Download as pdf or txt
Download as pdf or txt
You are on page 1of 33

International Journal of Production Research

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tprs20

A joint economic lot-size model for sustainable


industries of recycled content products

Mostafa Parsa , Ali Shahandeh Nookabadi & Zümbül Atan

To cite this article: Mostafa Parsa , Ali Shahandeh Nookabadi & Zümbül Atan (2020): A joint
economic lot-size model for sustainable industries of recycled content products, International
Journal of Production Research, DOI: 10.1080/00207543.2020.1802078

To link to this article: https://doi.org/10.1080/00207543.2020.1802078

Published online: 13 Aug 2020.

Submit your article to this journal

Article views: 8

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
https://doi.org/10.1080/00207543.2020.1802078

A joint economic lot-size model for sustainable industries of recycled content


products
Mostafa Parsaa,b , Ali Shahandeh Nookabadib and Zümbül Atan a

a Department of Industrial Engineering and Innovation Sciences, Eindhoven University of Technology, Eindhoven, The Netherlands;
b Department of Industrial and Systems Engineering, Isfahan University of Technology (IUT), Isfahan, Iran

ABSTRACT ARTICLE HISTORY


Coupling forward and reverse operations in closed-loop supply chains (CLSCs) can lead to a higher Received 2 July 2019
degree of sustainability. Previous studies on CLSCs have focused on the products that are either Accepted 15 July 2020
purely remanufactured or purely new. The emphasis on recycling necessitates research that consid- KEYWORDS
ers recycled content products, which are made from a mix of virgin and recycled material. We study Sustainability; multi-echelon
a CLSC consisting of a manufacturer, a retailer, a supplier, a material recovery facility (MRF) and a inventory; closed-loop
recycling facility. The supplier and the recycling facility provide the virgin and recycled raw mate- supply chains; recycled
rials, respectively, to the manufacturer who mixes them based on a predefined proportion known content products; joint
as the recycled content level to produce the finished products for the retailer. The recycling facility economic lot-sizing problem
recovers the manufacturer’s production waste as well as the used products collected by the MRF.
We aim at maximising the chain-wide profit by optimising the shipments among the supply chain
parties. We propose a Branch-and-Bound algorithm to obtain the optimal solution, and a heuristic
procedure, which provides the optimal solution in 96% of all tested instances. In addition, we prove
the optimality of a bang-bang policy for the recycled content level in most cases.

1. Introduction dollars in Europe and 4.5 trillion US dollars globally by


Sustainable development is defined as ‘development that 2030 (Li et al. 2020). Just through electronic waste recy-
meets the needs of the present without compromising the cling, CE can generate revenues of about 2.15 billion euro
ability of future generations to meet their own needs’ in Europe (Bressanelli, Perona, and Saccani 2019). The
(World Commission on Environment and Development transition can also cut CO2 emissions from heavy indus-
(WCED) 1987). Over the last few years, there has been a try by 296 million tons per year in the EU and 3.6 billion
significant increase in the demand for natural resources tons per year globally (Enkvist and Klevnäs 2018).
due to the population growth (Govindan and Hasanagic In addition to practitioners, the researchers have also
2018). Estimations indicate that by 2050 production sys- recognised the topic of circular economy as an important
tems will annually devour about 140 billion tons of fossil component of production and operations management
fuels, ores, minerals and biomass (Santibanez Gonza- research (Choi, Taleizadeh, and Yue 2020; Yang et al.
lez, Koh, and Leung 2019) tripling the current material 2019). Figure 1 represents different research areas within
consumption (Govindan and Hasanagic 2018). To tackle CE (Govindan, Soleimani, and Kannan 2015; Lieder and
the associated challenges, a transition towards Circular Rashid 2016). One particular area that has received a
Economy (CE) is considered as a key action for sustain- lot of attention studies the closed-loop supply chains
able development (Lechner and Reimann 2019) due to its (CLSCs). The CLSCs simultaneously consider the pro-
substantial potential for reducing negative environmen- duction and material flow from suppliers to manufac-
tal impact while generating economic benefits (Bai et al. turers to distributors to retailers and to end-consumer
2019). CE advocates regenerative and restorative produc- (forward supply chain) as well as the reverse flows of used
tion systems by closing the loop of the linear product products (reverse supply chain) (Govindan, Soleimani,
lifecycle (Batista et al. 2019). The transition to CE has a and Kannan 2015; Souza 2013). Among the most impor-
high potential to generate a net benefit of 2.0 trillion US tant operational decisions for the players of CLSCs are the

CONTACT Mostafa Parsa mostafa.parsa2008@gmail.com Department of Industrial Engineering and Innovation Sciences, Eindhoven University of
Technology, P.O. Box 513, 5600 MB, Eindhoven, The Netherlands; Department of Industrial and Systems Engineering, Isfahan University of Technology (IUT),
Isfahan 84156-83111, Iran

© 2020 Informa UK Limited, trading as Taylor & Francis Group


2 M. PARSA ET AL.

Figure 1. Different research areas of circular economy.

production and inventory management decisions (Guide (Richter 1996a, 1996b). Richter (1997) shows that the
and Van Wassenhove 2009). In this paper, we consider pure strategy of either no waste disposal (full remanu-
these decisions. facturing) or no remanufacturing (full waste disposal)
Selected surveys on cutting edge problems in produc- dominates the mixed approach. Subsequent work con-
tion research (Dolgui 2019) indicate the optimisation of firms the dominance of the pure strategy over the mixed
inventory and product return policies as one of the most strategies (Dobos and Richter 2000, 2003, 2004; Richter
essential aspects of efficient CLSC management strategies and Dobos 1999; Teunter 2001).
(Kazemi, Modak, and Govindan 2019). Schrady (1967) All of the above-mentioned studies consider an inven-
is the earliest study on production-inventory problems tory system of single-echelon supply chain with pro-
with returned materials. This study develops a determin- duction/manufacturing and remanufacturing. There are
istic model with one production/manufacturing batch multiple studies on inventory management of multi-
succeeding R remanufacturing batches. The policy is echelon CLSCs. Chung, Wee, and Yang (2008) consider a
known as the (1,R) policy. Both processes are assumed (1,1) policy and develop an inventory model for a single-
to be instantaneous. Nahmias and Rivera (1979) gener- item multi-echelon CLSC consisting of a manufacturer, a
alise this model by considering a finite remanufacturing supplier, a collector, and a retailer. In addition to the tra-
rate. Teunter (2004) studies infinite and finite manufac- ditional forward flow, reverse flow is considered. In the
turing and remanufacturing rates with no disposal under reverse flow, the collector returns the used products to the
(1,R) and (M,1) policies. The author obtains a square root manufacturer who remanufactures them and then ships
relation for the optimal manufacturing and remanufac- them to the retailer for sale. The results show a signifi-
turing lot-sizes and proposes a modification for lot-sizes cant increase in the total profit when centralised decision
to allocate integer values to R and M. Konstantaras and making is adopted compared to the decentralised one.
Papachristos (2006), Konstantaras and Skouri (2010), Multiple studies extend this work in different directions.
and Widyadana and Wee (2010) extend the model of The work of Chung, Wee, and Yang (2008) is gener-
Teunter (2004). alised by Yuan and Gao (2010) for the (1,R) and (M,1)
Choi, Hwang, and Koh (2007) extend the (M,R) policy policies, by Yang et al. (2010) for the (M,R) policy with
by treating the sequence of procuring and remanufactur- deterioration, and by Giri and Sharma (2015) for the
ing batches within a system cycle as decision variables. (M,R) policy with imperfect manufacturing and quality
For the (M,R) policy, Bazan, Jaber, and El Saadany (2015) based return rate. Wee and Chung (2009) model two-
limit the number of times an item can be remanufactured echelon integrated deteriorating production inventory
and consider costs related to greenhouse gas emissions system with remanufacturing, just in time (JIT) deliver-
and used energy level. For the same policy, multiple stud- ies and green component design. Mitra (2009) develops
ies treat the waste disposal rate as a decision variable stochastic and deterministic models for a single-depot
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 3

single-distributor supply chain with customer returns. (2019) develop closed-form expressions for the global
Mitra (2012) generalises this model to correlated returns solution of the manufacturing-remanufacturing prob-
and demand rates. Teng et al. (2011) study the effect of lem. Godichaud and Amodeo (2019) propose economic
increasing stock to induce customers and partial back- order quantity (EOQ) inventory models for disassem-
ordering in the model of Mitra (2009). Giri and Glock bly systems that can be viewed as inventory systems
(2017) consider a single-manufacturer single-retailer with returned products. Zhang, Wu, and Wei (2019)
CLSC with stochastic product returns with learning investigate the optimal design of quality recovery and
and forgetting in production and inspection of returned refurbishment strategies for the manufacturer’s produc-
products and develop a model to optimise the inven- tion planning. Sun, Xia, and Liu (2020) establish inven-
tory decisions. Ma, Li, and Wang (2017) study the pricing tory lot-sizing models for a CLSC production system
decisions for the CLSC members under different col- from a product life-cycle perspective including introduc-
lection structures where a central planner, a retailer, a tion, maturity, and decline phases. Reddy and Kumar
manufacturer or a third-party collects the returned used (2020) study the effect of capacity usage differences
products. between manufacturing and remanufacturing operations
Multiple studies consider different types of collab- on inventory, capacity and production planning where
orative efforts in CLSCs. Jaber, Zanoni, and Zavanella new and remanufactured items are sold in primary and
(2014) consider the waste disposal problem under the secondary markets, respectively. Zouadi, Yalaoui, and
(M,R) policy of a manufacturer managing a retailer’s Reghioui (2018) develop a lot-sizing model for manu-
inventory according to a consignment stock (CS) agree- facturing/remanufacturing systems integrating a global
ment. Yuan et al. (2015) extend the work of Yuan emission constraint. Chu, Zhong, and Li (2018) con-
and Gao (2010) to a partly centralised decision-making sider a CLSC where multiple manufacturers share a joint
structure and study its effect on inventory policies and third-party recycler. They study the impact of the joint
expected profits of all supply chain members. Hariga, third-party’s scale on the efficiency of reverse channels
As’ad, and Khan (2017) consider a single-manufacturer and compare joint third-party-managed collection with
single-retailer CLSC under a CS agreement where the retailer-managed (used by Kodak) and manufacturer-
system cycle time and the sequence of manufacturing and managed (used by HP, Xerox, and Canon) collections.
remanufacturing batches are jointly optimised. Bazan, To better understand the contribution of this study
Jaber, and Zanoni (2017) also consider the same sys- and to compare it with the state of the art, a classifica-
tem and extend the research of Bazan, Jaber, and El tion of the relevant literature from several viewpoints is
Saadany (2015) to study classical and CS coordination provided in Table 1. Previous studies consider products
policies. Hasanov, Jaber, and Tahirov (2019) addresses that are either purely remanufactured or purely new. We
coordination of the production and inventory policies in contribute to the literature by considering recycled con-
a four-echelon CLSC including multiple tier-1 and tier- tent products made from a mix of virgin and recycled
2 suppliers, a manufacturer, and multiple buyers. They materials with specific proportions. This proportion is
develop two inventory models with and without consid- called recycled content level and defined in (ISO 14021
ering environmental costs related to the greenhouse gas 2016) as ‘the mass percentage (X%) of recycled mate-
emissions and used energy based on the idea of Bazan, rials in a product’. When recycled material is diverted
Jaber, and Zanoni (2017). from the manufacturing waste or scrap, it is referred to
In a recent study, Mawandiya, Jha, and Thakkar as pre-consumer or post-industrial material and when it
(2020) propose a production-inventory model for an is diverted from consumers’ EOL products, it is referred
integrated CLSC consisting of a retailer and a manufac- to as post-consumer material.
turer/remanufacturer. This study aims at understanding Our study aims at benefiting industries that have sus-
the conditions under which the CLSC is more prof- tainability goals to produce recycled content products. In
itable than the forward supply chain. Taleizadeh and current practice, the paper industry is among those that
Moshtagh (2019) study the (1,1) policy and consider have almost completely met the society’s expectations
two different qualities for manufactured and remanu- for the environmental compatibility of their processes,
factured items and imperfect remanufacturing. As’ad, the sustainability of their raw material, and the recycla-
Hariga, and Alkhatib (2019) jointly optimise finished bility of their products (Bajpai 2014). Figure 2 depicts
product shipping and raw material replenishment strate- the lifecycle of recycled content paper (Arjowiggins
gies, as well as lot-sizing decisions and the sequence of Graphic 2014). Fundamentally paper manufacturing has
manufacturing and remanufacturing batches for a single- two major steps; pulp production and paper production.
manufacturer single-retailer CLSC under the CS agree- Pulp production can be of two types; virgin pulp produc-
ment. For the same system, Ben-Daya, As’ad, and Nabi tion and recycled pulp production (Rahman, Hussain,
4
M. PARSA ET AL.
Table 1. Classification of the relevant literature.
Product Type
Supply chain new or reman- Recycled Production Reman. Waste Disposal Replenishment
Study configuration ufactured content Sequence Demand Manu. Rate Rate Model policy Shortages deterioration
Schrady (1967) Single-echelon  - (1,R) D1 Infinite Infinite - LFL3 - -
Nahmias and Rivera (1979) Single-echelon  - (1,R) D Infinite Finite - LFL - -
Richter (1996a, 1996b) Single-echelon  - (M,R) D Infinite Infinite  LFL - -
Teunter (2004) Single-echelon  - (M,1) & (1,R) D Finite Finite - LFL - -
Konstantaras and Papachristos (2006) Single-echelon  - (M,1) & (1,R) D Finite Finite - LFL FB11 -
Choi, Hwang, and Koh (2007) Single-echelon  - General D Infinite Finite - LFL - -
Chung, Wee, and Yang (2008) Four-echelon  - (1,1) D Finite Finite - NDES4 - -
Wee and Chung (2009) Two-echelon  - Simultaneous D Finite Finite - NDES - 
Mitra (2009) Two-echelon  - Simultaneous D & S2 Infinite Infinite - DES5 FB -
Widyadana and Wee (2010) Single-echelon  - (M,1) & (1,R) D Finite Finite - LFL - -
Yang et al. (2010) Four-echelon  - (M,R) D Finite Finite - NDES - 
Yuan and Gao (2010) Four-echelon  - (M,1) & (1,R) D Finite Finite - NDES - -
Teng et al. (2011) Two-echelon  - Simultaneous D Infinite Infinite - DES PB12 -
Mitra (2012) Two-echelon  - Simultaneous D&S Infinite Infinite - DES FB -
Jaber, Zanoni, and Zavanella (2014) Two-echelon  - (M,R) D Finite Finite - LFL - -
Yuan et al. (2015) Four-echelon  - (M,1) & (1,R) D Finite Finite - NDES - -
Giri and Sharma (2015) Four-echelon  - (1,1) & (M,R) D Finite Finite - NDES - -
Bazan, Jaber, and El Saadany (2015) Single-echelon  - (M,R) D Finite Finite - LFL - -
Giri and Glock (2017) Two-echelon  - Simultaneous D Finite Finite - NDES - -
Ma, Li, and Wang (2017) Three-echelon  - Simultaneous D N/A N/A - N/A - -
Hariga, As’ad, and Khan (2017) Two-echelon  - General D Finite Finite - LFL - -
Bazan, Jaber, and Zanoni (2017) Two-echelon  - (M,R) D Finite Finite - Classical: NDES CS: LFL - -
Mawandiya, Jha, and Thakkar (2020) Two-echelon  - (1,1) S Finite Finite - NDES FB -
Zouadi, Yalaoui, and Reghioui (2018) Two-echelon  - Simultaneous S N/A N/A - N/A - -
Chu, Zhong, and Li (2018) Three-echelon  - Simultaneous D N/A N/A - N/A - -
Hasanov, Jaber, and Tahirov (2019) Four-echelon  - Simultaneous D Infinite Infinite - LFL - -
Taleizadeh and Moshtagh (2019) Four-echelon  - (1,1) D Finite Finite - NDES LS13 -
As’ad, Hariga, and Alkhatib (2019) Two-echelon  - General D Finite Finite - FP6 : LFL RM7 : OTM8 & MTO9 - -
Ben-Daya, As’ad, and Nabi (2019) Two-echelon  - (M,R) D Finite Finite - LFL - -
Godichaud and Amodeo (2019) Single-echelon  - Simultaneous D Infinite Infinite - LFL LS -
Sun, Xia, and Liu (2020) Single-echelon  - (M,R) D Finite Finite - LFL - -
Reddy and Kumar (2020) Single-echelon  - Simultaneous S N/A N/A - MTS10 LS -
This study Five-echelon -  Simultaneous D Finite Finite - NDES - -
1 Deterministic; 2 Stochastic; 3 Lot-For-Lot; 4 Non-Delayed Equal-Sized; 5 Delayed Equal-Sized; 6 Finished Products; 7 Raw Materials; 8 One-To-Multi is a policy in which one RM shipment covers multiple manufacturing cycles; 9
Multi-To-One is a policy in which multiple RM shipments covers one manufacturing cycle; 10 Make-To-Stock;11 Fully Backordered; 12 Partially Backordered; 13 Lost Sales.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5

