Aerodynamic shape optimization based on discrete adjoint and RBF

Abstract

This paper presents a new framework for Aerodynamic Shape Optimization (ASO) implemented inside the open-source software SU2. A parametrized surface is iteratively morphed to improve a desired aerodynamic coefficient. At each loop, the computational grid must be updated. The radial basis function (RBF) mesh deformation technique is introduced to extend the capability to explore the design space. RBF is implemented with some state-of-the-art data reduction systems to lower the computational cost. The Discrete Adjoint is adopted to compute the sensitivity in combination with Automatic Differentiation to calculate the required Jacobians. RBF is differentiated as well, resulting in a method-dependent surface sensitivity. The gradient-based algorithm “Sequential Least Squares Programming” drives the research of the minimum. The study demonstrates that the presented combination is more robust than an ASO, including linear elasticity analogy (ELA) as mesh deformation method. It can handle complex geometries and apply larger displacements, making possible the optimization of a wing-winglet configuration and a rotating wind turbine. Results are presented in two and three-dimensions for compressible and incompressible flows, showing a stronger reduction of the drag without affecting the lift.