基于时程深度学习的流场特征分析方法

流场的特征分析是流体力学的重要研究方向,传统方法主要根据流动参数量值的大小加以判断,所得结果受选取的参数形式及主观阈值影响大.本文提出了基于流场时程深度学习的流动特征分析新方法,建立了基于自动编码的流动特征提取模型;采用无监督训练方法充分挖掘流动时程信号中的隐含特征,进行流场中复杂时序特征的低维表征与特征分析.开展了R...     

基于扩散模型的流场超分辨率重建方法

低分辨率的流场数据具有较少的信息量,不能充分捕捉流场中的细节演化过程.尤其对于湍流的随机脉动特征和小尺度涡旋细节特征更加难以获取,这限制了对流场演化机理进行深入研究.为了解决这一局限性,并从低分辨率流场中重建高分辨率数据,文章提出一种流场超分辨率重建的生成扩散模型FlowDiffusionNet.该模型以低分辨率流场数据输入作为约束条件,采用去噪分数匹配方法,来实现高分辨率流场数据的复现.FlowDiffusionNet在结构设计上充分考虑了流场数据的低频信息与高频空间特征,采用基于扩散过程的建模方法,用于重建高分辨率流场数据的残差.该模型结构便于实现迁移学习,可在不同程度的退化流场上应用.将该方法在多种经典流场数据集上进行测试,并与双三次插值(bicubic)、超分辨率生成对抗网络(SRGAN)、超分辨率卷积神经网络(SRCNN)等方法进行比较.结果表明,该方法在各种流场上的重建性能达到最佳水平,特别是对于包含小尺度涡结构的4倍下采样流场数据,客观评价指标SSIM达到0.999.     

一类面向高阶精度自适应流动计算的流场插值方法

网格自适应技术和高阶精度数值方法是提升计算流体力学复杂问题适应能力的有效技术途径. 将这两项技术结合需要解决一系列技术难题, 其中之一是高阶精度流场插值. 针对高阶精度自适应流动计算, 提出一类高精度流场插值方法, 实现将前一迭代步网格中流场数值解插值到当前迭代步网格中, 以延续前一迭代步中的计算状态. 为实现流场插值过程中物理量守恒, 该方法先计算新旧网格的重叠区域, 然后将物理量从重叠区域的旧网格中转移到新网格中. 为满足高阶精度要求, 先采用k-exact最小二乘方法对旧网格上的数值解进行重构, 获得描述物理量分布的高阶多项式, 随后采用高阶精度高斯数值积分实现物理量精确地转移到新网格单元上. 最后, 通过一个具有精确解的数值算例和一个高阶精度自适应流动计算算例验证了本文算法的有效性. 第一个算例结果表明当网格规模固定不变时, 插值精度阶数越高, 插值误差越小; 第二个算例显示本文方法可以有效缩短高精度自适应流动计算的迭代收敛时间.      

高精度边界格式的研究

利用有精确解的Ｒｉｎｇｌｅｂ流动，构造了对流场数值解精度进行检验的“Ｒｉｎｇｌｅｂ机器”．重点讨论了边界格式对流场数值解的影响及高精度边界格式的建立．计算表明，在场内应用二阶精度格式情况下，采用二阶精度的边界格式所得到的流场解精度将大大高于采用一阶边界格式所得到的精度．为提高流场解的精度，不仅需要高精度的边界格式，还必须注意边界格式与场内格式的匹配．计算还表明，采用特征线修正的方法能有效地提高边界处理的精度     

水面舰船尾流电导率信号分布规律的研究

在舰船尾流区与非尾流区之间存在显著的速度差和盐度(或密度)差,利用电导率探头可获得对应于这些差别的尾流电导率信号。在水槽中形成了与海洋环境类似的盐度分层流场,由双螺旋桨自航水面船模产生尾流,分别在盐度分层流场和非分层均匀流场中测量了尾流电导率信号的横向分布,并对盐度分层流场中水面舰船尾流的纵向速度分布进行了数值计算。结果表明:在盐度分层流场和非分层均匀流场中水面舰船尾流的电导率信号沿其横向近似呈高斯分布;尾流速度对其电导率信号的影响比盐度梯度的影响大得多;尾流的无量纲纵向速度亏损的数值计算结果与尾流的相对电导率信号横向分布的实测结果具有很好的一致性。     

三维复杂流场内能激发松弛的DSMC数值模拟

本文研究了带内能激发松弛和不同来流条件下的复杂外形高超声速飞行器过渡区三维流场特性.采用一种对飞行器物面网格与DSMC计算域网格分别标识的方法,通过判断模拟分子与表面碰撞来完成飞行器物面网格与DSMC计算域网格间的信息传递和信息存贮,对于复杂外形飞行器精确描述的物面网格不需做进一步处理,直接应用于不依赖于飞行器外形的DSMC计算的通用子程序中.采用L-B碰撞模型,分析了内能激发松弛的流场参数特性,从分子碰撞、分子与飞行器表面碰撞的角度,分析了不同来流条件下的流场参数的变化情况.     

