Evaluation of the Capacity of RANS/URANS/LES in Predicting the Performance of a High-Pressure Turbine: Effect of Load and Off Design Condition

High-fidelity methods like Large-Eddy Simulation (LES) have undoubtedly done significant progress in the last years in simulating and predicting more realistic turbomachinery configurations. Although they provide highly valuable insights in flow physics, the common feedback from industry is that these methods are still expensive and require more validations especially if using law-of-the-wall to allow predictions of extremely high Reynolds number and transitional flows. Moreover, they are still wondering about how to use these new methods in the conception plan of real engines, compared to the more mature methods like Reynolds-Averaged Navier Stokes (RANS) or unsteady RANS.

The objective of this paper is to provide preliminary answers to these questions by addressing design issues by use of LES. To do so, a research state-of-the-art high-pressure turbine stage, without technological details (except fillet) and for which experimental data are available, is computed with the three methods: i.e. RANS, URANS and LES. Starting from the nominal operating design, a database is acquired varying the design space (three Zweifel numbers), load (three pressure rates) and rotation speed (three reduced speeds). For this study, special care is taken to ensure that same operating points are simulated and same post-treatment techniques are used.

The analysis of the database is carried out incrementally from a design perspective. Numerical results are systematically compared to experimental ones. First, a global view of performances is done through 0D maps of efficiency, reduced mass flow and flow angle as a function of the pressure rate, reduced speed and Zweifel number. Then, radial profiles of total pressure and flow angle are compared in absolute but also relative to the nominal operating point. Secondary flows and more particularly the radial migration of boundary layer on the rotating blade are studied to understand what’s missing in the numerical simulations. Finally, 2D maps of total pressure and flow angle are compared in a plane downstream the rotating blade. A fair comparison of CPU cost but also return time is also provided.

Main conclusions are twofold: 1/ Calibrated RANS provides excellent results at the nominal operating point but lacks of accuracy at off design conditions. Only unsteady methods (both URANS and LES) allows a good agreement with experiment along the whole database. 2/ LES and standard law-of-the-wall is validated against experiments in a high-pressure turbine without technological details but still representative of a realistic and recent industrial design. From an aero design point, this paper shows the interest in using URANS for off design conditions. It also represents a milestone for LES that had to be passed before addressing more complex issues which URANS hardly address, e.g. where turbulent mixing and transport of turbulence are dominant or in aerothermal flows.