A

Methodology for the

Systematic Evaluation of Engineering
Design Objects

A Translation into English of the Original German Thesis

Submitted to the Department of Aerospace Engineering at
Stuttgart University, Germany, in Fulfillment for the Degree
of

Doktor-Ingenieur (Dr.-Ing.)

by
Dipl.-Ing. Stephan Rudolph
born in Riedlingen

Certified by: Prof. Dr.-Ing. Bernd-Helmut Kröplin

Accepted by: Prof. Dr.-Ing. Hans-Jörg Bullinger
Thesis Start: March 1st, 1991
Thesis Due: December 7th, 1993
Oral Exam: August 29th, 1994

Institute of Statics and Dynamics of Aerospace Structures

Stuttgart University
1995
Editor: Prof. Dr.-Ing. Bernd-Helmut Kröplin
Translator: Dominica Smith

ISBN 3-930683-01-6

All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photocopying,
recording, or otherwise, without the prior written permission of the publisher.

c Institute of Statics and Dynamics of Aerospace Structures (ISD)

Stuttgart University, Germany, 1995

By sending a request by email to [email protected] or by writing to Stephan

Rudolph, Institute of Statics and Dynamics of Aerospace Structures (ISD), Stuttgart Uni-
versity, Pfaffenwaldring 27, D-70569 Stuttgart, Germany, a copy of this PhD thesis may be
obtained.
Preface

The research described in this document was done as a research assistant at the Institut
für Statik und Dynamik der Luft- und Raumfahrtkonstruktionen (I SD) at Stuttgart University,
Germany, and was thematically and financially embedded in the framework of the partial
project Design Methodology within the research project “Construction in Engineering - Paths
to an Holistic Approach”. I am grateful to the Deutsche Forschungsgemeinschaft (D FG) and
to the interdisciplinary research organisation Forschergruppe im Bauwesen (F OGIB).

I would like to thank Professor Dr.-Ing. Bernd-Helmut Kröplin for his constant interest, his
benevolent support and abandoned confidence in my research efforts. Also, I would like to
thank all the members of the research group as well as the other research assistants at the
Institute of Statics and Dynamics in Aerospace Engineering (I SD) at Stuttgart University who
helped to make this piece of work possible.

I would particularly like to thank Mr. Manfred Zippel from the Institut für Luftfahrtantriebe
(I LA) for our numerous discussions on the formation of models in the natural sciences and for
various comments on the literature. Also Dr.-Ing. Ernst Schrem for his readiness to discuss
and his constructive comments during the checking of the text and Dipl.-math. Georg Mayer
for his criticism of the given proofs. I am additionally grateful to Professor Dr.-Ing. Ingolf
Grieger for the viewing of the manuscript.

Stephan Rudolph
Stuttgart, December 7th, 1993 [email protected]

Preface of the english edition

Great effort and care has been invested by the translator, Dominica Smith, to obtain a rea-
sonably good translation despite the fact of not having an engineering background. In this
respect, the reader is encouraged to check any occurring problem in understanding either
with the original text (published as: Stephan Rudolph, Eine Methodik zur systematischen
Bewertung von Konstruktionen, Fortschritts-Berichte, Reihe 1, Nummer 251, VDI-Verlag,
Düsseldorf, 1995) or with one of the papers to be published in english [79, 80, 81] about this
work.

The reader is also encouraged to check the permanently provided and updated information
on the World Wide Web via http://www.isd.uni-stuttgart.de/˜rudolph
Illustrations

Both the illustrations 1.1 and 1.2 on pages 13 and 14 are taken from the book “Morris
Lapidus: der Architekt der Amerikanischen Traums” [18], published by M. Düttman and
F. Schneider at the Birkhäuser Verlag, Basel. Reprinting with kind permission of the publish-
ers.
With regard to the photographs of the illustrations 5.38 to 5.43 of the six coffee pots on
pages 88 to 90, taken from the earlier programme of the Rosenthal Company, I would like to
thank Dr. Rolf Garnich from the company, Dr. Garnich Design, Esslingen, for the temporary
loan of the objects. I would also like to thank Dipl.-Ing. Jürgen Streng and his Nikon for his
support during the photographing.
Contents 5

Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Illustrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Abstracts (english, french and german) . . . . . . . . . . . . . . . . . . . . . . 10
1 Introduction 13
1.1 Outline, Evaluation and Design . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2 Computer-Aided Design (CAD) and Artificial Intelligence (AI) . . . . . . . . . 17
2 State of Knowledge 21
2.1 Decision Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.1 Utility Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.2 Multiple Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.3 Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.4 Metrification Postulates . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Model Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Multiple Attribute Decision Making (MADM) . . . . . . . . . . . . . . . 29
2.2.2 Multiple Objective Decision Making (MODM) . . . . . . . . . . . . . . 30
2.2.3 Expert Systems (XPS) . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3 Evaluation Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.1 Analytic-Hierarchical Process (AHP) . . . . . . . . . . . . . . . . . . . 32
2.3.2 Outranking Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3 Assignment Setting 47
3.1 Evaluation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.1.1 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.1.2 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.3 Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6 Contents

4 Solution Approach 51
4.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 Homogeneity Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.2 Model Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.3 Measurability Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1.4 Evaluation Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.1.5 Authorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 Central Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.1 Product Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.2 Pi Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.3 Pi Theorem (algebraic) . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3.1 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3.2 Minimality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.3 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.4 Granularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.5 Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.6 Modularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.7 Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.8 Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4 Epistemological Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.1 Invariancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.2 Homogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.3 Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.4 Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4.5 Abstraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4.6 Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5.1 Area of Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.5.2 Limitations in Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.5.3 Freedom of Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.5.4 Modes of Consideration . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Contents 7

5 Examples of Usage 69
5.1 Classic Dimensional Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1.1 Mathematical Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.1.2 Ascending Bubble in a Pipe . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.3 Object Support Mechanism . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Modeling and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.1 Body of a Car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.2 Pressure Valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.3 Gas Turbine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.3 Groups of Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3.1 Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3.2 Coffee Pots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6 Summary 101
6.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Appendix 104

A Proofs 105
A.1 Product Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
A.2 Pi Theorem (algebraic) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.3 Fundamental Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.4 Dimensionless Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
B Evaluations 111
B.1 AHP Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
B.2 Outranking Relations Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . 113
Bibliography 116
8 Contents

Nomenclature

^^ indefinite set operator

A (k) evaluation matrix of k-th criterion
a positive real factor
aij matrix element
a(ijk) matrix element of k -th criterion
ji matrix element
B description of object
b real factor
j factor of linear equation system
c global concordance number
c requirement level
cj partial concordance number
d global discordance number
d requirement level
dj partial discordance number
E energy
F measurement mapping
F0 measurement mapping
' mapping
'j mapping
g constraints
I unit matrix
I index set
IR interval of real numbers
i index of physical variables or alternatives
j index of evaluation variables or alternatives
Kj scaling factor
k index of alternatives
 coupling matrix
l index of alternatives
 eigenvalue of a matrix
max largest eigenvalue of a matrix
M set of features
Contents 9

Mk feature of feature set

m number of evaluations or alternatives
mi unit of physical dimension
 mapping or scaling
IN set of natural numbers
n number of physical variables or design variables
n dimension of a matrix
pj threshold of indifference
 evaluation space
 dimensionless product
[] dimensional representation of a product
j evaluation variable
1 ; : : : ; m set of evaluation variables
qj preference threshold
IR set of real numbers
IR+ set of positive real numbers
Rkl rank order of two alternatives k and l
r rank or rank order
S set of objects
U utility function
Uk utility function of k-th alternative
uj weighting function of j -th criterion
vj veto threshold
w weighting
wj weighting
X space of physical variables
X lower bound of physical variable
X upper bound of physical variable
Xk alternative k
x(k) eigenvector of k-th criterion
xi physical variable
x1 ; : : : ; xn set of physical variables
zj evaluation, utility function
zj(k) evaluation, utility of k-th criterion
10 Contents

Abstract

A systematic evaluation method for the design process will be of increasing importance in
the future, since the demand and the capabilities of software systems which support the
designer are rapidly growing. For such a systematic evaluation method, it has to be first
proved how the evaluation of parts and components in the design process can be found and
represented () evaluation problem). Secondly, the way of combining the different partial
results into a single final evaluation in the presence of possibly conflicting goal criteria, has
to be determined () aggregation problem). Finally, the principal question of arrangement
and completeness of the defined hierarchy of goal criteria () structuring problem) has to be
answered.

As a possible solution approach, a formal object-model based on the Product- and the Pi-
theorem is suggested. In principle, this approach allows the modeling and evaluation of
objects during the design process. This method allows statements as to the applicability to
specific problem classes and its range of validity, possesses a hierarchy and an interpretation
of the goal criteria as well as a modular extensibility of the evaluation components. Also,
statements can be derived concerning completeness, minimality and consistency of the used
representation. Finally, a sensitivity analysis can be performed and a coupling matrix of the
design variables can be determined.

The derived decision model has some significant advantages over classical decision models
(i.e. the Analytical-Hierarchical Process) but could also be integrated into these. Additionally,
some of the conventional model assumptions can be justified by the results of the suggested
approach.

The most important consequence of the postulated hypothesis “any minimal description in
the sense of the Pi-theorem is an evaluation”, is the surprising result that the problem of
evaluation can be principally reduced to a problem of description. Therefore, the problem of
evaluation is solved exactly in those cases where a complete description exists.

The potential of the suggested approach is demonstrated and discussed using several exam-
ples taken from physics as well as from engineering backgrounds. Due to a closely related
mathematical formulation, an interesting link to the technique of “qualitative reasoning” in
artificial intelligence is established.
Contents 11

Résumé

Une méthode d’évaluation systématique dans le processus de conception deviendra dans le

futur de plus en plus importante grâce aux diverses possibilités d’application des systèmes
de logiciel de conception assistée par ordinateur. Dans ce cas, doit être non seulement mis
en lumière comment l’évaluation des composants est effectuée et representée () problème
d’évaluation), mais aussi déterminé comment l’agrégation des différentes évaluations pour
une évaluation globale dans la présence éventuelle de buts multiples et contradictoires est
réalisée () problème d’agrégation). En outre, la question essentielle de la structure et de
l’intégralité des critères de hiérarchie utilisés reste à résoudre () problème de structure).

Comme approche possible pour la problématique d’évaluation sera proposé un modèle

formel basé sur les théorèmes de produit et de Buckingham (Pi-Théorème). Avec ce modèle,
les propriétés des objets du domaine de l’ingénieur pourront être en principe entièrement
modelisées et évaluées. La méthode proposée permet des énoncés exacts sur les classes
des problèmes modélisables et sur les domaines de validité, crée une structure hiérarchique
ainsi que la structuration du contenu et l’élargissement modulaire de chaque composant
d’évaluation. De la même façon résultent de cette méthode des enoncés sur l’intégralité,
minimalité et consistence de la forme d’évaluation utilisée ainsi qu’une analyse d’influence
et une matrice de couplage des variables de conception.

Par rapport à des procédés classiques d’évaluation comme le processus hierarchique analy-
tique, il en résulte de nets avantages, bien que la méthode d‘évaluation proposée soutienne
la construction de modèles conventionels à cause de leurs nombreuses caractéristiques
identiques et mathématiquement démontrables et puisse être aussi intégrée dans ces
mêmes procédés classiques d‘évaluation.

La conséquence la plus importante de l‘hypothèse postulée, justifiée par un raisonnement

scientifique et confirmée par différentes vérifications, que “toute forme minimale de des-
cription dans le sens du Pi-théorème est une évaluation”, est le résultat surprenant, que la
solution du problème d‘évaluation peut être réduite à la solution du problème de descrip-
tion. Ainsi le problème d‘évaluation est exactement résolu dans le cas où une description
complète est connue.

Le potentiel d‘application de cette approche est discuté et mis en évidence à l’aide des
exemples du domaine de l‘ingénieur. De plus, il résulte de la notation mathématique utilisée
une liaison interdisciplinaire supplémentaire à la technique de “qualitative reasoning” dans
le domaine de l‘intelligence artificielle.
12 Contents

Kurzfassung

Eine systematische Bewertungsmethodik wird im Entwurfsprozeß wegen der vielfältigen An-

wendungsmöglichkeiten von Softwaresystemen zur Unterstützung des Entwerfers in der
Zukunft immer mehr an Bedeutung gewinnen. Dabei muß nicht nur geklärt werden, wie
die Bewertung von Teilkomponenten () Bewertungsproblem) erbracht und dargestellt wird,
sondern es ist auch die Aggregation von Teilbewertungen zu einer Gesamtbewertung unter
möglicherweise konkurrierenden Mehrfachzielen () Aggregationsproblem) zu bestimmen.
Darüber hinaus ist die prinzipielle Frage der Struktur und der Vollständigkeit () Gliederungs-
problem) der verwendeten Kriterienhierarchie zu klären.

Als möglicher Lösungsansatz für die Bewertungsproblematik wird ein auf dem Produkt-
und dem Pi-Theorem beruhendes formales Objektmodell vorgeschlagen, mit dem Eigen-
schaften von Objekten im Ingenieurbereich im Prinzip durchgängig modelliert und bew-
ertet werden können. Die vorgeschlagene Methodik ermöglicht exakte Aussagen über die
modellierbaren Problemklassen und Gültigkeitsbereiche, ergibt eine hierarchische Struktur,
die inhaltliche Gliederung und die modulare Erweiterbarkeit der einzelnen Bewertungskom-
ponenten. Ebenso folgen aus der Methodik interessante Aussagen über Vollständigkeit,
Minimalität und Konsistenz der verwendeten Bewertungsrepräsentation sowie eine Sensi-
tivitätsanalyse und eine Kopplungsmatrix der Entwurfsvariablen.

Gegenüber klassischen Bewertungsverfahren wie dem Analytisch-Hierarchischen Prozeß

ergeben sich dadurch einige deutliche Vorteile, obgleich die vorgeschlagene Bewertungs-
methodik aufgrund einiger identischer und mathematisch beweisbarer Merkmale den Aufbau
konventioneller Modelle unterstützt und auch in diese integriert werden kann.

Die wichtigste Konsequenz aus der aufgestellten, erkenntnistheoretisch begründeten und

durch verschiedene Kontrollen bestätigten Hypothese “Jede im Sinne des Pi-Theorems
minimale Beschreibung ist eine Bewertung” stellt das überraschende Ergebnis dar, daß
grundsätzlich die Lösung des Bewertungsproblems auf die Lösung des Beschreibungsprob-
lems zurückgeführt werden kann. Damit ist das Bewertungsproblem genau dann gelöst,
wenn eine vollständige Beschreibung vorliegt.

Das Anwendungspotential dieses Lösungsansatzes wird anhand von Beispielen aus dem
Ingenieurbereich ausführlich dargestellt und diskutiert. Dabei ergibt sich aus der mathema-
tischen Notation zusätzlich eine interessante Querverbindung zur Technik des “qualitative
reasoning” in der Künstlichen Intelligenz.
 13

1 Introduction

Problems frequently arising during the realisation of complex development and design pro-
cesses have, at an early stage, stimulated mankind to concern himself with the process
of design and to analyse the proceedings upon which the designer’s outlining actions are
based [65]. As this piece of work is concerned with evaluation, a partial aspect of the design
process, fig. 1.1 should serve as an introduction. It illustrates the final result of a preliminary
design sketch [18].

Figure 1.1: “Americana”, New York, NY, 1957 [18]

It is without a doubt that the solution of such a design process not only requires the in-
volvement of present knowledge [92] but also all the creative and cognitive abilities of the
designer. This knowledge about approach and problem solving strategies for achieving new
goals is called “meta-knowledge” [92].
14 CHAPTER 1. INTRODUCTION

The first drafts of the design of the hotel “Americana” are still retained and shown in fig. 1.2.
These drafts may serve as a means of studying certain steps in the approach of the designer.
This shows that the conceptual goals to be achieved are also of great importance, as well
as the importance of the aspects knowledge and meta-knowledge, already revealed.

Figure 1.2: Design Sketch “Americana”[18]

For the designer, it is a fastidious task to distinguish between the relevant influential variables
and processes which effect the design, to perceive the mutual dependencies and degrees
of freedom in the design variables and to make profitable use of them in order to achieve the
conceptual goals. The more abundant the imaginable canon of shapes, the more difficult it is.
This is because most designs are not only distinct through an infinite variety of parameters
but also through different principles of formation, called topologies.

In order to grasp the backgrounds of these facts, the intellectual and temporal course of the
design process will be roughly outlined and described.
1.1. OUTLINE, EVALUATION AND DESIGN 15

1.1 Outline, Evaluation and Design

If, at the end of the design process, the actual complete object exists, then, at the beginning,
the designer only has an idea at his disposal. In the course of the design process, the
intellectual conception of the designer is evolved at the same time as the formation of the
object and its design parameters x1 ; x2 in time. This is illustrated in fig. 1.3.

t 6

....................................................
....................
.............
............
.........
.......
......
.....
.....
....
...
..
..
......
.............................................................................................................................................. ..
....................
... ................. ..
.. ........ ......... ..
.... .
.......... ..........
.. .... . .....
.
... ....... .... .
.... .........
. ..
.... ........ ........
....... ........ ....
..........
................ .......... ...
....................... ............. ..
............................................... .................. ..
..........................................................................................
..

r
..
.....
.......
.....
.......
.......
........... ..........
...........

* x2
.............
...................
.......................................... . ......... .
... ...
...............
.............................................................


 - x1
Figure 1.3: Design Process

A loop-shaped iterative procedure is typical for the design process. This is due to the ex-
istence of different kinds of boundary conditions and due to the often implicit coupling of
demands which must be fulfilled simultaneously in different partial domains of a complex
technical system. The need to observe the impact of decisions made in a partial domain
upon all other partial domains leads to the formation of several small loops in the course of
design [58], see fig. 1.4. Characteristic for each loop is the necessity to evaluate situations

Idea Quality Cost :::::: Environment Aesthetics

....
.....
......
........ ..........................
....................
.....

.........................................
................................. .......

.......................................................................................................................
..........................
..........................
...............
......
 ....
. .. ... ....
......
........................... .... .. .. .. ......
.
................. . . ..............
-
... .. .. . ..
.......
. ...................................
...
... .
.........
........... .......... . .......
.
.. .. .. ..........
...
........................................................................
.
.. . ..
............................................... .. .. ........
...
...
..... .................................................................................................................................................
..... ................................. .....................
...... ................. ............
............................... ....
.
.........
....... ..........................
. ................
........
.

-
.... .......................
....... ....................................................................................................................................................................................
...
..
...
....
.....
......
......... .......................
..................
................ Models ..................................................
.........................
.............
.....

........
..
...
........
....... .........................
.................... + ..........
. .....
. .............
. .
. .. .
...........................
............
....
....
......
...
....
.....
Methods
....................................
......................................................................................................................................
.......................
...... ................... .............
........ ....................... .....
...
........
....... ......................... ...........
...
-
.
........
. .....................
..... .
....... .. .... ......................................................
........................................................................................................................
....
..
...
....
.....
......
........ .........................
...................
.......................................
.....

.................................. 
................................................................................................................................
.......
............ ................
........................
.............
......
.....
....
.....
....... ......................... ...
.
...
. ... ..
...
. ..........
...
? ....
-
. ..... .... ..................
........ ....................
...... .
....... .. ..... . ...... ..... .. .. . .
....................................................................................................................................... .............................. ..
.
...

Object

Figure 1.4: Loops in the Design

and decisions with regard to the given problem and to estimate possible consequences as
well as risks and couplings with other partial domains. This becomes even more difficult
because, depending on the partial domain in question, the kind of knowledge available, the
16 CHAPTER 1. INTRODUCTION

existing formalisms and methods and the precision of the available data vary considerably.
The selection of possible problem solving strategies thus varies as shown in fig. 1.5.

deterministic
................
..........
r
...........................................
..........

r r continual
........
....... .......
...... ......
.........
. .....
.....
.
.....
.....
?
..
....
intuitive .
...
......
.
.. ....
....
....
....
..
. ....

.
....
.
....
.
@
R
@ ?
?
...
...
...
...
..
. ...
..
. ..
..
. ..
...
..
. ..
... ..

r r symbolic
.. ..
. ..
..
. ..
... ..
..

numeric
...
..
..
..
-  .
.
.
.
..
.. ..
.
.. .
.. ...
.. ..
..
..
..
..
..
..
Kinds of Knowlegde ..
..
..
.. ..
... .
.. ..
... ...

r?? r explicit
...
... ...
...
...
....
....
....
@
I
@ ....
....
.
...
..

.... ....
.... ....
discrete .....
6 .....

r
..... ...
.
..... ..
...... .....
...... .....
........ ......
.......... ........
.............. ..........
..............................................

stochastic
.. .. ..
.. simpli- . . ..
exact .
. .
. . intuition
theories . . fied
. . heuristics
. . . trial & error
.. theories. . ..
 -
. . Precision of Knowledge .
. . .
. . Use of Knowledge .
formulae experience values .
optimisation norms .
prototypes
 Computer -

 Person -

Figure 1.5: Complexity of Knowledge

The way in which decisions are made, depending on evaluations in the separate disciplines
by means of available models and methods and according to which systematics the ag-
gregation of the single evaluations to a complete evaluation occurs, usually leading to the
necessity to form compromises, is therefore of fundamental significance in the design pro-
cess. In the second chapter, this will be more explicitly dealt with whilst analysing some
classic models.

As well as this, it will become clear in the course of this chapter that the evaluation and the
description of design objects take on a key role in the design process and in the development
of computer-aided design systems [4, 22, 74]. In order to clarify the context of such future
computer-aided design systems, the basic intellectual approach and application potential of
computer-aided design and artificial intelligence techniques will be briefly portrayed.
1.2. COMPUTER-AIDED DESIGN (CAD) AND ARTIFICIAL INTELLIGENCE (AI) 17

1.2 Computer-Aided Design (CAD) and Artificial

Intelligence (AI)

The application of computers in the fields of development and design has a tumultuous
history. At the beginning of computer-aided drawing, the main focus was upon facility in
editing and reusability of elements in drawings. Today, the desire for a homogeneous and
consistent product model [57] within the company is moved to the center of developmental
activity by means of presently available relational databases. This data model does not
only include known geometric data, it also includes all product related data such as material
properties, manufacturing times and cost calculations.

Based upon the idea of a homogeneous data model, associative systems [53] are being
developed. In such a system, all software applications (mesh generation, finite element cal-
culations, simulations of injection molding, etc.), derived from the parametrised stored data
model, are automatically updated when some of the design variables are modified. Thus a
ideas
HH
' j
H
- solutions $

?
analysis synthesis

& variants  %



objects

Figure 1.6: Design Sequence

sequential comparison of a design with its predecessor is more feasible and therefore the
formation of design variants, shown in fig. 1.6, is more effectively supported. To further fa-
cilitate this problem, further means have been developed in engineering sciences to support
the designer, by means of construction catalogues [75], knowledge-based systems [17] or
system models [110] for example, in the creation and selection of design alternatives.

Fundamental to the realisability of such software systems is the far-reaching usage of known
knowledge and the retaining of known analytic methods of calculation. These, when cor-
rectly formulated and converted according to the computer, advantageously make use of the
extreme processing speed of simple mathematic base operations.
18 CHAPTER 1. INTRODUCTION

In AI however, the objective of research is mainly the plane of meta-knowledge. In this

plane, the attempt is made to recognise the strategies of the designer and to map them
into a software system. In this field of research, therefore, the statement “the equivalence
of searching and knowledge”, shown in fig. 1.7 is often used and in partial areas success-
fully tested, although the principle workability of a reasoning kernel, indispensable to such
a software system with the imitation of the cognitive abilities of humans is hotly debated
between representatives of “strong” and “weak” AI [69]. In fig. 1.7, a dynamic search space

u
initial variant
HHH




 ? H

H
j
H
HHu variants
 H
 ? HH

j
H u
HHvariants

 H
e goal variant ? HH
j
H

Figure 1.7: Expanded Search Space

is generated, in which alternative solutions are produced. This generation presumes a initial
variant in a first expansion stage. The alternative solutions are evaluated, with regard to
their present state and that of the possible improvement potential. Afterwards, the variants
are organised and the highest valued variant further expanded and refined.

This process can be formulated as an algorithm [106] and be turned into the pseudocode
of the following form in fig. 1.2. During the search procedure, problems of every type occur

procedure Design
begin
determine initial variant and conceptual goal
while variant not equal to goal
begin
generate variants (?)
evaluate variants (? ?)
estimate improvement potential (? ? ?)
rank and sort variants
refine highest valued variant first
end
print result
end

Figure 1.8: Procedure Design

which are marked with one or more stars (?) according to their increasing difficulty. There-
fore, a description is necessary in order to represent the variants, in which the object can be
1.2. COMPUTER-AIDED DESIGN (CAD) AND ARTIFICIAL INTELLIGENCE (AI) 19

described in a sufficiently exact way, in the context of a world model. Furthermore, in order
to produce variants by means of systematic application of allowed operations, a number of
possible and permitted operators have to be declared. Therefore, heuristics as well as a
limitation of the depth of the search tree are applied, in order to limit the complexity of the
algorithm, which usually grows exponentially and also to limit the large number of variants
to be evaluated.

