Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In recent years, due to environmental concerns, remanufacturing of products is practiced in different
Received 25 July 2016 companies and closed loop supply chain network in these companies is signicant to optimize. There-
Received in revised form fore, current study is aimed to determine an optimal closed loop supply chain network, which is
7 June 2017
composed of multiple producers, remanufacturers, intermediate centers and customer centers.
Accepted 9 June 2017
Available online 10 June 2017
Furthermore, uncertainty in the demand and uncertainty in the quantity of returned products is
considered simultaneously in the network to make it signicantly useful in the uncertain environment. A
novel genetic articial bee colony (GABC) algorithm is introduced with a new food source representation
Keywords:
Close loop supply chain network
for the current problem. The proposed GABC algorithm considered neighbor food sources for local search
Remanufacturing and used crossover and mutation operations of genetic algorithm to enhance the exploration ability of
Demand uncertainty the proposed algorithm. Taguchi method is employed to compute the optimal parameters of GABC for
Genetic articial bee colony algorithm two different size test problems taken from literature and a Case problem which are modied according
to the current research problem. The performance of presented GABC algorithm is tested by comparing
the results of considered test problems with the results obtained from the original articial bee colony
(ABC) algorithm and genetic algorithm (GA). Moreover, to test the robustness of the proposed GABC
algorithm, different scenarios of small and large size problems based on the quantity of demand and
variations in demand are made to perform the experiments. Results indicate that proposed GABC out-
performs standard ABC and GA in different scenarios to give smaller value of the total cost of network
and gives more robust results to give smaller variations in the total cost of network due to uncertain
variations in the demand, as compared to original ABC and GA.d
2017 Elsevier Ltd. All rights reserved.
1. Introduction loop, used products are collected from customers and moved to
intermediate centers for cleaning and disassembly operations. Later
Close loop supply chain optimization problem is gaining popu- they are moved to the remanufacturing facilities for remanu-
larity in recent years due to environmental concerns of different facturing. Close loop of supply chain is signicant to reduce envi-
manufacturing industries. Close loop supply chain consists of for- ronmental pollution due to reuse of scrap and used products from
ward and reverse loop of supply chain, which works simulta- customers. Many companies now a days are using reverse ow of
neously. In the forward loop, the newly manufactured products are their used products from customers to recovery centers. The re-
moved to the customers in different areas. While in the reverse covery options of the used products includes reuse of the collected
products, repairing of the returned items, remanufacturing or the
combination of these options (Thierry et al., 1993; Zhiqiang and
Nathalie, 2007). The reverse supply chain network system can be
* Corresponding author. State Key Lab of Digital Manufacturing Equipment and
Technology, HUST-SANY Joint Lab of Advanced Manufacturing, Huazhong Univer-
designed for direct reusable network, repairing network, recycling
sity of Science and Technology, Wuhan, 430074, PR China; Department of Industrial network and remanufacturing network (Fleischmann et al., 1997;
Engineering, University of Engineering and Technology, Taxila, Pakistan. Thierry et al., 1993; Zhiqiang and Nathalie, 2007). In reusable
E-mail address: saifullah47@yahoo.com (U. Saif).
http://dx.doi.org/10.1016/j.jclepro.2017.06.079
0959-6526/ 2017 Elsevier Ltd. All rights reserved.
718 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
network, the products are directly reused, for example, soft drink has uncertain environment and the demand of the customers and
bottles, pallets or containers etc. The repairing logistics network quantity of the returned products is uncertain. The need to account
restores the failed products to their working conditions. For uncertainty in the supply chain network is essential to design
example, reconditioning of the domestic appliances including, closed loop supply chain network model in strategic view (Ozkir
washing machines, electric equipment, refrigerators etc, are and Basligil, 2013). The demand of the new products from
involved in the repairing network. In recycling process, material is different customer centers is varying and it is not constant in most
recovered without converting product structure, for example, steel of the real cases. Furthermore, the quantity of the returned prod-
scrap is used to manufacture different steel products, glass, papers ucts from different customer centers is also uncertain and varies
etc. In remanufacturing, a product is reformed into its original due to several factors. Therefore, it is noteworthy to consider un-
shape by several operations which can include disassembly, over- certain demand of new products and uncertainty in the quantity of
hauling cleaning and replacement operation etc. Remanufacturing returned products while making strategic decision on the closed
systems are used to remanufacture several kind of machineries like, loop supply chain network and therefore considered in the current
air craft, automobiles etc. Current research is focused on remanu- research.
facturing, as recovery option, for the used products in close loop The closed loop supply chain network problem is combinatorial
supply chain network. optimization problem and it can take long computational time if it
Closed loop supply chain network problem is the focus of many is solved with exact methods. Therefore, a novel metaheuristic
researchers due to the necessity of reducing the use of raw material, method which has not been used in literature to solve closed loop
controlling the environmental pollution, recycle management and network optimization problem is introduced in the current
due to the social responsibility (Zohal and Soleimani, 2016). The research. The performance of the proposed algorithm is tested on
supply chain network design is a strategic decision problem which different size problems under several scenarios of demand and
is aimed to decide number of different facilities required to make in returned products and the corresponding results are compared
the network. Furthermore, supply chain network optimization is with the performance of other algorithms.
important to identify the ideal location of these facilities in the The structure of the current research paper is as follows: Section
network and identify the effective connections of these facilities 2 shows the review of literature work on closed loop supply chain
with each other to satisfy the overall demand of the customers with problem. Section 3 shows a mathematical model for the considered
minimum total cost of the network. The location and position of the closed loop supply chain network problem. Section 4 illustrates
facilities in the network can affect, the investment cost, service proposed novel solution approach based on articial bee colony
level of the customers and has its much practical implication in the algorithm. Section 5 indicates design of experiments and tuning of
real remanufacturing industries now a days. Aras et al. (2007) parameters of the proposed algorithm on two different size prob-
determined an optimal locations of the collection centers and lems to get optimal values of algorithm parameters for these
maximize the purchase price of the products used in reverse lo- problems. Furthermore, it also shows the results of considered two
gistics network. Zohal and Soleimani (2016) presented a closed problems. Section 6 illustrate the implementation of the proposed
loop supply chain problem for a gold industry to minimize the cost GABC to solve the small and large size problems in detail with
and carbon emissions. Yi et al. (2016) presented a closed loop considering uncertainty in demand and in the returned products.
supply chain network design problem for a construction machinery Moreover, scenario planning is described in this section to test
remanufacturing company to minimize total cost of the network different category of problems based on different values of the
which includes, xed cost, transportation cost, processing cost, variations in demand and returned products and different ranges of
revenue cost and cost of new components. In closed loop supply the uncertain demand and returned products. Furthermore,
chain network, the location decision of all facilities signicantly different scenarios are considered in each category of test problem
affects the remanufacturing process. For example, the location and the results are explained in this section. Section 7 shows the
position of the customer centers need to be optimized to satisfy the conclusion of the current research and future recommendations.
demand of customers, located in the areas near the location posi-
tion of the customer centers. In addition, the location and number 2. Literature review
of intermediate facilities is substantial to know while making the
optimal network because in some areas, there is more population of In literature, closed loop supply chain network optimization
customers who are willing to return the used products to the problem has been investigated by several researchers. Fleischmann
company and in some other areas, the population of customers is et al. (1997) presented a comprehensive literature review on the
less prepared to return the used products for remanufacturing. application of mathematical modeling in reverse logistics man-
Therefore, the location decision of different facilities is signicant in agement and shows different models which have been presented in
the network design and optimization problem for different rema- literature. Krikke et al. (1999) proposed a mixed integer program-
nufacturing industries. Furthermore, there are some capacity con- ming model (MILP) for reverse logistics network for photocopier
straints related to each location position of the customer centers manufacturer and considered minimization of supply chain cost,
which are required to be considered while making the network. energy and residual waste in close loop supply chain. Uster et al.
