AI-enabled fault diagnosis for hydrogen fuel cells: recent advances, challenges, and future perspectives

Ensuring the operational reliability and safety of hydrogen fuel cells has become a critical research focus. The inherently coupled multi-physics nature of fuel cells gives rise to diverse and nonlinear fault mechanisms, presenting substantial challenges for conventional model-based diagnostic approaches. In recent years, artificial intelligence (AI) has emerged as a compelling alternative, owing to its strong capabilities in data-driven pattern recognition and adaptive learning. In this study, typical faults and diagnosis execution process in fuel cell systems were systematically reviewed. At the core, unlike existing reviews that primarily classify methods by algorithmic categories or application scenarios, we systematically evaluate the applicability and limitations of representative AI techniques at each diagnostic stage from data acquisition to control adaptation. Finally, the key challenges that hinder real-world deployment are discussed, and emerging research trends are prospected. This review's contribution lies in offering a new perspective by connecting diagnostic workflow requirements with AI capabilities, which reframes how researchers should match techniques to fuel cell system challenges.