Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Journal of Cleaner Production 172 (2018) 1567e1581

Contents lists available at ScienceDirect

Journal of Cleaner Production


journal homepage: www.elsevier.com/locate/jclepro

Sustainable multi-period reverse logistics network design and


planning under uncertainty utilizing conditional value at risk (CVaR)
for recycling construction and demolition waste
Mohsen Rahimi a, Vahidreza Ghezavati b, *
a
Department of Industrial Engineering, South Tehran Branch, Islamic Azad University (IAU), Tehran, Iran
b
Associate Professor of Industrial Engineering, South Tehran Branch, Islamic Azad University (IAU), Tehran, Iran

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

Article history: Decreasing the environmental effect and increasing the social effect as well as the profit of organizations
Received 23 May 2017 has gained a considerable attention in supply chain network problems. Construction and demolition
Received in revised form wastes (C&D wastes) management poses serious challenges to the government and private sector in
15 September 2017
terms of the environmental damages and potential gains due to recycling these types of wastes. On the
Accepted 20 October 2017
Available online 1 November 2017
other hand, considering the triple bottom line in reverse logistics network design problems simulta-
neously is an issue that has gained little attention in previous studies. Therefore in this research we
propose a multi-period multi-objective mixed integer linear programming to design and plan a network
Keywords:
Sustainable reverse logistics network design
of reverse logistics under uncertainty for recycling C&D wastes in which the objectives are represented as
Construction and demolition waste profit and social impact maximization and environmental effect minimization. In this paper the network
Two-stage stochastic programming design problem with stochastic demand of recycled products and rate on investment is regarded. In order
Conditional value at risk to cope with the uncertainty in the model, we consider a risk averse two-stage stochastic programming,
where we specify the conditional value at risk (CVaR) as the risk measure. We also apply off-site and on-
site separation as two common method for segregating C&D wastes. Appropriate numerical is conducted
to learn the effects of crucial parameters on the model and gain managerial insights as well.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction Considering uncertainty in the concept of the supply chain


network design and planning is inevitable. This issue would be
Nowadays, due to some economical, legal and social concerns more tangible in reverse logistics network design in which we
industries seem inevitable to deal with their End Of Life (EOL) suffer more from the lack of accurate and enough information.
products. For these reasons a network of collection and recycling There are various sources of uncertainties in a reverse supply chain
centers as well as landfills are designed for the purpose of recov- such as quality, price, time and amount of products, which do not
ering value or proper disposal. In this regard, dealing with C&D exist in a classical form of supply chain Soleimani and Govindan,
wastes, due to some reasons, is a controversial subject in the field of 2014. In this paper we consider the demand of fine and coarse
reverse logistics; firstly, burying C&D wastes can pose environ- aggregates as two fundamental products resulting from recycling
mental dangers, and secondly, some enormous capital will be lost if C&D wastes and rate on investment (ROI) as stochastic parameters.
proper actions aren't taken. For this purpose, a multi period mixed Two-stage stochastic programming deals with the uncertainty
integer linear programming model has been proposed in this paper in two stages; at first some decisions have been made and some
in order to determine the optimal number of recycling centers as scenarios are conducted considering the history of stochastic pa-
well as allocating between entities of the supply chain by the use of rameters. Then one of the predefined scenarios happens and the
two stage stochastic programming. second stage decisions have been made to mitigate the effects of
first stage decisions.
On the other hand, considering uncertainty in the mathematical
model does not guarantee to cope with the effect of variability of
* Corresponding author.
E-mail addresses: v_ghezavati@azad.ac.ir, vrghezavati@gmail.com
random outcomes. In other words, traditional two-stage stochastic
(V. Ghezavati). programming considers the expectation of random variables,

https://doi.org/10.1016/j.jclepro.2017.10.240
0959-6526/© 2017 Elsevier Ltd. All rights reserved.
1568 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

which makes it risk-neutral. For non-repetitive decision making end of life vehicles (ELVs). Their deterministic multi-period model
problems, such as location-allocation problems, risk averse consists of entities like collection and dismantling centers, shred-
approach would provide more robust solutions compared to risk ders, recycling centers and landfills. The objective of their model is
neutral approach. For this reason in our study, we use CVaR- profit-maximization with a real case study in the field of end of life
integrated two-stage programming to tackle this issue. vehicles recycling industry.
Sustainability, which is the main concept of this paper, in- Zhou and Zhou, 2015 try to propose a location-allocation mixed
tegrates economical aspect of a supply chain with environmental integer non-linear mathematical model to design a network of
and social responsibility. Seuring and Müller, 2008 define sustain- supply chain to recycle waste-paper. The proposed model is then
ability as “the designing and employing human systems as well as validated by a real case study. All parameters of the model are
industrial systems in order to use natural resources and to make considered deterministic and the model tries to find the optimal
sure that the normal cycles do not have negative impact on social number of recycling centers while the costs of network is
conditions, human health and ecosystem”. In order to include social minimized.
and environmental aspects in network design decisions, many in- Alumur et al., (2012) study designing and planning of a reverse
dicators have been used by researchers. Indicators like co2 emis- supply chain network which makes it possible for a decision maker
sion, energy used and waste generated have been conducted by to consider future adjustments in different time horizons. Adjust-
previous studies to measure the environmental effect of a supply ments like the numbers and capacity of facilities as well as the
chain. In this paper, the environmental impacts resulted from amount of products held in inventories can be considered in the
opening recycling centers, producing at recycling centers and model in different time periods.
handling C&D wastes and recycled products in terms of co2 emis- Alshamsi and Diabat, 2015 develop the model proposed by
sions related to energy consumption are modeled. Alumur et al. (2012) by considering other options in the model.
The social dimension has gained less attention in comparison to They consider two possible option for transportation of their
environmental dimension in previous studies and a comprehensive products and components in their model; 1) in-house fleet and 2)
definition has not been even presented (Eskandarpour et al. outsourcing. The model can decide on the number of trucks which
(2015)). Social sustainability in supply chains addresses issues of should be bought or rented in different time horizons. It is also
social justice and human rights with studies focusing on practices possible to decide on the size of investment, which is a specific
such as supplier human rights actions, labor conditions, codes of percentage of the gained revenue, to be made at the initial period of
practices and social auditing, supplier compliance with child labor time.
laws, and the delivery of social equity through sourcing from Qin and Ji, 2010 develop a single period single product reverse
diverse suppliers in terms of gender, size, ethnicity and avoidance logistics to deal with nondeterministic return quantity. They use
of conflicts of interest. Including social aspects in network design fuzzy programming approach to formulate three different mathe-
decisions allows to better evaluate the impact of a supply chain on matical model under different criteria. They utilize hybrid intelli-
its stakeholders: employees, customers and local communities. So gent algorithm to solve the proposed models and justified by
in order to measure the social impact we should evaluate the effects different numerical examples.
of that on employees and customers. Three common indicators Roghanian and Pazhoheshfar, 2015 propose a probabilistic
including work conditions, societal commitment and customer is- mixed integer linear programming model with stochastic demand
sues have been used by researchers for quantifying social effects. In to design and plan a network of reverse logistics. They utilize
the field of work conditions, employment is the main social indi- priority-based genetic algorithm to solve and validate the proba-
cator used in literature (Eskandarpour et al. (2015)). In order to bilistic reverse logistics network design problem by applying to a
incorporate the social effects in our model we use fixed and vari- numerical example.
able created job opportunities and the number of jobs which are Ayvaz et al. (2015) try to develop a capacitated system for
lost due to work damages. The rest of this paper is arranged as recycling waste of electrical and electronic equipment with sto-
follows: In Section 2, relevant previous researches are reviewed. In chastic return quantity, sorting ratio and transportation costs in
section 3, the model characteristics and formulation is demon- two-stage stochastic structure. The objective function of the model
strated. A comprehensive computational analysis is undertaken in is profit maximization and the proposed model is justified by a real
Section 4. Finally, Section 5 is assigned to conclusion and future case study performed in Turkey. In order to solve the problem they
research. A brief explanation about the incorporation of CVaR in the utilize sample average approximation method.
framework of two-stage stochastic programming is represented in Simic, 2015 developed a model to plan a network of reverse
the appendix. logistics for recycling and managing the end of life vehicle. The
model is formulated in two-stage stochastic programming concept
2. Literature review with nondeterministic numbers of ELVs. The proposed model is
tested in a hypothetical case study.
In this paper we attempt to propose a risk-averse stochastic
mathematical model, regarding green and sustainability issues in 2.2. Risk aversion in RL and CLSC models
order to design a network of reverse logistics for recycling C&D
wastes. Consequently our focus in literature survey in relations The most fundamental paper on the subject of risk-averse two-
with network design problems, will falls into three categories: 1) stage stochastic programming for reverse logistics network design
considering uncertainty in reverse logistics (RL) and closed-loop problem is done by Soleimani and Govindan (2014). They develop a
supply chain (CLSC) mathematical models; 2) risk-aversion in RL network of reverse logistics by incorporating CVaR as a coherent
and CLSC mathematical models; and 3) green and sustainable and convex risk measure in their model. Then they try to learn the
supply chain network design (SCND). effect of risk aversion factors in their model by designing 24 nu-
merical examples by using CPLEX as a solver.
2.1. Considering uncertainty in RL and CLSC models Soleimani et al. (2013) propose a risk-averse mathematical
model to design and plan a forward and revers supply chain. They
Demirel et al. (2014) utilize mixed integer linear programming try to consider VaR, CVaR and Mean Absolute Deviation as three
to design a network of reverse logistics for managing and recycling common risk measures in their model in order to compare them.
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1569

