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

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Science

Background:

• Proton transport in Nafion membranes is vital for proton exchange membrane fuel cell (PEMFC) performance.
• Chemical degradation significantly impacts membrane reliability and longevity.
• Predictive modeling of transport properties under various conditions is needed.

Purpose of the Study:

• Develop interpretable Machine Learning (ML) and Deep Learning (DL) frameworks.
• Predict key transport properties (conductivity, tortuosity) under varying conditions.
• Understand how chemical degradation affects proton transport.

Main Methods:

• Utilized a high-fidelity, multiscale simulation dataset.
• Trained a 3D convolutional neural network on voxelized nanostructure data.
• Applied Gradient-weighted Class Activation Mapping and Shapley-value analysis.
• Developed random forest models with extracted geometric and macroscopic features.

Main Results:

• Accurately predicted vehicular and Grotthuss conductivities and tortuosity.
• Identified key nanostructural regions influencing proton conduction.
• Quantified the impact of nanostructural and environmental factors on transport.
• Constructed a mechanism map linking hydration, temperature, and degradation to conduction regimes.

Conclusions:

• AI frameworks can capture geometry and semantics for materials analysis.
• Nanostructure aging reshapes proton conduction pathways.
• Developed a foundation for degradation-aware optimization and inverse design of membranes.
• Opened new avenues for predictive aging modeling in ion-conducting membranes.