Brazilian Journal of Operations & Production Management 15 (2018), pp 499-516

A HYBRID ANT COLONY SYSTEM AND TABU SEARCH

ALGORITHM FOR THE PRODUCTION PLANNING OF DYNAMIC
CELLULAR MANUFACTURING SYSTEMS WHILE CONFRONTING
UNCERTAIN COSTS

Aidin Delgoshaei ABSTRACT

[email protected]
Kharazmi University, Tehran, Iran
Highlights:
1. Cellular Manufacturing systems cover a wide range of industries.
2. Inflation rate can impose financial harms on cellular manufacturing systems.
Abolfazl Mirzazadeh
[email protected] 3. The over-allocation of workers, which usually happens in dynamic systems, causes re-
Kharazmi University, Tehran, Iran duction of the system performance.
4. The proposed algorithm in this research can successfully schedule cellular systems to
Ahad Ali reduce system costs.
[email protected] Goal:
Lawrence Technological University, The main aim is to determine the best trade-off values between in-house manufactur-
Michigan, United States of America
ing and outsourcing, and track the impact of uncertain costs on gained schedules. To be
more comprehensive, the performance of human resources is restricted and the partial
demands are considered uncertain.
Design / Methodology / Approach:
In this paper a new method for minimizing human resource costs, including operating,
salary, hiring, firing, and outsourcing in a dynamic cellular manufacturing system is pre-
sented where all system costs are uncertain during manufacturing periods and can be
affected by inflation rate. For this purpose, a multi-period scheduling model that is flexible
enough to use in real industries has been proposed. To solve the proposed model, a hybrid
Ant Colony Optimization and the Tabu Search algorithm (ACTS) are proposed and the out-
comes are compared with a Branch-and-Bound based algorithm.
Results:
Our findings showed that the inflation rate has significant effect on multi-period system
planning. Moreover, utilizing system capability by the operator, for promoting and using
temporary workers, can effectively reduce system costs. It is also found that workers’ per-
formance has significant effect on total system costs.
Limitations of the investigation:
This research covers the cellular manufacturing systems.
Practical implications:
The algorithm is applied for 17 series of dataset that are found in the literature. The pro-
posed algorithm can be easily applied in real industries.
Originality / Value:
The authors confirm that the current research and its results are original and have not
been published before. The proposed algorithm is useful to schedule cellular manufactur-
ing systems and analyse various production conditions.
Keywords: Human Resource Scheduling; Ant Colony Optimization; Skilled Worker Assign-
ing; Out-Sourcing; Uncertain Costs.

ABEPRO
DOI: 10.14488/BJOPM.2018.v15.n4.a4
500
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

1. INTRODUCTION As a different point of view, R. Askin and Huang (2001)

studied the performance of heuristics as the greedy algo-
Designing Appropriate Facilities is vital for scheduling rithm and the simulated annealing algorithm for creating
manufacturing systems engineering. Tompkins et al. (2003) team works and promoting teams, using cross-training
reported that over $250 billion is estimated to be spent for plan in CMS. Slomp and Suresh (2005) focused on assig-
facility designing, planning, and re-planning in the U.S ev- ining workers to team works with the aim of minimizing
ery year. Material Transferring takes 20% to 50% of the total the training and assigning costs while maximizing labor
cost of manufacturing systems. An appropriate scheduling flexibility. In the same year, Fitzpatrick and Askin (2005)
can reduce the mentioned costs by 10% to 30%. The Cel- argued that elements of a good team formation are not
lular manufacturing system (CMS) is a trustful way of using only limited to personal skills and characteristics but
group technology. In a cellular system, the layout is designed also technological and human interactions. Hence, using
in a way as to use the advantages of manufacturing flexi- pre-determined skill level measures, they tried to select
bility and efficient flow by grouping machines according to workers and assign them to appropriate teams in cells to
the families of products (Papaioannou and Wilson, 2010). maximize team performance. At the same time, Cesaní
Human resource management (HRM) is considered an im- and Steudel (2005) focused on work sharing, work balanc-
portant issue in CMS problems through the last 3 decades. ing and leveling the operator assignments in the presence
During the year 2013, U.S. spent $7.0 billion for training and of bottleneck operations. For continuing to prevent over
employment programs. Hence, during last decades, scien- loading and over assigning, Satoglu and Suresh (2009)
tists made their best efforts to define and solve HRM prob- used goal programming method in a mathematical model,
lems in different circumstances, conditions and situations to where the objectives were minimizing over assignment of
find out new ways to reduce such expenses. workers, cross training, hiring, and firing costs.

Perhaps, determining the optimal number of workers in Cross-trained workers are referred to those workers
different cells is the main idea of HRM problems in CMS. that trained to perform more than one task. By determin-
In 1994, Morris and Tersine simulated some cell layouts by ing the best set of cross-training, workers can improve
considering equipment and labor (Morris and Tersine, 1994). system performance. In 1996, Bartholdi and Eisenstein
Next year, to determine the optimal number of operators declared that, considering larger cells that are equipped
and assigning them to parts, Park and Lee (1995) developed with multi-skilled workers and various workstations caus-
a 2-stage model while in the first stage, a Taguchi method es a more stable work place, thus yielding emerging bal-
was used to determine a system performance that was then anced production lines and maximizing production vol-
used as the objective function of the assigning model. The ume as well (Bartholdi & Eisenstein, 1996). Afterward,
idea of maximizing saving costs between operating and out- Kleiner et al. (1998) took charge of skilled workers, in a
sourcing was investigated by Heady (1997). But their model computer-based system. The proposed system that was
did not investigate operator level, training, hiring and firing employed in an air craft component manufacturing com-
costs. Afterward, Norman et al. (2002) focused on assigning pany in U.S included cell lead time, part travel distance,
workers in CMS when the aim was maximizing system profit process yield, operator classification, and labor efficiency.
and then Ertay and Ruan (2005) developed the idea of de- In continue, it was proved that cross-trained workers can
termining the number of operators to optimize the number achieve higher performance than normal workers (Gel
of output products. For this purpose, using weighted input et al., 2000) and top management role and cross-trained
data, a data envelopment analysis (DEA) was applied. But employees have significant impact on the successful im-
they failed to consider skill levels of the operators and ma- plementation of CMS (Olorunniwo and Udo, 2002). In the
chines. year 2000, Kher found that by using cross-trained attribu-
tions, such as learning and re-learning, more effective
The idea of considering operator levels was developed by training schemes can be provided (Kher, 2000). The idea
Suer and Cedeño (1996). For this purpose, a mixed integer of distributing skilled workers within teams and the de-
programming method was used to generate alternative op- gree of the workforce belongs to Molleman and Slomp
erator levels and then an integer programming method was (1999), where they indicated that such items have import-
used to find the optimal operator assignments in cells. The ant effects on the performance of CMS. They showed that
idea of worker assigning and training problem in function- using appropriate plans to promote and distribute skilled
al and cellular layouts was then followed by R. G. Askin & workers yields remarkable results on the performance of
Huang (1997). In addition, Aryanezhad et al. (2009) consid- the system and, as a result, each of the workers will be
ered three skill levels for workers, which can be promoted an expert in more than one skill. Later, four cross-training
through the planning horizon by training, in a multi period policies, which were considered based on workload of the
scheduling model, for simultaneous cell forming and worker bottleneck workers in certain and uncertain work condi-
assignning. tions, were compared by Slomp and Molleman (2002).
 501
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

