assembly features in modelling and planning phd-holland

7/31/2019 Assembly Features in Modelling and Planning Phd-holland

1/173

assembly features in modelling and planning


2/173

Copyright 1997 by Winfried van Holland

ISBN 90-9011056-9NUGI 855

A PostScript version of this thesis is available via :

ftp://ftp.twi.tudelft.nl/TWI/publications/dissertations/W.vanHolland.ps.gz.

This thesis was typeset using teTEX (version 0.4), a TEX/ LATEX distribution that includes TEX, LATEX2 , -

and many other programs, such as dvips, xdvi and B IBTEX.

The main part of this thesis was written with the use of the XEmacs editor (version 19.14) using AUC-TEX(version 9.4g) on a LinuX (version 2.0.23) operating system.

Many of the designations used by the manufacturers and sellers to distinguish their products are claimed as

tradema rks. Where those designations app ear in this thesis, and the author was aw are of the trademark claim,

the designations have been printed with the symbol.


3/173

Assem bly Features

in

Mod elling and Plann ing

PROEFSCHRIFT

ter verkrijging van de graad van doctor

aan de Technische Universiteit Delft,

op gezag van de Rector Magnificus Prof.dr.ir. J. Blaauwendraadin het openbaar te verdedigen ten overstaan van een commissie,

door het College van Dekanen aangewezen,

op maan dag 3 november 1997 om 10:30 uu r

door

Winfried VAN HOLLAND

informa tica ingenieur

geboren te Maarssen


4/173

Dit proefschrift is goedgekeurd door de promotor:

Prof.dr.ir. F.W. Jansen

Toegevoegd prom otor:

Dr. W.F. Bronsvoort

Samenstelling promotiecommissie:

Rector Magnificus, voorzitter

Prof.dr.ir. F.W. Jansen, Technische Universiteit Delft, promotor

Dr. W.F. Bronsvoort, Technische Universiteit Delft, toegevoegd promotorProf.dr. I. Horvath, Technische U niversiteit Delft

Prof .dr.ir. H.J.J. Kals , Univer siteit Tw ente en Technische Un iversiteit Delft

Prof.dr. H. Koppe laar, Technische Universiteit Delft

Dr.ir. T. Storm, Technische U niversiteit Delft

Dr. D.E. Whitney, Massachusetts Institute of Technology

Prof. J.M. Hennesy, Technische U niversiteit Delft, reservelid


5/173

Preface

This thesis is concerned with the application of the feature m odelling

concept, originally developed for single-part manufacturing, to assembly

mod elling and planning. When you are expecting that the complete as-

sembly process will be automated from now on, then this thesis may dis-

app oint you . How ever, this thesis shows that some aspects of assembly

can be more automated and easier used with the here described assemblyfeatures.

The research was carried out at the Compu ter Graphics and CAD/ CAM

group of the Department of Compu ting Science of Delft University of

Technology. It was p rofessor Denis McConalogue w ho h eaded the CAD/

CAM group w hen I became a PhD studen t. Because of his sup erannu ation,

professor Erik Jansen became my first promoter. Without the ever motivat-

ing wor k of my co-prom oter and sup ervisor, Wim Bronsvoort, wh o invited

me to join the group as a PhD stud ent, this w ork could not have been fin-

ished. He has the perfect gift to drag people through difficult periods. Iowe him many red pens for all his corrections made in my documents.

Several p eople d id also provide d irect or indirect contributions to my

work, I want to thank:

Especially, the other PhD students involved in feature modelling re-

search Maur ice Dohm en, Klaas Jan de Kraker and later Rafael Bidarra

for being sparring partners for my ideas.

The people from the Mechanical Engineering Department Marcel

Tichem, Michiel Willemse, Jan Peter Baartman, Ton Storm, Bart Meijer and

professor Nick Reijers in their ever lasting patience in explaining me,

as a comp uting scientist, the differences between screws and bolts and the

more serious p arts of m echanical engineering.

The sup port staff of our group du ring my stay Aadjan van der Helm,

Peter Kailuh u, Kees Seebregts and ter Jonker they did spend a lot of

time in creating for me a w orking env ironment, as comfortable as it could


6/173

vi Preface

be, given the m eans available.

The master students who helped me in solving parts of my problems

Ronald Hupperichs doing research in the functional modelling area,

Ronald van Gimst doing research in motion planning and Marco van der

Zwet doing research in modelling with assembly features.

The other PhD students of our group Arjan Kok (my first room-

mate), Reinier van Kleij (my second roomm ate), I. Ari Sadarjoen, Andrea

Hin, Theo van Walsum, Wim de Leeuw, Freek Reinders and Erik Reinhard

for all the nice hours during work and after work, but almost always

talking about our w ork.

I will miss the laughter of Ton Bosman through the long aisles, and the

sneezings from Matthijs Sepers, wh o un consciously provided the bu ilding

with some liveliness.

I will miss the people from the First-Aid group, who were alwayslaughing du ring practising First-Aid on the most d isgusting wou nd s and

amputations.

And I will remember my car-pool p artners dr. Cees Witteveen, dr.

Leo Boellaard and the one w ithout a grad e Martien van Beeck they have

shortened the long traffic jams with all kinds of games and alternative

routes through the Groene Hart of the Netherland s.

I also want to thank my new colleagues, from Baan Research, who are

provided with the habit of continuously reminding me of the fact that

there w as still something like a thesis that I had to finish.

I want to thank m y parents for providing me the opp ortunity to go to uni-versity. And of course, my w ife Nelleke, for her love, sup port and patience

and all the work she did for me d uring the last years, that I was sup posed

to do, but d id not d o, because I was busy w ith what you are reading now.

Soli Deo Gloria to God alone be glory.

Winfried van Holland

Breukelen, 1997


7/173

Contents

1 Introduction 1

1.1 Assem bly m od ellin g . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Assem bly p lann ing . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Prod uct m od els . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Main objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.5 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

I Background 7

2 Assembly modelling 9

2.1 Top -d ow n a nd b ot tom -u p m od ellin g . . . . . . . . . . . . . . 9

2.1.1 Top -d ow n ap proach . . . . . . . . . . . . . . . . . . . 9

2.1.2 Bottom -u p ap p roach . . . . . . . . . . . . . . . . . . . 10

2.2 Modelling with detailed single-part models . . . . . . . . . . 10

2.2.1 Sin gle-p art m od els . . . . . . . . . . . . . . . . . . . . 112.2.2 H iera rch ical m od els . . . . . . . . . . . . . . . . . . . 13

2.2.3 Relation al m od els . . . . . . . . . . . . . . . . . . . . . 13

2.3 Modelling with functional information . . . . . . . . . . . . . 17

2.4 A ssem bly fea tu res t o fi ll t he g ap . . . . . . . . . . . . . . . . 21

3 Assembly planning 25

3.1 Assem bly p lan nin g m od u les . . . . . . . . . . . . . . . . . . 25

3.2 Gro u p in g a ss em b ly p la n nin g m o d u les . . . . . . . . . . . . . 28

3.2.1 On -lin e a nd o ff-lin e p la nn in g . . . . . . . . . . . . . . 29

3.2.2 U sin g a bst ra ct ion lev els . . . . . . . . . . . . . . . . . 29

3.3 Experiences with existing assembly planners . . . . . . . . . 30

3.3.1 Experiences in DIAC . . . . . . . . . . . . . . . . . . . 31

3.3.2 Exp er ien ces in Arch im e d es 2 . . . . . . . . . . . . . . 32

3.4 Exp e rie nces in m a n u fa ct u rin g p la n nin g . . . . . . . . . . . . 32

3.5 Fea tu res in a ssem bly p la nn in g . . . . . . . . . . . . . . . . . 33

vii


8/173

viii CON TEN TS

4 Towards an integrated modelling and planning environment 35

4.1 Lon g-term goals . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1.1 Top-down and functional modelling . . . . . . . . . . 36

4.1.2 Integration of single part and assembly modelling . . 37

4.1.3 Integration of modelling and planning . . . . . . . . 37

4.2 Sh ort-term goals . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.2.1 Top-down and functional modelling . . . . . . . . . . 38

