Optimizing Ni-Mo Electrocatalysts With Machine Learning-Guided Autonomous Platform for Next-Generation Alkaline Water Electrolyzers

Active and stable electrocatalysts for water electrolysis are essential to reduce the cost of green hydrogen. Scarcity and cost barriers drive extensive efforts to develop PGM-free catalysts, but matching the activity and stability of PGM-based catalysts remains a great challenge due to the slow and time consuming manual experimentation. Here, we present a closed-loop optimization workflow combining an automated electrocatalyst synthesis and testing system with Bayesian optimization for rapid discovery of nickel-molybdenum electrocatalysts. We find that exploration-focused acquisition strategies are critical under limited experimental budgets, and our results indicate that this is due to compositional parameters that exhibit delayed and conditional influence on performance, causing exploitative strategies to prematurely converge to suboptimal regions of the design space. By varying the exploration-exploitation tradeoff in our experimental campaigns, we identified several distinct regions in the Ni-Mo phase space with exceptionally high performance within limited experimental budgets.