Computational Insights into PEMFC Durability: Degradation Mechanisms, Interfacial Chemistry, and the Emerging Role of Machine Learning Potentials

TL;DR

This review discusses recent computational modeling advances in understanding PEMFC degradation mechanisms, emphasizing the coupling of chemical, mechanical, and electrochemical failure pathways and the potential of machine learning potentials.

Contribution

It highlights the integration of multiscale computational methods and machine learning to better understand and predict PEMFC durability issues.

Findings

Degradation pathways are interconnected through feedback loops.

Current models do not capture the coupling of multiple degradation mechanisms.

Machine learning potentials offer promising avenues for multiscale modeling.

Abstract

Proton exchange membrane fuel cells (PEMFCs) are a promising clean energy technology, offering high efficiency and near-zero operational emissions for stationery and automotive applications. However, their widespread adoption remains limited by insufficient durability, driven by the degradation of the catalyst layer and proton exchange membrane under realistic operating conditions. While the macroscopic consequences of degradation are well established experimentally, the atomistic and molecular mechanisms that initiate and propagate failure remain incompletely understood. This review synthesizes recent advances in computational modelling, spanning density functional theory, molecular dynamics, and emerging machine learning potentials, to examine how chemical, mechanical, electrochemical, and contamination driven degradation mechanisms operate across multiple length and time scales. Key topics include radical-induced membrane degradation, platinum dissolution and carbon support corrosion, mechanical fatigue under electrical and hygrothermal cycling, and the impact of ionic and gaseous contaminants. A central finding is that these degradation pathways are not independent, but form strongly coupled feedback loops that no existing computational framework has been designed to capture this coupling simultaneously. Future directions are proposed, with emphasis on multiscale modelling frameworks and the application of machine learning interatomic potentials to the electrified interface.