The capacity of the customer centers can affect the service level of (2007) developed a closed loop supply chain network model to
the customers and affects the quantity of the returned products optimize the ow of products in the network. They used collection
that can be collected from different customers. In addition, there is centers, remanufacturing facilities in their proposed network. Their
transportation cost which can occur in the network and the study focused to minimize the processing cost, transportation and
transportation cost depends on the decision of the connections xed costs of facilities in the close loop supply chain network. Lu
between different facilities in the network. The closed loop supply and Bostel (2007) proposed a mixed integer programming model
chain network is concerned to determine the relationship between containing three types of facilities in the network. They considered
all the facilities in the closed loop supply chain, including customer customers, intermediate centers, remanufacturing centers and
centers, the collection centers, the manufacturers and re- producers in the close loop supply chain network. Du and Evans
manufacturers so that the network is optimized and it can satisfy (2008) studied close loop supply chain problem to minimize
the overall demand of different customers located in different overall cost and total tardiness. They supposed manufacturers,
areas. However, in real environment, the closed loop supply chain collection sites i.e., third party logistics, repairing facilities in the
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 719
closed loop network. Gupta and Evans (2009) presented a non- network composed of customer centers, intermediate centers,
preemptive goal programming approach to model closed loop producers and remanufacturers or a facility containing both pro-
supply chain network. They used, manufacturing cost, refurbishing ducer and remanufacturer. The location sites for any facility is not
cost, disassembly cost, testing cost, remanufacturing cost, disposal xed and each facility has possibility to be allocated to any site in
cost and transportation cost in their developed model. Fleischmann the network while making the optimal network. This can give more
et al. (2001) reported a generic model for the design of closed loop diversity and can give global solution of the network in the current
supply chain and observed that considerable savings can be made research.
in cost of production due to simultaneous forward and reverse Moreover, in literature, demand uncertainty is dealt with sto-
network integration. They included customer centers, warehouses chastic programming, or scenario planning approach and in real
and production plants in the close loop supply chain network. environment, the value of demand in different scenarios depends
Nevertheless, in most of the above mentioned literature, the on several factors which has not been included in previous re-
customer demand and returned quantity of used products is searches. The demand of new products and quantity of returned
considered as deterministic. However, in real environment, de- products varies randomly and it depends on different factors of the
mand and quantity of returned products is uncertain. The closed market. Furthermore, each factor has different weightage on
loop supply chain network which is optimal for a xed value of changing the demand and returned quantity in different facility
demand may not be suitable to use for different values of the de- locations. For example, demand of returned products in certain
mands. This may give different value of cost of transportation, areas depends on unexpected events occurring in market, uncer-
storage cost or may not deliver products in time to the customers tain changes in season, uncertain fall in the prices of competitors or
when the demand of new products and quantity of returned uncertain changes in the taxes or other price. These demand related
products in the network varies. Furthermore, the closed loop factors can cause uncertain variations in the demand of the prod-
network which is based on deterministic demand and returns, can ucts. Product demand uncertainty occurs due to several factors and
give local optimal solution of the network and it cannot be used for each factor has its own probability to occur. Furthermore, each
large range of uncertain values of the demand of products. There- factor has its own impact in changing the value of the demand.
fore, it is signicant to use uncertainty in the demand of new Similarly, there are several factors which can cause uncertain
products and in the quantity of returned products simultaneously change in the quantity of the returned products. For example, a
in closed loop supply chain network problem. In literature, few different product model is introduced by some other manufac-
studies has considered uncertainty in the demand and returned turers to the market, then there is possibility that customers may
products for different customers in closed loop supply chain return some products to the company and this may change the
network problem. For example, Salema et al. (2007) extended the quantity of returned products randomly. These factors have
model proposed by Fleischmann et al. (2001) and incorporated different weightage and each factor has different range to cause
uncertainty in demand and returned products using mixed integer change in quantity of the returned products. The uncertain demand
programming approach. They considered scenario planning of new products and uncertain quantity of the returned products
approach to consider the uncertainty in demand of products. has not been investigated in literature based on these factors.
Francas and Minner (2009) proposed two stage stochastic model to Moreover, the scenarios for demand and uncertain quantity of
design closed loop supply chain network considering uncertain returned products based on these factors has not been investigated
demand and returns. Shi et al. (2010) considered uncertain demand in literature. Current research is novel to include demand uncer-
and returns and proposed stochastic programming approach for tainty in new products and uncertainty in the quantity of the
closed loop supply chain network optimization. Pishvaee et al. returned products based on several factors for making different
(2011) considered uncertainty in demand in closed loop supply scenarios. Furthermore, in literature, holding cost in the network
chain and proposed a robust approach to deal with the uncertainty. has not been considered during optimizing the closed loop supply
Vahdani et al. (2012) applied fuzzy approach for robust optimiza- chain network (Salema et al., 2007; Pishvaee et al., 2011; Vahdani
tion of closed loop supply chain network optimization problem. et al., 2012; Zohal and Soleimani, 2016). However, the quantity of
They considered suppliers, manufacturing facilities, customers, returned products and quantity of products manufactured in the
some bidirectional facilities in their considered closed loop supply plants also incurs holding costs but it has not been investigated in
chain network. They aimed to minimize total cost which includes literature in closed loop supply chain network optimization prob-
the xed cost, transportation cost in forward and reverse ow in lems (Salema et al., 2007; Pishvaee et al., 2011). However, the
the network. Recently, Saif et al. (2015) presented a brief descrip- holding cost of returned products in the collection centers and the
tion of a model for facility location problem in closed loop supply holding cost of the new and remanufactured products in the
chain network problem but have not solved it. Saman and Zhang manufacturing facilities are required to be considered during
(2013) proposed a facility location model in closed loop supply optimization of the network and therefore considered in the cur-
chain network considering uncertainty in demand and returns. rent research. In addition, the safety stock of new products and
They considered stochastic programming and scenario planning returned products for remanufacturing in the remanufacturing
approach to include uncertainty in the demand and returned centers is required to be included and is considered in the current
products. They used multi plants, collection centers, demand study. Current research is novel to include holding cost in the in-
markets and products. They proposed a mixed integer program- termediate centers and remanufacturers in the closed loop supply
ming model to minimize total cost. However, they assumed a xed chain network. Moreover, safety stock of the returned products in
location of the plants and collection centers which is known in intermediate centers, safety stock of the products in the producers,
advance in their model. However, in closed loop supply chain safety stock of remanufactured products in the remanufacturing
network problems, the xed location for certain facilities can limit centers is rst time considered simultaneously in the current study.
the diversity of solution of network and cannot give global optimal In addition, several additional constraints in the closed loop supply
solution. In the network optimization problems, if the location sites chain network problem are used in the current research which
are not xed for the facilities, then it may give a network solution includes the safety stock constraints, storage capacity constraints
which can consider all possibilities of placing facilities in all loca- and service level constraints to store the remanufacturing products,
tion sites and can give more global search for network solution. returned products in the producers and remanufacturing centers.
Therefore, current research considers a closed loop supply chain In literature, different solution methods has been used to solve
720 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
closed loop supply chain problem. For example, Salema et al. (2009) level and optimal inventory levels. In addition, the quantity of the
presented a mixed integer programming model and proposed returned products from different customers is optimally used
branch and bound method to solve their presented problem. Other during remanufacturing.
methods which are used in literature include, nested diagonaliza-
tion solution algorithm (Chen et al. (2007), an OR opt algorithm 3. Problem description
procedure (Alshamrani et al., 2007), scatter search algorithm (Du
and Evans, 2008), two phase heuristic method based on greedy In this section the proposed facility location problem in closed
construction (Wollenweber, 2008), heuristic algorithm based on loop supply chain network is presented. In the considered closed
imperialist competitive algorithm and variable neighborhood loop supply chain network, four kinds of facilities are considered.
search (Devika et al., 2014; Fakhrzad and Moobed, 2010) etc. In These facilities includes customer centers to collect the used
addition to these methods, metaheuristic methods including ge- products from customers, intermediate centers which are used for
netic algorithm (Lee and Chan, 2009; Aydin et al., 2016; Yi et al., cleaning, disassembly, checking, and ranking the returned products
2016), genetic algorithm with priority based encoding method before they enter into the remanufacturing centers. The third kind
(Lee and Chan, 2009), Simulated annealing with dynamic neigh- of facility used in the proposed network is remanufacturing centers
borhood search (Pishvaee et al., 2010), Ant colony algorithm (Zohal where the returned products from intermediate centers are
and Soleimani, 2016) etc, have also been used to solve closed loop checked again and they are remanufactured. The fourth type of
supply chain network optimization problem. Furthermore, in facility considered in the current research problem is producers
literature there are some studies which used simulation method for which are used to make new products according to the demand of
closed loop supply chain network optimization problem (Tonanont products. The produced products and the remanufactured products
et al., 2009; Suryabatmaz et al., 2004). Recently, Karaboga (2005) are directly transported to the customer centers for sale. There are
introduced an algorithm called articial bee colony algorithm two transportation roots in the considered network problem. One
(ABC) which is based on the foraging behavior of bees is an excel- root is used to move returned products from customer centers to
lent algorithm. The original structure of ABC is useful for contin- intermediate centers and then to the remanufacturing centers or
uous problems. However, ABC algorithm has also been successfully producers, called reverse ow. The other root is used to move
applied in discrete nature of problems including, jobshop (Li et al., products from remanufacturing facilities and producers to the
2011), and owshop problems (Fatih Tasgetiren et al. (2011). ABC customer centers called forward ow. Fig. 1 illustrates closed loop
algorithm needs less control parameters and is easy to implement supply chain network with forward and reverse ows of products
and to the best of authors knowledge, it has not been used to solve between different facilities. The notations and abbreviations used
closed loop supply chain network optimization problem. Current for the current problem are described as under.
