Decision Based Design Open Workshop

Design is Information Processing - What Would a Science of Engineering Design Look Like & How Does Decision-Based Design Fit Into it?

Dr. David W. Rosen
david.rosen@me.gatech.edu

Systems Realization Laboratory
The George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0405
Phone: (404) 894-9668

Abstract

Design is a human-centered activity that is, fundamentally, an exercise in information processing. In this brief position paper, I outline some of my beliefs concerning the development of a science of engineering design. I concentrate on three points. First, I believe that design is not just decision making. Decision making is a, but not the, fundamental construct in design. Second, I develop a simple ontology of design information and its processing, then ask how Decision-Based Design fits into this view of design. Third, I propose a concerted effort to explicate information in the context of design. What is information and how is it transformed during design?

Nomenclature

I - Information in the universe (known and unknown). Information is classified into objects, relationships, transformations (operations), and properties.

ID - Information known by the designer or design team. ID Õ I.

IP - Information that is relevant to the current design problem. IP Õ I.

IPD - Information that is relevant to the current design problem and known by designer.
IPD = ID « IP.

L - Relationship space. L Õ I.

O - Object space. O Õ I.

P - Properties. Objects, transformations, and relationships can have properties.
P:I Æ (¬n » ¡n), ¬n = space of real numbers in n dimensions, ¡n = integers in n dimensions.

R - Representation space. Structured information within a viewpoint. Applying a viewpoint, V, generates a representation of specified information:
V:I Æ R. Vj(i OE IPD) Æ Rij for a particular viewpoint Vj

T - Transformation space used to model processes. Each transformation converts one representation into another: T:(R ¥ ID ¥ IPD) Æ R. T Õ I.

V - Viewpoint. Aspect of product life cycle. Acts as a filter on ID or IPD.

The Nature of Engineering Design

Design is a human-centered activity that is essentially a series of transformations of information. I believe that the two fundamental types of transformations are tasks and decisions. During these transformations, new information ìatomsî are added and new relationships among existing information ìchunksî are developed. Information can be classified in many ways, but for our purposes, I will identify: objects, relationships, transformations (operations), and properties. A property is a measurable quantity of an object, relationship, or transformation.

In this paper, I present an abstract model of design information domains. With some work, this could become an ontology for engineering design. The presentation is along the lines of Stinyís work on developing algebras of design (Stiny, 1991). This serves as my contribution to the development of a science of engineering design. The issue that I would like to raise is: Within the context of this design information model, how does Decision-Based Design fit in? As we meet over the next few years, my approach to DBD will center on this issue.

Design Information Domains and Relationships
Similar to Yoshikawa (1981), I define the space of all information, I, in the universe, whether it is known or yet unknown, to be the space (ordered set) of all objects, relationships, and transformations. That is, I = space generated by O, L, T. Another way to divide up information is to introduce the designerís knowledge and the particular design problem of interest. Note that I use the term designer to denote an individual designer or a design team of any size. The notation ID is used to denote the designerís knowledge, while IP denotes the information that is relevant to the current design problem. The intersection IPD = ID « IP is essentially the current state of information about the design problem.

Along the time-line of a product, many activities can be identified; for example, design, manufacture, usage, service, recycling, etc. Finer gradations can be identified. For each of these activities, a viewpoint can be identified that focuses on the subset of information that is relevant to that activity. The set of all viewpoints and their combinations is denoted V.

Generally, designers act on information through representations of that information. For each viewpoint in V, a class of representations can be identified. That is, there are design representations, manufacturing representations, analysis representations, etc. The representations are a filtered, perhaps more structured collection of design information. In general, V:I Æ R. For a particular viewpoint, j, and at a given point in time, Vj(i OE IPD) = Rij is a representation of the known, problem-specific information about a design in viewpoint Vj.

Properties are attributes of objects, relationships, and transformations. Designers often reason about design problems, artifacts, and processes through their properties, rather than directly with the elements themselves. Generally, properties are physical, measurable quantities, such as length, weight, melting point, etc., but can be other types of measurable quantities. Properties will be modeled as a relation from information to measures (real or integer numbers): P:I Æ (¬n » ¡n). It is important to highlight an important class of properties, that of human interpretations or evaluations of information. These evaluations include preference, certainty, and imprecision.

The final set of notations that will be introduced is that of transformations. Design processes consist of a structured set of transformations that operate on information, some of which will be general information known by the designer, while other information will be problem-specific. As stated, transformations can be tasks or decisions. I believe that designers operate on information through representations of that information that are relevent to a specific viewpoint. So, at one level, transformations can be defined as: T:(R ¥ ID ¥ IPD) Æ R. In other words, the current representation can be augmented with other information that the designer knows. At the level of information, another type of transformation can be identified, that of learning something new about the design that the designer had not known in any other contexts. This type of transformation can be defined as: T:(IP ¥ IPD) Æ IPD.

Properties of Design Processes

Processes are constructed from transformations, both tasks and decisions. Let me enumerate a few fairly elementary transformations.

Conversion: convert between representation structures without changing the underlying information: Conversion:Vi(I) Æ Vj(I), i ¹ j.

Extraction: extract information of interest from a set of representations: Extraction:R Æ R.

Modification: change the values of the representation without changing the structure: ModificationR:R Æ R, ModificationI:I Æ I.

Creation: create/instantiate/add values to a representation. At the level of information, Creation: IP Æ IPD.

Selection: evaluate a set of alternatives relative to a set of attributes, assigning measures of preference to each alternative (representation): Selection:R Æ ¬.

With specific definitions of transformations, we can assemble them into design processes and experiment with them, assessing their performance. This may lead to an ability to predict performance of design processes, before executing them. A predictive ability indicates at least the hope for theory development.

