International Journal of Supply and Operations Management

IJSOM
August 2018, Volume 5, Issue 3, pp. 218-233
ISSN-Print: 2383-1359
ISSN-Online: 2383-2525
www.ijsom.com

An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation Problems

with Cold-standby Components

Mohammad N. Juybari a, Mostafa Abouei Ardakan*, a, Hamed Davari-Ardakani a

a
Department of Industrial Engineering, Faculty of Engineering, Kharazmi University, Tehran, Iran

Abstract
One of the most practical methods for improving system reliability is making a tradeoff between components reliability
and redundancy levels, which is known as reliability-redundancy allocation problem (RRAP). The RRAP aims to
maximize the overall system reliability by creating a balance between the component reliabilities and the number of
redundant components in each subsystem. In the RRAP, the redundant components are performed in a predetermined
order under a redundancy strategy. In this paper, a cold standby redundancy strategy is considered for the redundant
components. Besides, a penalty guided water cycle algorithm is adjusted for solving the problem. The proposed algorithm
is implemented on two famous benchmark problems to evaluate the performance of the proposed approach. The obtained
numerical results reveal the superiority of the proposed solution method over all previous studies.

Keywords: Reliability-redundancy allocation problem; Cold-standby strategy; Reliability optimization; Water cycle
algorithm.

1. Introduction

Most of the catastrophic incidents during the last century might imply that failures and their consequences are due to
poor reliability of their scales. The explosion of two NASA’s space shuttles Challenger and Columbia in 1986 and 2003,
respectively, are obvious examples that failure of a system can have widespread effect on the total expected mission of
the system. Another example of the role of reliability in structural design is the biggest nuclear disasters in the world,
like the explosion of four operating nuclear reactors at the Chernobyl site in 1986 (Elsayed, 2012). These examples
signify that reliability plays a critical role in industries and has far reaching effect on consumers of services. The formal
definition of reliability is the probability that a service or product will operate without failures for a definite time interval
under the specified operating stresses. In other words, reliability is the system’s success measurement in supporting its
function perfectly during its specified period of time or its design life (Elsayed, 2012).

Reliability optimization problem (ROP) is an important topic that has attracted the interest of many academic and applied
engineering studies. Two different categories of ROP are: (a) Redundancy Allocation Problem (RAP) and (b) Reliability-
Redundancy Allocation Problem (RRAP). In the RAP, the component choices and their characteristics such as reliability,
cost and weight are predetermined. The goal of the RAP is to find the optimal number of redundant components in each
subsystem in order to maximize the overall system reliability whereas in the RRAP, the component reliability is a design
variable and its characteristics are computed as increasing nonlinear functions of component reliability (Ardakan &
Hamadani, 2014a). The RRAP is formulated as a Mixed Integer Nonlinear Programming (MINLP) problem. Therefore,
solving such a complex problem is very difficult especially on a large scale. In this paper the RRAP is considered.

Corresponding author email address: [email protected]

218
 Juybari, Abouei Ardakan and Davari-Ardakani

A redundancy strategy determines the way of using redundant components in a subsystem. Generally, there are two
traditional redundancy strategies called active and standby. In the active strategy, all the redundant components start
their operation from time zero. There are three alternatives to the standby strategy, namely cold, warm and hot. In the
cold standby strategy, one component is in active mode and the rest of the components are kept idle and protected from
operational stresses. In this strategy, when the active component fails, the first redundant component is activated and
starts its mission as a new one. But in the warm standby strategy, although just one component is needed, all redundant
ones are somewhat affected by the operational stresses. Finally, in the hot standby strategy, the component failure is not
influenced by whether it was in operating mode or in idle model. As an extension to redundancy strategies, Ardakan and
Hamadani (2014b) introduced a new redundancy strategy for adding redundant components to subsystems in the RAP,
which is called ‘mixed’ strategy. This strategy is a comprehensive model of both active and standby strategies. Abouei
Ardakan et al. (2016) implemented the mixed strategy in the RRAP, which is more complicated than the RAP. Peiravi
et al. (2017) formulated a general form of the mixed strategy and introduced “k-mixed”. This novel redundancy strategy
utilizes concepts of k-out-of-n and mix strategy, simultaneously. They implemented k-mixed in the RAP.

Generally, in standby strategies, a switching system is needed for replacing the failed component by a new one
(Tavakkoli-Moghaddam, Safari, & Sassani, 2008). System designers consider two possible scenarios for switching
mechanisms. In the first suggested scenario (S1), the failure detector scans the system performance in order to find any
failure in the active components and replace them by a redundant one. In this scenario, the switch reliability is considered
a non-increasing function of time (𝜌𝑖 (𝑡)). On the other hand, in the second scenario (S2), the probability of switch failure
is considered a constant value 𝜌𝑖 (𝑡).. In this case, the switch failure happens only when it is used (Coit, 2001).

Kim (2018) studied optimal reliability design for a RAP system with mixed components and imperfect switching policy.
A Structured Markov chain was used to calculate each subsystem’s reliability. Most of the researchers formulated the
RRAP only by considering the active redundancy strategy. Both exact and meta-heuristic methods were applied to the
problem. Exact optimization methods including dynamic programming (Kuo, 2001) and branch and bond (Kuo, Lin, Xu,
& Zhang, 1987) have been used to solve the problem. Hikita et al. (1992) developed a dynamic programming approach
to solve the RRAP by a single constrained surrogate method. However, most of traditional methods failed to solve the
problem in a reasonable time, due to high-dimension of the RRAP. Hence, researchers applied meta-heuristic methods
to solve the reliability optimization problem (Wang & Li, 2012). In the field, many meta-heuristic and nature inspired
algorithms such as genetic algorithm (Ardakan, Hamadani, & Alinaghian, 2015; Ramirez-Marquez, Coit, & Konak,
2004; Tavakkoli-Moghaddam et al., 2008), artificial neural networks (Habib, Alsieidi, & Youssef, 2009), particle swarm
optimization (dos Santos Coelho, 2009), ant colony optimization (Liang & Smith, 2004; Nahas & Nourelfath, 2005)
have been developed for solving different RRAP problems.

