You are on page 1of 26

Accepted Manuscript

Production planning for stochastic manufacturing/remanufacturing system with


demand substitution using a hybrid ant colony system algorithm

Wenjie Liu, Wenyan Ma, Yi Hu, Mingzhou Jin, Kai Li, Xiangyun Chang, Xianyu Yu

PII: S0959-6526(18)33921-0
DOI: https://doi.org/10.1016/j.jclepro.2018.12.205
Reference: JCLP 15263

To appear in: Journal of Cleaner Production

Received Date: 12 September 2018


Revised Date: 19 December 2018
Accepted Date: 19 December 2018

Please cite this article as: Liu W, Ma W, Hu Y, Jin M, Li K, Chang X, Yu X, Production planning
for stochastic manufacturing/remanufacturing system with demand substitution using a hybrid
ant colony system algorithm, Journal of Cleaner Production (2019), doi: https://doi.org/10.1016/
j.jclepro.2018.12.205.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Production Planning for Stochastic Manufacturing/Remanufacturing System with
Demand Substitution Using a Hybrid Ant Colony System Algorithm
Wenjie Liu1, Wenyan Ma1, Yi Hu2, Mingzhou, Jin3,4,*, Kai Li5, Xiangyun Chang6, Xianyu Yu1
1. College of Economics and Management, Nanjing University of Aeronautics and Astronautics, Nanjing 211106,
China
2. Yangtze Memory Technologies Co., Ltd, Wuhan, 430000, China
3. College of Logistics and Transportation Engineering, Central South University of Forestry and Technology,
Changsha, 410004, China
4. Department of Industrial and Systems Engineering, University of Tennessee, Knoxville 37996, USA, jin@utk.edu

PT
5. Department of Business Administration,Hefei University of Technology, Hefei 230009, China
6. Department of Management Science and Engineering, East China University of Science and Technology,
Shanghai 200237, China

RI
* Corresponding Author

Abstract: A hybrid manufacturing/remanufacturing system (HMRS) is an effective tool to address the

SC
global challenge of resource depletion and environmental deterioration. This paper aims to make an
optimal production plan for a stochastic HMRS with demand substitution. To achieve the above
objective, a multi-period mixed integer programming model was first constructed. An ant colony

U
system algorithm with random sampling method (ACS-RSM) was proposed to minimize the total
expected cost of the stochastic HMRS. Finally, the proposed model and ACS-RSM algorithm were
AN
applied to an auto alternator case. The effects of the used product recovery rate and batch sizes of new
and remanufactured products on the total expected cost were analyzed. The research results showed
M

that the ACS-RSM algorithm performed well regarding computational efficiency and solution quality.
There were two major findings through the practical case study. The first finding was that with
increase of recovery rate of used product, total expected cost of the HMRS declined dramatically until
D

a certain point. When the recovery rate was greater than 91%, the total expected cost kept almost
constant. The second finding was that when the batch sizes of the new product and remanufactured
TE

product rose, the total expected cost had an obvious increase and the running time of the ACS-RSM
algorithm decreased monotonically. The study yields an effective decision-making tool for optimizing
EP

the production plan of the stochastic HMRS with demand substitution.


Key words: Production planning; Hybrid manufacturing/remanufacturing system; Demand
substitution; Ant colony system algorithm with random sampling method
C

1. Introduction
AC

The rapid development of industrialization has caused serious resource depletion and environmental
deterioration (Liu et al. 2017). Remanufacturing is a process of transforming used or end-of-life
products into a “like new or better” condition by a series of activities, such as dismantling, cleaning,
remanufacturing, reassembly, and testing (Hauser and Lund, 2003). It first appeared in the United
States in the 1930s and successfully addressed a serious crisis of commercial vehicles during World
War II. In comparison to a new product, a remanufactured product can be produced with 30% raw
materials, 40% energy, 50% cost, and less than 20% pollution emissions (Rathore et al., 2011;
Govindan et al., 2016). Remanufacturing contributed greatly to decreasing resource depletion and
environmental pollution. Many countries, such as the United States, Germany, Netherland, Japan,
1
ACCEPTED MANUSCRIPT
China and India, highly appreciated remanufacturing (Liu et al., 2017). A number of world-famous
enterprises have highly visible remanufacturing programs in automobiles, electronics, and
construction machinery, etc. Most automobile companies (e.g., Toyota, Ford, GM, and Volkswagen)
have launched their remanufacturing business for auto engines, gear-boxes, alternators and starters.
Hewlett-Packard sold remanufactured computers in their outlet store. Many of them have
implemented a hybrid manufacturing/remanufacturing system (HMRS) to simultaneously provide
new products and remanufactured products to consumers. For example, PICO (ChangShu) Auto

PT
Motor Co. Ltd in China engaged in manufacturing and remanufacturing of auto starters and
alternators for thirty years.
Current HMRS studies mainly focus on inventory control, production planning and job

RI
scheduling. Most studies on the production planning of the HMRS assumed that the demand of
remanufactured product and the return of used product were deterministic. They also assumed that

SC
new and remanufactured products were homogenous in quality and were accepted with the same
prices by consumers. In reality, the HMRS faces a complex production environment. The return of
used product and the demand of remanufactured product are usually stochastic and uncertain.

U
Consumers usually believe that the quality of a remanufactured product is inferior to a new one.
Furthermore, many countries had banned remanufactured products from being sold as new products.
AN
For instance, the federal government in the U.S. required that remanufactured goods should not be
marketed as new products (Ahiska et al., 2017). Additionally, downward demand substitution, where
the new product substitutes for remanufactured item, is sometimes performed to avoid a shortage of
M

the remanufactured products. Current studies are unable to efficiently derive an optimal production
plan for the HMRS with stochastic demand and return and with downward demand substitution. It is
D

becoming a critical issue for enterprises to implement the HMRS. Therefore, this paper aims to
determine the optimal production plan of the stochastic HMRS with downward demand substitution.
TE

The resulting decision making tool out of this paper is applicable at many auto part enterprises
involving HMRS and will contribute to realizing cleaner production and sustainable development
through more efficient remanufacturing planning.
EP

The remainder of the paper is structured as follows. Section 2 presents a brief literature review
and identifies research gaps. Section 3 constructs a multi-period mixed integer programming model
for the stochastic HMRS with demand substitution. A hybrid ACS-RSM algorithm is developed to
C

solve the above model in Section 4. Section 5 applies this algorithm to an auto alternator case in
AC

China. Optimal production plan is investigated and effects of recovery rate of used product and batch
sizes of new and remanufacturing products on the stochastic HMRS are also discussed. Section 6
summarizes the whole study and lists several future research directions.

2. Literature Review
This section consists of two subsections to better understand current researches on production
decision on the HMRS. The first subsection overviews relevant research literatures. The second
subsection lists research gaps to position this paper.

