Robust detection of acoustic partial discharge signals in noisy environments

Abstract

Partial discharge (PD) can be used to predict insulation failures in power transformers. Accurate detection of particular PD types has a significant role in anticipating forthcoming outages. However, the noise encountered with PD measurements negatively affects the detection accuracy. In this paper, we propose a robust PD detection technique that is immune to noise through efficient frequency domain-based feature extraction from acoustic emission signals. The PD spectrum is first obtained using Fourier transform and then, the low frequency band of 0.05-0.15MHz is used as a representative feature vector. Finally, four different classifiers are used to examine the PD detection accuracy. Experimental results on a benchmark dataset verify the robustness of the proposed method for PD detection, as it achieves 100% classification accuracy for clean PD signals and up to 99.62% for noisy PD signals.