Data-driven modeling and optimization of anode-side structural, transport and electrochemical properties in PEM water electrolyzers

Proton exchange membrane water electrolyzers are intricate multiphysics systems, whose commercial scale up and viability depend on optimizing the structural/material design of their cell components. Some of these structural properties also influence reactivity of the catalyst layers. The sluggish kinetics on anode side is of particular interest as the main water splitting reaction occurs on this part of the electrolyzer. Unlike PEM fuel cells, the public availability of experimental datasets is still limited. Therefore, in this work we undertake a data-driven approach using synthetic data generated from physics-based numerical model to investigate the effects of important structural, transport and electrochemical properties on cell performance. To facilitate model training through dimensionality reduction, Shapley additive explanations based feature selection procedure is adapted. Two separate artificial neural network models were trained on the reduced dataset to predict cell performance and voltage losses. Parametric analysis shows that based on the selected input properties, the ionomer volume fraction and absolute permeability of the porous transport layer are the important structural properties which influence cell performance. Voltage loss decomposition analysis corroborates these observations. Finally, the surrogate model was employed as the objective function within the genetic algorithm to determine the optimal input feature values. The optimized structural, transport and electrochemical properties demonstrated an improved cell performance when compared to the selected baseline values.