Short Circuit Location in Transformer Winding Using Deep Learning of Its Frequency Responses

Abstract

A turn-to-turn short circuit fault is One of the most important defects in transformer windings that is most difficult to diagnosis. Degradation decreases impedances of inter-turn insulations that finally may lead to a solid turn-to-turn short circuit. In this paper, early detection of turn-to-turn faults in transformers windings has been studied, in its high-impedance stage, using Convolutional Neural Network (CNN) based on extracting features from frequency response traces. For this purpose, a model winding has been used as test object to approve capability of the proposed approach. A variety of low impedance and high impedance short circuit faults were tested on the model winding. The results show that this method is able to detect turn-to-turn faults in transformer winding even in their early stages.