Professional Documents
Culture Documents
Department of Industrial & Systems Engineering, University of Ohio, Athens, OH, USA
School of Industrial Engineering, College of Engineering, University of Tehran, Tehran, Iran
c
Research Center for Organizational Processes Improvement, Sari, Iran
d
Department of Industrial Engineering, Iran University of Science and Technology, Tehran, Iran
e
Department of Electrical Engineering, University of North Carolina at Charlotte, Charlotte, NC, USA
b
a r t i c l e
i n f o
Article history:
Received 8 June 2013
Received in revised form 20 April 2015
Accepted 14 May 2015
Available online 5 June 2015
Keywords:
Robust optimization
Dynamic cellular manufacturing system
Production planning
Uncertainty
Machine reliability
Operator assignment
a b s t r a c t
In this study, a robust optimization approach is developed for a new integrated
mixed-integer linear programming (MILP) model to solve a dynamic cellular manufacturing
system (DCMS) with unreliable machines and a production planning problem simultaneously. This model is incorporated with dynamic cell formation, inter-cell layout, machine
reliability, operator assignment, alternative process routings and production planning
concepts. To cope with the parts processing time uncertainty, a robust optimization
approach immunized against even worst-case is adopted. In fact, this approach enables
the systems planner to assess different levels of uncertainty and conservation throughout
planning horizon. This study minimizes the costs of machine breakdown and relocation,
operator training and hiring, inter-intra cell part trip, and shortage and inventory. To verify
the performance of the presented model and proposed approach, some numerical examples
are solved in hypothetical limits using the CPLEX solver. The experimental results demonstrate the validity of the presented model and the performance of the developed approach
in nding an optimal solution. Finally, the conclusion is presented.
2015 Elsevier Inc. All rights reserved.
1. Introduction
Cellular manufacturing (CM) concept as an application of group technology philosophy in industrial environments uses
adjacency among the parts features to reduce parts defect rates, throughput times, factory space requirements, material trips
and system setup times and costs. On the other hand, the main goal of production planning (PP) as a capacity planning tool is
to reduce the inventory and shortage levels. In fact, PP is the means by which we prepare our production quantities. Due to
nowadays competitive manufacturing environment, having more efciency and productivity is become a vital concern for
industry owners, to produce with the lowest level of cost and casualties as well as the highest level of exibility.
Therefore, in recent years most industries show great inclination to cellular manufacturing systems (CMSs). A CMS design
consists of four important decisions: cell formation (CF), group layout (GL), group scheduling (GS) and resource allocation
(RA) assigning various resources (e.g., operators, equipment and materials) [13]. In manufacturing environments, these
Corresponding author.
E-mail address: mandana.sakhaii@gmail.com (M. Sakhaii).
http://dx.doi.org/10.1016/j.apm.2015.05.005
0307-904X/ 2015 Elsevier Inc. All rights reserved.
170
decisions are inter-linked. Therefore, the need arises to develop models considering them as an integrated whole. Indeed, it
cannot be guaranteed that optimal decisions obtained from a problem that only models with one of the aforementioned factors (i.e., CF, GL, RA and production planning) will be optimal for another problem, separately. Indeed, a real industrial environment comprises all of these factors together. For instance, the obtained solutions satisfying only limitations of the CF
problem cannot satisfy the limitations of a problem integrated all factors. Then, the nal design may not be efcient and
optimal. Consequently, in order to have a model whose decisions are comparable with practical decisions, a comprehensive
model is necessary.
Most of the previous studies usually concentrate only on CF decision. A comprehensive review of CF related studies can be
found in [48].
There have been some studies that concentrate on GL. Rosenblatt [9] investigated a dynamic layout and illustrated that
generating a static conguration at random is not effective. Wang et al. [10] developed a model in CMS, which considers
demand changes over the product life cycle. They minimized the total material handling cost and solved inter and
intra-cell facility layout problems, concurrently, using a simulated annealing (SA) algorithm. Solimanpur et al. [11] proposed
an inter-cell layout problem in cellular manufacturing. In their study, the problem is modeled as a quadratic assignment
problem and then an ant colony optimization (ACO) algorithm is developed to solve the proposed model.
Tavakkoli-Moghaddam et al. [12] formulated a facility layout model in a CMS by considering stochastic demands, in which
the objective function is to minimize inter and intra-cell material handling costs in both interintra cell layout problems,
concurrently.
There are some studies in the literature integrate CF and GL decisions together. Paydar et al. [13] developed a solution to
solve cell formation and layout of machines within each cell problem, concurrently, under the assumption that a number of
cells are known in advance. They treated CM as a multiple departures single destination multiple traveling salesman problem and a solution methodology based on SA is proposed to solve the formulated model. Kia et al. [2] proposed an integrated
CF and GL model, in which the multi-rows layout utilization to locate machines in the cells congured with exible shapes
and some design features is considered.
As mentioned previously, one of the decisions in developing a CMS is considering resource allocation (e.g., operator allocation) that has rarely been considered in CM models; usually job waiting occurs due to the non-availability of machines
and/or operators. Therefore, operator allocation to machines can play a key role in the CMS efciency. Aryanezhad et al.
[14] proposed a dynamic cell formation and operator assignment problem. They considered alternative process routings,
machine exibility and workers promotion from one skill level to another in which the objective function is to minimize production cost, inter-cell material cost, machine costs, hiring, ring, training and salary costs. Ghotboddini et al. [15] presented
a multi-objective mixed-integer model for the DCMS, which considers cell formation, simultaneously, with labor assignment
to minimize the reassignment cost of human resource, over time cost of equipment and labors, and maximize the utilization
rate of human resource. Then, they solve their model with the Benders decomposition approach.
One of the other important factors in designing CMS is the system reliability. This factor can affect the system total efciency strictly. Machine breakdowns result in higher production costs, longer production period and other manufacturing
problems. A few studies have been developed to overcome machine breakdown challenges in the CMS design. Das et al.
[16] developed the CMS design to minimize the total system costs and maximize the machine reliabilities along the selected
processing routes, which provides a exible routing that ensures a high overall performance of the CMS. Chung et al. [17]
proposed an efcient tabu search (TS) algorithm based on a similarity coefcient to solve the CF problem with alternative
process routings and machine reliability considerations. Raee et al. [18] proposed an integrated cell formation and
inventory lot sizing problem to minimize some CMS costs. Moreover the process deterioration and machine breakdowns
are considered to make the model more practical and applicable.
In most studies, researchers have developed CMS problems under a static environment, in which the product mix and
demand rate are constant and known in one single period. However, considering a dynamic environment is more realistic.
In dynamic environments, a multi-period horizon is considered where the number of the product mix-set and demand rate
are different in each period. Therefore, the best cell conguration for one period may not be optimal for other periods [19].
See [20] and [21] as recent surveys. A dynamic cell formation problem and production planning are strongly inter-linked and
should not be solved separately and sequentially for following reasons:
In the real industrial environment, the production quantity usually is not equal to the demand quantity in each planning
horizon, and it may be gratied from inventory.
Production planning decisions should ascertain the amount of production to determine the machines type and number
that should be installed in each cell. On the other hand, due to the capacity consideration number of different machines
to be installed in each cell should be known in turn to determine the amount of production [22].
The usual constraints employed are: (1) inventory balance equations for making the inventory and/or shortages balanced
with those from the previous period, production quantity, and the demand quantity and (2) capacity constraints that ensure
the total workload for each resource (e.g., labor and machine) [23].
There are some studies in literature, in which the CMS and production planning are considered simultaneously to reach
an efcient production plan. Ah Kioon et al. [24] developed a model consisted of cell formation, system reconguration, multiple process routings, production planning, machine capacities and availabilities. Safaei and Tavakkoli-Moghaddam [25]
171
proposed a multi-period cell formation and production planning to minimize machine, inter/intra-cell movement, reconguration, partial subcontracting, and inventory carrying costs. Mahdavi et al. [22] developed the multi-period cell formation
and production planning with considering worker assignment to minimize machine related costs, reconguration cost,
inter-cell material handling cost, inventory holding and backorder costs, hiring, ring and salary costs.
Most studies in this area have been limited to deterministic approaches where all model parameters are assumed to be
known with certainty. While, in the real world, CMS parameters (e.g., processing time, part demand, product mix,
inter-arrival time and available machine capacity) often subject to uncertainties. It is no longer acceptable, to present results
without considering the uncertainties involved. Different types of approaches have been developed to address uncertainty
are stochastic programming approach, fuzzy programming approach, and robust optimization approach [26,27]. Some
surveys dedicated to the CMS under uncertainty are as follows.
