Neural network based classification of partial discharge in HV motors

Abstract

This paper discusses the general application of using Neural Networks (NN) to classify six different types of Partial Discharge (PD). Stator winding failures contribute about 30–40% of the total motor failures according to IEEE and EPRI. Ninety percent (90%) of electrical failures on High-Voltage (HV) equipment are related to insulation deterioration. Large datasets were collected for motors with PD defects as well as PD-free machines. The datasets of PD were pre-processed and prepared for use with a NN using statistical means. It was possible to utilise the advantages offered by multiple NN models to classify the PD defects with a maximum recognition rate of 94.5% achieved, whereas previous research work did not exceed a classification accuracy of 79%.