Partial Discharge Signal Denoising Using Morphological Component Analysis

Abstract

On-site partial discharge (PD) measurement is an important tool to monitor the insulation conditions of electrical equipment. PD signal processing techniques have been evolving in recent years and, notably, techniques based on overcomplete dictionaries and sparse representations have achieved relevant results for PD signal filtering. A new PD denoising approach, based on Morphological Component Analysis (MCA), is presented in this paper. MCA aims to separate the PD pulse from noise, which are superimposed in measured PD signals, using overcomplete dictionaries, sparse representations and signal's prior information. The method was tested on synthetic signals containing amplitude modulated (AM), impulsive and Gaussian noise. These are commonly found in measured PD signals. It was also tested on on-site measured PD signals. MCA achieved efficient results in PD signal denoising when the noise amplitude is greater than the PD pulse amplitude for the impulsive and AM noise.