Balakrishnan and Cheng reviewed cell formation problems considering uncertainty in demands and resources [28]. Some
studies in literature dealt with the uncertainty using fuzzy optimization (FO) techniques. Szwarc et al. [29] considered the
uncertainty in demand and machines capacity in a CMS problem, and resolved by a fuzzy approach. Rabbani et al. [30]
proposed a bi-objective cell formation problem with stochastic demand quantities and solved with a two-phase fuzzy linear
programming approach. The need to use approaches, which are immune to data uncertainty increases when there is the
possibility that one single uncertainty may change the optimal solution from optimal to non-optimal or even infeasible
one. Safaei et al. [31] proposed a new mathematical model, in which part demand and facility availability are considered
as fuzzy parameters. The proposed model has been solved using a fuzzy programming approach. Later they proposed an
extended fuzzy parametric programming approach to solve a dynamic cell formation problem considering the uncertain part
demand and machine capacity [32]. Azadeh et al. [33] proposed an integrated fuzzy multivariatefuzzy simulation approach,
in which fuzzy setup and processing times as well as multiple products are considered for their CMS model.
There are few studies in literature, in which stochastic optimization (SO) techniques are applied to nd an optimal solution. Ghezavati and Saidi-Mehrabad [34] proposed a CM model and assumed that processing and arrival times for parts are
stochastic. They solved the uncertainty using a queuing-based analysis. Ghezavati and Saidi-Mehrabad [35] applied a
scenario-based stochastic programming technique to solve the cell formation problem integrated with the group scheduling
decision, concurrently. Also, Renna and Ambrico [36] proposed a new approach to design a CMS. At rst, a main model is
developed by which the machines are grouped into manufacturing cells. The multi-purpose machines should be selected
and assigned to a remainder cell. In their research, the product demand is considered as an uncertain parameter. Then
two sub-models (considering the sub-periods) and allocation model are developed for detail planning of the manufacturing
system. However, this study has some challenges. The rst one is related to its strategy to cope with uncertainty. Actually the
scenario based considering of the demand, limits the demand occurrence possibilities in real world applications. The other
challenge is its incapability in nding the global optimal solution because of the sequential nature of solving the proposed
models. In order to nd an optimal solution, these models should be regarded, simultaneously
Another optimization approach, which addresses data uncertainty, has revealed under the name of robust optimization
(RO). This approach tries to nd a solution that copes with all uncertain data possibilities. Various RO approaches are developed. The idea behind RO is to consider the worst-case scenario without a specic distribution assumption. A body of a
literature has been dedicated to this approach. Several robust counterparts of linear optimization problems were proposed
since 1973. Soyster [37] proposed the rst robust model for linear optimization problems with uncertain data, who considered a deterministic linear optimization model, which is feasible for all data lying in a convex set. While, the model is very
conservative and assumes that all random parameters are equal to their worst-case value. However, Mulvey et al. [38] used
the term robust optimization for the rst time to introduce a new stochastic programming framework comprising a
scenario-based approach with a goal programming formulation [39]. Several robust formulations and applications were
introduced by Ben-Tal and Nemirovski [4042] and Ghaoui and Lebret [43]. They developed a framework of the robust
optimization in linear optimization that is less conservative, introducing a nonlinear term in the objective function. To cope
with over conservatism and nonlinearity issues in previous studies, Bertsimas and Sim [44,45] proposed a linear robust
optimization. In the CMS eld, Ghezavati et al. [46] proposed a mathematical model including cell formation and group
scheduling problems in a supply chain framework. They solved the proposed model by applying an approach based on
Mulvey et al. [38] considering part processing time and demand uncertainty. Table 1 summarizes the recent studies
conducted to design an efcient cellular manufacturing system.
In this paper, a robust optimization model is developed within the framework of Bertsimas and Sim [45], in which they
developed a different approach to control the conservation level in the solution. They model data uncertainties by a set of
bounded intervals and the solutions are deterministically immune to uncertain data in certain sets. To the best of our knowledge, the RO technique proposed by Bertsimas and Sim [45] is not applied in a CMS environment. Our motive to apply RO
instead of fuzzy optimization (FO) and traditional two-stage stochastic optimization (SO) are: (i) in traditional two-stage SO
huge number of scenarios may increase the model complexity and computational burden. Also, they need accurate denitions of probability distribution for uncertain data to cater with precise insight, while it can be costly to attain exact distribution parameters. From the other point of view the RO approach remains computationally tractable irrespective of its
number of uncertain parameters and RO denes uncertain coefcients through bounded intervals if exact distributional
information is unavailable [39]. (ii) FO related approaches increase the solving complexity and are typically difcult to be
solved in a reasonable computational time, especially in comparison to the RO approach, which is less sophisticated and
FO uses fuzzy numbers differ from real-valued numbers used in RO that are difcult to understand [47,48].
172
Table 1
Summary of the literature review.
Study
Types of problem
CF
GL
X
X
X
X
X
X
X
X
Presented paper
X
X
GS
RA
Approaches
PP
Concurrent
X
X
X
X
X
X
X
X
X
X
X
X
Problem uncertainty
Sequential
Certain
Certain
Certain
Certain
Certain
Certain
parameters
parameters
parameters
parameters
parameters
parameters
Certain parameters
Stochastic programming (demand uncertainty)
Fuzzy programming (demand and machine capacity
uncertainty)
Queuing based analysis (part arrival time uncertainty)
X
X
X
X
X
X
Furthermore, this paper integrates different concepts including production planning, operator assignment,
distance-based relocation, machine reliability, inter-cell layout design, alternative process routings and operators learning
effect. The rest of this paper is organized as follows. The proposed model and its robust counterpart, designated as a robust
DCMS (RDCMS) method, are introduced in Section 2. In Section 3, two illustrative examples are provided to demonstrate the
effectiveness of the suggested RDCMS model. Finally, the main conclusions are drawn in Section 4.
2. Mathematical description of the problem
In this section, we rst describe the deterministic nonlinear mathematical model, which comprises cell formation,
inter-cell layout design, production planning, operator assignment, machine reliability and alternative process routings, with
the aim to minimize machine breakdown cost, inter-intra cell part trip costs, machines relocation cost, inventory holding and
backorder costs and also operators training and hiring costs. Moreover, the proposed nonlinear model is linearized and subsequently a robust optimization approach is applied throughout the model as an optimization technique to cope with the
product processing time uncertainty. Indeed, the main contribution of this model is as follows:
A new model is proposed to incorporate with three important decisions (i.e., CF, RA and GL) of the CMS with production
planning simultaneously. It is a comprehensive model that can be used in many real-world applications.
A robust optimization approach [46] is developed for the new presented model. To our best knowledge, this framework
has not been adopted in the integrated DCMS and PP model. Within this framework an optimal solution, which is robust
against data uncertainties in part processing time, can be obtained.
Machine breakdown is considered to make the model more practical. It is one of the major factors inuencing on the performance of the CMS and can result in higher production costs, longer production period and other manufacturing
problems.
Additionally, this paper provides a larger coverage of real world manufacturing factors, such as an operator learning
effect, alternative process routings, machine relocation and the like.
As mentioned in the rst contribution of the proposed model, the framework is useful for different industrial companies,
such as automobile companies. A CMS model considered CF, GL and RA simultaneously is applied to an automobile industry
as presented in [3]. In the proposed model, not only three of these decisions CF, GL and RA are integrated, but also production
planning (PP) and to cope with the CMS uncertain parameters and breakdowns, both parts processing time uncertainty and
machine breakdown rate are considered, simultaneously. Therefore, the proposed framework has more capability to be used
in the automobile industry. Since, the proposed methodology is close to the practical conditions of the automobile company.
For instance, the model gives an automobile company the chance not only to nd the optimal decision for cell formation,
group layout and resource allocation of the automobile company, but also to obtain the minimum production planning
related cost. Also, by considering the uncertainty of the processing time, the production line of the company will be modeled
more accurately. For example, it is possible that a worker cannot nish a part at the certain time, and it makes the production
173
line to stop. Also, stopping the production line may happen if the breakdown of the machine is not considered. Consequently,
the proposed model has the potential to be applied to an industrial company, such as an automobile one.
The following assumptions are taken into account for the development of the mathematical model.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
2.1. Notations
2.1.1. Indices
M
I
C
L
H
Number of machines
Number of parts
Number of machine cells should be constructed
Number of candidate locations to be a cell (L P C)
Number of production periods
HRhw
174
K hri
fU ri
t0m;h
A1
A2
1;
0;
1;
X hc;l
0;
1;
h
EMw
0;
N hm;c
ohw;m
Shw;m
V hr;i
Ihi
Bhi
Q hi
QBhi
175
Material handling costs: Eqs. (1a) and (1b) represent inter- and intra-cell material handling costs, respectively. An
inter-cell part trip only occurs, if operations of the same part need to be processed in more than one cell. Also, Eq. (1a)
tries to locate cells in an efcient way. And if two consecutive operations in one routing are allocated to the same cell,
but to different machines the intra-cell material part trip occurs. There are always some production limitations (e.g., total
available processing time of machines or operators), which prevent the system from satisfying all demands in a production period, therefore the production quantity in a period can satisfy the demand or not due to these limitations.
h
Ri K ri1 C
H X
I X
C X
L X
L
X
X XX
A1 :Disl;l0 :V hr;i : Q hi :NhUm ;c :X hc;l :NhUm1 ;c0 :X hc0 ;l0
Min OF
ri
h1 i1 r1 m1 c1 c0 1 l1 l0 1
h
ri
1a
Ri K ri1 C
H X
I X
X
XX
ri
ri
1b
Production planning related costs: Eq. (1c) represents the inventory holding cost, which occurs due to keeping inventory
in warehouses, having delay to fulll the demand of the customer leads to backorder cost Eq. (1d).
