A deep learning (DL) closure model for large-eddy simulation (LES) is developed and evaluated for incompressible flows around a rectangular cylinder at moderate Reynolds numbers. Near-wall flow simulation remains a central challenge in aerodynamic modeling: RANS predictions of separated flows are often inaccurate, while LES can require prohibitively small near-wall mesh sizes. The DL-LES model is trained using adjoint PDE optimization methods to match, as closely as possible, direct numerical simulation (DNS) data. It is then evaluated out-of-sample (i.e., for new aspect ratios and Reynolds numbers not included in the training data) and compared against a standard LES model (the dynamic Smagorinsky model). The DL-LES model outperforms dynamic Smagorinsky and is able to achieve accurate LES predictions on a relatively coarse mesh (downsampled from the DNS grid by a factor of four in each Cartesian direction). We study the accuracy of the DL-LES model for predicting the drag coefficient, mean flow, and Reynolds stress. A crucial challenge is that the LES quantities of interest are the steady-state flow statistics; for example, the time-averaged mean velocity $\bar{u}(x) = \displaystyle \lim_{t \rightarrow \infty} \frac{1}{t} \int_0^t u(s,x) ds$. Calculating the steady-state flow statistics therefore requires simulating the DL-LES equations over a large number of flow times through the domain; it is a non-trivial question whether an unsteady partial differential equation model whose functional form is defined by a deep neural network can remain stable and accurate on $t \in [0, \infty)$. Our results demonstrate that the DL-LES model is accurate and stable over large physical time spans, enabling the estimation of the steady-state statistics for the velocity, fluctuations, and drag coefficient of turbulent flows around bluff bodies relevant to aerodynamic applications.
The potential for data-driven applications to scale-resolving simulations of turbulent flows is assessed herein. Multigrid sequential data assimilation algorithms have been used to calibrate solvers for Large Eddy Simulation for the analysis of the high-Reynolds-number flow around a rectangular cylinder of aspect ratio 5:1. This test case has been chosen because of a number of physical complexities which elude accurate representation using reduced-order numerical simulation. The results for the statistical moments of the velocity and pressure flow field show that the data-driven techniques emp…
In this short note we apply the recently proposed data-driven RANS closure modelling framework of Schmelzer et al. (2020) to fully three-dimensional, high Reynolds number flows: namely wall-mounted cubes and cuboids at Re=40,000, and a cylinder at Re=140,000. For each flow, a new RANS closure is generated using sparse symbolic regression based on LES or DES reference data. This new model is implemented in a CFD solver, and subsequently applied to prediction of the other flows. We see consistent improvements compared to the baseline $k-ω$ SST model in predictions of mean-velocity in the complet…
We propose a method to construct a reduced order model with machine learning for unsteady flows. The present machine-learned reduced order model (ML-ROM) is constructed by combining a convolutional neural network autoencoder (CNN-AE) and a long short-term memory (LSTM), which are trained in a sequential manner. First, the CNN-AE is trained using direct numerical simulation (DNS) data so as to map the high-dimensional flow data into low-dimensional latent space. Then, the LSTM is utilized to establish a temporal prediction system for the low-dimensionalized vectors obtained by CNN-AE. As a test…
Aerodynamic pressure field over bluff bodies immersed in boundary layer flows is correlated both in space and time. Conventional approaches for the analysis of distributed aerodynamic pressures, e.g., the proper orthogonal decomposition (POD), can only offer relevant spatial patterns in a set of coherent structures. This study provides an operator-theoretic approach that describes dynamic pressure fields in a functional space rather than conventional phase space via the Koopman operator. Subsequently, spectral analysis of the Koopman operator provides a spatiotemporal characterization of the p…
The LES/PDF methodology is applied to the Cambridge/Sandia turbulent stratified flame series under swirling conditions. The methane/air chemistry is represented by a 16-species reduced ARM1 mechanism, and the in situ adaptive tabulation method is adopted to accelerate the chemistry calculation. Effects of differential diffusion and heat loss through the bluffbody surface are taken into account. The simulations are conducted for premixed (SwB3), moderately and highly stratified (SwB7 and SwB11, respectively) cases under swirling conditions. The results from LES/PDF simulations are compared with…