The results confirmed that better team performance can 2. THE PROPOSED MODEL
be expected by using higher levels of cross-training work-
ers. Then, the staffing level and shop layouts in the de- In this section a framework is used to examine the per-
partment, strictly and hybrid cell layouts were taken into formance of the model. The framework has steps to provide
consideration (Jensen, 2000), by changing the number of production plans by using skilled workers, hiring and firing
employees in each department and considering three lev- temporary workers (as needed) and outsourcing respec-
els of workload balance and two labor transferring rules. tively is developed (Figure 1). In many real cases managers
Jensen evaluated product flow and job tardiness. Recent- may prefer in-house manufacturing than outsourcing since
ly, Li et al. (2012) focused on minimizing average salary outsourcing may be more costly or may cause some quali-
while maximizing average of satisfaction. For this purpose ty problems or on-time producing. Despite, in some cases,
they developed a multi-objective mixed integer program- others may desire to use outsource services due to use sup-
ming method to determine the number of cross-trained pliers that provide cheaper parts in a reasonable quality or
labors and also tasks that must be assigned to the labors lack of technology. In this study the main reason for using
in flexible assembly cell layout. Another contribution of outsource services is the lack of enough system capacity for
their research was considering worker satisfaction and fulfilling demands in a specific period. For this purpose, the
task redundancy levels. proposed framework is prepared to balance the in-house ca-
pacity periodically based on customer demands promoting
Recently, the idea of considering dynamic part de- workers and/or hiring temporary workers before using sub-
mands in HRM-CMS, which can cause system imbalance, contractor’s services (as needed). Hence, the steps of the
is more popular, as it was less developed before. To solve method are designed as:
this problem, Mahdavi et al. (2010) developed a multi-
mode planning method aiming at workers’ assignments 1) Generating initial layout (using initial sets of workers
in a reconfigurable CMS. In the proposed model, hiring, to cells).
firing, and also salary costs were considered as a part of
total system costs. They also considered the available 2) Promoting workers by improving skills of the fixed
time for workers. Afterward, they focused on inter-cellu- operators using training budget.
lar movements of workers and parts while processing on
a specific machine (Mahdavi et al., 2012). The contribu- 3) Hiring new temporary workers for fulfilling the re-
tion of this study was adding workers as the 3rd dimension mained part demands.
of machine-part incidence matrix by using a cubic matrix.
4) Firing extra temporary workers to reduce system
Delgoshaei et al. (2016b) reviewed material transfer- costs (if exists).
ring models and approaches and illustrated the drawbacks
emerged while transferring materials in CMS. In the same 5) Using outsource services if any part demands are
year, Delgoshaei et al. (2016a) proposed a new method still remains.
for reducing the cell load variation in dynamic CMS. Del-
goshaei et al. (2017) used a hybrid genetic algorithm and After generating initial layouts by using forward serial
simulated annealing for scheduling CMS. programming, the main strategy is to satisfy customer needs
by promoting fixed operators (current workers) before hiring
An in-depth survey in the history of HRM problems in new ones. If promoting available workers in any planning pe-
CMS (Table 1) shows the trade-off issue between in-house riod can cover the customer needs, there will be no need to
manufacturing, considering the skilled workers, the hir- hire temporary workers or to use outsource services which
ing and firing of temporary workers, and outsourcing, is mostly expensive. Otherwise, with respect to cell sizes,
while part of the demands are considered uncertain and the algorithm focuses on hiring new workers that will be
human performance is not fixed and is less developed. considered as temporary workers. Note that cell size can be
Moreover, system costs are considered fixed while in the controlled by lower and upper cell boundaries through the
real world, system costs shall not be considered equal in planning horizon. Such workers are considered temporary
different time intervals, as they may be affected by many since they are temporarily hired to fulfill the customer de-
factors, such as inflation rate (Delgoshaei et al., 2014). In mands in a planning slot. Therefore they are not eligible for
this research, a new mixed integer mathematical model is further training unless the algorithm decided to keep them
addressed by considering uncertain costs to find the best for another planning period. In continue, if the maximum ca-
combination of worker allocation and outsourcing in the pability of in-door manufacturing is impotent to satisfy the
presence of the mentioned condition. customer needs, using the services of collaborating firms or
subcontractors will be suggested. The explained logic will
be kept in the solving process of all metaheuristics that will
502
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Table 1. Comparing the opted researches from the literature

Prog. Method/
No. Reference DY/EX Contribution/Significant Point

O1
O2
O3
O4
O5
O6
O7
type Approach
Aryanezhad et al. Considering worker’s skill level
1 DY  NL-MIP  SDCWP * *
(2009) and machine level
2 Askin and Huang (1997) EX IP   - * * Using aggregate planning
Considering psychological, organi-
3 Askin and Huang (2001) EX GP * *
zational and technical factors
Bartholdi and Einsen- Converting non-linear dynamical
4 DY NL-MIP  Simulation * *
stein (1996) systems to an exact model
Cesaní and Steudel Dealing with the problem of labor
5 EX LP Framework * *
(2005) flexibility in CMS
Determining optimal number of
6 Ertay and Ruan (2005) EX NL-IP DEA *
operators and labor assignment
Fitzpatrick and Askin
7 EX MIP  Heuristic * * * Forming effective worker teams
(2005)
Minimizing number of outsourced
8 Heady (1997) EX IP LINDO *
parts
Focused on advantages of ma-
9 Jensen (2000) EX  -  SLAM * *
chine and labor flexibility
Considering learning and forget-
10 Kher (2000) EX  -   - *
ting issues
Evaluating cross-training policies
11 Li et al. (2012) EX NL-MIP Genetic * *
during cell forming process
Focused on available time of
12 Mahdavi et al. (2010) DY IP * * workers/ dynamic system recon-
figuration
Considering multi-task workers
13 Mahdavi et al. (2012) EX NL-MIP BandB * as 3rd dimension of machine part
incidence matrix
Molleman and Slomp Determining number of workers
14 EX LP  2heuristics  * *
(1999) that must learn some specific skill
Maximizing organization effective-
15 Norman et al. (2002) EX MIP CPLEX * ness/considering technical and
human skills
Olorunniwo and Udo Considering Sociotechnical
16 EX  -  Field Study *
(2002) variables
Using dynamic programming/con-
17 Park and Lee (1995) EX - Taguchi *
sidering 2 levels of workers
Minimizing cross-training, hiring,
Satoglu and Suresh
18 EX GP GMAS/SA * * * * firing and over-assignment of
(2009)
workers
Slomp and Molleman Considering 4 training policies/
19 EX LP M-ANOVA * *
(2000) evaluating team performance
Slomp and Suresh Assigning operators in work
20 EX BP LINGO * * *
(2005) teams
Considering operator level (work
Suer and Cedeño
21 EX -  Clustering * * station level) in similarity coeffi-
(1996)
cient
D Y / P r o g . LP Linear Programming DP Dynamic Programming
Dynamic/Exact Programming type
EX type
IP Integer Programming NL-MIP Non-linear Mixed Integer Programming GP Goal Programming O7 Outsourcing
O1 Hiring and Firing O3 Salary O5 Labor Efficiency

