Partial discharge identification by using signal processing techniques

Abstract

Partial Discharge (PD) detection after denoising, characterization and identification are the three main signal processing requirements of PD analysis. Voluminous digital PD data are nowadays readily available with constant improvements in PD measurement techniques. Power Engineers may be able to detect prominent PDs using oscilloscope and existing couplers. But identification of the types of developing and random occurring PD is a real challenge to any practicing engineer. In this thesis, details on using wavelet transform in the form of either continuous wavelet transform or discrete wavelet transform with two methods to denoise, identify the location of PD and retrieve PD wave shape without magnitude distortion are presented. To identify the type of PD, some experimental studies and about six existing and developed signal processing methods are carried out. Laboratory experimental study provided reproducible data with enough number of sampled points on three types of pure PD and one multisources PD. The signal processing is done on PD random occurrence in 20 ms with multicycles, and is known as group PD analysis. In single PD analysis, the individual shape of PD is extracted and analysed. Group analysis includes the study of following distributions of PD: Φ-q, Φ-n and q-n, and the corresponding statistical operators Sk, Ku and CC, weibull parameters using qcumulative n and Φ-cumulative n distributions, cluster analysis using (Δt-ΔV), (ΔV n-ΔV n-1), (Δt n-Δt n-1) and (Δ(ΔV)-Δ(Δt)) distributions and Wavelet Transform