Recognition of partial discharge patterns in motors using neural network

A Neural Network approach using back-propagation algorithm is proposed to identify different partial discharge patterns in this paper. With the mapping characteristics embedded in NN, the typical discharge models and their corresponding characteristic parameters are interlinked by the NN. With the aid of a PC diagnosis system, the character parameters are extracted out from the figures that describe the phrase character of the discharge pulses. The result shows that the NN is capable of recognizing different discharge patterns in motors, such as surface discharge, inter discharge, slot discharge and ending discharge. Also the delta U model is studied to present different PDs and their discharge theories, because it can give the information of the variance of continues discharge voltage.