You are on page 1of 10

A model for integrated production planning in cellular

manufacturing systems
Mingyuan Chen
Department of Mechanical Engineering, Concordia University, Montreal, Canada
Introduction
Various manufacturing production planning
and inventory control problems have been
studied extensively by many production
management researchers. Different models
and methods developed to solve these
problems can be found in widely used
textbooks of production engineering or
manufacturing systems (Riggs, 1981; Singh,
1996). Inventory control models from simple
EOQ to more complicated MRP, kanban and
CONWIP (Monden, 1983; Spearman and
Zazanis, 1992) models have been developed
and widely used in today's manufacturing
industry. Some of them are very successful in
practical applications. Mathematical
programming is also a powerful tool for
solving complicated production planning
problems, when product structures with
multi-stages and multi-components are
considered. Review and discussion on
mathematical programming models for
certain kanban and MRP systems can be
found (Price et al., 1994). Other mathematical
programming models for MRP or kanban-
based production planning have also been
developed (Bard and Golany, 1991; Bitran and
Chang, 1987; Price et al., 1995). Many of the
mathematical models and solution methods
were developed to solve problems in general
manufacturing or service industries and can
be widely applied. Prominent manufacturing
features such as production flexibility and
manufacturing cell formation were usually
not considered in developing production
planning models. On the other hand,
manufacturing systems analysis tends to
study more specific system characters such
as job sequencing and scheduling, alternative
process plans and different ways of forming
manufacturing cells as well as handling
production materials and tools (for example,
Burbidge, 1989; Chen and Cheng, 1995; Jamal,
1993). As pointed out in Arvindh and Irani
(1994), an integrated approach should be
pursued in manufacturing system analysis,
since different aspects of a system are inter-
related in many ways. In addition, a
comprehensive model consisting of different
aspects of the system can help one to
understand the problem better. Integrated
system approaches can minimize the
possibility of certain important aspects of the
system being overlooked, while other issues
are being studied. On the other hand,
integrated models may be difficult to solve
optimally and efficiently with today's
computational power available to general
practitioners and they may have to rely on
heuristics to solve real life practical
problems. However, these models will soon
become more solvable in the near future with
the fast progress of computer technology in
hardware and software. Based on the above
considerations, we used an integrated
approach to study cellular manufacturing
production planning problems in this paper.
Optimal solution of the integrated model
gives the best part families and machine
cells. In addition, it also provides the best
production plan to minimize inventory and
machine set-up costs. In fact, this model is an
extension of the model discussed in Chen
(1998) where production quantities were
assumed known parameters, not decision
variables. In that model, these known values
and other system features were used for
manufacturing cell formations and
reconfigurations. Other similar studies can
be found in Harhalakis et al. (1990); Vakharia
and Kaku (1993); Dahel (1995); Atmani et al.
(1995), where cell formations and system
reconfigurations were discussed. We found
very few articles in literature on planning
cellular manufacturing production. The
survey (Olorunniwo, 1996), conducted at a
number of companies in the USA,
demonstrated how the implementations of
cellular manufacturing systems affect the
functions of production planning and control.
In Damodaran et al., (1992), a production
planning model was proposed for cellular
manufacturing systems but their model does
not include inventory aspects. In this paper,
an optimization model for simultaneous
inventory planning, part family allocation,
and manufacturing cell formation is
proposed. The objective function of the model
in the current research is to minimize inter-
cell material handling cost, finished-goods
inventory cost and machine set-up cost for
The current issue and full text archive of this journal is available at
http://www.emerald-library.com/ft
[ 275]
Integrated Manufacturing
Systems
12/4 [2001] 275284
# MCB University Press
[ISSN 0957-6061]
Keywords
Production planning,
Cellular manufacturing,
Integrated manufacturing system
Abstract
Inventory control models deal with
production planning in order to
minimize inventory and shortage
cost, while cellular manufacturing
analysis mainly addresses how
machines should be grouped and
parts be produced. A
mathematical programming model
is developed using an integrated
approach for production and
inventory planning in a cellular
manufacturing environment. The
mathematical programming model
minimizes inter-cell material
handling cost, finished-good
inventory cost and system set-up
cost. The non-linear mixed integer
programming model cannot be
directly solved for real size
practical problems due to its NP-
complexity. A decomposition-
based heuristic algorithm was then
developed to efficiently solve the
integrated planning and control
problem. Numerical examples are
provided to test and illustrate the
model and the solution method
presented in this paper.
This research was
supported in part by
Research Grant
#OGP0121863 from NSERC
of Canada. The author
thanks the anonymous
referee for comments and
suggestions which have
improved an earlier version
of this paper. The author
also thanks J. Gregory at
Concordia University for her
help in proof-reading the
manuscript of this paper.
cellular manufacturing production. Similar
to other optimization models, achieving this
optimal goal is subject to certain constraints.
In the next section, detailed problem
description and discussion on the objective
and constraint functions are presented,
followed by the formulation of a non-linear
mathematical programming model. The
development of a heuristic to solve the
production planning problem is presented in
the third section. A number of numerical
examples to illustrate the model and the
algorithm are shown in the fourth section
along with the discussions on certain
computational aspects related to the solution
method. Summary and conclusions of the
research presented in this paper are given in
the final section. Possible extensions of the
model and further development of the
algorithm are also discussed in the final
section.
Problem description and model
formulation
Consider a manufacturing system consisting
of a number of machines to process different
part-types. Each part-type may require some
or all of the machines for processing. In
addition, consider the manufacturing system
in a number of time periods t, where
t = 1,2,. . ., T, with T > 1. One time period
could be a day, a week or a month. Demands
for different part-types are assumed to be
known from work orders or from forecast.
Demands may be satisfied from production in
the same time period or from inventory. Back
orders and shortage cost are not considered
in this paper. Our problem is to determine
the optimal quantities of different part-types
to be processed by the machines at different
time periods. The production planning
problem is considered in a cellular
manufacturing environment and the solution
of the problem also determines how to form
the manufacturing cells and how to group the
part-types into part families. A mathematical
programming model is developed to solve
this cellular manufacturing production
planning problem. Owing to the above
problem features, the mathematical
programming model becomes a mixed
integer non-linear programming model. The
objective function of this model is to
minimize inter-cell material handling cost,
machine operating cost, production set-up
cost and part inventory cost. In this model,
we assume that there is a single process plan
for each part-type. Part processing cost,
therefore, is not included in the objective
function. Minimizing this cost function is
subject to known or predicted production
demands and manufacturing process
requirements. Since both the quantity of
different part-types to be processed by the
machines and the formation of the machine
cells are decision variables, non-linear terms
are presented in the objective function of the
mixed integer programming model.
However, these non-linear terms can be
linearized without much difficulty. Detailed
model linearization is provided in the
Appendix. Before the model is presented, we
first give the notations:
Known parameters and coefficients
t = time index, t = 1,. . .,T
i = part-type index, i = 1, . . .,I(t)
j = index of operations of part-type
i, j = 1,. . .,J
i
k = machine index, k = 1,. . .,K
l = cell index, l = 1,. . ., L
H
i
(t) = unit inventory holding cost of part-
type i for time period t
D
i
(t) = known demand of part-type i for
time period t
M
k
(t) = unit machine operating cost for
machine type k in time period t
S
i
= set-up cost to produce part-type i
R
i
= unit cost to move part-type i in
batches between cells
LB
l
= minimum number of machines in
cell l
UB
l
= maximum number of machines in
cell l
Continuous decision variables
x
i
(t) = amount of part-type i to be
processed in time period t
v
i
(t) = amount of part-type i in inventory
at the end of time period t
0-1 Decision variables
z
il
(t) =
1; if part-type i is processed in
cell l during time t;
0; otherwise
8
>
<
>
:

