Structures 13 (2018) 213–229

Contents lists available at ScienceDirect

Structures
journal homepage: www.elsevier.com/locate/structures

Optimum Design of Double-layer Barrel Vaults by Lion Pride Optimization T

Algorithm and a Comparative Study
⁎
A. Kaveh , S. Mahjoubi
Centre of Excellence for Fundamental Studies in Structural Engineering, Iran University of Science and Technology, Narmak, Tehran, P.O. Box 16846-13114, Iran

A R T I C L E I N F O A B S T R A C T

Keywords: In this article, the newly developed optimization method, so-called the Lion Pride Optimization Algorithm
Lion Pride Optimization Algorithm (LPOA), is applied to optimal design of double-layer space barrel vaults. In order to demonstrate the perfor-
Barrel vaults mance of the LPOA, three large-scale benchmark optimization design problems of double-layer barrel vaults are
Size-optimization optimized and the results are compared with those of some metaheuristics from literature. The second aim of this
Optimal design
article is to solve these examples using three other robust metaheuristic algorithms, namely Artificial Bee Colony
Constrained problems
Meta-heuristic search
(ABC), Cuckoo Search (CS), and Particle Swarm Optimization (PSO). A comparative study of these algorithms
shows the suitability of the LPOA for solving real-world practical spatial truss structures.

1. Introduction prone to instability, especially under the action of heavy unsymmetrical

A barrel vault is an effective semi-cylindrical form of the roof
system, which is widespread for multipurpose facilities including
warehouse, rail station, pools, sports center, airplane hangars and
community centers as providing a long-span and economical roof with a
significant amount of space. Barrel roofs can be used for covering rec-
tangular structures [1]. A barrel vault, typically consists of a single or
multiple braced layers of bar elements that are arched in the width
direction. However, the barrel vaults usually have uniform cross-sec-
tions along their lengths. The shape of the cross-section of a barrel vault
may vary along its longitudinal axis. The curvature of a barrel roof is
created by a motion movement of a single or multiple layer grid along a
space curve which is called a directrix. The directrix can be a circular
arc, an ellipse, a catenary, a parabola or a cycloid [2]. Barrel vaults are
given different names depending on the way their surface is formed [3].
There are disparate possible types of bracing that can be used in barrel
roofs. But theoretically, just fully triangulated barrel vault systems can
be analyzed as pin-jointed structures. The barrel vaults, having the
quadrilateral or hexagonal types of bracing, must have rigid joints to be
stable and consequently, their elements are under bending moments,
shear forces and torques.
As a rule, the earlier types of braced barrel vaults were constructed
as single-layer structures, but nowadays, double-layer barrel roofs are
usually preferred by designers in the construction of long-span and
support-free roofs [1,4]. Double-layer barrel vaults have better struc-
tural behavior inherently as their architectural forms. The experiments
have also shown that large span single layer braced barrel vaults are
loads and that the rigidity of joints can exert an important influence on
the overall stability of the structure [5]. The single-layer barrel vaults
are mainly under the action of flexural moments, the component
members of double-layer barrel vaults are almost exclusively under the
action of axial forces and the elimination of bending moments leads to a
full utilization of strength of all the elements [3].
Turning to the procedure of optimization, it should be stated that
choosing layout variables that satisfy all design constraints and de-
clining the aggregate costs is a grave concern for structural engineers,
specifically for sophisticated and complex design problems.
Optimization methodologies can be powerful initiatives that offer fea-
sible and economically good solutions for the designers.
Generally, Optimization techniques seek good feasible variable set
in order to minimize or maximize single or multiple objective functions
systematically considering a predefined search space and a set of con-
straints. These methods can be divided into two general distinct groups:
mathematical gradient-based techniques and stochastic non-gradient
approaches. If the objective function of an optimization problem is
smooth (i.e., differentiable) and gradient information is reliable, then
gradient-based optimization algorithms present an extremely powerful
collection of tools for solving the problem [6]. However, in real-world
structural design problems, finding the relation between the design
variables, loads, and behavioral parameters such as deflections,
stresses, failure modes or frequencies may not be possible or it may be
difficult to compute. In addition, the implementation success of the
mathematical programming methods is intensively depending on good
starting point. It should be noted that some mathematical programming

⁎
Corresponding author.
E-mail address: alikaveh@iust.ac.ir (A. Kaveh).

https://doi.org/10.1016/j.istruc.2018.01.002
Received 3 November 2017; Received in revised form 3 January 2018; Accepted 15 January 2018
2352-0124/ © 2018 Institution of Structural Engineers. Published by Elsevier Ltd. All rights reserved.
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

based methods have been developed for discrete optimum design pro- Table 1
blems that are not very efficient for obtaining the optimum solution of The LPOA parameters selected in this study.
the large size practical design problems [7–9]. The objective functions
Parameters Value
of real-world structural design problems are oversensitive, non-convex,
unreliable and rough. Hence, gradient-based methods are not efficient Number of prides 3–7
for these types of problems. In the last two decades, the researchers Lions in each prides 4–7
Male lions in each prides 1–2
have proposed several stochastic methodologies to solve complicated
Female lions in each prides 3–6
problems. For instance: the Genetic Algorithm (GA) [10] that is con- Territory ratio 0.5
ceptualized using the Darwin's natural select theory; the Simulated Mating probability 0.1
Annealing (SA) [11] is inspired by the annealing in metallurgy; the Ant Immigration rate 0.1
Colony Optimization Algorithm (ACO) [12] that mimics the behavior of Diversification factor 0.1 to 0.00001

ants seeking a path between their colonies and food sources; Particle
Swarm Optimization (PSO) [13] that is based on social behavior of fish
Table 2
schooling or bird flocking; the Harmony Search (HS) [14] that is in- The steel pipe sections taken from AISC-LRFD code.
spired by the improvisation process of musicians; the Big Bang Big
Crunch (BB-BC) algorithm that is inspired by big bang theory; Artificial Num. Type Nominal Area (in2) Moment of Gyration radius
Bee Colony (ABC) [15] that is based on the swarm intelligence of honey diameter (in) inertia (in4) (in)