research is novel to present genetic articial bee colony (GABC) Indices
algorithm to solve the considered closed loop supply chain network C f1; 2; :::; i; ::::; nC g: Set of different locations of customer
optimization problem. centers
Current research is focused to develop a model for closed loop I f1; 2; ::::; j; ::::; nI g: Set of intermediate center
supply chain network considering uncertainty in demand and P f1; 2; ::; k; :::::; nR g: Set of different producers or remanu-
quantity of the returned products in an automobile manufacturing facturing centers at location kVi fvi1 ; vi2 ; :::::; viM g: Set of different
company. The main questions which current research seeks to factors which may cause variance in demand at customer center i
address is Parameters
Fixed Costs:
How to develop closed loop supply chain network model fkP : Fixed cost to setup a producer center at location kfkR : Fixed
considering uncertainty in demand and in the quantity of the cost to setup a remanufacturing center at location kfjI : Fixed cost to
returned products? setup an intermediate center at location jfiC : Fixed cost to setup
How to solve it using a novel metaheuristic algorithm? customer center at location i
Unit Costs:
Current research is aimed: UCkP : Unit cost at producer center at location kUCkR : Unit rema-
nufacturing cost at remanufacturing center at location kUCjI : Unit
To identify the optimal location of different facilities in the reprocessing cost at intermediate center at location j
closed loop supply chain network Shipping Costs:
To optimize and decide the connections between different fa- SCki : Shipping cost from producer or remanufacturing center k
cilities in the closed loop supply chain network. to customer center iSCij : Shipping cost from customer center i to
To consider production cost, transportation cost, holding cost of intermediate center at location jSCjk : Shipping cost from interme-
inventory in facilities, xed cost, capacity constraints of the fa- diate center j to producer or remanufacturing center k
cilities, the service level constraints and some additional con- Disposal Costs:
straints in the closed loop supply chain simultaneously. DCjI : Disposal cost at intermediate center jDCkR : Disposal cost at
To consider uncertainty in the demand of new products and remanufacturing center k
quantity of the returned products by incorporating change in Holding Costs:
demand and returned products considering different factors Hk : Holding cost of returned and produced products at rema-
which are causing uncertainty in making the scenarios nufacturing center or producer kHjI : Holding cost of returned pro-
simultaneously. duced products at intermediate center j
To develop a novel genetic articial bee colony algorithm to Demand:
di Ndi ; dsi : Normal distributed demand of the product at
m
search an optimal solution of the considered closed loop supply
customer center idri Ndi r ; disr : Normal distributed returned
m
chain network optimization problem.
m
products at customer center idi : Mean value of demand of the
In the optimized network, the total cost of investment including s
product at customer centeridi : Variance value of demand of the
the total cost of the network is minimized. Furthermore, the un- product at customer center i
certain demand of customers is also fullled with the given service Returns:
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 721
Produced or
Remanufactured
Products Flow
Disposed Products in
Intermediate Centers
m
di r : Mean value of quantity of the returned product at customer customer centeri taken through intermediate center j to remanu-
center idsi r : Variance value of quantity of the returned product at facturer at location k
m
customer center iedi : Estimated average value of demand of the Binary Variables:
m
product at customer center iedi r : Estimated average value of
quantity of the returned product at customer center i
1 if ;
Demand Variations: YPk Producer is at position k
bim : The weight for the factor m for customer center i which 0 Otherwise
1 if ;
causes variance in the demand YRk Remanufact uring center is at position k
Number of Facilities: 0 Otherwise
1 if ;
NjC : Maximum number of customers which are allowed to YIj Intermediate center is at position j
0 Otherwise
deliver their returned products to intermediate center jNkI : 1 if;
Maximum number of intermediate centers which are allowed to YCj Customer center is at position i
0 Otherwise
deliver their returned products to remanufacturing or producer at
locationkNkC : Maximum number of customers which are allowed to Consider there are known number of locations sites for making
get delivery from products or remanufacturing center k facilities in the closed loop supply chain network. From these
Service Level: location sites, there are known number of producers, re-
drP
k : Service level for the returned product at remanufacturing manufacturers, intermediate centers and customer centers which
center or producer kdPk : Service level for the product at remanu- are desired to be located on known number of site location points to
facturing center or producer k make a closed loop supply chain network. The number of site lo-
Disposals: cations is known and it is required to decide, at which location
b : Percentage at which returned products can be disposed of at point, which facility is to be build and with which facilities it can be
the intermediate centers (Suppose it is equal on all intermediate connected in the optimized network. The location decision of these
centers) facilities depends on the initial cost called, xed cost of these fa-
g : Percentage at which checked returned products can be cilities at different location sites. Moreover, the location decision of
disposed of at the remanufacturing centers (Suppose it is equal on facilities also depends on their connections with other facilities so
all remanufacturing centers) that the transportation cost between them is minimum. For
Safety Stock: example, the location of customer centers not only depends on its
SPk : Safety stock of the product at producer kSrP k
: Safety stock of xed cost, but it also depends on its possible connections with
the returned products at producer or remanufacturing center kSrI j : intermediate centers so that it can transport the returned products
Safety stock of the returned products at intermediate center i to the intermediate center with smaller transportation cost. It is
Capacity of Facilities: possible that a customer center can have possibility to get con-
SCapPk : Storage capacity of products at producer or remanu- nected with different intermediate centers. However, the decision
facturing center kSCaprP k
: Storage capacity of the returned products of its connection also depends on the distance of the intermediate
at producer kSCaprI j : Storage capacity of the returned products at center and its storage capacity so that it can economically stores the
intermediate center j returned products and the transportation cost is also desired to be
Decision variables minimum. Furthermore, there is possibility that the demand of the
Continuous variables: considered customer center might not be fullled with one
XPik : Fraction of demand from customer centeri to remanufac- manufacturer, so it can be connected with more than one pro-
turer or producer at location kXRijk : Fraction of returned products at ducers. The considered network problem is to assign optimal
722 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
" ! rm
X
nR X
nC X
M SrR 1 bdi
Prod:cos t UCkP
m
edi bim vim XPki YCi YPk ZkrR k
(8)
1 bdri s
k1 i1 m1
# 2
nR X
X nI X
nC
rm
SPk UCkR 4SrP
k 1 g1 b
SrI di
ZkrI k
(9)
k1 j1 i1 dri s
! 3
nI X
X nC X
M
bim vim XRijk YCi YIj 5
rm
edi
j1 i1 m1
X
nR X nC hn
nI X o 3.2. Constraints
1 bUCjI b:DCjI YIj
k1 j1 i1
n o i 3.2.1. Production and demand constraints
1 b1 gUCkR g1 b DCkR YPk The new products and returned products are required to fully
! meet their demand. Equation (10) shows that, for each customer
X
M
edi
mr
bim vim XRijk YCi center, the demand is collectively satised from all producers or
m1 remanufacturing centers connected with it. The constraint shown
in Equation (11) indicates that for each customer, the quantity of
(4)
the returned products are moved to all intermediate centers con-
nected with it which are then moved to all of the producers or
remanufacturing centers connected with each intermediate
3.1.4. Holding cost centers.
Holding cost or the storage cost is incurred due to inventory of
P
nR
the returned products in the remanufacturing centers and the in- XPik ; 1 ci (10)
ventory in the producers. It is the sum of total holding cost of the k1
new products in the producers, holding cost of returned products in
the producers, holding cost of returned products in the remanu- P
nR P
nI
XRijk 1; ci (11)
facturing centers and holding cost of the returned products in the k1 j1
intermediate centers, as illustrated in Equation (5). Equation (6) to
Equation (9) are used to calculate the probability to ensure that the The constraint shown in Equation (12) shows that the sum of
safety stock levels of new products in producers, safety stock of safety stock of new products, safety stock of the returned products
remanufactured products in the producers, safety stock of rema- in remanufacturing centers and quantity of the returned products
nufactured products in the remanufacturing canters and safety entering in the remanufacturing centers should meet the total
stock level of returned products in the intermediate centers can demand of new products.
satisfy their corresponding demands in the network, respectively.