斑海豹胡须涡激振动及其尾流循迹机理直接数值模拟

在视力和听力受到限制的条件下, 斑海豹可以通过具有特殊外形的胡须识别和追踪水中游鱼的尾迹. 从仿生学的角度, 对斑海豹胡须在尾迹流场中的振动特性和循迹机理进行研究, 有助于开发研制新型水下探测器. 本文采用嵌入式迭代浸入边界法,在雷诺数$Re=300$和折合流速$U_{\rm r}=6.0$的条件下, 对均匀流场和尾迹流场中斑海豹胡须模型的涡激振动进行直接数值模拟,研究了胡须模型的振动特征和尾涡结构, 并与具有相同等效直径的圆柱、椭圆柱的振动响应进行对比, 分析了不同结构外形对振动特性和尾涡结构的影响, 探讨了斑海豹胡须的感知能力和循迹机理. 模拟结果表明, 胡须结构具有显著的减阻、抑振作用, 在均匀流场中做微幅混沌运动, 这为斑海豹胡须感知提供了纯净的信号背景. 而在尾迹流场中,胡须结构的振动响应显著增强, 稳定且周期性好, 与其他柱体相比, 具有更高的信噪比和敏感度. 这揭示了斑海豹利用胡须振动识别和追踪水中游鱼尾迹的机理, 对于开发研制新型水下探测器具有重要借鉴意义.      

空间模式下可压缩气固两相混合层流动特性研究

采用Euler-Lagrange颗粒-轨道双向耦合模型对空间模式下含有固粒的二维可压缩混合层流场进行了研究。气相流场采用具有空间三阶精度的WNND格式进行数值模拟,固相方程采用单边三点差分离散。在考虑流场对固粒作用的同时,也计及了固粒对流场的反作用。在对流马赫数为0.5时,研究了颗粒相对密度、颗粒尺寸、Stokes数等因素对粒子运动和流场结构的影响。研究结果表明:在可压缩空间模式混合层中,固粒的Stokes数仍然是主要影响参数之一;相同Stokes数下不同密度的固粒对流场的干扰不同,轻固粒对流场的干扰明显要小。     

基于经验包络法的非线性系统参数识别

基于平均法导出了Van der Pol-Duffing类振子响应时程瞬时特性与系统参数间的函数关系,在此基础上提出一种高效的非线性系统参数识别方法。借助经验包络法EE（Empirical Envelope Method）求解了响应时程瞬时特性,验证了EE相对传统Hilbert变换HT（Hilbert Transform）方法在求解瞬时频率上的优势。通过数值算例验证了本文方法的识别精度。分析了信号长度、初始条件、采样频率和噪声比例四种因素对识别精度的影响。结果表明,线性参数识别精度不受上述因素影响,非线性刚度项系数识别精度受各因素影响较为明显;本文方法具有良好的抗噪声性能,即使系统响应受到10%的噪声污染,本文方法也具有很好的识别精度。     

基于经验包络法的非线性系统参数识别

基于平均法导出了Van der Pol-Duffing类振子响应时程瞬时特性与系统参数间的函数关系,在此基础上提出一种高效的非线性系统参数识别方法。借助经验包络法EE(Empirical Envelope Method)求解了响应时程瞬时特性,验证了EE相对传统Hilbert变换HT(Hilbert Transform)方法在求解瞬时频率上的优势。通过数值算例验证了本文方法的识别精度。分析了信号长度、初始条件、采样频率和噪声比例四种因素对识别精度的影响。结果表明,线性参数识别精度不受上述因素影响,非线性刚度项系数识别精度受各因素影响较为明显;本文方法具有良好的抗噪声性能,即使系统响应受到10%的噪声污染,本文方法也具有很好的识别精度。     

基于卷积神经网络的钝体尾迹识别研究

针对相同特征长度不同钝体的尾迹结构相近,肉眼难于分辨的问题,提出了一种基于卷积神经网络的钝体尾迹识别方法,并在竖直肥皂膜水洞的典型钝体模型尾迹实验中获得高准确率验证.实验平台由自建竖直肥皂膜实验装置、钝体模型(方柱、圆柱和三角柱)及图像采集系统组成,可基于光学干涉法实现对不同速度下钝体肥皂膜尾迹的高清持续拍摄.所建立卷...     

