the 72-bar truss by all algorithms.

Si : The total ranks for each of the ith columns. 10-bar, 25-bar, 37-bar, 60-bar, 72-bar, 120-bar, 200-bar, and 942-
bar trusses) based on all performance metrics and Friedman’s
Si2
∑
12
FNRT = − 3k (j + 1) . rank test are demonstrated in Tables 2–11.
jk (j + 1)
• From Table 2, PFHV mean value obtained by MOHTS/D has
5.3. Results and discussion presented better solutions over MOMVO, MOPVS, MOSMA,
and MOSOS for all the considered trusses. This demonstrates
The results obtained by the MOPVS, MOSMA MOSOS, MOMVO, the high volume solutions enclosed near the PF of MOHTS/D
and MOHTS/D algorithms for all considered test examples (i.e. relatively. The STD value obtained from the PFHV test for
14
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 17. Box plots of the 72-bar truss by all algorithms.

Table 2
Results of (PFHV Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 0.64 0.62 0.63 0.62 0.60 0.68 0.63 0.67 0.66 0.65
fmax 0.64 0.63 0.63 0.63 0.62 0.69 0.66 0.68 0.68 0.67
fmean 0.64 0.62 0.63 0.63 0.61 0.68 0.64 0.68 0.67 0.66
fmedian 0.64 0.63 0.63 0.63 0.61 0.68 0.64 0.67 0.67 0.67
fstd 0.00 0.01 0.00 0.01 0.01 0.00 0.01 0.00 0.01 0.01
Friedman Value 500.00 250.00 375.00 225.00 150.00 500.00 125.00 400.00 250.00 225.00
25 bar 120 bar
fmin 0.70 0.69 0.69 0.69 0.68 0.55 0.54 0.54 0.54 0.54
fmax 0.70 0.69 0.69 0.69 0.69 0.56 0.55 0.55 0.55 0.54
fmean 0.70 0.69 0.69 0.69 0.68 0.56 0.55 0.54 0.55 0.54
fmedian 0.70 0.69 0.69 0.69 0.68 0.56 0.55 0.54 0.55 0.54
fstd 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Friedman Value 500.00 275.00 325.00 300.00 100.00 500.00 400.00 250.00 250.00 100.00
37 bar 200 bar
fmin 0.69 0.67 0.69 0.64 0.67 0.77 0.70 0.72 0.73 0.73
fmax 0.71 0.69 0.69 0.68 0.68 0.78 0.72 0.76 0.75 0.73
fmean 0.70 0.68 0.69 0.67 0.68 0.78 0.71 0.74 0.74 0.73
fmedian 0.71 0.67 0.69 0.67 0.68 0.78 0.72 0.75 0.74 0.73
fstd 0.01 0.01 0.00 0.02 0.01 0.01 0.01 0.02 0.01 0.00
Friedman Value 500.00 175.00 400.00 150.00 275.00 500.00 100.00 325.00 325.00 250.00
60 bar 942 bar
fmin 0.52 0.50 0.52 0.51 0.50 0.71 0.68 0.69 0.67 0.71
fmax 0.53 0.51 0.53 0.53 0.52 0.75 0.69 0.70 0.69 0.72
fmean 0.53 0.51 0.52 0.52 0.51 0.74 0.68 0.70 0.68 0.72
fmedian 0.53 0.51 0.52 0.51 0.51 0.74 0.68 0.70 0.68 0.72
fstd 0.00 0.00 0.00 0.01 0.01 0.02 0.00 0.01 0.01 0.00
Friedman Value 450.00 150.00 425.00 300.00 175.00 475.00 175.00 275.00 150.00 425.00
Average Friedman Value 3925.00 1650.00 2775.00 1950.00 1700.00 1975.00 800.00 1250.00 975.00 1000.00

MOHTS/D has the least value, which shows the superiority result. The GD metric results as per Table 3 illustrate that
of MOHTS/D in terms of search consistency. Also, MOHTS/D MOHTS/D has better mean values and STD values for all
ranks first with the highest ranks as per Friedman’s test problems. MOHTS/D achieved a decreased difference in the

15
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 18. The 120-bar 3D truss.

mean value of similar results obtained from MOPVS, MOSOS, MOHTS/D accomplished the best mean and STD functional
MOSMA, and MOMVO, governing its superior convergence values relatively. This is also justified by its superior Fried-
quality in comparison to others. Moreover, from Friedman’s man’s test values.
test at a 95% significance level, the proposed MOHTS/D • The SP metric results from Table 8 demonstrate the better
ranked first among the considered algorithms. mean and STD values of MOHTS/D relatively. The MOHTS/D
• In terms of the CVG performance measure from Table 4, STD values illustrate a significant decrease from MOMVO,
the MOHTS/D gets the best optimal mean values with the MOSOS, MOPVS, and MOSMA. Moreover, based on Fried-
lowest STDs for all considered algorithms. Also, MOHTS/D man’s test, MOHTS/D ranks first among other algorithms
obtained the best Friedman’s test rank. These outcomes and eventually displays its enhanced spacing quality. From
exhibit MOHTS/D improved coverage attributes relative to Table 9, MOHTS/D manifests better mean STD results than
others. According to Table 5, MOHTS/D reveals the best the rest. The outcomes reflect the well-distributed non-
mean value of CPF metric at the lowest STD values, and dominated solutions for MOHTS/D.
also accomplished a maximum of Friedman’s test rank. This • IGD mean value of MOHTS/D from Table 10 displays a signif-
defines the improved coverage over PF quality. icant decrease over MOMVO, MOSOS, MOSMA, and MOPVS
• The DM metric mean values are depicted in Table 6 from respectively for all the considered problems. Similarly, there
which it is evident that the MOHTS/D has a significant is significant percentage decrease is recorded between MO-
increase in the mean values compared to MOSMA, MOPVS, HTS/D and MOMVO, MOSOS, MOSMA, and MOPVS STD val-
MOSOS, and MOMVO. Also, MOHTS/D realizes the best STD ues. Moreover, Friedman’s test MOHTS/D realizes a superior
values relatively. MOHTS/D achieved the first rank with value and is ranked at first then rest. The outcomes from Ta-
Friedman’s test values for the considered problems. Thus, ble 10 depict the uniformity-convergence-spread attribute
MOHTS/D exhibits the superior diversity maintenance fea- of MOHTS/D. The RT metric from Table 11 depicts the bet-
ture. The proposed decomposition-based approach proves ter mean values of MOHTS/D comparatively and also for
best in terms of PD measure as shown in Table 7 as the least STD values. Friedman’s test illustrates the least

16
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 19. Best Pareto fronts of the 120-bar truss by all algorithms.

Fig. 20. Box plots of the 120-bar truss by all algorithms.

17
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 3
Results of (GD Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 13.86 18.61 42.44 17.14 41.58 34.65 15.50 19.95 21.83 100.16
fmax 15.59 20.44 76.53 38.35 79.13 48.41 26.18 28.65 30.01 123.23
fmean 14.60 19.91 54.67 23.68 62.77 41.55 22.04 22.28 24.66 114.65
fmedian 14.48 20.30 49.86 19.61 65.19 41.57 23.24 20.27 23.41 117.61
fstd 0.88 0.88 14.99 10.05 19.30 7.92 5.17 4.26 3.87 11.02
Friedman Value 100.00 250.00 425.00 250.00 475.00 400.00 200.00 175.00 225.00 500.00
25 bar 120 bar
fmin 5.00 6.35 5.99 6.35 27.23 82.70 88.88 178.77 88.25 351.61
fmax 7.18 6.94 12.66 7.00 30.40 102.20 141.81 259.41 113.03 406.14
fmean 5.63 6.71 8.04 6.73 28.27 88.20 115.82 200.92 104.23 382.81
fmedian 5.16 6.77 6.76 6.78 27.73 83.94 116.30 182.76 107.82 386.74
fstd 1.04 0.29 3.16 0.33 1.44 9.41 21.62 39.03 10.93 27.70
Friedman Value 150.00 250.00 300.00 300.00 500.00 125.00 275.00 400.00 200.00 500.00
37 bar 200 bar
fmin 3.34 4.87 5.29 3.93 19.28 127.73 39.66 60.69 137.94 290.17
fmax 6.23 5.34 7.09 5.84 28.31 312.91 93.23 136.26 211.79 358.79
fmean 4.56 5.19 5.93 4.51 22.86 199.61 71.20 104.39 163.30 316.98
fmedian 4.33 5.28 5.68 4.13 21.92 178.90 75.95 110.30 151.73 309.48
fstd 1.21 0.22 0.85 0.89 3.84 79.29 22.55 38.05 32.98 33.29
Friedman Value 225.00 250.00 375.00 150.00 500.00 375.00 125.00 175.00 325.00 500.00
60 bar 942 bar
fmin 21.37 7.54 7.44 11.31 39.02 51927.85 4179.84 4825.71 4702.49 21057.56
fmax 28.85 14.22 11.30 14.90 55.90 75632.34 4876.52 9766.77 5921.58 40966.83
fmean 25.19 9.29 8.67 12.52 45.61 58357.39 4406.98 7327.77 5333.73 34897.61
fmedian 25.28 7.70 7.96 11.94 43.75 52934.69 4285.77 7359.29 5355.43 38783.03
fstd 3.05 3.29 1.82 1.69 7.22 11555.69 328.58 2816.79 679.71 9453.63
Friedman Value 400.00 200.00 125.00 275.00 500.00 500.00 150.00 275.00 175.00 400.00
Average Friedman Value 2275.00 1700.00 2250.00 1900.00 3875.00 1400.00 750.00 1025.00 925.00 1900.00

