Skip to main content Accessibility help

Sub-grid scale model classification and blending through deep learning

  • Romit Maulik (a1), Omer San (a1), Jamey D. Jacob (a1) and Christopher Crick (a2)


In this article we detail the use of machine learning for spatio-temporally dynamic turbulence model classification and hybridization for large eddy simulations (LES) of turbulence. Our predictive framework is devised around the determination of local conditional probabilities for turbulence models that have varying underlying hypotheses. As a first deployment of this learning, we classify a point on our computational grid as that which requires the functional hypothesis, the structural hypothesis or no modelling at all. This ensures that the appropriate model is specified from a priori knowledge and an efficient balance of model characteristics is obtained in a particular flow computation. In addition, we also utilize the conditional-probability predictions of the same machine learning to blend turbulence models for another hybrid closure. Our test case for the demonstration of this concept is given by Kraichnan turbulence, which exhibits a strong interplay of enstrophy and energy cascades in the wavenumber domain. Our results indicate that the proposed methods lead to robust and stable closure and may potentially be used to combine the strengths of various models for complex flow phenomena prediction.


Corresponding author

Email address for correspondence:


Hide All
Bardina, J., Ferziger, J. H. & Reynolds, W. C.1980 Improved subgrid-scale models for large-eddy simulation. AIAA Paper 1980-1357.
Beck, A. D., Flad, D. G. & Munz, C.-D.2018 Neural networks for data-based turbulence models. arXiv:1806.04482.
Berselli, L. C., Iliescu, T. & Layton, W. J. 2006 Mathematics of Large Eddy Simulation of Turbulent Flows. Springer.
Bishop, C. M. 2006 Pattern Recognition and Machine Learning (Information Science and Statistics). Springer.
Boffetta, G. & Ecke, R. E. 2012 Two-dimensional turbulence. Annu. Rev. Fluid Mech. 44, 427451.
Canuto, V. M. & Cheng, Y. 1997 Determination of the Smagorinsky–Lilly constant CS. Phys. Fluids 9 (5), 13681378.
Cushman-Roisin, B. & Beckers, J.-M. 2011 Introduction to Geophysical Fluid Dynamics: Physical and Numerical Aspects. Academic Press.
Duraisamy, K., Iaccarino, G. & Xiao, H. 2019 Turbulence modeling in the age of data. Annu. Rev. Fluid Mech. 51, 357377.
Frisch, U. 1995 Turbulence. Cambridge University Press.
Fukami, K., Fukagata, K. & Taira, K.2018 Super-resolution reconstruction of turbulent flows with machine learning. arXiv:1811.11328.
Gamahara, M. & Hattori, Y. 2017 Searching for turbulence models by artificial neural network. Phys. Rev. Fluids 2 (5), 054604.
Germano, M. 2015 The similarity subgrid stresses associated to the approximate Van Cittert deconvolutions. Phys. Fluids 27 (3), 035111.
Germano, M., Piomelli, U., Moin, P. & Cabot, W. H. 1991 A dynamic subgrid-scale eddy viscosity model. Phys. Fluids 3 (7), 17601765.
Grossmann, S. & Mertens, P. 1992 Structure functions in two-dimensional turbulence. Z. Phys. B. Con. Mat. 88 (1), 105116.
Guermond, J.-L., Oden, J. T. & Prudhomme, S. 2004 Mathematical perspectives on large eddy simulation models for turbulent flows. J. Math. Fluid Mech. 6 (2), 194248.
Habisreutinger, M. A., Bouffanais, R., Leriche, E. & Deville, M. O. 2007 A coupled approximate deconvolution and dynamic mixed scale model for large-eddy simulation. J. Comput. Phys. 224 (1), 241266.
Hennigh, O.2017 Lat-net: compressing lattice Boltzmann flow simulations using deep neural networks. arXiv:1705.09036.
Hickel, S., Egerer, C. P. & Larsson, J. 2014 Subgrid-scale modeling for implicit large eddy simulation of compressible flows and shock-turbulence interaction. Phys. Fluids 26 (10), 106101.
Hornik, K., Stinchcombe, M. & White, H. 1989 Multilayer feedforward networks are universal approximators. Neural Netw. 2 (5), 359366.
King, R., Hennigh, O., Mohan, A. & Chertkov, M.2018 From deep to physics-informed learning of turbulence: diagnostics. arXiv:1810.07785.
King, R. N., Hamlington, P. E. & Dahm, W. J. 2016 Autonomic closure for turbulence simulations. Phys. Rev. E 93 (3), 031301.