A hybrid immersed‐boundary and multi‐block lattice Boltzmann method for simulating fluid and moving‐boundaries interactions

A numerical method is developed for modelling the interactions between incompressible viscous fluid and moving boundaries. The principle of this method is introducing the immersed‐boundary concept in the framework of the lattice Boltzmann method, and improving the accuracy and efficiency of the simulation by refining the mesh near moving boundaries. Besides elastic boundary with a constitutive law, the method can also efficiently simulate solid moving‐boundary interacting with fluid by employing the direct forcing technique. The method is validated by the simulations of flow past a circular cylinder, two cylinders moving with respect to each other and flow around a hovering wing. The versatility of the method is demonstrated by the numerical studies including elastic filament flapping in the wake of a cylinder and fish‐like bodies swimming in quiescent fluid. Copyright © 2006 John Wiley & Sons, Ltd.     

Moving least squares reconstruction for sharp interface immersed boundary methods

We propose a new approach for reconstructing velocity boundary conditions in sharp-inerface immersed boundary (IB) methods based on the moving least squares (MLS) interpolation method. The MLS is employed to not only reconstruct velocity boundary conditions but also to calculate the pressure and velocity gradients in the vicinity of the immersed body, which are required in fluid structure interaction problems to obtain the force exerted by the fluid on the structure. To extend the method to arbitrarily complex geometries with nonconvex shaped boundaries, the visibility method is combined with the MLS method. The performance of the proposed curvilinear IB MLS (CURVIB-MLS) is demonstrated by systematic grid-refinement studies for two- and three-dimensional tests and compared with the standard CURVIB method employing standard wall-normal interpolation for reconstructing boundary conditions. The test problems are flow in a lid-driven cavity with a sphere, uniform flow over a sphere, flow on a NACA0018 airfoil at incidence, and vortex-induced vibration of an elastically-mounted cylinder. We show that the CURVIB-MLS formulation yields a method that is easier to implement in complex geometries and exhibits higher accuracy and rate of convergence relative to the standard CURVIB method. The MLS approach is also shown to dramatically improve the accuracy of calculating the pressure and viscous forces imparted by the flow on the body and improve the overall accuracy of FSI simulations. Finally, the CURVIB-MLS approach is able to qualitatively capture on relatively coarse grids important features of complex separated flows that the standard CURVIB method is able to capture only on finer grids.     

Improvement of mass source/sink for an immersed boundary method

An improved immersed boundary method using a mass source/sink as well as momentum forcing is developed for simulating flows over or inside complex geometries. The present method is based on the Navier–Stokes solver adopting the fractional step method and a staggered Cartesian grid system. A more accurate formulation of the mass source/sink is derived by considering mass conservation of the virtual cells in the fluid crossed by the immersed boundary. Two flow problems (the decaying vortex problem and uniform flow past a circular cylinder) are used to validate the proposed formulation. The results indicate that the accuracy near the immersed boundary is improved by introducing the accurate mass source/sink. Copyright © 2006 John Wiley & Sons, Ltd.     

一种模拟动边界绕流的锐利界面浸入边界法

When the structural wall moves over a fixed grid, the structure coverage will change, resulting in many dead and emerging elements. To avoid the influence of malformation and reconstruction of body-fitted grids on the calculation efficiency and accuracy of the fluid-structure interaction problems with coupled boundary movement on the fixed grid, an improved numerical method for describing the interaction between an immersed rigid body and fluid based on a sharp-interface is proposed. In this method, both the fluid and solid are regarded as pure fluid domains in the whole computational domain, and the solid boundary is divided into several Lagrangian grid points. The flow parameter or velocity is reconstructed by interpolation at the interface element, which is then directly used as the boundary condition of the flow field, thus reflecting the influence of the wall boundary conditions. The method constructs the calculation structure of “virtual point, force point and vertical foot point”, and the velocity of the virtual point is obtained by bilinear interpolation. Then, the velocity of the force point is calculated by forcing the solid boundary to meet the no-slip condition, and the equations of the coupling system based on the immersion boundary method are finally solved to realize the numerical simulation of the flow with a complex moving boundary. The numerical program for this immersed boundary method is established using C++, then the accuracy and reliability of the proposed method are validated by comparison with the literature and experimental results of the basic numerical example of flow around a cylinder. Furthermore, the effects of the structural shape and the angle of attack on the trailing vortex structure, the vortex shedding frequency, and the lift/ coefficient characteristics of the flow around the elliptical cylinder have been analyzed. The anti-symmetric S-type, “P+S” Ⅰ-type and “P+S” Ⅱ-type trailing vortex shedding modes, as well as the variation laws of the vortex structure size, vortex shedding frequency and lift-drag coefficients ratio with axis ratio and angle of attack, are captured. The critical angle of attack (25°) corresponding to the maximum lift-drag ratio is determined as 25°.     

