The precise simulation of turbulent flows is of immense importance in a variety of scientific and engineering fields, including climate science, freshwater science, and the development of energy-efficient manufacturing processes. Within the realm of turbulent flow simulation, direct numerical simulation (DNS) is widely considered to be the most reliable approach, but it is prohibitively expensive for long-term simulation at fine spatial scales. Given the pressing need for efficient simulation, there is an increasing interest in building machine learning models for turbulence, either by reconstructing DNS from alternative low-fidelity simulations or by predicting DNS based on the patterns learned from historical data. However, standard machine learning techniques remain limited in capturing complex spatio-temporal characteristics of turbulent flows, resulting in limited performance and generalizability. This paper presents a novel physics-enhanced neural operator (PENO) that incorporates physical knowledge of partial differential equations (PDEs) to accurately model flow dynamics. The model is further refined by a self-augmentation mechanism to reduce the accumulated error in long-term simulations. The proposed method is evaluated through its performance on two distinct sets of 3D turbulent flow data, showcasing the model's capability to reconstruct high-resolution DNS data, maintain the inherent physical properties of flow transport, and generate flow simulations across various resolutions. Additionally, experimental results on multiple 2D vorticity flow series, generated by different PDEs, highlight the transferability and generalizability of the proposed method. This confirms its applicability to a wide range of real-world scenarios in which extensive simulations are needed under diverse settings.
Simulating Three-dimensional Turbulence with Physics-informed Neural Networks
Institution for Foundation of Data Science, Yale University（数据科学基础研究所，耶鲁大学）；Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania（机械工程与应用力学系，宾夕法尼亚大学）；Sibley School of Mechanical and Aerospace Engineering, Cornell University（机械与航空航天工程学院，康奈尔大学）；Advanced Computing, Mathematics and Data Division, Pacific Northwest National Laboratory（先进计算、数学和数据部门，太平洋西北国家实验室）
While the popularity of physics-informed neural networks (PINNs) is steadily rising, to this date PINNs have not been successful in simulating dynamical systems whose solution exhibits multi-scale, chaotic or turbulent behavior. In this work we attribute this shortcoming to the inability of existing PINNs formulations to respect the spatio-temporal causal structure that is inherent to the evolution of physical systems. We argue that this is a fundamental limitation and a key source of error that can ultimately steer PINN models to converge towards erroneous solutions. We address this pathology…
One of the most enigmatic science question concerning inertial particle transport by a turbulent boundary layer flow is the value of the turbulent Schmidt number as the ratio of particle diffusivity and turbulent eddy viscosity. Using direct acoustic measurement of turbulent particle flux profile, and two-phase flow turbulence-resolving numerical simulation, it is demonstrated that turbulent dispersion of particles is reduced rather than enhanced when predicted with existing literature model. The explanation lies in the misleading assumption of settling velocity in quiescent water to estimate …
With the aim of assessing internal wave-driven mixing in the ocean, we develop a new technique for direct numerical simulations of stratified turbulence. Since the spatial scale of oceanic internal gravity waves is typically much larger than that of turbulence, fully incorporating both in a model would require a high computational cost, and is therefore out of our scope. Alternatively, we cut out a small domain periodically distorted by an unresolved large-scale internal wave and locally simulate the energy cascade to the smallest scales. In this model, even though the Froude number of the out…
This paper extends the methodology to use physics-informed enhanced super-resolution generative adversarial networks (PIESRGANs) for LES subfilter modeling in turbulent flows with finite-rate chemistry and shows a successful application to a non-premixed temporal jet case. This is an important topic considering the need for more efficient and carbon-neutral energy devices to fight the climate change. Multiple a priori and a posteriori results are presented and discussed. As part of this, the impact of the underlying mesh on the prediction quality is emphasized, and a multi-mesh approach is dev…
Jülich Supercomputing Centre, Forschungszentrum Jülich GmbH（于利希研究中心于利希超级计算中心）；RWTH Aachen University（亚琛工业大学）