O2 Worker Satisfaction O4 Worker Assignment O6 Training/Skill level

 503
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Start
Hiring New
workers
No No
Ini�ate a layout Solu�on Check Training workers Solu�on Check
(Randomly)
Firing temporary
workers
Yes
Yes
No
Finish Outsourcing Solu�on Check

Figure 1. Flowchart of the proposed framework

Ma terial Pr 1
Di s tributor
Pr 2

Sub-contractor
Pr 4
C1 P1 T2 D1 P3 W1

Pa i n�ng
Ca bi n 1

1
Pr 6
Corridor

C2 P2 W1 D1 P4 W1

C3 T1 W2 Corridor

Sub-contractor
Pr 3

1
Corridor
P5
Stora ge 2

Stora ge 3

Stora ge 4
Stora ge 1

Pr 5 SH1
Pa i n�ng

Pa i n�ng

Sub-contractor
Ca bi n 2

Ca bi n 3

S1
SH2
1

W5

Pr: Part Route of Products

Figure 2. A Flow Diagram of Work Assigning and Outsourcing in a Cellular Manufacturing System

be deigned in the next part. Figure 2 shows a typical layout 1) Maximum number of cells is fixed through the plan-
of CMS where part demands are manufactured using the ning horizon.
system capacity and the rest of those parts that cannot be
completely manufactured inside the system are planned to 2) Maximum and minimum numbers of workers that
allocate to subcontractors. can be allocated to each cell are controlled using up-
per and lower boundaries.
The following items show the assumptions that are taken
into consideration in order to provide production plan alter- 3) The performance of fixed operators and temporary
natives by means of system capability and also outsource workers is not constant and will be affected by in-
services: creasing the volume of production.
504
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

4) Inflation rate is not constant and, hence, all system

costs will be changed periodically through planning
horizon.

5) Part demands are not fixed and supposed to change

periodically. Part demands will be calculated using
normal function distribution.

6) Promoting Workers is allowed through planning

horizon.

7) Training which is done to promote workers takes

no time and the training cost for each skill level is
known in advance.

8) The skill level and skill-production rate coefficients

are fixed and known in advance.

9) In firing condition, swapping fixed operators is pro-

hibited unless there was no temporary worker.

10) Each location can be filled by only one worker. : Inflation rate during period that will be calcu-
lated using randomly increasing inflation rate estimator:

2.1. Inputs (2)

2.3. Matrixes

Number of operators that are available at the beginning

of planning period ( )

Operator-Part incidence: Ability of the operator type S in

manufacturing part i (

Operator-Part ability incidence: Ability of manufacturing

part i while the skill level of the operator is n

2.2. Parameters
2.4. Variables

(1)
 505
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

(15)

(16)

(17)

(18)

(19)

The first sentence in the objective function is developed

to show the operating cost including machinery and wage of
: The performance coefficient of worker type processing of each part. The second sentence explains the
training costs that will be applicable only for fixed operators.
during period The third and fourth sentences show hiring and firing tem-
porary workers respectively and the final sentence of objec-
tive function is used for determining outsource values. The
inflation rate affects the different types of system costs in
(3)
the model. This will allow the decision makers to analyze the
role of uncertain costs in increasing system entropy.
.
The first series of constraints are used for assuring that
each of the products will be produced not less than its mar-
2.5. Mathematical Model ket demand (whether inside the company or using outsource
services) through planning slots. The second constraint
The proposing model can be developed now as a non-lin- allows managers to pre-determine the proportion of pro-
ear mixed integer programming method: ducing products in the system or using outsource services
based on the company policies. The third constraint controls
promoting worker’s skill according to the training budget in
(4)
each planning slot. The fourth constraint ensures that the
number of workers (whether temporary or fixed) will change
(5)
logically in each period. The fifth constraint sets the initial
number of each worker types in the first period. The sixth
constraint is used to guarantee that the number of workers
in each planning slot will not exceed or be less than the up-
(6)
per and lower bounds, respectively. The seventh constraint
ensures that none of the workers will be assigned more than
s.t: (7) his/her capability. The next three sets of constraints are de-
signed to set initial worker’s skills and to make sure that the
(8) skill levels will not increase infinitely. The rest of the con-
straints is used to control the domain of the variables.

(9)
2.6. Solution representing

In both solving methods, which will be explained next,

(10) the scheme of representing the solutions of the model is il-
lustrated by developing a solution-string that contains four
matrixes where the 1st matrix shows positions of operators
(11) in cell layout, the second matrix indicates the skill level of
(12) each operator, the third matrix reveals the amount of as-
signed parts to each operator and the last matrix shows the
(13) amount of each part types that decided to be processed by
subcontractors (Figure 3).
(14)
506
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Solution-position Solution-skill Solution-x outsourcing

Figure 3. A representing scheme sample

The degree of hardness of the proposed model is 3.1.1. Choosing number of neighbors
. Hence, for example, for a small size problem (sup-
pose 5 cells with the capacity of 8 operators in each cell, The proposed algorithm uses the replacing strategy for
4 operator types and 3 skill levels), the number of possible generating new feasible solutions (Figure 4). It is noted that
solutions (both feasible and non-feasible) can be calculated various strategies are available for searching neighbors of
as: a solution that must be chosen based on the nature and
circumstances of each problem. Replacing elements, im-
(20) proving an element or zero-one shifting are among popular
As mentioned by Fitzpatrick and Askin (2005) even small methods. In this case, such elements are defined by replac-
size problems in such cases are hard to solve using normal ing workers with different work skills.
optimizing algorithms. Hence, in next part, a typical branch
and bound algorithm (BandB) is developed and then pro- Suppose
pose an ACS that is strengthened using the positive features
of tabu search algorithm is proposed. Then, the results of is
the BandB and ACS will be compared. an intial solution.

Then
3. SOLVING ALGORITHMS

3.1. Branch and Bound Algorithm is a new neighbor that can be generated by shifting the
element in the initial solution.
Branch and bound algorithm is generally employed to
find optimal solutions in optimization problems where all Then is a new neighbor that can be generated by shifting
new solutions are compared with a lower or upper (or both) the element in the initial solution.
bounds. BandB can be designed as a single or multi start-point
algorithm and the number of neighbors for each iteration may
be varied (based on solution string), which helps to search 3.1.2. Objective function operator
solution spaces more comprehensively, but at the same time
BandB is not applicable for continuous variable models in its The objective function of the proposed mathematical
classic form since it was designed for solving discrete variable model is considered the main operator for evaluating the
problems. In addition, one big drawback with BandB is the new neighbors:
lack of a mechanism for escaping from local optima. Hence,
BandB is more sensitive for falling into local optima. More-
over, although BandB is a fast tracking and reliable algorithm
for small and medium scale problems, for large scale and
complicated problems, BandB is more risky to encounter with (21)
the “early stage convergence” phenomenon.
In the next step, using the following formula, the best
P1 P2 P3 P4 P5 P6 P7 P8 P1 P2 P3 P4 P5 P6 P7 P8
W1 0 0 10 0 0 0 5 0 W1 0 0 5 0 0 0 5 0
W2 0 3 0 5 0 0 10 0 W2 0 3 0 10 0 0 10 0
W3 20 0 0 0 40 0 0 5 W3 20 0 0 0 40 0 0 5
W4 10 8 0 10 0 5 5 0 W4 10 8 0 10 0 5 5 0
W5 0 5 10 5 0 0 7 0 W5 0 5 10 5 0 0 7 0
Figure 4. Choosing a new feasible neighbor in B&B algorithm
 507
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

neighbor for each solution (say ‘r’) will be chosen: will occur as a consequence of achieving optimum solution
or encountering with a local optimum. Choosing improper
(22) number of neighbors ( ) is another reason that may cause
the algorithm to be impotent.

