基于组合神经网络的雷诺平均湍流模型多次修正方法

求解雷诺平均(Reynolds-averaged Navier-Stokes, RANS)方程依然是工程应用中有效且实用的方法, 但对雷诺应力建模的不确定性会导致该方法的预测精度具有很大差异. 随着人工智能的发展, 湍流闭合模型结合机器学习元素的数据驱动方法被认为是提高RANS模型预测性能的有效手段, 然而这种数据驱动方法的稳定性和预测精度仍有待进一步提高. 本文通过构建一个全连接神经网络对RANS方程中的涡黏系数进行预测以实现雷诺应力的隐式求解,该神经网络记作涡黏系数神经网络(eddy viscosity neural network, EVNN). 此外, 也使用张量基神经网络(tensor basis neural network, TBNN)预测未封闭量与解析量之间的高阶涡黏关系, 并利用基张量保证伽利略不变性. 最后, 采用多次修正的策略实现修正模型对流场预测的精度闭环. 上述方法使用大涡模拟(large eddy simulation, LES)方法产生的高保真数据, 以及RANS模拟获得的基线数据对由EVNN和TBNN组合的神经网络进行训练, 然后用训练好的模型预测新的RANS模拟的流场. 通过与高保真LES结果进行对比, 结果表明, 相比于原始RANS模型, 修正模型对后验速度场、下壁面平均压力系数和摩擦力系数的预测精度均有较大提升. 可以发现对雷诺应力线性部分的隐式处理可以增强数值求解的稳定性, 对雷诺应力非线性部分的修正可以提升模型对流场各向异性特征预测的性能, 并且多次修正后的模型表现出更高的预测精度. 因此, 该算法在数据驱动湍流建模和工程应用中具有很大的应用潜力.      

用基于M-SST模型的DES数值模拟喷流流场

脱体涡数值模拟方法(dettached eddy simulation,DES)是把雷诺平均Navier-Stokes方程(RANS)方法及大涡模拟方法(LES)结合起来模拟有脱体涡的湍流流场的数值模拟方法,其主要思想是在物面附近解雷诺平均Navier-Stokes方程、在其他区域采用Smagorinski大涡模拟方法。本文在剪切应力传输(SST)湍流模型的基础上用DES及混合非结构网格数值模拟具有横向喷流的湍流流场,算法采用Osher逆风格式,利用该套程序(包括网格生成及算法),对导弹在不同马赫数下的喷流流场进行了数值模拟,并与同时开展的实验研究的结果进行了对比,结果表明用该方法处理这类问题是较准确的。     

高阶矩湍流模型研究进展及挑战

高阶矩模型是湍流模式理论研究中的难点和前沿.自周培源先生首次建立一般湍流的雷诺应力输运方程起,为了更精确的预测复杂流动,人们从未间断过对高阶矩模型的研究.尤其进入新世纪以来,随着计算机硬件水平的飞跃和高精度数值算法的突破,湍流模拟方法正由RANS向LES转变.而无论对于RANS框架、LES框架还是两者混合,高阶矩模式都...     

逆压梯度边界层未分离情况下的湍流模式适用性研究

边界层逆压梯度作用下的流动是许多工程中的一个基础问题,由于逆压梯度作用,流动形态复杂,使得数值模拟有很大的难度。基于雷诺平均纳维-斯托克斯RANS(Reynolds Averaged Navier-Stokes)方程对二维平板逆压梯度边界层作数值计算研究,选取6种代表性的湍流模式,得到局部摩擦系数的数值解,与实验值比较,发现k-ω模式具有很好的精度。基于该湍流模式,给出了湍动能分布,该结果有助于认识逆压梯度边界层流动的复杂特征。     

逆压梯度边界层未分离情况下的湍流模式适用性研究Turbulence model applicability for boundary layer flow with adverse pressure gradient in attached case

边界层逆压梯度作用下的流动是许多工程中的一个基础问题,由于逆压梯度作用,流动形态复杂,使得数值模拟有很大的难度。基于雷诺平均纳维‐斯托克斯RANS（Reynolds Averaged Navier‐Stokes）方程对二维平板逆压梯度边界层作数值计算研究,选取6种代表性的湍流模式,得到局部摩擦系数的数值解,与实验值比较,发现k‐ω模式具有很好的精度。基于该湍流模式,给出了湍动能分布,该结果有助于认识逆压梯度边界层流动的复杂特征。     