Large eddy simulations and three-dimensional proper orthogonal decomposition were used to study the interaction between a large stationary and moving bluff body and a high Reynolds number stably-stratified turbulent boundary layer. An immersed boundary method is utilized to take into account the effect of the bluff body on the boundary layer flow and turbulent inflow conditions upstream of the bluff body are imposed by employing a concurrent precursor simulation method both because a pseudo-spectral method is utilized in the horizontal directions. The dominant POD mode type was observed to occ…
A large eddy simulation (LES) study of the flow around a 1/4 scale squareback Ahmed body at $Re_H=33,333$ is presented. The study consists of both wall-resolved (WRLES) and wall-modelled (WMLES) simulations, and investigates the bimodal switching of the wake between different horizontal positions. Within a non-dimensional time-window of 1050 convective flow units, both WRLES and WMLES simulations, for which only the near-wall region of the turbulent boundary layer is treated in a Reynolds-averaged sense, are able to capture horizontal (spanwise) shifts in the wake's cross-stream orientation. E…
We propose a novel active-flow-control (AFC) strategy for bluff bodies to hide their hydrodynamic traces from predators. A group of windward-suction-leeward-blowing (WSLB) actuators are adopted to control the wake of a circular cylinder submerged in a uniform flow. An array of velocity sensors are deployed in the near wake to provide feedback signals. Through the data-driven deep reinforcement learning (DRL), effective control strategies are trained for the WSLB actuation to mitigate the cylinder's hydrodynamic signatures, i.e., strong shears and periodically shed vortices. Only a 0.29% defici…
FR3D: Three-dimensional Flow Reconstruction and Force Estimation for Unsteady Flows Around Extruded Bluff Bodies via Conformal Mapping Aided Convolutional Autoencoders
Flow over bluff bodies has multiple engineering applications and thus, has been studied for decades. The lift and drag coefficients are practically important in the design of many components such as automobiles, aircrafts, buildings etc. These coefficients vary significantly with Reynolds number and geometric parameters of the bluff body. In this study, we have analyzed the sensitivity of lift and drag coefficients on single and tandem elliptic cylinders to cylinder aspect ratios, angles of attack, cylinder separation, and flow Reynolds number. Sensitivity analysis with Monte-Carlo algorithm r…
On the spatial structure and intermittency of soot in a lab-scale gas turbine combustor: Insights from large-eddy simulations
Barcelona Supercomputing Center（巴塞罗那超级计算中心）；Eindhoven University of Technology（埃因霍温理工大学）；Eindhoven Institute for Renewable Energy Systems（埃因霍温可再生能源系统研究所）
We propose an aerodynamic force model associated with a Galerkin model for the unforced fluidic pinball, the two-dimensional flow around three equal cylinders with one radius distance to each other. The starting point is a Galerkin model of a bluff-body flow. The force on this body is derived as a constant-linear-quadratic function of the mode amplitudes from first principles following the pioneering work of Noca (1997, 1999) and Liang & Dong (2014). The force model is simplified for the mean-field model of the unforced fluidic pinball (Deng et al. 2020) using symmetry properties and sparse ca…
In RANS simulations for turbulent scalar transport, it is common that using an eddy-viscosity (EV) model to close the Reynolds stress yields reasonable mean flow predictions but large errors in scalar transfer results regardless of scalar flux model inadequacies. This failure mode of EV models is generally related to the fact that the transport of momentum and scalar depends on different Reynolds stress components. The present work addresses two common issues relevant to such failures in turbulent scalar transport around obstacles. The first issue is the general overprediction of scalar transf…
We present an efficient deep learning technique for the model reduction of the Navier-Stokes equations for unsteady flow problems. The proposed technique relies on the Convolutional Neural Network (CNN) and the stochastic gradient descent method. Of particular interest is to predict the unsteady fluid forces for different bluff body shapes at low Reynolds number. The discrete convolution process with a nonlinear rectification is employed to approximate the mapping between the bluff-body shape and the fluid forces. The deep neural network is fed by the Euclidean distance function as the input a…
Large-eddy simulation (LES) of a turbulent flow through an array of building-like obstacles is an idealized model to study transport of contaminants in the urban atmospheric boundary layer (UABL). A reasonably accurate LES prediction of turbulence in such an UABL must resolve a significant proportion of the small but energetic eddies in the roughness sublayer, which remains prohibitive even though computational power has been increased significantly. In this article, we present a large-eddy simulation methodology to study turbulence in UABLs, where the turbulence closure is based on coupling t…
