A novel prediction method for unsteady aerodynamic force on three-dimensional folding wing aircraft

Abstract

The aerodynamic characteristics of morphing aircraft frequently exhibit extremely unsteady and nonlinear features during morphing process, making it challenging to obtain accurate aerodynamic parameters fast based on traditional computational fluid dynamics methods. In this paper, a data driven model Multi-Task Cross network, which is based on deep learning methods of cross network and Multi-Task Learning, is proposed to predict the unsteady aerodynamic parameters of a three-dimensional folding wing aircraft at any angle of attack. The Multi-Task Cross model seeks to effectively fuse external flow field and aircraft properties while making full use of sparse input features and limited training data. More accurate results can be obtained by using the Multi-Task Cross model to predict different aerodynamic parameters of folding wing aircraft. The model takes three inputs: angle of attack, time and unfolding angle of wing. It outputs the unsteady forces and moments coefficients through multi-layer nonlinear mapping. The training and testing data used for the Multi-Task Cross model are calculated by computational fluid dynamics method. With the proposed prediction model, the unsteady aerodynamic parameters at any angle of attack and any morphing state can be obtained in seconds. The testing results indicate that the Multi-Task Cross model predicts aerodynamic parameters quickly and accurately. The proposed Multi-Task Cross model achieves higher prediction accuracy when compared to prediction models based solely on Multilayer Perceptron, Multi-Task Learning method, and Cross method.