A comprehensive review of machine learning in proton-exchange membrane fuel cells: From materials to intelligent systems

Machine learning (ML) is rapidly transforming proton-exchange membrane fuel cell (PEMFC) science by enabling advances across materials design, multiscale modeling, diagnostics, and intelligent control. Persistent barriers to widespread PEMFC deployment—including high cost, limited durability, and complex multiscale transport phenomena—arise from nonlinear interactions that conventional modeling and experimentation struggle to resolve. This review presents a comprehensive synthesis of ML approaches tailored to the PEMFC value chain, showing how numerical, experimental, imaging, and hybrid datasets support surrogate models that reduce Computational Fluid Dynamics (CFD) computational costs by orders of magnitude while maintaining physical consistency through physics-informed and multi-fidelity learning. ML-guided optimization accelerates the rational design of catalyst layers, membranes, gas-diffusion media, and bipolar plates, uncovering non-intuitive structure–property relationships. Deep-learning-based diagnostics and hybrid residual frameworks significantly improve fault detection, state-of-health estimation, and remaining-useful-life prediction. At the system scale, ML-enhanced model predictive control and deep reinforcement learning enable adaptive, autonomous operation under dynamic conditions, while emerging digital-twin architectures integrate physics-based solvers with real-time data assimilation to deliver predictive, self-correcting fuel-cell management. By unifying methodological advances, cross-scale opportunities, and critical challenges, this review establishes ML as a foundational enabler for the next generation of high-performance, durable, and economically viable PEMFC technologies.