Classification of Partial Discharge Patterns in Rotating Electrical Machines Using Machine Learning

This paper presents a machine learning approach for the automated classification of partial discharge images recorded on rotating electrical machines. It introduces a classification of different partial discharge patterns for phase resolved partial discharge plots. A major problem in the use of machine learning is the large amount of training data needed for a powerful classifier. An approach is carried out, comprises a realistic partial discharge patterns mentioning off-line and online measurement with a strong focus on electrical machines. This paper shows how to handle the leak of training data by an augmentation of training data, using a few images extracted from the International Electrotechnical Commission (IEC) standards, and discusses convolutional neural network classifier. The classifier will be trained by augmented data and will be tested for different fault conditions in electrical machines. The error conditions that can be applied to the algorithm are shown, and the limitations of this method are discussed.