STRUCTURAL AWARENESS IN COMPLEX PRODUCT DESIGN – …

9TH INTERNATIONAL DESIGN STRUCTURE MATRIX CONFERENCE, DSM’07 16 – 18 OCTOBER 2007, MUNICH, GERMANY

STRUCTURAL AWARENESS IN COMPLEX PRODUCT DESIGN – THE MULTIPLE-DOMAIN MATRIX

Institute of Product Development, TU München

Keywords: Multiple-Domain Matrix, Graph Theory

1 INTRODUCTION Complexity in product development represents a challenge that can be outlined by the following statements: In spite of well-known approaches on avoiding, reducing, and controlling complexity in product development, a steady increase of complexity can be assumed for the past and predicted for the future. Structures that result from the connectivity of elements represent an important tool for accessing complexity in product development. And if complexity can be controlled, it does not necessarily imply negative aspects but can provide competitive advantages in product development. Concerning the management of complexity in product development this paper introduces an approach that focuses on the constellations formed by existing linkages in objects. This provides far-reaching possibilities for analysis, control, and optimization of complex products and services in a holistic context, while the amount of data remains manageable. The analysis implies exploring the connectivity of an object in order to identify content, and thus to interpret and better understand the considered object. Controlling complexity means interacting with the elements of an object in order to understand consequences resulting from implemented actions. And optimization implies adapting the connectivity of an object in order to better suit its desired functionalities. The holistic context means to consider all relevant aspects of a complex object. In the following, the objectives and the technical realization of the approach will be presented.

2 OBJECTIVES OF STRUCTURE-BASED COMPLEXITY MANAGEMENT The presented approach on complexity management is based on the finding that on the one hand each technical product, as well as the processes of its creation, can be accessed by its structure (the term “structure” is understood as the network formed by dependencies between system elements and represents a basic attribute of each system). On the other hand, the structure generation itself emerges without being controlled (passive structure synthesis) and often does not apply to the requirements of the product or process. However, the structure determines the attributes and behavior of products and processes. Some of their characteristics can even be defined in the structural composition only, e.g. the robustness to component changes [1]. Cyclic dependency chains between components can cause self-energizing or self-impeding effects and hierarchical substructures can produce avalanche effects; and the creation of clusters can indicate possibilities of modularization [2]. Known approaches of complexity management focus either on isolated aspects (e.g. optimizing variant numbers from a production view), or address a comprehensive modeling approach implemented at an abstract level of detail. This dilemma between the considerations of extracts and global views that are difficult to manage can be solved by applying a structure-based approach comprising the following objectives: Determination of relevant domains: It is important to determine all relevant aspects, as improper simplification or the extraction of single aspects can hinder the re-transfer of the outcome to the initial problem. In contrast, all irrelevant aspects must be excluded from considerations. Selection of relevant system views: Complex systems comprise a multitude of aspects and dependency types that allow for many system views; users must be able to select the relevant ones. Analysis of dependency structures: Dependency structures must be analyzed concerning the existence of characteristic structural attributes that allow for suggesting the system behavior. Structure manual: A structure manual points out relevant system elements and structure constellations. This helps estimate the potentials and risks of component changes. Identification of suitable structural designs: The established structural optimization approaches (e.g. modularization, design for product robustness) must be amended and provided at a generic level for application to user-defined systems.

Maik Maurer, Udo Lindemann

87

3 THE MULTIPLE-DOMAIN MATRIX For the interaction with complex systems, many different visualization techniques and computational approaches exist; however, matrix-based methods have been commonly used in product development. Starting from the modeling of relations within one domain by use of the Design Structure Matrix (DSM), the mapping between different domains by Domain Mapping Matrices (DMMs) has acquired a multitude of development methods. Furthermore, some multi-domain approaches are mentioned in the literature that combine several DSMs or DMMs and allow for more comprehensive system consideration [3] [4]. However, a lack of appropriate analysis methods exists.

A

Derived dependencies

A 1

1

X

DSM area

Native DSM

Derived DSM

Domains

X

DMM area

Native DMM

Elements

MDM

Dependency

Matrix cell for indicatingself -ref lexive dependency

Bi-directional dependency

Domain mapping logic

Dif ferent dependency types

Figure 1. Construction of a Multiple Domain Matrix

In Figure 1, an integrative matrix combination of DSMs and DMMs, which will be denoted as Multiple-Domain Matrix (MDM). The MDM possesses all features of a common DSM; in fact, it represents a DSM on a higher level of abstraction.

