Yong Dissertation PDF

Computer-Aided Design for Rapid Tooling: Methods for Mold
Design and Design-for-Manufacture
A Thesis
Presented to
the Academic Faculty
By
Yong Chen
In Partial Fulfillment
of the Requirements for the Degree of
Doctor of Philosophy in Mechanical Engineering
Georgia Institute of Technology

August 2001
Copyright © 2001 by Yong Chen

Computer-Aided Design for Rapid Tooling: Methods for Mold
Design and Design-for-Manufacture
Approved:
David W. Rosen, Chair

Associate Professor, Mechanical
Engineering
Janet Allen
Senior Research Scientist, Mechanical
Engineering
Charles Eastman
Professor, Architecture and Computer
Science
Farrokh Mistree
Professor, Mechanical Engineering
John Muzzy
Professor, Chemical Engineering
Thomas Starr
Professor, Chemical Engineering
Date Approved
ii
ACKNOWLEDGMENTS
As I am putting the finishing touches on this document that has recorded my work
for the past three years, I still clearly remember the feeling when I first read the
dissertation of Dr. Stewart Coulter in April 1998. I was anxious and tried to understand
what was required for a Ph.D. in the United States. Although getting lost at that time, I
was shocked by the work and the writing skills shown in the dissertation. To write a
dissertation like that seemed only a dream so far away. Now I have completed my
doctoral research and finished my dissertation. My years here in Atlanta have been one of
the greatest personal growth in my life. These accomplishments, however, would not
have been possible without the help of several people listed as follows. I am grateful for
the tremendous roles they have played in my life.
First sincere appreciation is expressed to my advisor Dr. David W. Rosen for his
continuous encouragement, guidance, and patience throughout my graduate studies. He
has provided generous support and has been the primary motivation for this research. His
insights have guided me to a level of understanding higher than what I thought possible.
I would like to thank the other committee members for their comments and
suggestions. In particular, I owe special thanks to Dr. Farrokh Mistree and Dr. Janet
Allen, who opened the world of decision-based design and its usage in several areas for
me. Their continued guidance and inspiration have also helped me grow both
intellectually and personally. I also owe special thanks to Professor Charles Eastman,
who guided me in the field of solid modeling. The idea of the reverse glue operation was
first thought out in my taking a course given by Prof. Eastman. I am profoundly grateful
to Dr. John Muzzy and Dr. Thomas Starr for sharing their insights on polymer injection
molding and powder injection molding.
I feel very lucky to be able to work in two distinguished laboratories, System
Realization Laboratory (SRL) and Rapid Prototyping and Manufacturing Institute
(RPMI), at the same time in the past three years. Several colleagues in SRL gave me
iii
great support to this research. Special thanks is given to Shiva Sambu who worked with
me together on developing the design for Rapid Tooling system and performing the case
studies of a robot arm and camera roller (Chapter 6, 7, 8). I would like to thank Yao Lin
for helping me in understanding Design of Experiments and the usage of statistic
software. Thanks also to Brian Davis for teaching me how to use the Coordinate
Measuring Machine. I cherish the support and compassion given freely by other close
colleagues in SRL – Angran Xiao, Hongqing Wang, Zahed Siddique, Sunji Jangha,
Carolyn Conner Seepersad, Marco Fernandez, Benay Sager, Scott Cowan, Scott Duncan,
and Ruhul Kulkarni. Thanks for making my study in Georgia Tech enjoyable.
I would like to thank my colleagues in RPMI for their generous help in this
research: to Reggie Ponder, for teaching me how to use SLA machines; to Giorgos
Hatzilias, for teaching me how to use Morgan Press Injection molding machine; to Giang
Pham, for teaching me how to use Sumitomo Injection molding machine; to Kent
Dawson, for sharing his experiment data and his knowledge on the properties of injection
molded parts; to Young-Bin Park, for teaching me how to use Instron universal testing
machine.
I am also a student member in Graphics, Visualization and Usability (GVU)
Center at Georgia Tech. I am very grateful to the two courses given by Dr. Jarek
Rossignac, who helped me to understand the beauty of computational geometry. The
region generation algorithm in this research was first developed as a course project in one
of these courses. The geometric modeling course given by prof. Eastman inspired me to
further develop some ideas in the field of automatic mold design. Special thanks goes to
Guoquan Zhou, a graduate student of Prof. Eastman, for sharing his code of a simple
solid modeling system.
The financial support from the National Science Foundation (NSF DMI- 96-
18039) is greatly appreciated. It makes this work possible.
Three and a half years of graduate study was a long way to go. I consider myself
blessed to know my wife, Mrs. Ying Wu. Her love and support make these years really
enjoyable. I offer deepest thanks to her for her understanding and sacrifice throughout
iv
this process. I am also in debt to my parents and parents-in-law for what they have done
for me. Without their continued support and belief in me, I would not have gotten to this
point. This dissertation stands as a testament to their success. I am also thankful to Sai
Zeng, Lei Wu, Cheng Zhang, Lunyu Ma, Qi Zhu, Yong Huang, and Liang Zhu, who have
given me the great gift of friendship.
v
TABLE OF CONTENTS
ACKNOWLEDGMENTS iii
TABLE OF CONTENTS vi
LIST OF TABLES xi
LIST OF FIGURES xiv
NOMENCLATURE xix
SUMMARY xxi
CHAPTER 1
FUNCTIONAL PROTOTYPE USING RAPID TOOLING 1
1.1 PRIORS – FUNCTIONAL PROTOTYPE AND RAPID TOOLING 2
1.1.1 Product Realization and Prototype 2
1.1.2 Rapid Prototyping 3
1.1.3 Injection Molding and Rapid Tooling 6
1.1.4 Rapid Manufacturing 8
1.2 DESIGN FOR RAPID TOOLING AND RESEARCH OPPORTUNITIES 9
1.2.1 Current Usage of Rapid Tooling and Related Problems 9
1.2.2 Motivating project – Rapid Tooling TestBed 11
1.2.3 Challenges of RTTB and My Research Approaches 12
1.2.4 Research Opportunities in Mold Design and Design-for-Manufacture 16
1.3 RESEARCH FOCUS IN THE DISSERTATION 23
1.3.1 The Principal Goal, Research Questions and Hypotheses in the Dissertation 24
1.3.2 Validation Philosophy and Strategy 26
1.3.3 Contributions from the Research 29
1.4 OVERVIEW OF DISSERTATION 30
CHAPTER 2 33
A LITERATURE REVIEW: MOLD DESIGN AND DESIGN-FOR-MANUFACTURE
2.1 TOPICS IN CHAPTER 34
2.2 MOLD DESIGN METHODS AND ALGORITHMS 35
2.2.1 Parting Direction 35
2.2.2 Parting Line 37
2.2.3 Parting Surface 38
2.2.4 External and Internal Undercut Detection 39
2.2.5 Synthesis Approaches of Basic Elements 39
2.2.6 Intersection of V-Map 41
2.2.7 Detection of Non-Drafted Surfaces 42
vi
2.3 MOLD CONSTRUCTION METHODS AND TOOLS 43
2.3.1 Approach Based on Extending Parting Lines 43
2.3.2 Approach Based on Sweeping 43
2.3.3 Industrial Approach 44
2.4 CAD REPRESENTATION 46
2.4.1 Representation of 3D Surfaces and Solids 49
2.4.2 Manipulation of Solid Models 51
2.4.3 High Level Representations 53
2.5 DESIGN FOR MANUFACTURE: STRATEGIES AND TECHNIQUES 56
2.6 DESIGN TECHNOLOGIES 62
2.6.1 Decision-Based Design 62
2.6.2 Compromise Decision Support Problem 62
2.6.3 Robust Concept Exploration Method (RCEM) 69
2.7 LITERATURE REVIEW SUMMARY 71
CHAPTER 3
THE MULTI-PIECE MOLD DESIGN METHOD 72
3.1 OVERVIEW OF THE MULTI-PIECE MOLD DESIGN METHOD 73
3.2 PROBLEM FORMULATION FOR MULTI-PIECE MOLD DESIGN 75
3.2.1 Analysis of Existing Problem Formulations 75
3.2.2 Problem Formulations of Multi-Piece Mold Design 78
3.3 OVERVIEW OF THE MULTI-PIECE MOLD DESIGN PROCESS 79
3.4 BASIC ELEMENTS OF MPMDM 80
3.4.1 Demoldability of Mold Pieces 81
3.4.2 Analysis of the Basic Elements 87
3.4.3 The Generation Approach of Concave Regions 89
3.5 REGION COMBINATION OF MPMDM 94
3.5.1 Combining Criteria and Their Evaluation Approach 94
3.5.2 Verification of Draft Angle 98
3.5.3 Analysis of Region Combination Process and Related Representations 99
3.5.4 Region Combination Algorithm and Design Knowledge 103
3.6 MOLD PIECE CONSTRUCTION APPROACH BASED ON REVERSE GLUE 107
3.6.1 Principle and Related Representations of the Approach 108
3.6.2 Generation Approach of Glue Faces 111
3.6.3 Reverse Glue Algorithm and Parting Surface 115
3.7 SUMMARY OF CHAPTER 3 118
CHAPTER 4
RTMDS AND ITS USAGE FOR MOLD DESIGN 122
4.1 OVERIEW OF RAPID TOOLING MOLD DESIGN SYSTEM 123
4.2 SUPPORTING MODULES AND THEIR IMPLEMENTATIONS 124
4.2.1 ACIS Manipulation Module and Part Representation 124
4.2.2 Problem PDLP and LINGO® 128
4.2.3 GUI and Control Options 128
vii
4.3 THE IMPLEMENTATION AND LIMITATIONS OF MOLD DESIGN
MODULES 129
4.3.1 Region Generation 130
4.3.2 Region Combination 132
4.3.3 Mold Piece Construction 135
4.4 INTUITIVE EXAMPLE PARTS 136
4.4.1 Test Example 1: A Box with a Rib 137
4.4.2 Test Example 2: A Box with a Through Hole and Two Grooves 141
4.5 INDUSTRIAL CASES 145
4.5.1 Industrial Example 1: A Housing 146
4.5.2 Industrial Example 2: A Thin Wall Part 150
4.5.3 Industrial Example 3: A Complex Housing 152
4.6 EVALUATION OF EXAMPLES AND CASES 155
4.6.1 Region Generation Process 155
4.6.2 Region Combination Process 156
4.6.3 Mold Piece Construction Process 157
4.6.4 The Whole Process of RTMDS 158
CHAPTER 5
FORMULATING DESIGN REQUIREMENTS FOR RAPID TOOLING AS
GEOMETRIC TAILORING PROBLEM 162
5.1 PROPERTIES OF RAPID TOOLING 163
5.1.1 Mold Material Properties 163
5.1.2 Mold Fabrication Properties 165
5.1.3 Part Properties of the AIM Tooling 166
5.2 PRINCIPLES OF FUNCTIONAL TESTING AND GEOMETRIC TAILORING 167
5.2.1 Principle of Functional Testing – Buckingham Π Theorem 167
5.2.2 Similarity Methods 168
5.2.3 Fundamentals of Geometric Tailoring 169
5.3 DESIGN DECISION TEMPLATE FOR MGT 171
5.3.1 MGT Decision Template and its Methodology 174
5.3.2 Formulation of the MGTDT 176
5.4 USAGE OF MGTDT 178
5.4.1 Formulating Functional Properties in the MGTDT 178
5.4.2 Formulating and Solving the MGT Problem 181
5.5 INITIAL CASE STUDIES 181
5.5.1 Building Prototypes of a Tensile Bar 182
5.5.2 Building Prototypes of a Rib Part 188
5.5.3 Building Prototypes of a Ring Gear 198
CHAPTER 6
A DECISION-BASED DESIGN FOR RAPID TOOLING SYSTEM 212
viii
6.1 OVERVIEW OF DESIGN FOR RAPID TOOLING SYSTEM 213
6.2 PROCESSING PLANNING OF THE AIM TOOLING 216
6.2.1 SLA Process Planner 216
6.2.2 Stereolithography Mold Life Predictor 218
6.2.3 Rapid Tooling Cost Estimator 221
6.3 MPGT DECISION TEMPLATE AND MPGT PROBLEM FORMULATION 222
6.3.1 MPGTDT 223
6.3.2 MPGT Problem Formulation 226
6.4 SOLVING THE MPGT PROBLEM 229
6.4.1 Solution Strategy 229
6.4.2 Solution Process and Implementations 233
6.5 COMPARISON OF THE CURRENT USAGE AND DFRTS 238
CHAPTER 7
FUNCTIONAL PROTOTYPES OF A ROBOT ARM 246
7.1 A ROBOT ARM DESIGN – PROBLEM DESCRIPTION 247
7.2 MOLD DESIGN WITH AID OF RTMDS 248
7.3 GEOMETRIC TAILORING WITH AID OF DFRTS – MODELING 253
7.3.1 MPGT Decision Template for the Robot Arm 253
7.3.2 Modeling Design Functions 256
7.3.3 Modeling Fabrication Processes 261
7.4 GEOMETRIC TAILORING WITH AID OF DFRTS – SOLVING 265
7.4.1 Solving Discrete Variables 265
7.4.2 Solving Other Variables 269
7.4.3 Selecting A Solution 274
7.4.4 Post-Solution Analysis 274
7.5 PHYSICAL VALIDATION 278
7.5.1 Build Time Validation 278
7.5.2 Accuracy Validation 279
7.5.3 Surface Finish Validation 280
7.5.4 Material Property Validation 281
7.5.5 Geometry and Weight Validation 281
7.5.6 Mold Life Validation 283
7.5.7 Summary of Physical Validation 284
7.6 EVALUATION OF ROBOT ARM CASE – POST DFRTS 285
CHAPTER 8
FUNCTIONAL PROTOTYPES OF A CAMERA ROLLER 290
8.1 A CAMERA ROLLER DESIGN – PROBLEM DESCRIPTION 291
8.2 MOLD DESIGN WITH AID OF RTMDS 293
8.3 GEOMETRIC TAILORING WITH AID OF DFRTS – PROBLEM ANALYSIS 299
ix
8.4 GEOMETRIC TAILORING WITH AID OF DFRTS – MODELING 303
8.4.1 Modeling Design Functions 303
8.4.2 Modeling Fabrication Processes 311
8.5 GEOMETRIC TAILORING WITH AID OF DFRTS – SOLVING 317
8.5.1 Solving Discrete Variables 317
8.5.2 Solving Continuous Variables 321
8.5.3 Selecting A Solution 323
8.5.4 Post-Solution Analysis 323
8.6 PHYSICAL VALIDATION 326
8.7 EVALUATION OF CAMERA ROLLER CASE – POST DFRTS 329
CHAPTER 9
ACHIEVEMENTS AND RECOMMENDATIONS 334
9.1 ANSWERING THE RESEARCH QUESTIONS 335
9.1.1 Research Question Overview 335
9.1.2 Answering Research Questions 337
9.2 ACHIEVEMENTS: REVIEW OF RESEARCH CONTRIBUTIONS 340
9.3 CRITICAL ANALYSIS: LIMITATIONS OF THE RESEARCH 343
9.4 FUTURE WORK 345
APPENDIX A
RTMDS IMPLEMENTATION 348
A.1 CODE FILES OF RTMDS 349
A.2 CLASSES OF RTMDS 351
A.3 IMPLEMENTATION ON ACIS 352
A.4 IMPLEMENTATION ON LINGO 354
APPENDIX B
CAMERA ROLLER CASE STUDY: COMPLETE SOLUTIONS OF THE
MODIFIED MPGT PROBLMES 356
B.1 SOLUTIONS OBTAINED FROM OPTDESX FOR SLICING SCHEMES
1~6 OF PO1 357
1~6 OF PO2 358
1~6 OF PO3 359
1~6 OF PO4 360
REFERENCES 361
VITA 375
x
LIST OF TABLES
Table 1.1 – Relationship Between Hypotheses and Dissertation Sections. 26

Table 1.2 - Validation of Hypotheses 1 and 2 addressed in different chapters. 28
Table 4.1 – The Complexities of the Example Parts in Section 4.4 and 4.5. 136
Table 4.2 – The Information for Test Example 1. 138
Table 4.3 – The Information for Test Example 2. 141
Table 4.4 – The Information for Industrial Example 1. 146
Table 4.7 – Experimental Data of Region Generation Process. 155
Table 4.8 – Experimental Data of Region Combination Process. 156
Table 4.9 – Experimental Data of Mold Piece Construction Process. 157
Table 4.10 – Experimental Data of Mold Design Process. 159
Table 5.1– Tensile Properties Comparison for Atactic Polystyrene. 166
Table 5.2– Flexural Properties Comparison for Atactic Polystyrene. 166
Table 5.3 – Word Formulation of the MGTDT. 176
Table 5.4 – Mathematical Formulation of the MGTDT. 177
Table 5.5 – Experimental Plan For Testing the MGT. 182
Table 5.6 – MGT Tensile Bar Problem Formulation. 184
Table 5.7– Scenarios of Goal Weights and Related Results. 185
Table 5.8– Material property validation results for polystyrene. 187
Table 5.9 - MGT Rib Problem Formulation By the Designer. 190
Table 5.10 – Design Factors and Their Ranges. 192
Table 5.11 – Results of the Experiments for Rib Part. 192
Table 5.12 – Complete MGT Rib Problem Formulation. 195
Table 5.13– Scenarios of Goal Weights. 196
Table 5.14– Rib Part MGT Results. 196
Table 5.15 – Maximum Torque of The Speed Reducer. 199
Table 5.16 - MGT Ring Gear Problem Formulation By the Designer. 201
Table 5.17 – Design Factors and Their Ranges of Ring Gear. 203
Table 5.18 – Results of the Experiments for Ring Gear. 203
Table 5.19 – Complete MGT Ring Gear Problem Formulation. 205
Table 5.20– Scenarios of Goal Weights. 206
Table 5.21– Ring Gear MGT Results. 207
Table 6.1 – SLA Process Planning Word Formulation [Sambu, 2001 #967]. 216
Table 6.2 – SLA Process Planning Mathematical Formulation. 217
Table 6.3 – IJM Process Planning Word Formulation [Sambu, 2001 #967]. 218
Table 6.4 – Factors Selected for Mold Life Prediction [Sambu, 2001 #967]. 219
Table 6.5 – IJM Process Planning Mathematical Formulation. 220
xi
Table 6.6 – Cost and Time Estimates for Different Steps in Rapid Tooling Process
(Sambu, 2001). 221
Table 6.7– Word Formulation of the MPGTDT. 223
Table 6.8 - Mathematical Formulation of the MPGTDT. 224
Table 6.9 – MPGT Problem Word Formulation. 226
Table 6.10 – MPGT Problem Mathematical Formulation. 227
Table 7.1 – The Information for Robot Arm. 248
Table 7.2 – MPGT Robot Arm Problem Formulation by the Designer. 253
Table 7.3 – Design Factors and Their Ranges of Robot Arm. 256
Table 7.4 - List of Experiments for Response Surface Generation. 256
Table 7.5 – Results of the Experiments for Robot Arm. 258
Table 7.6 - Results of Validation Experiments for the Robot Arm. 260
Table 7.7 – MPGT Robot Arm Problem Formulation. 262
Table 7.8 – Modified MPGT Robot Arm Problem Formulation. 269
Table 7.9 - Starting Points Investigated for Each Slicing Scheme. 272
Table 7.10 – Solutions of System Variables for Different Slicing Schemes. 272
Table 7.11 – Solutions of Goals for Different Slicing Schemes. 273
Table 7.12 – Solutions of the MPGT Problem for the Robot Arm. 274
Table 7.13 - Target Modification Experiments. 276
Table 7.14 – Results of Target Modification Experiments. 277
Table 7.15 - Individual Functional Target Modification Experiments. 277
Table 7.16 – Results of Individual Functional Target Modification Experiments. 277
Table 7.17 - SLA Build Time for the Mold Pieces of Robot Arms. 278
Table 7.18 – Accuracy of the Robot Arms. 279
Table 7.19 - Surface Finish Experiment Results on Prototype Parts. 280
Table 7.20 - Material Property Validation Results for Polystyrene. 281
Table 7.21 - Injection-molded part dimensions and weight. 282
Table 7.22 - Injection Molding Parameters Used for the Robot Arm. 283
Table 7.23 - Summary of Physical Validation Results for MPGT of Robot Arm. 285
Table 7.24 - Solutions of Sequential and Concurrent Solution Process. 286
Table 7.25 - Goals Achievements for Sequential and Concurrent Solution
Processes. 286
Table 8.1 – The Information for Robot Arm. 294
Table 8.2 – Estimated Mold Life and Required Number of Each Mold Piece. 300
Table 8.3 – Design Factors and Their Ranges of Camera Roller. 304
Table 8.4 - List of Experiments for Response Surface Generation. 304
Table 8.5 - Experiments to Study the Effect of Mesh Size. 305
Table 8.6 – Results of the Z-Rotation Experiments for Camera Roller. 306
Table 8.7 - Results of Validation Experiments of Z-rotation. 308
Table 8.8 – Results of the Volume Experiments for Camera Roller. 309
Table 8.9 - Results of Validation Experiments of Volume. 311
Table 8.10 – MPGT Camera Roller Problem Formulation. 314
Table 8.11 - Dimensions of the Different Mold Features. 317
Table 8.12 - Surface Finish and Build Time Achievements for Slicing Schemes. 320
xii
Table 8.13 - Starting Points Investigated for Each Slicing Scheme. 321
Table 8.14 - Solutions Obtained for Slicing Scheme 1 for Different Starting
Points. 322
Table 8.15 - Objective Function Values Obtained for Different Slicing Schemes. 323
Table 8.16 – Solutions of the MPGT Problem for Camera Roller. 323
Table 8.17 - Values of Goals and Intermediate Responses for the Solution. 323
Table 8.18 - Target Modification Experiments. 325
Table 8.19 – Results of Target Modification Experiments. 325
Table 8.20 – Percent Change of the Goals for Target Modification Experiments. 325
Table 8.21 - Values of System Variables for Target Modification Experiments. 326
Table 8.22 - Injection Molding Parameters Used for the Camera Roller. 328
Table 8.23 - Solutions of Sequential and Concurrent Solution Process. 330
Table 8.24 - Goals Achievements for Sequential and Concurrent Solution
Processes. 330
xiii
LIST OF FIGURES
Figure 1.1 - The Cost of an Design Change Related to Development Phase 2

Figure 1.2 - A Schematic Drawing of a SLA 4
Figure 1.3 - A CAD Model and a SL Model of a Robot Arm 5
Figure 1.4 - A Stereolithography Model for Assembly Studies and Tests with
Regard to Collision Behavior 5
Figure 1.5 – Direct AIM Tooling Process 7
Figure 1.6 – Processes of Using Direct AIM Tooling to Build Prototypes 9
Figure 1.7 – RTTB System and My Focus 12
Figure 1.8 – Different Geometries of Some Industrial Parts 13
Figure 1.9 – Design for Rapid Tooling 15
Figure 1.10 – An Illustrative Mold for Injection Molding 16
Figure 1.11 – Mold Design Process and My Considerations 17
Figure 1.12 – Standardized Master Frame Used in Morgan Press at RPMI 18
Figure 1.13 – A Multi-piece Molds for a Part Given by Protoform GmbH 19
Figure 1.14 – Relations of DFRT, DFM and Geometric Tailoring 23
Figure 1.15 -Validation Square (Pedersen, 1999) 27
Figure 1.16 - Overview of Dissertation Chapters 30
Figure 1.17 - Pictorial Overview of the Dissertation 32
Figure 2.1 - Literature Review Topics and Relationship with Hypotheses. 34
Figure 2.2 – Two Examples of V-map. 41
Figure 2.3 – A Part with Mold Pieces Generated by Two Approaches. 43
Figure 2.4 – Mold Construction Process of Unigraphics/MoldWizard for an
Industrial Part. 45
Figure 2.5 – An Example Mold Design Given by Magics RP. 45
Figure 2.6 – An Example Mold Design Given by IMOLD. 46
Figure 2.7 – An Example Mold Design Given by Moldplus. 46
Figure 2.8 – An Example Mold Design Given by QuickSplit. 47
Figure 2.9 – An Example Mold Design Given by I-DEAS. 47
Figure 2.10 – Coedges of a Part. 50
Figure 2.11 – The Distinction between Geometric and Topological Information
within the ACIS Solid Modeler (Spatial Technology, 2000). 51
Figure 2.12 – An Example of Gluing Operation. 52
Figure 2.13 – Boolean Operations. 53
Figure 2.14 – An Example of Behavioral Modeling (PTC). 55
Figure 2.15 - Word formulation of a Compromise DSP problem (Mistree,
et al., 1993). 63
Figure 2.16 - Mathematical formulation of a Compromise DSP problem
(Mistree, et al., 1993). 64
xiv
Figure 2.17 - The Compromise DSP with Modifications based on the Linear
Physical Programming Model (Hernandez and Mistree, 2001). 66
Figure 2.18 - Class Function Regions for a Generic ith Objective (Hernandez
and Mistree, 2001). 68
Figure 2.19 - The Robust Concept Exploration Method (RCEM) (Chen, 1995). 69
Figure 2.20 – Summary of Chapter 2 and Preview of Chapter 3. 71
Figure 3.1 – The Elements of the Multi-Piece Mold Design Method. 73
Figure 3.2 – Relationship between the Elements of MPMDM and Dissertation
Hypotheses. 74
Figure 3.3 – An Approach Based on V-map. 76
Figure 3.4 – Examples for the Problem Formulation. 76
Figure 3.5 – Mold Pieces for a Pocket with Empty V-Map. 77
Figure 3.6 – Steps of the Mold Design Process. 80
Figure 3.7 – Demoldability of Mold Pieces. 81
Figure 3.8 – Demoldability of Neighboring Mold Pieces. 83
Figure 3.9 – Dividing of Concave Regions. 85
Figure 3.10 – Splitting of a Combined Region. 86
Figure 3.11 – Concave Regions and Combined Regions. 87
Figure 3.12 – Combined Region Example. 88
Figure 3.13 – Edge Classification. 89
Figure 3.14 – Different Splitting Faces and Orders. 91
Figure 3.15 – PD Evaluation Approach. 96
Figure 3.16 – A Region with Different Mesh Size. 98
Figure 3.17 – Different Draft Angles. 98
Figure 3.18 – Relationships Between Regions and Minimum Draft Angle. 99
Figure 3.19 – Edges and Faces of a Region in a CXF Combining Step. 100
Figure 3.20 – Edges and Faces of Two CRs in a Combining Step. 102
Figure 3.21 – Different Combining Results of a Part. 104
Figure 3.22– Face Dividing for A Region of Cavity. 105
Figure 3.23 – Two Example Parts for Combining Order. 107
Figure 3.24 – Regions and Mold Pieces of a Part. 107
Figure 3.25 – An Example of a Boolean Mold Base and Mold Pieces. 109
Figure 3.26 – Glue Faces of an Example Part. 111
Figure 3.27 – Generation of Inner Glue Faces. 113
Figure 3.28 – Process of Face Generation. 114
Figure 3.29 – An Example for Coedge Splitting. 116
Figure 3.30 – An Example of Different Parting Surface. 118
Figure 3.31 – Partial Theoretical Validation. 119
Figure 3.32 - Summary of Chapter 3 and Preview of Chapter 4 121
Figure 4.1 – The Organization and Data Flow of the RTMDS. 123
Figure 4.2 – The Process to Generate an Input File for RTMDS. 125
Figure 4.3 – Focus of the ACIS Entity Classes in RTMDS. 126
Figure 4.4 – The Relationship of Topology Entity Classes in ACIS. 127
Figure 4.5 – Screen Capture of the RTMDS Interface. 128
xv
Figure 4.6 – Setting Tolerance Values in RTMDS. 129
Figure 4.7 – A Simple Example of Face Splitting. 131
Figure 4.8 – Two Examples of Face Splitting. 131
Figure 4.9 – A Complicated Example of Face Splitting. 132
Figure 4.10 – The Combining Order of Faces in a Same Plane. 133
Figure 4.11 – An Example for Setting Tolerance Value. 134
Figure 4.12 – Mold Pieces for a Pocket with Empty V-Map. 136
Figure 4.13 – A Test Example of a Rib Part. 138
Figure 4.14 – Graphical Results of a Mold Design for Test Example 1. 139
Figure 4.15 – Graphical Results of Another Mold Design for Test Example 1. 140
Figure 4.16 – A Test Example of a Box with a Through Hole and Two Grooves. 141
Figure 4.17 – Illustration of the Running Results of Test Example 2. 142
Figure 4.18 – Graphical Results of Mold Design for Test Example 2. (Step 1~7) 143
Figure 4.19 – Graphical Results of Mold Design for Test Example 2 (Step 8). 144
Figure 4.20 – A Housing for A Phone Adapter. 146
Figure 4.21 – Graphical Results of a Mold Design for Industrial Example 1. 148
Figure 4.22 – Physical Validation of the Mold Design for Industrial Example 1. 149
Figure 4.23 – A Thin Wall Part for A Phone Adapter. 150
Figure 4.24 – Graphical Results of a Mold Design for Industrial Example 2. 151
Figure 4.25 – Generated Mold Pieces for Industrial Example 2. 152
Figure 4.26 – A Complex Housing. 153
Figure 4.27 – Region Combination (2 regions). 153
Figure 4.28 – Mold Pieces for Industrial Part 3. 154
Figure 4.29 – Relations of Region Generation Time with Face/Edge Number. 156
Figure 4.30 – Relations of Region Combination Time with Face/Edge Number. 157
Figure 4.31 – Relations of Mold Piece Construction Time with Face/Edge
Number. 158
Figure 4.32 – Relations of Mold Design Running Time with Face/Edge Number. 158
Figure 4.33 – Empirical Structural and Performance Validation for Hypothesis 1. 160
Figure 5.1– Maximum Tensile and Shear Strength of SL5170 Versus Temp. 164
Figure 5.2 – Stair Stepping Effect of Layer Manufacturing and Reason. 165
Figure 5.3 – Fundamental Terms and Concept of the ESM [Cho, 1999 #964]. 169
Figure 5.4 – Steps of a Scenario based on the MGT Decision Template. 172
Figure 5.5 – The Decision Template for the MGT and Design Freedom. 174
Figure 5.6 - A Cantilever Beam. 179
Figure 5.7 – The Illustration of a Tensile Bar Example. 183
Figure 5.8 – Photo of SLA Tools and Prototype Tensile Bars. 186
Figure 5.9 – Photo of Prototype Tensile Bars Used for Tensile test. 187
Figure 5.10 – Quality Distribution and Geometric Tailoring. 188
Figure 5.11 – The Illustration of a Rib Example. 189
Figure 5.12 – Geometry Variables of the Rib by Adding Draft Angle. 191
Figure 5.13 – Two Screen Dumps of Experiment Results for Rib Part. 193
Figure 5.14 – Graphical Relations of The Maximum Deflection and Variables. 194
xvi
Figure 5.15 – Parts Produced by SLA Molds. 197
Figure 5.16 – SLA Molds and Pullout Failure. 198
Figure 5.17 – The Ring Gear and the Speed Reducer of a Cordless Drill. 198
Figure 5.18 – Some Terminology Describing a Spur Gear (Shigley and Mischke,
1989). 199
Figure 5.19 – Fatigue of Ring Gear. 200
Figure 5.20 – Two Screen Dumps of Experiment Results for Ring Gear. 204
Figure 5.21 – Graphical Relation of the Maximum Stress with PD and W. 205
Figure 5.22 – A Photo of the Prototype Ring Gears. 208
Figure 5.23 – A SLA Mold Piece in a Standard Mold Plate. 208
Figure 5.24– Summary of Hypothesis Validation. 210
Figure 6.1 – Relations of DFRT, Geometric Tailoring and DFM. 213
Figure 6.2 – Infrastructure of the DFRTS and Related Sections. 214
Figure 6.3 – Research Scope of the DFRTS. 215
Figure 6.4 – Screen Capture of the SLA Process Planner. 218
Figure 6.5 – Screen Capture of SL Mold Life Predictor. 221
Figure 6.6 – Relations of Designer’s Requirements and MGTDT/MPGTDT. 222
Figure 6.7 – Relations of the MPGTDT and MPGT Problem. 225
Figure 6.8 – Possible Values of PO and LTi for a Example Part. 230
Figure 6.9 – The Solution Strategy for the MPGT Problem. 231
Figure 6.10 – Two Sub-problems of the MPGT. 232
Figure 6.11 – The Solution Process of the DFRTS. 234
Figure 6.12 – Surface Finish Test Piece. 235
Figure 6.13 – The Current Usage of RT and Related Decision Order of Variables. 239
Figure 6.14 – The DFRTS and Related Decision Order of Variables. 240
Figure 6.15 – A Two Variable Design Space [Karandikar, 1991 #269]. 241
Figure 7.1 – A Robot Arm Design. 247
Figure 7.2 – Generated Regions for the Robot Arm. 249
Figure 7.3 – Graphical Results of a Mold Design for the Robot Arm. 250
Figure 7.4 – Graphical Results of Another Mold Design for the Robot Arm. 251
Figure 7.5 – A Completer Mold Design for the Robot Arm. 252
Figure 7.6 – Physical Validation of the Mold Design for the Robot Arm. 252
Figure 7.7 – Process-related Goals of Robot Arm Design. 253
Figure 7.8 – Example ANSYS Output of Analysis for the Robot Arm. 257
Figure 7.9 – Graphical Relations of Responses and Variables. 259
Figure 7.10 – Validation Experiment Design for the Response Surfaces. 260
Figure 7.11 – Some Process Planning Goals. 262
Figure 7.12 – Two Considered Mold Designs for the Robot Arm. 266
Figure 7.13 – Four Considered Part Orientations for the Mold Designs. 267
Figure 7.14 – Eight Considered Slicing Schemes for the Part Orientations. 268
Figure 7.15 – Inter-Relationship Diagram of Goals and Variables. 275
Figure 7.16– Objective Function vs. Iteration No. for Modified MPGT Problem. 276
xvii
Figure 7.17 – Mold Piece Layout on SLA 3500 Platform [Sambu, 2001 #967]. 279
Figure 7.18 – Pieces of Robot Arm Used in Surface Finish Experiments. 280
Figure 7.19 - Gap Between the Parting Surfaces due to Mold Warpage. 282
Figure 7.20 - Short (left) and Complete (right) Shots obtained from IJM. 284
Figure 7.21 - Chipped Mold and Parts Obtained from the Mold. 284
Figure 7.22 – The Ranges of Goals for the Robot Arm Problem. 287
Figure 7.23 – Empirical Structural and Performance Validation for H1 and H2. 288
Figure 7.24 – Preview of Chapter 8. 289
Figure 8.1 – A Production Camera Roller with a Camera. 291
Figure 8.2 - Loading Conditions on the Camera Roller. 292
Figure 8.3 - Surface Finish and Tolerance Requirements. 292
Figure 8.4 – Dividing Cylindrical Faces for the RTMDS. 293
Figure 8.5 – Graphical Results of Region Combination Process. 295
Figure 8.6 – Changing Region Number of Selected Faces in the RTMDS. 296
Figure 8.7 – Graphical Results of a Mold Design for the Camera Roller –Phase 1. 297
Figure 8.8 – Graphical Results of a Mold Design for the Camera Roller – Phase 2. 298
Figure 8.9 – A Completer Mold Design for the Camera Roller. 299
Figure 8.10 – Physical Validation of the Mold Design for the Camera Roller. 299
Figure 8.11 - Mold Life of Different Features in Mold Pieces. 300
Figure 8.12 – Problem Identified in the Preliminary Experiment. 301
Figure 8.13 - Different Features in Camera Roller. 302
Figure 8.14 - Modified Camera Roller Part. 302
Figure 8.15 - Geometry Variables in the Modified Camera Roller. 303
Figure 8.16 - ANSYS Output of Z-Rotation for Camera Roller (Exp.1: NC=
NR = 3). 305
Figure 8.17 – Graphical Relations of z-rotation and Variables. 308
Figure 8.18 – Graphical Relations of Volume and Variables. 310
Figure 8.19 – Process Planning Goals for the Mold Pieces. 313
Figure 8.20 – Four Part Orientations for the Mold Design [Sambu, 2001 #967]. 319
Figure 8.21 - Promising Slicing Schemes for PO1 [Sambu, 2001 #967]. 320
Figure 8.22 - Objective Function vs. Iteration Number for MPGT Problem. 322
Figure 8.23 – Mold Design for the Tailored Camera Roller Part. 327
Figure 8.24 – Physical Mold Pieces and Injection Molded Camera Roller. 328
Figure 8.25 – Comparison the Camera Roller before and after Geometric
Tailoring. 329
Figure 8.26 – The Dimensions of the Third Mold Piece. 330
Figure 8.27 – Empirical Structural and Performance Validation for H1 and H2. 332
Figure 8.28 – Preview of Chapter 9. 333
Figure 9.1 – Context for Research Questions and Hypotheses. 335
Figure 9.2 – Validity Square. 337
Figure 9.3 – An Ideal System Working Between the Designer and Manufacturer. 347
xviii
NOMENCLATURE
ACES Accurate Clear Epoxy Solid

ACIS Three-dimensional Geometric Modeler
AIM Tooling ACES Injection Molding Tooling
ANOVA Analysis Of Variance
ANSYS Finite Element Analysis Software
B-rep Boundary Representation
CAD Computer Aided Design
CCD Central Composite Design
CE Concurrent Engineering
CMM Coordinate Measuring Machine
CPL Closed Parting Loop
CR Combined Region
CSG Constructive Solid Geometry
CVR Concave Region
CXF Convex Face
di+, di- positive and negative deviation variables in compromise DSP
DBD Decision-Based Design
DFM Design for Manufacture
DFRTS Design for Rapid Tooling System
DOE Design of Experiments
cDSP Compromise Decision Support Problem
DSP Decision Support Problem
FEA Finite Element Analysis
GF Glue Face
GUI Graphical User Interface
IJM Injection Molding
LINGO Engineering Optimization Software
LT Layer Thickness in SLA process
LP Linear Programming
LPP Linear Physical Programming
MB Mold Base
MCD Mold Configuration Design
MD Mold Design
MFC Microsoft Foundation Classes
MGT Material Geometric Tailoring
MGTDT Material Geometric Tailoring Decision Template
ML Mold Life
xix
MPC Mold Piece Construction
MPD Mold Parting Direction
MPGT Material-Process Geometric Tailoring
MPGTDT Material-Process Geometric Tailoring Decision Template
MPMDM Multi-Piece Mold Design Method
NF Neighboring Face
OptdesX Engineering Optimization Software
PC Personal Computer
PD Parting Direction
PE Parting Edge
PIM Powder Injection Molding
PL Parting Line
PO Part Orientation in SLA Process
PS Parting Surface
R2, R2adj The ratio of the model sum of squares to the total sum of squares, and
that ratio adjusted for the number of parameters in the model
RCEM Robust Concept Exploration Method
Response Performance parameter of the system, i.e., a system constraint or goal
RM Rapid Manufacturing
RMCD Region-based Mold Configuration Design
RSE Response Surface Equation
RSM Response Surface Methodology
RTMDS Rapid Tooling Mold Design System
RTTB Rapid Tooling TestBed
RP Rapid Prototyping
SLA Stereolithography Apparatus
V-map Visibility Map
xx
SUMMARY
Physical models and prototypes are fundamental to superior product development
and production. Rapid Tooling techniques have the potential to dramatically reduce the
time and cost in producing limited quantities of functional prototypes in final material.
However, the current usage of Rapid Tooling has two problems: (1) mold design for parts
with a wide variety of geometries may take a long time; (2) design iterations between
designers and manufacturers may take a long time before different design requirements
are achieved in the prototypes. They overshadow the time and cost benefits of Rapid
Tooling. In this dissertation, these two problems are addressed by developing a Mold
Design Method and a Design-for-Manufacturing Method respectively.
Based on Computation Geometry, a Multi-piece Mold Design Method is
developed to automate several important mold design steps, including determining
parting directions, parting lines, and parting surfaces, and constructing mold pieces for
multi-piece molds. The method has three steps. First concave regions and convex faces
are generated from a given part. Second the generated concave regions and convex faces
are combined into several regions. Finally mold pieces are constructed for the combined
regions based on a reverse glue operation. The method is employed to develop a Rapid
Tooling Mold Design System, which has been used to design molds to fabricate
prototype parts with widely varying complexities. Two test parts and five industrial parts
are presented in the dissertation to illustrate the usage of the mold design system. The
xxi
running time of the system is tested for each part. The mold design results are also
verified by physical experiments.
Based on Decision-Based Design, a Design for Rapid Tooling System is
developed to aid manufacturers to tailor a submitted part efficiently and effectively. A
basic idea of this work is that for Rapid Tooling the burden of design-for-manufacture
can be transferred to the manufacturer by geometric tailoring decision templates. The
designer’s requirements on functional prototypes are formulated in the decision
templates, which are in the compromise DSP format. Three testing examples illustrate
that the design freedom given by the designer to the manufacturer is important for
reducing the iterations between the designer and manufacturer. By synthesizing
decisions of part design, rapid prototyping process, and injection molding process
variables, a solution strategy and a three-stage solution process are proposed for the
system. Compared with the current usage of Rapid Tooling, better decision order of
design variables is used in the design for Rapid Tooling system for the geometric
tailoring problem. Two case studies, a robot arm and a camera roller, are used to test the
system. Physical prototypes of the tailored part designs are produced for the validation of
the DFM method.
xxii
Chapter 1 – Functional Prototype Using Rapid Tooling
CHAPTER 1
FUNCTIONAL PROTOTYPE USING RAPID TOOLING
The product realization process, driven by market factors, is changing dramatically.

Building better physical prototypes in shorter period of time becomes more important in
product development and production. One way to achieve this goal for injection molded
products is through Rapid Tooling. In this chapter we will give a brief introduction to
Rapid Tooling and design for Rapid Tooling. The project by which this research is
motivated is presented to highlight some of the challenging issues involved in the
development of a Rapid Tooling Testbed. Research opportunities and foundations for
this research will also be presented followed by the research questions, goals and
hypotheses. In the last Section an overview of the entire dissertation is given.
1
1.1 PRIORS – FUNCTIONAL PROTOTYPE AND RAPID TOOLING

The principal objective in this dissertation is to develop methods to facilitate the
usage of Rapid Tooling for a part design. However, before the current usage of Rapid
Tooling and research opportunities are presented in Section 1.2, several important notions
are introduced in this section to establish a context for the reader. These notions include
the importance of functional prototype (Section 1.1.1), Rapid Prototyping (Section 1.1.2),
Rapid Tooling (Section 1.1.3) and Rapid Manufacturing (Section 1.1.4). The focus of
this research, which includes research questions and related hypotheses, is presented in
Section 1.3. Finally the overview of the entire dissertation is presented in Section 1.4).
1.1.1 Product Realization and Prototype

The product realization process, driven by market factors, is changing dramatically.
There are increasing global competition, decreasing levels of demand for discrete
products, and declining profitability. Consequently, only the companies who are able to
launch quality products ahead of their competitors can continue to enjoy healthy markets.
Smith and Reinertsen (1991) studied competitiveness of different products and stated that
“A company able to launch a quality product ahead of their competition not only realizes
100% of the market before rival products arrive but also tends to maintain a dominant
position for a few years even after competitive products have finally been announced.”
This is also evidenced by a vice-president of Hewlett Packard who testified that “a
reduction of one month in the printer development time would result in additional profits
in excess of the entire product development cost” (Kazmer and Speight, 1997).
As many executives now realize how vital it is to move new products to market
rapidly, designers, who play a key role in the product realization process, face the
constant pressure to cut the product development time and costs. In the mean time,
designers also face the sustained pressure on avoiding design errors. It is well known that
cost of design errors and related changes increases substantially with each step of product
development. According to Wohlers Associates, the cost of an engineering change
increases roughly by an order of magnitude (see Figure 1.1) as the design progresses
from one significant development phase to the next.
Product Development Phase Cost of Design Change

CONCEPTUAL MODELLING $10
DETAIL DESIGN $100
TEST $1,000
MANUFACTURING $10,000
PRODUCT REALSE $100,000
Figure 1.1 - The Cost of a Design Change Related to Development Phase

(Wohlers Associates).
Prototypes of the design can aid designers to find defects as early as possible and
correct them when it is still relatively cheap and easy. As stated in a paper from Harvard
2
Business Review (Bowen, et al., 1994), “… prototypes not only enable products to be
developed and launched more quickly, but also result in products that are both higher-
quality and more effective in fulfilling their intended purpose in the marketplace.”
As computers become more powerful, Virtual Prototyping (VR), which involves
analyzing Computer Aided Design (CAD) models for different end applications, is
developing quickly. Various analysis and simulation packages enable assessment of
product functionality; for example, kinematic modeling enables motion simulation, CFD
(Computational Fluid Dynamics) can replicate a wind tunnel and assess fluid flow, FEA
(Finite Element Analysis) can be used to determine load-carrying capacity and to predict
temperature distributions. Other analysis tools simulate various manufacturing and
assembly processes; for example, software packages, like Moldflow and CMold, are
available to model most of the usual materials-forming processes such as injection
molding, investment casting, and so forth. Other packages simulate the various assembly
operations, providing insight on setting up a manufacturing line (Siddique and Rosen,
1997; Gadh and Ratnakar, 1998). In many cases, these VP techniques and simulation
software systems can help designers to evaluate their designs quickly and cheaply.
Unfortunately, despite the promise of these tools with the advances in computer
simulation, not all aspects of design testing can be formulated in mathematical models.
Also as products and processes become increasingly complex, the confidence and
accuracy of a Virtual Prototype system may still not be able to satisfy a designer’s
expectations. Therefore, physical prototypes are still fundamental to superior product
development and production. In the words of Michael Schrage of the MIT Sloan School
of Management, “For companies that genuinely care about incremental and breakthrough
innovations, organizational redesign and core process re-engineering are not enough.
Companies that want to build better products must learn how to build better physical
prototypes.” (Schrage, 1993) Similar themes also pervaded in an address at the IMS
International Conference on Rapid Product Development given by Daimler-Benz’s Dr.
Werner Pollmann. He examined prototyping at Daimler-Benz and wrote as follows:
“Purchase of a car depends strongly on subjective impressions. Next to technical
properties like horsepower or security equipment, properties like noise, handling,
or styling are key factors for a purchase decision. But these properties can only be
evaluated by physical prototypes. For that reason, availability of high quality
functional prototypes will remain an important element of product development
and cannot be substituted by digital models and analysis.” (Pollmann, 1994)
From late 1980s, a new kind of technology, which is based on additive fabrication,
was developed to make physical prototypes more quickly. These techniques were named
Rapid Prototyping (RP), or Solid Freeform Fabrication (SFF). They are discussed in
Section 1.1.2 as the basis of Rapid Tooling, which is discussed in Section 1.1.3.
1.1.2 Rapid Prototyping

Rapid Prototyping, also called Solid Freeform Fabrication, or Layered
Manufacturing (LM), refers to the fabrication of physical parts layer-by-layer directly
from CAD data. It involves successively adding new materials, in layers, to create a solid
of some predefined shape. Therefore it is a fundamentally different method of fabrication
3
from traditional approaches like turning and milling, which are based on removing
materials, or rolling or casting, which are based on deforming materials into desired
shapes.
Currently there are over 30 RP technologies available for model production based on
the principle of additive fabrication, including Stereolithography (SL), Selective laser
sintering (SLS), Ink-Jet Printing, Fused Deposition Modeling (FDM). The major
differences among these technologies are in two aspects: (1) material used; and (2) part
building techniques. Figure 1.2 is a schematic drawing to show the building technique of
Stereolithography (SL), the most prevalent technology on the market, which was
pioneered by 3D Systems. A short explanation on how the process operates is as follows.
Initially, the elevator is located at a distance from the surface of the liquid equal to the
thickness of the first, bottom-most layer. The laser beam will scan the surface following
the contours of the slice. The interior of the contour is then hatched using a hatch pattern.
The liquid is a photopolymer that solidifies when exposed to the ultra-violet laser beam.
The elevator is moved downwards, and the subsequent layers are produced analogously.
Finally, the part is removed from the vat, and cured in a special oven.
Figure 1.2 - A Schematic Drawing of a SLA.

Rapid Prototyping has several advantages due to its unique fabrication process. It is
faster. Also the geometric complexity of the part has a significantly smaller impact on
the fabrication process. That is, when creating a part layer-by-layer, a simple cube and a
sculptured solid are equally easy to manufacture for RP. So some designs, which are so
complicated that it was impossible to make them before, can now be fabricated via RP
technologies. Figure 1.3 is a robot arm design based on the principle of Tensegrity
Structures in Rapid Prototyping and Manufacturing Institute (RPMI) at Georgia Institute
of Technology. The new design is light yet strong, which is suitable for a high-speed
robot. Furthermore, multi-material structures and parts with embedded electronics can be
created using the layer-by-layer fabrication of RP (Merz, et al., 1994; Kumar and Dutta,
1998; Jackson, 2000).
4
Figure 1.3 - A CAD Model and a SL Model of a Robot Arm (RPMI at Gatech).
Currently there are roughly 7000 RP systems installed worldwide (according to
Wohlers Associates at www.WohlersAssociates.com). The use of RP technologies has
resulted in a significant reduction in the prototyping time for numerous industrial cases
(Jacobs, 1992; Jacobs, 1996; Wohlers, 1998). One demonstrative example shown in
Figure 1.4 is Blaupunkt’s front cap. According to (Edelmann, 2000), Blaupunkt, a
German mobile electronics company, developed three functional models of a swivel-type
front cap with the help of Stereolithography. The company was able to save 7 weeks’
development time using the Stereolithography process compared with the conventional
milling method. Also the cost for prototypes was cut tremendously. The traditional
method – milling from solids – would have cost the company between 30,000 and 40,000
Figure 1.4 - A Stereolithography Model for Assembly Studies and Tests with
Regard to Collision Behavior (Blaupunkt).
5
DM, the costs with Stereolithography were 12,000 DM.

Karapatis and coauthors (1998) classified the functions of prototypes in four main
classes: the visual prototype, which is just a “solid”, real CAD model used to reveal
design defects and gain marketing clearance; the functional prototype (“form & fit”),
which is mainly used to detect assembly problems; the material prototype, which has the
same geometrical and mechanical properties as the final part; the production prototype,
which is made by the same process as the final part. To make material and production
prototypes of injection molding parts, a new kind of techniques, Rapid Tooling, was
developed based on Rapid Prototyping technologies in 1990s. They are discussed further
in Section 1.1.3.
1.1.3 Injection Molding and Rapid Tooling

Many consumer products involve the design and fabrication of injection molded
thermoplastic parts. Injection molding is the most widely used process for manufacturing
thermoplastic products (Mitchell, 1996). It has many benefits like forming near net shape
product with nothing or very little secondary machining process, and producing parts that
combine the functionality of many parts into one. The main character of injection
molding process is that shapeless material is formed by a tool (i.e. molds) with a
complementing cavity in the shape of the desired part. Injection molded parts are
produced by closing mold pieces and injecting molten material into the formed cavity.
After the molten material solidifies, the part is ejected from the mold. This process is
also applicable to metals and ceramics by a new technology known as powder injection
molding (PIM) (German and Bose, 1997).
Tools, which define the shape of the part, play a vital role in the injection molding
process. Traditionally, the manufacture of tooling for both prototype parts and
production components represents one of the longest and most costly phases in the
development of most new products. It typically took a month to produce a prototype
mold and another three months to complete mold design and manufacturing (Beckert,
1999). The cost and time implications of the tooling process are particularly problematic
for low-volume products aimed at niche markets, or alternatively for rapidly changing
high-volume products.
By using Rapid Prototyping technologies in the fabrication of tools or patterns, a
new kind of techniques, named Rapid Tooling (RT), can reduce tooling cost and time
especially when only small quantities of a part are needed (Jacobs, 1996). More
importantly the parts made by Rapid Tooling can be made of end-state materials like
polystyrene, which are required in the testing of functional prototypes in some industrial
cases.
Since the first Rapid Tooling process (QuickCast) was developed by 3D Systems in
1992 (Jacobs 1996), there are more than 20 Rapid Tooling methods until today (Wohlers,
2000), like Epoxy Tooling, Composite Tooling, Kirksite Tooling, Spray Metal Tooling,
Direct ACES Tooling, DTM RapidTool. Different tooling processes have different time
and material requirements.
Williams (1999) classifies Rapid Tooling processes into two categories: transfer and
direct. Transfer Rapid Tooling process utilizes RP model as a master pattern. An epoxy
6
is cast around the master to form a mold. Direct Rapid Tooling process uses Rapid
Prototyping machines to fabricate the tools directly. Tromans and Wimpenny (1998)
compared direct and indirect tooling, and predicted that “the use of rapid prototyping
technology to manufacture tooling will evolve from its indirect use as master patterns for
soft tooling, eventually providing direct methods of manufacturing tools on rapid
prototyping machines.” Several reasons they gave included accuracy of the tools, time
and easiness of the process. For similar reasons, Wohlers (2000) also believed the direct
approaches would gain an edge in the long term (6 to 10 years).
In addition, since only the Direct Rapid Tooling processes are investigated in our lab
(Rapid Prototyping and Manufacturing Institute), in this dissertation the author will only
consider the direct Rapid Tooling processes. Specifically, this research of design for
Rapid Tooling is mainly focused on one typical direct Rapid Tooling process, direct AIM
(ACES Injection Molding). The main steps of the direct AIM tooling are shown in Figure
1.5. After finishing the mold design in CAD models, a SLA system is used to build the
mold using the ACES (Accurate Clear Epoxy Solid) build style. The mold is shelled out
on the bottom side to leave a cavity which can be backfilled with various materials
including aluminum-filled epoxy, ceramics, and low-melting metals. After some optional
supplement processes such as milling or polishing, the mold is ready to be used in an
injection molding machine.
Direct AIM Tooling
Mold Mold
CAD Physical
Model Model
Mold build by SLA Mold after Epoxy Back Fill
Figure 1.5 – Direct AIM Tooling Process.

Direct Rapid Tooling has been used in many industrial cases and proven useful in
building functional prototypes. An example given here came from Xerox Corporation
(Jacobs, 2001). An internal Xerox customer required 100 polystyrene switch actuators in
a very short time period. After evaluating various alternatives, Jeffery Heath and his
team decided to try Direct AIM. The team was able to injection mold the required 100
polystyrene parts just 5 days after the CAD design was completed!
Compared to traditional tool making approaches like CNC and EDM, direct Rapid
Tooling has its limitations in controlling the accuracy and material properties of the
resulting tools, which is discussed further in Section 1.2. These limitations are being
addressed in development efforts and are expected to be mitigated in the near future. As
RP and RT technologies become more mature, Rapid Manufacturing (RM), that is, using
RP and RT technologies to manufacture end use parts directly becomes more feasible.
7
1.1.4 Rapid Manufacturing

Although this research is based on the prototype development, the methods presented
in the dissertation also have implications on Rapid Manufacturing. Therefore the current
development of Rapid Manufacturing, especially Rapid Production Tooling, is described
briefly in this section.
If RP and RT today can produce functional models quickly and inexpensively, a
logical question is why not use them to produce actual production parts? For most
applications today, the idea is not practical due to the limitations of speed, part size,
accuracy, surface finish, and material properties. However, some RP and RT techniques
have already been used to manufacture finished products, especially for low-volume
manufacturing and customized products.
In the development of Rapid Manufacturing, Wohlers observed as follows,
"Initially, it will happen where the unit price of a part is high and the volume is low,
and also where the part is small — about the size that would fit within a 6-in.
cube… In the automotive industry, at first, such activity will be limited to high-end
personalized vehicles where volumes are lower … In the medical area, rapid
manufacturing will continue to make inroads in applications such as custom
implants. Small customized parts for helicopter interiors such as structural parts for
seats, cabinetry, or nameplates are other potential applications.”
Currently, several commercially available processes for producing products exist.
As an example, 3D Keltool given by 3D Systems, has achieved more than 3 million shots
for unfilled thermoplastics and over 500,000 shots with glass-fiber-filled thermoplastics.
Although most of these processes are transfer Rapid Tooling, the direct Rapid Tooling
techniques are continuously improved. New technologies are also in development. For
example, the researchers in Sandia are developing a system which uses laser to melt
powdered materials in layers to produce the final part. The part is as metallically dense as
one made by conventional means (Bylinsky, 1998). Recently this technology (Laser
Engineered Net Shaping – LENS) has been commercialized by OptomecTM.
The pressure of time, quality and cost, together with increasing product variety, more
customized products and world-wide competition is driving technology development and
implementation in the area of rapid manufacturing. Hilton (2001) presented a conceptual
model of “the disposable tool”, which uses lower-cost, lower-volume tools to produce
products instead of high-cost, high-volume tools. It is an optional approach considering
the trade-off between tool performance with cost and time. Currently, at least 25
different groups are investigating rapid production tooling (Jacobs, 2001). In (Ashley,
1994), a long list of top US companies, including Ford Motor Co. and Pitney Bowes Inc.,
had joined a total of nine new research consortia devoted to the development of processes
by which large-volume-capable production tooling could be produced quickly and
inexpensively.
In order to compress the product development process, the technologies including
Rapid Prototyping, Rapid Tooling and Rapid Manufacturing are introduced in this
section. In the next section, design for Rapid Tooling will be discussed in order to
outline the research focus for the dissertation in Section 1.3.
8
1.2 DESIGN FOR RAPID TOOLING AND RESEARCH OPPORTUNITIES

The motivation and foundation for investigating the proposed research are described
in this section. First an overview of the current usage of Rapid Tooling is described in
Section 1.2.1 with two related problems identified. The Rapid Tooling Testbed is
presented in Section 1.2.2 to provide a context and foster a better understanding of the
motivation. Two main challenges related to the testbed are described in Section 1.2.3 to
justify the research approaches used in this research. Finally research opportunities are
discussed in Section 1.2.4 before the focus of this research is presented in Section 1.3.
1.2.1 Current Usage of Rapid Tooling and Related Problems

Rapid Tooling has the potential to dramatically shorten the time required to produce
functional prototypes or products. However, several problems may extend the lead-time
to get the functional prototypes or products. These problems should be addressed to
realize this potential. In this section the current usage of Rapid Tooling is examined to
show these problems.
The whole process of using direct AIM tooling to get functional prototypes includes
part design, mold design, mold fabrication by SLA, and part fabrication by injection
molding (refer to Figure 1.6). Designers carry out part design, which results in a CAD
model of the part to be fabricated. The CAD model is then sent to manufacturers, for
example, a Rapid Tooling service bureau. According to the part design, the
manufacturers will design and build molds to produce prototype parts. Finally the
prototypes are sent back to the designers for functional testing. So the manufacturer, in
this dissertation, actually includes mold designer, rapid prototyping engineer, and
injection molding engineer.
Designer Manufacturer
Part
CAD
Model Mold CAD Model
Mold build by RP
Mold after Epoxy Back Fill Injection Part
Phase I: Phase II: Phase III: Phase IV:

Design Part Design Mold Build Mold Build Part
Figure 1.6 – Processes of Using Direct AIM Tooling to Build Prototypes.

As described in (Jacobs, 1996), a designer is required to do the data preparation for
using Rapid Tooling. That is, the designer should send a fabricatable CAD model
(usually in STL file format) to a manufacturer. If some features of the part design are
found inappropriate for the manufacturing process, the designer is required to change the
features according to the feedbacks given by the manufacturer. Often, many iterations of
changing are required before satisfactory prototypes can be fabricated.
Some tasks, like the mold design for a given part, are elaborate and difficult for the
manufacturer, especially if the part is designed without considering the manufacturing
requirements. By using RP machines, the fabrication of mold pieces may only take one
or two days. However, it may take one week to generate a qualified mold design that is
9
compatible with the RP machine and the injection molding machine. Some requirements
given by the designer (such as tight accuracy and surface finish) may make the mold
design and process planning rather difficult, or even impossible. Whenever the designer
is required to modify the part design, the manufacturer needs to go through the whole
processes (mold design, RP process planning, and IJM process planning) again for the
new part design.
Therefore, two kinds of problems are identified in the current usage of Rapid
Tooling, which may significantly impede the applications of Rapid Tooling in the
product development processes.
(1) Several steps in the processes of using Rapid Tooling may take a long time.
In the phase of part design, a designer often spends a lot of time in Design-for-
Manufacture (DFM) step in order to get a fabricatable part design. The task of Design-
for-Manufacture proves to be a heavy burden for designers since most designers are still
not familiar with Rapid Tooling requirements. Even worse, as a new kind of
manufacturing method, Rapid Tooling is developing so quickly that no accurate
information about how to design for Rapid Tooling is available on any published
handbooks. All these factors may make the Design-for-Manufacture step take a long
time. In some cases, it even makes designers hesitate to use Rapid Tooling because of
the unfamiliarity with the DFM.
The mold design for a given part also proves to be an elaborate and difficult job for
the manufacturer. For example, “designing the parting lines and surfaces for a mold is
one of those thankless industrial jobs that requires intelligence and experience but isn’t
the least bit interesting or fun.” (CAD/CAM Publishing Inc., 1999) Even worse, the mold
design step may repeat several times whenever the part design is changed. Materialise
(www.Materialise.com), a leading Rapid Prototyping provider, once gave a report named
‘Data Preparation as a Bottleneck’, which stated that:
“The actual production time of the rapid tool insert varies from 15 to 35 hours, and
the post processing takes around 20 to 40 hours of work. Where traditional tooling
uses 8 to 10 weeks to make the tool, rapid tooling does the job in one to two weeks.
The cost benefit of rapid tooling compared to traditional tooling is about 25% to
30%. … However, the lead-time necessary for obtaining the tool design lays a
shadow over the time and cost benefits of the rapid tooling process. … The lead-
time from part design to tool design can be more than three weeks, which
eliminates the time and especially the cost benefits of the rapid tooling process.”
(2) The iterations of design changes may take a long time.
Since most designers are not familiar with the Rapid Tooling technologies, some
requirements of a part design may be unreasonable, or even out of the capability of the
RT technologies. Whenever these factors are identified in the mold design and process
planning phases, the designer is notified to change them according to the feedbacks given
by the manufacturer. Sometimes, negotiation between designer and manufacturer is
necessary. It is well known that the two types of engineers “do not speak the same
language,” a situation which has often been described by saying that design engineer do
not bother to speak to manufacturing engineers and just throw their drawings over the
10
wall separating them. Currently many Rapid Tooling facilities are located at Service
Bureaus (Wohlers, 1999), which are separated from designers geographically and
organizationally. This makes communication between designer and manufacturer even
more difficult.
Thus we believe Rapid Tooling will never be rapid if the designer still needs one or
two weeks to modify his/her design to facilitate Rapid Tooling process after initially
submitting the design, and the tool design for the given part will take another week before
the mold can be fabricated. Therefore the key question to be investigated in this
dissertation is:
How to reduce the lead-time in the usage of Rapid Tooling to produce functional
prototypes for a part design?
Being related to the two problems identified in this section which cause long lead-
time in the application of Rapid Tooling, this question can be answered in two levels: (1)
How to reduce the time of several time-consuming steps? and 2) How to reduce the
number of iterations between the designer and manufacturer? To effectively answer
these two questions, research opportunities in design for Rapid Tooling are discussed in
Section 1.2.4. Based on the opportunities, the research questions and related hypotheses
are presented in Section 1.3.
In the next section, the motivating project (RTTB) of this research is presented to
provide a context and foster a better understanding of the DFM method to be presented in
this dissertation.
1.2.2 Motivating Project – Rapid Tooling TestBed

The research in this dissertation is motivated by the Rapid Tooling TestBed (RTTB),
which is sponsored by National Science Foundation (NSF DDF DMI-9618039). The
RTTB is intended to be an experimental testbed that supports exploration of design and
design-for-manufacture issues related to rapid prototyping and rapid tooling for injection
molding (Allen and Rosen, 1997; Rosen, 2000). It supports the design of parts and
molds, the selection of prototyping technologies and vendors, and the fabrication of those
parts and molds.
As depicted in Figure 1.7, the RTTB is divided into three stages: the design stage,
the design for manufacture stage, and the manufacturing stage. During the design stage,
requirements are entered into the testbed. These requirements consist of a part
representation along with the design specifications for the part. In the DFM stage,
suitable part and mold materials, rapid prototype technologies, and injection molding
technologies, are selected for the fabrication of both the injection molding tooling and the
part. In addition, the injection molding tooling is designed and the part is tailored to
facilitate the manufacturing of both the mold and part. The final stage involves the actual
manufacturing of both the tools and the part. The work in this dissertation is related to
geometric tailoring and generation and selection of mold configuration, which are
marked in black lines in Figure 1.7. Before these two modules, the appropriate material
and process have been determined for the given part (Herrmann and Allen, 1999).
11
Design Design for Manufacture Manufacture

CAD
Tools Preliminary
Part Selection of Geometric
Representation SFF Tailoring
Techniques Generation &
Preliminary Combining Selection of
Selection of SFF & IJM Part
Mold
Design Materials Techniques Configuration
Fabrication
Solution Premilinary
Design Selection of Resource Mold
Methods Specification IJM Selection Fabrication
Techniques
Event Timeline
Figure 1.7 – RTTB System and My Focus.

As a result of the RTTB project, “customer” of the rapid tooling testbed will receive
nearly production-representative components in a variety of polymer, ceramic, and metal
materials within 3-4 days of submitting a product model. Related to RTTB project,
challenges and the related research approaches are discussed in the next section.
1.2.3 Challenges of RTTB and My Research Approach

The Rapid Tooling TestBed is intended to assist a designer in obtaining useful
prototypes for his/her part design, where “useful” means that the prototype mimics some
desired production characteristics. A part given by the designer may have different kinds
of shapes. The designer may care about different kinds of production characteristics and
requirements. Therefore, the research in this dissertation is not considering the
application of Rapid Tooling for a part or just one kind of parts. Instead general methods
for handling different kinds of part designs are explored. To develop these methods, two
challenges of RTTB are identified in this dissertation.
(1) A part may have a wide variety of geometries.
To implement different design functions, a part designed by the designer may have
different geometries. Figure 1.8 shows some industrial parts considered in this
dissertation. It is quite obvious that different designs require different shapes.
In Section 1.2.1, the first problem identified by the author in the current usage of
Rapid Tooling is that several steps, especially the mold design step, take a long time.
One reason for the problem is the wide variety of possible geometries that a part may
have. Because the mold design is tightly related to the shape of the part, the
manufacturer may need to spend a long time in this step.
12
Figure 1.8 – Different Geometries of Some Industrial Parts.

(2) A part may have a wide variety of design requirements.
A designer can have a wide variety of design requirements for the part design.
Related to the different geometries in the part, there are different kinds of variables and
parameters. Also the attributes or design goals are quite different according to the design
functions. For example, the production characteristics of a part can be the maximum
stress, or the deflection of a feature. In addition the designer may give requirements such
as cost, time, surface finish and accuracy of some important surfaces.
13
In most situations, the designers’ preferences on these design requirements are quite
different. So the “customer” of RTTB can be classified in different categories with the
most frequently asked question they may have as shown below:
• Some care more about time: How soon I can get my parts?
• Some care more about tolerance: What tolerance can be achieved for specified
design features?
• Some care more about cost and tolerance: How are the tolerances correlated with
cost?
• Some care more about functions: What will be the material properties, internal
stresses, etc.?
This list goes on. It is quite obvious from the list that the tradeoffs between the
different design requirements are an important consideration in the RTTB system.
In Section 1.2.1, the iterations between the designer and manufacturer are identified
as a problem in the current usage of Rapid Tooling. These iterations mean the designer
and manufacturer need to make tradeoffs between their requirements. This is rather
difficult and time-consuming. The author believes one reason for this problem is the lack
of tools to help the designer and manufacturer to make the tradeoffs between their
requirements.
Therefore, the research approaches to be presented in this dissertation are based on
Computational Geometry and Decision-based Design, which correspond to the
challenges of wide varieties of geometries and design requirements respectively.
(1) Mold design based on Computational Geometry.
For the wide variety of geometries considered in the mold design, the author believes
that the computer can aid the manufacturer to dramatically reduce the time needed in this
step. Whenever the designer changes the part design slightly, the manufacturer should be
able to repeat the mold design in even less time. Solving geometric problems with the
aid of computer requires carefully designed geometric algorithms and data structures,
which are exactly the research areas of computational geometry.
Computational geometry emerged from the field of algorithms design and analysis in
the late 1970s. It has grown into a recognized discipline partly due to the advent of
powerful computers, and the explosion of application domains – computer graphics,
geographic information system, robotics, computer aided design and manufacture – in
which geometric algorithms play a fundamental role. According to (Goodman and
O'Rourke, 1997), computational geometry means “the study of geometric problems from
a computational point of view, including computational complexity, computational
topology, and questions involving the combinatorial complexity of arrangements and
polyhedra.”
Computational geometry and the research opportunities in mold design are further
analyzed in Section 1.2.4. Based on the analysis, the research questions and related
hypotheses on mold design are presented in Section 1.3.2.
14
(2) Design-for-Manufacture based on Decision-Based Design.

The cost, lead-time and quality of injection molded parts are tightly related to the
part design, mold design, and fabrication process planning (Pye, 1989; Rosato and
Rosato, 1995). These tight interrelations make the tradeoffs between the wide variety of
design and manufacturing requirements rather difficult to make.
In this dissertation, the fabrication process planning, in which the manufacturing
parameters and variables are determined, is considered as a design process to distinguish
it from the actual fabrication process itself. So the main design processes of using RT to
get the prototypes include part design, mold design, RP process planning, and IJM
process planning. Currently these design processes are sequential. That is, mold design
is started after the designer fixes all the part design variables and sends the CAD model
of the part to the manufacturer. Similarly, the RP process planning needs the mold
design before it starts, and the injection molding process planning needs the information
of part design and mold design. These sequential processes can be shown in Figure
1.9.(a) as one of the product development spirals.
Better Part
Design
Part Design Mold Design
Parameters & Parameters &
Variables Variables
Mold Design
Attributes
Part Design Function
Cost
Testing Time
Rapid Quality, etc
Designer Prototyping
Manufacturer Process
Planning
RP Process IJM Process
Parameters & Parameters &
Functional Injection Variables Variables
Prototypes Molding
Process
Planning
(a) Product Development Spiral (b) Relations of Parameters and Attributes

Figure 1.9 – Design for Rapid Tooling.
In the above sequential design processes, the designer and manufacturer make
decisions about part design variables, mold design variables, RP process variables, and
IJM process variables. These decisions will affect the attributes such as cost, time and
function of the prototypes, as shown in Figure 1.9.(b). So from the viewpoint of
decision-making, the author believes the decisions in each design process are not
necessarily sequential or in the same order as that of the information flow. To some
extent, they should be integrated or concurrent because of the interrelations between the
variables. The work presented in Chapter 5 and 6 is toward this direction.
In this dissertation, the framework of Decision-Based Design, DBD, is used for
describing the part realization process. In Decision-Based Design, decisions serve as
markers to identify the progression of a design from initiation to implementation to
termination. In DBD, decisions represent a unit of communication; one that has both
domain-dependent and domain-independent features (Shupe, 1988; Mistree et al., 1989).
15
By approaching design from the perspective of making decisions, it becomes possible to

envision a unified approach for design-for-manufacture problem, which is further
analyzed in Section 1.2.4.
Related to the research approaches used in this dissertation, research opportunities in
mold design and design-for-manufacture are identified in the next section.
1.2.4 Research Opportunities in Mold Design and Design-for-Manufacture

Considering the two challenges of the Rapid Tooling TestBed, the research
approaches used in this research are presented in the last section. In this section, the
opportunities for making contributions in mold design and Design-for-Manufacture are
further discussed before the research focus is presented in Section 1.3.
• Mold Design for Different Geometries
A mold is a highly sophisticated design that consists of several components for even
a very simple part. A typical layout and descriptions of parts in the mold is shown in
Figure 1.10. The function of the mold is threefold: injecting the plasticized plastics in the
desired shape, solidifying the injecting molding part (cooling for thermoplastics and
heating for thermosets), and ejecting the solidified parts out of the mold.
Correspondingly, the mold has five sets of components: (1) the components for injection
(e.g. sprue bushing); (2) the cavity and core; (3) the components for cooling (e.g. cooling
channels in core and cavity); (4) the components for ejection (e.g. ejector pins and ejector
Figure 1.10 – An Illustrative Mold for Injection Molding.
16
plate); and (5) the components for assembling other parts in the machine (e.g. leader
pins). The components of core and cavity are to form the shape of the part. Therefore
they are the most important component in the mold design.
To design such a mold for a part with varied geometries is a rather complicated
process involving several steps. The author classified the main steps as shown in Figure
1.11 which are based on the handbooks of (Rees, 1995) and (Rosato and Rosato, 1995).
They are simplified as sequential processes although some steps may actually be parallel
and some loops may exist.
Determine Determine mold Determine

impression type (integral/ Parting
number in a mold insert) directions
(a)
Determine Check Part for
Determine
Parting Draft Angle,
Mold Parting Lines
Surface Thickness, etc.
Configuration
Design
(choose layout) Determine Part Determine external and
Orientation and Determine Core internal side actions and
Position in Mold and Cavity movement requirements,
Base if any
Feed systems (sprue,

Cooling system Ejection types
runner, gate) types
configuration and position
and layout.
Determine part
Splitting mold
shrinkage and
Mold Base Selection base into mold
apply shrinkage
pieces
compensation
(b)
Mold Detail Add feed systems

Creating Cooling Add holes for
Design (sprue, runner, gate)
system ejection pins in mold
(construction) in mold.
Add ancillary items (guide Complete

Check Interference
bushes, guide pillars, Assembly
by Simulation
spure bush, etc.) Design
Figure 1.11 – Mold Design Process and My Considerations.

The mold design process is divided into two phases: configuration design and detail
design. The mold configuration design includes the selection of parting direction, parting
surface, ejection method, feeding method, and cooling method. These selections
significantly affect the mold cost and DFM features. The mold detail design is mainly to
construct the mold pieces according to the mold configurations. CAD systems like
Pro/Mold and SolidWorks can help mold designers to reduce the time in this phase.
17
In prototype tooling, many components of the mold are standardized. Specifically,

the usage of interchangeable mold inserts into a standardized mold base, or master frame,
significantly simplifies the mold design. The inserts encompass the mold core and
cavity, from which parts are molded. The master frame contains standardized
components, such as the ejector system, gating, nozzles, and built-in water jackets for
cooling. Standardizing these components in the master frame eliminated much of the
work associated with design and manufacturing the components in a new mold. Figure
1.12 is a picture showing the master frame used in Morgan Press at RPMI with a SLA
mold insert.
Cavity
Sprue
Master
Frame
Runner
B-Plate
Leader
Pin
Ejector
Plate
Figure 1.12 – Standardized Master Frame Used in Morgan Press at RPMI.

The usage of the standardized master frame saves time in several steps including the
design of sprue, runner, ejection pins, and cooling lines. However, the design and
construction of the mold inserts (core and cavity for two-piece mold) are still time-
consuming due to the varied geometries of a part.
Several commercial software systems have been developed to aid mold designers in
designing mold inserts. However, these software systems are mainly on mold
construction instead of mold configuration design. And lots of improvements are
expected from mold designers. In (Beckert, 1999), Steve Sivitter, vice president of
Delcam International Inc., gives suggestions for mold design software according to their
survey of mold designers as below:
Moldmakers look for design software that allows core and cavity surfaces to be
separated automatically and that can automatically determine the natural parting line
for shapes and use that curve to create the split surfaces. Because plastic parts must
have sloping surfaces to be removed from the mold, software must also be able to
check, create or modify the draft on surface models. What’s more, software should
also allow for even multi-surface, variable radius fillets to be created quickly and
easily.
18
Therefore, the configuration design of parting direction, parting lines, and parting
surface, and the construction of mold pieces are investigated in this dissertation. They
are shown in Figure 1.11 enclosed with thick boundary lines.
Another technique in prototype tooling that is also considered in this research is
multi-piece molding. Being different from two-piece molds, multi-piece molds are the
molds that contain more than two mold pieces. Each mold piece can be hand-loaded into
a mold base mounted on the injection molding machine platens. During mold injection
and part cooling process, the molds are accurately and securely clamped into the holding
device. Finally each mold can be hand-removed from the mold base to release the part.
Although the technique is only suitable for producing a small number of parts, it can
fabricate the prototypes of parts with more complicated geometries. An example of
multi-piece molding, which is provided by Protoform GmbH is shown in Figure 1.13.
The injection molded part has several undercuts (features that would prevent the part
from ejection out of the mold), therefore it cannot be formed by only two mold pieces.
(a) A part to be built (b) CAD models of disassembled mold pieces
(c) Photo of mold pieces (d) Injection molded parts
Figure 1.13 – A Multi-piece Molds for a Part Given by Protoform GmbH.

Since multi-piece molds have more than a pair of opposite parting directions, it is
more difficult and time-consuming to generate a good mold design. The author believes
the design for multi-piece molding also provides plenty of research opportunities.
Therefore,
19
it is proposed to develop a systematic approach from geometrical perspective to

automate several important mold design steps, including selecting parting
directions, parting lines, and parting surfaces, and constructing mold pieces for
multi-piece mold design.
The Multi-Piece Mold Design Method (MPMDM) is developed in this dissertation to
address this question. The MPMDM and its associated steps are introduced in Section
3.1 and 3.3. The associated system (Rapid Tooling Mold Design System) is introduced in
Section 4.1.
• Design-for-Manufacture for Different Requirements
Design-for-manufacture (DFM) “is concerned with understanding how product
design interacts with the other components of the manufacturing system and in defining
product design alternatives which help facilitate ‘global’ optimization of the
manufacturing system as a whole” (Stoll, 1991). In the past 10 years the area of DFM is
under intense investigation. A large number of DFM methods and tools have been
developed in various manufacturing domains for a part design.
Although there are a number of ways to categorize these approaches, the author will
use four categorization methods (analyzing approach, design modification scope,
measure of manufacturability, and responsibility of DFM) to identify research
opportunities for the usage of Rapid Tooling. More detailed descriptions on some of
these approaches are presented in Section 2.5.
(1) Analyzing approach.
Based on what analyzing approach may be taken, the DFM approaches are classified
roughly as follows:
• Rule-based approaches. In these approaches, rules are used to identify infeasible
design attributes from direct inspection of the design description. A typical
example of the rule-based approaches is the guidelines in a handbook. These
guidelines enumerated design configurations that posed manufacturability
problems. The designer had to carefully study these guidelines and avoid these
configurations that resulted in poor manufacturability.
• Plan-based approaches. In these approaches, manufacturing plans for the design
are generated first. If there is more than one possible plan, then the most
promising plan should be used for analyzing manufacturability.
Rapid Tooling has several manufacturing operations. Therefore it is rather difficult
to formulate design guidelines and to determine the manufacturability of a design directly
from the design description. Thus the plan-based approach is more suitable for Rapid
Tooling.
(2) Design modification scope.
A part design usually consists of three kinds of requirements. First is the basic shape
of the part, or topology. Most parts are shaped the way they are for a specific reason.
Second is the geometry, or parametric values of the part. Third are other requirements,
including fit (e.g. tolerance, accuracy), function (e.g. materials, physical and mechanical
20
properties, thermal properties, electrical properties), appearance (e.g. color, texture, and
contour), time, cost, safety and environmental issues.
Different DFM approaches focus on different design requirements. So they can be
classified according to their design modification scopes. For example, if the guidelines in
a handbook focus on the design configurations, then the design modification scope is the
topology of the part design.
The design requirements that need to be tested in the functional prototypes produced
by Rapid Tooling varied, including the form, fit, function, and appearance. In this
dissertation, the design modification scope of the DFM approach includes geometry,
tolerance, accuracy, physical and mechanical properties, time, and cost. The changing of
the part shape or topology is not considered in this research.
(3) Measure of Manufacturability.
Based on the scales on which manufacturability is measured, the DFM approaches
are classified as follows (Gupta, et al., 1995):
• Binary measures. The approaches using this manufacturability rating simply
report whether or not a given set of design attributes is manufacturable.
• Qualitative measures. Designs are given qualitative grades (such as “poor”,
“average”, “good”, or “excellent”) based on their manufacturability by a certain
production process.
• Abstract quantitative. This type of approaches involves rating a design by
assigning numerical ratings along some abstract scale, for example a
manufacturability index between 1 and 10.
• Time and cost. In general, a design’s manufacturability is a measure of the effort
required to manufacture the part according to the design specifications. Since all
manufacturing operations have measurable time and cost, this type of approaches
use time and cost as an underlying basis to form a suitable manufacturability
rating.
For the approaches based on the measure of manufacturability by binary measures,
qualitative measures, and abstract quantitative, it can be difficult to interpret the measures
or to compare and combine them. However, the ratings based on time and cost can easily
be combined into an overall rating. Since several requirements are considered in this
dissertation and the tradeoffs between them should be made, the measure of
manufacturability by time and cost is more suitable for Rapid Tooling.
(4) Responsibility of DFM.
According to the people who are responsible for DFM, the DFM approaches are
classified as follows.
• Designer-based approaches. In several DFM approaches, it is taken for granted
that the designer takes the full responsibility of DFM. In these approaches, “the
purpose of DFM is to ensure that the designer considers manufacturing issues
during the design.” (Dissinger and Magrab, 1996) So this type of approach tries to
developed tools from design guidelines to automated manufacturability analysis
systems, which are to be used by the designer. However, automated analysis
21
meets difficulties if a part or a manufacturing process is complex. It is rather

difficult, if not impossible, to formulate all manufacturer’s experience and
decisions into rules and algorithms.
• Approaches based on an integrated team of designer and manufacturer. Ideally
designer and manufacturer should cooperate in DFM problem. Concurrent
Engineering (CE) advocates designers to work with manufacturing engineers
during the design process. However, these approaches are more on organizational
structures and information sharing than on integration of decisions. The author
believes the collaborations between the designer and manufacturer are not
guaranteed even if they can work as a team. More importantly, a coordination
method and a uniform decision framework need to be developed for the
collaborative decision-making.
• Manufacturer-based approaches. Recently much of the research on DFM has
looked to the VLSI (very large scale integrated circuits) community because of
the tremendous success of the VLSI industry. A group of researchers in a NSF
workshop on structured design methods observed that a clean interface, which
separate design efforts at increasingly high levels of abstraction from the growing
complexities of the fabrication processes, is the key to the rapid success of the
VLSI development (Antonsson, 1996). That is, a VLSI designer does not have to
fully detail his/her design; a process compiler will fill in the process-dependent
details. So in this kind of approach, the designer actually transfers part of the
burden of DFM for a fabrication process to the manufacturer.
The Rapid Tooling TestBed is partly motivated by developing approaches to achieve
a clean interface between the designer and manufacturer for Rapid Prototyping and Rapid
Tooling processes (Allen and Rosen, 1997; Rosen, 1998). The research presented in this
dissertation also proceeds in this direction. Since the objective is to develop production-
representative prototypes, part of the DFM for Rapid Tooling is actually design-for-
prototyping. The author believes these DFM tasks (Geometric Tailoring which is defined
later) can be transferred to the manufacturer. In doing so, it is important to develop an
approach and a system to aid the manufacturer to integrate design and manufacturing
requirements, and to gain a better understanding of the tradeoffs between these
requirements.
As a summary, a DFM approach, which is based on process planning, considering
the tradeoffs of many design requirements, and executed by the manufacturer, is explored
in this dissertation for Rapid Tooling process. But the current state of research does not
address the issues of how to transform DFM to the manufacturer and how to integrate
DFM with the process planning in the manufacturer side. Therefore,
it is proposed to develop a formal, technical approach founded in Decision-Based
Design to tailor a submitted part design according to both product and process
considerations for Rapid Tooling process.
In light of the DFM approach explored in this dissertation, the following definitions
for geometric tailoring and design for Rapid Tooling are offered to provide a context for
the remainder of the dissertation.
22
Geometric Tailoring, in this dissertation, is to change the geometry of a part to

lower fabrication cost and time, and to produce functional prototypes to mimic the
production functions, because the material and fabrication process to get molds and parts
are different from those in producing products. Geometric Tailoring is actually part of
design-for-manufacture.
Design for Rapid Tooling, in this dissertation, is the integration of the requirements
of geometric tailoring, mold design, and fabrication process planning to make decisions
on their variables for producing the prototypes of a part design.
Considerations
on Topology,
Assembly,
Design for etc.
Manufacture
Geometric
Tailoring Mold Design
(Chp 5)
Design for
Rapid Tooling
(Chp 6)
IJM Process RP Process
Planning Planning
Figure 1.14 – Relations of DFRT, DFM and Geometric Tailoring.

The relations of Design for Rapid Tooling (DFRT), Design-for-Manufacture and
Geometric Tailoring are shown in Figure 1.14. The research work related to Geometric
Tailoring is elaborated in Chapter 5, and the research work related to DFRT is presented
in Chapter 6.
With the research opportunities identified and two definitions presented, the focus of
this dissertation will be described in the next section.
1.3 RESEARCH FOCUS IN THE DISSERTATION
The research focus in this dissertation is embodied as follows:
! a set of research questions that captures motivation and specific issues to be
addressed,
! a set of corresponding research hypotheses that offers a context by which the
research proceeds,
! a set of verification studies performed in this work, and
! a set of resulting research contributions that embody the deliverables from the
research in terms of intellectual value, a repeatable method of solution,
limitations, and avenues of further investigation.
The research questions are presented in Section 1.3.1 along with the corresponding
research hypotheses and related tasks. The validation strategy for this work is presented
in Section 1.3.2. The resulting research contributions are introduced in Section 1.3.3.
23
1.3.1 The Principal Goal, Research Questions and Hypotheses in the Dissertation
The principal goal in this dissertation is the development of methods to reduce the
lead-time in the usage of Rapid Tooling to produce functional prototypes for a part
design. The principal research goal is to be achieved by two sub-goals: 1) To reduce the
time of several time-consuming steps in mold design, and 2) To reduce the number of
iterations between the designer and manufacturer. As discussed in the previous section,
Computational Geometry and Decision-Based Design provide the foundation on which
this work is built. Given these foundations and goals, the motivation for this research is
embodied in two primary research questions which are stated as follows.
Primary Research Questions:
Q1. How to aid the mold designer to reduce the mold design time for a wide
variety of part geometries in the design for Rapid Tooling process?
Q2. How to reduce the time of iteration between the designer and manufacturer
in the usage of Rapid Tooling for a wide variety of design requirements?
These research questions are related directly to the sub-goals in this research. The
following hypotheses are investigated in this dissertation in response to the primary
research questions. There is a one-to-one correspondence between the hypotheses and
primary research questions.
Hypothesis 1: Multi-Piece Mold Design Method provides a method to automate
several key steps of the mold design process, which can greatly reduce the
mold design time for a wide variety of part geometries.
Hypothesis 2: Geometric tailoring for Rapid Tooling can be integrated with
process planning based on decision templates and solved by the manufacturer,
which can reduce the time of iteration between the designer and manufacturer.
Since Questions 1 and 2 are quite broad, three supporting research questions and
sub-hypotheses are proposed to facilitate the verification of Hypothesis 1. Three
supporting research questions and sub-hypotheses are proposed to facilitate the
verification of Hypothesis 2. The supporting questions and sub-hypotheses for
Hypothesis 1 are stated as follows.
Q1.1. What are appropriate basic elements to automate several mold design steps
for a wide variety of part geometries?
Q1.2. How to generate mold configurations by a systematic design process based
on the basic elements?
Q1.3. How to generate mold pieces from a given mold base effectively and
efficiently according to a mold configuration design?
Sub-Hypothesis 1.1: Concave region and convex face are two kinds of basic
elements that provide an efficient and effective approach for exploring and
developing molds design method for Rapid Tooling.
24
Sub-Hypothesis 1.2: The Region based combining process and related algorithms
provide a systematic method to generate mold configurations of multi-piece
molds design.
Sub-Hypothesis 1.3: The Reverse Glue Mold Construction Method and related
algorithms provide an efficient and effective method for constructing multi-
piece mold designs for Rapid Tooling.
There is a one-to-one correspondence between the supporting questions and sub-
hypotheses. Although developed for Rapid Tooling, sub-hypotheses 1.1, 1.2 and 1.3 also
have implications on the mold design for the production injection molding (refer to
Section 3.4, 3.5 and 3.6).
The second hypothesis is related to the Design-for-Manufacture problem for Rapid
Tooling. The supporting questions and sub-hypotheses for Hypothesis 2 are stated as
follows.
Q2.1. How to reduce the iterations between the designer and manufacturer in
producing functional prototypes that have different material properties
from products?
Q2.2. How to formulate the design for Rapid Tooling problem which integrates
decisions on design and manufacturing variables and other design and
manufacturing requirements including goals, constraints, and preferences?
Q2.3. How to solve the design for Rapid Tooling problem effectively and
efficiently?
Sub-Hypothesis 2.1: The designer can initiate a material geometric tailoring

(MGT) formulation based on a MGT decision template; therefore the
manufacturer, who completes and solves the MGT problem, can produce
production-representative prototypes more quickly.
Sub-Hypothesis 2.2: The design for Rapid Tooling problem can be formulated by
several compromise DSPs and tasks, which can then be integrated into a
design for Rapid Tooling system (DFRTS).
Sub-Hypothesis 2.3: A three-stage solution process can be utilized to get a
satisficing solution effectively and efficiently based on design and
manufacturing models and continuous/discrete variables.
The research questions and hypotheses provide a frame upon which the different
developments and contributions from this research will be achieved. The relationship
between the hypotheses and the various sections of the dissertation are summarized in
Table 1.1. Verification and validation of the issues are discussed in the next section, and
testing of the individual hypotheses commences in Chapter 3, lasting until Chapter 6.
Further verification of the hypotheses comes from application of the Rapid Tooling Mold
Design System and Design for Rapid Tooling System on producing prototypes for two
industrial parts. The application of the systems is presented in Chapter 7 and 8. Chapter 9
contains a review of verification of hypotheses. In the next section the strategy for
hypotheses validation in this dissertation is presented.
25
Table 1.1 – Relationship Between Hypotheses and Dissertation Sections

Hypothesis and Tasks Section Section Tested
Discussed
H1. Geometric reasoning algorithms for Chap 3 Chap 4
mold design Chap 7
Chap 8
H1.1 Basic elements and related algorithms §3.4 §4.4, §4.5,
§7.2, §8.2
H1.2 Region based combining process and §3.5 §4.4, §4.5,
related algorithms §7.2, §8.2
H1.3 Reverse Glue Mold Construction §3.6 §4.4, §4.5,
method and related algorithms §7.2, §8.2
H2. Design-for-manufacture method for Chap 5 Chap 5
Rapid Tooling Chap 6 Chap 7
Chap 8
H2.1 Material geometric tailoring formulation Chap 5 §5.5, §7.3, §8.3
and solution
H2.2 Design for Rapid Tooling problem §6.1, §6.2, §7.3, §8.4
formulation §6.3
H2.3 Design for Rapid Tooling solution §6.4, §6.5 §7.4, §8.5
Process
1.3.2 Validation Philosophy and Strategy

According to Simpson (1998) the validation process for engineering design "is to
show the research and its products to be sound, well grounded on principles of evidence,
able to withstand criticism or objection, powerful, convincing and conclusive, and
provable." The validation of a design method can be difficult and therefore a validation
strategy needs to be developed carefully. Referring to a validation strategy developed by
(Pedersen, 1999), Siddique presented a validation strategy in his dissertation (Siddique,
2000) to test a “Product Family Reasoning System” (PFRS). The validation strategy is
sound, and therefore is adapted in this dissertation to test RTMDS and DFRTS.
The argument of this dissertation is embodied in a structure characterized by
research questions, hypotheses, and hypothesis-testing (the Scientific Method). In the
context of Natural Science, research questions refer to observations (articulating the
‘truth’), hypotheses refer to explaining the observations (understanding the ‘truth’), and
hypothesis-testing refers to validating the explanation (accepting knowledge about the
‘truth’). Hence, in the context of Engineering Design, hypothesis-testing becomes the
vehicle by which new scientific knowledge is accepted and added to the current pool of
knowledge. This ties the research validity discussion strongly to a fundamental problem
addressed early in epistemology and later in the philosophy of science: what is scientific
knowledge, and what constitutes confirmation of a knowledge claim? Pedersen (1999)
discusses the answer to this question from two major philosophical schools: the
Reductionist / Formalist / Foundationalist school and the Holistic / Social / Relativist
school.
26
Research validation is asserted to be a process of building confidence in its

usefulness. As stated the purpose of this research is to develop the “Rapid Tooling Mold
Design System” (RTMDS) and “Design for Rapid Tooling System” (DFRTS) to facilitate
producing functional prototypes for a set of part designs, which may have wide varieties
of geometries and requirements. The usefulness of the RTMDS and DFRTS is associated
with whether the method is effective and efficient. Pedersen (1999) partitions the
validation process into (1) Qualitative Process and (2) Quantitative Process.
Structural Validation – A Qualitative Process (Adapted from Pedersen, 1999)
Effective is referred to as being structurally valid. In context of design method
validation, this implies three things. It implies (1) accepting the ‘correctness’ of the
individual constructs constituting the method; (2) accepting the internal consistency of
the way the constructs are put together in the method; and (3) accepting the
appropriateness of the examples / case studies that will be used to verify the performance
of the method. The validity of the method constructs – individually and integrated –
deals with the structural ‘soundness’ of the method in a more general sense, and is
therefore denoted as Theoretical Structural Validity. The validity of demonstrating the
method with the chosen case studies deals with the structural ‘soundness’ of the method
for some particular instances, and is therefore denoted Empirical Structural Validity, and
both ‘validity’ are evaluated qualitatively.
Performance Validation – A Quantitative Process (Adapted from Pedersen, 1999)
Efficient is referred to as being valid from a performance perspective. In the context
of design method validation, this implies three things. It implies (1) accepting that the
outcome of the method is useful with respect to the initial purpose; (2) accepting that the
hypothesized constructs are contributing to making the outcome useful; and (3) accepting
that the usefulness of the method is beyond the case studies. The validity of the method
performance and the outcome performance which is based on documenting the usefulness
for some particular instances, is denoted Empirical Performance Validity. Similarly, the
validity of the method performance and the outcome performance which is asserted
beyond some particular instances, deals with generality and is therefore denoted
Theoretical Performance Validity.
Theoretical Theoretical
Structural Performance
Validity Validity
Empirical Empirical
Validity Validity
Figure 1.15 -Validation Square (Pedersen, et al.,, 2000).
27
To perform the validation in a systematic manner Pedersen (1999) introduced a

"validation square" (Figure 1.15). The validation square is applied in this dissertation to
test the hypotheses. More specifically:
(1) Mathematical proofs and constructions of tools used in different steps of the
RTMDS and DFRTS will be used to provide structural validation.
(2) Partial performance validation of the hypotheses will be performed by
determining the usefulness of the different mathematical tools on simple
examples.
(3) Further performance verification of the RTMDS and DFRTS will be provided
using integrated case studies.
In the mold design problem, math models of basic elements and their combining
process will be developed. Some of the properties of these math models will be formally
proven to provide theoretical structural validation of RTMDS. Mathematical
construction of the tools along with some of the well-matured mathematical concepts,
used in both mold design and DFM problems, will provide partial structural validation of
the hypotheses.
Simple examples are used to test the usefulness of the mathematical tools and the
algorithms developed for mold design and DFM problem of Rapid Tooling. Although
these examples will be simple, results obtained from applying developed tools and
algorithms will be used to validate the performance of RTMDS and DFRTS.
Two case studies are implemented to further validate the performance of RTMDS
and DFRTS, and to validate and verify this research. The two case studies are:
• A robot arm design
• A camera roller design
Table 1.2 - Validation of Hypotheses 1 and 2 addressed in different chapters

Chap 3 Chap 4 Chap 5 Chap 6 Chap 7 Chap 8 Chap 9
Hypotheses 1
Theoretical X X
Performance Validity
Theoretical Structural X X
Validity
Empirical Structural X X X
Validity
Empirical X X X
Hypotheses 2
Theoretical X X
Theoretical Structural X X
Validity
Empirical Structural X X X
Validity
Empirical X X X
Validity of hypotheses 1 and 2 is addressed in different chapters of the dissertation,

which are summarized in Table 1.2. With the validation strategy for this dissertation
presented in this section, the contributions from this research are presented.
28
1.3.3 Contributions from the Research

Rapid Tooling technologies enable the injection molding of near production-
representative prototypes (accuracy and material properties, etc.) to be made within less
time and in a lower cost. But the current state of research does not address the issue of
how to reduce the lead-time in the usage of Rapid Tooling. The basic idea of this work is
that by using mold design methods, mold inserts for a given part can be designed quicker;
and by using design-for-manufacture methods, the time of iteration between designer and
manufacturer can be reduced. The hypotheses and sub-hypotheses, taken together, define
the research presented in this dissertation and hence the contributions from the research.
Contributions related to Computer-aided Mold Design:
• A problem definition for multi-piece mold design and a solution process for the
formulated problem.
• The notions of concave region and convex face in mold configuration design and
a means of generating them for a given part.
• A combining approach to generate regions from the basic elements based on
linear programming and mold design knowledge.
• A mold construction method based on the reverse glue operation and a geometric
reconstruction problem for any mold base.
Contributions related to Design-for-Manufacture:
• A demonstration of transferring part of DFM responsibility (geometric tailoring)
to the manufacturer is feasible for Rapid Tooling.
• The development of design decision templates which are used as the digital
interface between the designer and manufacturer for Geometric Tailoring
problem.
• The development of an integrated system for Design for Rapid Tooling problem
and a solution process with three stages.
Contributions related to Rapid Tooling:
• A tool to aid the mold designer in designing molds for a part design.
• An integrated system to aid the manufacturer to change part design and process
planning according to design and manufacturing requirements.
• Several industrial cases are investigated in the system which is important
components of the Rapid Tooling TestBed.
This being the first chapter of the dissertation, these contributions cannot be
substantiated; therefore, they are revisited in Section 9.2 after all of the research findings
have been documented and discussed. An overview of the dissertation is presented next.
29
1.4 OVERVIEW OF DISSERTATION

To facilitate this discussion, an overview of the chapters in the dissertation is shown
in Figure 1.16. Having been lain the foundation by introducing the research questions
and hypotheses for the work in this chapter, the next chapter contains a literature review
of research related to different product family avenues. Research areas that are reviewed
include: (1) mold configuration design methods (Section 2.2); (2) mold construction
methods and tools (Section 2.3); (3) CAD representation (Section 2.4); (4) Design-for-
Manufacture strategies and approaches (Section 2.5); and (5) design technologies
(Section 2.6). A discussion on how these concepts relate to the overall objective of the
Chapter 1
Functional Prototype Using Rapid # Intorduction, Motivation, Foundation
Identification
Tooling # Research question, hypotheses
Problem
# Review different mold design methods
Chapter 2 from existing literature
A Literature Review: Mold Design # Review different DFM methods and
and Design-for-Manufacture solving techniques
# Present technological foundations
Chapter 3 # Problem definitions and steps of MPMDM

Automatic Design and # Basic elements and generation method
Construction of Multi- # Combining critera, process and algorithms
#
Development of Method
piece Molds: Methods Mold construction method and algorithms
# Implementation, limitation and test

Chapter 4 examples
RTMDS and its Usages # Provide proof of concept and initial
verification of method
# Modeling of Geometric Tailoring
Chapter 5
# Design decision templates with design
Formulating Design Requirements
representation
as Geometric Tailoring
# Proof of concept and initial verification of
method
Chapter 6 # Modeling of DFRTS

Design for Rapid Tooling: A # MPGT decision template and problem
Decision-Based Method # A solution strategy and process for DFRTS
# Demonstrate usage of MDRTS and

Testing and
Verification
Chapter 7 DFRTS
Functional Prototypes of a Robot Arm # Provide verification of methods
# Demonstrate usage of MDRTS and

Chapter 8 DFRTS
Functional Prototypes of a Camera Roller # Provide further verification of methods
Closure
# Summarize research findings,

Chapter 9 contributions, and limitations
Closing Remarks # Identify future work
Figure 1.16 - Overview of Dissertation Chapters
30
dissertation is presented in Section 2.1.

