The subgrid-scale modelling of a low Mach number strongly anisothermal turbulent flow is investigated using direct numerical simulations. The study is based on the filtering of the low Mach number equations, suited to low Mach number flows with highly variable fluid properties. The results are relevant to formulations of the filtered low Mach number equations established with the classical filter or the Favre filter. The two most significant subgrid terms of the filtered low Mach number equations are considered. They are associated with the momentum convection and the density-velocity correlation. We focus on eddy-viscosity and eddy-diffusivity models. Subgrid-scale models from the literature are analysed and two new models are proposed. The subgrid-scale models are compared to the exact subgrid term using the instantaneous flow field of the direct numerical simulation of a strongly anisothermal fully developed turbulent channel flow. There is no significant differences between the use of the classical and Favre filter regarding the performance of the models. We suggest that the models should take into account the asymptotic near-wall behaviour of the filter length. Eddy-viscosity and eddy-diffusivity models are able to represent the energetic contribution of the subgrid term but not its effect in the flow governing equations. The AMD and scalar AMD models are found to be in better agreement with the exact subgrid terms than the other investigated models in the a priori tests.
This work presents a new approach for premixed turbulent combustion modeling based on convolutional neural networks (CNN). We first propose a framework to reformulate the problem of subgrid flame surface density estimation as a machine learning task. Data needed to train the CNN is produced by direct numerical simulations (DNS) of a premixed turbulent flame stabilized in a slot-burner configuration. A CNN inspired from a U-Net architecture is designed and trained on the DNS fields to estimate sub-grid scale wrinkling. It is then tested on an unsteady turbulent flame where the mean inlet veloci…
A nonlocal subgrid-scale stress (SGS) model is developed based on the convolution neural network (CNN), a powerful supervised data-driven approach. The CNN is an ideal approach to naturally consider nonlocal spatial information in prediction due to its wide receptive field. The CNN-based models used here only take primitive flow variables as input, then the flow features are automatically extracted without any $priori$ guidance. The nonlocal models trained by direct numerical simulation (DNS) data of a turbulent channel flow at $Re_τ=178$ are accessed in both the $priori$ and $posteriori$ test…
Spatial and spectral energy exchanges associated with the turbulence kinetic energy per unit mass, or half-trace of the velocity covariance tensor, are studied in an anisothermal low Mach number turbulent channel flow. The temperatures of the two channel walls are 293 K and 586 K. This generates a strong temperature gradient in the wall-normal direction. The effect of the temperature gradient on the energy exchanges is investigated using two direct numerical simulations of the channel, at the mean friction Reynolds numbers 180 and 395. The temperature gradient creates an asymmetry between the …
We use machine learning to perform super-resolution analysis of grossly under-resolved turbulent flow field data to reconstruct the high-resolution flow field. Two machine-learning models are developed; namely the convolutional neural network (CNN) and the hybrid Downsampled Skip-Connection Multi-Scale (DSC/MS) models. These machine-learning models are applied to two-dimensional cylinder wake as a preliminary test and show remarkable ability to reconstruct laminar flow from low-resolution flow field data. We further assess the performance of these models for two-dimensional homogeneous turbule…
This work determines the inaccuracy of using Reynolds averaged Navier Stokes (RANS) turbulence models in transition to turbulent flow regimes by predicting the model-based discrepancies between RANS and large eddy simulation (LES) models and then incorporates the capabilities of machine learning algorithms to characterize the discrepancies which are defined as a function of mean flow properties of RANS simulations. First, three-dimensional CFD simulations using k-omega Shear Stress Transport (SST) and dynamic one-equation subgrid-scale models are conducted in a wall-bounded channel containing …
