Perspectives on research methodology in the field of “AI and design” are discussed. This perspective is based on a view of a “science of design” focusing on methods of design and on characteristics of design tasks that affect what methods are relevant for a given task. Two methodological issues are discussed: the need to try applying a design method on multiple tasks and domains, and the need to work with collaborators who are experts in the task domain of each research system you build.

We analyze the diagnosis task in the context of adaptive design and redesign of physical devices. We identify three types of diagnosis tasks that differ in the types of information they take as input: the design does not achieve a desired function of the device, the design results in an undesirable behavior, and a specific structural element in the design misbehaves. We describe a model-based approach for solving the diagnosis task in the context of adaptive design and redesign. This approach uses functional models that explicitly represent the device functions and use them to organize teleological and causal knowledge about the device. In particular, we describe a specific kind of functional model called structure—behavior—function (SBF) models in which the causal behaviors of the device are specified in terms of flow of substances through components. We illustrate the use of SBF models with three examples from Kritik2, a knowledge system that designs new devices by retrieving, diagnosing, and adapting old device designs.

This paper is concerned with presenting guidelines to aide in the selection of the appropriate network architecture for back-propagation neural networks used as approximators. In particular, its goal is to indicate under what circumstances neural networks should have two hidden layers and under what circumstances they should have one hidden layer. Networks with one and with two hidden layers were used to approximate numerous test functions. Guidelines were developed from the results of these investigations.

Creative design has been characterized in computational terms as “that design activity which occurs when a new variable is introduced into the design” (Gero, 1994). This is opposed to “routine design,” where “knowledge about variables, objectives expressed in terms of those variables, constraints expressed in terms of those variables, and the processes needed to find values for those variables, are all known a priori.” A third alternative, “innovative design,” occurs when no new variables are introduced, but when one or more variables are used with values outside the usual scope. In computational terms, routine design can also be seen as “search,” (sometimes also called “exploitation”) in so far as a certain, predefined search space is searched for a design solution. For creative design, where the search proceeds outside the boundaries of a predefined search space, the term “exploration” can be used.

In order to enhance the knowledge acquisition capability of the expert system DOMES (Design Of Mechanisms by an Expert System), which is developed at the University of Rhode Island for creative type synthesis of mechanisms, methodologies have been developed to build a knowledge engineer module based on natural language understanding. Specifically, artificial intelligence concepts and Lisp programming techniques have been incorporated in this module to implement the following: (1) analysing and understanding design criteria supplied by human designers in the form of English sentences; and (2) transforming these design criteria into Lisp code and storing them in the knowledge base of DOMES. This knowledge engineer module enhances the capability and improves the performance of the expert system DOMES by providing an effective means for knowledge acquisition based on natural language understanding. The concepts and implementation techniques in developing this module are general and can also be utilized for other knowledge-based systems.

A logic-based approach for automating the processing of design standards is illustrated. This approach is composed of three steps: conceptualization, formalization and implementation. Conceptualization is referred to as the representation of the knowledge necessary for solving the problem of interest in terms of objects and relations. Formalization is referred to as the representation of the objects and relations of interest as axioms using the language of predicate calculus. And, Implementation is referred to as the representation of the axioms of interest and the strategy for manipulating axioms using the constructs of a programming language.

The paper illustrates the logic-based approach to engineering problem-solving automation by considering the portion of the AISC Specification that governs the design of axially loaded members. First, the relations of interest are identified (Conceptualization). Then, predicate calculus is used to formally represent the relations (Formalization) as axioms and to mechanically manipulate them. The checking and designing of structural components via mechanical manipulation of the axioms are illustrated in the paper. Finally, a constraint logic programming language is used to develop a computer program for automatic processing of the specification (Implementation). This program is composed of a set of rules that closely resemble the formulated axioms.

We describe ACDS, an automatic diagnostic system. ACDS is capable of diagnosing faults on analog circuits in dynamic conditions. The circuit's dynamic behavior is studied by means of a series of intrastate simulations during which the qualitative state of the circuit does not change. An acquistion board collects the value of a set of quantities corresponding to accessible test points. These measurements are converted into qualitative values and are used for two purposes: first, to determine the state of the circuit components; second, to trigger the diagnostic procedure whenever a discrepancy between observed and predicted behavior is found. The main difficulty in this phase of measurement interpretation is in obtaining meaningful numerical-qualitative data conversion for values of quantities approaching a boundary between two different qualitative intervals. System performance has been verified through a number of simulations, which have shown the proposed approach to be efficient both in terms of localized faults and of flexibility in adapting to different circuits.

Design is not done in a vacuum. Engineers often rely on prior designs to make new design decisions instead of solving every new problem from scratch. Prior designs that represent good solutions to the tightly coupled nature of mechanical devices are used as guides. Moreover, prior failures are used to avoid repeating old mistakes. In this paper we present a computer-based approach to exploiting the knowledge embodied in prior designs. Reasoning from design cases requires the ability to use cases, or pieces of cases that realize subfunctions of the device being designed. It is, however, difficult to recognize and retrieve relevant cases or case pieces using a given design specification. Because there is no one-to-one correspondence between the desired behavior of a device and the individual component behaviors, it is often not possible to find relevant design cases by using the given overall behavioral specification as an index into case memory. We approach this problem by elaborating the given behavior specification into a description that gives rise to indices with which relevant components can be retrieved. The elaborations are carried out in a behavior-preserving manner using two transformation operators that (a) rely on physical laws if it is known which ones are relevant, or (b) hypothesize behaviors and then search the case memory for ways in which the required behaviors may be achieved. These two approaches are used opportunistically in CADET, a case-based mechanical design system.

