Ecological Indicators 67 (2016) 753768

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

A fuzzy multi-objective optimization model for sustainable reverse

logistics network design
Kannan Govindan a, , Parichehr Paam b , Amir-Reza Abtahi b,
a
Centre for Sustainable Engineering Operations Management, Department of Technology and Innovation,
University of Southern Denmark, Odense, Denmark
b
Department Information Technology Management, Faculty of Management, Kharazmi University, Tehran, Iran

a r t i c l e

i n f o

Article history:
Received 30 July 2015
Received in revised form 26 February 2016
Accepted 5 March 2016
Available online 25 April 2016
Keywords:
Reverse logistics
Sustainability
Social responsibility
Fuzzy mathematical programming
Multi-objective metaheuristic algorithm
Epsilon-constraint method

a b s t r a c t
Decreasing the environmental impact, increasing the degree of social responsibility, and considering
the economic motivations of organizations are three signicant features in designing a reverse logistics
network under sustainability respects. Developing a model, which can simultaneously consider these
environmental, social, and economic aspects and their indicators, is an important problem for both
researchers and practitioners. In this paper, we try to address this comprehensive approach by using
indicators for measurement of aforementioned aspects and by applying fuzzy mathematical programming to design a multi-echelon multi-period multi-objective model for a sustainable reverse logistics
network. To reect all aspects of sustainability, we try to minimize the present value of costs, as well
as environmental impacts, and optimize the social responsibility as objective functions of the model. In
order to deal with uncertain parameters, fuzzy mathematical programming is used, and to obtain solutions on Pareto front, a customized multi-objective particle swarm optimization (MOPSO) algorithm is
applied. The validity of the proposed solution procedure has been analyzed in small and large size test
problems based on four comparison metrics and computational time using analysis of variance. Finally,
in order to indicate the applicability of the suggested model and the practicality of the proposed solution
procedure, the model has been implemented in a medical syringe recycling system. The results reveal
that the suggested MOPSO algorithm overtakes epsilon-constraint method from the aspects of quality
of the solutions as well as computational time. Proper use of the proposed process could help managers
efciently manage the ow of recycled products with regard to environmental and social considerations,
and the process offers a sustainable competitive advantage to corporations.
2016 Elsevier Ltd. All rights reserved.

1. Introduction
Reverse logistics (RL) is one of the signicant areas discussed
in subjects related to logistics and management of supply chain
in various industries. Because of the great effects on customer
relationships, reverse logistics and logistics related to operational
capabilities should be regarded as a managerial priority (Liu, 2014;
Bouzon et al., 2016). RL is a general term that covers a wide area,
including all operations related to re-using of goods and materials.
The efcient management of these operations can improve the system of distribution and collection of goods and materials. Generally,
the aim of RL is to manage reverse currents; that is, the backward
currents in the supply chain (lvarez et al., 2007). RL includes all

Corresponding author. Tel.: +45 65503188.

E-mail addresses: kgov@iti.sdu.dk (K. Govindan), abtahi@khu.ac.ir
(A.-R. Abtahi).
http://dx.doi.org/10.1016/j.ecolind.2016.03.017
1470-160X/ 2016 Elsevier Ltd. All rights reserved.

logistical activities that are related to resource reduction, recycling,

replacing, re-using the materials, and dissolving the wastes (Stock,
1992).
Over the past decade, companies have been challenged with
the complicated issues of customer returns; accordingly, RL developed much more as a set of models and techniques to manage
these issues. The rapid growth of RL activities has heightened the
levels of social and environmental degradation (Henriques and
Sadorsky, 1996) and, consequently, has drawn the attention of academic researchers and industrial practitioners to nd a solution for
these problems.
One of the most important objectives facing the researchers is
to design a RL network that can minimize the costs, and simultaneously can consider green and social issues. Incorporation of green
strategies and sustainability in logistics can be the keys to address
these problems (Murphy and Poist, 2000). In a broader vision, in
addition to providing green products and services for customers,
green logistics (GL) considers the overall logistics of an items ow.

754

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

In other words, not only do the entities in the supply chain have
to be green in their own logistics operations, but also they have to
cooperate with each other with regard to green considerations. In
this approach, RL is a primary issue of GL, and cannot be designed
by simply considering economic aspects. Hence, in a sustainable
way, environmental factors have to be considered as well (Zhou
et al., 2000; Bjrklund et al., 2012; Hervani et al., 2005; Zhang et al.,
2015). In addition, companies must care about their social responsibility. This implies that companies, besides making prot, have
to think about conformity with legal necessities, ethical principles,
and esteem for people and communities in all of their activities (Pai
et al., 2015).
Thus, the term sustainability is used when environmental and
social factors are taken into account in addition to economic
aspects. Seuring and Muller (2008) dened sustainability as the
designing and employing human systems as well as industrial systems in order to use natural resources and to make sure that the
normal cycles do not reduce the quality of life and future economic opportunities and also do not have any negative impact on
social conditions, human health, and ecosystem. Every company
and their individuals need to develop themselves with regard to the
health and safety of all creatures, to a cleaner environment, and help
to ensure a safe and balanced society (Kotzee and Reyers, 2016).
Todays companies take a range of different approaches in their
pursuit of sustainability as they attempt to make their supply chains
more responsive to the environment and society. Effective RL programs have great potential in helping them attain these goals,
because sustainability and RL are interconnected. Hence, designing the supply chains with optimized RL can help them enhance
resource recovery, reduce returns, integrate shipments, and adjust
transportation administration (Lee and Lam, 2012).
Many researchers trying to explore the relationships linking
sustainability and green supply chain in RL (Govindan, 2015;
Rostamzadeh et al., 2015). Most of the previous research has
focused on social and green issues through closed-loop supply
chain, but integrating RL in the design of green and sustainable supply chains has rarely been considered (Govindan et al.,
2015a,b,c, 2016; Govindan and Cheng, 2015). In a quantitative
approach, this integration can be done by adding some decision
variables, objective functions, and constraints into mathematical
models (Govindan et al., 2015b; Soleimani and Govindan, 2015).
Therefore, the contribution of this paper is to integrate the green
and sustainability issues in RL. In this study, a sustainable reverse
logistics network design (SRLND) is presented. The minimization
of the present value of costs, the minimization of environmental impacts, and the maximization of social responsibility are the
objective functions to consider the three aspects of sustainability.
Social responsibility is considered as increasing the career opportunities and reducing harms at work. In order to minimize the
environmental impacts, the eco-indicator 99 methodology is used,
which is a way for estimating the environmental impacts of a supply
chain network (Pishvaee and Razmi, 2012; Pishvaee et al., 2012).
Network design problems are classied as NP-Hard problems
and the duration of the solution process is increased exponentially
according to the size of the problem (Aras et al., 2008; Fattahi et al.,
2015). Consequently, the problem of multi-objective RL network
design is an NP-Hard problem. In order to solve this problem in
an acceptable length of time, meta-heuristic algorithms should be
used. The result of solving the multi-objective RL network design
problem with a meta-heuristic algorithm is the production of nondominated Pareto optimal solutions for decision-makers. By using
meta-heuristic algorithms, all possible solutions for the problem
are obtained and the decision-makers can make their nal decision
based on comparison metrics and comprehensive data.
The rest of this paper is organized as follows. In Section 2, the
literature in this area of research is reviewed. In Section 3, the

problem is dened and the model is formulated. In Section 4, the

solving methodology and comparison metrics are discussed. Section 5 is dedicated to experimental results. Finally, in Section 6, the
conclusions are presented and suggestions for future research are
provided.
2. Literature review
In this paper, we attempt to propose a model for a RL network design problem, regarding green and sustainability issues and
use a meta-heuristic algorithm to solve the model. Consequently,
the focus of the literature survey in this study is subdivided into
three sections: (1) RL mathematical models for supply chain (SC);
(2) Green and sustainable RL and SC; and (3) The applications of
meta-heuristic algorithms in RL and SC.
2.1. RL mathematical models for supply chain
In this section, some RL mathematical models for SC and the
solving methodologies have been reviewed. Fleischmann et al.
(1997) did one of the rst studies on various characteristics and
improvements in RL system. They separated the subject into three
major parts: production management, inventory management, and
distribution management. Then they elaborated on the implications of re-using efforts per each part, studied the models suggested
in the previous researches, and offered several suggestions for
future researchers. Minner (2001) mixed internal and external
product returns and their improvement with the problem of certainty inventory control in the SC. He then solved the problem
by a concave minimum optimization method and stated that the
re-using of products creates the surplus inventory. He concluded
that when the complete concurrence among various supply conditions is expensive (considering the characteristics of SC in terms of
processing time, level of services, and inventory costs), the creation
of surplus inventory is inevitable.
In solving the problem of reverse logistics recycling balance,
Chen (2012) mainly focused on the balance in conditions that the
prices of the market and recycling ows have interactive effects and
the ows of input and output recycled materials in the SC are not
balanced. Lee and Chan (2009) offered an RL network based on radio
frequency identication (RFID) technology. Their aim was to nd a
model for the optimization of product output. They suggested that
RFID could be used for the counting of stored items in storage points
and for sending signals to the returning center. Das and Chowdhury
(2012) offered a recycling, logistics model for various electronic
product wastes in order to minimize the overall processing costs.
Their model consisted of four recycling phases: collection, separation, recycling, and repair. The nal site included a dumping point,
primary market, and secondary market. They found that the transportation costs constitute a major part of recycling costs. Therefore,
they concluded that the reduction of transportation costs is the best
way to reduce the overall costs of the system.
Nikolaou et al. (2013) presented a combinatorial model for
the social responsibility of companies and sustainability in the RL
system on the basis of a complete operational framework. Their
framework included combinatory mathematical indices for the
evaluation of social responsibility in RL. Their work can enhance
the companies performance in social responsibility throughout
reverse logistics processes. Ramezani et al. (2013) developed a
model to design a forward and reverse asymmetric logistics system. They suggested a process for optimizing the quality, prot, and
customer responsiveness as the objective functions in their model.
Suyabatmaz et al. (2014) presented two modeling methods
for the stochastic RL system design to handle the uncertainties in the problem. One addressed the design of a distribution
network and another considered the development of a generic

