Bulletin of the American Physical Society
68th Annual Meeting of the APS Division of Fluid Dynamics
Volume 60, Number 21
Sunday–Tuesday, November 22–24, 2015; Boston, Massachusetts
Session H21: Turbulence: Modeling I |
Hide Abstracts |
Chair: Yi Li, University of Sheffield Room: 209 |
Monday, November 23, 2015 10:35AM - 10:48AM |
H21.00001: Synthesizing non-Gaussian inhomogeneous turbulence using optimization techniques Yi Li Synthetic turbulence is an important component of large eddy simulations, where it is used as the initial or inlet condition. Traditional synthetic models have not attempted to reproduce small scale dynamics even though it is important to sustain turbulence development. This problem is attacked recently by a Multi-scale Turnover Lagrangian Map (MTLM) model, which successfully reproduces a range of small-scale statistics in isotropic turbulence. In this talk, we introduce the constrained MTLM method (CMTLM), where optimization technique is used to generate inhomogeneous non-Gaussian MTLM synthetic fields. In CMTLM, the inhomogeneous statistics are set as the target, to be matched by the MTLM field. The MTLM field is found as the solution of an optimization problem with the random Gaussian input to MTLM as the control. We use several cases to show that the optimal MTLM field reproduces the inhomogeneous statistics while maintaining the realistic small scale statistics in many different flow conditions. The method thus proves to be a useful tool for large eddy simulations. [Preview Abstract] |
Monday, November 23, 2015 10:48AM - 11:01AM |
H21.00002: Wall-resolved adaptive simulation with spatially-anisotropic wavelet-based refinement Giuliano De Stefano, Eric Brown-Dymkoski, Oleg V. Vasilyev In the wavelet-based adaptive multi-resolution approach to turbulence simulation, the separation between resolved energetic structures and unresolved flow is achieved through wavelet threshold filtering. Depending on the thresholding level, the effect of residual motions can be either neglected or modeled, leading to wavelet-based adaptive DNS or LES. Due to the ability to identify and efficiently represent energetic dynamically important flow structures, these methods have been proven reliable and effective for the computational modeling of wall-bounded turbulence. The wall-resolved adaptive approach however necessitates the use of high spatial resolution in the wall region, which practically limits the application to moderate Reynolds numbers. In order to address this issue, a new method that makes use of a spatially-anisotropic adaptive wavelet transform on curvilinear grids is introduced. In contrast to all known adaptive wavelet-based approaches that suffer from the ``curse of anisotropy,'' i.e., isotropic wavelet refinement and inability to have spatially varying aspect ratio of the mesh elements, this approach utilizes spatially-anisotropic wavelet-based refinement. The method is tested for the turbulent flow past a rectangular cylinder at moderately high Reynolds number. [Preview Abstract] |
Monday, November 23, 2015 11:01AM - 11:14AM |
H21.00003: Machine Learning Models for Detection of Regions of High Model Form Uncertainty in RANS Julia Ling, Jeremy Templeton Reynolds Averaged Navier Stokes (RANS) models are widely used because of their computational efficiency and ease-of-implementation. However, because they rely on inexact turbulence closures, they suffer from significant model form uncertainty in many flows. Many RANS models make use of the Boussinesq hypothesis, which assumes a non-negative, scalar eddy viscosity that provides a linear relation between the Reynolds stresses and the mean strain rate. In many flows of engineering relevance, this eddy viscosity assumption is violated, leading to inaccuracies in the RANS predictions. For example, in near wall regions, the Boussinesq hypothesis fails to capture the correct Reynolds stress anisotropy. In regions of flow curvature, the linear relation between Reynolds stresses and mean strain rate may be inaccurate. This model form uncertainty cannot be quantified by simply varying the model parameters, as it is rooted in the model structure itself. Machine learning models were developed to detect regions of high model form uncertainty. These machine learning models consisted of binary classifiers that predicted, on a point-by-point basis, whether or not key RANS assumptions were violated. These classifiers were trained and evaluated for their sensitivity, specificity, and generalizability on a database of canonical flows. [Preview Abstract] |
Monday, November 23, 2015 11:14AM - 11:27AM |
