DOKUZ EYLÜL UNIVERSITY

GRADUATE SCHOOL OF NATURAL AND APPLIED

SCIENCES

DESIGN OF CELLULAR MANUFACTURING

SYSTEM WITH WORKER ASSIGNMENT

by

Müge AKPINAR

March, 2013

İZMİR
 DESIGN OF CELLULAR MANUFACTURING
SYSTEM WITH WORKER ASSIGNMENT

A Thesis submitted to the

Graduate School of Natural and Applied Sciences of Dokuz Eylül University

In Partial Fulfillment of the Requirements for the Degree of Master of

Science in Department Of Industrial Engineering

by

Müge AKPINAR

March, 2013

İZMİR
 ACKNOWLEDGEMENTS

I am appreciated to my family for their support on me at whole my life, but

especially on this study.

And also I feel myself much more lucky by having a guide as Assist. Prof. Dr.
Özcan Kılınçcı.

And at the end, I would like to thank to TÜBİTAK for their monetary support.

Müge AKPINAR

iii
 DESIGN OF CELLULAR MANUFACTURING SYSTEM WITH WORKER
ASSIGNMENT

ABSTRACT

Cellular manufacturing is one of the most effective approaches in manufacturing

area, when the system contains similar products by shapes, functions or production
methods. Cell Formation Problem is the main problem in cellular manufacturing
systems.

In this thesis, a specific cell formation problem, which consists of alternative

routes of parts, different process times at different machines for each part,
manufacturing cost, different inter-cell material handling cost for different parts,
different machine complexity degree, different worker talent degree, multifunctional
workers and worker training cost. To solve the problem, an algorithm based on
Simulated Annealing algorithm is presented. The number of the cells is determined
in the presented algorithm using Kaiser’s Rule by Bashir & Karaa (2008). To test the
proposed algorithm, three cell formation problems are randomly generated, the small
sized consists of five machines and seven parts; the medium sized consists of ten
machines and fifteen parts; and the large sized consists of eighteen machines and
thirty parts. The results show that the proposed algorithm produces good results
giving the minimum total cost which includes manufacturing cost, inter-cell material
handling cost, and worker training cost. Also effects of the inter-cell material
handling cost and the worker training cost on the total cost are analyzed by changing
each cost.

Keywords: Cell formation, cellular manufacturing, heuristics, human issue, worker

training, worker assignment, simulated annealing.

iv
 ÇALIŞAN ATAMALI HÜCRESEL İMALAT SİSTEMİ TASARIMI

ÖZ

Hücresel imalat sistemleri, imalat alanında çok etkili bir yöntemdir, özellikle
şekil, fonksiyon veya üretim yöntemi açısından benzer parçalar söz konusu olduğu
zaman. Hücresel imalat sisteminin temel problemi Hücre Oluşturma Problemi’dir.

Bu tezde, alternatif rota, makinalarda parça bazlı farklı imalat süreleri, üretim
maliyeti, hücreler arası parça bazlı taşıma maliyeti, makine bazlı zorluk derecesi,
operatör bazlı yetenek derecesi, çok fonksiyonlu operatörler, ve çalışan eğitim
maliyeti bilgilerini içeren nitelikleri belli bir hücre oluşturma problemi çalışılmıştır.
Problemi çözmek için Tavlama Benzetimi algoritmasına dayanan bir algoritma
sunulmuştur. Hücre sayısı belirlemek için sunulan algoritmanın içinde Bashir ve
Karaa (2008) tarafından bulunan Kaiser Kuralı kullanılmıştır. Oluşturulan
algoritmayı denemek için rastgele üç hücre oluşturma problemi geliştirilmiştir, küçük
boyda olan 5 makine ve 7 parçadan, orta boyda olan 10 makine ve 15 parçadan ve
büyük boyda olan 18 makine ve 30 parçadan oluşmuştur. Sonuçlar, sunulan
algoritmanın, üretim maliyeti, hücreler arası taşıma ve çalışan eğitim maliyetinden
oluşan minimum toplam sistem maliyetini veren iyi sonuçlar ortaya çıkarmıştır.
Ayrıca eğitim maliyeti ve hücreler arası taşıma maliyetinin toplam maliyet
üzerindeki etkisi, birim maliyetler değiştirilerek analiz edilmiştir.

Anahtar kelimeler: Hücre oluşturma, hücresel imalat, sezgiseller, insan faktörü,

çalışan eğitimi, çalışan ataması, tavlama benzetimi.

v
CONTENTS

Page

M.Sc THESIS EXAMINATION RESULT ................................................................. ii

ACKNOWLEDGEMENTS ........................................................................................ iii

ABSTRACT ................................................................................................................ iv

ÖZ ................................................................................................................................ v

LIST OF FIGURES .................................................................................................... ix

LIST OF TABLES ....................................................................................................... x

CHAPTER ONE - INTRODUCTION ..................................................................... 1

1.1 Cellular Manufacturing System ......................................................................... 1

1.2 Framework of the Thesis.................................................................................... 3

1.3 Outline of the Thesis .......................................................................................... 4

CHAPTER TWO – CELLULAR MANUFACTURING SYSTEM AND CELL

FORMATION PROBLEM ....................................................................................... 6

2.1 Introduction ....................................................................................................... 6

2.2 The Cost Components Of Cellular Manufacturing Systems ............................. 7

2.3 Worker Assignment In Cellular Manufacturing Systems ............................... 11

2.4 The Literature Review On Cellular Manufacturing Systems .......................... 14

2.4.1. Cellular Manufacturing Systems ............................................................ 14

2.4.2. Human Factor In Manufacturing ............................................................ 17

vi
CHAPTER THREE – PROBLEM DEFINITON AND SOLUTION
METHODOLOGY................................................................................................... 25

3.1 The Proposed Algorithm And Components .................................................... 29

3.1.1. The Kaiser’s Rule ................................................................................... 29

3.1.2. The Simulated Annealing Algorithm...................................................... 31

3.1.3. The Proposed Algorithm ........................................................................ 34

3.1.4. An Illustrative Example .......................................................................... 36

CHAPTER FOUR – ANALYZING THE COMPUTATIONAL RESULTS ..... 39

4.1. Computational Results For 5x7 Sized Part-Machine Matrix ......................... 41

4.1.1 Analysis of Inter-material Handling Cost ................................................ 46

4.1.2 Analysis of Training Cost ........................................................................ 48

4.2. Computational Results For 10x15 Sized Part-Machine Matrix ...................... 51

4.2.1 Analysis of Inter-material Handling Cost ................................................ 56

4.2.2 Analysis of Training Cost ........................................................................ 60

4.3. Computational Results For 18x30 Sized Part-Machine Matrix ..................... 63

4.3.1 Analysis of Inter-material Handling Cost ................................................ 68

4.3.2 Analysis of Training Cost ........................................................................ 72

CHAPTER FIVE – CONCLUSION ...................................................................... 77

REFERENCES ......................................................................................................... 79

APPENDIX ............................................................................................................... 85

vii
A1. Cms Matlab Codes ........................................................................................ 85

viii
LIST OF FIGURES

Page

Figure 1.1 A detailed classification of cell formation methods .............................. 3

Figure 2.1 Example of machine worker matrix ..................................................... 23

ix
LIST OF TABLES

Page

Table 2.1 Similarity Coefficient Formulas for General Purpose........................... 10

Table 2.2 Literature survey on Heuristics ............................................................. 16

Table 2.3 Factors affecting the development and deployment of labor flexibility
............................................................................................................................... 22

Table 2.4 Literature Survey ................................................................................... 24

Table 4.1 Characteristics of the problems ............................................................. 40

Table 4.2 Computational results of 5x7 part-machine matrix ............................... 44

Table 4.3 Components of the 1st route.................................................................. 44

Table 4.4 Components of the 2nd route ................................................................ 45

Table 4.5 Computational results of 5x7 part-machine matrix ............................... 46

Table 4.6 Computational results of 5x7 part-machine matrix ............................... 47

Table 4.7 Computational results of 5x7 part-machine matrix ............................... 49

Table 4.8 Computational results of 5x7 part-machine matrix ............................... 50

Table 4.9 Part-machine-cell matrix ....................................................................... 51

Table 4.10 Computational results of 10x15 part-machine matrix ......................... 54

Table 4.11 Components of the 1st route................................................................ 54

Table 4.12 Components of the 3rd route ............................................................... 55

Table 4.13 Computational results of 10x15 part-machine matrix ......................... 57

Table 4.14 Computational results of 10x15 part-machine matrix ......................... 58

Table 4.15 Computational results of 10x15 part-machine matrix ......................... 60

Table 4.16 Computational results of 10x15 part-machine matrix ......................... 61

Table 4.17 Computational results of 10x15 part-machine matrix ......................... 62

Table 4.18 Part-machine-cell matrix ..................................................................... 63

x
Table 4.19 Computational results of 18x30 part-machine matrix ......................... 66

Table 4.20 Components of the 6th route ............................................................... 67

Table 4.21 Computational results of 18x30 part-machine matrix ......................... 69

Table 4.22 Computational results of 18x30 part-machine matrix ......................... 70

Table 4.23 Computational results of 18x30 part-machine matrix ......................... 72

Table 4.24 Computational results of 18x30 part-machine matrix ......................... 73

Table 4.25 Computational results of 18x30 part-machine matrix ......................... 75

Table 4.26 Part-machine-cell matrix ..................................................................... 75

xi
 CHAPTER ONE
INTRODUCTION

1.1 Cellular Manufacturing System

A manufacturing system consists of labor, machine and raw materials. When they
are brought together to produce a product in a same place, the system gets ready to
work. Many types of manufacturing systems exist in the real world. For example;
assembly lines, flexible manufacturing systems, project-based (make to order)
manufacturing systems, batch type manufacturing systems, continuous (make to
stock) manufacturing systems, and cellular manufacturing systems. Cellular
manufacturing system is studied in this thesis.

Cellular manufacturing systems (CMS) are not included in traditional systems

which have been used for the several years. CMS is based on similarity of parts and
functions of machines. This system works by bringing some machines together with
some parts in a common area. Some rules or assumptions are used while this
integration is carried out. First, the similarity between parts should be determined.
Similar parts form “families”, when they get together. This similarity is generally
determined by their relationship with machines in the production system.

Then machines, which are intensively related with families, are assigned to these
families. At the end of this assignment, the group, formed by interrelated parts and
machines, is called as “cell”.

A rule should be discussed for cell formation problems. The Kaiser’s Rule is a
good guide to form cells. This rule uses within-group correlations and calculates the
optimal number of cells in a production system with several parts and machines.
After that, cell compositions and families can be determined easier.

As Cellular Manufacturing system provides grouping similar parts into part

families and the corresponding machines into cells, it constructs small imaginary

1
manufacturing plants which are responsible for only themselves in the whole
manufacturing system. Thus, gaining ascendancy over these small plants gets easier
and a better control approach is provided on manufacturing area in this way. This
new system provides reduced paper work, reduced labor, better supervisory control,
reduced tooling, reduced setup time, reduced delivery time, reduced lead time,
reduced rework and scrap materials, reduced lot size, reduced work in process,
reduced inventory, reduced material handling, easier scheduling and improved
quality efficiency and flexibility in manufacturing system.

After configuring a cellular manufacturing system, there are many issues that
should be taken into consideration. For example; capacity and quantity of machines,
routes, types and quantities of parts, abilities of workers, part carrying costs, etc.
Many methods are used to solve cell formation problems in the literature. They have
both advantages and disadvantages. These approaches could be mainly classified into
three groups,

1)Part oriented approaches

2)Process oriented approaches
3)Visual inspection method

Visual inspection method generally does not work effectively. The separation of
parts depends on the visual ability of worker. That means it relies on personal
experience and carefulness.

The part-oriented approaches use the shapes or functionalities of parts to form

families and groups by some classification and coding methods. However the
configuration of cell can not completely be done by these techniques.

Process oriented approaches work by manufacturing similarities as the similarities

of parts’ routes. The advantage is, these approaches only use machines which are
required for the part.

2
 A detailed classification for these approaches is shown in figure 1.

Figure 1.1 A detailed classification of cell formation methods

1.2 Framework Of The Thesis

Various studies on cellular manufacturing systems and their main points have
been evaluated and discussed for this thesis. It is understood that the implementing
cellular manufacturing system provides many advantages in manufacturing areas.
First, the possibility of applying cellular manufacturing system to current system
should be analysed. After that, an effective method should be found to form cells and
to implement cellular manufacturing system. The point is, whole characteristics of
the existing manufacturing system should be included by implemented system. Many
approaches are defined in searched studies. Some of them have several assumptions,
and some of them are suitable just for more stable systems. However the issue,
related with human or workers, is always ignored in studies. Because this topic is
hard to study with its generally behavioural based structure. Human issue has a big
importance on manufacturing systems. Because human is the main component of the
system.

3
 It can be seen that few studies exist on literature for human issue, which are not
appropriate to simulate real systems. Because of that, it is decided to study cell
formation problem with human issue in this thesis. Technical skills and trainability is
used for workers. This means workers can be multi-functional in this study. Because
a worker should be suitable for a cell which we assigned to. In real life situations,
training is usually used and assignments are done by workers’ skills. In this study, it
is tried to be as close as it gets to a real system. Some assumptions are also used.

A part may have alternative routes in a real life manufacturing system. This
provides flexibility in the system. Because of that, alternative routes are included in
this study. It is also called as “alternative machines for a part”, in the study.

Operations are carried out for multi-period in a real life manufacturing system.
But multi-period calculations make system more complicated and hard to solve.
Because of that, a cellular manufacturing system for single period is modelled.

Simulated Annealing(SA) heuristic is used to solve the derived problems. It can

be seen from literature survey that SA heuristic is easier and faster than other
heuristics to solve a model.

A couple of numerical examples are derived for a 5x7, 10x15 and 18x30
(machine-part) dimensioned problem.

In problems, this study tried to form cells which include workers with specified
number of machines and parts, by minimizing the objective cost function using
Matlab R2008a.

1.3 Outline Of The Thesis

This thesis consists of chapters. Chapter one gives some basic informations about
Cellular Manufacturing Systems and applications of solving cell formation problems.
Also framework of the thesis is explained in this part.

4
 The components of the problem are explained and a detailed literature review is
done for Cellular Manufacturing in Chapter two. Everything about Cellular
Manufacturing (benefits, applications, hardnesses, deficiencies, studied and
unstudied issues in literature, studies, etc.) is explained in this chapter.

The components of the problem and algorithm are explained in Chapter three.
Heuristics have an important role on solving problems in an acceptable time interval
and with an acceptable performance. So we need to learn and use them to have better
and applicable results. Also the characteristics of the derived problem are explained
in this chapter. Proposed Algorithm which is constructed for the derived problem is
shown in this chapter.

In Chapter four, computational results of the problem are explained with all
points. The problem is explained by numerical examples, and results.

In Chapter five, the conclusion is explained and interpreted.

5
 CHAPTER TWO
CELLULAR MANUFACTURING SYSTEM AND CELL FORMATION
PROBLEM

2.1 Introduction

Cellular manufacturing system (CMS) is an implementation of Group Technology

(GT) to the manufacturing area. Group technology is a manufacturing philosophy
that has established the potential to contribute positively in batch-type production,
and it endeavours to combine the flexibility of the job production system with the
high productivity of the flow production system (Ham, Hitomi, & Yoshida, 1985).
Mosier & Taube (1985), and Shunk (1985) define GT as;

"...a disciplined approach to identify things such as parts, processes, equipment,

tools, people or customer needs by their attributes, analyzing those attributes looking
for similarities between and among the things; grouping the things into families
according to similarities; and finally increasing the efficiency and effectiveness of
managing the things by taking advantage of the similarities. "

The most common two incorporation methods for determining part families and
cells are called classification and coding. Some rules are used while determining part
families and cells. It was explained in chapter one, that at the beginning of cell
formation the similarity between parts should be determined. Coding is used in here.
Cell determination, which is the main point of CMS, is carried out after that
similarity determination. Then some other usual problems may occur as in traditional
systems. The layout of machines and cells should be decided. Actually, it is as
important as forming cells. Cells should be well emplaced to avoid unnecessary
traffic in manufacturing area. An inefficient designed layout would cause worse
results.

The aim in CMS is minimizing the overall cost, while maximizing the
effectiveness in manufacturing area. This cost is called Objective Function in
models. Models are used to solve real systems. For example; mathematical models,

6
heuristic algorithms, simulation models etc. A model of the real system is
constructed before the execution. Trials will cause a big cost in real manufacturing
systems. We can’t change places of machines and routes of parts for several times in
a manufacturing area to find the right decision. Because of that, models of real
systems are used. They help us to find the optimal arrangement without any changes
in the manufacturing area. At the end, we can apply the optimal result to the real
manufacturing system at once.

