Partial Discharge Characterization of Ceramic Power Electronics Circuit Carriers Assisted by Machine Learning

This paper presents an approach for transferring knowledge about partial discharges in polymer insulators to ceramic insulators with the aid of machine learning. It is shown how various machine-learnable features can be generated from partial discharge measurement data and processed in varying artificial neural networks for classification. It is found that polymer-based partial discharges can be classified using this method. In addition, the Long Short-Term Memory based artificial neural network enables partial discharge cause finding and thus fault detection in ceramic power electronics substrates.