LIBRARY
Michigan State
University

PLACE II RETURN BOX to roman this Mom from your rocord.
TO AVOID FINES rotum on or bdoro date duo.

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU In An Afflrmdlvo ActionlEqa-l Opportunity Innuulon

 

ANALYSIS OF TOOL SCHEDULING STRATEGIES IN A
STOCHASTIC ENVIRONMENT WITH A FINITE

LIFE RESOURCE

By

Steven B. Lyman

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Management

1993

ABSTRACT
ANALYSIS or TOOL SCHEDULING STRATEGIES IN A
STOCI-IASTIC ENVIRONMENT WITH A FINITE
LIFE RESOURCE
By

Steven B. Lyman

Tooling has become a growing concern in most manufacturing environments.
This is most apparent in environments which have eliminated inventory and capacity
buffers which hide problems. Tooling is often the main determinate of Shop floor
capacity and performance. The recognition of tooling as a finite life resource has
brought about this research which involves tool control.

While there is an extensive body of research which examines machine and
labor resources, neither emulate tooling. Past dual resources constraint (DRC)
research examines only machine and labor as infinite life resources. The unique
nature of tooling necessitates the need for control and scheduling procedures which
considers these differences. Specifically, the unique traits of tooling are, machine
Specific because of size, finite life, and can be refurbished (maintenance).

The purpose of this research is to examine how a finite life resource, which
has maintenance performed intermittently, should be controlled. How should a DRC
flow Shop, with a finite life resource, be controlled when both tool life and
maintenance time are characterized by a stochastic distribution ill be addressed. The
DRC model used in this Simulation research attempts to answer this question by

examining the scheduling of jobs and control of tools.

Output from this model was analyzed using ANOVA and Tukey multiple
comparisons. The analysis found that tool control was the dominate factor in
determining Shop performance, followed by job scheduling. How tools were
controlled, via maintenance, determined tool availability which influenced Shop
performance. Those tool rules which promoted frequent preventive maintenance over
that of corrective maintenance enhanced performance.

As for job scheduling, job priority rules (dispatching) were most influential
on due date performance measures. Job priority rules which prioritized by sequence
dependency over considering all job due dates performed worse. While this result

contradicts past research, it can be attributed to the finite life of tooling.

DEDICATED TO MY WIFE

CAROL LYNN REWERS

THANK YOU

iv

ACKNOWLEDGEMENTS

I wish to thank several individuals who provided assistance and support
during my dissertation process. Professor Steven A. Melnyk, Chairman of the
Dissertation Committee, was instrumental throughout the dissertation. He provided
direction, support, input, and most of all motivation. His knowledge and
accessibility have added innumerable to the completion of this dissertation. A Sincere
thanks for all he has done.

Professors Phil Carter, Soumen Ghosh, and Robert Handfield, the members
of the dissertation committee, have continually provided valuable feedback and
insight. Their assistance have added to the quality of this dissertation.

Several fellow doctoral students in the Management Department have also
helped during my years in the program. A special thanks go out to, Michael D’Itri,
Vijay Kannan, David Mendez, and Keah Choon Tan. Their contributions and
friendship have made the program enjoyable.

Lastly, I would like to thank the members of my family, both my parents,
Richard and Irma Lyman, and my in-laws, Leonard and Doris Rewers. Without their
support, encouragement, and patience, both the Ph.D. program and dissertation

would not have been possible.

 

US
US

CH

 

 

 

CH

.——

 

 

 

TABLE OF CONTENTS

LIST OF TABLE
LIST OF FIGURES

CHAPTER 1 : INTRODUCTION AND PROBLEM STATEMENT
1.1 Introduction
1.2 Description of Tooling Environment
1.2.1 Tool Life
1.2.2 Traits of Tooling
1.3 Problem Statement
1.3.1 How do we Schedule Jobs?
1.3.2 How do we Schedule Tools?
1.3.3 How does Variation in Resource Life and
Renewal Effect the Scheduling and Assignment
Decision
1.3.4 Specific Research Questions
1.4 Research Methodology
1.5 Research Contribution _
1.6 Organization of Dissertation

CHAPTER 2 : BACKGROUND AND LITERATURE REVIEW
2.1 Introduction
2.2 Tooling Characteristics
2.2.1 Tool Types
2.2.2 Tool Life
2.2.2.1 Description
2.2.2.2 Tool Life Distributions
2.2.3 Tooling Economics
2.2.4 Tooling Characteristic Summary
2.3 Tool Scheduling
2.3.1 Flexible Manufacturing Systems
2.3.1.1 FMS and Tool Characteristics
2.3.1.2 FMS and Individual Machine Control
2.3.1.3 FMS and Tool System Management
2.3.2 Tool Scheduling in Non-FMS Machine Models
2.3.3 Summary of Tool Scheduling Literature

vi

v1i

\IOvfiH

11

15
16

17
18
19
20

22
23
23
24
24
26
28
28
29
29
31
33
35
36
38

2.4 Maintenance 40

2.4.1 Maintenance Strategies 41

2.4.2 Maintenance Models & Characteristics 41
2.4.2.1 Classification 41

2.4.2.2 Failure and Service Time Distributions 46

2.4.2.3 Maintenance and Tool Availability 47

2.4.3 Corrective & Preventive Maintenance Scheduling 48
2.4.3.1 Descriptive Models 48

2.4.3.2 Analytical Models 50

2.4.4 Production and Maintenance Scheduling 52

2.4.5 Summary of Maintenance Literature 54

2.5 Dual Resource Constraint & Labor Scheduling Models 56
2.5.1 Operational Issues in DRC 58
2.5.1.1 Dispatching Rules 58

2.5.1.2 Due Date Rules 59

2.5.1.3 Labor Assignment 60

2.5.2 Design Issues in DRC 63
2.5.2.1 Labor 63

2.5.2.2 Information Control 64

2.5.3 Summary of DRC Literature 65

2.6 Sequence Dependent Models 66
2.6.1 Sequence Dependent Scheduling Rules 67
2.6.1.1 Tooling Sequence Dependency 68

2.6.2 Group Scheduling 70

2.6.3 Summary of the Sequence Dependency Literature 73

2.7 Summary of Literature Review 74

CHAPTER 3 : TOOL PLANNING AND CONTROL: A CONCEPTUAL
FRAMEWORK

3.1 Introduction 76
3.2 Tooling Management Framework 76
3.2.1 Planning of Tools 78
3.2.2 Scheduling of Tools 79
3.2.3 Shop Floor Control of Tools 82
3.3 Detailed Scheduling of Forming Tools 83
3.3.1 Tool Timing and Placement Decision 83
3.3.2 Examples of Tool Control 85
CHAPTER 4 : RESEARCH METHODOLOGY AND SIMULATION MODEL
4.1 Introduction 91
4.2 Model Development 92
4.3 Validation of Experiment 93
4.3.1 External Validity 93

4.3.2 Construct Validity 94

vii

4.3.3 Internal Validity
4.3.4 Statistical Conclusion Validity
4.4 Simulation Design Issues
4.4.1 Verification
4.4.2 Initialization Bias
4.4.3 Variance Reduction
4.4.4 Sample Size
4.4.4.1 Normality
4.4.4.2 Auto-Correlation
4.5 Description of Simulation Environment
4.5.1 Shop Model Parameters
4.5.1.1 Due Date Setting
4.5.1.2 Processing and Setup Time
4.5.1.3 Machine and Tool Assignment
4.5.1.4 Dispatching Rule
4.5.1.5 Shop Control Heuristics
4.5.1.6 Mean Tool Life
4.5.1.7 Preventive Maintenance Point and
Percentage Estimate
4.5.1.8 Maintenance Service Time
4.5.1.9 Parameter Summary
4.5.2 Experimental Factor Levels
4.5.2.1 Job Priority Heuristics
4.5.2.2 Tool Control Heuristics
4.5.2.3 Tool Life Distribution
4.5.2.4 Maintenance Service Time Distribution
4.5.3 Model Assumptions
4.6 Performance Measures
4.7 Data Collection
4.8 Summary

CHAPTER 5 : RESEARCH HYPOTHESES AND DATA ANALYSIS

5.1 Introduction

5.2 Analysis of Effects

5.3 Research Hypotheses

5.4 Post Hoc Analysis
5.4.1 Tool Life Variance Analysis
5.4.2 Maintenance Service Variance Analysis
5.4.3 Job-Tool Interaction Analysis

5.5 Data Analysis Procedures
5.5.1 Testing for Normality
5.5.2 Testing for Homogeneity of Variance
5.5.3 Transformation of Date

viii

94
95
96
96
97
100
101
101
102
103
103
105
106
107
108
109
109
110

110
111
111
111
115
119
120
120
121
122
123

124
124
125
135
135
137
138
138
139
140
140

CH

811

5.5.4 Residual Analysis
5.5.5 Data Analysis Summary
5.6 Summary

CHAPTER 6 : EXPERIMENTAL ANALYSIS AND CONCLUSIONS
6.1 Introduction
6.2 Analysis of Effects for Performance Measures
6.2.1 Mean Time in System
6.2.2 Standard Deviation of Time in System
6.2.3 Mean Tardiness
6.2.4 Standard Deviation of Tardiness
6.2.5 Percentage of Jobs Late
6.2.6 Percentage of Tool Failures
6.2.7 Analysis of Effects Summary
6.3 A Prior Hypotheses Analysis
6.3.1 Hypothesis 1
6.3.2 Hypothesis 2
6.3.3 Hypothesis 3
6.3.4 Hypothesis 4
6.3.5 Hypothesis 5
6.3.6 Hypothesis 6
6.3.7 Hypothesis 7
6.3.8 Hypothesis 8
6.3.9 Hypothesis 9
6.3.10 Research Hypotheses Summary
6.4 Post Hoc Analysis
6.4.1 Relative Performance of Heuristics
6.4.1.1 Tool Life Variance
6.4.1.2 Maintenance Service Time Variance
6.4.2 Job Priority Rule By Tool Control Rule
6.4.3 Summary of Post Hoc Analysis
6.5 Discussion of Results
6.6 Summary of Analysis
6.7 Future Research Directions
6.8 Conclusions

BIBLIOGRAPHY

ix

140
142
143

144
145
145
150
154
158
162
166
167
173
174
174
186
186
196
196
204
205
214
214
221
222
222
230
230
240
241
246
249
252

256

6- 14d

6-14e

6-14f

LIST OF TABLES

Comparative Rating of Tooling Aspects by Tool Type

Forming Tool Classifications

Relationship of Tool-Maintenance Variance

Summary of Simulation Environment

Tool-Machine Assignment Matrix

Design of Experiment

Priority Levels for Job Priority Rule PSR4

Model Assumptions

Treatments in Experiment

Refinement of Research Issues to Hypotheses

Analysis of Variance for Mean Time in System

Treatment Means for Mean Time in System

Analysis of Standard Deviation of Time in System

Treatment Means for Standard Deviation of Time

In System

Analysis of Mean Tardiness

Treatment Means for Mean Tardiness

Analysis of Standard Deviation of Tardiness

Treatment Means for Standard Deviation of Tardiness

Analysis of Percentage of Jobs Late

Treatment Means for Percentage of Jobs Late

Analysis of Log Percentage of Tool Failures

Treatment Means for Percentage of Tool Failure

Synopsis of ANOVA Results From Analysis of Effects

Tukey Multiple Comparisons of Jobs Priority Rules for NOPM
Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for FPTPM
Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for VARLO
Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for VARHI
Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for VARPM
Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for MQBPM

17
105
107
112
114
120
126
136
146
147
150
151

154
155
158
159
162
163
167
164
171
178

178

179

179

180

180

6- 14g
6- 15
6- 16a
6— 16b
6- 16c
6- 16d
6- 17
6- 1 8
6- 19
620
6—2 1
6-22
6-23a
6-23b
6-24a
6—24b
6-25

6-26

Tool Rule.

Tukey Multiple Comparisons of Jobs Priority Rules for JDDTL
Tool Rule.

Analysis of Variance for Sequence Dependent Jobs Priority Rules
SDTC and SSTL.

Tukey Multiple Comparisons of Tool Control Rules for DRTC
Job Rule.

Tukey Multiple Comparisons of Tool Control Rules for SDTC
Job Rule.

Tukey Multiple Comparisons of Tool Control Rules for PSR4
Job Rule.

Tukey Multiple Comparisons of Tool Control Rules for SSTL
Job Rule.

Analysis of Variance for Early and Postponed Variable PM Tool
Control Rules

Analysis of Variance for Variable PM Tool Control Rules with
Maintenance Queue Information (MQBPM) and Without (VARPM)
Analysis of Variance for Fixed PM Point Tool Control Rules
Control Rules Which Considers Job Due Date (JDDTL) and Does
Not (FPTPM)

Analysis of Variance for Tool Life Variance

Analysis of Variance for Maintenance Service Time Variance
Summary of Hypothesis Results

Tukey Multiple Comparison of Job Priority Rules by

Tool Life Variance

Tukey Multiple Comparison of Tool Control Heuristics

by Tool Life Variance

Tukey Multiple Comparison of Job Priority Rules by
Maintenance Service Variance

Tukey Multiple Comparison of Tool Control Heuristics

by Maintenance Service Variance

Tukey Multiple Comparison of Job Priority Rules by

Tool Control Rules

Example of Tool Maintenance Time

xi

181
181
190
190
191
191
197
197
206
213
213
218
223
223
231
231
238

244

 

3.

-a‘

1-21
1-3;
1-31

T‘
J...

f_)
o
._.

J
J

L») '3’.) (7*) r ) r J t J t. J r
O I O I
a.) ta.) .__.. r) LII $.- 'J..p r

3-5b
4-1

6-1a
6-lb
6-1c
6-ld

6-2a

LIST OF FIGURES

Tool Taxonomy

Comparison of Maintenance Policies and Frequency of
Occurrence

Comparison of Maintenance Frequency to Costs
Information Involved in Job Selection: Basic Model
Information Involved in Job Selection: Decision Flow
Spheres of Research Focus

Tooling Issues Taxonomy

FMS Planning and Control Hierarchy

Tool Scheduling Taxonomy

Maintenance Policies

Maintenance Taxonomy

Taxonomy of DRC Research

Production Planning and Control

Tool Scheduling and Control

Shop Layout and Tool Control: Tools are Machine
Specific With No Duplicates

Shop Layout and Tool Control: Continued

Shop Layout and Tool Control: Continued

Shop Floor and Tool Control: Tool Flexibility

with No Duplicates

Shop Floor and Tool Control: Continued

Shop Floor and Tool Control: Tool Flexibility

with Multiple Duplicates

Shop Floor and Tool Control: Continued

Flow of Work Through the Shop

Comparison of Control Rules: Low Tool/Low Maintenance
Variance, Time in System

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Time in System

Comparison of Control Rules: Low Tool/High Maintenance
Variance, Time in System

Comparison of Control Rules: High T ool/High Maintenance

Variance, Time in System

Comparison of Control Rules: Low Tool/Low Maintenance

xii

10
14
14
20
30
32
39
43
55
5 7
77
80
86

86
86
88

88
90

90
104
148
148
149
149

152

 

 

1.... _-

6-31)
6-2:
62d
(+33
(th
63c
(rid
64a
64b

64C

(>53
65b
65c
65d
66a
661)
its

6M

6‘73

67c ;

67d
We

6-2b
6-2c
6-2d
6-3a
6-2b
6-3c
6-3d
6—4a
6-4b
6-4c
6-4d
6-5a
6-5b
6-5c
6-5d
6-6a
6-6b
6-6c
6—6d
6-7a
6-7b
6—7c

6-7d
6—7e

Variance, Standard Deviation Time in System

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Standard Deviation Time in System

Comparison of Control Rules: Low Tool/High Maintenance
Variance, Standard Deviation Time in System

Comparison of Control Rules: High Tool/ High Maintenance
Variance, Standard Deviation Time in System

Comparison of Control Rules: Low Tool/ Low Maintenance
Variance, Tardiness

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Tardiness

Comparison of Control Rules: Low Tool/ High Maintenance
Variance, Tardiness

Comparison of Control Rules: High Tool/High Maintenance
Variance, Tardiness

Comparison of Control Rules: Low Tool/Low Maintenance
Variance, Standard Deviation Tardiness

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Standard Deviation Tardiness

Comparison of Control Rules: Low Tool/High Maintenance
Variance, Standard Deviation Tardiness

Comparison of Control Rules: High Tool/High Maintenance
Variance, Standard Deviation Tardiness

Comparison of Control Rules: Low Tool/Low Maintenance
Variance, Percentage of Jobs Late

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Percentage of Jobs Late

Comparison of Control Rules: Low Tool/High Maintenance
Variance, Percentage of Jobs Late

Comparison of Control Rules: High Tool/High Maintenance
Variance, Percentage of Jobs Late

Comparison of Control Rules: Low Tool/Low Maintenance
Variance, Percentage of Tool Failures

Comparison of Control Rules: High Tool/Low Maintenance
Variance, Percentage of Tool Failures

Comparison of Control Rules: Low Tool/High Maintenance
Variance, Percentage of Tool Failures

Comparison of Control Rules: High Tool/High Maintenance
Variance, Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness

Mean Values for Percentage of Jobs Late

xiii

152
153
153
156
156
157
157
160
160
161
161
164
164
165
165
169
169
170
170
175
175
176

176
177

6-7f

6-8a

6-8b

6-8c

6-8d

6-8e

6-8f

6-9a

6-9b

6«9c

6-9d

6-9e

6—9f

6-10a
6-10b
6-10c
6-10d
6-10e
6—10f
6-11a
6-11b
6-11c
6-11d
6-11e
6-11f
6-12a
6-12b
6-12c
6-12d
6-12e
6-12f
6-13a
6-13b
6-13c
6-13d
6-13e
6-l3f
6-14a
6-l4b
6-14c
6-14d
6-14e
6—14f
6—15a

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness
Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Mean Values for Time in System

xiv

177
183
183
184
184
185
185
187
187
188
188
189
189
193
193
194
194
195
195
198
198
199
199
200
200
201
201
202
202
203
203
207
207
208
208
209
209
210
210
211
211
212
212
215

 

(>le 1
615C 1
615d 1
615i: 1
6151‘ 1
616a I

frifib .
616C

616d

6-15b
6-15c
6-15d
6-15e
6—15f
6-16a
6-l6b
6-16c
6-16d
6-16e
6-16f
6-17a
6-17b
6-17c
6-17d
6-17e
6-17f
6-18a
6-18b
6-18c
6-18d
6-18e
6-18f
6-19a

6- 19b

Mean Values for Standard Deviation of Time in System
Mean Values for Tardiness

Mean Values for Standard Deviation of Tardiness

Mean Values for Percentage of Jobs Late

Mean Values for Percentage of Tool Failures

Comparison of Job Priority Rules by Tool Life Variance
for Mean Time in System

Comparison of Job Priority Rules by Tool Life Variance
for Standard Deviation Time in System

Comparison of Job Priority Rules by Tool Life Variance
for Mean Tardiness

Comparison of Job Priority Rules by Tool Life Variance
for Standard Deviation Tardiness

Comparison of Job Priority Rules by Tool Life Variance
for Percentage of Jobs Late

Comparison of Job Priority Rules by Tool Life Variance
for Percentage of Tool Failures

Comparison of Tool Control Rules by Tool Life Variance
for Mean Time in System

Comparison of Tool Control Rules by Tool Life Variance
for Standard Deviation Time in System

Comparison of Tool Control Rules by Tool Life Variance
for Mean Tardiness

Comparison of Tool Control Rules by Tool Life Variance
for Standard Deviation Tardiness

Comparison of Tool Control Rules by Tool Life Variance
for Percentage of Jobs Late

Comparison of Tool Control Rules by Tool Life Variance
for Percentage of Tool Failures

Comparison of Job Priority Rules by Maintenance Service
Variance for Mean Time in System

Comparison of Job Priority Rules by Maintenance Service
Variance for Standard Deviation Time in System
Comparison of Job Priority Rules by Maintenance Service
Variance for Mean Tardiness

Comparison of Job Priority Rules by Maintenance Service
Variance for Standard Deviation Tardiness

Comparison of Job Priority Rules by Maintenance Service
Variance for Percentage of Jobs Late

Comparison of Job Priority Rules by Maintenance Service
Variance for Percentage of Tool Failures

Comparison of Tool Control Rules by Maintenance Service

Variance for Mean Time in System

Comparison of Tool Control Rules by Maintenance Service

XV

 

215
216
216
217
217
224
224

225

226

227

227

228

228

229

229

232

232

233

233

234

234

235

235

 

6- 19c

6-19d

6-19e

6-19f

6-20

Variance for Standard Deviation Time in System
Comparison of Tool Control Rules by Maintenance Service
Variance for Mean Tardiness

Comparison of Tool Control Rules by Maintenance Service
Variance for Standard Deviation Tardiness

Comparison of Tool Control Rules by Maintenance Service
Variance for Percentage of Jobs Late

Comparison of Tool Control Rules by Maintenance Service
Variance for Percentage of Tool Failures

Total Maintenance Time

xvi

236

236

237

245

 

CHAPTER 1

INTRODUCTION AND PROBLEM STATEMENT

1.1 INTRODUCTION

Control systems such as Just-In—Time (JIT) and Computer Integrated
Manufacturing (CIM) have eliminated buffers in manufacturing environments,
including work in process and excess capacity. With fewer buffers, tooling has a
greater effect on production performance (Mason, 1986) such as delay order
processing. Tooling has become a major issue in most manufacturing environments.
In a Flexible Manufacturing System (FMS) for example, machining center
flexibility is determined by the number of different tools in the storage magazine
(Gray, Seidmann and Stecke, 1986). FMS tooling is a critical resource that needs
special control procedures (Gruver and Senninger, 1990).

In the automotive industry (Vasilash, 1990) and in other‘repetitive
manufacturing environments, the concern over tooling centers on costs. As much as
20 percent of a firm’s material costs (Erhom, 1983) can be attributed to tooling. In
the case of FMS, an estimated 25 to 30 percent of fixed and variable costs are due
to tooling (Kouvelis, 1991). Annual costs for various forms of tooling can exceed
$1 Million for small firms and $100 million for large firms (Huber, 1989). The
US. metalworking industry spends an estimated $1.5 billion a year on cutting tools
alone (Mason, 1991). As firms strive to utilize assets more effectively, tooling

issues take on greater importance. Idle or unused resources are a drain on assets

inte
dett
tool

5

[hrs

had

$16

103:

2

and profits. Effective tooling asset management requires the right tool at the right
machine. Without adequate control, machines and workers are idle, and schedules
and delivery dates are missed (Kupferberg, 1986).

As the study of shop floor control focuses more on tooling its effect on
performance becomes apparent, particularly with respect tO capacity (Kupferberg,
1986; Blackburn, 1989). Tooling is Often the main determinant of Shop floor
capacity. Up to 16 percent of production schedules (Mason, 1991) may be missed
because of such problems as insufficient tool life, lost tools, and tool failure.

The importance of tooling was also confirmed during two separate
interviews. In the first case, an automotive parts supplier was in the process Of
developing a fully integrated automated tool control system to monitor all forms of
tooling. The tool control system will be a key component in a new production plant
beginning operation soon. The supplier had decided to invest in the new system for
three reasons.

First, to satisfy customer due date (delivery dates). In the past, the supplier
had not monitored tool wear and maintenance, resulting in unscheduled down time
and missed delivery dates. With the adoption of HT by the supplier’s customers,
lost capacity due to tooling can no longer be tolerated.

Second, some customers believe that properly maintained tools can provide a
better quality product. Research has shown that well-maintained tools not only yield
higher quality products but also ensure reliable capacity (Finch and Gilbert, 1986).

Since the supplier had no record of scheduled maintenance, the customer questioned

3

whether the supplier was capable of maintaining quality tolerances for products.

Third, the supplier estimated that each plant operation had up tO $25 million
in obsolete or unusable tools stored throughout the plants. Another cost factor
involved buffers, as noted earlier. When the supplier moved to reduce inventory
cost by lowering work in process and excess tooling, it became apparent that
controls were necessary to plan tooling use and maintenance. It also became
apparent that the scheduling of tooling for production was a function of required
maintenance and differs from the traditional procedures used for machines and
equipment. The variability of tool types 030th cutting and forming) contributed to
uncertainty, and the supplier had to reevaluate and develop new policies for
controlling tools.

In the second case, the researcher interviewed a firm supplying consumer
products and discovered similar problems. The manager had adopted modular tools
to cut tooling cost, up to 40 percent. The modular tools allowed the firm to make
variations of its products using the same tool by inserting or moving sections on the
tool. The new tools increased flexibility, but reduced tool life and increased
required maintenance. This in turn, caused problems in tool availability and
production scheduling. New policies had to be adopted to control for those
contingencies.

As these cases illustrate, tooling is a constraining resource long overlooked
by production managers. With the adoption of HT and other new approaches, it has

become apparent that tooling is a key component of manufacturing that companies

no longer can afford to ignore.

1.2 DESCRIPTION OF TOOLING ENVIRONMENT
Briefly, tooling is composed of a variety Of items which include: jigs,
fixtures, forms, dies, cutting tools, gauges, molds, templates, and more (Broom,

1967). Figure 1-1 Shows the various forms Of tooling. There are production and

 

 

 

TOOL
1
I I
PRODUCTION NON-PRODUCTION
l
I I
CUTTING FORMING __ GAUGES
_ JIGS
GENERAL DIES _ FIXTURES
SPECIAL INJECTION ..__ PALLETS
MODULAR MOLDS j... HOLDE R8
— TEMPLATES
_ SET-BLOCKS
._ AN OLE-PLATE

 

 

 

 

non-production tools, differentiated by the fact that production tools are those used
to shape or alter the product directly.

Table 1-1 compares different tool types on a number of dimensions. For
example, costs for a standard cutting tool (drills) are low, specialized cutting tools

(multi-bit mill) are moderate (up to $10,000), and costs for dies are high ($100,000

5

and up). It should be noted that a number of the dimensions are interrelated such as

replacement frequency and life Of tool.

Table 1-1 Comparative Rating Of Tooling Issues by Tool Type

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Tool Type
Standard Special Stamping Injection
Tooling Issues Cutting Cutting Dies Dies
Costs per Tool Low Moderate High High
Purchasing Lead Time Low Low High High
Tool Standardization High Moderate Low Low
Life of Tool Low High High High
Breakage Probability High Moderate Moderate Moderate
Breakage Cost ' Low Low High High
Maintenance Frequency High High Moderate Moderate
Maintenance Lead Time Low Low Moderate Moderate
Replacement Frequency High Moderate Low Low
Replacement Lead Time Low Low High High
Speed Variability High High Low Low
Scheduling Difficulty Low Low High High
Duplicates Likely High Moderate Low Low
Batch Run Length Low Low High High

 

 

 

 

 

 

As can be seen in Figure 1-1, production tools fall into two categories,
cutting or forming and it is this latter category which is of interest in this research.
Forming tools have two important features: cost and uniqueness. It is not
uncommon for forming tools to cost as much as the machine on which it is used

(Brown, Geoffrion and Bradley, 1981), in excess Of $1 million and for that reason

 

‘9.“‘J
midaquS
’he :05:
Wren a

cuttng t

service t
We a 1

directly

J

Classifix -

SC

C1353111C3
Shown in

”331131 fia

~_.

Una‘aiiab

prodUCiio;

6

multiple tool copies are rare. The single tool COpy places a capacity constraint on
the jobs produced and is especially apparent when the tOOl requires maintenance.
When a forming tool breaks or needs service, it is pulled for repair, and unlike

cutting tools, can not be easily replaced with a new copy.

1.2-119mm

A major variable in the production environment is tool life. Tool life is
defined as: the period between placing a tool into production and removing it from
service because it no longer yields a quality (usable) product. Once the tool fails to
make a quality part, it must be refurbished or repaired. Obviously, tool life is
directly related to control and scheduling Of both production and maintenance.

A unique difference between cutting and forming tools is that the latter are
not discarded until a model change makes them Obsolete. Depending on their
classification, forming tools generally do not wear out permanently. The
classification of forming tools found in the Tool Engineers Handbook (1949) is
shown in Table 1-2. This research centers on the type A classification.

During the life of a forming tool there may be several refurbishment or
maintenance operations. In this research, tool life is defined as the period between
placing a tool into production and removing it from service for maintenance. Tool
life is a finite life resource because there will be periods when the tool is
unavailable for production. The fact that tools are not always available for

production and the subsequent impact on capacity has been noted by researchers

1, ,1, , n
IBMb...

 

(Blackburn, 1989; Kupferberg, 1986).

Table 1-2 Forming Tool Classifications.

 

Class Description of Classification

 

A Best type and grade of materials used for long life. Designed for high
volume production and for ease of maintenance.

 

B Applicable for medium production quantities. Designed tO last for total
production run only. Less consideration given to ease of maintenance.

 

C Cheapest useable tools for low volume production. Limited life with
little or no maintenance done on tool.

 

Temporary Used for limited production runs typically found in job shops. Lowest
cost tool that can produce the part.

Source: 'Iool Engineers HandFOOF, 19:9

 

 

 

 

 

 

1.2-2 mm

Tooling differs from other production resources because tools: (1) are
specific to machines and jobs, (2) have'a finite life, and (3) are renewable (Melnyk,
Ghosh and Ragatz, 1989). Some of these traits were mentioned previously, and are
now described in detail.

1) Specificity: Tools usually are used on a designated machine because of size
or configuration. Frequently, the machine is the only resource capable of using the
tool for production. Use on a designated machine also simplifies shop floor control,
as the tool may be too large to transport, and use at one location Simplifies the flow
of production. Compared to cutting tools forming tools tend to be more machine
specific.

Another aspect of specificity involves the specific tasks of each job or order.

Cutting tools usually are flexible, that is, different types such as reamers and mills

EKI

machine
tooling is
3') F;
1001 yieid
mama;
filtration
Gillimore
may not 2
Combinau
Filer. 19
81111 is iii
the W of
1101mm “

mailman.

8

can perform the same production task (Slack, 1987). In reality, this flexibility is
limited because job’s require certain production tasks by certain tools. The
advantage of cutting tools are that multiple tool copies exist for each tool type. The
tooling requirements of a given job may be fulfilled by any copy of a specific tool
type. With forming tools, there exists only one copy of each tool, and it is designed
for a particular product or job.

Thus, forming tools are both machine and job specific. A tool is used on one
machine and is used to process a single job type. Introducing machine-specific
tooling is a unique approach to production Simulation.

2) Finite Life: Recall that tool life is defined as the period during which the
tool yields a usable product (from placement into production until removal for
maintenance). Various measures for defining tool life include number of hours of
operation, number of jobs processed, or number of tool hits (McCall, 1965;
Gillimore and Penlesky, 1988). Finite life means that the tool is a resource which
may not always be available. Past research with either machine, labor, or a
combination of the two assumes that tools will always be available (Nelson, 1967;
Fryer, 1974; Treleven and Elvers, 1985). The worker or machine does not wear out
and is assumed to have a infinite life, whereas finite life tooling must be removed at
the end of its life. When the tool fails to produce a usable product or reaches a
point at which the risk of failure is high, it must be serviced. The removal for
maintenance differentiates the tooling resource from the labor resource. It does,

however, make tooling life Similar to machine breakdown.

1"

.A
.r

Hagen:
size and i
productio

T1
(Piers '* ‘1,
1001 $1515.

Mainline:

9

3) Renewal: The frequency and duration of tool renewal varies with tool type.
Cutting tools have a relatively shorter tool life and tend to require renewal more
often than forming tools. The renewal processes for cutting tools involves
Sharpening which is usually short in duration (McCall, 1965). Due to forming tools
Size and intricacy, renewal maintenance requires more time which complicates
production scheduling.

The renewal process involves either preventive or corrective maintenance
(Pierskalla and Voelker, 1976). Corrective Maintenance (CM) takes place when a
tool fails, (i.e. when it breaks or no longer yields quality products). Preventive
Maintenance (PM) occurs before tool failure. AS the frequency of PM is increased,
the frequency of CM decreases as Shown in Figure l-2a. The figure indicates a
linear relationship, but a nonlinear shape is possible.

PM is less costly and shorter in duration than CM (Sherif and Smith, 1981).
It has been found that PM helps provide quality outputs at a lower operating cost
than CM (Sutton, 1983). Figure l-2b illustrates maintenance costs (PM plus CM)
versus the costs of breakdown and associated penalties (Newman, 1985; Gallimore
and Penlesky, 1988). As the level of maintenance increases, total costs decrease via
lower failure costs. For each dollar invested in maintenance, failure costs drop by a
greater amount up to a point when it is no longer advantageous to invest in
additional maintenance. The intersection of the maintenance and failure cost curve
is the point of lowest costs. Past the intersection, the costs and time consumed by

maintenance exceed the benefits of reduced failure cost and down time. The

10

 

Figure 1-2a Comparison of Maintenance
Policies and Frequency of Occurrence

ENQUIRY

INNTENANCE

 

 

‘7

mm OF
CORRECTIVE
MAINTENANCE

 

 

 

Figure 1-2b Comparison of Maintenance
Frequency to Costs

Costs

 

 

V

Frequency
of

Maintenance

(W + W

 

 

a”:” 1 I f
11145:” 93¢.

13 PRO

T1167
manage opt
mmralnl I
described u
acorn?“en
assumed tha
prac’lilon 1'5
(Rmnna
as a sewarat:
that tools to
must be do
here is 3 var
job shop 111
influences it
in order 10;)
nature which
heuristic det-
giien he 100

Performance 1

11

marginal benefit of maintenance at this point is zero or negative.

1.3 PROBLEM STATEMENT

The research question which will be answered is: How do you effectively
manage operations in a dual resource constraint shop where the tooling
constraint has a finite life and where resource life and resource renewal are
described using stochastic distribution? Tooling has traditionally been viewed aS
a component of a machine. Machine specific issues examined in the past have
assumed that results are applicable to both tool and machine. Some researchers and
practitioners have come to realize that tools and machines are separate resources
(Brown et a1., 1981; Gray et a1., 1989). Therefore, this research examines tooling
as a scout; mg! constrained resource in addition to the machine resource. Given
that tools and machines are separate resources, methods to control each resource
must be developed to enhance production Shop performance. The model adopted
here is a variation of previous work conducted in a dual resource constraint (DRC)
job shop (Melnyk et a1., 1989). A DRC job shop has two limited resources which
influences its output. In this research, both tooling and machines must be available
in order to process a job. This work also focuses on forming tools and the unique
nature which affects the control procedures needed in the shop. The scheduling
heuristic developed here will attempt to provide good (not optimal) performance,
given the tooling attributes that cause system variance. It has been Shown that Shop

performance can be improved by reducing components of variance within the

system (51:
is Maker
to process
one requi:
using stoch
Wat

3713 301118 0

1.3.1 m
The
POSS: 1e. A
“‘0 rt‘Sour
either one
”1‘01 both

xhedflilng
&

12

system (Melnyk, Denzler and Fredendall, 1992). One component of system variance
is whether tooling is available when needed and whether the tool has sufficient life
to process the job in its entirety. The other component of system variance is the
time required to repair tools. The two forms of system variance will be modeled
using stochastic distribution which emulates the real world environment.

With this in mind, more specific questions can be developed. The following

are some of the questions from which hypotheses will follow.

1.3.1 How do wo ghg‘oolo jobs?

The objective of scheduling is to move jobs through the shop as quickly as
possible. At issue is how to process jobs when each Stage of the operation requires
two resources, machine and tool. Since jobs are both machine and tool specific, if
either one of these limited resources is unavailable, the job must remain in queue
until both resources are free. Figure l-3a illustrates the scheduling process. Job
scheduling must consider machine condition (availability and setup), jobs in queue,
and tool condition (availability and life). The rank order of jobs in the queue is
based on priority, which is determined by four factors:

Job Priority = f(Job Traits, Job Interrelationship,

Tool Condition, Shop Condition).

The question of how to schedule jobs, focuses on issues related to the job

with the highest priority. The determination of priority depends on certain

conditions and types of information. The following discussion looks at each of the

i iaczors oh:
160 trait
pmeessng time
hforrhation is 1
Show to impn
seseral intenie
due date dispa:
Whether segue:
101) int:

The interreiazh
miluirihg the s
dimitient job
Ch”its. Jobs
T001 Cl

life and 115k 0
M PTOCessihg
the Processing
mm"“iiiahcel,
11110113 MIC-I

deliiewl FIE

been addifigge‘

13

four factors which determines priority.

Job traits refer to attributes specific to each job, such as job arrival,
processing time, or job due date (Blackstone, Phillips and Hogg, 1983). The
information is used in setting priorities for dispatching rules, which have been
shown to improve Shop performance (Conway, Maxwell and Miller, 1967). After
several interviews with manufacturing companies, this researcher determined that
due date dispatching rules are the predominant selection criterion. This held true
whether sequence dependency was present or not.

Job interrelationship refers to traits common to a number of jobs in queue.
The interrelationship of interest in this study is sequence dependency, that is, jobs
requiring the setup (or tool) currently on a machine. The objective of sequence
dependent job processing is to reduce the amount of time consumed by setup/tool
changes. Jobs requiring the same setup/tool will be given higher priority.

Tool condition information help determine job priority on the basis of tool
life and risk of tool failure. The issue involves tradeoffs between tool condition and
job processing. Should a job be processed if the risk of tool failure is very high? If
the processing time for a job exceeds the tool’s estimated usable life (before
maintenance), should it be processed? If so, under what criteria? The tradeoff
involves whether to start processing the job with the possibility of delivering the job
on time versus the risk of tool failure, causing added maintenance time and missed
delivery. The consideration of tool condition in establishing job priority has not

been addressed in the literature.

14

 

Figure 1-3a Information Flows Involved
in Job Selection: Basic Model

 

 

JOB GUEUE

 

TOOL ROON

 

 

 

 

BELEO‘I’ION /

DECISION K

 

 

 

 

131$]

 

 

 

 

[:7]

WORK CENTER

 

 

 

 

 

 

Figure 1-3b Information Involved in
Job Selection: Decision Flow

SETTING JO. PRIORITY

 

.EOUENCE DEPENDENCY
YII

IET PRIORITY L” 1 no I

DIOPATCNINO RULE

 

clamor-1m. RULE
no" on tumor ALL JOII
o I '

I
N
GUPPICIEN TOOL LIP! —_ TOOL CONDITION
iVIO VII 1N0

SELECT JOI

YE.
CHANGE TOOL GETUP DELAY
IO

1: 1

PROCEED JOD

 

 

 

lit”
ghost): 1
the floor.
xrforrnar;

informant"

15

Shop condition refers to the status of the shop floor when determining
priority. The idea is to examine the total shop condition before releasing work to
the floor. Order Review Release (ORR) has been found to be beneficial to Shop
performance by leveling work load (Melnyk and Ragatz, 1989). The use of
information for determining due dates has been examined by Bertrand (1983) and
Baker (1984).

Figure 1-3b shows the decision process in setting job priority and selection.
The exact sequence of decision can vary, but for the purposes of this model

decisions will be made according to Figure 1—3b.

1.3.2 How oo wo Soheoolo tools for orgioction go maintenance?

This two part question is combined because the elements of production and
maintenance are interrelated. Both production and maintenance requirements
determine tool usage. More specifically, when to put a tool into production and
when to pull it out. Four factors influence tool usage:

Tool Usage = f(Tool Condition, Demand for Tool, Demand

for Other Tools, Maintenance Activity).

Tool condition determines whether a tool can continue in production. The
tool is either usable or not. If not usable, it is then classified as a tool failure. A
usable tool goes through different stages of life but by definition is still capable of
processing jobs. The issue is whether there is sufficient life on the tool to process a

job.

Der:
high prom:
term in u
occurs (job:

Ber
11 other H
different it:

mmou

agfiiist 1.5

aCCQUm a

16

Demand for tool refers to sequence dependency, that iS, whether there is a
high priority job in queue that requires the tool currently on a machine. A tool will
remain in use as long as there are jobs that require it, except when truncation
occurs (jobs of higher priority are placed in queue).

Demand for other tools arise when truncation requires a different tool setup.
In other words, production of one item Stops temporarily in order to produce a
different item and a tool change takes place. Tool usage in this case is based on
priority as compared to simple demand.

Maintenance activity relates to and uses information about tool condition. If
a tool fails and needs CM, then tool usage is terminated and the tool is sent for
servicing. Termination also occurs during PM. Pulling a tool from production for
PM is a cloudy issue. Strict PM policies dictate that maintenance occurs at a
specified time for a given tool, whereas less restrictive PM policies consider other
factors, such as demand or job priority. 1

Maintenance activity also affects tool usage if the maintenance queue is long,
which reduces tool availability for production. In PM scheduling, managers must
decide whether to pull a tool before its specified time if the maintenance queue is
short. That is, managers must weigh the advantages of a Short servicing period
against the disadvantages of unplanned downtime. The decision also must take into

account any other tools slated for PM.

1.3.3 How oxs vao'ation in tosouroe lifo and ronewal affgt the soheooling god

pupa.

1e.el0'

maize:
heunsti
Questio:
92st cor
em
1001 11:}
kids c

hfitfigg

1011
H191

,__g
{,4
J:-
l(/s

l7
assignmont ogisions?

System variation effects shop performance. Variation can be viewed in two
different ways: type of distribution (gamma, log-normal, etc.) and degree of
variation. The focus of this research will be on the degree Of variation on tool life
and maintenance service time. This will be accomplished by testing different levels
of variance for the specific distributions mean value both for tool life and
maintenance service time. The question of how job scheduling and tool control
heuristics perform under different conditions can be examined. The objective of this
question involves the robustness of control heuristic under different environmental
test conditions. It is not the objective to find an Optimal solution or Single best
heuristic. Certain heuristic performance may deteriorate faster or Slower with either
tool life or maintenance variance or their combination. Table 1—3 illustrates the
levels of variance that each heuristic will be subjected to. The goal in testing

heuristic robustness is to develop a framework for tool control.

Table 1-3 Relationship of Tool-Maintenance Variance

 
  
  
  
    
 

 

 

 

  

 

 
  

HIGH
MAINTENANCE MAINTENANCE
VARIANCE VARIANCE
r LOW TOOL LIFE VARIANCE LOW-LOW LOW-HIGH
HIGH TOOL LIFE VARIANCE HIGH-LOW HIGH-HIGH

 

1-3-4 W

The previous sections explored general questions and procedural issues

1.411

51111;;
met:
PIC):

1' n
I

~11
I

18

relevant to this research. More specific questions will examine how control
procedures compare under different conditions of variability. In particular,
1. Does additional information used when setting job priority for
scheduling affect shop performance?
2. Does preventive maintenance enhance Shop performance over

corrective maintenance?

3. How does various preventive maintenance policies influence shop
performance.
4. Does variance in tool life and maintenance time affect the

relative performance of different job priority and tool control

heuristics.

1.4 RESEARCH METHODOLOGY

The SIMAN 3.5 simulation package is used to model a flow Shop
environment with additional subroutines written in Fortran (Pegden, 1987). This
method is necessary because of the inability to control conditions in a real
production Shop. An analytical model would not provide the necessary complexity
required to simulate a DRC Shop (Nelson, 1966; Treleven, 1989).

The experimental factors to be examined in Chapter 4, are:

- Job Priority Rules

- Tool Control Policies

- Distribution Variance for Tool Life

19

- Distribution Variance for Maintenance Service Time

The research is organized into three phases. Phase one involves model
development of control rules and policies, based on interviews with several
manufacturing firms. Phase two validates and verifies the simulation model. The
run length and batch size will also be determined to assure independence and
normality at this time. Phase three involves full factorial design for the experiment,

including data collection and analysis for the performance measures.

1.5 RESEARCH CONTRIBUTION

This research examines an area that has received little attention. The model
developed extends past work in the DRC shop and on maintenance scheduling by
adding tool control. The model explicitly considers tool life and its variability in the
shop environment, an effect not previously explored.

