Synergistic regulation mechanism and performance assessment of two-phase transport in Tesla valve flow fields for PEMFCs: Jointly driven by 3D+1D physics and machine learning models

Insufficient drainage capacity in flow channels leads to liquid water retention, deteriorated gas mass transfer, and severe performance degradation in proton exchange membrane fuel cells. Although the Tesla valve flow field (TVFF) exhibits inherent advantages in unidirectional flow guidance and backflow suppression, its global optimization for enhanced water management remains underexplored. This study proposes a machine learning-driven, multi-objective global optimization framework to maximize both the drainage efficiency and electrochemical output of the TVFF. A high-fidelity 3D+1D multi-physics model, integrating the volume of fluid method with dynamic contact angles, is developed to rigorously resolve gas-liquid two-phase dynamics and coupled electrochemical kinetics. Utilizing optimal Latin hypercube sampling, an artificial neural network surrogate model is trained and subsequently optimized via a genetic algorithm targeting current density, liquid saturation, and pressure drop. The optimal TVFF geometry (channel width 1.226 mm, angle 61.450°, and transition length 1.527 mm) fundamentally alters the internal gas-liquid transport regime, facilitating continuous water film removal. Consequently, compared to the base design, the optimized TVFF cuts water coverage by 77.5%, lowers both liquid saturation and pressure drop by 13.2%, and boosts current density by 3%. Ultimately, this study provides a vital synergistic regulation mechanism to extend the multi-source energy-lifespan of fuel cell systems, unlocking their potential for demanding applications such as hydrogen heavy-duty trucks.