(5)
!
X
nI X
nC
rm
X
M
SPk SrP
k 1 g1 b edi bim vim XRijk YCi YIj
j1 i1 m1
m
SPk di !
ZkP (6) X
nC X
M
dsi edi
m
bim vim Xik YCi YPk ; ck
i1 m1
rm (12)
SrP 1 bdi
ZkrP k
(7)
1 bdri s The difference between the net quantities of the new products
required with the net quantity of the returned products can give
724 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
80 ! 1 ! ! 9
< X
nC X
nI X
M X
nC X
M =
ProbrP @SrP bim vim XRijk A
r m m rP
1 g1 b edi edi bim vim Xik 0 dk ; ck (22)
: k ;
i1 j1 m1 i1 m1
center is responsible to deliver their products to some predened 4.2. Genetic articial bee colony algorithm
number of customer centers. These constraints are dened in
Equation (23) to Equation (27). In the original structure of ABC algorithm, each employee bee is
assigned only one food source which can be signicant to explore
P
nC the number of solutions simultaneously in the employee bee phase.
YCi YIj NjC ; cj (23)
i1
However, this exploration can further be improved with incorpo-
rating local search scheme for each employee bee. Due to the
complex nature of the current problem, the original ABC is modi-
P
nI
YIj YRk NkI ; ck (24) ed here by adding some steps of genetic algorithm in different
j1 phases of ABC algorithm and it is named as genetic articial bee
colony algorithm (GABC). The proposed GABC is designed for the
P
nI current close loop supply chain network optimization problem and
YIj YPk NkI ; ck (25) it is explained in this section.
j1
P
nC 4.2.1. Encoding scheme for solution representation
YRk YCi NkC ; ck (26) The encoding scheme introduced for the current problem is
i1
indicated in Fig. 2. The solution representation of a food source is
presented in Fig. 2 which gives the complete information of the
P
nC
location sites for customer centers from the known number of total
YPk YCi NkC ; ck (27)
i1 site locations. Furthermore, the food source also explains the sites
allocated for intermediate centers, manufacturers or producers.
Moreover, the network information to identify the connection of
customer centers with intermediate centers, connection of inter-
mediate centers with remanufacturers or producers and the
4. Method connection of producers and remanufacturers with different cus-
tomers is also clearly indicated in Fig. 2. As an example, total 10 site
4.1. Articial bee colony algorithm locations are available from which 6 customer centers, two inter-
mediate centers and two locations are used for remanufacturing
Articial bee colony (ABC) algorithm is rst time proposed be centers, or producers or both can be made on a single location. It
Karaboga (2005). This algorithm is inspired by foraging behavior of can be seen from Fig. 2 that there are two rows in the food source
honey bees. The algorithm is used to explore good quality solutions vector. The numbers in the rst 6 elements of the food source
which are represented as a food sources for the employee bees. The vector in the rst row indicate the location points to which the
nectar amount of a food source describes the quality of solution customer centers are assigned. For example, the numbers in the
which can be considered as the tness of the solution candidate. rst 6 elements of the rst row of food source are 1, 10, 5, 2, 6 and 9.
Three kinds of bees are used in this algorithm. The rst half of the These numbers describes the location site from the available loca-
population are employee bees and remaining half is considered as tion sites which are allocated for customer centers. The numbers in
onlooker bees. In the rst stage of ABC algorithm, employee bees the next two elements (7 and 8 element) in the rst row describe
are sent to the food sources and each employee bee is assigned one the location sites assigned for intermediate centers, i.e., 4 and 7 are
food source. Employee bees nd out the nectar amount of the food allocated to intermediate centers. Furthermore, the numbers in
sources assigned to them. The onlooker bees wait in the hive and element 9 and 10 in the rst row of the food source vector shows
get the signicant information about the food sources with better the location sites assigned to the remanufacturer centers or pro-
nectar amounts from the employee bees. Onlooker bees can decide ducers. The remaining 6 elements in the rst row shows the 6
the employee bees to move on which direction for more food customer centers. The second row of the food source vector in-
source search. This is decided by onlooker bees on the basis of dicates the information of the connections between different fa-
nectar amount of the food source. The employee bees which have cilities. For example, the numbers in the rst 6 elements in the
no more improvement in their nectar amount can become scouts second row shows the two intermediate centers with which the
bees and they can nd a new food source randomly. corresponding customer center is connected. For example, the
Fig. 2. Solution representation for closed loop supply chain netwrok problem.
726 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
YES
Update Archive
Send Scout Bee to Create New
YES
Stop
numbers 1, 2, 1, 1, 2, 1 in the rst 6 elements of the second row of of the second row in the food source shows that at their corre-
food source vector indicates that the customer centers allocated at sponding location sites, which facility is build, the numbers can be
the location sites 1, 10, 5, 2, 6 and 9 are connected with the inter- 0, 1 and 2 which indicates producers, remanufacturers and both.
mediate center 1, 2, 1, 1, 2 and 1 respectively. The numbers shown in For example, for the elements 9 and 10, the location sites as
element 7 and 8 in the second row of food source vector shows the described in the rst row for produces and remanufacturers are 8
connection of the corresponding intermediate centers with the and 3. The corresponding numbers in the second row, i.e., 0 and 1
producer or remanufacturers, i.e., intermediate center at site 4 and shows that at location sites 8 and 3 there is producer and a rema-
site 7 are connected with the remanufacturer or producer 1 and 2 nufacturing centers respectively. The numbers 1 and 2 presented in
respectively. Moreover, the numbers shown in the element 9 and 10 the last 6 elements of the second row of the food source vector
facilities in neighbors of a food sources
1 10 5 2 6 9 4 7 8 3
Different schemes of connection
Same location points for different
1 1 2 2 1 1 2 1 1 0
between facilities
1 10 5 2 6 9 4 7 8 3
1 2 1 1 2 1 1 2 0 1
1 10 5 2 6 9 4 7 8 3
2 1 2 1 1 2 1 1 1 0
shows rst and second facilities from the facilities described in the
element 9 and 10 of the second row. Furthermore, the numbers 1
NECr
shown in last 6 elements of the second row of the food source gives Fixed:cos t Shipping:cos t Prod:cos t Holding:cos tr
information of the connection of the corresponding facilities with (28)
the respective customers. For example, 1, 2, 2, 1, 2, 1 indicates that
the 6 customer centers i.e., 1, 2, 3, 4, 5, 6 are connected with the
producer or remanufacturer, (1 and 2 shows the rst and second
facility as described in the element 9 and 10, i.e., producer and 4.2.3.3. Update the archive. In proposed GABC, there is a food
remanufacturer respectively). This representation of food sources source archive which stores one best neighbor of each employee
for the current problem is signicant and can completely dene the bee. The archive size is equal to the employee bee population. After
solution of a problem. The ow chart of the proposed GABC is every cycle, the best neighbor food source of each employee bee
shown in Fig. 3 and the step by step procedure is explained in this obtained is compared with the food source stored in archive and
section. The owchart of the proposed GABC is indicated in Fig. 3 the best one among them is stored in the archive to update the
and the algorithm is explained in this section. archive.
4.2.2. Initialization of food source population 4.2.3.4. Select parent food source. Two of the best food sources from
In GABC algorithm, the number of food sources initially created the archive are taken from the archive and considered as parent
is equal to the number of employee bees. The initial population of food sources. Tournament selection is used for the selection of
food sources is randomly generated in GABC algorithm to induce parent food sources in GABC algorithm.
diversied solutions in the search space.
