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 M01: Reacting Flows: Computational & Analytical Methods |
Hide Abstracts |
Chair: Shashank Yellapantula, National Renewable Energy Lab Room: Georgia World Congress Center B201 |
Tuesday, November 20, 2018 8:00AM - 8:13AM |
M01.00001: Machine Learning Methods for Early Prediction of Sustained Ignition Pavel Petkov Popov, Houyi Du, Jonathan Ben Freund We use machine learning methods to seek means of early prediction for sustained ignition in reactive flow simulations. Three configurations are considered: a geometrically complex simulation of a turbulent H2-air jet in crossflow in addition to two simpler ersatz models for this implemented on computationally cheap 2D and 3D cubic meshes. The objective is to correctly classify the ignition attempts as either igniting or non-igniting, based on the an early stages of the simulation. Several machine learning implementations are evaluated, ranging from a simple two-layer perceptron to two- and three-dimensional convolutional neural networks (CNN) with multiple data channels. Logarithmic pre-conditioning of radical concentrations is found to significantly improve the machine learning performance. The performance of CNN's is evaluated both when trained and tested on the same type of data (full jet in crossflow simulation vs. ersatz case) and when a network trained on computationally cheap ersatz data is applied to the jet in crossflow case. In the latter case, we explore the fidelity required of the ersatz model in order to achieve good predictive accuracy when applied to the jet in crossflow. |
Tuesday, November 20, 2018 8:13AM - 8:26AM |
M01.00002: A-priori analysis of joint PDF of mixture fraction and progress variable trained using machine learning techniques Marc Henry de Frahan, Shashank Yellapantula, Ryan King, Ray Grout, Marc Day In this study, we use supervised machine learning (ML) techniques to investigate models in the context of presumed PDF based LES techniques. A key modeling uncertainty associated with the presumed PDF methods is the shape of the joint PDF. In the current study, data from reacting DNS of the Low-Swirl burner (LSB) with methane as fuel (Day et al. Combust Flame. 2012) are used to develop models using ML techniques for joint PDF shape prediction. DNS data in the form of the joint PDF of moments of passive and reacting scalars, namely mixture fraction and progress variable, are used as the training set. Using a-priori analysis, we demonstrate the performance of traditional ML models and recent deep learning (DL) methods. Joint PDF predictions are convoluted with conditional means of the progress variable source term to help comparing various PDF shapes. Comparisons of random forest regression, a traditional ML technique, with two different types of deep neural networks, a fully-connected feed forward neural network and a generative adversarial network, indicate that the DL techniques produce more accurate joint PDFs. |
Tuesday, November 20, 2018 8:26AM - 8:39AM |
M01.00003: A-priori analysis of a data driven closure model trained from a reacting DNS of a Low-Swirl Burner Shashank Yellapantula, Marc Henry de Frahan, Ryan King, Ray Grout, Marc Day Machine learning (ML) advances coupled with the exponential increase in supercomputing capabilities presents tremendous promise in developing complex models for reacting flows from direct numerical simulations (DNS). In this study, we describe a model development effort via application of machine learning techniques to a reacting flow DNS. Specifically, we describe our work in applying various ML models to data from reacting DNS of a Low-Swirl Burner (LSB) [1]. In the current study, supervised learning techniques are utilized within the class of deep learning algorithms to investigate reacting flow sub-grid models. DNS data, in the form of moments and dissipation rates of mixture fraction and progress variable, are used as the input parameters. A-priori analysis is used to demonstrate the efficacy of the ML techniques to generate a sub-grid representation of the source term for progress variable. Models generated from both Random forest and a Deep Neural Network with 10 hidden layers and 20 nodes are compared. Both methods show tremendous promise and are found to produce peak conditional means within 15% of the DNS data across multiple filter widths. References [1] Day M, Tachibana S, Bell J, Lijewski M, Beckner V, Cheng RK. Combust Flame. 2012;159(1):275-290 |
Tuesday, November 20, 2018 8:39AM - 8:52AM |
M01.00004: A Comparison of Physics-Based and Data-Based Methods of Dimension Reduction in Turbulent Combustion A. Cody Nunno, Bruce A. Perry, Jonathan F. MacArt, Michael E. Mueller Turbulent combustion simulation requires solving for many interdependent variables over disparate scales. To alleviate the computationally expensive nature of these simulations, reduced-order models for the thermochemical state are often necessary. Dimension reduction can be achieved through physics-based approaches, such as in Flamelet-Generated Manifolds or the Flamelet-Progress Variable model, or through data-based approaches, such as Principal Component Analysis. While both classes of approaches have been assessed separately, a combined analysis has not been reported. In this work, the efficacy of reduced-order models created via data-based and physics-based methods is assessed using 3D DNS databases of nonpremixed and premixed hydrogen-air jet flames and a more chemically-complex sooting flame, as well as 1D laminar flame calculations. The data-based techniques produce reasonably accurate representations of the thermochemical state using almost as few parameters as the physics-based models. The leading parameters of the data-based models have similar physical interpretation to the parameters in the physics-based models, indicating the potential to discover additional model parameters in more complex configurations. |
Tuesday, November 20, 2018 8:52AM - 9:05AM |
M01.00005: A comprehensive multi-scale method of model reduction and analysis for reactive flow Tianhan Zhang, Temistocle Grenga, Yiguang Ju The numerical simulation for reactive flow with detailed chemistry is challenging, one of the main reasons is that the highly nonlinear chemical source term together with diffusion and convection terms in Navier-Stokes equation contributes to a wide range of time scales, which impose a severe stiffness in the simulations. The traditional approaches to resolve the stiffness problem focus on chemistry aspect. However, those methods assume combustion under homogeneous ignition condition which lose the generality for various combustion scenario and conditions. The current work develops a comprehensive method to analyze characteristic time scales for reaction, convection and diffusion. Firstly, using the G-Scheme Participation Index (GPI) the chemical mechanism is analyzed and simplified. Then, the capability of GPI in analyzing chemical mechanisms is enriched with the introduction of a new Participation Index for convection and diffusion processes and the definition of their characteristic time scales, which provides more physics insights into the flow-reaction coupling system. This leads to a criterion for the automatic discrimination of combustion modes. |
Tuesday, November 20, 2018 9:05AM - 9:18AM |
M01.00006: Accelerating laminar flame speed solvers for large kinetic mechanisms Simon Lapointe, Russell Whitesides, Matthew McNenly A fast solver is proposed for the simulation of one-dimensional laminar flames with large kinetic mechanisms. An approximately factorized Jacobian is used as preconditioner to greatly reduce the computational cost associated with matrix operations. The constant, non-unity Lewis number assumption is introduced to reduce the computational cost and improve the convergence of the solver. The solver is applied to laminar flame speed calculations with kinetic mechanisms of varying sizes (up to thousands of species). Laminar flame speeds and species profiles are found to be in very good agreement with other codes. The computation times increase only linearly with the number of species. This is a significant improvement over traditional steady-state solvers for which the computational cost scales quadratically with the number of species. For the largest mechanism tested, the present solver is two orders of magnitude faster than commonly-used codes. The use of an approximate Jacobian does not significantly affect the domain of convergence, making the solver well suited for laminar flame speed sweeps with large kinetic mechanisms. |
Tuesday, November 20, 2018 9:18AM - 9:31AM |
M01.00007: Novel method for Lagrangian-particle analysis of highly compressible reacting turbulence Yoram Kozak, Sai Sandeep Dammati, Laura O'Neill, Luis G Bravo, Peter E Hamlington, Alexei Poludnenko Highly compressible turbulent reacting flows involve complex phenomena, such as non-linear turbulence-chemistry interactions, and deflagration-to-detonation transition. Lagrangian-particle tracking can provide a powerful analysis framework of the flow complementary to Eulerian based formulation. Accurate advection of massless Lagrangian fluid parcels in highly turbulent flows places stringent requirements on the integration algorithms in terms of numerical errors. We discuss the development of accurate integration and interpolation methods with particular focus on the flows with discontinuities or high gradients induced by shock waves or flame fronts. Under these conditions, traditional smooth interpolation schemes are inaccurate. This problem is solved by Weighted-Essentially-Non-Oscillatory-type (WENO) interpolation schemes. Also, we show that time integrators with symplectic properties ensure that trajectory errors remain bounded in time. The accuracy of these methods is verified and is then applied in direct numerical simulations (DNS) of fast, highly compressible, premixed turbulent reacting flows. |