4.2.2 Integration of single-part and assembly mod elling . . 38

4.2.3 Integration of modelling and planning . . . . . . . . 38

4.2.4 Towards solutions of the short-term goals . . . . . . . 39

II Modelling w ith assembly features 41

5 Feature-based product model 43

5.1 O bject-or ien tation . . . . . . . . . . . . . . . . . . . . . . . . . 44

5.2 Sin gle p art m od el . . . . . . . . . . . . . . . . . . . . . . . . . 46

5.2.1 Fo rm fe at u re s a s b u ild in g b lo ck s . . . . . . . . . . . . 47

5.2.2 C on st ra in ts fo r m u t u al r ela tio ns . . . . . . . . . . . . 48

5.2.3 Feature model: combining form features and con-

straints . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5.3 Assem bly m od el . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.3.1 C om p o ne nt s a s b u ild in g b lo ck s . . . . . . . . . . . . . 50

5.3.2 Connection features for mutual relations . . . . . . . 52

5.3.3 Generic combined model: combining components

a nd con n ect ion fea tu res . . . . . . . . . . . . . . . . . 53

5.4 Com bin ed p rod u ct m od el . . . . . . . . . . . . . . . . . . . . 54

6 Assembly features 57

6.1 Assembly information and assembly features . . . . . . . . . 57

6.1.1 Generic and instance level assembly information . . . 57

6.2 A ssem bly fea tu re d efin it ion s . . . . . . . . . . . . . . . . . . 58

6.3 H an dlin g featu res . . . . . . . . . . . . . . . . . . . . . . . . . 59

6.3.1 In for m at io n w it hin h an d lin g . . . . . . . . . . . . . . 60

6.3.2 H a nd lin g fea tu res cla ss . . . . . . . . . . . . . . . . . 60

6.4 Con nection featu res . . . . . . . . . . . . . . . . . . . . . . . . 63

6.4.1 In fo rm a tio n wit h in a co nn ect io n . . . . . . . . . . . . 63

6.4.2 Con n ect ion fea tu re cla ss . . . . . . . . . . . . . . . . . 64

6.4.3 A tt ach m en ts a nd a gen ts . . . . . . . . . . . . . . . . . 66


9/173

CON TEN TS ix

7 Assembly modelling prototype system 69

7.1 Prototy pe arch itectu re . . . . . . . . . . . . . . . . . . . . . . 69

7.2 Assembly modelling versus actual assembly of a produ ct . . 70

7.3 Com bin ed cla ss v iew er s . . . . . . . . . . . . . . . . . . . . . 717.3.1 Geom etr y v iew er . . . . . . . . . . . . . . . . . . . . . 71

7.3.2 Grap h view er . . . . . . . . . . . . . . . . . . . . . . . 73

7.4 Co mb in ed cla ss m od ellin g . . . . . . . . . . . . . . . . . . . . 78

7.4.1 O n-lin e m od ellin g . . . . . . . . . . . . . . . . . . . . 79

7.5 Im p lem en tation issu es . . . . . . . . . . . . . . . . . . . . . . 81

III Planning w ith assembly features 83

8 Stability analysis 858.1 Tr an sla tion al st ab ilit y . . . . . . . . . . . . . . . . . . . . . . . 86

8.1.1 In ter na l freed om o f m ot io n . . . . . . . . . . . . . . . 87

8.1.2 Visibility m ap s . . . . . . . . . . . . . . . . . . . . . . 88

8.1.3 Cen tr al p roject ion . . . . . . . . . . . . . . . . . . . . . 89

8.1.4 Using connection features for internal freedom of

m otion . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

8.2 Rotation al stability . . . . . . . . . . . . . . . . . . . . . . . . 93

8.2.1 Ro ta tio na l d e gr ees o f fr ee d om . . . . . . . . . . . . . 95

8.2.2 Using connection features for rotational degrees of

freed om . . . . . . . . . . . . . . . . . . . . . . . . . . 98

9 Grip planning 99

9.1 Gr ip p lan nin g in gen er al . . . . . . . . . . . . . . . . . . . . . 99

9.1.1 Grip per asp ects . . . . . . . . . . . . . . . . . . . . . . 99

9.1.2 Fin ger d om ain a sp ect s . . . . . . . . . . . . . . . . . . 1 01

9.1.3 A ct ua l g rip a sp ects . . . . . . . . . . . . . . . . . . . . 101

9.2 Finger domains and non-free regions . . . . . . . . . . . . . . 102

9.2.1 Exp an d ed Fa ce So lid s . . . . . . . . . . . . . . . . . . 1 02

9.2.2 Problems using Expanded Face Solids . . . . . . . . . 103

9.3 Usin g fea tu r es in g rip p la n nin g . . . . . . . . . . . . . . . . . 1059.3.1 U sin g for m fea tu res . . . . . . . . . . . . . . . . . . . 1 06

9.3.2 Usin g h an d lin g fe at u re s . . . . . . . . . . . . . . . . . 112

9.3.3 Usin g co nn ect io n fe at u re s . . . . . . . . . . . . . . . . 113

9.4 Resu lts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114


10/173

x CON TEN TS

10 Other planning modules 121

10.1 M otion p la nn in g . . . . . . . . . . . . . . . . . . . . . . . . . 121

10.1.1 Gro ss m o tio n p la n nin g . . . . . . . . . . . . . . . . . . 121

10.1.2 Fin e m ot io n p la nn in g . . . . . . . . . . . . . . . . . . 122

10.2 Asse m bly seq u en ce p la nn in g . . . . . . . . . . . . . . . . . . 123

10.2.1 Generate precedence relations . . . . . . . . . . . . . 124

10.2.2 Generate feasible assembly sequences . . . . . . . . . 124

10.2.3 Retrieve the optimal assembly sequence . . . . . . . . 126

10.2.4 Ad ditional profits of using features for sequence plan-

ning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

IV Concluding remarks 131

11 Conclusions and future research directions 133

11.1 Con clu sion s . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

11.2 Fu t u re r es ea rch d ir ect io ns . . . . . . . . . . . . . . . . . . . . 138

Bibliography 141

A Terminology 153


11/173

W ho is wise and understanding among you?

Let him show it by his good life, by deeds done in hum ility that

comes from wisdom.

Jam es 3:13 (NIV)


12/173


13/173

Chapter 1

Introduction

With the now commonly used tools in computer-aided design (CA D) andcomputer-aided manufacturing (CA M), accura cy in part design an d preci-

sion in production have increased. These improvements made it possible

to further reduce the lead time, or time-to-market, of products, for exam-

ple by further automating the assembly p rocess. The p rocess of linking

CAD and CAM is now going on for a couple of decades.

In the 80s, however, the automation of the assembly process suddenly

staggered. One of the reasons for th is was the disillusion in the possi-

bilities of using flexible assembly robots. Although it w as p ossible for

such robots to perform a large set of tasks, it w as very hard to automati-

cally generate programs to execute these tasks on these robots (Gottschlichet al. 1994). Each task, together w ith alternatives needed because of u n-

certainties in the assembly process, had to be completely spelled out. To

overcome these problems, this work had to be automated.

Before the production of a new or modified product can take place,

it must be preceded by an engineering phase. In this phase, a design or

re-design of a p roduct is executed, resulting in a model of the produ ct to

produ ce. Together w ith this model, a plan is made, d escribing h ow the

produ ct can actually be p roduced. The key for automating the assembly

planning lies in the use of the product model information for assembly

analysis and planning.

This thesis will focus on the automation and integration of modelling

and plann ing, especially for assembly. With the u se of new techniqu es us-

able in assembly mod elling and plann ing, the autom ation of the assembly

process can make another step forward.


14/173

2 Introduction

1.1 Assembly modelling

In assembly mod elling, a mod el is created rep resenting a prod uct consist-

ing of several smaller comp onents. Because of th ese smaller comp onents,

the focus in assembly mod elling w ill be not only on th ese comp onents, butalso on the relations between these comp onents. A comp onent tha t cannot

be subdivided into smaller components is called a single part. A group of

components merged together is called an assembly.

