Aerodynamic Optimization based on a Discrete Adjoint Framework and Radial Basis Function Mesh Deformation in SU2

A Radial Basis Function (RBF) mesh deformation is implemented and coupled with the discrete adjoint framework within the open-source toolkit SU2. The RBF allows to handle more complex geometry than the current elastic deformation approach, while enabling large deformations to expand the design space. Data reduction schemes including multilevel greedy algorithms are used to improve the efficiency of RBF mesh deformation on large data sets. Numerical experiments show a significant reduction of memory usage over the linear elasticity analogy both for two-dimensional cases and for large, three-dimensional problems. Additionally, the mesh deformation process is differentiated by Automatic Differentiation within the discrete adjoint approach, resulting in method-dependent surface sensitivity, thus allowing the Sequential Least Squares Programming optimizer to converge to a new local minimum by modifying the geometrical shape towards the final design.