H X
I
X
ahi : Ihi ;
1c
h1 i1
H X
I
X
bhi : Bhi :
1d
h1 i1
Machine breakdown cost: In many studies existed in the literature, it is assumed that machines are reliable and can
process parts without any breakdown; while, one of the major factors inuencing on the performance of the CMS
is machine breakdowns that result in higher production costs, longer production period and other manufacturing
problems. It is assumed that machine breakdown cost is known and is based on its repair, install/uninstall costs,
machines reliability follows an exponential distribution with a known failure (breakdown) rate, and in its operating
time can be dened as follows.
MTBF
1
:
k
Based on these denitions a machines total breakdowns over its production periods can be obtained through dividing the
production time by the MTBF and then multiplying this quantity by the machine breakdown cost. Eq. (1e) searches for
routing plans for all the part types that eventuate in a minimum overall system failure rate, and therefore maximizes
the reliability of the system.
h
Ri K ri
H X
I X
X
X V hr;i : Q hi : BRUm
ri
: thi; Um :
MTBF Umri
ri
h1 i1 l1 m1
1e
Machines relocation cost between periods is minimized using Eq. (1f). It is assumed that these relocations are dependent
on distances between two cells machines travel.
H1 X
M X
C X
C X
L X
L
X
h
h1
0
cm : Nhm;c : Nh1
m;c0 : X c;l : X c0 ;l0 : Disl;l :
1f
h1 m1 c1 c0 1 l1 l0 1
Operator hiring & training costs: At last operators hiring and training costs are minimized by terms (1 g) and (1 h),
respectively. It is assumed that if an operator is trained to work with a particular machine this learning effect will
be considered for succeeding periods without any extra training cost.
H X
W X
M
X
HRhw ohw;m ;
1g
h1 w1m1
H X
W X
M X
C
X
1h
176
s.t.Constraints (2) and (3) ensure that the number of machines for one cell is not exceeded lower and upper bounds of a cell
size, in which the upper bound is used to prevent all machines from being assigned to a single cell.
M
X
Nhm;c P Lowc
8h; c;
m1
M
X
Nhm;c 6 UPc
8h; c:
m1
Constraint (4) indicates that each machine type can be assigned to a single cell.
C
X
Nhm;c 1 8h; m:
c1
Constraint (5) indicates that just a single process routing will be selected as the optimal route for each part. This routing will
be selected only if that part type will be produced in the corresponding period.
h
Ri
X
V hr;i QBhi
8h; i:
r1
Constraint (6) ensures that each cell should be constructed in only a single location.
L
X
X hc;l 1 8h; c:
l1
Constraint (7) indicates that only single cell can be constructed in one location.
C
X
X hc;l 6 1 8h; l:
c1
Constraint (8) ensures that the available machine time capacity that is consisted of total machines time capacity is not
exceeded.
h
Ri
I X
X
i1 r1
8h; m:
Constraint (9) makes the inventory and/or backorders balanced with those from the previous period, production quantity,
and the demand quantity.
Q hi Dhi Ih1
Bih1 Ihi Bhi
i
8h; i:
Constraint (10) determines the number of operator types should be assigned to a particular machine.
W
X
EMhw;m NMOm
8h; m:
10
w1
Constraint (11) guarantees that each operator can be assigned to only one machine.
M
X
EM hw;m 6 1 8h; w:
11
m1
Constraint (12) is a logical constraint, which guarantees that the number of particular operator type can take a positive value
only if that operator type is assigned to a machine.
8h; w; m:
12
Constraint (13) ensures that the total available time for operators is not exceeded.
h
Ri
I X
W
X
X
V hr;i Q hi t hi;Um 6
t 0h;w ohw;m
i1 r1
ri
8h; m:
13
w1
Constraint (14) states that if an operator is capable of working with a specic machine or is trained to work with it, this
learning effect will be considered in succeeding periods. The trained operator for a particular machine in a period can work
with that machine in the next periods without any extra training cost.
8h 1; . . . ; H 1; w; m:
177
14
Constraint (15) is a logical constraint, which guarantees that the volume of a part type should be produced in a particular
period can be a positive quantity only when its corresponding binary variable is equal to 1.
Q hi 6 A1 QBhi
8h; i:
15
Nhm;c 2 0; 1 8h; m; c;
16
X hc;l 2 0; 1 8h; c; l;
17
QBhi 2 0; 1 8h; i;
18
V hr;i 2 0; 1 8h; r; i;
19
20
21
22
8h; w; m:
2.2. Linearization
In this section, the linearization of the nonlinear model is applied [2,3]. The nonlinearity of the model is due to Eq. (1a),
(1b), (1e), (1f), (1 h) and Constraints (8), (13) and (14). Therefore, to transform mentioned terms into linear terms
non-negative variables are introduced as follows.
VQ hr;i Q hi V hr;i ;
SEhw;m Shw:m EM hw;m ;
whUm ;Um1 ;c VQ hr;i NhUm ;c :NhUm1 ;c ;
ri
ri
ri
h
;c;l;U m1
;c0 ;l0
Um
ri
ri
VQ hr;i
ri
ri
h1
h
h1
ghm;c;c0 ;l;l0 Nhm;c Nm;c
0 X c;l X c 0 ;l0 ;
23
24
25
26
8h; w; m;
27
28
29
30
31
ri
ri
ri
ri
ri
ri
ri
ri
ri
ri
178
uhUm ;c;l;Um1 ;c0 ;l0 6 VQ hr;i A1 4 N hUmri ;c NhUm1 ;c0 X hc;l X hc0 ;l0 8h; r; i; m; c; c0 ; l; l0 ;
ri
ri
ri
uhUm ;c;l;Um1 ;c0 ;l0 P VQ hr;i A1 4 NhUmri ;c NhUm1 ;c0 X hc;l X hc0 ;l0
ri
ri
ri
8h; r; i; m; c; c0 ; l; l0 ;
32
33
34
ri
8h 1; . . . ; H 1; m; c; c0 ; l; l0 ;
h
h1
4ghm;c;c0 ;l;l0 6 Nhm;c Nh1
m;c0 X c;l X c0 ;l0
35
h1
h
h1
ghm;c;c0 ;l;l0 P Nhm;c Nm;c
8h 1; . . . ; H 1; m; c; c0 ; l; l0 ;
0 X c;l X c 0 ;l0 3
36
37
8h; w; m; c;
38
8h; w; m; c;
39
40
After substituting new variables, the linear form of the model is as follows.
h
Min OF
Ri K ri1 C
H X
I X
C X
L X
L
X
X XX
A1 :Disl;l0 :uhUm ;c;l;Um1 ;c0 ;l0
ri
h1 i1 r1 m1 c1 c0 1 l1 l0 1
ri
Rhi K hri1 C
H X
I X
X
XX
A2 :whUm ;Um1 ;c
ri
h1 i1 r1 m1 c1
ri
H X
I
X
Ri K ri
H X
I X
X
X VQ hr;i : BRUm
ri
: thi; Um
MTBF m
h1 i1 r1 m1
ri
ri
H1 X
M X
C X
C X
L X
L
X
"
#
H X
W X
M
C
X
X
c1
s.t.
Constraints (2)(7), (9)(12), (15)(40)
h
h
Sh1
w;m Sw;m SEw;m
8h 1; . . . ; H 1; w; m;
41
Ri
I X
X
VQ hr;i t hi;Um 6 t0h;m
i1 r1
ri
8h; m;
42
Ri
I X
W
X
X
VQ hr;i t hi;Um 6
t0h;w ohw;m
i1 r1
ri
8h; m:
43
w1
179
the enforced uncertainties can be controlled through a parameter named conservation level of the optimal solution [44].In
this section, a robust approach developed by Bertsimas et al. [45] for discrete optimization problems with uncertain parameters is presented, which controls the conservatism level of the solution. The obtained solutions from a robust optimization
approach guarantees more situations, even worst ones. The important goal of the robust framework is to present an optimal
solution that is robust against data uncertainties in part processing time.
Consider the following nominal linear optimization problem with a set of j variables.
