基于神经网络的跨音速非定常气动力的辨识

利用递归神经网络(RNN)模型具有时间记忆性,且会考虑之前的输入输出对当前输出影响的特点,以递归神经网络方法建立了NACA0012翼型在跨音速阶段的非定常气动力模型;利用CFD计算NACA0012翼型绕其刚心作变频俯仰运动的跨音速气动力系数为训练数据,建立跨音速非定常气动力模型。以建立的跨音速非定常气动力模型预测NACA0012翼型作俯仰简谐振动的气动力系数,并与CFD计算的气动力系数进行对比。结果表明,该模型具备优良的逼近非线性非定常气动力的能力;针对跨音速二维翼型,该模型相比CFD可以更快速地构建,并能迅速且较为准确地预测不同频率下作简谐振动时的气动力。

Identification of Unsteady Transonic Aerodynamic Forces Based on Neural Network

Based on RNN neural network model with time memory and considering the characteristics of the influence of previous input and output on current output, the unsteady aerodynamic model of NACA0012 airfoil in transonic stage is established by RNN method. The transonic aerodynamic coefficients of NACA0012 airfoil with frequency conversion pitching around its rigid center are calculated by CFD as training data, and the transonic unsteady aerodynamic model is established. The aerodynamic coefficients of NACA0012 airfoil for pitching simple harmonic vibration are predicted by the established unsteady transonic aerodynamic model and compared with those calculated by CFD. The results show that the model has a good ability to approximate the nonlinear unsteady aerodynamic force. For the transonic two-dimensional airfoil, the model can be constructed more quickly than CFD, and can predict the aerodynamic force of simple harmonic vibration at different frequencies more quickly and accurately.