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

method based on a recently proposed novel solution methodology. They discussed the results obtained from the two approaches
under different scenarios. Ghayebloo et al. (2015) presented a biobjective programming model for an asymmetric forward/reverse
logistics network. Multiple non-dominated solutions are obtained
to demonstrate the trade-off between prot and greenness objectives. They developed some managerial insights through various
computational experiments.
2.2. The applications of meta-heuristic algorithms in RL and SC
Some metaheuristic studies in RL and SC have been reviewed in
this section. Min et al. (2006) introduced a non-linear model to solve
RL problems. Their aim was to reduce the costs of RL processes. This
model can create a balance in the inventory and reduce retention
and transportation costs. Du and Evans (2008) proposed a model to
optimize the total costs and the total lateness of cycle time. Their
models results revealed that minimizing both costs and cycle times
make the network a more centralized structure.
Kim et al. (2009) introduced a route-nding program for vehicles in South Korea transporting electronic material wastes that
were at the end of their life cycle. Their aim was to minimize the
transportation routes by identifying the four main areas of recycling
centers. For each area, a route-nding problem was made. In order
to solve this problem, tabu search algorithm was employed. Kannan
et al. (2016) proposed a sustainable strategy for electronic waste
through formal recycling. Pishvaee et al. (2010) proposed a mathematical programming model for a multi-period logistics network
in order to decrease the transportation costs and xed costs. The RL
network included customer places, collection, checking, recycling,
and dumping with limited capacity. They used a simulated annealing algorithm as solution procedure and compared the results with
the solutions obtained by other exact methods. They concluded that
the solutions obtained by their suggested algorithm are better than
other ones.
Diabat et al. (2013) designed a multi-step RL network problem in order to determine the quantity and positions of primary
collection places, centralized return places, and maximum keeping time for small volumes of returned products. They presented
a model to reduce the costs in RL network designing problem,
including the costs of keeping, preparation, transportation, etc. To
solve this problem, they used genetic and articial immune system
algorithms. Roghanian and Pazhoheshfar (2014) designed a multiproduct, multi-period reverse logistics network to determine the
disassembly subsection and processing centers and also the transportation plan that will satisfy the demand of manufacturing and
recycling centers with a minimum xed opening cost and the total
shipping cost. Then, they used a genetic algorithm as their solution
methodology. Soleimani and Govindan (2015) developed a model
to improve closed loop supply chain network. The decision variables, including design and planning in location and allocation,
are considered in their proposed network. Then, they presented a
methodology based on particle swarm optimization and a genetic
algorithm to solve the model. The results showed that the proposed hybrid algorithm outperforms the single genetic algorithm
and single particle swarm optimization.
2.3. Green and sustainable RL and SC
Integrating green concepts and sustainability in RL and SC is an
evolving area in the literature. However, the increasing integration of sustainability into SC demonstrates an evolving area where
they exhibit explicit interactions (Ashby et al., 2012). Korchi and
Millet (2011) obtained 18 various general structures of a reverse
logistics network by changing the places of customers and the
centers of grouping, storing, and production. By comparing and

755

analyzing these structures, they managed to offer structures with

higher economic prots and lower environmental harms.
Chaabane et al. (2012) introduced a model in order to design a
sustainable RL. The model had two objective functions to optimize
the costs and the negative impacts on the environment (solid, liquid, and gaseous emissions emanating from production processes
and transportation systems). Pishvaee and Razmi (2012) designed
a model for a supply chain to optimize the costs and the environmental negative impacts. To measure the environmental impacts,
they employed a method called life cycle assessment.
Lopes Silva et al. (2013) studied reverse logistics and developed
a returnable packaging model to optimize the waste generation and
environmental impacts while minimizing the costs and consumption of the resources. They used eld data, ow mapping of RL, and
environmental assessment in their research. Ramos et al. (2014)
considered economic, environmental, and social aspects in designing the RL systems. They solved a multi-objective vehicle routing
problem with inter-depot routes and the Pareto front obtained
for the case study. Finally, trade-offs between the objectives were
reviewed.
Fahimnia et al. (2015) suggested a tactical planning model for
the supply chain, which can be used to examine compromises
between cost and environmental impacts. They developed a solution procedure known as nested integrated cross-entropy (NICE),
applied it to the proposed nonlinear mathematical model, and
focused on examining the association between lean attempts and
green consequences. They found that (1) not all lean involvements
in the tactical level result in green advantages, and (2) a comparison between lean and centralized supply chains with green supply
chains demonstrates that the green supply chain is a more exible
and efcient choice.
Govindan et al. (2015a) used a fuzzy multi-attribute decisionmaking method to indicate causal relationships between practices
in management of green supply chains and performances. They
found the main practices that could have a positive impact on supply chain performances in terms of environmental and economic
aspects. The results revealed management, collaboration, green
buying, and ISO 14001 certication are the most important practices in the administration of green supply chains. Mangla et al.
(2015) analyzed the risks relevant to adoption and effective implementation of green supply chain (GSC) practices from an industrial
viewpoint. In the rst phase of their research, six categories and
twenty-ve specic risks associated with the GSC were identied.
In the next phase, the fuzzy analytic hierarchy process, a qualitative
and quantitative analysis was used to analyze the identied risks
for determining their priority of concern. The analysis of the results
indicated that operational category risks are the most important
risks in GSC.
Coskun et al. (2015) designed the green supply chain network
based on consumers green expectations by proposing a goalprogramming model considering three consumer segments, i.e.,
green, inconsistent, and red consumers. They solved a hypothetical
real-lifelike example problem to show the value and applicability of the suggested model and studied a set of scenarios to offer
an insight on how the consumer determination level of greenness
affects the green supply network. The ndings presented a way to
measure the relations between green supply chains and consumer
behavior. Table 1 represents a summary of reviewed papers in term
of investigated problem, applied method, and main ndings of the
research.
A precise analysis of Table 1 leads to determining some gaps and
important points about the necessity of this research. These gaps
are described below:
The majority of the above cited papers in the eld of network
design ignored the design with the integration of both green and

756

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

Table 1
A summary of literature review.
Author

Investigated problem

1. RL mathematical models for supply chain

Fleischmann et al.
Systematic overview of RL system in three parts of
(1997)
production, inventory, and distribution
management.
Minner (2001)
Certainty inventory planning in the SC with
internal and external product returns with the
purpose of minimizing the cost
Lee and Chan (2009)
Dening an integrated articial intelligence model
for RL on the basis of radio frequency identication
technology
Das and Chowdhury
Modeling RL network for electrical products
(2012)
wastes with the purpose of minimizing the costs
Nikolaou et al. (2013)
A framework for the evaluation of social
responsibilities and sustainable development in RL
Ramezani et al. (2013)
A multi-objective model for design of forward and
reverse logistics network under uncertainty
Suyabatmaz et al.
(2014)
Ghayebloo et al. (2015)

Combining two simulations and analytical models

for the design of RL network
A mixed integer model for design of a forward and
reverse logistics network

Method

Main ndings

Planning approaches

Finding the differences and similarities of

RL and forward logistics

Concave minimum optimization

method

Creating balance between surplus

inventory and dumping costs

Genetic algorithm

Obtaining answers with higher quality

CPLEX

Reducing transportation and recycling

system costs
Evaluating and improving social
performance of the companies
Obtaining signicant vision for
development of the decision making
procedure
Obtaining results from the two approaches
under different scenarios
Developing some useful managerial
insights

Triple bottom line approach

An evaluation method for
evaluating the systematic supply
chain conguration
Analytical modeling approaches
The weighted sums, and the
e-constraint methods

2. The applications of meta-heuristic algorithms in RL

Min et al. (2006)
RL network design by dening a non-linear
programming model
Du and Evans (2008)
A model for designing a RL network with the aim
of minimizing the costs and maximizing customer
satisfaction
Kim et al. (2009)
Designing RL network by dening a vehicle
route-nding model

Tabu search, constraint method,

and dual simplex

Pishvaee et al. (2010)

Simulated annealing algorithm

Diabat et al. (2013)

Roghanian and
Pazhoheshfar (2014)

Govindan et al. (2015b)