2.3. Green and sustainable network design transportation accident risks).


Considering the literature review of RL mathematical model and
Bing et al. (2014) propose a single-objective mixed integer linear Table 1 leads us to identify some gaps in the field of sustainable
programming model to plan a network of reverse logistics for reverse logistics network design;
recycling household plastic waste. The environmental aspect is
quantified in terms of CO2 emission costs. The objective function of  All of the cited papers ignored the design of a sustainable
the proposed model is trying to minimize the costs of network and reverse logistics network by considering risk-aversion in their
environmental costs resulted from processing and transportation of mathematical model. Indeed our research claims that consid-
products simultaneously. ering all aspects of sustainability and risk-aversion simulta-
Amin and Zhung, 2013 develop a model by considering neously in two-stage stochastic programming framework in the
economical and environmental aspects in CLSC. The multi- field of reverse logistics network design is novel. Generally
objective model also is extended by stochastic programming (sce- designing a sustainable network of supply chain with risk-
nario-based) to examine the effects of uncertain demand and re- aversion in the field of recycling C&D wastes is the main
turn on the network configuration. The environmental impact is contribution of our work. To the best of our knowledge, there are
quantify by considering some factors like environmental friendly no papers which consider both sustainability and risk aversion
materials used in plants for producing products and clean tech- in their mathematical model in the field of recycling C&D
nology. The proposed model is solved by the means of epsilon- wastes.
constraint method and weighted sum average.  We consider two options in our mathematical model for col-
Yu and Solvang, 2017 tried to develop a risk-averse mathemat- lecting and separating C&D wastes; 1. On-site separation which
ical model in which they have considered environmental impacts in is used when the act of separating C&D wastes is done in the
terms of co2 emissions. They also incorporated uncertainty in their place of generating them. 2. Off-site separation which is done in
model by stochastic programming. However in their model they the recycling centers. So in this paper we try to develop a sus-
have only considered two aspects of sustainability (i.e economic tainable risk-averse two-stage stochastic programming model
and environmental aspects) and the social responsibility has been to design and plan a network of reverse logistics for recycling
ignored. C&D wastes with stochastic demand and rate on investment.
Govindan and Jafarian, 2015 developed a model in which they
integrated supply chain network design with order allocation 3. Model description and formulation
problem (OAP). In their mathematical model, they only considered
economic and environmental aspects. To solve the propped model In this study we propose a multi-period multi-objective two-
they used a noble hybrid multi-objective approach, called MOHEV, stage stochastic programming model with stochastic parameters
to solve the addressed problem. of demand and rate on investment and objective functions of profit
Kannegiesser et al., 2015 tried to design a sustainable network of maximization, environmental impact minimization and social
supply chain in which they proposed a mathematical which aims at impact maximization. The proposed model is under the following
minimizing the so called time to sustainability that some pre- assumptions and features:
defined sustainable targets of the model will be achieved. In their
paper, they considered all the three aspects of sustainability.  The amount of fine aggregates (FA) and coarse aggregates (CA)
Recently, Govindan et al. (2016) proposed a fuzzy multi-echelon can be aggregated.
multi-period multi-objective model to address the network design  Production capacity of recycling centers for producing FA and CA
problem of collection and recycling of medical syringes. In their can be aggregated.
mathematical model they consider all aspect of sustainability  Shortage cost is zero.
consists of minimizing the costs of network as well as environ-  A portion of generated C&D wastes is separated on-site (in place
mental impacts and optimizing social responsibility as objective where they are generated).
functions of the model. They utilized MOPSO algorithm to obtain  Two basic products resulted from recycling of C&D wastes is
Pareto optimal solutions. They also compare the proposed algo- considered in this model.
rithm with epsilon-constraint method based on some comparison  The model is aimed at finding optimal number of recycling
metrics. center among potential locations.
Soleimani et al. (2017) developed a mathematical model in the  A portion of obtained revenue is assigned for capacity expansion
field closed-loop supply chain which consider all aspects of sus- costs of recycling centers.
tainability, i. e economic, social and environmental factors. In their  Capacity expansion of recycling centers for t ¼ 1 is zero.
model, they considered components of End-of-life products and  On-site separation is done in generation points and off-site
raw materials which is the main contribution of the paper. In order separation is carried out in recycling centers.
to deal with the mathematical model, the authors used fuzzy  The collection centers and manufacturers are the third party
mathematical programming. They also have developed a genetic providers which for the first one we pay for disposing C&D
algorithm to solve and validate the proposed model. wastes and for the second one we sell FA and CA to.
Arampantzi and Minnis, 2017 proposed a mathematical model  Different environmental impacts resulted from opening recy-
which encompasses all aspects of sustainability to design a network cling centers, manufacturing at recycling centers and trans-
of supply chain. In their environmental objective function they have porting C&D wastes as well as FA and CA are modeled.
considered GHG emissions and waste generation due to construc-  To quantify environmental impacts, the proposed model ac-
tion of facilities both in terms of co2 emissions. To incorporate social counts for GHG (Green House Gases) emissions related to en-
responsibility, they considered various kinds of social aspects such ergy consumption for constructing and operating recycling
as labor conditions (e.g stable employment, dismissals, rest pe- centers in terms of CO2 emissions.
riods) and societal community development (e.g. support of local  Social responsibility of the organizations has multiple facets and
suppliers, establishing new facilities in less developed areas, includes several indicators regarding social and environmental
1570 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

Table 1
Review of previously published literature.

Author approach Risk aversion Model type Network type objective functions

two-stage fuzzy probabilistic non-deterministic deterministic forward reverse Cost social environmental
stochastic (profit) impacts impacts

Roghanian and Pazhoheshfar, 2015 * * * *


Ayvaz et al. (2015) * * * *
Qin and Ji, 2010 * * * *
Simic, 2015 * * * *
Sinha et al. (2010) * * * *
Demirel et al. (2014) * * *
Zhou and Zhou, 2015 * * *
Alumur et al. (2012) * * *
Alshamsi and Diabat, 2015 * * *
Yanik (2014) * * *
Soleimani and Govindan, 2014 * * * * *
Soleimani et al. (2013) * * * * * *
Zeballos et al. (2016) * * * * * *
Moghaddam and Fuzzy, 2015 * * * * *
Dinler and Gungor, 2017 * * * *
Amin and Zhung, 2013 * * * * * *
Ramezani et al. (2013) * * * * * *
Bing et al. (2014) * * * *
Devika et al. (2014) * * * * * *
Govindan et al. (2016) * * * * * *
Soleimani et al. (2017) * * * * * * *
Arampantzi and Minnis, 2017 * * * * *
Yu and Solvang, 2017 * * * * * *
Kannegiesser et al., 2015 * * * * *
Govindan and Jafarian, 2015 * * * * *

issues. ISO proposed ISO2600 in order to achieve a standard  Generation points: In this entity C&D wastes are generated and
framework to for social responsibility. To quantify social impacts separated. Then separated C&D wastes are sent to recycling
we follow ISO2600 and the work of Devika et al. (2014) who centers. In this entity we aim at recycling a desirable percentage
consider two criteria: of C&D wastes at the point of its origins.
 Created job opportunity which is divided into two categories  Regional collection centers and landfills: In these centers C&D
of fixed job opportunities and variable job opportunities wastes are gathered and sent to recycling centers.
which depend on the capacity expansions of the recycling  Recycling centers: In these centers C&D wastes gathered from
centers. generation points and collection centers are recycled. Recycled
 Works' damages which fall into two categories of works' products (FA and CA) are bought to project sites and generation
damages during the establishment of recycling centers and points as well as manufacturers as raw materials. The portion of
those which occur during the producing of CA and FA. C&D wastes which is not suitable for recycling is sent to landfills
to dispose.
Fig. 1 shows the proposed network for recycling C&D wastes.  Project sites: Construction projects which buys FA and CA to
The role of each entity is explained as follows: complete their construction activities.

puddle Manufacturer

C&D

FA,CA
Separate C&D

Generation point

C&D
Regional collection
C&D Recycling center FA,CA
center and landfill

Project site

FA,CA

Fig. 1. The proposed network for recycling C&D wastes.


