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.
[1]
John F. Sowa,et al.
Conceptual Structures: Information Processing in Mind and Machine
,
1983
.
[2]
Benjamin Kuipers,et al.
Qualitative Simulation as Causal Explanation
,
1987,
IEEE Transactions on Systems, Man, and Cybernetics.
[3]
Rattikorn Hewett.
Aggregation in model-based reasoning using prime models: a preliminary report
,
1992,
Proceedings ICCI `92: Fourth International Conference on Computing and Information.
[4]
Michael R. Genesereth,et al.
The Use of Design Descriptions in Automated Diagnosis
,
1984,
Artif. Intell..
[5]
Randall Davis,et al.
Reasoning from First Principles in Electronic Troubleshooting
,
1983,
Int. J. Man Mach. Stud..
[6]
Barbara Hayes-Roth,et al.
Representing and Reasoning about Physical Systems Using Prime Models
,
1991,
Principles of Semantic Networks.
[7]
William J. Clancey,et al.
Heuristic Classification
,
1986,
Artif. Intell..
[8]
Johan de Kleer,et al.
A Qualitative Physics Based on Confluences
,
1984,
Artif. Intell..
[9]
Gordon I. McCalla,et al.
A Computational Framework for Granularity and its Application to Educational Diagnosis
,
1989,
IJCAI.
[10]
P. Johnson-Laird.
Mental models
,
1989
.
[11]
James D. Hollan,et al.
STEAMER: An Interactive Inspectable Simulation-Based Training System
,
1984,
AI Mag..
[12]
Brian Falkenhainer,et al.
Setting up Large-Scale Qualitative Models
,
1988,
AAAI.
[13]
Kenneth D. Forbus.
Qualitative Process Theory
,
1984,
Artif. Intell..