Application of Supervised and Unsupervised Neural Networks for Broken Rotor Bar Detection in Induction Motors

This paper describes the application of automatic diagnosis procedures for the detection of broken bars in squirrel cage induction machines based on neural network (NN) classifiers. On the ground of representative data of the motor condition, obtained through an appropriate processing of experimental measures, NNs are effectively employed for discriminating healthy and faulty motors and providing a indication of the fault level. Both supervised and unsupervised training algorithms for NNs are used to evaluate their suitability to this kind of task. Two different diagnosis techniques, consisting in analyzing stator currents during start up and stator voltages after supply disconnection respectively, have been experimented to provide suitable input data to the NN for the fault detection. Differently from other diagnosis techniques, they possess the distinctive feature of being insensitive to load conditions. Experimental diagnosis results show the noticeable potentialities of the proposed automatic diagnosis approach that is able to identify the rotor fault in the early stages