M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1571

ðf Þ
 Manufacturers: They buy FA and CA from recycling centers as Micrn :Minimum allowable capacity expansion of recycling
raw materials to produce their products. center r with technology n for fine aggregates
ðcÞ
 Puddles: The part of C&D wastes, which is not recyclable is sent Micrn :Minimum allowable capacity expansion of recycling
to puddles for leveling low-lying areas. center r with technology n for coarse aggregates
ERrn :Environmental impact (expressed in co2 emissions) of
The complete proposed two-stage stochastic programming establishing recycling center r with technology n
model is presented as follows: EMrn :Environmental impact (expressed in co2 emissions) of
Sets and indices: recycling per unit of fine and coarse aggregates in recycling
center r with technology n
 g: set of generation points ETrm : Environmental impact (expressed in co2 emissions) of
 d: set of puddles transporting one unit of coarse and fine aggregates from recy-
 l: set of regional collection centers and landfills cling center r to manufacturer m
 m: set of manufacturers ETrp : Environmental impact (expressed in co2 emissions) of
 p: set of project sites transporting one unit of coarse and fine aggregates from recy-
 r: potential number of recycling centers cling center r to project sites p
 n: set of technology ETrg : Environmental impact (expressed in co2 emissions) of
 s: set of scenarios transporting one unit of coarse and fine aggregates from recy-
 T: set of time periods cling center r to generation point g
 i: index, which can refer to any nodes ETrd : Environmental impact (expressed in co2 emissions) of
 j: index, which can refer to any destination nodes transporting one unit of coarse and fine aggregates from recy-
 f: superscript for fine aggregates cling center r to puddle d
 c: superscript for coarse aggregates ETrl : Environmental impact (expressed in co2 emissions) of
transporting one unit of coarse and fine aggregates from recy-
Parameters: cling center r to landfill l
ETgr : Environmental impact (expressed in co2 emissions) of
fixrn : Fixed cost of opening recycling plant r2R with technology transporting one unit of separate C&D wastes from generation
n point g to recycling center r
Tct : Transportation cost per unit per kilometer during time ETlr : Environmental impact (expressed in co2 emissions) of
period t transporting one unit of C&D wastes from generation point l to
Scrt : Segregation cost during time period t recycling center r
Sgt : Onesite segregation cost per unit during time t fjrn :Number of fixed job opportunities created by establishing
Dct : Disposal cost of C&D wastes per unit during time period t recycling center r with technology n
Rcrtn : Recycling cost of C&D wastes per unit at recycling center r flrn : Lost days due to works damage during the establishment of
with technology n during time t recycling center r with technology n
ðf Þ
Hcrt : Inventory holding cost of fine aggregates per unit at vjrt : Number of variable job opportunities created through
recycling center rduring time period t working in recycling center r during time period t
ðcÞ
Hcrt : Inventory holding cost of coarse aggregates at recycling vlrt : Lost days due to work damages during working in recycling
center rduring time period t center r in time period t
ðfÞ ðf Þ
Kcrt :Processing capacity expansion cost of fine aggregates per demst : Demand of fine aggregates in scenario s during time t
ðcÞ
unit in recycling center r in time t demst : Demand of coarse aggregates in scenario s during time t
ðcÞ
Kcrt :Processing capacity expansion cost of coarse aggregates roIst : Rate on investment under scenario s during time t
per unit in recycling center r in time t ut : Percentage of the profit invested during time period t
ðfÞ
fixcrt : Fixed cost of processing capacity expansion for fine ag- ps : Probability of occurring scenario s
gregates in recycling center r in time period t l : The weight of risk
ðcÞ
fixcrt : Fixed cost of processing capacity expansion for coarse a : Confidence level
aggregates in recycling center r in time period t b : Percentage of C&D wastes targeted to separate at source.
Disij : Distance between location i and location j (i,j) Maxr: Maximum number of allowable activated recycling
2fðr; gÞ; ðr; pÞ; ðr; mÞ; ðr; lÞ; ðr; dÞg centers
ðf Þ
Caprn : Initial processing capacity of recycling center r with
technology n for producing fine aggregates Decision variables:
ðcÞ
Caprn : Initial processing capacity of recycling center r with
technology n for producing coarse aggregates Xrn : Binary variable equals "100 if recycling center r with tech-
Cvrt : Inventory holding capacity of recycling center r2R during nology n is open and "000 otherwise
ðfÞ
time t Yrstn : Binary variable equals "100 if capacity of recycling center r
ðf Þ
Revt : Unitary revenue gained by selling one unit of fine ag- with technology n in scenario s2 S is expanded during time t for
gregates during time period t producing fine aggregates and "000 otherwise.
ðcÞ ðcÞ
Revt : Unitary revenue gained by selling one unit of coarse Yrstn : Binary variable equals "100 if capacity of recycling center r
aggregates during time period t with technology n in scenario s is expanded during time t for
ðf Þ
Macrn :Maximum allowable capacity expansion of recycling producing coarse aggregates and "000 otherwise.
center r with technology n for fine aggregates Q ijst : Flow of C&D wastes sent from location i to location j in
ðcÞ
Macrn :Maximum allowable capacity expansion of recycling scenario s in time period t (i,j) 2fðl; rÞ; ðg; rÞ; ðr; lÞ; ðr; dÞg
ðfÞ
center r with technology n for coarse aggregates Vijst : Flow of fine aggregates sent from location i to location j in
scenario s during time period t (i,j) 2fðr; gÞ; ðr; mÞ; ðr; pÞg
1572 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

ðfÞ
Vijst : Flow of coarse aggregates sent from location i to location j 2
in scenario s during time period t (i,j) 2fðr; gÞ; ðr; mÞ; ðr; pÞg X X X X  XX X
fhrstn : Amount of fine aggregates produced in recycling plant r TRC ¼ 4 Q grst $Disgr $Tct ps þ
under scenario s in time t with technology n g2G r2R s2S t2T l2L r2R s2S
X XXX
chrstn : Amount of coarse aggregates produced in recycling plant  ðQ lrst $Dislr $Tct Þps þ
r under scenario s in time t with technology n t2T r2R l2L s2S
ðf Þ
hrstn : Amount of fine aggregates hold at the inventory of recy- X X X X
 ðQ rlst $Dislr $Tct Þps þ
cling center r produced with technology n during time period t
t2T r2R d2D s2S
in scenario s X X X X
ðcÞ
hrst : Amount of coarse aggregates hold at the inventory of  ðQ rdst $Disrd $Tct Þps þ
t2T i2R j2fG;P;Mg s2S
recycling center r produced with technology n during time
X ðfÞ
 X X X
period t in scenario s
ðfÞ
 Vijst $Disij $Tct ps þ
Cerstn : Amount of capacity added to recycling center r with t2T i2R j2fG;P;Mg s2S
technology n in terms of fine aggregates under scenario s in time 3
t X ðcÞ