Taking into account the existence of multiple, often diverse and competing goals, the evalu-
ation of variants therefore leads directly to the evaluation problem. Because the solution of
this problem plays a key role in the control and dominance of the combinatorial explosion of
the search space, a qualitative advance in the field of software-supported design systems
seems possible if this problem can be relatively satisfactorily solved in the future.

As well as this, for the evaluation of the developmental potential of the variants, the know-
ledge of a measurement is also required, with which the difference between the present state
reached and the conceptional goals can be expressed. In this area, the intuitive and cog-
nitive capabilities of the designer still play a dominant role and it is difficult to make definite
statements about expected future advances in this area.

Because the classic evaluation methods and models [52, 109] used in literature to evaluate
variants vary according to the type of knowledge available and the existing formalisms, they
are portrayed and analysed, along with their typical characteristics, in the following chapter.
20 CHAPTER 1. INTRODUCTION
 21

2 State of Knowledge

In this chapter, a comprehensive portrayal of the research situation will not be considered.
Instead a classifying portrayal of the relevant known models and those documented in the
literature will be attempted, the analysis of which manifests three classic core questions,
which in turn fundamentally motivate the assignment setting of this piece of work to a great
extent.

2.1 Decision Theories

The portrayal of the decision finding’s relevant theories can be organised into four differing
principle fields of thought [52, 109]. These are:

 Deductive decision theory. The features of deductive decision theory are

() formal science)

{ Axioms. A complete set of axioms, which are obeyed by the decision maker.
{ Model. Construction of contradiction-free decision models with regard to axiom
systems.
{ Logic. Logic conclusions within the axiom-supported model explain the behaviour
of the decision maker.

 Descriptive decision theory. The features of descriptive decision theory are

() real science)

{ Goals. Goal criteria of the decision maker are ascertained and determined.
{ Model. The values, persuasions and modes of behaviour are observed and mod-
eled.
{ Algorithm. The evaluation result is calculated and the best one chosen.
 Statistical decision theory. The features of statistical decision theory are
() statistics)

{ Coincidence. Processes are liable to coincidental occurrences which are de-

scribed by probability distributions.
{ Model. The behaviour is modeled by statistical distribution.
{ Stochastics. Behaviour is determined by statistically optimal considerations.
22 CHAPTER 2. STATE OF KNOWLEDGE

 Game theory of strong AI. Features of strong AI are

() information science)

{ Goal. Algorithmisation of intelligence

{ Model. Equivalence of searching and knowledge, use of search strategies.
{ Procedure. Usage of knowledge bases and inference engines.

Various aspects are often singled out of these four views and to support the decision, mixed
with one another during the formation of an evaluation or decision making model. At this
point, it is often the case that the decision must be made whilst considering multiple goals.
The decision theory [26, 27, 105, 107] therefore assumes that the decision maker’s order of
preference can be represented using a so-called utility function.

2.1.1 Utility Function

With the utility function, the degree of goal accomplishment zj of a design object, with regard
to the j -th goal criteria, can be given as a real number. Due to a supporting causality
relation [50], this degree of goal accomplishment is a function of the object and therefore of
the design parameter quantities X = fx1 ; x2 ; : : : ; xn g with

zj = zj (x1 ; x2 ; : : : ; xn) = zj (X ) j = 1; : : : ; m (2.1)

The determination of such a utility function is difficult and is described as an evaluation
problem within the framework of this work. With “evaluation”, the mapping of an object onto
an goal criteria or a partial quality is understood.

2.1.2 Multiple Goals

The entire quality of the design object is made up of the partial evaluations found. The
formation and the determination of the entire utility function U of a design object from the
single utility function zj found for the m partial goals with

U = U (z1 (X ); z2(X ); : : : ; zm (X )) (2.2)

is therefore described as an aggregation problem within the framework of this piece of work.
Thus, aggregation is understood as being the mapping of the single partial qualities onto the
entire quality of the object.

During the construction of the entire utility function, the mutual preferential indepen-
dence [23, 109] of the goals plays an important role because only in this case is a purely
2.1. DECISION THEORIES 23

additive construction of the overall aggregation function U from the weighting function uj
justified with
Xm
U = uj (zj (X )) (2.3)
j =1
An unavailable preferential independence of the partial goals can however be obtained by
newly redefining the partial goals, for instance by summarising the preferentially dependent
goals into one single goal. The arranging of the goal quantities is therefore also liable to
many non-rational influences [109], so that the formation of such a hierarchical graph show-
ing goal quantities can be described as a structuring problem in the framework of this piece
of work.

Through the structuring problem, questions beyond this about the completeness, about the
existence of a minimal form of the evaluation graph etc. of classical evaluation models are
posed, which are of a philosophical as well as a philological nature and therefore cannot
be answered within the framework of this representation. However, these questions will be
answered later in detail for the new evaluation methodology.

Due to the complexity of the entire questioning process, mostly simple models with a con-
stant weighting are therefore used. For the general additive form of the aggregation function
according to equation 2.3,
Xm
Uk = wj zj (Xk ) (2.4)
j =1
applies for the case of the constant weights wj . This is also described as a weighted sum-
mation model. In this, the weighting wj of the j -th partial goal is assumed constant. The
index k stands for the k-th alternative considered. As well as this, multiplicative aggregation
functions of the form
Ym z (X ) !wj
j k
Rkl = (2.5)
j =1 zj (Xl )
which are also be described as a weighted product model, have been investigated [98]. In
this, the weighting wj of the j -th partial goal is assumed constant. The indices k and l
serve to differentiate between two alternatives k and l. Furthermore a rank Rkl with Rkl > 1
represents the preference of the alternative k over the alternative l.

Through the common aggregation of the partial uses of both the models to an overall ag-
gregation function U or to a rank sequence R, these aggregation functions belong to the
compensatory models. In these, a bad judgement in one area can be compensated by a
better judgement in other areas.

This means that the reliability and measuring quality of the data used within partial eval-
uations as well as in the weighting of the goal quantities is of particular importance. The
24 CHAPTER 2. STATE OF KNOWLEDGE

representation of the different scaling features possible is therefore the object of the next
section.
2.1. DECISION THEORIES 25

2.1.3 Scales

While quantifying the properties of the design object, the various data have to be acquired
and processed. Here, the objects or processes occurring are mapped in the evaluation
process onto the evaluation graph or converted into a subjective utility value with the help
from the preference function.

The features of the single attributes or the objectively occurring processes or objects can
therefore be quantified into three differing scaling qualities:

 Nominal scale. The features of a nominal scale are:

{ unique correspondence object ?! measuring value

{ purely qualitative, besides correspondence no allowed operations to do with mea-
suring values
{ example: account number

 Ordinal scale. The features of an ordinal scale are:

{ unique correspondence object ?! measuring value

{ besides purely qualitative approaches about >; =; < relations no allowed opera-
tions to do with measuring values
{ example: finish 1st, 2nd, 3rd

 Cardinal scale. Unique correspondence object ?! real number. The subdivision oc-
curs here in three further subscales:

{ interval scales. The features of an interval scale are:

- intervals of the unit of measurement are constant
- no or only arbitrary zero point
- besides >; =; < relations, such as +; ? arithmetic and formation of mean
value, no allowed operations to do with measuring value
- example: Celsius scale ( C ).
{ ratio scale. The features of a ratio scale are:
- intervals of the unit of measurement are constant
- existence of an absolute zero point
- all operations concerning real numbers are allowed for measuring values
- example: Kelvin scale ( K ).
{ absolute scale: The features of an absolute scale are:
26 CHAPTER 2. STATE OF KNOWLEDGE

- dimensionlessness
- existence of an absolute zero point
- all operations concerning real numbers are allowed for measuring values
- example: probabilities

In order to ascertain the features of the separate attributes and the weightings, the following
two measuring processes are available:

 direct measuring methods. Features of direct measuring methods are:

{ correspondence: stimuli ?! number.

{ result: data in cardinal scale

 indirect measuring methods. Features of indirect measuring methods are:

{ correspondence: stimuli ?! interval.

{ result: data in ordinal to max. interval scale

From experimental psychological studies [109], the problem is known, that people can only
estimate reliably and without distortion using ordinal scales. However, ordinal and inter-
val scales cannot be transformed onto ratio scales. Therefore, absolute differences in the
complete evaluation of two alternatives can only be used very limited in comparative inter-
pretations of the alternatives.

2.1.4 Metrification Postulates

The theory of physical measurement [35, 70], on which ratio scales are based, builds upon
three axioms, which must be completed by three metric conventions in order to clearly define
measurement scales. The axiomatic arrangement of the theory of measurement is, as a brief
overview, summarised according to the description in [35], as follows.

First requirement. In order to describe objects, a set S of objects and a set of features M
with:

S = fA; B; C; : : :g (2.6)
M = [ Mk with Mk \ Mj = ; (2.7)
2.1. DECISION THEORIES 27

is given based on a certain set of indices k; j " J . In addition, for all sets of features Mk a
unique mapping  of S onto M through
 : S ?! M with S 6= ;; M 6= ; 8k"J (2.8)
is defined. Furthermore, for each mapping k "  onto the feature Mk , a measurement
regulation exists such that for each element A, B out of S , the decision whether k (A) =
k (B ) or k (A) 6= k (B ) applies can be made through observation. The description B of an
element A " S through the mapping k onto the feature types Mk through the definition
B(A) = f k (A) j k " J g (2.9)
is then complete, when every A " S through B(A) is uniquely determined and therefore
B(A) = B(B ) () A = B (2.10)
generally applies.

Second requirement. Every feature type Mk is arranged linearly. This ensures the exis-
tence of a binary relation with the qualities of asymmetry and transitivity. This is defined as
follows:

 asymmetry. For each two elements 1 and 2 exactly one of the three cases applies:
1 <  2 or
1 =  2 or (2.11)
1 >  2
 transivity. For the three elements 1 , 2 and 3 with
1 <  2 and
2 <  3 follows
1 <  3 (2.12)

Third requirement. Each feature type Mk is mappable onto an interval IR in a strictly

monotonously way. Through this, such a mapping
F0 : M ?! IR with IR " IR (2.13)
of the sets of features M onto the interval IR , declares that
F0(1) < F0 (2) in case  1 < 2 (2.14)
With this, the axiomatic foundations of the scaling are laid, in which all pairs of features
(i; j ) are each applied with a pair of real numbers (F0(i); F0(j )). With the three following
metric conventions, this makes possible the eventual construction and determining of the
unit scales.
28 CHAPTER 2. STATE OF KNOWLEDGE

First metric convention. Through the real numbers’ difference formation by means of the
mapping

x = F0 (i) ? F0 (j ) (2.15)

the formation of subclasses Ux " U can occur. From the strict monotony of F0 , the result is
that, with

Ux = f (i; j ) j x = F0(i) ? F0(j ); i; j "Mg (2.16)

exactly, Ux = Uy applies if x = y . Therefore, the order of features of M are obtained in the
order of real numbers, by means of the mapping F0 .

Second metric convention. In the type of feature Mk , a feature 0 is thus distinguished

so that the unit of measurement 0 is applied to it (zero feature or element)

F (0) = aF0 (0) + b = 0 (2.17)

Third metric convention. In the type of feature Mk , a feature 1 is thus distinguished, so

that the unit of measurement 1 is applied to it (unity feature or element)

F (1) = aF0 (1) + b = 1 (2.18)

By determining 0 ; 1 and solving the equation systems 2.17 and 2.18 according to

a = 1=fF0 (1) ? F0(0)g > 0 (2.19)

b = ?F0 (0)=fF0(1) ? F0 (0)g (2.20)
the measurement scale F () is uniquely determined.

2.2 Model Classes

The model classes, which are based upon the differing theoretical approaches taken, are
briefly represented here. For such decision making following several goals simultaneously,
models are traditionally divided into three different model classes [52, 109] in the following
three paragraphs.
2.2. MODEL CLASSES 29

........................................................................ .
.
. .
. .
.
.
. 1
........................ .......................................
.
.
.
.
.
.
.
Entire goal
.
..
..
..
.
....
..
..
.
........................................... ... . .
........ ............................
. ....................
... ..
. .
. . ..
..
..
. . ... .... .......................................................
..... .
. .
. .
.. ............ ........................ ...................... .
.
. ............. ............................
....... ..
............................ ............. .
. ........... ............................
............... .
......................................................................... .
.. ... .
........................................................................ .
........................................................................ . ................................................................................ ........................................................................................
.
.. .
. .
.. .
. .
. .
. .
. .
. .
. .
.
.. . . .
..
.. 2 .
.
..
..........................................................................
.
..
.. 3 .
.
..
.........................................................................
.
.
.
. 4
.......................................................................
.
.
.
.
.
.
.
.
.
.
. 5
.......................................................................
.
.
.
.
.
.
.
.
.
. 6
.......................................................................
.
.
.
.
.
.
.
Partial goals

Figure 2.1: One-Step Criteria Hierarchy

2.2.1 Multiple Attribute Decision Making (MADM)

If a MADM model is used to evaluate, the desired goal (here ‘1’ in fig 2.1) and the corre-
sponding decomposition of the entire goal into the single partial goals (here e.g. as ‘2’ to ‘6’)
has to be determined. Thus the one-step evaluation graph can be represented in fig. 2.1.
By further sub-dividing the single partial goals, a multiple stepped evaluation hierarchy can
occur. Afterwards, the single utility functions of the partial goals of all k alternatives, corre-
sponding to equation 2.4,

zj = zj (Xk ) j = 2; : : : ; 6 (2.21)
as well as the constant weightings wj of the partial goals for the k alternatives are to be
determined with
X
6
Uk = wj zj (Xk ) (2.22)
j =2
whilst applying a simple additive aggregation function, according to equation 2.4. With re-
gards to the goal hierarchy, the alternative with the highest evaluation results is the best
alternative among all alternatives, according to
X
6
U = max
k
wj zj (Xk ) (2.23)
j =2
It is often the case that the units of measurement of the single judgements zj can be deter-
mined through measurement, calculation, estimation or specialist judgement and which are
then, by means of a weighting factor wj defined by group consensus, aggregated to an entire
evaluation U . Thus the following characteristics for typical substitutes of MADM models can
be summarised as follows.

 Formation of a criteria hierarchy

 Determination of utility and aggregation functions

 Finite number of alternatives to be evaluated

 Result: Selection of one or more best alternatives. Classic area of application: deci-
sions regarding investment of goods.
30 CHAPTER 2. STATE OF KNOWLEDGE

In section 2.3.1, the Analytic-Hierarchic Process is analysed in detail due to its compatibil-
ity in representing the MADM model. The application upon an example is illustrated and
discussed in detail.

As well as this, there are also one-step hierarchical approaches which are extremely similar
to the MADM models but which explicitly support the modeling of imprecision due to the
differing form of their utility function. Instead, only a weak order relation is generated upon
the (partial) set of alternatives instead of an absolute evaluation measurement.

The visualisation of this order relation with the mutual dominance relationships of the al-
ternatives can occur in a graph as is illustrated in fig. 2.16 of the example. Here it can be
observed that the determined order relations do not always have to be transitive. These
kinds of evaluation models are therefore described as Outranking Relations [39, 40, 76, 77].
These approaches possess the following characteristics:

 Preference relations formation with concordance and discordance numbers

 Explicit modeling of imprecision

 Result: (Half) order concerning a set

In section 2.3.2, the application of the decision technological approach upon an example is
represented and discussed in more detail.

2.2.2 Multiple Objective Decision Making (MODM)

In order to apply MODM models, the global goal function U (X ) must be explicitly known. In
these models, not already existing alternatives are not evaluated. In contrary, it is the best
possible solution X which is calculated by means of variations of the initial solution X . This
is done through the formulation of an optimisation problem [25, 67] of an goal function U with
!
U (X ) = Max (2.24)
and also through the possible existence of constraints [62, 94] of both the forms
g(X )  0 (2.25)
X  X  X (2.26)
in which a starting solution of X is assumed. The processes used to solve optimisation
problems differ here in the requirements they place upon the goal function U , from which,
depending upon the algorithm [91] used, the first derivative U 0 or the second derivative U 00
must also be known. As well as this, a further precondition is that the solution space X is
continuous. Summarised, the MODM models possess mostly the following characteristics:
2.2. MODEL CLASSES 31

 The existence of an explicit goal function and possibly their derivatives.

 Finite number of constraints and a continuous solution space.

 Result: Calculation of one or more optimal alternatives. Classic area of application:

Optimisation of technical processes.

In contrast to MADM models, the stricter prerequisites concerning goal functions and so-
lution space limit to a great extent the establishment of MODM models and inhibit their
application in areas of imprecise knowledge from the very beginning. Therefore the MODM
approach is not pursued further in this work.

2.2.3 Expert Systems (XPS)

The application of XPS models requires firstly a knowledge acquisition of the entire available
or accessible specialist knowledge within a certain field. The facts and rules determined in
this area have to then, in order to be incorporated into a data base, be adapted and arranged
in such a way that they can be processed by the inference mechanism of the XPS. This
necessary preliminary work limits the establishment of XPS to certain areas of knowledge,
in which such preliminary work is possible (frame problem). Furthermore a conclusion drawn
from an XPS can only refer to elements of knowledge found in the XPS, so that a further
limitation exists here. Typical XPS models possess mostly the following characteristics:

 Modeling of facts and rules.

 Processing through the inference mechanism.

 Result: Derivation of statements. Classic area of application: Technical systems diag-

nosis.

Thus the XPS approach is not suitable for areas of weakly structured knowledge and there-
fore is not pursued further in the framework of this piece of work.

The DSS model (decision support systems) available as a derivative of the XPS model is
also irrelevant in this context because it is only used for complex evaluations from data
banks, according to the present day state of technology. They are therefore only tools used
to support decision making and not decision making models in their own rights, therefore
they are meaningless in the framework of this study.
32 CHAPTER 2. STATE OF KNOWLEDGE

2.3 Evaluation Models

In this section, two comprehensive application examples for evaluation models also suitable
for areas of imprecise knowledge are represented. These are the analytic-hierarchical pro-
cess and the outranking relations approach. Both the application examples serve on the one
hand to illustrate the given model equations and procedures and, by follow-up questioning
about the assumptions, lead, on the other hand directly to the following summarising analy-
sis of the evaluation problem which motivates the assignment setting in the next chapter.

2.3.1 Analytic-Hierarchical Process (AHP)

The Analytic-Hierarchical Process (AHP) [83] is applied to the supporting of complex evalu-
ation and decision processes. In a first step, the complete goal to be evaluated is split into
its partial goals. Due to the need to name all fundamental influence quantities, the differ-
ent criteria of a decision become explicit. At the same time, the process of goal finding is
also influenced by the linguistical, sociological and group-dynamic aspects. Therefore no
definite guiding principles exist which guarantee the attainment of mutual independence of
preference or the completeness of the criteria hierarchy.

The procedure can be represented by the example of the job search of an american graduate
from Philadelphia [37]. His complete goal hierarchy of accepting the job in the preferred best
city, splits into the partial criteria; the distance from his home town, Philadelphia, cost of
living, climate, offer of education and quality of life. The partial goal, offer of education is
split again into high schools and colleges, the partial goal, quality of life into traffic situation
and cultural and leisure possibilities. The result is the criteria hierarchy in fig. 2.2. For a
........................................................................................... .
.
. .
.
.
. .
.
.
. .
.
.
.
.
City .
.
.
.
.
. .
.... .. ...................................................................
............................................................. .
..... .. . . . . .. ...
.
. .
. ............. .................................................... .
.
.
.......................................
. ... .................. ................................................
. ... .
. .
...
..
..
.................................. ...... ... ............. .
.
. ............. ..........................
.
...
..
...
..
..
..
... .
..... ....................... .
. ...... ... ........... .
.
.
............
............
.........................
............................
............... ...... . . .....
......................................................................................................... ................................................................................................. . ............................................................................................ . .................................................................................................. .........................................................................................................
.
.. .
. .. .
. .
. .
. .
. .
. .
. .
.
.. .. . . .
.. ..
. .. .
.
. .
.
.
.
. .
.
.
.
. .
.
.
.
.
..
..
..
Distance ..
.
.
..
.
.. Cost .
.
.
.
.
.
.
.
Climate .
.
.
.
.
.
.
.
Education .
.
.
.
Quality of Life
.
.
.
.
.
.
.
.
.
. . .
. .
. .
. .
. .
. .
. .
......................................................................................... ......................................................................................... .......................................................................................... .................................................................................................................... .........................................................................................................
.. ........ . . . . . . . . . . . ..
. . . .. . . . . . . . . . . . . .. .. ......................................................................... ...... .... . . . . .. ................... ...........
.. ....... . .. .. . . . . .. .. .
.. ...................................... ................................... ... . .... .
. ..... .................... .... .......
....... ... ... .......
.................................. ................... ...................
.................................. ................... .................... ......
.............................................................................................................. ...................................................................................................... .................................................................................................... ...............................................................................................
.. . . . . . . .
.. ... .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.. .
. .
. .
. . .
. .
. .
.
.
... High School .
...
.
.
.
College .
.
.
.
.
.
.
. Traffic .
.
.
.
.
.
. Culture .
.
.
.
.
. .
. .
. .
. .
. .
. .
. .
.......................................................................................... ......................................................................................... .......................................................................................... ...........................................................................................

Figure 2.2: Hierarchical Criteria Structure

better overview of the formula used, the terms used in this hierarchy are represented in the
following by the numbers 1 to 10. The following illustration is therefore the basis of the given
example.

In a second step, the aggregation of the partial criteria has to be determined. This occurs
2.3. EVALUATION MODELS 33

....................................................................... .
.
. .
.
. .
.
.
.
. 1
....................................................................
.
.
.
.
. .... ...
. . . . . ... ..... .
... .
. .
. ................. ............................................. . . .......... ....
...
. ..
...........................
.
.. .................. .........................................
..
...
. ......... .................................. ...................... .
.
. ............. ............................
..... ... .
........................ .... ........... . ............ ..........................
.. ..... .....
............................................. ........................... . ..... ..
..................................................................... .
....................................................................... . ............................................................................... ........................................................................................
.. . .. . .
. . .
. . .
. .
.. . .. . . . . . . .
.. ... .. .
. . .
. . .
. . .
.
.. 2
........................................................................ .
. .. 3 .
.
.
.
.
.
. 4 .
.
.
.
.
.
.
5 .
.
.
.
.
. 6
........................................................................
.
.
.
....................................................................... ......................................................................
. . .......................................................................
. ..
. ... .. .
. . . ... .
.
..... ......... .......
.... ............................................................... .. ........ ..... . ............................
................. ........ .. .. . . .. .. ................................................................ .. ..... ... ... ..... ................. ... . . . .....
...
..
.................. ...................................... .... . . . . . . . . . . .. . ................ . . .. . . . . . .. . .. . .
.................. ... ........
................................................. .................... ....................................................................... . ...................................................................... . ........................................................................
.. . .
. . .
. . .
. .
.. .
. .
. .
. .
. .
. .
. .
.
.. . . . .
.. 7 .
.
.
.
.
. 8
......................................................................
.
. .
.
. 9
......................................................................
.
. .
.
. 10
.......................................................................
.
.
........................................................................ . .
. . .
. . .

Figure 2.3: Hierarchical Criteria Structure

by the decision maker estimating and interpreting the priority of the partial goals with regard
to the complete goal.

In a third step, those alternatives on offer are evaluated with regard to those criteria which
are leaves of the criteria tree. In this example they are the 4 cities, Boston, Los Angeles,
St. Louis and Houston. The same evaluation process is used for this as is used for the
evaluation of the partial goals.

The evaluation consists of the alternatives on offer being evaluated in pairs against one
another with regard to the criteria [42, 43]. The result of the comparison in pairs of the k-
th partial goal are summarised in matrix A (k) . Due to the substitutability of the evaluation
(k)
sequence, the elements aij are reciprocal to one another. Thus
8 (k)
< i 6= j; 1  i; j  n
a(ijk) = : 1=aji (2.27) for
1 i = j; 1  k  m for

applies. If the evaluations of matrix A (k) with the dimension n is also transitive, then the
evaluations are consistent and generally

a(ijk) = a(ihk) a(hjk) 1hn (2.28)

applies. This means that the elements a(ijk) of the matrix A (k) consist of pure quotients of the
form
(k)
a(ijk) = zi(k) (2.29)
zj
Looked at from another angle, this means that in the case of a consistent matrix, equa-
tion 2.30, by means of a comparison with equation 2.28, is determined by a single line of the
matrix A (k) .
X n
zi(k) = n1 a(ijk)zj(k) (2.30)
j =1

By comparing equation 2.30 with equation 2.31, a further result is that in this case, the
evaluation zi corresponds the components of the eigenvector xi to the largest eigenvalue
max of the matrix A (k) (see also [44, 45, 89]).
Xn
x(ik) =  1 a(ijk)x(jk) (2.31)
max j =1
34 CHAPTER 2. STATE OF KNOWLEDGE

If all alternatives i and j have been compared in pairs in regard to the criterion k in the
AHP, the eigenvector is calculated with reference to the largest eigenvalue of the matrix A (k)
(k)
which is formed from the elements aij . The eigenvector is used as an evaluation result. The
normed eigenvector x(k) then represents the relative rank sequence r (j ) of the alternatives
with regard to the k-th partial goal.