i[jk[l
(t) =
1; if operation j of part i to be
processed by machine k is
done in cell l during time t;
0; otherwise
8
>
>
<
>
>
:

i
(t) =
1; if part-type i is processed in
time t;
0; otherwise
8
<
:
n
kl
(t) =
1; if one unit of type k machine is
placed in cell l at time t;
0; otherwise
8
<
:
Subscripts i[jk] of variable
i[jk[l
indicate that
machine k is required to process operation j
of part-type i. This information is known
from the given part process plan. Using the
[ 276]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
above defined variables and parameters, the
cost function of the mathematical model can
be expressed as below:
min
X
T
t=1
X
I(T)
I=1
R
i
x
i
(t)[
X
L
l=1
z
il
(t) 1[

X
T
t=1
X
K
k=1
M
k
(t)
X
L
l=1
n
kl
(t)

X
T1
t=1
X
I(t)
i=1
H
i
(t)v
i
(t)
X
T
t=1
X
I(t)
i=1
S
i

i
(t)
The first term of the objective function is the
inter-cell material handling cost. This cost
function is similar to those in Vakharia and
Kaku (1993) and Chen (1998). In a typical cell
formation problem, the number of part-type i
to be produced in time t, x
i
(t), is normally
considered constant. In that case, the
material handling cost function will be
linear. The use of this form of material
handling cost function implicitly assumes
that the distances between every pair of the
cells are equal. The second term of the
objective function is the machine operating
cost. We assume that the machines can be
included when they are needed and can be
removed from the system when they are not
required. This assumption will be discussed
and explained further in a later section of
this paper. The third term is the finished
goods inventory cost and the last term in the
function is the system set-up cost. The first
two cost items are related to forming
manufacturing cells, while the latter two cost
items are related to production and
inventory control. The minimization of this
cost function is subject to certain constraint
conditions. One constraint is to ensure that,
if operation j of part-type i will be processed
by machine k in one of the cells in time
period t, then the corresponding binary
variable for system set-up must be 1. This
requirement can be expressed by:
X
L
l=1

