Methods for design optimization using high fidelity turbulent flow simulations

Abstract

Design optimization with high-fidelity turbulent flow simulations can be challenging due to noisy and expensive objective function evaluations. The noise decays slowly as computation cost increases, therefore is significant in most simulations. It is often unpredictable due to chaotic dynamics of turbulence, in that it can be totally different for almost identical simulations. This thesis presents a modified parallel Bayesian optimization algorithm designed for performing optimization with high-fidelity simulations. It strives to find the optimum in a minimum number of evaluations by judiciously exploring the design space. Additionally, to potentially augment the optimization algorithm with the availability of a gradient, a massively parallel discrete unsteady adjoint solver for the compressible Navier-Stokes equations is derived and implemented. Both the methods are demonstrated on a large scale transonic fluid flow problem in a turbomachinery component.