Figure 2. Lifecycle of recycled content paper adapted from Arjowiggins Graphic (2014).

and Basri 2014). In virgin pulp production, wood chips far (Armstrong World Industries Inc. 2019). Hence, the
are chemically and/or mechanically processed (Bajpai benefits of using recycled material in manufacturing are
2018). In recycled pulp production used papers are col- evident.
lected, sorted, de-inked and bleached (McKinney 1994). Our research is relevant to all types of recycled content
Both virgin and recycled pulps are then shipped to the products. EPA (2007b) has designated 61 recycled con-
paper manufacturer. The basic manufacturing principle tent products in eight groups: non-paper office products,
is to mix the virgin and recycled pulps together accord- paper and paper products, vehicular products, trans-
ing to the recycled content level (Evans, Sherin, and Lee portation products, park and recreation products, land-
2013). scaping products, construction products and miscella-
Sequana SA is a France-based company, producing neous products. Optimal lot-sizing decisions can provide
cutting-edge sustainable papers. It has two principal huge benefits to the CLSCs of these products. However,
wholly-owned subsidiaries: Arjowiggins manufactures these decisions have not been addressed in the scientific
papers and Antalis distributes them. Arjowiggins has literature yet. To close the research gap, we consider the
integrated the production of recycled pulp and recy- supply chain of a recycled content product depicted in
cled content paper at three manufacturing sites, Green- Figure 3. The supply chain consists of a manufacturer, a
field, Bessé-sur-Braye, and Le Bourray Mills (Arjowig- retailer, a supplier, a material recovery facility (MRF) and
gins Graphic 2019). Arjowiggins produces a wide num- a recycling facility.
ber of products such as high quality 60% recycled con- In the forward supply chain, the supplier and the recy-
tent papers from 60% recycled fiber and 40% virgin cling facility supply the virgin and recycled raw materials
fiber in silk and gloss finish such as ‘Cocoon® 60 Gloss’ to the manufacturer, respectively. The virgin and recycled
and ‘Cocoon® 60 Silk’. By producing Cocoon® recy- raw materials account for (1 − X)% and X% of the total
cled papers rather than virgin fiber papers, the com- required raw materials for production, respectively. The
pany has saved 94,709,228 kg of landfill, 153,880,640 kg manufacturer receives each lot of raw materials in l and
of wood, 13,315,001 kg of CO2 and greenhouse gases, k equal-sized batches from the supplier and the recycling
3,002,508,054 L of water, and 183,010,461 kwh of energy facility, respectively. The raw materials are mixed based
since 2015 (Cocoon® 2019). on the recycled content level and transformed into fin-
Another industry that operates with sustainability ished products. The finished products are shipped to the
targets is the construction industry. Armstrong World retailer in n equal-sized batches at every retailer cycle,
Industries Inc. takes back old fiberglass acoustical ceil- which lasts for Tr time units. The retailer sells the fin-
ing tiles and transforms them into new ones containing ished products to end consumers. In the reverse supply
greater than 50% total recycled content. This company chain, the MRF buys back some of the used products. A
has saved over 1 million tons of virgin raw materials so fraction of the used products returned by consumers are
6 M. PARSA ET AL.

Figure 3. CLSC of the recycled content products.

non-recyclable, i.e. they fail to meet the quality standards development of circular and sustainable supply chains.
required for recycling as specified by the recycling facil- On the other side, low X values mean large environmental
ity. The MRF disposes the non-recyclable returned items, cost to society since more natural resources need to be
processes the remaining recyclable ones and ships them used. Therefore, the two following problems could be
to the recycling facility as the post-consumer materials in worth studying:
m equal-sized batches. In addition, a fraction of the man- 1) [ECOLOGY → ECONOMY]-Tracing the econ-
ufacturer’s production is scrap and sold to the recycling omic impact of ecological concerns (Richter 1997;
facility as the pre-consumer or post-industrial materials. Richter and Dobos 1999): We can achieve this by solv-
Then, both post- and pre-consumer materials are deliv- ing our optimisation problem for multiple X values.
ered to the recycling site to be transformed into recycled Hence, we can show how the optimal profit and opti-
raw materials. mal decisions react to different levels of ecological
Each location delivers inventory to its customer loca- concerns.
tion in equal-sized batches. Our objective is to maximise 2) [ECONOMY → ECOLOGY]-Tracing the ecologi-
the total profit of the integrated system TP(Tr , k, l, m, n) cal impact of economic concerns (Richter 1997; Richter
by finding the optimal number of shipments from a lot, and Dobos 1999): Having a profit maximisation prob-
shipment schedule and batch size for all parties. Hence, lem implies a decision maker with economic concerns.
we are dealing with a Joint Economic Lot-sizing Problem We can explain how these concerns impact X by opti-
(JELP) in the context of integrated CLSCs. mising it. Hence, we can explain how the profit max-
Another objective of this paper is to investigate the imiser can optimally determine the ecological factor
interplay between ecological and economic aspects of X with respect to changing inventory-related decisions
the problem. Understanding and evaluating the effect of (Tr , k, l, m, and n). The optimal policy for X is either the
decisions, which are made based on economic concerns, lowest/highest possible value of X so-called bang-bang
on the ecological factors and vice versa is vital for the policy X ∗ ∈ {Xmin , Xmax } or a stationary point within the
stakeholders including supply chain players, society and defined range of possible values X ∗ ∈ (Xmin , Xmax ). Xmax
government to develop sustainable policies (Archibugi as the optimal solution is favoured by green groups and
and Nijkamp 1989; Beaton and Maser 2016; Parsa et al. environmentalists because natural resources can be saved
2019). The amount of recycled content, which is mea- for the future generations and the firms contribute to
sured by X in our problem, can be used to evaluate the sustainable development purposes at the most level
the ecological (green) aspects of the resulting solutions. (Dobos and Richter 2004). Some conflict might occur
High X values can imply significant contribution to the between economic and ecological approaches when the
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 7

optimal solution is not Xmax . In this case, it is essential to 2. Model formulation


study the ways to resolve the conflicts.
In this section, we develop a mathematical formulation
Accordingly, we address the following research ques-
for the JELP in the integrated CLSC of the recycled
tions:
content products. An overview of the parameters, deci-
sion variables, and dependent variables is provided in
1. What is the optimal shipment policy for the inte-
Appendix A.
grated CLSCs of recycled content products?
We consider the multi-echelon CLSC depicted in
2. How is the profit and shipment policy of each party
Figure 3. The supply chain consists of a manufacturer, a
affected by the manufacturing scrap (post-industrial
retailer, a supplier of virgin materials, a material recov-
materials)?
ery facility (MRF), and a recycling facility. We make the
3. What level of recycled content is the most cost-
following assumptions:
effective for sustainable industries?
4. How can the conflict between economic and eco-
logical approaches on the optimal recycled level be 1. The supply chain is integrated vertically in the sense
addressed? that it is jointly or partially owned by organisa-
tion(s). Hence, system-wide optimisation is accept-
To answer these questions, we formulate the JELP as a able to all members and benefit-sharing among them
mixed integer non-linear programming (MINIP) model. is not an issue (Ben-Daya and Al-Nassar 2008; Ben-
We present a B&B algorithm to find the exact solution to Daya, As’ad, and Seliaman 2013; Ben-Daya, Dar-
the problem. In addition, we develop a fast and accurate wish, and Ertogral 2008). This
heuristic procedure. It has given the optimal solution in assumption is consistent with practice. For exam-
96% of all tested instances. Hence, the heuristic can help ple, Sequana SA is a vertically integrated company
managers and policymakers to make quick and reliable that has two principal wholly-owned subsidiaries:
decisions. Our numerical results indicate that a decrease I) Arjowiggins Graphic that manufactures recy-
in manufacturing scrap fraction reduces the number of cled pulps in Greenfield mill and recycled papers
shipments to each party. It increases the manufacturer’s, in Bessé-sur-Braye and Le Bourray mills, and II)
MRF’s, and the whole supply chain’s profits. On the other Antalis that operates in the distribution business.
hand, decrease in the scrap fraction results in declines Other strong examples of successful vertically inte-
in the profits of the supplier and the recycling facility. grated supply chains are Zara SA, Alibaba Group
The scrap fraction has a negligible effect on the retailer’s Holding Limited, Luxottica Group SpA, Exxon
profit. We provide important insights for industries striv- Mobil Corporation, and British Petroleum PLC.
ing for sustainability. We prove, for the recycled content 2. We consider a non-delayed equal-sized shipment
level, that one of the two extreme (bang-bang) strategies policy (Lu 1995). This policy allows shipments dur-
is the best level, while the stationary point within the ing production phases and there is no need to wait
range can be optimal under certain limited conditions. until the whole lot is produced. Therefore, each lot
We use break-even analysis to present practical solutions in the supply chain is delivered to the adjacent party
to resolve the possible conflict of interest between eco- in a number of equal-sized batches and deliveries are
nomic and ecological approaches. Finally, we extend our allowed to occur during the production of the lot.
model to solve the same problem for assembly systems. 3. The time horizon is infinite since joint economic
The remainder of the paper is organised as follows. lot-sizing models are particularly useful when com-
The MINLP model is formulated in Section 2. The B&B panies have established long-term relationships with
and heuristic algorithms are presented in Sections 3 and their customers and suppliers (Glock 2012).
4, respectively. The heuristic procedure is compared with 4. Shortages are not allowed. Therefore, the demand
the B&B algorithm through computational results in rate is less than or equal to the perfect production
Section 5. In Section 6, we analyze the effects of scrap rate (D ≤ βP).
fraction and recycled content level on the optimal lot-
sizing decisions. In addition, we provide a decision tree Figure 4 illustrates the behaviour of inventory levels
to support the managers’ decision on the optimal level of all parties in the CLSC. Our objective is to deter-
of recycled content. In Section 7, we extend the proposed mine the optimal number of shipments, time interval
model to the assembly industry. We summarise our man- between shipments and batch sizes for all parties involved
agerial insights in Section 8. The concluding remarks and such that the chain-wide total profit per unit time TP
directions for future research are provided in Section 9. is maximised. The chain-wide total profit per period is
We defer the derivations and proofs to the Appendices. the sum of the profits of individual parties. Therefore,
8 M. PARSA ET AL.

Figure 4. Behaviour of the inventory levels over time in the multi-echelon CLSC.

we formulate each player’s profit per unit time as the associated with each production setup. The manufacturer
difference between its revenue and cost per unit time. delivers each production lot to the retailer in n equal-
sized batches at every retailer’s cycle time (Tr ) incurring
a fixed transportation cost Om per shipment.
2.1. Manufacturer’s profit We assume that a fraction β (0 ≤ β ≤ 1) of the man-
The recycled content products are made from a mix of ufacturer’s production is perfect and the rest indicated
virgin and recycled materials with specific proportions. by (1 − β) is manufacturing scrap, i.e. post-industrial
This proportion is called the recycled content level (X). materials. Therefore, the manufacturer should produce
D
It is calculated as the percentage, by mass, of recycled β product units per unit time to avoid shortages where
material in a product (ISO 14021 2016). Therefore, X D perfect finished products meet the retailer demand at
(1−β)D
percentage of raw materials required for production is a price Pr per unit and D β −D= β items per unit
recycled and the rest is virgin. The recycling facility and time are sold to the recycling facility at a price PRm per
supplier deliver the recycled and virgin materials to the unit. Therefore, the manufacturer’s sales revenue per unit
manufacturer in k and l equal-sized batches at price Pmr time is TRm = Pr D + PRm (1−β)Dβ .
and Pmp per unit material, respectively. A fixed ordering Figure 4 shows the same inventory pattern is repeated
cost Amr (Amp ) is charged whenever the manufacturer in each system cycle nTr . In the following, we develop
orders a batch of recycled (virgin) materials. Then, the a mathematical formulation for the manufacturer’s total
manufacturer proportionally mixes both materials based cost TCm . For that purpose, we need to determine the
on the recycled content level and transforms them into expression for production length Tp within the system
finished products. The manufacturer produces items in cycle. Given that shortages do not occur, the perfect pro-
lots at a finite rate P incurring production cost Cp per unit duction should be enough to satisfy the retailer demand
finished product as well as a fixed setup cost Am per lot during the system cycle. Hence, we need the following
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 9

equality to hold: given by TPm = TRm − TCm . Hence, we have:


   2 
βPTp = D(nTr ), (1) (1 − β)D D XnTr
TPm = Pr D + PRm − hmr
β 2β 2 kP
 2 
The left-hand side of Equation 1 calculates the number D (1 − X)nTr
of perfect items manufactured per system cycle and the − hmp
2β 2 lP
right-hand side is the consumption level per system cycle.  
Tr D2 (βP − D)nTr D Tr D
The manufacturer’s holding cost consists of the costs − hm + −
of carrying the finished product, recycled and virgin βP 2βP 2
materials. The unit costs per unit time are hm , hmr , and Om Am kAmr lAmp
− − − −
hmp , respectively. The manufacturer’s total cost per unit Tr nTr nTr nTr
time consists of holding, transportation, setup, ordering,  
Cp D (1 − X)DPmp + XDPmr
production, and purchasing costs can be written as: − − . (2)
β β
Holding cost In Equation (2), the first term represents the manufac-
←−−−−2−−−− −−−−−− −− −−−−−→
+ hmp D (1−X)nT
2
hmr D2βXnT r r turer’s sales revenue, the second term is the holding cost
2 kP 2β 2 lP
TCm =  2  of the recycled materials, the third term is the holding
+hm TβPrD
+ (βP−D)nT
2βP
rD
− Tr D
2 cost of virgin materials, the fourth term is the holding
cost of the finished product, the fifth term is the fixed
Om Am kAmr + lAmp
+ + + transportation cost related to the retailer’s order of any
Tr nTr nTr size, the sixth term is the fixed setup cost, the seventh
←→ ←→ ←−−−−−−−→
Transportation Setup Ordering
term is the ordering cost for recycled materials, the eighth
cost cost cost
  term is the ordering cost for virgin materials, the ninth
Cp D (1 − X)DPmp + XDPmr
+ + . term is the production cost, and the tenth term is the
β β summation of purchasing costs of virgin and recycled
←−→ ←−−−−−−−−−−−−−−−−→
Production
cost
Purchasing cost materials all per unit time.

