A design can be considered as simpler than another for a variety of reasons. Simplification, as a consequence, can be done in a variety of ways. We are studying simplification by Analogy with stored simplifications. The designs are represented as Function, plus Behavior, plus Structure. Simplifications may occur at each of these levels, with consequences for the other levels of the representation. Experiments are being made with a prototype system that can simpify designs by analogy.
Single Function Agents
The concept of Single Function Agents (SiFAs) has been developed to investigate design problem-solving using multi-agent architectures. SiFas provide elementary functionality specific to design tasks. With this approach we intend to explore elementary patterns of interaction, communication and conflict resolution (especially trade-offs between goals). As the agents are small, and restricted to a limited number of types, we expect to be able reveal critical phenomena in multi-agent design problem-solving that may have been obscured in systems with larger agents.
Learning in Multi-Agent Design Systems
The goal of this research is to investigate multi-agent learning paradigms relevant for design problem-solving. Agents can either learn or be learned about. The specific learning models are intended to be evaluated in the context of the Single Function Agent paradigm, with a particular emphasis on parametric design problems. The research will identify dimensions which are relevant for a description of learning in multi-agent systems; focus those aspects of multi-agent systems that will support design problem-solving; identify the need for improvements in such systems -- such as the improvement of the design process or the quality of the final design; and define, implement and test small multi-agent learning situations to assess their impact.
Documentation of a design normally consists of a description of the final design itself: effectively a "snapshot" of just the final decisions. Capturing the design rationale, the alternatives considered while designing and the reasons for accepting or rejecting them, offers a richer view into both the product and the decision making process. This information could be invaluable as an aid in revising, maintaining, documenting, evaluating, and learning the design.
Assembly of Design Problem-Solving Fragments
This research is about assembling fragments of design problem-solving knowledge into a complete design expert system for a class of (mostly parametric) configuration design problems. The fragments are extracted from design problem-solving knowledge written in the DSPL language. The assembly process is guided by a problem decomposition produced by a heuristic problem decomposition system, KDD, that was produced by previous Ph.D. research. The decomposition specifies a hierarchy of groups and sub-groups of design decisions, along with a suggested order in which they should be decided.
Methodology Discovery in Multidisciplinary Design
This research addresses the problem of producing a design methodology for the design of engineered objects, where the design process requires the use of knowledge from multiple disciplines. A methodology dictates how to order and coordinate design tasks. In many cases such a methodology does not exist. The theoretical basis for methodology production in multidisciplinary design is weak and therefore there are many issues to be explored. The approach is to decompose the available multidisciplinary knowledge into appropriate pieces. The pieces are inserted into a computer program framework that will use each piece at an appropriate time while designing. The resulting trace is an instance of a methodology. These instances will be compared and abstracted into a general methodology. Robot design is selected as an initial test area of multidisciplinary design for this research.
Intelligent Interfaces (with Prof. Wills)
The problem is that of automatically generating the presentation of some given information about a designed object. In particular, we have been interested in the presentation of information about computer networks. We assume that a presentation will be instantiated in the form of a display, on a given device. Different tasks may require different kinds and amounts of information, while different pieces of information require different presentations. When generating a presentation the user's task, the user's preferences, general and domain specific human-computer interaction principles, and the capabilities of the device will determine which displays will be appropriate for the presentation and which one of these will the best possible choice. A demonstration system has been developed that treats the generation of presentations as a design task, dividing the process into functional design, embodiment design, configuration design, and detailed design.
webmaster at cs.wpi.edu / Thu Jul 15 17:56:24 EDT 1999