See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/379515233

Sizing and Shape Optimization of Discrete Truss Employing a Target-Oriented

Krill Herd Algorithm

Article in ASCE-ASME J Risk and Uncert in Engrg Sys Part B Mech Engrg · April 2024
DOI: 10.1115/1.4064644

CITATIONS READS

0 30

4 authors, including:

Cheng Lixang Yan-Gang Zhao

Yunnan University of Finance and Economics Kanagawa University
2 PUBLICATIONS 0 CITATIONS 282 PUBLICATIONS 4,843 CITATIONS

SEE PROFILE SEE PROFILE

Peipei Li
Technische Universität Dortmund
22 PUBLICATIONS 103 CITATIONS

SEE PROFILE

All content following this page was uploaded by Cheng Lixang on 22 November 2024.

The user has requested enhancement of the downloaded file.

 Sizing and Shape Optimization
Lixiang Cheng of Discrete Truss Employing
Department of Architecture,
Kanagawa University,
Yokohama 221-8686, Japan
a Target-Oriented Krill Herd
Yan-Gang Zhao Algorithm
School of Civil Engineering,
Beijing University of Technology, The krill herd (KH) algorithm is widely used for optimizing truss structures as no gradient
Beijing 100124, China information is necessary, and only a few parameters require adjustment. However, when the
truss structure becomes discrete and complex, KH tends to fall into a local optimum.
Pei-Pei Li1 Therefore, a novel target-oriented KH (TOKH) algorithm is proposed in this study to
Chair for Reliability Engineering, optimize the design of discrete truss structures. Initially, a crossover operator is established
TU Dortmund University, between the “best krill” and “suboptimal krill” to generate a robust “cross krill” for global
Dortmund 44227, Germany exploration. Additionally, an improved local mutation and crossover (ILMC) operator is
e-mail: [email protected] introduced to fine-tune the “center of food” and candidate solutions for local exploitation.
The proposed method and other optimization approaches are experimentally compared
Lewei Yan considering 15 benchmark functions. Then, the performance of the TOKH algorithm is
College of Civil Engineering, evaluated based on four discrete truss structure optimization problems under multiple
Guangzhou University, loading conditions. The obtained optimization results indicate that the proposed method
Guangzhou 510006, China presents competitive solutions in terms of accuracy, unlike other algorithms in the literature,
and avoids falling into a local minimum. [DOI: 10.1115/1.4064644]

Keywords: discrete truss optimization, krill herd algorithm, target-oriented, robustness

1 Introduction [4]. Modern metaheuristic algorithms include the genetic algorithm

(GA) [5–7], biogeography-based optimization [8–10], harmony
Truss optimization has gained considerable attention because of
search (HS) [11–13], differential evolution [14,15], ant colony
its direct applicability to the design of structures. Designers and
optimization [16], particle swarm optimization [17–19], artificial
owners require optimized trusses to reduce building costs [1].
bee colony [20–22], teaching-learning based optimization (TLBO)
However, the implementation of discrete truss optimization
[23], artificial fish swarm [24], firefly algorithm [25–27], political
problems is challenging as several truss designs generally entail a
optimizer (PO) [28], cuckoo search [29,30], bat algorithm [31,32],
complex design space. Traditionally, researchers have used
manta ray foraging optimization (MRFO) [33] and krill herd (KH)
mathematical approaches that employ rounding-off techniques
algorithm [34]. Among these algorithms, the KH algorithm is known
based on continuous solutions to solve discrete truss optimization
for its powerful exploitation ability, adjustment of fewer parameters,
problems. However, these methods may become infeasible or
and easy implementation.
generate increasingly suboptimal solutions with numerous variables
The KH algorithm was inspired by the herding behavior of
[2]. Therefore, simulation-based metaheuristic algorithms to solve
Antarctic krill swarms. During optimization, each krill is primarily
truss optimization problems are required.
affected by other krill individuals, foraging action, and physical
Metaheuristic algorithms combine rules and randomness to
diffusion. Foraging and other krill-induced motions include two
imitate natural phenomena and attempt to identify the optimum
global and two local strategies that operate in parallel, rendering KH
design in a reasonable amount of computing time using trial-and-
a robust algorithm [34]. Compared to earlier metaheuristic
error techniques [3]. The ability to balance exploitation (intensifi-
algorithms, the KH algorithm imposes fewer mathematical require-
cation) and exploration (diversification) during a search determines
ments and can be easily adapted to solve various engineering
the efficiency of a specific metaheuristic algorithm. Exploration
optimization problems. Furthermore, rather than using a gradient
ensures the validity and breadth of the algorithm in the search area,
search, the KH algorithm uses a stochastic search based on the krill
which can be beneficial for global optimization. Exploitation
population, which eliminates the need for derivative information.
expands the local search for the currently explored optimal area
These features have increased the flexibility of the KH algorithm and
and further finds the minimum [4]. To address global search
produced better solutions. Consequently, it has been increasingly
requirements, modern metaheuristic algorithms have evolved to
investigated and successfully employed in various practical
incorporate three primary purposes, namely, solving problems
structural optimization problems, such as truss structures [35],
faster, solving larger problems, and enhancing algorithm robustness
pin-jointed plane frames [36], and welded beams [37].
Nevertheless, the KH algorithm can cause a risk of stagnancy after
1 the initial stage when applied to multi-extreme discrete optimization
Corresponding author.
Manuscript received July 29, 2023; final manuscript received December 6, 2023; problems. Therefore, several approaches have been explored to
published online April 2, 2024. Assoc. Editor: Chul-Woo Kim. increase the diversity of solutions, resulting in a few KH variations

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-1
Part B: Mechanical Engineering Copyright V
C 2024 by ASME
with improved performance. A chaotic KH (CKH) algorithm was S ¼ ½S1 , S2 , …, Sv ,
presented to improve global optimization [38], wherein the three Find
Si 2 Di , Di ¼ ½di,1 , di,2 , …, di,rðiÞ 
primary movements of a krill swarm were adjusted during the
Xnm
optimization process using various chaotic maps. An opposition KH To minimize WðSÞ ¼ ci  Si  Li (1)
(OKH) algorithm was proposed to increase the diversity of the i¼1
population [39]. Herein, three operators, namely, opposition-based rmin  ri  rmax i ¼ 1, 2, …, n
learning, position clamping, and Cauchy mutation were added to the Subject to
dmin  dj  dmax j ¼ 1, 2, …, m
normal KH algorithm to improve the global convergence. A
multistage KH (MSKH) algorithm was introduced in Ref. [40] by
Here, S represents the set of design variables, Di denotes an
adding a local mutation and crossover operator and an elite scheme
allowable set of discrete values for design variable Si, v indicates the
in the exploitation stage. This resulted in the complete utilization of
number of design variables or member groups, r(i) represents the
the global and local search capabilities of the krill swarm to solve the
number of available discrete values for the i-th design variable, W(S)
global numerical optimization problem. Guo et al. [41] proposed an
denotes the weight of the structure, n indicates the number of
improved KH (IKH) algorithm, wherein information could be
component members in the structure, m represents the number of
exchanged between the top krill during movement. The IKH
nodes, ci denotes the material density of member I, Li indicates the
algorithm employed a novel Levy flight distribution to extend the
length of member i, dj represents the nodal displacement/deflection
search range and added an elite scheme to update the krill-motion
at node j, ri denotes the stress developed in the i-th element, and dmin
calculation, generate better candidate solutions, and accelerate
and dmax represent the lower and upper bounds, respectively.
global convergence. Laith et al. [42] proposed a modified KH
The optimum design of truss structures must satisfy the
algorithm (MKHA) to improve global exploration by modifying the
optimization constraints stated in Eq. (1). This procedure comprises
genetic operator of the basic KH. After analyzing the influence of a
the following three rules:
step-size scaling factor ct on KH, Li et al. [43] advanced KH with a
linear decreasing step (KHLD), wherein ct was linearly decreased Rule 1: Any feasible solution is better than any infeasible solution.
over time to balance exploration and exploitation. Another study Rule 2: Between two feasible solutions, a solution with a better
[44] used stud KH (SKH) algorithm for 22 benchmark functions. objective function value is preferable.
Furthermore, a levy-flight KH (LKH) algorithm [45] was proposed Rule 3: Between two infeasible solutions, the solution with the
to improve the optimization performance of the KH algorithm; its smallest constraint violation is preferred.
effectiveness was verified using several benchmark functions.
The first and third rules direct the search toward feasible regions,
Although several variants of KH algorithm enhance its optimi-
whereas the second rule directs the search to a feasible region with
zation performance, their accuracy when handling discrete truss
suitable solutions [47].
optimization problems remains unsatisfactory. The aforementioned
algorithms ignore the effect of the “suboptimal krill” and “center of
food” on the aggregation of KH, resulting in an inadequate balance 3 Krill Herd Algorithm
between global exploration and local exploitation. Typically, the This section briefly introduces the principles of KH [34]
discrete truss optimization easily falls into a local minimum. To algorithm. KH algorithm uses an optimization process to attain a
prevent this, a novel target-oriented KH (TOKH) algorithm is global solution defined using an objective function similar to the
proposed in this study, which considers both the “suboptimal krill” process by which a krill swarm obtains food and gathers
and “center of food”. Initially, the “suboptimal krill” and “best krill” continuously. Over time, the location of an individual krill is
were crossed to generate a novel “cross krill” for better global determined by three primary movements, namely,
exploration. Subsequently, an improved local mutation and cross-
over (ILMC) operator was applied to fine-tune the “center of food” (i) Motion induced by other individuals,
and population to improve local exploitation for effectively solving (ii) Foraging action, and
truss optimization. Four discrete truss design problems are applied (iii) Physical diffusion.
to verify the robustness of the developed TOKH algorithm. The These motion types can be expressed using a Lagrangian model in
optimization efficiency of TOKH algorithm increased by 20.90, a n-dimensional decision space as follows [34]
17.37, 53.53, and 88.01% for the four problems when compared to
those of KH algorithm. The results verify that the proposed method dXi
is highly competitive with other optimization approaches reported in ¼ Ni þ Fi þ Di (2)
dt
the literature and avoids falling into a local optimum.
The remainder of this paper is organized as follows: Section 2 where dXdt denotes the speed of each krill, and Ni , Fi , and Di indicate
i

presents the formulation of the discrete sizing optimization problem. the motions induced by other krill individuals, foraging action, and
A brief overview of the basic KH algorithm is provided in Sec. 3. physical diffusion of the i-th krill individuals, which can be obtained
Section 4 describes the proposed TOKH algorithm, and the using Eqs. (3), (5), and (7), respectively.
superiority of TOKH algorithm is verified using 15 benchmark The approximate value of the direction of induced motion (ai ) can
functions in Sec. 5. Section 6 explains the robustness of TOKH be calculated using a target swarm density (target effect), local
algorithm validated using four discrete truss optimization problems. swarm density (local effect), and repulsive swarm density (repulsive
Finally, the conclusions of the study and directions for future effect). For krill individual i, this movement can be formulated
research are summarized in Sec. 7. as [38]

Ninew ¼ N max ai þ xn Niold (3)

2 Formulation of Discrete Truss Optimization
Problems where
Discrete sizing optimization of the truss attempts to identify the
optimal cross section of the system elements to minimize the ai ¼ alocal
i þ atarget
i (4)
structural weight. Moreover, the minimum design must satisfy
the inequality constraints that limit the design variable sizes and Here, N max denotes the maximum induced speed, xn indicates the
structural responses [46]. inertia weight of the motion induced in the range of [0,1], Niold
The discrete structural optimization problem for a truss can be represents the last motion induced, alocal
i and atarget
i denote the local
formulated as [28] and target effects, respectively.

