Bulletin of the American Physical Society
71st Annual Meeting of the APS Division of Fluid Dynamics
Volume 63, Number 13
Sunday–Tuesday, November 18–20, 2018; Atlanta, Georgia
Session D31: Large Eddy Simulations: Modeling 
Hide Abstracts 
Chair: Anthony Leonard, California Institute of Technology Room: Georgia World Congress Center B403 
Sunday, November 18, 2018 2:30PM  2:43PM 
D31.00001: A novel hybrid twolevel and kineticeddy simulation model for high Reynolds number wallbounded turbulent flows Reetesh Ranjan, Achyut Panchal, Suresh Menon A novel hybrid model for simulation of wallbounded turbulent flows is developed by blending the twolevel simulation (TLS) method [1] in the nearwall inner region with the kineticeddy simulation (KES) method [2] in the outer region. TLS is a multiscale approach, where a field variable is decomposed into its largescale (LS) and smallscale (SS) components, and both LS and SS fields are obtained explicitly by employing modeling assumptions for the smallscales. It does not employ the notion of spatial filtering and eddy viscosity. KES is a twoequation based model, where transport equations for the local subgrid kinetic energy and length scale are solved, thus allowing the method to approach to very largeeddy simulation (VLES), LES, and direct numerical simulation (DNS) in the limit of resolution of turbulence length scales corresponding to only large scales, grid size, and fully resolved, respectively. The generalized hybrid formulation is evaluated for its capabilities by simulating fully developed turbulent channel flow at different frictional Reynolds number, and comparing with the reference DNS results. [1] K. A. Kemenov, S. Menon, J. Comput. Phys., 220 (2006), 290311. 
Sunday, November 18, 2018 2:43PM  2:56PM 
D31.00002: Local Variational Germano Identity for Dynamic Large Eddy Simulations based on Finite Elements Onkar Sahni In this talk we will present a localized formulation of the variational Germano identity (VGI) that is applicable to inhomogeneous turbulent flows. This is done in the context of stabilized finite element methods where a combined subgridscale (SGS) model is employed. In particular, the combined SGS model uses the residualbased variational multiscale (RBVMS) approach along with the Smagorinsky eddyviscosity model. The RBVMS model is used to represent the crossstress terms while the eddyviscosity model is used for the Reynolds stresses. The unknown parameter related to the eddyviscosity model is computed using a dynamic procedure based on the local VGI. The overall idea of the local VGI is to compare the numerical solution at two different levels of discretization/grid in a local fashion. The local VGI based dynamic procedure is made practical by employing suitable assumptions. Further, the procedure is made robust by employing a Lagrangian averaging scheme along with a local spatial averaging. This is equivalent to averaging along local pathtubes and maintains the applicability of the current methodology to inhomogeneous turbulent flows. We demonstrate the current LES methodology on a variety of problems including flow over cylinders and surging airfoils. 
Sunday, November 18, 2018 2:56PM  3:09PM 
D31.00003: A simple extension to eddyviscosity models for Large Eddy Simulations based on tensor decompositions. Felipe A. V. de Bragança Alves, Stephen de Bruyn Kops Using tensor decompositions theorems and a nonlinear tensor formed by the Lie product of the strain rate and the rate of rotation tensor we form a simple 2element tensorial basis set to model the residual stress tensor attributing specific roles to each tensor. The strain rate tensor is responsible for reproducing the correct energy transfer from the resolved to the unresolved scales, while the nonlinear term is responsible for the correct energy redistribution among the resolved scales. The coefficients multiplying each tensor are uncoupled, so the nonlinear term can be added to any commonly used version of eddy viscosity model without the need for rethinking the eddy viscosity modeling. 
Sunday, November 18, 2018 3:09PM  3:22PM 
D31.00004: Datadriven deconvolution for the large eddy simulation of Kraichnan turbulence Romit Maulik, Omer San, Adil Rasheed, Prakash Vedula In this study, we demonstrate the use of artificial neural networks as optimal maps which are utilized for the convolution and deconvolution of coarsegrained fields to account for subgrid scale turbulence effects. We demonstrate that an effective eddyviscosity is characterized by our purely datadriven large eddy simulation framework without the explicit utilization of phenomenological arguments. In addition, our datadriven framework does not require the knowledge of true subgrid stress information during the training phase due to its focus on estimating an effective filter and its inverse so that gridresolved variables may be related to direct numerical simulation data statistically. Through this we seek to unite the structural and functional modeling strategies for modeling nonlinear partial differential equations using reduced degrees of freedom. Both apriori and aposteriori results are shown for the Kraichnan turbulence case in addition to a detailed description of validation and testing. Our findings indicate that the proposed framework approximates a robust and stable subgrid closure which compares favorably to the Smagorinsky and Leith hypotheses for capturing theoretical kineticenergy scaling trends in the wavenumber domain. 
