A Comparative Study of Denoising Techniques for UHF Signals from Partial Discharge

In this work, a comparative study of three typical denoising techniques for UHF signals from Partial Discharge (PD) sources is carried out: Wavelet Transform (WT), Variational Mode Decomposition (VMD), and Multiscale PCA (MSPCA). First, the techniques are applied to synthetic signals created from mathematical models to establish the best operating parameters. Subsequently, an experimental set-up is carried out to generate PDs and thus UHF signals, ensuring an environment with white noise and periodic type interference signals. The results, for both synthetic and real UHF signals, indicate that MSPCA is superior to WT and VMD. However, in the case of real UHF signals, MSPCA is observed to distort the waveform to such an extent that the wavefront is attenuated. The VMD technique also has a distortion effect on the signal, although not as pronounced. On the other hand, WT exhibits less noise mitigation, but less signal distortion than VMD and MSPCA. From this study, we conclude that before applying a denoising technique, it is important to analyze its effect on the waveform and not only take performance criteria into account. If this is not considered, downstream tasks such as classification, identification, and localization that extract features from waveforms may be affected.