i[jk[l
(t) =
i
(t):
Constraints must also ensure that, if
operation j of part-type i is to be processed by
machine k placed in cell l, then part-type i
will be processed in cell l and machine k must
be placed in cell l. These two requirements
can be expressed by:

i[jk[l
(t) _ z
il
(t)
and

i[jk[l
(t) _ n
kl
(t):
Normally there is an upper limit to the
number of machines in each cell due to the
limit of the physical space. In addition, there
should be at least one machine in each cell;
otherwise the cell will disappear. These two
requirements can be enforced by:
LB
l
_
X
k
k=1
n
kl
(t) _ UB
l
:
In planning the production, demand of part-
type i at time t should be deducted from the
finished parts in storage at time t. The
relationship of storage levels at different time
periods is given by:
v
i
(t 1) = v
i
(t) x
i
(t) D
i
(t):
Since backorders are not considered in this
model, we require:
v
i
(t) _ 0:
We also require that the production in the
entire planning horizon must meet the total
demand. This is expressed by:
X
T
t=1
x
i
(t) =
X
T
t=1
D
i
(t):
The last constraint of the model is the
relationship between the set-up variable
i
(t)
and the part processing quantity x
i
(t). From
the definitions of the variables, we have:

i
(t) =
1; if x
i
(t) > 0;
0; if x
i
(t) = 0:
(
Summarizing the above discussed objective
and constraint functions, we present the non-
linear mixed integer programming model for
planning cellular manufacturing production
as follows:
MP
min
X
T
t=1
X
I(t)
i=1
R
i
x
i
(t)[
X
L
l=1
z
il
(t) 1[

X
T
t=1
X
K
k=1
M
k
(t)
X
L
l=1
n
kl
(t)

X
T1
t=1
X
I(t)
i=1
H
i
(t)v
i
(t)
X
T
t=1
X
I(t)
i=1
S
i

i
(t)
(1)
s.t.
X
L
l=1

i[jk[l
(t) =
i
(t); j = 1; . . . ; J
i
; i = 1; . . . ; I; \t;
(2)

i[jk[l
(t) _ z
il
(t); j = 1; . . . ; j
i
; i = 1; . . . ; I;
l = 1; . . . L; \t;
(3)

i[jk[l
(t) _ n
kl
(t); k = 1; . . . ; k; l = 1; . . . L; \t (4)
LB
l
_
X
K
k=1
n
kl
(t) _ UB
l
; l = 1; . . . ; L; \t; (5)
[ 277]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
v
i
(t 1) = v
i
(t) x
i
(t) D
i
(t);
i = 1; . . . ; I; t = 1; . . . ; T 1;
(6)
X
T
t=1
x
i
(t) =
X
T
t=1
D
i
(t); i = 1; . . . ; I; (7)

i
(t) =
1; if x
i
(t) > 0;
0; if x
i
(t) = 0:
i = 1; . . . ; I; \t;