The components of Objective Function are explained in the next section.

2.2 The Cost Components Of Cellular Manufacturing Systems

The objective function consists of costs which occur in the manufacturing system.
These costs are; manufacturing cost, inter-cellular material handling cost, setup cost,
hiring-firing costs and training cost.

The manufacturing cost denotes the value of cost that occurs by machines which
are visited by parts. A matrix, which shows the relation between parts and machines,
is needed for a manufacturing cost calculation. This matrix is called part-machine
incidence matrix, and formed by 1’s and 0’s. 1(one) means, that part visits that
machine. 0(zero) means that part has no operation by that machine. If the sequence
of operations is important in a system, 1’s and 0’s turn to order numbers of
operations. Sequence may change cost calculation, family formation and lead time of
the system. The reason is, difference in the similarity calculation and assignment of
parts to machines. The existence of sequence for parts may change the objective
value, and the cell configuration. Another issue that changes the cost calculation is
importance level of parts. In some manufacturing systems, an importance level is
assigned to products. This level is shown by weight parameters. A part may have
higher weight parameter according to production volume or operation time of that
part (Yin & Yasuda, 2006). Furthermore importance level may be used for some
other issues. These issues show characteristics of the manufacturing system. This
also changes the objective value because of different cell formations.

7
 Another issue, that changes the manufacturing cost, is alternative routes. A part
may have more than one route to accomplish its process. This method is usually used
in real manufacturing systems. Because this gives flexibility to the system and
decreases the lead time.

The second part of objective function is inter-cellular material handling cost. This
is the cost caused by the traffic between cells. If a part needs an operation out of a
cell, then it should generate a move out from this cell. This move is called inter-
cellular material handling. Inter-cellular movements are caused by exceptional
elements or operations which are carried out by some machines in some other cells,
basically. However this does not mean that intra-cell movements are not important
for cost control. The layout problem, for both intra-cell and inter-cell systems, is
another problem of implementing Cellular Manufacturing systems. When we have a
better cell configuration, we will have lower inter-cellular material handling cost.
The companies which deal with unreliable customer demands, and want to survive,
have to briskly adapt themselves to changes and organise production system in
accordance with these changes.

The third part of objective function is setup cost. Setup cost occurs when a
machine operates several parts which is also called multi-functionality. This ability
may avoid opportunity cost. The part selection is an important issue. More similar
parts will cause less setup cost. Because a machine will need less setup. Setup cost
also occurs when multiple period is analysed in manufacturing system. Demand rates
change for each period. When different demand rates occur, machines will need
setup to be able to satisfy customer demand. Cell configuration is also effective for
this issue. When cells in the manufacturing system are well configurated according to
all periods, less changes of cells and machine setups will be needed.

The fourth part of objective function is hiring-firing cost. Number of workers is

important when forming a cellular manufacturing system. Each cell should have a
worker. When we don’t have enough workers or have more than needed, workers are
hired or fired. But this causes cost as fine or redundant wages.

8
 The last part of objective function is training cost. A trained worker can operate
various types of machines. This supplies flexibility to the manufacturing system on
scheduling. Training and cross-training also eliminate monotony and support
motivation. The main point in forming a system is, positioning the worker to the
appropriate cell and train if needed. However this brings cost. Because of that, cell
formation methods try to construct a balance between training a worker and forming
a cell. When a machine is more complex than talent level of worker in the cell, that
worker should be trained to that complexity level. A new trained worker would not
be as efficient as a high talented worker in real life.

Actually, all components of Objective Function cost equation mainly depend on

formed cells in the manufacturing area. Different cell formations may cause different
cost calculations. Because of that, cell formation methods have a big role on cost
calculation of the whole system.

The most common cell formation method is similarity determination. It is also a

very easy method to use. However it is not capable to real systems. Because this
method does not let use constraints and any other attributes of the real systems
except weights or sequences of parts. Many researchers have many formulations for
similarity calculations. However the characteristics of the system, which are
explained above, change these formulations. After the similarity determination, the
machines which have highest similarity values, form a cell. And cellular
manufacturing system is constructed. However this method does not form healthy
manufacturing cells. Because this calculations does not show the optimal number of
cells, and different combinations should be tried or a specified number should be
given to the system. An overview for these different formulations of different
characteristics is shown in the Table 2.1.

The formulas in the Table 2.1 denote;

a: number of parts visit both machines;
b: number of parts visit machine i but not j;
c: number of parts visit machine j but not i;

9
 d: number of parts visit none of machine i and j.

Table 2.1 Similarity coefficient formulas for general purpose (Yin & Yasuda, 2006)

Similarity Coefficient Definition Sij Range

Jacard a/(a + b + c) 0-1
Hamann [(a + d) -(b + c)]/[(a + d) + (b + c)] -1 to 1
Yule (ad - bc)/(ad + bc) -1 to 1
Simple matching (a + d)/(a + b + c + d) 0-1
Sorenson 2a/(2a + b + c) 0-1
Rogers and Tanimoto (a+d)/[a+2(b+c) + d] 0-1
Sokal and Sneath 2(a + d)/[2(a + d)+ b + c) 0-1
Rusell ind Rao a/(a + b + c + d) 0-1
Baroni-Urbani and
Buser [a + (ad)1/2]/[a + b + c + (ad)1/2] 0-1
1/2
Phi (ad - bc)/[(a + b)(a + c)(b + d)(c + d)] -1 to 1
1/2
Ochiai a/[(a + b)(a + c)] 0-1
2
PSC a /[(b + a)*(c + a)] 0-1
Dot-product a/(b + c + 2a) 0-1
Kulczynski 1/2[a/(a + b) + a/(a + c)] 0-1
Sokal and Sneath 2 a/[a + 2(b + c)] 0-1
Sokal and Sneath 4 1/4[a/(a + b) + a/(a + c) + d/(b + d) + d/(c + d)] 0-1
Relative matching [a + (ad)1/2]/[a + b + c + d + (ad)1/2] 0-1

In Table 2.1, similarity coefficient formulas for general purpose, which are used
in the literature surveys, are seen. But there exist more than these formulas. And
some formulas are derived from these general purposed ones. Hwang & Ree (1996),
Gupta (1993), Won & Kim (1997), and Won (2000) constructed similarity
coefficient formulations with alternative routes. Gupta’s formulation also includes
operation sequences, production volumes and operation times.

Vakharia & Wemmerlöv (1990), Selvam & Balasubramanian (1985) constructed

formulations with operations’ sequences. Seifoddini (1988) used Jaccard similarity
coefficient. His formulation also uses operation volumes. Choobineh (1988) and
Tam(1990) are other researchers worked on operations’ sequences. The difference is,
they worked operations’ sequences by distances. Balasubramanian & Panneerselvam
(1993) studied on operation sequences, additional cell arrangements, production
volume, over moves and costs of them.

10
 It was told that weight parameters were used in cell formation methods. It is also
used in similarity coefficient determination. The basis is general purposed similarity
coefficient formulations. The adaptation is done by injecting the weight into the
general purposed formulation.

The another main point in our study is Worker Assignment. This issue is
discussed in the next section.

2.3 Worker Assignment In Cellular Manufacturing Systems

In previous parts, it was told that the worker was one of components of Cellular
Manufacturing System. Although it is a very important issue, worker assignment is
always eliminated in many studies. Because analyzing human in manufacturing
systems is difficult. Assignment should be carried out by skills of workers. These
skills can be divided into technical and human skills. But making this classification is
also hard. Studies are done to be able to useful for real manufacturing areas. The
more we eliminate issues when modeling the system, the less studies match with real
life. Forming a suitable manufacturing system is the beginning of effectiveness.
Using resources effectively is as important as choosing the right system. Cellular
manufacturing is both an advantageous and an easy system to use. However forming
a cell is not easy as using it. Also worker assignment makes it harder. Because of this
hardness, many heuristics are used to build the system in the literature.

We know, a cell is composed from a family and related machines. On the other
hand, worker is main point of a cell. When we form a cell, we should supply suitable
machines and suitable workers for the family of parts. When machine or worker on
the hand is not available, multi-functional machines and training or cross training for
workers, may be a solution for the system. But this brings extra cost.

Many attributes of workers affect the system. For example, motivation, education,
trainability, multi-functionality, assiduousness, ability, etc. Generally multi-
functionality, ability and trainability attributes are used in system modeling studies.

11
 When modeling a real life problem, we can apply human issues in eight broad
areas: worker assignment strategies, skill identification, training, communication,
autonomy, reward/compensation system, teamwork, and conflict management.
(Bidanda, Ariyawongrat, Needy, Norman, & Tharmmaphornphilas, 2005)

Assignment is the most important part of manufacturing systems. It is also so

difficult. Ability, trainability, suitability of worker and needs of the manufacturing
area should be analysed. A wrong decision will cause an extra cost of training,
material handling or re-work. Assignment can be done by mathematical models or
some other heuristics.

Skill identification should be done properly. It is used to compare with the task
and assignment will be carried out by this identification. As told above, it can be
divided to human and technical skills. Technical skills are generally some skills on
accomplishing the task. Human skills are about personal communication, harmony
with the team or motivation. Training or cross-training becomes a part of assignment
in here. When worker is not suitable for the task, training will be a solution for the
problem. But this brings extra cost for the manufacturing system. And also ability of
worker should be analysed before this method. Because a training would not give the
same result for different workers. The worker should be able to talent-upgrade.
Everybody has some capacity, but not same as each other. Also the amount of
training would not be same for everyone. An analyse should be done for both worker
and task, before a training application.

Communication ability of workers in a manufacturing area is important.

Communications within workers and between workers and management are
important for task definitions and assignments, problem definitions, and solutions of
these problems. Also communication has a significant importance on training.

Type of autonomy should be well analysed. In a manufacturing system the control

can be given to the worker for a work area, or everything can be managed from the
top management. In cellular manufacturing, workers have their own responsibilities

12
in their cells. Cellular manufacturing system uses advantages of this autonomy type.
When workers have responsibility of their cells, they may be able to solve problems
occur in the cell, they can maintain machines in the cell, they can manage their cell.
Manufacturing area will be divided into small manufacturing organisations that each
have a keeper inside. Because of that, this system needs multi-functional workers.

Motivation is a very important issue in manufacturing systems. It affects to many

issues on humans, especially carefulness. Reward and compensation systems are
used for motivation of workers. It is a remarking system for both positively and
negatively; positively to support the good actions, and negatively to eliminate the bad
and wrong actions.

Generally workers work in teams in manufacturing systems. A team may consist

of personnel from many areas. That means communication within the team and
acting through team’s destination is an important issue in manufacturing systems to
achieve success. Team’s success means system’s success. This will be achieved by
the harmony of team members. The harmony is for both technical and personal skills.

In a manufacturing system, there may work many workers in same area. And one
will affect the other by his own function in the manufacturing system. Or some
different ideas may be occur in the system. Conflict management have a significant
role in this situation. Many workers work together with many different skills and
positions in the same area. Conflict is sometimes an unavoidable situation. It is
important to turn this situation to a helpful and an useful situation. This is called
conflict management and headmen take role in this position.

Worker assignment affects the manufacturing system both by cost and

productivity. It is an inter-related issue with cell formation in a cellular
manufacturing area. This issue is as important as part and machine similarity.
Because the compatibility of worker with machines may be strengthen or weaken the
stability of a cell. A literature review is carried out in the next section on Cellular
Manufacturing System Solution Methods and Human Issue in this system.

13
2.4 The Literature Review On Cellular Manufacturing Systems

A general literature review is carried out in the next part for cellular
manufacturing systems. And a literature review for human issue is carried out after
that part.

2.4.1 Cellular Manufacturing Systems

In Cellular Manufacturing (CM), in each cell, some operations are done on parts
by machines, so that the main objective is maximizing the intra-cell operations while
minimizing the number of inter-cell movements (Saeedi, Solimanpur, Mahdavi, &
Javadian, 2010).

Some researchers analyzed cellular manufacturing systems by their studies.

Guerrero, Lozano, Smith, Canca, & Kwok (2002) studied cellular manufacturing by
weighted similarity coefficients, a new self-organizing neural network and a linear
network flow model. Cell formation has two steps: first, part families are formed and
then machines are assigned. Also a Maximum Spanning Tree heuristic is used in
their study to compare the results. Self-organizing neural network is used in the main
part of the problem; forming part families.

Wu, Chu, Wang, & Yan (2007) studied hierarchical genetic algorithm for cellular
manufacturing. In their problem, routing (sequence), work load, machine capacity,
demand, batch size, and layout type are searched. Cell formation and layout design
are carried out simultaneously. First, a mathematical model is constructed. Then
genetic algorithm is used for cell formation problem. Crossover and mutation are
both used. Dynamic assignment is done in their problem.

Balakrishnan & Cheng (2007) studied cellular manufacturing problem with multi-
period. They also use demand and resource uncertainty. This manufacturing system
is harder than single-period to construct. Also the demand and resource are not
known. A mathematical model is constructed.

14
 It is seen that many studies were done for single period, which demand and some
other values were constant. Because multi-period systems are hard to solve. But it
does not occur like this in real life systems.

Safaei, Mehrabad, & Ameli (2008) studied a dynamic cellular manufacturing

system with a hybrid SA algorithm. A mixed-integer programming model is also
developed in their study. The advantage of their study is, they calculate inter and
intra-cell material handling by sequence of operations. That means, both side of
traffic between two cells are calculated. And machine replication is allowed in the
model. The hybrid system consists of mean field annealing algorithm and simulated
annealing algorithm. It was explained before that SA algorithm needed an initial
solution to be able to improve it. In this study the mean field annealing algorithm is
used for that initial solution which is needed by SA algorithm. Mean field annealing
algorithm is a combination of neural networks. It is seen that mean field annealing
algorithm increases the performance of the model and speeds up the algorithm to
reach the optimal result. Back order is not allowed and a limit is determined for the
maximum cell number. The problem is for multi-period. And both intra and inter-cell
movements are calculated in the model. But human issue is not included in the study.

Pailla, Trindade, Parada, & Ochi (2010) studied on a comparison between

simulated annealing and genetic algorithms. They use an algorithm to improve both
simulated annealing algorithm results and genetic algorithm results. This algorithm
uses some other crossover rule instead of classic methods. Both results for some
numerical examples reach to almost same performance. But SA algorithm finds even
better results than previous literature studies for the same problems.

Tavakkoli-Moghaddam, Rahimi-Vahed, Ghodratnama, & Siadat (2009) studied

solving a cell formation problem by simulated annealing with two types of cells. One
type can produce different types of parts, the other one can produce specific types of
products. First, a non-linear mathematical model is formed. Then the model is solved
by Simulated Annealing Algorithm. Three objective issues are included in the model.

15
These issues are; minimizing delay cost of parts, minimization of unproductive times
of cells, maximizing the unused capital.

A literature survey was carried out by Saeedi, Solimanpur, Mahdavi, & Javadian
(2010), on details of heuristics. Table 2.2, shows that the points which are taken or
not taken into consideration by researchers who studied cell formation until 2010.

Table 2.2 Literature survey on heuristics (Saeedi, Solimanpur, Mahdavi, & Javadian, 2010)

Sequence Exceptional
Applied of Production Elements Intercellular
Reference Methodology operation Volume (Voids) Movements
Islier Ant Algorithm No No No No
Prabhaharan et
Ant Algorithm Yes Yes No Yes
al.
Mak et al. Ant algorithm Yes No No No
Spiliopoulos and Ant algorithm Yes No No Yes
Sofianopoulou
Kesen et al. Ant algorithm Yes No No No
Satolgu and Goal
No No No No
Suresh Programming
Clustering
No Nc No No
Kao and Fu Algorithm
Pandian and Neural
Yes No Yes Yes
Mahapatra Networks
Genetic
Yes No Yes No
Mahdavi et al. Algorithm
Mahdavi and Heuristic
Yes No Yes No
Shirazi Aıgorithm
Simulated
No Yes No No
Arkat et al. Annealing
Ahi et al. TOPSIS Yes No Yes No
Wang et al. Scatter Search Yes Yes No No
Murugunandam GA + Tabu
Yes Yes No No
et al. Search

We can see that, some researchers did not work with sequence of operations
which are related with inter and intra-cell movements. This means, traffic between
two parts for one direction or for both directions has same importance. This
assumption may affect the result.

16
 We can see another point that, volume of parts are not taken into consideration by
some researchers. The costs of material handlings for one part and for many parts are
not same. And a part with high quantity can get ahead about reducing cost, instead of
a part with low quantity even though unit material handling cost is lower.

