Computation of unsteady turbomachinery flows: Part 2—LES and hybrids

Abstract

The choice of turbulence model can have a strong impact on results for many turbomachinery zones. Palliative corrections to them and also transition modeling can have a further profound solution impact. The spectral gaps necessary for theoretically valid URANS solutions are also lacking in certain turbomachinery zones. Large Eddy Simulation (LES) alleviates the serious area of turbulence modeling uncertainty but with an extreme increase in computational cost. However, there seems a lack of validation data to explore in depth the performance of LES and thus strategies to refine it. LES best practices are needed. Although LES is, obviously, much less model dependent than RANS, grids currently used for more practical simulations are clearly insufficiently fine for the LES model and numerical schemes not to be playing an excessively strong role. Very few turbomachinery simulations make use of properly constructed, correlated turbulence inflow. Even if this is attempted, most measurement sets are incomplete and lack an adequate basis for modeling this inflow. Gas turbines are highly complex coupled systems and hence inflow and outflow boundary condition specification needs to go beyond just synthesizing turbulent structures and preventing their reflection. Despite the strong limitations of the dissipative Smagorinsky model, it still sees the most wide spread use, generally, in excessively dissipative flow solvers. Monotone Integrated LES (MILES) related approaches, hybrid LES–RANS and more advanced LES models seem to have an equal but subservient frequency of use in turbomachinery applications. Clearly the introduction of a RANS layer can have a substantial accuracy penalty. However, it does allow LES to be rationally used, albeit in a diluted sense for industrial applications. The Reynolds numbers found in turbomachinery are substantial. However, in certain areas evidence suggests they will not be enough to ensure a long inertial subrange and hence the use of standard LES modeling practices. Despite the excessively coarse grids used in much of the LES work reviewed, with essentially RANS based codes, meaningful results are often gained. This can perhaps be attributed to the choice of cases, these being ones for which RANS modeling gives extremely poor performance. It is a concern that for practical turbomachinery LES studies grid densities used tend to have an Reynolds number scaling to a strong negative power.