Min cT x;
s:t:
A x 6 b;
44
lb 6 x 6 ub:
Suppose that the uncertainties affect both the elements of the vector C and matrix A = an;j and these elements are modeled as
~n;j , j 2 J n . Therefore, uncertain elements can be dened using mean value
symmetric and bounded random variables ~cn and a
and range of each uncertain element as follows.
45
~n;j an;j a
^n;j ; an;j a
^n;j a
~n;j 2 A:
a
46
Conversion level (CL), symbolized by Cn (n = 0, 1, . . ., CN) is introduced for robustness intentions and adjusting the model
robustness given the level of solution conservatism which adopts different values in the interval 0; jJ n j, where J n is a set
^n;j > 0g.
comprises uncertain elements of either the objective function (n = 0), or nth constraint (n = 1, . . ., CN), J n fjja
Therefore, C0 and Cn represent the conservation level value for the objective function and nth constraint, respectively.
The nonlinear robust counterpart of the model (44) is as follows. For more details, the interested reader may refer to [45].
Min
X
c j xj
j
max
(
)
X
^cn :jxj j C0 bC0 c ^ct0 jxt0 j ;
47
j2S0
s.t.
X
an;j : xj
j
(
X
max
lbj 6 xj 6 ubj
)
^n;j :jxj j Cn bCn c: a
^n;tn :jxtn j
a
6 bn n 1; . . . ; CN;
48
j2Sn
j 1; . . . ; DN:
By applying duality theory to (47) and (48), we have an equivalent linear optimization as follows.
X
X
Min cj xj z0 C0
p0;j ;
j
49
j2J 0
s.t.
X
X
an;j xj zn : Cn
pn;j 6 bn
j
n 1; . . . ; CN;
50
j2J n
z0 p0;j P ^cj yj
^n;j yj
zn pn;j P a
8j 2 J 0 ;
8j 2 Jn ; n 1; . . . ; CN;
51
52
yj 6 xj 6 yj
j 1; . . . ; DN;
53
lb j 6 xj 6 ubj
j 1; . . . ; DN;
54
zn P 0 n 0; 1; 2; . . . ; CN;
55
pn;j P 0 8j 2 J n ; n 0; 1; . . . ; CN;
56
yj P 0 j 1; . . . ; DN:
57
Therefore, a robust formulation for the proposed model to account for uncertainty in parts processing time can be developed
by introducing a set of symmetric bounded intervals, which represent the uncertainty of the parts processing time as follows.
h
i
~thi;m 2 thi;m ^t hi;m ; t hi;m ^t hi;m
8h; i; m:
58
180
Table 2
Part input data for Example 1.
Part
1
2
3
Period 1
Period 2
Process sequence
Demand
ai ($)
bi ($)
Process sequence
Demand
ai ($)
bi ($)
M1-M4-M3
M2-M3
M1-M3
M1-M2
M4-M2-M1
M4-M3-M1
M3-M1
M2-M1-M3
2, 1,
4,5
2, 5
2,3
4, 1,
4, 2,
2, 5
1, 8,
60
16
6.5
19
18
90
7.25
16
5
5
65
6.25
19
2, 7
1, 7
2, 1
8,2
7, 1, 5
5, 5, 7
120
105
M4-M2
M1-M3
M3-M1
M2-M4
M4-M2-M3
M1-M3-M4
120
6.5
17
60
7.5
17
M3-M1
M2-M4
5, 1, 8, 5
100
16
A simple problem is presented in the Appendix to justify clearly how the proposed robust optimization works.
3.2. Proposed RO for the integrated model
The proposed model integrated with the DCMS and PP problems has the robust counterpart formulation as follows.
h
Ri K ri1 C
H X
I X
C X
L X
L
X
X XX
Min OF
A1 :Disl;l0 :uhUm ;c;l;Um1 ;c0 ;l0
ri
h1 i1 r1 m1 c1 c0 1 l1 l0 1
Rhi
ri
K hri1
H X
I XXX
C
X
A2 :whUm ;Um1 ;c
ri
h1 i1 r1 m1 c1
ri
H X
I
X
h1 i1
h
59
Ri K ri
H X
I X
I
X
X VQ hr;i : BRUm
X
ri
: thi;Um z0 : C0
p0;i
MTBF m
h1 i1 r1 m1
ri
ri
i1
H1 X
M X
C X
C X
L X
L
X
"
#
H X
W X
M
C
X
X
h
h
h
c1
s.t.
Constraints (2)(7), (9)(12), (15)(41), and
h
Ri
I X
I
X
X
VQ hr;i t hi;Um zh;m Ch;m
phi;m 6 t 0h;m
i1 r1
ri
8h; m;
60
i1
Ri
I X
I
W
X
X
X
VQ hr;i t hi;Um zh;m Ch;m
phi; m 6
t 00h;w ohw;m
i1 r1
z0 p0;i P
ri
i1
H X
M
X
^th y h
i;m
0;t
h1 m1
i;m
8h; m;
61
w1
z0 P 0; p0;i P 0;
8i;
62
Table 3
Operator input data for Example 1.
Machines
Operators
W1
W2
W3
W4
W5
Hiring cost
M1
M2
M3
M4
Period 1
Period 2
105
0
100
115
120
100
120
115
100
0
105
0
120
115
125
0
105
115
120
130
50
60
70
80
70
80
70
80
50
50
181
To
L1
L2
L3
L1
L2
L3
0
1
2
1
0
1
2
1
0
Table 5
Machine input data for Example 1.
Machines
Time
Capacity (min)
MTBF (min)
Breakdown cost ($)
Period 1
Period 2
M1
M2
M3
M4
4550
4600
5400
750
4500
4450
3660
650
4600
4500
3000
600
4700
4750
2940
700
Table 6
Range of variation for Ch;m for Example1.
Machine
Period 1
Period 2
M1
M2
M3
M4
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
4]
4]
4]
2]
4]
4]
4]
4]
Table 7
Optimal inter-cell layout and machine grouping for Example 1.
Cell
Period 1
C1
C2
Period 2
Cell location
Machines
Cell location
Machines
L3
L2
M2, M4
M1, M3
L2
L3
M1, M3
M2, M4
Table 8
Optimal part routings for Example 1.
Part
Period 1
Period 2
R1
R2
R2
R1
R3
R1
R1
R1
Table 9
Objective Function value and its components for Example 1.
OF
($)
Inter-cell cost
($)
Intra-cell cost
($)
Inventory cost
($)
Backorder cost
($)
Breakdown cost
($)
Relocation cost
($)
Training cost
($)
Hiring cost
($)
3624
247
158
209
325
1435
230
1020
Table 10
Optimal production plan for Example 1.
Part
Period 1
Inventory
1
2
3
4
Period 2
Backorder
11
5
23
Production
Demand
71
100
88
60
60
105
65
60
Inventory
Backorder
Production
Demand
100
95
97
96
120
90
120
100
182
Table 11
Optimal operator allocation for Example 1.
Period 1
Period 2
M1
Operators
W1
W2
W3
W4
W5
M2
M3
M4
M1
M2
M3
1
3
3
1
zh;m P 0; phi; m P 0;
i;m
M4
8h; i; m;
63
y0;th 6
i;m
Ri
X
VQ hr;i : BRUm
ri
r1
MTBF Umri
6 y0;th
i;m
y0;th P 0;
i;m
8h; i; m;
64
yth 6
i;m
Ri
X
VQ hr;i 6 yth
i;m
r1
yth P 0;
i;m
8h; i; m:
65
PRhi
By comparing constraints (60) and (61) with (50), it is gured out that r1
VQ hr;i xj , n = h m and the number of the conservation levels for Constraints (60) and (61) are equal to C1 ; C2 ; . . . ; Chm . Moreover, in Constraints (60) and (61), for each m
and h, the number of uncertain sets elements are equal to the number of parts that machine m produces in period h, i.e.
J n 0; jJ n j 0; Ih;m , where Ih;m is the number of parts produced by machine m in period h. As well, by comparing the fourth
P PRhi PK hri VQ hr;i : BRUmri
row of (59) with (49), it is revealed that Hh1 r1
m1 MTBF m xj . Also, by considering Constraints (62) and (51), a set of
U
ri
uncertain elements can vary from zero to the total number of all uncertain variables of the set, i.e. J 0 0; jJ 0 j 0; I and
Constraint (62) is only function of I. While, ^thi;m is function of H, M and I. Therefore, the summations of H and M are used
in Constraint (62).
4. Numerical examples
In this section, to validate the performance of the proposed RDCMS model two illustrative examples are implemented in
the model. The simulations have been performed with CPLEX solver on a 64-bit computer with 4 GB of RAM and Intel Core i5
CPU. The data sets related to the rst example are shown in Tables 26. The rst example consists of four part types, four
machine types, ve operator types, two cells and two periods. Table 2 shows the information related to parts such as alternative routings, processing times, demand quantity, inventory holding and backorder costs, for example part type one with
the demand quantity of 60 has two different routings, its rst routing consists of three operation sequences that should be
processed on machine M1, M4 and M3, respectively. Due to nowadays competitive manufacturing environments and a short
life cycle of one product, rms based on the feedback taken from their customers may add/remove some characteristics
to/from their product(s) to better compete with other rivals. Therefore, the procedure of one product type may change from
one period to another period. In this study, the routings are adopted in a way that these changes can be incurred as well.