Kolmogorov, A. N. 1941 The local structure of turbulence in incompressible viscous fluid for very large Reynolds numbers. In Dokl. Akad. Nauk SSSR, vol. 30, pp. 301305. JSTOR.
Kraichnan, R. H. 1967 Inertial ranges in two-dimensional turbulence. Phys. Fluids 10 (7), 14171423.
LaBryer, A., Attar, P. J. & Vedula, P. 2015 A framework for large eddy simulation of Burgers turbulence based upon spatial and temporal statistical information. Phys. Fluids 27 (3), 035116.
Langford, J. A. & Moser, R. D. 1999 Optimal LES formulations for isotropic turbulence. J. Fluid Mech. 398, 321346.
Lapeyre, C. J., Misdariis, A., Cazard, N., Veynante, D. & Poinsot, T. 2019 Training convolutional neural networks to estimate turbulent sub-grid scale reaction rates. Combust. Flame 203, 255264.
Layton, W. & Lewandowski, R. 2003 A simple and stable scale-similarity model for large eddy simulation: energy balance and existence of weak solutions. Appl. Maths Lett. 16 (8), 12051209.
Leith, C. E. 1968 Diffusion approximation for two-dimensional turbulence. Phys. Fluids 11 (3), 671672.
Lilly, D. K. 1992 A proposed modification of the Germano subgrid-scale closure method. Phys. Fluids 4 (3), 633635.
Ling, J., Jones, R. & Templeton, J. 2016a Machine learning strategies for systems with invariance properties. J. Comput. Phys. 318, 2235.
Ling, J. & Kurzawski, A.2017 Data-driven adaptive physics modeling for turbulence simulations. AIAA Paper 2017-3627.
Ling, J., Kurzawski, A. & Templeton, J. 2016b Reynolds averaged turbulence modelling using deep neural networks with embedded invariance. J. Fluid Mech. 807, 155166.
Ling, J. & Templeton, J. 2015 Evaluation of machine learning algorithms for prediction of regions of high Reynolds averaged Navier–Stokes uncertainty. Phys. Fluids 27 (8), 085103.
Mathew, J., Lechner, R., Foysi, H., Sesterhenn, J. & Friedrich, R. 2003 An explicit filtering method for large eddy simulation of compressible flows. Phys. Fluids 15 (8), 22792289.
Maulik, R. & San, O. 2017a A dynamic framework for functional parameterisations of the eddy viscosity coefficient in two-dimensional turbulence. Intl J. Comput. Fluid Dyn. 31 (2), 6992.
Maulik, R. & San, O. 2017b A neural network approach for the blind deconvolution of turbulent flows. J. Fluid Mech. 831, 151181.
Maulik, R. & San, O. 2017c A stable and scale-aware dynamic modeling framework for subgrid-scale parameterizations of two-dimensional turbulence. Comput. Fluids 158, 1138.
Maulik, R., San, O., Rasheed, A. & Vedula, P. 2018 Data-driven deconvolution for large eddy simulations of Kraichnan turbulence. Phys. Fluids 30 (12), 125109.
Maulik, R., San, O., Rasheed, A. & Vedula, P. 2019 Subgrid modelling for two-dimensional turbulence using neural networks. J. Fluid Mech. 858, 122144.
McWilliams, J. C. 1990 The vortices of two-dimensional turbulence. J. Fluid Mech. 219, 361385.
Milano, M. & Koumoutsakos, P. 2002 Neural network modeling for near wall turbulent flow. J. Comput. Phys. 182 (1), 126.
Moser, R. D., Malaya, N. P., Chang, H., Zandonade, P. S., Vedula, P., Bhattacharya, A. & Haselbacher, A. 2009 Theoretically based optimal large-eddy simulation. Phys. Fluids 21 (10), 105104.
Pearson, B. & Fox-Kemper, B. 2018 Log-normal turbulence dissipation in global ocean models. Phys. Rev. Lett. 120 (9), 094501.
Pearson, B., Fox-Kemper, B., Bachman, S. & Bryan, F. 2017 Evaluation of scale-aware subgrid mesoscale eddy models in a global eddy-rich model. Ocean Model. 115, 4258.
Pope, S. B. 2004 Ten questions concerning the large-eddy simulation of turbulent flows. New J. Phys. 6, 35.
Sagaut, P. 2006 Large Eddy Simulation For Incompressible Flows: An Introduction. Springer.
San, O. 2014 A dynamic eddy-viscosity closure model for large eddy simulations of two-dimensional decaying turbulence. Intl J. Comput. Fluid Dyn. 28 (6–10), 363382.
San, O. & Staples, A. E. 2012 High-order methods for decaying two-dimensional homogeneous isotropic turbulence. Comput. Fluids 63, 105127.
San, O., Staples, A. E., Wang, Z. & Iliescu, T. 2011 Approximate deconvolution large eddy simulation of a barotropic ocean circulation model. Ocean Model. 40 (2), 120132.
San, O. & Vedula, P. 2018 Generalized deconvolution procedure for structural modeling of turbulence. J. Sci. Comput. 75 (2), 11871206.
