Within the domain of Computational Fluid Dynamics, Direct Numerical Simulation (DNS) is used to obtain highly accurate numerical solutions for fluid flows. However, this approach for numerically solving the Navier-Stokes equations is extremely computationally expensive mostly due to the requirement of greatly refined grids. Large Eddy Simulation (LES) presents a more computationally efficient approach for solving fluid flows on lower-resolution (LR) grids but results in an overall reduction in solution fidelity. Through this paper, we introduce a novel deep learning framework SR-DNS Net, which aims to mitigate this inherent trade-off between solution fidelity and computational complexity by leveraging deep learning techniques used in image super-resolution. Using our model, we wish to learn the mapping from a coarser LR solution to a refined high-resolution (HR) DNS solution so as to eliminate the need for performing DNS on highly refined grids. Our model efficiently reconstructs the high-fidelity DNS data from the LES like low-resolution solutions while yielding good reconstruction metrics. Thus our implementation improves the solution accuracy of LR solutions while incurring only a marginal increase in computational cost required for deploying the trained deep learning model.
This review focuses on Direct numerical simulations (DNS) of turbulent flows laden with droplets or bubbles. DNS of these flows are more challenging than those of flows laden with solid particles due to the surface deformation in the former. The classification of the discussed numerical methods is based on whether the initial diameter of the bubble/droplet is smaller or larger than the Kolmogorov length scale, and whether the instantaneous surface deformation is fully resolved or obtained via a phenomenological model. Numerical methods that account for the breakup of single droplet/bubble as w…
Starting with an integral formulation of mass flow rate through an ensemble of isotherms constituting a statistically planar, turbulent premixed flame, a scaling for the corresponding turbulent flame speed is derived without invoking Damk{ö}hler's hypotheses. Major approximations and interim results are validated using a large Karlovitz number, unity Lewis number, Direct Numerical Simulation (DNS) dataset of n-heptane/air mixture, computed with reduced chemistry. A new length scale quantifying the fluctuation distance of the isotherms within the premixed flame structure is introduced.
The entrainment of ambient fluid into a turbulent shear flow has been a topic of wide interest for several decades. To estimate the entrainment of ambient fluid into turbulent jet, it is essential to define the boundary of the jet. The question arises as to what should be the appropriate criterion to determine the edge of the turbulent jet. From the present DNS simulations, we observe that there is a need to define two boundaries for the turbulent jet, namely, the Rotational/Irrotational (R/IR) interface termed here as "Outer boundary" and the Turbulent/Non-Turbulent (T/NT) interface as "Inner…
The turbulent flow in an infinitely extended plane channel is analysed by solving the Navier-Stokes equations with a DNS approach. Solutions are obtained in a numerical solution domain of finite size in the streamwise as well as in the lateral direction setting periodic boundary conditions in both directions. Their impact on large scale structures in the turbulent flow field is analysed carefully in order to avoid their suppression. When this is done appropriately well known stripe patterns in these flows can be observed and analysed especially with respect to their relative motion compared to…
We present a new DNS data set for a statistically axisymmetric turbulent jet, plume and forced plume in a domain of size $40 r_0 \times 40 r_0 \times 60 r_0$, where $r_0$ is the source diameter. The data set provides evidence of the validity of the Priestley and Ball entrainment model in unstratified environments (excluding the region near the source), which is corroborated further by the Wang and Law and Ezzamel \emph{et al.} experimental data sets, the latter being corrected for a small but influential co-flow that affected the statistics. We show that the turbulence in the core region of th…
RWTH Aachen University（亚琛工业大学）；Institute for Combustion Technology（燃烧技术研究所）；CORIA – CNRS UMR 6614（CORIA——法国国家科学研究中心6614联合研究单位）；Jülich Supercomputing Centre, FZ Jülich（于利希研究中心于利希超级计算中心）
In this paper, deep learning (DL) methods are evaluated in the context of turbulent flows. Various generative adversarial networks (GANs) are discussed with respect to their suitability for understanding and modeling turbulence. Wasserstein GANs (WGANs) are then chosen to generate small-scale turbulence. Highly resolved direct numerical simulation (DNS) turbulent data is used for training the WGANs and the effect of network parameters, such as learning rate and loss function, is studied. Qualitatively good agreement between DNS input data and generated turbulent structures is shown. A quantita…
Direct numerical simulation (DNS) of turbulent flows is computationally expensive and cannot be applied to flows with large Reynolds numbers. Large eddy simulation (LES) is an alternative that is computationally less demanding, but is unable to capture all of the scales of turbulent transport accurately. Our goal in this work is to build a new data-driven methodology based on super-resolution techniques to reconstruct DNS data from LES predictions. We leverage the underlying physical relationships to regularize the relationships amongst different physical variables. We also introduce a hierarc…
In this report we briefly summarize the set-up and results for direct numerical simulations (DNS) of turbulent channel flows with riblets. The DNS are carried out with OpenFOAM which allows the use of an unstructured mesh. The present results were presented at the European Drag Reduction and Flow Control Meeting in 2015.
