A hybrid fuzzy/neural system used to extract heuristic knowledge from a fault detection problem

Abstract

Neural net have proven to be capable of solving the motor monitoring and fault detection problem using an inexpensive, reliable and noninvasive procedure. The neural net, unfortunately, cannot provide heuristic knowledge about the motor or the fault detection process. This paper introduces a novel hybrid fuzzy/neural fault detector that uses the learning capabilities of the neural net to detect if a motor has an incipient fault. Once the fuzzy/neural fault detector is trained, heuristic knowledge about the motor and the fault detection process can also be extracted. With better understanding of the heuristics through the use of fuzzy rules and fuzzy membership functions, we can have a better understanding of the fault detection process of the system; thus we can design better motor protection systems. The electric motors in industry are exposed to a wide variety of environments and conditions. These factors, coupled with the natural aging process of any machine, make the motor subject to incipient faults. These incipient faults, left undetected, contribute to the degradation and eventual failure of the motors. With proper monitoring and fault detection schemes, the incipient faults can be detected; thus maintenance and down-time expenses can be reduced while also improving safety. In this paper, motor bearing faults in single-phase induction motors are used to illustrate this novel system. This illustration demonstrates the successful training of a hybrid fuzzy/neural system that can provide accurate fault detection, and gives the heuristic reasoning for the fault detection procedure. >