Turbulence model assessment of the separated flow in the Stanford diffuser

The computational prediction of separated turbulent flows in internal ducts is a well-known challenging problem. The flow separation within the diffuser sections of wind tunnels is a significant problem and was the motivation for this study which aimed to identify the most accurate turbulence model for the prediction of the separated flow regions in a large industrial wind tunnel. To investigate this problem, the Stanford diffuser developed by Cherry et al., was used as a benchmark to compare simulations with different turbulence models. Corner flow separations, with the associated decrease in static pressure rise, total pressure losses and flow unsteadiness are common in rectangular diffusers. Results from the Stanford diffuser test-case show that linear eddy-viscosity models fail to predict flows in highly asymmetric diffusers due to their over-sensitivity to transverse pressure gradients. On the other hand, Reynolds stress model (RSM) simulations, particularly with the Gibson and Launder (GL-RSM) formulation are in better qualitative and quantitative agreement with the experimental data due to the anisotropic nature of their Reynolds stress formulation.