For solving the RRAP with the active redundant strategy, a two-phase approach based on an Immune Algorithm (IA)
has been developed by Hsieh and You (2011) . They implemented the IA in the first phase and developed a new procedure
to improve the final solution in the second phase. For the same problem, a hybrid algorithm which is a combination of
Harmony search and Differential Evolution algorithm was introduced by Wang and Li (2012). Yeh and Hsieh (2012)
developed a modified Artificial Bee Colony (ABC) for the RRAP problems. Results showed that the proposed approach
leads to better solutions compared to those of other meta-heuristic algorithms. A new version of Harmony search (HS)
called EGHS was proposed by Zou et al. (2011) to solve the RRAP problem. The concept of EGHS is inspired by mixing
the HS algorithm and the concept of swarm intelligence in the particle swarm optimization (PSO) algorithm. In another
study, for solving the RRAP problems, a novel particle swarm optimization algorithm called IPSO was introduced by
Wu et al. (2011).

Afonso et al. (2013) used an Imperialist Competitive Algorithm (ICA) for the RRAP problems. Valian and Valian (2013)
modified the Cuckoo Search (CS) algorithm for the RRAPs and tested the performance of the proposed algorithm in
different benchmark problems. Valian et al. (2013) improved the convergence rate and accuracy of the CS and compared
its performance in RRAP problems with the results of other studies. A new version of the Stochastic Fractal Search (SFS)
algorithm called PSFS was represented by M.A.Mellal and E.Zio (2016) to solve the reliability optimization problems,
including the RRAP with active components. The idea of PSFS was inspired by adding a penalty function to the SFS
algorithm. In most practical optimization problems, handling of conflicting objectives is a serious concern. Ardakan and
Rezvan (2017) formulated the RRAP as a bi-objective problem by considering system cost and total reliability as two
opposite goals.

It should be noted that all the above-mentioned studies considered the RRAP with the active redundancy strategy. For
the first time, Ardakan and Hamadani (2014a) developed the RRAP with a standby redundancy strategy and solved the
problem by using a penalty guided GA. The outcomes of their study confirmed that the standby redundancy strategy
outperforms the active one.

In the present study, the RRAP is considered with the cold-standby strategy for redundant components. A recently
introduced algorithm called Water Cycle Algorithm (WCA) (Eskandar et al., 2012) is adjusted and implemented on the

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 219

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

problem. The rest of the paper is organized as follows. In Section 2, the formulation of the RRAP with the cold standby
strategy and two benchmark problems are presented. Section 3 presents the WCA for solving the proposed non-linear
problems. The efficiency of the WCA in the RRAP is presented in Section 4. Finally, Section 5 contains the conclusions.

Notations:
t: total considered time
m: total number of subsystems
i : exponential distribution factor which indicates the failure rate of components in subsystem i;
1i m
 (t ) , i : Switch reliabilities at given time t for the first and second switching scenarios, respectively.
i
n: n=(n1,n2,…,nm) is a row vector of redundancy level representing the number of components
allocated to subsystem i.
r: r=(r1,r2,…,rm) is a row vector of the reliability for each component in subsystem i; 1  i  m
R total reliability of system without considering penalty function
Ri (t ) estimated lower bound of reliability for subsystem i at time t; 1  i  m
Rc,i(tm): cold-standby reliability at mission time (tm) calculated for subsystem i; 1  i  m
Rs: total reliability of system
M: total number of considered resource constraints
gj(r,n): The jth constraint of the mathematical formulation. 1  j  M
f(r,n): the objective function of the mathematical model
l: l=(l1,l2,…,lM) is a vector that implies the resource bounds
fix(t): the probability density function of the xth component failure time in subsystem i at time t;
1i m
Ri and Ri(ni,t): reliability of subsystem i at time t involving n components; 1  i  m
V, C, and W: The upper bound of the sum of the subsystems’ product of volume and weight, the upper
bound on the total cost of the entire system, and the upper bound on the total weight of the
system, respectively.
wi, vi, and ci: Weight, volume, and cost of each component in subsystem i, respectively.
 i , i : physical features of subsystem i; 1  i  m
Z+: discrete space of positive integers
Raindrop: solutions that are generated by the water cycle algorithm
Npop the quantity of raindrops to be generated in the initializing step of the algorithm
NS intensity of flow
d distance between individuals
Xindividual flow direction of a particular individual
i : penalty parameter greater than one

 (ri , ni ) : penalty term

UB and LB: Upper and lower bound of the variable, respectively
BP: the position of the best point among the considered group of involved points
Nvar Number of variables in a single solution
max_iteration: maximum number of iterations

1. Formulation of the reliability-redundancy allocation problem

Improving the system reliability is the main objective of reliability optimization problems. The RRAP is useful for
systems that have high reliability requirements (dos Santos Coelho, 2009). The RRAP is considered with the aim of
maximizing system reliability subject to some non-linear constraints such as weight, cost and volume as follows:

(1)
Maximize Rs  f (r, n)

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 220

 Juybari, Abouei Ardakan and Davari-Ardakani

subject to g j (r, n)  l j (2)

0  ri  1, ri , ni  Z  , 1 i  m
where Rs indicates the reliability of the system; gj represents the jth constraint; r = (r1, r2, . . . , rm) is the vector that
indicates the component reliabilities of the system. n = (n1, n2, ..., nm) is a vector that shows the redundancy level of
subsystems; ri and ni indicate the reliability of and the number of components in the ith subsystem, respectively. f (r, n)
is the objective function of the problem i.e. the system reliability; lj indicates the maximum level of resource. j and m is
the total number of subsystems. The aim of the model is to maximize the total reliability of the system by determining
the redundancy level and the components’ reliability in each subsystem while considering the resource limitations. The
RRAP model is categorized as the constrained MINLP optimization problem (Ardakan & Hamadani, 2014a).

