Data-Driven Engineering of MnCoMo Electrocatalysts: Predictive Modeling and Experimental Validation for the Oxygen Evolution Reaction

Abstract The development of sustainable green hydrogen through the use of catalytic processes demands a very active and earth-abundant electrocatalyst capable of facilitating the kinetic difficulties of the oxygen evolution reaction. Conventional trial-and-error and solely computational approaches suffer from high resource expenditures and inefficiencies in converting theoretical results into practical lab results. In order to address this issue, a closed-loop, data-driven experimental framework is proposed in order to facilitate the automated and expedited optimization process of transition metal electrocatalysts. Four experimentally measured MnCoMo oxide-based electrodes containing 0, 0.25, 0.5, and 0.75 mmol Mo were used as the baseline dataset for machine-learning model development and evaluation across interpolation and extrapolation scenarios. The fully unmeasured 1.0 mmol Mo composition was then withheld from model development, predicted blindly from the four baseline profiles, synthesized, and experimentally characterized for prospective validation. Gradient-boosted tree and deep learning neural network models are tested against a variety of physical synthesis descriptors within both interpolation and extrapolation scenarios. The physical synthesis and experimental verification of an out-of-sample concentration of the electrocatalyst show high accuracy in prediction, thus demonstrating the capability of encoding algorithmic models in order to avoid unnecessary manual screening efforts within the investigated MnCoMo compositional space.