Table 4
Results of (CVG Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 0.52 0.61 0.66 0.77 1.00 0.84 0.75 0.88 0.87 1.00
fmax 0.66 0.72 0.80 0.85 1.00 0.95 0.97 0.94 0.99 1.00
fmean 0.62 0.67 0.76 0.81 1.00 0.89 0.81 0.92 0.92 1.00
fmedian 0.64 0.67 0.79 0.81 1.00 0.88 0.77 0.92 0.92 1.00
fstd 0.06 0.05 0.07 0.05 0.00 0.05 0.11 0.03 0.06 0.00
Friedman Value 125.00 200.00 325.00 350.00 500.00 225.00 150.00 275.00 350.00 500.00
25 bar 120 bar
fmin 0.57 0.67 0.40 0.71 0.95 0.60 0.75 0.64 0.70 1.00
fmax 0.67 0.75 0.49 0.74 1.00 0.67 0.85 0.70 0.80 1.00
fmean 0.60 0.72 0.43 0.73 0.97 0.62 0.80 0.66 0.77 1.00
fmedian 0.57 0.72 0.42 0.73 0.97 0.61 0.79 0.65 0.78 1.00
fstd 0.05 0.03 0.04 0.02 0.03 0.03 0.04 0.03 0.04 0.00
Friedman Value 200.00 325.00 100.00 375.00 500.00 125.00 400.00 175.00 300.00 500.00
37 bar 200 bar
fmin 0.82 0.85 0.74 0.80 1.00 0.95 0.57 0.97 1.00 1.00
fmax 0.93 0.87 0.88 0.98 1.00 1.00 1.00 1.00 1.00 1.00
fmean 0.89 0.86 0.82 0.88 1.00 0.99 0.88 0.99 1.00 1.00
fmedian 0.90 0.86 0.83 0.86 1.00 1.00 0.97 1.00 1.00 1.00
fstd 0.05 0.01 0.07 0.09 0.00 0.03 0.21 0.02 0.00 0.00
Friedman Value 350.00 225.00 200.00 225.00 500.00 300.00 162.50 312.50 362.50 362.50
60 bar 942 bar
fmin 0.97 0.80 0.87 0.99 1.00 1.00 0.11 0.39 0.87 1.00
fmax 1.00 1.00 0.97 1.00 1.00 1.00 0.69 0.73 0.92 1.00
fmean 0.98 0.89 0.90 1.00 1.00 1.00 0.40 0.57 0.91 1.00
fmedian 0.98 0.88 0.88 1.00 1.00 1.00 0.39 0.58 0.92 1.00
fstd 0.01 0.10 0.05 0.01 0.00 0.00 0.24 0.19 0.03 0.00
Friedman Value 300.00 200.00 150.00 400.00 450.00 450.00 125.00 175.00 300.00 450.00
Average Friedman Value 2075.00 1787.50 1712.50 2662.50 3762.50 837.50 937.50 1312.50 1812.50 837.50

computational complexity of the MOHTS/D and MOMVO for • Figs. 4, 7, 10, 13, 16, 19, 22, and 25 illustrate the best
solving the truss problems. PFs obtained by all considered algorithms for the 10-bar,
18
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 5
Results of (CPF Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 0.89 0.49 0.60 0.49 0.34 0.74 0.29 0.43 0.39 0.37
fmax 0.92 0.51 0.67 0.60 0.60 0.78 0.34 0.49 0.45 0.50
fmean 0.90 0.50 0.64 0.54 0.49 0.75 0.31 0.45 0.41 0.41
fmedian 0.90 0.50 0.65 0.53 0.51 0.75 0.30 0.44 0.41 0.37
fstd 0.02 0.01 0.03 0.05 0.13 0.02 0.02 0.03 0.03 0.06
Friedman Value 500.00 150.00 400.00 250.00 200.00 500.00 100.00 375.00 275.00 250.00
25 bar 120 bar
fmin 0.92 0.55 0.66 0.54 0.52 0.89 0.59 0.56 0.55 0.55
fmax 0.97 0.60 0.68 0.63 0.61 0.91 0.65 0.64 0.59 0.57
fmean 0.95 0.57 0.68 0.60 0.59 0.90 0.62 0.59 0.57 0.57
fmedian 0.95 0.56 0.68 0.62 0.61 0.90 0.62 0.58 0.57 0.57
fstd 0.02 0.03 0.01 0.04 0.04 0.01 0.03 0.04 0.02 0.01
Friedman Value 500.00 150.00 400.00 250.00 200.00 500.00 325.00 300.00 175.00 200.00
37 bar 200 bar
fmin 0.79 0.41 0.48 0.35 0.38 0.68 0.13 0.08 0.18 0.15
fmax 0.93 0.51 0.53 0.45 0.49 0.84 0.16 0.22 0.26 0.22
fmean 0.89 0.44 0.50 0.42 0.46 0.78 0.15 0.18 0.21 0.18
fmedian 0.91 0.43 0.50 0.45 0.49 0.80 0.15 0.21 0.20 0.17
fstd 0.06 0.05 0.02 0.05 0.05 0.08 0.02 0.07 0.04 0.03
Friedman Value 500.00 200.00 375.00 175.00 250.00 500.00 125.00 275.00 350.00 250.00
60 bar 942 bar
fmin 0.55 0.26 0.39 0.34 0.37 0.41 0.02 0.09 0.08 0.15
fmax 0.61 0.35 0.40 0.43 0.53 0.57 0.11 0.14 0.10 0.21
fmean 0.58 0.31 0.39 0.36 0.41 0.51 0.05 0.10 0.09 0.19
fmedian 0.59 0.32 0.39 0.34 0.38 0.52 0.03 0.09 0.10 0.20
fstd 0.02 0.04 0.01 0.04 0.08 0.07 0.04 0.03 0.01 0.02
Friedman Value 500.00 100.00 350.00 225.00 325.00 500.00 150.00 200.00 250.00 400.00
Average Friedman Value 4000.00 1300.00 2675.00 1950.00 2075.00 2000.00 700.00 1150.00 1050.00 1100.00

Table 6
Results of (DM Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 0.80 0.50 0.46 0.52 0.52 0.80 0.36 0.42 0.40 0.46
fmax 0.83 0.52 0.52 0.60 0.55 0.81 0.40 0.46 0.47 0.49
fmean 0.82 0.51 0.49 0.54 0.54 0.80 0.38 0.44 0.42 0.48
fmedian 0.82 0.50 0.49 0.52 0.54 0.81 0.37 0.43 0.41 0.48
fstd 0.01 0.01 0.03 0.04 0.02 0.00 0.02 0.02 0.03 0.01
Friedman Value 500.00 200.00 175.00 300.00 325.00 500.00 100.00 300.00 250.00 350.00
25 bar 120 bar
fmin 0.77 0.58 0.46 0.59 0.49 0.82 0.58 0.48 0.59 0.60
fmax 0.79 0.62 0.49 0.62 0.63 0.85 0.67 0.52 0.65 0.67
fmean 0.77 0.60 0.47 0.60 0.55 0.84 0.64 0.50 0.63 0.64
fmedian 0.77 0.60 0.48 0.60 0.54 0.84 0.65 0.50 0.64 0.65
fstd 0.01 0.01 0.02 0.02 0.07 0.01 0.04 0.03 0.03 0.03
Friedman Value 500.00 325.00 100.00 300.00 275.00 500.00 300.00 100.00 225.00 375.00
37 bar 200 bar
fmin 0.81 0.49 0.44 0.45 0.48 0.76 0.30 0.19 0.32 0.28
fmax 0.83 0.58 0.52 0.51 0.57 0.79 0.33 0.31 0.41 0.37
fmean 0.82 0.51 0.46 0.47 0.52 0.77 0.31 0.27 0.35 0.31
fmedian 0.83 0.49 0.45 0.46 0.51 0.77 0.30 0.28 0.32 0.29
fstd 0.01 0.04 0.04 0.03 0.04 0.01 0.01 0.05 0.04 0.04
Friedman Value 500.00 325.00 150.00 200.00 325.00 500.00 300.00 150.00 350.00 200.00
60 bar 942 bar
fmin 0.68 0.39 0.45 0.40 0.46 0.66 0.20 0.26 0.24 0.26
fmax 0.75 0.42 0.48 0.49 0.50 0.75 0.29 0.29 0.30 0.33
fmean 0.73 0.41 0.46 0.42 0.48 0.72 0.22 0.27 0.27 0.31
fmedian 0.74 0.40 0.45 0.40 0.48 0.74 0.20 0.27 0.28 0.33
fstd 0.03 0.01 0.01 0.04 0.02 0.04 0.05 0.02 0.02 0.03
Friedman Value 500.00 125.00 275.00 200.00 400.00 500.00 125.00 250.00 300.00 325.00
Average Friedman Value 4000.00 1800.00 1500.00 2125.00 2575.00 2000.00 825.00 800.00 1125.00 1250.00

