Deep reinforcement learning for turbulence modeling in large eddy simulations

Kurz, Marius; Offenhäuser, Philipp; Beck, Andrea

doi:10.1016/j.ijheatfluidflow.2022.109094

Cited by 42 publications

(26 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By integrating the RANS simulations with RL-based training process, the reward can be defined by other quantities of interest, so that the discovered model would be more consistent with physical observations 37 . In addition, by extending RL with special types of neural networks, a nonlocal constitutive models could be developed for improving the performance of data-driven turbulence models 40,56 . x/h; Ū + x/h…”

Section: Discussionmentioning

confidence: 99%

“…Moreover, supervised learning could be ill-posed for turbulence modeling in implicit filtered LES due to labeled filter form is not available. This challenge can be alleviated by RL, as is done by Kurz, Offenhäuser, and Beck 40 , who applied RL with convolutional neural networks to find an optimal eddy-viscosity for implicitly filtered LES of HIT.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Discovering explicit Reynolds-averaged turbulence closures for turbulent separated flows through deep learning-based symbolic regression with non-linear corrections

Tang,

Wang,

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Discovering explicit Reynolds-averaged turbulence closures for turbulent separated flows through deep learning-based symbolic regression with non-linear corrections

Tang,

Wang,

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

“…As long as this term does not depend on the solution u the forcing commutes with W. This means we can simply add F = WF to the RHS of (23) without any contribution to the closure term. In addition, we can account for its contribution to the evolution of s by first computing its contribution F to the evolution of the SGS content (see (24)) as…”

Section: Forcingmentioning

confidence: 99%

Energy-Conserving Neural Network for Turbulence Closure Modeling

Gastelen¹,

Edeling²,

Sanderse³

2023

Preprint

View full text Add to dashboard Cite

In turbulence modeling, and more particularly in the Large-Eddy Simulation (LES) framework, we are concerned with finding closure models that represent the effect of the unresolved subgrid-scales on the resolved scales. Recent approaches gravitate towards machine learning techniques to construct such models. However, the stability of machine-learned closure models and their abidance by physical structure (e.g. symmetries, conservation laws) are still open problems. To tackle both issues, we take the 'discretize first, filter next' approach, in which we apply a spatial averaging filter to existing energy-conserving (fine-grid) discretizations. The main novelty is that we extend the system of equations describing the filtered solution with a set of equations that describe the evolution of (a compressed version of) the energy of the subgridscales. Having an estimate of the energy of the subgrid-scales, we can use the concept of energy conservation and derive stability of the discrete representation. The compressed variables are determined via a data-driven technique in such a way that the energy of the subgrid-scales is matched. For the extended system, the closure model should be energy-conserving, and a new skew-symmetric convolutional neural network architecture is proposed that has this property. Stability is thus guaranteed, independent of the actual weights and biases of the network. Importantly, our framework allows energy exchange between resolved scales and compressed subgrid scales and thus enables backscatter. To model dissipative systems (e.g. viscous flows), the framework is extended with a diffusive component. The introduced neural network architecture is constructed such that it also satisfies momentum conservation. We apply the new methodology to both the viscous Burgers' equation and the Korteweg-De Vries equation in 1D and show superior stability properties when compared to a vanilla convolutional neural network.

show abstract

“…This does not mean that ML cannot be successful in turbulence modeling, as it is still in its early stages and many possibilities have not been explored yet. Despite its limitations, such as its black-box nature and limited extrapolation capabilities, ML might have the a) Electronic mail: abkar@mpe.au.dk potential to overcome these challenges through the use of various types of neural networks [23][24][25][26] . It is important to note that the above-mentioned observations and criticisms about ML are not applicable to all ML techniques and are more commonly made about ML in general rather than specific ML methods.…”

Section: Introductionmentioning

confidence: 99%

“…It is important to note that the above-mentioned observations and criticisms about ML are not applicable to all ML techniques and are more commonly made about ML in general rather than specific ML methods. This paper aims to use a specific ML tool, reinforcement learning (RL), which has limited usage in turbulence modeling 20,23,27,28 but has various applications in CFD for control and optimization 9,29 . Novati et al 27 were the first to use RL for predicting subgrid scale (SGS) turbulence model in the context of large-eddy simulations (LES) followed by Refs.…”

Section: Introductionmentioning

confidence: 99%

Log-law recovery through reinforcement-learning wall model for large-eddy simulation

Vadrot¹,

Yang²,

Bae³

et al. 2023

Preprint

View full text Add to dashboard Cite

This paper focuses on the use of reinforcement learning (RL) as a machine-learning (ML) modeling tool for near-wall turbulence. RL has demonstrated its effectiveness in solving high-dimensional problems, especially in domains such as games. Despite its potential, RL is still not widely used for turbulence modeling and is primarily used for flow control and optimization purposes. A new RL wall model (WM) called VYBA23 is developed in this work, which uses agents dispersed in the flow near the wall. The model is trained on a single Reynolds number (Re τ = 10 4 ) and does not rely on high-fidelity data, as the back-propagation process is based on a reward rather than output error. The states of the RLWM, which are the representation of the environment by the agents, are normalized to remove dependence on the Reynolds number. The model is tested and compared to another RLWM (BK22) and to an equilibrium wall model, in a half-channel flow at eleven different Reynolds numbers (Re τ ∈ [180; 10 10 ]). The effects of varying agents' parameters such as actions range, time-step, and spacing are also studied. The results are promising, showing little effect on the average flow field but some effect on wall-shear stress fluctuations and velocity fluctuations. This work offers positive prospects for developing RLWMs that can recover physical laws, and for extending this type of ML models to more complex flows in the future.

show abstract

Deep reinforcement learning for turbulence modeling in large eddy simulations

Cited by 42 publications

References 35 publications

Discovering explicit Reynolds-averaged turbulence closures for turbulent separated flows through deep learning-based symbolic regression with non-linear corrections

Discovering explicit Reynolds-averaged turbulence closures for turbulent separated flows through deep learning-based symbolic regression with non-linear corrections

Energy-Conserving Neural Network for Turbulence Closure Modeling

Log-law recovery through reinforcement-learning wall model for large-eddy simulation

Contact Info

Product

Resources

About