A

B

1

A

B

1

A

B

1

A

B

1

2

A

B

1

2

A

B

1

2

1 2 3

4 5 6

A

1

Element of the domain in question

Element of the additional domain

Native dependency

Derived dependency

Figure 2. Computational logics for deriving DSMs from other DSMs and DMMs

The MDM allows for the systematic modeling of networks comprising different domains and dependency types. Although MDM models are not restricted to isolated views of single domain and dependency types, DSM-like analysis can still be performed, as subsets of homogeneous content are represented separately in the implied subsets. The MDM is generally not a two-dimensional matrix system, as different matrices can exist for the same DSM or DMM subset. This means that two different dependency types are depicted that exist within the same combination of domains in parallel.

88

The decisive advantage of a MDM is the possibility for deriving DSMs based on information from other DSMs and DMMs. Six computational logics exist, if dependencies possess a direction (Figure 2). Obtainable insight to complex systems will be shown in the presentation.

4 USING GRAPH THEORY FOR ENHANCED DSM ANALYSIS Once DSMs have been determined analysis methods can be applied in order to obtain structural characterizations. Most of the available algorithms refer directly to the matrix representation, i.e. structural characteristics are pointed out by realignment of element orders in matrices, e.g. [5]. A common base for the description of all DSM analyses is given by the graph theory. Its consequent application on DSM-conform structures does not only allow for more efficient and improved application of known methods (e.g. triangularization), but also provides a many further analysis methods. A collection of these methods will be shown in the presentation.

5 THE STRUCTURE MANUAL AND SUITABLE STRUCTURAL DESIGN A structure manual comprises of selected information about the occurrence of basic analysis criteria for further interpretation. Results of methodical analyses are systematically presented and facilitate the planning and tracing of system adaptations as well as their resulting impact. These methodical analyses are the feed-forward-analysis, the impact checklist, mine seeking, the structural ABC-analysis, and the trace-back-analysis. The provision of visualized networks that clarify domain-spanning dependencies and occurring structural characteristics can help developers to better understand a system, especially effects beyond their own scope of responsibility. As well, the structure manual can indicate opportunities and restrictions for product adaptations.

6 CONCLUSIONS The here presented approach meets both major shortcomings of existing methods in complexity management – the lack of promising and generic instruments for analysis, control and optimization of complex system structures on the one hand, and the applicability on comprehensive and domain-spanning problems on the other. The approach is founded on established matrix-based methods and provides therefore the derived Multiple-Domain Matrix as core of a systematic methodology. Available analysis methods and strategies have been transferred to the integrative fundament of graph theory and systematically complemented. Building on that, practical procedures for the analysis, control, and optimization of complex system structures have been generated.

REFERENCES [1] Coyle, R. G. System Dynamics Modelling – A Practical Approach, London: Chapman & Hall

1996. [2] Baldwin, C. Y.; Clark, K. B. Design Rules - The Power of Modularity. Vol. 1, Cambridge: MIT

Press 2000. [3] Bongulielmi, L.; Henseler, P., Puls, C.; Meier, M. The K- & V-Matrix Method – An Approach

in Analysis and Description of Variant Products, In: Proceedings of the ICED 01, Glasgow. Bury St. Edmunds: IMechE 2001, pp 571-578.

[4] Yassine, A.; Whitney, D.; Daleiden, S.; Lavine, J. Connectivity Maps: Modeling and Analysing Relationships in Product Development Processes, Journal of Engineering Design 14 (2003) 3, pp 377-394.

[5] Kusiak, A. Engineering Design – Products, Processes and Systems, San Diego: Academic Press 1999.

Contact: Maik Maurer Technische Universität München Institute of Product Development Boltzmannstr. 15 85748 Garching Germany Phone +49 89 289 15155 Fax +49 89 289 15144 e-mail: [email protected] URL: www.pe.mw.tum.de

89

Structural awareness in complex product design – The Multiple-Domain Matrix

Maik Maurer, Udo Lindemann

9th International DSM Conference 2007- 2

Overview

• The challenge of complexity in product development• Requirements on effective complexity management• The Multiple-Domain Matrix

– Systematic concept– Deduction of system views

• Management of multiple-domain environments

90


Facts about complexity in product development

• Steady increase of complexity• All artefacts comprise

structures describing the linkages between elements

• Structure generation emerges without being controlled

• The structure determines the attributes and behavior

• Structures often do not apply to the initial requirements Process complexityOrganizational complexity

1985:9 Segmente

2005:> 40 Segmente

2002:33 Segmente

1997:24 Segmente

1992:16 Segmente A00/A0 3-DOOR

A 5-DOOR

B-TOP-LIM

Kombi-A0/AKombi -B-TOP

MPV-A

A 3-DOOR.