A (k) x(k) = x(k) (Eigenvector) (2.32)

r(k) = Px (k)
(k)
(Leaves) 8 fk j k = 2; 3; 4; 7; 8; 9; 10g (2.33)
jxj j
j
The calculation of the normed eigenvectors to the highest eigenvalue corresponds, in the
case of consistent inputs, to exactly that percentage of elements of a line in the evaluation
matrix A (k) and represents a best possible mean value [83], in the general case of non-
consistent inputs. A consistency measure [83] is also available here for the estimation of
the incompatibility of the single judgements. This consistency measure is based upon the
fact that max  n always applies for the highest eigenvalue of a real, positive matrix with
mutually reciprocal element [83]. For a matrix of the dimension n, a consistency index c:i: is
therefore defined to

c:i: = max ?n (2.34)

n?1
Because this consistency index refers to at least reciprocal and therefore correlated eval-
uations, a relationship to a random index r:i:, coming from probability number matrices is
formed [99, 100, 101]. The calculation of the probability number matrices is comprehensive

Dimension n 1 2 3 4 5 6 7 8 9 10
r:i: 0.00 0.00 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49

Figure 2.4: Random Inconsistency Index r:i:

but, in the framework of this representation, only the knowledge of the numerical values
given in the table in fig. 2.4 of the highest eigenvalues of probability number matrices is
required. Thus a consistency ratio

c:r: = c:i:
r:i: (2.35)
can be defined where a value of c:r: < 0:1 is considered as acceptable. For higher values of
c:r:, the decision maker should work through his evaluations once more in order to reduce
inconsistencies. Now the calculation of the evaluation result can ensue. The matrices
(k )
A are formed by multiplying the rank sequences r(k) of the partial criteria according to
equation 2.33 with the criteria weights. The multiplication of this matrix with the respective
eigenvalue of the criteria x(k) , results in the relative rank sequence of the alternatives, with
regard to the k-th criterion. This process is carried out for all inner nodes from the bottom to
2.3. EVALUATION MODELS 35

the top. In this way, the rank sequence r (1) of all alternatives can be determined with regard
to the underlying criteria hierarchy `10 ? `100 shown in fig. 2.2.

r(1) = A (1) x(1) (2.36)

A (1) = [r(2) ; r(3) ; r(4) ; r(5); r(6) ] (2.37)
with
A (5) = [r(7) ; r(8) ] (2.38)
r(5) = A (5) x(5) (2.39)
A (6) = [r(9) ; r(10) ] (2.40)
r(6) = A (6) x(6) (2.41)
In the practical application of AHP on an evaluation problem, precisely three cases can occur
which correspond to the existence of qualitatively varying scales in section 2.1.3. These
three cases are illustrated here using the example, the evaluation of the partial goal distance
of the four city alternatives, Boston, Los Angeles, St. Louis and Houston from Philadelphia.

 Case A: The evaluation occurs by means of linguistic terms which are correspondingly
transposed into the scale of 1 ? 9 in fig. 2.5. This scale is justified by psychological
experience and tests [83] because people can only reliably estimate within an ordi-

Scale Meaning
1 Equally important or preferred
3 Weakly more important or preferred
5 Moderately more important or preferred
7 Strongly more important or preferred
9 Absolutely more important or preferred
2,4,6,8 Graduations
Reverse values Dominance of the second criterion over the first

Figure 2.5: Linguistic Evaluation Scale

nal scale level and also because this scale only permits relative comparisons. The
definition of the consistency ratio c:r: is therefore important because the evaluations’
freedom of opposition is comprehended.

 Case B: The evaluation occurs by means of estimating the data. ’From Philadelphia
to Los Angeles it is about nine times as far as from Philadelphia to Boston.’ ) BOS
LA =

1 () LA = 9 . Because a further distance is evaluated as unfavourable, the reverse

9 BOS 1
values of the distance are used to form the matrix in fig. 2.6.
36 CHAPTER 2. STATE OF KNOWLEDGE

Distance max 4.0164

Bos LA St.L Hou c:r: 0.0061
Bos 1.0000 9.0000 3.0000 4.0000 x 0.5869
LA 0.1111 1.0000 0.3333 0.5000 0.0670
St.L 0.3333 3.0000 1.0000 2.0000 0.2171
Hou 0.2500 2.0000 0.5000 1.0000 0.1290

Figure 2.6: Distance Evaluation (estimated)

 Case C: The evaluation occurs by means of exact measuring data. The exact dis-
tances are used, which can be taken from street maps for example ) BOS LA =
2706 [miles] = 9; 14189; see fig. 2.7. Here both the eigenvectors x of the estimated dis-
296 [miles]

Distance max 4.0000

Bos LA St.L Hou c:r: 0.0000
Bos 1.0000 9.1419 2.9324 5.0946 x 0.6073
LA 0.1094 1.0000 0.3208 0.5573 0.0664
St.L 0.3410 3.1175 1.0000 1.7373 0.2071
Hou 0.1963 1.7944 0.5756 1.0000 0.1192

Figure 2.7: Distance Evaluation (measured)

tance in fig. 2.6 differs only slightly from the exact distance in fig. 2.7. The linguistic
scale can therefore render also very precise estimations possible.

With AHP, a tool is therefore available with which the decision finding can be represented
and made transparent. The mathematical means used in this are summarised in fig. 2.8.
Further qualitative and quantitative statements of the model, for instance the interpretation

Human experiences Mathematical property

Evaluation in pairs of n criteria
in an evaluation matrix Matrix A with aij " IR+
Reversibility of comparisons in pairs aij = 1=aij
a1 3 times more important than a2 a =3, a = 1
1 2
a2 a 31

In a decision hierarchy Non-commutative operations

division into upper and lower goals A B 6= B A
Registration of inconsistencies Error norm: c:r:(max )
since max  dim(A )
Reliability of the end result Error propagation
Sensitivity analysis Total differential

Figure 2.8: Mathematical Equivalences in AHP

2.3. EVALUATION MODELS 37

of meanings of the relative distances within the rank sequence, can unfortunately not be
achieved by the calculation of the rank sequence r (1) , due to the ordinal quality of the scales
used.

Example. For the job search example [37], the following evaluation matrices, eigenvectors
and consistency ratios shown in appendix B.1 result for the single leaves from figure B.1
to B.7. Fundamental to this evaluation judgement for the partial goals distance and living ex-
penses are exact data material from maps and statistical year books. The other evaluations
including the criteria tree and the aggregation weights reflect, however, the subjective view
of the decision maker [37].

For the weighting of the partial goals amongst one another, the decision maker [37] stated
the matrices, computed the eigenvectors and consistency ratios, which are shown in ap-
pendix B.1 in fig. B.8 to B.10, for the single inner nodes. Thus, the evaluation of alternatives
and the weighting of the criteria amongst one another are identified.

Therefore the final calculation of the end result can ensue, as is shown in fig. 2.9.

City (1) 0.1570 ( Weighting

0.1067
0.3021
0.0883
(2) (3) (4) (5) (6) 0.3459 + Rank/Alternative
0.6073 0.3184 0.2591 0.4863 0.2459 0.3356 1. Boston
0.0664 0.1003 0.5367 0.1845 0.3732 0.3287 2. Los Angeles
0.2071 0.4432 0.1316 0.2411 0.2307 0.2206 3. Saint Louis
0.1192 0.1382 0.0725 0.0882 0.1503 0.1151 4. Houston
0.0000 0.0000 0.0388 0.0055 0.0202 0.0192 ( Consistency
 MB
 B
 B
Education (5) 0.4000 Culture (6) 0.3000
(7) (8) 0.6000 (9) (10) 0.7000
0.4082 0.5383 0.4863 0.0634 0.3241 0.2459
0.0816 0.2531 0.1845 0.0590 0.5078 0.3732
0.4082 0.1297 0.2411 0.5298 0.1025 0.2307
0.1020 0.0790 0.0882 0.3479 0.0656 0.1503
0.0000 0.0091 0.0055 0.0150 0.0224 0.0202

Figure 2.9: Evaluation Result City

The end result of the evaluation from fig. 2.9 can also be graphically illustrated for the upper
38 CHAPTER 2. STATE OF KNOWLEDGE

hierarchy level. The visualisation of the evaluation result is shown in fig. 2.10. Here, Boston
appears in first place due to the above average evaluation in various partial criteria. It is
followed by Los Angeles which occupies the most strongly weighted criteria climate and
quality of life but which otherwise shows a below average evaluation.

Cost of Living (3) Boston

..
.
Los Angeles


Saint Louis



..
....
..
. ..

Houston




..
..
....
..
...
Climate (4)
.....
...... ........
...... ..
.
...
........
..... ... .. .
.....
...... . ... ...
............... . . .
. ..
.............. .. ............. .
... ..... . . ......... ..... ............
... ..... ......
.. ...... ....... .... ......... ... .... ...
. ..... ...... .
.... . . . .. ......
. ..... ....... .... ..
.. .. ......
... ..... . . ..... ..... ... . ..... .. ....
... ...... .. . . ..
... ..... ........................ .. ..
.
..
.. ......
... ...... . ..... .. ......
... . ..... . . .....
. .
. ......
. . ..
... ........... ......... . . ...
. ..... ... .......... . . ......
. ... ......... ..
. .. .
.
.. . . . ......
... ..
.... ............ .... ....... .... ..
. ....
. ..
... .. ........ .. .. ... .. ......
. . . .
. ......... ... .... .... ..
.. ...... .
. . . .....
...................................................................................................
.
..
.
.
.. ..
. ..
... .
..
.
.. ......................................
. .
........... .... ...0.1
.
.......... .. .. .
.
.
.......................................................
.
0.2
. .. .. . .
0.3 0.4 0.5
......
.
.
.......0.6 Distance (2)
. .... ..
. ..... . . . . . . .. ....
... ....... .........
. ..... ....... ..... ... . . ......
. . ... ... .. ... . .....
.. .. ........ ...... .. . .......
.. ........ ...... ...... . ... ........
.. ................................ .. .. ... .... .
.. ....... ..... . .. ........
. .
... ..... . ........... ....... ...... . . .......
.. .
..... ..... .......... ........ ...........
.. ....... .... . ..
............................
.. ..... .................... ............. .. .
........ ...................... ... .. .........
.
............................... . ...
.... . ..... ......
..... .... ....
......... .......
..
.... .
..... .......
Offer of Education (5) ..
..
..
..
.
.......
..
..
..
..
.
.......
..
..

Quality of Life (6)

Figure 2.10: Visualisation of Evaluation Result City

Sensitivity analysis. The behaviour of the model can be tested by means of a study of
parameters. By using the chain rule upon equation 2.36 in order to derive the rank sequence
for a general hierarchy of the evaluation criteria,

@r(1) = @r(1) @r(i)


@r(k)

@x(j) @r(i) @r(k) @r(j)
= (x(1) (i) (k)
i )(xk )


(xj ) (2.42)
and therefore
(x(1) (i) (k)
i )(xk )


(xj )
x

r(1) = (j ) (2.43)
is valid, so that the influence of small changes of the partial evaluations can be studied. The
(k)
notation xj means the j -th component of the k-th eigenvectors.

Discussion. The basics of the analytical hierarchical process were first illustrated in detail
in [83] and since then, there have been published numerous studies, such as application
examples and user experiences (e.g. [37, 82, 87]). A controversial discussion [5, 6, 20, 21,
84, 85, 86, 88] was aroused by the fact that a simple extension of the amount of alternatives
from n ? 1 to n can, in certain cases, lead to a change in the original rank sequence of the first
2.3. EVALUATION MODELS 39

n ? 1 alternatives. The discussion is based upon a stability equation which will be derived
for a one-step hierarchy according to fig. 2.1 as follows. If all elements in the evaluation
matrices A (k) of the n ? 1 alternatives are consistent, the addition of the matrix element
a(1kn) is sufficient when increasing the number of alternatives by 1 to n due to equation 2.28,
because all the remaining elements of the n-th line and column of a consistent matrix B (k)
are automatically determined. Thus
8
< for 1  k  m
A (k) = (a(ijk)) : for 1  i; j  n ? 1 (2.44)
8 (k)
< for 1  k  m
B (k) = (bij(k)) = : a(ijk) (2.45)
ain for 1  i  n

holds. In the case of consistent matrices B (k) , the largest eigenvalue max of the matrix is
exactly equal to the dimension n of the matrix and, after transforming equation 2.31

B (k) x(k) = nx(k) (2.46)

applies for the calculation of the eigenvalue max . As the eigenvector is normed to a length
of unity using equation 2.33 for the single components of the normed eigenvector
8
a(k) a(1kn) (1=a(1kj)) < for 1  j  n
rj(k) = Pn jn(k) = Pn a(k) =a(k) : for 1  k  m (2.47)
a 1n
h=1 hn h=1 1h

therefore applies. If the weighting factor of the highest criteria according to equation 2.36 is
described as x(1) , for the determination of the end result
Xm
(1)
rj = rj(k)x(1)
k (2.48)
k=1
holds. If the original n ? 1 alternatives are available without limitation of their generality,
sorted according to their sequence,

r1(1)  r2(1)  : : :  rn(1)?1 or (2.49)

rj(1)  rj(1)+1 for 1j n?2 (2.50)
applies. In order to maintain the rank sequence in equation 2.50, by applying a further
alternative, the following stability equation can be given by combining equation 2.47 with 2.48
and 2.50
80
m < 1
1 ,X 9
X @ ? 1 A a(k)x(1) n aln =  0
(k )
for 1  j  n ? 1 (2.51)
k=1
: a(1kj) a(1k;j)+1 1n k (k) ;
h=1 alh
If this is disregarded, the insertion of a further alternative leads to a change of rank within the
otherwise unchanged n?1 alternatives. As the norming of the eigenvectors and therefore the
vector norm used influences the result, a lively discussion has been aroused [20, 21, 46, 88].
40 CHAPTER 2. STATE OF KNOWLEDGE

2.3.2 Outranking Relations

In outranking relations models [40, 76, 77], the modeling of imprecision is achieved by stating
explicit decision thresholds. In order to do this for every partial goal, the individual and
freely choosable veto thresholds vj , indifference thresholds pj and preference thresholds
qj , illustrated in fig. 2.11, are used to form the partial preference relations cj (Xk ; Xl ) and
dj (Xk ; Xl).
........ ........... ......
...
.....
....
dj (Xk ; Xl ) 1 6 cj (Xk ; Xl ) ....
....
...................................................................................................
....
.
.....
...... .
... ....
....
...... ....
... ....
...... .
....
.
..
... ....
..... ....
.... ....
....
....
......
-
?vj ?pj
..

pj qj
... ..

zj (Xk ) ? zj (Xl )
........... .......... ........... .............................................................................................

0
Veto Indifference thresholds Preference

Figure 2.11: Partial Preference Relations

Here, zj (Xk ) ? zj (Xl ) signifies the difference in the j -th fulfillment of the goal of two alter-
natives Xk and Xl . The corresponding mathematic formulation of the partial preference
relations yield according to fig. 2.11
8
>
> 0 zj (Xk ) ? zj (Xl ) < pj
< zj (Xk )?zj (Xl )?pj
for
cj (Xk ; Xl) = > qj ?pj for pj  zj (Xk ) ? zj (Xl )  qj (2.52)
>
: 1 for zj (Xk ) ? zj (Xl ) > qj
8
>
> 0 zj (Xk ) ? zj (Xl ) > ?pj
< ?zj (Xk )+zj (Xl )?pj
for
dj (Xk ; Xl) = > vj ?pj for ?vj  zj (Xk ) ? zj (Xl )  ?pj (2.53)
>
: 1 for zj (Xk ) ? zj (Xl ) < ?vj
These partial preference relations of the j partial goals are, by means of a weighted summa-
tion, summarised into a global concordance number c(Xk ; Xl ) and into a global discordance
number d(Xk ; Xl ) through the following relationship of the j partial goals as
X
c(Xk ; Xl ) = w1 wj cj (Xk ; Xl) (2.54)
j
X
d(Xk ; Xl ) = w1 wj dj (Xk ; Xl ) with (2.55)
j
X
w = wj (2.56)
j
Through the global concordance number c, all aspects evaluated according to their relative
significance towards the entire goal, are grasped according to their relationship to the global
prevalence relation, such that sufficient reasons exist stating that alternative Xk might not
be less favourably evaluated as alternative Xl . Accordingly, by means of the global discon-
cordance number d, all aspects evaluated according to their relative significance towards
2.3. EVALUATION MODELS 41

the entire goal, are grasped according to their relationship to the global prevalence relation,
such that sufficient reasons exist stating that alternative Xk might not be more favourably
evaluated than alternative Xl .

Now, all global concordance and discordance numbers c and d are used in a comparison
with predefined concordance and disconcordance numbers c and d, described as level of
requirement, to form the global preference relation S with

Xk SXl () c(Xk ; Xl )  c ^ d(Xk ; Xl)  d (2.57)

Therefore the hypothesis, Xk is not less favourably evaluated than Xl , is assumed, if the
global concordance number c reaches a given given level of concordance c and if the global
discordance number d does not overstep a given level of discordance d [40]. The visuali-
sation of such an order relation with the mutual dominance relations of the alternatives is
illustrated by means of graphs in fig. 2.16 of the application example.

Example. In order to illustrate the process in outranking relations models, the following
order process of seven car alternatives, A to G is illustrated in detail [39, 40]. Here, the
decision maker sets up a one-step criteria hierarchy which applies the partial goals, price,
comfort and speed. The weighting wj of the partial goal with regard to the entire evaluation,
car, can then be defined in a second step. The veto thresholds vj , the indifference thresholds
pj and preference thresholds qj , which are individual for each partial goal, have then to be
defined in order to form the three partial preference relations, as is illustrated in fig. 2.12.

Parameters
Criteria q p v w
Price 1.500 2.500 -10.000 0.400
Comfort 0.000 0.000 -2.500 0.300
Speed 20.000 20.000 -60.000 0.300

Figure 2.12: Preference Threshold and Weighting

Together with the judgements and the technical data of the alternatives A to G, the necessary
input data of the following table in fig. 2.13 is yielded according to the outranking relations
approach.

The negative sign in the line price means here that a higher price is evaluated less favourably.
According to the table in fig. 2.13 together with equations 2.52 and 2.53 the figures B.11
to B.16 are yielded as well as the partial concordance and discordance numbers of the
partial goals. For illustrative purposes, these are represented in appendix B.2.

Thus, the global concordance and discordance can be determined with the criteria weights
w in the table in fig. 2.12, by means of equation 2.54 to 2.56. Building upon the concordance
42 CHAPTER 2. STATE OF KNOWLEDGE

Alternatives
Description A B C D E F G
Price -30.00 -26.00 -25.00 -21.00 -20.00 -19.00 -12.00
Comfort 5.00 5.00 4.00 3.00 2.00 1.00 2.00
Speed 210.00 150.00 200.00 190.00 140.00 200.00 160.00

Figure 2.13: Data of Alternatives

and disconcordance numbers, the global preference relation S of the decision maker is
completed by comparing the requirement levels c and d according to equation 2.57. For

Global Concordance
A B C D E F G
A 0.000 0.300 0.300 0.600 0.600 0.300 0.600
B 0.400 0.000 0.300 0.300 0.300 0.300 0.300
C 0.400 0.300 0.000 0.300 0.600 0.300 0.600
D 0.400 0.700 0.400 0.000 0.600 0.300 0.600
E 0.400 0.400 0.400 0.000 0.000 0.300 0.000
F 0.400 0.700 0.400 0.200 0.300 0.000 0.300
G 0.400 0.400 0.400 0.400 0.700 0.700 0.000

Figure 2.14: Global Concordance Numbers Car

Global Discordance
A B C D E F G
A 0.000 0.200 0.333 0.867 1.000 1.000 1.000
B 1.000 0.000 0.750 0.500 0.467 0.750 1.000
C 0.400 0.400 0.000 0.200 0.333 0.467 1.000
D 0.800 0.800 0.400 0.000 0.000 0.000 0.867
E 1.000 1.000 1.000 0.750 0.000 1.000 0.733
F 1.000 1.000 1.000 0.800 0.400 0.000 0.600
G 1.000 1.000 0.800 0.400 0.000 0.500 0.000

Figure 2.15: Global Discordance Numbers Car

number values c  0:6 and d  1, fig. 2.16 results directly, in which the global preference
relation S of the decision maker is illustrated. At the same time it should be observed that the
underlying theory does not always guarantee the determined order relations to be transitive.

If the graph relation in fig. 2.16 is reorganised by means of topological sorting, fig. 2.17 is
yielded. It thus becomes apparent that only alternatives A and C are not dominated by
any others. By comparison with the technical data of the car in fig. 2.13, it becomes ap-
parent that, through an extreme veto threshold (veto threshold is identical with indifference
2.3. EVALUATION MODELS 43

C .
..
..
...
... B
... ....................
.
.. ..........................
.. ..............
... ..............
..
. . ...... ..
. ...............
........
... .................
...............
..............
.............. ...
.............. ..
D ..................................... ..
.. .... ........................
.. ..... . ..................
.. ..... ....
.. ..... .. .........................
.. ..... .. .............
.. ..... ... ..............
..............
. ..... ...
... ... .............
... ..........
.
............
..............
.. ... ......... ..............
.............
.. .
. ..... .............
.. . . ..... ............
.. ... ..... .............
... ... .....
...... A
... . .. .....
.. .. .....
. ......
... ..
. .....
. .. .....
.......... ....... .....
.....
.... .... .....
...... ......... .....
...... ..... .....
........ .....
......... .....
............................. .....
.....
............................
E .............
.............
.....
.....
.............. .....
............. .....
............. .....
............. .....
.............. ...........
.............
............. .....................
............. .....
..................
......
.. ... . . . .............
......... G
..............
. ........
. ....................
................
....................
F

c  0:6;
Figure 2.16: Global Preference Relation S ( d  1)
3 B  3 F
   
- D Q -G -3 E
A
QQ 
 7
QQ 
QQ 


 C

c  0:6;
Figure 2.17: Topologically Sorted Preference Relation S ( d  1)

threshold here), the higher price, compared to the other car, can be compensated by the
similarly heavy weighting of the above-average criteria comfort and speed. A direct prefer-
ence statement between A and C cannot be attained without varying the requirement levels
c and d.
44 CHAPTER 2. STATE OF KNOWLEDGE

2.4 Analysis

The advantages and disadvantages of the classic decision models and processes repre-
sented in the literature, are briefly summarised once again in fig. 2.18 for clarification.

MADM Advantages: + Simplicity (7! Selection)

+ Criteria and hierarchy plausible
+ Simple linear model, transparency
+ Stability, sensitivity analysis
Disadvantages: ? Single stage, no process
? A priori knowledge necessity
? Algorithmic influences (Norm)
? Aggregation of scales, compensation
Outranking Advantages: + Simplicity (7! Ranking)
Relations + Treatment of imprecision
+ Treatment of incomplete approaches
+ Knockout criteria
Disadvantages: ? Lack of hierarchy
? Aggregation of scales, compensation
? Level of concordance difficult to define
? Level of discordance difficult to define
MODM Advantages: + Process (7! Calculation)
+ Interaction
+ Determination of all functional efficient solutions
+ Inclusion of a posteriori knowledge
Disadvantages: ? Goal function necessary
? Parameters must be design variables
? Size of solution set
? Algorithmic convergence
XPS Advantages: + Knowledge of facts, derivation (7! Control)
+ Formulation of rule knowledge
+ Automatic use of symbolic knowledge
+ Explanation possible
Disadvantages: ? Monotone versus non-monotone conclusion
? Necessary acquisition of knowledge
? Necessary structuring of knowledge
? Unexplained, whether generation of ideas possible

Figure 2.18: Advantages and Disadvantages of the Classic Decision Models

The features of this classic model are compared here with the problem aspect of the real
evaluation situations occurring in the design process.
2.4. ANALYSIS 45

Here, the following problem areas in the evaluation problem are illustrated:

 Criteria catalogue. Problem aspects contained within are:

{ Completeness and minimality of the criteria

{ Organisation into a hierarchical arrangement
{ Determination of aggregation function

 Criteria. Problem aspects contained within are:

{ Occurrence of design and non-design quantities

{ Selectivity, correlation and coupling of the criteria
{ Measurability, scale and dimension of data

 Scales. Problem aspects contained within are:

{ Physical quantities with and without dimension or zero point

{ Abstract quantities and engineering terms
{ Existence of subjective and objective quantities of equal rank

 Data. Problem aspects contained within are:

{ Exact measurability and calculability

{ Calculation models and experiences in engineering
{ Separation from subjectivity or objectivity during collection of data

 Solution space. Problem aspects contained within are:

{ Discrete as well as continuous form of solutions in form of differing topologies and

parameter variants
{ Unknown nature of solution space, as illustrated in fig. 2.19 as clouds of dots
representing design solutions.
xj
pp pp rppp p pp pp
6

pp pp rpp pp pp p Topology A
Topologyp Bp p
pp pp r p p p p p - xi
 Topology C
xk
+

Figure 2.19: Clouds of Dots for Topologies

46 CHAPTER 2. STATE OF KNOWLEDGE
 47

3 Assignment Setting

The piece of work under consideration was thematically embedded into the research group
“Construction in Engineering - Paths to an Holistic Approach” within the framework of the
partial project 5.3, called Design Methodology. The holistic outline and evaluation pursued
within the research group of construction in engineering using the example of bridges [28,
29], resulted therefore in the context of theoretical questioning and investigation which is
demonstrated, represented and performed in this piece of work.

