基于组合神经网络的雷诺平均湍流模型 多次修正方法

张珍; 叶舒然; 岳杰顺; 王一伟; 黄晨光

doi:10.6052/0459-1879-21-073

基于组合神经网络的雷诺平均湍流模型多次修正方法

doi: 10.6052/0459-1879-21-073

张珍^*,†,
叶舒然^*,†,
岳杰顺^*,†,
王一伟^*,†,2), ,,
黄晨光^*,†,**

^*中国科学院力学研究所流固耦合系统力学重点实验室, 北京 100190
^†中国科学院大学工程科学学院, 北京 100049
^**中国科学院合肥物质科学研究院, 合肥 230031

基金项目: 1)国家自然科学基金资助项目(11772340);国家自然科学基金资助项目(11802311);国家自然科学基金资助项目(11672315)

详细信息

作者简介:
2)王一伟, 研究员, 主要研究方向: 高速水动力学、人工智能在流体力学中的应用. E-mail: wangyw@imech.ac.cn

通讯作者:
王一伟

中图分类号: O357
计量
- 文章访问数: 1916
- HTML全文浏览量: 418
- PDF下载量: 416
- 被引次数: 0
出版历程
- 收稿日期: 2021-02-18
- 刊出日期: 2021-06-01

A COMBINED NEURAL NETWORK AND MULTIPLE MODIFICATION STRATEGY FOR REYNOLDS-AVERAGED NAVIER-STOKES TURBULENCE MODELING

Zhang Zhen^*,†,
Ye Shuran^*,†,
Yue Jieshun^*,†,
Wang Yiwei^{*,†,2)
, ,},
Huang Chenguang^*,†,**

^*Key Laboratory Mechanics in Fluid Solid Coupling Systems, Institute of Mechanics, Chinese Academy of Sciences, Beijing 100190, China
^†School of Engineering Science, University of Chinese Academy of Sciences, Beijing 100049, China
^**Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China

摘要

摘要: 求解雷诺平均(Reynolds-averaged Navier-Stokes, RANS)方程依然是工程应用中有效且实用的方法, 但对雷诺应力建模的不确定性会导致该方法的预测精度具有很大差异. 随着人工智能的发展, 湍流闭合模型结合机器学习元素的数据驱动方法被认为是提高RANS模型预测性能的有效手段, 然而这种数据驱动方法的稳定性和预测精度仍有待进一步提高. 本文通过构建一个全连接神经网络对RANS方程中的涡黏系数进行预测以实现雷诺应力的隐式求解,该神经网络记作涡黏系数神经网络(eddy viscosity neural network, EVNN). 此外, 也使用张量基神经网络(tensor basis neural network, TBNN)预测未封闭量与解析量之间的高阶涡黏关系, 并利用基张量保证伽利略不变性. 最后, 采用多次修正的策略实现修正模型对流场预测的精度闭环. 上述方法使用大涡模拟(large eddy simulation, LES)方法产生的高保真数据, 以及RANS模拟获得的基线数据对由EVNN和TBNN组合的神经网络进行训练, 然后用训练好的模型预测新的RANS模拟的流场. 通过与高保真LES结果进行对比, 结果表明, 相比于原始RANS模型, 修正模型对后验速度场、下壁面平均压力系数和摩擦力系数的预测精度均有较大提升. 可以发现对雷诺应力线性部分的隐式处理可以增强数值求解的稳定性, 对雷诺应力非线性部分的修正可以提升模型对流场各向异性特征预测的性能, 并且多次修正后的模型表现出更高的预测精度. 因此, 该算法在数据驱动湍流建模和工程应用中具有很大的应用潜力.
- 组合神经网络 /
- 雷诺平均 /
- 各向异性 /
- 隐式修正 /
- 精度闭环
Abstract: Solving the Reynolds-averaged Navier-Stokes (RANS) equation remains an effective and practical approach in engineering applications, but the uncertainty of Reynolds stress modeling will lead to discrepancies in the prediction accuracy of this approach. With the development of artificial intelligence, the data-driven method of turbulence model combined with machine learning algorithm is more effective than the original RANS model, however, the stability and prediction accuracy of the data-driven method could still be further improved. In the present paper, a fully connected neural network is constructed to predict the eddy viscosity, and this neural network is called as Eddy Viscosity Neural Network (EVNN). Additionally, a tensor-based neural network (TBNN) is also applied to predict the higher-order eddy viscosity relationship between the unclosed quantity and the analytical quantity, and the basis tensors are used to ensure the Galilean invariance. Finally, the closed-loop accuracy of the predicted flow field is realized through multiple modifications. For the method above, the neural network which is combined by EVNN and TBNN, is trained by using the high-fidelity data generated by the large eddy simulation (LES) and the baseline data obtained by the RANS simulation. Compared with the high-fidelity LES results, the results of the modified model exhibit significantly higher accuracy in the posterior velocity field, the mean pressure coefficient, and the mean friction coefficient than the original RANS model. It can be found that the implicit treatment of the linear part of the Reynolds stress can enhance the numerical stability, and the modification of the nonlinear part of the Reynolds stress can better predict the anisotropic characteristics of the flow field. Furthermore, the prediction accuracy is further improved through the multiple modification strategy. Therefore, the combined neural network and multiple modification strategy developed in this paper, have strong potentials in data-driven turbulence modeling and engineering applications in the future.
- combined neural network /
- Reynolds average /
- anisotropy /
- implicit modification /
- closed-loop accuracy

