Degradation prediction and Remaining Useful Life (RUL) estimation methods for Proton Exchange Membrane Fuel Cells (PEMFC): A review

The paper discusses recent developments in diagnosing proton exchange membrane fuel cell (PEMFC) degradation and predicting remaining useful life (RUL), focusing on conventional and machine learning-based methods for improved prediction accuracy and performance. The commercialization of PEMFCs is hindered by durability and operational reliability issues, necessitating accurate degradation diagnosis and remaining useful life prediction to improve RUL prediction performance. A summary of the degradation mechanisms of PEMFCs and a discussion on state-of-art machine learning-based prognostic has been provided. Machine learning models utilize measurable parameters, including voltage, current, power, temperature and pressure, to enable real-time degradation diagnosis without requiring destructive material inspection. The strength and limitations of data-driven and model-based methods are critically compared to their prediction accuracy and computational efficiency. This review shows that among the reviewed techniques, particle swarm-guided long short-term memory (LSTM-PSO) effectively enhances model performance through parameter optimization. Based on the reviewed literature, an integrated LSTM-PSO framework is identified as a promising approach for improving accuracy and reliability of real-time PEMFC degradation diagnosis and RUL prediction.