25-bar, 37-bar, 60-bar, 72-bar, 120-bar, 200-bar, and 942- while the MOHTS/D fronts are continuous and well dis-
bar trusses, respectively. The Fronts obtained by MOMVO, tributed in nature for all considered problems. Moreover for
MOSOS, MOSMA, and MOPVS are broken and discontinuous comprehensive analysis box plots are illustrated in Figs. 5,

19
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 7
Results of (PD Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 78987298.44 37130873 59953936.8 47409181.6 30687751.5 47852119.43 25006314.26 38300476.63 28955615.28 30716975.84
fmax 103324022.5 51156743.7 63191218.7 56497225.1 58969974.5 52485468.96 39550554.42 44117931.16 43491031.28 32487961.83
fmean 93805106.12 45782482.7 62315142.4 50448061.2 48165287.1 51158051.46 29616637.18 41429135.84 34311162.38 31702995.63
fmedian 96454551.77 47421157.1 63057707.2 48942919.1 51501711.3 52147308.73 26954840.01 41649067.77 32399001.49 31803522.44
fstd 11812360.41 6758026.18 1575394.81 4284187.18 13613949.8 2226895.328 6872690.463 2388505.028 6927683.918 921146.5344
Friedman Value 500 175 400 200 225 500 150 350 275 225
25 bar 120 bar
fmin 38319506.69 30244465.7 24732987 23884891.8 21858842 404452115.6 353046700.1 381216342.1 337410792.3 307562563
fmax 46306613.41 35159977.3 27366565.6 24677126.5 25283094.6 501645981.4 389752810.6 520976175.3 433359336.2 368012476.8
fmean 42649077.9 31748375 25862011.8 24211095.3 24038028.9 453060734.7 363719576 431370562.9 391421740.3 336366206.2
fmedian 42985095.75 30794528.4 25674247.4 24141181.4 24505089.5 453072421 356039396.6 411644867.1 397458416.4 334944892.5
fstd 4258854.62 2332777.88 1339257.52 333351.522 1627464.2 56101413.68 17583347.81 66268665.44 49418550.74 33340468.42
Friedman Value 500 400 250 150 200 450 200 400 275 175
37 bar 200 bar
fmin 15201455.05 8661154.6 9956042.9 7405482.27 9113888.58 109690129.1 41425176.67 38540318.6 65579756.33 42762060.02
fmax 19871677.42 9887808.4 12212982.5 10890323.6 11715337.7 202353054 45198049.22 62640272.92 70127711.39 47788143.17
fmean 18484636.64 9020324.77 10963625.3 9378729.35 10741882.6 135946735.1 43309751.54 55819842.28 67370493.51 45578763.28
fmedian 19432707.05 8766168.05 10842737.9 9609555.76 11069152.1 115871878.6 43307890.14 61049388.79 66887253.16 45882424.96
fstd 2227571.808 580437.286 1180084.66 1785675.93 1244628.67 44366685.25 2176121.742 11616917.99 2213227.544 2598917.815
Friedman Value 500 150 300 200 350 500 175 250 400 175
60 bar 942 bar
fmin 21567233.19 11568314 17147614.3 14494412.5 16078880.4 10228960659 1853813092 2485683233 2864130500 2922023806
fmax 29846703.79 15079365.2 21347348.3 17315543.2 24253574 13819418479 3151760961 3364547349 3076457616 3779611264
fmean 23751757.07 12609022.4 19302477.7 16344138.2 18219816.6 12826329490 2427157556 3034540595 3011810779 3534813028
fmedian 21796545.65 11894205.2 19357474 16783298.5 16273406 13628469411 2351528085 3143965899 3053327499 3718808520
fstd 4069045.42 1675311.4 1715709.89 1258416.52 4026683.69 1740912652 673462206.8 420875677.6 100839739.2 412528559.1
Friedman Value 500 125 375 225 275 500 175 225 225 375
Average Friedman Value 3950.00 1550.00 2550.00 1950.00 2000.00 1950.00 700.00 1225.00 1175.00 950.00

Table 8
Results of (SP Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 743.06 571.83 900.97 1398.87 861.82 545.98 223.00 381.94 525.34 529.19
fmax 922.43 914.07 1231.01 1869.51 1728.26 1968.85 507.31 465.94 824.94 809.12
fmean 875.85 680.73 1132.41 1689.12 1298.31 1425.36 389.51 404.50 602.81 666.80
fmedian 918.96 618.52 1198.84 1744.04 1301.57 1593.30 413.87 385.05 530.47 664.45
fstd 88.59 161.66 157.25 226.79 496.50 684.88 141.72 41.07 148.11 114.31
Friedman Value 250.00 150.00 300.00 500.00 300.00 500.00 150.00 150.00 325.00 375.00
25 bar 120 bar
fmin 257.34 365.76 317.95 409.84 443.30 5846.14 3786.44 4512.01 4410.81 8059.31
fmax 311.35 506.62 540.66 731.00 599.74 6502.80 8853.13 9012.92 7959.22 12812.01
fmean 275.71 467.91 405.92 609.65 502.10 6242.95 7430.74 7443.82 6448.49 10695.84
fmedian 267.08 499.63 382.54 648.88 482.68 6311.42 8541.68 8125.18 6711.96 10956.02
fstd 25.47 68.42 108.52 154.07 74.94 279.50 2447.21 1998.84 1796.72 2480.20
Friedman Value 100.00 325.00 275.00 425.00 375.00 200.00 325.00 350.00 175.00 450.00
37 bar 200 bar
fmin 108.91 97.49 116.83 42.84 134.94 1014.45 370.71 380.66 814.09 627.39
fmax 165.63 186.59 164.41 215.96 217.64 1390.64 794.18 956.04 1055.47 955.97
fmean 144.92 121.34 135.72 143.08 172.39 1170.93 591.24 740.34 994.87 738.81
fmedian 152.56 100.64 130.82 156.75 168.49 1139.30 600.03 812.32 1054.96 685.94
fstd 24.78 43.60 23.23 87.07 34.06 187.91 234.78 275.42 120.52 154.94
Friedman Value 325.00 225.00 275.00 300.00 375.00 450.00 175.00 225.00 425.00 225.00
60 bar 942 bar
fmin 167.04 93.98 105.91 166.59 263.58 76335.22 24240.48 46435.74 46239.11 95534.42
fmax 409.57 173.27 143.98 263.02 471.70 115280.47 61736.92 62951.95 57326.91 181255.53
fmean 300.65 124.10 126.29 237.88 334.50 102410.23 39919.98 53801.39 51978.39 138062.49
fmedian 312.98 114.58 127.64 260.95 301.35 109012.62 36851.27 52908.94 52173.76 137730.01
fstd 100.03 34.18 20.54 47.54 93.19 18360.32 18785.84 8629.08 6184.18 49110.19
Friedman Value 450.00 125.00 175.00 300.00 450.00 450.00 175.00 225.00 200.00 450.00
Average Friedman Value 2725.00 1650.00 1975.00 2650.00 3000.00 1600.00 825.00 950.00 1125.00 1500.00