Direct numerical simulations (DNS) of fully-developed turbulent channel flows for very low Reynolds numbers have been performed with a larger computational box sizes than those of existing DNS. The friction Reynolds number was decreased down to 60, where the friction Reynolds number is based on the friction velocity and the channel half width. When the Reynolds number was decreased to 60 with small computational box size, the flow became laminar. Using a large box, we found that a localized turbulence was observed to sustain in the form of periodic oblique band. This type of locally disordered…
Turbulence in the restricted nonlinear (RNL) dynamics is analyzed and compared with DNS of Poiseuille turbulence at $R=1650$. The structures are obtained by POD analysis of the two components of the flow partition used in RNL dynamics: the streamwise-mean flow and fluctuations. POD analysis of the streamwise-mean flow indicates that the dominant POD modes, in both DNS and RNL, are roll-streaks harmonic in the spanwise. However, we conclude that these POD modes do not occur in isolation but rather are Fourier components of a coherent roll-streak structure. POD analysis of the fluctuations in DN…
Results of numerical simulations obtained by a staggered finite difference scheme together with an efficient immersed boundary method are presented to understand the effects of the shape of three-dimensional obstacles on the transition of a boundary layer from a laminar to a turbulent regime. Fully resolved Direct Numerical Simulations (DNS), highlight that the closer to the obstacle the symmetry is disrupted the smaller is the transitional Reynolds number. It has been also found that the transition can not be related to the critical roughness Reynolds number used in the past. The simulations …
The large structures in the outer layer of turbulent wall flows are of great physical importance, because they contain a substantial fraction of the streamwise kinetic energy and of the Reynolds stresses. Nevertheless, the organization of the outer region of wall turbulence has historically received less attention than that of the inner region, and as a consequence, it is still the subject of many open experimental and theoretical questions. In order to address some of those questions, we have performed direct numerical simulations of the turbulent incompressible flow in plane channels at Reyn…
We carry out Direct Numerical Simulation (DNS) of flows in closed rectangular ducts with several aspect ratios. The Navier-Stokes equations are discretized through a second-order finite difference scheme, with non-uniform grids in two directions. The duct cross-sectional area is maintained constant as well as the flow rate, which allows to investigate which is the appropriate length scale in the Reynolds number for a good scaling in the laminar and in the fully turbulent regimes. We find that the Reynolds number based on the half length of the short side leads to a critical Reynolds number whi…
This investigation is based on a DNS of a steady self-preserving incompressible axisymmetric turbulent jet at a Reynolds number of 2400. The DNS data enable accurate maps of the outer irrotational flow field, and also the vorticity field in the turbulent core of the jet. It is found necessary to define two separate boundaries of the jet. The first is an inner boundary (turbulent/nonturbulent, T/NT), from where vorticity rises steeply towards to the core. The second is an outer rotational/irrotational boundary, beyond which the flow may be considered irrotational. The velocity field beyond the …
A Direct Numerical Simulation (DNS) of the incompressible flow around a rectangular cylinder with chord-to-thickness ratio 5:1 (also known as the BARC benchmark) is presented. The work replicates the first DNS of this kind recently presented by Cimarelli et al (2018), and intends to contribute to a solid numerical benchmark, albeit at a relatively low value of the Reynolds number. The study differentiates from previous work by using an in-house finite-differences solver instead of the finite-volumes toolbox OpenFOAM, and by employing finer spatial discretization and longer temporal average. Th…
