 |
Center for Nonlinear Studies |
University of Wyoming
This talk will cover recent advances made at the University of Wyoming in sensitivity analysis techniques for computational fluid dynamics problems on unstructured meshes. The basic approach is based on the implementation and solution of the discrete adjoint problem. A consistent modular approach has been adopted for implementing the discrete adjoint and convergence of the adjoint problem which is equivalent to that of the analysis problem is obtained using the transpose of the preconditioner used for the analysis problem. The resulting adjoint-based sensitivity analysis approach is used to drive design optimization problems, for both steady and unsteady problems. In the case of unsteady problems, the entire time history of the solution must be stored for use by the adjoint solver as it proceeds backwards in time. This is done by writing out the solution to disk at each time step. The use of the adjoint problem for estimating multiple sources of error in a specific simulation objective output is also demonstrated. For an unsteady oscillating airfoil problem, the error sources due to temporal resolution as well as incomplete convergence of the linearsystem of equations at each implicit time step are estimated independently by the adjoint approach and used to drive an adaptive time step selection and convergence tolerance criteria at each time step. Finally, prospects for further advances in error estimation for multiphysics problems will be discussed in the concluding remarks of this talk.
Host: David Moulton