The general objective and steps of Multi-Piece Mold Design method are presented in
Chapter 3. Section 3.1 gives an overview of the problem definition, highlighting the
issues and design steps associated with the method. The basic elements of MPMDM and
an approach to generate them are presented in Section 3.2. The criteria and algorithm for
combining the basic elements into regions are discussed in Section 3.3. Finally, a mold
construction method based on the reverse glue operation and glue faces reconstruction is
presented in Section 3.4.
In Chapter 4 the implementation and initial testing of the mold design methods and
algorithms are discussed. The overview of RTMDS including software tools, modules
and limitations are presented in Section 4.1. Two simple test examples are discussed in
Section 4.4 for the partial testing of the system. Three more complicated industrial cases
are presented as the further verification of the RTMDS in Section 4.5 followed by a
summary of the system presented in the chapter (Section 4.6).
Material geometric tailoring problem for Rapid Tooling is presented in Chapter 5.
The properties of Rapid Tooling are discussed in Section 5.1. Related to the principle of
functional testing and similarity methods, the fundamentals of geometric tailoring are
presented in Section 5.2. Material geometric tailoring and MGT decision template are
discussed in Section 5.3. The usage of the MGT decision template, including formulating
function properties and solution approaches, is presented in Section 5.4. Three test
examples are discussed in Section 5.5 for the partial testing of the method.
After the material geometric tailoring study is completed in Chapter 5, the design for
Rapid Tooling problem, which consists of geometric tailoring, mold design and process
planning, is discussed in Chapter 6. An overview of the DFRTS is presented in Section
6.1. Software tools related to the modules of the system are introduced in Section 6.2.
The problem formulation of design for Rapid Tooling is presented in Section 6.3. Based
on the solution process for the DFRTS described in Section 6.4, a comparison of the
current usage of RT and DFRTS is provided in Section 6.5.
Application of RTMDS and DFRTS to a robot arm and a camera roller to illustrate
the use of the method in mold design and DFM is presented in Chapter 7 and 8
respectively. For each case study, an overview of the problem is given along with
pertinent analysis information, the steps of the methods are performed, and the
ramifications of the results are discussed.
Chapter 9 is the final chapter in the dissertation and contains a summary of the
dissertation, emphasizing answers to the research questions and resulting research
contributions in Sections 9.1 and 9.2, respectively. Limitations of the methods and
possible avenues of future work are discussed in Section 9.3 and Section 9.4. Finally,
some closing remarks are given in Section 9.5.
A pictorial overview of the dissertation is illustrated in Figure 1.17. It proceeds from
bottom to top, beginning with the foundation provided in this chapter: Computational
Geometry and Decision-Based Design. This figure provides a road map for the
dissertation, and it is referred to at the end of each chapter to help guide the reader
through the work as the research progresses from chapter to chapter.
31
Chapter 9: Achievements and Recommendations
Chp 7: Prototypes of a Chp 8: Prototypes of a

Robot Arm Camera Roller
Part Design
Parametric
Part Design
CAD Model
Requirements
of Part
A. Rapid Tooling Mold

Design System
Parting Directions
(Chp3 & 4) Parting Lines
(Section 6.3)
Parting Surfaces
Mold Piece Number
The Designer's MPGT
Problem Formulation
Given
Parametric CAD
Part Design
Models of Mold Pieces Find
(Section 6.2.1) (Section 6.2.2) Design Parameters
Satisfy
C. Injection Molding Constraints
B. RP Process Planner Goals
Process Analyzer
Surface Finish Draft Angle Minimize
Accuracy Rib Height/width Ratio Deviation
Cost Part Thickness
Time Mold Life
The RP Process The IJM Process

Compromise DSP Compromise DSP
Given Given
Mold Design Mold Design
Find Find
RP Process Parameters IJM Process Parameters (Section 6.2.3)
Satisfy Satisfy
Constraints Constraints E. Rapid Tooling
Goals G oals
Minimize Minimize Cost Predictor
Deviation Deviation Time
Cost
(Section 6.4 D. Coompromise

& 6.5) DSP Solver
Input and Output
Processor
Tailored Part Design and
related Mold Design, RP and
C-DSP Template
IJM Process Parameters
Chp 4: RTMDS and its Usage

Chp 6: Design for Rapid Tooling
Part P Part P
F1 F9 F1
F9
F3 F3 Given
F2 (1) (2)
F6 F8 F6 F8
An alternative to be improved throughmodification; Assumptions usedto modelthe domainof interest.
F2 The systemparameters:
n number of systemvariables p+q number of systemconstraints
p equality constraints q inequalityconstraints
Fn Fn
F4 F5 F5
m number of systemgoals Gi(X) systemconstraint function
fk(di ) function of deviation variables to be minimizedatpriority level k for the preemptive case.
F4 Find
Values for the systemvariables Xi i = 1, ... , n
F7 F7 Values for the deviation variables - +
di , di i = 1, ... , m
Satisfy
Systemconstraints (linear, nonlinear)
Part P gi(X) = 0 ; i = 1, ..., p gi(X) ≥ 0 ; i = p+1, ..., p+q
Systemgoals (linear, nonlinear)
PD1
Mold Base Bounds
Ai(X) + di- - di+= Gi ; i = 1, ..., m
F1 R3 Ximin ≤ Xi ≤ Ximax ; i = 1, ..., n

R1 F9 M1 F1 Deviationvariables
F3 F3
di-, di+ ≥ 0 ; di-. di+ = 0 ; i = 1, ..., m
(3) PD1 Minimize
F2 F8 F2 Preemptive deviation function(lexicographic minimum)
PL1
F6 Z=[ f1(d-i,d+i ),..., fk(di-, di+ )]
Archimedain deviation function
Z= ∑Wi(d−i + d+i ) where ∑W =1, W ≥0
F5 Fn M2 F5 Mk i i
F4
F4 PL2 Rk F7
PD2
Fn
R2 F7
PD2
Chp 3: Rapid Tooling Mold Chp 5: Geometric Tailoring

Design method
Mold Mold Construction Design

Configuration CAD DFM Strategies
Methods and Techniques
Design Methods Representation
Tools
§2.2 §2.3 §2.4 §2.5 §2.6
Foundations: Computational Geometry & Decision-Based Design
Figure 1.17 - Pictorial Overview of the Dissertation
32
Chapter 2 – A Literature Review: Mold Design and Design-for-Manufacture
CHAPTER 2
A LITERATURE REVIEW: MOLD DESIGN AND DESIGN-FOR-

MANUFACTURE
In this chapter a survey of relevant work in different aspects of mold design and
design-for-manufacture will be presented. In Section 2.1 a more thorough description of
how the research topics covered in this chapter are relevant to this dissertation is
provided. Mold configuration design methods proposed by different researchers for
determining parting directions, parting lines, parting surfaces, and undercut features are
presented in Section 2.2. Mold construction methods and commercial mold design
software systems are investigated in Section 2.3 for constructing mold pieces. A review
of representation and manipulation methods of solid models is presented in Section 2.4.
High-level CAD representation methods are also discussed in this section to provide a
background for the decision templates in Chapter 5. Different DFM metrics and methods
are presented in Section 2.5. The difficulties associated with these methods are also
highlighted. Three design technologies, the Decision-Based Design, Compromise DSP,
and Robust Concept Exploration Method, are reviewed in Section 2.6. They provide
foundation for the design for Rapid Tooling system in Chapter 6. In the final section of
this chapter (Section 2.7), a summary of the material presented and a preview of what is
next are provided.
33
2.1 TOPICS IN CHAPTER

In Section 1.2 motivation to mold design and design-for-manufacture (DFM) was
given along with the foundations of the research approaches used in this dissertation. In
this chapter a literature review on different topics related to mold design and design-for-
manufacture will be presented (Figure 2.1). Topics covered are:
(1) Mold configuration design methods: Methods used to determine parting
directions, parting lines, parting surfaces, and undercut features, have been
presented by different researchers. They are reviewed in this section as a starting
point for developing a multi-piece mold design method. This topic is related to
Hypotheses 1.1 and 1.2.
(2) Mold construction methods and tools: Methods used to construct mold pieces
have been identified and reviewed. Several commercial mold design software
systems are also investigated on their splitting approaches. This topic is related to
Hypothesis 1.3.
(3) CAD representation: Both the representation and manipulation of solid models
are reviewed. They provide a base for developing a computer-aided mold design
method and related algorithms. The solid modeling approaches are related to
Hypotheses 1.1 to 1.3. In this section high-level CAD representation methods are
also investigated to provide a background for decision template, which is related
to Hypothesis 2.1.
(4) DFM strategies and techniques: Both process-oriented methods and information-
orientation methods are reviewed. The DFM approaches in VLSI (Very Large
Scale Integrated Circuits) design are also investigated to provide a background
for the DFM system developed in this research. The DFM approaches provide a
background for Hypothesis 2.1 and 2.2.
(5) Design technologies: Several design technologies, including the Decision-Based
Design, Compromise DSP, and Robust Concept Exploration Method, are
reviewed. They provide foundation for the design for Rapid Tooling system,
which is related to Hypotheses 2.2 and 2.3.
H1 (H1.1, H1.2, H1.3) H2 (H2.1, H2.2, H2.3)

Methods and Technologies
Tools
§2.2 §2.3 §2.4 §2.5 §2.6
Figure 2.1 - Literature review topics and relationship with Hypotheses
34
With the importance and relationships of the topics covered in the literature review
highlighted, Mold configuration design methods are discussed in the next section.
2.2 MOLD DESIGN METHODS AND ALGORITHMS
The methods by which information is extracted from the model for use by the mold
design software can be collectively termed geometric reasoning. There appears to be no
definitive accepted definition of geometric reasoning, although the term has been used for
the last fifteen years or so. Woodwark (1989) explained it in the following manner.
While computer graphics and image processing have been with us for thirty years or
so, and look like mature disciplines, it is really only the algorithmic processes that
have been computerized. There are many more complex, less easily pinned-down
things that are still want to do with shape information and cannot. In the organization
of this conference, we called these activities Geometric Reasoning.
Jared, et al., (1994) give a refined definition of Geometric Reasoning as “the process
by which application requests related to geometric properties and attributes are satisfied
from a geometric properties and attributes are satisfied from a geometric model using
both inferential and algorithmic method”.
Geometric reasoning is one method of providing answers for solving variety of
geometries in design for Rapid Tooling. In this and the next section some geometric
reasoning methods on automatic mold design are presented based on literature survey.
The automation of mold design for injection molding process is studied in many
publications. However, works scatter in the determination of parting direction, parting
line, parting surface, and undercut detecting individually. In this section, the reviewed
works are arranged according to their focused area in the mold design. The definitions of
parting direction, parting line, paring surface and undercut are given in mold design
handbooks (Rosato and Rosato, 1995), and hence not repeated here.
2.2.1 Parting Direction

In the determination of parting direction, there are mainly three kinds of approaches.
(1) Compute the exact global accessibility cones based on V-map.
A representative problem formulation of this approach is given by Chen, et al.,
(1993). It formulates demoldability as a visibility problem. A notion of visibility map
(or V-map) of surfaces is presented to define the possible parting direction for pockets of
the given part, which are some spherically convex polygons. By computing the
intersection of V-maps, the problem of finding a pair of parting directions that minimizes
the number of cores is transformed to find a pair of antipodal points p and –p that
maximize the number of V-maps which contain either p or –p.
Based on this formulation, several papers presented other approaches for the
selection of parting directions of mold (Weinstein and Manoochehri 1996; Vijay, et al.,
1998). Visibility map is also used in other areas like NC machining (Woo, 1994; Gupta,
et al., 1996). The basic element of the approach is a “pocket”, which can be generated by
subtracting a part from its convex hull, or by testing adjoining surfaces as shown in
(Weinstein and Manoochehri, 1996).
35
(2) Compute the exact global accessibility cones based on automatic molding feature
recognition.
Gu, et al. (1999) use a universal hint-based feature recognition algorithm to
recognize all features of a molded part. Their features included holes, steps, pockets,
protrusions, etc. Different feature types have their candidate parting directions (CPD)
stored in the system. Finally each CPD is evaluated using an object evaluation function
and the CPD with maximum evaluation value is selected as the optimal parting direction
of the part.
Fu, et al. (1999) classify undercut features (features that prevent the removal of a
part from the molds along the parting directions) as Inside Internal Undercut, Outside
Internal Undercut, Inside External Undercut, and Outside External Undercut. Based on
the undercut feature characteristics and geometric entities (three-edge, four-edge and
more than four-edge), algorithms to recognize them are presented in the paper.
More recently, Yin, et al. (2001) present their approach to construct core and cavity
based on the recognized undercut features. For a given part, a volume-based feature
recognition method using non-directional blocking graph is developed to recognize
undercut features. Then the optimal parting direction is determined by minimizing the
number of undercuts for different candidate parting directions.
Krishnan (1997) describes automated two-piece and multi-piece mold design for
injection molding. The part is constructed by stacking 2.5D primitives called C-entities
along the Z direction through either a Constructive Solid Geometry (CSG) or Destructive
Solid Geometry (DSG) operation. Since the primitives considered are only 2.5D solid
that are stacked along the Z direction, the complexity of the part is limited. The parting
surface directions are also constrained to be along the X-axis or Y-axis direction.
Dhaliwal, et al. (2000) described a feature-based approach to automated design of
multi-piece sacrificial molds which are to be manufactured by CNC machining.
However, how to separate mold from the part is not considered since the author
considered only sacrificial molds, which can be destroyed after the part has been
produced.
(3) Compute approximate global accessibility cones by sampling a set of discrete
directions.
Hui and Tan (1992) heuristically generated candidate parting directions from normal
vectors of planar faces and from center-lines of holes and bosses. Two criteria, the
blockage factor and the performance value, are used to determine the main parting
direction and subsidiary parting direction. To evaluate the geometry of an undercut, Hui
(1996) developed a partitioning scheme to subdivide the cavity solid of a component
along a given direction. In the search for main and side core directions, the search space
is the set of all normals to individual faces of the object and the opening of cavity solid.
A search tree is built for side core selection.
Urabe and Wright (1997) selected three principal coordinate directions as the
candidate parting directions, then calculate the number of undercuts, the projected area,
the number of cone surface for each candidate direction and used them as major mold
36
factors to determine the main parting direction. This approach was limited to simple 3D
parts.
Lu and Lee (2000) proposed an approach for analyzing the interference element and
release direction in die cast or injection molded components. First a three-dimensional
ray-detection method is used to recognize and extract the interference elements. Then
distribution of the release directions can be computed, and the candidate release
directions can be prioritized based on the minimization of the number of side cores.
Review of Methods:
Three kinds of approaches are reviewed in this section. It is well know that feature
recognition of a part is rather difficult, especially as feature interactions are rather
important in the multi-piece mold design. Sampling a set of discrete directions cannot
guarantee that a suitable mold design will be find. Therefore this research will utilize the
approach of computing the exact global accessibility cones based on V-map. However,
the basic elements used in this research are different from those of the reviewed
approaches.
2.2.2 Parting Line

There are few published works on the determination of parting lines. Pye (1989)
suggested that the parting lines should be around the position of maximum dimension of
a product when viewed in the draw direction of an injection molding process. The basic
element in judging parting lines is part faces in all the approaches. Also in the
approaches, either the parting direction is already set, or the parting direction will not
cause any undercuts (because of convex polyhedra).
Tan, et al. (1988) proposed a parting line generation method for a triangular sub-
division of the product model’s surfaces. In the method, a draw direction is selected first.
Then, the algorithm triangulates and classifies the surfaces into visible faces and invisible
faces. Since the parting lines are located in the outermost and the innermost boundaries
of the product model when viewed in the draw direction, they are those edges on the
product model separating the visible faces from the invisible faces. The limitation is that
it does not apply to models with non-drafted faces (faces parallel to the mold opening
direction).
Ravi and Srinivasan (1990) proposed the sectioning and silhouette methods for
parting line generation. The section method locates the parting line by computing the
intersection of the product model with a plane normal to the draw direction. The
silhouette method is capable of dealing with non-planar parting surfaces. It projects the
product model onto a plane normal to the draw directions to obtain the projected
boundary of the product. The projected boundary is then swept back along the draw
direction. The intersection between rays from the projected boundary and the surfaces of
the product model form portions of the parting line.
Serrar (1995) developed a semi-automatic method for selecting parting loops. In this
method, the user first selects an edge from the parting loop. Then the system finds the
branches (adjacent edges), and successively highlights these branches until the user
confirms which branch belongs to the parting loop. This process is repeated until the
loop is closed.
37
Wong, et al. (1996) used a slicing strategy to locate the parting lines of a product
model along a draw direction. A recursive uneven slicing method is developed to locate
several parting surface for further evaluation. The approach is primarily proposed to deal
with free-form surfaces in product design.
Majhi, et al. (1999) discussed the problem of computing an undercut-free parting line
that is as flat as possible in mold design for a convex polyhedron. Two flatness criteria
for parting lines are given and algorithms are presented to compute a parting line based
on the criteria.
Review of Methods:
In the reviewed approaches, the parting directions have already been determined.
Therefore the parting lines are determined only for the given parting direction. Also only
two mold pieces were considered in these approaches. In this research, a mold design
method is developed in which parting lines and parting directions are determined at the
same time. The method is also suitable for multi-piece mold design.
2.2.3 Parting Surface

Parting surface is the surface separating the mold halves. Ganter and Truss (1990)
introduced a method for locating parting surface based on a set of evenly distributed
points lying on the surface of a sphere enclosing the object.
Ravi and Srinivasan (1990) presented a comprehensive decision model for selection
parting surfaces. A total of nine criteria were used, including projected area, flatness,
draw distance, draft, number of undercuts, volume of flash, and dimensional stability.
They presented algorithms for most of these criteria, but did not present deterministic
algorithms for generating a variety of candidate parting surfaces or parting directions.
Although it was not automated, the comprehensiveness of their approach aids designers
in decision making.
Tan, et al. (1988) presented a parting surface generation method by projection. After
getting the parting lines for a given parting direction, the algorithm creates planar parting
surface elements for each edge of the outer loop generated by a convex hull algorithm.
Therefore the whole parting surface of the mold is segmented and not in one plane
anymore. This approach is simple, however, the results do not fit with the molding
heuristics, that is parting surface is better to be planar to increase the shut-off force.
Nee and coauthors present the core and cavity creating approach of IMOLD@ (Nee,
et al., 1998; Nee, et al., 1999). After the parting lines are determined, all the parting
edges are classified into different groups and their extruding directions are determined.
The parting surfaces are the unions of the surfaces by extruding the parting edges to the
boundary of the core and cavity in directions perpendicularly outwards to the parting
direction.
Review of Methods:
Although several methods have been proposed for the determination of parting
surface from parting lines, none of them explored theoretic foundations of generating
parting surfaces. In this research, the generation of parting surface (in this dissertation
they are called glue faces because of the reverse glue operation) is transformed into a
38
geometric reconstruction problem. A formal problem formulation and several algorithms

are also presented.
2.2.4 External and Internal Undercut Detection

Undercut is one of the most important considerations in the mold design. Several
methods are developed for their recognition from given CAD models.
Hui and Tan (1992) choose points on the edge of a solid as a set of test points. These
test points are then checked for obstruction in the direction considered. A technique that
is similar to that of hidden-line removal is used. A semi-infinite ray that originates from
the point is cast in the parting direction under evaluation. A series of test points are then
generated on the ray to decide if the point is obstructed.
Rosen (1994) focused on finding undercuts and determining whether or not they are
external or internal. The shadow casting approach is used to find undercuts. However,
no parting lines are considered. So the results are actually only potential undercuts.
Shin and Lee (1993) developed a procedure to identify the interference faces
between the product and the mold from the core plate (or cavity plate) alone without
considering the product. Based on the results, an algorithm based on Euler operation is
developed to generate the side cores and the corresponding core and cavity plates of a
mold. In the approach, core plate need to be specified by a designer and glued to the
generated cores manually.
Fu, et al. (1999) presented undercut feature definition, classification, parameters and
the recognition criteria. It classified undercut feature as Inside Internal Undercut, Outside
Internal Undercut, Inside External Undercut, and Outside External Undercut. Based on
the undercut feature characteristics and geometric entities, algorithm for classification
and recognition are presented.
Stefano (1997) propose an approach based on the analysis of the topology of concave
object parts to identify and extract features from a solid model representation for casting
process. Two topological invariants are defined for feature classification. Extracted
features can then be used for automatic core pattern development and mold design.
Review of Methods:
Undercuts are main considerations in the determination of mold design. In this
research they are considered in the combination criteria in the mold design process.
Therefore no feature recognition algorithms are considered, which saves a lot of work.
2.2.5 Synthesis Approaches of Basic Elements

As stated in Section 2.2.1, the approaches for determining parting directions used
different basic elements, which include pockets (Chen, et al., 1993; Weinstein and
Manoochehri, 1996; Vijay, et al., 1998), features (Gu, et al., 1999; Lu and Lee, 2000;
Yin, et al., 2001), and faces (Hui and Tan, 1992; Urabe and Wright, 1997). All the
approaches use a similar synthesis approach to evaluate a parting direction. That is,
(1) For each candidate direction dk, V(dk) = ∑ weight_factori × mold_factori;
i
39
(2) Parting direction d: V(d) = Min [V(dk)].

In Step (1), mold factors are some rules used to determine the parting directions,
which may include number of undercuts, projected area, draft angles, etc. Weight factors
distinguish the importance of one mold factor relative to another. The assignment of
these weight factors is based on how the corresponding mold factor affects the cost,
quality, and productivity of the mold.
Depending on how to evaluate mold factors, there are two kinds of approaches.
(i) Evaluate mold factor for each pocket/features.
In (Chen, et al., 1993), candidate directions are the directions computed from the V-
map of pockets. Only one mold factor, number of undercuts, is considered. So the mold
factor for each candidate direction is actually the number of V-maps that do not contain
the candidate direction. In (Weinstein and Manoochehri, 1996; Lu and Lee, 2000; and
Yin, et al., 2001), a similar approach is used in evaluating parting directions. Besides
number of undercuts, Gu et al. (1999) also considers the projected area and thickness of
the molded part, which are evaluated based on the bounding box of a part.
Since face connectivity and faces that do not belong to pockets or features are not
considered in these approaches, they can explore the combinations of pocket/feature
rather quickly, usually with the aid of visibility map. However, this may cause problems
in constructing mold pieces for non-connected faces. Also if we want to consider more
mold design knowledge, we need to add more mold factors. Since several mold factors
like minimal draft angle are also related to faces that do not belong to pockets or features,
it is not clear how they should be added.
(ii) Evaluate mold factor for all faces.
In this approach, mold factors are evaluated for all faces. So more mold factors can
be considered. Urabe and Wright (1997) consider boxed area, projected area, number of
no-hidden faces, number of undercuts and cone surfaces. Hui and Tan (1992) consider
number of undercuts and projected area. A more comprehensive decision model is
presented in (Ravi and Srinivasan, 1990) for the selection of parting surfaces for casting
parts. A total of nine factors were used, including projected area, flatness, draw distance,
draft, number of undercuts, volume of flash, and dimensional stability. Related
approaches and expressions for most of these criteria are also presented.
Since it is quite time-consuming to test these mold factors for all faces, only a small
number of candidate parting directions or parting surfaces are considered in these
approaches. Urabe and Wright (1997) consider the principal axis directions as the
candidate directions. Hui and Tan (1992) consider the normals of planar surfaces and the
axis of a cylindrical or helical surface as the candidate directions.
Review of Methods:
In this research the first synthesis method is used. However, the basic elements used
in the mold design process are different from pockets or features, as discussed in Section
3.4.2. In the synthesis process mold design knowledge is also considered in this research.
40
2.2.6 Intersection of V-Map

Woo (1994) defines “Visibility Map (V-map)” from Gauss Map to formulate the
visibility problem. For a planar surface F, the V-map of F is a hemisphere centered on
the unit outward normal. So the V-map of a region is the range formed by the
intersection of the allowable draw ranges for each individual surface (Figure 2.2.a). By
calculating the intersection of V-maps, the range of feasible solutions can be represented
as a subset of a spherical surface.
Feasible withdrawal
directions Draw-range
CR
(a) V-map and draw-range
(b) A Case given in (Dhaliwal et al., 2000)
Figure 2.2 – Two Examples of V-map.

Since all V-maps and their intersections are calculated on spherical surfaces, several
approaches and algorithms based on spherical polygons have been presented for different
applications (Chen and Woo, 1992; Woo, 1994; Gupta, et al., 1996; Kweon and
Medeiros, 1998). Using the central projection to convert a spherical problem to a planar
one, Chen and Woo (1992) present four spherical algorithms, which include detection of
convexity on the sphere, computation for spherical convex hull, determination of the
spherical convexity of a union, and the intersection of hemispheres. They are used in
numerical control (NC) machining planning. Kweon and Medeiros (1998) utilize V-map
to represent accessible directions for measurements of tolerance. The concept of V-map
dimensionality is proposed to provide a method of clustering V-maps. Related algorithms
and data structures are presented and tested for CMM inspection.
41
The algorithms and related data structures to calculate the intersection of V-maps on
spherical surfaces are rather complicated. Typically, computations take long time
especially for a complex part. As an example, most recently Dhaliwal, et al. (2000)
present an algorithm for computing exact global accessibility cones for various faces of a
polyhedral object. One example given in the paper, which is shown in Figure 2.2.b,
would take 77 minutes on a Sun Ultra10 workstation.
Balasubramaniam, et al. (2000) developed a method by taking advantage of
computer graphics hardware to generate 5-axis roughing tool paths directly from a
tessellated representation of a body. Graphics cards make use of the depth-buffer
implemented using hardware to perform fast hidden surface removal and render the
object in a given scene. If all the individual faces on the object have been assigned
different colors, then the accessibility of each face in a given direction can be detected by
rendering the object using the given direction as the viewing direction, and querying the
colors that appear on the pixel map after rendering.
Review of Methods:
The idea of V-map is also used in this research. However, the calculation of draw
ranges is omitted. Instead a parting direction is calculated directly from the given faces
by solving an optimization problem. The optimization problem is further simplified by
approximating the spherical surface by a set of triangles. Therefore the intersection of V-
maps of a set of faces can be determined by solving a linear problem. The approach
present in this research is much easier and quicker than the spherical algorithms
presented in the reviewed approaches.
2.2.7 Detection of Non-Drafted Surfaces

Surfaces parallel to the parting direction must be drafted at least a minimum draft
angle for injection molding process (Rosato and Rosato, 1995). In order to guarantee a
fully drafted model, it is critical to automate the detection of non-drafted surfaces.
Relatively few published works cover this topic. Serrar (1995) presents a non-drafted
surface detection algorithm for a given part. In his approach, the parting direction is
given as the input. So the only task is to find faces with outward normals forming an
angle between 90o-min_draft_angle and 90o+min_draft_angle with the parting direction.
However, since draft angle of a face is tightly related to a given parting direction and
parting lines, it is better to detect non-drafted surfaces in the determination process of
parting direction and parting lines.
Review of Methods:
In the reviewed approaches, the parting direction is given as the input, which greatly
simplifies the problem. In this research, the detection of non-drafted surfaces is
integrated with the determination of parting direction and parting lines, which is more
general and useful.
After the methods on computer-aided mold configuration design are presented, the
methods and commercial tools on mold piece construction are described in the next
section.
42
2.3 MOLD CONSTRUCTION METHODS AND TOOLS

After parting direction and parting lines are selected (Section 2.2), mold designer can
use general CAD software such as Pro/Engineer, SolidWorks, and CATIA to construct
mold halves manually. However it is a tedious process that takes a long time. Currently
two approaches are proposed for automatically splitting the core and cavity inserts for
two-pieces mold design.
2.3.1 Approach Based on Extending Parting Lines

A mold base can be cut into two pieces by a parting surface. For parting lines not in
a plane, Tan, et al. (1988) presents a parting surface generation method. After getting the
parting lines for a given parting direction, an outer loop and inner loops are generated. A
convex hull algorithm is applied to the outer loop. Each edge of the hull is projected to
an adjacent side face of the mould block. The projection direction is perpendicular to the
parting direction but parallel to the surface normal of the side face of the mould block.
All planar faces generated by projecting hull edges can form a parting surface for the
part. Shin and Lee (1993), Serrar (1995) and Nee, et al. (1998) use a similar approach to
form the parting surface to split mold base into two halves.
Review of Methods:
This approach is quite straightforward. Extending the given parting lines outward
into faces can split mold base into two mold pieces. However, for non-flat parting lines,
the parting surface generated by this approach is also not flat. This is not accordant with
the best practice of mold design, that is parting surface is better to be planar to decrease
the manufacturing complexity and to increase the shut-off force to reduce material flash
(Ravi and Srinivasan, 1990). For example, for a part as shown in Figure 2.3.a, one mold
piece generated by the approach of extending parting lines is shown in Figure 2.3.b, and
the mold piece generated by algorithm Two_Mold_Piece_Generation (Section 3.6.3) is
shown in Figure 2.3.c. Compared with the mold piece in Figure 2.3.b, the mold piece in
Figure 2.3.c is cheaper to fabricate because of less benchwork, and its parting surface has
higher accuracy and surface finish. Consequently we can expect less material flash in its
injected parts.
(a) A part (b) Mold piece 1 (c) Mold Piece 2

Figure 2.3 – A Part with Mold Pieces Generated by Two Approaches.
2.3.2 Approach Based on Sweeping

Conceptually removing an injected part from a mold in a parting direction is similar
to sweeping part faces in the same parting direction. Hui and Tan (1992) described an
43
algorithm using sweep operations and Boolean operations to generate mold core and
cavity. First sweeping the mold part in the parting direction to generate a solid. Then
using two mold plates to subtract each end of the solid can generate two mold pieces.
The algorithm does not consider the internal parting lines. So for a shape with a through
hole, the algorithm will not generate the desired mold pieces. Urabe and Wright (1997)
also presented a mold construction method based on the sweeping of FACE_STRUCT
into BODYs. Then they are united with plates and mold walls to form core and cavity.
Review of Methods:
The sweeping approach has two problems:
(1) The mold construction process takes a long time by sweeping each face and
doing Boolean operations with mold plates, especially for a complex part. For
example, for an industrial part with normal mesh size as shown in Figure 4.26,
there are more than 5493 faces. Among them if only half faces need to be
swept, it is quite easily to take several hours to generate the mold pieces by a PC
machine.
(2) It is well known that Boolean operation for coincident geometry is a very
difficult problem. Most CAD packages still cannot handle it properly,
especially for the subtraction of two bodies with many vertices, edges or faces
in exactly same positions. So although the sweeping approach is theoretically
feasible, one may meet problems related to Boolean operations in the
implementation of the approach.
2.3.3 Industrial Approach

Currently there are several commercial mold design software systems that can
automatically split core and cavity for a part. The author investigated six leading systems
as listed below. For each system, an example is given to show the splitting approach it
uses.
(1) MoldWizard
Unigraphics/MoldWizard is a highly automated product for designing plastic molds.
It incorporates industry best practices to guide users through the steps required to
construct a mold. Some brief steps for constructing core and cavity for an industrial part
are shown in Figure 2.4. First parting lines for the given part are identified by the
interactive input from designers (Figure 2.4.b). Then for each parting edges a sweeping
direction is specified and parting surfaces are generated (Figure 2.4.c). These parting
surfaces together with others generated from inner loops can cut mould box into two
mold pieces. Figure 2.4.d shows one of the mold pieces.
44
(a) An industrial part (b) Parting lines of the part
(c) Parting surface to do Boolean operation (d) One of mold pieces
Figure 2.4 – Mold Construction Process of Unigraphics/MoldWizard for an Industrial Part.
Figure 2.5 – An Example Mold Design Given by Magics RP.
45
(2) Magics RP
Materialise (www.materialise.com) develops a Rapid Tooling module in its Magics
RP system, which automates the design of the insert tool. Figure 2.5 shows an example
generated by the system.
(3) IMOLD
IMOLDTM is a supplementary program for Unigraphics and SolidWorks. It has a
module, Core/Cavity Builder, to handle the paring of cores and cavities for both solid and
surface product models. An example generated by the system is shown in Figure 2.6.
Figure 2.6 – An Example Mold Design Given by IMOLD.
Figure 2.7 – An Example Mold Design Given by Moldplus.
46
(4) Moldplus
Moldplus (www.moldplus.com) is a supplementary program for Mastercam®. It can
generate parting surfaces for given parting lines. Figure 2.7 shows an example of parting
surface.
(5) QuickSplit
QuickSplit is a splitting tool developed by Cimatron Ltd. (www.cimatron.com) that
can separate core and cavity with sliders and inserts. Figure 2.8 shows an example
generated by the system.
Figure 2.8 – An Example Mold Design Given by QuickSplit.
(a) Part (b) Mold Design
Figure 2.9 – An Example Mold Design Given by I-DEAS.
47
(6) VGXTM Core/Cavity Design

VGXTM Core/Cavity Design is the mold design model of I-DEAS. An example part
and its mold design are shown in Figure 2.9.
Nee (1999) presents the generation approach of parting surface used in IMOLD,
which is based on extending parting lines. For other systems, the author did not find any
published works related to their splitting approach. However, according to the mold
results generated by the systems, it seems that all the systems use the approach based on
extending parting lines.
Test reports for each system are not given by its software company nor found in any
published works. Hogarth (1999) describes a test done by Minco Tool & Mold (Dayton,
OH) to split a part with more than 5000 faces. The splitting process by
Unigraphics/MoldWizard took four hours. In the same article, Cimatron claims a 90%
saving of core/cavity splitting time by using Quick-Split in projects that used to take
20~40 hours. Although not the same part nor computer, an industrial part with 5493
faces, which is shown in Figure 4.27, was tested with a splitting approach to be presented
in Section 3.6. The splitting process took less than 1 minute.
Review of Methods:
According to the mold design results given by the companies, the approach based on
extending parting lines is used in all the reviewed software systems. Consequently the
disadvantages of the approach of extending parting lines (Section 2.3.1) can be found in
these systems.
2.4 CAD REPRESENTATION
The CAD representation of parts is a critical element in this research. In a computer
aided mold design method, a part design need to be represented by some data structures
and manipulated by some algorithms. Also the generated mold pieces for the part should
be represented in CAD models. Besides the shape of a part, the design features and
design intents are also critical in a design-for-manufacture system. The decision-template
proposed in this dissertation (Section 5.3) is actually a high-level CAD representation of
a part design. Therefore, the CAD representation of a part design is reviewed in this
section.
The representation and manipulation of 3-dimensional surfaces and volumes are
studied in the field of solid modeling, which is a critical element in a growing number of
application areas, including CAD/CAM, robotics, simulation, analysis and graphics.
Conceptually, work in solid modeling has proceeded in three levels of abstraction
(Hoffman, 1989):
1. Symbolic and arithmetic foundations represent the lowest level of abstraction and
are concerned with the computer hardware support of integer and floating point
arithmetic as well as with the abilities of a programming language to express
computations and manipulate memory. Examples include the development of
data structures and handling of issues of mathematical robustness.
2. Mathematical and algorithmic infrastructure describes the fundamental operations
as implemented in the foundation above. Examples include operations for
48
creating solid, for performing interference tests among solids, and for developing
efficient algorithms for basic manipulations.
3. Application and user interface is the highest level of abstraction, focusing on the
development of applications in terms of the infrastructure created by the above.
Mold design is primarily concerned with problems that lie in the second category.
Decision template is concerned with problems that lie in the third category. There are
numerous important concepts from the core solid modeling literature that are directly
relevant to the work described in this dissertation. The remainder of this section very
briefly reviews some solid modeling concepts and terminology which will be used
throughout this dissertation – in particular, concepts concerned with manipulation and
representation of solid models. Some related work on high-level CAD representation is
also provided.
For more in-depth coverage of the field of solid modeling, readers can refer to the
texts by (Mantÿlä, 1988; Hoffman, 1989; Mortenson, 1997; and Woodwark, 1989), as
well as papers of (Requicha and Rossignac, 1992; and Miller, 1989). The classic text on
computer graphics of (Foley, et al., 1990) also covers solid modeling and its relationship
to graphics and rendering. For information regarding commercial solid modeling
systems, readers can refer to the product and reference information for Spatial
Technologies ACIS modeler (www.spatial.com) as well as the EDS/UNIGRAPHICS
Parasolid solid modeling kernel (www.ugs.com).
2.4.1 Representation of 3D Surfaces and Solids

Solid modeling systems are able to distinguish between the inside and outside of the
object. This capability distinguishes solid modelers from wire frame models, which store
only enough information to describe the edges bounding an object. There are three broad
classes of schemes for representation of solid models (Mantÿlä, 1988):
1. Decomposition approaches that model a solid as collections of primitive objects
connected in some way. Examples of these data structures include Quadtree for
two-dimensional objects and octree for solid objects.
2. Constructive approaches that model a solid as a combination of primitive solid
templates. The most famous constructive approach is Constructive Solid
Geometry (CSG). CSG adopts the “building block” approach to solid modeling in
its pure form. The user of a CSG modeler operates only on parameterized
instances of solid primitives and Boolean set operations on them. The part is
stored in the database as a tree in which the leaves of the tree correspond to
primitives and the nodes correspond to the Boolean operations.
3. Boundary-based approaches that model a solid using a data structure that
represents the geometry and topology of its bounding faces. A face is bounded by
edges, and edges are bounded by two vertices. Parts are stored in the database as
a linked-list of vertices, edges, and faces. The database must store information
about the connectivity of the faces and the equations defining the geometry of the
faces.
49
Currently the boundary-representation (b-rep) approach has emerged as the dominant

representation scheme in solid modeling and CAD systems. This is due in large part to
the representational power and flexibility of boundary models, as well as to recent
advances in numeric computation that overcame earlier problems with models becoming
unstable and inconsistent.
The three object types (face, edge, and vertex) form the basic constituents of
boundary models. Correspondingly there are polygon-based, vertex-based and edge-
based boundary models. Among them, one of the most popular b-rep structures is the
Winged-Edge representation and its variations. A basic idea in this representation is the
introduction of coedges. A coedge records the occurrence of an edge in a loop of a face.
Coedges have partner pointers that point to the other coedge list associated with the edge.
An example of a simple part is shown in Figure 2.10. Normally an edge is adjacent to
two faces; therefore, the edge has two coedges, each associated with a loop in one of the
faces. The two coedges always go in opposite directions along the edges.
Figure 2.10 – Coedges of a Part.

The work described in this dissertation assumes a boundary-representation solid
model represented in winged-edge data structure. A b-rep usually consists of a graphical
structure that models an entity’s topology. The connections between the nodes in the b-
rep graph represent the connections between the topological components of the entity’s
boundary. These topology nodes then contain pointers to their underlying geometric
entities; for example, a face of a solid is a topological entity (represented as a collection
of bounding edges) and it has associated with its surface (represented as an equation). An
illustration of the distinction between geometric and topological information from the
ACIS Solid Modeling Kernel is given in Figure 2.11.
50
Figure 2.11 – The Distinction between Geometric and Topological Information within the
ACIS Solid Modeler (Spatial Technology, 2000).
In this research, only manifold solids are considered. Intuitively, a manifold solid is
one in which each point on the boundary of the object has a neighborhood that is
equivalent to a two-dimensional disk. Practically, this means: (1) each edge belongs to
exactly two faces; (2) each vertex is surrounded by one sequence of edges and faces; (3)
faces intersect at common edges and vertices only; and (4) there is volume on only one
side of a face. The manifold conditions also exclude solid whose bounding surfaces are
self-intersecting. Non-manifold objects can be of mixed dimension and may have
vertices, edges, and faces that do not meet these requirements.
Review of Methods:
The representation method of solid models is considered in developing data
structures for a mold design method. They also determine the capability of a mold design
system in manipulating solid models.
After the representation of a part is introduced, the manipulation of parts is presented
in the next section.
2.4.2 Manipulation of Solid Models

This section briefly describes some of the common operators used to manipulate
geometric and solid models. In addition to operators such as transformations, rotations,
51
and scaling, there are additional functions specific to the nature of geometric modeling.
These operations include Euler operations and Boolean operations, which are used in
Chapter 3 and 4 in this dissertation.
• Euler Operators
Based on the theory of plane models, Euler operators act on the topology of a
boundary representation data structure. Starting from an idealized primitive “solid”
consisting of one face and one vertex, one can create a solid through a series of local and
global manipulations. Local manipulations can create (or delete) vertices, edges, and
loops; global manipulations can be used to create (or delete) holes or to divide a body
into multiple bodies.
Gluing is a high-level Euler operator, which can combine simpler solids to more
complicated ones. For two 6-sided cylinders (Figure 2.12), they can be glued into one
along a pair of entirely coincident faces. The gluing operation consists three steps
(Mantÿlä, 1988): (1) the merging of two half-edge data structures into one data structure
that has two shells; (2) the joining of the shells; (3) the merging of coincident edges and
vertices of the face.
Figure 2.12 – An Example of Gluing Operation.

• Boolean Operations.
Solid models can be considered as bounded point sets. Boolean operations union
(∪), intersection (∩) and difference (-) can be defined on solid models based on their
action on point set. For non-manifold models we are interested in, regularized Boolean
operations correspond to the counterparts of the ordinary Boolean operations. If A is a
solid, i(A) is the interior of A (point set A minus its boundary) and c(A) is the closure of A
(point set A plus its boundary); the regularized Boolean operations are defined as follows
(Mantÿlä, 1988):
∪*, Union: A∪*B = c(i(A∪* B));
∩*, Intersection: A∩*B = c(i(A∩*B));
-*, Difference: A-*B = c(i(A-*B));
Examples of regularized Boolean operations are shown in Figure 2.13.
52
Figure 2.13 – Boolean Operations.

The Boolean operations can also be represented as gluing operation. From the
boundary classification of A and B, all their Boolean combinations are:
∪, Union: A∪B = A out B ⊕ B out A;
∩, Intersection: A∩B = A in B ⊕ B in A;
-, Difference: A-B = A out B ⊕ (B in A)-1;
where ⊕ denotes the gluing operation, and (B in A)-1Examples denotes the “complement”
of B in A, i.e., B in A with the orientation of all faces reversed.
Review of Methods:
Euler operators and Boolean operation are used in the mold design method to be
presented in Chapter 3. The notations reviewed in this section will be used in the
presentation of the mold design method and revisited in Chapter 3.
2.4.3 High Level Representations

As solid modeling has become better understood, increasingly sophisticated and
robust solid modeling systems have emerged. Given that adequate solid modeling tools
are available at one’s disposal, one begins to ask questions such as: How should
functionality be represented? In this section several high-level representation methods
are reviewed, which provide a context for the decision template to be presented in
Chapter 5.
Features and feature-based approaches have proven popular in a variety of
CAD/CAM application domains. The concept of a feature attempts to reason about
53
design and manufacturing activities by modeling the relationship between the local
geometric and topological configurations of a design and the higher-level abstractions. In
this way, semantic information can be conveyed along with the shape. Significant work
has been directed toward feature-based design and manufacturability evaluation of part
designs (Shah, 1991; Rosen, 1992; Regli, 1995; Shah and Mantyla, 1995).
Currently most researchers are convinced that no single set of features can satisfy the
requirements of every possible design and manufacturing domain (Regli, 1995).
Generally a feature is regarded as a functional entity that is meaningful in certain
domains. Different feature types exist. Some typical feature types include (1) form
features, which are portions of the geometry; (2) precision features, which are deviations
from nominal form, size or location; (3) technological features, which are non-geometric
parameters related to the function or performance; (4) material features, which related to
material properties; and (5) assembly features, which include part relative orientations,
interaction surfaces, and fits. However most feature-based models address form features
only (e.g. hole and rib).
Rosen and coauthors (1994) proposed a computational framework, which was called
Goal-Directed Geometry, for early design stages. Tools for parametric geometry,
variational modeling, and feature-based design were combined with a multi-objective
optimization code to provide robust support for parametric design problems, where
parameter values are desired that best meet a set of goals and constraints. Geometric and
engineering models of a design are combined into a multi-objective optimization
formulation in Compromise Decision Support Problem (DSP). This idea is extended in
this research into the field of design-for-manufacturing.
Recently integrating part design knowledge into a CAD model is a developing
direction of mechanical design automation modeling. Parametric Technology Corp.
(PTC), one of the world’s leading CAD companies, recently divided the mechanical
design automation modeling into five generations. Initially, two-dimensional drafting,
then three-dimensional wireframe modeling, and finally three-dimensional solid
modeling. The current fourth generation is the parametric feature-based modeling. The
fifth generation is proposed and named by PTC as behavioral modeling
(http://www.ptc.com/products/proe/bmx/index.htm). The basic idea of the behavioral modeling is
to capture intelligence within features in the design side. Capturing product-intent is
taken as a natural part of the engineering process, and then automatically builds virtual
prototypes that satisfy multiple objectives. It advances feature-based modeling to
accommodate a set of adaptive process features that go beyond the traditional core
geometric features. The intent and performance of the design are also modeled.
To illustrate the behavior modeling, an example of a golf club design is shown in
Figure 2.14 (www.ptc.com/products/proe/bmx/examples/golf_club.htm). "Forgiveness"
and "feel" are two important characteristics in golf club design. Forgiveness refers to the
sensitivity of a club to off center hits and is influenced by location of the center of gravity
in the club head. Feel refers to the ease with which a golfer can comfortably swing the
club and is influenced by "swingweight", the measurement of a club's weight distribution
about a fulcrum point which is 14 inches from the grip end of the club. To solve the golf
club design problem, analysis features are created to capture mass and the center of
gravity of the head. To gain design insight to improving the forgiveness characteristics of
54
Figure 2.14 – An Example of Behavioral Modeling (PTC).

the club, a User Defined Analysis with a field point is created in this example to measure
the twisting force, or torque, applied to the club face when a ball is hit off center.
Sensitivity studies are then performed to determine the effect of toe thickness of the club
head on the location of the center of gravity and twisting force. To gain design insight to
improving the feel characteristics of the club a sensitivity study is performed to determine
the effect of the flange thickness of the toe of the club head on the weight of the head. A
final design is then identified through optimization studies.
Zhu and Kazmer (1999; 2000) presented a performance-based representation
approach, which is called Performance Orientation Chart (POC). POC is in the same
format as House of Quality (HOQ) but has different contents. Performances of the
design are visualized by relation figures. The designer can interactively develop the
design solution to satisfy multiple specifications guided by these graphical matrices.
Another interesting research work on transmitting CAD models from designers to
manufacturers is presented in (Storti, et al., 1999). Motivated by recent advances in
software development associated with object oriented programming style and data
encapsulation, Storti and coauthors developed an approach to the transmission of part
specifications for distributed solid freeform fabrication. Rather than translating a
standard format of CAD model, the authors specified a set of public methods necessary
for solid freeform fabrication. By specifying public members and methods that provide
fabrication systems with all the information needed to build parts, SFF systems can build
parts based on models constructed in any modeling environment for which the methods
are available.
55
Future design technologies will shift from the “geometry-centric” view to the
“knowledge-centric” view. Knowledge representations will play a larger role in the
design process for creation of product and assembly models. In this research, the author
uses a compromise-DSP (C-DSP) formulation to formulate design knowledge, which is
called decision template (Chapter 5) in this dissertation. A decision template can be
linked with a parametric feature-based CAD model, and transferred from designers to
manufacturers.
Review of Methods:
The high-level representation methods are briefly reviewed in this section to provide
a context for the decision templates to be discussed in Chapter 5.
In Section 1.2.4, design-for-manufacture methods were briefly evaluated in order to
identify research opportunities in design for Rapid Tooling. More detailed literature
review on design-for-manufacture is presented in the next section.
2.5 DESIGN FOR MANUFACTURE: STRATEGIES AND TECHNIQUES
The term design for manufacture, or DFM, is used to characterize efforts by design
and manufacturing to improve the product-process fit or to increase the degree to which
the product and process are designed simultaneously (Susman, 1992). Accordingy to
these two goals, the literature review on DFM is divided into two groups in this section.
• Improve Product-Process Fit
Today most information exchange between design and manufacturing occurs through
informal human communication, often requiring several iterations to get a part right.
Because of the drawbacks associated with this serial approach, design for manufacture
(DFM) has received considerable attention.
In order to improve product-process fit, a dominant approach in field of DFM is to
incorporate manufacturing concerns into the design process, with the goal of improving
product quality, decreasing product cost, and reducing product development time. In
essence, the purpose of this approach is to ensure that the designer considers
manufacturing issues during the design stage.
A DFM system related to this approach should guide the user through the design so
that a part is compatible with a process or it should provide the user with feedback so that
the user can decide if the part needs to be modified. The DFM system requires two
primary components: (1) a means to evaluate the part for manufacturability and (2) the
information needed to support the evaluation. Currently there are a number of
mechanisms that can be used for manufacturability evaluation.
One is to estimate the cost of a part. Cost estimation systems are available for a
number of processes. Most of these systems are based on empirical cost models for the
process and do not require a full three-dimension model of the part. Additionally, these
systems often require knowledge of the process not ordinarily possessed by the designer.
For example, one cost estimating system for injection molding requires the user to know
the injection temperature and pressure, the coolant temperature, and whether two-plate or
three-plate molds will be used for the part (Poli, et al., 1988).
56
A second approach for evaluating manufacturability is to apply manufacturing

guidelines to the geometry of the part to identify attributes of the geometry that are
difficult to manufacture. Often the guidelines are stated in terms of good and bad
practices which are based on years of experience. Unless some measure of
manufacturability is used, these systems do not enable two different part designs to be
compared. Most leading handbooks on DFM, such as (Pye, 1989; Broothroyd and
Dewhurst, 1991; Bralla, 1998), use this approach.
A third approach for evaluating manufacturability is to perform a manufacturing
simulating of the part and then to use the simulating to make redesign suggestions. For
example, plastic part designers already practice this approach to some extent by
modifying a part based on the results of flow simulations. Several commercial software
systems are available such as Part Advisor and QuickFill, which are developed by
Moldflow Corp. and C-Mold respectively.
Each approach has a set of benefits and drawbacks associated with it. Cost
estimation enables different parts to be compared and permits the comparison of parts
produced by different methods. The application of guidelines to the part geometry can be
accomplished at the design stage when the details are being generated. Simulation may
best be applied after the part has been designed, when the detail design is completely
known. Therefore many systems use a combination of the approaches. For example,
Poli and his coworkers have developed DFM systems for forging (Knight and Poli,
1985), die casting (Poli and Shanmugasundaram, 1991), sheet metal stamping (Poli, et
al., 1993), and injection molding (Poli, et al., 1988) based on a group technology
approach. In this approach, cost drivers were identified and then associated with
processing and tooling costs. Classification codes are associated with these cost drivers.
Parts are then classified according to a multi-digit code in order to provide their relative
costs so that candidate configurations can be compared. These systems are not linked to
a geometric model of the part. Instead, the users must fill out spreadsheets to define the
cost drivers associated with the part.
Besides the DFM metrics, a DFM method is any systematic procedure or strategy
that search for, analyzes, evaluates, or improves the manufacturability of a design (Shah
and Wright, 2000). DFM methods can be divided as the following:
- Technological Feasibility: Can a given process manufacture a given design? Can
we find a process from a finite set of processes to produce the desired design?
- Economic Feasibility: Is it cost effective to produce a design in desired batch size
by a given process plan? Can the desired quality be produced/maintained
consistently by the process plan without a high rejection rate? What is the time to
market if a given process plan is used?
- Trade-off Study/Optimization: Consideration of changes to design parameters to
reduce manufacturing cost by a greater amount than the loss of design quality.
The above DFM methods can be found in numerous literatures on DFM research for
different manufacturing processes. For example, (Boothroyd and Dewhurst, 1989)
studied the assembly process and proposes a Design for Assembly (DFA) method;
(Sarma, et al., 1996) presented an integrated approach for a CNC milling process; (Lee, et
57
al., 1998) presented a framework for concurrent process planning for injection molding
process; and (Dissinger and Magrab, 1996) discussed design for powder metallurgy
process. All these approaches largely depend on the characteristic of the manufacturing
process.
Similarly (van Vliet, et al., 1999) divided the DFM approaches into three phases:
Verifying, qualifying and optimizing the product manufacturability. The approaches
discussed before are in the categories of verifying and qualifying. In optimizing category
Grace and Billatos (in Van Vliet, et al., 1999) proposed a re-design approach for
optimization. In order to generate re-design suggestions, it is necessary to know the
functionality of the part. For this purpose, El Maraghy, et al. (in Gupta, et al., 1995) used
pre-defined functional features. Henderson, et al. (in Gupta, et al. 1995) developed a
method for representing functionality semantically within a solid modeling system.
Technologies from other research fields, such as neural networks and visualization,
can also be utilized in DFM research. He, et al. (1998) presented an intelligent system
employing fuzzy sets and neural networks, which is able to predict the process parameter
resetting automatically to achieve better product quality. Seven commonly encountered
injection molded product defects (short shot, flash, sink-mark, flow-mark, weld line,
cracking, and warpage) and two key injection mold parameters (part flow length and flow
thickness) are used as system input which are described using fuzzy terms. On the other
hand, nine process parameter adjusters (pressure, speed, resin temperature, clamping
force, holding time, mold temperature, injection holding pressure, back pressure, and
cooling time) are the system output. A back-propagation neural network has been
constructed and trained using a large number of {defects} -> {parameter adjusters}
expert rules. They system is able to predict the exact amount to be adjusted for each
parameter towards reducing or eliminating the observed defects.
Lu, et al. (1997) presented a volume-based geometric reasoning and visualization
approach to support design evaluation during preliminary design. The system extracted
the underlying geometric characteristics which usually affect part quality or increase
manufacturing difficulties from the part model. All the information is then presented to
designers in a form which can be easily interpreted via visualization techniques. The
presented system focused on thermal- and flow-related problems in die casting.
With the development of Internet, several Internet-based DFM system have been
developed. For example, an Internet-based design for Rapid Prototyping system is
developed at Stanford University to use Rapid Prototyping effectively (Frost and
Cutkosky, 1996; Rajagopalan and Pinilla, 1998). By formalizing RP process constraints
as design rules, the system creates a clean interface to decouple design from
manufacturing processes. Similarly, another web-based design-for-manufacture system
named CyberCut is developed for 3-axis CNC machining at the University of California,
Berkeley (Sarma, et al., 1996; Wang and Wright, 1998). The system combines facilities
for CAD model creation, computer aided process planning and NC code generation.
However, in these methods, researchers have to formulate requirements of the
investigated manufacturing process into knowledge (rule or algorithm), then develop
software systems to allow designers to analyze manufacturability during the design stage.
The approach seems to work well for processes with little requirements like rapid
58
prototyping and for conceptual design stage by using approximation codes. However,
automated analysis meets difficulties if a part or a fabrication process (like Rapid
Tooling) is complex. For a complex part, complex interactions of three-dimensional
geometries have largely prevented formalization of this knowledge. For a complex
fabrication process, it is rather difficult, if not impossible, to formulate all fabricator’s
experience and decision into rules and algorithms in the designer side.
In the usage of Rapid Tooling in producing functional prototypes, designers have
known functional requirements of a part before the prototypes are to be made. Therefore
if the part design requirements are formulated understandable for the manufacturer, then
the manufacturer may be in a better position to adjust the design to facilitate
manufacturing without compromising its functionality.
• Design Product and Process Simultaneously
Ideally designer and manufacturer should cooperate in DFM problem. Concurrent
Engineering (CE), which is “a systematic approach to the integrated, concurrent design of
products and their related processes, including manufacture and support” (Winner, et al.,
1988), was proposed to solve DFM problem. Over the years, numerous research efforts
have been focused on different methods of integrating product and process planning
(Prasad, 1996). While there are a number of ways to categorize these approaches, the
categorization developed by Eversheim (1997) is: the first category is organizational
structure-oriented integration, which is based on the formation of multidisciplinary
design teams, and enforced coordination. The second category is process-oriented
integration through parallel (as opposed to integrated) execution of design and process
planning activities to reduce development time. The final category is information-
oriented integration, which deals with the integration of computer based information to
realize information flow between design and process planning through improving data
exchange.
In the category of organizational structure-oriented integration, research is performed
mainly from the management (Adler, 1992), social (Susman and Dean, 1992) and cultural
context (Liker and Fleischer, 1992). Within industry, concurrent design often takes the
form of “Integrated Product Development Teams”, compromise of representatives from
all aspects of the product development process, who meet together to better design the
product. However, it is well known that the two types of engineer “don’t speak the same
language”. Rosenthal (1990) observed “one prerequisite for collaborative efforts to
succeed is that there be effective technological capabilities for the focused assembly of
information”. Thus, even if designer and manufacturer can work as a team, a
coordination method and a uniform decision framework are important for the
collaborative decision-making.
The approaches on the category of process-oriented integration are discussed before
and are not repeated here.
In the category of information-oriented integration, Collaborative Engineering is
proposed to cause product team members to consider all elements of the product life
cycle. Jin, et al. (1997) point out that there are three basic issues involved in providing
computer support for collaborative engineering design. The first issue is task
decomposition and representation. Task decomposition is concerned with identifying the
59
sub-tasks that can be divided in the way that minimizes interactions among the sub-tasks.
Task representation is related to defining a design task and its sub-tasks in the form that
can be easily handled by designers and computers to identify interactions among the sub-
tasks and cross the life-cycle of product development. The second issue is the need for a
communication infrastructure to facilitate communications among designers. Since in
most cases complete task decomposition, i.e., no interactions exist among sub-tasks, is
impossible, a sophisticated communication infrastructure is needed to facilitate flow of
information among designers. The third issue is coordination support. Coordination is
generally considered as the activity to resolve dependencies among sub-tasks.
Pahng, et al. (1998) presented an integrated and open product development
environment for distributed and collaborative design. The web-based framework, called
DOME (Distributed Object-based Modeling and Evaluation), allows designers to build
integrated models using both local and distributed resources and to collaborate by
exchanging services.
A computer-based design system developed by Sriram et al. provides a shared
workspace where multiple designers work in separate engineering disciplines (Sriram and
Logcher, 1993). In their DICE (Distributed and Integrated Environment for Computer-
aided Engineering) program, an object-oriented database management system with a
global control mechanism is utilized to resolve coordination and communication
problems. Design rationale provided during the product design process is also used for
resolving design conflicts.
Currently collaboration between team members is supported at a number of levels,
from simple data sharing, through single user view and mark-up, to co-viewing, to co-
modeling. It is believed that eventually team will be able to work in a product
development process that will support cooperative innovation through a shared work
environment.
With advances in computing and increased Internet usage, many companies take a
strong interest in the collaborative approach. Several commercial software systems were
developed and are available. For example, OneSapce by CoCreate Software Inc.
(www.cocreate.com) allows a number of people to participate in an on-line work session,
viewing a design (assembly) while cooperatively making comments (annotations)
directly on the 3D model and changing the model as desired. The basic principle of the
software is to operate it in a client-server environment. The server provides
communication coordination and prepares the data for the session. Others similar
products include Windchill by PTC (www.ptc.com) and e!Vista from SDRC
(www.sdrc.com). They support managing and communicating information about product
structures and changes throughout product life cycles. Both Windchill and e!Vista are
written in Java on both the client and server side. They appear as web pages within a
commercial web browser such as MS-Internet Explorer and Netscape.
• Design for Manufacture in other Fields
As to design and manufacturing processes, it is widely agreed that the computer
support of VLSI (Very Large Scale Integrated Circuits) design is generally more mature
than that of mechanical items. A group of researchers in a NSF workshop on structured
design methods observed that a clean interface, which separate design efforts at
60
increasingly high levels of abstraction from the growing complexities of the fabrication
processes, is the key to the rapid success of the VLSI development (Antonsson, 1996).
That is, VLSI designer can design circuits at the block-diagram level, a higher level than
details. This capability allows designers to quickly construct complex circuits, and
facilitates re-using parts of one design in another product. For the block design given by
the designer, a process compiler will fill in the process-dependent details. Mead (1994)
believed that a major factor in the success of the VLSI design is the ability to submit
designs with confidence that products would come back meeting specifications.
Inspired by the NSF workshop, some researchers (Rajagopalan and Pinilla 1998;
Frost and Cutkosky 1996) claimed a clean interface is achieved for RP process by
formalizing process constraints as design rules for designer. Jerard and coauthors (1998;
2000) also proposed a system, FACILE (Fast Associative Clean Interface Language and
Environment), to achieve a clean interface for NC machining.
However, argument that mechanical design cannot be like VLSI design can also be
found (Whitney, 1996). Whitney observed that the final designs are not optimized for
size or power consumption in VLSI design. He believed that this is the trade-off that
must be made to design at a function level. However in mechanical systems, minimizing
size, weight, and power consumption are often primary concerns. Therefore when it is
essential to minimize manufacturing costs or to meet stringent demands on tolerances or
materials properties, the designer will have to know detailed process characteristics,
constraints and costs.
As mentioned before, the research project of this dissertation, Rapid Tooling
TestBed, is also motivated by the NSF workshop and try to develop approaches to
achieve a clean interface between the designer and manufacturer for Rapid Prototyping
and Rapid Tooling processes (Allen and Rosen, 1997; Rosen, 1998). However, instead
of requiring the designer to know detailed process characteristics, constraints and costs, a
different approach is proposed in this research to achieve the clean interface. Our
approach is to develop a method to aid the manufacturer, who is familiar with the process
characteristics, constraints and costs, to integrate design and manufacturing requirements.
The author believes this may be a better approach to achieve the clean interface between
designers and manufacturers for a complex fabrication process such as Rapid Tooling.
Review of Methods:
Although Concurrent and Collaborative Engineering can offer substantial benefits on
sharing information and knowledge, one difficulty associated with them is an effective
coordination mechanism to handle conflicts between different designers and
manufacturers. As mentioned by Jin, et al. (1997), “coordination means to take into
consideration decisions made by others in making local decisions … The research is
concerned with exploring coordination strategies, and developing effective tools to carry
out coordination activities of designers”.
In this research, the author addresses the problem by proposing a decision support
system (Chapter 6), in which the benefits and drawbacks of designs and manufacturing
processes are quantified. Therefore tradeoffs between design and manufacturing
variables can be determined by solving requirements formulated in Compromise DSP
with the aid of optimization software systems.
61
After a DFM strategy is identified, several design technologies used in developing a

DFM system are presented in the next section.
2.6 DESIGN TECHNOLOGIES
Several design technologies, methods, and tools are utilized to support the design for
manufacture in this research. The design technologies of interest include the Decision-
Based Design (DBD), Compromise DSP (cDSP), and Robust Concept Exploration
Method (RCEM). They are described in more details as follows.
2.6.1 Decision-Based Design

Decision-Based Design provides the foundational base in this research for
developing an approach to integrate design and manufacturing requirements. Decision-
Based Design (DBD) is rooted in the notion that the principal role of a designer is to
make decisions (Shupe 1988; Mistree et al. 1989). In design, decisions are invariably
multileveled and multidimensional in nature. By approaching design from the
perspective of making decisions, it becomes possible to envision a unified approach for
design-for-manufacture problem.
An implementation of Decision-Based Design is the Decision Support Problem
(DSP) Technique (see, e.g., Bras and Mistree, 1991). In the DSP Technique, designing is
defined as the process of converting information that characterizes the needs and
requirements for a product into knowledge about a product (Mistree, et al., 1990). A
complete description of the DSP Technique can be found in, e.g., (e.g., Mistree, et al.,
1990). Among the tools available within the DSP Technique, the compromise DSP
(Mistree, et al., 1993) is a general framework for solving multiobjective, non-linear,
satisficing problems. In this dissertation, the compromise DSP is central to modeling
multiple design objectives and assessing the tradeoffs pertinent to design for
manufacture. Examples of these tradeoffs are discussed in the context of the two example
problems in Chapters 7 and 8. Decision Support Problems have been used in a variety of
domains, including design of ships, aircraft, mechanical systems etc (Mistree, et al.,
1990; Koch, et al., 1996). There are two basic types of DSP, selection and compromise.
These decisions can also be coupled and solved simultaneously. The procedures and tools
within the DSP Technique are detailed by Mistree, et al., (1993). The preliminary
selection and selection DSPs are discussed by Kuppuraju, et al., (1985) and Mistree, et
al., (1994); the compromise DSP is detailed within (Mistree, et al., 1993). The
compromise DSP will be used in this research, hence it is further described in the next
section.
2.6.2 Compromise Decision Support Problem

Solving design-for-manufacture problem will require decisions between conflicting
goals – determining the best set of design variables that will achieve both design and
manufacturing goals. Compromise DSP (cDSP) is a multiobjective decision model
which is a hybrid formulation based on Mathematical Programming and Goal
Programming (Mistree, et al., 1993) to satisfy a set of constraints while achieving a set of
conflicting goals as well as possible. The cDSP has been successfully used in several
design applications (Chen, et al., 1996; Koch, et al., 1996). In this research, cDSP is used
62
as a template to transfer the information and knowledge between designer and

manufacturer in DFM phase.
Formulation of a compromise DSP begins with a word formulation and proceeds to a
mathematical formulation. The word formulation consists of the keywords given, find,
satisfy, minimize and their associated descriptors, as shown in Figure 2.15. Given an
alternative and domain information for a problem at hand, the objective in the
compromise DSP is to find the values of system design variables which satisfy a set of
constraints and bounds and achieve as closely as possible a set of conflicting goals while
minimizing a deviation function.
COMPROMISE DSP
Keywords Descriptors
Given An alternative to be improved through modification;
assumptions, system parameters, constraints,
bounds, goals, and the deviation function.
Find Values of system variables and deviation variables.
Satisfy System constraints and bounds (feasibility), and goals
(desired target values or objectives).
Minimize A deviation function.
Figure 2.15 - Word formulation of a Compromise DSP problem (Mistree, et al., 1993).
The generic mathematical formulation of the compromise DSP is presented in Figure

2.16. In general, compromise DSP’s are written in terms of n system variables, which
are represented as a vector X. They define the physical attributes of an artifact that can
be altered. A set of p+q system constraints are used to model the limits placed on a
system design, and must be satisfied for feasibility. Mathematically, system constraints
are functions of system variables only, and may be a mix of linear and nonlinear
functions. Bounds are specific limits placed on the magnitude of each of the system
variables. A set of m system goals is used to model the aspirations for the design. It
relates the goal target, Gi, to the actual performance, Ai(X), of the system with respect to
the goal. The deviation variables, di- and di+, are introduced as a measure of
achievement, the difference between Ai(X) and Gi. The deviation variables, di+ and di-,
are always non-negative, and the product constraint, d +i • d-i = 0 , ensures that at least one
of the deviation variables for a particular goal is always zero. In the compromise DSP
the objective is to minimize a deviation function, Z(d-, d+), a function of the deviation
variables. Deviation function (objective function) formulations are classified as
Archimedean or Preemptive -- based on the manner in which importance is assigned to
satisfying the goals.
63
Given
An alternative that is to be improved through modification.
Assumptions used to model the domain of interest.
The system parameters:
n number of system variables
l number of discrete/integer system variables
p+q number of system constraints
p equality constraints
q inequality constraints
m number of system goals
gi (X) system constraint functions
gi (X) = Ci (X) - Di (X)
fk (di ) function of deviation variables to be minimized at priority
level k for the preemptive case
Wi weight for the Archimedean case
Find
The values of the independent system variables (they describe the physical
attributes of an artifact).
Xi i = 1,..., n
The values of the deviation variables (they indicate the extent to which
the goals are achieved).
di -, di + i = 1,..., m
Satisfy
The system constraints (linear, nonlinear) that must be satisfied for the
solution to be feasible. There is no restriction placed on linearity or convexity.
gi (X) = 0; i = 1,..., p
gi (X) 0; i = p+1,...,p+q
The system goals that must achieve a specified target value as far as possible.
There is no restriction placed on linearity or convexity.
Ai (X) + d i - - di + = Gi ; i = 1,..., m
The lower and upper bounds on the system.
Xi min Š Xi Š Xi max; i = 1,..., n
di - , di + 0 and di - • di + = 0
Minimize
The deviation function which is a measure of the deviation of the system
performance from that implied by the set of goals and their associated priority
levels or relative weights:
Case a: Preemptive (lexicographic minimum)
Z = [ fl ( di - , di + ), . . ,fm( di - , di + ) ]
Case b: Archimedean
Z = Σ Wi (di - + di +) ; Σ Wi = 1; Wi 0; i = 1,...,m
Figure 2.16 - Mathematical formulation of a Compromise DSP problem

(Mistree, et al., 1993).
Objective function captures all the goals in a single function and takes different
forms depending on the way different goals are synthesized. The solutions obtained for
different formulations of objective function could be significantly different. Hence the
objective function formulation is an important step that should be formulated to represent
the original problem requirements. The two types of objective function formulation in
cDSP are explained here.
In pre-emptive formulation, objective function is a list of rank ordered goals. While
minimizing this objective function, the goals with higher rank are minimized before
minimizing the goals with lower rank. Each time a goal is minimized, it is ensured that
its value is not changed while minimizing the subsequent goals (lower ranked goals)
64
(Mistree, et al., 1993). In this approach, all the available design freedom is used to
minimize the highest ranked goal. Then the design space is narrowed to the region that
yields the minimum value for the first goal. This design space is used to minimize the
second ranked goal. This is continued either till all the goals are minimized or till the
design space reduces to a single point.
In Archimedean formulation, the objective function is formulated as a weighted sum
of the appropriate deviation variables. Deviation variables are a measure of the goal
achievement. If the goal is a target-matching goal then both positive and negative
deviations are undesirable and hence both are considered in objective function
formulation. If the goal is a minimization goal then negative deviation indicates
overachievement and is desirable but positive deviation indicates underachievement and
is undesirable. Hence, only positive deviation is considered in objective function
formulation. For a maximization goal, negative deviations are undesirable and only these
are considered in objective function formulation. In archimedean formulation, most
important goal is given highest weight. Also, it is a common practice to normalize
weights so that their sum equals one.
These Archimedean and preemptive formulations that are typically used in the
Compromise DSP formulation suffer two major drawbacks associated to the arbitrary
definition of priority levels and targets for multiple objectives. The Archimedean one is
very difficult to implement with meaningful results because there is no consistent way to
determine a priori the right set of weights. Thus, choosing weights is either done
arbitrarily or through cumbersome iterations. The preemptive approach has the problem
that one objective is assumed infinitely more important than other. Also, in the
Compromise DSP is needed to define targets for each goal, and usually those targets are
selected based on "educated guesses" or through an inefficient process of iteration. These
shortcomings undermine the effectiveness of the Compromise DSP as a design tool.
These problems can be amended by modifying the objective function formulation in
Compromise DSP according to the Linear Physical Programming (LPP) formulation
proposed by Messac and coauthors (1996). Hernandez and Mistree (2001) modified the
objective function formulation of a cDSP to incorporate the features of LPP. This new
formulation of Compromise DSP is presented in Figure 2.17. The cDSP’s in the robot
arm and camera roller cases (Chapter 7 and 8) are formulated and solved using this
approach. The approach judiciously exploits the designer knowledge of the problem by
allowing to express preferences of each objective through various degrees of desirability:
unacceptable, highly undesirable, undesirable, tolerable, desirable, and ideal. It also
eliminates the need for iterative (or arbitrary) weights setting, making the use of
optimization technology more appealing to design engineer in an industrial setting.
65
Given
An alternative to be improved through modification.
Assumptions used to model the domain of interest.
The system parameters:
n number of system variables
p+q number of system constraints
p equality constraints
q inequality constraints
h number of system goals of class 1-S
m number of system goals of class 2-S
gi(x) system constraint function: g i ( x ) = C i ( x ) − Di ( x )
zk(di) function of deviation variables to be minimized at priority level k
Find
xi design variables: i = 1, …, n
d i , k + , d l ,k − deviation variables i = 1, …, h; l = 1,…,m; k = 1,…,4
Satisfy
System constraints (linear, nonlinear)
gi ( x ) = 0 ; i = 1, …, p
gi ( x ) ≥ 0 ; i = p+1, …, p+q
A (x)≤t
i
+ +
i ,5 i = 1, …, h
Ai− ( x ) ≥ ti−,5 i = 1, …, m
System goals (linear, nonlinear)
Ai+ ( x ) + d i−,k − d i+,k − d i+,k +1 = t i+,k i = 1, …, h; k = 1,…,4; d i+,5 = 0
Ai− ( x ) + d i−,k − d i+,k + d i−,k +1 = t i−,k i = 1, …, m; k = 1,…,4; d i−,5 = 0
Bounds
ximin ≤ xi ≤ ximax ; i = 1, …, n
+ −
d ,d ≥ 0 ;
i ,k l ,k i = 1, …, h; l = 1,…,m
d + . d i−,k = 0 ;
i ,k i = 1, … h,…m
Minimize
Lexicographic Formulation
Deviation function with four preemptive levels: Z = [ z1 ,..., z 4 ]
∑ wi ,5− j (d i+,5− j + d i−,5− j )

h+ m
where zj =
i =1
Z = ∑ ∑ wi ,k (d i+,k + d i−,k )
4 h+ m
Archimedean Formulation
k =1 i =1
Figure 2.17 - The Compromise DSP with Modifications based on the Linear Physical
Programming Model (Hernandez and Mistree, 2001).
In physical programming a designer provides targets for different levels of
satisfaction of each of the goals. Initially, all the target-matching goals are divided into
two goals: one minimization goal and one maximization goal. Each of these goals is
further divided into 6 regions: ideal, desirable, tolerable, undesirable, highly undesirable
and unacceptable based on target values. If the goal achievement is in ideal region, then
it is completely satisfied and its value need not be considered in the objective function
formulation. If the goal achievement is in unacceptable region, then it is considered
infeasible. Hence the six regions of a goal result in 4 sub-goals (corresponding to
desirable, tolerable, undesirable and highly undesirable regions) and a constraint
66
(corresponding to unacceptable region). Each of the 4 sub-goals has different weights

and is calculated from their target values. The values of weights increase from desirable
to highly undesirable sub-goals. The contribution of a goal achievement (in different
regions) to the objective function is shown in Figure 2.18. In the figure, goals of 4
different classes are presented. Class 1S corresponds to minimization goals, class 2S
corresponds to maximization goals, class 3S corresponds to target-matching goals and
class 4S corresponds to range-matching goals. Goals in class 3S and class 4S are divided
into two goals: one of class 1S and one of class 2S. From Figure 2.18, it can also be seen
that the rate of increase of objective function in farther regions (from ideal) is higher than
the closer regions.
The compromise DSP may be solved using the ALP algorithm (Mistree, et al., 1993).
A solution to the compromise DSP is called a "satisficing" solution since it is a feasible
point that achieves the system goals to the extent that is possible (Simon, 1996). By
specifying ranged sets of design parameters rather than point solutions, design flexibility
can be maintained, and part design can be more easily adapted to meet manufacturing
requirements. In this research, another term ‘Compromise DSP template’ indicates that
the cDSP formulation is a template and does not correspond to a specific problem. Based
on the requirements of the problem, the goals and constraints are formulated. Also, some
of the system variables in the template could become parameters (that are fixed) in the
cDSP formulation of a specific problem. Compromise DSP template formulations can be
used to represent the problem formulation for a group of similar problems and can be
formulated even before the specific problem requirements are defined.
67
zi Class-1S
~
z5
~z 4
~
z3
~z 2
gi ( x )
ti+1 t i+2 ti+3 t i+4 t i+5
Highly
Ideal Desirable Tolerable Undesirable Unacceptable
Undesirable
zi
Class-2S
~z 5
~z 4
~z 3
~z 2
gi ( x )
ti−5 t i−4 t i−3 t i−2 ti−1
Highly
Unacceptable Undesirable Tolerable Desirable Ideal
Undesirable
zi
Class-3S
gi ( x )
− − − − + + + +
t i5 t i4 t i3 t i2 ti1 t i2 t i3 t i4 t i5
Highly Highly
Unacceptable Undesirable Tolerable Desirable Desirable Tolerable Undesirable Unacceptable
Undesirable Undesirable
zi
Class-4S
gi ( x )
t i−5 t i−4 t i−3 t i−2 t − t +
i1
t i+2 t +
i3 t +
i4 t i+5
i1
Highly Highly
Unacceptable Undesirable Tolerable Desirable Ideal Desirable Tolerable Undesirable Unacceptable
Undesirable Undesirable
Figure 2.18 - Class Function Regions for a Generic ith Objective (Hernandez and Mistree,
2001).
68
2.6.3 Robust Concept Exploration Method (RCEM)

Robust Concept Exploration Method (RCEM) (Chen, 1995; Chen, et al., 1995) is a
methodology resulted from the integration of robust design techniques, design of
experiment techniques, and response surface methodology within the framework of the
compromise DSP. The RCEM facilitates an efficient and effective concept exploration
process as robust top-level design specifications are identified for the design of complex
systems. In this context, robustness of specifications is measured in terms of sensitivity
to changes in requirements - thus the focus is to minimize the effects on the conceptual
design of downstream design changes.
F. The Compromise DSP

Find
Control Variables
Overall Design Satisfy Robust, Top-Level
Requirements Constraints Design Specifications
Goals
"Mean on Target"
"Minimize Deviation"
“Maximize the independence”
Bounds E. Response Surface Model
Minimize
Deviation Function y
A. Factors and Ranges

x2
Noise z C. Simulation
Factors Programs
(Rigorous Analysis x1
x Product/ y Tools)
Control Process Response y=f(x, z)
Factors
µˆy = f(x,µz)
k 2 l 2
Žf σ2ˆz + Žf σ2ˆx
σ2ˆy=Σ ( ) i Σ Žxi ( )
B. Point Generator i=1 Žzi i=1 i
D. Experiments Analyzer
Design of Experiments
Plackett-Burman Input and Output
Eliminate unimportant factors
Full Factorial Design
Fractional Factorial Design Reduce the design space to the region Processor
Taguchi Orthogonal Array of interest
Central Composite Design Plan additional experiments Simulation Program
etc.
Figure 2.19 - The Robust Concept Exploration Method (RCEM) (Chen, 1995).
The computer infrastructure for implementing the RCEM is composed of four

generic processors surrounding a central ‘slot’ for inserting existing, domain-dependent
analysis tools as simulation programs, as shown in Figure 2.19. The simulation programs
(existing analysis programs) are used to evaluate the performance of a minimum number
of conceptual designs. The RCEM processors increase computational efficiency and
facilitate the generation of top-level design specifications. The point generator (processor
B) is used to design the necessary screening experiments. The experiments analyzer
(processor D) is used to evaluate the results of the screening and to plan additional
experiments. The response surface model processor (E) is used to create response surface
models, and the compromise DSP processor (F) is used to develop robust top-level design
specifications.
A. Factors and Ranges: Design variables are classified following the terminology and
principles used in Taguchi’s robust design to define the initial concept exploration space.
69
Design variables are defined as either control factors (under designer’s control) or noise
factors (not under designer's control), and the appropriate range of values for each is
specified. The responses (performance measures) are also identified, along with the
performance goals (signals). The range of interest for each response is also determined
for use in reducing the problem. The means for predicting the performance must also be
identified. The focus in robust design is to reduce both the effect on performance of the
noise factors, and the effect of variations in control factor values on performance.
B, C, D. Sequential Experimentation: A low order experiment is designed, the
experiments simulated (conceptual designs generated), and the results analyzed.
Significant design variables are identified (design drivers), and insignificant parameters
are fixed. Higher order experiments are designed and conducted as necessary and the
results analyzed. Thus, the number of experiments and order of the experiments is
gradually increased while the size of the problem is gradually reduced.
E. Elaborate Response Surface Models: Response surface models are created to
replace the original analysis tools when exploring concepts to generate top-level
specifications. The response surface equations map the factor-response relationship.
When the order of experimentation is satisfactory, the results are analyzed using
regression analysis and analysis of variance to determine the significance of the fit.
When the fit is significant, the final response surface models are defined.
F. Determine the Top-Level Specifications: The response surface models and overall
design requirements are formulated within the compromise DSP to generate the top-level
design specifications. The values of control factors identified in this step become the top-
level design specifications. Different design scenarios can be rapidly explored by
changing the priority levels of the goals.
Using the RCEM, a design space can be quickly and efficiently populated in the
early stages of design. Simulations are run at a set of design points, and response
surfaces are generated that relate product performance to design variable values. These
fast analysis modules are then integrated into the compromise DSP and the best regions
of design solutions are determined based on multiple measures of merit.
Response surface methodology (blocks A – E in Figure 2.19) is used for generating
relationship between several responses and design /manufacturing variables in this
research. A variation of RCEM to solve geometric tailoring problems is also explored.
Review of Methods:
Three design technologies that are related to the development of the design for Rapid
Tooling system (DFRTS) are reviewed in this section. As discussed in Section 1.2.3, the
decisions of design and fabrication variables provide a unified approach for DFM
problem, which is also shown in the review of Decision-Based Design. In this research
compromise DSP is used to formulate the DFM problem in the DFRTS. A similar
approach to RCEM is used in the first phase of the solution process of the DFRTS
(Section 6.4.2). Therefore they are reviewed in this section to provide a context for the
DFRTS to be presented in Chapter 6.
70
With some of the related and important topics reviewed in Section 2.2 ~ 2.6, the next
section will give a summary of the chapter followed by a preview of topics that will be
covered in the next chapter.
2.7 LITERATURE REVIEW SUMMARY
The literature review of the topics provides a foundation that can be used in
developing the mold design and design-for-manufacture methods. Research areas
reviewed in this chapter are:
! Mold configuration design methods
! Mold construction methods and tools
! CAD representation
! DFM strategies and techniques
! Design technologies
In the next chapter a Multiple-Piece Mold Design Method (MPMDM) will be
presented (Figure 2.20). The chapter (Section 3.1) starts with an overview of the
MPMDM, followed by problem formulations considered in the method (Section 3.2) and
steps involved (Section 3.3). Approaches associated with each step are described in
Section 3.4 ~3.6. The chapter closes with a summary, including hypotheses teasing.
Part P Part P
F1 F9 F1
F9
F3 F3
F2 (1) (2)
F6 F8 F2 F6 F8
Fn Fn
F4 F5 F5
F4
F7 F7
Part P
PD1
Mold Base
F1 R3
R1 F9 M1 F1
F3
(3) PD1
F3
F2 F8 F2
PL1
F6
F5 Fn M2 F5 Mk
F4
F4 PL2 Rk F7
PD2
Fn
R2 F7
PD2
Chp 3: Rapid Tooling Mold

Design method

Methods and Technologies
Tools
§2.2 §2.3 §2.4 §2.5 §2.6
Figure 2.20 – Summary of Chapter 2 and Preview of Chapter 3.
71
Chapter 3 – The Multi-Piece Mold Design Method
CHAPTER 3
THE MULTI-PIECE MOLD DESIGN METHOD

Part P Part P
F1 F9 F1
F9
F3 F3
F2 (1) (2)
F6 F8 F2 F6 F8
Fn Fn
F4 F5 F5
F4
F7 F7
Part P
PD1
Mold Base
F1 R3
R1 F9 M1 F1
F3
(3) PD1
F3
F2 F8 F2
PL1
F6
F5 Fn M2 F5 Mk
F4
F4 PL2 Rk F7
PD2
Fn
R2 F7
PD2
Chp 3: Multi-Piece Mold

Design method
Mold design can be a difficult, time-consuming process, particularly for multi-piece

molds. The automation of mold design for injection molding process is studied in many
publications. However, they usually address the determination of parting direction,
parting line, parting surface, and undercut detection individually. In this chapter, a
systematic approach is presented to automate several important mold design steps,
including selection of parting directions, parting lines, parting surfaces, and construction
of mold pieces. Additionally, this multi-piece mold design method (MPMDM) is suitable
for simple two-piece molds (consisting of core and cavity), as well as for molds with
many additional moving sections. The method is mainly based on regions, which are
central to the mold configuration design and mold piece construction processes. The
chapter starts with an overview of the MPMDM (Section 3.1), followed by problem
formulations considered in the method (Section 3.2) and steps involved in the MPMDM
(Section 3.3). With the general steps presented, approaches associated with each step are
described individually (Section 3.4~ 3.6). Finally the relationship between MPMDM and
the validation of the hypotheses are discussed (Section 3.7).
72
3.1 OVERVIEW OF THE MULTI-PIECE MOLD DESIGN METHOD

Combining multi-piece molding and Rapid Tooling techniques, it is possible to build
injection molding tools for complex parts in a very short period of time. However, since
multi-piece molds have more than one pair of opposite parting directions, it is more
difficult and time-consuming to generate a good mold design. Particularly for rapid
tooling applications, delivering prototype parts with turn-around times of less than two
weeks requires fast, proven mold design methods.
Given the geometry of a part, depending on the selection of mold design variables, a
different number of mold pieces may be required to form the part. It is desired to
minimize the number of required mold pieces because less mold pieces reduce the tooling
cost and simplify the operation of the mold. Therefore the problem considered in this
dissertation for multi-piece mold design is described as follows.
Problem MD: Mold Design. Given a solid part and a mold base, design minimum
number of mold pieces that can form the cavity of the part in the material injection
process, and can disassemble properly in the part ejection process.
Mainly from the geometric perspective, a systematical method, Multi-piece Mold
Design Method (MPMDM), is developed to automate several important mold design
steps, including selection of parting directions, parting lines, parting surfaces, and
construction of mold pieces. The elements of the MPMDM and their relations are shown
in Figure 3.1. The CAD models of part and mold base are the input. Correspondingly
the CAD models of mold pieces are generated by the MPMDM as the output. As
illustrated in the figure by a bounding box, the MPMDM actually consists of two
problem formulations, a three-stage design process, and three approaches for each stage.
Multi-Piece Problem MD
Mold Design
Method Problem MCD Problem MPC
(Section 3.2.2) (Section 3.2.2)
A Mold
A Mold Configuration Design Process Construction
Process
Part Basic Mold Piece Mold

elements Regions pieces
An approach An approach to An approach to

to generate combine basic construct mold
basic elements elements pieces
(Section 3.4.3) (Section 3.5.3) (Section 3.6.1)
Mold
Base
Figure 3.1 – The Elements of the Multi-Piece Mold Design Method.
73
In the MPMDM, the given Problem MD is divided into two sub-problems (Problem
MCD and MPC). Related to the sub-problems, a design process with three stages is
proposed and approaches for each stage are developed (Section 3.4 ~ 3.6).
In this chapter, the Multi-piece Mold Design Method (MPMDM) is presented before
the associated system (Rapid Tooling Mold Design System) is introduced in the next
chapter. Related to the elements given in Figure 3.1, the problem formulations of
MPMDM are presented in Section 3.2 to provide the context of the method. Based on the
problem formulations, the mold design processes, which consists of three steps to
generate mold pieces for a part, are described in Section 3.3. The basic elements used in
MPMDM and an alternative approach to generate them from a CAD model of a part are
presented in Section 3.4. An approach to combine these basic elements into regions
related to a mold design is discussed in Section 3.5. Based on the generated regions, a
mold piece construction approach is presented in Section 3.6 for a given mold base.
In Section 1.3.1, research questions and hypotheses for this dissertation were
presented. These research questions and hypotheses provide the basis for this work. To
provide a better understanding of how the MPMDM in this chapter is related to the
hypotheses, the relationship between the elements of MPMDM and the dissertation
hypotheses is shown in Figure 3.2. Hypothesis 1 and sub-hypotheses 1.1 ~ 1.3 were
presented in Section 1.3.1. In Section 3.7, how these hypotheses are validated will be
elaborated.
Multi-Piece Problem MD
Mold Design
Method
Problem MCD Problem MPC
H1
A Mold
A Mold Configuration Design Process Construction
Process
Basic Mold Piece Mold

Part
elements Regions pieces
H1.1
An approach H1.2
An approach to H1.3
An approach
to generate combine basic to construct
basic elements elements mold pieces
Mold
Base
Figure 3.2 – Relationship between the Elements of MPMDM and Dissertation

Hypotheses.
In light of multi-piece molding described in Section 1.2.4, four terms extensively
used in this chapter are defined first to provide a context for the reader. They may have
slightly different meanings from the definitions given in the handbooks like (Pye, 1989)
and (Rosato & Rosato, 1995), which mainly considered two-pieces molding.
74
A Parting Direction (PD), in this dissertation, is a direction along which a mold piece is
separated from the injection molded part.
A Parting Line (PL), in this dissertation, is the continuous closed curves on the surface
of a part which define faces to be split into mold pieces.
A Parting surface (PS), in this dissertation, is the contacting plane of the mold pieces
that forms a seal to prevent the thermoplastic material from escaping.
A Mold base (MB), in this dissertation, is the mold plate that forms the part cavity.
In light of these definitions, the problem formulation for multi-piece mold design is
discussed in the next section.
3.2 PROBLEM FORMULATION FOR MULTI-PIECE MOLD DESIGN
An accurate problem formulation is significant since it will affect the research
approach and the problem solving process to be used. The current state of research on
the automation of mold design mainly focuses on the two-piece molds because they are
more commonly used and relatively easier to design and manufacture than the multi-
piece molds. However, the problem formulations presented for the two-piece mold
design may not be proper for the multi-piece mold design. This is illustrated in Section
3.2.1 by analyzing a representative problem formulation given by Chen, et al., (1993).
Based on the analysis, the problem formulations of the MPMDM are presented in Section
3.2.2.
3.2.1 Analysis of Existing Problem Formulations

The automation of mold design for injection molding process is studied in many
publications (refer to Section 2.2 and 2.3). However, each paper usually addresses the
determination of parting direction, parting line, parting surface, and undercut detection
individually. This may bring problems in the efforts of combining them into one system
since different approaches may use different criteria and different basic elements.
Therefore problem formulations that consider the whole mold design process are highly
desirable.
Since a systematic approach to consider all the above important considerations is not
found, the remainder of this section will mainly discuss the existing problem
formulations for the determination of parting direction (PD). The determination of PD is
the first step in the automation of mold design. It is also the most important step since a
PD will affect all the subsequent steps in the design of a mold. A general description of
the existing problem formulations is: Find the best parting direction according to
some criteria among candidate parting directions, which are calculated from
individual pockets or features. According to this formulation, other faces of the part
that do not belong to the pockets or features will not be considered in the determination.
A representative approach (Chen, et al., 1993) with its problem formulation is
provided here to illustrate this general description. The approach was developed by
Linlin Chen, who was supervised by Tony Woo when they were at the University of
Michigan. For an object Ω (Figure 3.3.a), pockets of the object are the regularized
difference between its convex hull CH(Ω) and Ω, denoted by CH(Ω) -* Ω (Figure 3.3.b).
75
(a) A part (b) Pockets of the part (c) Spherical polygon of the pockets
Figure 3.3 – An Approach Based on V-map.