Several researchers have pointed to gaps in the literature. Melnyk et a1.
(1989) described aspects of tooling that are ripe for research, including the need for
additional tool assignment rules and an analysis of tooling in a detailed shop
environment. Ghosh, Melnyk and Ragatz (1991) point to the need to examine
different tool traits, such as tool life. Browne, Boon and Davis (1981) emphasized
the need to consider availability of tooling, among other resources, in developing
scheduling rules. They found no research in the area of shop scheduling with a
tooling resource.

In Figure 1-4, the Shaded area of the diagram represents the least studied

.3
Oh

:ss‘

10!

 

 

20

issues. Except for a few articles, research on tooling in the DRC shop model is
insufficient. Only recently has the combined issue of FMS and tooling received
much attention. Reasons for the FMS-tooling interest is the need to increase
equipment utilization due to the expense of such an environment. Another reason
for the interest lies in the adoption of JIT and waste elimination. As stated
previously, when waste is curtailed, the effect of tooling becomes much more

prominent.

 

Figure 1-4 Spheres Of Research Focus

    
   
       
 

 

TOOLING

 

FLEXIBLE

   

AND MANUFACTURING

     
   
   
   
   
 

PRODUCTION SYSTEMS

CONTROL

DUAL
RESOURCE
CONSTRAINT

 

 

 

1.6 ORGANIZATION OF THE DISSERTATION
Chapter 2 examines past research, in particular the literature on tooling and

the issues unique to this resource. Reviewed are articles on tool life, scheduling of

tooling 111 '
Riorne'.
aDRC sh

C:
undersa'.
ESL-2871.5
deal
used in 11

mee's 1h:

follooec'

research

21

tooling in FMS and job shop environments, and the effect of tooling on Shop
performance. Also examined is similarities to other DRC research such as labor in
a DRC Shop.

Chapter 3 presents a conceptual framework for tool management. An
understanding of how tooling fits into production planning and shop floor control
establishes a foundation for this research. Chapter 4 and 5 lays out the model in
detail, including the hypotheses to be tested, the methodology, and the techniques
used in testing the research hypotheses. This includes determining whether the data
meets the required assumptions of various statistical techniques.

Chapter 6 presents the experimental results. This includes all statistical tests
followed by a discussion of the results. The last part of chapter 6 point to future

research directions and answers the questions developed in chapter 1.

2.1 1

he”;
L Lt.
5

101'

the

._-¢

I"

!

CHAPTER 2

BACKGROUND AND LITERATURE REVIEW

2.1 INTRODUCTION

The issue of tool control received little attention until such methods as J IT
began to eliminate buffers. The two buffers most relevant to tooling are inventory
and capacity. Excess tooling is both a form of higher inventory cost and an unused
capacity resource. With the increase in automation, the cost of under-utilization
increases (Mason, 1991), as is evident in FMS. If tooling is unavailable, automated
equipment cannot be used efficiently. Cost is another factor, as tooling can account
for up to 30 percent of FMS cost. Several articles have pointed to the need to bring
tooling into the mainstream of production control and research (Browne et al.,1981;
and Melnyk et al, 1989).

This chapter focuses on the major tooling issues, including characteristics
common to tooling and control. Recent studies address how tooling affects not just a
single operation, but the whole shop. Gray et a1. (1990) suggest the need to view
tooling in a broader perspective because of its effect on FMS. Gray et a1. (1990)
propose a hierarchy ranging from tool-specific issues to the interrelationships of
tools, scheduling, and system development.

The literature reviewed in this chapter addresses issues involving
maintenance, sequence dependence, and DRC job Shops. Much of this research

lends credence to the values selected for parameters in this study and helps

22

ntersmd 2'1
DRC 00 Silt
resoxees. l

oder st hat

ll TOOL

Figure 1.1,
fillings, Chg“
Clam11$. aut.
Cil‘JIES. and
“it“ prod
1131‘; format.

The:
three are us

071113? 162117;

mafillne (01
01 setup .100
Dies

23

understand how control procedures were derived. In examining the literature on
DRC job Shops, parallels and contrasts will be drawn between labor and tooling
resources. This will assist in understanding how tooling control was developed and

under what conditions.

2.2 TOOLING CHARACTERISTICS
2.2.1 Mao:

The varieties of tooling and their applications were summarized previously in
Figure 1-1, but was not all inclusive. According to Bloom (1967), tooling includes:
fittings, chucks, micrometer rules, patterns, models, template setup tools, spring
clamps, automatic ejectors, magazine feeders, steps and guides, slides, gauges,
chutes, and tool Sharpeners. These are just some of the different tools that are part
of any production shop. The main components of tooling are those used in the direct
transformation process, such as cutting tools, dies, and forms. The most common
forms of tools are defined below.

There are power tools, hand tools, cutting tools, and setup tools. The first
three are used to remove material from a product (chip removal). Examples are:
drills, reamers, rasps, mill cutters, and grinders. Setup tools are used to prepare a
machine (or power tool) for production. There are innumerable types and varieties
of setup tools.

Dies - are metal patterned blocks that shape material through a Stamping or

vacuum process. Stamping dies typically are used to form metal parts and may

 

f

'8-
1 ILI

moi.

SD

J

24

involve a punching or cutting operation. Vacuum dies are used to draw flexible
materials such as heated plastics over the die’s pattern.

Molds - Similar to dies, are used in forming a product. Injection molding
forces molten plastic into molds with a specific pattern. Once the cavity within the
mold is filled, the mold is cooled and the plastic solidified making the part(s).

Gauges - are used to measure some aspect of a manufactured part. Gauges
are a common component of quality inspections. Gauges are used to determine
whether parts meet tolerances.

Fixtures - are used to hold parts on a machine during processing.

Jigs — are used to hold and guide tools during the cutting/processing of a part.

Deis (1983) defines tooling as specific to a particular task. He suggests the
same basic breakdown of tooling types and goes further to define tools as either
special or general in their applications. The specialization or generalization of a tool
is based on the breadth of the tools application. For example, Special purpose tools
are used for a specific task or order, whereas general application tools are more

flexible in how, where, and on what it is used.

2.2.2 Tml Lifo

2.2.2.1 m

A tool can be kept in production so long as it is capable of making parts that
meet quality standards, a functional view first expressed by Taylor (1907). Tool life

is related to machining economics, which correlates processing speed with rate of

tool Rea
5" 9.3:"
L1; I~I&\
cutting c
all 11 C73
1719.101 Cf

fail tbre

life. As

1001 re:
01'" "ig
(ilfi‘téfigj
PrOduc:
necesga
Product

511017, :-

Shi'per
fails. A
“"1131 3
5061156

lInclude?

25

tool wear. An increase in cutting speed shortens tool life. Cook (1967) elaborated on
this relationship by including cutting depth and temperature in the equation. As
cutting depth increases, tool wear increases, and tool life is shortened. Furthermore,
an increase in either cutting depth or speed raises the temperature, which can have a
major effect on tools. High temperatures make tools more brittle and cause them to
fail (break) more quickly. Cook also noted that tool vibration (chatter) affects tOOl
life. As chatter increases, tool life tends to decrease.

Cook (1973) defined several determinates (criteria) for the length of time the
tool remains in service. The first criteria, tool failure, consists of a fracture or
breakage which makes the tool incapable of cutting. The second criteria involves
dimensional tolerances. In this case, when a tool is no longer capable of maintaining
product quality, it is no longer used. A tool may not be able to maintain the
necessary tolerances long before it actually fails. The third criteria relates to the
product surface. If their are abnormalities on a product surface, faster tool wear and
shorter life may result.

Cook’s final criterion relates to economic concerns. If a tool can be
sharpened (reground), it may be advisable to remove it from production before it
fails. An estimated average cost per cutting edge can be developed to determine
what a tool’s life should be. Given all these variables, tool life is not an exact
science where predictions are accurate. Cook’s (1973) basic formula for tool life

includes several major variables.

T=AV‘Bt‘Cb‘D

WI".-

{6564

00m

h.)
r 4

Cool
COnd
idem

in th;

He“

26

Where: T = tool life (min)
V = cutting speed (ft/min)
t = feed rate (in/rev)
b = depth of cut (in)
A,B,C,D = constants

This equation applies to cutting tools, about which there is extensive
research. No similar work has been done on tool life estimations for forming tools.
The reason lies in forming tools variation, complexity, and environment.

Based on Deis’s (1983) definition, forming tools are classified as special
purpose and tend to be complex. AS the number of cutting edges, angle of bends,
and number of parts (nuts and bolts) that compose a tool increases, the life of the
tool decreases (Deis, 1983). This is based on the notion that as the number and

complexity of parts on a product rise, so does the risk of product failure.

22.22 W

Initial research used deterministic distribution to estimate the life of a tool.
Cook’s (1973) equation has a set of parameters representing the environmental
conditions encountered by a tool. The assumption is that if a cut is repeated under
identical conditions, then the exact same tool life will be obtained for each tool used
in that operation.

Fenton and Joseph (1979) argue that a deterministic tool life gives a distorted

view of machining economics. Optimal policies under this assumption do not hold

Fenton;
ameth
adhto
controlh
mhnbm
daehp

NMCyh

hehuh

“XSMe

8amedlj

0%,“,

27

true when stochastic tool life is used. Simulation results using stochastic tOOl life
distributions Show lower production and profits than with deterministic tool life
distribution. The stochastic distributions tested include: normal, uniform and
Weibull.

Bao (1980) studied multiple tool operations and drew the same conclusions as
Fenton and Joseph (1979). Bao’s research entailed having 2 - 6 tools operating at the
same time versus a single tool. The probability of work stoppage is greater with a
multi—tool operation because there are more tools that can fail. The operation is
controlled by the tool with the greatest wear rate. Using three different stochastic
distributions (log-normal, Weibull, and gamma) a dynamic programming model was
developed for determining tool replacement policy. The results Showed that the best
policy is to replace all tools when the first tool wears out.

Several articles have analyzed a metal cutting process with multiple tools
(Ramalingam, 1977; Ramalingam, Peng, and Watson et a1., 1978; Ramalingam and
Watson, 1978). The researchers found that tool failure was characterized by a
Weibull distribution when single tool failure occurred. A gamma distribution was
observed when multiple tool failure was presented, and a log-normal tool life was
possible under certain conditions.

The distributions observed for cutting tools also apply to forming tools. The
same distribution is also found in machine failure as between the two tool types

(Lie, Hwang, and Tillman, 1977).

 

sated AS]

3? ‘fit. loo

dezeminazg
2'15 more {1
03211111131; 1
MEI We
0f iipiacinc:
”8111:311vi

ConSllmed.

28
2.2.3 Tmling Economios

The economics of processing is determined by two factors, tool life and
speed. AS processing speed increases, production output rises resulting in additional
profit. Tool wear also increases, causing shorter tool life, frequent replacement, and
higher costs.

A number of articles have examined this relationship (Hitomi, 1976a, 1976b;
Levi and Rossetto, 1978; Rossetto and Levi, 1978; Ravignani, Zompi, and Levi,
1979; and Bon, 1980). An important component to these articles involve the
determination of tool life. The more variable (stochastic) the distributions selected,
the more frequent the need to replace tools and the lower the profit. When the
machining environment involves multiple tools simultaneously, the replacement
strategy worsens. The best policy is to replace all tools, when one fails. The benefit
of replacing all tools Simultaneously is that the total number of failures is reduced.
The drawback is that some tools are replaced before their entire processing life is

consumed.

2.2.4 Tooling Chmtog’stio Summm

Figure 2-1 provides a taxonomy of tooling characteristics. The first part of
this section defines the different types of tools and their basic functions (Bloom,
1967). Subsequent discussions centered on tool life. In particular, Cook (1973)
described how tool life is affected by the cutting environment which is composed of

cutting speed, depth, and temperature. Cook also developed a formula, based on

970C:

‘sori

0? 10
5011 I
i’lfi‘c

en‘s”

29

Taylor’s (1907), for calculating tool life based on these three main components.
When analyzing tool life and machining, both deterministic and stochastic
distributions have been used to model tool life. Fenton and Joseph (1979) argued
that deterministic distributions give an unrealistic view of the cutting environment.
Using stochastic tool life distributions give lower cost performance values than
deterministic distributions. Proper distribution selection also plays an important role
when looking at the economics of processing. Tradeoffs must be made between
processing Speed and tool replacement costs (Levi and Rossetto, 1978). The faster

work is processed, the more frequent the need for tool replacement.

2.3 TOOL SCHEDULING

This section examines the broad set of literature which looks at the allocation
of tooling resources. The tool scheduling process varies depending on the
environment (FMS, DRC, or MRP) it is examined under. The variation can be
attributed to Shop layout and its associated hardware. An element common to all

environments is that of tool characteristics, such as tool life.

2.3.1 WW

Tooling is a major concern in a FMS environment because it accounts for
25 - 30 percent of the fixed and variable costs (Ayres, 1988). Kiran and Krason
(1988) point out that tooling has a major effect on FMS performance and that on-

line monitoring of tool wear is needed if performance is to be improved. Gruver and

30

 

 

33: .33.? a EeuEiE-c
33:..- .o 5.2.22.3:
the: Eons-IE}.

80a: one

33: .332. a :85...

22:338..
r

8:: .32. .33
too: 3:3.—

0.3.5.5200

 

.000: com

Aasap:a «0 Eco—>33
.Oho: :0.- d 030.03...
“Ono: 2.0.003 d :5...
3650—6050: :32...—

oo_Eo:oou 9500...
7

Anna: Eva
:00: E003

won: .00...
_

 

33838820 .3...

>895me $33. 9500... TN 239”.

 

 

3.3.1233; (
1515 are 1.
11mm 11316
91 al. 1195
23:0 05131 1

31

Scaninger (1990) also agree that on-line tool monitoring is essential if the benefits Of
FMS are to be fully realized. The main advantage of on-line monitoring are reduced
down time and higher output levels (Kendall and Bayoumi, 1988). Articles by Gray
et a1. (1989, 1990) provide the most comprehensive examination of tooling issues
and past research on FMS tooling. They also divide FMS tool control into three
areas. These are shown in Figure 2-2, which is a hierarchical view of tool

management in FMS.

2.3.1.1 FMS goo Tool Chamotoristios

The major tool characteristics relevant to FMS are: tool life, cutting
economics, Standardization, and number and location of tools (data/information).
Tool life already has been discussed, and suffice it to say that the tool life equation
(Cook, 1973) and distribution issues (Ramalinjam, 1978) are the same in all cutting
environments. What is unique is that FMS can monitor on-line tOol wear and
communicates this information throughout the system. The system can react
automatically when a tool breaks or needs replacement (Turn and Tomizuka, 1989).

The second tool characteristic, cutting economics, is common to any cutting
environment. Primrose and Leonard (1986) looked at the trade off between tools,
materials, and labor costs in an effort to pinpoint variable processing costs.

McCarthey and Hinds (1982) considered demand due dates and processing
speed in an FMS. In their model, the machines in the shop are initially set at

maximum speed, and planned idle time allows process rates to be reduced so that no

 

”we.--“ __|

 

 

 

32

 

Figure 2-2 FMS Planning and Control Hierarchy

(Grey et el.. 1989)

FMS Syetorn Management

(Produetloe plenum. one tool eontrot)

1

Control a Schodullng of Work Centers

(The! loading. oleoereeet. refinement. and were eeouenoe)

 

Tool Control a Characteristics

"bot IIOe. eeonoetloe. eteederdlutlon. end Intonation)

 

 

 

idle time remains. The objective is to process all jobs to meet due dates at the
slowest processing time possible in order to limit tool wear and operating costs.
The third tool characteristic, Standardization, while applicable to any
‘ environment is especially important in FMS because of the high cost of cutting tools
(Ayres, 1988). By standardizing, fewer tools are needed, which means substantial
savings in inventory and control costs (Hartley, 1984). Group technology techniques
is one method proposed for finding tool commonality that leads to standardization
(Burbridge, 1975; Chang and Wysh, 1985). Dushin, Jones, and Lowe (1990)
developed an algorithm to find the smallest set of tools necessary to perform an
operation, subject to an FMS tool magazine capacity.
The last tool characteristic issue important to FMS deals with data (number
of duplicate tools and locations). Information is collected regarding tool wear,

number, and location. Tool data is necessary at subsequent levels for replacement

and tool
side SO

1938;.

machine

0‘)
"JJ
—‘

I J

33

and tool magazine loading. Tool breakage information can be communicated system
wide so that a possible alternate machine center can be found (Kendall and Bayoumi,
1988). The interface of data among all operations are necessary for tool delivery and

machine loading (Gaymon, 1986; Wick, 1987).

2.3.1.2 EMS ago Individoal Machine Control

At this level of FMS hierarchy, the individual machine is examined (Gray et
a1., 1989). A combination of tool characteristics (constraints) can be combined with
overall system control. At issue is tool loading and placement within the tool
magazine, work sequence, and tool replacement strategy.

The control of work flow is dependant on the loading of tools and job
sequencing. Tang and Demardo (1988) examined a single machine with a limited
tool magazine with known demand. The objective was to reduce the number of tool
changes prior to the start of processing.

Vinod and Sabbagh (1986) also looked at tool allocation to the tool magazine.
They considered an optimal allocation of spare tools. Because tool breakage is the
largest factor that decreases productivity, spare tools will thus increase productivity.
Vinod and Sabbagh proposed a closed queuing network optimization model which
determines allocation of multiple types of tools to machines.

The number of spare tools also relates to another machine level issue, that of
tool replacement. Replacement strategies that consider the variability of tool life and

machining parameter are thought to be more realistic (Bao, 1980; LaCommere,

of th-

its re

four
01 to
Pine

end

34

Diega, Nota, and Passabbabte, 1983). These models consider the option of changing
several tools simultaneously. This differs from past tool replacement models which
consider only a single tool and machine (Cook, 1966).

Sharit and Elhence (1989) explored a tool replacement strategy for an entire
system rather than a single machine. This model did not seek an optimal solution,
but rather, looked at the human and computer element of tool replacement. Sharit
and Elhence attempted to trade off the economic loss of tool replacement with that
of through-put time. Their tool replacement heuristic allows a tool to be replaced if
its remaining life is less than the job processing time.

A recent study by Amoako—Gyampah, Meredith, and Raturi (1992) examined
four alternative tooling allocation strategies. The first strategy used a bulk exchange
of tools per period. For each period a machine is given all the necessary tools to
process the jobs. This requires batching jobs according to the tools needed. At the
end of each period, the machine gets a new set of tools. One assumption of this rule
is that all tools have sufficient life. The second alternative tooling allocation Strategy
is referred to as tool migration. Tools are allowed to leave or enter a machine
throughout the period. If a job requires a different tool, then it is sent to the
machine. The third tooling allocation strategy is referred to as resident tooling. This
principle is based on group technology methods. The rule attempts to form clusters
of tool combinations at machines and permanently keep the tools at that location.
The last tool allocation strategy used a combination of bulk exchange and resident

tooling. Amoako-Gyampah et a1. (1992) simulation results Showed that rules one and

1010

221.
for

red

(its

011‘

35

four, which group tools so that job batching is possible, out perform the migration
(rule two) and tool clustering (rule three) rules. Bulk exchange performed best in

terms of both flow time and tardiness.

2.3.1.3 BMW

Tool system management at the upper level of the control hierarchy, seeks to
integrate the production planning system with tool control (Gray et a1., 1989).
Specifically, tool system management looks at tool inventory and scheduling.

Tool inventory is based on the number of tools and spares required.
Zavanella, Maccarini, and Bugini (1990) examined different replacement strategies
for tools with a stochastic life. The heuristic found to be most effective attempted to
reduce the amount of wasted or unused life of replacement tools. The heuristic
performs best with a limited“ tool supply and tool refurbishment delay.

Kouvelis (1991) sought to determine the optimal number of each tool type by
developing a two-tier planning/allocating procedure. The long-term aspect focused
on the optimal number of tools of each type and the short-term aspect attempted to
minimize tool switches and to balance workloads.

Production planning in an FMS depends on tool capacity, which is affected
by tool type and production part variety. Carrie and Perera (1986) explored how tool
and product variety influences tool changes and tool ware. Tool changes occur for
two reasons, tool wear and product variety. It was found that tool wear has a much

greater effect on the number of tool changes than does product variety. For this

pt
Ila;

T

99

e .

A‘

b\

11

36

reason, tool life was a more limiting factor on capacity than is product variety.
Thus, FMS planning Should take tool life into consideration.

Other planning issues related to FMS involve grouping parts and tooling for
effective production. Ventura, Chen, and Wu (1989) developed an algorithm for part
grouping and tool requirements. By grouping parts that require Similar tools, fewer
tool changes are necessary, and larger batching of production is possible. The model
also minimizes idle time, which helps reduce tool redundancy. Another model was
developed by DeSouza and Bell (1991), who used a Rank Order Clustering (ROC)
algorithm to group tools. The groupings are based on the job to be processed and
required tooling. The cluster algorithm reduces the number of tool changes (setup)
in the tool magazine. The model reduces the effort of managing and scheduling tools

in an FMS environment.

2.3.2 Itfmls Sohodoling in Non-FMS Maohino Models

Although most tool scheduling models are set in a FMS environment, a
number of other studies have focused specifically on tooling or incorporated it into
their framework. One such model is the single-plant mold allocation which assigns
molding tools to machines (Love and Vemuganti, 1978). The model attempts to
satisfy production demands while having limited tool capacity and changeover
restrictions. Tool capacity varies by period as new molding tools are added and Old
ones are reworked. The problem is formulated as a mixed integer program.

Another model which uses tooling as an element is by Brown et al. (1981).

37

They looked at production and sales planning for multiple periods with limited
shared tooling. In their model, forming tools (injection molds) were shared because
they could produce a number of similar products with the same dies. Each tool was
specific to a family of Similar parts. Brown et a1. formulated the problem as a mixed
integer linear program, with tooling as a constraint on each periods production. The
model determined how much of each product to produce and sell in each period.

Both Melnyk et a1. (1989) and Ghosh et a1. (1991) examined a single
machine with tooling availability. In their first model, Melnyk et a1. examined tool
availability and its influence on shop performance. Tool availability is determined by
the level of external demand (another machine) for each tool. A tool can remain at a
machine for Y number of jobs. AS Y decreases, tool availability decreases (tight tool
capacity) and all measures of Shop performance decrease. The implication is, the
lower the tool capacity, the worse shop performance becomes. Another contribution
from Melnyk et a1. (1989) involves tool assignment rules, which were found to be
more critical than job priority rules (dispatching). However, rules which consider
both job priority and tool availability performed best. Rules that attempt to avoid
tool changes (sequence dependent) perform poorly, except for tool change
performance measures. If the rule considers a job’s due date, its performance is
improved. It Should be noted that rules which attempt to avoid tool changes will
vary in performance depending on the setup time constraints.

The level of sequence dependency was tested by Ghosh, et a1. (1991) using

the same model as Melnyk et a1. (1989), except that various degrees of sequence

more

Oblctllt

38

dependency (severity of setup time between jobs) were added to the model. The
effect of higher levels of sequence dependency (increased setup time) caused shop
utilization to increase and shop performance to decrease. Other findings conformed
to past work (Melnyk et a1., 1989) which concluded that tool assignment rules are
more important than dispatching rule decisions. The best performance was Obtained

when both tool condition and job priority was considered together.

2.3.3 Summary of Tool Sohoouling Litoraturo

As Figure 2-3 shows, the literature on tool scheduling can be broken into:
FMS and non-FMS environments. The majority of tool scheduling literature falls
under the first group, FMS related. The reason for this is because of the high cost
of tooling in FMS (Ayres, 1988). Gray et a1. (1989, 1990) provide an in-depth
review and categorization of the FMS literature. Gray et a1. (1989) breaks decision
making into a three level hierarchy from tool magazine Size to tool
placement/ replacement decisions.

At the lowest level of the hierarchy there are a number of issues including:
tool life, cutting economics, tool standardization, and data/information. The last
issue, information, links and drives all three levels of the hierarchy. Information on
tool and shop condition is passed to higher levels of planning.

The second level of Figure 2-2 hierarchy involves Shop scheduling of work
and tool allocation. Tang and Demardo (1988) looked at tool allocation with the

objective of minimizing tool changes during processing. Amoako—Gyampah et a1.

39

 

 

20.3

:oExeo

:50an a .338.
._e .e 5:25

 

 

:en a euooMeo ..e .e geoEe>O-exeoE< 32.350
.3 .e 5.3....) eocefw a :Lego >023...
29.}. a eta-o ._e .0 205.30.... 2.3::
2.233. cacao-o .- no...) 3232 a eeo..Etn_
..e .e -__eee>neN etc-Eco a see... 3.3.th 4 Sop
EeEeoeceS .obooo 3:233:25
..e .e .325 Eeue>m .03 3:30 x3? .08.
.3 .e .3533 F _ _
..e .e ates
.232...» .- 26.. _
20002 mZnTcoz 20002 mi".
_ _
_

05.323 oc::oenom Eek

>Eocoxm._. mezzuecow .02. Tu 0.39m

 

 

40

(1992) considered four different tool allocation strategies and found bulk exchange
to be the best performer.

At the top of the hierarchy is the planning of production and tooling. Ventura
et a1. (1989) developed a means of grouping jobs so as to reduce tool changes. Also
considered at this level is the number of tools necessary for the system to meet
demand (Kouvelis, 1991).

For non-FMS models of tool scheduling, articles by Melnyk et a1 (1989) and
Ghosh et a1. (1991) are the most relevant. Both look at tool control procedures
simultaneously with dispatching rules. Other models by Brown et a1. (1981) and
Love and Vemuganti (1978) viewed tooling as a capacity issue and part of the

production planning process.

2.4 MAINTENANCE

The objective of any maintenance program is to transform equipment
(machines and tools) into a useful capacity. Machines and tools move through
several conditions over time with the probability of failure varying at each stage.
The final state for any equipment is failure. A maintenance program must consider
the varying conditions of the equipment. Equipment maintenance can take place
when one of two conditions exits: l) pre-failure, or 2) post-failure. Pre-failure
maintenance is usually referred to as preventive maintenance (PM). Post—failure
maintenance is referred to as corrective maintenance (CM). Pre and Post failure are

examples of two maintenance strategies, PM and CM.

"J
h

‘7?»

n
..»L.

nim-

41
2.4.1 Maintenmge Strategies

Maintenance strategies are usually grouped into two categories: 1) reduce the
frequency of failure, and 2) reduce the severity of failure (Hardy and Krajewski,
1975; Krajewski and Ritzman, 1988). Preventive maintenance (PM) would be
classified under the first category. Backup or equipment replacement would fall
under the second category and is usually referred to as preparedness maintenance
policy.

Gallimore and Panlesky (1988) defines five maintenance strategies: 1)
Reactive, 2) Preventive, 3) Inspection, 4) Backup, and 5) Upgrade. A reactive
maintenance strategy is identical to CM. Preventive and Inspection is usually
grouped under PM. The difference is, preventive maintenance is based on a regular
schedule while inspection maintenance is irregular and performed when the tools are
being used. Backup is a maintenance strategy based on the availability of redundant
equipment. Such an approach is justified when the cost of equipment breakdown
exceeds the cost of having excess capacity. The last maintenance strategy involves
the upgrade of equipment. With newer equipment, breakdown frequency diminishes,

resulting in less costly maintenance.

2.4.2M'nn M l h riis
2.4.2.1 Classification

A composite of breakdown and maintenance models can be found in a

number of different articles (McCall, 1965; Pierskalla and Voelker, 1976; Sherif and

SW
A!

de

f)

42

Smith, 1981; and Lie at el., 1977). Breakdown is relevant because it often
determines when maintenance activities take place. McCall (1965) conducted the
initial survey which identified the assumptions and relationships found in various
maintenance policies. This initial survey attempted to introduce the problem of
scheduling maintenance when equipment experiences stochastic failure. Figure 2-4
provides a breakdown of McCall’s various maintenance polices. McCall’s (1965)
two categories, known and unknown distribution of time to failure, describe
different approaches to maintenance scheduling. The following is a description of the
preventive or preparedness maintenance policies.
1). Periodic Policy: replace or inspect equipment at the time of failure or interval
(age) N, which ever comes first.
2). Sequential Policy: next inspection interval is recalculated just after each
maintenance action.
3). Opportunistic (multiple part - complex) Policy: when one of several component
fails or interval N, which ever comes first. At that point in time all components are
inspected. Each part has a stochastic life. All three of these policies, whether for
preventive or preparedness, assume a known distribution for mean time to failure
(M'I'I‘F).
The following assumptions are utilized by all three models.

a. The system is either operating or has failed.

b. Failure is an absorbing state, partial operation is not possible.

c. Maintenance action renews the system immediately after completion.

43

 

Figure 2-4 Maintenance Policies
(McCall, 1965)

l

 

 

l I
KNOWN DISTRIBUTION UNCERTAIN DISTRIBUTION
OF TIMES TO FAILURE OF TIMES TO FAILURE
I
F l
PREVENTIVE PREPAREDNESS ._ MINIMAX
MAINTENANCE MAINTENANCE
BOUNDING
TECHNIQUES

—— SEQUENTIAL __ _ ADAPTIVE
h— MULTIPLE PARTB _

 

 

 

 

 

 

d. The interval between successive renewal points are independent random

variables.

e. Maintenance costs or time penalty is higher if done after equipment failure

rather than before.

Sherif and Smith (1981) provide information on which policy is optimal
under various assumptions.

- For unlimited life systems, select the periodic policy.

- For systems with constant failure rates (exponential), maintain at failure.

- For systems with increasing failure rates (Weibull & Gamma), maintain a

progressive schedule.

- For systems with a finite life, select the sequential policy.

- For complex multi-part systems, if:

1. Parts are independent: choose periodic or sequential scheduling for

di

44

each part.
2. Parts are not independent: replace all parts when one fails.

McCall’s second category develops maintenance policies when the
distribution of time to failure is unknown. The following is a description of these
three maintenance policies.
1) Minimax: minimizes the maximum maintenance losses, whether it is costs,
downtime, or both. Nothing is known about a system’s failure distribution. The
optimal policy is to maintain at failure.
2) Bounded: partial information on the distribution is known (failure rate).
Chedyshev-type bounds are applied to one of the models previously discussed.
3) Adaptive: if subjective information exists about failure distribution, then the
Bayesian adaptive techniques are used.
Based on available information, either preventive or preparedness policies would be
selected and modified using one of the three techniques mentioned above.

Pierskalla and Voelker (1976) extended McCall’s work and adds maintenance
policy classification as either a discrete or continuous system review. Most of
McCall’s classification would fall under the discrete time model. Pierskalla and
Voelker (1976) separated the continuous time model, which attempts to minimize
costs, into three areas which are presented below.
1) Age dependent: this is a modification of the periodic and sequential maintenance
polices which consider critical threshold costs. Once beyond this cost, it is

advantageous to replace the equipment.

45

2) Shock: this model assumes that failure occurs due to an external shock to the
equipment. External shock occurs based on a Poisson process and have an
accumulative effect.

3) Interacting Repair: this is another opportunistic policy which includes
cannibalization, multistage replacement, and variable repair rates. Cannibalization
attempts to maintain equipment based on parts from an identical unit. The objective
is to provide the best possible configuration for operating equipment given no spare
parts (units).

Multistage replacement differs from cannibalization in that a new part is
always available. The objective with multistage replacement is to place the spare
part where it yields maximum benefit. This tends to be where failure costs are
highest.

Variable repair rates add the element of maintenance capacity as a decision
variable. Maintenance capacity is usually expressed as a service rate (that is, the
number of workers performing the service). The objective is to find a service rate
which minimizes the long run costs.

Sherif and Smith (1981) also discuss deterministic maintenance models and
assumptions. Under deterministic models, equipment life is known with certainty.
The optimal maintenance policy is periodic with equal length maintenance actions.
This is based on the following assumptions.

a. Outcomes of maintenance are non-random.

b. Maintenance restores the system to original condition.

faiiu
(Lie

dist:

COiil
USE
Wm
SjSil
suit;

ram

non
nom

Shor

46

c. Failure is observable and instantaneous.

d. Identical parts have the same known time to failure.

2.4.2.2 Ejg'lere egg Service Time Distributions

Most models that have examined the two components of maintenance, time to
failure and service time, have modelled the time duration as a stochastic process
(Lie et a1., 1977,; and Sherif and Smith, 1981). Several different types of
distributions have been used to model failure time (MTTF) including: exponential,
Erlang, Weibull, Gamma, Rayleigh, normal, log-normal, uniform, extreme value,
and general. Negative exponential is the most commonly employed distribution
because of it’s constant failure rate. While mathematically easier to use, data
collected from industry support the exponential application (Lie, et a1., 1977). The
use of normal distribution is justified based on data from the aircraft parts industry.
When the failure distribution is skewed, the gamma family is a better choice. For
systems characterized by fatigue failure, like tooling, the Weibull distribution is
suitable. Log-normal is not considered a good choice for mean time to failure, but
rather, fits better as a repair time distribution.

Distributions used to model maintenance renewal time also cover the same
group as mean time to failure: exponential, Erlang, Weibull, gamma, Rayleigh,
normal, uniform, and general (Lie et a1., 1977). The preferred distribution is log-
normal. Exponential is considered a good choice when there is a high frequency of

short repairs with a few long repairs. If repairs (renewal) of each item takes an

47

equal length of time, then uniform distribution is an appropriate choice.

A few articles use unique distribution to model failure (Sherif and Smith,
1981). One of the few research articles which addresses the issue of tool failure and
maintenance is by Vanderhenst, Van Steelandt, and Gelders (1981). The objective
was to minimize tool down time. The tool life or time to failure used in this model
was based on historical data which resembled an exponential distribution.

Denzler et al. (1987) modeled an FMS with breakdown uncertainty by using
a deterministic approach. Breakdowns are classified as either major or minor. Major
breakdowns occur once every ten shifts and require ten hours to perform the
maintenance. Minor breakdowns occur once every two shifts for a maintenance
duration of two hours. Deterministic distribution simplifies the model but tends to

over estimate the benefits of scheduling policies.

2.4.2.3 Maimenagee age Tml Availability

The frequency of tool failure and the length of time it takes to repair the
tool, determines the tool’s availability. It should be evident that tool availability
determines system performance. Lie et a1. (1977) classifies availability into three
(Markovian) categories: 1) instantaneous, 2) average uptime, and 3) steady-state.
Each of these three categories look at a different time intervals when estimating
availability. Another means of determining availability is with a ratio of uptime to
total possible time, which can be measured directly or estimated using the expected

value function (Goldman and Slattery, 1964). Two other methods of expressing

48

availability is with inherent availability:

: MTBF
1 MTBF+MTTR

 

where: MTBT mean time between failure

MTI'R = mean time to repair

and achieved availability;

A _ MTBM
a_____
MTBM+M

where: MTBM = mean time between maintenance
M = mean maintenance time of both CM and PM
The importance of the availability measure stems from the fact that it gives a
means of evaluating or comparing system performance. When different distributions
are used to test systems performance, the same expected value for availability allows

for a more accurate comparison.

2.4.3 WWW
2.4.3.1 W

There are a number of different models that provide a means of
implementing and controlling a maintenance program. Bojanowski (1984) used the

Materials Requirements Planning (MRP) logic to develop Service Requirements

49

Planning (SRP). SRP attempts to establish routine equipment inspections and
monitors wear to prevent machine failure and improve shop performance. By time
phasing maintenance inspections, plans for repair labor and materials can be
determined. Bojanowski (1984) estimates that 70 percent of machine failure is due to
either the lack of awareness of a need for service, or lack of proper service
inspection interval. Bojanowski also stated that equipment failure is especially
common for high wear parts like forming tools.

A maintenance model proposed by Newman (1985) also used MRP logic. He
states that by using a periodic planned inspection, the preventive maintenance
program could reduce the risk of machine failure. Newman (1985) goes farther than
Bojanowski ( 1984) by adding a master maintenance schedule to the preventive
maintenance requirements planning (PMRP). The master maintenance schedule
determines when a service activity needs to occur. For example, service is
performed after X number of products are produced, after a certain number of
operating hours, or when mean time between failure is reached. By selecting one of
these values, the point at which to preform preventive maintenance is determined.

A recent article by Maggard and Rhyne (1992) presents an integrated model
of maintenance. Total productive maintenance is an attempt to integrate all functions
with maintenance, especially production. The benefits obtained with this approach
resulted in a 6 percent increase in machine availability.

A case study by Christer and Whitelaw (1983) examined the benefits and

requirements of a maintenance program. They point to the critical need for historical

50

data and the ability to collect data continuously. Information on the causes and
consequences of machine failure helps prevent future failures. A maintenance
program should not only help eliminate failures, but provide an appropriate PM
schedule. Christer and Whitelaw’s estimate that breakdowns account for up to 20
percent of lost production time.

A subsequent article by Christer and Waller (1984) examined PM applied to
a vehicle fleet, in particular the timing and frequency of PM. A unique feature of
this model is the concept of delayed PM. If a vehicle breaks down, the part that
failed will be repaired, but should other parts that show wear be replaced at the
same time (a form of PM)? If not repaired, should the vehicle be rescheduled for
PM? Christer and Wallertesearch showed the complicated issues involved in
maintenance when multiple parts are present which may be interrelated. Their model

demonstrates that PM must be tailored to each situation.

2.4.3.2 MW

Vanderhenst et al. (1981) explored PM and CM strategies for tools. If
preventive maintenance takes place when ever a changeover or tool change occurs,
then little or no production time is lost because of these conditions. If, however, a
tool or machine should fail, then unplanned maintenance take places and system
availability decreases. It is estimated that availability loss due to breakdown is 8
percent. Depending on the penalties for CM, Vanderhenst etal. (1981) explored the

trade offs between tools during changeover (setup) and tool changes due to tool

51

failure.

Kay (1978) examined whether PM is more effective than CM. The time to
system failure is modeled with a Weibull distribution (where parameter b is
l<b<2.5). The objective is to optimize the percentage decrease in maintenance
costs. It is found that optimal cost performance does not equate to maximum
availability, which is contrary to many other studies. In addition, when its cost is
lower PM is preferred over CM because availability is greater. The benefits
associated with PM or CM depend on the values given to both costs and
availability.

Banerjee and Burton (1990) simulated a job shop and examined a number of
different maintenance strategies and maintenance capacity issues. Maintenance
capacity is determined by the number of workers and maintenance allocation rules (6
rules). Corrective maintenance is always given priority over preventive maintenance.
As for PM, the five rules are: (1) No PM performed, (2) PM every .3 periods of
operation, (3) PM every .5 periods of operations, (4) PM every 1 period of
operation, and (5) PM every 1.5 periods of operation. Mean time between failure
was modeled using an Erlang-4 distributions (MTBF mean varied from 70 to 130
hours). CM repair time was modeled with an exponential distribution (mean from 2
to 8 hours) and for PM repair time a uniform distribution was used (average time
from 1 to 4 hours). The findings show, that as the frequency of PM increases, the
average PM delay decreases, and flow time increases. In addition, as shop utilization

increases, the maintenance scheduling method used takes on greater importance.

52

This is logical, since down-time is more detrimental to shop performance.

2.4.4 PMuetien aad Mg'ntenagee Sehedeling

Few articles have considered the implications of maintenance actions on, or
in the presences of production schedules. Ram and Olumolade (1987) developed
maintenance schedules that consider the production plan, and a total expected costs
formulation that considers capacity per period, given the probability (Weibull
distribution) of machine failure. The model also included the average maintenance
time for PM and CM. The objective was to minimize the costs of PM and CM on a
per period bases. The model does not find an optimal cost, but simply a means of
evaluating the cost tradeoffs between CM and PM.

Pate-Cornell, Lee, and Tagaras (1987) developed combined maintenance and
production rules. The maintenance policy includes: no PM, scheduled maintenance,
and maintenance on demand during inspections. The latter policy is performed using
the production inspection process. If inspection indicates a need for maintenance,
(poor product quality) then maintenance is performed. This policy works well when
the shop is stable. In an unstable environment, the scheduled maintenance policy
works best. Planned maintenance allows the system to adjust production around
maintenance and machine availability.

An FMS model with machine failure was developed by Denzler Boa, and
Duplaga (1987), to determine the effect of machine failure on system performance.

Machine breakdown is classified as either minor or major, which is associated with

53

maintenance severity. Breakdowns do affect performance in a FMS, but short term
production scheduling procedures can lessen the impact of the breakdown.

Hsu (1992) examined how PM influences a production system and showed
that the optimal maintenance policy is very sensitive to the system’s monitoring
technology. Monitoring technology includes vision systems and torque sensing
equipment. Should such systems exists in the shop, it is optimal to perform
maintenance only when the monitoring equipment so indicates. If such monitoring
systems do not exist, then the model found that planned PM can improve system
performance.

The effect of maintenance on production was also analyzed by Wacker
(1987). Wacker looked at the elements of production throughput time of which
down-time is a component. He proposed that a truly effective PM program would
not add to a systems throughput time because it would be performed during setup.
For zero inventory (Hall, 1983) and JIT systems, such an approach is advocated, but
in reality the system is not likely to be this efficient. Variation in duration and
frequency of PM can also add to throughput time. Unscheduled maintenance,
whether CM or PM, will cause queues to increase and an increase in throughput
time. The objective of PM in reducing throughput time is to decrease unplanned
maintenance (CM) while not causing undue delays in production startup after a
setup change.

Finally, Ghosh and Gaimon (1989) considered as part of their model the

effect of various levels of PM on capacity. They found that PM can slow the

54

deterioration of machines, which can alleviate bottlenecks, and that higher levels of

PM can provide higher levels of capacity.

2.4.5 WW

Figure 2-5 provides a breakdown of the literature reviewed in this section.
The figure does not show that maintenance tends to be categorized as either
preventive (PM) or corrective (CM). Maintenance can be further broken down in an
attempt to either reduce frequency or severity (Hardy and Krajewski, 1975). Further
refinement of maintenance strategies can be found in Gallimore and Penlesky
(1988). While these articles are descriptive, more analytical models were described
by McCall (1965), whose maintenance scheduling approaches were outlined in
Figure 2—4.

Failure and service time have been modeled in a variety of ways. lee et a1.
(1977) and Sherif and Smith (1981) describe many of the approaches used in
distributions. The most common stochastic distributions used for failure and service
include: exponential, Erlang, Weibull, gamma, normal, log-normal, and uniform.
Stochastic distributions are particularly relevant to tool availability. Methods of
calculating availability is based on mean time to failure and expected service time.

The scheduling of production and maintenance is presented using both
descriptive and analytical models. In the descriptive model, Bojanowski (1984) and
Newman (1985) used MRP logic in the planning of maintenance. For the analytical

models, a number of different PM procedures were examined and compared to CM

55

 

 

b0...)» a .2220

._o .0 3.5.00 ..u .o a:
5....” 5.5m .- 2.5:”

 

cotam C onto—=0 oer-E I Econ.) d 2.0..” ..S:oo> w
x5. 55302 ._c .0 0... 2.5.20...
..I .0 .2.ofou=u> Into—:75 ..a «I a... .300!
Inez—5.320
Iona! c.0002 >2:36:52 oo_>.aw Evian. a
coE-o a 52.0 32.223 02.2302. 30.. a 95:3. co.«uo_..ooo_o
.383 — _ _
an:
._- .o .2230
..a .0 ..o:._oo-2:.
03.2.5.0 u E... 3.0322! 4 >93:
330.5; a
053—555 9.06:30
02.2.2562 a 95:30.3” 002235220
actuauok. 2.... use 20 d 22.02 00.02:...w
_ . _ .

 

2382... coca—.2522

>Eocoxm... 8:29:85. mum 9:9".

 

 

56
(Vanderhenst et al., 1981; Banerjee and Burton, 1990). The research found that

when the cost of PM was lower than CM, PM improved performance. Excessive or

frequent PM can also cause tool availability and shop performance to decrease.

