Modeling Membrane Degradation in PEM Electrolyzers with Physics-Informed Neural Networks

A. Polo-Molina, J. Portela, L.A. Herrero, R. Cicero González

Resumen:

This study introduces the first application of Physics-Informed Neural Networks (PINNs) to model membrane degradation in proton exchange membrane (PEM) electrolyzers, which are essential for sustainable hydrogen production. Traditional physics-based models offer physical interpretability but rely on numerous parameters that are difficult to measure, while data-driven models like machine learning provide flexibility but often lack generalizability and consistency with physical laws. The proposed PINN framework bridges this gap by integrating two ordinary differential equations: one describing membrane thinning through a first-order degradation law, and another modeling the time evolution of cell voltage due to degradation. The results show that the PINN effectively captures long-term degradation dynamics using limited and noisy data, while preserving physical meaning. This hybrid modeling approach provides a robust and accurate tool for understanding and predicting membrane degradation in PEM electrolyzers. It offers a promising foundation for improved diagnostics and performance optimization in electrochemical systems subjected to aging and reliability challenges.

Resumen divulgativo:

Este estudio desarrolla un nuevo modelo basado en PINNs para predecir la degradación de membranas en electrolizadores tipo PEM. La herramienta permite anticipar el deterioro con pocos datos, conservando la coherencia con las leyes físicas, ofreciendo un modelado robusto.

Palabras clave: Physics-Informed Neural Networks, PEM Electrolyzers, PEM Modelling, Membrane Degradation Modelling, Machine Learning

Fecha de Registro: 07-jul-2025