Decisions made during the creation of a model can have great impact

on the complete life cycle of the product. Wrong d ecisions m ade d ur-

ing design can be responsible for time- and money-consuming product

re-designs. To avoid such re-designs as much as possible, a designer has

to take into account requirements from other disciplines involved in the

life cycle of the prod uct. Involved d isciplines can be, among others, from

the start of the life cycle to its end: marketing, design, manufacturing, as-sembly, service and fina lly disassem bly, see Figure 1.1.

model

product

disassembl

y

marketing

desig

nmanuf

actu

ring

assembly

service

Figure 1.1: Prod u ct life cycle

Several analyses can already be performed during modelling to check

w hether the requirements are met. So, modelling and analysing have to be

integrated. Because of the analysing step, modelling cannot be restricted

to the design department only. It is possible that a designer is responsible


15/173

1.2 Assembly planning 3

for the final d esign of a prod uct, but the m odelling p rocess itself still needs

input from many other disciplines.

To fulfill all different, and sometimes conflicting, requirements, the de-

signer can u se the design for X (DFX) concept, wh ere the X can be any life

cycle phase (Boothroyd 1987, Andreasen et al. 1988, Boothroyd et al. 1994).

In DFX, already during design, requirements from X are taken into ac-

count, w here the X can, for example, stand for: Manu facture in DFM, As-

sembly in DFA, or Service in DFS. To realize this, the designer must have

knowledge of many issues involved in the disciplines or, and this is more

likely, the other d isciplines m ust cooperate du ring d esign. The latter is

usually meant by the term concurrent engineering.

The difference between single-part modelling an d assembly modelling lies

in the existence of several components in an assembly, together with as-

sembly relations. Because of these ad ditions, both the d esign an d the an al-

ysis steps within assembly modelling are more complex than for single-part mod elling. There are more d isciplines involved in assembly m od-

elling th an in single-par t m odelling, p roviding more (conflicting) require-

ments to satisfy. Due to th e existence of single p arts in every assembly,

assembly modelling cannot be separated from single-part modelling. The

term product modelling is used to refer to both.

1.2 Assembly planning

Modelling alone is not enough to prepare the p rodu ction of a produ ct. Theanalyses within modelling check whether the requirements can be met,

but do not exactly determine how they are met. This is determined within

planning.

Planning is responsible for the creation of the p rodu ction p lans, which

specify how the product can be produced.

In assembly p lanning, several plans are created specifying h ow the

product can be assembled given the product model.

The d ifference between single-part plann ing or manufacturing planning

an d assembly plann ing is comparable to the difference betw een single-part

modelling and assembly modelling. Product planning is therefore used to

refer to both.

Assembly plans are very difficult to generate, and often this is still done

man ually, consu ming a large am oun t of time. This time can d irectly effect

th e time-to-market, the time of a prod uct from its preliminary requirements

specified by marketing, until it finally can be sold on that market.


16/173

4 Introduction

In particular for smaller volumes, the time needed for assembly plan-

ning becomes more and more important. Such volumes are most of the

time produced on already available machines. These flexible assembly sys-

tems are capable of assembling families of products, in a flexible batch

order. During d evelopment of the produ ct, the time needed to generate

assembly plans is h ere relatively imp ortant.

At Delft University of Technology, a prototype flexible assembly cell,

th e DIAC (Delft Intelligent Assembly Cell), has been developed (Meijer

and Jonker 1991). The main goal was to develop a cell capable of assem-

bling a large variety of product families, in relatively small batches. This

research has been a collaboration between several faculties of Delft Uni-

versity of Technology: Mechanical Engineering, Applied Physics, Electri-

cal Engineering and Technical Mathematics and Informatics.

One of the problems arisen in this project was the difficulty to link all

developed planning modu les. There was no p redefined way for the m od-ules to retrieve need ed information or to store generated information. The

work described in this thesis is one of the spin-off projects of the DIAC

project, and is mainly concerned w ith this p roblem.

1.3 Product models

Both product modelling and planning are highly dependent on well de-

fined product models. Product models can be seen as information carriers

for m odelling and planning. Examples of information stored in prod uct

models are the geometry of the product and the used material. By com-bining information, other information can be generated that can also be

stored in the product model, e.g. volume, weight and center of gravity.

Nowadays, product models used in modelling and planning are hardly

integrated. But also th ere is hardly an y integration between single-part

mod els and assembly models. This results in several different p roduct

models for a product, at least one for every discipline involved in mod-

elling and planning.

Different product models give rise to severe problems, because of re-

du nd ancy of stored d ata and because of loss of information du e to conver-

sions between models. This makes it extremely difficult for one discipline,

to understand why certain decisions were taken by another discipline.

A possibility is the use of one integrated product model by all disci-

plines involved. Information is stored only once in the produ ct model,

resolving the problems due to redundant data. This makes it possible that

information can be stored by one discipline, and can be used by other


17/173

1.4 Main objective 5

disciplines. Also there is no unnecessary loss of information, because con-

versions from one product model to another are no longer needed.

Nowadays, in single-part models, there is a shift from storing only

geometry-oriented information toward s more fun ctional-oriented infor-

mation. The latter is d one using features in feature-based models. This

functional information is very useful du ring mod elling and planning. In

this thesis the feature-based concept will be ap plied to mod elling and

planning for assembly.

1.4 Main objective

The main objective of this thesis is to d evelop an integrated feature-based

produ ct model to be used in assembly d uring m odelling and planning.

Therefore three issues are involved in this research:

The structure of the product m odel itself. An integrated product

model for assembly is needed to avoid the unnecessary loss of in-

formation during modelling and planning, and to integrate single-

part and assembly models. Features are used because in single-part

models they have shown their benefits, and it is expected that this

w ill also be the case for assembly m odels.

The w ay this produ ct model can be used d uring assembly mod elling.

Therefore a p rototype system is built, in w hich the feature-based in-

tegrated model can be created.

The w ay this produ ct m odel can be u sed du ring assembly planning.

The same prototype system also contains planning functionality, in

which planning activities can be activated directly from the mod-

elling environment. It is evaluated wh ether the used features pro-

vide benefits d uring p lanning.

1.5 Overview

To describe the research done to realize the objective, the thesis is divided

into the following parts:

Part I backgroun d, gives an overview of comm only used mod elling tech-

niques in Chapter 2 and planning techniques in Chapter 3. After

this overview, the long-term goals are defined in Chapter 4, to pro-

vide a direction for future research and development in this area.


18/173

6 Introduction

These long-term goals are the starting points for the short-term goals,

which provide more clearness on the chosen solutions in the rest of

the thesis.

Part II modelling with assembly features, focusses on the m odelling en-vironment. First, in Chapter 5, a new object-oriented feature-based

produ ct m odel is presented. This m odel combines elements from

single-part and assembly modelling.

Thereafter, in Chapter 6, the focus is on assembly features, in the

described product model to keep track of the assembly information

of a produ ct.

In Chapter 7, a prototype assembly modelling system is described.

Within this system, assembly models can be created and manipu-

lated.

Part III planning w ith assembl y features, discusses the use of the object-

oriented feature-based product model with assembly features in as-

sembly planning m odu les.

It is not the intent to give a detailed description of every planning

module needed within assembly planning. Some have been chosen,

to verify the concept. Besides this verification, new and extended

planning algorithms are presented to show additional benefits of the

produ ct mod el.

First, in Chapters 8 and 9, extensive descriptions of using the prod-

uct m odel in stability an alysis and grip p lanning are given. This isfollowed by a discussion, in Chapter 10, of how the prod uct mod el

can be used in motion planning and assembly sequence planning.

Part IV concludi ng remarks, is the final part containing the conclusions

and future work in Chapter 11.

To avoid confusion about used terminology, Appendix A has been in-

cluded.


19/173

Part I

Background

Part I gives an overview of comm only u sed m odelling tech-

niques in Chapter 2 and planning techniques in Chapter 3. Af-

ter this overview, the long-term goals are defined in Chap ter 4,

to provide a direction for future research and development in

this area. These long-term goals are the starting points for the

short-term goals, which provide more clearness on the chosen

solutions in the rest of the thesis.