The detailed calculations for the manufacturer’s holding 2.2. Retailer’s profit
costs are given in Appendix B.1. In the formulation of
other terms, we rely on the fact that frequency and cycle The retailer meets the consumer demand with constant
time are inversely related. The frequency of retailer’s rate D finished products per unit time at retail price Pc per
purchase is T1r per unit time and accordingly, the manu- unit. Hence, the retailer’s sales revenue per unit time is
facturer’s transportation cost for delivering the retailer’s TRr = Pc D. The retailer incurs a fixed cost Ar for order-
order per unit time is OTmr . The production cycle equals ing a batch of finished products from the manufacturer at
the system cycle nTr according to Figure 4. Hence, the every cycle of length Tr . Also, the retailer incurs a hold-
number of production cycles per unit time is nT1 r and the ing cost hr per unit per unit time. The retailer’s total cost
manufacturer’s setup cost per unit time is nTAm
. The num- per unit time, TCr , is the sum of ordering, holding, and
r
ber of orders for recycled and virgin materials that occur purchasing costs. We have:
per unit time can be found by multiplying the number of Ordering cost Holding cost
these orders per production cycle by the frequency of the ←
→ ←−−→ Purchasing cost
Ar DTr ← →
production cycle per unit time. The numbers of orders TCr = + hr + Pr D .
Tr 2
for recycled and virgin materials are k × nT1 r and l × nT1 r ,
respectively. Consequently, the manufacturer’s ordering The number of retailer orders per unit time is 1
kA +lA Tr . Con-
cost is obtained by mrnTr mp . The manufacturer should sequently, the retailer’s ordering cost per unit time is Ar
Tr .
produce D β items per unit time to meet the demand rate The retailer’s average inventory level is DT r
Cp D 2 and the hold-
D. Hence, the manufacturer’s production cost is β . To ing cost per unit time is hr DT r
2 . The retailer buys D items
(1−X)D per unit time from the manufacturer. Hence, the purchas-
produce D β units, the manufacturer needs and XDβ β
units of virgin and recycled unit raw materials, respec- ing cost is Pr D. Finally, the retailer’s profit per unit time
tively. Therefore, the manufacturer’s raw material pur- TPr = TRr − TCr is given by

(1−X)DPmp +XDPmr
chasing cost per unit time becomes β . Ar DTr
TPr = Pc D − − hr − Pr D, (3)
Finally, the manufacturer’s profit per unit time is then Tr 2
10 M. PARSA ET AL.

where the first term is the retailer’s sales revenue, the 2.4. MRF’s profit
second term is the ordering cost, the third term is the
Some consumers return their EOL products to the MRF
holding cost of the finished products, and the fourth term
at a finite rate Ru per unit time (Ru ≤ D). The MRF buys
is the purchasing cost all per unit time.
back the returned products at price PM per unit. Since
some of the returned items fail to meet the quality stan-
2.3. Supplier’s profit dards required for recycling as specified by the recycling
facility, we consider that (1 − α)% of the returned prod-
The supplier meets the demand for (1 − X)% of raw
ucts are non-recyclable. The MRF separates and disposes
materials required for the manufacturer’s production of
D them at a cost Cd per unit. The MRF processes recy-
β per unit time with the virgin materials at price Pmp per clable returned items at a cost of CM per unit. Then, each
unit. Hence, the supplier’s revenue per unit time from the
lot of processed recyclable items is delivered to the recy-
sale is TRs = (1−X)D
β Pmp . The supplier extracts the vir-
cling facility in m equal-sized batches with price PR per
gin raw materials at processing cost Cs per unit material
unit. The MRF sells αRu recyclable used items per unit
and fixed setup cost As per lot. The supplier delivers each
time to the recycling facility. Therefore, its sales revenue
virgin material lot to the manufacturer in l equal-sized
per unit time is TRM = αRu PR . The MRF incurs fixed
batches incurring a fixed transportation cost Os per ship-
transportation cost OM to deliver each batch of recyclable
ment. The supplier also incurs holding cost hs per unit
items for the recycling facility as well as a fixed setup cost
material per unit time. The calculation for the supplier’s
AM per lot of recyclable items associated with each pro-
average inventory level is provided in Appendix B.2. The
cessing setup. There is also a holding cost hu per unit
supplier’s total cost per unit time, TCs , including holding,
recyclable item per unit time. Appendix B.3 provides the
transportation, setup, and extraction costs is as follows:
calculation for the MRF’s average inventory. The MRF’s
Holding cost
total cost per unit time, TCM , consisting of holding,
Transportation cost

 −2−−−−−−−−−−−−−→  ←→ transportation, setup, disposing, material processing, and
D (l − 1)(1 − X)nTr lOs
TCs = hs + purchasing costs is given as follows:
2β 2 lP nTr
Holding cost
←→
Setup cost
←−−−−−−→
Extraction cost ←−−−−−−−−−−−−−−−−−−−→
As Cs (1 − X)D hu nαRu [βmR + β(1 − m)αRu
+ + . +(1 − β)(1 − m)D]Tr
nTr β TCM =
2βmR
As explained before, the number of manufacturer’s mOM AM
orders for virgin materials is nTl r per unit time. Hence, + +
nTr nTr
the supplier’s transportation cost per unit time for deliv- ←−→ ←→
lOs Transportation cost Setup cost
ering order of the manufacturer is nT r
. The number of
times that the supplier extracts virgin raw materials per + Ru PM + (1 − α)Ru Cd + αRu CM .
←−→ ←−−−−−−→ ←−−→
unit time is nT1 r . Consequently, the supplier’s setup cost Purchasing cost Disposing cost Material processing
cost
per unit time is nTAs
r
. The manufacturer needs (1−X)D
β
units of virgin raw materials per unit time. Therefore, The number of times that the MRF collects returned
the supplier’s extraction cost is Cs (1−X)D
β . As a result, products per unit time can be calculated by multiplying
the supplier’s profit per unit time TPs = TRs − TCs is the number of these orders per system cycle by the fre-
given by: quency of the system cycle per unit time, i.e. m × nT1 r .
  Consequently, the MRF’s transportation cost is mO M
nTr .
(1 − X)DPmp D2 (l − 1)(1 − X)nTr The MRF’s collection cycle equals the system cycle nTr
TPs = − hs
β 2β 2 lP according to Figure 4. Hence, the number of collection
lOs As Cs (1 − X)D cycles per unit time is nT1 r and the MRF’s setup cost
− − − . (4) AM
per unit time is nT . The amounts of returned products
nTr nTr β r
(Ru ) that are separated by the MRF as non-recyclable
Here, the first term represents the supplier’s sales rev- items and recyclable items per unit time are (1 − α)Ru
enue, the second term is the holding cost of virgin materi- and αRu , respectively. Therefore, the MRF pays per unit
als, the third term is the fixed transportation cost related time the cost of Ru PM for purchasing returned products,
to the manufacturer’s order of any size, the fourth term the cost of (1 − α)Ru Cd for disposing the non-recyclable
is the fixed setup cost, and the fifth term is the extraction items, and the cost of αRu CM for processing the recy-
cost of the virgin raw materials all per unit time. clable items. As a result, the MRF’s profit per unit time
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 11

TPM is given by: sum of these two quantities appear on the right-hand side
of Equation (6). The recycling facility transforms them
hu nαRu [βmR + β(1 − m)αRu
into RTR units of recycled raw materials per system cycle.
+(1 − β)(1 − m)D]Tr
TPM = αRu PR − The recycling facility’s carries inventory of recycled
2βmR materials and recyclable returned products at holding
mOM AM costs rates of hRr and hu per unit per unit time, respec-
− − − Ru PM − (1 − α)Ru Cd
nTr nTr tively. The calculations for the recycling facility’s average
− αRu CM . (5) inventory levels, which require Equation (6), are pro-
vided in Appendix B.4. The recycling facility’s total cost
The first term is the MRF’s sales revenue, the second per unit time, TCR , consisting of holding, order process-
term is the holding cost of the recyclable returned prod- ing, setup, ordering, recycling, and purchasing costs is
ucts, the third term is the fixed transportation cost related given by:
to the recycling facility’s order of any size, the fourth term Holding cost
←−−2−−−−−−−−−−−−−−→
is the fixed setup cost, the fifth term is the purchasing cost hRr D XnTr [−R+kR−(−2+k)PX]
2β 2 kRP
of the used products returned by the consumers, the sixth TCR = hu n(D−Dβ+αβRu )2 Tr
term is the disposal cost of the non-recyclable returned + 2β 2 mR
products, and the seventh term is the processing cost of kOR AR mARu
the recyclable returned products all per unit time. + + +
nTr nTr nTr
←→ ←→ ←−−→
Transportation Setup Ordering
cost cost cost
2.5. Recycling facility’s profit  
XDCR (1 − β)D
The MRF’s recyclable returned products and the manu- + + PRm + αRu PR .
β β
facturing waste are delivered to the recycling facility as ←−−→ ←−−−−−−−−−−−−−−−−→
Recycling Purchasing
post-consumer materials and post-industrial materials, cost cost

respectively. A fixed ordering cost ARu is charged when- The number of manufacturer’s orders for recycled mate-
ever the recycling facility orders a batch. The recycling rials per unit time is nTk r . Consequently, the recy-
facility transforms them into recycled materials at a finite
cling facility’s transportation cost is kO R
nTr . According to
rate of R units per unit time. The recycling cost is CR per
Figure 4, the recycling facility’s cycle length is the same as
unit and the fixed setup cost per lot is AR . The recycling
the system cycle length nTr . Hence, the number of cycle
facility delivers each recycled lot to the manufacturer in k AR
is nT1 r and the setup cost is nT . As motivated before, the
equal-sized batches incurring a fixed transportation cost r
number of orders for recyclable returned products from
OR per shipment. The recycling facility meets the manu- m
the MRF is nT per unit time. Hence, the recycling facil-
facturer’s demand for ( XD β ) units of recycled raw material
r

per unit time at a price of Pmr per unit. Therefore, the ity’s ordering cost per unit time becomes mA Ru
nTr . As pre-
recycling facility’s sales revenue per unit time is given by viously explained, upon manufacturer’s request the recy-
TRR = XDP mr cling facility delivers XD
β unit of recycled raw materials
β . In the following, we obtain a mathemati- XDCR
cal formulation for the recycling facility’s total cost. For per unit time. The corresponding recycling cost is β .
that purpose, we need to determine the expression for the The recycling facility buys (1−β)D
β units of scarp from the
length of the recycling time TR within the system cycle. manufacturer and αRu recyclable returned items from
The recycling facility recycles the manufacturing waste the
 MRF per unit time.  The related purchasing cost is
as well as the recyclable returned products collected by (1−β)D
β PRm + αRu PR . As a result, the recycling facil-
the MRF during the system cycle. Hence, we need the
following equality to hold: ity’s profit per unit time TPR = TRR − TCR can be writ-
ten as:
Using Eq. (1)
RTR = (1 − β)PTp + αRu nTr ⇒ XDPmr hRr D2 XnTr [−R + kR − (−2 + k)PX]
TPR = −
(1 − β)PTp + αRu nTr β 2β 2 kRP
TR =
R hu n(D − Dβ + αβRu )2 Tr
  −
(1 − β)D + αβRu 2β 2 mR
= nTr . (6)
βR kOR AR mARu XDCR
− − − −
In fact, the manufacturer and the MRF deliver (1 − β) nTr nTr nTr β
 
PTp and αRu nTr units of pre-consumer and post- (1 − β)D
− PRm + αRu PR . (7)
consumer materials per system cycle, respectively. The β
12 M. PARSA ET AL.

Here, the first term represents the recycling facility’s


sales revenue, the second term is the holding cost of the
recycled materials, the third term is the holding cost of
the recyclable returned products, the fourth term is the
fixed transportation cost related to the manufacturer’s
order of any size for the recycled materials, the fifth term
is the fixed setup cost, the sixth term is the ordering cost
for the recyclable returned products, the seventh term is
the recycling cost, and the eighth term is the summation
of purchasing costs of the manufacturer’s scrap and the
MRF’s recyclable returned product all per unit time.
Now, all players’ individual profit functions have been
mathematically formulated. Hence, the chain-wide total
profit per unit time, TP, can simply be written as the sum
of the profits per unit time gained by the manufacturer,
the retailer, the supplier, the MRF, and the recycling facil-
ity formulated in Equations (2)-(5) and (7), respectively.
The optimisation problem is formulated as the following
mixed integer non-linear programming (MINLP) model:

MaxTP = TPm + TPr + TPs + TPM + TPR , (8)

S.t. :

k ≤ k0 , l ≤ l0 , m ≤ m0 , and n ≤ n0 , (9)
+
k, l, m, n ∈ Z , (10)
Tr ∈ R+ . (11)

The model contains five decision variables, four of


which are positive integers (k, l, m, and n), while the
last variable is a positive real number (Tr ). According
to Figure 4, time interval between shipments and batch
sizes for all parties are functions of the decision variables
k, l, m, n and Tr . Therefore, the objective is to find the
retailer’s optimal cycle time (Tr∗ ) and the optimal num-
ber of shipments (k∗ , l∗ , m∗ , n∗ ) such that total profit of
the whole supply chain is maximised. The optimal integer
number of shipments cannot be infinite due to existence
of the fixed costs. To broaden the applicability of the pro-
posed model in real-world inventory control systems, we
impose upper bounds to the integer decision variables
k, l, m and n. These bounds are k0 , l0 , m0 , and n0 , respec-
tively. Finally, we have the domain constraints in (10) Figure 5. Flowchart of the B&B algorithm.
and (11).

to nonlinear B&B procedure is Dakin (1965). The pro-


3. Branch-and-Bound (B&B) method
cedure is used to solve convex (Gupta and Ravindran
B&B approach dates back to Land and Doig (1960), where 1985; Kronqvist et al. 2019) and nonconvex (Androulakis
it is used to solve Mixed Integer Linear Programming 2009; Burer and Letchford 2012). The B&B algorithms
(MILP) models. To our knowledge, the first reference are used in many applications including vehicle routing
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 13

 
(Toth and Vigo 2002), data analysis (Brusco and Stahl D2 (1 − X)nTr#
2006), multi-objective decision making (Ghasemi Sag- − hmp
2β 2 lP
hand, Charkhgard, and Kwon 2019; Stidsen, Andersen,  
Tr# D2 (βP − D)nTr# D Tr# D
and Dammann 2014), lot-sizing (de Oliveira and San- − hm + −
tos 2018; Erenguc and Aksoy 1990; Mirmohammadi, βP 2βP 2
Shadrokh, and Kianfar 2009), water distribution net- Om Am kAmr lAmp
− − #− −
works (Costa et al. 2016; Costa, de Medeiros, and Pes- Tr# nTr nTr# nTr#
soa 2001), process systems engineering (Lee and Gross-  
Cp D (1 − X)DPmp + XDPmr
mann 2001; Smith and Pantelides 1997), and scheduling − − .
(Pinedo 2012). β β
(13)
Specifically, the B&B algorithm is a widely-used
methodology for finding exact solutions to noncon-
vex global optimisation problems (Lawler and Wood Here, Equation (13) formulates the manufacturer’s profit
1966; Moore 1991) and nonlinear small-scale problems per unit time in the sequential structure. TPm (k, l, n)
(Marendet et al. 2020). Our problem has both features; is strictly concave with respect to k and l. We refer to
the definiteness of the Hessian matrix of the objective Appendix C.1 for the proof of this result. As a result, we
function depends on the values of the decision variables obtain:
as well as the parameters. Hence, we cannot tell whether 
the problem is convex or concave MINLP. It is nonlinear ∂TPm DnTr# hmr X
= 0 ⇒ krx = , (14)
and has a few and fixed numbers of decision variables. For ∂k β 2Amr P
this reason, we choose this method to find the exact solu- 
tion to the problem. Figure 5 presents the flowchart of ∂TPm DnTr# hmp (1 − X)
= 0 ⇒ lrx = , (15)
this algorithm. We discuss the main features of the B&B ∂l β 2Amp P
algorithm in the subsequent sections.
where the optimal value of relaxed k and l for TPm ,
denoted by krx and lrx are obtained in Equations (14)
3.1. Initialisation and (15) by solving the equations ∂TP ∂k = 0 for k and
m