Some researchers calculate the affect of exceptional elements but most of them do
not. Exceptional elements mean inter-cellular movements. When a part, which is not
totally belong to a cell, needs an operation; it should enter that cell. Or if a machine,
which a part needs an operation from, takes place out of a cell; part should go out of
this cell.

It can be seen in Table 2.2 that, many researchers did not calculate inter-cellular
material handling. Inter-cellular movements are the most important cost part of the
objective function. A cell formation method which is applied with this assumption
would not be realistic. Because the main point in cost calculation and cell formation
is minimizing the inter-cellular movements which means trying parts to make stay in
their cells.

When we want to use heuristics to solve our problems, we need to have some
assumptions to be able to achieve results. If we have fewer assumptions, then our
model will respond closely to the real life problems. Also the importance of the
assumption for that problem is a point that should be critically determined. If the
issue that we make an assumption is a main point of our problem or a performance
criteria, then our model would not respond as good as we expect.

2.4.2 Human Factor In Manufacturing

In this part, a literature survey is carried out on human issue in cellular

manufacturing systems. And a summary table is done.

Dawis & Mabert (2000) studied on worker assignment and order releasement.
They implement two different mathematical models. Instead of productive resources,

17
they decided to study on inventory reducing formulations. Reassignment is allowed
in their models. Two issues which they studied, should be in a harmony. So they
implement two types of algorithm to achieve that harmony. First one is worker
assignment and order releasement are done simultaneously. The second method is
sequentially. Then they implement a heuristic, to be able to see the difference of
these two methods. At the end of the study it is seen that sequentially calculated
model gives better results and it is more sensitive to critical time intervals.

So, worker assignment can be divided into two categories:

1) Post-cell formation worker assignment (Norman, 2002).
2) Simultaneous formation of cells and worker assignment. (Aryanezhad, Deljoo,
& Mirzapour, 2009)

Stevens & Campion (1994) have found 14 KSA (knowledge,skill,ability) types in

their study. They say that these 14 different types can be used by an assignment
problem. Erin, Fitzpatrick, Ronald, & Askin (2005) say that a worker should be
analysed not only for technical skills but also for interpersonal skills. They worked
on a mathematical model. In their model, it is known which machine is located in
which cell, at the beginning of the model. And they studied on a heuristic model that
is called Balanced Heuristic Model. When multiple teams of workers needed, this
heuristic is used but it does not give good results if we have workers more than
needed in the system. Workers are chosen by their inertias to the teams. Also they
use Kolbe Conative Index to be able to measure the instinctive behaviour of workers.

Multi-functionality has a big role on assigning workers to the teams. Also training
is the main point of it. Slomp, Bokhorst, & Molleman (2005) have a study on cross-
training of workers. They used an integer programming model to allocate workers to
cells. Their model also decides if that worker should be cross-trained, to be able to
balance the work load on them. The objective is minimizing the cost, while
allocation is being done. They say that some skill identifications should be done. And
it is assumed in their model that, if a worker is cross-trained, his productivity would
be lower than a worker which is already able to operate that machine. The model also

18
has some constraints like limit on multi-functionality and machine redundancy. And
the training is given only for multi-functionality, not for upgrading workers’
productivity. Azizi, Zolfaghari, & Liang (2010) has a study on job rotation. They say
that boredom should be eliminated as possible to be able to make workers learn
operations. They studied on a mathematical model with skill identification and
boredom, and a metaheuristic SAMED-JR for large scaled problems. This
metaheuristic is a combination of Simulated Annealing Algorithm(SA) and Genetic
Algorithm(GA). They found that the metaheuristic, they used, gave better results
than using only SA or only GA.

Multi-functionality is also called as labor flexibility. This is the ability of

assigning workers to different operations. Cesani & Steudel (2005) define that as
intra-cell operator’s mobility. They classify labor strategies according to the machine
and operator assignment as dedicated, shared and combined. They used simulation to
see the effect of workload balancing. At the end of their study, it can be seen that
labor flexibility has a huge effect on productivity of the whole manufacturing system.
They used different operator numbers and different labor assignment strategies as
told before.

Askin & Huang (2001) made a study on forming effective worker teams. A mixed
integer programming was used in the study. They separated workers abilities as
technical and administrative. The model that they instructed includes worker
assignment and training for multi-functionality. They used meta-heuristics to be able
to achieve results for problems with big capacity. At the end of their study, it can be
seen that meta-heuristics give good results with reasonable time for NP-hard
problems. One of the model they used is Simulated Annealing(SA) algorithm. They
solved more complex models with SA. It can be seen that SA could achieve optimal
solution in their study. Another study was done by Aryanezhad, Deljoo, Mirzapour,
& Al-e-hashem (2009) with multi-functionality of workers and also machines. They
proposed a solution by Linear Integer Programming model. Objective function
includes manufacturing costs, material handling costs and personnel costs. The main
point of the study is, more than one period is included. But this model can be used

19
only for small sized problems. Because a real life problem would convert the system
to NP-hard problem.

Corominas, Pastor, & Rodriguez (2006) studied on a real system with multi-
functional workers. This study is not based on cellular manufacturing but applied in a
real life manufacturing system and investigates multi-functional worker assignment.
The difference from other studies is, they assigned tasks to operators. Problem is for
multi-period. Therefore, a mathematical model would come out with NP-hard
situation. Researchers applied another method. They solved problem for one period
and then allowed results as an input for the next period. So they divided the planning
horizon.

Another study with multi-period time horizon is done by Mahdavi, Aalaei,

Paydar, & Solimanpur (2010). Mathematical model was constructed. All cost issues
like hiring, firing, intercellular material handling etc. were used in objective function.
But cell number determination was not included in the study. This means, the cell
number was specified and allowed to the problem as an input. Model applies
flexibility on worker assignment, but does not include the cost. Backorder
opportunity exists in the model. But it is NP-hard problem for big scaled problems.

It can be seen that studies for multi-functionality are based on the training and
motivation of workers. It is accepted that a worker can be equal to a couple of
workers, at least more than one, by cross-training. Same tasks for long periods will
bring monotony for workers. This concept usually brings inattentiveness and work-
related accidents, too. Demand will also be flexible in a short time interval. In real
life situations, companies should be able to respond demands as fast as they change.
Multi-functional machines are an one of alternative applications. But still, multi-
functionality of workers are needed. A multi-functional machine can be operated by
a multi-functional operator, or different operators should be used for different
operations. But it is not a realistic application for a real life situation. Because
training is costed generally lower than hiring a worker. A decision making position

20
appears in here; which worker should be trained for which operation. This becomes
the basic unit of assignment nowadays.

To be able to have feasible work situations with workers, we should analyse their
abilities. Suitable tasks for right abilities will upgrade workers’ performance and also
systems performance. This point is studied in this thesis. When a training capacity is
occurred, a worker can be able to work for multiple points that usually brings high
motivation. In our study, training and upgrading of talent level is possible. And a
worker can work with multiple machines, which called multi-functionality.

Multi-functionality is also has a big role when number of machines are more than
number of workers. We have two options in this situation; unemployed machines in
the manufacturing area, or cross-training of workers. A worker can operate several
machines by cross-training. His talent also should be taken into consideration. The
main objective is minimizing the cost of both cross-training and unemployed time of
machines. Unemployed machines mean keeping the system away from demand
satisfaction. And unsatisfacted demand means receiving lower demand at the next
time. Especially nowadays, in a competitive market, time is a kind of money figure.
Workers should be work on whole production time, because unproductive time
means cost for a manufacturing system. It can’t be provided unique duty for a worker
every time. Demands may be changed through some time interval. So a worker
should operate several machines or be able to do several operations. In this way, a
manufacturing system can satisfy demand.

In the Table 2.3, Cesani & Steudel (2005) have listed factors affecting the
development and deployment of labor flexibility.

Cesani & Steudel (2005) studied on labor assignment. They say that, although
many factors are identified as influential in determining labor flexibility decisions,
some of them are qualitative in nature and thus, difficult to model. The model and
framework presented in their work, concentrate on those aspects that can be
quantified and for which information is readily available or could be determined. The

21
propositions were considered in the framework evolved from the empirical study and
their impact on system performance were investigated with the purpose of
developing knowledge about the complexities of the labor allocation process in labor
limited manufacturing cells.

Table 2.3 Factors affecting the development and deployment of labor flexibility (Cesani & Steudel,
2005)

Factor Issues
Equipment proximity Size of the cell/cellular
area
Layout
Wort flow (organization) Location/size inter-station
buffers
Level of automation
(manual, semi-automatic, Age/condition
Equipment CNC machinery)
Utilization
Type of labor
Dedicated assignments Combined assignments
assignments possible
(shared and dedicated)

Number of operations Bottleneck machines

Cross-training level of
Proficiency level Machine tending
individual operators
requirements
Quality of training Individual cycle times
Relative machine
Workload in the cell Variations in demand utilization
Rush jobs
Job rotation frequency Division of activities in
the cell
When and where to move Inter-cell vs. intra-cell
Job design
rules mobility
Response to load
imbalances
Responsibilities of the Operator's ownership:
Labor aspects operators machine vs. cell
Type of supervision Job autonomy
Personnel practices
Wort teams
Effective machine
Effective operator
utilization (through
utilization
Managerial concerns wortload sharing)
Leveled operator Focus on bottleneck
assignments operations Crosstraining
Administrative aspects
vs. compensation systems
Union restrictions

22
 The investigated companies currently do not use formal models such as
spreadsheet-based rough-cut analysis, linear programming or simulation to assign
operators to machines. Most labor assignments are made based on the experience of
the personnel involved with the cells. Many times, particularly at the cell
implementation level, labor decisions involve a lot of trial and error and therefore,
companies do not take the best use of their labor and machine resources.
Management in the companies investigated expressed the desirability for models and
guidelines to assist in the labor allocation process since supervisors and operators
disagreed on the most appropriate labor allocation strategies. Furthermore, while
developing a completely flexible workforce is a goal in both of these companies,
neither of them have objective measures to evaluate the impact that increasing
operators’ cross-training has on cell performance. Thus, cross-training decisions are
many times made arbitrarily. (Cesani & Steudel, 2005)

M\W A B C D Dm M\W A B C D Dm

1 X(12) X(0) 12 1 X(6) X(6) 12

2 X(0) X(14) 14 2 X(7) X(7) 14

3 X(6) X(4) 10 3 X(2) X(8) 10

4 X(10) X(0) 10 4 X(2) X(8) 10

5 X(0) X(6) 6 5 X(3) X(3) 6

6 X(2) X(6) 8 6 X(4) X(4) 8

WL 18 18 12 12 WL 15 15 15 15
Figure 2.1 Example of machine worker matrix (Slomp, Bokhorst, & Molleman, 2005).

In Figure 2.1, Slomp, Bokhorst, & Molleman (2005) have shown an example of
multi-functionality of workers. The difference between two matrixes shows the
multi-functionality of that worker. For example, in the left-handed figure, it cen be
seen that worker can operate machine 1 and 3. In the right handed figure, we can see,
worker A has cross-trained and is able to operate machine 2, too. But his total
workload gets lower because a worker might be less productive on a new duty.

23
 Several authors presented a hierarchical scheme for work force organisation
problems that consists of three phases: (1) planning; (2) scheduling; (3) allocation.
The assignment of tasks to multi-functional workers is done during phase (3), once a
schedule has been assigned to each worker. (Corominas, Pastor, & Rodriguez, 2006)

Table 2.4 shows the attributes of researches. We can see that almost all
researchers used mathematical model for their problems. Some of them are for single
period, and the rest are for multiple periods. Some of them used heuristics, but these
heuristics have some deficits.

Table 2.4 Literature survey

MATHEMATICAL MODEL

MULTI-FUNCTIONALITY
RESEARCHERS

HEURISTIC

ANOTHER

PERIOD
Sequentially
Assignment, more
+ -
Dawis and Mabert Simultaneously than 1
(2000) Assignment
14 KSA
Stevens and Campion (knowledge,ski
(1994) ll,ability) types
Erin, Fitzpatrick, Balanced Heuristic
+ Team selection worker 1
Ronald, Askin, (2005) Model
Slomp, Bokhorst,
+ worker 1
Molleman (2005)
Azizi, Zolfaghari,
+ SAMED-JR - 1
Liang, (2010)
Cesani and Steudel
simulation worker 1
(2005)
Askin and Huang Simulated
+ worker 1
(2001) Annealing
Aryanezhad, Deljoo,
more
Mirzapour, Al-e- + both
than 1
hashem (2009)
Corominas, Pastor, Result Is Input For more
Rodriguez (2006) The Next than 1
Mahdavi, Aalaei,
more
Paydar, Solimanpur +
than 1
(2010)

24
 CHAPTER THREE
PROBLEM DEFINITON AND SOLUTION METHODOLOGY

In this thesis, simulated annealing algorithm of a cell formation problem, which

number of the cell is determined by Kaiser’s Rule (Bashir & Karaa, 2008), is
proposed. The cost function, which we want to minimize, consists of inter-cellular
movements, training and manufacturing costs. In this problem, products move by
batches and demand levels are predetermined. It is studied single period. Alternative
routes are included in the problem and which machine produce which part is
specified. Some assumptions are included by the problem. They are;

- The problem has one period.

- 0-1 part-machine incidence matrix is used.
- Machines and parts change place in problem. Workers are stable in cells.
- The demand for each part type is known.
- The number of machines in the system is known.
- Training doesn’t take any time.
- Trained worker is assumed to be reached same productivity level as high
talented worker.
- Processing time for parts are randomly distributed and vary on different
machines.
- The inter-cell material handling cost per batch is known and vary for different
parts.
- A part may have alternative routes.
- A worker can operate more than one machine (multi-functionality).
- Cost of training depends on levels.
- Parts are moved in batches between cells.
- The machine relocation cost is 0(zero).
- Number of workers are as much as number of cells in the system.

To be able to understand the algorithm, it should be seen indices and input data;

25
 Indices are;

Alternative machine index: a=1; 2; . . . ;A

Cell index: p = 1; 2; . . . ; P
Iteration index: t = 1; 2; . . . ;T
Machine index: k = 1; 2; . . . ;K
Part type index: j = 1; 2; . . . ; J
Talent index: y = 1; 2; . . . ; Y
Worker index: v = 1; 2; .......;V

Input data;

B(k)= operating cost of machines per time unit

BN= a large number
Bs= batch size
C(j)= number of transportation for each part
D(j)= demand vector of parts
G(j)= inter-cell material handling cost by each part
Te(t)=temperature level of SA procedure for each itration
W(v,y-1)= Cost of worker training level skips 1 to 2, 2 to 3,…, (y-1) to y
X(k,j)= part-machine incidence matrix, XЄ(0,1)
Xa(a,j)= alternative machine matrix
Z(k,j)= production times of parts by each machine

Using these datas, the objective function is formed as;

 J J
 D( j )   Y
obj (t )  ( DZXB)    C ( j ) x  G ( j ) x     W ( y  1)
 j 1 j  1  Bs   y 2

(Equation 1)

obj(t) objective function value of iteration t

26
 This equation consists of three parts. The first part is sum of production cost, the
second part is sum of intercellular material handling costand the third part is sum of
training cost for workers.

First, our algorithm calculates the manufacturing cost of the system. This is shown
as;

 J K 
DZXB  D( j ) x   Z (k , j ) xX (k , j )  xB(k )
 j 1 k 1 
(Equation 2)

Manufacturing cost does not change by different cell formations. For each
iteration of alternative part-machine matrix, manufacturing cost remains same. When
we use different alternative for part-machine matrix, manufacturing cost changes.

Then algorithm calculates the inter-material handling cost for each cell formation.
Each alternative part-machine matrix and each iteration for these matrixes generate
different cost function. The steps for inter-material handling cost calculation isshown
below;

1. Find which parts need which machines,

2. Check the machines, which cell that they belong to,
3. If they spread out n different cells, set the inter-material handling (n-1)
4. Multiply inter-material handling number for each part by demand and by unit
inter-material handling cost. This equation is shown as;

 D( j ) 
Inter  MaterialHandlingCost   handling (i )  xG ( j ) x  
 Bs 

(Equation 3)

Inter-Material Handling cost changes by different cell formations.

