Automated Physical Modeling

The creation of abstract models of physical systems is an important AI research area. I describe a program which can automatically construct such models for machines like mechanical clocks or watches. The program finds an appropriate set of state variables and determines how they change as time passes. The abstract model of the mechanical device may be used to numerically simulate its behavior. My program uses short, controlled simulations to identify repet itive behavior patterns which can be used for long-term behavior prediction. 1 Introduction Reasoning about physical systems generally involves the creation and manipulation of abstract models, normally constructed by the person studying the system. 1 will discuss the automatic generation of such models, as well as their manipulation and analysis. The physical systems I will focus on are mechanical devices. An abstract model of a physical system is usually based on a set of state variables. A set of particular values for the state variables represents a particular state of the system. The model must also include a description how the state variables are related and how they change. The values taken by state variables may be either numerical quantities or qualitative symbols. The abstract models used in engineering and the physical sciences normally use state variables which take numerical values. Much of the work by artificial intelligence researchers on reasoning about physical systems has focused on models in which state variables take only qualitative values. Qualitative models of physical systems are often useful for reasoning in situations where the information available about the system is limited or imprecise. However, many physical systems cannot be adequately represented by qualitative models. Geometry is especially difficult to deal with qualitatively. Geometry plays a central role in the behavior of mechanical devices, so they cannot be described by purely qualitative models, although work has been done with mixed quantitative/qualitative models for such machines. In this paper I will focus on models with numerical state variables. It is often assumed that such quantitative models are not relevant to artificial intelligence because 1. We already know how to use quantitative models. Open problems are highly technical and of interest only to mathematicians. 2. The analysis of quantitative models doesn't yield results at the level needed for artificial intelligence. 3. People reason about physical situations qualitatively. In fact, quantitative models are quite worthy of consideration by AI researchers. Let us consider the objections listed …