∂TPm
We use the solution of sequential optimisation problem, ∂l = 0 for l, respectively. kandl are positive integers
i.e. the decentralised solution, as the first incumbent. The and TPm (k, l, n) is strictly concave in k and l. Therefore,
incumbent is the best available feasible solution at a given Algorithm SE presented in Appendix D.1 can be used
time. In the sequential structure, the solution obtained to obtain the optimal k, l, and n in the sequential struc-
for one location is forced to the next one in the hierar- ture. The optimal values are denoted by k# , l# , and n# ,
chy of decisions. We investigate the case where we first respectively.
solve for the retailer’s cycle length Tr by maximising its By substituting k# , n# , and Tr# in Equation (7), we
own profit. Then, assuming this cycle length is a given, obtain the following expression for the recycling facility’s
we optimise the manufacturer’s decision by optimising its profit function:
own profit. Then, we do the same for the recycling facility.
Next, we explain these steps in more detail. hRr D2 Xn# Tr# [−R + k# R
Since ∂∂TTP2 r = − 2A
2
r
< 0, TPr is strictly concave in Tr . XDPmr −(−2 + k# )PX]
r Tr3 TPR (m) = −
Accordingly, we obtain: β 2β 2 k# RP
 hu n# (D − Dβ + αβRu )2 Tr# kOR
∂TPr Dhr Ar 2Ar − 2
− # #
=− + 2 = 0 ⇒ Tr# = . (12) 2β mR n Tr
∂Tr 2 Tr Dhr AR mARu
− − # #
n# Tr# n Tr
Tr# is the optimal cycle length if we maximise the retailer’s  
profit only. Next, we consider the manufacturer’s profit. XDCR (1 − β)D
− − PRm + αRu PR .
By substituting Tr# in Equation (2), we obtain: β β
  (16)
(1 − β)D
TPm (k, l, n) = Pr D + PRm
β Hence, Equation (16) formulates the recycling facility’s
 2 #  profit per unit time in the sequential structure. Since
D XnTr
− hmr ∂ 2 TPR
=−
hRu n# (D−Dβ+αβRu )2 Tr#
< 0, TPR (m) is strictly
2β 2 kP ∂m2 β 2 m3 R
14 M. PARSA ET AL.

concave in m. Then we get: or equal to the current lower bound (UB ≤ LB). In the
 first case, it becomes the new incumbent and the LB is
(D − Dβ + αβRu )n# Tr# hRu updated when it is better than the current incumbent.
mrx = , (17)
β 2ARu R
where the optimal value of relaxed m in the sequential 3.5. Termination
structure, denoted by mrx is obtained in Equation (17)
When the first branching variable in the B&B tree, i.e.k,
by solving the equation ∂TP∂m = 0. Since m is a posi-
R
becomes greater than k0 , there is no remaining node
tive integer and TPR (m) is strictly concave in m, the
and the last incumbent is the global optimal solution.
optimal value of m in the sequential structure, denoted
Otherwise (k ≤ k0 ), we perform the next iteration.
by m# is derived at m1 = max[1, min{mrx , m0 }] and/or
m2 = min{mrx , m0 }. Finally, we have:
4. Heuristic procedure
TP(k# , l# , m# , n# , Tr# )
The definiteness of Hessian matrix of the objective func-
= TPr (Tr# ) + TPm (k# , l# , n# ) + TPR (m# )
tion TP depends on the values of the parameters and
+ TPs + TPM . (18) decision variables. Therefore, we cannot call whether
Equation (18) calculates the total profit under decentr- our problem is convex/concave MINLP. However, TP is
alised decision making. Now, we set the first incumbent always strictly concave with respect to Tr for given val-
at k# , l# , m# , n# , and Tr# and the initial lower bound on ues of k, l, m, and n (See Appendix C.2 for the proof).
the total profit at TP(k# , l# , m# , n# , Tr# ). Then, we use Consequently, we obtain:
branching, bounding, pruning, and termination steps per 
2klmRP
iteration. These steps are explained next. TrH = Aβ , (19)
nD(B + C)

3.2. Branching where



We select the next variable in the ordering—k, l, m, and
Am + AM + AR + As + n(Ar + Om )
n— as a branching variable. We choose one of the unfath- A= ,
+m(ARu + OM ) + k(Amr + OR ) + l(Amp + Os )
omed sub-problems (nodes) to create new sub-problems
by fixing the branching variable to its possible integer
values according to the depth-first search strategy. B = β 2 klmRP[hr + hm (n − 1) + nhu ]
+ βklR[−Dhm m(n − 2) − hu mnPR(1 − X)
3.3. Bounding
− nDPXhu (m − 1)],
The total profit for a given incumbent provides a lower
bound (LB) on the problem. In addition, we have a sub- C = nD{kmRhmp (1 − X) + kmRhs (−1 + l + X − lX)
problem per node. So, we can calculate an upper bound
(UB) for each node by solving the relaxed sub-problem + lX[(hmr + hRr (k − 1))mR − mPXhRr (k − 2)
in which the Constraints (9) and (10) are removed. + hkP(−1 + m + 2X − mX)]}.
The relaxed sub-problem reads as follows: min TP; s.t. :
k, l, m, n ≥ 1 and Tr ≥ 0. When the variable(s) approach Hence, the optimal Tr for given values of k, l, m, and
positive infinity, the limit of TP either is negative infin- n, which is defined as TrH , occurs at the only posi-
ity or does not exist. Therefore, we need to investi- tive root of ∂TP
∂Tr = 0 and this value is calculated using
gate all possible combinations in which no/some inte- Equation (19). In mixed integer programming (MIP)
ger variables are set to 1 and other variables meet the formulations, the technique of fixing a subset of inte-
second-order necessary and sufficient optimality condi- ger variables to create sub-problems which are easier to
tions (Bazaraa, Sherali, and Shetty 2013). Thus, we exe- solve has proven to result in good heuristics. Some of
cute Algorithm UB provided in Appendix D.2 to obtain these heuristics are the distance induced neighbourhood
the upper bound for each node. search (DINS) (Ghosh 2007), relaxation induced neigh-
bourhood search (RINS) (Danna, Rothberg, and Le Pape
2005), mutation, and crossover (Berthold 2006; Roth-
3.4. Pruning
berg 2007). This technique can be generalised to MINLP
A node is fathomed when the upper bound from directly (Berthold et al. 2011) and it is the main moti-
Algorithm UB is 1) a feasible solution or 2) less than vation behind our heuristic. We consider the following
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 15

general MINLP model adopted from our main model


formulated in (8)–(11): k0 = 30 CM = 10$/unit PM = 10$/unit Am = 20000$
l0 = 30 Cp = 95$/unit Pc = 250$/unit As = 450$
m0 = 30 CR = 13$/unit PR = 40$/unit AM = 1000$
Min Z = f (x, Tr ), x = (x1 , x2 , . . . , xq ) n0 = 30 Cs = 45$/unit Pmp = 77$ ARu = 100$
S.t. : α = 0.6 hmp = 0.6$/unit/year Pmr = 66$ OM = 25$
Li ≤ xi ≤ Ui , i = 1, 2, . . . , q (20) X = 0.55 hmr = 0.5$/unit/year Pr = 240$ Os = 300$
β = 0.8 hm = 1.50$/unit/year PRm = 40$ Om = 500$
Tr ∈ R+ , D = 500 units/year hr = 4.5$/unit/year Ar = 750$ OR = 100$
xj ∈ Z+ . P = 1000 units/year hRr = 0.30$/unit/year AR = 10000$
R = 800 units/year hu = 0.20$/unit/year Amr = 240$
Cd = 5$/unit hs = 0.5$/unit/year Amp = 250$
In this formulation, we have f : (x, Tr ) → R, x1 , x2 ,
. . . , xq are integer variables, q is the number of inte-
ger variables, and Tr is the only continuous variable. We solution tabulated in Table 2 in 984.6 and 49 s, respec-
assume without loss of generality that both Li and Ui ∈ tively. In order to evaluate the heuristic’s performance
Z+ . We also require f to be strictly concave with respect in terms of quality and speed, we need to solve more
to Tr and that is always true for our problem. problems that are provided in the next Section 5.2.
The algorithm starts by finding an optimal solution Also, we find that the integrated decision-making pro-
(x̄, T̄r ) to the problem obtained by relaxing the integral- cess results in higher total profit compared to the sequen-
ity constraint from xj ∈ Z+ to xi ∈ R+ . If x̄ is integral tial one as expected but it might result in lower profits of
for all x̄1 , x̄2 , . . . , x̄q , (x̄, T̄r ) is the global optimal solu- some members including the manufacturer, the retailer,
tion to the MINLP. Otherwise, the algorithm continues and the MRF in this example. When the CLSC is inte-
with a systematic search to force some integer variables grated vertically in the sense that it is jointly or partially
to take on integral values. Since optima can occur at the owned by organisation(s), this is not an issue. Other-
boundary of the domain (Li and Ui ), the algorithm gen- wise, lower profits for some members might cause serious
erates all possible sub-MINLPs in which different subsets problems, harming the overall strength and sustainabil-
of integer variables are fixed to their lower (Lj ) and upper ity of the CLSC. Motivated by these concerns, we advise
bounds (Uj ). Then, it finds the optimal solution (x̂, T̂r ) practitioners to coordinate their decentralised CLSC and
to the continuous relaxation of each sub-MINLP. The design mechanisms to fairly divide the resulting (poten-
algorithm continues by exploring all possible combina- tially higher) profit of integrated process among the
tions in which the optimal solution is rounded up/down involved parties.
to nearest integer respectively indicated by x̂i or x̂i . After
fixing all integer variables, we obtain optimal Tr by
using Equation (19). The pseudo-code of the heuristic 5.2. Additional results
algorithm is provided in Appendix D.3. We assess the performance of the heuristic compared to
There is no guarantee that the heuristic finds the opti- the B&B algorithm by solving several instances. Since
mal solution. It aims to find feasible solutions of good the time complexity of the B&B algorithm depends on
quality for our optimisation problem within a reason- k0 , l0 , m0 , and n0 , we create ten groups of instances where
ably short amount of time. We evaluate the performance all these parameters are set to 10, 20, 30, 40 . . . , 100. The
of our heuristic and report the results in the subsequent number of randomly generated instances in each group
sections. is 100. Therefore, in total 1,000 instances are generated,
using the following data set:
5. Computational results
5.1. Numerical example
PR , PRm ∼ Uniform[0.10Pc , 0.20Pc ]
In this section, we consider an example and provide the D ∼ Uniform[10, 10000]
α, β ∼ Uniform[0, 1] hu ∼ Uniform[0.01PR , 0.10PR ]
details of our solution. These details can be useful for X ∼ Uniform[1 − β, αβ + 1 − β] Pmp , Pmr ∼ Uniform[0.20Pc , 0.40Pc ]
  hmp ∼ Uniform[0.01Pmp , 0.10Pmp ]
readers who aim at implementing our exact and heuristic XD 10XD
R ∼ Uniform , hmr , hRr ∼ Uniform[0.01Pmr , 0.10Pmr ]
algorithms. The data is provided below. Both algorithms  β β Ps ∼ Uniform[0.15Pc , 0.30Pc ]
D R 1 hs ∼ Uniform[0.01Ps , 0.10Ps ]
are coded in MATLAB 2018 and run on a laptop com- P ∼ Uniform ,
β X Pr ∼ Uniform[0.70Pc , 0.90Pc ]
puter with Intel® CoreTM i5-8250U CPU@1.60GHz, RAM Ar , AR , Amr , Amp , Am , As , AM , ARu ∼ hr , hm ∼ Uniform[0.01Pr , 0.10Pr ]
Uniform[100, 10000] CM , Cp , CR ∼ Uniform[0.10Pc , 0.20Pc ]
8.00 GB, 64-bit operating system, and Windows 10. OM , Os , Om , OR ∼ Uniform[10, 1000] Cd ∼ Uniform[0.01Pc , 0.05Pc ]
We refer to Appendix E.1 for the calculations. Both Pc ∼ Uniform[1, 10000]
PM ∼ Uniform[0.01Pc , 0.05Pc ]
B&B and heuristic algorithms could obtain the optimal
16 M. PARSA ET AL.

Table 2. Optimal solution of the numerical example.


Structure k l m n Tr TPm TPr TPs TPM TPR TP
Sequential 6 5 5 15 0.8165 16778.594 3162.883 8410.209 1919.826 3378.249 33649.760
Integrated 5 2 5 12 1.0228 16747.045 3116.058 8644.762 1919.644 3383.009 33810.517

Table 3. Performance of the algorithms.


Average run time of the Average run time of the
Group k0 , l0 , m0 , and n0 No. of Non-Opt.% Avg. Dev.% Max. Dev.% heuristic (seconds) B&B (seconds)
1 10 3 3.0 5.1 44.8 281.3
2 20 2 4.9 6.2 32.5 337.0
3 30 3 2.2 4.7 41.0 939.5
4 40 4 4.5 6 49.6 1157.1
5 50 3 3.1 9.3 65.1 1880.2
6 60 5 2.6 4.1 40.9 3017.7
7 70 4 4.7 10.9 57.4 3143.1
8 80 6 6.1 8.5 34.0 3568.8
9 90 5 4.0 7.2 52.4 3811.3
10 100 5 5.2 10.4 40.5 4775.4

We do not set a limit on the run time and conse- concerning many business scenarios in today’s competi-
quently, the B&B algorithm obtains the optimal solution tive market.
in all instances. We define the following performance
measures:
6. Sensitivity analysis
• Percentage of instances in which the heuristic doesn’t Studying the effect of manufacturing scrap fraction
find the optimal solution, denoted by No. of Non-Opt. (1 − β) is among our research questions. In addition,
%. the recycle content level (X) is a key parameter for mix-
• Average and maximum percentage deviations from ing virgin and recycled materials in production of recy-
optimal objective function value when the heuristic cled content products. There are usually recommended
doesn’t find the optimal solution. These measures are and/or regulated ranges for the recycled content level
denoted by Avg. Dev. % and Max. Dev. %, respectively. (X). For example, (EPA 2007a) recommends 20%–100%
• Average run time for both algorithms. recycled content for bathroom tissue, 5%–40% for brown
papers, 25%–50% for corrugated containers, 20%–25%
In Table 3, we present the computational results for for fiberglass insulation, 25%–100% for polyester car-
ten groups of instances. Table 3 shows that the average pet, and 80%–100% for structural fiberboard. Sustainable
computational time for the heuristic is between 32.5 and industries need to know what recycled content level (X)
65.1 s which is significantly shorter than that for the is cost-effective. Therefore, this section aims to study the
B&B algorithm. The average computational time for the effect of the scrap fraction of production (1 − β) and
B&B algorithm increases from 281.3–4,775.4 s when val- the recycled content level (X). We consider the detailed
ues of k0 , l0 , m0 , and n0 increase from 10 to 100 since example in Section 5.1 and perform sensitivity analysis
this algorithm runs in O(k0 l0 m0 n0 ). The heuristic expe- on these two parameters.
riences only small changes in runtime when we con-
sider the ten groups because it has a time complexity of
O(1). The average runtime of the heuristic is 45.8 s. In 6.1. Effect of the scrap fraction
4% of the total instances (40 out of 1,000), it does not To study the effect of β, we obtain the optimal solu-
obtain the optimal solution with an average and max- tions and the corresponding objective function values for
imum percentage deviation from optimality of 4% and selected values of β ranging from 50% to 100% with an
10.9%, respectively. increment of 1%. We summarise our findings below.
Our results show that the proposed heuristic algo- Figures 6 and 7 show the optimal solutions when β
rithm performs quite well since it has obtained the global increases for the integrated and decentralised systems,
optimal solution for 96% instances and near-optimal respectively. Both figures indicate that β can significantly
solutions for other ones within a reasonably short amount affect the optimal decision variables except Tr in the
of time. In fact, the heuristic procedure provides a quick- sequential structure since according to Equation (17), Tr
and-reliable tool for managers to make instant decisions does not depend on β directly.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 17

Figure 6. Behaviour of the optimal solutions in the integrated


structure with respect to β. Figure 8. Behaviour of the profits in the integrated structure with
respect to β.

Figure 7. Behaviour of the optimal solutions in the sequential


structure with respect to β.

