Deep-learning-assisted insights into molecular transport in heterogeneous electrolyte films on electrodes

Abstract

Mass transfer in electrolyte films on electrodes is crucial to the performance of electrochemical energy devices, which is difficult or impossible to observe experimentally. Here, we develop a framework utilizing deep learning to analyze vast molecular dynamics (MD) data to reveal the molecular-level transport properties in electrolyte films. This framework contains physical feature analysis and selection based on MD simulations, surrogate model training, structure-transport relationship analysis, and structure discovery. This framework is then applied to explore oxygen transport in fuel cells, which allows the transport properties and their relationships to the structural characteristics of electrolyte films to be revealed, and thus, the critical features limiting oxygen transport are identified. Accordingly, increasing the catalyst surface hydrophilicity and suppressing the electrolyte film density fluctuation are favorable for oxygen transport. Moreover, this framework is transferable to revealing similar molecular-level transport phenomena in electrolyte films that widely exist in other electrochemical energy devices.