Applications, progress and technical challenges of machine learning in full lifetime of proton exchange membrane fuel cells

Proton exchange membrane fuel cells (PEMFCs), as core devices for hydrogen energy utilization, play an increasingly strategic role in transportation electrification, distributed power generation, and large-scale energy storage. However, the inherent strong multi-physics coupling, complex nonlinear degradation, and high sensitivity to operating conditions of PEMFC impose significant challenges on its cost, durability, and efficiency. Machine learning (ML) provides new opportunities to address these challenges and accelerate intelligent PEMFC development. This review examines ML applications across the full lifetime of PEMFCs, covering material design, manufacturing and quality control, system state management and control, and digital twin (DT) development. Beyond summarizing recent progress, it critically compares ML approaches in terms of data requirements, interpretability, computational cost, robustness, generalization, and uncertainty quantification, while highlighting emerging physics-informed ML, hybrid physics-data-driven models, self-adaptive DTs, and deep reinforcement learning (DRL). Three major challenges are identified: (1) data scarcity, heterogeneity, and insufficient standardization; (2) limited physical consistency, interpretability, generalization, and uncertainty quantification; and (3) computational, real-time, and safe-deployment requirements. Overall, integrating data, physical knowledge, and intelligent decision-making can advance ML-enabled PEMFCs from isolated prediction and optimization toward reliable, adaptive, and full-lifetime intelligent operation.