According to Figures 6 and 7, increasing β decreases


Figure 9. Comparison of the optimal values of Tr and n in the
the optimal number of shipments to each party per sys- integrated and sequential structures.
tem cycle in both sequential and integrated systems.
When β increases, the perfect production rate (βP)
and the manufacturer’s inventory level increase. In this
case, it is beneficial for the manufacturer to push more
inventory downstream, i.e. to the retailer, to avoid high increase of β (decrease of 1 − β) reduces manufactur-
holding costs. This implies an increase in the retailer’s ing scrap. As a result, the manufacturer consumes less
batch size. Since the demand that needs to be met virgin and recycled raw materials. Therefore, sales and
remains the same, the optimal number of shipments consequently the profits of the supplier and the recycling
to the retailer (n∗ and n# ) decreases. In addition, the facility decrease. In addition, a very slight fluctuation is
manufacturer needs less raw material when β increases. observed in the retailer’s profit because it depends only
Hence, the optimal number of shipments to the manu- on Tr∗ and changes in Tr∗ are not enough to affect it sig-
facturer from the supplier (l∗ and l# ) and the recycling nificantly. The trends are also valid for the sequential
facility (k∗ and k# ) and consequently the optimal num- system.
ber of shipments to the recycling facility from the MRF According to Figure 9, the optimal Tr in the integrated
(m∗ and m# ) decrease. system (Tr∗ ) is more than that in the sequential sys-
Figure 8 shows the behaviour of profits of each party tem (Tr# ). This shows that the retailer places orders with
and the whole supply chain in the integrated system as smaller batch sizes in the sequential system to control its
a function of β. According to Figure 8, the profits of holding cost. As a result, the optimal number of ship-
the manufacturer, the MRF, and the whole supply chain ments to the retailer per system cycle in the sequential
increase as β increases. However, the profits of the sup- system (n# ) is higher than that in the integrated system
plier and the recycling decrease. This is because the (n∗ ).
18 M. PARSA ET AL.

the forward supply chain (Supplier → Manufacturer →


Retailer → Consumers). In this case, both green groups
and supply chain parties benefit since high recycled con-
tent level implies an environmental solution, which is
also the best solution in terms of chain-wide total profit.
In contrast, the optimality of Xmin might results in a
disagreement and conflicts between green and environ-
mentalist groups and decision-makers with economic
concerns. In this case, the decision maker might try to
resolve the conflict. We explain how to achieve this with
an example.
We consider the previous detailed numerical example.
Figures 12–14 show surfaces of TP with respect to X and
Tr for different values of k, l, m, and n. Figure 12 is for the
Figure 10. Behaviour of the total profit (TP) with respect to X for case with k = 5, l = 2, m = 5, and n = 12. The surface is
different values of CR . strictly-concave in the Tr -direction and strictly-convex
in the X-direction. In addition, it has a saddle point at
6.2. Effect of the recycled content level (TR = 1.062, X = −0.470, TP = 32834.438). This figure
indicates that the optimal recycled level is always Xmax
To study the effect of the recycled content level, we calcu-
for this example. Figure 13 is for the case with k =
late the optimal total profit of the integrated supply chain
2, l = 2, m = 2 and n = 12. The surface is strictly-
for selected values of X ranging from 20% to 60% with an
concave in the Tr -direction. TP is always linear with
increment of 10% and multiple unit recycling cost values;
respect to X. In addition, it has a saddle point at
CR ∈ {12, 13, 14, 15, 16, 17, 18}. Let [Xmin , Xmax ] denote
(Tr = 2.735, X = −7.700, TP = 28708.772). This figure
an allowed range for X.
indicates that the optimal recycled level is always Xmax
Figure 10 shows that the optimal recycled content
since in the red surface of feasible region (i.e. Tr ≥ 0&0 ≤
level equals either the upper limit Xmax or lower limit
X ≤ 1), TP always increases when X increases. Hence, for
Xmin . Based on this observation, we develop a hypoth-
both cases (Figures 12 and 13), in addition to the supply
esis on the optimality of a so-called bang-bang policy for
chain players, the green and environmentalist groups are
the recycled content level. We investigate this hypothesis
satisfied.
and obtain the following theoretical results. We refer to
Figure 14 is for the case with k = 1, l = 2, m = 1, and
Appendix C.3 for the proof.
n = 12. The surface is strictly-concave in both direction
Theorem 6.1: For k ≥ 2&m ≥ 2, the bang-bang policy is and has a global maximum at (Tr = 0.969, X = 0.224,
the optimal policy for the recycled content level, i.e. the TP = 33272.258) where |H2 | = 2.33 × 107 > 0. If Xmin
optimal recycled content level is either Xmin or Xmax . ≤ X̂ = 22.4% ≤ Xmax , the optimal recycled level is
X̂ = 22.4%, otherwise it occurs at either Xmin or
This result leads to the following corollaries: Xmax . For example, when we consider Xmin = 0.25
and Xmax = 0.95, based on Corollary 6.1, the optimal
Corollary 6.1: Let (T̂r , X̂) denote the stationary points recycled level is 0.25 since X̂ = 22.4% ≤ Xmin ≤ Xmax
at which the gradient of function TP(Tr , X) becomes zero,
∂ 2 TP
and TP(Xmin = 0.25) = 33271.3 > TP(Xmax = 0.95) =
∂ 2 TP 32524.7. In this case, green and environmentalist groups
∂Tr2 ∂Tr ∂X
i.e. ∇TP(T̂r , X̂) = 0. H2 = ∂ 2 TP ∂ 2 TP
is the Hes- might oppose this profit-maximising decision.
∂X∂Tr ∂X 2 To deal with this problem, we perform a break-even
sian matrix of TP(Tr , X), and |H2 |
is the determinant of
analysis on the unit recycling cost CR (Figure 15). It
matrix H2 . Figure 11 shows the resulting decision tree for
helps to determine the conditions under which Xmax
the optimal recycled content level.
is more profitable than Xmin . Figure 15 shows that the
Corollary 6.2: TP is strictly-convex in X for (k > 2& break-even point is CR = 11.30, where the two lines
m ≥ 2) or (k = 2&m > 2), linear in X for (k = m = 2), TP(Xmax = 0.95 and TP(Xmin = 0.25 cross. It also indi-
and strictly-concave in X for other values of k and m. cates that for CR < 11.30$ we have TP(Xmax = 0.95) >
TP(Xmin = 0.25) and consequently, for any CR below
When the optimal recycled content level is Xmax , 11.30, the optimal recycled content level would be Xmax .
the reverse supply chain (Consumers → MRF → Recy- Therefore, one way of resolving the conflict between
cling Facility → Manufacturer) is more profitable than economic and environmental solutions is to reduce CR
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 19

Figure 11. Decision tree for the optimal recycled content level.

Figure 12. Graph of TP with respect to Tr and X for k = 5, l = 2, m = 5, and n = 12.

from 13$ to any value below 11.30$. This reduction can 7. Extension to the assembly systems
be enabled by offering tax cuts, subsidies, no or low-
The model developed in Section 2 is relevant to the pro-
interest loans, green grants, and accelerated investment
cess industries of a single material type like paper, plastic,
incentives to the recycling facility.
rubber, and metal. In this section, the model is extended

Figure 13. Graph of TP with respect to Tr and X for k = 2, l = 2, m = 2 and n = 12.


20 M. PARSA ET AL.

Figure 14. Graph of TP with respect to Tr and X for k = 1, l = 2, m = 1, and n = 12.

Figure 15. Break-even analysis on CR .

to the assembly systems. Hence, a final product is man- that a fraction β (0 ≤ β ≤ 1) of the manufacturer’s pro-
ufactured by the assembly of multiple components. We duction is perfect
v
and the rest (1 − β) is defective. We
consider the case where some of the components are i=1 λi Xi
define X̄ = v as the average remanufacturing per-
i=1 λi
new and the rest are remanufactured. Examples for such centage. Therefore, the total number of components per
products include electronics, medical devices, and vehi- returned product that are remanufactured can be calcu-
cles. We consider a CLSC consisting of a manufacturer, v v
a retailer, multiple suppliers, an MRF, and a remanufac- lated as λi Xi = X̄ λi . Based on these definitions,
i=1 i=1
turer. Figure 16 depicts the structure of the CLSC. we can rewrite the manufacturer’s profit per unit time
We consider a product, a unit of which is assembled Assembly
(TPm ) as follows:
from ν different components (v ∈ N). λi units of com-
ponent type i (i = 1, 2, 3, 4, . . . , ν; λi ∈ N) are needed for
each unit of product. Each component type is obtained
from a different supplier. (1 − Xi ) percentage of com-  
Assembly (1 − β)D
ponent type i is new and is provided by supplier i. TPm = Pr D + PRm
β
The remanufacturer provides the rest. The MRF col-  
lects and disassembles α percentage of returned products v
D λi X̄nTr
2
− hmri
and ships them to the remanufacturer. We also consider i=1
2kP
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 21

Figure 16. CLSC of the assembly industry.

ν
  unit time. As a result, the total profit per unit time of all
D2 λi (1 − X̄)nTr
− hmpi suppliers involved in the CLSC is given by:
i=1
2li P
 
v
v
Tr D2 (P − D)nTr D Tr D Assembly
TPsi = λi (1 − X̄)DPmpi
− hm + −
P 2P 2 i=1 i=1
Om Am kAmr lAmp
v  
− − − − D2 (li − 1)λi (1 − X̄)nTr
Tr nTr nTr nTr − hsi
v i=1
2li P

− Cp D − λi (1 − X̄)DPmpi
v
li Os v
A si
i=1
− i


i=1
nTr i=1
nTr

v
+ λi X̄DPmri . (21)
v

i=1
− Csi λi (1 − X̄)D. (23)
i=1
Here, hmri and hmpi are the manufacturer’s unit holding Using the procedure, which resulted in Equation (7), we
costs for remanufactured and new components of type obtain the following expression for the remanufacturer’s
i per unit time, respectively. Pmri and Pmpi represent the profit per unit time:
manufacturer’s unit purchase price for remanufactured
and new components of type i from the remanufacturer Assembly

v

and the supplier i, respectively. Accordingly, the supplier TPR = λi X̄DPmri


i=1
i’s profit per unit time can be written as:
hRr D2 λi X̄nTr
Assembly
TPsi = λi (1 − Xi )DPmpi
v
[−R + kR − (−2 + k)PX̄]

 2  2kRP
D (li − 1)λi (1 − Xi )nTr i=1
− hsi
2li P hu n(αRu )2 Tr kOR AR mARu
− − − −
li Osi As 2mR nTr nTr nTr
− − i − Csi λi (1 − Xi )D. (22)  
nTr nTr XDCR (1 − β)D
− − PRm + αRu PR .
β β
Here, li is the number of shipments from supplier i to (24)
the manufacturer per system cycle (decision variable), Osi
is the fixed transportation, Asi is the fixed setup cost, hsi In Equation (24), the first term represents the reman-
is the unit holding cost for the type i component per unit ufacturer’s sales revenue, the second term is the holding
time, Csi is the supplier i’s procurement cost per unit of cost of the remanufactured components, the third term
type i component, and TPsi is the supplier i’s profit per is the holding cost of the recoverable returned products,
22 M. PARSA ET AL.

the fourth term is the fixed transportation cost related to According to our derivations, the optimal level of recy-
the manufacturer’s order of any size for the remanufac- cled content is often either at its maximum or minimum
tured components, the fifth term is the fixed setup cost, allowable level. When the optimal recycled content is at
the sixth term is the ordering cost for the recoverable its minimum allowable level, the reverse supply chain
returned products, the seventh term is the remanufac- (Consumers → MRF → Recycling Facility → Manufac-
turing cost, and the eighth term is the summation of turer) is less profitable than the forward supply chain
purchasing costs of the manufacturer’s defective items (Supplier → Manufacturer → Retailer → Consumers)
and the MRF’s recoverable returned products all per unit causing conflicts between environmentalist groups and
time. decision-makers with economic concerns. Hence, we
The expressions for profits per unit time of the retailer advise the managers to deal with this challenge by pro-
and the MRF are exactly the same as in Equations (3) and viding financial supports to the members involved in the
(5), respectively. Hence, the optimisation problem for the reverse supply chain.
assembly system becomes the following MINLP: Our findings regarding the optimal recycled con-
tent level could explain why EU members recognise
Assembly

v
Assembly the importance of the extended producer responsibility
MaxTPAssembly = TPm + TPr + TPsi
(EPR) policy that transfers the financial burden of man-
i=1
aging post-consumer waste inventory to the members
Assembly
+ TPM + TPR , (25) of the forward supply chains. For example, French pro-
ducers of graphic papers such as Arjowiggins Graphic
S.t. : Company are required to compensate costs for the col-
lection and treatment of the post-consumer paper waste
k ≤ k0 , li ≤ l0 , m ≤ m0 , and n ≤ n0 , i = 1, 2, . . . , ν subjected to the French domestic law with article L.541-
10-1 of the Environment Code. Each graphic paper man-
ufacturer should pay an ‘eco-contribution’ fee based on
k, li , m, n ∈ Z+ , i = 1, 2, . . . , ν
the amount of distributed paper in the country to Éco-
folio, an organisation that is responsible for managing
Tr ∈ R+ .
reverse supply chains. This policy helps the reverse sup-
The objective is to maximise the chain-wide total ply chain to gain more profit than the forward one.
profit per unit time (TPAssembly ), which is the sum of the Therefore, as suggested by our findings, involved play-
profits per unit time of the manufacturer, the retailer, the ers can choose the maximum allowable level of recycled
suppliers, the MRF, and the remanufacturer formulated content and achieve higher profits and environmental
in Equations (21), (3), (23), (5), and (24), respectively. benefits.
Next, we define Xi,min and Xi,max as the minimum and Another example is provided by German’s large-scale
maximum values of Xi , respectively. These parameters EPR closed-loop chain in the framework of ‘Green Dot
can be selected according to the product properties. With Programme’ established by the Duales System Deutsch-
Theorem 7.1, we extend the previous analytical result in land AG (DSD) under the low of ‘Ordinance on the
Theorem 6.1 to the assembly systems. The proof is given Avoidance and Recovery of Packaging Waste 1991-
in the Appendix C.4. amended 1998’ as well as European Directive 94/62/EC
on packaging and packaging waste. The Green Dot’s
Theorem 7.1: For k ≥ 2 & m ≥ 2, the bang-bang policy imprint on a packaging material signals that its manu-
is optimal for the recycled content level of each component, facturer has paid an annual license fee to DSD for the
i.e. the optimal value of Xi for component i is either Xi,min collection, sorting, and recycling of used packaging mate-
or Xi,max . rial (paper/cardboard, glass, aluminium, tinplate, plastic,
and composites). The EPR closed-loop production sys-
tem increases the profits of parties involved in the reverse
8. Discussion and managerial implications
supply chain by reducing post-consumer waste inventory
In addition to providing the managers with a decision costs. This leads to a high recycled content level implying
tree to support decisions on the optimal level of recy- an environmental solution, which is also the best solu-
cled content, our study provides multiple managerial tion in terms of chain-wide total profit. Therefore, the
insights that can benefit companies and logistic providers Green Dot system has achieved great success as a Euro-
that produce, distribute, and recollect all recycled con- pean model so that most EU countries have adopted this
tent products. Examples include office products, paper, system for their reverse supply chain of packaging waste
construction products, packaging materials, etc. material.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 23