Even though compressible plasma turbulence is encountered in many astrophysical phenomena, its effect is often not well understood. Furthermore, direct numerical simulations are typically not able to reach the extreme parameters of these processes. For this reason, large-eddy simulations (LES), which only simulate large and intermediate scales directly, are employed. The smallest, unresolved scales and the interactions between small and large scales are introduced by means of a subgrid-scale (SGS) model. We propose and verify a new set of nonlinear SGS closures for future application as an SGS…
Large eddy simulations (LES) are a powerful tool in understanding processes that are inaccessible by direct simulations due to their complexity, for example, in the highly turbulent regime. However, their accuracy and success depends on a proper subgrid-scale (SGS) model that accounts for the unresolved scales in the simulation. We evaluate the applicability of two traditional SGS models, namely the eddy-viscosity (EV) and the scale-similarity (SS) model, and one recently proposed nonlinear (NL) SGS model in the realm of compressible MHD turbulence. Using 209 simulations of decaying, supersoni…
Large-scale river models are being refined over coastal regions to improve the scientific understanding of coastal processes, hazards and responses to climate change. However, coarse mesh resolutions and approximations in physical representations of tidal rivers limit the performance of such models at resolving the complex flow dynamics especially near the river-ocean interface, resulting in inaccurate simulations of flood inundation. In this research, we propose a machine learning (ML) framework based on the state-of-the-art physics-informed neural network (PINN) to simulate the downscaled fl…
A hybrid lattice Boltzmann/finite-difference solver for low Mach thermo-compressible flows developed in earlier works is extended to more realistic and challenging configurations involving turbulence and complex geometries in the present article. The major novelty here as compared to previous contributions is the application of a more robust collision operator, considerably extending the stability of the original single relaxation time model and facilitating larger Reynolds number flow simulations. Additionally, a subgrid model and the thickened flame approach have also been added allowing for…
The direct numerical simulation (DNS) of compressible isotropic turbulence up to the supersonic regime $Ma_{t} = 1.2$ has been investigated by high-order gas-kinetic scheme (HGKS) [{\it{Computers}} \& {\it{Fluids, 192, 2019}}]. In this study, the coarse-grained analysis of subgrid-scale (SGS) turbulent kinetic energy $K_{sgs}$ budget is fully analyzed for constructing one-equation SGS model in the compressible large eddy simulation (LES). The DNS on a much higher turbulent Mach number up to $Ma_{t} = 2.0$ has been obtained by HGKS, which confirms the super robustness of HGKS. Then, the exact c…
In the present work we assess the capabilities of neural networks to predict temporally evolving turbulent flows. In particular, we use the nine-equation shear flow model by Moehlis et al. [New J. Phys. 6, 56 (2004)] to generate training data for two types of neural networks: the multilayer perceptron (MLP) and the long short-term memory (LSTM) network. We tested a number of neural network architectures by varying the number of layers, number of units per layer, dimension of the input, weight initialization and activation functions in order to obtain the best configurations for flow prediction…
We compare the statistics of driven, supersonic turbulence at high Mach number using FLASH a widely used Eulerian grid-based code and PHANTOM, a Lagrangian SPH code at resolutions of up to 512^3 in both grid cells and SPH particles. We find excellent agreement between codes on the basic statistical properties: a slope of k^-1.95 in the velocity power spectrum for hydrodynamic, Mach 10 turbulence, evidence in both codes for a Kolmogorov-like slope of k^-5/3 in the variable rho^1/3 v as suggested by Kritsuk et al. and a log-normal PDF with a width that scales with Mach number and proportionality…
Super-resolution of turbulent reacting flows on complex meshes using graph neural networks
FLAME Laboratory, Department of Computational and Data Sciences, Indian Institute of Science, Bengaluru, 560012, India（印度科学研究院流体动力学与能源实验室，班加罗尔）；CAPS Laboratory, Department of Mechanical and Process Engineering, ETH Zürich, 8092 Zürich, Switzerland（苏黎世联邦理工学院机械与过程工程实验室，苏黎世）；Jülich Supercomputing Centre, Forschungszentrum Jülich GmbH, Jülich, 52425, Germany（尤利希超算中心，尤利希）
