Machine learning for fuel cell remaining useful life prediction: A review

Accurate prediction of the Remaining Useful Life (RUL) of fuel cell (FC) systems is essential to ensure operational reliability, optimize maintenance strategies, and extend system lifetime in safety-critical hydrogen applications. As FC degradation is governed by complex, nonlinear, and stochastic mechanisms, machine learning (ML) has emerged as a powerful paradigm for data-driven prognostics. This paper presents a structured and comprehensive review of recent ML-based approaches for FC RUL estimation, encompassing supervised, unsupervised, and hybrid methodologies, including regression techniques, support vector machines, ensemble models, neural networks, and advanced deep learning architectures. Despite notable progress, our analysis reveals persistent limitations in the current literature, particularly the widespread neglect of underlying electrochemical and physical degradation laws, as well as the scarcity and ambiguity of explicit RUL and End-of-Life (EoL) labels in publicly available datasets. These challenges significantly constrain model generalization, interpretability, and real-world applicability. To address these gaps, we conduct a comparative analysis of more than 20 recent state-of-the-art studies and propose a unified and generalizable RUL estimation pipeline. This framework integrates data acquisition, preprocessing, feature engineering, model design, and validation, while explicitly accounting for physical consistency and operational constraints. In addition, the paper formulates practical, multi-level recommendations, including first-order guidelines for data modeling and learning strategies, second-order recommendations targeting validation protocols and real-world deployment, and the systematic integration of uncertainty quantification (UQ) techniques to enhance robustness, interpretability, and trustworthiness. By consolidating methodological insights, emerging paradigms, and deployment-oriented considerations, this review provides a comprehensive reference and a forward-looking roadmap for the development of reliable, physics-consistent, and scalable RUL prognostic frameworks for fuel cell systems. • In-depth review of ML-based RUL prediction methods for fuel cell systems. • Identification of critical gaps in autonomous End of Life detection and missing RUL targets. • Proposing hybrid physics machine learning use cases for accurate RUL estimation. • Delivering actionable recommendations for RUL estimation in FC applications.