A NEURAL NETWORK BASED ACCELERATION TECHNIQUE OF GENETIC ALGORITHM CONVERGENCE IN AERODYNAMIC DESIGN OPTIMIZATION

Wing section optimization is aceomplished using a combined strategy consisting of a genetic algorithm (GA ) and an artificial neural network (A::-..r:K) . A real coded genetic algmi.t hm is utilized for an op­ timum search in design space. The numerical so­ lution of inviscid flow governing equations is used for evaluation of the design candidates. In order to red uce the number of these
time consuming eval­ uations required by GA, every M generat ion, all chromosomes fitness are trained to a neural net ­ work. Then, a control based genetic local search is handled by AX':\ as a fitness estimator to find new promising regions in design space. It is demon­ strated that this approach could save considerable computational time in application fields, such as
aerodynamic design . Results are presented for a constrained optimization of an airfoil at transonic flow conditions. The PARSEC method of airfoil generator and unstructured grid movement tech­ nique are used in this work. Event ually, optimum airfoil geometry is achieved by about 50% less com­ putational effort compared with the conventional GA method.