27
 Finally algorithm calculates the training cost for each cell formation. Each
alternative part-machine matrix and each iteration for these matrixes generate
different cost function as inter-material hadling cost. Because each combination of
the part-machine-cell string cause different talent necessities. It should be determined
that how complex machines are and the difference between this complexity and
workers’ talent degree. Then the cost is calculated by;

Y V
Wtotal   W (v, y  1)
y  2 v 1
(Equation 4)


Xmstep
According to Pinedo (2004), the third jki of a local search procedure is the
j 1
search process within ) neighbourhood
S (k , ithe after neighbourhood design. According to
 
J
Ym ji to be
this process; the value, that is tried jk  Xm jkiby SA algorithm, is objective
Zmdecreased
j 1
value. That value is also manufacturing system cost in this problem. Algorithm
checks if the objective value is lower than or equal to the value that is decided to be
reached. When we reach to that value, algorithm stops. Otherwise it changes the
assignment of machines in cells. This is the acceptance-rejection criterion according
to Pinedo (2004), which is the last step of the local search procedure. Neighbourhood
is used to change parts and machines in cells by exchangement and mutation. Saeedi,
Solimanpur, Mahdavi, & Javadian (2010) say that, some heuristic algorithms like
Hill Climbing technique, may found the Local Optimum instead of the Global
optimum because the movements leading to a new point worse than the current point
are not allowed. SA algorithm allows to choose a worse result with a probability.
This method helps to keep solution from local optimum. Because the goal is finding
global optimum.

In the next step, an overview is done on the algorithm that is constructed for
cellular manufacturing system.

28
3.1 The Proposed Algorithm And Components

In this part, detailed analyses are done for the problem, which is told above, and
the poposed algorithm for this problem. The proposed algorithm tries to minimize the
objective cost function with alternative routes and specified number of cells. This
specified number is found by Kaiser’s Rule. The total cost which is calculated for
that specified number of cells, is tried to be minimized by the Simulated Annealing
Algorithm. Before the detailed steps of algorithm, two methods are explained below
which are used in this algorithm. The Kaiser’s Rule and Simulated Annealing
Algorithm.

3.1.1 The Kaiser’s Rule

The Kaiser’s Rule is an approach, which gives the most suitable number p for the
system to divide. We can call it as optimal cell number. The Rule makes it by finding
the most similar parts and machines.

There are many approaches to find the similarity between parts and machines
which should be calculated to form cells. The similarity coefficient method is always
prefered among these approaches. Because this method is easy to use and gives
useful results. As told before, Jaccard’s similarity coefficient approach is used in our
algorithm. It is denoted by Ski. This approach considers the relationship between
Y itVdoesn’t consider this relationship as a traffic, which has a
parts and machines. But
total  
Wformulation
direction. The is;W (v, y  1)
y  2 v 1

 Xmj 1
jki

S (k , i )  1 < i,k < K

 Ym  Zm jk  Xm jki 
J

ji
j 1
(Equation 5)

29
J = number of parts.
K = number of machines;
Xmjki = 1 if part j has operation on both machines i and k, and 0 otherwise;
Ymji = 1 if part j has operation on machine i, and 0 otherwise;
Zmjk = 1 if part j has operation on machine k, and 0 otherwise;

The similarity coefficients matrix is formed by Ski’s. The matrix elements range
from 0 to 1. According to the matrix theory, if the similarity coefficient matrix is real
symmetric, it has n real eigenvalues. Moreover, the eigenvectors corresponding to
these eigenvalues are linearly independent and each eigenvector represents a cell.
These cells have low intercorrelations because the eigenvectors are uncorrelated, and
therefore there should be low similarities between machines that are associated with
different cells (Bashir & Karaa, 2008). This approach is simply called Kaiser’s Rule.
The equation is;

(S - λI) = 0 (Equation 6)

I denotes; the identity matrix,

S denotes; the similarity coefficient matrix,
λ denotes; the root (eigenvalue) of the equation,
denotes; n eigenvector.

Kaiser’s Rule says that the number of eigenvectors which are greater than 1 (one),
both shows the number of cell in a system that should be and suitability of this
system for the cellular manufacturing. If we have more than one eigenvector that fits
to that condition, the system is suitable for a cellular manufacturing.

Kaiser’s Rule is used with Simulated Annealing Algorithm in this thesis. Instead
of trying different cell number alternatives, the right number is given to the problem
by Kaiser’s Rule. This approach made the algorithm easier and faster to reach to the
feasible solution. In our study, it was seen that giving an optimal number (the value
that is found by Kaiser’s Rule) to the problem as a cell number returned a lower

30
cost(objective function) with same number of iterations. Simulated Annealing
Algorithm is explained in the next part.

3.1.2 The Simulated Annealing Algorithm

Simulated annealing (SA) is one of metaheuristics that have been used extensively
to solve combinatorial optimization problems. By simulating the phenomenon that
takes place in the cooling of pure substances from the liquid to the solid state, SA
improves a solution to an optimization problem gradually until it finds the best
solution in the search space (Kirkpatrick, Gelatt, & Vecchi, 1983). In each iteration,
the algorithm accepts a randomly generated solution in the neighborhood of the
current solution directly if it is better or probabilistically if it is worse. So it can be
seen that not only better results are accepted, but also worse results are accepted.
This provides global optimum solution to the algorithm.

Heuristic methods are formed by simulating the nature. Problem solutions are
derived from these simulated systems. Saeedi, Solimanpur, Mahdavi, & Javadian
(2010) say that the Simulated Annealing algorithm is derived from metallurgy and
thermodynamics which incorporated a temperature parameter into the minimization
parameter. A high temperature expands the search space, and with a lower
temperature the search space gets smaller. The procedure starts from a high
temperature and ends at a low temperature. At each temperature, a number of
iterations are done.

It was told that SA algorithm worked by global search algorithm. The basis of
global seach algorithm is local search algorithm. After the examination of local
search procedure, the main difference is explained below.

Pinedo (2004) says that local search procedures can be compared by the following
criterias;

1. The schedule representation needed for the procedure.

31
 2. The neighbourhood design.
3. The search process within the neighbourhood.
4. The acceptance-rejection criterion.

Pinedo (2004) says that the design of the neighbourhood is a very important
aspect of a local search procedure. Because it is the main point of algorithms. The
coding the system for the neighbourhood and using it in solution steps should be
well-calculated. This procedure may be easy for small models, but careness is needed
for more complex ones.

We can view a literature survey with local optimization by Wu, Chang, & Chung
(2008). They used neighbourhood calculations, which are the basis of SA modelling.
They checked performance of their solutions by grouping efficacy calculations. In
improvement studies, two types of movements were used for neighbourhood
calculations to decrease the objective function value. First one is, just one part may
change its cell; second one is, two parts may change their cells simultaneously. This
may faster improve the system.

The simulated annealing algorithm gives result at the end of some iterations. The
result of each iteration is an input of the next one. Pinedo (2004) explained the
formulation. Sk is the solution of iteration k. The best solution found is S0 at that
time. G(Sk) and G(S0) states the objective function of these solutions. Also G(S0), is
used as comparison criteria. If the result which is found in an iteration, is better than
this value, then it is accepted as the new solution. If not, then it is accepted by some
probability or not. This probability is calculated as ;

( ( ( )
P( =exp { } (Equation 7)

with probability 1 - P(S0, Sk) schedule Sk is rejected and Sk+1 = Sk. Best solution S0
does not change because it is better than Sk. The

32
 0 (Equation 8)

are cooling parameters or temperatures which are used to get the solution space
smaller.

As shown in equation 8, cooling parameter gets lower by each iteration. This

makes the solution space smaller, and the probability of acceptance of the worse
result, lower.

Pinedo (2004) presents the SA algorithm as below. Two different stopping criteria
are used for the algorithm; specified number of iterations or stopping when no
improvement occurs. The algorithm below, uses the first criteria.

The algorithm is formed on calculation, comparison and optimization of the

objective function as explained above, in this study. Algorithm starts by checking
input datas. An itial solution should be given to the algorithm to start. Kaiser’s Rule
is used for this initial solution. It is an effective and useful method as a start to find
the optimal solution. Kaiser’s Rule gives the optimal number of cells that should be
formed in the problem. This number is used to form cells on the initial solution.
Algorithm first starts calculations by this initial solution. These calculations give
inputs for the objective function. It was told that alternative routes existed in this
problem. The algorithm runs for each alternative route for a specified number of
iterations. And algorithm runs for same number for cell configuration that is given by
Kaiser’s Rule on same route.

Then each result for iteration number (t-1), is an input for iteration (t). It was told
that the objective function consisted of three parts. The first part is sum of production
cost, the second part is sum of intercellular material handling cost and the third part
is sum of training cost for workers. The sum of these costs is compared with a big
number and the neighbourhood solution is created according to the result. This cycle
turns till the last iteration number. The best result among overall iterations (which

33
alternative routes are included) is chosen as the solution of problem. The comparison
is done by simulated annealing algorithm.

The algorithm is;

Begin
Step 1.
Set k = 1 and select β1.
Select an initial sequence S1 using some heuristic.
Set S0 = S1.
Step 2.
Select a candidate schedule Sc from the neighbourhood of Sk.
If G(S0) < G(Sc) < G(Sk), set Sk+1 = Sc and go to Step 3.
If G(Sc) < G(S0), set S0 = Sk+1 = Sc and go to Step 3.
If G(Sc) > G(Sk) generate a random number Uk from a Uniform(0,1) distribution;
If Uk . P(Sk, Sc) set Sk+1 = Sc otherwise set Sk+1 = Sk and go to Step 3.
Step 3.
Select βk+1≤ βk.
Increment k by 1.
If k = N then STOP, otherwise go to Step 2. (Plaquin & Pierreval, 2000)

According to these definitions, the constructed and proposed algorithm is

explained in the next part.

3.1.3 The Proposed Algorithm

The proposed algortihm for solving the problem is constructed as shown below.

Set nnn=0
Step.1 If nnn <nnnpSet nnn=nnn+1 and generate alternative part-machine matrix,
else stop.
Step.2 Calculate the Similarity Coefficients of each machine

34
 Step.3 Apply the Kaiser’s Rule to find the optimal cell number
Step.4 Build the initial part-machine-cell matrix PMCi
Step.5 Set iteration=1:iterationmax
Step.6 Calculate the manufacturing cost DZXB according to alternative part-
machine matrix
Step.7 Calculate the inter-cellular material handling cost handling(i) according to
traffic between cells for each part
Step.8 Find training cost W(i) if training needed in cells and add to the total
training cost Wtotal
Step.9 Calculate the objective function obj(iteration)
Step.10 If it is lower than predetermined value (BN), accept the obj(iteration),
assign new value as BN, and generate new part-machine-cell matrix solution
Step.11 If not, reject the solution and generate new part-machine-cell matrix
solution from former iteration or accept the solution according to a probability value
of SA and generate a new part-machine-cell matrix solution.
Step.12 If iteration < iterationmax, go to step 5, else go to step 1.

The algoritm starts by taking the input datas. All steps run by these input datas.
Input datas are formed from quantitative datas and matrixes. Some of them are part-
machine matrix, alternative manufacturing matrix, demand matrix, matrix production
times, matrix of machine production cost, matrix of inter-cell material handling cost,
worker training cost matrix, worker talent matrix (randomly distributed) and so on.

First, algorithm generates the alternative part-machine matrix by alternative routes

given as an input. After specified number of iterations, algorithm generates the next
alternative part-machine matrix and run for same number of iterations.

The next step is Kaiser’s Rule. To learn the optimal number of cells, similarity of
machines are calculated and Kaiser’s Rule finds the optimal value according to these
similarities. This optimal number is not change till the next alternative route. This
means, it is used same number of cells for sprecified number of iterations.

35
 Manufacturing cost (DZXB), inter-cellular material handling cost (handling(i))and
training cost (W(i)) is calculated for each iteration according to alternative part-
machine matrix. The sum of these costs is objective function (obj(iteration)).

The simulated annealing procedure starts after objective function calculation. If

the objective value, that is calculated in this iteration, is lower than the predetermined
value (a big number), the new solution is generated from that iterations solution. If it
is higher, two options exist. If the probability value is higher than the randomly
generated value, the same procedure is done as in lower objective function value than
the predetermined value. If this is not occured, the result of earlier iteration is used
for a new solution generation.

This continues till the maximum iteration value. At the end, the lowest objective
function value and its datas are taken from overall results. This is the result of that
route. If any other routes exist, algorithm continues for them and goes to the
beginning of algorithm. This cycle turns for nnnp times, which nn denotes; alternative
route number for one part, np denotes; the number of parts that have alternative
routes.

3.1.4 An Illustrative Example

The steps and details of the proposed algorithm are explained in previous parts. A
numerical example for a manufacturing system with 5 parts and 3 machines is shown
below to understand better how algorithm works.

Set nnn=0
Step.1 If nnn <nnnpSet nnn=nnn+1 and generate alternative part-machine matrix,
else stop.

nnn=1

Xp= 0 1 1 1 0
1 0 0 0 1
1 0 1 0 0 ;

36
Step.2 Calculate the Similarity Coefficients of each machine
S= 1 0 0.25
0 1 0.3333
0.25 0.3333 1 ;

Step.3 Apply the Kaiser’s Rule to find the optimal cell number
C = 0.5833
1.0000
1.4167 ;

Number of values that are higher than 1 is two. That means Kaiser’s Rule offers 2
cells for the manufacturing system.

Step.4 Build the initial part-machine-cell matrix PMCi

PMCi= [ ]

Step.5 Set iteration=1:iterationmax

iteration=1

Step.6 Calculate the manufacturing cost DZXB according to alternative part-machine

matrix
DZXB = 815

Step.7 Calculate the inter-cellular material handling cost handling(i) according to

traffic between cells for each part
handling= 20

Step.8 Find training cost W(i) if training needed in cells and add to the total training
cost Wtotal
Wtotal= 2200

Step.9 Calculate the objective function obj(iteration) (BN=1.000.000)

37
obj = 3035

Step.10 If it is lower than predetermined value, accept the obj(iteration) and generate
new part-machine-cell matrix solution
3035<1.000.000

PMCi= [ ];

Step.11 If not, reject the solution and generate new part-machine-cell matrix solution
from former iteration or accept the solution according to a probability value of SA
and generate new part-machine-cell matrix solution.

Step.12 If iteration < iterationmax, go to step 5, else go to step 1.

iteration =1
go to step 1.

This example is the first iteration of first route. Algorithm goes on working in this
presentation. At the end of all iterations of all routes, the minimum value of objective
function and its datas are choosen by algorithm.

38
 CHAPTER FOUR
ANALYZING THE COMPUTATIONAL RESULTS

The cellular manufacturing problem and its solution method, which are studied in
this thesis, were explained in previous section. Matlab R2008a version is used to run
the proposed algorithm for solving the problem. This study is executed on an Intel
Centrino Duo, 1.73 Ghz, and Windows 7, using 1.5 GB RAM. Computations are
done for different numerical combinations and different sized matrices of datas. In
this chapter, computational results of numerical examples are shown, which are
derived for a system with 5 machines and 7 parts, 10 machines and 15 parts, 18
machines and 30 parts.

In this part, cell formation is done which cells include workers, machines and
parts, by minimizing the objective cost function. In the problems, machine
complexity levels and worker talent levels are included. Each cell has one stable
worker who’s talent degree is randomly assigned. If machines’ complexities are
higher than worker’s talent in a cell, then this worker is trained to operate these
machines. Inter-cell material handling cost is also included in the problem. Number
of cells is not specified but with part-machine matrix, optimal number of cells is
calculated by Kaiser’s Rule. This makes the algorithm faster and provides an
advantage on forming cells. A part can go alternative machines (different routes) to
be operated for the same operation. A worker can operate more than one machine.
This is called multi-functionality. And cost of training depends on complexity levels.
A single period is studied in this thesis. The table below, elaborates on the
characteristics of problem.

Several numerical examples are solved for each of 5x7, 10x15 and 18x30 sized
problems to see and analyze the numerical results. This is called Sensitivity Analyse,
and done for each problem to see the effects of objective function cost components
on the total cost. These components are; intercellular-material handling cost and
training cost. Intercellular-material handling costs and training costs are used for

39
different level of values, to see their effects on results. Detailed analyze of
computational results are shown in next sections.

Table 4.1 Characteristics of the problems

Number of machines specified 5-10-18
Number of parts specified 7-15-30
Machine complexity degree specified Max 3
Worker talent degree specified Max 3
Worker talent degree upgrading by training Max 2
Period capacity of the problem Single period
Material handling cost Specified
Unspecified(Calculated by
Number of cells Kaiser's Rule)
Equal to
the
Unspecified(equal to the number of
Number of workers number of cells) cells
A machine
can
produce
several
Multi-functionality of machines Enable parts
Alternative routing Enable
Multi-functionality of workers (trainig is Enable(one worker can
based on the most complex machine in the operate more than one
cell) machine)
Training Enable (takes 0 time unit)
Trained worker's
productivity is equal to the
Productivity of workers trueborn talented

Training cost Changes by talent level

40
4.1 Computational Results For 5x7 Sized Part-Machine Matrix

In this part, a manufacturing area is studied with 5 machines and 7 parts. The
solutions for all alternative routes are analyzed. Several examples are shown below
for different cost component level combinations. A sensitivity analyse is done by
these combinations. It can be seen that The Kaiser’s Rule divided the manufacturing
area to two cells for each example. Different inputs will form different solutions.

