Test Point Optimization for Model-based Fault Diagnosis Expert System

Abstract

Expert system, as the basis for other fault diagnosis methods, can take advantages of the expert domain knowledge and intuitive rule-based reasoning model. However, when test points of a faulty system are limited, combinatorial explosion problem of minimal diagnosis is caused by the use of model-based fault diagnosis expert system. In this paper, we develop a method to gradually reduce the minimal diagnosis by adding the system test points to realize fault location. Firstly, the computing procedure is formalized by combining set enumeration tree (SE-tree) with closed nodes to generate all the minimal hitting sets (i.e., minimal diagnosis). Then, as fault diagnosis synthetic information quantity and correlation matrix are introduced, we show that with progressive decomposition of the matrix, test point optimization strategy can be found out. Finally, using new observations and removal rules, minimal diagnosis can be gradually reduced, until the only minimal diagnosis is retained. The proposed test point optimization strategy for model-based fault diagnosis expert system can be applied to different engineering applications and its effectiveness is demonstrated with an example.