Ensemble Gradient for Learning Turbulence Models from Indirect Observations

Ensemble Gradient for Learning Turbulence Models from Indirect Observations

Year:    2021

Author:    Carlos A. Michelén Ströfer, Xin-Lei Zhang, Heng Xiao

Communications in Computational Physics, Vol. 30 (2021), Iss. 5 : pp. 1269–1289

Abstract

Training data-driven turbulence models with high fidelity Reynolds stress can be impractical and recently such models have been trained with velocity and pressure measurements. For gradient-based optimization, such as training deep learning models, this requires evaluating the sensitivities of the RANS equations. This paper explores the use of an ensemble approximation of the sensitivities of the RANS equations in training data-driven turbulence models with indirect observations. A deep neural network representing the turbulence model is trained using the network’s gradients obtained by backpropagation and the ensemble approximation of the RANS sensitivities. Different ensemble approximations are explored and a method based on explicit projection onto the sample space is presented. As validation, the gradient approximations from the different methods are compared to that from the continuous adjoint equations. The ensemble approximation is then used to learn different turbulence models from velocity observations. In all cases, the learned model predicts improved velocities. However, it was observed that once the sensitivity of the velocity to the underlying model becomes small, the approximate nature of the ensemble gradient hinders further optimization of the underlying model. The benefits and limitations of the ensemble gradient approximation are discussed, in particular as compared to the adjoint equations.

You do not have full access to this article.

Already a Subscriber? Sign in as an individual or via your institution

Journal Article Details

Publisher Name:    Global Science Press

Language:    English

DOI:    https://doi.org/10.4208/cicp.OA-2021-0082

Communications in Computational Physics, Vol. 30 (2021), Iss. 5 : pp. 1269–1289

Published online:    2021-01

AMS Subject Headings:    Global Science Press

Copyright:    COPYRIGHT: © Global Science Press

Pages:    21

Keywords:    Ensemble methods turbulence modeling deep learning.

Author Details

Carlos A. Michelén Ströfer

Xin-Lei Zhang

Heng Xiao

  1. Data-driven turbulence modeling in separated flows considering physical mechanism analysis

    Yan, Chongyang | Li, Haoran | Zhang, Yufei | Chen, Haixin

    International Journal of Heat and Fluid Flow, Vol. 96 (2022), Iss. P.109004

    https://doi.org/10.1016/j.ijheatfluidflow.2022.109004 [Citations: 17]
  2. Explainability analysis of neural network-based turbulence modeling for transonic axial compressor rotor flows

    Wu, Chutian | Wang, Shizhao | Zhang, Xin-Lei | He, Guowei

    Aerospace Science and Technology, Vol. 141 (2023), Iss. P.108542

    https://doi.org/10.1016/j.ast.2023.108542 [Citations: 8]
  3. Optimal sensor placement for ensemble-based data assimilation using gradient-weighted class activation mapping

    Xu, Zhaoyue | Wang, Shizhao | Zhang, Xin-Lei | He, Guowei

    Journal of Computational Physics, Vol. 514 (2024), Iss. P.113224

    https://doi.org/10.1016/j.jcp.2024.113224 [Citations: 1]
  4. Physical interpretation of neural network-based nonlinear eddy viscosity models

    Zhang, Xin-Lei | Xiao, Heng | Jee, Solkeun | He, Guowei

    Aerospace Science and Technology, Vol. 142 (2023), Iss. P.108632

    https://doi.org/10.1016/j.ast.2023.108632 [Citations: 6]
  5. A PDE-free, neural network-based eddy viscosity model coupled with RANS equations

    Xu, Ruiying | Zhou, Xu-Hui | Han, Jiequn | Dwight, Richard P. | Xiao, Heng

    International Journal of Heat and Fluid Flow, Vol. 98 (2022), Iss. P.109051

    https://doi.org/10.1016/j.ijheatfluidflow.2022.109051 [Citations: 4]
  6. Data augmented turbulence modeling for three-dimensional separation flows

    Yan, Chongyang | Zhang, Yufei | Chen, Haixin

    Physics of Fluids, Vol. 34 (2022), Iss. 7

    https://doi.org/10.1063/5.0097438 [Citations: 20]
  7. Combining direct and indirect sparse data for learning generalizable turbulence models

    Zhang, Xin-Lei | Xiao, Heng | Luo, Xiaodong | He, Guowei

    Journal of Computational Physics, Vol. 489 (2023), Iss. P.112272

    https://doi.org/10.1016/j.jcp.2023.112272 [Citations: 11]
  8. A unified method of data assimilation and turbulence modeling for separated flows at high Reynolds numbers

    Wang, Zhiyuan | Zhang, Weiwei

    Physics of Fluids, Vol. 35 (2023), Iss. 2

    https://doi.org/10.1063/5.0136420 [Citations: 18]
  9. Ensemble Kalman method for learning turbulence models from indirect observation data

    Zhang, Xin-Lei | Xiao, Heng | Luo, Xiaodong | He, Guowei

    Journal of Fluid Mechanics, Vol. 949 (2022), Iss.

    https://doi.org/10.1017/jfm.2022.744 [Citations: 39]
  10. Data augmented prediction of Reynolds stresses for flows around an axisymmetric body of revolution

    Liu, Yi | Wang, Shizhao | Zhang, Xin-Lei | He, Guowei

    Ocean Engineering, Vol. 296 (2024), Iss. P.116717

    https://doi.org/10.1016/j.oceaneng.2024.116717 [Citations: 0]