An adjoint method augmented with grid sensitivities for aerodynamic optimization.

摘要

The discrete adjoint equations for an aerodynamic optimizer are augmented to explicitly include the sensitivities of the grid perturbation. The Newton-Krylov optimizer is paired with grid perturbations via the elasticity method with incremental stiffening. The elasticity method is computationally expensive, but exceptionally robust---high quality grids are produced, even for large shape changes. For the gradient calculation, instead of encompassing grid sensitivities in finite differenced terms for the adjoint equations, they are treated explicitly. This results in additional adjoint equations that must be solved. This augmented adjoint method requires less computational time than a function evaluation, and retains its speed as dimensionality is increased. The accuracy of the augmented adjoint method is excellent, allowing the optimizer to converge more fully. A discussion of the trade-off between lengthy development time and increased performance indicates that the method would be particularly well-suited to complicated three-dimensional configurations.