Tuesday, November 20, 2018 9:31AM - 9:44AM |
M01.00008: A backward photon Monte Carlo solver to understand radiative heat transfer under optically-thick conditions Xinyu Zhao, Bifen Wu A backward photon Monte Carlo radiation solver that can account for the spectral effects of gases, soot and droplets is developed in this study, based on the principle of reciprocity. Particularly suitable for the optically thick environments that are commonly encountered in high-pressure diesel engines and gas turbine combustors, the solver can provide the radiative heat source terms within a localized area in a computationally efficient manner. A hierarchy of test cases are constructed, with increasing complexity in the participative media, to systematically verify and validate the solver. A forward Monte Carlo solver is employed to provide the benchmark conditions for comparison. The solver is then applied to a model gas turbine combustor, to investigate the heat transfer pattern on the wall, through frozen field analysis. Parametric studies on the directional and spectral dependency of the radiative heat transfer are performed to provide further details to facilitate the comparison with experiments. |
Tuesday, November 20, 2018 9:44AM - 9:57AM |
M01.00009: Progress Towards Efficient Simulation of Large-Scale Fires Caelan B Lapointe, Nicholas T Wimer, Peter E Hamlington Wildland fires and other fire phenomena (e.g., pool fires, fire whirls) are an increasing concern as extreme weather events become more frequent, and understanding how large-scale fires behave will be key to mitigating human and structural losses, as well as supporting fire suppression efforts. Simulations provide an appealing framework to study fire phenomena, but can quickly become intractable; the spatial and temporal scales of large fires span many orders of magnitude, motivating efforts to reduce computational expense. In this talk, we present fireDyMFoam, a new version of the native OpenFOAM solver fireFoam, that has been extended to incorporate Adaptive Mesh Refinement (AMR). The use of AMR in the solver allows run-time mesh refinement in areas of interest, while large regions of the domain remain coarse, greatly decreasing computational cost. New functionality is demonstrated through a series of test cases with significantly reduced computational costs when compared to uniform mesh simulations with identical finest-scale resolution. Future work will focus on incorporating adjoints for optimization (e.g., where the objective is to minimize fire spread) and coupling extended dynamic meshing functionality with native OpenFOAM pyrolysis modeling capability. |
Tuesday, November 20, 2018 9:57AM - 10:10AM |
M01.00010: Progress Towards Direct Numerical Simulations of Plumes and Pool Fires Nicholas Terry Wimer, Marc Day, Amanda Makowiecki, Jeffrey Glusman, John Daily, Gregory Rieker, Peter E Hamlington Computational simulations of fire have the potential to provide new physical understanding of fire dynamics and evolution in both natural and built environments. However, such simulations are challenging due to the multi-physics, multi-scale nature of essentially all fires. In this talk, we present new results from a computational effort focused on understanding and characterizing wildland fire spread at small scales (roughly 1m–1mm) using direct numerical simulations (DNS). The cost of the simulations is reduced using adaptive mesh refinement (AMR), where resolution is provided only when and where it is needed to resolve physically-relevant fine-scale features. Simulation results are shown for both large-scale plumes and pool fires studied experimentally in the FLAME facility at Sandia National Lab. In the plume configuration, helium gas is released into ambient air from a 1m inlet. In the pool fire configuration, methane is released into ambient air from the same 1m inlet, before burning as a non-premixed diffusion flame. Comparisons are made between results from the simulations with and without AMR, and between the simulations and experiments. Focus is placed, in particular, on the computational savings enabled by the use of AMR, in addition to simulation accuracy. |
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