Improving Manufacturing Systems Using Integrated Discrete Event Simulation and Evolutionary Algorithms Parminder Singh Kang A Thesis Submitted in Partial Fulfilment of the Requirement of De Montfort University for the Degree of Doctor of Philosophy May 2012 De Montfort University
212
Embed
Improving Manufacturing Systems Using Integrated Discrete Event Simulation and Evolutionary
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Improving Manufacturing Systems Using
Integrated Discrete Event Simulation
and Evolutionary Algorithms
Parminder Singh Kang
A Thesis Submitted in Partial Fulfilment of the Requirement of De Montfort
University for the Degree of Doctor of Philosophy
May 2012
De Montfort University
i
Abstract
High variety and low volume manufacturing environment always been a challenge for
organisations to maintain their overall performance especially because of the high level
of variability induced by ever changing customer demand, high product variety, cycle
times, routings and machine failures. All these factors consequences poor flow and
degrade the overall organisational performance. For most of the organisations,
therefore, process improvement has evidently become the core component for long term
survival.
The aim of this research here is to develop a methodology for automating operations in
process improvement as a part of lean creative problem solving process. To achieve the
stated aim, research here has investigated the job sequence and buffer management
problem in high variety/low volume manufacturing environment, where lead time and
total inventory holding cost are used as operational performance measures. The research
here has introduced a novel approach through integration of genetic algorithms based
multi-objective combinatorial optimisation and discrete event simulation modelling tool
to investigate the effect of variability in high variety/low volume manufacturing by
considering the effect of improvement of selected performance measures on each other.
ii
Also, proposed methodology works in an iterative manner and allows incorporating
changes in different levels of variability.
The proposed framework improves over exiting buffer management methodologies, for
instance, overcoming the failure modes of drum-buffer-rope system and bringing in the
aspect of automation. Also, integration of multi-objective combinatorial optimisation
with discrete event simulation allows problem solvers and decision makers to select the
solution according to the trade-off between selected performance measures.
iii
Acknowledgments
In my humble acknowledgement, I would like to convey my gratitude to all the people
who were with me directly or indirectly throughout this long journey.
First and foremost, I wish to thank god who has guided me throughout this journey as
being always with me as strength, determination and courage to pursue this work with
high level of confidence and commitment.
At the professional and academic level, I am really grateful to Dr Riham Khalil and Prof
Dave Stockton (my supervisors) to provide me this opportunity at first instance to work
on this research problem. Essentially, it was impossible to achieve this without their
precious encouragement, advice and guidance and endless support, who never accepted
less than my best effort. Thanks Riham and Dave for your endless guidance and support
in this journey, It is been a pleasure working with both of you.
Especial thanks to De Montfort University and Technology Strategy board to fund this
project (TSB K1532G, Accelerating process excellence using virtual discrete event
process simulation), which enabled me to peruse this research and to all project
collaborates for their valuable feedback.
At personal, I would like to show gratitude to my father and mother for their continuous
support and encouragement, and to my brother who’s endless support allowed me to
focus on my studies, thanks for being there as my elder brother. Words fail to express
my appreciation to my wife whose love and persistence confidence in me, has
encouraged me and always taken-off stress from my shoulders.
iv
I wish to express deep gratitude for all my family members in UK and India for their
love and support. Very special thanks to my uncle Baljit Singh for his invaluable
guidance and has always been a real inspiration to me.
It is a pleasure to thank my second and special family at the lean research group/centre
for manufacturing for their support, suggestions and care. Especially, thanks to
Lawrance Mukhongo for all the great time we spend together and always being there as
my elder brother.
Finally, I would like to thank everybody who was important to the successful realisation
of the thesis, as well as expressing my apology that I could not mention personally one
by one.
v
Declaration
I declare that the work described within this thesis was originally undertaken by me,
(Parminder Singh Kang) between the dates of registration for the degree of Doctor of
Philosophy at De Montfort University, July 2009 to May 2012.
vi
Abstract i
List of Tables xi
List of Figures xiv
Abbreviation and Glossary xvii
Research Dissemination xix
Chapter 1 – Introduction
1.1 Introduction 1
1.2 Need of Synchronous Flow 3
1.3 Lean Philosophy in Synchronous High Variety/Low Volume Manufacturing 4
1.4 Simulation and Combinatorial Optimisation 5
1.5 The Scope of Research 6
1.6 The Aim and Objective 7
1.7 The Structure of Thesis 8
Chapter 2 – Lean Creative Problem Solving and Process Improvement
2.1 Introduction 12
2.2 Brief History of Manufacturing Systems 12
2.3 Lean Philosophy 13
2.3.1 Lean’s Five Principals 15
2.3.2 Waste in Lean 18
vii
2.4 Manufacturing Problems 23
2.5 Lean Creative Problem Solving 26
2.5.1 Characteristics of Effective Problem Solving Process 26
2.5.2 Existing Problem Solving Methods 29
2.5.3 Process Improvement Using Lean Creative Problem Solving Process 34
2.6 Summary 36
Chapter 3 – Combinatorial Optimisation for Process Improvement
3.1 Introduction 37
3.2 Root Cause Analysis as Part of Process Improvement 38
3.2.1 Existing Root Cause Analysis Methods for Process Improvement 39
Step 6: to identify the bottleneck, correlation analysis has been performed on the
collected results according to the rules described in the step 6 of Section 4.4.
a. Average Queuing Time; Figure 5.1a exemplifies the degree of correlation
between total inventory holding cost and average queuing time for different
batch sizes.
Figure 5.1a (Total Inventory Holding Cost vs. Average Queuing Time)
Similarly, from Figure 5.1b demonstrates the degree of correlation between lead
time and average queuing time for different batch sizes.
Figure 5.1b (Lead Time vs. Average Queuing Time)
-1.00
-0.50
0.00
0.50
1.00
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Total Inventory Holding Cost vs. Average Queuing Time
Batch Size 1 Batch Size 5 Batch Size 10
-1.00
-0.50
0.00
0.50
1.00
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Lead Time vs. Average Queuing Time
Batch Size 1 Batch Size 5 Batch Size 10
98
b. Average Queue Size; Figure 5.2a, exemplifies the degree of correlation
between total inventory holding cost and average queue size for different batch
sizes.
Figure 5.2a (Total Inventory Holding Cost vs. Average Queue Size)
Similarly, Figure 5.2b, represents the degree of correlation between lead time
and average queue size for different batch sizes.
Figure 5.2b (Lead Time vs. Average Queue Size)
-1.00
-0.50
0.00
0.50
1.00
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Total Inventory Holding Cost vs. Average Queuing Size
Batch Size 1 Batch Size 5 Batch Size 10
-1.00
-0.50
0.00
0.50
1.00
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Lead Time vs. Average Queuing Size
Batch Size 1 Batch Size 5 Batch Size 10
99
c. % Working; Figure 5.3a, illustrates degree of correlation between total
inventory holding cost and % working for different number of parts.
Figure 5.3a (Total Inventory Holding Cost vs. %Working)
Similarly, Figure 5.3b, illustrates degree of correlation between lead time and %
working for different parts quantity.