LES/RANS方法混合函数特性研究

将两方程k-ω SST湍流模型和Sagaut的混合尺度亚格子模型通过一个混合函数相结合, 构造出一种混合大涡/雷诺平均N-S方程模拟方法(hybird large eddy simulation/reynolds-averaged navier-stokes, Hybrid LES/RANS), 采用这种混合模拟方法结合5阶WENO格式对Ma=2.8平板湍流边界层进行了数值模拟, 并在计算区域上游入口处采用“回收/调节”方法生成湍流脉动边界条件, 通过考查RANS区域向LES区域的过渡参数及网格分辨率对这种混合模拟方法进行了评价. 计算结果表明: 该文采用的混合模拟方法可以捕捉到湍流边界层中的大尺度结构且入口边界层平均参数不会发生漂移, 混合函数应当将RANS区域和LES区域的过渡点设置在对数律层和尾迹律层的交界处, 而过渡应当迅速以获得正确的雷诺剪切应力分布, 在该文采用的模型及数值方法的条件下, 流向及展向的网格小至与Escudier混合长相当时, 能够获得可以接受的脉动速度的单点-二阶统计值.     

旋转湍流中雷诺应力模型压力应变关联项的修正研究

利用张量的不变量理论,推导得出传统雷诺应力模型中压力应变关联项模型应用于旋转湍流模拟中的一些基本问题,即在纯旋转条件下,传统模型所描述的初始各向异性的湍流中雷诺应力张量演化规律是一个无衰减振荡过程,而快速畸变理论推导结果显示,其演化应是一个阻尼振荡衰减的过程。以衰减雷诺应力为目的,构造出包含旋转率张量高阶量的关联项。然后,结合变形率张量的高阶项,将修正模型扩展至椭圆形流线类型流动。最后,将修正模型应用于轴向旋转圆管内湍流流场的模拟,并将结果与实测结果进行了对比。     

基于卡门尺度和滤波方法的SST方程改进

基于滤波方法和卡门尺度对原始剪切应力输运(shear stress transport, SST)湍流模型进行了改进,提出了一种卡门尺度修正的滤波SST 方法. 湍流多尺度效应必须在分离流场模拟中给予反映,该方法减弱了雷诺平均(Reynolds averaged Navier-Stokes, RANS)方法时间平均特性对于流场脉动量的压迫作用,在流场中引入了大涡模拟(large eddy simulation, LES)方法的亚格子模型,形成一种新型的脱体涡模拟方法(detached eddy simulation,DES)方法;同时,为了降低原始DES方法在网格加密过程中产生网格诱发的雷诺应力损耗,利用卡门尺度对滤波因子进行修正. 平板边界层算例中,卡门尺度对于RANS方法的跟随性远远强于DES方法,在边界层内的速度型和RANS方法吻合很好,而DES方法在加密过程中速度型的鲁棒性较差,说明卡门尺度在有效地保护了边界层内使用RANS求解,降低速度型偏离对数率现象的产生;HGR-01翼型算例证明BY-SST方法可以有效的避免网格诱导分离现象的产生;证明BY-SST方法在分离流动中的精度高于DES方法.     

N-S方程数值研究翼型对微型扑翼气动特性的影响

首先基于嵌套网格发展了一套适用于三维扑翼研究的非定常雷诺平均Navier-Stokes(RANS)方程数值模拟方法.为了解决微型扑翼在低马赫数下的收敛问题,使用了预处理方法,湍流模型为BL模型.在该方法的基础上,保持状态参数和扑翼表面形状一定的情况下,分别研究了一系列不同厚度、不同弯度的翼型对于微型扑翼气动特性的影响....     

基于S-A湍流模型的Runge-Kutta有限元算法

采用一方程S-A模型(Spalart-Allmaras模型)封闭雷诺时均N-S方程(RANS方程)进行湍流数值计算,可以减少方程求解数量,节约计算时间。本文对其进行了有限元数值算法研究,首先通过沿流线坐标变换,得到无对流项RANS方程,并引入三阶Runge-Kutta法对其进行时间离散;然后利用沿流线的Taylor展开解决坐标变换带来的网格更新的困难;最后采用Galerkin法进行空间离散,得到湍流模型的有限元算法。基于方柱绕流和覆冰输电线绕流模型,与试验结果进行对比,验证了该算法的有效性,与一阶数值算法相比,该算法在精度和收敛性方面更具优势。     

