跨声速流动替代湍流模型的一种神经网络方法

朱林阳; 孙旭翔; 刘溢浪; 张伟伟

doi:10.1007/s10409-021-09057-z

One neural network approach for the surrogate turbulence model in transonic flows

doi: 10.1007/s10409-021-09057-z

Zhu Linyang^{1, 2},
Sun Xuxiang¹,
Liu Yilang³,
Zhang Weiwei^{1,*
, ,}

1.
School of Aeronautics, Northwestern Polytechnical University, Xi’an 710072, China
2.
State Key Laboratory of Aerodynamics, China Aerodynamics R&D Center, Mianyang 621000, China
3.
Department of Engineering Mechanics, Zhejiang University, Hangzhou 310027, China

Funds:

the National Numerical Wind Tunnel Project Grant

and the National Natural Science Foundation of China Grant

More Information

Corresponding author: Zhang Weiwei, E-mail address: aeroelastic@nwpu.edu.cn (Weiwei Zhang)

摘要

摘要: 随着深度神经网络等人工智能技术的发展, 数据驱动的机器学习方法也开始广泛应用于湍流模型的改进和构建中. 针对航空工程中的高雷诺数湍流, 我们之前的工作已经建立了适用于不同来流条件下二维亚声速翼型绕流的数据驱动模型. 本工作中, 采用全连接神经网络构建了三维跨声速机翼绕流中涡黏封闭的代理模型, 这在现有的工作中研究较少. 所构建模型仅基于ONERA-M6机翼两个状态的数据集, 并与Navier-Stokes方程求解器耦合. 采用四个不同机翼的亚、跨声速标模来测试模型性能. 迭代过程稳定并获得了收敛解, 实现了对原RANS模型的替代. 通过与SA模型计算的源数据对比, 结果表明, 所提出的模型能够很好地泛化到测试算例中. 在所有算例中, 截面上摩擦阻力系数的平均误差小于4%. 本工作表明将数据驱动方法应用于湍流模型化是可行的, 本文的建模模式是有效的.
Abstract: With the rapid development of artificial intelligence techniques such as neural networks, data-driven machine learning methods are popular in improving and constructing turbulence models. For high Reynolds number turbulence in aerodynamics, our previous work built a data-driven model applicable to subsonic airfoil flows with different free stream conditions. The results calculated by the proposed model are encouraging. In this work, we aim to model the turbulence of transonic wing flows with fully connected deep neural networks, where there is less research at present. The proposed model is driven by two flow cases of the ONERA (Office National d’Etudes et de Recherches Aérospatiales) wing and coupled with the Navier-Stokes equation solver. Four subcritical and transonic benchmark cases of different wings are used to evaluate the model performance. The iteration process is stable, and final convergence is achieved. The proposed model can be used to surrogate the traditional Reynolds averaged Navier-Stokes turbulence model. Compared with the data calculated by the Spallart-Allmaras model, the results show that the proposed model can be well generalized to the test cases. The mean relative error of the drag coefficient at different sections is below 4% for each case. This work demonstrates that modeling turbulence by data-driven methods is feasible and that our modeling pattern is effective.
- Deep neural network /
- Turbulence modeling /
- Transonic /
- High Reynolds number

HTML全文

1. Schematic of the adopted mixing length.

下载: 全尺寸图片幻灯片

2. Schematic diagrams of a fully connected deep neural network and b single neuron.

下载: 全尺寸图片幻灯片

3. Schematic of some activation functions.

下载: 全尺寸图片幻灯片

4. Coupling of the proposed model with CFD solver.

下载: 全尺寸图片幻灯片

5. The error evolution of the training and validation sets (refer to the left y-axis) and the evolution of stability loss (refer to the right y-axis).

下载: 全尺寸图片幻灯片

6. View of meshes over the ONERA-M6 wing.

下载: 全尺寸图片幻灯片

7. Comparison of C_p distributions at several sections obtained with the SA model and the ML model as well as the experimental results for case P1.

