Hybrid genetic algorithms and artificial neural networks for complex design optimization in CFD

The present paper is devoted to the study of design optimization strategies in the particular framework of complex computational fluid dynamics. Genetic algorithms are chosen as the optimization strategy, thanks to their robustness and flexibility. Two ways are explored to improve the behaviour of genetic algorithms in order to increase the efficiency of the search. First, approximated pre‐evaluations based on artificial neural networks are used to benefit from the knowledge acquired from the problem and to reduce the number of expensive evaluations by the flow solver required at each generation. Then, a hybridization technique is proposed for the final local search, which is performed by a deterministic method. These approaches are validated and applied on two‐ and three‐dimensional problems, involving Reynolds‐averaged Navier–Stokes computations with near‐wall turbulence modeling.