2.5 DUAL RESOURCE CONSTRAINT AND LABOR SCHEDULING
MODELS

A DRC shop involves more than just a machine limited shop environment.
The DRC shop has two limited resources which must both be available before
processing of work can start. The simultaneous availability of two resources
differentiates this research from the single constraint machine limited studies which
dominates the literature. Review articles by Blackstone, Phillip, and Hogg (1982),
Day and Hottenstein ( 1970), Graves (1981), and Panwalker and Iskander ( 1977),
provide an excellent background on machine-limited research. Only Blackstone et al.
and Day and Hottenstein briefly discuss the DRC research. Treleven (1989) was the
first to provide a detailed review of past research in DRC articles.

In the DRC field, a number of issues have received attention, as indicated in
Figure 2-6. Treleven’s classification of the various DRC models will be applied in
this literature review. The two major issues which have been addressed are operation
and design.

It should be noted that most DRC models look at machine and labor as the
constraining resources. Melnyk et al. (1989) and Ghosh et al. (1991) are the

exceptions and consider tooling as yet another limiting resource. Hogg, Phillip,

57

 

 

..0 .0 20000!

 

 

 

 

 

._0 .0 coo: 0.02m 0 c233... 53.0....
=0E>4 0 .3502 .0»; 5...». a .3on Sim .05“. 0 9.003
.0 .0 022.90: ._0 .0 30:00! .3202 ._0 .0 Fiona! .0»... 4 3.00.5 Lg...
.3202 ..0 .0 can... :32 ..0 .0 can... .320: .5202
2.0x .333 3.3.2.5 2:332“. 005000.... :25! 00.3.. 0.2.! 00:... cost.
. . _ . _ _
. _
.0504 .coE:u_00< 005000...
35.50
.02... 0 3.00! ..0 .0 2.2.00:
:0ue0u0fi .03... .- 5.003
>20u 0 50.0.9... 0300! .EE
Era ._0>_.u 0 3.00! 50.02
.2uc00 Eggs—:0...- O—CO 0:0 00-3: 05:03:20
p _
00:... €2.00 00...... .0co_.a.oa0
T _ _
0.0.52 can.

:23on 0mm .0 >Eocoxm.. mum 9:9".

 

 

58
Maggard, and Lesso (1975a, 1975b) discuss the possibility that tooling can be a

constraint, but they do not directly address any tooling issues. Instead, what Hogg et
al. developed was a multiple constraint model which allows for a third constraint

like tooling.

2.5 .1 gmgatieaal Issees in DRC

Operational issues in DRC models look at control procedures such as
dispatching rules, due date assignment, and labor allocation. These procedures are
enacted while the model is in operation. They determine how and when decisions in
the DRC model function. Operational issues are dynamic in that procedures are
enacted when certain conditions are present in the model. What separates the DRC
models from the machine-limited research is that while both are concerned with

dispatching and due date rules, the DRC model must also decide on labor allocation.

2.5.1.1 Wis;

Early DRC models examined various dispatching rules to see how multiple
resources alter performance. LeGrande (1966), Nelson (1967, 1970), Fryer (1973),
Rochette and Sadowski (1976), and Weeks and Fryer (1976, 1977) found that
dispatching rules had a significant affect on shop performance. Nelson found that the
shortest operating time gave the best results in terms of mean flow time, while the

first in system was better with respect to variance of flow time.

59
Weeks and Fryer (1976, 1977) found that the relative performance of various

dispatching rules depends on the tightness of the due date procedure. When first
come first served (FCFS), shortest processing time (SP1), and least slack per
remaining operations (SOPN) were tested with tight due dates, SPT was the better
shop performer. As due dates were loosened and limited labor transfers, SOPN
became the best performer.

LaGrande (1966) and Bulkin, Cooley, and Steinhoff (1966) used actual data
in a simulation to determine whether dispatching rules have a major influence on
performance. SPT (MINPRT) was ranked first, minimum slack (MINSOP) ranked
second, which is consistent with Nelson (1966). The ranking was based on an
average of ten performance measures. When the weight of the ten performance
measure were adjusted to favor early job completion, MINSOP became the best
performer. Bulkin et al. (1966) applied the MINSOP dispatching rule to an actual
operation, orders completed on time rose 10 percent, and both machine and labor

utilization increased.

25.12 W

Most DRC models assign due dates based on the total work content rule
(TWK) described by Conway et al. (1967). As stated previously, Weeks and Fryer
( 1976) examined both dispatching rules and due date assignments. They found that
the due date assignment procedure (TWK) has a significant effect on shop

performance. Due date has the greatest influence on the lateness performance

60

measures. The tightness of the due date assignment has a more important effect on
sh0p performance than do dispatching rules. A subsequent study by Weeks and
Fryer (1977) examined the effect of the K value used in the Total Work Content
(TWK) due date assignment rule. The objective was to determine the minimum cost
due date multiplier (K) value that enhanced shop performance. They found that the
value selected for K (in TWK) was dependent on cost structure, dispatching method,
and labor assignment rules.

Weeks (1979) further analyzed due date assignment rules in relation to shop
conditions. Seven different rules were tested, three of which used TWK with
different K values. Those rules that include information on shop congestion or job
flow time provided more predictable due dates than previously tested TWK methods.
In addition, it was found that dispatching rules which incorporate due dates (least

slack) perform better than process oriented rules (SPT).

2.5.1.3 WM

Labor assignment is a major factor in all the DRC models. The two most
important issues are when and where labor is assigned. Other relevant issues are;
which worker to select and whether the labor decision is centralized for
decentralized. The latter issues will be addressed under information control.

The decision about when to transfer workers has been shown to be a more
important labor assignment issue than where to send the worker. The when decision

determines the eligibility of the worker to be transferred. Nelson (1966, 1970)

61

examined the effect of cross training and the when labor transfer. He found that the
level of cross training, or the ability to move between machines, has a major
influence on shop performance, as does the when labor assignment. The decision of
when to move a worker was based on shop and cross training information.

Fryer (1973, 1974a, 1974b, 1976) and Weeks and Fryer (1976) showed that
the when rule had a significant effect on shop performance. Fryer (1973) found that
when rules, both intra and inter-divisional, were more important than where labor
rules. Fryer also concluded that the when labor rule was more influential on shop
performance than the dispatching rules. From the various studies by Fryer, it was
determined that "when idle" (all jobs in queue are done) labor transfer rules were
consistently better performers.

Treleven (1987) did a more complete comparison of when labor rules by
examining: when idle (QUE), when the current job is done (JOB), and when to pull
worker (PULL). The last rule, an attempt to allocate a worker to. areas of need, is a
combination of where and when to relocate workers. The PULL labor rule
outperformed the other two when rules.

The fact that Treleven’s (1987) PULL labor rule combines issues of when
and where to send workers presents a paradox in the literature. Nelson (1967)
proposed that where labor rules could improve mean and variance of flow time.
Studies by Fryer (1973), Weeks and Fryer (1976), and Treleven and Elvers (1985)
showed that where rules had little effect on shop performance. Treleven and Elvers

examined several where rules and found that they had no significant effect, except

62

on the number of labor transfers. Holstein and Berry (1972) found that the where
labor rule did have an impact on shop performance. The where rule reduced the
number of transfers without greatly increasing flow time. Where to send workers is
based on the longest queue in the shop. Holstein and Berry‘s where rule is similar to
Treleven’s (1987) PULL rule which seeks to combine when and where labor rules.
Melnyk and Lyman (1991) supported this approach by showing how varying labor
efficiency at work centers makes the decision about where to allocate labor more
important than has been recognized in past research.

The last labor selection issue addressed here is which resource (worker) to
select from. This assumes that more then one worker is idle and that the selection is
made from those available. If the labor resource is homogeneous, then selection is
not an issue. If the labor pool is heterogeneous, then the selection takes on
importance. Hogg, Phillips, and Maggard (1977) found homogeneous work-forces to
be superior to heterogeneous ones. In most cases, it is neither practical nor possible
to have workers or other resources which are equally efficient.

Maggard, Lesso, Hogg, and Phillips (1973, 1976), Maggard, Lesso, Keating,
and Wexler (1974), and Hogg, Phillips, et al. (1977) modeled labor with varying

efficiency. One goal of this research was to illustrate labor blocking which occurs
Only when resource efficiency varies. This occurs when less efficient resources are
allocated to perform work and more efficient resources are prevented or blocked
fl'om performing the work. If labor blocking can be prevented, flow and queue times

can be reduced. As the variation in labor efficiency increases, shop performance

63

deteriorates. The key to shop performance depends on the availability of efficient

labor, a conclusion also supported by Melnyk and Lyman (1991).

2.5.2 Design Issues in DRC

Design issues, like labor flexibility, efficiency, worker to machine ratio, and
information control deal with static aspects of the model. Design issues set the
parameters in which the operational issues must contend and, thus, can have a direct

bearing on Operations.

2.5.2.1 21129;

Design issues related to labor have three components: flexibility (cross
training), efficiency, and machine-staffing levels. These components are interrelated,
for example, the degree of flexibility or cross training can influence the ratio of
workers to machines. Allen (1963) was the first to demonstrate the benefits of
worker cross training. He claimed workers who were cross trained could be used
more efficiently because workers can be allocated where nwded. Nelson (1967) and
Fryer (1974) showed that as the level of cross training increases, the machine-
staffing levels can be reduced without decreasing performance.

Nelson (1967, 1968), Fryer (1973, 1976), Hogg et a1. (1977), and Park and
Bobrowski (.1989) found that with increased levels of labor flexibility, shop
performance increased. The benefits of greater flexibility can be achieved with a

small addition in cross training. Beyond a certain point, the benefits of cross training

on shop performance is only slight.

With regard to efficiency and flexibility, Nelson (1968) demonstrated that as
the variability of labor efficiency increases, so does the need for additional
flexibility. This was supported by Hogg et al. (1977) and Maggard et al. (1980).
Hogg et al. developed three different models: varying efficiency by worker (LD),
varying efficiency of worker by machine (MCD), and varying efficiency of worker
by both machine and worker (L&MCD). L&MCD represents the greatest variance
in efficiency; because of this, labor allocation rules based on the most efficiency
take on greater importance.

Whether the workforce is homogeneous or heterogenous, the level of cross
training affects the machine-staffing requirements. Maggrad et al. (1973), Hogg et
al. (1975), Fryer (1975), Weeks (1979), Elvers and Treleven (1985), and Treleven
and Elvers (1985) all analyzed the effect of staffing levels. In general, they conclude
that the best shop performance for machine-staffing levels is obtained with a worker
to machine ratio of between 1:2 to 2:3. When staffing levels exceed the 2:3 ratio,
worker idleness increases dramatically. At staffing levels less than 1:2, resource

utilization is at their maximum which causes shop congestion and deteriorates

performance.
252.2 mm

Control of information as part of the design determines where a decision is

made or how it affects the decision process. For example, the decision to move a

65
worker can be either centralized or decentralized. Gunther’s (1979, 1981) work on

transfer delays revealed that as the delay of moving a worker between machines
increases, shop performance (mean and variance of flow time) deteriorates for
traditional labor assignment rules. A parametric rule which considers transfer delay
information cause worker transfers to be delayed and keeps shop performance from
deteriorating.

Another example of information control is found in Fredendall (1991), whose
DRC model used an order review/release (ORR) method to control work on the
shop floor. ORR releases work based on various types of information, such as shop
load. The study revealed that how information is used is more important than what
information is used. It was also found that by using ORR, both dispatching and
labor assignment were not as significant as indicated in past research.

When looking at the control of labor assignments, centralized verse
decentralized information determines the degree of flexibility. Nelson (1967) found
that as control of labor transfer becomes more centralized, mean and variance of
flow time decreases because information regarding the entire shop is used. With
decentralized control, information regarding transfers are localized on divisional
levels and do not consider the needs of the entire operation. Fryer (1974a, 1974b,
1976) reached similar conclusions when he examined the when and where labor

assignment rules.

253 W

66

The DRC literature, generally centers on labor allocation and enhancing shop
performance. The labor allocation issues focus on when and where to send workers.
Most studies showed that when to move a worker was more important than where to
send the worker (Treleven, 1989). Two basic rules about when are: when idle, and
when the current job is done. A third rule, PULL, which was developed by
Treleven (1987), was also found to be effective. PULL is a combination of the when
and where decisions. As for where to send workers, the rule most commonly used is
the longest queue. A third issue of labor allocation deals with varied labor efficiency
(Nelson, 1967; Hogg et al., 1977). When worker efficiency varies, selection of
either a capable or most efficient worker becomes relevant.

Another major focus in the DRC research looks at shop performance through
dispatching and due date tightness. Both dispatching and due date tightness were
found to have a significant effect on performance (Weeks and Fryer, 1976, Weeks,

1979).

2.6 SEQUENCE DEPENDENT MODELS

Sequence dependency involves examining the relationship between jobs.
Every job requires certain unique resources at each step in its processing. In the
context of this research, the unique resource is the combination of machine and
tooling. The relationship between jobs on a particular machine involves which
tooling resides currently on the machine. Sequence dependent rules examine the

tooling attribute (or other attributes) of jobs to find those requiring the same tooling

67

resource and to set their priority ahead of other jobs. The objective behind sequence
dependency is to schedule jobs based on the current tooling setup or commonly
required resources.

It should be noted that sequence dependent scheduling rules and group
scheduling rules are different (Wemmerlov, 1992). Group scheduling attempts to
avoid setups by grouping jobs into families that require similar tooling setups.
Sequence dependent scheduling rules are myopic in that they examine the current
setup and changeover time. Both, however, consider the interrelationship of jobs in

queue.

2.6.1 Segeenee Demndent Scheduling Rules

While the benefits of sequence dependent scheduling are apparent, there is
limited research which examines this environment. Gavett (1965) conducted some of
the earliest research in this area. He looked at selecting the next job based on the
current setup which requires minimum setup time. The objective was to minimize
facility downtime (or setup time) over a finite number of jobs. Using deterministic,
uniform, and normally distributed setup time, Gavett showed that such a selection
procedure performed better than a random job selection rule, although the benefits
depended on the variability of both setup time and batch size.

A study by Hollier (1968) also selected jobs on the basis of current setup.
Hollier compared his current setup dispatching rule to several common dispatching

rules (FCFS, SPT, EDD, etc.). With normally distributed setup times, the model

68

found that rules which considered current setup performed better on several
measures, such as machine idle time and job lateness.

Wilbrecht and Prescott (1969) examined dispatching rules that do and do not
consider sequence dependency. Prioritization based on similar setups (SIMSET)
performed significantly better overall than did other dispatching rules. Although
SIMSET did best in only three out of the nine performance measures, its overall

consistency allows it to be the best overall rule.

2.6.1.1 Tmling Smeenee Dexndeney

Sequence dependency is usually a function of the tooling on a machine and
not the machine itself. The machines ability to process, is a function of the tooling
on a machine. This point was made in the discussion on FMS. An FMS tool
magazine determines what jobs can be processed through the work center. For this
reason, attention is focused on determining the optimal magazine load. For less
automated machining systems, like stamping and molding, tool loading is not an
issue but tool changeover (setup) is. White and Wilson (1977) discussed how cutting
tools have various levels of sequence dependency. The levels reflect the degree of
setup changes necessary which may include tool changes, fixture changes, and
machine modifications such as speed. White and Wilson collected actual data on
setup changes to develop an equation for setup time predictions. The data shows
how sequence dependent scheduling can reduce setup time.

Daoud and Purcheck (1981) examined how reducing the number of tool

69

changes via sequence dependent scheduling can improve shop performance. They
found that sequence dependent scheduling increases machine utilization and timely
job completion rates. They used a traveling salesman matrix to assign jobs on the
basis of lowest change-over costs. An assumption is that a job is tool specific, not
machine specific, and thus can be processed on any machine which reduces change-
over costs. While this model is useful in the planning process, it does not consider
resource availability or loading. The models objective is to reduce setup time
without considering other issues.

Melnyk et al.(l989) and Ghosh et al.(l99l) developed two models which
looked at tool sequence dependency and resource availability. Melnyk et al. looked
at tool control rules combined with dispatching rules that attempted to avoid setup
changes. One rule attempted to avoid a setup change, while the other incorporated
due date priority. These sequence dependent rules reduced the number of setups as
compared to traditional dispatching rules. The sequence dependent tool rule which
considered due date priority, also performed well in terms of a job tardiness and
flow time (depending on the dispatching rule used).

Ghosh et al. (1991) modified the model to look at the impact of sequence
dependency. The degree of sequence dependency was based on different percentages
of setup time to processing time. The higher the percentage, the longer the time for
setup. As the severity of setup time increased, the number of tool changes

decreased, and the extent of sequence dependency increased.

70
2.6.2 Group Sehegeling

Group scheduling categorizes jobs according to common attributes. In this
case, the attribute is common setups (major setup changes) with the possibility of
small changes (minor setup changes). This is an important feature in determining
how cellular manufacturing is obtained. While group scheduling has been applied to
cellular manufacturing, group scheduling is also applicable to other environments.
Hitomi and Ham (1977) showed that a flow pattern environment with group
scheduling has a significant effect on shop performance. Their results showed that
rules that seek to reduce setup through job sequencing do improve performance. As
the ratio of setup to processing time increases, so do the benefits of sequence
dependency. -

Baker and Dzielinski (1960) examined a single machine with sequence
dependent family rules. At issue was whether family-oriented rules perform better
than process oriented rules (SPT). The family rules analyzed in this study were
based on an exhaustive procedure which processed all jobs within a family before
changing. Baker and Dzielinski were also interested in the selection process of the
next family. Their research showed that rotating among the families of jobs was
superior to choosing the next family by minimum setup time. Furthermore, they
concluded, that at high levels of shop congestiOn, family (group) rules work better
for flow time measures.

Sawicki (1973) compared Baker’s (1960) exhaustive family rules to rules

that allow truncation. Truncation of the current family setup takes place when a

71

certain amount of processing time has elapsed. The objective was to improve due
date performance. This type of truncation process would be applicable to
environments where a resource has a finite life, such as tooling. Sawicki determined
that exhaustive family rules are more efficient in machine utilization and flow time,
but are only slightly better on due date issues.

Three scheduling rules were developed and tested by Mosier, Elvers, and
Kelly (1984) that looked at different information in group selection. The three rules
were: highest average job priority (AVE), highest work content per family (WORK),
and economic benefit of changing setup (ECON). ECON differs from the other two
rules in that it allows switching between families. Results showed that group
scheduling rules perform better on flow time and mean lateness than do regular
dispatching rules. Overall, WORK was the best performer, but ECON was a close
second.

Whereas, Mosier et al. looked at group selection based on the combined
characteristics of the group, Mahmoodi, Dooley, and Starr (1990) and Mahmoodi,
Tierney, and Mosier (1992) tested group selection based on a single job’s attribute
within the group. The single attribute was based on a priority determined by the
dispatching rule. The rules tested include: first come first served of all families
(FCFAM), earliest due date from all families (DDFAM), and minimize number of
setups from all families (MSFAM). MSFAM attempts to utilize sequence
dependency by selecting the next family which requires the least setup time change.

A comparison of family rules showed that FCFAM was the worst rule, while

72

DDFAM was best overall. MSFAM showed excellent performance on average flow
time because it attempted to avoid setup changes more than the other rules. This also
explains why MSFAM was such a poor performer on average tardiness.

The previous group/family selection rules tend to be exhaustive because they
process all jobs within a family before changing setup. The problem is, such rules
ignore other job priorities which may be higher. Mahmoodi and Dooley (1991)
examined this issue by comparing exhaustive versus non-exhaustive family
scheduling heuristics. They compared the exhaustive rules (DDFAM and MSFAM)
to two non-exhaustive rules (SLFAM and DKFAM). SLFAM processes all job
within a family until another family has a job with negative slack, at which time the
setup is changed to the new family. DKFAM processes the current family until the
due date of the first job in the current family reaches C (a constant that is
empirically determined) time units greater than the next most critical job in another
family. As compared to exhaustive rules, both DKFAM and SLFAM attempt to
reduce tardiness. Results show that MSFAM still performs best with respect to mean
flow time and preportion of tardiness. DKFAM was best in terms of mean tardiness
but worse with respect to proportion of tardiness. MSFAM and SLFAM were poor
performers regarding mean tardiness. Mahmoodi and Dooley concluded that
exhaustive rules are preferable to non-exhaustive rules in most cases.

Another issue known to influence the affect of group scheduling or sequence
dependency include shop conditions and the ratio of setup to processing time. Ruben

et al. (1991) showed that as shop utilization increases, so do the benefits of group

73

scheduling. This also held true when the ratio of setup to processing time increased.
Wemmerlov (1992) showed that as the number of groups diminish, so do the
number of setups with the result being lower mean flow time. Wemmerlov also
showed that when demand patterns are skewed toward one family, setup time is

reduced, resulting in lower flow times.

2.6.3 Summary 9! the Sequence Demndency Literature

Regardless of whether sequence dependency or group (family) scheduling
terminology is used, the objective is to reduce the frequency of setup changes. By
reducing the number of setup changes, shop performance is enhanced (Baker,
1984b; Mahmoodi et a1., 1990). The current machine setup is compared to the
queue to find the next job which minimizes the setup change. Gavett (1965) and
Mahmoodi et al. (1990) found that this method of selecting the next family of jobs
result in better shop performance.

The difference between sequence dependent rules and group scheduling lies
in how the queue is examined. Group scheduling selects the next group (family)
based on a family characteristics or within group job attribute. If group selection is
based on a job attribute within the group such as processing time or least slack, then
the dispatching rules for job selection should be based on the same attribute for the
best results (Mahmoodi et al., 1990). One assumption of group scheduling involves
major and minor setups (Mosier et al., 1984; Mahmoodi et al., 1990; Mahmoodi

and Dooley, 1991). Major setup time is incurred between groups, while minor setup

74

within groups is incurred (and often is ignored). In contrast, sequence dependency
does not distinguish between these two types of setups, but simply models it as a
deterministic or stochastic distribution.

The issue of whether to use sequence dependency or group scheduling is an
important issue in tooling control. Tool setup changes have been modeled as
sequence dependent, and not as group scheduling (White and Wilson, 1977; Ghosh
et al., 1991). To date, no research has examined tool control and group scheduling

techniques together.

2.7 SUMMARY OF LITERATURE REVIEW

While there is extensive body of literature in a number of areas related to
this research, no work has specifically looked at production scheduling and tool
control. The model developed for this research is a result of past research and
includes some of the following variables:

- Tool life distributions,

- Maintenance service time,

- Maintenance policies (CM vs. PM)

- Frequency of Maintenance,

- Tool allocation/scheduling,

- Tool sequence dependency rules.

While the literature provides a foundation from which this research was

derived, it also points out gaps that exist. The following are some of the gaps that

75

exist.
- No research in DRC schedules limited resources other than labor.
— DRC research which examines scheduling has not addressed the issue of
finite life resources or maintenance.
- There is a lack of specific shop scheduling procedures for both production
and maintenance.
- Sequence dependencies effectiveness, has not been analyzed in the
presences of tool failure or scheduling maintenance.
By examining these gaps in the research, a better understanding of shop floor
control is possible. Managers of production shops face many of these problems daily
and must resolve them by any means. The intent of this research is to provide

insight into the problems managers face and suggest methods to resolve them.

CHAPTER 3

TOOL PLANNING AND CONTROL: A CONCEPTUAL FRAMEWORK

3.1 INTRODUCTION

In Chapter I, a brief discussion of tooling and its role in production was
presented. Understanding that role is essential to appreciate how tooling influences
manufacturing. The purpose of this chapter is to view tools within the context of a
firm’s overall planning and control system. This includes a comprehensive
discussion of the planning and control activities necessary for tool management.
Figure 3-1 presents the framework on which the discussion is based.

The first part of this chapter examines how and why a firms long—range
planning must include a tooling strategy. This is especially true for such production
environments as FMS and stamping and injection die shops (forming tools).
Subsequent sections explore the operational aspects of tooling, including scheduling
and control. Such techniques as Materials Requirement Planning (MRP) will be
evaluated. Actual shopifloor control of tooling also will be addressed. In addition, a
major portion of the discussion will focus on analyzing different tool control

scenarios.

3.2 TOOL MANAGEMENT FRAMEWORK
Traditionally, planning and control of tooling has not been viewed as part of

the mainstream of production planning. Although not at issue in this research, an

76

77

 

 

 

82m xo<2 JOGhzOO

 

 

 

 

 

 

 

 

 

 

 

20....<oo....< .00... «Bolt
d<2°—P<¢W1° /\\ £020
0:... E... _ —
\ /
>10h m>z_ >¢Oh m>z.
$2300.. m.<_mw._.<2
.uzazm. \ /
02...:am20a 02.zz<..a 02.zz<..a. 0522‘...
D. n. hzmxu¢.:0u¢
02.400... >....0<n.<o 045.5951
oz.zz<..& >205}. 02......amzoo
.50 Icsoc . ' 20....03005.
zmho<l
BE. .22.... _ _
02.225... Oz.z.z<4.. Oz.zz<..n.
4 .P ”CE: W! 02—22(4a
OO 00 20....0300mn.
Gz...OOh >....O<n.<o 02.m:...0<..52<2

 

 

63:00 Use 9:53.. 8:260... Tm 2:9”.

 

 

 

exter
and
The
and
the 1

Earl

mar.
“hi.
hon
The

Silal

the

den

 

 

78

understanding of how tooling fits into the planning process demonstrates the full
extent of tooling’s effect. Figure 3-1 is a variation of the manufacturing planning
and control system framework developed by Vollmann, Berry, and Whybark (I988).
The model breaks planning and control into three parts: manufacturing, capacity,
and tooling. Looking vertically, Figure 3-1 shows planning at the top (front end),
the control functions of scheduling in the middle with operations at the backend.

Each segment is discussed in detail in order to explain the importance of tooling.

3.2.1 Plagning ef Tmls

As noted in Figure 3-1, three areas interface at the planning level:
manufacturing, capacity, and tooling. The driving force is manufacturing strategy,
which determines the strategies of both capacity and tooling. At this level, planning
horizons usually are long term, from a minimum of one month to more than a year.
The planning process determines the length of time it takes to implement the
strategies. The production plan dictates what type and how much capacity is needed.
Capacity is composed of available labor, machining time, and tooling. Capacity is
the link between manufacturing planning and tool planning because tooling is a
determining factor in capacity (Blackburn, 1988). If any of the three capacity
resources are insufficient, either more workers must be hired, more machines
purchased, or additional tools obtained. The lead time for adding machines and tools
may range from three months to a year, depending on the tool intricacy. The lead

time may be hours for the purchase of simple cutting tools or up to six months for

79

forming tools (Brown et al., 1981). The planning process must consider and
compensate for leadtime. By preplanning capacity through the manufacturing plan,
tooling needs can be integrated at an early stage, which is increasingly desirable as
the manufacturing environment changes, to time-based competition. Companies must
develop products more quickly (Cross, 1986; Stalk, 1988), and firms such as
General Motors have adopted strategies like simultaneous engineering, JIT and other
methods to accomplish this. Tools are developed at a much earlier phase in product

planning to reduce the lead time for product development.

3.2.2 Sehefleling ef Tmls

Tooling decisions are driven by the production plan through the Material
Requirement Plan (MRP) schedule. The MRP schedule determines which products,
and in what quantity, will be produced. However, MRP does not consider whether
the necessary resources are available. These are determined by linking Capacity
Requirements Planning (CRP) to MRP and then developing, detailed Tool
Requirement Planning (TRP) (Wassweiler, 1982; Savoie, 1988). Figure 3-2
illustrates the process. The Bill of Tooling (BOT) defines which tools are to be used
on a product and the tool life expected to be consumed at the end of processing
(Gayman, 1980). It is usually expressed in processing time or number of parts
produced. If this information is not on the BOT, than TRP can not track tool life.
An alternative tracking method proposed by Erhorn (1983), is to combine the

information on the Bill of Materials (BOM) with the BOT. In either case, the main

8O

 

Figure 3-2 Tool Scheduling
and Control

MASTER PRODUCTION
SCHEDULE

1

MATERIAL CAPACITY TOOLING
REQUIREMENTS .— REQUIRENENTs .—. REQUIREMENTS
PLANNING PLANNING PLANNING

BILL OF ' MATERIALs TOOL BILL OF
MATERIALS INVENTORY INVENTORY TOOLING
(BOM) (BOT)

 

 

17 I?
SHOP FLOOR CONTROL. ; ; TOOL SELECTION
& ALLOCATION

 

 

Objective is tO allow the system tO plan tOOl requirements, monitor tOOl inventory,
and plan tOOl replacement and purchases. If the amount, location, and life Of tools is
known, substantial savings can result (Huber, 1989; Vasilask, 1990).

In an FMS, the scheduling stage Of production planning involves tOOl A
selection for the work centers’ magazines. The tOOls placed in each magazine
determine the capabilities Of that work center. Under FMS, the key tO production

scheduling is tOOl allocation (Carrie and Perera, 1986). Kouvelis (1991) presents a

 

81

two-level decision hierarchy that links tOOl scheduling with long-term production
planning as a means Of production scheduling.

In a traditional jOb or flow shop using cutting tOOls, production scheduling
is affected by tooling tO a lesser degree than under FMS. In most cases, the
production schedule or MRP drives the traditional shop, not too] availability. With
only minor exceptions, tooling is treated like any Other inventoried material, Erhom
(1983) believes cutting tools need tO be controlled with systems like MRP, but not
forming tOOls, which are considered a capital asset and tend not tO be as perishable
as cutting tOOls. He adds that cutting tOOls should be controlled as inventory items,
but the remaining useful life Of each tOOl does not need tO be tracked. A major
benefit Of Erhom’s method is that it is an inexpensive and simple way tO monitor
tOOl control. Savoie (1987) agrees with this approach because it Often is impractical
tO track tOOl life.

The negative aspect Of Erhom’s system is the risk Of accumulating excess
tOOls in inventory with little remaining useful life. Wassweiler (1982) advocates
tracking tOOl usage as an inventory control method, which can be accomplished with
a BOT (Wessweiler, 1982). If tOOl life is tracked, lower or no tOOl safety stocks are
required (Savoie, 1987).

Unlike labor and machine time, tooling can be inventoried. This unique
feature allows it tO be scheduled and controlled differently than Other forms Of
capacity. There are two common ways in which tOOls are inventoried. The first

method treats tOOls as a material component controlled like any Other MRP item.

82

This method does not consider, tOOl life, only the number Of tOOls currently in
inventory.

The second method is based on tracking tOOl life and is more costly than the
first method. It may not be cost justifiable for most cutting tOOls. The advantage is a
more accurate knowledge Of available tOOl capacity and its current condition.

For forming tools, TRP and the scheduling process are the same as used for
cutting tOOls. Even though Erhom (1983) does not see the need tO track and
schedule nonperishable tOOls (forming tOOls), there is research that indicates
Otherwise (Brown, et al. 1981; Huber, 1989). TRP can be used tO track tOOl life,
which can aid in production and maintenance scheduling (Newman, 1985; Ram and

Olumolade, 1987).

3.2.3 Shep Flmr genteel ef Tmls

The final stage Of tOOl management involves detailed scheduling and selection
Of tools on the shop floor, where the physical removal or movement Of tOOls takes
place. TOOl selection usually is random, especially when tOOl life is nOt tracked. In
automated machining systems, like FMS, monitoring Of internal tOOl wear is
common, in which case random tOOl selection is easy and non-detrimental tO system
performance (Tarn and Tomizuka, 1989). That is, since tOOl life is being tracked,
random selection is from among tOOls with sufficient expected life tO process a jOb
in its entirety. Usually there is only one copy Of each type Of forming tOOl, selection

is not an issue.

83

TO date, no research has addressed tOOl selection policies involving a tOOl
type (that is, multiple copies Of cutting tOOls). Intuitively, however, it seems that
choice would be based on whether the tOOl has sufficient life tO process the job fully
or on some specific policy. The first policy is straight-forward, select a tOOl that will
allow job completion without downtime for tOOl replacement. The second policy
could include such specific policies as selection tO achieve lower inventory (by using
Older tOOls first) or tO reduce the risk Of tOOl failure (by using newer tOOls first).
Using Older tOOls first would reduce excess tOOl accumulation and could incorporate
a policy to replace the tOOl before the full tOOl life usage is reached (Lyman, 1993),
thus reducing inventory.

Using newer tOOls would reduce the risk Of tOOl failure as well as product
scrap and through-put time, and machine utilization would be increased (Banerjee
and Burton, 1990). An added benefit would be better use Of finite storage capacity.

As noted, nO studies have examined tOOl selection from among multiple
copies. Such research is needed, and a useful extension would be tO explore policies

and interactions Of multiple cutting tOOl copies under multi-tOOl jOb sequencing.

3.3 DETAILED SCHEDULING OF FORMING TOOLS
3.3.1 WW

Decisions about when and where to transfer tOOls resemble similar issues
related tO labor as a resource (Nelson, 1965; Fryer, I975; Treleven & Elvers,

1985). Whereas, past work has treated labor as a limited resource with a finite life,

84

tOOls, on the other hand, need replacement or renewal (refurbishment) after a certain
period Of use. The finite life Of tOOls makes them uniquely different from the labor
resource. This is not tO say that there are nO similarities; rather, they are not fully
interchangeable.

When tO place a tOOl intO production service involves control heuristics
typically based on jOb priority. Since a job is both machine and tOOl specific for an
Operation, job heuristics determine which tOOl should be used next for which job. If
the desired tOOl is not available, lost capacity may result (Mason, 1991). Thus, shOp
control heuristics need tO be developed tO reduce or eliminate the effect Of lost
capacity. However, to eliminate the problem would require a forward looking
scheduling capability, and the linkage Of capacity requirements tO expectations Of
tOOl life.

The when tOOl decision looks at: (1) when tO place a tOOl in production, (2)
when tO pull a tOOl from production, and (3) when to place a tOOl into maintenance.
In all three cases, the when decision is time oriented. In the first instance, a tOOl is
placed into production depending on demand. If a job requires a particular tOOl for
processing, that tOOl type is removed from storage and placed on the machine.

Removing a tOOl from production depends on both the lack Of demand (nO
jOb requiring the tOOl) and on the tool’s condition. When a tOOl fails or comes due
for PM, it is removed from production. Thus, the third instance relates tO the
second. TOOl failure dictates the need for maintenance (CM), whereas PM usually

takes place at a scheduled moment (or after an accumulated amount Of processing

85

time. PM also may occur just after tOOl inspection, which can take place at any time
but usually is done just after a tOOl is pulled Off the machine (when production is
completed or truncated).

The tOOl placement decision lOOks at where tO place or remove the tOOl in
production. These where decisions are not as prevalent as the when decision because
where issues depend on certain conditions such as tOOl flexibility.

Deciding where tO place a tOOl involves demand for a tOOl type. If there is nO
tOOl flexibility between or among machines, then there is nO Option and where is not
a concern. If multiple machines have a demand for a tOOl type, however, the issue
Of where tO place a single tOOl copy becomes importance. This is also true if
multiple tOOl copies havedifferent remaining processing life. The where decision
must analyze the tOOl life for each tOOl and jOb processing time. TO date, nO
research has addressed this matter. DRC research has shown that the decision about
where tO place labor does nOt have a significant effect on shop performance
(T releven and Elvers, 1987), but whether this holds true for tOOls remains tO be

investigated.

3.3.2 Examples pf Teal Cpntrel

Figures 3-3 a-c illustrates how tOOls are controlled in a shop. In Figure 3-3a,
the theoretical production shop is empty and idle. There are three work centers, with
four dedicated tools per work center and one maintenance center. T1-3 is too]

number 3 used at machine center one. Tools that need service are sent tO the

86

 

 

Figure 8-33 Shop Layout and Tool Control

(Tool. or. muohlno opoclflo wlth no duplicate)

 

 

 

 

 

 

WORK OINTIR i WORK CENTER E WORK CENTER E
OUEUE OUEUE OUEUE
T1-1 TC-I TE-1 TE-I TR-1 T.-.
[WWII—1 WIT] W
T‘-. T1-4 TE-E TE-4 TO-E TO-4

 

 

 

 

 

 

 

 

 

 

 

 

1 l r
l

MAINTENANOI GUIDE
‘ MAIN euauo
FROG...

Koya
Tx-y. Tool y on work center at
Jy. Job roqulrlng y tool (In quouo)

 

 

 

 

 

 

 

 

Figure 3-3b Shop Layout and Tool Control

 

 

 

 

 

 

 

 

 

 

 

 

WORK CENTER 1 WORK CENTER 8 WORK CENTER 3
J1 44 J: Jr J2 J8 41 do 43
71-3 72-3 13-1
[ jar-4 ] [ TIE-1 1 [ 73-3 1
78-8 73-4

 

 

 

 

 

 

 

 

 

I ..-.__ l J

Ti-E TE-4

 

 

 

TO-E

 

 

 

 

 

 

 

 

 

Figure 3-30 Shop Layout and Tool Control

 

 

 

 

 

 

 

 

 

 

 

 

WORK CENTER ‘ WORK CENTER E WORK CENTER 3
4‘ J1 J8 J4 J1 JE J3 J4 J4 J3 J1
T1-O T3-1
I Tfi-E J [ Tfl-E J [ TS-E 1
T1-4 T2-4 “rs-a T3-4

 

 

 

 

 

 

 

 

 

 

TE-1 TE-S

 

 

T1-1

 

 

 

 

 

 

 

 

87

maintenance queue and wait until capacity is available. When service is completed,
the tOOl is sent back to its designated machine center.

Figure 3-3b shows an active production shop. Each machine has a job queue
specific tO each work center. I 1 represents a jOb requiring tOOl number one. Looking
at work center 1, the current setup is tOOl number 4, which the second jOb in the
queue requires. When the current job is done processing, the tOOl should be changed
tO handle the priority job in queue, but Tl-l is on its way to maintenance facility
and is not likely tO be available. At work center 2, the current tOOl setup is number
1, which the next job in the queue (highest priority) requires, so nO tOOl change is
needed.

In Figure 3-3c, a difficult decision can be seen at work center 2. When the
current jOb is completed, which job should be processed next? Should the tOOl be
kept on the machine and the third jOb in the queue run so that a setup is avoided? Or
should the tOOl be pulled and setup time incurred in order tO process J4? The
tradeoff is between the expense Of an additional setup and the possibility Of late
delivery on a higher priority job (Mahmoodi et al., 1990). The situation at work
center 3 is similar except that the two highest priority jobs require the same tOOl.
Work center 1, the first two jobs in the queue are blocked out Of production until
the necessary tOOl is available from maintenance. TOOl blocking, due tO the lack Of a
tOOling resource, is similar to the situation Goodman (1979) noted regarding labor
blocking in a DRC.

Figure 3-4 a and b illustrates how tOOls are controlled when there is too]

88

 

Figure 3-4a Shop Floor and Tool Control
(TOOI flexibility with no duplicate)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TOOL
CRIB WORK
T1 CEN1TER
T2
T3
T4 WORK
1-5 TOOL PRODUCTION PATH CENTER
3 ‘ 2 I—
T6
T7
T8
T9 WORK
T10 : CENTER
3
TOOL TOOL MAINTENANCE PATH

 

 

MAINTENANCE

 

 

 

 

 

 

Figure 3-4b Shop Floor and Tool Control
(TOOI flexibility with no duplicate)
ch

TOOL
CRIB

 

 

L : T10

 

 

T1
W62

 

 

 

 

T3
T4

T8

T7

 

 

 

W03

 

 

T8

TOOL ._T2
MAINTENANCE

T6“

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

flexibility. From
center, so job rOu
work center has
needed. If so, th _
crib. If not, it gr
re1621564 to the s':
the same tool sir
T10). a machinci
resource, T001 t
F1glues 1
there are multip
due to excess dl
tradEOff is COS“
capacity, and a
firm vauires c
greater the mm

tool to SEIect n

should tool life

reWren.

 

 

89

flexibility. From centrally controlled tOOl crib, tOOls can be assigned tO any work
center, so jOb routing flexibility is possible. As was the case in Figure 3—3, when a
work center has completed a job, the tOOl is inspected tO determine if servicing is
needed. If so, the tOOl is sent to the maintenance center and then back tO the tOOl
crib. If not, it goes directly back to the tOOl crib. Depending on how jobs are
released tO the shop floor, it could be possible for different work centers tO require
the same tool simultaneously. With nO duplicates for each tOOl type (T1 through
T10), a machine can be blocked from production due tO lack Of available too]
resource. TOOl blocking is a factor in lost capacity (Mason, 1991).

Figures 3-5 a and b shows that the risk Of tOOl blocking is reduced when
there are multiple copies Of each tOOl type. With one copy per work center, blocking
due tO excess demand is unlikely. Less tOOl blocking means less lost capacity. The
tradeoff is costs, which include: increased tOOl investment, additional storage
capacity, and a more difficult selection process. The number Of duplicate tOOls a
firm acquires can be constrained by its storage capacity and available capital. The
greater the number Of duplicate tools, the more complicate is the decision Of which
tOOl tO select next for processing. Should Older tOOls be chosen over newer tOOls, or
should tOOl life be the criterion? This question should be addressed in future

research .

90

 

 

Figure 3-5a Shop Floor and Tool Control
(Tool flexibility with multiple duplicates)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TOOL CRIB
WORK

71.1 71.2 71.3 "—"“ “"1"“
72.1 72.2
73.1 79.2 73.9

. . . WORK
1:;‘1 $;.§ T4 3 TOOL PRODUCTION PATH CENTER
79.1 79.2 79.9 ‘ i 2 —
77.1 77.2 77.3
79.1 79.2 79.3
79.1 79.2 79.9 wonx
710.1 710.2 ~n—-—-~ OENSTER

TOOL TOOL MAINTENANCE PATH

 

 

MAINTENANCE

 

 

 

 

 

 

Figure 3-5b Shop Floor and Tool Control
(Tool flexibility with multiple duplicates)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

wc1
TOOL ORIB
, 74.3 r

71.1 71.2 71.3 ’T—"
72.1 72.2 WC?

73.2 73.3
74.1
75.175.2 % — “-2 _
79.1 79.2 79.3 ‘ _ '—

17.: T3.‘I
79.1 79.2 T8.3
79.1 79.2 79.3 77.3 __.
710.1 710.2 ~-———- 77-1 -

I wca

TOOL

 

 

MAINTENANCE A

 

 

 

 

 

 

CHAPTER 4

RESEARCH METHODOLOGY AND SIMULATION MODEL

4.1 INTRODUCTION

The research considered in this study focuses on experimentation in a
production shop system which is machine and tool limited. Looking at either one of
these resources separately gives an unrealistic view of shop scheduling. A more
realistic approach is to examine each resource (machine and tool) in combination.
However, this requires that the production system have both tools and machines
available for processing. A vast majority of past research investigating the Dual
Resource Constraint (DRC) environment has examined only machine and labor
availability through dispatching rules and labor assignment (Treleven, 1989). This
research differentiates itself from past research by replacing labor with tooling. The
benefit of examining a DRC environment is the identification of different tooling
factors which could influence shop performance.

To test the effects of finite life tooling, a simulation model is used to gather
data on a hypothetical DRC flow shop. The DRC shop will be analyzed under a
number of different experimental conditions. The output will then be subjected to
analysis of variance (ANOVA) statistical tests for the performance measures and
Tukey HSD multiple comparisons for both a_p;i_o;i, hypotheses and m tests.

The simulation method of analysis was used in order to better understand

tooling and to develop effective control heuristics. In addition, several firms

91

92

interviewed prior to developing the model expressed several concerns, including: 1)
the need to understand how variance in tool life and maintenance service time affects

the shop and 2) the need for usable and simple control procedures.

4.2 MODEL DEVELOPMENT

Proper development of a simulation experiment necessitates that several steps
be initially considered. An essential part of the simulation development requires that
a detailed evaluation of the model be conducted on an ongoing basis. Fishman and
Kivint (1968) classify the simulation model evaluation process into three areas: 1)
validation, 2) verification, and 3) problem analysis. A brief description of each is
provided below with a detailed discussion presented later.