2.1. System reliability with the cold standby redundancy strategy

Coit (2001) presents a general formulation of a subsystem reliability with the cold standby strategy that is appropriate
for any distribution of component time-to-failure, as in Eq. (3).
ni 1 t
Ri (t )  ri (t )    ri (t  u ) f i ( x ) (u ) du (3)
x 1 0

where ri (t ) indicates the component reliability at time t in subsystem i; ni shows the number of components in the ith
subsystem and fi ( x ) (t ) is the probability density function for the xth failure arrival for subsystem i.

For determining the reliability of the subsystem with imperfect switching, Coit (2001) represented Eqs. (4) and (5) for
two scenarios as follows:

Scenario 1 (S1): Continuous scanning and switching:

ni 1 t
Ri (t )  ri (t )    i (u ) ri (t  u ) f i ( x ) (u ) du (4)
x 1 0

Scenario 2 (S2): Switch activation only in response to failures:

ni 1 t
(5)
Ri (t )  ri (t )   i x  ri (t  u ) f i ( x ) (u ) du
x 1 0

where i (t ) and i are switch reliabilities at time t for the first and second scenarios, respectively. This paper considers
the first switching scenario (S1). Coit (2001), also determined an appropriate lower bound on subsystem reliability, Ri (t )
, as follows:
ni 1 t
Ri (t )  ri (t )  i (t )   ri (t  u ) f i ( x ) (u ) du (6)
x 1 0

As it is clear, for all u<=t, i (t )  i (u) , therefore, Eq. (6) is an estimation for Eq. (4). Coit (2001), considered the time-
to-failure of components as the Erlang distribution. In the RRAP, all non-linear constraints were extended by assuming
the exponential time-to-failure (Dhingra, 1992). Hence, in this study, Eq. (6) is developed based on the exponential
distribution and put into practice. In this case, Eq. (6) can be derived by applying the Poisson process. Thus, the
occurrences of subsystem failures are strictly less than ni failures (Coit, 2001), and Eq. (7), reformulates the Eq. (6) as
follows:

ni 1
e  i t (i t ) x
Ri (t )  ri (t )  i (t ) (7)
x 1 x!
where i is the component failure rate in the exponential distribution and t represents the total mission time of the
system.

In the present study, the calculation of subsystem reliability is based on Eq. (7). Therefore, instead of using ri as a decision
variable, the failure rate ( i ) is considered the decision variable. As a result, there are two decision variables in this
study as i and ni. We can obtain ri based on i . In the present work, two famous benchmark problems are considered.
In the next subsections, these benchmarks are introduced.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 221

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

2.2. Test problem 1 (P1): Series system

The first test problem includes five subsystems as series components which form a mixed-integer non-linear
programming problem as follows:
m
Maximize f (r, n)   Ri (ni , t ) (8)
i 1

Subject to
m
g1 (r, n)   wi . vi2 . ni2  V (9)
i 1
i
m
 1000 
g 2 (r, n)    i .    .  ni  e
0.25 ni
  C (10)
i 1  ln ri 
m
g3 (r, n)   wi . ni . e0.25 ni  W (11)
i 1

0  ri  1, ri , ni  Z  , 1 i  m (12)

This problem was also considered by (Afonso et al., 2013; Ardakan & Hamadani, 2014a; Y-C Hsieh & You, 2011;
Mellal & Zio, 2016; Valian et al., 2013; Valian & Valian, 2013; Yeh & Hsieh, 2011). A schematic view of the Series
system is depicted in Fig. 1.

1 2 3 4 5

Figure 1. A schematic view of the Series system (P1)

Here, the objective function, Eq. (8), contains the redundancy level of subsystem i and the reliability of the component
applied to gain the maximum system reliability at time t. As mentioned before, the reliability is calculated using Eq. (7).
Volume, cost and weight limitations are considered in Eqs. (9- 11). A combination of weight, redundancy allocation and
volume is given by Eq. (9). Eq. (10) and Eq. (11) are cost and weight limitations, respectively (Afonso et al., 2013). In
this formulation, V is the upper limit of the sum of the subsystem’s product of volume and weight, W represents the
upper limit on the weight of the system and C denotes the upper limit on the cost of the whole system. Also, weight of
each component in subsystem i is denoted by wi, volume of each component in subsystem i is symbolized by vi and
finally, ci is the cost of each component in subsystem i. Physical features of the ith subsystem are implied by i and i
. Discrete space of positive integers is represented by Z  .

2.3. Test problem 2 (P2): Series-Parallel system

Same as the first test problem, the second one includes five subsystems and is a mixed-integer nonlinear programming
problem as follows:
(13)
Maximize f (r, n) 1  (1  R1R2 )(1  (R3  R4  R3 R4 )R5 )
Subject to
g1 (r, n)  V , g2 (r, n)  C , g3 (r, n)  W
0  ri  1, ri , ni  Z  , 1 i  m

where, the calculation of Ri is based on Eq. (7), and all variable conditions are the same as those mentioned in P1. We
can find the same test problem in (Afonso et al., 2013; Ardakan & Hamadani, 2014a, 2014b; Mellal & Zio, 2016; Valian
et al., 2013; Valian & Valian, 2013; Wang & Li, 2012). Fig. 2 depicts this Series-Parallel system schematically.
1 2

3
5

Figure 2. A schematic view of the Series-Parallel system (P2)

3. Solution method: Water Cycle Algorithm

Water Cycle Algorithm, as one of the most powerful evolutionary algorithms, was first introduced by Eskandar et al.,
2012. The idea of Water Cycle Algorithm (WCA) was inspired by water circulation in nature. Fig. 3 shows a simple

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 222

 Juybari, Abouei Ardakan and Davari-Ardakani

diagram of the hydrologic cycle based on the observations in nature. By melting the glaciers in the mountains, streams
are formed and move downhill. By joining some streams through their journey downhill, the rivers are made. Finally,
the rivers end up in a sea. During this mechanism, water in rivers and lakes is evaporated and comes back to the earth as
raindrops.