Figure 5.3b (Lead Time vs. %Working)
-1.00
-0.50
0.00
0.50
1.00
M1 M2 M3 M4 M5
Total Inventory Holding Cost vs. %Working
500 Parts 1000 Parts 2000 Parts
-1
-0.5
0
0.5
1
M1 M2 M3 M4 M5
Lead Time vs. %Working
500 Parts 1000 Parts 2000 Parts
100
d. % Waiting: Figure 5.4a, illustrates degree of correlation between total
inventory holding cost and % waiting for different number of parts.
Figure 5.4a (Total Inventory Holding Cost vs. %Waiting)
Similarly, Figure 5.4b, illustrates degree of correlation between lead time and %
waiting for different number of parts.
Figure 5.4b (Lead Time vs. %Waiting)
-1.00
-0.50
0.00
0.50
1.00
M1 M2 M3 M4 M5
Total Inventory Holding Cost vs. % Waiting
500 Parts 1000 Parts 2000 Parts
-1
-0.5
0
0.5
1
M1 M2 M3 M4 M5
Lead Time vs. % Waiting
500 Parts 1000 Parts 2000 Parts
101
e. % Changeover; Figure 5.5a, illustrates degree of correlation between total
inventory holding cost and % changeover for different number of parts.
Figure 5.5a (Total Inventory Holding Cost vs. % Changeover)
Similarly, Figure 5.5b, illustrates degree of correlation between lead time and %
changeover for different number of parts.
Figure 5.5b (Lead Time vs. % Changeover)
-1.00
-0.50
0.00
0.50
1.00
M1 M2 M3 M4 M5
Total Inventory Holding Cost vs. % Changeover
500 Parts 1000 Parts 2000 Parts
-1
-0.5
0
0.5
1
M1 M2 M3 M4 M5
Lead Time vs. % Changeover
500 Parts 1000 Parts 2000 Parts
102
f. % Stopped; In current research % stopped refers to the long stoppages. There is
no direct relation between the % stopped and change in customer demand or
batch size. Machine failure is used as type of variability for data collection,
which drives % stopped in the final results. Along this, from Figure 5.6a and
Figure 5.6b, it is important to note that % stopped having weak positive
correlation between total inventory holding cost and lead time.
Figure 5.6a (Total Inventory Holding Cost vs. %Stopped)
Figure 5.6b (Lead Time vs. % Stopped)
-1.00
-0.50
0.00
0.50
1.00
M1 M2 M3 M4 M5
Total Inventory Holding Cost vs. % Stopped
500 Parts 1000 Parts 2000 Parts
-1
-0.5
0
0.5
1
M1 M2 M3 M4 M5
Lead Time vs. % Stopped
500 Parts 1000 Parts 2000 Parts
103
Step – 7: this segment describes the results after optimisation according to the variables
described in Table 4.6. Experiments have re-run again to collect the results after
optimisation, and are represented with respect to customer demand in terms of the total
number of parts. Along this, results from different runs are compared according to
different type of variability i.e. machine failure and batch sizes.
It is important to note that each combinatorial optimisation run represents two output
values. As elucidated in the Section 3.3.3.2, number of output values related to each
result is equal to the number of fitness functions, .i.e. from this research’s perspective
it’s lead time and total inventory holding cost.
Therefore, each experiment represents two dominant solutions, i.e. one with respect to
lead time and other with respect to total inventory holding cost and the selection of
solution from these dominant solutions is the choice of a decision-maker.
Note: it is important to note that this chapter only includes graphs for batch size 1.
Processed data for batch size 1, 5 and 10 is included in Appendix A1. Batch size 5 and
10 exhibits the similar trend as batch size 1.
NOTE: Figure 5.7 a – b, 5.8 a – b, 5.9 a – b, 5.10 a – b, 5.11 a – b and 5.12 a – b are
using logarithmic axis.
104
a. 500 Parts;
Table 5.4 (Lead Time and Total Inventory Holding Cost Before and After
Optimisation for 500 Parts)
Experim
ent N
o.
Batch
Size
Mach
ine F
ailure
optim
isation criteria
Before
Optimisation
Job Sequence
Optimisation
Buffer Size
Optimisation
Job Sequence and
Buffer Size
Optimisation
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
1.1
1
Yes
LT
20,489
2,032,863
8,001 620,692 8,545 6,061 8,035 54,135
1.2
TI H
C
8,008 597,043 8,682 5,520 8,342 5,273
2.1
No
LT
16,749
1,562,810
6,835 501,668 7,297 7,690 6,972 22,861
2.2
TIH
C
6,841 477,449 7,310 4,002 7,013 3,881
3.1
5
Yes
LT
10,742 1,311,448
8,001 792,895 8,431 25,590 8,018 22,004
3.2 TIH
C
8,008 724,090 8,640 21,417 8,357 20,800
4.1
No
LT
9,386 1,159,705
6,834 655,137 7,197 20,664 6,842 43,007
4.2
TIH
C
6,841 642,073 7,199 17,475 7,029 17,082
5.1
10
Yes
LT
9,287 1,084,242
8,001 861,840 8,537 81,340 8,001 116,457
5.2
TIH
C
8,008 795,537 8,545 41,491 8,125 40,521
6.1
No
LT
7,966 925,438
6,834 704,325 7,297 37,644 6,834 99,217
6.2
TIH
C
6,879 671,244 7,301 34,420 6,991 33,304
Table 5.4 illustrates the lead time and total inventory holding cost results collected for
500 jobs using different levels of variability. The results are presented according to the
optimisation criteria defined in Table 4.6.
105
I. 500 Parts without Machine Failure: Figure 5.7a – c compares the results based on
the identified performance measures before and after the job sequence, buffer size and
both job sequence and buffer size optimisation for 500 parts without machine failure.
Figure 5.7a and 5.7b exemplifies the reduction in average queuing time and queue size
respectively after applying the combinatorial optimisation.
Figure 5.7a (Average Queuing Time before and after Optimisation for 500 Parts
without Machine Failure)
Figure 5.7b (Average Queue Size before and after Optimisation for 500 Parts without
Machine Failure)
0.01
0.1
1
10
100
1000
10000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
20
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
106
Figure 5.7c illustrates the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.7c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 500 Parts without Machine
Failure)
0
10
20
30
40
50
60
70
80
90
100
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation% Changeover before Optimisation % Blocking before Optimisation% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation% Changeover after Job Sequence Optimisation % Blocking after Job Sequence Optimisation% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation% Changeover after Job Sequence and Buffer Size Optimisation % Blocking after Job Sequence and Buffer Size Optimisation
107
II. 500 Parts with Machine Failure: Similarly, Figure 5.8a – c compares the results
based on the identified performance measures before and after the job sequence, buffer
size and both job sequence and buffer size optimisation for 500 parts with machine
failure. Figure 5.8a and 5.8b exemplifies the reduction in average queuing time and
queue size respectively after applying the combinatorial optimisation.
Figure 5.8a (Average Queuing Time before and after Optimisation for 500 Parts with
Machine Failure)
Figure 5.8b (Average Queue Size before and after Optimisation for 500 Parts with
Machine Failure)
1
10
100
1000
10000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
10
100
1000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
108
Figure 5.8c shows the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.8c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 500 Parts with Machine
Failure)
0
10
20
30
40
50
60
70
80
90
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation
% Changeover before Optimisation % Blocking before Optimisation
% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation
% Changeover after Job Sequence Optimisation % Blocking after Job Sequence Optimisation
% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation
% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation
% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation
% Changeover after Job Sequence and Buffer Size Optimisation % Blocking after Job Sequence and Buffer Size Optimisation
109
b. 1000 Parts
Table 5.5 (Lead Time and Total Inventory Holding Cost Before and After
Optimisation for 1000 Parts)
Experim
ent N
o.