HTML全文

参考文献(51)

[1]	Durbin PA. Some recent developments in turbulence closure modeling. Annual Review of Fluid Mechanics, 2018, 50: 77-103
[2]	Launder BE, Sharma BI. Application of the energy dissipation model of turbulence to the calculation of flow near a spinning disc. Letters in Heat and Mass Transfer, 1974, 1(2): 131-137
[3]	强光林, 杨易, 陈阵等. 基于车身绕流的低雷诺数湍流模型改进研究. 力学学报, 2020, 52(5): 1371-1382 (Qiang Guanglin, Yang Yi, Chen Zhen, et al. Research on improvements of LRN turbulence model based on flow around automobile body. Chinese Journal of Theoretical and Applied Mechanics, 2020, 52(5): 1371-1382 (in Chinese))
[4]	Spalart PR. A one-equation turbulence model for aerodynamic flows//30th AIAA Aerospace Sciences Meeting & Exhibit, 1992: 429
[5]	Menter FR. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA Journal, 1994, 32(8): 1598-1605
[6]	Hamlington PE, Dahm WJA. Reynolds stress closure for non-equilibrium effects in turbulent flows. Physics of Fluids, 2008, 20(11): 107-300
[7]	Gatski TB, Speziale CG. On explicit algebraic stress models for complex turbulent flows. Journal of Fluid Mechanics, 1993, 254: 59-78
[8]	Abbasi S, Pirker S, Lichtenegger T. Application of recurrence CFD (rCFD) to species transport in turbulent vortex shedding. Computers & Fluids, 2020, 196: 104348
[9]	王巍, 唐滔, 卢盛鹏等. 主动射流控制水翼空化的数值模拟与分析. 力学学报, 2019, 51(6): 1752-1760 (Wang Wei, Tang Tao, Lu Shengpeng, et al. Numerical simulation and analysis of active jet control of hydrofoil cavitation. Chinese Journal of Theoretical and Applied Mechanics, 2019, 51(6): 1752-1760 (in Chinese))
[10]	Vahajia S, Han J, Cheung S, et al. Numerical investigation on the bubble size distribution around NACA0015 hydrofoil. Ocean Engineering, 2019, 172: 59-71
[11]	张佳悦, 李达钦, 吴钦等. 航行体回收垂直入水空泡流场及水动力特性研究. 力学学报, 2019, 51(3): 803-812 (Zhang Jiayue, Li Daqin, Wu Qin, et al. Numerical investigation on cavity structures and hyrodynamics of the vehicle during vertical water-entry. Chinese Journal of Theoretical and Applied Mechanics, 2019, 51(3): 803-812 (in Chinese))
[12]	Beetham S, Capecelatro J. Biomass pyrolysis in fully-developed turbulent riser flow. Renewable Energy, 2019, 140: 751-760
[13]	Capecelatro J, Desjardins O, Fox RO. On fluid-particle dynamics in fully developed cluster-induced turbulence. Journal of Fluid Mechanics, 2015, 780: 578-635
[14]	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, 53(2): 339-351 (Yue Jieshun, Quan Xiaobo, Ye Shuran, et al. A multi-scale network for the prediction of hydrodynamics in underwater. Chinese Journal of Theoretical and Applied Mechanics, 2020, 53(2): 339-351 (in Chinese))
[15]	Pope SB. A more general effective-viscosity hypothesis. Journal of Fluid Mechanics, 1975, 72(2): 331-340
[16]	Craft TJ, Launder BE, Suga K. Development and application of a cubic eddyviscosity model of turbulence. International Journal of Heat and Fluid Flow, 1996, 17(2): 108-115
