Insulation failure detection in transformer winding using cross-correlation technique with ANN and k-NN regression method during impulse test

In this paper, two new schemes are proposed for insulation failure detection in power transformer windings. In the first step, a real high frequency HV transformer winding is modeled based on the detailed model. Thereafter, a simulator is obtained for the insulation failure and is embedded in different locations of the winding in static and dynamic forms. In the first proposed scheme, the obtained ground current signals are used for feature selection based on the cross-correlation technique. Afterwards, a four-layered multiplier perceptron Artificial Neural Network (ANN) is trained using these features for fault detection. The accuracy rate of the ANN network was 84.33% for different faults. In the second scheme, k-Nearest Neighbors (k-NNs) is used as a regression method. The accuracy rate of this method reaches to 80.4%. Clearly, the first proposed scheme is able to detect the faults more accurately. Furthermore, in the first scheme, the accuracy rate was higher for the shunt faults in comparison to the series ones while the second scheme is more successful in detection of series failures.