Sie sind auf Seite 1von 19

Engineering Structures 151 (2017) 802–820

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Optimal parameters and performance of artificial bee colony algorithm


for minimum cost design of reinforced concrete frames
Assanai Tapao, Raungrut Cheerarot ⇑
Concrete and Computer Research Unit, Civil Engineering, Faculty of Engineering, Mahasarakham University, Kantarawichai, Mahasarakham 44150, Thailand

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

Article history: This study investigates the effect of controlling parameters on the performance of an artificial bee colony
Received 5 May 2017 (ABC) algorithm in the optimum design of reinforced concrete frames under combination loads according
Revised 24 August 2017 to the ACI318-08 building code requirements for structural concrete. The objective function is the total
Accepted 28 August 2017
cost of reinforced concrete frames, which consists of concrete cost, reinforcing bars cost, and formwork
cost. The cross section of structure, diameter, and number of reinforcing bars are considered as the design
variables. The effect of the number of bees, food source quantities, trial limit, and stopping condition on
Keywords:
the cost design are studied and presented with statistical results. Three design examples are collected
Reinforced concrete frames
Cost optimization
from related literature to evaluating the performance of ABC algorithm. The results demonstrate that
Constraints optimization the number of trial limits is critical to the quality of food source, while the numbers of bees, food sources,
Artificial bee colony algorithm and trial limits impact the obtained optimum solution and usage time. The statistical results reveal that
when the food source quantities are lower than the number of bees, ABC algorithm provides high perfor-
mance for all the design examples.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction algorithm for the optimum design of simply supported reinforced


concrete (RC) beams according to ACI318-08 provision [22]. Coello
Artificial bee colony (ABC) algorithm was first proposed by Kar- et al. [23] and Chakrabarty [24] demonstrated ABC algorithm’s
aboga in 2005 [1]; this algorithm imitates the foraging behaviors of superior performance compared to a simple genetic algorithm. Jah-
a honey bee swarm in nature. It has been widely utilized for solv- jouh et al. [25] introduced the minimum cost design of RC contin-
ing unconstrained and constrained optimization problems in uous beams by using the ABC algorithm. In a more recent study,
numerous fields, such as some benchmark functions [2–5], com- Aydoğdu et al. [26] presented the ABC algorithm in conjunction
posite functions [6], data clustering [7], job scheduling [8] and with Levy flight distribution (LFABC) for scout bee [27] to design
numerical optimization [9,10]. In addition, a number of researchers 3D steel frames for real-world application as per LRFD-AISC design
have improved ABC algorithm to obtain a global optimum by using parameters. The obtained results demonstrated that the LFABC was
numerous techniques such as the Gbest-guided technique for solv- more robust and efficient than the standard ABC algorithms and
ing the numerical function problem [11], improved searching func- other compared algorithms [28].
tion [12,13], quick local search [14], and memory mechanism [15]. The performance of ABC algorithm is controlled by the four
Those results demonstrated that ABC and improved ABC algo- parameters, namely, number of bees, number of food sources,
rithms are more efficient and robust than the compared meta- number of trial limit, and stopping condition. It is challenging to
heuristic algorithms. select the appropriate value of these controlling parameters for
In recent years, the application of ABC algorithm in structural the minimum cost design of reinforced concrete frames. This study
engineering has been investigated by numerous researchers. Son- investigates the effect of these controlling parameters on the per-
mez [16,17] applied the ABC algorithm for obtaining the minimum formance of ABC algorithm in the optimum design of reinforced
weight of 2D or 3D steel trusses. The results demonstrated ABC concrete frames under combination loads according to the building
algorithm’s higher effectiveness with respect to the compared code requirements for structural concrete: ACI318-08 [22]. The
algorithms [18–20]. Ozturk et al. [21] presented the use of ABC data of the three RC frame designs with dissimilar lateral and ver-
tical loads are collected from the research paper of Kaveh and Sabzi
[29] to compare the performance of algorithms. The objective
⇑ Corresponding author. function of this study is the total cost of the RC frame. The design
E-mail address: raungrut.c@msu.ac.th (R. Cheerarot).

http://dx.doi.org/10.1016/j.engstruct.2017.08.059
0141-0296/Ó 2017 Elsevier Ltd. All rights reserved.
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 803

constraints of ACI318-08 are designated as the penalty functions. have width (b) less than height (h), which is expressed in the form
The optimal parameters and performance of ABC algorithm are of penalty function k1 as follows:
presented with statistical and the optimum solution results.
b h
k1 ¼  1 and  1 6 0 or b < h and h 6 3b ð3Þ
h 3b
2. The reinforced concrete frame design constraints
The bending moment capacity of RC beam (/Mn) is defined as
 a
In this research study, the finite element method is used to eval- /Mn ¼ /As f y d  ð4Þ
uate the internal forces of a RC frame, which consists of axial and 2
flexural forces. The rectangular cross sections of beams and col- where Mn is the nominal bending moment capacity of the RC beam,
umns are considered as the pattern of reinforcing bars in Fig. 1. / is the strength reduction factor for the RC beam which is defined
The RC beam is considered to be a singly reinforced beam, in which as 0.90, As is the total area of reinforcing bars, fy is the yield strength
As1 and As2 denote the resistances of a negative and positive bend- of reinforcing bars, d is the distance from the edge to the centroid of
ing moment, respectively (as illustrated in Fig. 1A). The RC column the reinforcing bars under consideration, and a is the depth of the
is considered as a short tied reinforced column which the longitu- equivalent rectangular compressive stress block, which can be cal-
dinal reinforcing bars in the column are held in position by sepa- culated as follows:
rate lateral ties (as illustrated in Fig. 1B). The upper and lower
bounds of the design variables in this study are listed in Table 1. As f y
a¼ 0 ð5Þ
From the engineering perspective, the RC frame optimization 0:85f c b
problem is to design an economically viable structure that is safe
and corresponds to the design code. Therefore, the objective func- The penalty function for evaluating the bending moment capac-
tion of this study is the frame cost function as with other similar ity is defined as
studies [30–37]; the objective function is expressed in Eq. (1). In jMu j
conjunction with the objective function, ACI318-08 design stan- k2 ¼  1 6 0 or /Mn > Mu ð6Þ
/Mn
dards are taken into consideration from the perspective of security.
where Mu is the applied ultimate bending moment.
X
n
The ACI code defines the minimum amount of reinforcing bars
F ¼ Min Li ðC c V i;c þ Sc V i;s þ W c V i;w Þ ð1Þ
i¼1
as

1:4
where F is the total cost of RC frame, Li is the length of the ith mem- As;min ¼ bd ð7Þ
ber, Cc is the unit cost of concrete, Vi,c is the quantity of concrete, Sc fy
is the unit cost of reinforcing bars, Vi,s is the total quantity of rein- The maximum amount of reinforcing bars is defined as
forcing bars, Wc is the unit cost of formwork, and Vi,w is the quantity
0
of formwork. Eq. (1) under the design constraints as follows: f c 600
As;max ¼ 0:75ð0:85b1 Þ bd ð8Þ
f y 600 þ f y
K ¼ fk1 ; k2 ; k3 ; . . . ; kn g 6 0 ð2Þ
where As,min is the minimum allowable total area for reinforcing
where K is the constraints function and k1, k2, k3, . . . , kn are the bars, As,max is the maximum allowable total area for reinforcing
design constraint of kth. Each constraint is represented by an bars, f0 c is the concrete compressive strength, and b1 is the factor
expression which must evaluate to less than zero. corresponding to the equivalent depth of rectangular compressive
Apart from satisfying the objective function, the design must stress block to the neutral axis depth and is defined as
conform to the ACI318-08 standard so that it can be used in actual
0
construction. This study also enhances all design constraints in the b1 ¼ 0:85 P 0:85  0:008ðf c  30Þ P 0:65 ð9Þ
form of the penalty functions, k1 to k13 defined in the next section.
The minimum number of reinforcing bars penalty function is
defined as
2.1. Beam formulation and design constraints
As;min
k3 ¼  1 6 0 or As P As;min ð10Þ
To compare the performance of the algorithm with the study by As
Kaveh and Sabzi [29], in this study, the RC beam is designed to and the maximum reinforcing bars penalty function is defined as
resist only the applied bending moments, while the vertical shear-
ing force and deflection are not considered. The beams generally As
k4 ¼  1 6 0 or As 6 As;max ð11Þ
As;max

b The penalty function for reinforcing bar spacing is defined as


b  0
40 b  ns ds  2d
k5 ¼    1 6 0 or 40 6 ð12Þ
0
bns ds 2d ns  1
ns 1
As1
where 40 is the minimum required distance between reinforcing
d d
h h bar, ns is the number of reinforcing bars in the arrangement, and
ds is the reinforcing bar diameter.
As
As2
2.2. Column formulation and constraints

(A) Beam (B) Column The strength capacity of a rectangular tied column takes into
consideration the interaction diagram illustrated in Fig. 2. In this
Fig. 1. Position of reinforcing bars in the cross section. study, the RC column is satisfactory when an applied axial force
804 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

Table 1
Upper and lower bound of design variables.

Variables Min Max Step size


Beam
b (mm) 200 1000 50
h (mm) 200 1000 50
Bar size Top (mm) 22 22 –
Bottom (mm) 19 19 –
No. of bar Top (mm) 2 12 2
Bottom (mm) 2 12 2
Column
b (mm) 200 1000 50
h (mm) 200 1000 50
Bar size (mm) 25 25 –
No. of bar 4 24 2

Axial force

Maximum axial force (0, Pmax)

( Mn,, Pn)
Pn = 0.80Pmax
Compression failure region
n
M
Pn
e

Balance condition Mb ( Mb, Pb)


eb Pb
Tension failure region

( M0, 0) Bending moment

Fig. 2. RC column interaction diagram.