Table 2. Pseudo code for the BandB algorithm 1ST 2nd k-1th kth
itera�on itera�on itera�on itera�on No more
Improvement
Initialize parameters K,G is possible

• Initialize m, k and i (m: number of initial solutions; k=

number of neighbours for each solution; i: number of Replacing the …
iterations) worst solu�on
with the best
solu�on in
Do previous period The Best
Observed
Solu�on
Loop until (p<i)

• Generate m feasible solutions as Ini�al Good Worst Best Solu�on in

Solu�on Solu�onss Solu�on
n itera�on

• For each solution

Figure 5. Solving Process Scheme of B&B algorithm, considering
o Loop until (q<m) one neighbor for each solution

 Generate a neighbour
3.2. The Hybrid Ant Colony Optimization and Tabu
 Check Validity test for the solution Search (ACTS)

 Run function operator Ant colony optimization is inspired from the swarm intel-
ligence of real ants that live in big colonies (hundred thou-
 Calculate OFV sand of ants). As an optimizing method, it was firstly used by
Dorigo (1992) and a few years later it was recognized among
• If well reputed metaheuristics. The main aim of the classic ver-
sion of ACO, which was designed to solve discrete optimi-
zation problems, was to find a smaller path in a graph, but
other versions were promoted to solve continuous problems
with various objectives. Figure 6 shows how ACO finds the
• path with minimum distance (as an objective function).

• Check Stopping Criteria

3.1.3. Terminating criteria

1) If maximum number of iteration is reached.

2) If there is no further opportunity for improvement:

(23)

The equation above means that the searching process

will be terminated if none of the generated solutions in the
current iteration is better than pervious stages so that no im-
provement will be expected (Figure 5). Note that such status Figure 6. Solving Process Scheme of Ant Colony
508
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

3.2.1. Initial Colony Size and Pheromone Update

25

Choosing an appropriate number of colony members

plays a key role in providing better solutions in Ant colony is a uniformed random number between 0 and 1. The
systems. But at the same time, choosing big size of colony above formula chooses a value among normalized pher-
members causes more computations and more computation omones that are sorted in an ascending way. It is obvious
time accordingly. Each colony member is a representative of that members with higher values have more chance to be
a solution string that has the opportunity of development in chosen.
next iteration. While a new neighbor is generated, the effi-
ciency of the new path will be evaluated by calculating the 3) Using short-term and long-term memories as what
value of the sprinkled pheromone using the formula below: was used in the Tabu search algorithm: Due to high
capability of Tabu search for finding solutions in Np-
hard problems, which comes as a consequence of
using memories (short-term and long-term), TS has
been widely used as part of a hybrid with other al-
gorithms. Roux et al. (1999) developed a hybrid Ant
System (AS) and TS called ANTabu to solve optimiza-
The above formula uses the amount of objective function tion problems. A few years later, Kaji (2001) also em-
value (total system costs) as the main operator for calcu- ployed the hybrid of AS and TS for solving travelling
lating the amount of pheromone sprinkle of each colony salesman problem.
member. Therefore, the more objective function value sav-
ing occurs in a path, the more pheromone will be sprinkled. In this study the mechanism of short-term and long-term
Otherwise, the best amount among pheromones in the last memories of TS was used as a part of searching process in
iteration will be considered for the colony member that pro- ACO.The logic of TS is based on using short-term memory
vide a chance for worse solution to stay in an optimizing pro- to prohibit revisiting those solutions that had been rejected
cess for the next iterations and does not throw them away before and also those whom are banned by the algorithm
instantly. Such strategy is suitable for escaping from local for some reason (long-term memory).
optimum.
Ini�al Next Last
Itera�on Itera�on Itera�on
3.2.2. Evaporating Pheromones Short term
memory
(forbidden
ACO has a powerful improving engine that enables it for movement)
Best
providing good solutions, but at the same time, such im- Solu�on
proving speed increases the speed of solution convergence. ...
To prevent it, three different strategies can be used:

1) Choosing larger sets of colony members that in-

crease computation time as well.

2) Using a list of solutions with higher pheromones: Long term memory

considering only one member (with the highest (forbidden area)
pheromone level) in iteration as a candidate for the Figure 7. Solving Process Scheme of Tabu Search
rest of optimizing process causes immediate elimi-
nation of the other colony members. Therefore, the In 1986, Glover developed TS to overcome defects of
algorithm should not consider the best pheromone searching neighborhood spaces (Glover, 1986). Tabu search
as the only candidate for the next iteration, but in- is an effective way of local searching technique that im-
stead a probability function will be used to create a proves the chances of finding optimum or near optimum
tournament list of better pheromones in a way that solutions by defining such memory structures in a way that,
colony members with higher pheromone have more if a potential solution has been rejected in a certain period
chance to be chosen for the rest of the optimization before or if it violates a rule, it will be marked as a “tabu”
process. Such strategy also prevents early stage con- (forbidden) movement so that the algorithm will not consid-
vergence and helps to escape local optimum traps: er it for a certain period in next iterations (Figure 7).
 509
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

3.2.3. Intensification
(28)
During the process of searching local neighbors, the algo-
rithm may encounter a lot of elite neighbours. In such con- Proposition 2: Using inspiration rate helps to fade old
dition, it is better to have a mechanism for increasing the members of the tabu list. Hence, if an area is banned in an it-
search for the number of neighbors for a short time. Using eration, it will not last until the end of the searching process,
such strategy helps providing a good structure for concen- but instead, after a period (even if the Tabu list is not updat-
trating more comprehensively on elite members in local ed), the older members will be faded one by one. Through
areas. An appropriate short-term memory, provides an ef- the use of such strategy the areas that have been forbidden
fective way for searching elites in local areas as it prevents in the early stages of the solving process will have a chance
searching worse elements (Figure 7). to be involved in later iteratations again, thus helping to pre-
vent early stage solution convergence.

3.2.4. Diversification Proof:

Diversification is another common problem that emerges

while using TS algorithm. As mentioned before, TS algorithm a) Suppose
works based on functions that prohibit searching the areas
in a solution space that has been searched before with no is the oldest member of long-term memory
sign of improvements. To prevent such problem, two strate-
gies are considered: ( ).

