STUDY ON FLOW FIELD PARAMETERS OF WAKE TIME HISTORY TARGET RECOGNITION
-
摘要: 浸入流场中的固体壁面会形成高度复杂且具有一定特征的尾流流场, 利用尾流所包含的信息对物体的外形特征进行识别具有重要的应用价值. 然而, 在较高雷诺数情况下尾流流场形态及其时序特征复杂, 难以通过传统的数学物理方法对流场信号进行特征的识别与提取. 本文提出了基于尾流时程数据深度学习的流场特征提取与分析方法, 实现了基于一点的物理量时程进行流场中物体外形的识别; 同时, 对流场中不同物理参数时程的识别精度与识别结果进行分析与研究, 得到适用于目标识别的最优物理量参数. 通过对圆柱和方柱的尾流数据研究结果表明, 本文提出的基于卷积神经网络的模型具有好的训练收敛性和高的预测精度, 能够识别并提取得到时程数据中包含的流场特征, 采用流场横向速度时程作为物体外形识别信号的模型准确率高. 证明了本方法用于浸入流场中物体外形识别的可行性, 是一种目标识别的高精度方法.Abstract: Wall immersed in fluid will form highly complex wake flow with specific features. Therefore, the extraction and analysis of flow feature has important research value. However, in the case of high Reynolds number, the wake flow field are complex, so it is difficult to identify and extract the flow features by traditional mathematical and statistical method. In this paper, a new flow field feature extraction and analysis method based on deep learning of wake time history data is proposed, and the shape recognition based on local time history is realized; At the same time, accuracy of different time history parameter is analyzed, and the optimal physical parameters suitable for target recognition are obtained. Research results on the flow field data of cylinder and square cylinder show that the model based on convolution neural network proposed in this paper has good training convergence and high prediction accuracy, and model using transverse velocity time history has highest accuracy. At the same time, it is proved that method proposed in this paper is a new high-precision method for target recognition immersed in fluid.
-
表 1 样本与标签设置
Table 1. Sample and label settings
Label Shape Number of parameters Number of samples 0 circular cylinder 6 21600 1 square cylinder 6 21600 表 2 完全卷积神经网络模型参数
Table 2. Time convolution neural network model parameters
Layer Kernel number Size of kernel Activation function Output size input — — — 300 convolution 128 8 ReLU 300, 128 convolution 256 5 ReLU 300, 256 convolution 128 3 ReLU 300, 128 pooling — — — 128 FCL — — — 6 表 3 全连接网络模型参数
Table 3. Time convolution neural network model parameters
Layer Dropout parameter Number of neural Trainable parameter input − − 0 FCL 1 0.1 500 150500 FCL 2 0.2 500 250500 FCL 3 0.3 500 250500 output 0.3 2 1002 表 4 算例设置
Table 4. Parameters of cases
Case Model Parameters Training dataset Testing dataset F1 FCN Pressure P 4325 2875 M1 MLP Pressure P 4311 2889 F2 FCN velocity U 4316 2884 M2 MLP velocity U 4328 2872 F3 FCN velocity V 4320 2880 M3 MLP velocity V 4321 2879 F4 FCN velocity W 4326 2874 M4 MLP velocity W 4308 2892 F5 FCN magnitude 4323 2877 M5 MLP magnitude 4328 2872 F6 FCN vorticity 4307 2893 M6 MLP vorticity 4317 2883 -
[1] 章新华. 水下目标自动识别的核心技术. 兵工学报, 1998, 3: 275-280 (Zhang Xinhua. A survey on the core techniques in the automatic recognition of sonar targets. Acta Armamentarii, 1998, 3: 275-280 (in Chinese) [2] 朱进, 章新华. 被动声呐目标识别技术的现状与展望. 舰船科学技术, 2003, 25 (5): 55-58 (Zhu Jin, Zhang Xinhua. Status and prospects on techniques of recognition for passive sonar targets. Ship Science and Technology, 2003, 25 (5): 55-58 (in Chinese) [3] 冀航, 马泳, 杨克成. 载波调制激光雷达水下目标探测的仿真分析. 光学技术, 2007, 33(5): 648-650 (Ji Hang, Ma Yong, Yang Kecheng. Simulation analysis of modulated lidar on optical carrier for target detection in water. Optical Technique, 2007, 33(5): 648-650 (in Chinese) doi: 10.3321/j.issn:1002-1582.2007.05.028 [4] 吕沛, 周仁魁, 何俊华等. 水下单像素成像系统研究. 光电子.激光, 2011, 22(9): 1425-1430 (Lu Pei, Zhou Renkui, He Junhua, et al. Research on underwater single-pixel imaging system. Journal of Optoelectronics·laser, 2011, 22(9): 1425-1430 (in Chinese) [5] 王菲, 曾庆军, 黄国建等. 基于调制线谱特征提取的被动声呐目标识别技术研究. 舰船科学技术, 2002(2): 36-41 (Wang Fei, Zeng Qingjun, Huang Guojian, et al. Research on technique of passive sonar target recognition based modulation line spectrum feature extraction. Ship Science and Technology, 2002(2): 36-41 (in Chinese) [6] Kutz JN. Deep learning in fluid dynamics. Journal of Fluid Mechanics, 2017, 814: 1-4 doi: 10.1017/jfm.2016.803 [7] Hong F, Liu C, Guo L, et al. Underwater acoustic target recognition with a residual network and the optimized feature extraction method. Applied Sciences, 2021, 11(4): 1442-1451 doi: 10.3390/app11041442 [8] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions//IEEE Computer Society, Los Alamitos, CA, USA, 2015 [9] Lecun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436-444 [10] 金磊磊, 梁红, 杨长生. 基于卷积神经网络的水下目标声呐图像识别方法. 