[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]	Brunton SL, Noack BR, Koumoutsakos P. Machine learning for fluid mechanics. Annual Review of Fluid Mechanics, 2020, 52: 477-508
[19]	Brenner MP, Eldredge JD, Freund JB. Perspective on machine learning for advancing fluid mechanics. Physical Review Fluids, 2019, 4(10): 100501
[20]	Duraisamy K, Iaccarino G, Xiao H. Turbulence modeling in the age of data. Annual Review of Fluid Mechanics, 2019, 51: 357-377
[21]	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
[22]	Zhang Z, Song XD, Ye SR, et al. Application of deep learning method to Reynolds stress models of channel flow based on reduced-order modeling of DNS data. Journal of Hydrodynamics, 2019, 31(1): 58-65
[23]	K?hler F, Munz J, Sch?fer M. Data-driven augmentation of RANS turbulence models for improved prediction of separation in wall-bounded flows// AIAA Scitech 2020 Forum, 2020
[24]	Parmar B, Peters E, Jansen KE, et al. Generalized non-linear eddy viscosity models for data-assisted Reynolds stress closure//AIAA Scitech 2020 Forum, 2020
[25]	Rajabi E, Kavianpour MR. Intelligent prediction of turbulent flow over backward-facing step using direct numerical simulation data. Engineering Applications of Computational Fluid Mechanics, 2012, 6(4): 490-503
[26]	Xiao H, Wu JL, Laizet S, et al. Flows over periodic hills of parameterized geometries: A dataset for data-driven turbulence modeling from direct simulations. Computers & Fluids, 2020, 200: 104431
[27]	Wang JX, Wu JL, Xiao H. Physics-informed machine learning approach for reconstructing Reynolds stress modeling discrepancies based on DNS data. Physical Review Fluids, 2017, 2(3): 034603
[28]	Weatheritt J, Sandberg RD. A novel evolutionary algorithm applied to algebraic modifications of the RANS stress-strain relationship. Journal of Computational Physics, 2016, 325: 22-37
[29]	Weatheritt J, Sandberg RD. The development of algebraic stress models using a novel evolutionary algorithm. International Journal of Heat and Fluid Flow, 2017, 68: 298-318
[30]	Duraisamy K, Zhang ZJ, Singh AP. New approaches in turbulence and transition modeling using data-driven techniques//53rd AIAA Aerospace Sciences Meeting, 2015: 1284
[31]	Ling J, Ruiz A, Lacaze G, et al. Uncertainty analysis and data-driven model advances for a jet-in-crossflow. Journal of Turbomachinery, 2017, 139(2): 021008
[32]	King R, Hennigh O, Mohan A, et al. From deep to physics-informed learning of turbulence: Diagnostics. 2018, arXiv:1810.07785
[33]	Maulik R, San O. A neural network approach for the blind deconvolution of turbulent flows. Journal of Fluid Mechanics, 2017, 831: 151-181
[34]	Bode M, Gauding M, Kleinheinz K, et al. Deep learning at scale for subgrid modeling in turbulent flows: regression and reconstruction//International Conference on High Performance Computing: High Performance Computing, 2019: 541-560
