Improving shock-capturing robustness for higher-order finite element solvers

Abstract

Simulation of high speed flows where shock waves play a significant role is still an area of development in computational fluid dynamics. Numerical simulation of discontinuities such as shock waves often suffer from nonphysical oscillations which can pollute the solution accuracy. Grid adaptation, along with shock-capturing methods such as artificial viscosity, can be used to resolve the shock by targeting the key flow features for grid refinement. This is a powerful tool, but cannot proceed without first converging on an initially coarse, unrefined mesh. These coarse meshes suffer the most from nonphysical oscillations, and many algorithms abort the solve process when detecting nonphysical values. In order to improve the robustness of grid adaptation on initially coarse meshes, this thesis presents methods to converge solutions in the presence of nonphysical oscillations. A high order discontinuous Galerkin (DG) framework is used to discretize Burgers' equation and the Euler equations. Dissipation-based globalization methods are investigated using both a pre-defined continuation schedule and a variable continuation schedule based on homotopy methods, and Burgers' equation is used as a test bed for comparing these continuation methods. For the Euler equations, a set of surrogate variables based on the primitive variables (density, velocity, and temperature) are developed to allow the convergence of solutions with nonphysical oscillations. The surrogate variables are applied to a flow with a strong shock feature, with and without continuation methods, to demonstrate their robustness in comparison to the primitive variables using physicality checks and pseudo-time continuation.