Based on the Visibility Map (V-map) of pocket surfaces (refer to Section 2.2), spherical
polygons of the pockets can be generated (Figure 3.3.c).
So the problem PPD (pair of parting directions), which is described as “given an
object, find a pair of opposite parting directions that minimizes the number of mold
pieces”, is transformed to a new problem formulation as follows:
Problem SPCA (spherical polygon covering by antipodes): Given a set of spherically
convex polygon V1, V2, …, Vm, find a pair of antipodal points p and –p that minimize
the number of Vi containing either p or –p.
This problem formulation is widely referenced. Several other approaches (Weinstein
and Manoochehri, 1996; Vijay, et al., 1998; Yin, et al., 2001) for the selection of PD
were also developed based on it. However, three problems exist especially when it is
applied to the multi-pieces mold design.
First, the position relationships among pockets and features are not considered in the
formulation. So all related approaches look for a pair of parting directions (PD)
according to the intersection of V-maps of pockets or features. It is assumed that if
several pockets share the same PD, they can be formed by a single mold piece without
PD
PD1-
PD2-
PD2+
PD1+
(a) Pockets with same PD (b) Different Criteria
Figure 3.4 – Examples for the Problem Formulation.
76
considering their actual positions. However, it is not always true. For two pockets with
the same PD, if another pocket with a different parting direction lies between them, it is
difficult or even infeasible to construct a single mold piece for both of them. For
example, in Figure 3.4.a, the PD can satisfy the V-maps of P1 and P3. However, since P2,
which lies between P1 and P3, cannot utilize the same PD, a different PD needs to be
assigned to it. Even if the V-maps of P1 and P3 intersect, a single mold piece for both of
them is hard to generate. To avoid constructing a mold piece to form separated faces, it
is assumed in this dissertation that all the faces of a mold piece are connected.
Second, the problem of finding a PD with the minimum number of cores is
formulated so as to find the minimum number of pockets or minimum number of faces,
which are not covered in PD+ and PD-. However, the criterion of minimum non-covered
pocket number is not always the same as the criterion of minimum mold piece number.
For example, for a simple part as shown in Figure 3.4.b, by using the criterion of
minimum pocket number, PD1 will be chosen as the parting direction. Therefore two
additional cores are needed to form pocket P1 and P5. However if we choose PD2, only
one core is needed to form P2, P3 and P4. So actually the best solution to minimize the
mold pieces (PD2) is different from the solution to minimize non-covered pockets (PD1).
Again, since the geometric relations are not considered in the problem formulation, it is
difficult to know the relationship between the number of mold pieces and the number of
non-covered pockets.
Third, the parting direction for a face F is usually governed by the notion of
complete visibility. That is, for every point p on F, if the ray from p to infinity in the
direction d does not intersect the part, d is a good parting direction for face F. However,
for multiple piece mold design or form pin design, the requirement is less strict. If mold
pieces can form the cavity and they can be disassembled in their parting directions in
some order, then it is a feasible design. As an illustrative example, the shape P in Figure
3.5 is a pocket with empty V-map according to (Chen, et al., 1993) and is therefore not
considered in the formulation. But by using multiple piece mold design, the mold pieces
M1 and M2 can form the shape. In the disassembly process, M1 is first translated in
direction PD1. Then M2 can be moved out first in direction PD2, then direction PD1. So
PD2 is a feasible solution for face F even if F cannot be swept to infinity in direction PD2
without interference with the part.
PD1
M1
d M2
F
PD2
P
Figure 3.5 – Mold Pieces for a Pocket with Empty V-Map.
77
Based on the example given in Figure 3.5, it is clear that multi-piece mold design is
much more complicated than two-piece mold design. In the mold disassembly process,
mold pieces can be translated in different orders, and a mold piece can also be translated
in more than one direction. Since face sweeping operations and body interference tests
are rather time consuming, these tests will not be considered in this dissertation. But
based on the mold design result, an individual simulation module can be developed to
find a suitable disassembly order to translate the generated mold pieces in related parting
directions. If interference between the part and the mold pieces is found, we know that
either a new mold design needs to be regenerated, or the given part is not moldable and
therefore some modifications are necessary. Therefore in this dissertation it is assumed
that checking a mold piece and its neighboring part faces is sufficient to determine if the
mold piece can be disassembled.
In light of the problems of the existing formulations, the problem formulations of
MPMDM for the mold configuration design and the mold piece construction are
presented in the next section.
3.2.2 Problem Formulations of Multi-Piece Mold Design

The generation of mold pieces for a part can be divided into two phases (Figure 3.1):
(1) Mold configuration design. The mold design variables of parting directions and
parting lines are determined according to the geometry of the part.
(2) Mold piece construction. The mold pieces are generated according to a mold
base and the results given by the mold configuration design.
Accordingly the problem MD described in Section 3.1 actually consists of two sub-
problems, problem MCD and problem MPC. The formal formulations of the two
problems are presented as follows.
Problem MCD: Mold Configuration Design. Given a solid part in the
Boundary Representation (Mäntylä, 1988), it can be transformed as a graph G(N, A, R, E)
where N is the set of nodes, A is the set of arcs, R is the set of attributes for nodes, and E
is the set of attributes for arcs, such that:
a. for each face si ∈ Sface, there exists only one node in N;
b. for each common edge between faces si, sj ∈ Sface, there exists a unique arc aij-k
connecting the nodes ni and nj (there may be more than one arc between ni and nj);
c. for each node corresponding to a face si, an attribute ri (an integer number) is
assigned to represent the region number;
d. for each arc aij, an attribute ek is assigned to represent the edge between two faces.
If ri ≠ rj, ek = 1. The related edge is defined as a boundary edge; otherwise ek = 0,
and the related edge is defined as an internal edge.
Among all the combinations of G(N, A, R, E), find a graph G(N, A, Ri, Ej) such that:
(1) for each face pair si and sj, if rsi = rsj = r, we can find a path to link si and sj with
all nodes that have the same r. In other words, all faces with r are connected;
(2) A direction PD exists for all faces with same value ri so that the related mold
piece Mi can be disassembled (demoldability);
78
(3) The total number of different r values is minimized.

According to the result G(N, A, Ri, Ej) of the problem MCD, several sets Ri(F, E, D)
can be generated easily for the part, where F is the nodes N with the same value Ri, E is
the set of arcs A with Ej=1, and D is the direction PD generated for each set ri. Therefore
a definition of mold piece region is given as follows:
Definition 3.1. A Mold Piece Region (MP region) is a set Ri(F, E, D) where F is the
nodes N with the same value Ri, E is the arc set A with Ej=1, and D is the direction
PD generated for each set ri based on the result G(N, A, Ri, Ej) given by the problem
MCD.
Therefore the phase of mold piece construction can be formulated as:
Problem MPC: Mold Piece Construction. A solid part P and a mold base MB
are given in the Boundary Representation. Suppose set Ri(F, E, D) (1≤ i≤ k) are given,
where
(1) each node of F is a face of P, and each face of P is a node of F in Ri;
(2) each node of E is a boundary edge of the faces in F;
(3) D is a direction which makes an angle of at most 90o with the outward normals of
all faces in F;
(4) All faces in F are connected, that is, for any two nodes of F, we can find a path to
link them with all nodes in the same set.
Generate bodies M1~Mk such that:
(1) Faces of Ri are formed and only formed by Mi.
(2) After M1 ~ Mk are assembled together, they form MB with a cavity P inside. That
is, MB – P = M1 ∪ M2 ∪ …∪ Mk, where – and ∪ are two Boolean operators,
subtraction and union respectively.
After the mold design problems considered in this dissertation are formerly defined,
the mold design process related to the problems is introduced in the next section.
3.3 OVERVIEW OF THE MULTI-PIECE MOLD DESIGN PROCESS
The mold design process has been divided into two phases, mold configuration
design and mold piece construction. However, it is still quite difficult to generate mold
piece regions from the given faces of a part. Therefore the first phase is further divided
into two steps, generating basic elements from the faces and generating mold piece
regions from the basic elements (refer to Figure 3.1). In Section 3.4, the author will
explore what kinds of basic elements are appropriate for problem MD based on the
demoldability of a mold piece.
So the Multi-Piece Mold Design Method (MPMDM) has three steps as illustrated in
Figure 3.6 with relevant section given for each step. The inputs to MPMDM are the
boundary faces of a part P. Suppose the part has only one lump (a bounded, connected
region in space). Its boundary faces can be represented as shown in Figure 3.6.a.
Furthermore suppose that the part considered in this dissertation is a polyhedron. So all
faces are planar surfaces and all edges are straight lines. For other kinds of surfaces, they
can be approximated by planar surfaces in different mesh sizes. This limitation will be
revisited in Section 4.2.
79
Part P Part P
Section Section
3.4 3.5
F1 F1
F9
F9
F3 F3
F2 F8 (1) F2 F8 (2)
F6 F6
Fn Fn
F4 F5 F5
F4
F7 F7
(a) Part faces (b) Basic elements

Part P Section
PD1 3.6 Mold Base
F1
R1 F9 R3 M1 F1
F3
(3) F3
PD1
F2 F8 F2
F6
PL1
Fn M2 F5
F5 F4 Mk
F4 PL2 Rk
F7
PD2
Fn
F7
R2
PD2
(c) Mold Piece regions (d) Mold pieces
Figure 3.6 – Steps of the Mold Design Process.

From the faces of P, basic elements are generated first as the starting point for the
determination of regions (Figure 3.6.b). An approach with related algorithms for this step
is presented in Section 3.4. The basic elements are then combined into several mold
piece regions (Figure 3.6.c). All the part faces are combined in the way such that all the
generated regions have at least one feasible parting direction. Therefore it is guaranteed
that the mold pieces constructed in the next step can be disassembled properly. An
approach with related algorithms for this step is presented in Section 3.5.
After the mold configuration design, mold pieces M1, M2, …, Mk are to be
constructed for the mold piece regions R1, R2, …, Rk according to their PDs and PLs
(Figure 3.6.d). There is a one to one correspondence between the mold pieces and the
regions, and the faces of Ri should be formed by the mold piece Mi. By assembling all
the mold pieces together, they should form a mold base with a hollow cavity space in the
shape of P. An approach with related algorithms for this step is presented in Section 3.6.
As stated the generation of basic elements is the first step of MPMDM. In the next
section, the basic elements developed for multi-piece mold design are introduced with
algorithms to generate them from the faces of a part.
3.4 BASIC ELEMENTS OF MPMDM
In this section the demoldability of a mold piece Mi is analyzed first. Based on the
analysis, the basic elements of MPMDM are presented in Section 3.4.2 with a
80
comparison with the basic elements of other approaches. Finally an approach and two
algorithms to generate them are described in Section 3.4.3.
3.4.1 Demoldability of Mold Pieces

The condition of demoldability is the basis to problem MCD (Section 3.2.2).
Several lemmas on the demoldability of a mold piece are derived as follows.
Suppose region Ri is related to mold piece Mi. All the faces of Ri and all the faces in
other regions, which share at least one boundary edge with Ri, will affect Mi’s
demoldability. Although the faces that are not neighboring may also cause interference
after translating Mi for some distance, they are not considered in this dissertation (refer to
Section 3.2.1). Therefore two lemmas, lemma 3.1 and 3.2, are sufficient for determining
the demoldability of Mi. Lemma 3.1 considers the possible interference between a mold
piece and the faces of the related region.
Lemma 3.1. A mold piece Mi can be removed from the faces it forms (FRi) by a
translation in direction PD if and only if PD makes an angle from 0o to 90o with the
outward normals of all faces FRi.
Proof. The proof for the lemma is similar to a lemma presented for casting in (de Berg et
al., 1997). For the “only if” part: if PD would make an angle greater than 90o with
some outward normal of a face f, then any point q in the interior of f collides with the
mold when translated in direction PD.
For the “if” part, suppose at some moment FRi collides with Mi when translated in
direction PD. Let p be a point of Mi that collides with a face f of Fi. This means that
p is about to move into the interior of f, so the outward normal of f must make an
angle greater than 90o with PD. !
Let η(f) = (ηx, ηy, ηz ) be the outward normal of a region face f. The direction PD =
(dx, dy, dz) makes an angle of at most 90o with η(f) if and only if the dot product of PD
and η is non-negative. Hence, a face of the region induces a constraint of the form
ηx dx + ηy dy + ηz dz ≥ 0. (3.1)
Lemma 3.2 considers the possible interference between a mold piece and its
PDi
PDi + η(Fj) PDj
+ PDi
- -
Fj PDi Mj
Ri PEi Mi
Fi PEi Fi Fj Fi Fj Fi Fj
Rj
PEi PEi
(1) (2) (3)
(a) A neighboring face (b) Relation of PD with neighboring face
Figure 3.7 – Demoldability of Mold Pieces.
81
neighboring faces. As shown in Figure 3.7.a, suppose Fi is a face of region Ri and Fj is a

face of another region. If Fi and Fj share an edge PEi, Fj is called a neighboring face of
Ri, and PEi is called a neighboring edge of Ri. For any two neighboring faces, if the
dihedral angle of the faces is less than 180o, the edge between them is a concave edge.
Otherwise the edge is a convex edge. Therefore the translation direction PDi of the mold
piece Mi needs to satisfy some additional constraints.
Lemma 3.2. The mold piece Mi can be removed from a neighboring face Fj by a
translation in direction PD without interference if and only if (1) PEi is a convex
edge; or (2) PEi is a concave edge and PD • η(Fj) ≥ 0.
Proof. For the “only if” part, if PEi is a concave edge and PD •η(Fj) < 0, then any point p
in PEi of the mold piece will collide with the interior of Fj when translated in
direction PD, as shown in Figure 3.7.b2.
For the “if” part, suppose at some moment Fj collides with Mi when translated in
direction PD. Let p be a point of Mi that contacts Fi. This means that p is about to
move into the interior of Fi from the outside, so (1) if PEi is a convex edge, since Fj is
below Fi, Mi is translated in a direction that makes an angle greater than 90o with the
outward normal of Fi. So PD is not a valid direction for Mi according to Lemma 3.1
(refer to Figure 3.7.b1). (2) If PEi is a concave edge, the outward normal of Fi must
make an angle greater than 90o with PD, so PD • η(Fj) < 0. !
It is noticeable that following Lemma 3.2, even if a mold piece can be generated that
does not interfere with its neighboring faces, two neighboring mold pieces generated in
this way may not be able to be taken out at the same time. For example, as shown in
Figure 3.7.b3, face Fi is related to Mi which has parting direction PDi, while face Fj is
related to Mj with parting direction PDj. In the disassembly process, it is quite obvious
that we need to translate Mi in PDi first before we can translate Mj in PDj.
To design a suitable mold construction order, Lemma 3.3 provides guidance in
identifying the neighboring regions that cannot be removed at the same time. Therefore a
disassembly order needs to be considered further for them. For two neighboring regions
Ri and Rj, two of their faces Fi and Fj share an edge PEi as shown in Figure 3.8.a. Two
vectors (called Coedge in the Boundary Representation) CEi and CEj share the same edge
and have reverse directions. They define the interior side of the related faces.
Lemma 3.3. Two neighboring mold pieces Mi and Mj can be removed by translations in
directions PDi and PDj individually without interference with each other at edge PEi
if and only if CEi • (PDi x PDj) ≥ 0 , or CEj • (PDj x PDi) ≥ 0.
Proof. Since CEi = - CEj, the two inequalities are actually the same. So we only need to
prove one. For the “only if” part, if CEi • (PDi x PDj) < 0, the vector (PDi x PDj ) is
in the reverse direction as CEi, so PDi and Mi are in the different sides of PDj.
Therefore the sweeping of PEi in PDi will intersect with the sweeping of PEi in PDj,
as shown in Figure 3.8.b2.
For the “if” part, if two mold pieces can not be removed in directions PDi and PDj
individually, that is, the sweeping of edges PEi in PDi will intersect with the
sweeping of PEi in PDj. So PDi and Mi are in the different sides of PDj. Therefore
CEi • (PDi x PDj) < 0.
82
PDj
PDi PDj PDi
PDi PDj PDj PDi
PEi Rj
Ri
CEj Fj
CEi
Fi Mi Mi Mj Mi Mj
CEi Mj CEi CEi
(1) (2) (3)
(a) Two neighboring regions (b) Relation of Parting Direction of the regions
Figure 3.8 – Demoldability of Neighboring Mold Pieces.

If CEi • (PDi x PDj) = 0, that is PDi and PDj are in the same direction, so the
sweeping of edges PEi in PDi will not intersect with the sweeping of the edge in PDj,
and vice versa (refer to Figure 3.8.b3). !
For Problem MCD, conceptually if we try all different combinations of attribute R
for a part represented by graph G(N, A, R, E), we can always get the best design
according to our requirements and criteria. However, it is quite obvious that this problem
is strongly NP-hard since even one of its sub-problems (pocket combination to get a pair
of directions) is strongly NP-hard (Chen, et al., 1993). To develop an approach to solve
this NP-hard problem, we need to explore the properties of convex and concave edges
further.
Lemma 3.4. Suppose all the edges of a face Fi are convex. Fi can be added to any
neighboring region Rj without increasing the mold piece number or changing the
demoldability of mold pieces, if a parting direction PD exists which makes an angle
of at least 90o with the outward normals of all the faces of the region and Fi.
Proof. First according to Lemma 3.1, if a PD exists for all the faces of the region (Fj) and
Fi, a new mold piece can form Fi and Fj, and be removed from the faces by a
translation in direction PD.
Suppose F1 is the face to be combined with R1. It has an edge E1 sharing with face F2,
which belongs to another region R2. Since E1 is a convex edge, PD1 for R1 and PD2
for R2 will all satisfy Lemma 3.2. So the disassemblability of related mold pieces M1
and M2 is not changed. It is quite obvious that the mold piece number will not change
also. For other neighboring edges we can follow the same proof. !
Lemma 3.4 can also be extended to two regions. Suppose faces F1, F2, …, Fk
compose a region R1 and Fk+1, Fk+2, …, Fn compose a region R2. The bounding edges
between F1, …, Fn and all other faces are E1, E2, …, Ek.
Lemma 3.5. Suppose all bounding edges E1, E2, …, Ek are convex. Two neighboring
regions R1 and R2 can be combined as one region without increasing the mold piece
number or changing the demoldability of mold pieces, if a parting direction PD exists
which makes an angle of at least 90o with the outward normals of all the faces F1, F2,
…, Fn.
83
Proof. First according to Lemma 1, if a PD exists for all the faces of the regions R1 and
R2, a new mold piece can form all the faces, and be removed from the faces by a
translation in direction PD.
For a neighboring face Fi, suppose Ei is the neighboring edge. Since Ei is a convex
edge, the demoldability of related mold pieces is not changed according to Lemma
3.4. Also it is quite obvious that the mold piece number will not increase. Instead it
will decrease by 1. !
The mold design process is actually a process to find a good combination of R in the
graph G(N, A, R, E) according to our requirements. From the above analysis, several
properties of the combining process become evident:
Property 3.1. To get the minimum number of mold pieces, different regions and faces
are to be combined into larger regions to the extent possible.
Property 3.2. To maintain the connectivity of a region, only the region’s neighboring
faces need to be evaluated.
Property 3.3. All faces in one region satisfy Equation 3.1. According to Lemma 3.2, if
a face shares a concave edge with a region, the face normal must satisfy Equation 3.1
also. Therefore after combining faces into regions, two faces with a concave edge are
more likely to be in the same region.
Property 3.4. The combinability of a face with other neighboring faces is affected by
whether the neighboring edges are concave or convex. Since all the faces that share
concave edges need to be considered, a face with only convex edges is more easily
combined than a face with concave edges.
For further discussion, three definitions are given as follows.
Definition 3.2. A Concave Region, CVR(r), is a subgraph of the graph G(N, A, R, E)
given in Problem MCD, such that: (1) for every node Ni that belongs to the subgraph,
related attribute ri = r; (2) for each arc Aij related to Ni and Nj, if attribute ri = rj = r,
the edge associated with the arc is a concave edge; (3) for each arc Aij related to Ni
and Nj, if attribute ri = r and rj ≠ r, the edge associated with the arc is a convex edge.
Definition 3.3. A Convex Face is a face of which all the edges are convex.
Definition 3.4. A Combined Region, CR(r), is a subgraph of the graph G(N, A, R, E)
given in Problem MCD, such that: (1) for every node Ni that belongs to the subgraph,
related attribute ri = r; (2) for each arc Aij related to Ni and Nj, if attribute ri = r and rj
≠ r, the edge associated with the arc is a convex edge.
From the definitions, a combined region can be a concave region, or a region with
several convex faces, or a combination of some concave regions and convex faces.
Based on Lemma 1~5, the combining process to get a good mold design can be
pursued in two steps:
(1) Combine neighboring faces with concave edges into concave regions Ri’;
(2) Combining concave regions and convex faces into combined regions.
84
Considering Properties 3.3 and 3.4 of the combining process, it is assumed that the
faces of Ri’ generated in Step (1) will not be divided into different regions in Step (2).
That is, among all the combinations of R, those combinations, which have two faces
sharing a concave edge but belonging to different regions, will not be considered.
Without further dividing of Ri’, a concave region generated in Step (1) must satisfy
Lemma 1 in order to get a feasible mold piece. If no PD satisfies Lemma 1 for a region
Ri’, we have the following lemma.
Lemma 3.6. Suppose faces F1, F2, …, Fk compose a region with only concave internal
edges and convex boundary edges. If no direction PD makes an angle of at most 90o
with the outward normals of all the faces, no mold pieces to form them can be
removed individually without interference.
Proof. First according to Lemma 1, if no PD makes an angle of at most 90o with the
outward normals of all the faces, two or more mold pieces need to be used for region
R’. Accordingly, R’ can be divided into two or more regions. Each of them has a PD
to satisfy Lemma 1.
For the situation of two regions R1’ and R2’, suppose faces F1 and F2 belong to them
individually with a neighboring edge E12, as shown in Figure 3.9.a. Accordingly, to
satisfy Equation 3.1, any PD1 for R1’ makes an angle from 0o to 90o with the outward
normal η(F1); any PD2 for R2’ makes an angle from 0o to 90o with the outward
normal η(F2), as shown in Figure 3.9.b. If only F1 and F2 are considered, the feasible
region of PD for R’ is from η1 to η2. By adding a face (e.g. F3) in R1’ which shares a
concave edge with F1, the feasible range of PD1 will be reduced from η3 to η2
according to Equation 3.1. The feasible range is moving away from F1 toward F2.
Similarly adding any faces in R2’ (e.g. F4, F6) will make the feasible region of PD2
move away from F2 toward F1. If PD1 and R1’ are in the same side of PD2 (that is
CE1 • (PD1 x PD2) ≥ 0), any direction PD between PD1 and PD2 will satisfy Equation
3.1 for both R1’ and R2’. Therefore it makes an angle of at most 90o with the outward
F6
F3
R 1' R 2'
η1
R 1' η6 η4
η3 F4
η2
F2 F6 PD2 R 2'
F1 E12 F1 PD1
F5
F4
E12 F2
F3
R'
(a) Two neighboring regions (b) Relation of Parting Direction of the regions
Figure 3.9 – Dividing of Concave Regions.
85
normals of all the faces.

According to Lemma 3.3, two mold pieces related to R1’ and R2’ cannot be removed
individually without interference. !
Products designed for injection molding process seldom have a situation as the case
shown in Figure 3.9.b. More often, for a part that needs side cores or form pins, as
shown in Figure 3.5, two concave regions R1 and R2 exist. Each has a PD that satisfies
Lemma 1. However since PD1 and PD2 do not satisfy Lemma 3.3, a form pin is needed
for the part. So in this dissertation, we assume for a given part, any concave region must
have a PD to make an angle of at most 90o with the outward normals of all region faces.
The definition of concave region requires that all the internal edges of the region are
concave, and all the boundary edges are convex. From Lemma 3.7, we can see a
combined region can always be divided into concave regions and convex faces.
Lemma 3.7. A combined region can be divided into concave regions and convex faces.
Proof. For a combined region R with F1, F2, …, Fk,
(1) If all the internal edges are concave, it is a concave region already;
(2) If all the internal edges are convex, F1, F2, …, Fk are all convex faces;
(3) Otherwise, suppose an internal edge E12 is a convex edge.
a. If the related faces F1 and F2 are in one plane, we can delete E12 and make a new
face to replace F1 and F2 without affecting other faces and edges.
b. Otherwise, we can use the plane of F1 or F2 to split all the faces of the region R
into two halves. Assign all the faces in the up side to region R1, and all other
faces to region R2. Now F1 and F2 must be in different regions since E12 is a
convex edge. Therefore E12 is changed from an internal edge to a boundary edge.
For all other edges, the splitting will not change the edges’ convex or concave
properties. For all other faces Fi, (i) if the plane properly intersects Fi as shown
in Figure 3.10.a, Fi is divided into Fi+ and Fi- by adding a new edge Ei. Since Fi+
and Fi- belong to different regions, the new convex edge Ei is a boundary edge.
R1
R1 + P
Fi Fi
Fi +
-
+ Ei Fj ' R1 R2
Fj
R2
Fi - Fi ' Fi
convex
Fi '
edge
- concave concave
R2 P
Fi concave Fj edge edge
edge convex
Fj ' Fj
edge
(a) Intersection of Face and Plane (b) Fake Faces
Figure 3.10 – Splitting of a Combined Region.
86
(ii) if the plane intersects a convex edge Eij, the edge is changed from an internal
edge to a boundary edge; (iii) if the plane intersects a concave edge Eij as shown
in Figure 3.10.b, a fake face Fi’ is introduced for Fi which has zero area and is in
the same plane as Fi. Similarly a fake face Fj’ is introduced for Fj. By adding Fi,
Fj’ to R1 and Fj, Fi’ to R2, we can maintain the relationships that all edges within a
region are concave, and all edges between two regions are convex.
Continue the above process for regions R1 and R2, until all the internal edges of a
region are concave. !
By finding a loop of convex edges, a part can easily be divided into two combined
regions. According to Lemma 3.7, it is evident that a part can be divided into concave
regions and convex faces.
Based on the demoldability of a mold piece, the basic elements of MPMDM are
concave region, convex face and combined region. Compared with other approaches, the
author believes these basic elements are more appropriate for multi-piece mold design. A
more detailed analysis is described in the next section.
3.4.2 Analysis of the Basic Elements

Given the faces of a part, concave/combined regions and convex faces are generated
as the basic elements for Problem MCD. Comparing the basic elements of MPMDM
with those of other approaches (refer to Section 2.2), the author lists similarities and
differences as follows.
(1) Comparison with pockets
A “pocket” defined in (Chen, et al., 1993), or a “concave region” defined in
(Weinstein and Manoochehri, 1996), is actually a combined region defined in this
dissertation. However by adding the concept of concave regions, our approach is able to
handle more general situations.
R1
R3
R1
R4 R
R2
R2
F
R3
(a) Four concave regions (b) A combined region and three concave regions
Figure 3.11 – Concave Regions and Combined Regions.
87
A concave region is different from a combined region in its internal edges. If the
internal edges of a combined region R are all concave, R is a concave region also. For
example, pockets R1~ R4 as shown in Figure 3.11.a are all concave regions. However, a
combined region may consist of several concave regions. Two examples are given in
Figure 3.11.b and Figure 3.12 for depression and protrusion features respectively. Pocket
R in Figure 3.11.b has two convex internal edges and composes concave region R1~R3.
Similarly pocket R in Figure 3.12 has six convex internal edges and can be decomposed
into concave regions R1~R5. All the pockets are not considered in (Chen, et al., 1993)
and (Weinstein and Manoochehri, 1996) since their corresponding V-maps are empty.
By considering the concave regions of R instead of R itself, MPMDM can generate mold
designs for the parts.
F3
F1 F2
R2
R5
R R1 R4
R3
a) Before Splitting (b) After Splitting
Figure 3.12 – Combined Region Example.

(2) Comparison with features
Several approaches determine mold design for a part based on feature and related
feature recognition algorithms. The author believes that the dividing of part faces into
concave regions and convex faces is more general than classifying faces into features.
For example, features considered in (Gu, et al., 1999) are divided into through hole, blind
hole, step, slot, pocket, boss, rib, etc. Using a different classification, (Fu, et al., 1999)
considers inside internal undercut features, outside internal undercut features, inside
external undercut features, and outside external undercut features. Based on the edge
characteristics, each of them is further divided into three-edge, four-edge and more than
four-edge. However, all these features are actually a special case of concave region. So
if we just consider the relation of parting direction and undercut, concave region and
combine region is more general and can handle more cases.
(3) Comparison with faces
Faces are the most basic elements of a 3D CAD model. The number of faces of a
part is much more than that of regions, pockets, or features. For example, the pocket R
shown in Figure 3.12.a has 10 faces. The number of regions is 5 (Figure 3.12.b) and the
number of features is 2. In the mold configuration design process, different combinations
of basic elements are explored to determine a mold design. Therefore using faces as
88
basic elements may bring difficulties in the exploration process because too many faces
exist.
In Sections 3.4.1 and 3.4.2, the basic elements of MPMDM are introduced. With
their definitions in mind, an approach to generate concave regions from the faces of a part
is presented in the next section. Two algorithms and the analysis of the algorithms are
also presented.
3.4.3 The Generation Approach of Concave Regions

The generation of concave regions is central to the identification of the minimum
number of mold pieces. In this section, the overall approach for generating concave
regions is presented first. Then the details of the third step in this approach are discussed
with two algorithms presented.
• Summary of the approach
Three steps can generate concave regions of a part P. They are (1) classify all edges
of P into concave and convex; (2) Add neighboring faces with a concave edge to generate
combined regions; (3) Generate concave regions from the combined regions. After
summarizing Steps 1 and 2, Step 3 is discussed in detail as follows.
(1) Edge classification.
It is straightforward to determine if the dihedral angle of two neighboring faces is
less than 180o. Suppose two faces (F1 and F2) intersect at edge BC. Point A and D are
two points in F1 and F2 respectively (Figure 3.13). For edge BC,
i. if (AB x AC) • AD < 0, BC is a concave edge, where ‘x’ is cross product and ‘•’
is dot product of two vectors;
ii. otherwise, BC is a convex edge.
A F1
F2
C
Figure 3.13 – Edge Classification.

(2) Generation of combined regions.
This step is similar to the algorithm FIND_CVR for the generation of concave
regions given in (Weinstein and Manoochehri, 1996). Basically for any two neighboring
faces Fi and Fj, if the edge between them is concave, they should belong to a same
region.
89
The approach given in (Chen, et al., 1993) can also replace the above two steps.
Their approach is to construct a convex hull CH(P) first, then generating pockets by the
regularized difference between CH(P) and P.
(3) Generation of concave regions.
The proof of Lemma 3.7 provides a way to generate concave regions from a
combined region. The essential step is to split the combined region into concave regions
by utilizing a convex edge that is internal to the combined region. It is likely that these
convex edges either indicate the presence of undercuts or prevent mold pieces from being
demoldable for other reasons. Step 3 is discussed in details as follows.
• Split Region Algorithms
The algorithm Split_Region (SR) uses a face bounded by a convex internal edge (i.e.,
a convex edge is internal to a combined region) to split a given region. For each newly
generated region, the function is executed recursively.
Algorithm: Split_Region
Input: A combined region CR.
Output: A set of concave regions SCVR.
(1) Find all convex internal edges of CR;
(2) If no convex internal edges exist, then add CR to SCVR, and return.
(3) Find a face Fsplit which has convex internal edges, and construct a split surface
fsplit from Fsplit.
(4) Split each face Fi of CR into Fi+ and Fi- by fsplit.
(5) Generate a region CR+ by adding all faces Fi+.
(6) Generate a region CR- by adding all faces Fi-.
(7) Add connecting faces in CR+ and CR- to new regions CRi.
(8) Call Split_Region for each CRi.
By running the algorithm for a region, at least two regions are generated (CR+ and
-
CR ). Sometimes, more than two regions are generated according to the selected face and
the part geometry. For example, in Figure 3.12.a, if the plane of F1 is used to split R,
three regions (R1, R2, and a combined region of R3 ~ R5) are generated in Step (7) since R2
and the combined region in CR+ are not connected.
Step (4), face splitting, is studied in (de Berg et al., 1997). All other steps are pretty
straightforward except Step (3). In Step (3) of the algorithm, there are often many
choices in selecting a splitting face according to the internal edges. By using different
splitting faces or different splitting orders, different split regions will result, as will
different mold designs.
Observing the combined region given in Figure 3.14.a, there are four convex edges
within the region. Related to the edges, we can select any of faces (F1, F2, F3, or F4) as
the splitting face. Totally there are 24 (4!) different combinations. For some of the
selections, the splitting results of face F are shown in Figure 3.14.b. So some selections
(like first F1, then F3) can generate concave regions by splitting regions twice. Others
may require three splitting operations. Also since the generated concave regions are
different, the mold pieces generated for the part may be different. For example, if PD is
selected for a mold piece M1 to form F (Figure 3.14.a), sub-face (1) may be added to M1
90
A F B A F B
B
R (1)
A F4
F1 C C
(F1 -> F3) (F1 -> F2-> F4)
F F2
PD A F B A F B
C
F3
(2)
C C
(F1 -> F2-> F3) (Divide all)
(a) Example part (b) Different results
Figure 3.14 – Different Splitting Faces and Orders.

by using splitting faces F1, F3. However, if using splitting faces F1, F2, and F3, the sub-
face (2) may be added to M1 (Figure 3.14.b).
A straightforward way to get a unique dividing result is to find all the faces that have
one or more convex internal edges, then use related planes to split all faces of the region
(including the newly generated faces). So by using the splitting faces F1, F2, F3, and F4,
face F in Figure 3.14.a can be divided as shown in Figure 3.14.b (Divide all). The result
is not related to the choice of splitting order or splitting face.
Therefore, a modified algorithm, Complete_Split_Region (CSR), for complete
splitting is listed below:
Algorithm: Complete_Split_Region
Input: A combined region CR.
Output: A set of concave regions SCVR.
(1) Find all convex internal edges of CR;
(2) If no convex internal edges exist, then add CR to the set SCVR, and return.
(3) Find faces related to the convex internal edges, add surfaces corresponding to the
faces to set SF.
(4) for a surface fsplit in SF,
(5) for each face Fi of CR,
(6) Split Fi into Fi+ and Fi- by fsplit, and replace Fi in CR.
(7) Add regions CRi generated by adding all neighboring faces of CR with only
concave edges.
However, by algorithm analysis of Complete_Split_Region presented below, it is
clear that complete splitting is not a good approach. Readers can refer to the text of
(Cormen, et al., 1990) for the standard notations used in the area of Algorithm Analysis.
Suppose a region R is composed by n faces. Among them, m faces are related to
convex internal edges (e.g. for region R given in Figure 3.14.a, n = 6, m = 4). By using a
plane to split the region, some faces are divided into two faces. Others are either entirely
above the plane, or entirely below the plane. Suppose the average number of dividing
91
faces for each splitting plane is d, which is a small percentage of n. Assume a = d/n,
which represents the percentage of faces that are split by the splitting plane, and b =1+a.
Algorithm Analysis of Complete_Split_Region:
In the algorithm CSR, the main steps are (4), (5) and (6). Other steps can be omitted
compared to them. For the first splitting, n faces will be handled and a•n will be divided.
Therefore we will get n + a•n = b•n faces. For the second splitting, b•n faces will be
handled and a• (b•n) will be divided. So for the third splitting, b•n + a• (b•n) = b2•n
need to be handled and a• (b2•n) will be split. Similarly, for the mth splitting, bm•n need
to be handled. Suppose the analysis of each face is Θ(1), the running time of the
algorithm is:
T ( n, m) = n + b ⋅ n + b2 ⋅ n + b3 ⋅ n +...+ bm ⋅ n = n ⋅ (1 + b + b2 +...+ bm )
= n ⋅ [( bm − 1) / ( b − 1)] = Θ( n ⋅ bm )
Suppose m= c•n, where c is a constant. So T ( n ) = Θ( n ⋅ bc⋅n )

For any given part, 0 < a < 1 and 1 < b < 2. For large part with a big n, a algorithm
with Θ(n • bc•n) running time (b>1) is obviously not acceptable. Also, the number of
resulting faces will be very large (n•bc•n).
Algorithm Analysis of Split_Region:
Compared to CSR, the analysis of SR is a little more complicated because: (i) the
number of new generated regions can be 2 or more, depending on the topology of part
and the surface used to split. (ii) For each splitting operation, all internal convex edges
related to the splitting face are converted to boundary convex edges. So one or more faces
that share these edges are not potential splitting faces anymore. However, to simplify the
analysis of the algorithm, it is supposed that (1) each splitting generates only 2 regions;
(2) the reduction in the number of splitting faces is not considered. The actual running
time of the algorithm should be much better than the analysis given here because of these
two issues.
Follow the same representations of m, n, a, b, and c for a region R. In the algorithm
SR, the main steps are (4), (7) and (8). For the first splitting, n faces will be handled. a•n
and a•m faces will be divided, and the face used to split is not a splitting face anymore.
Therefore for the next recursive execution, each region will get (n + a•n)/2 = b•n/2 faces
and (m + a•m)/2 = (b•m)/2 splitting faces. So b•n/2 faces will be handled. a• (b•n/2)
and a• (b•m/2) will be divided. Each region in the next iteration will have b2•n/22 faces
and b2•m/22 splitting faces. This process will continue for m’ iterations until bm’m/2m’ =
1, that is, only one splitting face exists in the region. No more recursive execution is
needed. So m’ = logb/2(1/m).
Suppose the analysis for each face is Θ(1), then the analysis for the algorithm is:
b b
T ( n, m) = n + 2 ⋅ T [ ⋅ n, ⋅ ( m − 1)]
2 2
92
b b b
= n + b ⋅ n + b2 ⋅ n +...+2m' −1{( )m' −1 ⋅ n + 2T [( )m' ⋅ n,( )m' ⋅ m)]
2 2 2
= n + b ⋅ n + b2 ⋅ n +...+ bm' −1 ⋅ n + bm' ⋅ n = n ⋅ [( bm' − 1) / ( b − 1)]
1
lg b / 2
= Θ( n ⋅ bm' ) = Θ( n ⋅ b m
)
1
lg
b
1 lg b / 2 b 1 b lg 1− lg b lg 1− lg b
lg b / 2 1 lg b / 2 lg
since b m
= =m b
=m 2
= m lg b− lg 2 , T ( n, m) = Θ( n ⋅ m lg b− lg 2 ) .
m
lg 1− lg b
For any 1< b <1.412, 0 < r =
lg b− lg 2
<1 . So T ( n, m) = Θ( n ⋅ mr ) = Ο( n ⋅ m) = Ο( c ⋅ n 2 ) .
And the face number at the end of splitting is also O(c•n2).

Comparing the cost of algorithm SR and CSR, we can see algorithm SR is more
feasible from the algorithmic perspective. Therefore a good approach to select the
splitting faces and their orders for algorithm SR is necessary.
An ideal way to select a splitting face for a region is to check its neighboring regions
and use mold design knowledge to determine which splitting face is better or if further
splitting is necessary. So some splitting cases shown in Figure 3.14.b may not be
suitable. Characterizing this type of mold design knowledge is beyond the scope of this
dissertation. However, it is important to point out that if mold design knowledge is
available, it should be applied to the selection of appropriate splitting faces and splitting
orders.
According to the analysis of algorithm SR, an obvious set of candidates for the
splitting planes and orders is the set of faces and orders that can always divide the convex
internal edges of a region into two sides evenly. However, it may take a lot of time to
find the best face combination. Therefore in Step (3) of Split_Region, we actually use the
heuristic: find a face composing the largest number of convex internal edges to construct
fsplit. Comparing to the result of Complete_Split_Region, we may miss some face
combinations.
For parts that approximate a cylinder or a sphere, over-split regions may result.
Also, even if all concave regions are generated, it is still impossible to explore all the
combinations of concave regions and convex faces in the region combination as
discussed in Section 3.5. Therefore the author believes that it is unnecessary to split a
combined region until all concave regions are generated.
Usually a combined region needs to be split only when it does not have a feasible
parting direction, as shown in the examples given in Figure 3.11, Figure 3.12 and Figure
3.14. In the process of region combination, the main combining criterion is to test if a
parting direction exists for both regions. Accordingly the parting direction of a region
can determine if further splitting is necessary. So another step is added to algorithm SR
before Step (1), that is:
(0) Get a parting direction PD of CR, if PD is not (0, 0, 0), add CR to SCVR and return.
93
The approach of MPMDM to analyzing CR parting directions is presented in the next

section for the region combination process.
3.5 REGION COMBINATION OF MPMDM
After concave regions and convex faces are generated for a part, Problem MCD in
Section 3.2.2 can be redefined for Step (2) in Figure 3.6.
Problem RMCD: Region-based Mold Configuration Design. Given a solid
part in the Boundary Representation, it can be transformed as a graph G(N, A, R, E), such
that:
a. Each node in N is a convex face or a combined region;
b. For every common edge between nodes ni, nj, there exists an arc aij-k connecting
them;
c. An attribute ri, an integer number, is assigned to each node;
d. For every arc aij, an attribute ek is assigned to represent the boundary edges of a
node. If ri ≠ rj, ek = 1; otherwise ek = 0.
Among all combinations of G(N, A, R, E), find a graph G(N, A, Ri, Ej) such that:
(1) for every pair of nodes ni and nj, if rsi = rsj = r, we can find a path to link ni and nj
with all nodes having the same r. In other words, all nodes with r are connected;
(2) A direction PD makes an angle of at most 90o with the outward normals of all
faces of nodes with the same ri;
(3) The total number of different ri is minimized.
The approaches related to Problem RMCD are presented in this section. They are
organized in the following manner. First the criteria considered in the combining process
and approaches to evaluate parting direction are presented in Section 3.5.1. The
application of the approach in verifying draft angle of faces is introduced in Section
3.5.2. Then the region combination process and related representations are described in
Section 3.5.3. Finally an algorithm for combining regions is presented in Section 3.5.4
with related design knowledge.
3.5.1 Combining Criteria and Their Evaluation Approach

The criteria that are important in the mold configuration design are mainly based
upon the fabrication cost of mold pieces, quality of molded part and productivity of
molding operations. Three criteria that have been identified in Problem RMCD are face
connectivity, parting direction and minimum number of mold pieces. Besides them,
additional criteria should be considered for Rapid Tooling.
Rapid tooling has some additional properties. Since the material strength of the
mold (epoxy resin) is much lower than that of steel, some mold features are more easily
broken off in the ejection process, especially when the features are under-drafted (Palmer,
1999). On the other hand, the mold fabrication cost is small since Rapid Prototyping
techniques are capable of producing virtually any shapes with rather low cost. Therefore
the ease of ejection plays a dominant role in choosing a good mold configuration for
rapid tooling. Consequently, the only additional criterion to be considered in this
dissertation is the ease of ejection.
94
The ease of ejection can be determined by the draft angle and the area in shear
contact between the molded part and tools during the mold-opening operation.
Considering a face with normal η (ηx, ηy, ηz) which forms an angle α with a direction PD
(dx, dy, dz), the value dp = η•PD = ηx dx + ηy dy + ηz dz = |η| |PD| cos(α). dp can evaluate
the value of α. If we take η and PD as unit vectors (|η|=1, |PD|=1), dp = cos(α). Also
since a face Fi with bigger area (Ai) should have a bigger influence on the selection of
parting direction, Ai•dpi can be used to evaluate the ease of ejection for the face in PD.
Ai•dpi is also the projected area of the face in PD because dp = cos(α).
The criterion face connectivity is also important in the region combination process.
Since the faces of a region should be connected after combining, the face connectivity of
two regions or a region and a face needs to be determined in the combining process. For
a part in the boundary representation, the face connectivity has been recorded in the data
structure of the part. Therefore the face connectivity can be accessed quickly and easily
(refer to Section 4.2).
In the remainder of this section, the criterion parting direction will be discussed.
Since an approach that can quickly determine the parting direction of several faces is not
found in literature (refer to Section 2.2), the author will focus on the evaluation of parting
direction in this section. An approach based on Linear Programming is presented for
determining parting direction of two regions or a region and a face quickly and easily.
• Evaluation of Parting Direction
Suppose a combined region CR in Problem RMCD composes planar faces Fi (1≤ i ≤
n). Let η(Fi) = (ηix, ηiy, ηiz) be the outward normal of Fi and a direction PD = (dx, dy, dz).
According to Lemma 3.1, PD is a parting direction of CR if it satisfies:
η1x dx + η1y dy + η1z dz ≥ 0
η2x dx + η2y dy + η2z dz ≥ 0
…
ηnx dx + ηny dy + ηnz dz ≥ 0
The set of feasible directions for the above inequalities can be null, one, or a range of
directions. As stated in Section 2.2, the concept of V-maps can formulate the problem,
and spherical algorithms can calculate the intersection of V-maps. However, for Problem
RMCD, a parting direction of a region is calculated for two reasons: (1) to determine if a
removable mold piece exists for a region to be combined; (2) to find good parting
directions to construct mold pieces. By further analysis, we can see:
i. In the process of region combination, the main concern is if a parting direction
exists for two regions or a region and a face. The actual ranges of the feasible
parting directions are not important.
ii. Since different regions and faces are combined in trial and error, an approach to
determine the direction quickly and easily is essential.
iii. After region combination, a parting direction PD needs to be selected from the
feasible range of the parting directions according to some criteria. Therefore a
proper direction is more important than the whole feasible range in constructing a
mold piece for a mold piece region.
95
So in determining two regions or a region and a face to be combined, instead of the

whole range, only one PD needs to be calculated. Also if the criteria used for the
selection from the feasible range are followed in the calculation of a direction, the same
PD should be achieved for each mold piece after the region combination process.
Therefore the evaluation of PD can be formulated as an optimization problem. Since
only one direction is calculated, it is quite obvious that the optimization approach is much
faster and easier than the spherical algorithms, especially for regions with many faces.
Therefore for a region with face Fi (unit face normal ηi and area Ai, 1≤ i ≤ n), an
optimization problem based on ease of ejection for determining a unit parting direction
PD (dx, dy, dz) can be formulated as follows.
Problem PDOP: PD Optimization Problem.
n
Maximize: f ( d x , d y , d z ) = ∑ Ai ⋅ dpi
i =1
Subject to: dpi = η xi d x + η yi d y + η zi d z ≥ 0 for each face Fi (plane constraints).
x2 + y2 + z2 = 1 (sphere constraints)
An approach to solve Problem PDOP quickly is presented as follows.
PD2
PD1 F1
Si
F2
(0, 0, 0)
PD1
R
(a) Planar faces of a sphere (b) Relations between Problem PDOP and PDLP
Figure 3.15 – PD Evaluation Approach.

• The Evaluation Approach based on Linear Programming
In Problem PDOP, the sphere constraint is to get a unit vector. It can be
approximated by a set of linear surfaces as shown in Figure 3.15.a with acceptable errors.
Suppose the equations of a planar surface Si are sxi x + syi y + szi z – si = 0 with face
normal (sxi, syi, szi) toward inside. We can formulate a new problem as:
Problem PDLP: PD Linear Problem.
n
Maximize: f ( d x , d y , d z ) = ∑ Ai ⋅ dpi
i =1
Subject to: dpi = η xi d x + η yi d y + η zi d z ≥ 0 for each face Fi (plane constraints).
sxi d x + s yi d y + szi d z ≥ si for each face Si (sphere constraints).
The relations between Problem PDOP and Problem PDLP are:
96
(1) If problem PDOP has no solution, problem PDLP will yield the solution dx = dy =
dz = 0;
(2) If problem PDOP has a solution PD(dx, dy, dz) such that f(dx, dy, dz) > 0, problem
PDLP will yield a solution PD’ near PD within the sphere approximation error.
This is because if f(dx, dy, dz) > 0, there must be at least one dpi > 0. Suppose
Problem PDLP gives us a solution PD1 as shown in Figure 3.15.b. It must be a
vector within the sphere because of the sphere constraints. Suppose we construct
a ray from the origin to PD1. It will intersect with one planar surface Si at point
PD2. Since dpi > 0, f(PD2) > f (PD1). PD2 is a better solution than PD1. So the
solution given by Problem PDLP must intersect a planar surface Si. That is, the
solution is within the sphere approximation error.
(3) If problem PDOP has a solution PD(dx, dy, dz) such that f(dx, dy, dz) = 0, problem
PDLP will yield the solution dx = dy = dz = 0. In this case, since all dpi = ηi•PD =
0, PD must be vertical to the normals of all faces (Refer to an example given in
Figure 3.15.b). So if Problem PDLP gives a solution PD(0, 0, 0), two faces F1
and F2 which are not in the same surface should be considered. The unit vectors
PD1 = η1 x η2 and PD2 = η2 x η1 are to be checked by the plane constraints. If the
constraints are satisfied, PD1 or PD2 is also the solution of Problem PDOP.
Therefore solving problem PDLP can generate the solution of Problem PDOP.
Since the solution given by Problem PDLP is only to be compared with (0, 0, 0), the
approximation of the sphere can be rather rough. An approximation of a sphere with 144
faces is sufficient for determining a PD for the construction of mold pieces. These
surfaces can be pre-generated and used for any regions.
Problem PDLP is a linear optimization problem. It is well studied in operations
research (Reklaitis, et al., 1983). For a linear programming problem in only 3
dimensions, several algorithms can solve it with O(n) time (n is the number of
constraints) and linear storage (Megiddo, 1984; de Berg, et al., 1997).
Although the algorithm to get a PD for a region is limited to planar surfaces, quadric
and parametric surfaces can also be handled by approximating them with a series of
planar surfaces. By setting a smaller mesh size, a more accurate model results. Even for
a region with a large number of faces, the running time to solve Problem PDLP is rather
satisfactory. A test example is shown in Figure 3.16. A cylinder face and a planar face
compose a region. Setting surface deviation with different values can approximate the
cylinder face by different number of faces. The face number and corresponding running
time on a PC-700 are listed below. The results obtained for the three cases are all the
direction (0.0, 0.0, 1.0). The testing time is based on calling LINGO system
(www.lindo.com) in a PC with a 700 MHz Intel-III processor.
(a) Maximum deviation: -0.002”; Number of faces: 34; Time = 0.16 second.
(b) Maximum deviation: -0.0001”; Number of faces: 143; Time = 0.17 second.
(c) Maximum deviation: -0.00001”; Number of faces: 224; Time = 0.20 second.
97
R R R
(a) (b) (c)
Figure 3.16 – A Region with Different Mesh Size.

By using a linear program to determine a parting direction, it is evident that the
solution process becomes much faster and easier. Therefore it is feasible to explore more
combinations of regions and faces in less time. The approach for region combination is
presented in Sections 3.5.3 and 3.5.4.
In the next section, the approach described in this section is applied in verifying draft
angles of a part. The author believes that it is an effective and efficient approach for the
automatic detection of non-drafted and under-drafted faces.
3.5.2 Verification of Draft Angle

One thing to be noticed in the plane constraints of problem PDOP and PDLP is that
‘0’ in the right side of the inequality can be changed to a variable df. That is,
dpi = η xi d x + η yi d y + η zi d z ≥ df . The value of df is actually related to the minimum draft
angle γ of a part by df = sin (γ).
For a part to be fabricated by the injection molding process, its surfaces parallel to
the parting direction must be drafted at least an angle γ in order to ease the ejection of the
part and reduce the damaging possibility of the part and molds (Rosato and Rosato,
1995). The minimal draft angle γ of a part depends on many factors: molding process,
material, wall-depth, depth of textured surface, etc. For a complex model, the designer
PD1
F F F F
PD2
Non- Under- Wrongly-

drafted Drafted
drafted drafted
Figure 3.17 – Different Draft Angles.
98
may lose track of which surfaces are drafted and which are not. Consequently a tool to
detect those non-drafted and under-drafted surfaces is necessary.
The suitability of a draft angle for a face depends entirely on the parting direction of
the face. As shown in Figure 3.17, the non-drafted face F for PD1 is a well-drafted face
for PD2. Therefore the verification of draft angle is actually an integrated problem that
should be considered in the mold configuration design process.
Suppose the minimum draft angle for a part is given as γ. By setting df = sin (γ) to
replace 0 in the plane constraints of Problem PDLP, we can find all concave regions with
non-drafted or under-drafted faces. As shown in Figure 3.18.a, concave region R is found
as non-drafted because R has no solution for Problem PDLP. Without the help of a
computer system, it is rather difficult to find out all these features since the difference
between non-drafted and drafted appropriately with γ = 1.5o is non-noticeable when they
are displayed on a monitor.
PD2
R
R2
R1
PD1
(a) (b)
Figure 3.18 – Relationships Between Regions and Minimum Draft Angle.

Also by following the same criteria in the region combination process, it is
guaranteed that all regions have only well-drafted faces in their parting directions.
However the assignment of minimum draft angle γ may influence the number of mold
pieces. For example, if γ = 0, regions R1 and R2 in Figure 3.18.b can be combined into
one region in direction PD2. However, if γ =2.5o, R1 and R2 cannot be combined in PD2
any more. So one more mold piece needs to be constructed in PD1 for R1.
In the next section, the combining process, which is considered as a concave region
growing process, is discussed with the data structures of MPMDM presented.
3.5.3 Analysis of Region Combination Process and Related Representations

After getting combined regions (CR) and convex faces (CXF) from the faces (F) of a
part, a combination that satisfies the requirements of Problem MCD needs to be
determined. It is quite obvious that n(CR) + n(CXF) ≤ n(F), where n() denotes ‘number
of.’
(1) For a part P without CR, n(CR)= 0 and n(CXF) = n(F).
99
Since P is a convex polyhedron, in any two opposite directions d and –d, two mold
pieces can be constructed to form P without any undercuts (Chen, et al., 1993). So the
requirements presented in Problem RMCD can be satisfied quite easily. In this case,
several other criteria need to be considered, like the flattest parting line (Majhi, et al.,
1999), maximum projection area, etc. Industrial injection molding parts are rarely
convex polyhedrons.
(2) For a part P with one or more CR, n(R) = n(CR) + n(CXF) < n(F).
Suppose n(R) = c•n(F) where c is a constant ratio depending on the given shape.
Sometimes, c can be very small, that is n(R) is far less than n(F). As an example, the part
given in Figure 3.16.c has 1 region with 224 faces and 6 convex faces. So n(F) = 230,
n(R) = 1+6 = 7 and c =7/230 ≅0.03. But more often c is around 0.3 ~ 0.6.
Even if n(R) is less than n(F), an algorithm to explore all the combinations of CR and
CXF for minimum mold piece number is still strongly NP-hard. Therefore some
heuristics should be considered in combining regions. Since a CXF can be combined to
any neighboring region if a PD exists for them, CXFs are more flexible to be combined
than CRs. Therefore the focus of region combination is combined regions. In this
dissertation, an approach based on CR-growing process is developed. This approach
allows CRs to grow individually by combining neighboring CXFs and CRs. The process
continues until no further combining happens. The resulting regions correspond to a
mold design for P.
CR of Part P Neighboring CR of Part P

Faces
NFm NF6 PEs

PEs NFm Convex
CXF Faces
NF1 COF NF5 Combining NF1 COF
NF2 NF3 CXF1
NF2
NF3
Core Faces
(a) Before combining (b) After combining
Figure 3.19 – Edges and Faces of a Region in a CXF Combining Step.

Some definitions of edges and faces related to the region combination are provided
first. Suppose before CR-growing, a region CRi is given as shown in Figure 3.19.a. CRi
is composed by faces F1, F2, …, Fn.
Definition 3.5. A Core Face (COF) of CRi is one of the faces that compose CRi before
the CR-growing process. A core face of CRi will always belong to CRi.
The edges of CRi consist of all the edges of faces F1, F2, …, Fn. They can be divided
into three types: concave edges (CVE), convex edges (CXE), and parting edges (PE).
100
Suppose Ei is an edge of CRi. In the boundary representation two faces own Ei (refer to
Section 4.2). One of the owner faces must belong to CRi.
I If the two owner faces all belong to CRi,
a. if their dihedral angle is less than 180o, Ei is a CVE;
b. otherwise, Ei is a CXE.
II If one of the owner faces does not belong to CRi, Ei is a PE. And Ei must be a
convex edge based on the definition of combined region.
Definition 3.6. A Neighboring Face (NF) of CRi is one of the faces that does not belong
to CRi but shares the ownership of an edge with CRi.
Therefore by checking the owner faces of all parting edges, all neighboring faces of
CRi can be identified.
There are two possible cases in the region growing process.
(i) CRi combined with a CXF.
Suppose NF1, … , NFk are neighboring faces of CRi. They are convex faces that do
not belong to any regions. To determine if CRi can be combined with a face NF1, we can
use its unit normal (ηxNF1, ηyNF1, ηzNF1) to form an additional plane constraint:
dpi+1 = η xNF1d x + η yNF1d y + η zNF1d z ≥ 0 and add it to the formulation of Problem PDLP for
CRi.
The linear program in Section 3.5.1 can solve the problem in O(n+1) time. We can
also simplify the solution approach according to the properties of an Linear Programming
problem.
Since Problem PDLP has only 3 variables, each linear constraint is actually a half
plane in 3D space. So the feasible region is a polyhedron obtained by the intersection of
all the half planes. According to (Reklaitis, et al., 1983), one of the corner points of the
feasible region of a linear program is always an optimal solution. Suppose PD(dx, dy, dz)
is the solution of Problem PDLP with n half plane constraints h1, …, hi. If a half plane
hi+1 is added as a new constraint, de Berg (1997) gives that:
(a) If PD satisfies the constraint hi+1, the new optimal solution PD’ = PD.
(b) If PD does not satisfy the constraint hi+1, PD’ must be one of the intersection
points of hi+1 with h1 ~ hi, or the linear problem is infeasible.
For case (b), an algorithm running in linear time is also given to find the new optimal
solution. So the approach also runs in O(n) time, but it should have a smaller coefficient
of n compared to that of the LP solver.
For some neighboring faces NF1, … , NFk, if PD’ exists, CRi can combine them into
one region. All the faces F1, …, Fn and NF1, …, NFk are called faces of CRi, and faces
NF1, … , NFk are the Convex Faces (CXF) of CRi. So the faces of a combined region are
actually composed by core faces and convex faces. One difference is that after a face is
set as a core face, it will always belong to the region. On the contrary, a convex face may
leave the region to join another region as shown later.
101
As convex faces switch among regions, some old PEs are changed to CXE of CRi
and some new PEs are generated related to the new combined faces. Similarly some
neighboring faces are changed to convex faces of CRi and some new NFs are generated
related to the new PEs, which are shown in Figure 3.19.b.
CR1 and CR2 of Part P Neighboring CR1' of Part P
Faces
PEs NFm PEs

CR1 NFm
Region
COF Combining NF1
NF1
NF2
Convex CXF1 CR1'
NF2
Faces NF3
CXF1 NF3 COF
COF
Core Faces
CXF2
CR2
(a) Before combining (b) After combining
Figure 3.20 – Edges and Faces of Two CRs in a Combining Step.

(ii) CRi combined with another region.
As shown in Figure 3.20.a, suppose a neighboring face (NF3) of CR1 is also a face
(CXF1) of another region CR2. That is, CR1 and CR2 are two neighboring regions and are
candidates for joining into a larger region.
Since the convex faces of CR1 and CR2 are more easily switched between regions
than their core faces, CXFs are not considered in the test for region combination. That is,
only COFs of CR1 and CR2, and the convex faces to connect COFs, are formulated as
planar constraints in Problem PDLP. If the problem has a non-zero solution PD, two
regions are combined into one region CR1’ (refer to Figure 3.20.b). And the core faces of
CR1’ are the faces of the planar constraints in Problem PDLP. All the convex faces of
CR1 and CR2 are not faces of CR1’ anymore. Instead they are CXF of the part and can be
combined by any regions including CR1’.
To get convex faces to connect COFs of two regions, the original face of any convex
face needs to be recorded also. That is, if F1 shares a parting edge PE1 with a region face
F2, besides recoding F1 as the neighboring faces of CR, origin(F1) is recorded as F2. So
as shown in Figure 3.20.a, if NF3 of CR1 and CXF1 of CR2 are the same face, the original
face of NF3 in CR1 is added to planar constraints of CR1. The same process can be
recursively executed for the face of origin(NF3) until the original face is a core face.
Similarly, the original faces of CXF1 can be added to CR2.
Therefore the main data structures for a combined region include:
Vector PD; // Parting direction dx, dy, dz
Box Bound; // Bounding box of core faces
Edge_List LPE; // Parting edges
102
Face_List LCOF; // Core faces

Face_List LCXF; // Convex faces, origin of CXF is assigned as an attribute
Face_List LNF; // Neighboring faces, origin of NF is assigned as an attribute
In light of the combining process based on CR-growing, an algorithm and its
analysis are presented in the next section.
3.5.4 Region Combination Algorithm and Design Knowledge

Based on the data structures given in the last section, the algorithm Combine_Region
(CR) of MPMDM is listed as follows.
Algorithm: Combine_Region
Input: A set of combined regions SCVR and convex faces of P.
Output: A set of combined regions SCVR and convex faces of P.
(1) change ← FALSE.
(2) for a region CRi in SCVR,
(3) for each NFk in LNF, // decide the list of neighboring faces
(4) rn ← region_num(NFk);
(5) if rn = 0, then // NFk is not in any region
(6) if combinable_convex_face(CRi, NFk) = TRUE, then
(7) combine NFk with CRi, update LPE and LNF of CRi, change ← TRUE.
(8) else // NFk belongs to region CRrn
(9) if combinable_region(CRi, CRrn) = TRUE, then
(10) combine CRrn with CRi, update LPE and LNF of CRi, change ← TRUE.
(11) if change = TRUE, then Go to Step (1).
(12) else return.
The algorithm has a main loop to determine if any region combination should occur.
Within each iteration, there are two loops in Step (2) and (3). So the running time
depends on the number of regions and the number of neighboring faces for each region.
As small regions are combined into larger regions, the region number is decreasing and
the NF number for each region is increasing.
Although it is difficult to calculate the exact time cost of the algorithm, making some
simplifications can derive an approximate computational complexity measure. Suppose
P has n faces, among them a•n faces are convex faces (0< a <1). If the input SCVR has m1
regions, and the output SCVR has m2 regions, we can use the average region number
m + m2
m= 1 to represent the region number in all the iterations. Each region will have at
2
most n/m faces. Similarly we can use NF to represent the average number of neighboring
faces for all regions in the whole process. Suppose the program exits when all convex
faces are combined into regions. Within each iteration, one region will combine a
fraction of its neighboring convex faces, suppose b•NF. So if algorithm CR will iterate k
a⋅n
times before exiting, we know a ⋅ n − k ⋅ m ⋅ b ⋅ NF = 0 . That is k ⋅ m ⋅ NF = . Also
b
103
n
since the combining cost of a convex face is O( ) as shown in Section 3.5.3, the time
m
cost of algorithm CR is:
n a⋅n n n2
T ( n, m) = k ⋅ m ⋅ NF ⋅ = ⋅ = O( ). So T ( n ) = O( n2 ).
m b m m
For two regions or a region and a convex face, if the results in Step (6) or (9) are
false in an iteration, there is no need to determine their combinability in any later
iterations. Therefore repeated executions can be avoided by adding a face array and a
region number array in each region. The arrays record neighboring convex faces and
combined regions that are not combinable with the region individually.
F5
F4 Combining result 1:R1+F1+F3+F4+F5+F6, R2
F1
R1 Combining result 2:R1+F1+F3+F4+F5, R2+F6
F6 Combining result 3:R1+F1+F3+F4+F6, R2+F5

...
F3
R2 F2 Combining result 16:R1+F1, R2+F3+F4+F5+F6
Figure 3.21 – Different Combining Results of a Part.

In Algorithm CR, Steps (6) and (9) need to be discussed further. Conceptually, by
using different combining orders and different rules in functions
combinable_convex_face and combinable_region, different mold designs can be
obtained. A simple example is shown in Figure 3.21. Suppose a box has two similar
cavities in its top (F1) and bottom (F2) sides. Faces 3 ~ 6 are vertical relative to faces 1
and 2. No draft angle is considered. Several combining results for the part can be
obtained by trying different combinations of faces 3~6 with regions 1 and 2. Since the
related mold designs have the same number of mold pieces and each mold piece can be
disassembled properly, they all satisfy the requirements given in Problem RMCD. So
just from the geometric perspective, they are equally good designs.
Therefore, besides the requirements given in Problem MD, some mold design
knowledge should also be considered in Functions combinable_convex_face and
combinable_region. The heuristic rules that are considered in this dissertation are listed
with some explanations.
(1) Core/Cavity property of a region.
A combined region CRi can be classified as an internal region or an external region
according to its parting edges. If a loop formed by some parting edges of CRi is an
internal loop of a neighboring face, CRi is an internal region. For example, the parting
edges of R1 in Figure 3.21 form an internal loop of F1. So R1 is an internal loop.
Otherwise CRi is an external region, like R2 in Figure 3.21.
104
Generally internal regions are related to core mold pieces, and external regions are
related to cavity mold pieces. To facilitate the ejection of part from a core, convex faces
with vertical normals to PD usually go with cavity side instead of core side in the mold
design. So a normal mold design for the part given in Figure 3.21 is R1, F1 in core side,
and R2, F2 ~ F6 in cavity side.
(2) Main parting direction of a region.
For a part with many combined regions CRi, we may get several parting directions
PDi related to the combined regions. Among them, a pair of opposite directions is the
main parting direction (MPD) of the mold design. Others are all side directions.
Our approach to get the MPD is to find a pair of directions with the maximum region
volumes among all PDi. That is, we find parting directions PDk which are in the same or
opposite directions according to some tolerance. For each PDk, a volume of the related
region is calculated from its boundary box. The sum of all region volumes is assigned to
the direction. Finally we can get MPD by finding the direction with maximum volume.
Accordingly we can assign a value if_main_pd to each region. If MPD can be set as a
parting direction of a region, its if_main_pd is true. Otherwise it is false.
In general, it is preferred to combine a vertical face with a region in MPD than a
region in a side direction. So a convex face Fi is combinable with a region CRj if:
(a) the normal of Fi is not vertical to the PD of CRj;
(b) Fi is a vertical face, if_main_pd of CRj is true, and CRj is an external region.
PD1 PD1
Top view of F
PD2
PD2 F CR1 F2 L1
CR2 CR1 CR2 O1
C2
F M C1
O2
L2 F1
(a) case 1 (b) case 2 (c) Dividing approach
Figure 3.22– Face Dividing for A Region of Cavity.

(3) Neighboring face number of a region.
Suppose face F is a neighboring face of region CR2 and also a face of CR1.
According to the solution of Problem PDLP, suppose CR1 and CR2 are not combinable.
If the neighboring face F is the only NF of region CR2, CR2 must be a cavity in face F as
shown in Figure 3.22. So:
(a) if F is a convex face of CR1, delete F from CR1 and add it to CR2;
(b) if F is a core face of CR1 (as shown in Figure 3.22.a), or if F is also the only
neighboring face of region CR1 after F is deleted (as shown in Figure 3.22.b), F
should be divided into two faces F1 and F2 for CR1 and CR2.
105
The approach to do the above face splitting is briefly described here and shown in
Figure 3.22.c. Suppose internal loops L1 and L2 of F are related to CR1 and CR2. The
smallest enclosing discs C1, C2 for each loop can be constructed respectively by using
algorithm MiniDisc given in de Berg (1997). The algorithm runs in O(n) time, where n is
the number of points in a loop. Suppose O1, O2 are the centers of C1, C2, respectively,
and M is the middle point of line segment O1O2. If we assume C1 and C2 do not overlap, a
line can divide F into F1 and F2 by passing through M and being vertical to O1O2.
(4) Combining order of a region.
The order of regions in SCVR and the order of neighboring faces in LNF of a region
also affect the combining results. For example, a part as shown in Figure 3.23.a has four
regions, R1 ~ R4. Depending on the order of R2, R3 related to R1, R4, a combining result
with R1, R2, R3 as a region and R4 as another region may be generated. A result with R1,
R4 as a region and R2, R3 as another region may also be generated. Similarly if R1 is the
first region in SCVR, the above two results will be generated depending on the order of
neighboring faces in LNF of R1.
So the regions in SCVR can be reordered according to heuristic rules before
combining regions. For example, regions can be ordered by their volumes; that is,
moving regions with larger volumes to the beginning of SCVR. Hence, regions with larger
volumes can be combined first. Regions can also be ordered according to whether or not
they are in the main parting directions. Regions in the main parting directions would
combine first. Also, regions of SCVR can be classified into two sets SCVR1 and SCVR2
according to some criteria, like the value of if_main_pd. Then let regions in SCVR1
combine with each other first. After the combining process finishes, regions in SCVR2 are
added to start a new combining process.
In Section 4.2, the design knowledge considered in MPMDM will be revisited in the
discussion of RTMDS and its implementations.
• Analysis
By considering more mold design knowledge, a mold design that is more compatible
with good mold design practices can be achieved. Since different mold designs
correspond to different combinations of CR and CVX, the region-based approach is very
flexible in enabling the addition of new design knowledge. In algorithm
Combine_Region, functions combinable_convex_face and combinable_region control the
generation of different combinations of CR and CVX. So to add more design knowledge,
related heuristic rules can be formulated and added into the two functions. Since other
steps remain unchanged, it is rather easy to implement.
Besides mold design knowledge, our approach can handle mold fabrication
knowledge by following similar processes. For example, suppose mold pieces for the rib
part shown in Figure 3.23.b are to be fabricated with SLA machines. Assume surface
finish requirements of 2 µm are specified for the two opposite faces F1 and F2.
According to our knowledge of the SLA process, such high surface finish can be achieved
only by building the mold pieces such that F1 and F2 are the top surfaces of the part
(West, 1999). Therefore to build mold pieces such that F1 and F2 are both at the top
surface, a constraint that F1 and F2 cannot be in the same region is added. By adding an
additional combining rule in functions combinable_convex_face and combinable_region,
106
F2
R2
PD2 F1
PD1 R4
R1
R3
(a) Different combining orders (b) Mold fabrication knowledge
Figure 3.23 – Two Example Parts for Combining Order.

a better design that satisfies fabrication requirements is achieved. However, this may lead
to a three-piece mold design for the part instead of a design with only two mold pieces.
Finally, besides one combining results, several results can be generated
automatically, which allows the mold designer to select a mold configuration design
among them. In Section 9.4 this approach will be revisited.
After mold piece regions are generated for a part in this section, an approach for
constructing mold pieces effectively and efficiently will be presented in the next section.
3.6 MOLD PIECE CONSTRUCTION APPROACH BASED ON REVERSE GLUE
After region combination process presented in Section 3.5, several mold piece (MP)
regions have been generated for a part with their parting directions and parting lines. If
there exist convex faces that do not belong to any regions, they can form one or more MP
regions according to the connectivity. The parting directions of the new regions can be
evaluated by solving problem PDLP (Section 3.5.1). An example is given in Figure 3.24.
After region combination process, we got two regions (R1, R2) and four convex faces,
PD1 PL1(PL2)
R1
R1
M1
PL1(PL3)
R2
PD2 PL2(PL3)
M2
R2 M3
Convex R3
Face PD3
(a) Regions and convex faces (b) Regions (c) Related Mold Pieces
Figure 3.24 – Regions and Mold Pieces of a Part.
107
which are shown in Figure 3.24.a. The convex faces can form a new region R3. The
parting lines of the regions are shown in Figure 3.24.b. Correspondingly, three mold
pieces need to be generated as shown in Figure 3.24.c.
This section focuses on Step (3) shown in Figure 3.6. An approach based on the
reverse glue operation is developed for solving Problem MPC (Section 3.2.2) effectively
and efficiently. The remainder of this section has been organized in the following
manner. First, the reverse glue operation and the principle related to the operation are
presented in Section 3.6.1. A key step of the approach, generation of glue faces, is then
discussed. A problem formulation and related algorithms are presented for the generation
of glue faces in Section 3.6.2. Finally, the algorithms for constructing two-piece molds
and multi-piece molds are presented respectively in Section 3.6.3. The role of parting
surface in the approach is also described in this section.
3.6.1 Principle and Related Representations of the Approach

Although the approach of MPMDM is quite flexible for any mold base, for the
discussion below, it is supposed that a mold base (MB) is given as a rectangular block
that can contain the injected part (P) entirely inside. The position and orientation of P
within MB is affected by gate design, runner design and ejection pin design. The
approach of MPMDM is also flexible for the position and orientation of the part in MB.
Suppose the main parting direction of P generated in Section 3.5.4 is rotated to the z-axis
of MB, and the center point of the bounding box of P is translated to the center point of
the bounding box of MB.
After P is positioned at a suitable position and orientation in MB, a Boolean mold
base M’ = MB – P (‘–’ is the subtraction operation) can be generated as shown in Figure
3.25.a. From the equation, it is obvious that M’ contains two kinds of faces, outside faces
and inside faces. Outside faces (Fout) of M’ are the faces corresponding to the faces of
MB. Inside faces (Fin) of M’ are the faces corresponding to the faces of P. Suppose all
the faces of a body or a region are denoted as F( ), and the union of two face sets is
represented by +. So F(M’) = Fout-M’ + Fin-M’. (3.2)
The approach of MPMDM is also quite flexible for any parting surface, which will
be discussed in Section 3.6.3 in more details. Suppose a parting surface PS is given.
According to the position related to PS, all outsides faces of M’ can be divided into two
sets, faces above PS (denoted as ↑) and faces below PS (denoted as ↓). Therefore
Fout-M’ = ↑Fout-M’ + ↓Fout-M’. (3.3)
For a part P and MB’, suppose mold piece M1 is to be constructed for region R1 first.
All other regions can be considered as a new region R2’. So only two mold pieces M1 and
M2’ need to be constructed for regions R1 and R2’ respectively. Replacing P with R2’ and
MB with M2’, all other mold pieces can be constructed by recursively executing the above
process. Therefore, for discussion below, only the construction of two mold pieces (M1
and M2) for two regions (R1 and R2) is considered.
It is evident that each face of Fin in M’ has a corresponding face in P. Two faces are
in the exact same position but with opposite face normals. We use ~ to denote this
108
PF
P
↑Fout M’
MB
Glue Face 2
M1
Part ~F(R1)
M ' = MB-P Glue
Glue Face 1 Fout M’
M2
Reverse ~F(R2)
Glue
PF Fin M’
Glue Face 1
↓Fout M’
Glue Face 2
(a) Boolean Mold Base M’ (b) Mold Pieces M1 and M2
Figure 3.25 – An Example of a Boolean Mold Base and Mold Pieces.

corresponding relationship. Since faces of R1 are to be formed by M1 and faces of R2 are
to be formed by M2, all faces of Fin can be divided into two sets:
Fin-M’ = ~F(R1) + ~F(R2). (3.4)
From Equation (3.2), (3.3), and (3.4),
F(M’) = ↑Fout-M’ + ↓Fout-M’ + ~F(R1) + ~F(R2). (3.5)
For two bodies A, B in Boundary Representation, A = bA ∪ iA and B = bB ∪ iB,
where b denotes the set of points on the boundary and i denotes the set of interior points.
Therefore the union of A, B can be represented as: C = A ∪ B = bA ∪ bB ∪ iA ∪ iB
(Mortenson, 1997). For the boundary of C, we have:
bC = b(bA ∪ bB ∪ iA ∪ iB) = bA ∪ bB. (3.6)
Gluing operation is a standard high-level Euler operation for half-edge data
structure, which is a de facto data structure of the boundary representation. More details
about the half-edge data structure including coedges are given in Section 4.2. The main
operations of the gluing operation are to swap the partner coedges for all common edges
of two bodies, then merge all faces into one body. Mäntylä (1988) gave an equation to
represent union operation by gluing operation,
A ∪ B = (A out B) ⊕ (B out A), (3.7)
where ⊕ denotes gluing operation.
Considering the relation with bB, the boundary faces of A can be divided into three
kinds of faces: faces outside of bB, faces on bB, and faces inside of bB. That is,
bA = (bA out bB) + (bA on bB) + (bA in bB). (3.8)
Similarly, bB = (bB out bA) + (bB on bA) + (bB in bA). (3.9)
109
In Problem MPC, M’ = M1 ∪ M2. From Equation (3.6) and (3.7),

bM’ = bM1 ∪ bM2 = (bM1 out bM2) ⊕ (bM2 out bM1). (3.10)
Since M1 and M2 are two different bodies that are assembled together, (bM1 in bM2)
and (bM2 in bM1) in Equation (3.8) and (3.9) are null. So
bM1 = (bM1 out bM2) + (bM1 on bM2); (3.11)
bM2 = (bM2 out bM1) + (bM2 on bM1). (3.12)
If bM1 and bM2 are given, based on Equation (3.10) ~ (3.12), bM’ can be generated
by gluing operation (from Figure 3.25.b to Figure 3.25.a) with three steps.
(1) Finding faces (bM1 on bM2) and (bM2 on bM1) from bM1 and bM2, delete them
from M1 and M2 respectively (Equation (3.11) and (3.12));
(2) Swap partner coedges for all common edges of M1 and M2;
(3) Merge all faces of M1 and M2 into one body M’ with two shells (Equation (3.10)).
As shown in Figure 3.25.b, faces (bM1 on bM2) and (bM2 on bM1) are actually the
contact faces between M1 and M2 when they are assembled together. A definition for
them is given.
Definition 3.7. A Glue Face (GF) of a mold piece M1 related to another mold piece M2
is one of the faces (bM1 on bM2).
According to the definition, for each glue face of M1, there is a corresponding glue
face of M2. These two faces are in the same position but with opposite face normals.
However, in Problem MPC, bM1 and bM2 are unknown. They are actually what is to
be generated. Since MB and P are given in Problem MPC, M’ and bM’ can be generated
quite easily based on the equation M’ = MB – P.
The boundary of a 3D model is faces. So bM’ = F(M’). From equation (3.5) and
(3.10), it is evident that
↑Fout-M’ + ↓Fout-M’ + ~F(R1) + ~F(R2) = (bM1 out bM2) ⊕ (bM2 out bM1)
Therefore, suppose
(bM1 out bM2) = ↑Fout-M’ + ~F(R1); (3.13)
(bM2 out bM1) = ↓Fout-M’ + ~F(R2). (3.14)
If all the glue faces (bM1 on bM2) and (bM2 on bM1) are also known, M1 and M2 can
be generated based on Equation (3.11) and (3.12) with three steps.
(1) Remove faces ↑Fout-M’ + F(R1) and ↓Fout-M’ + F(R2) from M’, and use them to
generate two new bodies M1 and M2 respectively (Equation (3.13) and (3.14));
(2) Generate glue faces (bM1 on bM2) and (bM2 on bM1), and add them to M1 and M2
respectively (approaches are discussed in Section 3.6.2 in more details);
(3) Swap partner coedges for all common edges of M1 and M2 (Equation (3.11) and
(3.12)).
In this dissertation the above processes are named Reverse Glue operation.
110
In the reverse glue operation, Step (1) and (3) are straightforward. However, Step
(2), the generation of glue faces, may be rather difficult for some geometry. The
approaches of MPMDM for the generation of glue faces are presented in the next section.
3.6.2 Generation Approach of Glue Faces

For equation M’ = M1 ∪ M2, if M1 and M2 are given, the result M’ obtained from the
gluing process ( M1 U M 2 Glue
→ M ' ) is unique. However, if M’ is given, the result M1
and M2 from the reverse glue process ( M ' Re 
verseGlue
→ M1 U M 2 ) is not definitive.
Using different glue faces will give us different mold piece designs. For example, the
approach described in Section 2.3 generates parting surfaces by extending parting lines.
These faces are also glue faces since they also satisfy (bM1 on bM2) and (bM2 on bM1)
when assembling M1 and M2 together. So the key of problem MPC is actually to find
appropriate glue faces. In this dissertation, only planar parting surfaces are considered
since they can reduce mold fabrication cost and material flash in the injection molding
process (refer to Section 2.3).
In this section, a formal problem formulation for generating glue faces is presented
first. Three kinds of glue faces are identified. Algorithms for generating them are
described, with the focus on the generation of the inner glue faces.
• Problem Formulation of the Glue Face Generation
Suppose the given parting surface PS intersects with the outside faces of M’ at edges
EFout. The boundary edges of a region on the cavity of M’ are called its parting edges
(PEs). They may compose one or several closed parting loops (CPL), as shown in
Figure 3.26.a. One loop is the outer loop and others are inner loops.
F2
PF
P F1
F7
PF F3 F6
MB F2 F7 F6
F1
F3 M1
CPL1 M'
PF F8
F8
F4 F5
F1
F6
F4 F5
F1
F3 F6
F2
EFout F8
F7
CPL2
F8
M2
(a) Parting edges and EFout in M’ (b) Glue faces (c) Generated mold pieces
Figure 3.26 – Glue Faces of an Example Part.
111
To form mold piece M1 from ↑Fout-M’ and F(R1), and M2 from ↓Fout-M’ and F(R2), the
glue faces should have the given parting edges and EFout as their boundary edges, as
shown in Figure 3.26.b. So the generation of glue faces is actually a geometric
reconstruction problem, that is, generating faces according to their boundary edges. For
given parting edges and EFout, there may exist several solutions. As an example, for the
part given in Figure 3.26.a, the glue face F3 shown in Figure 3.26.b is not required to be
perpendicular to PS. So glue faces will be changed correspondingly, and result in another
mold piece design for the given part.
So the generation of glue faces considered in this dissertation can be formulated as:
Problem GFG: Glue Face Generation. A solid part MB’ is given in the
Boundary Representation, which has two shells with outside faces (Fout) and inside faces
(Fin) respectively. Suppose a planar parting surface PS and a parting direction PD are
given. Further suppose a closed edge loop EFout, which is the intersection of Fout with PS,
is given with several closed edge loops CPLi (1≤ i≤ k), which are composed by some
edges of Fin. Generate faces F1 ~ Fm such that:
a. Fi are planar (1≤ i≤ m) .
b. Suppose set E1(EFout, CPL1,…, CPLk) are all the edges of EFout and CPLi (1≤ i≤ k),
and set E2(F1,…,Fm) are the boundary edges of Union(F1, …, Fm). Two sets
should be the same, that is E1 = E2.
c. Fi (1≤ i≤ m) satisfy the disassembly requirements for a mold piece, that is,
normal(Fi) •PD ≥ 0 (Equation (3.1)).
Among CPLi (1≤ i≤ k), suppose CPL1 is the loop with the biggest projection area in
PS. To connect faces Fout and Fin, and separate faces ~F(R1) and ~F(R2) in M’, three
kinds of glues faces, GFps, GFproj, and GFinner, exist.
(1) Glue face on the parting surface (GFps):
GFps connects the faces of Fout with the outer parting loops CPLi (if CPLi is on PS),
or with project glue faces GFproj (if CPLi is not on PS). Usually GFps has only an outer
loop (Lpso) and an inner loop (Lpsi). The outer loop L pso is formed by the intersection of
Fout with PS, and the inner loop Lpsi is formed by the projection of the CPL1 onto PS. In
Figure 3.26.b, F1 is a GFps.
(2) Glue faces by projection (GFproj):
Disassembling a mold piece in direction PD is similar to sweeping the faces of the
mold piece along PD. Thus we can generate glue faces for edges of CPL1 in the same
principle. GFproj connect CPL1 with GFps. So for each parting edge ei of CPL1, if it is not
on PS, project it onto PS to get a new edge eip. A new glue face is generated by ei, eip and
edges to connect them. In Figure 3.26.b, F2 ~ F7 are GFproj.
An algorithm to generate GFps and GFproj from CPL1 and EFout is presented below.
Algorithm: Glue_Faces_Outer_Loop
Input: A linked parting loop CPL1, parting surface PS and M’.
Output: A set SGF of glue faces GFps and GFproj.
(1) initialize edge array pe, se;
112
(2) for each vertex vi of CPL1,

(3) pv ← projection(vi); // project the vertex into PS
(4) if(pvi ≠ vi) pei ← make_edge(vi, pv);
(5) otherwise pei ← NULL;
(6) for each edge ei of CPL1,
(7) if edge_in_plane(ei, PS) = TRUE, sei ← ei; // edge is in PS
(8) else // edge is not in PS
(9) sei ← projection(ei, PS); // project ei onto PS for sei
(10) gf_proj ← make_face(ei, pei, sei, pei+1);
(11) Add gf_proj to SGF;
(12) Lpi ← make_loop(se); // generate inner loop of GFpf
(13) Lpo ← intersection (PS, Fout(M’)); // generate outer loop of GFpf 4
(14) gf_pf ← make_face(Lpo, Lpi);
(15) Add gf_pf to SGF.
There are only two individual loops in Algorithm Glue_Faces_Outer_Loop. So the
cost of the algorithm is O(ne), where ne is the edge number of loop CPL1.
(3) Inner glue faces (GFinner):
Inner parting loops imply the existence of holes in part P. Correspondingly, inner
glue faces should be generated to separate two mold pieces. If an inner parting loop CPLi
is in a plane, such as loop CPL2 shown in Figure 3.26.a, it is straightforward to generate a
face GFinner according to the loop (F8 shown in Figure 3.26.b). However, in more general
cases, loop CPLi may not be in a plane. Therefore several glue faces are to be generated
for the loop. The generation of inner glue faces according to a given loop is discussed in
more details as follows.
After generating the glue faces for region R1, the glue faces for region R2 can be
generated just by face copying, because glue faces for M1 and M2 are in same positions
with reversed directions.
• Generation of Inner Glue Faces
The shape of inner glue faces is determined by inner parting loops. For an arbitrary
E1
LR1
R1
f3
E2
CPL2 E4
E3 f2
LR2
R2
f1
(a) Part with a snap fit (b) Linked parting loops (c) Resulted inner glue faces
Figure 3.27 – Generation of Inner Glue Faces.
113
inner parting loop, the generation of inner glue faces may be much more complicated. As
an example, a part with a snap fit feature is shown in Figure 3.27.a. Suppose the part
faces are divided into two regions, R1 and R2. Accordingly two closed parting loops are
formed. Between them, CPL2 is an inner parting loop, where the edges of CPL2 are not
in one plane. For CPL2, there are two coedge loops LR1 and LR2 related to R1 and R2
respectively as shown in Figure 3.27.b.
The generation of inner glue faces from inner parting loops is actually a geometric
reconstruction problem. Let A be a linked parting loop, and T a transformation such that
T(A) →B, where B is some planar faces whose boundary is defined by A. One common
reconstruction approach of surfaces from points consists of projecting the points to a
plane, triangulating them, and converting this into a triangulated surface by projecting
each vertex back into three dimensions (Goodman and O'Rourke, 1997). The approach is
often used for cartographic problems. However, in our problem, a plane and the
projection of points are not defined. So it is difficult to find such a plane to avoid the
coincidence of point projections.
To construct inner glue faces for Problem GFG, new edges are to be generated and
they should satisfy two requirements: (1) each new edge is not within a face of M’; (2)
each new edge is not an existing edge in M’. For requirement (1), if a new generated face
or edge is within a face of M’, the mold piece generated by reverse gluing operation will
be invalid because of face overlapping. Similarly, for requirement (2), if a new edge is
an existing edge in M’, we will get a body with an edge shared by more than two faces.
So the mold piece will be non-manifold.
Similar to many geometric reconstruction problems, faces that satisfy all the above
requirements for a loop may be different. In this dissertation a greedy heuristic is used to
minimize the number of generated faces. That is, starting from each coedge ei of a linked
loop, the number of succeeding coedges (nci) that can form a planar face with ei is
counted and assigned to ei. For example, for LR2 shown in Figure 3.27.b, the number for
each coedge is shown in Figure 3.28.a. So every time the approach starts from a coedge
emax with the biggest nci to generate a face. emax and all succeeding coedges are added to
an empty set SE until a succeeding coedge is not in the same plane. If the first and the last
(0) (2) f1 f2 f1
(2) (1) (1)
(1) (0)
(0)
(4)
(0) (0)
(2) (1)
(0) (2)
LR2 (4) f3 (4)
(4)
(a) nc of coedges (b) f1 is generated (c) f2 and f3 are generated
Figure 3.28 – Process of Face Generation.
114
coedges in SE are not linked, a new edge is generated with two related coedges to connect
them. One coedge is added to SE so that the coedges in SE can form a face, which is then
added to a face set. Another coedge is added to the linked loop to replace all the coedges
in SE. Some edges in the loop may need to update their number nc. This process
continues until no more coedges are in the loop. An example for the above process is
shown in Figure 3.28 for the inner parting loop given in Figure 3.27.
When calculating the number of succeeding coedges to form a planar face, it is
required to determine if a new edge to form a new face is valid. For example, in Figure
3.27.b, E1 is an existing edge of the part, and E2 is within a face of R2. So they are all
invalid according to the requirements of new generated edges. For a face with an invalid
new edge, nc of all the coedges except the last one is assigned 0.
The algorithm to generate GFinner is presented as follows.
Algorithm: Glue_Faces_Inner_Loop
Input: A linked parting loop CPLi.
Output: A set SGF with glue faces GFinner.
(1) for each edge ei of CPLi,
(2) nci ← edge_nc(ei, CPLi); // calculate nc for ei in CPLi
(3) while edges of CPLi are not null,
(4) esn ← biggest_edge(nc, CPLi); // find the edge with biggest nc
(5) m ← nc(esn), if m = 0, return error;
(6) set Sce ← null;
(7) add coedge esn, …, esn+ m to Sce;
(8) if esn and esn+ m have a same vertex, ne ← null;
(9) else ne ← make_edge (esn, esn+m); // new edge to form a closed loop
(10) generate coedges cne1 and cne2 from ne; // if ne = null, cne1 and cne2 are null
(11 ) gf_inner ← make_face(Sce, cne1); // generate new face
(12) add gf_inner to SGF;
(13) replace all coedges in Sce with cne2;
(14) update_edge_nc(cne2, CPLi); // update nc for cne2 and its preceding edges
The main loops of the algorithm are Step (3) and (4). Since it takes O(lgn) to get an
element with the biggest number in an array with n elements, the cost of the algorithm is
O(ne•lgne), where ne is the edge number of loop CPLi.
In this section the algorithms for generating glue faces are presented. Based on them
the algorithms for constructing two-piece molds and multi-piece molds are presented
respectively in the next section.
3.6.3 Reverse Glue Algorithm and Parting Surface

Based on the steps of the reverse glue operation (Section 3.6.1) and Algorithms
Glue_Faces_Outer_Loop and Glue_Faces_Inner_Loop (Section 3.6.2), the algorithms of
MPMDM for constructing mold pieces and the selection of parting surface are discussed
in this section.
115
• Algorithm for Two-piece Molds

Based on Equation (3.11) ~ (3.14), the algorithm for generating two mold pieces is
given as follows.
Algorithm: Two_Mold_Piece_Generation
Input: Parting surface (PS) and Boolean mold base M’ with faces Fout(M’), ~F(R1)
and ~F(R2).
Output: Mold piece M1 and M2.
(1) for each vertex vi of the parting edges of ~F(R1),
(2) get edges that share vi;
(3) for each edge ei,
(4) if ei intersects PS, split ei into edge ei+ and ei- by PS;
(5) Use PS to cut Fout(M’) into faces ↑Fout-M’ and ↓Fout-M’.
(6) generate parting loops CPLk from the parting edges of ~F(R1);
(7) classify CPLk into an outer loop CPL1 and several inner loops CPLi;
(8) initialize set SGF ← NULL;
(9) Glue_Faces_Outer_Loop(CPL1, SGF); // generate GFps and GFproj
(10)For each inner loops CPLi,
(11) Glue_Faces_Inner_Loop(CPLi, SGF); // generate GFinner
(12)set SGF’← copy_reverse_faces(SGF); // copy glue faces
(13)Swap partner coedges for all common edges of SGF, ↑Fout-M’ and ~F(R1);
(14)Swap partner coedges for all common edges of SGF’, ↓Fout-M’ and ~F(R2);
(15)Generate body M1 with the faces of SGF, ↑Fout-M’ and ~F(R1), then Regulate M1.
(16)Generate body M2 with the faces of SGF’, ↓Fout-M’ and ~F(R2), then Regulate M2.
Suppose the edge number of the parting edges is ne. Among them, the edge number
of the outer loop CPL1 is nLo, and the edge number of the inner loops CPLi is nLi. So the
cost of Step (1) to (11) is O(ne+ nLi•lg nLi). The cost of Step (13) and (14) depends on ne
and the edge number of the part npe. One simple implementation to find all common
edges is to use two iterations. One checks all edges of ne, and another checks all edges of
npe. So the cost of Step (13) and (14) is O(ne•npe). Therefore the cost of the algorithm is
PS
E
E1
V GFinter1
E3
E2
PE1
Figure 3.29 – An Example for Coedge Splitting.
116
O(ne+ nLi•lg nLi + ne•npe).