2
ACCEPTED MANUSCRIPT

2.1 Research on production plan of the HMRS


The earliest research on the HMRS was performed by Van der Laan et al. (1999a). After introducing
major activities of the HMRS, they investigated production plan and inventory control for the HMRS
and compared PUSH and PULL control policies for a HMRS. Over the past two decades, the business
values of the HMRS were gradually recognized by global enterprises and academics. Its importance
on environmental protection and resource conservation has also been realized by governments around
the world. A number of HMRS studies placed emphasis on inventory control (Van der Laan et al.,

PT
1999b; Tang and Grubbström, 2005; Sadok and Nidhal, 2016; Inderfurth, 2004; Ahiska and Kurtul,
2014; Ahiska et al., 2017) and job shop scheduling (Kouedeu et al., 2013; Polotski et al., 2015;

RI
Polotski et al., 2017; Giglio et al., 2017). Although inventory control and job shop scheduling are
important decisions of the HMRS, this paper is more interested in product planning.
Research on product planning can be classified into two categories: one is for manufacturing

SC
system. The other is for hybrid manufacturing/remanufacturing system (HMRS). In recent years,
Fazlollahtabar and Mahdav (2009) optimized production time and costs of an automated

U
manufacturing system by using stochastic programming. Deroussi and Fonseca (2009) adopted a
hybrid Ant Colony System to solve machine assignment problem in flexible manufacturing systems.
AN
Other researchers explored product planning of manufacturing systems from the view of supply
chains. The production planning for the HMRS has been increasingly discussed in the literature.
Depending on whether or not carbon emissions are considered, these researches can be divided into
M

two groups, as shown in Table 1.


Table 1 Literatures on production plan of the HMRS
D

Homogeneous New
Demand and Production Planning
TE

Authors and Remanufactured


Return Substitution Horizon
Products?
Group 1: With Zouadi et al. (2016) Deterministic No Yes Multi-period
Carbon Emission Chang et al. (2015) Deterministic Yes Yes and No Two-period
EP

Consideration Wang et al. (2017) Stochastic Yes No Single-period

Kiesmüller (2003) Stochastic No Yes Multi-period


C

Group 2: Without Cai et al. (2014) Stochastic No Yes Multi-period


Kenné et al. (2012) Deterministic No Yes Time
Carbon Emission
AC

Francie et al. (2015) Deterministic No Yes Time


Consideration Roshani et al. (2017) Deterministic No Yes Multi-period
This paper Stochastic Yes No Multi-period

For Group 1 in Table 1, production plans are affected by carbon emission constraints. Zouadi et al.
(2016) discussed a lot-sizing problem of the HMRS under a carbon emission constraint, in which
deterministic demand and return over a finite planning horizon were considered. The paper
consturcted a mathematical model and provided two hybrid solutoin methods. Research results
showed that these two methods could provid high-quality solutions in moderate computational time.
3
ACCEPTED MANUSCRIPT
Chang et al. (2015) addressed production planning under a carbon cap and trade mechanism. They
proposed two revenue maximization models for independent demand market and substitutable
demand market, respectively. Optimal production quantities of new product and remanufactured
product were determined and effects of the carbon cap and trade mechanism on production plan were
also analyzed. Wang et al. (2017) further examined production decision issues under the carbon cap
and trade mechanism and capital constraints, where new product can substitute for remanufactured
product. They presented three single-period models to determine reasonable production plans of

PT
new product and remanufactured product, and investigated the effects of four key parameters (such as
initial capital, carbon cap, carbon price, and carbon emission rate) on optimal production decisions.

RI
The results showed that 1) the capital constraints encouraged manufacturers to produce
remanufactured products at a level of high quality; 2) the carbon cap exerted strong influence over

SC
manufacturing and remanufacturing quantities of a capital-constrained manufacturers; and 3) the
carbon price had little effect on production policies of the capital-constrained manufacturers.
For Group 2 in Table 1, Kiesmüller (2003) addressed production control in a stochastic HMRS

U
with different manufacturing and remanufacturing leadtimes. A new approach was provided to
AN
determine optimal manufacturing and remanufacturing quantities and time. Using a numerical
example, the author illustrated that production performances could be improved substantially by a
proposed new approach. Cai et al. (2014) studied acquisition and production plan for an HMRS,
M

where they insisted that new and remanufactured products were of the same price and returned
products were of high quality or low quality. They modeled this problem as a Markov decision
D

process and derived optimal acquisition price and production strategies. Kenné et al. (2012) and
Francie et al. (2015) dealt with production planning of an HMRS in which machines suffered from
TE

stochastic breakdowns and repairs. Assuming that demand and return were deterministic and the new
and remanufactured products were homogeneous, Kenné et al. (2012) constructed a stochastic
EP

optimization model and designed an algorithm to investigate optimal production rates of


manufacturing and remanufacturing machines. Francie et al. (2015) adopted a Markov decision
process to determine optimal production rates. They provided a numerical case to illustrate
C

effectiveness of the proposed Markov decision model. Addtionally, Roshani et al. (2017) focused on a
AC

dynamic lot-sizing problem of the HMRS with a capacitated machine, which performed both
manufacturing and remanufacturing operations. A mathematical programming model for this problem
was established and a heuristic approach was proposed to solve it. Some experimental instances were
generated to evaluate the algorithm’s efficiency.

2.2 Research gap analysis and highlights of this study


Most research related to the HMRS concentrated on inventory control and job shop scheduling.
Production planning of the HMRS is facing a complex production environment, such as stochastic
demand of remanufactured product, random return of used product, multiple stages, and downward
4
ACCEPTED MANUSCRIPT
demand substitution. As seen in Table 1, few papers explored optimal production planning for the
stochastic HMRS with demand substitution over a multi-period horizon. This paper constructed a
stochastic programming model for the stochastic HMRS with downward demand substitution and
proposed an ACS-RSM algorithm to determine optimal production plan. Its highlights are as follows:
1) Establishment of a multi-period mixed integer program by thoroughly analyzing the HMRS with
stochastic demand and return and with downward demand substitution.
2) Development and application of an ant colony system algorithm combined with random sampling

PT
method to determine the optimal production plan of the stochastic HMRS.
3) Verification of the proposed model and algorithm with an auto alternator case and discussion of

RI
the effect of two key parameters (i.e., recovery rate of used product and batch sizes of new and
remanufactured products) on performances of the stochastic HMRS.

