Fault detection and fuzzy rule extraction in AC motors by a neuro-fuzzy ART-based system

This paper introduces a system for fault detection and classification in AC motors based on soft computing. The kernel of the system is a neuro-fuzzy system, FasArt (Fuzzy Adaptive System ART-based), that permits the detection of a fault if it is in progress and its classification, with very low detection and diagnosis times that allow decisions to be made, avoiding definitive damage or failure when possible. The system has been tested on an AC motor in which 15 nondestructive fault types were generated, achieving a high level of detection and classification. The knowledge stored in the neuro-fuzzy system has been extracted by a fuzzy rule set with an acceptable degree of interpretability and without incoherency amongst the extracted rules.