8, 11, 14, 17, 20, 23, and 26 obtained by all algorithms for the Pareto-optimal fronts. Friedman’s non-parametric rank
all performance metrics. It is evident from the results that test exhibits the reliability and efficacy of the proposed
the proposed MOHTS/D algorithm realizes superior results methodology by generational distance, coverage, and spac-
in each performance metric qualitatively and quantitatively. ing of solutions against true PF. MOHTS/D displays the high-
• Comprehensively, the MOHTS/D methodology achieved the est CPF and DM metric value of 4000 among other con-
maximum PFHV value of 39253, which is significantly large trasted MO techniques, which manifest its outstanding qual-
in comparison to MOPVS, MOSMA, MOSOS, and MOMVO ity of coverage over PF and diversity in the solutions. More-
optimizers that illustrates its superiority and high efficiency over, MOHTS/D illustrates a higher pure diversity between
in terms of closeness and diversity of the solutions towards the solutions than MOPVS, MOSMA, MOSOS, and MOMVO,

20
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 9
Results of (SD Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 0.57 0.73 0.44 0.71 0.73 0.64 0.84 0.80 0.84 0.82
fmax 0.64 0.80 0.59 0.86 0.88 0.74 0.92 0.80 0.87 0.88
fmean 0.62 0.76 0.50 0.81 0.80 0.72 0.89 0.80 0.86 0.85
fmedian 0.63 0.76 0.48 0.83 0.79 0.74 0.90 0.80 0.87 0.85
fstd 0.04 0.03 0.07 0.07 0.08 0.05 0.03 0.00 0.01 0.02
Friedman Value 175.00 325.00 125.00 425.00 450.00 100.00 475.00 200.00 400.00 325.00
25 bar 120 bar
fmin 0.46 0.67 0.45 0.78 0.49 0.54 0.62 0.37 0.65 0.67
fmax 0.53 0.83 0.59 0.86 0.85 0.58 0.68 0.51 0.78 0.75
fmean 0.49 0.78 0.51 0.81 0.65 0.56 0.66 0.45 0.72 0.72
fmedian 0.48 0.81 0.50 0.80 0.63 0.57 0.67 0.45 0.73 0.73
fstd 0.04 0.08 0.06 0.04 0.18 0.02 0.03 0.08 0.07 0.04
Friedman Value 125.00 400.00 225.00 450.00 300.00 200.00 325.00 100.00 425.00 450.00
37 bar 200 bar
fmin 0.58 0.83 0.64 0.83 0.76 0.47 0.95 0.95 0.95 0.93
fmax 0.68 0.85 0.76 0.86 0.83 0.62 0.98 0.96 0.97 0.93
fmean 0.62 0.85 0.67 0.84 0.79 0.54 0.97 0.95 0.96 0.93
fmedian 0.60 0.85 0.64 0.84 0.78 0.53 0.97 0.95 0.96 0.93
fstd 0.05 0.01 0.06 0.01 0.03 0.08 0.01 0.01 0.01 0.00
Friedman Value 125.00 450.00 175.00 450.00 300.00 100.00 450.00 350.00 400.00 200.00
60 bar 942 bar
fmin 0.78 0.85 0.80 0.83 0.72 0.56 0.95 0.94 0.94 0.93
fmax 0.92 0.89 0.86 0.86 0.86 0.67 0.98 0.96 0.98 0.97
fmean 0.82 0.87 0.83 0.85 0.81 0.62 0.97 0.95 0.96 0.95
fmedian 0.80 0.86 0.83 0.85 0.83 0.62 0.97 0.95 0.97 0.94
fstd 0.07 0.02 0.03 0.02 0.07 0.04 0.01 0.01 0.02 0.02
Friedman Value 200.00 475.00 275.00 275.00 275.00 100.00 450.00 275.00 375.00 300.00
Average Friedman Value 1125.00 3350.00 1725.00 3200.00 2600.00 500.00 1700.00 925.00 1600.00 1275.00

Table 10
Results of (IGD Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 794.88 5358.48 1673.40 2655.17 1677.64 1044.66 8144.76 4074.54 3597.14 4486.25
fmax 1077.04 10380.50 2609.44 7095.26 14746.49 1255.71 9113.05 6244.88 7069.96 7521.40
fmean 957.54 7348.44 2046.26 4968.37 7700.99 1132.92 8659.38 5373.32 5667.02 5444.36
fmedian 979.11 6827.39 1951.11 5061.53 7189.92 1115.66 8689.86 5586.93 6000.49 4884.89
fstd 142.42 2450.29 397.62 2460.62 7005.06 88.43 526.21 1065.13 1709.05 1397.39
Friedman Value 100.00 400.00 250.00 400.00 350.00 100.00 500.00 300.00 325.00 275.00
25 bar 120 bar
fmin 151.41 428.00 358.04 415.88 690.16 4223.48 8391.18 9628.37 17793.55 21818.83
fmax 179.48 700.29 435.90 1329.23 1311.99 6193.19 27717.01 20562.69 42273.88 25547.73
fmean 162.74 588.28 408.41 740.13 926.89 5415.78 15012.41 16442.86 26687.26 23707.91
fmedian 160.05 612.41 419.85 607.71 852.71 5623.23 11970.72 17790.18 23340.80 23732.53
fstd 13.81 135.21 36.83 432.37 298.99 959.47 8636.19 5241.32 11633.03 1522.58
Friedman Value 100.00 400.00 250.00 325.00 425.00 100.00 250.00 325.00 400.00 425.00
37 bar 200 bar
fmin 116.11 1097.56 679.37 1500.59 843.63 4343.86 36777.78 25290.32 27906.80 35251.43
fmax 614.40 2032.89 794.39 3778.27 1569.59 7710.55 44174.67 45557.42 30385.95 35578.67
fmean 250.88 1792.36 754.41 2257.92 1072.15 5545.57 39477.11 32764.76 29316.79 35459.16
fmedian 136.51 2019.50 771.94 1876.42 937.68 5063.93 38478.00 30105.66 29487.21 35503.26
fstd 243.10 463.37 51.14 1073.72 343.28 1595.01 3518.11 9661.54 1265.53 142.97
Friedman Value 100.00 425.00 200.00 450.00 325.00 100.00 475.00 300.00 250.00 375.00
60 bar 942 bar
fmin 1234.46 3973.40 2766.86 2167.00 1801.72 822459.94 4727581.09 4837110.84 4859053.11 4297926.15
fmax 2625.20 4274.12 3433.80 3193.55 3318.24 1816455.13 5813051.40 5133029.87 5325291.67 5159273.51
fmean 1844.85 4121.67 3226.46 2874.98 2703.88 1108089.91 5489909.97 5004126.00 5030662.48 4759929.22
fmedian 1759.87 4119.59 3352.59 3069.68 2847.77 896722.28 5709503.69 5023181.66 4969152.58 4791258.62
fstd 576.19 171.25 315.82 486.22 747.31 477405.52 510558.03 152063.33 203160.91 353499.53
Friedman Value 125.00 500.00 400.00 250.00 225.00 100.00 425.00 375.00 325.00 275.00
Average Friedman Value 825.00 3375.00 2400.00 2725.00 2675.00 400.00 1650.00 1300.00 1300.00 1350.00

21
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 11
Results of (RT Metric) on truss bar problems.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO MOHTS/D MOPVS MOSMA MOSOS MOMVO
10 bar 72 bar
fmin 10.48 13.97 10.66 10.59 10.47 71.35 74.19 71.47 71.71 71.10
fmax 10.57 14.67 10.76 10.70 10.60 72.46 75.30 72.56 72.41 72.59
fmean 10.54 14.30 10.69 10.65 10.54 72.14 74.61 72.18 72.20 72.17
fmedian 10.55 14.28 10.67 10.66 10.54 72.37 74.48 72.36 72.34 72.49
fstd 0.04 0.28 0.05 0.05 0.07 0.53 0.48 0.49 0.33 0.72
Friedman Value 150.00 500.00 350.00 350.00 150.00 250.00 500.00 200.00 225.00 325.00
25 bar 120 bar
fmin 26.59 31.27 26.74 26.64 26.58 127.23 132.07 127.98 127.36 127.81
fmax 26.75 32.10 26.97 26.81 26.68 127.83 137.51 130.17 128.35 129.78
fmean 26.68 31.87 26.84 26.75 26.64 127.62 133.97 128.88 128.02 128.46
fmedian 26.69 32.05 26.82 26.77 26.65 127.71 133.16 128.68 128.18 128.12
fstd 0.07 0.40 0.09 0.07 0.05 0.28 2.41 0.92 0.47 0.89
Friedman Value 200.00 500.00 400.00 300.00 100.00 100.00 500.00 375.00 300.00 225.00
37 bar 200 bar
fmin 35.04 38.03 35.02 35.20 35.13 194.52 196.72 195.73 195.24 195.10
fmax 35.17 38.75 35.47 35.41 35.31 233.75 230.06 210.99 231.98 202.85
fmean 35.08 38.37 35.33 35.26 35.19 205.41 215.00 199.61 208.07 200.44
fmedian 35.06 38.36 35.42 35.20 35.17 196.69 216.61 195.86 202.54 201.89
fstd 0.06 0.29 0.21 0.11 0.08 18.92 13.74 7.59 16.30 3.58
Friedman Value 150.00 500.00 300.00 325.00 225.00 250.00 450.00 200.00 375.00 225.00
60 bar 942 bar
fmin 62.91 64.47 62.39 62.93 62.82 1003.05 1006.56 1000.64 1006.62 1004.65
fmax 63.29 65.05 62.84 65.11 63.96 1118.03 1103.80 1092.01 1115.26 1087.76
fmean 63.11 64.82 62.65 63.52 63.19 1033.65 1049.45 1026.81 1085.72 1051.91
fmedian 63.13 64.88 62.68 63.02 63.00 1006.75 1043.72 1007.28 1110.50 1057.60
fstd 0.16 0.24 0.19 1.06 0.52 56.28 40.24 43.58 52.78 42.43
Friedman Value 325.00 475.00 125.00 325.00 250.00 225.00 325.00 175.00 475.00 300.00
Average Friedman Value 1650.00 3750.00 2125.00 2675.00 1800.00 825.00 1775.00 950.00 1375.00 1075.00

