Artificial Intelligence for Proton Exchange Membrane Fuel Cells Multiphysics Modeling and Optimization: Current Status and Future Directions

Hydrogen is a promising energy carrier for improving power grid stability and enhancing energy storage systems. Proton Exchange Membrane Fuel Cells (PEMFCs) are electrochemical converters used to produce electricity from green hydrogen. Therefore, PEMFCs have been used in several industrial applications covering the transportation sector, aerospace applications, and distributed electrical energy generation. However, the multidisciplinary characteristics of PEMFCs make modelling and efficiency analysis a difficult task, resulting in limitations in the wider deployment of this new electricity generator. PEMFCs are characterized by strongly coupled transport, thermal, electrochemical, and degradation phenomena. High-performance computational machines have been used to investigate and analyze the degradation phenomena in PEMFCs. Despite this, this approach provides a detailed physical understanding of degradation levels, but the costly repeated simulation results in limitations in the utilization of this approach. Artificial intelligence (AI) and machine learning (ML) have been used to overcome such problems. Indeed, AI and ML have become powerful tools for the modelling and analysis of multidisciplinary systems due to their ability to predict results based on experimental results, sensor outputs, images and numerical simulations. In this context, this paper sheds light on the applications of AI, ML and digital twins in the field of modelling, design and degradation phenomena analysis of PEMFCs. The main phenomena taking place in PEMFCs, including reactant transport, gas–liquid flow, gas diffusion layer anisotropy, water management, heat transfer, transient behavior, and scale bridging and their effects on PEMFC performance degradation, are analyzed. Artificial neural networks, convolutional and recurrent architectures, Gaussian-process and ensemble methods, physics-informed neural networks (PINNs), Bayesian optimization, metaheuristics, reinforcement learning, and hybrid digital-twin frameworks have different performances and they can be used for improving PEMFC modelling and phenomena analysis. Particular attention is given to the integration of physics-based and AI-driven models for PEMFC performance prediction, degradation assessment, optimization, and digital twin development. The review critically discusses existing applications, emerging approaches, and their limitations regarding experimental validation, uncertainty quantification, data availability, and generalization under varying operating conditions. Finally, future research directions are identified, emphasizing physics-informed modeling, degradation-aware digital twins, and reliable AI-assisted optimization and control.