Figure 3. A schematic diagram of water cycle in nature (Eskandar et al., 2012)

Same as other metaheuristic algorithms, creating the initial population is the first step of this algorithm. Here, the initial
population is called raindrops. The best raindrop is considered sea. Then, several good individuals (raindrops) are chosen
as rivers and the rest of individuals are implied as streams. Each river absorbs streams based on their magnitude and
flows to the sea (Eskandar et al., 2012).

3.1. Creating the first population

In the GA and PSO, the values of a single population were formed as an array called “Chromosome” and “Particle
Position”, respectively. Accordingly, in this algorithm, this array is called “Raindrop” for each individual. The test
problems applied in this paper contain 10 variables. Hence, the particular raindrop for these problems is a 1×10 vector
as follows:

Raindrop  1  2  3  4  5 n1 n2 n3 n4 n5 (14)

Where 𝜆i is failure rate of the component and ni indicates the redundancy in each subsystem. To start the optimization
process, a population with Npop size is needed. We can obtain the population of raindrops by representing a matrix of size
Npop × 10 as Fig. 4.

  1, 1  1, 2  1, 3  1, 4  1, 5 n1, 1 n1, 2 n1, 3 n1, 4 n1, 5 

  2, 1  2, 2  2, 3  2, 4  2, 5 n 2, 1 n 2, 2 n 2, 3 n 2, 4 n 2, 5 
Population of raindrops   
 
 
 Npop , 1  Npop , 2  Npop , 3  Npop , 4  Npop , 5 nNpop , 1 nNpop , 2 nNpop , 3 nNpop , 4 nNpop , 5 
Figure 4. The population representation

The objective function calculates the reliability of each raindrop as follows:

Ri  f (i1, i 2, i3, i 4, i5, ni1, ni 2, ni3, ni 4, ni 5) i  1,2,3,..., Npop (15)

In the searching process of the algorithm, streams and rivers may violate the considered constraints of the problem. In
this paper, a penalty function is added to the objective function in order to remove the infeasible solutions. By applying
this penalty, the feasibility of the final solution is guaranteed. The formulation of the modified fitness function is as
follows:

(16)
fitnessfunction   f (i1, i 2, i3, i 4, i5, ni1, ni 2, ni3, ni 4, ni5)   (i1, i 2, i3, i 4, i5, ni1, ni 2, ni3, ni 4, ni5) i  Npop
The water cycle algorithm is originally designed to consider the minimum value of cost function as the best solution. For
this reason, a sign (-) is added before the objective function in order to find the maximum optimum. In this case, if one

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 223

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

of the streams or rivers uses more resources than the upper bound of the resource limitations, the term
 (i1, i 2, i 3, i 4, i 5, ni1, ni 2, ni 3, ni 4, ni 5) adds a proper penalty to the fitness function. The penalty function is calculated as
follows:

3 g j (r, n) (17)
 (r1 , r2 ,..., rm , n1 , n2 ,..., nm )  i .max(0,  1)
i 1 lj

g j (r, n)
In the present study, the violation of a constraint such as g j (r, n)  l j is formulated as  1 and i is a number
lj
greater than 1, which is used to remove the infeasible streams or rivers. The feasibility of the final solution (i.e. sea) is
ensured by this approach.

For the first population, the maximum values of raindrops are considered as rivers and a sea. Nsr shows the number of
individuals which are chosen as sea and rivers. It is formulated in Eq. (18). The number of remained raindrops is
calculated by using Eq. (19).

Nsr= Number of Rivers + 1 (18)

NRaindrops=Npop- Nsr (19)

By using Eq. (20), the number of streams which are assigned to the specific rivers and sea is calculated.
 
 R (n) 
 
NSn  round  Nsr  NRaindrops  , n  1, 2,..., Nsr (20)
  R (i ) 

 i 1 

where NSn is the number of streams which flow to the rivers or sea based on the intensity of the flow (Eskandar et al.,
2012).

3.2. Stream flow

By joining the streams to each other, new rivers are created, and some streams can flow directly to the sea, which is the
best individual. A schematic view of stream’s flow is illustrated in Fig. 5.

Figure 5. Simplified diagram of the stream’s flow to a specific river

The stream needs a route to find its way towards a specified river. Therefore, a flow direction helps the stream to reach
a new position as follows:

X (0, C  d ), 1  C  2 (21)

Where C is a constant value between 1 and 2, and d is the current distance between the stream and river. The value of X
is the flow direction.

This concept is also used in the process of flowing rivers to sea. Hence, after obtaining a new route, the new positions of
stream and river are derived as following equations:

Xi+1 stream= Xi Stream + rand × C × (XiRiver-XiStream) (22)

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 224

 Juybari, Abouei Ardakan and Davari-Ardakani

Xi+1 River= Xi River + rand × C × (XiSea-XiRiver) (23)

Where a uniformly distributed random number in interval [0,1] is represented by rand. The positions of the stream and
the river are exchanged if the solution of the stream is better than the river. Such exchange may happen for a river and a
sea (Eskandar et al., 2012).

3.3. Evaporation condition

In order to prevent the algorithm from rapid convergence, evaporation is considered in the algorithm procedure. As
depicted in Fig. 3, water which is evaporated from rivers and lakes finally converts into clouds and raindrops. The rains
create new streams which flow to rivers or directly to the sea. By inspiring this concept, the proposed method avoids
being trapped in local optimum solutions. If the distance between a specified river and a sea is less than a constant number
(i.e. dmax), then the river joins the sea. In this situation, evaporation and raining processes occur. The following pseudo-
code shows the condition of evaporation and raining processes (Eskandar et al., 2012).

