Data fusion of multi-fidelity model and its application in low speed reflexed airfoil shape optimization

In this paper, a data fusion method was presented to combine multi Computational Fluid Dynamics models together in a target to constitute a scheme which could give an accurate and expeditious prediction of CFD simulation and reduce the total computational cost in a global optimization using Genetic Algorithm. Numerical models were defined as lower and higher fidelity according to the accuracy and speed of simulation. The framework of artificial neural network was applied to map and fuse variable fidelity models together. The lower fidelity model was employed as a neuron on the network, and a mapping strategy mixed multiplicative and additive scaling between lower and higher fidelity was built. A low speed reflexed airfoil optimization was deployed with a target to maximize the lift-to-drag ration under the constraints of consideration about stability margin using this technique. The optimal result proved that the method in this paper could decrease the computational cost and retain an acceptable accuracy of higher fidelity model at the same time.