EI、Scopus 收录
中文核心期刊

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于卷积神经网络的钝体尾迹识别研究

杜祥波 陈少强 侯靖尧 张帆 胡海豹 任峰

杜祥波, 陈少强, 侯靖尧, 张帆, 胡海豹, 任峰. 基于卷积神经网络的钝体尾迹识别研究. 力学学报, 2022, 54(1): 1-9 doi: 10.6052/0459-1879-21-404
引用本文: 杜祥波, 陈少强, 侯靖尧, 张帆, 胡海豹, 任峰. 基于卷积神经网络的钝体尾迹识别研究. 力学学报, 2022, 54(1): 1-9 doi: 10.6052/0459-1879-21-404
Du Xiangbo, Chen Shaoqiang, Hou Jingyao, Zhang Fan, Hu Haibao, Ren Feng. Wake recognition of a blunt body based on convolutional neural network. Chinese Journal of Theoretical and Applied Mechanics, 2022, 54(1): 1-9 doi: 10.6052/0459-1879-21-404
Citation: Du Xiangbo, Chen Shaoqiang, Hou Jingyao, Zhang Fan, Hu Haibao, Ren Feng. Wake recognition of a blunt body based on convolutional neural network. Chinese Journal of Theoretical and Applied Mechanics, 2022, 54(1): 1-9 doi: 10.6052/0459-1879-21-404

基于卷积神经网络的钝体尾迹识别研究

doi: 10.6052/0459-1879-21-404
基金项目: 国家自然科学基金(52071272, 12102357), 基础前沿(JCKY2018*18), 陕西省自然科学基础研究计划(2020JC-18), 中央高校基本科研业务费专项资金(3102021HHZY030002)和河南省水下智能装备重点实验室开放基金(KL01B2101)资助项目
详细信息
    作者简介:

    胡海豹, 教授, 主要研究方向: 水下仿生与流动控制. E-mail: huhaibao@nwpu.edu.cn

    任峰, 副教授, 主要研究方向: 主动流动控制. E-mail: renfeng@nwpu.edu.cn

  • 中图分类号: TV13

WAKE RECOGNITION OF A BLUNT BODY BASED ON CONVOLUTIONAL NEURAL NETWORK

  • 摘要: 针对相同特征长度不同钝体的尾迹结构相近, 肉眼难于分辨的问题, 提出了一种基于卷积神经网络的钝体尾迹识别方法, 并在竖直肥皂膜水洞的典型钝体模型尾迹实验中获得高准确率验证. 实验平台由自建竖直肥皂膜实验装置、钝体模型(方柱、圆柱和三角柱)及图像采集系统组成, 可基于光学干涉法实现对不同速度下钝体肥皂膜尾迹的高清持续拍摄. 所建立卷积神经网络识别模型由输入层、卷积层、池化层、全连接层和分类层组成, 其中, 卷积层和池化层用于提取尾迹的深层次特征信息, 而全连接层和分类层构成识别分类模式来分类输出图像对应的钝体类型或雷诺数. 通过将9000张尾迹图像数据集导入卷积神经网络模型, 以数据驱动方式建立了具有钝体形状或雷诺数识别能力的尾迹特征识别模型. 结果表明, 该模型对相同雷诺数下识别钝体形状的准确率达97.6%(识别300张不同形状钝体尾迹图像), 对不同雷诺数下识别钝体形状的准确率达96%(识别1200张不同雷诺数尾迹图像), 即使将不同钝体形状和雷诺数下尾迹图像混放一起, 其钝体形状和雷诺数识别准确率也可以达到91%(识别1500张混放尾迹图像). 该方法为进一步利用人工智能提取流体尾迹中的物理信息提供借鉴.

