Transformer Inter-Turn Fault Diagnosis Using Continuous Wavelet Transforms and Convolutional Neural Networks

Abstract

This paper proposes a new technique to detect transformer inter-turn faults using Continuous Wavelet Transforms (CWT) and a Convolutional Neural Network (CNN). This paper proposes a novel technique that uses a single artificial intelligence model for both fault localization and severity prediction of transformer inter-turn faults. To create the necessary training data, a 630 kVA 10.5/0.4 kV step-down transformer was simulated in Simulink, generating primary and secondary line current waveforms. The CWT decomposed these waveforms for the CNN model. The CNN model was able to predict the health status, affected transformer side, affected transformer phase, and percentage of turns affected with 100%, 99.9%, 99.8%, and 98.6% accuracy respectively. This new technique shows great potential for improving transformer inter-turn fault detection.