0 0
and bending moment is in the diagram. Each point in the diagram /Pb ¼ /½0:85f c ab b þ A0s f s  As f y  ð19Þ
can be calculated as follows:
The maximum axial capacity (/Pmax) is calculated as follows: where ab is defined as
!
0
/Pmax ¼ /0:85f c ðAg  Ast Þ þ Ast f y ð13Þ 600
ab ¼ b1 d ð20Þ
600 þ f y
where / is the strength reduction factor for the RC column which is
defined as 0.65. and
The allowable axial capacity (/Pn) is calculated as follows:       
h ab h 0 h
0 /Mb ¼ / C bc  þ C bs  d þ Tb d  ð21Þ
/Pn ¼ /0:80½0:85f c ðAg  Ast Þ þ Ast f y  ð14Þ 2 2 2 2
and the allowable bending moment capacity (/Mn) is calculated as where Cbc is calculated using Eq. (22), Cbs is calculated using Eq.
follows: (23), and Tb is calculated using Eq. (24).
      
h a h h 0
/M n ¼ / C nc  þ C ns
0
 d þ Tn d  ð15Þ C bc ¼ 0:85f c ðab Þb ð22Þ
2 2 2 2
0
where Ag is the column gross section, Ast is the total area of reinforc- C bs ¼ A0s f s ð23Þ
ing bars, Cnc is calculated using Eq. (16), Cns is calculated using Eq.
(17), Tn is calculated using Eq. (18), and d0 is the distance from T b ¼ As f y ð24Þ
the edge to the centroid of the reinforcing bars under consideration. 0
where f s is defined as
0
C nc ¼ 0:85f c ðb1 cÞb ð16Þ  0 
0 cb  d
f s ¼ min Es ; f y ð25Þ
0 cb
C ns ¼ A0s ðf y  0:85f c Þ ð17Þ
where Cb is defined as
 
ðc  dÞ0:003 0 !
T n ¼ As Es  0:85f c ð18Þ 600
c Cb ¼ d ð26Þ
600 þ f y
0
where A s is the area of reinforcing bars at compression side, c is the
distance from extreme compression fiber to neutral axis, and Es is The bending moment capacity at zero axial force (/M0) can be
modulus of elasticity of reinforcing bars. defined as follows:
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 805

    
h a h dst
/M 0 ¼ / C nc  þ Tb d  ð27Þ k13 ¼  1 6 0 or dst 6 dsb ð35Þ
2 2 2 dsb

The penalty function for an axial force capacity can be defined where dst and dsb are the reinforcing bar diameters in the columns
as on the upper and lower levels, respectively.

Pu Pu Pu
k6 ¼  1 or  1 or 160 3. Artificial bee colony algorithm
/Pmax /P n /P b
or /Pmax > Pu or /Pn > Pu or /Pb > Pu ð28Þ ABC algorithm is inspired from the foraging behavior of honey
and the bending moment capacity can be defined as bees in nature [1]. A bee swarm consists of three groups of bees:
employed bees, onlooker bees, and scout bees. The foraging behav-
Mu Mu Mu ior begins with the employed bees, while the initial food sources
k7 ¼  1 or  1 or 160
/M n /Mb /M 0 are generated randomly in the search space of a problem. The
or /M n > M u or /Mb > M u or /M0 > Mu ð29Þ employed bees collect the food sources within their memory and
share the information with the onlooker bees in the hive. The
where Pu is the applied ultimate axial force, /Pmax is the maximum
new food sources are generated based on the information from
axial capacity, /Pn is the allowable maximum axial capacity, /Pb is
the employed bees. The food sources in the proximity of the hive
the axial capacity at the balance condition, /Mb is the bending
are collected by the onlooker bees and those at large distances
moment capacity at the balance condition, /M0 is the bending
from the hive by the scout bees. The onlooker and scout bees
moment capacity at the zero axial force capacity, and / is the load
return to the hive and dance to share their food source information
capacity reduction factor.
with the employed bees. If the new food source information is bet-
ACI code states that the total area of reinforcing bars (As) in
ter than the current food source information, the latter will be
compression member is to be more than 1% and less than 8% of
replaced. If the reverse is true, an employed bee will abandon
the gross section (Ag). The penalty function for the minimum num-
and transform itself into a scout bee based on the amount of trial
ber of reinforcing bars is defined as
limit.
0:01  Ag Initialization: the ABC algorithm begins by creating a bee pop-
k8 ¼  1 6 0 or As P 0:01  Ag ð30Þ ulation randomly distributed over the search space according to
As
a pre-defined number. The initial swarm is defined as a set of
and that for the maximum number of reinforcing bars is defined as employed bees (PN). Each bee contains various design variables
which are defined for each problem. Initial solution values (xij)
As are chosen randomly in the range of each design variable:
k9 ¼  1 6 0 or As 6 0:08  Ag ð31Þ
0:08  Ag
xij ¼ xmin
ij þ randð0; 1Þ  ðxmin
ij  xmax
ij Þ ð36Þ

2.3. Joint connection constraints where xij is the position of the jth design variable for the ith bee, xmin
ij
and xmax
ij are the lower and upper bounds of the jth design variable
In actual construction, it is not feasible for the width of a beam for the ith bee.
to be larger than that of a column or a column on an upper level to The food source contains one solution as same as a bee. But cre-
be larger than one at a lower level. Therefore, it is critical to obtain ation is different, the new food sources are created around the hive
a design solution that can be used in actual construction. by considering the positions of neighbors. A new possible food
The penalty function for the connection of beam and column is source is in the range of a current solution and neighbor. The
controlled as new food sources are generated by
 
k10 ¼ min
bb
;
bb
 1 6 0 or bb 6 minðbc;left ; bc;right Þ ð32Þ v ij ¼ xij þ /ij ðxij  xkj Þ ð37Þ
bc;left bc;right
where vij is the position of the jth design variable of the ith food
where bb is the beam width, and bc,left and bc,right are the width of the source, /min
ij is a random number in (1, 1) and xkj is the current
columns on the left and right sides, respectively. position of the jth design variable for the kth bee.
Thus, thee penalty function for defining the variation of column The employed bees communicate the information on food
size from a particular column to the next lower column is defined sources to the onlooker bees and dispatch them to the food source
as area. The food source once selected will not be selected again. The
bct hct onlooker bee can select food only once. Food sources with high fit-
k11 ¼  1 and  1 6 0 or bct 6 bcb and hct 6 hcb ð33Þ ness values are selected first. The suitability of the food source can
bcb hcb
be expressed as the following fitness function:
where bct is the column width on the upper level, bcb is the column
width on the lower level, hct is the column height on the upper level, Fit i
Pi ¼ PPN ð38Þ
and hcb is the column height on the lower level. n¼1 Fit n
The penalty function for defining the variation of the number of
where Pi is the selected probability of the ith food source, Fiti is the
reinforcing bars from a particular column to the next lower column
fitness value of the ith bee, and Fn is the fitness value of the nth bee.
is defined as
When an onlooker bee concludes that a new food source is more
nst suitable than the current food source, it immediately brings the
k12 ¼  1 6 0 or nst 6 nsb ð34Þ
nsb food back to its hive. However, if it is not successful in identifying
a better food source, the scout bee state is considered. Transform-
where nst and nsb are the numbers of reinforcing bars on the upper ing the onlooker bee to scout bee is based on the number of trial
and lower levels, respectively. limits (Ti). The onlooker bee will transform into a scout bee when
The penalty function for reinforcing bar diameter of a pair of it is unsuccessful in identifying a better new food source until an
upper and lower columns is defined as attempt is defined to identify a distant new food source in order
806 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

to increase the opportunities for discovering better new food Step 2 - FN new solutions are generated depending on the dif-
sources. The scout bee identifies the positions of new food sources ference between the previous solution and a neighbor. We call
using Eq. (36). a new solution, a food source, which contains the same set of
Information exchange: in the original concept, based on design variables as a bee.
observed natural behaviors, when the onlooker bees and scout Step 3 - For scout bees, we create new solutions as in Step 1.
bees bring food back to the hive, they will display signs that indi- Step 4 - The solutions from Step 2 through 3 will be evaluated
cate the position of the food. The round dances of onlooker bees by Pi.
indicate the food source location in the proximity of the hive. Step 5 - The new solutions will be selected depending on rand
The tail wagging dance of scout bees indicate the food sources at (0,1) and Pi. One bee can select only one solution and it cannot
larger distances from the hive. Considering the food source select another.
exchange of employed bees will be based on the quality of the Step 6 - Memorize recent food positions and report the best
acquired and the probability of their own decisions as food source
Step 7 - Repeat Steps 2 through 6 until the termination criteria
randð0; 1Þ P Pi ð39Þ is reached.

A conceptual flow of the ABC algorithm is illustrated in Fig. 3, 4. Applying the ABC algorithm to the optimum design of RC
the steps are: frames

Step 1 - Generate PN initial solutions randomly until the solu- The ABC algorithm for the optimum design of RC frames is illus-
tions are in accord with the ACI code. The cross section and trated in Fig. 4, and it can be explained as follows:
number of reinforcing bars are set randomly in the design vari-
able ranges. A solution corresponds to a bee; it contains the Step 1 - Before starting, the ABC algorithm parameters such as
design variables of beams and columns of RC frame. PN, FN, maximum of Ti, and termination criteria are defined.

Step 1: Send the employee bees onto the initial food sources by random design variables

from Ndv number of design variables :

For each i (1, PN )


For each j (1, Ndv )

xij xijmin rand ( 0 ,1 ) ( xijmin xijmax )


Next j

Send xij employee bee onto the initial food sources

Next i

Repeat

Step 2: Generating new food sources around their hive and send onlooker bees to

collect these foods:

For each i (1, FN )

For each j (1, Ndv )


vij xij ij ( xij xkj )

Next j
Next i

Step 3: Send scout bees to collect new food sources far away from the hive as

employee bees

Step 4: Employee bees evaluate the nectar of the collected new food by using
Fiti
Pi PN
Fit n
n 1
Step 5: The new foods will be chosen from Pi the nectar amounts: rand ( 0 ,1 ) Pi
Step 6: Memorize and report the best food source

Until termination criteria is reached

Fig.3. Pseudo code of general ABC algorithm.


A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 807

Start

Step 1 Input the ABC algorithm parameters


New food sources are established
Step 6
No Yes
It is a scout bee
Step 7 Step 8
Initial population
The new design variables are The new design variables are
Step 2 The design variables are random by Eq. (37) random by Eq. (36)
random by Eq. (36)

Evaluation of ACI318-08 provision and the


Step 9
The fitness is evaluated by the objective function
Step 3
objective function

Step 4 No The number of


No Step 10
It corresponds to new solutions are equal the food source
ACI318-08 provision limit

Yes
Yes
Step 5
No The populations Yes Information exchanging
correspond to ACI318-08
provision
Step 11 Pi is evaluated by Eq. (38)

No
Random(0,1) < Pi Step 12

The scout bee evaluation


Step 13 Yes

No
Ti is equal to Step 16 Replacement
the maximum Ti

Step 17
Step 15 Step 14 Yes All new No
solutions are evaluated
Ti is increased by 1 Becomes the
level scout bee
Yes
Step 18
No
According to the termination
criteria

Yes

Step 19 Reports the optimum solution

End

Fig. 4. The procedure of the ABC algorithm for the optimum design of RC frames.