Algorithm 1. Pseudo-code of evaporation and raining processes

1: If |XiSea-XiRiver| < dmax
2: Start evaporation and raining processes
3: End if
Figure 6. Pseudo-code of evaporation and raining condition

The value of dmax is decreased iteratively, as Eq. (24):

di max
di+1 max = di max - (24)
max iteration

3.4. Raining process

After applying the evaporation process, the raining process begins. In this state, new raindrops form streams in various
locations. The following equation specifies the locations of new streams (Eskandar et al., 2012):

Xnew Stream = LB + rand × (UB - LB) (25)

Where LB and UB are lower and upper bounds of the defined problem, respectively. Again, the best raindrop joins the
sea while the rest of the raindrops form streams and rivers.

In order to improve the convergence rate of the algorithm, Eq. (26) is applied to those streams which straightly flow to
the sea. Furthermore, to enhance the exploration near the sea region (the best solution), this equation is formulated and
used as follows:

Xnew Stream = X Sea + √𝜇 × randn(1,Nvar) (26)

Where parameter 𝜇 controls the range of searching area near the sea and randn stands for a normally distributed random
number. In this problem, the number of variables (Nvar) is 10. As well as the RRAP is a constrained problem, Eq. (26)
improves the computational performance of the algorithm to search more in the feasible region.

3.5. Steps of the adjusted WCA

The pseudo-code of the penalty guided water cycle algorithm, applied to solving the RRAP is presented in algorithm 2,
followed by the flowchart as in Fig. 8 (Eskandar et al., 2012).

Algorithm 2. Pseudo-code of the adjusted water cycle algorithm

1: Set the initial parameters of WCA: Npop, Nsr, max_iteration and dmax
2: generate initial population by Eqs. (14), (18) and (19)
3: When the number of iterations is less than max_iteration,:
4: evaluate the fitness function of raindrops by Eq. (16)
5: calculate intensity of flow by Eq. (20)
6: flow the streams to the rivers by Eq. (22)
7: flow the rivers to the sea by Eq. (23)
8: replace the position of a stream with a river which gives a better solution.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 225

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

9: replace the position of a river with a sea which gives a better solution.
10: check the evaporation condition based on algorithm 1.
11: if the evaporation condition is met, then create clouds and perform the raining process by Eqs. (25) and (26)
12: update the value of dmax by Eq. (24)
13: End while
14: Print the final obtained results
Figure 7. Pseudo-code of the adjusted water cycle algorithm

4. Experimental results

In this section, the experimental results obtained from implementing the proposed WCA are presented for the two
benchmark problems described in Section 2. The proposed water cycle algorithm was coded in MATLAB software on
an Intel Core i7 with 4 GB of RAM on a personal computer. Taguchi method, one of the most applicable designs of
experiments, is employed to tune the initial WCA factors including Npop, Nsr, max_iteration and dmax. In this paper,
each factor is considered a 3-level parameter. Table 1 shows the input values of the Taguchi method, utilized for the
proposed WCA.

Table 1. Input values of the Taguchi method

Factor 1 2 3
Level
1 Npop 500 1000 1500
2 Nsr 4 5 6
3 max_iteration 30 40 50
4 dmax 1e-15 1e-10 1e-5

Main effects for means of each factor indicate the influence of the parameter on the total reliability at different levels, as
plotted in Fig. 9. The highest value represents the best level for each factor.

As a result, the following parameters are set in the proposed WCA: Npop=1500, Nsr=5, max_iteration= 30 and dmax=1e-
10. Same as other studies (Ardakan & Hamadani, 2014; Coit, 2001; Tavakkoli-Moghaddam et al., 2008), the switch is
assumed to be imperfect and its reliability is considered to be 0.99. In order to obtain desired results of the proposed
algorithm, each benchmark problem was solved independently five times. Table 2 and 3 present the input parameters of
the benchmark P1 (Series system) and P2 (Series-Parallel system), respectively.

Table 2. Input parameters of benchmark P1 (Series system)

Stage 105.i i w .v 2
i i
wi V C W
1 2.330 1.5 1 7 110 175 200
2 1.450 1.5 2 8
3 0.541 1.5 3 8
4 8.050 1.5 4 6
5 1.950 1.5 2 9

Table 3. Input parameters of benchmark P2 (Series-Parallel system)

Stage 105. i
i w .v 2i
wii
V C W
1 2.500 1.5 2 3.5 180 175 100
2 1.450 1.5 4 4.0
3 0.541 1.5 5 4.0
4 0.541 1.5 8 3.5
5 2.100 1.5 4 4.5

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 226

 Juybari, Abouei Ardakan and Davari-Ardakani

Figure 8. Flowchart diagram of the adjusted water cycle algorithm

Figure 9. Main effects plot for means

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 227

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

The simulation results of all five runs of the proposed WCA for the test problems P1 and P2 are represented in Table 4
and 5, respectively. Also Tables 4 and 5 report the STD of the fitness function in different five runs.

Table 4. Different runs of WCA for test problem P1 (Series system)

Parameter Run #1 Run #2 Run #3 Run #4 Run #5 STD
Rs 0.96957819 0.96957873 0.96817799 0.96957812 0.96957858 0.00056
i 0.00026535 0.00026720 0.00026993 0.00026757 0.00026718
0.00011931 0.00011941 0.00023774 0.00011953 0.00011930
0.00008923 0.00008857 0.00008920 0.00008836 0.00008884
0.00036702 0.00036627 0.00036952 0.00036674 0.00036644
0.00025285 0.00025355 0.00013288 0.00025259 0.00025298
n (3, 2, 2, 3, 3) (3, 2, 2, 3, 3) (3, 3, 2, 3, 2) (3, 2, 2, 3, 3) (3, 2, 2, 3, 3)
r 0.76693891 0.76552336 0.76343354 0.76523618 0.76553481
0.88752974 0.88743978 0.78841410 0.88733991 0.88754511
0.91463445 0.91523597 0.91466407 0.91543219 0.91499220
0.69279863 0.69331569 0.69107159 0.69299238 0.69319857
0.77658645 0.77604143 0.87557043 0.77678395 0.77648630
Slack (g1) 27 27 27 27 27
Slack (g2) 0.00005083 0.00000051 0.00000002 0.00000003 0.00021772
Slack (g3) 7.51891824 7.51891824 10.5724757 7.51891824 7.51891824