Designing multi-stage logistics network by

dening a linear programming model for
minimizing the costs
RL network design by dening a non-linear
programming model for minimizing the costs
A multi-product, multi-period reverse logistics
network problem for the return products

Closed-loop supply chain network optimization by

dening a deterministic multi-product,
multi-echelon, multi-period model

3. Sustainable reverse logistics models

Korchi and Millet
Designing a RL network for reproduction of
(2011)
electrical and electronic equipment

Chaabane et al. (2012)

Pishvaee and Razmi

(2012)
Lopes Silva et al. (2013)

Ramos et al. (2014)

Fahimnia et al. (2015)

Sustainable reverse logistics network modeling

with the purposes of minimizing the costs and
environmental negative impacts
A fuzzy model for environmental supply chain
Designing a returnable packaging model of RL with
the purposes of minimizing costs and resource
consumption and environmental impacts
The vehicle routing problem with the aim of
minimizing costs and environmental harms
resulted from carbon dioxide emissions, and
minimizing the maximum number of drivers
working hours
Supply chain planning with presenting a
mixed-integer nonlinear model

Genetic algorithm
Scatter search algorithm

Genetic algorithms and articial

immune systems
Genetic algorithm

A hybrid algorithm including GA

and PSO

Changing the locations of

customers and grouping, storing,
and production centers for creating
18 structures of the reverse
logistics network
Lingo optimization software

Epsilon-constraint and
Eco-indicator 99 method
Technical analysis and Life Cycle
Assessment (LCA) technique

Obtaining solutions with higher quality

compared to the solutions obtained by
previous methods
Finding near-optimal solutions and
reducing the reopening and transportation
costs
Finding near-optimal initial places and
optimal keeping time
Satisfying the demand imposed by
manufacturing recycling centers with
minimum total cost under uncertainty
condition
The dominance of the offered hybrid
algorithm compared to the GA and PSO

Presenting structures with higher

economic benets and lower
environmental harms

Creating balance among objective

functions and designing an effective
environmental strategy
Improving the efciency of forward and
reverse logistics
Minimizing costs and waste generation of
the nal customer

GAMS software

Creating balance among objective

functions and system improvement

The Nested Integrated

Cross-Entropy (NICE) method

They found that a exible supply chain is

the greenest and most efcient alternative
when compared to strictly lean and
centralized situations
Management collaboration, green buying
and ISO 14001 certication are the most
important administration of green supply
chains
Operational category risks are the most
important risks in GSC
Presented a way to measure the relations
between green supply chains and
consumer behavior

Govindan et al. (2015a)

Green supply chain management

DEMATEL method

Mangla et al. (2015)

Analyzing the risks relevant to adoption and

effective implementation of green supply chain
Designing the green supply chain network based
on consumers green expectations by proposing a
goal-programming model

Fuzzy AHP

Coskun et al. (2015)

Creating balance among inventory and

transportation costs
Improving RL system

ILOG CPLEX solver

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

sustainable (such as social issues) and just focused on the pure

green and environmental issues. Therefore, incorporating social
issues of network conguration besides economic and environmental aspects into a model of decision-making is considered as
a gap in the literature that has to be covered.
Uncertainty is a signicant issue in the design of RL networks
and is not considered in many of the literature studies. However,
we have used fuzzy mathematical programming to cope with the
uncertain parameters.
In terms of solution methodologies, it should be noted that some
of the above mentioned studies considered epsilon-constraint as
an exact method of solving. However, in our study, a modied
epsilon-constraint method is used. In addition, meta-heuristic
algorithms used by above studies are mostly genetic algorithms,
Tabu search and simulated annealing; only a few of them applied
MOPSO in the context of RL design problems.
In all of the above mentioned studies, the economic objective is
cost minimization. Nevertheless, we have dened our economic
objective function with regard to Net present value (NPV) concept.
It should be mentioned that a great number of literature papers
are a single-period analysis, but in this study, all decision variables are determined in a multi-period study.
In this study, we have dened different collecting technologies
in collecting centers. No article reviewed in the literature considered various technologies in their RL network centers.
Using comparative metrics for comparing solutions obtained
from exact method and meta-heuristic algorithm is another gap
that is addressed in this study.

Generally, the aforementioned features distinguish this paper

from the existing ones in the related literature. Therefore, a fuzzy
multi-period, multi-objective, mixed-integer mathematical programming model is developed in this paper, which considers
economic, environmental, and social aspects of the proposed RL
network conguration. Besides, in order to elevate solving capabilities, this paper proposes an epsilon-constraint method and MOPSO
with some modications.
Proper use of the process suggested in this paper could help
managers manage the ow of recycled products efciently with
regard to environmental and social respects, a benet that will offer
a sustainable competitive advantage to corporations. Beyond social
and environmental responsibilities, companies can raise their competitive advantages by thinking about RL in designing their SC. This
approach can mitigate the risks (Soleimani and Govindan, 2014),
increase cost savings, result in capital gain growth, and cause more
favorable relationships with customers.
3. Problem denition and model formulation
In this section, rstly, the problem is dened and then the
indices, parameters, and decision variables of the model as well
as model formulation have been presented.
3.1. Problem denition
In this article, a real industrial problem in the sustainable reverse
logistics network is investigated. To adapt to real life problems and
to increase the applicability of the model, a real case of a medical
syringe and needle producer in Iran, provided by (Pishvaee and
Razmi, 2012), has been used. We used the information about this
case to determine the indices, parameters, and decision variables
and to build the meaningful formulation of the reverse logistics
networks mathematical model. Around 70% of domestic demands
are addressed by this producer. Recently, this producer has received

757

Fig. 1. Reverse logistics network.

orders from two neighboring countries. Currently, it produces 600

million items annually.
Medical syringes have substantial negative impact on the environment at the end of their life cycle. According to the statistics
presented by the World Health Organization, 16 billion injections
are made across the world annually (WHO, 2005). It is estimated
that infected medical syringes and needles cause 816 million cases
of Hepatitis B, 2.34.4 million cases of Hepatitis C, and 80160 thousand cases of AIDS. Therefore, managing the end of the life cycle of
these products is a vital issue (Hanson and Hitchcock, 2009).
In order to deal with the possible dangers, safety packages
are applied to collect infected syringes and needles after their
use (WHO, 2003). There are three end of life methods to manage
the used syringes: (1) incineration method; (2) non-incineration
methods; and (3) recycling. Incineration is a common method for
managing the end of the life cycle of medical syringes. Although
this method is cheap, easy and capable of energy recovery, it harms
the environment and causes air pollution (WHO, 2005). The nonincineration methods, such as disinfection by microwaves, can help
companies to recycle energy with various degrees of effectiveness.
Generally, the process of recycling medical wastes is considered as a
hazardous activity (Zhao et al., 2009). Recycling of medical syringes
and needles in India has been investigated under the direction of
WHO. The results showed that this method can be used for medical syringes and needles, if is accompanied by disinfecting methods
(WHO, 2005), (Pishvaee and Razmi, 2012). This article focuses on
two life cycle methods: (1) incineration and (2) recycling.
The structure of multi-period, single-product, three-echelon,
reverse logistics network has been shown in Fig. 1. On this RL network, at the expiration of the life cycle, products are transferred
from customers to collecting centers. In collecting centers, the steel
and plastic parts of syringes are separated according to different
collection technologies. Then, they are transported to recycling centers. Those parts of the syringes which cannot be recycled are also
transported to incineration centers. It is assumed that all the wasted
syringes are collected from Customer areas in every period.
Therefore, in designing the RL network, we have to optimize
three objective functions: (1) minimization of present value of
costs, (2) minimization of environmental impacts, and (3) maximization of social responsibility. Based on Life Cycle Assessment
(LCA), a method called Eco-indicator 99 has been used to quantify
and to model the environmental impact (Goedkoop and Spriensma,
2000). LCA is a procedure that enables researchers that measure the
overall environmental burden and their possible impacts over the
entire life cycle of a process, activity, or product (Rebitzera et al.,
2004; Pishvaee and Razmi, 2012). To quantify and model factors
of social responsibility, job opportunities and average lost days
resulted from harms at work have been considered (Pishvaee et al.,
2012).
However, because of imprecise and sometimes unavailable data
and special parameters, criteria that all network design models face
in the real world, simple crisp models cannot be used. Therefore,

758

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

because of the existence of uncertainties in the problem of designing a RL network, we have used fuzzy mathematical programming
(Pishvaee et al., 2011; Zadeh, 1978).

