使用深度残差网络的乘波体气动性能预测
|
摘要
本文探究深度学习人工智能技术在飞行器气动外形预测中的应用。以激波装配法乘波体设计为背景,建立气动数据快速生成工具,使用拉丁超立方采样得到海量样本数据。使用深度残差神经网络构建气动外形参数到气动性能数据的代理模型,并与随机森林和双隐层神经网络等普通机器学习模型对比;同时将数据转换为图片,研究基于图片识别的深度学习模型搭建,省略飞行器外形的参数化表达。测试结果说明,深度残差网络作为数据代理模型的精度是随机森林和双隐层神经网络的3倍以上,而基于图片识别的代理模型精度提高有限。研究表明,深度残差网络在乘波体等易于生成大量数据的气动外形的性能预测中效果明显,为深度学习技术在气动外形设计中的应用奠定了基础。
The applications of deep learning in aerodynamic predictions are explored in the article.A big data set is generated from 3D waverider design using the Latin Hypercube Sampling.The deep residual neural network(ResNet)is applied to construct a surrogate model for aerodynamic shape parameters and aerodynamic performances.Compared with the Random Forests and two-hidden-layer Neural Network,the high efficiency of ResNet is studied.In addition,the idea of surrogate models based on the image recognition is proposed with the image set.The aerodynamic performance prediction omits the shape parameterization,and the surrogate model is easy to be implemented.The efficiency of the ResNet surrogate model is over 3 times higher than that of the common machine learning methods,such as Random Forests and twohidden-layer Neural Network.However,the efficiency of the ResNet image recognition model is not improved.The results suggest that the feasibility of the deep learning in aerodynamic shape design is confirmed,especially in some simple configurations,such as waverider.
The applications of deep learning in aerodynamic predictions are explored in the article.A big data set is generated from 3D waverider design using the Latin Hypercube Sampling.The deep residual neural network(ResNet)is applied to construct a surrogate model for aerodynamic shape parameters and aerodynamic performances.Compared with the Random Forests and two-hidden-layer Neural Network,the high efficiency of ResNet is studied.In addition,the idea of surrogate models based on the image recognition is proposed with the image set.The aerodynamic performance prediction omits the shape parameterization,and the surrogate model is easy to be implemented.The efficiency of the ResNet surrogate model is over 3 times higher than that of the common machine learning methods,such as Random Forests and twohidden-layer Neural Network.However,the efficiency of the ResNet image recognition model is not improved.The results suggest that the feasibility of the deep learning in aerodynamic shape design is confirmed,especially in some simple configurations,such as waverider.
引文
[1]高正红,王超.飞行器气动外形设计方法研究与进展[J].空气动力学学报,2017,35(4):516-528.GAO Z H,WANG C.Aerodynamic shape design method for aircraft:status and trends[J].Acta Aerodynamica Sinica,2017,35(4):516-528.(in Chinese)
[2] HILTON G E.OSINDERO S,THE Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[3] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[C]//Proceedings of Advances in Neural Information Processing System.South lake Tahoe,USA:Curran Associates Press,2012:1097-1105.
[4] SILVER D,HUANG A,MADDISON C,et al.Mastering the game of Go with deep neural networks and tree search[J].Nature,2016,529(7587):484-489.
[5] HOYLE N, BRESSLOFF N W, KEANE A J.Design optimization of a two-dimensional subsonic engine air intake[J].AIAA Journal,2006,44:2672-2681.
[6] PATNAIK S N,CORONEOS R M.A subsonic aircraft design optimization with neural network and regression approximators[R].AIAA 2004-4606,2004.
[7]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)
[8]陈冰雁,刘传振,纪楚群.基于激波装配法的乘波体设计与分析[J].空气动力学学报,2017,35(3):421-428.CHEN B Y,LIU C Z,JI C Q.Waverider design and analysis based on shock-fitting method[J].Acta Aerodynamica Sinica,2017,35(3):421-428.(in Chinese)
[9] LES PIEGL,WAYNE TILLER.The NURBS book(second edition)[M].USA:Springer-Verlag,1995:32-84.
[10]KRIZHEVSKY ALEX,SUTSKEVER I,HINTON G.ImageNet classification with deep convolutional neural networks[J].Communications of the ACM,2017,6(6):84-90.
[11]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large scale image recognition[C]//International Conference on Learning Representations.San Diego,USA,2015.
[12]CHRISTIAN S,LIU W,JIA Y,et al.Going deeper with convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Boston,USA,2015.
[13]HE K M,ZHANG X,REN S Q,et al.Deep residual learning for image recognition[R]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Boston,USA,2015.
[14]JIA Y Q,SHELHAMER E,DONALHUE J,et al.Caffe:convolutional architecture for fast feature embedding[C]//Proceedings of the ACM InternationalConferenceon Multimedia.Glasgow,UK,2014.
[15]HE K M,ZHANG X,KOMODAKIS N.Delving deep into rectifiers:surpassing human-level performance on imagenet classification[C]//IEEE International Conference on Computer Vision.Santiago,Chile,2015.
[16]BREIMAN L.Bagging predictors[J].Machine Learning,1996,24(2):123-140.
[2] HILTON G E.OSINDERO S,THE Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
[3] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.Imagenet classification with deep convolutional neural networks[C]//Proceedings of Advances in Neural Information Processing System.South lake Tahoe,USA:Curran Associates Press,2012:1097-1105.
[4] SILVER D,HUANG A,MADDISON C,et al.Mastering the game of Go with deep neural networks and tree search[J].Nature,2016,529(7587):484-489.
[5] HOYLE N, BRESSLOFF N W, KEANE A J.Design optimization of a two-dimensional subsonic engine air intake[J].AIAA Journal,2006,44:2672-2681.
[6] PATNAIK S N,CORONEOS R M.A subsonic aircraft design optimization with neural network and regression approximators[R].AIAA 2004-4606,2004.
[7]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)
[8]陈冰雁,刘传振,纪楚群.基于激波装配法的乘波体设计与分析[J].空气动力学学报,2017,35(3):421-428.CHEN B Y,LIU C Z,JI C Q.Waverider design and analysis based on shock-fitting method[J].Acta Aerodynamica Sinica,2017,35(3):421-428.(in Chinese)
[9] LES PIEGL,WAYNE TILLER.The NURBS book(second edition)[M].USA:Springer-Verlag,1995:32-84.
[10]KRIZHEVSKY ALEX,SUTSKEVER I,HINTON G.ImageNet classification with deep convolutional neural networks[J].Communications of the ACM,2017,6(6):84-90.
[11]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large scale image recognition[C]//International Conference on Learning Representations.San Diego,USA,2015.
[12]CHRISTIAN S,LIU W,JIA Y,et al.Going deeper with convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Boston,USA,2015.
[13]HE K M,ZHANG X,REN S Q,et al.Deep residual learning for image recognition[R]//IEEE Conference on Computer Vision and Pattern Recognition(CVPR).Boston,USA,2015.
[14]JIA Y Q,SHELHAMER E,DONALHUE J,et al.Caffe:convolutional architecture for fast feature embedding[C]//Proceedings of the ACM InternationalConferenceon Multimedia.Glasgow,UK,2014.
[15]HE K M,ZHANG X,KOMODAKIS N.Delving deep into rectifiers:surpassing human-level performance on imagenet classification[C]//IEEE International Conference on Computer Vision.Santiago,Chile,2015.
[16]BREIMAN L.Bagging predictors[J].Machine Learning,1996,24(2):123-140.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |