Bulletin of the American Physical Society
72nd Annual Meeting of the APS Division of Fluid Dynamics
Volume 64, Number 13
Saturday–Tuesday, November 23–26, 2019; Seattle, Washington
Session A19: Advanced Turbulence Models I |
Hide Abstracts |
Chair: Todd Oliver, University of Texas, Austin Room: 401 |
Saturday, November 23, 2019 3:00PM - 3:13PM |
A19.00001: Physics-informed deep neural networks applied to scalar subgrid flux modeling in a mixed DNS/LES framework Gavin Portwood, Misha Chertkov, Balasubramanya Nadiga, Juan Saenz, Daniel Livescu The application of artificial neural networks (ANNs) to turbulence closure has been an emergent and active area of research in recent years due to the success of such data-driven methods in fields of computer vision, natural language processing, and other industrial and scientific disciplines. In this research, we apply ANNs to spacio-temporal dynamic modeling of the subgrid passive scalar flux as it relates to large-eddy simulations (LESs). By training on direct numerical simulations (DNSs) of homogeneous isotropic turbulence coupled to a passive scalar, we optimize ANNs to predict the subgrid scalar flux as a function of resolved-scale features. Trained models are then implemented in simulation and evaluated with \textit{a-posteriori} analysis. In these simulations, filtered scalar advection is coupled to explicitly filtered and statistically-stationary turbulence such that scalar dynamics have no dependence on potentially inaccurate subgrid stress models. By analysis with single- and multi-point statistics, we demonstrate that the data-driven models compete, and often out-perform, properly optimized canonical models. We suggest that this simulation framework may serve as a simplified closure testbed for the investigation and evaluation of data-driven turbulence closures. [Preview Abstract] |
Saturday, November 23, 2019 3:13PM - 3:26PM |
A19.00002: Fractional physical-inform neural networks (fPINNs) for turbulent flows Fangying Song, Guofei Pange, Charles Meneveau, George Karniadakis We employ fractional operators in conjunction with physics-informed neural networks (PINNs) to discover new governing equations for modeling and simulating the Reynolds stresses in the Reynolds Averaged Navier-Stokes equations (RANS) for wall-bounded turbulent flows at high Reynolds number. In particular, we develop a simple one-dimensional model for fully-developed wall-turbulence that involves a fractional operator with fractional order that varies with the distance from the wall. We use available DNS data bases to infer the function that describes the fractional order, which has an integer value at the wall and decays monotonically to an asymptotic value at the centerline. We show that this function is universal upon re-scaling and hence it can be used to predict the mean velocity profile at all Reynolds numbers. We also extend the fractional RANS for fully-developed turbulent channel flow to a turbulent boundary layer and infer the fractional order in the wake region. [Preview Abstract] |
Saturday, November 23, 2019 3:26PM - 3:39PM |
A19.00003: Wall-modeled large-eddy simulation of a turbulent channel flow based on artificial neural network Young Mo Lee, Jungil Lee, Jae Hwa Lee Because the computational cost of large-eddy simulation (LES) in the near-wall region of wall-bounded flows is proportional to approximately square of the friction Reynolds number (\textit{Re}$_{\tau })$, utilizing wall-modeled LES (WMLES) is promising to simulate a turbulent flow at sufficiently high Reynolds number with a reasonable cost. The most widely used wall model is an equilibrium stress model (i.e., wall-stress model) based on the momentum conservation. However, this method still needs to improve the accuracy and applicability for complex flows (e.g., swirled or separated flow) due to the limitations of the equilibrium assumption. In the present study, we employ an artificial neural network (ANN) to obtain information of the wall shear stress for WMLES. The proposed method shows good prediction on the mean velocity and Reynolds stress profiles compared to previous models in a turbulent channel flow in the range of the friction Reynolds numbers (395\textless \textit{Re}$_{\tau }$\textless 5200), even though the turbulent statistics at untrained Reynolds numbers are predicted. [Preview Abstract] |
Saturday, November 23, 2019 3:39PM - 3:52PM |