H21.00004: Towards Sparse-Direct Interaction Perturbation (SDIP) for Variable-Density Flow David Petty, Carlos Pantano A numerical method has been developed to solve the set of integro-differential equations which result from applying the Sparse Direct-Interaction Perturbation (SDIP) technique to the low-speed, variable-density Navier-Stokes equations. This type of turbulence is at the heart of mixing and combustion applications. SDIP is a second-order moment closure theory that has particular relevance to the modeling of fluid turbulence. The strongly nonlinear numerical problem has been formulated as a system of equations using finite differences in time decorrelation, interpolation, variable-order quadratures, and mesh adaptation. The solution to this system has been made practicable by the construction of the full Jacobian of the numerical method using the Automatic Differentiation by OverLoading in C++ (ADOL-C) library. Special coordinate transformations were found to be essential for robust calculations of integrals that are not absolutely convergent; cancellations of singularities must be treated accurately. Progress towards the determination of the turbulence kinetic energy spectrum and velocity-scalar cospectra of the low-speed, variable-density Navier-Stokes equations derived from the SDIP solver will be discussed. [Preview Abstract] |
Monday, November 23, 2015 11:27AM - 11:40AM |
H21.00005: Spectral models of strongly inhomogeneous turbulence Andrew Bragg, Susan Kurien, Timothy Clark We compare results from a spectral model for inhomogeneous turbulence (Besnard et al., Theor. Comp. Fluid. Dyn., vol. 8, pp 1-35, 1996) with DNS data of a shear-free mixing layer (SFML) (Tordella et al., Phys. Rev. E, vol. 77, 016309, 2008). The SFML is used as a test case in which the efficacy of the model closure for the physical-space energy transport can be tested in a flow with strong inhomogeneity, without the additional complexity of mean-flow coupling. The model is able to capture certain features of the SFML quite well for intermediate to long-times, including the evolution of the mixing-layer width and turbulent kinetic energy. At short-times, and for more sensitive statistics such as the generation of the velocity field anisotropy, the model does not work so well. It may be argued that the discrepancy arises due to the local approximation to the intrinsically non-local pressure transport in physical-space, the effect of which would be particularly strong at short-times when the inhomogeneity of the SFML is strongest. Motivated by these results, we briefly discuss a new model that captures the non-local transport effects, for arbitrarily strong inhomogeneities of the flow. [Preview Abstract] |
Monday, November 23, 2015 11:40AM - 11:53AM |
H21.00006: Asymptotic stability of spectral-based PDF modeling for homogeneous turbulent flows Alejandro Campos, Karthik Duraisamy, Gianluca Iaccarino Engineering models of turbulence, based on one-point statistics, neglect spectral information inherent in a turbulence field. It is well known, however, that the evolution of turbulence is dictated by a complex interplay between the spectral modes of velocity. For example, for homogeneous turbulence, the pressure-rate-of-strain depends on the integrated energy spectrum weighted by components of the wave vectors. The Interacting Particle Representation Model (IPRM) (Kassinos \& Reynolds, 1996) and the Velocity/Wave-Vector PDF model (Van Slooten \& Pope, 1997) emulate spectral information in an attempt to improve the modeling of turbulence. We investigate the evolution and asymptotic stability of the IPRM using three different approaches. The first approach considers the Lagrangian evolution of individual realizations (idealized as particles) of the stochastic process defined by the IPRM. The second solves Lagrangian evolution equations for clusters of realizations conditional on a given wave vector. The third evolves the solution of the Eulerian conditional PDF corresponding to the aforementioned clusters. This last method avoids issues related to discrete particle noise and slow convergence associated with Lagrangian particle-based simulations. [Preview Abstract] |
Monday, November 23, 2015 11:53AM - 12:06PM |