First, we can analyze the table below, which shows results for 8 alternative part-
machine matrix of our algorithm with 5 machines and 7 parts.

Numerical results are shown in Table 4.2. We have 8 alternative part-machine

matrices, according to the alternative routes. Each coloumn shows the alternative
part-machine matrix results. Maximum number of iterations specified as 200 for each
alternative matrix which means the the algorithm runs for 200x8=1600 times. Each
iteration uses the result of previous one, as an input. In other words, algorithm
improves results by each iteration. (Also when algorithm chooses the worse result.)
“Minimum objective function value” means the minimum value within the each 200
iterations for each alternative route. We told that each iteration used the result of
previous one, as an input. And also we told that Simulated Annealing Algorithm
sometimes chose the worse result, not to stuck into local minimum. Within this
coloumn, the minimum value is shown that is found in successive iterations. For
example, the minimum objective value that is achieved within 200 iterations is 8675
by the 2nd route. Cell number is the number which is found by the Kaiser’s Rule
with the inputs given below. Kaiser’s Rule finds the cell number at the beginning of
the algorithm, and all iterations for the same route base on this number. In table 4.2,
it can be seen that the optimal cell number for the inputs given below, is 2. And
“Number of inter-material handling” shows the inter-cellular movements of the
manufacturing system within that cell number. For example, algorithm gave 5 times
material handling in the system at the end of 200 iterations which costed 210 for the
2nd route. But with same cell number, in the 1st route, algorithm gave 0 material
handling in the system at the end of 200 iterations which costed 0. This means,

41
algorithm formed a system with 2 cells for all alternative routes but cells had
different combinations of machines and parts because of the different routes.

The input datas for this problem are explained below.

The part-machine incidence matrix is (rows are machines);

Xp=

[ ]

Alternative machine matrix, which forms the alternative routes, is shown as Xa.
Rows are parts and coloumns are machines.

Xa=

[ ]

Production times of parts at each machine are shown in matrix Z. Rows are
machines and coloumns are parts.

Z=

[ ]

The temperature of Simulated Annaeling Algorithm probability equation is shown

as;
Te=

The demands of parts are shown as;

Dp=[10 15 5 20 35 15 30]

The operating costs of machines per time unit are;

42
B=

[ ]

The batch size of a part is;

Bs=5;

The inter-cell material handling costs for each part are shown as;
G=[10 15 20 10 15 15 10];

Costs of worker training skips between levels 1-2 and 3 are;

W=[1000 600];

Workers’ talent levels are;

T=[ ]

Part-machine-cell matrix as an initial solution in the problem;

PMC= [ ]

Machines’ complexity levels are;

M=[2 3 1 3 2];

Our algorithm found the minimum value as 8675 cost unit with 5 times carriage
and no training in overall iterations and routes. The rest routes have different
combinations of training and inter-material handling and costs of them.

We can analyze some of runs to be able to analyze the whole system.

The details of first route are shown in table 4.3. We can see that algorithm divided
the system to 2 cells and complexity level is 3 for both cells. 1200 unit cost spent to
make workers suitable for cells. This is the training cost. It can be seen that system

43
doesn’t need to any inter-material handlings. And manufacturing cost has the main
portion which costed 8375.

Table 4.2 Computational results of 5x7 part-machine matrix

Alternative
matrix 1 2 3 4 5 6 7 8 min average
Minimum
Objective 9575 8675 8840 8990 9920 10010 8785 8935 8675 9216,25
function value
Cost of inter-
material 0 210 150 210 0 0 150 210 210 116,25
handling
Cost of
8375 8465 8690 8780 8320 8410 8635 8725 8465 8550
manufacturing
Cost of
1200 0 0 0 1600 1600 0 0 0 550
training
Number of
2 0 0 0 2 2 0 0 0 0,75
training
Number of
2 2 2 2 2 2 2 2 2 2
cell
Number of
inter-material 0 5 4 5 0 0 4 5 5 2,875
handling

The part-machine-cell matrix of the 1st route is shown below as PMCi.

PMCi = [ ]

A=[ ]

Table 4.3 Components of the 1st route

minimum objective complexity level of complexity
function value cell 1 level of cell 2

9575 3 3

8375 0 1200
material handling
manufacturing cost training cost
cost

It can be seen that manufacturing area has two cells in Table 4.2. It can be also
seen in matrix A (cell number matrix). This matrix shows the cell numbers that parts

44
and machines belong to. These two matrices show that the parts 2-4-6 and machine
1-2 are in cell 1, parts 1-3-5-7 and machines 3-4-5 are in cell 2.

The details of 2nd route are shown in table 4.4. This is the minimum costed route.
We can see that material handling cost is 210 with 5 times inter-cell movement. And
0 unit training cost occurs. Manufacturing cost is different from first route, because
different parts use different machines (alternative machines-routes). Complexity
level is 2-3 for cells.

Table 4.4 Components of the 2nd route

minimum objective complexity level of complexity
function value cell 1 level of cell 2

8675 3 2

8465 210 0
material handling
manufacturing cost cost training cost

The part-machine-cell matrix of the 2nd route is shown below as PMCi;

PMCi = [ ]

A= [ ]

We used Kaiser’s Rule in our algorithm because we wanted to see how suitable
our system was for a cell formation problem. When we run our algorithm without the
Kaiser’s Rule and with the manually specified cell number (as 3 cells), our algorithm
found the minimum value as 9310 cost unit in overall iterations. We can see the
difference from results in Table 4.2. This also means that more inter-material
handling occurs. This is a worse result than our result with Kaiser’s Rule. This shows
the Kaiser’s Rule’s improvement effect on our problem.

45
4.1.1 Analysis Of Inter-Material Handling Cost

We can analyze one more example to check the sensitivity of our system with 5
machines and 7 parts. We change some of inputs to see how our system reacts. Inter-
material handling cost is multiplied by 4 in this example.

The input datas for Table 4.5 are same as datas for Table 4.2. The only difference
is;

G=4x[10 15 20 10 15 15 10];

Table 4.5 Computational results of 5x7 part-machine matrix

Alternative
matrix 1 2 3 4 5 6 7 8 min average
Minimum
Objective 9575 9305 9290 9380 9920 10010 9235 9565 9235 9535
function value
Cost of inter-
material 0 840 600 0 0 0 600 840 600 360
handling
Cost of
8375 8465 8690 8780 8320 8410 8635 8725 8635 8550
manufacturing
Cost of
1200 0 0 600 1600 1600 0 0 0 625
training
Number of
2 0 0 1 2 2 0 0 0 0,875
training
Number of
2 2 2 2 2 2 2 2 2 2
cell
Number of
inter-material 0 5 4 0 0 0 4 5 4 2,25
handling

It can be seen in the Table 4.5 that, algorithm gave no material handling in the
system at the end of 200 iterations for the first route. But with same cell number, in
the second route, algorithm gave 5 times inter-material handlings in the system at the
end of 200 iterations which costed 840. These two combinations are same as in Table
4.2. When we redound the cost of inter-material handling, algorithm enforces the
alternative routes of the system to make less inter-material handling and more
training according to that decreasement. We can see that the fourth route changed its
combination by the decision of having training instead of carriage. Because the one
time training could do the job of 5 times carriage which would cost 4 times higher

46
than the value in Table 4.2. The minimum value of whole system is came up with
seventh route, which was second route in Table 4.2. We can see that the algorithm
changed its decision by change of carriage costs.

We can analyze one more example to check the sensitivity of our system with 5
machines and 7 parts. Inter-material handling cost is multiplied by 8 in the next
example.

The input datas for Table 4.6 are same as datas for Table 4.2. The only difference is;

G=8x[10 15 20 10 15 15 10];

Table 4.6 Computational results of 5x7 part-machine matrix

Alternative
matrix 1 2 3 4 5 6 7 8 min average
Minimum
Objective 9575 9665 9690 9380 9920 10010 9635 9725 9380 9700
function value
Cost of inter-
material 0 0 0 0 0 0 0 0 0 0
handling
Cost of
8375 8465 8690 8780 8320 8410 8635 8725 8780 8550
manufacturing
Cost of
1200 1200 1000 600 1600 1600 1000 1000 600 1150
training
Number of
2 2 1 1 2 2 1 1 1 1,5
training
Number of
2 2 2 2 2 2 2 2 2 2
cell
Number of
inter-material 0 0 0 0 0 0 0 0 0 0
handling

We can see in Table 4.6, that the cost combination of 2nd,3th,7th and 8th routes
are changed when they are compared with values in Table 4.5. When we redound the
cost of inter-material handling more, algorithm enforces the system to make less
inter-material handling once again. We can see that the number of inter-material
handling of all routes in Table 4.6 is 0 (zero). This decreasement caused an
increasement on number of training for these routes in the whole system, because at
the end of comparison of training and increased inter-material handling costs, the

47
algorithm chose the least costed trainings. Because system should use either inter-
material handlings or worker trainings to compose cells. For example in eighth route,
algorithm chose the 1000 unit cost of one time training instead of 840x2 unit cost of
5 times carriage. If inter-material handlings are decreased, then training has more
importance and responsibility. Also the minimum value came up with fourth route
with 0 (zero) inter-material handling, in this example. We can say that the system is
non-sensitive to inter-material handling cost increasement from now on, because the
system eliminated all inter-material handlings in the system.

4.1.2 Analysis 0f Training Cost

We can analyze one more example to check the sensitivity of our system with 5
machines and 7 parts. It is said that we changed some of inputs to see how our
system reacts. In this example Inter-material handling cost is same as first example,
training cost is multiplied by 2.

The input datas for Table 4.7 are same as datas for Table 4.2. The only difference
is;

W=2x[1000 600];

We can see in Table 4.7 that the cost combination of first route is changed when it
is compared with values in Table 4.2. Algorithm chose the 1250 unit cost of 9 times
carriage instead of 2000 unit cost of one time training for this route. In Table 4.2, 6
times training is done in the whole system, but it is four in this example. When we
redound the cost of training, algorithm enforces the routes of the system to make less
training. Because of that algorithm makes the number of inter-material handling
higher in some alternative routes. But some routes (5th and 6th) accept to endure the
training cost instead of having carriage. Training cost and inter-material handling
cost are inversely proportional cost componenets. It is seen that the second route has
minimum value in Table 4.2, it is again second route that has minimum value in
Table 4.7. Because system already made training cost of some alternative routes 0

48
(zero). Because of small size of the system, some changes will not show any effects
on system.

Table 4.7 Computational results of 5x7 part-machine matrix

Alternative
matrix 1 2 3 4 5 6 7 8 min average
Minimum
Objective
function value 9625 8675 8840 8990 11520 11610 8785 8935 8675 9747,5
Cost of inter-
material
handling 1250 210 150 210 0 0 150 210 210 272,5
Cost of
manufacturing 8375 8465 8690 8780 8320 8410 8635 8725 8465 8550
Cost of
training 0 0 0 0 3200 3200 0 0 0 800
Number of
training 0 0 0 0 2 2 0 0 0 0,5
Number of
cell 2 2 2 2 2 2 2 2 2 2
Number of
inter-material
handling 9 5 4 5 0 0 4 5 5 4

We can analyze one more example to check the sensitivity of our system with 5
machines and 7 parts. In this example Inter-material handling cost is same as first
example, training cost is multiplied by 3.

The input datas for Table 4.8 are same as datas for Table 4.2. The only difference
is;

W=3x[1000 600];

It can be seen in Table 4.8 that, that the cost combination of 5th route is changed
when it is compared with values in Table 4.7. In Table 4.7, 4 trainings are done in the
whole system. But it is turned to 3 in Table 4.8. Because the algorithm chose the
1050 unit cost of 8 times carriage instead of 1800 unit cost of one time training for
the 5th route. We can say when we redound the cost of training, algorithm enforces
the system to make less training. As in Table 4.7, algorithm makes the number of
inter-material handling higher again in some routes(5th). But the alternative route,
which has the minimum value, is not changed. The changement of unit costs, could

49
not change the decision but changed the whole system’s structure. We can say that
the system is sensitive to cost changes of training but the second route is dominant
and makes the system to act as nonsensitive. The effect of training cost changements
in the manufacturing system are not clearly seen, because of the system’s small size.
Also the reason is, system could make the training costs 0 (zero) at the beginning and
without any changes on costs. The system goes on with same choises for higher
training costs than 3xW. That means system gets nonsensitive for higher costs than
3xW.

Table 4.8 Computational results of 5x7 part-machine matrix

Alternative
matrix 1 2 3 4 5 6 7 8 min average
Minimum
Objective
function value 9625 8675 8840 8990 12370 13210 8785 8935 8675 9966,25
Cost of inter-
material
handling 1250 210 150 210 1050 0 150 210 210 441,25
Cost of
manufacturing 8375 8465 8690 8780 8320 8410 8635 8725 8465 8550
Cost of
training 0 0 0 0 3000 4800 0 0 0 975
Number of
training 0 0 0 0 1 2 0 0 0 0,375
Number of cell 2 2 2 2 2 2 2 2 2 2
Number of
inter-material
handling 9 5 4 5 8 0 4 5 5 5,25

The part-machine-cell matrix of Table 4.8, 8th route is shown in table 4.9. Cell 1
has 3 machines, cell 2 has 2 machines. Machines 1-3-4 are in cell 1, machines 2-5 are
in cell 2. This configuration has 5 intercell movements between cell 1 and cell 2 as
shown in table 4.8.

Our problem that is based on Simulated Annealing Meta-heuristic has a 5x7 sized
part machine matrix. Two cells have two workers (one for each), who are trained as
hard as the most complex machine in that cells. An example with higher capacity
should be analyzed to see the effects. It is analyzed a problem with 10x15 sized part
machine matrix in the next part.

50
Table 4.9 Part-machine-cell matrix

cell 1 cell 2
machines 1 3 4 2 5 0 0

4.2 Computational Results For 10x15 Sized Part-Machine Matrix

In this part, a manufacturing area is studied with 10 machines and 15 parts.

Several examples are shown below for different cost component level combinations.
A sensitivity analyse is done again by these combinations as previous part.

First, we can analyze the table below, which shows results for 8 alternative part-
machine matrix of our algorithm with 10 machines and 15 parts.

Numerical results are shown in Table 4.10. To be able to analyze the system we
can check these results. For example, the minimum objective value that is achieved
within 200 iterations is 40580 by the 3rd route. Cell number is 3, which is found by
the Kaiser’s Rule with the inputs given below. Algorithm gave 14 times material
handling in the system at the end of 200 iterations which costed 675 for the 3rd route.
However with same cell number, in the 1st route, algorithm gave 28 material
handlings in the system at the end of 200 iterations which costed 1365. This means,
algorithm formed a system with 3 cells for all alternative routes but cells had
different combinationsof machines and partsbecause of the different route existance.

The input datas for this problem are explained below.

Alternative machine matrix, which forms the alternative routes, is shown as Xa.
Rows are parts and coloumns are machines.

