基于卷积神经网络的结冰翼型气动参数预测

Aerodynamic parameters prediction of airfoil ice accretion based on convolutional neural network

针对传统CFD计算获取结冰翼型气动参数过程繁琐的问题,提出一种基于数据转换和深度学习的图形化预测方法,开展结冰翼型气动特性参数的预测研究。该方法利用卷积神经网络可有效提取图形特征的特点,通过数据转换提取翼型结冰图形,建立结冰翼型气动参数预测模型,通过训练网络建立冰形与气动参数之间的非线性映射关系。最后,以NACA0012翼型为例开展仿真试验,证明了方法的可靠性和高效性。研究结果表明,利用该方法预测气动参数平均相对误差在5%以内,预测时间在毫秒量级,可为后续飞机结冰飞行安全分析和防/除冰设计提供参考。

For the complicated process of obtaining the aerodynamic parameters of the icing airfoil in the traditional CFD calculation, a graphical method based on data conversion and a deep learning was proposed to carry out the prediction research of icing airfoil aerodynamic parameters. According to the characteristic that convolution neural network could effectively extract graphic features, the icing image of airfoil leading edge was obtained through data conversion, and the prediction model of aerodynamic parameters of icing airfoil was established. The nonlinear mapping relationship between ice shape and aerodynamic parameters was realized through training network. At last, the simulation test was carried out by taking NACA0012 airfoil as an example to prove the reliability and efficiency of this method. It is shown that the average relative error of aerodynamic parameters predicted by this method is less than 5%, and the prediction time is in the order of millisecond. The result can provide reference for the subsequent aircraft icing flight safety analysis and anti/de-icing design.