     

  • 图  1  肥皂膜水洞结构图

    Figure  1.  Schematics of the vertical soap-film tunnel

    图  2  照明和拍摄装置示意图

    Figure  2.  Positions of the lighting device and the shooting device

    图  3  不同流量下的速度曲线

    Figure  3.  Velocity curves at different flow rates

    图  4  肥皂膜厚度与流速关系

    Figure  4.  Relationship between the film thickness and the flow rate

    图  5  典型钝体绕流尾迹图像

    Figure  5.  Image of wake around a typical blunt body

    图  6  尾迹识别CNN模型

    Figure  6.  CNN model for wake recognition

    图  7  CNN流程图

    Figure  7.  Flow chart to establish the CNN model

    图  8  卷积神经网络的损失函数值和验证集正确率

    Figure  8.  Loss function value and verification data set accuracy of the CNN

    图  9  测试识别结果

    Figure  9.  Test recognition results

    图  10  图像裁剪后的识别结果

    Figure  10.  Test image recognition results with reduction

    图  11  不同Re识别结果

    Figure  11.  Different Re recognition results

    图  12  相近工况识别结果

    Figure  12.  Identification results of similar working conditions

    图  13  卷积神经网络的损失函数值和验证集正确率

    Figure  13.  Loss function value and verify set accuracy of CNN

    图  14  混放识别结果

    Figure  14.  Mixed identification results

    图  15  卷积池化过程

    Figure  15.  Convolutional pooling process

    表  1  CNN网络结构参数

    Table  1.   CNN network structure parameters

    Network layerKey parameterOutputActivation function
    input 400 × 100, RGB:3 400 × 100 × 3
    conv1 kernel:5 × 5, step size:1 400 × 100 × 32 ReLU
    pooling1 kernel:2 × 2, step size:2 200 × 50 × 32 ReLU
    conv2 kernel:5 × 5, step size:1 200 × 50 × 64 ReLU
    pooling2 kernel:2 × 2, step size:2 100 × 25 × 64 ReLU
    conv3 kernel:5 × 5, step size:1 100 × 25 × 128 ReLU
    pooling3 kernel:2 × 2, step size:2 50 × 12 × 128 ReLU
    conv4 kernel:5 × 5, step size:1 50 × 12 × 128 ReLU
    pooling4 kernel:2 × 2, step size:2 25 × 6 × 128 ReLU
    dense1 1024 ReLU
    dense2 512 ReLU
    dropout scale parameter:0.5
    softmax 3 softmax
    下载: 导出CSV

    表  2  实验工况

    Table  2.   Experimental conditions

    Square columnCylinder columnTriangular column
    Re=160 case1 case2 case3
    Re=220 case4 case5 case6
    Re=275 case7 case8 case9
    Re=480 case10 case11 case12
    Re=550 case13 case14 case15
    下载: 导出CSV
  • [1] Schulte-Pelkum N, Wieskotten S, Hanke W, et al. Tracking of biogenic hydrodynamic trails in harbour seals (Phoca vitulina). Journal of Experimental Biology, 2007, 210(5): 781-787 doi: 10.1242/jeb.02708
    [2] Wieskotten S, Dehnhardt G, Mauck B, et al. Hydrodynamic deter mination of the moving direction of an artifificial fifin by a harbour seal (Phoca vitulina). Journal of Experimental Biology, 2010, 213(13): 2194-2200 doi: 10.1242/jeb.041699
    [3] Wieskotten S, Dehnhardt G, Mauck B, et al. The impact of glide phases on the trackability of hydrodynamic trails in harbour seals (Phoca vitulina). Journal of Experimental Biology, 2010, 213(21): 3734-3740 doi: 10.1242/jeb.047134
    [4] Dehnhardt G, Mauck M, Hanke W, et al. Hydrodynamic trail-following in harbor seals (Phoca vitulina). Science, 2001, 293: 102-104 doi: 10.1126/science.1060514
    [5] Jia LB, Yin XZ. Passive oscillations of two tandem flexible filaments in a flowing soap film. Physical Review Letters, 2008, 100(22): 228104 doi: 10.1103/PhysRevLett.100.228104
    [6] Jia LB, Li F, Yin XZ. Coupling modes between two flapping filaments. Journal of Fluid Mechanics, 2007, 581: 199-220 doi: 10.1017/S0022112007005563
    [7] Rutgers M, Wu XL. Two dimensional velocity profiles and laminar boundary layers in flowing soap films. Physics of Fluids, 1996, 8(11): 2847-2854 doi: 10.1063/1.869105
    [8] Rutgers M, Wu XL, Daniel WB. Conducting fluid dynamics experiments with vertically falling soap films. Review of Scientific Instruments, 2001, 72(7): 3025-3037 doi: 10.1063/1.1379956
    [9] 高宗海, 王思莹, 贾来兵等. 用于小Re数流动显示研究的水平式肥皂膜水洞. 实验流体力学, 2009, 23(3): 80-84 (Gao Zonghai, Wang Siying, Jia Laibing, et al. A horizontal soap film water hole for the study of small Re number flow display. Experimental Fluid Mechanics, 2009, 23(3): 80-84 (in Chinese) doi: 10.3969/j.issn.1672-9897.2009.03.017
    [10] 贾来兵. 二维流场中板状柔性体与流体相互作用的研究. [博士论文]. 中国科学技术大学, 2009

