Fundamental Study on Condition Assessment of Insulating Material using Deep Learning based on Waveform Characteristics of Partial Discharge

The waveform characteristics of the partial discharge (PD) reflect the process of electron avalanche in the discharge space. The authors developed a measurement system for the PD waveforms with the broadest possible frequency band and showed that the waveform characteristics depend on the condition of the insulating material. For example, when a semi-conductive sheet was attached to an insulator to simulate the decrease in surface resistance due to deterioration, it was clarified that shoulders appeared at the rising part of the PD waveform as the surface resistance decreased. Furthermore, the surface resistance of the insulator could be assessed by simulating the change of the rising part of the PD waveform using the equivalent circuit model. In this paper, the authors classified the PD waveforms obtained using samples with different surface resistivity by deep learning based on the waveform characteristics. By appropriately processing PD waveforms with the broadest possible frequency band, it is possible to identify the deterioration condition by deep learning using only the information in the PD waveform.