Machine learning assisted health status analysis and degradation prediction of aging proton exchange membrane fuel cells

Abstract

Proton exchange membrane fuel cells (PEMFCs) represent a significant application scenario for hydrogen energy and an important sector in achieving net-zero carbon emission. Prognostics and health management are crucial for enhancing their durability and reducing maintenance costs. This study proposes a framework for health status analysis and degradation prediction of aging PEMFCs, addressing the challenge of accurately identifying internal parameter states faced by current life prediction methods. Six aging factors are incorporated into the developed PEMFC mechanism model to characterize its intricate degradation process. The variations in these factors over a 3750-h experimental period are then estimated using the Particle Filtering method. Results demonstrate a notable reduction in the electrochemical surface area, decreasing from 5.76 m² to 4.08 m², accompanied by a significant increase in leakage current to nearly 6 A m⁻². These findings indicate substantial degradation of both the catalyst layer and membrane. Furthermore, ionic and contact resistances have increased as a result of reduced membrane conductivity and bipolar plate corrosion, respectively. The mass transport capacity has diminished, leading to an elevated concentration loss within the cell. Subsequently, the Transformer model is employed to forecast future changes in the aging factors and realize the degradation prediction over the next 1000 h. The effectiveness of the proposed method is fully validated under various conditions, with the average prediction error less than 4 %, which demonstrates higher long-term prediction accuracy compared to previous studies. This study provides an effective framework for the health management of PEMFCs and facilitates their widespread commercialization.