Recognition of single and multiple partial discharge sources in transformers based on ultra‐high frequency signals

Abstract

Partial discharge (PD) is a symptom of insulation defect or degradation in high‐voltage equipment. Thus, PD detection is an important diagnostic tool. Furthermore in practical situations, the PD can be generated from a single or multiple sources. Being able to detect and classify such PD events will help to determine the necessary corrective action to prevent insulation breakdown. To demonstrate, three different simulated discharge conditions in transformers were investigated: void, floating metal and their combination. The PD signals were captured using an ultra‐high frequency (UHF) sensor and denoised using wavelet transform method by application of Matlab wavelet multi‐variate denoising tool. Two types of mother wavelet, that is, db and sym, were applied to decompose the signals and extract the signal features in terms of their skewness, kurtosis and energy. These features were then used as input to train a neural network to analyse and determine the PD source type. Results show this technique is able to classify and recognise single and multiple PD source types with a high degree of success.