(8)
x
i
(t) _ 0; v
i
(t) _ 0;
i[jk[l
(t); n
kl
(t); z
il
(t)
= 0; 1; \i; j; l:t:
(9)
We found that directly solving the above
mixed integer programming model is NP-
hard by comparing the above model with
those discussed in Nemhauser and Wolsey
(1988), where their NP-hardness was proven.
It is computationally prohibitive to use
brute-force methods to solve a real-life
problem based on this NP-hard model, since
it has a large number of 0-1 integer variables.
In order to solve practical problems without
exponentially increased computation, an
efficient heuristic algorithm was developed
based on decomposition and well-known
Silver-Meal heuristic (Winston, 1994) for
production planning. The heuristic solution
method is discussed in detail in the next
section.
Heuristic solution method
The non-linear terms in model MP presented
in the previous section can be linearized
without much difficulty. In theory, the
linearized model can be directly solved using
any integer programming software such as
LINDO (Schrage, 1991). However, it is more
difficult to solve a linearized non-linear
problem than solving a problem without non-
linear terms in its original formulation.
Linearization results in a larger number of
constraint functions and consequently
requires more computation time to search for
optimal solutions. In the early stages of this
research, we attempted to directly solve the
linearized model using LINGO (LINDO
Systems Inc., 1995), an advanced version of
LINDO. A number of small problems were
solved directly using LINGO based on the
linearized formulation shown in the
Appendix and the results were used to test
the quality of the heuristic algorithm. Our
experiences show that it would take too long
to solve even very small experimental
problems and almost impossible to solve it
for real-life problems using widely available
PC computers. Some of these computational
experiences using the direct solution
approach are reported in the next section.
Attempts were also made to re-formulate the
MP model and to avoid using non-linear
terms. Such efforts were not successful, since
the quantities of part-types to be processed by
the machines and location of the machines
are both decision variables in this integrated
model. With the heuristic algorithm
developed in this research, larger size real-
life problems can be solved within
polynomial time of computation. Similar to
any other heuristic algorithms, however, the
optimality of the solution generated by our
heuristic is not guaranteed.
The first step of the heuristic algorithm is
to group the part-types into part families
based on their machine-part incidence
matrix. This can be done using any available
part grouping methods such as those
discussed in Singh (1996), and a part list can
be generated from such methods. The part-
types are sequenced according to their
closeness to the preceding part-type in the
list. Let the part-type sequence be {i
1
, i
2
, . . .,
i
v
, . . ., i
I
}, where I = max
t
{I(t)}. The heuristic
algorithm starts from part-type i
1
. Based on
given demand D
i1
(t), set-up cost S
i1
and
holding cost H
i1
, we use the Silver-Meal
heuristic to solve the planning problem for
part-type i
1
. The Silver-Meal heuristic is
utilized in this solution method because of its
simplicity, good quality, and fast speed in
obtaining the solutions. Since the whole
solution approach is heuristic, we decided to
use this efficient and effective heuristic to
solve the decomposed sub-problems. Using
the production plans generated by the Silver-
Meal method, machines required for
processing part-type i
1
will be placed in the
same cell if cell limit is not reached. When
the cell limit is reached, machines will be
placed in subsequent cells. When the
machines for processing the first part-type i
1
in the part list are all placed, the algorithm
will continue to place machines for other
part-types. To process other part-types, the
already placed machines will be utilized, if
part travel cost is lower than machine
operating cost. Otherwise, a second piece of
the same machine will be placed in the
system. This process will continue until the
arrangement for all the part-types is made.
The detailed solution process is presented in
the algorithm given below.
Heuristic for planning cellular
manufacturing production
Step 1
Start the algorithm with input data of total
time period T, part-type I(t), operation
number J
i
, machine type K, total number of
cells L, maximum and minimum numbers of
machines (UB
l
and LB
l
, respectively) for each
[ 278]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
cell l. Other input data are production
demand D
i
(t), inventory holding cost H
i
(t),
set-up cost S
i
, machine operating cost M
k
(t)
and material handling cost R
i
.
Step 2
For all the part-types to be processed by the
machines, group part-types into part families
based on rank order clustering (ROC)
algorithm (Singh, 1996; Heragu, 1997).
Step 3
Generate part list based on the output of the
ROC algorithm in Step 2. Let the sequenced
part list be {i
1
, i
2
, i
3
. . .i
I
} and let v = i
1
.
Step 4
Based on D
v
(t), S
v
and H
v
(t), use Silver-Meal
heuristic to decide the solution of x
v
(t) and
v
v
(t).
Step 5
For the solution of x
v
(t) found in Step 4, let k
1
,
k
2
, . . ., k
v
be the machines required to process
part-type v, k
1
, k
2
, . . ., k
v
{1, . . ., K}. For time
period t with x
v
(t) > 0, check:
.
Step 5.1. If machines k
1
, k
2
. . ., k
v
have been
placed in one single cell in the system,
then no more new machines will be placed
in the system. Go to Step 6.
.
Step 5.2. If machines k
1
, k
2
. . ., k
v
have been
placed in different cells and all cells have
reached their limits of maximum number
of machines, then no more machines can
be placed in the system. Go to Step 6.
.
Step 5.3. If machines k
1
, k
2
. . ., k
v
have been
placed in different cells, not all cells have
reached their machine limits, and for
\k
/
; k
/
k
1
; k
2
. . . ; k
v
; if R
v
x
v
(t) > m
k
/ (t),
then one more unit of machine k
/
will be
placed in a cell which has a space for an
extra machine, and has the largest
number of machines required to process
part-type v.
.
Step 5.4. If machine k
/
k
1
; k
2
. . . ; k
v
has
not been placed in the system, and not all
of the cells have reached their machine
limits, then place one unit of machine k
/
in
a cell which has a space for an extra
machine, and has the largest number of
machines required to process part-type v.
.
Step 5.5. If machine k
/
k
1
; k
2
. . . ; k
v
has
not been placed in the system, and all cells
have reached their machine limits, then
replace a duplicated machine k
//
by
machine k
/
, where
X
L
l=1
n
k
//
l
(t) _ 2:
If there are two or more such machines,
select the one corresponding to the
highest machine cost M
k
(t).
.
Step 5.6. If machine k
/
k
1
; k
2
. . . ; k
v
has
not been placed in the system, all cells
have reached their machine limits, and
there are no duplicated machines in any of
the cells in the system, then the original
problem has no feasible solution. Stop.
Step 6
If v = i
I(t)
, stop; otherwise, let v = v +1. Go to
Step 4.
Remarks
The above heuristic algorithm can always
find a feasible solution to the problem, if the
original problem has any feasible solutions.
Since duplicated machines will be removed
from the system when cell limits are reached,
the indication of no feasible solution from
Step 5.6 means that the original problem has
no feasible solutions.
Step 5 in the above algorithm attempts to
solve a much simplified minimization
problem with fixed t = t
/
, i = v and known
values of x
i
(t) = x
v
(t
/
), as given below:
.
For GP:
minG(z(t
/
); n(t
/
)) = R
v
x
v
(t
/
)
X
L
l=1
z
vl
(t
/
) 1
" #