Table 12
The Best Compromise Solution (BCS) results of all algorithms.
Algorithms MOHTS/D MOPVS MOSMA MOSOS MOMVO
10-bar truss (Best compromise solution)
fweight 7284.04 6565.23 3558.76 7039.02 5139.92
fcompliance 67200.76 74784.36 142077.39 69621.18 101386.06
25-bar truss (Best compromise solution)
fweight 4485.09 4418.94 2903.96 4007.17 2715.42
fcompliance 17809.72 18121.07 27913.42 19995.98 30971.76
37-bar truss (Best compromise solution)
fweight 2080.11 2358.67 1356.24 2006.21 1551.70
fcompliance 10316.75 9035.13 15910.00 10619.85 14367.78
60-bar truss (Best compromise solution)
fweight 5212.66 4345.27 4712.26 4390.29 4532.78
fcompliance 27131.81 30962.39 28422.08 30702.45 31092.10
72-bar truss (Best compromise solution)
fweight 11354.82 11856.50 9356.59 11695.69 9652.24
fcompliance 35797.20 33839.92 43174.81 34260.62 43119.15
120-bar truss (Best compromise solution)
fweight 50095.64 43673.45 20193.31 44942.20 45863.46
fcompliance 586512.45 674365.78 1456884.57 654491.23 642280.34
200-bar truss (Best compromise solution)
fweight 37925.11 38935.84 29967.57 34969.50 36175.52
fcompliance 40871.90 37675.12 49037.25 42344.60 41714.81
942-bar truss (Best compromise solution)
fweight 8845448.46 8723520.19 7363115.05 8911197.92 5920328.92
fcompliance 1554578.28 1346214.64 1497796.21 1345530.51 1897376.43

which is governed by its maximum value of 3750. Fried- and MOMVO algorithms as MOHTS/D realize the least value
man’s rank test also reflects the better spread of solution of 1125. For IGD metrics overall, the minimum value of
for MOHTS/D in comparison to MOPVS, MOSMA, MOSOS, 825 is achieved by MOHTS/D presents its uniformity in the

22
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Table 13
Results of (PFHV Metric) on various truss bar problems.
Algorithms MOHTS/D MOHTS MOEA/D MOHTS/D MOHTS MOEA/D
10 bar 72 bar
fmin 0.6499425 0.63682004 0.63648817 0.68920143 0.68790155 0.66892966
fmax 0.64227248 0.63874373 0.63865521 0.69086857 0.69205951 0.68304769
fmean 0.64111418 0.63765583 0.63784466 0.68425902 0.68975347 0.67476804
fmedian 0.64112086 0.63752976 0.63811762 0.68848304 0.68952641 0.6735474
fstd 0.00108186 0.00081587 0.00097359 0.01012641 0.00198485 0.00610089
FNRT 500 350 350 500 475 200
25 bar 120 bar
fmin 0.70502042 0.69428544 0.69060704 0.55237041 0.55244286 0.54756314
fmax 0.70575509 0.69517566 0.69189222 0.56285488 0.55415167 0.54998063
fmean 0.70540008 0.6947503 0.69150433 0.56412256 0.55338614 0.54905443
fmedian 0.70254124 0.69477004 0.69175904 0.56546199 0.55347501 0.54933697
fstd 0.00036513 0.00038916 0.00060177 0.00117938 0.0008176 0.00110617
FNRT 500 400 200 500 425 200
37 bar 200 bar
fmin 0.6903485 0.69762574 0.69977843 0.77777623 0.72871705 0.76490123
fmax 0.71382605 0.70100658 0.70362961 0.78991707 0.76358039 0.78056528
fmean 0.70682695 0.6996071 0.70173602 0.78895504 0.75086943 0.77055099
fmedian 0.71656662 0.69989804 0.70176802 0.78906344 0.75559014 0.76836872
fstd 0.00551566 0.00145075 0.00207876 0.01207844 0.01620452 0.00702604
FNRT 500 350 475 500 250 425
60 bar 942 bar
fmin 0.52989938 0.53350635 0.41349182 0.71923325 0.68455839 0.71389486
fmax 0.53195877 0.54536367 0.50704563 0.75996827 0.7223464 0.72963682
fmean 0.53563573 0.53859538 0.48209576 0.74382084 0.69679701 0.72003647
fmedian 0.53034239 0.53775575 0.5039228 0.74079913 0.69014163 0.7183071
fstd 0.02029765 0.0057487 0.04576146 0.01836825 0.01731709 0.00744751
FNRT 450 475 200 475 225 425
Average FNRT 3925 2950 2475 1 2 3

Fig. 21. The 200-bar 3D truss.

convergence-spread of the non-dominated set of solutions minimum structure mass f1 (Z ) and having the maximum
concerning MOPVS, MOSMA, MOSOS, and MOMVO optimiz- nodal deflection f2 (Z ) are chosen for all considered MO
ers. Moreover, MOHTS/D realizes the minimum RT metric optimization techniques and illustrated in Fig. 4 for 10-bar,
value of 1650 that manifests its potential to reach a glob- Fig. 7 for 25-bar, Fig. 10 for 37-bar, Fig. 13 for 60-bar, Fig. 16
ally optimal solution within the least computation time in for 72-bar, Fig. 19 for 120-bar, Fig. 22 for 200-bar, and Fig. 25
respect of other contrasted algorithms. for 942-bar along with their respective BCS points presented
• As explained in Section 3.1, the best-compromised solution in Table 12. On the basis of these results, it is evident
(BCS) that satisfied each objective is reported from the ob- that the proposed MOHTS/D methodology affords the best-
tained optimal PFs based on the fuzzy decision technique. compromised solution for all the eight considered bench-
The results obtained from the best optimization runs having marks and has the potential to create harmony between the

23
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 22. Best Pareto fronts of the 200-bar truss by all algorithms.

Fig. 23. Box plots of the 200-bar truss by all algorithms.

24
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

MOHTS/D with MOHTS and MOEA/D [79] methods. Therefore,

for all the eight truss case studies, MOHTS and MOEA/D were
employed and the HV metric results are presented in Table 13.
The statistical results show that the PFHV mean value obtained
by MOHTS/D is higher than MOHTS and MOEA/D in all truss
problems that demonstrate the high volume solutions enclosed
near the PF of MOHTS/D relatively. The STD value obtained from
the PFHV test for MOHTS/D has the least value for all truss
problems, which shows the superiority of MOHTS/D in terms of
search consistency. Also, MOHTS/D achieved the maximum total
PFHV value of 3925, which is significantly large in comparison
to MOHTS and MOEA/D optimizers and ranked first, which illus-
trates its superiority and high performance in solving large scale
optimization problems.

