Numerical prediction of largely separated flows in low-pressure turbine blades with high loading

Abstract

Low-pressure turbine blades pose a great challenge for designers as they are highly loaded and are prone to large flow separations at sufficiently low Reynolds numbers, typically at higher altitudes. Simulating these flows is computationally challenging due to the limitations of turbulence models in predicting laminar separations and subsequent transitions. Current work assesses the ability of the widely used $k_T\text{- }k_L\text{- }\omega$ turbulence model for the prediction of the separation region in two different Pratt & Whitney blades, T106-A and Pak-B, at different loadings. Unlike most simulations reported in the literature, current work utilizes a multiblock structured grid on a computational domain where suction and pressure sides are separated by pitch-wise distance. This meshing approach helps in capturing the leading and trailing edge curvatures, apart from improved accuracy with a structured grid approach. The computed results are compared with experimental data as well as available numerical results. For the T106-A blade flow at inlet Reynolds number $Re\simeq 50{,}000$, RANS simulation successfully predicts the separation and reattachment on the suction side very accurately. On the Pak-B blade, steady RANS simulations are performed for two inlet Reynolds numbers: 50,000 and 100,000. The pressure coefficient matches very well with the experimental data for the higher Re, but disagreements are found in the prediction of separation bubble size in the aft region of the suction side for $Re=50{,}000$. Some discrepancies are consistently observed in the predictions of wake losses downstream of both blades, suggesting limitations of RANS models in steady-state simulations in predicting wake mixing. It is noted that unsteady simulations reduce the wake error and predict better mixing, apart from improvements in predictions of boundary layer parameters.