Denoising of acoustic partial discharge signals corrupted with random noise

Power transformers are one of the most important and expensive electrical equipment that require online condition monitoring. Partial discharge (PD) measurement is considered the most effective and non-destructive approach to observe the condition of power transformers in service. However, many sources of noise interfere with the captured PD signals leading to waveform deformation. Thus, it is important for the detection and classification of PD signals to initially suppress the noise encountered with PD measurement. In this paper, we investigate a method, named power spectral subtraction denoising (PSSD) that uses fast Fourier transform to restrain the random noise encountered in measured acoustic PD signals. The denoising performance of PSSD is compared with those of wavelet-based denoising techniques in addition to the mathematical morphological filter. The denoising techniques are first examined on PD signals contaminated with low and high levels of simulated random noise. The denoising evaluation metrics show the superiority of PSSD over the other techniques. Moreover, a modified PSSD (M-PSSD) method is presented to address the actual PD signals corrupted with real random noise. High reduction in noise levels are achieved using M-PSSD.