Partial discharge (PD) diagnostics is considered a major and effective tool for the monitoring of insulating conditions of power cables. As such, a large amount of off-line or online PD measurements have been deployed in power cables during the past decades. However, challenges still exist in PD diagnostics for power cables. Noise is one of the challenges involved in PD measurement. This thesis develops new algorithms based on the characteristics of both PD signals and noise to improve the effectiveness of wavelet-based PD denoising. In the meantime, it presents new findings in the application of empirical mode decomposition (EMD) in PD denoising. Wavelet-based technique has received high attention in the area of PD denoising, it still faces challenges, however, in wavelet selection, decomposition scale determination, and noise estimation. It is therefore the first area of interest in this thesis to improve the effectiveness of existing wavelet-based technique in PD detection by incorporating proposed algorithms. These new algorithms were developed based on the difference of entropy between transformed PD signals and noise, and the sparsity of transformed PD signals corrupted by noise. One concern commonly expressed by critics of wavelet-based technique is a pre-defined wavelet is applied in wavelet-based technique. EMD is an algorithm that can decompose a signal based on the signal itself. Thus, the second area of interest in this thesis is to further investigate the application of EMD in PD denoising; a technique that does not require the selection of a pre-defined signal to represent the “unknown” signal of interest. A new method for relative mode selection (RMS) was proposed based on the entropy of each intrinsic mode function (IMF). Although this new method cannot outperform the existing ones, it reveals that RMS is not as important as claimed in the application of EMD in signal denoising. Also, PD signals, especially those with lower magnitudes,can receive serious distortion through EMD-based denoising. Finally, comparisons between wavelet-based and EMD-based denoising were implemented in the following aspects, i.e., executing time, distortion, effectiveness, adaptivity and robustness. Results unveil that improved wavelet-based technique is more preferable as it can present better performance in PD denoising.
|Date of Award||28 Feb 2020|
- University Of Strathclyde
|Sponsors||University of Strathclyde|
|Supervisor||Wah Hoon Siew (Supervisor) & John Soraghan (Supervisor)|