021108-2 / Vol. 10, JUNE 2024 Transactions of the ASME

 The foraging action involves two primary components, namely, be improved to enhance the global optimization of KH algorithm. In
the current “center of food” location and previous experience with the exploration phase, a crossover operator was established between
respect to the “center of food” position. The foraging action for the i- the “suboptimal krill” and “best krill” to generate a “cross krill”. The
th krill can be expressed as follows [38]: crossover Lagrangian model can be expressed as

Fi ¼ Vf bi þ xf Fold (5) Xc ¼ Xi þ x  ðXbest  Xi Þ  k  ðXi  Xsub Þ (9)

i
where Xc, Xi, Xbest, and Xsub denote the “cross krill”, “current krill”,
where “best krill”, and “suboptimal krill” respectively; and x and k
indicate different random numbers in the range of [0,1]. If the fitness
bi ¼ bfood
i þ bbest
i (6) value of Xc was better than that of Xbest, Xc was replaced by Xbest.
Algorithm 1 presents the pseudo-code used to achieve the
Here, Vf denotes the foraging speed, xf indicates the inertia weight aforementioned crossover operator.
within the range [0,1], bfoodi represents the “center of food”
attraction, bbest denotes the effect of the best fitness, and Fold Algorithm 1 Pseudo-code of the crossover operator
i i
indicates the last foraging action.
The third part of physical diffusion is primarily affected by the Begin
maximum diffusion rate and random vector. The physical diffusion Xc ¼ Xi þ x* (Xbest -Xi)- k*(Xi- Xsub)
can be expressed as [38] if Xc is better than Xbest then
Di ¼ Dmax d (7) Xbest ¼ Xc
end if
where Dmax denotes the diffusion speed, and d indicates a random End
vector within the range [1,1].
Foraging actions and movements induced by other krill After global exploration, the LMC operator was improved [40] to
individuals involve two global and two local optimization strategies. enhance the local exploitation. The LMC operator was inspired by
The simultaneous operation of these optimization strategies renders the reproduction mechanism of GA; it prevented the premature fall
the KH algorithm a powerful optimization method. During a specific into local optima by increasing the diversity of the population.
period, different effective movement parameters of the movement However, the LMC operator only considered the “best krill” as the
can be used. The changes in the location of a krill individual from t to target for random crossover and mutation, which was not conducive
t þ Dt can be expressed as follows [34]: to global optimization in the exploitation stage. The improved local
mutation and crossover (ILMC) operator was developed to improve
dXi the local exploitation of TOKH. In ILMC operator, the optimum
Xi ðt þ DtÞ ¼ Xi ðtÞ þ Dt (8) between the “center of food” and “best krill” served as the candidate
dt
solution, and the population learned from the candidate solution to
where Dt is an important parameter that completely depends on the accelerate aggregation to the optimum region in the exploitation
search space. Here, Dt can be increased appropriately when the stage. Algorithm 2 presents the pseudo-code of ILMC operator.
search space is wide, and when the search space is small, it can be Algorithm 2 Pseudo-code of the ILMC operator
appropriately reduced. Additionally, the crossover and mutation
mechanisms of GA [5] can be incorporated to improve the
Begin
performance of KH algorithm.
if Xbest is better than Xfood
Xcross ¼ Xbest
4 Target-Oriented Krill Herd Algorithm
else
Local exploitation and global exploration are the two critical
Xcross ¼ Xfood
components of modern metaheuristic algorithms. Exploitation
reinforces the local search for minimum or near-optimal solutions, end if
whereas exploration involves a global search to ensure the efficient for j ¼ 1 to d do
and effective exploration of the search space [4]. Excessive if rand  0.5 then
diversification causes solutions to jump around from one potentially Wi(j) ¼ Xcross(j)
optimal solution to another, increasing the convergence time
else Wi(j) ¼ Xcross(u)
required to reach optimum. However, excessive reinforcement
may trap the algorithm in local optimum, as only a portion of the end if
local space may be visited. Therefore, an effective algorithm end for j
requires an appropriate balance between these two components to Obtain the individual fitness value Wi
ensure efficient convergence, avoid falling into a local optimum, and if Wi is better than Xi then
guarantee the solution accuracy.
Xi ¼ Wi
Krill herd algorithm has demonstrated its ability to identify near-
global regions in continuous-truss optimization problems [35]. end
However, the insufficient balance between global exploration and End
local exploitation causes the algorithm to easily fall into a local
optimum when solving discrete truss optimization problems. The In the Algorithm 2, Xbest denotes the “best krill”, Xfood indicates the
proposed TOKH algorithm intends to balance the associated “center of food”. d represents the decision variables. Xi(j) denotes the
exploitation and exploration components to solve the discrete truss j-th variable of the solution Xi. Wi indicates the offspring. u represents a
optimization problem more efficiently. random integer number between 1 and d obtained from a uniform
The subsequent subsections present the basic principle and time distribution, and rand denotes a random real number in the interval (0,
complexity of TOKH algorithm. 1) obtained from a uniform distribution.
Based on this analysis, the steps of the TOKH algorithm can be
summarized as follows.
4.1 Basic Principle. This subsection describes the basic
principle of TOKH algorithm. The learning efficiency of the Step 1: Initialization: The generation counter was set to t¼1, and
population oriented to the “suboptimal krill” and “best krill” should the population P of NP krill individuals was randomly initialized.

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-3
Part B: Mechanical Engineering
 The foraging speed Vf, maximum diffusion speed Dmax, and (1) All addition and subtraction constants during runtime are
maximum induced speed Nmax were also set. replaced by constant 1,
Step 2: Krill population fitness evaluation: Each krill individual (2) Only the highest order items are retained, and
was evaluated according to its position. (3) The highest term constant is removed.
Step 3: Motion calculation: The motion induced by the presence
Firstly, TOKH algorithm is started with parameters initialization
of other individuals was obtained using Eq. (3), the foraging
which requires constant time complexity O(1). Secondly, in steps
motion was calculated using Eq. (5), and the physical diffusion
2–4, three different movements of krill are implemented based on
was determined based on Eq. (7).
the rules of KH. The time complexity grows linearly according to the
Step 4: The genetic operators [34] were implemented.
number of krill n, consequently, it requires O(3n). Thirdly, referring
Step 5: Crossover operator: The “best krill” and “suboptimal krill”
to crossover operator and ILMC operator, the time complexity in
were crossed according to Algorithm 1.
step 5 and step 6 are equal to O(1) and O(n), respectively. Moreover,
Step 6: Population fine-tuning: The krill population was fine-
to implement the population update, the time complexity in step 7 is
tuned using ILMC operator in Algorithm 2. Each krill was
equal to O(n). Finally, in step 8, the population evolves according to
evaluated considering its new position.
a specific number of iterations tmax. Therefore, the total time
Step 7: Update the population position: The positions of krill
complexity of TOKH is expressed by O(1þ(3n þ 1þn þ n)tmax) 2
individuals were updated in the search space.
O(n). In summary, it is obvious that most of the time complexity of
Step 8: Repeat: steps 2–7 were repeated until a stop criterion was
TOKH comes from the basic KH, and the two operators introduced
satisfied or a predefined number of iterations were completed.
do not change the time complexity of TOKH.
Figure 1 illustrates the flowchart of the TOKH algorithm. The Rastrigin benchmark function [45] was selected for the
verification. The functional dimensions were 30 and 100, and the
4.2 Time Complexity Analysis of Target-Oriented Krill population size was 50. All experiments were implemented and
Herd Algorithm. The time complexity of the algorithm was executed using MATLAB R2018a running on a personal computer with
investigated to analyze the real-time efficiency of the TOKH a 7-th generation core i7 CPU and Windows 10 OS. Figure 2 depicts
algorithm in a better manner. The worst-case running time was the experimental results for the time complexity.
expressed as a function of its input using a big Omicron (big-O) As indicated in Fig. 2, the ordinate denotes the total time required
notation [48]. Typically, the big-O notation is used to denote an in seconds for the algorithm to run 20 times on the Rastrigin
upper bound on the growth rate of a function and can be primarily benchmark function [45]. The abscissa represents the number of
applied to describe asymptotic behavior [48]. Logically, the big-O iterations, ranging from 0 to 1000; the running time was recorded
order is derived according to three rules: once every 50 generations. The red and blue lines represent the
fitting results of TOKH and KH algorithms, respectively. The
running time and iteration of TOKH algorithm satisfied the linear
relationship, and the time complexity of TOKH algorithm was O(n).
The experimental results validated that the two optimization
operators introduced do not change the time complexity.

