Residual Faraday Efficiency Enabling Interpretable Data-Driven Optimization of Mass Transport for CO2 Electroreduction

Gas diffusion electrodes (GDEs) are critical for gas-involved electrocatalysis, where the system efficiency hinges on balancing between electrocatalysts and mass transport. While machine learning (ML) has emerged as a powerful tool to search for efficient electrocatalysts, it lacks response variables to describe mass transport effects in GDEs. Here, we propose residual Faradaic efficiency (res-FE), derived by subtracting the potential-dependent mean FE from apparent FE values, to isolate porosity-mediated mass transport effects that are otherwise obscured by kinetic dominance in conventional metrics. Combining computational fluid dynamics simulations, interpretable ML, and multiobjective genetic algorithms, we establish the GDE porosities to CO2 reduction on Ag catalysts. ML interpretability based on res-FE uncovers a uniform distribution of porosities and overpotential─insights unattainable through apparent FE. Our optimizations further identify Pareto-optimal solutions balancing FE, partial current density, and energy efficiency across operational potentials, which reveal distinct porosity thresholds for gas diffusion layers (0.72–0.78) and catalyst layers (0.64–0.66).