Optimizing transformer fault detection: an investigation into current signal feature extraction

Abstract

Identifying faults is a crucial element in the realm of preventive maintenance and the condition monitoring of transformers. For fault detection of transformers many different conventional or advanced techniques such as short circuit impedance measurement, vibration and sound analysis, frequency response analysis (FRA), dissolved gas analysis and machine learning or deep learning have been used. Offline methods of fault detection are being experimented since faults can be detected at the earliest stages, the detection process does not disrupt power supply. By using feature extraction of the fault current waveform, the performance of the fault detection algorithm can be improved, and the accuracy of fault discrimination can be increased. The purpose of this study is to evaluate the use of feature extraction of current in fault transformers using wavelet transform in order to enhance the effectiveness of the fault detection in transformers. A simulated PSCAD model derived using lumped parameter network in [1] is used for the generation of different types of faults and obtaining their fault current waveforms for feature extraction