Acoustic Diagnosis of Partial Discharges in Transformers

Abstract

Partial discharge in the transformer may generate noise. In view of this characteristic, it is proposed to apply the audio detection method to the diagnosis of partial discharge. Discharge faults in the transformer were simulated, and a method for diagnosis of these faults based on audio is studied. Three models for typical partial discharges are designed, namely the plate electrode discharge, the pin plate electrode discharge and the surface discharge. The audio produced by these discharges was recorded and hence an abnormal audio database was built. Using wavelet packet algorithm, audio features of partial discharge are extracted, and then the neural network algorithm is used to distinguish one kind of discharges from others successfully.