Machine Learning-Guided Design of Perovskite Oxides for High-Temperature Oxygen Sensing

Reliable high-temperature oxygen sensors are crucial for demanding applications like combustion, steel production, and petrochemical refining, yet identifying stable, high-performance materials for such environments remains challenging. We use machine learning (ML) to accelerate the discovery of next-generation perovskite oxide sensors. The workflow used combines data from the Materials Project and published datasets. A gradient boosting classifier was used to identify stable perovskite candidates from over 180 million potential doped ABO₃ compositions (F1 score: 0.95). An XGBoost regression model was developed to predict the oxygen partial pressure-dependent conductivity profiles in the temperature range from 300-1000 °C and for pressures up to 1 atm (R²=0.99). A web-based tool was also developed to rank the most sensitive materials, with top candidates showing a three order-of-magnitude change in conductivity. We validated the high-ranking candidates against the literature and the results of density functional theory calculations for the oxygen vacancy migration mechanisms. This ML-guided approach significantly accelerates the design of advanced high-temperature oxygen sensors.