The automation of physical system modeling: Modeling strategies and AI implementation.

Abstract

System modeling plays a key role in various engineering activities including system analysis, performance evaluation, control and design sensitivity studies. However, there are no tools available to assist engineers in creating dynamic models. The modeling, analysis and simulation programs that are commercially available have no or little modeling ability. Users of these programs are required to make almost all the model assumptions before preparing the program input. Modeling has always been considered a more artful than scientific activity. However, the author argues that the modeling process is more structured than is commonly believed. The objective of this thesis is to identify the basic structure of the general modeling process and to propose several strategies that could serve as the basis for an artificial intelligence (A.I.) based computer environment for automating the modeling process. The identified modeling structure and the proposed strategies for automated modeling include: (1) Decomposing, the process of transforming real systems into a collection of objects or components and assigning them "motion" attributes. This can be accomplished with a user-interactive dialog with the computer. (2) Mapping, the process of associating physical objects (components) with ideal phenomena (basic physics). This is a totally automated strategy based on associating energy phenomena with each identified motion of the components. (3) Structured Modeling, a systematic and thus totally automated process of lumping the energy phenomena, defining state variables and assembling the model based on motion constraints. (4) Modifying, a heuristic-based strategy, for altering the Decomposing and Structured Modeling processes to alter a model which failed to meet the modeling specifications. An example is presented to illustrate these strategies. Based on these proposed strategies, a prototype of a COMputer Modeling Assistant (COMMA) is implemented as a production system using LISP on a APPOLO workstation., and its capabilities is demonstrated by means of examples. It is concluded that the artful process of modeling has more structure than is conventionally assumed and that this process can be largely automated. The proposed strategies for automated modeling, as part of an automated modeling system, should facilitate novice and expert modelers to develop proper models (models with necessary yet sufficient complexity) more efficiently and increase the efficacy of using system models for design.