Fig. 1. Objective function value versus conservation level and range of uncertain processing time.
183
Operators
W1
W2
W3
W4
W5
W6
W7
W8
W9
W10
W11
W12
W13
W14
W15
Hiring cost
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
M13
M14
Period 1
Period 2
0
100
90
0
100
100
80
0
120
110
0
90
0
80
0
100
0
90
120
100
0
60
90
0
90
80
0
110
90
80
0
90
110
100
0
90
0
150
90
0
50
60
100
0
70
130
0
100
90
80
70
0
110
120
0
100
140
90
100
80
150
120
0
0
110
100
90
0
100
120
110
150
120
0
60
120
0
120
90
0
70
0
100
0
120
100
100
140
120
0
0
100
90
80
80
70
0
70
100
0
90
60
100
0
90
90
95
0
100
0
100
120
0
110
90
0
60
90
80
60
100
0
90
100
120
0
70
70
0
120
100
0
90
120
50
0
120
100
0
100
120
0
100
100
90
80
78
90
0
90
90
0
120
100
0
100
130
0
100
90
100
90
0
100
0
100
120
0
100
120
0
110
0
90
80
90
0
90
120
0
90
80
0
110
0
70
0
90
90
70
0
110
130
90
60
0
120
110
0
110
0
100
100
0
90
90
110
90
120
50
50
60
70
60
75
50
80
65
55
80
70
75
55
55
60
80
65
65
80
70
60
70
65
50
60
75
80
50
55
50
Operator training and hiring costs are given in Table 3, Note that zero values in training cost means operators are able to
work with the machine at the beginning of the horizon. The three locations exist in the shop oor and the distances between
these locations are presented in Table 4. Regarding the machines, Table 5 provides their data. The data contains machines
time capacity, Mean Time Between Failure and breakdown cost.
Based on denition in Section 3.1, C1 ; C2 ; . . . ; C8 related to Example 1 (n = h m = 2 4 = 8), can vary according to Table 6
and Reliability expkt can vary through MTBF 1k. For Example 1, we adjust the processing time uncertainty ^t hi;Um equal
ri
to 50% of the nominal processing time and Cn 0:5 jJ n j. Note that when Cn = 0 the constraints will be equivalent to the
deterministic form and by changing Cn 2 0; jJ n j there will be the exibility of modifying the robustness of the method pertaining to the conservatism level of the solution.
The model solution determines the cell formation, number of operators and their allocation to machines, inter-cell layout
and production plan for each period. Due to different costs and simultaneously considering machine breakdown, layout and
production planning issues, the obtained objective function values cannot be compared to the previous works. The optimum
solutions are shown in Tables 711.
Table 7 shows the optimal solution related to inter-cell layout and machine grouping. For instance, in the period 1, cell C1
is constructed on location L3 and C2 is constructed in location L2, and also machines M2, M4 and M1, M3 are allocated to C1
and C2, respectively. Table 8 represents the optimal selected routings for all part types in two periods. For instance, in the
period 2, to process part type 3, it should go through machines M4, M2 and M3 subsequently under the rst routing. The
optimal objective function value and its components are shown in Table 9. As can be seen in Table 9, the relocation cost
is equal to zero, since machines have not been relocated at the beginning of the period 2. Table 10 shows the optimal production plan, the demand quantity for each part type can be satised through production, backorder and/or inventory. For
instance, in the period 1, the demand of the part type 3 is 65, while 88 parts are produced; therefore, 23 parts will be kept for
the next period as an inventory, which can satisfy a part of the next periods demand. On the other hand, only 100 parts of
part type2 have been produced while the demand quantity is 105. Hence, the demand cannot be satised completely in the
rst period and 5 units are considered as backorder, but this backorder can be satised in the second period.
Table 11 shows both the optimal number of operators and their assignment to different machines. For instance, 3 operators of type W2 are assigned to machine M1. By considering Tables 2, 7 and 8, it can be gured out how inter and intra-cell
part trips happen. For instance, in the period 2, part type three should be processed on machines M4, M2 and M3, respectively. On the other hand, machines are assigned to C1, C1 and C2, respectively. Therefore, part type 3 after being processed
on M4 in cell C1 will experience an intra-cell part trip to machine M2 placed in the same cell and at last the part would experience inter-cell trip to machine M3 in cell C2. Hence, processing of part type 3 in period 2 incurs both intra-cell and
inter-cell costs. By considering (1a) and (1b), these two costs can be computed as follows.
13 ;1
13
0:25 1 120 1 1 30
Note that VQ hr;i 6 Q hi A1 :1 V hr;i 8h; r; i indicates to machine M4, the rst machine in the rst routing of part type 3.The
impact of the aforesaid uncertainty is depicted in Fig. 1.
According to this gure, the objective function (OF) value is the function of both conservation level and the range of
uncertain parts processing time. By increasing conservation level and/or range of uncertain parts processing time, the OF
increases. With the robust optimization approach the desire to stay on the safe side can be achieved by enlarging uncertainty
184
set. In the non-presence of conservation level (i.e., deterministic model), the optimal value is 2954.069. On the other hand,
with maximum conservatism (i.e., worst-case) and 1 ^thi;Um the optimal value is increased by 64% to 4849.597.
ri
To verify the applicability of the proposed model and approach in a larger scale, the second example consists of 24 part
types, 14 machine types, 15 operator types, 2 cells and 2 periods. The related input data for Example 2 is given in Tables 12
16. Then, the optimal obtained results are shown through Tables 1720 and Fig. 2.
Based on denition in Section 3.1, C1 ; C2 ; . . . ; C28 of the second example (n = h m = 2 14 = 28) can vary according to
the Table 16 and C0 varies in 2SEhw;m 6 Shw;m EMhw;m 8h; w; m. Moreover, in the second example, the processing time uncertainty ^thi;Um is equal to 50% of the nominal processing time and Cn 0:5 jJ n j.
ri
As it is evident, Fig. 2 depicts the optimal inter-cell layout, machine grouping and operator assignment for Example 2 in
two periods, in which cell conguration for the rst period differ from the second period. For instance, 4 operators of type
W13 are assigned to machine M1 in the period 1,while 5 operators are assigned to machine M11 in the period 2, in period 1
operator W12 is not hired while in period two, one operator W12 is hired to work with machine M12. Although no cells are
dedicated to location L2 in both periods, the locations of some machine types and cells are changed in period two relative to
period one.
To better ascertain the performance of the proposed model and developed approach, 10 test instances are solved. It is
obvious that by increasing the problem size the computational time to nd the optimal or near-optimal solution increases.
As it is evident in Table 21, the presented model with the developed approach is able to return the optimal solution in a reasonable computational time for small and medium problems (instances No. 16) with an optimality gap of 0%. Medium-large
problems (i.e., instances Nos. 7 to 9) are solved with the maximum optimality gap of 0.072% in the maximum computational
time of 1783 s. Moreover, a larger-sized problem (i.e., instance No. 10) is solved with the computational time of 4371 and
optimality gap of 11.308%.
One of the main questions possible to be asked is the importance of concurrent consideration of the three main decisions
of the CMS. As stated by Bagheri and Bashiri [3], these goals are interrelated and should be considered simultaneously in
order to obtain an optimal solution. In this paper, this fact is investigated as follows. We rst solve the CF problem and
the obtained optimal value is equal to 11,250. Then, the model is solved by xing the obtain results for CF problem and nding the optimal solution for the RA problem. The total objective function value is 20,550. However, by concurrent solving of
these two decisions, the objective value of 16,482 is obtained. In large-scale industries, it is so necessary to design a manufacturing system in an optimal way. So as mentioned previously, this model can be applied in such companies.
Moreover, as a sensitivity analysis, the impact of MTBF on the OF components and optimal processing routes are illustrated through Tables 22 and 23. According to Table 22, by increasing MTBF, in other words, by decreasing failure rate of
machines, the total OF value decreases and by considering changes to components costs. We can understand that these
changes in the OF are mainly related to changes in breakdown costs, as machine MTBF increases, from 0.25 MTBF to
1.75 MTBF, the OF value and breakdown cost decreases about 52% and 80%, respectively. Table 23 demonstrates changes
in parts routing with relative to the current status (1 MTBF). For instance, when 1 MTBF increases to 1.75 MTBF, to
overcome this change in MTBF, part type 3 in period one will be processed on machines M2 and M11 under routing R2
instead of being processed on machines M3 and M13 under routing R1 (also see Table 18).
