Machine learning for adjoint vector in aerodynamic shape optimization

Adjoint method is widely used in aerodynamic design because only once solution of flow field required for it to obtain the gradients all variables. However, computational cost adjoint vector approximately equal that computation. In order accelerate and improve efficiency adjoint-based optimization, machine learning modeling presented. Deep neural network (DNN) employed construct mapping between local DNN can efficiently predict its generalization examined by a transonic drag reduction NACA0012 airfoil. The results indicate with negligible vector, proposed DNN-based achieve same optimization as traditional method.