Towards Prediction of Turbulent Flows at High Reynolds Numbers Using High Performance Computing Data and Deep Learning

Bode, Mathis; Gauding, Michael; Göbbert, Jens Henrik; Liao, Baohao; Jitsev, Jenia; Pitsch, Heinz

doi:10.1007/978-3-030-02465-9_44

Cited by 8 publications

(4 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Various efforts were done by researchers to simulate three-dimensional isotropic turbulence based on the three basic approaches: Direct Numerical Simulation (DNS), Large Eddy Simulation (LES), and Reynolds-Averaged Navier-Stokes (RANS). Furthermore, high-performance computing (HPC) has been used recently to satisfy the continuous demand of reaching more precise predictions of turbulence behavior and reduce the time consumption of simulation processes [1,2,3]. A major achievement for applying high-performance technologies in order to understand complex turbulent flow systems was done by Sergei et al.…”

Section: Introductionmentioning

confidence: 99%

Lattice Boltzmann high-performance simulations of decaying isotropic turbulence

Kareem

Mohammed

Sherbiny

et al. 2023

Preprint

View full text Add to dashboard Cite

This study investigates high-performance computing (HPC) strategies for improving and reducing the time consumption of the simulations of turbulent flows in periodic boxes at different resolutions using the lattice Boltzmann method (LBM). Isotropic turbulence is a fundamental problem in fluid dynamics that can be considered a good candidate for examining HPC applications. The LBM is a computational method that can be computationally parallelized, hence proper parallelization and vectorization are provided for reducing data volume and data transfer. Simulations of decaying isotropic turbulence at resolutions starting from 323 to 5123 using the LBM are carried out for these purposes. Clear progress for time reduction is achieved in all cases, while different features of the flow fields are depicted, and their characteristics are discussed. Very thin tubes are visualized, and the energy spectra are studied. More than 88%-time reduction is obtained in favor of the HPC using the same hardware resources with the lattice Boltzmann relaxation time 𝝉 = 0.503. All fields are initialized by a forced turbulent field that was simulated in a previous study using the LBM.

show abstract

Section: Introductionmentioning

confidence: 99%

Lattice Boltzmann high-performance simulations of decaying isotropic turbulence

Kareem

Mohammed

Sherbiny

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…[48][49][50][51][52] It also provides an alternative which bypasses the computational limitations of other data-driven models for turbulence generation that have been applied in lower Re environments and rely on large datasets. [53][54][55][56][57] In order to distinguish our work from many others in the rapidly developing subdiscipline of ML for fluid mechanics, 58 we itemize the main contributions of this work. Our main contributions include:…”

Section: Introductionmentioning

confidence: 99%

Learning the structure of wind: A data-driven nonlocal turbulence model for the atmospheric boundary layer

Keith,

Khristenko,

Wohlmuth

2021

Preprint

View full text Add to dashboard Cite

We develop a novel data-driven approach to modeling the atmospheric boundary layer. This approach leads to a nonlocal, anisotropic synthetic turbulence model which we refer to as the deep rapid distortion (DRD) model. Our approach relies on an operator regression problem which characterizes the best fitting candidate in a general family of nonlocal covariance kernels parameterized in part by a neural network. This family of covariance kernels is expressed in Fourier space and is obtained from approximate solutions to the Navier-Stokes equations at very high Reynolds numbers. Each member of the family incorporates important physical properties such as mass conservation and a realistic energy cascade. The DRD model can be calibrated with noisy data from field experiments. After calibration, the model can be used to generate synthetic turbulent velocity fields. To this end, we provide a new numerical method based on domain decomposition which delivers scalable, memory-efficient turbulence generation with the DRD model as well as others. We demonstrate the robustness of our approach with both filtered and noisy data coming from the 1968 Air Force Cambridge Research Laboratory Kansas experiments. Using this data, we witness exceptional accuracy with the DRD model, especially when compared to the International Electrotechnical Commission standard.

show abstract

“…The subfilter-scale content recovery was benchmarked against several popular struc-tural closure modeling strategies. Bode et al [5] studied the accuracy of various network architectures for predicting statistics of turbulent flows. Machine learning (ML) and DL have also been applied to flow control [14,27], development of low-dimensional models [38], generation of inflow conditions [11], or structure identification in two-dimensional (2-D) decaying turbulence [19].…”

Section: Introductionmentioning

confidence: 99%

Deep Learning at Scale for Subgrid Modeling in Turbulent Flows: Regression and Reconstruction

Bode

Gauding

Kleinheinz

et al. 2019

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

0000−0001−9922−9742] , Michael Gauding 2[0000−0003−0038−5249] , Konstantin Kleinheinz 1[0000−0003−0569−2221] , and Heinz Pitsch 1[0000−0001−5656−0961]Abstract. Modeling of turbulent flows is still challenging. One way to deal with the large scale separation due to turbulence is to simulate only the large scales and model the unresolved contributions as done in large-eddy simulation (LES). This paper focuses on two deep learning (DL) strategies, regression and reconstruction, which are data-driven and promising alternatives to classical modeling concepts. Using threedimensional (3-D) forced turbulence direct numerical simulation (DNS) data, subgrid models are evaluated, which predict the unresolved part of quantities based on the resolved solution. For regression, it is shown that feedforward artificial neural networks (ANNs) are able to predict the fully-resolved scalar dissipation rate using filtered input data. It was found that a combination of a large-scale quantity, such as the filtered passive scalar itself, and a small-scale quantity, such as the filtered energy dissipation rate, gives the best agreement with the actual DNS data. Furthermore, a DL network motivated by enhanced super-resolution generative adversarial networks (ESRGANs) was used to reconstruct fullyresolved 3-D velocity fields from filtered velocity fields. The energy spectrum shows very good agreement. As size of scientific data is often in the order of terabytes or more, DL needs to be combined with high performance computing (HPC). Necessary code improvements for HPC-DL are discussed with respect to the supercomputer JURECA. After optimizing the training code, 396.2 TFLOPS were achieved.

show abstract

Towards Prediction of Turbulent Flows at High Reynolds Numbers Using High Performance Computing Data and Deep Learning

Cited by 8 publications

References 13 publications

Lattice Boltzmann high-performance simulations of decaying isotropic turbulence

Lattice Boltzmann high-performance simulations of decaying isotropic turbulence

Learning the structure of wind: A data-driven nonlocal turbulence model for the atmospheric boundary layer

Deep Learning at Scale for Subgrid Modeling in Turbulent Flows: Regression and Reconstruction

Contact Info

Product

Resources

About