1) Validation: Determines if the model accurately represents the real system

or environment.

2) Verification: The process of ensuring that the model behaves the way it

was intended.

3) Problem Analysis: The process of drawing statistically significant

inferences from the data generated by the simulation model.

The order of verification and validation is not paramount as far as research
methodology is concerned (Law and Kelton, 1991). In fact, the two should proceed
simultaneously throughout the model’s development. Problem Analysis, on the other
hand, is related to the output of the model. It can only occur after the model has

been both verified and validated. Problem Analysis will be covered in Chapter 5 in

93

Design of Experiment section.

4.3 VALIDATION OF EXPERIMENT

The validation process attempts to answer two questions, 1) does the model
behaves in the same manner as the real life system?, and 2) what inferences can be
drawn from the model? An interesting component to validity is the fact that models
can never be proven correct, but rather incorrect.

Cook & Campbell (1979) provide four methods which can be used as
validating techniques. These consists of the following forms of validity: (1)
External, (2) Construct, (3) Internal, and (4) Statistical Validity. Each of these forms

of validity is discussed below in detail with respect to simulation modeling.

4.3. l Extgmfl Validity

External validity (face validity) deals with the question of whether the model
represents the real world sufficiently to apply the results. One way to have external
validity is if the model is based on information from an actual production
environment. By using actual shop data, any conclusions drawn or policies
developed can, in general, be applied to the shop. One problem with this approach is
that results frequently tend to be shop specific and may not be generalized to other
conditions or shops.

In developing the shop model for this research, extensive interviews were

conducted. Discussions with plant managers and shop floor personnel provided

94

insight into the workings of flow shops and job shops. Using the information derived
from these interviews, a general model was developed with a high level of external

validity.

4.3.2 Canstgtgt Validity

Construct validity looks at the causes and effects of manipulated variables
which determine how generalizable the results are. Simulation only considers a
certain number of constructs in the model. Real shops involve many more
constructs. Lack of generalization is due, in part, to the limited constructs in the
simulation model. This is a concern that exists when using simulation. This problem

can be addressed through proper selection of relevant measures and constructs.

4.3.3 Intgrnal Validity

Internal validity deals with the causal relationship between two measured
variables. Does the independent variable cause a variation in the dependant variable?
A causal relationship exists between variable X and variable Y, if Y is a direct
result of X. There are several threats to internal validity which include: history,
testing, instrumentation, and diffusion of treatment (Kirk, 1982). The unique feature
of simulation is that it does not suffer from these concerns. The major concern of

internal validity is whether the model operates correctly.

4.3.4 Statistical Qanglusian Validity

95

The existence of co-variance between input (x) and output (y) allows for
inferences to be made with regard to the model. The ability to draw conclusions or
inferences about the co-variance is dependant on the following (Cook and Campbell,
1979).

- Statistical Power: Inadequate sample size may cause the Null hypothesis not to be
rejected when there is a true significant difference in means. A minimum sample
size was determined (see Section 4.4.4) to ensure sufficient power. Another means
of increasing power is the reduction of irrelevant sources of variation by using
common random numbers.

- Violated Assumptions of Statistical Tests: Certain assumptions must be met if data
analysis results can be interpreted. These assumptions include: normal, identical and
independently distributed (IID) variables, and homogeneity of variance (see Section
4.4). Other assumptions which affect statistical validity involves random assignment
of variables and deciding if the model reaches steady state condition.

- Fishing and the Error Rate Problem: As the number of statistical tests increase, the
probability of drawing an incorrect conclusion increases. Likewise, as the number of
factors and levels in an experiment increase, so does the number of comparisons. To
reduce the likelihood of a false conclusion, a larger confidence interval is needed.

- Random Irrelevancies in the Experimental Setting: Random features, like
processing time, can affect the outcome of the dependent variables and increase
estimated error. The advantage of simulation is that common random number

streams are used, so that appropriate comparisons between treatment effects can be

96
made (Law & Kelton, 1991; Kleijnen, 1987).

4.4 SIMULATION DESIGN ISSUES

There are a number of issues which must be addressed when developing a
simulation model. These included: verification, initialization bias, variance
reduction, and sample size. By examining these issues, proper statistical techniques

can be applied and valid inferences regarding output made.

4.4.1 Varifiaatian

As stated previously, verification is an ongoing process that should occur
while the research modelis being developed. Verification centers on whether the
algorithms used operates correctly and whether the control procedure is modeled
correctly.

Most simulation text books (Pegden, Shannon, and Sadowski, 1990; Law and
Kelton, 1991; and Banks & Carson, 1984) provide the basic steps for verification of
a model. The steps used for this model include:

1. Informal Analysis

Informal analysis starts with a review by individuals who possess the
appropriate knowledge and abilities necessary to find errors. Use of fellow students
aided in this process.

2. Structured Walkthrough

This technique is the debugging process for verification. Advance simulation

97

software packages possess the ability to find system error and allow monitoring
during initial runs. The Siman Trace command allows event by event monitoring of
the model and an in-depth analysis of the flow of jobs. Furthermore, the progress of
jobs were plotted to see if they meet expected results.
3. Dynamic Analysis

This technique involves verification by running the model and observing the
models parameters at different levels. This stress testing pushes the model to its
limits and further provides evidence that the model is working properly.
4. Comparison to Known Output

To further verify that the model is operating correctly, comparisons to other
similar models can be made. This measure requires that the model have similar
experimental conditions. The DRC model developed for this research cannot be
compared to other models directly because of the many differences. Instead, the
results were presented to the interviewed companies for their input and comparison.

Each of the techniques used for verification are iterative.
Each technique was performed on the model as needed to ensure that the model was

verified.

4.4.2 14119211291011.1293
Non-terminating models must be concerned with start-up conditions. Many
models are started idle and empty which means that machines are free of work and

there are no jobs in queue. For non-terminating systems, the initial condition is not

98

representative of the expected operating conditions and will cause a bias in the
measured parameters. Only after the model is run does it reach an equilibrium or
steady state point. Until that point, the data collected is of little statistically analytic
value due to bias which lower performance values. Once steady state is reached, the
system exhibits long term behavior.

There are a number of techniques for reducing start-up bias [Wilson &
Pritsher, 1978]. The three most recommended approaches include: A) using a long
simulation run, B) truncating (discarding) a portion of the data and C) selecting
initial model conditions to reduce bias.

Of these three techniques, truncation was selected due to its ease of
application and effectiveness. This technique requires that the initial data be
discarded. By dropping this transient data, the biased data is eliminated from the
study. A good estimate of the start-up condition length is required.

Several methods were used to determine the truncation point (steady state).
The first method used is the visual inspection of a graph (Pegden et al.,1990). This
graphical procedure requires several replications followed by applying a moving
average on the mean values (Welch, 1983). The data is then plotted graphically to
see when the performance variable reaches steady state. Based on this method, it
was estimated that 100 jobs would need to be truncated. While this method is
simple, it lacks statistical proof that initialization bias is eliminated. For this reason,
the Schruben, Singh, and Tierney (1983) technique was used.

This second method is based on the Schruben (1982) and Schruben et al.

99

(1983) technique for detecting the presence of initialization bias in a time series. The
technique tries to determine if significant differences exist between batch means
from an initial run. The data is divided into 2 parts (halves) where k < N, with the
second half usually being much larger than the first. The following steps are used in
detecting negative initialization bias on time in system performance measures.

Step 1. Determine the sample variance and degrees of freedom using the batched

_ (,2
var =—
<le N

method. The degrees of freedom for t,, is equal to (n/2)-1 where n is the number of
batches.

Step 2. Determine the t statistic based on Schruben et al. (1983).

 

tv=( £0): (pawn-7k)
Step 3. If found not significant, the null hypothesis can be rejected, indicating no
difference between YN and Yk means. In addition, if the null hypothesis is rejected,
than the first half of the data is dropped and another batch of jobs (next k jobs) is
used as the first half. This process is repeated until the null hypothesis is accepted
(Kleijnen, 1987)

The Schruben et al. (1983) technique was performed for each treatment (112
cells). The number of jobs needed to be discarded ranged from 85 to 135. The

actual number of jobs discarded was 2000 for the start of each treatment run because

100

this was the value chosen for the batch size. Truncating by batch ensures that all
remaining batches are of equal size which is relevant when statistically analyzing the

data.

4.4.3 W

The use of Common Random Numbers (CRN) as a variance reduction
technique in the experimental design can improve the analysis of the results [Nelson,
1990]. CRN can also reduce the number of replications required by a model while
achieving the same level of precision of the model (Pegden et al., 1990). The
objective of CRN is to reduce variance in the point estimate of the mean response
except for that caused by the treatment. Reducing variance allows for smaller
confidence intervals of the performance measures. Smaller intervals allow for more
confidence in inferences drawn from the interval.

Use of common random numbers starts with a numerical seed value for a
given random number stream. The numerical swd value is the identical starting
point for each treatment. A random number stream is assigned to each unique
distribution (i.e. arrival time, processing time, and set up time) that requires
generating a random variable. This ensures that each varied model configuration will
have matched (synchronized) random numbers.

The use of CRN poses a problem through the loss of independence of
samples. By not synchronizing one random number stream, independence between

treatments is increased (Mihram, 1974). A total of six random number streams are

101

used. The only random number stream not synchronized was that used for defining
which machine a job would be assigned to. Arrival rate, processing time, setup
time, tool life, and maintenance service time were all matched for each treatment.
Another major problem with the use of CRN is the fact that it induces
autocorrelation between batches. This problem can be solved by having sufficiently

large batch sizes (refer to Section 4.4.4.2).

4.4.4 Sample Siza

Sample size (or run length) is a key concern for statistical analysis. An
inappropriate sample size may cause biased data and a non-normal distribution.
When the sample size istoo small, autocorrelation becomes a factor which affects
the statistical analysis. A larger sample size can resolve these issues and provide
high statistical power. A pilot run for mean time in system was conducted in order
to determine the sample size for each treatment condition which ensure that the
effects of normality and autocorrelation did not influence the statistical analysis. A
batch size of 2000 jobs was selected with 100 batches per run (treatment). This size

is considered sufficient by Schmeizer (1982) for estimating confidence intervals.

4.4.4.1 Narmgity

Traditional statistical analysis, like ANOVA, requires that the data be
normally distributed. Some variation in normality is allowed because of the

robustness of ANOVA (Neter, Wasserman, & Kutner, 1990). With small batch

102

sizes, it is less likely that observed values will be distributed normally. In
simulation, data tends to be highly correlated, and thus batch sizes have to be
sufficiently large for the means to be normally distributed.

Several methods can be used to determine if data is normally distributed.
The simplest method is to graphically plot the output (mean of n batches) and
visually inspect the results (Pegden et al., 1990; and Wilkinson, 1989). While this
method was used initially, the technique lacks statistical proof.

Instead, the method used for testing normality was Filliben’s (1975)
probability plot correlation coefficient test. Filliben’s method requires testing of the
correlation of batch means by comparing them with Filliben’s critical correlation
tables. If the correlation of the batched means surpasses the critical value from the
table, then the data is considered normal. If not, then the batch size is increased and
the test repeated. After testing each experimental condition, the minimum batch size

needed to ensure normality was between 900 to 1000.

4.4.4.2 W

Autocorrelation allows the variables of batch K to directly influence the
variables of batch K + 1. The smaller the batch size, the stronger the influence.
Law and Carson ( 1979) and Mihram (1984) suggest that the batch size be increased
until they become uncorrelated. To test for autocorrelation, the Von Neumann
statistic (q) (as recommended by Kleijnen (1987)) was used to test the output (flow

time). The value of q can be computed as a function of the number of batch means

103

(n). A single large batch is run and the test statistic divides the run into the greatest
number of batches (it) possible without being autocorrelated. If batch means are not
independent or normally distributed, then the null hypothesis is rejected and the
batch size is increased. Kleijnen (1987) further recommends that a value
corresponding to n= 100 provides the autocorrelation test with sufficient power.
Each treatment was tested using the Von Neumann test. The minimum batch size

necessary to ensure that autocorrelation was not a factor was 500 jobs.

4.5 DESCRIPTION OF SIMULATION ENVIRONMENT

The simulation model is described in the following three sections. In the first
section, a computer simulation using a discrete event model with user written
subroutines is described. The second section discusses the control procedures for

both job and tool, and the final section describes the assumptions made in the design

of the model.
4.5.1 SEW

The simulation model is a flow shop in which identical orders (job) follow a
specific path. This is in contrast to a job shop which randomly routes each job for
processing. Discussions with managers from several firms showed that forming
tools, common in stamping or injection model plant, tend to operate using a flow
shop pattern. On the other hand, cutting tools were found to be more common in job

shop environments. Since the focus of this research is on forming tools, a flow shop

104

 

 

o mam; i oumao 111,
v mafia 1' wumao .11
$3533 ’ 1L campus}! ..§Em<
cos a m: 9;. i O m 30.1111 3.. i 3..
szmO
II a mafia“: tll «Myrna i

,.I. . mz__zuol<sllill .Mw—uao IL.

 

 

8% 9: £92.; {25 B 26E 73 2:9“.

 

 

105

model was selected to enhance the study’s external validity.

Figure 4-1 describes the sequence of events required to process a job. Jobs
arrive based on a Poisson distribution at a negative exponential inter-arrival rate.
Upon arrival, job attributes (parameters) are defined, including: due date, processing
and setup time, and machine and tool requirements. The following sections describe
each of the parameters in detail. Table 4-1 provides a summary of the values

selected for the different model parameters.

Table 4-1 Summary of Simulation Environment

   
 
 

 

 

 

 

 

 

 

 

 

 

Description

Job Arrival Exponential, Mean based on 90% machine utilization
Due Date TWK, k=7
Processing Time Normal Distribution, Mean = 10 hrs., Std.Dev.= 1.5 hr
Setup Time Normal Distribution, Mean = 2 hrs, Std.Dev.=.2 hrs.
Actual Tool Life Normal Distribution, Mean = 120 hrs.

Std.Dev.= 14.12 & 38.2
PM Point of Tool Life Constant, 80% of Mean Tool Life
Preventive Maintenance Service Time Log-normal Distribution with Mean of 3.0 hrs.,

Variance=.3 & .9
Corrective Maintenance Service Time 2.0 * Preventive Maintenance Service Time
fl

 

 

4.5.1.1Wfl2

Due date is determined by taking the jobs arrival date plus a multiple of the
jobs processing time (TWK method). The following equation is used to calculate a
jobs due date (Conway et al., 1967).
DUE DATE = ARRIVAL TIME + K *TOTAL PROCESSING TIME
Where: K = constant

The value for K in the TWK due date assignment procedure is a major

106

determinate in the percentage of jobs past their due date. TWK was selected because
it has been shown to be an effective method for setting a due date with respect to
performance measures such as tardiness (Baker, 1984). A K value of 7 was selected
which resulted in 15 to 30 percent of jobs being tardy (Lyman, 1993). Several past
DRC models have also found K=7 to be a reasonable value (Melnyk & Lyman,

1991).

4.5.1.2 Ergessing Q51 Sstup Tim;

Processing and Setup time is determined from a random normal distribution.
A normal distribution was chosen because demand patterns from customers tend to
vary around a mean value, with equal probability and variance. Conversations with
plant schedulers indicate that processing time per customer varies only slightly. It
was estimated that the typical job (order) took an average of 10 hours of processing
time. Once processing starts for a job, the machine runs until the job is complete
(unless tool failure occurs).

Setup time was not a major factor for most plants visited because of quick
die changes. Setup time ranged from a low of 3 minutes to as high as 6 hours.
While most setup times were less than an hour, when it does exceed one hour, tool
setups tend to be a key concern. For this reason, setup time was set at two hours (20

percent of processing time) with a normal distribution (see Table 4-1).

4.5.1.3 Maghina aad Tml Assignment

107

Machine and Tool selection defines the location where a job will be
processed. The job is first randomly assigned a specific tool from a uniform
distribution. Once assigned a tool, the job is assigned a machine based on Table 4-2.
Table 4-2 shows that there are four tools per machine and that a tool is specific to
only one machine as designated by the value of l. A zero value designates that a
tool cannot be assigned to that machine.

Table 4-2 Tool-Machine Assignment Matrix

0
0
0
0
0
0
0
0
l
1
1
1
0
0
0
0
0
0
0

OOO————OOOOOOOOOOOO&
-~OOOOOOOOOOOOOOCOM

ooooooooooooooo——~—-
OOOOOOOOOOO———-—OOOO

 

Once assigned a machine (based on tool requirements), the job is sent to the
appropriate machine queue. The job will wait in queue until the time when it’s
priority establishes it as the next job for processing. The establishment of priority
within a queue is part of the job priority control policies which will be discussed in

Section 4.5.2. Once a job is done processing, the job then exits the system and is

108

considered complete.

Should a tool fail while a job is being processed, the job must return to the
machine queue. The job’s processing time is decreased by the amount of time the
tool was able to process the job. The failed tool is then sent to the maintenance
queue and waits for Corrective Maintenance (CM) to be performed. The job waits in
the machine queue until such time as the tool becomes available and job priority
rules establish the job as the next for processing. This describes the basic workings
of the simulation model. The next section will discuss how jobs are selected for
processing, and the rules which determine when tools are sent in for corrective or

preventative maintenance.

4.5.1.4 Dispatghing Rala

Only one dispatching rule, minimum slack, was tested in this model.
Minimum Slack (MINSLK) was chosen for several reasons. First, MINSLK has
been shown to be a robust dispatching rule over a number of performance measures,
particularly with mean and standard deviation of tardiness (Conway et al., 1967 ;
Lyman, 1993).

The other reason for the exclusive use of the minimum slack rule was based
on conversations with several plant managers, who were all using due date oriented
rules. Promised delivery dates drive the production planning process throughout the
plant. The choice of either Earliest Due Date (EDD) or MINSLK were considered.

MINSLK was chosen because it considers a job’s processing time.

109
4.5.1.5 Shag Qantrgl Hsuristics

In a DRC model, job priority and tool control decisions must be made
simultaneously. To initiate job processing, both machine and tool resources must be
free (idle). Both resources can then be seized and processing started. It is this
availability which is considered when establishing job priority. The specific control

heuristics will be presented in Sections 4.5.2.1 and 4.5.2.2.

4-5-1.6 ME

The average tool life, or mean time between tool failures, is based on an
arbitrary value. The arbitrary value is derived from discussions with various plant
personnel and managers. From these discussions, it became clear that few firms or
individuals knew with any certainty what the average time between failures were for
most forming tools. Frequently, companies use a standard life value for maintenance
policies which considers the number of production hours or parts produced. Even
with this information, tool life ranged from as low as fifty hours to a high of one
thousand hours. It was the lower bound of tool life that posed the greatest problems
for shop personnel. For this reason a value of 120 hours was selected for mean tool
life. Banerjee and Burton (1990) used a mean tool life (MTBF) of between 70 and
130 hours. The value of 120 hours was within their range and was considered short

enough to cause significant scheduling problems.

4.5.1.7 Egavaatjya Mg'ntanaasa Paint and Psrgntaga Estimate

llO

Preventive maintenance (PM point) is usually performed before the mean tool
life. The exact time is difficult to estimate because PM is based on tool inspection
procedures. For this model, sample runs will establish the exact value for the PM
point. A value 20 percent (100 hours) below the mean tool life would not be
unreasonable. This falls on the conservative (low frequency) side of Banerjee and
Burton’s (1990) PM policies.

For the variable PM heuristics, a value of 10 percent (above and below the
PM point) was selected for the PM range. The value of 10 percent is equivalent to
10 hours for VARHI, VARLO, and JDDTL, and 20 hours for VARPM and
MQBPM. The 10 percent value is that it corresponds to the mean processing time
(10 hours). On average, half the jobs will have a processing time which is less than
or equal to the variable PM range. Should a variable PM range of less than 10
percent be used, the benefits of variable PM would decrease. Future studies should

examine the issue of appropriate ranges for PM time.

4.5.1.8 Maiatanmga Saga; Time

Like mean tool life, maintenance service time is based on discussions with
production firms. Information on the length of maintenance service time was
available for both corrective (tool failure) and preventive maintenance. A mean PM
value of three hours was considered reasonable for most moderately complex
forming tools. The CM service time tends to range from no difference from PM, to

three times as long. The middle range (2.0 times) was agreed upon as appropriate.

111

This range is also consistent with past research which compare PM and CM (Kay,

1978; Banerjee and Burton, 1990).

4.5. 1.9 Eagamatar Sammagy

It should be noted that several of the parameters discussed in this chapter
(dispatch rules, mean tool life and service time), could also be examined as
experimental factors. This research has focused on control rules under stochastic

environments. Future research could examine variations of these parameters.

4.5 .2 Expgrimantal Eagtar Lgvels

Four experimental factors will be examined: job priority rules, tool control
rules, tool life distributions, and service time distributions. Table 4-3 illustrates the
factors and the number of levels for each factor. The following sections will describe

each level within a factor.

4.5.2.1 Jab Prigtjty Hgaristigs

Job priority heuristics determine which job in queue will be processed next.
Prior to the release of a machine and tool from production (job completion or tool
failure), the machine queue is reviewed. The review process prioritizes jobs in queue
by considering a number of factors including: dispatching rule, sequence
dependency, and tool condition. Listed below is a detailed explanation of the four

job priority heuristics. It should be noted that all heuristics must consider resource

112

 

 

Table 4-3 Design of Experiment

 

 

 

 

 

 

FACTORS TREATMENTS "“7““
LEVELS
01091170111913 nuLe
JOB PRIORITY uoueuce perenosucv
HIURISTIOS PRIORITY 95771113 RULE 4
sequence scHEOULmo
a TOOL LOAD
no 1m penronueo
FIXED Pu POLICY
“mane PM - Low
TOOL CONTROL \ARIABLE Pu - man 7
HEURISTICS \ARIABLE 1m — Lowmiou
uamnuauce IAGKLOO
.1017 cu: DATE - TOOL use
TOOL LIFE Low tarmac:
DISTRIBUTION HIGH “mane: 2
MAINTENANCE Low uummce
DISTRIBUTION HIGH nnmuce 2

 

 

 

 

availability for both machine and tool.

1.

Dispatching Rule with Tool Condition Information (DRTC). Priority is given

to the job with the lowest value minimum slack (MINSLK, due date minus

both job processing and current time) given the tool needed has sufficient life
(before PM). No consideration is given to the issue of setup changes, nor the
current machine setup. The objective is to process those jobs which have the

earliest due date, yet, not risk the possibility of tool failure while processing a

job. This rule is an adaptation of a simple dispatching rule by considering

finite tool life.

 

 

113
Sequence Dependency and Tool Condition (SDTC).

This rule combines the first job priority rule (DRC) with sequence dependency.
Priority is given to all jobs with the current tool setup (machine setup). Should
there be more than one job requiring the current tool-machine setup, then
priority is based on MINSLK dispatching. The current tool setup will remain,
until: 1) no jobs require the current tool, 2) the tool reaches its PM point, 3)
or the tool fails. If no jobs require the current tool setup, and if the tool has
not reached its PM point or failure, then this heuristic reverts back to job
priority DRTC. This rule attempts to reduce the frequency of tool changes
while still processing jobs based on due dates. It is an exhaustive rule which
attempts to process all jobs requiring the current tool setup prior to changing
tools. Such exhaustive rules have been found to be effective in past research
(Mahmoodi and Dooley, 1991). The problem with exhaustive rules is that they

tend to delay jobs with earlier due dates but require different tool setups.

Priority Setting Rule (PSR4).

This rule uses four levels of job priority with tool condition information.
Where SDTC can inflict delays in meeting job due dates because it is
exhaustive, this rule attempts to elevate this problem. PSR4 attempts to utilize
current setups and reduce excessive tool changes while simultaneously
considering the job’s due date. Should a job become past due (negative slack),

it is given priority over all other jobs regardless of the tool setup. Table 4-4

114

illustrates the four priorities.

As can be seen in Table 4-4, the first priority is given to past due jobs
which require the current setup. Second priority is given to those negative
slack jobs requiring a tool change. Jobs that are not past due are given lower
priority. This includes third priority for jobs requiring the current setup, and
fourth priority for jobs requiring a tool change. Melnyk et al. (1989) and
Ghosh et al. (1991) found this to be an effective rule, both in flow time and
due date performance. An element not considered by either Ghosh et al. or
Melnyk et al., that this rule considers, is tool condition (PM point). This rule
follows the same procedures as specified in SDTC. MINSLK dispatching is

used when more than one job exists within a priority.

Table 44 Priority Levels For Job Priority Rule PSR4.

 

 

 

 

 

 

Job Due Date Status
Tool Setup Past Due Not Past Due
Condition Negative Slack Positive Slack
Current Setup 1 3
Change Setup 2 4

 

 

 

 

Sequence Scheduling based on Tool Load (SSTL).
This rule is comparable to SDTC when sequencing jobs based on the current
setup. SSTL selects the job which requires the same tool setup based on

MINSLK and tool condition. However, it differs in the job selection process

115

when a tool change takes place. An analysis of the machine queue for
production time per tool is performed. The job processing time is summed for
each tool. The total processing time is then compared to the remaining tool life
(PM point) for each tool. Should total processing time be less than the
remaining tool life, then the algorithm continues to add job processing time
until either all jobs are exhausted or total processing time exceeds remaining
tool life. If total processing time is greater than remaining tool life, then the
algorithm subtracts the lowest priority job’s processing time from total
processing time for each needed tool. The closest match between processing
time and tool life without exceeding the tool life (PM point) is the next tool
selected for production. Jobs are then dispatched according to the MINSLK
dispatching rule.

The intent of this rule is to compare production demand to tool life.
This will reduce the number of tool setup changes and allow sequence

dependency to consider production demand.

45.2.2 W

Tool control heuristics determine when a tool is to be sent in for maintenance
(CM and PM). By controlling when a tool is sent in for maintenance, the rule
effectively controls tool availability. It should be noted that all tool control rules
examine a single tool at a time. Listed below is a detailed explanation of the seven

tool control rules.

116
No decision, allow tools to fail and perform CM (NOPM). The tool is

removed from production only when it has failed. PM is not considered. While
a penalty is incurred by a tool requiring CM, the frequency of maintenance
will be less than those rules which allow for PM. Several models have found

that a CM policy is preferred over PM policies (McCall, 1966).

Fixed Point In Time PM policy (FPTPM). This myopic PM policy uses a
single fixed value in the maintenance decision. Any time at or after this point
in time, the tool will be sent in for maintenance. If a tool is in production and
the tool researches the PM point, the tool will continue processing until the job
is completed. Oncecthe job is completed, the tool is then send in for PM. If,
however, the tool life is less than the PM point, the tool is kept available for

production.

Variable PM point with Low Frequency of Maintenance (VARLO). This
policy uses a range of time for the designated maintenance point. The range
for the variable point is 01*PM past the tools normal fixed PM point where a
= 10%. With PM point being equal to 100 hours of processing time, this rule
allows an extra 10 hours (100*10%) of tool usage. If the tool is past its PM
point, there are two options. First, if no job exists which can utilize the extra
time (past PM), then the tool is sent in for maintenance. The second option is

that if there is a job which can utilize the extra tool usage, then the tool

117

remains in production. If the current tool life is prior to the PM point, then the
tool remains in production.

The intent of this rule, as compared to FPTPM, is to allow more jobs
to be processed before a tool needs maintenance. While this rule increases the

risk of tool failure, a benefit is earlier job completion.

Variable PM point with high frequency of PM (VARHI). VARHI is the
opposite of VARLO in that it allows a tool to be sent in for maintenance prior
to the fixed PM point. The same 01 level is chosen (10%) as in VARLO. A
tool can be pulled for PM after 90 hours (PM-PM*a) of processing time. If
there exists a job which can utilize the tool for this time (up to PM point), then
the tool remains in production. If tool life is equal to or past 90 hours with no
job capable of utilizing the remaining time up to PM, then the tool is sent in
for PM. If the tools life is less than 90 hours, the tool remains in production
service.

The objective of this tool rule is to see whether early PM improves
performance over allowing a tool to continue processing. The tradeoff is, the

lower risk of tool failure versus more jobs past due because of tool availability.

Variable PM point (VARPM) combines VARLO and VARHI. This rule allows
for either early of late PM because the range for the maintenance decision

include 90 to 110 hours. If tool life is less than 90 hours, the tool remains in

118

production. If the tool life is greater than or equal to 110 hours, the tool is sent
in for maintenance. This leaves a 20 hour range in which the tool can remain
in production or be sent in for PM. Should no jobs exist which can use this
time (20 hours), then the tool is sent in for maintenance.

The purpose of this rule is to determine whether the combination of
VARLO and VARHI is better than either rule alone. By allowing greater
flexibility in the PM decision, the positive attributes of each rule are

incorporated.

Maintenance Queue Backlog and variable PM point (MQBPM). This rule
combines tool control VARPM with an additional consideration of the
maintenance queue. This rule preempts the selection of a job which, under
VARPM, would have utilized the time between 90 and 110 hours depending on
maintenance queue. If the maintenance queue is empty, the tool is sent in for
PM, regardless of whether a job is available to utilize the tool. The one
exception to this is if there is a job past due that can utilize the tool. The tool
is then used for processing. This rule is attempting to consider the possibility
of quick PM turn around as a means of reducing lost tool availability.
MQBPM also tends to stabilize the work load for the maintenance process.
MQBPM allows flexibility in PM decision which occurs at the shop
floor. Considering tool and shop conditions when deciding PM lowers the

decision making to the shop level where information is available.

119
7. Compare Job Due Date to Tool Life (JDDTL). This rule looks at the job’s due

date when deciding when to send a tool in for PM. JDDTL operates identically
as VARLO, with the exception of jobs which are past their due dates. JDDTL
keeps the tool in production until all past due jobs are processed, regardless of
tool life. Once all past due jobs are completed, the tool is then sent in for PM.
Processing of late jobs may take the tool life far beyond the PM point. This
assumes that the tool has not failed, in which case, the tool would be sent on
for CM. While this rule increases the risk of tool failure, it attempts to process
jobs which are late. JDDTL is the only rule which considers the due date

status of jobs.

4.5.2.3 Tml Lifs Distnbatign

Two levels of variance will be examined using a Normal distribution with a
mean value of 120 hours. The first level will look at low tool life variance (LOW).
The second level will consider the impact of high variance on system performance
(HIGH). The two levels, low and high, allow for comparison of the impact of
variance in tool life on shop performance. The LOW value corresponds to a 10
percent chance of a tool failure prior to the PM point. The HIGH value corresponds

to a 30 percent chance of tool failure prior to the PM point.

4.5.2.4 WWW

As with tool life, maintenance service time will have two levels of variance

120

using the log-normal distribution. The first level will examine low variance (LOW),
and the second level looks at high variance (HIGH). Using these two levels, a
comparison of the impact of maintenance time variance on shop performance will be

examined.

4.5.3 ModeLAssiLmntioas

The assumptions made in this model are similar to other hypothetical job
shops. The shop is composed of five non-identical machines and twenty non-identical
tools. Four tools are assigned to each machine with no flexibility or movement
between machines (see Table 4-2). Table 4-5 provides the assumptions used in this
research.

Table 4-5 Model Assumptions

 

Assumptions

 

Tools are machine specific, with no cross machine movement.

 

Once a tool fails, processing stops and the tool is sent in for maintenance.

 

Actual tool life is never known under a stochastic distribution.

 

Only tools fail, no machine failures.

 

Job preemption by other jobs is not allowed.

 

Upon tool failure, no damage occurs to either the job or machine.

 

Setup time is incurred only when tools are removed from a machine.

 

No setup time is incurred when changing jobs on a machine as long as
they are using the same tool (Kannan & Lyman, 1992).

 

 

Should a tool fail while processing a job, the job’s processing time is

121

decreased by the amount of time the job was in for processing until tool failure
occurred. The job’s remaining processing time will consist of the unfinished portion
of its total processing time. A job is not considered complete until it is fully
processed.

These assumptions parallel the assumptions of past DRC research, but contain
some noticeable exceptions regarding finite tool life. Parallel assumption can be
found in Melnyk et al. (1989), Melnyk and Lyman (1991), Kannan and Lyman

(1992), and Lyman (1993).

4.6 PERFORMANCE MEASURES

The performance measures selected in this study are those measures
considered to be common means of evaluating shop performance. The use of time in
system (mean and standard deviation) is a common measure of shop performance.
The use of time in system measures, provide an avenue for comparison to past DRC
results. For this reason, mean and standard deviation time in system measures were
used.

When discussing performance criteria with the interviewed managers, they
placed a high emphasis on meeting due dates. This ability to meet promised due
dates has also been emphasized by other researchers (Mahmoodi et al., 1991). For
this reason, the next three measures (three to five) focus on issues relating to
delivery performance.

The last measure, percentage of tool failures, tracks the ratio of CM to total

122

maintenances (CM plus PM). The uniqueness of this measure focuses on how it
should help explain variations in the other five performance measures. Percentage of
tool failures also allow a comparison of tool control heuristics with respect to tool
failures.

The six performance measures are:

1. Mean Time in System

2. Standard Deviation of Time in System

3. Mean Tardiness

4. Standard Deviation of Tardiness

5. Percentage of Jobs Late

6. Percentage of Tool Failures

The use of cost as a performance measure was not considered due to the
complex nature of forming tool costs (materials, etc.) and proprietary concerns.
Future research is needed to compare the costs of corrective maintenance (CM)

verses preventive maintenance (PM).

4.7 DATA COLLECTION

The simulation was conducted using SIMAN 3.5 software package on a
microcomputer. Fortran subroutines were used to customize the model and to collect
output data. The use of common random numbers were used for five of six random
number streams to reduce variance. A batch size of 2000 jobs was used to collect

data with 100 batches per run. This ensured both independence and normality so that

123

inferences are meaningful.

4.8 SUMMARY

The hypothetical DRC flow shop proposed for this study will be examined
using a simulation package. Issues pertinent to simulation methodology such as shop
operations, model assumptions, parameters, and performance measures were
presented. Other issues covered pertain to validity and statistical inferences (variance
reduction, normality, etc.).

The hypothesis and data analysis will be covered in Chapter 5. In addition,
Chapter 5 will also describe the statistical analysis of the data and residuals for

normality and homogeneity of variance.

CHAPTER 5

RESEARCH HYPOTHESES AND DATA ANALYSIS

5.1 INTRODUCTION

As outlined in Chapter 1, this research will investigate the effects that a finite
life resource has on a DRC shop. In addition, both job and tool heuristics which
consider finite resource life have been developed to evaluate shop performance. This
investigation will provide insights into how finite tool life impacts shop performance
and what control methods work best under varied conditions.

The first two sections of Chapter 5 describe the questions and hypotheses that
will provide the insight into the finite life resource and DRC shop that will be
discussed in Chapter 6. The last section of this chapter will entail an analysis of the
simulation data. This includes checking the residuals for normality and homogeneity
of variance, as well as describe the data transformation process. These steps are an
essential part of the process, so that conclusions regarding the hypotheses are

validated.

5.2 ANALYSIS OF EFFECTS

Each of the six performance measures, as described in Chapter 4, will be
statistically analyzed. This will include analysis of variance and multiple
comparisons. The questions to be answered by the analysis, is what factors (main

effects) significantly effect each measure. The expected results will show that all

124

125

four factors do significantly effect each of the six performance measures.

Also, higher order interaction will be examined for each measure. Should
higher order interactions be significant, then the use of linear contrasts will be used
to aid in explaining or understanding the interaction. By examining the performance
measures prior to answering the hypotheses, this will assist in the hypotheses tests

and conclusions.

5.3 RESEARCH HYPOTHESES

The hypotheses to be examined are based on answers obtained from the
questions developed in Section 1.3. These hypotheses were developed a_p;io_ri to the
simulation experiment and are non-orthogonal linear contrasts with a confidence
level of 0.05 for each comparison. The following hypotheses will examine: 1)
whether increased information in setting job priority for scheduling improves shop
performance, 2) which maintenance policy significantly affects shop performance,
and 3) whether the effects of tool and/or maintenance time alters the relative
performance of the various heuristics. The expected outcome will be discussed with
each specific hypothesis.

The following hypotheses will be tested using analysis of variance (ANOVA)
and linear contrasts using Tukey HSD multiple comparisons. Treatment means are
defined in Table 5-1 with i and j subscripts representing columns and rows from the
table respectively.

Hymthesisl: There is no significant difference between sequence dependent

126

 

__._ __. _‘,.

 

 

l

Job Priority

DRTC

SDTC

PSR4

SSTL

Table 5-1 Treatments in Experiment: Labeled by Column, Row

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ROW

 

 

 

 

Tool Life Variance ||
Low High
Maintenance Variance Maintenance Variance 1
Tool Control LOW High LOW High

NOPM #101 #111 #301 #401 l
FPTPM #102 #202 #302 #402 2
VARLO #103 #2113 #303 #403 3
VARHI #104 #134 #304 #404 4
VARPM #105 #295 #305 #405 5
MQBPM #106 #216 #306 #406 6
JDDTL #107 #207 #307 #407 .1 7
NOPM #103 #an #308 #4011 8
FPTPM II #109 #209 #309 #409 9
VARLO II #110 #210 #310 #410 10
VARHI II #111 #211 #311 #411 ll
VARPM I #112 #212 #312 #412 '2
MQBPM #113 #213 #313 #413 13
JDDTL i #114 #214 #314 #414 14
NOPM i #115 #215

FPT PM E #116 #216

VARLO #117 #217 #317 #417 17
VARI'II I #1111 #218 #318 #418 18
VARPM 4 #119 #219 #319 #419 19
MQBPM #120 #220 #320 #420 20
JDDTL : #121 1422. #321 #421 1 21
NOPM i #122 #222 #322 #422 22
FPTPM I #123 #223 #323 #421 23
VARLO I #124 #224 #324 #424 24
VARI-II H #125 #225 #325 #425 25
VARPM “ #136 #m #326 #426 26
MQBPM l #127 #221 #327 #427 27
JDDTL 1 #1211 #228 #3211 #4211 28

COLUMN 1 2 3 4

127

job priority rules which considers job due date versus tool condition when selecting

the next tool.

8 14 4 28
Ho:¢1=;;pij/28—z z pij/28=o

ng¢1¢0 i=1 j=22

This hypothesis attempts to answer whether the selection of the next job for
processing should be based on tool or due date, given that a tool change will take
place. The objective of sequence dependent rules is to enhance performance by
reducing the time consumed by setups. Job rules which reduce the number or
amount of setup time have been found to improve shop performance (Mahmoodi et
al., 1990, Kannan and Lyman, 1992). Consideration of tool condition (tool life)
first, permits the system to fully exploit the benefits of reduced setups. A study by
Lyman (1993) found that selecting jobs by looking at tool condition first enhance
performance over two other rules which examined a job’s due date first. The tool
condition first rule, TOOLIOB, performed best for all performance measures
because it tended to reduce the number of setups. Lyman’s model differs from this
research because tool life was deterministic with instantaneous replacement. For this
reason, the question of whether the same result was true with stochastic tool life and
tool maintenance delays was examined.

The expected outcome will reject the null hypothesis which indicates that
considering tool condition first, before job priority, does improve flow time
performance. The exception will be due date based performance measures. By

placing tool condition priority over job due date, delays in jobs requiring different

128

tools are expected.

W: The addition of information, specifically late due date status
and job interrelationship, does not significantly influence shop performance.

4 7 4 14 4 21
Ho:¢2=£ Z tug/28:22]);8 tin/28:2 .2 jig/29:0

1:1 j=1 1:1 1:15
H1:¢2¢0

This hypothesis will examine what effects, if any, each incremental form of
information used in job priority scheduling has on shop performance. The three job
rules which select jobs based on due date, and not tool condition are compared.
ANOVA will test to see if there is a significant difference between the three job
priority heuristics. Multiple comparison tests compare rules and show whether
additional information improves performance for job rules.

Use of information such as job interrelationships (tool requirement/sequence
dependency) allow the heuristic to reduce the number of tool setups. Hollier (1968),
Ghosh et a1. (1991), and Lyman (1993) have shown that certain forms of
information, such as job interrelationship, can enhance shop performance.

The use of due date status, like negative verse positive slack, has been shown
to be effective in improving shop tardiness (Melnyk et a1. 1989). This hypothesis
will determine if multi-level priority information used in the job’s selection (PSR4),
enhances shop performance over less information intense job priority heuristics
(DRTC and SDTC).

The outcome for this hypothesis is expected to show that additional

129

information used in setting job priority due dates has a significant benefit. Thus, the
null hypothesis will be rejected. By rejecting the null hypothesis, the incremental
benefit of additional information in setting job priority will surpass the benefits of
simplicity. In addition, results will show that PSR4 is better than SDTC, which is

better than DRTC. This ranking is dependent on the specific performance measure.

Hymthesisj: There is no significant difference between preventive
maintenance (PM) and corrective maintenance (CM).

While this hypothesis seems intuitively correct and of unquestionable

1:1 1:2 1=1 1:9 1=1 j=16 2:1 1:23
4 4 4 4

' (2 “11+ 1‘ 18+; 1‘ 115* I“ 122) /16‘O
1=1 1=1 i=1 1=1

outcome, this has not always been the case. Cases presented by Bojanowski (1984)
and Christer and Whitelaw (1983) cite the benefits of preventive maintenance over
corrective maintenance. The question of whether PM is preferred over CM also has
been addressed by Kay (1978) and Banerjee and Burton (1990). Kay showed that if
the penalty (added repair time for CM) was high enough, PM was preferred.
Banerjee and Burton’s model showed that, in many cases, PM decreased shop
performance (flowtime) over CM. They varied the PM point and found that PM

caused more delays than CM.

130
The expected outcome is that PM is significantly better than CM. When

looking at the test hypothesis, the predicted outcome is that PM policies (FPTPM
through JDDTL) are different than NOPM. Examining this question involves the use

of multiple comparisons to see which PM heuristic improves performance over CM.

am: There is no significant difference between using a fixed point

in time for preventive maintenance versus a variable (range) time.

I.”

4 4 4 4 4
Hoi¢4=(£ 1112+: “if; “116+: #123)/16’(2 "13'
1:1 1:1 2:1 i=1 '

i=1 J=3
C 12 4 19 4 26

+2 ”if: “if; E 1‘15) ”8‘0
1:1 =10 1:1 j=17 i=1 j=24

I-I1 : tbfio

Several articles have modeled the PM point as fixed in time (Kay, 1976;
Wells and Bryant, 1985; Banerjee and Burton, 1990). Under a fixed point in time
policy, PM can only take place after the fixed point in time is reached. Banerjee and
Burton (1990) tested several different fixed PM points and found that the shorter the
duration between PM points, the worse the shop performed.

Discussions with several plant personnel showed a more flexible policy. The
PM point was only a target or reference point. Scheduled PM dates tended to vary
around the designated PM point, depending on demand. This hypothesis will be used
to determine if a fixed PM point (FPTPM) performs as well as a variable PM point
(VARLO, VARI-II, and VARPM).

The expected outcome rejects the null hypothesis. This would add credence

131

to the shop personnel use of flexible PM points. The reason the null is rejected is
because variable PM considers job demand in queue. lf work exists for the tool,
VARLO and VARPM allow temporary postponement of the PM point, while
FPTPM does not. Also, VARHI and VARPM allow early PM if no demand exist
for the tool. Early PM can result in higher tool availability when production demand

exist.

W: There is no significant difference between early (VARHI) and

postponed (VARLO) variable preventive maintenance policies.

4 4 4 4
H0:4’5== (Z; I“ 13+Z; 1‘ MHz-:1 1‘ 1'17‘*Z:1 1‘ 124)/16

4 4 4 4
- (2 “14+; P 1111+: "113+: 11125) /16=O
i=1 1:1 i=1 i=1
H1:¢5ro

This hypothesis explores whether the policies of early tool withdrawal verses
postponed removal perform differently. Hypothesis 5 extends hypothesis 4 by
exploring the differences which exist between variable PM policies. An issue that
became apparent during the plant trip interviews was the lack of a clearly defined
range for the variable PM point. Since past research has not examined this issue,
hypothesis 5 will determine whether there is a significant difference between
performing PM early or late.