20/173

8


21/173

Chapter 2

Assembly modelling

For centuries, products have been designed with the sole use of technical

draw ings on a piece of pap er. With the recent introduction of compu t-ers, new possibilities became available, includ ing th e p ossibility to mod el

prod ucts w ith the aid of a comp uter. This chap ter w ill describe techniqu es

available for this, with the focus on assembly m odelling.

2.1 Top-down and bottom-up modelling

There are two main approaches in which one can create a product model,

the top-down and the bottom-up approach, as was described by, for ex-

ample, Libardi et al. (1988) and Lim et al. (1995).

2.1.1 Top-dow n approach

The top-down approach is based on the designers point of view. The de-

signer initially thinks in an abstract, functional manner to find ways to

satisfy the requirements of the produ ct. The product has to fulfill some

specified m ain function. By recursively dividing th e main function into

sub-functions, the designer generates a produ ct m odel in a top-dow n de-

sign or m odelling w ay.

The initial functional highly conceptual information is difficult tospecify in a general way, and therefore hard to store and use in a comp uter

environment. Later on, at the sub-function level, m ore and more d etails

are determined, which can be represented more simply in a general way,

and therefore can be better hand led by a compu ter. At the end , sub. . . sub-

functions can be represented by geometry, which can now very well be


22/173

10 Assembly modelling

stored in a comp uter. Research on how to mod el the highly conceptual in-

formation, sometimes called functional modelling, is in a p reliminary stage.

Examples of this approach are described in Section 2.3.

2.1.2 Bottom-up approach

The bottom-up app roach is based on availability of technologies. In this

app roach, complete highly detailed geometric representations of

components are already available. These representations of the compo-

nents are usually m ade w ith some CAD package, and later on the sp atial

relations between these comp onents are m odelled to specify the comp lete

product.

This approach is now widely used because of the technical possibili-

ties of CAD systems. In the last d ecades, there has been m uch research

in single-part modelling, resulting in many improvements. This researchwas focused on h ow to m odel the geometry of a comp letely detailedcom-

ponent. Modelling techniques like constructive solid geometry (CSG),

boun dar y rep resentations (B-Rep) or hybrids of these, can be u sed for th is.

Functional intents that have led to these detailed components can, how-

ever, not be stored in these representations. In this way the bottom, the

detailed representation of geometry of sub...sub-functions, is first stored

in the prod uct mod el. Thereafter the relations between th e comp onents are

add ed to the mod el. Examples of the bottom-up app roach are described

in Section 2.2.

2.2 Modelling with detailed single-part models

First the focus will be on the bottom-up mod elling app roach, because it is

easier to explain top-down modelling after bottom-up modelling.

The assembly m odels used in bottom-up modelling can be subdi-

vided into tw o d ifferent grou ps (Srikanth and Turn er 1990, Requicha and

Whalen 1991):

hierarchical models

relational m odels.

Both make use of already defined product models of the single parts, and

combine these into an assembly m odel of the complete prod uct. Therefore

a brief description of prod uct m odels for single par ts is given first.


23/173

2.2 Modelling w ith detailed single-part models 11

2.2.1 Single-part models

Compared to assembly modelling, there has been much research in the

area of single-part modelling. For m ore details on the d escribed m od-

els, see Requicha (1980), Mortenson (1985), Mantyla (1988) and Bronsv oortet al. (1991).

solid models

The first single-part models in CAD were in fact computer models of 2D

pap er draw ings. Later 3D computer models, representing the comp lete

topology and geometry (the geometry for short) of parts, were introduced.

These product models were called solid models, and the nowadays mostly

used representations are Const ructive Solid Geometry (CSG), Boundary Rep-

resentation (B-Rep) or hybrid forms of these.

Within CSG, models are built from a collection of primitive solid ob-

jects, e.g. blo cks, cylin d er s an d sp heres. The set op erat ion s u nion , d iffer-

ence and intersection can be applied on these objects to define new, com-

posite objects. The data structure is a binary tree; the root represents the

complete single part, and the leaves represent the primitive solid objects.

The main disadvantage of this method is that there is no explicit informa-

tion in th e representation abou t the faces, edges and vertices of the single

part. This information is needed du ring some analyses in m odelling and

in several modules in planning, both for single parts and for assemblies.

A produ ct model that d oes have explicit information about the faces,

edges and vertices in the single p art is the B-Rep. The d ata structure isa graph structure. Every node in the graph is a boundary element of the

solid object, i.e. face, edge or v ertex, and arcs in the gra ph represent ad ja-

cencies between these elements.

Hybrid data structures of CSG an d B-Rep combine the advantages of

both structures: the convenience of mod elling with CSG structures, and

the availability of explicit information within the B-Rep structures.

feature model s

A disadvantage of solid models is the lack of functional information inthe data structure: only the resulting geometry is stored. The intent of the

designer, i.e. why he has chosen for the specified geometry, is not stored

within solid models. This information is, however, needed during analy-

ses and planning, but also to verify whether certain changes are allowed

on the mod el.


24/173


To overcome this disadvantage, feature models were introduced. Fea-

tures contain, besides geometry, fun ctional information.

Feature m odels represent a produ ct by a set of feature instances, and

between these feature instances constraints can be specified. In literature

there are m any different d efinitions for features, but a common element in

these definitions is that they combine shap e geometry information

w ith fun ctional significance. This fun ctional significance is not restricted

to d esign significance, but can be significance for any involved d iscipline

du ring the p roduct life cycle. Each discipline can have its own way of

looking at a product, called a view. Each view may have its own set of

features (d e Kraker et al. 1995, Bronsvo ort et al. 1996). Features u sed in the

design view, so-called design features, are different from features used in

the manufacturing view, so-called manufacturing features. In cases where

functional significance is only related to a generic shape, the correspond-

ing feature is called form feature.It is un desirable that every view-specific feature m odel is created from

scratch. Several methods have been developed to create and convert fea-

ture models, and to provide consistency between several feature models:

feature recognition The featu re recognition method constructs feature

mod els out of already ava ilable solid mod els, or even 2D CAD draw -

ings. Several techniqu es have been developed to recognize features,

mostly manufacturing features.

One technique uses an (Attributed) Face Adjacency Graph (FA G)

(Joshi and Chang 1988), a graph with the nodes representing thefaces and the ed ges representing the ad jacencies between faces. For

every feature to recognize, a predefined FAG is known. By us-

ing graph-based pattern-matching techniques, features in the solid

mod el are recognized.

Another technique is decomposing the solid model into volumes,

and trying to map these volumes onto known feature volumes. This

technique is used by, among others, Kim (1992).

design by features When completely new product models have to be

generated, one can immediately start with feature modelling. Prod-uct models are created by combining instances from a set of view-

specific generic features, with constraints betw een them . This con-

cept is called design by featu res, but is not restricted to the design

view; a feature model for the manufacturing view can also be cre-

ated with design by features.


25/173


feature conversion The method offeatu re conversion, or feature map ping,

is used to create different feature models of the same product for

several different views. Feature recognition a nd design by features

are combined in this method. A feature m odel is created for one

view, using design by features, and is converted into a feature m odel

in another view by a kind of feature recognition, see, for example,

Dohm en et al. (1996) and d e Kraker et al. (1997).

More d etails about feature m odelling, and links for further reading, can

be found in Shah et al. (1991), Bronsvoort and Jansen (1993) and Shah and

Mantyla (1995).

Now that th e single-part mod els have been d escribed, the focus w ill be

back on combining these models in assembly models.

2.2.2 Hierarchical mode lsIn a hierarchical model, the comp lete produ ct is represented by a tree struc-

ture. Nodes in the tree represent parts (the leaves of the tree), subassem-

blies or the complete product (the root of the tree). The position and ori-

entation of every nod e is specified by a homogeneous transformation

matr ix. This transformation m atrix can be global, i.e. w ith respect to the

w orld coordinate system, or local, i.e. relative to the position and orienta-

tion of the d irect an cestor in th e tree.

One of the disadvantages of hierarchical models is the difficulty of

specifying and calculating the transformation m atrices. Another d raw-

back of the simple h ierarchical mod el is that the m odel d oes not containany information about relations between the individual components.