A0 5-DOOR

C-LIM

Kombi-B-LOW

A/B LOW 4-DOOR

SUV -C

B LOW-5-DOOR DIV.A00-A(FUN)

MPV-B-6+SEATS

Cabrio-A/B

Roadster-B

VAN A

SUV TOP 5-DOOR

D-LIM

MPV-B 5 SEATS

CONVERTIBLE -B/CCabrio -A

SUV LOW 5-DOOR

SUV MED. 5-DOOR

Roadster-A

SUV 3-DOOR

Coupe´-D

MPV-TOP

PICK UP

Cabrio -D

Coupe´-C

Prestige

Vielseitigkeit / Nutzen

Preis

Fahrspaß / Individualität1985:9 Segmente

2005:> 40 Segmente

2002:33 Segmente

1997:24 Segmente

1992:16 Segmente

1985:9 Segmente1985:9 Segmente

2005:> 40 Segmente2005:> 40 Segmente



1992:16 Segmente1992:16 Segmente A00/A0 3-DOOR

A 5-DOOR

B-TOP-LIM

Kombi-A0/AKombi -B-TOP

MPV-A

A 3-DOOR.

A0 5-DOOR

C-LIM

Kombi-B-LOW

A/B LOW 4-DOOR

SUV -C

B LOW-5-DOOR DIV.A00-A(FUN)

MPV-B-6+SEATS

Cabrio-A/B

Roadster-B

VAN A

SUV TOP 5-DOOR

D-LIM

MPV-B 5 SEATS

CONVERTIBLE -B/CCabrio -A

SUV LOW 5-DOOR

SUV MED. 5-DOOR

Roadster-A

SUV 3-DOOR

Coupe´-D

MPV-TOP

PICK UP

Cabrio -D

Coupe´-C

Prestige

Vielseitigkeit / Nutzen

Preis

Fahrspaß / Individualität

Multiple-domainlinkages

Market complexityProduct complexity

If complexity can be controlled, it does not necessarily imply negative aspects but provides competitive advantages


Objectives of the structure-based complexity management

• Known approaches– implement comprehensive modeling on an abstract level of detail– or focus on isolated aspects

• Objectives of the structure-based approach

– Determination of relevant domainsWhich aspects must be considered?

– Selection of relevant system viewsWhich information is required?

– Analysis of dependency structuresWhat can be concluded from constellations?

– Structure manualWhat to do and expect when interacting in complex systems?

– Identification of suitable structural designsDoes the structural design suit for the product’s intention?

91


The need for systematic multiple-domain modeling

A

B

C

1

2

3

αβ

γ

δ4

Component 1

Component 2

Component 3

Component 1

Person 1

Data 1

Affects

Affects

AffectsProcessed

by

Uses

DescribesClosedimpactchain ?

3

2

1 4

5

67

89

10

11

Single-domain network

e.g. feedback loop analysis

Multiple-Domain network• Most problems require

consideration of multiple domains and dependency types

• People often mix different domains in one network

• So far, most analyses can not be applied for multiple-domain networks


Systematic subdividing of multiple-domain networks: DSMs

• Intra-domain dependencies can be extracted to DSMs• Different dependency types form different DSMs

X

X

X

X X

A B C

A

B

C

X

X X

X

α

A

B

C

1

2

3

α

β

γ

δ

4

α

β

β

γ

γ δ

δ

2 31 4

1

4

2

3

X

X

X

1

2 3

4

1

2

3

4

Representation in DSMs

92


Systematic subdividing of multiple-domain networks: DMMs

B

C

1

2

3

α

β

γ

δ

4

X

X

X

X X

X

α

β

γ

δ

2 31 4

α β γ δ

A

B

C

1

2

3

2 31 4

X X

A

B

C

2 31 4

A

• Inter-domain dependencies can be extracted to DMMs• Different dependency types form different DMMs

Representation in DMMs


Aggregation of the Multiple-Domain Matrix

• A MDM is a DSM on a more abstract level of detail• A MDM comprises DSMs and DMMs• Different dependency types can be modeled (3D-matrix)• DSM and DMM subsets can result from native or computed data• The MDM provides the methodical basis for domain-spanning

system analysis and optimization

A

B

C

1

2

3

α

β

γ

δ4

Representation in MDMs

93


Composition of the Multiple-Domain Matrix

A

A 1

1

X

DSM area

Native DSM

Derived DSM

Domains

X

DMM area

Native DMM

Elements

MDM

Dependency

Matrix cell for indicatingself-reflexive dependency

Bi-directional dependency

Domain mapping logic

Different dependency types

Derived dependencies


Deduction of DSMs from multiple-domain contexts

Xd

+ =

1

2

3

α

β

γ

δ4

α

β

γ

δ

α β γ δ

Derive indirect dependencies by MDM

Derived DSM

How do depend on each other based on the dependencies of and ?