X
K
k=1
M
k
(t
/
)
X
L
l=1
n
kl
(t
/
)
(10)
.
For s.t.
X
L
l=1

v[jk[l
(t
/
) = 1; j = 1; . . . ; J
i
; (11)

v[jk[l
(t
/
) _ z
vl
(t
/
); j = 1; . . . ; J
v
; l = 1; . . . L;
(12)

v[jk[l
(t
/
) _ n
kl
(t
/
); k = 1; . . . ; k; l = 1; . . . L;
(13)
LB
l;v
_
X
K
k=1
n
kl
(t
/
) _ UB
l;v
; L = 1; . . . L; (14)
n
kl
(t
/
); z
vl
(t
/
) = 0; 1; \j; l: (15)
The model GP is a simplified cell formation
model. One can find a number of
optimization or heuristic algorithms aimed
at solving this model or its more complicated
extensions (Dahel, 1995; Logendran, 1993). In
this paper, we will not elaborate further on
the detailed features of this model or its
different extensions.
Computational time of using this algorithm
is polynomial, since it needs to use the ROC
and Silver-Meal algorithm, once in Steps 2
and 4, respectively. Machine placement in
Step 5 requires a maximum of T I K times
of comparisons, where I = max
t
{I(t)}.
[ 279]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
Next we present numerical examples to
illustrate the mathematical model and the
heuristic algorithm discussed above.
Numerical examples
In this section, we first present an example to
illustrate the solution procedure. We will
also present the main features of several
other examples to compare the
computational aspects of the heuristic
method and the brute-force methods usually
used by general purpose integer
programming software.
Example 1
In this example, there are ten different part-
types to be processed in five time periods (T =
5). Machines are to be placed in four
manufacturing cells. Each cell can have a
maximum of four machines of any type.
Detailed production demands for the ten part-
types are shown in Table I. Unit inter-cell
material handling cost for each part-type and
machine set-up cost to process the part-types
are given in Table II. Unit inventory costs to
hold the part-types are also given in Table II.
Table III presents the number of operations
of each part-type, part-machine relationship,
and machine cost information. Machine
operating costs are given in the second
column of this Table. For example, it costs 15
units to operate one unit of Machine 1 in the
system; ten units for Machine 2, and so on.
The third column of Table III shows that
there are four operations for processing Part
1. It also indicates that Machines 1, 2, 3, and 4
are required to perform operations 3, 1, 2, and
4, respectively, for Part 1. Similarly, the fifth
column of this Table shows that Part 3
requires Machines 5, 6 and 7 for its
operations 1, 2, and 3, respectively. All the
data in these three tables are randomly and
hypothetically generated and are used for
illustration purposes only.
Using the data in Tables I-III, the solution
was obtained by the heuristic algorithm. The
rank order clustering method was first used
to generate the part list shown in Table IV.
Table IV also shows the times and amounts of
production to satisfy demands of all time
periods. These are the results of the Silver-
Meal heuristic method. The total inventory
and set-up cost is 625.3. The second part of the
algorithm identifies machine cells for
different time periods presented in Table V.
The entries in this Table are the machine
numbers. One can notice that multiple units
of the same machine are allowed for the
system. Since there is no machine capacity
constraint in the model, more than one unit
of each type of machine in the same cell is
neither necessary nor economical. In the
second line of Table V (time t = 1), it shows
that Cell 1 has machines 3, 5 and 7 and Cell 2
has machines 4, 6, and 7, etc. Compared with
the data in Tables III and IV, we find that the
22 units of part-type 1 processed in time
period 1 will flow from Cell 1 or Cell 4 to Cell
3, and finish at Cell 2 or Cell 4 to be processed
by Machines 3, 1, 2 and 4, respectively. Other
part-type flows can also be identified with
data shown in Tables III, IV and V. Total
machine operating and material handling
cost is 1929.6 for this example problem. The
algorithm is coded in FORTRAN-77 and the
computational time for this simple example
using the heuristic algorithm is less than 0.01
seconds on a PC (Pentium-133) computer.
Table I
Part processing demand for Example 1
Part
Time period t 1 2 3 4 5 6 7 8 9 10
1 22 20 18 12 23 24 22 30 19 25
2 28 16 20 24 30 22 18 28 18 14
3 36 22 30 39 32 26 16 26 33 22
4 14 24 22 25 18 20 10 14 22 28
5 27 18 15 16 29 28 20 31 19 22
Table II
Cost information for Example 1
Part
1 2 3 4 5 6 7 8 9 10
Unit inter-cell M.H. cost 1.8 2.0 2.5 1.5 1.0 1.6 2.0 2.2 3.0 1.7
System set-up cost 25 30 20 27 20 25 18 25 25 25
Unit inventory cost 1.0 1.5 0.5 1.0 1.2 0.5 2.5 1.2 1.5 0.8
[ 280]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
Other examples and calculation
experiences
Although the size of the problem in Example
1 is relatively small, optimal solutions could
not be obtained using a brute-force method
for this problem. We used the linearized
optimization model shown in the Appendix to
solve a number of even smaller problems.
The problem characters and computational
features of these testing problems are shown