ðcÞ  Vijst $Disij $Tct ps 5
Cerstn : Amount of capacity added to recycling center r with
t2T
technology n in terms of coarse aggregates under scenario s in
time t (3.5)
ðf Þ
Csrstn : Processing capacity of recycling center r with technology Expected segregation costs:
n in scenario s for fine aggregates during time period t
ðcÞ XX XX X X X
Csrstn : Processing capacity of recycling center r with technology SGC ¼ ðQ lrst ÞScrt $ps þ
n in scenario s for coarse aggregates during time period t l2L r2R s2S t2T g2G r2R s2S
sst : Investment made at time t in scenario s XQ grst 
dvhs : Auxiliary variable for calculating CVaR under scenario s  Sgt $ps (3.6)
based on Eq. (a.6) (which is broadly explained in the appendix). t2T
b
h : Value at risk (VaR)
The first term of Eq. (3.6) calculates the segregation cost in
recycling center for C&D waste which is sent from collection cen-
ters and landfills and the second term calculates the segregation
3.1. Objective functions cost at the point of C&D wastes origins (on-site separation).
Expected Disposal costs:
The first objective function tries to maximize total expected XXXX
profit and it is determined as follows: DSC ¼ ðQ rlst $Dct Þps (3.7)
Maximize total expected profit ¼ Total expected sales e Total r2R l2L s2S t2T
costs. A portion of C&D wastes are not recycled due to some reasons
Total expected sales is calculated as follows: and they need to be disposed. For this, we have to pay in order to
X X XX dispose of them. This term calculate this kind of cost.
ps RevðfÞ ðfÞ
t Vijst Expected recycling costs:
i2Rj2fG;P;Mgs2St2T
X X XX X X X X  XX X
þ ps RevðcÞ ðcÞ RLC ¼ Q grst $Rcrt ps þ
t Vijst (3.1)
i2Rj2fG;P;Mgs2St2T g2G r2R s2S t2T l2L r2R s2S
X
 ðQ lrst $Rcrt Þps (3.8)
Eq. (3.1) has got two parts; the first part which calculates the
t2T
expected sales gained from fine aggregates and the second part
refers to expected sales obtained from selling coarse aggregates to Eq. (3.8) calculates the recycling costs of C&D wastes.
manufacturers, project sites and generation points. Expected inventory holding costs:
Total costs are calculated in the following form as we discuss in h   i
X XX ðfÞ ðf Þ ðcÞ ðcÞ
appendix: IHC ¼ ps hrst $Hcrt þ hrst $Hcrt (3.9)
r2R s2S t2T
Total costs: (1þ l) Fixed costs þ Total expected costs þ (l) Total
Eq. (3.9) calculates the costs of inventory held in recycling
CVaR costs (3.2)
centers.
Fixed costs (or First stage costs in terms of stochastic pro- Expected capacity expansion costs:
gramming) is calculated as follows: X XX X h  
ðfÞ ðfÞ ðfÞ ðfÞ
XX CEC ¼ ps fixcrt $Yrstn Þ þ ðKcrtn $Cerstn þ
ð1 þ lÞfixrn $Xrn (3.3) r2R s2S t2T n2N
 i
r2Rn2N ðcÞ ðcÞ ðcÞ ðcÞ
 fixcrt $Yrstn Þ þ ðKcrt $Cerstn
Total expect costs (or Second stage costs in terms of stochastic
(3.10)
programming) include the following costs:
Total expected costs: Expected transportation costs þ Expected The first and the second term of Eq. (3.10) calculates the fixed
segregation costs þ Expected disposal costs þ Expected recycling and variable costs of capacity expansion for producing fine aggre-
costs þ Expected inventory holding costs þ Expected capacity gates in recycling centers, respectively. The third and the forth
expansion costs (3.4). terms are the fixed and variable capacity expansion costs of recy-
Where, each of them is calculated as follows: cling centers for producing coarse aggregates.
Expected transportation costs: And the last term of Eq. (3.2) is calculated as follows based on Eq.
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1573

(a.6) which is brought in the appendix: 3.2. Constraints

! The proposed model includes the following constraints:


1 X Based on Eq. (a.6), we define the following constraint to calcu-
l hþ ps $dvhs (3.11)
1  a s2S late CVaR costs:
2
Where, l is risk coefficient which shows the tendency of the X X X X  XX X
decision makers to take risk h is value at risk which can be inter- dvhs  4 Q grst $Disgr $Tct þ
preted as a first-stage variabledvhs is an auxiliary variable which g2G r2R s2S t2T l2L r2R s2S
X XXXX
can be interpreted as a second-stage variable in terms of two-stage  ðQ lrst $Dislr $Tct Þ þ ðQ rlst $Dislr $Tct Þ
stochastic programming. t2T r2R l2L s2S t2T
The second objective, environmental impacts of the network, is
X X XX X X X
þ ðQ rdst $Disrd $Tct Þ þ
as follows: r2R d2D s2S t2T i2R j2fG;P;Mg s2S
X ðf Þ
 X X X
X X X XX X  Vijst $Disij $Tct þ
minimize : ERrn $Xrn þ ps $EMrn ðfhrstn t2T i2R j2fG;P;Mg s2S
r2R n2N r2R s2S t2T n2N
3
X X XX  X ðcÞ
 XX XX
þ chrstn Þ þ ps $ETrm VðfÞ  Vijst $Disij $Tct 5 þ ðQ lrst ÞScrt $
rmst
r2R m2M s2S t2T t2T l2L r2R s2S t2T
 X X XX  
ðcÞ X X X XQ grst  XXX
þ Vrmst þ ps $ETrp Vðfrpst
Þ ðcÞ
þ Vrpst þ Sgt þ
r2R p2P s2S t2T g2G r2R s2S t2T
b r2R l2L s2S
X X XX   X X X X X X  X
þ ps $ETrm VðfÞ
rmst þ
ðcÞ
Vrmst þ  ðQ rlst $Dct Þ þ Q grst $Rcrt þ
r2R m2M s2S t2T r2R t2T g2G r2R s2S t2T l2L
X XX   X X X XX X X Xh 
 ps $ETrg VðfÞ ðcÞ
rmst þ Vrgst þ  ðQ lrst $Rcrt Þ þ
ðfÞ ðf Þ
hrst $Hcrt
g2G s2S t2T r2R d2D r2R s2S t2T r2R s2S t2T
XX XXX  i X X X X h
 ps $ETrd $Q rdst þ þ
ðcÞ ðcÞ
hrst $Hcrt þ
ðf Þ ðfÞ
fixcrt $Yrstn Þ
s2S t2T r2R l2L s2S
X XX X 
r2R s2S t2T n2N
 i
 ps $ETrl $Q rlst þ þ
ðf Þ ðf Þ
ðKcrtn $Cerstn þ
ðcÞ ðcÞ ðcÞ ðcÞ
fixcrt $Yrstn Þ þ ðKcrt $Cerstn
t2T gεG r2R s2S
X XX X
 ps $ETgr $Q grst þ h cs
t2T lεL r2R s2S (3.14)
X
 ps $ETlr $Q lrst  
X X ðfÞ ðcÞ ðfÞ ðcÞ
t2T
Vijst þ Vijst  demst þ demst cs; t (3.15)
(3.12) i2R j2fP;M;Gg

The environmental impacts caused by opening recycling centers Constraint (3.15) guarantees that the flows of FA and CA sent
are presented in the first term. The second term represent the from recycling centers to project sites, manufacturers and genera-
environmental Impacts due to recycling C&D wastes in recycling tion points will satisfy the demand of these entities.
centers. The third to sixth terms and the seventh to tenth terms
X X X X X ðf Þ
stand for environmental impacts of transporting products (FA and Q grst þ Q lrst ¼ Q rlst þ Q rdst þ Vrpst
CA) and C&D wastes respectively. g2G l2L l2L d2D p2P
The social objective of the model is written as follows: X ðcÞ
X ðf Þ
X ðcÞ
þ Vrpst þ Vrmst þ Vrmst
p2P m2M m2M
" X X
X X X XX X  ðfÞ ðcÞ
Maximize : fjrn $Xrn þ ps $vjrt Ceðfrstn
Þ þ Vrgst þ Vrgst cr; s; t
g2G g2G
r2R n2N r2R s2S t2T n2N
# "
 X X X XX (3.16)
ðcÞ
þ Cerstn  flrn $Xrn þ
Constraint (3.16) is a balanced constraint which assures that all
r2R n2N r2R s2S t2T
# the flows entering to a recycling center in scenario s and during
X
 ps $vlrt ðfhrstnþ chrstn Þ time period t equal to all flows issuing from that recycling center in
n2N that scenario and time period.
(3.13) ðfÞ ðfÞ
Csrstn ¼ Caprn $Xrn t ¼ 1; cs; r; n (3.17)
The first term of the social objective function is the fixed job
opportunities which are created when a recycling center is estab- ðcÞ ðcÞ
lished and are independent from utilized capacity (job positions Csrstn ¼ Caprn $Xrn t ¼ 1; cs; r; n (3.18)
like manager). The second term refers to variable job opportunities
which depend on capacity expansion. Depending on how much ðfÞ ðfÞ ðfÞ
Csrstn ¼ Csrsðt1Þn þ Cerstn t > 1; cs; r (3.19)
capacity expansion a recycling center will have, they are going to
hire more workers. The third and the forth terms stand for work's
ðcÞ ðcÞ ðcÞ
damages either during establishment of recycling centers or during Csrstn ¼ Csrsðt1Þn þ Cerstn t > 1; cs; r (3.20)
the recycling and handling of FA and CA.
1574 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