The edge splitting process in Step (1) ~ (4) is necessary. Without the splitting, some
coedges of GFproj may not be able to find a partner coedge. So the generated M1 and M2
will be invalid bodies. For example, if the parting surface PS is chosen for a part as
shown in Figure 3.29, glue face GFinter1 is generated by projecting PE1 into PS. If edge E
is not divided into E1 and E2, edge E3 of GFinter1 will not be able to find a partner coedge.
A mold piece generated by the reverse glue approach may have several coplanar
faces. A body regulation operation is called in Step (15) and (16) in order to merge these
coplanar faces.
• Algorithm for Multi-piece Molds
The algorithm to generate two-piece molds can be extended for generating multi-
piece molds.
Algorithm: Multi_Mold_Piece_Generation
Input: Boolean mold base M’ with faces Fout(M’), and ~F(Ri) (1≤ i ≤ n).
Output: Set of mold piece Mi (1≤ i ≤ n).
(1) rank region number i according to the main parting direction and region volumes;
(2) for 1≤ i ≤ n-1, // only need to run n-1 times
(3) assign attributes ~F’(R1) to faces ~F(Ri);
(4) assign attributes ~F’(R2) to all faces ~F(Rj) (i+1≤ j ≤ n);
(5) assign attributes Fout(M’) to all glue faces;
(6) generate parting surface (PSi);
(7) Two_Mold_Piece_Generation(PSi,M’,M1’, M2’); //call two-piece mold function
(8) Mi ← M1’;
(9) if i = n-1, Mn ← M2’;
(10) else, M’ ← M2’.
Step (1) considers the generation order of mold pieces. It is quite obvious using
different orders will give us different results. The two heuristics considered are (i)
generating mold pieces for regions in main parting direction first; (ii) generating mold
pieces for regions with bigger volumes (calculate by bounding box) first. Accordingly
the given regions can be ranked before generating mold pieces for them.
Suppose the number of mold piece is nm, the average number of parting edges for
each region is ne, the average edge number of CPLi is nLi, and the average edge number of
the part is npe. The cost of the algorithm is O(nm•ne + nm•nLi•lg nLi+ nm•ne•npe).
The algorithm based on the Reverse Glue operation is very efficient. First it uses
Euler operations directly rather than sweeping and Boolean operations. Second the
algorithm complexity can be only related to the parting edge number instead of the edge
number of the part, if more advanced data structures are used. That is, for step (13) and
(14) in Algorithm Two_Mold_Piece_Generation, there is no need to determine all edges
of the part if we record all candidate edges for the reverse glue operations in advance.
The number of parting edges is usually far less than the edge number of a part, which is
evident from the examples given in Section 4.4. Therefore the mesh sizes of quadric
surfaces or parametric surfaces can be refined without affecting the execution time of
Algorithm Multi_Mold_Piece_Generation significantly.
117
Step (6) in Algorithm Multi_Mold_Piece_Generation is generating a parting surface

for region Ri. An approach for this step is discussed in more details as follows.
• Selection of Parting Surface
For the approaches based on extending parting lines, the parting surface is decided
entirely by the parting lines. For a given parting direction and some parting lines, there is
only one parting surface (refer to Section 2.3). However in the approach of MPMDM,
parting surface (PS) is also a mold design variable, similar to parting direction (PD) and
parting line (PL).
The principle of the reverse glue operation given in Section 3.6.1 does not have any
requirement on the parting surface. So the approach can generate mold pieces for an
arbitrarily chosen parting surface. For example, three mold designs for a simple part P
are shown in Figure 3.30.a ~ Figure 3.30.c. Their PD and PL are the same. The only
difference is PF1 ~ PF3 are in different positions.
M1 M1 M1
PF3 PF3
M1
e2 e1 e2 e1
PF1 PF2
P
P P P
M2 M2 M2 M2
(a) (b) (c) (d)
Figure 3.30 – An Example of Different Parting Surface.

Although all the designs satisfy the requirements given in Problem MPC, the
fabrication cost related to each design is different. For M1 in Figure 3.30.c, the shape
formed by e1 and e2 is a small protrusion, which may break easily in injection process.
As a worse situation, if e1 becomes vertical to PD, e2 and e1 will form a protrusion with
no volume as shown in Figure 3.30.d. Although theoretically M1 is still a valid shape, it
is not manufacturable anymore. So the designs given in Figure 3.30.a and Figure 3.30.b
are valid for the part, while the designs shown in Figure 3.30.c and Figure 3.30.d are not.
To generate a parting surface for a part with given PD and PL, the heuristic rule used
in the approach is given as follows. First all the vertices of the outer parting loop are
rotated to make PD as the z-axis of the new coordination system. By adding different z
value of each vertex to an array, zpf in the array that has the highest count number is
identified. So the parting surface to be generated is the plane defined by zpf and face
normal PD. User can also change it interactively in RTMDS, which will be discussed in
Section 4.2.
3.7 SUMMARY OF CHAPTER 3
In this chapter a systematic method (MPMDM) is presented for the design of multi-
piece molds for Rapid Tooling. An introduction to MPMDM is given first (Section 3.1).
Then problem formulations and the steps of the method are presented in Section 3.2 and
Section 3.3 respectively. Approaches associated with the different steps are presented in
Section 3.4 ~ 3.6. This chapter provides foundations for developing a mold design
118
system that can be used to design molds to fabricate prototype parts of different
complexity. This mold design system (RTMDS) will be presented in Chapter 4.
Although not explicitly presented in the chapter, the discussions on the different
requirements, information flow and information processing for the different steps of
MPMDM provide partial theoretical structural validation for Hypothesis 1, and the
theoretical analyses of the algorithms provide partial theoretical performance validation
for Hypothesis 1. These results are summarized below (Figure 3.31):
Hypothesis 1: Steps of MPMDM were presented in Section 3.3. The input and output
information of each step provides partial validity of the entire structure of MPMDM.
From the description it is evident that the output from each of the steps are used in
subsequent steps, which provides theoretical structural validation of MPMDM.
Further the final output of the MPMDM is mold pieces that satisfy Problem MCD and
Problem MPD (Section 3.2.2), which compose Problem MD. Therefore the results of
MPMDM were the targeted output of Problem MD. The theoretical structure
validation of the three steps of MPMDM combined, also provides partial structure
validation of MPMDM. The algorithms for each step were analyzed theoretically
(Section 3.4~3.6). These algorithm analyses provide an insight on the performances
Information flow for The method

different steps of performance in
basic element basic element
generation phase generation phase is
are consistent. The acceptable based Information flow for
final elements are on algorithm different steps of The method
the desired output. analysis. region combining performance in
phase are consistent. region combining
The mold piece phase is acceptable
regions are the based on algorithm
desired output. analysis.
H1.1
Theoretical Theoretical
Validation Validation H1.2
Information flow for different The method
phases of MPMDM are performance of
consistent. Information input MPMDM is
for the steps are either from acceptable based on
Information flow for The method
output of other steps or the the theoretical different steps of performance in
initial part information analysis of composing mold piece mold piece
provided. The final mold algorithms.
pieces are the desired output. construction phase construction phase
are consistent. The is acceptable
mold pieces are the based on algorithm
desired output. analysis.
H1
H1.3
Figure 3.31 – Partial Theoretical Validation.
119
of the algorithms for general inputs. Therefore they also provide partial performance
validation of MPMDM. The empirical structural and performance validation of the
MPMDM will be presented in Chapter 4, 7 and 8.
Hypothesis 1.1: Face connectivity and demoldability of mold pieces are two main
concerns in multi-pieces mold design. They were considered in Problem MCD. For
a given part, the convex and concave properties of its faces are a main factor to
decide demoldability of mold pieces, and therefore its mold design. Correspondingly
concave regions and convex faces are proper elements for mold design. Seven
lemmas were presented to provide partial theoretical structural validation of
Hypothesis 1.1 (Section 3.4.1). Although complete splitting can generate a unique
result of concave regions and convex faces for a part, the running time of the
approach is not feasible. In the approach for generating concave regions and convex
faces, some heuristics were considered in selecting splitting surface and splitting
criterion in the region splitting. The approach gives a feasible solution with satisfied
running time. The algorithm analysis of SR (Section 3.4.3) provides partial theoretical
performance validation of Hypothesis 1.1.
Hypothesis 1.2: By generating small regions, the efficiency to explore face
combinations is improved. As a criterion to determine feasible combinations of
regions and faces, parting direction of a region needs to be calculated. The process of
calculating V-Map and making selection from it can be combined into a linear
program (Section 3.5.1). Solving a linear program can give us a satisfactory solution
much more quickly and easily. The algorithm for Problem PDLP, which runs in
linear time and linear storage, provides partial theoretical performance validation of
Hypothesis 1.2. Detection of non-drafted surfaces is an important step in mold design.
Since draft type is tightly related to a parting direction and parting lines, the detection
should be executed in the determination process of parting direction (Section 3.5.2).
In region combination process, it is noticed that mold design knowledge and related
heuristic rules are needed since it is infeasible to explore and generate all the
combinations of regions and faces (NP-hard problem). The algorithm based on
region combination is flexible in adding more design knowledge (Section 3.5.4). The
analysis of the combining process and related algorithms provides partial theoretical
structural validation of Hypothesis 1.2.
Hypothesis 1.3: Automatically splitting the core and cavity inserts is an important step
in the mold design process. Reverse glue operation is tightly related to glue
operation, which has sound theoretic basis in the area of geometric modeling. The
mathematical proof of the reverse glue operation (Section 3.6.1) provides partial
theoretical validation of Hypothesis 1.3. Different glue faces will lead to different
mold piece design. It is better to generate a planar glue surface since it can reduce
mold fabrication cost and material flash in injection process. The generation of inner
glue faces is a geometric reconstruction problem. A formal problem definition of the
glue face generation is presented with two algorithms (Section 3.6.2). From the
algorithm analyses (Section 3.6.3), it is evident that the approach based on the reverse
glue operation is very efficient. It provides partial theoretical performance validation
of Hypothesis 1.3. The approach can provide instantaneous visual feedback on the
mold design results even for a part with rather complex geometries (Section 4.4.3).
120
In the next chapter (Figure 3.32) the Rapid Tooling Mold Design system (Section
4.1), an experimental system that can be used on the design of molds to fabricate
prototype parts with widely varying complexities, will be presented. Two test examples
(Section 4.4) and three industrial cases (Section 4.5) whose molds were automatically
generated by the system will be presented. These examples provide part of the empirical
structural and performance validation of the hypotheses.
Part P Part P
F1 F9 F1
F9
F3 F3
F2 (1) (2)
F6 F8 F2 F6 F8
Fn Fn
F4 F5 F5
F4
F7 F7
Part P
PD1
Mold Base
F1 R3
R1 F9 M1 F1
F3
(3) PD1
F3
F2 F8 F2
PL1
F6
F5 Fn M2 F5 Mk
F4
F4 PL2 Rk F7
PD2
Fn
R2 F7
PD2
Chp 3: Rapid Tooling Mold

Design method

Tools
§2.2 §2.3 §2.4 §2.5 §2.6
Figure 3.32 - Summary of Chapter 3 and Preview of Chapter 4.
121
Chapter 4 – RTMDS and Its Usage for Mold Design of Industrial Parts
CHAPTER 4
RTMDS AND ITS USAGE FOR MOLD DESIGN
Rapid Tooling and multi-piece molding can dramatically reduce the fabrication time
for complex injection molded prototypes. A systematic approach based on regions is
developed for automated design of multi-piece Rapid Tooling molds in Chapter 3. In this
chapter, an experimental system, the Rapid Tooling Mold Design System (RTMDS)
which is based on the MPMDM presented in Chapter 3, is described. First an overview
of the RTMDS is given in Section 4.1. Then the supporting modules and the related
software tools used in the system are presented in Section 4.2. The implementations of
the mold design modules are discussed in Section 4.3. These modules are the cores of the
system. The problems and related limitations in the implementations of the modules are
also presented in Section 4.3. With all the modules presented, two test examples and
three industrial examples are described in Section 4.4 and Section 4.5 respectively.
Finally the relationship between the RTMDS and the validation of the hypotheses are
discussed in Section 4.6.
122
4.1 OVERIEW OF RAPID TOOLING MOLD DESIGN SYSTEM

The method described in Chapter 3 has been employed to develop a Rapid Tooling
Mold Design System (RTMDS). The system is part of the RTTB project (Rapid Tooling
Testbed) under development at Georgia Institute of Technology (Section 1.2.2). The role
of the system is to design mold pieces for a part whose prototypes are to be produced by
Rapid Tooling.
The RTMDS consists of several modules and related data (Figure 4.1). The main
modules of RTMDS are (1) region generation, (2) region combination and (3) mold piece
construction. They correspond to the three steps of the Multi-piece Mold Design method
shown in Figure 3.6. The modules are linked to a graphical user interface (GUI). The
GUI controls the execution order, set options for each module, and graphically displays
the results. ACIS manipulation module and LINGO manipulation module are two
supporting modules which are used by these three main modules. Related to the given
CAD model, the ACIS manipulation module provides functions to evaluate geometry and
topology of the part, and to make necessary modifications to the part. Related to Problem
PDLP (Section 3.5.1), the LINGO manipulation module generates the Linear
Programming formulations that are to be solved by LINGO@. The module also transfers
the results into a parting direction and sends it back to the main modules. Related to the
system modules, the data flow of RTMDS is from CAD models of a part and a mold base
to CAD models of mold pieces. Besides these CAD models, the main data structures of
RTMDS include several arrays for regions (Section 3.5.3) and convex faces. They are
generated by the region generation module and modified by the region combination
LINGO
User
RTMDS
LINGO
LP
Manipulation
Graphical User Interface Formulation
Module
Mold
(1) Region (2) Region (3) Mold Piece Input
Step 2 Base
Step 1 Generation Combination Step 3 Construction
CAD
Module Module Module
Model
ACIS Manipulation
Module
Region Array Mold

Output
Piece
Input Part CAD Convex Face Bodies
Model Array Processor
etc.
Data
RTMDS Data Structures
Face Loop Coedge Edge Vertex
Software
Figure 4.1 – The Organization and Data Flow of the RTMDS.
123
module. Finally they are used by the mold piece construction module for generating
mold pieces.
In this chapter, RTMDS is presented in the following manners. First the supporting
modules (ACIS manipulation, LINGO manipulation, and GUI) and their related data are
introduced in Section 4.2. Their implementations and related software tools (ACIS,
LINGO, and MS-MFC) are also presented Section 4.2. Based on the supporting
modules, the implementation and limitations of the three main modules (region
generation, region combination, and mold piece construction) are discussed in Section
4.3. Finally two test examples and three industrial examples are presented in Section 4.4
and Section 4.5 respectively. They illustrate the usage of RTMDS for mold design
besides verifying the system. Finally a summary of the chapter is given in Section 4.6.
4.2 SUPPORTING MODULES AND THEIR IMPLEMENTATIONS
As shown in Figure 4.1, three main support modules in RTMDS are ACIS
manipulation, LINGO manipulation, and graphical user interface. They support the
implementations of MPMDM (Chapter 3) in the following manners. MPMDM is
developed mainly based on the geometric properties of a part. Consequently the ability
of evaluating and manipulating the part geometries is critical for the system. The author
chose ACIS as the geometry kernel for RTMDS because it is the most powerful solid
modeling kernel today (Section 4.2.1). The part and the mold base inputted to the system
are all represented in ACIS format. Correspondingly the ACIS manipulation module is
developed to support the operations that are related to geometries. To determine parting
directions in the generation and combining processes, a linear programming (LP)
problem needs to be solved (Section 3.5.1). The LINGO manipulation module can
generate the LP formulation related to the parting direction problem. It can also call the
LINGO system to solve the problem and send back the results (Section 4.2.2). Finally,
the Graphical User Interface module allows the user of the system to visually examine the
results of each step. The user can then interactively change design options, or the results
directly based on the displayed results (Section 4.2.3).
4.2.1 ACIS Manipulation Module and Part Representation

Three-dimensional (3D) geometric modelers are central to applications that are
related to 3D geometries. However it is difficult and usually impossible for the
developers of the applications to create a modeler from scratch. Therefore an existing
geometric modeler should be used in an application. In this section, the geometric
modeler used in the RTMDS, ACIS, is introduced briefly first. Then the preparation
process to generate a part that can be read by ACIS is described. Finally the
implementation of the ACIS manipulation module is presented after the data structures of
ACIS are briefly introduced.
• ACIS
ACIS® 3D Geometric Modeler (ACIS) from Spatial Technology Inc. (renamed
Spatial Corporation recently, www.spatial.com) is a well-known solid modeling kernel
for creating an end user application. As the leading 3D geometric modeler in the world,
ACIS has been used in over 200 applications. Some of them are today's leading
CAD/CAM/CAE software applications including AutoCAD 2000.
124
The ACIS kernel consists of a set of C++ classes (including data and member
functions, or methods) and functions (Spatial Technology, 2000). These classes and
functions provide a foundation of common modeling functionality (e.g. Boolean
operations). They also provide users the flexibility to be adapted and extended for
particular application requirements.
ACIS stores the information of a model to ACIS save files. ACIS also restores the
model information from these files. There are two types of ACIS save files: Standard
ACIS Text files (file extension .sat) and Standard ACIS Binary files (file extension .sab).
The only difference between these two files is that the data is stored as ASCII text in a
.sat file and as binary form in a .sab file. The organization of a .sat file and .sab file is
identical. Therefore the term SAT file is generally used to refer to both in this
dissertation.
To get a valid SAT file that can be handled by RTMDS, the converting process for a
CAD model of a part or a mold base is presented as follows.
• The Preparation of Input Files
The RTMDS can only handle the SAT file as the input. In addition, since MPMDM
only considers planar surfaces (Section 3.5), the boundary faces of a part or a mold base
that are inputted to the system must also be planar. Therefore the format for RTMDS
should be a SAT file with only planar faces (called PF-SAT file in this dissertation).
The process to generate a PF-SAT file for RTMDS is shown in Figure 4.2. A given
CAD model can be in any format, e.g. a SAT file, or a PRT file (ProEngineer), or a
SLDPRT file (SolidWorks). The first step is to transfer the CAD file to a STL file. The
STL or stereolithography format is an ASCII or binary file used in manufacturing. It is
the standard input for most rapid prototyping machines. In this process, quadric or
parametric boundary surfaces are approximated with a series of planar faces. The second
step is to transfer the STL file into a SAT file. In this process, all planar faces that are in
the same surface are combined into one face. The topology of faces, loops and edges are
also added to the file. Both STL and SAT are the formats that are supported in most
(1) (2)
CAD STL PF-SAT

file file file
Figure 4.2 – The Process to Generate an Input File for RTMDS.
125
major commercial CAD systems.

In STL file, the shape of the object is defined by a list of the triangular surfaces that
must meet up exactly with each other, without gaps or overlaps. The number of triangles
used can be user-defined. As the number of triangles increases and the relative triangle
size decreases, the shape begins to represent the outline of the quadric or parametric faces
more accurately. This is often referred to as the "facet resolution." CAD programs often
contain optional settings that will give the designer some level of control over the quality
of the STL file output. The most common settings or variables in the CAD systems are
"Chord Height" and "Angle Control." Their values are determined by the tradeoff
between the accuracy requirements and the file size of the model.
• Part Representation and Implementation
ACIS is a boundary-representation (B-rep) modeler, which means that it defines the
boundary between solid material and empty space. This boundary is made from a closed
set of surfaces in some spatial relations. The geometry and topology entities in ACIS are
given in Figure 4.3. Another entity class shown in the figure, which is extensively used
in RTMDS, is Attributes. An attribute is an object that is attached to an entity in an ACIS
model. It stores information about the entity that is not provided within the ENTITY
class itself. Therefore developers can attach application-specific data to entities by
deriving their own attribute classes.
Figure 4.3 – Focus of the ACIS Entity Classes in RTMDS (Spatial Technology,
2000).
126
Since only straight lines and planar faces are considered in RTMDS, the
manipulations related to the geometry entities are simple. However, in order to maintain
a valid model, the manipulations related to the topology entities may be difficult.
Therefore the ACIS manipulation module mainly focuses on managing the topology
entities. The relationship between them is shown in Figure 4.4. Most functions in the
module are related to face, loop, coedge, edge and vertex.
Focus of
RTMDS
Figure 4.4 – The Relationship of Topology Entity Classes in ACIS (Spatial

Technology, 2000).
C++ applications may interface to ACIS through Application Procedural Interface
(API), functions, C++ classes, and Direct Interface (DI) functions. C++ applications for
127
Microsoft Windows platforms may also take advantage of the ACIS interface to
Microsoft Foundation Classes (MFC), which is implemented in the ACIS Microsoft
Foundation Class Component (AMFC). The author adopted C++ classes, functions, and
AMFC in the development of RTMDS. The detail files and classes of the module are
listed in Appendix A.
In the next section, another supporting module of RTMDS is introduced.
4.2.2 Problem PDLP and LINGO®

Problem PDLP was presented in Section 3.5.1 to evaluate a parting direction of some
faces. It is a linear optimization problem since the goal and the constraints are all linear
functions of the variables. Many algorithms and systems can solve the problem. Among
them the Simplex method is the most famous method to solve linear programs (Reklaitis,
et al., 1983). Mainly based on the availability, a linear programming solver, LINGO®, is
used in RTMDS for solving Problem PDLP, which is developed by Lindo Systems Inc
(www.lindo.com).
LINGO provides both the Dynamic Link Library (DLL) and Object Linking and
Embedding (OLE) standards for interfacing with other applications. In RTMDS, the
LINGO DLL is used to access LINGO's functionality. In the implementation of the
LINGO manipulation module, the author found writing all constraints explicitly in a
model file is more robust than using a set with a template model file. The detail
implementation information is described in Appendix A.4.
4.2.3 GUI and Control Options

The Graphical User Interface (GUI) of RTMDS consists of menus, toolbars, and a
drawing area as shown in Figure 4.5. The user can control the running of the system by
Menu
Toolbar
Drawing
Area
Figure 4.5 – Screen Capture of the RTMDS Interface.
128
using the menus and toolbars. The generated results are displayed in the drawing area for
visualization. Usually different colors are used to help the user to interpolate the results.
All the figures shown in Section 4.4 and 4.5 are the screen captures of the results shown
in the drawing area.
The GUI module of RTMDS is implemented based on Microsoft Foundation Classes
(MFC). MFC consist of nearly 200 C++ classes, which are designed by Microsoft to
make Windows programming easier and quicker. Application programs inherit
functionality from MFC as needed. In the development of RTMDS, the ACIS Microsoft
Foundation Class (AMFC) is also integrated with the MFC provided in the Microsoft
Developer Studio. The detail implementation information of the module is also given in
Appendix A.
In the mold configuration design and mold piece construction processes, the user
may want to change some control options based on the running results (Section 4.3).
These control options can be changed interactively in RTMDS. An example of setting
tolerances is given in Figure 4.6.
Figure 4.6 – Setting Tolerance Values in RTMDS.

In light of the supporting modules describe in this section, the implementations of
the main modules, especially the limitations of the system, are discussed in the next
section.
4.3 THE IMPLEMENTATION AND LIMITATIONS OF MOLD DESIGN
MODULES
The three mold design modules discussed in this section are the region generation
module, region combination module, and mold piece construction module (Figure 4.1).
129
All the codes of the modules are written in the C++ language using the Microsoft Visual
C++ 6.0 compiler from personal computers (PC).
The algorithms of MPMDM presented in Chapter 3 are implemented in eight steps in
RTMDS. They correspond to the eight buttons in the toolbar of the system. The steps
and the related descriptions in Chapter 3 are listed as follows.
Step 1: Find concave edges, and generate combined regions and convex faces
(Section 3.4.3).
Step 2: For each region, find parting edges, neighboring faces and a parting direction
(Section 3.5.3).
Step 3: For regions without a parting direction, divide them into several regions such
that each one has a parting direction (Section 3.4.3 and Section 3.5.1).
Step 4: Get region properties (main parting direction and core/cavity) and change
region orders accordingly (Section 3.5.4).
Step 5: Combine neighboring regions and faces into several regions (Section 3.5.4).
Step 6: Judge parting surface for each region (Section 3.6.3).
Step 7: Get suitable mold base for the part; Put it in a proper position and subtract the
part from the mold base (Section 3.6.1).
Step 8: Generate mold pieces for each regions (Section 3.6.2 and Section 3.6.3).
Among them, Steps 1 ~ 3 belong to the region generation module; Steps 4 and 5
belong to the region combination module; and Steps 6 ~ 8 belong to the mold piece
construction module. From Section 4.3.1 to Section 4.3.3, the problems and limitations
identified in the implementation of the three modules are presented individually.
4.3.1 Region Generation

The region generation module is to generate combined regions and convex faces
from the faces of a part. The implementation of the module was based on the algorithms
presented in Section 3.4. The detail implementation information of the module is
provided in Appendix A. In testing the module with different cases, the author identified
three problems related to the region generation process. They also present the limitations
of the RTMDS.
• The Validity of Input Models
For a model input to RTMDS, it is assumed that each edge must be the border
between two and only two polygons. Related to the two polygons, two coedges are
linked to the edge and they should take each other as the partner coedge. Most CAD
models satisfy this requirement. However, in the testing of RTMDS the author observed
some CAD models with coedges whose partner coedges are null. Also since ACIS
support Non-manifold object, some CAD models may have edges that have more than
two coedges. If these CAD models are input to RTMDS, the system may exit abnormally
or even generate a system error.
To generate a PF-SAT file, a CAD model is converted to a STL file first. Then the
STL file is converted to a SAT file. This converting process, especially the translation
from the modeling format to STL, may lead to problems including missing geometry and
loop-orientation inconsistencies. If different CAD systems are used in the process, some
130
small edges of the model may be omitted due to the different tolerances used in the
systems.
The validation of a CAD model is a research area in solid modeling. The Euler’s
Rule can be used to check if a model is manifold (Mäntylä, 1988).
• The Difficulties in the Face Splitting
In Section 3.4.3, Algorithm Split_Region is presented to generate concave regions
from a combined region that does not have a feasible parting direction. The face splitting
in Step (4) is a major operation in the algorithm. A splitting surface (SS) can divide all
the faces of the region into two sides. The faces in one side are added to region CR+, and
the faces in another side are added to region CR-. An illustrative example of the face
splitting of face F is shown in Figure 4.7.
F+
SS SS
F F-
Figure 4.7 – A Simple Example of Face Splitting.

In RTMDS, a face is represented by a polygon (outer loop) with zero or many inside
polygons (inner loops). Depending on the positional relationship of the polygons and the
splitting surface, lots of situations should be considered. For example, an edge of the
outer loop can be in the plane of SS, which is shown in Figure 4.8.a. Also, an edge of an
inner loop can be in the plane of SS as shown in Figure 4.8.b. The face splitting
algorithm in RTMDS is based on the intersection edges of two faces. It can handle
general cases as shown in Figure 4.7, and some special cases as shown in Figure 4.8.
However, a face in a given part may be rather complicated with many special cases like
edge in SS or point in SS (Figure 4.9). Therefore it is rather difficult to develop a general
face splitting algorithm that considers all the special cases.
+ +
F1 F2 F+
SS SS
SS SS
F F -
F
F F-
(a) An edge of the outer loop in SS (b) An edge of an inner loop in SS
Figure 4.8 – Two Examples of Face Splitting.
131
SS
F
Figure 4.9 – A Complicated Example of Face Splitting.

The face splitting is also a research problem in geometric modeling. It is actually
one step of the Boolean operations. However, the author did not find a function in ACIS
to implement this operation. Also if the MPMDM is extended for a triangulated CAD
model, it is straightforward to develop a general face splitting algorithm for triangles.
• Data Structure and Memory Usage
Arrays instead of dynamic pointers are used in the current implementation of
RTMDS, since checking the values of an array is much easier in debugging. However,
this may bring into a problem in the memory usage.
For a complicated part as shown in Section 4.5.3, there are more than 5000 faces.
The size of several arrays, like array of concave edge, convex faces and combined
regions, should be bigger than 3000. Each element of the arrays may take 40 bytes for a
concave edge to 500KB for a combined region (Section 3.5.3). So the memory required
for running the part will be more than 3k x 500kB = 1500 MB. It is beyond the
capability of the computer used to test the system, which has 512 MB DRAM. So when
running the part in RTMDS, an error of low memory was observed and the program
exited abnormally.
If changing all the arrays to pointers, the memory can be applied dynamically from
the operation system (Windows). After finishing the usage of an array, its memory is
released and returned back to the operating system. Therefore the above problem related
to the low memory can be avoided.
4.3.2 Region Combination

From the regions and convex faces generated by the module of region generation, the
region combination module explores their different combinations to minimize the number
of mold pieces. The implementation of the module was based on the algorithms
presented in Section 3.5. The detail implementation information of the module is
provided in Appendix A. The problems and related limitations of the module are
discussed as follows.
132
• Different Combining Order and Options in RTMDS

As discussed in Section 3.5.4, the combining order of regions affects the combining
results, and correspondingly the generated mold pieces. However, it is rather difficult, or
even impossible, to get some general rules on the combining order. Sometimes opposite
rules are observed for different parts. For example, in most cases, neighboring faces in a
same plane should be combined into a region first. However, for a camera roller as
shown in Figure 4.10, it is desired not to combine the neighboring faces in the same plane
first (F1, F2, and F3, F4).
F2
F3 F4 F1
Figure 4.10 – The Combining Order of Faces in a Same Plane.

Consequently several options were provided in RTMDS to control the combining
orders for different parts. These options include (1) if regions in the main parting
direction are combined first; (2) if the core and cavity properties of a region are
considered in the combining process; (3) the combining iteration number before regions
that are not in the main parting direction are added to the combining process; (4) if faces
in the same plane are combined first. The user can set these options quite easily by the
menus of RTMDS. Several interactive tools are also provided in the system to inquire
and change the order of the generated regions.
Besides the order of regions, the order of neighboring faces of a region will also
affect the combining results (Section 3.5.4). However, their relations are not so evident
as the relations between the order of regions and the combining results. Therefore the
order of neighboring faces is not considered in the current implementation of RTMDS.
• Robustness and Accuracy
A main combining criterion for the region combination process is whether a feasible
parting direction exists for all faces. The evaluation approach used in RTMDS is based
on solving Problem PDLP (Section 3.5.1). In the problem, ε1 is a tolerance variable for
the plane constraints: dpi = η xi d x + η yi d y + η zi d z ≥ ε 1 for each face Fi. After solving the
problem, ε2 is another tolerance variable for judging if faces are combinable according to
the solution. That is, suppose the solution of Problem PDLP is v(x, y, z). The
combinability of the faces can be determined by judging the length of v with ε2.
if (v.x•v.x + v.y•v.y + v.z•v.z < ε2) then
Faces are combinable;
else
Faces cannot be combined.
133
Setting proper values of ε1 and ε2 is important for the combining process.

Considering the approximation errors of quadric or parametric surfaces, it is desirable to
set ε1 and ε2 smaller to make the combining process more robust. However, this may also
lead to some undesired results. As an example, a camera roller is shown in Figure 4.11,
in which F is a planar face for the approximation of a cylindrical face. It is desired to
combine face F with region R1. However, if ε1 is set as –0.0174, and ε2 is set as 0.85,
face F is also combinable with region R2 based on Problem PDLP. The plane constraints
and the solution of problem PDLP are also given in Figure 4.11. From the plane
constraints, especially the inequality of F, it is quite obvious that ε1 makes x of v non-
zero.
y x
R1
z
F
R2
Problem PDLP for Determining the Combinability of R1 and F
: MAX= (-0.134028*(x1-x2))+(0.380045*(y1-y2))+(-0.000000 *(z1-z2));
: !The plane constraints;
: (0.000000*(x1-x2))+(1.000000*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.980783*(x1-x2))+(0.195102*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.000000*(x1-x2))+(1.000000*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.000000*(x1-x2))+(0.000000*(y1-y2))+(-1.000000*(z1-z2))>=(-0.017400);
: (-0.980783*(x1-x2))+(0.195102*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.000000*(x1-x2))+(1.000000*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.000000*(x1-x2))+(0.000000*(y1-y2))+(1.000000*(z1-z2))>=(-0.017400);
: (0.000000*(x1-x2))+(1.000000*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (-0.831473*(x1-x2))+(0.555566*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (-0.555566*(x1-x2))+(0.831473*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (-0.195102*(x1-x2))+(0.980783*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.195102*(x1-x2))+(0.980783*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.555566*(x1-x2))+(0.831473*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (0.831473*(x1-x2))+(0.555566*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400);
: (-0.987689*(x1-x2))+(-0.156430*(y1-y2))+(0.000000*(z1-z2))>=(-0.017400); ← F
Solution of Problem PDLP
: v = (-0.2057, 0.94488, 0.0). Length |v| = 0.935.
Figure 4.11 – An Example for Setting Tolerance Value.
134
The combining errors in the beginning stages will affect the determinations
thereafter. Therefore the errors may spread and undesired results may be generated. For
example, as the result given in Figure 4.11, the parting direction of region R1 is v1 (-
0.2057, 0.94488, 0.0) instead of v2 (0.0, 1.0, 0.0). So some faces that are not combinable
in v2 can now be combined in v1; on the contrary, some faces that are combinable in v2
cannot be combined in v1.
In RTMDS, interactive tools are provided for setting values of these tolerance
variables (Figure 4.6). Picking tools are also provided to change faces of a region to a
different region.
4.3.3 Mold Piece Construction

The mold piece construction module is to generate mold pieces from the regions of a
part and a given mold base. The implementation of the module was based on the
algorithms presented in Section 3.6. The detail implementation information of the module
is provided in Appendix A. After testing the module, the author discusses some
limitations of the module as follows.
• Setting Parting Surface
Parting surface is important in the mold piece construction (Section 3.6.3). There are
cases that the parting surfaces automatically generated by the RTMDS were not the same
as what the user expected. The RTMDS also provides interactive tools for the user to
change the parting surface. Two approaches are provided in the system. The user can
pick an existing face or two existing edges. A parting surface will be constructed by the
system accordingly.
• Position the Part Relative to the Mold Base
As discussed in Section 3.6.1, the RTMDS will translate a part to the center point of
a mold base automatically. However, the position of the part in the mold base may need
to be changed in the actual mold design. For example, the positions of possible ejector
pins are fixed by the ejection pattern of an injection molding machine. Therefore the part
should be positioned in the mold base based on the geometry of the part and the positions
of the pins. RTMDS allows the user to change the position of the part interactively.
• Generation of Inner Glue Faces
In Section 3.6.2, Algorithm Glue_Faces_Inner_Loop was presented for the
generation of inner glue faces. In the algorithm, the greedy heuristic is used to minimize
the number of generated faces. The algorithm can handle many cases, including the cases
shown in Figure 3.25 and Figure 3.26. However, it is also observed that an inner loop of
a part can be rather complicated. Consequently the generation of inner glue faces for the
loop can be difficult. An illustrative example is given in Figure 4.12. The inner parting
loop is shown in the figure with the parting surface. RTMDS was unable to generate
inner glue faces from the loop that satisfies Problem GFG given in Section 3.6.2.
It is noticed that no matter how complicated the inner loop can be, the generated
inner glue faces should satisfy Problem GFG. Therefore the author believes Problem
GFG is a general problem formulation for the generation of inner glue faces. No
publications in the areas of computational geometry and CAD were found that discussed
135
Inner Parting
Loop Parting
Surface
Figure 4.12 – Mold Pieces for a Pocket with Empty V-Map.

Problem GFG or similar geometric problems. However, in order to construct mold
pieces for an injection molded part, the problem should be studied further.
• Mold Piece Construction Order
In Algorithm Multi_Mold_Piece_Generation (Section 3.6.3), mold pieces are
constructed in the order of region R1, R2, …, Rn. Before the algorithm is executed, the
user of the RTMDS can also interactively change the construction order via the tools
provided by the system.
In light of the RTMDS and its modules described in the last section and this section,
some testing examples of the RTMDS will be presented in Section 4.4 and 4.5. These
parts have widely varying complexities as shown in Table 4.1.
Table 4.1 – The Complexities of the Example Parts in Section 4.4 and 4.5.
Complexity Example Part Example Part Industrial Part Industrial Part Industrial Part
1 (Section 2 (Section 1 (Section 2 (Section 3 (Section
4.4.1) 4.4.2) 4.5.1) 4.5.2) 4.5.3)
Low X X
Medium X X
High X
4.4 INTUITIVE EXAMPLE PARTS

The main purpose of the examples given in this section is to test the methods
presented in Chapter 3. Since the examples are intuitive, it is easier to verify the
implementation of the system described in the last two sections. There are four steps that
136
are taken in each example to verify the results from the system are correct and make
sense. These steps or tasks are:
Task 1: Verify the regions and the convex faces generated by the system are the
same as what are expected from the method;
Task 2: Verify the regions after combining are the same as what are expected
from the method;
Task 3: Verify the generated mold pieces can form the combined regions inside
and the mold base outside;
Task 4: Examine the running time of each module to determine whether they are
in accordance with the algorithm analysis of MPMDM (Chapter 3).
Besides industrial parts as shown in Section 4.5, the author constructed more than 10
example parts and tested them in the RTMDS. The two parts presented in this section are
chosen because (1) they represent the two most common undercuts, protrusion and
extrusion; (2) they are the simplest parts that are related to two-piece molds and multi-
piece molds respectively.
A detail description of the results for each step of the RTMDS is given in Section
4.4.2 because the complexity of the test example 2 is suitable for explaining the running
results. It also illustrates the correctness of the implementations of the system. For other
examples given in this and the next sections, the results of each example are organized in
the following manners. First, the information of the part (the size of the file, face number,
etc.) is presented in a table. The screen captures of the graphical results of some key
steps given by the RTMDS are given in figures. The information related to these steps is
also provided in the table. Finally the running time of each step and the total time are
listed in the table. All the tests in this dissertation are based on a personal computer with
a 700 MHz Intel-III processor.
4.4.1 Test Example 1: A Box with a Rib

Input SAT file size: 11 KB.
The first test example is a rib part as shown in Figure 4.13. It is a simple part with
only 11 faces. The part has only one protrusion feature, and two mold pieces are
sufficient for producing the part. Table 4.2 lists the information regarding the part,
generated regions, and the execution time of each step. Figure 4.14 shows the graphical
outputs of the RTMDS. Different regions and convex faces (CVX) are marked with
different colors.
137
Figure 4.13 – A Test Example of a Rib Part.
Table 4.2 – The Information for Test Example 1.

Part Info. Face No. Concave face No. Concave edge No.
11 5 4
Region Info. Initially generated After dividing After combining
1 1 1
Reverse glue Info. CPL No. Edge # of CPL1 Edge # of CPLi
1 4 0
GFps No. GFproj No. GFinner No.
1 0 0
Step 1 Step 2 Step 3
Running Time (s) 0.01 0.44 0.00
0.12 0.38 0.01
Step 7 Step 8 Total Time
0.02 0.29 1.27
Besides the results automatically generated by the RTMDS, the user can also change
some faces of a region to the set of convex faces interactively. In Figure 4.15.a, the
region and convex faces of the part are shown, which are slightly different from the
results shown in Figure 4.14.b. Correspondingly, different mold designs are generated by
the RTMDS, which are also shown in Figure 4.15.
138
Region
Region
Convex Convex
Faces Face
(a) Region Generation (1 region+ 6 CXFs) (b) Region Combination (1 region+ 1 CXF)
(c) Mold Base and the Parting Surface
(d) Two Generated Mold Pieces

Figure 4.14 – Graphical Results of a Mold Design for Test Example 1.
139
Region
Parting
Surface
Convex
Faces
(a) Region Combination Result with the Parting Surface (1 region + 5 CXFs)
(b) Mold Base and the Parting Surface

Figure 4.15 – Graphical Results of Another Mold Design for Test Example 1.
140
4.4.2 Test Example 2: A Box with a Through Hole and Two Grooves
Figure 4.16 – A Test Example of a Box with a Through Hole and Two Grooves.
The second test example is a box with a through hole and two grooves as shown in
Figure 4.16. It is a simple part with only 18 faces. The part has three extrusion features.
It cannot be made by two mold pieces. Therefore multi-piece mold design is needed for
producing the part. Table 4.3 lists the information regarding the part, generated regions,
reverse glue operation, and the execution time of each step.
Table 4.3 – The Information for Test Example 2.

18 12 14
3 3 2
CPL No. Edge # of CPL1 Edge # of CPLi
Reverse glue Info. For R1 2 4 4
1 0 1
Reverse glue Info. For R2 1 12 0
1 2 0
0.01 0.49 0.0
Running Time (s) Step 4 Step 5 Step 6
0.05 0.45 0.00
0.02 0.32 1.34
A detail description of the results generated in each step is given as follows. It can
familiarize the reader with the notions given in the result table. It also illustrates the
correctness of the implementations of the RTMDS.
141
Step 1: Based on the criterion of judging convex and concave edges (Section 3.4.3),
there are 14 concave edges (5+5+4) in the part as shown in Figure 4.17. If a face
contains one or more concave edges, it is a concave face. Therefore 12 concave faces
(4+4+4) are identified. Based on their connectivity, three regions are generated for
all the concave faces (Figure 4.17). All the remaining faces are convex faces.
Therefore the total face number is 12+6 = 18.
R3
5 concave edges
R2 4 concave faces
5 concave edges 1 region
4 concave faces
1 region
R1
4 concave edges
4 concave faces
1 region
6 convex faces
z
x
y
Figure 4.17 – Illustration of the Running Results of Test Example 2.

Step 2: For each region, the parting edges and neighboring faces are recorded in the data
structure of the RTMDS. Also a parting direction for each region is generated by the
LINGO system. They are: PD1 for R1 is (0.0, 1.0, 0.0), PD2 for R2 is (-0.707, 0.0,
0.707), PD3 for R3 is (0.707, 0.0, 0.707).
Step 3: Since all regions have a parting direction, this step is skipped. However, for
industrial example 2, this step is critical since a generated region R has no feasible
parting direction (Figure 4.24).
Step 4: The main parting direction of the part is in y-axis because the volume of the
bounding box of R1 is the biggest. Also R1 is a core since one of its edge loops is an
inner loop of a face. Instead R2 and R3 are cavities since all their edge loops are outer
loop of faces.
Step 5: Based on the connectivity of faces and combining criterion (Section 3.5.1), R1
and a convex face are combined into a region. R2, R3 and a convex face are combined
into a region (Figure 4.18.b). In the figures, different regions and convex faces
(CVX) are marked with different colors. After the region combination process, 2
regions (R1, R2) and 4 convex faces are generated. For the four convex faces, a new
region R3 is generated.
142
Step 6: The parting surfaces generated for R1 and R2 are shown in Figure 4.18.d and
Figure 4.19.c respectively. The parting surface for R3 is not used in the mold piece
construction process, therefore it is not shown here.
Step 7: A mold base is generated based on the size of the part. After it is positioned as
shown in Figure 4.18.d, the Boolean operation is executed with the part.
R3
R2 R2
R1
R1
Convex
Faces
Convex
Faces
(a) Region Generation (3 regions + 6 CXFs) (b) Region Combination (2 regions + 4 CXFs)
R2
R1
Convex
Faces
(c) Another View of the Region Combination Results (d) Mold Base and the Parting Surface
Figure 4.18 – Graphical Results of Mold Design for Test Example 2. (Step 1~7)
143
GFinner
GFps
(a) Mold Piece M1 for R1
GFinner
GFps
(b) Mold Piece M2’ for R2 and R3 (c) Mold Base and the Parting Surface
GFps GFproj GFproj
GFps
(d) Mold Piece M2 for R2 (e) Mold Piece M3 for R3
(f) Putting Three Mold Pieces Together

Figure 4.19 – Graphical Results of Mold Design for Test Example 2 (Step 8).
144
Step 8: Since three regions are generated in the region combination process (Figure
4.18.b and c), there are two phases in Algorithm Multi_Mold_Piece_Generation.
Figure 4.19.a and b show the graphical outputs of the first phase. Figure 4.19.d and e
shows the graphical outputs of the second phase. In the generated mold pieces, the
generated glue faces are in the same color as those of the related regions. In the first
phase, there are two parting loops (related to R1). Each of them has 4 edges. Since
all the edges of CPL1 are in the parting surface, there is no GFproj (Section 3.6.2).
Related to the two parting loops, faces GFps and GFinner are generated as shown in
Figure 4.19.a and b. In the second phase, there is only one parting loops with 12
edges (related to R2). Among them, six of the parting edges are not in the parting
surface, and they form two GFproj and one GFps (Figure 4.19.d and e). There is no
GFinner in the second phase.
After the correctness of the implementations of the RTMDS is illustrated, the
running time of each step is also given in Table 4.3. From the results, it is evident that
the RTMDS is also very efficient. Since the running time for this example is too short,
further analysis on the computation time of the RTMDS is given in Section 4.5.3.
In the next section, three industrial parts are presented to further verify the mold
design method, and also to demonstrate that the RTMDS is capable to handle more
complex cases.
4.5 INDUSTRIAL CASES
In this section three industrial parts whose molds were designed by the RTMDS are
presented. For each example the four tasks described in Section 4.4 are also followed to
verify that the results from the system are correct and make sense. In addition, the author
also physically validated the mold design of industrial example 1 and produced
prototypes for the part design.
The examples were chosen as the case studies of the RTMDS mainly for the reasons
given as follows.
(1) The three example parts are typical injection molded parts;
(2) The complexity of the parts covers a typical range of part complexities based on
the face number of a part. The three examples have face number from 330 to
5000, which correspond to medium complexity to high complexity.
(3) The size of the first example is suitable for an existing mold base of a Morgan-
Press injection molding machine, which is located at RPMI. Therefore it can be
used to test the RTMDS in loading an existing CAD model of a mold base.
(4) The second example has a combined region which does not have a feasible
parting direction. So the region needs to be divided into several concave regions.
Also the part has several inner parting loops. Therefore the algorithms of dividing
regions and generating inner glue faces can be tested by the second example.
(5) The third example has more than 5000 faces. It is used mainly to test if the
RTMDS can handle parts with high complexity.
The three examples are organized based on their complexities. From industrial
example 1 to 3, the part complexity is increased. The results of each example are
organized in the same way as that of the examples given in Section 4.4. The information
145
of the part and the mold design steps is presented in a table. The screen captures of the
graphical results of some key steps given by the RTMDS are given in figures. All the
tests are also based on a personal computer with a 700 MHz Intel-III processor.
4.5.1 Industrial Example 1: A Housing

The first industrial example is a housing for a telephone adapter as shown in Figure
4.20. The original part was produced by the injection molding process. There are both
protrusion and extrusion features in the part. Table 4.4 lists the information regarding the
part, generated regions, and the execution time of each step. Figure 4.21 shows the
graphical outputs of the RTMDS. Different regions are marked with different colors.
The standard mold base used in a Morgan-Press injection molding machine is also shown
Figure 4.20 – A Housing for A Phone Adapter.
Table 4.4 – The Information for Industrial Example 1.

330 207 223
28 28 2
Reverse glue 1 54 0
Info. GFps No. GFproj No. GFinner No.
1 23 0
Glue_Faces_Outer_Loop Glue_Faces_Inner_Loop Multi_Mold_Piece_Generation
0.03 0.0 1.57
0.06 2.40 0.0
Running Time Step 4 Step 5 Step 6
(s) 0.04 7.15 0.36
0.94 1.75 12.7
146
in Figure 4.21.c.
The mold design generated by the RTMDS (Figure 4.21.d) is further verified by
physical validations. First the mold design is built directly by a SLA-3500 machine. A
photo of the built mold pieces is given in Figure 4.22.a. Using the SLA mold pieces,
more than 10 functional prototypes of the part are produced by the Morgan-Press
machine, which are shown in Figure 4.22.b. The material of the prototypes is
polystyrene. The parameters used in the injection molding machine are:
Temperature (Barrel / Nozzle): 430 / 450 oF.
Pressure (Injection / Pilot): 2450 / 70 psi; Clamp Force: 13 tons.
Time (injection / Cooling): 13 / 390 second.
Therefore, it is validated that the mold design for the housing (Figure 4.21.d) can
produce the part by the Rapid Tooling process.
147
R1
R2
(a) Region Generation (28 regions) (b) Region Combination (2 regions)
(c) Mold Base of the Morgan-Press

Figure 4.21 – Graphical Results of a Mold Design for Industrial Example 1.
148
(a) SLA Mold Pieces
(b) Injection Molded Parts

Figure 4.22 – Physical Validation of the Mold Design for Industrial Example 1.
149
4.5.2 Industrial Example 2: A Thin Wall Part

(a) Top View (b) Bottom View
Figure 4.23 – A Thin Wall Part for A Phone Adapter.

The second industrial example is a thin wall part as shown in Figure 4.23. Both the
top view and the bottom view are given. The part is a component of the telephone
adapter whose housing is presented in Section 4.5.1. The original part was produced by
the injection molding process. The part is rather complicated with several through holes
and slots. Table 4.5 lists the information regarding the part, generated regions, and the
execution time of each step. Figure 4.24 shows the graphical outputs of the RTMDS.
Different regions are marked with different colors. A combined region shown in Figure
4.24.a is divided into eight concave regions as shown in Figure 4.24.b. In the dividing,
face F1 and F2 (Figure 4.24.a) are the splitting surfaces because they have the biggest
number of convex internal edges (Section 3.4.3). Finally the generated mold pieces for
the part are given in Figure 4.25.

606 489 543
31 38 2
Reverse glue 12 50 184
1 14 11
Glue_Faces_Outer_Loop Glue_Faces_Inner_Loop Multi_Mold_Piece_Generation
0.03 18.0 21.0
Running Time 0.18 2.84 5.15
(s) Step 4 Step 5 Step 6
0.13 10.4 0.02
1.08 21.04 40.84
150
F2
R
F1
(a) Region Generation (31 regions)
R8
R4
R3 R1
R2
R7
R6
R5
(b) Region Dividing (38 regions, R is divided into R1 ~ R8)
R1
R2
(c) Region Combination (2 Regions)
Figure 4.24 – Graphical Results of a Mold Design for Industrial Example 2.
151
Figure 4.25 – Generated Mold Pieces for Industrial Example 2.
4.5.3 Industrial Example 3: A Complex Housing

Input SAT file size: 5.3 MB.
The third industrial example is a complex housing as shown in Figure 4.26. The
original part was produced by the injection molding process. As discussed in Section
4.3.1, the current implementation of the RTMDS requires more than 1500MB memory
for this part. Therefore in the PC used for testing the system, an error of low memory
was given by the operation system when loading the part in the system. However, the
author used interactive tools that are provided by the RTMDS to divide the faces into two
regions. The two regions are marked with different colors and shown in Figure 4.27.
The par was used to test the module of mold piece construction in the RTMDS. Table
4.6 lists the information regarding the part, the reverse glue operation, and the execution
time of the algorithms used in the module (Section 3.6). Finally the generated mold
pieces for the part are given in Figure 4.28.a. The author also built the mold pieces using
a SLA-3500 machine. A photo of the built mold pieces is shown in Figure 4.28.b.
152
Figure 4.26 – A Complex Housing.
R1
R2
Figure 4.27 – Region Combination (2 regions).

Face No. Edge No.
Part Info. 5493 9344
Concave face No. Concave edge No.
4880 4448
Reverse glue 1 234 0
1 13 0
Computation Glue_Faces_Outer_Loop Glue_Faces_Inner_Loop Multi_Mold_Piece_Generation
Time (s) 0.18 0.0 40.5
153
(a) CAD models of two mold pieces
(b) Two mold pieces made by SLA

Figure 4.28 – Mold Pieces for Industrial Part 3.
From the running results shown in Table 4.6, it can be observed that the computation
times of algorithms Glue_Faces_Outer_Loop and Glue_Faces_Inner_Loop are mainly
affected by the edge number of CPL1 and CPLi respectively. The industrial part 2 is an
interesting example with several inner parting loops. From its results, it is noticeable that
our algorithm to generate the inner glue faces GFinner, which is O(ne•lgne) (refer to
Section 3.6.2), is more time-consuming than the algorithm to generate GFps and GFproj,
154
which is only O(ne) (Section 3.6.2). Besides the edge number of CPL1 and CPLi, the
computational time of algorithm Multi_Mold_Piece_Generation is also affected by the
edge number of the part npe. These observations are all in accordance to our algorithm
analysis presented in Section 3.6.
Based on the test examples and the industrial examples presented in Section 4.4 and
4.5, an evaluation of the algorithms of the RTMDS is provided in the next section.
4.6 EVALUATION OF EXAMPLES AND CASES
In Chapter 3, algorithm analyses of several key steps of the Multi-Piece Mold Design
Method were presented. Comparing the theoretic analysis results with the experiment
results obtained from the RTMDS, we can partially validate the implementation of the
RTMDS. Based on the examples presented in this chapter and the case studies presented
in Chapter 7 and 8, the running time of the three stages is discussed individually.
4.6.1 Region Generation Process

The region generation process is the first step of the Multi-Piece Mold Design
Method (Section 3.3). For the algorithms presented in Section 3.4.3, the running time of
the region generation process is O(ne + nf2), where ne and nf are the edge number and the
face number of a given part respectively. In the RTMDS, steps 1 to 3 are related to the
region generation process. By adding the running time of step 1 ~ 3, we get the running
time of the region generation process as shown in Table 4.7. It is noticed that industrial
example 2 is not considered here because it is the only part which has regions to be
divided in step 3.
Table 4.7 – Experimental Data of Region Generation Process.

Total Face Concave Face Total Edge Concave Edge Region Generation
No. (nf) No. No. (ne) No. Time (second)
Test Exp1 11 5 24 4 0.45
Test Exp2 18 12 48 14 0.5
Robot Arm 92 66 206 40 1.84
Industrial Exp1 330 207 712 223 2.46
Camera Roller 544 336 1274 446 3.05
The relations of the running time of the region generation process with the face and
edge number of parts are shown in Figure 4.29. The lines in the graph are in accordance
with the algorithm analysis result.
155
Region Generation
3.5
3
Running Time (sec)
2.5
2
Total Face No.
1.5 Concave Face No.
1 Total Edge No.
Concave Edge No.
0.5
0
0 500 1000 1500
Face/Edge No.
Figure 4.29 – Relations of Region Generation Time with Face/Edge Number.
4.6.2 Region Combination Process

The approaches for the region combination process utilized in the RTMDS are
presented in Section 3.5. Based on the algorithm analysis (Section 3.5.4), the running
time of the region combination process is O(nf2), where nf is the face number of a given
part. In the RTMDS, steps 4 to 6 are related to the region combination process. By
adding the running time of step 4 ~ 6, we get the running time of the region combination
process as shown in Table 4.8. It is noticed that the camera roller is not considered here
because its combination process has three iterations instead of one (Figure 8.5).
Table 4.8 – Experimental Data of Region Combination Process.

Total Face Concave Total Edge Concave Region Combination
No. (nf) Face No. No. Edge No. Time (second)
Test Ex1 11 5 24 4 0.51
Test Ex2 18 12 48 14 0.5
Robot Arm 92 66 206 40 0.72
Industrial Exp1 330 207 712 223 7.55
Industrial Exp2 606 489 1337 543 10.55
The relations of the running time of the region combination process with the face
and edge number of parts are shown in Figure 4.30. The lines in the graph are in
accordance with the algorithm analysis result.
156
Region Combination
12
10
Running Time (sec)
8 Total Face No.

Concave Face No.
6
Total Edge No.
4 Concave Edge No.
0
0 500 1000 1500
Face/Edge No.
Figure 4.30 – Relations of Region Combination Time with Face/Edge Number.
4.6.3 Mold Piece Construction Process

The region generation process is the last step of the Multi-Piece Mold Design
Method (Section 3.6). For the algorithms presented in Section 3.6.3, the running time of
the region generation process is O(npe + nLi • lg nLi + npe • ne), where ne, nLi and npe are
the total edge number, the inner loop edge number, and the parting edge number of a
given part respectively. In the RTMDS, steps 7 and 8 are related to the mold piece
construction process. By adding the running time of step 7 and 8, we get the running
time of the mold piece construction process as shown in Table 4.9.
Table 4.9 – Experimental Data of Mold Piece Construction Process.

Parting Mold Piece
Total Concave Total Edge K* ne *
Edge No. ne * npe Construction
Face No. Face No. No. (ne) npe
(npe) Time (Second)
Test Exp1 11 5 24 4 96 0.48 0.31
Test Exp2 18 12 48 8 384 1.92 0.34
Industrial Exp1 330 207 712 54 38448 192.24 2.69
Industrial Exp2 606 489 1337 234 312858 1564.3 22.12
Industrial Exp3 5493 4880 9344 234 2186496 10932.5 40.5
The relations of the running time of the mold piece construction process with the
face and edge number of parts are shown in Figure 4.31. A constant K is multiplied to
each value in the column of ne• npe to make the value of ne• npe fit in the figure. The lines
in the graph, especially k• ne• npe, are in accordance with the algorithm analysis result.
157
Mold Piece Construction
45
40
Running Time (sec)
35
30 Total Face No.
25 Concave Face No.
20 Edge No.
15 K*ne*npe
10
5
0
0 2000 4000 6000 8000 10000 12000
Face/Edge No.
Figure 4.31 – Relations of Mold Piece Construction Time with Face/Edge Number.
4.6.4 The Whole Process of RTMDS

Finally, the total running time of the mold design process is shown in Table 4.10. It
corresponds to the sum of the running time of steps 1 ~ 8 in the RTMDS. The relations
of the total running time with the face and edge number of parts are shown in Figure
4.32. The lines in the graph illustrate the running time behavior of the RTMDS for a
given part.
Total Running Time
45
40
35
Running Time (sec)
30
Total Face No.
25 Concave Face No.
20 Total Edge No.
15 Concave Edge No.
10
5
0
0 500 1000 1500
Face/Edge No.
Figure 4.32 – Relations of Mold Design Running Time with Face/Edge Number.
158
Table 4.10 – Experimental Data of Mold Design Process.

Total Concave Total Edge Concave Total Running
Face No. Face No. No. (ne) Edge No. Time (Second)
Test Exp1 11 5 24 4 1.27
Test Exp2 18 12 48 14 1.34
Industrial Exp1 330 207 712 223 12.7
Industrial Exp2 606 489 1337 543 40.84
In the next section, a brief summary is given which discusses the relevance of these
results with regard to the hypotheses of the dissertation.
Mold design is a laborious process that requires significant time from the mold
designer. Automated mold design significantly reduces the mold design time, and
therefore reduces the lead-time of Rapid Tooling process in producing functional
prototypes. In this chapter, a Rapid Tooling Mold Design System (RTMDS) is presented
for the automated design of multi-piece molds for Rapid Tooling. An introduction to the
RTMDS is given first (Section 4.1). Then the supporting modules and the software tools
used in the system are presented in Section 4.2. The mold design modules are the
implementation of the Multi-piece Mold Design Method (Chapter 3). Problems in the
implementation of the modules are described in Section 4.3. Finally two test examples
and three industrial examples are presented in Section 4.4 and Section 4.5 respectively.
The RTMDS and the given case studies provide partial empirical structural and
performance validation of Hypothesis 1 (Figure 4.33). The results from testing the
hypothesis are summarized as follow.
Hypothesis 1: The implementations of the RTMDS were presented in Section 4.2 and
4.3. The case studies for testing the system are described in Section 4.4 and 4.5.
Although not explicitly presented in the sections, the discussions on the relations and
the implementations of the modules of the RTMDS provide partial empirical
structural validation of Hypothesis 1, and the discussions on the results and the
running time of different examples provide partial empirical performance validation
of Hypothesis 1. The summary of testing the sub-hypotheses (H1.1 ~ H1.3) presented
below will provide more details.
Hypothesis 1.1: The region generation module is developed based on the basic elements
of concave regions and convex faces (Section 4.3.1). The module can generate the
basic elements of mold design for several test parts and industrial parts in acceptable
time (Section 4.4 and 4.5). The generated elements can be used in the region
combination process for mold configuration design.
Hypothesis 1.2: The region combination module is developed based on the concave
regions and convex faces generated in the region generation module (Section 4.3.2).
The module can generate mold configuration design for several test parts and
industrial parts in acceptable time (Section 4.4 and 4.5). The generated mold
configuration design can be used in constructing mold pieces for the parts.
159
H1.3
H1
Mold pieces of the
Develop the mold piece
case stuides are
constructing module for
generated by the
Mold designs of the the combined regions
module in acceptable
Develop the RTMDS based case stuides are and a mold base. The
time, and they can be
on the MPMDM. The generated by the module can generate
used in the Rapid
system can be used to RTMDS in acceptable mold pieces for several
Tooling process to
generate mold design for time, and they can be test parts and industrial
produce prototypes.
several test parts and used to produce parts.
industrial parts. prototpyes in the Rapid
Tooling process.
Empirical Empirical
Validation Validation H1.2
Develop the region Mold configurations
combining module
H1.1 based on the concave
of the case stuides
are generated by the
regions and convex module in acceptable
faces. The module can time, and they can be
Develop the region Basic elements of the
generate mold used in constructing
generation module based case stuides are
configuration design for mold pieces for the
on the basic elements of generated by the
several test parts and parts.
concave regions and module in acceptable
industrial parts.
convex faces. The time, and they can be
module can generate the used in the region
basic elements of mold combining process of
design for several test mold configuration
parts and industrial parts. design.
Figure 4.33 – Empirical Structural and Performance Validation for Hypothesis 1.

Hypothesis 1.3: The mold piece construction module is developed according to the
combined regions of a part and a given mold base (Section 4.3.3). The module can
generate mold pieces for several test parts and industrial parts in acceptable time
(Section 4.4 and 4.5). The generated mold pieces can be used in the Rapid Tooling
process to produce functional prototypes.
Automatic generation of mold design is also a key step of Design-for-Manufacturing
system for injection molding process (Chapter 6). After CAD models of molds are
generated, detailed Design-for-Manufacturing feedback can be provided to the designer,
and RP process planning and Injection molding simulation is enabled. In the next chapter
(Figure 4.34), geometric tailoring, which is part of design-for-manufacture (Figure 1.14),
will be presented. After discussing the principles of functional testing (Section 5.2) and
design decision templates (Section 5.3), the usage of the template for Rapid Tooling is
presented in Section 5.4. Three case studies, a tensile bar, a rib part and a ring gear, are
used to test the presented scenario (Section 5.5). They also provide partial validation of
Hypothesis 2.1.
160
Part P Part P
F1 F9 F1
F9
F3 F3 Given
F2 (1) (2)
F6 F8 F6 F8
Analternative to be improved throughmodification; Assumptions used tomodel the domain of interest.
F2 The systemparameters:
Fn Fn
p equality constraints q inequalityconstraints
F4 F5 F5
fk(di ) function of deviation variables to be minimized at priority level k for thepreemptive case.
F4 Find
F7 F7 Values for the deviation variables di-, di+ i = 1, ... , m
Satisfy
Part P gi(X) = 0 ; i = 1, ..., p gi(X) ≥ 0 ; i = p+1, ..., p+q
PD1
Mold Base Bounds
Ai(X) + di- - di+ = Gi ; i = 1, ..., m
F1 R3 Xi ≤ Xi ≤ Xi ;
min max
i = 1, ..., n
R1 F9 M1 F1 Deviation variables
F3 F3
di-, di+ ≥ 0 ; di- . di+ = 0 ; i = 1, ..., m
(3) PD1 Minimize
F2 F8 F2 Preemptive deviation function (lexicographic minimum)
PL1
F6 Z=[f1(d-i,d+i ),..., fk(d-i, d+i )]
Archimedain deviation function
F5 Z= ∑Wi (di− + di+) where ∑W =1, W ≥ 0
F5 Fn M2 F4 Mk i i
F4 PL2 Rk F7
PD2
Fn
R2 F7
PD2

Design method

Tools
§2.2 §2.3 §2.4 §2.5 §2.6
161
Chapter 5 – Formulating Design Requirements for Rapid Tooling as Geometric Tailoring Problem
CHAPTER 5
FORMULATING DESIGN REQUIREMENTS FOR RAPID

TOOLING AS GEOMETRIC TAILORING PROBLEM
Given
Analternative tobe improvedthroughmodification; Assumptions usedtomodelthe domainof interest.
The systemparameters:
p equalityconstraints q inequalityconstraints
fk(di ) functionof deviationvariables tobe minimizedat prioritylevel kfor the preemptive case.
Find
Values for the deviationvariables di-, di+ i = 1, ... , m
Satisfy
gi(X) = 0; i = 1, ..., p gi(X) ≥ 0; i = p+1, ..., p+q
Ai(X) + di- - di+ = Gi ; i = 1, ..., m
Bounds
Ximin ≤ Xi ≤ Ximax ; i = 1, ..., n
Deviationvariables
di-, di+ ≥ 0; di-. di+ = 0; i = 1, ..., m
Minimize
Preemptive deviationfunction(lexicographic minimum)
Z=[f1(d-i ,d+i ),..., fk(d-i,d+i )]
Archimedaindeviationfunction
Z= ∑W(d i
−
i
+ di+) where ∑W =1, W ≥0
i i
Chp 5: Geometric Tailoring
In the current usage of Rapid Tooling, the iterations of design changes between the
designer and manufacturer may take a long time before production-representative
prototypes are produced. In this chapter, a geometric tailoring approach, material
geometric tailoring, is presented to reduce the iterations due to the material property
differences between the products and prototypes. First the properties of Rapid Tooling,
especially the part properties of the AIM tooling, are introduced to provide a context of
material geometric tailoring (Section 5.1). Related to the principle of functional testing
and similarity methods, the fundamentals of geometric tailoring are presented in Section
5.2. The material geometric tailoring decision template is introduced in Section 5.3,
which enables a “clean digital interface” between design and fabrication, effectively
separating design activities from manufacture activities. The usage of the MGT decision
template, including formulating function properties and solving approaches, is presented
in Section 5.4. Finally three test examples are discussed in Section 5.5 to demonstrate a
scenario of design-manufacture collaboration with the MGT decision template.
162
5.1 PROPERTIES OF RAPID TOOLING

As discussed in Section 1.2, the two problems in the current usage of Rapid Tooling
(RT) that are addressed in this dissertation are:
(1) The mold design step in the process of using RT may take a long time for parts
with a wide variety of geometries;
(2) The iterations of design changes may take a long time for parts with a wide
variety of design requirements.
In Chapter 3 and 4, the Multi-piece Mold Design method (MPMDM) and the related
Rapid Tooling Mold Design system (RTMDS) were presented, which addressed the first
problem. In this chapter and the next chapter, approaches that address the second
problem will be presented. The properties of Rapid Tooling are discussed first in this
section to provide a context for Chapter 5 and 6. They also foster a better understanding
of Geometric Tailoring which is to be presented in the remainder of this chapter.
In the RTTB project (Section 1.2.2), the direct AIM tooling is the focus of the
research on design for RT (Section 1.1.3). Therefore the properties of direct AIM
Tooling and its differences from the steel tooling are discussed in this section. In other
literature like (Karapatis, et al., 1998) and (Barlow, et al., 1996), the properties of other
RT processes are also discussed.
Compared with the conventional steel tooling, the direct AIM tooling has three
unique properties: (1) the mold material properties, (2) the fabrication method to get the
tools, and (3) the part material properties. They are discussed in detail as follows.
5.1.1 Mold Material Properties

In the Direct AIM tooling, tools are made of epoxy-based Stereolithography
photopolymers. Currently a series of materials are available for SLA machines (e.g. SL
7540, SL 7510, SL 5530, SL 5520, SL 5510, SL 5190, refer to www.3dsystems.com).
Each material has slightly different thermal and mechanical properties. However, if
compared with the properties of steel, they are pretty similar. Some of the chief physical
properties of the stereolithographic materials that may have an influence in the injection
molding process are:
! Compressive strength
! Tensile Strength
! Flexural Strength
! Shear Strength
! Impact Strength
! Wear Resistance
! Surface Hardness
! Coefficient of thermal expansion
! Thermal Conductivity
! Specific Heat
! Thermal Diffusivity
! Heat Deflection Temperature & Glass Transition Temperature * (Unique to
plastic mold. It is a measure of temperature up to which the material can be used
without significant degradation in strength).
163
! Thermal Fatigue
! Creep behavior under load for extended periods of time
Not all the properties are discussed in this section. Instead the author will focus on
the properties of thermal conductivity, tensile strength, and shear strength, and discuss
their effects on the injection molding process as follows.
Stereolithography material has a thermal conductivity of 0.185 W/m-K, while the
value of steel mold is typically 50 W/m-K (Dawson, 1998). Because this value is 300
times lower than the thermal conductivity of steel, longer cycle times must be employed
to allow the part to cool before ejection.
The maximum shear and tensile strengths of epoxy resins are much lower than those
of steel (e.g. the tensile strength of a typical alloy steel is 650 MPa). Furthermore, the
shear and tensile strengths of epoxy resins are tightly related to temperature of the
material. As the temperature increases, both the tensile and shear strengths of the
material decrease. The relation of strength versus temperature of epoxy SL5170 is given
in Figure 5.1 (Rahmati and Dickens, 1997). As hot plastic which is above 200 oC is
injected into the mold, the mold loses much of its strength. However, due to the low
thermal conductivity of the mold material and the short period of injection, this tooling
method is able to produce parts. Cooling time requirements must be adjusted to deal with
the reduction in material strength at higher temperatures.
Figure 5.1– Maximum Tensile and Shear Strength of SL5170 Versus Temperature.
Because of the low shear and tensile strengths of the mold materials, lower injection
pressures should be used to keep the mold from breaking. Also while the stress-strain
data are normally used for metal tool design, the creep-rupture data are more suitable for
composites and plastics (Jayanthi and Hokuf, 1997). Hence the design strength of the
mold is dependent upon both the magnitude of the applied load and the duration of its
application. The strength required must be adequate to resist the compressive, bending
164
and shearing stresses set up by the molding material under pressure as it moves into the
mold cavity and hardens.
In summary, the mechanical properties of the mold materials are quite different from
those of conventional steel molds. The low thermal conductivity and limited material
strength of the mold need to be considered in the injection molding process of the direct
AIM tooling. Recently Dawson discussed the rapid mold selection and injection molding
conditions in (Dawson, 2001). It provides more details of the relations between the
injection molding conditions and the mold material properties.
5.1.2 Mold Fabrication Properties

The layer based SL fabrication induces sidewall surface imperfections due to the
stair-stepping effects as shown in Figure 5.2.a. This is due to the parabolic profile of the
cured region in the photopolymer, as shown in Figure 5.2.b. The effect can act as
localized undercuts in the cavities which can make ejection of the mold part difficult.
(a) (b)
Figure 5.2 – Stair Stepping Effect of Layer Manufacturing and Reason.
Also the algorithm that generates the build file from the solid model CAD data turns
all surfaces into a series of triangles. This approximation of the CAD model leads to
inaccuracies in built geometry, as well as to the above “stair step” roughness on curved
surfaces. Inaccuracies of building geometry also are caused by part shrinkage as the
stereolithographic polymer cures.
The direct AIM tooling has different mold material and fabrication properties from
steel tooling. Therefore the mold life of direct AIM tooling is much lower than that of the
steel tooling. The actual failure of a feature may occur during injection or ejection in the
direct AIM tooling (Palmer, 1999). In the injection process, failures are caused by a
feature’s inability to resist the force of the polymer flow. In the ejection process, failures
are caused when a feature cannot overcome the shrinkage forces of the part upon
ejection. Therefore the three main types of failure are flow, pullout, and chipping
(Palmer, 1999), which are introduced briefly as follows.
(a) Flow failure: When the polymer is injected into the cavity, the pressure, which is
a factor of injection pressure, gate dimensions, and shot size, may cause the feature
to bend or break.
165
(b) Pullout failure: When the parts are ejected, the shrinkage force of the polymer
onto the mold causes the frictional force between the mold and the part. The
maximum tensile strength of the mold is reached before the part can be ejected,
and the mold features break off.
(c) Chipping failure: The most probable cause of chipping occurs due to a crack in
the material. Chipping failures usually do not occur on the first shot. It is assumed
that the failure is due to a defect or crack within the bulk material of the feature.
The low mold life of the direct AIM tooling will be considered in Chapter 6, which
will address more general issues of the design for rapid tooling.
5.1.3 Part Properties of the AIM Tooling

Thermoplastic materials that can be used in the direct AIM tooling include
polypropylene, polyethelene, delrin, polystyrene, ABS, polycarbonate, glass filled nylon,
glass filled PBT, etc. All of these materials are used in production. Although the parts
produced from the direct AIM tooling can be in the same material as that of the
production parts, their mechanical behavior and performance may be different (Dawson,
1998). The mechanical properties of an injection molded part are affected by the
microstructure of the part. In polymers the microstructures that affect the properties are
crystallinity, molecular orientation, and defects such as voids. Since the use of SL inserts
may lead to longer cycle time, different properties of the injection molded parts is
unavoidable. Dawson (1998, 2001) gave a comparison of the mechanical properties of
the molded materials that were simultaneously injected in a SL mold and a steel mold.
Some results taken from his work are given in Table 5.1 and
Table 5.2. The results indicate a variation in the material properties of atactic
polystyrene parts produced from steel and SLA molds. Another result in (Dawson,
1998) is that the residual stress of the parts obtained from the Direct AIM tooling is
smaller than that of the parts obtained from the steeling tooling.
Table 5.1– Tensile Properties Comparison for Atactic Polystyrene (Dawson, 1998).
Tooling Ultimate Stress Young’s Ultimate Density
(Mpa) Modulus (Gpa) Elongation (%) (g/cc)
Steel 37.4 3.2 1.3 1.045
Direct AIM 32.8 3.4 1.1 1.042
Table 5.2 – Flexural Properties Comparison for Atactic Polystyrene (Dawson, 1998).
Tooling Ultimate Stress Young’s Modulus Ultimate
(Mpa) (Gpa) Elongation (%)
Steel 67.5 3.0 2.6
Direct AIM 71.0 4.0 1.9
From the part properties of the AIM tooling, it is noticed that the prototypes
produced by the RT process have properties different from those of the production parts.
This may lead to difficulties in the prediction of product behaviors through the functional
testing of the prototypes, which is discussed in the next section in more details. To
address this problem, the principles of functional testing are presented first. Based on the
166
principles, a design-for-prototyping approach, geometric tailoring, is introduced in the

next section.
5.2 PRINCIPLES OF FUNCTIONAL TESTING AND GEOMETRIC TAILORING
Although RT has the potential to be used in low-volume manufacturing (Section
1.1.4), the parts produced by rapid tooling are still mainly used for functional testing. To
develop reliable functional tests, systematic methods based on sound principles are
necessary.
5.2.1 Principle of Functional Testing – Buckingham Π Theorem

Currently the Buckingham Π theorem is the basis in prototype testing to get the
correlation between physical models and products (Baker, et al., 1991). The basic idea of
the theorem is using dimensionless variables in the analysis. A brief description of the
Buckingham Π theorem is given as follows, which is adapted from (Cho, et al., 1999).
Consider two systems (a test model and a product) that can be described as
Model: Xm = f (dm,1, dm,2,…, dm,n),
Product: Xp = f (dp,1, dp,2, …, dp,n),
where X is the state of interest, and di is a system parameter. In the equations,
subscripts m and p denote the model and the product respectively. From the Buckingham
Π theorem, the above system equations can be equivalently represented as
Model: πm,x = F (πm,1, πm,2, …, πm,N),
Product: πp,x = F (πp,1, πp,2, …, πp,N),
where πi is a dimensionless parameter and N = n-k. The πi is a power function of
system parameter set D = {d1, d2, …, dn}, and the πx is a power function of system
parameter set D and the state X.
From the non-dimensional equations, πm,x and πp,x are identical if πm,i = πp,i for any i
= 1, 2, …, N. As a consequence, one can correlate the model and product states from
πp,x (Dp) = πm,x (Dm), (5.1)
if πp,i (Dp) = πm,i (Dm), ∀i = 1, 2, …, n, (5.2)
where D denotes the full-set of system parameters.
Equations 5.1 and 5.2 are the fundamental basis of the functional testing. The former
is called the prediction equation, and the latter are conditions for scaled models or
similarity constraints. One should design prototypes not to violate any of the similarity
constraints, in order to predict the product states through the prediction equation.
A simple example is given as follows to aid the reader to understand the theorem,
and the limitations of the similitude method which will be presented in the next section.
This example is adapted from web site www.mech.uwa.edu.au/courses/e101/.
Suppose a car travels straight with a velocity U ms-1. The car has a mass M kg. If
the driver suddenly applies the brakes so that a constant force F kgms-2 is applied, one
167
wants to know how long will the car take to stop. To design a prototype to test the
distance, a brief analysis based on Buckingham Π theorem is given.
Suppose L is the distance the car takes to stop. It is the variable to test.
So one can write f (L, U, M, F) = 0. To get dimensionless variables, one can know
LMLF OP
= 1 from the units of the variables. Therefore π =
LF
NMU 2
Q MU 2
, which is a
dimensionless variable.
Suppose π0 is the solution or root of the equation f (π) = 0, one can know
MU 2
L = π0 .
F
Therefore an experiment can be designed with M’, U’ and F’ to test L’, and calculate
π0 accordingly. After substituting M’, U’ and F’ with the given values we are interested,
the value of L can be calculated from the equation. One thing to be noticed is that in the
testing, M’, U’ and F’ can be much smaller than M, U, and F.
Based on the Buckingham Π theorem, a complex system can be constructed and
tested by geometrically scaling down, or changing materials, or simplifying models. By
doing that one can dramatically reduce the cost and time in building prototypes while
getting reasonable predictive values.
5.2.2 Similarity Methods

The similarity method, which is based on the Buckingham Π theorem, is widely used
for both effective empirical modeling and scale testing (Baker, et al., 1991). It can
experimentally predict the behavior of a target system through an indirect scaled testing
provided the two systems are ‘similar’.
To use the similarity method, one should be able to list all dominant system
parameters, which is shown in the example given in the last section. If any dominant
system parameters are not considered, the system states cannot be well represented, and
the correlation between the systems may become erroneous. Also the parameter values
of the model and product should be known beforehand. An additional assumption made
in the method is that the unknown governing equations f of the model and the product
should be identical. However, the assumption may not be satisfied especially for
prototypes that are built from different materials by various rapid prototyping techniques.
This is also the main reason why rapid prototyping is mostly utilized in early design
phases for visualization, while rapid tooling is utilized in later design phases for
functional testing.
Motivated by using RP techniques in functional testing, which presents significant
benefits through the material changes, a group of researchers at University of Texas at
Austin presented an empirical similarity method (ESM) to improve the accuracy of scale
testing (Cho, et al., 1998; Cho, et al., 1998; Cho, et al., 1999; Cho, et al., 1999). ESM
utilizes specimens to derive the similarity transformation, which correlates model and
product states, empirically. As shown in Figure 5.3, the model and the product
specimens are the geometrically simplified versions of the scaled model and the product.
168
Figure 5.3 – Fundamental Terms and Concept of the ESM (Cho, et al., 1999).
In the ESM, the product state is predicted through the specimen pair (the model and the
product specimens) and the scaled model.
The ESM is proposed for the scale testing with RP. In the case studies given in the
papers, some tested prototypes had widely different material and geometry scale from
those of the production parts. However, ESM requires additional effort to fabricate a
specimen pair, although the additional effort is much smaller than that required to
fabricate geometrically complex test products.
Although guided by the similarity method, the reliability of scale testing results has
been challenged frequently (Baker, et al., 1991). This may be because some of the model
parameters or loading conditions are difficult to control, or there are some uncertain
noises that affect the accuracy. Considering the properties of rapid tooling, a design-for-
prototyping approach, geometric tailoring, is proposed. It is also based on the
Buckingham Π theorem. Its fundamentals are presented in the next section.
5.2.3 Fundamentals of Geometric Tailoring

A mechanical designer may describe a new product with many attributes, which may
include the shape of the component, material, tolerance information, cost and time. Some
of the attributes are related to the design functions, such as geometric variables and
material properties. Some of the attributes are related to the fabrication process
conditions such as cost and time. In this chapter, the author only considers the attributes
that are related to the design functions. Other attributes will be considered in the next
chapter.
169
With the representations given in Section 5.2.1, suppose two parts (a test model and
a product) that are described as
Xm = f (dm,1, dm,2,…, dm,n),
Xp = f (dp,1, dp,2, …, dp,n).
Where X is the state of interest, and di is a system parameter. In the equations,
subscripts m and p denote the model and the product respectively. Therefore two
observations on X and di are the basis of this research.
Observation 5.1. In order to get accurate prediction results of XP, it is generally desired
to control the system parameters dm,i and dp,i identically.
This is because in functional testing, the representation of the function f is usually
unknown. Therefore there may exist some unknown system parameters, or uncertain
noises, or uncontrollable loading conditions. All these factors will affect the accuracy
of scale testing results. Therefore it is desired to make dm,i = dp,i.
Observation 5.2. System parameters di (1≤ i≤ n) are not equally important to X.
For an attribute Xj, some parameters are rather important. Other parameters may have
only negligible effects. For example, to test the fatigue of a gear design, the
maximum stress of the gear under the designed constraints and loads is important.
However, the small changes of the gear dimension may have only negligible effects
on the fatigue.
Therefore the system parameters can be divided into two categories based on their
relations with X. Suppose parameters d1, d2, …, dk are important to X, while parameters
dk+1, dk+2, …, dn are negligible to X. Therefore the principle of the approach
developed in this research is to change parameters dm,j (k+1≤ ≤ j ≤ n) to make dm,i as
≤ i≤
close to dp,i (1≤ ≤ k) as possible. And these parameter changes are named Geometric
Tailoring in this dissertation.
A formal definition of geometric tailoring given in Section 1.2.4 is revisited as
follow.
Geometric tailoring, in this dissertation, is to change the geometry of a part to lower
fabrication cost and time, and to produce functional prototypes to mimic the
production functions, because the material and fabrication process to get molds and
parts are different from those in producing products.
The manufacturability of a given design depends on the following three factors: (1)
the ability to produce the design within the specifications; (2) the ability to produce the
design with a low production cost; (3) the ability to produce the design within a short
production time. Related to these three factors, two kinds of geometric tailoring are
considered in this dissertation. Material Geometric Tailoring (MGT), which is the
geometric tailoring to mimic the production functions related to material properties, is
presented in this chapter. Material Process Geometric Tailoring (MPGT), which is the
geometric tailoring to lower the fabrication cost and to mimic the production functions
including accuracy and surface finish, is discussed in Chapter 6. MGT is actually a
special case of MPGT. However, because the cost and some functions such as accuracy,
170
which are related to the fabrication process, are also considered in MPGT, the process
planning should also be included in the solving process of MPGT. Therefore its solving
strategy is much more complicated (Section 6.4).
Since a part design may have a wide variety of design requirements for its prototypes
(Section 1.2.3), both MGT and MPGT have their applications in actual situations. For
example, for a part design, if the functional properties are much more important than the
cost and time, MGT should be used. However if the fabrication cost and time are also
concerns for the prototypes, MPGT should be used instead. Correspondingly process
requirements such as mold life are added in the problem formulation of MPGT (Section
6.2).
Although the geometric tailoring approaches in this dissertation are developed for
rapid tooling, they also have implications in design for rapid prototyping. For example,
SOMOS 8120 and SOMOS 9120 are two materials used in Stereolithography Apparatus
(SLA). They have properties close to polyethylene and polypropylene respectively. The
geometric tailoring of prototypes may be necessary in order to get better results by testing
the prototypes that are built with SOMOS 8120 and SOMOS 9120 for polyethylene and
polypropylene parts (Sambu, 2001).
Parts produced from the direct AIM tooling have slightly different mechanical
properties from those of the production parts produced with steel tools (Section 5.1.3).
Although these differences are much smaller than those between the parts built by rapid
prototyping and the production parts, the accuracy of the functional tests will be affected.
Lots of design functions such as the maximum stress, fatigue, deflection of a rib, are
related to the material properties. A systematic approach that is developed for handling
the differences of material properties in using Rapid Tooling is presented in the next
section.
5.3 DESIGN DECISION TEMPLATE FOR MGT
Based on the Buckingham Π theorem, it is important for a prototype to behave
similarly to a product in order to verify the design with more confidence. However since
the relationship between the attributes and parameters is usually complex, it may not be
straightforward to choose some parameters and tailor them to get the same values of the
attributes. Also a designer may have several functions of interest in a prototype.
Therefore several attributes of the prototype need to be tailored for the same values as
those of the product. Usually the changing of a system parameter for an attribute may
affect some other attributes. More importantly, not all system parameters are in the
control of the designer. For example, the material properties of the prototype are actually
determined by the manufacturing process, and therefore are in the control of the
manufacturer. Considered all these factors, a scenario of design-manufacture
collaboration based on the MGT decision template is proposed for the material geometric
tailoring. The steps in the method are illustrated in Figure 5.4.
As stated in Section 1.2.1, the designer and manufacturer are usually distributed
geographically and organizationally in the current usage of Rapid Tooling. The RTTB
project is trying to develop a “clean digital interface” to divide the designer and
manufacturer of RP and RT (Allen and Rosen, 1997; Rosen, 2000). The DFM
approaches presented in this dissertation are actually only one scenario of separating the
171
designer and manufacturer in a distributed computing environment. However the focus

of this work is how to formulate and solve the DFM problem provided the scenario had
been chosen (e.g. material and process had been selected by other modules).
There are eleven steps to the scenario based on the MGT decision template as
illustrated in Figure 5.4. The input to the problem is the part design to be tested. Based
on it, the designer first needs to determine which and how many functions are to be tested
in a prototype. Sometime several functions can be integrated into one prototype in order
to reduce cost. However this may bring difficulties in producing production-
Designer Manufacturer
CAD Model, Chosen Material

Part design to be tested MGT template and Process
Step 1
Determine properties of interest.
Step 6
Determine attributes affected by the
material differences.
Step 2
Determine loading conditions, attributes,
and system parameters that are relevant Step 7
to the properties of interest. Formulate additional relations in the MGT
template, hence complete deisgn decision
formulation for MGT.
Step 3
Determine the importance of the attributes
and their relations with system
parameters. Step 8
Solve the formulated MGT problem.
Step 4
Formulate attributes, system parameters
and their requirements in MGT design Step 9
template. Produce prototypes for tailored part
design.
Step 5
Send MGT design template, CAD model.
and/or FEA model to the manufacturer. Step 10
Send prototypes to the designer.
Step 11
Execute the functional testing of the
prototypes.
Figure 5.4 – Steps of a Scenario based on the MGT Decision Template.
172
representative prototypes for all the attributes.

According to the properties of interest (e.g. fatigue), the designer can determine
design attributes and system parameters that are related to the properties (e.g. maximum
stress, refer to Section 5.5 for more details). Also the loading conditions in the tests are to
be determined. Generally the loads and boundary conditions should be the same as those
of the production design. They may have already been formulated in a Finite Element
Analysis (FEA) model in the part design process. Otherwise in Step 7 the manufacturer
may build response surface models according to analysis based on the loads and
boundary conditions (Section 5.4).
Step 3 is to determine the importance of the attributes and related system parameters.
Some attributes and parameters are important for the prototypes. For example, the
attributes that are related to the properties of interest should be production-representative.
Some geometry parameters that are related to assembly are also important. They should
also remain unchanged in the prototypes. Other geometry and material parameters are
much less important for the purpose of the functional testing. Therefore they provide
more design freedoms for the geometric tailoring.
Currently Steps 1 to 3 mainly depend on the knowledge and experience of the
designer. Tools to help the designer in these steps are important. However they are
beyond the scope of this dissertation.
Based on the MGT decision template, which is to be presented in this section, the
designer can formulate the requirements of the attributes and parameters in a compromise
DSP format (Step 4). Then the formulated decision template and CAD model can be sent
to the manufacturer (Step 5). If a FEA model is constructed in the part design process for
the production material, it can also be sent to the manufacturer and used in the analysis of
the prototype with its material properties. The rapid development of the Internet provides
a distributed computing environment for sending these models. The formats and
approaches developed in the RTTB for this step are presented in the theses of (Gerhard,
2001) and other on-going students.
The approaches of selecting the material and process to produce the prototypes in the
RTTB are presented in (Herrmann and Allen, 1999). For a chosen fabrication material
and process, the manufacturer first needs to determine which attributes and parameters
are needed in the geometric tailoring (Step 6). Based on the decision template and CAD
model, additional relations or missing values are added to the template to formulate a
complete decision problem (Step 7). Then the formulated compromise DSP can be
solved (Step 8). Accordingly the manufacturer can change the part design and produce
prototypes for it (Step 9). After the prototypes are produced, they are sent back to the
designer for the functional tests (Step 11).
Of these steps, the author will focus on Steps 4, 7, and 8, which are enclosed with
thicker lines in Figure 5.4. The MGT decision template, which is discussed in the
reminder of this section, is related to Step 4 and 7. Its methodology and formulation are
presented in Section 5.3.1 and 5.3.2 respectively. The approaches to generate the
relations in the MGT template and to solve the MGT problem are presented in Section
5.4. They are related to Step 7 and 8 respectively.
173
5.3.1 MGT Decision Template and its Methodology

For the design-manufacture collaboration problem (transfer design information from
the designer to the manufacturer), a “Solid Interchange Format” (SIF) has been proposed
as a “clean digital interface” between design and manufacturing in the area of RP (Sequin
and McMains, 1995). SIF contains part geometry and possibly surface finish, tolerance,
material, and perhaps other related information types. Although this information is a
significant extension beyond STL, IGES, or even STEP file formats, it is still insufficient
to support design for functional prototyping (Section 5.2.3).
In this research, a design decision template is proposed as an effective digital
interface. It incorporates a series of compromise decisions. Using the template, enough
design information is transferred to enable the manufacturer to explore alternative part
designs in the event that the functional testing requirements cannot be achieved.
Effectively, this transfers the burden of design for manufacturing to the manufacturer
with more design freedom. Therefore the time of iteration between the designer and
manufacturer may be reduced.
The methodology of using the decision template as the digital interface between the
designer and manufacturer is further illustrated in Figure 5.5. Suppose the design
functions that the designer plans to test in the prototype are Fi. The design variables of
the part design that are related to Fi are Di. As stated before, only the design functions
that are related to the material properties are considered in the MGT. So suppose
material properties that are related to Fi are Mi. The relations between Fi, Di, ard Mi can
be formulated as Fi = f(Di, Mi). This relation f can be rather simple and may be given as a
simple equation (Section 5.4.1). It can also be complicated and unknown to the designer.
In this case, the response surface model is employed in this dissertation to formulate their
relation (Section 5.4.2). One thing to be noticed is that in Rapid Tooling, the material
Function Design
Desired Space Solution Solution
Point Space Point
Desired
Range Design Design
Variables Desired
Desired Functions Range
Range Fi Di
Fi = f (Di, Mi)
Designer
Manufacturer
Material Process
Properties Variables
Mi Pi
Material
Properties
Space
Figure 5.5 – The Decision Template for the MGT and Design Freedom.
174
properties of the prototypes are determined by the fabrication process variables Pi.
Therefore Mi is actually controlled by the manufacturer, while Di is controlled by the
designer.
In the current usage of Rapid Tooling, a STL file or a SIF file is transferred from the
designer to the manufacturer. In the file, the designer has already set the values of the
design variables and functions (by the requirements on material properties). For a
selected fabrication process and material, the only design freedom for the manufacturer is
the process parameters, which are adjusted to achieve the desired material properties.
However, the relation of Mi and Pi may be complicated especially for Rapid Tooling
which composes several steps. It is rather difficult, sometimes even impossible to
achieve the desired material properties (Section 5.1.3). Therefore the design is sent back
to the designer and a new iteration begins. In most cases, tradeoffs between different
functions have to be made if the desired values cannot be achieved for all Fi. To reach
such a “satisfying” solution for the designer and manufacturer, several iterations and lots
of negotiations may be necessary. This process may take a long time for some
complicated part designs.
By using the decision template, the author believes that the designer and
manufacturer can make tradeoffs more quickly to reach the “satisfying” solution. This is
because the designer has formulated the design requirements on the prototype in the
decision template. Based on the template, the manufacturer will have more design
freedom in producing prototypes. Besides the selection of the process parameters, the
additional design freedom may come from the following two sides.
(1) Design variables of the part design are given in desired ranges instead of specified
points.
As discussed in Section 5.2.3, the principle of geometric tailoring is to change
some unimportant parameters to make the more important variables more
production-representative. Therefore based on the decision template, the
manufacturer has the freedom to change the unimportant parameters in the
specified ranges.
(2) Design functions of the part design are given in desired ranges instead of
specified points.
The decision template is based on the feasible and satisficing regions of the
design space. The term satisficing was coined by (Simon, 1996) to describe a
particular form of less-than optimal solutions (Section 2.4). Instead of sending an
optimal solution based on the design information, the designer can send the
decision template to the manufacturer for a satisficing solution. Therefore the
manufacturer will have more design freedom in producing prototypes.
The additional design freedom enables the manufacturer to slightly change the
material properties of the prototypes. That is, if the material properties Mi are changed to
Mi’, the manufacturer can adjust Pi to Pi’ to achieve the same function Fi. In many cases,
the manufacturer may feel much more confident in producing prototypes with properties
Mi’ instead of Mi (Section 5.1.3). Therefore it is more likely that the required prototypes
can be produced without being sent back to the designer for a redesign. Three examples
175
and two case studies are given in Section 5.5, and Chapter 7 and 8 respectively. They
illustrate the applications of the decision template and its advantages.
Related to the MGT and MPGT, two kinds of design decision templates are
introduced in this dissertation. They have the same methodology. The decision templates
are:
• MGTDT – Material Geometric Tailoring Decision Template, a compromise
decision in which the component’s dimensions are modified to provide similar
functional performance when a prototype material replaces the production
material.
• MPGTDT – Material-Process Geometric Tailoring Decision Template, a
compromise decision in which the component’s dimensions are modified to suit a
prototype material and fabrication process.
They embody the relevant design information and designer preferences. The
formulation of the MGTDT is presented in the next section. MPGT and MPGTDT will
be discussed in Chapter 6.
5.3.2 Formulation of the MGTDT

The compromise Decision Support Problem (DSP) is a general framework for
solving multi-objective, non-linear optimization problems (Section 2.4). In this research
it was employed to formulate the MGTDT. The compromise DSP is central to modeling
multiple design objectives and assessing the tradeoffs pertinent to the design for Rapid
Tooling. The compromise DSP word formulation of the MGTDT is presented in Table
5.3.
Table 5.3 – Word Formulation of the MGTDT.

