Machine-learning prediction of thermal expansion coefficient for perovskite oxides with experimental validation

Abstract

Perovskite oxides have been of high-interest and relatively well studied over the last 20 years due to their various applications, specifically for solid oxide fuel cells (SOFCs) and solid oxide electrolysis cells (SOECs). One of the key properties for a perovskite to perform well as a component in SOFCs, SOECs, and other high-temperature applications is its thermal expansion coefficient (TEC). The use of machine learning (ML) to predict material properties has greatly increased over the years and has proven to be a very useful tool for materials screening. The process of synthesizing and testing perovskite oxides is laborious and costly, and the use of physics-based models is often highly computationally expensive. Due to the amount of elements able to be accommodated in the ABO₃ structure and the ability for crystallographic mixing in both the A and B-sites, there are a massive amount of possible ABO₃ perovskites. In this paper, a ML model for the prediction of the TECs of AA′BB′O₃ perovskites is produced and applied to millions of potential compositions resulting in reliable TEC predictions for 150 451 of the compositions.