Machine learning based fault classification approach for power electronic converters

This research proposes a fault classification approach for power electronics converters (PECs) operating with the distributed generation systems. The proposed approach classifies and localizes the faults in PECs by adapting wavelet transforms and artificial neural networks. Initially, various failure mechanisms for power modules in PECs are identified to generate the fault data. Further, the detailed and approximate coefficients of the fault data at each frequency band are extracted using wavelet transform and used as inputs to the classifier. The ANN classifier estimates the non-linear relationship between the features and targets patterns to develop the fault classification mechanism. The numerical simulations are carried out for injecting various faults and degradation scenarios in both the legs of the inverter and developing the fault classifier in Plecs/MATLAB integration. The results showed 97.4% training accuracy with component failure classifier and 94.2% training accuracy with the classifier trained for 50% degradation in power module.