Learning closure models for Turbulence: Calibration of BHR using Machine Learning techniques
Explore the source record for details and available documents.
Djibrilla Boureima, Ismael↗
Engineering topics
Kurien, Susan
Publications and source records attributed to Kurien, Susan.
The local wavenumber model for computation of turbulent mixing
We present an overview of the current status in the development of a two-point spectral closure model for turbulent flows, known as the local wavenumber (LWN) model. The model is envisioned as a practical option for applications requiring multi-physics simulations in which statistical hydrodynamics quantities such as Reynolds stresses, turbulent kinetic energy, and measures of mixing such as density-correlations and mix-width evolution, need to be captured with relatively high fidelity. In this review, we present the capabilities of the LWN model since it was first formulated in the early 1990s, for computations of increasing levels of complexity ranging from homogeneous isotropic turbulence, inhomogeneous and anisotropic single-fluid turbulence, to two-species mixing driven by buoyancy forces. The review concludes with a discussion of some of the more theoretical considerations that remain in the development of this model.
42 ENGINEERING↗
Dynamic calibration of differential equations using machine learning, with application to turbulence models
We present a methodology for calibration of parametric ordinary and partial differential equation models, using off-the-shelf software for back-propagation in Neural Networks (NN). As a prototypical example, we consider calibration of a Reynolds-averaged Navier-Stokes (RANS) turbulence closure model, against ground truth data from direct numerical simulations (DNS) of two different turbulent flows. Numerical time integration is represented as a custom NN, where only the RANS model parameters are trainable. A loss function is defined to quantify the mismatch between the NN prediction and the ground truth over a predefined, finite time integration window. This loss function is then minimized using a gradient descent method utilizing the back-propagation algorithm. Furthermore, this dynamic approach to training is to be contrasted with a static approach, wherein a least square regression estimate for parameters is obtained in the limit of an infinitesimal time integration window. In a first test of static and dynamic approaches against ground truth data generated by the model, the former proves to be significantly faster and more accurate than the latter at recovering the parameters. When both calibration approaches are tested against DNS data, for which it is known that the model cannot achieve a perfect fit, the static approach yields a good prediction only for short times, while the dynamic approach results in physical and stable predictions over the entire integration window. After optimization of the dynamic approach for time step, spatial resolution, stability, and physics-based constraints, we obtain a 50% improvement of outcomes over those obtained from the existing, manually calibrated set of parameters, demonstrating the merits of this systematic and automated procedure.
97 MATHEMATICS AND COMPUTING↗
Local wave-number model for inhomogeneous two-fluid mixing
We analyze the local wave-number (LWN) model, a two-point spectral closure model for turbulence, as applied to the Rayleigh-Taylor (RT) instability, the flow induced by the relaxation of a statically-unstable density stratification. Model outcomes are validated against data from 3D simulations of the RT instability. In the first part of the study we consider the minimal model terms required to capture inhomogeneous mixing and show that this version, with suitable model coefficients, is sufficient to capture the evolution of important mean global quantities including mix-width, turbulent mass flux velocity, and Reynolds stress, if the start time is chosen such that the earliest transitions are avoided. However, this simple model does not permit the expected finite asymptote of the density-specific-volume covariance b. In the second part of the study, we investigate two forms for a source term for the evolution of the spectrum of density-specific-volume covariance for the LWN model. The first includes an empirically motivated calibration of the source to achieve the final asymptotic state of constant b. Here, the second form does not require calibration but, in conjunction with enhanced diffusion and drag captures the full evolution of all the dynamical quantities, namely, the mix-layer growth, turbulent mass-flux velocity, Reynolds stress, as well as the desired behavior of b.
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
Machine-Learning for Rapid Optimization of Turbulence Models
1. Background and Motivation: What is turbulence; Why is turbulence important; How turbulence is modeled; BHR model; Motivation for a ML framework 2. DNS Database from OES-C4; 3. Machine learning framework: Formulation; Verification; Case scenarios; 4. Summary and Conclusions
42 ENGINEERING↗
Extracting a mixing parameter from 2D radiographic imaging of variable-density turbulent flow
We extract a suitably averaged fluctuating density from the two-dimensional radiographic image of a flow. The X-ray attenuation is given by the Beer–Lambert law which exponentially damps the incident beam intensity by a factor proportional to the density, opacity and thickness of the target. By making reasonable assumptions for the mean density, opacity and effective thickness of the target flow, we estimate the density fluctuation contribution to the attenuation. The extracted density fluctuations averaged across the thickness of the flow in the direction of the beam may be used to form the density–specific-volume correlation $b$. In a statistical description of variable-density turbulence, $b$ quantifies the degree of mixedness. The ability to extract a measure of mixedness from experimental data would be a powerful tool that could be used in the validation of mix models. The scheme proposed is tested for DNS data computed for variable density buoyancy-driven mixing. We quantify the deficits in the extracted value of $b$ due to target thickness, Atwood number and modeled signal noise. This analysis justifies using the proposed scheme to infer the mix parameter from thin targets at moderate to low Atwood numbers. Furthermore, to illustrate how the scheme might be used in a practical problem, we demonstrate its application to a radiographic image of counter-shear flow obtained from experiments at the National Ignition Facility.
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