4.2.3.5. Crossover. In this step, crossover operation is performed
4.2.3. Sending employee bees between the two parents food sources selected. The crossover
In this step, each employee bee is sent to its corresponding food operation is performed in two steps in GABC algorithm.
source and the corresponding step wise procedure is explained in Step 1: In this step the crossover only occurs in the portion of
this section. food source where the information of location points of the facil-
ities is shown. Order crossover is performed in rst step of cross-
4.2.3.1. Create neighbors of food source of each employee bee. over operation in GABC. As an example, Fig. 5 illustrates the rst
In the proposed GABC, each food source assigned to an employee step of crossover.
bee creates its E number of food source neighbors. The neighbor of Step 2: In this step, two point crossover operations is performed
an employee bee changes the information of connection schemes of in the offspring food sources obtained in step 1. The crossover only
various facilities while the location sites positions are same for the occurs in the portion of food source where the information of
neighbors of an employee bee. For example, for each employee bee connections between the facilities is described. Fig. 6 shows the
the location sites assigned are same in all neighbors of an employee two point crossover in the offspring food sources obtained in step 1.
bee and the difference only exists in the connections of the facil- Two point crossover is performed in two different positions in the
ities. Fig. 4 illustrates an example of the neighbor food sources of an food source as shown in Fig. 6.
employee bee. This method of creating neighbor food sources is
used for local search of the food sources for each solution of 4.2.3.6. Mutation. In this step, two point swap mutation is per-
assigned locations to the facilities. Food source assigned to each formed in the offspring food source. The swap mutation is sepa-
employee bee have different location points for the facilities while rately performed in the rst and second row in the food source to
for each employee bee, the neighbor food sources of the corre- introduce the diversity in the food sources in both location points
sponding employee bee have same location points of facilities and assigned to facilities and the connection schemes of the facilities.
the difference exists in the connection schemes of the facilities.
5 3 7 2 6 9 4 8 1 10 1 2 3 4 5 6
1 2 1 1 2 1 1 2 0 1 1 2 2 1 2 1
6 9 4 10 5 3 7 8 1 2 1 2 3 4 5 6
2 1 2 1 1 2 1 1 1 0 2 1 1 2 1 2
5 3 7 2 6 9 4 8 1 10 1 2 3 4 5 6
1 2 2 1 1 1 1 2 0 1 1 1 1 2 2 1
6 9 4 10 5 3 7 8 1 2 1 2 3 4 5 6
2 1 1 1 2 2 1 1 1 0 2 2 2 1 1 2
Table 1
Effective parameters of the proposed algorithm.
Levels Number of food sources Maximum number of algorithm Cycles Limit number of Cycles for food sources
1 200 400 10
2 300 600 15
3 400 800 20
4 500 1000 25
Table 2
Orthogonal array (OA) for Taguchi design of experiments for the proposed algorithm.
Population Size or Number of food sources Maximum number of algorithm Cycles Limit number of Cycles for food sources
1 1 1 1
2 1 1 2
3 1 2 3
4 1 2 4
5 2 3 1
6 2 3 2
7 2 4 3
8 2 4 4
9 3 1 1
10 3 1 2
11 3 2 3
12 3 2 4
13 4 3 1
14 4 3 2
15 4 4 3
16 4 4 4
5. Experiments and results can produce variation in the results without eliminating those
causes. Set of experiments is used which contains different levels of
5.1. Tuning of algorithm parameters the algorithm parameters. An orthogonal array (OA) of different
level of parameter values is used in Taguchi method to determine
Experiments are performed to investigate the performance of the signal to noise ratio (S/N). S/N ratio is the ratio of the value of
the proposed GABC algorithm in this section. The performance of objective function with the variations in its values. In OA the
an algorithm can be observed from the quality of results and vari- number of columns represents the parameters and the number of
ations in the results it can produce for various problems. Taguchi rows describes the number of experiments. Taguchi experiment is
method (Taguchi et al., 2005) is signicant to improve the quality of designed to determine the best level of parameters at which the S/N
the proposed algorithm by minimizing the effects of causes which is maximum i.e., quality of optimizing objective is better with less
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 729
F 24/15 Table 3
Unit transportation cost of products and returned products in CNY (Chinese Yuan)
between different facilities.
G H I
Location Sites A B C D E F G H I J K L
190/95/47/30 210/105/52/40 165/82/41/35
A 0 5 15 20 15 30 15 18 27 30 32 37
11/5 16/8 13/7 B 5 0 5 15 10 25 13 15 16 17 18 20
C 15 5 0 10 5 15 20 15 7 18 15 10
D 20 15 10 0 7 5 22 17 10 17 13 8
E 15 10 5 7 0 8 10 5 10 12 14 16
J K L F 30 25 15 5 8 0 20 15 5 18 16 8
G 15 13 20 22 10 20 0 5 10 8 11 20
230/115/57/25 125/62/31/20 215/108/53/40 H 18 15 15 17 5 15 5 0 6 5 8 13
I 27 16 7 10 10 5 10 6 0 8 6 8
22/17 19/10 20/8 J 30 17 18 17 12 18 8 5 8 0 5 10
K 32 18 15 13 14 16 11 8 6 5 0 6
L 37 20 10 8 16 8 20 13 8 10 6 0
Fig. 7. Small example problem data for facility location problem.
730 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
Table 4 four different values of xed costs and they are underlined, i.e., 100,
Storage capacities of the intermediate centers, producers and remanufacturers. 50, 25 and 20. These costs represents the xed cost of producers,
Location Sites H I J K L remanufacturers, intermediate centers and customer centers
Capacity of intermediate centers 20 25 30 40 50
respectively if they are located on position A in the network.
Capacity of producers or remanufacturing centers 10 15 20 10 13 Moreover, it is assumed that all the costs included in the current
Table 5
Unit processing cost in different facilities in CNY (Chinese Yuan).
the costs included in the problem. The numbers underlined above example problem are in CNY (Chinese Yuan). Furthermore, in Fig. 7
the boxes in Fig. 7 are the xed cost respectively for the four the numbers shown in the boxes represents the estimated average
m
different type of facilities. For instance, the location position A has demand edi of new products and the estimated average quantity of
130/65/32/30 150/75/37/30
100/50/25/20 T
120/140 108/90
225/140 215/108/53/40
120/85
C D
M
E 145/120 F 315/255
A B P Q
G H I
L
225/112/56/60
190/95/47/30 210/105/52/40 165/82/41/35
215/108/53/40
R 95/46
110/205 110/210 200/300
200/120
J K
200/100/50/40
230/120 200/230
O
N
175/87/43/40
190/95/47/30
S 204/105
180/90
Fig. 8. Large size example problem data for facility location problem.
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 731
m
the returned products edri from the customer, if there is a customer considered problem 20 nodes are considered as possible location
center in that location point. For example, the numbers mentioned points for the customer centers, intermediate centers, producers
in the box which is made at location position A are 15 and 20 which and remanufacturers for the considered problem. The data used for
shows the estimated average demand of the new products and the reverse ows at each facility site including the quantity of returned
estimated average demand of the returned products if the customer items, demand of products and xed costs of facilities is used here
center is made at the location A in the network. based on the Case company information and it is indicated in Fig. 8.
In the current small example problem there are total 12 sites In Fig. 8, the numbers shown underlined and are written outside
and it is assumed that there are total 7 customer centers, 2 inter- the box represents the xed cost of facilities. For example, there are
mediate centers and 3 location points are required to assign to four numbers, i.e., 130, 66, 32 and 30 written and underlined
locate the considered facilities. The unit transportation cost of the outside of the location site C. These represents the xed cost of
new and returned products from one facility location to other fa- producer, remanufacturer, intermediate center and customer cen-
cility location is described in Table 3 and the unit of cost is ter respectively. Similarly, the numbers written in the box repre-
m
considered in CNY (Chinese Yuan). The demand of new products sents the estimated average demand edi of new products and the
m
and returned products is uncertain in the current example problem estimated average quantity of the returned products edri from the
and it depends on several factors. It is assumed that there are ve customer, if there is a customer center in that location point. For
factorsbim with their values randomly taken in the range of [5, 9] example, there is average estimated demand of 120 products and
and the corresponding weightages vim for each factor is randomly average estimated quantity of the returned products on a customer
taken from the range of [1.5, 3]. The mean value of the demand of center if it is located on the location site C in the network.
new products and mean value of the quantity of the returned There are total 20 sites and the considered company needs to
m m P make total 12 customer centers, 4 intermediate centers and 4
products is calculated from di edi M m1 bim vim
m m P location points are required to assign to locate the producers or
anddri edri M b v
m1 im im respectively.
