This study investigates the aerodynamic performance of a low-pressure turbine, namely the T106C, by large eddy simulation (LES) and coarse grid direct numerical simulation (CDNS) at a Reynolds number of 100,000. Existing experimental data were used to validate the computational fluid dynamics (CFD) tool. The effects of subgrid scale (SGS) models, mesh densities, computational domains and boundary conditions on the CFD predictions are studied. On the blade suction surface, a separation zone starts at a location of about 55% along the suction surface. The prediction of flow separation on the turbine blade is always found to be difficult and is one of the focuses of this work. The ability of Smagorinsky and wall-adapting local eddy viscosity (WALE) model in predicting the flow separation is compared. WALE model produces better predictions than the Smagorinsky model. CDNS produces very similar predictions to WALE model. With a finer mesh, the difference due to SGS models becomes smaller. The size of the computational domain is also important. At blade midspan, three-dimensional (3D) features of the separated flow have an effect on the downstream flows, especially for the area near the reattachment. By further considering the effects of endwall secondary flows, a better prediction of the flow separation near the blade midspan can be achieved. The effect of the endwall secondary flow on the blade suction surface separation at the midspan is explained with the analytical method based on the Biot–Savart Law.