Table 5. Different runs of WCA for test problem P2 (Series-Parallel system)

Parameter Run #1 Run #2 Run #3 Run #4 Run #5 STD
Rs 0.99998828 0.99998828 0.99998827 0.99998828 0.99998828 4.0e-9
i 0.00019138 0.00019160 0.00019276 0.00019114 0.00019202
0.00016465 0.00016306 0.00016602 0.00016391 0.00016510
0.00010657 0.00009639 0.00009682 0.00009606 0.00009603
0.00009648 0.00010742 0.00010513 0.00010734 0.00010598
0.00014830 0.00014885 0.00014694 0.00014887 0.00014806
n (3, 3, 1, 2, 3) (3, 3, 2, 1, 3) (3, 3, 2, 1, 3) (3, 3, 2, 1, 3) (3, 3, 2, 1, 3)
r 0.82582078 0.82564109 0.82467802 0.82601610 0.82528969
0.84818976 0.84954379 0.84702574 0.84881753 0.84781158
0.89891322 0.90811131 0.90772127 0.90840989 0.90843699
0.90802848 0.89814457 0.90020768 0.89821940 0.89943930
0.86217186 0.86169644 0.86334976 0.86167768 0.86237954
Slack (g1) 53 62 62 62 62
Slack (g2) 0.00013618 0.00057513 0.00119624 0.00000193 0.00005565
Slack (g3) 7.11084884 6.10414028 6.10414028 6.10414028 6.10414028

Where i shows the failure rate of components used in subsystem i when the time-to-failure of components are
exponentially distributed. ni represents the total number of components in subsystem i, and ri is the reliability of each
component in subsystem i at mission time (1000 hour) which is calculated based on i . Finally, the slack parameters
imply unused resources for each constraint. The best obtained structure for each benchmark is illustrated in Fig. 10 and
11.

Figure 10. The best structure of Series system (P1)

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 228

 Juybari, Abouei Ardakan and Davari-Ardakani

Figure 11. The best structure of Series-Parallel system (P2)

The improving trends of WCA on the RRAP for benchmarks P1 and P2 are illustrated in Fig. 12 and 13, respectively.
Accordingly, the tuned algorithm tries to meet the optimum value in the first 30 iterations, which indicates the fast
convergence rate of the adjusted WCA.

0/975
0/970
Total Reliability

0/965
0/960
0/955
0/950
0/945
0/940
0 5 10 15 20 25 30 35
Number of generation

Figure 12. Plot of the adjusted water cycle algorithm in solving test problem 1 (Series system)

0/999990

0/999985
Total Reliability

0/999980

0/999975

0/999970

0/999965
0 5 10 15 20 25 30 35
Number of generation

Figure 13. Plot of the adjusted water cycle algorithm in solving test problem 2 (Series-Parallel system)

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 229

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

The best obtained result among five different runs is selected to be compared with those reported in the literature. The
results of comparing the obtained solutions with those reported in previous studies are reported in Tables 6 and 7. The
Maximum Possible Improvement (MPI) index is used in order to measure the impact of the improvements achieved by
using the adjusted WCA. The MPI is calculated as follows (Ardakan & Rezvan, 2017; dos Santos Coelho, 2009; Mellal
& Zio, 2016; Yeh & Hsieh, 2011):

MPI (%)  [Rs ( New Approach)  Rs (Other )] / [1  Rs (Other)] (27)

Where Rs (New Approach) is the total reliability obtained from the proposed solution method in this study, and Rs (Other)
shows the total reliability reported in other similar RRAP studies.

Based on the MPI values reported in Tables 6 and 7, it is clear that the proposed WCA outperforms all those in the
previous studies. The advantage of the standby strategy over the active one is also revealed. More importantly, it can be
seen that compared to a recent study by Ardakan & Hamadani (2014a), the proposed water cycle algorithm obtained
better solutions in the standby structure. More specifically, in comparison with the solutions reported in Ardakan &
Hamadani (2014a) for Series and Series-Parallel benchmarks, the MPI index values of the study show improvements by
0.0032% and 0.2638%, respectively. These remarkable improvements reveal the superior performance and robustness
of the adjusted water cycle algorithm in dealing with the RRAP problems.

Table 6. Comparison of best results obtained by the adjusted water cycle algorithm with those mentioned in the literature for Series
system (P1)
Cold-standby
Active strategy
strategy

(Ardakan &
(Y-C Hsieh

Hamadani,
(Afonso et

Zio, 2016)
(Wu et al.,

This study
(Valian &
Parameter

(Mellal &
al., 2013)
(Yi-Chih

Chen, &
Bricker,

(Yeh &
& You,

Valian,

2014a)
(Chen,
Hsieh,

Hsieh,
1998)

2006)

2011)

2011)

2011)

2013)