remanufacturers. The unit transportation cost of the new and
The variance in demand of new products and variance in the
returned products from one facility location to other facility loca-
quantity of the returned products is randomly taken from a uniform
tion is described in Table 6 and the costs are assumed in CNY
m m m m
distribution of 18di ; 14di and 18dri ; 14dri respectively. 90% service (Chinese Yuan). In the current problem, the costs included in the
network are assumed in the proportion of the costs of the case
level is required for all facilities which contains inventory and the company. The transportation cost includes the shipping cost of
corresponding safety stock of the facility is calculated parts which are moved using Trucks and Trailers by road. The de-
fromZ Saftystockmean
variance
. mand of new products and returned products is uncertain in the
The storage capacity of intermediate centers and remanufac- current example problem and it depends on several factors. It is
turer centers are assumed as 1:5 Safetystock of the facility as assumed that there are ve factors bim with their values randomly
shown in Table 4 for each location point, if there is an intermediate taken in the range of [10, 18] and the corresponding weightages vim
center or remanufacturer center on that location point respectively. for each factor is randomly taken from [1, 3]. The mean value of the
Table 5 indicates the other costs related with the product in demand of new products and mean value of the quantity of the
different facilities and values percentage of disposal of returned m m P
returned products is calculated from di edi M m1 bim vim
products in intermediate centers and remanufacturing centers in rm rm P M
anddi edi m1 bim vim respectively.
CNY (Chinese Yuan).
The variance in demand of new products and variance in the
quantity of the returned products is randomly taken from a uniform
m m m m
5.1.2. Large problem distribution of 18di ; 14di and 18dri ; 14dri respectively. 90% service
The large problem is made from the supply chain of a small
remanufacturing Case company in China. There are 20 different level is required for all facilities which contains inventory and the
towns which are considered in their network and therefore, in the corresponding safety stock of the facility is calculated
Table 6
Unit transportation cost of products and returned products between different facilities in CNY (Chinese Yuan).
Location Sites A B C D E F G H I J K L M N O P Q R S T
A 0 10 20 25 20 35 10 14 20 20 30 35 40 38 45 45 50 50 53 58
B 10 0 15 17 13 25 12 10 16 25 27 30 30 26 35 40 45 42 48 54
C 20 15 0 10 10 15 22 20 18 30 35 28 24 38 32 32 40 36 45 38
D 25 17 10 0 12 13 30 28 20 32 30 20 10 39 36 24 35 36 40 37
E 20 13 10 12 0 12 15 12 10 28 23 16 15 27 26 27 30 29 36 35
F 35 25 15 13 12 0 27 24 15 34 20 10 10 30 27 20 28 20 30 25
G 10 12 22 30 15 27 0 10 15 10 18 28 35 20 35 53 58 52 58 62
H 14 10 20 28 12 24 10 0 13 12 15 20 30 15 24 32 36 38 40 45
I 20 16 18 20 10 15 15 13 0 15 10 12 20 20 18 27 30 26 30 35
J 20 25 30 32 28 34 10 12 15 0 13 27 50 10 29 54 60 54 58 65
K 30 27 35 30 23 20 18 15 10 13 0 14 35 10 12 30 36 30 25 40
L 35 30 28 20 16 10 28 20 12 27 14 0 20 25 10 20 25 15 25 30
M 40 30 24 10 15 10 35 30 20 50 35 20 0 48 40 14 20 25 37 16
N 38 26 38 39 27 30 20 15 20 10 10 25 48 0 16 40 45 35 28 58
O 45 35 32 36 26 27 35 24 18 29 12 10 40 16 0 30 35 15 10 40
P 45 40 32 24 27 20 53 32 27 54 30 20 14 40 30 0 10 10 25 15
Q 50 45 40 35 30 28 58 36 30 60 36 25 20 45 35 10 0 13 27 13
R 50 42 36 36 29 20 52 38 26 54 30 15 25 35 15 10 13 0 17 20
S 53 48 45 40 36 30 58 40 30 58 25 25 37 28 10 25 27 17 0 40
T 58 54 38 37 35 25 62 45 35 65 40 30 16 58 40 15 13 20 40 0
732 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
Table 7
Storage capacities of the intermediate centers, producers and remanufacturers.
Location Sites M N O P Q R S T
11800
40
11600 38
Mean of Means
36
11400 34
Mean of S/N
11200 32
30
11000 28
10800 26
24
10600 22
1 2 3 4 20
Level of Parameters 1 2 3 4
Level of Parameters
population size Cycles Limit Cycles
Fig. 9. Mean of means for small size problem. Population size Cycles Limit Cycles
50000
Mean of Means
49500
41
49000 39
37
Mean of S/N
48500 35
33
31
48000
29
27
47500 25
1 2 3 4 1 2 3 4
Level of Parameters Level of Parameters
Population size Cycles Limit Cycles Population size
Table 8
fromZ Saftystockmean
variance
. Optimum level of parameters used for each category of problem.
The storage capacity of intermediate centers and remanufac- Problem Category Optimal level of parameters
turer centers are assumed as 3 Safetystock of the facility as shown Population size Cycles Limit Cycles
in Table 7 for each location point, if there is an intermediate center
Small Size Problem 2 4 4
or remanufacturer center on that location point respectively. The Large Size Problem 3 4 4
other costs related with the product in different facilities and values
percentage of disposal of returned products in intermediate centers
and remanufacturing centers is same as the costs used for large size
level, as indicated in Fig. 10. The mean of S/N values of objective for
problem are same as used for small size problems and presented in
small problem and large problem are indicated in Figs. 11 and 12
Table 5.
respectively. The optimum level of parameters obtained after
Graphical method is used to identify the optimal level of algo-
observing the mean of means and mean of S/N values for each
rithm parameters for the small and large size problem. The opti-
category of problem is illustrated in Table 8.
mum level of parameters for small problem is obtained by
observing mean value of means at each parameter level, as indi-
cated in Fig. 9. The optimum level of parameters for large problem 5.2. Results
is obtained by observing mean value of means at each parameter
In this section, small and large sizes of network optimization
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 733
A
Remanufacturers 5.2.1. Results for small problem
It can be seen from Table 9 that proposed GABC algorithm gives
E
Cost 10900
a close loop supply chain network which have total cost of 10900
F
Time 1870
CNY (Chinese Yuan) which is less as compared to the close loop
I
supply chain networks obtained from ABC or GA. This indicates that
H
the proposed GABC algorithm gives better results in terms of
G
quality. These results indicate that the close loop supply chain
network which is obtained by GABC require less amount of the total
J K L cost. Moreover, the proposed GABC gives the optimal network by
considering the overall effect of all costs including the xed cost of
Fig. 13. Closed loop supply chain network for small size problem obtained from GABC. facilities, shipping cost, holding cost and production costs etc. For
instance, the cost of network is 10900 CNY (Chinese Yuan) which is
obtained from the GABC algorithm, while the close loop network
problems are analyzed with the proposed GABC, ABC and GA using formed by ABC and GA algorithms have total costs of 11855 CNY and
the optimal parameter values obtained in the previous section. The 11267 CNY. These results indicate that the close loop network ob-
network solution of the small size problem obtained from GABC, tained from GABC is economical as compared to the networks
ABC and GA algorithm is indicated in Figs. 13, 14 and 15 respectively. formed by ABC and GA algorithm. However, the computational
Close Loop Supply chain Network obtained from ABC Customer Centers
B C D Intermediate Centers
A
Remanufacturers
F
Cost 11855
Time 1092
H E
G I
J L
K
Fig. 14. Closed loop supply chain network for small size problem obtained from ABC.
A B C D Intermediate Centers
Remanufacturers
Producer
E
F
I
G H Cost 11267
Time 1817
J K L
Fig. 15. Closed loop supply chain network for small size problem obtained from GA.
734 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
Customer Centers
Cost 48469
Time 1857
Intermediate Centers
Remanufacturers
C D M T
Producer
A B
H F P
G
I L R
J K
N O S
Fig. 16. Closed loop supply chain network for large size problem obtained from GABC.
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 735
Remanufacturers D T
C
Producer
A B Q
F
E
P
I
L
G H
J O
N S
Fig. 17. Closed loop supply chain network for large size problem obtained from ABC.
variations, medium range of variations and large range of varia- network which is obtained from GABC is less than the cost obtained
tions. Furthermore, for each category of problem, ten different from ABC and GA. The total cost of network includes the xed costs
scenarios of each small and large size problem are tested. of facilities, the holding costs, shipping costs and production costs.