2.2.3 Relational models

The disadvantage of having to manu ally p rovide the transformation ma-

trices for the hierarchical model, has led to the creation of the relational

model. The only way to automatically calculate the transformation matri-

ces, i.e. the position and orientation of the components in the complete

product, is to provide in some way the relations between these compo-

nents, which resulted in a change from a pure tree structure to a moregraph-oriented structure.

Ambler and Popplestone (1975) and Popplestone et al. (1980) used a

product model with relations between the components to calculate the ac-

tual p osition an d orientation of every comp onent in th e prod uct. These re-

lations were defined between so-called features on components. Features


26/173


were defined here as simple topological entities: planar face, cylindrical

shaft or hole, edge, sph erical face and vertex. They d efined a set of re-

lations, of type against m ating faces w ith op posite normals, coplanar

mating faces with identical normals , and fits fitting of a cylindrical

shaft and hole, and specified how transformation matrices can be calcu-

lated given these relationships.

Wesley et al. (1980) used an extended tree structure in their system

called AUTOPASS (AUTOmated Parts A Ssembly System). Their extended

tree structure allowed additional edges between parts and subassemblies

to represent certain relations. The leaves in this tree structure are not on

the level of single rigid parts, but on the level of so-called sub-parts

primitive polyhedrals, representing shapes on a part. These sub-parts are

an early form of wh at are now called form features.

Lee and Gossard (1985) provided relations between components in a

produ ct m odel with their virtual linkconcept. They use a tree structure torepresent the subd ivision of a produ ct into su bassemblies and parts. In

this tree structure, an add itional graph structure is specified; any mating

pair of two components is represented by a virtual link, see Figure 2.1. A

virtual link is representing the information required to describe the rela-

tion and the mating features between components needed to calculate the

transformation matrices.

product

virtual link virtual link virtual link

virtual link virtual link virtual link

component component

component component

Figure 2.1: Virtual link structure for assembled p rodu cts


27/173


The relations that could be represented with the virtual links were

again very elementary: the against relation a plane-plane mating re-

lation, and a fit s relation a cylindrical shaft-cylindrical hole mating re-

lation. Lee and A nd rews (1985) provid ed an algorithm to autom atically

generate the transformation matrices from these virtual links for every

comp onent in a produ ct. With the algorithm, th e need for the d ifficult

procedure to manually give all the transformation matrices has vanished;

however, before the algorithm can generate the matrices, many of these

basic relations had to be sp ecified.

With the solution of the positioning problem of the components in the

assembly, the research went more and more from only the basic creation of

the mod els towards the use of mod els for analysis, and even tow ards the

generation of assembly p lans. Especially th e assembly sequen ce problem

came into focus, i.e. can we use a product model to generate a possibleassembly sequence for the product.

Partially based on this, development was focused on purely graph-

based produ ct m odels. Within graphs, comp onents are represented by

nodes and relations between components are represented by edges. The

relations betw een comp onents w ere still very elementary. These elemen-

tary relations or basic relations are given by, for example, Homem de Mello

and Sanderson (1989) and Srikanth et al. (1991):

contact Two components have a contact relation if there are non-rigidly

attached faces, edges or vertices between them; i.e. there is some

freedom of motion betw een two comp onents w ith a contact relation.

attachment Two components have an attachment relation if there are

rigidly attached faces, edges or vertices between them; i.e. there is

no longer a freedom of motion between two components with an

attachment relation.

assembly dimension Two components have an assembly dimension if

there is a constraint between them to fully specify some other re-

lations. These dimensions are needed, for example, when a contact

relation between two faces is defined: additional assembly dimen-sions mu st be d efined to un iquely specify the position of one face on

the other. See Figure 2.2 on the following page for an example.

The relations are called elementary in the sense that one has to provide

many of these relations to fully describe the product model.


28/173


Terminology for the elementary relations differs from author to author,

but th ese types are comm only used. Notice, that the d efinition of an at-

tachment given here d iffers considerably from the one given later in Sub-

section 6.4.3, and subsequently used in this thesis.

contact

dimension

Figure 2.2: Examp le of a d imension relation

Sand erson and H omem de Mello (1990) used a g raph representation to

find assembly sequences. They used several representation levels for the

product model:

the solid models of the components, see Figure 2.3(a) on the next

page,

the relational model of the product, where they specified, besides

the set of components, a set of contacts ( ), a set of at-

tachments ( ; in this case eliminating all degrees of freedom ofa contact) and a set of relationships between elements of the set

( ), see Figure 2.3(b).

the connection model of the product, a reduced relational model,

showing only the connections between components, see Fig-

ure 2.3(c).

The difference between a relational mod el and a connection m odel is thus

the level of abstraction. In a connection mod el, only th e connections be-

tween components are defined; these connections can contain several re-

lations. These connection models are similar to the liaison graphs describedby Bou rjou lt; for extensions on h is mod el see De Fazio and Whitn ey (1987),

in which the precedence relations between connections can be given. The

relational model is on a lower abstraction level; here all the elementary

relations between comp onents are sp ecified. The connection m odel can be

generated from the relations in th e relational mod el.


29/173

2.3 Modelling w ith functional information 17

receptaclestick handlecap

(a) solid models

r4

r5r3

r1 r2

r10r6

a1

r7r9

a2

r13 r14

r12

r8

r11 c5

c1

c2 c3 c4

(b) relational model

3 42

1

5

(c) connection model

Figure 2.3: The product model representations used by Sanderson and

Hom em d e Mello

There exists much variety in product models using a graph for rep-

resenting the relations, but most are similar to the one described here.

Some use feature models for the components, and relations are then de-fined between features on the components. Roy and Liu (1988) already

made use of form features for their single-part components models. They

use a kind of Face Adjacency Graph to represent their single-part feature

mod els. Relations between different components are specified between

these features, and are called functional relations. The generated product

model is called a Functional Relationship Graph (FRG). Later t his ha s been

extended to the Modified-FRG (M -FRG) in Roy et al. (1989), where they

showed how information stored in the model can be used for elementary

assembly planning.

2.3 Modelling w ith functional information

In the previous section, we described the bottom-up approach of mod-

elling assemblies. Now we focus on the top-down approach.


30/173


In their Systematic App roach, Pahl and Beitz (1988) divided the d e-

sign process into four ph ases:

product planning and clarification of task,

conceptual d esign,

embodiment design and

detailed design.

Each phase takes as input the results from the previous phase. Now the

main problem of the bottom-up modelling approach becomes clear de-

tailed models of components cannot be generated before their function-

ality w ithin the prod uct, and their relations w ith other comp onents have

been specified. This can be done w ithin the conceptual d esign an d em-

bodiment d esign p hases by functional modelling.

A functional model represents a product structure from highly ab-

stract to concrete, i.e. from undetailed to highly detailed (Henderson and

Taylor 1993). To u se a fun ctional m odel in d esign, th e flow of the w ork

should preferably also go from abstract to concrete. This looks obvious,

but sometimes the flow goes from bottom to top: specifying a functional

structure from the d etailed p rodu ct description to th e conceptual level.

A problem in functional modelling is how to store the abstract and

und etailed information. There mu st be a possibility to compare stored

designs, in such a way that designs or pieces of designs can be re-usedin new designs. But on the highest level, the functions are represented

as black boxes, describing only (textually) the task of the function. This

makes it hard to find a unique way to describe a specific function in some

kind of language, and difficult to search for a conceptial solution earlier

used . H owever, some techniques d eveloped in artificial intelligence can

be used to find similarities between solutions.

Although functional mod els are in an early stage of developm ent, there

are some promising examples.

Mantyla (1991) described a modelling environment for functional, con-

ceptual and detailed design. Within the p rototype environment, called

WAYT (Why-Are-You -There?), a hierarchy of functions can be specified,

with relations between them. Feature models of the comp onents can be

associated to these functions.

