Cross-correlation-aided Fuzzy c-Means for Classification of Dynamic Faults in Transformer Winding During Impulse Testing

An approach for the classification of dynamic faults in transformer winding insulation during impulse testing based on the features extracted from the cross-correlation sequence is proposed. The features for fault classification are extracted from the cross-correlation sequence of impulse fault currents acquired from healthy, as well as faulted, conditions of transformer winding. The efficacy of the extracted features for identification of dynamic insulation failures are assessed by using two classification tools—the fuzzy c-means algorithm and the artificial neural network. The tools have successfully identified dynamic faults with different fault establishment times along with the location of the failure in the winding. Impulse fault currents for various faults having different fault establishment times are acquired by simulating static and dynamic insulation failure in an analog model of 33-kV winding of a 3-MVA transformer using the developed non-linear, bilateral fault simulator. Results show that the extracted features are capable of identifying the impulse fault patterns with acceptable accuracy. The concepts of cross-correlation feature extraction, simulation of dynamic faults, and fault classification methods are explained.