RANS simulation of unsteady cavitation around a Clark-Y hydrofoil with the assistance of machine learning

In the present paper, a novel machine-learning-assisted RANS method is proposed and utilized to investigate the unsteady cavitating flow around a Clark-Y hydrofoil. A data-driven machine learning model for turbulent eddy viscosity (TEV) is constructed based on high-fidelity LES results first and then makes a prediction for TEV in RANS simulation. The machine-learning-predicted TEV successfully imitates the TEV distribution characteristics of LES that there is a concentration around cavities and negative TEV in the wake. Based on the more accurate TEV field, the machine-learning-assisted RANS method makes more accurate predictions about velocity profile and basic performance parameters of the hydrofoil and gives a better representation of shedding process of sheet cavitation when compared with original RANS results. Finally, generalization performance of the machine learning model is assessed by simulating cavitating flows under different inlet velocity conditions. The results indicate that cavitation exhibits more periodicity with the decrease of inlet velocity in the range of investigated conditions, which may be associated with different TEV distribution characteristics.