eitst

3.2. Indices, parameters, and variables of the model

eitin

The following notations, parameters, and variables are used

in modeling multi-objective sustainable reverse logistics network
design:
j
k
l
m
n
tc
t
d jt
i
rjt
tc
g kt
ajkt
bklt

hkmt

oknt
tc

kt

lt

mt
 nt

k
l
m
 n
p tktc
 ttc
cc
eijkt

cs
eiklt

cp

eikmt
ci
eiknt

co
eitct

index of xed places of customers areas j = 1, 2, 3, . . ., J

index of candidate places for collecting centers k = 1, 2, 3,
. . ., K
index of recycling centers for steel parts l = 1, 2, 3, . . ., L
index of recycling centers for plastic parts m = 1, 2, 3, . . .,
M
index of centers for incineration n = 1, 2, 3, . . ., N
index of various collection technologies at collecting centers tc = 1, 2, 3, . . ., Tc
index of time periods t = 1, 2, 3, . . ., T
demands of customer area j in period t
rate of annual discount
rate of returns from customer area j in period t
xed costs of establishment of collecting center k with
collection technology tc in period t
cost of transportation of each unit of used syringe from
customer area j to collecting center k in period t
cost of transportation of steel part of each unit of used
syringe from collecting center k to steel recycling center
l in period t
cost of transportation of plastic part of each unit of used
syringe from collecting center k to plastic recycling center
m in period t
cost of transportation of each unit of used syringe from
collecting center k to incineration center n in period t
cost of processing of each unit of used syringe at collecting
center k with technology tc in period t
cost of processing of steel part of each unit of used syringe
at steel recycling center l in period t
cost of processing of plastic part of each unit of used
syringe at plastic recycling center m in period t
cost of processing of each unit of used syringe at incineration center n in period t
highest capacity of collecting center k
highest capacity of steel recycling center l
highest capacity of plastic recycling center m
highest capacity of incineration center n
number of job opportunities created if a collecting center
is opened at location k with technology tc in period t
average lost days resulted from harms at work per worker
when technology tc is used in period t
environmental impact of transporting each unit of used
syringe from customer area j to collecting center k in
period t
environmental impact of transporting steel part of each
unit of used syringe from collecting center k to steel
recycling center l in period t
environmental impact of transporting plastic part of each
unit of used syringe from collecting center k to plastic
recycling center m in period t
environmental impact of transporting each unit of used
syringe from collecting center k to incineration center n
in period t
environmental impact of processing each unit of used
syringe at collecting centers by technology tc in period
t

environmental impact of recycling steel part of each unit

of used syringe in period t
environmental impact of recycling plastic part of each
unit of used syringe in period t
environmental impact of incinerating each unit of used
syringe in period t
quantity of transported used syringe from customer area
j to collecting center k with technology tc in period t
quantity of transported steel part of used syringe from
collecting center k with technology tc to steel recycling
center l in period t
quantity of transported plastic part of used syringe from
collecting center k with technology tc to plastic recycling
center m in period t
quantity of transported used syringe from collecting center k with technology tc to incineration center n in period
t

1 If a collecting center with technology
tc is established in position k in period t
0 Otherwise

pl

eit

c
qtjkt
c
vtklt

tc
wkmt

tc
zknt

tc
ykt

It must be noted that those parameters, which have been written

under tilde, are fuzzy parameters.
3.3. Model formulation
The model has three objective functions and ten constraints.
3.3.1. Objective functions
The proposed model has economic, environmental, and social
objective functions.
3.3.1.1. Economic objective function. In order to dene an economic
objective function, rst the concept of Net present value (NPV) is
explained.
The NPV of an investment can be dened as follows:
NPV =

Present-Value-of-Benets

Present-Value-of-Costs

The following formulation converts future value (F) to present

value (P):
n

P = F[(1 + i) ]

(1)

where i is discount rate per period, and n is number of periods.

Since there is no selling parameter (and therefore no income
parameter) in the proposed model, the Present Value of Benets is
equal to zero. Therefore, the maximization of net present value is
equal to the minimization of the present value of costs:
NPV =

Present-Value-of-Costs

The overall costs of SRLND include xed costs of establishing collection centers as well as the variable costs of operational
processing and transporting materials among various centers of the
network.
The costs of transportation among various centers are obtained
by multiplying the cost of transporting each unit of used syringe
between two centers by the quantity of transport products between
the two centers. Therefore, the rst objective function is:
Min W1



tc tc
tc
c
= Min
ykt +
(
kt
+ ajkt )qtjkt
g kt
k

tc

tc

K. Govindan et al. / Ecological Indicators 67 (2016) 753768


k

lt + bklt )vtc +
(
klt

tc


n



tc
(mt + hkmt )wkmt

tc

tc
( nt + oknt )zknt
/(1 + i)

(2)

tc

that in this study, the selection of technology type in collecting

centers has a major impact on working conditions.
The third objective function:
Max W3 =


 
t

3.3.1.2. The environmental objective function. Tools based on the

concept of life cycle thinking (LCT) are routinely used to measure the environmental impact of systems, and LCA is used to
quantify the environmental impacts (Rigamonti et al., 2016). This
method enables us to give numerical values to those factors
which affect the environment, and to measure their potential
effects during the life cycle of a product (Rebitzer et al., 2004).
The eco-indicator 99 methodology sums up the results obtained
by LCA. In other words, it transforms them into understandable
units. These units are called environmental indices. Environmental
indices are numerical values that explicitly identify the environmental aspects of a process or a product. This method considers
three areas of damage: (1) human health, (2) ecosystem quality, and
(3) resources.
The eco-indicator 99 methodology has three different views: (1)
hierarchical, (2) individualist, and (3) egalitarian. In this study, the
calculation of values of environmental indices is based on the mean
hierarchical view. In this view, human health, quality of ecosystem,
and resources are assigned different scores (i.e., 40%, 40%, and 20%,
respectively) (Goedkoop and Spriensma, 2000).
In order to use eco-indicator 99 methodology, rst, the limits
of the system, performance unit, and the objective of using the
index should be determined. In the next step, the stages of life
cycle should be delineated. The stages of reverse logistics investigated in this study include (1) the transportation of used syringes
from customers to collecting centers (cc); (2) checkup operation
and categorizing (co); (3) transportation from collecting centers to
incineration centers (ci); (4) the incinerating of used syringe (in);
(5) transportation from collecting centers to steel recycling centers (cs); (6) steel recycling (st); (7) transportation from collecting
centers to plastic recycling centers (cp); (8) plastic recycling (pl).
In the third step, materials and processes in the stages of life cycle
should be quantied. In the last step, the nal score is obtained,
which includes: (1) nding the related Eco-indicator (Goedkoop
and Spriensma, 2000); (2) multiplying the quantities by the indicator values, and (3) adding up the supplementary results (Pishvaee
and Razmi, 2012).
The second objective function:

759

ytc
p tc
kt kt

tc


k

tc
 ttc ykt

(4)

tc

3.3.2. Constraints
The proposed model includes the following ten constraints:
3.3.2.1. Return satisfaction constraint.


k

j, t

c d
jt rjt
qtjkt

(5)

tc

Constraint (5) guarantees that all products used by customers

in every period are collected.
3.3.2.2. Technology assignment constraint.

tc
ykt
1 k, t

(6)

tc

Constraint (6) guarantees that in every period, at most one type

of technology is allocated to each collecting center.
3.3.2.3. Flow balance constraints.

c =
qtjkt

tc
wkmt
+

tc
wkmt
=

tc
zknt

k, tc , t

(7)

n
c
vtklt
k, tc , t

(8)

Constraints (7) and (8) guarantee that the ow in collecting

centers is balanced. Because two EOL methods are included in the
suggested model, the used syringes should rstly be transported to
incineration centers or the centers for separating steel and plastic
parts. Then, they should be transported to related recycling centers.
Hence, the number of plastic parts and the number of steel parts
transported to recycling centers should be the same.
3.3.2.4. Capacity constraints.

c y tc 
qtjkt

kt k

k, tc , t

(9)

c
l l, tc , t
vtklt

(10)

Min W2

 

co
cc
cs
c +
c

=
(eitct
+ eijkt
)qtjkt
(eitst + eiklt
)vtklt
t



tc

tc

pl

cp

tc
(eit + eikmt )wkmt
+

tc


k

ci
tc
(eitin + eiknt
)zknt

tc

(3)

3.3.1.3. Social objective function. In order to investigate the social

aspects, two indices are taken into account for social responsibility:
(1) the number of created job positions; (2) the average lost days
resulted from harms at work. The rst index is applied for both
the working conditions and the improvement of society (Veleva
and Ellenbecker, 2001). The second index is related to any harm
inicted on employees as a result of working conditions. It is clear

tc
m
wkmt

m, tc , t

(11)

tc
zknt
 n

n, tc , t

(12)

Constraints (9)(12) are related to the capacity of collecting

centers, steel recycling centers, plastic recycling centers, and incineration centers, respectively. In addition, constraint (9) guarantees
that the used syringes are not transferred to non-established collecting centers.
3.3.2.5. Decision variables constraints.
tc
ykt
{0, 1} k, tc, t
tc
tc
qtc
, vtc
, wkmt
, zknt
jkt
klt

0 j, k, l, m, n, tc, t

(13)
(14)

Constraints (13) and (14) are related to the binary and nonnegativity restriction on the corresponding decision variables.

760

K. Govindan et al. / Ecological Indicators 67 (2016) 753768


n


pes
((nt

mos
+ 2nt

opt
tc
+ nt /4) + oknt )zknt

/(1 + i)

(15)

tc

Min W2

 

co
cc
cs
c +
c

=
(eitct
+ eijkt
)qtjkt
(eitst + eiklt
)vtklt
t

The possibilistic programming method has been used to deal

with uncertain and vague parameters in the proposed sustainable reverse logistics network design (SRLND) model. As shown
in Fig. 2, a three-stage approach is proposed to solve the model.
In the rst stage, the initial fuzzy model is transformed into an
equal crisp model. In the second stage, because the problem is NPhard, the MOPSO algorithm, which is a meta-heuristic algorithm,
is used. Moreover, the model is solved by the epsilon-constraint
(epsilon-constraint method using codication and LINGO 14.0 optimization software). In the third stage, the rst four comparison
metrics (exactness and diversity metrics) are dened and afterward
the solutions obtained by the MOPSO algorithm (codied in MATLAB software) and e-constraint methods are compared on the basis
of four comparison metrics and computational time by ANOVA in
MINITAB 16.0 software.