94


Special use cases of the MDM application

V-Matrix1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3

Attr. 11.1 1 1 1 11.2 1 1 1 1 1 1

Attr. 22.1 12.2 12.3 1 1 1

Attr. 33.13.23.3

Attr

.1 Attr

.2 Attr

.3

V-Matrix1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3

Comp. 1

1.1 1 1 11.2 1 1 1 1

Comp. 2

2.1 1 12.2 1 12.3 1 1

Comp. 3

3.13.23.3

Com

p.1

Com

p.2

Com

p.3

K-Matrix1.1 1.2 2.1 2.2 2.3 3.1 3.2 3.3

Comp. 1

1.1 1 1 11.2 1 1

Comp. 2

2.1 12.2 12.3 1

Comp. 3

3.1 1 13.2 1 1 13.3 1 1

Attr

.1 Attr

.2 Attr

.3

Customer view

Technical view

K&V-matrixElement type A

Element B1 X XElement B2 XElement B3 X XElement B4 XElement B5 XElement B6 XElement B7 X XElement B8 X

Element A1

Element A2

Element A3

Element A4

Element A5

Element A6

Element type B

Element type C

Element A1 XElement A2 X XElement A3 X XElement A4 X XElement A5 XElement A6 X

Element C1

Element C2

Element C3

Element C4

Element C5

Element C6

Element type A

Element C7

Element type C

Element B1 1 2 2Element B2 4 4Element B3 2 2 3 3Element B4 3 3Element B5 3 3Element B6 4 4Element B7 4 4 5Element B8 6

Element C1

Element C2

Element C3

Element C4

Element C5

Element C6

Element type B

Element C7

Connectivity maps

Deduction of a DSM by1 DSM & 1 DMM

(Bongulielmi et al. 2001)

Deduction of a DMM by2 DMMs

(Yassine et al. 2003)


Systematics for deriving DSMs using one auxiliary domain

• Non-directed and bi-directional relations would extend the quantity of computational use cases

• Additional domains can be considered by sequential application of the systematics

A

B

1

A

B

1

A

B

1

A

B

1

2

A

B

1

2

A

B

1

2

1 2 3

4 5 6

A

1

Element of the domain in question

Element of the additional domain

Native dependency

Derived dependency

• Computations based on matrix multiplications

95


Application of Multiple-Domain approaches

Nodes/EdgesActive sumActivityArticulation nodeAttainabilityBridge edgeBusClosenessCriticalityEnd nodeIsolated nodeLeafPassive sumStart nodeTransit node

SubsetsBi-connected componentClusterDistanceFeedback loopHierarchyLocalityPathQuantity of indirect dependenciesSimilaritySpanning treeStrongly connected part

SystemsBandingClusteringDegree of connectivityDistance matrixMatrix of indirect dependenciesPartitioning, triang., sequencing

31 basic analysis criteria considering..

XX

X X

X

α

β

γ

δ

α β γ δ

I. MDM modeling

II. DSM deduction

III. DSM analysis

IV. Interpretation,application on design

a) Optimized structure management- Structure manual- Feed-forward analyses- Trace-back analyses- …

b) Active structural design- Structural ABC-analyses- Identification of structural