Batch
Size
Mach
ine F
ailure
optim
isation criteria
Before
Optimisation
Job Sequence
Optimisation
Buffer Size
Optimisation
Job Sequence
and Buffer Size
Optimisation
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
7.1
1
Yes
LT
29,744 5,849,512
15,899 2,439,180 16,574 37,135 16,115 46,474
7.2
TIH
C
15,903 2,109,480 16,994 10,942 16,751 10,873
8.1
No
LT
28,246 4,739,098
13,564 1,821,940 14,107 15,048 13,761 35,328
8.2
TIH
C
13,564 1,821,940 14,220 7,940 14,173 7,887
9.1
5
Yes
LT
20,912 4,530,910
15,903 2,774,240 16,452 75,733 16,136 68,694
9.2
TIH
C
16,136 2,765,210 16,919 43,462 16,597 42,980
10.1
No
LT
18,756 4,100,013
13,564 2,597,910 14,107 41,871 13,739 45,132
10.2
TIH
C
13,571 2,377,530 14,119 35,041 14,171 34,743
11.1
10
Yes
LT
18,898 4,290,147
15,899 3,416,140 16,694 101,108 16,006 196,339
11.2
TIH
C
15,903 2,879,160 16,705 83,823 16,576 84,922
12.1
No
LT
16,396 3,770,432
13,564 2,875,700 14,207 76,280 13,639 207,377
12.2
TIH
C
13,776 2,531,200 14,210 70,485 14,141 68,227
Similar to Table 5.4, Table 5.5 illustrates the results collected for 1000 jobs using
different levels of variability. The results are presented according to the optimisation
criteria defined in Table 4.6.
110
I. 1000 Parts without Machine Failure: Figure 5.9a – c compares the results based on
the identified performance measures before and after the job sequence, buffer size and
both job sequence and buffer size optimisation for 1000 parts without machine failure.
Figure 5.9a and 5.9b exemplifies the reduction in average queuing time and queue size
respectively after applying the combinatorial optimisation.
Figure 5.9a (Average Queuing Time before and after Optimisation for 1000 Parts
without Machine Failure)
Figure 5.9b (Average Queue Size before and after Optimisation for 1000 Parts
without Machine Failure)
0.01
0.1
1
10
100
1000
10000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
10
100
1000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
111
Figure 5.9c shows the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.9c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 1000 Parts without Machine
Failure)
0
10
20
30
40
50
60
70
80
90
100
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation
% Changeover before Optimisation % Blocking before Optimisation
% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation
% Changeover after Job Sequence Optimisation % Changeover after Job Sequence Optimisation
% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation
% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation
% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation
% Changeover after Job Sequence and Buffer Size Optimisation % Blocking after Job Sequence and Buffer Size Optimisation
112
II. 1000 Parts with Machine Failure: Here Figure 5.10a – c compares the results
based on the identified performance measures before and after the job sequence, buffer
size and both job sequence and buffer size optimisation for 1000 parts with machine
failure. Figure 5.10a and 5.10b exemplifies the reduction in average queuing time and
queue size respectively after applying the combinatorial optimisation.
Figure 5.10a (Average Queuing Time before and after Optimisation for 1000 Parts
with Machine Failure)
Figure 5.10b (Average Queue Size before and after Optimisation for 1000 Parts with
Machine Failure)
1
10
100
1000
10000
100000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
10
100
1000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
113
Figure 5.10c shows the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.10 c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 1000 Parts with Machine
Failure)
0
10
20
30
40
50
60
70
80
90
100
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation
% Changeover before Optimisation % Blocking before Optimisation
% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation
% Changeover after Job Sequence Optimisation % Blocking after Job Sequence Optimisation
% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation
% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation
% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation
% Changeover after Job Sequence and Buffer Size Optimisation % Blocking after Job Sequence and Buffer Size Optimisation
114
c. 2000 Parts
Table 5.6 (Lead Time and Total Inventory Holding Cost Before and After
optimisation for 2000 Parts)
Experim
ent N
o.
Batch
Size
Mach
ine F
ailure
optim
isation criteria
Before
Optimisation
Job Sequence
Optimisation
Buffer Size
Optimisation
Job Sequence
and Buffer Size
Optimisation
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
Lead
Tim
e
Total In
ven
tory
Hold
ing C
ost
13.1
1
Yes
LT
85,304 33,980,772
31,831 9,621,250 32,565 43,552 32,345 259,211
13.2
TIH
C
31,832 8,711,330 33,995 21,884 33,640 215,97.3
14.1
No
LT
66,167 25,839,806
27,094 7,125,940 28,037 29,519 27,446 160,039
14.2
TIH
C
27,094 7,125,940 28,047 15,109 27,998 15,045
15.1
5 Y
es
LT
41,348 20,455,456
31,838 11,268,000 32,724 14,987 324,84.3 131,383
15.2
TIH
C
31,838 11,268,000 32,929 83,794 33,090 84,049
16.1
No
LT
34,195 15,542,509
27,094 11,013,200 28,037 81,832 27,577 123,432
16.2
TIH
C
27,101 9,230,970 28,046 68,167 27,885 67,161
17.1
10
Yes
LT
37,446 17,800,888
31,831 13,234,000 33,001 164,191 32,343 266,318
17.2
TIH
C
31,838 11,663,100 33,001 164,191 32,725 161,613
18.1
No
LT
32,491 15,542,509
27,094 10,697,400 28,038 135,634 27,488 355,010
18.2
TIH
C
27,101 9,678,570 28,037 148,640 27,967 132,773
Finally, Table 5.6 illustrates the results collected for 2000 jobs using different levels of
variability. The results are presented according to the optimisation criteria defined in
Table 4.6.
115
I. 2000 Parts without Machine Failure: Figure 5.11a – c exemplifies the results based
on the identified performance measures before and after the job sequence, buffer size
and both job sequence and buffer size optimisation for 2000 parts without machine
failure. Figure 5.11a and 5.11b exemplifies the reduction in average queuing time and
queue size respectively after applying the combinatorial optimisation.
Figure 5.11a (Average Queuing Time before and after Optimisation for 2000 Parts
without Machine Failure)
Figure 5.11b (Average Queue Size before and after Optimisation for 2000 Parts
without Machine Failure)
0.01
0.1
1
10
100
1000
10000
100000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
10
100
1000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
116
Figure 5.11c shows the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.11c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 1000 Parts without Machine
Failure)
0
10
20
30
40
50
60
70
80
90
100
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation
% Changeover before Optimisation % Blocking before Optimisation
% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation
% Changeover after Job Sequence Optimisation % Blocking after Job Sequence Optimisation
% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation
% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation
% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation
% Changeover after Job Sequence and Buffer Size Optimisation % Blocked after Job Sequence and Buffer Size Optimisation
117
II. 2000 Parts with Machine Failure: Figure 5.12a – c exeplifies the results based on
the identified performance measures before and after the job sequence, buffer size and
both job sequence and buffer size optimisation for 2000 parts with machine failure.