Gu i (1993) described th e system, a computer environment for top-

dow n functional modelling. A m ulti-graph structure is used to represent


31/173

2.3 Modelling w ith functional information 19

both functional and relational models. Every node in a multi-graph repre-

sents a (conceptual) comp onent or a (conceptual) connector betw een com-

ponents. Therefore every node can be a multi-graph itself. In this way, a

hierarchical structure can be created from abstract to concrete. Both con-

ceptu al comp onents and connectors can be subdivided into other (concep-

tual) components and connectors. On the lowest, detailed level, compo-

nents represent single parts an d connectors represent elementar y relations

between these parts. An example is given in Figure 2.4.

component

connector

concrete

abstract

Figure 2.4: The multi-graph structure used in the system; components

are represented by blocks and connectors by circles

And reasen (1992) and Mortensen a nd And reasen (1996) divide a prod -

uct model into domains, a way of looking at something:

functions describe the effects that the p rodu ct is to create,

organs describe the entities that create the effects, and

parts are the m aterialization of the organs, i.e. the detailed p arts.

Within th is subd ivision, fun ctions, organs an d par ts are all described by a

hierarchy; so there is a hierarchy of functions, a hierarchy of organs and

a hierarchy of parts. Between these hierarchies, there exists also relations.


32/173


A function can be realized by several organs, and an organ can belong to

several fun ctions. Also an organ w ill norm ally n eed several par ts to fulfill

its realization, and a part will contain several details belonging to sev-

eral organs. These interrelations w ith a m any-to-man y character make the

functional models extremely complex, as can already be seen in a simple

example in Figure 2.5.

systems

sub-

partsorgansfunctions

total

elements

Figure 2.5: An examp le of function, organ, and part hierarchies, with their

complex interrelations

Another point that is stated by Andreasen is the difficulty in dividing

functions into su b-fun ctions. Before the function can be d ivided into sub-

functions, the used means the technology used to realize this function

has to be defined . Withou t choosing a specific mean , you cannot subd ivide

a function into sub-functions. Every p ossible mean for a function, will

result in a specific subd ivision. This results in a function/means tree.

Although a good theoretical foundation is given for the top-down

modelling process by Andreasen (1992), there is still a lack of proper com-

puter tools to support this modelling concept.


33/173

2.4 Assembly features to fill the gap 21

2.4 Assembly features to fill the gap

In this chapter, two concepts for modelling have been described, the

bottom-up and the top-down concept. The first has the problem of adding

higher-level information and relations to the product model of highly de-tailed comp onents. The latter has the problem of converting highly ab-

stract information to d etailed geometry information. We did not even con-

sider the p roblems in au tomatic conversion of functions to geometry!

There is a large gap between the detailed geometry information and

the elementary relations on the one hand , and the abstract functional in-

formation on the other han d. This gap is not only present w ithin assembly

mod elling, but also w ithin single-part m odelling.

The way a designer has to specify a part in a CSG or B-Rep model does

not correspond to his way of thinking. H e prefers a m ore abstract concept

to specify his product, and this was to some extent provided with th e in-trodu ction of feature m odelling. Features are on a higher abstraction level

than geometric elements, and features provide the possibility to contain

functional information.

Several peop le wh o n oticed the p ossibilities of using features in single

part modelling, also tried to apply the feature concept within assembly

modelling.

De Fazio (1990) described a prototype feature-based DFA system,

where form features for single-part modelling were used together with

features to specify the m ating relations between comp onents. The lat-

ter features provid ed add itional assembly-specific information such as

d egrees of freedom and relative extraction directions to the elementar yrelations normally used within assembly planning.

Although the ad ditional information could be u sed in assembly plan-

ning, these features did not bring assembly modelling to a higher abstrac-

tion level. Still all the elementary relations had to be specified separately.

The first appropriate use of the term assembly featu re was by Sodhi

and Turner (1991). Before that the term w as sometimes u sed, but only

to specify, in fact, elementary relations (Popplestone et al. 1980, Lee and

Andrews 1985).

Sodh i and Turn er u sed assembly features for specification of relations

between comp onents on a h igher abstraction level. In their opinion, as-sembly features served as a higher-level interface capturing assembly

relations at the functional level, and removing from the designer the bur-

den of identifying the und erlying elementary relations. See, for exam-

ple, Figure 2.6 on the following page, where a Pin Joint assembly feature is

shown. The designer can, with this feature, specify several form features


34/173


and several elementary relations in one step.

contacts

alignment

hole pin

cylindrical

planarcontacts

features feature

Figure 2.6: Pin Joint assembly feature

Shah and Tadepalli (1992) defined an assembly feature as an associa-

tion between two form features on different parts. They u sed assembly

features as an abstraction to specify several elementary relations.

They also pointed out that assembly features could be used within a

design-by-features concept. When a d esigner has chosen tw o form fea-

tures, the system can provide the designer with possible assembly fea-

tures. Although this is a bottom-up approach, it can be very useful.

Later, Shah and Rogers (1993) compared assembly mod elling withsingle-part modelling. They found that the two modelling concepts were

very comparable. A single p art can be thought of as an assembly of

form features, with mutual relations. The constraints used between form

features in single-part m odelling, can also be u sed as elementar y relations

in assembly m odelling. By combining su ch elementary relations, one can

generate h igher-level relations called assembly features. These assembly

features can then be extended with additional assembly-specific informa-

tion, such as d egrees of freedom and fit information.

Assembly features as defined here w ere only used to ease mod elling.

They were closer to a designers way of thinking: with one assembly fea-ture he could specify several elementary relations. The information stored

within these features was not yet used for analysis of the model, nor cre-

ation of assembly plans.

Assembly features combine information abou t assembly an d geometry,

i.e. both abstract functional and detailed geometrical information. There-


35/173

2.4 Assembly features to fill the gap 23

fore they can be used to fill the gap between abstract functional and de-

tailed geometric models. It seems easier to link pure functional models to

geometric models through assembly feature models, than without them.


36/173



37/173

Chapter 3

Assembly planning

In Chapter 1, it has already been described that automation of the gen-

eration of assembly plans is needed, will the assembly process itself beeffectively automa ted. In this chap ter, mod ules needed w ithin assembly

planning to analyse a model and to generate assembly plans will be de-

scribed.

Generally, assembly planning does not start at the moment that the

design process h as been completely finished. In the contemporary DFX

concept, already in an early phase of the d esign p rocess the designer takes

into account requirements from other disciplines involved in the prod-

uct life cycle, see Figure 1.1 on page 2. During the design phase, certain

steps in the planning for other disciplines are done simultaneously, thus

analysing the proposed produ ct model. The generated analysis informa-

tion should be stored in the product model. Modelling and planning arenot completely separated, but concurrently performed by several disci-

plines that are closely related and simultaneously operating on the same

product model.

But even within one discipline, for example assembly, there exists sev-

eral kinds of p lanning activities. For all these activities, the prod uct m odel

is analysed to see w hether the activity can be p erformed. Each differ-

ent activity has its own module to perform the analysis. The next section

w ill briefly d escribe some of these mod ules u sed in assembly an alysis and

planning.

3.1 Assembly planning modules

In literature, the term assembly planning is almost exclusively used for

assembly sequence planning (Hom em de Mello and Sanderson 1991b,


38/173

26 Assembly planning

Wolter 1991, Delcham bre 1992). Although assembly sequence p lanning

finding feasible sequences in which the product can be assembled

is a very important planning module in assembly planning, there exists

many other m odules (Nevins and Whitney 1989). In Figure 3.1, several

planning modules are shown. It is indicated that assembly sequence plan-

ning is highly depen dent on these other mod ules as well, especially w here

these mod ules concern about the assemblability of an assembly.

planning

stabilityanalysis

planning

feeding

subassembly

planning

gripplanning

planning

sequence

assembly

planningfixture

motion

Figure 3.1: Assembly sequence planning and its relation with some other

planning modules

The following list briefly enumerates assembly planning modulesfoun d in literatu re (H eemskerk 1990, Marten s 1991, Boneschan scher 1993,

Lee 1994, Gottschlich et al. 1994), with their specific goals.

fixture planning A fixture planner determines fixtures, together with

base components to be used for assembling. A base component is the

first component of an assembly that is assembled. Fixture planning

determ ines the base comp onents leading to a minimal nu mber of fix-

turing setups d uring assembly and providing a m aximal set of com-

pon ent app roach d irections. Anoth er problem to solve for the fixture

planner is to determine the optimal number of base components thatcan be placed on one fixture at a time.

feeding planning A feeding planner determines usable feeders for the

components to assemble. The planner m ust take into account the

approach directions of the components on the feeder. In connection


39/173

3.1 Assembly planning modules 27

with the nu mber of base components on a fixture, the feeding plan-

ner must find the optimal number of components on one feeder, to

eliminate u nnecessary feeding op erations. Feeding is more time crit-

ical in assembly, comp ared to m anu facturing, because assembly op -

erations use relatively short times, comparable with feeding times.