The demands of the research group upon a model which is able to represent the design
process and to evaluate the different design variants can be located in the context of the
abilities of classical decision making models as shown in fig. 3.1.

MADM
uJ
J
J
Outranking-
Relations
r J
J
J
} J
J
J
Evaluation J
Problem J
u J
J u
MODM XPS

Figure 3.1: The Evaluation Problem in the Context of Classic Models

The successful evaluation of (partial) decisions in the design process requires an evaluation
model to be able to incorporate discrete knowledge (e.g. norms) such as in expert sys-
tems (X PS), precise knowledge (e.g. functional relationships) such as in Multiple Objective
Decision Making Models (M ODM), as well as imprecise knowledge (e.g. from the field of
socio-economics) such as in Multiple Attribute Decision Making Models (M ADM).

In this way, the complex evaluation situations which occur in real life exhibit all of the differing
aspects of these three classes of classical evaluation models and in this respect, represent
a challenge to the further scientific investigation and development of more and more sophis-
ticated evaluation models.
48 CHAPTER 3. ASSIGNMENT SETTING

3.1 Evaluation Problems

If one accepts the principle mode of procedure which is to dissect an entire evaluation into
partial evaluations or is alternately the aggregation of partial evaluations into an entire eval-
uation, formed models of evaluation can only then do justice to their central meaning within
the decision related finding, when at least partially sufficient answers to the following three
core questions can be given:

 How are the structure and completeness of the criteria tested ?

 How are the single partial evaluations obtained ?
 How are partial evaluations aggregated to an entire evaluation ?

To portray these questions in their complete context is the objective of the following sections.

3.1.1 Structure

Fundamental to the methodology to be presented is a strict systematic separation of the

evaluation process into a description and an evaluation hierarchy [58] which prevents a mix-
ing of both concepts. Whereas the description graph contains all the necessary information
concerning the design object and its constructive variants, all in differing levels and grades
of abstraction, the evaluation graph represents the structure and linkage of the evaluation
criteria as concerns the formation of the entire evaluation. The mapping of the right hand
description graph onto the left hand evaluation graph occurs by means of transfer functions,
see fig. 3.2.
entire
object
PPP
entire
   PP
evaluation
H  PP
 HH partial partial partial

transfer
H functions
 H object 1 object 2 object 3
partial partial  Q
 Q
 Q
evaluation Y evaluation Z  Q
partial partial
object 2a object 2a

Figure 3.2: Systematic-methodological Terminology Separation

In all cases, the need for clarification consists, above all, in the evidence of structure, com-
pleteness and independence of the evaluation criteria and the evaluation graph, as well as in
the primarily functional geometric oriented form of structuring and representing the descrip-
tion graph, which both do not always have to be only tree formations as simply represented
in fig. 3.2.
3.1. EVALUATION PROBLEMS 49

3.1.2 Evaluation

The evaluation is concerned, above all, with the clarification of the prerequisites of the trans-
fer function ' which maps the description graph onto the evaluation graph. In compliance
with an objectifiable evaluation result, the evaluation  should be determined only by the
object attributes X , therefore
'
  X
has to be valid. The interpretation that the object attributes X are a necessary and sufficient
condition of the resulting evaluation , is supported as an objectifiable causal relation by the
generally accepted interpretation [13, 50] of the definition of causality.

3.1.3 Aggregation

Partial evaluations can be condensed into an entire evaluation by means of aggregation.

Correspondingly, the mathematically abstracted representation shown in figure 3.3 to the
representation shown in figure 3.2 results in consistency conditions [58], which follow from
'0
  X
'3 6 6
'1
'j
(1 ; : : : ; m )  (x1; : : : ; xn)

Figure 3.3: Mathematic-schematic Abstraction

the possible, differing mapping sequences ('3  'j versus '0  '1 ) and therefore, despite
different calculatory paths, have to result in compatible evaluations.
50 CHAPTER 3. ASSIGNMENT SETTING
 51

4 Solution Approach

4.1 Motivation

As an introduction to the solution approach and to present the principle arguments used, the
thought out mathematical-physical context of the description compared with the evaluation
should be made clear first. In this respect, in a short overview, an attempt will be made
to represent in summary the fundamental principle ideas in theory formation in physics —
described here as the homogeneity1 hypothesis, the model hypothesis and the measurability
hypothesis — to enlarge upon the contribution made later by the evaluation hypothesis.

4.1.1 Homogeneity Hypothesis

Under the term epistemological formation, the differentiation and classification which is be-
ing established of varying objects and processes encountered in reality will be understood
according to their natural state of being. Thus, for instance, the characteristics “mass” and
“length” of such an object are principally experiencable and comprehendable in different
ways and therefore are also not directly comparable with one another. The comparison of

 Observation of

 Reality
Homogeneity Hypothesis
) theoretical of sets
HH
rrrrrrrrrr rrrrrrr rr
H -
r rr r r xClasses
1 ; : : : ; xk

Figure 4.1: Formation of Sets of Features

two elements is thus only then viable when both elements belong to the same group as re-
gards their comparison. The theory of dimensional homogeneity of physical functions [35]
is founded upon this epistemological principle of the construction of homogeneous sets of
features which is represented in fig. 4.1.

1 Formation of terminology used according to [35] is described in [68] as the coherence principle also.
52 CHAPTER 4. SOLUTION APPROACH

4.1.2 Model Hypothesis

By means of an inductive or deductive approach, it has often been attempted to develop

a model idea for objects or processes occurring in reality. This means that all relevant
parameters of influence must be identified and determined. This in turn means that for

Classes
 x1 ; : : : ; xk


Model Hypothesis
) qualitative
HH 
 

H - Object Properties
f (x1; :::; xn)



Figure 4.2: Production of Relevance Lists

a certain object feature f , represented in fig. 4.2, the model formation occurs by means of
drawing up a relevance list [35] of x1 ; x2 ; : : : ; xn of all n quantities of influence xj . This means
presuming a certain functional relationship f between these quantities.

4.1.3 Measurability Hypothesis

Building upon the principles of the homogeneity hypothesis, the mapping of a set of features
onto the domain IR of the real numbers is explained by the metric postulate [35]. With

Object Properties
 f (x1 ; :::; xn)


Measurability Hypothesis
) quantitative
HH 
 

H - Object Description
z
y
6
h
f (x1; :::; xn) = 0
>
-x


t
b

Figure 4.3: Quantitative Description

this principle, the grasp of the object feature represented in fig. 4.3 can also occur in a
quantitative object description by means of the functional relationship f . The recommended
implied notation f (x1 ; : : : ; xn ) = 0 used there in the place of xn = f (x1 ; ::; xn?1 ) will later make
possible a more elegant notation of the proofs of both the product and of the Pi theorem.
4.1. MOTIVATION 53

4.1.4 Evaluation Hypothesis

An evaluation represents in the broadest sense “qualitative” and “quantitative” information

about an object or a process (i.e. what it is and how well it is evaluated). If the form of repre-
sentation taken for the evaluation is still redundant, then this redundancy can be eliminated
without any loss of information. Each evaluation therefore has the feature which enables it

Object Description



f (x1 ; :::; xn) = 0
Evaluation Hypothesis Q
Q
) minimal Pi-Theorem
HH
H -
F (1; :::; m )Q
s
Q
=0
Object Evaluation
Figure 4.4: Evaluation

to possess a redundancy-free representation form and in this sense has to be minimal. This
means that the number of independent parameters cannot be further reduced. Therefore
it makes sense to formulate the evaluation hypothesis corresponding to fig. 4.4 as follows:
“Any minimal description in the sense of the Pi theorem is an evaluation”. The guaranteed
representation of the form F (1 ; ::; m ) = 0 which, through the Pi theorem2 to each existing
description f (x1 ; ::; xn ) = 0, is minimal in the required sense and represents an evaluation.

Complete Illustration. The entire context of the assumptions from the homogeneity hy-
pothesis used, up to the evaluation hypothesis, is shown in fig. 4.5.

For dimensionally homogeneous function relations and their dimensionless parameters, the
following mathematical properties evident in the appendices A.1 to A.4, and which are
founded on the product and on the Pi theorem, are valid. The interpretation of these proper-
ties, whose aim it is to solve the evaluation problem, is the purpose of the following sections.

4.1.5 Authorisation

For the benefit of the evaluation hypothesis, an answer should be briefly given to the question
about its authorisation. As to its establishment, the principles named in section 4.4 are drawn
upon, i.e. causality, consistency and invariancy etc., which make possible the partly provable
and the partly epistemological checks of the various conclusions drawn from the hypothesis.
These general statements form, together with the results of the various examples of usage,
the evaluation hypothesis’ basis of evidence.
2 Representation and deduction of the Pi theorem is the objective of the section 4.2.2, 4.2.3 and A.2
54 CHAPTER 4. SOLUTION APPROACH


Observation of

 Reality
Homogeneity Hypothesis
) theoretical of sets
rr r r
rrrrrrrrrrrrrrrrr r
HH
H -
Classes

x1 ; : : : ; xk


Model Hypothesis
) qualitative
HH 
 

H - Properties
f (x1 ; : : : ; xn)
 

 
Measurability Hypothesis
) quantitative  
HH  
H - Description
z
6
y .
.....
h
f (x1 ; : : : ; xn) = 0
 >

-x
....
.....


 b
t

Evaluation Hypothesis
) minimal
HH Evaluation
H -
F (1 ; : : : ; m) = 0

Figure 4.5: Evaluation and Description

The fundamental requirements for the applicability of the Pi theorem are the construction of
a system using differing classes of features, the qualitative and quantitative conjoinability to
statements f of the form f (x1 ; : : : ; xn ) = 0, as well as the invariancy of the corresponding
statements F under transformations x0 = ax of the reference scales of these classes. Due
to the unit system of physic’s fulfilling the said requirements, explicit reference is constantly
being made to this physical unit system and also the given application examples are derived
from the area of physics. As well as this, the derived statements are generally valid also for
all other feature systems which fulfill the given assumptions.

4.2 Central Proofs

Formulated solely as a theorem here, the results of both complete derived function theo-
retical arguments of the product and Pi theorem [15] are given and can be found in ap-
pendices A.1 and A.2. For this reason, the modern pure algebraic representation of the Pi
theorem, with the help of the notation about the term dimensional matrix [24, 68], is devel-
oped in the section to follow.
4.2. CENTRAL PROOFS 55

4.2.1 Product Theorem

Theorem. If the invariance of ratios of physical quantities under scaling of the units of mea-
surement is assumed, the relation between primary physical quantities x1 ; x2 ; : : : ; xn " IR+
and secondary physical quantities  = f (x1 ; : : : ; xn) only exists through the functional rela-
tionships of the following differentiable form:

f = Cx1 1 x2 2


xn n (4.1)
Whereby for fixed n > 1 is n " IN and the C; 1 ; 2 ; : : : ; n " IR are real constants.

4.2.2 Pi Theorem

Theorem. From the existence of a complete and dimensionally homogeneous functional re-
lationship f of exactly n physical quantities xi " IR+ , the existence of a functional relationship
F of only m < n dimensionless parameters j (also called similarity variables or dimension-
less products) results with

f (x1 ; :::; xn) = 0

F (1; :::; m ) = 0 (4.2)
whereby m = n ? r is reduced by the rank r of a dimensional matrix formed by means of the
n quantities. The dimensionless products take the form
Yr
j = xj x?i ji (4.3)
i=1
whereby for fixed j is j = 1; : : : ; m " IN and the j 1 ; j 2; : : : ; jr " IR are real constants.

4.2.3 Pi Theorem (algebraic)

For the Pi theorem and the consequences derived from it, the possibility of a somewhat
more evident approach is offered through the concept of the dimensional matrix [24, 68].

Concept of proof. In the dimensional matrix, the exponents of the unit representation
of physical quantities form a set of homogeneous linear equations due to the dimensional
homogeneity of physical formulae. The characteristics concerned with detatchability, inter-
changeability of bases etc., proven for sets of homogeneous linear equations are thus able
to be utilized in order to deduce a relation of exactly m dimensionless products, with the help
of the product theorem.
56 CHAPTER 4. SOLUTION APPROACH

Deduction. Firstly, a dimensional matrix, represented on the left in fig. 4.6 is formed from
the xi which are contained in the relation f (x1 ; : : : ; xn ) = 0. This dimensional matrix is in
general not square and contains n lines to receive the x1 to xn and also up to 7 columns
which represent the physical basic quantities of the unit system3 in use. The elements of the
matrix aij are, in the representation of xi in fig. 4.6, left, the exponents in the j -th dimension.
Then the dimensional matrix is brought into diagonal form by means of rank-preserving
m1 m2 : : : : : m7 m1 m2 : : : : : mr
x1 66 x1 1
x2 x2 1
: :: ::
: r : :
: : :
: : :
: aij n : :
:
: xr
: ? 1
: 6 x1
: :
: m :: ji
:
xn ?? m
x

Figure 4.6: Definition of the Dimensional Matrix

transformations. Following this, all the zero columns are erased, see fig. 4.6, right. This
means that the set of physical quantities Xn = fx1 ; : : : ; xn g in the physical relation

f (Xn) = 0 n2 (4.4)

can be separated into the two sets Xi = fx1 ; : : : ; xr g of the primary and thus linearly inde-
pendent base quantities and in the set Xj = fx1 ; : : : ; xm g of secondary and thus dependent
physical quantities in the relation

m=n?r
f (Xi; Xj ) = 0 i = 1; : : : ; r (4.5)
j = 1; : : : ; m
It is to note that the indexing of Xj is changed from r + 1; : : : ; n to 1; : : : ; m, see fig. 4.6, right.
Each of the two sets Xi and Xj contains at least one element here, because the set Xn with
n  2 guarantees that neither the set Xi nor the set Xj can be an entirely empty set, due to
the dimensional homogeneity given by equation 4.4.

If the units mi of all primary quantities xi " Xi are scaled corresponding to the scaling prop-
erties of physical units according to equation A.2 from xi to mi xi , all secondary quantities
3 The SI unit system has at present 7 known basic quantities (units): length (m), mass (kg), time (s), current
intensity (A), temperature (K), amount of material (mol) and light intensity (cd).
4.3. CONCLUSIONS 57

xj " Xj are scaled with it corresponding to their exponential representation, due to the prod-
uct theorem. For each physical quantity, this leads to a transformation of the form
Yr i = 1; : : : ; r
f (mixi ; xj mi ji ) = 0 (4.6)
i=1 j = 1; : : : ; m
If one selects the scaling factors in such a way that mi xi = 1 applies, conversion with mi =
1=xi leads eventually to
Yr
F (1; xj x?i ji ) = 0 or (4.7)
i=1
F (1; : : : ; m ) = 0 with m = n ? r (4.8)
whereby the dimensionless variables j are explained in the form
Yr
j = xj x?i ji j = 1; : : : ; m (4.9)
i=1
with m = n ? r.

4.3 Conclusions

Through the evaluation hypothesis, the embedding of the evaluation problem in the world of
physical model concepts occur. In this way, from the hypothesis, “any minimal description in
the sense of the Pi theorem is an evaluation”, the statement is produced that the evaluation
problem can be basically traced back to a description problem. With this, a principally formal
solution for the evaluation problem, which is coupled to the solution of the corresponding
description problem f (x1 ; : : : ; xn ) = 0, is available.

In the following sections, those characteristics will be more closely elucidated, which occur
by applying the Pi theorem to the evaluation problem.

4.3.1 Evaluation

The evaluation problem can be traced back to a problem of description. The m dimen-
sionless products also called dimensionless parameters, are derived from the description
f (x1; : : : ; xn) = 0 and represent the units of measurement of the method of evaluation. They
take the form
Yr
j = xj x?i ji
j = 1; : : : ; m (4.10)
i=1
Thereby, it has also been proved that the quantity of all equally evaluated objects or pro-
cesses always corresponds exactly to the quantity of physically similar objects or processes.
This represents the sought-after solution of the evaluation problem.
58 CHAPTER 4. SOLUTION APPROACH

4.3.2 Minimality

The dimensionless parameters j form a fundamental system in a linear algebraic sense

(see Appendix A.3 or [35]) and therefore have features inherent to all bases of a linear
vector space. The minimal and necessary number of linearly independent bases of a linear
vector space is therefore a feature of the dimension of the space which is covered by means
of the bases system.

F ( |1; : {z
: : ; m} ) = 0 (4.11)
System
The equivalence and resulting exchangeability of all possible base systems of a space with
one another can be interpreted as corresponding to the endeavours to neutralise coupling
in natural language terminology formation.

4.3.3 Completeness

Linear combinations of base vectors of a linear vector space can cover the entire vec-
tor space. They maintain in this sense the property of completeness. Therefore it can
be guaranteed that from a set of dimensionless parameters, taken from a description
f (x1; : : : ; xn) = 0, F (1; : : : ; m ) = 0 represents a minimal and complete evaluation.
Up to now, in engineering, the dimensionless parameters are interpreted most generally as
ratios of characteristic forces, energies, times or lengths such as is known with the Reynolds’
number

Re = ul
 = ul
  inertia
friction
forces (4.12)
The possible physical interpretations of the dimensionless parameters show therefore the
meaning contained in the evaluation. This represents the sought-after solution of the struc-
ture problem.

4.3.4 Granularity

If the modeling is further refined, further quantities xn+i are added to the description
f (x1; : : : ; xn) = 0. At the same time, two different cases [35] need to be decided upon
with regard to fig. 4.7. If the rank r of the dimensional matrix, extended by additional xn+i
remains constant, the earlier m remain unchanged and are only added to by new m+i . If
the order r increases and if there are at least two exponents available in the newly added
order column, the earlier dimensional parameters m remain once again unchanged and are
4.3. CONCLUSIONS 59

m1 m2 : : : : : m7 m1 m2 : : : : : mr mr+1
x1 66 x1 1
x2 x2 1
: : :
:
: r :: :
:
: : :
: n : :
: aij : :
: xr
: ? 1
: 6 x1
: :
: m :: ji
:
xn ??
xm
xn+1 xm+1 1

Figure 4.7: Case Differentiation of Extended Dimensional Matrices

only added to by the new dimensionless parameters m+i . If, however, there is only one
exponent ji 6= 0 in a new rank column, then no further dimensionless parameter m+i can
exist because, for the extended order, the condition of dimensional homogeneity cannot be
fulfilled, due to a lack of partner with a second aji 6= 0. In this case, the entire relevance list
has to be tested in order to obtain a dimensionally homogeneous model.

The granularity of the evaluation method has the advantage that the former m remain un-
changed during a further refining of the description and are extended by adding mere  ’s. In
this way, the hierarchical principle of detailing in compliance with successive refining of the
description, is fully supported from the top to the very bottom.

4.3.5 Hierarchy

The description graph corresponding to the representation f (x1 ; : : : ; xn ) = 0 and that of

its representation as evaluation graph F (1 ; : : : ; m ) = 0 accordingly, possess exactly one
hierarchy level which immediately results from the chosen solution of the implicit functions f
and F . Due to the mapping of the description graph onto the evaluation graph, to determine
the structure of the evaluation graph, the conversion in xj = f (x1 ; : : : ; xj ?1 ; xj +1 ; : : : ; xn ) is
sufficient to determine the dimensionless coefficient (see Appendix A.4 or [35]) associated
to xj with

cj = '0 ('1(x1 ; : : : ; xn )) (4.13)

= '3 (1; : : : ; m ) (4.14)
as an evaluation aspect. With this, only those aspects can be evaluated where a descrip-
tion is known. An example of a single stage hierarchical set-up of the description and the
60 CHAPTER 4. SOLUTION APPROACH

evaluation graph is represented in fig. 4.8.

'0
  X
'3 6 6
'1
'j
(1 ; : : : ; m )  (x1 ; : : : ; xn)

Figure 4.8: Mapping Scheme

The proven existence of equation 4.14 with the calculation of a total evaluation taken
from partial evaluations represents the solution to the aggregation problem which is being
searched for.

4.3.6 Modularity

If an object consists of numerous sub-objects, these can be modeled and evaluated sep-
arately as well as together. The compatibility of the dimensionless parameters which are
generated by the varying descriptions is made secure because the respective synthesis of
the dimensionless products is known and they can simply be compared with one another. It
is therefore also conceivable to store the functional descriptions f as well as the functional
evaluations F of parts or processes which occur frequently in a database, thus resulting in
an analogy on the concept of finite elements [2, 90]. This will be more fully portrayed in
example 5.2.2.

4.3.7 Coupling

The structure of all dimensionless products j taken from the design variables x1 ; : : : ; xn
can be summarised in a Boolean topological matrix a . Thus  = a T a holds for the coupling
matrix  for the entire system. If the unity matrixII is also inserted into this relation, a coupling
matrix of the design variables is retained with:
 = aT I a (4.15)
The elements of the matrix I can, at the same time, be interpreted as the natural degrees
of freedom of the design object, whose combining results in the properties of the entire
object. This is in analogy to the building up of the entire stiffness matrix from the topology
matrix a and of the individual stiffness matrices K n formulated in natural coordinates of
each finite element [2, 90]. During the calculation of such a coupling matrix in figure 5.16 of
example 5.2.2, this analogy will be demonstrated in more detail.
4.4. EPISTEMOLOGICAL CHECKS 61

4.3.8 Sensitivity

Two design objects having different quantity features of the same amount of design quantities
X = fx1 ; : : : ; xn g, are called similar, when the dimensionless products
Yr i = 1; : : : ; r
j = xj x?i ji
(4.16)
i=1 j = 1; : : : ; m
turn out to be identical. If the definition of the scaling factors Kj with

xjb = Kj xja j = 1; : : : ; n (4.17)

are introduced for the ratio of design quantities of these two objects (index a and b), the given
model laws [59]
Yr i = 1; : : : ; r
Kj = Ki ji (4.18)
i=1 j = 1; : : : ; m
result by insertion into equation 4.16. If the total differential for the expression in equa-
tion 4.16 is formed, the result is
Xr @
d j = @ j
@xj d x j+
j
d xi j = 1; : : : ; m: (4.19)
i=1 @xi
For d j = 0, equation 4.19 yields the formal representation of the iso-hyperplane for the par-
tial quality j of the design. With infinitesimally small changes of only two design quantities
xj and xi in a given design object, the result is according to equation 4.19 from the claim to
similarity for the calculation of the partial derivatives that

@xj = xj i = 1; : : : ; r (4.20)
ji
@xi xi j = 1; : : : ; m
applies. The sensitivity analysis with dj = 0 thus results from the differential formulation of
the known model laws with j = const. The formal reference beyond that from equation 4.20
on the technique of qualitative reasoning within artificial intelligence [7] will be represented
in section 5.2.2.

4.4 Epistemological Checks

In order to test the plausibility of the evaluation hypothesis, some arguments have been
summarised hereafter which serve to check the hypothesis in question. On the one hand,
motivated from an epistemological viewpoint and on the other from a purely mathematical
viewpoint, these deliberations represent a collection of necessary conditions, to which an
evaluation model and its units of measurement have to comply.
62 CHAPTER 4. SOLUTION APPROACH

4.4.1 Invariancy

Under the prerequisite that an objectively concise evaluation of an object actually exists at
all, the evaluation quantity cannot depend upon the arbitrarily chosen definitions of basic
quantities of the units of measurement system (the Étalons de standard, e.g. the master
meter) in which the object is described. In this respect it should be demanded of the eval-
uation, that the evaluation of an object’s properties produce results that are not dependent
upon the fact whether, for example, the length of an object is measured in inches or in me-
ters. This means however, that evaluations can only be dimensionless quantities which are
invariant under such scaling transformations.

4.4.2 Homogeneity

An evaluation quantity which is objectively reproducible, can only exist when founded upon
a law. In the course of growing human knowledge, the most commonly occurring physical
model formation in the form of equations means that the description of the object used as
well as the mapping of the object description onto the object evaluation have to be, as with all
laws of physics, dimensionally homogeneous. The requirement of dimensional homogeneity
which appears firstly as a severe restriction is always fulfillable through the introduction of
fundamental physical constants. Moreover, in the thought structure of physics, this demand
is a fundamental component [68] in the construction of non-contradictory statement systems.

4.4.3 Causality

The required law for the existence of a reproducible evaluation quantity is supported by
the principle of causality [13]. The rigorous interpretation of the principle of causality [50]
represented in fig. 4.9, which states that the “occurrence ” is necessarily and sufficiently
determined by the “occurrence X”, corresponds with the condition of a complete relevance

  X
Figure 4.9: The Principle of Causality

list of physical quantities in the sense of the evaluation model formation using the Pi theorem.
4.4. EPISTEMOLOGICAL CHECKS 63

4.4.4 Consistency

With evaluations made in differing hierarchy levels it should be guaranteed that despite dif-
fering calculatory paths, evaluations will result which are always consistent with one another.
This can be tested using a comparison of results from both the possible mapping sequences
of ' = '3  'j versus ' = '0  '1 in fig. 4.10. In this mapping, '1 corresponds to the func-

'0
(Evaluation Graph)
  X (Description Graph)
'3 6 6
'1
'j
(1 ; : : : ; m )  (x1 ; : : : ; xn)

Figure 4.10: Mapping Scheme

tion f and '3 to the function F (vertical consistency), the 'j and '0 are the corresponding
dimensionless product transformations j , (horizontal consistency).