5 Benchmark Mathematical Functions

Fifteen benchmark functions were used to compare the optimized
performance of TOKH algorithm with nine other versions of KH
algorithm, including KH [34], CKH [38], OKH [39], MSKH [40],
IKH [41], MKH [42], KHLD [43], SKH [44], and LKH [45]. Table 1
summarizes the specific parameter settings of each algorithm.
Benchmark functions are widely used for evaluating algorithm
performance owing to their ease of implementation and high
reliability. In this study, 15 different dimensional benchmark
functions were selected to compare the performances of the
algorithms. The dimensions (n) of the benchmark functions were
set to 10, 30, and 50, respectively; the optimization performance of
the 30-dimension was particularly analyzed, which is a pretty
representation of the average performance of the algorithm. The
expressions and properties of these benchmark functions have been
reported in a previous study [38].
The population size NP of all algorithms was 50. The maximum
number of iterations was 50, and 30 simulations were independently
performed by each benchmark function. Tables 2–5 summarize the
simulation results of the algorithm considering 30-dimension.
Table 2 lists the best-optimized performances of these algorithms,
whereas Table 3 presents the general optimization performance,
Table 4 indicates the worst optimization performance, and Table 5
describes the stability. Tables 6 and 7 summarize the general
optimization performance when the dimensions of the benchmark
functions were 10 and 50, respectively.
When the dimension of the benchmark function was 30, TOKH
and MSKH exhibited the best optimization performance for nine and
seven benchmark functions, respectively (Table 2). TOKH
algorithm performed better than the other KH algorithms on
F1–F2, F6–F7, and F10–F14; MSKH algorithm performed best on
F2–F5, F8–F9, and F15. About the general optimization perform-
ance (Table 3), TOKH algorithm performed best on ten benchmark
Fig. 1 TOKH algorithm flowchart functions (F1–F2, F4, F6–F7, and F10–F14), and MSKH algorithm

021108-4 / Vol. 10, JUNE 2024 Transactions of the ASME

 Fig. 2 Relationship between the running time and number of iterations

Table 1 Specific parameter settings of each algorithm

Algorithm name Basic parameters

KH The maximum induced speed N ¼ 0.01, the foraging speed Vf ¼ 0.02, the maximum diffusion speed Dmax ¼ 0.005
max

KHLD Ctmax ¼ 2, Ctmin ¼ 0, Other parameters are the same as KH.

TOKH All parameters are the same as KH.
OKH Reference [39], OKH7 is selected. d1¼0.4, d2 ¼0.6. Other parameters are the same as KH.
CKH Reference [38], M10 chaotic map is selected. Other parameters are the same as KH.
IKH Reference [41], IKH2 is selected. Other parameters are the same as KH.
MSKH All parameters are the same as KH.
SKH Reference [44], SKH1 is selected. Crossover probability of single point crossover Pc ¼ 1. Other parameters are the same as KH.
LKH Max Levy-flight step size A ¼ 1.0. Other parameters are the same as KH.
MKH Elitism parameter Keep ¼ 2, Other parameters are the same as KH.

performed best on five benchmark functions (F3, F5, F8–F9, and benchmark functions (Table 7). Figure 3 illustrates the convergence
F15). Furthermore, TOKH algorithm performed best on twelve process of several benchmark functions for 30-dimension. As
benchmark functions (F1–F2, F4, F6–F8, and F10–F15) with respect indicated in the figure, TOKH algorithm was significantly superior
to the worst optimization performance (Table 4). In terms of stability to all other algorithms with respect to the optimization process.
(Table 5), TOKH algorithm exhibited the best stability for twelve Based on the data presented in Tables 2–7 and Fig. 3, we
benchmark functions (F1–F2, F4, F6–F8, and F10–F15). As concluded that the developed metaheuristic TOKH algorithm was
indicated in Table 6, when the dimension of the benchmark function more robust and stable than other metaheuristic search methods.
was 10, TOKH algorithm performed best on ten benchmark
functions (F1–F4, F6–F7, F10–F12, and F14), MSKH algorithm
performed best on F5 and F8–F9, and SKH exhibited the best 6 Truss Optimization Problems
performance on F15. When the dimension of the benchmark To further investigate the robustness of TOKH algorithm for truss
function was 50, TOKH algorithm performed best for all 15 optimization, we solved the weight minimization problems of four

Table 2 Best values of the benchmark functions (the dimension of function is 30)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

F1 1.95 0.29 0.00 0.040 1.60 1.57 0.012 1.36 1.84 0.051
F2 2.67 0.99 0.00 0.61 1.23 1.62 0.00 1.05 2.20 0.89
F3 6.71  101 1.01  101 4.58  103 8.38  101 2.15  101 1.33  101 7.59  1024 1.78  101 2.37  101 3.09
F4 1.47  103 3.67  101 4.26  105 9.34  101 1.33  102 1.34  102 1.10  1025 5.06  101 2.17  102 3.33  101
F5 4.52  103 7.82 2.32  105 1.67 1.89  101 2.82  101 8.57  1026 1.64  101 2.21  101 3.11
F6 8.04  102 5.82  102 1.80  1024 1.06  102 3.38  102 7.05  102 6.84  104 5.04  102 3.74  102 1.22  102
F7 1.60  102 6.85 8.63  1027 8.88  101 9.24  101 4.64  101 5.37  103 6.24  101 8.88  101 5.05  101
F8 5.25  101 2.79  101 2.71  101 1.13  101 3.33  101 3.61  101 2.48  1023 2.86  101 3.82  101 2.95  101
F9 8.66  103 8.23  103 1.21  101 2.20  103 6.04  103 2.20  103 2.21  1022 2.84  103 5.20  103 1.25  103
F10 2.35  104 4.30 1.86  1025 7.46  103 1.44  104 2.89  103 1.26 6.19  102 1.80  104 8.67  103
F11 1.26  104 9.45  103 1.80  1023 1.88  101 3.62  101 8.65 3.14  102 3.57  101 3.39  101 7.92
F12 6.94 2.26 8.94  1024 2.78  101 6.00 6.53 3.57  102 7.36 5.28 3.81
F13 3.26  102 2.24 2.03  1025 5.03  101 5.04  101 8.42  101 1.49  102 3.46 6.73  101 1.31
F14 4.41  102 4.84  101 4.70  1028 1.48  101 3.51  101 1.43  101 1.50  104 1.06 1.04  102 8.65
F15 4.72  102 3.98 6.67  101 1.35 3.99  101 6.43  101 9.73  1022 4.34 1.09  102 3.37
Total 0 0 9 0 0 0 7 0 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-5
Part B: Mechanical Engineering
 Table 3 Mean values of the benchmark functions (the dimension of function is 30)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

F1 2.74 1.79 0.00 0.34 2.30 2.13 0.022 2.65 2.67 0.72
F2 6.25 1.11 0.00 0.88 2.65 3.51 0.020 1.66 3.79 1.08
F3 1.50  102 2.64  101 9.88 103 1.37  101 5.28  101 3.29  101 3.11  1023 6.71  101 7.99  101 7.34
F4 3.02  105 1.30  102 2.30  104 3.84  102 3.36  102 1.47  103 1.56  101 1.49  102 5.70  102 1.09  102
F5 1.50  105 2.47  101 1.58  104 4.01  102 8.76  101 1.73  103 3.93  1025 3.63  101 1.50  103 9.94
F6 3.59  101 9.25  101 8.26  1024 6.78  102 1.11  101 2.09  101 4.04  103 1.84  101 9.38  102 4.08  102
F7 2.03  102 5.39  101 3.99  1024 1.58  102 1.49  102 8.04  101 3.00 1.24  102 1.30  102 7.12  101
F8 8.82  101 2.92  101 2.74  101 3.07  101 4.65  101 6.04  101 2.69  101 3.34  101 6.14  101 3.76  101
F9 9.35  103 9.48  103 2.80  101 3.51  103 6.82  103 3.86  103 1.18  1021 4.37  103 6.41  103 2.24  103
F10 3.86  104 7.85  103 8.53  1024 1.71  104 2.41  104 7.43  103 2.10  103 3.54  103 2.83  104 1.32  104
F11 9.55  109 3.62  109 1.12  1022 3.96  101 4.91  101 1.76  101 7.09  102 6.98  101 4.82  101 2.42  101
F12 1.01  101 5.80 1.38  1022 3.98  101 8.82 1.01  101 2.83  101 1.24  101 1.01  101 6.66
F13 6.52  102 9.66 8.86  1024 1.74 1.62  102 3.43  102 3.56  102 5.71  101 3.17  102 9.67
F14 8.70  102 3.25 1.16  1025 1.88  101 1.10  102 7.11  101 1.50  101 1.07  101 3.05  102 5.62  101
F15 2.27  103 8.99 6.67  101 1.63  102 1.77  102 3.73  102 4.63  1021 1.70  101 4.56  102 1.83  101
Total 0 0 10 0 0 0 5 0 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

Table 4 Worst values of the benchmark functions (the dimension of function is 30)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

F1 3.39 2.76 0.010 1.19 3.45 2.86 0.031 3.36 3.34 1.65
F2 16.09 1.68 0.010 1.23 4.65 6.00 0.063 2.94 6.55 1.24
F3 2.40  102 5.93  101 1.67  102 8.43  101 1.49  102 7.51  101 6.55 3 1023 2.59  102 1.74  102 1.89  101
F4 1.91  106 3.31  102 6.85  1024 9.33  102 5.06  102 2.54  104 9.35  101 5.63  102 1.07  103 2.95  102
F5 7.95  105 4.91  101 5.12  104 7.30  103 9.46  102 1.31  104 9.79 3 1025 5.19  101 7.13  103 2.33  101
F6 7.97  101 1.50  102 2.77  1023 1.74  101 1.80  101 6.17  101 1.79  102 4.28  101 2.77  101 7.70  102
F7 2.32  102 1.86  102 1.58  1023 2.04  102 1.84  102 1.25  102 2.99  101 2.58  102 1.61  102 1.07  102
F8 1.40  102 3.30  101 2.77  101 5.23  101 6.40  101 8.74  101 2.81  101 4.78  101 1.11  102 5.06  101
F9 1.01  104 1.01  104 3.82  101 5.61  103 8.58  103 4.91  103 2.67 3 1021 5.46  103 7.19  103 3.30  103
F10 6.09  104 1.43  104 5.16  1023 2.73  104 3.58  104 1.28  104 1.50  104 1.23  104 3.59  104 1.87  104
F11 1.28  1011 8.28  1010 2.53  1022 6.32  101 7.61  101 3.32  101 1.04  101 1.02  102 6.62  101 5.85  101
F12 1.39  101 9.94 4.85  1022 7.74  101 1.34  101 1.71  101 3.48 1.99  101 1.36  101 1.11  101
F13 1.44  103 3.33  101 3.56  1023 8.94 4.38  102 7.61  102 9.15  102 2.97  102 5.89  102 3.67  101
F14 1.42  103 1.05  101 8.07  1025 7.34  101 2.48  102 1.66  102 2.98 7.77  101 5.74  102 1.52  102
F15 7.37  103 1.74  101 6.67 3 1021 2.74  103 6.38  102 1.19  103 7.51  101 4.85  101 1.07  103 4.35  101
Total 0 0 12 0 0 0 3 0 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

truss structures under multiple loading conditions using discrete LKH [45].) in the literature to evaluate the robustness of TOKH
variables. The algorithms were coded in MATLAB and the structures algorithm. Twenty independent runs were performed for each
were analyzed using the direct stiffness method. The optimization design problem with the population size of each algorithm set to 30.
results were compared to the results obtained from other
optimization methods (including TLBO [23], PO [28], MRFO 6.1 Planar 10-Bar Truss. The planar 10-bar truss structure is
[33], KH [34], CKH [38], OKH [39], IKH [41], KHLD [43], and one of the most popular test problems in structural optimization,