Therefore, the components to assemble must be continuously fed,

otherw ise the assembly cell w ill be idle.

stabili ty analysis A stability an alyzer checks w hether a certain assembly

is stable and thus can be used as partial assembly or subassembly.

The analyzer takes into account different stability conditions: static

conditions only gravitational forces, transport conditions forces

due to accelerations because of transport, and assembly operational

conditions forces due to the assembly operation itself, i.e. addi-

tional forces n eeded to establish contacts. Chapter 8 will describe

this topic in more d etail.

grip planning A grip planner tries to determine appropriate tools for

gripping the components, and the areas on the components where

to grasp them. To find these areas, also informa tion concerning u sed

feeder, fixture and already assembled partial assemblies is derived.

Chap ter 9 will describe this top ic in more d etail.

subassembly planning A subassembly planner tries to divide the com-

plete product into subassemblies. The main requirement for sub-

assemblies is that they remain stable when manipulated, but other

requirements can be add ed, such as that the subassembly mu st ful-

fill some functionality, or is importan t for service pu rposes.

gross motion planning Gross motion planning is the first phase of mo-

tion planning. A gross m otion planner d etermines a collision free

path from the feeding position towards a position near the partial

assembly. Finding a collision free path is also known as the piano

movers problem.

fine motion planning Fine motion planning is the second phase in mo-

tion planning, and starts at the point where the gross motion plan-ner stopped. A fine motion p lanner thus determines an assembly

path from a position near the p artial assembly to the final assembled

position. In contrast w ith gross motion planning, fine motion plan-

ning will often use contacts to reduce uncertainties in positioning

the component. Fine motion p lanning uses these contacts to lead


40/173


the component to its final position. Section 10.1 will describe motion

planning in more detail.

assembly sequence planning An assembly sequence planner determines

feasible assembly sequences for a product. However, these se-quences are highly dependent on output generated by almost all

other planning and analysis modu les. Sometimes, when problems

occur during assembly, it is needed to create an ad-hoc sequence,

used to fin ish the a ssembly as far as p ossible. Section 10.2 will de-

scribe this topic in m ore d etail.

scheduling A scheduler determines an optimal assembly sequence for a

complete batch of products. Out of a set of feasible sequences gen-

erated by th e assembly sequen ce plann er, the sched uler m ust choose

optimal sequences to be used d ur ing actual assembly. Therefore

resources must also be allocated by the scheduler. This can only bedone when, for example, the fixture, feeding and grip planners have

already selected their resources.

It can be preferable to do assembly sequence planning and schedul-

ing simultaneously, to make use of the dependencies between these

mod ules and to d ecrease the exponential complexity. Searching fea-

sible sequences is related to available resources known already for

the schedu le. By exploiting this information, the search space can be

restricted.

In add ition to these plann ing mod ules, there can be several other m odu les

within assembly p lanning, su ch as sensor planning, d ynamical analysis,

functional analysis, assembly line planning, etc. These will not be elabo-

rated here.

The question may rise, whether there exists some fixed execution se-

quence for these mod ules, so that all required information will be avail-

able at the right moment. As can be seen in the description of assem-

bly sequence plann ing, this is not always the case, because esp ecially this

mod ule is highly interdependent and therefore very intertwined with all

the oth er m odu les concerning th e assemblability.

3.2 Grouping assembly planning modules

This section will describe some proposals made to group assembly plan-

ning modules, in such a way that some ordering between groups can be

found.


41/173

3.2 Grouping assembly planning modules 29

3.2.1 On-line and off-line planning

One way to divide the planning modules is into on-line an d off-line mod-

ules (Gottschlich et al. 1994).

In essence, the on-line modu les are depend ent on information gener-ated d uring the assembly process. On-line planning mod ules therefore

also generate their output during the actual assembly process. The output

is used almost immed iately w ithin that p rocess. On-line planning mod -

ules are therefore time critical the longer their required planning time,

the larger the risk that the assembly process, executed in parallel, has to

wait.

The off-line planning modules are less time critical. They are less de-

pend ent on actual information, and therefore can generate outpu t before

the assembly process is executed .

How ever, it is difficult to strictly separate the planning m odu les into

on-line and off-line modules. Some of the planning modules can be used

both on -line and off-line. For examp le, the schedu ler can m ake a schedu le

off-line for a batch of products, given some specified flexible assembly cell

with resources. However, during the actual execution of such an off-line

generated schedule, it must be able to generate a schedule on-line, for ex-

amp le when sp ecific resources are not available or d amaged (van H olland

et al. 1992).

3.2.2 Using abstraction levels

Heemskerk (1990) suggests a hierarchical reference model with four levels

of abstraction, and places the modules in such a level, see Figure 3.2 on the

next page.

batch leve l The batch level generates plans for a complete batch of prod-

ucts to be assembled.

product lev el The p roduct level deals only with one (type of) product of

the complete batch at a time, and generates plans for this product.

part level The part level deals only with one part (component) from aproduct at a time, and generates plans for this part.

primitive level The primitive level generates plans for four primitive ac-

tions that must be executed on each part during assembly: feed,

grasp, move and mount.


42/173


level

part

level

primitive

level

batch

level

part A

part B

product X

product Y

30 products X

20 products Y

product

feed A

grasp A

Figure 3.2: A hierarchical reference m odel for assembly p lanning m odu les

Although these abstraction levels provide an overview of the information

needed within assembly planning, there are some disadvantages of this

mod el (Martens 1991, Boneschanscher 1993). The main disadv antage is

the restriction of placing planning modules at one abstraction level only. It

w ill be incorrect to p lace, for example, assembly sequence plann ing, at th e

product level only. By doing so, the system will provide the scheduler with

only one assembly sequence plan for a specific product, which may result

in suboptimal schedules on batch level. Better results are retrieved when

assembly sequence planning and scheduling cooperate with each other.

This will decrease the total search space needed to find proper solutions

(van Holland et al. 1992)

3.3 Experiences w ith existing assembly planners

Mostly, assembly p lanning modu les are d escribed independently from

each other. Only a few descriptions of systems have been given that con-

tain several planning modules. Two of these systems are the DIAC system


43/173

3.3 Experiences w ith existing assembly planners 31

and the Archimedes 2 system. In the following subsections, some experi-ences with these systems are d escribed.

3.3.1 Experiences in D I A C

In Section 1.2, it h as already been described that the w ork in this thesis is

a spin-off from the DIAC (Delft Intelligent Assembly Cell) project (Meijer

and Jonker 1991).

Several prototype p lanning modu les have been d eveloped du ring this

project, for examp le gross m otion p lanning (Martens 1991), assembly se-

quence planning, scheduling and batch planning (Boneschanscher 1993),

and fine motion planning and grip planning (Baartman 1995).

The product models used the Product Data Model (PDM) and the

Connection Model (Martens 1991) were based on low-level geometric

informa tion for the componen ts, with elementary relation information be-

tween these. The PDM contained information about:

Insertion Point, a point n ear the p artial assembly w here the assembly

of a component can start.

Final Point, a point on the partial assembly where the component to

assemble must be positioned.

Insertion Path, the path to bring the component to assemble from

Insertion Point to Final Point.

Ap pr oach Direction Set, a set of d irections available to bring the com-

ponent from a place relatively far from the Insertion Point to the In-

sertion Point.

These attributes had to be entered m anually. A d isadvantage of the used

produ ct m odel was the lack of ad ditional assembly-specific information

required for the p lanning modu les. So, every m odule had its own pro-

cedures for geometric reasoning to analyse the product geometry on spe-

cific assembly information. This information w as stored locally w ithin the

module, and could not be used by other modules. In addition, there was

no description of how results from a module could be stored, for use in

other mod ules.