Inspection planning is a process of reasoning about inspection activities. As a result, a sequence of inspection actions is formulated, which, when performed, will achieve the desired measurements. In manufacturing, automated inspection technologies, such as Computer-Aided Inspection (CAI) or Co-ordinate Measuring Machines (CMMs), will be facilitated by inspection planning. Inspection planning involves the following four aspects: representation of inspection features; process formalization; modeling of inspection activities; and, finally, plan synthesis. This paper discusses an approach to knowledge-based inspection planning. Accordingly, a prototype inspection planning system has been developed, which is also described in this paper.

A critical problem with existing computer-based geometric modeling is the labor intensive task involved in specifying data input for the description of three-dimensional (3-D) objects. This paper describes a new, 3-D input system aimed at alleviating this problem. It is based on the use of a three-dimensional digitizer for the direct input of spatial coordinates, and an intelligent interactive user interface. The intent of this system is to create a high level, intelligent interface between the designer and a geometric solid modeler which would lighten the designer's burden in performing arduous 3-D geometric description tasks. The user interface is developed around the Knowledge Craft expert system building tool, using the rule-based Carnegie Representation Language OPS5 (CRL-OPS). The system uses schematic networks or frames and production rules to encode knowledge about geometric primitive digitization methods, object feature operators, solid modeler requirements, and input command functions. It also employs a forward-chaining inference strategy to direct the knowledge. This ensures that only a minimal amount of valid data entry is required by the user. However, if excessive data is entered the intelligent interface has the capability to extract the required information. As a result, the solid modeler can automatically create the appropriate object “primitive” or the specific object “feature” upon recognition by the expert system. It will be demonstrated that these capabilities can simplify the 3-D model description process.

Contributions of general design theory (GDT) proposed by Yoshikawa for the development of advanced CAD (computer-aided design) and for innovative design from the research results of a group at the University of Tokyo are illustrated. First, the GDT that formalizes design knowledge based on axiomatic set theory is reviewed. Second, this theoretical result is tested against experimental work on design processes. Although in principle the theoretical results agree with the experimental findings, some problems can be pointed out. From these problems a new design process model, called the refinement model, is established, which has better agreement with the experimental findings. This model implies three guiding principles in developing a future CAD system. One is that future CAD requires a mechanism for physics-centered modeling and multiple model management. Second, a mechanism for function modeling is also required, and the FBS (function-behavior-state) modeling is proposed. Third, intention modeling is also proposed for recording decision-making processes in design. These advanced modeling techniques enable creative, innovative designs. As an example, the design of self-maintenance machines is illustrated. This design example utilizes design knowledge intensively on a knowledge-intensive CAD. This is a new way of engineering and can be called knowledge-intensive engineering. The design of self-maintenance machines is, therefore, an example of knowledge-intensive design of knowledge-intensive products, which demonstrates the power of the design methodology derived from the GDT.

In this paper the need for Intelligent Computer Aided Design (Int.CAD) to jointly support design and learning assistance is introduced. The paper focuses on presenting and exploring the possibility of realizing “learning” assistance in Int.CAD by introducing a new concept called Shared Learning. Shared Learning is proposed to empower CAD tools with more useful learning capabilities than that currently available and thereby provide a stronger interaction of learning between a designer and a computer. “Controlled” computational learning is proposed as a means whereby the Shared Learning concept can be realized. The viability of this new concept is explored by using a system called PERSPECT. PERSPECT is a preliminary numerical design tool aimed at supporting the effective utilization of numerical experiential knowledge in design. After a detailed discussion of PERSPECT's numerical design support, the paper presents the results of an evaluation that focuses on PERSPECT's implementation of “controlled” computational learning and ability to support a designer's need to learn. The paper then discusses PERSPECT's potential as a tool for supporting the Shared Learning concept by explaining how a designer and PERSPECT can jointly learn. There is still much work to be done before the full potential of Shared Learning can be realized. However, the authors do believe that the concept of Shared Learning may hold the key to truly empowering learning in Int.CAD.

In many creative design processes, cross-domain knowledge is required to inspire the new design result. Thus, in knowledge-based design, how we represent the cross-domain knowledge becomes a key issue. In this paper, we present a formalism for design knowledge representation. By analyzing function representation in different design domains, from graphic design and industrial design to architectural and engineering device designs, we find that although the focus of each kind of design is different, the function representation can be generalized into a small number of categories. This formalism can be used in an explorative model of design by analogy, where designs from different design domains are sources to help produce a new design.

Spreadsheets are difficult to use in applications, where only incomplete or inexact data (e.g., intervals) are available-a typical situation in various design and planning tasks. It can be argued that this is due to two fundamental shortcomings of the computational paradigm underlying spreadsheets. First, the distinction between input and output cells has to be fixed before computations. Second, cells may have only exact values. As a result, spread-sheets support the user only with primitive iterative problem solving schemes based on trial-and-error methods. A constraint-based computational paradigm for next generation interval spreadsheets is presented. The scheme makes it possible to exploit incomplete/inexact data (intervals), and it can support problem solving in a top-down fashion. Current spreadsheets constitute a special case of the more general interval constraint spreadsheets proposed.