Explainable Machine Learning and Deep Learning Models for Understanding Operational and Degradation Effects in Nafion Membranes

Understanding how chemical degradation influences proton transport in Nafion membranes is critical for improving the reliability of proton exchange membrane fuel cells (PEMFCs). Here, we develop interpretable Machine Learning (ML) and Deep Learning (DL) frameworks to predict key transport properties, including vehicular and Grotthuss mechanism conductivities, and the tortuosity factor under varying water content, temperature, and degradation levels. Leveraging a high-fidelity, multiscale simulation data set, we trained a 3D convolutional neural network on voxelized membrane nanostructure data to predict proton conduction properties. Gradient-weighted class activation mapping is applied to highlight the spatial regions within the nanostructure that most strongly influence the predictions. We further develop random forest models based on extracted geometric features and macroscopic input conditions, with Shapley-value analysis used to quantify the impact of nanostructural and environmental factors. A fully coupled mechanism map is constructed to link hydration, temperature, and degradation with distinct conduction regimes. This visualization reveals how nanostructure aging reshapes proton conduction and provides a foundation for inverse design, which enables degradation-aware optimization of the membrane. Our study provides a multimodal AI framework capable of capturing both geometry and semantics in materials analysis and opens new avenues for predictive aging modeling and intelligent design in ion-conducting membranes.