Max W3 =

= Min




j

t mos

opt

tc
+ gktc /4)ykt

ci
tc
+ eiknt
)zknt

(16)

tc

pes

t mos

(pktc + 2pktc

opt

tc
+ pktc /4)ykt

tc




pes

t
(t c

t mos
+ 2t c

opt

t
+ t c

tc
/4)ykt

(17)

tc

s.t.



opt

tc

[(rjtmos

opt
pes
+ rjt /2) + (1 )(rjt

+ rjtmos /2)]

c =
qtjkt

(19)

c +
vtklt

tc
wkmt
+

tc
wkmt
=

(18)

j, t

tc
ykt
1 k, t

tc

pes

c
qtjkt
[(djtmos + djt /2) + (1 )(djt + djtmos /2)]

tc
zknt

k, tc , t

(20)

c
vtklt
k, tc , t

(21)

l
pes

c ytc [(
qtjkt
kt
k

opt

+ mos
/2) + (1 )(mos
+ k /2)] k, tc , t
k
k

(22)

c [(pes + mos /2) + (1 )(mos + opt /2)]

vtklt
l
l
l
l

l, tc , t

(23)

pes

opt

tc
mos
wkmt
[( m + mos
m /2) + (1 )( m + m /2)] m, tc , t

tc

pes

opt

tc
zknt
[(n + nmos /2) + (1 )(nmos + n /2)] n, tc , t

(25)

k
t

pes

t mos

((ktc + 2ktc

opt

tc
+ ktc /4) + ajkt )qjkt

pes

opt

tc
((lt + 2ltmos + lt /4) + bklt )vklt

tc


k

pes

tc


k

(gktc + 2gktc

tc


(eitin

(24)

Min W1

cp

tc

In this study, the method presented by Jimenez et al. (2007) is

chosen to solve the possibilistic programming model. This method
is established on mathematical concepts, such as expected values and interval of fuzzy numbers, and moreover general ranking
method which can provide various membership functions. Based
on this method we used an expected value of fuzzy parameters to
convert fuzzy objective functions to its crisp form. Then, based on
the method proposed by Jimenez et al. (2007), Jimenez (1996), and
(Pishvaee and Razmi, 2012), the fuzzy constraints are converted
into their crisp form by determination of the least satisfactory feasibility degree of decision vector.
Based on the mentioned points, the proposed SRLND crisp model
is as follows:

pl

tc

tc
(eit + eikmt )wkmt

 

4.1. The corresponding crisp model


k

Fig. 2. A schematic view of the proposed approach.

4. Solving methodology and comparison metrics

tc



pes

opt

mos
tc
((mt + 2mt
+ mt /4) + hkmt )wkmt

tc
ykt
{0, 1} k, tc , t
c , vtc , w tc , z tc
qtjkt
klt
kmt
knt

0 j, k, l, m, n, tc , t

(26)
(27)

4.2. Multi-objective particle swarm optimization algorithm

In nature, birds search for food based on their own experiences
as well as the experiences of other members of the group. The
particle swarm optimization algorithm, which is a meta-heuristic
algorithm, originates from the views of Kennedy and Eberhart

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

(1995). It belongs to a category of algorithms used to nd the

optimal or near-optimal solutions of numerical problems. First, a
number of birds (solutions) are produced randomly. Then, their
speeds are assigned. Based on the current speed of the bird and
its distance from the best position seen by the bird and its distance
from the best position found by neighboring birds, a new speed
is calculated for the bird. Because the obtained speed is equal to
the distance, which has been covered by the bird in one stage, the
new position of the bird in the next stage can be determined. This
process is repeated for a number of times. Finally, the best position found by the birds is selected as the solution of the problem
(Khalili-Damghani et al., 2013).

761

based on their goal values. Consider the following mathematical

multi-objective programming problem:
min(f1 (x), f2 (x), . . ., fp (x))
(29)

s.t.
x S,

In the epsilon-constraint method, we optimize one of the

objective functions considering the other objective functions as
constraints, as shown below (Mavrotas, 2009):
min f1 (x)
s.t.

4.2.1. PSO algorithm from a mathematical perspective

Consider a group consisting of m particles which are looking
for the optimal value of objective functions in a n-dimensional
search space. Each particle has three vectors: (1) position vector,
 = (x , x , . . ., x ), i = 1, 2, 3,. . ., m which is associated with a
X
i
i1
i2
in
solution; (2) velocity vector, V i = (vi1 , vi2 , . . ., vin ), i = 1, 2, 3, . . ., m
which shows the movement of each particle in each dimension for
the improvement of current position; (3) particles best position
 i = (pi1 , pi2 , . . ., pin ), i = 1, 2, 3, . . ., m which shows the best
vector, p
tted position of the particle from the rst step of the algorithm
(Khalili-Damghani et al., 2013).
The tness of a position can be measured using the objective
function of the main optimization problem. A vector of a global best
 g = (pg1 , pg2 , . . ., pgn ), is reserved to store all information
particle p
transferred among all particles of the swarm. Based on the mentioned points, both the velocity vector and position vector of each
particle in the swarm are updated with the model (28) through the
multiple iterations of the algorithm:


 (t)) i = 1, 2, . . .m
V (t + 1) = W V i (t) + c1 r1 (Pi X
i

 (t))
+ c2 r2 (Pg X
i

 (t + 1) = X
 (t) + V (t + 1)
X
i
i
i

(28)

f2 (x) 2
f3 (x) 3

(30)

...
fp (x) p
x S,
By changing the right hand side of constrained objective functions (i ), the efcient solutions are obtained for the problem. All
of the solutions are gained and the decision maker can make the
decision based on available information (Khalili-Damghani et al.,
2013; Pishvaee and Razmi, 2012).
4.4. Comparison metrics
In order to assess the exactness and diversity of various methods
of Pareto front production in multi-objective problems, comparison
metrics are used. Several metrics proposed by Yu and Gen (2010)
have been used to study the exactness and the diversity of both
suggested processes in regenerating the Pareto front of the SRLND
problem.

i = 1, 2, . . .m

W is the weight of inertia, which shows the tendency of particle

to remain in its previous exploration direction (Shi and Eberhart,
1999). C1 and C2 are cognitive and social factors, respectively; r1
and r2 [0,1] are random numbers, and t shows the number of
iterations.
4.2.2. Multi-objective PSO
The solution methods of multi-objective problems, including
meta-heuristic, should have two main features. Firstly, they should
produce non-dominated solutions with high quality on the Pareto
front of multi-objective decision-making problems. Secondly, the
produced solutions should be properly dispersed on the Pareto
front of multi-objective decision-making problems. Simplicity and
successfulness of simple PSO algorithm prepared the ground for the
development of its multi-objective versions (Coello et al., 2004). In
this research, the dynamic self-adaptive multi-objective particle
swarm optimization method proposed by (Khalili-Damghani et al.,
2013) have been customized to solve the problem.

4.4.1. Diversity metrics

Diversication (DM)
By this metric, diversication among non-dominated solutions
is measured. In the following formula, |xi yi | is Euclidean space
between the non-dominated solution xi and the non-dominated
solution yi , and n is the number of Pareto solutions. The higher is
the value of this metric, the better. A high value shows that there
are fewer equal solutions and there is more variety among the
solutions (Zitzle et al., 1998; Tavana et al., 2013):


 n

d=
max(|x y |)
i

(32)

i=1

Spacing (SM)
This metric, introduced by Schott (1995), shows the relative
distance between successive solutions. In fact, this metric shows
the uniform distribution of Pareto solutions in the search space.
It can be obtained by the following equation:

n1
i=1

|di d|

4.3. Epsilon-constraint method

S=

The epsilon-constraint method is identied as a posteriori

method that can present a suitable picture of a Pareto optimal set
for decision makers; then, the decision maker can choose the most
favored solution (Hwang and Masud, 1979; Pishvaee and Razmi,
2012). In the epsilon-constraint method, one of the multiple objectives is taken as objective function; the other objectives convert
to constraints with right hand side numbers, which are calculated

In this equation, di is the Euclidean space between two sides

of the Pareto solutions, d is the mean of spaces di and n is the
number of Pareto solutions.
When the solutions are distributed uniformly, the value of s
is small. Therefore, algorithms are better in which the spacing
values among nal, non-dominated solutions are small (Schott,
1995).

(n 1) d

(33)

762

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

Moreover, the computational time of the algorithm (T) is considered as a metric for evaluation. The shorter value for this time,
the better.

5. Experimental results

Fig. 3. The number of Pareto solutions in multi-objective subjects.

