High dimensional shape optimization in turbomachinery assisted by gradient enhanced metamodeling

The aim is to develop innovative metamodeling techniques that integrate gradient information, facilitating the optimization of turbomachinery with precision and robustness akin to local gradient-based methods, while maintaining a global perspective. This endeavor faces two primary challenges: integrating "noisy" gradient information from adjoint solvers and operating in a high-dimensional input space. Recent literature includes a few attempts derived from the Gradient Enhanced Kriging, but they have yet to yield truly satisfactory results. The challenges persist as the models prove ill-conditioned and struggle to manage large mathematical systems. Additionally, there is a sensitivity to gradient accuracy; despite the advancement of adjoint solvers, the gradients of Quantities of Interest (QoI) obtained are acknowledged to be slightly inaccurate.