We investigate and model pressure effects on the turbulent burning velocity over a wide range of pressures and turbulence intensities with the direct numerical simulation (DNS) of statistically planar turbulent premixed flames for lean hydrogen/air mixture. DNS results indicate that the stretch factor has an impact on the turbulent burning velocity and flame surface area at elevated pressures. In particular, the enhanced stretch factor at high pressures increases the ratio of turbulent and laminar burning velocities, diminishing the "bending" effect. Based on a good consistency between turbule…
This paper concerns the study of direct numerical simulation (DNS) data of a wavepacket in laminar turbulent transition in a Blasius boundary layer. The decomposition of this wavepacket into a set of "modes" (a basis that spans an approximate solution space) can be achieved in a wide variety of ways. Two well-known tools are the fast Fourier transform (FFT) and the proper orthogonal decomposition (POD). To synergize the strengths of both methods, a hybrid POD-FFT is pioneered, using the FFT as a tool for interpreting the POD modes. The POD-FFT automatically identifies well-known fundamental, s…
We have conducted a direct numerical simulation (DNS) study of dilute turbulent particulate flow in a vertical plane channel, considering thousands of finite-size rigid particles with resolved phase interfaces. The particle diameter corresponds to approximately 11 wall units and their terminal Reynolds number is set to 136. The fluid flow with bulk Reynolds number 2700 is directed upward, which maintains the particles suspended upon average. Two density ratios were simulated, differing by a factor of 4.5. The corresponding Stokes numbers of the two flow cases were O(10) in the near-wall region…
Direct numerical simulations of turbulent heat transfer in fully-developed channel flows have been performed in a range of friction Reynolds number between 60 and 180, based on the friction velocity and the channel half width $δ$, with emphasis on a puff-like structure, large-scale spatial intermittency. For the Reynolds numbers lower than 80 with a large computational domain of 51.2 x 2 x 22.5, the turbulent puff was observed and its significant influences on the momentum and heat transports were found. The spatial structure of the equilibrium puff, or the localized turbulence, was examined w…
Direct numerical simulation of a thermal turbulent boundary layer: an analogy to simulate bushfires and a testbed for artificial intelligence remote sensing of bushfire propagation
An a priori analysis of a DNS database of turbulent lean premixed methane flames is presented considering the relative effects of turbulence and LES filtering, along with a potential modelling approach for LES with finite-rate chemistry. The leading-order effect was found to be due to the filter operation; flame response to turbulence was a secondary effect, and manifested primarily as an increase in standard deviations about conditional means. It was found that the radicals O, H and OH were less impacted by the filter than other high-temperature radicals, which were significantly reduced in m…
We investigate the fluctuations of thermodynamic state-variables in compressible aerodynamic wall-turbulence, using results of direct numerical simulation (DNS) of compressible turbulent plane channel flow. The basic transport equations governing the behaviour of thermodynamic variables (density, pressure, temperature and entropy) are reviewed and used to derive the exact transport equations for the variances and fluxes (transport by the fluctuating velocity field) of the thermodynamic fluctuations. The scaling with Reynolds and Mach number of compressible turbulent plane channel flow is discu…
We study turbulent flows in a smooth straight pipe of circular cross--section up to $Re_τ \approx 6000$ using direct--numerical-simulation (DNS) of the Navier--Stokes equations. The DNS results highlight systematic deviations from Prandtl friction law, amounting to about $2\%$, which would extrapolate to about $4\%$ at extreme Reynolds numbers. Data fitting of the DNS friction coefficient yields an estimated von Kármán constant $k \approx 0.387$, which nicely fits the mean velocity profile, and which supports universality of canonical wall-bounded flows. The same constant also applies to the p…
