Discharge image reconstruction and frequency domain analysis based on event data

Abstract

Optical image method has been the earliest and most used direct method for observing gas discharge. Currently, research on gas discharge monitoring based on visible light mainly relies on high-speed cameras, but the large size, significant data storage requirements, and susceptibility to interference from complex backgrounds and lighting conditions limit their further application. Dynamic vision sensing (DVS) technology is a neuromorphic sensing technology that asynchronously measures the luminance changes at each pixel. It offers advantages such as a large dynamic range (>120 dB), high temporal resolution (up to 1 µs), and small data volume (MB level). In this study, dynamic vision sensing technology was employed to monitor both 30 mm short-gaps and 1080 mm long-gaps discharge processes simultaneously. This study developed the CountImage encoding method for event data and conducted image reconstruction, time-domain analysis, and frequency-domain characteristic analysis based on the event data. The results show that the event-reconstructed images are highly consistent with the high-speed camera images, and the arc development process and its path can also be clearly observed. Additionally, this study discovered a correlation between the electrical characteristics and event information during the discharge process. In the time domain, the duration of the maximum DVS event count closely matches the duration during which the voltage drops to zero during flashover. In the frequency domain, the Pearson correlation coefficient between the event stream spectrum and the voltage signal spectrum is greater than 0.95. Both the maximum number of brightening events (ON_max) and the maximum number of darkening events (OFF_max) are positively correlated with the voltage applied between the electrodes. This study demonstrates that, compared to the GB/s data rate of high-speed cameras, this approach can record the discharge process and accurately reconstruct the discharge process, arc morphology, and discharge path at MB/s data rates, while also adapting to changes in brightness without the need for exposure adjustment. Additionally, there is a positive correlation between the frequency-domain characteristics of the event data and the voltage characteristics. These results indicate that dynamic vision sensing holds promise as a replacement for high-speed cameras in laboratory discharge observations and could be even effectively applied to discharge monitoring in electrical equipment in real grid.