bees; Cuckoo Search (CS) [16] algorithm that is inspired by life of the 1 ST 1/2 0.25 0.017 0.2608
cuckoo birds; Charged System Search (CSS) [17] method that imitate 2 EST 1/2 0.32 0.02 0.2500
the laws of physics and mechanics; Colliding bodies optimization al- 3 ST 3/4 0.33 0.037 0.3333
gorithm (CBO) that is inspired by the collision between objects [18]; 4 EST 3/4 0.43 0.045 0.3224
5 ST 1 0.49 0.087 0.4197
Teaching Learning Based Optimization (TLBO) [19] that is based on the
6 EST 1 0.64 0.11 0.4073
philosophy of learning by teaching; Mine Blast (MB) algorithm [20] 7 ST 1 1/4 0.67 0.19 0.5399
that is based on landmines explosion; Dragonfly Algorithm (DA) [21] 8 ST 1 1/2 0.8 0.31 0.6229
that mimics the swarm intelligence of dragonflies; Virus Optimization 9 EST 1 1/4 0.88 0.24 0.5241
10 EST 1 1/2 1.07 0.67 0.7889
Algorithm (VOA) [22] that imitates the behavior of viruses attacking a
11 ST 2 1.07 0.39 0.6045
living cell; and Drone Squadron Optimization (DSO) [23] is artifact- 12 EST 2 1.48 0.87 0.7658
inspired method inspired by Drone Squadron in taking earth mon- 13 ST 2 1/2 1.7 1.54 0.9515
itoring. Besides, there is a number of hybrid or modified algorithms that 14 ST 3 2.23 3.02 1.1637
are proposed by researchers such as [24–31]. 15 EST 2 1/2 2.25 1.92 0.9238
16 DEST 2 2.66 1.31 0.7018
In the recent decades, different meta-heuristic algorithms have been
17 ST 3 1/2 2.68 4.79 1.3369
developed and applied to the solution of structural optimization pro- 18 EST 3 3.02 3.89 1.1349
blems. For instance, Teaching–Learning Based Optimization (TLBO) 19 ST 4 3.17 7.23 1.5102
was used for designing steel frames. Adaptive Dimensional Search 20 EST 3 1/2 3.68 6.28 1.3063
(ADS) method was designed for discrete truss sizing optimization [32]. 21 DEST 2 1/2 4.03 2.87 0.8439
22 EST 5 4.3 15.2 1.8801
The optimum layout design of multi-span reinforced concrete beams 23 EST 4 4.41 9.61 1.4762
under dynamic loadings was carried out using Ant Colony Optimization 24 DEST 3 5.47 5.99 1.0465
(ACO) algorithm [33]. Optimum column layout design of reinforced 25 ST 6 5.58 28.1 2.2441
concrete frames under wind loading is performed by Sharafi et al. 26 EST 5 6.11 20.7 1.8406
27 DEST 4 8.1 15.3 1.3744
[34,35]. The graph theory and the ant colony based algorithm are
28 ST 8 8.4 72.5 2.9378
combined to solve the problems of thin-walled steel sections [36]. Ar- 29 EST 6 8.4 40.5 2.1958
tificial Neural Networks (ANN) was applied to a post-tensioned con- 30 DEST 5 11.3 33.6 1.7244
crete road bridge design problem [37]. Notably, in the field of size 31 ST 10 11.9 161 3.6782
optimization of double-layer barrel vault frames, some studies are 32 EST 8 12.8 106 2.8777
33 ST 12 14.6 279 4.3715
carried out earlier by Refs. [3,32,38–40]. Although, the dimensions of 34 DEST 6 15.6 66.3 2.0616
structures may be governed stringently by construction requirements, 35 EST 10 16.1 212 3.6287
design standards, the purpose of the structure, and/or client demands, 36 EST 12 19.2 362 4.3421
topology, shape and size optimization method can be performed con- 37 DEST 8 21.3 162 2.7578
currently. In accordance with the study of Kaveh et al. [41] this ap-
ST = standard weight, EST = extra strong, DEST = double-extra strong.
proach should be an effective way in cases when the height of barrel
vaults can also be considered as a variable.
sections provided by American Institute of Steel Construction (AISC)
This paper utilizes a recently developed nature-inspired metaheur-
code [43] in the method. Three numerical examples are investigated to
istic algorithm, so-called Lion Pride Optimization Algorithm (LPOA),
verify the robustness of the mentioned technique in finding good op-
for size optimization of double-layer barrel vault structures. To further
timal solutions for this kind of design problems. The outcomes of the
validate the applicability of the LPOA, Artificial Bee Colony (ABC) [15],
LPOA are also compared to those of some state-of-the-art optimization
Cuckoo Search (CS) [16], and Particle Swarm Optimization (PSO) [13]
techniques to illustrate the efficiency of the approach.
are used to solve the design problems. In these algorithms the bar
The remainder of the paper is organized as follows: the utilized
sections are the variables, and the design limitations are the constraints
optimization approaches are presented briefly in Section 2; three
of the optimization problems. Finally, the discrete search space is a
practical designs of double-layer barrel vault roofs are performed and
multi-dimensional area, and the number of dimensions is equal to the
comparative study of the mentioned optimization approaches are con-
number of variables, and each variable is selected from the pre-defined
ducted in Section 3; finally, the concluding remarks are derived in
list of sections.
Section 4.
The optimization procedure obtains a minimum weight of double-
layer barrel roof structures subjected to the AISC-ASD [42] specifica-
tions.
The optimization variables are selected from the industrial cross

214
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 1. 3D view (a), plan view with group numbers (b), and
flatten cross-sectional view (c) of the 384-bar double-layer
barrel vault.

2. Optimization approaches non-gradient algorithm. The total population is grouped into different
pride social groups and the behavior of each agent is inspired by the
2.1. Lion Pride Optimization Algorithm social life of the lions and lionesses. A lion pride consists of lionesses,
offspring and a small number of male lions. Each member has a specific
This sub-section provides a brief introduction of the LPOA [44] as a responsibility in the group. Female lions have cooperative strategies for

215
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 2. Convergence curves for the 384-bar double-layer barrel

vault (load Case 1).

Fig. 3. Convergence curves for the 384-bar double-layer barrel

vault (load Case 2).

hunting. Male lions protect their territory against invading lions, and variable of the jth search agent (lion), and amin and amax are the
for this reason, they usually excurse in their dominated areas. In the minimum and the maximum permissible values, respectively. In other
following, the mathematical model of the algorithm is provided [44]. words, in this section, first positions of lions are selected randomly.
Lions live together and form social groups named as the pride:
2.1.1. Generate initial lions and prides
The first step is to initialize the vector of lion matrices with the ⎛ lion1, k ⎞ ⎛ a1,1, k … a1, n, k ⎞
number of design variables and then evaluate their associated fitness Pridek = ⎜ ⋮ ⎟ = ⎜ ⋮ ⋱ ⋮ ⎟
⎜ lion ⎟
functions. ⎝ pk , k ⎠ ⎝ apk,1, k ⋯ apk, n, k ⎠ (3)
amin, i < ai < amax, i (1)
where Pridek is the kth pride group and pk represents the population
lionj = [a1, j,⋯, an, j] (2) number of residents in the kth pride. The residents of each pride group
are selected randomly.
where lionj is the initial position of the jth lion, n represents the number
of design variables, ai,j is the vector's components of the ith design

216
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Table 3
Comparative results for the 384-bar braced barrel vault for the load Case 1.

Element group Kaveh & Moradveisi [3] Present work

Engineering design CBO ECBO ABC CS PSO LPOA

1 ST 1 1/4 ST 1 1/4 ST 1 1/4 EST 1 ST 1/2 ST 1/2 ST 1/2

2 EST 2 EST 2 ST 2 1/2 ST 2 1/2 EST 1 1/2 ST 3 ST 2 1/2
3 EST 2 EST 3 EST 2 EST 3 ST 2 1/2 EST 1 1/2 EST 1 1/2
4 ST 1 1/4 ST 1 1/4 ST 1 1/4 EST 1 1/2 EST 1 1/2 ST 1 1/2 ST 1 1/2
5 EST 4 DEST 2 DEST 2 1/2 DEST 3 EST 2 1/2 EST 5 DEST 3
6 DEST 8 EST 1 1/2 ST 1 1/4 ST 1 1/2 ST 1 1/2 ST 1 1/2 ST 1 1/2
7 ST 12 EST 10 EST 12 ST 12 ST 12 DEST 8 ST 10
8 EST 8 DEST 5 DEST 5 ST 10 DEST 4 EST 8 DEST 5
9 ST 10 DEST 6 ST 10 EST 10 EST 12 DEST 6 EST 10
10 EST 10 EST 10 ST 12 ST 12 EST 8 DEST 8 DEST 6
11 ST 8 DEST 5 ST 8 ST 10 DEST 5 EST 6 DEST 5
12 EST 8 ST 12 ST 12 ST 10 DEST 6 EST 8 EST 8
13 EST 5 EST 5 DEST 4 EST 5 ST 6 ST 6 DEST 3
14 ST 8 ST 5 ST 6 DEST 2 1/2 EST 4 EST 3 1/2 ST 5
15 ST 3 1/2 ST 3 1/2 ST 3 ST 3 ST 2 1/2 ST 2 1/2 ST 2 1/2
16 ST 6 DEST 2 1/2 DEST 2 1/2 ST 4 DEST 3 ST 4 EST 4
17 ST 8 EST 5 EST 5 EST 5 EST 5 DEST 2 1/2 EST 5
18 EST 1 1/2 EST 2 ST 2 DEST 2 EST 1 1/4 ST 2 1/2 EST 1 1/2
19 ST 1 1/4 ST 1 1/4 ST 1 1/4 EST 2 ST 1 1/4 ST 1 1/4 ST 1 1/4
20 EST 2 ST 1 1/4 ST 2 EST 2 ST 1 1/4 EST 2 EST 1 1/2
21 EST 2 EST 2 EST 1 1/4 ST 2 1/2 EST 1 1/2 ST 2 1/2 EST 1 1/2
22 EST 2 ST 1 1/4 ST 1 1/4 EST 2 ST 1 1/4 ST 1 1/4 ST 1 1/4
23 EST 2 EST 2 ST 2 EST 1 1/2 EST 1 1/2 EST 1 1/2 EST 1 1/2
24 ST 4 EST 3 ST 4 EST 3 EST 2 1/2 EST 3 ST 3 1/2
25 ST 2 1/2 EST 2 EST 2 1/2 EST 2 ST 2 1/2 ST 2 1/2 ST 2 1/2
26 ST 3 EST 2 ST 2 1/2 ST 2 1/2 ST 2 1/2 EST 1 1/2 ST 2 1/2
27 DEST 2 1/2 DEST 2 ST 3 1/2 DEST 2 EST 3 ST 2 1/2 DEST 2
28 ST 2 1/2 EST 2 EST 2 ST 3 ST 3 EST 1 1/2 EST 2
29 ST 2 1/2 ST 2 1/2 ST 2 EST 2 ST 2 1/2 EST 2 EST 2
30 EST 2 EST 2 ST 2 1/2 EST 2 EST 2 1/2 EST 1 1/2 EST 2
31 EST 2 EST 2 EST 2 ST 2 1/2 EST 2 EST 1 1/2 EST 1 1/2
Weight (kips) 71.121.0 64.061.8 62.644.8 65.420.66 63.995.51 64.859.22 60.977.72
Weight (tons) 32.260 29.058 28.415 29.674 29.028 29.420 27.659
Number of analyses – 30,000 30,000 30,000 30,000 30,000 30,000