We present results of three-dimensional Direct Numerical Simulations (DNS) and Large Eddy Simulations (LES) of turbulent gravity currents with a Discontinuous Galerkin (DG) Finite Elements method. In particular, we consider the three-dimensional lock-exchange test case as a typical benchmark for gravity currents. Since, to the best of our knowledge, non-Boussinesq three-dimensional reference DNS are not available in the literature for this test case, we first perform a DNS experiment. The three-dimensional DNS allows to correctly capture the loss of coherence of the three-dimensional turbulent…
We propose a methodology for generating time-dependent turbulent inflow data with the aid of machine learning (ML), which has a possibility to replace conventional driver simulations or synthetic turbulent inflow generators. As for the ML model, we use an auto-encoder type convolutional neural network (CNN) with a multi-layer perceptron (MLP). For the test case, we study a fully-developed turbulent channel flow at the friction Reynolds number of ${\rm Re}_τ = 180$ for easiness of assessment. The ML models are trained using a time series of instantaneous velocity fields in a single cross-sectio…
The Leray-$α$ model reduces the range of active scales of the Navier-Stokes equations by smoothing the advective transport. Here we assess the potential of the Leray-$α$ model in its standard formulation to simulate wall-bounded flows. Three flow cases are considered: plane channel flow at $\Re_τ=590$, Rayleigh-Bénard convection at $\Ra=10^7$ and $\Pr=1$, and a side-heated vertical channel at $\Ra=5 \times 10^6$ and $\Pr=0.7$. The simulations are compared to results from a well resolved and coarse DNS. It is found that for all three flow cases, the variance in the velocity field increases as t…
Although the evolution of premixed flames in turbulence has been frequently studied, it is not well understood how small flames interact with large-scale turbulent flow motion. Since this question is of practical importance for the occurrence of cycle-to-cycle variations in spark ignition engines, the objective of the present work is to fundamentally differentiate early flame kernel development from well-established turbulent flame configurations. For this purpose, a DNS database consisting of three flames propagating in homogeneous isotropic turbulence (Falkenstein et al., Combust. Flame, 201…
University of Notre Dame（圣母大学）；Missouri University of Science and Technology（密苏里科技大学）；Virginia Tech（弗吉尼亚理工大学）
Modeled Reynolds stress is a major source of model-form uncertainties in Reynolds-averaged Navier-Stokes (RANS) simulations. Recently, a physics-informed machine-learning (PIML) approach has been proposed for reconstructing the discrepancies in RANS-modeled Reynolds stresses. The merits of the PIML framework has been demonstrated in several canonical incompressible flows. However, its performance on high-Mach-number flows is still not clear. In this work we use the PIML approach to predict the discrepancies in RANS modeled Reynolds stresses in high-Mach-number flat-plate turbulent boundary lay…
An innovative \textit{deep learning} approach has been adopted to formulate the eddy-viscosity for large eddy simulation (LES) of wall-bounded turbulent flows. A deep neural network (DNN) is developed which learns to evaluate the eddy-viscosity from a dataset generated for a channel flow at friction Reynolds number $Re_τ = 395$ using the Dynamic Smagorinsky subgrid-scale model. Later this DNN is employed to predict the eddy-viscosity for a number of grid configurations for channel flow at $Re_τ = 395$ and $590$. The statistics computed from the DNN based LES model show an excellent match with …
In this contribution we have briefly introduced the problem of turbulent thermal convection with a particular look at its transition to the ultimate regime and the resolution requirements needed for the direct numerical simulation of this flow.
