A Comparison of node-based and CAD-based parametrisations in shape optimisation

Aerodynamic shape optimisation with gradient-based methods is rapidly gaining popularity in the aerospace and automotive industries. The most effective method to compute the required sensitivities is the adjoint method which allows to compute sensitivities of an arbitrary number of design variables at constant computational cost. This in turn opens up a wide range of possibilities to parametrise the shape as the number of design variables is no longer a limiting factor as long as the parametrisation algorithm can also differentiated in reverse mode. A wide range of parametrisations have been proposed for aerodynamic shape optimisation.1 Of particular interest to us are parametrisations that do not require manual setup but can be derived automatically from existing information. One the one hand we will consider a node-based parametrisation which uses the surface nodes of the CFD grid as design variables. On the other hand we will consider CAD-based parametrisations based on the boundary representation (BRep) as given in the STEP file format