• Proposition 1: The first strategy is using capaci- b) If

ty restricted long-term memory with the ability of
updating during iterations. Through the use of such (29)
strategy a new candidate will be replaced with the
oldest member of long-term memory whenever the
long-term memory is full. c) If

• Proof: Imagine an initial allocation of workers that

are eligible to develop by training, hiring (or firing) (30)
and outsourcing in a layout. If using an initial distri-
bution of workers for providing the next neighbors
causes worse objective function, then this cell layout 3.2.5. Short-term Memory
will be considered a member of long-term memory:
• Proposition: In this study short-term memory is con-
sidered during training procedure in a way that, if
a) Suppose promoting an operator (in a specific planning slot)
is the oldest member of long-term memory ( causes less production volume than outsourcing,
). this level of training is banned temporarily for that
operator until the following planning slot.
b) If (suppose ω is the number
of cells that are already filled up in the Tabu list) and Proof:

a) Suppose operator is considered as a candidate for

are new
promoting the skill level in a specific period
candidates for long-term memory, then:
b) If
(26)

and
; then:
(27)
510
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

< settings, it is important to use appropriate parameters, oth-

erwise the quality of results may not be good enough (Del-
goshaei et al., 2014). The number of iterations for BandB and
; (31)
ACTS is set depending on the size of the problem (150, 200,
Then . But there is no logical and 250 for small, medium and large size problems, respec-
reason to prohibit the worker until the end of the optimizing tively). The same strategy is used for determining the num-
process. Therefore, the worker will be considered as a ber of neighborhood sizes, where 150, 200, and 250 neigh-
member of the short-term memory in the mentioned period bors are considered for BandB and ACTS. Since the chance of
( ) encountering with local optima are varied from each case to
another, three different evaporation rates are supposed for
ACTS (0o.1, 0.3, and 0.5) and similar rates are considered as
3.2.6. Long-term Memory local escaping rate. For ACTS, the Tabu list size is supposed
to change based on the scale of problems (2, 3, and 5).
• Proposition: As mentioned before, the framework
starts by generating initial ants (cell layouts) ran- In order to minimize the effects of using estimated pa-
domly, which are then promoted using training, hir- rameters (such as number of colony members, iterations,
ing and using outsource services. Suppose that, after and tabu list size) for the proposed algorithm, a similar way
preparing a solution, if the objective function of an that was used by Li et al. (2012), where they solve all exper-
ant is worse than all other colony members in previ- imented ten times, was followed.
ous iterations, it means that this cell layout is not a
good candidate for improvement. In such condition,
there is no need to focus on the mentioned area. 4. EXPERIMENTS AND RESULTS
Hence, the initial layout that was generated during
the first step of the framework is forbidden for the To examine and evaluate the performance of the pro-
coming iteration. posed ACTS algorithm for the proposed method, a num-
ber of experiments from the literature is solved (Table 3).
Proof: All examples were solved by BandB and ACTS using Matlab®
R2009 software, which is installed on a Core™ i7 personal
a) laptop that was supported by 2.40 GH CPU and 8 GB RAM.
is an initial cell layout,

4.1. Discussion

Results showed that in a certain number of iteration, in

47.06% of the solved cases, ACTS provided better solutions
then: by escaping from local optima and in 29.01% both algo-
rithms reported the same results (Table 4). The main rea-
son for such improvement is the use of short- and long-term
memories that enable ACTS to provide better solutions in
the same number of iterations (up to 2.24%).

Although in most of the cases, the average of the sched-

(32)
ules that was achieved by using ACTS causes smaller system
b) if costs in a specific number of program repetition, at the same
; time, ACTS consumes more computations and needs more
(33) time accordingly. Besides, comparing the variance between
the results of both methods shows that the value of im-
then places in provements is varied for different problems and depends on
the structure of problems; hence, it cannot be predicted in
. advance. The results also show that, while large scale prob-
lems are taken into account, ACTS provides better solutions
most of the times (Figure 6). Such results reveal the impor-
3.3. Parameter Setup tance of using an effective structure for escaping from lo-
cal optima in ACTS. Although the increase in the number of
Since metaheuristics are sensitive to initial parameter solutions in iterations and in the number of neighborhoods
 511
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Table 3. Table of Input Data for Experiments that were gathered from the literature*

No. Problem Source K W O S T C.S NOP d(i,t) T.R PR(i) SR S.C

Askin and Huang [2000 1500 1700 0.8 12
1 2 2 2 2 3 8 [2 4] d~N [15 15] [1 2] 2
(2001) 1800]
Askin and Huang [2000 1500 1700 0.8 12
2 2 2 2 4 4 4 [2 4] d~N [15 15] [1 2 3 4] 2
(2001) 1800]
Suer and Cedeno [2000 1500 1700 0.8 26
3 1 4 4 3 4 12 [3 2 3 2] d~N [30 40 20 23] [1 2 3] 4
(1996) 1800]
Small

Mahdavi et al. 0.8 22

4 2 4 4 3 2 12 [4 5 3 6] d~N [2000 1500] [20 40 30 40] [1 2 3] 4
(2010)
Mahdavi et al. [2000 1500 1700 0.8 22
5 2 4 4 3 4 16 [4 3 6 5] d~N [20 40 30 40] [1 2 3] 4
(2012) 1800]
Aryanezhad et al. 0.8 16
6 3 3 3 5 3 9 [5 4 5] d~N [2000 1500 1700] [10 15 15] [1 2 3 4 5] 3
(2009)
[2000 1500 1700 0.8 25
7 Li et al. (2012) 2 5 5 5 4 15 [3 5 4 2 3] d~N [20 25 30 20 15] [1 2 3 4 5] 5
1800]
Mahdavi et al. [2000 1500 1700 0.7 5
8 2 5 5 3 4 15 [7 8 5 9 4] d~N [10 10 8 10 6] [1 2 3] 5
(2012) 1800]
Mahdavi et al. 0.7 12
9 2 6 6 4 3 18 [5 4 3 6 5 4] d~N [2000 1500 1700] [10 8 14 11 14 7] [1 2 3 4] 6
(2010)
Norman et al. [2000 1500 1700 0.7 14
10 2 6 6 4 5 14 [3 5 2 4 3 4] d~N [12 14 21 14 24 16] [1 2 3 4] 6
Medium

(2002) 1800 1400]

Askin and Huang [2 4 3 5 4 3 [2000 1500 1700 [14 21 15 16 8 12 [1 2 3 4 5 6 0.7 15
11 2 8 8 8 4 20 d~N 8
(2001) 2 4] 1800] 14 22] 7 8]
Askin and Huang [2 4 3 5 4 3 [2000 1500 1700 [23 31 15 21 13 14 0.7 12
12 2 8 8 4 4 20 d~N [1 2 3 4] 8
(2001) 2 4] 1800] 14 16]
Askin and Huang [2000 1500 1700 0.7 12
13 8 2 2 2 4 4 [14 6] d~N [5 3] [1 2] 2
(2001) 1800]
Askin and Huang [2000 1500 1700 0.7 12
14 8 2 2 4 4 4 [14 6] d~N [5 3] [1 2 3 4] 2
(2001) 1800]
Aryanezhad et al. [2000 1500 1700 0.7 12
15 4 5 5 6 4 12 [2 4 3 4 4] d~N [25 14 14 23 14] [1 2 3 4 5 6] 5
(2009) 1800]
Aryanezhad et al. [2000 1500 1700 [1 2 3 4 5 0.7 12
Large

