A fuzzy expert system for facility diagnosis of rotating machines

This paper proposes a system for diagnosing faults in rotating machines based on spectra method and production rules for crisp and fuzzy inferences. The knowledge of skilled engineers for diagnosing rotating machines is represented by a syndrome matrix, and production rules for both crisp and fuzzy inferences are composed based on the matrix. Membership functions for fuzzy rules are determined by using the knowledge of engineers for diagnosing rotating machines. In the method, spectrum bandwidths are introduced to allow the variations of spectra for individual machines and/or errors in measurement. A set of candidate of faults is obtained after crisp inferences based on production rules for them. Then, the confidence factor of every member in the set is determined by fuzzy inference. Faults in rotating machines can be diagnosed even by unexperienced engineers using the candidate faults and their confidence factors.