The above-mentioned evidence related to the optimal Computational results show that the heuristic performs
recycled content level indicates that government author- remarkably well; it obtains the global optimal solution in
ities with policymakers should set a sensible regulatory 96% of the tested instances and near-optimal solutions
framework of objectives, roles and responsibilities shared for remaining ones within a reasonably short amount of
among members, financing mechanisms, performance time. Hence, our heuristic can be used by companies that
targets, approvals, and monitoring for the CLSC accord- need quick-and-reliable solutions to explore the feasibil-
ing to the characteristics of the industry. The formulation ity of available business options in today’s competitive
of the regulatory policies should take the procurement, market.
production, collection, and recycling into account simul- We also perform sensitivity analysis on the scrap frac-
taneously to obtain global optimisation of policies and tion and indicate that a decrease in manufacturing scrap
ensure that they are properly implemented. fraction results in a smaller number of shipments to each
In addition, we find that the total profit from the party per system cycle. It increases the profits of the man-
decentralised decision-making process (sequential struc- ufacturer, the MRF, and the whole supply chain, and
ture), as suggested in our model, is less than the total decreases the profits of the supplier and the recycling
profit that can be obtained from the integrated decision- facility. It has a negligible effect on the retailer’s profit.
making process. Although the letter results in higher Our results can benefit companies in multiple different
overall profit, it can reduce the profits of some mem- ways. The managers can use the decision tree presented in
bers. When the CLSC is integrated vertically in the sense this paper for selecting the optimal level of recycled con-
that it is jointly or partially owned by the organisation(s), tent. Under most conditions, we prove the optimality of
this might not be an issue. Otherwise, it becomes a big bang-bang policy for the recycled content level. We dis-
challenge to sustain long-term cooperation in the CLSC. cuss that a conflict between profit-maximising decision-
The supply chain needs to be coordinated to fairly divide makers and environmentalists might occur if the optimal
the resulting profit among the members. Designing con- recycled content is at the minimum allowable level, which
tracts to coordinate the whole supply chain can encour- means that the forward supply chain attains higher profit
age the parties with lower profits to accept operating than the reverse one accompanied by natural resources
in the CLSCs with integrated decision-making processes depletion at the fastest rate. To cope with this problem,
(Shen, Choi, and Minner 2019). the parties involved in the reverse supply chain should
be financially supported with some economic and/or
administrative instruments under the EPR or non-EPR
9. Conclusions
umbrellas such as virgin materials taxes, advance recy-
Coupling forward and reverse operations in closed-loop cling fees, recycled content standards, recycling subsi-
supply chains (CLSCs) can lead to a higher degree of sus- dies, tax rebates, deposit-refund, take-back systems, and
tainability. Previous studies on inventory management of environmental improvement funds. The results of this
the CLSCs have focused on the products that are either paper encourage policy-makers to pay closer attention to
purely remanufactured or purely new. The emphasis on the effects of governmental regulations on collected and
recycling necessitates research that considers recycled processed inventory flows in the reverse supply chain.
content products, which are made from a mix of virgin They should work together on introducing legislation
and recycled material. We consider a CLSC that consists for developing suitable mechanisms and instruments to
of a manufacturer, a retailer, a supplier of virgin raw mate- support the reverse supply chain according to the char-
rials, an MRF (collector of the used product returned by acteristics of the industry. We also suggest that the profit-
consumers), and a recycling facility that makes recycled sharing contract is a viable option for the managers to
raw materials. Each location delivers inventory to its cus- coordinate the decentralised CLSCs.
tomer location in equal-sized batches. Our objective is to In many realistic settings, the return rate depends on
maximise the total profit of the integrated system by find- the buyback price and acceptable quality level. Incorpo-
ing the retailer’s optimal cycle time and optimal number rating this dependence into our model is a possible future
of shipments from a lot for all parties. Hence, we study a research direction. It provides a useful inspection tool
Joint Economic Lot-sizing Problem (JELP) in the context for the MRF or third-party reverse logistics providers to
of integrated CLSCs. accept/reject each returned product by setting a mini-
We propose a branch-and-bound (B&B) algorithm to mum percentage of useful parts for a recyclable item.
find the exact solution of the problem with an initial Extending our model to the case where all or some par-
lower bound that is the optimal solution of the sequential ties are allowed to backorder is a possible future research
problem, which implies solving for the decision variables direction. In our study, we consider a system with a sin-
one at a time. We also develop a heuristic procedure. gle retailer, while real-world supply chains usually consist
24 M. PARSA ET AL.

of multiple retailers. Extending our analysis to these sys- serves as the Head of Transportation Engineering Department
tems is another interesting research direction, especially at Isfahan University of Technology.
when the retailers compete and have price-dependent Zümbül Atan is an associate professor of
demands. In this paper, we study a deterministic model. supply chain management in the OPAC
It would be interesting, yet very challenging, to study group (Operations, Planning, Accounting,
the same supply chain with stochastic demand on the and Control) of the Department of Indus-
trial Engineering and Innovation Sciences
retailer’s side. Then, optimising the inventory decisions at Eindhoven University of Technology
of all parties and analyzing the effect that the recycled (TU/e). She is the Research Director of the
content level might have on the optima decisions can pro- European Supply Chain Forum’s FMCG
vide new and useful insights. In addition to considering Community. Zümbül’s research is broadly focused on multi-
stochastic demand, extensions that consider uncertain- echelon inventory theory, retail operations and revenue man-
agement. Zümbül has conducted a large number of projects
ties on the supply side are other future research directions
with industry mainly through Master students. Her research
(Atan and Rousseau 2016; Atan and Snyder 2012; Atan is published in leading journals (including Manufacturing &
and Snyder 2014). Service Operations Management, Production and Operations
Management, IISE Transactions, Naval Research Logistics,
European Journal of Operational Research). Zümbül chairs the
Note Dutch Operational Management and Logistics Conference and
she has been in the scientific committees of multiple inter-
1. Since shortages are not allowed, we have: βP ≥ D & R ≥ national conferences. She has earned multiple university- and
XP ⇒ R ≥ XD D
β & β ≤ P ≤ X.
R
departmental-level teaching awards.

Acknowledgments
ORCID
The authors would like to sincerely thank anonymous refer-
ees, the guest editor and the editor-in-chief for their valuable Zümbül Atan http://orcid.org/0000-0003-1328-1756
suggestions and comments.

References
Disclosure statement
Androulakis, I. P. 2009. “MINLP: Branch and Bound Global
No potential conflict of interest was reported by the author(s). Optimization Algorithm.” In Encyclopedia of Optimization,
edited by C. A. Floudas, and P. M. Pardalos, 2132–2138.
Boston: Springer.
Notes on contributors Archibugi, F., and P. Nijkamp. 1989. Economy & Ecology:
Towards Sustainable Development. Dordrecht: Kluwer.
Mostafa Parsa received his BSc and MSc
Arjowiggins Graphic. 2014. “Manufacturing process.” Accessed
in Industrial Engineering from the Depart-
16 November 2019. https://docplayer.net/29622915-Manu
ment of Industrial and Systems Engineer-
facture-of-recycled-pulp-manufacture-of-recycled-pulp-re
ing at Isfahan University of Technology in
cycled-paper.html.
2009 and 2012, respectively. Now, Mostafa
Arjowiggins Graphic. 2019. “Our production mills.” Accessed
is a PhD student in this department. As a
16 November 2019. https://recycled-papers.co.uk/green-
visiting researcher, he has worked closely
matters/our-production-mills.
with the OPAC group (Operations, Plan-
Armstrong World Industries Inc. 2019. “Circular Economy.”
ning, Accounting, and Control) of the Department of Industrial
Accessed 19 June 2019. https://www.armstrongworldindust
Engineering and Innovation Sciences at Eindhoven University
ries.com/en-us/responsibility/sustainability/circular-econ
of Technology (TU/e) since 2018. His research interests include
omy.html.
sustainability, supply chain management, inventory manage-
As’ad, R., M. Hariga, and O. Alkhatib. 2019. “Two Stage
ment, and transportation. In industry, he is the founder of Nakh
Closed Loop Supply Chain Models Under Consignment
Jarahan Pars (Pars Sutures) Co. in the field of medical device
Stock Agreement and Different Procurement Strategies.”
manufacturing in Isfahan Science and Technology Town.
Applied Mathematical Modelling 65: 164–186.
Ali Shahandeh Nookabadi is a professor Atan, Z., and M. Rousseau. 2016. “Inventory Optimization
in the Department of Industrial and Sys- for Perishables Subject to Supply Disruptions.” Optimization
tems Engineering at Isfahan University of Letters 10 (1): 89–108.
Technology where he has been a faculty Atan, Z., and L. V. Snyder. 2012. “Disruptions in one-
member since 1997. He served as the head warehouse multiple-retailer systems.” Available at SSRN:
of this department for 8 years. His research http://ssrn.com/abstract = 2171214.
interests include location theory, facilities Atan, Z., and L. V. Snyder. 2014. “EOQ Models with Sup-
planning, and quality assurance. His cur- ply Disruptions.” In Handbook of EOQ Inventory Problems:
rent research focuses on facilities location, facilities planning, Stochastic and Deterministic Models and Applications, edited
and performance management of supply chain. He currently by T. M. Choi, 43–55. New York: Springer.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 25

Bai, C., J. Sarkis, F. Yin, and Y. Dou. 2019. “Sustainable Supply Circular Economy era.” International Journal of Production
Chain Flexibility and its Relationship to Circular Economy- Research 58 (1): 118–127.
Target Performance.” International Journal of Production Chu, X., Q. Zhong, and X. Li. 2018. “Reverse Channel Selection
Research. doi:10.1080/00207543.2019.1661532. Decisions with a Joint Third-Party Recycler.” International
Bajpai, P. 2014. “4 - Process Steps in Recycled Fibre Processing.” Journal of Production Research 56 (18): 5969–5981.
In Recycling and Deinking of Recovered Paper, edited by P. Chung, S.-L., H.-M. Wee, and P.-C. Yang. 2008. “Optimal Pol-
Bajpai, 55–83. Amsterdam: Elsevier. icy for a Closed-Loop Supply Chain Inventory System with
Bajpai, P. 2018. Biermann’s Handbook of Pulp and Paper: Volume Remanufacturing.” Mathematical and Computer Modelling
1: Raw Material and Pulp Making. Amsterdam: Elsevier. 48 (5): 867–881.
Batista, L., Y. Gong, S. Pereira, F. Jia, and A. Bittar. 2019. “Cir- Cocoon® . 2019. “About Cocoon.” Accessed 19 June 2019.
cular Supply Chains in Emerging Economies – a Compara- https://www.cocoonpaper.com/en/about-cocoon/.
tive Study of Packaging Recovery Ecosystems in China and Costa, L. H. M., B. de Athayde Prata, Helena M. Ramos, and M.
Brazil.” International Journal of Production Research 57 (23): A. H. de Castro. 2016. “A Branch-and-Bound Algorithm for
7248–7268. Optimal Pump Scheduling in Water Distribution Networks.”
Bazan, E., M. Y. Jaber, and A. M. A. El Saadany. 2015. Water Resources Management 30 (3): 1037–1052.
“Carbon Emissions and Energy Effects on Manufactur- Costa, A. L. H., J. L. de Medeiros, and F. L. P. Pessoa.
ing–Remanufacturing Inventory Models.” Computers & 2001. “Global Optimization of Water Distribution Networks
Industrial Engineering 88: 307–316. Through a Reduced Space Branch-and-Bound Search.”
Bazan, E., M. Y. Jaber, and S. Zanoni. 2017. “Carbon Emissions Water Resources Research 37 (4): 1083–1090.
and Energy Effects on a two-Level Manufacturer-Retailer Dakin, R. J. 1965. “A Tree-Search Algorithm for Mixed Inte-
Closed-Loop Supply Chain Model with Remanufacturing ger Programming Problems.” The Computer Journal 8 (3):
Subject to Different Coordination Mechanisms.” Interna- 250–255.
tional Journal of Production Economics 183: 394–408. Danna, E., E. Rothberg, and C. Le Pape. 2005. “Exploring Relax-
Bazaraa, M. S., H. D. Sherali, and C. M. Shetty. 2013. Nonlinear ation Induced Neighborhoods to Improve MIP Solutions.”
Programming: Theory and Algorithms. New York: Wiley. Mathematical Programming 102 (1): 71–90.
Beaton, C. R., and C. Maser. 2016. Economics and Ecology: de Oliveira, W. A., and M. O. Santos. 2018. “A new Branch-
United for a Sustainable World. Boca Raton, FL: CRC Press. ing Rule to Solve the Capacitated lot Sizing and Scheduling
Ben-Daya, M., and A. Al-Nassar. 2008. “An integrated inven- Problem with Sequence Dependent Setups.” TEMA (São
tory production system in a three-layer supply chain.” Pro- Carlos) 18: 515–529.
duction Planning & Control 19 (2): 97–104. Dobos, I., and K. Richter. 2000. “The Integer EOQ Repair and
Ben-Daya, M., R. As’ad, and K. A. Nabi. 2019. “A Single-Vendor Waste Disposal Model—Further Analysis.” Central Euro-
Multi-Buyer Production Remanufacturing Inventory System pean Journal of Operations Research 8 (2): 173–194.
Under a Centralized Consignment Arrangement.” Comput- Dobos, I., and K. Richter. 2003. “A Production/Recycling
ers & Industrial Engineering 135: 10–27. Model with Stationary Demand and Return Rates.” Central
Ben-Daya, M., R. As’ad, and M. Seliaman. 2013. “An integrated European Journal of Operations Research 11 (1): 35–46.
production inventory model with raw material replenish- Dobos, I., and K. Richter. 2004. “An Extended Produc-
ment considerations in a three layer supply chain.” Interna- tion/Recycling Model with Stationary Demand and Return
tional Journal of Production Economics 143 (1): 53–61. Rates.” International Journal of Production Economics 90 (3):
Ben-Daya, M., M. Darwish, and K. Ertogral. 2008. “The joint 311–323.
economic lot sizing problem: Review and extensions.” Euro- Dolgui, A. 2019. “Selected Surveys on Cutting Edge Problems
pean Journal of Operational Research 185 (2): 726–742. in Production Research.” International Journal of Production
Berthold, T. 2006. “Primal heuristics for mixed integer pro- Research 57 (15-16): 4621–4626.
grams.” Master thesis, Technische Universität Berlin. Enkvist, P., and P. Klevnäs. 2018. “The circular economy - a
Berthold, T., S. Heinz, M. E. Pfetsch, and S. Vigerske. 2011. powerful force for climate mitigation: transformative inno-
“Large Neighborhood Search Beyond MIP.” In Proceedings of vation for prosperous and low-carbon industry.”.
the 9th Metaheuristics International Conference (MIC 2011), EPA. 2007a. “Consolidated recovered materials advisory notices
edited by L. D. Gaspero, A. Schaerf, and T. Stützle, 51–60. (RMANs) for the comprehensive procurement guidelines
Udine, Italy: Springer. (CPG).”.
Bressanelli, G., M. Perona, and N. Saccani. 2019. “Challenges EPA. 2007b. “EPA comprehensive procurement guideline
in Supply Chain Redesign for the Circular Economy: a Lit- (CPG) program.”.
erature Review and a Multiple Case Study.” International Erenguc, S. S., and Y. Aksoy. 1990. “A Branch and Bound
Journal of Production Research 57 (23): 7395–7422. Algorithm for a Single Item Nonconvex Dynamic lot Sizing
Brusco, M. J., and S. Stahl. 2006. Branch-and-Bound Applica- Problem with Capacity Constraints.” Computers & Opera-
tions in Combinatorial Data Analysis. New York: Springer. tions Research 17 (2): 199–210.
Burer, S., and A. N. Letchford. 2012. “Non-convex Mixed- Evans, P., A. Sherin, and I. Lee. 2013. The Graphic Design Refer-
Integer Nonlinear Programming: A Survey.” Surveys in ence & Specification Book: Everything Graphic Designers Need
Operations Research and Management Science 17 (2): 97–106. to Know Every Day. Beverly, MA: Rockport.
Choi, D. W., H. Hwang, and S. G. Koh. 2007. “A Generalized Ghasemi Saghand, P., H. Charkhgard, and C. Kwon. 2019. “A
Ordering and Recovery Policy for Reusable Items.” European Branch-and-Bound Algorithm for a Class of Mixed Integer
Journal of Operational Research 182 (2): 764–774. Linear Maximum Multiplicative Programs: a bi-Objective
Choi, T. M., A. Taleizadeh, and X. Yue. 2020. “Game The- Optimization Approach.” Computers & Operations Research
ory Applications in Production Research in the Sharing and 101: 263–274.
26 M. PARSA ET AL.

