Book Description
The fast and reliable simulation of turbulent flow using Reynolds Averaged Navier Stokes (RANS) models is a major financial issue for many industries. With the increasing complexity of geometries and simulated flows, as well as requirements in terms of fidelity, the generation of appropriate meshes has become a key link in the chain of computation. We show in this thesis the ability of modern numerical schemes to simulate turbulent flows on fully unstructured meshes generated automatically using mesh adaptation methods. We present the implementation of different versions of the Spalart-Allmaras model as well as the numerical choices guaranteeing a sufficient robustness of the solver in order to not require a structured boundary layer. We then introduce the error analysis necessary to propose different error estimators for mesh optimization. This methodology is tested on various external aerodynamic and turbomachinery test cases and compared to traditional mesh generation methods. We show the ability of mesh adaptation methods to automatically generate optimal mesh sizes for RANS simulations on realistic and complex geometries.