Table 5 Standard deviations of the benchmark functions (the dimension of function is 30)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

F1 0.34 0.49 0.00 0.37 0.46 0.34 0.00 0.48 0.35 0.56
F2 2.72 0.14 0.00 0.14 0.96 1.20 0.010 0.54 1.26 0.080
F3 5.00  101 1.20  101 3.27  103 2.02  101 2.87  101 1.36  101 1.67  1023 5.90  101 2.86  101 3.75
F4 4.58  105 7.71  101 1.31  1024 1.88  102 9.86  101 4.78  103 3.54  101 1.02  102 2.15  102 6.24  101
F5 1.85  105 7.91 9.96  105 1.41  103 1.73  102 2.99  103 2.35  1025 8.19 2.22  103 5.55
F6 2.07  101 3.05  101 5.88  1024 4.06  102 4.66  102 1.30  101 4.07  103 9.48  102 4.72  102 2.04  102
F7 1.85  101 5.57  101 4.17  1024 2.50  101 2.12  101 1.90  101 9.11 4.77  101 2.00  101 1.34  101
F8 2.27  101 1.01 1.39  1021 5.71 8.83 1.48  101 5.08 4.51 1.55  101 5.61
F9 3.67  102 4.13  102 7.16  102 5.18  102 5.58  102 7.37  102 5.56  1022 6.88  102 5.09  102 4.59  102
F10 9.81  103 2.67  103 1.30  1023 4.66  103 5.62  103 2.21  103 3.43  103 2.67  103 4.69  103 3.08  103
F11 2.58  1010 1.52  1010 6.17  1023 1.12  101 7.96 5.88 1.55  102 1.80  101 9.94 1.10  101
F12 1.93 1.84 1.00  1022 1.13  101 1.57 2.39 6.51  101 3.57 2.13 1.57
F13 2.69  102 7.66 9.11  1024 1.57 9.20  101 1.86  102 1.81  102 6.51  101 1.52  102 7.71
F14 2.71  102 2.56 1.65  1025 1.82  101 4.93  101 4.20  101 5.52  101 1.58  101 1.13  102 3.81  101
F15 1.79  103 3.61 5.61  1025 5.05  102 1.34  102 2.77  102 2.32  101 1.00  101 2.73  102 9.04
Total 0 0 12 0 0 0 3 0 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

021108-6 / Vol. 10, JUNE 2024 Transactions of the ASME

 Table 6 Mean values of the benchmark functions (the dimension of function is 10)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

2 1 1 1 2
F1 2.11 2.00  10 0.00 1.20  10 5.40  10 4.10  10 4.50  10 1.37 1.42 6.10  102
F2 6.20  101 1.40  101 0.00 5.30  101 5.00  101 2.20  101 3.40  102 2.70  101 4.10  101 2.60  101
F3 9.68 9.00  102 0.00 3.50  101 3.42 2.06 9.40  101 2.15  101 3.15 9.10  102
F4 2.56  101 1.56 1.63  1024 4.30  101 4.48 2.09 3.46 1.19  101 5.39 3.10  101
F5 7.06 9.00  102 1.44  104 2.00  101 5.00  101 5.90  101 5.43  1026 1.24 2.01 1.33  104
F6 7.00  103 1.70  103 1.00  1024 3.79  104 8.41  104 5.18  104 2.00  102 2.40  103 1.20  103 3.14  104
F7 5.30  102 9.90  101 1.03  1025 9.80  102 1.18 2.98 1.00  103 4.97 2.05 1.20  101
F8 8.12 5.77 6.56 1.42 5.66 1.30  101 9.08  1025 4.40 5.46 2.01
F9 1.99  103 1.65  103 1.50  102 8.55  102 6.74  102 4.60  102 1.60  1023 3.36  102 7.38  102 1.50  102
F10 4.47  102 1.72  101 8.77  1026 2.35  101 1.67  102 6.11 6.99  105 3.80  101 1.27  102 6.10  101
F11 1.81  101 6.27 5.18  1024 1.94 5.00  101 2.90  103 1.82  103 1.85  103 4.96 4.56  103
F12 4.40  102 1.70  102 1.70  1023 5.90  102 5.30  102 7.00  103 1.10  102 5.00  103 1.80  102 6.90  103
F13 1.10  102 2.50  103 3.14  105 8.20  103 7.90  103 5.33  1026 1.50  103 3.05  106 3.85  104 2.77  105
F14 6.45 2.20  103 1.37  1029 8.08  104 2.00  102 2.29  104 4.04  107 1.87  104 5.60  101 2.55  105
F15 7.10  101 6.60  101 4.08  104 6.70  101 1.00  101 1.00  101 1.01  104 7.41  1025 1.30  101 2.20  103
Total 0 0 10 0 0 1 3 1 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

Table 7 Mean values of the benchmark functions (the dimension of function is 50)

KH KHLD TOKH OKH CKH IKH MSKH SKH LKH MKH

23 2
F1 1.11  10 1
7.45 8.60  10 3.83 1.05  10 1
1.02  10 1
8.03  10 1.07  101
1.06  10 1
8.48
F2 4.41  101 2.05  101 1.90  1023 1.03 3.43  101 4.52  101 9.29  102 2.99  101 3.50  101 1.87  101
F3 4.91  101 8.99 1.30  1023 3.33 2.45  101 2.86  101 7.11  102 2.16  101 2.52  101 9.93
F4 8.02  106 2.86  103 4.60  1023 3.28  102 1.98  106 7.56  106 6.78  101 6.37  103 2.88  106 2.51  105
F5 2.07  106 1.70  104 2.20  1023 5.14 8.07  105 1.54  106 1.57 1.13  105 6.99  105 2.05  105
F6 2.51 4.09  102 3.00  1023 5.80  102 1.52 2.01 2.84  102 1.78 1.59 8.52  101
F7 4.48  102 2.17  102 2.00  1023 3.03  102 3.37  102 2.61  102 1.40  101 3.22  102 2.75  102 2.04  102
F8 1.95  106 2.45  105 4.88  101 7.64  102 1.21  106 1.66  106 6.79  101 5.10  105 1.11  106 4.13  105
F9 1.75  104 1.73  104 9.00  1021 7.40  103 1.39  104 1.05  104 1.19  102 1.07  104 1.36  104 8.23  103
F10 1.56  105 3.53  104 2.31 4.12  104 7.81  104 4.98  104 5.17  104 3.55  104 2.17  104 5.88  104
F11 4.20  1022 2.76  1020 2.46  1022 1.50  109 1.07  1010 8.99  101 1.71  101 1.71  102 9.83  107 1.26  102
F12 2.48  101 1.42  101 2.10  1022 3.32  101 2.29  101 2.44  101 1.16 2.54  101 2.02  101 2.11  101
F13 4.78  103 2.03  103 1.70  1023 1.45  101 4.19  103 4.72  103 1.13  101 3.14  103 4.44  103 1.87  103
F14 3.27  103 1.14  102 6.29  1025 4.42  101 6.15  102 7.68  102 2.45  101 2.14  102 5.27  102 3.35  102
F15 3.02  104 3.02  103 8.42  1021 6.05 1.57  104 2.08  104 2.54 7.15  103 1.60  104 6.28  103
Total 0 0 15 0 0 0 0 0 0 0

Bold font in the table is used to highlight the “optimal values” among all methods.

previously solved in Ref. [23]. Figure 4 depicts the geometry and and displacement constraints, the minimum weight of the 10-bar
support conditions followed for this two-dimensional, cantilevered truss was obtained by adjusting the cross-sectional area of each
truss under loading conditions. As indicated in the figure a static load member. A set of 41 discrete values were used for the possible cross-
of 100 kips was applied downward to two nodes. To satisfy the stress sectional areas for each member, as follows: S ¼f1.62, 1.80, 1.99,

Table 8 Optimized designs for the 10-bar truss

KH KHLD CKH LKH IKH TOKH OKH TLBO PO MRFO

S1 30.00 33.50 33.50 30.00 30.00 33.50 28.50 33.50 33.50 33.50
S2 1.62 3.55 5.94 2.13 1.80 1.62 13.50 1.62 1.62 1.62
S3 28.50 28.50 22.90 28.50 22.90 22.90 30.00 22.90 22.90 22.90
S4 15.50 11.50 15.50 11.50 18.80 14.20 18.80 14.20 15.50 14.20
S5 1.62 1.62 1.62 1.62 1.62 1.62 5.94 1.62 1.62 1.62
S6 1.62 5.94 1.62 1.99 1.62 1.62 5.94 1.62 1.62 1.62
S7 19.90 18.80 22.00 16.90 22.90 22.90 22.90 22.90 18.80 22.90
S8 7.97 11.50 7.97 16.00 11.50 7.97 7.22 7.97 7.22 7.97
S9 2.13 3.55 1.62 4.18 1.62 1.62 5.94 1.62 1.62 1.62
S10 22.00 18.80 22.00 22.90 18.80 22.90 18.80 22.90 28.50 22.90
NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00
Computational time (s) 1.33 1.42 1.51 1.81 1.64 1.80 3.01 1.54 0.50 0.58
Optimal value (lb) 5.63  103 5.83  103 5.65  103 5.78  103 5.59  103 5.49  103 6.54  103 5.49  103 5.62  103 5.49  103
Mean (lb) 7.13  103 7.19  103 6.13  103 7.15  103 6.85  103 5.64  103 6.83  103 5.91  103 6.56  103 5.66  103
Std (lb) 1.51  103 9.66  102 2.53  102 1.31  103 1.14  103 4.84  101 2.86  102 2.50  102 8.01  102 1.09  102
Increased efficiency  0.84% 14.03% 0.28% 3.93% 20.90% 4.21% 17.11% 7.99% 20.62%

