A generative artificial intelligence model for efficient gas sensitivity prediction in materials without parameters from first principle calculation

Abstract

High gas sensitivity materials are crucial for the development of high-performance gas sensors. Traditionally, obtaining such materials has relied on trial and error, a process that is time-consuming, resource-intensive and laborious. In this study, an intelligent predictive framework using a generative artificial intelligence model is presented, which predicts the gas sensitivity of target materials with remarkable speed and accuracy. To validate the proposed framework, a material database is constructed using ZnO doped with eight different metal atoms, which serves as representative materials. Subsequently, 25 physical parameters, including crystal structure and electronic structure, are calculated using first-principles calculations. Eight intelligent learning models are then trained to predict sensitivity, with the Extremely Randomized Trees model performing the best, achieving a mean square error of 0.02. By combining feature engineering with generative artificial intelligence models, a novel model based on intrinsic atomic features (radius, electronegativity, first ionization energy) is developed, which enables the prediction of CO gas sensitivity in doped ZnO using the generated 12 key parameters, without any parameters from first principle calculation, thus significantly reducing the time and cost associated with discovering gas-sensitive properties. This predictive framework can be easily extended to other materials, facilitating the functional development of materials.