SC
3. Problem description
A stochastic HMRS with downward demand substitution is considered to satisfy multi-period

U
demands of new and remanufactured products, as illustrated in Fig. 1. In this hybrid system,a new or
remanufactured product comprises core component and non-core components. The core component
AN
can be original or remanufactured component, while the non-core ones can only be original
components. There are three inventories in this system (returned product, remanufactured product and
M

new product) and two production activities (manufacturing and remanufacturing). Manufacturing
provides brand-new products to replenish serviceable inventory of new product, which are primarily
used to satisfy demand of new product
D

. Remanufacturing offers remanufactured products, which


are manufactured from collected used products, to replenish serviceable inventory of remanufactured
TE

product. They are used to meet demand of remanufactured product. A returned product can provide a
used core component after dismantling. Only a certain percentage of used core components are
suitable for remanufacturing, which is called recovery rate θ .
EP

ztN(ξ )
Original
C

Components DtN
I tN(ξ )
AC

e(
t
ξ

DtR(ξ )
Rt(ξ )
z (ξ )
t
R

I tO(ξ ) zt (ξ ) I tR(ξ )
O

Fig. 1 A multi-period HMRS with demand substitution


In real-world situations, returned amount of used product and demand for remanufactured
product are both stochastic and depend on scenario . The stochastic feature may lead to
5
ACCEPTED MANUSCRIPT
excessive inventory or out of stock if a poor production decision is made. When there is an overstock
of returned products in the hybrid system, used products will be disposed of due to a limited inventory
capacity. In addition, when remanufactured products are out of stock, downward substitution can be
adopted to lower risk of lost sales. In other words, a new product, if necessary, can be provided to the
customer at the price of a remanufactured product (Ahiska and Kurtul, 2014). If there are still unmet
demands for the new product and remanufactured product, they should be met by purchasing new
products and remanufactured products from a spot market.

PT
The problem arising from this stochastic HMRS is to determine how many new product batches to
manufacture, how many batches to remanufacture, and at which periods these production activities

RI
take place. In order to solve above issues, the following additional assumptions regarding values and
costs are made. The quality of remanufactured products has an inferior perceived value compared

SC
with that of new products from the customer viewpoint. Unit purchase cost of remanufactured
products is less than unit purchase cost of new products . Unit manufacturing cost of
remanufactured products is lower than unit manufacturing cost of new products . In addition,
is less than unit purchase cost of remanufactured

U
unit manufacturing cost of new products
AN
products . Above cost structure ensures that downward substitution is first performed when
remanufactured product is out of stock.
M

4. Proposed optimization model


This study focuses on minimizing total expected cost of the stochastic HMRS with downward demand
D

substitution by determining an optimal production plan. In this section, a stochastic programming


model for the stochastic HMRS is proposed, which is followed by an ACS-RSM optimization
TE

algorithm to seek for the optimal production plan. The following notations are used for the proposed
stochastic programming model.
Parameters:
EP

: Planning horizon, = 1, … , ;
Φ: Set of scenarios, which can be finite or infinite, is its element;
∈ , ,
C

: , for new products, for remanufactured ones, and for returned


AC

products;
: Demand of new products in period ;
: Stochastic demand of remanufactured products in period in scenario ;
: Stochastic returned amount of used products in period in scenario ;
: Recovering rate of used products;
: Setup cost for manufacturing product per period, ∈ , ;
: Unit manufacturing cost of product , including component cost ∈ , ;

6
ACCEPTED MANUSCRIPT
ℎ : Unit inventory holding cost of product , ∈ , , ;
: Unit disposal cost of returned products;
: Unit purchase cost of product from the spot market in period , ∈ , ;
: Unit downward substitution cost, which equals to − ;
! : Inventory capacity of product , ∈ , , ;
" : Manufacturing capacity of product in each period, ∈ , ; and
# : Batch size of product , ∈ ,

PT
.
Variables

$ : ∈ ,

RI
Number of manufacturing batches of product in period , ;

% : Binary variable to indicate whether manufacturing product happens in period or not


(% = 1 if so; % = 0, otherwise);

SC
' : Quantity of product obtained from the spot market in period in scenario , ∈ , ;
') : Quantity of disposed returned products in period in scenario ;

*
+
: Inventory of product at the end of period
U in scenario

: Quantity of new products substituting for remanufactured ones in period


, ∈ , , ;
AN
in scenario .
Production plan $ , % ; $ , % is decided at the beginning of planning horizon and does not
M

change based on scenarios. However, inventory, purchasing, and substitution quantities are scenario
sensitive.
D

4.1 Stochastic programming model


The stochastic HMRS with downward demand substitution is modeled as a multi-period mixed
TE

integer stochastic program model. This model has been validated with commercial solver Guribi 8.0
and results out of the model were correct. The model realization file is enclosed as supplemental
EP

materials.
9

min , = - . - / % + # $ 1+ E 5 - 6 * + + + ') + - ' 78 (1)


3∈4
:; ∈ , ∈ , ,) ∈ ,
C

s.t. * = * <; +# $ − −+ +' = 1,2, … , ; ∈ >


AC

(2)

* = * <; +# $ − ++ +' = 1,2, … , ; ∈ > (3)

*) = * )<; + − $ − ')
?@
= 1,2, … , ; ∈ >
A
(4)

# $ ≤" % = 1,2, … , ; ∈ , (5)

* <; ≤! = 1,2, … , ; ∈ >; ∈ , , (6)

7
ACCEPTED MANUSCRIPT

$ , $ ∈ C D ; % ∈ 0,1 ; + ,' ,* ≥ 0. = 1,2, … , ; ∈ >; ∈ , ,

Objective function (1) is to minimize the total expected cost of the HMRS, including set-up costs
to have manufacturing and remanufacturing operations in each period, variable manufacturing and
remanufacturing costs, inventory holding costs of new product, remanufactured product and returned
product, substitution cost, disposal cost for excessive inventory of returned products, and purchase
costs for the new products and remanufactured products over all possible scenarios. Note that the
scenario set > can be finite or infinite. Constraints (2), (3) and (4) capture the dynamics of

PT
serviceable inventories of the new product, remanufactured product and used product. Constraint (5)

RI
states that production outputs of the new product and remanufactured product are limited by
manufacturing capacity and remanufacturing capacity in each period. Constraint (6) represents
inventory capacities of the new items, remanufactured items and returned items. Constraint (7)

SC
declares the types of variables. Due to existence of spot market, the stochastic program (1-6) is
feasible under any scenario.

U
4.2 ACS-RSM algorithm
AN
The ant colony algorithm is a meta-heuristic algorithm proposed by Dorigo et al. (2000). Through
studying foraging behaviors of ant colonies, they found that ants usually released some chemical trails
called pheromones on the path, which was detected by other ants nearby and guided their own
M

foraging behaviors. Ants tended to choose paths with high-intensity pheromones. If a path attracted
more ants, it had more cumulative pheromones, resulting in a higher probability of choosing this path
D

by other ants. Thus, a positive feedback was formed within an ant colony by exchanging the
pheromone information and this feedback helped ants to find an optimal foraging path. This algorithm
TE

has successfully used in different optimization problems. For instance, vehicle routing problems (Goel
and Maini, 2018; Kalayci and Kaya, 2016; Schyns, 2015), quadratic assignment problems, job shop
scheduling (Huang and Yang, 2008; Leung et al., 2010) and production planning (Du and Guo, 2016;
EP

Wang et al., 2014; Shishvan and Sattarvan, 2015). Production planning for a stochastic HMRS with
downward substitution is a typical NP-hard problem. Considering complexity and randomness of this
C

problem, an ant colony algorithm combining with the random sampling method (ACS-RSM) is
proposed to explore an optimal production plan. This hybrid algorithm can not only reduce total
AC

expected cost of the HMRS but also improve computational efficiency.