Constraints (3.17) and (3.18) indicate that the production ca- X ðfÞ ðcÞ
 X  ðf Þ ðcÞ
 X
ðfÞ ðcÞ

pacity of recycling centers are equal to initial capacity in time Vrgst þ Vrgst þ Vrmst þ Vrmst þ Vrpst þ Vrpst
period t ¼ 1. In other words we do not have capacity expansion for g2G m2M p2P
 
neither FA nor CA in time period t ¼ 1. On the other hand, the ca- ðfÞ ðcÞ
 Caprn þ Caprn $Xrn cr; s; n; t ¼ 1
pacity of FA and CA in period t (t > 1) is equals to the capacity of FA
and CA in period (t-1) added to the capacity expansion we have in (3.30)
time period t.
X ðfÞ ðcÞ
 X  ðf Þ ðcÞ
 X
ðfÞ ðcÞ

Vrgst þ Vrgst þ Vrmst þ Vrmst þ Vrpst þ Vrpst
ðfÞ g2G m2M p2P
Yrstn ¼ 0 t ¼ 1; cs; r; n (3.21)  
ðfÞ ðcÞ
 Csrstn þ Csrstn cr; s; n; t > 1

ðcÞ (3.31)
Yrstn ¼ 0 t ¼ 1; cs; r; n (3.22)
Constraints (3.30) and (3.31) guarantee that all flows issuing
Constraints (3.21) and (3.22) guarantee that capacity expansions from recycling centers will not exceed the capacity of these centers.
for both FA and CA doesn't occur in time period t ¼ 1.
X X
Xrn  Maxr (3.32)
ðfÞ r2R n2N
Yrstn  Xrn t>1 cr; s; n (3.23)
Constraint (3.32) is a managerial limitation on the maximum
number of allowable recycling centers.
ðcÞ
Yrstn  Xrn t > 1 cr; s ; n (3.24) X X X
ðfhrstn þ chrstn Þ ¼ Q grst þ Q lrst cr; s; t; n (3.33)
Constraints (3.23) and (3.24) assure that the capacity expansion n2N g2G l2L
occurs in recycling centers which are already established. Constraint (3.33) indicates that the amount of FA and CA equals
to the flows of C&D wastes which are entered to recycling centers.
ðfÞ ðcÞ
hrstn þ hrstn  Cvrt $Xrn cr; s; t; n (3.25)
4. Computational analysis
Constraints (3.25) guarantees that the whole amount of FA and
CA held in inventories of recycling centers will not exceed the In order to validate the proposed model, a mid-size problem has
storage capacity of recycling centers. been designed. The model is coded and run by Gams 24.1.2 software
optimizer which can find the optimum of the proposed two-stage
ðfÞ ðfÞ ðfÞ ðfÞ stochastic mixed-integer linear programming. The proposed
MicðfÞ
rn $Yrstn  Cerstn  Macrn $Yrstn cr; s; n; t > 1 (3.26) multi-objective model is solved by using epsilon-constraint
method, which is a classical method, in order to obtain optimal
Pareto solutions. The structure of epsilon-constraint method is
ðcÞ ðcÞ ðcÞ ðcÞ
MicðcÞ
rn $Yrstn  Cerstn  Macrn $Yrstn cr; s; n; t > 1 (3.27) explained in the appendix. We construct a network with three units
in each entities consisting of three recycling centers, three collec-
Constraints (3.26) and (3.27) indicate that the capacity expan-
tion centers and landfills, three generation points, three project
sion for producing FA and CA will not be more or less than the
sites, three manufacturers and three puddles as well as three types
allowable amount of capacity expansion which already has been
of technology in order to validate the proposed mathematical
set.
model. To deal with the uncertainty in the model we design 9
scenarios by considering three cases for each stochastic parameters.
XX X Tables 2 and 3 represents the generated parameters and designed
ðf Þ
sst1 ¼ ut  ð1 þ roIst1 Þ  Revt1  fhrst1n þ scenarios, respectively.
r n r2R In order to quantify the environmental impacts, we consider
!
X ðcÞ
GHG emissions related to energy consumption due to construction
 ð1 þ roIst1 Þ  Revt1  chrst1n s; t > 1 of recycling centers and producing FA and CA in recycling centers
n2N and shipping them among supply chain entities which all of them
(3.28) are expressed in terms of CO2 emissions. For constructing a recy-
Constraint (3.28) indicates the amount of money we are going to cling center the proposed model accounts for GHG emissions due to
invest for the purpose of capacity expansion in recycling centers in consuming energy during construction and production of con-
time period “t” which is a percentage of the revenue gained from struction materials. To produce each unit of FA and CA, the model
the previous period. calculates the CO2 emissions based on the amount of energy which
is used during the processing of FA and CA (e.g., lighting, heating,
h     cooling, moving etc.). The emission used in the model is obtained
X X ðfÞ ðfÞ ðfÞ ðf Þ ðcÞ ðcÞ
ps fixcrt $Yrstn þ ðKcrtn $Cerstn þ fixcrt $Yrstn from recognized data sources including manufacturing web sites
r2R n2N (e.g. JEMA, 2010), reports (e.g. The Consumer Goods Forum, 2012),
i research articles (e.g. Jeong et al., 2012), and commercial emission
ðcÞ ðcÞ
þ ðKcrtn $Cerstn
calculators (see EPA, 2015). For example, Jeong et al. (2012) evaluate
 ss;t1 cs; t > 1 the environmental impact (energy and CO2) of materials consumed
in construction. Their results indicate that CO2 emissions of the
(3.29)
various construction materials is estimated to be about 500 kg-
Constraints (3.29) assures that the expected capacity expansion CO2/m2 of built space on average. Using this benchmark we esti-
costs will not exceed the investment mad for this purpose. mate the amount of CO2 emissions generated to establish one
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1575

Table 2
Parameters of computational study.

Row Parameters Uniform distribution

1 fixrn U (45000,90000)
2 Tct U (1,3)
3 Scrt U (1.5,2)
4 Sgt U (0.5,1.2)
5 Dct U (0.5,1.5)
6 Rcrt U (3,4.2)
7 ðf Þ U (40,55)
Revt
8 ðcÞ U (80,100)
Revt
9 ðfÞ U (65000,120000)
Capr
10 ðcÞ U (75000,300000)
Capr
11 ERn U (417777.8,471111.2)
12 EM rn U (10,30)
13 ETrp ; ETrg ETrm U (0.2,2)
14 ETrl ; ETrd U (0.3,4)
15 ETlr ,ETgr U (0.4,5)
16 fjrn U (1,100)
17 flrn U (10,100)
18 U (0.1,0.7) Fig. 2. Obtained Pareto optimal solutions in 3-D figure.
vjrtn
19 vlrtn U (0.1,0.9)

*Dollars.
**Million grams.
1.18E+08
2
9.76E+07
capacity unit of a recycling center (expressed in m ). According to 7.76E+07
the standard unit fuel consumption, each gram of CO2 per kilo- Profit 5.76E+07
meter (Gasoline) is converted to liter per 100 km as 1 g/kilometer 3.76E+07
CO2 equals 0.043 L/100 km, i.e. each track produces 23.2 g CO2 in 1.76E+07
each kilometer which is driven and all the parameters related to the -2.40E+06

-1.35E+06
-1.26E+06
-9.11E+05
-8.71E+05
-8.02E+05
-7.05E+05
-6.05E+05
-4.69E+05
-2.97E+05
-1.24E+05
transportation emissions, which are the coefficient of distances
between entities, are estimated based on this number.
The optimal Pareto solutions set is obtained by solving the
proposed example, which is illustrated in Figs. 2e5. The obtained
figures show the conflicting objective functions in the mathemat- Social Impact
ical model.

