Separation and Classification of Corona Discharges Under Low Pressures Based on Deep Learning Method

With the growing concern toward the global warming crisis, the electrification of commercial aircraft is targeted to reduce greenhouse gas emissions from the aviation industry. However, the environment that an aircraft operates in provides significant design challenges. Moreover, the technologies that enhance the power density of the powertrain (such as higher voltage levels and wide bandgap devices) lead to severe tension on the insulation systems. The combination of harsh environmental conditions and insulation-threatening technologies raises concern about the reliability of electrical equipment, such as power generators, motors, and cables. Since the failure of the insulation system translates into the failure of the entire equipment, it is crucial to investigate the behavior of discharge sources under low-pressure conditions. In this regard, this study develops a dense convolutional neural network (DenseNet) model based on experimental data to separate and classify various sources of corona discharge under low-pressure conditions. The results show that DenseNet models can achieve high accuracy within a reasonable training time. The accurate detection and classification of discharge sources provide the backbone of a dielectric online condition monitoring system (DOCMS) that can actively monitor the health of electrical equipment in an electric aircraft.

Automated summary

This study develops a DenseNet model based on experimental data to separate and classify various sources of corona discharge under low-pressure conditions, providing the foundation for a dielectric online condition monitoring system.