Parallel Optimization Strategy for Large -Scale Computational Design

The focus of this paper is to document and demonstrate sensitivity analysis tech niques that ensure multidisciplinary flexibility , and that are utilized within a parallel optimization strategy that could possibly reduce the design costs by orders of magnitude for large -scale shape optimization requiring high -fidelity simulations. Furth ermore, unique to this optimization method, as applied to high -fidelity computational design, is the use of the second -derivative in formation produced from the Complex Taylor Series Expansion (CTSE) method. This additional information (i.e., the diagonal o f the Hessian) may be used for both design variable scaling and for incorporation into the optimization algorithm. These enhancements have been found to increase robustness and to accelerate the convergence of the optimization . Demonstrative results are sh own for a single -point aerodynamic shape optimization of a wing immersed in a transonic flow. For this relatively small scale design example with m oderately complex physics, the results illustrate that the parallel optimization strategy reduced the total d esign time by an order of magnitude . For larger scale design and/or optimizations where the high -fidelity physics require much more CPU time, the savings will become much more pronounce