Machine Learning Based Inter-Turn Short Circuit Detection for Three-Phase Power Transformers with Primary Side Currents

A novel method is proposed in this paper for inter-turn short circuit detection in three-phase power transformers. This detection method only requires primary side currents and can determine the fault severity and locate the faulty phase(s). Phase shift information is extracted out of primary side currents and can act as an effective fault indicator. Machine learning techniques are adopted for feature generation and decision region separation. Continuous degradation monitoring is feasible with the proposed approach. To improve the robustness and accuracy of the detection algorithm, Gaussian mixture model is introduced from the perspective of probabilistic machine learning. Both simulation and experiment results are provided to validate the algorithm.