The difference between the early variable PM (VARHI) from the postponed
or late PM rule (VARLO) involves risk. VARLO increases the risk of tool failure

because the PM point is surpassed if the tool is needed. The benefit is a greater

132

utilization of the tool’s full life. The risk is the down time due to tool failure.
Results will show that the added risk is beneficial and thus rejects the null

hypothesis. VARLO adds sufficient benefits over the added risk of tool failure.

Hypothgsis 6: The preventive maintenance policy which examines
maintenance queue performs significantly better than the variable preventive
maintenance policies, which does not.

This hypothesis tests whether the addition of maintenance queue information

Hlubgo

in the PM decision helps elevate performance over the basic variable PM rule
(VARPM). Pate-Cornell et al. (1987) showed that unplanned early maintenance was
beneficial only when the shop was stable. Unplanned early maintenance performance
deteriorates quickly when shop variability (demand and maintenance time) increased.
The MQBPM rule attempts to elevate the impact of variability by reducing
maintenance delay. By allowing early maintenance only when the maintenance queue
is empty, tool maintenance delays are reduced.

Expected results will show that the null hypothesis will be rejected. This will

demonstrate that early PM based on maintenance queue can be beneficial. The

133

frequency of PM will increase slightly with the result of decreased maintenance
delay. Past studies by Banerjee and Burton (1990) and Pate-Cornell et al. (1987)
have shown that increased PM frequency causes shop performance to decrease. The

reason for this was that the models did not explicitly consider maintenance delays.

Hymthesis 7: The use of job due date in the decision of when to perform
PM significantly improve shop performance.

As with hypothesis 6, this hypothesis tests whether additional information in

the PM decision is beneficial. The additional information in this case is the status
(negative slack) of job’s in queue. With due dates being so critical a factor in
customer satisfaction, JDDTL rule is an attempt to enhance delivery performance.
The rule delays PM in favor of processing past due jobs. In effect, JDDTL is
increasing the risk of tool failure for the gains in improved delivery performance.
The expected outcome will show that the gains in delivery performance
outweigh the drawbacks of higher risk of tool failure (reject null hypothesis). By
considering issues external to the tool condition, information will improve shop

performance.

134
Hymthgsis 8: There is no significant difference between the performance of

the shop under low or high tool life variance.

2 28 4 28
Heaps: (2;; 1‘11) /S6- (1);}; Pu) /S6=0

H1:¢8#0

Hypothesis 8 tests whether tool life variance impacts shop performance. The
expected outcome is that tool life variance will significantly affect shop
performance, rejecting the null hypothesis. The variability of tool life will cause
increased tool failures that will result in the deterioration of all performance
measures. Tool failures will decrease tool availability. Past research has shown that
lower tool availability significantly impacts shop performance (Melnyk et al., 1989;

Ghosh et al., 1991).

Hymthasis 2: There is no significant difference between the high or low

maintenance service time variance.

28 28 28 28

Ho=¢9= <2 131292 1137') /56- (2 112142 1143') /56=o
j=1 J=1 1:1 j=1

H1:4>9¢0

As in hypothesis 8, this hypothesis test will determine if maintenance service
time significantly impacts performance under high and low variance. As is the case
with tool life variance, maintenance time variance can play an important role in shop

performance (Banerjee and Burton, 1990). Maintenance service time represents tool

135

down time (availability). As the down time variance increases, system performance
decreases (Vanderhenst et al., 1981), because tools are less available for production.
Variance increases maintenance queue causing greater waiting delays and lowers tool
availability. The expected outcome will be to reject the null hypothesis which
indicates that maintenance time is a significant factor in shop performance.

The nine hypotheses just discussed are based on the issues and questions
presented in Chapter 1. Table 5-2 shows the progression from the initial research
issues to hypotheses. Each step in the process involves refinement of the questions
until they become a focused hypothesis. The answers to these hypotheses can then be
directed back to the initial research question: How do we effective manage
operation in a DRC shop where the tooling constraint has a finite life and where

resource life and resource renewal are described using stochastic distribution?

5.4 POST HOC ANALYSIS

Hypotheses 8 and 9 address whether tool life and maintenance
variance are significant factors. What they do not consider is the impact of these two
forms of variance on the relative performance of the various heuristic. The mtg
analysis will examine the relative performance and thus, determine the robustness of

the various heuristics. The use of multiple comparisons will assist in this analysis.

5.4.1 T991 Lifa Vag'mas Analysis

By examining the relative performance of tool and job heuristics under two

136

Table 5-2 Refinement of Research Issues to Hypotheses.

 

 

 

|

_——1
Questions Hypotheses

mm

There is no significant difference

between job priority rules which

considers job due date before tool

 

How do we schedule How does additional information condition versus the a rule that looks
jobs? used in setting job priority affect at tool condition first. then job‘s due
shop performance? date.

 

The addition of information, like late
due date status and job
interrelationship, does not
significantly influence shop

 

 

performance.
I“ =
There is no significant difference
Does PM enhance performance between preventive maintenance and
over CM? corrective maintenance.

 

There is no significant difference
between using a fixed point in time
for preventive maintenance versus a
variable (range) time.

 

There is no significant difference

How do we schedule between early (VARHI) and
tools for production and How do various PM policies affect postponed (VARLO) variable
maintenance? sh0p performance? preventive maintenance policies.

 

The preventive maintenance policy
which examines maintenance queue
performs significantly better than
other variable preventive maintenance
policies.

 

The use ofjob due date in the
decision of when to perform PM
significantly improve shop

 

performance.

I I
How does variation in Does variance in tool life and There is no significant difference
resource life and renewal maintenance time affect the relative between shop performance under low
affect scheduling and performance of different job or high tool life variance.

assignment decisions? priority and tool control heuristics?

 

There is no significant difference
between shop performance under low
or high maintenance service time
variance.

 

 

 

 

 

 

levels of tool life variance, the robustness of the rules can be established. The

 

137

objective is to find out whether rules which consider more information in their
decision process perform as well under higher or lower tool life variation. This
objective is achieved by comparing job priority and tool control heuristics for low
tool life variance, to the relative performance of the heuristics for high variance.
Any change in relative performance indicates the rules are sensitive to tool life
variance.

Tool control heuristics which tend to increase tool failure risk (VARLO,
VARPM, and JDDTL) will diminish in the relative performance with increased tool
life variance. This conclusion is based on Levi and Rossetto (1978) and Bon (1980)
who determined that conservative tool strategies (reduce tool failure risk) remain
effective as tool life variance increases. Also, Vanderhenst et al. (1981) found that

strategies which increase CM over PM caused shop performance to deteriorate.

5.4.2 Maintanaas; Sawiga Vag'aaga Analysis

The objective is to determine which rules are robust under maintenance time
variance. Job and tool heuristics will be compared under low and high maintenance
variance. Any change in relative performance indicates the rules are sensitive to tool
life variance, thus, lacks robustness.

Job priority heuristic do not consider the maintenance process in establishing
priority. Thus, the relative ranking of job priority rules will not likely be effected.

As for tool control heuristics, maintenance service time variance will alter

the relative performance. MQBPM explicitly considers maintenance backlog and

138

should improve its relative performance as variance of maintenance time increases.
JDDTL, on the other hand, causes an increased risk of tool failure, thus higher
maintenance time lowers its performance. In addition, any rules (VARHI and
VARPM) which cause more frequent PM’s will increase in the relative performance

to other rules (Banerjee and Burton, 1990).

5.4.3 lglz-ngl Intaragtion Analysis

The objective of this analysis is to determine which combination of job and
tool rules provides robust performance. By examining the combined rules, it will
become apparent which factors, job priority or tool control, influences the relative
performance for each specific measure. Also, results will show how the different
combination of rules can alter the relative performance of either a good performing

job or tool rule.

5.5 DATA ANALYSIS PROCEDURES

Once the data for each experimental condition has been collected, the data is
analyzed to determine if it meets certain conditions for the statistical tests to be
valid. These conditions or assumptions must be met before the research questions
and am hypotheses and m analysis can be performed. A check of the
residuals for normality and homogeneity of variance is done to ensure that ANOVA
and other statistical tests are valid. For the data that violate these assumptions,

transformation using one of three methods was selected.

139

The validity of ANOVA is dependant on normality and homogeneity of
variance and, thus, requires additional analysis to ensure that the statistical tests are
meaningful. Minor violations of these assumptions do not preclude the use of
ANOVA (Neter et al., 1990). Should the residuals be non—normally distributed (by a
minor amount), the impact is a small decrease in the tests power. The implications
of reduced power occurs when the ANOVA p values are close to .05, resulting in

false conclusions.

5.5.1 flfssting fgr Ngrmality

To test for the assumptions of normality, the residuals were first analyzed
using a normal probability plot. The plot pits expected values against residuals.
While this method is helpful, it does not provide statistical proof. For this reason,
each treatment was tested using Filliben’s (1975) Probability Correlation Coefficient
Test (PCCT). The test uses the normal probability plot correlation coefficient which
is the product moment correlation coefficient between the ordered observation
residual X and ordered statistic medians M, which forms a normal distribution. The
rationale behind this test is that normality will tend to yield near linear normal
probability plots which will give near unity values for the probability plot correlation
coefficient. Comparison of the correlation with percentage points from the normal
probability plot correlation coefficient is evaluated. A statistically significant

correlation indicates that the data is normally distributed.

140
5.5 .2 Tasting far flamggansity gf Variance

Bartlett’s test was used to test for homogeneity of variance. The test
determines if there is a significant difference between sample variances. A major
concern for Bartlett’s test is any departure of the residuals from normality. In such
cases, Bartlett’s test can not be considered valid. If the performance measures were
non-normally distributed, the data was transformed and retested. The statistical

package SPSS 5.1 uses Barlett’s test as its mean for testing homogeneity of variance.

5.5.3 flfmsfgrmatign gf Dam

Should either assumption of normality or homogeneity of variance be
violated, the data for that performance measure was transformed. Neter et al. (1990)
recommends transforming data using one of three methods: log, square root, or
reciprocal. All three methods are used and the best method which provides a normal

distribution for residual is selected.

5 .5 .4 Rgsidaal Anflysis

Each treatment residual was examined for normality and homogeneity of
variance. The subscript values from Table 5-1 will be used as the treatment
reference for the following discussion.

For mean time in system, all but twelve treatments (301, 307, 308, 315, 322,
328, 401, 407, 408, 415, 422, 428) were normally distributed. The twelve

combinations of rules were all within 4% of what was required to accept the

141
hypothesis of normality. All but five (301, 307, 407, 415, 428) of the twelve

treatments had residual variances that were homogenous. Those treatments that
were heterogenous were among the worst performers for this measure.

For standard deviation of time in system, all but four treatments (314, 328,
414, 428) were non-normally distributed. These four treatments had PCCT values
within 3% of that required to accept the hypothesis of normality. All treatments had
equal variances for the residuals.

For mean tardiness, all treatment residuals were normally distributed. Twenty
treatments (102, 107, 114, 122, 128, 201, 207, 222, 228, 301, 307, 314, 322, 401,
407, 409, 410, 414, 422, 428) had heterogenous variance, but once again they were
among the worse performers.

For standard deviation of tardiness, all but fifteen treatments had normally
distributed residuals. Of these non-normal treatments, eight (102, 109, 209, 307 ,
309, 328, 405, 407) had PCCT values within 2% of that required to accept the
hypothesis of normality. The remaining seven (107, 207, 114, 128, 214, 314, 414)
were within 7% of the critical value. As for homogeneity of variance, only a few
treatments (110, 128, 227, 315 , 327, 415 , 427) violated this assumption and were
not among the top performers.

For percentage of jobs late, all but twenty f0ur treatments had normally
distributed residuals. Seventeen of the non-normally distributed treatments (103,
108, 116, 120, 124, 203, 208, 216, 220, 224, 305, 308, 316, 324, 403, 405, 408)

were within 5% of the critical value. The remaining seven treatments (201, 303,

142
330, 405, 416, 420, 424) were within 7% of the critical value. The heterogenous

variances again came from treatments (115, 116, 120, 121, 203, 216, 220, 221,
315, 321, 330, 403, 415, 421) which performed poorly.

For percentage of tool failures, only 41 treatments had normally distributed
residuals, with another 11 having PCCT values within 6% of the critical value. With
so few treatments normally distributed, the data was transformed. Of the three
methods used, log transformation significantly improved the fit of the data. Thirty
one treatments were non-normally distributed, but twenty six (102, 103, 107, 109,
110, 112, 114, 116, 121, 123, 124, 126, 128,202,203, 209, 307, 221, 228, 302,
323, 324, 326, 328, 402, 428) had PCCT values within 3% of that required to
accept the hypothesis of normality. The remaining five treatments (303, 321, 403,
409, 421) were within 6% of the critical value. Again, the treatments which violate
the assumption of homogeneity of variance tend to be poor performers (103, 107,

203, 207, 222, 223, 228, 322, 403, 407, 421, 423, 428).

5.5.5 Data Analysis Sammag

The impact of non-normal residual distributions cause a small increase in the
significance level while decreasing the power of the ANOVA test slightly. The large
sample size used in this model reduces the significance of non-normality. This also
hold true for heterogeneity of variance. ANOVA is still valid even when minor
violations in these assumptions are present. Only when p values are close to .05 will

false conclusion likely results. As for the large violations of the ANOVA

143

assumptions (percentage of tool failures), the data was transformed.

5.6 SUMMARY

This chapter discussed the issues and questions that the model will examine
in Chapter 6. A full factorial design was used in analyzing the model. A_pr_io_r_i
hypotheses is presented with an explanation and expected outcome. In the next
chapter, the use of ANOVA and Tukey HSD multiple comparisons will be used to

test the hypotheses and answer questions that were developed.

CHAPTER 6

EXPERIIVIENTAL ANALYSIS AND CONCLUSIONS

6.1 INTRODUCTION
After assuring the assumptions of ANOVA were not violated in Chapter 5,
this chapter will use three parts to analyze the data. This includes:
The significance of the main effects and interactions for each performance
measure. This will include both statistical (ANOVA tables) and graphical
(figures) analysis. Examining the questions of which factors significantly
impact performance will assist the hypotheses test. The identification of

specific heuristics and conditions which improve performance will be made.

The amt; hypotheses, which were expressed as non-orthogonal linear
contrasts, will require the use of multiple comparisons in their analysis if
higher order interactions are present. All of the hypotheses compare treatments
(levels) within a factor. While ANOVA will detect differences (main effect
significance), it does not lend itself to comparisons of levels within a factor.

For this reason, multiple comparisons will be used in the analysis.

39.21.1158: analysis will require further use of multiple comparisons. Specific
issues relating to the interactions of factors (tool control rules with job priority

rules under different variance) will be analyzed. The objective is to provide

144

145

insight into performance patterns of the various heuristics.
Upon completion of the analysis, specific conclusions will be drawn. This
includes a comparison to past research as well as new findings. The last part of this

chapter will discuss the direction that future research should consider.

6.2 ANALYSIS OF EFFECTS FOR PERFORMANCE MEASURES

The results presented are based on a full factorial design consisting of 112
treatment conditions. All statistical tests were performed with a confidence level (a)
of .05. The statistical package, SPSS 5.01 for Windows, was used to perform all
tests.

Common random number streams, as described in Chapter 4, are used as a
blocking factor in ANOVA. If the blocking variable is significant, experimental
error variance is reduced (Neter et al., 1990). NBATCH was used as the blocking
factor. If significant, then the batches are independent which.is airequirement of
ANOVA. In the ANOVA tests performed, NBATCH is significant for all
performance measures which demonstrates higher levels of treatment independence
(Mihram, 1974). The following discussion will examine the ANOVA tables and
figures (graphs) for each performance measure. Also provided is the treatment

means from which the figures are derived.

6.2.1 Mgaa Tia]: ia Sysmm

Table 6-1 contain the ANOVA results for time in system. Only two main

146

Table 6-1 Analysis of Variance for Mean Time in System

Source of Variation 5 D MS F

P
BATCH 146741.96 99 16304.66 4279.71 .000
MTOOL 9385.46 6 1564.24 410.59 .000
NLIFE 6.23 1 6.23 1.63 .201
NRULE 331.71 3 110.57 29.02 .000
NSERVE .01 1 .01 .00 .971
MTOOL*NLIPE 146.09 6 24.35 6.39 .000
MTOOL*NRULE 311.13 18 17.28 4.54 .000
MTOOL*NSERVE 10.28 6 1.71 .45 .845
NLIFE*NRULE 10.12 3 3.37 .89 .448
NLIFE*NSERVE .34 1 .34 .09 .765
NRULE*NSERVE 2.96 3 .99 .26 .855
MTOOL*NLIFE*NRULE 71.20 18 3.96 1.04 .413
MTOOL*NRULE*NSERVE 22.13 18 1.23 .32 .997
NLIFE*NRULE*NSERVE 1.52 3 .51 .13 .940
MTOOL*NLIFE*NSERVE 10.19 6 1.70 .45 .848
MTOOL*NLIFE*NRULE*NSERVE 41.45 17 2.44 .64 .862
(Model) 159429.2 119 1339.74 351.66 .000
(Total) 163239.02 11199 145.88
R—Squared a .977
Adjusted R-Squared = .974

effects, tool control rules (MTOOL) and job priority rules (NRULE), are significant.
The higher order interactions of MTOOL by NRULE and MTOOL by NLIFE were
also significant. All other main effects and interaction are not significant. Figure 6-
1(a-d) illustrate how the different tool control rules and job priority rules influence
mean time in system. The figures show that tool control rules cause more
fluctuations in performance than other factors. Those tool rules which provide
greater PM flexibility, such as VARHI, VARPM, and MQBPM, reduce mean time
in system. This holds true for all tool life and maintenance time variances. The
worst performing tool control rule is NOPM.

As for job priority rules, SSTL consistently performs worse than any of the

other job rules. This holds for most tool control rules or variance levels. The best

147

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 6-2 Treatment Mean for Mean Time in System
TOOUMAINT JOB TOOL RULES
VARIABILITY RULE NOPM FPTPM VARLO I vxam VARPM MQBPM IDDTL__II
DRTC 61.90 56.28 58.29 53.27 54.80 54.42 59.21
TOOL-LOW SDTC 63.39 56.18 58.96 53.55 53.67 53.44 59.77
MAM-LOW PSR4 62.99 60.77 58.98 53.37 54.76 53.33 56.82
SSTL 66.04 56.34 59.78 53.96 56.65 54.17 60.47 1'
DRTC 62.67 57.77 59.39 54.71 55.61 54.48 60.05
TOOL-HIGH SDTC 65 .80 58.27 59.53 54.29 54.70 54.24 59.57
MAINT-LOW PSR4 62.29 57.68 58.76 54.54 55.74 54.15 58.41
SSTL 64.04 58.58 59.99 55.02 56.77 54.55 60.98
=— in:
DRTC 62.07 55.46 58.32 53.27 54.96 54.90 59.73
TOOL-LOW SDTC 64.42 56.87 58.98 53.55 53.67 53.44 58.80
MAINT-HIGH PSR4 63.33 55.50 58.65 53.37 54.73 53.32 57.79
I SSTL 66.69 56.41% 59.44 53.93 56.66 54.27 61.35
DRTC 62.71 58.27 58.21 55.55 55.59 54.55 59.78
TOOL-HIGH SDTC 63 .36 S9 .03 59.10 54.30 54.70 54.07 60.25
MAINT-HIGH PSR4 63.10 57.72 59.77 54.54 55.04 54.12 58.57
1 SSTL 63.76 58.57 59.84 55.59 57.39 54.41 62.13
Legend:

Tool Control Rules: NOPM - no PM performed. allow tool to fail.
FPTPM - fixed PM point, PM occurs only after PM point.
VARLO - variable PM point, allow postponed PM up to 10% beyond.
VARHI - variable PM point, allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM, 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but consider job due date.
DRTC - prioritized by due date
SDTC - prioritized by sequence dependency, then due date.
PSR4 - prioritized using four priority levels
SSTL - prioritized by sequence dependency. then by tool condition.
TOOL-LOW: low tool life variance.
TOOL-HIGH: high tool life variance.
MAINT-LOW: low maintenance service time variance.
MAINT-HIGH: high maintenance service time variance.
—

Job Priority Rules:

performing job priority rule is PSR4, followed by DRTC (Table 6-2). However, this

rule does not hold true in all cases (Figures 6-1 a-d).

148

 

 

MeanTime in System

Figureb6-1a Comparison of Control Rules

on cl/Low Muntenance Variance

 

3

 

8

8

8

8

8

 

O

 

VAR”
ToolControlRule

-DR'IC -smt -PSR4 :lssn.

 

 

 

Mran Tim e m System

Figure 6-1 b Comparison of Control Rules

figh Tool/Low Maintenance V ariance

 

 

 

 

 

 

0

mm FPTPM VAR}. 0 VARPM m
Tool Control Rule

-DR’I‘C -smc -1>sn4 -ssn.

 

 

 

149

 

 

Figure 6-1c Comparison of Control Rules
Low Tool/Kg}: Maintenance V srranee

 

 

 

 

 

 

 

 

 

 

 

 

 

 

80

70
E 60
a
(5‘ 50
E
0 0
E
c 30
I!
I:
2 m

10

0

mm VARLO VARHI VARPM ”PM man.
Tool Control Rule
-DR‘IC -smc -PSR4 [:3 5511.
Figure 6-1d Comparison of Control Rules
thTool/EghMarntenanceVariance

70

60
3 so
65‘
E 40
E 30
S
g 20

 

mm VARLO

VARHI VARPM PM rwn
Tool Control Rule m

-DR'PC -sn'rc -PSR4 C3551}.

 

 

 

150

6.2.2 Smdard Deviatign of Time in System

Table 6-3

of Time in System

Analysis of Variance for Standard Deviation

Source of Variation 8 D MS F p
BATCH 290635.78 99 32292.86 3127.17 .000
MTOOL 14698.24 6 2449.71 237.22 .000
NLIFE 7.15 1 7.15 .69 .405
NRULE 19202.93 3 6400.98 619.86 .000
NSERVE .00 1 .00 .00 .993
MTOOL*NLIFE 98.00 6 16.33 1.58 .149
MTOOL*NRULE 8758.29 18 486.57 47.12 .000
MTOOL*NSERVE 18.21 6 3.04 .29 .940
NLIFE*NRULE 10.57 3 3.52 .34 .796
NLIFE*NSERVE .31 1 .31 .03 .862
NRULE*NSERVE .83 3 .28 .03 .994
MTOOL*NLIFE*NRULE 123.12 18 6.84 .66 .850
MTOOL*NRULE*NSERVE 137.48 18 7.64 .74 .772
NLIFE*NRULE*NSERVE 1.42 3 .47 .05 .987
MTOOL*NLIFE*NSERVE 36.25 6 6.04 .59 .742
MTOOL*NLIFE*NRULE*NSERVE 149.65 17 8.80 .85 .632
(Model) 341908.01 119 2873.18 278.23 .000
(Total) 352234.56 11199 314.78

R-Squared = .971

Adjusted R-Squared a .967

The ANOVA Table 6-3 provides an analysis for standard deviation of time in
system. The main effects of MTOOL and NRULE are significant as well as higher
order interactions between MTOOL and NRULE. All other main effects and
interactions are not significant. Figures 6—2 (a—d) show that job priority rules which
are due date based (DRTC and PSR4) cause less variation (standard deviation) in the
time in system than rules which promote sequence dependency (SDTC and SSTL).

The standard deviation of time in system is reduced by tool control rules
which promote more frequent (early) PM (Table 64). Figures 6-2 (a-d) show that
the combination of job due date based priority rules with frequent PM tool rules

consistently provide better performance. This combination includes: DRTC with

151

 

Table 6-4 Treatment Means for Standard Deviation for
Time in System

 

TOOUMAINT JOB TOOL RULES —||
VARIABILITY RULE

 

 
   
   
 

VARHI

        

DRTC 46.46 55.77 44.95 42.79 59.50

 

TOOL-LOW SDTC 57.79 60.46 55.33 49.37 50.21 50.12 65.53

 

MAlNT-LOW PSR4 47.47 45 .99 49.26 43.95 46.67 44.02 48.57

 

4.57

SSTL 6 56.29 55.55 54.35 53.39 52.38 63.41
I#—-—AF F=é=l====l=fi=g=

DRTC 46.08 57.80 45.71 41.95 43.46 46.14 61.87

-_7

 

TOOL-HIGH SDTC 61.57 60.98 56.20 50.27 51.38 51.16 63.44

 

MAM-LOW PSR4 47.29 47.36 49.01 45.34 47.26 44.81 48.87

 

 

 

 

SSTL 65.33 58.57 55.01 55.91 52.84 55.78 65.64
rm ===—_..-*——-‘— ...____=_——-¥_,r

DRTC 45.95 54.63 44.98 40.88 42.99 49.67 62.14

 

TOOL-LOW SDTC 59.55 60.59 55.34 49.37 50.21 50.06 64.17

 

MAINT-HIGH PSR4 47.43 46.01 48.81 43.95 46.60 44.07 49.34

SSTL 65.34 56.24 55.85 54.04 53.45 51.58 64.08
F Ea ;

DRTC 45.99 57.43 43.97 43.35 57.17 49.47 62.40

 

 

 

TOOL-HIGH SDTC 57.90 62.88 54.91 50.29 51.44 50.91 65 .22

 

MAINT-HIGH PSR4 47.64 47.34 49.02 45.32 46.69 45 .21 48.82

' SSTL 63.72 58.89 55.70 57.16 54.28 53.07 63.94

Legend:
Tool Control Rules: NOPM - no PM performed, allow tool to fail.
FPTPM - fixed PM point, PM occurs only afler PM point.
VARLO - variable PM point. allow postponed PM up to 10% beyond.
VARHI - variable PM point. allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM, 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but consider job due date.
lob Priority Rules: DRTC - prioritized by due date
SDTC - prioritized by sequence dependency. then due date.
PSR4 - prioritized using four priority levels
SSTL - prioritized by sequence dependency, then by tool condition.
TOOL-LOW: low tool life variance.
TOOL-HIGH: high tool life variance.
MAINT-LOW: low maintenance service time variance.
MAINT 4116“: high maintenance service time variance.
—

 

 

 

 

 

 

 

 

 

 

 

VARHI, MQBPM, VARLO, and VARPM; and PSR4 with VARHI and MQBPM.

152

 

 

9d Dev Time in System

FigureLo 6-2a Comparison of Control Rules
wTo ollLow Mamtenance Van lance

 

 

 

 

 

 

mm VARLO VARHI VARPM mm m1.

Tool Control Rule

-DR'IlC -smc -PSR4 E15511

 

 

 

ad Dev Time in System

Figure 6-2b Comparison of Control Rules
Egh Tool/Low Maintenance V arrance

 

 

 

 

4s;

‘ 7%”: 1.2335'
Aw» 35:: .

 

FPI'PM VARLO VARH! VARPM NIEPM ML
Tool Control Rule

-muc -snrc -PSR4 1:35511.

 

 

 

153

 

 

Figure 6-2c Comparison of Control Rules
Low Tool/Sigh Mamtenance Variance

 

 

 

 

 

 

 

 

 

 

 

70
so
a
:3. so
0'?
E a)
E
E", 30
3
c:
.5 no
05
10
0
mm vruuo mun VARPM mam rum
ToolControlRule
-DR’DC -snrc -PSR4 E35511
Figure 6-2d Comparison of Control Rules
thTool/thMamtenance Variance
10
so
a
8 so
«71‘
E a]
E
E" 3'0
2
a
1, 20
as

 

mm VARLO

 

VARM mam m

VARHI
Tool Control Rule

-1>ch -5131c -PSR4 -5511.

 

 

 

 

154

The tool control rule, JDDTL, consistently caused poorer performance when
combined with all, but PSR4 job priority rules. Tool rule, NOPM, also performed

poorly when combined with sequence dependent rules SDTC and SSTL (Figure 6-
2a).

6.2.3 Mmlardiness

Table 6-5 Analysis of Variance for Mean Tardiness

Source of Variation S D MS F p
BATCH 432155.42 99 48017.27 3241.60 .000
MTOOL 37338.53 6 6223.09 420.11 .000
NLIFE 8.41 1 8.41 .57 .451
NRULE 105296.46 3 35098.82 2369.49 .000
NSERVE - 1.18 1 1.18 .08 .778
MTOOL*NLIFE 186.09 6 31.01 2.09 .052
MTOOL*NRULE 23867.35 18 1325.96 89.51 .000
MTOOL*NSERVE 71.98 6 12.00 .81 .562
NLIFE*NRULE 11.56 3 3.85 .26 .854
NLIFE*NSERVE .83 1 .83 .06 .813
NRULE*NSERVE 1.20 3 .40 .03 .994
MTOOL*NLIFE*NRULE 303.71 18 16.87 1.14 .308
MTOOL*NRULE*NSERVE 137.89 18 7.66 .52 .951
NLIFE*NRULE*NSERVE .16 3 .05 .00 1.000
MTOOL*NLIFE*NSERVE 24.38 6 4.06 .27 .949
MTOOL*NLIFE*NRULE*NSERVE 267.48 17 15.73 1.06 .387
(Model) 609399.55 119 5121.00 345.71 .000
(Total) 624212.39 11199 557.83

R-Squared x .976

Adjusted R-Squared - .973

Table 6-5 shows the results of the ANOVA test for mean tardiness. The only
main effects and interactions that are significant is MTOOL and NRULE and
MTOOL by NRULE. Figures 6-3 (a-d) show that due date oriented job priority rules
(DRTC and PSR4) tend to lower mean tardiness more than sequence dependant rules
(SDTC and SSTL). The PSR4 rule performed best among job priority rules (Table

6-6). The figures also illustrate how SDTC and SSTL rules are consistently poor

 

155

 

 

 

 

 

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 6-6 Treatment Means for Mean Tardiness
TOOUMAINT TOOL RULES H
VARIABILTI‘Y NOPM FPTPM VARLO worm MEI

=——T'——'_—_'
44.61 61.39 43.52 38.64 40.74 41.95 69.79
TOOL-LOW 59.35 65.30 60.23 52.90 53.61 54.17 72.84
MAINT-LOW 41.25 35 .73 40.45 34.45 37.56 34.35 37.74
70.21 61.67 59.40 58.38 57.59 53.57 73.99
w*=—_—_——
DRTC 44.10 66.64 44.09 39.59 41.48 39.57 73.27
TOOL-HIGH SDTC 64.97 65.79 59.92 53.61 55.17 54.56 68.81
MAINT-LOW PSR4 41.05 37.98 39.87 36.98 38.65 35 .62 38.30
1 SSTL 71.79 63.68 58.20 if: 57.14 56.00 77.20
DRTC 43.54 63.86 43.55 38.64 40.96 43.60 73.79
TOOL—LOW SDTC 62.27 65.21 60.16 52.90 53.61 53.66 72.77
MAlNT-HIGH PSR4 41.03 36.07 39.91 34.45 37.71 34.24 40.12
I SSTL 70.76 63.05 59.72 58.39 57.55 53.85 74.75 J
DRTC 44.83 65.09 42.26 41.27 59.93 41.29 74.79
TOOL-HIGH SDTC 60.17 68.24 58.75 53.76 55.48 54.60 71.64
MAINT-HIGH PSR4 41.58 37.24 39.92 36.96 38.07 , 35.72 39.08
SSTL 70.69 66.10 59.13 59.93 58.21___55.50 74.50 g

 

Legend:

Tool Control Rules: NOPM - no PM performed, allow tool to fail.
FPTPM - fixed PM point, PM occurs only afier PM point.
VARLO — variable PM point, allow postponed PM up to 10% beyond.
VARHI - variable PM point. allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM, 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but consider job due date.
DRTC - prioritized by due date
SDTC - prioritized by sequence dependency. then due date.

Job Priority Rules:

 

 

 

 

PSR4 - prioritized using four priority levels
SSTL - prioritized by sequence dependency. then by 1001 condition.
TOOL-LOW: low tool life variance.
TOOL-HIGH: high tool life variance.
MAINT-LOW: low maintenance service time variance.
MAlNT-HIGH: high maintenance service time variance.
—

performers. This holds for all tool control rules.

When tool control rules were combined with job priority rules, DRTC,

156

 

 

Figure 643a Comparison of Control Rules

Low To :1wa Maintenance V arrance

 

 

 

 

 

 

 

 

 

S

O

 

 

 

 

 

mm VARLO

-DR'I‘C -sm1c -PSR4 -5511.

 

80
70
60
E 50
*6
33 an
5 1:11
2 3"
3°
10 If;
u .
FPTPM VARLO VARHI VARPM m 113cm.
ToolControlRule
-m<1c -51)11c -1>sa4 E3 5511.
Figure 6-3b Comparison of Control Rules
K311 To ol/Low Maintenance Variance
W
80
70
2 60
i
ii 50
1-4
g 0
U
2 30
20

 

VAR.” VARM . rum.
Tool Control Rule W

 

 

 

157

 

 

Mean Tardiness

Figure 6-3c Comparison of Control Rules
LowTo 01/th Mantenance Variance

 

 

 

 

 

 

 

 

 

mm VARLO VARHI VAW MQBPM man.

To 61 Control Rule

-DKIC -smc -1>sa4 [235511.

 

 

 

 

Mean Tardiness

88838

Figurem 6-3d Comparison of Control Rules

UthhMarn tenance Van ant:

 

 

 

 

 

mm VARID VARH VARPM m .rer
Tool Control Rule

-DR’IC -snrc -ps1>.4 -5511.

 

 

158
SDTC, and SSTL, results showed that flexible PM point rules (VARHI and

VARPM) work better than when combined with FPT PM or NOPM tool rules. The
exception to this situation involves the PSR4 job rule. The PSR4 rule tends to

equalize tardiness performance of all tool control rules.

6.2.4 Stande Deviation of Tardiness

Table 6-7 Analysis of Variance for Standard Deviation
of Tardiness

Source of Variation S D MS F p
BATCH 710918.21 99 78990.91 1932.57 .000
MTOOL 43265.75 6 7210.96 176.42 .000
NLIFE 23.11 1 23.11 .57 .452
NRULE 71851.41 3 23950.47 585.96 .000
NSERVE .01 1 .01 .00 .985
MTOOL*NLIFE 191.34 6 31.89 .78 .585
MTOOL*NRULE 35981.86 18 1998.99 48.91 .000
MTOOL*NSERVE 48.12 6 8.02 .20 .978
NLIFE*NRULE 169.78 3 56.59 1.38 .246
NRULE*NSERVE 14.51 3 4.84 .12 .949
NLIFE*NSERVE .43 1 .43 .01 .919
MTOOL*NLIFE*NRULE 521.23 18 28.96 .71 .805
MTOOL*NRULE*NSERVE 588.13 18 32.67 .80 .703
MTOOL*NLIFE*NSERVE 90.85 6 15.14 .37 .898
NLIFE*NRULE*NSERVE 13.60 3 4.53 .11 .954
MTOOL*NLIFE*NRULE*NSERVE 445.63 17 26.21 .64 .860
(Model) 884508.81 119 7432.85 181.85 .000
(Total) 925382.40 11199 826.97

R-Squared x .956

Adjusted R-Squared = .951

 

Table 6—7 shows the ANOVA results for standard deviation of tardiness. The
only significant main effects and interactions are MTOOL, NRULE, and NRULE by
MTOOL respectively. The due date based job priority rules, DRTC and PSR4,
perform better than sequence dependent rules SDTC and SSTL (Table 6-8). As was

the case with mean tardiness, the PSR4 job rule provides consistent performance,

159

Table 6-8 Treatment Means for Standard Deviation of

 

Tardiness
TOOL/MAINT JOB TOOL RULES
VARIABILU‘Y RULE NOPM FPTPM v0.1.0 VARHI VARPM MQBPM JDDTL ll

 

39.62 63.74 38.10 34.42 37.57 51.51 67.41

 

TOOL-LOW SDTC II 58.21 74.68 57.07 50.10 52.65 52.80 78.70

 

MAINT-LOW 44.93 40.29 46.83

 

57.28 61.56 69.07

37.58 47.38 72.03

 

 

TOOL-HIGH SDTC 63.77 73.54 58.32 51.15 54.06 54.41 76.79

 

MAM-LOW PSR4 38.37 43.14 46.11 41.95 44.90 41.12 46.44

 

SSTL 69.64 64.01 57.54 66.44 56.15 70.43 74.27
—_.____J

 

 

DRTC 38.62 63.03 38.10 34.42 37.77 53 .73 71.27

 

TOOL-LOW SDTC 60.86 73.62 57.03 50.10 52.65 52.68 77.28

 

MAlNT-HIGH PSR4 38.31 42.32 45.77 39.76 44.51 40.38 47.73

 

 

 

 

 

 

SSTL 64.88 60.82 59.49 64.14 57.37 59.54 70.12
— — ———————— EI==__= —-._ .____.r
DRTC 38.39 67.31 36.69 37.55 68.94 55.85 73.22

 

TOOL-HIGH SDTC 58.96 76.14 56.17 51.11 54.10 53.58 79.77

 

MAINT-HIGH PSR4 38.31 43.50 46.16 41.87 44.15 42.23 46.06

 

SSTL 66.20 64.26 59.07 68.94 58.20 64.22 69.42
are -. -~—- --—~—~ - »———————- == ==

Legend:

Tool Control Rules: NOPM - no PM performed, allow tool to fail.
FPTPM - fixed PM point. PM occurs only afier PM point.
VARLO - variable PM point, allow postponed PM up to 10% beyond.
VARHI - variable PM point, allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM. 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but consider job due date.

Job Priority Rules: DRTC - prioritized by due date
SDTC - prioritized by sequence dependency, then due date.
PSR4 - prioritized using four priority levels
SSTL - prioritized by sequence dependency. then by tool condition.

TOOL-LOW: low tool life variance.

TOOL-HIGH: high tool life variance.

MAINLLOW: low maintenance service time variance.

MAINT-HIGH: high maintenance service time variance.

 

 

 

 

 

 

 

 

 

regardless of the tool control rule used (Figure 6—4a). The worst performing job

160

 

 

Figure 6-4a Comparison of Control Rules
Low Tool/Low Maintenance Variance

 

 

 

 

 

 

 

 

 

 

5

O

 

 

 

 

 

 

 

W
80
_ 7O
'4‘.
.5 60
3
l"- 50
to.
O
p 0
a”
'6 30
05
11
10
0
mm VARLO VARHI VARPM 1.4mm m
ToolControlRule
-mmc -snrc -PSR4 C3 5511.
Figure 64b Comparison of Control Rules
3311 Tool/Ion? Maintenance Variance
W
80
a 70
E
‘8 60
‘3
{-5 50
“6
p 0
0
Q
.u- I}
05
I)

 

WARM
Tool Cvontrol Rule

-DR'IC -51>11c -psa4 [35511.

 

 

 

161

 

 

Figure 6-4c Comparison of Control Rules
Low Tool/Kali Maintenance Variance

 

 

 

 

 

 

 

 

 

 

E

O

 

 

 

 

 

FPTPM VARLO

 

90
80
a 70
E
E 60
C
l-t 50
“o‘
p l)
0
a
.6 3)
or
m
10
0
FPTPM VARLO VAN-ll VARPM mm rum
Tool Control Rule
-mu1c -sn1c -P$R4 [:1 5511.
Figure 64d Comparison of Control Rules
High To ol/l-Egh Maintenance V arrance
90
80
u 70
‘E 60
fi 50
“5
> (1
0
a
‘5 30
as
I!

 

mam morn

VARHl
Tool ControlRule

-DRTC -5011c -PSR4 E35511.

 

 

 

162

priority rule was SSTL.

Variable PM tool control rules, VARHI, VARLO, and VARPM, performed
best when combined with due date job priority rules DRTC and PSR4 (Figures 6-4
a-d). As for SDTC, this rule tends to perform worse when combined with fixed PM

point rules (FPTPM and J DDTL).

62.5 W

Table 6-9 Analysis of Variance for Percentage of Jobs

Late
Source of Variation S D MS F p
BATCH .51 99 .06 53.27 .000
MTOOL 106.90 6 17.82 16780.70 .000
NLIFE .03 1 .03 31.39 .000
NRULE .16 3 .05 49.01 .000
NSERVE .00 1 .00 .01 .925
MTOOL*NLIFE .81 6 .14 127.55 .000
MTOOL*NRULE .33 18 .02 17.17 .000
MTOOL*NSERVE .00 6 .00 .07 .999
NLIFE*NRULE .00 3 .00 .19 .903
NLIFE*NSERVE .OO 1 .00 .01 .911
NRULE*NSERVE .00 3 .00 .04 .990
MTOOL*NLIFE*NRULE .01 18 .00 .54 .942
MTOOL*NRULE*NSERVE .00 18 .00 .05 1.000
NLIFE*NRULE*NSERVE .00 3 .00 .00 1.000
MTOOL*NLIFE*NSERVE .00 6 .00 .07 .999
MTOOL*NLIFE*NRULE*NSERVE .00 17 .00 .04 1.000
(Model) 125.69 119 1.06 994.82 .000
(Total) 126.75 11199 .11
R-Squared . .992
Adjusted R-Squared a .991

Table 6-9 shows the ANOVA results for percentage of jobs late. All the main
effects, except NSERVE, are significant. The higher order interactions of MTOOL
by NLIFE and MTOOL by NRULE are also significant. The sequence dependent

rules, SDTC and SSTL, provide lower percentage of jobs late for most treatments.

163

 

Table 6-10 Treatment Means for Percentage of Jobs Late

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TOOUMAINT JOB TOOL RULES

VARIABILITY RULE NOPM FPTPM VARLO vanm VARPM MQBPM | 10011. I

DRTC 33.1 24.3 30.2 26.3 27.4 27.0 25.2

TOOL-LOW SDTC 31.4 24.2 27.5 24.7 24.6 24.3 25 .4

MAM-LOW PSR4 38.3 33.1 34.8 31.4 31.7 31.2 34.3
28.6 . 26.5 24.2 26.5 J

DRTC 34.0 24.0 31.1 27.9 28.3 27.9 24.7

TOOL-HIGH SDTC “ 32.2 25.8 28.3 25.2 25.3 24.9 26.4

MAINT-LOW PSR4 37.7 35.1 35 .0 31.7 32.4 31.7 35.3

SSTL 30.1 27.1 29.1 24.5 26.6 24.0 26.3
===— F

DRTC 33.8 23.2 30.2 26.3 27.6 27.0 24.7

TOOL-LOW SDTC 31.6 24.9 27.6 24.7 24.6 24.4 24.9

MAINT-HIGH PSR4 38.8 33.1 34.7 31.4 31.7 31.3 34.3
SSTL 32.3 25.4 28.1 23.8 26.5 24.2 27.1 J

i DRTC 33.8 24.8 30.7 28.2 24.9 27.7 24.0

TOOL-HIGH SDTC 31.0 25 .8 27.9 25 .2 25 .3 24.8 26.1

MAINT-HIGH PSR4 38.3 35.1 35.0 31.6 31.8 31.5 35.3

SSTL 29.9 26.2 28.8 24.9 27.0 24.0 27.8

 

 

Legend:
Tool Control Rules: NOPM - no PM performed, allow tool to fail.
FPTPM - fixed PM point, PM occurs only after PM point.
VARLO - variable PM point. allow postponed PM up to 10% beyond.
VARHI - variable PM point, allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM, 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but consider job due date.
Job Priority Rules: DRTC - prioritized by due date
SDTC - prioritized by sequence dependency. then due date.
PSR4 — prioritized using four priority levels
SSTL - prioritized by sequence dependency. then by 1001 condition.
TOOL-LOW: low tool life variance.
TOOL-HIGH: high tool life variance.
MAINT-LOW: low maintenance service time variance.
MAINT 41161-1: high maintenance service time variance.
—

The PSR4 job priority rule performs worse than any other job rule. This is a sharp

contrast to PSR4’s performance on mean and standard deviation of tardiness. The

164

 

 

FigureLo 6-5a Comparison of Control Rules
wTo ollLow Maintenance Variance

 

 

 

 

    

 

 

 

1 0V.