potentials- …


Application of MDMs in projects

Components KomponentenEigenschaften (statisch) der Komponenten

Eigenschaften (dynamisch) der Komponenten

Versuchs-parameter Einsitzprozedur Lastfall-

CharakteristikaDummy

(-Eigenschaften) Kriterien

Komponenten K hat geometrische Abhängigkeit zu K

K-Änderung kann E-Änderung bewirken

K-Änderung kann E-Änderung

bewirken--- --- --- --- ---

Eigenschaften (statisch) der Komponenten

E-Änderung kann K-Änderung

bewirken

E-Änderung (stat.) kann E-Änderung (stat.) bewirken

E-Änderung (stat.) kann E-Änderung (dyn.) bewirken

---

E-Änderung (stat) kann ESP-Änderung bewirken

--- ---

E-Änderung (stat.) kann Krit-

Änderung bewirken


E-Änderung (dyn.) kann K-Änderung

bewirken

E-Änderung (dyn.) kann E-Änderung (stat.) bewirken

E-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

--- --- --- ---

E-Änderung (dyn.) kann Krit-

Änderung bewirken

Versuchs-parameter --- ---

V-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

V-Änderung kann V-

Änderung bewirken

--- --- ---

V-Änderung kann Krit-Änderung bewirken

Einsitzprozedur --- ---

ESP-Änderung (dyn.) kann E-

Änderung (dyn.) bewirken

ESP-Änderung kann V-

Änderung bewirken

ESP-Änderung kann ESP-Änderung bewirken

--- ---

ESP-Änderung kann Krit-Änderung bewirken

Lastfall-Charakteristika --- ---

L-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

--- … --- --- Lastfall bestimmt Kriterien

Dummy (-Eigenschaften) --- ---

D-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

D-Änderung kann V-

Änderung bewirken

--- ---

D-Änderung kann D-

Änderung bewirken

D-Änderung kann Krit-Änderung bewirken

Kriterien --- --- --- --- --- --- ---

Krit.-Änderung kann Krit.-Änderung bewirken

KomponentenEigenschaften (statisch) der Komponenten


Versuchs-parameter Einsitzprozedur Lastfall-

CharakteristikaDummy

(-Eigenschaften) Kriterien

Komponenten K hat geometrische Abhängigkeit zu K

K-Änderung kann E-Änderung bewirken

K-Änderung kann E-Änderung

bewirken--- --- --- --- ---

Eigenschaften (statisch) der Komponenten

E-Änderung kann K-Änderung

bewirken

E-Änderung (stat.) kann E-Änderung (stat.) bewirken

E-Änderung (stat.) kann E-Änderung (dyn.) bewirken

---

E-Änderung (stat) kann ESP-Änderung bewirken

--- ---

E-Änderung (stat.) kann Krit-

Änderung bewirken


E-Änderung (dyn.) kann K-Änderung

bewirken

E-Änderung (dyn.) kann E-Änderung (stat.) bewirken

E-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

--- --- --- ---

E-Änderung (dyn.) kann Krit-

Änderung bewirken

Versuchs-parameter --- ---

V-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

V-Änderung kann V-

Änderung bewirken

--- --- ---

V-Änderung kann Krit-Änderung bewirken

Einsitzprozedur --- ---

ESP-Änderung (dyn.) kann E-

Änderung (dyn.) bewirken

ESP-Änderung kann V-

Änderung bewirken

ESP-Änderung kann ESP-Änderung bewirken

--- ---

ESP-Änderung kann Krit-Änderung bewirken

Lastfall-Charakteristika --- ---

L-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

--- … --- --- Lastfall bestimmt Kriterien

Dummy (-Eigenschaften) --- ---

D-Änderung (dyn.) kann E-Änderung (dyn.) bewirken

D-Änderung kann V-

Änderung bewirken

--- ---

D-Änderung kann D-

Änderung bewirken

D-Änderung kann Krit-Änderung bewirken

Kriterien --- --- --- --- --- --- ---

Krit.-Änderung kann Krit.-Änderung bewirken

Efficient change management for automotive safety systems

Attributes (s)

Attributes (d)

Test parameters

Seating

Loading charact.

Dummy

Criteria

Pr

Co implicates Pr

---

Ch implicates Pr

Production boundary conditions

Pr

Pr determines Co (production constraint)

---Pr determines Ch

(production constraint)

Pr has influence on Pr

Types of high pressure pump Co Fu Ch

ComponentsCo

Co has influence on Co (geometry,

package)Co serves Fu Co has influence

on Ch

FunctionsFu

--- Fu requires Fu(functional model)

---

CharacteristicCh

Ch has influence on Co

=(Co Ch) T--- Ch has influence

on Ch

Components

Functions

Characteristics

Production boundary conditions

Optimization of high pressure pumps regarding modular design

Milestones

Documents

Attributes

Tasks

Dates

Roles

Improvement of process landscapes

Presentations on projects during this conference

96


Conclusions & Outlook

• The MDM is the basis of domain-spanning network analysis– Focusing on relevant subsets allows for efficient acquisition– Deriving system views for analysis by DSM and graph theory– Consideration of domain-spanning effects and change propagation

• Application of the MDM requires explicit system definition– Relevant domains– Relevant dependency types– Specification of useful DSM computations

• Future enhancements– Quantification of elements and dependencies by attributes– Computational schemes for DMM deduction– Computational schemes for consideration of several domains

• Software– LOOMEO serves for MDM acquisition, analysis, and interpretation

97