Partial Discharge Classification in Insulation Material via Discrete Wavelet Transform Noise Reduction and Artificial Neural Network

Insulation material degradation stemming from partial discharge (PD) is a major factor in power equipment failure. External electromagnetic interference introduces noise, compromising the accuracy of PD diagnosis, thus, addressing partial discharge noise suppression is crucial. This study examines four PD types which include corona, floating electrode, surface, and void discharge measured from insulation materials. PD signals were combined with Additive White Gaussian Noise (AWGN) and treated with Discrete Wavelet Transform (DWT) for noise reduction. Six features including root mean square error, variance, standard deviation, waveform factor, kurtosis, and skewness were extracted from denoised signals to classify PD types using Artificial Neural Network (ANN) with 10-fold cross-validation. Classification accuracy was assessed and compared. Results indicate effective AWGN removal by DWT, achieving up to 90% classification accuracy for PD defects post-noise reduction using ANN.