A Multi Objective Optimization of Multi Period Multi Product Closed Loop Supply Chain

Abhinav Kumar Sharma, S. Santhosh, B. Shahul Hamid Khan1

Department of Mechanical Engineering
Indian Institute of Information Technology Design and Manufacturing Kancheepuram
Chennai, Tamilnadu, India
Email- mds13m001@iiitdm.ac.in , mdm13d001@iiitdm.ac.in and bshahul@iiitdm.ac.in

ABSTRACT: A Supply Chain Network (SCN) consists of suppliers, manufacturing plants,

warehouses and customers. These all work in an order to fulfil the customer requirements. Here
customer requirement is getting the product as soon as possible. An SCN is designed in such a way that
it meets the customer requirement inefficient and the effective way. But these are associated with the
cost. This cost is calculated on the basis of various parameters like inventory management,
production management, transportation, etc. Hence it becomes necessary to have an SCN which has less
operational cost and simultaneously meets the customer requirements. Remanufacturing can save the
raw material cost to the company. In this way the industry can enjoy more profit. The inventory carrying
cost is one of the major player in this scenario. Optimum inventory levels are required so as to balance
between satisfying the demand and low inventory holding cost. Some products can be remanufactured
and can be sold in the market. But some products which cant be remanufactured are disposed off.
In this paper a Supply Chain is considered. It is a multi-echelon supply chain network which
consists of, manufacturing plant, warehouse, and retailers. Raw material supplier delivers the raw
material to the manufacturing plant. The manufacturing plant processes the raw material, turns them
into finished products and sends to the warehouses from where it is sent to the retailers. Here retailers
send these items to their customers. The objective of the model is to minimise the total cost.This paper
deals with the optimization of the model with Particle Swarm Optimization and comparing the results
with the GA and CPLEX.

Keywords: Heuristics, Remanufacturing, Multi Period, Multi Product, Multi Echelon, Shortages.
INTRODUCTION
A supply chain consists of all parties involved, in fulfilling a customer request. The supply
chain includes not only the manufacturer and suppliers, but also transporters, warehouses, retailers,
and even customers themselves. Within each organization, such as a manufacturer, the supply chain
includes all functions involved in receiving and filling a customer request. These functions include,
but are not limited to, new product development, marketing, operations, distribution, finance, and
customer service. Consider a customer walking into a store to purchase a product. The supply chain
begins with the customer and his or her need for product. The next stage of this supply chain is the
retail store that the customer visits. Store stocks its shelves using inventory that may have been supplied
to it. The distributor in turn is stocked by the manufacturer. The manufacturing plant receives raw
material from a variety of suppliers, who may themselves have been supplied by lower-tier suppliers.
The supply chain activities constitute a mega process and lots of decision involved in their successful
design and the operation. In recent years the supply chain designing problems have been gaining
importance due to market competition. Design of the SCN is influenced by its warehouse location,
plant location, retailer and customer location. Parameters like inventory, production rate, transported
quantities etc. have to be decided accordingly. SCN network has its own constraints too. It may be
supplier constraints, production constraints and inventory constraints etc. So the variables has to be
fine tune in order to satisfy the constraints. Subramanian Pahang, N. Ramkumar, T.T. Narendran & K.
Ganesh (2013) [7] addressed about decision making in (1) location/operating decisions for warehouses,
hybrid facilities and manufacturing facilities and (2) production and distribution of products between

stages in the supply chain. Goal programming models and compromise programming technique
_________________
Corresponding Author

were used to solve the problem. Fan Wang, Xiaofan Lai & Ning Shi (2012) [1], modelled a supply chain
network problem with the environmental concers. They made decision regarding the environmental
concern with in the design stage only and proposed a multi objective optimization model which will
capture the trade-off between the total cost and the environmental degradation. Successful
implementation of the model can be done in planning the green supply chain management. Su-YolLee,
RobertD.Klassen, AndreaFurlan, AndreaVinelli (2014) [6] took into the consideration, the bull whip
effect in the supply chain. The bullwhip effect is a critical measure that amplifies the demand variability
as customer order flows in the chain. In order to minimize the inventory holding cost and to fulfil the
order in time the supply chain constraints should be perfectly decided so that this kind of situations can
be avoided in future. There are also conditions of uncertainty in some supply chain models. In this case
supply chain decisions may vary from time to time. Even a decision taken for the particular stage of the
supply chain which may seems to be feasible, may prove wrong under this conditions. One of the
example is backorder items.
From above it is obvious that the inventory management and fulfilment of the customer demand
is an issue of prime concern. It can be put in this way, the concern is logistic which comprises of storage
transportation and flow of good. Furthermore, in recent years, one of the most important developments
in manufacturing systems has been the Just-in-Time (JIT) system increases productivity. The JIT
approach was initially used by and at Toyota. JIT has been used by many companies. Industries want to
reduce the inventory cost because it is again of prime concern. So JIT finds application in this situations.
As soon as the manufacturer gets the order, it starts manufacturing it and dispatched it as soon as
possible. Similarly distributor asks for inventory when required and dispatches it when as soon as it
receives inventory. Most of the classical production models assume that all goods generated by
production machines are characterised by perfect quality. In reality, imperfect goods production is quite
normal in manufacturers. Therefore, the costs of defective goods are considered essential in supply chain
optimisation. In most cases, reworking of defective goods can be quite profitable.
The model considered here is taken from work of Salah Alden Ghasimi, Rizauddin Ramli,
Nizaroyani Saibani [5]. In this paper the results are optimized using the Particle Swarm Optimization
and Genetic Algorithm. Then the results are compared with the results of GA and the CPLEX solver.
PSO has lots of advantages. It was developed by James Kennedy and Russell Eberhart [8] in
1996 which mimics the movement of the birds for the search for food. The birds will follow the leader
and will continuously communicate within themselves. Eventually they will reach their destination at
last. So PSO mimics the same behaviour. One of the advantage of PSO over GA is that it remembers
the position of the particles which is very important for and algorithm.