 

 

 

 

 

 

 

507.
(W.
3
C
.4
3 30%
1“
O
I)
w
3 207.
3
‘6
e.
107.
0%
VAR}! PM
Tool Control Rule M3
-DR’DC -sorc -PSR4 C3 5511.
Figure 6-5b Comparison of Control Rules
15311 To ol/Low Maintenance V ariance
50%
0%
2.’
U
._l
B
,2 307.
be
0
U
2"
E‘ 207.
3
3
o.

VARHI
To 61 Control Rule

-DRTC -51)11c apsru E35511.

 

 

 

165

 

 

Figure 6-5c Comparison of Control Rules
Low Tool/Hall Mamiemnce V mance

 

 

   

 

 

 

 

 

l 07.

 

 

 

 

FPTHW VARLO

 

507.

m
8
I
p.)
3 m 3' l
‘5' i 7}- _l
§ ~
g 207. l
3
i II I I

10% ll I I

07.

FPTPM VARLO VAN-ll VARPM JIDTL
Tool Control Rule m
- mm: -smc -PSR4 5511.
Figure 6-5d Comparison of Control Rules
532: Tool/fish Maintenance Variance

507.

07.
8
l
.4
32 307.
‘a-a
O
0
9
E 207.
S

 

VARHI m morn.
Tool ControlRule M26

-DR'PC -SD'IC -PSR4 E3 SSTL

 

 

 

166

reason for the abrupt difference between tardiness measures and percentage of jobs
late involves the specifics of the measure. The percentage of jobs late measurement
only counts whether a job is late or not. Whereas, the mean and standard deviation
of tardiness measures the degree of late jobs. Thus, the job priority rule, PSR4, has
a greater percentage of jobs late, but the mean and standard deviation of late jobs is
lower (Figure 6—5a). The reverse is true of the sequence dependent rules, SDTC and
SSTL, which cause fewer late jobs but tend to be much later (Figures 6-5 a-d).

The tool control rules which allow PM perform better than the tool rule
NOPM (Table 6-10). The best performing tool control rules are those which

promote early PM (VARHI, VARPM, and MQBPM).

6.2.6 Percentage of 1m] Failures

Table 6-11 shows the ANOVA results for log percentage of tool failure. All
the main effects, except NSERVE, are significant. The higher order interactions of
MTOOL by NRULE and MT 00L by NLIFE are also significant. The SSTL job
priority rule performs worse than the other job rules when combined with the service
PM tool rules: VARLO, VARHI, and VARPM (Table 6-12). The job rule, DRTC,
also performs poorly when combined with MQBPM and JDDTL tool rules.

Figures 6-6 (a-d) show how the different tool control rules perform for
percentage of tool failures. The tool control rule, NOPM, causes 100 percent tool
failure, as expected. The tool control rules, VARLO, JDDTL, and FPTPM, all have

higher levels of tool failure. These tool rules do not provide for early PM as do the

167

Table 6-11 Analysis of Variance for the Log of Percentage
of Tool Failures

Source of Variation S D Ms F p
BATCH 4.42 99 .49 41.11 .000
MTOOL 238.66 6 39.78 3326.38 .000
NLIFE .89 l .89 74.83 .000
NRULE 1.58 3 .53 43.96 .000
NSERVE .00 1 .00 .22 .638
MTOOL*NLIFE 2.28 6 .38 31.74 .000
MTOOL*NRULE 4.84 18 .27 22. 50 .000
MTOOL*NSERVE .00 6 .00 .07 .999
NLIFE*NRULE .00 3 .00 .11 .955
NLIFE*NSERVE .00 l .00 .41 .524
NRULE*NSERVE .00 3 .00 .11 .957
MTOOL*NLIFE*NRULE .28 16 .02 1.48 .100
MTOOL*NLIFE*NSERVE .01 6 .OO . 12 .994
MTOOL*NRULE*NSERVE .03 18 .00 . 15 1.000
NLIFE*NRULE*NSERVE .00 3 .00 .08 .972
MTOOL*NLIFE*NRULE*NSERVE .01 15 .00 .03 1.000
(Model) 450.55 119 3.92 327.63 .000
(Total) 460.58 11199 .48

R-Squared = .978

Adjusted R-Squared = .975

rules: VARHI, VARPM, and MQBPM (Figure 6-6a). Another feature of these tool
control rules is, they all deteriorate in performance as tool life variance increases

(Figures 6-6 a—b).

6.2.7 Angysis 9f Effects Summm

The following is a summary of the significant effects for the six performance
measures.

- The main effects of tool control rules (MTOOL) and job priority rules
(NRULE) is significant for all performance measures. Table 6-13 provides a
synopsis of the ANOVA test results for significant main effects and interaction.

The higher order interaction between tool control rules (MTOOL) and job priority

168

 

Table 6-12 Treatment Means for Percentage of Tool

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Failures
TOOL RULES
FPTPM VARLO VARHI VARPM M.“
.___. i-f~ _..__* _..___ _.__ .__ .__. ._____.
' 17.2 46.3 0.2 3.0 0.4 37.4
TOOL-LOW sore 100 17.5 39.6 0.1 0.3 0.0 33.3
MAINT-LOW 959.4 I 100 16.8 44.7 0.3 2.0 0.0 29.6
ssrr. ' 100 16.9 53.7 0.3 6.0 0.2 36.5
|-—-— — —7 2 ~ --—» -————~ ‘____ —- ~- ~- ===$=E=l
DRTC 100 33.5 49.5 2.6 4.1 1.3 46.7
TOOL-HIGH sore 100 33.9 43.8 1.6 1.9 0.8 42.2
MAM-LOW PSR4 100 33.0 47.5 2.0 3.5 0.8 39.5
SSTL 100 33.0 54.3 3.1 7.3 0.8 45.9
mm==w
l DRTC 100 17.4 46.3 0.2 3.1 0.3 37.8
TOOL-Low sorc 100 17.1 39.6 0.1 0.3 0.0 33.2
MAINT-HIGH psru 100 17.3 44.4 0.3 2.0 0.0 29.7
ssn. 100 16.6 53.6 0.2 5.9 0.2 36.4
_ an: ==T
DRTC 100 32.9 49.5 4.2 2.9 1.0 46.3
TOOL-HIGH sorc 100 32.7 44.7 1.6 1.9 0.7 42.8
MAINT-HIGH 175114 100 32.9 47.8 2.0 3 4 0.6 40.6
SSTL 100 33.3 54.3 2.9 6.9 0.8 46.0
imam—=—
Legend:

Tool Control Rules: NOPM - no PM performed, allow tool to fail.
FPTPM - fixed PM point, PM occurs only after PM point.
VARLO - variable PM point, allow postponed PM up to 10% beyond.
VARHI - variable PM point, allow early PM up to 10% before.
VARPM - variable PM point, allow early or postponed PM, 10%.
MQBPM - use VARPM rule but consider maintenance queue
JDDTL - use FPTPM but considerjob due date.

Job Priority Rules: DRTC - prioritized by due date
SDTC - prioritized by sequence dependency. then due date.
PSR4 - prioritized using four priority levels
SSTL - prioritized by sequence dependency. then by tool condition.

TOOL—LOW: low tool life variance.

TOOL-HIGH: high tool life variance.

MAINT—LOW: low maintenance service time variance.

MAM-HIGH: high maintenance service time variance.

rules (NRULE) is significant for all performance measures.

169

 

 

Percentage of Tool Failure:

0
a

D
‘

 

Figure 6-6a Comparison of Control Rules
Low Tool/Low Maintenance V ariance

 

 

 

 

FPTPM VARLO

 

 

 

 

VAR” M 112011
ToolControlRule m

M
-DR'IC -snm -1>sn4 -ssn.

 

 

 

Percentage of Tool Failures

Fogure 6-6b Comparison of Control Rules

 

   

th To ol/Low Maintenance V arrance

 

 

 

 

 

 

mm VARLO

VAN-ll
Tool ControlRule

-DR'IC -smc -pss4 -ssn.

 

 

 

170

 

 

Perc entage of Tool Farlures

Figure 6-6c Comparison of Control Rules
Low Tool/th Maintenance V arrance

 

 

 

 

 

 

 

.=._r‘1

 

 

 

 

m VARLO VAN-ll mm
To 01 Control Rule

-Dmc -5mr: -PSR4 [35511.

 

 

 

Percentage of Tool Failures

Figure 6-6d Comparison of Control Rules
High Tool/th Maintenance Variance

 

 

 

 

 

 

 

 

mm VARLO

VAR}! M mt.
ToolControlRule m

-DR'DC -sorc -PSR4 D5511.

 

 

 

171
_

Table 6-13 Synopsis of ANOVA Results from Analysis of Effects

fi

 

 

 

 

 

 

SOURCE OF VARIATION TSY STS TRD STD PLT PCM
NBATCH 4 4 4 4 4 ..
MTOOL 4 4 4 4 4 4
NLIFE 4 4
NRULE 4 4 4 4 4 4
NSERVE

MTOOL '1- NLIFE =1- 4 4
MTOOL * NRULE 4 4 4 4 4 ..

 

MTOOL "‘ NSERVE

 

NLIFE * NRULE

 

NLIFE * NSERVE

 

NRULE * NSERVE

 

MTOOL * NLIFE "' NRULE

 

MTOOL * NRULE "' NSERVE

 

NLIFE * NRULE * NSERVE

 

MTOOL * NLIFE * NSERVE

 

 

 

 

 

 

 

 

 

MTOOL * NLIFE * NRULE * NSERVE
w ——-————

 

ll

Legend: "' - Significant Effect
TSY - Mean Time in System
STS - Standard Deviation Time in System
TRD - Mean Tardiness
STD - Standard Deviation Tardiness
PLT - Percentage of Jobs Late
PCM - Percentage of Tool Failures

 

- The main effects of tool life variance (NLIFE) is significant for
performance measures dealing with percentage of jobs late and percentage of tool
failures.

- The higher order interaction between tool control rules (MTOOL) and tool

172

life variance (NLIFE) is significant for mean time in system, percentage of jobs late,
and percentage of tool failures.

- All other main effects and interactions are not significant. This includes all
three and four way interactions.

An analysis of these effects for the performance measures allow a number of
conclusions and observations to be made. These conclusions include:

- Tool control rules which allow variable PM perform better than others,
particularly PM rules which promote early maintenance prior to the PM point.

- Job priority rules which are due date based (DRTC and PSR4) perform
better than sequence dependent rules (SDTC and SSTL). This is attributed to the fact
that due date rules evaluate all jobs in queue every time a job is done processing or
tool failure occurs. By prioritizing all jobs in queue, a more efficient evaluation of
jobs in queue is made.

- Maintenance service time (NSERVE) does not significantly influence shop
performance. The reason is that maintenance utilization is too low to cause delays in
tool repair. Higher maintenance utilization would cause maintenance queue to
increase resulting in longer delays and lower tool availability. Increasing mean
service time would likely change this result.

- Tool life variance (NLIFE) is only significant for two out of six
performance measures. While this result may seem surprising, the reason can be
attributed to tool availability. Availability can be viewed as the time the tool is not in

maintenance (down time). On average, a tool is in maintenance 3 hours for every

173

100 hours of processing time (3% down time) based on a preventive maintenance
(PM) heuristic. In the worst case scenario, that of corrective maintenance (CM),
down time occurs on average 6 hours for every 120 hours of processing time (5%).
Should tool availability decrease substantially, this factor would become significant.
In general, the influence of NSERVE and NLIFE on the shop performance measures
is not very strong. This indicates that the control rules (MTOOL and NRULE) are
the main decision rules that affect shop performance. However, if the means are
large, NSERVE and NLIFE can also affect shop performance.

- The best combination of rules consisted of: PSR4-MQBPM, PSR4-VARHI,
DRTC-MQBPM and DRTC-VARHI. This is attributed to the fact that PSR4 and
DRTC, and MQBPM and VARHI are consistently the better performing job priority

rules and tool control rules respectively.

6.3 A PRIOR] HYPOTHESES ANALYSIS

When higher order interactions are significant, as indicated in the ANOVA
analysis, interpretation of linear contrasts may not be valid. One Way ANOVA is
used when comparing two treatment means. Also used, is multiple comparison
(Tukey HSD) to analyze 1m hypotheses (Neter et al., 1990). This involves
comparing appropriate treatment means as discussed in Chapter 5. The use of Tukey
multiple comparison will show which treatments, if any, are significantly different.
With significant higher order interaction between job priority rules (NRULE) and

tool control rules (MTOOL), multiple comparisons of job rules for each tool rule

174

and comparison of tool rules for each job rules is necessary. It should be noted that
the significant higher order interaction between tool rules (MTOOL) and tool life

(NLIFE) did not alter the rank order significant groups for the multiple comparison.

6.3.1 Hymthesis l

Hypothesis 1 tests whether sequence dependant job priority rules used to
select the next group of jobs is affected by the decision to use job due date or tool
condition information. When replacing a tool on a machine, is it better to select the
next tool based on job due date (SDTC) the tool which can be utilized the longest on
the machine (SSTL). The objective is to fully utilize the advantages of sequence
dependency which reduces setup time.

In summary, there is no difference between sequencing dependant rules,
accept the null hypothesis. The results of Tables 6-14(a-g) and Table 6-15 show that
for any of the performance measures, there is a significant difference between SDTC
and SSTL. This holds for all six performance measures. Examining Tables 6-14(a-g)
and Figures 6-7(a-f), it can be seen that the rank order shows SDTC performs
slightly better than SSTL on every measure. The results show that there is no
significant difference between using due date or tool condition information when

selecting the next tool for a machine.

632 W812.

Hypothesis 2 continues the examination of job priority rules by examining the

175

 

 

Mean Time in System

8

Figure 6-7a Mean Values for Time in System

 

 

 

 

 

Job Pnorrty Rules

 

 

 

Std Dev. Time in System

B

8

Figure 6-7b Mean Values for Standard Deviation of Time in System

 

 

 

 

 

Job Priority Rules

 

 

 

176

 

 

Mean Tardiness

Figure 6-7c Mean Values for Tardiness

 

 

  

Job Phonty Rules

 

 

 

9d Dev Tar¢ness

Figure 6-7d Mean Values for Standard Deviation of Tardiness

 

 

 

Job Pnonty Rules

 

 

 

 

 

 

177

 

Percentage oflobr Late

Figure 6-7e Moan Vahes for Percentage of Jobs Late

 

 

 

Job Pno my Rules

 

 

 

Percentage of Tool Failures

Figure 6-7f Mean Values for Percentage of Tool Failures

 

 

    

Job Pnorrty Rules

 

 

 

 

178

 

 

V

(l 1l Illlllli l

 

 

 

0.5.0 35.. .Ebm 0.5m 0.5m _ 0.5m _
0.5m 0.500 Uh0m _ 00.0w _ 00.0w 90.0w
00.0m ._me 0.5.0 0.5.0 0.500 0.500
30mm 00.0w 354. 354— 350 $50
83:3. $55.50. 5393 E
.50. he 23 £2. 5:250 25,—. 5:250 502%
522554. m3 06 095854— 2555 $55.50. :82 .5985 E 58:. :82

 

 

 

 

 

B5 3:2 #3 _ #5 a

i—Emm DEMO DEOm DEOm DEOm DEOm _

QM mm JEmm _ DEMO DEMO DEMO _ DEMO _

DEOm DEOw vaM VMmA— .vM mm .vM mn— m

8.23"— muéEEE. Eoumam E _
.5 ._. 5 23 32. 5:250 22... 5:250 584.3 _
32:354. «3 0o 525854— 3555 $05.30. 52 955% E 22... :82 _
. 1.-.. ill ll. l1..1.l| - l1L

.23. 38 252 ._e 423. are: 8. :6 28558 6.4.24.2 43.: 4:6 2e;

 

 

 

 

 

179

 

 

 

 

05mm 1.5mm 405mm
00.0m 0.0.0m _ 00.0m
00.00 350 0.500
350 0.500 30mm
85.50 305.5 0. 522$ 50
.5 0. 0o 83 So. 5:250 55. 0. 5:250 553$

525850 «3

 

._hmm .

DEMO
30mm
UEQm

 

0o cusp—350

3.5.-- .

0.500
Ham
00.0m

 

 

4.85.50. 55.).

 

 

:0 05.0. 503.

 

53.50
.50. 0o

 

2a... 2.2.

095350 we... .5 0935850

.23. .2; o5<> .8 6.3. are... so. .6 88.358 2.53: 493 6:6 635.

 

$55.50.
5:250

 

 

505$ :—
050,—. 5:250

 

 

:0 05. 0. 52

180

 

   
       

 

 

 

 

 

 

 

     

          

  

 

_ 5.5 3...... ...-mm . 15m 1.5m 1.5m
-....mm 00.00 00.0w 00.0w _ 00.0w _ 00.0w
_ «0%. 1.00m 00.00 00.00 00.00 _ 00.00
00.0m 00.0w 040%. v0m0 e09. v0mn.
85...... 82.650. 522$ 5.
_ .5 0. 0o 23 2.6.. 5:250 05.... 5:250 522$
50258.54. 03 .5 5025254. .5555 «55:50. :32 255$ a. 55. 0. 5.).
_ . -1 1 1 l Illa“ n
.23. .80 2.502 .6. .23. .52... 8. .6 88.5.88 222:2 .33 .2-.. 22¢
lull .. 1 - 1 .. - 11.-- .- .1. .. .... 11.1. 1 .. 1--
00.0.9- «0%. _ 1....mm 1....mm .....mm 1....mm _
00.00 00.00 00.0w 00.0w 00.0w _ 00.0w
¢0m0 1.0.wm _ ¢0m0 00.00 _ 00.00 _ 00.00
00.0w 00.0w 00.00 #09. ¢0m0 0.0%. .
85.50 522.5... 522$ e. _
.5... 0o 23 4.5.. 5:250 55.... 5:250 529$
002585.. won. 5 502585.. .5555 m5:.:5... 5: E2053 5. 05.... 52).
Ill l .11.- .....-:1| 1| l 1l1 1 l -1 .111| 1-.ll1 l-l.1l. 11111-1111. ...-.1 ..l--- -

 

 

 

 

 

.23. .80 2.5; a. .23. .52... so. .6 8625.58 222:: .25 22-6 22¢

181

 

 

 

858005052 505 852. 5

 

1 u ”meow...“

 

 

 

 

 

 

 

 

fi
603 .33 83.0 8m. .0 $3.. .53
...... m3. .-.. 02d 2.4.. 3nd
85.50 885.5... 522$ e.
.50. 0o 23 8.0. 5:250 55.... 5:250 522$
5025680 06-. 06 8025684. 555$ w8=.550. =82 555$ :. 85. 0. 553.

 

......mm :5 0 0.0m 8.20 3.5.5 5.. 5252.50 852.com 50 35253 5 22:23. 2.: 250.

 

 

 

 

 

 

I
00.00 v0m4. 00.90 ......mm 1.5mm 1.00m
_ .....mm 00.00 00.0w 00.0w _ 00.0w _ 00.0w
00.0w .....mw 00.00 00.00 00.00 _ 00.00
v09. 00.0w .402. .409. v0m0 .05.
85.5... m8e..5 0. 522$ a.
.5... .5 251. 5.3 5:250 55.... 5:250 522$
002568.. 03 .5 8025.500 555$ m8e..5 0. 582 52.5% a. 85.0. :82 _

.23. .660 1:5. .6. .23. 5.66.... ..6. .6 4.6625688 22262 .62.: 22.6 2640

 

 

 

 

 

 

 

182
differences between the first three rules: DRTC, PSR4, and SDT C. The analysis of

these three rules allow a comparison of different job due date based rules. This
differs from the job priority rule SSTL which used tool load as a criteria for
selection and not due date.

For Hypothesis 2, the null is rejected because there is a significant difference
between the three job priority rules. Tables 6—14(a-g) and Figures 6-8(a-f) show that
rules PSR4 and DRTC are significantly different for most performance measure. For
mean time in system, there is no significant difference between the three rules. As
for standard deviation of time in system and standard deviation of tardiness, PSR4
and DRTC are both significantly different than SDTC, but are not significantly
different than each other. For mean tardiness and percentage of jobs late, all three
job rules are significantly different from each other (in most cases). For mean
tardiness, PSR4 is the best performing rule followed by DRTC and the SDTC
(Figure 6-8c), whereas, the opposite ranking is observed for percentage of jobs late
(Figure 6—8d). The log percentage of tool failures shows that SDTC and PSR4 are
significantly different than DRTC (except for NOPM tool rule), but are not
significantly different than each other.

Figures 6—8(aod) show that PSR4 is the best performing job priority rule,
followed by DRTC and SDTC respectively. In Figure 6-8e, the rank order is
reversed with PSR4 being the worst performer. Only in log percentage of tool
failures (Figure 6—80 does the order of the job rules change where SDTC is first,

followed by PSR4 and then DRTC in performance (except under FPTPM tool rule).

183

 

 

a

8

MemTtmeinSystem

8

B

8

8

Figure 6—8a Mean Values forTime in System

 

 

 

 

DITC PSR4

SDTC
Job Pnonty Rules

 

 

 

Sd Dev Time in System

Figure 6-8b Mean Values for Standard Deviation of Time in System

 

 

 

 

 

 

PSR4

um: SUI'C
Job Pnonty Rules

 

 

 

184

 

 

Mean Tardiness

Figure 6-8c Mean Values for Tardiness

 

 

 

 

 

 

 

m“: 5U“:
Job Priority Rules

 

 

 

Std Dev. Tardine s:

 

Figure 6-8d Mean Vabes for Standard Deviation ofTardiness

 

 

 

 

 

 

 

1 ob 9:53.33): Rules

 

 

 

185

 

 

Percentage chobr Late

Figure 6-80 Mean Values for Percentage ofJobs Late

 

 

 

 

 

 

[INC SUIC
Job Pnority Rules

 

 

 

Percentage of Tool Failures

Figure 6-8fMean Values for Percentage of Tool Failures

 

 

 

 

Job Priority Rules

 

 

 

186
6.3.3 Hmthesis 3

Hypothesis 3 tests whether there is any benefit to preventative maintenance
over corrective maintenance. Do PM tool control rules perform better the CM rule
(NOPM)? The conclusion is, PM does improve performance significantly (Tables 6-
16 a-d), thus rejecting the null hypothesis. The rule NOPM is usually the worst
performing tool rule formean time in system, percentage of jobs late, and log
percentage of tool failures. For standard deviation time in system, mean tardiness,
standard deviation of tardiness, NOPM is a significantly worse performer than tool
rules VARLO, VARHI, VARPM, and MQBPM (in most cases). The PM tool rules,
FPTPM and JDDTL, perform significantly worse than NOPM for mean and
standard deviation of tardiness. J DDTL performed significantly worse than NOPM
for standard deviation of time in system. While there is no significant difference
between NOPM and FPTPM, NOPM is only ranked lower for standard deviation of
time in system under job rule SSTL.

Overall, PM tool control rules perform better than NOPM, but there are a
number of exceptions. NOPM tool rule never performs better than the variable PM
tool control rules: VARPM, VARHI, and MQBPM (Figures 6—9 a-f). The two fixed
PM point tool rules: FPTPM and JDDTL, performed worse on half of the

performance measures (Figures 6-9 b-d).

6.3.4 11mm

Hypothesis 4 answers whether a fixed PM policy is better than a variable PM

187

 

 

MeanTimemSystem

3

Figure 6-9a Mean Values for Time in System

 

 

 

 

VAR}! YARN m man.
To ol Control Rules

 

 

 

Std Dev Tune in System

8

8

23

8

8

B

5

Figure 6-9b Mean Values for Standard Deviation ofTime in System

 

 

 

 

VARM m m

VARH
To 01 Control Rules

 

 

 

 

 

188

 

 

Mean Tardiness

Figure 6-9c Mean Values for Tard'ness

 

 

 

 

 

 

V VARPM ”PM man.
Tool Coml Rules

 

 

 

 

Std Dev Taniness

Figure 6-9d Mean Values for Standard Deviation of Tardiness

 

 

 

 

 

VARPM m JDUTL

mm VARLO VARHI
Tool Control Rules

 

 

189

 

 

Percentage ofJobs Late
D
K)

Figure 6-9e Mean Value for Percentage of Jobs Late

 

 

 

 

 

mm VARLO VARPM mam

V
Tool Cowl Rules

 

 

 

Percentage of Tool Failures

 

Figure 6-9fMean Values for Percentage of Tool Failures

 

 

 

 

 

 

 

 

 

 

 

 

FPTPM VARLO

V VARM
Tool Cofifiil Rules mm

 

 

 

 

 

. “l
2002 2002 40.000 40.00.. 40.00.. 2002
O40<> O40<> 200.00 200.00 200.00 40.00..
40.00.. 40.000 2002 2002 _ 2002 O40<>
200.00 ~ 200<> O40<> O40<> O40<> 200.00
200<> 200.00 _ 20002 200<> 20002 200<>
0204.5 0I0<> _ 200<> _ 20002 _ 200<> _ 0:0<>
20002 20002 0:0<> =00<> 0:0<> 20002
8.250 .80. 054 5055.. 0. 522$ :. 250.
0c 00358.00 03 32. 00 00253.00 505.50 555.20 805.55 0. 52.5.60 5.5.20 522$ ... 2... 0.
fl“

 

 

 

 

 

 

 

 

.23. no. 55 5. W23. .828 .82 .o 28.5.58 222...... .25 6.2. 22¢

 

 

 

 

m
E“

502 5oz 255 255 255 502

35> 35> 5...... 5.5 _ 55". 2.55

2.55 55> 502 502 502 35>

5.5... _ 2.55 53: 35> 5.62 _ 5.5”.

25> 5.5 35> 55> 35> 55>

_ 55> 5.5:. 55> 55> _ 55> _ 55>

53: 55> ...5> 5.62 55> 5.52

85.5 .82 23 8.525 52:... ... as:
.o 5:895 «3 2.2 .o 25522. 8.5.3 235m 362...: 852.8 26:5 522m .2 as;

 

 

 

 

 

.23. no. 85. a. .23. .238 .9... .0 2.8.5.58 222...). .25 8.2. 2...:

 

 

191

 

 

 

 

 

 

 

 

 

.lnlllllflu F fl
2002 2002 40.00.. 40.00.. 40.00.. 2002
O40<> O40<> _ 2002 200.00 _ 2002 40.00..
40.00.. 40.00.. 200.00 2002 200.00 O40<>
200.00 _ 200<> O40<> O40<> _ O40<> 200.00
200<> 200.00 _ 02> .00<> 0=0<> 200<>
_ 55> 5.5.2 _ 53: 55> _ 520.2 _ 55>
20002 =00<> 200<> 20002 200<> 20002
85.50 .8 0. 254 825550. 522$ :0 5. 0.
.5 90259500 004 2.2. 00 ems—.0500 5.5.50 5.55.0 5055.. 0. 55.50 555.20 522$ ... 05. 0.

 

.23. no. .55 s. 23. .828 .2... .o 88.558 22.3: .83 8.... 2......

 

 

 

 

 

 

 

 

2002 2007. 40.000 40.000 40.00..l 20oz

O40<> O40<> O40<> 2002 _ 200.00 40000

40.00.. 40.00.. 200.00 O40<> 20oz O40<>

200.00 200.00 2002 200.00 O40<> 200.00

200<> 200<> 200<> 200<> _ 200<> 200<>

_ =00<> 000.2; 20002 00.004; _ 20002 _ $525

20002 20002 0.525 20002 00.025 20002

_ 85.50 .80. 254 8650. 522$ ... 05.0.
0.. 9025800 004 2.2. 5 0023200 5.5.50 5.5.20 .585550. 8.5.50 5.5.20 522$ ... .2... 0.

 

I

.23. ..o. 3.2. .8 823. .828 .8... .o 28.5.58 222...). .28 8:6 2...:

 

 

192

policy. This test requires the comparison of three variable PM tool control rules:
VARLO, VARHI, and VARPM, to the fixed PM rule FPT PM.

FPTPM rules were significantly different than the variable PM rules (Tables
6-16 a-d) except for percentage of jobs late. Tool control rule VARHI performed
significantly better than FPTPM for all measures except percentage of jobs late and
tardiness (when using PSR4 job rule). For this performance measure (Figures 6-10
a—f), VARHI is not significantly different but does perform better (Figure 6-10c).
FPTPM performs significantly worse than any variable PM tool rule for standard
deviation of time in system, and mean and standard deviation of tardiness. The
exception to this is when job priority rule PSR4 is used. FPTPM tool rule performs
significantly better than variable PM rule VARLO for mean time in system, log
percentage of tool failures, and percentage of jobs late. FPTPM tool rule performed
significantly worse than VARPM for all performance measures except percentage of
jobs late. For this measure, FPTPM outperforms VARPM (Figure 6-10c).

The results show that variable PM rules outperform fixed point PM rules.
The PM tool rules, VARHI and VARPM consistently outperformed FPTPM. These
two PM tool rules allow for early PM. The third PM tool rule, VARLO, allows PM
to be postponed past the PM point causing more tool failures (Figure 6-lOf). This
explains why VARLO performs worse than FPTPM three of six performance

measures (Figures 6-10 b-d).

193

 

 

a

8

Mean Time in System

is

e

8

8

Figure 6-10a Mean Values for Time in System

 

 

FPTM V

 

 

MUD VARH
Tool Control Rules

 

 

 

Std Dev. Time in System

Figure 6-10b Mean Values for Standard Deviation Time in System

 

 

 

 

 

 

VARHA

m VARLO VARHI
Tool ControlRules

 

 

 

194

 

 

Mean Tardiness

Figure 6-10c Mean Vahes for Tard'ness

 

 

 

 

 

ARLO VARHI
Tool Control Rules

 

 

 

ad Dev. Tardiness

Figure 6-10d Mean Values for Standard Deviation of Tardiness

 

 

 

 

 

 

FPTPM VAR].

O VARHI
Tool Control Rules

 

 

 

195

 

 

Percent ofJobs Late

Figure 6—10e Mean Vahes for Percentage ofJobs Late

 

 

 

 

 

 

VARLO VARHI VARM
Tool Control Rules

 

 

 

Percent ofTaol Failures

Figure 6-10f Mean Values for Percentage ofTool Failures

 

 

 

 

 

 

 

 

V 0 VARHI
ARL Tool Control Rules

 

 

 

196
6.3.5 Hymthesis 5

Hypothesis 5 tests whether there is a significant difference between early
variable PM tool rule (VARHI) and postponed variable PM tool rule (VARLO). The
question attempts to answer whether it is advantageous to allow early or postponed
tool PM. The results show that the null hypothesis is rejected, there is a significant
difference between early versus postponed variable PM tool control rules (Table 6-
17). For all performance measures, VARHI performs better (Figures 6-11 a-f and
Tables 6-16 a-d). Only performance measures, mean and standard deviation is there
no significant differences between VARHI and VARLO tool rules (Table 6-17).
While these two measures are not significantly different, VARHI still outperforms
VARLO (Figure 6-10 c-d).

VARHI performs better than VARLO because of the lower percentage of tool
failures. This shows that early PM is preferred to that of postponed PM. The only
time that postponed PM does not worsen performance is with tardiness (mean and

standard deviation).

6.3.6 113mm

Hypothesis 6 compares a variable PM tool control rule which allows both
early or postponed PM when maintenance queue is empty (MQBPM) with that of
variable PM (VARPM). The null hypothesis tests whether there is a significant
difference between tool rules which considers maintenance queues versus a tool rule

that does not.

197

 

895
an

 

L959
Cod

    

 

 

 

mezzan— 88... .8
ems—589m we;
F t. l .i i

  

  

 

 

 

3a.. 33.
8 038083

30528.
8:350
Emu—5m

 

mmoEEfl.

 

 

806$ :—
oEE. 5:525
8355

 

     
  

823m 5 «EC.

 

.9588 3223 as Eamozv 88.585 326 85:25“:
55» 83. .838 38. 2.. 22E; he 8§a> ._o “3:92 ”I. 0:3

 

 

 

 

   

 

   

 

 

 

 

,- .888 233 A83
mmmfimv $0.2 Sun.
32:33. 802% E
83.3... 88,—. .8 23 28.. 8:330 25,—. c2325
nuance?— wo‘. 8 nag—3.5m 23ng 86535. Hagan 53mg 5 25,—.

 

  

 

  

.83. .828 .2; 2“. 032:; 8898; Ea ram he 8§E> co 2322 2-0 22:

198

 

 

a

Figure 6-11a Mean Values for Time in System

 

8

Mean Time in System

8

B

B

8

 

 

 

VARHI

Tool Control Rules

 

 

 

Std Dev Time in System

Figure 6-11b Mean Values for Standard Deviation of Time h System

 

 

  

Tool Control Rules

 

 

 

 

 

 

199

 

 

Mean Tardiness

Figure 6-11c Mean Values for Tardiness

 

8

B

8

B

 

 

 

Tool Control Rules

 

 

 

Std Dev Tartfiness

Figure 6-11d Mean Values for Standard Deviation of Tardiness

 

 

 

To cl Co ntrol Rules

 

 

 

 

 

22
ooaowmm
O

33 £2 .8 .5 tom

     
    

O

     
   

    

‘ 3 J. I
0 o O o
hung—ck 86L. .8 ~53?»

200

 

 

Percent of Jobs Late

Figure 6-11e Mean Values for Percentageof Jobs Late

 

 

 

 

 

VARLO V
ToolConzrolRuler M

 

 

 

Percent of Tool Failure:

Figure 6-1 1f Mean Values for PercentageofTool Failures

 

 

 

 

 

Tool Control Rules

 

 

 

 

201

 

 

8

El

Mean Time in System
8

Figure 6—12a Mean Values for Time in System

 

 

 

Tool Control Rules

 

 

 

Std Dev Timem System

Figure 6-12b Mean Vahes for Standard Deviation ofTime in System

 

 

 

Tool Control Rules

 

 

 

 

 

202

 

 

Mean Tardiness

Figure 6-12c Mean Values for Tardiness

 

 

 

 

m
Tool Control Rules M3

 

 

 

3d Dev Tardiness

S

B

8

3

Figure 6-12d Mean Vahes for Standard Deviation ofTardiness

 

 

 

 

Tool Control Rules

 

 

 

 

 

203

 

 

Percent of Jobs Late

0
u

a
a

O

005

Figure 6-12e Mean Vahes for Percentage of Jobs Late

 

 

 

Tool Control Rules

 

 

 

Percent ofTool Failures

Figure 6—1 2f Mean Values for Percentage ofTooI Failures

 

 

    

 

 

 

 

Tool Control Rules

 

 

 

 

204

The null hypothesis is rejected, there is a significant difference between
VARPM and MQBPM. MQBPM performs significantly better than VARPM for
percentage of jobs late and log percentage of tool failures (Table 6-18). VARPM
performs significantly better than MQBPM for standard deviation of time in system
(1‘ able 6—18 and Figure 6—12b).

For the next three performance measures, mean time in system, mean
tardiness, and standard deviation of tardiness, there was no significant differences
(Table 6-18). Of these three non-significant performance measures, the MQBPM
rule performed best for mean time in system and mean tardiness (Figure 6-12 b-c).
For standard deviation of tardiness, VARPM performs better than MQBPM for the
two measures of variance, standard deviation of time in system and tardiness. While

VARPM causes less variation, it performs worse on all other measures.

6.3.7 Hymthesis 7

Hypothesis 7 looks at whether a tool rule which considers past due jobs
performs significantly different than PM rules which do not. The hypothesis tests
whether tool control rule JDDTL is significantly different than FPTPM. JDDTL is a
modified version of the tool rule FPTPM. Only JDDTL considers job due date status
and will keep a tool in production until all past due jobs are processed or until the
tool fails.

The null hypothesis is rejected, there is a significant difference between the

two tool control rules (FPTPM and JDDTL). FPTPM performs significantly better

205

than JDDTL for all performance measures except standard deviation of tardiness and
percentage of jobs late (I‘ able 6-19). While FPT PM did not perform significantly
different than JDDTL, FPTPM still outperformed JDDTL for both standard
deviation of tardiness and percentage of jobs late (Figure 6-l3e).

The poor performance of JDDTL can be contributed to the greater percentage
of tool failures. The fact that JDDTL continues processing past due jobs causes the
risk (and number) of tool failures to rise. The added risk is only incurred for past
due jobs.

The conclusion is, using job due date as a means of tool control is not
beneficial. Allowing past due jobs the ability to keep a tool in production in excess
of its PM point does not improve shop performance. The additional risk of tool

failure outweighs the benefit of job completion.

6.3.8 W

Hypothesis 8 examines whether tool life variance effects the performance of
the shop. The null hypothesis tests whether there is a significant difference between
LOW and HIGH tool life variance. Only performance measures, percentage of jobs
late and log percentage of tool failures performance, are significantly influenced by
tool life variance (I‘ able 620). The logic that tool life variance has a significant
effect on percentage of tool failures is apparent (Figures 6-14 a-f). What is
surprising is the limited effect tool life had on mean and standard deviation of

tardiness and time in system. The explanation relates to the amount of lost tool

206

 

 

 

 

 

 

 

 

85.5% 88min? 328:5. - m 65in
289 Some :89 .53 223 ...Gmoe
Cod—N mum. 09;. End $06 Sod
30528. 823m :—
8S=£ 88,—. .8 2a..— 28. 8:230 25,—. 8:230
33:86; we: .8 03:83“. 23.8% 38.—Gab .5985 823$ 5 25...

 

 

623 8.3. 35:8 88 as; 2.. Be”. .8 8.3.; .6 warez 2e 22¢

ASA—EB 82 moon— Ea Shh—n5 3am 25 8.. 302280

 

207

 

 

Mean Time in System

3

S

8

B

B

8

Figure 6-13a Mean Vahes for Time in System

 

 

 

 

 

Tool Control Rules

 

 

 

Std Dev. Time in 31mm

is

8

Figure 6-13b Mean Values for Standard Deviation of Time in System

 

 

 

 

 

Tool Control Rules

 

 

 

208

 

 

Mean Tardne s:

Figure 6-13c Mean Values for Tardiness

 

 

 

 

 

Tool Control Rules

 

 

 

Std Dev Tardiness

Figure 6-13d Mean Vabes for Standard Deviation of Tardiness

 

 

 

Tool Co ntrol Rules

 

 

 

 

209

 

 

Percent ofJobr Late

01)

Figure 6-13e Mean Values for Percentage of Jobs Late

 

 

 

Tool Control Rules

 

 

 

Percent of Tool Failures

Figure 6-13fMean Values for Percentage of Tool Faihres

 

 

 

 

 

 

 

ToalControlRules

 

 

 

210

 

 

Mean Time in 51mm

3

8

as

e

B

8

8

Figure 6-14a Mean Values forTime in System

 

 

 

 

Tool Life V ariance

 

 

 

3d. Dev Time in Sy stem

Figure 6-14b Mean Values for Standard Deviation ofTime in System

 

 

 

 

To ol hfe V ariance

 

 

 

211

 

 

Mean Tardnesr

Figure 6-14c Mean Values forTardiness

 

 

 

 

HIGH
Tool hfe V ariance

 

 

 

Std Dev Tardness

Figure 6-14d Mean Values for Standard Deviation of Tardiness

 

 

 

 

RICH

Tool life Variance

 

 

 

 

 

212

 

 

Percent of Jobs Late

Figure 6-14e Mean Values for Percentage of Jobs Late

 

 

 

 

Tool bfe Variance

 

 

 

Perc ent of Tool Failures

Figure 6-14f Mean Values for Percentage ofTool Failures

 

 

 

 

 

 

Tool Life Variance

 

 

 

213

85.8%:82 .83 328. ... - 52.8on
I 1

 

 

 

 

 

 

 

 

 

 

$3.. 25.. 83.. $3.. 32... 33..
.8. 08. .2. 8m. 2w. :0.
32.8.8... 523% E
3.2:... .8... 8 2.... 38.. 8:580 2.5... 8:580 E233
82.2.8.8“. we: 8 89.2.3.8; .85—2.2m 385:3... 82 888% 5 25... 822
H [H

 

 

 

.25 ... 2.38m 88:28.2 .8 82.553 8 53322. 3-: 888.0.

 

 
 
  
     

 

82.28%:52 E»... 8.8. E . one
I
.88.. cm... 23.. $9... 5.3 83
Bed 8.... no. ._ Re. So. ”3..
3258... 8293 ..—
8.=_:E 88... 8 23 3.8.. 8:580 2.5... 8:580 E233
82:88; was. 8 8988.8; c.5235 385:3... =82 ...acsfim 5 25... :82

 

 

 

 

 

 

35...; 0.... .ooe 5. 85:9 .o 3:22 8;. 2.5

 

214

availability which is not impacted significantly by tool life variance.

6.3.9 W2

Hypothesis 9 is similar to Hypothesis 8 but looks at how maintenance service
time variance effects shop performance. The hypothesis tests whether there is a
significant difference between LOW and HIGH maintenance time variance. The null
hypothesis is accepted, there is no significant difference for all six performance
measures (Fable 6-21). Figures 6—15 (a-t) clearly show that each performance
measure has nearly identical results under both LOW and HIGH maintenance
variance. The unexpected result can be contributed to the low utilization of the
maintenance process (ranging from 11% to 22%). With a low utilization rate, the
maintenance queue waiting time is minimal, resulting in less loss tool availability is
higher. Should maintenance variance increase or frequency of tool failure increase,
maintenance utilization would increase causing increased queue delay. Only when
utilization exceeds 60% would maintenance variance become a significant factor.
The problem with this utilization rate is that it would not emulate the visited shop

floors.

6.3.10 Wm

A summary of the hypothesis results are provided in Table 6-22. The results
show that of the tool control rules, MQBPM and VARPM, consistently provide the

best performance. The most consistent performer for the job priority rules is PSR4,

215

 

 

Mean Time in System

Figure 6-15a Mean Vahes for Time in System

 

8

S

8

a

8

8

 

 

Mantenancc Time V trance

 

 

 

Std. Dev Time In System

Figure 6-15b Mean Values for Standard Deviation of Time in System

 

 

 

 

 

M amtenance Time Variance

 

 

 

216

 

 

Mean Tudmen

Figure 6-15c Mean Vahes for Tardiness

 

 

 

  

men

 

M aintenance Time Variance

 

 

 

Std Dev Tardiness

Figure 6-15d Mean Values for Standard Deviation of Tardiness

 

 

 

 

M aimenance 'Ilme Variance

 

 

 

 

 

217

 

 

Percent ofJobs Late

Figure 6-15e Mean Values for Percentage of Jobs Late

 

 

 

 

M eimenance Time V ariance

 

 

 

Percent of Tool Failure:

Figure 6~15f Mean Values for Percentage of Tool Failures

 

 

 

    

M untenance Time V anmce

 

 

 

218

Table 6-22 Summary of Hypothesis Results

 

Hypotheses

Conclusion

 

1. Sequence dependent job rules based on due
date or tool load are equal.

Accept Ho, sequence dependent job rule based
on due date perform only slightly better.

 

2. Job rules which prioritize by due date are
equal. regardless of additional information.

Reject H0, only certain forms of information
enhances performance not.

 

3. CM tool rule is equal to PM tool rules.

Reject H0, most PM tool rules perform better,
but not all.

 

4. Fixed PM point tool rules equals variable
PM tool rules.

Reject Ho, variable PM rules with early PM
perform better than fixed PM point rules.

 

5. Early variable PM tool rule equals
postponed variable PM tool rules.

Reject Ho, early PM performs better than
postponed PM tool rule.

 

6. Variable PM tool rule which does and does
not consider maintenance queue information
are equal.

Reject H0, using maintenance queue in
variable PM rule improves performance.