16 5 4 4 7 4 8 [4 3 4 3 3] d~N [16 21 28 21 18] 4

(2009) 1800] 6 7]
Aryanezhad et al. [1 2 3 4 5 6 0.7 12
17 5 5 5 8 3 10 [4 3 4 3 3] d~N [2000 1500 1700] [16 21 28 21 18] 5
(2009) 7 8]
Number of Number of Time
K: Cell Number W: O: S: Skill Level T: C.S: Cell Size
Workers Operations Slots
Initial Number Part De- Training Production Skill Number of
NOP: d(i,t): T.R: PR(i): SR: SC:
of Operators mands Budget rate rate sub-contractors
* While gathering data in each case, the authors tried to use applicable data. It is noted that other required parameters
are generated by the members of this research.

can strengthen the possibility of escaping from the local op- factors that can affect system performance in cellular manu-
tima, ACTS can provide still better schedules in the specif- facturing systems. To evaluate the impact of inflation rate on
ic number of iterations. In addition, it was found that the system costs, all problems are solved, again considering two
speed of convergence in BandB is faster than ACTS. assumptions. In the first case, all system costs are consid-
ered fixed during planning horizon, while in the other status,
It is figured out that the proposed framework can effec- system costs are uncertain and are supposed to be changed
tively improve the objective function by reducing system periodically (Table 5). It is found that increasing inflation rate
costs. Such improvement is a result of finding the best values causes increasing system costs and will change the balance
of appointing skilled operators, temporary workers (hiring of schedules (in terms of indoor manufacturing and using
and firing) and subcontractor’s services. Results clear that outsource services).
inflation rate and operator performances are two important
512
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Table 4. Results of solving the experiments gathered from the literature

Problem Best Vari-

No. Itr. N.S BandB C.M itr. L.E ACTS OFV max min D IMP%
Source Value ance
Askin and Huang
1 150 150 28850 150 150 0.3 28839 28839 28844.50 7.78 28850 28839 11 0.04
(2001)
Askin and Huang
2 150 150 44666 150 150 0.3 44666 44666 44666.00 0.00 44666 44666 0 0.00
(2001)
Suer and Cedeno
3 150 150 16258 150 150 0.3 16416 16258 16337.00 111.72 16416 16258 158 0.96
(1996)
Mahdavi et al.
4 150 150 10641 150 150 0.3 10654 10641 10647.50 9.19 10654 10641 13 0.12
(2010)
Mahdavi et al.
5 150 150 27797 150 150 0.3 27738 27738 27767.50 41.72 27797 27738 59 0.21
(2012)
Aryanezhad et
6 200 200 10113 200 200 0.3 9886.2 9886.2 9999.60 160.37 10113 9886.2 226.8 2.24
al. (2009)
7 Li et al. (2012) 200 200 56311 200 200 0.3 56311 56311 56311.00 0.00 56311 56311 0 0.00
Mahdavi et al.
8 200 200 25964 200 200 0.3 25972 25964 25968.00 5.66 25972 25964 8 0.03
(2012)
Mahdavi et al.
9 200 200 25211 200 200 0.3 25406 25211 25308.50 137.89 25406 25211 195 0.77
(2010)
Norman et al.
10 200 200 49457 200 200 0.3 49457 49457 49457.00 0.00 49457 49457 0 0.00
(2002)
Askin and Huang
11 200 200 42769 200 200 0.3 42394 42394 42581.50 265.17 42769 42394 375 0.88
(2001)
Askin and Huang
12 200 200 80853 200 200 0.3 80808 80808 80830.50 31.82 80853 80808 45 0.06
(2001)
Askin and Huang
13 200 200 89663 200 200 0.3 89663 89663 89663.00 0.00 89663 89663 0 0.00
(2001)
Askin and Huang
14 200 200 17739 200 200 0.3 17736 17736 17737.50 2.12 17739 17736 3 0.02
(2001)
Aryanezhad et
15 250 250 39980 250 250 0.3 39980 39980 39980.00 0.00 39980 39980 0 0.00
al. (2009)
Aryanezhad et
16 250 250 37130 250 250 0.3 37116 37116 37123.00 9.90 37130 37116 14 0.04
al. (2009)
Aryanezhad et
17 250 250 27099 250 250 0.3 27073 27073 27086.00 18.38 27099 27073 26 0.10
al. (2009)
Improvements
Itr. Iteration N.S Neighbourhood Size IMP%
Percentage

Moreover, while using subcontractor services is more in-house manufacturing, which cause system to be more de-
expensive or whenever the total the effective capacity of a pendent on outsource services in the period of the planning
company is not used, the framework tends to use in-house horizon (see the right column of the table 6). In contrast,
manufacturing by improving fixed operator’s skills or hiring in the second case, the out sourcing costs are considered
new temporary workers. Although such scenario leads re- too much. As a result, the algorithm decided to produce
ducing system costs, the operators may become over-allo- more in-house parts that cause operators over-allocating. It
cated. In contrast, if using the outsource services is cheaper is also found that the possibility of increasing system costs
than in-door manufacturing, the system may become too as a cause of mismatching between worker’s promoting and
dependent on the use of outdoor services. Table 6 compares outsourcing is increased in the dynamic condition of market.
the results of solving a problem, where all the settings and
conditions are equal except for the outsourcing price. In the Table 6. The system is too dependent on outsource services
first cast, the outsourcing price is too low and close to the (Problem 1) / Operators are over-allocated (Problem 2)
 513
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Table 5. Comparing Results of ACTS, Considering Fixed and Uncertain Costs

CASE 1 (Fixed Costs) CASE 2 (Uncertain Costs)

N.S/ N.S/
No. Problem Source Inf. Rate Itr. L.E TSC.* Itr. L.E TSC. D
C.M C.M
1 Askin and Huang (2001) [1.01 1.02 1.05] 150 150 0.3 28613 150 150 0.3 28839 226
2 Askin and Huang (2001) [1.01 1.02 1.05 1.14] 150 150 0.3 42790 150 150 0.3 44666 1876
3 Suer and Cedeno (1996) [1.01 1.02 1.05 1.14] 150 150 0.3 15304 150 150 0.3 16258 954
4 Mahdavi et al. (2010) [1.01 1.02 1.05] 150 150 0.3 10515 150 150 0.3 10641 126
5 Mahdavi et al. (2012) [1.01 1.02 1.05 1.14] 150 150 0.1 26336 150 150 0.1 27738 1402
6 Aryanezhad et al. (2009) [1.01 1.02 1.05] 200 200 0.3 9391 200 200 0.3 9886.2 495.2
7 Li et al. (2012) [1.01 1.02 1.05 1.14] 200 200 0.3 53193 200 200 0.3 56311 3118
8 Mahdavi et al. (2012) [1.01 1.02 1.05 1.14] 200 200 0.1 24648 200 200 0.1 25964 1316
9 Mahdavi et al. (2010) [1.01 1.02 1.05] 200 200 0.3 24955 200 200 0.3 25211 256
10 Norman et al. (2002) [1.01 1.02 1.05 1.14 1.07] 200 200 0.3 46587 200 200 0.3 49457 2870
11 Askin and Huang (2001) [1.01 1.02 1.05 1.14] 200 200 0.3 40388 200 200 0.3 42394 2006
12 Askin and Huang (2001) [1.01 1.02 1.05 1.14] 200 200 0.3 76931 200 200 0.3 80808 3877
13 Askin and Huang (2001) [1.01 1.02 1.05 1.14] 200 200 0.3 85370 200 200 0.3 89663 4293
14 Askin and Huang (2001) [1.01 1.02 1.05 1.14] 200 200 0.3 16943 200 200 0.3 17736 793
15 Aryanezhad et al. (2009) [1.01 1.02 1.05 1.14] 250 250 0.3 37912 250 250 0.3 39980 2068
16 Aryanezhad et al. (2009) [1.01 1.02 1.05 1.14] 250 250 0.3 35017 250 250 0.3 37116 2099
17 Aryanezhad et al. (2009) [1.01 1.02 1.05 1.14] 250 250 0.3 26598 250 250 0.3 27073 475
TSC: Total System Cost

