Aeroelastic Prediction System with Multiinput-Multioutput Characteristics Based on the Gated Recurrent Neural Network

A recurrent neural network-based nonlinear aerodynamic order reduction model with robustness to different Mach numbers is developed. The Mach number is added as an additional input variable to prediction system for simulate nonlinear characteristics with different flow condition. A weighted filtered white gaussian noise with a large frequency and amplitude range is used as the training case. The prediction accuracy of the model with numerous inputs and different outputs is examined to validate the generalization capacity of model. Predicting the unsteady aerodynamic forces of the NACA0012 airfoil during transonic flow verifies the efficiency of the suggested method. The model accurately reflects the major characteristics of transonic flow, as well as the time lag characteristics of the NACA0012 airfoil under different incoming flow conditions and reduce frequency, as shown by a comparison of harmonic aerodynamic responses in time domain. This method significantly reduces calculation time when compared to typical CFD calculation methods, and the calculation time of this method is just around 6.47% of the entire time cost of a full-order simulation using a CFD solver.

I. Introduction

In most circumstances, the aerodynamic force/moment of an airplane in motion is a function of time. This is a common aeroelastic dynamic issue that occurs when the wing is aroused by unexpected wind or when the wing is abruptly maneuvered. Aeroelastic dynamics is an unstable process that involves the interaction of elasticity and aerodynamics. The key is to improve the accuracy and efficiency of unsteady aerodynamics calculation. In the field of aeroelasticity, the standard technique based on CFD/CSD coupling calculation is mainstream technology, however its low computing efficiency limits the further development of aeroelasticity analysis. At the same time, the huge amounts of data it generates cannot be successfully exploited or examined.