GIVEN:
• Parametric CAD model of part • Material properties
• Functional property models • Goal preferences as weights
• Target values for functional properties
FIND:
System Variables: Deviation Variables:
Geometry variables Deviation of goals from targets
SATISFY:
Goals: Constraints:
Meet target functional properties Meet geometry and/or assembly constraints
Meet targets of geometry variables Bounds:
Bounds for all system variables
MINIMIZE:
Deviation Function: Weighted sum of goal deviations
176
Mathematically, the compromise DSP is a multi-objective decision model based on

Mathematical Programming and Goal Programming. A general mathematical
formulation of MGTDT is presented in Table 5.4, which is an extension of the word
formulation given in Table 5.3.
Table 5.4 – Mathematical Formulation of the MGTDT.

GIVEN:
• nf - number of functions;
ng - number of geometry variables; nm - number of material variables
p – equality constraints; q – inequality constraints.
• Geometry variables Gp, i i = 1, …, ng
• Material properties Mp, i i = 1, …, nm

• Design functions Fp, i = fi (Gp,j, Mp,n) i = 1, …, nf; j = 1, …, ng; n = 1, …, nm
• Material properties Mm, i** i = 1, …, nm
• Design functions Fm, i = fi (Gm,j, Mm,n) ** i = 1, …, nf; j = 1, …, ng; n = 1, …, nm
• Weight Wi i = 1, …, ng+nf
FIND:
System Variables:
Gm, i i = 1, …, ng
Deviation Variables:
di+, di- i = 1, …, ng+nf
SATISFY:
Goals:
Fp, i / Fm,i - di+ + di- = 1 i = 1, …, nf [5.3]
Gp,i / Gm,i - dnf+i+ + dnf+i- = 1 i = 1, …, ng [5.4]

Constraints:
di+ • di- = 0, di+ ≥ 0, di- ≥ 0 i = 1, …, ng+nf
gj(Gi) = 0, gk(Gi) ≤ 0 i = 1, …, ng; j=1, …, p; k=1, …, q

Bounds:
Gimin ≤ Gi ≤ Gimax i = 1, …, ng
Fimin ≤ Fi ≤ Fimax i = 1, …, nf
MINIMIZE:
Deviation Function: Wi • (di+ + di-) where ΣWi = 1, Wi ≥ 0 (i= 1, …, ng+nf)
Note:
1. Subscripts m and p denote the prototype model and the product respectively.
2. Symbol ‘**’ denotes the entries that are to be completed by the manufacturer.
177
The math formulation contains systems variables, goals and constraints. The system
variables in the MGTDT are the geometry variables (Gj) and material properties (Mn).
The geometry variables typically include modifying dimensions (e.g. lengths and angles)
of features in a part. In some cases geometry variables could be a whole feature. In such
cases the manufacturer has freedom to redesign the feature, e.g. use a rib instead of boss
to fulfill the same function. Material properties Mi include Young’s modulus, tensile
strength, and elongation yield, etc.
The goals in the MGTDT include functional and geometry goals. The functional
goals (Equation 5.3) include matching the functional properties of prototype parts to their
target values. Target values correspond to functional properties of the production part.
These functional properties Fi can be stress, deflection, weight, or some others. The
geometry goals (Equation 5.4) include matching the geometry variables of the tailored
part to their target values. Target geometry values are the most desirable values of part
dimensions provided by the designer.
The constraints in the MGTDT include geometry constraints that arise due to space
and weight limitations. While performing geometric tailoring of an assembly, the
assembly requirements are also considered in the constraints.
One thing to be noticed is that in the formulation, the designer does not have all the
information (the material properties of the prototypes are out of his/her control). The
entries indicated by ‘**’ in Table 5.4 denote information that the manufacturer must
supply in order to complete the problem formulation and generate a solution. However,
based on his/her information, the designer can instantiate the MGTDT to create the
problem formulation suitable for communicating to the manufacturer.
In light of the methodology and formulation of the MGTDT presented in this section,
the usage of MGTDT is discussed in the next section, especially the approaches to
formulate the relation f in the template.
5.4 USAGE OF MGTDT
In the MGT decision template, the designer provides target values for geometry
variables and design functions, and also their preferences. The Manufacturer provides
material properties of prototype parts and related quantitative relations between goals and
system variables. Therefore the information required to solve the material geometric
tailoring problem comes from the design and manufacturing organizations. Essentially
the MGTDT is actually a method to organize the design and manufacturing information
into one formulation. After the MGT problem is formulated, it can be solved with the aid
of engineering optimization software (e.g. OptdesX). The approaches to formulate the
required quantitative models for functional properties are presented in Section 5.4.1. The
process of solving the MGT problem is discussed in Section 5.4.2.
5.4.1 Formulating Functional Properties in the MGTDT

The functional properties Fi in Table 5.4 are related to the geometry variables (Gj)
and material properties (Mn). Suppose the relationship can be stated as Fi = fi(Gj, Mn).
Here fi is the function of Fi for a set of design variables Gj and Mn.
178
For simple geometry, it is possible to obtain an analytical equation for the design
functions such as stress or weight. But for complicated geometry it is usually not
possible. In such circumstances, design functions can be represented as a response
surface generated by fitting a surface through experimental data points. Both approaches
are described in more details as follows. They correspond to the two types of
“quantitative” models in common use: Analytical models and Simulation models.
Figure 5.6 - A Cantilever Beam.

(1) Functions Represented as Analytic Equations
In numerous handbooks of mechanical design, different design functions are usually
represented as analytic equations of geometric variables and material properties. For
example, a cantilever beam is shown in Figure 5.6. If a force ‘F’ is applied at one end of
the beam, the equations of the maximum stress (σ), maximum strain (ε) and maximum
displacement (δ) in the beam can be calculated by Equations 5.5 ~ 5.7 respectively
(Shigley and Mischke, 1989).
FLc
σ= (5.5)
I
FLc
ε= (5.6)
YM * I
FL3
δ= (5.7)
3 I * YM
Therefore for a cantilever beam, if the design functions of interest in the MGTDT are
the maximum stress, or maximum strain, or maximum displacement, the functions fi may
be represented as analytic equations directly. In the equations, L, c and I are geometry
variables (refer to Figure 5.6, I is the moment of inertial of the cross-section area about z
axis). YM is material property, which represent Young’s modulus.
Depending on loading conditions, the same design functions may be represented in
different equations. For example, if a displacement δ, instead of a force ‘F’, is applied in
at one end of the cantilever beam, the maximum stress, maximum strain and force
required to cause this displacement are calculated by Equations 5.8 ~ 5.10 respectively.
3δc * YM
σ= [5.8]
L2
179
3δc
ε= [5.9]
L2
3 Iδ * YM
F= [5.10]
L3
Therefore in the MGT problem the loads and boundary conditions for testing the design
functions are also very important. The designer should formulate them in functions fi in
the MGTDT, or in a FEA model that is to be transferred to the manufacturer. Based on
the loading conditions, the manufacturer can tailor the part and produce production-
representative prototypes more quickly.
For more complicated geometry, the analytical equations are usually not accurate
enough for the MGT problem. Therefore, the response surface models are employed to
represent function fi in this dissertation.
(2) Functions Represented as Response Surface Equations
Response surface methodology, or RSM, is a collection of mathematical and
statistical techniques that are useful for the modeling and analysis of problems in which a
response of interest is influenced by several variables and the objective is to optimize this
response (Section 2.5). The result response surface equations (RSEs) allow for a better
understanding of the relationships between the inputs and the response. In the examples
given in Section 5.5.2 and 5.5.3, second order response surfaces are used (k = 2). So the
RSE can be written in the form of a polynomial function as:
k k k k
y = b0 + ∑ bi x i + ∑ bii x i + ∑ ∑b x x
2
i i j [5.11]
i =1 i =1 i=1 j = 1, i≠ j
In much of this work, response surfaces are used to replace computationally

complex, but high fidelity analyses for usage during design synthesis. Response surfaces
approximate the actual design space and are based on the high fidelity analyses, but
enable much faster syntheses. After synthesis, a check is performed to ensure that the
performance indicated by the synthesis result is not too far off. This type of synthesis is
particularly useful during concept exploration (Chen, 1995). The method is called the
Robust Concept Exploration Method (RCEM) (Section 2.5).
In order to create the response surfaces for the system goals, a number of
experiments must be run to gather data for an empirical model. When the system goals
are dependent on two or more factors (system variables), Design of Experiments (DOE)
techniques can be practiced to determine the experiment sequence for the empirical
model (Montgomery, 1991). Factorial experiment designs involve testing a number of
variables, or factors, at different values, or levels. In the examples given in this chapter,
the experiments used to construct response surfaces (model building experiments) were
fractional factorial experiment designs with a face centered central composite design.
Three levels of each factor were considered. Other techniques for constructing response
surface equations can be found in (Box and Draper, 1987; Khuri and Cornell, 1987;
Myers and Montgomery, 1995).
Based on the experiment design, the experiments can be performed either on analysis
packages or through physical experimentation. In the examples given in Section 5.5.2
180
and 5.5.3, ANSYS (www.ansys.com), a finite element analysis (FEA) software system, is
used to get the responses of maximum stress and maximum deflection. After the
experiments, MINITAB (www.minitab.com), a statistical software system, is used to fit
response surface equations through the experimental data. These response surface
equations are incorporated into the MGTDT for the MGT problem.
5.4.2 Formulating and Solving the MGT Problem

The mathematical formulation of the MGTDT presented in Table 5.4 is a general
formulation that can be used for a specific case. For a part design, the designer first
identifies all the functional requirements of the prototype. Once the functional
requirements are identified, the geometry variables that have some design freedom and
that affect the functional requirements should be identified. Appropriate bounds should
be chosen for the geometry variables. The other requirements such as space or weight
constraints should also be identified. Once all the problem specific information is
identified, the quantitative relationship between goals and system variables should be
generated. For simple factors like weight, analytical equations can be developed. For
complicated factors like stress, experimental approach should be used. Design of
experiments can be used to identify the list of experiments required to generate a
response surface model. The experiments could be performed on analysis packages such
as ANSYS. After all the design information is filled in the MGTDT, the designer can
send it to the manufacturer with other information such as a CAD model and FEA model.
Based on the template, CAD model, and FEA model, the manufacturer can identify the
problem specific information and generate quantitative models in the same way. Three
examples are given in Section 5.5 to demonstrate the above process to formulate the
MGT problem based on the MGTDT.
After completing the formulation of the MGT problem, the manufacturer can solve it
with the aid of computer. Since the MGT problem is formulated in compromise DSP
format, it can be solved using the ALP algorithm (Mistree, et al., 1993), which is part of
the DSIDES (Decision Support in Designing Engineering Systems) software. The
examples given in Section 5.5.2 and 5.5.3 are comparatively simple, which are used to
verify the idea of material geometric tailoring. Since the efficiency of the solving process
is not the main concern in these examples, the author employed the exhaustive search
method to solve the compromise DSP. In the two case studies to be presented in Chapter
7 and 8, OptdesX (www.et.byu.edu/~optdes), an engineering optimization software system,
was employed in the solving process.
With the MGT and MGTDT presented, three examples are given in the next section
which verify the idea of the material geometric tailoring for Rapid Tooling.
5.5 INITIAL CASE STUDIES
In this section, the author went through three case studies to verify the correctness of
the material geometric tailoring (MGT). They also illustrate the application of the
material geometric tailoring decision template (MGTDT). In Section 5.1, a tensile bar
example is presented which is used to familiarize the reader with the basic idea of the
MGT and the formulation of the MGTDT. It is the simplest example, in the author’s
opinion, that can illustrate the MGT problem. In Section 5.2 a rib part problem is
181
considered for both design requirements (maximum deflection) and manufacturing

requirements (mold life). Tradeoffs are made by solving a compromise DSP based on the
MGTDT. In Section 5.3, a ring gear part, which is taken from a cordless drill, is studied
in order to get functional prototypes using the AIM tooling.
The examples were planned to test different aspects of the MGT problem (Table
5.5). In the tensile bar problem a simple equation is derived, while simulation software
(ANSYS) is used to study the design requirements with relation to geometry variables in
the rib and gear problems. The number of design functions considered in the problems of
tensile bar and rib is only one. However two design functions are considered in the ring
gear example. In the rib problem, the author also added the mold life consideration in
formulating and solving the MGT problem. It sets the stage for the research to be
presented in Chapter 6, which will consider more manufacturing requirements of the
Rapid Tooling process.
Table 5.5 – Experimental Plan For Testing the MGT.

Exp 1: Tensile Bar Exp 2: Rib Part Exp 3: Ring Gear
Function Representation
Analytic Equations X
Response Surface Equations X X
Number of Design Functions
One X X
Many X
Design Functions
Force X
Deflection X
Stress X
Other Manufacturing Requirements
Yes X
No X X
5.5.1 Building Prototypes of a Tensile Bar

In this section the author uses a simple example to illustrate the principle of the MGT
and MGTDT.
• Problem Introduction
Suppose the designer finished a product design, within which a tensile bar is used to
hold some other components to the housing of the product (Figure 5.7). In the design
process, the designer was not sure about the weight of these components (F). So he/she
had to estimate a value for F and continued the design of the tensile bar. The design
equation for the tensile bar is pretty simple (Shigley and Mischke, 1989). From
F
σ= , the designer can know F = σ • h • t , where σ is the stress of the part under the
h•t
load. It should be smaller than the ultimate yield strength of the material, Su.
The value of Su is related to the material and fabrication process for the part design.
Although different scenarios for the selection of material and fabrication process are
feasible and explored in the RTTB project, the author assumes that the designer will
select them in this example. Suppose atactic polystyrene and H13 steel tools are selected
182
Fixed
L
h
t
Figure 5.7 – The Illustration of a Tensile Bar Example.

for the production parts. By referring to design handbooks, the designer can get the value
of Su. Suppose Su = 37.4 MPa (Dawson, 1998).
Accordingly the designer can finish the tensile bar design by using a safe factor
σ max
γ = . Suppose the values of the variables are h = h0 and t = t0. After finishing the
Su
initial design, the designer hopes to do functional testing to verify that the design works.
In industries prototypes are also used as the milestones for different design phases.
Instead of waiting for production injection mold tooling for the tensile bar, the
designer decides to fabricate a prototype tensile bar using the AIM tooling with the same
material atactic polystyrene. Since the designer is unsure about the maximum load F
which may vary widely in the running of the product, the designer would like the
prototype tensile bar to have the same safe factor γ as that of the production tensile bar.
Therefore problems (e.g. the value of the maximum load is assumed too small in the
design) can be identified in the functional testing of the product.
Base on the equations of σ and γ, one can get
Fmax
γ = . [5.12]
Su • h • t
Therefore the functional property of interest in this example is the maximum force
Fmax. Since Su of the prototypes is unknown for the designer, he/she can formulate the
above design information in a MGTDT and let the manufacturer to take care how to
produce qualified prototype tensile bars.
• Designer Activities
To begin, the designer must define the problem from his/her perspective, ensuring
that the design intent is communicated and design freedom is properly specified. For the
tensile bar, the designer specifies three design variables, length (L), width (h), and
thickness (t). For the product design as shown in Figure 5.7, L and t also affect the
assembly of the product. Therefore the designer may do not want the manufacturer to
change their values. However, h can be tailored by the manufacturer in a range specified
by the designer without affecting other components. Therefore the designer can
183
formulate L and t into the MGT formulation but given them tight ranges. The designer
can also formulate h as the only geometry variable in the formulation that can be changed
by the manufacturer. The latter way is used in the formulation shown in Table 5.6.
Suppose the range of h is 0.4 to 0.6 inch, and the range of Fm is 0.99•Fp to 1.01•Fp.
These ranges also specify the allowable design freedom within which the manufacturer
may tailor the design. One thing to be noticed is that if the design freedom given to the
manufacturer is too small, the manufacturer may not be able to find a satisfying solution
for the MGT problem. Therefore the design will be sent back to the designer to begin a
new iteration.
In the MGT formulation (Table 5.6), the two goals include one goal on meeting
width value of the production tensile bar (Equation 5.13) and the other goal on achieving
Table 5.6 – MGT Tensile Bar Problem Formulation.
Given:
! Length L = 5.0 inch, Thickness t = 0.2 inch, Width h = 0.5 inch
! Ultimate yield strength of the product Su, p = 37.4 Mpa
! Ultimate yield strength of the prototype Su, m**
! The maximum force of the product, Fp= Su,p • h • t
! Weight W1, W2
Find:
! The geometric variables:
• Width, h
! The maximum force of the prototype, Fm
! Deviation variables
• d1+, d1-, d2+, d2-
Satisfy:
! Goals:
0.5
• Width: + d −1 − d +1 = 1 [5.13]
h
F
• Maximum Force: p + d − 2 − d + 2 = 1 [5.14]
Fm
! The Force Equation:
• Fm= f(Su,m, h, t) **
! 0.99 • Fp ≤ Fm ≤ 1.01 • Fp
! di-, di+ > 0, i = 1, 2
di- • di+ = 0, i = 1, 2
! The bounds on the system variables:
0.4 inch ≤ h ≤ 0.6 inch
Minimize:
! The deviation function (Archimedean formulation):
• Z = W1( d −1 + d +1 ) + W2 ( d − 2 + d + 2 ),( ΣWi = 1)
Note: Symbol ‘**’ denotes the entries that are to be complemented by the manufacturer.
184
the same maximum force in the prototype tensile bar as will exist in the production
tensile bar (Equation 5.14). An analytic equation for the last goal is also given in the
formulation based on Equation 5.12. The designer can assign different weight for the
goals based on the requirements. In this example, the force goal is much important than
the width goal. Therefore a weight scenario can be W1=0.05 and W2=0.95. In Chapters 7
and 8, a method using preference model to generate weights for different goals is
employed in the two case studies (Hernandez and Mistree, 2001).
With this information, the designer instantiates the MGT template to create the
problem formulation that is suitable for communicating with the manufacturer. This
instantiated template is shown in Table 5.6. The entries indicated by ‘**’ denote
information that the manufacturer must supply in order to complete the problem
formulation and generate a solution.
• Manufacturer Activities
With the aid of agents in a distributed computing environment (Gerhard, 2001), the
manufacturer can receive the MGT formulation sent by the designer, and also the
material and process that are selected for the prototypes. In the MGT formulation, two
pieces of information from the manufacturer are required: the prototype material property
Su,m, and the force behavior as a function of system variables. Suppose the AIM tooling
for polystyrene is chose for the prototype. Based on his/her knowledge and experience
on the material and process, the manufacturer knows:
Su,m=32.8 Mpa
Fm= Su,m • h • t.
Therefore the empty entries in Table 5.6 can be completed. And the complete
problem formulation can be solved using an exhaustive search algorithm. The results for
the weight scenario W1=0.05 and W2=0.95 are:
h = 0.57 inch, Max Force = 3.739, Deviation Z= 0.6%.
Comparing to the target values h = 0.5 inch and Fp = 3.74, the errors of h and Fm are
14% and 0.2% respectively. Considering the weigh W2 = 0.95, this result makes sense.
Another comparison given here is to compare the error of Fm with that of the tensile
bars produced without geometric tailoring. For this example, Fm will be 3.28 if no
geometric tailoring is executed. Therefore the error of Fm is 12.3%, which is much
bigger than 0.2%.
Besides the weight scenario given by the designer, the manufacturer may also
explore other scenarios. Three weight scenarios and the related results are shown in Table
5.7. In these scenarios, the dimension goal is weighted more highly. As can be seen, the
Table 5.7– Scenarios of Goal Weights and Related Results.
Scenario Width Goal Max Force Goal h Fm
1 0.1 0.9 0.57 3.739
2 0.5 0.5 0.57 3.739
3 0.9 0.1 0.564 3.703
185
results across all scenarios are almost identical, indicating a solution that is stable with
respect to preferences.
Based on the results, the manufacturer can reconstruct a CAD model of the tensile
bar using the variable values. The designer may be notified about the results by a
message from the manufacturer before the prototypes are produced. In the fabrication
process, the manufacturer should also make efforts to make the prototypes have the
material properties that are formulated in the MGT (Su,m=32.8 Mpa for this example).
• Physical Validation
The research presented in this section is mainly based on the work given in (Dawson,
1998). The author also did some physical validation for the example. The validation
results presented in this section will also used in Section 7.5 for an example of a robot
arm.
First a pair of SLA molds was built in the tooling mode in a SLA-3500 machine
(Figure 5.8). Then a Morgan-press injection-molding machine (G-100T) from Morgan
Industries Inc. was used to fabricate the prototypes of the tensile bar. The material was
general-purpose polystyrene from the Dow Chemical Company (www.dow.com). The
main injection molding parameters used in the process were:
Barrel/Nozzle temperature: 430 / 450 oF
Clamping Force: 11 tons
Injection Pressure: 2.5 x 103 psi
Pilot Valve Pressure: 6x10 psi
Injection Time: 35 second
Cooling Time: 200 second
Cycle Time: 6 minutes
Figure 5.8 – Photo of SLA Tools and Prototype Tensile Bars.
186
More than 10 parts were produced. Some of them are shown in Figure 5.8.
Among the prototypes, four tensile bars are used to determine material properties
(Figure 5.9). The obtained values are presented in Table 5.8. In the table, the values in
‘estimated’ row correspond to the material properties from (Dawson, 1998) and are used
in the MGT problem; ‘mean’ row corresponds to the mean values of the material
properties obtained from physical experimentation. The variation of the material
properties (maximum % deviation) for different specimens is 4 – 9%. This is a small
error and hence the values can be considered to be consistent. Comparing the actual and
estimated values, tensile strength and Young’s modulus have an error of 1.7% and 2.5%
respectively. These values are very small indicating a very good match between the
estimated and actual values. Strain values are not used in the MGT problem and hence
no comparisons are provided.
Table 5.8– Material property validation results for polystyrene.

Tensile Strength Young's Modulus
Replication (MPa) (MPa) % Strain @ Yield
1 30.7 3392 1.01
2 32.2 3730 0.97
3 33.6 3698 1.10
4 32.5 3123 1.15
Mean 32.3 3486 1.06
Max. Deviation Percentage 4.81% 7.03% 8.51%
Estimated 32.8 3400
Deviation Percentage 1.71% 2.46%
Figure 5.9 – Photo of Prototype Tensile Bars Used for Tensile test.
187
• Discussion
The design freedom given by the designer to the manufacturer is important for
reducing the iterations between the designer and manufacturer. In this example, the
author noticed that if the bounds of h given by the designer are changed from 0.4 ≤ h ≤
0.6 to 0.45 ≤ h ≤ 0.55, no solution could be found for the given requirements. Therefore
the design may be sent back to the designer to begin a new iteration.
From the testing results shown in Table 5.8, it can be noticed that the material
properties of the prototypes are actually in a certain range instead of a point. This is
because of some inevitable noise factors in the rapid tooling process (Figure 5.10).
Suppose the material property of interest is Y and its distribution can be represented as
the normal distribution as shown in Figure 5.10. The mean and standard deviation of Y
are µY and σY. They are mainly determined by the capability of the manufacturing
process. For a given target value y, bias(y) = y-µY. If bias(y) is larger than the given
tolerance, the produced parts will have very low quality. Therefore the principle of
geometric tailoring is to change the target value y to y’ in order to make bias(y’) < bias(y).
It is obvious that the best value of y’ is µY. In the current usage of Rapid Tooling, y is set
by the designer who may have no experience of the process and material. Therefore y
may have a big bias. However, by using the MGTDT, the manufacturer can set the value
of y with a lower bias based on his/her understanding of the process and material.
Therefore geometric tailoring can help the manufacturer to produce qualified prototypes
much more easily. The designer can also benefit from it because less iteration is needed.
Standard
Quality within
deviation
Probability σY specification
Distribution
bias
x Control Quality Tolerance

Factors distribution deviation
Y
Part P y' µY ,σY
RT Process
y Mean µY
Performance
y' Target y
Geometric
Noise Tailoring
Factors µ z ,σ z
z
Figure 5.10 – Quality Distribution and Geometric Tailoring.

With the detail usage of the MGTDT presented in this section, two other examples
are presented in the next two sections. Instead of analytic equations, the Response
Surface Equations (RSEs) are used to represent functional properties.
5.5.2 Building Prototypes of a Rib Part

Suppose a rib part as shown in Figure 5.11 is used to support a load F = 25 pounds.
Similar to the tensile bar example presented in Section 5.5.1, atactic polystyrene and H13
188
b
F
w
h
Figure 5.11 – The Illustration of a Rib Example.

steel tools were chosen for producing the production parts. The designer finished the rib
design as b = 0.192 inch, h = 0.75 inch, w= 0.852 inch. No draft angle was considered (α
= 0). According to the design, the deflection of the rib under the load will be 0.02 inch,
which satisfies the assembly requirements of the product.
However since this deflection is critical to the performance of the product, the
designer needs 50 prototypes to validate his/her design within one week. Suppose the
error of the deflection of the prototypes is 5%. Since the lead-time to get the production
tools for the part is too long, the designer decides to use the AIM tooling in producing the
prototypes.
In this example, suppose the designer is only interested in the deflection of the rib
when testing the prototypes. Therefore the functional property of interest in this example
is the maximum deflection of the rib, which is related to the flexural Young’s modulus of
the material. The flexural Young’s modulus of a production part made of atactic
polystyrene with H13 steel molds is YMsteel = 3.0 Gpa = 436710 psi (Dawson, 1998).
In the design process, the designer had already used a Finite Element Analysis
software system (e.g. ANSYS) to simulate the maximum deflection related to the load.
This simulation was used by the designer in his/her design. Suppose in the production
part the maximum deflection of the rib is 0.02 inch. Based on this value and the errors
that are allowed in the functional testing, the designer may set the range of the maximum
deflection for the prototypes as 0.019 ≤ MDm ≤ 0.021.
Among the three design variables (b, h, w) of the rib, assume w is fixed because of
other assembly requirements. Therefore b and h are the design variables that can be
changed by the manufacturer. Their ranges are shown as follows.
0.182 ≤ b ≤ 0.202;
0.71 ≤ h ≤ 0.79.
189
In the rib design, the designer did not consider any draft angles because he/she was
not aware of its necessity. Therefore based on all the information, the designer may
initiate a MGT rib problem formulation as shown in Table 5.9.
Table 5.9 - MGT Rib Problem Formulation By the Designer.

Given:
! b = 0.192 inch, h = 0.75 inch, w= 0.852 inch
! Load F = 25 lb
! Flexural Young’s modulus of the product YMsteel = 436710 psi
! Flexural Young’s modulus of the prototype YMSLA**
! The maximum deflection of the rib MDp = 0.02 inch
! Weight W1, W2, W3
! 50 prototypes are needed
Find:
• Thickness, b
• Height, h
! The maximum deflection, MDm
• d1+, d1-, d2+, d2-, d3+, d3-
Satisfy:
! Goals:
0192
.
• Thickness: + d −1 − d +1 = 1 [5.15]
b
0.75
• Height: + d −2 − d +2 = 1 [5.16]
h
0.02
• Maximum Deflection: + d −3 − d +3 = 1 [5.17]
MDm
! The maximum deflection equation:
• MDm = f(b, h) **
! 0.019 inch ≤ MDm ≤ 0.021 inch
! di-, di+ > 0 i = 1, 2, 3
di- • di+ = 0 i = 1, 2, 3
0.182 inch ≤ b ≤ 0.202 inch
0.71 inch ≤ h ≤ 0.79 inch
Minimize:
• Z = W1( d −1 + d +1 ) + W2 ( d − 2 + d + 2 ) + W3 ( d − 3 + d + 3 ),( ΣWi = 1)
In the MGT formulation (Table 5.9) three goals are considered (Equations 15 ~ 17).
The designer can assign different weights to them based on his/her requirements.
Different weight scenarios will be discussed further in the solving process of the MGT.
190
The MGT formulation and CAD model of the part can be sent to the manufacturer.
In addition, the designer can also send the FEA model that was used in determining the
maximum deflection of the rib to the manufacturer.
After receiving the above design information, the manufacturer can begin the
geometric tailoring for the part. Suppose the AIM tooling for polystyrene is chose for the
prototype. Correspondingly the manufacturer knows that the flexural Young’s modulus
of atactic polystyrene of parts produced with SLA molds are YMSLA = 4.0 Gpa = 582280
psi (Dawson, 1998). The manufacturer can also employ the FEA model to understand the
relation of the maximum deflection with the material differences of the prototype ribs.
After getting an equation of MDm, all the empty entries in Table 5.9 can be completed.
The solving process will be very similar to that of the tensile bar example (Section 5.5.1).
However, suppose the manufacturer notice that there is no draft angle in the rib
design. This may lead to some difficulties in the injection molding process (Rosato and
Rosato, 1995). Especially the mold life of SLA molds will be significantly lower for a
part design without draft angle (Palmer, 1999). Suppose this consideration needs to be
added to the MGT formulation.
To tailor the draft angle based on the mold life, the manufacturer must have an
analytical equation to represent the relation of the mold life and the draft angle. Based on
the work of (Palmer, 1999), suppose a response surface equation is generated as:
Number of Shot = 23.13 – 14.54•hr + 16.06•α + 1.4•hr•hr + 0.67•α•α - 2.24•hr•α,
where hr is the height ratio (h/b) and α is the draft angle. Although this model is
developed for a mold protrusion and may be rather crude, the author will use it in this
example to demonstrate how to integrate the design and manufacturing considerations.
The manufacturer can add draft angle α as an additional geometry variable in the
MGT formulation with its target value as 0.5o. Accordingly there are two thickness
variables, b1 and b2, which are related to the top and bottom thickness of the rib (Figure
5.12). They have the relation as:
b1 = b2 - 2•h•tan(α)
where h is the height and α is draft angle.
b1
h
a
b2
Figure 5.12 – Geometry Variables of the Rib by Adding Draft Angle.
191
Furthermore another constraint is related to the value of b1. In the injection molding
process, the top surface of the rib will be used as the ejection position. After the
calculation, mold designer decided that the ejection pins need to be 0.125 inch in
diameter. Therefore, the top thickness b1 should be greater than 0.125 inch.
Correspondingly the relation MDm = f(b, h) in Table 5.9 is changed to MDm = f(b2, h,
α). This relation can be represented by a Response Surface equation. Based on the FEA
model given by the designer, the manufacturer can simulate the maximum deflection
related to the load and boundary conditions for different geometries. This process is
described in more details as follows.
The three design factors chosen for the experiments and their ranges are given in
Table 5.10. The experimental design that is used in this example is CCF (Central
Composite Faced) design (Montgomery, 1991). There are totally 15 experiments. The
experimental design and related results are shown in Table 5.11.
Table 5.10 – Design Factors and Their Ranges.

Design Factor Low Band (-1) Middle (0) High Band (1)
Bottom Thickness (b2) (in.) 0.182 0.192 0.202
Height (h) (in.) 0.71 0.75 0.79
α) (o)
Draft Angle (α 0 1.25 2.5
Table 5.11 – Results of the Experiments for Rib Part.

Design Factors Response
Exp. b2 (in.) h (in.) α()o
Max. Deflection (inch)
1 0.202 (1) 0.79 (1) 2.5 (1) 0.0210
2 0.202 (1) 0.79 (1) 0 (-1) 0.0171
3 0.202 (1) 0.71 (-1) 2.5 (1) 0.0184
4 0.202 (1) 0.71 (-1) 0 (-1) 0.0136
5 0.182 (-1) 0.79 (1) 2.5 (1) 0.0295
6 0.182 (-1) 0.79 (1) 0 (-1) 0.0204
7 0.182 (-1) 0.71 (-1) 2.5 (1) 0.0243
8 0.182 (-1) 0.71 (-1) 0 (-1) 0.0154
9 0.202 (1) 0.75 (0) 1.25 (0) 0.0178
10 0.182 (-1) 0.75 (0) 1.25 (0) 0.0206
11 0.192 (0) 0.79 (1) 1.25 (0) 0.0205
12 0.192 (0) 0.71 (-1) 1.25 (0) 0.0154
13 0.192 (0) 0.75 (0) 2.5 (1) 0.0211
14 0.192 (0) 0.75 (0) 0 (-1) 0.0154
15 0.192 (0) 0.75 (0) 1.25 (0) 0.0178
In the FEA analysis, the element type that is used is Brick 20 node (solid 95). The
author also noticed that the maximum stress given by ANSYS is around 4000 psi, which
is less than 6000 psi. Therefore the stress for the rib design is not a main concern. Two
screen dumps given by ANSYS for two experiments (experiment 6 and 7) are shown in
Figure 5.13.
192
Figure 5.13 – Two Screen Dumps of Experiment Results for Rib Part.
A statistical software system, MINITAB (www.minitab.com), was used to generate
the response surface model for the above experiments. The response surface equation
given by MINITAB is:
Max_Deflection = 0.43 - 4.799•b2 + 0.093•h + 0.021 • α + 14.722•b2•b2 + 0.139•h•h –
1.281•b2•h – 0.093•b2•α - 0.002•h•α
2 2
(R = 98.8%, R (adj) = 96.7%, Max. dev = 4.5%, Avg. dev = 1.98%)
193
The response surface of the maximum deflection has very high R2 and R2 (adj)
values. This indicates that the response surface fits very well through actual data. The
graphical relations of the response and the variables given by MINITAB are shown in
Figure 5.14. They are provided for a better understanding of the response surface
equation.
inch
inch
inch
(a) Relation of Deflection with h and b2
inch
inch
inch
(b) Relation of Deflection with b2 and α
Figure 5.14 – Graphical Relations of The Maximum Deflection and Variables.

Therefore the manufacturer can formulate the MGT problem by adding all the
manufacturing requirements. The complete MGT problem formulation is shown in Table
5.12. In the formulation five design goals are considered instead of the three goals given
194
in Table 5.9. The goals of draft angle (Equation 5.18) and shot number (Equation 5.19)
are added by the manufacturer.
Table 5.12 – Complete MGT Rib Problem Formulation.

Given:
! b2 = 0.192 inch, h = 0.75 inch, w= 0.852 inch
! Load F = 25 lb.
! Flexural Young’s modulus of the product YMsteel = 436710 psi
! Flexural Young’s modulus of the prototype YMSLA = 582280 psi
! The maximum deflection of the rib MDp = 0.02 inch
! Weight W1, W2, W3, W4, W5
! 50 prototypes are needed.
Find:
• Bottom Thickness, b2
• Height, h
• Draft angle, α
! The maximum deflection, MDm
! Shot number of a mold before failure, NS.
• d1+, d1-, d2+, d2-, d3+, d3-, d4+, d4-, d5+, d5-
Satisfy:
! Goals:
0192
.
• Bottom thickness: + d −1 − d +1 = 1
b2
0.75
• Height: + d −2 − d +2 = 1
h
0.02
• Maximum Deflection: + d −3 − d +3 = 1
MDm
0.5
• Draft Angle: + d −4 − d +4 = 1 [5.18]
α
50
• Shot number: + d −5 − d +5 = 1 [5.19]
NS
! The maximum deflection Response Surface Model:
• MD = 0.43- 4.799•b2+ 0.093•h +0.021•α +14.722•b2•b2+ 0.139•h•h- 1.281•b2•h –
0.093•b2•α - 0.002•h•α
! 0.019 inch ≤ MDm ≤ 0.021 inch
! Top thickness: b1 = b2 – 2•h•tan(α)
! Requirement of ejector pin: b1 > 0.125
! Height ratio: hr = h/b2
! Number of shot: NS = 23.13 – 14.54•hr + 16.06•α + 1.4•hr•hr + 0.67•α•α - 2.24•hr•α
! di-, di+ > 0 i = 1, …, 5
di • di = 0
- +
i = 1, …, 5
195
0.182 inch ≤ b2 ≤ 0.202 inch

0.71 inch ≤ h ≤ 0.79 inch
0o ≤ α ≤ 2.5o
Minimize:
• Z = W1( d −1 + d +1 ) + W2 ( d − 2 + d + 2 ) + W3 ( d − 3 + d + 3 ) + W4 ( d − 4 + d + 4 ) + W5 ( d − 5 + d + 5 )
( ΣWi = 1)
A C program using the exhaustive searching method is developed to solve the

formulated problem. Four different goal preference scenarios were investigated. All
goals were weighted evenly in Scenario 1. In Scenario 2, the dimensional goals were
weighted more highly. Scenarios 3 weighted the shot number goal more highly. In
Scenario 4 the shot number goal was not considered and all other four goals were
weighted evenly. Weights for the four scenarios are shown in Table 5.13.
Table 5.13– Scenarios of Goal Weights.

Dimension Goals Max. Shot
Scenario Thickness Height Goal Draft Angle Deflection Number
Goal Goal Goal Goal
1 0.2 0.2 0.2 0.2 0.2
2 0.3 0.3 0.3 0.05 0.05
3 0.1 0.1 0.1 0.1 0.6
4 0.25 0.25 0.25 0.25 0
Results of the MGT problem are shown in Table 5.14 for each of the four scenarios.
As can be seen, the results across all scenarios are almost identical, indicating a solution
that is stable with respect to preferences. However, the deviations of the solutions are
rather high. This is because the goal for shot number is set too high (50) and the response
surface equation of the shot number is crude. However, the author believes these
solutions can help the designer and manufacturer to understand the possible difficulties in
making tradeoffs. Therefore the negotiations between them may be speeded up.
Table 5.14– Rib Part MGT Results.

Scenario Bottom Height Draft Max. Shot
Thickness h (inch) Angle α Deflection Number Deviation
b2 (inch) o
() MDm (inch) NS
1 0.19 0.74 2.5 0.0191 10 0.97
2 0.19 0.74 2.5 0.0191 10 0.45
3 0.19 0.74 2.5 0.0191 10 2.48
4 0.195 0.79 1.7 0.0191 1 0.21
Since the main goal of this example is to illustrate the capability of the MGTDT in
integrating the design and manufacturing requirements, no further errors are made in
improving the accuracy of the solutions for the example. Based on the results, the
manufacturer selects the following rib part dimensions:
196
b2 = 0.19 inch, h = 0.74 inch, α= 2.5o, MDm = 0.0191, NS = 10.

• Discussion
Based on the simulation of ANSYS, the maximum deflection of the rib part
produced by SLA molds is 0.0178 inch if no geometric tailoring is executed. The
maximum deflection of the production part is 0.02 inch. Therefore the error between the
prototype and production parts is (0.02-0.0178)/0.02 = 11%, which is bigger than the
error requirement (5%). In some cases, it may bring problems in the design verification
based on the functional testing. For example, the maximum deflection of the prototypes
in this example is 0.0178 inch, which is less than what the production design will perform
(0.02 inch). Therefore even the running results given by the functional testing are
satisfactory, the product design may still fail to achieve the design specifications.
However by using the tailored dimensions, the error between the prototype and
production parts is only (0.02-0.0191)/0.02 = 4.5%, which is within the error requirement
(5%).
Although the response surface equation of the shot number is not accurate, it can
foster a better understanding of the necessity of the geometric tailoring. When the mold
life requirement is considered, the draft angle given by solving the MGT problem is 2.5o.
Correspondingly the SLA molds can produce 10 parts before they fail. However if this
manufacturing requirement is not considered, draft angle is 1.7o and the shot number for
the SLA molds is only 1 (Scenario 4). This will be not acceptable in producing the
prototypes. Even worse, the draft angle given by the designer is 0o. The manufacturer
may not be able to get a single part for this design. In Chapter 6, the author will present
an approach for Rapid Tooling in order to consider more requirements that are related to
the manufacturing process.
Figure 5.15 – Parts Produced by SLA Molds.
197
For the tailored part design, the author used a SLA-3500 and Morgan-press injection-
molding machine in producing molds and parts respectively. The material is a general-
purpose polystyrene from the Dow Chemical Company (www.dow.com). Over 18 parts
were made before a pullout failure happened in a SLA mold piece. Two parts are shown
in Figure 5.15. The SLA molds and the pullout failure are shown in Figure 5.16.
Pullout
Failure
Figure 5.16 – SLA Molds and Pullout Failure.

However, because of the lack of experiment device, the deflection of the prototypes
is not tested under the load. In the next section a ring gear example is presented, in which
two functional properties are considered in one prototype.
5.5.3 Building Prototypes of a Ring Gear

The design problem under consideration here is the design of ring gears for a speed
reducer in a family of cordless drills. These speed reducers are planetary gear trains,
Ring gear of
interest
To
Chuck
From
Motor
Layer 1 Layer 2 Layer 3
Figure 5.17 – The Ring Gear and the Speed Reducer of a Cordless Drill.
198
typically with three stages, as shown in Figure 5.17. Suppose ring gears for layer 1 and 2
are made of injection molded, atactic polystyrene. In this problem the author only
considered the ring gear for layer 1.
For different drill with voltage of 9.6V, 12V, 14.4V and 18V, the maximum torque
to chuck is given in Table 5.15. A gearshift is used for providing two different output
speeds (1400 rpm and 400 rpm) for the user by engaging and disengaging the second
planetary layer.
Table 5.15 – Maximum Torque of The Speed Reducer.

Voltage (V) Maximum Torque (in-lb)
9.6 156
12 200
14.4 237
18 330
Suppose a gear train has been designed and a prototype gear train is required for
functional testing. Rather than waiting for production injection mold tooling for the ring
gear to be machined, the designer wants to fabricate prototype ring gears using a Rapid
Tooling process. In the cordless drill design, the reliability of its transmission is rather
important. Therefore suppose the prototype ring gears are to be used for fatigue testing.
The error requirement of the fatigue is 5%.
Figure 5.18 – Some Terminology Describing a Spur Gear (Shigley and Mischke,
1989).
A spur ring gear can be determined by three design variables, teeth number (N), pitch
diameter (PD), and face width (W). They are shown with other terminologies in Figure
199
5.18 (Shigley and Mischke, 1989). The diameter of the pitch circle is called pitch
diameter.
Suppose the ring gear design in layer 1 are: N=54, PD = 1.335inch, F= 0.215 inch.
Since the speed reducing ratio of the gear train is related to the teeth number by equation
N2 ω1
= . The designer may not want the manufacturer to change the number of N
N1 ω 2
because of the speed requirements. Therefore face width (F) and pitch diameter (PD) are
the design factors that can be tailored by the manufacturer. Suppose their ranges are:
0.204 ≤ F ≤ 0.226;
1.268 ≤ h ≤ 1.402.
Then because the fatigue property of the ring gear is investigated in the functional
testing, the designer should determine an approach to evaluate the fatigue of the ring
gear. An approach based on the maximum stress is presented as follows.
When the cordless drill is in work, the ring gear is under fluctuating stresses as
shown in Figure 5.19.a. Suppose the maximum Von Mises stress is σmax. Since σmin = 0
for the ring gear, one can get
mean stress (σm) = (σmax + σmin ) / 2 = σmax /2,
stress amplitude (σa) = (σmax - σmin ) / 2 = σmax /2.
σ Sa
σmax
σa
0
Time N
0
(a) Fluctuating Stress in the Gear (b) The Stress-Life Curve
Figure 5.19 – Fatigue of Ring Gear.

The fatigue, or gear life, can be represented by the number of cycles (N) that the
teeth can run before it breaks. The gear life is related to the fluctuating stresses in the
gear teeth (Shigley and Mischke, 1989). Suppose a factor is c, nominal stress Sa = cσa.
The relation of Sa and N can be illustrated in Figure 5.19.b.
Dowling (1993) gave a way to calculate the number of cycles for given fluctuating
stresses. For different materials including polymer,
N = (Sa/a)1/b,
where a= k1 • Sut, b = k2.
In the equations, Sut is the tensile strength of the material, k1 and k2 are all factors.
Since Sa = c•σa = c•σmax /2, for a teeth of the ring gear,
200
N = [Sa / (k1• Sut)]1/k2 = k• (σmax / Sut) 1/k2. [5.20]

Suppose k = kakbkckdkekf, where ka is surface factor, kb is size factor, kc is load factor,
kd is temperature factor, ke is material factor, kf is miscellaneous-effect factor (corrosion,
residual stress, metal spraying, cyclic frequency, stress concentration, etc.). Since the
production and prototype gears have similar surface, size, load, temperature, and
material, ka, kb, kc, kd, ke, kf in the above equations can be thought as equal.
Based on Equation 5.20,
Nprototype = kprototype• (σmax/Sut)prototype 1/k2prototype,
Nproduct = kproduct • (σmax/Sut) product1/k2product,
Therefore, similar fatigue performance (Nproduct = Nprototype) can be achieved by
making the ratio σmax/Sut of prototype gears and production gears as similar as
possible.
In this example, since the same ring gear is operated in two different speeds, two
different loads are added to the ring gear. In the functional testing, it is desired that the
prototype ring gears will behavior similar to the performance of the production ring gears
in both loads. Therefore two functional properties (MS1, MS2) should be considered in
the MGT formulation.
Related to different output speeds, two speed ratios are 0.0613 (use layer 1, 3) and
0.019 (use layer 1, 2, 3). Suppose the number of planet gear is 5. For the given design,
High torque T1= 156/5*0.0613=1.913 lb-in;
Low torque T2= 156/5*0.019=0.593 lb-in.
The maximum stresses of the product ring gear for low and high speed given by
ANSYS are 3282 psi and 1017 psi respectively. The tensile strength of a production ring
gear made of atactic polystyrene is 37.4 Gpa (Dawson, 1998). That is, Sut, product = 37.4
Gpa = 5430 psi.
Based on all the information, the designer may initiate a MGT ring gear problem
formulation as shown in Table 5.16.
Table 5.16 - MGT Ring Gear Problem Formulation By the Designer.

Given:
! N = 54, PD = 1.335 inch, W= 0.215 inch
! Load T1 = 1.913 lb-in, T2 = 0.593 lb-in
! Yield Strength of the product Su, p = 5430 psi
! Yield Strength of the prototype Su, m **
! The maximum stress of the product gear under T1: MS1, p = 3282 psi
MS1, p • Su,m
! Target maximum stress of the prototype gear under T1: TMS1,m =
Su, p
MS2, p • Su,m
Su, p
201
! Weight W1, W2, W3, W4

Find:
• Pitch Diameter, PD
• Face Width, W
! The maximum stress under T1, MS1,m
• d1+, d1-, d2+, d2-, d3+, d3--, d4+, d4-
Satisfy:
! Goals:
1335
.
• Pitch Diameter: + d −1 − d +1 = 1 [5.21]
PD
0.215
• Face Width: + d −2 − d +2 = 1 [5.22]
W
TMS1,m
• Maximum Stress under T1: + d −3 − d +3 = 1 [5.23]
MS1,m
TMS2,m
• Maximum Stress under T2: + d −4 − d +4 = 1 [5.24]
MS2,m
• MS1,m = f1(PD, W, T1) **
• MS2,m = f2(PD, W, T2) **
! 2733 psi ≤ MS1,m ≤ 3021 psi
846.9 psi ≤ MS2,m ≤ 936.1 psi
! di-, di+ > 0 i = 1, 2, 3, 4
di- • di+ = 0 i = 1, 2, 3, 4
1.268 inch ≤ PD ≤ 1.402 inch
0.204 inch ≤ W ≤ 0.226 inch
Minimize:
• Z = W1( d −1 + d +1 ) + W2 ( d − 2 + d + 2 ) + W3 ( d − 3 + d + 3 ) + W4 ( d − 4 + d + 4 ),( ΣWi = 1)
In the MGT formulation (Table 5.16), four goals include two goals on geometry
(Equations 5.21 and 5.22) and two goals on maximum stress (Equations 5.23 and 5.24).
The MGT formulation and CAD model of the part can be sent to the manufacturer. In
addition, the designer can also send the FEA model that was used in determining the
maximum stress of the ring gear to the manufacturer.
202
After receiving the design information, the manufacturer can begin the geometric
tailoring for the part. Suppose the AIM tooling for polystyrene is chose for the prototype.
Correspondingly the manufacturer knows the yield strength of atactic polystyrene of
gears produced with SLA molds are Su,m = 32.8 Gpa = 4760 psi (Dawson, 1998).
The manufacturer can also employ the FEA model to understand the relation of the
maximum stress with the material and geometry differences of the prototype ring gear.
The design variables given by the designer are PD and W. Correspondingly the design
factors and their ranges for the experiments are given in Table 5.17. CCF (Central
Composite Faced) design is used in the experimental design in this example
(Montgomery, 1991). There are totally 9 experiments for each load. The experimental
design and related results given by ANSYS are shown in Table 5.18.
Table 5.17 – Design Factors and Their Ranges of Ring Gear.

Thickness W (inch) 0.204 0.215 0.226
Pitch diameter PD (inch) 1.268 1.335 1.402
Table 5.18 – Results of the Experiments for Ring Gear.

Design Factors Response
Max. Stress (psi)
Exp. W (in.) PD (in.) Torque = Torque =
1.913 lb-in 0.593 lb-in
1 0.226 (1) 1.402 (1) 2581 800
2 0.226 (1) 1.268 (-1) 3325 1031
3 0.204 (-1) 1.402 (1) 2939 911
4 0.204 (-1) 1.268 (-1) 3591 1113
5 0.215 (0) 1.402 (1) 2924 906.4
6 0.215 (0) 1.268 (-1) 3535 1096
7 0.226 (1) 1.335 (0) 2956 916.4
8 0.204 (-1) 1.335 (0) 3183 986.8
9 0.215 (0) 1.335 (0) 3282 1017
In the FEA analysis, the element type that is used is Tet-10 node (solid 72). Two
screen dumps given by ANSYS for two experiments (experiment 1 and 7) are shown in
Figure 5.20.
Based on the results of the experiments, the response surface equations for the
maximum stress given by the statistical software (MINITAB) are:
MS1,m = -50522 + 565972*W - 3537*PD -1249312*W*W + 1968*PD*PD - 31208*W*PD;
(R2 = 99%, R2 (adj) = 97.3%, Max. dev = 1.5%, Avg. dev = 0.8%)
MS2,m = -15634 + 175337*W - 1119*PD - 386502*W*W + 631*PD*PD - 9837*W*PD.
(R2 = 99.0%, R2 (adj) = 97.3%, Max. dev = 1.5%, Avg. dev = 0.8%)
203
Figure 5.20 – Two Screen Dumps of Experiment Results for Ring Gear.
The response surfaces of the maximum stresses have very high R2 and R2 (adj)
values. This indicates that the response surface fits very well through actual data. The
graphical relation of MS1,m and the design variables (PD and W) given by MINITAB is
shown in Figure 5.21. It is provided here for a better understanding of the response
surface equation. Although not shown, MS2,m has a similar relation with the design
variables.
Based on all the information, the manufacturer can formulate a complete MGT
problem as shown in Table 5.19.
204
Figure 5.21 – Graphical Relation of the Maximum Stress with PD and W.
Table 5.19 – Complete MGT Ring Gear Problem Formulation.

Given:
! N = 54, PD = 1.335 inch, W= 0.215 inch
! Load T1 = 1.913 lb-in, T2 = 0.593 lb-in
! Yield Strength of the product Su, p = 5430 psi
! Yield Strength of the prototype Su, m = 4760 psi
MS1, p • Su,m
Su, p
MS2, p • Su,m
Su, p
! Weight W1, W2, W3, W4
Find:
• Pitch Diameter, PD
• Face Width, W
205

• d1+, d1-, d2+, d2-, d3+, d3--, d4+, d4-
Satisfy:
! Goals:
1335
.
• Pitch Diameter: + d −1 − d +1 = 1
PD
0.215
• Face Width: + d −2 − d +2 = 1
W
TMS1,m
• Maximum Stress under T1: + d −3 − d +3 = 1
MS1,m
TMS2,m
• Maximum Stress under T2: + d −4 − d +4 = 1
MS2,m
• MS1,m = -50522 + 565972*W - 3537*PD -1249312*W*W + 1968*PD*PD - 31208*W*PD
• MS2,m = -15634 + 175337*W - 1119*PD - 386502*W*W + 631*PD*PD - 9837*W*PD
! 2733 psi ≤ MS1,m ≤ 3021 psi
846.9 psi ≤ MS2,m ≤ 936.1 psi
! di-, di+ > 0 i = 1, 2, 3, 4
di • di = 0
- +
i = 1, 2, 3, 4
1.268 inch ≤ PD ≤ 1.402 inch
0.204 inch ≤ W ≤ 0.226 inch
Minimize:
• Z = W1( d −1 + d +1 ) + W2 ( d − 2 + d + 2 ) + W3 ( d − 3 + d + 3 ) + W4 ( d − 4 + d + 4 ),( ΣWi = 1)
A C program using exhaustive searching method is developed to solve the

formulated problem. Three scenarios of goal preference were investigated. All goals
were weighted evenly in scenario 1. The two other scenarios weighted the dimension
goals more highly, as in scenario 2, or weighted the maximum stress goals more highly,
as in scenario 3. Weights for the three scenarios are shown in Table 5.20.
Table 5.20– Scenarios of Goal Weights.

Dimension Goals Maximum Stress 1 Maximum Stress 2
Scenario Face Width Pitch Diameter Goal Goal
Goal Goal
1 0.25 0.25 0.25 0.25
2 0.4 0.4 0.1 0.1
3 0.1 0.1 0.4 0.4
206
Results of the MGT problem are shown in Table 5.21 for each of the three scenarios.
As can be seen, the results across all scenarios are almost identical, indicating a solution
that is stable with respect to preferences.
Table 5.21– Ring Gear MGT Results.

Scenario Face Width w Pitch Diameter Max. Stress Max. Stress Deviation
(inch) pd (inch) 1 MS1 (psi) 2 MS2 (psi) (%)
1 0.216 1.401 2837.5 900.1 1
2 0.215 1.402 2885 903.7 2
3 0.216 1.401 2837.5 900 1.6
Based on the results, the manufacturer selects the following ring gear dimensions:
W = 0.216 inch, PD = 1.401 inch, MS1 = 2837.5, MS2 = 900.
Discussion:
If no geometric tailoring is executed, the maximum stress of the prototype ring gear
is 3282 psi to transfer T1, and 1017 psi to transfer T2. Therefore the errors between the
prototype and production gears are (3282 – 3282•4760/5430)/ (3282•4760/5430) = 14%
for T1 and (1017 - 1017•4760/5430)/ (1017•4760/5430) =14% for T2. Therefore the
fatigue property of the prototype gears will be different from that of the product gears.
After tailoring the dimensions, the error reduced to 1.4% (T1) and 1% (T2) respectively.
For the tailored part design, the author used a SLA-3500 and a Sumitomo (75 Ton)
injection-molding machine (www.sumitomopm.com) in producing molds and parts
respectively. The material is general-purpose polystyrene from the Dow Chemical
Company. Over 10 shots were made. The SLA mold is still in good condition after the
injection-molding process was stopped. The main injection molding parameters used in
the process were:
Temperature setting:
Zone 1: Throat of Hopper 220 oC
Zone 2: Melt Zone 225 oC
Zone 3: Transition Zone 230 oC
Zone 4: Metered Zone 235 oC
Zone 15 Nozzle 235 oC
Clamping Force: 50 tons
Hold Time: 15 second
Hold Pressure: 8.0 kgf/cm2
Shot Size: 40.5 mm
Cooling Time: 450 second
Cycle Time: 8 minute
Injection Speed: 25 mm/sec
Some of the produced ring gears are shown in Figure 5.22. One of the SLA mold
pieces installed in the standard mold plate of the injection-molding machine is shown in
Figure 5.23. However, because of the lack of experiment device, the maximum stresses
of the prototypes are not tested under the loads.
207
Figure 5.22 – A Photo of the Prototype Ring Gears.
Figure 5.23 – A SLA Mold Piece in a Standard Mold Plate.
208
After the test examples are presented for the MGT and MGTDT, a brief summary is
given in the next section, which discusses the relevance of these results with regard to the
hypotheses of the dissertation.
prototypes are produced. In this chapter and the next chapter, methods on geometric
tailoring for Rapid Tooling are presented to address the problem. One geometric
tailoring problem, material geometric tailoring, is considered in Chapter 5. The
properties of Rapid Tooling, especially the part properties of the AIM tooling, are
introduced in Section 5.1 to provide a context of material geometric tailoring. Related to
the principle of functional testing and similarity methods, the fundamentals of geometric
tailoring are presented in Section 5.2. The material geometric tailoring decision template
is introduced in Section 5.3, which enables a “clean digital interface” between design and
fabrication, effectively separating design activities from manufacture activities. The
usage of the MGT decision template, including formulating function properties and
solving approaches, is presented in Section 5.4. Finally three test examples are discussed
in Section 5.5 to demonstrate a scenario of design-manufacture collaboration with the
MGT decision template.
The research question and hypothesis that are related to this chapter are Q2.1 and
sub-hypothesis 2.1, which are repeated here:
Q2.1. How to reduce the iterations between the designer and manufacturer in
producing functional prototypes that have different material properties
from products?
Sub-Hypothesis 2.1: The designer can initiate a material geometric tailoring
(MGT) formulation based on a MGT decision template; therefore the
manufacturer, who completes and solves the MGT problem, can produce
production-representative prototypes more quickly.
Although not explicitly presented in the chapter, the discussions on the properties of
Rapid Tooling, the Buckingham Π theorem, information flow and information processing
for the different steps in using MGTDT provide partial theoretical structural validation
for Hypothesis 2.1. In Section 5.5, the usage of MGTDT in designing a prototype tensile
bar, rib part and ring gear is presented. The discussions on the problem requirements of
the examples and the usage of the MGTDT provide partial empirical structure validation
for Hypothesis 2.1, and the discussions on solving process and physical validation of the
examples provide partial empirical performance validation for Hypothesis 2.1. These
results are summarized in Figure 5.24.
The author also made an effort on providing theoretical performance validation for
Hypothesis 2.1. Some explanations on the performance validation of the hypothesis are
given as follows.
Currently the decisions on part design are made by the designer. In the design
process, the designer formulates goals, constraints and preferences to make the decisions.
However these decision factors are not transferred to the next stages (e.g. Design-for-
209
Hypothesis 2.1
Theoretical Structure Validation Theoretical Performance Validation
The MGT is based on Buckingham PI

theorem; the MGTDT is based on decision
based design and other design technologies
including compromise DSP and RSM.
Empirical Structure Validation Empirical Performance Validation
The requirements on producing functional Production-representative prototypes
prototypes of a tensile bar, a rib part, and of the tensile bar, rib part and ring
a ring gear using the AIM tooling are gear were produced based on the
representative of the problems. The scenario of using MGTDT to transfer
MGTDT can be used to formulate the DFM to the manufacturer without
design and manufacturing requirements. iterations.
Potential Time
Savings
100%
Knowledge
Increase about Design
Maintain Knowledge
CUMULATIVE
Freedom
Design
Freedom
0%
Design Timeline
Figure 5.24– Summary of Hypothesis Validation.
Manufacture). So design freedom decreases quickly. Consequently the cost and time to
make decisions in the later stages increase dramatically. In this research, the MGTDT is
proposed to formulate sufficient design information for the material geometric tailoring
problem. By communicating the formulated design information to later stages, the
knowledge about design increases without decreasing design freedom dramatically. With
the maintained design freedom, some difficulties in the later stages can be solved easily
and quickly (Figure 5.24). This was illustrated in the discussions of the tensile bar
example (Section 5.5.1).
The validation of hypothesis 2.1 also partially supports Q2 and hypothesis 2, which
are repeated here:
Q2. How to reduce the time of iteration between the designer and manufacturer
in the usage of Rapid Tooling for a wide variety of design requirements?
210

Among different design requirements considered in Q2, only functional properties
that are related to material properties are considered in Q2.1. Accordingly, for the
geometric tailoring considered in hypothesis 2, only material geometric tailoring is tested
in hypothesis 2.1.
With the principles of geometric tailoring introduced in this chapter, a design for
Rapid Tooling system is presented in the next chapter (Figure 5.25). Besides the
requirements that are related to material properties, other requirements that are related to
process planning are also considered in the system (Section 6.1). The Material-Process
Geometric Tailoring (MPGT) problem and the MPGT decision template are described in
Section 6.3. A solving approach for the MPGT problem is presented in Section 6.4.
Part Design
Parametric
Part Design
CAD Model
Requirements
of Part

Design System
Parting Directions
(Section 6.3)
Parting Surfaces
M old Piece Number
The Designer's MPGT
Problem Formulation
Parametric CAD Given
Part Design
Satisfy
Process Analyzer Minimize
Surface Fin ish Draft A ngle
Cost Part Thickness
Time Mold Lif e

Given Given
Find Find
RP Proc ess Parameters IJM Process Param eters (Section 6.2.3)
Satisfy Satisfy
Goals Goals
Minimize Minimize
Cost Predictor
Cost

& 6.5) DSP Solver
Input and Output
Proces sor
related Mold Design, RP and C-DSP Template

Part P Part P
F1 F9 F1
F9
F3 F3 Given
F2 (1) (2)
F6 F8 F6 F8
Analternative tobeimprovedthroughmodification; Assumptions usedtomodel the domainof interest.
F2 Thesystemparameters:
p equalityconstraints q inequalityconstraints
Fn Fn
F4 F5 F5
fk(di ) functionof deviationvariables tobeminimizedat priority level k for thepreemptive case.
F4 Find
Values for thesystemvariables Xi i = 1, ... , n
F7 F7 Values for thedeviation variables di-, di+ i = 1, ... , m
Satisfy
Part P gi(X) = 0; i = 1, ..., p gi(X) ≥ 0 ; i = p+1, ..., p+q
PD1
Mold Base Bounds
Ai(X) + d i- - di+ = Gi ; i = 1, ..., m
F1 R3 Ximin ≤ Xi ≤ Ximax ; i = 1, ..., n

R1 F9 M1 F1 Deviationvariables
F3 F3
di-, di+ ≥ 0; di- . d i+ = 0; i = 1, ..., m
(3) PD1 Minimize
F2 F8 F2 Preemptive deviationfunction(lexicographicminimum)
PL1
F6 Z=[f1(d-i,d+i ),..., fk(d-i, d+i )]
Archimedaindeviationfunction
Fn M2 F5 Z= ∑W(d −
+ di+ ) where ∑W =1, W ≥0
F5 F4 Mk i i i i
F4 PL2 Rk F7
PD2 Fn
R2 F7
PD2

Design method

Tools
§2.2 §2.3 §2.4 §2.5 §2.6
211
Chapter 6 – A Decision-Based Design for Rapid Tooling System
CHAPTER 6
A DECISION-BASED DESIGN FOR RAPID TOOLING SYSTEM
Part Design
Parametric
Part Design
CAD Model
Requirements
of Part

Design System
Parting Directions
(Section 6.3)
Parting Surfaces
Mold Piece Number
The Designer's MPGT
Problem Formulation
Part Design
Satisfy
Process Analyzer
Surface Finish Draft A ngle Minimize
Cos t Part Thickness
Time Mold Life

Given Given
Find Find
RP Process Parameters IJM Process Parameters (Section 6.2.3)
Satisfy Satisfy
Goals Goals
Minimize Minimize Cost Predictor
Cost

& 6.5) DSP Solver
Input and Output
Processor
prototypes are produced. Based on the principle of geometric tailoring presented in the
last chapter, a design for Rapid Tooling system (DFRTS) was presented for material-
process geometric tailoring (MPGT) to address the problem in this chapter. First the
infrastructure and scope of the DFRTS are introduced in Section 6.1. The components of
the DFRTS that are related to the process planning of the AIM tooling are described in
Section 6.2. The MPGT decision template for the designer and the integrated MPGT
problem formulation are presented in Section 6.3. A solution strategy for the MPGT
problem and a three-stage solution process for the DFRTS are described in Section 6.4.
Finally a comparison of the DFRTS and the current usage of RT is given from the
perspective of decision-making.
212
6.1 OVERVIEW OF DESIGN FOR RAPID TOOLING SYSTEM

As stated in Section 1.2.3, the “customers” of the RTTB (Section 1.2.2) have a wide
variety of design requirements. These requirements for the prototype parts may include
the shape of the products, function properties, tolerances, surface finish, batch size, cost
and time. Among these requirements for Rapid Tooling, the requirements on function
properties are related to the material properties of the prototypes. In Chapter 5, the
material geometric tailoring (MGT) problem and MGT decision template were presented
to address the differences of the material properties between the products and prototypes.
The principles of geometric tailoring (Section 5.2.3) and MGTDT (Section 5.3.1) are
applicable to the material-process geometric tailoring (MPGT) and MPGT decision
template that are to be presented in Section 6.3. The example of the rib part (Section
5.5.2) also sets the stage for the research in this chapter with more manufacturing
requirements considered.
Some requirements for Rapid Tooling, such as tolerances, surface finish, batch size,
cost and time, are tightly related to the fabrication process. Therefore the process
planning of the Rapid Tooling should also be considered. Consequently a Design for
Rapid Tooling System (DFRTS) is proposed in this chapter to address the geometric
tailoring problem related to these requirements. Its relation with geometric tailoring and
DFM is described in Section 1.2.4 and repeated here in Figure 6.1.
Considerations
on Topology,
Assembly,
Design for etc.
Manufacture
Geometric
Mold Design
Tailoring
Design for
Rapid Tooling
(Chp 6)
IJM Process RP Process
Planning Planning
Figure 6.1 – Relations of DFRT, Geometric Tailoring and DFM.

In Chapter 5, the approach based on the MGTDT formulated the decisions in design
stages into a compromise DSP, and then transferred it to later stages. The requirements
of the later stages were integrated into the problem formulation, which was then solved
for a satisficing solution (Section 5.3.1). In this chapter, the DFRTS employs the same
principle. That is, the decisions of the part design, RP process planning, and IJM process
planning are formulated into compromise DSPs. Instead of solving them for ‘optimal’
solutions for each stage, the formulations are transferred to later stages and integrated
into a bigger problem. Finally a compromise DSP solver is developed in the DFRTS to
get a satisficing solution for all the stages. The process is illustrated in Figure 6.2.
213
Part Design
Parametric
Part Design
CAD Model
Requirements
of Part
(Chp3 & 4)
Design System
Parting Directions
Parting Lines
Parting Surfaces
(§6.3.1)
Mold Piece Number
The Designer's MPGT
Problem Formulation
Part Design
Design Parameters
(§6.2.1) (§6.2.2) Satisfy
Building Directin Cooling Time
Layer Thickness Draft Angle Deviation
Fill Overcure Rib Height/width Ratio
Hatch Overcure Part Thickness

Planning Formulation Planning Formulation
Given Given
Find Find
RP Process Parameters IJM Process Parameters
Satisfy Satisfy (§6.2.3)
Goals Goals
Minimize Minimize
Cost Estimator
Cost
(§6.3.2, §6.4 ) D. MPGT Problem

Solver
Input and Output
Processor
Figure 6.2 – Infrastructure of the DFRTS and Related Sections.

The related sections in this dissertation for the components of the DFRTS are also
shown in Figure 6.2. The input to the system is the part design whose functional
prototypes are to be produced by the Rapid Tooling. For the given CAD model of the
part, the method and related system for generating CAD models of mold pieces are
presented in Chapter 3 and 4 respectively. The MPGT decision template for design
214
requirements is presented in Section 6.3.1. The RP process planner (B) and injection
molding process analyzer (C) are discussed in Section 6.2.1 and Section 6.2.2
respectively. The MPGT problem formulation and the related MPGT problem solver (D)
are presented in Section 6.3.2 and 6.4. By solving the integrated problem, the tailored
part design and the related mold design, Rapid Prototyping and injection molding process
parameters are determined.
The DFRTS is a decision-based system because the decisions and their formulations
provide a unified framework for integrating design and manufacturing requirements for
design-for-manufacturing (DFM) problem. In this chapter, the author will focus on
formulating the decisions in the part design and solving the integrated problem.
Before the components of the DFRTS are introduced, the research scope of the
system is described as follows. As shown in Figure 6.3, four phases in using the direct
AIM tooling to build prototypes are (1) design part, (2) design mold, (3) build mold, and
(4) build part. For each phase, there are several variables. The variables that are
considered in the DFRTS are marked by underlines in the figure. Some other variables in
phases II, III, and IV that are not considered are also shown in the figure. The
determination of the research scope was mainly based on the importance of the variables,
and also the research work that had done at our lab (Rapid Prototyping and
Manufacturing Institute). In the next section, the computational modules that are
developed for RP and Injection molding (IJM) process are presented. Related to these
modules, the software systems of SLA process planner, mold life predictor, and RT cost
estimator are also described.
Part
CAD
Mold build by RP
Cost
Parting Direction Layer Thickness Cooling Time
Time
Parting Lines Building Orientation Thermal Cure
Part Number
Parting Surface Hatch Overcure Nozzle Temperature
Stress
Part Orientation Fill Overcure Clamping Force
Strain
Ejection Pin No. Z-level wait Injection Pressure
Weight
Ejection Pin Position Sweep period Injection Time
Surface finish
Sprue Type Boder Overcure Holding Time
Flatness tolerance
Gate Size Pre-dip delay Packing Pressure
Parallelism tolerance
Cooling Channels Post-dip delay etc.
Positional tolerance
etc. etc.
Circularity tolerance
Concentricity tolerance
Perpendicularity tolerance
Figure 6.3 – Research Scope of the DFRTS.
215
6.2 PROCESSING PLANNING OF THE AIM TOOLING

The process planning of the AIM tooling includes the process planning of Rapid
Prototyping (RP) and injection molding (IJM) processes. The computational modules of
building time, tolerances, and surface finish that are related to the RP process are
presented in (McClurkin, 1997; Lynn, 1998; West, 1999). The computational modules of
mold life that are related to the IJM process are presented in (Cedorge, 1999; Le Baut,
1999; Palmer, 1999; Pham, 2001; Rodet, 2001). The software systems that were
developed based on these computational modules are presented in (Sambu, 2001). In the
work of the DFRTS the author worked closely with Shiva Sambu, a M.S. student at our
lab. We also used the same case studies, robot arm and camera roller, which will be
presented in Chapter 7 and 8 respectively.
To foster a better understanding of the DFRTS, the word and mathematical
formulations of the RP and IJM processes shown in Figure 6.2 are presented in Section
6.2.1 and 6.2.2. The software systems of SLA process planner, mold life predictor, and
RT cost estimator are briefly discussed in Section 6.2.1, 6.2.2, and 6.2.3 respectively.
6.2.1 SLA Process Planner

Related to the RP process used in the AIM tooling (Section 1.1.3), SLA process
planner corresponds to the RP process planner (B) shown in Figure 6.2. SLA process
planner develops a plan that is used to fabricate a part in the stereolithography process
(Section 1.1.2). These plans consist of a set of parameter values that influence or control
how the part is to be fabricated.
Many properties of the fabricated parts, such as surface finish and tolerances, depend
on the values of the process variables (e.g. build orientation and layer thickness).
However changing the value of a process variable may affect several properties in the
ways that may conflict with each other. For example, decreasing the layer thickness in
the fabricating process can build parts with better surface finish. However, it will also
increase the building time and cost significantly. Therefore tradeoffs are necessary in
order to generate a good plan.
As a multi-objective problem, the SLA process planning can be formulated using the
compromise Decision Support Problem (cDSP). The word formulation of the SLA
process planning is given in Table 6.1.
Table 6.1 – SLA Process Planning Word Formulation (Sambu, 2001).

GIVEN:
• CAD model of part • SLA material property models
• Target values for SLA process goals • SLA process models
• Target mold life • Goal preferences as weights
FIND:
SLA process variables Deviation of goals from targets
SATISFY:
216
Goals: Constraints:
Meet target material properties Meet SLA process constraints
Meet targets of SLA process goals Bounds:
MINIMIZE:
As shown in Figure 6.3, the SLA process variables that are considered are Part
Orientation (PO), Layer Thickness (LT), Hatch Overcure (HOC) and Fill Overcure
(FOC). The SLA process goals are Surface Finish (SF), Accuracy (AC), Build Time
(BT), and material properties of Young’s Modulus (YM) and Tensile Strength (TS). For
the process variables and goals, the mathematical formulation of SLA process planning
problem is presented in Table 6.2. The formulation is a general math formulation which
can be extended for other RP processes. The exact quantitative relationship between
goals and system variables is given in (West, 1999) and (Sambu, 2001).
Table 6.2 – SLA Process Planning Mathematical Formulation (Sambu, 2001).

GIVEN:
• CAD model of part
• ACTi, SFTi • YMT, TST, EYT
• AC = f (PO, LT, HOC, FOC, ZL, SP) • YM = f (LT, HOC)
• SF = f (PO, LT) • TS = f (LT, HOC)
• BT = f (PO, LT, HOC, FOC, ZL, SP)
FIND:
Part Orientation in vat (PO) d1+, d1-, d2+, d2-, d3+, d3-, d4+, d4-,
Slicing scheme (LT) dk+, dk-, dj+, dj-
Hatch Overcure (HOC)
Fill Overcure (FOC)
Z-Level wait (ZL)
Sweep Period (SP)
SATISFY:
Goals: Constraints:
(YMT / YM) - d1+ + d1- = 1 di+ • di- = 0
(TST / TS) - d2+ + d2- = 1 Large horizontal planes
(BTmax - BT) / (BTmax - BTmin) - d4+ + d4- = 1 Support structures
(ACTk / ACk) - dk+ + dk- = 1 Bounds:
(SFTj / SFj) - dj+ + dj- = 1 Bounds for all system variables
MINIMIZE:
Deviation Function: g (d1+, d1-, d2+, d2-, d3+, d3-, d4-, dk-, dj-)
217
A software system based on the ACIS solid modeling kernel was developed in C++
on a PC by (West, 1999) and refined by (Sambu, 2001). A screen capture of the SLA
process planner is given in Figure 6.4.
Figure 6.4 – Screen Capture of the SLA Process Planner.
6.2.2 Stereolithography Mold Life Predictor

Related to the SLA molds that are used in the injection molding (IJM) process of the
AIM tooling, stereolithography mold life predictor corresponds to the injection molding
process analyzer (C) shown in Figure 6.2. As discussed in Section 5.1, the mold life of
the direct AIM tooling is much lower than that of the steel tooling because of the material
and fabrication properties of stereolithography molds. Therefore the mold life is a main
concern in the IJM process planning, and is formulated in the word formulation for the
IJM process planning (Table 6.3).
Table 6.3 – IJM Process Planning Word Formulation (Sambu, 2001).

GIVEN:
• CAD model of molds • Mold life models
• Target mold life • Goal preferences as weights
FIND:
IJM Process variables Deviation of goals from targets
218
SATISFY:
Goals: Constraints:
Meet target mold life Meet IJM process constraints
Bounds:
MINIMIZE:
Many factors affect the mold life of SLA molds (Cedorge, 1999; Le Baut, 1999;
Palmer, 1999; Pham, 2001; Rodet, 2001). These factors can be broadly classified into
four categories: geometry variables, injection molding parameters, SLA parameters and
material properties. Material properties include the properties of both part and mold
material. The factors that are investigated in the mold life predictor are shown in Table
6.4. These factors are chosen based on the data obtained from experiments performed at
Georgia Tech. Some of these factors are investigated qualitatively while the others are
investigated to obtain quantitative models.
Table 6.4 – Factors Selected for Mold Life Prediction (Sambu, 2001).
Injection molding
Geometry variables parameters SLA parameters Material properties
Feature type Injection pressure Layer Thickness Young’s modulus
Feature size Injection velocity Hatch Overcure Tensile strength
Draft angle Injection temperature Border Overcure Heat deflection temperature
Surface area Ejection temperature Laser beam size Glass transition temperature
Hydraulic radius Ejection force Degree of cure Friction coefficient
Bonding Area Cooling time Thermal cure Poisson’s ratio
Wall thickness Hold pressure Thermal expansion coefficient
Gating type
Among the injection molding parameters, only the cooling time is related to the
fabrication cost and time directly. Although the other variables are also related to the
mold life, they are mainly determined by some other requirements. For example, the
injection pressure should be large enough to enable the polymer to fill the whole cavity.
These requirements are not considered in the IJM process planning formulation of the
DFRTS. Therefore there is only one system variable (cooling time) in the cDSP
mathematical formulation of IJM process planning problem (Table 6.5).
In the formulation, cooling time (CT) is the time from injection of molten plastic into
the mold to the ejection of solidified part from the mold. The cooling time has a lower
bound of 180 seconds and an upper bound of 330 seconds. The lower bound is to ensure
that the parts are solidified before they are ejected out of the mold. The mold life
predictor is applicable for CT ≤ 330 and hence this value is chosen as the upper bound.
The sub-script ‘k’ corresponds to the number of molds (NM) required to injection mold
219
the desired number of parts (NP). So, there is a total of NM system variables in Table 6.5.
The formulation has two goals: cost and time. A modified cDSP formulation proposed in
(Hernandez and Mistree, 2001) is used to formulate the goals and their weights, which is
discussed in Section 6.3.2. The method to evaluate cost and time is presented in Section
6.2.3.
Table 6.5 – IJM Process Planning Mathematical Formulation (Sambu, 2001).

GIVEN:
• CAD models of mold pieces • RP variables used to build the mold
• Mold life ML = f (CT) • Number of parts NP
• Time T = f (ML, CT, NP) • Cost C = f (ML, CT, NP)
FIND:
CTk d j+, p , d j−, p
SATISFY:
Goals: Constraints:
+
C − max(C − t ,0) 1, p +1 C ≤ t1+, 5
Cost: +
+ d1−, p − d1+, p = 1
t1, p
T − max(T − t2+, p+1,0) T ≤ t2+, 5
Time: +
+ d2−, p − d2+, p = 1
t2 , p
Bounds: d j+, p * d j−, p = 0
180 ≤ CT ≤ 330 d j+, p , d j−, p ≥ 0
MINIMIZE:
cd h
4 2
Deviation Function: ∑∑w
p =1 j =1
j, p
+
j, p + d j−, p
A software system based on the mathematical modes was developed in C++ on a PC

by (Sambu, 2001). A screen capture of the SL mold life predictor is given in Figure 6.5.
220
Figure 6.5 – Screen Capture of SL Mold Life Predictor.
6.2.3 Rapid Tooling Cost Estimator

The cost and time of the fabrication process are related to the process planning.
Developing quantitative models of cost and time requires analysis of the whole rapid
tooling process. The steps involved in obtaining injection molded parts from the CAD
model of the part in rapid tooling process are listed in Table 6.6. The estimated values of
time and cost for each step are also given in Table 6.6.
Table 6.6 – Cost and Time Estimates for Different Steps in Rapid Tooling Process
(Sambu, 2001).
Step Time (hr) Cost ($ / hr)
Mold design 0.5 50
SLA Preparation 0.25 50
SLA 250 - 35
SLA Building BT
SLA 3500 - 65
Human – 0.25 Human – 20
RP Cleaning
TPM – 0.5 TPM – 5
Postcuring 1 10
Thermal curing TC * 8 10
Human – 0.25 Human – 20
Backfilling
Setting – 12 Setting – 0
Human – 20
Machining 0.25
Machine – 20
Human – 20
Assembling 0.25
Machine – 30
Molding parts CyT * NP 30
221
The time and cost of all the steps is straightforward except the SLA building step.
The build time of prototypes using a SLA machine depends upon the geometry of the
parts and also the SLA process variables. The SLA build time estimator, presented in
(McClurkin and Rosen, 1998), reads the vector (.v) and range (.r) files created by
3Dlightyear or Maestro, 3D System’s software, and calculates the build time of the
prototype to within roughly 2%. There are three common types of build vectors used in
the stereolithography process: fill vectors, hatch vectors, and border vectors. These three
types of vectors may be directly related to the properties of the sliced CAD model.
A software system based on the data in Table 6.6 and the SLA build time estimator
was developed in C++ on a PC by (Sambu, 2001). The Rapid Tooling cost estimator can
receive RT process parameters and return the time and cost related to the process
parameters.
After the process planning modules of the DFRTS are presented, MPGT and MPGT
decision template are described in the next section.
6.3 MPGT DECISION TEMPLATE AND MPGT PROBLEM FORMULATION
The designer may have different requirements for the prototypes of a part design. As
stated in Section 5.3, if only functional properties (e.g. maximum stress and deflection)
are the main concerns for the prototypes, the designer can initiate a problem formulation
based on the material geometric tailoring decision template (MGTDT), and transfer it to
the manufacturer. Since the process variables are not needed in the formulation, the
problem can be solved quickly and easily (refer to the example given in Section 5.5).
However, if the requirements of surface finish, tolerances, cost and time are also
considered for the prototypes, the material-process geometric tailoring decision template
(MPGTDT) should be used instead (Figure 6.6). Since more goals are added into the
formulation, the tradeoffs have to be made between the functional properties and the
other requirements, which are related to the process planning of RT. It is more difficult
Designer's Requirements for Prototypes
{
Stress
Functional Strain
}
MGTDT Properties Load
(§5.3) etc.
Surface Finish
Flatness tolerance MPGTDT
Parallelism tolerance
Tolerances Positional tolerance
Circularity tolerance
Concentricity tolerance
Perpendicularity tolerance
Cost and Time
Figure 6.6 – Relations of Designer’s Requirements and MGTDT/MPGTDT.
222
because of the coupling between sub-problems (Section 6.4).

The word and mathematical formulations of the MPGTDT are presented in Section
6.3.1. Based on the MPGT formulation given by the designer and the process planning
formulations given in the last section, an integrated MPGT problem formulation is
described in Section 6.3.2.
6.3.1 MPGTDT
As stated in Section 5.3.1, the definition of Material-Process Geometric Tailoring
Decision Template is:
• MPGTDT – Material-Process Geometric Tailoring Decision Template, a
compromise decision in which the component’s dimensions are modified to suit a
prototype material and fabrication process.
The MPGTDT has the same principle as that of the MGTDT (Section 5.3.1). They
also have similar formulations, except more goals and constraints are added in the
MPGTDT. The compromise DSP word formulation of the MPGTDT is presented in
Table 6.7, in which tolerances and surface finish are called process goals.
Table 6.7– Word Formulation of the MPGTDT.

GIVEN:
• Parametric CAD model of part • Functional property models
• Target mold life • Material Properties
• Target values for functional properties • Prototype material properties
• Target values for process goals • Target cost and time
FIND:
SATISFY:
Goals: Constraints:
Meet targets of geometry variables
Meet targets of process goals Bounds:
Meet targets of mold life Bounds for all system variables
Meet target cost and time
MINIMIZE:
Correspondingly to the word formulation, a general mathematical formulation of

MPGTDT is presented in Table 6.8.
223
Table 6.8 - Mathematical Formulation of the MPGTDT.

GIVEN:
• nf - number of functions;
ng - number of geometry variables; nm - number of material variables
nsf - number of surface finish variables; ntol - number of tolerance variables
p – equality constraints; q – inequality constraints.
• Gp, i i = 1, …, ng
• Mp, i i = 1, …, nm
• Fp, i = fi (Gp,j, Mp,n) i = 1, …, nf; j = 1, …, ng; n = 1, …, nm
• Mm, i** i = 1, …, nm
• Fm, i = fi (Gm,j, Mm,n)** i = 1, …, nf; j = 1, …, ng; n = 1, …, nm
• SFi,T , SFi** i = 1, …, nsf
• Toli,T , Toli** i = 1, …, ntol
• MLT, ML**
• Time: [Tmin , Tmax], T** • Cost: [Cmin , Cmax], C**
• Wi i = 1, …, nf+ng+nsf+ntol+2
FIND:
System Variables:
Gm, i i = 1, …, ng
Deviation Variables:
di+, di- i = 1, …, nf+ng+nsf+ntol+2
SATISFY:
Goals:
Fp, i / Fm,i - di+ + di- = 1 i = 1, …, nf [6.1]
Gp,i / Gm,i - + dnf+i- = 1
dnf+i+ i = 1, …, ng [6.2]
SFi,T / SFi - dnf+ng+i+ + dnf+ng+i- = 1 i = 1, …, nsf [6.3]
Toli,T / Toli - dnf+ng+nsf+i+ + dnf+ng+nsf+i- =1 i = 1, …, ntol [6.4]
(Tmax–T)/(Tmax–Tmin)- di+ + di- = 1 i = nf+ng+nsf+ntol+1 [6.5]
(Cmax–C)/(Cmax–Cmin)- di+ + di- = 1 i = nf+ng+nsf+ntol+2 [6.6]
Constraints:
di+ • di- = 0, di+ ≥ 0, di- ≥ 0 i = 1, …, nf+ng+nsf+ntol+2
gj(Gi) = 0, gk(Gi) ≤ 0 i = 1, …, ng; j=1, …, p; k=1, …, q

Bounds:
Gimin ≤ Gi ≤ Gimax i = 1, …, ng
Fimin ≤ Fi ≤ Fimax i = 1, …, nf
SFimin ≤ SFi ≤ SFimax i = 1, …, nsf
224
Tolimin ≤ Toli ≤ Tolimax i = 1, …, ntol
Tmin ≤ T ≤ Tmax Cmin ≤ C ≤ Cmax

MINIMIZE:
Deviation Function: Wi • (di+ + di-) where ΣWi = 1, Wi ≥ 0 (i= 1, …, nf+ng+nsf+ntol+2)
Note:
1. Subscripts m and p denote the prototype model and the product respectively.
2. Symbol ‘**’ denotes the entries that are to be completed by the manufacturer.
Compared with the mathematical formulation of the MGTDT given in Table 5.4,
more variables and goals (Equations 6.1 ~ 6.6) are considered in the MPGTDT besides
geometry variables (Gj), material properties (Mn), and functional properties (Fi). These
variables and goals include surface finish (SFj), tolerance (Toli), mold life (ML), time (T)
and cost (C). As stated before, they are tightly related to the process planning. Therefore
more entries are indicated by ‘**’, which denotes information that the manufacturer must
supply in order to complete the problem formulation and generate a solution.
MPGTDT RP Process Planning IJM Process Planning

Given Given Given
Fh = f (Gj, MPg) ML = f (LT, TA) Cp = f (CT)
AC = f (PO, LTi, HOCi, FOCi) Tp = f (CT)
SF = f (PO, LTi)
Nm = f (ML)
Analysis up
lin
g
co
BT = f (PO, LTi, HOCi, FOCi,upNli m)
ng
co lin
g Cm = f (BT, TA)
up Tm = f (BT, TA)
co
System Variables System Variables System Variables

Gi, θj PO, LTi, HOCi, FOCi, TA CT
Goals & Constraints Goals & Constraints Goals & Constraints
Fh AC, SF, Cm, Tm coupling Cp , T p
MPGT Problem
Given
Fh = f (Gj, MPg)
AC = f (PO, LTi, HOCi, FOCi)
SF = f (PO, LTi, Gj, θk)
ML = f (LT, TA, Gj, θk, CT)
Nm = f (ML)
BT = f (PO, LTi, HOCi, FOCi, Nm)
Cm = f (BT, TA)
Tm = f (BT, TA)
Synthesis Cp = f (CT)
Tp = f (CT)
C = Cm + Cp
T = Tm + Tp
System Variables
Gj, θk, PO, LTi, HOCi, FOCi, TA, CT
Goals & Constraints
AC, SF, C, T, Fh
Figure 6.7 – Relations of the MPGTDT and MPGT Problem.
225
Based on the design information, the designer can instantiate the MPGTDT to create
the problem formulation suitable for communicating to the manufacturer. The usage
scenario of the MPGTDT can be the same as the one presented in Section 5.3, therefore
will not be repeated here. The problem formulation given by the designer and the
formulations of the process planning can be integrated into a MPGT problem formulation
(Figure 6.7), which is to be discussed in the next section.
6.3.2 MPGT Problem Formulation

In the MPGTDT problem, the requirements on production-representative functions,
accuracy, surface finish, and fabrication cost and time are considered. By integrating the
problem formulations of the MPGTDT (Table 6.7), RP process planning (Table 6.1) and
IJM process planning (Table 6.3), one can obtain a Material Process Geometric Tailoring
(MPGT) problem formulation. The word formulation of the MPGT problem is shown in
Table 6.9.
Table 6.9 – MPGT Problem Word Formulation.

GIVEN:
• Parametric CAD model of part • Functional property models
• Target values for process goals • RP process goals models
• Target mold life • Mold life models
• Target values for functional properties • RP process models
• Target material properties • Prototype material properties
• Target cost and time • Goal preferences as weights
FIND:
RP process variables
IJM process variables
SATISFY:
Goals: Constraints:
Meet targets of geometry variables Meet RP process constraints
Meet targets of process goals Meet IJM process constraints
Meet target mold life Bounds:
Meet target cost and time Bounds for all system variables
MINIMIZE:
In the MPGT formulation, the system variables include geometry variables, RP

process variables and IJM process variables. The goals include meeting the target
functional properties, geometry variables, mold life, cost/time, and process goals. The
process goals are the RP process goals. Experiments performed by (Cedorge, 1999)
226
showed that the surface finish of the part surfaces is very close to the surface finish of the
mold surfaces. Therefore the accuracy and surface finish of the part depends on the
accuracy and surface finish of the mold, and hence these goals are considered for the
mold instead of the part. The mathematical models of accuracy and surface finish for RP
process can be used here directly. The material properties of RP process are not
considered in the formulation because they are related to the molds instead of the parts.
The constraints include meeting the geometry constraints, RP process constraints and IJM
process constraints.
Related to the above word formulation, a MPGT problem mathematical formulation
is presented in Table 6.10 by integrating the math formulations given for the design
process (Table 6.8), RP process (Table 6.2) and IJM process (Table 6.5). One thing to be
noticed in Table 6.10 is that the modified cDSP formulation proposed in (Hernandez and
Mistree, 2001) is used (Section 2.5). Hernandez and coauthors modified the objective
function (and correspondingly the goals) in cDSP formulation according to the Linear
Physical Programming (LPP) formulation developed by Messac, et al. (1996). In LPP,
all the goals are classified into class 1S, 2S, 3S and 4S. Class 1S corresponds to
minimization goals, class 2S corresponds to maximization goals, class 3S corresponds to
target matching goals and class 4S corresponds to range matching goals. Range matching
goals have range of values as target instead of a single point. Hernandez and coauthors
proposed to split all the goals of class 3S and class 4S into two independent goals of class
1S and class 2S. Therefore the designer can express preferences of each goal through
various degrees of desirability: unacceptable, highly undesirable, undesirable, tolerable,
desirable, and ideal.
Table 6.10 – MPGT Problem Mathematical Formulation.