    Jia Laibing. Study on the interaction between plate flexible body and fluid in two-dimensional flow field. [PhD Thesis]. University of Science and Technology of China, 2009 (in Chinese)
    [11] Wolf BJ. Shape Classification using hydrodynamic detection via a sparse large-scale 2D-sensitive artificial lateral line. IEEE Access, 2020, 8: 11393-11404 doi: 10.1109/ACCESS.2020.2965316
    [12] Li B, Zhang X. Classifying wakes produced by self-propelled fish-like swimmers using neural networks. Theoretical and Applied Mechanics Letters, 2020, 10(3): 149-154 doi: 10.1016/j.taml.2020.01.010
    [13] 赵西增, 徐天宇, 谢玉林等. 基于卷积神经网络的涵洞式直立堤波浪透射预测. 力学学报, 2021, 53(2): 330-338 (Zhao Xizeng, Xu Tianyu, Xie Yulin, et al. Prediction of wave transmission of culvert breakwater based on convolutional neural network. Chinese Journal of Theoretical and Applied Mechanics, 2021, 53(2): 330-338 (in Chinese) doi: 10.6052/0459-1879-20-235
    [14] Roggen D. Deep convolution and LSTM recurrent neural networks for multimodal wear rable activit recognition. Sensors, 2016, 16(1): 115-121 doi: 10.3390/s16010115
    [15] Dong Z, Pei MT, He Y, et al. Vehicle type classification using unsupervised convolutional neural network//Proceedings of the 22nd International Conference on Pattern Recognition. Stockholm, Sweden, 2014: 172-177
    [16] 魏晓良, 潮群, 陶建峰等. 基于LSTM和CNN的高速柱塞泵故障诊断. 航空学报, 2021, 42(3): 435-445 (Wei Xiaoliang, Chao Qun, Tao Jianfeng, et al. Fault diagnosis of high speed piston pump based on LSTM and CNN. Acta Aeronautica et Astronautica Sinica, 2021, 42(3): 435-445 (in Chinese)
    [17] 杨义红, 尹协振, 陆夕云. 肥皂膜水洞实验技术. 实验流体力学, 2005, 19(4): 36-41 (Yang Yihong, Yin Xiezhen, Lu Xiyun. Soap film water tunnel experiment technology. Experimental Fluid Mechanics, 2005, 19(4): 36-41 (in Chinese) doi: 10.3969/j.issn.1672-9897.2005.04.007
    [18] Leweke T, Williamson C. Three-dimensional instabilities in wake transition. European Journal of Mechanics-B/Fluids, 1998, 17(4): 571-586 doi: 10.1016/S0997-7546(98)80012-5
    [19] 高宗海. 水平肥皂膜水洞研制及其应用. [博士论文]. 中国科学技术大学, 2008

