Accelerating Discovery of WX2-Based Gas Sensors (X = S, Se, and Te) for C4F7N Decomposed Species via Density Functional Theory and Machine Learning

Abstract C4F7N decomposition products such as HF, CO, C2F3N and C2F4 pose serious risks to the operation status of C4F7N-based insulation devices in the electrical system. This method, which combines machine learning (ML) with density functional theory (DFT) calculations, proposes a novel and efficient method for accelerating screening of transition metal (TM)-doped WX2 (X = S, Se, and Te) as potential sensing materials for such four typical gases. By establishing input feature descriptors and conducting training and optimization of eight machine learning models, the optimal models for predicting two crucial adsorption and sensing parameters, namely, adsorption energy (Eads) and bandgap modulation (ΔBg) are determined with high accuracy. In addition, the sensing response of the selected 8 materials is further analyzed to illustrate their potential for sensing typical gas species. This work not only accelerates the discovery of WX2-based sensing materials upon C4F7N decomposed species but also lays a foundation for the rational design of advanced gas sensors, typically realizing the insulation evaluation in the power system.