Fault Identification Based on PD Ultrasonic Signal Using RNN, DNN and CNN

Traditional fault identification algorithms have low recognition accuracy in practical engineering applications. In this paper, new fault identification methods are proposed and proved to be more excellent. The new methods are based on: 1) Downsample the partial discharge (PD) ultrasonic signal to obtain sound signal that human can hear; 2) Extract features from sound data based on methods including Mel Frequencies Cepstral Coefficient (MFCC) to extract expanding signal characteristics; 3) Use Recurrent Neural Network (RNN), Deep Neural Network (DNN) and Convolutional Neural Network (CNN) classifiers on each feature extraction method and compare the effects and performances.