Sarghini, F., De Felice, G. & Santini, S. 2003 Neural networks based subgrid scale modeling in large eddy simulations. Comput. Fluids 32 (1), 97108.
Singh, A. P. & Duraisamy, K. 2016 Using field inversion to quantify functional errors in turbulence closures. Phys. Fluids 28 (4), 045110.
Singh, A. P., Medida, S. & Duraisamy, K. 2017 Machine-learning-augmented predictive modeling of turbulent separated flows over airfoils. AIAA J. 55 (7), 22152227.
Smagorinsky, J. 1963 General circulation experiments with the primitive equations: I. The basic experiment. Mon. Weath. Rev. 91 (3), 99164.
Sotgiu, C., Weigand, B. & Semmler, K. 2018 A turbulent heat flux prediction framework based on tensor representation theory and machine learning. Intl Commun. Heat Mass Transfer 95, 7479.
Stolz, S. & Adams, N. A. 1999 An approximate deconvolution procedure for large-eddy simulation. Phys. Fluids 11 (7), 16991701.
Tabeling, P. 2002 Two-dimensional turbulence: a physicist approach. Phys. Rep. 362 (1), 162.
Tracey, B. D., Duraisamy, K. & Alonso, J. J.2015 A machine learning strategy to assist turbulence model development. AIAA Paper 2015-1287.
Vollant, A., Balarac, G. & Corre, C. 2017 Subgrid-scale scalar flux modelling based on optimal estimation theory and machine-learning procedures. J. Turbul. 18 (9), 854878.
Vorobev, A. & Zikanov, O. 2008 Smagorinsky constant in LES modeling of anisotropic MHD turbulence. J. Theor. Comput. Fluid Dyn. 22 (3–4), 317325.
Vreman, A. W. 2004 An eddy-viscosity subgrid-scale model for turbulent shear flow: algebraic theory and applications. Phys. Fluids 16 (10), 36703681.
Wang, J.-X., Wu, J., Ling, J., Iaccarino, G. & Xiao, H.2017a A comprehensive physics-informed machine learning framework for predictive turbulence modeling. arXiv:1701.07102.
Wang, J.-X., Wu, J.-L. & Xiao, H. 2017b Physics-informed machine learning approach for reconstructing Reynolds stress modeling discrepancies based on DNS data. Phys. Rev. Fluids 2 (3), 034603.
Wang, Z., Luo, K., Li, D., Tan, J. & Fan, J. 2018 Investigations of data-driven closure for subgrid-scale stress in large-eddy simulation. Phys. Fluids 30 (12), 125101.
Weatheritt, J. & Sandberg, R. 2016 A novel evolutionary algorithm applied to algebraic modifications of the RANS stress–strain relationship. J. Comput. Phys. 325, 2237.
Weatheritt, J. & Sandberg, R. D. 2017a The development of algebraic stress models using a novel evolutionary algorithm. Intl J. Heat Fluid Flow 68, 298318.
Weatheritt, J. & Sandberg, R. D. 2017b Hybrid Reynolds-averaged/large-eddy simulation methodology from symbolic regression: formulation and application. AIAA J. 55 (11), 37343746.
Wu, J.-L., Xiao, H. & Paterson, E.2018 Data-driven augmentation of turbulence models with physics-informed machine learning. Preprint, arXiv:1801.02762.
Xiao, H., Wu, J.-L., Wang, J.-X., Sun, R. & Roy, C. J. 2016 Quantifying and reducing model-form uncertainties in Reynolds-averaged Navier–Stokes simulations: a data-driven, physics-informed Bayesian approach. J. Comput. Phys. 324, 115136.
Yu, C., Xiao, Z. & Li, X. 2016 Dynamic optimization methodology based on subgrid-scale dissipation for large eddy simulation. Phys. Fluids 28 (1), 015113.
Zhang, Z., Song, X.-d., Ye, S.-r., Wang, Y.-w., Huang, C.-g., An, Y.-r. & Chen, Y.-s. 2019 Application of deep learning method to Reynolds stress models of channel flow based on reduced-order modeling of DNS data. J. Hydrodyn. 31, 18.
Zhu, L., Zhang, W., Kou, J. & Liu, Y. 2019 Machine learning methods for turbulence modeling in subsonic flows around airfoils. Phys. Fluids 31 (1), 015105.
MathJax is a JavaScript display engine for mathematics. For more information see

JFM classification

Related content

Powered by UNSILO

Sub-grid scale model classification and blending through deep learning

  • Romit Maulik (a1), Omer San (a1), Jamey D. Jacob (a1) and Christopher Crick (a2)


Full text views

Total number of HTML views: 0
Total number of PDF views: 0 *
Loading metrics...

Abstract views

Total abstract views: 0 *
Loading metrics...

* Views captured on Cambridge Core between <date>. This data will be updated every 24 hours.

Usage data cannot currently be displayed.