Step 2 - The initial employed bee is generated by Eq. (36), in the provisions, it will be passed, and we go to Step 5. If they
which the characteristics of a bee correspond to the design vari- are not in agreement, we go back to Step 2.
ables of the RC frame members. Step 5 - Evaluating the number of initial employed bee: If the
Step 3 - The initial employed bee is evaluated by calculating the number of initial employed bee is equal to the maximum num-
load capacity of the RC frame members, and the fitness value is ber of bees, the generation of employed bees is complete, and
calculated by the objective function. we go to Step 6. Otherwise, we go back to Step 2.
Step 4 - ACI318-08 provisions are employed to evaluate the Step 6 - Evaluating a type of bee: In the case of scout bee type,
design variables of the initial employed bee. If the design we go to Step 8; in the other cases, we goes to Step 7.
variables of the initial employed bee are in agreement with Step 7 - Local search: A new solution is generated by Eq. (37),
and we go to Step 9.
808 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

Step 8 - Global search: A new solution is generated by Eq. (36), values is tested 10 times. In total, 5  4  10 = 200 runs are con-
and we go to Step 9. ducted for each design example.
Step 9 - A new solution is evaluated in terms of the load capac-
ity, provision, and objective function. 5.2. Number of trial limit adjustment
Step 10 - Evaluating the numbers of new solutions, if the num-
ber of new solutions is equal to the maximum food source limit, Number of trial limits (Ti) is a parameter that affects the behav-
we go to Step 11. In the other cases, we go to Step 6. ior of ABC. If Ti is low, it will motivate the onlooker bees to trans-
Step 11 - The information exchange probability of a new solu- form into scout bees in large numbers. If Ti is high, the scout bees
tion is evaluated by Eq. (38). are likely to be present in small numbers and are not likely to raise
Step 12 - A threshold value is generated as a random number all the problem-solving process. To illustrate the effects of Ti, 10
between 0 and 1, which is compared with the Pi value. If the independent runs for each of various values of Ti (5, 10, 15, 20,
Pi value is higher than the threshold value, we go to Step 16; 25, 50, 75, 100, 150, and 200 times) are conducted.
if the Pi value is less than the threshold value, we go to Step 13. Fig. 5 illustrates the relationship between the average cost of
Step 13 - Evaluating the number of trial limits, if the number of each bee population and each Ti value. It can be seen that when
trial limits is equal to the number of unimproved current solu- Ti increases the average cost has decreased for each the number
tions, we go to Step 14. Otherwise, we go to Step 15. of bees. However, when Ti is more than 100 times, it reveals that
Step 14 - An employed bee becomes a scout bee if it is not fea- the average values are either constant or have decreased negligi-
sible to determine a new solution that is an improvement over bly. Therefore, the use of Ti = 100 times is adequate for solving
the current solution. the optimum cost design of reinforced concrete frames.
Step 15 - The number of the unimproved current solution is
increased by a level.
5.3. Termination criterion adjustment
Step 16 - The information on current food sources is replaced by
the information on new food sources. The number of the unim-
The termination criterion is a critical parameter. This parameter
proved current solution is set as zero.
can be defined through various approaches depending on user
Step 17 - Evaluating the number of new solutions, if all the new
requirements. In this study, ABC algorithm ceased operating when
solutions have been evaluated, we go to Step 18, and for the
it could not determine a suitable solution within a specified num-
other cases, we go to Step 11.
ber of iterations (Rp). If the value of Rp is high, the time to deter-
Step 18 - Evaluating the termination criteria, if the termination
mine the optimum solution will increase. On the other hand, if
criteria are achieved, the optimal design process is terminated,
this value is low, it is challenging to determine the optimum solu-
and we go to Step 19. For the other cases, we go to Step 6.
tion. Design example 1 is selected as a case study for the optimum
Step 19 - Finally, the optimum solution is reported.
selection of Rp. The values of Rp were selected from 20 to 80 times,
with increments of 10 times, for the case study, and each value of
5. Parameter calibration Rp was tested 10 times.
Table 3 presents the results of the statistical tests when Rp was
The parameters required for ABC algorithm are the number of varied. When the value of Rp for each population increases, the
bees (PN), number of food sources (FN), number of trial limit (Ti), average costs are likely to reduce. It is noteworthy that the average
and termination criterion. These parameters control the effective- and standard deviation with Rp = 60–80 are better than those with
ness of ABC algorithm in directly obtaining the optimum solution. Rp = 20–50, except for the case of PN = 200. The use of Rp = 60 pro-
Selection of appropriate values is critical for obtaining the opti- vides the statistical tests close to that of Rp = 70 and 80 but it gives
mum solution in the minimum time. This study proposed a param- the lower average time. Therefore, this study selects Rp = 60 to stop
eter tuning method for the optimum design of RC frames. working of program for all number of bees. The relationship
between average time and repetition are illustrated in Fig. 6. Each
number of bees can be predicted from the function forms as
5.1. Number of bees and food sources ratio follows:

The number of bees (PN) and food sources (FN) are parameters For 100 individuals; Time ¼ 0:0018R2p þ 0:3665Rp
affecting the time and accuracy of ABC algorithm. The optimum 2
þ 0:0587 with R ¼ 0:9951 ð40Þ
proportion of the number of bees and food sources for optimum
structure size, will provide the less time spent. However, the most
For 200 individuals; Time ¼ 0:0046R2p þ 0:7982Rp
challenging task is the selection of appropriate values of these
2
parameters. This study has considered various numbers of bees  0:8264 with R ¼ 0:9986 ð41Þ
(PN; 50, 100, 200, 300, and 400 individuals) and various ratios of
FN:PN (0.5:1, 1:1, 1.5:1, and 2:1), as presented in Table 2, to deter- For 300 individuals; Time ¼ 0:0041R2p þ 0:9004Rp
mine the optimal ratio between FN and PN. These values are used 2
in all the frame design examples. Each combination of PN and FN þ 0:2111 with R ¼ 0:9990 ð42Þ

Table 2
Number of bees (PN) and food sources (FN).

Number of bees Number of food source (FN)


(PN) FN:PN = 0.5:1 1:1 1.5:1 2:1
50 25 50 75 100
100 50 100 150 200
200 100 200 300 400
300 150 300 450 600
400 200 400 600 800
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 809

23200 6. Design examples and numerical results


50 individuals
Data on three RC frame designs are collected from the paper by
100 individuals Kaveh and Sabzi [29], namely, three-bay four-story, three-bay
23000
200 individuals eight-story, and three-bay twelve-story. These example structures
were designed by Heuristic particle swarm ant colony optimization
300 individuals (HPSACO), and Heuristic big bang-big crunch (HBB-BC). The f0 c and
22800
400 individuals fy are defined as 23.5 MPa and 392 MPa, respectively. A non-
factored uniform dead load of D = 22.3 kN/m and live load of
Average cost ($)

L = 10.7 kN/m were applied to every beam in each frame. The con-
22600 crete unit weight was set as 2323 kg/m3 and the steel unit weight
to 7849 kg/m3. They formed beams and columns cross-section as
Stability the databases following ACI318-08 [22]. The unit cost of concrete,
22400 reinforcing bars, and formwork are USD 105 m3, USD 0.9 kg1,
and USD 92 m2, respectively. The load combination according to
ACI318-08 [22] is considered as
22200
U ¼ 1:2D þ 1:6L ð44Þ

22000
U ¼ 1:2D þ 1:0L þ 1:4E ð45Þ
0 50 100 150 200 250
Numbers of trial limit (Ti) U ¼ 0:9D þ 1:4E ð46Þ

Fig. 5. The trends of Ti for design example 1. where D is the uniform dead load, L is the uniform live load, and E is
the lateral earthquake load, respectively.
ABC algorithm is operated using a Microsoft Windows 7, 64-bit
For 400 individuals; Time ¼ 0:0002R3p  0:0368R2p operating system, 8 GB DDR3 of RAM, and an Intel core i7-4500U
2 1.80 GHz CPU; it is coded using Microsoft Visual Basic 6. The statis-
þ 2:6007Rp  12:84 with R ¼ 0:9988:
tical results of each design example demonstrate the efficiency of
ð43Þ ABC algorithm.

Table 3
Tuning Rp for ABC algorithm.

