PHYSICS-ENHANCED DEEP LEARNING METHODS FOR MODELLING AND SIMULATING FLOW FIELDS
-
摘要: 流体运动理论上可用Navier−Stokes方程描述, 但由于对流项带来的非线性, 仅在少数情况可求得方程解析解. 对于复杂工程流动问题, 数值模拟难以高效精准计算高雷诺数流场, 实验或现场测量难以获得流场丰富细节. 近年来, 人工智能技术快速发展, 深度学习等数据驱动技术可利用灵活网络结构, 借助高效优化算法, 获得对高维、非线性问题的强大逼近能力, 为研究流体力学计算方法带来新机遇. 有别于传统图像识别、自然语言处理等典型人工智能任务, 深度学习模型预测的流场需满足流体物理规律, 如Navier−Stokes方程、典型能谱等. 近期, 物理增强的流场深度学习建模与模拟方法快速发展, 正逐渐成为流体力学全新研究范式: 根据流体物理规律选取网络输入特征或设计网络架构的方法称为物理启发的深度学习方法, 直接将流体物理规律显式融入网络损失函数或网络架构的方法称为物理融合的深度学习方法. 研究内容涵盖流体力学降阶模型、流动控制方程求解领域.Abstract: Fluid flows can be theoretically described by the Navier−Stokes equations. However, due to the nonlinear convection term, analytical solutions of the equations can only be obtained for a few cases. For complex engineering flow problems at high Reynolds numbers, it is difficult to calculate the flow field efficiently and accurately by numerical simulation, and it is difficult to obtain rich details by experiment or field measurement. With the rapid development of artificial intelligence technology, data-driven technologies such as deep learning can make use of flexible network structures and efficient optimization algorithms to obtain strong approximating ability for high-dimensional and nonlinear problems, bringing opportunities for the development of computational methods for fluid mechanics. Different from traditional data-driven deep learning modeling methods for image classification and natural language processing, the flow fields predicted by deep learning models should obey physical laws of fluids, such as the Navier−Stokes equations and typical energy spectrum. Recently, physics-enhanced deep learning methods have developed rapidly and are gradually becoming a new research paradigm of fluid mechanics: the method of selecting network input features or designing network architecture according to the laws of fluid physics is called the physics-inspired deep learning method, and the method of explicitly integrating the laws of fluid physics into the network loss function or network architecture is called the physics-informed deep learning method. The research content covers the fields of reduced order modelling of fluid mechanics and solution of flow governing equations.
-
图 4 用于建立压力−速度场模型的卷积神经网络架构. “conv”表示卷积层, “ReLU”是非线性激活函数, “pooling”表示最大池化层, “FC”表示全连接层, “concat”表示不同路径融合[62]
Figure 4. The architecture of the fusion CNN to establish the pressure−velocity model. “conv”denotes convolutional layer; “ReLU” denotes nonlinear rectified linear unit; “pooling” denotes max pooling layer; “FC” denotes fully-connected layer; “concat” denotes the concatenating of different paths[62]
图 5 不同雷诺数下模型预测流场与CFD计算结果对比[62]. CFD计算结果: (a) Re = 65, (c) Re = 170, (e) Re = 500, (g) Re = 1000; 模型预测结果: (b) Re = 65, (d) Re = 170, (f) Re = 500, (h) Re = 1000
Figure 5. Comparisons of instantaneous flow fields between the model predictions and CFD results for various Reynolds numbers[62]. CFD results for (a) Re = 65, (c) Re = 170, (e) Re = 500, and (g) Re = 1000; model predictions for (b) Re = 65, (d) Re = 170, (f) Re = 500, and (h) Re = 1000
表 1 采用RAR增加残差点时NSFnets求解的速度和压力L2相对误差[66]
Table 1. Relative L2 errors of velocity and pressure solutions for NSFnets with residual points added via RAR[66]
Sampling VP-NSFnet VV-NSFnet $ \epsilon_{u} $/% $ \epsilon_{v} $/% $ \epsilon_{p} $/% $ \epsilon_{u} $/% $ \epsilon_{v} $/% Random 0.0095 ± 0.0026 0.0823 ± 0.0167 0.0401 ± 0.0061 0.0254 ± 0.0058 0.0704 ± 0.0160 RAR 0.0056 0.0369 0.0315 0.0083 0.0280 
下载: 导出CSV
-