4.2.1 Algorithm framework
Major difficulty in determining an optimal production plan is to deal with stochastic demand and
return in the HMRS. According to Monte Carlo Sampling Methods developed by Shapiro (2003),
random sampling data can effectively substitute for stochastic data when the size of random sampling
is large enough. This paper proposed an ACS-RSM algorithm framework to overcome computational
difficulty caused by stochastic feature of the HMRS. This framework consists of two modules: Ant
Colony System (ACS) module and Random Sampling Method (RSM) module, as illustrated in Fig. 2.

8
ACCEPTED MANUSCRIPT
The RSM module includes Random Sampling and Solution Evaluation (RSSE) sub-module and local
search (LS) sub-module. There are six major processes in the framework of ACS-RSM algorithm, as
shown below.
1) Import the stochastic HMRS with demand substitution to the RSM module.
2) Generate G (a big number) deterministic problems of production planning by random sampling
method, where M random sampling data can substitute for stochastic demand of remanufactured
product and stochastic return amount of used product. Demand of remanufactured product and

PT
returned quantity of used product in each period are generated according to their stochastic
distribution functions. These deterministic problems are transferred to the ACS module.
3) Obtain initial optimal production plan for each iteration by the ACS module. Using ant colony

RI
system algorithm (described in Appendix 1), the ACS module determines an optimal production
plan for each deterministic problem. Based on all obtained production plans, the most frequently

SC
occurring number of manufacturing and remanufacturing batch are identified as the initial optimal
production plan at this iteration.
4) Generate W (another big number) alternative production plans by the LS sub-module. Based on
the initial optimal production plan, LS sub-module generates H alternative production plans for

U
AN
the RSSM sub-module by using a neighborhood search, which can prevent the ACS-RSM
algorithm from trapping in local optimum.
5) Identify the best production plan in each iteration and update the pheromone. Using each
M

alternative production plan, the total cost of each previously generated deterministic problem is
calculated to obtain average total cost of the G deterministic problems. The alternative
D

production plan with minimum average total cost is chosen as the best production plan of this
stochastic HMRS in the current iteration. According to this best plan, the ACS-RSM algorithm
TE

updates the pheromone and then enters next iteration. It keeps previously generated M
deterministic problems for determining the optimal production plan of the next iteration.
6) Output final optimal production plan for stochastic MRHS with downward substitution. After
performing I iterations or satisfying algorithm stop condition, the final production plan of the
EP

stochastic HMRS can be exported.


C
AC

9
ACCEPTED MANUSCRIPT

PT
RI
U SC
Fig. 2 A framework of ACS-RSM algorithm
AN
4.2.2 Algorithm steps
The flows of the ACS-RSM algorithm are shown in Fig 3. The detailed steps are described below.
M
D
TE
C EP
AC

10
ACCEPTED MANUSCRIPT

PT
RI
U SC
y
TCmin
AN
y
TCmin =TCmin,y = y + 1
ACS-RSM
|TCmin y
-TCmin | ≤ ε or y = Y
M
D

Fig. 3 Flow of ACS-RSM algorithm for stochastic HMRS


TE

Step 1: Initialize parameters. The parameters of the ACS-RSM algorithm are divided into two types.
The first type is algorithm parameters, which include the population size of ants z, initial
pheromone value JK , pheromone trail factor L , heuristic information factor M ,
EP

pseudo-random rule proportion NK , preferred probabilityOK , selection proportions O; and pQ ,


pheromone evaporation rate R , algorithm stop criterion S , maximum iterations I , and
constants G and H. These algorithm parameters are described in detail in Appendix 1. The
C
AC

second type is system parameters defined in Section 3.1.


Step 2: Determine optimal production plan for G deterministic problems of production plan for
iteration % by ACS algorithm. The RSSE sub-module generates G deterministic problems
through sampling stochastic distribution functions of the demand of remanufactured products
and the return quantity of used products. These deterministic problems are recorded and are
provided to the ACS module. If W < G, the ACS module determines an optimal production
plan for each deterministic problem by using the ACS algorithm, as described in Appendix 1.
Step 3: Form an initial optimal production plan for the iteration % by the ACS module. Based on M
optimal production plans obtained in Step 2, most frequently occurring numbers of
11
ACCEPTED MANUSCRIPT
manufacturing and remanufacturing batch are identified at each period t. These batch numbers
form an initial optimal production plan for this stochastic HMRS at iteration %.
Step 4: Generate W alternative production plans at iteration y. To avoid local optimum, the LS
sub-module generates H alternative production plans from the initial optimal production
plan by using a neighborhood search and provides them to the RSSE sub-module.
Step 5: Calculate average total cost of each of the H alternative production plans. For each
alternative production plan, the total cost of each of the M deterministic problems is determined

PT
and their average total cost is then calculated.
Step 6: Identify optimal production plan from the W alternative production plans at iteration y and
update the pheromone. The alternative production plan with the minimum average total cost is

RI
selected as the optimal production plan of the stochastic HMRS at iteration y. Its average total
cost is recorded as total expected cost ,YZ[ . The pheromone of the ACS-RSM algorithm is
\

SC
updated by using formula (A5) in Appendix 1.
Stop or continue the iteration. If either inequalities ] ,YZ[ − ,YZ[ ] < S or % < I is
^_`< `a \
Step 7:
satisfied, go to Step 8. Otherwise, retain the M deterministic problems for determining the

U
optimal production plan of next iteration and go to Step 2.
AN
Step 8: Output the final optimal production plan of the stochastic HMRS and its total expected cost.

5. A practical case study


M

5.1 Data resource


Most automobile companies, such as Toyota, Ford, GM and Volkswagen, have established their own
D

HMRSs. In China, the HMRS was first established by some automotive part enterprises in the late
1980s. For example, PICO (Changshu City) Auto Motor Remanufacturer Co., Ltd built the earliest
TE

HMRS of auto alternators and auto starters in 1988 for various automakers, such as Ford, Delco,
Mitsubishi, and Bosch. The products were usually sold to the United States, Canada, Europe, and
EP

Southeast Asia. From 2005 to 2012, remanufacturing volume of auto parts in China boosted from
5,000 units to 1.5 million units and annual sales soared from 50 million RMB to 8 billion RMB (Liu
et al. 2017).
C

In this section, the auto alternator HMRS of PICO Auto Motor Remanufacturer Co., Ltd was
AC

selected as a practical case to validate the proposed stochastic programming model and ACS-RSM
algorithm in Section 4. System parameters of the stochastic programming model were investigated
and their values were obtained from this auto alternator HMRS. In this practical case, each auto
alternator consists of one core component and three non-core components. The demand of the new
products is deterministic,as shown in Table 2. However the demand of remanufactured
products and the returned quantity of used product in each period are stochastic and they are
subject to two identical and independent Poisson distributions with parameters of b; = 200 and λQ =
150, respectively. In other words, = e = b;f + <gh /e! and = e = bfQ + <gk /e!. Batch
12
ACCEPTED MANUSCRIPT
sizes of new product and remanufactured product # , # are set at 10 items/batch. Other parameters
are listed in Table 3.

