Industry-University Collaborative Project on Numerical Predictions of Cavitating Flows in Hydraulic Machinery: Part 1—Benchmark Test on Cavitating Hydrofoils

Through an industry-university collaborative project, extensive benchmark studies have been made for numerical prediction of cavitating flows around two-dimensional Hydrofoils: Clark-Y 11.7% and NACA0015. The emphases are placed on the ability of present cavitation models to predict the breakdown characteristics for these hydrofoils. The benchmarking was done for a light and a moderate loading condition of these hydrofoils at a chord-based Reynolds number in the order of 10⁶. Four commercial CFD flow solvers, ANSYS CFX, ANSYS Fluent, and STAR-CCM+, and SCRYU/Tetra, along with four open-source or in-house flow solvers in universities participated in this benchmark. All the cavitation models, except one, implemented in these flow solvers are based on an assumption of homogenous media of one fluid, for which inception, growth, decay and destruction of cavitation are expressed by density change of the mixture fluid composed of liquid and gas phases. They differ with each other in how they determine the mixture fluid density and can be categorized into of barotropic type or of source-sink type. Despites these differences in the cavitation models themselves and differences in the Navier-Stokes solvers, turbulence models and computational grids, the results of the benchmark show a consistent trend of discrepancy between the predicted and measured breakdown characteristics. Namely, none of the cavitation models is able to predict sudden drop of the lift coefficient near the breakdown point confirmed in the measured characteristics. The lift coefficients predicted by all the cavitation models show a gradual decrease with decreasing cavitation number. This discrepancy between the predicted and measured breakdown characteristics is most prominent at the higher loading condition for NACA0015. But, it is consistently confirmed for the other cases investigated in this benchmark. The difference seems to be the results of under prediction of the cavity length, which probably comes from an intrinsic limitation associated with a cavitation model based on an assumption of homogeneous media of one fluid.