GIVEN:
• Parametric CAD model of part, GTi, θTs • Function F h = fh (Gj, MPg)
• Accuracy AC = f (PO, LT, HOC, FOC, ZL, SP) • Material properties MPg
• Surface finish SF = f (Gi, θs, PO, LT) • Mold life ML = f (Gi, θs, LT, TC, CT)
• Build time BT = f (Gi, PO, LT, HOC, FOC, ZL, SP) • ACTq, SFTr, FTh, NP, CT, TT, MPTg
• Total time T = f (BT, CT, TC, ML, NP) • Cost C = f (BT, CT, TC, ML, NP)
FIND:
Gi, θS d j+, p , d j−, p
POkm, LTkmn, HOCkm, FOCkm, ZLkml, SPkml
TCk, CTk
SATISFY:
Goals: Constraints:
Fh − max( Fh − th+, p +1,0)
+
+ d h−, p − d h+, p = 1 [6.7] h (Gi) = 0
t h, p
Fh + min( Fh − th−, p +1,0)

+ d h−, p − d h+, p = 1 [6.8] t h−,5 ≤ Fh ≤ th+,5
t h−, p
227
Gi − max(Gi − ti+, p+1,0)

+
+ di−, p − di+, p = 1 [6.9] Large horizontal planes
ti , p
Gi + min(Gi − ti−, p+1,0)
−
+ di−, p − di+, p = 1 [6.10] Support structures
ti , p
θ s − max(θ s − ts+, p +1,0)
+ d s−, p − d s+, p = 1 [6.11] ACq ≤ tq+,5 , SFr ≤ tr+,5
ts+, p
θ s + min(θ s − ts−, p +1,0)
+ d s−, p − d s+, p = 1 [6.12] C ≤ t4+, 5 , T ≤ t5+, 5
ts−, p
ACq − max( ACq − tq+, p+1,0)
+ dq−, p − dq+, p = 1 [6.13] d +j , p • d −j , p = 0
tq+, p
SFr − max( SFr − tr+, p+1 ,0)
+ dr−, p − dr+, p = 1 [6.14] d j+, p , d j−, p ≥ 0
tr+, p
C − max( C − t1+, p+1 ,0) Bounds:
+ d1−, p − d1+, p = 1 [6.15]
t1+, p
T − max(T − t2+, p+1 ,0)
+
+ d2−, p − d2+, p = 1 [6.16] Bounds for all system variables
t 2, p
MINIMIZE:
4 2( h + i + o )+ q + r + 2
Deviation Function: ∑
p =1
∑w
j =1
j, p dd +
j, p + d −j , p i
Where,
g – index for material property variables
h – index for function variables
i – index for geometry dimension variables
j – running index for all deviation variables
k – index for the number of molds
l – index for the number of blocks of different Z-level wait (ZL) and Sweep
period (SW) in a mold piece
m – index for the number of mold pieces in each mold design
n – index for the number of blocks of different layer thicknesses in a mold piece
p – 1,…4; used for goal formulation in LPP (linear physical programming)
q – index for the number of accuracy requirements
r – index for the number of surface finish requirements
s – index for draft angle variables
The system variables in the MPGT problem (Table 6.10) include the system
variables from the MPGTDT (geometry variables: part dimensions Gi and draft angle θs),
RT process planning problem (RP process variables: part orientation PO, layer thickness
228
LT, hatch overcure HOC, fill overcure FOC, z-level wait ZL, sweep period SP) and IJM
process planning problem (IJM process variable: thermal cure TC and cooling time CT).
If single mold assembly cannot produce the desired number of parts (NP), multiple
mold sets should be built. It is possible to build different mold sets with different values
of geometry variables. But, in this formulation, it is assumed that all the molds are
fabricated with same values of geometry variables to ensure that all the prototypes have
identical properties.
There is a total of 2H+2I+2S+Q+R+2 goals in Figure 7.6 that include ‘2H’ function
goals (Equations 6.7 and 6.8), ‘2I’ part dimension goals (Equations 6.9 and 6.10), ‘2S’
draft angle goals (Equations 6.11 and 6.12), ‘Q’ accuracy goals (Equation 6.13), ‘R’
surface finish goals (Equations 6.14) and one of each of time and cost goals (Equations
6.15 and 6.16). The goals of part dimension, draft angle, and function are target
matching goals and in LPP formulation of cDSP, each of these goals is divided into two
independent goals of class 1S (minimization) and class 2S (maximization). The goals of
accuracy, surface, cost and time are minimization goals (class 1S).
Function goals are affected by only geometry and material variables. Accuracy goal
is affected by only the RP variables. Surface finish goal is affected by geometry
variables (e.g. draft angle) and RP variables (e.g. part orientation). Cost and time are
affected by all the system variables. The constraints in MPGT for RT problem include
geometry and assembly constraints affected by geometry variables, and RP constraints.
The other constraints arise due to the LPP formulation of cDSP problem. Objective
function is formulated as a weighted sum of goal deviations (archimedean formulation).
For each deviation, weight wj,p can be determined from target values of the related goal.
For example, the target values of cost goal can be $98, $125, $150, $170, and $180
corresponding to the target levels of ideal, desirable, tolerable, undesirable, and
unacceptable. An algorithm that can be used for the calculation was presented in
(Hernandez and Mistree, 2001).
After the MPGT problem formulation is given, a solution approach for the DFRTS is
presented in the next section.
6.4 SOLVING THE MPGT PROBLEM
After the MPGT problem was formulated in the last section, a solution strategy and a
related solution process are presented in Section 6.4.1 and 6.4.2 for the problem
respectively.
6.4.1 Solution strategy

The formulating and solution approaches for the MGT problem (Table 5.4) are
presented in Section 5.4.2. To get the quantitative relationship between goals and system
variables, analytical equations or the response surface equations can be employed in the
formulation (Section 5.4.1). Software systems such as DSIDES or OptdesX can be used
to solve the completed problem formulation. However the approaches cannot be directly
used for the MPGT problem.
229
Different from the MGT problem, the MPGT problem (Table 6.10) has geometric
and discrete variables besides continuous variables. This is because the MPGT problem
has the requirements that are tightly related to the process planning (e.g. surface finish
and build time), hence related process variables are added in the formulation. Some
process variables may be discrete or related to geometry. In the research scope of the
DFRTS, part orientation (PO) and layer thickness (LT) are two discrete variables. They
are also two important process variables in the RP process planning. For a part as given
in Figure 6.8.a, some possible part orientations are shown in Figure 6.8.b. From the
(a) An Example Part
Orientation 1
Orientation 2 Orientation 3 Orientation 4
Orientation 5 Orientation 6
(b) Different Part Orientations (PO) in SLA Building Process
Slicing scheme 1 Slicing scheme 2 Slicing scheme 3 Slicing scheme 4
(c) Different Slicing Schemes (LTi) for a Part Orientation (PO)
Figure 6.8 – Possible Values of PO and LTi for a Example Part (Sambu, 2001).
230
figure, it is evident that the effects of PO on accuracy (AC) or surface finish (SF) of faces
are not continuous. Small changes of PO can cause large changes of AC and SF. Two
constraints, large horizontal planes and support structures, are also related to the part
orientations (West, 1999). For one of the part orientations, different slicing schemes are
shown in Figure 6.8.c. For SLA process the values of layer thickness (LT) can only be
0.002, 0.004 and 0.008 inch.
Besides the discrete variables, the number of variables LTi in the MPGT is not fixed.
This is because the adaptive slicing is used in the RP process planning. Compared to
uniform slicing process, which uniformly slices the CAD model into a finite number of
slices at a certain thickness, the adaptive slicing can reduce the stair-step effect and
improve the surface quality, without greatly sacrificing the amount of fabrication time.
However because different layer thickness is used, the number of LTi depends on the
local surface geometry. For different part orientations, there may be different layer
thickness.
Based on the above considerations, a solution strategy is developed for the MPGT
problem as shown in Figure 6.9. Suppose in a problem P, X is discrete variables and Y is
continuous variables. The solution strategy divides the problem P into three sub-
problems P1, P2, and P3. In the first sub-problem P1, a set of candidate values of the
discrete variables (X) are determined based on the default values of Y. For each candidate
value of X, satisficing values of the continuous variables (Y) are determined in the second
sub-problem P2. Finally a solution is chosen among all the solutions given by P1 and P2
in the third sub-problem P3. The solution is then returned as the solution of problem P.
Problem P
Find
X, Y
Minimize
Z=f(X, Y)
P1 P2 P3
Given Given Given
Y' X' Xi', Yi'
Find Find Select
X Y X, Y
Minimize Minimize Minimize
Z=f(X, Y') Z=f(X', Y) Z=f(Xi', Yi')
Figure 6.9 – The Solution Strategy for the MPGT Problem.

In many cases the solution given by the solution strategy is not optimal for problem
P. However, it is a satisficing solution which is evident from the formulations of P1, P2,
and P3. When several sub-problems are integrated into a problem P, the problem may
have several variables, goals, and constraints (refer to Table 6.10). Therefore it will be
rather difficult, or even infeasible, to find the optimal solution of problem P. Instead an
231
approach to find a satisficing solution may be more appropriate. The case studies of a
robot arm and a camera roller (Chapter 7 and 8) will further illustrate it.
Considering the discrete variables PO and LT in Table 6.10, the MPGT problem is
divided into modified RP process planning problem (A) and modified MPGT problem
(B) (Figure 6.10). Modified RP process planning is a sub-set of RP process planner and
has only part orientation (PO) and slicing scheme (LTi) as system variables. The purpose
of modified RT process planning problem is to obtain a set of candidate orientations and
MPGT Problem
Given
Fh = f (Gj, MPg)
SF = f (PO, LT i, G j, θk)
Nm = f (ML)
BT = f (PO, LT i, HOCi, FOCi, Nm)
Cm = f (BT, TA)
Tm = f (BT, TA)
Cp = f (CT)
Tp = f (CT)
C = C m + Cp
T = T m + Tp
System Variables
Goals & Constraints
AC, SF, C, T, Fh
Large horizontal planes
Support structures
A. Modified RP Process Planning B. Modified MPGT Problem

Given Given
AC = f (PO, LTi) PO=PO', LTi=LTi' y
SF = f (PO, LTi) Fh = f (Gj, MPg)
x2
BT = f (PO, LTi) AC = f (PO, LTi, HOCi, FOCi) x1
System Variables ML = f (LT, TA, Gj, CT, θk)
PO, LTi
Goals & Constraints Nm = f (ML)
AC, SF, BT BT = f (PO, LTi, HOCi, FOCi, Nm)
Large horizontal planes Cm = f (BT, TA)
Support structures Tm = f (BT, TA)
Cp = f (CT)
Tp = f (CT)
C = Cm + Cp
T = Tm + T p
System Variables One
Gj, θk, HOCi, FOCi, TA, CT
Solution
Goals & Constraints
AC, SF, C, T, Fh
Figure 6.10 – Two Sub-problems of the MPGT.
232
a set of promising slicing schemes for all the mold pieces. The goals considered in the
problem are surface finish, accuracy and build time. The default values of other RP
process variables (e.g. hatch overcure HOC and fill overcure FOC) are used in computing
the values of the goals. The constraints of large horizontal planes and support structures
are also considered in determining part orientations. The results of modified RP process
planning include a set of slicing schemes for a set of part orientations of each mold piece
(Sambu, 2001).
The solutions of PO and LTi are then used to solve modified MPGT problem. The
only difference between MPGT problem and modified MPGT problem is that mold
orientation and slicing scheme are fixed in modified MPGT problem. Therefore the
problem is simplified because of less system variables and constraints. By generating
equations of goals and the continuous variables, one can get the solution of the
continuous variables related to the given PO’ and LTi’.
Finally a selection is performed to determine the best of the obtained solutions. This
solution is accepted as the solution for the MPGT problem.
From the infrastructure of the DFRTS (Figure 6.2), one can see the MPGT problem
is actually formulated for one mold piece generated by the RTMDS. Because the mold
design variables (parting direction, parting line, parting surface) in the RTMDS are also
variables that are related to part geometries (Chapter 3 and 4), they can be treated in the
same way as PO and LTi. The solution processes of the DFRTS and related
implementations are presented in the next section.
6.4.2 Solution Process and Implementations

Related to the solution strategy presented in Figure 6.9, the solution process of the
DFRTS can be divided into three phases, modeling design functions and fabrication
processes, solving modified MPGTs, and selecting among different solutions (Figure
6.11). They are described in more details as follows.
• Phase I: Modeling Design Functions and Fabrication Processes.
To make tradeoffs between different goals, the quantitative relationships between the
goals and system variables should be generated first. In the MPGT problem (Table 6.10),
equations are needed for function (Fh), accuracy (AC), surface finish (SF), mold life
(ML), build time (BT), total time (T) and cost (C). These equations can be divided into
two categories, models of design functions and models of fabrication processes.
- Models of Design Functions.
Design function Fh in the MPGT problem can be any functional properties that the
designer is interested in. For example, it can be maximum stress (refer to Section 5.5.3),
maximum deflection (refer to Section 5.5.2), or load (refer to Section 5.5.1). The
approaches to formulate equations for the MPGT problem are the same as those for the
MGT problem (Section 5.4.1). In the case studies to be presented in Chapter 7 and 8, an
approach similar to RCEM (Chen, 1995) was used to formulate the design functions of
stress, deflection, and rotation. The approach is shown in Figure 6.11. It has five main
steps, which are (1) identify factors and ranges of system variables; (2) design of
233
Phase I: Modeling design Phase III:

Phase II: Solving modified MPGTs
functions and fabrication Selecting
processes among
Part Design
Design different
Freedom solutions
A. Rapid Tooling
1.1 Identify Mold Design Solution of
factors and System MPGT
ranges A set of mold
designs
1.4
Analyze 1.2 Design of
B. Refined RP
...
Experiment Experiments
Process Planner Selection
Results
(other
A set of PO
1.3 Physical solutions)
Experiment /
RSEs of
and LT ...
Simulation
Fh, AC, SF Solutions
Programs y
Solution for a for different
x2
D1. OptdesX mold design mold design
x1
and PO, LT and PO, LT
1.5 Build
Response System Objective funciton
Surface Models variables and constraints
y
C. Mold Life D2. C Data Variables E. RT Cost
x2 Variables
x1 Predictor Module Estimator
Mold life Cost and
y Time
x2
x1
RSE of ML
Figure 6.11 – The Solution Process of the DFRTS.

experiments; (3) execute simulations (ANSYS); (4) analyze experiment results; and (5)
build response surface models.
It is obvious that models of design functions are related to design requirements.
Therefore the above steps should be executed for different parts and design requirements.
- Models of Fabrication Processes.
Models of process properties are important in the process planning. The process
properties in the scope of DFRTS are accuracy, surface finish, mold life, build time of
SLA, time and cost of RT. One thing to be noticed is that these models are needed for
different fabrication processes; however, for different parts using the same fabrication
process, the models can be reused.
A significant amount of work is done at Georgia Tech. in obtaining quantitative
models for accuracy, surface finish and build time on SLA-250 (SOMOS 7110) and
SLA-3500 (SL 7510) (McClurkin, 1997; Lynn, 1998; West, 1999; Sambu, 2001). They
are introduced briefly to foster a better understanding of the capabilities and limitations
of the DFRTS.
Accuracy AC = f (PO, LT, HOC, FOC, ZL, SP):
234
Lynn (1998) performed a number of experiments on SLA 250 – SOMOS 7110 and
developed mathematical models to predict accuracy of SLA prototypes. The first set of
experiments performed by Lynn was screening experiments. From the screening
experiment results, Lynn identified hatch overcure, fill overcure, sweep period, z-level
wait, layer thickness and part orientation to be the RP variables that have significant
effect on part accuracy. A Face Centered Composite design of experiment is used to
determine the next set of experiments (main experiments) used to generate response
surface models. Different types of accuracies are studied, which include flatness,
parallelism, perpendicularity, concentricity, circularity, and positional tolerance. By
using the similar experimental methodology, (Davis, 2001) generated quantitative models
for accuracy of the parts built on SLA 3500 – SL 7510.
Surface finish SF = f (Gi, θs, PO, LT):
West (1999) performed surface finish experiments on SLA 250 – SOMOS 7110 and
developed quantitative models to predict surface finish. The RP variables that affect
surface finish are part orientation and layer thickness. Experiments were run for three
different layer thickness (2, 4 and 8 mils) for a test piece as shown in Figure 6.12. The
test piece consists of ten squares each rotated in an increment of 5° across the length of
the piece. This geometry allows a surface finish measurement to be taken from each of
the rectangular planar surfaces. In performing the surface finish experiments, effect of
support structure was eliminated by taking measurements from the region of the surface
that is not affected by supports. Additional experiments were performed to predict
surface penalty due to support structures. By using the similar experimental method,
Sambu (2001) generated quantitative models for surface finish of the parts built on SLA
3500 – SL 7510.
Figure 6.12 – Surface Finish Test Piece.

Build time BT = f (Gi, PO, LT, HOC, FOC, ZL, SP):
McClurkin (1997) developed a Build Time Estimator (BTE), which reads the vector
(.v) and range (.r) files created by Maestro, 3D System’s software for stereolithography,
and calculates the build time of prototypes with errors of roughly 2%. Since these files
are still not available in the SLA process planning, West (1999) developed quantitative
models based upon empirical data collected from the BTE. In West’s experiments, the
235
build time was calculated from hatch time, fill time, border time and recoat time. Layer
thickness, cross-section area and perimeter of the layer were considered as the variables
that affect build time. Based on West’s work, Sambu (2001) performed sets of build time
experiments on SLA 250 and SLA 3500, and generated better build time models that are
applicable over larger range of RP variables.
Total time T = f (BT, CT, TC, ML, NP) and Cost C = f (BT, CT, TC, ML, NP):
Sambu (2001) developed quantitative models of cost and time based on the analysis
of the whole Rapid Tooling process. The whole direct AIM tooling process is divided
into ten steps as shown in Table 6.6. The estimated values of time and cost /hr for each
step are also presented in the table. They were determined based on the practice at our
lab and literature survey.
Mold life ML = f (Gi, θs, LT, TC, CT):
A significant amount of work is done at Georgia Tech. in obtaining quantitative
models for stereolithography mold life on direct AIM tooling (Cedorge, 1999; Le Baut,
1999; Palmer, 1999; Crawford, 2001; Pham, 2001; Rodet, 2001). Different types of
failures and factors that affect mold life were identified (Section 6.2.2). Sambu (2001)
developed a software system based on the quantitative models from these work to predict
ejection force and stereolithography mold life.
• Phase II: Solving Modified MPGTs.
For a part design, the software modules in this phase include Rapid Tooling mold
design system, refined RP process planning, mold life predictor, optimization software
(OptdesX), a C data module, and Rapid Tooling cost predictor. For a part design, a set of
mold designs is generated by the Rapid Tooling mold design system. For each mold
design, a set of part orientations and slicing schemes is generated by the refined RP
process planner. For each of them, a modified MPGT problem can be formulated by
considering the equations of goals and system variables. It is then solve by OpdesX with
the aid of the C data module, mold life predictor, and RT cost predictor. The software
modules in phase II are described in more detail as follows.
A. Rapid Tooling Mold Design System.
The RTMDS generates a set of mold designs for the CAD model of a part. The
CAD models of mold pieces are generated by the system (refer to Chapter 4).
B. Refined RP Process Planner.
Refined RP-PP software generates a set of promising process plans for each mold
piece for each mold design. The process plans obtained are only partial, as they do not
have information about RP process variables (HOC and FOC). Along with the process
plans, an information file is also generated by RP-PP software. This file contains the
information regarding the scan and recoat times for each of the mold pieces. This
information is used in the C module to calculate the build time for the required quantities
of all the mold halves (Sambu, 2001).
236
C. Mold Life Predictor.

According to given process variables, the mold life prediction software module can
predict the shot number of stereolithography molds before they fail (Section 6.2.2). The
module was implemented in C and can be called by the C data module (D2).
D1. OptdesX.
OptdesX (www.et.byu.edu/~optdes/) is an engineering optimization software system
which uses Simulated Annealing (SAN) algorithm. Since it was used in the case studies
in Chapter 7 and 8, a detail description of the algorithm is given as follows in order to
foster a better understanding of the results given in the case studies.
SAN is a heuristic based optimization algorithm that does not use gradients in
finding the optimum solution. SAN models a phenomenon in nature – the annealing of
solids – to optimize a complex system. From a given starting point, the algorithm makes
small random changes in the design variables in the problem, causing a change in the
value of objective function. If the change is negative (i.e., if objective function is
reduced), then the current solution is better and it is accepted as the starting point for the
next iteration. If the change is positive (i.e., if the objective function is increased), the
new solution is worse; however, it may still be accepted according to the Boltzmann
probability factor presented in equation 6.1.
P = exp
FG −∆E IJ
H kTK
b
[6.17]
Where,
∆E is the change in objective function value,
kb is Boltzmann constant, and
T is the temperature (corresponds to actual annealing process).
This equation is used in annealing process. The Boltzmann probability ‘P’ is
compared to a random number between 0 and 1 drawn from a uniform distribution; if it is
higher, the solution is accepted. Accepting worse solutions in the earlier phases of
problem execution, SAN escapes some local minima.
In the OptdesX implementation of SAN, temperatures T in equation 6.17 are not
specified. Instead, starting and final probability are specified. Apart from these values,
number of cycles and maximum perturbation values should also be specified. Number of
cycles is the number of SAN iterations performed (number of different solution points
investigated) before the SAN execution is stopped. Each cycle has different temperature
and hence the Boltzmann probability is calculated for each cycle. Maximum perturbation
is the maximum allowable variation in the value of a system variable (per iteration).
SAN is an effective algorithm in solving discrete and mixed problems but it can also
be used to solve continuous problems. As SAN is a heuristic based algorithm that uses
random number generation, it is possible to obtain different solutions for different
attempts of solving a problem with the same starting point.
237
D2. C Data Module.

As an analysis module for OptdesX, the C data module is used to calculate goals and
intermediate responses for OptdesX. The values of the variables are passed from
OptdesX to the C module and the values of goals, constraints and objective function are
calculated in C module and passed to OptdesX (Figure 6.11). Some equations are
embedded in the C program of the module. The C data module can also call the mold life
predictor and the RT cost estimator for the values of mold life and cost/time respectively.
E. RT Cost Estimator.
According to given process variables, the rapid tooling cost estimator can predict the
cost and time for the whole RT process (Section 6.2.3). The module was implemented in
C and can be called by the C data module (D2).
• Phase III: Selecting the Solution of the MPGT.
For each mold design, part orientation and slicing scheme, OptdesX gives a solution
with its deviation value Zi. After all the solutions are generated, the solution with least
value of Zi is selected as the solution for the MPGT problem.
The MPGT problem and the above solution processes are further illustrated in the
case studies in Chapter 7 and 8. In the next section, the author compares the solution
process of the DFRTS with the current usage of RT, and explains the advantages of the
DFRTS base on decision order and design freedom of variables.
6.5 COMPARISON OF THE CURRENT USAGE AND DFRTS
In the current usage of Rapid Tooling, sub-problems of part design, RP process
planning, and IJM process planning are formulated and solved sequentially (Figure 6.13).
That is, decisions on design variables are made first based on the design requirements.
Then mold variables are determined and mold pieces for the part are generated. For the
mold pieces, decisions on RP process variables are made based on the requirements for
RP process. Finally decisions on IJM process variables are made based on the remaining
design freedoms.
However, the sub-problems of using Rapid Tooling are coupled. The arrows in Figure
6.13 indicate the coupling between geometric tailoring, RP process planning, and IJM
process planning. In the geometric tailoring formulation, geometry variables Gi affect the
mold life of stereolithography molds (Section 6.2.2). They could also affect surface
finish if they affect surface orientation. θj is the draft angles for different features. Draft
angle affects surface finish of drafted surfaces (Section 6.4.2). In the RP process
planning problem, ML is mold life. As explained in Section 6.2.2, mold life of a SL mold
feature is affected by its height, width, draft angle (geometric variables), layer thickness
used to build the feature (RP process variables) and the cooling time (IM process
variable) used in the injection molding of the parts. Also the goals of ‘meeting targets of
time and cost’ are coupled because the specified time and cost are for the whole
fabricating process, which is the sum of those for each individual process.
238
Part
CAD
Mold build by RP
(Chp 3 &4)
Geometric Tailoring RP Process Planning IJM Process Planning

Given Given Given
ML = f (LT, TA) Cp = f (CT)
Fh = f (Gj, MPg)
SF = f (PO, LTi)
Nm = f (ML) c
g BT = f (PO, LTi, HOCi, FOCi,oNump)
plin lin
cou g Cm = f (BT, TA) g
plin
cou Tm = f (BT, TA)

Fh AC, SF, Cm, Tm coupling Cp, Tp
Decision Mold
Gi, θ j Variables PO, LTi, HOCi, FOCi, TA CT
Order:
Figure 6.13 – The Current Usage of RT and Related Decision Order of Variables.
The couplings between different sub-problems may cause the sequential solution
process unable to find solutions in some later stages. Therefore iterations are necessary
which may take a long time.
In the DFRTS, the sub-problems are synthesized into a MPGT problem, and the
couplings between the sub-problems are formulated in the synthesis problem formulation.
The coupled goals/responses are bolded in the problem formulation as shown in Figure
6.14. Two additional goals, total cost (C) and total time (T), are also included in the
formulation. These goals are not part of any of the sub-problem formulations but are
synthesized from mold cost/time (RP-PP) and part cost/time (IJM-PP).
For the MPGT problem, an approach of obtaining satisficing solution is developed
by decomposing the problem into two sub-problems: modified RP process planning
problem and modified MPGT problem (Section 6.4.1). The decomposition of MPGT
problem is also shown in Figure 6.14. By solving the two new sub-problems, one can get
the solution for the MPGT problem.
So in the design for Rapid Tooling process, what is the benefit to synthesize the sub-
problems into a combined problem, and then decompose the synthesis problem into sub-
problems again in order to solve it?
239
MPGTDT RP Process Planning IJM Process Planning

Given Given Given
Fh = f (Gj, MPg) ML = f (LT, TA) Cp = f (CT)
SF = f (PO, LTi)
Nm = f (ML)
Analysis up
lin
g
co
BT = f (PO, LTi, HOCi, FOCi,upNli m)
ng
co lin
g Cm = f (BT, TA)
up Tm = f (BT, TA)
co

Fh AC, SF, Cm, Tm coupling Cp, Tp
MPGT Problem
Given
Fh = f (Gj, MPg)
Nm = f (ML)
Cm = f (BT, TA)
Tm = f (BT, TA)
Synthesis Cp = f (CT)
Tp = f (CT)
C = Cm + Cp
T = Tm + Tp
System Variables
Goals & Constraints
AC, SF, C, T, Fh
Modified RP-PP Modified MPGT Problem

Given Given
AC = f (PO, LTi) Fh = f (Gj, MPg)
SF = f (PO, LTi) AC = f (PO, LTi, HOCi, FOCi)
Solving
Solution BT = f (PO, LTi)
ML = f (LT, TA, Gj, CT, θk)
Approach Nm = f (ML)
Cm = f (BT, TA)
Tm = f (BT, TA)
System Variables Cp = f (CT)
PO, LTi
Tp = f (CT)
Goals & Constraints C = Cm + Cp
AC, SF, BT T = Tm + Tp
System Variables
Gj, θk, HOCi, FOCi, TA, CT
Goals & Constraints
AC, SF, C, T, Fh
Decision Mold
Variables PO, LTi Gi, θ j ,HOCi, FOCi, TA, CT
Order:
Figure 6.14 – The DFRTS and Related Decision Order of Variables.
240
Comparing the sub-problems before and after the synthesis (Figure 6.14), one can
notice that the decisions on system variables are reordered. That is, although the decision
on mold variables, PO, and LTi is still sequential, the decisions on all other variables (Gi,
θi, HOCi, FOCi, TA, CT) are concurrent in the DFRTS. Concurrently formulating and
solving the variables enables us to explore more design spaces because the variables are
not fixed in the former stages. Consequently it is more likely to find a satisfying solution;
hence the iterations and lead-time in the current usage of RT can be reduced. Therefore
using a different decision order based on identified conflicting variables and goals is the
key for the above question.
Beside less iteration and time, other research also indicates that modeling
concurrency may result in better designs. Karandikar (1989) studied the design of a
cylindrical pressure vessel. He formulated the design and manufacturing requirements
related to the pressure vessel and solved the problems sequentially and concurrently. By
comparing the results, Karandikar concluded that “coupling design and manufacture and
solving for the design and manufacturing variables concurrently, as opposed to
sequentially, result in an increase in design freedom and possibly in the quality of the
resulting design.” Sobieszczanski-Sobieski (1989) also gave a good example to
graphically illustrate the argument that the sequential approach may lead to a suboptimal
design. The example is presented here to foster a better understanding of the advantages
of the DFRTS over the current usage of RT.
Suppose a two-variable design and manufacturing problem with X1 as design
variable and X2 as manufacturing variable. A certain performance measure P, expressed
as the single goal, has to be maximized. Suppose the designer has constraints C1 and C2.
The design space based on the design requirements can be sketched as shown in Figure
6.15.a. It is obvious that the optimal solution is at O1 for the designer.
(a) Design Requirements (b) Design and Manufacturing requirements

Figure 6.15 – A Two Variable Design Space (Karandikar and Mistree, 1991).
241
Assume the manufacture has an additional constraint C3 due to manufacturing

considerations. Therefore the design space by considering all the requirements can be
sketched as shown in Figure 6.15.b. Since constraint C3 makes the earlier solution O1
infeasible, the new solution for the problem is at O3. With the concurrent formulation,
one would get this solution. However using the sequential formulation and solution
approach, the value of design variable X1 is frozen in the design stage, thus only X2 can
be changed in the formulation, and O2 is obtained as the solution. It is seen that O2 is on
a lower P contour than O3 and, hence, represent a poorer design.
In the DFRTS, the decisions on continuous design and process variables are made
concurrently by the manufacturer. By modeling the concurrency, a better design (with
less cost) is obtained in the case study of a camera roller (Section 8.7). It can also be
explained by the analysis shown in Figure 6.15. Two case studies are presented in the
next two chapters to illustrate the usage of the DFRTS. In the next section, a brief
summary is given for discussing the hypotheses of the dissertation and the validation
strategy to be used in the case studies.
In the current usage of Rapid Tooling, the decisions are made in the same order as
the information flow. In each design stage, goals, constraints and preferences are
formulated to make the decision. But these decision factors are not transferred to the next
stages, so design freedom decreases quickly. Consequently the cost and time to make the
decisions in the later stages increase dramatically. To address the problem, the decisions
in different design stages may be reordered so that the decisions of some coupling
variables can be formulated and solved concurrently. Based on this idea, a design for
Rapid Tooling system (DFRTS) was presented for material-process geometric tailoring
(MPGT) in this chapter. First the infrastructure and scope of the DFRTS are introduced in
Section 6.1. The components of the DFRTS that are related to the process planning of
the AIM tooling are introduced in Section 6.2. The MPGT decision template for the
designer and the integrated MPGT problem formulation are presented in Section 6.3. A
solution strategy for the MPGT problem and a three-stage solution process for the
DFRTS are described in Section 6.4. Finally a comparison of the DFRTS and the current
usage of RT is given from the perspective of decision-making.
The research question and hypothesis that are related to this chapter are Q2.2, Q2.3
and sub-hypothesis 2.2, 2.3. They are repeated as follows.
Q2.2. How to formulate the design for Rapid Tooling problem which integrates
Q2.3. How to solve the design for Rapid Tooling problem effectively and
efficiently?
Sub-Hypothesis 2.2: The design for Rapid Tooling problem can be formulated by
several compromise DSPs and tasks, which can then be integrated into a
design for Rapid Tooling system (DFRTS).
242
Sub-Hypothesis 2.3: A three-stage solution process can be utilized to get a

satisficing solution effectively and efficiently based on design and
manufacturing models and continuous/discrete variables.
Based on the validation strategy presented in Section 1.3.2, the DFRTS and the
related hypotheses were validated in the following categories.
! Theoretical Structural Validity
The structural validity includes validating the ingredients, the components, and the
integrity. They are described in more details as follows.
- Validating the ingredients
Validating the ingredients includes validating specific modules of a cDSP
formulation (like response surfaces) and validating the functions in a software
implementation. Different aspects of validating the response surface models are
presented below.
a) Verifying that the R2, R2 (adj), maximum and average deviations of the response
surface models are reasonably small.
b) Verifying the behavior of the response surface models with respect to the design
variables and studying if the behavior is intuitively reasonable.
c) Verifying if the response surface is a good estimation of the actual design space.
Additional validation experiments should be performed to do this validation.
The response surfaces of the fabrication processes (Section 6.4.2) are problem-
independent. Sambu (2001) described their validations in more details. The response
surfaces of design functions are problem-dependent and are formulated and validated for
each problem separately (refer to Chapter 7 and 8).
Validating the functionality of a software implementation involves identifying the
potential problem-causing functions and testing their functionality for different sets of
inputs. The validation of RTMDS is discussed in Chapter 4. The validations of RP
process planner, mold lifer predictor, and RT cost estimator are also presented in (Sambu,
2001).
- Validating the components
Validating the components includes validating the formulation and implementation
of different cDSP’s and tasks. Different aspects of validating the cDSP formulations are
presented as follows.
a) Verifying that the quantitative relationships between all goals /constraints and the
system variables are available.
b) Verifying that the response surfaces used in the cDSP have reasonable fits.
Otherwise the cDSP problem should be solved again with the response surfaces
generated in a smaller design space that better approximates the actual design
space.
243
In the DFRTS, the cDSP’s are solved in OptdesX using Simulated Annealing
algorithm. The aspects of the validation of the implementation of cDSP’s are presented
below.
a) Verifying that the cDSP implementation satisfies the constraints on deviation
variables.
a. di+, di- ≥ 0
b. di+ • di- = 0
b) Verifying that the Linear Physical Programming (LPP) formulation of cDSP has
resulted in a solution with reasonable values of goal achievements. This is
achieved by observing the region of the goals in which the solution falls.
c) Verifying that the SAN algorithm is functioning correctly for the MPGT problem
for the chosen algorithm parameters.
d) Verifying the effect of starting point on the final solution of the problem.
e) Verifying the accuracy of the SAN algorithm by comparing its results to grid
search results.
The above verifications were tested for the two case studies and are presented in
Chapter 7 and 8. In the case studies, the result of the cDSP was also validated by
comparing the values of estimated (response surfaces) and actual (experiment) values of
goal achievements at the obtained solution.
- Validating the integrity
The integrity between different modules of the DFRTS is to test if the information
flow between the components is complete and correct. The integrity between different
sub-problems of MPGT problem is presented in 6.4.2. It is also validated through the
two case studies in Chapter 7 and 8.
! Empirical Structural Validity
Two case studies are chosen for the validation of the DFRTS. They are a robot arm
(Section 7.1) and a camera roller (Section 8.1). Both of them are industrial parts that are
produced by polymer injection molding. The functional requirements of the parts are the
maximum stress, deflection and rotation, which are all typical in the functional testing of
prototypes. Other manufacturing requirements (e.g. surface finish, tolerances) are also
required for the prototypes because of assembly requirements. The case study of the
robot arm is comparably simple, which makes the physical validation of each design
requirement feasible. The case study of the camera roller is rather complex. It is used to
test that the DFRTS can be utilized for complex cases. Therefore the author believes that
the case studies are appropriate to verify the performance of the system.
For each case study, the validations of ingredients, components and integrity as
stated earlier were tested and are presented in Chapter 7 and 8 respectively.
! Empirical Performance Validity
The solutions given by the RTMDS for each case study were checked with the
achieved values of design requirements. Physical prototypes were also produced for the
244
tailored part designs and the related process planning. For the case study of the robot
arm, further physical validations are performed on testing the given requirements (surface
finish, accuracy, material properties, weight, etc.) and performances of process planning
(mold life, build time, etc.). All of these tests validate the usefulness of the DFRTS in
producing functional prototypes for Rapid Tooling.
! Theoretical Performance Validity
The discussions on decision order and related design freedoms presented in Section
6.5 provide partial theoretical performance validation for Hypothesis 2.2 and 2.3.
The validations of hypothesis 2.2 and 2.3 also partially validate hypothesis 2, which
is repeated as follow:
In addition to the material geometric tailoring problem that was tested in hypothesis
2.1 (Chapter 5), the material-process geometric tailoring was tested in hypothesis 2.2 and
2.3 in this chapter. Therefore these two kinds of geometric tailoring problems for Rapid
Tooling were all tested for hypothesis 2.
With the DFRTS introduced in this chapter, a robot arm case study is presented in
the next chapter (Figure 6.16). In the case study, both the RTMDS (Chapter 4) and the
DFRTS (chapter 6) will be tested.
Chp 7: Prototypes of a
Robot Arm
Part Design
Parametric
Part Design
CAD Model
Requirements
of Part

Design System
P arting Directions
(Chp3 & 4) P arting Lines
(Section 6.3)
P arting Surfaces
M old Piec e Number
The Designer's MPGT
Problem Formulation
Part Design
Satisfy
Surface Finis h Draft Angle
Cost Part Thickness
Tim e Mold Life

Given Given
Mold Design M old Design
Find Find
RP Process Param eters IJM Proces s Parameters (Section 6.2.3)
Satisfy Satisfy
Goa ls Goals
Minimize Minimize
Cost Predictor
Cost

& 6.5) DSP Solver
Input and Output
Processor

245
Chapter 7 – Functional Prototypes of a Robot Arm
CHAPTER 7
FUNCTIONAL PROTOTYPES OF A ROBOT ARM
Chp 7: Prototypes of a Robot Arm
The RTMDS (Chapter 4) and DFRTS (Chapter 6) are applied to determine the mold
design and geometric tailoring for a robot arm design in this chapter. A problem of
producing functional prototypes of a robot arm is introduced in Section 7.1. By using the
RTMDS, two mold designs are generated for the robot arm (Section 7.2). A standard
mold base for Morgan Press injection molding machine is used in the mold piece
construction process. The formulation of the geometric tailoring problem for the robot
arm is presented in Section 7.3. Among the three design functions, two of them are
represented by response surface equations, and one is represented by an analytical
equation. The generation of the equations is also presented in Section 7.3. The solution
process of the MPGT problem, which consists of three stages, is described in Section 7.4.
The modified MPGT problems that are related to eight slicing schemes are solved by
OptdesX. The results of physical experiments for the validation of the problem solution
are presented in Section 7.5. Finally an evaluation of sequential and concurrent solution
processes for the robot arm case is provided in Section 7.6, and the validation of the
hypotheses is discussed in Section 7.7.
246
7.1 A ROBOT ARM DESIGN – PROBLEM DESCRIPTION

A challenging task in many robot designs is to design and control a robot arm
equipped with a hand for skillful manipulation tasks. Accurate position, velocity and
force are required in the running of the robot arm. Because they are related to so many
factors, besides the calculations and simulations, prototypes are usually required to test
the robot design. A case study of producing functional prototypes for a robot arm design
is presented in this chapter. In performing the case study, the author worked together
with Shiva Sambu, a M.S. student in our lab, who also studied the Rapid Prototyping
scenario for the robot arm case (Sambu, 2001).
t
D
(a) A Robot Arm and the Geometry Variables for Geometric Tailoring
Loads
Fixed
(b) Loading Conditions on the Robot Arm
Figure 7.1 – A Robot Arm Design.

Suppose a new robot arm design as shown in Figure 7.1.a is used in a robotic
mechanism. Several of them are needed in order to obtain desired degrees of freedom.
Usually, robot arms are made of steel (metal) to bear high loads with little deflection. For
low load applications (e.g. toy), plastic arms are used. The loading conditions on the
robot arm are shown in Figure 7.1.b. The larger cylindrical hole is completely fixed. A
5N tensile load and a 20N bending load are applied uniformly on the smaller cylindrical
surface. For the loading conditions, the designer determined that the production robot
arms are injection molded in atactic polystyrene material. Since the two cylindrical holes
in the robot arm are connected with other components, they are required to have good
surface finish (ideal – 20 µin; desirable – 40 µin; tolerable – 80 µin; undesirable – 130
µin; unacceptable – 200 µin). Also the top and bottom surfaces (flat surfaces of the robot
arm) have parallelism and flatness tolerance requirements according to assembly
requirements. Suppose the parallelism tolerance is (0.002, 0.004, 0.008, 0.013, 0.020)
inch and the flatness tolerance is (0.001, 0.002, 0.004, 0.007, 0.010) inch, which
correspond to (ideal, desirable, tolerable, undesirable, unacceptable) (refer to (Hernandez
and Mistree, 2001) for the goal formulation).
247
After finishing the design, the designer required fifty functional prototypes of the
robot arms fabricated in the production material for functional testing. Suppose in the
testing the designer is interested in the stress, displacement and weight of the prototype
parts under the loading conditions. Because the length of the robot arm affects the
orientation of the links in the robotic mechanism, its dimension cannot be modified. Also
the diameter of the holes affects the size of the bushings and hence cannot be changed.
After considering the dimensions of the robot arm, the designer determined that the
geometry variables of D, d and t have some design freedom. As shown in Figure 7.1.a,
‘D’ is the diameter of the link at the larger end, ‘d’ is the diameter of the link at the
smaller end, and ‘t’ is the thickness of the robot arm. Within the ranges given by the
designer, the manufacturer can modify them in the produced prototypes.
Finally considering the time and budget constraints, the designer wants to get the
prototypes within a week for the maximum cost of $1500.
7.2 MOLD DESIGN WITH AID OF RTMDS
The Rapid Tooling Mold Design System (RTMDS) was presented in Chapter 4. It is
based on the Multi-piece Mold Design Method (MPMDM) which was presented in
Chapter 3. The RTMDS is developed to reduce the design time in several important
mold design steps, including determining parting directions, parting lines, and parting
surfaces, and constructing mold pieces.
The size of the robot arm CAD file (.sat) is 103 KB. After loading it in the RTMDS,
the mold designer can generate a mold design with two mold pieces as shown in Figure
7.3.d within 18 seconds. The information regarding the part, generated regions, reverse
glue operation, and the execution time of each step is listed in Table 7.1. The reader can
refer to the descriptions given in Section 4.4.2 for the meaning of each items. The
running time given in the table is based on a personal computer with a 700 MHz Intel-III
processor. One thing to be noticed is that the running time of step 7 includes the time for
the user to interactively select the input file name of the mold base. Similarly the running
time of step 8 includes the time for the user to interactively select the output file names of
the mold pieces.
Table 7.1 – The Information for Robot Arm.

Face No. Concave face No. Concave edge No.
Part Info.
92 66 40
Initially generated After dividing After combining
Region Info.
26 26 1 region and 1 CXF
3 24 18+16
Reverse glue Info.
1 0 2
0.01 1.83 0.00
Running Time (s)
0.33 0.38 0.01
4.5* 10.1* 17.16*
Note: ‘*” denotes that the time of interactively selecting files is included in the measurement.
248
The graphical outputs given by the RTMDS are shown in Figure 7.2, Figure 7.3, and
Figure 7.4. Different regions and convex faces (CVX) are marked with different colors
in the figures. A brief description of the figures is given as follows.
(a) Generated Regions

R1
R2
F3
F2 F4
F1
(b) A Closer View of the Regions
(c) Two Combined Regions
Figure 7.2 – Generated Regions for the Robot Arm.

Based on the criterion of judging convex and concave edges (Section 3.4.3), there
are 40 concave edges in the robot arm. They are mainly related to the approximation of
the cylindrical holes. Therefore 66 concave faces are identified for the approximation.
Based on their connectivity, 26 regions are generated initially. These regions are shown
in Figure 7.2.a, with a closer view shown in Figure 7.2.b. In the figure, region R1 consists
of F1 and F2. Because F2 and F3 are in the same plane, they are convex with each other.
Therefore F3 and F4 will generate a new region R2 based on the region generation
249
algorithms (Section 3.4.3). They are combined into two regions during the region
combination process as shown in Figure 7.2.c.
CXF
R1
(a) Region Combination (1 region+ 1 CXF)
(b) A Mold Base for Morgan Press (c) Part with the Parting Surface
GFinner1 GFinner2 GFps
Figure 7.3 – Graphical Results of a Mold Design for the Robot Arm.
250
One region combination result is shown in Figure 7.3.a. The mold base used for the
robot arm is given in Figure 7.3.b. It is one of the standard mold bases for the Morgan
Press@ injection molding machine. The generated mold pieces for the region
combination result are shown in Figure 7.3.d. The generated glue faces (GFps, GFinner1,
GFinner2) are also shown in the figure.
R1
CXFs
(a) Region Combination (1 region + 25 CXFs)
GFinner2 GFps
GFinner1
Figure 7.4 – Graphical Results of Another Mold Design for the Robot Arm.
The user can change the combination results by setting different control options. For
the robot arm, if the option of not combining vertical faces is set, a different combination
result will be given as shown in Figure 7.4.a. Correspondingly, different mold pieces are
generated by the RTMDS (Figure 7.4.b) with the glue faces (GFps, GFinner1, GFinner2) for
the mold pieces.
The generation of CAD models of mold pieces is an important task in the mold
design process. After it is finished, other tasks can be started. In IronCAD system
251
(www.ironcad.com), the author added ejector-pin holes, gate and runner in the mold
pieces. A complete mold design for the robot arm is shown in Figure 7.5. The mold
design is further verified by physical validations. A photo of the built mold pieces and
prototype parts is given in Figure 7.6.
Ejector Pin Holes
Gate Runner
Figure 7.5 – A Complete Mold Design for the Robot Arm.
Figure 7.6 – Physical Validation of the Mold Design for the Robot Arm.
252
7.3 GEOMETRIC TAILORING WITH AID OF DFRTS – MODELING

The requirements given by the designer for the prototypes of the robot arm include
functional properties (stress, displacement, weight), surface finish, accuracy (parallelism
tolerance, flatness tolerance), cost and time. It is a material-process geometric tailoring
(MPGT) problem because of the process-related requirements. The MPGT decision
template (MPGTDT) can be used to formulate the design requirements, and the DFRTS
can be used to solve the MPGT problem for the robot arm.
The geometric tailoring of the robot arm will be presented in Section 7.3 and 7.4,
which corresponds to the modeling and solving stages in the solution process of the
DFRTS (Figure 6.10).
7.3.1 MPGT Decision Template for the Robot Arm

As stated in Section 7.1, the functional prototypes of the robot arm should have
similar performances of maximum stress, deflection and weight as those of the
production parts which are 5.99 MPa, 0.51mm and 3.40g respectively. There are also
other requirements such as surface finish, accuracy, time and cost, which are shown in
Figure 7.7. The design variables that can be tailored by the manufacturer are D, d, t.
Therefore based on the mathematical formulation of the MPGTDT (Table 6.8), the
designer may initiate a MPGT robot arm problem formulation for the above design
information (Table 7.2).
Surface Finish SF Cost ≤ $1500

Parallelism
tolerance Time≤ 7 days
PTOL
d
t
D
Flatness tolerance FTOL
Figure 7.7 – Process-related Goals of Robot Arm Design.
Table 7.2 – MPGT Robot Arm Problem Formulation by the Designer.

PROBLEM STATEMENT:
Design requirements given by the designer for producing functional prototypes of the robot arm
as shown in Figure 7.1.
GIVEN:
! Geometry variables that affect part functionality: D, d, t
! Tensile load = 5N, bending load = 20N
! Production parts injection molded in Atactic polystyrene with steel molds
• Young’s modulus YMp = 3200 Mpa
• Tensile strength TSp = 37.4 Mpa
• Density Denp = 1.04 g/cc
! Prototype parts **
• Young’s modulus YMm**
253
• Tensile strength TSm**

• Density Denm**
! Goals of prototypes under loads:
• Von-mises stress requirement (S),
• Weight requirement (W)
• Y-displacement requirement (YD)
• Geometry variables requirements (D, d, t)
! Goals of produced prototypes:
• Parallelism tolerance between top and bottom flat surfaces (PTOL)
• Flatness tolerance for top and bottom flat surfaces (FTOL1, FTOL2)
• Surface finish for the surfaces of the cylindrical holes (SF1, SF2)
• Total cost (C)
• Total time (T)
! Targets for the goals: (Ideal, desirable, tolerable, undesirable, unacceptable) (19 goals)
• S (1S) # 5.99, 6.29, 6.89, 7.79, 8.99 (MPa) [7.1]
• S (2S) # 5.99, 5.69, 5.09, 4.19, 3.00 (MPa)
• W (1S) # 3.40, 3.57, 3.91, 4.42, 5.10 (g)
• W (2S) # 3.40, 3.23, 2.89, 2.38, 1.70 (g)
• YD (1S)# 0.51, 0.54, 0.59, 0.66, 0.77 (mm)
• YD (2S)# 0.51, 0.48, 0.43, 0.36, 0.26 (mm)
• D (1S) # 20.32, 20.83, 21.84, 23.37, 25.40 (mm)
• D (2S) # 20.32, 19.81, 18.80, 17.27, 15.24 (mm)
• d (1S) # 10.16, 10.41, 10.92, 11.68, 12.70 (mm)
• d (2S) # 10.16, 9.91, 9.40, 8.64, 7.62 (mm)
• t (1S) # 3.048, 3.099, 3.200, 3.353, 3.557 (mm)
• t (2S) # 3.048, 2.996, 2.896, 2.743, 2.539 (mm)
• PTOL # 0.002, 0.004, 0.008, 0.013, 0.020 (in)
• FTOL1, FTOL2 # 0.001, 0.002, 0.004, 0.007, 0.010 (in)
• SF1, SF2 # 20, 40, 80, 130, 200 (µin)
• C # 150, 300, 600, 1000, 1500 ($)
• T # 3, 4, 5, 6, 7 (days)
FIND:
• D, d, t
! The requirements: S, YD, W, PTOL, FTOL1, FTOL2, SF1, SF2, C, T
• din + , din − i = 1,…,19, n = 1,…,4
SATISFY:
! Goals:
• C, T, PTOL, FTOL, SF are minimization goals (class 1S)
• S, W, YD, D, d and t are target-matching goals (class 3S) and each of these goals is split
into two independent goals of class 1S and class 2S.
• Class 1S goals formulation:
c
Aq ( x ) − max Aq ( x ) − tq,n+1 ,0 h+d−
q ,n − dq+,n = 1 q = 1,…,13, n=1,…,4
tq ,n
• Class 2s goals formulation:
254
c
Ar ( x ) + min Ar ( x ) − tr ,n+1,0 h+d −
r ,n − dr+,n = 1 r = 1,…,6, n=1,…,4
tr , n
Where Ap(x) is the pth goal and tp,n is the target for the pth goal in nth region (LPP).
! Requirement equations:
• Maximum stress S = f1(D, d, t)** [7.2]
• Maximum Y-displacement YD = f2(D, d, t)** [7.3]
• Volume V = f3(D, d, t)** [7.4]
• Weight W = Denm * V [7.5]
• PTOL, FTOL, SF, T, C**
! Constraints:
• 3.00 Mpa ≤ S ≤ 8.99 Mpa
• 1.72 g ≤ W ≤ 5.15g
• 0.34 mm ≤ YD ≤ 1.02 mm
• PTOL ≤ 0.020
• FTOLj ≤ 0.010 j = 1, 2
• SFj ≤ 200 j = 1, 2
• C ≤ 1500
• T≤7
+ −
• din • din = 0
• din + , din − ≥ 0
! Bounds:
• 15.24 mm ≤ D ≤ 25.40 mm
• 7.62 mm ≤ d ≤ 12.70 mm
• 2.539 mm ≤ t ≤ 3.557 mm
MINIMIZE:
c h
4 19
Z = ∑ ∑ wi,n+1 di+,n + d i−,n
n =1 i =1
The Problem formulation in Table 7.2 contains the specific information regarding the
goals and targets but does not contain the quantitative models of the response surface of
goals (Equations 7.2 ~ 7.5) in terms of system variables. In the problem formulation, the
targets for the goals given by the designer are formulated in Linear Physical
Programming formulation (Section 2.6.2). Each goals has five target values which are
corresponds to idea, desirable, tolerable, undesirable, and unacceptable. For example, in
Equation 7.1, S (1S) # 5.99, 6.29, 6.89, 7.79, 8.99 (MPa) where 1S means this is a
minimization goals, and the ideal value of S is less than 5.99 Mpa. It is desirable if S is in
range (5.99, 6.29] MPa; it is tolerable if S is in range (6.29, 6.89] MPa; it is undesirable if
S is in range (6.89, 7.79] MPa; it is highly undesirable if S is in range (7.79, 8.99] MPa;
and it is unacceptable if S is bigger than 8.99 MPa. The targets of other goals have the
same meanings. As discussed in Section 2.6.2, it is more straightforward for the designer
to assign target values for some goals than to assign weights for the goals.
After the MPGT problem is formulated, the designer can send it with the CAD
model of the part, and the FAE model for analyzing stress and deflection to the
255
manufacturer. After being produced, the prototypes of the robot arm will be sent back
from the manufacturer for the functional testing.
7.3.2 Modeling Design Functions

After receiving the design information, the manufacturer can begin the geometric
tailoring for the robot arm. First the design functions and the fabrication processes need
to be modeled. Suppose the direct AIM tooling (SLA 3500 – SL 7510 molds) for Atactic
polystyrene is chosen for the prototypes. In this section the models of design functions
are introduced. The models of the AIM tooling process for the robot arm are presented in
the next section.
The three design requirements for the robot arm are the maximum stress (S), the
maximum Y-displacement (YD), and volume (V). In this study, the equations of S and
YD are represented by response surface models, and the equation of V is represented by a
analytic equation. The steps of getting the equations are described as follows.
• Generating Maximum Stress and Y-Displacement Equations
As shown in Figure 6.10, the approach used in this study is similar to the RCEM
(Section 2.5). Five steps are used to generate the response surface equations of S and YD.
(1) Identify factors and ranges
The design factors considered in the response surface equations of S and YD are D, d
and t. Their ranges are given in Table 7.3.
Table 7.3 – Design Factors and Their Ranges of Robot Arm.

D (mm) 15.24 20.32 25.40
d (mm) 7.62 10.16 12.70
t (mm) 2.539 3.048 3.557
(2) Design of Experiments

A three factor – three level central composite response surface design with α =
1.6818 (for rotatability) is used as the design for the experiments. In this design, only
one replication is used because the experiments are performed on a software system
(ANSYS) instead of physical experiments. Table 7.4 lists the values of the design
variables for each of the experiments.
Table 7.4 - List of Experiments for Response Surface Generation.

Expt. No. D (mm) d (mm) T (mm)
1 15.24 7.62 2.54
2 25.4 7.62 2.54
3 15.24 12.7 2.54
4 25.4 12.7 2.54
5 15.24 7.62 3.556
6 25.4 7.62 3.556
7 15.24 12.7 3.556
8 25.4 12.7 3.556
256
9 11.7856 10.16 3.048

10 28.8544 10.16 3.048
11 20.32 5.8928 3.048
12 20.32 14.4272 3.048
13 20.32 10.16 2.19456
14 20.32 10.16 3.90144
15 20.32 10.16 3.048
(3) Simulation
There are a total of 15 experiments. For each experiment, the analysis of the robot arm is
performed by a FEA software system (ANSYS). The parametric FEA models are first
generated in a CAD software system, Pro/Engineer (www.ptc.com), and then executed in
ANSYS. The loading conditions shown in Figure 7.1.b are applied on the robot arm.
Representative stress and y-displacement plots obtained from ANSYS are given in Figure
7.8. The distributions of stress and displacement in the robot arm for all the cases are
similar to the plots shown in Figure 7.8.
Considered Regions
(a) Maximum Von-Mises Stress (b) Maximum y-displacement
Figure 7.8 – Example ANSYS Output of Analysis for the Robot Arm.
For the distributions, one can see that the Von-Mises stress is highest near the large
cylindrical hole and at the edges of the robot arm, and the y-displacement is highest at the
smaller end of the robot arm. The stress concentration at the large cylindrical hole is very
high due to modeling conditions (the cylindrical surface is completely constrained). In
practice, these stress concentrations do not exist and only arise due to modeling
limitations. Hence, the stress values around the large cylindrical surface are discarded
while determining the maximum Von-Mises stress in the robot arm. The considered
regions in the robot arm are shown in Figure 7.8.a. They are caused by the bending and
tensile forces instead of the boundary conditions, and therefore in our interests. The
displacement distribution is more uniform. The maximum displacement in the end of the
smaller hole is used as the results. The maximum stresses and y-displacements given by
ANSYS for each experiment are shown in Table 7.5.
257
Table 7.5 – Results of the Experiments for Robot Arm.

Expt. No. D (mm) d (mm) T (mm) S (Mpa) YD (mm)
1 15.24 7.62 2.54 12.781 1.2020
2 25.4 7.62 2.54 5.599 0.4418
3 15.24 12.7 2.54 11.32 0.8641
4 25.4 12.7 2.54 4.625 0.3443
5 15.24 7.62 3.556 9.075 0.8576
6 25.4 7.62 3.556 4.023 0.3184
7 15.24 12.7 3.556 8.070 0.6172
8 25.4 12.7 3.556 3.274 0.2469
9 11.7856 10.16 3.048 15.66 1.3774
10 28.8544 10.16 3.048 3.369 0.2534
11 20.32 5.8928 3.048 7.429 0.6952
12 20.32 14.4272 3.048 5.485 0.3963
13 20.32 10.16 2.19456 8.325 0.6627
14 20.32 10.16 3.90144 4.712 0.3768
15 20.32 10.16 3.048 5.999 0.4808
(4) Build Response Surface Models

A statistical software system, MINITAB, was used to perform regression analysis
and ANOVA of the obtained data. The response surfaces generated by the system are
presented in equations 7.6 and 7.7 along with their R2, R2 (adj), maximum deviation and
average deviation values.
Max_Stress (MPa) = 68.5522 - 3.1668 • D - 0.8671 • d - 9.7250 • t + 0.0457 • D • D +
0.0150•d• d + 0.4571• t • t + 0.0072 • D • d + 0.1952 • D • t + 0.0661 • d • t [7.6]
Ydisp (mm) =7.22889 -0.33009•D -0.20477•d -0.84273•t +0.00434•D•D + 0.00254•d•d
+ 0.02793 • t• t + 0.00397•D • d + 0.01794• D • t + 0.01196• d• t [7.7]
(5) Validation of response surfaces
The response surfaces of von-mises stress and y-displacement (equations 7.1 and
7.2) have very high R2 and R2 (adj) values. This indicates that the response surface fits
very well through actual data. The average errors of the fitted surfaces of stress and y-
displacement are 4.4% and 5.0%. These are reasonably small values considering the
design space considered in the problem. The maximum errors are high (16% and 18%).
It is observed that the same data point results in the maximum errors for both stress and
y-displacement response surfaces. Excluding this data point, the maximum error for
stress is decreased to 6.7%, and the maximum error for y-displacement is decreased to
8.5%. These values are considerably smaller than the original values.
To gain a better understanding between the responses and design variables, the
response surfaces are plotted in Figure 7.9.a and Figure 7.9.b. The variation of the stress
and displacement with D, d and t are almost the same. The values of stress and
258
MPa MPa
MPa
mm mm
mm mm mm
mm
(a) Relations of Stress with D, d, t
mm mm mm
mm mm
mm mm mm
mm
(b) Relations of Y-displacement with D, d, t
Figure 7.9 – Graphical Relations of Responses and Variables.

displacement decrease with increase in any of the three design variables. This is
intuitive, as the stress reduces with increase in stress bearing area. The increase in any of
the variables (D, d and t) increases the stress bearing area. The behavior of displacement
is also similar. This argument provides qualitative validation for the response surface
models.
In order to validate the quantitative models, additional experiments are designed and
performed to determine the error between actual design space (experimental data) and
approximated design space (response surface model). The values of the geometry
variables for the validation experiments are selected such that they spread evenly
throughout the design space. This approach of selecting the validation experiments is
illustrated in Figure 7.10 for a two factor- three level central composite design. The
points A1 – A9 are the experiments corresponding to the two factors - three levels central
composite response surface design. The points B1 – B4 are the validation experiments for
them. These points are evenly spaced with respect to each other and with respect to the
original set of experiments that are used to generate the response surface. For the robot
arm example which uses a three factor – three level central composite design, this
approach results in 8 experiments. A fractional factorial experimental design with 4
experiments is used in this case to obtain validation experiments.
For the four validation experiments, the values of D, d and t along with the FEA
results obtained from ANSYS are presented in Table 7.6. The maximum and average
errors of stress response surface are 7.0% and 4.3% respectively. These values for
259
A8
A4 A1
B1 B2
A7 A9 A5
B3 B4
A3 A2
A6
Figure 7.10 – Validation Experiment Design for the Response Surfaces.

displacement response surface are 8.1% and 5.1%. These error values are comparable to
the values obtained from the original set of experiments (used for response surface
generation). This indicates that the response surface is in fact an approximation of the
actual design space and is not merely a fitted line through the data points considered in
the DOE.
Table 7.6 - Results of Validation Experiments for the Robot Arm.

Exp. Design Variables Stress YDisp
No. D (mm) d (mm) t (mm) Actual Estimated % Error Actual Estimated % Error
1 22.86 11.43 3.302 4.38 4.07 7.0 0.51 0.48 6.5
2 22.86 8.89 2.794 5.57 5.49 1.5 0.69 0.69 0.0
3 17.78 11.43 2.794 8.05 8.41 4.5 0.97 1.03 5.9
4 17.78 8.89 3.302 7.27 7.58 4.3 0.95 1.03 8.1
As a summary, the stress and displacement response surfaces are validated by:
a) The response surfaces have a reasonably good fit through the actual data points.
This is tested by checking the R2, R2 (adj), maximum deviation and average
deviation values for these response surfaces.
b) It is qualitatively verified (from the response surface plots) that the behavior of
the response surfaces is in accordance with the expected behavior.
c) The low values of maximum deviation and average deviation for the validation
experiments indicate that the response surface models are in fact a reasonably
good approximation of the original design space.
• Generating Volume and Weight Equations
An analytical equation was developed to determine the volume of the robot arm.
Based on the geometry of the robot arm (Figure 7.1), an equation is given in Equation
7.8.
Volume (mm3)= {π/4•D2 + [(L/2) • (D+d) • Sin (cos-1((D-d)/2L))] – [(1/4) • (D2 – d2) •
cos-1((D-d)/2L)] – [(π/4) • (D02 – d02)]} • t [7.8]
260
Where,
L = 63.5 mm, is the center-to-center distance between the holes in the robot arm,
D0 = 10.32 mm, is the diameter of the larger hole,
d0 = 5.16 mm, is the diameter of the smaller hole.
Based on the volume, the weight of prototypes made of SL7510 can be calculated by
equations 7.9.
Weight (kg) = 1.22 * 10-3 * Volume [7.9]
The models of volume and weight represent the actual design space. They are
derived from the geometry relations, therefore valid.
7.3.3 Modeling Fabrication Processes

For the fabrication process of prototypes (direct AIM tooling), suppose SLA 3500 –
SL 7510 molds are used in the injection molding process. As discussed in Section 6.4.2,
the models of surface finish, accuracy, mold life, build time and cost for the fabrication
process were formulated based on other research work. These models are the same for
different parts. Therefore they were used in the geometric tailoring for the robot arm
directly. The models of process that are related to the robot arm case are listed as
follows.
Surface finish SF = f (PO, LT, θ1, θ2)
Surface finish of parting surface SFP = f (PO, LT)
Parallelism tolerance PTOL = f (PO, LT, HOC, FOC)
Flat tolerance FTOL = f (PO, LT, HOC, FOC)
Mold life ML = f (LT, CT, TA, D, d, t, θ1, θ2)
Mold number Nm =
LM
Np − 1
+1
OP
ML N Q
SLA build time BT = f (PO, LT, HOC, FOC, Nm)
Mold cost Cm = f (BT, TA)
Part cost Cp = f (CT)
Total cost C = Cm + Cp
Mold time Tm = f (BT, TA)
Part time Tp = f (CT)
Total time T = Tm + Tp
For the direct AIM tooling, the RP process variables are mold orientation (PO),
slicing scheme (LTp), hatch overcure (HOCp), fill overcure (FOCp), and thermal aging
(TA). In the variables, the sub-script ‘p’ corresponds to the block number of different
layers. TA is a boolean variable, which related to if the mold pieces are thermal cured
before they are used in the IJM process. The IJM process variable is cooling time (CT).
The bounds on the cooling time depend on part geometry and are different for different
parts. Experimentally it is verified that a cooling time of 5 – 7 minutes is required to
injection-mold the robot arm with SLA molds on Morgan press injection-molding
machine. Attempts to mold parts with lower cooling times resulted in deformed parts.
Beside the process variables, the manufacturer may add two draft angle variables θ1, θ2
(Figure 7.11) for mold life consideration.
261
The goals of the robot arm are transferred to goals of mold pieces directly based on
the accuracy and surface finish of the part surfaces are very close to those of the mold
surfaces (Cedorge, 1999). For a mold design given by RTMDS (Figure 7.3.d), some
process-related goals are shown in Figure 7.11. The manufacturer may add an additional
surface finish requirement (SFP ≤ 100 µin) for the parting surfaces of the mold pieces.
Surface Finish SF
Parallelism tolerance PTOL
Draft Angle α
Flatness tolerance FTOL1~3 Parting Surface Finish SFP1,2
Figure 7.11 – Some Process Planning Goals.

Based on the information given by the designer, and the models of design functions
and fabrication processes, the manufacturer can formulate a complete MPGT problem as
shown in Table 7.7.
Table 7.7 – MPGT Robot Arm Problem Formulation.

PROBLEM STATEMENT:
Find a tailored part design and related process parameters for producing functional prototypes of
the robot arm as shown in Figure 7.1. In the problem the mold design of the robot arm is already
determined (Figure 7.12).
GIVEN:
! Tensile load = 5N, bending load = 20N
262

! Prototype parts injection molded in Atactic polystyrene with SLA 3500 – SL 7510 molds
• Young’s modulus YMm = 3400 MPa
• Tensile strength TSm = 32.8 MPa
• Density Denm = 1.04 g/cc
! Required number of prototype parts Np = 50
• Flatness tolerance of mold pieces (FTOL1, FTOL2, FTOL3)
• Surface finish of mold pieces (SF1, SF2)
• Surface finish of parting surfaces (SFP1, SFP2)
• Total cost (C)
• Total time (T)
• Draft Angle (θ1, θ2)
• S (1S) # 5.99, 6.29, 6.89, 7.79, 8.99 (MPa)
• S (2S) # 5.99, 5.69, 5.09, 4.19, 3.00 (MPa)
• W (1S) # 3.40, 3.57, 3.91, 4.42, 5.10 (g)
• W (2S) # 3.40, 3.23, 2.89, 2.38, 1.70 (g)
• YD (1S)# 0.51, 0.54, 0.59, 0.66, 0.77 (mm)
• YD (2S)# 0.51, 0.48, 0.43, 0.36, 0.26 (mm)
• D (1S) # 20.32, 20.83, 21.84, 23.37, 25.40 (mm)
• D (2S) # 20.32, 19.81, 18.80, 17.27, 15.24 (mm)
• d (1S) # 10.16, 10.41, 10.92, 11.68, 12.70 (mm)
• d (2S) # 10.16, 9.91, 9.40, 8.64, 7.62 (mm)
• t (1S) # 3.048, 3.099, 3.200, 3.353, 3.557 (mm)
• t (2S) # 3.048, 2.996, 2.896, 2.743, 2.539 (mm)
• θ1 # 0.0, 0.5, 1.5, 3.0, 5.0 (degree)
• θ2 # 0.0, 0.5, 1.5, 3.0, 5.0 (degree)
• PTOL # 0.002, 0.004, 0.008, 0.013, 0.020 (in)
• FTOL1, FTOL2, FTOL3 # 0.001, 0.002, 0.004, 0.007, 0.010 (in)
• SF1, SF2 # 20, 40, 80, 130, 200 (µin)
• SFP1, SFP2 # 10, 20, 40, 65, 100 (µin)
• C # 150, 300, 600, 1000, 1500 ($)
• T # 3, 4, 5, 6, 7 (days)
! Allowable layer thickness (LT): 2, 4 and 8 mils
FIND:
! The system variables:
• D, d, t
• Draft angle variables: θ1, θ2
• RP process variables: PO, LTP, HOCp, FOCp, TA
• IJM process variables: CT
! The requirements: S, YD, W
263
• din + , din − i = 1,…,24, n = 1,…,4
SATISFY:
! Goals:
• C, T, PTOL, FTOL, SF, SFP, θ1, θ2 are minimization goals (class 1S)
c
Aq ( x ) − max Aq ( x ) − tq,n+1 ,0 h+d −
q ,n − dq+,n = 1 q = 1,…,18
tq ,n
c
Ar ( x ) + min Ar ( x ) − tr ,n+1,0 h+d −
r ,n − dr+,n = 1 r = 1,…,6
tr , n
• S (Mpa) = 68.5522 - 3.1668 • D - 0.8671 • d - 9.7250 • t + 0.0457 • D • D + 0.0150•d• d
+ 0.4571• t • t + 0.0072 • D • d + 0.1952 • D • t + 0.0661 • d • t
• YD (mm) = 7.22889-0.33009•D-0.20477•d-0.84273•t +0.00434•D•D + 0.00254•d•d +
0.02793 • t• t + 0.00397•D • d + 0.01794• D • t + 0.01196• d• t
• V (cc) = {0.7854•D2 + [1.25 • (D+d) • sin (cos-1((D-d)/5))] – [0.25 • (D2 – d2) • cos-1((D-
d)/5)] – 0.1571} • t• 16.39
• W (g) = 1.04 * V
• SFi = f (PO, LT, θ1, θ2), i =1, 2
• SFPi = f (PO, LT), i =1, 2
• PTOL = f (PO, LT, HOC, FOC)
• FTOLi = f (PO, LT, HOC, FOC) , i =1, 2,3
• ML = f (LT, CT, TA, D, d, t, θ1, θ2)
• Nm =
LM N − 1 + 1OP
p
N ML Q
•BT = f (PO, LT, HOC, FOC, Nm)
•Cm = f (BT, TA)
• Cp = f (CT)
• C = Cm + Cp
• Tm = f (BT, TA)
• Tp = f (CT)
• T = Tm + Tp
! Constraints:
• 3.00 Mpa ≤ S ≤ 8.99 Mpa
• 1.72 g ≤ W ≤ 5.15g
• 0.34 mm ≤ YD ≤ 1.02 mm
• C ≤ $1500
• T ≤ 7 days
• PTOL ≤ 0.020 inch
• FTOLj ≤ 0.010 inch, j = 1, 2, 3
• SFj ≤ 200 µin j = 1, … ,4
• SFPj ≤ 100 µin j = 1, 2
264
+ −
• din • din = 0
• din + , din − ≥ 0
! Bounds:
• 15.24 mm ≤ D ≤ 25.40 mm
• 7.62 mm ≤ d ≤ 12.70 mm
• 2.539 mm ≤ t ≤ 3.557 mm
• 0 degree ≤ θ1 ≤ 5 degree
• PO (θx, θy, θz) – Discrete variable
• TA (0 or 1) – Discrete variable
• 2 ≤ LT ≤ 8 (mils) – Discrete variable
• LT=2, 0.002 (mils) ≤ HOC ≤ 0.006 (mils), 0.012 (mils) ≤ FOC ≤ 0.016 (mils)
• 300 second ≤ CT ≤ 420 second
MINIMIZE:
c h
4 24
Z = ∑ ∑ wi,n+1 di+,n + d i−,n
n =1 i =1
After the MPGT problem for the robot arm is formulated, the solution process for
system variables are presented in the next section.
7.4 GEOMETRIC TAILORING WITH AID OF DFRTS – SOLVING
In the DFRTS, the solution process of the MPGT problem has three stages, solving
discrete variables, solving continuous variables, and selecting a solution (Figure 6.8). For
the robot arm, these steps are introduced in Section 7.4.1~7.4.3 respectively.
7.4.1 Solving Discrete Variables

The discrete variables in the DFRTS are mold design variables, part orientations and
layer thickness. They are determined with the aid of RTMDS and refined RP process
planner.
• Mold Design Variables
As presented in Section 7.2, the RTMDS can aid the manufacturer to generate
feasible mold designs for the robot arm. Two mold designs generated by the system are
further considered in this case. They are shown in Figure 7.12.
265
(a) MD1
Robot
Arm
(b) MD2
Figure 7.12 – Two Considered Mold Designs for the Robot Arm.
• Part Orientation in SLA Building
For each mold design, the mold pieces are inputted to the refined RP process planner
to generate a set of part orientations (PO) and layer thickness (LT). As describe in
Section 6.4.2, only the discrete variables (PO and LT) are considered in the modified RP
process planner. Hatch overcure (HOC) and fill overcure (FOC) are set as default values.
Its problem formulation is similar to RP-PP problem (Table 6.2). Sambu (2001) gave a
detail description of the refined RP process planner. Solving the modified RT-PP
problem results in a set of promising mold orientations and slicing schemes for each of
the mold pieces.
Two part orientations are further considered for each mold design. This results in 4
possible results as shown in Figure 7.13. For mold design 1, one orientation is obtained
for mold cavity and two are obtained for mold core. For mold design 2, two orientations
are obtained for mold cavity and one is obtained for mold core.
266
Mold
Design (a) MD1-PO1
(b) MD1-PO2
Mold
Design (c) MD2-PO1
(d) MD2-PO2
Figure 7.13 – Four Considered Part Orientations for the Mold Designs.
267
Each of these orientations yielded two promising slicing schemes. This results in a
total of 8 slicing schemes. All the slicing schemes obtained are shown in Figure 7.14. In
the figure, the dark shade corresponds to 2-mil layer thickness, light shade corresponds to
8-mil layer thickness and the medium level shade corresponds to 4-mil layer thickness.
Slicing Scheme 1 Slicing Scheme 2
Slicing Scheme 7
Slicing Scheme 8
Figure 7.14 – Eight Considered Slicing Schemes for the Part Orientations.
Slicing schemes 1 and 2 correspond to mold orientation shown in Figure 7.13.a. The
cylindrical surfaces and the parting surfaces have surface finish specifications. The
parting surface is horizontal for this orientation. Hence, in these slicing schemes, lower
layer thickness (2 and 4-mils) are used for the block with the cylindrical surface and the
remaining blocks (top and bottom) are built with the maximum allowable layer thickness
of 8-mils. The same reasoning applies for slicing schemes 5 and 6. They have the same
slicing schemes but correspond to mold orientation in Figure 7.13.c. Slicing schemes 3
and 4 correspond to mold orientation in Figure 7.13.b. The parting surface of the mold
core is vertical in this orientation and only a layer thickness of 2-mils satisfies the surface
finish constraint. Hence, most of the regions of the molds have 2-mil layer thickness.
268
The slicing schemes 7 and 8 are similar to 3 and 4 but correspond to mold orientation in
Figure 7.13.d.
For each slicing scheme, a modified MPGT problem can be formulated (Section
6.4.1) and solve with the aid of OpdesX.
7.4.2 Solving Other Variables

The values of other variables of the MPGT problem (including a discrete variables
TA) are determined by solving a modified MPGT problem. The formulation of the
modified MPGT problem is identical to the MPGT problem (Table 7.7) except for one
difference. The set of promising mold orientations and the slicing schemes are available
in the modified MPGT problem. As these values are already determined, PO and LT are
dropped from the system variables. The modified MPGT formulation for robot arm
problem is presented in Table 7.8.
Table 7.8 – Modified MPGT Robot Arm Problem Formulation.

PROBLEM STATEMENT:
Find a tailored part design and related process parameters for producing functional
prototypes of the robot arm as shown in Figure 7.1. In the problem the mold design of the robot
arm is already determined as well as the part orientation (PO) and the layer thickness (LTi) used
in SLA process.
GIVEN:
! Parametric CAD models of mold pieces
! Mold orientation (PO)
! Slicing Schemes (LT)
• Flatness tolerance of mold pieces (FTOL1, FTOL2, FTOL3)
• Surface finish of mold pieces (SF1, SF2)
• Total cost (C)
• Total time (T)
• Draft Angle (θ1, θ2)
• S (1S) # 5.99, 6.29, 6.89, 7.79, 8.99 (MPa)
• S (2S) # 5.99, 5.69, 5.09, 4.19, 3.00 (MPa)
• W (1S) # 3.40, 3.57, 3.91, 4.42, 5.10 (g)
• W (2S) # 3.40, 3.23, 2.89, 2.38, 1.70 (g)
• YD (1S)# 0.51, 0.54, 0.59, 0.66, 0.77 (mm)
• YD (2S)# 0.51, 0.48, 0.43, 0.36, 0.26 (mm)
• D (1S) # 20.32, 20.83, 21.84, 23.37, 25.40 (mm)
• D (2S) # 20.32, 19.81, 18.80, 17.27, 15.24 (mm)
269
• d (1S) # 10.16, 10.41, 10.92, 11.68, 12.70 (mm)

• d (2S) # 10.16, 9.91, 9.40, 8.64, 7.62 (mm)
• t (1S) # 3.048, 3.099, 3.200, 3.353, 3.557 (mm)
• t (2S) # 3.048, 2.996, 2.896, 2.743, 2.539 (mm)
• θ1 # 0.0, 0.5, 1.5, 3.0, 5.0 (degree)
• θ2 # 0.0, 0.5, 1.5, 3.0, 5.0 (degree)
• PTOL # 0.002, 0.004, 0.008, 0.013, 0.020 (in)
• FTOL1, FTOL2, FTOL3 # 0.001, 0.002, 0.004, 0.007, 0.010 (in)
• SF1, SF2 # 20, 40, 80, 130, 200 (µin)
• SFP1, SFP2 # 10, 20, 40, 65, 100 (µin)
• C # 150, 300, 600, 1000, 1500 ($)
• T # 3, 4, 5, 6, 7 (days)
FIND:
• D, d, t
• Draft angle variables: θ1, θ2
• RP process variables: HOCp, FOCp, TA
! The requirements: S, YD, W
• din + , din − i = 1,…,24, n = 1,…,4
SATISFY:
! Goals:
• C, T, PTOL, FTOL, SF, SFP, θ1, θ2 are minimization goals (class 1S)
c
Aq ( x ) − max Aq ( x ) − tq,n+1 ,0 h+d −
q ,n − dq+,n = 1 q = 1,…,18
tq ,n
c
Ar ( x ) + min Ar ( x ) − tr ,n+1,0 h+d −
r ,n − dr+,n = 1 r = 1,…,6
tr , n
• S (Mpa) = 68.5522 - 3.1668 • D - 0.8671 • d - 9.7250 • t + 0.0457 • D • D + 0.0150•d• d
+ 0.4571• t • t + 0.0072 • D • d + 0.1952 • D • t + 0.0661 • d • t
• YD (mm) = 7.22889-0.33009•D-0.20477•d-0.84273•t +0.00434•D•D + 0.00254•d•d +
0.02793 • t• t + 0.00397•D • d + 0.01794• D • t + 0.01196• d• t
• V (cc) = {0.7854•D2 + [1.25 • (D+d) • sin (cos-1((D-d)/5))] – [0.25 • (D2 – d2) • cos-1((D-
d)/5)] – 0.1571} • t• 16.39
• W (g) = 1.04 * V
• SFi = f (PO, LT, θ1, θ2), i =1, 2
• SFPi = f (PO, LT), i =1, 2
• PTOL = f (PO, LT, HOC, FOC)
• FTOLi = f (PO, LT, HOC, FOC) , i =1, 2,3
• ML = f (LT, CT, TA, D, d, t, θ1, θ2)
270
• Nm =
LM N − 1 + 1OP
p
N ML Q
•BT = f (PO, LT, HOC, FOC, Nm)
•Cm = f (BT, TA)
• Cp = f (CT)
• C = Cm + Cp
• Tm = f (BT, TA)
• Tp = f (CT)
• T = Tm + Tp
! Constraints:
• 3.00 Mpa ≤ S ≤ 8.99 Mpa
• 1.72 g ≤ W ≤ 5.15g
• 0.34 mm ≤ YD ≤ 1.02 mm
• C ≤ $1500
• T ≤ 7 days
• PTOL ≤ 0.020 inch
• FTOLj ≤ 0.010 inch, j = 1, 2, 3
• SFj ≤ 200 µin j = 1, … ,4
• SFPj ≤ 100 µin j = 1, 2
+ −
• din • din = 0
• din + , din − ≥ 0
! Bounds:
• 15.24 mm ≤ D ≤ 25.40 mm
• 7.62 mm ≤ d ≤ 12.70 mm
• 2.539 mm ≤ t ≤ 3.557 mm
• 0 ≤ θ1 ≤ 5
• 0 ≤ θ2 ≤ 5
MINIMIZE:
c h
4 24
Z = ∑ ∑ wi,n+1 di+,n + d i−,n
n =1 i =1
The problem is solved by an engineering optimization software system to determine

a solution for the related slicing scheme. In this case study, the modified MPGT problem
is solved by OptdesX (with SAN algorithm) for the robot arm. A C module is developed
to calculate the values of stress, Y-displacement, volume, weight, surface finish, and
accuracy in the formulation (Table 7.8). The values of SLA mold life and cost/time are
calculated by a mold life predictor and a RT cost estimator respectively (Section 6.4.2).
As explained in Section 6.4.2, OptdesX uses Simulated Annealing (SAN) algorithm
in this study. SAN algorithm runs for 5000 cycles with a maximum perturbation value of
0.3. Then, the obtained solution is further optimized by running SAN several times with
271
same maximum perturbation value (1000 cycles each time) till the solution is converged.
The similar procedure is then repeated for maximum perturbation values of 0.1, 0.03 and
0.01 (in that order) to obtain the final solution. Multiple values of maximum perturbation
can ensure problem convergence. For each modified MPGT problem, three different
starting points are also investigated to test the convergence of the solution. The starting
points used in the study are presented in Table 7.9.
Table 7.9 - Starting Points Investigated for Each Slicing Scheme.

Starting Point
Variable 1 2 3
D 15.24 20.32 25.4
d 7.62 10.16 12.7
t 2.539 3.048 3.557
TC 0 0 1
draft1 0 2.5 5
draft2 0 2.5 5
CT 300 300 300
HOC1 0.001 0.001 0.001
HOC2 0.002 0.002 0.002
HOC3 0.001 0.001 0.001
FOC1 0.002 0.002 0.002
FOC2 0.012 0.012 0.012
FOC3 0.002 0.002 0.002
Starting points 1, 2 and 3 correspond to low, medium and high values of the
variables respectively. It can be observed that same cooling time and HOC and FOC
values are used for all the starting points. The available models indicate that using the
lowest possible value of cooling time would reduce the value of objective function and
hence this value is used for all the starting points. Also, HOC and FOC do not affect the
accuracy of parts /molds built on SLA 3500 – SL 7510 and hence these values affect only
the build time. Build time is minimized by using the minimum possible values for these
variables and hence the lower bounds (for the corresponding layer thickness) are chosen
for all HOC and FOC variables.
The complete solutions (system variables) obtained from OptdesX for all the slicing
schemes are presented in Table 7.10. The related goal for all the slicing schemes are
provided in Table 7.11 (PTOL= 0.0016 inch, FTOL1 = FTOL2 = FTOL3 =0.0018 inch).
Table 7.10 – Solutions of System Variables for Different Slicing Schemes.

Slice Start D d t TA draft1 draft2 CT HOC1 HOC2 HOC3 FOC1 FOC2 FOC3
o o
# Point (mm) (mm) (mm) (0/1) () () (s) (mil) (mils) (mils) (mils) (mils) (mils)
1 20.62 9.9 2.987 0 0.61 0 300 0.001 0.002 0.001 0.002 0.012 0.002
1 2 20.68 9.92 2.990 0 0.61 0.01 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.46 9.91 2.980 0 0.61 0.03 300 0.001 0.002 0.001 0.002 0.012 0.002
2 1 20.47 10 2.997 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
2 20.18 9.91 3.035 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
272
3 20.31 9.97 3.042 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
1 20.69 9.82 2.936 0 0.59 0.01 300 0.001 0.002 0.001 0.002 0.012 0.002
3 2 20.68 9.9 2.990 0 0.61 0.02 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.69 9.89 2.988 0 0.6 0 300 0.001 0.002 0.001 0.002 0.012 0.002
1 21.12 9.48 2.901 0 0.73 0.5 300 0.001 0.002 0.001 0.002 0.012 0.002
4 2 20.64 9.8 2.901 0 0.74 0.5 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.88 9.41 2.896 0 0.72 0.5 300 0.001 0.002 0.001 0.002 0.012 0.002
1 20.30 9.9 2.989 0 0.61 0 300 0.001 0.002 0.001 0.002 0.012 0.002
5 2 20.47 9.87 2.963 0 0.6 0 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.35 9.89 2.957 0 0.6 0 300 0.001 0.002 0.001 0.002 0.012 0.002
1 20.26 10.2 2.986 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
6 2 20.32 9.99 2.997 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.29 10.2 3.021 0 0.5 0 300 0.001 0.002 0.001 0.002 0.012 0.002
1 20.56 9.93 2.973 0 0.6 0 300 0.001 0.002 0.001 0.002 0.012 0.002
7 2 20.31 9.89 3.000 0 0.61 0.04 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.46 9.9 2.980 0 0.6 0.05 300 0.001 0.002 0.001 0.002 0.012 0.002
1 21.46 9.43 2.904 0 0.72 0.5 300 0.001 0.002 0.001 0.002 0.012 0.002
8 2 21.21 9.4 2.915 0 0.72 0.5 300 0.001 0.002 0.001 0.002 0.012 0.002
3 20.90 9.51 2.920 0 0.73 0.51 300 0.001 0.002 0.001 0.002 0.012 0.002
Table 7.11 – Solutions of Goals for Different Slicing Schemes.
Slice Start Stress y-disp Weight SF1 SF2 SFP1 SFP2 Cost Obj.
# Point (Mpa) (mm) (g) (µin) (µin) (µin) (µin) ($) Mold # Function
1 6.05 0.49 3.34 65 64 6 6 570 1 0.035398
1 2 5.99 0.49 3.36 65 64 6 6 570 1 0.03542
3 6.16 0.5 3.31 65 64 6 6 570 1 0.035458
1 6.09 0.49 3.35 132 131 6 6 595 2 0.432727
2 2 6.21 0.51 3.33 132 131 6 6 595 2 0.432777
3 6.1 0.5 3.37 132 131 6 6 595 2 0.432715
1 6.14 0.5 3.29 65 64 6 64 1108 1 0.82324
3 2 6 0.49 3.36 65 64 6 64 1108 1 0.8232
3 6 0.49 3.35 65 64 6 64 1108 1 0.823168
1 6.04 0.49 3.27 132 132 6 64 1059 1 1.039292
4 2 6.26 0.51 3.24 132 132 6 64 1059 1 1.039792
3 6.21 0.51 3.22 132 132 6 64 1059 1 1.038692
1 6.24 0.51 3.3 65 64 6 6 563 1 0.035179
5 2 6.2 0.5 3.29 65 64 6 6 563 1 0.035193
3 6.29 0.51 3.27 65 64 6 6 563 1 0.035129
1 6.2 0.5 3.33 132 131 6 6 596 2 0.432818
6 2 6.2 0.5 3.32 132 131 6 6 596 2 0.432806
3 6.1 0.49 3.37 132 131 6 6 596 2 0.432769
1 6.11 0.5 3.32 65 64 6 64 1144 1 0.980352
7 2 6.22 0.51 3.31 65 64 6 64 1144 1 0.980389
3 6.17 0.5 3.31 65 64 6 64 1144 1 0.9804
1 5.84 0.48 3.32 132 132 6 64 1119 1 1.305671
8 2 5.97 0.49 3.29 132 132 6 64 1119 1 1.305311
3 6.12 0.5 3.26 132 132 6 64 1119 1 1.30588
273
The analysis and validation of the solutions are presented in Section 7.4.4. From the
solutions for the eight slicing schemes, one solution should be selected as the solution of
the MPGT problem.
7.4.3 Selecting A Solution

From Table 7.11, it can be seen that slicing schemes 1 and 5 have lowest values of
objective function. Slicing schemes 2 and 6 use 4-mil layer thickness for the cylindrical
surfaces and hence have worse surface finish than slicing schemes 1 and 5. This results
in slightly higher objective function values for these slicing schemes. All other slicing
schemes (3, 4, 7 and 8) have bad surface finish and build time (cost) goal achievements
and hence have very high values of objective function.
The slicing schemes 1 and 5 correspond to orientations in Figure 7.13.a and Figure
7.13.c, and mold design in Figure 7.12.a and Figure 7.12.b respectively. Both the mold
orientations result in same value of objective function (Table 7.11) indicating that the net
effect of the achievement of all the goals considered in the problem formulation is the
same. However, there are additional concerns that are not quantifiable (for which no
quantitative models are available) and hence are not incorporated in the problem
formulations. For the robot arm example, one such concern is the positional tolerance of
the cylindrical holes. The location of the cylindrical holes with respect to the part
boundary should be accurate but no quantitative models are available to estimate this
error. For the mold orientation in Figure 7.13.a, the positional error (of the cylindrical
surface with respect to the part boundary) occurs due to shrinkage and SLA building
inaccuracies. For mold orientation in Figure 7.13.c, apart from these errors, alignment
error also exists. When the two mold pieces come together while injection molding, it is
possible to have some alignment errors between the mold halves. This misalignment
results in higher (worse) values of the positional tolerance.
Therefore, it is determined that solution corresponding to slicing scheme 1 is
superior to other solutions, and it is regarded as the solution of the MPGT problem. With
the tailored robot arm design, the values of mold design variables, RP process variables,
and IJM process variables are also generated. They are summarized as shown in Table
7.12.
Table 7.12 – Solutions of the MPGT Problem for the Robot Arm.
D d t TA Draft1 draft2 CT HOC1 HOC2 HOC3 FOC1 FOC2 FOC3
o o
(mm) (mm) (mm) (0/1) () () (s) (mil) (mils) (mil) (mil) (mil) (mil)
20.62 9.9 2.987 0 0.61 0 300 0.001 0.002 0.001 0.002 0.012 0.002
7.4.4 Post-Solution Analysis

The objective function values of the solutions obtained for all the starting points are
very close to each other (Table 7.11). This indicating that all the starting points converge
to same value. In Table 7.10, TA is a Boolean variable to indicating if thermal aging is
performed for the molds. Thermal aging is an eight hours curing process for SLA molds,
which may slightly increase the strength of SLA molds (Rodet, 2001). However it also
increase the mold cost and time. For the robot arm, its affect on the mold life cannot
274
compensate for the increase of cost. Therefore TA has a value of 0 for all the slicing
schemes, which indicates that the mold pieces should not be thermally aged.
CT is the cooling time used in the injection molding process. With longer cooling
time, the injection molded part will become strong enough for the ejection operation.
However, as the part further shrinks onto the molds, the SLA mold life decreases.
Therefore the value of cooling time depends on the above tradeoff and varies for different
parts. For the robot arm, a lower CT bigger than 300 seconds will increase the mold life
based on the model used in the study. It also reduces part cost and time. Therefore the
value of CT will always decrease to the lower bound of the given range.
For SLA 3500, HOC and FOC only affect build time (they also affect tolerance for
SLA 250). Therefore, in the obtained solution they decrease to their lower bounds, which
results in minimum possible build time for each mold piece for the slicing schemes. The
inter-relationships between all goals and variables are shown in Figure 7.15, which can
foster a better understanding of the results listed in Table 7.10 and Table 7.11.
BOC ZL
Parallelism LT SP
Tolerance
HOC
Mold Time
Surface Finish
Flatness
FOC
Tolerance Mold Cost
PO Build Time
TA Mold Life
Total Cost
Stress
D
d Total Time
Y-Disp
t Number of Molds
Part Cost
Weight θ1
θ2 Part Time
Goal
System
CT Number of Parts variables
Others are intermediate
variables
Figure 7.15 – Inter-Relationship Diagram of Goals and Variables.

Before we can accept these solutions, we need to confirm that the operation of SAN
algorithm in solving MPGT problem for robot arm is accurate. The variation of the value
of objective function with iteration (cycle) number for slicing scheme 1 for starting point
3 is shown in Figure 7.16. The fluctuation in the value of objective function is high in the
beginning and small in the later iterations. Also the objective function value has
gradually reduced. This indicates the proper functioning of SAN algorithm in solving
modified MPGT problem.
275
Figure 7.16– Objective Function vs. Iteration No. for Modified MPGT Problem.
The effect of modifying the target ranges on the obtained solutions is also studied.
In the modified MPGT problem, all the goals can be divided into four categories:
geometry, functional, RP process and cost. The goals in RP process category are
accuracy and surface finish goals. Accuracy goals are not affected by hatch and fill
overcures (for SLA 3500 – SL 7510) and hence will not be affected by modifying targets
in MPGT problem. Also, the only variable in MPGT problem that affects surface finish
is draft angle. For small changes, draft angle does not have a significant effect on surface
finish. Hence, the effect of modifying the target ranges on RP process goals is not
significant and therefore not considered.
The experiments by modifying target ranges of the goals in the other three categories
are presented in Table 7.13. In each of these experiments, the target range for all the
goals in the corresponding category is reduced by 40%. By running OptdesX for the new
values of the goals, the results for the experiments are presented in Table 7.14. The
improvement and deterioration of the goals due to this modifying in target ranges are
represented by positive and negative percentages respectively.
Table 7.13 - Target Modification Experiments.

Goals Original Values Modified Values
Minimum ideal Maximum Minimum ideal Maximum
D (mm) 15.24 20.32 25.4 17.272 20.32 23.368
Geometry
Experiment d (mm) 12.7 10.16 12.7 11.684 10.16 11.684
t (mm) 2.539 3.048 3.557 2.7426 3.048 3.3534
stress (Mpa) 3.00 5.99 8.99 4.196 5.99 7.79
Functional
Experiment y-disp (mm) 0.26 0.51 0.77 0.36 0.51 0.666
weight (g) 1.70 3.4 5.10 2.38 3.4 4.42
Cost Exp. cost ($) 150 150 1500 150 150 960
276
Table 7.14 – Results of Target Modification Experiments.

Geometry Functional Cost
Target Values of Experiment Experiment Experiment
Goals Values Current New New New
Solution values Improvement values Improvement values Improvement
D (mm) 20.32 20.62 20.52 33.3% 20.56 20% 20.65 -10%
d (mm) 10.16 9.9 10.02 46.2% 9.91 3.8% 9.94 15.4%
t (mm) 3.048 2.987 3.012 41.0% 2.988 1.6% 2.977 -16.4%
stress(MPa) 5.99 6.05 6.027 38.3% 6.079 -48.3% 6.04 16.7%
y-disp (mm) 0.51 0.49 0.487 -15% 0.491 5% 0.49 0%
Weight (g) 3.4 3.34 3.37 50% 3.34 0% 3.34 0%
Cost ($) 150 570 570 0% 570 0% 570 0%
From the values in Table 7.14, it can be seen that all the geometry goals improve (by
more than 30%) when their target ranges are reduced. The stress goal is worsened
significantly and displacement goal is improved a little by reducing the target ranges of
functional goals. The weight goal hasn’t changed. Reducing the target range of cost goal
has not changed its value.
It is noticed that reducing the target range of functional goals has worsened the
achievement of stress goal. To identify the reason for this behavior, additional
experiments by reducing the target range of one goal at a time by 80% are performed for
all the functional goals. The new experiments are presented in Table 7.15. The obtained
results are presented in Table 7.16. They confirm that the achievements of all the goals
are improved by reducing their target ranges. This indicates that the deterioration of
stress goal when all the functional goal target ranges are reduced is due to internal trade-
offs (trade-off within the functional goals).
Table 7.15 - Individual Functional Target Modification Experiments.

Individual stress (Mpa) 3.00 5.99 8.99 5.392 5.99 6.59
Functional y-disp (mm) 0.26 0.51 0.77 0.46 0.51 0.562
Experiment weight (g) 1.70 3.4 5.1 3.06 3.4 3.74
Table 7.16 – Results of Individual Functional Target Modification Experiments.

