Air-Discharge Decomposition Gases Detected by Sensor Array with Depthwise Separable Convolution

When electrical equipment suffers from a discharge fault, the air insulation medium produces characteristic decomposition products represented by NO₂ and CO, and the composition and content of the discharge decomposition products are closely related to the severity of the discharge fault. To more accurately establish the response relationship between air discharge decomposition products and electrical equipment faults, this paper tests the concentration and composition of gaseous characteristic decomposition products during air discharge based on a self-designed high-performance gas sensor array. A dataset with 8 different gas mixture concentration ratios was prepared. Based on the Depthwise Separable Convolution, a lightweight neural network algorithm for classifying feature gas concentrations is constructed, realizing high-precision recognition of air discharge fault severity based on sensor signals, with an accuracy of up to 100% on this dataset. By comparing the recognition performance of this network with several classical machine learning models, it can be found that the network model proposed in this paper has higher recognition accuracy and faster response speed, and has broad application prospects in the field of air discharge fault detection of electrical equipment.