usingqualitativerepresentationsin.docVIP

Using Qualitative Representations in Controlling Engineering Problem Solving Yusuf Pisan y-pisan@ /~yusuf/ Northwestern University Institute for The Learning Sciences 1890 Maple Ave. STE 300 Evanston, IL 60201-3150 (847) 491-7698 Abstract Engineering problem solving requires both domain knowledge and an understanding of how to apply that knowledge. While much of the recent work in qualitative physics has focused on building reusable domain theories, there has been little attention paid to representing the control knowledge necessary for applying these models. This paper shows how qualitative representations and compositional modeling can be used to create control knowledge for solving engineering problems. This control knowledge includes modeling assumptions, plans and preferences. We describe an implemented system, called TPS (Thermodynamics Problem Solver) that illustrates the utility of these ideas in the domain of engineering thermodynamics. TPS to date has solved over 30 problems, and its solutions are similar to those of experts. We argue that our control vocabulary can be extended to most engineering problem solving domains, and employed in a wide variety of problem solving architectures. Introduction Engineering problem solving requires both domain knowledge and an understanding of how to apply that knowledge. Recent work in qualitative physics has focused on building reusable domain theories, concentrating on the fundamental physical laws of various domains. However, there has been little research on representing the control knowledge necessary for applying these models. Qualitative representations have been successfully used for designing binary distillation plants (Sgouros, 1993) and for designing controllers (Kuipers Shults, 1994). However, since de Kleer’s (1975) original work highlighting how qualitative reasoning was needed to solve physics problems, the task of engineering analysis, as exemplified by textbook engineering problems, has