Neural network based incipient fault detection of induction motors

A pattern recognition technique based on artificial neural networks (ANNs) is playing a significant role in identifying the incipient faults of induction motors. Requirements of the pattern recognition algorithm to detect these faults are that it should not only show high accuracy to determine the extent of the fault, but also it must report if it can not identify a particular fault so that preventive steps can be taken in recognizing the undetected faults and updating the underlying neural network in the shortest possible time. A system based on the popular feedforward neural network (FNN) suffers a problem in satisfying these requirements of pattern recognition. In this paper a new pattern recognition scheme based on the ART2 neural network is proposed to detect the incipient faults of induction motors. The design, implementation and dynamic updating of this type of system are illustrated with an example