Flowfield Reconstruction Method Using Artificial Neural Network

Abstract

Multidimensional aerodynamic database development has become more and more important for the design, control, and guidance of modern aircraft. To relieve the curse of the dimensionality, a novel flowfield reconstruction method based on an artificial neural network is proposed. The idea is to design a simplified problem that is related to the target problem. Then, the map from the simplified problem to the target problem is built using an artificial neural network. Finally, the target problem can be predicted efficiently through solving the simplified problem instead. Examples of the efficiency of this approach include two-dimensional viscous nozzle flows, an inviscid M6 wing flow, a viscous hypersonic flow of a complex configuration, and an unsteady two-dimensional Riemann problem to evaluate the performance of the proposed method. A Gaussian process model is also incorporated for a comparative study. With an artificial neural network of moderate complexity, the solution of the target problem can be generated with good accuracy. Among other observations, it is found that shocks can be predicted well with sharp resolution for steady and unsteady cases. Overall, using a simplified problem that accounts for all the interested parameters as inputs tends to be more reliable than using input parameters.