The malfunction of air-insulated power equipment can produce a series of gas decomposition products, represented by NO2 and CO. However, the severe cross-sensitivity of sensing materials toward these two gases is a puzzle that restricts the development of fault diagnosis. This work has fabricated a micro-sensor array loading with four gas-sensing materials (WO3, NiO, TiO2, SnO2 doped with SnCl2-In2O3), and a portable gas detection device has also been implemented towards NO2 and CO, which was flexible for several detection scenarios (direct connection, calibration, and sampling mode). We have acquired the response curves of the portable device towards 50 ppm NO2 (5870.53%) and 50 ppm CO (-8.12%). First, the principal component analysis (PCA) has verified the classification ability of NO2 and CO mixtures. Then, four machine learning algorithms (KNN, SVM, RF, and NBM) were utilized to improve the selectivity towards NO2. The accuracy of four methods was 99.63%, 99.62%, 99.23%, and 98.86, respectively. Ultimately, the portable device with NVIDIA Jetson nano demonstrated great potential in the rapid diagnosis of the air-discharge fault. This work proposed a portable detection system for air discharge gas decomposition with high selectivity, which allowed to recognize the concentration of NO2 accurately under the interference of CO.

暂无