Rp (Times) Cost (USD) Average time (s)


Min Max Average Median Mode Standard deviation
For 100 individuals
20 22,144 25,399 23,443 23,180 – 1120.83 6.89
30 22,144 24,635 22,699 22,353 22,353 758.49 9.26
40 22,144 23,025 22,348 22,262 22,171 273.57 11.31
50 22,144 23,633 22,343 22,171 22,144 460.91 14.11
60 22,144 22,353 22,170 22,144 22,144 65.36 15.69
70 22,144 22,353 22,173 22,144 22,144 64.70 16.91
80 22,144 22,353 22,167 22,144 22,144 65.89 17.53
For 200 individuals
20 22,144 24,878 23,045 22,826 22,380 876.02 13.52
30 22,144 24,025 22,666 22,445 22,144 663.02 18.47
40 22,144 24,080 22,397 22,144 22,144 617.81 24.15
50 22,144 22,171 22,146 22,144 22,144 8.56 27.74
60 22,144 22,171 22,146 22,144 22,144 8.56 30.49
70 22,144 22,171 22,146 22,144 22,144 8.56 32.67
80 22,144 22,171 22,146 22,144 22,144 8.56 33.78
For 300 individuals
20 22,144 24,855 23,139 22,844 22,144 1010.61 18.40
30 22,144 23,085 22,419 22,248 22,144 344.36 26.05
40 22,144 23,822 22,524 22,353 22,144 544.15 31.46
50 22,144 22,171 22,149 22,144 22,144 11.42 36.53
60 22,144 22,144 22,144 22,144 22,144 0.00 41.77
70 22,144 22,144 22,144 22,144 22,144 0.00 45.39
80 22,144 22,144 22,144 22,144 22,144 0.00 47.88
For400 individuals
20 22,144 23,597 22,737 22,613 22,144 483.66 25.94
30 22,144 22,900 22,424 22,367 22,353 263.08 37.79
40 22,144 22,996 22,276 22,144 22,144 269.04 44.86
50 22,144 22,171 22,149 22,144 22,144 11.42 50.44
60 22,144 22,144 22,144 22,144 22,144 0.00 53.21
70 22,144 22,144 22,144 22,144 22,144 0.00 57.98
80 22,144 22,144 22,144 22,144 22,144 0.00 61.65
810 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

100 individuals 200 individuals and the number of food sources is up to 200. The relationship
between the number of bees and average costs is illustrated in
300 individuals 400 individuals
Fig. 10A. It demonstrates that when the numbers of bees are
70
increased, the average costs are decreased and approximately con-
verge at PN = 200 individuals. Fig. 10B is the relationship between
60 the number of bees and average time, wherein it is determined
Average time (s)

50 that the average time depends on the number of bees and food
sources. The higher the bee population and number of food
40
sources, the more the time required to determine the minimum
30 cost. From the result, it can be concluded that the optimum param-
eter values for this design example are PN = 300 at FN:PN = 0.5:1 as
20
it requires the lowest time.
10 The comparison of the optimum design variables among
0 HPSACO, HBB-BC, and ABC algorithms are presented in Table 9. It
0 20 40 60 80 100 is observed that ABC algorithm yields the lowest minimum cost
Number of repetitions among the three at USD 22,144; this is less than that of HPSACO
and HBB-BC algorithms (USD 22,207 in both these cases) by USD
Fig. 6. The relationship between average time and number of repetition.
63. Although the variation in minimum cost of each algorithm is
marginal, but ABC algorithm can be reached to the minimum cost
less than HPSACO and HBB-BC algorithms. The economic part of
6.1. Design example 1 ABC algorithm come from the beam cross section of B2 group is
smaller than that of the HPSACO and HBB-BC algorithm which it
Design example 1 is a small size building, which is illustrated demonstrates that ABC algorithm can finds the different cross sec-
in Fig. 7. This frame is the model for a small hotel, resort, or tion of beam and column more than the HPSACO and HBB-BC
dormitory. The frame height is 13.20 m, and the total bay length algorithm.
is 22.5 m. The width of each bay is 7.5 m while each story has a
height of 3.3 m. The total number of members is 28, consisting of
12 beams and 16 columns. Fig. 7A illustrates the various lateral 6.2. Design example 2
loads, uniform dead load and live load on each story. The design
members are divided into two groups for beam and two groups Design example 2 is a medium-size building and is illustrated in
for columns, as illustrated in Fig. 7B. Each column group is Fig. 11. This frame is a model suitable for a condominium, medium
divided into four stories while each beam group is divided every size hotel, or medium size dormitory. The frame height is 26.40 m.
two stories. The total number of possible solutions in the search The bay width and story height are the same as in design example
space is 12  16  16  5  5 + 16  16  16  10 = 117,760 1. The various lateral loads, uniform dead load and live load
solutions. applied on each story of the frame are illustrated in Fig. 11A. The
ABC algorithm is one of the swarm intelligence algorithms total number of members is 56, consisting of 24 beams and 32 col-
which it is well-known that it has a fast convergence rate when umns. These members are divided into seven groups—three groups
the number of populations is increased. Fig. 8 illustrates general for beams and four groups for columns, illustrated in Fig. 11B. The
convergence histories for each number of bees when ABC algo- total number of possible solutions in the search space is
rithm was employed in a design example 1 (FN:PN = 1:1). The con- 24  16  16  5  5 + 32  16  16  10 = 235,520 solutions.
vergence rate of the 400 individuals is the highest while the The optimum cost is denoted as a bold number in Tables 10–13;
convergence rate of the 50 individuals is the lowest. Fig. 9 illus- the optimal cost is USD 47,342. The statistical results demonstrate
trates the distribution of each bee belonging to group of 400 indi- that the average and standard deviation of costs determined using
viduals, from the initial to the final position. It is determined that ABC algorithm decreases continuously when the number of bees
the average of the bee swarm in the final position is USD 26,466. and the proportion of food sources increase. Fig. 12 illustrates
The total number of solutions is below average and equal to 256 the convergence histories of each value of number of bees for
individuals, which comprises 64% of all the bees. When compared design example 2 (at FN:PN = 1:1). It is determined that the con-
with BB-BC algorithm [35], it is determined that the spread of indi- vergence rate of the higher population is higher than that of the
viduals of the population of ABC algorithm is better than that of the lower population; this trend is similar to that of example 1.
BB-BC algorithm, which demonstrates that ABC algorithm is highly Table 14 presents the accuracy (%) of ABC algorithm for design
efficient for optimal design of reinforced concrete frames. How- example 2; it is determined that for PN = 50, 100, and 200 individ-
ever, the use of the higher number of population will necessitate uals, accuracy increases with increase in FN. The performance of
a larger amount of time for calculation, and if the selected stop ABC algorithm is 100% for PN = 200 at FN:PN = 2:1 and PN = 300
condition is inappropriate, the received cost will not be the lowest. and 400 at all FN:PN. However, the optimum parameter values
The statistical results with regard to the adjusting ratio of the for this design example are PN = 300 individuals and FN:
number of bees and food sources for design example 1 is presented PN = 0.5:1 as these values yield performance of 100% in the least
in Tables 4–7, in which the optimal cost of the design example is amount of time, a criterion similar to the one used for design
denoted in bold number (USD 22,144). Table 8 summarizes the example 1.
accuracy of ABC algorithm, which is calculated from the percentage Fig. 13A reveals a correlation between the average and number
of the number of times optimal cost appeared out of the 10 trials. It of bees for each FN:PN of design example 2. The average costs at
is determined that the accuracy of ABC algorithm depends on the PN = 200 and 300 for FN:PN = 1:1, 1.5:1, and 2:1 are approximately
FN:PN ratio, wherein ABC algorithm has 100% efficiency when convergent, and they converge to the minimum cost at PN = 300
PN = 300 and 400 individuals for all values of FN as well as the case for all FN:PN. An observation of Fig. 13B reveals that the usage time
of PN = 200 and FN:PN = 2:1. The accuracy of ABC algorithm is at PN = 300 for FN:PN = 0.5:1 is approximately equal to that at
higher than 70% when the number of bees is up to 200 individuals PN = 100 for FN:PN = 1.5:1; however, the accuracy at the former
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 811

D = 22.3 kN/m + L = 10.7 kN/m


12.4 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


9.3 kN B2 B2 B2

4 @ 3.3m = 13.20 m

4 @ 3.3m = 13.20 m
D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1
6.2 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


3.1 kN B1 B1 B1

C1 C2 C2 C1

7.5 m 7.5 m 7.5 m 7.5 m 7.5 m 7.5 m

(A) Applied loads (B) Design group


Fig. 7. Topology and lateral loads for design example 1.

40000
38000 50 individuals
36000 100 individuals
34000 200 individuals
Cost ($)

32000 300 individuals


30000 400 individuals
28000
26000
24000
22000
20000
0 20 40 60 80 100 120 140 160 180 200
Number of iterations

Fig. 8. Convergence histories of each the number of bees for design example 1 (FN:PN = 1:1).

Initial cost position Final cost position


62000

57000

52000

47000
Cost ($)

42000

37000

32000
Average
27000

22000
0 100 200 300 400 500
Number of bees

Fig. 9. Initial position to the final position of the costs (FN:PN = 1:1 and FN = 400).

condition is 100%, while that at the latter is 60%. This demonstrates 6.3. Design example 3
that the search space is wider when the number of bees is more
than that of the food sources. Table 15 presents the optimum Design example 3 is a tall building, which is illustrated in
values of the design variables for design example 2; here, the Fig. 14. The frame’s height is 39.6 m. The bay width and story
minimum cost for ABC algorithm is less than those for HPSACO height are the same as in previous example. Fig. 14A illustrates
and HBB-BC by 2.42% and 1.91%, respectively. the combination both the applied lateral and vertical loads. The
812 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

Table 4
Statistical results of design example 1 for the case of FN:PN = 0.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 22,171 283 5.27 22,171 332 11.72 22,171 220 15.31 22,144 330 31.91 22,144 270 33.69
2 22,171 245 5.08 22,353 276 9.70 22,171 286 19.59 22,144 218 20.42 22,144 219 27.83
3 22,353 257 6.16 22,508 194 7.41 22,144 175 11.46 22,144 254 24.62 22,144 227 28.82
4 22,481 261 6.92 22,144 284 10.14 22,144 293 19.60 22,144 270 25.40 22,144 256 32.50
5 22,171 270 7.63 22,171 243 8.55 22,144 318 21.19 22,144 272 26.50 22,144 229 28.60
6 22,862 199 5.43 22,171 176 6.49 22,144 214 14.03 22,144 271 25.46 22,144 288 36.20
7 22,717 225 5.18 22,353 161 5.39 22,171 215 14.41 22,144 252 24.76 22,144 284 35.84
8 22,353 343 7.73 22,481 223 8.44 22,144 258 17.34 22,144 221 21.13 22,144 264 32.68
9 23,015 215 5.00 22,171 279 9.85 22,144 271 17.35 22,144 283 27.32 22,144 278 35.60
10 22,144 223 4.76 22,144 211 7.11 22,144 265 17.08 22,144 249 24.05 22,144 253 32.01
Min 22,144 199 4.76 22,144 161 5.39 22,144 175 11.46 22,144 218 20.42 22,144 219 27.83
Max 23,015 343 7.73 22,508 332 11.72 22,171 318 21.19 22,144 330 31.91 22,144 288 36.20
Average 22,444 252 5.91 22,267 238 8.48 22,152 252 16.74 22,144 262 25.16 22,144 257 32.38
Med 22,353 251 5.35 22,171 233 8.49 22,144 262 17.21 22,144 262 25.08 22,144 260 32.59
Mode 22,171 – – 22,171 – – 22,144 – – 22,144 – – 22,144 – –
SD 318 41 1.12 144 55 1.92 13 44 2.98 0 32 3.21 0 25 3.10