The consistency of the evaluation over and above the hierarchy levels can be traced back to
the consistency of the description f , due to the properties of the mapping made evident by
the Pi theorem. Thus, the consistency of multi-leveled hierarchies results from the recursive
interlinking of the argumentation for simple hierarchies.

4.4.5 Abstraction

Abstraction means the validity of a property in numerous elements, which can therefore
be summarised within a heading. The mapping n : 1 means mathematically, according
to fig. 4.11 (drawn from [47]) the condition of a surjective and non-injective mapping. The

r Xr
 
r Xr 
r Xr 
r Xr
rHYH r r r rHYH r r r
r HHr r H Hr r r
not injective injective not injective bijective
not surjective not surjective surjective

Figure 4.11: Mapping Properties (according to [47])

property of abstraction is fulfilled for the partial evaluation because of the product structure
of the dimensionless parameters, for the entire evaluation due to the existence of function F
and for the global descriptions through the existence of the function f .
64 CHAPTER 4. SOLUTION APPROACH

The theory of abstraction [63, 103] indicates analogically that extensional abstraction stipu-
lates the existence of an indefinite universal quantor ^
^ for which
^^X;Y X Y ?! (A (X ) $ A (Y )) (4.21)

applies. In this   represents the equivalence relation, with reference to the statement form
A (X ) and A (Y ), to which both the infinite statement sets X and Y are equivalent.

4.4.6 Identity

The usage of dimensionless parameters j as a unit of measurement for evaluation guar-

antees identical results exactly for those objects or processes which belong to the same
similarity class. This means that through a hyperplane in the description area defined by a
j = const, each hyperplane is mapped onto the very same point in the evaluation space,
see fig. 4.12.

u 'j hhh
 E
 h E
  E hh
 E ((
(( ((E
 E

(1 ; : : : ; m) (x1 ; : : : ; xn)

Figure 4.12: Mapping of Similarity Classes

Due to the special property of dimensionlessness, the ascertained dimensionless products

are independent of the physical measurements of comparison used and are therefore a pure
property of the physical object or process.

4.5 Discussion

In this section, the overlapping aspects of the suggested method of evaluation are to be
discussed. Because the method of evaluation is mainly founded upon the assumptions
made as a prerequisite of proof for the Pi theorem, these assumptions need to be discussed
in a more general framework.
4.5. DISCUSSION 65

4.5.1 Area of Validity

The requirement of the dimensional homogeneity of functions is ’weak’ in that it is purely

epistemologically founded and can always be fulfilled by introducing an appropriate funda-
mental physical constant. It is therefore not possible to ever experience a formal contradic-
tion by means of the method of dimensional analysis within the entire thought structure of
physics, as shown in fig. 4.13.
'
' $
' $
' $
' $
simplified
exact
theories 
&theories %
&discrete equations %
&discrete experiments %
&thought structure of physics %
&physical reality

Figure 4.13: Fields of Validity

The methodology’s field of validity is therefore not limited to the areas of ‘precise’ physical
knowledge and due to the character of being a necessary condition in a mathematical sense,
it is also a useful methodical means of control and help in border areas of ‘imprecise’ physical
knowledge.

4.5.2 Limitations in Usage

The practical use of the evaluation methodology however is limited to those areas of know-
ledge, in which at least statements about the class affiliation of the objects (sets of features)
and the linkage rules (functional relationships) defined for them are explained. The formation
of feature sets leads here to the formation of a physical unit system, while the definitions of
operations concerning these quantities signify the construction of an algebraic structure [68].
In this structure, the formation of models in the form of functions f can be established by
means of a mapping of the object properties onto the domain of the real numbers.

The presently known 7 physical basic quantities and their unit names with their attributed SI
units represented in fig. 4.14, are mass M (kg), length L (m), time T (s), temperature  (K),
current intensity i (A), amount of material  (mol) and light intensity I (cd). These primarily
66 CHAPTER 4. SOLUTION APPROACH

independent classes of properties are, for instance, interchangeable and transferable into
one another through equations [68] of the following form:

E  kT
E  m c2 (4.22)
E  h
The aggregation problem of evaluation in the areas of economy, ecology, sociology and
aesthetics etc. with the mentioned evaluations from the area of engineering will remain thus
unexplained until all the descriptions forms can be expressed and converted into a common
unit system which can be compared with the physical unit system. Within the physical unit
system this can, for instance, occur through a chosen equation of the form 4.22 by means
of which these units can be connected and converted into physical units commonly used in
the various areas of engineering.

Of interest is that within the thought structure of mathematics and physics, a necessary as
well as sufficient condition for this step, the postulation of further laws of nature and the
necessity to identify the natural constants, which are obligatorily contained within, can be
stated.

4.5.3 Freedom of Selection

The reduction to m = n ? r dimensionless variables, attainable by means of the Pi theo-

rem, is affected — apart from the number of n physical quantities, which are defined by the
granularity of model formation — by the rank r of the dimensional matrix. It is therefore an
advantage, with regard to the number of existing columns, to achieve the highest possible
rank of the dimensional matrix. This can possibly be done by means of a transition to another
equivalent physical unit system. See a selection of equivalent unit systems to this effect in
fig. 4.14.

Because of the increased pictorial interpretation in mechanics, there were fF; L; T g as well
as fM; L; T g unit systems in use before the obligatory introduction of the SI measuring unit
system. In a pure force-path-problem, one of the advantageous usages of the fF; L; T g as
opposed to the fM; L; T g system resulted from the immediate ability to identify the dimen-
sionless products taken from the right hand dimensional matrix in fig. 4.15, in which the
relevant physical quantities of the bar deflection are listed in a fF; L; T g system:

1 = ElP 2 2 = lI4 3 = ul (4.23)

In fig. 4.15, the transformation matrices necessary for the transformation of unit representa-
tions of physical quantities from fM; L; T g into the fF; L; T g system and vice versa are given
4.5. DISCUSSION 67

fM; L; T; ; i; ; I g Units
SI-

?
fM; L; T; g thermo-
dynamics
more powerful
unit systems ?
. . . . . . . . . . . . . . . . . . p = nk# . . . . . . . . .
equivalence unit systems
fF; L;
*
V g
.....
.....
........................
..........
.......
......
.....
.....
..... ....
....

v = at
....
......
. ....
.... ....
... ...
...
... ...
... ..
... ...
.. ..
.. ..
..
..
??
..

fM; L; T g
..

v = st?1
..
...
...
..
technical
..
..
.. . mechanics
..
..
..
...
..
..
..
6
.. ...
.
...

F = ma
..
... ...
... ...
.... ...
.... ....
....
..... ..
.....
..
..... .....

fF; j
..... .....

L; T g 
.... .....
.......
.
..
...
...
..
...........
.
...........
.

?
engineering unit system
. . . . . . . . . . . . . . . . . . F = G mr m 1
2
12
2
. . . . . . . .
less powerful
unit systems
?
fL; T g astro-
physics

Figure 4.14: Units and Equivalent Unit Systems

1 -1 2 1 1 -2
1 1
[M] [L] [T] 1 () [F] [L] [T] 1
E 1 -1 -2 E 1 -2
l 1 ? l; E; I P l 1
P 1 1 -2 ?............................................................................
?
? P 1
u
....................
................
? ..........?
I I
..............
..............

4 ? 4
?
u 1 u 1

Figure 4.15: Bar Deflection (Linear Theory)

as an example. Their product always produces the unit matrix I . Physical quantities which
are expressed in equivalent unit systems can therefore always reversibly and unambiguously
be transformed into the representation form of any other equivalent unit system.

Corresponding to the differentiation quality of varyingly sized unit systems according to

fig. 4.14 in the best possible case a reduction of, at the most, 7 is produced, in classical
thermodynamics, at the most 4, in classic mechanics, 3 and in astrophysics, a reduction of,
at the most, 2 variables. There have been, in that respect, investigations to see how far
for example, by means of a separate treatment of the unit of length [L] in each direction in
68 CHAPTER 4. SOLUTION APPROACH

space as [Lx ; Ly ; Lz ], the rank r of the dimensional matrix increases [51] and how a further
reduction of the number of dimensionless parameters can be achieved.

4.5.4 Modes of Consideration

Although outside the set task of this work in the context of the research group [28, 29]
but however, interesting and interdisciplinary, a formal mathematically orientated interpre-
tation of the possible origin of an intergroup conflict will also be illustrated here. Because
descriptions and evaluations of buildings are undertaken in the framework of an holistic
consideration of engineering (bridge) constructions, tensions between the interdisciplinary
group members result from the differentiating viewpoints and modes of consideration held
by human decision makers (i.e. the architects) and engineers (i.e. the technical people).

“The decision maker” can only make limited statements about the completeness and struc-
ture of his evaluation criteria because formally, a fundamental system would be necessary
for this. Due to the complete lack of knowledge about the inner structure of the transfer
functions for this purpose, this evidence cannot be produced. Therefore the path from left to
right and back in fig. 4.16 is basically barred if the inner structure of the m is unknown.
.......................................
....... ......
......

1 ; 2 ; : : : ; m
.....
..... .....
..... ....
..... ....
-
. ...
...
.
..
...
.
. ...
..
..
“Engineer”
+
..
Evaluation ..
. ..
..

*
.
. ..
. ..
..
. .
.
.
.. .
.. ..
.. ...
.. ..
...
 ..
Description
x1 ; x 2 ; : : : ; x n
.. ..
... ..
“Decision Maker” ...
....
....
..... ..
.....
....
...
..... ....
...... .....
......... ......
.....................................

Figure 4.16: Interdisciplinary Conflict of Viewpoints

“The engineer” is in an only seemingly more advantageous situation. The path from right to
left (and back) in fig. 4.16 is basically open and in some cases the establishment of a transfer
function may also succeed. However, the engineer is far from knowing the completeness of
his description xn and therefore far from retaining the knowledge about a complete funda-
mental system.

In an interdisciplinary research group, this fundamental difference of viewpoints, which

touches on a basic dilemma, can at first only be overcome piece by piece through col-
lectively arranged conventions on the carefully established completion of descriptions and
evaluations by means of definitions.
 69

5 Examples of Usage

In this chapter, the history, further development and some application examples of dimen-
sional analysis will be shown. The first part includes three selected examples which should
clarify the advantages of applying dimensional analysis to the simultaneously occurring,
gradual transition from areas of “precise” into areas of “imprecise” physical knowledge. Like-
wise, the second part is concerned also, in three examples with the application of the Pi
theorem upon the actual modeling and evaluation of objects in the design process. The third
part briefly deals with the generalised method of group transformations and concludes with
an example from the latest state of knowledge on holistic evaluation being aspired to.

5.1 Classic Dimensional Analysis

The two indispensable requirements of dimensional analysis thought out, the terminology
concepts of dimension and of homogeneity go back through F OURIER, N EWTON and G ALILEI
right to the ancient Greeks. T HEON (100 A.D.) has already quoted [48]:

“Relationship is the certain behaviour of two homogeneous quantities to-

wards one another. Because how inhomogeneous quantities behave to-
wards one another is impossible to know, states A NDRASTOS.”

For V IETA (1591) this reasoning adheres closely to the homogeneity principle as a conse-
quence. Corresponding to the present day form of the equation a3 + d
a = c (in latin, in
means
, aequatur means =), he writes [102]:

“Si A cubus + D plano in A, aequatur C solido.”

whereby he explicitly states the dimensions of the variables C and D. As the first researcher
in modern times, F OURIER (1822) is awarded with having defined the concept of the phys-
ical dimension, by applying the principle of dimensional homogeneity to sums of physical
equations and introducing the dimension test for physical equations, he showed that con-
clusions on physical relationships can be drawn through the analysis of the dimensions [30]
concerned.

Before the turn of the century, the experimenters and users were still a long way ahead of
the theorists. Ten years ahead and independent of R EYNOLDS (1883), H ELMHOLZ (1873)
ascertained the dimensionless groups of hydrodynamics and determined the transition point
70 CHAPTER 5. EXAMPLES OF USAGE

of laminar to turbulent flow in pipes [36]. That the necessary exact theoretical fundamen-
tals still had to be proved was not considered to be of prime importance by men such as
H EAVISIDE (1885), one of the defenders of experimental mathematics [35]:

“Am I to refuse to eat because I do not fully understand the mechanism

of digestion ?”

After the turn of the century and further important works from P RANDTL (1907) for example,
the hour of the theorists came at last. Attempts at proving and proofs of the Pi theorem sup-
ported by extremely differentiatingly strong assumptions, which were also sometimes wrong
or incomplete or often gained through unknowledgeable research of the time, came from
R IABOUSCHINSKY (1911), BUCKINGHAM (1914), RUNGE (1916) and B RIDGEMAN (1922) to
name but a few important names [36]. The even earlier evidence from VASCHY (1890) was
not incorporated by the expert world. The unnecessarily strong assumptions of this evidence
are for example its limitation to integral exponents ji  IN in the unitary representation and
the assumption of the C 1 -constancy of the function f .

The mathematical consolidation of the reasoning and the reduction of assumptions were
first concluded by B IRKHOFF (1950) and H AINZL (1968). This was done by proving the
fundamental statement of the Pi theorem in a pure algebraic sense [9, 68] and, as briefly
shown in the third part of this chapter, generalising it to the even broader method of group
transformations [41] as E HRENFEST-A FANASSJEWA (1926) had already suggested. Further
details on the history of the Pi theorem can be found in [24, 35, 36]. Firstly though, the first
three out of the seven examples mentioned will now follow.

5.1.1 Mathematical Pendulum

The mathematical pendulum [35, 64, 93] in fig. 5.1 consists of a point mass m, which has
been placed upon a massless rod of length l, and can be turned at the point O. With a lateral
displacement corresponding to the angle  from the point of balance and a complete neglect
of frictional forces, a periodic oscillation sets in, under the influence of gravity. The period of
the oscillation T is sought.

Dimensional Analysis Solution. From the dimensional matrix shown in fig. 5.1 consisting
of n = 5 physical quantities and a rank of r = 3, n ? r = 2 dimensionless products directly
result.
q
1 = T g=l (5.1)
2 =  (5.2)
5.1. CLASSIC DIMENSIONAL ANALYSIS 71

 p
[M] [L] [T] SI units A@
A @
O '
m 1 kg ..........A
......*
..

l 1 m 
..................... A
........
....@
A........... ..
.....

@
g 1 -2 m=s2 g A
@
A
l e
 rad A
A
@..............

um
...
.....

T s
.....

1 ? A................
..
.
.....
.....

....... ..
..........
............
....................................................

Figure 5.1: Mathematical Pendulum

The now necessary determination of the function F (1 ; 2 ) = 0 generally only results from
the approximation of experimental data or from theory. However, in cases where lesser
influence quantities are concerned, the dimensional analysis provides even further reaching
information about the function F . This is shown in the following.

The determination of the period of oscillation T from the given quantities corresponds to the
question whether a dimensionally homogeneous relation exists in the form

f (m; l; g; ; T ) = 0 or
T = f1(m; l; g; ) (5.3)
assuming that this “intuitive” list of influence quantities is complete. This assumption can
however only in the end be verified by the experiment.

Investigation of equation 5.3 by means of a first  times virtual scaling ( > 0) of the unit
of mass by m to ?1 m, one recognises immediately that T cannot be dependent upon m.
The required invariancy under scale transformations can otherwise not be attained (no other
physical quantity possesses this unit). In this way, equation 5.3 is reduced to

T = f2 (l; g; ) (5.4)
By means of a second  times virtual scaling, ( > 0) of the unit length, T and  remain
unchanged, whereby l converts into ?1 l and g into ?1 g . Therefore only the invariant ratio
l=g can exist and equation 5.4 becomes
T = f3 (l=g; ) (5.5)
By means of a third and last  times virtual scaling ( > 0) of the unit of time, T converts
into ?1 T and l=g into ?2 l=g . Therefore the equation 5.5 can only be unitarily invariant in
the form
q
T = l=g f4 () (5.6)
The function f4 () still remaining in this equation, cannot be determined by the dimensional
analysis. It must be the result of experiments or theoretical consideration, due to the fact
72 CHAPTER 5. EXAMPLES OF USAGE

that f4 () is already dimensionless when developing f4 () into an exponential series of the
form

f4() = a0 + a1  + a2 2 + a3 3 + : : :
X 1
= a0 (1 + aai i) (5.7)
i=1 0

being a0 6= 0, mostly only the first terms of the series ought to be taken into consideration
for a sufficiently exact model formation. Therefore only a few experiments are necessary to
solve the approximation assignment. For linearised systems, i.e. small lateral displacement
  1, only one single measurement is necessary to determine a0 .

Analytic Approximation Solution. To check the dimensional analytical solution and to

directly compare the expenditure, the solutions of the linearised and of the exact differential
equation (DE) are given [108]. Additionally, the DE is deduced with the help of the angular
momentum theorem by point O in fig. 5.1.

ml' = ?mg sin ' or

' + gl sin ' = 0 (exact DE) (5.8)

' + gl ' = 0 (linearised DE)

Therefore the independence of the period of oscillation T is immediately confirmed by m in

equation 5.4. Given the initial conditions '(0) =  and '_ (0) = 0, linearising the differential
equation for small angles by means of the approximation (sin '  ') and comparing the
coefficients with the general formulation, 5.9, solution 5.10

' = A sin(!t) + B cos(!t) (5.9)

' =  cos !t (5.10)
q
! = g=l (5.11)

can be derived. From the angular eigenfrequency !

T = 2! (5.12)
q
= 2 l=g (5.13)

results for the period of oscillation T sought. Thus, the approximation of f4 () for small
angular displacement is determined by the constant a0 = 2 (equation 5.7).
5.1. CLASSIC DIMENSIONAL ANALYSIS 73

Exact Analytical Solution. To calculate the exact solution, one best makes a deduction
from the formulation in the energy principle, in which one formulates the equality of energies
in the beginning and in the general aspect, under frictionless conditions

mgl cos  = mgl cos ' ? m2 l2 '_ 2 and therefore

s
'_ =  2lg (cos ' ? cos ) (5.14)
Integration of the differential equation 5.14 by separating the dependent and independent
variable '_ = d '=d t leads, for the period of oscillation T , to the normal form of a complete
elliptic integral of the first kind. This solution is still a function of the module sin(=2), also
described as K (sin(=2)) and therefore also of the initial conditions
s Z=2
T = 4 gl q d'
0 1 ? sin2 (=2) sin2 '
s
= 4 gl K (sin(=2)) (5.15)
The values of the integrals can be taken from tables [54] for the varying values of the module.
Approximation of the solution for small values of the modules using an exponential series
yields
s "  2  1
3 2 #
T = 2 l 1+ 1 2
sin (=2) + 4
sin (=2) + : : : (5.16)
g 2 2
4
The period of oscillation T increases therefore with increasing lateral displacement. Fur-
thermore, the sought coefficients of the function f4 () result from the comparison with equa-
tion 5.7.

5.1.2 Ascending Bubble in a Pipe

The velocity of a vertically ascending bubble in a vertical, water-filled pipe with circular cross-
section is of interest in pump construction [19]. With the justifiable assumptions by fluid flow
mechanics, (semi-infinite bubbles; disregard of the secondary running effect; inviscid, incom-
pressible, rotationally-free and stationary flow; disregard of capillary forces), the influence
quantities given in fig. 5.2 can be stated.

Dimensional Analysis Solution. The dimensionless products result from the dimensional
matrix of the problem
q
1 = v= gd (5.17)
2 = l =w (5.18)
74 CHAPTER 5. EXAMPLES OF USAGE

- d 
??
?
?
?
? w ??
?
?
? ?
? ??
? ??
[M] [L] [T] SI units ?
?
? ??
?
d 1 m ?
? v ?
?
?
? g
?
?
g 1 -2 m=s2 ?
? 6 ?
?
?
?
?
l 1 -3 kg=m3 ?
? ............................. ?
? ... ?
?
?
?
w kg=m3 ?
? .... .
.. ...
.. ?
.
.
.. ?
..
1 -3 ?....
? .
.. ?
. ..
..?
?...... @
..

v 1 -1 m=s ?
?
?
.
.
@
B .. ?
..
.
.?

? ??
? ??
?
? ?
?
? ?
?
?
?
l
?
?
?
?
? ??
?
?

Figure 5.2: Ascending Bubble in Pipe

Because the relevance list is small, additional information, such as in the preceding example,
about the function F (1 ; 2 ) = 0, can be obtained.

The determination of velocity v

from the given quantities, corresponds once again to the
question about the existence of a dimensionally homogeneous relation taking the form

f (d; g; w; l ; v) = 0 or
v = f1 (d; g; w; l ) (5.19)
assuming also that this intuitive list of influence quantities is complete. This assumption can
ultimately however only be verified by the experiment. In equation 5.19, one recognises
through a first virtual  times scaling ( > 0) of the unit of mass from m to ?1 m, so that l
with ?1 l and w with ?1 w change. An invariance through the transformation of the unit of
mass can therefore only be obtained by the modification of equation 5.19 to

v = f2 (d; g; l ) (5.20)

w
By means of a second virtual  times scaling ( > 0) of the unit of time, v changes by 1 v
p
and g by 2 g . Therefore an invariancy can only exist through g and equation 5.20 becomes

v = pgf3(d; l ) (5.21)

w

Through the third and last  times scaling ( > 0) of the unit of length v transfers to ?1 v ,
pg into ?1=2 pg and d to ?1 d. Therefore the equation 5.21 can only be unit-invariant in the
form
q
v = g d f4( l ) (5.22)
w
5.1. CLASSIC DIMENSIONAL ANALYSIS 75

The remaining determination of the function f4 (l =w ) still in it, has to be a result of experi-
ments or of theoretical considerations. Because l  w applies with the material constants
(as constant assumed factors), it can be expected with the concepts analogical to equa-
tion 5.7, that a single chosen experiment is sufficient to determine f4 in equation 5.22.

Analytical Approximation Solution. An exact analytical solution for the sought-after as-
cension velocity is still unknown even today [19, 35]. The known analytical approximation
solution [19] for the potential equation in cylinder coordinates

 = @@r2 + 1r @@r + @@z2 = 0

2 2
(5.23)
is comprehensive and difficult due to the boundary conditions of the integration problem with

@ = 0
@r for r = d=2; ?1  z  1
!1 @z = ?v 0  r  d=2
@
zlim for
(5.24)
( @@r )2 + ( @@z )2 = ?2gz(1 ? wl ) on B
@ = 0
@n on B; n ? B

retaining the sought after quantity v as well as the unknown form of the bubbles’ outer surface
B which are again themselves part of the solution required. The approximation solution is
only available by dividing the rotationally symmetrical bubble into two parts. The first part
is an exponential function from infinity up to the level of the midpoint of the bubble. The
tangential transition to the second part, an arc of circle of approximately the same radius of
curvature as the bubble at the transition point, is continuous in the first and second derivative.

5.1.3 Object Support Mechanism

The object of a microscope [16] consists of a dovetail guided sliding plate of mass m which
can be moved to and fro in a dovetail keyway through a threaded pin with screw pitch t, by
turning the milled wheel of diameter D left or right. By means of two ball bearings, the pin is
immovably supported in a small casing which also incorporates the dove tail guide. This is
schematically illustrated in fig. 5.3. The dynamic viscosity  represents the properties of the
lubricant, the frictional forces are modeled by F and the material properties by the spring
stiffness c. If a model of the object support mechanism already exists, the question is posed,
through which detail improvement in the construction can the central positioning error x of
the object support be most effectively reduced.
76 CHAPTER 5. EXAMPLES OF USAGE

[M] [L] [T] SI units

m 1 kg
F 1 1 -2 kg m=s2
 1 -1 -1 kg=m s
t 1 m
c 1 -2 m=s2
D 1 m
x 1 m
F; ; m; c - x
@
@
@
@ ?
?
Q
 Q
 6
Q
Q Q
Q
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
??
??
?
?
?? D
Q

-  t Q

Q
Q Q
Q ?

Figure 5.3: Object Support Mechanism (scheme)

Dimensional Analysis Solution. From the dimensional matrix are yielded the four dimen-
sionless products (n = 7; r = 3)

1 =  t
2 3
(5.25)
mF
2 = m Fc3
2 2
(5.26)
3 = m2 DF3 (5.27)

4 = m2 xF3 (5.28)

to whose completion the following heuristics are related: Those variables for which propor-
tionalities (x  t; x  c?1 ; x  D ?1 ) can be given due to the physical understanding, are
not used to choose the necessary three basic variables (for rank r = 3) of the dimensional
matrix. As basis variables, the choice of a set of r physical quantities is understood, whose
partial matrix also possesses the rank r, within the dimensional matrix. Those heuristics
mentioned guarantee that at least the sign of the exponents ; , and  of the approxima-
tion can be correctly determined for the real tangential hypersurface
C0 = (1 ) (2 ) (3 ) (4 ) (5.29)
covered at the design point of the prototype. If this is solved according to the sought-after
quantity x, choosing = = =  = 1=3, then
! ! ! !
x = C1 m Ft 2 F 2 mF mF
2 3 1
3
1
3
1
3
1
3

0 m c3 2 D 3 2
5.2. MODELING AND EVALUATION 77

= C1 ct FD (5.30)
0
is returned. In the case in question, in which the deduction of an analytical solution expres-
sion using conventional means could be questioned, a first solution statement can be rapidly
produced by means of dimensional analysis. If, in addition the measurement data of a pro-
totype is available, the constant C0 can be determined and therefore the influence of small
constructive adjustments can be roughly estimated.