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-7
Part B: Mechanical Engineering
2.13, 2.38, 2.62, 2.88, 2.93, 3.09, 3.13, 3.38, 3.47, 3.55, 3.63, 3.84, Table 9 Element information for the 25-bar truss
3.87, 3.88, 4.18, 4.22, 4.49, 4.59,4.80, 4.97, 5.12, 5.74, 7.22, 7.97,
11.5, 13.5, 13.9, 14.2, 15.5, 16.0, 16.9, 18.8, 19.9, 22.0, 22.9, 26.5, Design variables Members
30.0, 33.5g (in2). The combined size of all feasible solutions was
(41)10. The Young’s modulus of the material was 107 psi, and the S1 (1,2)
S2 (1,4), (2,3), (1,5), (2,6)
weight density was 0.1 lb/in3. The displacement constraints in the X S3 (2,5), (2,4), (1,3), (1,6)
and Y directions at each node were limited to 2 in. The restraining S4 (3,6), (4,5)
stress of each member was less than 25 ksi. S5 (3,4), (5,6)
Table 8 summarizes the optimal designs and their corresponding S6 (3,10), (6,7), (4,9), (5,8)
number of structural analyses (NSA) using the TOKH with nine S7 (3,8), (4,7), (6,9), (5,10)
other methods. TOKH, TLBO, and MRFO achieved the lightest S8 (3,7), (4,8), (5,9), (6,10)
designs. However, statistical results after 20 runs also demonstrate
that TLBO and MRFO exhibited less stability compared to TOKH.
Based on the average of 20 independent runs, the NSA is 3100 for Table 10 Nodal coordinates for the 25-bar truss
TOKH and 1580 required for KH. Although TOKH has 1520 more
structural analyses than KH, the optimization efficiency of TOKH Node X (in.) Y (in.) Z (in.)
algorithm is 20.90% higher than that of KH algorithm. Figure 5
shows the average convergence curves obtained for TOKH and the 1 37.5 0.0 200.0
nine different methods when the planar 10-bar truss is applied. This 2 37.5 0.0 200.0
3 37.5 37.5 100.0
is a relatively simple structural sizing optimization problem, and its
4 37.5 37.5 100.0
optimal sizing is easy to find. It is obvious that through efficient 5 37.5 37.5 100.0
global exploration, TOKH quickly converges to the minimum with 6 37.5 37.5 100.0
fewer iterations. The TOKH algorithm has the highest global 7 100.0 100.0 0.0
convergence rate. For other algorithms, MRFO works very well, 8 100.0 100.0 0.0
because it ranks 2 among ten methods. 9 100.0 100.0 0.0
10 100.0 100.0 0.0

6.2 Spatial 25-Bar Truss. Figure 6 illustrates the 25-bar

transmission tower spatial truss, which has been analyzed by several

Fig. 3 Performance comparison considering the 30-dimension benchmark functions

021108-8 / Vol. 10, JUNE 2024 Transactions of the ASME

 Fig. 3 (Continued)

researchers [23]. All structural elements were organized into eight 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0, 2.1, 2.2, 2.3,
groups, where the members of each group shared the same material 2.4, 2.5, 2.6, 2.7, 2.8, 2.9, 3.0, 3.1, 3.2, 3.3, 3.4g (in2). The size of the
and cross-sectional properties. Table 9 presents each element group design search space was (34)8.
according to the member number; (each member is defined based on Table 12 compares the final optimum design and the correspond-
its start and end node numbers). Table 10 lists the coordinates of the ing results calculated by the ten methods. TOKH algorithm still has
25-bar truss nodes. The Young’s modulus of the material was the robustness compared with the other nine methods. The average
107 psi, and the weight density was 0.1 lb/in3. A single load case was weight gained by TOKH algorithm is 488.34 lb, 17.37% lighter than
applied to the structure in the design of the 25-bar truss (Table 11).
The allowable stresses for each member were 640 ksi and the
allowable displacements for each node in the X, Y and Z directions
were 60.35 in. Discrete values for each cross-sectional area were
obtained from the available set S ¼ f0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7,

Fig. 4 10-bar truss Fig. 5 Algorithm optimization process of the 10-bar truss

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-9
Part B: Mechanical Engineering
 Fig. 6 25-bar truss Fig. 9 Algorithm optimization process of the 72-bar truss

that of KH algorithm. The standard deviation with the TOKH

algorithm is 2.59, which is the lowest among all methods. Figure 7
depicts the average convergence curves for the 25-bar truss. TOKH
quickly converges to a better global region in the early iterations and
continues to search the minimum in about 10–20 iterations through
local exploitation. MRFO algorithm converges slowly in the early
stage and gradually overtakes TLBO and PO in the late iteration.

6.3 Spatial 72-Bar Truss. The optimization of the 72-bar,

four-level tower (Fig. 8) was reported in Ref. 23. The material
density and modulus of elasticity of this 72-bar truss were identical
to those of the 10- and 25-bar truss structures. The allowable stress of
each member was 625 ksi, and the permissible displacement of each
point on the top layer in all directions was 60.25 in. The 72-bar
space truss was divided into 16 groups. Table 13 lists the two
independent load cases applied to the spatial 72-bar space truss. The
range of the discrete cross-sectional areas was 0.1–3.0 in2 with an
increment of 0.1 in2 for each of the 16 element groups, resulting in 30
discrete cross section. The size of the resulting search space was
(30)16 designs.
Fig. 7 Algorithm optimization process of the 25-bar truss

Fig. 8 72-bar truss: (a) side view and (b) typical story

021108-10 / Vol. 10, JUNE 2024 Transactions of the ASME

 Table 11 Loading conditions for the 25-bar truss is 53.53% higher than KH. These results indicate that TOKH
algorithm has an apparent advantage in search ability compared with
Node Px (kips) Py (kips) Pz (kips) the other nine methods. Figure 9 shows the optimization results for
the spatial 72-bar truss, which is a complex structural sizing
1 1.0 10.0 10.0 optimization problem. From Fig. 9, different from the planar 10-bar
2 0.0 10.0 10.0
3 0.5 0.0 0.0
truss as shown in Fig. 5, the figure shows that TOKH quickly
6 0.6 0.0 0.0 converges to a better global region in the early iterations by global
exploration and continues to find the minimum in about 10–30
iterations through efficient local exploitation. MRFO algorithm still
Table 14 compares the final optimum design and the correspond- has the disadvantage of slow convergence in the early stage, and as
ing results identified by the ten methods. The lightest weight and the iterations increase, the convergence speed gradually surpasses
average weight achieved by TOKH algorithm, which are 387.94 lb that of TLBO and PO.
and 402.30 lb, respectively. Based on the average weight of 20
independent runs, although TOKH algorithm has 1520 more 6.4 A 942-Bar Truss Tower. The final example considered in
structural analyses than KH, the optimization efficiency of TOKH this study is that of a 26-story space truss tower consisting of 942

Table 12 Optimized designs for the 25-bar truss

KH KHLD CKH LKH IKH TOKH OKH TLBO PO MRFO

S1 1.60 1.40 0.50 1.80 0.20 0.10 1.20 0.90 0.10 0.20
S2 1.50 0.70 0.50 2.80 0.10 0.50 1.20 0.30 0.40 1.10
S3 3.00 3.40 3.40 2.00 3.40 3.40 3.00 3.20 3.40 3.20
S4 0.10 0.70 0.10 0.40 0.10 0.10 0.90 0.10 0.10 0.20
S5 0.70 1.60 2.80 0.40 0.60 1.90 1.00 2.50 2.40 1.30
S6 0.90 0.70 1.10 1.00 1.00 1.00 0.90 0.90 1.10 0.90
S7 0.60 1.00 0.20 0.40 1.20 0.40 0.90 0.80 0.30 0.40
S8 3.40 3.40 3.40 3.40 3.40 3.40 3.20 3.40 3.40 3.40
NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00
Computational time (s) 0.72 0.77 0.86 1.40 0.98 1.90 1.38 2.18 0.82 1.13
Optimal value (lb) 520.66 531.48 494.30 540.05 503.38 485.05 529.56 501.86 487.33 493.83
Mean (lb) 590.98 558.37 530.87 571.76 549.41 488.34 579.05 523.36 509.75 500.76
Std (lb) 39.89 19.72 18.54 16.74 36.36 2.59 37.39 14.32 25.59 5.94
Increased efficiency  5.52% 10.17% 3.25% 7.03% 17.37% 2.02% 11.44% 13.74% 15.27%

Bold font in the table is used to highlight the “optimal values” among all methods.