in Table VI. We conducted the computation
using PC-based HyperLINGO software
(LINDO Systems Inc., 1995). From Table VI,
one can notice that the computational time
increases exponentially to the sizes of these
problems. In running the testing problems,
the search process for the solution of
Problem 5 was terminated after 11 hours of
calculation, when the best IP solution and the
lower bound shown on the LINGO screen
were still far apart. The number of
constraints in Problem 6 exceeds the limit of
our software and we were unable to run this
problem by LINGO (the LINGO model and
data files of Example 1 can be obtained by
contacting the author). From the experience
of computing these small testing problems,
we observed that the computational time will
not be tolerable for finding optimal solutions
using general integer programming software
Table III
Part operation requirement and machine cost for Example 1
Machine Machine Part
number cost 1 2 3 4 5 6 7 8 9 10
1 15 3 1 1 1 1
2 10 1 2 1 5 1 1 2
3 20 2 4 2 2
4 15 4 3 3 3 3
5 12 1 2 1 2 3 2
6 15 2 2 4
7 16 3 3 4 3 5
Table IV
Part processing sequence and batch size in Example 1
Part
Time period t 1 2 10 9 7 5 6 8 4 3
1 22 36 39 19 22 23 46 30 36 38
2 28 0 0 18 18 30 0 28 0 0
3 50 22 22 33 16 32 46 26 71 67
4 0 42 50 22 10 18 0 14 0 0
5 27 0 0 19 20 29 28 31 0 0
Table V
Cell formation of Example 1
Cell 1 Cell 2 Cell 3 Cell 4
t = 1 3,5,7 4,6,7 1,2,4 1,3,4
t = 2 2,3,7 3,5,7 2,6,7 1,3,4
t = 3 3,5,7 4,6,7 1,2,4 1,3,4
t = 4 2,3,5 1,3,7 4,6,7 1,2,4
t = 5 2,3,7 3,5,7 2,6,7 1,3,4
Table VI
Other examples and computation features
Problem
Num.of time
periods
Num. of
part types
Num. of
cells
Num. of
machine types
Num. of
vars.
Num of
integer vars.
Num. of
constraints
CPU time
(hh:mm:ss)
1 2 3 2 3 83 62 143 00:00:00.10
2 2 4 2 3 108 80 191 00:01:02.00
3 2 5 2 4 137 102 239 00:03:39.00
4 2 6 2 4 162 120 287 01:14:01.00
5 3 7 2 4 246 180 430 (>11 hours)
6 5 10 4 7 1,390 950 2,346 N/A
[ 281]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
on widely available PC platforms. One may
have to rely on efficient heuristic algorithms
to generate satisfactory or near-optimal
solutions in solving practical problems. The
heuristic algorithm developed in this
research found optimal solutions within less
than 0.1 seconds for all the trivial problems
(Problems 1 to 4), while no comparison can be
made for Problems 5 or 6.
Summary and discussion
In this paper, we developed a mathematical
programming model to solve production
planning problems in cellular manufacturing
systems. The non-linear integer
programming model can be linearized but
solving the model is NP-hard. An algorithm
based on decomposition and the Silver-Meal
heuristic was proposed to solve the problem.
Examples tested in this research show that
this method is very promising. There are a
number of modelling and other issues that
can be further investigated. These are listed
below:
.
Different travel distances between cells.
Quadratic terms can be used to express
the costs of material handling among the
cells (Atmani et al., 1995). In a way similar
to that shown in the Appendix, those non-
linear terms can easily be linearized by
adding more integer variables and linear
constraints into the model. The numbers
of added integer variables and constraints
are normally significant to the model size
and make the problem more difficult to
solve.
.
Operation sequence. Part operation
sequence can be enforced by adding more
constraints to the model following a
simple process (Vakharia and Kaku, 1993).
This will also make the problem larger
and more difficult to solve.
.
Machine cost. In this paper, we assume
that the cost of machines occurs when
they are used in the system. The
algorithm decides that certain machines
will be included in the cells only logically,
not physically. If there are costs
associated with actual machine
installation and removal, then the model
must be further modified.
.
Machine capacity constraint. Machine
capacity constraint can be used in
formulating the cell formation model.
Such constraints can be in the form of:
X
i;j
q
i[jk[
x
i[jk[l
(t) _ n
kl
(t); \k; l; t;
where q
i[jk]
is the capacity requirement for
machine k by operation j of part-type i and
n
kl
(t) is the number of type k machines in
cell l. The above inequality can be used to
substitute Equation (4) in the MP model.
This more general machine capacity
constraint will also make the model more
difficult to solve (Dahel, 1995).
.
Set-up cost. The Silver-Meal heuristic does
not consider the difference in set-up costs
between continuing production or
interrupted production. This same
approach was also followed in modelling
the problem in this paper. This cost
difference should be included in a more
general model. Again, the model then may
have more integer variables and become
more difficult to solve.
All the above possible extensions make the
model more realistic but result in a
substantially larger number of integer
variables. The model extensions related to
machine installation and set-up costs will
cause significant change of the heuristic
