A novel method for the recognition of PD patterns by neural network

Partial Discharge (PD) measurements are generally considered as a tool to detect defects in the insulation of HV components. In recent years, a pursued goal has been the identification of defects for diagnostic purposes. This could be done through the analysis of a set of PD data and then using neural networks. In this paper a novel defect identification method is presented. Starting from PD data detected on a large family of specimens, the method aims at the evaluation of a suitable set of parameters to be used as input variables of a small neural network, able to identify the probable defect inside the insulation. With this aim the statistical Weibull analysis is performed on PD pulse amplitude histograms to obtain the scale parameter α and the shape parameter β. Then, two fractal parameters (fractal dimension and lacunarity) and two statistical operators (Skewness and Kurtosis) are evaluated from the PD phase histogram and from the 3-dimension histogram respectively. Tests of the above method performed either on samples containing artificial defects built in the laboratory or on natural defects in HV components are then reported. Comments are consequently put forward