Solu- Out- [12 12 29

Solu- [2 2 1 [5 5 5 12 23 29
Problem tion-posi- Solution-x sourc-
tion-skill 2 3 1 5 5 5 12 23 12
tion ing
2 3 2 5 5 5 12 23 23
[3 3 0 [4 1 0 [18 5 0 2 3 3 5 5 5 3 6 29
1 0 2 1 0 5 1 0 12 3 1 1 1 1 5 9 12 12]
2 1 1 5 1 5 12 8 29 3 2 2] 2 5 5]
3 3 3 3 5 5 12 23 23 [ 12 12
3 1 1 5 5 3 23 29 15 2) Askin and [2 2 1 [5 5 5 29 12 23 [0 3
2 1 2] 5 2 5] 12 11 12] Huang (2001) 2 3 1 5 5 5 29 0
2/2/2/2/4/8/ 2 3 2 5 5 5 12 23 0 8
1) Askin and [3 3 0 [4 1 0 [18 2 0 [0 51 [2 4] 2 3 3 5 5 5 12 12 23 0
Huang (2001) 1 0 2 1 0 5 1 0 12 0 OP=[3 5 2]; 3 1 1 1 1 5 23 0 9
2/2/2/2/4/8/ 2 1 1 5 1 5 12 8 29 0 56 OS=[40 20 30]; 3 2 2] 3 5 5] 3 6 29 0]
[2 4] 3 3 3 4 5 5 18 23 23 0 12 12 12]
OP=[3 5 2]; 3 1 1 5 5 3 23 29 15 0 57 [2 2 1 5 5 5
OS=[4 2 3]; 2 1 2] 5 2 5] 12 11 12] 3] 2 3 1 5 5 5 [12 12 29
2 3 2 5 5 5 12 23 29
[3 3 0 [1 1 0 [7 7 0 2 3 3 5 5 5 12 23 12
1 0 2 1 0 5 1 0 12 1 1 1 1 1 5 12 23 23
2 1 1 5 2 5 12 11 29 3 2 2] 3 5 5] 8 8 29
3 3 3 4 5 5 18 23 23 12 12 12]
3 1 1 5 5 4 23 29 22
2 1 2] 5 2 5] 12 11 12]
As mentioned before, the performance rate of opera-
tors is another factor that can influence the scheduling of
a system and system costs, accordingly. Figure 8 indicates
different costs of a system where all other settings are con-
sidered the same; however, different performance rates are
taken into consideration. The positive slope of the trend line
shows that, while the performance rate of workers is de-
514
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

creased, the total system cost increases. Such phenomenon mension and therefore the developed model is an NP-hard
happens as a result of decreasing the production volume of problem. To overcome such shortcoming, a hybrid ACO
the system that causes an increase in the amount of out- and TS algorithms is developed. It is observed that infla-
sourcing accordingly. tion rate has negative effects on the scheduling of cellular
manufacturing systems. Over-allocating workers and creat-
ing reliant systems (on subcontractors) are two events that
16000 emerge as a consequence of the increasing inflation rate.
15800 15832 Over-allocating workers that usually happens in dynamic
systems causes reduction of the system performance that
15600
can significantly affect production schedules and increase
15400 15415
the total system costs. It is also found that, while partial
15279
System Cost

15200 demands are uncertain, finding the best quota of manu-

15000 14930 15045 facturing production and using outsource services is vital
and can significantly improve the system productivity by
14800 14722
reducing system costs. In this manner, hiring or firing tem-
14600 porary workers (as needed) and promoting skilled workers
14400 play a key role. Future research in this area is suggested
14200 by considering different payment methods for both skilled
14000 workers and subcontractors.
1 0.95 0.9 0.85 0.8 0.75
Alpha rate Acknowledgements: The authors would like to sincerely
appreciate the editorial team and also anonymous review-
Figure 8 Trend line shows increase system cost, while decreasing ers for their positive and constructive comments which are
performance rate used to improve the manuscript.

