Accelerating the Development of Oxygen Electrodes for Solid Oxide Cells Using Two-Stage Machine Learning

Xueyu Hu,Yucun Zhou,Zheyu Luo,Meilin Liu
DOI: https://doi.org/10.1149/ma2024-01372254mtgabs
2024-08-09
ECS Meeting Abstracts
Abstract:Reversible Solid Oxide Cells (RSOCs) stand out as highly efficient and cost-effective devices for energy storage and conversion. However, the pursuit of high-performance and durable RSOCs is impeded by the lack of efficient and stable oxygen electrodes. The conventional trial-and-error approach used for the development of oxygen electrodes is resources and time-consuming. In this presentation, we introduce a novel "two-stage machine learning" approach designed to expedite the discovery of optimal oxygen electrodes for efficient oxygen reduction/evolution reaction (ORR/OER) in SOCs. In the initial stage, we employed a high-throughput calculation approach to generate four ORR/OER related descriptors (energy above the convex hull, p-band center, d-band center, and vacancy formation energy) for a set of 4455 perovskite oxide candidates. Subsequently, we successfully predicted these descriptors using machine learning techniques based on the fundamental physico-chemical features of the candidates. Moving to the second stage, we measured the polarization resistances - directly related to ORR/OER activity - of 142 oxygen electrodes and employed them as prediction targets. Through our machine learning method, we accurately predicted experimental results based on the fundamental physico-chemical features of the candidate materials. Notably, among the 4455 candidates studied, a specific oxygen electrode material is predicted to exhibit both high activity and stability. When applied to an RSOC based on a proton-conducting electrolyte operated at 600 °C, a high peak power density of >1.5 W cm −2 is demonstrated in the fuel cell mode, and a high current density of >2.5 A cm −2 is achieved at a cell voltage of 1.3 V in the water electrolysis mode while maintaining stable operation for 500 hours.
What problem does this paper attempt to address?