CAD-based aerodynamic shape optimization using geometry surrogate model and adjoint methods

Current optimization tasks at the industrial level pose several challenging requirements, including high accuracy of flow evaluation (single drag counts), short optimization time to actively progress with the design loop (single days), robustness of the process, definition of design variables in CAD. Despite recent advances in aerodynamic shape optimization based on adjoint methods, it has not yet been matured enough to efficiently use it in the industrial environment. A typical bottleneck is the efficient and robust link between CAD model and the deformation of the computational mesh. Optimization frameworks including this coupling are in the scope of current research interest. One of the approaches is to use NURBS control points extracted from CAD as design variables. The advantage of this method is a simple framework (CAD is used outside optimization loop) and analytical surface gradient evaluation, with the main disadvantages of having limited control over the parameterization (as it is not using CAD model parameters) and limited ability to handle complex geometries (intersections). The other notable approach is based on using CAD engine in the optimization loop through CAPRI interface. It has the advantage of using actual CAD parameterization as design variables and can handle complex geometries, although implementation is more complicated especially in the industrial environment due to calls to the CAD engine at each iteration. The latter approach has been used in gradient-based optimization with surface deformation gradient computed with finitedifference method. It is possible to obtain fully analytical geometry sensitivities, although it requires significant implementation effort (either to develop differentiable CAD engine or to analytically differentiate all the operations performed by existing CAD engines).