Sunday, November 18, 2018 3:22PM  3:35PM 
D31.00005: Consistency and accuracy of LES by explicit filtering Joseph Mathew, Sumit Kumar Patel An explicit filtering method was proposed as an equivalent implementation of the approximate deconvolution method for large eddy simulation (LES). The essential requirement of this method was that all spatial numerical operations (finding derivatives, interpolations) be of high resolutionbe spectrally accurate over a large part of the computed, lowwavenumber range. A natural expectation is that the range of scales over which the dynamics are more accurately represented increases with gridrefinement. Second, beyond some level of refinement, large scale dynamics do not change significantly because the energy in small scale content is orders of magnitude smaller. Third, as Reynolds number is increased, in many flows where the spectral range of the active scales increases at the small scale end only, an LES should return the same large scale dynamics. Systematic studies at different resolutions and Reynolds numbers in forced, homogeneous isotropic turbulence and round jets (lowspeed and highspeed ones with multiple shock cells, at Reynolds numbers of the order of a million), will be presented that show these expectations to be met in LES by explicit filtering. Such a posteriori consistency and accuracy make this a reliable method for LES. 
Sunday, November 18, 2018 3:35PM  3:48PM 
D31.00006: OpenFOAM based Evaluation of PANS Method with NonLinear Eddy Viscosity Closure for Separated Turbulent Flows Sagar Saroha, Sawan S. Sinha, Sunil Lakshmipathy 
Sunday, November 18, 2018 3:48PM  4:01PM 
D31.00007: Abstract Withdrawn Prediction of nearwall turbulence is critical to the success of practical simulationbased investigation of complex flows as resolving all the relevant scales of motion is computationally intractable. Underresolved LES methods that combine phenomenology with the predicted nearwall field show limited success in approximating the nearwall variant of the wellknown turbulence closure problem. Alternative paradigms based on datadriven modeling are increasingly being explored for building turbulence closures. We hypothesize that existing models can be improved by leveraging more information available in the coarse predictions. In this study, we explore the role of data quantity and choice of regression framework for machine learningbased near wall modeling. To this end, we systematically explore the usefulness of strain rate information in addition to the velocity field in modeling nearwall turbulence structure. Additionally, we also assess the suitability of nonlinear regression methods such as artificial neural networks for improved representation of the dynamics as against more interpretable sparse functional regression approaches. This a priori analysis leverages DNS datasets of canonical flat channel turbulent flows. 

D31.00008: ABSTRACT WITHDRAWN

Sunday, November 18, 2018 4:14PM  4:27PM 
D31.00009: Assessment of different cutoff (filterwidth) prescription approaches for the scaleresolving PANS method Branislav Basara, Sharath Girimaji, Zoran Pavlovic The PartiallyAveraged NavierStokes (PANS) is a scaleresolving turbulence computational approach designed to resolve large scale fluctuations and model the remainder with appropriate closures. Depending upon the prescribed cutoff length (filter width) the method adjusts seamlessly from the ReynoldsAveraged NavierStokes (RANS) to the Direct Numerical Solution (DNS) of the NavierStokes equations. The unresolved to total kinetic energy ratio f_{k} is the cutoff control parameter and its specification requires the knowledge of the local turbulent (resolved + unresolved) kinetic energy. While the unresolved kinetic energy is computed directly from model equations, in most current formulations, the resolved kinetic energy is obtained by suitably averaging the resolved field – as in dynamic Smagorinsky LES computations. As the averaging process is expensive, recently alternate specification approaches have been developed. One such approach is to solve an additional equation for resolved turbulent kinetic energy as first proposed by Basara and Girimaji (2013) and further developed by Basara, Pavlovic and Girimaji (2018). In this presentation, we analyse the various f_{k}specification approaches. Important conclusions regarding the merits of each method are drawn. 
Sunday, November 18, 2018 4:27PM  4:40PM 
D31.00010: The application of data assimilation to combine experimental data and LES for improved stateestimation. Jeffrey Labahn, Hao Wu, Shaun Harris, Bruno Coriton, Werner M. Ihme, Jonathan H Frank In the current study, data assimilation techniques are investigated to integrate highspeed highresolution experimental data into a Large Eddy Simulation (LES). LES of an inert jet is performed without data assimilation and shown to accurately reproduce statistical flowfield quantities. To capture the transient dynamics, assimilation of experimental data is performed using an Ensemble Kalman Filter (EnKF) algorithm and the performance of the method is investigated to understand its impact on the state estimation. Our first objective is to investigate the impact that data assimilation has on the resulting flow field for this inert jet. This is accomplished by comparing transient predictions and instantaneous flow structures obtained from a baseline LES without data assimilation to those obtained via EnKF. The second objective is to identify the impact that data localization has on the resulting predictions. Following this, we investigate how the stateestimation is affected by changes in experimental uncertainty, assimilation frequency and sparsity of experimental data. 
Follow Us 
Engage
Become an APS Member 
My APS
Renew Membership 
Information for 
About APSThe American Physical Society (APS) is a nonprofit membership organization working to advance the knowledge of physics. 
© 2019 American Physical Society
 All rights reserved  Terms of Use
 Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 207403844
(301) 2093200
Editorial Office
1 Research Road, Ridge, NY 119612701
(631) 5914000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 200452001
(202) 6628700