Evaluation of Extended Weak-Equilibrium Conditions for Fully Developed Rotating Channel Flow

The weak-equilibrium condition, which is the basis in the development of algebraic Reynolds stress models is assessed in the prediction of fully developed turbulent channel flow under the influence of system rotation. The budget of the various terms of the exact transport equation for the anisotropy tensor is evaluated by using a DNS database. Two diffusive transport constraints are evaluated by using the DNS data. The results show that neither of them can hold for the near-wall region. An asymptotic analysis of the near-wall behavior is performed and an alternative form of the diffusive transport constraint is proposed. The analysis shows that the proposed alternative diffusive transport constraint has the potential to improve the predictive ability of the resultant ARSM.     

Anisotropy of the Reynolds stresses in a turbulent boundary layer on a rough wall

The measured anisotropy invariants of the Reynolds stress tensor in a self-preserving rough wall turbulent boundary layer indicate that the anisotropy is significantly smaller than in a smooth wall layer.RAA is grateful to Dr. P. Spalart for the DNS data. The support of the Australian Research Council is acknowledged.     

A reformulated synthetic turbulence generation method for a zonal RANS–LES method and its application to zero-pressure gradient boundary layers

A synthetic turbulence generation (STG) method for subsonic and supersonic flows at low and moderate Reynolds numbers to provide inflow distributions of zonal Reynolds-averaged Navier–Stokes (RANS) – large-eddy simulation (LES) methods is presented. The STG method splits the LES inflow region into three planes where a local velocity signal is decomposed from the turbulent flow properties of the upstream RANS solution. Based on the wall-normal position and the local flow Reynolds number, specific length and velocity scales with different vorticity content are imposed at the inlet plane of the boundary layer. The quality of the STG method for incompressible and compressible zero-pressure gradient boundary layers is shown by comparing the zonal RANS–LES data with pure LES, pure RANS, and direct numerical simulation (DNS) solutions. The distributions of the time and spanwise wall-shear stress, Reynolds stress distributions, and two point correlations of the zonal RANS–LES simulations are smooth in the transition region and in good agreement with the pure LES and reference DNS findings. The STG approach reduces the RANS-to-LES transition length to less than four boundary-layer thicknesses.     

Reynolds Stress Budgets in Couette and Boundary Layer Flows

Reynolds stress budgets for both Couette and boundary layer flows are evaluated and presented. Data are taken from direct numerical simulations of rotating and non-rotating plane turbulent Couette flow and turbulent boundary layer with and without adverse pressure gradient. Comparison of the total shear stress for the two types of flows suggests that the Couette case may be regarded as the high Reynolds number limit for the boundary layer flow close to the wall. The limit values of turbulence statistics close to the wall for the boundary layer for increasing Reynolds number approach the corresponding Couette flow values. The direction of rotation is chosen so that it has a stabilizing effect, whereas the adverse pressure gradient is destabilizing. The pressure-strain rate tensor in the Couette flow case is presented for a split into slow, rapid and Stokes terms. Most of the influence from rotation is located to the region close to the wall, and both the slow and rapid parts are affected. The anisotropy for the boundary layer decreases for higher Reynolds number, reflecting the larger separation of scales, and becomes close to that for Couette flow. The adverse pressure gradient has a strong weakening effect on the anisotropy. All of the data presented here are available on the web [36]. This revised version was published online in July 2006 with corrections to the Cover Date.     

Hybrid LES-RANS: An approach to make LES applicable at high Reynolds number

The main bottle neck for using large eddy simulations (LES) at high Reynolds number is the requirement of very fine meshes near walls. Hybrid LES-Reynolds-averaged Navier-Stokes (RANS) was invented to get rid of this limitation. In this method, unsteady RANS (URANS) is used near walls and away from walls LES is used. The matching between URANS and LES takes place in the inner log-region. In the present paper, a method to improve standard LES-RANS is evaluated. The improvement consists of adding instantaneous turbulent fluctuations (forcing conditions) at the matching plane in order to provide the equations in the LES region with relevant turbulent structures. The fluctuations are taken from a DNS of a generic boundary layer. Simulations of fully developed channel flow and plane asymmetric diffuser flow are presented. Hybrid LES-RANS is used both with and without forcing conditions.     

