Fault Diagnosis of Power Transformer by Acoustic Signals with Deep Learning

Abstract

Acoustic signals of an operated transformer are continuously produced and closely related to its normal and typical fault conditions, including overload operation, mechanical failures, local overheating and discharge caused by insulation breakdown. Those signals can be easily collected with microphone sensors in substation. To better recognize the proper conditions of power transformer with acoustic signals, this paper presents a fault diagnosis method based on the GFCC voiceprint spectrum and convolution neural network (CNN). Here, the gammatone filterbank is used to match the real features of acoustic signals of transformer to calculate the feature parameters of GFCC for recognition by VGG16 CNN with high accuracy. The acoustic signals of a dry type transformer with the voltage rating of 10kV in load experiment under normal and typical fault conditions are measured and analyzed to verify the correctness of the proposed methods. Calculated results have shown that the recognition accuracy for the different conditions of power transformer is more than 95%.