Processing Fault-Trees by Approximate Reasoning in Solving the Technical Diagnostic Problem(A Possibilistic Diagnostic Decision-Making Method)
IFAC Proceedings Volumes, Jul 1, 1993
Abstract The work may be viewed as a qualitative jump from the use of artificial intelligence (AI... more Abstract The work may be viewed as a qualitative jump from the use of artificial intelligence (AI) techniques to that of fuzzy logic based methods in the design of diagnost expert systems. In response to the classical AI approaches of solving the technical diagnostic problem by forward and backward processing causality propagation through a probabilistic knowledge tree, a fuzzy logic - based approach is proposed which is able to overcome the main disadvantages of the AI approaches. The main steps of the diagnostic methodology are briefly exposed in the following. A-priori a symptoms-faults fuzzy interrelational diagnostic. model is OFF-LINE designed as a basis for the diagnostic reasoning. The acquired diagnostic information is first organized in diagnostic causal networks. These, via a fuzzification procedure are transformed in diagnostic relevant networks (DRN) . By compressing via fuzzy composition the information from the DRNs the diagnostic model is obtained and encoded within the knowledge base as a collection of fuzzy IF-THEN rules. The inference mechanism performes ON-LINE diagnosis in two steps. First by approximate reasoning on the fuzzy diagnostic model with the detected values of the symptoms the model is adapted to the current diagnostic situation. Then, by aggregating its information according to the connectives of the URNs, the adapted diagnostic model is compressed to a vector containing fault possibilities. As fault candIdates the ones with the maximal fault possibilities are considered by being validated via generalized modus tollens. The proposed diagnostic methodology is developed on a very simple example for a DC motor.
Uploads
Papers by Mihaela Ulieru