CAD-based and CAD-free Aerodynamic Shape Optimization of Turbomachinery Blade Rows using the Adjoint Method

Abstract

This PhD thesis deals with the gradient-based aerodynamic shape optimization of 3D turbomachinery blade rows, where the sensitivity derivatives of the objective functions with respect to (w.r.t.) the design variables are computed by coupling a new discrete adjoint CFD solver with a differentiated CAD model. First, to increase the efficiency of the Rolls-Royce in-house Hydra adjoint code, an implicit iterative solution scheme is developed (replacing the pre-existing explicit one) and the resulting implicit adjoint solver accelerates the convergence significantly. Then, a node-based CAD-free parameterization as well as alternative CAD-based approaches for computing geometric sensitivities (and, thus, making the shape optimization steps fully compatible with the CAD model) are tested and evaluated according to industrial design requirements (such as geometric constraints' imposition). All the developed tools are incorporated into gradient-based optimization workflows, which are demonstrated on the examined applications. Finally, to be able to optimize w.r.t. more than one objectives, a new multi-objective optimization method is developed, which traces the Pareto front in an efficient way using Hessian approximations computed by the BFGS iterative scheme.