PROBLEM DESCRIPTION

Fig 1. The model adopted from [5]

The problem shown is a multi-echelon, multi-period and multi product model. At

manufacturer some of the goods produced are not perfect and hence discarded. Now these discarded
goods can either be remanufactured or can be considered as scrap depending upon the economics
of the remanufacturing. Then the perfect goods are sent to the distributor and further to the retailers.
From retailers customer buy these goods which is the end of supply chain. In order to maintain the
minimum level of inventory goods are built to order. As soon as the demand arises, the production
starts. Once the enough quantities are produced, items are sent to the distributors for further
distribution.

Assumptions in ModelFollowing are the assumptions considered in the model1. The amount of demand was assigned to the manufacturers at the beginning of the period.
2. The duration of each period was equal to the sum of the production and rework times.
3. Retailer shortage is allowed.
4. The model is designed for multiple manufacturers, distributors, retailers, products and multiperiods.
5. The inspection process for the products is perfect, and the inspection time is zero.
6. It is assumed that all imperfect quality products are not repairable, whereas those considered as
scraps.
7. The locations of plants, distributors, retailers and suppliers are fixed.
8. The capacities of the manufacturers, the distributors and the retailers are known.
9. Products are temporarily stored at the distributors before delivery to the retailers.
10. There is no inventory at the distributors at the beginning or end of the planning horizon.
11. The capacities of the manufacturer, distributor and retailers are limited.
12. The storage capacities for each perfect product are limited.
13. Store capacities and allocated storage capacities for defective goods are limited.
14. All demands must be satisfied during the planning horizon.
15. The production and reworking times are limited.

MATHEMATICAL MODELLING
Set of indices are,
Manufacturing Plant - 1, 2, 3.M
Warehouse - 1, 2, 3.W
Retailer - 1, 2, 3.R
Product - 1, 2, 3.P
Time Period - 1, 2, 3.T

Parameters:

pempt Percentage of the defective goods p produced by manufacturer m during period t

Rmpt Rework cost per defective good p by manufacturer m during period t
hwpt Holding cost of product p at warehouse w during period t
h 'mpt Holding cost of product p for the defective goods stored at manufacturere m during

mpt

period t
Time required to produce goods p by manufacturer m during period t

T t Total production time during period t

R mpt Rework time required for the goods p by manufacturer m during t

R t Total rework time during period t

Cmpt Production cost per unit item by manufacturer m during period t
CTmwpt Shipping cost of each product p from manufacturer m to warehouse w during period t
CT ' wcpt Shipping cost of each product p from warehouse w to retailer c during period t
CT "mpt Shipping cost of each defective goods at manufacturer m during period t

mpt Percentage of scrap products l produced by by manufacturer m during period t

mpt Discount cost per scrap goods l for sale by manufacturer m during period t
cpt Shortage cost of each product p at retailer w during period t
CI t Indirect cost during period t
In pwt Inventory of the product p at warehouse w during period t
demcpt Demand of retailer c for product p during period t
QCmpt Production capacity of manufacturer m for product p during period t
Qwt Total storage capacity of warehouse w during period t
Q 'ct Total storage capacity of retailer k during period t
Following are the variables in the Model:

QTmwpt Amount of products p transported by factory m to warehouse w during period t