西北工业大学学报, 2021, 39(2): 285-291 (Jin Leilei, Liang Hong, Yang Changsheng. Sonar image recognition of underwater target based on convolutional neural network. Journal of Northwestern Polytechnical University, 2021, 39(2): 285-291 (in Chinese) doi: 10.1051/jnwpu/20213920285 [11] 唐旭东, 朱炜, 庞永杰等. 水下机器人光视觉目标识别系统. 机器人, 2009, 31(2): 171-178 (Tang Xudong, Zhu Wei, Pang Yongjie, et al. Target recognition system based on optical vision for AUV. Robot, 2009, 31(2): 171-178 (in Chinese) doi: 10.3321/j.issn:1002-0446.2009.02.012 [12] Yan J, Sun H, Chen H, et al. Resonance-based time-frequency manifold for feature extraction of ship-radiated noise. Sensors (Basel, Switzerland) , 2018, 18(4): 936 doi: 10.3390/s18040936 [13] Yang H, Shen S, Yao X, et al. Competitive deep-belief networks for underwater acoustic target recognition. Sensors (Basel, Switzerland) , 2018, 18(4): 952 doi: 10.3390/s18040952 [14] 王年华, 鲁鹏, 常兴华等. 基于机器学习的非结构网格阵面推进生成技术初探. 力学学报, 2021, 53(3): 740-751 (Wang Nianhua, Lu Peng, Chang Xinghua, et al. Preliminary investigation on unstructured mesh generation technique based on advancing front method and machine learning methods. Chinese Journal of Theoretical and Applied Mechanics, 2021, 53(3): 740-751 (in Chinese) doi: 10.6052/0459-1879-20-402 [15] 张珍, 叶舒然, 岳杰顺等. 基于组合神经网络的雷诺平均湍流模型多次修正方法. 力学学报, 2021, 53(6): 1532-1542 (Zhang Zhen, Ye Shuran, Yue Jieshun, et al. A combined neural network and multiple modification strategy for reynolds-averaged navier-stokes turbulence modeling. Chinese Journal of Theoretical and Applied Mechanics, 2021, 53(6): 1532-1542 (in Chinese) doi: 10.6052/0459-1879-21-073 [16] 江昊, 王伯福, 卢志明. 基于数据驱动的流场控制方程的稀疏识别. 力学学报, 53(6): 9(Jiang Hao, Wang Bofu, Lu Zhiming. Data-driven sparse identification of governing equations for fluid dynamics. Chinese Journal of Theoretical and Applied Mechanics, 53(6): 9 (in Chinese)) [17] 谢晨月, 袁泽龙, 王建春等. 基于人工神经网络的湍流大涡模拟方法. 力学学报, 2021, 53(1): 1-16 (Xie Chenyue, Yuan Zelong, Wang Jianchun, et al. Artificial neural network-based subgrid-scale models for large-eddy simulation of turbulence. Chinese Journal of Theoretical and Applied Mechanics, 2021, 53(1): 1-16 (in Chinese) [18] Ling JL, Kurzawski A, Templeton, et al. Reynolds averaged turbulence modelling using deep neural networks with embedded invariance. Journal of Fluid Mechanics, 2016, 807: 155-166 [19] Maulik R, San O, Jacob JD, et al. Sub-grid scale model classification and blending through deep learning. Journal of Fluid Mechanics, 2019, 870: 784-812 doi: 10.1017/jfm.2019.254 [20] Ren F, Wang C, Tang H. Active control of vortex-induced vibration of a circular cylinder using machine learning. Physics of Fluids, 2019, 31(9): 93601 doi: 10.1063/1.5115258 [21] Ren F, Wang CL, Tang H. Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth. Physics of Fluids, 2020, 33: 0060690 [22] Huang J, Liu H, Cai W. Online in situ prediction of 3-D flame evolution from its history 2-D projections via deep learning. Journal of Fluid Mechanics, 2019, 875: 545 [23] Zhang Y, Azman AN, Xu KW, et al. Two-phase flow regime identification based on the liquid-phase velocity information and machine learning. Experiments in Fluids, 2020, 61(10): 212 [24] Raissi M, Wang Z, Triantafyllou MS, et al. Deep learning of vortex-induced vibrations. Journal of Fluid Mechanics, 2019, 861: 119-137 doi: 10.1017/jfm.2018.872 [25] Jin X, Cheng P, Chen WL, et al. Prediction model of velocity field around circular cylinder over various Reynolds numbers by fusion convolutional neural networks based on pressure on the cylinder. Physics of Fluids, 2018, 30(4): 47105 doi: 10.1063/1.5024595 [26] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 2014, 15(1): 1929-1958 [27] Alom MZ, Taha TM, Yakopcic C, et al. The history began from alexNet: A comprehensive survey on deep learning approaches. Computer Vision and Pattern Recognition. arXiv: 1803.01164 cs.cv, 2018 [28] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions//IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 2015: 1-9 [29] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv: 1409.1556, 2014 [30] Lecun Y, Kavukcuoglu K, Farabet C. Convolutional networks and applications in vision//IEEE, Paris, 2010 [31] 战庆亮, 周志勇, 葛耀君. Re = 3900圆柱绕流的三维大涡模拟. 哈尔滨工业大学学报, 2015, 47(12): 75-79 (Zhan Qingliang, Zhou Zhiyong, Ge Yaojun. 3-Dimensional large eddy simulation of circular cylinder at Re = 3900. Journal of Harbin Institute of Technology, 2015, 47(12): 75-79 (in Chinese) [32] Wang Z, Yan W, Oates T. Time series classification from scratch with deep neural networks: a strong baseline//International Joint Conference on Neural Networks, Anchorage, AK, USA, 14-19 MAY 2017: 1578-1585 -