2.1.2. Main loop Wingernew = Prey + H2 × |W | × rand + (D) × (2 rand − 1) (8)

Metaphorically speaking, lions and lionesses are the search agents of
the LPOA and the search space is their habitat. W = Prey − Winger = [w1+⋯+ wn] (9)

2.1.2.1. Hunting. Resident lionesses usually go hunting as a team. Each
lioness takes-over the specific role in hunting process. Three different
types of rules are assumed as Chaser, Cheater (refrainer) and Winger.
The chaser pursues the prey directly and Wingers attack the victims in
other directions. In the meantime, cheaters just run to narrow down the
prey, and usually never hunt in the presence of a cooperator. Therefore,
individuals that do not participate in group hunts withhold effort that
increases the group's success rate [45].
The formation of the hunter groups is assumed as the following: all
female lions in each pride sort as their fitness and grouped into three
main groups. The best female lions are named as chasers, the second
best group is assumed as wingers and the third group is set as cheaters.
One member of each of these three general groups is selected un-
methodically and form hunting groups.
The mathematical model of the chaser hunters is as follows:
Chasernew = Chaser + H1 × rand + (D) × (2 rand − 1)
|W | = w12+⋯+wn 2 (10)
where Wingernew represents the new position of the Winger hunter lion
and H2 is a random unit vector perpendicular to the vector W.
The new position of the cheaters follows the equation below:
Cheaternew = Prey + H3 × rand + (D ) × (2 rand − 1) (11)
with
H3 = (Prey − Cheater ) (12)
where Cheaternew is the new position of the cheaters and Cheater is the
present position of the cheater.
2.1.2.2. Excursion. Each male lion in a pride moves to their pride's
territory to protect their home range. To simulate this behavior, males
visit their territories randomly and explore the territorial areas. The
(4) movement of male lions is formulated as the following:

H1 = (Prey − Chaser ) (5) Malelionnew = Territory + E × (D ) × (2 rand − 1) (13)

D = DF × amp (6) Territoryi = BestPositions (rand × TR)i (14)

amp = amax − amin
where Chasernew and chaser are the new and existing positions of Chaser
lion respectively, D represents diversification matrix, DF is diversifica-
tion factor, rand is a random value in the interval [0,1], and Prey is a
randomly chosen from the best positions of search agents until now.
The new position of wingers follows the equation below:
(7)
where Malelionnew is the new position of the male lion, Territory is
randomly selected from the best positions in the pride members,
BestPositions is the cumulative best position of the resident lions
(including both sexes) that is sorted according to the fitness values
and ranked from best to worst, E is the excursion constant, and TR
represents the territory ratio.

217
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Table 4
Comparative results for the 384-bar braced barrel vault for the load Case 2.

Element group Kaveh & Moradveisi [3] Present work

Engineering design CBO ECBO ABC CS PSO LPOA

1 ST 1 1/4 ST 1 1/4 ST 1 1/4 ST 1/2 EST 1/2 ST 1/2 ST 1/2

2 EST 2 EST 2 ST 2 EST 1 1/2 EST 2 EST 1 1/2 EST 1 1/2
3 ST 1 1/4 ST 1 1/4 ST 1 1/4 ST 1 1/4 ST 1 1/4 ST 1 1/4 ST 1 1/4
4 DEST 2 ST 3 1/2 ST 3 1/2 EST 2 1/2 DEST 2 EST 3 1/2 ST 3 1/2
5 EST 2 EST 2 EST 2 EST 2 EST 2 ST 2 1/2 EST 1 1/2
6 EST 2 EST 2 ST 2 EST 1 1/2 EST 2 EST 1 1/2 EST 1 1/2
7 EST 5 EST 5 EST 6 EST 4 ST 6 EST 5 ST 6
8 EST 5 ST 8 DEST 4 EST 5 EST 6 EST 6 DEST 4
9 EST 5 DEST 3 EST 3 ST 5 EST 3 1/2 ST 4 ST 4
10 ST 5 ST 3 1/2 ST 3 ST 2 1/2 ST 2 1/2 ST 2 1/2 ST 2 1/2
11 ST 5 ST 3 1/2 ST 3 ST 2 1/2 ST 2 1/2 ST 2 1/2 ST 2 1/2
12 ST 5 EST 3 1/2 ST 4 ST 4 ST 4 EST 3 EST 3
13 DEST 2 EST 1 1/2 EST 1 1/2 EST 2 EST 2 EST 2 EST 1 1/2
14 ST 2 1/2 ST 3 1/2 ST 2 1/2 ST 2 1/2 ST 3 1/2 ST 3 ST 2 1/2
15 EST 3 ST 4 ST 4 DEST 3 EST 3 ST 5 DEST 2 1/2
16 DEST 2 1/2 EST 3 1/2 ST 5 ST 5 EST 3 DEST 2 1/2 ST 5
17 ST 5 ST 4 EST 3 1/2 EST 3 1/2 ST 5 EST 3 1/2 ST 4
18 EST 1 1/2 ST 1 1/2 ST 1 1/2 ST 1 1/4 EST 1 1/2 EST 2 1/2 EST 1 1/2
19 EST 2 EST 2 ST 2 ST 1 1/4 EST 1 1/2 EST 1 1/2 EST 1 1/2
20 EST 2 EST 2 ST 2 EST 1 1/2 EST 1 1/2 EST 2 EST 1 1/2
21 EST 2 EST 2 ST 2 EST 2 EST 2 EST 1 1/2 EST 1 1/2
22 ST 2 1/2 EST 2 ST 2 EST 1 1/2 EST 2 1/2 EST 1 1/2 EST 1 1/2
23 ST 1 1/2 EST 2 ST 1 1/4 ST 1 1/4 EST 1 1/2 ST 1 1/4 ST 1 1/4
24 EST 1 1/4 EST 1 ST 1 ST 1 1/4 EST 1 1/4 ST 1 1/4 ST 1 1/4
25 EST 2 EST 2 ST 1 1/2 ST 1 1/2 EST 1 1/2 ST 1 1/2 ST 1 1/2
26 EST 1 1/2 ST 1 1/2 ST 1 1/2 EST 1 1/2 EST 1 1/2 EST 1 1/2 EST 1 1/2
27 EST 2 EST 2 1/2 ST 2 ST 2 1/2 ST 2 1/2 ST 2 1/2 EST 2
28 EST 2 EST 2 ST 2 EST 2 ST 1 1/2 ST 1 1/2 ST 1 1/2
29 EST 2 1/2 EST 2 ST 2 EST 3 1/2 DEST 2 ST 2 1/2 EST 2 1/2
30 DEST 2 ST 2 1/2 ST 3 EST 2 ST 2 1/2 ST 1 1/2 EST 2
31 ST 2 1/2 EST 2 ST 2 EST 2 EST 1 1/2 EST 1 1/2 EST 1 1/2
Weight (kips) 36.636 32.937 29.423 31.413 31.127 29.167 28.560
Weight (tons) 16.617 14.940 13.346 14.249 14.119 13.230 12.955
Number of analyses – 30,000 30,000 30,000 30,000 30,000 30,000

nm
Table 5 1−β
Statistical results of different methods for the 384-bar braced barrel vault for the load
Offspring1 = β × FemaleLion + ∑ nm × MaleLionl × Sl
l=1
∑l = 1 Sl (15)
Case 1 (lbs).

Algorithm Average Maximum Minimum Standard deviation nm

1−β
Offspring2 = (1 − β ) × FemaleLion + ∑ nm × MaleLionl × Sl
ABC 67,231.74 68,315.15 65,420.66 908.81 l=1
∑l = 1 Sl
CS 66,872.09 68,151.43 63,995.51 1178.48
(16)
PSO 76,268.36 98,622.48 64,859.22 10,368.10
LPOA 62,258.80 65,451.46 60,977.72 1010.11
where FemaleLion is the best position of the selected mother and
MaleLion represents the best position of males in the pride, Sl equals
Table 6 to 1 if the male l is in the coalition, otherwise it is equal to 0; nm is the
Statistical results of different methods for the 384-bar braced barrel vault for the load number of resident males in the pride and β is a randomly generated
Case 2 (lbs). number with a normal distribution with mean value 0.5 and standard
deviation 0.1. It is assumed that the gender of these two newborns is
Algorithm Average Maximum Minimum Standard deviation
random.
ABC 37,709.47 44,980.53 31,413.21 4543.21
CS 31,896.35 33,896.35 31,127.40 669.11
PSO 32,432.54 46,415.71 29,166.73 3631.69
2.1.2.4. Intragroup interaction. Male lions grow into mature lions and
LPOA 29,556.00 31,130.91 28,560.00 786.04 become aggressive and fight other males in their pride. As male mature
lions, any male cubs that became mature are a new rival to his throne
and so must be eliminated. Male lions in the same pride fight each
2.1.2.3. Mating. The basic concept of mating proposed by Yazdani and other, weaker ones beaten with the pride and become a nomad. This
Jolai [46] is used to model the mating of lions. Female lions sorted as behavior is represented by the following: number of male lions in each
their best fitness values derived until the current iteration and M% pride are in equilibrium, and weaker agents (according to their fitness
(mating probability) of female lions in each pride that have a better best values) must leave the pride and become a nomad.
solution so far in the current iteration mate with one or several resident
males. These males are selected randomly from the same pride as the
2.1.2.5. Migration. To simulate this natural phenomenon of pride
selected females. Two offspring breed in each mating according to the
resident lions' migration, some female lions in each pride migrate
following equations:
with the probability of immigration rate (I%) in every iteration. Also,
surplus female lions in each pride get out of the pride and become a
nomad.