Xa= [0 0;0 0;2 6;0 0;0 0;2 5;0 0;0 0;1 8;0 0;0 0;0 0;0 0;0 0;0 0]

The part-machine incidence matrix is (rows are machines);

51
Xp=

0 1 1 0 0 1 1 1 1 0 1 0 1 0 1
1 0 1 0 1 1 0 0 1 0 1 1 0 1 1
1 1 0 1 0 1 0 0 0 1 1 0 0 0 1
0 0 0 1 1 0 0 1 1 0 0 1 1 1 1
1 0 1 0 1 0 1 0 0 1 0 0 0 1 0
1 1 0 0 1 1 1 1 0 1 1 0 0 0 1
1 0 0 1 0 1 0 1 0 0 1 0 0 0 0
1 0 1 0 1 0 1 0 0 1 0 1 1 0 1
1 1 0 1 0 1 0 0 1 0 1 0 1 0 0
0 0 1 1 1 1 0 1 1 0 1 0 1 0 1

Production times of parts at each machine are shown in matrix Z. Rows are
machines and coloumns are parts.
Z=

1 3 2 3 1 4 3 7 2 2 4 1 1 3 2
3 5 5 1 4 3 2 3 6 3 4 1 3 2 3
7 2 2 4 1 1 3 5 1 1 6 3 4 7 2
1 4 2 3 6 3 4 5 1 4 3 2 3 6 3
8 5 1 1 4 6 3 2 2 4 1 9 3 5 1
3 4 1 3 2 3 4 1 1 3 5 1 1 6 4
3 2 3 6 3 4 1 3 2 3 3 7 1 9 7
5 1 1 4 6 8 5 1 4 1 9 3 5 5 4
4 2 3 6 3 4 5 1 4 3 2 2 4 1 1
3 5 5 8 5 1 4 1 6 3 4 5 1 2 2

The temperature of Simulated Annaeling Algorithm probability equation is shown

as;
Te= ;

The demands of parts are shown as;

Dp=[10 15 5 20 35 15 30 20 15 10 30 20 15 10 15] ;

The operating costs of machines per time unit are;

B= [8; 5; 9; 7; 6; 7; 9; 10; 4; 11] ;

The batch size of a part is

Bs=5;

52
 The inter-cell material handling costs for each part are shown as;
G=[10 15 20 10 15 15 10 10 15 20 15 20 10 25 5];

Costs of worker training skips berween levels 1-2 and 3 are;

W=[1000 600];

Workers’ talent levels are;

T= [1 2 3 1 2 2 1 3 3 2]

Part-machine-cell matrix as an initial solution in the problem;

PMC= 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 2 3 4 5 6 7 8 9 10 0 0 0 0 0

Machines’ complexity levels are;

M=[2 3 3 3 2 2 3 2 3 3]

As an example, the details of sixth route are shown in table 4.11. We can see that
complexity levels are 2-2-3 for cells. 1000 unit cost spent to make workers suitable
for cells. This is the training cost. It can be seen that system needs to 15 inter-
material handlings. And manufacturing has the main portion which costed 39860.

The part-machine-cell matrix of the result that is shown in Table 4.11 is shown
after the table. The matrix A, shows the cell numbers that parts and machines belong
to. Matrix A and PMCi show that the parts 3-6-9-12-15 and machine 5-8 are in cell 1,
parts 2-5-8-11-14 and machines 1-2-3-6-9 are in cell 2, the parts 1-4-7-10-13 and
machine 4-7 are in cell 3.

53
Table 4.10 Computational results of 10x15 part-machine matrix

Alternative
matrix
1 2 3 4 5 6 7 8 min average
Minimum
Objective
function
value

40955 41675 40580 42290 41440 41585 41880 40840 40580 41405,625
handling
material
Cost of
inter-

1365 725 675 1025 940 725 1065 665 675 898,125
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39905 39882,5
training
Cost of

0 1000 0 1000 1000 1000 1000 0 0 625

Number of
training

0 1 0 1 1 1 1 0 0 0,625
Number of
cell

3 3 3 3 3 3 3 3 3 3
Number of

handling
material
inter-

28 15 14 21 20 15 22 14 14 18,625

Table 4.11 Components of the 1st route

minimum objective complexity level of complexity complexity
function value cell 1 level of cell 2 level of cell 3

41585 2 2 3

39860 725 1000

material handling
manufacturing cost training cost
cost

54
PMCi = 3 6 9 12 15 2 5 8 11 14 1 4 7 10 13
5 8 0 0 0 0 1 2 3 6 9 10 4 7 0

A= 3 2 1 3 2 1 3 2 1 3 2 1 3 2 1
2 2 2 3 1 2 3 1 2 2 3 1 1 1 1

The details of third route are shown in table 4.12. This is the minimum costed
route. We can see that material handling cost is 675 with 14 times inter-cell
movement. And 0 unit training cost occurs. Manufacturing cost is different from first
route, because different parts use different machines (alternative machines).
Complexity level is 3-2-2 for cells.

Table 4.12 Components of the 3rd route

minimum objective complexity level of complexity complexity
function value cell 1 level of cell 2 level of cell 3

40580 3 2 2

39905 675 0
material handling
manufacturing cost training cost
cost

The part-machine-cell matrix of the result that is shown in Table 4.12 is shown
below;

PMCi = 3 6 9 12 15 2 5 8 11 14 1 4 7 10 13
0 0 0 10 1 2 4 5 8 9 0 3 6 7 0

A= 3 2 1 3 2 1 3 2 1 3 2 1 3 2 1
2 2 3 2 2 3 3 2 2 1 3 1 1 2 1

55
 These two matrices show that the parts 3-6-9-12-15 and machine 10 are in cell 1,
parts 2-5-8-11-14 and machines 1-2-4-5-8-9 are in cell 2, the parts 1-4-7-10-13 and
machine 3-6-7 are in cell 3.

We used Kaiser’s Rule in our algorithm because we wanted to see how beneficial
our system was for a cell formation problem. When we run our algorithm not with
the Kaiser’s Rule but with the manually specified cell number as 2 and 4 cells, our
algorithm found the minimum value as 41225 cost unit for the system with 2 cells,
and 41040 cost unit for the system with 4 cells in overall iterations. We can see the
difference from results in Table 4.10.

4.2.1 Analysis Of Inter-Material Handling Cost

We can analyze one more example to check the sensitivity of our system with 10
machines and 15 parts. We change some of inputs to see how our system reacts.
Inter-material handling cost is multiplied by 4 in this example.

The input datas for Table 4.13 are same as datas for Table 4.10. The only
difference is;

G=4x[10 15 20 10 15 15 10 10 15 20 15 20 10 25 5];

It can be seen in the Table 4.13 that, 1st, 2nd, 4th, 5th, 6th, 7th routes changed
their combinations of training and inter-material handling by the decision of having
training instead of carriage. Because, for example in first route the 3 times training
could provide the job cheaper with 14 times carriage instead of just 28 times carriage
which would cost 4 times higher than the value in Table 4.10. When we redound the
cost of inter-material handling, algorithm enforces the system to make less inter-
material handling and more training according to that decreasement. However the
third and eighth routes show the same combinations as Table 4.10. And algorithm
didn’t change the selection of alternative route. The minimum value of whole system
is came up with third route again.

56
Table 4.13 Computational results of 10x15 part-machine matrix

Alternative
matrix
1 2 3 4 5 6 7 8 min average
Minimum
Objective
function
value

44370 42850 42605 44685 43920 42760 44915 42835 42605 43617,5
handling
material
Cost of
inter-

2580 700 2700 2220 2220 700 2900 2660 2700 2085
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39905 39882,5
training
Cost of

2200 2200 0 2200 2200 2200 2200 0 0 1650

Number of
training

3 3 0 3 3 3 3 0 0 2,25
Number of
cell

3 3 3 3 3 3 3 3 3 3
Number of

handling
material
inter-

14 7 14 12 12 7 15 14 14 11,875

Inter-material handling cost is multiplied by 8 in the next example, to see the

effect better.

The input datas for Table 4.14 are same as datas for Table 4.10. The only
difference is;

G=8x[10 15 20 10 15 15 10 10 15 20 15 20 10 25 5];

57
Table 4.14 Computational results of 10x15 part-machine matrix

Alternative
matrix
1 2 3 4 5 6 7 8 min average
Minimum
Objective
function
value

46310 43550 44715 46905 46140 43460 47815 45275 43460 45521,25
handling
material
Cost of
inter-

3920 1400 4700 4440 4440 800 5800 3500 800 35625
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39860 39882,5
training
Cost of

2800 2200 600 2200 2200 2800 2200 1600 2800 2075
Number of
training

4 3 1 3 3 4 3 2 3 2,875
Number of
cell

3 3 3 3 3 3 3 3 3 3
Number of

handling
material
inter-

12 7 10 12 12 5 15 11 5 10,5

We can see in Table 4.14, some routes changed their combinations of training and
inter-material handling again by the decision of having training instead of carriage.
When we redound the cost of inter-material handling more, algorithm enforces the
system to make less inter-material handling once again. For example in first route the
1 more time training could provide the job cheaper with 12 times carriage instead of
3 times training and 14 times carriage which would cost 4 times higher carriage than
the value in Table 4.10. But in the 6th route, 7 times carriage and 3 times training

58
costs are equal to 5 times carriage and 4 times training cost, and algorithm chose to
add one more training to the system.

Also the minimum value came up with sixth route, in this example. We can say
that the system is sensitive to inter-material handling cost increasement, but we
should analyse one more increasement to see the nonsensitive point.

It is analyzed if the inter-material handling cost is multiplied by 12 in the next

example.

The input datas for Table 4.15 are same as datas for Table 4.10. The only
difference is;

G=12x[10 15 20 10 15 15 10 10 15 20 15 20 10 25 5];

We can see in Table 4.15, that the cost combination of 7th route is changed when
all values are compared with values in Table 4.14. When we redound the cost of
inter-material handling more, algorithm enforces the system to make less inter-
material handling once again. This decreasement caused an increasement on number
of training for these routes in the whole system, because at the end of comparison of
training and increased inter-material handling costs, the algorithm chose the least
costed trainings. Because system should use either inter-material handlings or worker
trainings to compose cells. For example in 7th route, algorithm chose the 1000 unit
cost of one more time training and 13 times carriage instead of 5800x2 unit cost of
15 times carriage and 3 times training. If inter-material handlings are decreased, then
training has more importance and responsibility. Also the minimum value came up
with 6th route again, in this example. We can say that the system is sensitive to inter-
material handling cost increasement till now but nonsensitive from now on, because
the algorithm gives same results and makes same choise with higher values than
12xG.

59
Table 4.15 Computational results of 10x15 part-machine matrix

Alternative
matrix

1 2 3 4 5 6 7 8 min average
function value
Minimum
Objective

48270 44250 47555 49125 48360 43860 50275 47025 43860 47340
Cost of inter-

handling
material

5880 2100 7050 6660 6660 1200 7260 5250 1200 5257,5
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39860 39882,5
training
Cost of

2800 2200 600 2200 2200 2800 3200 1600 2800 2200
Number of
training

4 3 1 3 3 4 4 2 4 3
Number
of cell

3 3 3 3 3 3 3 3 3 3
inter-material
Number of

handling

12 7 10 12 12 5 13 11 5 10,25

4.2.2 Analysis Of Training Cost

We can analyze one more example to check the sensitivity of our system with 10
machines and 15 parts. In this example Inter-material handling cost is same as first
example, training cost is multiplied by 2.

60
 The input datas for Table 4.16 are same as datas for Table 4.10. The only
difference is;

W=2x[1000 600];

Table 4.16 Computational results of 10x15 part-machine matrix

Alternative
matrix

1 2 3 4 5 6 7 8 min average
function value
Minimum
Objective

40955 42675 40580 43290 42440 42585 42880 40840 40580 42030,6
Cost of inter-

handling
material

1365 725 675 1025 940 725 1065 665 675 898,125
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39905 39882,5
training
Cost of

0 2000 0 2000 2000 2000 2000 0 0 1250

Number of
training

0 1 0 1 1 1 1 0 0 0,625
Number
of cell

3 3 3 3 3 3 3 3 3 3
inter-material
Number of

handling

28 15 14 21 20 15 22 14 14 18,625

We can see in Table 4.16 that the cost combinations have no changes according to
Table 4.10. Algorithm gives the same results as in Table 4.10. Also the minimum

61
value came up with 3th route again, in this example. Because system already gives
solution with no training in the first example.

Table 4.17 Computational results of 10x15 part-machine matrix

Alternative
matrix

1 2 3 4 5 6 7 8 min Average
function value
Minimum
Objective

40955 42850 40580 44290 43440 43585 43880 40840 40580 421552,5
Cost of inter-

handling
material

1365 2900 675 1025 940 725 1065 665 675 1170
manufacturing
Cost of

39590 39950 39905 40265 39500 39860 39815 40175 39905 39882,5
training
Cost of

0 0 0 3000 3000 3000 3000 0 0 1500

Number of
training

0 0 0 1 1 1 1 0 0 0,5
Number
of cell

3 3 3 3 3 3 3 3 3 3
inter-material
Number of

handling

28 37 14 21 20 15 22 14 14 21,375

The results for;

W=3x[1000 600] is shown in Table 4.17.

62
 Algorithm gives same minimum objective function value (3th route) also for the
example which training cost is multiplied by 3, which is shown in Table 4.17. Just
for the 2nd route, the number of inter-material handlings are increased and the
combination of system is changed. Instead of one time training with 3000 unit cost
and 15 times inter-material handling, algorithm chose not to make training and to
increase the inter-materail handling number to 37 with 2900 unit cost. The rest of
routes have same combinations. We can say that the system is sensitive to cost
changes of training but the 3th route is dominant and makes the system to act as
nonsensitive. The same outcome is received by higher unit training costs. That means
system gets nonsensitive for higher costs than 3xW.

The part-machine-cell matrix of minimum objective value in Table 4.16 and 4.17,
the 3th route, is shown in table 4.18. Cell 1 has 2 machines, cell 2 has 5 machines
and cell 3 has 3 machines. This configuration has 14 intercell movements between
cell 1, cell 2 and cell 3 as shown in table 4.16 and 4.17.

Table 4.18 Part-machine-cell matrix

cell 1 cell 2 cell 3

machines 9 10 1 2 5 6 8 3 4 7

An example with higher capacity should be analyzed to see the effects. It is

analyzed a problem with 18x30 sized part machine matrix in the next part.

4.3 Computational Results For 18x30 Sized Part-Machine Matrix

Now, we can analyze the table below, which shows results for 8 alternative part-
machine matrix of our algorithm with 18 machines and 30 parts.

Numerical results are shown in Table 4.19. It can be seen that the optimal cell
number for the inputs given, is 4. Algorithm gave 58 material handlings in the
system at the end of 200 iterations which costed 4629 in the sixth route (the
minimum value within all routes). The overall objective function cost of sixth route
is 66609 unit cost.

63
 The input datas for Table 4.19 are explained below.

The part-machine incidence matrix is (rows are machines);

Xp=

0 1 1 0 1 0 1 1 0 1 1 1 1 0 1 1 1 0 0 0 0 1 0 1 1 0 1 1 1 0
1 0 1 1 1 1 0 0 0 1 1 0 0 0 1 1 0 1 1 1 1 0 1 1 1 0 0 0 1 1
1 1 1 0 0 0 0 1 0 1 1 0 1 1 0 0 0 0 1 0 1 0 1 1 0 1 0 1 0 1
0 0 1 1 0 0 0 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 0 1 1 0 1 1 1 0
0 0 0 0 1 0 1 0 1 1 0 1 0 1 0 0 0 1 1 0 0 0 0 1 1 1 1 0 0 0
1 0 1 1 1 1 1 0 1 1 0 1 1 0 0 0 1 1 0 0 0 0 1 1 1 0 0 0 1 1
1 0 1 1 0 1 1 0 0 0 0 1 0 1 1 0 0 1 0 0 0 0 0 0 0 1 0 1 1 0
1 0 1 1 1 1 0 0 1 0 1 1 0 1 0 1 0 0 0 0 1 1 0 0 0 0 1 0 1 1
0 0 1 1 0 1 1 1 0 0 1 0 1 1 0 1 1 0 0 0 0 0 1 1 0 0 0 1 0 1
0 0 0 1 0 0 0 1 1 0 1 1 1 1 1 0 1 0 1 1 1 1 0 1 1 0 1 1 0 0
0 1 1 0 1 1 0 0 0 1 0 1 1 1 1 0 0 1 0 1 1 0 0 0 1 1 0 1 0 0
1 1 0 0 1 1 0 0 0 1 1 0 1 1 0 0 0 1 1 0 0 0 1 0 1 0 1 1 0 1
1 1 0 1 1 0 1 1 0 0 0 0 1 1 1 1 0 0 0 0 1 0 1 1 0 0 0 0 1 1
0 1 0 1 0 1 1 0 1 0 1 0 0 1 1 0 1 0 1 0 0 0 0 1 0 1 1 1 1 0
0 0 0 1 0 1 0 1 1 0 1 0 1 1 1 0 1 0 1 1 1 1 0 0 1 0 1 1 0 0
0 0 1 1 1 1 1 0 0 1 1 1 0 1 0 1 0 0 1 1 1 0 1 1 0 1 0 0 0 1
1 0 1 1 0 0 0 0 1 0 1 1 1 1 1 0 1 0 1 0 0 1 0 1 0 0 0 1 1 0
0 0 1 0 1 1 0 1 1 1 1 1 0 1 0 1 1 0 1 0 1 0 1 1 0 0 1 0 0 0

Alternative machine matrix, which forms the alternative routes, is shown as Xa.
Rows are parts and coloumns are machines.

