Optimization methods are used to drive mesh adaptation to reduce the discretization error in a solution to the 2D Euler equations. The mesh adaptation is driven by minimizing a functional based on truncation error, since it is the local source of the discretization error. The supersonic expansion fan was used to perform the investigation of the proposed methods. This paper investigates several possible design variables to be used in the optimization process. The improvements in discretization error achieved on the different optimized meshes and equidistributed meshes are compared. The costs of doing adaptation versus uniform refinement are also discussed and shown that adaptation is cheaper than uniform refinement. It is observed that equidistribution obtains similar reductions in error compared to mesh optimization but is much cheaper to perform. It is also noted that costs can be saved by performing equidistribution on a coarse mesh and then refining that mesh using a spline fit.
展开▼