Figure 5.12a and 5.12b exemplifies the reduction in average queuing time and queue
size respectively after applying the combinatorial optimisation.
Figure 5.12a (Average Queuing Time before and after Optimisation for 2000 Parts
with Machine Failure)
Figure 5.12b (Average Queue Size before and after Optimisation for 2000 Parts with
Machine Failure)
0.01
0.1
1
10
100
1000
10000
100000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queuing Time before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
1
10
100
1000
Queue for M1 Queue for M2 Queue for M3 Queue for M4 Queue for M5
Average Queue Size before and After Optimisation
Before Optimisation After Job Sequence Optimisation
After Buffer Size Optimisation After Job Sequence and Buffer Size Optimisation
118
Figure 5.12c shows the results for % working, % waiting, %changeover and % blocking before and after optimisation.
Figure 5.12c (% Working, % Waiting, % Changeover and % Blocking before and after Optimisation for 2000 Parts with Machine
Failure)
0
10
20
30
40
50
60
70
80
90
M1 M2 M3 M4 M5
% Working, % Waiting, % Changeover and % Blocking before and after Optimisation
% Waiting before Optimisation % Working before Optimisation
% Changeover before Optimisation % Blocking before Optimisation
% Waiting after job Sequence Optimisation % Working after Job Sequence Optimisation
% Changeover after Job Sequence Optimisation % Blocking after Job Sequence Optimisation
% Waiting after Buffer Size Optimisation % Working after Buffer Size Optimisation
% Changeover after Buffer Size Optimisation % Blocking after Buffer Size Optimisation
% Waiting after Job Sequence and Buffer Size Optimisation % Working after Job Sequence and Buffer Size Optimisation
% Changeover after Job Sequence and Buffer Size Optimisation % Blocked after Job Sequence and Buffer Size Optimisation
119
Chapter 6 – Discussion
6.1 Introduction
As discussed earlier, high variety/low volume (HV/LV) manufacturing systems are
more vulnerable to failures due to their dynamic and complex nature, which can be seen
as high level of variability induced by the high product mix, changing customer demand
and the manufacturing conditions itself. This not only vitiates the organisational
performance but also increases the manufacturing cost significantly (Bertrand and
Sridharan, 2001; Li, 2003; and Heike et al., 2001). In recent years, researchers have
proposed a number of methods to improve the manufacturing performance under highly
variable environments. For instance, according to Khalil et al. (2008), performance of
HV/LV manufacturing systems can be improved by reducing the level of variability and
by improving synchronisation of flow.
Proposed method here aligns with the research aim, which is automated lean CPS to
achieve process improvement. Genetic algorithm (GA) based combinatorial
optimisation has been integrated with a discrete even simulation (DES) tool. The DES
tool here works in an iterative manner with combinatorial optimisation model, which
may provide the quicker response to rapidly changing customer demand by determining
the optimal buffer sizes and job sequences. Results from the Chapter 5 have shown that
proposed model may have positive effect to improve the operational level measures by
reducing the level of variability and improving the synchronous flow.
This chapter exemplifies the experimental results and further discussion has been made
on the basis of collected data and exiting buffer management models.
120
6.2 Ability to Respond Quickly to the Variability without Compromising the
Organisational Goals
Steering the system in order to respond rapidly toward the high level of variability is
one of the essential factors to maintain organisational performance. In this research,
Lead Time (LT) and total inventory holding cost (TIHC) are considered as two
organisational goals as well as two objectives for combinatorial optimisation, which
may play the vital role in the success of an organisation. Combinatorial optimisation
with DES modelling here provides a tailored system to reduce the existing variability
and assists improving the flow of material. This research has investigated the variability
at the level of;
a. Customer Demand: Customer demand can be seen as a factor for variability in
terms of change in product quantity or product mix. Change in customer
demand quantity or product mix may have adverse effect on the lead time and
total inventory holding cost due to complexity of HV/LV manufacturing
environment, where parts may have different routes to follow and may have
variable setup and cycle times. Increasing the product mix may lead to the
larger number of machine setups. Optimal job sequence and buffer locations
need to be determined to accommodate all these changes in the manufacturing
environment.
b. System Variability: Similar to customer demand, variability induced from the
manufacturing environment itself needs to be examined to achieve the
synchronous flow, as different WorkCentre may have different parts to process,
variable breakdown time and capacity requirements, which may interrupt the
121
synchronous flow. Therefore, buffer sizes need to be optimised in order to
accommodate the proceeding and succeeding WorkStation’s requirements to
sustain the system in case of WorkCentre breakdown or product changeover.
Along this, optimal job sequence needs to be determined to reduce the effect of
interruptions due to the product change.
In summary, proposed combinatorial optimisation model has considered system-level
variability alongside customer demand to sustain the system against inconsistent
machine failures, setups, processing times and product routings. This may work as a
rapid tool to determine the optimal job sequences and buffer sizes to support
dynamically changing manufacturing environments.
6.3 Achieving the Synchronous Flow to Improve the Performance of System in
HV/LV Manufacturing Environment
According to the researchers and as discussed in Section 2.4, it has been observed that
in HV/LV manufacturing environment, non-Synchronous flow of material is one of the
contributors towards extended LTs and higher inventory holding costs (Khalil et al.,
2008). This research has proposed multi-objective GA based combinatorial optimisation
model to reduce the level of variability, which is one of the effective methods that can
be used to accomplish the synchronous flow. Furthermore, the effect of variability can
be reduced by coordinating the flow of material between different resources.
In this research, combinatorial optimisation has reduced the lead time and total
inventory holding cost significantly by optimising the job sequences and buffer sizes
under different type of variability for 500, 1000 and 2000 parts (customer demand) as
122
shown in Table 5.4, 5.5 and 5.6 respectively. After optimisation, the accomplishment of
synchronous flow can be seen as;
a. Once the optimal job sequence has been determined, fewer interruptions
required because of product change.
b. Optimal buffer sizes are determined to accommodate the proceeding and
succeeding WorkCentre in case of machine failure or changeover. Along this, it
provides control over the material release into the system, as the material release
is limited by available buffer capacity.
c. Optimal job sequence and buffer sizes together lead to accomplishment of
synchronous flow.
Along this, other advantages can be seen as lower work-in-progress (WIP) inventories,
improved flow of material and improved overall performance, which also has a direct
impact on the lead time and total inventory holding cost. Here, Figure 6.1 and Figure
6.2 illustrates the lead time and total inventory holding cost improvements for before
and after optimisation for batch size = 1 and customer demand = 500 parts. Similarly,
other results exhibit the same trend, which can be seen from the Table 5.4, 5.5 and 5.6.