QPmpt EPQ of products p by manufacturer m during period t
Def mpt Amount of the defective goods p produced by the manufacturer m during period t
SQmpt Amount of scrap goods p produced by manufacturer m during period t
QPPmpt Amount of perfect products p produced by manufacturer m during t before rework
TQPPmpt Amount of perfect products p produced by manufacturer m during t after rework
QT ' wcpt Amount of products p transported from warehouse w to retailer c during period t
SPcpt Amount of shortage of product p in retailer c during period t
Tt ALOEC according to importance of lead time by customer in period t
The objective function is as follows:
M P T
W P T
t W
t M
M P T
Z min Cmpt QPmpt hwpt X mwp Ywcp h 'mpt Def mpt
m 1 p 1 t 1
w 1 p 1 t 1
1 w1
1 m 1
m1 p 1 t 1
M

R
m 1 p 1 t 1
M

mpt Def mpt CTmwpt QTmwpt CT ' wcpt QT ' wcpt

m 1 w 1 p 1 c 1
M

w 1 c 1 p 1 t 1

CT "mpt Def mpt mpt SQmpt SPcpt cpt CI t Tt

m 1 p 1 t 1

m 1 p 1 t 1

c 1 p 1 t 1

t 1

Constraints are as follows:

QPmpt QCmpt
M

QT

Qwt

mwpt

m 1
W

QT '
w 1

m, p, t

wcpt

(1)

w, p, t

Q 'ct

(2)

c, p , t

Def mpt pempt QPmpt

(3)

m, p , t

SQmpt mpt Def mpt

(4)

m, p , t

QPPmpt (1 pempt ) QPmpt

(5)

m, p , t

TQPPmpt QPPmpt Def mpt SQmpt

C

dem

cpt

c 1

m 1

m 1

SQmpt QPmpt

c 1

m 1

demcpt TQPPmpt
M

m 1

c 1

p, t

c 1

c 1

m 1

wcpt

demcpt

(10)

p, t

(12)
w, p, t T

m 1 1

w 1 1

m 1 1
W

mwp

m 1 p 1
M

mpt

TQPPmpt

p, l

(15)

m, p , t

(16)

(17)

QPmpt T t

R
m 1 p 1

(14)

w 1 1

QT
M

w, p, t T

QTmwp QT 'wcp
w 1

(13)

m 1 1

QTmwp QT 'wcp Inljt

M

(11)

c, p , t

QT 'wcp QTmwp
c 1 1

(8)
(9)

SPcpt demcpt QPPmpt

w 1

(7)

p, t

m, p , t

p, t

QPPmpt demcpt

QT '

(6)

mpt

Def mpt Rt

T t Rt Tt

(18)

QPPmpt , TQPPmpt , QPmpt , SPcpt , Def mpt , SQmpt , QT ' wcpt , QTmwpt 0

(19)
m, w, c, p, t

(20)

The constraint (1) restrict the production capacity for the manufacturing plant. Constraint (2) & (3)
deals with the delivery capacity constraints for warehouse and the retailers. Constraint (4) represent
the amount of defective goods produced. Constraint (5) represent the amount of scrap goods from

the defective goods. Constraint (6) represent amount of perfect good before reworking and (7)
represent the amount of good after reworking of the defective goods. Constraint (8) state the total
production is the addition of the demand and the scrap goods. Constraint (9) shows the total demand
is satisfied by the perfect goods only and no defective goods or scrap reaches the customer.
Constraint (10) & (11) are for the shortages allowed at the retailer. Constraint (12) denotes the
demand is satisfied at retailer by the quantities transported to the retailer by the warehouse.
Constraint (13) & (14) are the inventory constraints. If these are satisfied then only the model
follows the JIT logistics. Constraint (15) is balance between incoming and the outgoing at the
warehouse. Constraint (16) states that the total goods supplied from the manufacturer to warehouse
are only the perfect ones. Constraints (17), (18) and (19) are the time constraints. Constraint (20) is
the non-negativity constraint.

SOLUTION METHODOLOGY
The model is solved by using CPLEX, GA and PSO and then the results are compared. PSO gave
better results as compared to the GA. First approach is to get the exact solution by CPLEX and then
optimizing the problem with GA and PSO. PSO has got advantages over GA algorithm like saving
the best solution from all the iterations. Both the algorithms aims for the global optima. Flowchart
for the algorithm is shown in the figure. The PSO starts with the generation of the initial population.
It is generated randomly in the solution space. After this fitness function which is nothing but the
objective function is calculated for the each population. Then best of the objective function fitness
value is calculated based on the condition whether it is maximization problem or the minimization
problem. Then the deviation of the other population from global best is calculated with the help of
following formulas.