Table 2 Demand of new product in the case

Period t
1 2 3 4 5 6
100 100 120 120 110 110

PT
Table 3 Other system parameters in the case

l
Parameter Value Unit Parameter Value Unit

mn × 10 RMB × 10 RMB
6 week 90 %

RI
1,350 1,250
wn 25 /time× 10 RMB 20 × 10 RMB
xn
y × 10 RMB × 10 RMB

SC
40 32
z 4 × 10 RMB 8 × 10 RMB
{ n 7 /item × 10 RMB ℎ 6 /item × 10 RMB
{ | × 10 RMB ! )

U
2 40 item
} n 80 item ! 80 item
~n "
AN
200 item 210 item

5.2 Sensitivity analysis of key parameters of ACS-RSM algorithm


Key parameters of ACS-RSM algorithm comprised pheromone trail factor L, heuristic information
M

factor M, pseudo-random rule proportion NK and preferred probability OK . Considering that solution
D

quality of the ACS-RSM algorithm is affected by the above key parameters, sensitivities of this
algorithm to the variations of four key parameters are analyzed as below. The algorithm is compiled
TE

on the MATLAB 6.1 platform, running on a computer with 2.5GHz frequency, dual-core CPU, 4G
memory and WIN7 OS.
EP

5.2.1 Pheromone trail factor and heuristic information factor


Importance of the pheromone trail factor L and heuristic information factor M comes from their
impact on computational efficiency and solution quality of the ACS-RSM algorithm. Based on study
of Dorigo and Stutzle (2004), the ranges of L and M should be separately set in [2, 4] and [1, 3] to
C
AC

find optimal production plan efficiently. In this study, the effect of using higher or lower pheromone
trail factor and heuristic information factor are shown in Table 4. Here, NK and OK are deterministic
and equal to 0.7.

Although the best pheromone trail factor and heuristic information factor cannot be obtained
through Table 4, it shows that high value for pheromone trail factor and heuristic information factor
does not produce a considerable improvement on the total expected cost. So L = 2 and M = 1 were
selected for the following experiments for the ACS-RSM algorithm.

13
ACCEPTED MANUSCRIPT
Table 4 Effect of pheromone trail factor α and heuristic information factor β

L M L M L M
Experiment Total Experiment Total Experiment Total
Number Expected cost Number Expected cost Number Expected cost

1 2 1 57,870 4 3 1 59,245 7 4 1 61,814

2 2 2 59,218 5 3 2 60,212 8 4 2 62,535

3 2 3 60,856 6 3 3 62,596 9 4 3 63,923

PT
5.2.2 Pseudo-random rule proportion and preferred probability
Table 5 lists the effect of the pseudo-random rule proportion N0 on the total expected cost. The lower

RI
and higher pseudo-random rule proportions both lead the ACS-RSM algorithm to stagnate at local
optimal solutions. It is clear that NK = 0.7 is relatively appropriate for the auto alternator case. Here,
L, M and OK are fixed at 2, 1, and 0.7, respectively.

SC
Table 5 Effect of pseudo-random rule proportion NK

Experiment 1 2 3 4 5
NK

U
0.5 0.6 0.7 0.8 0.9
AN
Total Expected Cost 58,868 57,877 57,870 58,212 58,649

The effect of the preferred probability OK is shown in Table 6. It shows that OK = 0.6 is the most
suitable for the ACS-RSM algorithm when L, M and NK are fixed at 2, 1, and 0.7.
M

Table 6 Effect of preferred probability •‚


D

Experiment 1 2 3 4 5
OK 0.5 0.6 0.7 0.8 0.9
TE

Total Expected Cost 57,881 57,865 57,870 60,182 63,829

5.3 Optimal results and discussions


EP

5.3.1 Optimal results


To verify efficiency and effectiveness of the ACS-RSM algorithm, an Adaptive Genetic algorithm
C

(AGA) proposed by Jing et al. (2013) was used for comparison. The paper of Jing et al. (2013) is
closely similar to this study. Major parameters of the two algorithms are listed in Table 7.
AC

Table 7 Comparison ACS-RSM with AGA


Average Total Average
Algorithm Algorithm Parameter Values
Expected Cost Running Time
ACS-RSM Ant amount ' = 20, L = 2, M = 1, NK = 0.7, OK = 0.6, 57,865 167s
O; = 0.7, OQ = 0.6, R = 0.05, G = 1,000, H = 200

AGA Population size = 100, crossover probability = 0.6, 59,704 421s


penalty factor = 50, mutation probability = 0.01

14
ACCEPTED MANUSCRIPT
Each program ran independently 10 times and their average total expected cost and average
running time were calculated. As can be seen in Table 7, computational efficiency and solution quality
of the ACS-RSM algorithm significantly outperformed those of Adaptive Genetic algorithm. Using
the ACS-RSM algorithm, the optimal production plan of the auto alternator case was obtained, as
shown in Table 8. The total expected cost of the optimal production plan reached 57,865.

Table 8 Optimal production plan of the auto alternator HMRS

Period y

PT
Decision
Variables 1 2 4 4 5 6
$ 19 0 12 18 0 15
%

RI
1 0 1 1 0 1
$ 17 16 18 20 18 18

SC
1 1 1 1 1 1

5.3.2 Discussions
Recovery rate of used product and batch sizes of new and remanufactured products are two key

U
system parameters. They play important roles in reducing total expected cost of the stochastic HMRS.
AN
In order to find optimal system parameters for the auto alternator case, effects of the two key
parameters were conducted.
(1) Effect of the recovery rate of used product
M

Relationship between the recovery rate of use product and total expected cost of the stochastic
HMRS was illustrated in Fig 4.
D
TE
C EP
AC

Fig. 4 Effect of recovery rate on total expected cost of stochastic HMRS

An interesting phenomenon was found that the total expected cost rapidly decreased when the
recovery rate increased from 0.5 to 0.91. However, it was almost unchanged when the recovery
rate was greater than 0.91. There are two main reasons for this phenomenon. 1) When the recovery
rate rose from 0.5 to 0.91, the substitution cost and purchasing cost for remanufactured products

15
ACCEPTED MANUSCRIPT
from spot market reduced significantly, which caused a rapid decline of the total expected cost. 2)
When the recovery rate θ was greater than 0.91, the substitution and purchasing costs did decrease.
But the inventory holding cost of the remanufactured products increased because of high recovery rate
of returned product. So the total expected cost did not decrease dramatically as before but seemed
almost constant.
(2) Effect of batch size of the new and remanufactured products
Fig 5 represented the effect of batch sizes of the new and remanufactured products on the total

PT
expected cost and running time of the ACS-RSM algorithm. It was clear that with the increase of
batch sizes, the total expected cost of the stochastic HMRS had an obvious increasing trend, while run

RI
time dropped sharply. The main reason for this phenomenon was: The smaller the batch size was, the
closer manufacturing and remanufacturing quantities were to the real demands of the new and

SC
remanufactured products in each period. This contributed to lower inventory holding cost and
purchasing costs and also led to a sharp reduction in total expected cost. Unfortunately, smaller batch
sizes enlarged the feasible solution space, resulting in much longer running time.