The impact of conservation level on the OF for Example 2 is depicted in Fig. 3. As it is expected by increasing the model
conservatism, the model becomes more immunized against parts processing time uncertainty and the OF increases. In the
Table 13
Distance matrix between candidate locations for Example 2.
From
L1
L2
L3
L4
L5
L6
To
L1
L2
L3
L4
L5
L6
0
1
2
1
2
3
1
0
1
2
1
2
2
1
0
3
2
1
1
2
3
0
1
2
2
1
2
1
0
1
3
2
1
2
1
0
Table 14
Machine input data for Example 2.
Machines
Time Capacity (min)
MTBF (min)
Breakdown cost ($)
Period 1
Period 2
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
M13
M14
7000
6900
5400
705
7800
7500
3030
800
7600
7000
3300
600
7500
7300
4900
750
7600
7500
3200
850
7000
7500
5010
650
7200
7400
4050
760
7700
7900
3700
890
7600
7500
4100
750
8000
7900
3200
860
7700
7800
3010
650
6000
7900
4200
670
7500
7300
3500
600
7700
7500
4300
700
185
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Period 1
Period 2
Process sequence
Demand
ai ($)
bi ($)
Process
Sequence
Demand
ai ($)
bi ($)
M1-M7
M4-M7
M5-M7
M3-M13
M2-M11
M2-M3-M13
3-4
4-5
2-5
3-6
4-5
3-4-5
100
95
6
5
14
14
M1-M7
M5-M7
3-4
2-3
60
100
5
4
15
14
100
4.5
16
M2-M11
4-3
70
15
95
16
14
2-3-6
4-2-6
5-4
2-1-3-4
4-5-1
85
16
2-3-4
2-3-5
3-2-4
100
M6-M9-M14
M12-M11-M14
M1-M13
M1-M7-M10M13
M3-M5-M12
M4-M7
M5-M8
M6-M8-M9-M13
M2-M12-M14
M6-M9
M6-M8
M7-M10
M9-M14
M6-M8-M9
M6-M12-M14
M6-M8
M9-M14
M6-M8
M8-M13
M6-M8-M9
M3-M10-M14
M6-M8
M6-M13
M4-M8-M7
M4-M5-M7
M13-M9
M1-M5
M4-M5
M4-M9-M7
M4-M5-M7
M6-M10
M7-M11
M6-M9
M9-M14
M1-M6-M13
M1-M6-M11
M10-M3
M2-M3
M2-M3-M13
M2-M3-M11
M12-M11-M14
65
13
85
90
6
5
13
14
M1-M13
M4-M5-M12
M3-M5-M12
5-4
2-4-3
1-4-5
70
80
5
5
15
15
3-5
4-6
3-5-6-8
3-9-2
3-2
3-5
4-6
3-5
3-7-6
3-4-6
4-6
2-5
5-6
6-3
2-3-4
4-5-6
5-4
5-3
1-4-6
1-3-6
8-1
3-4
6-7
4-3-2
4-2-2
5-6
4-2
5-6
6-4
2-3-4
2-3-6
5-6
2-6
100
14
13
4.5
16
2-4
2-3
5-1-3
60
95
M4-M7
M4-M8
M2-M12-M14
80
15
75
17
M6-M8
4-2
80
16
90
17
M9-M10
7-6
70
14
95
17
15
15
3-2-5
3-1-4
3-7
80
90
M6-M8-M14
M6-M12-M9
M9-M14
70
14
110
14
M9-M14
5-4
90
10
100
10
M3-M10-M14
4-5-6
90
15
85
18
14
13
3-8
5-4
2-5-3
70
100
M12-M5
M6-M13
M4-M5-M7
90
15
90
17
M1-M5
2-4
95
13
100
100
5
4.5
15
15
4
6
13
12
14
90
11
90
15
4-5
1-3-1
1-7-1
3-5
2-5
2-5
90
90
100
M4-M5
M7-M5-M7
M7-M9-M7
M7-M11
M3-M11
M9-M14
90
15
80
14
16
11
3-6-7
2-3-7
4-5
90
90
M1-M6-M11
M4-M5-M11
M2-M3
80
12
Table 16
Range of variation of Ch;m for Example 2.
Machine
Period 1
Period 2
M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11
M12
M13
M14
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
[0,
5]
4]
5]
5]
7]
11]
8]
9]
10]
5]
4]
4]
9]
7]
4]
4]
5]
5]
8]
4]
6]
3]
6]
2]
6]
5]
3]
7]
186
Table 17
Optimal inter-cell layout and machine grouping for Example 2.
Cell
Period 1
C1
C2
C3
C4
C5
Period 2
Cell location
Machines
Cell location
Machines
L5
L6
L4
L1
L3
M4-M5
M1-M10-M14
M2-M11
M3-M7-M9-M12-M13
M6-M8
L1
L6
L3
L5
L4
M7-M9-M12-M13
M1-M10-M14
M6-M8
M2-M4-M11
M3-M5
Table 18
Optimal part routings for Example 2.
Part
10
11
12
Period 1
Period 2
R1
R1
R1
R1
R1
R1
R1
R2
R1
R1
R1
R1
R1
R2
R1
R2
R2
R1
R1
R1
R2
R1
R2
R1
Period 1
Period 2
13
R2
R1
14
R1
R1
15
R1
R1
16
R2
R2
17
R2
R1
18
R2
R1
19
R1
R1
20
R2
R1
21
R2
R2
22
R2
R1
23
R1
R1
24
R2
R1
Table 19
Objective function value and its components for Example 2.
OF ($)
Inter-cell
cost($)
Intra-cell
cost($)
Inventory
cost($)
Backorder
cost($)
Breakdown
cost($)
Relocation
cost($)
Training
cost($)
Hiring
cost($)
22845
4659
271
178
297
9205
1160
820
6255
Table 20
Optimal production plan for Example 2.
Part
Period 1
Inventory
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Period 2
Backorder
6
10
20
2
Production
Demand
100
95
100
95
85
92
90
100
95
75
90
95
96
100
100
85
100
90
100
100
100
90
100
88
100
95
100
95
85
85
90
100
95
75
90
95
90
110
100
85
100
90
100
100
100
90
80
90
Inventory
Backorder
Production
Demand
60
100
70
100
65
63
80
60
80
80
70
80
64
100
81
70
90
95
90
90
90
90
70
82
60
100
70
100
65
70
80
60
80
80
70
80
70
90
90
70
90
95
90
90
90
90
90
80
absence of conservation level, the model converts to deterministic form with the OF value of 18,175 and by considering the
highest level of conservatism the OF value increases to 27,925.Although, by increasing the conservation level (CL), the objective function increases; it gives the risk-averse decision maker the chance to reduce the risk of the problem in the real condition. He/she considers the model very conservative to immunize the obtaining decisions against risks may happen in the
real world. On the other hand, the risk-seeker decision maker tries to decrease the cost of the problem and does not pay more
attention to the risks. Hence, the risk-seeker manager prefers to consider a little conservation level for the model.
187
Fig. 2. Cell conguration schema for Example 2. Note: numbers in parentheses indicates to the numbers of operators assigned to machines.
Table 21
Computational results from different-sized problems.
No.
IMWCH
No. variables
No. constraints
Objective function
Gap (%)
1
2
3
4
5
6
7
8
9
10
44522
86722
12 8 9 3 2
16 10 11 4 2
20 12 13 4 2
24 14 15 5 2
28 16 17 5 2
32 18 19 5 2
36 20 21 5 3
40 22 23 6 3
1402
2762
12807
43686
54768
116343
132243
154656
286148
372144
3560
7013
35413
124199
155562
331142
375623
439469
807685
1065439
3624
6925
10631
16352
19159
22845
26256
32195
42101
46907
24
95
264
863
1142
1305
1647
1783
2450
4371
0
0
0
0
0
0
0.013
0.042
0.072
11.308
Therefore, adjusting the conservation level is a trade-off between increasing the cost or decreasing the risks; and an experienced decision maker can adjust CL by his/her experience, historical and forecasted data. For instance, if the experienced
manager of an automobile company that workers can work on a specic time or the MTBF of the machines is low, then
he/she can consider a risk-seeker model with a little CL to increase the companys prot and vice versa.
188
Table 22
Effect of MTBF on OF value and its components.
MTBF
OF ($)
Inter-cell
cost ($)
Intra-cell
cost ($)
Inventory
cost ($)
Backorder
cost ($)
Breakdown
cost ($)
Relocation
cost ($)
Training
cost ($)
Hiring
cost ($)
0.25 MTBF
0.50 MTBF
0.75 MTBF
1 MTBF
1.25 MTBF
1.5 MTBF
1.75 MTBF
38,321
31,677
25,776
22,845
20,851
19,609
18,358
4102
3840
3518
4659
3860
3465
3606
273
410
335
271
395
349
305
105
90
161
178
173
164
167
376
276
246
297
162
140
283
25165
18196
12366
9205
7340
6251
5024
1920
2070
2180
1160
2030
2510
2150
980
630
840
820
826
750
723
5400
6165
6130
6255
6065
5980
6100
Table 23
Effect of MTBF on parts routing (Period 1/Period 2).