Target Values of Stress Experiment Ydisp Experiment Weight Experiment
Goals Values Current New New New
Solution values Improvement values Improvement values Improvement
D (mm) 20.32 20.62 20.82 -66.6667 20.48 46.66667 20.72 -33.3333
D (mm) 10.16 9.9 9.93 11.53846 9.9 0 9.88 -7.69231
T (mm) 3.048 2.987 2.95 -60.6557 2.941 -75.4098 2.994 11.47541
stress(MPa) 5.99 6.05 6 83.33333 6.24 -316.667 5.97 66.66667
y-disp (mm) 0.51 0.49 0.485 -25 0.507 85 0.484 -30
Weight (g) 3.4 3.34 3.34 0 3.27 -116.667 3.37 50
Cost ($) 150 570 570 0 570 0 570 0
277
For the obtained solution (with nominal target values), all the RP process variables
(HOC and FOC) are already at their lower bounds resulting in minimum possible build
time for each mold piece for the current slicing scheme (#1). Also, the number of
required molds is the minimum possible value (1). Therefore the obtained build time is
the minimum possible build time for the slicing scheme. As the solution (Table 7.12)
indicates that no thermal aging is needed, the build time is already at its minimum
possible value. The cost is also at its minimum possible value and cannot be reduced
further for this slicing scheme. Hence, no change is detected in its value in spite of
reducing its target range.
After the solving process of the MPGT problem is validated, the results of physical
experiments are presented in the next section. They further validate the solution of the
problem.
7.5 PHYSICAL VALIDATION
Several physical experiments were performed including build time, accuracy
(flatness tolerance and parallelism tolerance), surface finish, material properties,
geometry, weight, and mold life. They verify the values of the goals given in the solution
(Table 7.12), and are described as follows.
7.5.1 Build Time Validation

The mold pieces are built on SLA 3500 – SL 7510. The layout of the mold pieces on
the SLA platform (corresponding to the obtained solution from the modified RP-PP
problem) is shown in Figure 7.17. For this layout, the estimated build time from build
station is 4:39 hours and that from the RS models is 4:53 hours. The actual build time is
4:50 hours. The values indicate the close correspondence between the actual and
estimated (RS models) values. A summary of build time validation is given in Table
7.17.
Table 7.17 - SLA Build Time for the Mold Pieces of Robot Arms.
Estimated build time (RS models) 4:53 hr
Actual build time 4:50 hr
Error 1.0%
278
Figure 7.17 – Mold Piece Layout on SLA 3500 Platform (Sambu, 2001).
7.5.2 Accuracy Validation

The accuracy of the robot arms is measured on a Brown & Sharpe’s Microval PFxTM
454 moving bridge Coordinate Measuring Machine (CMM). The repeatability of the
machine is 0.16 mils in all the three directions. A spherical probe with a 5 mm tip
diameter is used to take the measurements. The CMM machine is calibrated before
making the measurements. The robot arm is clamped in a vise and 10 data points are
taken from each of the two flat surfaces to determine the flatness and parallelism
tolerances. One of the surfaces has ejector pin marks but they are carefully avoided while
taking the measurements.
The measurements resulted in a flatness tolerance of 0.7 mil for the top surface
(without ejector pin narkss) and 1.9 mil for the bottom surface. The parallelism error
between the two surfaces is 2.2 mils. The estimated flatness and parallelism tolerances
for these surfaces (obtained from RP-PP software) are 1.7 mils and 1.6 mils respectively.
Though the flatness error of the first surface is small, the other values are close to the
expected values of 1.7 (flatness) and 1.6 mils (parallelism). A summary of accuracy
validation is given in Table 7.18.
Table 7.18 – Accuracy of the Robot Arms.

FTOL1 (mils) FTOL2 (mils) PTOL (mils)
Estimated (RP-PP) 1.7 1.7 1.6
Actual (CMM) 0.7 1.9 2.2
Error 143% -10.5% -27.3%
279
7.5.3 Surface Finish Validation

The surface finish of the robot arm is measured on a Taylor-Hobson Form Talysurf
(a surface profilometer) with a 60 mm stylus and a 0.002in tip. The repeatability of the
Talysurf is 0.86 µin. The robot arm has surface finish specification on the two inner
cylindrical surfaces. As the probe does not fit in these holes, the robot arms were broken
at the cylindrical holes, as shown in Figure 7.18, and the surface finish of the inner
surfaces (both larger and smaller cylindrical surfaces) is measured. Two measurements
were made on each surface, one with a stroke length of 1 mm and another with a stoke
length of 1.5 mm. For both measurements, a value of 0.25 mm is used as the filter
(limiting value between waviness and roughness). The two measurements serve as
replications to determine sudden variations in the surface. After ensuring that there are
no sudden variations in the surface (by comparing the results of both replications), the
measurement with stroke length of 1.5 mm is used as the surface finish value, as it has
longer trace and hence is a better representation of the complete surface.
Figure 7.18 – Pieces of Robot Arm Used in Surface Finish Experiments.

The surface finish measurements of the large and small cylindrical surfaces for 1 mm
and 1.5 mm stroke length are presented in Table 7.19 along with the estimated surface
finish values. The measurements are made for two purposes: 1) to study the effect of
shot number on the surface finish, and 2) to compare the estimated and actual surface
finish values. From the obtained data, no trend is seen in relating surface finish values to
the shot number. Hence, it is concluded that the shot number does not affect surface
finish. This indicates that either the part or the mold stair-steps deform elastically during
the part ejection to retain the surface profile.
Table 7.19 - Surface Finish Experiment Results on Prototype Parts.

Large Cyl. Surface Small Cyl. Surface
(µin) (µin)
Stroke length 1 mm 1.5 mm 1 mm 1.5 mm
Shot 2 70 75 - -
No. 7 75 63 83 80
46 82 73 71 68
Average 75 70 77 74
Max % Error 9.0 10.5 7.5 8.7
Estimated 64 64 65 65
% Error 15.2 8.7 15.7 12.0
280
The surface finish values for different stroke lengths serve as replications but the
value corresponding to 1.5 mm stroke length is a better representative of the whole
surface and hence is used for comparison purposes. The errors (between estimated and
actual values of surface finish) for the large and small cylindrical surfaces are 9% and
12% respectively. These values are small considered the experimental errors.
7.5.4 Material Property Validation

Experiments are performed on a screw driven instron testing machine to determine
the Young’s modulus and tensile strength of the injection molded polystyrene material.
The same process parameters as the obtained solution are used in the fabrication process.
However the robot arms are not good candidates for testing the material properties.
These parts have a varied cross-section and hence determining the stress and strain would
be difficult. Also, the length is too small to insert an extensometer on the part to measure
the displacement. Due to these reasons, it is decided to use standard tensile testing
specimens instead of robot arms. The same process parameters are used in producing the
tensile bars in order to ensure that the material properties of the robot arm and the tensile
bar are the same.
Four tensile testing specimens (for four replications) are used to determine material
properties (Figure 5.8). The obtained values are presented in Table 5.8 and also repeated
in Table 7.20. In the table, ‘estimated’ row corresponds to the values of the material
properties used in the MPGT problem and ‘mean’ row corresponds to the mean values of
the material properties obtained from physical experiments. The variation of the material
properties (maximum % deviation) for different replications is 4 – 9%. This is a small
error and hence the values can be considered to be consistent. Comparing the actual and
estimated values, tensile strength and Young’s modulus have an error of 1.7% and 2.5%
respectively. These values are very small indicating a very good match between the
estimated and actual values. Strain values are not used in MPGT problem and hence no
comparisons are provided.
Table 7.20 - Material Property Validation Results for Polystyrene.

Tensile Strength Young's Modulus
Replication (MPa) (MPa) % Strain @ Yield
1 30.7 3392 1.01
2 32.2 3730 0.97
3 33.6 3698 1.10
4 32.5 3123 1.15
Mean 32.3 3486 1.06
Max % Deviation 4.81 7.03 8.51
Estimated 32.8 3400 -
% Deviation 1.71 2.46 -
7.5.5 Geometry and Weight Validation

The dimensions of the fabricated robot arm (D, d, and t) are measured with vernier
calipers. The weight of the robot arm is measured on a weighting machine with an
accuracy of 0.0001 gm. The dimensions and the weight of the injection-molded robot
281
arm along with the expected values (from CAD model) are presented in Table 7.21. The
presented values are obtained for one of the prototype parts but all the parts are observed
to be similar.
Table 7.21 - Injection-molded part dimensions and weight.

D (mm) d (mm) t (mm) Weight (g)
CAD model 20.64 9.9 2.99 3.34
Actual 20.71 10.04 3.65 4.09
Error 0.3% 1.4% 18.1% 18.3%
The error of the D and d values is very small (< 1.5%). The outer surfaces of the part
are drafted (by 1.50) and hence it is difficult to measure the dimensions D and d
accurately. Hence, these values of error are reasonable (considering the mold building,
injection-molding shrinkage and measurement errors). The error in part thickness ‘t’ is
very high (18%), and cannot be considered as experimental error. It has been observed
that this error is due to flashing, which is “extra plastic enter between or under mold
parts” (Rees, 1995). The flashing occurred due to the mismatch between the parting
surfaces and not due to the opening of the mold. It is observed that the mold pieces are
curved in shape due to warpage. This resulted in a gap between the two mold pieces that
inevitably causes flashing. Also, it is observed that the clamping force did not reduce this
gap as the parting surfaces are below the mold base height, hence the clamping force is
not exerted on the molds. Flashing is observed on all the molded parts and its thickness
is measured to be around 0.6 mm. This value explains most of the error in the value of
thickness. The misalignment between the parting surfaces (cause of flash) of the mold
pieces is shown in Figure 7.19.
Mold Core
Gap due to
warpage
Mold Cavity
Figure 7.19 - Gap Between the Parting Surfaces due to Mold Warpage.
A probable reason for the considerable warpage in the mold could be due to the low
values of hatch overcure used in building the mold. From the solution obtained for the
MPGT problem (Table 7.12), it is noted that minimum possible hatch and fill overcure
values are used in building both the mold pieces. Using small values of overcure
decrease the building time. However, it also leads to smaller degree of cure of the molds
built on the SLA machine. Later, when the parts are thermally cured, all the uncured
resin is cured, which leads to bulk shrinkage and warpage. Higher quantities of uncured
282
material lead to higher warpage. Hence, a probable solution to reduce warpage is to use
higher hatch and fill overcures. In order to add this consideration in the DFRTS,
quantitative models are needed to predict warpage as a function of RP process variables.
Then this problem can be prevented.
The error in the weight of the robot arm is ≈18%. This error is also very high. But it
is primarily due to the flashing and the related increase in part thickness. Using the
values of dimensions and weights, the actual and estimated density values are computed.
The estimated density (used in MPGT problem formulation) is 1.04 gm/cc. The density
computed from the experimental values of dimensions and weight is 1.025 gm/cc. The
error in the value of density is 1.5%. This value is very small indicating that the error in
weight is primarily due to variation in thickness and not due to density.
7.5.6 Mold life validation

To verify the mold life of the generated molds, several robot arms are injection-
molded from the molds. The injection-molding parameters used are presented in Table
7.22. The value of the cooling time obtained from the MPGT solution is used in physical
experimentation. The other parameters are adjusted to obtain good parts.
Table 7.22 - Injection Molding Parameters Used for the Robot Arm.
Parameter Value Units
Clamping Force 11 Tons
Injection Pressure 3 Mpsi
Pilot Valve Pressure 70 psi
0
Plastic Temp. in barrel 430 F
0
Plastic Temp. in nozzle 450 F
Injection Time 20 Second
Cooling Time 300 Second
Release Agent Used No units
A total of 50 parts are molded from one set of molds. These include the partially
filled shots obtained during the initial phases of parameter fine-tuning and the
intermittent unexpected failed shots. Out of the 50 parts, 26 are the complete shots.
Samples of complete and incomplete shots are shown in Figure 7.21. All the 50 parts
molded have plastic solidified around the big boss feature but only 26 of these have
material solidified around the small boss feature. This indicates that the large boss
feature is tested for 50 shots and the small feature is tested for 26 shots. It is observed
that a small portion of the big boss feature is chipped off after 46 shots. The chipped
mold and the last four parts obtained from this mold are shown in Figure 7.20.
No features failed due to either pullout or flow failure for 50 shots, which is
predicted by the mold life predictor. This validates the solution that one mold is
sufficient to mold 50 parts with no pullout and flow failure.
283
Figure 7.21 - Short (left) and Complete (right) Shots obtained from IJM.
Extra material
on the part
Chipped of due to
region of the chipping
mold feature failure
Figure 7.20 - Chipped Mold and Parts Obtained from the Mold.
7.5.7 Summary of Physical Validation

The values of all the goals except stress and displacement were experimentally
measured and validated in this section. The values of the different goals /variables
obtained from physical experimentation are summarized in Table 7.23 along with their
estimated values. From the values in the table, we can see that the variables D, d,
Young’s modulus, tensile strength, build time and density have error < 1.5%. This error
is very small and can be considered as experimental or part building error. The surface
finish values have an error of around 10%. Though not very small, this error is also
reasonable. It should be noted that the surface finish models are developed for the SLA
molds and are applied for the injection-molded parts. Part fabrication and ejection could
284
cause small variations in surface finish values. Weight and part thickness have an error
of 18%. This can be attributed to the flashing because of mold warpage. The error in the
accuracy /tolerance values could be due to the small surface area used for the
measurement.
Table 7.23 - Summary of Physical Validation Results for MPGT of Robot Arm.
Goal Actual value Estimated value Error
D (mm) 20.71 20.64 -0.3%
d (mm) 10.04 9.9 -1.4%
t (mm) 3.65 2.99 -18.1%
Young's Modulus (MPa) 3486 3400 -2.5%
Tensile Strength (MPa) 32.3 32.8 1.5%
Flatness Tol1 (mils) 0.7 1.7 142.9%
Flatness Tol2 (mils) 1.9 1.7 -10.5%
Parallelism Tol (mils) 2.2 1.6 -27.3%
Large hole SF (µin) 70 64 -8.7%
Small hole SF (µin) 74 65 -12.0%
Build Time (min) 290 293 1.0%
Weight (g) 4.09 3.34 -18.3%
Density (g/cc) 1.025 1.04 1.5%
Values of stress and displacement were not measured due to lack of appropriate
measurement equipment. There was no easy way of measuring the stress and
displacement of the robot arm under the loading conditions considered in the problem.
Hence these values are not experimentally validated. However it should be noticed that
these values are partially validated by validating the Young’s modulus of tensile bars
(Section 7.5.4). The remaining part of validation (obtaining stress and/or displacement
from Young’s modulus) concerns with the finite element analysis (FEA) of the robot arm.
Some of the validating aspects of the FEA model include: verifying if the loading
conditions represent the real problem, verifying if the loading conditions are
appropriately applied, verifying if the material properties are entered correctly in correct
units, and verifying if the mesh size is small and the element shape has small aspect ratio.
For robot arm problem, these aspects are taken care while preparing the FEA model in
Pro/Engineer. Apart from these, no other physical validation is performed to explicitly
determine the stress and displacement at the considered loading conditions.
Based on the experiment results, the solution given by the DFRTS (Section 7.4.3)
satisfies the problem requirements given in Section 7.1. Fifty prototypes are produced
with satisfied surface finish and accuracy. The fabrication cost is $570 which is less than
the budget ($1500). After all the validation work for the robot arm case is presented, an
evaluation of the RTMDS and DFRTS based on the robot arm study is given in the next
section.
7.6 EVALUATION OF ROBOT ARM CASE – POST DFRTS
With the aid of the DFRTS, the material-process geometric tailoring (MPGT) was
successfully performed for the robot arm (Section 7.4). To further understand the robot
arm case, a comparing experiment is performed and presented here.
285
In the experiment, the decisions on design variables and process variables are made
sequentially. That is, the MGT problem is first formulated and solved. For the solution
of the MGT problem, the manufacturer then performs the process planning of the Rapid
Tooling. The solutions (values of geometry and RP variables) for the experiment are
presented in Table 7.24 and the corresponding goal achievements are presented in Table
7.25. The current solution and goal achievements are also listed in the table.
Table 7.24 - Solutions of Sequential and Concurrent Solution Process.

D d t TA Draft1 draft2 CT HOC1 HOC2 HOC3 FOC1 FOC2 FOC3
Variables o o
(mm) (mm) (mm) (0/1) ( ) () (s) (mil) (mils) (mil) (mil) (mil) (mil)
Sequential 20.34 10.17 3.049 0 0.62 0.1 300 0.001 0.002 0.001 0.002 0.012 0.002
Concurrent 20.62 9.9 2.987 0 0.61 0 300 0.001 0.002 0.001 0.002 0.012 0.002
Table 7.25 - Goals Achievements for Sequential and Concurrent Solution Processes.
Stress y-disp Weight SF1 SF2 SFP1 SFP2 Cost
Goals Mold #
(Mpa) (mm) (g) (µin) (µin) (µin) (µin) ($)
Sequential 6.022 0.485 3.404 65 64 6 6 570 1
Concurrent 6.05 0.49 3.34 65.3 64.4 6 6 570 1
Target 5.99 0.51 3.4 20 20 10 10 570 1
Comparing the solutions of the sequential and concurrent solution processes, it is

noticed that the sequential process has a better solution of design variables (D, d, t) and
related goal achievements (Stress, Y-displacement, weight) as their values are closer to
the target values. This is obvious because the design variables are solved first in the
sequential solution process. As no process goals are considered in the determination
process of the design variables, higher weights are assigned to the design functional
goals. This results in solutions that are closer to the target values. However, the draft
angle and surface finish got from the sequential solution process are slightly worse.
Therefore the improvement on design functions is actually compensated by the
deteriorations of process goals. At first it seems that the improvements of design
functions and variables are larger than the deteriorations of process goals. After further
analysis, a possible reason is provided as follows. As stated before, the linear physical
programming (Section 2.5) is used in assigning weights for different goals. The feasible
range of a goal is divided into sub-ranges of ideal, desirable, tolerable, and undesirable.
Based on the algorithm for weight calculation (Hernandez and Mistree, 2001), the weight
of a goal in desirable sub-range is approximately 10 times bigger than that of a goal in
ideal sub-range. This is reasonable in making tradeoffs between different goals. In
Figure 7.22, the solution of the robot arm (Table 7.12) is positioned in related sub-ranges
based on the specified goals in the MPGT problem formulation (Table 7.7). As draft
angle and surface finish are in desirable sub-range, their weights are much higher than
those of design goals which are in ideal sub-range. Therefore the small changes of draft
angle and surface finish will modify the objective function of the whole problem further
than those of design variables and goals.
In contrast to the author’s expectation, the solutions of sequential and concurrent
solution processes are almost the same (Table 7.24 and Table 7.25). This is mainly
286
Target
stress
Lower
weight weight
Y-disp
Higher
D
d weight
t
Unacceptable
θ2 θ1
Undesirable Desirable Ideal
SF1
Tolerable
PTOL
SF2
FTOL1
SFP1
FTOL2 cost
SFP2
FTOL3
SFP3
Unacceptable
Ideal Desirable Undesirable

Tolerable
Figure 7.22 – The Ranges of Goals for the Robot Arm Problem.
because the manufacturing requirements for the robot arm (50 prototypes with $1500 and
7 days) are rather easy to achieve. Therefore the process considerations are not
significant in the solving process of the MPGT problem. As discussed in Table 7.14, the
reduction of the cost range by 40% did not change its value at all. As only one mold core
and cavity are needed for 50 prototypes, and all RP and IJM variables have their values in
the lower bounds, the cost of rapid tooling process cannot be reduced further. Therefore
fixing design variables in the sequential solution process has little effects on the later
solving process for RP and IJM variables. This also explains why both solution
processes generate similar results. However, the process considerations may be much
significant for a more complicated part. Consequently the sequential and concurrent
solution processes will result in rather different solutions. A case of camera roller is used
to illustrate it, which is presented in Chapter 8. In the next section, a brief summary is
given for discussing the relevance of these results with regard to the hypotheses of the
dissertation.
In this chapter, the RTMDS and DFRTS were applied to determine the mold design
and geometric tailoring for a robot arm design. A problem of producing functional
prototypes of a robot arm was first introduced in Section 7.1. By using the RTMDS, two
mold designs were generated for the robot arm (Section 7.2). A standard mold base for
Morgan Press injection molding machine was used in the mold piece construction. The
formulation of the geometric tailoring problem for the robot arm was presented in Section
7.3. Among the three design functions, two of them were represented by response
surface equations, and one was represented by an analytical equation. The generation of
the equations was also presented in Section 7.3. The solution process of the MPGT
problem, which consists of three stages, was described in Section 7.4. The modified
MPGT problems that are related to eight slicing schemes were solved by OptdesX. The
287
results of physical experiments for the validation of the problem solution were presented
in Section 7.5. Finally an evaluation of sequential and concurrent solution processes for
the robot arm case was provided in Section 7.6.
The robot arm example highlighted different aspects of mold design and material-
process geometric tailoring problem and related solution procedures. In the discussion of
application of RTMDS and DFRTS to the robot arm example, the relationships with the
hypotheses were identified. The results from testing the hypotheses are summarized
below (Figure 7.23):
Hypothesis 1: Two mold designs were generated for the robot arm by using the Rapid
Tooling Mold Design system, which is developed based on the Multi-piece Mold
Design Method. The whole running process only took less than one minute. The
summary of testing the sub-hypotheses (H1.1 ~ H1.3) presented below will provide
more details.
Hypothesis 1.1: 40 concave edges and 66 concave faces are identified. Based on the
identified concave faces, 26 concave regions are generated for the robot arm. The
region generation process only took seconds for the robot arm.
Hypothesis 1.2: The generated concave regions of the robot arm are combined into 1
concave region and 1 convex face for one mold design, and 1 concave region and 25
convex faces for another mold design. The region combination process only took
seconds for the robot arm.
Hypothesis 1
Mold designs of robot
The requirements on mold
arm are generated by
design of the robot arm are
the RTMDS in
representative of the mold
acceptable time, and
design problems. The
they can be used to
system can be used to
produce prototpyes in
generate mold design for
the Rapid Tooling
the robot arm.
process.
Empirical Empirical
Structural Performance Hypothesis 2
Validation Validation
The requirements on Solutions of geometric

geometric tailoring of the tailoring of robot arm are
robot arm are representative generated by the DFRTS
of the geometric tailoring in acceptable time, and
problems. The system can tailored part design and
be used to perform process parameters can
geometric tailoring for the be used to produce
robot arm. functional robot arms in
Empirical Empirical
Figure 7.23 – Empirical Structural and Performance Validation for H1 and H2.
288
Hypothesis 1.3: Mold pieces are constructed for the generated mold design for a standard
mold base which can be installed in a Morgan Press injection molding machine. The
mold piece construction process only took seconds for the robot arm.
Hypothesis 2: Tailored part design with mold design variables, RP process variables and
IJM process variables were generated for the robot arm by using the Design for Rapid
Tooling system, which integrates geometric tailoring and process planning by the
manufacturer. The satisfied prototypes are produced without any iteration for the
robot arm. The summary of testing the sub-hypotheses (H2.1 ~ H2.3) presented
below will provide more details.
Hypothesis 2.1: For the robot arm the designer formulated a partial MPGT problem
based on the MPGT decision template. The manufacturer formulated a complete
MPGT problem which considered the material difference of products and prototypes.
Hypothesis 2.2: The designer’s MPGT problem formulation, RP process planning
formulation, and IJM process planning formulation were integrated into a complete
MPGT problem formulation for the robot arm.
Hypothesis 2.3: The MPGT problem was solved in a three-stage solution process. The
solution was reasonable and validated by physical experiments.
In the next chapter RTMDS and DFRTS will be applied to a more complicated case,
a camera roller (Figure 7.24). The case study will demonstrate RTMDS’s usefulness
towards determining mold design for a part with an undercut. A multi-piece mold design
is generated for the part (Section 7.2). The difficulties in fabricating the camera roller are
much more significant (Section 7.3). Therefore in addition to the geometric tailoring due
to the material properties, the camera roller will also illustrate the necessity of geometric
tailoring because of the fabrication process requirements (Section 7.4 and 7.5).
Chp 8: Prototypes of a
Camera Roller
The RP Process
Compromise DSP
Given
A. Mold Configuration B. RP Process
Mold Design
Generator Generator
Parametic Find
Parting direction Surface Finish
CAD Model Process parameters
Parting lines Accuracy
Satisfy
Mold halves Cost
Constraints
Side Actions Time
Goals
Minmize
The Designer's Deviation
C. Injection Molding
Compromise DSP
Given Process Analyzer
The IJM Process
Part Design Draft Angle
Relations between Compromise DSP
Rib Height/Width ratio
Find Part and Mold Given
Design parameters Wall Thickness
Size Parameters Mold Design
Satisfy Boss Height
Surface Finish Find
Constraints Mold Life
Accuracy Mold parameters
Goals Satisfy
Minmize Constraints
Deviation Goals
Minmize
Deviation
D. Hierarchical Input and Output

Compromise DSP
Processor Processor
Tailored Part Design /
Mold Design / RP C-DSP Template
Process Design
Chp 4: RTMDS and its Usage Chp 6: Design for Rapid Tooling
Figure 7.24 – Preview of Chapter 8.
289
Chapter 8 – Functional Prototypes of a Camera Roller
CHAPTER 8
FUNCTIONAL PROTOTYPES OF A CAMERA ROLLER
Chp 8: Prototypes of a Camera Roller
The RTMDS (Chapter 4) and DFRTS (Chapter 6) are applied to determine the mold
design and geometric tailoring for a camera roller design in this chapter. This example is
more complicated than the robot arm example discussed in the previous chapter.
Therefore it tests the applicability of the RTMDS and DFRTS for complex part designs.
A problem of producing functional prototypes of a camera roller is first introduced in
Section 8.1. By using the RTMDS, a mold design with three mold pieces is generated for
the camera roller (Section 8.2). A standard mold base for Morgan Press injection
molding machine is used in the mold piece construction. A preliminary study and
experiment are performed which results in a modified camera roller design (Section 8.3).
This modified design is considered in the DFRTS. The formulation of the geometric
tailoring problem for the modified camera roller is presented in Section 8.4. In the
formulation the design functions are represented by response surface models. The
solution process of the MPGT problem, which consists of three stages, is described in
Section 8.5. The modified MPGT problems that are related to 24 slicing schemes are
solved by OptdesX. The results of physical experiments for the validation of the problem
solution are presented in Section 8.6. Finally an evaluation of sequential and concurrent
solution processes for the camera roller case is provided in Section 8.7.
290
8.1 A CAMERA ROLLER DESIGN – PROBLEM DESCRIPTION

A more complicated part, a camera roller, is studied and presented in this chapter. It
is used to demonstrate the applicable of the RTMDS and DFRTS to more complicated
parts. This chapter is organized in a similar way as that of Chapter 7, in which a case
study of a robot arm was presented. Also similar to the case study of the robot arm, the
author worked with Shiva Sambu, a M.S. student in our lab, in performing the case study.
Ribs Grooves
Undercut
Figure 8.1 – A Production Camera Roller with a Camera.

A camera roller as shown in Figure 8.1 is a component in a disposable camera. It is
used as a spool on which a roll of film is wound. The movement of the film in the
camera is achieved by the rotation of the roller (either manually or automatically). The
primary purpose of the roller is to deliver the film at the desired rate. Although the
rotation speed while taking a picture is low (manually by fingers), the rotation speed to
reel the film onto the roller is pretty high while the camera is produced (automatically by
machines). As thousands of disposable cameras are produced each day, it is desired to
use a higher winding speed to reduce the production time for each camera. As rather
high velocities and accelerations are involved, the stress of the camera roller in the
assembly process is usually rather high. Therefore in the design of the camera roller, the
designer should verify that the stress generated by the film movement in the high winding
speed is within the material strength. The camera roller should also have small weight
to achieve low inertia, low cost and to facilitate easy handling.
Suppose during the film reeling process, loading conditions on the roller part for the
camera assembly process are shown in Figure 8.2. A tangential force of 20 N and a
291
compressive force of 5 N are applied on the cylindrical surface (red color). The
cylindrical surface of the undercut is fixed (blue color). Tangential force is due to the
acceleration and tension in winding the film. The compressive force is due to the
compression of wound film on the roller.
Loads
Fixed
Figure 8.2 - Loading Conditions on the Camera Roller.

Under these loading conditions, the designer finishes a new camera roller design
with a lower weight (2.276g) and an angle of rotation about the axis of the cylinder
(0.013 radian). Suppose in production, the camera rollers are to be fabricated in atactic
polystyrene through injection-molding process with steel molds. To test the new design,
the designer requires fifty functional prototypes of the camera roller in testing the
assembly process. In the testing, the designer is interested in the weight of the roller, and
the angle of rotation of prototypes about the axis of the cylinder. Therefore it is desired
that the prototypes should be product-representative on the values of weight and the
rotation angle under the loading conditions.
Flatness ≤ 10 mils
Surface Finish ≤ 450 µ in
Parallelism ≤ 20 mils
Positional ≤ 30 mils
Figure 8.3 - Surface Finish and Tolerance Requirements.

Considering the assembly requirements, the designer also has surface finish and
tolerance requirements for the prototypes, which are shown in Figure 8.3. The cylindrical
surface of the body and the cylindrical surface of the undercut (shown in Figure 8.1)
should have a surface finish of less than 450 µin. The surface finish for the cylindrical
surface of the body is used to ensure the reeling of the film. The surface finish for the
292
undercut is for the rotation of the roller. Besides the surface finish, the outer circular
faces of the cylindrical features should have a flatness tolerance of less than 10 mils,
parallelism tolerance of less than 20 mils, and positional tolerance of less than 30 mils.
These tolerances are required for proper assembly and alignment of the roller in the
camera.
Considering the time and budget constraints, the designer wants to get the prototypes
within a week for the maximum cost of $2500. To enable the manufacturer to produce
the prototypes within such limited time and cost, the designer decides to given much
design freedom to the manufacture for geometric tailoring. As the detail features
(grooves and ribs) of the part design will not be tested in the functional testing, the
designer does not care about them in the prototypes. That is, as long as the weight and
the angle of rotation are production-representative, the manufacturer has the freedom to
change them to facilitate the fabrication of the prototypes.
For the problem as stated in this section, the mold design for the camera roller by
using the RTMDS is presented in Section 8.2; the geometric tailoring for the given design
requirements by using the DFRTS is presented in Section 8.3 to 8.5. Finally physical
validation of the results is presented in Section 8.6.
8.2 MOLD DESIGN WITH AID OF RTMDS
The Rapid Tooling Mold Design System (RTMDS) was used in the mold design for
the camera roller in the study. Based on the part shown in Figure 8.1, a CAD model of
the camera roller was created. An additional step was added in the preparation process,
that is, face A was used to split all the faces of the part (Figure 8.4). This is because a
face is the basic element of the RTMDS. In the region generation and combination
process they usually will not be split (Section 3.5). Therefore if face F1 as shown in
Figure 8.4 is a single face, it will belong to a region and a corresponding mold piece. If it
is desired to form F1 and F2 with two mold pieces, they are required to be two different
faces in the inputted CAD model. The preparation of the camera roller was performed in
SolidWorks@, a 3D CAD system provided by SolidWorks Corporation
(www.solidworks.com). The size of the generated CAD file (.sat) is 657 KB.
F2
F1 A
Figure 8.4 – Dividing Cylindrical Faces for the RTMDS.
293
With the aid of the RTMDS, the mold designer can generate a mold design more
quickly for the camera roller. The information regarding the part, generated regions,
reverse glue operation, and the execution time of each step is listed in Table 8.1. The
reader can refer to the descriptions given in Section 4.4.2 for the meaning of each items.
The running time given in the table is based on a personal computer with a 700 MHz
Intel-III processor. One thing to be noticed is that the running time of step 7 includes the
time for the user to interactively select the input file name of the mold base. Similarly the
running time of step 8 includes the time for the user to interactively select the output file
names of the mold pieces.
Table 8.1 – The Information for Camera Roller.

544 336 446
32 32 3
Reverse glue Info. 1 20 0
For R1 GFps No. GFproj No. GFinner No.
1 0 0
Reverse glue Info. 2 67 4
For R2 GFps No. GFproj No. GFinner No.
1 0 1
0.14 2.91 0.00
Running Time Step 4 Step 5 Step 6
(seconds) 2.66 33.02 0.04
Step 7 Step 8-1 Step 8-2 Total Time
9.4* 11.6* 12.5* 63.27*
Note: ‘*” denotes that the time of interactively selecting files is included in the measurement.
The graphical results given by the RTMDS are also provided in Figure 8.5 to foster a
better understanding of the region generation and combination processes of the system.
In the figures, different colors are used to indicate different regions or convex faces
(CVX). First in the region generation process, 32 regions are generated based on the
connectivity of concave faces (336 concave faces are identified). These regions are
shown in Figure 8.5.a with two different views. In the first cycle of region combination
process, 32 regions are combined into 11 regions (Figure 8.5.b). The region combining
criteria are mainly based on the connectivity of regions and feasible parting directions
(Section 3.5.1). In the second cycle of region combination process, the 11 regions are
combined into 6 regions (Figure 8.5.c). Finally 3 regions are generated (Figure 8.5.d)
after three cycles. They cannot be further combined due to the undercut shown in Figure
8.1. Therefore three mold pieces are needed in producing the camera roller.
294
(a) Generated Regions (32 regions)
(b) Region Combination Result - Cycle 1 (11 regions, running time - 16.65 seconds)
(c) Region Combination Result - Cycle 2 (6 regions, running time - 10.69 seconds)
R2 R3
R1
(d) Region Combination Result – Cycle 3 (3 regions, running time – 5.68 seconds)
Figure 8.5 – Graphical Results of Region Combination Process.
295
Before the regions are used for the mold piece construction, two faces (F1, F2) as
shown in Figure 8.6 need to switch from region R3 to R2. The reason of F1 is combined
to R3 was discussed in Section 4.3.2. F2 is a vertical face and is neighboring to both R2
and R3. Therefore it is combinable to both regions according to our combining criteria.
In RTMDS, interactive tools are provided to select faces and switch the region numbers
of the selected faces.
R3
F1
R2 F2
Figure 8.6 – Changing Region Number of Selected Faces in the RTMDS.

The mold base used for the camera roller is shown in Figure 8.7.a. The generated
mold pieces for R1 is shown in Figure 8.7.c, and the mold piece for R2 and R3 is shown in
Figure 8.7.d. The generated glue face (GFps) is also shown in the mold pieces.
296
(a) A Mold Base for Morgan Press (b) Mold Base after Boolean Operation
GFps
(c) Mold Piece M1 for R1 (d) Mold Piece M2 for R2 and R3
Figure 8.7 – Graphical Results of a Mold Design for the Camera Roller –Phase 1.
After loading M2 as the new mold base for regions R2 and R3, the mold piece
construction algorithm (Section 3.6.3) is executed again. The parting surface used in the
construction is shown in Figure 8.8.b. The two generated mold pieces are shown in
Figure 8.8.c. The glue faces (one GFps and one GFinner) used for the mold piece
construction are also shown in the figure.
297
(a) A Mold Base for Morgan Press (b) Part with the Parting Surface
GFps
GFinner
(c) Two Generated Mold Pieces
Figure 8.8 – Graphical Results of a Mold Design for the Camera Roller – Phase 2.
The generation of CAD moldes of mold pieces is an important task in the mold
design process. After it is finished, other tasks can be started. In IronCAD system
(www.ironcad.com), the author added ejector-pin holes, gate and runner in the mold
pieces. A complete mold design for the camera roller is shown in Figure 8.9. The mold
design is further verified by physical validations. A photo of the built mold pieces and
prototype parts is given in Figure 8.10.
298
Runner
Gate
Ejector Pin Holes
Figure 8.9 – A Completer Mold Design for the Camera Roller.
Figure 8.10 – Physical Validation of the Mold Design for the Camera Roller.
8.3 GEOMETRIC TAILORING WITH AID OF DFRTS – PROBLEM ANALYSIS

Suppose the direct AIM tooling (SLA 3500 – SL7510 molds) for Atactic polystyrene
is chosen for the prototypes. Because of the complicated shape of the part, the
manufacturer cannot determine if satisfactory prototypes can be produced in such a short
time and limited budget. Therefore a preliminary analysis and experiment are performed
299
to identify the challenges in injection molding the camera roller parts without any
geometric tailoring.
From Figure 8.9, one can see that the mold pieces have several thin and tall rib
features. Injection-molding parts with such features in SLA molds could be difficult. As
a preliminary study, the mold life predictor (Section 6.4.2) is used to determine the life of
all the protrusion features in the mold pieces. The calculations are based on the models
for the mold pieces fabricated on SLA 3500 with SL 7510 resin with no thermal aging. A
cooling time of 300 seconds, injection pressure of 3 MPa, layer thickness of 2 mils and a
draft angle of 10 are used in the calculation. The mold life of different features in mold
core, mold cavity and third mold piece are shown in Figure 8.11. All the mold features
have very low life (≤ 30 parts).
10 21 23 10 29
30 26 6
4
24
21
23 26 22
9
9 9 11
10
Mold Core Mold Cavity
16
Third Piece
Figure 8.11 - Mold Life of Different Features in Mold Pieces.

As the mold life of a mold piece is determined by the lowest life of all its features,
the required mold piece number for producing 50 prototypes is estimated in Table 8.2.
Table 8.2 – Estimated Mold Life and Required Number of Each Mold Piece.
Mold Piece Mold life of Mold Piece Required Number of Mold Piece
Mold Core 9 6
Mold Cavity 4 13
Third Piece 16 4
300
A physical experiment was also performed for the camera roller. The experimental
result verified that the mold pieces, especially the mold cavity, have very low mold life.
In the experiment, two thin ribs in the mold cavity are deformed only after one shot,
which resulted in incomplete fill of all later shots (Figure 8.12). These two ribs
correspond to the features of mold cavity with mold life 4 and 6 in Figure 8.5. This
coincidence also verified that the mold life calculation presented in Figure 8.5 is
reasonable.
Related Ribs in Mold Piece
Incomplete Fills
Figure 8.12 – Problem Identified in the Preliminary Experiment.

Based on his/her experience, the manufacturer determines that it is infeasible to
produce the required prototypes in one week with only $2500. In the current usage of
Rapid Tooling, the design will be sent back to the designer with a note that it cannot be
produced within the budget. However, the designer may not know what should be
changed in order to make it fabricatable, as he/she may have no idea about why the part
cannot be made. Therefore it may take a long time with several iterations before the
required prototypes are produced. Sometimes it may also lead to the designer choosing
another fabrication process, or just increasing the budget to save the trouble.
However, as stated in Section 8.1, suppose the designer has already formulated the
design intension (weight and angle of rotation) and other requirements (surface finish and
tolerance) of the camera roller into a formulation which is then transferred to the
manufacturer. The manufacturer can try to tailor the part design based on the design and
manufacturing requirements. Within the given design freedom, the tailored design may
be fabricatable. Therefore the iteration between the designer and manufacturer can be
avoided.
In the case of camera roller, as the designer gave the design freedom of changing
detail features (grooves and ribs) of the part design (Section 8.1), the manufacturer
determines to tailor their topology before performing parametric tailoring. These
topology tailoring and the related reasoning are provided as follows.
The camera roller is designed with large number of ribs to obtain good strength with
low weight. Most of the rib features have no specific purpose apart from providing
301
strength to the part. To enable injection molding of the parts through SL molds, some of
the slots in the part are filled with material to avoid the corresponding mold feature
failure. In Figure 8.13, the ‘Rib’ feature is required to hold the film in position and hence
is an important feature. A ‘Support Rib’ is added to provide strength to the rib. Addition
of support rib results in a small ‘Slot A’. This slot corresponds to the rib feature with
mold life of 4 parts in Figure 8.11. Fabricating this slot feature with SL molds would be
difficult and hence the whole feature is removed by filling ‘Slot A’ with material.
Slot C Reel slot

Support Rib
Slot A
Thin Wall Slot B
Rib
Figure 8.13 - Different Features in Camera Roller.

In Figure 8.13, ‘Reel slot’ is required to hold one end of the film in the roller and
hence is an important feature. To avoid the failure of the corresponding mold feature,
slots B and C are filled with material. Filling slots B and C provides freedom to increase
the width of the ‘reel slot’ feature that would result in increased mold life. Also, filling
slots B and C avoids the risk of the corresponding mold feature failure. The slots located
on the opposite side of slots B and C are also filled for the same reason.
Another feature in the part that should be modified to facilitate injection molding is
‘Thin Wall’ shown in Figure 8.13. This thin wall has 0.5 mm wall thickness, 23 mm
length and 4 mm height. Filling such thin and long wall features with plastic requires
high injection pressures. But using high pressures would cause flow failure of the mold
features. Hence, based on experience, a guideline on thin wall is that all the features in
the part should have a wall thickness of larger than 0.75 mm. Accordingly the thickness
Figure 8.14 - Modified Camera Roller Part.
302
of the thin wall in camera roller is increased to 0.75 mm. The part obtained after these
modifications is shown in Figure 8.14.
The topology and parametric modifications presented so far are performed before
formulating the MPGT problem. These modifications are required to ensure the
manufacturability of the part. In MPGT problem, the functional property requirement
and mold life concern are further considered to ensure that satisfactory prototypes are
produced. The formulation and solving of the MPGT problem are presented in Section
8.4 and 8.5 respectively.
8.4 GEOMETRIC TAILORING WITH AID OF DFRTS – MODELING
To formulate the MPGT problem, the design functions and the fabrication processes
should be modeled first.
8.4.1 Modeling Design Functions

The design requirements for the camera roller are the weight (W) and the Rotation
along z-axis (ZR). In this study, the equations of W and ZR are represented by response
surface models because the camera roller has rather complicated geometry. The steps of
getting the equations are described as follows.
• Generating Z-Rotation Response Surfaces
As shown in Figure 6.10, the approach used in this study is similar to the RCEM
(Section 2.5). Five steps are used to generate the response surface equations of ZR.
(1) Identify factors and ranges
Based on the available design freedom for the modified camera roller (Figure 8.14),
the geometry variables that are considered in the geometric tailoring are shown in Figure
8.15. In the figure, D is diameter of the cylindrical hole, Wi is width of the ‘Reel Slot’, t
is thickness of the center plane, NC is number of columns of slots in the camera roller
array, and NR is number of rows. These variables are the key factors that affect the
functional properties and mold life of the features. D, Wi and t are continuous variables.
NC and NR are discrete variables. The ranges of the geometry variables are presented in
Tt
NR
Wi
NC
Figure 8.15 - Geometry Variables in the Modified Camera Roller.
303
Table 8.3.
Table 8.3 – Design Factors and Their Ranges of Camera Roller.

D (mm) 2.75 3.75 4.75
Wi (mm) 2.0 3.0 4.0
t (mm) 1.0 2.0 3.0
NC 2 - 3
NR 2 - 3
(2) Design of Experiments

A three factor – three level central composite response surface design with α = 1.0
(for rotatability) is used as the design for the experiments to generate Z-Rotation response
surface models. In this design, only one replication is considered because the
experiments are performed on a software system (ANSYS), which would result in the
same solution for all the replications. Two discrete variables are NC and NR. They can
have 4 combinations as (2, 2), (2, 3), (3, 2), and (3, 3). Three continuous variables are D,
Wi, and t. Table 8.4 lists the values of the continuous design variables for each of the
experiments.
Table 8.4 - List of Experiments for Response Surface Generation.

Expt. No. D (mm) Wi (mm) t (mm)
1 4.75 4 3
2 4.75 4 1
3 4.75 2 3
4 4.75 2 1
5 2.75 4 3
6 2.75 4 1
7 2.75 2 3
8 2.75 2 1
9 4.75 3 2
10 2.75 3 2
11 3.75 4 2
12 3.75 2 2
13 3.75 3 3
14 3.75 3 1
15 3.75 3 2
(3) Simulation
Considering both discrete variables (NC and NR) and continuous variables (D, Wi,
and t), a central composite design is used for three level variables and a full factorial
design is used for the two level variables. Therefore there are totally 4x15 = 60
experiments. For each experiment, the analysis of the robot arm is performed by a FEA
software system (ANSYS). First the parametric CAD models are generated in a CAD
software system, SolidWorks (www.solidworks.com). Then FEA modeling is performed
304
in another CAD system, Pro/Engineer (www.ptc.com). Finally finite element analysis is

performed in ANSYS. The loading conditions shown in Figure 8.2 are applied on the
camera roller. A cylindrical coordinate system is used in adding the loads. The variation
of Z-Rotation in the camera roller for experiment 1 for NC = NR = 3 is shown in Figure
8.16. The Z-Rotation has a value of zero near the cylindrical hole (where it is fixed) and
it has a maximum value of 6.517x10-3 radians at the other end. The variation of Z-
Rotation within the camera roller is similar for all the experiments. In the plot, blue color
(right end of the part) indicates small rotations while red color (left end) indicates big
rotations.
Figure 8.16 - ANSYS Output of Z-Rotation for Camera Roller (Exp.1: NC=NR = 3).
Before performing the experiments listed in Table 8.4, a set of experiments are
performed to determine a good mesh size considering the trade-off between the accuracy
of the solution and running time. The experiments performed on the modified camera
roller part and the results obtained are presented in Table 8.5.
Table 8.5 - Experiments to Study the Effect of Mesh Size.

Max size Min size Z-Rotation
Expt No. (mm) (mm) (radian) # Elements Smart mesh
1 3 0.5 0.007437 8868 No
2 2 0.5 0.007679 13967 No
3 1.6 0.5 0.007794 10045 No
4 1 0.5 0.007870 10445 No
5 1 0.3 0.007870 27728 No
6 1 0.5 0.007957 30679 Cylinders (max mesh size = 0.5)
Cylinders and surfaces near it
7 1.5 0.5 0.008013 28688 (max mesh size = 0.5)
8 0.75 0.5 0.008034 54810 No
Cylinders and surfaces near it
9 1 0.5 0.008178 70061 (max mesh size = 0.5)
305
In Table 8.5, the experiments are arranged in ascending order of the accuracy of Z-
Rotation value. In FEA, using large number of elements results in more accurate value of
stress or strain /displacement. Using lower number of elements usually results in smaller
values of stress or strain/displacement. This is also evident from Table 8.5. In this table,
max size and min size are the maximum and minimum allowable element sizes. Smart
mesh column indicates if any smart meshing is used for the part. Smart meshing involves
identifying small or crucial features and locally reducing the element size in that region.
Number of elements in the part after meshing is presented in # elements column. Z-
Rotation is the value of the rotation of the part along its axis obtained after FE analysis.
Experiments 1 ~ 4 and 8 have same values of minimum element size but different
maximum size. The number of elements and z-rotation increase with reducing the max
size. Experiment 3 is an exception since the number of elements reduces by reducing the
max size. Though, the value of z-rotation increases. This could be due to more uniform
element size throughout the part. Experiments 4 and 5 have same values of max size but
different min size. The number of elements in the part increases by reducing the value of
min size for experiment 5 but the z-rotation value did not change indicating that reducing
the value of min size is not helpful in obtaining a better solution. Comparing experiments
4 and 6, the values of min size and max size are same but in experiment 6, smart mesh is
used. The small cylindrical features in the part are meshed with a max element size of
0.5 mm. This increased the number of elements and also the accuracy of z-rotation. The
comparison between experiments 3 and 7 is similar. In experiment 7, smart mesh is used
for the cylindrical features and the surfaces around them. Experiments 6 and 9 are
similar except that surfaces around the cylindrical feature have smaller mesh size for
experiment 9.
Comparing the values of z-rotation for the first and last experiment, we can see that
experiment 1 has a 9% lower value. This means that using a max size of 3 mm with no
smart meshing results in a 9% error in z-rotation value when compared to using a max
size of 1 mm with smart meshing. However as the computation time for an experiment is
proportional to the number of elements used in the experiment, it is noticed that
experiment 9 requires approximately 25 minutes running time (on a PC with a 700MHz
Intel-III processor). This time is unacceptable when running 60 experiments to fit
response surface models. Comparing the accuracy of the solution and number of
elements in the part, experiment 7 (using max element size of 1.5 mm, min element size
of 0.5 mm and max element size of 0.5 mm around the cylindrical features) is chosen as
the standard for the experiments in this study. This experiment has a z-rotation error of
2% (compared to experiment 9) and takes approximately 10 minutes to run.
The z-rotation given by ANSYS for each experiment is shown in Table 8.6.
Table 8.6 – Results of the Z-Rotation Experiments for Camera Roller.

Z-Rotation (Radian)
Expt. No. D (mm) Wi (mm) t (mm) NC=2, NC=2, NC=3, NC=3,
NR=2 NR=3 NR=2 NR=3
1 4.75 4 3 0.007229 0.006919 0.006819 0.006517
2 4.75 4 1 0.012065 0.010893 0.010386 0.009123
3 4.75 2 3 0.00588 0.005612 0.005539 0.005187
306
4 4.75 2 1 0.009963 0.008994 0.008474 0.007387

5 2.75 4 3 0.007864 0.007501 0.007381 0.007178
6 2.75 4 1 0.012597 0.011494 0.010961 0.009877
7 2.75 2 3 0.006401 0.006137 0.006073 0.005792
8 2.75 2 1 0.010419 0.009629 0.00897 0.008013
9 4.75 3 2 0.008588 0.007962 0.007741 0.007029
10 2.75 3 2 0.009109 0.008594 0.008441 0.007631
11 3.75 4 2 0.009894 0.009249 0.008854 0.008132
12 3.75 2 2 0.008018 0.007537 0.007172 0.006588
13 3.75 3 3 0.006667 0.006325 0.006264 0.005923
14 3.75 3 1 0.010888 0.009929 0.009409 0.008348
15 3.75 3 2 0.008752 0.008202 0.007809 0.007242
(4) Build Response Surface Models

A statistical software system, MINITAB, was used to perform regression analysis
and ANOVA of the obtained data. The response surfaces generated by the system are
presented in equations 8.1 ~ 8.4 along with their R2, R2 (adj), maximum deviation and
average deviation values. In the models, the units of geometry variables (D, Wi and t) are
mm and of z-rotation (ZR) are radians.
NC =2, NR=2:
ZR2,2 = (12.1910 - 0.7585•D + 0.1764•Wi - 1.5277• t + 0.0819•D•D + 0.1828•Wi•Wi +
0.0149•t•t - 0.0223•D•Wi - 0.0197•D•t - 0.1727•Wi•t) •10-3 [8.1]
NC =2, NR=3:
ZR2,3 = (11.6270 - 0.8488•D - 0.0506•Wi - 1.1300•t + 0.0729•D•D + 0.1808•Wi•Wi -
0.0693•t•t - 0.0028•D•Wi + 0.0153•D•t - 0.1287•Wi•t) •10-3 [8.2]
NC =3, NR=2:
ZR3,2 = (11.2048 - 1.5515•D + 0.5053•Wi - 0.7899•t + 0.1768•D•D + 0.1035•Wi•Wi -
0.0629•t•t - 0.0126•D•Wi - 0.0031•D•t - 0.1548•Wi•t) •10-3 [8.3]
NC =3, NR=3:
ZR3,3 = (9.43317 - 1.0040•D + 0.2710•Wi - 0.5420•t + 0.09824•D•D + 0.12634•Wi•Wi -
0.08486•t•t - 0.02167•D•Wi + 0.01327•D•t - 0.10400•Wi•t) •10-3 [8.4]
307
(5) Validation of response surfaces

The response surfaces of z-rotation (equations 8.1 ~ 8.4) have very high R2 and R2
(adj) values (≥ 99.8 %). This indicates that the response surface fits very well through
actual data. The values of maximum and average deviation for Z-Rotation response
surfaces are ≤ 1.4% and 0.5% respectively. These values are very small which reflects
the accuracy of the response surface models.
response surface Z2,2 is plotted in Figure 8.17. The response surfaces for other values of
NC and NR are very similar.
Figure 8.17 – Graphical Relations of z-rotation and Variables.

In Figure 8.17, Z-Rotation is higher for higher values of width and for lower values
of thickness and diameter. The variation of Z-Rotation with diameter is very small
though. Increasing width removes more material in camera roller resulting in lower
inertia and hence higher values of Z-Rotation. On the other hand, increasing thickness
adds material in camera roller in the center plane resulting in increased inertia and hence
lower values of Z-Rotation. Increasing diameter changes the location of the fixed
cylindrical surface in the camera roller. Apparently, this did not have noticeable effect on
Z-Rotation. The behavior of the response surfaces is intuitive and reasonable.
To validate the quantitative models more completely, additional experiments are
performed to determine the error between actual and approximated design space. Four
validation experiments are performed (one for each response surface). The experiments
are performed in the same way as the original set of experiments. The values of
geometry variables used in these experiments along with the obtained results are
presented in Table 8.7. The maximum and average errors of Z-Rotation response surface
are 2.1% and 1.1% respectively. The error values for Z-Rotation are very small and are
comparable to the values obtained from the original set of experiments. This indicates
that the response surfaces are in fact an approximation of the actual design space and are
not merely the surfaces fitted through the data points used in the DOE.
Table 8.7 - Results of Validation Experiments of Z-rotation.

-3
Z-Rotation (* 10 ) (rad.)
Expt. No. D Wi t NC NR
Actual Estimated % Error
1 4.25 3.5 2.5 3 3 6.86 6.85 0.08
2 4.25 2.5 1.5 3 2 8.22 8.14 0.96
308
3 3.25 3.5 1.5 2 3 9.97 9.76 2.11

4 3.25 2.5 2.5 2 2 7.55 7.46 1.23
• Generating Weight Response Surfaces

Unlike the robot arm example where the weight is computed using analytical
equation, in this case weight is computed using response surfaces. This is due to the
complicated part geometry of the camera roller. The same 60 experiments used for Z-
Rotation are also used in generating the weight response surfaces. The volume of the
CAD models is obtained from SolidWorks, in which a tool of mass properties function is
provided to calculate the volume of a part.
Since the first three steps are similar to those of generating z-rotation response
surface, they are not repeated here. The volume given by SolidWorks for each
experiment is shown in Table 8.8.
Table 8.8 – Results of the Volume Experiments for Camera Roller.

3
Volume (mm )
Expt. No. D (mm) Wi (mm) t (mm) NC=2, NC=2, NC=3, NC=3,
NR=2 NR=3 NR=2 NR=3
1 4.75 4 3 2407.16 2454.03 2434.44 2477.95
2 4.75 4 1 2151.64 2214.25 2187.44 2245.6
3 4.75 2 3 2580.34 2627.21 2607.63 2651.13
4 4.75 2 1 2336.22 2398.84 2372.03 2430.18
5 2.75 4 3 2452.92 2499.79 2480.2 2523.71
6 2.75 4 1 2197.4 2260.01 2233.2 2291.36
7 2.75 2 3 2626.11 2672.97 2653.39 2696.89
8 2.75 2 1 2381.98 2444.6 2417.79 2475.94
9 4.75 3 2 2367.48 2422.22 2399.02 2449.86
10 2.75 3 2 2413.24 2467.98 2444.79 2495.62
11 3.75 4 2 2306.35 2361.09 2337.89 2388.72
12 3.75 2 2 2485.23 2539.97 2516.77 2567.6
13 3.75 3 3 2519.34 2566.2 2546.62 2590.13
14 3.75 3 1 2269.51 2329.99 2303.18 2361.34
15 3.75 3 2 2394.43 2449.17 2425.97 2476.8
Similarly MINITAB was used to perform regression analysis and ANOVA of the
obtained data. The response surfaces generated by the system are presented in equations
8.5 ~ 8.9 along with their R2, R2 (adj), maximum deviation and average deviation values.
In the models, the units of geometry variables (D, Wi and t) are mm, of volume (V) is
mm3, and of weight (W) is g.
NC =2, NR=2:
V2,2 = 2.47088 + 0.00763•D - 0.10332• Wi + 0.11638•t - 0.00407•D•D +
0.00136•Wi•Wi + 0.00285•Wi •t [8.5]
309
NC =2, NR=3:
V2,3 = 2.54752 + 0.00457•D - 0.10577•Wi + 0.11135•t - 0.00366•D•D + 0.00177•Wi•Wi
- 0.00066• t • t + 0.00285•Wi•t [8.6]
NC =3, NR=2:
V3,2 = 2.51710 + 0.00455•D - 0.10575•Wi + 0.11496•t - 0.00366•D•D + 0.00177•Wi•Wi
- 0.00066• t • t + 0.00285•Wi•t [8.7]
NC =3, NR=3:
V3,3 = 2.58255 + 0.00454•D - 0.10572•Wi + 0.10763•t - 0.00366•D•D + 0.00176•Wi•Wi
- 0.00066• t • t + 0.00285•Wi•t [8.8]
W = 1.04 * V
[8.9]
The response surfaces of volume (equations 8.4 ~ 8.8) have very high R2 and R2
(adj) values (100 %). This indicates that the response surface fits very well through
actual data. The values of maximum and average deviation for volume response surfaces
are ≤ 0.04% and 0.01% respectively. These values are very small which reflects the
accuracy of the response surface models.
response surface V2,2 is plotted in Figure 8.18. The response surfaces for other values of
NC and NR are very similar.
Figure 8.18 – Graphical Relations of Volume and Variables.
310
Variation of volume with geometry variables, as shown in Figure 8.18, is opposite to

that of Z-Rotation. This is intuitive, as Z-Rotation is inversely proportional to inertia
while volume is directly proportional to inertia. Volume is higher for lower values of
diameter and width and for higher values of thickness. Using lower values of diameter
and width removes smaller amounts of material and hence the part has higher volume.
Using higher value of thickness adds more material in the center plane and hence
increases the volume of the part. This indicates that the behavior of the response surfaces
is intuitive and reasonable.
To validate the quantitative models more completely, additional experiments are
performed to determine the error between actual and approximated design space. Four
validation experiments are performed (one for each response surface). The experiments
are performed in the same way as the original set of experiments. The values of
geometry variables used in these experiments along with the obtained results are
presented in Table 8.9. The maximum and average errors of volume response surface are
0.018% and 0.008% respectively. The error values for volume are quite insignificant and
it can be very well assumed that the estimated values of volume from response surface
are equal to the actual values.
Table 8.9 - Results of Validation Experiments of Volume.

3
Volume (mm )
Expt. No. D Wi t NC NR
Actual Estimated % Error
1 4.25 3.5 2.5 3 3 2.4774 2.4772 0.008
2 4.25 2.5 1.5 3 2 2.3991 2.3987 0.018
3 3.25 3.5 1.5 2 3 2.3559 2.3557 0.006
4 3.25 2.5 2.5 2 2 2.5117 2.5117 0.000
As a summary, the Z-rotation and weight response surfaces are validated by:
a) The response surfaces have a reasonably good fit through the actual data points.
This is tested by checking the R2, R2 (adj), maximum deviation and average
deviation values for these response surfaces.
b) It is qualitatively verified (from the response surface plots) that the behavior of
the response surfaces is in accordance with the expected behavior.
c) The low values of maximum deviation and average deviation for the validation
experiments indicate that the response surface models are in fact a reasonably
good approximation of the original design space.
8.4.2 Modeling Fabrication Processes

For the fabrication process of prototypes (direct AIM tooling), suppose SLA 3500 –
SL 7510 molds are used in the injection molding process. As discussed in Section 6.4.2,
the models of surface finish, accuracy, mold life, build time and cost for the fabrication
process were formulated based on other research work. These models are the same for
different parts. Therefore they were used in the geometric tailoring for the camera roller
directly. The models of process that are related to the robot arm case are listed as
follows.
311
Surface finish SF = f (PO, LT, θ1, θ2)

Surface finish of parting surface SFP = f (PO, LT)
Parallelism tolerance PTOL = f (PO, LT, HOC, FOC)
Flat tolerance FTOL = f (PO, LT, HOC, FOC)
Mold life ML = f (LT, CT, TA, D, d, t, θ1, θ2)
Mold number Nm =
LM
Np − 1
+1
OP
ML N Q
SLA build time BT = f (PO, LT, HOC, FOC, Nm)
Mold cost Cm = f (BT, TA)
Part cost Cp = f (CT)
Total cost C = Cm + Cp
Mold time Tm = f (BT, TA)
Part time Tp = f (CT)
Total time T = Tm + Tp
For the direct AIM tooling, the RP process variables are mold orientation (PO),
slicing scheme (LTp), hatch overcure (HOCp), fill overcure (FOCp), and thermal aging
(TA). In the variables, the sub-script ‘p’ corresponds to the block number of different
layers. TA is a boolean variable, which related to if the mold pieces are thermal cured
before they are used in the IJM process. The IJM process variable is cooling time (CT).
The bounds on the cooling time depend on part geometry and are different for different
parts. Experimentally it is verified that a cooling time of 5 – 7 minutes is required to
injection-mold the robot arm with SLA molds on Morgan press injection-molding
machine. Attempts to mold parts with lower cooling times resulted in deformed parts.
Beside the process variables, the manufacturer may add three draft angle variables θ1, θ2,
and θ3 (Figure 8.19) for mold life consideration. In this study a uniform draft angle is
used for a mold piece. That is, draft angles θ1, θ2 and θ3 are considered for all protrusion
features of the mold core, mold cavity, and the third mold piece respectively. All
together, there are a total of 15 protrusion features in the mold pieces. It is possible to
fabricate each of them with different draft angles. However considering 15 different draft
angle variables significantly increases the computational time and hence only three
different variables (one for each mold piece) are considered in this study.
Obtaining low values of draft angles is considered as a goal in the MPGT problem.
Using higher draft angles increases the volume (and hence weight) of the part and also
distorts the whole part geometry and hence is undesirable. Other goals of the camera
roller given by the designer can be transferred to goals of mold pieces directly based on
the accuracy and surface finish of the part surfaces are very close to those of the mold
surfaces (Cedorge, 1999). For a mold design given by RTMDS (Figure 8.9), some
process-related goals are shown in Figure 8.19. The manufacturer may add an additional
surface finish requirement (SFP ≤ 200 µin) for the parting surfaces of the mold pieces.
The manufacturer may also add some other requirements in the mold pieces for
producing the required prototypes parts. These requirements include positional tolerance
(PoTOL1, PoTOL2), flatness tolerance of parting surfaces (FTOLP1, FTOLP2, FTOLP3,
FTOLP4, FTOLP5), and perpendicularity tolerance (PeTOLP1, PeTOLP2) (refer to Figure
8.19).
312
Flatness ≤ 20 mils (FTOLP1)

Perpendicularity ≤ 10 mils (PeTOLP1) Flatness ≤ 20 mils (FTOLP3)
Draft angle Perpendicularity ≤ 10 mils
Surface Finish Surface Finish
θ1 ≤ 200 µin (PeTOLP2) ≤ 200 µin
(FTOLP2) (SFP3)
Surface Finish Draft angle

≤ 450 µin
θ2
Primary Parting
Surface
Primary Parting Secondary
Secondary Parting Surface
Surface Surface Finish Parting Surface
≤ 450 µin (FTOLP4)
(FTOLP2)
Flatness ≤ 10 mils Flatness ≤ 10 mils

Parallelism ≤ 20 mils Parallelism ≤ 20 mils
Positional ≤ 30 mils Positional ≤ 30 mils
(PoTOL1) (PoTOL2)
(FTOLP5)
Flatness ≤ 20 mils
Surface Finish
≤ 200 µin
Draft angle (SFP5)
θ3
Secondary
Parting Surface
Surface Finish
≤ 450 µin
Figure 8.19 – Process Planning Goals for the Mold Pieces.

It is noticed that in the third mold piece, two of the surface finish goals are coupled
with draft angle variables (θ3). These surface finish goals correspond to the cylindrical
undercut in the camera roller and the secondary parting surface. The drafts in the other
mold pieces affect the rib features in the mold but not the cylindrical surface in the mold
(that has surface finish requirement). Hence the other surface finish requirements are not
coupled.

Based on the information given by the designer, and the models of design functions
and fabrication processes, the manufacturer can formulate a complete MPGT problem as
shown in Table 8.10. In the formulation the injection pressure (IP) used in fabricating the
parts is also added because high injection pressure may lead to flow failure of the molds.
Based on the preliminary experiment, it is determined that an injection pressure of 2.0
Mpsi is sufficient for molding camera roller parts.
313
Table 8.10 – MPGT Camera Roller Problem Formulation.

PROBLEM STATEMENT:
Find a tailored part design and related process parameters for producing functional
prototypes of the camera roller as shown in Figure 8.1. In the problem the mold design of the
camera roller is already determined which is shown in Figure 8.19.
GIVEN:
! Parametric CAD model of modified camera roller
! Parametric CAD models of the mold pieces for the camera roller
! Geometry variables that affect part functionality: D, Wi, t, NC, NR
! Tangential force = 20N, compressive force = 5N
! Prototype parts injection molded in Atactic polystyrene with SLA 3500 – SL 7510 molds
• Young’s modulus YMm = 3400 MPa
• Tensile strength TSm = 32.8 MPa
• Density Denm = 1.04 g/cc
! Injection pressure (IP) = 2.0 Mpsi
• Z-Rotation requirement (ZR),
• Geometry variables requirements (D, Wi, t)
• Parallelism tolerance between cylindrical surfaces (PTOL1, PTOL2)
• Flatness tolerance of mold pieces (FTOL1, FTOL2, FTOL3, FTOL4)
• Positional tolerance of mold pieces (PoTOL1, PoTOL2)
• Flatness tolerance of parting surfaces (FTOLP1, FTOLP2, FTOLP3, FTOLP4, FTOLP5)
• Perpendicularity tolerance of mold pieces (PeTOLP1, PeTOLP2)
• Surface finish of mold pieces (SF1, SF2, SF3, SF4)
• Surface finish of parting surfaces (SFP1, SFP2, SFP3, SFP4, SFP5)
• Total cost (C)
• Total time (T)
• Draft Angle (θ1– Features in mold core, θ2– mold cavity, θ3– third mold piece)
• ZR (1S) # 0.01308, 0.01439, 0.01570, 0.01766, 0.01962 (radians)
• ZR (2S) # 0.01308, 0.01177, 0.01046, 0.00850, 0.00654 (radians)
• W (1S) # 2.276, 2.390, 2.504, 2.663, 2.845 (g)
• W (2S) # 2.276, 2.162, 2.049, 1.889, 1.707 (g)
• D # 2.75, 3.15, 3.55, 4.15, 4.75 (mm)
• Wi # 2.0, 2.4, 2.8, 3.4, 4.0 (mm)
• t # 1.0, 1.4, 1.8, 2.4, 3.0 (mm)
• θ1 # 0.0, 1.0, 2.0, 3.5, 5.0 (degrees)
• θ2 # 0.0, 1.0, 2.0, 3.5, 5.0 (degrees)
• θ3 # 0.0, 1.0, 2.0, 3.5, 5.0 (degrees)
• PTOL1, PTOL2 # 0.002, 0.004, 0.008, 0.013, 0.020 (in)
• FTOL1, FTOL2, FTOL3, FTOL4 # 0.001, 0.002, 0.004, 0.007, 0.010 (in)
314
• PoTOL1, PoTOL2 # 0.003, 0.005, 0.010, 0.017, 0.030 (in)

• FTOLP1, FTOLP2, FTOLP3, FTOLP4, FTOLP5# 0.002, 0.004, 0.008, 0.013, 0.020 (in)
• PeTOLP1, PeTOLP2 # 0.001, 0.002, 0.004, 0.007, 0.010 (in)
• SF1, SF2, SF3, SF4 # 75, 125, 200, 300, 450 (µin)
• SFP1, SFP2, SFP3, SFP4, SFP5 # 20, 40, 80, 130, 200 (µin)
• C # 500, 900, 1300, 1900, 2500 ($)
• T # 3, 4, 5, 6, 7 (days)
! Allowable layer thickness (LT): 2, 4 and 8 mils
FIND:
• D, Wi, t, NC, NR
• Draft angle variables: θ1, θ2, θ3
• RP process variables: PO, LTP, HOCp, FOCp, TA. Where p corresponds to block number.
! The requirements: ZR, W
• din + , din − i = 1,…,35, n = 1,…,4
SATISFY:
! Goals:
• C, T, PTOL, FTOL, PoTOL, FTOLP, PeTOL, SF, SFP, D, Wi, t, θ1, θ2, θ3 are
minimization goals (class 1S)
• ZR and W are target-matching goals (class 3S) and each of these goals is split into two
independent goals of class 1S and class 2S.
c
Aq ( x ) − max Aq ( x ) − tq,n+1 ,0 h+d −
q ,n − dq+,n = 1 q = 1,…,33
tq ,n
c
Ar ( x ) + min Ar ( x ) − tr ,n+1,0 h+d −
r ,n − dr+,n = 1 r = 1, 2
tr , n
• ZR (radian) = f (D, Wi, t, NC, NR)
• W (g) = f (D, Wi, t, NC, NR)
• SFi = f (PO, LT), i =1, 2
• SFi = f (PO, LT, θ3), i =3, 4
• SFPi = f (PO, LT), i =1, …, 5
• PTOLi = f (PO, LT, HOC, FOC), i =1, 2
• FTOLi = f (PO, LT, HOC, FOC), i =1, …, 4
• PoTOLi = f (PO, LT, HOC, FOC), i =1, 2
• FTOLPi = f (PO, LT, HOC, FOC), i =1, …, 5
• PeTOLPi = f (PO, LT, HOC, FOC), i =1, 2
• ML1 = f (LT, CT, IP, TA, Wi, t, NC, NR, θ1) [8.10]
• ML2 = f (LT, CT, IP, TA, Wi, t, NC, NR, θ2) [8.11]
• ML3 = f (LT, CT, IP, TA, D, θ3) [8.12]
• N mi =
LM N − 1 + 1OP
p
N ML Q
i= 1, 2, 3
i
315
•BT = f (PO, LT, HOC, FOC, Nmi)

•Cm = f (BT, TA)
• Cp = f (CT)
• C = Cm + Cp
• Tm = f (BT, TA)
• Tp = f (CT)
• T = Tm + Tp
! Constraints:
• 0.00654 radian ≤ ZR ≤ 0.01962 radian
• 1.707 g ≤ W ≤ 2.845g
• C ≤ $2500
• T ≤ 7 days
• PTOLj ≤ 0.020 inch, j = 1, 2
• FTOLj ≤ 0.010 inch, j = 1, …, 4
• PoTOLj ≤ 0.030 inch, j = 1, 2
• FTOLPj ≤ 0.010 inch, j = 1, …, 5
• PeTOLPj ≤ 0.010 inch, j = 1, 2
• SFj ≤ 450 µin j = 1, … ,4
• SFPj ≤ 200 µin j = 1, …, 5
+ −
• din • din = 0
• din + , din − ≥ 0
! Bounds:
• 2.75 mm ≤ D ≤ 4.75 mm
• 2.0 mm ≤ Wi ≤ 4.0 mm
• 1.0 mm ≤ t ≤ 3.0 mm
• 2 ≤ NC ≤ 3 – Discrete variable
• 2 ≤ NR ≤ 3 – Discrete variable
• PO (θx, θy, θz) – Discrete variable
• 2 ≤ LT ≤ 8 (mils) – Discrete variable
MINIMIZE:
c h
4 35
Z = ∑ ∑ wi,n+1 di+,n + d i−,n
n =1 i =1
In the problem formulation (Table 8.10), the targets for the goals are formulated in
Linear Physical Programming formulation (Section 2.6.2). The meaning of target values
is also explained in the robot arm study (7.3.1). The mold life (ML1, ML2, ML3, Equation
8.10 ~ 8.12) is calculated for all the protrusion features. There are a total of 4 rib features
316
in mold core (ML1), 10 rib features in mold cavity (ML2) and 1 boss feature in third mold
piece (ML3). The dimensions of these mold features are presented in Table 8.11. These
are the dimensions of the features in modified camera roller part shown in Figure 8.15.
The system variables NC and NR affect the quantity and dimensions of these features.
Therefore based on the values of NC and NR, the dimension of each feature should be
calculated and used in the evaluation of the mold life of the mold pieces. As stated
before, life of a mold piece is equal to the smallest value of mold life of the features in
the corresponding mold piece. Mold life affects the number of different mold pieces
required to fabricate desired number of functional prototypes. Number of molds in turn
affects build time and fabrication cost.
Table 8.11 - Dimensions of the Different Mold Features.

Height Width /Dia Thick Part Thick
Feature No Type Mold Piece
(mm) (mm) (mm) (mm)
1 Rib Core 4.90 12.20 2.00 1.1
2 Rib Core 4.45 5.80 2.25 1.1
3 Rib Core 4.45 5.80 2.25 1.1
4 Rib Core 4.45 7.40 2.25 1.1
5 Rib Cavity 4.90 6.50 2.00 1.1
6 Rib Cavity 4.45 2.90 1.25 1.1
7 Rib Cavity 4.45 5.80 1.25 1.1
8 Rib Cavity 4.45 5.80 1.25 1.1
9 Rib Cavity 4.90 7.40 2.00 1.1
10 Rib Cavity 4.90 7.40 2.00 1.1
11 Rib Cavity 4.90 7.40 2.00 1.1
12 Rib Cavity 4.45 5.80 1.25 1.1
13 Rib Cavity 4.45 5.80 1.25 1.1
14 Rib Cavity 4.45 5.80 1.25 1.1
15 Boss Third Piece 3.50 2.75 *** 3.0
After the MPGT problem for the camera roller is formulated, the solution process for
system variables are presented in the next section.
8.5 GEOMETRIC TAILORING WITH AID OF DFRTS – SOLVING
In the DFRTS, the solution process of the MPGT problem has three stages, solving
discrete variables, solving continuous variables, and selecting a solution (Figure 6.8). For
the camera roller, these steps are introduced in Section 8.5.1~8.5.3 respectively.
8.5.1 Solving Discrete Variables

The discrete variables in the DFRTS are mold design variables, part orientations and
layer thickness. They are determined with the aid of RTMDS and refined RP process
planner.
• Mold Design Variables
As presented in Section 8.2, the RTMDS can aid the manufacturer to generate
feasible mold designs for the camera roller. The camera roller is a complicated part with
undercuts and intricate geometry. Consequently there is little freedom in choosing the
317
mold design variables. It should be noted that more mold designs are possible with larger
number of mold pieces. However since it is desirable to use the least possible number of
mold pieces (Section 3.2), only one mold design, which is shown in Figure 8.9, is further
considered.
• Part Orientation in SLA Building
For the mold design, the mold pieces are inputted to the refined RP process planner
to generate a set of part orientations (PO) and layer thickness (LT). As describe in
Section 6.4.2, only the discrete variables (PO and LT) are considered in the modified RP
process planner. Hatch overcure (HOC) and fill overcure (FOC) are set as default values.
Its problem formulation is similar to RP-PP problem (Table 6.2). Sambu (2001) gave a
detail description of the refined RP process planner. Solving the modified RT-PP
problem results in a set of promising mold orientations and slicing schemes for each of
the mold pieces.
Two orientations (one favorable for build time and the other favorable for surface
finish) are obtained for mold core and mold cavity. One orientation is obtained for third
mold piece. These orientations result in 4 possible combinations as shown in Figure
8.20.
The mold pieces are combined together in IronCAD to create a single ACIS file.
Slicing module is run for all the four combinations or orientations. Six promising slicing
schemes are generated for each of the combinations. The slicing schemes obtained for
combination 1 are shown in Figure 8.21. In this figure, dark shade corresponds to 2-mil
layer thickness, light shade corresponds to 8-mil layer thickness and the medium level
shade corresponds to 4-mil layer thickness.
318
(a) PO1
(b) PO2
(c) PO3
(d) PO4
Figure 8.20 – Four Part Orientations for the Mold Design (Sambu, 2001).
319
Slicing Scheme 1
Slicing Scheme 2
Figure 8.21 - Promising Slicing Schemes for PO1 (Sambu, 2001).
Slicing schemes 1 ~ 4 correspond to the solutions that result in the smallest values of
the objective function. They have small layer thickness and hence achieve best possible
accuracy and surface finish. Slicing schemes 5 and 6 have best possible build time
achievement. However the surface finish requirements are barely met. It is observed that
the primary trade-off in the slicing scheme is between surface finish and build time goals.
A comparison of surface finish and build time goal achievements for these slicing
schemes is presented in Table 8.12. The slicing schemes for the other combinations are
similar.
Table 8.12 - Surface Finish and Build Time Achievements for Slicing Schemes.
Slicing SF_part_ SF_part_
SF_mold1 SF_mold2 SF_mold3 Build Time
Scheme mold1 mold2
1 156 156 64 64 64 13:59
2 156 156 64 68 68 13:39
3 156 156 131 74 72 13:13
4 156 156 64 77 75 12:52
5 390 390 400 187 177 6:28
6 382 390 130 187 177 6:29
320
There are totally 4x6 = 24 slicing schemes. For each slicing scheme, a modified
MPGT problem can be formulated (Section 6.4.1) and solve with the aid of OpdesX.
8.5.2 Solving Continuous Variables

The values of other variables of the MPGT problem (including three discrete
variables: NC, NR, TA) are determined by solving a modified MPGT problem. The
formulation of the modified MPGT problem is identical to the MPGT problem (Table
8.10) except for one difference. The set of promising mold orientations and the slicing
schemes are available in the modified MPGT problem. As these values are already
determined, PO and LT are dropped from the system variables. Similar to the robot arm
study (Section 7.4), the modified MPGT formulations for camera roller problem are
developed from the MPGT problem (Table 8.10).
The formulated modified MPGT problems are then solved by an engineering
optimization software system to determine a solution for the related slicing scheme. In
the camera roller study, OptdesX (with SAN algorithm) was used to solve the modified
MPGT problem for all the 24 slicing schemes obtained from the modified RP PP
problem. The solving process is similar to that of the robot arm study (Section 7.4.2).
Therefore only the differences are explained and presented here.
Three starting points, low, medium and high, are investigated to determine the effect
of starting point on the obtained solution. The starting points used in the study are
presented in Table 8.13. The available models indicate that using the smallest possible
cooling time, HOC and FOC values results in the best solution for molds built on SLA
3500 with SL 7510 resin and hence different values for these variables are not
investigated in starting points.
Table 8.13 - Starting Points Investigated for Each Slicing Scheme.

Variable Low Medium High
D 2.75 3.75 4.75
Wi 2 3 4
t 1 2 3
NR 2 2 3
NC 2 3 3
θ1 0 2.5 5
θ2 0 2.5 5
θ3 0 2.5 5
TC 0 0 1
CT 300 300 300
HOC1-6 0.002 0.002 0.002
FOC1-6 0.012 0.012 0.012
The obtained solutions for slicing scheme 1 are presented in Table 8.14. The values
of the system variables for the solutions obtained for all the starting points are very close
indicating a good convergence. Also, the values of objective function are very close too.
This indicates that the solution of modified MPGT/RT problem is independent of starting
point.
321
Table 8.14 - Solutions Obtained for Slicing Scheme 1 for Different Starting Points.
Variable Low Medium High
D 2.76 2.78 2.76
Wi 2.80 2.80 2.78
t 1.00 1.00 1.00
NR 2 2 2
NC 2 2 2
θ1 1.3 1.3 1.3
θ2 1.2 1.2 1.2
θ3 1.0 1.0 1.0
TC 0 0 0
CT 300 300 300
HOC1-6 0.002 0.002 0.002
FOC1-6 0.012 0.012 0.012
Z 0.0152 0.0152 0.0154
The variation of the objective function with iteration (cycle) number for medium
starting point is shown in Figure 8.22. The plot is for 5000 cycles and the total run time
is 2 minutes and 30 seconds. This plot shows an expected behavior of the SAN algorithm
(large fluctuations in the beginning and small fluctuations at the end) indicating its proper
functioning for this problem.
Figure 8.22 - Objective Function vs. Iteration Number for MPGT Problem.
The objective function values for the solution obtained for the 24 slicing schemes are
presented in Table 8.15. The complete listing of the obtained solutions is presented in
Appendix B. The analysis and validation of the solutions are presented in section 8.5.4.
Among the 24 solutions, a solution is chosen as the solution of the MPGT problem,
which is presented in the next section.
322
Table 8.15 - Objective Function Values Obtained for Different Slicing Schemes.
Z SS1 SS2 SS3 SS4 SS5 SS6
PO1 0.0152 0.0157 0.0193 0.0168 4.4450 3.4922
PO2 0.0348 0.0339 0.0327 0.0384 1.1760 1.0641
PO3 0.0371 0.0336 0.0361 0.0357 3.9700 4.0338
PO4 0.0363 0.0324 0.0352 0.0344 4.1710 3.8244
8.5.3 Selecting A Solution

From the 24 values in Table 8.15, it can be seen that slicing scheme 1 for PO1
(Figure 8.20.a) has the smallest value of objective function. Therefore, it is determined
that the solution corresponding to PO1-SS1 is regarded as the solution of the MPGT
problem.
With the tailored camera roller design, the values of mold design variables, RP
process variables, and IJM process variables are also determined. They are summarized
as shown in Table 8.16. The values of goals, significant intermediate responses and
objective function (along with the ideal target values) for this solution are presented in
Table 8.17.
Table 8.16 – Solutions of the MPGT Problem for Camera Roller.

D Wi t θ1 θ2 θ3 CT HOC1-6 FOC1-6
NR NC o o o TA
(mm) (mm) (mm) () () () (sec) (mil) (mil)
2.78 2.80 1.00 2 2 1.3 1.2 1.0 0 300 0.002 0.012
Table 8.17 - Values of Goals and Intermediate Responses for the Solution.
ZR W SF1 SF2 SF3 SFP1 SFP2 SFP3
(rad.) (g) (µin) (µin) (µin) (µin) (µin) (µin)
Achieved 0.0104 2.40 156 156 65 6 64 6
Target 0.0131 2.28 75 75 75 20 20 20
SFP4 SFP5 FTOL PTOL PoTOL FTOLP PeTOLP Cost
(µin) (µin) (inch) (inch) (inch) (inch) (inch) ($)
Achieved 64 6 0.0018 0.0016 0.0064 0.0018 0.0020 1291
Target 20 20 0.0010 0.0020 0.0030 0.0020 0.0010 500
D Wi t
Z # molds1 # molds2 # molds3
(mm) (mm) (mm)
Achieved 2.78 2.80 1.00 0.0152 1 1 3
Target 2.75 2.00 1.00 0 - - -
8.5.4 Post-Solution Analysis

From Table 8.16, one can see that the solution has NR = NC = 2 indicating that this
combination results in a better solution than the original values used in the production
part (NR = NC = 3). Considering the mold life concern for thin and tall features in SL
molds, this is reasonable. Using lower number of columns and rows increases the width
323
of the rib features in the mold and hence increases their life. The solution has the draft
angle values > 10 indicating that drafts are necessary in the mold pieces for the camera
roller. As in the robot arm example (Section 7.4.3), no thermal aging is preferred.
Cooling time, HOC and FOC converge to their lower bounds as expected.
From the values of goal achievements for the solution presented in Table 8.17, one
can see that z-rotation (ZR) is lower than its target value and weight (W) is higher than its
target. The difference between the achieved and target values is 21% for z-rotation and
5.3% for weight. We also run ANSYS for the initially modified part (Figure 8.14). The
values of ZR and W for the part are 0.0080 (radian) and 2.57 (g) respectively. The
difference between them and the target values is 40% for z-rotation and 13% for weight.
Therefore by performing geometric tailoring, the difference of these values is reduced.
The cylindrical surfaces on mold core (SF1) and mold cavity (SF2) have a surface
finish of 156 µin and the cylindrical surface on third mold piece (SF3) has a surface finish
of 65 µin. The cylindrical surfaces in core and cavity are horizontal and hence have
worse surface finish than the cylindrical surface in third mold piece, which is vertical.
The primary parting surface for all the mold pieces are horizontal and hence have a good
surface finish of 6 µin. The secondary parting surfaces for mold core and mold cavity are
vertical and have a surface finish of 64 µin. The flatness tolerance on parting surface and
parallelism tolerance between the cylindrical faces are lower than the desired value
(achieved better than what is desired) while the other tolerances are more than their ideal
values. The ideal value for cost is $500 and maximum allowable value is $2500. The
cost for the obtained solution is $1291.
Diameter (D), width (Wi) and thickness (t) are all system variables and goals in the
MPGT problem. For the obtained solution, the difference between ideal and achieved
value for D, Wi and t are 1.1%, 40% and 0% respectively. This indicates that the
diameter and thickness need not be varied much to satisfy the functional property and
mold life requirements, though width needs to be modified significantly. The goals are
well satisfied by modifying width (Wi), number of rows (NR) and number of columns
(NR) in the part.
Table 8.17 also contains the values of objective function (Z) and the number of
different mold pieces required to fabricate the desired number of parts. According to the
obtained solution, one of mold core and cavity and three of third mold pieces are required
to fabricate the desired number of functional prototypes. Compared to the required
number of each mold pieces (6 for mold core, 13 for mold cavity, 4 for the third piece as
shown in Table 8.2), this is significantly smaller. It is also possible to mold 50 parts with
only one of third mold piece. However higher draft angle (θ3) is required. Using three
mold pieces results in a better trade-off between draft angle and cost goals. Therefore the
obtained solution shows a reasonable trade-off between different goals and also has
reasonable values for the individual goal achievements.
Further validation of the solution is performed by experiments of reducing the target
ranges of different goals. For the camera roller, as surface finish and tolerance goals are
affected by only mold orientation and slicing scheme, the effect of target range on these
goals is not studied. The experiments by modifying target ranges of each other goal are
presented in Table 8.18. In each of these experiments, the target range for the goal in the
324
corresponding category is reduced by 40%. By running OptdesX for the new values of
the goals, the results for the experiments are presented in Table 8.19. The improvement
and deterioration of the goals due to this modifying in target ranges are shown in Table
8.20. Positive values indicate improvement and negative values indicate deterioration of
the corresponding goal.
Table 8.18 - Target Modification Experiments.

ZR (rad.) 0.00654 0.01308 0.01962 0.009156 0.01308 0.017004
W (g) 1.707 2.276 2.845 1.9346 2.276 2.6174
D (mm) 2.75 2.75 4.75 2.75 2.75 3.95
Wi (mm) 2.0 2.0 4.0 2.0 2.0 3.2
t (mm) 1.0 1.0 3.0 1.0 1.0 2.2
cost ($) 500 500 2500 500 500 1700
Table 8.19 – Results of Target Modification Experiments.

Target whose range is reduced
Target Current
ZR exp. W exp. D exp. Wi exp. t exp. Cost exp.
Goals Values Solution
ZR (rad.) 0.01308 0.01041 0.01041 0.01041 0.01042 0.01012 0.01042 0.01029
W (g) 2.276 2.40 2.40 2.40 2.40 2.43 2.40 2.39
D (mm) 2.75 2.78 2.76 2.77 2.76 2.78 2.75 3.14
Wi (mm) 2.00 2.80 2.80 2.80 2.80 2.48 2.80 2.81
t (mm) 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
cost ($) 500 1291 1291 1291 1291 1291 1291 1153
Table 8.20 – Percent Change of the Goals for Target Modification Experiments.
Goals ZR exp. W exp. D exp. Wi exp. t exp. Cost exp.
ZR (%) 0.0 0.0 0.1 -2.2 0.1 -0.9
W (%) 0.0 0.0 0.0 -1.4 0.0 0.3
D (%) 0.6 0.3 0.7 0.3 1.2 -12.9
Wi (%) 0.2 0.1 0.0 16.0 0.0 -0.6
t (%) -0.1 0.1 0.0 0.0 0.0 -0.1
Cost (%) 0.0 0.0 0.0 0.0 0.0 27.7
Z-rotation, weight and thickness have no improvement by reducing their target

range. Thickness is already at the ideal value for nominal targets and hence cannot
improve further. Apparently z-rotation and weight have trade-off with other goals that do
not let them improve. This is evident by observing the corresponding rows. The values
of ZR and W never improved in all the experiments. Even the deterioration of these
variables is not significant. The improvement in diameter goal with reducing its target is
very small (0.7%). Width and cost have an improvement of 16% and 28% respectively.
These are high values indicating positive effect on these goals by reducing their target
325
range. Width is improved by worsening z-rotation and weight goals. Cost is improved
by worsening diameter and draft angle (shown in Table 8.21) goals. In fact, cost is
reduced by reducing the required number of third mold pieces from 3 to 1. This is
achieved by increasing the draft angle of the corresponding boss feature and reducing the
part thickness (by increasing diameter).
The variation in the values of the other system variables (NR, NC, θ1, θ2, θ3, CT, TA,
HOC and FOC) is not significant. Their values for different experiments are presented in
Table 8.21. One thing to be noticed for different experiments is the value of θ3, which is
the draft angle that corresponds to the boss feature in the third mold piece. It increases
from 1.0o to 1.7o when the target range for cost goal is reduced. As explained earlier,
increasing draft angle results in fewer mold pieces and hence lower cost.
Table 8.21 - Values of System Variables for Target Modification Experiments.

Current
Goals Solution ZR exp. W exp. D exp. Wi exp. t exp. Cost exp.
D (mm) 2.78 2.76 2.77 2.76 2.78 2.75 3.14
Wi (mm) 2.80 2.80 2.80 2.80 2.48 2.80 2.81
t (mm) 1.00 1.00 1.00 1.00 1.00 1.00 1.00
NR 2 2 2 2 2 2 2
NC 2 2 2 2 2 2 2
θ1 ( )
o
1.3 1.3 1.3 1.3 1.3 1.3 1.3
θ2 ( )
o
1.2 1.2 1.2 1.2 1.4 1.2 1.2
θ3 ( )
o
1.0 1.0 1.0 1.0 1.0 1.0 1.7
TA 0 0 0 0 0 0 0
CT (sec.) 300 300 300 300 300 300 300
HOC1-6 (inch) 0.002 0.002 0.002 0.002 0.002 0.002 0.002
FOC1-6 (inch) 0.012 0.012 0.012 0.012 0.012 0.012 0.012
The solutions of the target range reduction experiments indicate that the goals do not
worsen (if not improve) by reducing their target range. This is an expected behavior and
hence provides more faith in the correctness of the algorithm and analysis functions (used
in OptdesX) as well as the actual values of the obtained solution.
In the next section, physical experiments are presented which further validate the
solution of the problem.
8.6 PHYSICAL VALIDATION
The RTMDS was used to generate the mold pieces for the tailored part design
(Figure 8.23.a). The running process is the same as the mold design process for the
original camera roller design (Section 8.2). Therefore it is not repeated here. The
generated mold design is shown in Figure 8.23.b. Compared them with the mold pieces
in Figure 8.9, it is expected that the mold pieces for the tailor part will have a much better
mold life than the mold pieces for the original part. The designed mold pieces were
fabricated on SLA 3500 with SL 7510 resin using the mold orientations, slicing schemes
and the process parameters obtained by solving the MPGT problem. The molds are post
326
processed in the usual manner (cleaning with TPM, water and alcohol, drying with air,
post-curing in UV oven for an hour and thermal curing for 2 hours).
(a) Tailored Camera Roller (Two Views)
Mold Cavity Mold Core
Third Mold Piece
(b) Mold Piece for the tailored Camera Roller
Figure 8.23 – Mold Design for the Tailored Camera Roller Part.
Placing the mold pieces in the mold base of the Morgan Press machine was
challenging because the three mold pieces are assembled and dissembled manually. The
mold core and mold cavity were fixed in the corresponding mold halves and the third
mold piece was placed in the bottom half without fixing it. This piece was fixed after
clamping the mold halves. The injection molding parameters used to mold the parts are
presented in Table 8.22. The ejection of the parts from the mold is in the reverse order.
The fixture of the third piece is loosed before the molds are opened. The ejection of the
327
parts from the mold core is also performed manually. After removing the mold core with
the part from the mold base, the third piece is ejected first and the part is ejected from the
mold core with the aid of four ejector pins. A photo of the physical mold pieces and the
parts fabricated from them are shown in Figure 8.24.
Table 8.22 - Injection Molding Parameters Used for the Camera Roller.
Parameter Value Units
Clamping Force 11 Tons
Injection Pressure 2 Mpsi
Pilot Valve Pressure 50 psi
0
Plastic Temp. in barrel 430 F
0
Plastic Temp. in nozzle 450 F
Injection Time 10 Second
Cooling Time 300 Second
Release Agent Used No units
Figure 8.24 – Physical Mold Pieces and Injection Molded Camera Roller.
In the experiment, it is observed that the mold pieces for the tailored part design are
much stronger than the mold pieces shown in Figure 8.9. All the features in the mold
pieces are still intact after three shots (it was ruined by a mistake made by the author).
This also verifies the correctness of the solutions obtained from the MPGT problem.
328
8.7 EVALUATION OF CAMERA ROLLER CASE – POST DFRTS

With the aid of the DFRTS, the material-process geometric tailoring (MPGT) was
successfully performed for the robot arm (Section 8.5). To further understand the camera
roller case, some comparing experiments are performed and presented here.
When putting the original design and the tailored design together (Figure 8.25), one
can obviously see that the geometric tailoring tremendously reduced the fabrication
difficulties in the Rapid Tooling process for the camera roller. A simple calculation is
provided here to illustrate the necessity of geometric tailoring for the case.
(a) Before Tailoring
(b) After Tailoring
Figure 8.25 – Comparison the Camera Roller before and after Geometric Tailoring.
As estimated by the mold life predictor, the mold life of mold core, mold cavity and
the third piece is 9, 4, and 16 for the original design (Section 8.3). Therefore the required
number of mold core, mold cavity and the third piece is 6, 13, and 4 respectively. In the
solution of the MPGT problem, the cost of rapid tooling process is $1291 ($1091 for SLA
fabrication process and $200 for injection molding process). It should be noticed that this
cost is for one set of mold pieces (1 of mold core, 1 of more cavity and 3 of the third
piece) for the tailored part. Based on our experience, the cost of the SLA fabrication
process for 6 of mold core, 13 of mold cavity and 4 of the third piece for the original part
would be 5 ~ 7 times of $1091, which is $5455~7637. After a mold piece is ruined, a
new mold piece needs to replace the old one, which leads to a cost for assembling and
disassembling the mold piece. More frequent replacements leads to higher cost and
longer time. Therefore it is estimated that the cost of Rapid Tooling process for the
original design would be around $6500. Compared to the cost of $1291 for the tailored
design, it is significantly decreased. Suppose the designer’s budget is $2500, the
manufacturer will have a profit of $1209 instead of a loss of $4000.
Another comparing experiments are performed for the modified camera roller part
(Figure 8.14). Similar to the experiment given in Section 7.6, the sequential and
329
concurrent solution processes are tested in the experiments. First in the sequential
experiment, the MGT problem is formulated and solved. For the solution of the MGT
problem, the manufacturer then performs the process planning of the Rapid Tooling. In
the concurrent experiment, the MPGT problem is formulated and solved. Since the cost
factor is important here, the cost experiment (reducing the target range of cost by 40%) is
adopted (Table 8.18). The solutions (values of geometry) for the experiments are
presented in Table 8.23 and the corresponding goal achievements are presented in Table
8.24. The target value of each goal is also listed in the table.
Table 8.23 - Solutions of Sequential and Concurrent Solution Process.

