A Multidimensional Machine Learning-Based Approach to Optimization of Ternary Nanoalloy Catalysts for Oxygen Reduction Reaction

Abstract

The arduous process of designing the optimal Pt-based alloy catalyst poses a great barrier to the mass commercialization of hydrogen fuel cells. Given the vast possibilities of metal combinations and their intricate interactions, the trial-and-error experimental search for the optimal catalyst is practically unfeasible. A major challenge lies in understanding the multimetallic alloy catalysts’ dynamic atomic structure and metal composition under oxygen reduction reaction conditions. We report herein a multidimensional machine learning-based approach to the analysis and optimization of ternary nanoalloy catalysts. The effectiveness of this approach is demonstrated through assessing the relationships between chemical composition, lattice structure, and catalytic performance of the alloy catalysts, which provides a data-driven framework and insight for understanding the dynamic nature of the alloy catalysts. Despite the complex, dynamic, and evolving nature of Pt-based alloy catalysts under the reaction conditions, the results provide some important insights into potential pathways to achieving the optimization of such ternary alloy catalysts. Moreover, the proposed data-driven approach is versatile and can readily be extended to discovery and development of other alloy catalysts, including high-entropy alloy catalysts.