A modified probabilistic neural network-based algorithm for detecting turn faults in induction machines

Abstract

In this paper, an intelligent technique for detecting the turn fault and supply voltage unbalances in an induction machine is proposed under varying load conditions. The proposed probabilistic neural network (PNN) technique is based on well-established statistical principles rather than heuristic approaches that are adopted in the multi-layer perceptrons. The PNN is derived from the Bayes decision strategy and nonparametric kernel-based estimators of probability density functions. The machine model is developed for inter-turn short circuit fault and it is compared with its a priori models for its robustness in detecting the fault at incipient stages under various operating conditions of the machine. This algorithm is proposed for first time and training, testing, and validation results using the method appear promising and easily realizable in the industries.