218
A. Kaveh, S. Mahjoubi
Fig. 4. Displacement ratios of the best-obtained solution with the LPOA in the 384-bar barrel vault design problem.
Fig. 5. Strength ratios of the best-obtained solution with the LPOA in the 384-bar barrel vault design problem (a) load Case 1, (b) load Case 2.
2.1.3. Algorithm parameters
Parameter values of the LPOA in the evaluation of the double-layer
barrel vault examples are provided in Table 1. It should be noted that
the parameters in here are adjusted for this type of design optimization
problems.
2.2. Artificial Bee Colony
The Artificial Bee Colony algorithm (ABC) is an optimization algo-
rithm based on the intelligent foraging behavior of honey bee swarm,
proposed by Karaboga [15]. The colony consists of three groups of bees:
employed bees, onlookers, and scouts. It is assumed that there is only
one artificial employed bee for each food source. In other words, the
number of employed bees in the colony is equal to the number of food
sources around the hive. Employed bees go to their food source and
come back to hive and dance in this area. The employed bee whose food
source has been abandoned becomes a scout and starts to search for
finding a new food source. Onlookers watch the dances of employed
bees and choose food sources depending on dances [15]. The mathe-
matical formulation of this technique is provided in the following
subsection.
219
Structures 13 (2018) 213–229
2.2.1. Generate initial population of food sources
The first step is to generate a random initial population of workers,
where, SN denotes the swarm size and food sources are the solutions.
2.2.2. Main loop
The first half of the swarm consists of the employed bees, and the
second half constitutes the onlooker bees. It should be noted that the
number of worker bees or the onlooker bees is equal to the number of
solutions in the swarm.
2.2.2.1. Employed bees phase. Each employed bee, Xi, generates a new
candidate solution, Vi in the neighborhood of its present position as
equation below:
for i ≠ j: Vik = Xik + Φik × (Xik − X jk ) (17)
where Vi represents a candidate solution, Xk is a randomly selected food
source, Φik is a random number within [−a,a] in which a represents the
acceleration coefficient upper bound. Once the new candidate solution
Vi is generated, if the fitness value of Vi is better than that of its parent,
then update Xi = Vi; otherwise, keep Xi unchanged.
A. Kaveh, S. Mahjoubi
Fig. 6. Maximum strength and slenderness ratios of the best-obtained solution with the LPOA in the 384-bar barrel vault design problem for different group elements (a) load Case 1, (b)
load Case 2.
2.2.2.2. Onlooker bees phase. An onlooker bee evaluates the
information taken from all employed bees and chooses a food source
with a probability related to its objective function values. This
probabilistic selection named a roulette wheel selection mechanism is
formulated as below:
fiti
Pi =
∑ fit
where Pi is the probability of selection and fiti is the nectar amount of
the ith worker.
2.2.2.3. Scout bees phase. If a position cannot be improved over a
predefined number (limit) of cycles, then the food source is abandoned.
If Xi is selected as the abandoned food source, the Xi is replaced by a
randomly selected solution in defined search space.
2.2.3. Algorithm parameters
The acceleration coefficient upper bound is chosen as 0.5 and the
scout limit is selected as 500.
2.3. Cuckoo Search
Cuckoo Search is a metaheuristic algorithm inspired by some spe-
cies of Cuckoo that is developed by Yang & Deb [16]. These species lay
Structures 13 (2018) 213–229
their eggs in the nests of other host birds with amazing abilities like
selecting the recently spawned nests and removing existing eggs that
increase the hatching probability of their eggs. The host takes care of
the eggs presuming that the eggs are its own. However, some of the host
birds are able to combat with this parasite behavior of cuckoos, and
throw out the discovered alien eggs or build their new nests in new
locations. A generation is represented by a set of host nests. Each nest
(18)
carries an egg (solution). The quality of the solutions is improved by
generating a new solution from an existing solution and modifying
certain characteristics. The number of solutions remains fixed in each
generation [47]. The formulation of the algorithm is described as fol-
lows.
2.3.1. Generate initial nests or eggs of host birds
The initial locations of the nests are determined by the set of values
randomly generated.
2.3.2. Main loop
2.3.2.1. Generate new cuckoos by Lévy flights. All the nests except for the
best nest are replaced by new cuckoo eggs produced from their
positions as the following:
nesti(t + 1) = nesti(t ) + α. S. (nesti(t ) − nestbest
(t )
). r (20)
where nestit is the ith nest current position, α is the step size parameter; r
220
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 7. 3D view (a), plan view with group numbers of the top
layer (b), and flatten cross-sectional view with the group number
of bracing and the top layer elements (c) of the 693-bar double-
layer barrel vault.

is a random number from a standard normal distribution and nestbest is 1 if rand < pa
Pij = ⎧
the position of the best nest so far; and S is a random walk based on the
⎩0
⎨ if rand ≥ pa (24)
Lévy flights. The Lévy flight essentially provides a random walk while
the random step length is drawn from a Lévy distribution. One of the where rand is a random number in [0, 1] interval and pa is the
most efficient and yet straightforward ways of applying Lévy flights is discovering probability. Existing eggs are replaced considering quality
to use the so-called Mantegna algorithm [48]. In Mantegna algorithm, by the newly generated ones from their current positions through
the step length S can be calculated by: random walks with step size such as:
u S = rand. (nests (randperm1(n),:) − nests (randperm2(n),:) )
S=
|v|1/ β (21) nest t + 1 = nest t + S. ∗P (25)
where β is a parameter between [1, 2] interval; u and v are drawn from where randperm1 and randperm2 are the random permutation functions.
the normal distribution as:

u~N (0, σu2), v ~N (0, σv2) (22) 2.3.3. Algorithm parameters

The CS parameters are as follows: The number of nests (n) is se-
1/ β
Γ(1 + β ) sin(πβ /2) ⎫ lected as 7, the step size parameter (α) is chosen as 0.1, and discovering
σu = ⎧ (β − 1)/2 ⎬
, σv = 1 probability (pa) is considered as 35%.
⎨
⎩ Γ[(1 + β )/2] β 2 ⎭ (23)

2.4. Particle Swarm Optimization

2.3.2.2. Alien eggs discovery. The alien eggs discovery is performed for
each component of each solution in terms of probability matrix as: Particle Swarm Optimization is one of the most popular algorithms

221
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 8. Convergence curves for the 693-bar double-layer barrel

vault.

Table 7
Comparative results for the 693-bar braced barrel vault structure.