 

7. Fixed PM point tool rule equals fixed PM
point tool rules which considers past due jobs.

Reject H0, considering past due jobs decreases
performance for fixed PM point rule.

 

8. Tool life variance does not impact shop
performance.

Accept Ho.

 

 

9. Maintenance service variance does not
impact performance.

 

Accept Ho.

 

 

 

which confirms finding by Melnyk et a1. (1989). Specific results found in the
analysis which bare mentioning include:

- The sequence dependent rules SDTC and SSTL, while showing promise in
past research (Kannan and Lyman, 1992; Lyman, 1993), fail to benefit an
environment with stochastic life tooling. This may be attributed to the fact that the
ratio of setup time to processing time is 20%. Higher levels of this ratio are likely to
change these findings.

- The due date rule PSR4 which utilizes sequence dependency while

219
considering job slack, performs best. The job rule DRTC, which looks at job due

date when prioritizing, performs second best. The results show that while sequence
dependency can reduce the number of tool changes, job due date is more important
to shop performance. This outcome is most apparent in the tardiness measures.

- PM tool rules perform better than the non-PM rule (NOPM) in most cases.
PM rules reduce the amount of tool failures and thus maintenance delays. It should
be noted that the NOPM tool rule performs better than tool rules VARLO and
JDDTL on several measures.

Vanderhenst et a1. (1981), and Banerjee and Burton (1990) point out that CM
allow for a more efficient utilization of the tooling resource. The maintenance delay
incurred by NOPM is less than the tool rules VARLO and JDDTL, which allow for
PM. Their total time in maintenance (CM + PM + maintenance queue time) is
greater than that incurred by NOPM. VARLO and JDDTL thus suffers from the
worse conditions associated with both CM and PM. For example, frequently there is
not enough PM to reduce tool failures, yet too much causes a higher level of total
maintenance time.

- Fixed PM point (FPTPM) does not perform as well as variable PM rules.
The exception is, when the variable PM tool rule allows for postponed PM. Once
again, the superior performance of VARHI and VARPM over FPTPM can be
attributed to total maintenance time (tool failure). The FPTPM tool rule forces a tool
to remain available for production until it passes the PM point, after which, it can

go in for maintenance. This inflexibility causes the risk and number of tool failures

220
to increase. Both VARHI and VARPM allow for early removal of tools for PM,

thus reducing tool failures.

- The form of variable PM, whether early or late, is a major factor in shop
performance. The early PM rule VARHI consistently gives better performance than
the postponed PM tool rule VARLO. While both rules consider shop demand, the
difference in performance can be contributed to tool failure risk. The higher risk
results in greater frequency of tool failures and higher total maintenance time.

- The use of maintenance queue information improves the performance of a
variable PM tool rule. The MQBPM rule is a modified VARPM tool rule with the
addition of maintenance queue information. Examining maintenance queue and
allowing removal of tools for PM causes reduced tool failure risk. Total maintenance
time is also lower for MQBPM than for VARPM. While the use of maintenance
queue information improves performance on a number of measures, it causes a
greater variation in performance. Thus, MQBPM performs worse than VARPM for
the standard deviation performance measures.

- Allowing tools to remain in production until all past due jobs are done does
not improve delivery performance (tardiness and percentage of jobs late). The tool
rule JDDTL performed significantly worse than FPTPM because it forces a tool to
remain in production. The high risks of tool failures negates any benefit that delayed
PM could provide. Whenever presented with the choice to perform PM or process a
job when past the PM point, the results show that the tool should be removed for

PM.

221

- The last issue, the effects of variance, is not as clear-cut as this research
might indicate. While tool life and maintenance variance is inconsequential, this only

reflects the parameters used in the model.

6.4 Post Hoc Analysis

The following analysis consists of two parts. The first part will examine the
relative performance of tool control and job priority rules since these rules dominate
the influence on shop performance. The control rules will be examined under low
and high tool life and maintenance service variance. The objective is to determine
the robustness of the control rules and further investigate their affect.

The second part will examine the performance of combined tool control rules
and job priority rules. The objective here is to determine what combination of rules
provide the best overall performance.

While 2522.119; analysis does not provide the same level of construct validity
as am analysis, it does allow further investigation and insights. The purpose of
the M analysis is to provide insight into how the tool and job heuristics
perform. Tukey HSD multiple comparisons will be used to conduct the analysis. The
significant differences found in the Tukey test which are not found in the ANOVA
tests can be attributed to the techniques used in comparing treatments. ANOVA
compares the mean of the means to each treatment mean. Whereas, Tukey compares

each treatment mean to each other.

222

6.4.1 Relative Pgrfgrmance of Heuristics

The introduction and HIGH and LOW variance consist of two parts: 1) tool
life, and 2) maintenance. Two specific items will be examined when analyzing the
multiple comparison tables. The first concern looks at how the relative performance
of each heuristic is affected by variance. Does the rule perform well under LOW or
HIGH variance? The second issue, is there any cross over effect to different levels
of variance? This looks at whether a heuristic performs better for HIGH variance

than LOW variance.

6.4.1.1 2135:] Life Variflce

Each performance measure, except mean time in system, show significantly
different groups of job priority rules (Table 6-23a). The rank order of the job rules
show that those job rules which perform well under LOW variance do equally well
under HIGH variance. The exception is, mean time in system and log percentage of
tool failures (Table 6—23a). What is seen for these two performance measures is
grouping of job rules under LOW and HIGH variance. Also, in no case does a job
rule perform better under HIGH variance than under LOW variance (Figures 6—16 a-
0.

When examining tool control rules by tool life variance, there are significant
differences between certain rules (Table 6-23b). The tool rules which perform well
under LOW variance also do so under HIGH tool life variance. Tool rules perform

better under LOW variance than HIGH variance.

223

 

_ 10.1-2002
304-2002
121-04042
304-040<>
191-40.00..
304-4000..
10.1-2000".
304-2000....

_ 10.1-200<>
304-200<>
_ 10.1-.10<>

_ 10.1-20002

_ 304-.10<>
304-20002

33...... 2.80.
.0 00859.00 03

           

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

fl fi - ll.-
10.1-2002 10.1-4000. 10.1-4000. 10.140.00.-
304-2002 304-4000.. 304-4000.. 304-4000 .-

10.1-040<> 10.1-2000". 10.1-2000”. 10.1-2000...
304-040<> 304-20000 304-2000". 304-20000
10.1-4000. 10.1-20002 10.1-2002 10.1-2002
10.1-2000... 10.1-200<> 304-2002 304-2002
304-4000.. 304-20002 10.1-040<> 10.1-040<>
10.1-200<> 10.1-2002 10.1-2004; 304-040<>
304-200<> 304-2002 304-040<> 10.1-200<>
10.1-.1~.<> 10.1-04-0<> 10.1-.10<> 10.1-20002
10.1-20002 304-04~.<> 304-200<> 304-20002
304-20000 10.1-.10<> 10.1-20002 10_1-.1~.<>
304-20002 304-200<> 304-20002 304-200<>
304-.10<> 304-.10<> 304-.10<> 304-.10<>
2.»... 22.8.8... 529.0 ... 2......
use. 8 00359.00 5:...80 ......ESm 20:830. 5:180 23:80

 

 

 

  

 

 

 

3.3.25 o... .8... ...: 3.2.25: .228 .2; .o meofiasoo 2......2 £3 2......

 

 

 

 

 

 

 

    

 

10.1-2002
304-2002
10.140.00.-
30440.00.
10.1-040<>
304-040<>
10.1-2000".
304-2000".
10.1-200<>
304-20u.<>
10.1-.10<>
10.1-20002
304-20002
304-.10<>

 

 

 

529.0

... 2...... :32

 

 

 

 

 

     

 

 

 

 

 

 

8:2; o.... .8... B 3.3. 5.8.... ..o. 3 ”garage 0......e... .33 5e 2......

 

 

:e.:-.:2 :e5-3.2 52:02.2 5.5-fina- :o5-.....8 555:2
505-955 33-3.2 .53-.5... 30..-...me Ben-.92 Ben-48m
55-3.2 55-0.35 5052.9. 55-95.. 55-0.5... 55-8.8
55-2.9. Ben-open Ben-ope... 33-8-em Ben-open 55-0.55
Ben-.92 553.8... 5.5-eta 55-2.35 55-03:. :e.:-3.2
.33-ob... 33-.....2 33.3%. 33-83:. 33-0.3... Ben-ope...
333.2 55-89. 553.2 _ 55.3.2 55.3.2 33¢ka
Ben-open Ben-o8... 333.2 .53-3.2 33.3.2 33.3.2
«2...... .2: a... .852; 52:... ... 2...... sue...»-
.o 2.3822 «3 so. .a 25822 8.3.69 3.2.3... 32.2... =8: 8.3.5: 285.... =. 25 =8:

 

 

224

 

 

a!

8

MunT-ne 057mm

8

8

8

8

Figure 6-163 Con'parison of Job Priority mles by Tool Life Variance

for Men Tune m System

 

 

 

 

 

SDTC
Job Pnonly Rule;
-IDWTOOLVARIANCE -HIGH'IOOLVARIANCE

 

 

 

Std Dev Time 0 System

Figure 6-16b Carparism of Job Riority Rules by Tool Life Variance

for Standard Devmxon Time In System

 

 

 

 

    

SUTC
Iob Pnonty Rule:
-DOW'IOOLVARIANCE -HIGH'IOOLVAR.IANCE

 

 

 

225

 

 

Mean Tardmes

Figure 6-16c Comparison of Job Riority Rules by Tool Life Variance
f

or M can Tudmeu

 

 

 

 

M'C SDTC PSM S11
Job Pnorny Rules

-LOWTOOLVARIANCE -HIGHTDOLVARIANCE

 

 

 

Std Dev Tardinen

Fugue 6-16d Convarison of Job Rion‘ty RJles by Tool Life Variance

for Snndard Devnnon Tudmess

 

 

 

 

SDTC
Job Pnonty Rules
-IDW’IOOLVARIANCE -HIGH'IOOLVARIANCE

 

 

 

226

 

 

c o
N u

Percentage of Jobs Late

0

Fugue 6-16e Oorrparison of Job Priorly Rules by Tool Life Variance

for Percentage orobs late

 

 

 

 

 

 

mTC SDTC PSRA SSTL
Job Phonty Ruler

-LOW'IOOL VARIANCE -HIGH'DOOLVARIANCE

 

 

 

Percentage of Tool Failure:
0

Fngre 6-16f Corrparism of Job Rioriy Rules by Tool Life Variance

for Percentage ofTo ol Fallurex

 

 

 

 

Job Pnonty Rules
-wW'IDOLVARIANCE -HIGHIOOLVAMANCE

 

 

 

227

 

 

MemTrmemSystzn-i
8

Figure 6-17a Cormarison of Tool Control Rules by Tool L'rfe Variance

for Mun Time 0 System

 

 

 

 

m m VARLO VAR}!
Tool Control Rules

-LOW mow-um- NCE -HIGH‘IOOLVARIANCE

 

 

 

Std Dev Trmexn System

Figure 6-17b Comparison of Tool Control Rules by Tool Life Variance

for Standard DevxatlonTrme in System

 

 

 

i
l
i
i
4
i
i
i

 

m m VARLO VARH
Tool ControlRules

-LOW'IOOLVARIANCE -HIGH'IOOLVAR1ANCE

 

 

 

 

 

228

 

 

 

Figure 6-17c Corrparison of Tool Control Riles by Tool Life Variance

for Mun Tardiness

 

 

 

 

 

 

 

 

 

70
60
v. 50
2
g 41
l
H
s an
fi
20
IO
0
RIM FPTPM VAR-LO VAR-HI VAR-PM ”PM rum.
Tool Control Rules
-LOW'I‘OOLVARIANCE -HIGH‘IOOLVARIANCE
Figure 6-17d Comparison of Tool Oontrol Rules by Tool Life Variance
for Standard Devution Tardiness
8O
70
u w ‘
5 so
a
5" 41
2
Q 30
‘0
cs
20

 

 

MIPM rum.

m m VARLO VARHI VAR-PM
To 01 Control Rules

-LOW'IOOLVARIANCE -HIGH'DOOLVARIA NCE

 

 

 

 

 

229

 

Fig-ire 6-17e Corrparison d Tod Control Rules by Tool Life Variance

for PQrcentlge orobs Late

 

 

 

 

 

 

 

 

0‘

035

i 03

-§ 025

“5 02
w
s

g or:
E

0 01

005

0

m FPTPM VAR-LO VAN! mm mm.
Tool Control Rules
-IDW'IOOLVARIANCE -HIGH'ICOLVAR.IANCE
Figire 6-17f Cormarison of Tool Control Rules by Tool Life Variance
for Percentage ofTool Failure:

12

3 l
E

I: on
'5
O
F

«5 as
2
5

5 04
a.”

 

 

 

 

 

 

 

 

 

 

VAR-H m rwn.
Tool Control Rules

-IDWTCOLVARIANCE -HIGH'IOOLVARIANCE

 

 

 

230
6.4.1.2 Maintenggee Serviee Time Variance

All performance measures except, mean time in system and log percentage of
tool failures show significant differences between treatments for job rule by
maintenance variance (T able 6-24a). lob priority rules do not differ significantly in
performance between LOW and HIGH variance (Figures 6-18 a-f). Job rules which
perform well under LOW variance do so under HIGH maintenance variance. There
is no cross over effect where a rule under HIGH variance performs better than LOW
variance (Figure 6-18 a-t).

All performance measures show groupings of significant differences between
tool rule by maintenance variance (Table 6—24b). Tool control rules do not
significantly differ between HIGH and LOW maintenance variance (Table 6-24b).
Tool rules which perform well under LOW variance, also perform almost equally as

well under HIGH variance. No tool rule exhibited cross over effect.

6.4.2 1912 Priemy Rule by [gel Qentml Rele

All performance measures show significant groupings of various tool control
by job priority rules (Table 6-25). The higher order interaction of MTOOL by
NRULE is also significant for ANOVA on all performance measures in the analysis
of effects. The groupings of significantly different job-tool rules make comparisons
difficult. For this reason, the rank order of combined rules will be used to draw
conclusions.

For mean time in system measure, the tool control rule has more influence on

231

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

0000...; 00.300 000000.502 00 3.00.501 .3500 .00 0. 00 0000200800 0.05:2 .03-.0 0.0.. 0.

_ 10.1-2002 10.1-2002 191-4000.. 10.1-4000.- 10.140.00.. 10.1-2002
304-2002 304-2002 3044000.. 304-4000 0 3044000.. 304-2002
_ 10.1-040<> 10.1-040<> 10.1-20000 10.1-20000 10.1-20000 121-4000..
304040<> 304-04~.<> 304.2000 304.2000 304.2000 3044000..
_ 10.14009. 10.14009. 10.1-20002 10.1-2002 304-2002 1,0.1-040<>
3044000.. 304-4000.. 304-20002 304-2002 10.1-2002 304-040<>
_ 10.1-2000... 10.1-20~.<> 10.1-20~.<> 10.1-040<> 10.1-040<> 10.1-20000
304-20000 10.1-20000 304-2002 304-040<> 304-040<> 304-20000
10.1-200<> 304-200<> 10.1-2002 10.1-200<> 10.1-200<> 10.1-200<>
304-20~.<> 304-20000 10.1-040<> 304-200<> 10.1-20002 . 304-200<>
10.1-.10<> 10.1-.10<> 304-040<> 10.1-_1~.<> 304-20002 10.1-.10<>
304-.10<> 304-.1~.<> 10.1-.10<> 304-.10<> 304-20~.<> 10.1-20002
10.1-20002 10.1-20002 304-20~.<> 10.1-20002 10.1-.10<> 304-20002
304-20002 304-20002 304-.1~.<> 304-20002 304-.1~.<> 304-.10<> H
.05—.00 0.00... 0.04 200.0... 0. 80.2.0 ... 08.0. 50.9.0
00 003009.00 004 2.0.. ..0 005000.00 00.5300 0.00030 .3505... 00.5.00 0.00.5.0 ... 0.... 0. :82
Hi flan

 

 

 

 

 

 

 

 

 

 

 

 

 

10.1-4000]! 10.1.3000 10.14000 10.14000 10.14000 10.14000
10.1-0000 304-3000 30440.00 30440.00 30440.00 304-40.00
30440.00 10.1-0000 10.1-0000 10.1-0000 10.1-0000 10.1-0000
10.1-0000 304-0000 304-0000 304-0000 _ 304-0000 304-0000
304-0000 10.1-4000 10.1-0000 10.1-0000 10.1-0000 10.1-0000
10.1-0000 30440.00 304-0000 304-0000 304-0000 304-0000
3040.000 10.1-0000 10.1.3000 _ 10.1.3000 10.1.3000 10.14000
304-0000 304-0000 304.3000 304.3000 304.3000 30413.00
00.3.00 .00.... 0.04 2.00.0.0. c.0300 ... 0.0.0. 50.9.0
00 003000.00 004 2.0.. 00 008000.00 00.5.00 0.50030 2050...... 0002 00.5.50 0.0.530 ... 0.... 0. :32
n “H

005......> 003.00 000000.502 an. 00.00 0.0010 000 00 0000100500 0.0.0.02 00.... 0. awn-o 0300.

 

 

232

 

MeanTnme mSynem

Figure 6-18a Corrparison of Job Priority Rules by Nhintenance Service Variance

for Tune m Synem

 

 

 

 

 

DITC SUTC P514 SSTL
Job Pnonly Rules

-LOWMA IN'I‘ENANCE VA RIA NCE - HIGH MA IN'I‘ENANCE VA RIA NCE

 

 

 

9d Dev Tune 0 System

Figure 6-18b Oonparison of Job Riority Rules by Maintenance Service Variance

for Standard DEVIIKIOD Txme 111 System

 

 

 

 

 

mTC SDTC PSR4 53'“.
Job Pnonly Rules

-IDWMAIN'IENANCE VARIANCE -HIGHMA IN’ENANCE VARIANCE

 

 

 

233

 

Me an Tudne s:

Figure 6—18c Comparison of Job Riority Rules by Neintenance Service Variance

for Mann Tudmen

 

 

 

 

ulTC SDTC PSRA SSTL
Job Pnemy Rules

-LOWMAIN'IENA NCE VARIANCE -HIGHMAINTENANCE VARIANCE

 

 

 

Std Dev Tardiness

Figure 6-18d Corrparison of Job Riority miles by Maintenance Service Variance

for Standard Deviation Tardiness

 

 

 

um: SDTC FSFJ SSTL
Job Pnoniy Rules

- IDWMAIN'I‘ENA NCE VARIANCE -HIGHMAIN'IENANCE VARIANCE

 

 

 

234

 

Figure 6-18e Oorrpariscn of Job Riority Rules by Maintenance Service Variance
for Percentage ofJobs Laxe

 

 

 

 

 

 

 

 

 

 

 

§ 03
03

43 025
H
‘0-

3 oz
9

E 015

g 01

005

0

mm SDTC PSR4 SSTL
Job Pnomy Rules
-wWMAINIENANCE VARIANCE -HIGHMAINTENANC£ VARIANCE
Figure 6-18f Con'parison of Job R'iority Rules by Maintenance Service Variance
for Percentage ofTool Failures

04

E 03
1
I:
‘8
H

"6 02
E

E 0!
Do

 

DITC SDTC PSR4 SSTL
Job Pnomy Rule:

-IDWMAINTENANCE VARIANCE - HIGHMAINuaNANCE VARIANCE

 

 

 

 

235

 

Mean Time In System

Figure 6-19a Corrparison of Tool Control Rules by Nbintenance Service Variance

for Mean Tune in System

 

 

 

 

m m VARLO m mun.

VAR”
Tool Control Rules
-IoWMAINIENANCE VARIANCE - HIGHMAIN’IENANCE VARIANCE

 

 

 

 

 

Std Dev TIme In System

Figure 6-19b Comparison of Tool Control Rules by Nhintenance Service Variance

for Standard Deflation Tune In System

 

 

 

 

KIM FPTPM VARLO

VARHI
Tool Control Rules
-IoWMAIN'IENA NCE VARIANCE -HIGHMAINTENA NCE VARIANCE

 

 

 

 

 

236

 

Mun Tardiness

Figure 6-19c Comparison of Tool Control miles by Maintenance Service Variance

[zir Mean Tuizness

 

 

 

 

 

 

m mm VARLO VARHI
Tool Control Rules

-wWMAINIENANCE VARIANCE - HIGH MAINIENA NCE VARIANCE

 

 

 

Std Dev Tardne ss

Figure 6-19d Con'parison of Tool Control Rules by Maintenance service Variance

for Standard CeViaIICn Tardiness

 

 

 

 

 

 

 

W m VARLO

VAM
Tool ControlRules
-wWMAINIENANCE VARIANCE - HIGHMAINTENANCE VARIANCE

 

 

 

l

 

237

 

Percentage of Jobs Late

Figure 6-19e Convarison of Tool Control Rules by Nhintenance Service Variance
for Percentage oflobs Late

 

 

 

 

 

 

m FPTPM VARLO mm 111m.

VARHI
Tool Control Rules
-LOWMAIN'IENA NCE VARIANCE -HIGHMAINEENA NCE VARIANCE

 

 

Perc entage of Tool Fulute s

 

Figure 6-19f Comparison of Tool Control Rules by Maintenance Service Variance
for Percentage ofToolFaIlures

 

 

 

 

 

 

 

 

 

 

mm FPTPM VARLO

 

VARHI PM
Tool ControlRules m

-DOWMAINI£NANCE VARIANCE - HICHMAINIENANCE VARIANCE

 

 

238

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2.52.095 252.33.. 4922.093 4.25.493 2.52.493 2.52..me
2.52.552 0.3.2.33. 2:3-095.c- .99.5..9_.5 4959.095 2.52.95...
2.2.2-2.2.. 4.25735; ...—befiuba dang-Chan ARGO—{Pam 2.202.va
2202 ...-9mm 2E... -35.. -..—52.3.93 2.52 ..me ...—.0920th 2.52 0.55
3:23.493 . 25209.5 2.52.493 2.5.2-.493 2.5.693 4959.493
_ C. 2.25.093: . 51(5va _=~.<>.i:.mm 22,526th 2202-925 0.575..me
33<>§£ 2.52-553 2.2.3.995 29:34-93 2.5.493 4955.095
_ .: .5595 55>-33.. 2.302 -493 2.52 ...... 7- 2.53595 49.57.93
0.3.3.552 ~ 2.30233. . 22......me 0.3;-..93 0.3.9.493 0.5.2.953
.955...me . 2.52.493 . 2.52.093 Q3<>..Emm 0.3.3.993 0325-33..
_ 49.5.559- m 5.323.895 ,2 3.75293 .zm<>-49mm _:~.<>49mm . 03.9-59.5
459.33.. . 0.3.3493. . 2.3.3.493 2.3.9..me 2.3.3493... W 4955733..
2.9.3.553 . 325,595... . :4:<>.u95m 2.302.493 2302..me . 2.5.0.53
2....EC95 . 2302-5-55 H 2.302.053 2.3.9.093 2.3.3.099. . 2...... 49mm
22:“..va M $525.0th 2.333.922 2.502.922 2.2.02.9:5 ” 2.2.5.6920
2.5.... -493 . 2..z<> 0.5.0 2.502 .95—Q _§<>.C._.am 51(50th _. 2.5—(5.493
2.313.493. . 4955.493. 553.95... 2.325-595 0.343-va _ - 2.3.3-3.3
_ 2.3.3-L95 2.3.3493 4955.-va 2.52.595 4955733.. 2.3.3.035
2.3.3.33. 2.3.2-49¢... 2.3.3-33.. 0.3.9-995 2.302-955 2.2—(>33.
.57.)...th fiscawfiram 2..x<>-b._.xa 2.502 .9th 2202.5”. . _=~.<>.4._.mm
55>-.993 2.9.9.95... 2.3.3-33.. 252.33.. 2..~.<>-s.m.. . 2.302.995
55>-var. _ 2.3.3-95.. 2.9.3.33. 05:533.. 2.9...-va 2.302.493
2.5—(>302; _Ix<>..:bm 2.502...sz .95—(50.5.0 2.5—(>0th .95—(>0th
=3<>u95m 49.5755 5.75-33. 49.5.33. 2.52.8.5 2.352-953
2302-995 ” 2.302.993 2.52.995 2.3.3-Em. Q3<>¢9~5 555-33..
2.502.493 _=u<>..:.mm 2202-35.. 2255.53 §<>é¢mm _=¢<>..u._.om
2.302.952 2.302 .493 Q3<>695 553.32.. 2.302633. 2.302093
2.502-332 wag-OED _:~.<>.U-..¢Q 2.302.0sz €50.55 2.302-va
«2.4—...". 38... 23 maniacs-r 829$ :— 05; 8290.0
.5 53:089.. no.4 3o. 5 53:920.. c2350 c.3935 $2.63... cataton— Ea—ESm .__ 25... :32

 

mmasm Houucoo H009 >2
mmasm xufluofium now we mcomwummeoo mamfiuazz xmxse mmim manna

 

239
performance. The tool rules, MQBPM, VARHI, and VARPM are the best

performers regardless of job priority rule. The job priority rules influence cause
variation in the relative performance. For example, when job rules PSR4 and SDTC
are combined with MQBPM, VARHI, and VARPM, performance for mean time in
system tends to be the best. The worst tool control rule performers are NOPM,
JDDTL, and VARLO. Once again, the job priority rule causes only small increases
or decreases in the relative position.

For standard deviation of time in system, the consistency of performance of
the rules is less certain. The job priority rules tend to play a larger role in relative
performance for this measure. Job priority rules PSR4 and DRTC improve
performance the most. When these two job rules are combined with VARHI and
MQBPM tool rules, the best performance results. The job rules SSTL and SDTC
tend to cause performance to worsen. Only the JDDTL tool control rule showed
consistently poor performance.

For mean tardiness, job priority rules are more influential than tool control
rules in determining performance. The first group (top) of significantly different
combined rules is dominated by the PSR4 job rule, followed by DRTC. This
outcome is what would be expected for job rules which prioritize by due date. It
should also be noted that the first significant group contained all seven tool control
rules because they were combined with job rule PSR4 (Table 6-25). The
combination of SSTL and SDTC job rules with JDDTL and FPTPM tool rules

results in the worst performance.

 

240

For standard deviation of tardiness, the job priority rules are also more
influential than tool control rules. The job rules which promote better performance
are PSR4 and DRTC, with SDTC and SSTL resulting in poorer performance (Table
6-25). As for the tool control rules, there is no clear dominate rule.

For percentage of jobs late, job priority rules SDTC provides consistently
high performance, while PSR4 gives the worst performance. No clear combination
of job priority and tool control rules perform best for this measure.

For log percentage of tool failures, the tool control rules influence
performance more than job priority rules. The first two significantly different
groupings are dominated by the tool control rules, MQBPM and VARHI. The worse
performing tool rules are NOPM, VARLO, and JDDTL. No clear dominate job

priority rule can be found (Table 6—25).

6.4.3 Summary Qf Egg; Hm Analysis

Results from the m analysis confirm findings found in the analysis of
effects and a prior analysis. The following is a summary of the m analysis.

- The relative performance of both job priority and tool control rules are not
affected by tool life and maintenance service variance (also evidenced in the
ANOVA results). Control rules perform better under LOW variances than HIGH
with little difference between each level. This is to be expected since shop
performance generally deteriorates under high variance. The better performing tool

rules (MQBPM and VARHI) exhibit robustness for both forms of variance. MQBPM

 

241

and VARHI tool rules also exhibit robustness across performance measures. The job
priority rules (PSR4 and DRTC) also exhibit robustness in the presences of variance.
The job priority rules do not exhibit robustness with respect to performance
measures. The PSR4 rule gives the best performance on most measures, except
percentage of jobs late where it performs worse.

- No job priority or tool control rule exhibited cross-over effect. While the
result was expected, this may not always hold true under different model parameters.

- No single combination of tool and job rules perform best for all
performance measures. The most consistently high performance is obtained when the
PSR4 job rule is combined with either MQBPM or VARHI tool rules. lt should be
noted that both PSR4 job rule and MQBPM tool rules are information intensive.
Both heuristics consider more conditions in their decision process then most other

rules. Clearly, this shows additional information is beneficial.

6.5 Discussion of Results

The analysis of effects, m hypothesis, and ppm analysis found that
two tool control rules, MQBPM and VARHI, consistently outperform all other tool
rules. As for job priority rules, PSR4 provides the best performance except for
percentage of jobs late. There are however, a number of results which did not
conform to expectations. The following is a discussion of why the outcomes
differed.

The expected outcome that SSTL job rule would outperform SDTC rule was

 

 

 

242
based on past research (Lyman, 1993). Job rules which considered tool utilization
improved performance under deterministic tool life. This model used stochastic
(unknown) tool life. By selecting the tool which can be utilized the longest on a
machine, a slight increase in tool failures was observed. This lowered the
performance of SSTL. The other reason SSTL performed poorly is that it did not
give consideration to job due date when tool change took place. This causes further

worsening of SSTL's performance.

 

The job rule SDTC considers due date for all jobs in queue when tool
changes take place. For this reason, SDTC performed slightly better than SSTL.
Both SDTC and SSTL suffer because they are myopic. They attempt to fully utilize
an existing tool setup without considering all jobs in queue due dates.

Neither SDTC nor SSTL perform better than the PSR4 job rule. PSR4 rule,
like SSTL and SDTC, allow sequence dependency to utilize existing tool setup
(Melnyk et al., 1989). The advantage of PSR4 is the rule re-examines the machine
queue in its entirety after each job is completed. The re-examination allows the
PSR4 rule more information in its decision process, thus giving it top performance
among all the job rules.

The exception for PSR4 job rule is the percentage of jobs late performance
measure. The reason for this poor performance can be explained by the priority
setting. Priorities are set by job slack and setup requirements. While PSR4 rule
responds quickly to jobs which are past due (negative slack), this increases the

number of jobs late while simultaneously keeping mean and standard deviation of

243

tardiness low. To improve PSR4 job rule on percentage of jobs late requires only a
minor modification to the job slack value.

The use of due date information is essential in shop performance. Both PSR4
and DRTC job priority rules use due date status of all jobs when deciding which job
to process next. Both rules perform consistently well on the performance measures
as well as being robust to system variance. The evidence shows that the key piece of
information for job priority rules is due date status with setup requirements playing a
minor role in performance.

While results in this study generally confirms results from past research
(Melnyk et al., 1989; Lyman, 1993), it also contradicts certain research. The
conclusions that sequence dependency is a key determinate in shop performance is
not supported in this model. Articles by Mahmoodi et al. (1990), Mahmoodi and
Dooley (1991), and Kannan and Lyman (1992) support the position that sequence
dependant rules (family rules) improve performance. In the model used for this
research, sequence dependant job rules result in worse performance. This conclusion
is based on a lower setup time to processing time ratio (20% vs. 33%) and in the
presences of finite life tooling resource.

There were a number of surprising results that occurred with the tool control
heuristics. While it was expected that PM tool rules would outperform the CM tool
rule, there are a number of exceptions. The results confirm past research of Kay
(1978) and Banerjee and Burton (1990) by pointing out that PM is preferred in most

cases. Banerjee and Burton showed that some PM rules can decrease performance

 

 

because of maintenance frequency.

The poor performing PM tool rules, JDDTL and FPTPM, suffer from higher
total maintenance time. These rules cause high frequency of tool failures and thus
suffer the plight that NOPM tool rule does. JDDTL and FPTPM tool rules also do
not benefit from PM to the extent that tool rules MQBPM and VARHI do. The
result is higher maintenance time due to tool failures compared to VARHI and more
frequent maintenance trips than NOPM. The higher frequency allows NOPM to
outperform JDDTL and FPTPM on several performance measures. Table 6-26
provide an illustration of how PM tool rule can perform better or worse than the

CM tool rule.
—

Table 6-26 Example of Total Maintenance Time

 

 

 

 

TOOL NUMBER MAINTENANCE NUMBER MAINTENANCE TOTAL TIME
RULE OF CM TIME FOR CM OF PM TIME FOR PM

(6 Hours) (3 Hours)
NOPM 30 180 0 0 180
JDDTL 17 102 30 90 192
VARHI 2 12 45 135 147

 

 

 

 

 

 

 

 

Other results which did not conform to expectations was that of hypothesis 4.
Results showed that allowing early PM (VARHI and VARPM) outperform the fixed
point PM rule (FPTPM). The unexpected result is that variable PM rule VARLO
does not outperform FPTPM. This points out an interesting conclusion. Early PM
reduces the risk of tool failure, which increases tool availability via less maintenance

delay. Higher tool availability increases shop performance. Past work by Banerjee

 

 

245

and Burton (1990) and Pete-Cornell et al. (1987) found that shop performance
decreases with high PM frequency. Conversely, the more frequent PM under tool
rules VARHI and VARPM contradict these past findings. The tool rules VARLO
and FPTPM cause a higher frequency of tool failures, thus lowering performance.
As stated previously, VARHI tool rule outperforms VARLO for the reason I]
cited above. In Hypothesis 5, it is stated that the reverse would hold true, VARLO i

would outperform VARHI. Clearly, the added risk to process an additional job is not

 

beneficial, but detrimental to performance. This same conclusion can also be applied b
to Hypothesis 7. The added tool failure risk that rule JDDTL allows, actually causes
all performance measures to worsen when compared to FPT PM.

The results of Hypothesis 6 are as expected, the use of maintenance queue
information does improve shop performance. The logic behind MQBPM tool rule is
similar to that of VARHI, which encourages early PM when conditions permit.
While these two rules may not operate identically, the results show they perform
equally well.

The key component to shop performance for this model as well as in past
models is tool (resource) availability (Melnyk et al., 1989). This conclusion can be
seen in the performance of the various tool control rules. The tool rules, MQBPM
and VARHI, cause the least tool failures and thus the shortest total maintenance
service delay. The less time spent in the maintenance process the greater tool
availability. If PM is too frequent, the result will be decrease shop performance

(Banerjee and Burton, 1990).

246

The tool heuristics developed for this model take the perspective actually used
on shop fioors. If there is a past due job which needs a tool. keep the tool in
production as long as the tool's limits have not been surpassed. When the choice is
between processing jobs early (prior to due date) versus sending a tool in for
maintenance, choose maintenance. This is especially true for the MQBPM tool rule
which uses maintenance queue information. If there is no backlog of tools (queue is
empty) waiting for maintenance. then its advantageous to seize the opportunity. The
other advantage to the use of maintenance queue information is that tool repairs can
be spread out, allowing balanced maintenance workload.

This last point brings up the issue of planned and unplanned maintenance.
While not considered explicitly by the PM tool rules, they do implicitly plan
maintenance. Past articles have pointed to the need and benefits of planned
maintenance (Bojanowski, 1984; Christer and Whitelaw, 1988). This research has
demonstrated that PM is beneficial compared to CM. Further, the ability to balance
maintenance workload through the MQBPM tool rule is similar to planned
maintenance. If planned maintenance is to be truly effective, then it must balance
maintenance workload while simultaneously considering shop floor demand. Linking
the maintenance process to the shop floor schedule or MRP schedule would
accomplish this task.

The results from this experiment has contributed to further understanding in a
number of ways. The use of information, both type and frequency, can improve

shop performance. While past research supports this proposition (e. g. Fredendall,

 

,\
1:.
.o.’
3.:
‘3:
-b‘

‘ ,' 1‘
5.

‘ ~C‘.
‘J
.r}
U

 

247

1991), this research highlights the benefits of information in a stochastic
environment. This includes the use ofjob due date status and setup requirements.

When job due date is checked intermittently, shop performance deteriorates

significantly.
The use of information in the tool decision is also a major factor in shOp r]
performance. Maintenance queue backlog information allows the tool rule MQBPM -‘-:

to significantly out perform rule VARPM. Information on shop floor demand also

-‘

 

leads to the success of the variable PM rules. By evaluating job demand, the tool J
rules can evaluate whether there is a need to keep the tool in production or allow for
early PM.

This research has also contributed by developing a number of unique tool
control rules. Past research has focused almost exclusively on fixed point PM rules
(Banerjee and Burton, 1990; Christer and Whitelaw, 1984). The results from this
research indicate that variability in the PM decision can improve performance over
the fixed PM point rules. The variability allows for consideration of such factors as:
shop demand, tool condition and maintenance load. This provides the decision maker
the opportunity to evaluate the environment and make an informed decision.

The last contribution of this research is the recognition that control of a finite
life resource requires a different approach. Whereas past DRC research has viewed
the resources of labor and machine as having infinite life, this model replaces labor
with a finite life resource (tooling). This changes the characteristics of the DRC

model. To control a finite life resource requires the development of new heuristics.

248
Control heuristics which work for labor are inappropriate for tooling. Research by
Nelson (1966), Goodman (1972) and Fryer (1974, 1978) point out that the where to
send and when to send a resource. With finite life resource, an additional series of
questions must be asked: 1) is the resource available, 2) how much of the resource is
remaining, and 3) is that sufficient life? While this adds complexity to the control
heuristics, these rules are more representative of the real decision made on the shop

floor.

6.6 Summary of Analysis

Results show that the factors, tool control rules (MTOOL) and job priority
rules (NRULE), are the determinates of shop performance. Of these rules, several
stand out as effective and robust. The early variable PM rules, MQBPM and
VARHI, when combined with PSR4 job rule provide consistently high performance.
While variable PM gives better performance over fixed point PM tool rules FPTPM
and JDDTL, this flexibility is not totally beneficial. The expectation that variable
PM tool rules which postpones PM would improve performance, especially
tardiness, was not realized. In fact, the opposite was true.

The expectation that sequence dependant job priority rules SSTL and SDTC
would improve performance was not realized. The poor performance is attributed to
the fact that: l) the setup to processing ratio is 20 percent, and 2) to their myopic
focus on machine/tool setup. The job priority rule DRTC, while not performing as

well as PSR4 rule, does provide fairly robust performance. DRTC focuses on job

 

 

 

249

due date as its priority and thus performs better on those measures which focus on

delivery performance.

6.7 Future Research Directions
The experimental condition and results suggest several future research 1
avenues. The following is a brief description of possible extensions.

- The relationship of the PM point to mean tool life needs further

 

examination. This can be accomplished by adjusting the PM point relative to mean H
tool life. By changing the parameters of the model, examination of the tool control
rules will illustrate how they differ under various environments. Should the PM
point be moved further from mean tool life, such tool rules as VARLO or FPTPM
may improve in their performance. Examining different PM points will allow for a
more appropriate placement of the PM point or aid in the development of simpler
yet effective PM tool rules. While Banerjee and Burton (1990) varied the PM point,
they only examine the effect of such a change on a fixed PM point rule.

- Another tool control rule issue which needs to be explored further involves
the degree of variability in the variable PM tool rules. The results of this study has
found that the use of variable PM proVides the best performance. Since this research
is the first known work which tests a variable range for the PM decision, additional
tests are needed. Adjusting the current 10% range for variable PM to several values
both higher and lower will be necessary. This will provide a wider range of

information on what is an appropriate range for a variable PM tool control rule.

250

Such information would benefit managers who are establishing parameters for a
maintenance program.

- A third issue which must be explored is tool life distribution. The use of a
normal distribution was selected because of discussions during plant visits and
several articles (Lie et al., 1977; Fenton and Joseph, 1979). Past research has shown
that varied tool life distributions like log-normal, Weibull, and gamma can alter
results (Ramalingam et al., 1978; Bao, 1980). Examining the tool heuristics under
different distributions will test the robustness of the control rules. Once again, the
benefits to this analysis would be to provide a manager information from which to
make a decision.

- The frequency of maintenance issue needs to be examined in greater detail.
The results of this research contradicts the outcome found by Banerjee and Burton

(1990). Figure 6-19 illustrates how frequency of maintenance and total maintenance

 

 

Figure 6-20 Total Maintenance Time

A)
Pro uenc ,
q ‘y Total

Of CM Maintenance '

_ Time _/ . .

Maintenance \ \- -- "
Time \ Frequency
\ \X of PM
‘ N

 

 

”Frequency of Maintenance

 

 

 

251

time are related. While this relationship seem straight forward, the variable PM tool
rules seem to alter the shape of the total maintenance curve. Additional research
needs to focus on how the various tool rules influence the relationship.
- Results from this research has shown PSR4 rule is effective for all
measures, but percentage of jobs late. Modification of PSR4, plus the addition of job "I
priority rules which consider tool availability, should be studied. While the job i it

priority rules studied in this model considered the PM point, future job priority rules

 

should examine remaining tool life and maintenance status. The addition of these u
forms of information are likely to improve performance.

- Other parameters that should be altered to develop a comprehensive
understanding of tool control is tool flexibility. The use of flexibility in the PM rule
has been found to improve shop performance. Would the ability to move tools
between machines improve performance? Past DRC research has found labor
flexibility to be beneficial (e. g. Fryer, 1974; Goodman, 1972). The likely results
will show that the addition of flexibility to a resource improves performance
(Swamidass, 1988). Past DRC models (e.g. Melnyk and Lyman, 1991) have
developed control heuristics which are useful for labor control but are not likely to
be useful with finite tool life. New control heuristics will need to be developed
which consider tool life and maintenance delay.

- Another parameter issue which needs further study involves multiple c0pies
of each tool. While this has been studied in past research (e. g. Bao, 1980; David

and Purcheck, 1981). Multiple tool copies have never been studied in a DRC

252
environment or when maintenance (PM or CM) needs to be performed. Once again,
heuristics will need to be developed which incorporates the additional information.
Such information as location, remaining life, and specific application will need to be
considered for the effective management of the shop floor.

- The last research issue which needs further investigation is that of shop
floor performance measures. The traditional DRC model has focused on measures
specific to job performance (Treleven, 1988). Such measures as time in system and
tardiness, while important, are not the only consideration. Fredendall (1992) used
labor cost data to evaluate shop performance. The cost of tool changes and
maintenance needs to be considered to evaluate the appropriateness of both tool and
job heuristics. Other non-job oriented performance measures that will aide in
comprehending tool rule influence includes: total maintenance time, maintenance
utilization rate, and utilization rate. The different measures provide a means to

further evaluate shop performance.

6.8 Conclusion

The main findings of this study consist of the following:

- Preventive maintenance improves performance over corrective maintenance
in most cases.

- Preventive maintenance policies which promote early or frequent
maintenance improve performance over fixed point or postponed PM policies.

- The use of certain forms of information, both in job priority and tool

 

 

253

control rules, can improve shop performance.

- Tool life and maintenance service variance is not an influence on shop
performance due to the small loss of tool availability from maintenance delay.

- Job priority rules which consider due date and setup requirements prior to
job processing enhances performance the most.

The above results are the conclusions drawn from the analysis. To complete ”—1

the purpose of this research there are a number of questions which must be answered

 

that were developed in Chapter 1.

The first question: how do we schedule jobs, can be answered by examining
the results of job priority rules. The PSR4 job priority rule provides the best
example of how to schedule jobs in the most effective manner. The PSR4 rule
considers due date, setup requirements, and tool availability when selecting a job.
Scheduling of jobs requires that tool be available and then prioritizes based on either
due date or setup requirements or both. The use of all three sources of information
provides the best method of scheduling jobs.

The second question examines how do we effectively schedule tools for
production and maintenance? The scheduling of tools for production is determined
by the job priority rules. The scheduling of maintenance is determined by the tool
control rules. Those rules which promote early PM or frequent maintenance provide
the best performance. Allowing tools to remain in production until they fail cause
shop performance to deteriorate.

Thus, tool rules provide the best means of scheduling tools for maintenance. The

 

254

tool rule JDDTL, which attempts to schedule tools to be available for production at
the expense of maintenance, also cause shop performance to deteriorate. Additional
research needs to be conducted to explore tool scheduling that considers both
production demand and planned maintenance. No rule was able to consider more
than one tool at a time. By considering multiple tools simultaneously, scheduling of
production and maintenance is likely to improve performance. P}
The third question is, how does tool life and maintenance variation affect

scheduling decisions? Based on the We; analysis, the control heuristics exhibited

 

robust behavior and is not influenced by either form of variance. Li
By answering these three specific questions, the general research question:

how do we effectively manage a DRC shop given stochastic tool life and

maintenance service, can be resolved. To answer this question, the most effective

means to managing a DRC shop is to divide the decision process into two segments.

