Fault characterization of partial discharge

Abstract

Partial discharge (PD) has been a main concern in electrical insulation degradation and this research work was to establish a method to understand and characterize the PD phenomena for fault diagnosis. For this purpose digital processing techniques are applied to study the clear distinction of the discharge behavior. Signal pre-processing, spectral analysis and clustering method were developed to generate a PD database in order to support the discharge evaluation. The algorithmic design can affect the preprocessing, feature extraction, feature selection and classification of the PD data. Data were obtained form laboratory experiments with test models simulating the PD sources raised to the inception voltage. The stochastic PD signals were measured with sufficient number of sampled points in 20 ms. The individual PD pulse was filtered, extracted and analyzed by digital signal processing techniques.