So, although several planning modules were provided in the system,

the mod ules were not w ell integrated.


44/173


3.3.2 Experiences in Archimede s 2

At the Sand ia National Laboratories, the Archimedes 2 mechanical assem-bly plann ing system h as been d eveloped (Kaufman et al. 1996)1. The sys-

tem is able to read commonly-used CAD data files, which makes it easierto work with more complex products. The modules available within the

system are: mating planning, assembly sequence planning and task plan-

ning.

Ames et al. (1995) have described some shortcomings of the planner.

Some p lanning still has to be entered by h and : grip p lanning, fixture p lan-

ning, fine and gross motion plann ing and stability analysis. An other prob-

lem is the integration of all planning modu les, including the p roblem of

resolving conflicting constraints betw een m odu les.

Further there is no possibility to store non-geometric data in the used

produ ct m odel, which is sometimes needed in mod ules. This auxiliarydata has to be computed from the product model, which is very difficult

and time-consuming. In a way this is also superfluou s, because du ring

product modelling this information was already, at least partially, known

to the designer.

Finally, there is no possibility for an engineer to interact with generated

plans d uring the planning process.

3.4 Experiences in manufacturing planning

The problems described are not unique for assembly planning, and some

of them can also be found in a closely related area: manufacturing plan-

ning.

Here the lack of manufacturing-specific information in the used low-

level geometric product m odels can be found . The u se of higher-level

feature models solved a major part of these problems. With feature mod-

elling, see Subsection 2.2.1, the product models were converted to or even

designed by m anu facturing features. These manu facturing features con-

tain, besides a description of the shape, manufacturing-specific informa-

tion that can directly be used within manufacturing planning.

See Shah and Mantyla (1995), van Houten (1991), Rosen (1992) andJasperse (1995) for an overview of manufacturing features and their use in

manu facturing planning.

1See their hom e page with d escriptions of the system at: http://www.sandia.gov/2121/

archimedes/archimedes.html


45/173

3.5 Features in assembly planning 33

3.5 Features in assembly planning

As described in this chapter, most of the product models used in assem-

bly planning are not able to directly provide the information n eeded for

assembly planning. The promising results in manufacturing planning us-ing prod uct m odels containing m anufacturing features, did not yet have

much influence in assembly planning.

However, there are some authors describing features used within as-

sembly m odelling, as described in Section 2.4. Their assembly features

were mainly used to provide a higher modelling level, reducing the time-

consuming specification of elementary relations.

For example, De Fazio (1990) described a prototype feature-based

design-for-assembly system, where features are used for modelling both

single-parts and assemblies. Later more details of this system were given

in De Fazio et al. (1993), and it became clear that the system could also behelpful in planning.

In their bottom-up modelling concept, the assembly features are ele-

mentar y relations containing ad ditional assembly information, such as d e-

grees of freedom and relative extraction directions, i.e. the op posite of ap -

proach directions. This ad ditional information could successfully be u sed

within several of their prototype assembly p lanning mod ules, as are: the

bottom-up design of the assemblies, the positioning of the components

relative to each other using the features, derivation of feasible assembly

sequences, and performing economic analysis on the found sequences to

select one an d possibly generate a conceptu al assembly line for it. N eeded

informa tion could be d irectly retrieved from the assembly mod el, withou texecuting complex and time-consuming geometric reasoning procedures.

This all brought the assembly models to a higher abstraction level with

respect to planning.

As within manu facturing p lanning, information already kn own du ring

d esign and needed d ur ing planning, could n ow be stored in features in the

product model.

However, information generated during planning should ideally also

be stored in the p rodu ct model. Later on this information can then be

used by other p lanning mod ules.

When m ore information is integrated in th e mod el, also planning mod -ules can be better integrated information can be retrieved from and

stored in the same m odel. Especially intertwined m odules could h ave

mu ch profit from integrated mod els, for examp le a considerable redu ction

in computation time.

Therefore features could also be useful in assembly planning, both to


46/173


store required functional information together with the geometric infor-

mation, and to integrate information used by several planning modules.


47/173

Chapter 4

Towards an integrated modelling

and planning environment

It is not enough to investigate only modelling and planning concepts and

techniques used in the past, to retrieve possible areas for improvements.

When you want to know where to improve, you should know the differ-

ences between where you are now and what you w ant to achieve. This

chapter w ill d escribe a futu re mod elling and planning environment, and

this will provide directions to work at.

The future modelling and planning environment should include con-

cepts th at at least accomp lish th at:

the quality of the produced products improves d ue to more ad-

vanced modelling techniques,

the time-to-market d ecreases due to better plann ing tools.

This chapter will first describe the long-term goals for a modelling and

planning environment. These goals w ill p rovide the overall research di-

rections. The derived short-term goals will be described in Section 4.2.

4.1 Long-term goals

The best way to d etermine the long-term goals is, of course, to investigate

w hat is lacking in current systems, and wh at is needed by users. The maind irection is also d escribed by Hen derson and Taylor (1993). They stated

that th ere should be an integrated environm ent for ph ysical objects the

componen ts and th eir conceptual counterp arts the intent of the com-

ponents.

Key top ics to realize such an integrated env ironment are:


48/173

36 To ward s an i nte grate d mod el li ng an d p lan ni ng e nv iro nmen t

top-down and functional modelling,

integrated single part and assembly modelling, and

integration of mod elling and plann ing activities.In the n ext subsections, these topics w ill be further explained .

4.1.1 Top-down and functional modelling

One of the major disadva ntages of the cur rent m odelling environm ents, is

that the user is provided with a far too simple and limited model, which

cannot handle all information already known by him or her, and that, as

a consequence, information is lost. This information m ust often be recov-

ered again later in the product life cycle through complex calculations,

wh ich wou ld be unn ecessary if it could be stored in the system from thestart (Wilson an d Pratt 1988). For examp le, when a u ser specifies a spe-

cific bolt-nut connection between two plates, he or she already knows the

assembly sequence restrictions for su ch a connection, i.e. the bolt and nu t

connection mu st be established after the bolt and plate connections. H ow-

ever, this information can gen erally not be stored in th e mod el, and there-

fore the time-consuming assembly sequence planning module tries every

combination of bolt, nut an d p lates to find a p roper assembly sequence. At

the end, it will provide the user with a solution that he or she was already

familiar with during modelling.

The main goal is therefore to provide a modelling environment that is

able to store inpu t on a high abstraction level, close to th e w ay of thinkingof the user. The user must be able to create a model in a top-down way,

from conceptual to d etailed. Therefore it m ust be possible to translate

functional information to lower-level, more detailed information. In be-

tw een the top and the bottom level, it mu st be possible to specify partially

detailed information, such as sketches of components and functional faces,

see for example Horvath et al. (1995) and Horvath (1996). The app roach

from Mantripragada et al. (1997) is also been going in this top-down de-

sign direction. They use the most imp ortant characteristics of a produ ct,

so-called Key Characteristics (KCs), to represent the design intent. These

KCs are then used to identify important datums on components, and todefine relationships between them.

One of the major problems in this field is to maintain the consistency

of a top-dow n generated m odel. After changing something at a m ore de-

tailed level of the model, higher levels should be changed accordingly.

This can be extremely difficult, because changes at a detailed level can


49/173

4.1 Long-term goals 37

have several effects on higher levels, and there is no one-to-one mapping

of detailed information to conceptu al information.

4.1.2 Integration of single part and assembly mod elling

As a consequence of the top-down concept, modelling of assemblies and

the corresponding single parts must be highly integrated. Creation of the

component, and finally the single-part, models will be driven by the cre-

ation of the assembly model. Changes made in the assembly, directly in-

fluence the models of the corresponding components.

The relations between comp onents p artially provide the behavior of an

assembly, and can also directly specify some of the geometric properties

of the components. So by providing global shapes, and relations betweenthem, elements of the detailed geometry can be generated.

These possibilities also provide th e system w ith the opp ortun ity to p er-

form some planning activities on a partially detailed model. In a p

assembly features in modelling and planning phd-holland

Documents