Models for finite-rate-chemistry in underresolved flows still pose one of the main challenges for predictive simulations of complex configurations. The problem gets even more challenging if turbulence is involved. This work advances the recently developed PIESRGAN modeling approach to turbulent premixed combustion. For that, the physical information processed by the network and considered in the loss function are adjusted, the training process is smoothed, and especially effects from density changes are considered. The resulting model provides good results for a priori and a posteriori tests o…
Significant improvements in the computational performance of the lattice-Boltzmann (LB) model, coded in FORTRAN90, were achieved through application of enhancement techniques. Applied techniques include optimization of array memory layouts, data structure simplification, random number generation outside the simulation thread(s), code parallelization via OpenMP, and intra- and inter-timestep task pipelining. Effectiveness of these optimization techniques was measured on three benchmark problems: (i) transient flow of multiple particles in a Newtonian fluid in a heterogeneous fractured porous do…
Enhanced heat transport by turbulent two-phase Rayleigh-Bénard convection
本文报道了乙烷（C $_2$ H $_6$）样品的湍流传热测量结果，当外加温差$ΔT$跨越液-气共存曲线$T_ϕ(P)$时，从下面加热乙烷样品。当样品顶部温度$T_t$降低到$T_ϕ$以下时，液滴发生凝结，蒸发潜热$H$提供了额外的传热机制，有效电导率$λ_{eff}$随$T_t$的降低而线性增加，并达到最大值$λ_{eff}^*$，比单相$λ_{eff}$大一个数量级。当$P$接近临界压力时，即使$H$消失，$λ_{eff}^*$也急剧增加。我们把这种现象归因于接近临界点时液滴成核速率的提高。
In this study, a deep learning-based approach is applied with the aim of reconstructing high-resolution turbulent flow fields using minimal flow fields data. A multi-scale enhanced super-resolution generative adversarial network with a physics-based loss function is introduced as a model to reconstruct the high-resolution flow fields. The model capability to reconstruct high-resolution laminar flows is examined using data of laminar flow around a square cylinder. The results reveal that the model can accurately reproduce the high-resolution flow fields even when limited spatial information is …
We present measurements of the azimuthal orientation θ(t) and thermal amplitude δ(t) of the large-scale circulation (LSC) of turbulent rotating convection within an unprecedented large Rossby number range 1<Ro<314. Results of θ(t) reveal persistent rotation of the LSC flow in the retrograde direction over the entire Ro range. The rotation speed ratio remains a constant of 0.13(${\pm}$0.01) for 10<Ro<70, but starts to increase with increasing Ro when Ro>70. We identify the mechanism through which the mean retrograde rotation speed can be enhanced by stochastic cessations in the presence of weak…
A data-driven, model-free framework is introduced for calculating Reduced-Order Models (ROMs) capable of accurately predicting time-mean responses to external forcings, or forcings needed for specified responses, e.g., for control, in fully turbulent flows. The framework is based on using the Fluctuation-Dissipation Theorem (FDT) in the space of a limited number of modes obtained from Dynamic Mode Decomposition (DMD). Using the DMD modes as the basis functions, rather than the commonly used Proper Orthogonal Decomposition (POD) modes, resolves a previously identified problem in applying FDT to…
Reynolds-averaged Navier-Stokes (RANS) equations are presently one of the most popular models for simulating turbulence. Performing RANS simulation requires additional modeling for the anisotropic Reynolds stress tensor, but traditional Reynolds stress closure models lead to only partially reliable predictions. Recently, data-driven turbulence models for the Reynolds anisotropy tensor involving novel machine learning techniques have garnered considerable attention and have been rapidly developed. Focusing on modeling the Reynolds stress closure for the specific case of turbulent channel flow, …
Reconstruction of turbulent flow based on data assimilation methods is of significant importance for improving the estimation of flow characteristics by incorporating limited observations. Existing works mainly focus on using only one observation data source, e.g., velocity, wall pressure, lift or drag force, to reconstruct the flow. In practical applications observations are disparate data sources that often vary in dimension and quality. Simultaneously incorporating these disparate data is worth investigation to improve the flow reconstruction. In this work, we investigate the disparate data…
It is known that turbulent energy is rapidly transferred in the direction of the rotation axis in a rotating system, in comparison with the non-rotating case. In this study, this phenomenon is investigated as a problem of energy diffusion expressed by the Reynolds averaged Navier-Stokes (RANS) model. The conventional gradient-diffusion approximation for the turbulent energy flux cannot account for the enhanced energy transport observed in rotating inhomogeneous turbulence. In order to adequately describe the phenomenon, we propose a new model for the energy flux due to the pressure associated …
