Accurate partial discharge classification from acoustic emission signals

Abstract

Accurate partial discharge (PD) classification provides significant information to asses power transformers' insulation condition. The work presented in this paper aims to improve classification from acoustic emission signals for oil-paper insulated systems. Three different types of PDs are considered; surface discharge, PD from a sharp point to ground electrode, and PD from parallel plates. The PD types are simulated with aged insulation material (oil/paper), large tank size, and high surrounding noise level. The signals collected from each PD type are preprocessed using Continuous Wavelet Transform. The preprocessed signals are compressed using zonal coding applied over Direct Cosine Transform coefficients to create the feature vectors for classification, where a feed-forward with back-propagation trained neural network is utilized. The results indicates high recognition rate for classifying the different PD types using the proposed method.