Table 13 Load cases for the 72-bar truss

Case Node Px (kips) Py (kips) Pz (kips)

1 17.0 0.0 0.0 5.0

18.0 0.0 0.0 5.0
19.0 0.0 0.0 5.0
20.0 0.0 0.0 5.0
2 17.0 5.0 5.0 5.0

Table 14 Optimized designs for the 72-bar truss

KH KHLD CKH LKH IKH TOKH OKH TLBO PO MRFO

S1 1.10 1.60 1.80 0.90 1.90 2.00 1.20 0.50 1.70 1.70
S2 0.40 0.70 0.60 0.80 0.40 0.50 1.00 0.60 0.70 0.60
S3 0.30 0.60 0.50 0.20 0.20 0.10 0.10 0.10 0.10 0.10
S4 0.70 1.30 1.00 0.20 0.60 0.10 0.10 1.80 0.10 0.30
S5 0.60 1.80 1.20 1.70 1.90 1.10 1.30 0.40 1.30 1.60
S6 0.80 0.80 0.50 0.40 0.80 0.50 0.80 0.70 0.60 0.40
S7 0.90 1.20 0.70 1.60 0.10 0.10 0.10 0.30 0.10 0.20
S8 0.90 1.60 0.20 1.50 0.20 0.10 0.40 0.60 0.10 0.20
S9 1.40 1.00 1.40 1.00 0.80 0.60 0.80 0.50 1.40 0.70
S10 1.60 0.60 0.50 0.50 0.40 0.60 1.10 0.10 0.40 0.50
S11 0.80 0.40 0.40 1.70 0.20 0.10 0.10 0.40 0.10 0.10
S12 1.70 0.40 0.50 1.60 0.60 0.10 0.70 1.60 0.10 0.30
S13 0.30 0.60 0.40 1.50 0.20 0.20 1.10 0.50 0.20 0.40
S14 0.60 0.30 0.60 1.00 0.60 0.50 1.10 0.70 0.50 0.90
S15 0.50 1.30 0.30 1.30 0.30 0.50 0.10 1.90 0.30 0.40
S16 1.80 0.70 0.60 0.90 0.70 0.60 0.50 0.50 0.80 0.40
NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00
Computational time (s) 1.81 1.90 1.82 3.33 2.27 4.81 7.12 5.58 1.96 3.09
Optimal value (lb) 739.63 681.36 520.59 785.15 461.01 387.94 644.58 449.94 412.66 419.00
Mean (lb) 865.71 782.99 584.17 821.81 502.50 402.30 705.74 523.17 463.53 442.38
Std (lb) 67.61 65.83 49.78 33.38 24.93 6.04 40.20 48.81 51.43 18.21
Increased efficiency  9.56% 32.52% 5.07% 41.96% 53.53% 18.48% 39.57% 46.46% 48.90%

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-11
Part B: Mechanical Engineering
 Fig. 10 A 942-bar truss tower

bars and 244 nodes schematically depicted in Fig. 10. This problem
aims to identify the lightest design with the design variables defined
as the member cross-sectional areas and divided into 59 group. A
single load case is considered such that it consists of lateral loads of
5.0 KN (1.12 kips) applied in both x- and y-directions and a vertical
load of –30 KN (–6.74 kips) applied in the z-direction at all nodes of
the tower. The density and elastic modulus of the material are
2767.99 kg/m3 (0.1 lb/in3) and 69 GPa (1.0  104 ksi), respectively.
The constraint conditions include allowable stresses and displace-
ments for the truss tower. The maximum allowable stress in each
member under tension and compression equals 172.37 MPa (25 ksi)
while the maximum allowable displacement in x, y, z directions for
the all the nodes is 38.1 cm (15.0 in). A discrete set of 137
economical standard steel sections selected from W-shape profile
list based on area and radii of gyration properties is used to size the
variables. The lower and upper bounds on size variables are taken as
6.16 in2 and 215.0 in2, respectively Ref. [49]. The size of the design
search space is (137)59. Further details regarding member grouping
and design constraints can be found in Ref. [50].
Compared with the above examples, this truss tower includes
more elements and load cases. Table 15 lists the final optimum
design and the corresponding results calculated by the ten methods. Fig. 11 Algorithm optimization process of the 942-bar truss

021108-12 / Vol. 10, JUNE 2024 Transactions of the ASME

Even though the lightest weight gained by TOKH algorithm is 12,330 structural analyses and KH requires 6,230 structural
marginally heavier than PO algorithm, TOKH algorithm has the analyses. Although TOKH requires a greater NSA iteration
smallest average weight and standard deviation of the optimum compared to KH algorithm, the truss weight is 88.01% lighter
weight among all the approaches. Regarding computational burden, than KH. While considering both the search robustness and the
due to a more discrete and nonlinear search space, TOKH requires optimize efficiency, the proposed achieves markedly better

Table 15 Optimized designs for the 942-bar truss

KH KHLD CKH LKH IKH TOKH OKH TLBO PO MRFO

S1 17.60 17.60 101.00 38.20 35.10 6.16 23.20 59.20 31.10 11.50
S2 48.80 20.00 18.30 42.10 38.80 6.48 11.80 91.50 6.16 16.80
S3 16.20 32.70 51.30 61.80 17.90 6.16 20.80 51.80 7.68 15.60
S4 68.80 21.80 50.00 15.60 83.30 6.16 9.13 48.80 6.16 7.69
S5 8.25 15.60 6.16 6.49 17.60 6.16 11.20 35.90 6.49 13.00
S6 8.25 25.90 118.00 10.60 20.80 6.16 14.70 147.00 10.60 9.71
S7 178.00 19.10 56.80 6.16 18.30 8.25 10.30 6.16 6.16 8.25
S8 196.00 24.10 7.61 10.60 12.60 6.16 9.13 162.00 6.16 14.10
S9 26.50 68.80 35.10 101.00 32.00 27.30 25.30 32.90 13.20 29.80
S10 7.08 13.30 74.10 17.90 8.25 11.20 7.61 7.08 6.16 17.10
S11 14.70 101.00 83.30 88.20 20.80 7.61 23.20 6.16 6.16 9.71
S12 43.20 28.20 44.70 15.80 27.30 8.84 17.90 59.20 6.16 8.25
S13 44.70 55.80 38.80 8.79 31.10 6.16 38.80 6.16 6.16 14.10
S14 29.80 9.71 6.16 11.50 9.13 7.69 20.80 59.20 6.16 7.08
S15 44.70 15.80 81.90 26.50 6.49 8.25 7.08 117.00 6.16 14.10
S16 43.20 15.80 6.16 10.60 39.90 7.08 25.30 23.20 6.16 8.85
S17 61.80 50.00 109.00 19.70 18.30 6.49 31.10 74.10 13.30 11.20
S18 32.90 17.90 6.16 10.00 42.70 6.16 10.60 6.16 6.16 8.85
S19 35.10 88.20 125.00 35.90 21.80 6.48 11.80 162.00 6.49 11.70
S20 22.60 35.90 28.50 46.70 51.30 6.16 11.20 72.80 6.16 11.80
S21 42.10 35.30 38.20 6.16 44.70 20.00 10.30 53.60 31.10 14.70
S22 7.69 8.79 6.16 15.60 24.10 6.16 12.60 6.16 6.16 9.12
S23 162.00 11.20 74.10 13.50 14.70 8.79 14.10 80.80 6.48 16.20
S24 20.80 16.70 7.08 11.80 118.00 7.08 14.10 178.00 14.10 28.20
S25 7.08 17.60 22.30 15.60 17.60 23.20 13.20 178.00 29.40 29.10
S26 55.80 31.10 91.50 11.20 15.60 23.20 9.71 9.71 6.16 12.60
S27 61.80 83.30 55.80 8.85 14.40 9.13 12.60 18.30 6.16 21.80
S28 162.00 72.80 39.90 53.60 6.16 6.16 25.30 56.80 6.48 19.70
S29 98.80 31.10 6.16 15.60 8.25 6.16 21.10 89.60 7.69 13.00
S30 59.20 16.20 134.00 29.10 10.60 6.16 14.70 46.70 6.16 19.70
S31 44.70 10.30 50.00 25.60 19.10 35.30 18.30 178.00 31.10 20.80
S32 42.10 31.20 39.90 20.00 31.10 10.30 11.20 6.16 6.16 9.71
S33 13.50 16.80 6.16 11.20 14.40 6.48 31.20 62.10 6.16 14.70
S34 7.61 14.40 43.20 32.00 18.30 9.71 13.50 43.20 6.16 14.70
S35 26.20 23.20 125.00 10.30 31.10 7.65 9.71 14.10 6.16 12.60
S36 39.90 28.50 6.16 11.80 8.25 7.08 25.30 51.80 6.16 10.00
S37 29.10 18.30 17.10 32.90 18.30 39.90 24.30 15.60 31.10 39.90
S38 29.80 32.90 6.49 8.79 14.10 6.16 11.20 20.00 6.16 7.08
S39 35.90 21.50 16.20 11.20 8.79 6.48 27.30 125.00 6.16 10.60
S40 42.10 6.49 35.90 35.90 16.80 11.80 14.70 101.00 6.16 7.69
S41 19.70 11.80 20.80 26.50 11.20 6.16 9.71 28.50 6.16 8.25
S42 20.80 134.00 75.90 14.40 24.10 6.16 23.20 6.16 6.49 8.25
S43 22.30 25.30 59.20 28.50 42.10 48.80 35.10 125.00 51.80 32.00
S44 17.10 11.50 6.16 9.13 17.60 6.16 9.71 9.71 6.16 10.60
S45 44.70 32.00 14.10 16.70 8.25 6.16 44.70 80.80 6.16 9.71
S46 80.80 21.50 6.16 7.69 53.60 6.16 9.12 14.70 6.16 8.25
S47 21.80 7.08 6.16 23.20 12.60 7.08 7.08 101.00 6.16 17.90
S48 35.10 25.30 21.80 11.80 8.79 6.48 32.90 48.80 6.16 10.00
S49 46.30 38.20 68.80 48.80 28.50 65.40 39.90 51.30 75.90 35.90
S50 7.69 7.69 6.16 6.48 16.20 7.08 11.50 23.20 6.16 13.20
S51 29.80 59.20 25.90 22.60 16.70 6.48 10.30 6.16 6.16 11.20
S52 61.80 10.60 101.00 10.60 46.30 8.25 24.10 38.20 6.49 11.80
S53 10.00 20.00 35.10 14.70 21.80 10.30 32.70 196.00 16.20 11.20
S54 38.80 20.80 26.20 14.70 11.20 6.48 9.71 68.80 17.90 35.10
S55 19.70 72.80 162.00 26.50 38.20 31.10 27.30 162.00 31.10 26.20
S56 39.90 22.60 43.20 35.90 10.00 6.16 27.30 74.10 6.16 18.30
S57 62.10 51.80 6.16 31.10 74.10 55.80 29.80 17.10 32.90 29.80
S58 21.50 13.50 147.00 6.16 53.60 10.30 12.60 97.90 6.16 15.60
S59 215.00 32.90 38.80 16.20 31.10 6.49 9.13 107.00 6.16 9.13
NSA 6230.00 6230.00 6230.00 12230.00 7630.00 12330.00 24230.00 17910.00 7030.00 12030.00
Computational time (s) 120.11 120.64 132.06 236.92 156.79 242.04 471.50 311.18 203.98 214.48
Optimal value (lb) 7.52  105 4.47  105 6.11  105 3.29  105 4.57  105 1.70  105 2.92  105 9.09  105 1.63  105 2.06  105
Mean (lb) 1.51  106 1.41  106 7.75  105 1.38  106 1.42  106 1.81  105 3.50  105 9.63  105 2.08  105 2.33  105
Std (lb) 3.70  105 4.47  105 7.11  104 5.13  105 4.37  105 4.70  103 2.68  104 3.76  104 5.41  104 1.57  104
Increased efficiency  6.62% 48.68% 8.61% 5.97% 88.01% 76.82% 36.22% 86.23% 84.57%

Bold font in the table is used to highlight the “optimal values” among all methods.