0.93157800 0.931 0.93168 0.931 0.93168 0.93168 0.931 0.93168 0.96957 0.96957
Rs
67800 234 68000 200 239 67939 239 758 858
(3, 2, 2, 3, (3, 2, (3, 2, 2, (3, 2, (3, 2, 2, (3, 2, 2, (3, 2, (3, 2, 2, (3, 2, 2, (3, 2, 2,
n 3) 2, 3, 3, 3) 2, 3, 3, 3) 3, 3) 2, 3, 3, 3) 3, 3) 3, 3)
3) 3) 3)
0.77942700 0.779 0.77946 0.780 0.77939 0.77941 0.779 0.77939 0.76459 0.76553
r
26600 230 37307 900 694 87400 888 335 481
0.86948200 0.872 0.87188 0.871 0.87183 0.87183 0.872 0.87183 0.88752 0.88754
51300 346 78343 700 328 05700 701 892 511
0.90267400 0.902 0.90280 0.902 0.90288 0.90288 0.903 0.90288 0.91539 0.91499
63400 088 40890 500 508 42600 536 527 220
0.71403800 0.710 0.71135 0.711 0.71140 0.71139 0.710 0.71140 0.69350 0.69319
64800 017 47356 300 387 96000 252 544 857
0.78689600 0.788 0.78786 0.787 0.78780 0.78780 0.786 0.78779 0.77603 0.77648
40600 159 38760 000 371 90200 948 145 630
55.5385 55.47 55.4706 55.47 55.4709 55.4706 55.47 55.4706 0.0032 ---
MPI(%)
34 22 25
Slack (g1) 27 27 27 27 27 27 27 27 27 27
0.12145400 0.001 0.00000 0.000 - 0.00000 0.000 0.00000 0.00002 0.00021
Slack (g2) 55900 053 10100 0.00021 027 09900 000 478 772
84
7.51891800 7.518 7.51891 7.518 7.51891 7.51891 7.518 7.51891 7.51891 7.51891
Slack (g3)
91800 800 91800 820 824 91800 824 824 824

5. Concluding remarks

In this paper, one of the major problems in the field of reliability optimization problem called reliability-redundancy
allocation problem (RRAP) was considered. The RRAP aims to maximize the system reliability by selecting appropriate
components reliability and suitable redundancy level, subject to some resource limitations. Unlike most recent studies
which used the active redundancy strategy, in this research the cold standby strategy was chosen for redundant
components. For solving the cold standby RRAPs, the present study used an adjusted water cycle algorithm. In order to
check the validity of the proposed solution method, its performance was compared with those algorithms reported in
other similar studies. For future work, implementing heterogeneous components in the benchmark problems can be an
interesting idea. Furthermore, the proposed algorithm might be applied to handle other reliability optimization problems.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 230

 Juybari, Abouei Ardakan and Davari-Ardakani

Table 7. Comparison of best results obtained by the adjusted water cycle algorithm with those mentioned in the literature for Series-
Parallel system (P2)
Active strategy Cold-standby strategy

Hsieh &

Hamada
(Afonso

(Ardaka
Paramet

(Yeh &

(Valian

(Mellal
Valian,

2014a)
(Wu et

(Wang
(Chen,

& Zio,
Hsieh,
2006)

2011)

2011)

2011)

2013)

2012)

2013)

2016)
et al.,

study
& Li,
(Y-C

You,

This
n&
al.,

ni,
&
er

Rs 0.99997 0.99997 0.9999766 0.9999766 0.9999766 0.99997 0.99997 0.9999766 0.9999882490 0.999
65800 73100 400 490 490 66500 66100 491 98828
n (2,2,2,2, (2, 2, 2, (2, 2, 2, 2, (2, 2, 2, 2, (2, 2, 2, 2, (2, 2, 2, (2, 2, 2, (2,2,2,2,4) (3,3,2,1,3) (3,3,1,
4) 2, 4) 4) 4) 4) 2, 4) 2, 4) 2,3)
r 0.81248 0.81974 0.8191852 0.8195915 0.8199270 0.81959 0.82201 0.8196593 0.82484673 0.825
500 570 6 6 9 600 264 9 82078
0.84315 0.84500 0.8436642 0.8449510 0.8452676 0.84500 0.84365 0.8449808 0.84281657 0.848
500 800 1 7 6 000 640 5 18976
0.89738 0.89545 0.8947299 0.8954285 0.8954915 0.89551 0.89129 0.8955064 0.90817308 0.898
500 810 2 5 5 400 092 3 91322
0.89451 0.90090 0.8953762 0.8955223 0.8954406 0.89551 0.89869 0.8955064 0.89869900 0.908
600 320 8 4 9 900 886 5 02848
0.87059 0.86840 0.8691272 0.8684902 0.8683187 0.86845 0.86824 0.8684476 0.86546301 0.862
000 690 4 3 8 600 939 9 17186
MPI 49.9573 48.3472 49.8288 49.8094 49.8094 49.8072 49.8931 49.8092 0.2638 ---
(%) 9
Slac 40 40 40 40 40 40 40 40 62 53
k
(g1)
Slac 0.00262 - 0.0005610 0.0000000 0.0000161 0.00000 0.00039 0.0000000 0.00006415 0.000
k 700 1.46952 0 0 0 700 600 0 13618
(g2) 200
Slac 1.60928 1.60928 1.6092890 1.6092890 1.6092890 1.60928 1.60928 1.6092889 6.10414028 7.110
k 900 900 0 0 0 900 900 7 84884
(g3)

References

Abouei Ardakan, M., and Rezvan, M. T. (2018). Multi-objective optimization of reliability–redundancy allocation
problem with cold-standby strategy using NSGA-II. Reliability Engineering & System Safety, Vol. 172, pp. 225–238.

Abouei Ardakan, M., Sima, M., Zeinal Hamadani, A., and Coit, D. W. (2016). A novel strategy for redundant components
in reliability--redundancy allocation problems. IIE Transactions, Vol. 48(11), pp. 1043–1057.

Afonso, L. D., Mariani, V. C., and dos Santos Coelho, L. (2013). Modified imperialist competitive algorithm based on
attraction and repulsion concepts for reliability-redundancy optimization. Expert Systems with Applications, Vol. 40(9),
pp. 3794–3802.

Ardakan, M. A., and Hamadani, A. Z. (2014a). Reliability–redundancy allocation problem with cold-standby redundancy
strategy. Simulation Modelling Practice and Theory, Vol. 42, pp. 107–118.