[1] Krizhevsky A, Sutskever I, Hinton GE. Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 2012, 25: 1097-1105 [2] Cho K, van Merriënboer B, Gulcehre C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint, 2014, arXiv: 1406.1078 [3] Sallab AE, Abdou M, Perot E, et al. Deep reinforcement learning framework for autonomous driving. Electronic Imaging, 2017, 2017(19): 70-76 doi: 10.2352/ISSN.2470-1173.2017.19.AVM-023 [4] Kutz JN. Deep learning in fluid dynamics. Journal of Fluid Mechanics, 2017, 814: 1-4 doi: 10.1017/jfm.2016.803 [5] 张伟伟, 寇家庆, 刘溢浪. 智能赋能流体力学展望. 航空学报, 2021, 42(4): 524689 (Zhang Weiwei, Kou Jiaqing, Liu Yilang. Prospect of artificial intelligence empowered fluid mechanics. Acta Aeronautica et Astronautica Sinica, 2021, 42(4): 524689 (in Chinese) [6] Edeling WN, Cinnella P, Dwight RP. Predictive RANS simulations via bayesian model-scenario averaging. Journal of Computational Physics, 2014, 275: 65-91 doi: 10.1016/j.jcp.2014.06.052 [7] Edeling WN, Cinnella P, Dwight RP, et al. Bayesian estimates of parameter variability in the k-ε turbulence model. Journal of Computational Physics, 2014, 258: 73-94 doi: 10.1016/j.jcp.2013.10.027 [8] Ling J, Kurzawski A, Templeton J. Reynolds averaged turbulence modelling using deep neural networks with embedded invariance. Journal of Fluid Mechanics, 2016, 807: 155-166 doi: 10.1017/jfm.2016.615 [9] Pope SB. Turbulent Flows. Cambridge: Cambridge University Press, 2000 [10] Zhu LY, Zhang WW, Kou JQ, et al. Machine learning methods for turbulence modeling in subsonic flows around airfoils. Physics of Fluids, 2019, 31(1): 015105 doi: 10.1063/1.5061693 [11] Jiang C, Vinuesa R, Chen RL, et al. An interpretable framework of data-driven turbulence modeling using deep neural networks. Physics of Fluids, 2021, 33(5): 055133 doi: 10.1063/5.0048909 [12] Xie CY, Yuan ZL, Wang JC. Artificial neural network-based nonlinear algebraic models for large eddy simulation of turbulence. Physics of Fluids, 2020, 32(11): 115101 doi: 10.1063/5.0025138 [13] Duraisamy K, Iaccarino G, Xiao H. Turbulence modeling in the age of data. Annual Review of Fluid Mechanics, 2019, 51: 357-377 doi: 10.1146/annurev-fluid-010518-040547 [14] 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) [15] 金晓威. 物理启发的钝体绕流场机器学习计算方法. [博士论文]. 哈尔滨: 哈尔滨工业大学, 2021(Jin Xiaowei. Physics-inspired machine learning methods for flow field around a bluff body. [PhD Thesis]. Harbin: Harbin Institute of Technology, 2021 (in Chinese)) [16] Hinton GE, Osindero S, Teh YW. A fast learning algorithm for deep belief nets. Neural Computation, 2006, 18(7): 1527-1554 doi: 10.1162/neco.2006.18.7.1527 [17] Bengio Y, Lamblin P, Popovici D, et al. Greedy layer-wise training of deep networks. Advances in Neural Information Processing Systems, 2007: 153-160 [18] Lecun Y. Generalization and network design strategies. Technical Report CRG-TR-89-4, University of Toronto, 1989 [19] Zhang CW, Yang MY, Zeng HJ, et al. Pedestrian detection based on improved LeNet-5 convolutional neural network. Journal of Algorithms & Computational Technology, 2019, 13: 1-10 [20] Zhang K, Zuo WM, Chen YJ, et al. Beyond a gaussian denoiser: Residual learning of deep CNN for image denoising. IEEE Transactions on Image Processing, 2017, 26(7): 3142-3155 doi: 10.1109/TIP.2017.2662206 [21] Chen YJ, Yu W, Pock T. On learning optimized reaction diffusion processes for effective image restoration//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015: 5261-5269 [22] Goodfellow I, Bengio Y, Courville A. Deep Learning. Cambridge, MIT Press, 2016 [23] Long ZC, Lu YP, Dong B. PDE-Net 2.0: Learning PDEs from data with a numeric-symbolic hybrid deep network. Journal of Computational Physics, 2019, 399: 108925 doi: 10.1016/j.jcp.2019.108925 [24] Cai JF, Dong B, Osher S, et al. Image restoration: total variation, wavelet frames, and beyond. Journal of the American Mathematical Society, 2012, 25(4): 1033-1089 doi: 10.1090/S0894-0347-2012-00740-1 [25] Dong B, Jiang QT, Shen ZW. Image restoration: Wavelet frame shrinkage, nonlinear evolution PDEs, and beyond. Multiscale Modeling & Simulation, 2017, 15(1): 606-660 [26] Rumelhart DE, Hinton GE, Williams RJ. Learning representations by back-propagating errors. Nature, 1986, 323(6088): 533-536 doi: 10.1038/323533a0 [27] Schuster M, Paliwal KK. Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 1997, 45(11): 2673-2681 doi: 10.1109/78.650093 [28] Hochreiter S, Schmidhuber J. Long short-term memory. Neural Computation, 1997, 9(8): 1735-1780 doi: 10.1162/neco.1997.9.8.1735 [29] Goodfellow IJ, Pouget-Abadie J, Mirza M, et al. Generative adversarial networks. Advances in Neural Information Processing Systems, 2014, 3: 2672-2680 [30] Wang XT, Yu K, Wu SX, et al. Esrgan: Enhanced super-resolution generative adversarial networks//Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 2018 [31] Mnih V, Kavukcuoglu K, Silver D, et al. Human-level control through deep reinforcement learning. Nature, 2015, 518(7540): 529-533 doi: 10.1038/nature14236 [32] Silver D, Schrittwieser J, Simonyan K, et al. Mastering the game of go without human knowledge. Nature, 2017, 550(7676): 354-359 doi: 10.1038/nature24270 [33] Cybenko G. Approximation by superpositions of a sigmoidal function. Mathematics of Control, Signals and Systems, 1989, 2(4): 303-314 doi: 10.1007/BF02551274 [34] Barron AR. Approximation and estimation bounds for artificial neural networks. Machine Learning, 1994, 14(1): 115-133 [35] Hornik K. Approximation capabilities of multilayer feedforward networks. Neural Networks, 1991, 4(2): 251-257 doi: 10.1016/0893-6080(91)90009-T [36] Delalleau O, Bengio Y. Shallow vs. deep sum-product networks. Advances in Neural Information Processing Systems, 2011, 24: 666-674 [37] Eldan R, Shamir O. The power of depth for feedforward neural networks//Conference on Learning Theory, 2016: 907-940 [38] Bottou L. Online learning and stochastic approximations. On-line Learning in Neural Networks, 1998, 17(9): 142 [39] Kingma DP, Ba J. Adam: A method for stochastic optimization//The 3rd International Conference on Learning Representations, 2015, arXiv:1412.6980 [40] Du S, Lee J, Li HC, et al. Gradient descent finds global minima of deep neural networks//International Conference on Machine Learning, 2019: 1675-1685 [41] Sirovich L. Turbulence and the dynamics of coherent structures. I. Coherent structures. Quarterly of Applied Mathematics, 1987, 45(3): 561-571 doi: 10.1090/qam/910462 [42] Hudy LM, Naguib A, Humphreys WM. Stochastic estimation of a separated-flow field using wall-pressure-array measurements. Physics of Fluids, 2007, 19(2): 024103 doi: 10.1063/1.2472507 [43] Durgesh V, Naughton JW. Multi-time-delay LSE-POD complementary approach applied to unsteady high-Reynolds-number near wake flow. Experiments in Fluids, 2010, 49(3): 571-583 doi: 10.1007/s00348-010-0821-4 [44] He CX, Liu YZ. Proper orthogonal decomposition-based spatial refinement of TR-PIV realizations using high-resolution non-TR-PIV measurements. Experiments in Fluids, 2017, 58(7): 86 doi: 10.1007/s00348-017-2371-5 [45] Gerhard J, Pastoor M, King R, et al. Model-based control of vortex shedding using low-dimensional Galerkin models//The 33rd AIAA Fluid Dynamics Conference and Exhibit, 2003: 4262 [46] Druault P, Guibert P, Alizon F. Use of proper orthogonal decomposition for time interpolation from PIV data. Experiments in Fluids, 2005, 39(6): 1009-1023 doi: 10.1007/s00348-005-0035-3 [47] Borée J. Extended proper orthogonal decomposition: a tool to analyse correlated events in turbulent flows. Experiments in Fluids, 2003, 35(2): 188-192 doi: 10.1007/s00348-003-0656-3 [48] Hosseini Z, Martinuzzi RJ, Noack BR. Sensor-based estimation of the velocity in the wake of a low-aspect-ratio pyramid. Experiments in Fluids, 2015, 56(1): 13 doi: 10.1007/s00348-014-1880-8 [49] Discetti S, Raiola M, Ianiro A. Estimation of time-resolved turbulent fields through correlation of non-time-resolved field measurements and time-resolved point measurements. Experimental Thermal and Fluid Science, 2018, 93: 119-130 doi: 10.1016/j.expthermflusci.2017.12.011 [50] Discetti S, Bellani G, Örlü R, et al. Characterization of very-large-scale motions in high-Re pipe flows. Experimental Thermal and Fluid Science, 2019, 104: 1-8 doi: 10.1016/j.expthermflusci.2019.02.001 [51] Deng ZW, Chen