4.4.2. Exactness metrics

The mean ideal distance (MID)
The value of this metric is obtained by measuring the distance between Pareto points of the algorithm and the ideal point
(0,0). The MID metric can be calculated by the following equation
(Khalili-Damghani et al., 2015):
2

(f1i ) + (f2i ) + (f3i )

i=1

(34)

In this equation, f1i is the value of the rst objective function

by putting the solution of i in it; f2i is the value of the second
objective function by putting the solution of i in it; f3i is the value
of the third objective function by putting the solution of i in it,
and n is the number of Pareto solutions.
Since in multi-objective subjects based on the Pareto approach
one of the aims is to have fronts which are as much as possible
close to the basis of coordinates, this metric is used to show the
distance between fronts and the best value of the population. The
smaller the value of this metric, the more efcient the algorithm
(Zitzle et al., 1998).
The number of non-dominated Pareto solutions (NNS)
The value of this metric shows the number of Pareto optimal
solutions that can be found in every algorithm. Fig. 3 shows the
way in which this metric is calculated (Zitzle et al., 1998).
Table 2
The dimensions of test problems.
Indices

Small-size

Large-size

J
K
L
M
N
Ttc
T

15
11
4
4
3
3
3

25
20
6
6
6
9
9

DM

MID

NNS

MOPSO

e-Constraint

MOPSO

1
2
3
4
5
6
7
8
9
10

1,259,911.843
1,245,593.600
1,217,877.616
1,315,200.867
1,223,753.451
1,207,711.856
1,342,830.795
1,154,593.004
1,169,528.761
1,244,405.136

1,919,570.606
1,846,624.047
1,818,847.576
1,870,520.199
1,836,415.369
1,697,314.198
1,629,181.139
1,711,523.811
1,586,519.466
1,567,841.825

13
15
24
24
18
11
19
20
17
20

Ave.
Std. Dev.

1,238,140.69
58,403.17

1,748,435.82
126,297.51

18.10
4.28

e-Constraint
10
9
9
8
8
8
7
7
7
7
8.00
1.05

Table 5
Computational times (min) for the small-size problems.
Problem

T
MOPSO

e-Constraint

1
2
3
4
5
6
7
8
9
10

15.36
14.12
14.93
15.10
13.32
14.02
14.28
13.59
15.65
13.93

26.88
27.11
28.57
26.56
27.06
27.98
26.95
27.24
26.80
26.74

Ave.
Std. Dev.

14.43
0.78

27.19
0.62

Table 6
The results of the comparison metrics and computational times (min) for the largesize problems for MOPSO algorithm.

Table 3
The results of the small-size problems in diversity metrics.
Problems

Table 4
The results of the small-size problems in exactness metrics.
Problem

n 


MID =

To check the applicability and the quality of the proposed

solution procedures, we statistically simulated the parameters
of the model using uniform probability density function and
generated several small-size and large-size test problems. The
elements of the generated test problems have been determined
based on the proposed real case of the syringe and needle producer. The dimensions of the test problems are demonstrated

Problems

SM

MOPSO

e-Constraint

MOPSO

e-Constraint

1
2
3
4
5
6
7
8
9
10

2369.4379
3285.0066
4532.612
6930.3414
2628.6919
1807.852
3517.706
1887.3237
1567.5108
3501.7059

4079.2796
4002.2814
3971.1458
3879.7453
3852.4628
3835.8426
3758.7358
3710.4001
3708.4896
3685.0903

1.0206
1.3198
1.6467
1.5912
0.7148
0.9485
0.9009
0.6225
0.8618
0.9643

1.8397
1.8426
1.8405
1.8412
1.8433
1.8396
1.8414
1.8363
1.8396
1.8358

Ave.
Std. Dev.

3202.82
1606.68

3848.35
136.06

1.06
0.35

1.84
0.00

MOPSO

DM

SM

MID

NNS

1
2
3
4
5
6
7
8
9
10

4210.4693
2474.0561
4210.469
2474.0561
3183.1896
7125.7559
4648.5662
3049.2418
3906.7567
3573.814

0.6169
0.7948
0.6169
0.7948
0.6932
1.3844
0.8912
0.9676
1.0772
0.9097

4,812,878.66
4,658,313.26
4,812,878.66
4,658,313.26
4,705,307.66
4,972,501.24
4,724,075.39
4,635,683.84
4,695,719.4
4,934,414.85

22
13
22
14
17
20
29
22
23
11

434.06
439.62
443.99
448.13
392.05
430.87
421.96
479.72
387.23
411.45

Ave
Std. Dev.

3585.64
1356.76

0.87
0.23

4,761,008.62
118,184.49

19.50
5.50

428.91
27.43

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

in Table 2. The results from the application of both aforementioned solving methods to the test problems, executing the MOPSO
algorithm and the epsilon-constraint method 30 times, are as
follows:
5.1. Results of small-size test problems
The mean of 30 values of diversity metrics obtained by both solving methods is given in Table 3. According to Table 3, it is obvious
that the average of the DM metric in epsilon-constraint methods is higher than MOPSO algorithm. It means that the variation
among the solutions obtained by the epsilon-constraint method is

763

higher. In addition, non-dominated solutions obtained by MOPSO

have a smaller average for SM metric compared to non-dominated
solutions obtained by the epsilon-constraint method. Hence, the
non-dominated solutions found by MOPSO have a higher degree of
uniformity in terms of distribution. The mean of 30 values of exactness metrics obtained by both solving methods is represented in
Table 4.
It can be seen in Table 4 that the average of the MID metric is
smaller in MOPSO methods compared with the epsilon-constraint
method. Therefore, the mean of distances between MOPSO nondominated solutions and (0,0) is smaller than the mean of distances
between epsilon-constraint non-dominated solutions and (0,0).

Fig. 4. The diagram of obtained Pareto solutions in three-dimensional space: (a) small-size, and (b) large-size.

Fig. 5. Generated objective values for 30 non-dominated solutions in the small-size problems: (a) economic, (b) environmental, and (c) social.

764

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

Comparison of Diversity

Comparison of Spacing
2.00

e-constraint

MOPSO

e-constraint

6000

MOPSO

1.83515

1.75

5000

Diversity

Spacing

1.50
1.25

4000
3497.05

3000

1.00

0.93384

2460.49

0.75

2000

0.50

1000
Panel variable: Codes

Panel variable: Codes

(a) Comparison of spacing

(b) Comparison of diversity

Comparison of MID

Comparison of Non-dom inated pareto solutions' numbers

e-constraint

MOPSO

MOPSO

2000000

25

1800000

20

MID

Number of non-dominated pareto

30

e-constraint

15

1600000

1400000

10
1200000
5
1000000
Panel variable: Codes

Panel variable: Codes

(d) Comparison number of Pareto solutions

(c) Comparision of MID

Comparison of time
140

e-constraint

MOPSO

120

Time

100
80
65.4553

60
40

29.2713

20
0
Panel variable: Codes

(e) Comparison of time

Fig. 6. The plot box of comparison metrics (a) DM, (b) SM, (c) MID, (d) NNS, and (e) T.

This shows that MOPSO performs more efciently. The average of

NNS metrics in MOPSO method is larger than the epsilon-constraint
method. Therefore, MOPSO method produces more non-dominated
solutions. Table 5 shows the computational time of both solving
methods for small-size test problems. As it can be seen, the mean

of computational time of MOPSO algorithm is shorter than epsilonconstraint method.

To sum up, in small-size test problems, the MOPSO algorithm
performs better than epsilon-constraint method based on three
metrics out of four. Moreover, it has shorter computational time.

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

765

Fig. 7. KolmogorovSmirnov test for all metrics in both procedures.

Accordingly, MOPSO algorithm is relatively more efcient than

epsilon-constraint method.
Fig. 4 shows the diagram of obtained Pareto solutions in threedimensional space in small- and large-size test problems.

5.2. Results of large-size test problems

The large-size test problems were solved only by MOPSO algorithm and the epsilon-constraint method was not successful to

766

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

Table 7
Analysis of variance.
Source

Degree of freedom

Sum of square

I. DM
Codes
Error
Total

1
193
194

37,192,711
37,192,711
5,1807,473
268,433
89,000,184

S = 518.1; R2 = 41.79%; R2 (adj) = 41.49%

II. SM
Codes
Error
Total

1
193
194

III. MID
Codes
Error
Total

Mean square

P-value
138.56

0.000

28.1200
28.1200
4.0666
0.0211
32.1867

S = 0.1452; R2 = 87.37%; R2 (adj) = 87.30%

1334.56

0.000

1
193
194

1.6537
1.6537
8.3750
0.043
1.0028

S = 208,312; R2 = 16.49%; R2 (adj) = 16.06%

38.11

0.000

IV. NNS
Codes
Error
Total

1
193
194

4819.85
4819.85
1134.53
5.88
5954.38

S = 2.425; R2 = 80.95%; R2 (adj) = 80.85%

819.92

0.000

V. T
Codes
Error
Total

1
193
194

45,321
45,321
131,745
683
177,066

S = 26.13; R2 = 25.60%; R2 (adj) = 25.21%

66.39

0.000

obtain any solution. Therefore, ANOVA is not applied to these problems and only the average values of metrics and computational time
are obtained by 30 iterations of the MOPSO algorithm. These values
are shown in Table 6.
Since in small-size test problems, MOPSO algorithm performs
relatively better than epsilon-constraint method, it can be concluded that MOPSO performs more efciently in large-size test
problems as well.