Ardakan, M. A., and Hamadani, A. Z. (2014b). Reliability optimization of series–parallel systems with mixed
redundancy strategy in subsystems. Reliability Engineering & System Safety, Vol. 130, pp. 132–139.

Ardakan, M. A., Hamadani, A. Z., and Alinaghian, M. (2015). Optimizing bi-objective redundancy allocation problem
with a mixed redundancy strategy. ISA Transactions, Vol. 55, pp. 116–128.

Chen, T.-C. (2006). IAs based approach for reliability redundancy allocation problems. Applied Mathematics and
Computation, Vol. 182(2), pp. 1556–1567.

COIT, D. W. (2001). Cold-standby redundancy optimization for nonrepairable systems. IIE Transactions, Vol. 33(6),
pp. 471–478.

Dhingra, A. K. (1992). Optimal apportionment of reliability and redundancy in series systems under multiple objectives.
IEEE Transactions on Reliability, Vol. 41(4), pp. 576–582.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 231

 An Adjusted Water Cycle Algorithm for Solving Reliability-redundancy Allocation ...

dos Santos Coelho, L. (2009). An efficient particle swarm approach for mixed-integer programming in reliability–
redundancy optimization applications. Reliability Engineering & System Safety, Vol. 94(4), pp. 830–837.

Elsayed, E. A. (2012). Reliability engineering (Vol. 88). John Wiley & Sons.

Eskandar, H., Sadollah, A., Bahreininejad, A., and Hamdi, M. (2012). Water cycle algorithm – A novel metaheuristic
optimization method for solving constrained engineering optimization problems. Computers & Structures, Vol. 110–
111, 151–166.

Habib, A., Alsieidi, R., and Youssef, G. (2009). Reliability analysis of a consecutive r-out-of-n: F system based on neural
networks. Chaos, Solitons & Fractals, Vol. 39(2), pp. 610–624.

Hikita, M., Nakagawa, Y., Nakashima, K., and Narihisa, H. (1992). Reliability optimization of systems by a surrogate-
constraints algorithm. IEEE Transactions on Reliability, Vol. 41(3), pp. 473–480.

Hsieh, T.-J., and Yeh, W.-C. (2012). Penalty guided bees search for redundancy allocation problems with a mix of
components in series–parallel systems. Computers & Operations Research, Vol. 39(11), pp. 2688–2704.

Hsieh, Y.-C., Chen, T.-C., and Bricker, D. L. (1998). Genetic algorithms for reliability design problems.
Microelectronics Reliability, Vol. 38(10), pp. 1599–1605.

Hsieh, Y.-C., and You, P.-S. (2011). An effective immune based two-phase approach for the optimal reliability–
redundancy allocation problem. Applied Mathematics and Computation, Vol. 218(4), pp. 1297–1307.

Kim, H. (2018). Maximization of system reliability with the consideration of component sequencing. Reliability
Engineering & System Safety, Vol. 170(Supplement C), pp. 64–72.

Kuo, W. (2001). Optimal reliability design: fundamentals and applications. Cambridge university press.

Kuo, W., Lin, H.-H., Xu, Z., & Zhang, W. (1987). Reliability optimization with the Lagrange-multiplier and branch-
and-bound technique. IEEE Transactions on Reliability, Vol. 36(5), pp. 624–630.

Liang, Y.-C., and Smith, A. E. (2004). An ant colony optimization algorithm for the redundancy allocation problem
(RAP). IEEE Transactions on Reliability, Vol. 53(3), pp. 417–423.

Mellal, M. A., and Zio, E. (2016). A penalty guided stochastic fractal search approach for system reliability optimization.
Reliability Engineering & System Safety, Vol. 152(Supplement C), pp. 213–227.

Nahas, N., and Nourelfath, M. (2005). Ant system for reliability optimization of a series system with multiple-choice
and budget constraints. Reliability Engineering & System Safety, Vol. 87(1), pp. 1–12.

Peiravi, A., Karbasian, M., and Abouei Ardakan, M. (2017). K-mixed strategy: A new redundancy strategy for reliability
problems. Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability,
1748006X17736166. Article in press.

Ramirez-Marquez, J. E., Coit, D. W., and Konak, A. (2004). Redundancy allocation for series-parallel systems using a
max-min approach. Iie Transactions, Vol. 36(9), pp. 891–898.

Tavakkoli-Moghaddam, R., Safari, J., and Sassani, F. (2008). Reliability optimization of series-parallel systems with a
choice of redundancy strategies using a genetic algorithm. Reliability Engineering & System Safety, Vol. 93(4), pp. 550–
556.

Valian, E., Tavakoli, S., Mohanna, S., and Haghi, A. (2013). Improved cuckoo search for reliability optimization
problems. Computers & Industrial Engineering, Vol. 64(1), pp. 459–468.

Valian, E., and Valian, E. (2013). A cuckoo search algorithm by Lévy flights for solving reliability redundancy allocation
problems. Engineering Optimization, Vol. 45(11), pp. 1273–1286.

Wang, L., and Li, L. (2012). A coevolutionary differential evolution with harmony search for reliability–redundancy
optimization. Expert Systems with Applications, Vol. 39(5), pp. 5271–5278.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 232

 Juybari, Abouei Ardakan and Davari-Ardakani

Wu, P., Gao, L., Zou, D., and Li, S. (2011). An improved particle swarm optimization algorithm for reliability problems.
ISA Transactions, Vol. 50(1), pp. 71–81.

Yeh, W.-C., and Hsieh, T.-J. (2011). Solving reliability redundancy allocation problems using an artificial bee colony
algorithm. Computers & Operations Research, Vol. 38(11), pp. 1465–1473.

Zou, D., Gao, L., Li, S., and Wu, J. (2011). An effective global harmony search algorithm for reliability problems. Expert
Systems with Applications, Vol. 38(4), pp. 4642–4648.

Int J Supply Oper Manage (IJSOM), Vol.5, No.3 233