algorithm, while the others should not affect
the use of the heuristic approach proposed in
this paper. Certain modifications of the
algorithm may still be needed. Nevertheless,
in our future research, we plan to investigate
the features of these more realistic models
and to develop efficient algorithms to solve
them. In addition, a number of practical
issues need to be considered in applying the
solution technique developed in this paper.
For example, demand forecast, non-
quantitative constraints, alternative
manufacturing processes, cost estimates,
machine layout and re-layout, workforce
management, etc., all should be carefully
addressed in practical implementations.
These are important practical issues for
production planning in a dynamic and
integrated cellular manufacturing system
environment as well as challenging research
topics in this area.
References
Arvindh, B. and Irani, S.A. (1994), ``Cell
formation: the need for an integrated solution
of the subproblems'', International Journal of
Production Research, Vol. 32, pp. 1197-218.
Atmani, A., Lashkari, R.S. and Caron, R.J. (1995),
``A mathematical programming approach to
joint cell formation and operation allocation
in cellular manufacturing'', International
Journal of Production Research, Vol. 33,
pp. 1-15.
Bard, J. and Golany, B. (1991), ``Determining the
number of kanbans in a multiproduct,
multistage production system'', International
Journal of Production Research, Vol. 29,
pp. 881-95.
Bitran, G.R. and Chang, L. (1987), ``A
mathematical programming approach to a
[ 282]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
deterministic kanban system'', Management
Science, Vol. 33, pp. 427-41.
Burbidge, J.L. (1989), Production Flow Analysis
for Planning Group Technology, Oxford
Science Publication, Oxford.
Chen, M. (1998), ``A mathematical programming
model for system reconfiguration in a
dynamic cellular manufacturing
environment'', Annals of Operations Research,
Vol. 77, pp. 109-28.
Chen, S.-J. and Cheng, C.S. (1995), ``A neural
network-based cell formation algorithm in
cellular manufacturing'', International
Journal of Production Research, Vol. 33,
pp. 293-318.
Dahel, N.-E. (1995), ``Design of cellular
manufacturing systems in tandem
configuration'', International Journal of
Production Research, Vol. 33, pp. 2079-95.
Damodaran, V., Lashkari, R.S. and Singh, N.
(1992), ``A production planning model for
cellular manufacturing systems with
refixturing considerations'', International
Journal of Production Research, Vol. 30,
pp. 1603-16.
Harhalakis, G., Nagi, R. and Proth, J.M. (1990),
``An efficient heuristic in manufacturing cell
formation for group technology applications'',
International Journal of Production Research,
Vol. 28, pp. 185-98.
Heragu, S. (1997), Facilities Design, PWS.
Jamal, A.M.M. (1993), ``Neural network and
cellular manufacturing'', Industrial
Management & Data Systems, Vol. 93, pp. 21-5.
LINDO Systems Inc. (1995), LINGO User's Guide,
LINDO Systems Inc.
Logendran, R. (1993), ``Methodology for
converting a functional manufacturing
system into a cellular manufacturing
system'', International Journal of Production
Economics, Vol. 29, pp. 27-41.
Monden, Y. (1983), Toyota Production Systems,
Industrial Engineering and Management
Press, Norcross, GA.
Nemhauser, G.L. and Wolsey, L.A. (1988), Integer
and Combinatorial Optimization, Wiley, New
York, NY.
Olorunniwo, F.O. (1996), ``Changes in production
planning and control systems with
implementation of cellular manufacturing'',
Production and Inventory Management,
Vol. 37, pp. 65-70.
Price, W., Gravel, M. and Nsakanda, A.L. (1994),
``A review of optimization models of Kanban-
based production systems'', European
Journal of Operational Research, Vol. 75,
pp. 1-12.
Price, W., Gravel, M., Nsalanda, A.L. and Cantin
F. (1995), ``Modeling the performance of a
Kanban assembly shop'', International
Journal of Production Research, Vol. 33,
pp. 1171-7.
Riggs, J. (1981), Production Systems: Planning,
Analysis and Control, Wiley, New York, NY.
Schrage, L. (1991), LINDO: An Optimization
Modeling System, Boyd & Fraser.
Singh, N. (1996), Systems Approach to Computer-
Integrated Design and Manufacturing, Wiley,
New York, NY.
Spearman, M.L. and Zazanis, M.A. (1992), ``Push
and pull production systems: issues and
comparisons'', Operations Research, Vol. 40,
pp. 521-32.
Vakharia, A.J. and Kaku, B.K. (1993),
``Redesigning a cellular manufacturing
system to handle long-term demand changes:
a methodology and investigation'', Decision
Sciences, Vol. 24, pp. 909-23.
Winston, W.L. (1994), Operations Research:
Applications and Algorithms, 3rd ed.,
Duxbury Press, Pacific Grove, CA.
Appendix. Model linearization
The non-linear terms in the objective
function of model MP presented in the second
section can be linearized. We first re-write
the first term of the function as follows:
X
T
t=1
X
I(t)
i=1
R
i
x
i
(t)[
X
L
l=1
z
il
(t) 1[
=
X
T
t=1
X
I(t)
i=1
R
i
[
X
L
l=1
x
i
(t)z
il
(t)[ x
i
(t)
(16)
The non-linear items x
i
(t)z
il
(t) can be
linearized. Let:
w
il
(t) = x
i
(t)z
il
(t):
This implies that:
w
il
(t) =
x
i
(t); if z
il
(t) = 1;
0; if z
il
(t) = 0:

