Machine Learning-Assisted Discovery of High-Performance Perovskite Electrolytes for Intermediate-Temperature Protonic Ceramic Electrochemical Cells

Abstract

Solid oxide electrochemical cells (SOECs) have attracted intensive attention because of their high efficiency and low environmental impact. Much effort has been made to decrease the operating temperature of SOECs down to the intermediate-temperature range. The electrolyte materials with high ionic conductivity are the crucial factor in deciding the operating temperature of SOECs among all components. Proton-conducting perovskite oxides are one of the most promising electrolyte materials for intermediate-temperature SOECs due to their excellent ionic conductivity, usually called protonic ceramic electrochemical cells (PCECs). The most common method to discover the new potential proton-conducting perovskite electrolytes is that the introduced aliovalent dopants facilitate the creation of oxygen vacancy to maintain electroneutrality and give rise to protonic defects when exposed to a reduced atmosphere. In addition to the dopant elements and the dopant content, the total conductivities of doped perovskite electrolytes also depend on the synthesis conditions, the measurement condition and so on. Therefore, the investigation of potential perovskite electrolytes for the intermediate-temperature range should not only be based on the search for the most promising compositions but also requires a detailed study of the influences of its microstructure and processing conditions on the total conductivities. The traditional models of trial-and-error experimental processes limit the development of promising electrolyte materials. Machine learning has emerged as a powerful tool to gain hidden rules from extensive data. By focusing on the investigation of the potential perovskite electrolytes, we collected 400 training data and defined 34 desirable features. Features related to basic elemental properties, crystal structure, microstructure, synthesis conditions, and measurement conditions have been included. Based on these data from the published literature, machine learning models are utilized to predict the total conductivity of perovskite compositions at targeted conditions. Thereafter, feature importance is evaluated to establish a guideline for the rapid and accurate screening of potential perovskite electrolytes in the intermediate-temperature range. Consequently, the screened compositions with high total conductivity from the trained model are prepared, and the experimental results are consistent with the predicted results. This study establishes a new avenue to discover perovskite electrolytes with high total conductivity for intermediate-temperature PCECs. Keywords: Perovskite electrolyte, Proton conducting, Protonic ceramic cells, Machine learning. Figure 1