New Breakdown Diagnosis Method based on Electrical Measurement and Deep Learning Recognition for Multiple Electrodes in Vacuum

Vacuum breakdown mechanism evolution during the conditioning is important for vacuum interrupter insulation improvement. Traditionally, the breakdown mechanism can only be obtained after the conditioning, which requires the manual operation and classification step by step. It takes about 10 hours for one electrode pair with about 400 breakdowns. This paper proposes a new breakdown diagnosis method based on electrical measurement and deep learning recognition in conditioning. The voltage and current are measured and processed to be simplified breakdown waveforms, and the breakdown mechanism is classified through deep learning with simplified breakdown waveforms. The whole process of breakdown diagnosis is automatic with no manual operation. The results show that the new method accuracy has a close relationship with the train samples (above 77.81% at least), and it can diagnose breakdown during the conditioning, which takes about 5 seconds for one breakdown. The new method can make the breakdown diagnosis efficiency, real time and high accuracy.