De-noising of time-domain spectroscopy data for reliable assessment of power transformer insulation

Abstract

Polarisation-depolarisation current (PDC) measurement and its analysis is a popular technique for assessing the condition of transformer insulation. Owing to the low magnitude of PDC, recording noise-free PDC data from in-situ power transformers is a challenge. Once the relaxation current data get affected by noise, it becomes difficult to formulate insulation model (as recorded data loses its characteristic shape). This further makes the data difficult to analyse and predict insulation condition. In this study, two de-noising techniques are discussed (one is based on Wavelet Transform while the other is based on Stockwell Transform) for eliminating low-frequency non-stationary noise from recorded PDC data. Comparison between these two techniques suggests de-noising using Stockwell Transform is advantageous over wavelet analysis. The proposed methodology is first tested on data recorded from the sample prepared in the laboratory and then on data measured from reallife in-service power transformer.