123
Figure 6.1(Lead Time before and after optimisation for Batch Size = 1 and Customer
Demand = 500 Parts)
Figure 6.2(Total Inventory Holding Cost before and after optimisation for Batch Size
= 1 and Customer Demand = 500 Parts)
0
3000
6000
9000
12000
15000
18000
21000
Machine Failure No Machine Failure
Lead Time befor and after Optimisation
Before Optimisation Job Sequrence Optimisation
Buffer Size Optimisation Job Sequence and Buffer Size Optimisation
0
250000
500000
750000
1000000
1250000
1500000
1750000
2000000
Machine Failure No Machine Failure
Total Inventory Holding Cost before and after Optimisation
Before Optimisation Job Sequrence Optimisation
Buffer Size Optimisation Job Sequence and Buffer Size Optimisation
124
6.4 Contributions of Proposed Methodology
The contribution of proposed methodology towards knowledge can be given as;
a. Integration of Simulation Tool and Combinatorial Optimisation Method; in
this research, a generic GA based combinatorial optimisation method has been
proposed, which is integrated with DES tool (Simul8) to automate the process
improvement and for a rapid response to dynamically changing manufacturing
environment. The fitness of the solutions is measured based on lead time and
total inventory holding cost, which are the two optimisation objectives too. This
integration provides the adoptability and applicability of proposed methodology
in the wide range of problems, as any change in real-world scenario can easily
be incorporated to the DES model. The main features of proposed integrated
model are;
I. Represents the buffer management problem, where optimal buffer size
needs to be determined to reduce the lead time and total inventory
holding cost.
II. Allows genetic algorithms based combinatorial optimisation model to
generate an optimal job sequence to reduce the level of variability due to
changeovers.
III. Enables different products to follow different routes with variable
processing times and setup times.
IV. Allows change in customer demand, which can be in terms of quantity
or/and product mix.
125
V. Optimisation objectives can be varied according to the organisational
goals and problem to be solved.
VI. Quick response to change in variability and Provides visual
representation for the selected performance measures.
b. Use of Combinatorial Optimisation for Buffer Management and Job
Sequencing; in this research, buffer sizes and job sequence are the two inputs to
the proposed combinatorial optimisation model, i.e. either of these or both can
be optimised at the same time. Customer demand is used as one type of
variability in terms of quantity and product mix. Therefore, even minor changes
in customer demand might distraught the performance of the whole system, i.e.
buffer sizes and job sequence may need to be re-optimised to accommodate the
change in customer demand. Combinatorial optimisation model here provides a
flexible approach for problem solvers and/or decision-makers to select the
specific parameters for improvement. The results have been collected according
to the input to combinatorial optimisation model;
I. Job sequence.
II. Buffer size.
III. Both job sequence and buffer size.
Table 5.4, 5.5 and 5.6 illustrates results for customer demand of 500, 1000 and
2000 parts respectively under the different levels of variability included in the
proposed model. It is important to note that;
I. Job sequence optimisation has improved lead time significantly, as the
focus remains on the minimising the changeovers. There is reduction in
126
total inventory holding cost too, which is only coming from the reduced
changeovers.
II. On the other hand, determining the optimal buffer sizes may assist in
synchronisation of the flow of material, therefore, results has shown
expressively reduced lead time and total inventory holding cost. The
effect of buffer size optimisation can be given as;
1. In case of changeover and machine failure, optimal buffer sizes
may provide the adequate material and capacity for succeeding
and proceeding WorkCentre respectively, which may reduce the
lead time.
2. Along this, buffer size may limit the excessive WIP in the system
and restricts the amount of work released into the system, which
may significantly reduce the lead time and total inventory holding
cost by achieving synchronous flow. This allows system to
behave as a pull system, as material is only released when buffer
capacity is available.
III. Finally, determining the optimal job sequence and buffer size together
inherits the benefits of job sequence optimisation and buffer size
optimisation. This shows improved lead time and total inventory holding
cost on the previous two methods.
c. Dealing with Different Types of Variability; Khalil (2005) has addressed the
deterministic effects of variability and proposed a model to improve the
performance of flow lines in the light of different types of variability. In this
research, however, one type of variability is addressed by investigating trade-off
127
between multiple objectives of the combinatorial optimisation model by varying
the buffer sizes and job sequence. Unlike the single objective optimisation,
where only one objective is optimised (i.e. the main aim remains to find the best
solution) without considering the knock-on effect of optimisation on the other
performance measures.
On the other hand, there are other factors that have been considered in the
proposed optimisation model, which are not directly involved throughout the
process of optimisation. This variability can be exemplified as;
I. Product Mix; a customer order can consist of different type of parts,
having different processing requirements.
II. Customer Demand; customer demand can be changed in terms of
number of parts with respect to individual part or part type itself.
III. Routings; parts may follow different routes according to the WorkCentre
required to process the particular part type.
IV. Machine Failure; machine failure may cause blocking and waiting for
the proceeding and succeeding WorkCentre respectively because of
inadequate buffer capacities.
V. Setup Time; different part types may have different setup times, which
may cause increased lead times and longer processing queues.
VI. Processing Time; WorkCentre may need different processing times for
different products.
128
All these factors are considered by the proposed combinatorial optimisation
model as it exhibits the ability to respond according to change any of these
factors.
d. Inbuilt Root Cause Analysis (RCA); proposed model inherits some of the
principals of the Lean philosophy. While finding the optimal solution it
considers the cause-and-effect relationship between different performance
measures. In proposed combinatorial optimisation model two objectives have
been used i.e. reducing the lead time and total inventory holding cost, which
takes in account the effect of one objective on another. RCA implementation can
be observed from two different aspects, which are;
I. With respect to each objective function; proposed model here considers
the effect of improving one objective on other, as improving one
objective may have adverse effect on other. For instance, reducing buffer
sizes to all “1” or no buffers between WorkCentre can reduce total
inventory holding cost to its minimum level. However, at the same time
lead time can be increased significantly, because system won’t be able to
accommodate high level of variability and complexity of manufacturing
systems.
II. Relation between succeeding and proceeding WorkCentre; while,
deciding the optimal job sequence and buffer size, it’s essential to
consider the interrelationships between the succeeding and proceeding
WorkCentre because of high level of variability and complexity of
manufacturing environment. The proposed model here has taken in
account the relation between the succeeding and proceeding WorkCentre
129
implicitly to accommodate the variability due to setup, processing time,
machine failure and customer demand as product mix and quantity by
providing optimal buffer capacities.
e. Using Combinatorial Optimisation and Simulation Tool as Iterative Method;
proposed model has utilised combinatorial optimisation framework and the DES
model as an iterative method, which inherits the concept of continuous
improvement from the Lean philosophy. As illustrated in the Section 3.3.3,
combinatorial optimisation model can be used both to determine optimal buffer
sizes or job sequence or both job sequence and buffer sizes. Along this, solution
provided by each generation is the improvement over the previous generation,
which mimic the continuous improvement feature of the Lean philosophy.
6.5 Discussion of Results
This segment discusses the results collected through proposed methodology, as shown
in Chapter 5;
a. The proposed method is started by collecting data from the Technology Strategy
Board (TSB) project (Ref: K1532G) collaborators to develop the DES model.
Generic factors have been used to represent the different level of variability in
DES model, i.e. customer demand, product mix, routings, breakdowns,
processing time and setup time as described in Table 4.1, 4.2, 4.3 and 4.4. These
generic factors could be used in different manufacturing environments and are
applicable in both manufacturing and service industry.
b. Similarly, generic PMs (Table 4.5) have been chosen which are equally
applicable in different manufacturing environments and service industry.
