Abstract
Model-based reasoning is currently a very active area of A.I. research. Explicit models of a physical system are used to provide predictions about the behaviour of that system. However, the derivation of such models is still very much a ‘black art’. This paper attempts to develop some principles which guide the development of appropriate models, and upon which modelling assumptions can be made. A review of the conventional approach to modelling physical systems is given and this is interpreted and extended to include techniques currently being developed within qualitative modelling. The notions of resolution and abstraction are defined and discussed. It is argued that abstraction is the key extension that A.I. brings to modelling complex systems. Two types of complexity of physical systems are identified and these are related to the process of adjusting the model resolution and abstraction. Finally, a case study of a Rotary Cement Kiln is presented to illustrate some of these concepts.
Get full access to this article
View all access options for this article.