6. Conclusions

A novel MOHTS/D algorithm is proposed and examined for

the MO structure optimization design issues. To evaluate the
effectiveness of the proposed method, eight challenging both
2-D and 3-D benchmark problems that are subject to multiple
constraints with discrete design variables were examined to ac-
count for the practicality of the investigation. At the same input
condition, the results obtained from the simulation are compared
with the other six population-based distinguished algorithms.
The performance metrics, i.e., RT, HV, GD, CVG, PD, IGD, SP, SD,
CPF, and DM, are employed in all test examples to concurrently
examine the progression and diversity of the non-dominated set
of solutions procured from the computation experiments. Out-
comes obtained demonstrate that MOHTS/D outperforms other
examined algorithms in terms of search consistency and conver-
gence rate. Moreover, based on Friedman’s rank test, MOHTS/D
ranked first in all the design problems and showed its compe-
tency in solving large real-life challenging optimization problems
efficiently. Overall the investigation illustrated the efficacy of the
proposed MOHTS/D technique over others both qualitatively and
quantitatively. The proposed MOHTS/D algorithm does not have
algorithm-specific parameters that make it robust and suitable for
a comprehensive application.
In the future, one can investigate this algorithm for higher
dimensions and intricate engineering design problems including
the design of box-beam structures for aero-elastic optimization of
aircraft structures, battery management systems, lighter compos-
ite bumper beams with higher crashworthiness characteristics,
to optimize appropriate size, shape, and profile of ultrasonic
food drying equipment and operating conditions. As this algo-
rithm is relatively new in this field, more research is needed
for the prediction of its global optimization behaviour, which
can be contrasted with other prominent algorithms. Even though
the presented approach performed better, more refining like an
adaptive or Elite selection strategy is expected to improve the
search efficiency even more. Interested researchers can addition-
ally work on improving the distribution of the MOHTS/D algo-
rithm’s Pareto-optimal solution set and using it to solve dynamic
multi-objective optimization issues. Because objective functions
in human judgement contain competing behaviours, a decision-
Fig. 24. The 942-bar tower truss.
maker can efficiently extract a Pareto-optimal solution by deter-
mining his priorities for corresponding fuzzy objectives in a fuzzy
set.
local intensification and global diversification of search in
complex large structural MO optimization problems.
Declaration of competing interest
Since the suggested MOHTS/D algorithm is inspired by the strat-
egy of decomposition and developed to improve the performance The authors declare that they have no known competing finan-
of MOHTS (which has been employed for structure optimization cial interests or personal relationships that could have appeared
problems [57]), it is essential to compare the performance of to influence the work reported in this paper.
25
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

Fig. 25. Best Pareto fronts of the 942-bar truss by all algorithms.

Fig. 26. Box plots of the 942-bar truss by all algorithms.

26
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

References [24] V.K. Chouhan, S.H. Khan, M. Hajiaghaei-Keshteli, Sustainable planning and
decision-making model for sugarcane mills considering environmental
[1] N. Srinivas, K. Deb, Muiltiobjective optimization using nondominated issues, J. Environ. Manag. 303 (2022) 114252, http://dx.doi.org/10.1016/
sorting in genetic algorithms, Evol. Comput. 2 (3) (1994) 221–248. j.jenvman.2021.114252.
[2] F.W. Glover, G. A. Kochenberger (Eds.), Handbook of Metaheuristics, Vol. [25] A.M. Fathollahi-Fard, M. Hajiaghaei-Keshteli, R. Tavakkoli-Moghaddam,
57, Springer Science & Business Media, 2006. Red deer algorithm (RDA): a new nature-inspired meta-heuristic, Soft
Comput. 24 (19) (2020) 14637–14665, http://dx.doi.org/10.1007/s00500-
[3] K. Deb, A. Pratap, S. Agarwal, T.A.M.T. Meyarivan, A fast and elitist
020-04812-z.
multiobjective genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput. 6 (2)
[26] V.K. Chouhan, S.H. Khan, M. Hajiaghaei-Keshteli, Metaheuristic approaches
(2002) 182–197.
to design and address multi-echelon sugarcane closed-loop supply chain
[4] V. Escandon-Bailon, H. Cervantes, A. García-Nájera, S. Zapotecas-Martínez,
network, Soft Comput. 25 (16) (2021) 11377–11404, http://dx.doi.org/10.
Analysis of the multi-objective release plan rescheduling problem, Knowl.-
1007/s00500-021-05943-7.
Based Syst. 220 (2021) 106922, http://dx.doi.org/10.1016/j.knosys.2021.
[27] C.M. Fonseca, P.J. Fleming, Genetic algorithms for multiobjective optimiza-
106922.
tion: formulation discussion and generalization, in: Icga, Vol. 93, No. July,
[5] H. Dong, J. Li, P. Wang, B. Song, X. Yu, Surrogate-guided multi-objective
1993, pp. 416–423.
optimization (SGMOO) using an efficient online sampling strategy, Knowl.-
[28] S. Bandyopadhyay, S. Saha, U. Maulik, K. Deb, A simulated annealing-based
Based Syst. 220 (2021) 106919, http://dx.doi.org/10.1016/j.knosys.2021.
multiobjective optimization algorithm: AMOSA, IEEE Trans. Evol. Comput.
106919.
12 (3) (2008) 269–283, http://dx.doi.org/10.1109/TEVC.2007.900837.
[6] D. Paul, A. Jain, S. Saha, J. Mathew, Multi-objective PSO based online feature [29] T. Robič, B. Filipič, Differential evolution for multiobjective optimization,
selection for multi-label classification, Knowl.-Based Syst. (2021) 106966, in: International Conference on Evolutionary Multi-Criterion Optimization,
http://dx.doi.org/10.1016/j.knosys.2021.106966. Springer, Berlin, Heidelberg, 2005, pp. 520–533, http://dx.doi.org/10.1007/
[7] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, P. Jangir, Multi-objective 978-3-540-31880-4_36.
teaching-learning-based optimization for structure optimization, Smart Sci. [30] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, Multi-Objective Passing
10 (1) (2022) 56–67, http://dx.doi.org/10.1080/23080477.2021.1975074. Vehicle Search algorithm for structure optimization, Expert Syst. Appl. 169
[8] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, Performance enhancement (2021) 114511, http://dx.doi.org/10.1016/j.eswa.2020.114511.
of meta-heuristics through random mutation and simulated annealing- [31] C.C. Coello, M.S. Lechuga, MOPSO: A proposal for multiple objective
based selection for concurrent topology and sizing optimization of truss particle swarm optimization, in: Proceedings of the 2002 Congress on
structures, Soft Comput. (2022) 1–23, http://dx.doi.org/10.1007/s00500- Evolutionary Computation. CEC’02 (Cat. No. 02TH8600), Vol. 2, IEEE, 2002,