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 5
Statistical results of design example 1 for the case of FN:PN = 1:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 22,171 218 6.65 22,536 155 9.84 22,144 225 27.88 22,144 223 39.86 22,144 287 67.85
2 22,144 299 10.25 22,144 256 16.16 22,144 272 32.85 22,144 223 39.69 22,144 243 57.23
3 22,380 216 7.92 22,171 286 17.84 22,144 323 38.45 22,144 150 26.62 22,144 210 47.95
4 22,144 343 13.33 22,144 245 16.07 22,144 208 25.14 22,144 269 47.85 22,144 181 41.08
5 22,380 200 8.52 22,481 203 12.37 22,144 253 30.92 22,144 244 43.69 22,144 203 47.21
6 22,144 296 9.79 22,144 239 15.05 22,144 248 30.25 22,144 186 33.35 22,144 250 58.75
7 22,673 224 8.01 22,144 227 14.10 22,171 234 28.50 22,144 228 41.15 22,144 204 47.68
8 22,144 285 10.22 22,144 285 17.63 22,144 261 31.89 22,144 195 34.12 22,144 201 46.91
9 22,144 278 9.00 22,144 302 19.78 22,144 234 27.92 22,144 267 48.12 22,144 225 52.88
10 22,380 207 7.60 22,144 274 17.41 22,144 257 31.12 22,144 239 42.16 22,144 278 64.51
Min 22,144 200 6.65 22,144 155 9.84 22,144 208 25.14 22,144 150 26.62 22,144 181 41.08
Max 22,673 343 13.33 22,536 302 19.78 22,171 323 38.45 22,144 269 48.12 22,144 287 67.85
Average 22,270 257 9.13 22,219 247 15.63 22,146 252 30.49 22,144 222 39.66 22,144 228 53.21
Med 22,157 251 8.76 22,144 251 16.11 22,144 251 30.58 22,144 226 40.51 22,144 218 50.41
Mode 22,144 – – 22,144 – – 22,144 234 – 22,144 223 – 22,144 – –
SD 179 49 1.89 153 44 2.92 9 31 3.61 0 37 6.70 0 35 8.63

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 6
Statistical results of design example 1 for the case of FN:PN = 1.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 22,171 273 12.36 22,144 242 22.83 22,144 248 45.22 22,144 260 70.34 22,144 230 81.23
2 22,171 240 12.11 22,171 249 22.78 22,144 241 42.99 22,144 228 60.41 22,144 175 60.86
3 22,144 300 15.23 22,144 232 21.42 22,144 257 46.34 22,144 269 72.30 22,144 277 99.55
4 22,144 225 12.20 22,144 246 22.86 22,144 229 40.83 22,144 264 70.75 22,144 213 75.46
5 22,171 257 13.69 22,353 149 13.26 22,144 184 32.48 22,144 211 55.43 22,144 259 92.33
6 22,353 271 14.81 22,171 186 16.68 22,144 229 41.22 22,144 197 51.63 22,144 185 64.90
7 22,144 188 9.69 22,144 267 24.97 22,144 248 44.53 22,144 171 45.20 22,144 243 86.72
8 22,144 218 10.70 22,144 248 23.18 22,171 205 36.28 22,144 223 58.88 22,144 269 95.86
9 22,585 165 8.00 22,144 224 20.82 22,144 261 46.91 22,144 221 58.27 22,144 234 83.24
10 22,171 188 9.08 22,171 202 17.88 22,144 259 46.45 22,144 222 59.38 22,144 152 52.42
Min 22,144 165 8.00 22,144 149 13.26 22,144 184 32.48 22,144 171 45.20 22,144 152 52.42
Max 22,585 300 15.23 22,353 267 24.97 22,171 261 46.91 22,144 269 72.30 22,144 277 99.55
Average 22,220 233 11.79 22,173 225 20.67 22,146 236 42.33 22,144 227 60.26 22,144 224 79.26
Med 22,171 233 12.16 22,144 237 22.10 22,144 245 43.76 22,144 223 59.13 22,144 232 82.24
Mode 22,171 188 – 22,144 – – 22,144 248 – 22,144 – – 22,144 – –
SD 143 44 2.41 65 36 3.62 9 25 4.77 0 31 8.74 0 42 15.68

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 813

Table 7
Statistical results of design example 1 for the case of FN:PN = 2:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 22,353 238 15.12 22,144 217 25.59 22,144 285 66.57 22,144 219 77.60 22,144 196 91.15
2 22,171 255 15.82 22,144 296 36.18 22,144 198 45.84 22,144 237 84.25 22,144 216 101.64
3 22,327 228 14.83 22,144 273 33.59 22,144 284 66.46 22,144 225 79.73 22,144 182 84.94
4 22,144 292 19.22 22,144 247 29.66 22,144 239 55.31 22,144 248 87.73 22,144 219 103.21
5 22,144 317 20.39 22,144 242 28.97 22,144 230 52.90 22,144 274 97.88 22,144 252 119.28
6 22,144 293 18.68 22,144 264 31.36 22,144 232 54.68 22,144 174 60.28 22,144 243 115.10
7 22,144 245 15.98 22,144 251 29.74 22,144 275 63.27 22,144 257 92.30 22,144 242 114.17
8 22,144 249 16.17 22,144 290 35.41 22,144 243 56.66 22,144 158 54.64 22,144 253 119.14
9 22,171 259 16.23 22,171 216 25.56 22,144 208 48.28 22,144 223 79.33 22,144 208 98.12
10 22,171 237 14.80 22,144 264 32.17 22,144 244 58.09 22,144 226 79.51 22,144 227 106.21
Min 22,144 228 14.80 22,144 216 25.56 22,144 198 45.84 22,144 158 54.64 22,144 182 84.94
Max 22,353 317 20.39 22,171 296 36.18 22,144 285 66.57 22,144 274 97.88 22,144 253 119.28
Average 22,191 261 16.72 22,146 256 30.82 22,144 244 56.81 22,144 224 79.33 22,144 224 105.30
Med 22,157 252 16.08 22,144 258 30.55 22,144 241 55.98 22,144 226 79.62 22,144 223 104.71
Mode 22,144 – – 22,144 264 – 22,144 – – 22,144 – – 22,144 – –
SD 80 29 1.98 9 27 3.66 0 30 7.05 0 35 13.26 0 24 11.77

FN = Food source, PN = Population size, and Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.
Cost is USD and time is seconds.

Table 8
Accuracy of ABC algorithm for design example 1.

Number of bees FN:PN


(PN) 0.5:1 1:1 1.5:1 2:1
50 10% 50% 40% 50%
100 20% 70% 60% 90%
200 70% 90% 90% 100%
300 100% 100% 100% 100%
400 100% 100% 100% 100%

22500 FN:PN = 0.5:1 FN:PN = 1:1


22450 FN:PN = 0.5:1 FN:PN = 1.5:1 FN:PN = 2:1
FN:PN = 1:1 120
22400
FN:PN = 1.5:1
Average cost ($)

100
22350
Average time (s)

FN:PN = 2:1
80
22300

22250 60

22200 40

22150 20

22100 0
0 100 200 300 400 500 0 100 200 300 400 500
Number of bees (PN) Number of bees (PN)
(A) (B)
Fig. 10. Influence of the population sizes and food sources for design example 1.

Table 9
Comparison of the optimum design variables for design example 1.

HPSACO [29] HBB-BC [29] ABC algorithm


Group b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS
B1 300 500 3-D19/5-D22 300 500 3-D19/5-D22 300 500 3-D19/5-D22
B2 300 500 4-D19/5-D22 300 500 4-D19/5-D22 250 550 3-D19/4-D22
C1 350 350 8-D25 350 350 8-D25 300 300 10-D25
C2 300 300 6-D25 300 300 6-D25 350 350 4-D25
Cost (USD) 22,207 22,207 22,144

b = section width, h = section height, BS = bottom position of reinforcing bars, and TS = top position of reinforcing bars.
814 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

D = 22.3 kN/m + L = 10.7 kN/m


28 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


24.5 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


21 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


17.5 kN B2 B2 B2

8 @ 3.3m = 26.40 m

8 @ 3.3m = 26.40 m
D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3
14 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


10.5 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


7 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


3.5 kN B1 B1 B1

C1 C2 C2 C1

7.5 m 7.5 m 7.5 m 7.5 m 7.5 m 7.5 m

(A) Applied loads (B) Design group


Fig. 11. Topology and lateral loads for design example 2.

Table 10
Statistical results of design example 2 for the case of FN:PN = 0.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 47,342 391 11.96 47,401 238 14.32 47,342 315 35.65 47,342 302 49.19 47,342 295 63.64
2 47,767 261 8.57 48,205 241 14.33 47,342 304 33.23 47,342 307 49.78 47,342 271 58.93
3 47,719 251 9.19 47,342 271 15.74 47,401 282 30.86 47,342 270 44.76 47,342 219 46.57
4 47,342 322 12.05 47,401 276 15.78 47,342 344 38.23 47,342 328 54.13 47,342 326 70.17
5 47,401 274 10.37 47,342 274 15.83 47,342 278 31.52 47,342 225 37.08 47,342 281 60.14
6 47,584 279 9.01 47,401 302 17.64 47,906 183 20.36 47,342 268 44.59 47,342 394 84.47
7 47,832 268 9.09 48,133 317 18.23 47,342 300 33.46 47,342 211 34.16 47,342 269 58.54
8 47,342 351 10.98 47,342 265 16.06 47,401 313 35.01 47,342 229 38.34 47,342 306 66.09
9 48,613 225 6.88 47,342 313 18.67 47,342 332 37.15 47,342 341 55.80 47,342 294 64.27
10 47,342 252 8.34 47,401 286 16.68 47,401 268 30.16 47,342 283 45.89 47,342 346 74.84
Min 47,342 225 6.88 47,342 238 14.32 47,342 183 20.36 47,342 211 34.16 47,342 219 46.57
Max 48,613 391 12.05 48,205 317 18.67 47,906 344 38.23 47,342 341 55.80 47,342 394 84.47
Average 47,628 287 9.64 47,531 278 16.33 47,416 292 32.56 47,342 276 45.37 47,342 300 64.77
Med 47,492 271 9.14 47,401 275 15.95 47,342 302 33.35 47,342 277 45.32 47,342 295 63.96
Mode 47,342 – – 47,401 – – 47,342 – – 47,342 – – 47,342 – –
SD 397 51 1.66 338 27 1.49 174 45 5.04 0 44 7.18 0 48 10.26