Although the stability properties of the wake past impervious bluff bodies have been widely examined in the literature, similar analyses regarding the flow around and through porous ones are still lacking. In this work, the effect of the porosity and permeability on the wake patterns of porous rectangular cylinders is numerically investigated at low to moderate Reynolds numbers in the framework of direct numerical simulation combined with local and global stability analyses. A modified Darcy-Brinkman formulation is employed here so as to describe the flow behavior inside the porous media, wher…
The purpose of this work is to expand the work of Streher (2017) in order to investigate the aeroelastic instabilities generated by the flow around a moving NACA0012 airfoil. The profile has a chord length of $c=0.1 m and is exposed to a flow at a Reynolds number of Re=30,000. The airfoil has only two degrees of freedom: Translation in relation to the vertical direction and a rotation around the span-wise axis. A partitioned approach based on two separate solvers and a fluid-structure interaction (FSI) coupling scheme is applied. The in-house CFD solver FASTEST-3D computes the fluid sub-proble…
A numerical study of stably stratified flows past spheres at Reynolds numbers $Re=200$ and $Re=300$ is reported. In these flow regimes, a neutrally stratified laminar flow induces distinctly different near-wake features. However, the flow behaviour changes significantly as the stratification increases and suppresses the scale of vertical displacements of fluid parcels. Computations for a range of Froude numbers $Fr\in [0.1,\infty]$ show that as Froude number decreases, the flow patterns for both Reynolds numbers become similar. The representative simulations of the lee-wave instability at $Fr=…
In computational fluid dynamics, there is an inevitable trade off between accuracy and computational cost. In this work, a novel multi-fidelity deep generative model is introduced for the surrogate modeling of high-fidelity turbulent flow fields given the solution of a computationally inexpensive but inaccurate low-fidelity solver. The resulting surrogate is able to generate physically accurate turbulent realizations at a computational cost magnitudes lower than that of a high-fidelity simulation. The deep generative model developed is a conditional invertible neural network, built with normal…
The interaction between turbulent axisymmetric wakes plays an important role in many industrial applications, notably in the modelling of wind farms. While the non-equilibrium high Reynolds number scalings present in the wake of axisymmetric plates has been shown to modify the averaged streamwise scalings of individual wakes, little attention has been paid to their consequences in terms of wake interactions. We propose an experimental setup that tests the presence of non-equilibrium turbulence using the streamwise variation of velocity fluctuations between two bluff bodies facing a laminar flo…
Formation of recirculating wakes is a prominent feature of inertial flow around bluff bodies. Below the onset of vortex shedding in uniform unbounded flows, the fluid in the recirculating wake region moves on closed planar orbits. The steady wake is thus an isolated zone in the flow and does not exchange fluid with the free stream. In this work, we utilize lattice-Boltzmann simulations and microfluidic experiments to demonstrate that in microchannel inertial flow of Newtonian fluids, the recirculating wake is replaced by a three-dimensional vortical flow. Spiraling streamlines generate a conti…
Machine-learning-based simulation of turbulent flows over periodic hills using a hybrid U-Net and Fourier neural operator framework
Southern University of Science and Technology（南方科技大学）；Harbin Engineering University Qingdao Innovation and Development Base（哈尔滨工程大学青岛创新发展基地）；National University of Singapore（新加坡国立大学）；Nanchang Hangkong University（南昌航空大学）
In this paper, an end-to-end nonlinear model reduction methodology is presented based on the convolutional recurrent autoencoder networks. The methodology is developed in the context of the overall data-driven reduced-order model framework proposed in the paper. The basic idea behind the methodology is to obtain the low dimensional representations via convolutional neural networks and evolve these low dimensional features via recurrent neural networks in the time domain. The high dimensional representations are constructed from the evolved low dimensional features via transpose convolutional n…
IB-Flows: an open-source multi-GPU immersed boundary code for fluid-structure interaction
Gran Sasso Science Institute (GSSI)（格兰萨索科学研究所）；INFN–Laboratori Nazionali del Gran Sasso（意大利国家核物理研究所-格兰萨索国家实验室）；Department of Physics and INFN, University of Rome, Tor Vergata（罗马托尔维加塔大学物理系及意大利国家核物理研究所）；NVIDIA Corporation（英伟达公司）；Physics of Fluids Group, University of Twente（特温特大学流体力学组）