The results of the objective function values and CPU time for Therefore, the network obtained from GABC is better in quality as
each category of problems for small problem among ten scenarios compared to the network obtained from ABC and GA. Moreover, the
is obtained and illustrated in box plots as indicated in Figs. 19, 20 CPU time difference is also small and not much signicant. The
and 21 respectively. Furthermore, the results of objective function difference between the objective function and computation time
values and CPU time for each category of problems for large increases as the variations in the demand increases from small to
problem among ten scenarios is obtained and illustrated in box plot medium and from medium to large. Furthermore, for small and
in Figs. 22, 23 and 24 respectively. medium variations in demand, the objective function value of the
small problem is not much variable and variations in the objective
function values is small as compared to the ABC and GA algorithm.
6.1. Small problem with small demand
These small variations can show that the optimal network is ob-
tained from GABC remains better as compared to the network ob-
The small size problem with small size of demand with different
tained from ABC and GA even there are changes in the demand
ranges of demand variations is considered here and the total cost of
occurs in the network. Therefore, it can show that the network
the network against different demand scenarios is indicated in
obtained from GABC algorithm can cope with the demand varia-
Fig. 19. It can be seen from Fig. 19 that for small problem, when the
tions better as compared to the network obtained from ABC and GA
variations in demand increases from small to medium and from
and this shows the better robust performance of the network ob-
medium to large variation, the objective function value i.e., total
tained form GABC as compared to ABC and GA. However, for large
cost of the network, increases. Moreover, the objective value ob-
variations in demand, there is more variations in the objective
tained from GABC is smaller as compared to the objective function
function obtained from GABC as compared to ABC and GA. These
values obtained from ABC and GA when there is small and medium
results indicate that when there is small variation and medium size
variations in demand. However, for large variations in demand, the
variation in demand, the proposed GABC algorithm can give the
objective function obtained from GA is less as compared to GABC
closed loop supply chain network with consistence value of the
and ABC but their difference is very small. Moreover, the CPU time
total cost. These results indicate that the proposed GABC algorithm
taken by GA is less than both GABC and ABC for small range, me-
gives more robust solution results as compared to ABC and GA,
dium range and large range of variations in demand but the dif-
when there are small and medium size variations in demand.
ference in CPU time is not signicant. For small range of variations
However, when the variations in demand is large, the proposed
in demand, GABC performs better as compared to both ABC and GA
GABC algorithms is not signicantly robust.
in terms of solution quality and computation i.e., the cost of
736 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
Remanufacturers
C D M T
Producer
A B
F
P
Q
E
G L
H
R
N
K
S
J O
Fig. 18. Closed loop supply chain network for large size problem obtained from GA.
Table 11
Scenario of different problems containing different value and range of uncertain demand and returned products.
Small demand variation with small range 10 [5, 10] [1.5, 3] [10, 12] [1, 3]
Small demand variation with medium range 10 [5, 20] [1.5, 6] [10, 24] [1, 6]
Small demand variation with large range 10 [5, 30] [1.5, 9] [10, 36] [1, 9]
Medium demand variation with small range 10 [15, 10] [2, 3] [20, 12] [1.5, 3]
Medium demand variation with medium range 10 [15, 20] [2, 6] [20, 24] [1.5, 6]
Medium demand variation with large range 10 [15, 30] [2, 9] [20, 36] [1.5, 9]
Large demand variation with small range 10 [25, 10] [2.5, 3] [30, 12] [2, 3]
Large demand variation with medium range 10 [25, 20] [2.5, 6] [30, 24] [2, 6]
Large demand variation with large range 10 [25, 30] [2.5, 9] [30, 36] [2, 9]
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 737
4
Small problem& Small demand
x 10
28
GABC GABC GA
GABC GA
ABC ABC
ABC
4 GA 26
3.5
24
Objective Value
3
22
ABC
CPU Time
2.5 GABC
GA 20
2
18
1
14
ABC GA
0.5
GABC
12
Fig. 19. Total cost of the network against different demand scenarios for small problem with small demand.
8
26
Objective Value
GA
ABC
7 GABC
CPU Time
24
6
22
GABC ABC
5 GA
20
4
18
3 16
GABC ABC
2 GA
14
1 12
Fig. 20. Total cost of the network against different demand scenarios of small problem with medium demand.
performance based on the total cost of the network as compared to comparable with the CPU time taken by ABC and GA. For small
ABC and GA. Furthermore, the CPU time taken by GABC for small range of variations in demand, medium range and large range of
range of demand, medium range and large range of demand, is demand variations GABC performs better as compared to both ABC
738 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
4
Small problem& Large demand
x 10
6 ABC 32
30
5.5
GABC GA
28
5 ABC
GABC
GA
26
Objective Value
4.5
24
CPU Time
4
22
GABC ABC GA
3.5
20
3 18
16
2.5
14
2 GABC ABC GA GABC
ABC GA GABC
ABC GA
12
1.5
Fig. 21. Total cost of the network against different demand scenarios of small problem with large demand.
4
Large problem &Small demand
x 10
ABC GABC
GABC
ABC GA
12
28
GA
26
10
ABC
GA
Objective Value
24
GABC
8
CPU Time
22
GA
6 GABC ABC
20
18
4
16
14
GABC ABC GABC
GA ABC
GA
0
12
and GA in terms of solution quality, i.e., the total cost of the network the better stability of the results obtained from GABC. Furthermore,
obtained from GABC is smaller as compared to the network ob- it can be seen from Fig. 20 that the proposed GABC algorithm gives
tained from ABC and GA. Furthermore, there is less variations in the less variations in the objective function value when there is change
CPU time and objective values obtained from GABC which indicates in the range of variations in demand. These less variations in the
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 739
6.5 12
GABC
6 11.9
GABC
5.5 GABC GABC
Objective Value
11.8
5 ABC
CPU Time
11.7
11.3
2.5 GABC ABC GA
11.2
2
Fig. 23. Total cost of the network against different demand scenarios of large problem with medium demand.
4
x 10
Large problem & Large demand
10 ABC
GABC
13.8
GA
9
13.6
8
13.4
7
Objective Value
13.2
GABC ABC GA
6 GABC
GABC
CPU Time
13
GABC
5 ABC ABC
12.8
GA
4 ABC
12.6
GA GA
3
12.4
2 12.2
GABC ABC GA
1 12
objective function, i.e., total cost of network, indicate that the consistently gives the optimal network even there are changes in
network obtained from GABC gives consistent results and the demand. Moreover, the network obtained from GABC can cope
740 Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742
with the uncertain variations in demand with total network cost as GA. These results indicate the robustness of the results obtained
minimum as compared to network obtained from ABC and GA. from GABC is better as compared to ABC and GA. When the varia-
These results indicate that the proposed GABC algorithm gives tions in demand increases to large, the variations in the objective
more robust solutions as compared to both ABC and GA and gives function values obtained from GABC, ABC and GA also increases.
less variations in the total cost of the network. These results indicate that for large variations in demand, the so-
lutions given by GABC, ABC and GA are not signicantly robust but
6.3. Small problem with large demand compared to ABC and GA, GABC is relatively gives less variations in
objective function when there is large variations in demand.