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-13
Part B: Mechanical Engineering
 Table 16 Statistical results of the optimized designs for the four trusses using different algorithms

Example Design Variable KH KHLD CKH LKH IKH TOKH OKH TLBO PO MRFO Ranking

Planar NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00 8
10-bar truss Computational 1.33 1.42 1.51 1.81 1.64 1.80 3.01 1.54 0.50 0.58 8
time (s)
Optimal value (lb) 5.63  103 5.83  103 5.65  103 5.78  103 5.59  103 5.49  103 6.54  103 5.49  103 5.62  103 5.49  103 1
Mean (lb) 7.13  103 7.19  103 6.13  103 7.15  103 6.85  103 5.64  103 6.83  103 5.91  103 6.56  103 5.66  103 1
Std (lb) 1.51  103 9.66  102 2.53  102 1.31  103 1.14  103 4.84  101 2.86  102 2.50  102 8.01  102 1.09  102 1
Spatial NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00 8
25-bar truss Computational 0.72 0.77 0.86 1.40 0.98 1.90 1.38 2.18 0.82 1.13 8
time (s)
Optimal value (lb) 520.66 531.48 494.30 540.05 503.38 485.05 529.56 501.86 487.33 493.83 1
Mean (lb) 590.98 558.37 530.87 571.76 549.41 488.34 579.05 523.36 509.75 500.76 1
Std (lb) 39.89 19.72 18.54 16.74 36.36 2.59 37.39 14.32 25.59 5.94 1
Spatial NSA 1580.00 1580.00 1580.00 3080.00 1930.00 3100.00 6080.00 4410.00 1780.00 3030.00 8
72-bar truss Computational 1.81 1.90 1.82 3.33 2.27 4.81 7.12 5.58 1.96 3.09 8
time (s)
Optimal value (lb) 739.63 681.36 520.59 785.15 461.01 387.94 644.58 449.94 412.66 419.00 1
Mean (lb) 865.71 782.99 584.17 821.81 502.50 402.30 705.74 523.17 463.53 442.38 1
Std (lb) 67.61 65.83 49.78 33.38 24.93 6.04 40.20 48.81 51.43 18.21 1
942-bar NSA 6230.00 6230.00 6230.00 12230.00 7630.00 12330.00 24230.00 17910.00 7030.00 12030.00 8
truss tower Computational 120.11 120.64 132.06 236.92 156.79 242.04 471.50 311.18 203.98 214.48 8
time (s)
Optimal value (lb) 7.52  105 4.47  105 6.11  105 3.29  105 4.57  105 1.70  105 2.92  105 9.09  105 1.63  105 2.06  105 2
Mean (lb) 1.51  106 1.41  106 7.75  105 1.38  106 1.42  106 1.81  105 3.50  105 9.63  105 2.08  105 2.33  105 1
Std (lb) 3.70  105 4.47  105 7.11  104 5.13  105 4.37  105 4.70  104 2.68  104 3.76  104 5.41  104 1.57  104 1

performance. Figure 11 shows the optimization process for the 942- and 72-bar trusses, after 50 iterations, TOKH requires 3100
bar truss tower, which is a complex structural sizing optimization structural analyses while KH requires 1580 structural analyses.
problem with high dimensional design and nonlinear buckling But considering the computational time, TOKH takes only 0.47 s,
constraints. When attempting to solve this optimization problem, 1.18 s, and 3 s longer than KH, respectively. In the 942-bar truss, due
methods may easily trap into a local optimum. Hence, a method to the higher dimensional design space, TOKH requires more
capable of maintaining both efficient global exploration and local structural analyses, and its computational time takes 121.93 s longer
exploitation is likely to produce better results. As can be seen in than that of KH. As the truss becomes more complex, the
Fig. 11, in the early iterations, TOKH can quickly converge to a computational cost of TOKH increases more significantly, which
better global region by global exploration. Then, the minimum is means that the computational efficiency is sacrificed to improve the
further found in about 10–100 iterations through effective local robustness. But the computational time of TOKH increases within
exploitation. We can draw the conclusion that, TOKH is superior to an acceptable range.
the other algorithms during the process of optimization, while PO
and MRFO performs the second and the third best in this complex
truss sizing optimization, respectively. 7 Conclusions and Future Work
From above-analyses about the Figs. 5, 7, 9, and 11, the intrinsic A novel variant of KH algorithm, referred to as TOKH algorithm,
property of TOKH, which distinguishes it from the other methods in was developed for the sizing optimization of discrete truss
the literature, is that in optimizing the sizing of complex trusses, structures. First, a crossover operator was established between the
TOKH quickly converges to a better global region in the early “best krill” and “suboptimal krill” to produce a robust “cross krill”.
iterations by global exploration, and then the local exploitation can Second, an ILMC operator was introduced to fine-tune the “center of
be used to further find the minimum. food” and candidate solutions. The objective of TOKH algorithm
was to optimize the balance between exploration and exploitation.
Therefore, the crossover operator was used to focus on the global
6.5 Robustness and Computational Cost Analysis in Truss
exploration, whereas ILMC operator was used for the local
Sizing Optimization. To examine the robustness and computa-
exploitation. It is found that:
tional cost of TOKH, this section compares the statistical results of
the four trusses examples yielded by TOKH with those gained by 1. The time complexity experiment performed using the
nine other widely used metaheuristic algorithms. Table 16 lists the Rastrigin benchmark function demonstrated that the running
statistical results of the optimized designs for the four truss examples time of TOKH algorithm was linear based on the number of
from 20 runs using TLBO, PO, MRFO, KH, CKH, OKH, IKH, iterations, which was consistent with the inference of big-O.
KHLD, LKH, and TOKH. Among ten methods, TOKH identifies the The two operators introduced do not change the time
lightest designs in the 10-, 25- and 72-bar trusses. As the structure complexity of TOKH.
becomes more complex, the performance of TOKH degrades, 2. TOKH algorithm was compared with nine other algorithms
ranking it second in the 942-bar truss. The average weight and using 15 benchmark functions. The performance of TOKH
standard deviation obtained by TOKH ranks first among 10-, 25-, algorithm for most functions and different dimensions,
72-, and 942-bar trusses. These results indicate that TOKH possesses particularly for different types of high-dimensional func-
robustness and stability compared to the methods mentioned in the tions, was statistically superior to those of the other
literature. Results also indicate that in the optimization of 10-, 25-, metaheuristic algorithms.
72-, and 942-bar trusses, the optimization efficiency of TOKH 3. TOKH algorithm was applied to four discrete truss
algorithm has improved by 20.90, 17.37, 53.53, and 88.01%, optimization problems under multiple loading conditions.
respectively, compared to KH algorithm. With the higher the We compared the numerical results for various trusses
structural dimension and the stronger the discretization, the more obtained using TOKH algorithm with other methods in the
prominent the robustness of TOKH algorithm. literature to verify the effectiveness, efficiency, and robust-
However, to improve the robustness of optimized truss sizing, ness. The results indicated that, among the ten algorithms,
TOKH requires more computational cost than KH. In the 10-, 25-, TOKH algorithm is competitive in terms of optimal weight,