Element Hasançebi Hasançebi et al. [40] Hasançebi & Kaveh et al. [38] Çarbaş & Present work
group & Çarbaş Kazemzadeh Azad Kazemzadeh
[39] [41] Azad [32]

ACO2 ACO HS GA BB-BC MBB-BC MCSS IMCSS ADS ABC CS PSO LPOA

1 ST 4 DEST 2 1/2 EST 3 1/2 EST 3 DEST 2 EST 3 1/ EST 3 EST 3 1/2 DEST 2 1/2 ST 4 ST 2 1/2 ST 3 EST 3 1/2
1/2 2
2 ST 1 ST 1 EST 3/4 ST 1 1/4 ST 1 ST 1 ST 1 ST 1 ST 1 ST 1 ST 1 EST 1 ST 1
3 ST 1 1/4 ST 1 ST 1 EST 1 ST 3/4 ST 3/4 EST 3/4 EST 1 ST 3/4 ST 1 1/4 EST 1 1/2 EST 1 1/2 EST 3/4
4 EST 1 1/4 ST 1 ST 1 ST 1 ST 1 1/4 ST 1 EST 1/2 ST 3/4 ST 1 ST 1 1/4 ST 1 1/2 ST 1 ST 1
5 ST 3/4 ST 1 EST 3/4 ST 3/4 ST 3/4 ST 3/4 EST 1/2 ST 1 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4
6 ST 5 ST 3/4 ST 4 EST 4 EST 3 1/ EST 3 1/ EST 3 DEST 2 EST 3 1/2 EST 1 1/2 ST 5 ST 2 1/2 ST 2 1/2
2 2
7 ST 1 1/4 EST 1 ST 1 1/4 EST 1 ST 1 ST 1 EST 1 1/4 ST 1 ST 1 EST 1 1/2 ST 1 ST 1 ST 1
8 EST 1 1/4 ST 3/4 ST 3/4 ST 3/4 ST 1 ST 1 ST 1 ST 1 1/4 ST 1 ST 1 1/2 ST 1 1/4 EST 2 EST 1
9 EST 3 1/2 ST 3 1/2 ST 3 1/2 DEST 2 ST 1 ST 1 ST 3/4 EST 1/2 ST 1 ST 1 ST 1 ST 1 ST 1
10 ST 1 1/4 DEST 2 1/2 ST 1 EST 1 ST 3/4 ST 3/4 EST 1/2 ST 1/2 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4
11 EST 3/4 ST 1 ST 1 1/4 ST 1 1/4 EST 2 1/ ST 3 EST 2 1/2 ST 3 DEST 2 EST 2 1/2 EST 1 1/2 EST 2 EST 2
2
12 ST 1 1/2 EST 1 EST 1 1/4 ST 1 1/2 ST 1 1/2 ST 1 1/2 EST 1 1/2 EST 1 1/4 ST 1 1/4 EST 1 ST 1 1/4 ST 1 1/4 EST 1
13 ST 1 1/2 EST 1 1/4 EST 1 1/2 EST 1 1/2 EST 1 1/ EST 1 1/ ST 2 1/2 EST 2 EST 1 1/2 ST 2 1/2 EST 2 DEST 2 EST 1 1/2
2 2
14 ST 1 1/4 EST 1 EST 1 1/4 ST 1 1/2 ST 1 ST 1 ST 3/4 ST 1/2 ST 1 ST 1 ST 1 ST 1 1/2 ST 1
15 EST 3/4 ST 3/4 ST 3/4 ST 1 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4
16 ST 1 1/2 EST 1 EST 1 1/4 ST 1 1/4 EST 1 1/ EST 1 1/ ST 1 1/4 EST 1 1/4 EST 1 1/2 ST 3/4 EST 3/4 ST 3/4 EST 1
2 4
17 EST 2 EST 1 1/4 EST 1 1/4 EST 1 1/2 ST 1 1/4 ST 1 1/4 ST 1 1/2 ST 1 1/2 EST 1 EST 1 1/4 ST 1 1/2 EST 1 EST 1
18 ST 1 1/4 ST 1 ST 1 1/4 ST 1 1/2 ST 2 1/2 ST 3 ST 3 ST 3 ST 2 1/2 EST 3 1/2 DEST 2 ST 3 1/2 DEST 2
19 ST 1 1/4 ST 1 1/4 EST 1 ST 1 1/4 ST 1 1/4 ST 1 EST 3/4 ST 3/4 ST 1 ST 1 1/4 ST 1 ST 1 ST 1 1/4
20 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 1/2 ST 3/4 ST 3/4 ST 3/4 ST 3/4 EST 3/4 ST 3/4
21 EST 2 1/2 ST 2 1/2 ST 2 1/2 ST 3 ST 1 ST 1 ST 1 1/4 ST 1 ST 1 ST 1/2 ST 3/4 ST 1/2 ST 3/4
22 ST 1 1/2 ST 1 1/4 ST 1 ST 1 1/4 ST 1 ST 3/4 EST 3/4 EST 1 ST 3/4 EST 1 ST 1 ST 3/4 ST 1
23 ST 3/4 ST 1 EST 1 EST 3/4 ST 3/4 ST 3/4 ST 3/4 EST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4 ST 3/4
Weight 12.13347 10.9992 11.23271 12.02949 10.85942 10.59533 10.81239 10.55086 10.56858 11.44762 10.69024 10.81718 9.6922
(kips)
Weight 5.50365 4.98915 5.09507 5.45648 4.92575 4.80596 4.90442 4.78581 4.79382 5.19255 4.84901 4.90659 4.39631
(tons)
Number of 50,000 – – – 27,150 27,150 14,300 9200 5000 15,000 15,000 15,000 8000
ana-
lyses

222
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Table 8 for self-weight minimization with LPOA, AB arts shown in PSO

Statistical results of different methods for the 693-bar braced barrel vault (lbs). methods. The Formex program is used for generating nodal coordinates
and connectivity matrix of the barrel vaults. The programming lan-
Algorithm Average Maximum Minimum Standard deviation
guage of this software is Formian which was described in details in
ABC 11,695.74 11,862.38 11,447.62 147.16 [52]. The design variables are the cross-sectional areas of the bar ele-
CS 11,384.41 11,552.82 10,690.24 372.70 ment which are selected from steel pipe sections from AISC-LRFD [43].
PSO 12,279.36 14,642.41 10,817.18 1830.23
These pipe sections are shown in Table 2. ST, EST, and DEST ab-
LPOA 11,262.00 11,491.90 9692.20 813.21
breviations stand for standard weight, extra strong, and double-extra
strong, respectively. Steel pipe sections are produced in three different
that originally accredited to Kennedy and Eberhart [13], inspired by the strength categories. The thickness of the steel pipe wall determines the
swarm behavior of fish and bird schooling in nature. The idea was to let category of the pipes. MATLAB software is used as a programming tool
the agents (birds) find food sources with social intelligence in the pre- for the analysis of the structures and performing the optimization
defined search space. Each agent was able to remembering its best procedure. It can be noted that the structures are analyzed using the
position and knowing the best position of the whole swarm. The ex- direct stiffness method. The penalty approach is employed for handling
tremum of the mathematical function to be optimized can be thought of the design constraints due to its simplicity. The design problem is
as the food source [49]. The modified version of the PSO that was in- outlined as follows:
troduced by Shi & Eberhart [50] is considered in this research. A brief find{X } = [x1,⋯, x nv ]
introduction to PSO is provided in the following. to minimize f subjected to design constraints (28)

2.4.1. Generate initial position of particles where, {X}is the vector of the optimum solution, nv is the number of
The first step is to generate random initial positions. design variables, and f denotes the penalized weight of a barrel vault as
defined by the following equation:
2.4.2. Main loop f = fpenalty × f ({X }) (29)
The updating formula for calculating the next position of a particle
ns
is introduced as:
f ({X }) = ∑ ρli Ai
vik, j+ 1 = wvik, j + c1 r1 (xbestik, j − x ik, j ) + c2 r2 (xgbest jk − x ik, j ) n=1 (30)
(26)
nc

x ik, j+ 1 = x ik, j + vik, j+ 1 (27) fpenalty (x ) = (1 + ε1 × v ) ε2 , v = ∑ vi

i=1 (31)
where xi,jk and vi,jk are the jth component of the ith particle's position
|λ |
and velocity respectively in the kth iteration; w is the inertia weight; vi = max ⎛ i∗ − 1, 0⎞
⎜ ⎟

xbesti is the best position of the ith particle; xgbest represents the best
position of the whole swarm; r1 and r2 represents two random numbers
in the interval [0,1]; and c1 and c2 are cognition parameter and social
behavior parameter, respectively.
2.4.3. Algorithm parameters
The PSO parameters are considered as the following: c1 and c2 were
set equal to 2; and inertia weight decreases linearly from 0.9 to 0.4
during the optimization process, as suggested by Ref. [51].
3. Optimal design of double-layer barrel vaults
Three large-scale double-layer barrel vault structures are considered
⎝ λi ⎠ (32)
where f({X}) is the self-weight of a structure; fpenalty represents the
penalty function; and ρ, Ai, and Li are the material density, the cross-
sectional area, and the length of the nth bar element, respectively; ns
presents the number of elements; νi is the value of violation of the ith
constraint; nc is the number of constraints; λi is the ith limited aspect of
the structure (e.g. stress, frequency or displacement) and λi⁎ is its upper
bound; And ε1 and ε2 are parameters that extremely penalize the un-
feasible solutions. The reason behind this is that if the variable ν is
greater than zero, then it is certain that the outcome of the particle is
unfeasible. As the purpose of the methodology which is to find the
minimum of the function f, when ν is greater than zero, fpenalty plunge
the penalized objective function, and with this measure, the

Fig. 9. Displacement ratios of the best-obtained solution with the LPOA in the 693-bar barrel vault design problem.

223
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 10. Strength ratios of the best-obtained solution with the LPOA in the 693-bar barrel vault design problem.

