基于深度学习的翼型参数化建模方法

Deep learning-based method for parametrized modeling of airfoil

为解决现有翼型几何参数化描述方法优化设计效率低、计算工作量大的问题,提出了一种基于深度学习的翼型参数化建模方法。该方法以伊利诺伊大学厄巴纳-香槟分校(University of Illinois at Urbana-Champaign,UIUC)翼型数据库中翼型上下表面坐标点转化的翼型二维图像作为输入,首先使用卷积运算提取大量翼型图像的几何特征,然后通过多层感知机对提取的几何特征进行分类和压缩,将翼型形状压缩成若干个简化的拟合参数,最后通过解码器恢复翼型图像并输出翼型上下表面的点坐标。在此基础上,探讨了拟合参数数量对翼型几何精度的影响,确定了含6个拟合参数的卷积神经网络(convolutional neural network,CNN)结构,并基于计算流体力学数值仿真验证了所提出方法的拟合精度。最后,开发了可视化翼型几何设计软件,实现了拟合参数的调整与修正,并分析了各拟合参数对翼型形状的影响规律。结果表明,6个拟合参数均会对翼型形状产生全局影响,单独或联合调整6个拟合参数可获得新的翼型设计空间。研究结果可为翼型的优化设计提供技术支持与理论参考。

In order to solve the problems of low efficiency and heavy computational workload during the optimization design process in the existing airfoil geometric parametrized modeling methods,a deep learning-based airfoil parametrized modeling method was put forward.In this method,the twodimensional airfoil images converted from coordinate points of airfoil upper and lower surfaces in the airfoil database of the University of Illinois at Urbana-Champaign(UIUC)were taken as the input.Firstly,the convolution operations were used to extract geometric features of a large amount of airfoil images.Then,the extracted geometric features were classified and compressed by multi-layer perceptron,and the airfoil shape was compressed into several simplified fitting parameters.Finally,the airfoil image was restored and the coordinates of points on the upper and lower surfaces of airfoil were output by a decoder.On this basis,the influence of the number of fitting parameters on the geometric accuracy of airfoil was discussed,and a convolutional neural network(CNN)structure with six fitting parameters was determined.At the same time,the fitting accuracy of the proposed method was verified by the computational fluid dynamics numerical simulation.Finally,the visual airfoil geometry design software was developed to adjust and modify the fitting parameters,and the influence law of each fitting parameter on the airfoil shape was summarized.The results indicated that all the six fitting parameters had a global impact on the airfoil shape,and the new airfoil design space could be obtained by adjusting the six fitting parameters individually or jointly.This research results can provide technical support and theoretical guidance for airfoil optimization design.