Turbulent-laminar intermittency, typically in the form of bands and spots, is a ubiquitous feature of the route to turbulence in wall-bounded shear flows. Here we study the idealised shear between stress-free boundaries driven by a sinusoidal body force and demonstrate quantitative agreement between turbulence in this flow and that found in the interior of plane Couette flow -- the region excluding the boundary layers. Exploiting the absence of boundary layers, we construct a model flow that uses only four Fourier modes in the shear direction and yet robustly captures the range of spatiotempor…
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 …
The LBM is combined with the Volume Penalization (VP LBM) approach to simulate flows in the presence of obstacles. The single relaxation time LBM and the multiple relaxation time LBM are used. For cases where the fluid motion is enhanced by moving boundaries, the comparison of our results with analytical results is satisfactory. The flow around a cylinder at a relatively high Reynolds number is also calculated. For that case,the VP LBM method was parallelized. A comparison of our results with those found in the literature, demonstrates the reliability of the VP LBM to treat complex flows.
Condensation is an important aspect of many flow applications due to the universal presence of humidity in the air at ambient conditions. For direct numerical simulations of such flows, simulating the gas phase as a mixture characterized by temperature and humidity coupled by the release and absorption of latent heat has been shown to yield results consistent with multiphase direct numerical simulations at reduced costs. Modeling sessile droplets as quasi-static deformations based on the underlying condensation rates at the surface extends the single-phase approach to include the interaction b…
Parameter extension simulation (PES) as a mathematical method for simulating turbulent flows has been proposed in the study. It is defined as a calculation of the turbulent flow for the desired parameter values with the help of a reference solution. A typical PES calculation is composed of three consecutive steps: Set up the asymptotic relationship between the desired solution and the reference solution; Calculate the reference solution and the necessary asymptotic coefficients; Extend the reference solution to the desired parameter values. A controlled eddy simulation (CES) method has been de…
Deep neural network models have shown a great potential in accelerating the simulation of fluid dynamic systems. Once trained, these models can make inference within seconds, thus can be extremely efficient. However, they suffer from a generalization problem when the flow becomes chaotic and turbulent. One of the most important reasons is that, existing models lack the mechanism to handle the unique characteristic of turbulent flow: multi-scale flow structures are non-uniformly distributed and strongly nonequilibrium. In this work, we address this issue with the concept of visual attention: in…
Neural Multigrid Memory For Computational Fluid Dynamics
Ecole Polytechnique（巴黎综合理工学院）；University of Engineering and Technology, Vietnam National University, Hanoi（河内越南国立大学工程技术大学）；Microsoft（微软）；OpenAI；Google（谷歌）；Littelfuse（力特公司）；Halıcıoğlu Data Science Institute, University of California, San Diego（加州大学圣地亚哥分校哈利哲奥卢数据科学研究所）
This paper is focused on the fundamental mechanism(s) of viscoelastic turbulence that lead to polymer induced turbulent drag reduction phenomenon. A great challenge in this problem is the computation of viscoelastic turbulent flows, since the understanding of polymer physics is restricted to mechanical models. An effective state-of-the-art numerical method to solve the governing equation for polymers modelled as non-linear springs, without using any artificial assumptions as usual, was implemented here for the first time on a three-dimensional channel flow geometry. The capability of this algo…
Spatial clustering of identical particles falling through a turbulent flow enhances the collision rate between the falling particles, an important problem in aerosol science. We analyse this problem using perturbation theory in a dimensionless parameter, the so-called Kubo number. This allows us to derive an analytical theory quantifying the spatial clustering. We find that clustering of small particles in incompressible random velocity fields may be reduced or enhanced by the effect of gravity (depending on the Stokes number of the particles) and may be strongly anisotropic.