Certain phenomena can be identified regarding information during design. As design proceeds, more and more information is accumulated and structured. Information becomes more quantitative as design proceeds. In the notation introduced here, for transformation k, IPDk » i Õ IPDk+1, where i OE IP.; that is, transformations never result in reductions in the information known about a design. The scope of focus changes, at times widening and at others contracting. Generally, as design proceeds, fewer and fewer alternatives exist (reduction of design freedom) but those alternatives that survive are understood better and better, that is, their certainty increases. I believe that these phenomena can be observed during design at any scale; i.e., from an individual planning a new route to get to work, to the design of a Boeing 777.

Foundation for a Science of Engineering Design

A short perusal of the development in the previous section gives ample clues as to the foundations, at least at a basic level, for a science of engineering design.

Set Theory
At the core of this foundation is set theory, including mathematical relations and the structures they induce (partially-ordered sets, lattices, topologies, etc.). Since graphs are a natural way to represent many structured sets, graph theory is another important topic.

Problem Formulation
These topics are particularly relevant to problem formulation. Trade-off decisions and much of optimization-based problem solving are typically confined to operating within subsets of ¬n. A significant set of challenges is to generate feasible concepts in discrete design spaces. In order to search such spaces, the spaces must be defined mathematically, a problem that is neither understood, nor necessarily even appreciated.

Project Planning and Resource Allocation

This addresses the need to formulate the design problem correctly and to design the design organization and design process rationally. As plans are composed of discrete structures (transformations), plan formulation is necessarily performed within a discrete design space and will require synthesis methods capable of operating in these spaces.

Resource allocation is a much-studied area that is typically amenable to methods of optimization within subsets of ¬n. However, simultaneously planning a design project and allocating resources is a complex mixed-integer-discrete problem.

Theory of Design Strategies
What I mean by this is to provide a solid foundation for characterizing different approaches to design, whether they are called ìrobust design,î ìadaptive design,î ìdesign for flexibility,î etc. Is Decision-Based Design a strategy or something broader? How do we represent and reason with strategies? It seems that strategies give preference to certain types of transformations, certain patterns of transformations, certain types of evaluation criteria; generally, preferences and priorities.

Theory of Utility and Preference
Human interpretations and evaluations of information play a key role in guiding design processes. I include this topic here for completeness, as there are probably many Workshop participants that are much more knowledgeable about this topic than I am.

Design Information

The long development of domains of design information earlier in the paper should be an indication of the importance of a rigorous, comprehensive model of information. Two bodies of literature are particularly relevant to such a model: Information Theory, and Suhís Axiomatic Design.

Information Theory
Information theory was developed in the 1940ís by Shannon (1947) and was subsequently used to explain classical thermodynamics (Tribus, 1960). The concept of minimizing information content was shown to be related to the principle of maximum entropy (2nd law of thermodynamics) by Jayne (1950). More recently, an interpretation of Suhís Information Axiom has been proposed in terms of information theory (Vadde et al, 1994). Whether or not this is a fruitful path to pursue remains to be seen.

Axiomatic Design
Axiomatic Design as developed by Suh (1990) postulates two axioms about design:

ï Independence Axiom - functional requirements (FRs) should be satisfied independently with respect to a change in a design parameter.

ï Information Axiom - the information content in a design should be minimized.

Suh provides definitions of information content and proposes measures of this content. Without debating the merits of Axiomatic Design, it is encouraging to note that attempts have been made to measure information content of an in-progress design. Can Suhís measures be derived from Information Theory? Probably not. But can other measures of information content be derived from Information Theory? I believe so. The structure of engineering design information is much richer than the information bits that Shannon studied, but with proper mathematical modeling of design information, the concepts of Information Theory and entropy should transfer.

Closure

I believe that it will be possible to formulate a science of engineering design and that this science will be useful. The development of design information domains presented here does not represent a theory, but it is meant to indicate one direction of research that must be pursued in order to construct such a theory as a basis for a science.

What is Decision-Based Design? One way to rigorously provide an answer is to construct an ontology, similar to the domains and transformations presented here, for DBD. This will contribute to achieving the goal of defining what design is from a decision making perspective. This is my interest. Such an ontology will also contribute to the goal of establishing relationships between theories developed in other science domains with a theory of design.

As a final comment, my position paper may be viewed as a digression away from the topic at hand. Perhaps. But my point is to look at the big picture of design theory and to search for a general foundation from which specific perspectives, such as DBD, can be rooted.

References

Jayne, E T, (1957) Information Theory and Statistical Mechanics, Vol. 106, p. 620.

Shannon, C E (1958) A Mathematical Theory of Communication, Vol. 27, pp. 379-423, 623-656.

Stiny, G (1991) ìThe Algebras of Design,î Research in Engineering Design, 2(3): 171-181.

Suh, N P (1990), The Principles of Design, Oxford University Press, New York, 1990.

Tribus, M (1961) ìInformation Theory as the Basis for Thermostatics and Thermodynamics,î J. of Applied Mechanics, Vol. 28, pp. 539-51.

Vadde S, Allen, J K, Lucas, T, & Mistree, F (1994) ìOn Modeling Design Evolution Along a Design Timeline,î 5th AIAA/VASA/USAF/ISSMO Symposium on Multidisciplinary Analysis and Optimization, Panama City, FL, Sept. 7-9.

Yoshikawa, H. (1981) General Design Theory and a CAD System, Man-Machine Communications in CAD/CAM, Tokyo, October 2-4, 1980, Proceedings of IFIP WG 5.2, North-Holland, Amsterdam, pp. 35-58.