Model-Based Diagnostic Reasoning and its Application to Network Fault Diagnosis

Abstract

This paper describes a diagnostic method based on models of discrete event driven systems. Diagonosis is performed with generate-and-test strategy. The method has the following three advantages. (1) Diagnosis systems can be made by only inputting system models. (2) Verification is easy, because only a model is necessary to be checked. (3) Performance of diagnosis system depends entirely on the minuteness of models.
We showed that the proposed method had generality beyond one domain by applying the proposed method to two different domains, one network fault diagnosis and the other fault diagnosis of numeric calculation circuits. But two problems remained unsolved: The modeling method was not defined completely, and the speed of diagnosis was slow.
In this paper, a modeling method and improvement of calculation speed are reported. As for a modeling method, the structure is modeled with object-oriented technique, and the function is modeled with transition networks that explain state transition of components. Transition networks are generated by combining structure data with rules that describe causal relations of state transitions. Creating transition networks in advance improves calculation speed, because interpretation of causal rules is not needed during diagnosis.
Application to network fault diagnosis is also explained. Two models whose description levels are different are created, and it is shown that diagnosis results depend only on the model.