022-06930-2. pp. 1051–1056, http://dx.doi.org/10.1109/CEC.2002.1004388.
[9] A.A. Heidari, S. Mirjalili, H. Faris, I. Aljarah, M. Mafarja, H. Chen, Harris [32] K. Deb, A. Pratap, S. Agarwal, T.A.M.T. Meyarivan, A fast and elitist
hawks optimization: Algorithm and applications, Future Gener. Comput. multiobjective genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput. 6 (2)
Syst. 97 (2019) 849–872, http://dx.doi.org/10.1016/j.future.2019.02.028. (2002) 182–197, http://dx.doi.org/10.1109/4235.996017.
[10] A. Faramarzi, M. Heidarinejad, B. Stephens, S. Mirjalili, Equilibrium op- [33] D. Angus, C. Woodward, Multiple objective ant colony optimisation, Swarm
timizer: A novel optimization algorithm, Knowl.-Based Syst. 191 (2020) Intell. 3 (1) (2009) 69–85, http://dx.doi.org/10.1007/s11721-008-0022-4.
105190, http://dx.doi.org/10.1016/j.knosys.2019.105190. [34] G.G. Tejani, N. Pholdee, S. Bureerat, D. Prayogo, Multiobjective adaptive
[11] G. Dhiman, V. Kumar, Seagull optimization algorithm: Theory and its symbiotic organisms search for truss optimization problems, Knowl.-Based
applications for large-scale industrial engineering problems, Knowl.-Based Syst. 161 (2018) 398–414, http://dx.doi.org/10.1016/j.knosys.2018.08.005.
Syst. 165 (2019) 169–196, http://dx.doi.org/10.1016/j.knosys.2018.11.024. [35] G.G. Tejani, V.J. Savsani, V.K. Patel, S. Mirjalili, Truss optimization with
[12] Z. Yang, L. Deng, Y. Wang, J. Liu, Aptenodytes Forsteri Optimization: natural frequency bounds using improved symbiotic organisms search,
Algorithm and applications, Knowl.-Based Syst. 232 (2021) 107483, http: Knowl.-Based Syst. 143 (2018) 162–178, http://dx.doi.org/10.1016/j.knosys.
//dx.doi.org/10.1016/j.knosys.2021.107483. 2017.12.012.
[13] F.A. Hashim, A.G. Hussien, Snake Optimizer: A novel meta-heuristic opti- [36] E. Zitzler, M. Laumanns, L. Thiele, SPEA2: Improving the Strength Pareto
mization algorithm, Knowl.-Based Syst. 242 (2022) 108320, http://dx.doi. Evolutionary Algorithm, TIK-Report, 2001, p. 103, http://dx.doi.org/10.
org/10.1016/j.knosys.2022.108320. 3929/ethz-a-004284029.
[14] M. Braik, A. Hammouri, J. Atwan, M.A. Al-Betar, M.A. Awadallah, White [37] D. Gong, B. Xu, Y. Zhang, Y. Guo, S. Yang, A similarity-based cooperative
Shark Optimizer: A novel bio-inspired meta-heuristic algorithm for global co-evolutionary algorithm for dynamic interval multiobjective optimization
optimization problems, Knowl.-Based Syst. 243 (2022) 108457, http://dx. problems, IEEE Trans. Evol. Comput. 24 (1) (2019) 142–156, http://dx.doi.
doi.org/10.1016/j.knosys.2022.108457. org/10.1109/TEVC.2019.2912204.
[15] H. Karami, M.V. Anaraki, S. Farzin, S. Mirjalili, Flow direction algorithm [38] M. Rong, D. Gong, Y. Zhang, Y. Jin, W. Pedrycz, Multidirectional prediction
(FDA): A novel optimization approach for solving optimization problems, approach for dynamic multiobjective optimization problems, IEEE Trans.
Comput. Ind. Eng. 156 (2021) 107224, http://dx.doi.org/10.1016/j.cie.2021. Cybern. 49 (9) (2018) 3362–3374, http://dx.doi.org/10.1109/TCYB.2018.
107224. 2842158.
[39] S. Winyangkul, K. Wansaseub, S. Sleesongsom, N. Panagant, S. Kumar, S.
[16] L. Abualigah, A. Diabat, S. Mirjalili, M. Abd Elaziz, A.H. Gandomi, The
Bureerat, N. Pholdee, Ground structures-based topology optimization of
arithmetic optimization algorithm, Comput. Methods Appl. Mech. Engrg.
a morphing wing using a metaheuristic algorithm, Metals 11 (8) (2021)
376 (2021) 113609, http://dx.doi.org/10.1016/j.cma.2020.113609.
1311, http://dx.doi.org/10.3390/met11081311.
[17] F. MiarNaeimi, G. Azizyan, M. Rashki, Horse herd optimization algorithm:
[40] P. Jangir, A.A. Heidari, H. Chen, Elitist non-dominated sorting harris hawks
A nature-inspired algorithm for high-dimensional optimization problems,
optimization: Framework and developments for multi-objective problems,
Knowl.-Based Syst. 213 (2021) 106711, http://dx.doi.org/10.1016/j.knosys.
Expert Syst. Appl. 186 (2021) 115747, http://dx.doi.org/10.1016/j.eswa.
2020.106711.
2021.115747.
[18] S. Li, H. Chen, M. Wang, A.A. Heidari, S. Mirjalili, Slime mould algorithm: A [41] K. Wansasueb, S. Bureerat, S. Kumar, Ensemble of four metaheuristic using
new method for stochastic optimization, Future Gener. Comput. Syst. 111 a weighted sum technique for aircraft wing design, Eng. Appl. Sci. Res. 48
(2020) 300–323, http://dx.doi.org/10.1016/j.future.2020.03.055. (4) (2021) 385–396, https://ph01.tci-thaijo.org/index.php/easr/article/view/
[19] M. Hajiaghaei-Keshteli, M. Aminnayeri, Keshtel Algorithm (KA); a new 242706.
optimization algorithm inspired by Keshtels’ feeding, in: Proceeding in IEEE [42] V.K. Patel, V.J. Savsani, Heat transfer search (HTS): a novel optimization
Conference on Industrial Engineering and Management Systems, 2013, pp. algorithm, Inform. Sci. 324 (2015) 217–246, http://dx.doi.org/10.1016/j.ins.
2249–2253. 2015.06.044.
[20] B. Mosallanezhad, M. Hajiaghaei-Keshteli, C. Triki, Shrimp closed-loop [43] S.O. Degertekin, L. Lamberti, M.S. Hayalioglu, Heat transfer search algo-
supply chain network design, Soft Comput. 25 (11) (2021) 7399–7422, rithm for sizing optimization of truss structures, Latin Amer. J. Solids
http://dx.doi.org/10.1007/s00500-021-05698-1. Struct. 14 (3) (2017) 373–397, http://dx.doi.org/10.1590/1679-78253297.
[21] B. Mosallanezhad, V.K. Chouhan, M.M. Paydar, M. Hajiaghaei-Keshteli, [44] A. Hazra, S. Das, M. Basu, Heat transfer search algorithm for non-convex
Disaster relief supply chain design for personal protection equipment economic dispatch problems, J. Inst. Eng. (India): Ser. B 99 (3) (2018)
during the COVID-19 pandemic, Appl. Soft Comput. 112 (2021) 107809, 273–280, http://dx.doi.org/10.1007/s40031-018-0320-1.
http://dx.doi.org/10.1016/j.asoc.2021.107809. [45] B.D. Raja, V. Patel, R.L. Jhala, Thermal design and optimization of fin-and-
[22] A.M. Fathollahi-Fard, M. Hajiaghaei-Keshteli, R. Tavakkoli-Moghaddam, The tube heat exchanger using heat transfer search algorithm, Therm. Sci. Eng.
social engineering optimizer (SEO), Eng. Appl. Artif. Intell. 72 (2018) Prog. 4 (2017) 45–57, http://dx.doi.org/10.1016/j.tsep.2017.08.004.
267–293, http://dx.doi.org/10.1016/j.engappai.2018.04.009. [46] R. Chaudhari, J.J. Vora, S.S. Mani Prabu, I.A. Palani, V.K. Patel, D.M.
[23] A. Salehi-Amiri, A. Zahedi, N. Akbapour, M. Hajiaghaei-Keshteli, Designing a Parikh, L.N.L. de Lacalle, Multi-response optimization of WEDM process
sustainable closed-loop supply chain network for walnut industry, Renew. parameters for machining of superelastic nitinol shape-memory alloy using
Sustain. Energy Rev. 141 (2021) 110821, http://dx.doi.org/10.1016/j.rser. a heat-transfer search algorithm, Materials 12 (8) (2019) 1277, http://dx.
2021.110821. doi.org/10.3390/ma12081277.