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

total number of members is 86, consisting of 36 beams and 48 col- the accuracy of ABC algorithm; it reveals that for number of food
umns. The design members are divided into nine groups—three source higher than 100, performance is 80%. The least time and
groups for beams and six groups for columns. Each beam and col- 100% performance are yielded at PN = 100 for FN:PN = 1.5:1, at
umn group is divided as every four stories, illustrated in Fig. 14B. which average time and average number of iterations are 55.81 s
The total number of possible solutions in the search space is and 255 rounds, respectively.
36  16  16  5  5 + 48  16  16  10 = 353,280 solutions. Fig. 16A illustrates the trends for each of the values of FN:PN
Tables 16–19 display the statistical results for each ratio of in design example 3. ABC algorithm exhibits reasonable perfor-
number of bees to number of food sources. The optimum cost is mance at PN = 300 and 400 for all FN:PN because the accuracy
denoted in bold and is equal to USD 79,033. Although the design for each of them is 100%. FN:PN = 2:1 yields the highest rate of
variables are 1.5 times those of the design example 2, the efficiency convergence to the minimum cost. Fig. 16B illustrates the rela-
does not decrease and the PN = 50 since FN:PN = 1:1 discover the tionship of PN and average time for each value of FN:PN. It
optimum cost for 8–9 times of 10 times. Fig. 15 illustrates the con- demonstrates that the trend for average time in design example
vergence histories of each number of bees for design example 3 3 is similar to those in design examples 1 and 2. Therefore, it
(FN:PN = 1:1); it is determined that the trend of convergence in can be stated that an increase in FN:PN results in an increase
example 3 is similar to that in examples 1 and 2. Table 20 presents in the rate of convergence as well as usage time for calculation
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 815

Table 11
Statistical results of design example 2 for the case of FN:PN = 1:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 47,401 274 15.73 47,584 211 22.80 47,401 238 49.00 47,342 256 80.42 47,342 302 125.25
2 47,401 326 19.18 47,342 335 36.20 47,342 316 66.14 47,342 298 92.43 47,342 297 121.30
3 47,342 264 15.47 47,401 226 24.46 47,401 222 45.67 47,342 283 89.27 47,342 285 116.17
4 47,767 244 14.10 48,024 190 20.07 47,342 309 64.21 47,342 252 77.47 47,342 290 118.63
5 47,342 361 19.97 47,342 353 38.51 47,342 322 66.98 47,342 325 101.80 47,342 283 116.10
6 47,342 275 15.79 47,342 290 31.35 47,401 287 58.91 47,342 265 83.48 47,342 265 108.75
7 47,342 262 14.68 47,401 319 35.11 47,342 352 72.74 47,342 264 81.43 47,342 305 125.23
8 48,629 209 12.26 47,342 349 37.08 47,342 308 63.70 47,342 273 84.35 47,342 289 118.58
9 47,401 343 19.74 47,342 271 30.25 47,342 267 56.58 47,342 306 95.03 47,342 323 131.22
10 47,401 300 17.00 47,342 320 34.57 47,342 292 60.88 47,342 265 82.02 47,342 311 127.24
Min 47,342 209 12.26 47,342 190 20.07 47,342 222 45.67 47,342 252 77.47 47,342 265 108.75
Max 48,629 361 19.97 48,024 353 38.51 47,401 352 72.74 47,342 325 101.80 47,342 323 131.22
Average 47,537 286 16.39 47,446 286 31.04 47,360 291 60.48 47,342 279 86.77 47,342 295 120.85
Med 47,401 275 15.76 47,342 305 32.96 47,342 300 62.29 47,342 269 83.91 47,342 294 119.97
Mode 47,401 – – 47,342 – – 47,342 – – 47,342 265 – 47,342 – –
SD 404 47 2.56 216 59 6.50 28 40 8.29 0 24 7.65 0 16 6.57

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 12
Statistical results of design example 2 for the case of FN:PN = 1.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 47,401 244 20.56 47,401 260 41.55 47,342 259 80.07 47,342 236 108.49 47,342 266 164.57
2 47,401 282 23.45 47,342 294 46.36 47,401 194 59.72 47,342 256 118.67 47,342 268 165.88
3 47,342 309 25.60 47,342 286 45.59 47,342 313 97.06 47,342 272 127.35 47,342 264 163.26
4 47,342 308 24.96 47,342 280 44.47 47,342 276 86.29 47,342 300 139.19 47,342 279 172.42
5 47,342 354 29.02 47,401 286 44.96 47,342 370 114.43 47,342 241 111.24 47,342 265 164.60
6 47,525 262 21.97 47,525 247 39.40 47,342 269 83.10 47,342 279 128.86 47,342 275 171.58
7 47,342 273 22.52 47,342 299 47.70 47,342 283 88.03 47,342 252 116.52 47,342 236 146.55
8 47,342 262 21.98 47,342 269 43.68 47,342 290 90.66 47,342 211 97.35 47,342 213 129.83
9 47,401 274 22.36 47,342 255 40.75 47,342 239 74.15 47,342 280 129.77 47,342 234 144.03
10 47,401 326 27.03 47,584 237 37.52 47,342 290 90.23 47,342 257 119.76 47,342 278 171.55
Min 47,342 244 20.56 47,342 237 37.52 47,342 194 59.72 47,342 211 97.35 47,342 213 129.83
Max 47,830 354 29.02 47,584 299 47.70 47,401 370 114.43 47,342 300 139.19 47,342 279 172.42
Average 47,475 289 23.94 47,396 271 43.20 47,348 278 86.37 47,342 258 119.72 47,342 258 159.43
Med 47,401 278 22.98 47,342 275 44.07 47,342 280 87.16 47,342 257 119.21 47,342 266 164.58
Mode 47,401 262 – 47,342 286 – 47,342 290 – 47,342 – – 47,342 – –
SD 179 34 2.65 88 21 3.27 18 46 14.33 0 26 12.16 0 22 14.34

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 13
Statistical results of design example 2 for the case of FN:PN = 2:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 47,830 257 27.71 47,342 302 63.23 47,342 308 126.72 47,342 270 167.72 47,342 311 255.76
2 47,401 292 30.90 47,401 253 52.72 47,342 272 111.96 47,342 230 141.30 47,342 262 214.61
3 47,342 277 29.76 47,342 290 60.38 47,342 276 113.44 47,342 278 173.47 47,342 262 216.47
4 47,342 329 34.49 47,342 317 65.12 47,342 346 141.56 47,342 300 184.54 47,342 279 230.94
5 47,342 225 23.93 47,342 318 66.25 47,342 312 128.20 47,342 283 176.16 47,342 220 179.40
6 47,719 210 22.29 47,342 286 59.89 47,342 250 100.85 47,342 264 165.17 47,342 281 231.34
7 47,401 281 29.62 47,342 323 66.92 47,342 284 117.16 47,342 242 150.82 47,342 248 204.26
8 47,342 298 31.68 47,401 263 54.65 47,342 358 145.57 47,342 288 179.07 47,342 263 216.18
9 47,342 243 25.92 47,342 294 61.44 47,342 288 119.37 47,342 279 173.43 47,342 253 207.66
10 47,342 354 38.01 47,342 220 45.60 47,342 265 107.50 47,342 222 137.71 47,342 293 242.17
Min 47,342 210 22.29 47,342 220 45.60 47,342 250 100.85 47,342 222 137.71 47,342 220 179.40
Max 47,830 354 38.01 47,401 323 66.92 47,342 358 145.57 47,342 300 184.54 47,342 311 255.76
Average 47,440 277 29.43 47,354 287 59.62 47,342 296 121.23 47,342 266 164.94 47,342 267 219.88
Med 47,342 279 29.69 47,342 292 60.91 47,342 286 118.26 47,342 274 170.57 47,342 263 216.33
Mode 47,342 – – 47,342 – – 47,342 – – 47,342 – – 47,342 262 –
SD 180 45 4.75 25 33 6.78 0 35 14.36 0 26 16.20 0 25 21.40

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.
816 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

81000
50 individuals
76000
100 individuals
71000 200 individuals

Cost ($)
66000 300 individuals
400 individuals
61000

56000

51000

46000
0 20 40 60 80 100 120 140 160 180 200
Number of iterations

Fig. 12. Convergence histories for each number of bees for design example 2 (FN:PN = 1:1).

Table 14
Accuracy of ABC algorithm for design example 2.

Number of bees FN:PN


(PN) 0.5:1 1:1 1.5:1 2:1
50 40% 40% 50% 60%
100 40% 60% 60% 80%
200 60% 70% 90% 100%
300 100% 100% 100% 100%
400 100% 100% 100% 100%

47650 FN:PN = 0.5:1 FN:PN = 1:1


FN:PN = 0.5:1 FN:PN = 1.5:1 FN:PN = 2:1
47600
FN:PN = 1:1 250.00
Average cost ($)

47550 FN:PN = 1.5:1


200.00
Average time (s)

47500 FN:PN = 2:1

47450 150.00

47400 100.00

47350 50.00
47300
0 100 200 300 400 500 0.00
0 100 200 300 400 500
Number of bees (PN)
Numner of bees (PN)
(A)
(B)
Fig. 13. Influence of population sizes and food sources for design example 2.

Table 15
Comparison of the optimum design variables for design example 2.