下载: 全尺寸图片幻灯片

8. Comparison of C_f distributions at several sections obtained with the SA model and ML model for case P1.

下载: 全尺寸图片幻灯片

9. Comparison of C_f distributions at several sections obtained with the SA model and the ML model for case P2.

下载: 全尺寸图片幻灯片

10. Comparison of C_f contours at the upper surface of the ONERA-M6 wing for cases P1 and P2.

下载: 全尺寸图片幻灯片

11. Comparison of C_f distributions at several sections obtained with the SA model and the ML model for case P3.

下载: 全尺寸图片幻灯片

12. Comparison of C_f distributions at several sections obtained with the SA model and ML model for case P4.

下载: 全尺寸图片幻灯片

13. Comparison of C_f contours at the upper surface of the DPW-W1/W2 wing for cases P3 and P4.

下载: 全尺寸图片幻灯片

14. Comparison of friction drag coefficient C_d,f at several sections for test cases.

下载: 全尺寸图片幻灯片

15. Coupling of the data-driven turbulence model with the CFD solver.

下载: 全尺寸图片幻灯片

16. Residual history of density for test cases.

下载: 全尺寸图片幻灯片

Table 1. The flow features used as the regression input, where ${‖ \cdot ‖}_{\infty}$ denotes the infinity norm

Feature	Description	Sign
Q1	Density	ρ
Q2	Nonorthogonality between velocity and its gradient	$\| u_{i} u_{j} \frac{\partial u_{i}}{\partial x_{j}} \| / {‖ \nabla U ‖}_{\infty}$
Q3	Pressure gradient along a streamline	$u_{i} \frac{\partial P}{\partial x_{i}}$
Q4	Entropy	S′
Q5	Local velocity scaled by the friction velocity	$\sqrt{u^{2} + v^{2}} / u_{τ}$
Q6	Local velocity	$\sqrt{u^{2} + v^{2}}$
Q7	Strain rate	${‖ S^{+} ‖}_{\infty}$
Q8	Rotation rate	${‖ R^{+} ‖}_{\infty}$

下载: 导出CSV

Table 2. The training parameters

Parameters	Value
Batch normalization [48]	Adopted
Activation function	Tanh
Optimizer	Adamax
Learning rates	ReduceLROnPlateau
λ₁	1×10⁻⁷
λ₂	2×10⁻⁶
λ₃	1

下载: 导出CSV

Table 3. Spanwise locations of different wings

Section	y/b
Section	M6	DPW-W1/DPW-W2
1234567	0.200.440.650.800.900.950.99	0.0260.1570.2980.4200.5110.6820.945

下载: 导出CSV

Table 4. The training, validation and test cases

	Case	Ma	α (°)	Re (×10⁶)	Wing
Training	T1	0.76	1.25	11.72	ONERA-M6
Training	T2	0.79	3.70	11.72	ONERA-M6
Validation	−	0.83	3.02	11.72	ONERA-M6
Test	P1	0.84	3.06	11.72	ONERA-M6
	P2	0.699	3.06	11.72	ONERA-M6
	P3	0.76	0.5	5	DPW-W1
	P4	0.76	0.5	5	DPW-W2

下载: 导出CSV

参考文献(64)

