A review on machine learning-based prognostics and health management for hydrogen fuel cells

Owing to their high efficiency, energy density, and zero-emission, hydrogen fuel cells are regarded as a promising energy solution to alleviate environmental pollution. The reliability of fuel cells is crucial to their safety in life cycle. To accelerate the process of engineering application, it is significant to investigate prognostics and health management (PHM) methods for hydrogen fuel cells to enhance both safety and reliability. PHM is an emerging discipline encompassing state of health (SOH) estimation, remaining useful life (RUL) prediction and fault diagnosis. Within the realm of Artificial Intelligence (AI), machine learning (ML) has been increasingly integrated into PHM domain. Due to the strength of high accuracy and efficiency, ML-based PHM methods show significant promise in systems with complex data and multiple parameters like fuel cells. In this paper, we reviewed relevant literature in the field of ML-based PHM methods for hydrogen fuel cells, focusing primarily on developments over the last decade. While various algorithms continue to integrate and innovate to achieve optimal performance in modeling, data preprocessing, and optimization, significant scope for development remains due to unresolved open issues. Consequently, we identified future research directions in ML-based PHM methods for fuel cells, including uncertainty quantification, digital twin, generative models and prescriptive maintenance.