Fig. 11. Maximum strength and slenderness ratios of the best-obtained solution with the LPOA in the 693-bar barrel vault design problem for different group elements.

implemented algorithms select feasible ones systematically. In this the density of steel is equal to 0.288 lb per cubic inch (7833.413 kg/
study, ε1 is taken as unity, and ε2 starts from 1.5 and linearly increases m3). Strength and slenderness limitations are according to the AISC-
to 3 in the design problems. ASD provision [42]. Allowable tensile and compressive stresses are
depicted in Eqs. (32) and (33).

3.1. A 384-bar double-layer barrel vault +

⎧ σi = 0.6Fy for σi ≥ 0
⎨ −
⎩ σi for σi < 0 (33)
A spatial and stable 384-bar double-layer barrel vault is investigated
in this section. This discrete optimization problem was solved by Kaveh
& Moradveisi [3] using the Colliding bodies optimization (CBO) and ⎧⎡ ⎛1 − λi2 ⎞ F ⎤/ ⎛ 5 + 3λi λi3
y − ⎞ for λ < Cc
Enhanced Colliding bodies optimization approaches. These approaches ⎪ 2Cc2 ⎠ 8Cc3 ⎠
σi− = ⎣⎝ ⎦ ⎝3 8Cc

were compared with the outcomes of the method namely engineering ⎨ 12π 2E
⎪ 23λi2
for λ ≥ Cc
design. These solutions were found by CSI SAP2000 software using ⎩ (34)
auto-select. This structure consists of two rectangular nets, and for
making it stable, angles of the bottom nets are put into the center of one
of the above nets, and these are connected through diametrical ele- 2π 2E
Cc =
ments [38] as shown in Fig. 1. The span of the barrel vault is 24.82 m, Fy (35)
its rise is 5.12 m and its length is 26.67 m, and the depth that is the
distance between the top and bottom layers is equal to 1.35. This where E is the modulus of elasticity, Fy is the yield stress of steel, Cc is
structure consists of 111 pinned joints and 384 bar members and the the slenderness ratio dividing the elastic and inelastic buckling regions,
whole elements are grouped into 23 independent sizing variables which λi is the slenderness ratio (λi = kLi/ri) in which k is the effective length
are identified in Fig. 1(b). The structural material properties are as- factor, Li is the member length and ri is the radius of gyration. The
sumed as follows: The modulus of elasticity is considered as 30,450 ksi slenderness limitations that are according to AISC-ASD provision is
(210,000 MPa), the yield stress of steel is taken as 58 ksi (400 MPa), and described below:

224
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 12. 3D view (a), top plan view of quarter of the barrel vault
with the related group numbers (b), and flatten cross-sectional
view (c) of the 1536-bar double-layer barrel vault.

⎧ λ c = KLi / ri ≤ 200 for σi < 0
⎩ λt = KLi / ri ≤ 300
⎨ for σi > 0 (36)
The displacement constraints of ± 0.1969 in. (5 mm) are imposed
on all the nodes in x, y and z directions.
All connections are assumed as ball-jointed and top-layer joints are
subjected to concentrated vertical loads and supports are considered at
the two external edges of the top layer of the barrel vault. This design
problem is subjected to two types of loadings: Case 1 is a symmetric
loading condition to which the vertical concentrated loads of −20 kips
(−88.964 kN) are applied on free joints (non-support joints) of top
layer; in Case 2, which is asymmetric, the concentrated loads of
−10 kips (−44.482 kN) are applied at the right-hand half and at the
left-hand half of the structure the loads of −6 kips (−26.689 kN) are
applied on non-support top-layer joints, respectively.
Total populations of 30 particles are utilized for the 384-bar double-
layer barrel vault structural design example and the termination con-
dition is set to 30,000 number of function evaluations. Figs. 2 and 3
show the convergence curves obtained using these four utilized algo-
rithms LPOA, ABC, CS, and PSO. These figures associated with the first

225
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Table 9 Table 10
Comparative results for the 1536-bar braced barrel vault. Statistical results of different methods for the 1536-bar braced barrel vault (lbs).

Element group Present work Algorithm Average Maximum Minimum Standard deviation

ABC CS PSO LPOA ABC 167,500.58 171,919.14 160,930.61 4863.51

CS 170,331.45 176,335.97 161,125.95 2902.44
1 EST 6 ST 6 EST 6 DEST 4 PSO 147,812.176 178,172.41 133,331.04 15,645.08
2 DEST 2 1/2 EST 4 EST 3 EST 3 LPOA 135,156.90 165,199.48 125,665.32 11,208.05
3 ST 4 EST 3 1/2 EST 3 EST 3
4 EST 4 EST 4 ST 6 ST 6
5 EST 5 DEST 3 ST 6 DEST 3 and the second load cases, respectively. It can be seen from Figs. 2 and 3
6 DEST 3 EST 6 EST 6 DEST 3
7 EST 1/2 EST 3/4 EST 1/2 ST 1/2
that the convergence speed of the LPOA to find the better solutions is
8 ST 3 DEST 2 EST 1 1/2 EST 1 1/2 slightly lower than the others in almost 2000 first number of function
9 ST 3 1/2 ST 3 EST 2 1/2 ST 1 1/2 evaluations when under 7% of termination criterion was already
10 EST 2 DEST 2 ST 3 EST 1 1/2 reached. It can be interpreted that the LPOA optimization approach
11 ST 3 1/2 ST 2 1/2 ST 3 EST 2 1/2
performs the exploitation and exploration simultaneously in different
12 EST 3 1/2 ST 3 1/2 EST 2 ST 1 1/2
13 EST 2 ST 5 ST 3 1/2 EST 1 1/2 pride groups and consequently, the total population was obstructed
14 ST 5 ST 6 EST 3 1/2 EST 4 from moving into a small region. However, the other techniques con-
15 EST 5 EST 5 DEST 4 ST 6 tribute to a fast convergence speed as reducing the size of the search
16 ST 5 EST 4 ST 5 DEST 3 space. Moreover, according to both Figs. 2 and 3, CS algorithm has
17 ST 5 EST 3 1/2 EST 3 ST 5
worst behavior in the phase of the global search. Other techniques show
18 ST 3 1/2 ST 3 ST 3 EST 2
19 ST 3 1/2 EST 3 ST 3 1/2 EST 2 a comparable response in the first stages. Turning to exploitation stage,
20 EST 2 1/2 DEST 2 EST 1 1/2 ST 1 1/2 LPOA's search particles show a great behavior in such a way that they
21 DEST 2 1/2 ST 3 1/2 EST 2 ST 1 1/2 progressively find better solutions. It should be stated that LPOA finds
22 EST 2 EST 3 1/2 EST 2 EST 1 1/2
better answers where the best answer's penalized weight is closer to the
23 DEST 2 1/2 ST 5 EST 2 ST 1 1/2
24 ST 3 1/2 EST 2 ST 2 1/2 EST 1 1/2 final best solution of this effort in comparison with other techniques.
25 EST 2 1/2 EST 3 1/2 EST 2 EST 1 1/2 Having contemplated these statements, this algorithm has a robust
26 EST 3 DEST 2 ST 3 ST 4 behavior in the phase of probing local search. Finally, the figure reveals
27 ST 2 1/2 ST 1 1/2 ST 1 1/2 EST 1 1/2 that the answers of the present method are lower than those of others in
28 ST 1 1/2 EST 2 EST 1 1/2 ST 1 1/2
about 90% of the process.
29 ST 4 EST 2 EST 1 1/2 ST 2 1/2
30 EST 1 1/2 EST 2 1/2 ST 1 1/2 EST 1 1/2 Optimization results which are given in Tables 3 and 4 indicate that
31 EST 2 1/2 ST 4 EST 1 1/2 ST 2 1/2 the LPOA converged to best design in comparison to the other methods.
32 EST 1 1/2 EST 1 1/2 ST 1 1/2 EST 1 1/2 Additionally, the LPOA shows better performance in finding optimal
33 EST 2 1/2 ST 2 1/2 EST 2 EST 1 1/2
design solutions in local search phase and this fact leads to find better
34 EST 1 1/2 EST 2 1/2 EST 1 1/2 ST 1 1/2
35 EST 2 EST 2 EST 1 1/2 ST 1 1/2
answers.
Weight (kips) 158.936 161.126 133.331 125.665 However, Tables 5 and 6 contain the statistical results of the algo-
Weight (tons) 72.092 73.085 60.478 57.000 rithms over 20 independent runs for load Case 1 and load Case 2, re-
Number of analyses 15,000 15,000 15,000 15,000 spectively. From these tables, it can be observed that the search quality
of the LPOA is better than other techniques, and the average solution
found by LPOA is better than the best solution found by ABC, CS, and
PSO and other existing solutions. Moreover, according to the statistics,

Fig. 13. Convergence curves for the 1536-bar double-layer

barrel vault.

