基于卷积神经网络的结冰翼型气动特性建模研究

何磊; 钱炜祺; 董康生; 易贤; 柴聪聪

doi:10.7527/S1000-6893.2021.26434

航空学报 >

0 0 - 0

DOI: https://doi.org/10.7527/S1000-6893.2021.26434

基于卷积神经网络的结冰翼型气动特性建模研究

何磊 ,
钱炜祺 ,
董康生 ,
易贤 ,
柴聪聪

展开

1. 中国空气动力研究与发展中心计算空气动力研究所；中国空气动力研究与发展中心空气动力学国家重点实验室
2. 中国空气动力研究所与发展中心计算所
3. 中国空气动力研究与发展中心计算空气动力研究所；
4. 中国空气动力研究与发展中心
5. 中国空气动力研究与发展中心空气动力学国家重点实验室

收稿日期: 2021-09-24

修回日期: 2021-11-09

网络出版日期: 2021-11-10

收起

Aerodynamic characteristics modeling of iced airfoil based on convolution neural networks

HE Lei ,
QIAN Wei-Qi ,
DONG Kang-Sheng ,
YI Xian ,
CHAI Cong-Cong

Expand

Received date: 2021-09-24

Revised date: 2021-11-09

Online published: 2021-11-10

Fold

摘要

提出了基于卷积神经网络（Convolutional Neural Networks，CNN）的翼型结冰气动特性预测方法，设计了输入层结冰翼型图像规范，克服了复杂冰形在翼面同一位置法线方向存在多值，单值函数难以描述的问题。预测模型可同时预测多个迎角对应的升阻力系数，实现了直接从冰形图像到气动特性的快速预测，对升力系数（CL）和阻力系数（CD）预测结果的平均相对误差均可控制在8%以内。重点研究了不同卷积层数量、卷积核数量、卷积核尺寸对模型性能的影响规律：CNN的不同层次特征对应不同滤波频率，卷积层数增加会捕获更多高频特征量；增加卷积核数量可提取更多冰形特征，提升模型性能，但数量过多会增加冗余特征，降低模型泛化性能；CD预测模型对卷积核数量的最低要求大于CL，其原因在于，相较CL，CD不仅受翼面压差影响，还受摩阻特性影响，其建模所需的关键特征数量需求多于CL；增大卷积核尺寸，可扩大卷积操作“视野”，增强对冰形整体特征信息的提取，有利于提升模型泛化性能。相关结论为飞机结冰气动特性实时动态预测与监测提供了新的思路和方法支撑。

关键词： 飞机结冰; 气动特性; 机器学习; 深度学习; 卷积神经网络

本文引用格式

何磊 , 钱炜祺 , 董康生 , 易贤 , 柴聪聪 . 基于卷积神经网络的结冰翼型气动特性建模研究[J]. 航空学报, 0 : 0 -0 . DOI: 10.7527/S1000-6893.2021.26434

Abstract

A prediction method of iced airfoil aerodynamic characteristics is proposed based on the convolutional neural networds (CNN). This new method is able to solve the multiple-value problem of the complex ice shape along the wall-normal direction at a specific location of the airfoils. The CNN prediction model can predict the lift and drag coefficient at multiple angle of attacks by a rapid process from iced airfoil image to aerodynamics characteristics directly. The relative mean errors of the resulting lift coefficient (CL) and drag coefficient (CD) is less than 8%. By comparison, the influence of number of convolutional layers, number of convolutional filters and convolutional filter size on the model performance was investigated.The features of different layers in CNN correspond to different filter frequency, increasing number of convolutional layers will capture more features of high frequency. When increasing the number of filters，more features of ice shape will be extracted and the performance of model will be promoted. However, there will be more redundant features if the number of the filters exceed a critical value, which results in degration of the generalization performance.The required number of filters for predicting CD is higher than that of CL. This is because of the effects of characteristics of friction on CD are even more pronounced, so the number of key features requied to predicting CD is higher than that of CL. Moreover, larger size of filters will widen the field of view of convolution operation. Consequently, the capability of extracting global features will be enhanced, which will promote the generalization performance of the model. The current new method can be applied to the real-time prediction and monitor of aerodynamic characteristic of aircraft icing.

Key words： aircraft icing; aerodynamic characteristics; machine learning; deep learning; convolutional neural networks