5.2 Modeling and Evaluation

In this section, the application examples of dimensional analysis upon the evaluation problem
are summarised. Here, each of these three examples illustrates a particular point of view.
With the body of a car, for example, it is the consistency of the transformational and aggrega-
tion functions which is represented, in the example of the pressure valve, the coupling matrix
of the design variables as well as a connection to artificial intelligence is represented, while
in the third example of the turbine, the successive refining of description and evaluation in
the process of modeling in the design process is represented.

5.2.1 Body of a Car

If one regards the resistance minimising form of the car body shown in fig. 5.4, the dimen-
sional matrix of the relevant physical influence quantities density , kinematic viscosity  ,


;   LL
v W
@
@ m -
mA

 l -

Figure 5.4: Sketch of a Car

a characteristic length l, velocity v and air resistance W can be completed by means of a

dimensional analytical consideration. Due to m = n ? r = 5 ? 3 = 2 in fig. 5.5, the two
following dimensionless products can be obtained

1 = vl (5.31)
2 = vW2 l2 (5.32)
78 CHAPTER 5. EXAMPLES OF USAGE

[M] [L] [T] SI units

l 1 m
v 1 -1 m=s
 1 -3 kg=m3
 2 -1 m2 =s
W 1 1 -2 N
Figure 5.5: Dimensional Matrix

If the usual notations found in fluid mechanics are used as follows,

Re = vl ( = 1 ) (5.33)
cw = vW2 l2 = qlW2 ( = 2 ) (5.34)
the determination of air resistance W , dynamic pressure q and characteristic cross sectional
surface l2 is then sufficient in a macroscopic consideration, because more does not need
to be made known about the flow condition which is finally determined by the exact form of
the car body. This corresponds to the global transfer function '0 in fig. 5.6. If however, the
'0
cw  Car
6'3 H
* Y
'
 1 HH
'2  H
Re  Air Shape

Figure 5.6: Car Graphs

function F (Re; cw ) = 0 is solved according to cw , the immediate result from the dimensional
analysis is that the resistance coefficient cw with

cw = G(Re) (5.35)
depends only upon the Reynold’s number Re and therefore the resistance coefficient of a
particular geometry only depends thus upon the flow condition. It is worth noting that be-
cause of this, the function G(Re) only has to be experimentally determined once for each
geometry family of interest, as can be inferred from the literature [35] for the standard ge-
ometry “sphere” represented in fig 5.7. The ascertion of the function G(Re) in fig. 5.7 cor-
responds therefore to the determining of the aggregation function '3 in fig. 5.6, while the
mapping specification Re = vl= represents the transformation function found there. The
consistency of the evaluations of both levels of consideration '0 and '2 is therefore secure
because they result from the purely mathematical transformation of the consistent descrip-
tion '1 of the physical model, found in fig. 5.6.
5.2. MODELING AND EVALUATION 79

103
....
....

102
....
....
.....
....
.....
....
....
6 .....
....
.....

101
.....
.....
.....
.....

cw
.....
.....
.....
.....
.....
.....
......

100
......
......
.......
........
........
..........
..............
.............................................................................................
..
..

10?1
...
.. ......
... .................... .........
.............

10?1 100 101 102 103 -104 105 106 107

Re
Figure 5.7: Resistance Coefficient cw of a Sphere (from [35])

5.2.2 Pressure Valve

The pressure valve [7, 104], shown in fig. 5.8, is modeled in three steps. Firstly the flow
of a fluid through the pipe is modeled by means of the relevant quantities, density , valve

pi ;  a - q; po
 
 
 

k; x 6

p

Figure 5.8: Pressure Valve

diameter cross section a, inlet pressure pi , outlet pressure po and the volume flow rate q .
This yields the dimensional matrix A of rank r = 3, as shown in fig. 5.9, from which the two

A [M] [L] [T] SI units

 1 -3 kg=m3
a 2 m2
pi 1 -1 -2 N=m2
po 1 -1 -2 N=m2
q 3 -1 m3 =s
Figure 5.9: Dimensional Matrix Partial System A

dimensionless products A1 and A2 due to m = n ? r = 5 ? 3 = 2

A1 = q 1=2
1=2
(5.36)
a pi
A2 = ppo (5.37)
i
can be obtained. Both the dimensionless products A1 and A2 are not independent of one
another, rather they are coupled by (at least one) base variable, because of the condition of
80 CHAPTER 5. EXAMPLES OF USAGE

dimensional homogeneity. This is represented as an undirected graph in fig. 5.10. Here the
 pi

A1 A2
 
Figure 5.10: Partial Graph A

’s are generally the m nodes and the common base variables form the edges of the undi-
rected graph. This relationship can also be constructed as a coupling matrix of the partial
object A in fig. 5.11 according to the FALKian Scheme [10] corresponding to equation 4.15.
In a second step, the valve is modeled by means of the relevant physical quantities of the

q  a p i po
a 1 1 1 1 A1
1 1 A2
I 1 1 1 1 1 Ia
1 1 1
1 1 1 1 1
1 1 1 1 1
aT 1 1 1 1 1 A
1 1 1 1 1 2 1
1 1 1

Figure 5.11: Coupling Matrix  A

stiffness of the spring k, the lateral displacement x and the control pressure p. This yields

B [M] [L] [T] SI units

k 1 -2 kg=s2
x 1 m
p 1 -1 -2 N=m2
Figure 5.12: Dimensional Matrix Partial System B

the dimensional matrix B in fig. 5.12 from which precisely one dimensionless product B1 is
obtained.

B1 = xkp (5.38)

This can be represented once more as an undirected graph in fig. 5.13. The coupling matrix
of the partial object B in fig. 5.14 is again analogically constructed according to the FALKian
Scheme.

In the last step, the assembly of the single functional components takes place. In this, the
functional-geometric connection conditions of the partial objects are also expressed as a
5.2. MODELING AND EVALUATION 81


B1

Figure 5.13: Partial Graph B

p x k
a 1 1 1 B 1
I 1 1 1 1 Ia
1 1 1 1
aT 1 1 1 1 B
1 1 1 1

Figure 5.14: Coupling Matrix  B

dimensionless product. Thus the valve cross sectional surface is geometrically coupled with
the lateral displacement of the slide plate. The functional coupling of the pressures po and p
result from the topology of the construction. Therefore, two further products

C 1 = pp (5.39)
o
C 2 = a1x=2 (5.40)
are formed. The whole system of the five dimensionless parameters evaluating the design
are therefore not independent of one another, rather they are coupled by the base variables
common to the five parameters respectively. This is shown in fig. 5.15 once again as an
undirected graph. The representation of the design variable coupling matrix k of the entire
 pi

A1 A2
 
a po
x p 
C 2 B1 C 1
  
Figure 5.15: Graph of the Entire System

system according to the Falkian Scheme in fig. 5.16, therefore results from the matrices of
the partial object and the functional geometric connection conditions. The coupling matrix
 therefore results from the super positioning of the matrices of the partial object being
a function of the functional-geometric connection conditions. This represents an analogy
on the construction of the entire stiffness matrix  from the element stiffness matrix K n in
natural element coordinates and the boolean topology matrix a with

 = aT K n a (5.41)
82 CHAPTER 5. EXAMPLES OF USAGE

q  a p i po p x k
1 1 1 1 A1
1 1 A2
a 1 1 1 B 1
1 1 C 1
1 1 C 2
1 1 1 1 1
1 1 1
I A1 2 ; 1 C1 2; 1 1 1 Ia
B1 1 1 1
1 1 1
1 1 1 1 1
1 1 1 1 1
1 1 1 1 2 1 1
1 1 1 1 1 2 1 C1
aT 1 1 1 2 1 B1 
1 1 A1 2
;
1 2 1 1
1 1 1 1 2 1
1 C2 1 1 1

Figure 5.16: Coupling Matrix 

according to the concept of finite elements [2]. This analogy is shown in fig. 5.16 by means
of the statement of origin of the single dashed boxes.

Artificial Intelligence (AI). This reference and test example, used often in AI also refers
directly to the technique of “qualitative reasoning” [7, 72, 104] in AI about the notation used.
Understood by this term is the machine produced answer to the type of question: “How
does the outlet pressure po of the object behave when the inlet pressure pi increases ?” The
necessary calculation of the partial derivation @po =@pi here corresponds therefore to a graph
searching problem, which has to be solved by finding the expansion path C 1 ! B1 !
C 2 ! A1 of the chain rule in order to form the derivation sequence.
@po = ( @po ) ( @p
@k ) ( @x ) ( @a
@q
@ ) = (? 1 ) po < 0 (5.42)
@pi @p @k @x @a @q @ @pi 4 pi

In this way, a negative result for all design quantities po ; pi " IR+ indicates the existence of a
restoring quantity.
5.2. MODELING AND EVALUATION 83

5.2.3 Gas Turbine

The modeling of a gas turbine [32] can result in a first step in accordance with to the block
scheme in fig. 5.17. In this way only those thermodynamic quantities which overlap the

Q_ Z - W_ -
N
Gas Turbine -
Q_ A
Figure 5.17: Block Scheme Gas Turbine

system’s limitations need to be covered. These are thermal power Q_ Z , chemically fed in the
_ N and into (waste heat) power Q_ A.
form of fuel, which is transformed into the (shaft) power W
The resulting dimensional matrix G in fig. 5.18 with n = 3, has the rank r = 1. Therefore the

G [M] [L] [T] SI units

Q_ Z 1 2 -3 W
W_ N 1 2 -3 W
Q_ A 1 2 -3 W
Figure 5.18: Dimensional Matrix Gas Turbine G

two following dimensionless products G1 and G2 yield

G1 = W_ N =Q_ Z (5.43)

G2 = Q_ A =Q_ Z (5.44)
(Because Q_ Z is partly transformed into W _ N and partly released as Q_ A, Q_ Z = ?W_ N ? Q_ A
is applicable when considering a system-egoistic position.) Therefore a measurement of
W_ N and Q_ Z is sufficient for an evaluation on a macroscopic level, in order to obtain the
dimensionless product G1 for the evaluation of the power transformation. This is shown in
fig. 5.19. In this way, the mode of consideration G(G1 ; G2 ) = 0 is consistent as applying of

'0
G1  Gas Turbine

Figure 5.19: One-step Evaluation Hierarchy

the definition of the thermal degree of effectivity th shows.

_N Q_ A
th = ?W = 1 + (5.45)
Q_ Z Q_ Z
= ?G1 = 1 + G2 (5.46)
84 CHAPTER 5. EXAMPLES OF USAGE

Refinement. In a more in-depth, detailed consideration of the gas turbine structure [32]
with its components compressor V , combustion chamber B and turbine T , the result is the
modeling found in fig. 5.20. Here, the air taken in (state 1) is firstly compressed in the com-
q -
B
Q_ Z @ -
3 ?
@ 2 ?
@ W_ V ? W_ N
V  T -
? @
1q -?
?
@
@ - 4q

Figure 5.20: Component Scheme of Gas Turbine

pressor V (state 2), then heated in the combustion chamber B (state 3) and finally expanded
in the turbine T (state 4). The thermodynamic process in this is first ideally assumed as adi-
abatic, reversible compression (1!2), isobaric heating (2!3), adiabatic-reversible expansion
(3!4) and isobaric cooling (4!1). In fig. 5.21 this process is shown in the (T; s)-diagram.

T 6
.... .
.....
.... .
r3
..
.....
.
.
.
....... . .
.
.
....... .
.
.
...
.... .
.
.
.
......
. .
.
.
.
... ..... .
.
.
.. ...... .
.
.
........ .
.

r r
. . .

2 .
. .
....... .
.
.
..
. ....... .
.
.
.
...... .
..
..
.. .
.
.... .
......... .
.
.
........ ..
.

4
..
.
.
. .........
.
.
. ..
.......

r
.
.
. ........
. ........
.
. ...........
.
. .............
.
. ...............
.
.
. .
...
..
.. .
.....
................
.....................

1
s-

Figure 5.21: Idealised Gas Turbine Process in (T; s)-Diagram

Compressor. For the separate modeling of the compressor V in fig. 5.22, the mass flow
rate m
_ , the specific heat capacity cp, the inlet temperature T1, the shaft work in the com-
pressor W _ V and the outlet temperature T2 are sufficient. These five quantities form the
dimensional matrix V in fig. 5.23. With a rank of r = 3 and n = 5 it follows that two dimen-

@ -
@
@ 
2
V
?
W_ V
1q -?
?

Figure 5.22: Compressor V

5.2. MODELING AND EVALUATION 85

V [M] [L] [T] [] SI units

m_ 1 -1 kg = s
cp 2 -2 -1 J = kg K
T1 1 K
W_ V 1 2 -3 W
T2 1 K
Figure 5.23: Dimensional Matrix Compressor V

sionless products V 1 and V 2 yield

V 1 = mc W_ V (5.47)
_ p T1
V 2 = TT2 (5.48)
1
By using the relation 5.49 for the reversible adiabate as well as the definition of two thermo-
dynamic identities 5.50,
!?
T 2 = p2
1

T1 p1 (5.49)
R = c p ? cv and = cp (5.50)
cv
_ V , when ideal gases are assumed, can be fixed to
the compressor shaft work W
8 ! ? 9
< =
W_ V = ( ? 1) m_ R T1 : pp2
1

? 1;
1
 T2  
= m_ cp T1
T1 ? 1 (5.51)
Therefore the result from the two dimensionless products is the relation V (V 1; V 2) = 0
when transformed from equation 5.51 to

W_ V =  T2  ? 1 or
m_ cp T1 T1
V 1 = V 2 ? 1 (5.52)

Combustion Chamber. For the separate modeling of the combustion chamber B as an

ideal heat exchanger in fig. 5.24, the mass flow rate m
_ , the specific heat capacity cp, the inlet
_
temperature T2 , the heat power QZ fed in and the outlet temperature T3 are sufficient, when
the mass of fuel fed in is disregarded. These quantities form the dimensional matrix B in
fig. 5.25. Due to there being a rank r = 3 and with n = 5, the two dimensionless products
86 CHAPTER 5. EXAMPLES OF USAGE

Q_ Z-
- B -
3
2
Figure 5.24: Combustion Chamber B

B [M] [L] [T] [] SI units

m_ 1 -1 kg = s
cp 2 -2 -1 J = kg K
T2 1 K
Q_ Z 1 2 -3 W
T3 1 K
Figure 5.25: Dimensional Matrix Combustion Chamber B

B1 and B2 again yield

B1 = mc Q_ Z (5.53)
_ p T2
B2 = TT3 (5.54)
2
Here, the temperature increase in the isobaric combustion process can be determined as

Q_ Z = m_ cp (T3 ? T2 ) (5.55)
when ideal gases are assumed. Therefore, the two dimensionless products yield the relation
B (B1 ; B2 ) = 0 when transformed from equation 5.55 to
Q_ Z = T3 ? 1 or
m_ cp T2 T2
B1 = B2 ? 1 (5.56)

Turbine. For the separate modeling of the turbine T in fig. 5.26, the mass flow rate m _ , the
specific heat capacity cp, the inlet temperature T3 , the outlet temperature T4 and the outlet
shaft work W_ T are sufficient (Because the outlet power W_ T is used partly as W_ N and partly
3 ??
W_ V ? W_ N
 T -
@
@@ - 4q

Figure 5.26: Turbine T

for compressing as W_ V , W_ N = W_ V + W_ T applies when taking into consideration a system

5.2. MODELING AND EVALUATION 87

egoistical standpoint). The n = 5 quantities stated in fig. 5.26 form the dimensional matrix T
in fig. 5.27 with a rank of r = 3 and therefore, the two dimensionless products T 1 and T 2

T [M] [L] [T] [ ] SI units

m_ 1 -1 kg = s
cp 2 -2 -1 J = kg K
T3 1 K
W_ T 1 2 -3 W
T4 1 K
Figure 5.27: Dimensional Matrix Turbine T

yield

T 1 = mc W_ T (5.57)
_ p T3
T 2 = TT4 (5.58)
3
Using the relation 5.49 for the reversible adiabate, the given shaft work (outlet power) W_ T ,
when ideal gases are assumed, can be determined as
8 !? 9
< =
W_ T = ( ? 1) m_ R T3 : pp4
1

? 1;
3
 T4  
= m_ cp T3 T ? 1 (5.59)
3
Therefore, from two dimensionless products, the resulting relation is T (V 1; V 2) = 0 when
transformed from equation 5.59 to

W_ T =  T4  ? 1 or
m_ cp T3 T3
T 1 = T 2 ? 1 (5.60)

Evaluation. According to equation 5.45, the global evaluation from the detailed description
of the gas turbine represented yields

th = 1 ? TT4 ? T1 (5.61)

3 ? T2
and therefore

 T 2 ? (V 2 B2 )?1

G1 = 1 ? (B2 )?1
?1 (5.62)
Assuming ideal gases, this is possible by means of its three components, compressor V ,
combustion chamber B and turbine T , due to the now better known procedure. The aggre-
gation function '3 of the various partial evaluations to a complete evaluation thus determined
88 CHAPTER 5. EXAMPLES OF USAGE

'0
G1  Gas Turbine
' H
* Y
3  6HH 
*HY
6HH 1 '
 H '2  H
V 2 B2 T 2  V B T

Figure 5.28: Two-step Evaluation Hierarchy

is shown in fig. 5.28 together with the description graph. The transformation functions '0
and '2 coincide with the established  quantities. Here, the impression given of an optical
conformity between the description and evaluation graphs, is purely coincidental because
each of the three partial components V , B and T could be modeled by means of a relation
F (1; 2 ) = 0 containing exactly two dimensionless products.

Polytropic Process. A polytropic change of state is more realistic than the modeling of
the compression (1!2) and the expansion (3!4) as reversible adiabates. The slight change
in the process by the new changes of state (1!20 ) and (3!40) is shown in fig. 5.29. With the

T 6
.... .
.... ...
...
......
r3
......
. ..
.... ..
..... ..
..... ..
......... ..
...
...
..... ...
.....

r 40
... ..

2r0
..... ..
..
..
. ...
.......... ...
....
...... ..
...
............ ....
....... ....
.
...
.. ... . .
..... .
......
... ........
. ....
.. ......
......

r
.. .......
.. .........
.. ..........
..
. .
.....
...
.....
.
... ..............
. .................
.......................

1
s-

Figure 5.29: Real Gas Turbine Process in (T; s)-Diagram

general formulation for the polytropic relation

! nV ?
T2 = p2 nV
1

with  < nV < 1 for compression (5.63)

T1 p1
! nT ?
T4 = p4 nT
1

with 1 < nT <  for expansion (5.64)

T3 p3
The expressions for the shaft power, when ideal gases are assumed, are determined as
8 ! nV ? 9
 m_ R T > < p2 n V >
= 1

_WV = 1> ? 1 (5.65)

( ? 1) : p1 >
;
and

8 ! nT ? 9
 m_ R T >
< p4 n T >
= 1

_WT = 3> ? 1 (5.66)

( ? 1) : p3 >
;
5.2. MODELING AND EVALUATION 89

Using the definition of the, due to the process, reciprocal pressure ratios V 3 and T 3 from
compressor and turbine

V 3 = pp2 (5.67)
1
T 3 = pp 4
(5.68)
3
as well as the dimensionless quotients V 4 and T 4 from compressor and turbine

V 4 = nVn? 1 (5.69)
V
T 4 = nTn? 1 (5.70)
T
a form
W_ V =  T2 ? 1 (5.71)
m_ cp T1 T1
W_ T = T4 ? 1

(5.72)
m_ cp T3 T3
is yielded. This form corresponds to the known shaft work relations 5.51 and 5.59 up until
the further influence quantities. Thus, further refinement in modeling will succeed well. In
addition, four further dimensionless quantities given with

V 1 = (V 2 ? 1) = (VV3 ? 1) 4
and correspondingly (5.73)
T 1 = (T 2 ? 1) = (TT3 ? 1) 4
(5.74)
are incorporated in the modeling. In this way, suction losses in the compressor intake can
also be taken into account in the modeling. Thus additional hierarchy levels appear in the
description and evaluation graphs as shown in fig. 5.30. Such in-depth studies and simi-

G1 Gas Turbine

*H
 Y *H
 Y
 6HH  6HH
 H  H
V 2 B2 T 2 () V B T
@
?I @
? I 6 6
? @ ? @
V 3 V 4 T 3 T 4 polytropic polytropic

Figure 5.30: Three-step Evaluation Hierarchy

larity theoretical investigations have already been successfully carried out in various fields
of power engines [49, 56, 96]. The aggregation function corresponding to the evaluation
hierarchy in fig. 5.30 is obtained by using equations 5.73 and 5.74 and yields
 T ? ( V  )?1
G1 = 3 1 ? (V 3 )?B12 ? 1
T
4 4

(5.75)
B2
90 CHAPTER 5. EXAMPLES OF USAGE

It is important to note, that the view of the turbine expressed in equation 5.75 is consistent
with the view expressed in equation 5.46, if the data are taken from experiments. At the
same time, it cannot be assumed, that a pure tree formation for the evaluation hierarchy will
always arise, as already noted in section 3.1.1. (This however is what coincidentally results
in fig. 5.30 for example.) As a matter of fact, evaluation hierarchies can also occur in the
form of a graph [78], depending on the example.

5.3 Groups of Transformations

All dimensionally homogeneous physical equations can be made invariant to changes in

units of their basic system of dimensions. Furthermore, the determining equations of a
physical relation can be further invariant when undertaking additional transformations. This
application exceeds the known statement of the Pi theorem and has likewise been proven
to be versatile in many cases of physics and engineering applications. Transforming the
general N AVIER -S TOKES-equations in dimensionless form, for example, yields P RANDTL’s
equations for boundary layer flow, taking into consideration, that certain terms can be ne-
glected at high R EYNOLD numbers Re. For high M ACH numbers Ma, such a transformation
yields the equations for supersonic flow. In general such transformations can yield asymp-
totic solutions [36] valid for the particular problem in question.

Determination of the invariants, known as method of groups of transformations, requires,

however, an exact knowledge of all physical equations. Thus, a far deeper knowledge of
all relevant physical phenomena is needed than was necessary to draw up a complete rel-
evance list for the Pi theorem. Therefore, the method of groups of transformations will only
be briefly presented and illustrated using the equation for instationary heat transfer as an
example.

5.3.1 Heat Transfer

From the balance of a small volume element the differential equation of instationary heat
transfer [38, 97], when inner heat sources are not present, yields

 c @# = @  @#  + @  @#  + @  @# 
@t @x @x @y @y @z @z
= div ( grad #) (5.76)
5.3. GROUPS OF TRANSFORMATIONS 91

Assuming isotropic material properties ( independent of location, direction and tempera-

ture), this can be described as
@# = a  @ 2 # + @ 2 # + @ 2 # 
@t @x2 @y2 @z2
= a
# (5.77)
by means of the definition of the temperature guide number a = = c. To represent the
above mentioned, e.g. by means of the method of groups of transformations, limitation to
the one-dimensional case suffices. When the boundary conditions for the following example
are assumed, in which a warmer body cools down in a cooler medium by instationary heat
transfer, the sought after solution function of the DE results as a relation

f (X; #c; ; a; x; #; t; ) = 0 (5.78)

of 8 physical quantities. In this, X signifies a characteristic length, #c a characteristic tem-
perature,  thermal conductivity, a the temperature guide number, x local coordinates, # the

[M] [L] [T] [] SI units

X 1 m
#c 1 K
 1 1 -3 -1 W=m K
a 2 -1 m2 =s
x 1 m
# 1 K
t 1 s
1 -3 -1 W=m2 K
Figure 5.31: Dimensional Matrix of Instationary Heat Transfer

temperature, t the time and the coefficient of heat transfer. From the above dimensional
matrix in fig. 5.31, the following four dimensionless exponential products can be obtained
due to m = n ? r = 4.

1 = Xx   (5.79)
2 = ##   (5.80)
c
3 = Xa t2    Fo (5.81)
4 = X  Bi (5.82)
The commonly used nomenclature is given here:  is the dimensionless local coordinate 1 ,
 is the dimensionless temperature variable 2 ,  or the Fourier number Fo is the dimension-
less time 3 and the Biot number Bi for the dimensionless product 4 . These are also used in
92 CHAPTER 5. EXAMPLES OF USAGE

the following equations. Through transformation of the one dimensional differential equation

@# = a @ 2 # (5.83)
@t @x2
onto the new coordinates ;  and  , substitution of the partial derivatives @# = #c @, @x =
X @ and @t = (X 2=a) @ and algebraic conversions of the differential equation representing
one dimensional instationary heat transfer into dimensionless form yields

@ = @ 2  (5.84)
@ @ 2
In this way the context of a solution function corresponding to equation 5.78 is reduced to a
solution function of only four variables to

g(; Fo; Bi; ) = 0 (5.85)

For some geometrically simple, half-infinite bodies and in order to justify one-dimensional
heat transfer (e.g. plate, cylinder and sphere) the solutions from equation 5.85 can even
be given analytically enclosed and represented in a G ROEBER-diagram [1]. This is shown
in fig. 5.32 in the case of the planar wall of thickness 2X for the temporal decrease in the
central temperature  = #m =#c of the wall. The temperature gradient which, for example,

1:0 ........................................................................................................................................................................
............................................................
............ .............
............................
........................
...................
......... ............ ..................
....... .......... ..............
....... .......... ............
...... ........ ............
...... ........ ..........

