A hybrid method combining degradation mechanisms and deep learning for lifetime prediction of proton exchange membrane fuel cells under dynamic load cycle conditions

TY - JOUR

T1 - A hybrid method combining degradation mechanisms and deep learning for lifetime prediction of proton exchange membrane fuel cells under dynamic load cycle conditions

AU - Ke, Chang

AU - Han, Kai

AU - Wang, Yongzhen

N1 - Publisher Copyright: © 2025 Elsevier B.V.

PY - 2025/3/30

Y1 - 2025/3/30

N2 - Prognostics and health management (PHM) is an effective method to improve the durability of proton exchange membrane fuel cells (PEMFCs). Accurate lifetime prediction is an essential prerequisite for health management. This paper proposes a hybrid prediction method that combines degradation mechanisms with deep learning neural networks to predict the degradation trends and estimate the remaining useful life (RUL) of PEMFCs under dynamic load cycle conditions. Firstly, the polarization curve model is employed to extract degradation-related parameters and quantify the overvoltage. The relationship between overvoltage and membrane electrode assembly (MEA) degradation is analyzed, revealing that cathode catalyst and membrane are the key components influencing the degradation. Secondly, a comprehensive degradation index (CDI) is developed. A novel method for quantifying the weight coefficients of the CDI is proposed for the first time. The effects of catalyst and membrane degradation on the overall performance degradation are quantified, which are 82.2 % and 17.8 %, respectively. Finally, the long short-term memory (LSTM) and gated recurrent unit (GRU) models are employed to predict the degradation trend. The results show that GRU outperforms LSTM in this study. The maximum RUL estimation error of the proposed hybrid method is 9.50 %, with all errors within the 10 % confidence interval.

AB - Prognostics and health management (PHM) is an effective method to improve the durability of proton exchange membrane fuel cells (PEMFCs). Accurate lifetime prediction is an essential prerequisite for health management. This paper proposes a hybrid prediction method that combines degradation mechanisms with deep learning neural networks to predict the degradation trends and estimate the remaining useful life (RUL) of PEMFCs under dynamic load cycle conditions. Firstly, the polarization curve model is employed to extract degradation-related parameters and quantify the overvoltage. The relationship between overvoltage and membrane electrode assembly (MEA) degradation is analyzed, revealing that cathode catalyst and membrane are the key components influencing the degradation. Secondly, a comprehensive degradation index (CDI) is developed. A novel method for quantifying the weight coefficients of the CDI is proposed for the first time. The effects of catalyst and membrane degradation on the overall performance degradation are quantified, which are 82.2 % and 17.8 %, respectively. Finally, the long short-term memory (LSTM) and gated recurrent unit (GRU) models are employed to predict the degradation trend. The results show that GRU outperforms LSTM in this study. The maximum RUL estimation error of the proposed hybrid method is 9.50 %, with all errors within the 10 % confidence interval.

KW - Deep learning

KW - Degradation index

KW - Hybrid prediction method

KW - Proton exchange membrane fuel cell

UR - https://www.scopus.com/pages/publications/85217081045

U2 - 10.1016/j.jpowsour.2025.236464

DO - 10.1016/j.jpowsour.2025.236464

M3 - Article

AN - SCOPUS:85217081045

SN - 0378-7753

VL - 633

JO - Journal of Power Sources

JF - Journal of Power Sources

M1 - 236464

ER -