The first segment is job selection. The use of PSR4 job priority rule provides the

highest level of performance. The second segment consists of tool allocation via tool

control rules. The most effective means of improving performance is through the use

of variable PM rules which promote early maintenance. The best rules are VARHI

and MQBPM. The combined rules PSR4-VARHI and PSR4-MQBPM provide the

 

best and most robust set of rules. What this points to is additional information and
frequent/early maintenance enhances performance.
The final part of this conclusion will focus on the application of the results

obtained from this model. The model is based on information obtained from two

255

flow shops. Thus, the results from this research are generally applicable to these two
manufacturing environments. Results are not specifically applicable because the
model is a combination of both environments. While both shops were similar, they
differed in several ways. This includes their approach to preventive maintenance and
tool control. The results do indicate that the shop which promotes frequent PM is
following the best approach. To confirm these results, additional tests will be
necessary using assumptions which more closely simulate the specific environment.
The generalizability of the results obtained from this study is limited to those
shop environments which emulate the assumptions used. The DRC model is not
indicative of shop environments because there are only two constrained resources.
The shops which were visited have additional constraints like labor and materials.
These shops also operated under different assumptions, like machine failure and tool
flexibility, than those used in this model. To conclude that the results from this study
are applicable under all constraints and assumptions, would not be valid. It should
also be noted, that the results from this flow shop model may not be applicable to
job shops. What this illustrates is the need for additional research. By adding
additional constraints and modifying the assumptions, it may then be possible to
apply the results to a specific shop or develop a framework from which results can

be generalized to all shop environments.

BIBLIOGRAPHY

 

 

BIBLIOGRAPHY

A.S.T.E. Handbook Committee, Tool Engineers Handbook, Mcgraw—Hill Book
Company, New York, New York, First Edition, 1949, Section 97, p. 1494.

Adair-Heeley, C.B., ”The JIT Challenge for Maintenance,” Production and Inventory
Mmagement Raview Q APICS News, Vol.9, (1989), No.9 (Sept), pp. 34-35.

Allen, M., "The Efficient Utilization of Labor Under Conditions of Fluctuating
Demand,” In J. Muth and G. Thompson(Eds.), Industrial Sgheduling, Englewood
Cliffs, NJ, Prentice—Hill, I963.

Amoaka-Gyampah, K., Meredith, J .R. and Raturi, A., "A Comparison of Tool
Management Strategies and Part Selection Rules for a Flexible Manufacturing System,"
International Joarnal of Production Research, Vol.30, No.4 (1992), pp.733-748.

Ashton, LE. and Cook, F.X., "Time to Reform Job Shop Manufacturing," Harvard
Businass Raviaw, Vol.67, No.2 (1989), pp. 106-1 1 l.

Avi-Itzhak, B., Maxwell, W.L., and Miller, L.W.,"Queuing With Alternating
Priorities,” W, Vol.13 (1965), No.2, pp. 306—318.

Ayres, R.V., "Future Trends in Factory Automation," Manufacturing Review, Vol.1
(1988), No.2, pp. 93-103.

Baker, C.T., and Dzielinski, B. P., "Simulation ofa Simplified Job Shop," Journal of
Qwrationg Maaagamant, Vol. 6, 1960, pp. 311-323.

Baker, K.R., Intrmagtign ta Saguencing and Scheduling, John Wiley and Sons, 1974.

Baker, K.R., ”Sequencing Rules and Due Date Assignments in a Job Shop,"
Managemant Sgiange, Vol. 30 (1984b), No.9.

Baker, R.D., "Testing the Efficacy of Preventive Maintenance," Journal Qf the
ngratignal Raggaggh Smiety, Vol.42 (1991), No.6 (Jun), pp.493-503.

Bao, H., ”Application of Dynamic Programming to Optimize Tool Replacement

256

 

 

257

Schedules for Multi-Tool Operations Involving Distributed Tool Lives," Journal of

Mfihanieal Design, ASME, Vol.102, July 1980, pp. 446-452.

Banerjee, A., and Burton, 1.8., "Equipment Utilization Based Maintenance Task

Scheduling in a Job Shop," Bergman Journal Qf Operational Research, Vol.45 (1990),
No.2-3 (Apr 13), pp. 191-202.

Banks, J., and Carson, J. S., Discrete-Event System Simulation, Prentice-Hall, Inc.,
Englewood Cliffs, NJ, 1984.

Barlas, Y., "An Autocorrelation Function Test for Output Validation," Simulation, July
1990. PP. 7-15.

Bertrand, J.W.M., ”The Use of Workload Information to Control Job Lateness in
Controlled and Uncontrolled Release Production Systems," Journal of Operations
Management, Vol.3 (1983), No.2.

Blackburn, J.H., Capacity Management, South-Western Publishing Co., Cincinnati.
OH. 1989.

Blackstone, J.H., Phillips, D.T., and Hogg, G.L., "A State-of-the-Art Survey of
Dispatching Rules for Manufacturing Job Shop Operations,” Journal of Operations
Management, Vol.3 (1983), No.2, pp.79-92.

Bojanowski, R.S., "Improving Factory Performance with Service Requirements

Planning (SRP)," Ergfluetion and Inventog Management, Vol.25 (1984), No.2 (Second
Quarter), pp. 31-44.

Brad, J .F., ”A Heuristic for Minimizing the Number of Tool Switches on a Flexible
Machine," IE Tgaasaetien, Vol.20 (1988), No.4 (December), pp.382-391.

Broom, H.N., Prgflaetien Maaagement, Homewood, IL: Richard D. Irwin, Inc. 1967.
Brown, G.G., Geoffrion, A.M., and Bradley, G.H., "Production and Sales Planning

with Limited Shared Tooling at the Key Operations," Management Scienee, Vol.27
(1981), No.3 (March), pp.247-259.

Browne, J ., Boon, J .15., and Davis, B.J., "Job Shop Control," International Journal of

Preduction Reggteh, Vol.19 (1981), No.6, pp.633-643.
Buffa,E.S.,M 'n th m i've h In ' uf 'n St te for US.

Cemeaaies, Dow-Jones Irwin, 1984.

Bulkin, M.H., Cooley, J.L., and Steinhoff, H.W., "Load Forecasting Priority

 

 

258

Sequencing, and Simulation in a Job Shop Control System," Management Science,
Vol.13 (1966), No.2 (October), pp.B29-b51.

Burbridge, J., "A Manual Method of Production Flow Analysis," The Production
Engineer, Vol.56 (1977), No.10.

Byrkett, D.L., Ozden, M.H., and Patton, J.M., ”Integrating Flexible Manufacturing
Systems with Traditional Manufacturing, Planning, and Control," Production and

Inventety Management Jeumal, Third Quarter, 1988, pp. 15-20.

Carrie, AS, and Perera, D.T.S., "Work Scheduling in FMS under Tool Availability
Constraints," International Journal of Production Research, Vol.24 (1986), No.6,
pp. 1299-1308.

Chang, T.C., and Wysk, R.A., An Introduction to Automated Process Planning
Systems, Prentice-Hall, Englewood Cliffs, NJ, 1985.

Christer, A.H., and Waller, W.M., "An Operational Research Approach to Planned

Maintenance: Modelling P.M. for a Vehicle Fleet," Journal of the Operational Research
mm, Vol.35 (1984), No.11, pp.967-984.

Christer, A.H., and Whitelaw, J ., "An Operational Research Approach to Breakdown

Maintenance: Problem Recognition," Jeurnal pf the Operational Research Society,
(UK), Vol.34 (1983), No.11, pp. 1041-1052.

Cohen, P.H., and Black, J.T., "Tool Life Distribution Discussion," Journal of

Engingring fer Inthtstgt, Vol.97 (1977), No.3.

Conway R.W., Maxwell, W.L., and Miller, L.M., Theoryof Schedulinfig, Reading,
MA: Addison-Wesley Publishing Co., 1967.

Cook, N.H, Mapafaettsg'ng Analysis, Addison-Wesley, Reading, MA, 1966.

Cook, N.H., "Tool Ware and Tool Life," ASME, Jeurnal of Engineering for Industry,
Vol.93 (1973), No.11 (November), pp.931-938.

Cook, T. D., and Campbell, D. T., Quasi-Experimentation; Design & Analysis Issues
fer Field Settings, Houghton Mifflin Company, Boston, Mass., 1979.

Cross, K.F., "Making Manufacturing More Effective by Reducing Throughput Time,"
National Eredaetivity Review, Winter 1986-87, pp. 35-47.

Day, I .13., and Hottenstein, M.P., "Review of Sequencing Research," Naval Research

W, Vol.1? (1970), No.2 (March), pp.ll-39.

 

259

Daoud, Z.A., and Purcheck. G.F.K., "Multi-Tool Job Sequencing for Tool-Change
Reduction," n rn inl umal f Pr uction Research, Vol.19 (1981), No.4,
pp.425-435.

Daskin, M., Jones, PC, and Lowe, T.J., "Rationalizing Tool Selection in a Flexible
Manufacturing System for Sheet-Metal Products," Operations Research, Vol. (1992)

De Meyer, A., Nakane, J., Miller, J.G., and Ferdows, K., ”Flexibility: The Nest
Competitive Battle, The Manufacturing Futures Survey," Strategic Management
Jamal, Vol.10 (1989).

De Souza, R.B.R., and Bell, R., "A Tool Cluster Based Strategy for the Management
of Cutting Tools in Flexible Manufacturing Systems," Jettrnal of erations
Management, Vol. 10 (1991), No.1 (January), pp. 73-91.

Deis, P., Production and Inventory Management in the Technological Age, Englewood
Cliffs, NJ, Prentice-Hill, 1983.

Denzler, D.R., Boe, W.J., and Duplaga, E., "An Experimental Investigation of FMS

Scheduling Rules Under Uncertainty," Journal of Operations Management, Vol.7
(1987), No.1-2 (Oct), pp.139-151.

Elvers, D.A., and Treleven, M., "Job-Shop vs. Hybrid Flow-Shop Routing in a Dual
Resource Constrained System," Dgision Science, Vol.16 (1985), No.2 (Spring),
pp.213-222.

Erhom, C.R., "Tooling Planning and Scheduling Systems," in Execution and Control

Systems; Cempttters in Manafaeturing, Auerbach Publishers Inc. May 1983.

Fenton, R.G., and Joseph, ND, "The Effects of the Statistical Nature of Tool-Life on

the Economics of Machining," Internatipnal Journal of Machine TQI Design Research,
Vol. 19 (1979), pp. 43-50.

Filliben, J.J., "The Probability Plot Correlation Test for Normality," Technometrics,
Vol.17 (1975), No.1, pp.111-117.

Finch, 8., "Japanese Management Techniques in Small Manufacturing Companies: A
Strategy for Implementation,” Production and Inventory Management, Vol.27 (1986),
No.3 (Third Quarter), pp. 30-38.

Finch, B., and Gilbert, J.P., ”Developing Maintenance Craft Labor Efficiency Through
an Integrated Planning and Control System: A Prescriptive Model," Jeumal of

WW, Vol.6 (1986), No.3/4 (May/Aug), pp. 449-459.

 

 

260

Fishman, G. S., "Grouping Observations in Digital Simulation," Management Science.
Vol. 24, No. 5, (January 1978), pp. 510—521.

Fishman, G.S., Erineiples pf Diserete Event Simulation, New York, 1978.

Fishman, G. S., and Kiviat, P. J., ”The Statistics of Discrete-Event Simulation,"
Simalatipa, October 1968, pp. 185-195. ‘

Fredendall, L.D., "An Experimental Investigation of Information Use in a Job Shop
Operating Under Dual Resource Constraints: A Simulation Study," Ph D Dissertation

Miehigan State llniversity, 1991.

Fredendall, L.D., Melnyk, S.A., and Ragatz, G., ”Information and Scheduling in a
Dual Resource Constrained Job Shop,” Werking Pamr, Dec. 1992.

 

Fryer, J.S., "Operating Policies in Mutliechelon Dual-Constraint Job Shop,"
Management Seienee, Vol.19 (1973), No.9 (May), pp. 1001-1012.

Fryer, J.S., "Organizational Structure of Dual-Constraint Job Shop," Decision Science,
Vol.5 (1974), No.1 (January), pp.45-57.

 

Fryer, J.S., "Labor Flexibility in Multiechelon Dual-Constraint Job Shops,"
Management Seienee, Vol.20 (1974), No.7 (March), pp. 1073-1080.

Fryer, J.S., "Effects of Shop Size and Labor Flexibility in Labor and Machine Limited
Production Systems," Management Seienee, Vol.21 (1975), No.5 (January), pp.507-
515.

Fryer, J.S., "Organizational Segmentation and Labor and Machine Limited Production
Systems," Deeisien Seience, Vol.7 (1976), No.4 (October), pp.725-738.

Gallimore, K.F., and Panlesky, R.J., "A Framework for Developing Maintenance
Strategies," Pr u in Inv nto Mana ement, Vol.29, (1988), No.1 (First
Quarter), pp. 16-22.

Gavett, J .W., "Three Heuristic Rules for Sequencing Jobs to a Single Production
Facility," Management Seience, Vol.11 (1965).

Gayman, D.J., "Computers in the Tool Crib," Manufacturing Engineering, September,
1986. pp. 41-44.

Gerwin, D., "An Agenda for Research on the Flexibility of Manufacturing Processes,"
In m i n ion (1 P i n Mana m nt, Vol.7 (1987), No.1.

 

 

 

261

Ghosh, S. and Gaimon, C., "Simultaneous Determination of Production Maintenance,
In-Process Inventory, and Capacity in a Serial Production System," Working Paper,

Ghosh, 8., Melnyk, S.A., and Ragatz, G.L., "Tooling Constraints and Shop Floor
Scheduling: Evaluating the Impact of Sequence Dependency," International Journal of

Web. July 1991.

Gilbert, J.P., and Finch, B.J., "Maintenance Management: Keeping Up with
Production's Changing Trends and Technologies," Journal of Operations Management,
Vol.6 (1985), No.1 (Nov.),pp.1-l2.

Glazebrook, K.D., "Evaluating the Effect of Machine Breakdown in Stochastic
SchedulingProblems," v R r hL itic uarterl ,Vol.34,1987,pp.319-335.

Goldman, AS, and Slattery, T.B., Maintainability, John Wiley, 1964.

Goodman, S.H., ”The Effects of Machine and Worker Assignments on the Performance
of a Dual Constrained Job Shop Possessing Flexibility in Both Resources," Ph.D.

Qissettatign. Pennsylvania State University, 1972.

Graves, S.C., "A Review of Production Scheduling,: Operations Research, Vol.29
(1981), No.4 (July-August), pp. 646-675.

Gray, A.E., Seidmann, A., and Stecke, K.E., "Tool Management in Automated

Manufacturing: A Tutorial,’ Ermgdings ef the Third thSA/TIMS Conference on
Flexible Mapufaettlring Systems, 1989, pp. 93-98.

Gray, A.E., Seidmann, A., and Stecke, K.E., "A Synthesis of Tool- Management

Issues and Decision Problems in Automated Manufacturing," Working Paper,
University of Michigan, 1990.

Gruver, W.A., and Senninger, M.T., "Tooling Management in an FMS," Mghanical
Engineering, March 1990, pp. 40-44.

Gunther, R.E., "Server Transfer Delays in a Dual Resource Constrained Parallel
Queuing System," W, Vol.25 (1979), No.12 (December), pp. 1245-
1257.

Gunther, R.E., "Dual-Resource Parallel Queue with Server Transfer and Information
Access Delays," Wee, Vol.12 (1981), No.1 (January), pp.97-111.

Hall, R.W., Ere Inventeries, Homewood, IL: Dow Jones-Irwin, I983.

Hartley, J.R., EMS at Werk, IFS Publications, Bedford, UK, 1884.

 

 

262

Ham, 1., Hitomi, K., Nakamura, N., Yoshida, T., "Optimal Group Scheduling and
Machining Speed Decision Under Due Date Constraints," Transactions of the American

Smiety of Meehanieal Engineers, Vol. 101 (1979), No.5.

Hardy, S.T. and Krajewski, L.J., "A Simulation of Interactive Maintenance Decision,"

Qeeisign Seienee, Vol.6 (1975), No.1 (Jan.), pp. 92-105.
Harrigan, K.R., Sttategie Flexibility, Lexington Books, 1985.

Hartley, J., EMS at Wetk, New York, NY, IFS Productions Ltd. and North-Holland,
1984.

Hartmann, E.H., "Maintenance Productivity: Why It Is So Low and How to Improve

It," Natienal Ergluetivity Review, Vol.5 (1985), No.3 (Summer), pp.224-232.

Haynes, R.D., Kumar, C.A., and Byrd, J., "The Effectiveness of Three Heuristic

Rules for Job Sequencing in a Single Production Facility," Mgagement Scienee, Vol.
19 (I973).

Heidelberger, P. and Welch, P. D., "Simulation Run Length Control in the Presence
ofan Initial Transient," filmmtipns Researeh, Vol. 31, No. 6, (Nov. - Dec. 1983), pp.
1109-1144.

Hitomi, K., "Analysis of Production Models; Part I: The Optimal Decision of
Production Speeds; Part II: Optimization of a Multistage Production System," AIIE
Transaetions, Vol.8 (1976), No.1 (March), pp.96-112. .

Hitomi, K., "Analysis of Optimal Machining Speeds for Automatic Manufacturing,’
Internatienal thlmfl at Breduetipn Researeh, Vol.27 ( 1989), No.10, pp. 1685-1691.

Hitomi, K. and Ham, 1., "Group Scheduling Technique for Multiproduct, Multistage

Manufacturing Systems," Jeumal ef Engingring fer Industg, August, 1977, pp. 759-
765.

Hogg, G.L., Phillips, D.T., and Maggard, M.J., "Parallel-Channel, Dual Resource
Constrained Queuing Systems with Heterogeneous resources," AIIE Tgnsactiens, Vol.
9 (1977), No.4 (December), pp.352-362.

Hogg, G.L., Phillips, D.T., Maggard, M.J., and Lesso, W.G., "GERTS OR: A Model
for Multi-Resouree Constrained Queuing Systems, Part 1: Concept, Notation, and

Examples," Was, Vol. 7 (1975), No.2 (June), pp.89-99.

Hogg, G.L., Phillips, D.T., Maggard, M.J., and Lesso, W.G., "GERTS OR: A Model
for Multi-Resource Constrained Queuing Systems, Part II: An Analysis of Parallel-

 

 

 

263

Channel, Dual Resource Constrained Queuing System wit Homogeneous Resources."

W. Vol. 7 (1975), No.2 (June), pp.100-109.

Hollier, R.H., "A Simulation Study of Sequencing in Batch Production," Oxrational
ReseateLQpaflerly, Vol.19 (1968), No.4, pp. 389-404.

Holstien, W.K., and Berry, W.L., ”The Labor Assignment Decision: An Application
of Work Flow Structure Information," Management Science, Vol.18 (1972), No.7
(March), pp.390-400.

Hopp, W.J., and Wu, Sung-Chi, "Machine Maintenance with Multiple Maintenance
Actions," W, Vol.22 (1990), No.3 (Sept), pp.226—233.

Hsu, Lie-Fem, "Optimal Preventive Maintenance Policies in an M/G/l Queue-Like
Production System,” European Journal of Operational Research, Vol.58 (1992), No.1
(Apr 10). Pp. 112-122.

Huber, R.F., ”Control ofTooling Promises Bonanza," Production, December 1989, pp.
51-53.

Johnston, S.K., ”JIT: Maximizing Its Success Potential," Production and Inventory
W, Vol.30 (1989), No.1 (First quarter), pp. 82-86.

Kannan, V., and Lyman, S.B., "An Analysis of the Effects of Lot Splitting in Group

Scheduling,” ELQeeegings pf the Twenty-Third Annual Mid-West Decision Science
lnstitute Meeting, May 1992.

Kannan, V., and Lyman, 8.8., "An Analysis of the Effects of Lot Splitting in Group

Scheduling," Brmefljngs ef the flfwenty-fljhirg Annual Natipnal Decisipn Science
Instittite Meeting, November 1992.

Kay, E., "The Effectiveness of Preventive Maintenance," International Jettmal of
Prggluetion Reseaseh, Vol.14 (1975), No.3, pp.329-344.

Kelton, W. D., "Random Initialization Methods in Simulation," IE Transactipns, Vol.
21, No. 4, December 1989, pp. 355-367.

Kelton, W. D., and Law, A. M., "The Transient Behavior of the M/M/s Queue, with
Implications for Steady-State Simulation," Qwrations Resgpch, Vol. 33, No. 2,
(March - April 1985), pp. 378-396.

Kendall. L.J . and Bayoumi, A. , "Automated Tool-Wear Monitoring and Tool Changing

Using Intelligent Supervisory Control," lnternatipnal qurnal pf Prfluetipn Researeh,
Vol.26 (1988), No.10, pp. 1619-1628.

 

264

Kiran, AS, and Krason, R.J., ”Automating Tooling in a Flexible Manufacturing

System," nglastrial Engineering, April 1988, pp.52-57.

Kirk, R. E., Expeg'mental Design; Ermegtlres for the Behavioral Sciences, 2nd edition,
Wadsworth, Inc., Belmont CA, 1972.

Kleijnen, J. P. C., ”Analyzing Simulation Experiments with Common Random
Numbers," Management Seienee, Vol. 34, No. 1, January 1988, pp. 65-74.

Kleijnen, J. P. C., Statistical Tools for Simulation Practitioners, Marcel Dekker, Inc.,
New York, New York, 1987.

Kouvelis, P., "An Optimal Tool Selection Procedure for the Initial Design Phase of a

Flexible Manufacturing System," Eurpman Journal of Operations Management, Vol.55
(1991), pp. 201-210.

Krajewski, LI. and Ritzman, L.P., erations Management, Addison-Wesley
Publishing Co., Reading, Mass. 1990.

Kupferberg, M., ”Tooling: The Frontier of Capacity Management," American
r i Inv nt ntrl ie 29th Annual International Conference
W, St. Louis, MO, 1986, pp.l86-189.

La Commare, U., Diega, L., Nota, S., and Passabbabte, A., "Optimum Tool
Replacement Policies with Penalty Cost for Unforeseen Tool Failure," International

Jettmal pf Predectipn Eesgreh, Vol.23 (1983), No., p.237.

Law, A. M., ”Design and Analysis of Simulation Experiments for Manufacturing
Applications," Ermmings pf the 12%) Winter Simelatipn Cpnferenee, pp. 33-37.

Law, A. M., ”Statistical Analysis of the Output Data from Terminating Simulations,"
WW. Vol. 27. pP- 131-143-

Law, A. M., "Statistical Analysis of Simulation Output Data,"
erations Resgreh, Vol. 31, No. 6, (Nov. - Dec. 1983), pp. 983-1029.

Law, A. M., and Carson, J. 8., ”A Sequential Procedure for Determining the Length
of a Steady-State Simulation," Qpepations Researeh, Vol. 27, No. 5, (Sept. - Oct.
1979), pp. 1011-1025.

Law, A. M., and Kelton, W. D., "Confidence Intervals for Steady-State Simulation:
1. A Survey of Fixed Sample Size Procedures," W, Vol. 32, No. 6
(Nov./Dec.), 1984, pp. 1221-1240.

 

 

265

Law, A. M., and Kelton, W. D., "Confidence Intervals for Steady-State Simulation:
II: A Survey of Fixed Sample Size Procedures,” Management Science, Vol. 28, No.
5, May 1982. PP. 550-562.

Law, A. M., and Kelton, W. D., Simulati_on Modeling and Analysis, Second Edition,
McGraw-Hill, Inc., New York, New York, 1991.

LeGrande, E., "The Development of a Factory Simulation Using Actual Operating
Data," Managemept fljeehnplegy, Vol.3 (1963), No. 1 (May), pp.l-19.

Levi, R., and Rossetto, S., "Machining Economics and Tool Life Variation: Part 1,
Basic Considerations and their Practical Implications," Journal of Engineering for
Industg, ASME, Vol.100 (1978), No.11 (November), pp. 393-396.

Levi, R., and Rossetto, S., "Machining Economics and Tool Life Variation: Part 2,

ApplicatiOn to Models for Machining Processes," Journal of Engineering for Industry,
ASME, Vol.100 (1978), No.11 (November), pp. 397-401.

Lewis, F.A., "Some Factors Affecting the Design of Production Systems in Batch

Manufacturing," Erpegding ef the 14th Machine Tool Design and Research
Cenferenee, 1973.

Lie, C.H., Hwang, C.L., and Tillman, F.A., "Availability of Maintenance Systems:
A State-of-the-Art Survey,” AIIE Transactions, Vol.9, 1977, pp.247-259.

Love, RR. and Vemuganti, R.R., "The Single-Plant Mold Allocation Problem with

Capacity and Changeover Restrictions," eratipns Research, Vol.26 (1978), No.1
(Jan-Fed). Pp. 159-165.

Lyman, S.B., "An Analysis of Finite T 001 Life and Scheduling Heuristics in a Dual

Resource Constraint Job Shop," Preefling of the Twenty-Feasth Annufl Mid-West
Decisien Seienees Institttte, April, 1993.

Macaulay, 8., "Amazing Things Can Happen If You Keep It Clean," Prfluetion,
Vol. 100 (1988), No.5 (May), pp. 72-74.

Maggard, B.N. and Rhyne, D.M., "Total Productive Maintenance: A Timely

Integration of Production and Maintenance,“ Jeamal pf Pmttetien at! Inventery
Mapagem, Vol.33 (1992), No.4 (Forth Quarter), pp. 6-10.

Maggard, M. J., Hogg, G. L, and Phillips, D. T., "The Efficiency and Economics of
Small Heterogenous Labor Limited Queuing Systems, " nternatipnal Jettm a ef
W, Vol. 18 (1980), No. 5 (September-October), pp. 619- 636.

 

266

Maggard, M.J., Lesso, W.G., Keating, R.J., and Wexler, M.J., "Network Analysis
with GERTS III OR," lndestrial Engineering, Vol.6 (1974), No.5 (May), pp.24-29.

Mahmoodi, F., Dooley, K.J., and Starr, P.J., ”An Investigation of Dynamic Group

Scheduling Heuristics," lntematipnal Jetlrnal pf Produetion Research, Vol.28 (1990),
No.9 (August).

Mahmoodi, F., and Dooley, K.J., "A Comparison of Exhaustive and Non-Exhaustive
Group Scheduling Heuristics in a Manufacturing Cell," International Journal of

W, Vol.29 (1991), No.9. PP. 1923-1939.

Mahmoodi, F., Tierney, L.J., and Mosier, C.T., "Dynamic Group Scheduling
Heuristics in a Flow Through Cell Environment," Decision Sciences, Vol.23 (1992),
No.1.

Malhotra, M.K., and Ritzman, L.P., "Resource Flexibility Issues in Multistage
Manufacturing," Deeisipn Seienees, Vol.21 (1990).

Martin, J.M., "Managing Tools Makes the Cell Click," Manufacturing Engineering,
April 1989, pp. 59-62. ,

Mason, F., ”Computerized Cutting Tool Management," American Machinist &

Atttpmated Manpfaetpring, Vol.130 (1986), No.5 (May), pp.105-l32.

Mason, F., "Getting Control Over Tools is 3 Trend," American Machinist, Vol. 135
(1991), No.5 (May), pp. 45-49.

McCall, J.J., "Maintenance Policies for Stochastically Failing Equipment: A Survey,"
, Vol.11 (1965), No.5 (March), pp. 493-524.

McCartney, J., and Hinds, B.K., ”Tooling Economics in Integrated Manufacturing

System," lpternatipnal qumal pf Prfluetion Research, Vol.20 (1982), No.4, pp. 493-
505.

Melnyk, S.A., Ghosh, S., and Ragatz, G.L., "Tooling Constraints and Shop Floor

Scheduling: A Simulation Study," Jeemal pf ngtions Management, Vol.8 (1989),
No.2, pp.69-89.

Melnyk, S.A., and Lyman, 8.8., "Analysis of Varying Labor Efficiency and Capability
in a Dual-Constraint Job Shop: A Simulation Experiment," Ereeflings pf the Twenty-
Seeend Midwest Deeisipn Seienee Institpm Meeting, May 1-3, 1991, pp. 291-296.

Melnyk, S.A., Denzler, DR, and Fredendall, L., "Variance Control vs. Dispatching
Efficiency,” redeetipn and Inventety Mgagement Journal, Vol.33, No.3, Third

 

 

Quarter, 1992, pp. 6—13.

Melnyk, S.A. and Ragatz, G.L., "Order Review/Release: Research Issues and

Perspectives," lntematipnal Journal ef Production Research, Vol.27, No.7 (1989),
pp.lO81-1096.

Melnyk, S.A., Ragatz, G.L., Fredendall, L., "Load Smoothing by the Planned and
Order Review/Release System,” Working Paper, Nov. 1990.

Mihram, G.A., "Blocking in Simulation Experimental Design," Journal of Statistical

Camputingandjimulatim Vol.3 (1974), 1313-29-32.

Miller, J .G., and Berry, W.L., "Heuristic Methods for Assigning Men to Machines:
An Experimental Analysis,” AIIE Transaetions, Vol.6 (1974), No.2 (June), pp.97-104.

Monahan, G.E., "A Survey of Partially Observable Markov Decision Processes:
Theory, Models, and Algorithms," Management Science, Vol.28 (1982), No.1 (Jan),
pp. 1-16.

Mosier, C .T., Elvers, D.A., and Kelly, D., "Analysis of Group Technology Scheduling
Heuristics," latematt'eng Jetlrnal pf Ermeetien Resgreh, Vol.22 (1984), No.5

Moulis, P. "Is Hidden rework Draining Company Profits," Quality, May 1992, pp. 15-
19.

Nagi, R., Harhalakis, G., and Proth, J.L., "Multiple Routings and Capacity
Considerations in Group Technology Applications," Internatipnal Journal of Prfluction
teem, Vol.28 (1990), No.12.

Nakagawa, T., "A Summary of Discrete Replacement Policies," Eurpm Jearng of
Qmpatjppalieseapclt, Vol.17 (1984), No.3, pp.382-392.

Nelson, B. L., "Variance Reduction for Simulation Practitioners," Preeeflings of the
1989 Winter Simulatt'ep Qenfereng. PP. 43-51.

 

Nelson, B. L., "Variance Reduction in the Presence of Initial-Condition Bias," L
Trangetiens, Vol. 22, No. 4, December 1990, pp. 340-350.

Nelson, R.T., "Labor Assignment as a Dynamic Control Problem," Qpetations
Ewen, Vol.14 (1966), No.3 (May-June), pp.369-375.

Nelson, R.T., "Labor and Machine Limited Production Systems," Mapagement
Seienee, Vol.12 (1967), No.6 (May), pp.648-671.

 

 

268

Nelson, R.T., ”Dual-Resource Constrained Series Service System," Oxrations
fiesegelt, Vol.16 (1968), No.2 (March-April), pp.324-341.

Nelson, R.T., "A Simulation of Labor Efficiency and Centralized Assignment in a

Production System,” Management Seienee, Vol.17 (1970), No.2 (October), pp.b97-
b106.

Neter, J., Wasserman, W, and Kutner, M.H., Appliegl Linea; Statistical Models, 3rd
edition, Irwin, Homewood, Illinois, 1985.

Newman, R.G., "MRP Where M = Preventative Maintenance," Production and
lnventery Management, Vol.26, No.2, Second Quarter 1985, pp. 21-28.

Panwalker, 8.8., and Iskander, W., "A Survey of Scheduling Rules," erations
Reseatep, Vol.25 (1977), No.1 (January-February), pp.45-61.

Park, P.S., and Bobrowski, P.M., ”Job Release and Labor Flexibility in a Dual

Resource Constrained Job Shop," qumal pf erations Management, Vol.8, No.3,
August 1989, pp 230-249.

Pate-Comell, M.E., Lee, H.L., and Tagaras, G., "Warning of Malfunction: The
Decision to Inspect and Maintain Production Processes on Schedule or on Demand,"

Management Seienee, Vol.33 (1987), No.10, pp. 1277-1290.
Pegden, C.D., Intrtfiuetipn te SIMAN, Systems Modeling Corporation, 1987.

Pegden, C. D., Shannon, R. E., and Sadowski, R. P., Introduetion tQ Simttlation Using
SIMAN, McGraw-Hill, Inc., New York, New York, 1990.

Pierskalla, W.P. and Voelker, J .A., "A Survey of Maintenance Models: The Control

and Surveillance of Deteriorating Systems," Naval Research Logistics Quarterly, Vol.23
(1976), No. 1 (March), pp.353-388.

Primrose, PL. and Leonard, R., "Reappraising Cutting Tool Economics within the

Bounds of Accountancy Theory," nternational Jettrnal of Prpguction Researeh, Vol.24
(1986), No.2, pp.269-278.

Oliva-Lopez, E. and Purcheck, G.F.K., "Load Balancing for Group Technology

Planning and Control," Intematleng thtmal pf Maehine T991 Design ang Resgrch,
Vol.19 (1979), No.4.

Owen, J .V.,"Moving to Modular", Mapttfaetttring Engingring, February, 1992, pp.
51-55.

269

Ozekici, 8., "Optimal Periodic Replacement of Multicomponent Reliability Systems,"
MW, Vol.36 (1088), No.4 (Jul-Aug), pp.542-552.

Ragatz, G.L., and Mabert, V.A., "An Evaluation of order Release Mechanisms in a
Job Shop Environment," Qeeisien Seienees, Vol. 19 (1988).

Ram, B, and Olumolade, M., "Preventive Maintenance Scheduling in the Presence of

a Production Plan," Brglttetion ape Inventoty Management, Vol.28 (1987), No.1,
pp.81-87.

Ramalingam, 8., "Tool Life Distributions: Part 2: Multiple-Injury Tool-Life Model,"

Jeamal ef Engineering fer Industg, ASME, Vol.99 (1977), No.9 (August), pp. 523-
528.

Ramalingam, S., Peng, Y.I., and Watson, J.D., "Tool Life Distributions: Part 3:
Mechanism of Single Injury Tool Failure and Tool Life Distribution in Interrupted
Cutting," rn n in rin f In A M , Vol.100(1978), No.5 (May),
pp. 193-200.

Ramalingam, S., Peng, Y.I., and Watson, J.D., ”Tool Life Distributions: Part 4:
Minor Phases in Work Material and Multiple-Injury Tool Failure," Journal of

Engingripg fet JnglestLy, ASME, Vol.100 (1978), No.5 (May), pp. 201-209.

Ravignani, G.L., Zompi, A., and Levi, R., "Multi-Tool Machining Analysis: Part 2,
Economic Evaluation in View of Tool Life Scatter," mal of En ineerin for
Industty, ASME, Vol.10] (1979), No. 5 (May), pp. 237-240.

Rochette, R., and Sadowski, R.P., ”A Statistical Comparison of the Performance of
Single Dispatching Rules for a Particular Set of Job Shop," ntematienal Journal pf
W, Vol.15 (1976), No.1, pp.63-75.

Rubin, R.A., Mosier, C.T., and Mahmoodi, P., "A Comprehensive Analysis of Group
Scheduling Heuristics in a Job Shop Cell," Dmisien Seienee, 1992.

Russell, G.R., and Philipoon, P.R., "Sequencing Rules and Due Date Setting
Procedures in a Flow-Line Cell Shop With Family Setups," Jeumal of glperations
Management, Vol.10 (1992), No.2.

Savoie, R.M., "How to Simplify Tool Requirements Planning in a High Volume,

Repetitive Shop," Ameg'eatt Ermttetie apgl Inventpgt gentrol Smiety Thit'tieth Annual
Wags. 1987. PP- 100-103-

Sawicki, J.D., "The Problems of Tardiness and Saturation in a Multi-Class Queue with
Sequence Dependent setups," W, Vol.4 (1973), No.4.

270

Schmeiser, B.W., "Batch Size Effect in the Analysis of Simulation Output,” Operation
32MB. Vol.30, No.3, 1982.

Schruben, L. W.,”Detecting Initialization Bias in Simulation Output," Operations
Emil. Vol. 30, No. 3, May-June 1982, pp. 569-590.

Schruben, L., Singh, H., and Tierney, L., "Optimal Tests for Initialization Bias in
Simulation Output,” W, Vol. 31, No. 6, 1983, pp. 1167-1178.

Shannon, R. E., "Tests for Verification and Validation of Computer Simulation

Models," Espeeedings uf Winter Simulatien glenferenee 1981, pp. 573-577.

Sharit, J ., and Salvendy, G., "A Real-Time Interactive Computer Model of a Flexible
Manufacturing System," IE Transactions, Vol.19 (1987), No.2, pp.167-177.

Sharit, J ., and Wlhence, S., "Computerization of Tool-Replacement Decision Making
in Flexible Manufacturing Systems: A Human-System Perspective," International

qurnal pf Erpgluetion Eeseareh, Vol.27(1989), No.12, pp. 2027-2039.

Sherif, Y.S. and Smith, M.L., "Optimal Maintenance Models for Systems Subject to

Failure - A Review," aval Resgrch Legistics Quarterly, Vol.28 (1981), No.1
(March), pp. 47-74.

Skinner, W.H., "The Focused Factory," Harvard Business Review, May-June, 1974.

Slack, N., "Flexibility as a Manufacturing Objective," International Journal of

Qperatiens ad Prgfluction Management, Vol.3 (1983), No.3.

Slack, N., "The Flexibility of Manufacturing Systems," ntemational Journal of
' ' n , Vol.7 (1987), No.4.

Stalk, G., "Time-The Next Source of Competitive Advantage," Harvagd Business

VI w,

Sutton, J.R., ”Total Preventive Maintenance," nglustrial Engineering, Vol.22 (1990),
No.10, pp.18-19.

Suresh, NO, and Meredith, J.R., "Achieving Factory Automation Through Group
Technology Principles" MW. Vol.5 (1985). No.2.

Swamidass, P.M., "Manufacturing Flexibility," thatiuns Mmagement Assggiation,
W. 1988.

Tagaras, G., "An Integrated Cost Model for the Joint Optimization of Process Control

271

and Maintenance," Journal of Operational Research Society, (UK), Vol.39 (1988),
No.8, pp- 757'766.

Tang, CS, and Denardo, E.V., "Models Arising from a Flexible Manufacturing
Machine, Part I: Minimization of the Number of Tool Switches," and "Part II:
Minimization of the Number of Switches Instants," ngatiens Resgreh, Vol.36
(1988), No.5 (September-October), pp.767-784.

Tarn, J.H., and Tomizuka, M., "On-Line Monitoring of Tool and Cutting Conditions

in Milling," qurng ef Engineering fer Industty, ASME, Vol.111 (1989), No.8
(August), pp. 206-212.

Taylor, F.W., "On the Art of Cutting Metals," ASME Transactions, Vol.28, 1907,
pp.310-350.

Treleven, M.D., "The Timing of Labor Transfers in Dual Resource-Constrained
Systems: "Push" vs. "Pull" Rules," Dgisipn Seience, Vol.18 (1987), No.1 (Winter),
pp.73-88.

Treleven, M.D., "A Review of the Dual Resource Constrained System Research," L
Transaetipns, Vol. 21, No.2, 1989.

Treleven, M.D., and Elvers, D.A., "An Investigation of Labor Assignment Rules in

a Dual-Constrained Job Shop,” Jeumal pf gmtatipns Management, Vol.6 (1985), No.1
(November), pp.51-68.

Vasilash, G.S., ”A new Age for Cutting Tools or Business as Usual?," Production,
October 1990, pp. 32-36.

Van Wassenhove, L.N., and Vanderhenst, P., "Planning Production in a Bottleneck

Department," Eurom Jeumal pf Qpeeatienal Research, Vol.12 (1982), No.2, pp.
127-137.

Ventura, J.A., Chen, F.F., and Chin-Hang Wu, "Grouping Parts and Tools in Flexible

Manufacturing Systems Production Planning," Intemational Juurnal pf Prgfluetien
mu, Vol.28 (1990), No.6, pp. 1057-1065.

Venderhenst, P. ,Van Steelandt, F. V. ,and Gelders, L. f., ”Efficiency Improvement of

a Transfer Line via Simulation, Jeum al et Qpergation onfl Rh,ese§e (UK), Vol. 32
(1981), No. 7 (July), pp. 555- 582.

Vinod, B., and Sabbagh, M., "Optimal Performance Analysis of Manufacturing
Systems Subject to Tool Availability," r m f rai n M ment,
Vol.24 (1986), pp. 398-409.

272

Vollmann, T.E., Berry, W.L., and Whybark, D.C., Manufacturing Planning and
Central Systems, Dow Jones-Irwin, Homewood, 111. 1988.

Wacker, J.G., "The Complementary Nature of Manufacturing Goals By Their
Relationship to Throughput Time: A Theory of Internal Variability of Production

System.” Wm. Vol.7 (1987). No.1-2 (Oct). tap-91406.

Wassweiler, W.R., "Tooling Requirements Planning," American Production and
nv nt n r i h Annual nf renc Proceedin , Chicago, IL, 1982,
PP.160-162.

Weeks, J.R., and Fryer, J.S., "A Simulation Study of Operating Policies in a
Hypothetical Dual Constrained Job Shop," Management Science, Vol.22 (1976), No.12
(August), pp. 1362-1371.

Weeks, J.R., and Fryer, J.S., "A Methodology for Assigning Minimum Cost Due-
Dates," Management Seienee, Vol.23 (1977), No.8 (April), pp.872-881.

Weeks, J.R., ”A Simulation Study of Predictable Due-Date," Management Science,
Vol.25 (1979), No.4 (April), pp.363-373.

Welch, P.D., "The Statistical Analysis of Simulation Results," Computer Performance
Mmeling flauupmk, Academic Press, Inc., 1983, pp. 267-329.

Wells, OE, and Bryant, J.L., "Optimal Preventive Maintenance Policies for Systems
with Missions of Random Duration," J E Transaetions, Vol.17,(1985), No.4 (Dec),
pp.338-345.

Wemmerlov, U., and Vakharia, A.J., ”Job & Family Scheduling of a Flow Line
Manufacturing Cell: A Simulation Study," IIE Ttansactions, Vol.23 (1991), No.4.

Wemmerlov, U., "Fundamental Insights Into Part Family Scheduling: The Single
Machine Case," Dmisipn Seienees, Vol.23 (1992), No.2.

White, C.H., and Wilson, R.C., "Sequence Dependent Setup Times and Job
Sequencing,” ntematlenal Jeurnal ef Prmuetion Research, Vol.15 (1977), No.3.

Wick, C., "Advances in Machining Centers," Mauufacturing Engineering, October
1987, pp. 24-32.

Wilbrecht, J.K., and Prescott, R.C., "The Influence of Setup Time on Job Sh0p
Performance," Mauagmieutjelertee, Vol. 16 (1969).

Wilkinson, Leland, SXSflfAT; The System fer Statistics fur the PC, SYSTAT Inc.,

273
1987.

Wilson, J. R., and Pritsker, A.A.B., "Evaluation of Startup Policies in Simulation
Experiments," Simulatiun, Sept. 1978, pp. 79-89.

Zavanella, L. Maccarini, GO, and Bugini, A., "FMS Tool Supply in a Stochastic
Environment: Strategies and Related Reliabilities," International Journal of Machine

1:991; r9; Mauufaeturing, Vol.30 (1990), No.3, pp.389-402.

Zompi, A., Levi, R., and Ravignani, G.L., "Multi-Tool Machining Analysis: Part 1,

Tool Failure Patterns and Implications," qurnal pf Engineering for Industty. ASME,
Vol.101 (1979), May, pp.230-236.