226
A. Kaveh, S. Mahjoubi Structures 13 (2018) 213–229

Fig. 14. Displacement ratios of the best-obtained solution with the LPOA in the 1536-bar barrel vault design problem.

Fig. 15. Strength ratios of the best-obtained solution with the LPOA in the 1536-bar barrel vault design problem.

Fig. 16. Maximum strength and slenderness ratios of the best-obtained solution with the LPOA in the 1536-bar barrel vault design problem for different group elements.

227
A. Kaveh, S. Mahjoubi
it can be observed that the solutions found by PSO have a higher
standard deviation and a higher maximum result which indicates lower
stability of this algorithm compared to the other techniques.
The dominance of the design constraints in controlling the final
optimized results of the best answer is also investigated. The nodal
displacement ratio, axial stress limitations of elements, and maximum
stress and the slenderness ratio of the groups are depicted in Figs. 4–6.
According to these figures, it can be observed that the joint displace-
ment constraints have an impact in the way the algorithm discovers
design solutions. It can be stated that the maximum amount of nodal
displacement ratios is somewhere in the vicinity of 0.9999. Further-
more, strength and slenderness limitations were highly effective in the
process of evaluating the optimization variables.
3.2. A 693-bar double-layer barrel vault
The optimal design of the braced 693-bar braced barrel vault was
first presented by Hasançebi & Carbas [39]. This roof structure is al-
ready built for roofing the platform shelters at the Thirumailai Railway
Station in India [2] which consists of 259 joints and 693 members with
23 independent design variables. The free span of the barrel vault is
19.03 m, its rise is 5.75 m and its length is 22.9 m. The geometry and
the member grouping scheme of the structure are shown in Fig. 7. It is
assumed that the barrel vault is loaded by uniformly distributed vertical
loads applied to the top of the roof and supports are considered at the
two external edges of the top and bottom layers. The structural material
properties are assumed as follows: The modulus of elasticity is taken as
29,000 ksi (203,893.6 MPa), the yield stress of steel is assumed as 36 ksi
(253.1 MPa), and the density of steel is 0.288 lb/in3 (7833.413 kg/m3).
The applied loads are considered as follows: a uniform dead load (DL)
pressure of 35 kg/m2; a positive wind load (WL) pressure of 160 kg/m2;
and a negative wind load (WL) pressure of 240 kg/m2. For design
purposes, these loads are combined as follows:
• Load Case 1: 1.5(DL + WL) = 1.5(35 + 160) = 292.5 kg/m
2
2
(2.87 kN/m )
• Load Case 2: 1.5(DL − WL) = 1.5(35 − 240) = −307.5 kg/m
2
2
(3.00 kN/m )
These two types of loads are applied along the vertical direction.
Stress and slenderness constraints are considered according to AISC-
ASD [12] as discussed earlier. The nodes are subjected to the dis-
placement limitations of ± 0.1 in. (0.254 cm) in every direction.
The problem of structural design was formerly investigated by
Hasançebi & Çarbaş using Ant Colony Optimization (ACO) method
[39], Hasançebi et al. [38] using Ant Colony Optimization (ACO), Ge-
netic Algorithm (GA) and Harmony Search (HS) optimization techni-
ques, Hasançebi & Kazemzadeh Azad [39] utilizing Big Bang-Big
Crunch (BB-BC) and Modified BB-BC (MBB-BC) algorithms, Kaveh et al.
[36] using Improved Magnetic Charged System Search (IMCSS) and
Magnetic Charged System Search (MCSS) methods, Çarbaş & Ka-
zemzadeh Azad [32] employing Adaptive Dimensional Search ap-
proach.
In this research, the population of search agents is taken as 20. In
the first attempt, 8000 numbers of objective function evaluations serve
as the termination criterion of the optimization process. However, ABC,
PSO and CS algorithms did not reach comparative solutions and for this
reason, the termination condition of these methods was increased to
15,000 numbers. The convergence curves of the best run of the opti-
mization approaches are provided in Fig. 8. As it can be seen from this
figure, the convergence rate of the LPO algorithm is much higher than
those of ABC, PSO and CS methods in all phases of the exploration and
exploitation, unlike the previous problem. It can be described that the
process of finding solutions in the early iterations is more efficient for
complex optimization problems.
Table 7 compares the optimal design solutions obtained in this
228
Structures 13 (2018) 213–229
research with the outcome of previous studies and the statistical data on
20 independent runs are provided in Table 8. The results show the
validity and effectiveness of the LPOA in comparison to other robust
algorithms. The statistics show that the PSO algorithm is not reliable
compared to the ABC, CS, and LPOA. As well as this, ABC has better
stability behavior among others, but the best solution found by this
method is the worst in comparison to those of LPOA, CS, and PSO.
Finally, the nodal displacement ratio, axial stress limitations of
elements, and maximum stress and the slenderness ratio of the best
design of the LPOA which are classified into the mentioned groups as
depicted in Figs. 9 to 11. The charts indicate that nodal displacement
ratios reach in the proximity of 1, specifically in the circumstances
defined in load case 2. Besides, Fig. 11 reveals that slenderness lim-
itations are more active than the strength limitations.
3.3. A 1536-bar double-layer barrel vault
Optimal design of a large-scale pin-jointed barrel vault is studied in
this section. This structure consists of 1536 bar elements and 413 joints
with a span of 40 m and a length of 50 m with 35 independent sizing
groups. The geometric details and member groups are presented in
Fig. 12. The properties of steel for the design are considered identical to
those of the first example. The modulus of elasticity is taken as
30,450 ksi (210,000 MPa), the yield stress of steel is equal as 36 ksi
(248.21 MPa), and the density of materials is considered to be 0.288 lb/
in3 (7833.413 kg/m3). Likewise, the design constraints (including stress
and slenderness limitation) are considered according to AISC-ASD [12]
are the same as noted in Section 3.1. The nodal displacements are
limited to ± 0.1969 in. (5 mm) in every direction. The supports are
fixed at the two external edges of the top layer of the braced barrel
vault and all joints of the top layer are subjected to concentrated ver-
tical loads of 5 kips.
In this design example, the population of search agents is set to 30
and the optimization process is terminated when the number of ob-
jective function evaluations reached at 15000. The comparative results
and the convergence curves obtained by the considered meta-heuristics
(PSO, CS, ABC and LPOA) are presented in Table 9 and Fig. 13, re-
spectively. It can be observed from Table 9 it can be seen that LPOA
found a better solution among the considered four techniques. Also, the
best try of Particle Swarm Optimizer converged to a better answer in
comparison to the ABC and CS algorithms.
For this design example, the algorithms were run 20 times and the
statistical results (maximum and average objective functions and
standard deviations) are provided in Table 10. It can be observed from
this table that LPOA and PSO show better performance on average, but
the maximum and corresponding standard deviation of the solutions
that were found with PSO algorithm is more than other optimization
techniques and it can be said that this method did not show reliable
performance. Furthermore, Artificial Bee Colony and Cuckoo Search
algorithms show good stability, but they were not able to find the best
results. It may be noted that the LPOA was found more diverse solutions
compared to the ABC and CS, and these two algorithms were not able to
reach better solutions.
Displacement ratios of the best solution of the LPOA outcomes have
been illustrated in Fig. 14. Moreover, axial stress limitations and the
maximum amount of strength and slenderness ratios which have been
classified in the defined groups are depicted in Figs. 15–16 respectively.
The charts illustrate that the design procedure which was based on
distinct constraints have been affected with nodal displacement and
slenderness limitations under multiple load cases. Notwithstanding,
strength constraints were not highly active in shifting the process.
4. Conclusions
In this paper, discrete optimization of double-layer barrel vaults is
performed using non-gradient optimization techniques, the Lion Pride
A. Kaveh, S. Mahjoubi
Optimization Algorithm (LPOA), Artificial Bee Colony (ABC), Cuckoo
Search (CS), and Particle Swarm Optimization (PSO). Three real-world
optimization design problems of double-layer barrel vaults are solved