Table 8
Calculated condence interval for the comparison metrics and the computational time.
Individual 95% CIs for DM mean based on pooled Std. Dev.
Level
N
Mean
e-Constraint
150
3497.0
45
2460.5
MOPSO

Std. Dev.
278.9
956.1

Individual 95% CIs for SM mean based on pooled Std. Dev.

N
Mean
Level
e-Constraint
150
1.8351
45
0.9338
MOPSO

Std. Dev.
0.0125
0.3031

Individual 95% CIs for MID mean based on pooled Std. Dev.
N
Mean
Level
e-Constraint
150
1,424,092
45
1,205,516
MOPSO

Std. Dev.
230,980
98,349

Individual 95% CIs for NNS mean based on pooled Std. Dev.
N
Mean
Level
e-Constraint
150
6.067
45
17.867
MOPSO

Std. Dev.
1.616
4.115

Individual 95% CIs for T mean based on pooled Std. Dev.

N
Mean
Level
e-Constraint
150
65.46
45
29.27
MOPSO

Std. Dev.
29.73
0.84

Fig. 5 denotes the values of the objective functions of both

methods and Fig. 6 shows the plot box of comparison metrics in
small-size test problems.
5.3. Statistical analysis
In this section, ANOVA is used to make a meaningful comparison
between solving methodologies in small-size test problems. Fig. 7

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

represents the KolmogorovSmirnov test results. These results

stated that the normality of the population was not rejected. Therefore, the parametric test is applied for performance evaluation of
the procedures.
Afterward, ANOVA has been applied to check the existence of
meaningful differences between the performances of the solving
methods in all of the metrics. Table 7 represents the results of
ANOVA. Since the p-value 0.05, we can conclude that the hypothesis of equal means for all of the metrics is rejected. A 95% condence
interval (CI) of the means of all metrics is illustrated in Table 8.
It is clear that there is a meaningful difference between the performances of methods in all of the metrics because none of the CIs
include zero in their intervals.
Furthermore, Table 8 shows the mean values of the metrics.
Based on these values, it can be concluded that MOPSO method
outperforms the epsilon-constraint method in aspects of quality
of solutions and computational time. In other words, compared
to the epsilon-constraint method, MOPSO produces more nondominated solutions, which are more uniformly distributed in the
search space, and the average value of its non-dominated solutions
is closer to the point (0,0).

6. Conclusions and future research

In this article, the subjects of reverse logistics and sustainability
were combined and a sustainable reverse logistics model was
presented with three objective functions: the minimization of
the present value of costs, the minimization of environmental
impacts, and the maximization of social responsibility. Incorporating social issues into the RL network (in addition to the economic
and environmental aspects) covers a gap in the literature we
discovered; this study considers the social responsibility of the
recycling system and quanties it as an objective function of the
proposed model. To cope with uncertainty, fuzzy mathematical
programming was employed.
In the solving procedure, rst the fuzzy model was converted into a corresponding crisp model. Then, the meta-heuristic
algorithm of MOPSO and the epsilon-constraint method were
used to solve the model. The Pareto solutions obtained by the
two methods were compared by analysis of variance in terms
of comparison metrics. The obtained results demonstrated that
the modied MOPSO algorithm is relatively more efcient than
epsilon-constraint method according to the solutions quality and
computational time. In order to show the applicability of the model
in the real world, the suggested model was applied to a medical
syringe recycling system. Based on the case study, there were some
difculties with the medical syringe and needle producer about
obtaining access to the plants information.
For future study, researchers can change the denition of objective functions, especially the social objective to include issues such
as the maximization of the workplaces sanitary conditions or the
education of workers. Moreover, the application of the proposed
model and solution procedures in other medical and recyclable
products can be an interesting research topic. In addition, in this
paper, the fuzzy logic was applied to confront with uncertainties.
However, other approaches can apply to deal with uncertainties
such as stochastic programming.

References
lvarez, M., Berrone, P., Husillos, F., Lado, N., 2007. Reverse logistics, stakeholders
inuence, organizational slack and managers posture. J. Bus. Res. 60, 463473.
Aras, N., Aksen, D., Tanugur, A.G., 2008. Locating collection centers for incentivedependent returns under a pick-up policy with capacitated vehicles. Eur. J. Oper.
Res. 191, 2340.

767

Ashby, A., Leat, M., Hudson-Smith, M., 2012. Making connections: a review of supply
chain management and sustainability literature. Supply Chain Manag. 17 (5),
497e516.
Bjrklund, M., Martinsen, U., Abrahamsson, M., 2012. Performance measurements
in the greening of supply chains. Supply Chain Manag. 17, 2939.
Bouzon, M., Govindan, K., Rodriguez, C.M.T., Campos, L.M.S., 2016. Identication and
analysis of reverse logistics barriers using fuzzy Delphi method and AHP. Resour.
Conserv. Recycl.
Chaabane, A., Ramudhin, A., Paquet, M., 2012. Design of sustainable supply chains
under the emission trading scheme. Int. J. Prod. Econ. 135, 3749.
Chen, P., 2012. The investment strategies for a dynamic supply chain under stochastic demands. Int. J. Prod. Econ. 139, 8089.
Coello, C.A.C., Pulido, G.T., Lechuga, M.S., 2004. Handling multiple objectives with
particle swarm optimization. IEEE Trans. Evol. Comput. 8, 256279.
Coskun, S., Ozgur, L., Polat, O., Gungor, A., 2016. A model proposal for green supply chain network design based on consumer segmentation. J. Clean. Prod. 110,
149157.
Das, K., Chowdhury, A.H., 2012. Designing a reverse logistics network for optimal
collection, recovery and quality-based product-mix planning. Int. J. Prod. Econ.
135, 209221.
Diabat, A., Kannan, D., Kaliyan, M., Svetinovic, D., 2013. An optimization model for
product returns using genetic algorithms and articial immune system. Resour.
Conserv. Recycl. 74, 156169.
Du, F., Evans, G.W., 2008. A bi-objective reverse logistics network analysis for postsale service. Comput. Oper. Res. 35, 26172634.
Fahimnia, B., Sarkis, J., Eshragh, A., 2015. A tradeoff model for green supply chain
planning: a leanness-versus-greenness analysis. Omega 54, 173190.
Fattahi, M., Mahootchi, M., Govindan, K., Husseini, S.M.M., 2015. Dynamic supply
chain network design with capacity planning and multi-period pricing. Transp.
Res. Part E: Logist. Transp. Rev. 81, 169202.
Fleischmann, M., Dekker, R., Van Der Laan, E., Van Nunen, J.A.E.E., BloemhofRuwaard, J.M., Van Wassenhove, L.N., 1997. Quantitative models for reverse
logistics: a review. Eur. J. Oper. Res. 103, 17.
Ghayebloo, S., Tarokh, M.J., Venkatadri, U., Diallo, C., 2015. Developing a bi-objective
model of the closed-loop supply chain network with green supplier selection and
disassembly of products: the impact of parts reliability and product greenness
on the recovery network. J. Manuf. Syst. 36, 7686.
Goedkoop, M., Spriensma, R., 2000. The Eco-indicator 99, A damage oriented method
for Life Cycle Impact Assessment. Methodology Report, third ed. PR Consultants, Amersfoort, Netherlands.
Govindan, K., 2015. Green sourcing: taking steps to achieve sustainability management and conservation of resources. Resour. Conserv. Recycl. 104, 329333.
Govindan, K., Cheng, T.C.E., 2015. Sustainable supply chain management: advances
in operations research perspective. Comput. Oper. Res. 54.
Govindan, K., Jafarian, A., Nourbakhsh, V., 2015a. Bi-objective integrating sustainable
order allocation and sustainable supply chain network strategic design with
stochastic demand using a novel robust hybrid multi-objective metaheuristic.
Comput. Oper. Res. 62, 112130.
Govindan, K., Seuring, S., Zhu, Q., Azevedo, S.G., 2016. Accelerating the transition
towards sustainability dynamics into supply chain relationship management
and governance structures. J. Clean. Prod. 112, 18131823.
Govindan, K., Khodaverdi, R., Vafadarnikjoo, A., 2015b. Intuitionistic fuzzy based
DEMATEL method for developing green practices and performances in a green
supply chain. Expert Syst. Appl. 42 (20), 72077220.
Govindan, K., Soleimani, H., Kannan, D., 2015c. Reverse logistics and closed-loop
supply chain: a comprehensive review to explore the future. Eur. J. Oper. Res.
24, 603626.
Hanson, J.J., Hitchcock, R.W., 2009. Towards sustainable design for single-use
medical devices. In: 31st Annual International Conference of the IEEE EMBS,
Minneapolis, MN, USA, September 26.
Henriques, I., Sadorsky, P., 1996. The determinants of an environmentally responsive
rm: an empirical approach. J. Environ. Econ. Manage. 30 (3), 381395.
Hervani, A.A., Helms, M.M., Sarkis, J., 2005. Performance measurement for green
supply chain management. Benchmarking 12, 330353.
Hwang, C.L., Masud, A., 1979. Multiple Objective Decision Making. Methods and
Applications: A State of the Art Survey. Lecture Notes in Economics and Mathematical Systems 164. Springer-Verlag, Berlin.
Jimenez, M., 1996. Ranking fuzzy numbers through the comparison of its expected
intervals. Int. J. Uncertain. Fuzziness Knowl. Based Syst. 4 (4), 379388.
Jimenez, M., Arenas, M., Bilbao, A., Rodriguez, M.V., 2007. Linear programming
with fuzzy parameters: an interactive method resolution. Eur. J. Oper. Res. 177,
15991609.
Kannan, D., Govindan, K., Shankar, M., 2016. India: formalize recycling of electronic
waste. Nature 530 (7590), 281.
Kennedy, J., Eberhart, R.C.,1995. Particle swarm optimization. In: Proceedings of the
IEEE International Conference on Neural Networks. IEEE Service Center, Piscataway, NJ, pp. 19421948.
Khalili-Damghani, K., Abtahi, A.R., Tavana, M., 2013. A new multi-objective particle swarm optimization method for solving reliability redundancy allocation
problems. Reliab. Eng. Syst. Saf. 111, 5875.
Khalili-Damghani, K., Abtahi, A.R., Ghasemi, A., 2015. A new bi-objective
location-routing problem for distribution of perishable products: evolutionary
computation approach. J. Math. Model. Algorithms Oper. Res. 14 (3), 287312.
Kim, H., Yang, J., Lee, K., 2009. Vehicle routing in reverse logistics for recycling
end-of-life consumer electronic goods in South Korea. Transp. Res. Part D 14,
291299.

