Application of EOS Based on Machine Learning Method on CFD Study of Rapid Hydrogen Refueling Process

Hydrogen is attracting attention as an eco-friendly energy source that can replace fossil fuels. In particular, hydrogen fuel cell electric vehicles (FCEVs) have been developed to reduce carbon dioxide emissions in the transportation sector. Currently, commercially available FCEVs store hydrogen as highly compressed gas form to increase volumetric energy density. To provide a refueling time similar to that of internal combustion engine vehicles (ICEVs), hydrogen refueling stations (HRSs) are installed to supply gaseous hydrogen into FECVs up to 35 MPa or 70 MPa in a relatively short time. The refueling process of filling compressed gas within a confined volume of the on-board storage tank is inevitably accompanied by the temperature increase. However, the refueling process should be carried out under a limited temperature considering the thermal and mechanical safety of the storage tank. Since the hydrogen storage tank installed in the commercial FCEV is equipped with a single temperature sensor, only the average temperature can be measured and monitored during the refueling process. Therefore, modeling the refueling process is useful for understanding the gas filling phenomenon and finding the optimal refueling strategy. In particular, the CFD study method that considers the motion of the fluid inside the tank enables prediction of local temperature changes inside the storage tank, which cannot be measured in the commercial vehicle refueling process. The CFD research is conducted by combining expressions representing the fluid properties and a model describing the flow characteristics. Therefore, an appropriate combination of equations should be examined before developing a CFD model and simulating the refueling process. In this study, the hydrogen refueling process is simulated using three equations of state (EOSs) and five turbulent models. The results are compared and quantitatively analyzed using experimental data to propose an appropriate EOS with an accurate turbulence model. Experiments of hydrogen filling into Type III tank of 74 L up to 35 MPa within 1 min have been chosen to make the assumption of axial symmetry for CFD model valid. Comparing the three EOSs (SRK, PR, and ML), it is found that the reduction of simulation time can be attained with good accuracy when using ML EOS which has been developed to describe the volumetric property of hydrogen. Among the five turbulence models (yPlus, k–ε, realizable k–ε, low-Reynolds k–ε, and k–ω) generally used in many CFD studies, the realizable k–ε model shows satisfactory results on the reproduction of mean and local thermal behaviors inside of on-board storage tanks.