D Wi t
θ1 ( ) θ2 ( ) θ3 ( )
o o o
Variables NR NC
(mm) (mm) (mm)
Sequential 2.75 2.8 1.00 2 2 1.3 1.2 1.4
Concurrent 3.14 2.81 1.00 2 2 1.3 1.2 1.7
Table 8.24 - Goals Achievements for Sequential and Concurrent Solution Processes.
Goals ZR (rad.) W (g) Cost ($) # Mold1 # Mold2 # Mold3 Obj. Function
Sequential 0.01043 2.399 1221.66 1 1 2 0.0365
Concurrent 0.01029 2.394 1152.5 1 1 1 0.0408
Target 0.01310 2.28 500 - - - 0.0
Comparing the results of the sequential and concurrent solution processes, it is as

expected that the sequential process has a better solution of design variables (D) and
design goal achievements (Z-Rotation). However, its cost and the total objective function
are higher than those of the solutions obtained from the concurrent process. After further
analysis, we can see the increase of cost is mainly because the required number of the
third mold piece is increased from 1 to 2. The main dimensions of the boss in the third
mold piece are D and θ3 as shown in Figure 8.26. The mold life of the mold piece is
determined by their values. In the concurrent solution process, D and θ3 are formulated
and solved together. Therefore the values of D and θ3 can all be increased to reduce the
θ3
Figure 8.26 – The Dimensions of the Third Mold Piece.
330
required mold number. However, in the sequential solution process, D has already been
fixed in the former process. The manufacturer can only change the value of θ3. If the
design freedom of θ3 is not sufficient to make the required mold number as 1, an
additional mold piece will be required. Correspondingly the value of θ3 can be reduced
for the required mold number (2). As the objective function value of the concurrent
solution process is smaller that of the sequential solution process, it verifies that the
concurrent formulation and solving processes utilized in the DFRTS is better than the
current sequential solving process.
The experiments also verify that the mapping from design space to the
manufacturing space may have discontinuities. That is, for the continuous variables D
and θ3, some design spaces correspond to required mold piece number 1. However other
design spaces may require two mold pieces. Consequently the fabrication cost will be
rather different for them. Considering the discontinuity of the design-for-manufacture
(DFM), it is generally desired to transfer more design freedom to the manufacturer.
In the next section, a brief summary is given for discussing the relevance of these
results with regard to the hypotheses of the dissertation.
In this chapter, the RTMDS and DFRTS were applied to determine the mold design
and geometric tailoring for a camera roller design. A problem of producing functional
prototypes of a camera roller was first introduced in Section 8.1. By using the RTMDS, a
mold design with three mold pieces was generated for the camera roller (Section 8.2). A
standard mold base for Morgan Press injection molding machine was used in the mold
piece construction. A preliminary study and experiment were performed which results in
a modified camera roller design (Section 8.3). This modified design was considered in
the DFRTS. The formulation of the geometric tailoring problem for the modified camera
roller was presented in Section 8.4. In the formulation the design functions were
represented by response surface models. The solution process of the MPGT problem,
which consists of three stages, was described in Section 8.5. The modified MPGT
problems that are related to 24 slicing schemes were solved by OptdesX. The results of
physical experiments for the validation of the problem solution were presented in Section
8.6. Finally an evaluation of sequential and concurrent solution processes for the camera
roller case was provided in Section 8.7.
Similar steps are utilized in the camera roller example as those in the robot arm
example. This indicates the generality of the approaches developed in the RTMDS and
DFRTS. Hence, this example extends the faith in the developed approaches from the
particular examples presented in this thesis to other problems that fall under this category
(theoretical performance validation). In the discussion of application of RTMDS and
DFRTS to the camera roller example, the relationships with the hypotheses were
identified. The results from testing the hypotheses are summarized below (Figure 8.27):
Hypothesis 1: A mold design with three mold pieces was generated for the camera roller
by using the Rapid Tooling Mold Design system, which is developed based on the
Multi-piece Mold Design Method. The whole running process only took several
minutes. The summary of testing the sub-hypotheses (H1.1 ~ H1.3) presented below
will provide more details.
331
Hypothesis 1
Mold designs of
The requirements on mold
camera roller are
design of the camera roller
generated by the
are representative of the
RTMDS in acceptable
mold design problems.
time, and they can be
The system can be used to
used to produce
generate mold design for
prototpyes in the Rapid
the camera roller.
Tooling process.
Empirical Empirical
Structural Performance Hypothesis 2
The requirements on Solutions of geometric

geometric tailoring of the tailoring of camera roller
camera roller are are generated by the
representative of the DFRTS in acceptable time,
geometric tailoring problems. and tailored part design
The system can be used to and process parameters
perform geometric tailoring can be used to produce
for the camera roller. functional camera roller in
Empirical Empirical
Figure 8.27 – Empirical Structural and Performance Validation for H1 and H2.
Hypothesis 1.1: 446 concave edges and 336 concave faces are identified. Based on the
identified concave faces, 32 concave regions are generated for the camera roller. The
region generation process only took seconds for the camera roller.
Hypothesis 1.2: The generated concave regions of the camera roller are combined into 3
concave regions. The region combination process only took less than one minute for
the camera roller.
Hypothesis 1.3: Mold pieces are constructed for a standard mold base which can be
installed in a Morgan Press injection molding machine. The mold piece construction
process only took less than one minute for the camera roller.
Hypothesis 2: Tailored part design with mold design variables, RP process variables and
IJM process variables were generated for the camera roller by using the Design for
Rapid Tooling system, which integrates geometric tailoring and process planning by
the manufacturer. The satisfactory prototypes are produced without any iteration for
the camera roller. The summary of testing the sub-hypotheses (H2.1 ~ H2.3)
presented below will provide more details.
Hypothesis 2.1: For the camera roller the designer formulated a partial MPGT problem
based on the MPGT decision template. The manufacturer formulated a complete
MPGT problem which considered the material difference of products and prototypes.
Hypothesis 2.2: The designer’s MPGT problem formulation, RP process planning
formulation, and IJM process planning formulation were integrated into a completer
MPGT problem formulation for the camera roller.
Hypothesis 2.3: The MPGT problem was solved in a three-stage solution process. The
solution was reasonable and validated by physical experiments.
332
In the next and final chapter, contributions form this work will be presented along
with a critical review of the research and future work (Figure 8.28).
Chapter 9: Achievements and Recommendations
Chp 7: Prototypes of a Chp 8: Prototypes of a

Robot Arm Camera Roller
Figure 8.28 – Preview of Chapter 9.
333
Chapter 9 – Achievements and Recommendations
CHAPTER 9
ACHIEVEMENTS AND RECOMMENDATIONS
In this dissertation, a mold design system and a design-for-manufacture system have

been developed, presented, and tested to facilitate producing function prototypes for a
part design. The development and presentation of these systems is brought to a close in
this chapter. In Section 9.1, closure is sought by returning to the research questions
posed in Chapter 1 and reviewing the answers that have been offered. The resulting
contributions are then summarized in Section 9.2. Limitations of the research are
discussed in Section 9.3, and possible avenues of future work are described in Section
9.4.
334
9.1 ANSWERING THE RESEARCH QUESTIONS

In this section focus is returned to the research questions justified through the
discussions of Chapter 1 and formulated specifically in Section 1.2.1. In essence, this
section is a review of hypothesis testing presented throughout the dissertation, as the
hypotheses offered in direct support of the research questions. The relationship between
the research questions, and the hypotheses formulated are shown in Figure 9.1. Each
hypothesis has been supported through the investigations of the previous chapters.
9.1.1 Research Question Overview

As stated in the introduction to Chapter 1, the main objective of this research is to
reduce the time for functional prototypes in the product realization process. The key
research question being investigated is:
How to reduce the lead-time in the usage of Rapid Tooling to produce functional
prototypes for a part design?
The key question directly ties to the principal technology development goal for this
research project (RTTB): to enable “customer” of the rapid tooling testbed receive
nearly production-representative components in a variety of polymer, ceramic, and metal
materials within 3-4 days of submitting a product model. In this research, the focus is on
Primary Research Question
How to reduce the lead-time in the usage of Rapid
Tooling to produce functional prototypes for a part
design?
Q1. How to aid the mold Q2. How to reduce the time of
designer to reduce the mold iteration between the designer and
design time ? manufacturer ?
H1. Multi-Piece Mold Design Method provides a method H2. Geometric tailoring can be integrated with
to automate several key steps of the mold design. process planning and solved by the manufacturer.
Q1.1 What are Q1.2 How to Q1.3 How to Q2.1 How to Q2.2 How to Q2.3 How to
appropriate generate mold generate mold reduce the formulate the solve the
basic elements configurations pieces from a iterations design for Rapid design for
to automate based on the mold configuration because of Tooling Rapid Tooling
mold design? basic elements? design? different material problem? problem?
properties?
H1.3 The H2.1 Material H2.2 A three-

H1.2 The region Reverse Glue H2.2
H1.1 Concave Geometric stage solution
based combining Mold Construction Formulation of
region and Tailoring process for a
process and Method and design for Rapid
convex face. Decision satisficing
algorithms. algorithms. Tooling system
Template. solution.
Figure 9.1 – Context for Research Questions and Hypotheses.
335
the mold design and design-for-manufacture for Rapid Tooling. The idea is to reduce the
time of several time-consuming mold design steps and the number of iterations between
the designer and manufacturer in Rapid Tooling. Towards achieving this objective, a
Rapid Tooling Mold Design System (RTMDS) and Design for Rapid Tooling System
(DFRTS) are developed.
The key question has two related questions associated with it:
Q1. How to aid the mold designer to reduce the mold design time for a wide variety of
part geometries in the design for a Rapid Tooling process?
Q2. How to reduce the time of iteration between the designer and manufacturer in the
usage of Rapid Tooling for a wide variation of design requirements?
To address these questions, research hypotheses are proposed in support of achieving
the principal objective for the dissertation.
H1 Multi-Piece Mold Design Method provides a method to automate several key
steps of the mold design process, which can greatly reduce the mold design time
for a wide variety of part geometries.
H2 Geometric tailoring for Rapid Tooling can be integrated with process planning
based on decision templates and solved by the manufacturer, which can reduce
the time of iteration between the designer and manufacturer.
Their elaboration and verification have provided the context in which the research
work has proceeded. This work has been based in computational geometry and decision-
based design. The sub-questions associated with these two questions are:
Q1.1 What are appropriate basic elements to automate several mold design steps for a
wide variety of part geometries?
Q1.2 How to generate mold configurations by a systematic design process based on the
basic elements?
Q1.3 How to generate mold pieces from a given mold base effectively and efficiently
according to a mold configuration design?
Q2.1 How to reduce the iterations between the designer and manufacturer in producing
functional prototypes that have different material properties from products?
Q2.2 How to formulate the design for Rapid Tooling problem which integrates
Q2.3 How to solve the design for Rapid Tooling problem effectively and efficiently?
To facilitate answering these questions, each has hypothesis associated with them:
H1.1 Concave region and convex face are two kinds of basic elements that provide an
efficient and effective approach for exploring and developing molds design
method for Rapid Tooling.
336
H1.2 The Region based combining process and related algorithms provide a systematic
method to generate mold configurations of multi-piece molds design.
H1.3 The Reverse Glue Mold Construction Method and related algorithms provide an
efficient and effective method for constructing multi-piece mold designs for
Rapid Tooling.
H2.1 The designer can initiate a material geometric tailoring (MGT) formulation based
on a MGT decision template; therefore the manufacturer, who completes and
solves the MGT problem, can produce production-representative prototypes more
quickly.
H2.2 The design for Rapid Tooling problem can be formulated by several compromise
DSPs and tasks, which can then be integrated into a design for Rapid Tooling
system (DFRTS).
H2.3 A three-stage solution process can be utilized to get a satisficing solution
effectively and efficiently based on design and manufacturing models and
continual/discrete variables.
With an overview of the research questions and hypotheses presented, answers to
these research questions are summarized in the next section. As mentioned the research
questions are answered through testing the hypotheses.
9.1.2 Answering Research Questions

In general a research question is answered when the corresponding hypothesis is
validated. Hence, in order to answer the fundamental question, we validate each of the
supporting hypotheses. This is done according to the process given in Section 1.3.2,
where each of the ‘validities’ in the Validity Square (Figure 9.2) is used to test each
supporting hypotheses separately. Then, by adding necessary additional information, the
validity of the fundamental hypothesis – i.e., the validity of the RTMDS and DFRTS – is
asserted through induction.
THEORETICAL THEORETICAL
STRUCTURAL PERFORMANCE
VALIDITY VALIDITY
EMPIRICAL EMPIRICAL
STRUCTURAL PERFORMANCE
VALIDITY VALIDITY
Figure 9.2 – Validation Square.

Answering the research questions is performed following the hierarchy of the
questions presented in Figure 9.1. First the two questions are answered using the answers
337
of the supporting questions. These two questions are then used to answer the key
question.
• Answer to Question 1:
This research question has three supporting questions, answer to these questions are
summarized as follows:
Answer to Basic Element Question (1.1)
By identifying concave region, combined region and convex face, this work extends
the basic elements used in other works, which are pockets (Chen, et al., 1993; Weinstein
and Manoochehri, 1996) and features (Fu, et al., 1999; Gu, et al., 1999). The basic
elements of regions are more general for different geometries. Therefore more parts can
be handled by the developed mold design method. The theoretical structural validity was
discussed in Section 3.3 through mathematical definition and proofs related with it.
Empirical structural validity and performance validity was presented using the illustrative
examples (Section 4.4 and 4.5) and case studies (Chapter 8 and 9). In all cases the
concave regions and convex faces provided basic elements for mold configuration design.
Answer to Region Combination Question (1.2)
In this research, the mold configuration design is transferred to a region-growing
process, which captures the connectivity of faces and the demoldability of regions. This
transferring enables the automation of the generation of parting directions and parting
lines for mold configuration design. The mold design knowledge can also be added to
the region combination process easily. The theoretical structural validity was discussed
in Section 3.4 through mathematical definition and relations related with it. Empirical
structural validity and performance validity was presented using the illustrative examples
(Section 4.4 and 4.5) and case studies (Chapter 7 and 8). In all cases combined regions
were generated which provided mold configuration design for mold piece construction.
Answer to Mold Piece Construction Question (1.3)
Gluing operation is a standard high-level Euler operation for half-edge data structure.
A novel approach based on reversing the gluing operation was developed in this research
and used in the mold piece construction. The mold piece construction problem is then
transferred to a glue face generation problem. It is illustrated in this dissertation that the
glue face generation is actually a geometric reconstruction problem. The theoretical
structural validity was discussed in Section 3.5 through mathematical definition and
relations related with it. Empirical structural validity and performance validity was
presented using the illustrative examples (Section 4.4 and 4.5) and case studies (Chapter
7 and 8). In all cases mold pieces were generated for mold configuration designs within
satisfied time.
Answer to Mold Design Question
Answering the research sub-questions related to Hypothesis 1 provides a foundation
for answering the high-level question. In this research computation geometry and solid
modeling have been utilized for multi-piece mold design for Rapid Tooling. By using the
RTMDS to automate several steps of the mold design, mold pieces are generated in a
short time. Other mold design components (e.g. gate, runner, ejector pin holes) can be
338
added to the generated mold pieces. Consequently the total mold design time is
dramatically decreased.
In Section 3.3 the relationships with the different steps of the mold design method
were discussed which validated the structural (theoretical and empirical) validity of the
method. The application of the RTMDS was demonstrated in illustrative examples
(Section 4.4 and 4.5) and case studies (Chapter 7 and 8). The mold design results verify
the usefulness of the mold design method and the related system in designing molds for a
set of products. The mold designs are further verified by producing physical prototypes
with the molds. Although used for a limited number of examples, the obtained results
indicate that the mold design method presented in this dissertation is applicable for other
problems.
• Answer to Question 2
The second research question also has three related questions associated with it,
answers to these three questions are first provided because they will help answer
Question 2, which is related to geometric tailoring problem.
Answer to Material Difference Question (2.1)
In different Rapid Tooling process, the material difference between production and
prototypes are normal. The geometric tailoring, which is based on Buckingham π
theorem, was utilized to transfer the material difference problem to a process of making
decisions on unimportant system variables (Section 5.2). Then a decision template in the
compromise DSP formulation is proposed for integrating the design and manufacturing
requirements (Section 5.3). The Buckingham Π theorem and the compromise DSP
provide the theoretical structural validity of material geometric tailoring and its decision
template. Empirical structural validity and performance validity was presented using
three illustrative examples (Section 5.5) and two case studies (Chapter 7 and 8). In all the
cases both design and manufacturing requirements were satisfied without iterations
between the designer and manufacturer.
Answer to Problem Formulation Question (2.2)
In this dissertation decisions on design and process variables are used to model the
design-for-manufacture problem for different viewpoints. The decisions in the three
steps of the Rapid Tooling process (material-process geometric tailoring, RP process
planning and IJM process planning) can be formulated as compromise-DSPs, and mold
design can be considered as a task. Therefore a design for rapid tooling system was
presented in which the decisions on different steps are integrated into a single
compromise-DSP (Section 6.3). The theoretical structural validity was discussed in
Section 6.1 through different modules and their relations. Empirical structural validity
and performance validity was presented using a robot arm and a camera roller (Chapter 7
and 8). In the case studies formulations that were considered in the DFRTS were
presented, which provided a unified decision framework for solving the geometric
tailoring problem.
Answer to Problem Solving Question (2.2)
By identifying coupling between goals and variables, this work presented a three-
stage solution process for solving the material-process geometric tailoring problem. A
339
satisficing solution is generated instead of an optimal solution because of the complexity

of the problem. The theoretical structural validity was discussed in Section 6.4 through
different modules and their relations. Empirical structural validity and performance
validity was presented using two case studies (Chapter 7 and 8). In all cases following
the solution process of the DFRTS generated satisfied solutions of the geometric tailoring
problem, which were further validated by physical experiments.
Answer to Geometric Tailoring Question
Answering the research sub-questions, associated with Hypothesis 2, provide a
foundation that can be used to answer the overall geometric tailoring question. The
iterations between the designer and manufacturer are actually a process of integrating
design and manufacturing requirements. Geometric tailoring is also a method to integrate
design and manufacturing requirements. However much less time is needed because
computers instead of negotiations between the designer and manufacturer solve the
tradeoffs between the requirements. Therefore a satisficing solution can be found more
quickly.
The application of the DFRTS was demonstrated in two case studies (Chapter 7 and
8). Solutions were generated by the system based on the design and manufacturing
requirements. Physical prototypes were produced for the solutions. These results verify
the usefulness of the geometric tailoring and the Design for Rapid Tooling System in
producing functional prototypes for a set of products.
• Answering Key Question
Answer to the key question proposed for this work is embodied by the Rapid Tooling
Mold Design System and Design for Rapid Tooling System, which reduce the time of
mold design and iterations between the designer and manufacturer respectively. The two
systems are important components of the Rapid Tooling Testbed, which uses Rapid
Prototyping technology in the product tooling design and manufacture. Several case
studies (including the robot arm and camera roller in Chapter 7 and 8) were tested in the
RTTB. The results affirmed that the goal of producing production-representative
components in a variety of polymer, ceramic, and metal materials within 3-4 days of
submitting a product model is feasible. The experiments also verify the importance of
the research questions addressed in this dissertation, that is, how to reduce the time of
mold design and how to avoid iterations between designer and manufacturer.
Producing functional prototypes for a part design within days is always desired to
compress the product development time and cost. Today’s competitive and highly
volatile market makes this desire much stronger. This research is an effort toward
achieving this goal. With this in mind, a summary of the research contributions is offered
in the next section.
9.2 ACHIEVEMENTS: REVIEW OF RESEARCH CONTRIBUTIONS
The contributions offered in this dissertation are introduced in Section 1.3.3 and
realized throughout the dissertation. The list provided in Chapter 1 was partial and
identified only the high level contributions. In this section contributions from this work
will be revisited to highlight the significance of this work.
340
The primary contribution from this work is a systematic method and a set of
algorithms for multi-piece mold design and a systematic way (formulations and a
solution process) to integrate design and manufacturing requirements. They are
embodied in the Rapid Tooling Mold Design System and the Design for Rapid Tooling
System respectively. The methods are tested by several industrial cases via the
developed systems. The contributions can be categorized in three areas.
Contributions related to Computer-aided Mold Design (H1 and H1.1~1.3):
• A problem definition for multi-piece mold design is proposed as the foundation of
a computer-aided mold design method. Compared to other representative problem
definitions, the problem definition presented in this research added the face
connectivity as an important factor in the mold design (Section 3.2).
• A solution process for the formulated problem definition is proposed (Section
3.3). As an additional requirement, face connectivity, is considered, the mold
design process is more complicated. Another important contribution of this
research is identifying a systematic process to generate mold design for the
problem definition.
• The notions of concave region and convex face in mold configuration design and
a means of generating them for a given part. As an extension of “pocket”, which
is a common notion in the computer-aided mold design, concave region can
handle more general cases (Section 3.4). An algorithm, which is based on several
lemmas, is also presented to generate concave regions for a part effectively.
• A combining approach to generate regions from the basic elements based on
linear programming and mold design knowledge (Section 3.5). In judging
feasible parting direction of several faces, a simple yet efficient approach based
on linear programming is proposed. It enables more combinations of regions and
faces are explored because of its fast solution process. An algorithm is developed
which allows more mold design knowledge considered.
• A mold construction method based on the reverse glue operation and a geometric
reconstruction problem for any mold base. A novel mold construction approach
based on reverse gluing operation is proposed in this research, which is much
more efficient than the current industrial approaches (Section 3.6). Also it is
illustrated in this research that the mold construction problem is actually a
geometric reconstruction problem. Therefore computer-aided mold design
methods may find theoretical basis in the area of computational geometry.
• A set of examples and industrial cases are valuable for the future research in
computer-aided mold design area. In this research the author constructs several
example parts to test different aspects of a mold design method (Section 4.4).
Several representative industrial cases are also utilized in the dissertation (Section
4.5, 7.2 and 8.2). They provide a basis to compare different mold design
approaches.
341
Contributions related to Design-for-Manufacture (H2 and H2.1 ~2.3):

• A demonstration of transferring part of DFM responsibility (geometric tailoring)
from the designer to the manufacturer is feasible, at least for Rapid Tooling.
Traditionally the designer takes all the responsibility of DFM. This was taken for
granted by lots of researchers for a long time. However from the perspective of
information integration, design-for-manufacture is actually a process in which
design and manufacturing requirements are integrated and tradeoffs are made on
the design and manufacturing goals and variables. Therefore it is not necessarily
to let the designer have all the DFM responsibility. This research demonstrates
that by given some design freedom to the manufacturer, the cost/time of
fabrication process may be reduced (Section 5.5 and Chapter 8).
• Geometric tailoring is identified as part of DFM tasks that may be transferred
from the designer to the manufacturer in the process of producing functional
prototypes. The definition and category of geometric tailoring are presented with
its fundamentals (Section 5.2).
• Decision templates based on the compromise DSP are proposed as the digital
interface to transform design information to the manufacturer for geometric
tailoring problem. Their methodology is identified based on design freedom
(Section 5.3). Two kinds of decision templates are formulated in this research and
are applied to a set of examples (Section 5.5 and 7.3).
• An integrated design for Rapid Tooling system is developed based on Decision-
Based Design. The DFRTS is a collaborative system which can handle conflicts
between designers and manufacturers. A problem formulation which considers
both the design and manufacturing requirements is presented. This research
demonstrates that a concurrent approach to formulate and solve problem may
result in a better solution than a sequential approach (Section 6.5 and 8.6).
• A solution strategy and related solution process are proposed for the integrated
problem formulation. The solution process is systematic and the same three
stages can be followed to obtain solutions for different parts (Section 7.4 and 8.5).
Contributions related to Rapid Tooling (H1 and H2):
• The Rapid Tooling Mold Design System is a useful tool in designing molds for a
part design. The implementation of the mold design method into a convenient
tool is another kind of contributions of this research.
• The Design for Rapid Tooling System, which integrates several software
packages, is developed to aid the manufacturer to produce functional prototypes
by the direct AIM tooling. In this research, different software systems are used in
solving the geometric tailoring problem, which reduces the time in achieving
satisfied prototypes.
• The developed systems are important components of the Rapid Tooling TestBed.
The RTTB is an experimental testbed that supports exploration of design and
design-for-manufacture issues related to rapid prototyping and rapid tooling for
injection molding (Section 1.2.2). The efforts in producing prototypes within 3-4
342
days of submitting a product model will significantly reduce the product

realization time and cost.
• Several industrial cases are investigated, which demonstrates the advantage and
disadvantage of the direct AIM tooling. As is a new technique, the direction AIM
tooling has only limited case studies published in literature. In this research
prototypes of a ring gear (Section 5.5.3) and a camera roller (Section 8.5) were
produced. These studies illustrated that the direction AIM tooling can be used for
rather complicated part. However its limited mold life should be considered in
selecting a fabrication process.
Anticipated Impact:
In summary, the author expects this research will have two impacts.
(1) It is expected that this research will enhance our formal understanding of the
basic computational issues that lie behind the computer-aided mold design
problem. In this way, improved and more rigorous mold design systems can be
developed and used in the injection molding industry.
(2) Design-for-Manufacture (DFM) is actually a process to integrate both design and
manufacturing requirements. From this perspective, there is no particular reason
for the designer to play a dominant role in performing DFM. A different
approach is presented in this research, which, in the author’s hope, will enhance
our understanding of the basic strategies of DFM. It is believed that further
research in this direction can enhance our understanding of the advantages and
disadvantages of different DFM approaches.
After the contributions of this research were highlighted in this section, limitations of
the research will be identified and discussed in the next section.
9.3 CRITICAL ANALYSIS: LIMITATIONS OF THE RESEARCH
With the research questions answered and the contributions of this work presented,
in this section a critical evaluation of this work is performed, with the focus on the
limitations of RTMDS and DFRTS and their different elements.
RTMDS
The mold design method developed in this research only considers planar surfaces.
Although quadric and parametric surfaces can be approximated by a series of planar
surfaces, the face number and file size may increase dramatically. More importantly, it
may bring problems in the integration of the RTMDS with other software systems. For
example, the RP process planner developed by West (1999) checks all directions of
planar surfaces of a part in selecting a building orientation in a SLA machine. Therefore
it may take a long time for a mold piece generated by the RTMDS because too many
planar faces are generated in the approximation of a quadric surface.
The RTMDS can only handle a SAT file as the input. This may add a burden to the
mold designer in preparing the input file of a part. Also the transfer of a STL file to a
SAT file may bring in some errors in the CAD model. In the transferring, all planar faces
that are in the same surface are combined into one face in the SAT file. This may not be
343
desired for some parts (e.g. the camera roller in Section 8.2). Therefore the mold design
method needs to be extended to handle a part represented in STL file.
Some implementation limitations of the modules of the RTMDS are listed as
follows. The limitations of the region generation module include (1) the face splitting
algorithm is not general; (2) data structure of the system is not efficient which results in
large memory requirements. The limitations of the region combination module include
(1) limited more mold design knowledge is considered in the system; (2) the robustness
and accuracy of the combination process are different for parts; (3) the combining orders
of regions and faces are not changeable by users. The limitations of the mold piece
module include (1) the algorithm of judging parting surface for a region needs to be
refined; (2) the generation of inner glue faces can only handle regular cases; (3) the part
can not be positioned freely within the mold base.
DFRTS
In the DFRTS, much effort is required to model the manufacturing process before a
geometric tailoring problem can be solved. Currently some manufacturing requirements
for the direction AIM tooling have been formulated in the DFRTS (Section 6.1). They
can be reused for different parts. However, to add more requirements or process
variables of the direction AIM tooling (e.g. injection molding process variables besides
the cooling time), more process models are needed. To extend the DFRTS to other
processes and materials, their manufacturing models should also be built by experiments
or simulation. Related to this limitation is that as only quantitative equations can be
considered in the compromise DSP, some properties of a fabrication process or a design
function may not be able to be considered in the DFRTS if they cannot be formulated as
equations.
Geometric tailoring is mainly due to the material difference of products and
prototypes (Section 5.5), or the fabrication difficulties for a part design (Section 8.3).
Not surprisingly, if the material properties of products and prototypes are similar, the
material geometric tailoring will generate similar results as the original design. Similarly
if there is only little difficulty in producing a part design, the material-process geometric
tailoring will generate similar results as those of material geometric tailoring (Section
7.6). Since it takes much time and effort to couple the decisions in the geometric
tailoring problem, this effort may not be needed if the two results will be similar.
Therefore a criteria and a related judging method may be needed in the DFRTS to
determine if the geometric tailoring is necessary for a part design.
The dimensions in the decisions templates are related to a feature in CAD models.
Currently their relations are not recorded. In this research the author acts as the designer
and the manufacturer at the same time, hence this is not a concern in the current DFRTS.
However, in real applications, a mechanism to build and record the relation of
dimensions in decision templates and features in CAD models is necessary. The research
in feature representation and recognition may be incorporated in this development.
In the camera roller example (Chapter 8), the topology of the part is changed because
small dimension changes cannot satisfy the manufacturing requirements for the part. In
the DFRTS a systematical approach was developed for the dimensional changes.
However the topological changes are still based on the experience and knowledge of the
344
manufacturer. Therefore a systematic approach is desired for judging if the topology of a

part needs to be changed, and how is should be changed if it is necessary.
With some of the limitation of the RTMDS and DFRTS highlighted in this section,
recommendations for future work will be presented in the next section.
9.4 FUTURE WORK
The limitations of the RTMDS and DFRTS described in the last section can be used
to guide future research related to the systems. Some of the future research avenues are
described as follows.
Future Work in Computer-Aided Mold Design
Refining the RTMDS
As mentioned in Section 9.3, several refinements can be added in the current
implementation of the RTMDS. Some tasks are further described as follows.
! The mold design method can be extended to handle a CAD model represented in STL
format. As the concave regions are generated based on the convex and concave
property of two neighboring faces, a part represented in triangles can also be handled
by the region-based mold design method. The only task here is to generate topology
relation from a STL file. Research on the topology generation for STL file is rather
mature in computer graphics area.
! In the later mold design process, it is desired to move the part position in the mold
base based on the ejector pattern of the injection molding machine. Therefore in the
RTMDS, it is important to enable the user to position the part freely within the mold
base.
! Parting surface is a important mold design variable. The current algorithm of judging
parting surface for a region is rather simple, and can be improved to handle more
cases.
! The current data structure of the RTMDS is not efficient. By using pointer instead of
array, the memory requirements for large CAD models will be reduced.
Extending the RTMDS
! Shrinkage compensation is also an important consideration in the mold design,
especially for low-pressure powder injection molding (shrinkage can be as high as
14%). Based on the models developed in (Judson, 1999), it is possible to predict the
shrinkages and change the shape of the mold correspondingly. In ACIS, the
introduction of laws (mathematical functions) provides a powerful tool to model
complex geometries. By using different law classes, complex functions can be used
for creating curve and surface geometry.
! More design knowledge can be added to the multi-piece mold design. One possible
criterion is the ejection property of a mold piece. Considering the fragility of the
injection molded part in the low pressure power injection molding, more mold pieces
may be necessary for a complete part.
345
New Directions
! A geometric reconstruction problem, Problem GFG (Section 3.6.2), is defined in this
research. No approaches for this problem are found in literatures of computational
geometry. However it is an important problem in computer-aided mold design
research in order to construct mold pieces for a part (refer to Figure 4.12). In
RTMDS an algorithm is proposed for simple cases. Further research is necessary to
develop a theoretical basis and more advanced approach.
Future Work in Design-for-Manufacture
Refining the DFRTS
! It is desired to formulate the updated research results on the direct AIM tooling and
incorporate them into the DFRTS.
! To further testing the idea of geometric tailoring, it is desired to test the DFRTS with
different fabrication process (e.g. low-pressure powder injection molding) and
different materials.
! The automation levels in utilizing different software systems of the DFRTS need to
be increased in order to handle different parts more efficiently.
Extending the DFRTS
! A mechanism to build and record the relation of dimensions in decision templates and
features in CAD models needs to be developed. As mentioned before, a decision
template can be linked with a parametric feature-based CAD model which is similar
to behavioral modeling (Section 2.4). Therefore the intent and performance of the
design are an integral part of the product model. Their values can also reference
external applications as part of the product model definition.
! A systematic approach is desired for judging if the topology of a part needs to be
changed, and how is should be changed if it is necessary. The task would be rather
challenging considering the representation and manipulation of the topology are all
much more difficult than those of dimensions. However research on graph theory
may be useful.
! In the DFRTS, a criteria and a related judging method can be developed to determine
if the geometric tailoring is necessary for a part design. It would save the efforts of
the designer and manufacturer in formulating and solving the geometric tailoring
problem.
New Directions
! Another significance of transferring the responsibility of geometric tailoring from the
designer to the manufacturer is that it enables the designer to consider several
fabrication processes at the same time. It is well known that different prototyping
processes have their advantages and disadvantages. Typically no single process
meets all the requirements of a part. Therefore it is difficult for the designer to
consider several fabrication processes and perform design-for-manufacture tasks for
them. Instead it is feasible to develop a distributed system (Figure 9.3). With the aid
of decision templates, the designer formulates the design requirements and sends
them to a process broker. The process broker can then select several candidate
processes and send the manufacturers the design requirements. Based on the design
346
requirements and the property of a fabrication processes, the manufacturer performs

the geometric tailoring and sends the results back to the process broker. The process
broker can aid the designer in selecting a process based on the tailored part design.
Therefore more fabrication processes are considered for a par design.
Rapid
4.Tailored Part Design
& Goal Achievements
Tooling
1. CA
D Mode
Services
Requir l of part & 3. Geometric
emen 2. CA
ts D Mode Tailoring
requir l of part &
emen
ts
5. Recomendation
Designer
Manufacturer
Internet 3. Geometric
Tailoring
Process Manufacturer
Broker Internet
3. Geometric
Tailoring
Manufacturer
Figure 9.3 – An Ideal System Working Between the Designer and Manufacturer.
347
Appendix A
APPENDIX A
RTMDS IMPLEMENTATION
In this appendix the files and classes are summarized for the RTMDS implementation.
The RTMDS was implemented in Microsoft Visual C++ for Windows NT on personal
computers. ACIS geometric modeler and LINGO are also used in the system
implementation.
• Appendix A.1 contains the code files and simple descriptions of their
functions.
• Appendix A.2 contains the main classes of the RTMDS and related
descriptions of their functions.
• Appendix A.3 contains the implementation information based on ACIS
geometric modeler.
• Appendix A.4 contains the implementation information based on LINGO
software system.
348
Appendix A
A.1 Code Files of RTMDS

The RTMDS was written in C++ using Microsoft Wizard and the Microsoft Foundation
Classes (MFC). The codes include C++ source (.CPP) files, header (.H) files, resource
(.RC) files, and a project (.DSP) file. They are described with their source code size as
follows:
• Source (.CPP) Files:

RTMold.cpp – The main program source file. It also contains the initialization and
deletion of ACIS and LINGO libraries. (9 KBytes)
MainFrm.cpp – Derive the CmainFrame class, which handles the creation of toolbar
buttons and the status bar. (5 KBytes)
RTMoldDoc.cpp – Derive and implement the document class, which initialize and
serialize a document. This is also the main file which contains the functions of
mold design modules that may change the context of a CAD model. (52 KBytes)
RTMoldView.cpp – Derive and implement the view class, which display and print the
document data. (13 KBytes)
Attrib_Connect_Region.cpp – Implement of an attribute (connected region
properties) attached to an entity of an ACIS model. (3 KBytes)
Attrib_PEdge.cpp – Implement of an attribute (parting edges) attached to an entity of
an ACIS model. (3 KBytes)
Attrib_Region.cpp – Implement of an attribute (region properties) attached to an
entity of a ACIS model. (3 KBytes)
Attrib_Rtmold.cpp – Implement of a basic attribute used for deriving other attributes.
(1 KBytes)
Concave_Edge.cpp – Code to generate and record concave edges. (1 KBytes)
Concave_Face.cpp – Code to generate and record concave faces from generated
concave edges. (2 KBytes)
Concave_Region.cpp – Code to generate and record concave regions from generated
concave faces. (50 KBytes)
Connect_Edge.cpp – Code to generate and record connect edges of a region. (1
KBytes)
Geometry.cpp – Code to judge geometry properties. It also contains common
functions to call ACIS functions. (41 KBytes)
LP_Parting_Direct.cpp – Code to generate parting direction of a region by calling
LINGO software system. (7 KBytes)
Region_Split.cpp – Code to divide a region or a face by using a given plane. (31
KBytes)
Rev_Glue.cpp – Code to generate glue faces and generate mold piece bodies based on
the glue faces and the mold base. (48 KBytes)
RTMold_Dialog.cpp – Code to generate an error dialog box. (3 KBytes)
RTMSelectTool.cpp – Code to select entities interactively by using a mouse. (13
KBytes)
SetTolDlg.cpp – Code to generate a dialog box for setting combination tolerances of
faces and regions. (2 KBytes)
ChangeRegDlg.cpp – Code to generate a dialog box for changing region numbers. (2
KBytes)
349
Appendix A
Set_CurR.cpp – Code to generate a dialog box to set the current region number for
judgment. (2 KBytes)
Set_MP_CurR_Dlg.cpp – Code to generate a dialog box to set the current region
number for mold piece construction. (2 KBytes)
StartLoopDlg.cpp – Code to generate a dialog box to set the iteration number before a
region which is not in main parting direction is considered in the combination
process. (2 KBytes)
• Header (.H) Files:
For each source file (.cpp) , there is a header file in the RTMDS. The header file (.h)
contains the definitions of parameters that are used in the source files, and also the
definitions of functions that are implemented in the source files. The function
descripoint of each header file is omitted here because the header file has the same
function as the source file in the same name. Therefore only the source code size is
provided.
RTMold.h (2 KBytes) MainFrm.h (2 KBytes)
RTMoldDoc.h (5 KBytes) RTMoldView.h (4 KBytes)
Attrib_Connect_Region.h (1 KBytes) Attrib_PEdge.h (1 KBytes)
Attrib_Region.h (1 KBytes) Attrib_Rtmold.h (1 KBytes)
Concave_Edge.h (1 KBytes) Concave_Face.h (1 KBytes)
Concave_Region.h (3 KBytes) Connect_Edge.h (1 KBytes)
Geometry.h (3 KBytes) LP_Parting_Direct.h (2 KBytes)
Region_Split.h (2 KBytes) Rev_Glue.h (4 KBytes)
RTMSelectTool.h (3 KBytes) SetTolDlg.h (1 KBytes)
ChangeRegDlg.h (2 KBytes) Set_CurR.h (2 KBytes)
Set_MP_CurR_Dlg.h (2 KBytes) StartLoopDlg.h (2 KBytes)
• Resource (.RC) Files:

Resource.h – Definitions of parameters in RTMold.rc. (4 Kbytes)
RTMold.rc – code for the definitions of menus, accelerators, dialogs, icons, and
toolbars. (20 KBytes)
• Project (.DSP) File:

RTMold.dsp – code used within the development environment. It stores the
information specific to the project. (13 KBytes)
Total size of source code: 379 KBytes

Executable Size: 913 KBytes, compiled optimized.
350
Appendix A
A.2 Classes of RTMDS

The major implementation of the RTMDS is in classes, which a format of Object-
Oriented Programming. The classes in the code are presented as follows with a brief
description of the purpose of each class:
class CRTMoldApp : public CwinApp – This class initializes the main application. It
also contains the initialization and deletion of ACIS and LINGO libraries.
class CMainFrame : public CframeWnd – This class creates the toolbar buttons and
the status bar. It also set their styles and positions.
class CRTMoldDoc : public CDocument – This class creates and modifies a
document. The main functions of three mold design modules are also in this class
because they change the context of a document.
class CRTMoldView : public CView – This class displays and prints the document
data. All the functions related to showing a CAD model are in this class.
class ATTRIB_RTMOLD : public ATTRIB – This class defines the common
attribute class used in the RTMDS. All other attribute classes are derived from it.
class ATTRIB_PEDGE: public ATTRIB_RTMOLD – This class defines an attribute
(parting edges) which is attached to an entity of an ACIS model.
class ATTRIB_CONNECT_REGION: public ATTRIB_RTMOLD – This class
defines an attribute (connected region properties) which is attached to an entity of
an ACIS model.
class ATTRIB_REGION: public ATTRIB_RTMOLD – This class defines an
attribute (region properties) which is attached to an entity of an ACIS model.
class Concave_Edge – This class generates and records concave edges in a CAD
model.
class Concave_Face – This class generates and records concave faces from generated
concave edges.
class Concave_Region – This class generates and records concave regions from
generated concave faces.
class Connect_Edge – This class generates and records connect edges of a region.
class RTMSelectTool : public MouseTool – This class provides functions to enable
users to interactively select entities in a CAD model by using a mouse.
class CSetTolDlg : public Cdialog – This class creates a dialog box for setting
combination tolerances of faces and regions.
class CStartLoopDlg : public Cdialog – This class creates a dialog box to set the
iteration number before a region which is not in main parting direction is
considered in the combination process
class CChangeRegDlg : public Cdialog – This class create a dialog box for changing
the number of a region.
class CSet_CurR : public Cdialog – This class creates a dialog box to set the current
region number for judgment.
class CSet_MP_CurR_Dlg : public CDialog – This class creates a dialog box to set
the current region number for mold piece construction.
351
Appendix A
A.3 Implementation on ACIS

The RTMDS is developed based on ACIS 6.2 and Microsoft Visual Studio 6.0. Before
using ACIS Microsoft Foundation Class Component (AMFC), the environment and
VC++ project should be set.
• Environment Variables
1. Set system variables: A3DT = [acis install directory]; ARCH = NT_DLLD;
A3DT_MAJOR = 6; A3DT_MINOR = 2.
2. A path to the libraries associated with the ARCH should be set up by adding the
string “;%A3DT%\lib\%ARCH%” in path environment variable.
• Microsoft Visual C++ Project Settings
Project settings are assessed under the “Project->Setting” menu command.
1. “Use run-time library” Setting: Select the “C/C++” tab and “Code Generation” in
the category pop-up. Select “Debug Multithreaded DLL” under the “Use run-time
library” pop-up.
2. Precompiled Header Option: In the “C/C++” tab, select “Precompiled Header”
from the category pop-up. Select the “Not using precompiled headers” radio
button.
3. Preprocessor definitions: In the “C/C++” tab, select “Preprocessor” from the
category pop-up. Click into the field for “Preprocessor definitions:” at the end of
the existing text. Add the string “,ACIS_DLL,NT,STRICT” into the project
setting field.
4. Additional Include Directories: In the “Preprocessor” category of the “C/C++”
tab, move the insertion cursor to the “Additional include directories:” field. Add
the string “., $(A3DT)\abl, $(A3DT)\aec, $(A3DT)\ag, $(A3DT)\ar,
$(A3DT)\amfc, $(A3DT)\amfc\acismfc, $(A3DT)\base, $(A3DT)\blnd,
$(A3DT)\bool, $(A3DT)\br, $(A3DT)\clr, $(A3DT)\covr, $(A3DT)\cstr,
$(A3DT)\ct, $(A3DT)\ds, $(A3DT)\eulr, $(A3DT)\fct, $(A3DT)\ga, $(A3DT)\gi,
$(A3DT)\gl, $(A3DT)\igl, $(A3DT)\ihl, $(A3DT)\intr, $(A3DT)\iges,
$(A3DT)\kern, (A3DT)\kern\kernel, $(A3DT)\law, $(A3DT)\lop, $(A3DT)\lopt,
$(A3DT)\mesh, $(A3DT)\ofst, $(A3DT)\oper, $(A3DT)\part, $(A3DT)\phl,
$(A3DT)\pid, $(A3DT)\rbase, $(A3DT)\rbi, $(A3DT)\rem, $(A3DT)\sbool,
$(A3DT)\scm, $(A3DT)\shl, $(A3DT)\skin, $(A3DT)\swp, $(A3DT)\vda,
$(A3DT)\vm, $(A3DT)\warp, $(A3DT)\xgeom” into the field.
5. Additional Library Path: Select the “Link” tab and “Input” from the category pop-
up menu. Insert the string “$(A3DT)\lib\$(ARCH)” into the “Additional library
path” field.
After the environment is set, a new project can be created to implement intended
functions.
• ACIS AppWizard
1. Copy the file “AcisAW.awx” from “$(A3DT)\amfc\aw-i386” to “Microsoft Visual
Studio\Common\MSDev98\Template” folder.
2. Initialize a new project by using ACIS AppWizard in VC++ environment.
• Function Implementations
352
Appendix A
1. Add all the source files (.cpp) and header files (.h) in folders
“$(A3DT)\amfc\acismfc” and “$(A3DT)\amfc\acismfc\tools” to the project.
2. Finally, ACIS’s API functions can be called in the source code to implement the
intended functions. Spatial Technology provides an online help for all the
functions and classes. By checking the description of functions, the developer can
determine which function should be used and how it should be used.
353
Appendix A
A.4 Implementation on LINGO

LINGO provides the Dynamic Link Library (DLL) standard for interfacing with other
applications. In C++ code, the steps to use LINGO to solve a linear problem are listed as
follows:
(1) First loading the LINGO DLL into the system by:
// Load the LINGO DLL
HINSTANCE m_hInstLINGO = LoadLibrary( "lingodll.dll");
// Get a pointer to the LINGO script processor – LGCSCRIPT
LINGOCALL* m_pLINGOCall = (LINGOCALL*) ::GetProcAddress((HMODULE)
m_hInstLINGO, "LGCSCRIPT");
(2) Generate a script file (.LNG);
(3) Use the C/C++ wrapper function LGCSCRIPT to call the LINGO script processor
to solve the linear problem. Some sample codes are given as:
pArgs[0] = (void*) "SET ECHOIN 1\nSET TERSEO 1\nTAKE
D:\\cv_Parting_dir.LNG\nGO\nQUIT\n";
// Capture all standard output in a file
pArgs[1] = (void*) "D:\\cv_Parting_Dir.LOG";
// Let LINGO allocate working memory
pArgs[2] = NULL;
double* pTransfer[8] = { &x1, &x2, &y1, &y2, &z1, &z2, &dStatus, NULL};
pArgs[3] = pTransfer;
int nArgs = 4;
// Pass the command script to LINGO
int nErrorCode;
(*m_pLINGOCall)( pArgs, &nArgs, &nErrorCode);
(4) The results given by LINGO are stored in the arguments pArgs[3], which can be
manipulated by C++ code;
(5) Finally after solving all linear problems, we should release the LINGO DLL from
the system by:
// Release LINGO library
FreeLibrary( m_hInstLINGO);
A sample of the generated LNG problem are given as follows:
MODEL:
MAX= (-0.004378*(x1-x2))+(0.004656*(y1-y2))+(1.001311 *(z1-z2));
!The surface constraints;

(-0.341977*(x1-x2))+(0.939683*(y1-y2))+(0.006876*(z1-z2))>=(-0.174000);
(-0.642831*(x1-x2))+(0.765977*(y1-y2))+(0.006878*(z1-z2))>=(-0.174000);
…
(0.980711*(x1-x2))+(-0.195275*(y1-y2))+(0.008613*(z1-z2))>=(-0.174000);
(0.980711*(x1-x2))+(0.195275*(y1-y2))+(0.008613*(z1-z2))>=(-0.174000);
!The sphere constraints;

(-0.130446*(x1-x2))+(0.034953*(y1-y2))+(-0.990839*(z1-z2))>=-0.990839;
(-0.130446*(x1-x2))+(-0.034953*(y1-y2))+(-0.990839*(z1-z2))>=-0.990839;
…
354
Appendix A
(-0.095493*(x1-x2))+(0.095493*(y1-y2))+(0.990839*(z1-z2))>=-0.990839;
(-0.130446*(x1-x2))+(0.034953*(y1-y2))+(0.990839*(z1-z2))>=-0.990839;
!Here is the data;

DATA:
@POINTER (1) = x1;
@POINTER (2)= x2;
@POINTER (3)= y1;
@POINTER (4)= y2;
@POINTER (5)= z1;
@POINTER (6)= z2;
@POINTER (7) = @STATUS();
ENDDATA
END
355
Appendix B
APPENDIX B
CAMERA ROLLER CASE STUDY:

COMPLETE SOLUTIONS OF THE
MODIFIED MPGT PROBLMES
In this appendix the results of solving the 24 modified MPGT problems in Section 8.5.2
are provided.
• Table B.1 contains the results of variables and goals given by OptdesX for the
slicing schemes 1 to 6 for PO1 (Figure 8.20.a).
slicing schemes 1 to 6 for PO2 (Figure 8.20.b).
slicing schemes 1 to 6 for PO3 (Figure 8.20.c).
slicing schemes 1 to 6 for PO4 (Figure 8.20.d).
356
Appendix B
Table B.1 – Solutions Obtained from OptdesX for Slicing Schemes 1~6 of PO1.
PO 1
SS1 SS2 SS3 SS4 SS5 SS6
Dia (mm) 2.782406 2.75355 3.280292 2.762365 3.631252 2.833394
Width (mm) 2.800527 2.796417 2.801306 2.799082 2.802805 2.80029
Thick (mm) 1.00125 1.000293 1.000419 1.001088 1.03211 1.000053
Rows 2 2 2 2 2 2
Cols 2 2 2 2 2 2
o
Draft1 ( ) 1.292404 1.286405 1.289208 1.302345 1.455682 0.913498
o
Draft2 ( ) 1.17641 1.182111 1.183049 1.179806 1.823202 0.962735
o
Draft3 ( ) 0.992764 1.017975 1.167627 1.008618 0.001585 1.002316
TA 0 0 0 0 0 0
CT (sec.) 300 300 300 300 300 300
HOC1-6 (mil) 0.002 0.002 0.002 0.002 0.002 0.002
FOC1-6 (mil) 0.012 0.012 0.012 0.012 0.012 0.012
Z-Rot (rad.) 0.010408 0.010417 0.010239 0.010415 0.010078 0.010391

Weight (g) 2.398691 2.399411 2.389682 2.399121 2.386154 2.397752
mold1_SF (µin) 156 156 156 156 390 382
mold2_SF (µin) 156 156 156 156 390 390
mold3_SF (µin) 64.91126 64.92935 132.7919 64.92262 407.0193 132.5313
mold1_PP (µin) 6 6 6 6 6 6
mold1_SP (µin) 64 68 74 77 187 187
mold2_PP (µin) 6 6 6 6 6 6
mold2_SP (µin) 64 68 72 75 177 177
mold3_P (µin) 6 6 6 6 6 6
Flat_Acc (inch) 0.001771 0.001771 0.001771 0.001771 0.001771 0.001771
Par_Acc (inch) 0.001616 0.001616 0.001616 0.001616 0.001616 0.001616
Per_Acc (inch) 0.002002 0.002002 0.002002 0.002002 0.002002 0.002002
Pos_Acc (inch) 0.006434 0.006434 0.006434 0.006434 0.006434 0.006434
Cost ($) 1290.801 1268.699 1238.415 1215.334 1381.207 1293.074
# molds1 1 1 1 1 2 3
# molds2 1 1 1 1 1 3
# molds3 3 3 3 3 17 6
Z 0.015166 0.015664 0.019315 0.016753 4.445024 3.492163
357
Appendix B
Table B.2 - Solutions Obtained from OptdesX for Slicing Schemes 1~6 of PO2.
Assembly 2
Dia (mm) 3.283212 3.257703 3.166841 3.538173 2.947257 2.921188
Width (mm) 2.79911 2.799287 2.801307 2.809834 2.798542 2.800543
Thick (mm) 1.001664 1.001234 1.000015 1.007053 1.003162 1.000218
Rows 2 2 2 2 2 2
Cols 2 2 2 2 2 2
o
Draft1 ( ) 1.288486 1.291703 1.285756 1.285702 0.965153 0.966904
o
Draft2 ( ) 1.183907 1.182914 1.188512 1.17707 1.108309 1.091008
o
Draft3 ( ) 1.564165 1.588649 1.69928 1.297851 0.98618 1.01393
TA 0 0 0 0 0 0
CT (sec.) 300 300 300 300 300 300
HOC1-6 (mil) 0.002 0.002 0.002 0.002 0.002 0.002
FOC1-6 (mil) 0.012 0.012 0.012 0.012 0.012 0.012
Z-Rot (rad.) 0.010233 0.010242 0.010275 0.01016 0.010341 0.010358

Weight (g) 2.389997 2.390428 2.391826 2.384321 2.39644 2.396307
mold1_SF (µin) 69 69 69 69 145 145
mold2_SF (µin) 69 69 69 69 145 145
mold3_SF (µin) 65.34362 65.36319 65.45268 133.0053 132.5065 132.5493
mold1_PP (µin) 64 66 68 68 158 155
mold1_SP (µin) 6 6 6 6 6 6
mold2_PP (µin) 64 66 68 68 158 155
mold2_SP (µin) 6 6 6 6 6 6
mold3_P (µin) 6 6 6 6 6 6
Flat_Acc (inch) 0.001771 0.001771 0.001771 0.001771 0.001771 0.001771
Par_Acc (inch) 0.001616 0.001616 0.001616 0.001616 0.001616 0.001616
Per_Acc (inch) 0.002002 0.002002 0.002002 0.002002 0.002002 0.002002
Pos_Acc (inch) 0.006434 0.006434 0.006434 0.006434 0.006434 0.006434
Cost ($) 1526.607 1510.464 1490.847 1557.075 1278.308 1278.703
# molds1 1 1 1 1 3 3
# molds2 1 1 1 1 2 2
# molds3 1 1 1 2 5 5
Z 0.034796 0.033874 0.032748 0.038377 1.175995 1.064109
358
Appendix B
Assembly 3
Dia (mm) 3.334592 3.174583 3.303421 3.328339 3.543661 3.595448
Width (mm) 2.800256 2.791463 2.798809 2.801502 2.784523 2.76438
Thick (mm) 1.001374 1.001518 1.002335 1.004717 1.003623 1.040405
Rows 2 2 2 2 2 2
Cols 2 2 2 2 2 2
o
Draft1 ( ) 1.285534 1.286113 1.285229 1.306726 1.248626 1.370419
o
Draft2 ( ) 1.182152 1.187626 1.17893 1.180519 1.939747 1.921682
o
Draft3 ( ) 1.521045 1.688734 1.540514 1.512036 0.000096 0.000575
TA 0 0 0 0 0 0
CT (sec.) 300 300 300 300 300 300
HOC1-6 (mil) 0.002 0.002 0.002 0.002 0.002 0.002
FOC1-6 (mil) 0.012 0.012 0.012 0.012 0.012 0.012
Z-Rot (rad.) 0.01022 0.01026 0.010225 0.010216 0.010142 0.010034

Weight (g) 2.388818 2.392824 2.38971 2.389256 2.3862 2.391746
mold1_SF (µin) 156 156 156 156 382 390
mold2_SF (µin) 146 146 146 147 206 205
mold3_SF (µin) 65.30937 65.44407 65.3248 65.30225 406.9467 406.9701
mold1_PP (µin) 6 6 6 6 6 6
mold1_SP (µin) 64 64 68 70 196 196
mold2_PP (µin) 64 71 66 67 185 185
mold2_SP (µin) 6 6 6 6 6 6
mold3_P (µin) 6 6 6 6 6 6
Flat_Acc (inch) 0.001771 0.001771 0.001771 0.001771 0.001771 0.001771
Par_Acc (inch) 0.001616 0.001616 0.001616 0.001616 0.001616 0.001616
Per_Acc (inch) 0.002002 0.002002 0.002002 0.002002 0.002002 0.002002
Pos_Acc (inch) 0.006434 0.006434 0.006434 0.006434 0.006434 0.006434
Cost ($) 1535.333 1480.554 1516.019 1504.937 1538.917 1535.311
# molds1 1 1 1 1 2 2
# molds2 1 1 1 1 1 1
# molds3 1 1 1 1 17 17
Z 0.037134 0.033593 0.036098 0.035687 3.969953 4.033753
359
Appendix B
Assembly 4
Dia (mm) 3.228181 3.204102 3.179343 3.246603 3.537486 3.553896
Width (mm) 2.797147 2.802364 2.798943 2.799383 2.77814 2.797763
Thick (mm) 1.00315 1.001073 1.000267 1.001422 1.03303 1.1209
Rows 2 2 2 2 2 2
Cols 2 2 2 2 2 2
o
Draft1 ( ) 1.2909 1.288987 1.286832 1.289972 1.435979 1.503045
o
Draft2 ( ) 1.191222 1.181283 1.191475 1.207883 1.861469 1.871831
o
Draft3 ( ) 1.628649 1.654986 1.682676 1.606225 0.00358 0.00042
TA 0 0 0 0 0 0
CT (sec.) 300 300 300 300 300 300
HOC1-6 (mil) 0.002 0.002 0.002 0.002 0.002 0.002
FOC1-6 (mil) 0.012 0.012 0.012 0.012 0.012 0.012
Z-Rot (rad.) 0.010245 0.010262 0.010268 0.010245 0.010077 0.009911

Weight (g) 2.391459 2.391151 2.39185 2.390661 2.390754 2.399862
mold1_SF (µin) 69 69 69 69 197 145
mold2_SF (µin) 156 156 156 156 382 390
mold3_SF (µin) 65.39534 65.41664 65.43913 65.37729 407.1168 406.9625
mold1_PP (µin) 64 66 66 67 192 185
mold1_SP (µin) 6 6 6 6 6 6
mold2_PP (µin) 6 6 6 6 6 6
mold2_SP (µin) 64 64 68 70 200 185
mold3_P (µin) 6 6 6 6 6 6
Flat_Acc (inch) 0.001771 0.001771 0.001771 0.001771 0.001771 0.001771
Par_Acc (inch) 0.001616 0.001616 0.001616 0.001616 0.001616 0.001616
Per_Acc (inch) 0.002002 0.002002 0.002002 0.002002 0.002002 0.002002
Pos_Acc (inch) 0.006434 0.006434 0.006434 0.006434 0.006434 0.006434
Cost ($) 1532.095 1479.042 1513.217 1497.957 1497.03 1521.837
# molds1 1 1 1 1 2 2
# molds2 1 1 1 1 1 1
# molds3 1 1 1 1 17 17
Z 0.036288 0.032353 0.035243 0.034411 4.171018 3.824423
360
References
REFERENCES
Adler, P. S., 1992, "Managing DFM: Lerning to Coordinate Product and Process
Design," Inetgrating Design and Manufacturing for Competitive Advantage (G. I.
Susman, ed.), Oxford University Press, pp. 140-156.
Allen, J. K. and D. W. Rosen, 1997, "Design Methodology Issues in a Distributed

Studio for Rapid Tooling and Molded Parts," International CIRP Design Seminar, Los
Angeles, CA.
Antonsson, E. K., 1996, "Structured Design Methods for MEMS", California

Institute of Technology, Pasadena, CA,
Ashley, S., 1994, "Prototyping with Advanced Tools", Mechanical Engineering, No.
June, pp. 48-55.
Baker, W. E., P. S. Westine and F. T. Dodge, 1991, Similarity Methods and

Engineering Dynamics: Theory and Practice of Scale Modeling, Elsvier.
Balasubramaniam, M., P. Laxmiprasad, S. Sarma and Z. Shaikh, 2000, "Generating

5-axis NC Roughing Paths Directly from a Tessellated Representation," Computer-Aided
Design, Vol. 32, No. 4, pp. 261-277.
Barlow, J. W., J. J. Beaman and B. Balasubramanian, 1996, "A Rapid Mould-making

System: Material Properties and Design Considerations," Rapid Prototyping Journal,
Vol. 2, No. 3, pp. 4-15.
Beckert, B. A., 1999, "Demystifying Moldmaking Software", Computer-Aided

Engineering, No. 6, pp.
Boothroyd, G. and P. Dewhurst, 1989, "Product Design for Assembly", Boothroyd

Dewhurst, Inc., Wakefield, RI.
Bowen, H., K. Clark, C. Holloway, D. Leonard-Barton and S. Wheelwright, 1994,

"Regaining the Lead in Manufacturing," Harvard Business Review, No.
September/October, pp. 108-144.
Box, G. E. P. and N. R. Draper, 1987, Empirical Model-Building and Response

Surface, John Wiley & Sons, New York.
Bralla, J. G., 1998, Design for Manufacturability Handbook, McGraw-Hill.
Broothroyd, G. and P. Dewhurst, 1991, "Product Design for Manufacture and

Assembly," Design for Manufacture: Strategies, Principles and Techniques (J. Corbett,
M. Dooner and C. Pym, eds.), Addison Wesley.
361
References
Bylinsky, G., 1998, "Industry's Amazing New Instant Prototypes", Fortune, No.
January, pp.
CAD/CAM Publishing Inc., 1999, "Fast Parting Lines", Computer Aided Design
Report, Vol. 19, No. 10, pp. 6-7.
Cedorge, T., 1999, "Surface Roughness and Draft Angle Effects on

Stereolithography Molds", M.S. Thesis, School of Mechanical Engineering, Georgia
Institute of Technology, Atlanta, GA.
Chen, L., S. Chou and T. C. Woo, 1993, "Parting directions for mould and die
design," Computer Aided Design, Vol. 25, No. 12, pp. 762-768.
Chen, L. L. and T. C. Woo, 1992, "Computational Geometry on the Sphere with

Applications to Automated Machining," Journal of Mechanical Design, No. 114, pp.
288-295.
Chen, W., 1995, "A Robuts Concept Exploration Method for Configuring Complex
Systems", Ph.D. Dissertation, School of Mechanical Engineering, Georgia Institute of
Technology, Atlanta, Georgia.
Chen, W., J. K. Allen, D. Mavris and F. Mistree, 1996, "A Concept Exploration
Method for Determining Robust Top-Level Specifications," Engineering Optimization,
Vol. 26, pp. 137-158.
Chen, W., J. K. Allen, F. Mistree and K.-L. Tsui, 1995, "Integration of Response
Surface Methods with Compromise Decision Support Problem in Developing a General
Robust Design Procedure," ASME Design Automation Conference, Boston,
Massachusetts, ASME, New York.
Cho, U., K. L. Wood and R. H. Crawford, 1998, "A Novel Empirical Similarity
Method for Reliable Product Testing with Rapid Prototypes," ASME Design Automation
Conference, Atlanta, GA, Paper No. DETC98/DAC-5605.
Cho, U., K. L. Wood and R. H. Crawford, 1998, "Online Functional Testing with
Rapid Prototypes: a Novel Empirical Similarity Method," Rapid Prototyping Journal,
Vol. 4, No. 3, pp. 128-138.
Cho, U., K. L. Wood and R. H. Crawford, 1999, "Error Measures for Functional
Product Testing," ASME Design for Manufacturing Conference, Las Vegas, NV.
Cho, U., K. L. Wood and R. H. Crawford, 1999, "System-Level Functional Testing

for Scaled Prototypes with Configurational Distrotions," ASME Design Automation
Conference, Las Vegas, NV, Paper No. DETC99/DAC-8606.
362
References
Cormen, T. H., C. E. Leiserson and R. R. L., 1990, Introduction to Algorithms, MIT

Press.
Crawford, J. C., 2001, "Injection Failure of Stereolithography Molds", M.S. Thesis,

School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA.
Davis, B., 2001, "Characterization and Calibration of SLA Products and Processes",
M.S. Thesis, School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, GA.
Dawson, E. K., 1998, "The Effect of Rapid Tooling on Final Product Properties",
Master Thesis, School of Chemical Engineering, Georgia Institute of Technology,
Atlanta.
Dawson, E. K., 2001, "The Effect of Rapid Tooling on Final Product Properties",
Ph.D. Dissertation, School of Chemical Engineering, Georgia Institute of Technology,
Atlanta.
de Berg, M., M. van Kreveld, M. Overmars, and O. Schwarzkopf , 1997,

"Computational Geometry- Algorithms and Applications," Springer.
Dhaliwal, S., S. Gupta and J. Huang, 2000, "Computing Exact Global Accessibility
Cones for Polyhedral Objects," ASME Design for Manufacturing Conference, Baltimore,
Maryland, DETC00/DFM-14034.
Dhaliwal, S., S. K. Gupta, J. Huang and M. Kumar, 2000, "A Feature Based
Approach to Automated Design of Multi-Piece Sacrificial Molds", Univ. of Maryland,
College Park, MD, ISR T.R. 2000-23.
Dissinger, T. E. and E. B. Magrab, 1996, "Geometric Reasoning for

Manufacturability Evaluation - Application to Powder Metallurgy," Computer-Aided
Design, Vol. 28, No. 10, pp. 783-794.
Dowling, N. E., 1993, Mechanical Behavior of Materials, Prentice Hall.
Edelmann, O., 2000, "The RNS 3 Travelpilot by Blaupunkt Driven by Success with
3D Systems", 3D@work, Vol. No. 3, pp. 4-5.
Eversheim, W., W. Bochtler, R. Grabler and W. Kolscheid, 1997, "Simultaneous

Engineering Approach to an Integrated Design and Process Planning," European Journal
of Operational Research, Vol. 100, pp. 327-337.
Foley, J. D., A. van Dam, S. K. Feiner and J. F. Hughes, 1990, Computer Graphics:
Principles and Practice, Addison-Wesley Publishing Co.
363
References
Frost, H. R. and M. R. Cutkosky, 1996, "Design for Manufacturability via Agent

Interaction," ASME-DETC 96, Irvine, CA, Paper No. DFM-1302.
Fu, M. W., J. Y. H. Fuh and A. Y. C. Nee, 1999, "Undercut Feature Recognition in

an Injection Mould Design System," Computer-Aided Design, Vol. 31, pp. 777-790.
Gadh, R. and S. Ratnakar, 1998, "Geometric Shape Abstractions for Internet-based

Virtual Prototyping," Computer-Aided Design, Vol. 30, No. 6, pp. 473-486.
Ganter, M. and L. L. Truss, 1990, "Computer Assisted Parting Line Development for
Cast Pattern Production," AFS Transactions, Vol. 98, pp. 795-800.
Gerhard, J., 2001, "Towards a decision-based distributed product realization

environment for engineering systems", Master thesis, School of Mechanical Engineering,
Georgia Institute of Technology, Atlanta, GA.
German, R. M. and A. Bose, 1997, Injection Molding of Metals and Ceramics,

MPIF, Princeton, NJ.
Goodman, J. E. and J. O'Rourke, 1997, Handbook of Discrete and Computational

Geometry, CRC Press, New York.
Gu, Z., Z. Zhou, S. Gao and J. Shi, 1999, "Determination of Mold Parting Direction
Based on Automatic Molding Feature Recognition," ASME Design Engineering
Technical Conference, Las Vegas, Nevada, DETC99/CIE-9119.
Gupta, P., R. Janardan, J. Majhi and T. Woo, 1996, "Efficient Geometric Algorithms
for Workpiece Orientation in 4- and 5-axis NC Machining," Computer Aided Design,
Vol. 28, No. 8, pp. 577-587.
Gupta, S. K., W. C. Regli, D. Das and D. S. Nau, 1995, "Automated

Manufacturability Analysis: A Survey", University of Maryland, ISR-TR-95-14.
He, W., Y. F. Zhang, K. S. Lee, J. Y. H. Fuh and A. Y. C. Nee, 1998, "Automated

Process Parameter Resetting for Injection Moulding: a Fuzzy-Neuro Approach," Journal
of Intelligent Manufacturing, Vol. 1998, No. 9, pp. 17-27.
Hernandez, G. and F. Mistree, 2001, "A Method for Formulating the Deviation
Function of Multiobjective Nonlinear Problems in Goal Programming," Engineering
Optimization, in press.
Herrmann, A. and J. Allen, 1999, "Selection of Rapid Tooling Materials and

Processes in a Distributed Design environment," ASME-DETC, Las Vegas, Paper No.
DFM-8930.
364
References
Hilton, P. D., 2001, "The Future of Rapid Manufacturing," Rapid Tooling:

Technologies and Industrial Applications (P. D. Hilton and P. F. Jacobs, eds.), Marcel
Dekker, New York, pp. 257-264.
Hogarth, S., 1999, "Molding the future for the shop floor," Manufacturing
Engineering. Vol. 123, No. 11, pp. 54-62.
Hoffman, C. M., 1989, Geometric and Solid Modeling: An Introduction, Morgan

Kaufmann Publishers Inc.
Hui, K. , 1996, “Geometric Aspects of the Mouldability of Parts,” Computer Aided

Design, Vol. 29, No. 3, pp. 197-208.
Hui, K. C. and S. T. Tan, 1992, "Mould Design with Sweep Operations - a Heuristic
Search Approach," Computer-Aided Design, Vol. 24, No. 2, pp. 81-91.
Jackson, T. R., 2000, "Analysis of Functionally Graded Material Object

Representation Methods", PhD thesis, Department of Ocean Engineering, Massachusetts
Institute of Technology, Boston, MA.
Jacobs, P., 1992, Rapid Prototyping & Manufacturing: Fundamentals of

Stereolithography, SME, Dearborn, MI.
Jacobs, P., 1996, Stereolithography and other RP&M Technologies: from Rapid
Prototyping to Rapid Tooling, Society of Manufacturing Engineers, Dearborn, MI.
Jacobs, P. F., 2001, "Rapid Soft Tooling and Rapid Bridge Tooling," Rapid Tooling:
Technologies and Industrial Applications (P. D. Hilton and P. F. Jacobs, eds.), Marcel
Dekker, Inc, New York, pp. 51-94.
Jared, G. M., M. G. Limage, I. J. Sherrin and K. G. Swift, 1994, "Geometric

Reasoning and Design for Manufacture," Computer-Aided Design, Vol. 26, No. 7, pp.
528-536.
Jayanthi, S. and B. Hokuf, et. al., 1997, "Stereolithographic Injection Molds for
Direct Tooling," Solid Freeform Fabrication Proceedings, Austin, Texas, University of
Texas at Austin.
Jerard, R. B. and S. J. Cox, 1998, "FACILE: A Clean Interface for Design and
Fabrication," ASME-Computer In Engineering Conference, Atlanta, GA, Paper No.
DETC98/CIE-5529.
Jerard, R. B. and O. Ryou, 2000, "Internet Based Fabrication of Discrete Mechanical

Parts," NSF Design & Manufacturing Research Conference, Vancouver, Canada.
365
References
Jin, Y., R. Ganeshan and P. Li, 1997, "ASCAD: Agent-Supported Collaborative

Design," CIRP 1997 International Design Seminar on Multimedia Technologies for
Collaborative Design and Manufacturing, Los Angeles, CA.
Judson, E., 1999, "Dimensional Accuracy in Rapid Prototyping of Ceramics Formed

by Injection Molding Using Rapid Tooling", Ph.D. Dissertation, Materials Science and
Engineering, Georgia Institute of Technology, Atlanta.
Karandikar, H. M., 1989, "Hierarchical Decision Making for the Integration of

Information from Design and Manufacturing Processes in Concurrent Engineering",
Ph.D. Dissertation, Department of Mechanical Engineering, University of Houston,
Houston, Texas,
Karandikar, H. M. and F. Mistree, 1991, "Modeling Concurrency in the Design of

Composite Structures," Structural Optimization: Status and Promise (M. P. Kamat, ed.),
AIAA, Washington D.C..
Karapatis, N. P., J. P. S. van Griethuysen and R. Glardon, 1998, "Direct Rapid

Tooling: A Review of Current Research," Rapid Prototyping Journal, Vol. 4, No. 2, pp.
77-89.
Kazmer, D. O. and R. G. Speight, 1997, "Polymer Injection Molding Technology for

the Next Millennium: A Vision to the Future," Journal of Injection Molding Technology,
Vol. 1997, No. June,
Khuri, A. and J. A. Cornell, 1987, Response Surface: Design and Analysis, Marcel
Dekker, Inc., New York.
Knight, W. A. and C. Poli, 1985, "A Systematic Approach to Forging Design,"

Machine Design, No. January.
Koch, P. N., A. Barlow, J. K. Allen and F. Mistree, 1996, "Configuring Turbine

Propulsion Systems using Robust Concept Exploration," Advances in Design Automation,
New York, ASME.
Krishnan, S., 1997, "Design for Manufacture: An Integrated System for Injection
Molding and Milling", PhD Dissertation, Mechanical Engineering, University of
Maryland, College Park.
Kumar, V. and D. Dutta, 1998, "An Approach to Modeling and Representation of

Heterogeneous Objects," ASME Journal of Mechanical Design, Vol. 120, No. 4, pp. 659-
667.
Kuppuraju, N., P. Ittimakin and F. Mistree, 1985, "Design through Selection ... A
Method that Works," Design Studies, Vol. 6, No. 2, pp. 91-106.
366
References
Kweon, S. and D. J. Medeiros, 1998, "Part Orientations for CMM Inspection Using
Dimensioned Visibility Maps," Computer-Aided Design, Vol. 30, No. 9, pp. 741-749.
Le Baut, Y. P., 1999, "Thermal Aspect of Stereolithography Molds", M.S. Thesis,

School of Mechanical Engineering, Georgia Insitute of Technology, Atlanta, GA.
Lee, R.-S., Y.-M. Chen, H. Y. Cheng and M.-D. Kuo, 1998, "A Framework of a
Concurrent Process Planning System for Mold Manufacturing," Computer Integrate
Manufacturing Systems, Vol. 11, No. 3, pp. 171-190.
Liker, J. K. and M. Fleischer, 1992, "Organizational Context: Barriers to DFM,"

Inetgrating Design and Manufacturing for Competitive Advantage (G. I. Susman, ed.),
Oxford University Press, pp. 228-264.
Lu, H. Y. and W. B. Lee, 2000, "Detection of Interference Elements and Release

Direction in Die-cast and Injection-moulded Components," Journal of Engineering
Manufacture, Vol. 214, No. B6, pp. 431-441.
Lu, S. C., A. B. Rebello, R. A. Miller, G. L. Kinzel and R. Yagel, 1997, "A Simple
Visualization Tool to Support Concurrent Engineering Design," Computer-Aided Design,
Vol. 29, No. 10, pp. 727-735.
Lynn, C. M., 1998, "Accuracy Models for SLA Build Style Decision Support", M.S.
Thesis, School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA.
Majhi, J., P. Gupta and R. Janardan, 1999, "Computing a Flattest, Undercut-free

Parting Line for a Convex Polyhedron, with Application to Mold Design," Computational
Geometry Theory and Applications, Vol. 13, pp. 229-252.
Mantÿlä, M., 1988, An Introduction to Solid Modeling, Computer Science Press.
McClurkin, J. and D. Rosen, 1998, "Computer-aided Build Style Decision Support

for Stereolithography," Rapid Prototyping Journal, Vol. 4, No. 1, pp. 4-13.
McClurkin, J. E., 1997, "A computer aided build style decision support method for
stereolithography", M.S. Thesis, School of Mechanical Engineering, Georgia Institute of
Technology, Atlanta, GA.
Mead, C., 1994, "Preface to Workshop Report on New Paradigms for

Manufacturing", National Science Foundation Report, Arlington, VA, NSF 94-123.
Megiddo, N., 1984, “Linear Programming in Linear Time When the Dimension is
Fixed,” ACM, Vol. 31, pp. 114-127.
367
References
Merz, R., F. B. Prinz, K. Ramaswami, M. Terk and L. Weiss, 1994, "Shape

Deposition Manufacturing," Proceedings of the Solid Freeform Fabrication Symposium,
Austin, TX, University of Texas at Austin.
Messac, A., S. M. Gupta and B. Akbulut, 1996, "Linear Physical Programming: A

New Approach to Multiple Objective Optimization," Transactions on Operational
Research, No. 8, pp. 39-59.
Miller, J. R., 1989, "Architectural Issues in Solid Modelers," IEEE Computer

Graphics and Applications, Vol. 9, No. 5, pp. 72-87.
Mistree, F., O. F. Hughes and B. A. Bras, 1993, "The Compromise Decision Support
Problem and the Adaptive Linear Programming Algorithm," Structural Optimization:
Status and Promise, pp. 249-289.
Mistree, F., O. F. Hughes and B. A. Bras, 1993, "The Compromise Decision Support
Problem and the Adaptive Linear Programming Algorithm," Structural Optimization:
Status and Promise (M. P. K., ed.), Washington, D.C., pp. 247-289.
Mistree, F., K. Lewis and L. Stonis, 1994, "Selection in the Conceptual Design of
Aircraft," AIAA/NASA/USAF/ISSMO Symposium on Multidisciplinary Analysis and
Optimization, Panama City, Florida, September 7-9.
Mistree, F., D. Muster, J. A. Shupe and J. K. Allen, 1989, "A Decision-Based

Perspective for the Design of Methods for Systems Design," Recent Experiences in
Multidisciplinary Analysis and Optimization, Hampton, Virginia, NASA.
Mistree, F., W. F. Smith, B. Bras, J. K. Allen and D. Muster, 1990, "Decision-Based

Design: A Contemporary Paradigm for Ship Design," Transactions, Society of Naval
Architects and Marine Engineers, Jersey City, New Jersey, pp. 565-597.
Mitchell, P., 1996, Tool and Manufacturing Engineers Handbook, Society of

Manufacturing Engineers, Dearborn, MI.
Montgomery, D. C., 1991, Design and Analysis of Experiments, John Wiley & Sons,
New York.
Mortenson, M. E., 1997, Geometric Modeling, John Wiley & Sons, Inc., New York.
Myers, R. H. and D. C. Montgomery, 1995, Response Surface Methodology: Process

and Product Optimization using Designed Experiments, John Wiley & Sons, New York.
Nee, A. Y. C., M. W. Fu, J. Y. H. Fuh, K. S. Lee and Y. F. Zhang, 1998, "Automatic

determination of 3-D parting lines and surfaces in plastic injection mould," Annals of
CIRP, Vol. 47, No. 1, pp. 95-98.
368
References
Nee, A. Y. C., S. K. Ong and Y. G. Wang, 1999, Computer Applications in Near

Net-Shape Operations, Springer.
Pahng, G. F., S. Bae and D. Wallace, 1998, "Web-Based Collaborative Design

Modeling and Decision Support," ASME-DETC98, Atlanta, GA, Paper No. EIM-5681.
Palmer, A., 1999, "The Effect of Feature Geometry on the Life of Stereolithography
Molds," MS Thesis, School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta.
Pedersen, K., 1999, "Designing Platform Families: An Evolutionary Approach to

Developing Engineering Systems", Ph.D., Mechanical Engineering, Georgia Institute of
Technology, Atlanta, GA.
Pedersen, K., J. Emblemsvag, R. Bailey, J. K. Allen and F. Mistree, 2000,

"Validating Design Methods and Research: The Validation Square," ASME/DETC,
Baltimore, Maryland, Paper No. DETC2000/DTM-14579.
Pham, G., 2001, "Ejection Failure of Stereolithography Molds", M.S. Thesis, School
of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA.
Poli, C., P. Dastidar, P. Mahajan and R. Graves, 1993, "Design for Stamping -
Analysis of Part Attributes that Impact Die Construction Costs for Metal Stampings,"
Journal of Mechanical Design, Vol. 115, No. December, pp. 735-743.
Poli, C., J. Escudero and F. Fernadez, 1988, "How Part Design Affects Injection
Molding Tool Costs," Machine Design, Vol. 24, No. Nov.,
Poli, C. and S. Shanmugasundaram, 1991, "Design for Die Casting - A Group

Technology Based Approach," Proc. ASME Design Theory and Methodology
Conference.
Pollmann, W., 1994, "Prototyping at Daimler-Benz: State of the Art and Future
Requirements," IMS International Conference on Rapid Product Development, Stuttgart,
Germany.
Prasad, B., 1996, Concurrent Engineering Fundamentals: Integrated Product and

Process Organization, Prentice Hall PTR.
Pye, R. G. W., 1989, Injection Mould Design, Longman Scientific & Technical.
Rahmati, S. and P. Dickens, 1997, "Stereolithography Injection Mould Tool Failure

Analysis," Solid Freeform Fabrication Symposium, Univ. of Texas at Austin, Austin,
Texas.
369
References
Rajagopalan, S. and J. M. Pinilla, et al., 1998, "Integrated Design and Rapid

Manufacturing over the Internet," ASME - DETC, Atlanta, GA.
Ravi, B. and M. N. Srinivasan, 1990, "Decision Criteria for Computer-Aided Parting

Surface Design," Computer-Aided Design, Vol. 22, No. 1, pp. 11-18.
Rees, H., 1995, Mold Engineering, Hanser, New York.
Regli, W. C. I., 1995, "Geometric Algorithms for Recognition of Featrues from Solid
Models", Ph.D. Dissertation, Department of Computer Science, University of Maryland,
College Park.
Reklaitis, G. V., A. Ravindran, and M. Ragsdell, 1983, Engineering Optimization:

Methods and Applications, New York, John Wiley and Sons.
Requicha, A. A. and J. R. Rossignac, 1992, "Solid Modeling and Beyond," IEEE

Computer Graphics and Applications, Vol. 12, No. 9, pp. 31-45.
Rodet, V., 2001, "Tool Life and Failure Mechanisms of Stereolithography Molds",
M.S. Thesis, School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, GA.
Rosato, D. V. and D. V. Rosato, 1995, Injection Molding Handbook, Chapman &

Hall.
Rosen, D. W., 1992, "A Feature-based Representation to Support the Design Process
and the Manufacturability Evaluation of Mechanical Components", Ph. D. Dissertation,
Department of Mechanical Engineering, University of Massachusetts, Amherst.
Rosen, D. W., 1994, "Automated Construction of Moving Mold And Die Section
From Solid Component Models".
Rosen, D. W., 1998, "Progress Towards a Distributed Product Realization Studio:

The Rapid Tooling TestBed," Knowledge Intensive CAD IFIP WG 5.2 Workshop.
Rosen, D. W., 2000, "Design-to-Manufacture Information Transfer in the Context of

Solid Freeform Fabrication Technologies," Knowledge Intensive Computer Aided Design
(S. Finger, T. Tomiyam and M. Mantyla, eds.), Kluwer Academic Publishers, Boston.
Rosen, D. W., W. Chen, S. Coulter and S. Vadde, 1994, "Goal-Directed Geometry:

Beyond Parametric and Variational Geometry CAD Technologies," ASME Design
Automation Conference, pp. 417-426.
Rosenthal, S. R., 1990, "Bridging the Cultures of Engineers: Challenges in

Organizing for Manufacturable Product Design," Managing the Design-Manufacturing
Process (J. E. Ettlie and H. Stoll, eds.), Mcgraw-Hill, New York, pp. 21-52.
370
References
Sambu, S., 2001, "A Design for Manufacture Method for RP and RT", Master
Thesis, Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA.
Sarma, S. E., S. Schofield, J. Stori, J. MacFarlane and P. Wright, 1996, "Rapid

Product Realization from Detail Design," Computer-Aided Design, Vol. 28, No. 5, pp.
383-392.
Schrage, M., 1993, "The Culture of Prototyping," Design Management Journal, Vol.
4, No. 1, pp. 55-65.
Sequin, C. H. and S. McMains, 1995, "What Can We Learn from the VLSI CAD
Revolution?" NSF Workshop on Design Methodologies for Solid Freeform Fabrication,
Carnegie Mellon University, Pittsburgh, PA.
Serrar, M., 1995, "Applying Moldability Evaluation to Solid Models", PhD

Dissertation, Mechanical Engineering, Rensselaer Polytechnic Institute, Troy, NY.
Shah, J. and P. Wright, 2000, "Developing Theoretical Foundations of DFM,"

ASME- Design for Manufacture Conference, Baltimore, MD.
Shah, J. J., 1991, "Assessment of Feature Technology," Computer Aided Design.
Shah, J. J. and M. Mantyla, 1995, Parametric And Feature-Based CAD/CAM, John

Wiley & Sons, Inc., New York.
Shigley, J. E. and C. R. Mischke, 1989, Mechanical Engineering Design, McGraw

Hill, Inc., New York.
Shin, K. H. and K. Lee, 1993, "Design of Side Cores of Injection Moulds from
Automatic Detection of Interference Faces," Journal of Design and Manufacturing, Vol.
3, No. pp. 225-236.
Shupe, J. A., 1988, "Decision-Based Design: Taxonomy and Implementation",

Ph.D. Dissertation, Department of Mechanical Engineering, University of Houston,
Houston, Texas.
Siddique, Z., 2000, "Common Platform Development: Designing for Product

Variety", Ph.D. Dissertation, Mechanical Engineering, Georgia Institute of Technology,
Atlanta.
Siddique, Z. and D. W. Rosen, 1997, "A Virtual Prototyping Approach to Product

Disassembly Reasoning," Computer-Aided Design, Vol. 29, No. 12, pp. 847-860.
Simon, H. A., 1996, The Sciences of the Artificial, MIT Press, 3rd Ed., Cambridge,
MA.
371
References
Smith, P. and D. Reinertsen, 1991, "The Time-to-market Race," Developing

Products in Half the Time), Van Nostrand Reinhold, New York, pp. 3-13.
Simon, H. A., 1996, The Sciences of the Artificial, MIT Press, 3rd Ed., Cambridge,
MA.
Sobieszczanski-Sobieski, J., 1989, "Multi-Disciplinary Optimization for Engineering

Systems: Achievements and Potential", NASA, Report No.101566.
Spatial Technology, Inc., 2000, ACIS Technical Overview, Boulder, CO.
Sriram, R. D. and R. Logcher, 1993, "The MIT DICE Project," IEEE Computer, pp.
64-65.
Stefano, P. D., 1997, "Automatic Extraction of Form Features for casting,"

Computer-Aided Design, Vol. 29, No. 11, pp. 761-770.
Stoll, H. W., 1991, "Design for Manufacture: an Overview," Design for

Manufacture: Strategies, Principles and Techniques (J. Corbett, M. Dooner, J. Meleka
and C. Pym, eds.), Addison-Wesley, pp. 107-129.
Storti, D., C. Redl, M. Ganter, G. Turkiyyah and T. Woo, 1999, "Encapsulated

Transmission of Part Specifications for Distributed Solid Freeform Fabrication," ASME-
Design Automation Conference, Las Vegas, Nevada.
Susman, G. I., 1992, Integrating Design and Manufacturing for Competitive

Advantage, Oxford University press, New York.
Susman, G. I. and J. W. Dean, 1992, "Development of a Model for Predicting

Design for Manufacturability Effectiveness," Inetgrating Design and Manufacturing for
Competitive Advantage (G. I. Susman, ed.), Oxford University Press, pp. 228-264.
Tan, S. T., M. F. Yuen, W. S. Sze and K. W. Kwong, 1988, "A Method for
Generation of Parting Surface for Injection Moulds," Conference on Computer Aided
Production Engineering, Edinburgh, UK.
Tromans, G. and D. I. Wimpenny, 1998, "Rapid Tooling-The Future for Industry",

Rapid News, Vol. 3, No. 3, pp. 40-46.
Urabe, K. and P. K. Wright, 1997, "Parting Direction and Parting planes for the
CAD/CAM of Plastic Iinjection Molds," Proceeding of DETC'97, Sacramento, CA.
van Vliet, J. W., C. A. van Luttervelt and H. J. J. Kals, 1999, "State-of-the-art Report
on Design for Manufacturing," DETC99-DFM, Las Vegas, NV, Paper No. DFM-8970.
372
References
Vijay, J. V. K., U. Shrinivasa and B. Gurumoorthy, 1998, "Automatic Draw

Direction Generation for Die Design," ASME Computers and Information in
Engineering, Atlanta, GA.
Wang, F.-C. F. and P. K. Wright, 1998, "Web-based CAD Tools for a Networked
Manufacturing Service," ASME - DETC, Atlanta, GA.
Weinstein, M. and S. Manoochehri, 1996, "Geometric Influence of a Molded Part on

the Draw Direction Range and Parting Line Loctions," Journal of Mechanical Design,
Vol. 118, No. 3, pp. 29-39.
West, A. P, 1999, "A Decision Support System for Fabrication Process Planning of
Stereolithography," Master Thesis, School of Mechanical Engineering, Georgia Institute
of Technology, Atlanta.
West, A. P., S. P. Sambu, and D. W. Rosen, 2000, “A process planning method for
improving build performance in stereolithography,” Computer Aided Design, Vol. 33, No.
1, pp. 65-79.
Whitney, D. E., 1996, "Why Mechanical Design Cannot Be Like VLSI Design",
Massachusettes Institute of Technology, Boston.
Williams, J., 1999, "Rapid Tooling for Plastic Injection Molds: Application and
Decision-Maker Tools", Time-Compression, Vol. 4, No. 6, pp. 46-52.
Winner, R. I., J. P. Pennell, H. E. Bertrand and M. M. G. Slusarczuk, 1988, "The

Role of Concurrent Engineering in Weapons System Acquisition", Institute for Defense
Analyses, Alexandria, Virginia, IDA Report R-338.
Wohlers, T., 1998, "State of the Industry Report: 1998." Wohlers Associates.
Wohlers, T., 1999, "Rapid Prototyping and Tooling State of the Industry 1999:
Executive Summary," Time-Compression Technologies, Vol. 4, No. 3, pp. 40-43.
Wohlers, T. T., 2000, "Rapid Production With Rapid Prototyping", Computer-Aided

Engineering, No. April, pp. 20-24.
Wong, T., S. T. Tan and W. S. Sze, 1996, "Parting Line Formation by Slicing a
Trimmed Surface Model," ASME-DETC, Irvine, CA.
Woo, T. C., 1994, "Visibility Maps and Spherical Algorithms," Computer-Aided

Design, Vol. 26, No. 1, pp. 6-16.
Woodwark, J. R., 1989, Geometric Reasoning, Clarendon Press.
373
References
Yin, Z., H. Ding and Y. Xiong, 2001, "Virtual Prototyping of mold design:
geometric mouldability analysis for near-net-shape manufactured parts by feature
recognition and geometric reasoning," Computer-aided Design, Vol. 33, No. pp. 137-154.
Zhu, L. and D. O. Kazmer, 1999, "A Performance-Based Representation for

Engineering Design," ASME-Design Theory and Methodology Conference, Las Vegas,
Nevada.
Zhu, L. and D. O. Kazmer, 2000, "A Performance-Based Representation of

Constraint Based Reasoning and Decision Based Design," ASME-Design Theory and
Methodology Conference, Baltimore, Maryland.
374
Vita
VITA
Yong Chen was born in Guiyang, Guizhou, China on October 20, 1972. He grew up in
Zunyi, Guizhou and attended Zuiyi No. 4 High School. He earned his Bachelor of
Science degree in Mechanical Engineering from Zhejiang University (Hangzhou,
Zhejian, China) in 1993. He earned his Master of Science degree in Mechanical
Engineering from Huazhong University of Science and Technology (Wuhan, Hubei,
China) in 1996. After that he worked as a lecturer at school of Mechanical Engineering
and a research scientist at Graphical Software Center at Huazhong University of Science
and Technology before attending Georgia Institute of Technology in March 1998. His
graduate work was funded by National Science Foundation Grant DMI-96-18039. Upon
graduation, Yong has accepted a position as a Senior Software Engineer with the 3D
Systems Inc. in Valencia, California.
375

Yong Dissertation PDF

Cargado por

Información del documento

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

Yong Dissertation PDF

Cargado por

Copyright:

Formatos disponibles

Computer-Aided Design for Rapid Tooling: Methods for Mold

Design and Design-for-Manufacture

Georgia Institute of Technology

Copyright © 2001 by Yong Chen

David W. Rosen, Chair

Table 1.1 – Relationship Between Hypotheses and Dissertation Sections. 26

Figure 1.1 - The Cost of an Design Change Related to Development Phase 2

ACES Accurate Clear Epoxy Solid

Physical models and prototypes are fundamental to superior product development

Design Method and a Design-for-Manufacturing Method respectively.

Based on Computation Geometry, a Multi-piece Mold Design Method is

developed to automate several important mold design steps, including determining

verified by physical experiments.

Based on Decision-Based Design, a Design for Rapid Tooling System is

developed to aid manufacturers to tailor a submitted part efficiently and effectively. A

can be transferred to the manufacturer by geometric tailoring decision templates. The

designer’s requirements on functional prototypes are formulated in the decision

reducing the iterations between the designer and manufacturer. By synthesizing

the DFM method.

FUNCTIONAL PROTOTYPE USING RAPID TOOLING

The product realization process, driven by market factors, is changing dramatically.

1.1 PRIORS – FUNCTIONAL PROTOTYPE AND RAPID TOOLING

1.1.1 Product Realization and Prototype

Product Development Phase Cost of Design Change

Figure 1.1 - The Cost of a Design Change Related to Development Phase

1.1.2 Rapid Prototyping

Figure 1.2 - A Schematic Drawing of a SLA.

DM, the costs with Stereolithography were 12,000 DM.

1.1.3 Injection Molding and Rapid Tooling

Direct AIM Tooling

Mold build by SLA Mold after Epoxy Back Fill

Figure 1.5 – Direct AIM Tooling Process.

1.1.4 Rapid Manufacturing

1.2 DESIGN FOR RAPID TOOLING AND RESEARCH OPPORTUNITIES

1.2.1 Current Usage of Rapid Tooling and Related Problems

Phase I: Phase II: Phase III: Phase IV:

Figure 1.6 – Processes of Using Direct AIM Tooling to Build Prototypes.

1.2.2 Motivating Project – Rapid Tooling TestBed

Design Design for Manufacture Manufacture

Figure 1.7 – RTTB System and My Focus.

1.2.3 Challenges of RTTB and My Research Approach

Figure 1.8 – Different Geometries of Some Industrial Parts.

(2) Design-for-Manufacture based on Decision-Based Design.

(a) Product Development Spiral (b) Relations of Parameters and Attributes

By approaching design from the perspective of making decisions, it becomes possible to

1.2.4 Research Opportunities in Mold Design and Design-for-Manufacture

Figure 1.10 – An Illustrative Mold for Injection Molding.

Determine Determine mold Determine

Feed systems (sprue,

Mold Detail Add feed systems

Add ancillary items (guide Complete

Figure 1.11 – Mold Design Process and My Considerations.

In prototype tooling, many components of the mold are standardized. Specifically,

Figure 1.12 – Standardized Master Frame Used in Morgan Press at RPMI.

(a) A part to be built (b) CAD models of disassembled mold pieces

(c) Photo of mold pieces (d) Injection molded parts

Figure 1.13 – A Multi-piece Molds for a Part Given by Protoform GmbH.

it is proposed to develop a systematic approach from geometrical perspective to

meets difficulties if a part or a manufacturing process is complex. It is rather

Geometric Tailoring, in this dissertation, is to change the geometry of a part to

Figure 1.14 – Relations of DFRT, DFM and Geometric Tailoring.

Sub-Hypothesis 2.1: The designer can initiate a material geometric tailoring

Table 1.1 – Relationship Between Hypotheses and Dissertation Sections