Abstract
This paper addresses limitations of widely used Reynolds-averaged turbulence models (RANS) for prediction of gas turbine internal air systems. Results from direct numerical simulation (DNS), wall-resolved large-eddy simulation (LES), wall-modelled large-eddy simulation (WMLES), and RANS for benchmark test cases are compared. For rotor-stator disc cavity flows results for mean velocities, velocity fluctuations, rotor torque and laminar-turbulent transition are considered and compared with published data. For cavities between co-rotating discs attention is focused on buoyancy-driven convection in the centrifugal force field. It is concluded that WMLES is suitable for application in engine conditions, offering better accuracy than RANS in some critical applications. This confirms recently published results for turbine rim sealing and is further illustrated by application to convection in a sealed cavity at higher Rayleigh number than is practical with DNS or wall-resolved LES. The results show that the approximate near-wall treatment gives reasonable results for complex flows and extend previous studies to higher speed rig conditions where Eckert number effects become significant.