查看文章

In the field of optimal shape design, we are interested in iteratively changing an aerodynamic shape in order to improve the performance of an aircraft. In realistic three-dimensional design problems, it is typical for shape optimizations to require hundreds of design variables. 1, 2 The core impetus for these large dimension spaces is that we cannot know a-priori which variables are needed to most efficiently optimize the design. We thus construct parameterizations with many variables in order to increase the chance that we reach a useful result. However, this increases the cost required to find an optimum, and increases the possibility that multi-modality appears in complex design problems. Today’s optimization techniques manage this “curse of dimensionality” in different ways. Local optimizers such as gradient-based optimization techniques efficiently find local minima when paired with adjoint-based sensitivity …