Convergence acceleration of fluid dynamics solvers using a reduced-order-model

Abstract

A new method based on Reduced Order Modeling (ROM) is presented for the purpose of accelerating the convergence of an iterative solver via solution prediction in the framework of a Computational Fluid Dynamics (CFD) solver. Using a number of flow solutions as snapshots at pre-convergence iterations of the explicit time-marching solver, the ROM- based acceleration technique is used to project a more accurate solution. The method is first applied to accelerate the convergence of an incompressible Navier-Stokes solver and the classical two-dimensional lid-driven cavity flow case is revisited for this purpose. Next, the proposed convergence acceleration technique is utilized in the framework of a density- based CFD solver and subsonic as well as transonic ows over a NACA0012 airfoil are studied.