[1]	H. Zhou, and H. X. Zhang, What is the essence of the so-called century lasting difficult problem in classic physics, the “problem of turbulence”?, Sci. Sin.-Phys. Mech. Astron. 42, 1 (2012).
[2]	R. Han, Y. Wang, Y. Zhang, and G. Chen, A novel spatial-temporal prediction method for unsteady wake flows based on hybrid deep neural network, Phys. Fluids 31, 127101 (2019).
[3]	X. Jin, P. Cheng, W. L. Chen, and H. Li, Prediction model of velocity field around circular cylinder over various Reynolds numbers by fusion convolutional neural networks based on pressure on the cylinder, Phys. Fluids 30, 047105 (2018).
[4]	M. Xu, S. Song, X. Sun, and W. Zhang, A convolutional strategy on unstructured mesh for the adjoint vector modeling, Phys. Fluids 33, 036115 (2021).
[5]	Y. Zhang, W. Chan, and N. Jaitly, Very deep convolutional networks for end-to-end speech recognition, in: IEEE International Conference on Acoustics Speech and Signal Processing (ICASSP), (New York, 2017), p. 4845
[6]	T. Y. Lin, A. RoyChowdhury, and S. Maji, Bilinear convolutional neural networks for fine-grained visual recognition, IEEE Trans. Pattern Anal. Mach. Intell. 40, 1309 28692962(2018).
[7]	X. Wen, Y. Liu, Z. Li, Y. Chen, and D. Peng, Data mining of a clean signal from highly noisy data based on compressed data fusion: A fast-responding pressure-sensitive paint application, Phys. Fluids 30, 097103 (2018).
[8]	V. Sekar, Q. Jiang, C. Shu, and B. C. Khoo, Fast flow field prediction over airfoils using deep learning approach, Phys. Fluids 31, 057103 (2019).
[9]	B. Liu, J. Tang, H. Huang, and X. Y. Lu, Deep learning methods for super-resolution reconstruction of turbulent flows, Phys. Fluids 32, 025105 (2020).
[10]	Z. Deng, C. He, Y. Liu, and K. C. Kim, Super-resolution reconstruction of turbulent velocity fields using a generative adversarial network-based artificial intelligence framework, Phys. Fluids 31, 125111 (2019).
[11]	Z. Deng, Y. Chen, Y. Liu, and K. C. Kim, Time-resolved turbulent velocity field reconstruction using a long short-term memory (LSTM)-based artificial intelligence framework, Phys. Fluids 31, 075108 (2019).
[12]	C. Rao, H. Sun, and Y. Liu, Physics-informed deep learning for incompressible laminar flows, Theor. Appl. Mech. Lett. 10, 207 (2020).
[13]	L. Sun, and J. X. Wang, Physics-constrained bayesian neural network for fluid flow reconstruction with sparse and noisy data, Theor. Appl. Mech. Lett. 10, 161 (2020).
[14]	X. Zhao, L. Du, X. Peng, Z. Deng, and W. Zhang, Research on refined reconstruction method of airfoil pressure based on compressed sensing, Theor. Appl. Mech. Lett. 11, 100223 (2021).
[15]	J. Kou, and W. Zhang, A hybrid reduced-order framework for complex aeroelastic simulations, Aerospace Sci. Tech. 84, 880 (2019).
[16]	W. Zhang, J. Kou, and Z. Wang, Nonlinear aerodynamic reduced-order model for limit-cycle oscillation and flutter, AIAA J. 54, 3304 (2016).
[17]	J. Ling, Using machine learning to understand and mitigate model form uncertainty in turbulence models, in: IEEE International Conference on Machine Learning and Applications, (New York, 2015), p. 813
[18]	J. Ling, and J. Templeton, Evaluation of machine learning algorithms for prediction of regions of high Reynolds averaged Navier Stokes uncertainty, Phys. Fluids 27, 085103 (2015).
[19]	A. P. Singh, S. Medida, and K. Duraisamy, Machine-learning-augmented predictive modeling of turbulent separated flows over airfoils, AIAA J. 55, 2215 (2017).
[20]	J. Ling, A. Kurzawski, and J. Templeton, Reynolds averaged turbulence modelling using deep neural networks with embedded invariance, J. Fluid Mech. 807, 155 (2016).
[21]	H. Xiao, J. L. Wu, and J. X. Wang, Physics informed machine learning for predictive turbulence modeling: Progress and perspectives. in: Proceedings of the 2017 AIAA SciTech (2017)