4.2. Performance Metric for Evaluating Results 6. REFERENCES

Aryanezhad, M.; Deljoo, V.; Mirzapour Al-e-hashem, S.
Finding suitable metrics that can provide a base for com-
(2009), “Dynamic cell formation and the worker assignment
paring results is essential. For this manner, a formula is de-
problem: a new model”, The International Journal of Advan-
veloped to check the amount of reducing total system cost,
ced Manufacturing Technology. Vol. 41. No. 3-4, pp. 329-342.
comparing with the average of results that was achieved by
both algorithms. Askin, R.; Huang, Y. (2001), “Forming effective worker
teams for cellular manufacturing”, International Journal of
(34) Production Research. Vol. 39. No. 11, pp. 2431-2451.
Improvements%=
Askin, R. G.; Huang, Y. (1997), “Employee training and as-
signment for facility reconfiguration”, Proceedings of the 1997
6th Annual Industrial Engineering Research Conference, IERC,
The formula above shows how better ACTS can reduce
Miami, FL. 426-431.
total system costs compared to the other algorithm. Table
4 shows that the performance of ACTS in most of the cases Bartholdi, J. J.; Eisenstein, D. D. (1996), “A production line
is better than BandB (up to 2.24%), which means that com- that balances itself”, Operations Research. Vol. 44. No. 1, pp.
paring with BandB, ACTS can improve the system costs more 21-34.
effectively. Cesaní, V. I.; Steudel, H. J. (2005), “A study of labor assign-
ment flexibility in cellular manufacturing systems”, Computers
& industrial engineering. Vol. 48. No. 3, pp. 571-591.
5. CONCLUSION
Delgoshaei, A.; Ariffin, M.; Baharudin, B.; Leman, Z. (2016a),
In this research, a new programming method is proposed “A new method for decreasing cell-load variation in dynamic
to assign skilled operators and temporary workers and use cellular manufacturing systems”, International Journal of In-
subcontractors’ services in cellular manufacturing systems. dustrial Engineering Computations. Vol. 7. No. 1, pp. 83-110.
Then the impact of uncertain costs and operator’s perfor- Delgoshaei, A.; Ariffin, M. K.; Baharudin, B. H. T. B.; Le-
mance on dynamic cellular manufacturing systems is eval- man, Z. (2014), “A Backward Approach for Maximizing Net
uated. The developed problem (as it was found for many Present Value of Multi-mode Pre-emptive Resource-Cons-
similar problems in the literature) is sensitive to model di- trained Project Scheduling Problem with Discounted Cash
 515
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Flows Using Simulated Annealing Algorithm”, International cellular manufacturing”, Computers & industrial enginee-
Journal of Industrial Engineering and Management. Vol. 5. ring. Vol. 34. No. 1, pp. 207-217.
No. 3, pp. 151-158.
Li, Q.; Gong, J.; Fung, R. Y.; Tang, J. (2012), “Multi-ob-
Delgoshaei, A.; Ariffin, M. K. A.; Ali, A. (2017), “A multi- jective optimal cross-training configuration models for an
-period scheduling method for trading-off between skilled- assembly cell using non-dominated sorting genetic algo-
-workers allocation and outsource service usage in dyna- rithm-II”, International Journal of Computer Integrated
mic CMS”, International Journal of Production Research. Manufacturing. Vol. 25. No. 11, pp. 981-995.
Vol. 55. No. 4, pp. 997-1039.
Mahdavi, I.; Aalaei, A.; Paydar, M. M.; Solimanpur, M.
Delgoshaei, A.; Ariffin, M. K. A. M.; Leman, Z.; Baharu- (2010), “Designing a mathematical model for dynamic cel-
din, B. H. T. B.; Gomes, C. (2016b), “Review of evolution lular manufacturing systems considering production plan-
of cellular manufacturing system’s approaches: Material ning and worker assignment”, Computers & Mathematics
transferring models”, International Journal of Precision En- with Applications. Vol. 60. No. 4, pp. 1014-1025.
gineering and Manufacturing. Vol. 17. No. 1, pp. 131-149.
Mahdavi, I.; Aalaei, A.; Paydar, M. M.; Solimanpur, M.
Dorigo, M. (1992), Optimization, learning and natural al- (2012), “A new mathematical model for integrating all in-
gorithms, PhD Thesis, Politecnico di Milano, Italy. cidence matrices in multi-dimensional cellular manufactu-
ring system”, Journal of Manufacturing Systems. Vol. 31.
Ertay, T.; Ruan, D. (2005), “Data envelopment analysis
No. 2, pp. 214-223.
based decision model for optimal operator allocation in
CMS”, European Journal of Operational Research. Vol. 164. Molleman, E.; Slomp, J. (1999), “Functional flexibility
No. 3, pp. 800-810. and team performance”, International Journal of Produc-
tion Research. Vol. 37. No. 8, pp. 1837-1858.
Fitzpatrick, E. L.; Askin, R. G. (2005), “Forming effecti-
ve worker teams with multi-functional skill requirements”, Morris, J. S.; Tersine, R. J. (1994), “A simulation compari-
Computers & industrial engineering. Vol. 48. No. 3, pp. son of process and cellular layouts in a dual resource cons-
593-608. trained environment”, Computers & industrial engineering.
Vol. 26. No. 4, pp. 733-741.
Gel, E.; Hopp, W.; Van Oyen, M. (2000), Workforce agility
in systems with hierarchical cross-training: working paper, Norman, B. A.; Tharmmaphornphilas, W.; Needy, K. L.;
Northwestern University Evanston, IL. Bidanda, B.; Warner, R. C. (2002), “Worker assignment in
cellular manufacturing considering technical and human
Glover, F. (1986), “Future paths for integer programming
skills”, International Journal of Production Research. Vol.
and links to artificial intelligence”, Computers & Operations
40. No. 6, pp. 1479-1492.
Research. Vol. 13. No. 5, pp. 533-549.
Olorunniwo, F.; Udo, G. (2002), “The impact of manage-
Heady, R. B. (1997), “Forming minimum-cost machine
ment and employees on cellular manufacturing implemen-
cells with exceptional parts using zero-one integer pro-
tation”, International Journal of Production Economics. Vol.
gramming”, Journal of Manufacturing Systems. Vol. 16. No.
76. No. 1, pp. 27-38.
2, pp. 79-90.
Papaioannou, G.; Wilson, J. M. (2010), “The evolution
Jensen, J. B. (2000), “The impact of resource flexibility
of cell formation problem methodologies based on recent
and staffing decisions on cellular and departmental shop
studies (1997–2008): Review and directions for future re-
performance”, European Journal of Operational Research.
search”, European Journal of Operational Research. Vol.
Vol. 127. No. 2, pp. 279-296.
206. No. 3, pp. 509-521.
Kaji, T. (2001), “Approach by ant tabu agents for Trave-
Park, T.; Lee, H. (1995), “Design of a manufacturing cell
ling Salesman Problem”, Systems, Man, and Cybernetics,
in consideration of multiple objective performance measu-
2001 IEEE International Conference on. 3429-3434.
res”, Manufacturing research and technology. Vol. 24. No.,
Kher, H. V. (2000), “Examination of flexibility acquisition pp. 181-202.
policies in dual resource constrained job shops with simul-
Roux, O.; Fonlupt, C.; Talbi, E.-G.; Robilliard, D. (1999),
taneous worker learning and forgetting effects”, Journal of
ANTabu--enhanced version.
the Operational Research Society. Vol. No., pp. 592-601.
Satoglu, S. I.; Suresh, N. C. (2009), “A goal-programming
Kleiner, B. M.; Drury, C. G.; Palepu, P. (1998), “A compu-
approach for design of hybrid cellular manufacturing sys-
ter-based productivity and quality management system for
tems in dual resource constrained environments”, Compu-
ters & industrial engineering. Vol. 56. No. 2, pp. 560-575.
516
Brazilian Journal of Operations & Production Management
Volume 15, Número 4, 2018, pp. 499-516
DOI: 10.14488/BJOPM.2018.v15.n4.a4

Slomp, J.; Molleman, E. (2002), “Cross-training policies Suer, G. A.; Cedeño, A. A. (1996), “A configuration-based
and team performance”, International Journal of Produc- clustering algorithm for family formation”, Computers & in-
tion Research. Vol. 40. No. 5, pp. 1193-1219. dustrial engineering. Vol. 31. No. 1, pp. 147-150.
Slomp, J.; Suresh, N. C. (2005), “The shift team forma- Tompkins, J.; White, J.; Bozer, Y.; Tanchoco, J. (2003), Fa-
tion problem in multi-shift manufacturing operations”, Eu- cilities planning, Willey, New York.
ropean Journal of Operational Research. Vol. 165. No. 3,
pp. 708-728.

Received: 27 May 2018

Approved: 10 Jun 2018
DOI: 10.14488/BJOPM.2018.v15.n4.a4
How to cite: Delgoshaei, A.; Mirzazadeh, A.; Ali A. (2018), “A Hybrid Ant Colony System and Tabu Search Algorithm
for the Production Planning of Dynamic Cellular Manufacturing Systems While Confronting Uncertain Costs”,
Brazilian Journal of Operations & Production Management, Vol. 15, No. 4, pp. 499-516, available from: https://
bjopm.emnuvens.com.br/bjopm/article/view/501 (access year month day).