An Improved Methodological Approach for Denoising of Partial Discharge Data by the Wavelet Transform
Partial Discharge (PD) measurements may be affected by external noise and disturbances of various natures such as interference from broadcasting stations, stochastic noise, pulses from power electronics, etc. Extracting PD pulses from such a noisy environment is therefore a crucial issue. This paper presents a wavelet based technique for automatic noise rejection. The core of the paper is the use of an improved methodological approach for the selection of a suitable wavelet, which aims at summing up the benefits and overcoming some limitations of previous techniques. Firstly, a very wide set of training signals is used for the identification of the decomposition level and for the calculation of suitable performance parameters that identify each wavelet; then a Performance Fingerprint is introduced in order to summarize the ability of a specific wavelet to reconstruct a partial discharge waveform, and a distance criterion is used for the selection of the most suitable wavelet. Afterwards, useful information is collected for the reconstruction of the PD signal, and finally, results on the application of the algorithm for a set of numerical and experimental signals are presented.