Evolvable free-form deformation control volumes for evolutionary design optimization

Evolutionary design optimization for improving the performance of real world objects, like e.g. car shapes in the context of aerodynamic efficiency, usually depends on a well balanced combination of representation, optimizer and design evaluation method. Shape representation requires a fair trade-off between minimum number of design parameters and design flexibility which likewise guarantees a good optimization convergence while allowing manifold design variations. Recently, shape morphing methods have gained increased attention because of their capability to represent complex shapes with a reasonable number of parameters, especially powerful if coupled with numerical simulations for measuring design performance. Free-form deformation, as prominent shape morphing representative, relies on an initial grid of control points, the control volume, which allows the modification of the embedded shape. The set-up of the control volume is a crucial process which in practice is done manually based on the experience of the human user. Here, a method for the automated construction of control volumes is suggested based on a proposed measure E_CV which relies on the concept of evolvability as a potential capacity of representations to produce successful designs in a reasonable time. It is shown for target shape matching experiments that optimizations based on E_CV-tuned control volumes provide a significantly better performance in design optimization.