Leveraging convolutional neural networks for enhancing performance of Cs3Cu2I5/TiO2 nanocrystal-based carbon monoxide gas sensor

Abstract

Carbon monoxide is a colorless, odorless gas that can cause irreversible effects on the brain if inhaled in excess. Accurate measurement of carbon monoxide is very important for human health. The current carbon monoxide sensors work with long response time and narrow detection range. Therefore, in this work, TiO₂ was coated on perovskite nanocrystals Cs₃Cu₂I₅ by solution method to obtain high performance carbon monoxide gas sensitive material Cs₃Cu₂I₅/TiO₂ nanocrystals. Moreover, high precision and anti-interference measurement of carbon monoxide was realized by combining with machine learning. The carbon monoxide sensor based on Cs₃Cu₂I₅/TiO₂ has a short response/recovery time of 3.8/17.5 s and a sensitivity of 0.36 at 10 ppm. Then, an intelligent classification algorithm is used to quickly identify the concentration of carbon monoxide gas, and the recognition accuracy is as high as 99.5 %, which is higher than that of traditional sensitivity determination of gas concentration methods, due to the fact that machine learning involves multiple features of the electrical response curve, rather than the single feature of traditional sensitivity determination methods. Finally, the sensor was integrated into the self-rescuer for coal mine to realize the intelligent opening in the presence of CO, and improved the rescue efficiency. We believe that this sensor will be promising in the field of carbon monoxide, and the idea of using machine learning to intelligently recognize gas concentrations can be extended to other gas sensors.