021108-14 / Vol. 10, JUNE 2024 Transactions of the ASME

 average weight, and stability. Furthermore, TOKH algorithm [20] Tapao, A., and Cheerarot, R., 2017, “Optimal Parameters and Performance of
demonstrated significantly faster convergence to the optimal Artificial Bee Colony Algorithm for Minimum Cost Design of Reinforced
Concrete Frames,” Eng. Struct., 151, pp. 802–820.
solution compared to other methods. Compared to KH [21] Bajer, D., and Zorić, B., 2019, “An Effective Refined Artificial Bee Colony
algorithm, although TOKH required slightly more computa- Algorithm for Numerical Optimization,” Inf. Sci., 504, pp. 221–275.
tional cost, its optimization efficiency improved by 20.90, [22] Gorkemli, B., and Karaboga, D., 2019, “A Quick Semantic Artificial Bee
17.37, 53.53, and 88.01%, respectively. As the complexity of Colony Programming (qsABCP) for Symbolic Regression,” Inf. Sci., 502, pp.
346–362.
the truss increased, the advantage of TOKH became more [23] Camp, C. V., and Farshchin, M., 2014, “Design of Space Trusses Using Modified
evident. The proposed TOKH algorithm can serve as an ideal Teaching-Learning Based Optimization,” Eng. Struct., 62–63, pp. 87–97.
method for handling discrete truss problems. [24] Shen, W., Guo, X., Wu, C., and Wu, D., 2011, “Forecasting Stock Indices Using
Radial Basis Function Neural Networks Optimized by Artificial Fish Swarm
Although TOKH algorithm takes on better global optimization Algorithm,” Knowl.-Based Syst., 24(3), pp. 378–385.
capability, it has a high computational cost. Therefore, further [25] Yang, L., Qi, C., Lin, X., Li, J., and Dong, X., 2019, “Prediction of Dynamic
Increase Factor for Steel Fibre Reinforced Concrete Using a Hybrid Artificial
research is needed on TOKH algorithm to improve robustness while Intelligence Model,” Eng. Struct., 189, pp. 309–318.
reducing the computational cost. [26] Gandomi, A. H., Yang, X. S., and Alavi, A. H., 2011, “Mixed Variable Structural
Optimization Using Firefly Algorithm,” Comput. Struct., 89(23–24), pp.
Funding Data 2325–2336.
[27] De Souza, R. R., Miguel, L. F. F., Lopez, R. H., Miguel, L. F. F., and Torii, A. J.,
2016, “A Procedure for the Size, Shape and Topology Optimization of
 The National Natural Science Foundation of China (Grant No. Transmission Line Tower Structures,” Eng. Struct., 111, pp. 162–184.
52278135; Funder ID: 10.13039/501100001809). [28] Awad, R., 2021, “Sizing Optimization of Truss Structures Using the Political
 The Alexander von Humboldt Foundation (Funder ID: Optimizer (PO) Algorithm,” Structures, 33, pp. 4871–4894.
[29] Tran-Ngoc, H., Khatir, S., De Roeck, G., Bui-Tien, T., and Wahab, M. A., 2019,
10.13039/100005156). “An Efficient Artificial Neural Network for Damage Detection in Bridges and
Beam-Like Structures by Improving Training Parameters Using Cuckoo Search
Data Availability Statement Algorithm,” Eng. Struct., 199, p. 109637.
[30] Savković, M. M., Bulatović, R. R., Gasić, M. M., Pavlović, G. V., and Stepanović,
The datasets generated and supporting the findings of this article A. Z., 2017, “Optimization of the Box Section of the Main Girder of the Single-
are obtainable from the corresponding author upon reasonable Girder Bridge Crane by Applying Biologically Inspired Algorithms,” Eng. Struct.,
request. 148, pp. 452–465.
[31] Yang, X. S., and Gandomi, A. H., 2012, “Bat Algorithm: A Novel Approach for
Global Engineering Optimization,” Eng. Comput., 29(5), pp. 464–483.
References [32] Bekdaş, G., Nigdeli, S. M., and Yang, X. S., 2018, “A Novel Bat Algorithm Based
[1] Kaveh, A., and Talatahari, S., 2009, “A Particle Swarm Ant Colony Optimization Optimum Tuning of Mass Dampers for Improving the Seismic Safety of
for Truss Structures With Discrete Variables,” J. Constr. Steel., 65(8–9), pp. Structures,” Eng. Struct., 159, pp. 89–98.
1558–1568. [33] Cao, H., Sun, W., Chen, Y., Kong, F., and Feng, L., 2023, “Sizing and Shape
[2] Li, L. J., Huang, Z. B., and Liu, F., 2009, “A Heuristic Particle Swarm Optimization of Truss Employing a Hybrid Constraint-Handling Technique and
Optimization Method for Truss Structures With Discrete Variables,” Comput. Manta Ray Foraging Optimization,” Expert Syst. Appl., 213, p. 118999.
Struct., 87(7–8), pp. 435–443. [34] Gandomi, A. H., and Alavi, A. H., 2012, “Krill Herd: A New Bio-Inspired
[3] Gandomi, A. H., Yun, G. J., Yang, X. S., and Talatahari, S., 2013, “Chaos- Optimization Algorithm,” Commun. Nonlinear Sci. Numer. Simul., 17(12), pp.
Enhanced Accelerated Particle Swarm Optimization,” Commun Nonlinear Sci. 4831–4845.
Numer. Simul., 18(2), pp. 327–340. [35] Gandomi, A. H., Talatahari, S., Tadbiri, F., and Alavi, A. H., 2013, “Krill Herd
[4] Yang, X. S., 2010, Nature-Inspired Metaheuristic Algorithms, Luniver Press, ,UK. Algorithm for Optimum Design of Truss Structures,” Int. J. Bio. Inspired Comput.,
[5] Benzo, P. G., Pereira, J. M., and Sena-Cruz, J., 2022, “Optimization of Steel Web 5(5), pp. 281–288.
Core Sandwich Panel With Genetic Algorithm,” Eng. Struct., 253, p. 113805. [36] Gandomi, A. H., Alavi, A. H., and Talatahari, S., 2013, “Structural Optimization
[6] Korus, K., Salamak, M., and Jasinski, M., 2021, “Optimization of Geometric Using Krill Herd Algorithm,” Swarm Intelligence Bio-Inspired Computation,
Parameters of Arch Bridges Using Visual Programming FEM Components and Elsevier, Amsterdam, The Netherlands, pp. 335–349.
Genetic Algorithm,” Eng. Struct., 241, p. 112465. [37] Abdel-Basset, M., Wang, G. G., Sangaiah, A. K., and Rushdy, E., 2019, “Krill
[7] Khodzhaiev, M., and Reuter, U., 2021, “Structural Optimization of Transmission Herd Algorithm Based on Cuckoo Search for Solving Engineering Optimization
Towers Using a Novel Genetic Algorithm Approach With a Variable Length Problems,” Multimed. Tools Appl., 78(4), pp. 3861–3884.
Genome,” Eng. Struct., 240, p. 112306. [38] Wang, G. G., Guo, L., Gandomi, A. H., Hao, G. S., and Wang, H., 2014, “Chaotic
[8] Simon, D., 2008, “Biogeography-Based Optimization,” IEEE Trans. Evol. Krill Herd Algorithm,” Inf. Sci., 274, pp. 17–34.
Comput., 12(6), pp. 702–713. [39] Wang, G. G., Deb, S., Gandomi, A. H., and Alavi, A. H., 2016, “Opposition-Based
[9] Ergezer, M., Simon, D., and Du, D., 2009, “Oppositional Biogeography-Based Krill Herd Algorithm With Cauchy Mutation and Position Clamping,” Neuro-
Optimization,” International Conference on Systems, Man and Cybernetics, San computing, 177, pp. 147–157.
Antonio, TX, Oct. 11–14, pp. 1009–1014. [40] Wang, G. G., Gandomi, A. H., Alavi, A. H., and Deb, S., 2016, “A Multi-Stage
[10] Du, D., Simon, D., and Ergezer, M., 2009, “Biogeography-Based Optimization Krill Herd Algorithm for Global Numerical Optimization,” Int. J. Artif. Intell.
Combined With Evolutionary Strategy and Immigration Refusal,” International Tools, 25(02), p. 1550030.
Conference on Systems, Man and Cybernetics, San Antonio, TX, Oct. 11–14, pp. [41] Guo, L., Wang, G. G., Gandomi, A. H., Alavi, A. H., and Duan, H., 2014, “A New
997–1002. Improved Krill Herd Algorithm for Global Numerical Optimization,” Neuro-
[11] Geem, Z. W., Kim, J. H., and Loganathan, G. V., 2001, “A New Heuristic computing, 138, pp. 392–402.
Optimization Algorithm: Harmony Search,” Simulation, 76(2), pp. 60–68. [42] Laith, M. A., Ahamad, T. K., and Essam, S. H., 2019, “Modified Krill Herd
[12] Murren, P., and Khandelwal, K., 2014, “Design-Driven Harmony Search (DDHS) Algorithm for Global Numerical Optimization Problems,” Advances in Nature-
in Steel Frame Optimization,” Eng. Struct., 59, pp. 798–808. Inspired Computing and Applications, Springer, Cham, Switzerland, pp. 205–221.
[13] Molina-Moreno, F., Garcıa-Segura, T., Martı, J. V., and Yepes, V., 2017, [43] Li, J., Tang, Y., Hua, C., and Guan, X., 2014, “An Improved Krill Herd Algorithm:
“Optimization of Buttressed Earth-Retaining Walls Using Hybrid Harmony Krill Herd With Linear Decreasing Step,” Appl. Math. Comput., 234(10), pp.
Search Algorithms,” Eng. Struct., 134, pp. 205–216. 356–367.
[14] Storn, R., and Price, K., 1997, “Differential Evolution: A Simple and Efficient [44] Wang, G. G., Gandomi, A. H., and Alavi, A. H., 2014, “Stud Krill Herd
Heuristic for Global Optimization Over Continuous Spaces,” J. Glob. Optim., 11(4), Algorithm,” Neurocomputing, 128, pp. 363–370.
pp. 341–359. [45] Wang, G., Guo, L., Gandomi, A. H., Cao, L., Alavi, A. H., Duan, H., and Li, J.,
[15] Talatahari, S., Gandomi, A. H., Yang, X. S., and Deb, S., 2015, “Optimum Design 2013, “Levy-Flight Krill Herd Algorithm,” Math. Probl. Eng.., 2013, pp. 1–14.
of Frame Structures Using the Eagle Strategy With Differential Evolution,” Eng. [46] Lee, K. S., Geem, Z. W., Lee, S., and Bae, K. W., 2005, “The Harmony Search
Struct., 91, pp. 16–25. Heuristic Algorithm for Discrete Structural Optimization,” Eng. Optimiz., 37(7),
[16] Dorigo, M., and Stutzle, T., 2004, Ant Colony Optimization, MIT Press, pp. 663–684.
Cambridge, MA. [47] Cheng, M. Y., Prayogo, D., Wu, Y. W., and Lukito, M. M., 2016, “A Hybrid
[17] Eberhart, R., and Kennedy, J., 1995, “A New Optimizer Using Particle Swarm Harmony Search Algorithm for Discrete Sizing Optimization of Truss Structure,”
Theory,” Proceedings of the Sixth International Symposium on Micro Machine Autom. Constr., 69, pp. 21–33.
and Human Science, Nagoya, Japan, Oct. 4–6, pp. 39–43. [48] Knuth, D. E., 1976, “Big Omicron and Big Omega and Big Theta,” ACM Sigact
[18] Minh, H. L., Khatir, S., Wahab, M. A., and Cuong-Le, T., 2021, “An Enhancing News, 8(2), pp. 18–24.
Particle Swarm Optimization Algorithm (EHVPSO) for Damage Identification in [49] Construction, A., 1989, Manual of Steel Construction: Allowable Stress Design,
3D Transmission Tower,” Eng. Struct., 242, p. 112412. American Institute of Steel Construction, Chicago, IL.
[19] Jiang, W., Xie, Y., Li, W., Wu, J., and Long, G., 2021, “Prediction of the Splitting [50] Hasançebi, O., Çarbaş, S., Dogan, E., Erdal, F. E., and Saka, M. P., 2009,
Tensile Strength of the Bonding Interface by Combining the Support Vector “Performance Evaluation of Metaheuristic Search Techniques in the Optimum
Machine With the Particle Swarm Optimization Algorithm,” Eng. Struct., 230, Design of Real Size Pin Jointed Structures,” Comput. Struct., 87(5–6), pp.
p. 111696. 284–302.

ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, JUNE 2024, Vol. 10 / 021108-15
Part B: Mechanical Engineering

View publication stats