with these techniques and the results are compared with each other and
those of the existing studies in the literature. The results obtained in-
dicate that the LPOA is computationally more efficient in searching for
optimal solutions and outperforms other robust algorithms. Moreover,
in accordance with the convergence curve, the trade-off between two
sub-processes, exploration and exploitation, of the algorithm leads to
have a favorable behavior in probing global and local areas.
Furthermore, it can claim that the average weight and the standard
deviation of the solutions obtained by the LPOA in a pre-defined in-
dependent optimization runs are lower in these design problems.
Therefore, it can be concluded that the Lion Pride Optimization
Algorithm is competitive with earlier studies and other methods con-
sidered in this research. Also, the convergence rate of this algorithm is
quite good for the investigated large-scale problems.
References
[1] Makowski ZS. Analysis, design and construction of braced barrel vaults. UK: CRC
Press; 1986.
[2] Ramaswamy D, Eekhout M. Analysis, design and construction of steel space frames.
Thomas Telford; 2002.
[3] Kaveh A, Moradveisi M. Optimal design of double-layer barrel vaults using CBO and
ECBO algorithms. Iran J Sci Technol Trans Civil Eng 2016;40:167–78.
[4] Nooshin H. Space structures and configuration processing. Prog Struct Eng Mater
1998;1:329–36.
[5] Lan TT. Space frame structures. Handbook of structural engineering. 24. Boca
Raton, FL: CRC Press; 2005. p. 1–50.
[6] Hare W, Nutini J, Tesfamariam S. A survey of non-gradient optimization methods in
structural engineering. Adv Eng Softw 2013;59:19–28.
[7] Saka MP. Optimum design of steel sway frames to BS5950 using harmony search
algorithm. J Constr Steel Res 2009;65:36–43.
[8] Horst R, Pardalos PM. Handbook of global optimization. Springer Science &
Business Media; 2013.
[9] Horst R, Tuy H. Global optimization: deterministic approaches. Springer Science &
Business Media; 2013.
[10] Holland JH. Adaptation in natural and artificial systems. An introductory analysis
with application to biology, control, and artificial intelligence. Ann Arbor, MI:
University of Michigan Press; 1975.
[11] Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by simulated annealing. Science
1983;220:671–80.
[12] Dorigo M, Di Caro G. Ant colony optimization: a new meta-heuristic. IEEE; 1999. p.
1470–7.
[13] Kennedy J, Eberhart R. Particle swarm optimization. vol. 4. 1995. p. 1942–8194.
[14] Geem ZW, Kim JH, Loganathan G. A new heuristic optimization algorithm: har-
mony search. Simulation 2001;76:60–8.
[15] Karaboga D. An idea based on honey bee swarm for numerical optimization.
Technical report-tr06. Erciyes university, engineering faculty, computer en-
gineering department; 2005.
[16] Yang X-S, Deb S. Cuckoo search via Lévy flights. IEEE; 2009. p. 210–4.
[17] Kaveh A, Talatahari S. A novel heuristic optimization method: charged system
search. Acta Mech 2010;13:267–89.
[18] Kaveh A, Mahdavi VR. Colliding bodies optimization: a novel meta-heuristic
method. Comput Struct 2014;139:18–27.
[19] Rao RV, Savsani VI, Vakharia D. Teaching–learning-based optimization: a novel
method for constrained mechanical design optimization problems. Comput Aided
Des 2011;43:303–15.
[20] Sadollah A, Bahreininejad A, Eskandar H, Hamdi M. Mine blast algorithm for op-
timization of truss structures with discrete variables. Comput Struct
2012;102:49–63.
[21] Mirjalili S. Dragonfly algorithm: a new meta-heuristic optimization technique for
solving single-objective, discrete, and multi-objective problems. Neural Comput
Applic 2016;27:1053–73.
[22] Liang Y-C, Cuevas Juarez JR. A novel metaheuristic for continuous optimization
problems: virus optimization algorithm. Eng Optim 2016;48:73–93.
229
Structures 13 (2018) 213–229
[23] de Melo VV, Banzhaf W. Drone squadron optimization: a novel self-adaptive algo-
rithm for global numerical optimization. Neural Comput Applic 2017:1–28. http://
dx.doi.org/10.1007/s00521-017-2881-3.
[24] Kaveh A, Ilchi Ghazaan M. Enhanced colliding bodies optimization for design
problems with continuous and discrete variables. Adv Eng Softw 2014;77:66–75.
[25] He Q, Wang L. A hybrid particle swarm optimization with a feasibility-based rule
for constrained optimization. Appl Math Comput 2007;186:1407–22.
[26] Cheng M-Y, Prayogo D, Wu Y-W, Lukito MM. A hybrid harmony search algorithm
for discrete sizing optimization of truss structure. Autom Constr 2016;69:21–33.
[27] Kaveh A, Zolghadr A. Truss optimization with natural frequency constraints using a
hybridized CSS-BBBC algorithm with trap recognition capability. Comput Struct
2012;102:14–27.
[28] Guo P, Wang X, Han Y. The enhanced genetic algorithms for the optimization de-
sign. IEEE; 2010. p. 2990–4.
[29] Baghlani A, Makiabadi M, Sarcheshmehpour M. Discrete optimum design of truss
structures by an improved firefly algorithm. Adv Struct Eng 2014;17:1517–30.
[30] Martinez-Martin FJ, Gonzalez-Vidosa F, Hospitaler A, Yepes V. Multi-objective
optimization design of bridge piers with hybrid heuristic algorithms. J Zhejiang
Univ Sci A 2012;13:420–32.
[31] Kaveh A, Talatahari S. Particle swarm optimizer, ant colony strategy and harmony
search scheme hybridized for optimization of truss structures. Comput Struct
2009;87:267–83.
[32] Hasançebi O, Kazemzadeh Azad S. Adaptive dimensional search: a new meta-
heuristic algorithm for discrete truss sizing optimization. Comput Struct
2015;154:1–16.
[33] Sharafi P, Hadi MNS, Teh LH. Optimum spans' lengths of multi-span reinforced
concrete beams under dynamic loading. In: Caicedo JM, Catbas FN, Cunha A, Racic
V, Reynolds P, Salyards K, editors. Topics on the dynamics of civil structures. vol. 1.
New York, NY: Springer New York; 2012. p. 353–61.
[34] Sharafi P, Hadi MNS, Teh L. Optimum column layout design of reinforced concrete
frames under wind loading. Topics on the dynamics of civil structures. vol. 1.
Springer; 2012. [9781461424123].
[35] Sharafi P, Hadi MNS, Lip H. Cost optimization of column layout design of reinforced
concrete buildings. Metaheuristic applications in structures and infrastructures. vol.
1. 2013. p. 129–46.
[36] Sharafi P, Teh LH, Hadi MNS. Shape optimization of thin-walled steel sections using
graph theory and ACO algorithm. J Constr Steel Res 2014;101:331–41.
[37] García-Segura T, Yepes V, Frangopol DM. Multi-objective design of post-tensioned
concrete road bridges using artificial neural networks. Struct Multidisip Optim
2017:1–12. http://dx.doi.org/10.1007/s00158-017-1653-0. (Published Online
2017).
[38] Kaveh A, Mirzaei B, Jafarvand A. Optimal design of double layer barrel vaults using
improved magnetic charged system search. Asian J Civil Eng (BHRC)
2014;15:135–54.
[39] Hasançebi O, Carbas S. Ant colony search method in practical structural optimi-
zation. Int J Optim Civil Eng 2011;1:91–105.
[40] Hasançebi O, Çarbas S, Saka MP. A reformulation of the ant colony optimization
algorithm for large scale structural optimization. Stirlingshire: Civil-Comp Press;
2011.
[41] Kaveh A, Mirzaei B, Jafarvand A. Shape-size optimization of single-layer barrel
vaults using improved magnetic charged system search. Int J Civil Eng
2014;12:447–65.
[42] Construction A. Manual of steel construction: allowable stress design. Chicago (IL):
American Institute of Steel Construction (AISC); 1989.
[43] Construction AIoS. Load & resistance factor design: manual of steel construction.
American Institute of Steel Construction; 1998.
[44] Kaveh A, Mahjoubi S. Lion pride optimization algorithm: a meta-heuristic method
for design optimization problems. Sci Iran 2017. [Accepted for publication].
[45] Stander PE. Cooperative hunting in lions: the role of the individual. Behav Ecol
Sociobiol 1992;29:445–54.
[46] Yazdani M, Jolai F. Lion Optimization Algorithm (LOA): a nature-inspired meta-
heuristic algorithm. J Comput Des Eng 2016;3:24–36.
[47] Kaveh A, Bakhshpoori T. Optimum design of space trusses using cuckoo search
algorithm with levy flights. Iran J Sci Technol Trans Civil Eng 2013;37:1–15.
[48] Mantegna RN. Fast, accurate algorithm for numerical simulation of Levy stable
stochastic processes. Phys Rev E 1994;49:4677.
[49] Kaveh A. Advances in metaheuristic algorithms for optimal design of structures.
2nd edition Switzerland: Springer International Publishing; 2017.
[50] Shi Y, Eberhart R. A modified particle swarm optimizer. IEEE; 1998. p. 69–73.
[51] Eberhart RC, Shi Y. Comparing inertia weights and constriction factors in particle
swarm optimization. IEEE; 2000. p. 84–8.
[52] Nooshin H. Formex configuration processing in structural engineering. 273. Essex,
UK: Elsevier Applied Science Publishers Ltd; 1984.