We seek possible statistical consequences of the way a forcing term is added to the Navier--Stokes equations in the Direct Numerical Simulation (DNS) of incompressible channel flow. Simulations driven by constant flow rate, constant pressure gradient and constant power input are used to build large databases, and in particular to store the complete temporal trace of the wall-shear stress for later analysis. As these approaches correspond to different dynamical systems, it can in principle be envisaged that these differences are reflect by certain statistics of the turbulent flow field. The ins…
Recent attempts to use deep learning for super-resolution reconstruction of turbulent flows have used supervised learning, which requires paired data for training. This limitation hinders more practical applications of super-resolution reconstruction. Therefore, we present an unsupervised learning model that adopts a cycle-consistent generative adversarial network that can be trained with unpaired turbulence data for super-resolution reconstruction. Our model is validated using three examples: (i) recovering the original flow field from filtered data using direct numerical simulation (DNS) of …
The main objective of this work is to perform a detailed comparison of the lattice Boltzmann equation (LBE) and the recently developed discrete unified gas-kinetic scheme (DUGKS) methods for direct numerical simulation (DNS) of the decaying homogeneous isotropic turbulence (DHIT) in a periodic box. The flow fields and key statistical quantities computed by both methods are compared with those from pseudo-spectral (PS) method. The results show that the LBE and DUGKS have almost the same accuracy when the flow field is well-resolved, and that the LBE is less dissipative and is slightly more effi…
While the saturated magnetic energy is independent of viscosity in dynamo experiments, it remains viscosity-dependent in state-of-the-art 3D direct numerical simulations (DNS). Extrapolating such viscous scaling-laws to realistic parameter values leads to an underestimation of the magnetic energy by several orders of magnitude. The origin of this discrepancy is that fully 3D DNS cannot reach low enough values of the magnetic Prandtl number $Pm$. To bypass this limitation and investigate dynamo saturation at very low $Pm$, we focus on the vicinity of the dynamo threshold in a rapidly rotating f…
In this paper we employ renormalized viscosity and thermal diffusivity to construct a subgrid-scale model for large eddy simulation (LES) of turbulent thermal convection. For LES, we add $ν_\mathrm{ren} \propto Π_u^{1/3} (π/Δ)^{-1/3}$ to the kinematic viscosity; here $Π_u$ is the turbulent kinetic energy flux, and $Δ$ is the grid spacing. In our model, the turbulent Prandtl number is unity. We performed LES of turbulent thermal convection on a $128^3$ grid and compare the results with direct numerical simulation (DNS) on a $512^3$ grid. There is a good agreement between the LES and DNS results…
The sources of wall-pressure fluctuations in turbulent channel flow are studied using a novel framework. The wall-pressure power spectral density (PSD) $(ϕ_{pp}(ω))$ is expressed as an integrated contribution from all wall-parallel plane pairs, $ϕ_{pp}(ω)=\int_{-δ}^{+δ}\int_{-δ}^{+δ}Γ(r,s,ω)\,\mathrm{dr}\,\mathrm{ds}$, using the Green's function. Here, $Γ(r,s,ω)$ is termed the net source cross spectral density (CSD) between two wall-parallel planes, $y=r$ and $y=s$ and $δ$ is the half channel height. Direct Numerical Simulation (DNS) data at friction Reynolds number of $180$ and $400$ are used…
Simulating the collision behaviour of cloud drops in a turbulent environment is numerically expensive. Because of the typical sizes of cloud drops, their motion is predominantly influenced by the smallest turbulent scales in the flow. We can exploit this property by using an Arnold--Beltrami--Childress (ABC) flow instead of a full direct numerical simulation (DNS) to simulate the turbulent effect on cloud drop collisions. This allows simulation of drop motion using much less computational resources than needed by DNS and therefore, allows for simulations with many more drops and larger model d…
Turbulent flows under transcritical conditions are present in regenerative cooling systems of rocker engines and extraction processes in chemical engineering. The turbulent flows and the corresponding heat transfer phenomena in these complex processes are still not well understood experimentally and numerically. The objective of this work is to investigate the turbulent flows under transcritical conditions using DNS of turbulent channel flows. A fully compressible solver is used in conjunction with a Peng-Robinson real-fluid equation of state to describe the transcritical flows. A channel flow…
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…
Inclined turbulent thermal convection by large Rayleigh numbers in extremely small-Prandtl-number fluids is studied based on results of both, measurements and high-resolution numerical simulations. The Prandtl number $Pr\approx0.0093$ considered in the experiments and the Large-Eddy Simulations (LES) and $Pr=0.0094$ considered in the Direct Numerical Simulations (DNS) correspond to liquid sodium, which is used in the experiments. Also similar are the studied Rayleigh numbers, which are, respectively, $Ra=1.67\times10^7$ in the DNS, $Ra=1.5\times10^7$ in the LES and $Ra=1.42\times10^7$ in the m…
