Accelerated Screening of Highly Sensitive Gas Sensor Materials for Greenhouse Gases Based on DFT and Machine Learning Methods.

Greenhouse gases (GHGs) have caused great harm to the ecological environment, so it is necessary to screen gas sensor materials for detecting GHGs. In this study, we propose an ideal gas sensor design strategy with high screening efficiency and low cost targeting four typical GHGs (CO, CH, NO, SF). This strategy introduces machine learning (ML) methods based on density functional theory (DFT) to achieve accurate and rapid screening from a large number of candidate gas sensor materials. Specifically, the candidate materials include 28 different transition metal-doped WSe monolayers (TM-WSe), and four gas molecules and their optimal adsorption structures on TM-WSe are constructed. Ten fine-tuned ML models are implemented to train and predict the adsorption energy () and adsorption distance () of target gases on TM-WSe, thereby selecting the optimal ML model and identifying these promising gas sensor materials. In addition, the gas-sensing properties of these materials are verified by band structure, work function, and recovery time. This research provides a reasonable and low-cost new way for rapid screening of ideal gas sensor materials with the help of artificial intelligence and proves its effectiveness through experiments.