Review of Denoising Techniques for Noisy Partial Discharge Signals

Low, Chen Yong (2020) Review of Denoising Techniques for Noisy Partial Discharge Signals. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Low Chen Yong.pdf Restricted to Registered users only Download (3MB)

Abstract

Partial discharge (PD) can cause the dielectric strength to weaken and the high voltage system finally breakdown due to occurrences of PD. Therefore, partial discharge measurement gives a warming to the corrosion of the insulation. PD testing also act as an indicator of initial errors. However, there is a difficult and challenging task during PD measurements which is the PD signal always suffers from the strong coupling of exterior noise. In order to eliminate those external noise, various methods of de-noising of partial discharge signal will be reviewed, studied, analysed and evaluated in this paper. All the testing will be conducted by using MATLAB. A better de-noising technique will be selected and implemented based on the current stage of study. Based on the comparative studies, it seems that determining of threshold value for de-noising of PD signals is a difficult and challenging task to be solved. A thresholding procedure plays an important role in de-noising of noisy partial discharge. The selection of threshold coefficients through hard or soft threshold functions should be determined properly in order to achieve the goals which is fast computation speed and great robustness during de-noising process. In this study, four main conventional threshold selection rules based on wavelet transform will be evaluated to select the best threshold coefficient for de-noising purpose. For example, 'rigrsure', 'sqtwolog', 'heursure' and 'minimaxi'. Instead of wavelet transform, various techniques such as Singular Spectral Analysis (SSA), Morphological filter and Power spectral subtraction (PSS) will be implemented also. Same PD and noise model with high level of noise with SNR = –10 dB and low level of noise with SNR = 10 dB will be conducted to evaluate the performance of the proposed methods. PSSD method proved that it is more powerful compared to SSA, MF and existing wavelet-based techniques, it is observed that PSSD yields better MSE and higher SNR according to the denoised PD pulses even in high noise level.