HPSACO [29] HBB-BC [29] ABC algorithm


Group b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS
B1 300 500 3-D19/6-D22 300 500 3-D19/6-D22 350 500 3-D19/6-D22
B2 300 500 3-D19/6-D22 300 500 3-D19/6-D22 300 500 3-D19/5-D22
B3 300 500 3-D19/5-D22 300 500 3-D19/5-D22 250 550 3-D19/4-D22
C1 400 400 8-D25 400 400 8-D25 350 350 6-D25
C2 500 500 8-D25 450 450 12-D25 500 500 6-D25
C3 350 350 8-D25 350 350 8-D25 350 350 6-D25
C4 350 350 8-D25 350 350 8-D25 350 350 4-D25
Cost (USD) 48,514 48,263 47,342

b = section width, h = section height, BS = bottom position of reinforcing bars, and TS = top position of reinforcing bars.

required in the optimization process. Table 21 presents a com- the results, it can be concluded that ABC algorithm is a straight-
parison of design variables. ABC algorithm is more economical forward tool yielding high performance when the appropriate
than HPSACO and HBB-BC by 5.06% and 2.59%, respectively. From parameters are selected.
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 817

D = 22.3 kN/m + L = 10.7 kN/m


36 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C5 C6 C6 C5


33 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C5 C6 C6 C5


30 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C5 C6 C6 C5


27 kN B3 B3 B3

D = 22.3 kN/m + L = 10.7 kN/m C5 C6 C6 C5


24 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3

12 @ 3.3m = 39.6 m
12 @ 3.3m = 39.6 m
21 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


18 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


15 kN B2 B2 B2

D = 22.3 kN/m + L = 10.7 kN/m C3 C4 C4 C3


12 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


9 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


6 kN B1 B1 B1

D = 22.3 kN/m + L = 10.7 kN/m C1 C2 C2 C1


3kN B1 B1 B1

C1 C2 C2 C1

7.5 m 7.5 m 7.5 m 7.5 m 7.5 m 7.5 m

(A) Applied loads (B) Design group


Fig. 14. Topology and lateral loads for design example 3.

Table 16
Statistical results of design example 3 for the case of FN:PN = 0.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 79,033 294 12.53 79,033 261 20.76 79,217 199 30.50 79,033 206 46.78 79,033 275 83.20
2 79,217 306 13.94 79,217 245 19.48 79,033 257 40.68 79,033 298 68.23 79,033 285 86.51
3 79,217 297 12.72 79,033 250 19.85 79,033 276 42.51 79,033 251 56.78 79,033 270 81.13
4 79,217 239 9.88 79,033 297 24.13 79,033 283 43.66 79,033 280 64.07 79,033 241 72.71
5 79,217 234 10.51 79,033 258 20.23 79,033 287 44.80 79,033 272 62.95 79,033 244 72.98
6 79,217 230 10.02 79,033 269 21.27 79,033 270 42.42 79,033 211 47.96 79,033 268 80.79
7 79,217 234 10.63 79,033 268 21.89 79,033 261 40.67 79,033 286 65.95 79,033 260 77.92
8 79,033 266 11.92 79,033 265 21.10 79,033 259 40.15 79,033 303 69.85 79,033 259 78.03
9 79,033 261 11.18 79,455 225 17.61 79,033 227 35.11 79,033 251 57.54 79,033 218 65.55
10 79,089 291 12.63 79,033 311 25.44 79,033 293 48.22 79,033 271 61.53 79,033 249 75.81
Min 79,033 230 9.88 79,033 225 17.61 79,033 199 30.50 79,033 206 46.78 79,033 218 65.55
Max 79,217 306 13.94 79,455 311 25.44 79,217 293 48.22 79,033 303 69.85 79,033 285 86.51
Average 79,149 265 11.60 79,094 265 21.18 79,052 261 40.87 79,033 263 60.16 79,033 257 77.46
Med 79,217 264 11.55 79,033 263 20.93 79,033 266 41.55 79,033 272 62.24 79,033 260 77.97
Mode 79,217 234 – 79,033 – – 79,033 – – 79,033 251 – 79,033 – –
SD 89 30 1.35 139 25 2.26 58 29 4.98 0 33 7.91 0 19 6.03

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Note: To compare the performance of the ABC algorithm, we designed using nonlinear analysis, it is cheaper and safer than
used the ultimate loads in the design of RC frames to measure one designed by elastic analyses because the applied loads for
the performance of finite element analysis by the direct stiffness nonlinear analysis are closer to a structural really more realistic
method as the same as the other algorithms [29]. The direct stiff- [38–39]. In case of nonlinear analyses, the parameters from
ness method is a linear analysis which simplifies in a practical structural analysis must enter into the constraint functions of the
common than non-linear analysis. In fact, if the structure is ABC algorithm too.
818 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

Table 17
Statistical results of design example 3 for the case of FN:PN = 1:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 79,033 311 23.43 79,033 260 38.21 79,033 288 84.54 79,033 265 115.55 79,033 234 133.19
2 79,397 233 18.22 79,033 289 44.20 79,033 263 76.83 79,033 239 104.11 79,033 265 152.75
3 79,033 339 26.55 79,217 203 30.19 79,033 314 91.38 79,033 255 111.41 79,033 236 134.22
4 79,033 343 27.89 79,033 271 42.76 79,033 311 90.98 79,033 242 105.83 79,033 254 145.03
5 79,033 332 27.06 79,033 277 43.59 79,033 223 64.18 79,033 275 121.13 79,033 267 152.86
6 79,033 314 24.27 79,033 275 42.14 79,033 290 85.18 79,033 287 126.59 79,033 268 153.53
7 79,033 266 20.51 79,089 164 23.84 79,033 246 70.83 79,033 220 95.13 79,033 223 128.02
8 79,033 254 19.63 79,033 307 46.43 79,033 253 74.16 79,033 252 110.47 79,033 257 146.34
9 79,033 278 21.86 79,033 227 33.88 79,033 227 66.25 79,033 246 106.25 79,033 264 151.94
10 79,217 294 22.11 79,033 261 39.07 79,033 258 74.94 79,033 236 103.82 79,033 261 152.76
Min 79,033 233 18.22 79,033 164 23.84 79,033 223 64.18 79,033 220 95.13 79,033 223 128.02
Max 79,397 343 27.89 79,217 307 46.43 79,033 314 91.38 79,033 287 126.59 79,033 268 153.53
Average 79,088 296 23.15 79,057 253 38.43 79,033 267 77.93 79,033 252 110.03 79,033 253 145.06
Med 79,033 303 22.77 79,033 266 40.61 79,033 261 75.89 79,033 249 108.36 79,033 259 149.14
Mode 79,033 – – 79,033 – – 79,033 – – 79,033 – – 79,033 – –
SD 123 38 3.28 59 43 7.14 0 32 9.71 0 20 9.17 0 16 9.71

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 18
Statistical results of design example 3 for the case of FN:PN = 1.5:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 79,033 285 32.76 79,033 290 64.02 79,033 244 106.40 79,033 211 136.51 79,033 237 203.50
2 79,033 248 29.16 79,033 236 51.31 79,033 259 113.49 79,033 240 157.42 79,033 253 220.16
3 79,033 303 35.44 79,033 295 64.81 79,033 253 110.35 79,033 231 149.21 79,033 260 226.16
4 79,033 311 35.30 79,033 245 53.69 79,033 268 116.84 79,033 249 162.07 79,033 209 179.99
5 79,033 265 30.18 79,033 242 53.18 79,033 276 120.77 79,033 225 147.03 79,033 204 176.18
6 79,217 221 25.43 79,033 275 60.92 79,033 249 108.32 79,033 233 151.63 79,033 229 197.95
7 79,033 319 36.85 79,033 209 44.69 79,033 246 108.00 79,033 245 160.70 79,033 240 209.27
8 79,033 239 27.47 79,033 308 67.70 79,033 259 113.77 79,033 271 177.64 79,033 244 211.14
9 79,033 319 37.25 79,033 237 51.24 79,033 288 126.51 79,033 257 168.57 79,033 272 236.46
10 79,033 264 31.63 79,033 215 46.52 79,033 255 111.23 79,033 233 152.08 79,033 271 236.59
Min 79,033 221 25.43 79,033 209 44.69 79,033 244 106.40 79,033 211 136.51 79,033 204 176.18
Max 79,217 319 37.25 79,033 308 67.70 79,033 288 126.51 79,033 271 177.64 79,033 272 236.59
Average 79,052 277 32.15 79,033 255 55.81 79,033 260 113.57 79,033 240 156.29 79,033 242 209.74
Med 79,033 275 32.19 79,033 244 53.44 79,033 257 112.36 79,033 237 154.75 79,033 242 210.20
Mode 79,033 319 – 79,033 – – 79,033 259 – 79,033 233 – 79,033 – –
SD 58 35 4.08 0 34 8.02 0 14 6.29 0 17 11.68 0 23 21.10

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.

Table 19
Statistical results of design example 3 for the case of FN:PN = 2:1.

PN 50 100 200 300 400


No. Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time Cost Itr. Time
1 79,033 315 47.52 79,033 266 77.38 79,033 273 157.77 79,033 216 185.00 79,033 246 283.86
2 79,089 249 37.28 79,033 239 70.09 79,033 275 159.44 79,033 245 212.78 79,033 260 296.33
3 79,033 257 39.24 79,033 246 71.66 79,033 240 137.14 79,033 216 184.85 79,033 276 317.82
4 79,033 344 51.82 79,033 260 75.61 79,033 278 159.58 79,033 252 218.73 79,033 236 267.70
5 79,033 299 45.70 79,033 236 68.56 79,033 229 130.34 79,033 236 203.89 79,033 306 352.65
6 79,033 331 49.85 79,033 263 76.33 79,033 304 175.12 79,033 264 229.73 79,033 270 308.73
7 79,033 300 46.10 79,033 281 82.70 79,033 284 164.07 79,033 250 216.66 79,033 280 322.03
8 79,033 274 41.56 79,033 252 73.31 79,033 282 162.49 79,033 221 190.20 79,033 236 268.52
9 79,033 283 43.18 79,033 302 89.50 79,033 282 161.61 79,033 246 213.25 79,033 270 308.48
10 79,033 242 36.73 79,033 273 79.79 79,033 279 161.68 79,033 204 175.77 79,033 277 317.78
Min 79,033 242 36.73 79,033 236 68.56 79,033 229 130.34 79,033 204 175.77 79,033 236 267.70
Max 79,089 344 51.82 79,033 302 89.50 79,033 304 175.12 79,033 264 229.73 79,033 306 352.65
Average 79,039 289 43.90 79,033 262 76.49 79,033 273 156.92 79,033 235 203.09 79,033 266 304.39
Med 79,033 291 44.44 79,033 262 75.97 79,033 279 160.59 79,033 241 208.34 79,033 270 308.60
Mode 79,033 – – 79,033 – – 79,033 282 – 79,033 216 – 79,033 236 –
SD 17 35 5.19 0 20 6.30 0 22 13.21 0 20 17.98 0 22 26.15

Itr. = Iteration, Med = Median, SD = Standard deviation, respectively.