U
AN
M
D
TE
EP

Fig. 5 Effect of batch sizes on total expected cost and running time
C

6. Management implications
AC

The above discussions provide a scientific basis for formulating two useful managerial implications
for senior leaders of remanufacturing enterprises.
(1) Recovery rate θ of used auto alternator should be controlled at 0.91 to obtain an ideal total
expected cost. Based on Figure 4, when recovery rate is lower than 0.91, remanufacturing
enterprises should raise it to 0.91 for obtaining an ideal total expected cost. This can be achieved by
various methods, such as adopting a new demolition method, performing the Product Design for
Remanufacturing (PDF) strategy, etc. However, it is noted that the recovery rate does not need to
be higher than 0.91 because of no obvious further reduction in the total expected cost. Typically,
16
ACCEPTED MANUSCRIPT
further improving recovery rate requires more advanced remanufacturing technologies and perhaps
additional equipment, which are translated into significant extra costs for the HMRS without bringing
in great benefits.
(2) Batch sizes of new and remanufactured products should be set to seven to reach the minimum
total expected cost. Determining optimal batch sizes of the new and remanufactured products is an
effective approach to reduce the total expected cost. According to Fig. 5, a useful business suggestion
can be made that optimal batch sizes should be set to seven to reach the minimum total expected cost.

PT
The running time of 170 seconds is acceptable by remanufacturing enterprise.

7. Conclusions and Future Research

RI
Production planning for the stochastic HMRS with downward substitution has been one of major
challenges faced by enterprises implementing hybrid production systems. The major contribution of

SC
this paper is to offer a useful decision-making tool to optimize such an HMRS. This paper constructed
a multi-period mixed integer stochastic programming model and proposed an ACS-RSM algorithm to
determine the optimal production plan and explored several specific business suggestions for the

U
future. An auto alternator HMRS in China was used as a case with the implementation of the
AN
ACS-RSM algorithm. Research results can help enterprises to formulate effective business measures
to improve the HMRS. The major findings of the practical case study include:
(1) ACS-RSM algorithm proposed by this paper performed well regarding computational efficiency
M

and solution quality.


(2) Recovery rate θ of used auto alternator should be controlled at 0.91 to obtain an ideal total
D

expected cost. The total expected cost rapidly declined when the recovery rate rose from 0.5 to
0.91. However, it nearly stayed unchanged when the recovery rate was greater than 0.91. This
TE

indicated that for remanufacturing enterprise, the recovery rate θ should be controlled at 0.91 to
obtain an ideal total expected cost. It is meaningless for manufacturers to further improve
recovery rate when the rate is already over 0.91.
EP

(3) Batch sizes of new and remanufactured products should be set to 7 to the minimum total expected
cost. With the increase of batch sizes, the total expected cost of the stochastic HMRS had an
C

obvious increasing trend, while run time dropped sharply. When batch sizes of new and
remanufactured products were set to 7, the total expected cost reached the minimum.
AC

This study still has some limitations. It mainly addressed production planning of a single product.
It does not involve production planning of multiple products, which may have totally different
optimization result. In addition, the proposed model does not consider the shortage option, even
though the option is pretty common in real life. Further researches are needed to resolve these
limitations.

Acknowledgment

This work is supported by grants from the National Natural Science Fund (71871117, 71273131,
71473085, 71671090), Humanities and Social Science Foundation of Ministry of Education of China
17
ACCEPTED MANUSCRIPT
(18YJA630066), Key Project of University Social Sciences Foundation of Jiangsu Province of China
(2017ZDIXM083), Aeronautical Science Fund of China (2017ZG52080).
Reference
Ahiska, S.S., Kurtul, E., 2014. Modeling and analysis of a product substitution strategy for a
stochastic manufacturing /remanufacturing system. Computers & Industrial Engineering, 72, 1-11.
Ahiska, S.S., Gocer, F., King, R.E., 2017. Heuristic inventory policies for a hybrid
manufacturing/remanufacturing system with product substitution. Computers & Industrial

PT
Engineering, 114, 206-222.
Cai, X., Lai, M., Li, X., Li, Y., Wu, X., 2014. Optimal acquisition and production policy in a hybrid
manufacturing/remanufacturing system with core acquisition at different quality levels. European

RI
Journal of Operational Research, 233(2), 374-382.
Chang, X., Xia, H., Zhu, H., Fan, T., Zhao, H., 2015. Production decisions in a hybrid manufacturing

SC
-remanufacturing system with carbon cap and trade mechanism. International Journal of Production
Economics, 162, 160-173.
Deroussi, L., Fonseca, J.B.D., 2009. A hybrid Ant Colony System for machine assignment problem in

U
flexible manufacturing systems. International Conference on Computers & Industrial Engineering,
AN
Troyes, France, 205-210.
Du, B., Guo, S., 2016. Production planning conflict resolution of complex product system in group
manufacturing: A novel hybrid approach using ant colony optimization and Shapley value. Computers
M

& Industrial Engineering, 94, 158-169.


Dorigo, M., Bonabeaub, E., Theraulaz, G., 2000. Ant algorithms and stigmergy. Future Generation
D

Computer Systems, 16(8), 851–871.


Dorigo, M., Stutzle, T., 2004. Ant Colony Optimization. The MIT Press, London, England.
TE

Fazlollahtabar, H., Mahdav, I., 2009. Applying stochastic programming for optimizing production
time and cost in an automated manufacturing system. International Conference on Computers &
Industrial Engineering, Troyes, France, 1226 – 1230.
EP

Francie, K.A., Jean-Pierre, K., Pierre, D., Victor, S., Vladimir, P., 2015. Stochastic models and
numerical solutions for manufacturing/remanufacturing systems with applications to the printer
C

cartridge industry. Journal of Manufacturing Systems, 37(3), 662-671.