V (i) V (i) ( c1* rand (0,1) * (pbest(i) present(i))) ( c2 * rand (0,1) * (xbest(i) present(i)))
X (i) X(i 1) V (i)

Figure 2.Flow chart for PSO

Figure 3 Flow chart for GA

The GA also starts with the initial population generation. But the way population is updated is different
from the PSO. In GA the genetic operator, crossover is done to update the population. The population
is updated and then it is further updated, depending upon the stopping criteria. To speed up the process
mutation was also performed in the algorithm with some probability.
The initial population in both GA and PSO was generated based upon the constraints (1) and
(8). It was produced randomly and then fitness function Zmin was calculated for each population. In case
of PSO the random velocity and random position was allocated to each and every particle. After this
using the relations for PSO the each and every individual of the population was updated. The value of
c1 and c2 was kept 2. In case of the GA, the population was updated by using crossover operator. To
speed up we used mutation operator. These methodologies were done on MATLAB R2013a
(8.1.0.604) running on system with Windows 8.1, core I7 processor 3.5 GHz and 8 GB RAM. The
comparison between the methods is enlisted in the following table.

Table 1. Description of Sample Problems

Sample
Problem

Manufacturer

Warehouses

Retailers

Products

Time Period

Table 2. Parameters and Values

Parameters

Value

pempt

Uniform (0,0.09)

QPmpt

Normal (15,3)

Rmpt

Normal (6,2)

hmpt

Normal (6,2)

hwpt

Normal (10,3)

mpt

Normal (6,3)

Rmpt

Normal (5,2)

CTmwpt

Normal (12,3)

CTwcpt

Normal (8,2)

CTmpt

Normal (5,2)

mpt

Normal (11,3)

mpt

Uniform (0,0.9)

cpt

Normal (6,1)

demcpt

Normal (2000,1000)

CIt

Normal (0,0.6)

Table 3. Comparison of Results between CPLEX, GA and PSO

Sample
Problem

CPLEX
Zmin

Run time (sec)

GA
Zmin

PSO

Run time (sec)

Zmin

Run time (sec)

339526

0.007

340864

118

340840

102

785294

0.011

804349

197

793300

184

512837

0.009

525108

159

517876

147

483710

0.007

489579

143

484889

134

727192

0.011

741740

186

728490

177

985961

0.013

989458

211

988401

203

870296

0.013

879472

205

874121

193

1391027

0.016

1651802

243

1449190

229

Table 4. Deviation from Results of CPLEX for GA and PSO

Sample
Problem

Gap% GA

Gap% PSO

0.39

0.38

2.42

1.01

2.39

0.98

1.21

0.24

2.00

0.17

0.35

0.2

1.05

0.44

2.28

0.51

CONCLUSION
In this paper we proposed the GA and PSO for solving the supply chain network. The aim was to
minimize the total cost using most appropriate algorithm and simultaneously to meet all the constraints.
Inventory level were supposed to be zero so as to justify the JIT logistics. From results it is clear that
the PSO gave better results as compared to the GA for the given problems. A table is also presented to
give the gap between the CPLEX & GA solution and CPLEX and PSO solution. It can be concluded
that in a model like this the heuristics methods can be applied and can meet the requirements. This
problem can be converted to the multi-objective problem in future and can be solved further.

REFERENCES
[1] Fan Wang, Xiaofan Lai, Ning Shi (2010). A Multi-Objective Optimization for Green Supply Chain
Network Design, Decision Support Systems, Volume: 51, pages: 262-269.
[2] Fulya Altiparmak, Mitsuo Gen, Lin Lin, Turan Paksoy (2006). A Genetic Algorithm Approach For
Multi-Objective Optimization of Supply Chain Networks. Computers & Industrial Engineering,
Volume 51, pages: 196215.
[3] Kalyanmoy Deb, Multi-Objective Optimization Using Evolutionary Algorithms, Wiley India Pvt.
Ptd.
[4] Rajeshwar S. Kadadevaramath, Jason C.H. Chen, B. Latha Shankar, K. Rameshkumar (2012).
Application of Particle Swarm Intelligence Algorithms in Supply Chain Network Architecture
Optimization, Expert Systems with Applications, Volume 39, pages: 1016010176.
[5] Salah Alden Ghasimi, Rizauddin Ramli, Nizaroyani Saibani (2013). A Genetic Algorithm for
Optimizing Defective Goods Supply Chain Costs using JIT logistics and Each-Cycle Lengths. Applied
Mathematical Modelling, Volume 38, pages: 1537-1547.
[6] Su-YolLee, RobertD.Klassen, AndreaFurlan, AndreaVinelli (2014). The Green Bullwhip Effect:
Transferring Environmental Requirements along a Supply Chain, International Journal of Production
Economics, Volume: 156, pages: 39-51.
[7] Subramanian Pazhani , N. Ramkumar , T.T. Narendran & K. Ganesh (2013). A Bi-Objective
Network Design Model for Multi-Period, Multi-Product Closed-Loop Supply Chain, Journal of
Industrial and Production Engineering, Volume 30:4, pages: 264-280.