Adaptive Bayesian Denoising for General Gaussian Distributed Partial Discharge Signals in 11 kV Stator Coil using Wavelet Transform

Abstract

Noise removal from Partial discharge signal which is corrupted with Gaussian white noise, using Wavelet Transform (WT) based Bayes estimation is presented.Such corrupted partial discharge signal is considered to exhibit Generalized Gaussian Distribution (GGD). Denoising of partial discharge (PD) signals is one of the prime pre-processing stages which help in analyzing the effect of PD on insulation by studying its characteristics and features. The pre-and post-processing of PD signal can be performed significantly using wavelet transform. The WT based adaptive thresholding and coefficients modeling methods for denoising provides a better chance to reduce interferences. The estimation of BayesShrink based estimation of the denoised signal is performed from its noisy wavelet coefficients for the signals with Gaussian distribution. The BayesShrink threshold is independent of the coefficient distribution and is a function of noise and noiseless signal variance. BayesShrink is the most effective thresholding method providing threshold value by reducing the Bayes risk. The coefficient estimation based on Bayes method for GGD signals performs well in removing white noise from the corrupted PD signal. The PD signals are collected by conducting the experiment on damaged 11 kV stator coil which has mica as insulation. The collected signal is processed using WT based Bayes estimation. The result indicate that the optimum Bayes estimator with GGD behave similar to soft thresholding method.