Proton exchange membrane fuel cells (PEMFCs) are promising for zero-emission vehicles, but their sub-zero start-up capability remains a major hurdle. Freezing of product water inside the membrane electrode assembly can block reactant transport, deactivate catalyst sites, and cause irreversible mechanical damage. A study published in Frontiers of Chemical Science and Engineering presents a comprehensive framework that combines cathode catalytic H₂-O₂ reaction heating with machine learning (ML) and multi-objective optimization to enhance cold-start efficiency.
Unlike conventional self-starting methods that couple heat and water generation, the cathode catalytic approach uses a non‑electrochemical combustion reaction, separating heat production from water formation. This provides high-power heating during initial startup while effectively inhibiting ice accumulation. The research team built a 450-cell stack model using gFUELCELL software, validated against experimental polarization data (Pearson correlation coefficient 0.99). They then designed a two-phase cold-start procedure: a preheating phase using catalytic H₂-O₂ reaction followed by an electrochemical heating phase.
Comparative simulations showed the cathode catalytic strategy far outperformed the anode‑initiated method. At -20 °C, the cathode approach raised the stack coolant temperature to 70 °C in 59.7 seconds, achieving a heating rate of approximately 2.3 °C·s⁻¹. Ice volume fraction in the cathode catalyst layer peaked at only 3.28 vol% at 6 seconds and melted rapidly, with ice persisting for only about 12 seconds. In contrast, the anode catalytic method failed to exceed 0 °C even at 37 seconds.
To enable rapid prediction and optimization, the team trained four ML models: random forest, support vector regression, artificial neural network, and XGBoost. XGBoost was selected as the surrogate model due to its superior ability to capture complex nonlinear relationships. SHAP (SHapley Additive exPlanations) analysis revealed that anode back pressure and hydrogen temperature dominate ice volume fraction, while pump flow coefficient and reactant temperature significantly affect preheating efficiency.
Multi-objective optimization using the NSGA‑II algorithm identified Pareto‑optimal solutions balancing preheating time, electrochemical heating time, and ice volume fraction. Compared to the base case, optimized parameters reduced preheating time by about 5 seconds, and cut both time indicators by approximately 14–18 %. However, the study also identified limitations: the static mapping capability of XGBoost, combined with error propagation during genetic algorithm iterations, caused the optimized frontier to deviate from the real physical boundary in some regions, particularly in controlling ice volume fraction below 1 vol% within 30 seconds.
This work demonstrates the feasibility of combining cathode catalytic heating with data‑driven optimization, while highlighting the need for integrating physical mechanisms and expanding extreme‑condition data to improve engineering applicability. Future efforts will focus on ultra‑low temperature scenarios (‑30 °C and below), dynamic hydrogen injection for safety, and reactant pre‑humidification.
DOI
10.1007/s11705-026-2643-9