Classification of Vacuum Breakdown During Conditioning Based on Deep Learning

Conditioning is an efficient technology to improve vacuum gap insulation, which is a collection of a series of breakdown events. Each breakdown event contains and be represented by one breakdown waveform with voltage and current. This article aims to classify the vacuum breakdown during the conditioning based on deep learning with its advantages on image feature extraction and recognition. The differences among pulsed current induced breakdown (PB), field emission induced breakdown (FEBD), and particle induced breakdown (PBD) during the prebreakdown period and breakdown numbers during breakdown period in voltage and current could be extracted and recognited by deep learning. Thus, four kinds of breakdown waveforms PB, PBD1 (PBD with one or several breakdowns), PBD2 (PBD with continuous breakdowns), and FEBD are separated for deep learning to recognize and classify. The classification takes about 10 s for each breakdown waveform and its accuracy is above 84.7%. The validation is confirmed from different perspectives.