Mathematics

When dealing with turbulent parametrized flow problems, the strategy to be employed for an efficient and reliable solution can result difficult to be selected. It is mostly due to the possible high number of required realizations in case of shape optimization or different physical scenarios simulation. In addition, to resolve turbulent flow problems, a large variety of turbulence models is nowadays available. Although the direct projection of the additional set of equations describing the kinetic energy dissipation is possible, such an approach would produce a ROM which depends on the selection of the turbulence model. In this talk, we present instead a more generic approach where the contribution due to chaotic motion of fluid particles is operated by data-driven approaches. Specifically, the idea is to combine classical projection-based techniques applied to compressible Navier-Stokes equations together with artificial intelligence methods for turbulence treatment. This choice allows us to develop a procedure for turbulent flows simulations which is completely independent on the turbulence modeling selection.