Fig. 3. Optimal Pareto solutions (Profit vs. Social Impact).


4.1. Sensitivity analysis

In this section we conduct some examples in order to provide CVaRa accounts for the risk of larger realizations of the total costs.
managerial insights about the important parameters of the math- Thus, larger a values add to the conservative behavior of the model.
ematical model. We conduct our experimental calculations on risk- In order to learn how alpha (a) affects profit maximization
aversion parameters (a and l) as well as segregation rate parameter objective function and its components, we consider seven different
to realize how they affect the objective functions of the proposed values for alpha (a) with fixed values for lambda (l ¼ 0.1) and beta
model. (b ¼ 0.1).
In our proposed model we have two risk parameters (a and l) As it is illustrated in Figs. 6 and 7, by increasing the value of
which we should realize how they affect the optimal policies. alpha (a), the model behaves more conservative. Therefore as we
Generally in terms of financial field we can define CVaRa as the discussed before, the value of CVaR increases by the augmentation
mean value of the worst (1 a) of the total loss. The specified a level of alpha values. Fig. 6 also implies that by increasing the value of
represents the risk preference in percentage terms, i.e., CVaRa alpha, the values of designing stage costs, which stands for the costs
quantifies the mean value of the worst (1a) % of the total costs. of first-stage decision variables, and second stage costs, which
When a increases the corresponding value-at-risk increases, and stand for the costs caused by the second-stage decision variables,

Table 3
Complete explanation of nine scenarios.

Scenario Demand of FA Demand of CA Probability ROI Probability Scenario Probability

1 low (10e15) low (100e150) 25% 5e15 20% 0.05


2 mid (15e20) mid (150e200) 65% 5e15 20% 0.13
3 high (20e25) high (200e250) 10% 5e15 20% 0.02
4 low (10e15) low (100e150) 25% 15e30 50% 0.125
5 mid (15e20) mid (150e200) 65% 15e30 50% 0.325
6 high (20e25) high (200e250) 10% 15e30 50% 0.05
7 low (10e15) low (100e150) 25% <30 30% 0.075
8 mid (15e20) mid (150e200) 65% <30 30% 0.195
9 high (20e25) high (200e250) 10% <30 30% 0.03
1576 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

8.0E+07 Profit Revenue

Environmetal Impact
6.0E+07
4.0E+07 2.6E+08
2.0E+07 2.1E+08
1.6E+08
0.0E+00
1.1E+08
6.04E+06
1.75E+07
2.88E+07
3.93E+07
4.88E+07
5.80E+07
6.54E+07
7.09E+07
8.78E+07
9.98E+07
5.5E+07
5.0E+06
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Profit Profit 9.977 9.960 9.950 9.940 9.928 9.912 9.89E
Revenue 2.367 2.367 2.367 9.940 9.928 9.912 9.89E
Fig. 4. Optimal Pareto solutions (Profit vs. Env Impact). alpha

Optimal Pareto solutions ( Social vs. Env Impact) Fig. 7. Alpha vs. Revenue and Profit.

are shown in Fig. 8. We also calculate the effect of alpha (a) on


expected amount of FA and CA produced in recycling centers in
order to evaluate more. As the value of alpha (a) increases, we
expect that the model behaves more conservative. Therefore, due to
this behavior of the model, the amounts of FA and CA increase
which lead to greater values for environmental impacts. This issue
has been depicted in Fig. 9.
We analyze the effect of alpha on social impact by considering
six different values for this parameter and with fixed values for
(l ¼ 0.1) and beta (b ¼ 0.1). By increasing the value of alpha, and as a
result by rising the risk-aversion behavior of the model as well as
the expected amount of FA and CA, we expect downward trend for
social impact values. As illustrated in Fig. 10, the value of social
Fig. 5. Optimal Pareto solutions (Social vs. Env Impact).
impact decreases by the augmentation of parameter alpha (a).
Another important risk aversion parameter that we should
do not change. analyze its effects on the model is lambda (l).
Since by increasing the value of alpha we tend to rise the level of In order to analyze lambda effect on profit maximization
risk-aversion behavior in the model, we expect lower level for the objective function and its components, we consider six values with
values of profit and revenue. As it is illustrated in Fig. 7, due to more fixed values for (a ¼ 0.1) and (b ¼ 0.1). Figs. 11e13 and Table 5 show
conservative behavior of the model, by increasing the value of the results of this experiment.
alpha, the values of profit and revenue decrease. Lambda (l) is a risk-averse coefficient which implies that
Table 4 illustrates the complete results of alpha effects on the increasing the parameter l will lead to a higher level of risk-
profit maximization objective function and its components. aversion which means that by increasing the value of parameter
To analyze the effects of alpha (a) on Environmental impacts, we l, model behaves more conservative. Thus, increasing the param-
consider seven values for alpha parameter with fixed values for eter l and/or the parameter a implies a higher level of risk-aversion.
lambda (l ¼ 0.1) and beta (b ¼ 0.1). The results of this experiment However as it is illustrated above, by increasing the value of l
the value of CVaR decreases. By reviewing Figs. 11 and 12 as well as
Table 5, we can also conclude that the model tries to decrease the
second-stage costs by increasing the value of l. Indeed, adding to
1.5E+08 the value of lambda lead to a more risk-averse behavior so the
performance of the model will be more conservative. Consequently,
1.0E+08 the costs decrease as it is illustrated in Figs. 11 and 12.
On the other hand, by adding to the value of l, which leads to
5.0E+07
more risk-aversion behavior, the values of revenue and profit (first
0.0E+00 objective function) decrease.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 We evaluate the changing behavior of environmental impact by
considering six different values for lambda (l).
ALPHA Fig. 14 shows increase in the value of environmental impacts by
augmentation of lambda (l). Indeed by increasing the risk-aversion
CVaR Planning Stage Costs behavior of the model the environmental impact increases. We can
investigate the reason of this increase in expected amount of FA and
Designing Stage cost CA. As it is shown in Fig. 15 by increasing the value of lambda,
expected value of FA and CA increases. So an increase in the envi-
ronmental impact is expected.
Fig. 6. Alpha vs. costs of the network.
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1577

Table 4
Complete results of alpha effect on profit.

alpha (a) 0.1 0.2 0.3 0.4 0.5 0.6 0.7

CVaR 2.372990 Eþ7 2.542603 Eþ7 2.644553 Eþ7 2.739842 Eþ7 2.861517 Eþ7 3.026018 Eþ7 3.215826 Eþ7
First-stage costs 2187649.3 2187649.3 2187649.3 2187649.3 2187649.3 2187649.3 2187649.3
Second-stage costs 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8
Revenue 2.367841 Eþ8 2.367841 Eþ8 2.367841 Eþ8 9.940904 Eþ7 9.928737 Eþ7 9.912286 Eþ7 9.89 Eþ7
Profit 9.977589 Eþ7 9.960628 Eþ7 9.950433 Eþ7 9.940904 Eþ7 9.928737 Eþ7 9.912286 Eþ7 9.89 Eþ7

6.30E+07 3.0E+08

Environmental impact
6.25E+07 2.0E+08
6.20E+07
1.0E+08
6.15E+07
0.0E+00
6.10E+07 0 0.1
6.05E+07 lambda
0.1 0.2 0.4 0.5 0.7 0.8
alpha Revenue Second stage costs
Profit CVaR
Fig. 8. Alpha vs. Environmental impacts.
First stage cost

Alpha vs. expected value of FA and CA Fig. 11. Lambda vs. Profit and costs variables.

To investigate about the effect of lambda on social impact


objective function, we consider seven values for lambda (l) as well
as fixed values for alpha (a ¼ 0.1) and beta (b ¼ 0.1).
As it is illustrated in Fig. 16, the value of social objective function
decreases by increasing the value of lambda.
We also consider some values for beta (b), which is the rate of
on-site separation, with fixed values of alpha (a ¼ 0.1) and lambda
(l) to analyze the effect of this parameter on profit objective
function. The results of this experiment is shown in Fig. 17. The
results show upward trend of change in the value of profit when the
values of beta (b) increase. Indeed by adding to the value of b the
model tends to decrease the opening cost as well as capacity
Fig. 9. Alpha vs. expected value of FA and CA.