130
Proposed combinatorial optimisation based method is evaluated based on the
two OFs, which are;
I. Lead time: time required to fulfil the customer demand.
II. Total inventory holding cost: total cost incurred to accommodate WIP.
c. Along this, other PMs, such as %working, %waiting and %changeover (Table
4.5) are used to exemplify the knock-on effect of one PM on other PMs to
determine the effect of improvement of PMs on each other.
d. The research has accompanied with different experiments to include the
complexity and depth of a real-world problem by introducing the different type
of variability that could occur in real environment. Along this, running different
experiments would give an insight of different performance measures that how
they can affect the lead time and total inventory holding cost as well as their
knock-on effect on each other.
e. Initial results are analysed to identify the bottleneck based on the performance
measures described in Table 4.5. Correlation analysis has been used to identify
the bottleneck resource according to the Step 6 of Section 4.4. From the result’s
analysis, a clear inference cannot be drawn for bottleneck identification. In
complex manufacturing environment, due to high level of variability different
WorkCentre spectacle an asymmetric trend, this may make it almost impossible
for problem solvers and\or decision-makers to decide precisely over the
bottleneck process. To determine the bottleneck effectively detailed analysis is
required by breaking down processes with respect to different type of variability.
This manual approach is not only time consuming but also there is higher
probability of mistakes. Along this, bottleneck may shift because of changes
131
induced in the manufacturing environment because of high level of variability
and complexity. Bottleneck analysis is given based on the data collected before
optimisation;
I. Figure 5.1a and 5.1b, Queue for M1 and M2 have strong positive
correlation with lead time and total inventory holding cost, which makes
WorkCentre M1 and M2 potential candidates for the bottleneck.
II. Similarly, from Figure 5.2a and 5.2b, for all batch size’s Queue for M2
has strong positive correlation with lead time and total inventory holding
cost. At the same time, Queue for M1 exhibits similar a trend as Queue
for M2 but only for batch size 5 and 10.
III. From Figure 5.3a and 5.3b, for all WorkCentre’s, % working exhibits a
very strong negative correlation with lead time and total inventory
holding cost for 500 and 1000 parts. While, for 2000 parts;
1.%working shows very strong negative correlation with lead time
and total inventory holding cost form M2 only.
2.%working shows very strong negative correlation with total
inventory holding cost only for M3.
IV. From Figure 5.4a and 5.4b,
1.For 1000 parts, M5 exhibits strong correlation between %waiting
and total inventory holding cost.
2.Similarly, for the 1000 parts M2 shows strong negative
correlation between the lead time and % waiting.
V. From Figure 5.5a and 5.5b,
132
1.M2 and M5 have very strong correlation between %changeover
and LT for all parts, whereas M3 exhibits similar trend but only
for 500 and 1000 parts.
2.WorkCentre M3, M4 and M5 show very strong positive
correlation between total inventory holding cost and
%changeover for 500 parts, 1000 parts and both 500 and 2000
parts respectively. While, WorkCentre M2 and, M3 and M5 show
strong positive correlation for all parts and 1000 parts
respectively.
As discussed earlier, from results it’s extremely difficult to identify the
bottleneck process, as high level of variability may make system behaviour
unpredictable. Along this, it is almost impossible to determine the effect of
different PMs on each other. Further, in-depth analysis is required to identify the
bottleneck process accurately and to select performance measures by
considering the knock-on effect on each other may be without making any false
perceptions.
f. To overcome this problem, research here has applied an integrated approach
using GA based combinatorial optimisation and DES to achieve synchronous
flow and to reduce the effect of variability, where initial bottleneck identification
is not required. The proposed combinatorial optimisation and DES model
elevates the system performance by implicitly considering the knock-on effect of
selected PMs on each other. After applying proposed methodology lead time and
total inventory holding cost has been improved significantly by determining the
133
optimal buffer size and/or job sequences, as shown in Table 5.4, 5.5 and 5.6 for
500, 1000 and 2000 parts respectively. After optimisation result’s analysis can
be given as;
I. For different customer demand, lead time and total inventory holding
cost are improved radically after applying optimisation. Three
optimisation approaches (Table 5.4, 5.5 and 5.6) has been used, which
are;
1. Job sequence optimisation; the main focus remains on the lead
time improvement by reducing the setups because of product
change. For instance, from Table 5.4 Experiment Number 1.1,
lead time is reduced from 20489 min to 8001 and total inventory
holding cost from 2032863 to 620692. Similar trend been shown
in the other experiments. This lead time improvement is from the
reduced setups and there is no control on the material flow as
buffer sizes are default i.e. not optimised.
2. Buffer size optimisation; In this case, only the buffer sizes been
optimised by keeping job sequence default. Buffer size
optimisation has radically improved the lead time and total
inventory holding cost both, as optimal buffer sizes for each of
the workstation may provide synchronous flow by controlling the
material flow. For example, from Table 5.4 Experiment Number
1.1, lead time is improved from 20489 to 8545 and total
inventory holding cost from 2032863 to 6061. Also, other
experiments show similar trend. However, there may be still
134
improvement opportunity as default job sequence may not be the
optimal one.
3. Finally, job sequence and buffer size optimisation together
epitomises a significant improvements in both lead time and total
inventory holding cost by inheriting the benefits of job sequence
and buffer size optimisation.
II. Proposed method has shown improvement according to identified PM’s
(Table 4.5);
1. Improved average queuing time after optimisation Figure 5.7a
to 5.12a. It is important to note that, Figure 5.7a and 5.8a
average queening time is reduced to its minimum after buffer size
optimisation, while Figure 5.9a to Figure 5.12a average queuing
time are reduced to minimum for both buffer size optimisation
and job sequence and buffer size optimisation. This may be
because on increased number of parts against product mix (i.e.
1000 and 2000 parts instead of 500).
2. Similarly, from Figure 5.7b to 5.12b reduced the average queue
sizes. Average queue size improvement shows similar trend as
average queuing time for 500, 1000 and 2000 parts as queue size
and queuing time are directly related to each other.
3. Finally, Figure 5.7c to 5.12c has shown an improvement in %
working and reduced changeovers due to the product mix. In fact
changeovers are significantly reduced after the job sequence
optimisation as the main target remains setup reduction. Also, job
135
sequence and buffer size optimisation together follows the similar
trend as synchronous flow and optimal job sequence contributes
towards setup reduction. However, only buffer size optimisation
does not reduce changeovers significantly there are still setup’s
involved due to the default job sequence, which may not the
optimal one.
6.6 Improving Different Performance Measures (PM) by Reducing the Effect of
Variability
As discussed in Section 3.4 PM’s are the fundamental building block for process
improvement and they help to identify the success or failure of a system. This research
has investigated the variability that can occur in flow lines on the basis of PMs
identified in Table 4.5. Along this, in this research PMs are used as a validation tool for
the proposed methodology.
In this research, performance measures are included to quantify the fitness of each
solution, which is reducing the lead time and total inventory holding cost.
Using optimal job sequence and buffer sizes, lead time and total inventory holding cost
are decreased radically as obtained from the results (Table 5.4, Table 5.5 and Table
5.6). However, there are other performance measures that contribute directly or
indirectly towards lead time and total inventory holding cost. Here, selected PMs can be
seen as;
136
a. lead time and total inventory holding cost represent the overall system
performance, which are affected by other operational level PMs such as %
working, % waiting, % changeover and queuing time.
b. The operational level PMs may be improved by reducing the level of variability.