27
S. Kumar, P. Jangir, G.G. Tejani et al. Knowledge-Based Systems 253 (2022) 109591

[47] J.K. Pattanaik, M. Basu, D.P. Dash, Heat transfer search algorithm for [67] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, Improved metaheuristics
combined heat and power economic dispatch, Iran. J. Sci. Technol. Trans. through migration-based search and an acceptance probability for truss
Electr. Eng. 44 (2) (2020) 963–978, http://dx.doi.org/10.1007/s40998-019- optimization, Asian J. Civil Eng. 21 (7) (2020) 1217–1237, http://dx.doi.
00280-w. org/10.1007/s42107-020-00271-x.
[48] A. Hazra, S. Das, A. Laddha, M. Basu, Economic power generation strategy [68] S. Kumar, G.G. Tejani, S. Mirjalili, Modified symbiotic organisms search
for wind integrated large power network using heat transfer search for structural optimization, Eng. Comput. 35 (4) (2019) 1269–1296, http:
algorithm, J. Inst. Eng. (India): Ser. B (2020) 1–7, http://dx.doi.org/10.1007/ //dx.doi.org/10.1007/s00366-018-0662-y.
s40031-020-00427-y. [69] R.T. Marler, J.S. Arora, Survey of multi-objective optimization methods
[49] G. Tejani, V. Savsani, V. Patel, Modified sub-population based heat transfer for engineering, Struct. Multidiscip. Optim. 26 (6) (2004) 369–395, http:
search algorithm for structural optimization, Int. J. Appl. Metaheuristic //dx.doi.org/10.1007/s00158-003-0368-6.
Comput. 8 (3) (2017) 1–23, http://dx.doi.org/10.4018/IJAMC.2017070101. [70] E. Zitzler, K. Deb, L. Thiele, Comparison of multiobjective evolutionary
[50] P. Savsani, M.A. Tawhid, Discrete heat transfer search for solving travelling algorithms: Empirical results, Evol. Comput. 8 (2) (2000) 173–195, http:
salesman problem, Math. Found. Comput. 1 (3) (2018) 265, http://dx.doi. //dx.doi.org/10.1162/106365600568202.
org/10.3934/mfc.2018012. [71] D.H. Wolpert, W.G. Macready, No free lunch theorems for optimization,
[51] D. Maharana, P. Kotecha, Simultaneous heat transfer search for compu- IEEE Trans. Evol. Comput. 1 (1) (1997) 67–82, http://dx.doi.org/10.1109/
tationally expensive numerical optimization, in: 2016 IEEE Congress on
4235.585893.
Evolutionary Computation, CEC, IEEE, 2016, pp. 2982–2988, http://dx.doi.
[72] A. Hertz, D. de Werra, Using tabu search techniques for graph
org/10.1109/CEC.2016.7744166.
coloring, Computing 39 (4) (1987) 345–351, http://dx.doi.org/10.1007/
[52] E. Alnahari, H. Shi, K. Alkebsi, Quadratic interpolation based simultaneous
BF02239976.
heat transfer search algorithm and its application to chemical dynamic
[73] R. Kuik, M. Salomon, L.N. van Wassenhove, J. Maes, Linear programming,
system optimization, Processes 8 (4) (2020) 478, http://dx.doi.org/10.3390/
simulated annealing and tabu search heuristics for lot sizing in bottleneck
pr8040478.
assembly systems, IIE Trans. 25 (1) (1993) 62–72, http://dx.doi.org/10.
[53] G.G. Tejani, V.J. Savsani, V.K. Patel, S. Mirjalili, An improved heat transfer
1080/07408179308964266.
search algorithm for unconstrained optimization problems, J. Comput. Des.
Eng. 6 (1) (2019) 13–32, http://dx.doi.org/10.1016/j.jcde.2018.04.003. [74] J.K. Lee, Y.D. Kim, Search heuristics for resource constrained project
[54] V. Savsani, V. Patel, B. Gadhvi, M. Tawhid, Pareto optimization of a half car scheduling, J. Oper. Res. Soc. 47 (5) (1996) 678–689, http://dx.doi.org/10.
passive suspension model using a novel multiobjective heat transfer search 1057/jors.1996.79.
algorithm, Modell. Simul. Eng. 2017 (2017) http://dx.doi.org/10.1155/2017/ [75] X.S. Yang, Review of metaheuristics and generalized evolutionary walk
2034907. algorithm, Int. J. Bio-Inspir. Comput. 3 (2) (2011) 77, http://dx.doi.org/10.
[55] M.A. Tawhid, V. Savsani, ∈-constraint heat transfer search (∈-HTS) algo- 1504/IJBIC.2011.039907.
rithm for solving multi-objective engineering design problems, J. Comput. [76] K. Sörensen, Metaheuristics—the metaphor exposed, Int. Trans. Oper. Res.
Des. Eng. 5 (1) (2018) 104–119, http://dx.doi.org/10.1016/j.jcde.2017.06. 22 (1) (2015) 3–18, http://dx.doi.org/10.1111/itor.12001.
003. [77] M. Mernik, S.H. Liu, D. Karaboga, M. Črepinšek, On clarifying misconcep-
[56] B.D. Raja, R.L. Jhala, V. Patel, Thermal-hydraulic optimization of plate tions when comparing variants of the Artificial Bee Colony Algorithm by
heat exchanger: A multi-objective approach, Int. J. Therm. Sci. 124 (2018) offering a new implementation, Inform. Sci. 291 (2015) 115–127, http:
522–535, http://dx.doi.org/10.1016/j.ijthermalsci.2017.10.035. //dx.doi.org/10.1016/j.ins.2014.08.040.
[57] G.G. Tejani, S. Kumar, A.H. Gandomi, Multi-objective heat transfer search [78] M. Črepinšek, S.H. Liu, M. Mernik, Replication and comparison of compu-
algorithm for truss optimization, Eng. Comput. (2019) 1–22, http://dx.doi. tational experiments in applied evolutionary computing: common pitfalls
org/10.1007/s00366-019-00846-6. and guidelines to avoid them, Appl. Soft Comput. 19 (2014) 161–170,
[58] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, Multi-objective modified http://dx.doi.org/10.1016/j.asoc.2014.02.009.
heat transfer search for truss optimization, Eng. Comput. (2020) 1–16, [79] Q. Zhang, H. Li, MOEA/D: A multiobjective evolutionary algorithm based
http://dx.doi.org/10.1007/s00366-020-01010-1. on decomposition, IEEE Trans. Evol. Comput. 11 (6) (2007) 712–731, http:
[59] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, Multiobjecitve structural //dx.doi.org/10.1109/TEVC.2007.892759.
optimization using improved heat transfer search, Knowl.-Based Syst. [80] A. Trivedi, D. Srinivasan, K. Sanyal, A. Ghosh, A survey of multiob-
(2021) 106811, http://dx.doi.org/10.1016/j.knosys.2021.106811. jective evolutionary algorithms based on decomposition, IEEE Trans.
[60] B.S. Yıldız, S. Kumar, N. Pholdee, S. Bureerat, S.M. Sait, A.R. Yildiz, A Evol. Comput. 21 (3) (2016) 440–462, http://dx.doi.org/10.1109/TEVC.2016.
new chaotic Lévy flight distribution optimization algorithm for solving 2608507.
constrained engineering problems, Expert Syst. (2022) e12992, http://dx. [81] L. Wang, Multi-objective optimization based on decomposition for
doi.org/10.1111/exsy.12992. flexible job shop scheduling under time-of-use electricity prices, Knowl.-
[61] T. Kunakote, N. Sabangban, S. Kumar, G.G. Tejani, N. Panagant, N. Pholdee, Based Syst. 204 (2020) 106177, http://dx.doi.org/10.1016/j.knosys.2020.
et al., Comparative performance of twelve metaheuristics for wind farm 106177.
layout optimisation, Arch. Comput. Methods Eng. 29 (1) (2022) 717–730, [82] W. Li, Z. Chai, Z. Tang, A decomposition-based multi-objective immune
http://dx.doi.org/10.1007/s11831-021-09586-7. algorithm for feature selection in learning to rank, Knowl.-Based Syst. 234
[62] S. Kumar, P. Jangir, G.G. Tejani, M. Premkumar, H.H. Alhelou, MOPGO: A
(2021) 107577, http://dx.doi.org/10.1016/j.knosys.2021.107577.
new physics-based multi-objective plasma generation optimizer for solving
[83] M. Menor-Flores, M.A. Vega-Rodríguez, Decomposition-based multi-
structural optimization problems, IEEE Access 9 (2021) 84982–85016, http:
objective optimization approach for PPI network alignment, Knowl.-Based
//dx.doi.org/10.1109/ACCESS.2021.3087739.
Syst. (2022) 108527, http://dx.doi.org/10.1016/j.knosys.2022.108527.
[63] S. Kumar, P. Jangir, G.G. Tejani, M. Premkumar, MOTEO: A novel physics-
[84] S. Vinodh, S. Sarangan, S.C. Vinoth, Application of fuzzy compromise
based multiobjective thermal exchange optimization algorithm to design
solution method for fit concept selection, Appl. Math. Model. 38 (3) (2014)
truss structures, Knowl.-Based Syst. 242 (2022) 108422, http://dx.doi.org/
1052–1063, http://dx.doi.org/10.1016/j.apm.2013.07.027.
10.1016/j.knosys.2022.108422.
[85] M. Premkumar, P. Jangir, R. Sowmya, H.H. Alhelou, A.A. Heidari, H. Chen,
[64] M. Fasihi, R. Tavakkoli-Moghaddam, S.E. Najafi, M. Hajiaghaei, Optimizing
MOSMA: Multi-objective slime mould algorithm based on elitist non-
a bi-objective multi-period fish closed-loop supply chain network design
by three multi-objective meta-heuristic algorithms, Scientia Iranica (2021) dominated sorting, IEEE Access 9 (2020) 3229–3248, http://dx.doi.org/10.
http://dx.doi.org/10.24200/sci.2021.57930.5477. 1109/ACCESS.2020.3047936.
[65] Z. Wu, X. Zhao, Y. Ma, X. Zhao, A hybrid model based on modified multi- [86] D.H. Tran, M.Y. Cheng, D. Prayogo, A novel Multiple Objective Symbiotic
objective cuckoo search algorithm for short-term load forecasting, Appl. Organisms Search (MOSOS) for time–cost–labor utilization tradeoff prob-
Energy 237 (2019) 896–909, http://dx.doi.org/10.1016/j.apenergy.2019.01. lem, Knowl.-Based Syst. 94 (2016) 132–145, http://dx.doi.org/10.1016/j.
046. knosys.2015.11.016.
[66] S. Kumar, G.G. Tejani, N. Pholdee, S. Bureerat, P. Mehta, Hybrid Heat [87] S. Mirjalili, P. Jangir, S.Z. Mirjalili, S. Saremi, I.N. Trivedi, Optimization
Transfer Search and Passing Vehicle Search optimizer for multi-objective of problems with multiple objectives using the multi-verse optimization
structural optimization, Knowl.-Based Syst. 212 (2021) 106556, http://dx. algorithm, Knowl.-Based Syst. 134 (2017) 50–71, http://dx.doi.org/10.1016/
doi.org/10.1016/j.knosys.2020.106556. j.knosys.2017.07.018.

28