3.0E+08
Alpha vs. social impacts
2.0E+08

1.0E+08

0.0E+00
3 6
-1.0E+08
lambda

Revenue Second stage costs


Profit CVaR
First stage cost

Fig. 10. Alpha vs. social impacts. Fig. 12. Lambda vs. Profit and costs variables.
1578 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

Lambda vs. Environmental impact


3.0E+08
2.0E+08
1.0E+08
0.0E+00
0.6 2
lambda

Revenue
Second stage costs
profit
CVaR Fig. 14. Lambda vs. Environmental impact.
First stage cost
Fig. 13. Lambda vs. Profit and costs variables.
3500000

Expected value of FA and


3000000
expansion cost which adds to the value of the profit maximization
objective function. 2500000
The effects of beta on environmental and social impacts are 2000000
illustrated in Figs. 18 and 19, respectively.
1500000

CA
Fig. 18 shows that value of environmental impact decreases
while the value of the beta (b) increases. On the hand the value of 1000000
social objective function increases by augmentation of beta (b) 500000
parameter which is depicted in Fig. 19.
So by reviewing Figs. 17e19, a high value for parameter beta (b)
0
is highly recommended. So it would be better to improve on-site 0 0.05 0.1 0.2 0.3 0.4
separation activities rather than separating them in the recycling lambda
center which impose more cost on the network of the supply chain.

5. Conclusion Fig. 15. Lambda vs. expected value of FA and CA.

In this paper we studied design and planning a sustainable


reverse logistics network design for recycling C&D wastes. We tried
to consider all three pillars of sustainability in the mathematical
model. So the objective functions of our model consists of the
maximization of profit and social impacts as well as minimization
of environmental impacts of the network. Considering sustain- -2.1E+05
ability and risk-aversion in the field of reverse logistics network

Social impact
design for recycling C&D wastes covers a gap in the literature we
discovered. We used CVaR as a coherent risk measure to consider -7.1E+05
risk-aversion in our two-stage stochastic programming model. The
developed model considers two stochastic parameters: demand of
-1.2E+06
FA and CA and rate on investment. We also considered on-site
separation as well as off-site separation in the model which is
novel on the subject of recycling C&D wastes. -1.7E+06
To solve the proposed model we used epsilon constraint method
0 0.05 0.1 0.25 0.35 0.4 0.5 0.6
and obtained optimal solutions on Pareto front. We also selected
three important parameters of the model to conduct sensitivity lambda
analysis. Some instances were conducted in order to evaluate the
effects of these parameters on the objective functions of the pro-
posed model. The results showed that by increasing the value of a Fig. 16. Lambda vs. Social impact.

Table 5
Complete results of lambda effect on profit.

lambda (l) 0 0.1 0.6 2 3 6

CVaR 5.283643 Eþ7 2.372990 Eþ7 2.372990 Eþ7 2.313619 Eþ7 1.694778 Eþ7 1.604001 Eþ7
First-stage costs 2.187649 Eþ6 2.187649 Eþ6 2.187649 Eþ6 2.187649 Eþ06 2.187649 Eþ6 2.187649 Eþ6
Second-stage costs 1.322288 Eþ8 1.322288 Eþ8 1.322288 Eþ8 1.294886 Eþ8 1.128435 Eþ8 1.127567 Eþ8
Revenue 2.367841 Eþ8 2.367841 Eþ8 2.367841 Eþ8 2.330299 Eþ8 2.034653 Eþ8 1.984676 Eþ8
Profit 1.023676 Eþ8 9.977589 Eþ7 8.940887 Eþ7 8.422536 Eþ7 3.102792 Eþ7 2.584270 Eþ7
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1579

and its weight parameter (l). Therefore, the model has high flexi-
1.1E+08 bility based on the risk adversity of the managers.
1.1E+08 For future research as we mentioned before there are a few
1.1E+08 factors for definition of social impacts in mathematical models
1.0E+08 which have been used by researchers. Concepts like workplaces'

Profit
sanitary conditions, job security and education of workers can be
1.0E+08 quantified in models to measure social impacts. The developed
1.0E+08 model considered just two fundamental products of recycling C&D
9.8E+07 wastes while we can produce more products from them. So
9.6E+07 designing a multi-product model is also recommended. Presenting
9.4E+07 a real case study with a meta-heuristic algorithm for solving large-
size problems can be another interesting research gap.
0.05 0.15 0.3 0.45 0.65 0.8 0.9 1
beta
a. Appendix

Mean-CVaR two stage-stochastic programming


Fig. 17. Beta vs. Profit.

The general structure of a two-stage stochastic programming is


as follows (Birge and Louveaux, 1997):
7.0E+07 X
minEðf ðx; uÞÞ ¼ cT x þ
T
6.0E+07 ps bs ys
S:t:Ax ¼ d
Env. impact

5.0E+07
4.0E+07 B2 x þ Ds ys ¼ hs
(a.1)
3.0E+07 s2S
2.0E+07 x0
1.0E+07 ys  0; s2S
0.0E+00 The first term of Eq. (a.1) refers to the first-stage costs resulted
0.05 0.1 0.2 0.4 0.5 0.7 0.8 1 from first-stage decision variables (vector x), which must be made
here and now and before taking other actions and the second one
beta
calculates the second-stage costs constituted of second-stage de-
cision variables (vector y) which are made to compensate for the
effects of the first-stage decision variables with the probability of ps
Fig. 18. Beta vs. Environmental impact.
on sampling space of u:
The mean-risk two-stage stochastic programming is formulated
the value of profit and social impacts decrease and the value of the as follows (Noyan, 2012; Soleimani and Govindan, 2014):
environmental impact increases. On the other hand an increase to
minfE½f ðx; uÞ þ lpðf ðx; uÞÞg (a.2)
the value of l leads to a decrease to the value of profit and an in-
crease to the environmental impacts. The evaluation of the
where, l is a non-negative weighted coefficient of risk part, which is
parameter b showed that by increasing to the value of this
determined based on the decision maker's preferences to take risk,
parameter the model increases the value of profit and social im-
and r is a specific risk measure (e.g. CVaR).
pacts and decreases the environmental impact. So from the
By replacing CVaR as a risk measure in Eq. (a.2) we have the
managerial point of view, it would be better for decision makers to
following equation:
choose a high value for parameter b. Besides, the managers can
choose their level of risk in the CVaR through turning parameter a minfE½f ðx; uÞ þ lCVaRðf ðx; uÞÞg (a.3)
According to the Eq. (a.1) we can rewrite Eq. (a.3) as follows:
Beta vs. Social impact fE½f ðx; uÞ þ lCVaRðf ðx; uÞÞg
¼ C T x þ EðQ ðx; uÞÞ þ lC T x þ lCVaRðQ ðx; uÞÞ (a.4)
¼ ð1 þ lÞC T x þ EððQ ðx; uÞÞÞ þ lCVaRðQ ðx; uÞÞ

where Q ðx; uÞ is the second-stage costs which is also known as


recourse function.

We can also formulate CVaR of random variable Z with the


confidence level a as follows:

1
fða; h; xÞ ¼ h þ EfmaxfZðx; uÞ  h; 0gg (a.5)
1þa
Based on the definition of CVaR in Eq. (a.5) we equivalently
reformulate Eq. (a.4) as follows (Noyan, 2012; Soleimani and
Fig. 19. Beta vs. Social impact. Govindan, 2014):
1580 M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581