Part
0.25 MTBF
0.50 MTBF
0.75 MTBF
1 MTBF
1.25 MTBF
1.5 MTBF
1.75 MTBF
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
R1/R1
R2/R1
R1/R1
R1/R1
R1/R1
R1/R1
R1/R2
R1/R1
R2/R1
R1/R1
R2/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R2
R2/R1
R2/R1
R1/R1
R1/R1
R2/R2
R2/R1
R1/R1
R2/R1
R1/R1
R2/R1
R1/R1
R1/R1
R1/R1
R1/R1
R1/R2
R1/R2
R2/R1
R1/R1
R2/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R2
R2/R1
R1/R1
R1/R1
R2/R1
R2/R2
R1/R1
R1/R2
R2/R1
R1/R1
R1/R1
R2/R1
R1/R1
R1/R1
R1/R1
R2/R2
R1/R2
R2/R1
R2/R1
R2/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R1
R2/R1
R2/R1
R1/R1
R2/R1
R1/R1
R1/R1
R1/R1
R1/R2
R1/R1
R1/R1
R1/R2
R1/R2
R2/R1
R1/R1
R2/R1
R2/R1
R2/R1
R1/R1
R1/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R1
R2/R2
R2/R1
R1/R1
R2/R1
R1/R1
R2/R1
R1/R1
R1/R1
R2/R1
R1/R1
R1/R1
R1/R1
R2/R1
R2/R1
R2/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R2
R2/R1
R1/R1
R1/R1
R2/R1
R2/R2
R2/R1
R1/R2
R2/R1
R1/R1
R2/R1
R1/R1
R1/R2
R2/R1
R1/R1
R2/R2
R2/R2
R2/R1
R1/R1
R2/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R2
R2/R1
R2/R1
R1/R1
R2/R1
R2/R1
R2/R1
R1/R1
R2/R1
R1/R1
R1/R1
R2/R1
R1/R2
R2/R1
R1/R1
R1/R1
R1/R2
R2/R1
R1/R1
R2/R1
R2/R1
R2/R1
R1/R1
R1/R1
R2/R2
R2/R1
R1/R1
R1/R1
R2/R1
R2/R2
R2/R1
R1/R1
R2/R1
Fig. 3. Effect of the conservation level on the objective function (cost $).
189
5. Conclusion
In this paper, a robust optimization approach has been developed for a new integrated mathematical model to solve a
dynamic cellular manufacturing system (DCMS) integrated with production planning problem considering the parts processing time uncertainty. Considering this kind of realistic uncertainty is necessary to analyze the efciency of manufacturing
system under a real condition. Moreover; many real-world manufacturing factors (e.g., alternative process routings,
inter-cell layout, machine reliability and relocation, machine capacity and operator assignment) have been considered. To
verify the performance of the presented method, two numerical examples have been generated in hypothetical limits and
solved by the CPLEX solver. The experimental results have veried the performance of the presented model and the behavior
of the developed approach in nding an optimal solution. Indeed, the presented mathematical model can be implemented in
small, medium, large-sized problems since the gap of small and medium problems is zero and after that the gap increases
with a reasonable and slow rate. According to sensitivity analysis of the proposed linear model, it can be inferred that the
breakdown rate has a signicant impact on overall system efciency. In fact, in order to obtain an optimal cellular production
environment; the machines should be sustained in a reliable level by an appropriate maintenance process. Investigating the
interactions between maintenance cost and stated objective functions, product/process quality, production rates, considering more options of the uncertainty sources associated with the part demand, product mix, inter-arrival time, available
machine capacity and MTBF to further guarantee the DCMS models and considering the effects of learning and forgetting
on the production system and the position of a potential bottleneck [49] can be interesting as future studies.
Appendix A
A simple problem is presented to justify clearly how the proposed robust optimization works. The deterministic model is
shown as follows:
x2 P 0:
The above model is not practical since all coefcients are considered deterministic ones. While in realistic problems, these
elements are non-deterministic and uncertain.
~ 1 2x
~ 2;
Min OF 3x
s.t.
~ 1 1x
~ 2 6 4;
2x
~ 1 1x
~ 2 6 3;
1x
x1 P 0;
x2 P 0;
~ 2,
~ J 2;
~ 1
~ and J 1;
~ 1
~ are the sets of uncertain elements of objective function, rst and second
where J 0 3;
1
2
constraints, respectively. Also, C0 2 0; jJ 0 j, C1 2 0; jJ 1 j and C2 2 0; jJ 2 j. In other words, C0 2 0; 2, C1 2 0; 2 and C2 2 0; 2.
Assume that C0 1:5, C1 1, C2 0:5 and ranges of uncertain elements are equal to 50% of the nominal elements:
~ 3 1:5;
^ 3 1:5;
^ 2
~ 2 1;
^ 2 1
^
For terms of the objective function: 3
~
^
^
~
^
^
For rst constraint: 2 2 1; 2 1; 1 1 0:5; 1 0:5
~ 1 0:5;
^ 1 0:5;
^ 1
~ 1 0:5;
^ 1 0:5
^
For second constraint: 1
As mentioned in several parts of the paper, the main goal of designing the robust optimization (RO) approach is to make
our model immune to data uncertainty [38]. In this optimization methodology, Immune means that the proposed RO
framework solves an optimization problem considering values that are worse than nominal values to guarantee that real
decisions are better than the conservative obtained decisions of the RO framework (e.g., in the proposed minimizing model,
processing time of part 1 on machine 1 in the period 1 is t 11;1 P 3, while according to Table 15, the nominal value of the mentioned processing time is equal to 3). To model this goal in a minimizing problem, the RO approach considers a maximizing
part for modeling a worse-case:
190
80
1 0
19
0:5
0:5
<
=
z}|{
z}|{
^ 1 j 1:5 1 1jx
^ 2 jA; @1jx
^ 2 j 1:5 1 1:5jx1 jA ;
Min OF 3x1 2x2 max @1:5jx
:
;
s.t.
80
1 0
19
0
0
<
=
z}|{
z}|{
^
^
^
@
A
@
1jx1 j 1 1 0:5jx2 j ; 0:5jx2 j 1 1 1jx1 jA 6 4;
2x1 1x2 max
:
;
80
1 0
19
0:5
< z0:5
=
}|{
z}|{
1x1 1x2 max @0:5 0 0:5jx1 jA; @0:5 0 0:5jx2 jA 6 3;
:
;
x1 P 0;
x2 P 0:
The proposed framework is capable of controlling the robustness of the obtained optimal solution by adjusting the degree of
conservation for each uncertainty set. For instance, by assuming C0 jJ 0 j and Cn jJ n j (in the above example,
C0 C1 C2 2), the worst-case is considered to optimize in the proposed RO approach. The above formulation is the nonlinear robust counterpart of the proposed model. Therefore, linearization techniques are used to obtain a linear model.
Besides, this formulation is a combination of a minmax problem. Hence, the duality theory is applied to the maximizing
part of this framework in order to acquire a minimizing problem. Consequently, by applying duality theory and linearization
methods, the linear counterpart of the proposed model is as follows:
z1 p1;1
z1 p1;2
z2 p2;1
^ ;
P 1y
1
^ ;
P 0:5y
2
^ ;
P 0:5y
1
^ ;
z2 p2;2 P 0:5y
2
x1 P 0;
x2 P 0;
y1 6 x1 6 y1 ;
y2 6 x2 6 y2 ;
z0 P 0; z1 P 0; z2 P 0;
p0;1 P 0; p0;2 P 0; p1;1 P 0;
y1 P 0;
p1;2 P 0;
p2;1 P 0;
p2;2 P 0;
y2 P 0;
References
[1] U. Wemmerlov, N.L. Hyer, Procedures for the part family/machine group identication problem in cellular manufacture, J. Oper. Manage. 6 (1986) 125
147.
[2] R. Kia, A. Baboli, N. Javadian, R. Tavakkoli-Moghaddam, M. Kazemi, J. Khorrami, Solving a group layout design model of a dynamic cellular
manufacturing system with alternative process routings, lot splitting and exible reconguration by simulated annealing, Comput. Oper. Res. 39 (11)
(2012) 26422658.
[3] M. Bagheri, M. Bashiri, A new mathematical model towards the integration of cell formation with operator assignment and inter-cell layout problems
in a dynamic environment, Appl. Math. Model. 38 (2014) 12371254.
[4] C.H. Chu, Cluster analysis in manufacturing cellular formation, OMEGA 17 (3) (1989) 289295.
[5] C.H. Chu, Recent advances in mathematical programming for cell formation, Manuf. Res. Technol. 24 (1995) 346.