YJ, Liu YZ, et al. Time-resolved turbulent velocity field reconstruction using a long short-term memory (LSTM)-based artificial intelligence framework. Physics of Fluids, 2019, 31(7): 075108 doi: 10.1063/1.5111558 [52] Cai T, Deng ZW, Park Y, et al. Acquisition of kHz-frequency two-dimensional surface temperature field using phosphor thermometry and proper orthogonal decomposition assisted long short-term memory neural networks. International Journal of Heat and Mass Transfer, 2021, 165: 120662 doi: 10.1016/j.ijheatmasstransfer.2020.120662 [53] Wu P, Sun JW, Chang XT, et al. Data-driven reduced order model with temporal convolutional neural network. Computer Methods in Applied Mechanics and Engineering, 2020, 360: 112766 doi: 10.1016/j.cma.2019.112766 [54] Zhao X, Du L, Peng XH, et al. Research on refined reconstruction method of airfoil pressure based on compressed sensing. Theoretical and Applied Mechanics Letters, 2021, 11(2): 100223 doi: 10.1016/j.taml.2021.100223 [55] Jin XW, Laima SJ, Chen WL, et al. Time-resolved reconstruction of flow field around a circular cylinder by recurrent neural networks based on non-time-resolved particle image velocimetry measurements. Experiments in Fluids, 2020, 61(4): 114 doi: 10.1007/s00348-020-2928-6 [56] 惠心雨, 袁泽龙, 白俊强等. 基于深度学习的非定常周期性流动预测方法. 空气动力学学报, 2019, 37(3): 462-469 (Hui Xinyu, Yuan Zelong, Bai Junqiang, et al. A method of unsteady periodic flow field prediction based on the deep learning. Acta Aeronautica et Astronautica Sinica, 2019, 37(3): 462-469 (in Chinese) doi: 10.7638/kqdlxxb-2019.0003 [57] Sekar V, Jiang Q, Shu C, et al. Fast flow field prediction over airfoils using deep learning approach. Physics of Fluids, 2019, 31(5): 057103 doi: 10.1063/1.5094943 [58] Fukami K, Fukagata K, Taira K. Super-resolution reconstruction of turbulent flows with machine learning. Journal of Fluid Mechanics, 2019, 870: 106-120 doi: 10.1017/jfm.2019.238 [59] Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation//International Conference on Medical Image Computing and Computer-assisted Intervention, 2015: 234-241 [60] Deng ZW, He CX, Liu YZ, et al. Super-resolution reconstruction of turbulent velocity fields using a generative adversarial network-based artificial intelligence framework. Physics of Fluids, 2019, 31(12): 125111 doi: 10.1063/1.5127031 [61] Ledig C, Theis L, Huszár F, et al. Photo-realistic single image super-resolution using a generative adversarial network//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 4681-4690 [62] Jin XW, 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): 047105 doi: 10.1063/1.5024595 [63] Roshko A. Perspectives on bluff body aerodynamics. Journal of Wind Engineering and Industrial Aerodynamics, 1993, 49(1-3): 79-100 doi: 10.1016/0167-6105(93)90007-B [64] Zhu LY, Zhang WW, Tu GH. Generalization enhancement of artificial neural network for turbulence closure by feature selection. Advances in Aerodynamics, 2021, in press [65] Raissi M, Perdikaris P, Karniadakis GE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, 2019, 378: 686-707 doi: 10.1016/j.jcp.2018.10.045 [66] Jin XW, Cai SZ, Li H, et al. NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations. Journal of Computational Physics, 2021, 426: 109951 doi: 10.1016/j.jcp.2020.109951 [67] Griewank A. On automatic differentiation. Mathematical Programming: Recent Developments and Applications, 1989, 6(6): 83-107 [68] Wang SF, Teng YJ, Perdikaris P. Understanding and mitigating gradient pathologies in physics-informed neural networks. SIAM Journal on Scientific Computing, 2021, 43(5): A3055-A3081 doi: 10.1137/20M1318043 [69] Lu L, Meng XH, Mao ZP, et al. DeepXDE: A deep learning library for solving differential equations. SIAM Review, 2021, 63(1): 208-228 doi: 10.1137/19M1274067 [70] E WN, Yu B. The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics, 2018, 6(1): 1-12 doi: 10.1007/s40304-018-0127-z [71] Zang YH, Bao G, Ye XJ, et al. Weak adversarial networks for high-dimensional partial differential equations. Journal of Computational Physics, 2020, 411: 109409 doi: 10.1016/j.jcp.2020.109409 [72] Mao ZP, Jagtap AD, Karniadakis GE. Physics-informed neural networks for high-speed flows. Computer Methods in Applied Mechanics and Engineering, 2020, 360: 112789 doi: 10.1016/j.cma.2019.112789 [73] Jagtap AC, Kharazmi E, Karniadakis GE. Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems. Computer Methods in Applied Mechanics and Engineering, 2020, 365: 113028 doi: 10.1016/j.cma.2020.113028 [74] Wei SY, Jin XW, Li H. General solutions for nonlinear differential equations: a rule-based self-learning approach using deep reinforcement learning. Computational Mechanics, 2019, 64(5): 1361-1374 doi: 10.1007/s00466-019-01715-1 [75] Meng XH, Li Z, Zhang DK, et al. PPINN: Parareal physics-informed neural network for time-dependent PDEs. Computer Methods in Applied Mechanics and Engineering, 2020, 370: 113250 doi: 10.1016/j.cma.2020.113250 [76] Raissi M, Wang ZC, Triantafyllou MS, et al. Deep learning of vortex-induced vibrations. Journal of Fluid Mechanics, 2019, 861: 119-137 doi: 10.1017/jfm.2018.872 [77] Raissi M, Yazdani A, Karniadakis GE. Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations. Science, 2020, 367(6481): 1026-1030 [78] Cai SZ, Wang ZC, Fuest F, et al. Flow over an espresso cup: inferring 3-D velocity and pressure fields from tomographic background oriented Schlieren via physics-informed neural networks. Journal of Fluid Mechanics, 2021, 915: A102 doi: 10.1017/jfm.2021.135 [79] Yang L, Meng XH, Karniadakis GE. B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data. Journal of Computational Physics, 2021, 425: 109913 doi: 10.1016/j.jcp.2020.109913 [80] Bar-Sinai Y, Hoyer S, Hickey J, et al. Learning data-driven discretizations for partial differential equations. Proceedings of the National Academy of Sciences, 2019, 116(31): 15344-15349 doi: 10.1073/pnas.1814058116 [81] Kochkov D, Smith JA, Alieva A, et al. Machine learning–accelerated computational fluid dynamics. Proceedings of the National Academy of Sciences, 2021, 118(21): e2101784118 doi: 10.1073/pnas.2101784118 [82] Zhuang JW, Kochkov D, Bar-Sinai Y, et al. Learned discretizations for passive scalar advection in a two-dimensional turbulent flow. Physical Review Fluids, 2021, 6(6): 064605 doi: 10.1103/PhysRevFluids.6.064605 [83] Wang YF, Shen ZJ, Long ZC, et al. Learning to discretize: solving 1D scalar conservation laws via deep reinforcement learning. Communications in Computational Physics, 2020, 28(5): 2158-2179 doi: 10.4208/cicp.OA-2020-0194 [84] Greenfeld D, Galun M, Basri R, et al. Learning to optimize multigrid PDE solvers//International Conference on Machine Learning, 2019: 2415-2423 [85] Li WJ, Laima SJ, Jin XW, et al. A novel long short-term memory neural-network-based self-excited force model of limit cycle oscillations of nonlinear flutter for various aerodynamic configurations. Nonlinear Dynamics, 2020, 100: 2071-2087 doi: 10.1007/s11071-020-05631-5 [86] Chen YT, Huang D, Zhang DX, et al. Theory-guided hard constraint projection (HCP): A knowledge-based data-driven scientific machine learning method. Journal of Computational Physics, 2021, 445: 110624 doi: 10.1016/j.jcp.2021.110624 [87] Hennigh O, Narasimhan S, Nabian MA, et al. NVIDIA SimNet™: An AI-accelerated multi-physics simulation framework//International Conference on Computational Science, 2021: 447-461 [88] Dong SC, Li ZW. Local extreme learning machines and domain decomposition for solving linear and nonlinear partial differential equations. arXiv preprint, 2020, arXiv: 2012.02895 [89] Shin Y, Darbon J, Karniadakis GE. On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type PDEs. Communications in Computational Physics, 2020, 28(5): 2042-2074 doi: 10.4208/cicp.OA-2020-0193 -
下载:
点击查看大图
图(8) / 表(1)
计量
- 文章访问数: 1858
- HTML全文浏览量: 584
- PDF下载量: 550
- 被引次数: 0