#m=#c
..... ....... ..........
.

Bi = 0:1
..... ....... .........
..... ....... ........
..... ....... ........
..... ...... .......
..... ...... ........
..... ....... .......
..... ...... .......
.... ...... .......
.... . ...... ......
.... ..... ......

Bi = 1:0
.... ...... ......
......

0:5
..... ..... ......
.... ..... ......
..... ...... ......
..... ..... ......
..... ...... ......
..... ...... .....

Bi ! 1
..... ..... ......
..... .......
..... ......
..... ......
..... . .......
..... ......
..... .......
..... .......
...... .......
..... .......
...... .......
...... ........
...... ........
...... .........
....... .........
........ ..........
........ ...........
.............

0:0
.......... ..............
........... ....................
................. ..
.....................................

0:1 1:0 Fo 10:0

Figure 5.32: G R ÖBER-Diagram of Half-Infinite Planar Wall [1]

is naturally also experimentally determinable for a particular wall, is, due to equation 5.85,
applicable not only for this particular wall but also for all other walls whose quantities form
identical parameters and therefore can also be produced from affine transformations (3 and
4 ), besides the required thermal and geometric similarity (1 und 2 ).
The results produced with the method of group transformations according to equation 5.85,
will be applied in the following during the ascertainment of a partial evaluation which is
comparing the thermal properties of various body forms in the next and last example.
5.3. GROUPS OF TRANSFORMATIONS 93

5.3.2 Coffee Pots

In this example it is to be exemplarily shown, how, according to the present state of know-
ledge and whilst lacking a common feature system (i.e. no common unit system), physically
justified evaluations and, for example, evaluation from the fields of aesthetics and economics
are aggregated together into a holistic evaluation using the AHP process.

Criteria. The three most important evaluation criteria in the chosen example comparing
coffee pots are aesthetics, the insulation capability and the price of the various design ob-
jects. Therefore the evaluation hierarchy yields the forms as shown in fig. 5.33, which could
still be extended or refined if needed. The aggregational weightings are firstly undefined

Quality
:X

 y
X
 6 XXXX
  XXX
 X
Heat
Aesthetics Price
Insulation

Figure 5.33: Evaluation Hierarchy of Coffee Pots

for all three evaluation criteria in the absence of a superordinate feature class system and
has therefore be determined by the human decision makers themselves. This aggrega-
tional weighting is assessed in the framework of this example according to the linguistic
scale shown in fig. 2.5 as “much more important” (9) for the aestheticism, as “somewhat
more important” (6) for the insulation capability and as “normal important” (1) for the price.
Therefore only the partial evaluation of the sub-criteria still need to be determined in order
to form a complete evaluation. For this the description and terminology concepts used in the
respective branches, is made use of.

Aesthetics. Since ancient times, pains have been taken to identify the reasons for beauty.
The philosophical concepts such as form, order, symmetry and proportion as the pure and
typical embodiment of beauty were common property in greek philosophy, thus P LATON and
A RISTOTLE endeavoured to quantify more exactly the relation of the whole to its parts in
geometric figures, rhythms or musical intervals in the form of numeric relationships. Since
then, certain numeric relationships were allotted a particularly aesthetic value, the golden
section [66].

Building on this thought, B IRKHOFF (1928) defined an aesthetic measurement M for certain
simple classes of objects [8] (e.g. vases) with

M = O
C (5.86)
94 CHAPTER 5. EXAMPLES OF USAGE

where O is the degree of order in the object and C the complexity of the object. No further
investigations will be made into the themes of and the exact determining of the aesthetic
measure for the coffee pots under consideration, because this is extensively documented
in [33]. Of importance is firstly only the decisive condition that O and C alone are determined
by the forming of rules of measurement specifications and are therefore only dependent
upon the object and not upon the observer. This property necessary for an evaluation has
already been fully founded and discussed in section 4.4.3 in epistemological reflections.

Corresponding to the further development and refining of the aesthetic measurement M ,

through M ASER [61] and G ARNICH [33], the measurement M formed according to equa-
tion 5.86 is used in fig. 5.34 to aesthetically evaluate the coffee pots, although in the past the

Design in M
ABC 1.11
Berlin 0.95
Composition 1.20
Lotus 1.30
Rheinland 0.87
Variation 1.10

Figure 5.34: Designs and Aesthetic Measurements (according to G ARNICH [33])

otherwise very frequent critical assessments of the underlying aesthetic information theory
have let the broader scientific interest [95] in this field slacken.

The coffee pots under consideration are illustrated in fig. 5.38 to 5.43 for the benefit of
personal formation of opinion. Instead of objects existing as photographs, the design objects
could also be modeled using the solid-modeller in the programme pack I-DEAS [53] and by
means of a ray tracing process, virtual photo-realistic pictures could be produced.

Heat Insulation. To evaluate the different insulation capabilities of the different forms of
coffee pots, the dimensionless time  has to be determined analogically to fig. 5.32. The
dimensionless time is defined according to the preselected mid temperature ratio as  =
0:5 and prevailing boundary conditions. In the case of existing design objects, this can
be done by experimentation or, during earlier stages of the design process, by means of
numerical simulation. In this specific case the real design objects were available and thus
the results shown in fig. 5.35 obtained by conducting an experimentation. Alternatively, a
finite difference method can be used to determine the necessary physical quantities. This
could be done unproblematically using an explicit difference scheme, known for the one-
dimensional case as B INDER -S CHMITT -method. Choosing the time- and spatial-steps with
5.3. GROUPS OF TRANSFORMATIONS 95

Design in Fo
ABC 0.0834
Berlin 0.1052
Composition 0.0725
Lotus 0.1154
Rheinland 0.0754
Variation 0.0810

Figure 5.35: Designs and Insulation (Dimensionless Time Fo)

the module p as
a
t  1
p = (
(5.87)
x)2 2
the numerical stability of this method can be proven in general.

Price. The exact formation of the cost calculation is difficult as the in-house knowledge
necessary for this is not available. In order to construct the price table in fig. 5.36, the oral

Design in DM
ABC 81.30
Berlin 78.70
Composition 90.00
Lotus 76.90
Rheinland 96.00
Variation 90.00

Figure 5.36: Price Table of the Designs (According to [73])

statements of the press department of the firm ROSENTHAL [73] are therefore resorted to.
The designs of coffee pots [34] considered in this example were originally made for this firm
and formed part of the delivery programme even up to the year 1988 (DM 1:6  US-$ 1).

Complete Evaluation. With the partial evaluations of the single criteria and the prescribed
aggregation weighting, the following complete evaluation result for the design object is
yielded in fig. 5.37. The limitations which have already been mentioned in section 2.3.1
are important for a meaningful interpretation of the evaluation result, obtained by means of
the AHP process.

Due to the absolute dominance of the design object Lotus, a good-natured decision situa-
tion [71] is at hand which has no need to form a compromise by means of a compensatory
96 CHAPTER 5. EXAMPLES OF USAGE

Complete Evaluation 0.5625

0.3750 Weighting
Aesthetics Insulation Price 0.0625
0.1700 0.1565 0.1746 0.1652 ABC
0.1455 0.1974 0.1797 0.1671 Berlin
0.1838 0.1360 0.1572 0.1642 Composition
0.1991 0.2166 0.1839 0.2047 Lotus
0.1332 0.1415 0.1473 0.1372 Rheinland
0.1685 0.1520 0.1572 0.1616 Variation
0.0000 0.0000 0.0000 0.0000 Consistency Ratio

Figure 5.37: Complete Evaluation of the Designs

model. The creation of such a dominant design object variant is, however, definitely not the
rule in complex decision situations and the stability of the decision behaviour can therefore
be investigated by means of comprehensive sensitivity analysis.

At the same time, a main objective of the AHP process should be to make the decision finding
more transparent by the obligation to name explicitly the partial criteria, the partial evaluation
and the aggregation weighting. The formalised scheme on hand should therefore be used
until the evaluation result, retained in this way, corresponds with the intuitively obtained
complete evaluation result by the decision maker and remains stable in the framework of a
sensitivity analysis. In this sense, a final evaluation result of a complex evaluation hierarchy
which is computed by means of the AHP process should never be imposed upon the decision
maker. In contrary, it should serve the singular purpose of facilitating the systematic and
critical self-reflection of the decision maker.

As an example, the evaluation model which has been consciously chosen for the partial
area of aesthetics illustrates additionally by means of the aesthetic measurement M the
permanent dilemma present in any evaluation situation, which is the need to put trust into
judgements and suggested compromises in the form of aggregation weights suggested by
“experts”. This has even more drastic consequences when taking into consideration that the
justification of the respective model formations based on boundary conditions and simplify-
ing assumptions found in the special disciplines are not known by and therefore cannot be
judged correctly by the human decision maker.
5.3. GROUPS OF TRANSFORMATIONS 97

Figure 5.38: Design ABC

Figure 5.39: Design Berlin

98 CHAPTER 5. EXAMPLES OF USAGE

Figure 5.40: Design Composition

Figure 5.41: Design Lotus

5.3. GROUPS OF TRANSFORMATIONS 99

Figure 5.42: Design Rheinland

Figure 5.43: Design Variation

100 CHAPTER 5. EXAMPLES OF USAGE
 101

6 Summary

As has been shown in the previous sections, evaluation problems deal mainly with three core
questions. Apart from the problem of structure and completeness of the partial evaluation
criteria used, these are the problem of finding appropriate units of measurement for each
single partial evaluation component, as well as the problem of aggregating them to a com-
plete evaluation. In classic decision making procedures, such as the Analytical-Hierarchical
Process, single judgements are determined by measurements, estimation or consideration
of expert judgements. By means of weighting factors, prescribed by general consensus,
these are aggregated to a complete evaluation.

In contrast, the suggested evaluation method, based upon dimensional analysis, uses di-
mensionless quantities as units of measurement. These quantities, as opposed to dimen-
sional quantities, possess the property of invariance and therefore represent an advance
towards objectivity. The insight into the evaluation problem obtained using this method of
evaluation confirms therefore the principal procedure of the Analytical-Hierarchical Process,
which hierarchically structures the complete evaluation into partial criteria. Beyond this,
the suggested method of evaluation can also be integrated into the Analytical-Hierarchical
Process and, being a formally consistent evaluation model in the field of precise physical
knowledge, has the advantage of being applicable to the exact physical model. Therefore, it
offers further foundation of the evaluation methodology.

The postulated evaluation hypothesis (“Any minimal description in the sense of the Pi the-
orem is an evaluation”) is fundamental to the application of dimensional analysis upon an
evaluation problem. This hypothesis has been epistemologically founded and has, in the
framework of this piece of work, been confirmed by its successful application to several
cases. Further results concerning this subject have been categorised in the following sec-
tion.

6.1 Results

The main consequence of the evaluation hypothesis is the unexpected result that the so-
lution of the evaluation problem can basically be reduced to the solution of the description
problem. Thus, the evaluation problem can only then be solved, when a complete description
is known.

Therefore, the further consequence of the evaluation problem is that the form as well as the
102 CHAPTER 6. SUMMARY

content and the interpretation of the evaluation graph are a result of a mapping of the de-
scription graph. As well as this, due to the consistency of the basic physical model formation,
not only the units of measurement pertaining to the single evaluations (the criteria), but also
the aggregation functions (the weighting) of the single criteria are defined in relation to one
another and are consistent throughout all hierarchic levels.

Thus, linkage effects, which are not clearly visible in advance, due to the obvious relations
within the graph structure of the dimensionless products can be revealed and partially qual-
ified during the design process. The sensitivity analyses, important during the formation of
variants in the design process, can be derived from the differential formulation of the model
laws. Due to the notation used, the result represents an interesting cross reference to the
the technique of “qualitative reasoning” in artificial intelligence.

Even though the provable, correct relevance list of parameters used in dimensional analy-
sis is restricted to areas of precise physical knowledge, this method provides a set of aids
also for areas of imprecise physical knowledge. Its application as particular mathemati-
cally necessary condition is unable to lead to formal contradictions. Therefore, dimensional
analysis, e.g. as opposed to classical expert systems, proves to be a versatile tool even in
weakly structured areas of knowledge. Several indications about the possible developmental
directions of successful future applications can be found in the next section.

6.2 Outlook

In dimensional analysis, the function F (1 ; : : : ; m ) = 0 must be experimentally determined

when mapping the description space X onto the evaluation space . Within dimensional
analysis, it is therefore feasible to unite advantageously the techniques of “rule-based” and
“case-based” reasoning, both of which have until now been regarded as individual tech-
niques in the field of artificial intelligence.

In this way, the construction of a “Symbolic Front-End for a CAD-System [79]” would be feasi-
ble by means of the dimensional analysis method. In this software model, the entire symbolic
representation of the physical conditions and processes concerned with the design model
can occur. These conditions and processes can be mapped onto the geometric and material
quantities of the solid modeler by means of associative links. The transferral of the concept
of finite elements onto the reusage of premodeled partial objects and construction groups
might be successfully attempted. Together with the associative links of the solid modeler,
already realised, which is concerned with the end applications, based upon the product data
model, the realisation of a flexible and holistic design system is conceivable. Additionally,
analytical and experimental results can be converted over to all topologies which stem from
6.2. OUTLOOK 103

known topologies by means of certain mappings. This is possible because a physically

founded similarity measure is now available through the dimensionless parameters.

Furthermore, it is possible to supportively erect and justify real formations of analogies and
associations by means of the similar structure of two evaluation graphs as well as to develop
possible decomposition strategies for optimisation due to the known coupling matrix of the
design quantities. This could be proved one day to be a considerable advance in the support
of the designer during complex design assignments in the future.
104 CHAPTER 6. SUMMARY
 105

A Proofs

Here both proofs of the Product and the Pi theorems are presented and explained. The
material for the proof of the Product theorem is based mainly on the work of B RIDGMAN [14],
while the material for the proof of the Pi theorem is due to the work of G ÖRTLER [35] and
PAWLOWSKY [68]. Other methods and further comments on the mathematical background
and history of both proofs, as well as further references on the subject, can be found in the
excellent book by G ÖRTLER [35].

A.1 Product Theorem

Theorem. If the invariance of ratios of physical quantities under scaling of the units of mea-
surement is assumed, the relation between primary physical quantities x1 ; x2 ; : : : ; xn " IR+
and secondary physical quantities  = f (x1 ; : : : ; xn) only exists through the functional rela-
tionships of the following differentiable form:

f = Cx1 1 x2 2


xn n (A.1)
Whereby for fixed n > 1 is n " IN and the C; 1 ; 2 ; : : : ; n " IR are real constants.
Idea. The fact that ratios of physical quantities are invariant under scale transforms of physi-
cal units is used to set up an equation for a general function expression relating primary and
secondary physical units. The unique solution to this equation is then proved to be exactly
of the form to be shown.

Proof (see B RIDGMAN [14]). Physical quantities can be divided into two classes of primary
and secondary quantities. Primary physical quantities are of fundamental and of irreducible
simplicity and can therefore be expressed using one single unit of the employed unit system.
Secondary physical quantities are derived from primary quantities by some function f and
can only be expressed using combinations of units of the employed unit system. In this
respect, time (measured in s) and length (in m) are primary quantities, while velocity (in
m=s) is a secondary quantity. The two values of measurement xi and yi of a primary physical
quantity can be altered by a scale transform (with mi > 0)
x0i = mi xi (A.2)
yi0 = mi yi (A.3)
of the appropriate reference unit, e.g. by switching from meter to inch as reference unit or
vice versa. While the two values of measurement x0i and xi of a primary physical quantity
106 APPENDIX A. PROOFS

depend on such a scale transform, the ratio of the two values of measurement of the primary
physical quantity
x0i = xi (A.4)
yi0 yi
to both primary and secondary physical quantities and is called the absolute significance of
relative magnitude (see B RIDGMAN [14]). Secondary physical quantities are derived from
primary physical quantities using a very general functional relationship f of the form

f (x1; x2 ; : : : ; xn): (A.5)

Assuming the absolute significance of relative magnitude, it follows that the following ratio is
invariant under scale transforms
f (m1x1 ; m2 x2; : : : ; mnxn ) = f (x1 ; x2; : : : ; xn) (A.6)
f (m1y1; m2y2; : : : ; mnyn) f (y1; y2; : : : ; yn)
where xi ; x0i ; yi and yi0 are values of measurement of primary quantities according to equa-
tion (A.2) and (A.3). Equation (A.6) may also be written as

f (x01; x02 ; : : : ; x0n) = f (y10 ; y20 ; : : : ; x0n) ff((xy1;; xy2;; :: :: :: ;; yxn)) (A.7)
1 2 n
Partial differentiation, with respect to m1 , leads to
0 0 0 0 0 0
x1 @f (x1 ; x@x2 ;0: : : ; xn ) = y1 @f (y1 ; y@y2;0: : : ; xn ) ff((xy1 ;; xy2;; :: :: :: ;; yxn)) (A.8)
1 1 1 2 n
Setting all m1 = m2 = :::  1 in equation (A.8) gives
x1 @f (x1 ; x2 ; : : : ; xn ) = y1 @f (y1; y2; : : : ; yn) (A.9)
f (x1; x2 ; : : : ; xn) @x1 f (y1; y2; : : : ; yn) @y1
Holding the right hand side of equation (A.9) constant and varying only the x1 on the left
hand side, one obtains
x1 @f (x1 ; x2 ; : : : ; xn ) = const (A.10)
f (x1; x2 ; : : : ; xn) @x1
This can also be written as
x1 @f = const (A.11)
f @x1
The only solution to this equation (const 6= 0; const 6= ?1) is
f = C1 x1 const (A.12)
where the constant C1 is a function of the remaining x2 ; x3 ; : : : ; xn , which were held constant.
Repeating the process from equation (A.8) up to equation (A.12) for the remaining xi thus
yields the desired form of the function

f = Cx1 1 x2 2


xn n (A.13)
with n > 1; n 2 IN and C; 1 ; 2 ; : : : ; n 2 IR.
A.2. PI THEOREM (ALGEBRAIC) 107

A.2 Pi Theorem (algebraic)

Theorem. From the existence of a complete and dimensionally homogeneous functional re-
lationship f of exactly n physical quantities xi " IR+ , the existence of a functional relationship
F of only m < n dimensionless parameters j (also called similarity variables or dimension-
less products) results with

f (x1 ; :::; xn) = 0

F (1; :::; m ) = 0 (A.14)

whereby m = n ? r is reduced by the rank r of a dimensional matrix formed by means of the

n quantities. The dimensionless products take the form
Yr
j = xj x?i ji (A.15)
i=1
whereby for fixed j is j = 1; : : : ; m " IN and the j 1 ; j 2; : : : ; jr " IR are real constants.

Idea. The exponents of the unit representation of physical quantities form a homogeneous
linear equation system. Starting with a relationship of n physical quantities, the properties of
rank preservation or rank reduction of a set of homogeneous linear equation system can be
used to derive a relationship between only m dimensionless quantities.

Proof (see PAWLOWSKY [68]). From the exponents of the unit representation of n physical
quantities xi of a complete, dimensionally homogeneous equation in the form

f (x1; : : : ; xn) = 0 (A.16)

a so called dimensional matrix can be constructed. This is shown on the left of figure A.1.
The dimensional matrix has n rows for the xi and up to k  7 columns, which represent the
m1m2 : : : : : mk m1m2 : : : : : mr
x1 x
66 1 1
x2 x2 1
:: : ::
:: r ::: ::
:: aij n :: ::
:: x
? r 1
:: 6x1:
:: m :: ji
xn ??
x m

Figure A.1: Definition of dimensional matrix

108 APPENDIX A. PROOFS

base units of the unit system1 employed. Then the dimensional matrix is brought by rank
preserving equivalence transforms into diagonal form, as shown on the right of figure A.1.

According to equation (A.16), the set Xn = fx1 ; : : : ; xng of physical quantities of a complete,
dimensionally homogeneous equation

f (Xn) = 0 n2 (A.17)

can be separated into the two subsets of the i-set Xi = fx1 ; : : : ; xr g, with quantities xi which
are linearly independent of each other, and the j -set Xj = fx1 ; : : : ; xm g with quantities xj
which are linearly dependent on the i-set.

f (Xi; Xj ) = 0 i = 1; : : : ; r (A.18)
j = 1; : : : ; m
In this respect, the i-set is represented by the upper half of the dimensional matrix in fig-
ure A.1, while the j -set is represented by the lower half of the dimensional matrix in fig-
ure A.1. Both subsets contain at least one element each, since the property of dimensional
homogeneity of any complete physical equation guarantees that at least one quantity has to
be expressed in terms of the dimensions of the other quantities.

If the units of all primary quantities xi 2 Xi are now subject to a scale transform as shown in
equation (A.2), the value of measurement of each primary quantity changes from xi to mi xi ,
while all secondary units xj 2 Xj are scaled according to the derived general form of any
secondary physical quantity, as proved by the Product theorem. This leads to an expression
of the form
Yr i = 1; : : : ; r
f (mixi ; xj mi ji ) = 0 (A.19)
i=1 j = 1; : : : ; m
If the scaling factors mi are chosen in such a way that mi xi = 1 holds, the use of mi = 1=xi
leads directly to
Yr
F (1; xj x?i ji ) = 0 and thus (A.20)
i=1
F (1; : : : ; m ) = 0 (A.21)
where j represents a dimensionless variable of the form
Yr
j = xj x?i ji
j = 1; : : : ; m (A.22)
i=1
with m = n ? r.
1 At present, seven different base units are known and defined in the SI-Unit-System (symbol and dimension)

as: mass (M in kg), length (L in m), time (T in s), thermodynamic temperature ( in K), electric current (I in A),
amount of substance ( in mol) and luminous intensity (Iv in cd).
A.3. FUNDAMENTAL SYSTEMS 109

A.3 Fundamental Systems

From a dimensional matrix A , formed by n physical quantities, arbitrary exponential products

with and without dimension of the form

 = x1 x2


xnn
1 2
(A.23)
can be obtained. These exponential products chosen are only then dimensionless when, in
their dimensional representation in the physical units m1 to mr ,

[] = m01 m02


m0r = 1 (A.24)

applies. The condition A.24 can also be formulated as a homogeneous linear equation
system (LES) of the exponents in the dimensional representation of the dimensional matrix
A as
( 1; 2; : : : ; n) A = 0 (A.25)
Thus, the question of the existence and the variety of solution in dimensionless products [35]
is traced back to the question of the existence, solvability and solution variety of a homoge-
neous LES. Because homogeneous LES are always solvable [11], the existence of a dimen-
sionless exponential product is always assured. As well as this, homogeneous LES with
rg(A ) = r always form a (n ? r)-parametrised set of solutions, so that a fundamental system
with m = n ? r different dimensionless exponential products

j = x1 j x2 j


xnjn
1 2
j = 1; : : : ; m (A.26)
always ensues. Furthermore, also those ( 1 ; 2 ; : : : ; n ) are possible solutions correspond-
ing to equation A.23, which are represented as linear combinations of the m solutions of the
fundamental system by
X
m
( 1; 2; : : : ; n) = j ( j 1; j 2 ; : : : ; jn) with j " IR (A.27)
j =1
with suitably chosen j . The choice and suitable selection of a particular fundamental sys-
tem is therefore free. Fundamental systems are often thus chosen so that the known, well-
established dimensionless parameters result as parts of a fundamental solution. One exam-
ple derived from such a fundamental system is shown in section 5.3.1, Heat Transfer.

A.4 Dimensionless Coefficients

If the form
Yr
j = xj x?i ji
(A.28)
i=1
110 APPENDIX A. PROOFS

is retained by means of the represented dimensional considerations on physical processes

with the dimensionless exponential products of the Pi theorem, special forms of some j
are known as dimensionless parameters under their own names, such as is the case for the
R EYNOLD’s number Re = vl ?1 . In the general case of any interesting quantity y  xj and
cy  j with
cy = Qr y ji (A.29)
xi
i=1
is also described as dimensionless y -coefficient in the literature, as is the case for the resis-
tance coefficient cw = W=qA in aerodynamics.
 111

B Evaluations

In this part of the appendix, the separate partial evaluation results taken from the numerical
examples of the AHP process and the method of Outranking Relations have been separately
listed for purposes of clarification. The corresponding requirements and the representation
of the respective complete evaluation results, as well the discussion about them have been,
however, excluded.

B.1 AHP Evaluations

For the example of the job search [37], the evaluation matrices, eigenvectors and consis-
tency ratios represented here result for the single leaves of the partial criteria from fig. B.1
to B.7. These evaluation judgements for the partial goals, distance and living expenses
are based upon exact data extracted from maps and statistical yearbooks. The other eval-
uations, including the criteria tree and the aggregation weightings, reflect the subjective
viewpoint of the decision maker [37].