Cost is USD and time is seconds.
A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820 819

125000
120000 50 individuals
115000 100 individuals
110000 200 individuals

Cost ($)
105000 300 individuals
100000 400 individuals
95000
90000
85000
80000
75000
0 20 40 60 80 100 120 140 160 180 200
Number of iterations

Fig. 15. Convergence histories for each number of bees for design example 3 (FN:PN = 1:1).

Table 20
Accuracy of ABC algorithm for design example 3.

Number of bees FN:PN


(PN) 0.5:1 1:1 1.5:1 2:1
50 30% 80% 90% 90%
100 80% 80% 100% 100%
200 90% 100% 100% 100%
300 100% 100% 100% 100%
400 100% 100% 100% 100%

79160
FN:PN = 0.5:1 FN:PN = 1:1
FN:PN = 0.5:1
79140 FN:PN = 1.5:1 FN:PN = 2:1
FN:PN = 1:1
350.00
Average cost ($)

79120
FN:PN = 1.5:1
300.00
Average time (s)

79100 FN:PN = 2:1


250.00
79080 200.00
150.00
79060
100.00
79040
50.00
79020 0.00
0 100 200 300 400 500 0 100 200 300 400 500
Number of bees (PN) Number of bees (PN)
( A) (B)
Fig. 16. Influence of population sizes and food sources for design example 3.

Table 21
Comparison of the optimum design variables for the design example 3.

HPSACO [29] HBB-BC [29] ABC algorithm


Group b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS b (mm) h (mm) BS/TS
B1 350 550 3-D19/7-D22 350 600 3-D19/6-D22 400 550 3-D19/6-D22
B2 350 550 3-D19/6-D22 300 550 5-D19/6-D22 350 550 3-D19/5-D22
B3 350 550 3-D19/5-D22 300 500 3-D19/6-D22 300 550 3-D19/4-D22
C1 500 500 8-D25 450 450 10-D25 450 450 6-D25
C2 650 650 10-D25 600 600 12-D25 600 600 8-D25
C3 450 450 8-D25 400 400 10-D25 350 350 6-D25
C4 500 500 10-D25 500 500 10-D25 500 500 6-D25
C5 350 350 6-D25 350 350 8-D25 350 350 6-D25
C6 400 400 4-D25 400 400 4-D25 350 350 4-D25
Cost (USD) 83,250 81,138 79,033

b = section width, h = section height, BS = bottom position of reinforcing bars, and TS = top position of reinforcing bars.
820 A. Tapao, R. Cheerarot / Engineering Structures 151 (2017) 802–820

7. Conclusions [13] Pan QK, Tasgetiren MF, Suganthan PN, Chua TJ. A discrete artificial bee colony
algorithm for the lot-streaming flow shop scheduling problem. Inf Sci
2011;181(12):2455–68.
This study investigates the parameters that control the effi- [14] Karaboga D, Gorkemli B. A quick artificial bee colony (qABC) algorithm and its
ciency of ABC algorithm in optimum design of RC frames. performance on optimization problems. Appl Soft Comput 2014;23
(8):227–38.
ACI318-08 standard is utilized for evaluating the optimum design.
[15] Li X, Yang G. Artificial bee colony algorithm with memory. Appl Soft Comput
The design constraints are defined as the penalty function. Three 2016;41(1):362–72.
sizes of RC frames are used as the design examples. The load com- [16] Sonmez M. Discrete optimum design of truss structures using artificial bee
colony algorithm. Struct Multidisc Optim 2010;43(1):85–97.
bination and lateral earthquake are applied on each RC frame. The
[17] Sonmez M. Artificial Bee Colony algorithm for optimization of truss structures.
efficiency of ABC algorithm is controlled by four parameters, Appl Soft Comput 2011;11(2):2406–18.
namely, a stopping condition, number of trial limits, number of [18] Lee KS, Geem ZW. A new structural optimization method based on the
bees, and number of food sources. The stopping condition is critical harmony search algorithm. Comput Struct 2004;82(9–10):781–98.
[19] Li LJ, Huang ZB, Liu F, Wu QH. A heuristic particle swarm optimizer for
to the optimization process time. From the statistical results, it is optimization of pin connected structures. Comput Struct 2007;85(7–8):340–9.
demonstrated that the stopping condition of 60 times is most suit- [20] Perez RE, Behdinan K. Particle swarm intelligence for structural design
able for ABC algorithm. The number of trial limits of 100 times is optimization. Comput Struct 2007;85(19–20):1579–88.
[21] Ozturk HT, Durmus AY, Durmus AH. Optimum design of a reinforced concrete
sufficient to solve this problem. The efficiency of ABC algorithm beam using artificial bee colony algorithm. Comput Concrete 2012;10
is higher than 70% when the number of food sources is higher than (3):295–306.
or equal to 200 and is 100% for all the frames when PN is higher [22] ACI 318-08 Building code requirements for structural concrete and
commentary. ACI Committee 318. Structural Building Code. Farmington Hills
than 300; moreover, FN:PN = 0.5:1 yields the least time. In addi- (MI, USA): American Concrete Institute; 2008.
tion, it is concluded that the original ABC algorithm is a straightfor- [23] Coello Coello CA, Christiansen AD, Santos Hernández F. A simple genetic
ward tool providing high performance when appropriate algorithm for the design of reinforced concrete beams. Eng Comput 1997;13
(4):185–96.
parameters are selected.
[24] Chakrabarty BK. Models for optimal design of reinforced concrete beams.
Comput Struct 1992;42(3):447–51.
Acknowledgments [25] Jahjouh MM, Arafa MH, Alqedra MA. Artificial Bee Colony (ABC) algorithm in
the design optimization of RC continuous beams. Struct Multidisc Optim
2013;47(6):963–79.
The authors would like to acknowledge the Concrete and Com- [26] Aydoğdu I, Akın A, Saka MP. Design optimization of real world steel space
puter Research Unit, Faculty of Engineering, Mahasarakham frames using artificial bee colony algorithm with Levy flight distribution. Adv
University for providing facilities and equipment. The authors also Eng Softw 2016;92(1):1–14.
[27] Hakli H, Uğuz H. Levy flight distribution for scout bee in artificial bee colony
thank Assoc.Prof. John Morris for his valuable support and algorithm. Lect Notes Softw Eng 2013;1(3):254–8.
suggestions. [28] Aydoğdu I. Optimum design of 3-d irregular steel frames using ant colony
optimization and harmony search algorithms. Middle East Technical
University; 2010. Ph.D. Thesis.
References [29] Kaveh A, Sabzi O. A comparative study of two meta-heuristic algorithms for
optimum design of reinforced concrete frames. Int J Civ Eng 2011;9
[1] Karaboga D. An idea based on honey bee swarm for numerical optimization, (3):193–206.
Technical Report TR06. Erciyes University, Engineering Faculty, Computer [30] Albuquerque AT, Debs MK, Melo A. A cost optimization-based design of
Engineering Department; 2005. precast concrete floors using genetic algorithms. Autom Const 2012;22
[2] Karaboga D, Basturk B. A powerful and efficient algorithm for numerical (1):348–56.
function optimization: artificial bee colony (ABC) algorithm. J Glob Optim [31] Sahab MG, Ashour AF, Toropov VV. A hybrid genetic algorithm for reinforced
2007;39(3):459–71. concrete flat slab buildings. Comput Struct 2005;83(8–9):551–9.
[3] Karaboga D, Basturk B. On the performance of artificial bee colony (ABC) [32] Perera R, Vique J. Strut-and-tie modelling of reinforced concrete beams using
algorithm. Appl Soft Comput 2008;8(1):687–97. genetic algorithms optimization. Constr Build Mater 2009;23(8):2914–25.
[4] Karaboga D, Akay B. A comparative study of artificial bee colony algorithm. [33] Guerra A, Kiousis PD. Design optimization of reinforced concrete structures.
Appl Math Comput 2009;214(1):108–32. Comput Concrete 2006;3(5):313–34.
[5] Karaboga D, Ozturk C. A novel clustering approach: Artificial Bee Colony (ABC) [34] Nedushan BA, Varaee H. Minimum cost design of concrete slabs using particle
algorithm. Appl Soft Comput 2011;11(1):652–7. swarm optimization with time varying acceleration coefficients. World Appl
[6] Akay B, Karaboga D. A modified artificial bee colony algorithm for real- Sci J 2011;13(12):2484–94.
parameter optimization. Inf Sci 2012;192(1):120–42. [35] Kaveh A, Sabzi O. Optimal design of reinforced concrete frames Using big
[7] Zhang C, Ouyang D, Ning J. An artificial bee colony approach for clustering. bang-big crunch algorithm. Int J Civ Eng 2012;10(3):189–200.
Expert Syst Appl 2010;37(7):4761–7. [36] Akin A, Saka MP. Harmony search algorithm based optimum detailed design of
[8] Taheri J, Lee YC, Zomaya AY, Siegel HJ. A Bee Colony based optimization reinforced concrete plane frames subject to ACI 318–05 provisions. Comput
approach for simultaneous job scheduling and data replication in grid Struct 2015;147(1):79–95.
environments. Comput Oper Res 2013;40(6):1564–78. [37] Kwak HG, Kim J. Optimum design of reinforced concrete plane frames based on
[9] Xiang WL, An MQ. An efficient and robust artificial bee colony algorithm for predetermined section database. Comput Aided Des 2008;40(3):396–408.
numerical optimization. Comput Oper Res 2013;40(5):1256–65. [38] Magliulo G, Maddaloni G, Cosenza E. Comparison between non-linear dynamic
[10] Das S, Biswas S, Kundu S. Synergizing fitness learning with proximity-based analysis performed according to EC8 and elastic and non-linear static analyses.
food source selection in artificial bee colony algorithm for numerical Eng Struct 2007;29(11):2893–900.
optimization. Appl Soft Comput 2013;13(12):4676–94. [39] Fajfar P. A nonlinear analysis method for performance based seismic design.
[11] Zhu G, Kwong S. Gbest-guided artificial bee colony algorithm for numerical Earthq Spectra 2000;16(3):573–92.
function optimization. Appl Math Comput 2010;217(7):3166–73.
[12] Gao WF, Liu SY. A modified artificial bee colony algorithm. Comput Oper Res
2012;39(3):687–97.

Das könnte Ihnen auch gefallen