An optimal algorithm for applying wavelet transform in identifying the arrival time of PD pulse in a UHF detection system

The detection and localisation of Partial Discharge (PD) activities is gaining importance in maintaining operational transformers and diagnosing those which are faulty. Ultra-high-frequency (UHF) techniques have been applied successfully to detection of PD activities in HV transformers because of their sensitivity to fast transient pulses. When an UHF PD monitoring system is applied, one of the crucial pieces of information that need to be preserved is the arrival time which allows the Triangulation algorithm to be applied for localisation of PD source(s). Unfortunately, the presence of noise in the UHF signals can complicate the application. This paper presents the results of a study which has been carried out as a joint effort between Glasgow Caledonian University and University of Strathclyde to investigate the efficacy of applying Discrete Wavelets Transform (DWT) to the extraction of PD pulses from noisy environments and the determination of their arrival times in UHF signals. Studies on practically acquired UHF signals from an in-service power transformer prove that DWT is effective in locating the arrival times of PD pulses when the standard deviation (SD) of added noise is no greater than the maximum magnitude of the PD pulses frequency domain [4]. However, the method is more complicated than the conventional filter technique, and requires expertise and experience to be applied to produce its best effect. In this paper, an effective DWTbased denoising algorithm is proposed to remove noise in UHF signals while preserving the arrival time in order to locate the PD source accurately. The paper firstly introduces the application of UHF method to the measurement of PD pulses in transformers and gives typical UHF responses to PD activities. Then a DWT based algorithm is proposed to extract the UHF pulses associated with PD activities from noisy data, including the selection of mother wavelet, the thresholding algorithm and the reconstruction of the signals. Finally, the effectiveness of the proposed method is evaluated for various levels of noise presence prior to discussions of the shortcomings and limitation of the proposed method