    Gao Zonghai. Development and application of horizontal soap-film water tunnel. [PhD Thesis]. University of Science and Technology of China, 2008 (in Chinese)
    [20] 潘松, 田新亮. 基于直立型肥皂膜水洞的典型钝体绕流阻力实验研究. 水动力学研究与进展: A辑, 2020, 35(5): 601-607 (Pan Song, Tian Xinliang. Experimental study on flow resistance around a typical blunt body based on a vertical soap-film tunnel. Journal of Hydrodynamics:Series A, 2020, 35(5): 601-607 (in Chinese)
    [21] Zhang Y, Sung W, Mavris D. Application of convolutional neural network to predict airfoil lift coefficient//AIAA/ASCE/AHS/ASC Structures. Structural Dynamics and Materials Conference, 2018: 1903-1912
    [22] Schirrmeister RT, Gemein L, Eggensperger K, et al. Deep learning with convolutional neural networks for decoding and visualization of EEG pathology. Human Brain Mapping, 2017, 38(11): 5391-5420 doi: 10.1002/hbm.23730
    [23] 岳杰顺, 权晓波, 叶舒然等. 水下发射水动力的多尺度预测网络研究. 力学学报, 2020, 12(4): 1-11 (Yue Jieshun, Quan Xiaobo, Ye Shuran, et al. A multi-scale network for the prediction of hydrody namics in underwater. Chinese Journal of Theoretical and Applied Mechanics, 2020, 12(4): 1-11 (in Chinese)
    [24] Monti F, Boscaini D, Masci J , et al. Geometric deep learningon graphs and manifolds using mixture model CNNs//Proceedings of the Conference on Computer Vision andPattern Recognition. Honolulu, USA, 2017: 5425-5434
    [25] Hinton GE, Srivastava N, Krizhevsky A, et al. Dropout: A simple way to prevent neural networks from overfifitting. Journal of Machine Learning Research, 2014, 15(1): 1929-1958
    [26] 周飞燕, 金林鹏, 董军. 卷积神经网络研究综. 计算机学报, 2017, 40(6): 1229-1251 (Zhou Feiyan, Jin Linpeng, Dong Jun. Review of convolutional neural networks. Chinese Journal of Computers, 2017, 40(6): 1229-1251 (in Chinese) doi: 10.11897/SP.J.1016.2017.01229
    [27] Trischler A, Wang T, Yuan X, et al. Newsqa: a machine comprehen sion dataset//Proceedings of the 2nd Workshop on Representation Learning for NLP, 2017
    [28] Ganin Y, Ustinova E, Ajakan H, et al. Domain-adversarial training of neural networks. Journal of Machine Learning Research, 2017, 17(1): 2096-2030
    [29] Rumelhart DE, Hinton GE, Williams RJ. Learning representations by back propagating errors. Nature, 1986, 323(6088): 533-536 doi: 10.1038/323533a0
    [30] 徐冰冰, 岑科廷, 黄俊杰等. 图卷积神经网络综述. 计算机学报, 2020, 43(5): 755-780 (Xu Bingbing, Cen Keting, Huang Junjie, et al. A review of graph convolutional neural networks. Chinese Journal of Computers, 2020, 43(5): 755-780 (in Chinese) doi: 10.11897/SP.J.1016.2020.00755
    [31] Srivastava N, Hinton G, Krizhevsky A. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 2014, 15(6): 1929-1958
    [32] Sainatha TN, Saona G. Deep convolutional neural networks for large-scale speech tasks. Neural Networks, 2015, 64: 39-48
  • 加载中
图(15) / 表(2)
计量
  • 文章访问数:  145
  • HTML全文浏览量:  23
  • PDF下载量:  28
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-08-20
  • 录用日期:  2021-10-30
  • 网络出版日期:  2021-10-31

目录

    /

    返回文章
    返回