A modeling technique for generating causal explanations of physical systems

This paper introduces an approach for physical system modeling that gives a basis for generating causal explanations of system behaviors. We explicitly represent knowledge about structures and functions of physical systems in two levels of abstraction: domain and prime models. Domain models explicitly represent the actual structures and processes that make up particular systems, whereas prime models explicitly represent the abstract normal and abnormal structures and processes underlying certain classes of physical systems with analogous behavior. Each domain model is viewed as an instance of a particular prime model and thus during the reasoning about the domain model, any causal behavior of a corresponding prime system model can be inherited. Our approach differs from other systems that generate causal explanation for tutorial purposes in two ways. First it generates a causal explanation of a particular physical system’s behavior from general knowledge of causal mechanisms derived from appropriate physical principles rather than a specific causal knowledge about the particular system. Second, unlike the systems that generate causal explanation from qualitative simulation, our approach explicitly represents system structures and processes, instead of sets of qualitative constraints, and thus can directly generate causal explanations for behaviors of the system structures and processes with no additional interpretation required. This paper discusses the approach with an illustrated example and an ongoing research in extending this technique when dealing with multiple granularities. The limitations and other advantages of this approach are described.