For instance, queuing time and % changeover may be reduced by selecting the
optimal job sequence and buffer sizes.
c. Improving the right operational level PMs may improve the lead time and total
inventory holding cost significantly.
6.7 Applicability of Proposed Model with the Existing Systems
As discussed earlier, proposed model provides a reliable and quick responsive
framework for a complex manufacturing system to deal with the different types of
variability, which can be indirectly or directly affecting the system. Researchers have
developed different techniques for buffer management system such as Optimised
Production Technology (OPT), Theory of Constraint (TOC), Drum-Buffer-Rope
(DBR), Evolutionary Optimisation Methods and Pull System. Proposed model may
enhance the use of those methods, which can be given as;
a. Optimised Production Technology (OPT); OPT is a manufacturing control
philosophy by Goldratt in early 1980’s. The objective of OPT is to
simultaneously raise throughput while reducing inventory and operating costs,
and achieve a smooth, continuous flow of work. According to Watson et al.
(2007), OPT is based on nine rules (Table 6.1), which are developed by Goldratt
in 1986.
137
Table 6.1 (Optimal Production Technology Rules)
a. Balancing flow, not capacity.
b. Utilisation of a non-bottleneck resource is determined by constraint in
the system.
c. Utilisation and activation of a resource are not synonymous.
d. An hour lost at a bottleneck process is an hour lost for the total system.
e. An hour saved at a non-bottleneck is just a mirage.
f. Bottlenecks govern both throughput and inventory in the system.
g. A transfer batch may not, and many times should not, be equal to the
process batch.
h. The process batch should be variable, not fixed.
i. Schedules should be established by looking at all the constraints
simultaneously. Lead times are a result of a schedule and cannot be
predetermined.
The main focus of OPT rules remains the planning and optimisation of
constraint or bottleneck resource directly through rules b, d, e, f and i and
indirectly through rules a, c, g and h (Fresco, 2010). Proposed methodology,
therefore, aligns with the underlined foundation of OPT i.e. principal objective
remains to achieve synchronous manufacturing as a part of continuous
improvement. Along this, proposed methodology provides an advantage over
OPT having the ability to respond in highly variable complex manufacturing
environment.
b. Theory of Constraints (TOC); TOC is operation’s planning and control
philosophy that assists problem solvers, when the resources are limited and
conflicting. The main focus is to maximise the throughput by maximising the
throughput of constrained resource and minimising the non-value added
activities (Wei et al., 2002; and Linhares, 2009). According to Rahman (1998),
138
TOC strictly follows the five steps as shown in Table 6.2. Here, proposed model
can assist in the complex manufacturing environments, where;
I. TOC may be difficult to apply i.e. detailed analysis is needed or it’s
almost impossible to identify the system constraint or multiple system
constraints exist.
II. Constraints may quickly change due to high level of variability involved
in the manufacturing process.
III. Failure to identify the buffer capacities.
Along this, proposed model aligns with TOC concept as the main focus remains
same i.e. maximising the overall system performance and minimising the non-
value added activities.
Table 6.2 (Theory of Constraints Rules (Fresco, 2010))
a. System constraint identification.
b. Decide how to exploit systems constraints.
c. Subordinate everything else to the above decision.
d. Elevate the systems constraints.
e. If in any of the previous steps a constraint is broken, return to “Step a”.
Do not let inertia become the next constraint.
c. Drum-Buffer-Rope (DBR); DBR is a finite capacity scheduling mechanism for
planning and control in order to protect throughputs. DBR provides an improved
methodology over the TOC management philosophy. It is based on the three
basic elements, which are (Betterton and Cox, 2009; Stratton and Knight, 2010;
and Fresco, 2010);
I. Drum; defines the constrained resource, which limits the capacity of the
system.
139
II. Buffer; provides protection to Drum from different type of variability
involved in the system.
III. Rope; specifies the release of raw material to the production system
according to capacity of Drum.
DBR follows the sequence of tasks for material flow control in constraint based
systems, which are (Betterton and Cox, 2009; and Betterton and Cox, 2009);
I. Bottleneck or capacity constrained resource (CCR) identification.
II. Schedule CCR to maximise its use.
III. Synchronise all other resources according to the CCR production
schedule.
IV. Identify and quantify the buffer location where inventory needs to be
held.
Proposed model customises the concept of DBR methodology by;
I. Targeting improvement strategies for whole system instead of a
constraint resource only. This allows dealing with the bottleneck shift
due to high level of variability such as uncertain customer demand and
machine failure. i.e. bottleneck doesn’t need to be identified explicitly.
II. Determining the optimal sequence with which jobs need to be scheduled
to maximise the utilisation of the bottleneck resource. Similarly,
identification of optimal buffer sizes to accommodate variability induced
due to product changeovers and machine failures.
140
III. Integration of DES and GA based combinatorial optimisation model
allows the system to be adoptable to highly variable customer demand
and manufacturing environment.
d. Evolutionary Optimisation Methods; over the years, researchers have proposed
various evolutionary optimisation methods to achieve synchronous flow and
continuous improvement. For instance, Zang et al. (2009) has exemplified the
two-phase particle swarm optimisation algorithm for flow shop scheduling. On
the other hand, Fontanilli and Ponsonnet (2000) have used DES GAs as a
production optimisation tool. Similarly, there are other various examples where
different evolutionary techniques have been used such as ant colony mechanism,
GAs combined with swarm technology and simulated annealing. The proposed
multi-objective GA based combinatorial optimisation method can assist existing
evolutionary approaches as;
I. Multi-objective optimisation to deal with effect of PMs on each other.
Current research has used lead time and total inventory holding cost as
two objectives. However, proposed model is equally applicable with
other objectives, as different problems and organisations can have the
different goal to achieve.
II. Providing the optimal buffer size and job sequence may allow to create
the optimal schedule as well. As scheduling is merely the task of
arranging given sequence with respect to time and resource availability.
Optimal job sequence and optimal buffer sizes here improve the material
flow and provide with the reduced lead time and total inventory holding
cost, which may lead to the optimal schedule.
141
III. Integration of DES and combinatorial optimisation tool provides an
opportunity for problem solvers and decision-makers to validate the
solution before implementation.
e. Pull System; pull system is an integral element of lean philosophy to regulate
the flow of material by providing material according to what has been
consumed. According to Askin and Krishnan (2009), it is utmost important to
locate the optimal buffer levels, which can operate as a control point for pull
system implementation. Determining these control points can improve LT and
WIP levels significantly by providing the synchronous flow. In this research,
optimal buffer levels are determined to improve the flow of material, which
allow the system to behave like pull system. In proposed system, products
follow a sequential flow, but it’s not essential for all products to be processed on
all WorkCentre. Along this, proposed model allows to adjust the control points
(buffer levels) according to change in the level of variability, such as product
mix and customer demand.
6.8 Adoption of Proposed Method in Different Industrial and Service Sectors
Proposed combinatorial optimisation model is not only applicable in manufacturing
industry but also equally can be applied in different operational sectors, such as service
industry. The applicability issues of the proposed model are;
a. It is important to note that proposed model is integrated with DES tool, which
broadens the scope and applicability of proposed research in different
operational sectors. Here, DES model gives opportunity to represent the real
world problem that can fit with proposed methodology.
142
b. Proposed model uses the generic performance measures, which are applicable or
can be used in both service and manufacturing industry. This allows to,
I. Identify the goals and objectives w.r.to selected problem and operational
sector.