Giglio, D., Paolucci, M., Roshani, A., 2017. Integrated lot sizing and energy-efficient job shop
AC

scheduling problem in manufacturing/remanufacturing systems. Journal of Cleaner Production, 148(1),


624-641.
Govindan, K., Shankar, K.M., Kannan, D., 2016. Application of fuzzy analytic network process for
barrier evaluation in automotive parts remanufacturing towards cleaner production - a study in an
Indian scenario. Journal of Cleaner Production, 114, 199-2
Goel, R., Maini, R., 2018. A hybrid of ant colony and firefly algorithms (HAFA) for solving vehicle
routing problems. Journal of Computational Science, 25, 28-37.
Hauser, W.M., Lund, R.T., 2003. The remanufacturing industry: Anatomy of a giant: a view of
18
ACCEPTED MANUSCRIPT
remanufacturing in America based on a comprehensive survey across the industry. Report, Boston
University, Boston, US.
Huang, R.H., Yang, C.L., 2008. Overlapping production scheduling planning with multiple objectives
- An ant colony approach. International Journal of Production Economics, 115(1), 163-170.
Inderfurth, K., 2004. Optimal policies in hybrid manufacturing/remanufacturing systems with product
substitution. International Journal of Production Economics, 90(3), 325-343.
Jing Y., Wang X., Li W.C., Deng L., 2013. Hybrid lot sizing production planning of manufacturing

PT
remanufacturing to automobile engine. Computer Integrated Manufacturing Systems, 19(4), 774-781.
Kouedeu, A.F., Kenne, J.P., Dejax, P., Songmene, V., Polotski, V., 2013. Analytical modeling and

RI
control of a hybrid manufacturing/ remanufacturing system. IFAC Proceedings, 46(9), 401-406.
Kiesmüller, G.P., 2003. A new approach for controlling a hybrid stochastic manufacturing
/remanufacturing system with inventories and different lead times. European Journal of Operational

SC
Research, 147(1), 62-71.
Kenné, J.P., Dejax, P., Gharbi, A., 2012. Production planning of a hybrid manufacturing/

U
remanufacturing system under uncertainty within a closed-loop supply chain. International Journal of
Production Economics, 135(1), 81-93.
AN
Kalayci, C.B., Kaya, C., 2016. An ant colony system empowered variable neighborhood search
algorithm for the vehicle routing problem with simultaneous pickup and delivery. Expert Systems
with Applications, 66, 163-175.
M

Liu, W., Wu, C., Chang, X., Chen, Y., Liu, S., 2017. Evaluating remanufacturing industry of China
using an improved grey fixed weight clustering method-a case of Jiangsu Province. Journal of cleaner
D

production,142, 2006-2020.
Van der Laan, E., Salomon, M., Dekker, R., 1999a. An investigation of lead-time effects in
TE

manufacturing/remanufacturing system under simple PUSH and PULL control strategies. European
Journal of Operational Research, 1999, 115(1),195- 214.
EP

Van der Laan, E., Salomon, M., Dekker, R., Wassenhove, L.V., 1999b. Inventory control in hybrid
systems with remanufacturing. Management Science, 45(5), 733-747.
Leung, C.W., Wong, T.N., Mak, K.L., Fung, R.Y.K., 2010. Integrated process planning and scheduling
C

by an agent - based ant colony optimization. Computers & Industrial Engineering, 59 (1), 166-180.
AC

Polotski, V., Kenne, J.P., Gharbi, A., 2015. Optimal production scheduling for hybrid manufacturing/
remanufacturing systems with setups. Journal of Manufacturing Systems, 37(3), 703-714.
Polotski, V., Kenne, J.P., Gharbi, A., 2017. Production and setup policy optimization for hybrid
manufacturing/remanufacturing systems. International Journal of Production Economics, 183(B),
322-333.
Rathore, P., Kota, S., Chakrabarti, A., 2011. Sustainability through remanufacturing in India: a case
study on mobile handsets. Journal of Cleaner Production, 19 (15), 1709-1722.
Roshani, A., Giglio, D., Paolucci, M., 2017. A relax-and-fix heuristic approach for the capacitated
dynamic lot sizing problem in integrated manufacturing/remanufacturing systems.
19
ACCEPTED MANUSCRIPT
IFAC-PapersOnLine, 50(1), 9008-9013.
Sadok, T., Nidhal, R., 2016. Optimal production control of a manufacturing-remanufacturing system
with delivery activity. IFAC-PapersOnLine, 49(12), 1233-1238.
Schyns, M., 2015. An ant colony system for responsive dynamic vehicle routing. European Journal of
Operational Research, 245 (3), 704-718.
Shapiro, A., 2003. Monte Carlo Sampling Methods. Handbooks in Operations Research and
Management Science, 10, 353-425.

PT
Shishvan, M.S., Sattarvand, J., 2015. Long term production planning of open pit mines by ant colony
optimization. European Journal of Operational Research, 240 (3), 825-836.

RI
Tang, O., Grubbström, R.W., 2005. Considering stochastic lead times in a
manufacturing/remanufacturing system with deterministic demands and returns. International Journal
of Production Economics, 93-94, 285-300.

SC
Wang, Y., Chen, W., Liu, B., 2017. Manufacturing/remanufacturing decisions for a capital-constrained
manufacturer considering carbon emission cap and trade. Journal of Cleaner Production, 140(3),

U
1118-1128.
Wang, H., Rong, Y., Xiang, D., 2014. Mechanical assembly planning using ant colony optimization.
AN
Computer Aided Design, 47(2), 59-71.
Zouadi, T., Yalaoui, A., Reghioui, M., Kadiri, K.E.E., 2016. Hybrid manufacturing/remanufacturing
lot-sizing problem with returns supplier’s selection under carbon emissions constraint.
M

IFAC-PapersOnLine, 49(12), 1773-1778.


D
TE
C EP
AC

20
ACCEPTED MANUSCRIPT
Appendix. Ant colony system algorithm

Ant colony system (ACS) algorithm is a meta-heuristic algorithm simulating ant behaviors. In the
ACS algorithm, encoding and decoding methods of this algorithm have a great influence on the
correctness and efficiency of the result. In addition, state transition rule and pheromone updating
strategy are two key strategies for forming the ACS algorithm of this article.
A.1 Encoding and decoding of ACS algorithm

PT
The encoding is in essence a solution description for production planning of the HMRS. In the
ACS algorithm of this article, integer encoding was adopted with an aim of being understood easily.
$; , %; ; $; , %; , $Q , %Q ; $Q , %Q , … , $ , % ; $ , % , … , $9 , %9 ; $9 , %9 was used to represent
a production plan of the HMRS for planning horizon . The element $ , % ; $ , % showed the

RI
values of decision variables $ , % , $ and % in the period , where number of manufacturing
batch $ and number of remanufacturing batch $

SC
had lower and upper bounds determined by the
following formula (A3), and % and % were binary variable. The decoding of the ACS algorithm

U
was described as below: according to obtained production planning of the HMRS ,
$; , %; ; $; , %; , $Q , %Q ; $Q , %Q , … , $9 , %9 ; $9 , %9
AN
, the total expected cost of the HMRS is
calculated.