[22]	B. D. Tracey, K. Duraisamy, and J. J. Alonso, A machine learning strategy to assist turbulence model development, in: 53rd AIAA aerospace sciences meeting, (2015), p. 1287
[23]	L. Zhu, W. Zhang, J. Kou, and Y. Liu, Machine learning methods for turbulence modeling in subsonic flows around airfoils, Phys. Fluids 31, 015105 (2019).
[24]	Z. Wang, K. Luo, D. Li, J. Tan, and J. Fan, Investigations of data-driven closure for subgrid-scale stress in large-eddy simulation, Phys. Fluids 30, 125101 (2018).
[25]	C. Xie, J. Wang, H. Li, M. Wan, and S. Chen, Artificial neural network mixed model for large eddy simulation of compressible isotropic turbulence, Phys. Fluids 31, 085112 (2019).
[26]	C. Xie, J. Wang, H. Li, M. Wan, and S. Chen, Spatial artificial neural network model for subgrid-scale stress and heat flux of compressible turbulence, Theor. Appl. Mech. Lett. 10, 27 (2020).
[27]	K. J. Nathan, Deep learning in fluid dynamics. J. Fluid Mech. 814, 1 (2017)
[28]	X. Yan, J. Zhu, M. Kuang, and X. Wang, Aerodynamic shape optimization using a novel optimizer based on machine learning techniques, Aerospace Sci. Tech. 86, 826 (2019).
[29]	L. Zhu, W. Zhang, X. Sun, Y. Liu, and X. Yuan, Turbulence closure for high Reynolds number airfoil flows by deep neural networks, Aerospace Sci. Tech. 110, 106452 (2021).
[30]	T. Cebeci, A. M. O. Smith, and P. A. Libby, Analysis of turbulent boundary layers, J. Appl. Mech. 43, 189 (1976).
[31]	B. Baldwin, and H. Lomax, Thin-layer approximation and algebraic model for separated turbulent flows, in: 16th Aerospace Sciences Meeting, (1978), p. 257
[32]	H. Oertel, Prandtl-Essentials of Fluid Mechanic (Springer Science & Business Media, Heidelberg, 2010), p. 158
[33]	E. R. Van Driest, On turbulent flow near a wall, J. Aeronaut. Sci. 23, 1007 (1956).
[34]	M. J. Nituch, S. Sjolander, and M. R. Head, An improved version of the Cebeci-Smith eddy-viscosity model, Aeronautical Q. 29, 207 (1978).
[35]	P. S. Granville, A modified Van Driest Formula for the mixing length of turbulent boundary layers in pressure gradients, J. Fluids Eng. 111, 94 (1989).
[36]	Z. S. She, X. Chen, and F. Hussain, Quantifying wall turbulence via a symmetry approach: a Lie group theory, J. Fluid Mech. 827, 322 (2017).
[37]	M. J. Xiao, and Z. S. She, Precise drag prediction of airfoil flows by a new algebraic model, Acta Mech. Sin. 36, 35 (2020).
[38]	F. H. Clauser, The Turbulent Boundary Layer, Advances in Applied Mechanics (Elsevier, Amsterdam, 1956), pp. 1–51
[39]	S. Pirozzoli, Revisiting the mixing-length hypothesis in the outer part of turbulent wall layers: mean flow and wall friction, J. Fluid Mech. 745, 378 (2014).
[40]	G. Maise, and H. McDonald, Mixing length and kinematic eddy viscosity in a compressible boundary layer., AIAA J. 6, 73 (1968).
[41]	Z. Zhang, X. Song, S. Ye, Y. Wang, C. Huang, Y. An, and Y. Chen, Application of deep learning method to Reynolds stress models of channel flow based on reduced-order modeling of DNS data, J. Hydrodyn. 31, 58 (2019).
[42]	Z. J. Zhang, and K. Duraisamy, Machine learning methods for data-driven turbulence modeling, in: 22nd AIAA Computational Fluid Dynamics Conference, (2015), p. 2460
[43]	Y. LeCun, Y. Bengio, and G. Hinton, Deep learning, Nature 521, 436 26017442(2015).
[44]	I. Guyon, and A. Elisseeff, An Introduction to Feature Extraction, Feature Extraction: Foundations and Applications (Springer, Heidelberg, 2006), pp. 1–25
[45]	J. Ling, R. Jones, and J. Templeton, Machine learning strategies for systems with invariance properties, J. Comput. Phys. 318, 22 (2016).
[46]	J. L. Wu, J. X. Wang, H. Xiao, and J. Ling, A priori assessment of prediction confidence for data-driven turbulence modeling, Flow Turbul. Combust. 99, 25 (2017).