768

K. Govindan et al. / Ecological Indicators 67 (2016) 753768

Korchi, A.E., Millet, D., 2011. Designing a sustainable reverse logistics channel: the
18 generic structures framework. J. Clean. Prod. 19, 588597.
Kotzee, I., Reyers, B., 2016. Piloting a social-ecological index for measuring ood
resilience: a composite index approach. Ecol. Indic. 60, 4553.
Lee, C.K.M., Chan, T.M., 2009. Development of RFID-based reverse logistics system.
Expert Syst. Appl. 36, 92999307.
Lee, C.K.M., Lam, J.S.L., 2012. Managing reverse logistics to enhance sustainability of
industrial marketing. Ind. Mark. Manag. 41 (4), 589598.
Liu, D., 2014. Network site optimization of reverse logistics for E-commerce based
on genetic algorithm. Neural Comput. Appl. 25, 6771.
Lopes Silva, D.A., Santos Ren, G.W., Sevegnani, G., Sevegnani, T.B., Serra Truzzi,
O.M., 2013. Comparison of disposable and returnable packaging: a case study of
reverse logistics in Brazil. J. Clean. Prod. 47, 377387.
Mangla, S.K., Kumar, P., Barua, M.K., 2015. Risk analysis in green supply chain using
fuzzy AHP approach: a case study. Resour. Conserv. Recycl., http://dx.doi.org/
10.1016/j.resconrec.2015.01.001.
Mavrotas, G., 2009. Effective implementation of the e-constraint method in multiobjective mathematical programming problems. Appl. Math. Comput. 213,
455465.
Min, H., Ko, H.J., Ko, C.S., 2006. A genetic algorithm approach to developing the
multi-echelon reverse logistics network for product returns. Int. J. Manag. Sci.
34, 5669.
Minner, S., 2001. Strategic safety stocks in reverse logistics supply chains. Int. J. Prod.
Econ. 71, 417428.
Murphy, P.R., Poist, R.F., 2000. Green logistics strategies: an analysis of usage patterns. Transport. J., 516.
Nikolaou, I.E., Evangelinos, K.I., Allan, S., 2013. A reverse logistics social responsibility
evaluation framework based on the triple bottom line approach. J. Clean. Prod.
56, 173184.
Pai, P.F., Chen, L.C.H., Lin, K.P., 2015. A hybrid data mining model in analyzing corporate social responsibility. Neural Comput. Appl., 112.
Pishvaee, M.S., Kianfar, K., Karimi, B., 2010. Reverse logistics network design using
simulated annealing. Int. J. Adv. Manuf. Technol. 47, 269281.
Pishvaee, M.S., Rabbani, M., Torabi, S.A., 2011. A robust optimization approach to
closed-loop supply chain network design under uncertainty. Appl. Math. Model.
35, 637649.
Pishvaee, M.S., Razmi, J., 2012. Environmental supply chain network design using
multi-objective fuzzy mathematical programming. Appl. Math. Model. 36,
34333446.
Pishvaee, M.S., Razmi, J., Torabi, S.A., 2012. Robust possibilistic programming for
socially responsible supply chain network design: a new approach. Fuzzy Sets
Syst. 206, 120.
Ramezani, M., Bashiri, M., Tavakkoli-Moghaddam, R., 2013. A new multiobjective stochastic model for a forward/reverse logistic network design
with responsiveness and quality level. Appl. Math. Model. 37 (12),
328344.
Ramos, T.R.P., Gomes, M.I., Barbosa-Pvoa, A.P., 2014. Planning a sustainable reverse
logistics system: balancing costs with environmental and social concerns.
Omega 48, 6074.
Rebitzer, G., Ekvall, T., Frischknecht, R., Hunkeler, D., Norris, G., Rydberg, T., Schmidt,
W., Suh, S., Weidema, B.P., Pennington, D.W., 2004. Life cycle assessment: Part
1: Framework, goal and scope denition, inventory analysis, and applications.
Environ. Int. 30 (5), 701720.

Rigamonti, L., Sterpi, I., Grosso, M., 2016. Integrated municipal waste management
systems: an indicator to assess their environmental and economic sustainability.
Ecol. Indic. 60, 17.
Roghanian, E., Pazhoheshfar, P., 2014. An optimization model for reverse logistics
network under stochastic environment by using genetic algorithm. J. Manuf.
Syst. 33 (3), 348356.
Rostamzadeh, R., Govindan, K., Esmaeili, A., Sabaghi, M., 2015. Application of fuzzy
VIKOR for evaluation of green supply chain management practices. Ecol. Indic.
49, 188203.
Schott, J.R., (Masters thesis) 1995. Fault tolerant design using single and multicriteria genetic algorithms optimization. Department of Aeronautics and
Astronautics, Massachusetts Institute of Technology, Cambridge, MA.
Seuring, S., Muller, M., 2008. From a literature review to a conceptual framework for
sustainable supply chain management. J. Clean. Prod. 16, 16991710.
Shi, Y., Eberhart, R.,1999. Empirical study of particle swarm optimization.
In: Congress on Evolutionary Computation. IEEE Press, Piscataway, NJ,
pp. 19451950.
Soleimani, H., Govindan, K., 2015. A hybrid particle swarm optimization and genetic
algorithm for closed-loop supply chain network design in large-scale networks.
Appl. Math. Model. 39 (14), 39904012.
Soleimani, H., Govindan, K., 2014. Reverse logistics network design and planning
utilizing conditional value at risk. Eur. J. Oper. Res. 237 (2), 487497, http://dx.
doi.org/10.1016/j.ejor.2014.02.030.
Stock, J.R., 1992. Reverse Logistics. Council of Logistics Management, Oak Brook, IL.
Suyabatmaz, A.C., Tevhide Altekin, F., Sahin, G., 2014. Hybrid simulation-analytical
modeling approaches for the reverse logistics network design of a third-party
logistics provider. Comput. Ind. Eng. 70, 7489.
Veleva, V., Ellenbecker, M., 2001. Indicators of sustainable production: framework
and methodology. J. Clean. Prod. 9, 519549.
Tavana, M., Khalili-Damghani, K., Abtahi, A.R., 2013. A fuzzy multidimensional
multiple-choice knapsack model for project portfolio selection using an evolutionary algorithm. Ann. Oper. Res. 206 (1), 449483.
World Health Organization (WHO), 2003. Procuring Single-Use Injection Equipment
and Safety Boxes.
World Health Organization (WHO), 2005. Safe Management of Bio-medical Sharps
Waste in India: A Report on Alternative Treatment and Non-Burn Disposal Practices.
Yu, X., Gen, M., 2010. Introduction to Evolutionary Algorithms. London Limited:
Springer-Verlag (Chapter 6).
Zadeh, L., 1978. Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets Syst. 1,
328.
Zhang, S., Lee, C.K.M., Chan, H.K., Choy, K.L., Wu, Z., 2015. Swarm intelligence applied
in green logistics: a literature review. Eng. Appl. Artif. Intell. 37, 154169.
Zhao, W., van der Voet, E., Huppes, G., Zhang, Y., 2009. Comparative life cycle assessments of incineration and non-incineration treatments for medical waste. Int. J.
Life Cycle Assess. 14, 114121.
Zhou, Z., Cheng, S., Hua, B., 2000. Supply chain optimization of continuous process industries with sustainability considerations. Comput. Chem. Eng. 24,
11511158.
Zitzle, E., Thiele, L., Eiben, A.E., Back, T., Schoenauer, M., Schwefel, H.P., 1998. Multiobjective optimization using evolutionary algorithms a comparative case study.
In: Fifth International Conference on Parallel Problem Solving from Nature
(PPSN-V), Berlin, Germany, pp. 292301.