A.2 Two key strategies of the ACS algorithm


M

A.2.1 State transition rule


In the initial phase of ACS algorithm, a number of ants are randomly placed at different positions in
period , corresponding to different combinations of batch numbers $1 , $; . In the ACS algorithm,
D

any ant e randomly moves from current position ƒ in the period − 1 (namely node „Zf −1 ) to
TE

another not yet visited position in the period (namely node „ f ), shown in Fig. A1. Here,
∈ Z
f
is the feasible position scope. This process is called stochastic state transition.
EP

ν ik ( t −1)
C

(xN
t −1 , xtR−1 )
AC

ν kj ( t )

(xN
1 , x1R ) (xt
N
, xtR )

Fig. A1 Schematic diagram of state transition for ant k


In order to avoid trapping into local optimal solution and improve computational efficiency and
solution quality of the ACS algorithm, ant k uses a pseudorandom proportional rule to choose next
21
ACCEPTED MANUSCRIPT
position , showed in following formula (A1). It is notedthat j in the formula (A1) only represents the
probability of ant k at position i choosing next position j.

arg max /JZ ‹


, ŒZ •
1 N ≤ NK
= … ∈ ‰Š
Ž N > NK .
(A1)

Here, N is a random variable subjected to a uniform distribution supported by 0 and 1. NK is a


deterministic parameter in •0,1‘, called pseudo-random rule proportion. JZ is pheromone when
ant e at position ƒ in period − 1 selects position = 1⁄,f is heuristic

PT
in the period . ŒZ
information when position is selected in period . ,f is the total cost of ant e in period . L is

RI
the pheromone trail factor determining the influence of pheromone trail while M is the heuristic
information factor representing influence of the heuristic information. Ž is a variable reflecting the
probability of ant e choosing position , generated by the roulette wheel selection method and is

SC
calculated by formula (A2).

ŒZ •
J ‹
∗ ∈ f
Z
f
=. Z
0
∑ •∈ Š

JZ• ‹

U
∗ ŒZ• • Z

Otherwise.
(A2)
AN
Here, Z
f
is the transition probability by which ant e at position ƒ in period − 1 selects
f
is the feasible position scope. For combined variable of batch numbers
M

position in period , Z

$ ,$ in the feasible position set Z


f
, š$ , $ › and š$ , $ › are the ranges for manufacturing
batch number $ and remanufacturing batch number $ , respectively, and have impact on
D
TE

computational efficiency of the ACS algorithm. They are determined by formula (A3).
$ ∈ šO; $ , $ ›
. œ•O ≥ 0
$ ∈ šOQ $ , $ ›
EP

"
¡$ ∈ š$ , $ › = ¢0, £
Ÿ #
(A3)

ϥO < 0
* )<; + − ! ) * )<; + "
C

Ÿ$ ∈ š$ , $ › = ¢max ¤ , 0¥ , min ¤ , ¥£
ž # / # / #
AC

Note that O; and OQ are both deterministic and between•0,1‘. Here, " /# and " /# are
respectively maximum numbers of manufacturing and remanufacturing batches in each period.
O; and OQ are the select value proportion determined according to decision maker’s experience.
œ•O = ∑9¦: / ¦ − " 1 is accumulated deficiency between demand of remanufactured products and
remanufacturing capacity from period to period . œ•O ≥ 0 indicates that there will be a
shortage of remanufactured products. To avoid purchasing from the spot market, the ranges of
$ and $ are suggested to take upper limits with a high preferred probability OK . If ϥO < 0, the
22
ACCEPTED MANUSCRIPT
range of š$ , $ › for $ depends on inventory in period − 1, return amount, recovery rate of used

products, production capacities, and inventory capacities of manufacturing and remanufacturing.

A.2.2 Pheromone updating strategy


In the ACS algorithm, two activities including pheromone evaporation and pheromone deposition
manipulate initial pheromone value. The former is an activity of reducing pheromone helping the ACS
module to disregard bad solutions. The latter is an activity of adding additional pheromone

PT
contributing to construction of new production plans (Shishvan and Sattarvand, 2015). Proper
updating of pheromone trail value determines convergence speed of the ACS algorithm and quality of
the solutions found. Pheromone updating strategies consist of local pheromone updating and global

RI
pheromone updating.
1) Local pheromone updating: Any ant e will perform the local pheromone update strategy
immediately after a production plan is constructed. The value of local pheromone trial from position ƒ

SC
to position in the period is calculated by formula (A4).
JZf = 1 − R JZf + R JK 0 < R < 1.

U
(A4)
Here, R is the pheromone evaporation rate, JK is the initial value of pheromone trial, and JZf is
AN
the pheromone value on the path from position ƒ to position for ant e.
2) Global pheromone update: In the ACS algorithm, only the best-so-far ant is permitted to
M

increase pheromones after all iterations. The value of global pheromone trial is given by formula
(A5).
JZ = 1 − R JZ + R∆JZ¨© , ∀ ƒ, ∈ ¨©
D

(A5)
where
1/ , ¨© ∀ ƒ, ∈
TE

¨©
∆JZ¨© = ¤
0 Otherwise.
¨©
represents the best-so-far production plan; ∆JZ¨© is the pheromone deposition on arc ƒ, of
EP

Here,
the best-so-far production plan; and , ¨© is the total cost of the best-so-far production plan.
C

A.3 Steps of the ACS algorithm


AC

The flow of the ACS algorithm is illustrated in Fig. A2. The detailed steps are as follows.
Step 1: A deterministic problem of production plan for the HMRS is first inducted. The initialized
parameters divide into two types. one is algorithm parameters, which includes population size
of ants z, initial pheromone value τ0 , pheromone trail factor L, heuristic information factor M,

pseudo-random rule proportionNK , select probabilities O; and OQ , pheromone evaporation


rate R, algorithm stop criterion S; (a minimum value) and maximum iteration X. Others are
system parameters in Section 3.1.
Step 2: In the period 1, C ants are randomly assigned with number of manufacturing and
23
ACCEPTED MANUSCRIPT
remanufacturing batch $; and $; , which are in the ranges of š$ , $ › and š$ , $ › that are

calculated based on (A3). Any ant e constructs its path sequence (corresponding to an
th
optimal production plan) by formula (A1-A2) at the x iteration.
Step 3: After ant e constructs a path sequence, local pheromone update is performed by formula
(A4).
Step 4: The ant with the minimum total cost is identified as the best-so-far ant. Its path sequence is the
, ¨© .

PT
best-so-far production plan and its total cost is recorded as the best-so-far total cost
The global pheromone update is performed by using formula (A5).
Step 5: If ] ,YZ[
^_`
− ,YZ[
«
] < S; or $ < ¬, go to step 6; Otherwise, go to Step 2.

RI
Step 6: Output the optimal production plan for the deterministic problem of the HMRS and record its
total cost.

U SC
AN
M
D
TE

ACS
|TCmin x
-TCmin |<ε 1 or x = X
ACS
TCmin x
=TCmin,x = x + 1
C EP
AC

Fig. A2 Flow of ACS algorithm for the deterministic HMRS

24
ACCEPTED MANUSCRIPT
Highlights

• A multi-period mixed integer stochastic programming model for the hybrid


manufacturing/remanufacturing system (HMRS) with stochastic demand and return and
with downward substitution.
• Development of an ant colony system algorithm combined with the random sampling
method to determine optimal production plans for the stochastic HMRS.
• Business measures for manufacturing/remanufacturing enterprises to coordinate

PT
manufacturing and remanufacturing
• A real-world case to validate the model and algorithm and to provide managerial insights.

RI
U SC
AN
M
D
TE
C EP
AC

You might also like