Wavelet Packet Denoising of Partial Discharge Data

Abstract

On-site partial discharge (PD) measurements are affected by external noise and disturbances of various nature such as interferences from radio transmissions, stochastic noise, steep front pulses from power electronics, etc. In such a hostile electromagnetic environment, the partial discharge can be completely submerged by the surrounding signals and the sensitivity and reliability of the measurement can be strongly affected. Once the signal has been recorded, a post-processing tool is needed in order to recover the PD pulses. In this paper a wavelet based technique has been used in order to extract the PD data from a noisy environment: the Wavelet Packet Transform (WPT) has been adopted, which allows a more detailed analysis of the signal with respect to the Discrete Time Wavelet Transform (DTWT). As a first step, the post-processing tool has been applied to a typical uncorrupted PD pulse which has been decomposed up to its deepest level; subsequently, an energy-based criterion in conjunction with hard-thresholding has been used in order to select a suitable mother wavelet and an adequate decomposition level; useful patterns have then been collected for the reconstruction of the signal with minimum shape distortion. Finally, the suggestions provided have been used for the extraction of numerical simulating signals, characterised by a very low signal to noise ratio (SNR), reproducing a PD pulse corrupted by external interferences and noise, in order to show the effectiveness of the proposed method.