参考文献

[1] 郭向东, 柳庆林, 刘森云, 等. 结冰风洞中过冷大水滴云雾演化特性数值研究[J]. 航空学报, 2020, 41 (8): 123655.
GUO X D, LIU Q L, LIU S Y, et al. Numerical study of supercooled large droplet cloud evolution characteristics in icing wind tunnel[J]. Acta Aeronautica et Astronautica Sinica, 2020, 41(8): 123655(in Chinese).
[2] 易贤, 王斌, 李伟斌, 等. 飞机结冰冰形测量方法研究进展[J]. 航空学报, 2017, 38 (2): 520700.
YI X, WANG B, LI W B, et al. Research progress on ice shape measurement approaches for aircraft icing[J]. Acta Aeronautica et Astronautica Sinica, 2017, 38(2): 520700 (in Chinese).
[3] 桂业伟, 周志宏, 李颖晖. 关于飞机结冰的多重安全边界问题[J]. 航空学报, 2017, 38 (2): 520723.
GUI Y W, ZHOU Z H, LI Y H, et al. Multiple safety boundaries protection on aircraft icing[J]. Acta Aeronautica et Astronautica Sinica, 2017, 38(2): 520723(in Chinese).
[4] KIM H S, BRAGG M B. Effects of leading-edge ice accretion geometry on Airfoil Performance[R].Reston, VA: AIAA.1999, AIAA-1999-3150
[5] BRAGG M B, HUTCHISON T, MERRET J, et al. Effects of Ice Accretion on Aircraft Flight Dynamics[J]. 2000: AIAA 2000-0360.
[6] 钟长生, 王立新. 结冰对飞机动力学特性影响的分析方法及其进展[J]. 飞行力学, 2004, 12 (4): 22-24.
ZHONG C S, WANG L X. Analysis methods and its development about effect of ice accretion on aircraft flight dynamics characteristics[J]. Flight　Dynamics, 2004, 22(4): 22-24, 84(in Chinese).
[7] POKHARIYAL D, BRAGG M B, HUTCHISON T, et al. Aircraft Flight Dynamics with Simulated Ice Accretion[R].Reston, VA: AIAA.2001, AIAA-2001-0541
[8] 袁坤刚, 曹义华. 积冰几何特性对翼型性能影响的神经网络预测[J]. 北京航空航天大学学报, 2008, 34 (8): 900-903.
YUAN K G, CAO Y H. Effect of ice geometry to airfoil performance using neural networks prediction[J] Journal of Beijing University of Aeronautics and Astronautics, 2008, 34(8): 900-903(in Chinese).
[9] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 31 (5786): 504-507.
[10] 吴正文. 卷积神经网络在图像分类中的应用研究[D]. 成都: 电子科技大学, 2015.
WU W Z. Application research of convolution neural network in image classification[D]. Chengdu: University of Electronic Science and Technology, 2015. (in Chinese)
[11] RADFORD A, METZ L, CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks[DB/OL]. arXiv:1511.06434, 2015.
[12] GATYS L A, ECKER A S, BETHGE M. A neural algorithm of artistic style[DB/OL]. arXiv:1508.06576, 2015.
[13] KARPATHY A, TODERICI G, SHETTY S, LEUNG T. 2014 Large scale video classification with convolutional neural networks[C]// In Proc. IEEE Conf. on Computer Vision and Pattern Recognition. IEEE, 2014 of Conference.1725–1732.
[14] FRANK S, GANG L, Dong Y. Conversational speech transcription using context-dependent deep neural networks[C]// The 12th Annual Conference of the International Speech Communication Association. Florence, Traly: 2011 of Conference.
[15] 王怡星, 韩仁坤, 刘子扬, 等. 流体力学深度学习建模技术研究进展[J]. 航空学报, 2021, 42(4): 524779.
WANG Y X, HAN R K, LIU Z Y, et al. Progress of deep learning modeling technology for fluid mechanics[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 524779(in Chinese)
[16] 张伟伟, 寇家庆, 刘溢浪. 智能赋能流体力学展望[J]. 航空学报, 2021, 42 (4): 524689.
ZHANG W W, KOU J Q, LIU Y L. Prospect of artificial intelligence empowered fluid mechanics[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 524689(in Chinese).
[17] KUTZ J N. Deep learning in fluid dynamics[J]. Journal of Fluid Mechanics, 2017, 814: 1-4.
[18] YILMAZ E, GERMAN B. A convolutional neural network approach to training predictors for airfoil performance[R].Reston, VA: AIAA.2017, AIAA 2017-3660
[19] ZHANG Y, SUNG W J, Mavris D N. Application of convolutional neural network to predict airfoil lift coefficient[C]// 2018 AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference. Kissimmee, USA: 2018 of Conference.1903.
[20] XIAO H, WU J L, WANG J X, et al. Physics informed machine learning for predictive turbulence modeling: progress and perspectives[R]. Reston, VA: AIAA.2017, AIAA-2017-1712
[21] HUANG J J, DUAN L, WANG J X, et al. High-mach-number turbulence modeling using machine learning and direct numerical simulation database[R].Reston, VA: AIAA.2017, AIAA-2017-0315