Then equation (16) can be replaced by the


following linear expression:
X
T
t=1
X
I(t)
i=1
R
i
X
L
l=1
w
il
(t) x
i
(t)
( )
(17)
with the three following added inequalities:
w
il
(t) _ x
i
(t) Mz
il
(t) M; (18)
w
il
(t) _ x
i
(t); (19)
w
il
(t) _ Mz
il
(t) (20)
where M is a large positive number. In the
three inequalities, Equations (18) and (19)
enforce w
il
(t) to be x
i
(t) when z
il
(t) is 1; while
Equation (20) enforces w
il
(t) to be 0, when z
il
(t) is 0. Equation (20) has no effect on w
il
(t)
when z
il
(t) is 1; while Equation (18) and (19)
have no effect on w
il
(t) when z
il
(t) is 0.
The other non-linear part is Equation (8) in
the model MP (x(t), z(t), (t), v(t), n(t)). The
conditional 0-1 requirement for variable
i
(t)
can be simply converted to the following set
of inequalities:

i
(t) _ x
i
(t) (21)
M
i
(t) _ x
i
(t) (22)
[ 283]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284
while
i
(t) are 0-1 variables and M is a large
positive number. One can easily verify that
Equations (21) and (22) serve the same
purpose as Equation (8). Using the above
simple linearization transformation, the
original model can be converted into:
MLP(x(t); z(t); (t); v(t); n(t))
minM(x(t); z(t) =
X
T
t=1
X
I(t)
i=1
R
i
[
X
L
l=1
w
il
(t)
x
i
(t)[ S
i

i
(t)
(23)

X
T1
t=1
X
I(t)
i=1
H
i
(t)v
i
(t)
X
T
t=1
X
K
k=1
M
k
(t)
X
L
l=1
n
kl
(t)
s.t.
X
L
l=1

i[jk[l
(t) =
i
(t); j = 1; . . . ; J
i
; i = 1; . . . ; I; \t;
(24)

i[jk[l
(t) _ z
il
(t); j = 1; . . . ; J
i
; i = 1; . . . ; I;
l = 1; . . . L; z\t;
(25)

i[jk[l
(t) _ n
kl
(t); k = 1; . . . ; k; l = 1; . . . L; \t (26)
LB
l
_
X
K
k=1
n
kl
(t) _ UB
l
; l = 1; . . . ; L; \t; (27)
v
i
(t 1) = v
i
(t) x
i
(t) D
i
(t);
i = 1; . . . ; I; t = 1; . . . ; T 1;
(28)
X
T
t=1
x
i
(t) =
X
T
i=1
D
i
(t); i = 1; . . . ; I; (29)
w
il
(t) _ x
i
(t) Mz
il
(t) M; (30)
W
il
(t) _ x
i
(t) (31)
w
il
(t) _ Mz
il
(t); (32)

i
(t) _ x
i
(t) (33)
M
i
(t) _ x
i
(t) (34)
x
i
(t) _ 0; w
il
(t) _ 0; v
i
(t) _ 0; (35)

i[jk[l
(t);
i
(t); n
kl
(t); z
il
(t) = 0; 1; \i; j; l:t: (36)
[ 284]
Mingyuan Chen
A model for integrated
production planning in cellular
manufacturing systems
Integrated Manufacturing
Systems
12/4 [2001] 275284

You might also like