H21.00007: Adaptive variable-fidelity wavelet-based eddy-capturing approaches for compressible turbulence Eric Brown-Dymkoski, Oleg V. Vasilyev Multiresolution wavelet methods have been developed for efficient simulation of compressible turbulence. They rely upon a filter to identify dynamically important coherent flow structures and adapt the mesh to resolve them. The filter threshold parameter, which can be specified globally or locally, allows for a continuous tradeoff between computational cost and fidelity, ranging seamlessly between DNS and adaptive LES. There are two main approaches to specifying the adaptive threshold parameter. It can be imposed as a numerical error bound, or alternatively, derived from real-time flow phenomena to ensure correct simulation of desired turbulent physics. As LES relies on often imprecise model formulations that require a high-quality mesh, this variable-fidelity approach offers a further tool for improving simulation by targeting deficiencies and locally increasing the resolution. Simultaneous physical and numerical criteria, derived from compressible flow physics and the governing equations, are used to identify turbulent regions and evaluate the fidelity. Several benchmark cases are considered to demonstrate the ability to capture variable density and thermodynamic effects in compressible turbulence. [Preview Abstract] |
Monday, November 23, 2015 12:06PM - 12:19PM |
H21.00008: Full field inversion: A tool to diagnose and improve closure models Anand Pratap Singh, Karthik Duraisamy Existing single-point closure models of turbulence are inaccurate in complex flows. The errors inherent in these models cannot be rectified by modifying parameters in the model -- it is rather the functional form of the model that is in question. In this work, full-field inversion is used to infer the functional form of modeling discrepancies. The inference process is driven by Bayesian inversion applied to data from Direct Numerical and Large Eddy simulations and experimental measurements. A physics-constrained approach is used to regularize the heavily ill-posed problem. It is to be noted that the full-field inversion involves extreme-scale optimization and Hessian computations. Efficient surrogate-enhanced adjoint techniques are employed to obtain the maximum aposteriori estimate and covariance of the inferred functions. The procedure is applied in a number of problems involving adverse and favorable pressure gradients and separation. The extracted information is used as part of a data-driven inversion/machine learning framework to improve closure models. [Preview Abstract] |
Monday, November 23, 2015 12:19PM - 12:32PM |
H21.00009: Closure modeling using field inversion and machine learning Karthik Duraisamy The recent acceleration in computational power and measurement resolution has made possible the availability of extreme scale simulations and data sets. In this work, a modeling paradigm that seeks to comprehensively harness large scale data is introduced, with the aim of improving closure models. Full-field inversion (in contrast to parameter estimation) is used to obtain corrective, spatially distributed functional terms, offering a route to directly address model-form errors. Once the inference has been performed over a number of problems that are representative of the deficient physics in the closure model, machine learning techniques are used to reconstruct the model corrections in terms of variables that appear in the closure model. These machine-learned functional forms are then used to augment the closure model in predictive computations. The approach is demonstrated to be able to successfully reconstruct functional corrections and yield predictions with quantified uncertainties in a range of turbulent flows. [Preview Abstract] |
Monday, November 23, 2015 12:32PM - 12:45PM |
H21.00010: Subgrid-scale modeling for flows with strong density variations Sidharth GS, Graham Candler High-speed reacting flows exhibit strong spatio-temporal density variations that arise from heat release, compressibility and differences in composition. Strong density gradients involve baroclinic and dilatational sources of vorticity, thereby influencing the flow dynamics. The present work develops subgrid-scale models for this class of flows. We employ the formulation based on filtered velocity ($\bar u_i$) as the resolved-scale velocity variable. This is because the conventional Favre-filtered velocity ($\tilde u_i$) is deficient in capturing the resolved-scale velocity dynamics and ignores the subgrid-scale interactions of pressure gradient with density. Furthermore, we investigate the contribution of subgrid-scale density fluctuations to the local subgrid-scale stress (and subgrid-scale scalar fluxes) via generation of small-scale velocity gradients. This effect is studied in the framework of the stretched-vortex subgrid-scale model. \emph{ A posteriori} performance of the proposed modifications is analysed on large eddy simulations of inert and reacting mixing layers. [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. |
© 2024 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
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700