The small size problem with large size of demand with different Furthermore, the CPU time taken by GABC for small range of de-
ranges of demand variations is considered here and the total cost of mand, medium range and large range of demand, is comparable
the network against different demand scenarios of the problem are with the CPU time taken by ABC and GA and there is no much
indicated in Fig. 21. It can be seen from Fig. 21 that for small difference in CPU time for GABC, ABC and GA for small, medium and
problem, when the variation in demand increases from small to large variations in demand problem scenarios.
medium and from medium to large variation, the objective function
value is increases, this indicates that the cost of the network in-
creases as the variation in demand increases. Moreover, the 6.5. Large problem with medium demand
objective value obtained from GABC is smaller as compared to the
values obtained from ABC and GA when there are small variations The large size problem with medium size of demand with
in demand. Whereas, the objective function value obtained from different ranges of demand variations is considered and the total
GABC is comparable with ABC and GA for medium variation and cost of the network against different demand scenarios of the
large variation in demand. The total cost of network obtained from problem are indicated in Fig. 23. It can be seen from Fig. 23 that for
GABC is smaller as compared to the network obtained from ABC large problem, when the variation in demand increases from small
and GA when there are small variations in the demand, which in- to medium and from medium to large variation, the objective
dicates that GABC gives better results in terms of solution quality as function value is increases, this indicates that the cost of the
compared to ABC and GA. Furthermore, the CPU time taken by network increases as the variation in demand increases. Moreover,
GABC for small range of demand, medium range and large range of the objective value obtained from GABC is smaller as compared to
demand, is comparable with the CPU time taken by ABC and GA and the values obtained from ABC and GA for problems with small
there is no much difference in CPU time for GABC, ABC and GA for variation. However, there is small difference in the objective value
small, medium and large variations in demand problem scenarios. obtained from GABC, ABC and GA for the scenario with medium and
It can be seen from Fig. 21 that there is less variations in the large variations in demand. Moreover, the objective function values
objective function value obtained from GABC algorithm when there obtained from GABC gives less variations as compared to the vari-
is small and medium size variations in the demand as compared to ations in total cost of network obtained from ABC and GA when the
ABC and GA. These results indicate that the proposed GABC algo- variations in demand increases from small, medium and large
rithm gives the network which has smaller value of the total variations. These results indicate better robustness of the solutions
network cost as compared to ABC and GA even there are changes in in terms of the total cost of network obtained from GABC as
the demand. This shows that the network obtained from GABC compared to ABC and GA. Furthermore, the CPU time taken by GA
gives more robust solutions as compared to both ABC and the total for small range of demand, medium range and large range of de-
cost of the network is not changing even there is variations in the mand, is smaller but comparable with the CPU time taken by GABC
demand in small or medium range. However, as the size of varia- and ABC and there is no much difference in CPU time for GABC, ABC
tions increases to large size variations, the objective function values and GA for small, medium and large variations in demand problem
obtained from GABC, ABC and GA shows large variations. These scenarios.
results indicate that the total cost of the network obtained from
GABC, ABC and GA varies in large range as the demand varies in
large range. The robustness of the results obtained from GABC is 6.6. Large problem with large demand
better than ABC and GA in terms of the total cost of network when
there is small size and medium size variations in the demand. The large size problem with large size of demand with different
ranges of demand variations is considered and the total cost of the
6.4. Large problem with small demand network against different demand scenarios of the problem are
indicated in Fig. 24. It can be seen from Fig. 24 the objective value,
The large size problem with small size of demand with different i.e., total cost of the network, obtained from GABC is smaller as
ranges of demand variations is considered and the total cost of the compared to the values obtained from ABC and GA for problems
network against different scenarios of demand are indicated in with the problem scenarios with small variation. The objective
Fig. 22. It can be seen from Fig. 22 that for large problem, when the function value obtained from GABC is smaller and also there are less
variations in demand increases from small to medium and from variations in the objective function values for the solutions given by
medium to large variation, the objective function value is increases, GABC as compared to the objective function values obtained from
this indicates that the cost of the network increases as the variation ABC and GA for small, medium and large size variations in demand.
in demand increases. Moreover, the objective value, i.e., total cost of These results indicate that the results obtained from GABC gives
the network, obtained from GABC is smaller as compared to the more optimal closed loop supply chain network and the total
values obtained from ABC and GA for problems with small variation network cost is robust, ie., total network cost is not changing in
and medium range of variations. However, there is small difference large range when there is uncertain change in demand. Further-
in the objective value obtained from GABC, ABC and GA for the more, the CPU time taken by GA for small range of demand, me-
scenario with large variations in demand. Furthermore, there is dium range and large range of demand, is smaller but comparable
small variations in the objective function obtained from GABC al- with the CPU time taken by GABC and ABC and there is no much
gorithm when there is small and medium size variations in demand difference in CPU time for GABC, ABC and GA for small, medium and
compared to the objective function values obtained from ABC and large variations in demand problem scenarios.
Y.Y. Cui et al. / Journal of Cleaner Production 162 (2017) 717e742 741
Expert Syst. Appl. 36, 9299e9307. http://dx.doi.org/10.1016/j.eswa.2008.12.002. Strategic Production and Operations Management Issues in Product Recovery
Li, Jun-Qing, Pan, Quan-Ke, Gao, Kai-Zhou, 2011. Pareto-based discrete articial bee Management. Management Report Series No. 145. Erasmus Universiteit/Rot-
colony algorithm for multi-objective exible job shop scheduling problems. Int. terdam School of Management.
J. Adv. Manuf. Technol. 55 (9e12), 1159e1169. Tonanont, A., Yimsiri, S., Rogers, K.J., 2009. Reverse logistics optimization with data
Lu, Z., Bostel, N., 2007. A facility location model for logistics systems including envelopment analysis. In: Proceedings of 2009 Industrial Engineering Research
reverse ows: the case of remanufacturing activities. Comput. Oper. Res. 34, Conference, Miami, 30 May-3 June 2009, pp. 1268e1273.
299e323. Saif, Ullah, Guan, Zailin, Xianhao, Xu, Zongdong, He, Baoxi, Wang, Jahanzeb, Mirza,
Ozkir, Vildan, Basligil, Huseyin, 2013. Multi-objective optimization of closed-loop 2015. Modeling of facility location problem for reverse logistics system with
supply chains in uncertain environment. J. Clean. Prod. 41, 114e125. uncertainty in forward and reverse ows of products. Int. J. Mech. Eng. Robot.
Pishvaee, Mir Saman, Kianfar, Kamran, Karimi, Behrooz, 2010. Reverse logistics Res. 4 (3), 242e246.
network design using simulated annealing. Int. J. Adv. Manuf. Technol. 47 (1e4), ster, Halit, Easwaran, Gopalakrishnan, Akali, Elif, etinkaya, Sila, 2007. Benders
269e281. decomposition with alternative multiple cuts for a multi-product closed-loop
Pishvaee, Mir Saman, Rabbani, Masoud, Torabi, Seyed Ali, 2011. A robust optimi- supply chain network design model. Nav. Res. Logist. 54 (8), 890e907.
zation approach to closed-loop supply chain network design under uncertainty. Vahdani, Behnam, Tavakkoli-Moghaddam, Reza, Modarres, Mohammad,
Appl. Math. Model. 35 (2), 637e649. Baboli, Armand, 2012. Reliable design of a forward/reverse logistics network
Salema, Maria Isabel Gomes, Barbosa-Povoa, Ana Paula, Novais, Augusto Q., 2007. under uncertainty: a robust-M/M/c queuing model. Transp. Res. Part E Logist.
An optimization model for the design of a capacitated multi-product reverse Transp. Rev. 48 (6), 1152e1168.
logistics network with uncertainty. Eur. J. Oper. Res. 179 (3), 1063e1077. Wollenweber, Jens, 2008. A multi-stage facility location problem with staircase
Salema, Maria Isabel Gomes, Po voa, Ana Paula Barbosa, Novais, Augusto Q., 2009. costs and splitting of commodities: model, heuristic approach and application.
A strategic and tactical model for closed-loop supply chains. OR Spectr. 31 (3), OR Spectr. 30 (4), 655e673.
573e599. Yi, Pengxing, Huang, Min, Guo, Lijun, Shi, Tielin, 2016. A retailer oriented closed-
Shi, Jianmai, Zhang, Guoqing, Sha, Jichang, Amin, Saman Hassanzadeh, 2010. loop supply chain network design for end of life construction machinery
Coordinating production and recycling decisions with stochastic demand and remanufacturing. J. Clean. Prod. 124, 191e203. http://dx.doi.org/10.1016/
return. J. Syst. Sci. Syst. Eng. 19 (4), 385e407. j.jclepro.2016.02.070.
Suryabatmaz, A.C., Altekin, F.T., Sahin, G., 2014. Hybrid simulation-analytical Zhiqiang, Lu, Nathalie, Bostel, 2007. A facility location model for logistics systems
modeling approaches for the reverse logistics network design of a third-party including reverse ows: the case of remanufacturing activities. Comput. Oper.
logistics provider. Comput. Indus. Eng. 70, 74e89. Res. 34, 299e323.
Taguchi, G., Chowdhury, S., Wu, Y., 2005. Taguchi Quality Engineering Handbook. Zohal, Mostafa, Soleimani, Hamed, 2016. Developing an ant colony approach for
John Wiley and Sons, New York. green closed-loop supply chain network design: a case study in gold industry.
Thierry, M.C., Salomon, M., Van Nunen, J.A.E.E., Van Wassenhove, L.N., 1993. J. Clean. Prod. 133, 314e337.