[6] N. Singh, Design of cellular manufacturing systems: an invited review, Eur. J. Oper. Res. 69 (3) (1993) 284291.
[7] H.M. Selim, R.G. Askin, A.J. Vakharia, Cell formation in group technology: review, evaluation and directions for future research, Comput. Ind. Eng. 34 (1)
(1998) 320.
[8] G. Papaioannou, J. Wilson, The evolution of cell formation problem methodologies based on recent studies (19972008): review and directions for
future research, Eur. J. Oper. Res. 206 (3) (2010) 509521.
[9] M.J. Rosenblatt, The dynamics of plant layout, Manage. Sci. 32 (1986) 7686.
[10] T.Y. Wang, K.B. Wu, Y.W. Liu, A simulated annealing algorithm for facility layout problems under variable demand in cellular manufacturing systems,
Comput. Ind. 46 (2) (2001) 181188.
[11] M. Solimanpur, P. Vrat, R. Shankar, Ant colony optimization algorithm to the inter-cell layout problem in cellular manufacturing, Eur. J. Oper. Res. 157
(3) (2004) 592606.
[12] R. Tavakkoli-Moghaddam, N. Javadian, B. Javadi, N. Safaei, Design of a facility layout problem in cellular manufacturing systems with stochastic
demands, Appl. Math. Comput. 184 (2) (2007) 721728.
191
[13] M.M. Paydar, I. Mahdavi, I. Sharafuddin, M. Solimanpur, Applying simulated annealing for designing cellular manufacturing systems using MDmTSP,
Comput. Ind. Eng. 59 (4) (2010) 929936.
[14] M.B. Aryanezhad, V. Deljoo, S.M.J. Mirzapour Al-e-hashem, Dynamic cell formation and the worker assignment problem: a new model, Int. J. Adv.
Manuf. Technol. 41 (34) (2009) 329342.
[15] M.M. Ghotboddini, M. Rabbani, H. Rahimian, A comprehensive dynamic cell formation design: Benders decomposition approach, Expert Syst. Appl. 38
(3) (2011) 24782488.
[16] K. Das, R.S. Lashkari, S. Sengupta, Reliability consideration in the design and analysis of cellular manufacturing systems, Int. J. Prod. Econ. 105 (1)
(2007) 243262.
[17] S.H. Chung, T.H. Wu, C.C. Chang, An efcient tabu search algorithm to the cell formation problem with alternative routings and machine reliability
considerations, Comput. Ind. Eng. 60 (1) (2011) 715.
[18] K. Raee, M. Rabbani, H. Raei, A. Rahimi-Vahed, A new approach towards integrated cell formation and inventory lot sizing in an unreliable cellular
manufacturing system, Appl. Math. Model. 35 (2011) 18101819.
[19] F. Defersha, M. Chen, A linear programming embedded genetic algorithm for an integrated cell formation and lot sizing considering product quality,
Eur. J. Oper. Res. 187 (1) (2008) 4669.
[20] V. MajaziDalfard, New mathematical model for problem of dynamic cell formation based on number and average length of intra and intercellular
movements, Appl. Math. Model. 37 (4) (2013) 18841896.
[21] N. Safaei, M. Saidi-Mehrabad, M.S. Jabal-Ameli, A hybrid simulated annealing for solving an extended model of dynamic cellular manufacturing
system, Eur. J. Oper. Res. 185 (2) (2008) 563592.
[22] I. Mahdavi, A. Aalaei, M.M. Paydar, M. Solimanpur, Designing a mathematical model for dynamic cellular manufacturing systems considering
production planning and worker assignment, Comput. Math. Appl. 60 (2010) 10141025.
[23] B. Kim, S. Kim, Extended model for a hybrid production planning approach, Int. J. Prod. Econ. 73 (2) (2001) 165173.
[24] S. Ah kioon, A. Bulgak, T. Bektas, Integrated cellular manufacturing systems design with production planning and dynamic system reconguration, Eur.
J. Oper. Res. 192 (2) (2009) 414428.
[25] N. Safaei, R. Tavakkoli-Moghaddam, Integrated multi-period cell formation and subcontracting production planning in dynamic cellular manufacturing
systems, Int. J. Prod. Econ. 120 (2) (2009) 301314.
[26] N. Sahinidis, Optimization under uncertainty: state-of-the-art and opportunities, Comput. Chem. Eng. 28 (2004) 971983.
[27] S.M.J. Mirzapour Al-e-hashem, H. Malekly, M.B. Aryanezhad, A multi-objective robust optimization model for multi-product multi-site aggregate
production planning in a supply chain under uncertainty, Int. J. Prod. Econ. 134 (2011) 2842.
[28] J. Balakrishnan, C.H. Cheng, Multi-period planning and uncertainty issues in cellular manufacturing: a review and future directions, Eur. J. Oper. Res.
117 (2007) 281309.
[29] D. Szwarc, D. Rajamani, C.R. Bector, Cell formation considering fuzzy demand and machine capacity, Int. J. Adv. Manuf. Technol. 13 (2) (1997) 134147.
[30] M. Rabbani, F. Jolai, N. Manavizadeh, F. Radmehr, B. Javadi, Solving a bi-objective cell formation problem with stochastic production quantities by a
two-phase fuzzy linear programming approach, Int. J. Adv. Manuf. Technol. 58 (2012) 709722.
[31] N. Safaie, M. Saeidi-Mehrabad, R. Tavakkoli-Moghaddam, F. Sasani, A fuzzy programming approach for a cell formation problem with dynamic and
uncertain conditions, Fuzzy Sets Syst. 159 (2) (2008) 215236.
[32] N. Safaie, R. Tavakkoli-Moghaddam, An extended fuzzy parametric programming-based approach for designing cellular manufacturing systems under
uncertainty and dynamic conditions, Int. J. Comput. Integrated Manuf. 22 (6) (2009) 538548.
[33] A. Azadeh, M. Sheikhalishahi, M. Koushan, An integrated fuzzy DEAFuzzy simulation approach for optimization of operator allocation with learning
effects in multi products CMS, Appl. Math. Model. 37 (2013) 99229933.
[34] V.R. Ghezavati, M. Saidi-Mehrabad, An efcient hybrid self-learning method for stochastic cellular manufacturing problem: a queuing-based analysis,
Expert Syst. Appl. 38 (2011) 13261335.
[35] V.R. Ghezavati, M. Saidi-Mehrabad, Designing integrated cellular manufacturing systems with scheduling considering stochastic processing time, Int. J.
Adv. Manuf. Technol. 48 (2010) 701717.
[36] P. Renna, M. Ambrico, Design and reconguration models for dynamic cellular manufacturing to handle market changes, Int. J. Comput. Integrated
Manuf. 28 (2) (2015) 170186.
[37] A.L. Soyster, Convex programming with set-inclusive constraints and applications to inexact linear programming, Oper. Res. 21 (5) (1973) 11541157.
[38] J.M. Mulvey, R.J. Vanderbei, S.A. Zenios, Robust optimization of large-scale systems, Oper. Res. 43 (1995) 264281.
[39] D. Alem, R. Morabito, Production planning in furniture settings via robust optimization, Comput. Oper. Res. 39 (2012) 139150.
[40] A. Ben-Tal, A. Nemirovski, Robust convex optimization, Math. Oper. Res. 23 (1998) 769805.
[41] A. Ben-Tal, A. Nemirovski, Robust solutions of uncertain linear programs, Oper. Res. Lett. 25 (1) (1999) 113.
[42] A. Ben-Tal, A. Nemirovski, Robust solutions of linear programming problems contaminated with uncertain data, Math. Program. 88 (3) (2000) 411
424.
[43] L. El Ghaoui, H. Lebret, Robust solutions to least-squares problems with uncertain data, SIAM J. Matrix Anal. Appl. 18 (4) (1997) 10351064.
[44] D. Bertsimas, M. Sim, The price of robustness, Oper. Res. 52 (1) (2004) 3553.
[45] D. Bertsimas, M. Sim, Robust discrete optimization and network ows, Math. Program. 98 (2003) 4371.
[46] V.R. Ghezavati, S.J. Sajjadi, M. DehghanNayeri, Integrating strategic and tactical decisions to robust designing of cellular manufacturing under
uncertainty: xed suppliers in supply chain, Int. J. Comput. Intell. Syst. 4 (2011) 837854.
[47] M. Soleimani-damaneh, G.R. Jahanshahloo, S. Abbasbandy, Computational and theoretical pitfalls in some current performance measurement
techniques and a new approach, Appl. Math. Comput. 181 (2006) 11991207.
[48] G.J. Doole, Evaluation of an agricultural innovation in the presence of severe parametric uncertainty: an application of robust counterpart optimization,
Comput. Electron. Agric. 84 (2012) 1625.
[49] C.H. Glock, M.Y. Jaber, Learning effects and the phenomenon of moving bottlenecks in a two-stage production system, Appl. Math. Model. 37 (2013)
86178628.