Xa=

[ ]

The demands of parts are shown as;

Dp=[10 15 5 20 10 15 30 20 10 5 25 15 15 35 25 20 30 10 15 10 5 40 30 40 25 25 5 15 35 15]

Production times of parts at each machine are shown in matrix Z. Rows are
machines and coloumns are parts.

64
Z=

1 3 2 3 1 4 3 7 2 2 4 1 1 3 3 6 3 4 1 4 3 2 3 5 5 1 2 3 6 3
3 5 5 1 4 3 2 4 2 3 6 3 4 4 1 1 3 3 6 3 4 5 5 1 2 2 4 4 2 1
7 2 2 4 1 1 3 5 1 1 4 2 3 5 5 1 4 3 2 4 2 3 2 4 1 1 3 3 4 4
1 4 2 3 6 3 4 3 5 5 1 4 3 3 4 3 8 5 1 1 7 9 2 4 1 6 4 2 5 7
8 5 1 1 4 2 3 4 2 7 5 1 1 8 3 6 1 1 5 6 6 4 6 8 2 6 4 3 8 9
3 5 1 1 1 1 8 3 4 3 3 4 3 4 1 1 3 5 1 1 4 2 5 1 1 7 9 2 5 1
3 5 1 1 3 5 5 1 4 3 2 4 2 3 2 4 1 6 3 4 3 5 5 1 4 5 6 6 4 6
4 1 1 3 5 1 3 6 3 4 3 5 5 1 4 3 3 4 3 8 3 4 2 7 5 1 1 8 3 5
3 2 4 2 1 1 4 2 3 5 5 1 4 1 1 5 6 6 4 6 5 5 1 4 5 6 6 3 4 3
5 5 1 4 3 2 4 2 1 4 3 3 4 3 8 1 1 4 2 3 4 2 7 5 1 1 8 3 6 1
7 2 2 4 1 1 3 2 7 5 1 1 8 3 6 1 5 1 4 3 2 4 2 3 2 4 1 1 4 2
1 1 4 2 3 5 4 6 5 5 1 4 5 6 6 3 4 3 5 1 3 6 3 4 3 5 5 4 3 5
7 2 2 4 1 1 3 4 3 4 1 1 3 5 4 2 7 5 1 2 1 4 3 3 1 1 5 6 6 4
3 2 4 2 3 2 1 4 2 3 4 2 7 5 1 1 6 5 5 8 3 4 3 3 4 3 4 1 1 3
9 2 4 4 2 3 4 2 7 5 5 5 1 4 5 6 6 4 6 6 3 4 3 5 5 1 4 3 5 1
6 3 2 3 2 4 1 6 3 3 5 5 1 4 1 1 5 6 6 1 1 5 6 6 4 6 8 3 6 1
6 3 4 1 4 3 2 5 1 4 1 4 5 6 6 3 4 3 6 6 3 4 3 5 1 3 6 1 4 1
4 3 7 2 2 4 1 4 3 8 1 1 4 2 3 4 2 4 3 3 4 3 8 1 1 4 2 4 3 3

The temperature of Simulated Annaeling Algorithm probability equation is shown

as;
Te=108

The operating costs of machines per time unit are;

B=[ 1; 2; 5; 5; 2; 4; 2; 5; 3; 4; 2; 3; 2; 5; 4; 2; 3; 2]

The batch size of a part is;

Bs=5

The inter-cell material handling costs for each part are shown as;
G=[10 15 30 20 30 15 25 15 30 15 14 23 16 15 17 29 22 12 23 14 18 23 25 30 10 35 27 32 13 24]

Costs of worker training skips between levels 1-2 and 3 are;

W=[1000 600]

Workers’ talent levels are;

T=[1 3 2 1 2 2 1 3 3]

Part-machine-cell matrix as an initial solution in the problem;

65
PMC=

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 0 0

21 22 23 24 25 26 27 28 29 30
0 0 0 0 0 0 0 0 0 0

Machines’ complexity levels are;

M=[2 3 1 3 2 3 2 2 3 3 2 3 1 3 2 3 3 2]

Table 4.19 Computational results of 18x30 part-machine matrix

Alternative
matrix

1 2 3 4 5 6 7 8 min average
function value
Minimum
Objective

67311 68638 67575 66978 68248 66609 67358 67990 66609 67588,4
Cost of inter-

handling
material

5661 4828 5985 4828 4828 4629 5798 6070 4629 5328,38
manufacturing
Cost of

59650 60010 59590 59950 59620 59980 59560 59920 59980 59785
training
Cost of

2000 3800 2000 2200 3800 2000 2000 2000 2000 2475
of training
Number

2 5 2 3 5 2 2 2 2 2,875
Number
of cell

4 4 4 4 4 4 4 4 4 4
inter-material
Number of

handling

74 60 76 72 60 58 72 75 58 68,375

66
 We can analyze some of runs to be able to analyze the whole system.

The details of sixth route are shown in table 4.20. This is the minimum costed
route. We can see that algorithm divided the system to 4 cells and complexity levels
are 1-2-2-3.

Table 4.20 Components of the 6th route

minimum
objective complexity level complexity complexity complexity
function value of cell 1 level of cell 2 level of cell 3 level of cell 4

66609 1 2 2 3

59980 4629 2000

manufacturing material
cost handling cost training cost

The part-machine-cell matrix of the result that is shown in Table 4.20 is shown
below;

A=
4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1
3 2 1 4 4 2 2 4 4 3 4 3 4 2 3 4 4 3 1 4

4 3 2 1 4 3 2 1 4 3
2 3 1 3 3 4 3 3 3 2

PMCi=

4 8 12 16 20 24 28 3 7 11 15 19 23 27 2 6 10 14 18 22
0 0 2 6 7 14 0 0 1 10 12 15 18 0 0 0 0 0 0 3

26 30 1 5 9 13 17 21 25 29
4 5 8 9 11 13 16 17 0 0

67
 When we ran our algorithm without Kaiser’s Rule and specified the cell number
manual as 3, the minimum objective value that our algorithm found is 70214. That is
worse than the result shown in Table 4.19. When the cell number is set to 5, the
minimum objective value is 68119. That is also worse than the result shown in Table
4.19. It was seen that the inter-material handling cost has decreased but the training
cost has increased with 3 cells. Lower cell number causes less inter-material handling
but more training to have workers which can operate more machines. The difference
shows the Kaiser’s Rule’s improvement effect on our problem.

4.3.1 Analysis Of Inter-Material Handling Cost

We can analyze one more example to check the sensitivity of our system with 18
machines and 30 parts. We change some of inputs to see how our system reacts.
Inter-material handling cost is multiplied by 4 in this example.

The input datas for Table 4.21 are same as datas for Table 4.19. The only
difference is;

G=4x[10 15 30 20 30 15 25 15 30 15 14 23 16 15 17 29 22 12 23 14 18 23 25 30 10 35 27
32 13 24];

It can be seen in Table 4.21 that, algorithm gave 58 material handlings in the
system at the end of 200 iterations which costed 18516 for the 6th route, which gave
the minimum objective function value. When we multiply the cost of inter-material
handling by four, algorithm enforces the system to make less inter-material handling.
For example in the 1st route, algorithm chose to make 3 times more training and 56
times inter-material handling instead of 74 times inter-material handling. Because the
second desicion costed lower. So the algorithm changed combinations of 1st, 3rd and
8th routes according to cost comparisons. However the route is not changed. The
minimum value came up with 6th route again, in this example. The 6th route has 58
times carriage in Table 4.21 as in Table 4.19.

68
Table 4.21 Computational results of 18x30 part-machine matrix

Alternative
matrix
1 2 3 4 5 6 7 8 min average
function value
Minimum
Objective

82402 83122 82102 81462 82732 80496 84752 81432 80496 82366,5
Cost of inter-

handling
material

18952 19312 19312 19312 19312 18516 23192 19312 18516 19481,5
manufacturing
Cost of

59650 60010 59590 59950 59620 59980 59560 59920 59980 59785
training
Cost of

3800 3800 3200 2200 3800 2000 2000 2200 2000 2875
Number of
training

5 5 4 3 5 2 2 3 2 3,625
Number
of cell

4 4 4 4 4 4 4 4 4 4
inter-material
Number of

handling

56 60 60 72 60 58 72 60 58 62,25

We can analyze one more example to check the sensitivity of our system with 18
machines and 30 parts. Inter-material handling cost is multiplied by 8 in this
example.

The input datas for Table 4.22 are same as datas for Table 4.19. The only
difference is;

69
G=8x[10 15 30 20 30 15 25 15 30 15 14 23 16 15 17 29 22 12 23 14 18 23 25 30 10 35 27
32 13 24];

Table 4.22 Computational results of 18x30 part-machine matrix

Alternative
matrix

1 2 3 4 5 6 7 8 min average
function value
Minimum
Objective

101354 98418 101414 100374 98196 98892 96608 101152 96608 99551
Cost of inter-

handling
material

37904 34008 38624 37224 34176 35112 33848 38432 12693 36166
manufacturing
Cost of

59650 60010 59590 59950 59620 59980 59560 59920 59560 59785
training
Cost of

3800 4400 3200 3200 4400 3800 3200 2800 3200 3600
Number of
training

5 6 4 4 6 5 4 4 4 4,75
Number
of cell

4 4 4 4 4 4 4 4 4 4
inter-material
Number of

handling

56 53 60 59 53 53 54 59 54 55,875

In table 4.22, it can be seen that, the optimal cell number for the inputs given
above, is not changed. Because the part-machine incidence matrix is same. We
should check the other datas. We can see that the algorithm changed the

70
combinations of 2nd, 4th, 5th,7th and 8th routes. And the minimum value in overall
routes and iterations come up with 7th route with 96608 unit cost. Algorithm decided
to increase training number according to the unit cost increasement in inter-material
handling. For example, algorithm gave 54 material handlings in the system at the end
of 200 iterations which costed 33848 for the minimum costed route (7th route). But it
was 72 material handlings in the system at the end of 200 iterations in Table 4.21.
The total cost increased by the increasement of unit cost for inter-material handling.
But it can be seen that the number of inter-material handling is decreased. This is the
main point that we want. Algorithm enforces the system to make less inter-material
handling. So, the problem is sensitive to inter-material handling cost.

It is analyzed if the inter-material handling cost is multiplied by 12 in the next

example.

The input datas for Table 4.23 are same as datas for Table 4.19. The only
difference is;

G=12x[10 15 30 20 30 15 25 15 30 15 14 23 16 15 17 29 22 12 23 14 18 23 25 30 10 35 27 32 13
24];

We can see in Table 4.23, that the cost combination of 3rd route is changed when
all values are compared with values in Table 4.22. When we redound the cost of
inter-material handling more, algorithm enforces the system to make less inter-
material handling once again. Because system should use either inter-material
handlings or worker trainings to compose cells. For example in 3rd route, algorithm
chose the 600 unit cost of one more time training and 54 times carriage instead of
57936 unit cost of 60 times carriage and 4 times training. If inter-material handlings
are decreased, then training has more importance and responsibility. However the
minimum value came up with 2nd route, in this example. It is seen that the
differences got smaller when we multiply inter-material handling cost by 12.
Different unit costs cause different route choises and changes the numbers of training
and inter-material handling traffics. But we can say that the system is nonsensitive

71
 Number of Cost of inter- Minimum
Number Number of Cost of Cost of Alternative
inter-material material Objective
of cell training training manufacturing matrix
handling handling function value

1
56 4 5 3800 59650 56856 120306

2
53 4 6 4400 60010 48924 113334
3
higher values than 12xG.

54 4 5 3800 59590 51228 114618

4

4.3.2 Analysis Of Training Cost

59 4 4 3200 59950 57396 120546
5

53 4 6 4400 59620 51264 115284

72
6

53 4 5 3800 59980 52668 116448

Table 4.23 Computational results of 18x30 part-machine matrix

54 4 4 3200 59560 50772 113532

8

59 4 4 2800 59920 57648 120368

min

50 4 6 4400 60010 48924 113334

average

55,125 4 4,875 3675 59785 53344,5 116805

training cost with 18 machines and 30 parts. It is said that we changed some of inputs
from now on, because the algorithm gives same results and makes same choise with

We can analyze one more example to check the sensitivity of our system for
 Number of Cost of Minimum
inter- Number of Number of Cost of Cost of inter- Objective Alternative
material cell training training manufacturing material function matrix
handling handling value
difference is;

1
74 4 2 4000 59650 5661 69311

W=2x[1000 600];

2
81 4 2 4000 60010 7401 71411

3
87 4 1 1200 59590 7898 68688

4
72 4 3 4400 59950 4828 69178

73
5

73 4 2 4000 59620 6953 70573

6
Table 4.24 Computational results of 18x30 part-machine matrix

86 4 1 2000 59980 6537 68517

first example (Table 4.19), training cost is multiplied by 2.

85 4 1 1200 59560 7514 68274

8

75 4 2 4000 59920 6070 69990

85 4 1 1200 59560 7514 68274

79,125 4 1,75 3100 59785 6607,75 69492,8

min average
to see how our system reacts. In this example Inter-material handling cost is same as

The input datas for Table 4.24 are same as datas for Table 4.19. The only
 We can see in Table 4.24 that the cost combination of 2nd, 3th, 5th, 6th and 7th
routes are changed when it is compared with values in Table 4.19. Algorithm chose
the 7th route as minimum value with 7514 unit cost of 85 times carriage and 1 time
training instead of 72 times carriage and 2 times of training for this route. This is the
minimum objective function value in overall routes and iterations for this example (it
was the 6th route in the first example). When we redound the cost of training,
algorithm enforces the routes of the system to make less training. Because of that
algorithm makes the number of inter-material handling higher in some alternative
routes. But some routes (1st, 4th and 8th) accept to endure the training cost instead of
having carriage. Training cost and inter-material handling cost are inversely
proportional cost components.

We can analyze one more example to check the sensitivity of our system with 18
machines and 30 parts. Training cost is multiplied by 3 in this example.

The input datas for Table 4.25 are same as datas for Table 4.19. The only
difference is;

W=3x[1000 600];

In this example, algorithm changed the combination of the 3th, 4th and 8th routes.
It is seen that the algorithm choose third route as minimum objective function
solution. It was the seventh route in Table 4.24. In the 3th route, algorithm decided to
make 95 times inter-material handling and not to make any training instead of 87
times inter-material handling and 1 time training.

According to these explanations, it is seen that our algorithm with 18x30 sized
matrix is sensitive to training unit cost changes. But for further trials, the algorithm
has same solutions and same choise. This shows that the algorithm is nonsensitive on
increasements, which are more than 3 times, for training.

74
Table 4.25 Computational results of 18x30 part-machine matrix

Alternative
matrix
1 2 3 4 5 6 7 8 min Average
Minimum
Objective
function

71052,6
71311

73411

68695

72835

72573

69517

68874

71205

68695
value
handling

7292,63
material
Cost of
inter-

5661

7401

9105

6885

6953

6537

7514

8285

9105
manufacturing
Cost of

59650

60010

59590

59950

59620

59980

59560

59920

59590

59785
training
Cost of

6000

6000

6000

6000

3000

1800

3000

3975
0

0
Number of
training

1,375
2

0
Number of
cell

4
Number of

handling
material

83,375
inter-

74

81

95

87

73

86

85

86

95

The part-machine-cell matrix of Table 4.25, 3rd route is shown in table 4.26. Cell
1 has 2 machines, cell 2 has 3 machines, cell 3 has 4 machines, cell 4 has 9
machines. This configuration has 95 intercell movements between these four cells as
shown in table 4.25.

Table 4.26 Part-machine-cell matrix

cell 1 cell2 cell3 cell4

machines 12 18 3 9 10 6 14 16 17 1 2 4 5 7 8 11 13 15

75
 17 numerical examples are processed by the proposed SA algorithm. 5 of them are
constructed in small scaled size, 6 of them are constructed in medium scaled size, 6
of them are constructed in large scaled size. When values are getting higher, the
difference will be much more higher. The running time by the proposed simulated
annealing algorithm is longer for the large scaled sizes.

76
 CHAPTER FIVE
CONCLUSION

In the literature review, it was not seen that much studies about cell formation
with human issue. In real life manufacturing systems, it can not be thought the
system without workers and effects of them. A study was done in this paper about
that topic.

In this study, a cell formation was done for single period and with alternative
routes of parts. It was tried to determine cells by the cost calculation of demand for
parts, training for workers and inter-cellular material handling for parts again. Also
the basis is similarty coefficients. Jaccard’s similarity coefficient formulation was
used.