[35]	Han Q, Li XL, Yu CP. Subgrid-scale model based on the vorticity gradient tensor for rotating turbulent flows. Acta Mechanica Sinica, 2020, 36(3): 692-700
[36]	Wang CH, Ge MW. Applying resolved-scale linearly forced isotropic turbulence in rational subgrid-scale modeling. Acta Mechanica Sinica, 2019, 35(3): 486-494
[37]	Yin YH, Yang P, Zhang YF, et al. Feature selection and processing of turbulence modeling based on an artificial neural network. Physics of Fluids, 2020, 32(10): 105117
[38]	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
[39]	Yang M, Xiao Z. Improving the k-$omega $-$gamma $-Ar transition model by the field inversion and machine learning framework. Physics of Fluids, 2020, 32(6): 064101
[40]	Zhang ZJ, Duraisamy K. Machine learning methods for data-driven turbulence modeling//22nd AIAA Computational Fluid Dynamics Conference, 2015: 2460
[41]	Parish EJ, Duraisamy K. A paradigm for data-driven predictive modeling using field inversion and machine learning. Journal of Computational Physics, 2016, 305: 758-774
[42]	Wu JL, Xiao H, Sun R, et al. Reynolds-averaged Navier-Stokes equations with explicit data-driven Reynolds stress closure can be ill-conditioned. Journal of Fluid Mechanics, 2019, 869: 553-586
[43]	Beetham S, Capecelatro J. Formulating turbulence closures using sparse regression with embedded form invariance. Physical Review Fluids, 2020, 5(8): 084611
[44]	Gatski TB, Speziale CG. On explicit algebraic stress models for complex turbulent flows. Journal of Fluid Mechanics, 1993, 254: 59-78
[45]	Pope SB. Turbulent Flows. Cambridge: Cambridge University Press, 2000
[46]	Maas AL, Hannun AY, Ng AY. Rectifier nonlinearities improve neural network acoustic models//30th International Conference on Machine Learning (ICML), 2013
[47]	Krogh A, Hertz JA. A simple weight decay can improve generalization//4th International Conference on Neural Information Processing Systems, 1992: 950-957
[48]	Breuer M, Peller N, Rapp C, et al. Flow over periodic hills-numerical and experimental study in a wide range of Reynolds numbers, Computers & Fluids, 2009, 38(2): 433-457
[49]	时北极, 何国威, 王士召. 基于滑移速度壁模型的复杂边界湍流大涡模拟. 力学学报, 2019, 51(3): 754-766 (Shi Beiji, He Guowei, Wang Shizhao. Large-eddy simulation of flows with complex geometries by using the slip-wall model. Chinese Journal of Theoretical and Applied Mechanics, 2019, 51(3): 754-766 (in Chinese))
[50]	Temmerman L, Leschziner MA. Large eddy simulation of separated flow in a streamwise periodic channel construction//International Symposium on Turbulence and Shear Flow Phenomena, 2001: 399-404
[51]	吴霆, 时北极, 王士召等. 大涡模拟的壁模型及其应用. 力学学报, 2018, 50(3): 453-466 (Wu Ting, Shi Beiji, Wang Shizhao, et al. Wall-model for large-eddy simulation and its applications. Chinese Journal of Theoretical and Applied Mechanics, 2018, 50(3): 453-466 (in Chinese))