References
!
X
N
1 X N
minð1 þ lÞC T x þ pi qTi yi þ hþ pv Alshamsi, A., Diabat, A., 2015. A reverse logistics network design. J. Manuf. Syst. 37,
i¼1
1  a i¼1 i i 589e598.
Alumur, A., Nickel, S., Saldanha-da-Gama, F., Verter, V., 2012. Multi period reverse
S:t: Wi yi ¼ hi  Ti x; i ¼ 1; :::; N; logistics network design. Eur. J. Oper. Res. 220, 67e78.
Amin, H.S., Zhung, G., 2013. A new multi-objective stochastic model for a forward/
x2X (a.6) reverse logistic network design with responsiveness and quality level. Appl.
Math. Model. 37, 328e344.
yi  0; i ¼ 1; :::; N Arampantzi, C., Minnis, I., 2017. A new model for designing sustainable supply chain
networks and its application to a global manufacturer. J. Clean. Prod. 156,
vi  qTi yi  h; i ¼ 1; :::; N 276e292.
Ayvaz, B., Bolat, B., Aydin, N., 2015. Stochastic reverse logistics network design for
h2R; Vi  0; i ¼ 1; :::; N waste of electrical and electronic equipment. Resour. Conserv. Recycl. 104 (Part
B), 391e404.
Note that in Eq. (a.6) you can interpret h (value at risk) as the Bing, X., Bloemhof-Ruwaard, J.M., van der Vorst, J.G.A.J., 2014. Sustainable reverse
first-stage variable and vi as the second-stage variable. logistics network design for household plastic waste. Flex. Serv. Manuf. J. 26,
119e142.
Birge, J.R., Louveaux, F., 1997. Introduction to Stochastic Programming. Springer
Series in Operations Research. Springer-Verlag, Berlin/Heidelberg.
Demirel, E., Demirel, N., Gocken, H., 2014. A mixed integer linear programming
model to optimize reverse logistics activities of end-of-life vehicles in Turkey.
b. Appendix J. Clean. Prod. 112 (3), 2101e2113.
Devika, K., Jafarian, A., Nourbakhsh, V., 2014. Designing a sustainable closed-loop
supply chain network based on triple bottom line approach: a comparison of
There are many method in the literature to solve multi-objective
meta-heuristics hybridization techniques. Eur. J. Oper. Res. 235, 594e615.
optimization problems. In this article we have applied epsilon- Dinler, E., Gungor, Z., 2017. Planning decisions for recycling products containing
constraint method in order to obtain Pareto optimal solutions. hazardous and explosive substances a fuzzy multi-objective model: Resources.
Conserv. Recycl. 117, 93e101.
The epsilon-constraint method is identified as a posteriori method
EPA, 2015. Greenhouse Gas Equivalencies Calculator [WWW Document]. https://
that can present a suitable picture of a Pareto optimal set for de- www.epa.gov/energy/greenhouse-gas-equivalencies-calculator.
cision makers; then, the decision maker can choose the most Eskandarpour, M., Dejax, P., Miemczyk, J., Pe ton, O., 2015. Sustainable supply chain
favored solution (Pishvaee and Razmi, 2012). In the epsilon- network design: an optimization-oriented review. Omega 54, 11e32.
Govindan, K., Jafarian, A., Nourbakhsh, V., 2015. Bi-objective integrating sustainable
constraint method, one of the multiple objectives is taken as order allocation and sustainable supply chain network strategic design with
objective function and the other are incorporated as the constraints stochastic demand using a novel robust hybrid multi-objective metaheuristic.
of the model. Consider the following mathematical multi-objective Comput. &Oper. Res. 62, 112e130.
Govindan, K., Paam, P., Abtahi, A.R., 2016. A fuzzy multi-objective optimization
programming problem: model for sustainable reverse logistics network design. Ecol. Indic. 67, 753e768.
JEMA, 2010. Consider the Life Cycle of the Refrigerator [WWW Document]. https://
  www.jema-net.or.jp/English/businessfields/environment/data/summary_
max f 1 ðxÞ; f 2 ðxÞ; …f p ðxÞ consider.pdf.
s:t: (b.1) Jeong, Y.-S., Lee, S.-E., Huh, J.-H., 2012. Estimation of CO2 emission of apartment
buildings due to major construction materials in the Republic of Korea. Energy
x2S;
Build. 49, 437e442.
Kannegiesser, M., Gunther, H., Authenrieb, N., 2015. The time-to-sustainability
where x is the vector of decision variables, f1(x), f2(x) … fp(x) are optimization strategy for sustainable supply network design. J. Clean. Prod.
the p objective functions and S is the feasible region. In the e- 108, 451e461.
Mavrotas, G., 2009. Effective implementation of the e-constraint method in multi-
constraint method we optimize one of the objective functions using objective mathematical programming problems. Appl. Math. Comput. 213,
the other objective functions as constraints, incorporating them in 455e465.
the constraint part of the model as shown below (Mavrotas, 2009): Moghaddam, S., Fuzzy, K., 2015. multi-objective model for supply selection and
order allocation in reverse logistics systems under supply and demand uncer-
tainty. Expert Syst. Appl. 42, 6237e6254.
max f 1 ðxÞ Noyan, N., 2012. Risk-averse two-stage stochastic programming with an application
s:t: to disaster management. Comput. Oper. Res. 39, 541e559.
Pishvaee, M.S., Razmi, J., 2012. Environmental supply chain network design using
f 2 ðxÞ  e2 multi-objective fuzzy mathematical programming. Appl. Math. Model. 36,
f 3 ðxÞ  e3 (b.2) 3433e3446.
… Qin, Z., Ji, X., 2010. Logistics network design for product recovery in fuzzy envi-
f 2 ðxÞ  e2 ronment. Eur. J. Oper. Res. 202 (2), 479e490.
Ramezani, M., Bashiri, M., Tavakkoli-Moghadam, R., 2013. A new multi-objective
x2S; stochastic model for a forward/reverse logistic network design with respon-
siveness and quality level. Appl. Math. Model. 37, 328e344.
By changing the values of RHS of the constraint objective Roghanian, E., Pazhoheshfar, P., 2015. An optimization model for reverse logistics
functions (ei) the Pareto optimal solutions are obtained. But finding network under stochastic environment by using genetic algorithm. J. Manuf.
the RHS values is not a trivial tasks. For this we need to calculate the Syst. 33, 348e356.
Seuring, S., Müller, M., 2008. From a literature review to a conceptual framework for
range of at least (p-1) objective function. The range of objective sustainable supply chain management. J. Clean. Prod. 16 (15), 1699e1710.
functions consists of the best value and the nadir value. For Simic, V., 2015. A two-stage interval-stochastic programming model for planning
calculating the range of the objective function incorporated in the end-of-life vehicles allocation under uncertainty. Resour. Conserv. Recycl. 98,
19e29.
constraint, there are two possible methods: 1) the conventional
Sinha, S., Shankar, R., Taneerananon, P., 2010. Modelling and case study of reverse
method which optimize each of the objective function separately to logistics for construction aggregates. Int. J. Logist. Syst. Manag. 6 (1), 39e59.
obtain the ranges 2) lexicographic method which optimize the first Soleimani, H., Govindan, K., 2014. Reverse logistics network design and planning
utilizing conditional value at risk. Eur. J. Oper. Res. 237, 487e497.
objective function and then optimize the second objective function
Soleimani, H., Seyyed-Esfahani, M., Kannan, G., 2013. Incorporating risk measures in
with obtaining the optimal solution of the first objective function closed-loop supply chain network design. Int. J. Prod. Res. 52, 1843e1867.
by incorporating it to the constraint of the model and then this Soleimani, H., Govindan, K., Saghafi, H., Jafari, H., 2017. Fuzzy multi-objective sus-
process continues so that we end up with the objective functions. tainable and green closed-loop supply chain network design. Comput. Ind. Eng.
109, 191e203.
Lexicographic method for obtaining Pareto sets is highly The Consumer Goods Forum, 2012. KPI-team: Sustainability Measures for Logistical
recommended.
M. Rahimi, V. Ghezavati / Journal of Cleaner Production 172 (2018) 1567e1581 1581

Activities [WWW Document], p. 2016. http://ecr-all.org/upload/blogfiles/685/ logistics network design under uncertainty. J. Clean. Prod. 164, 1248e1267.
CO2-GHG-EnergyConsumption-2016-KPI-Team_FINAL.pdf. Zeballos, L.J., A.Mendez, C., P.Barbasa-Povoa, A., 2016. Design and planning of
Yanik, Seda, 2014. Reverse logistics network design under the risk of hazardous closed-loop supply chains: a risk-averse multistage stochastic approach. Ind.
materials transportation, human and ecological risk assessment. Int. J. 21, Eng. Chem. Res. 55, 6236e6249.
1277e1298. Zhou, X., Zhou, Y., 2015. Designing a multi-echelon reverse logistics operation and
Yu, H., Solvang, W., 2017. A carbon-constrained stochastic optimization model with network: a case study of office paper in Beijing: Resources. Conserv. Recycl. 100,
augmented multi-criteria scenario-based risk-averse solution for reverse 58e69.

You might also like