The objective function was about minimizing the cost of the system overall, using
Matlab R2008a program. Kaiser’s Rule was used to find the optimal cell number and
to form cells according to that number. To check the efficiency of Kaiser’s Rule in
this problem, a pre-determined numbers as cell numbers were given to the system. It
was seen that we received better results (lower cost values) when we used The
Kaiser’s Rule for giving the cell number as initial number. Numerical examples
showed that Kaiser’s Rule made the algorithm faster. Because the cell number that
Kaiser’s Rule gave, gave a lower cost value at the beginning. This means, Kaiser’s
Rule provided a better assignment of machines and parts.

Simulated Annealing Algorithm (SAA) was used to construct the system. To

determine cells and their part families, SAA minimized the sum of system cost for
the all parts. SAA was chosen, because it was seen that this heuristic could be simply
performed. Also SA algorithm gives good results among other heuristics in general.

Cell formation was done with talented workers. We could train them if needed.
Also system let to choose different routes for manufacturing. These routes were

77
predetermined for the problem, like demand, manufacturing times, manufacturing
costs, trainig costs, material handling costs.

The neighbourhood solution was found by mutation. This also allowed to the
system, to change the number of machines in each cell and see the different
combinations for number of cell members.

Different combinations were used for cost components in numerical examples to

see the effects on results. The training cost and inter-material handling cost are used
to see the effects. Because these two components affects the decision variables in the
problem. It can be seen that the increased cost forced the system to decrease action
scores of that cost. But until a cost level. This level is different for each problem.
When the cost combinations are higher than that level, increasement does not affect
the result mainly by scores, but increases the total cost.

Multi-period cell formation will be studied for future researches. Multi-period is

very important for the real world systems. A new cell formation is not efficient for a
plant for each demand period. So a cell formation that considers the whole demand
periods should be planned.

Also the layout planning and change of place costs for machines should be added
for future studies. Because the layout is the part of cell formation problem and have a
big importance on real life systems. Changing the place of a machine means extra-
cost for the manufacturing system. Configuration should be done according to that
issue.

78
REFERENCES

Aryanezhad, M. B., Deljoo, V., Mirzapour Al-e-hashem, & S. M. J., (2009).

Dynamic cell formation and the worker assignment problem: a new model. The
International Journal of Advanced Manufacturing Technology, 41, 329–342.

Askin, R. G. & Huang, Y., (2001). Forming effective worker teams for cellular
manufacturing. International Journal of Production Research, 39(11), 2431–
2451.

Azizi, N., Zolfaghari, S., & Liang, M., (2010). Modeling job rotation in
manufacturing systems: The study of employee’s boredom and skill variations.
Int. J. Production Economics, 123, 69–85.

Balakrishnan, J. & Cheng, C. H., (2007). Multi-period planning and uncertainty

issues in cellular manufacturing: A review and future directions. European
Journal of Operational Research, 177, 281-309.

Balasubramanian, K. N. & Panneerselvam, R., (1993). Covering technique-based

algorithm for machine grouping to form manufacturing cells. International
Journal of Production Research, 31, 1479–1504.

Bashir, H. A. & Karaa, S., (2008). Assessment of clustering tendency for the design
of cellular manufacturing systems. Journal of Manufacturing Technology
Management, 19 (8), 1004 – 1014.

Bhadury, J. & Radovilsky, Z., (2006). Job rotation using the multi-period assignment
model. International Journal of Production Research, 44, 4431–4444.

Bidanda, B., Ariyawongrat, P., LaScola Needy, K., Norman, B. A., &
Tharmmaphornphilas, W., (2005). Human related issues in manufacturing cell

79
 design, implementation, and operation: a review and survey. Computers and
Industrial Engineering, 48, 507–523.

Black, J. T.,(1991). The design of the factory with a future. New York: McGraw-Hill.

Cesani, V.I. & Steudel, H.J., (2005). A study of labor assignment flexibility in
cellular manufacturing systems. Computers and Industrial Engineering, 48, 571–
591.

Cheng, C. H., Gupta, Y.P., Lee, W.H., & Wong, K.F., (1998). A TSP-based heuristic
for forming machine groups and part families. International Journal of
Production Research, 36(5), 1325–37.

Choobineh, F., (1988). A framework for the design of cellular manufacturing

systems. International Journal of Production Research, 26, 1161- 1172.

Corominas, A., Pastor, R., & Rodriguez, E., (2006). Rotational allocation of tasks to
multifunctional workers in a service industry. Int. J. Production Economics, 103,
3–9.

Dawis, D.J. & Mabert, V.A., (2000). Order dispatching and labor assignment in
cellular manufacturing systems. Decision Sciences, 31(4; ABI/INFORM), 745

Fitzpatrick, E. L. & Askin, R. G., (2005). Forming effective worker teams with
multi-functional skill requirements. Computers and Industrial Engineering, 48,
593–608.

Defersha, F. M. & Chen, M., (2006). A comprehensive mathematical model for the
design of cellular manufacturing systems. Int. J. Production Economics, 103,
767–783.

80
Guerrero, F., Lozano, S., Smith, K. A., Canca, D., & Kwok, T., (2002).
Manufacturing cell formation using a new self-organizing neural network.
Computers and Industrial Engineering, 42, 377–382.

Gupta, T., (1993). Design of manufacturing cells for flexible environment

considering alternative routing. International Journal of Production Research, 31,
1259–1273.

Ham, I., Hitomi, K., & Yoshida, T., (1985). Group technology application to
production management. Nijhoff, EN: Kluwer.

Hwang, H. & Ree, P., (1996). Routes selection for the cell formation problem with
alternative part process plans. Computers and Industrial Engineering, 30, 423–
431.

Kirkpatrick, S., Gelatt, C. D., & Vecchi, M. P., (1983). Optimization by simulated
annealing. Science, 220, 671–680.

Mahdavi, I., Aalaei, A., Paydar, M. M., & Solimanpur, M., (2010). Designing a
mathematical model for dynamic cellular manufacturing systems considering
production planning and worker assignment. Computers and Mathematics with
Applications, 60, 1014-1025.

Mosier, C. T. & Taube, L., (1985). The facets of group technology and their impact
on implementation. OMEGA, 13, 381-391.

Norman, B. et al, (2002). Worker assignment in cellular manufacturing considering

technical and human skills. International Journal of Production Research, 40
(6),1479–1492.

81
Pailla, A., Trindade, A. R., Parada, V., & Ochi, L. S., (2010). A numerical
comparison between simulated annealing and evolutionary approaches to the cell
formation problem. Expert Systems with Applications, 37, 5476–5483.

Pinedo, M., (2004). Planning and scheduling in manufacturing and services.

Springer, New York, 424-431.

Plaquin, M. & Pierreval, H., (2000). Cell formation using evolutionary algorithms
with certain constraints. Int. J. Production Economics, 64(1-3), 267–278.

Saeedi, S., Solimanpur, M., Mahdavi, I., & Javadian, N., (2010). Heuristic
approaches for cell formation in cellular manufacturing. J. Software Engineering
and Applications, 3, 674-682.

Safaei, N., Saidi-Mehrabad, M., & Jabal-Ameli, M. S., (2008). A hybrid simulated
annealing for solving an extended model of dynamic cellular manufacturing
system. European Journal of Operational Research, 185, 563–592.

Seifoddini, H., (1988). Recent developments in production research. Elsevier

Science Publishers B.V.. The Netherlands, 562–570.

Selvam, R. P. & Balasubramanian, K. N., (1985). Algorithmic grouping of operation

sequences. Engineering Cost and Production Economics, 9, 125–134.

Shunk, D., (1985). Group technology provides organized approaches to realizing

benefits of CIMS. Industrial Engineering, 17, 74-81.

Slomp, J., Bokhorst, J. A. C., & Molleman, E., (2005). Cross-training in a cellular
manufacturing environment. Computers and Industrial Engineering, 48, 609–624.

82
Stevens, M. J. & Campion, M. A., (1994). The knowledge, skill, and ability
requirements for teamwork: implications for human resource management.
Journal of Management, 20(2), 503-530.

Suer, G. & Bera, I., (1998). Optimal operator assignment and cell loading when lot-
splitting is allowed. Com. and Ind. Eng., 35(3– 4),431–434.

Sütçü, A., Tanrıtanır, E., Durmuşoğlu, B., & Koruca, H. I., (2011). An integrated
methodology for layout design and work organisation in a furniture manufacturing
plant. South African Journal of Industrial Engineering, 22(1), 183-197.

Tam, K. Y., (1990). An operation sequence based similarity coefficient for part
families formations. Journal of Manufacturing Systems, 9, 55–68.

Tavakkoli-Moghaddam, R., Rahimi-Vahed, A. R., Ghodratnama, A., & Siadat, A.,

(2009). A simulated annealing method for solving a new mathematical model of a
multi-criteria cell formation problem with capital constraints. Advances in
Engineering Software, 40, 268–273.

Tavakkoli-Moghaddam, R., Javadian, N., Javadi, B., & Safaei, N., (2007). Design of
a facility layout problem in cellular manufacturing systems with stochastic
demands. Applied Mathematics and Computation, 184, 721–728.

Vakharia, A.J. & Wemmerlöv, U., (1990). Designing a cellular manufacturing

system: A materials flow approach based on operation sequences. IIE
Transactions, 22, 84–97.

Won, Y. K., (2000). New p-median approach to cell formation with alternative
process plans. International Journal of Production Research, 38, 229–240.

83
Won, Y. K. & Kim, S. H., (1997). Multiple criteria clustering algorithm for solving
the group technology problem with multiple process routings. Computers and
Industrial Engineering, 32, 207–220.

Wu, X., Chu, C. H., Wang, Y., & Yan, W., (2007). A genetic algorithm for cellular
manufacturing design and layout. European Journal of Operational Research,
181, 156-167.

Wu, T. H., Chang, C. C., & Chung, S. H., (2008). A simulated annealing algorithm
for manufacturing cell formation problems. Expert Syst. Appl., 34(3), 1609–1617.

Yin, Y. & Yasuda, K., (2006). Similarity coefficient methods applied to the cell
formation problem: a taxonomy and review. Int. J. Production Economics, 101,
329-352.

84
 APPENDIX

A1.Cms Matlab Codes

clear;
nn=2;
np=3;

Xp= ;

[ ]

nnn=0;
Xa=[0 0;0 0;2 4;0 0;0 0;2 5;1 3];
a=find(Xa(:,1)>0);
aa=length(a);

for x1=1:nn
for x2=1:nn
for x3=1:nn
Xp(Xa(a(1),:),a(1))=0;
Xp(Xa(a(2),:),a(2))=0;
Xp(Xa(a(3),:),a(3))=0;
Xp(Xa(a(1),x1),a(1))=1; %alternative machine matrix
Xp(Xa(a(2),x2),a(2))=1;
% finds parts which have alternative routes
Xp(Xa(a(3),x3),a(3))=1;

nnn=nnn+1;
BN=10000000000; %a big number
ii=1;
jj=7;
kk=5;
LL=2;
iterationmax=200;
temperature=100000000;

obj=BN*(ones(1,iterationmax));
TT=zeros(1,iterationmax);

Dp=[10 15 5 20 35 15 30]; %demands of parts%

Z= ; %production times of parts, rows are machines%

[ ]

B=[ 8; 5; 9; 7; 6]; %operating cost of machines per time

unit coloumns are machines%

85
bs=5; %batch size%

G=[10 15 20 10 15 15 10];

W=[1000 600]; %training costs by levels%

T=fix(1+(3*rand(1,18))); %talent levels of workers%

PMC=[1 2 3 4 5 6 7;1 2 3 4 5 0 0]; %Initial part-machine-cell matrix

M=[2 3 1 3 2 3 2 2 3 3 2 3 1 3 2 3 3 2]; %complexity level of machines

WC=[1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18]; %worker matrix%

obj=0;

Am=Xp; %Kaiser’s Rule optimal cell

number
calculations%
[u,v]=size(Am);
Xm=zeros (u);
Ym=zeros (u);
Zm=zeros (u);
for i=1:u
for j=1:u
for am=1:v
if Am(i,am)==Am(j,am) && Am(i,am)==1;
Xm(i,j)=Xm(i,j)+1;
Ym(i,j)=Ym(i,j)+Am(i,am);
Zm(i,j)=Zm(i,j)+Am(j,am);
else
Ym(i,j)=Ym(i,j)+Am(i,am);
Zm(i,j)=Zm(i,j)+Am(j,am);
end
S(i,j)=Xm(i,j)/(Ym(i,j)+Zm(i,j)-Xm(i,j));
end
end
end
C=eig(S);
th(nnn)=hucresayisi; %optimal cell number%
hucresayisi=0;

if jj>kk;
PMCi=zeros(2,jj);
A=zeros(2,jj);
hhh=jj;
else
PMCi=zeros(2,kk);
A=zeros(2,kk);
jj=kk;
hhh=kk;
end

for j=1:2;
huc=(th(nnn));
for i=1:jj;

86
 if huc>=1;
A(j,i)=huc;
huc=huc-1;
else
huc=(th(nnn));
A(j,i)=huc;
huc=huc-1;
end
end
end

Ayedek=A;

for iteration=1:iterationmax
egitimadett=0;
a1=1;
b1=1;
d=0;
a0=zeros(1,max(A));
b0=zeros(1,max(A));
c0=zeros(1,max(A));
alt=Xp;
for p=1:(th(nnn));
A=A';
A=reshape(A,1,hhh+hhh);
h=find(A==p);
a0(p)=length(find(h<(hhh+1)));
b0(p)=length(find(h>hhh));
c0(p)=length(h);
i1=1;
for i=a1:(a0(p)+a1-1)
for k=i1:i1
if h(k)<=hhh
PMCi(1,i)=PMC(1,h(k));
end
end
i1=i1+1;
end
a1=a1+a0(p);
b1=b1+b0(p);
for j=(a0(p)+1):c0(p)
d=d+1;
if d<=hhh;
PMCi(2,d)=PMC(2,(h(j)-hhh));
end
end
end

ZX=(Z.*Xp);

DZX=Dp*ZX'; %Total time on machines in the

period%

DZXB=DZX*B;

%Inter-Cellular Mat. Handling calculations%

87
l=zeros(1,jj);
li=zeros(1,jj);
for j=1:jj;
U=find(Xp(:,j)==1);
U=U';
m=length(U);
for p=1:(th(nnn));

tas=zeros(1,p);
h=find(A==p);
Xpt=PMC(2,((h(find(h>hhh)))-hhh));
for i=1:m;
if length(find(U(i)==Xpt))>0;
l(j)=1+l(j);
else
l(j)=0+l(j);
end
tas(p)=tas(p)+l(j);
end
if l(j)>0;
li(j)=li(j)+1;
end
end
tasima(j)=li(j)-1;
end

%Training calculations%
Wtotal=0;
b=1;
egitimadett=0;

for p=1:(th(nnn));
h=find(A==p);
a0=zeros(1,(max(A)+1));
a0(p+1)=length(find(h<(hhh+1)));
x=0;
b=a0(p)+b;

for i=b:(a0(p+1)+b-1);
if PMCi(2,i)~=0;
if M(PMCi(2,i))>x;
x=M(PMCi(2,i));
end
end
end

if (T(WC(p)))<x; %if a machine has higher

complexity than
for i=(T(WC(p))):(x-1); worker’s talent level, then
training is done%
Wtotal=Wtotal+W(i);
egitimadett=egitimadett+1;
end
end
end

%Objective function%

88
obj(iteration)=(DZXB)+(sum(tasima.*G.*(Dp/bs)))+(Wtotal);
A=reshape(A,hhh,2);
A=A';

if obj(iteration)<BN;
Ayedek=A;
BN=obj(iteration);
if iteration>1;
TT(iteration)=TT(iteration-1);
end
if obj(iteration)<=min(obj);
egitimadet(nnn)=egitimadett;
alt=Xp;
Wmin=Wtotal;
Tasmin(nnn)=sum(tasima);
Ctasima=(sum(tasima.*G.*(Dp/bs)));
Curetim=DZXB;
Cegitim=Wtotal;
Egitimsayisi=sum(egitimadet);

end

for j=1:hhh;
A(2,j)=fix((th(nnn)*rand)+1);
end
sonuc=BN;

else
temperature=temperature*(0.95);
TT(iteration)=temperature;
if exp((BN-obj(iteration))/temperature)>(2*rand(1));
BN=obj(iteration);
Ayedek=A;
for j=1:hhh;
A(2,j)=fix((th(nnn)*rand)+1);
end

else
for j=1:hhh;
A=Ayedek;
A(2,j)=fix((th(nnn)*rand)+1);

end
end
end

end
objmin(nnn)=min(obj);
Ctasimamin(nnn)=Ctasima;
Curetimmin(nnn)=Curetim;
Cegitimmin(nnn)=Cegitim;
end
end
end

89
90