The structure of homogeneous turbulent shear flow is studied using data generated by Direct Numerical Simulations (DNS) and a linear analysis for both compressible and incompressible cases. At large values of the mean shear rate, the Rapid Distortion Theory (RDT) limit is approached. Analytical solutions are found for the inviscid compressible RDT equations at long times. The RDT equations are also solved numerically for both inviscid and viscous cases. The RDT solutions, confirmed by the DNS results, show that the even order transverse derivative moments of the dilatational and solenoidal vel…
Data from a three-dimensional Direct Numerical Simulation of a turbulent premixed Bunsen flame at a low global Lewis number are analyzed to address the effects of the curvature on the local flame front. For this purpose, the chemical kinetics is modeled according to a reduced scheme, involving 5 reactions and 7 species, to mimic a H$_{2}$/Air flame at equivalence ratio $ϕ=0.5$. An increase of the local temperature and reaction rate is found for fronts elongated into the fresh gases (concave), while local quenching is observed for fronts elongated in the opposite direction (convex), i.e. toward…
University of Oxford（牛津大学）；University of Notre Dame（圣母大学）
Deep learning (DL) has recently emerged as a candidate for closure modeling of large-eddy simulation (LES) of turbulent flows. High-fidelity training data is typically limited: it is computationally costly (or even impossible) to numerically generate at high Reynolds numbers, while experimental data is also expensive to produce and might only include sparse/aggregate flow measurements. Thus, only a relatively small number of geometries and physical regimes will realistically be included in any training dataset. Limited data can lead to overfitting and therefore inaccurate predictions for geome…
In the current study, model expressions for fifth-order velocity moments obtained from the truncated Gram-Charlier series expansions model for a turbulent flow field probability density function are validated using data from direct numerical simulation (DNS) of a planar turbulent flow in a strained channel. Simplicity of the model expressions, the lack of unknown coefficients, and their applicability to non-Gaussian turbulent flows make this approach attractive to use for closing turbulent models based on the Reynolds-averaged Navier-Stokes equations. The study confirms validity of the model e…
We present results from direct numerical simulations (DNS) of oscillatory pipe flow at several dimensionless frequencies Wo = {6.5, 13, 26} and one fixed shear Reynolds number Re_τ = 1440. Starting from a fully-developed turbulent velocity field at that Re_τ, the oscillatory flow either relaminarises or reaches a conditionally turbulent or strongly asymmetric state depending on Wo. The numerical method is validated by demonstrating excellent agreement of our DNS results with experimental data and analytical predictions from literature for the limiting cases of non-oscillating but turbulent and…
We propose a predictive model of the turbulent burning velocity $S_T$ in homogeneous isotropic turbulence (HIT) based on Lagrangian statistics of propagating surfaces. The propagating surfaces with a constant displacement speed are initially arranged on a plane, and they evolve in non-reacting HIT, behaving like the propagation of a planar premixed flame front. The universal constants in the model of $S_T$ characterize the enhancement of area growth of premixed flames by turbulence, and they are determined by Lagrangian statistics of propagating surfaces. The flame area is then modelled by the…
The direct numerical simulation (DNS) of the Taylor--Couette flow in the fully turbulent regime is described. The numerical method extends the work by Quadrio & Luchini (Eur. J. Mech. B / Fluids, v.21, pp.413--427, 2002), and is based on a parallel computer code which uses mixed spatial discretization (spectral schemes in the homogeneous directions, and fourth-order, compact explicit finite-difference schemes in the radial direction). A DNS is carried out to simulate for the first time the turbulent Taylor--Couette flow in the turbulent regime. Statistical quantities are computed to complement…
The evolution with Reynolds number of the dissipation function, normalized by wall variables, is investigated using direct numerical simulation (DNS) databases for incompressible turbulent Poiseuille flow in a plane channel, at friction Reynolds numbers up to Reτ= 2000. DNS results show that the mean part, directly dissipated by the mean flow, reaches a constant value while the turbulent part, converted into turbulent kinetic energy before being dissipated, follows a logarithmic law. This result shows that the logarithmic law of friction can be obtained without any assumption on the mean veloc…
Recent DNS results [Gerolymos G.A., Vallet I. : J. Fluid Mech. 807 (2016) 386--418] have provided data for the terms in the transport equations for the components of the dissipation tensor $\varepsilon_{ij}$ in low-Reynolds turbulent plane channel flow. The present paper extends the previous results by a detailed analysis of the behaviour of various mechanisms in the $\varepsilon_{ij}$-transport equations (production, diffusion, redistribution, destruction), with particular emphasis on the component-by-component comparison with the corresponding mechanisms in the transport equations for the Re…