[22] LING J, KURZAWSKI A, TEMPLETON J. Reynolds averaged turbulence modeling using deep neural networks with embedded invariance[J]. Journal of Fluid Mechanics, 2016, 807: 155-166.
[23] 张伟伟, 朱林阳, 刘溢浪. 机器学习在湍流建模型构建中的应用进展[J]. 空气动力学报, 2019, 37 (03): 444-451.
ZHANG W W, ZHU L Y, LIU Y L, et al. Progresses in the application of machine learning in turbulence modeling[J]. Acta Aerodynamica Sinica, 2019, 37(3): 444-454 (in Chinese).
[24] 廖鹏, 姚磊江, 白国栋, 等. 基于深度学习的混合翼型前缘压力分布预测[J]. 航空动力学报, 2019, 34 (8): 1751-1758.
LIAO P, YAO L J, BAI G D, et al. Prediction of hybrid airfoil leading edge pressure distribution based on deep learning[J]. Journal of Aerospace Power, 2019, 34(8): 1751-1758(in Chinese).
[25] RAISSI M, YAZDANI A, KARNIADAKIS G. E. Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations[J]. Science, 2020, 367 (6481): 1026-1030.
[26] GUO X X, LI W, LORIO F. Convolutional neural networks for steady flow approximation[C]// ACM SigKDD International Conference. San Francisco, CA: 2016 of Conference.481-490.
[27] LEE S, YOU D. Data-driven prediction of unsteady flow over a circular cylinder using deep learning[J]. Journal of Fluid Mechanics, 2019, 879: 217-254.
[28] 叶舒然, 张珍, 王一伟, 等. 基于卷积神经网络的深度学习流场特征识别及应用进展[J]. 航空学报, 2021, 42 (4): 524736.
YE S R, ZHANG Z, WANG Y W, et al. Progress in deep convolutional neural network based flow field recognition and its applications[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 524736(in Chinese).
[29] LECUN Y, BOTTOU L, BENGIO Y. GradientGbased learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86 (11): 2278-2324.
[30] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521 (7553): 436-446.
[31] 陈海, 钱炜祺, 何磊. 基于深度学习的翼型气动系数预测[J]. 空气动力学学报, 2018, 36 (2): 294-299.
CHEN H, QIAN W Q, HE L. Aerodynamic coefficient prediction of airfoils based on deep learning[J]. Acta Aerodynamica Sinica, 2018, 36(2): 294-299(in Chinese).
[32] GLOROT X, BENGIO Y. Understanding the difficulty of training deep feedforward neural networks[C]// International Conference on Artificial Intelligence and Statistics. Brookline, MA: 2010 of Conference.249-256.
[33] NAIR V, HINTON G E. Rectified linear units improve restricted Boltzmann machines[C]// Proceedings of the 27th International Conference on Machine Learning. Haifa, ILMS: 2010 of Conference.807-814.
[34] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[C]// The Advances in Neural Information Processing Systems. Lake Tahoe, NV, USA: 2012 of Conference.1097–1105.
[35] 何磊, 钱炜祺, 汪清, 等. 机器学习方法在气动特性建模中的应用[J]. 空气动力学学报, 2019, 37 (3): 470-479.
HE L,QIAN W Q,WANG Q,et al.Applications of machine learning for aerodynamic characteristics modeling[J].Acta Aerodynamica Sinica,2019,37(3):470-479
[36] CHEN H, HE L, QIAN W Q, et al. Multiple Aerodynamic Coefficient Prediction of Airfoils Using Convolutional Neural Network[J]. Symmetry, 2020, 12 (4): 544.
[37] 何磊, 钱炜祺, 易贤, 王强. 基于转置卷积神经网络的翼型结冰冰形图像化预测方法[J]. 国防科技大学学报, 2021, 43 (3): 98-106.
HE L, QIAN W Q, YI X, et al. Graphical prediction method of airfoil ice shape based on transposed convolution neural networks[J]. Journal of National University of Defense Technology, 2021, 43(3): 98-106(in Chinese).
[38] BRAGG M B. An Experimental Study of the Aerodynamics of a NACA 0012Airfoil with a Simulated Glaze Ice Accretion[R].Reston, VA: AIAA.1986, NASA-CR-179897
[39] BRAGG M B, BROEREN A P, BLUMENTHAL L A. Iced-Airfoil Aerodynamics[J]. Progress in Aerospace Sciences, 2005, 41 (5): 323–362.
[40] BRAGG M B, KHODADOUST A, Spring S A. Measurem ents in a leading-edge separation bubble due to a simulated airfoil ice accretion[J]. AIAA journal, 1992, 30 (6): 1462–1467.
[41] CARL O G. GRUMMP Version 0.2.1 User's Guide[R]. Columbia: University of British Columbia.1997,
[42] 张耀冰, 邓有奇, 吴晓军, 孟凡菊. DLR-F6翼身组合体数值计算[J]. 空气动力学学报, 2011, 29 (2): 163-169.
ZHANG Y B, DENG Y Q, WU X J, et al. Drag prediction of DLR-F6 using MFlow unstructured mesh solver[J]. Acta Aerodynamica Sinica, 2011, 29(2):163-169.
[43] KINGMA D P, BA J. Adam: Amethod for stochastic optimization[DB/OL]. arXiv preprint arciv:1412,6980, 2014.

Options

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract

参考文献