Significant progress has been made on the model development for simulating turbulent reacting flows. As a consequence, we are currently in a position where key-physical aspects of fairly complex combustion processes are well understood at a qualitative and -- in many cases -- also at a quantitative level. Examples are the prediction of temperature and major species, statistically stationary flames, gas-phase combustion, turbulent transport, and turbulence/flame coupling. However, current challenges lie in capturing transient processes and stability boundaries, minor species and emissions, mult…
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…
Turbulence in fluids is an ubiquitous phenomenon, characterized by spontaneous transition of a smooth, laminar flow to rapidly changing, chaotic dynamics. In 1883, Reynolds experimentally demonstrated that, in an initially laminar flow of water, turbulent motions emerge without any measurable external disturbance. To this day, turbulence remains a major unresolved phenomenon in fluid mechanics; in particular, there is a lack of a mathematical model where turbulent dynamics emerge naturally from a laminar flow. Recently, we proposed a new theory of turbulence in gases, according to which turbul…
This paper is concerned with the partitioned iterative formulation to simulate the fluid-structure interaction of a nonlinear multibody system in an incompressible turbulent flow. The proposed formulation relies on a three-dimensional (3D) incompressible turbulent flow solver, a nonlinear monolithic elastic structural solver for constrained flexible multibody system and the nonlinear iterative force correction scheme for coupling of the turbulent fluid-flexible multibody system with nonmatching interface meshes. While the fluid equations are discretized using a stabilized Petrov-Galerkin formu…
Spatial clustering and intermittency in the relative velocity of heavy particles of the same size settling in turbulent flows can be strongly affected by gravity. We present a model for the timescale of the fluid velocity gradient seen by particle pairs and propose an effective Kubo number based on this timescale to explain the mechanism of gravity-enhanced clustering. We explore the mechanisms of the gravity-induced reduction or enhancement of the intermittency in the particle radial relative velocity (RRV) at different Stokes numbers based on backward-in-time relative dispersion and preferen…
For rapidly rotating turbulent Rayleigh--Bénard convection in a slender cylindrical cell, experiments and direct numerical simulations reveal a boundary zonal flow (BZF) that replaces the classical large-scale circulation. The BZF is located near the vertical side wall and enables enhanced heat transport there. Although the azimuthal velocity of the BZF is cyclonic (in the rotating frame), the temperature is an anticyclonic traveling wave of mode one whose signature is a bimodal temperature distribution near the radial boundary. The BZF width is found to scale like $Ra^{1/4}Ek^{2/3}$ where the…
High-Fidelity Simulations of Two Miscible Fluids in Small Scale Turbulent Mixers Using a Variational Multiscale Finite Element Method
School of Mechanical Engineering, Purdue University（普渡大学机械工程学院）；Department of Agricultural and Biological Engineering, Purdue University（普渡大学农业与生物工程系）；Davidson School of Chemical Engineering (by courtesy), Purdue University（普渡大学戴维森化学学院（荣誉））
This study presents a deep learning-based framework to reconstruct high-resolution turbulent velocity fields from extremely low-resolution data at various Reynolds numbers using the concept of generative adversarial networks (GANs). A multiscale enhanced super-resolution generative adversarial network (MS-ESRGAN) is applied as a model to reconstruct the high-resolution velocity fields, and direct numerical simulation (DNS) data of turbulent channel flow with large longitudinal ribs at various Reynolds numbers are used to evaluate the performance of the model. The model is found to have the cap…
The aim of this Letter is to characterize the flow regimes of suspensions of finite-size rigid particles in a viscous fluid at finite inertia. We explore the system behavior as function of the particle volume fraction and the Reynolds number (the ratio of flow and particle inertia to viscous forces). Unlike single phase flows where a clear distinction exists between the laminar and the turbulent states, three different regimes can be identified in the presence of a particulate phase, with smooth transitions between them. At low volume fractions, the flow becomes turbulent when increasing the R…
