Wavelet Selection for Analysis of Acoustic Signals Emitted by Partial Discharges in Oil Insulation

This paper presents and discusses wavelet selection for investigation on acoustic signals (AE) emitted by partial discharges (PD) in transformer mineral oil. The main aim of this research is to indicate the optimal wavelet regarding the accuracy of PD classification. Three scenarios are used to simulate different AE signal sources, including: point-plate PD source, air bubbles in oil and point-plate PD source with air bubbles in oil. All of the measurements were conducted under las conditions. AE signals emitted by each scenario were registered and processed using discreet wavelet transform (DWT) based on six different wavelet families. Finally, a support vector machine (SVM) classifier was used to classify the captured signals. As a result the highest accuracy was noted for the symlet wavelet. The presented results may be used to enhance the PD classification process using the AE method and potentially improve the diagnostics of PD in power transformers.