Investigations on the Suitability of Modern Denoising Methods for Impulse Fault Current Waveforms

Noise removal is pertinent for conditional monitoring and reliable signal processing. The study investigates the effectiveness of seven denoising methods and assesses the comparative performance between thresholding based time-frequency denoising using Wavelet Transform, Wavelet Synchrosqueezing Transform, Fourier Synchrosqueezing Transform, time domain supervised denoising comprising Gaussian Process Regression, Convolutional Autoencoder, Long Short Term Memory (LSTM) Network, and a hybrid time-frequency supervised learning technique. Impulse fault current signals of a transformer is considered for analysis. The results illustrate the ability of each method to reconstruct the denoised version of noise corrupted fault signal with high degree of accuracy.