[47]	A. Paszke, S. Gross, and S. Chintala, PyTorch: Tensors and dynamic neural networks in Python with strong GPU acceleration (2017)
[48]	S. Ioffe, and C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift, in: International Conference on Machine Learning, (Baltimore, 2015), pp. 448–456
[49]	S. Zheng, Y. Song, T. Leung, and I. Goodfellow, Improving the robustness of deep neural networks via stability training, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, (New York, 2016), pp. 4480–4488
[50]	V. Schmitt, and F. Charpin, Pressure Distributions on the ONERA-M6-wing at Transonic Mach numbers, Experimental Data Base for Computer Program Assessment, (AGARD-AR-138, 1979)
[51]	V. N. Vatsa, Accurate numerical solutions for transonic viscous flow over finite wings, J. Aircraft 24, 377 (1987).
[52]	M. Mani, J. Ladd, A. Cain, R. Bush, M. Mani, J. Ladd, A. Cain, and R. Bush, An assessment of one- and two-equation turbulence models for internal and external flows, in: 28th Fluid Dynamics Conference, (AIAA, Snowmass Village, 1997), p. 2010
[53]	J. Mayeur, A. Dumont, D. Destarac, et al. RANS simulations on TMR test cases and M6 wing with the Onera elsA flow solver. AIAA Paper, 1745 (2015)
[54]	R. G. Silva, J. L. F Azevedo, and E. Basso, Simulation of ONERA M6 Wing Flows for Assessment of Turbulence Modeling Capabilities. AIAA Aerospace Sciences Meeting, 0549 (2016)
[55]	T. Scheidegger, and L.S.G. Zori, in: 3rd AIAA CFD Drag Prediction Workshop Part 2: DPW-W1/W2, 3-4 (2006)
[56]	J. Morrison, Statistical Analysis of CFD Solutions from the Third AIAA Drag Prediction Workshop, in: 45th AIAA Aerospace Sciences Meeting and Exhibit, (AIAA, Reno, 2007), p. 254
[57]	K. B. Thompson, and H. A. Hassan, Simulation of a variety of wings using a Reynolds stress model, J. Aircraft 52, 1668 (2015).
[58]	M. Gamahara, and Y. Hattori, Searching for turbulence models by artificial neural network, Phys. Rev. Fluids 2, 054604 (2017).
[59]	M. Schmelzer, R. P. Dwight, and P. Cinnella, Machine learning of algebraic stress models using deterministic symbolic regression, arXiv: 1905.075101905.07510
[60]	J. Ling, and A. Kurzawski, Data-driven adaptive physics modeling for turbulence simulations. in: 23rd AIAA Computational Fluid Dynamics Conference, (AIAA, San Diego, 2017), p. 3627
[61]	J. X. Wang, J. L. Wu, and H. Xiao, Physics-informed machine learning approach for reconstructing Reynolds stress modeling discrepancies based on DNS data, Phys. Rev. Fluids 2, 034603 (2017).
[62]	J. Weatheritt, R. D. Sandberg, J. Ling, G. Saez, and J. Bodart, A comparative study of contrasting machine learning frameworks applied to RANS modeling of jets in crossflow, in: Asme Turbo Expo: Turbomachinery Technical Conference & Exposition, (Seoul, 2017), p. 50794
[63]	M. A. Cruz, R. L. Thompson, L. E. B. Sampaio, and R. D. A. Bacchi, The use of the Reynolds force vector in a physics informed machine learning approach for predictive turbulence modeling, Comput. Fluids 192, 104258 (2019).
[64]	J. Wu, H. Xiao, R. Sun, and Q. Wang, Reynolds-averaged Navier-Stokes equations with explicit data-driven Reynolds stress closure can be ill-conditioned, J. Fluid Mech. 869, 553 (2019).

施引文献

资源附件(0)

访问统计

点击查看大图

图(16) / 表(4)

计量

文章访问数: 62
HTML全文浏览量: 47
PDF下载量: 0
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

留言板

One neural network approach for the surrogate turbulence model in transonic flows

doi: 10.1007/s10409-021-09057-z

Corresponding author: Zhang Weiwei, E-mail address: aeroelastic@nwpu.edu.cn (Weiwei Zhang)

计量

目录

留言板

One neural network approach for the surrogate turbulence model in transonic flows

doi: 10.1007/s10409-021-09057-z

Corresponding author: Zhang Weiwei, E-mail address: aeroelastic@nwpu.edu.cn (Weiwei Zhang)

计量

出版历程

目录