A Comparison of Large Scale Extraction Methods in the Study of Annular Wake Flow

Large scale periodic structures can exist in selected flow fields. Examples are the Precessing Vortex Core in swirling flows, vortex shedding behind a cylinder or the wake of an annular jet. A number of techniques are available to extract these large scales from the turbulent fluctuations in the flow field. In this paper, an analysis is made of three such methods: Eulerian Time Filtering (ETF), Proper Orthogonal Decomposition (POD) and non-linear least-squares regression POD (NLSR-POD). The accuracy of the three different extraction methods is compared quantitatively with phase averaged data of an annular wake flow. This flow was chosen as a test case, since it is widely used in industrial applications, such as for example bluff-body burners. It was shown that all three methods were able to reconstruct the flow field with reasonable accuracy. These techniques are therefore applicable to a number of periodic flows. The big advantage of these extraction methods is that they require 20 times less experimental data compared to phase averaging. All three methods require more or less the same computational time and since the computational time is a few orders of magnitude lower than the measurement time, application of these techniques results in a very large reduction in the total time to obtain the flow field characteristics. This results in a significant reduction of time in the design process of such flows.     

Frequency effects on lift and drag for flow past an oscillating cylinder

A transversely oscillating cylinder in a uniform flow is modeled to investigate frequency effects of flow-induced wake on lift and drag of the cylinder. Specifically, verified unsteady fluid dynamic simulations using an immersed-boundary method in a fixed Cartesian grid predict the flow structure around the cylinder and reveal how the integration of surface pressure and shear distributions provides lift and drag on the oscillating cylinder. In this study, frequency ranges to be considered are both near and away from the natural frequency of wake vortex shedding. Subsequently, the effects of frequency lock-in, superposition and demultiplication on lift and drag are discussed based on the spectral analysis of time histories of lift and drag.     

Numerical modeling of wave interactions with coastal structures by a constrained interpolation profile/immersed boundary method

A high‐order difference method based multiphase model is proposed to simulate nonlinear interactions between water wave and submerged coastal structures. The model is based on the Navier–Stokes equations using a constrained interpolation profile (CIP) method for the flow solver, and employs an immersed boundary method (IBM) for the treatment of wave–structure interactions. A more accurate interface capturing scheme, the volume of fluid/weighed line interface calculation (VOF/WLIC) scheme, is adopted as the interface capturing method. A series of computations are performed to verify the application of the model for simulations of fluid interaction with various structures. These problems include flow over a fixed cylinder, water entry of a circular cylinder and solitary waves passing various submerged coastal structures. Computations are compared with the available analytical, experimental and other numerical results and good agreement is obtained. The results of this study demonstrate the accuracy and applications of the proposed model to simulate the nonlinear flow phenomena and capture the complex free surface flow. Copyright © 2015 John Wiley & Sons, Ltd.     

Fully resolved simulations of viscoelastic suspensions by an efficient immersed boundary-lattice Boltzmann method

An efficient immersed boundary-lattice Boltzmann method (IB-LBM) is proposed for fully resolved simulations of suspended solid particles in viscoelastic flows. Stress LBM based on Giesekus and Oldroyd-B constitutive equation are used to model the viscoelastic stress tensor. A boundary thickening-based direct forcing IB method is adopted to solve the particle–fluid interactions with high accuracy for non-slip boundary conditions. A universal law is proposed to determine the diffusivity constant in a viscoelastic LBM model to balance the numerical accuracy and stability over a wide range of computational parameters. An asynchronous calculation strategy is adopted to further improve the computing efficiency. The method was firstly applicated to the simulation of sedimentation of a single particle and a pair of particles after good validations in cases of the flow past a fixed cylinder and particle migration in a Couette flow against FEM and FVM methods. The determination of the asynchronous calculation strategy and the effect of viscoelastic stress distribution on the settling behaviors of one and two particles are revealed. Subsequently, 504 particles settling in a closed cavity was simulated and the phenomenon that the viscoelastic stress stabilizing the Rayleigh–Taylor instabilities was observed. At last, simulations of a dense flow involving 11001 particles, the largest number of particles to date, were performed to investigate the instability behavior induced by elastic effect under hydrodynamic interactions in a viscoelastic fluid. The elasticity-induced ordering of the particle structures and fluid bubble structures in this dense flow is revealed for the first time. These simulations demonstrate the capability and prospects of the present method for aid in understanding the complex behaviors of viscoelastic particle suspensions.