II. Formulate the problem according to the identified performance
measures.
c. The focus remains on the two main organisational objectives i.e. reducing the
lead time and total inventory holding cost by determining the Job sequence and
buffer sizes.
Proposed model here can be used to improve the operational performance by improving
the flow of material or information through the organisation.
143
Chapter 7 – Conclusion
Maintaining the performance of HV/LV (high variety and low volume) manufacturing
environment is one of the most challenging tasks, as high level of process/product
variability and can increase the lead time (LT) and manufacturing cost significantly. At
the same time, this variability cannot be ignored, as it is derived from the customer
demand. To stay in competition, therefore, it is essential to maintain the high-
performance levels under the light of high variability by achieving the synchronous
flow. The main aim of current research is to develop a methodology for automating
operations process improvement (PI) in order to cope with high level of variability and
complexity of HV/LV manufacturing environment.
The research has successfully developed a buffer management system based on
combinatorial optimisation and discrete event simulation (DES) modelling that may
help problem solver and decision-makers to accomplish the synchronous flow by
reducing effect of variability. There are other HV/LV manufacturing issues have been
addressed, which are;
a. GA based multi-objective combinatorial optimisation to determine optimal
buffer sizes and job sequences to reduce the effect of variability and promote the
synchronous flow. The optimal buffer sizes are determined to accommodate the
high level of variability and job sequence to reduce the number of setups
required in HV/LV manufacturing environment. Furthermore, proposed model
has used the trade-off between lead time and total inventory holding cost. This
also provides an opportunity for problem solvers and decision makes to select
solutions based on organisational priorities.
144
b. Provides the ability to manage system constraints to deal with different levels of
variability, where optimal solution is derived by considering the effect of
improving one performance measure on another through GA based
combinatorial optimisation.
c. Integration of DES and GA based combinatorial optimisation model to respond
quickly to changes in customer demand and variability within the different
process/activities to fulfil that demand.
d. Improvement over the existing DBR systems. Proposed model has exemplified
these advancements as;
I. Addressing the issue of shifting bottleneck or false bottleneck
identification to overcome the DBR failure modes.
II. Determining the optimal buffer sizes and job sequences to minimise the
lead time and total inventory holding cost.
e. Inbuilt RCA method within the proposed combinatorial to address the cause and
effect with respect to;
I. Each objective functions and selected performance measures.
II. Relation between proceeding and succeeding WorkCentre.
f. Adopting the lean creative problem solving where continuous improvement
plays a big role. The proposed model and simulation tool are used in an iterative
manner.
In summary, research here has achieved most of the objectives by using a complex
manufacturing environment model. The positive results have exemplified the
effectiveness and robustness under highly unstable circumstances. Previous research in
DBR illustrates that as volatility in manufacturing environment increases, the
145
effectiveness of DBR system decreases. However, proposed research model has tackled
high level of variability in HV/LV manufacturing environment and overcome the DBR
failure modes, as exemplified in Chapter 5 and Chapter 6 i.e. methodology has
successfully generated the optimal buffer sizes and job sequence under the light of high
variability by maintaining the reduced lead time and total inventory holding cost.
146
Chapter 8 – Future Work
This research has proposed a methodology for automated lean creative problem-solving
as a part of process improvement and has been validated in the complex HV/LV
manufacturing environment by inducing different levels of variability, as described in
chapter 4. According to the results in Chapter 5 and the discussion in chapter 6 and
chapter 7 proposed GA based multi-objective combinatorial optimisation model has
achieved research objectives, which are examined by investigating the job sequence and
buffer sizes.
The proposed research framework can be enhanced further as;
a. Batch size optimisation; Current results are collected using processing batch
sizes of 1, 5 and 10, whereas the transfer batch sizes are kept as 1. It will be
interesting to investigate the behaviour of the proposed methodology with
variable transfer batch sizes too, as GA may allow adapting the proposed model
by the inclusion of variable transfer batch sizes. In addition to this, no
optimisation criteria have used while choosing the processing and transfer batch
sizes. Selected experimental batch sizes are derived from the literature review.
In the future, there is an opportunity to include batch size optimisation with the
proposed model.
b. Include operator factor as a type of variability; in proposed methodology
resources are not considered while investigating different types of variability. In
future, effect of operators as part of different identified resource types examined
with respect to selected performance measures as;
I. Effect of travelling time on the lead time.
147
II. Effect of operator skills on the lead time and total inventory holding cost.
III. Deciding over the optimal number of operators needed.
IV. Measure resource/operator utilisation
148
References:
Agnetis, A., Alfieri, A. and Nicosia, G. (2004) A Heuristic Approach to Batching and
Scheduling Single Machine to Minimize the Setup Cost. International Journal of
Computers and Industrial Engineering, Vol. 46, Issue 4, pp. 793 – 802.
Ahuja, I. P. S. and Khamba, J. S. (2008) Total Productive Maintenance: Literature
Review and Directions. International Journal of Quality and Reliability Management,
Vol. 25, Issue 7, pp. 709 – 756.
Alford, D., Sackett, P. and Nelder, G. (2000) Mass Customisation – An Automotive
Perspective. International Journal of Production Economics, Vol. 65, No. 1, pp. 99 –
110.
Al-Kabbi, M., Khalil, R. and Stockton, D. (2009) Implementing Lean in Software
Development Operations. The Proceedings of 6th
International Conference on Product
Lifecycle Management, 6th – 8
th July, 2009, University of Bath, UK, pp. 690 – 698.
Al-Kabbi, M., Khalil, R. and Stockton, D. (2010) Improving Operations Management
Planning and Control of a Service and Project with Lean Principals. Proceedings of the
Junior Scientist Conference of Science and Technology for Future, 4th – 7
th April, 2010,
Vienna University of Technology, Austria, pp. 61 – 62.
Ammerman, M. (1998) Root Cause Analysis Handbook – A Simplified Approach to
Identifying, Correcting and Reporting Workplace Errors. Productivity Press, New York,
USA.
Antony, J., Somasundarum, V. and Fergusson, C. (2004) Applications of Taguchi
Approach to Statistial Design of Experiments in Czech Republican Industries.
International Journal of Productivity and Performance Management, Vol. 53, No. 5,
pp. 447 – 457.
149
Askin, R. G. and Krishnan, S. (2009) Defining Inventory Control Points in
Multiproduct Stocha - stic Pull System. International Journal of Production Economics,
Vol. 120, No. 1, pp.
Banks, J. (1999) Introduction to Simulation. Proceedings of the 1999 Winter
Simulation Conference, 05th – 08
th Dec., 199, Atlanta, USA, Vol. 1, pp. 7 – 13.
Banks, J., Carson, J. S., and Nelson, B. L. (1996) Discrete-Event System Simulation.
Prentice-Hall Inc., New Jersey, USA, 2nd
Edition.
Bashford, H., Sawhney, A., Mund, A. and Walsh, K. (2002) Process Mapping of
Residential Foundation Slab Construction Processes. Proceedings of the IEEE Winter
Simulation Conference, 8th – 11
th Dec., 2002, Arizona, USA, Vol. 2, pp. 1752 - 1758.
Bashin, S. and Burcher, P. (2006) Lean Viewed as a Philosophy. Journal of