Ghosh, S. 2007. “DINS, a MIP improvement heuristic.” Paper Land, A. H., and A. G. Doig. 1960. “An Automatic Method of
presented at the 12th International IPCO Conference, Solving Discrete Programming Problems.” Econometrica 28
Ithaca, NY, USA. (3): 497–520.
Giri, B. C., and C. H. Glock. 2017. “A Closed-Loop Sup- Lawler, E. L., and D. E. Wood. 1966. “Branch-and-bound Meth-
ply Chain with Stochastic Product Returns and Worker ods: a Survey.” Operations Research 14 (4): 699–719.
Experience Under Learning and Forgetting.” International Lechner, G., and M. Reimann. 2019. “Integrated Decision-
Journal of Production Research 55 (22): 6760–6778. Making in Reverse Logistics: an Optimisation of Interacting
Giri, B. C., and S. Sharma. 2015. “Optimizing a Closed- Acquisition, Grading and Disposition Processes.” Interna-
Loop Supply Chain with Manufacturing Defects and Quality tional Journal of Production Research. doi:10.1080/00207543.
Dependent Return Rate.” Journal of Manufacturing Systems 2019.1659518.
35: 92–111. Lee, S., and I. E. Grossmann. 2001. “A Global Optimiza-
Glock, C. H. 2012. “The Joint Economic lot Size Problem: A tion Algorithm for Nonconvex Generalized Disjunctive Pro-
Review.” International Journal of Production Economics 135 gramming and Applications to Process Systems.” Computers
(2): 671–686. & Chemical Engineering 25 (11): 1675–1697.
Godichaud, M., and L. Amodeo. 2019. “EOQ Inventory Li, Q., X. Guan, T. Shi, and W. Jiao. 2020. “Green Product
Models for Disassembly Systems with Disposal and Lost Design with Competition and Fairness Concerns in the
Sales.” International Journal of Production Research 57 (18): Circular Economy era.” International Journal of Production
5685–5704. Research 58 (1): 165–179.
Govindan, K., and M. Hasanagic. 2018. “A Systematic Review Lieder, M., and A. Rashid. 2016. “Towards Circular Economy
on Drivers, Barriers, and Practices Towards Circular Econ- Implementation: a Comprehensive Review in Context of
omy: a Supply Chain Perspective.” International Journal of Manufacturing Industry.” Journal of Cleaner Production 115:
Production Research 56 (1-2): 278–311. 36–51.
Govindan, K., H. Soleimani, and D. Kannan. 2015. “Reverse Lu, L. 1995. “A one-Vendor Multi-Buyer Integrated Inventory
Logistics and Closed-Loop Supply Chain: A Comprehen- Model.” European Journal of Operational Research 81 (2):
sive Review to Explore the Future.” European Journal of 312–323.
Operational Research 240 (3): 603–626. Ma, P., K. W. Li, and Z. J. Wang. 2017. “Pricing Decisions in
Guide, V. D. R., and L. N. Van Wassenhove. 2009. “OR Closed-Loop Supply Chains with Marketing Effort and Fair-
FORUM—The Evolution of Closed-Loop Supply Chain ness Concerns.” International Journal of Production Research
Research.” Operations Research 57 (1): 10–18. 55 (22): 6710–6731.
Gupta, O. K., and A. Ravindran. 1985. “Branch and Bound Marendet, A., A. Goldsztejn, G. Chabert, and C. Jermann. 2020.
Experiments in Convex Nonlinear Integer Programming.” “A Standard Branch-and-Bound Approach for Nonlinear
Management Science 31 (12): 1533–1546. Semi-Infinite Problems.” European Journal of Operational
Hariga, M., R. As’ad, and Z. Khan. 2017. “Manufacturing- Research 282 (2): 438–452.
remanufacturing Policies for a Centralized two Stage Sup- Mawandiya, B. K., J. K. Jha, and J. J. Thakkar. 2020. “Opti-
ply Chain Under Consignment Stock Partnership.” Interna- mal Production-Inventory Policy for Closed-Loop Supply
tional Journal of Production Economics 183: 362–374. Chain with Remanufacturing Under Random Demand and
Hasanov, P., M. Y. Jaber, and N. Tahirov. 2019. “Four-level Return.” Operational Research 20: 1623–1664.
Closed Loop Supply Chain with Remanufacturing.” Applied McKinney, R. W. J. 1994. Technology of Paper Recycling. New
Mathematical Modelling 66: 141–151. York: Springer.
ISO 14021. 2016. “ISO 14021:2016 Environmental labels and Mirmohammadi, S. H., S. Shadrokh, and F. Kianfar. 2009.
declarations — self-declared environmental claims (Type II “An Efficient Optimal Algorithm for the Quantity Discount
environmental labelling).”. Problem in Material Requirement Planning.” Computers &
Jaber, M. Y., S. Zanoni, and L. E. Zavanella. 2014. “A Con- Operations Research 36 (6): 1780–1788.
signment Stock Coordination Scheme for the Production, Mitra, S. 2009. “Analysis of a two-Echelon Inventory System
Remanufacturing and Waste Disposal Problem.” Interna- with Returns.” Omega 37 (1): 106–115.
tional Journal of Production Research 52 (1): 50–65. Mitra, S. 2012. “Inventory Management in a two-Echelon
Kazemi, N., N. M. Modak, and K. Govindan. 2019. “A Review Closed-Loop Supply Chain with Correlated Demands and
of Reverse Logistics and Closed Loop Supply Chain Manage- Returns.” Computers & Industrial Engineering 62 (4):
ment Studies Published in IJPR: a Bibliometric and Content 870–879.
Analysis.” International Journal of Production Research 57 Moore, R. E. 1991. “Global Optimization to Prescribed Accu-
(15-16): 4937–4960. racy.” Computers & Mathematics with Applications 21 (6):
Konstantaras, I., and S. Papachristos. 2006. “Lot-sizing for a 25–39.
Single-Product Recovery System with Backordering.” Inter- Nahmias, S., and H. Rivera. 1979. “A Deterministic Model for
national Journal of Production Research 44 (10): 2031– a Repairable Item Inventory System with a Finite Repair
2045. Rate.” International Journal of Production Research 17 (3):
Konstantaras, I., and K. Skouri. 2010. “Lot Sizing for a Sin- 215–221.
gle Product Recovery System with Variable Setup Numbers.” Parsa, M., A. Shahandeh Nookabadi, S. D. Flapper, and Z. Atan.
European Journal of Operational Research 203 (2): 326–335. 2019. “Green hub-and-Spoke Network Design for Aviation
Kronqvist, J., D. E. Bernal, A. Lundell, and I. E. Grossmann. Industry.” Journal of Cleaner Production 229: 1377–1396.
2019. “A Review and Comparison of Solvers for Convex Pinedo, M. L. 2012. Scheduling: Theory, Algorithms, and Sys-
MINLP.” Optimization and Engineering 20 (2): 397–455. tems. 4th ed. New York: Springer.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 27

Rahman, M. O., A. Hussain, and H. Basri. 2014. “A Critical Toth, P., and D. Vigo. 2002. The Vehicle Routing Problem,
Review on Waste Paper Sorting Techniques.” International SIAM Monographs on Discrete Mathematics and Applica-
Journal of Environmental Science and Technology 11 (2): tions. Philadelphia: SIAM.
551–564. Wee, H. M., and C. J. Chung. 2009. “Optimising Replenish-
Reddy, K. N., and A. Kumar. 2020. “Capacity Investment ment Policy for an Integrated Production Inventory Deterio-
and Inventory Planning for a Hybrid Manufacturing – rating Model Considering Green Component-Value Design
Remanufacturing System in the Circular Economy.” Interna- and Remanufacturing.” International Journal of Production
tional Journal of Production Research. doi:10.1080/00207543. Research 47 (5): 1343–1368.
2020.1734681. Widyadana, G. A., and H. M. Wee. 2010. “Revisiting lot Sizing
Richter, K. 1996a. “The EOQ Repair and Waste Disposal Model for an Inventory System with Product Recovery.” Computers
with Variable Setup Numbers.” European Journal of Opera- & Mathematics with Applications 59 (8): 2933–2939.
tional Research 95 (2): 313–324. World Commission on Environment and Development
Richter, K. 1996b. “The Extended EOQ Repair and Waste Dis- (WCED). 1987. Our Common Future. New York: Oxford
posal Model.” International Journal of Production Economics University Press.
45 (1-3): 443–447. Yang, Y., L. Chen, F. Jia, and Z. Xu. 2019. “Complementar-
Richter, K. 1997. “Pure and Mixed Strategies for the EOQ ity of Circular Economy Practices: an Empirical Analysis of
Repair and Waste Disposal Problem.” OR Spektrum 19 (2): Chinese Manufacturers.” International Journal of Production
123–129. Research 57 (20): 6369–6384.
Richter, K., and I. Dobos. 1999. “Analysis of the EOQ Repair Yang, P. C., H. M. Wee, S. L. Chung, and P. C. Ho. 2010.
and Waste Disposal Problem with Integer Setup Num- “Sequential and Global Optimization for a Closed-Loop
bers.” International Journal of Production Economics 59 (1): Deteriorating Inventory Supply Chain.” Mathematical and
463–467. Computer Modelling 52 (1): 161–176.
Rothberg, E. 2007. “An Evolutionary Algorithm for Polishing Yuan, K. F., and Y. Gao. 2010. “Inventory Decision-Making
Mixed Integer Programming Solutions.” Informs Journal on Models for a Closed-Loop Supply Chain System.” Interna-
Computing 19 (4): 534–541. tional Journal of Production Research 48 (20): 6155–6187.
Santibanez Gonzalez, E. D. R., L. Koh, and J. Leung. 2019. Yuan, K. F., S. H. Ma, B. He, and Y. Gao. 2015. “Inventory
“Towards a Circular Economy Production System: Trends Decision-Making Models for a Closed-Loop Supply Chain
and Challenges for Operations Management.” International System with Different Decision-Making Structures.” Inter-
Journal of Production Research 57 (23): 7209–7218. national Journal of Production Research 53 (1): 183–219.
Schrady, D. A. 1967. “A Deterministic Inventory Model for Zhang, Z., J. Wu, and F. Wei. 2019. “Refurbishment or Qual-
Repairable Products.” Naval Research Logistics Quarterly 14 ity Recovery: Joint Quality and Pricing Decisions for new
(3): 391–398. Product Development.” International Journal of Production
Shen, B., T. M. Choi, and S. Minner. 2019. “A Review on Supply Research 57 (8): 2327–2343.
Chain Contracting with Information Considerations: Infor- Zouadi, T., A. Yalaoui, and M. Reghioui. 2018. “Hybrid
mation Updating and Information Asymmetry.” Interna- Manufacturing/Remanufacturing lot-Sizing and Supplier
tional Journal of Production Research 57 (15-16): 4898–4936. Selection with Returns, Under Carbon Emission Con-
Smith, E. M. B., and C. C. Pantelides. 1997. “Global Opti- straint.” International Journal of Production Research 56 (3):
misation of Nonconvex MINLPs.” Computers & Chemical 1233–1248.
Engineering 21: S791–S7S6.
Souza, G. C. 2013. “Closed-loop Supply Chains: a Critical
Review, and Future Research.” Decision Sciences 44 (1): 7–38. Appendices
Stidsen, T., K. A. Andersen, and B. Dammann. 2014. “A
Branch and Bound Algorithm for a Class of Biobjective Appendix A: notations
Mixed Integer Programs.” Management Science 60 (4): 1009– We use the following notation to develop the model separately
1032. for each party.
Sun, H., X. Xia, and B. Liu. 2020. “Inventory lot Sizing Policies
for a Closed-Loop Hybrid System Over a Finite Product Life Manufacturer:
Cycle.” Computers & Industrial Engineering 142: 106340. Retailer:
Taleizadeh, A., and M. S. Moshtagh. 2019. “A Consignment Supplier:
Stock Scheme for Closed Loop Supply Chain with Imperfect MRF:
Manufacturing Processes, Lost Sales, and Quality Depen- Recycling facility:
dent Return: Multi Levels Structure.” International Journal
of Production Economics 217: 298–316.
Teng, H.-M., P.-H. Hsu, Y. Chiu, and H. M. Wee. 2011. “Opti- Appendix B: derivations of the holding costs
mal Ordering Decisions with Returns and Excess Inventory.”
B.1. Manufacturer’s holding costs
Applied Mathematics and Computation 217 (22): 9009–9018.
Figure 17 including top and down graphs represents common
Teunter, R. H. 2001. “Economic Ordering Quantities for Recov-
inventory patterns of the manufacturer’s finished products in
erable Item Inventory Systems.” Naval Research Logistics 48
two forms. In the down graph, the hatched area is the total
(6): 484–495.
inventory of finished products supplied for the retailer, while
Teunter, R. H. 2004. “Lot-sizing for Inventory Systems with
the sum of the rectangle BCEF and the triangle ABF is the total
Product Recovery.” Computers & Industrial Engineering 46
accumulated inventory of finished products at the manufac-
(3): 431–441.
turer’s end. Therefore, the non-hatched area in the down graph
28 M. PARSA ET AL.

k number of shipments to the manufacturer per system cycle


n number of shipments to the retailer per system cycle (a (decision variable)
decision variable) TR recycling time within the system cycle (depended variable)
Tp production time within the system cycle (depended variable) R recycling rate
1−β scrap fraction of production OR fixed transportation cost to deliver the manufacturer’s order
P production rate AR recycling facility’s fixed setup cost
X recycled content level, expressed as a percentage ARu fixed ordering cost from the MRF for recyclable returned
Om fixed transportation cost to deliver the retailer’s order products
Am manufacturer’s fixed setup cost hRr holding cost per unit recycled material per unit time
Amr fixed ordering cost from the recycling facility for recycled hu holding cost per unit recyclable returned product per unit time
materials CR recycling cost per unit
Amp fixed ordering cost from the supplier for virgin materials PR purchase price per unit recyclable returned product from the
hm holding cost per unit finished product per unit time MRF
hmr holding cost per unit recycled material per unit time PRm purchase price per unit production scrap from the
hmp holding cost per unit virgin material per unit time manufacturer
Cp production cost per unit finished product IRr (t) recycling facility’s inventory level of recycled materials w.r.t
Pmr purchase price per unit recycled material from the recycling time
facility IRu (t) recycling facility’s inventory level of recyclable returned
Pmp purchase price per unit virgin material from the supplier products w.r.t time
Im (t) manufacturer’s inventory level of finished products w.r.t time TRR recycling facility’s total revenue per unit time
Imr (t) manufacturer’s inventory level of recycled materials w.r.t TCR recycling facility’s total cost per unit time
time TPR recycling facility’s total profit per unit time
Imp (t) manufacturer’s inventory level of virgin materials w.r.t time
TRm manufacturer’s total revenue per unit time
n(n − 1)DTr2 (n − 1)βPTp Tr
TCm manufacturer’s total cost per unit time = = ,
TPm manufacturer’s total profit per unit time 2 2
   
Tp Tp
BC = nTr − Tp − Tr − = (n − 1) Tr − ,
Tr cycle time for the retailer (decision variable) n n
D demand rate  
Tp
Ar fixed ordering cost from the manufacturer AreaBCEF = BF × BC = βPTp × (n − 1) Tr − ,
hr holding cost per unit finished product per unit time n
Pc retail price per unit finished product to the consumer
Pr purchase price per unit finished products from the manufacturer 1 1 1
Ir (t) retailer’s inventory level of finished products w.r.t time
AreaABF = × BF × AB = × (βPTp ) × Tp = βPTp2 ,
2 2 2
TRr retailer’s total revenue per unit time
TCr retailer’s total cost per unit time Im = AreaBCEF + AreaABF − AreaHatched
TPr retailer’s total profit per unit time  
Tp 1 (n − 1)βPTp Tr
= βPTp (n − 1) Tr − + βPTp2 −
n 2 2
l number of shipments to the manufacturer per system cycle (decision  
variable) 1 (n − 1) (n − 1)
Os fixed transportation cost to deliver the manufacturer’s order = βPTp Tp + (n − 1)Tr − Tp − Tr
As Supplier’s fixed setup cost
2 n 2
 
hs holding cost per unit virgin material per unit time (n − 2) (n − 1)
Cs extraction cost per unit virgin material = βPTp − Tp + Tr . (B.1)
Is (t) supplier’s inventory level of virgin materials w.r.t time 2n 2
TRs supplier’s total revenue per unit time
TCs supplier’s total cost per unit time
Using Equation (B.1), The average inventory of the manu-
TPs supplier’s total profit per unit time facturer’s finished products (Īm ) is given by:
 
m number of shipments to the recycling facility per system cycle Im βPTp (n − 2) (n − 1)
Īm = = − Tp + Tr . (B.2)
(decision variable) nTr nTr 2n 2
Ru return rate of used products (Ru < D)
1−α disposal percentage of returned product Equation (B.2) can be rewritten by Tp = nDTr
from
α recyclable percentage of returned product βP
OM fixed transportation cost to deliver the recycling facility’s order Equation (1) as follows:
AM MRF’s fixed setup cost
Tr D2 (βP − D)nTr D Tr D
hu holding cost per unit recyclable returned product per unit time Īm = + − . (B.3)
CM processing cost per unit recyclable returned product βP 2βP 2
Cd disposal cost per unit non-recyclable returned product
PM unit purchase price per EOL product from the consumer So, the manufacturer’s holding cost per unit time for the
IM (t) MRF’s inventory level of recyclable returned products w.r.t time finished products would be:
TRM MRF’s total revenue per unit time  
TCM MRF’s total cost per unit time Tr D2 (βP − D)nTr D Tr D
TPM MRF’s total profit per unit time hm Īm = hm + − . (B.4)
βP 2βP 2

is equal to the area under the top graph representing the manu- We obtain the manufacturer’s average inventory of the vir-
facturer’s total inventory of finished products in a system cycle gin materials per unit time using Tp = nDT r
βP from Equation (1)
(Im ): as follows:
 
T T
AreaHatched = (DTr )Tr + (2DTr )Tr l lp (1 − X)P lp D2 (1 − X)nTr
= . (B.5)
+ . . . + ((n − 1)DTr )Tr 2nTr 2β 2 lP
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 29

B.3. MRF’s holding cost


Figure 18 shows the MRF’s inventory level over time. The total
inventory of the MRF per system cycle (IM ) is the sum areas of
the rectangle ACEG and the triangle ABC minus the hatched
triangle areas.
 
