Partial discharge recognition of stator winding insulation based on artificial neural network

Abstract

Partial discharge (PD) reflects insulation condition of high voltage electrical apparatus. Many large electrical machines tend to be subjected to large numbers of load cycles which often shorten the service time of the stator winding. PD accelerates insulation deterioration and there will be more serious PD phenomena at the continual development of insulation fault and before final breakdown of insulation than that at the beginning. PD contains characteristic quantities that can be used to inspect the insulation condition to avoid sudden failure especially for on-line monitoring. In this paper, five types of different physical simultaneous insulation models, which reflect PD in stator windings of large generators, were made. Simulated PD types included surface discharge at endwinding, slot discharge, delamination discharge in three different positions of ground wall insulation. Different levels of voltage were applied to models to obtain different extent level of PD. An artificial neural network (ANN) with backpropagation algorithm was designed to identify the types. The levels of PD extent have also been judged by their distribution. The recognition ability of the ANN was studied. Different types and levels of PD within the winding insulation of large generators were identified with a satisfactory recognition rate.