Distance (2) max 4.0000

Bos LA St.L Hou c:r: 0.0000
Bos 1.0000 9.1419 2.9324 5.0946 x 0.6073
LA 0.1094 1.0000 0.3208 0.5573 0.0664
St.L 0.3410 3.1175 1.0000 1.7373 0.2071
Hou 0.1963 1.7944 0.5756 1.0000 0.1192

Figure B.1: Evaluation of Alternatives Distance

Living Cost (3) max 4.0000

Bos LA St.L Hou c:r: 0.0000
Bos 1.0000 3.1738 0.7184 2.3039 x 0.3184
LA 0.3151 1.0000 0.2264 0.7259 0.1003
St.L 1.3920 4.4180 1.0000 3.2070 0.4432
Hou 0.4340 1.3776 0.3118 1.0000 0.1382

Figure B.2: Evaluation of Alternatives Living Cost

112 APPENDIX B. EVALUATIONS

Climate (4) max 4.1046

Bos LA St.L Hou c:r: 0.0388
Bos 1.0000 0.3333 2.0000 5.0000 x 0.2591
LA 3.0000 1.0000 4.0000 5.0000 0.5367
St.L 0.5000 0.2500 1.0000 2.0000 0.1316
Hou 0.2000 0.2000 0.5000 1.0000 0.0725

Figure B.3: Evaluation of Alternatives Climate

High School (7) max 4.0000

Bos LA St.L Hou c:r: 0.0000
Bos 1.0000 5.0000 1.0000 4.0000 x 0.4082
LA 0.2000 1.0000 0.2000 0.8000 0.0816
St.L 1.0000 5.0000 1.0000 4.0000 0.4082
Hou 0.2500 1.2500 0.2500 1.0000 0.1020

Figure B.4: Evaluation of Alternatives High School

College (8) max 4.0247

Bos LA St.L Hou c:r: 0.0091
Bos 1.0000 2.0000 5.0000 6.0000 x 0.5383
LA 0.5000 1.0000 2.0000 3.0000 0.2531
St.L 0.2000 0.5000 1.0000 2.0000 0.1297
Hou 0.1667 0.3333 0.5000 1.0000 0.0790

Figure B.5: Evaluation of Alternatives College

Traffic (9) max 4.0406

Bos LA St.L Hou c:r: 0.0150
Bos 1.0000 1.0000 0.1429 0.1667 x 0.0634
LA 1.0000 1.0000 0.1250 0.1429 0.0590
St.L 7.0000 8.0000 1.0000 2.0000 0.5298
Hou 6.0000 7.0000 0.5000 1.0000 0.3479

Figure B.6: Evaluation of Alternatives Traffic

Culture (10) max 4.0605

Bos LA St.L Hou c:r: 0.0224
Bos 1.0000 0.5000 4.0000 5.0000 x 0.3241
LA 2.0000 1.0000 5.0000 6.0000 0.5078
St.L 0.2500 0.2000 1.0000 2.0000 0.1025
Hou 0.2000 0.1667 0.5000 1.0000 0.0656

Figure B.7: Evaluation of Alternatives Cultural Selection

B.2. OUTRANKING RELATIONS EVALUATIONS 113

For the weighting of the partial goals amongst one another, the decision maker [83] gave
the following matrices, eigenvectors and consistency ratios from fig. B.8 to fig. B.10 for the
separate inner nodes.

Education (5) max 2.0000

School College c:r: 0.0000
School 1.0000 0.6667 x 0.4000
College 1.5000 1.0000 0.6000

Figure B.8: Criteria Weighting Education

Quality of Life (6) max 2.0000

Traffic Culture c:r: 0.0000
Traffic 1.0000 0.4286 x 0.3000
Culture 2.3333 1.0000 0.7000

Figure B.9: Criteria Weighting Quality of Life

City (1) max 5.1813

(2) (3) (4) (5) (6) c:r: 0.0405
(2) 1.0000 2.0000 0.5000 2.0000 0.3333 x 0.1570
(3) 0.5000 1.0000 0.2500 2.0000 0.3333 0.1067
(4) 2.0000 4.0000 1.0000 2.0000 1.0000 0.3021
(5) 0.5000 0.5000 0.5000 1.0000 0.2500 0.0883
(6) 3.0000 3.0000 1.0000 4.0000 1.0000 0.3459

Figure B.10: Criteria Weighting City

B.2 Outranking Relations Evaluations

Using the table from fig. 2.13, the following illustrations B.11 to B.16 with the partial concor-
dance and discordance of the partial goals result, with application of equation 2.52 and 2.53.
114 APPENDIX B. EVALUATIONS

Price
Concordance A B C D E F G
A 0.000 0.000 0.000 0.000 0.000 0.000 0.000
B 1.000 0.000 0.000 0.000 0.000 0.000 0.000
C 1.000 0.000 0.000 0.000 0.000 0.000 0.000
D 1.000 1.000 1.000 0.000 0.000 0.000 0.000
E 1.000 1.000 1.000 0.000 0.000 0.000 0.000
F 1.000 1.000 1.000 0.500 0.000 0.000 0.000
G 1.000 1.000 1.000 1.000 1.000 1.000 0.000

Figure B.11: Partial Concordance Numbers Price

Comfort
Concordance A B C D E F G
A 0.000 0.000 1.000 1.000 1.000 1.000 1.000
B 0.000 0.000 1.000 1.000 1.000 1.000 1.000
C 0.000 0.000 0.000 1.000 1.000 1.000 1.000
D 0.000 0.000 0.000 0.000 1.000 1.000 1.000
E 0.000 0.000 0.000 0.000 0.000 1.000 0.000
F 0.000 0.000 0.000 0.000 0.000 0.000 0.000
G 0.000 0.000 0.000 0.000 0.000 1.000 0.000

Figure B.12: Partial Concordance Numbers Comfort

Speed
Concordance A B C D E F G
A 0.000 1.000 0.000 1.000 1.000 0.000 1.000
B 0.000 0.000 0.000 0.000 0.000 0.000 0.000
C 0.000 1.000 0.000 0.000 1.000 0.000 1.000
D 0.000 1.000 0.000 0.000 1.000 0.000 1.000
E 0.000 0.000 0.000 0.000 0.000 0.000 0.000
F 0.000 1.000 0.000 0.000 1.000 0.000 1.000
G 0.000 0.000 0.000 0.000 1.000 0.000 0.000

Figure B.13: Partial Concordance Numbers Speed

B.2. OUTRANKING RELATIONS EVALUATIONS 115

Price
Discordance A B C D E F G
A 0.000 0.200 0.333 0.867 1.000 1.000 1.000
B 0.000 0.000 0.000 0.333 0.467 0.600 1.000
C 0.000 0.000 0.000 0.200 0.333 0.467 1.000
D 0.000 0.000 0.000 0.000 0.000 0.000 0.867
E 0.000 0.000 0.000 0.000 0.000 0.000 0.733
F 0.000 0.000 0.000 0.000 0.000 0.000 0.600
G 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Figure B.14: Partial Discordance Numbers Price

Comfort
Discordance A B C D E F G
A 0.000 0.000 0.000 0.000 0.000 0.000 0.000
B 0.000 0.000 0.000 0.000 0.000 0.000 0.000
C 0.400 0.400 0.000 0.000 0.000 0.000 0.000
D 0.800 0.800 0.400 0.000 0.000 0.000 0.000
E 1.000 1.000 0.800 0.400 0.000 0.000 0.000
F 1.000 1.000 1.000 0.800 0.400 0.000 0.400
G 1.000 1.000 0.800 0.400 0.000 0.000 0.000

Figure B.15: Partial Discordance Numbers Comfort

Speed
Discordance A B C D E F G
A 0.000 0.000 0.000 0.000 0.000 0.000 0.000
B 1.000 0.000 0.750 0.500 0.000 0.750 0.000
C 0.000 0.000 0.000 0.000 0.000 0.000 0.000
D 0.000 0.000 0.000 0.000 0.000 0.000 0.000
E 1.000 0.000 1.000 0.750 0.000 1.000 0.000
F 0.000 0.000 0.000 0.000 0.000 0.000 0.000
G 0.750 0.000 0.500 0.250 0.000 0.500 0.000

Figure B.16: Partial Discordance Numbers Speed

116 APPENDIX B. EVALUATIONS
Bibliography 117

Bibliography

[1] Algermissen, J.: Grundlagen des Wärmeaustauschs. Lecture Notes, Stuttgart Univer-
sity, 1985.

[2] Argyris, J.; Mlejnek, H.: Die Methode der finiten Elemente. Band 1. Vieweg, Braun-
schweig, 1986.

[3] Bauer, F.: 400 Jahre moderne Algebra. Informatik Spektrum 14, 6 (1991), 343-344.

[4] Becker,, T.; Caesar, C.; Schuh, G.: Beherrschung der Variantenvielfalt — Regelkreis
zwischen Produktgestaltung und Produktbewertung. Konstruktion 41, (1989), 239-242.

[5] Belton, V.: A comparison of the analytic hierarchy process and a simple multi-attribute
value function. European Journal of Operational Research 26 (1986), 7-21.

[6] Belton, V.; Gear, T.: On a Short-coming of Saaty’s Method of Analytic Hierarchies.
Omega (1984), Vol. 11, No. 3, 228-230.

[7] Bhaskar, R.; Nigam, A.: Qualitative Physics Using Dimensional Analysis. Artificial Intel-
ligence, 45, 73-111, 1990.

[8] Birkhoff, D.: Aesthetic measure. Cambridge Press, Massachusetts, 1933.

[9] Birkhoff, G.: Hydrodynamics, a Study in Logic, Fact and Similitude. Princeton University
Press, Princeton, 1950.

[10] Braun, K.: Einführung in die Matrizenrechnung. Lecture Notes, Stuttgart University,
1985.

[11] Brenner, J; Lesky, P.: Mathematik für Ingenieure und Naturwissenschaftler, Band I.
Akademische Verlagsgesellschaft, Wiesbaden, 1978.

[12] Buckingham, E.: On Physically Similar Systems; Illustration of the Use of Dimensional
Equations. Phys. Review 4, (1914), 345-376.

[13] Bunge, M.: Causality. The place of the causal principle in modern science. Harvard
University Press, Cambridge, Massachusetts, 1957.

[14] Bridgman, P.: Dimensional Analysis. Yale University Press, New Haven, 1922.

[15] Bridgman, P.: Theorie der physikalischen Dimensionen. Teubner, Leipzig, 1932. Trans-
lation of [14] into german.
118 Bibliography

[16] Dolinskii, I.: Use of dimensional analysis in the construction of mechanical assemblies
for optical instruments. Soviet Journal of Optical Technology, 57, 8 (1990), 512-514.

[17] Doster, A.: Expertensystemunterstützte Benutzungsoberfläche für Tragwerksberech-

nungen auf objektorientierter Basis. PhD Thesis, Technical University of Munich, 1990.

[18] Düttmann, M; Schneider, F. (Hrsg.): Morris Lapidus: der Architekt des Amerikanischen
Traums. Birkhäuser, Basel, 1992.

[19] Dumitrescu, D.: Strömung an einer Luftblase im senkrechten Rohr. Zeitschrift für Ange-
wandte Mathematik und Mechanik, 23, (1943), 139-149.

[20] Dyer, J.: Remarks on the Analytic Hierarchy Process. Management Science (1990),
Vol. 36, No. 3, 249-258.

[21] Dyer, J.: A clarification of “Remarks on the Analytic Hierarchy Process.” Management
Science (1990), Vol. 36, No. 3, 274-275.

[22] Ehrlenspiel, K.: Kostengesteuertes Design — Konstruieren und Kalkulieren am Bild-

schirm. Konstruktion 40, (1988), 359-364.

[23] Eisenführ, F.; Weber, M.: Rationales Entscheiden. Springer, Berlin, 1993.

[24] Elvers, B.; Hawkins, S.; Schulz G. (Eds.): Ullmanns Encyclopedia of Industrial Chem-
istry. Volume B1: Fundamentals of Chemical Engineering. Verlagsgesellschaft, Wein-
heim, 1990.

[25] Eschenauer, H.; Juhani, K.; Osyczka, A.: Multicriteria Design Optimization. Springer,
Berlin, 1990.

[26] Fandel, G.: Optimale Entscheidung bei mehrfacher Zielsetzung. Lecture Notes in Eco-
nomical and Mathematical Systems, Springer, Berlin, 1972.

[27] Fishburn, P.: Foundations of Decision Analysis: Along the Way. Management Science
(1989), Vol. 35, No. 4, 387-405.

[28] Deutsche Forschungsgemeinschaft (DFG): Jahresbericht 1992. Aufgaben und Ergeb-

nisse. Band 1. Deutsche Forschungsgemeinschaft, Bonn, 1992.

[29] Deutsche Forschungsgemeinschaft (DFG): Jahresbericht 1992. Programme und Pro-

jekte. Band 2. Deutsche Forschungsgemeinschaft, Bonn, 1992.

[30] Fourier, J.: Théorie analytique de la chaleur. Paris, 1822. Quoted from [36].

[31] Franke, H.: Lexikon der Physik. Band 1, 2 und 3. Keller Verlagshandlung, Stuttgart,
1969.
Bibliography 119

[32] Frohn, A.: Einführung in die technische Thermodynamik. Akademische Verlagsge-

sellschaft, Wiesbaden, 1977.

[33] Garnich, R. Konstruktion, Design und Ästhetik. PhD Thesis, Stuttgart University, 1968.

[34] Garnich, R. Personal Communications. Esslingen, 1993.

[35] Görtler, H.: Dimensionsanalyse. Theorie der physikalischen Dimensionen mit Anwen-
dungen. Springer, Berlin, 1975.

[36] Görtler, H.: Zur Geschichte des -Theorems. Zeitschrift für Angewandte Mathematik
und Mechanik, 55 (1975), 3-8.

[37] Golden, B; Wasil, E; Harker, P. (Eds.): The Analytic Hierarchy Process. Springer, Berlin,
1989.

[38] Grigull, U.; Sandner, H.: Wärmeleitung. Springer, Berlin, 1990.

[39] Habenicht, W.: Die Bewertung neuer Technologien mit Verfahren des Multi-Criteria-
Decision-Making. Operations Research Proceedings (1990), 342-349.

[40] Habenicht, W.: Bewertung von Projekten auf der Basis von Mehrfachzielen. Technical
Report, Institut für Industriebetriebslehre, University of Hohenheim, 1991.

[41] Hainzl, J.: Verallgemeinerung des -Theorems mit Hilfe spezieller Koordinaten in
L IEschen Transformationsgruppen. Arch. Rat. Mech. Anal. 30, (1968), 321-344.

[42] Harker, P.: Alternative Modes of Questioning in the Analytic Hierarchy Process. Mathe-
matical Modelling (1987), Vol. 9, No. 3-5, 353-360.

[43] Harker, P.: Incomplete Pairwise Comparisons in the Analytic Hierarchy Process. Math-
ematical Modelling (1987), Vol. 9, No. 11, 837-848.

[44] Harker, P.: Root of a Positive Reciprocal Matrix: With Application to the Analytic Hierar-
chy Process. Applied Mathematics and Computation (1987), 22, 217-132.

[45] Harker, P.; Vargas, L.: The theory of ratio scale estimation: Saaty’s Analytic Hierarchy
Process. Management Science (1987), Vol. 33, No. 11, 1383-1403.

[46] Harker, P.; Vargas, L.: Reply to “Remarks on the Analytic Hierarchy Process” by
J.S. Dyer. Management Science (1990), Vol. 36, No. 3, 269-273.

[47] Heuser, H.; Wolf, H.: Algebra, Funktionalanalysis und Codierung. Teubner, Stuttgart,
1986.
120 Bibliography

[48] Hiller (Ed.): Theonis Smyrnaei Expositio rerum mathematicarum ad legendum Pla-
tonem utilium. Teubner, Leipzig, 1878. Quoted from [36].

[49] Hinsenkamp, G.: Kennfeldoptimierung adiabater Schrauben-Expansionsmaschinen

beim Einsatz unterschiedlicher Arbeitsfluide. Forschung im Ingenieurwesen — Engi-
neering Research 59, 1/2 (1993), 1-7.

[50] Hodapp, V.: Analyse linearer Kausalmodelle. Huber, Bern, 1984.

[51] Huntley, H.: Dimensional Analysis. Macdonald, London, 1952.

[52] Hwang, C.; Kwangsun, Y.: Multiple Attribute Decision Making. Lecture Notes in Eco-
nomics and Mathematical Systems 186, Springer, Berlin, 1981.

[53] I-DEAS User’s Guide: Model Solution and Optimization. Structural Dynamics Research
Corporation, Milford, Ohio, 1990.

[54] Jahnke; Emde; Lösch: Tafeln höherer Funktionen. Teubner, Stuttgart, 1966.

[55] Kirk-Othmer: Encyclopedia of Chemical Technology. Volume 7 and 19. John Wiley, New
York, 1979.

[56] Kleinschmidt, W.: Einige Problemstellungen bei Verbrennungsmotoren vor dimension-

stheoretischem Hintergrund. Universität Gesamthochschule Siegen, 1988.

[57] Klement, K.: Präsentation von Produktmodelldaten – Schnittstelle zwischen Computer-

Graphik und CAD/CIM. PhD Thesis, Technical University of Darmstadt, 1992.

[58] Kröplin, B.: Personal Communications. Institut für Statik und Dynamik der Luft- und
Raumfahrtkonstruktionen, Years 1991-1993.

[59] Langhaar, H.: Dimensional Analysis and Theory of Models. Wiley, London, 1951.

[60] Lykoudis, P.: Non-Dimensional Numbers as Ratio of Characteristic Times. International

Journal of Heat and Mass Transfer, 33, (1990), 1568-1570.

[61] Maser, S.: Numerische Ästhetik. Neue mathematische Verfahren zur quantitativen Be-
schreibung und Bewertung ästhetischer Zustände. Karl Krämer, Stuttgart, 1970.

[62] Mlejnek, H.: Grundlagen der Tragwerksoptimierung. Lecture Notes, Stuttgart Univer-
sity, 1985.

[63] Mittelstraß, J. (ed): Enzyklopädie Philosophie und Wissenschaftstheorie. Band 1. Bibli-

ographisches Institut, Mannheim, 1980.

[64] Moog, W.: Ähnlichkeits- und Analogielehre. VDI-Verlag, Düsseldorf, 1985.

Bibliography 121

[65] Oelsner, R.: Geschichte des Konstruierens in Deutschland. Vom künstlerischen Han-
deln zum formalisierten Wissen. Konstruktion, 44 (1992), 387-390.

[66] Paciolo, L.: De divina proportione. Venedig, Italien, 1509. Quoted from [8].

[67] Papalambros, P.; Wilde, D.: Principles of optimal design. Cambridge University Press,
Cambridge, 1988.

[68] Pawlowsky, J.: Die Ähnlichkeitstheorie in der physikalisch-technischen Forschung.

Grundlagen und Anwendungen. Springer, Berlin, 1971.

[69] Penrose, R.: Computerdenken. Spektrum der Wissenschaft, Heidelberg, 1991.

[70] Piotrowsky, J.: Theory of physical and technical measurement. Elsevier, Amsterdam,
1992.

[71] Rittel, H.: Planen, Entwerfen, Design. Kohlhammer, Stuttgart, 1992.

[72] Roque, W.; dos Santos, R.: Qualitative Reasoning, Dimensional Analysis and Com-
puter Algebra. Proceedings Int. Symp. on Symbolic and Algebraic Computation, ISSAC
91, 460-461.

[73] Rosenthal, Pressestelle: Personal Communications on Sales Prices in 1988. Archive

of the Rosenthal Company, 1993.

[74] Roth, K.: Elementarmodelle der Maschinenelemente— Möglichkeit zum Konstruieren

durch den Rechner. Konstruktion 40, (1988), 309-316.

[75] Roth, K.: Konstruieren mit Konstruktionskatalogen. Springer, Berlin, 1982.

[76] Roy, B.: Selektieren, Sortieren und Ordnen mit Hilfe von Prävalenzreaktionen: Neue
Ansätze auf dem Gebiet der Entscheidungshilfe für Multikriteria-Probleme. Zeitschrift
für betriebwirtschaftliche Forschung (1980), 32, 465-498.

[77] Roy, B.; Hugonnard, J.: Ranking of suburban line extension projects on the Paris metro
system by a multicriteria method. Transpn. Res. (1982), Vol. 16A, 4, 301-312.

[78] Rudolph, S.; Kröplin, B.: Über die systematische Bewertung von Konstruktionen. Bauin-
genieur 69 (1994), 1, 3-11.

[79] Rudolph, S.: On a Symbolic CAD-Front-End for Design Evaluation based on the Pi-
Theorem. 189-203, in: Gero, J.; Sudweeks, F.(Eds.): Advances in Formal Design Meth-
ods for CAD. Preprints of the IFIP WG 5.2 Workshop on Formal Design Methods for
Computer-Aided Design, Mexico City, Mexico, June 13-16, 1995.
122 Bibliography

[80] Rudolph, S.: On a Method for Design Evaluation Based on the Pi-Theorem. Submitted
March 1995.

[81] Rudolph, S.: Upper and Lower Bound Estimates of “T HE P RINCIPLES OF D ESIGN”.
Submitted March 1995.

[82] Saaty, R.: The Analytic Hierarchy Process—What it is and how it is used. Mathematical
Modelling (1987), Vol. 9, No. 3-5, 161-176.

[83] Saaty, T.: The Analytic Hierarchy Process. McGraw-Hill, New York, 1980.

[84] Saaty, T.; Vargas, L.: The Legitimacy of Rank Reversal. Omega (1984), Vol. 11, 513-
516.

[85] Saaty, T.; Vargas, L.: Inconsistency and Rank Preservation. Journal of Mathematical
Psychology (1984), Vol. 28, 205-214.

[86] Saaty, T.: Rank Generation, Preservation, and Reversal in the Analytic Hierarchy Pro-
cess. Decision Sciences (1987), Vol. 18, 157-177.

[87] Saaty, T.: Axiomatic Foundation of the Analytical Hierarchy Process. Management Sci-
ence (1986), Vol. 32, No. 7.

[88] Saaty, T.: An exposition of the AHP in reply to the paper “Remarks on the Analytic
Hierarchy Process”. Management Science (1990), Vol. 36, No. 3, 259-268.

[89] Saaty, T.; Vargas, L.: Comparison of Eigenvalue, Logarithmic Least Squares and Least
Squares Methods in Estimating Ratios. Mathematical Modelling (1984), Vol. 5, 309-324.

[90] Schwarz, H.: Methode der Finiten Elemente. Teubner, Stuttgart, 1984.

[91] Schwefel, H.: Numerische Optimierung von Computer-Modellen mittels der Evolutions-
strategie. Birkhäuser, Basel, 1977.

[92] Shapiro, (eds.): Encyclopedia of Artificial Intelligence. Volume 1 and 2. John Wiley, New
York, 1987.

[93] Shive, J.; Weber, R.: Ähnlichkeiten in der Physik. Springer, Berlin, 1993.

[94] Siddall, J.: Analytical decision-making in engineering design. Prentice-Hall, Englewood

Cliffs, 1972.

[95] Steller, E.: Kunst und Computer. Bibliographisches Institut, Mannheim, 1992.

[96] Spurk, J.: Dimensionsanalyse in der Strömungslehre. Springer, Berlin, 1992.

Bibliography 123

[97] Stephan, K.; Mayinger, F.: Thermodynamik. Einstoffsysteme Band 1. Springer, Berlin,
1986.

[98] Triantaphyllou, E.; Mann, S.: An Examination of the Effectiveness of Multi-Dimensional

Decision-Making Methods: A Decision-Making Paradox. Decision Support Systems 5
(1989), 303-312.

[99] Vargas, L.: Reciprocal matrices with random coefficients. Mathematical Modelling
(1982), Vol. 3, 69-81.

[100] Vargas, L.: Analysis of Sensitivity of Reciprocal Matrices. Applied Mathematics and
Computation (1983), 12, 301-320.

[101] Vargas, L.: Utility Theory and Reciprocal Pairwise Comparisons: The Eigenvector
Method. Socio-Economical Planning Sciences (1986), Vol. 20, No. 6, 387-391.

[102] Vieta, F.: In artem analyticem isagoge. Tours, 1591. Quoted from [3].

[103] Wedekind, H.: Objektorientierte Schema-Entwicklung. Wissenschaftsverlag, Mann-

heim, 1992.

[104] Weld, S.: Theories of Comparative Analysis. MIT Press, 1990.

[105] Winkler, R.: Decision modeling and rational choice: AHP and utility theory. Manage-
ment Science (1990), Vol. 36, No. 3, 247-248.

[106] Winston, P.: Künstliche Intelligenz. Addison-Wesley, München, 1984.

[107] Yilmaz, M.: Multiattribute Utility Theory: A Survey. Theory and Decision (1978), 9,
317-347.

[108] Ziegler, H.: Vorlesungen über Mechanik. Birkhäuser, Basel, 1977.

[109] Zimmermann, H.; Gutsche, L.: Multi-Criteria Analyse. Springer, Berlin, 1991.

[110] Zwicky, F.: Entdecken, Erfinden, Forschen im morphologischen Weltbild. Droemer-

Knaur, München, 1966.