1 αRu nTr TR αRu nTr TR
AreaHatched = m × × × = ,
2 m m 2m

BF × AC αRu (nTr − TR )nTr


AreaABC = = ,
2 2
αRu TR
AreaACEG = CE × AC = × nTr ,
m
IM = AreaABC + AreaACEG − AreaHatched
αRu (nTr − TR )nTr αRu nTr TR
= + . (B.11)
2 2m
Using Equation (B.11), The average inventory of the MRF is:
IM αRu (nTr − TR ) αRu TR
ĪM = = + . (B.12)
nTr 2 2m

Equation (B.12) can be rewritten by TR = (1−β)D+αβR
βR
u

nTr from Equation (6) as follows:


nαRu [βmR + β(1 − m)αRu + (1 − β)(1 − m)D]Tr
Figure 17. Manufacturer’s inventory level (top), accumulation ĪM = .
2βmR
and supply of manufacturer’s inventory (down) over time for the
(B.13)
finished products.
Therefore, the MRF’s holding cost per unit time would be
as follows using Equation (B.13):
Using Equation (B.5), the manufacturer’s holding cost per hu nαRu [βmR
unit time for the virgin materials is obtained by: +β(1 − m)αRu + (1 − β)(1 − m)D]Tr
 2  hu ĪM = . (B.14)
D (1 − X)nTr 2βmR
hmp . (B.6)
2β 2 lP
B.4. Recycling facility’s holding costs
Similarly, the manufacturer’s average inventory of the recy- Using Equation (6), the recycling facility’s average inventory of
cled materials per unit time is: the recyclable returned products per unit time is formulated by:
   
T
k kp XP kp
T m RT R TR
n(D − Dβ + αβRu )2 Tr
D2 XnTr m m
=
= . (B.7) . (B.15)
2nTr 2β 2 kP 2nTr 2β 2 mR
Using Equation (B.7), the manufacturer’s holding cost per So, the recycling facility’s holding cost per unit time for the
unit time for the recycled materials would be: recyclable returned items is obtained as follows using Equation
(B.15):
 2 
D XnTr hu n(D − Dβ + αβRu )2 Tr
hmr . (B.8) . (B.16)
2β 2 kP 2β 2 mR
B.2. Supplier’s holding cost Using Equation (1) and the same calculation procedure
We can obtain the supplier’s average inventory of the vir- employed in Appendix B.1, the recycling facility’s average
inventory of the recycled materials per unit time is obtained
gin materials per unit time as follows using Tp = nDT r
βP from as follows:
Equation (1): Tp Tp

 
2 2
k X P (R − XP)XPTp k XP Tp
T
(1 − X)PTp2 − l lp (1 − X)P lp
T + −
D2 (l − 1)(1 − X)nTr R 2R 2 nTr
= .
2nTr 2β 2 lP D2 XnTr [−R + kR − (−2 + k)PX]
(B.9) = . (B.17)
2β 2 kRP
Using Equation (B.9), the supplier’s holding cost per unit Consequently, the recycling facility’s holding cost per unit
time is: time for the recycled materials is given by:
 2 
D (l − 1)(1 − X)nTr hRr D2 XnTr [−R + kR − (−2 + k)PX]
hs . (B.10) . (B.18)
2β 2 lP 2β 2 kRP
30 M. PARSA ET AL.

Figure 18. Inventory level of the recyclable returned products over time for MRF (top) and recycling facility (down).

According to Equation (C.7), ∂∂TTP


2
Appendix C: proofs 2 is always negative and TP
r
C.1. TPm is strictly concave with respect to k and l. is strictly concave with respect to Tr . 
The first order partial derivatives of TPm for given values of n
and Tr is as follows: C.3. Proof of Theorem 6.1
The second order partial derivative of TP with respect to X is
∂TPm −2Amp β 2 l2 P+ D2 hmp n2 Tr2 (1 − X)
= , (C.1) calculated by:
∂l 2β 2 l2 nPTr  
∂ 2 TP D2 nTr hRr (k − 2) hu (m − 2)
∂TPm Amr D2 hmr nTr X = + . (C.8)
=− + , (C.2) ∂X 2 β 2R k m
∂k nTr 2β 2 k2 P
∂ 2 TP ∂ 2 TP

Using Equations (C.1) and (C.2), the second order partial ∂Tr2 ∂Tr ∂X
derivatives of TPm for given values of n and Tr is: Let H2 = ∂ 2 TP ∂ 2 TP
denote the Hessian matrix of
∂X∂Tr ∂X 2
∂ 2 TPm ∂ 2 TPm TP(Tr , X). The determinant of matrix H2 , denoted by |H2 | is
= = 0, (C.3) derived as follows:
∂k∂l ∂l∂k
 2   2 
∂ 2 TPm D2 hmp nTr (1 − X) ∂ TP ∂ 2 TP ∂ TP ∂ 2 TP
= − < 0, (C.4) |H2 | = . − .
∂l2 β 2 l3 P ∂Tr2 ∂X 2 ∂Tr ∂X ∂X∂Tr
 2   2
∂ 2 TPm D2 hmr nTr X ∂ TP ∂ 2 TP ∂ 2 TP
=− < 0. (C.5) = . − . (C.9)
∂k 2 β 2 k3 P ∂Tr2 ∂X 2 ∂Tr ∂X
2

∂ TPm ∂ TPm
2
For given values of k, l, m, and n, the maximum of TP(Tr , X)
H1 = ∂l2 ∂l∂k is the Hessian matrix of TPm . The
∂ 2 TPm ∂ 2 TPm can occur at critical points or the boundary of the domain.
∂k∂l ∂k2 The matrix H2 is always indefinite for k ≥ 2&m ≥ 2 because
determinant of denoted by |H1 | is given as follows
matrix H1 , ∂ 2 TP
< 0 and ∂∂XTP
2
using Equations (C.3)-(C.5): ∂Tr2 2 ≥ 0 according to Equations (C.7) and (C.8)

respectively and then |H2 | < 0 according to Equation (C.9). In


D4 hmp hmr n2 Tr2 (1 − X)X this case, each stationary point of TP(Tr , X) is a saddle point
|H1 | = > 0. (C.6)
β 4 k3 l3 P2 and TP(Tr , X) has no local extremum. Therefore, the maxi-
mum of TP(Tr , X) occurs at its boundary of the domain Xmin
The Hessian matrix of TPm is negative definite because the
or Xmax . This completes the proof of Theorem 6.1. 
first and second order leading principal minors of matrix H1 are
always negative and positive respectively ( ∂ ∂lTP2 m 0&|H1 | 0)
2

according to Equations (C.4) and (C.6). Therefore, TPm is


C.4. Proof of Theorem 7.1
The second order partial derivatives of TPAssembly (Tr , X̄) with
strictly concave in l and k. 
respect to Tr and X̄ are as follows:

C.2. TP is strictly concave with respect to T r . ∂ 2 TPAssembly −2
The second order partial derivative of TP with respect to Tr is = nAr + Am + AM + AR + As + kAmr
∂Tr2 nTr3
calculated by:
+ lAmp + mARu + nOm + mOM + kOR
∂ 2 TP −2 
= (nAr + Am + AM + AR + As + kAmr + lAmp
∂Tr 2 nTr3 v
+ li Osi < 0, (C.10)
+mARu + nOm + mOM + kOR + lOs ) < 0. (C.7) i=1
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 31


v ⎤
D.2. Algorithm UB
hRr (k − 2) λi
∂ 2 TPAssembly D2 nTr ⎢ hu (m − 2) ⎥
= ⎢ i=1
+ ⎥.
∂ X̄ 2 R ⎣ k m ⎦

1) Set Y on the number of remaining integer variables that haven’t


(C.11) been allocated to the branching variables yet.
⎡ ⎤ 2) Do steps
 (2-1) and (2-2) for j from 0 to Y:
∂ 2 TPAssembly ∂ 2 TPAssembly Y
∂T
= ⎣ ∂ 2 TPAssembly ∂Tr ∂ X̄ ⎦
Assembly 2 2-1) relaxed problem(s) is/are created after relaxing Con-
Let H2 r
∂ 2 TPAssembly
denote the Hes- j
∂ X̄∂Tr ∂ X̄ 2 straints (9) and (10) and setting j remaining integer variable(s)
on 1.
sian matrix of TPAssembly (Tr , X̄). The determinant of matrix 2-1) Find local maximum(s) by checking the second-order neces-
Assembly Assembly
H2 , denoted by |H2 | is derived as follows: sary and sufficient optimality conditions for each problem of
  the step (2-1).
Assembly ∂ 2 TPAssembly ∂ 2 TPAssembly 2) Find the most TP among the local maximums recorded in the step
|H2 | = . (2-2) as the upper bound (UB) for the node.
∂Tr2 ∂ X̄ 2
 
∂ 2 TPAssembly ∂ 2 TPAssembly
− .
∂Tr ∂ X̄ ∂ X̄∂Tr D.3. Heuristic algorithm
 
∂ 2 TPAssembly ∂ 2 TPAssembly
= .
∂Tr2 ∂ X̄ 2
 2 Input: MINLP as in (20)
∂ 2 TPAssembly Start
− . (C.12) /* solve continuous relaxation of (20) */
∂Tr ∂ X̄ (x̄, T̄r ) ← ArgMax{f (x, Tr ) : Li ≤ xi ≤ Ui , xi ∈ R+ i = 1, . . . , q; Tr ∈ R+ }
v v If x̄ ∈ Z+ then
λX λX (x̄, T̄r ) is the optimal solution
Let X̄max = v i i,max
i=1
and X̄min = v i i,min
i=1
so that
i=1 λi i=1 λi Else
Assembly {
X̄min ≤ X̄ ≤ X̄max . The Matrix H2 is always indefinite
∂ 2 TPAssembly
{
< 0 and ∂ TP∂ X̄2
2 Assembly
for k ≥ 2&m ≥ 2 because ∂Tr2
≥0 For all possible combinations do
according to Equations (C.10) and (C.11) respectively and Fix remaining integer variables to x̄i or x̄i
Assembly Compute TrH by Equation (19) and then TP by Equation (8)
then |H2 | < 0 according to Equation (C.12). In this case, }
each stationary point of TPAssembly (Tr , X̄) is a saddle point /* solve the sub-MINLPs */
and TPAssembly (Tr , X̄) has no local extremum. Therefore, the For j from 1 to q − 1
For r from 1 to j
maximum of TPAssembly (Tr , X̄) occurs at its boundary of the For s from 1 to q − j
domain X̄min or X̄max that means {Xi,min |∀i ∈ {1, 2, . . . , v}} For t from 1 to s
or {Xi,max |∀i ∈ {1, 2, . . . , v}}. This completes the proof of {
Theorem 7.1.  Set xr , . . . , xj to their lower bounds (upper bounds)
Set xj+t , . . . , xj+s to their upper bounds (lower bounds)
Find the optimal solution (x̂, T̂r ) for continuous relaxation of the
Appendix D: algorithms sub-MINLP
{
D.1. Algorithm SE For all possible combinations do
Fix remaining integer variables on x̂i or x̂i
Compute TrH by Equation (19) and then TP by Equation (8)
}
}
1) For n from 1 to n0 do steps (1-1) and (1-2): Find the most TP and its corresponded decision variables as the best
1-1) Calculate krx and lrx from Equations (14) and (15). possible solution obtained by the heuristic
1-2) For given value of n, calculate TPm (k, l, n) via Eq. (13) }
in (k = k1 , l = l1 ), (k = k1 , l = l2 ), (k = k2 , l = l1 ), and End.
(k = k2 , l = l2 ) where
k1 = max[1, min{krx , k0 }], k2 = min{krx , k0 },
l1 = max[1, min{lrx , l0 }], l2 = min{lrx , l0 }.
2) Find the most TPm resulted from step 1 and its corresponded
combination of k# , l# , and n# as the optimal solution of the
sequential structure.
32 M. PARSA ET AL.

Appendix E: calculations algorithm obtains the optimal decision variables for the inte-
grated system as k∗ = 5, l∗ = 2, m∗ = 5, n∗ = 12, and Tr∗ =
E.1. Calculations for the numerical example 1.0228. The corresponding total profit is TP(k∗ , l∗ , m∗ , n∗ , Tr∗ )
B&B algorithm: = 33810.517. The profits for the manufacturer, the retailer,
First, we perform the initialisation step of the developed the supplier, the MRF, and the recycling facility in the inte-
B&B algorithm. We obtain Tr# = 0.8165 by Equation (12). grated decision-making process are TPm = 16747.045, TPr =
Then, using Algorithm SE, we obtain n# = 15, k# = 6, and l# = 3116.058, TPs = 8644.762, TPM = 1919.644, and TPR =
5. mrx = 4.707 is obtained from Equation (17). Since TPR 3383.009, respectively.
(m = 5) = 3378.249 is greater than TPR (m = 4) = 3377.372, Heuristic algorithm:
m# is set to 5. The maximum total profit from the decentralised According to the heuristic algorithm, as the first step,
/sequential decision making, obtained using Equation (18), we find the optimal solution of the continuous relaxation.
is TP(k# , l# , m# , n# , Tr# ) = 33649.760. The corresponding prof- We obtain the following solution: k = 5.421, l = 1.558, m =
its for the manufacturer, the retailer, the supplier, the MRF, 5.419, n = 12.168, and Tr = 1.013. Since this solution is not
and the recycling facility in the sequential structure are integral, we continue the algorithm. If we set k = 5.421 =
TPm = 16778.594, TPr = 3162.883, TPs = 8410.209, TPM = 5, l = 1.558 = 2, m = 5.419 = 5, and n = 12.168 = 12, we can
1919.826, and TPR = 3378.249, respectively. compute TrH = 1.0228 via Equation (19). At the end of the
We use k# = 6, l# = 5, m# = 5, n# = 15, and Tr# = 0.8165 algorithm, the solution k = 5, l = 2, m = 5, n = 12, and TrH =
as the first incumbent and TP(k# , l# , m# , n# , Tr# ) = 33649.760 1.0228 represents the best possible solution obtained by the
as the initial lower bound (LB). Finally, the developed B&B heuristic. Note that this is also the global optimal solution.

You might also like