The development of algebraic stress models using a novel evolutionary algorithm

This work presents developments to a novel evolutionary framework that symbolically regresses algebraic forms of the Reynolds stress anisotropy tensor. This work contributes to the growing trend in machine-learning for modelling physical phenomena. Our framework is shown to be computational inexpensive and produce accurate and robust models that are tangible mathematical expressions. This transparency in the result allows us to diagnose issues with the regressed formulae and appropriately make amendments, as we further understand the regression tools. Such models are created using hybrid RANS/LES flow field data and a passive solving of the RANS transport equations to obtain the modelled time scale. This process shows that models can be regressed from a qualitatively correct flow field and fully resolved DNS is not necessarily required. Models are trained and tested using rectangular ducts, an example flow genus that linear RANS models even qualitatively fail to predict correctly. A priori and a posteriori testing of the new models show that the framework is a viable methodology for RANS closure development. This a posteriori agenda includes testing on an asymmetric diffuser, for which the new models vastly outperform the baseline linear model. Therefore this study presents one of the most rigorous and complete CFD validation of machine learnt turbulent stress models to date.     

Numerical simulation on evolution of subharmonic low-speed streaks in minimal channel turbulent flow

The evolution of low-speed streaks in the turbulent boundary layer of the minimum channel flow unit at a low Reynolds number is simulated by the direct numer- ical simulation （DNS） based on the standard Fourier-Chebyshev spectral method. The subharmonic sinuous （SS） mode for two spanwise-aligned low-speed streaks is excited by imposing the initial perturbations. The possibilities and the physical realities of the turbulent sustaining in the minimal channel unit are examined. Based on such a flow field environment, the evolution of the low-speed streaks during a cycle of turbulent sus- taining, including lift-up, oscillation, and breakdown, is investigated. The development of streamwise vortices and the dynamics of vortex structures are examined. The results show that the vortices generated from the same streak are staggered along the streamwise direction, while the vortices induced by different streaks tilt toward the normal direction due to the mutual induction effect. It is the spatial variations of the streamwise vortices that cause the lift-up of the streaks. By resolving the transport dynamics of enstrophy, the strength of the vortices is found to continuously grow in the logarithmic layer through the vortex stretching mechanism during the evolution of streaks. The enhancement of the vortices contributes to the spanwise oscillation and the following breakdown of the low-speed streaks.     

Critical transition Reynolds number for plane channel flow

The determination of the critical transition Reynolds number is of practical importance for some engineering problems. However, it is not available with the current theoretical method, and has to rely on experiments. For supersonic/hypersonic boundary layer flows, the experimental method for determination is not feasible either. Therefore,in this paper, a numerical method for the determination of the critical transition Reynolds number for an incompressible plane channel flow is proposed. It is basically aimed to test the feasibility of the method. The proposed method is extended to determine the critical Reynolds number of the supersonic/hypersonic boundary layer flow in the subsequent papers.     

A Priori Assessment of Prediction Confidence for Data-Driven Turbulence Modeling

Although Reynolds-Averaged Navier–Stokes (RANS) equations are still the dominant tool for engineering design and analysis applications involving turbulent flows, standard RANS models are known to be unreliable in many flows of engineering relevance, including flows with separation, strong pressure gradients or mean flow curvature. With increasing amounts of 3-dimensional experimental data and high fidelity simulation data from Large Eddy Simulation (LES) and Direct Numerical Simulation (DNS), data-driven turbulence modeling has become a promising approach to increase the predictive capability of RANS simulations. However, the prediction performance of data-driven models inevitably depends on the choices of training flows. This work aims to identify a quantitative measure for a priori estimation of prediction confidence in data-driven turbulence modeling. This measure represents the distance in feature space between the training flows and the flow to be predicted. Specifically, the Mahalanobis distance and the kernel density estimation (KDE) technique are used as metrics to quantify the distance between flow data sets in feature space. To examine the relationship between these two extrapolation metrics and the machine learning model prediction performance, the flow over periodic hills at Re = 10595 is used as test set and seven flows with different configurations are individually used as training sets. The results show that the prediction error of the Reynolds stress anisotropy is positively correlated with Mahalanobis distance and KDE distance, demonstrating that both extrapolation metrics can be used to estimate the prediction confidence a priori. A quantitative comparison using correlation coefficients shows that the Mahalanobis distance is less accurate in estimating the prediction confidence than KDE distance. The extrapolation metrics introduced in this work and the corresponding analysis provide an approach to aid in the choice of data source and to assess the prediction performance for data-driven turbulence modeling.