A19.00004: Zonal Turbulence Modeling via Decision Trees.. Racheet Matai, Paul Durbin The idea of a zonal model is a given model, but with its coefficients varying in different regions of a flow. That idea suggests using a form ofclassifier to identify zones. The bag-of-trees algorithm has been used to devise a zonal k-omega model. The training data are optimized, coefficient discrepancy fields, obtained by the method of Duraisamy, et al, 2015.The optimization is done with LES data as the target for flow over a circular arc bump. The discrepancy data are binned, with each bin assigned a particular range of values. The zones are parameterized by training the machine learning model with a local feature set.~The features are coordinate invariant flow parameters. The classification that is derived by ML is close toassociating zones with adverse and favorable pressure gradients. The correction produced by the machine learning algorithm is self-consistent; i.e. once the solution converges, the zones remain fixed. [Preview Abstract] |
Saturday, November 23, 2019 3:52PM - 4:05PM |
A19.00005: Fractional-Order LES Subgrid-Scale Modeling: Theory and Numerics Mehdi Samiee, Ali Akhavan-Safaei, Mohsen Zayernouri Filtering the Navier-Stokes equations in the large-eddy simulation of turbulent flows inherently introduces nonlocal features in the subgrid scale fluid motion. Such long-range effects become even more pronounced when the filter- width enlarges. That urges the development of new nonlocal closure models, which respect the corresponding non-Gaussian statistics of the subgrid stochastic motions. Starting from the filtered Boltzmann equation, we model the corresponding equilibrium distribution function with a \textit{L\'evy} stable distribution, which leads to the proposed fractional-order modeling of subgrid-scale stresses. We approximate the filtered equilibrium distribution function with a power-law term, and we derive the corresponding filtered Navier-Stokes equations. Subsequently in our functional modeling, the divergence of subgrid-scale stresses emerges as a single-parameter fractional Laplacian, $(-\Delta)^{\alpha}(\cdot)$, $\alpha \in (0,1]$, of the filtered velocity field. Finally, the introduced model undergoes \textit{a priori} evaluations based on the direct numerical simulation database of forced and decaying homogeneous isotropic turbulent flows at high and moderate Reynolds numbers, respectively. [Preview Abstract] |
Saturday, November 23, 2019 4:05PM - 4:18PM |
A19.00006: Motivation and development of the time-averaged active model-splitting hybrid RANS/LES sigfried haering, Todd Oliver, Ramesh Balakrishnan, Robert Moser We discuss motivation and development of the time-averaged active model-split (TAMS) hybrid RANS/LES approach. TAMS has been specifically constructed to overcome challenges associated with existing hybrid approaches related to LES/RANS blending techniques and inconsistencies between the resolved and modeled turbulence. Core to TAMS is a hybridization strategy in which the RANS and LES components act through separate models formulated using the mean and fluctuating velocity, respectively, as approximated by time averaging over the local turbulent timescales. Justification for this splitting strategy is discussed based on true subgrid terms from filtered DNS and simple LEVM arguments. Multiple validation cases are used to demonstrate the potential of the method. [Preview Abstract] |
Saturday, November 23, 2019 4:18PM - 4:31PM |
A19.00007: Numerically measured scale-dependent eddy viscosity in homogeneous isotropic turbulence yasaman shirian, Ali Mani In the present work, we report the directly measured eddy viscosity in incompressible homogeneous isotropic turbulence using the macroscopic forcing method. Our results provide a scale-dependent eddy viscosity; specifically, in the low-wavenumber-limit, eddy viscosity is a constant, while in the high-wavenumber limit, it becomes inversely proportional to the wavenumber. Our results present a contrast to the previously reported eddy diffusivity via renormalization group theories (Yakhot and Smith 1992), this difference is attributed to proper quantification of transport at small scales by large-eddies in the present study. We also report the scale-dependent turbulent Schmidt number as well as collapse of the measured eddy-viscosity operator with increase of flow Reynolds number. \newline Supported by DOE [Preview Abstract] |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2020 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
1 Research Road, Ridge, NY 11961-2701
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700