Recognition of partial discharge in stator winding models based on 3-dimensional pattern using artificial neural network

Abstract

Partial discharge (PD) is an important phenomenon and has a close relation to the insulation condition of electrical apparatus. Usually PD accelerates insulation deterioration and before final breakdown, their activities will be much more serious than that in ordinary operation. Therefore PD is an adequate characteristic quantity for the inspection of insulation condition in order to avoid sudden failures, especially for online monitoring. An artificial neural network (ANN) group with the backpropagation algorithm was developed to identify the types and extent of PDs. Six different physical models, which could reflect PDs in the stator windings of large electrical machines, were made. Simulated PD types included surface discharges at end windings, slot discharges, delamination in three different positions of ground wall insulation as well as a standard PD level of new machines. Different levels of voltage were applied to models to obtain various extents of PD activities. The fingerprints of experimental PD data were extracted with the /spl phi/-q-n 3-dimensional pattern. The recognition ability of the ANN group was investigated. Different types and extents of discharge within the winding insulation of large electrical machines were identified with a satisfactory recognition rate.