瞬态热传导问题的精细积分-双重互易边界元法

采用双重互易边界元法结合精细积分法求解二维含热源的瞬态热传导问题。针对边界积分方程中热源项和温度关于时间导数项引起的域积分,采用双重互易法处理,将域积分转换为边界积分。采用边界元法将边界积分方程离散后,得到关于时间的微分方程组,并利用精细积分法处理其中的指数型矩阵;对于微分方程组中由边界条件和热源项引起的非齐次项,采用解析的方法计算。为了比较精细积分-双重互易边界元法的计算效果,同时使用有限差分法计算温度对时间的导数项。通过数值算例验证了本文方法的有效性和精确性。计算结果表明:时间步长对于精细积分-双重互易边界元法的结果影响较小,而有限差分法对时间步长比较敏感且只在时间步长选取较小时有效;当选取较大时间步长时,精细积分-双重互易边界元法依然具有良好的计算精度。     

奇异边界法求解多连通域的瞬态热传导问题

提出一种基于奇异边界法结合双重互易法的数值模型来求解瞬态热传导问题。奇异边界法属于配点型边界无网格方法,相对于网格方法,其具有无需划分网格,只需边界配点的优势。运用差分格式来处理热传导方程中的时间变量,将原热传导方程化为非齐次修正Helmholtz方程。修正Helmholtz方程的解由齐次解和特解两部分组成,齐次解通过奇异边界法求出,特解由双重互易法求出,源项由径向基函数近似。通过数值算例检验了本文数值模型的精度及有效性;算例结果表明,该数值模型计算精度较高,误差基本都在1%以内,具有很好的稳定性,能有效地应用于求解多连通域的瞬态热传导问题。     

基于CUDA的有限元矩阵并行装配算法研究

构建航天飞行器的结构有限元模型是准确模拟飞行仿真、完成飞行器在轨飞行阶段结构故障监测和诊断的基础。采用细长体飞行器简化梁模型，提出新的基于CUDA（Compute Unified Device Architecture）的有限元单元刚度矩阵生成和总刚度矩阵组装算法。依据梁单元矩阵的对称性，结合GPU硬件架构提出并行生成算法并进行改进。为有效减少装配时间，在装配过程中采用着色算法，提出了基于GPU（Graphics Processing Unit）共享内存的非零项组装策略，通过在不同计算平台下算例对比，验证了新算法的快速性。数值算例表明，本文算法的求解效率较高，针对一定计算规模内的模型可满足快速计算与诊断的实时性要求。     

????????????????????????е????

提出了将杂交边界点法和双重互易法结合求解势问题的一种新的算法. 将势问题的解分为通解和特解两部分，通解使用 杂交边界点方法求解，特解则利用局部径向基函数近似. 该方法输入数据只是求解域上离散 的点，不需要额外的方程来计算域内物理量，后处理十分简便. 数值算例表明了该方法的稳 定性和有效性.     

一种基于块雅可比迭代的高阶FR格式隐式方法

最近, 基于非结构网格的高阶通量重构格式(flux reconstruction, FR)因其构造简单且通用性强而受到越来越多人的关注. 但将FR格式应用于大规模复杂流动的模拟时仍面临计算开销大、求解时间长等问题. 因此, 亟需发展与之相适应的高效隐式求解方法和并行计算技术. 本文提出了一种基于块Jacobi迭代的高阶FR格式求解定常二维欧拉方程的单GPU隐式时间推进方法. 由于直接求解FR格式空间和隐式时间离散后的全局线性方程组效率低下并且内存占用很大. 而通过块雅可比迭代的方式, 能够改变全局线性方程组左端矩阵的特征, 克服影响求解并行性的相邻单元依赖问题, 使得只需要存储和计算对角块矩阵. 最终将求解全局线性方程组转化为求解一系列局部单元线性方程组, 进而又可利用LU分解法在GPU上并行求解这些小型局部线性方程组. 通过二维无黏Bump流动和NACA0012无黏绕流两个数值实验表明, 该隐式方法计算收敛所用的迭代步数和计算时间均远小于使用多重网格加速的显式Runge-Kutta格式, 且在计算效率方面至少有一个量级的提升.      

轴对称薄壁结构自由振动的边界元分析

采用双互易法分析薄壁轴对称结构自由振动的特征频率以及特征模态.首先,采用径向基函数插值域积分里的位移,利用双互易法将域积分转化为子午面边界的积分.然后,将边界物理量、基本解和特解展开为傅里叶级数,沿环向积分后得到的边界积分方程可用于轴对称结构带体积力问题和受非对称载荷的动力学分析,其积分域为轴对称结构子午面边界上的线积分,进一步降低了问题的维度和离散的难度.文章详细探讨了源点处于对称轴的特殊情况,根据基本解和特解的退化形式,针对无体积力和有体积力分别给出了处理奇异矩阵的方案.对于薄壁结构,采用双曲正弦变换处理近奇异积分有效提高积分精度.最后将双互易法和双曲正弦变化应用于薄壁轴对称结构带体积力的静力学和自由振动分析.数值结果表明,文章提出的处理奇异矩阵的方法能够有效处理源点处于对称轴的情况;当圆筒厚高比为10~(-3),边界元计算的特征频率的相对误差为10~(-3),且优于有限元的结果.     

块Lanczos法在大跨屋盖风致响应中的应用

提出块Lanczos向量直接叠加法分析大跨屋盖结构的风致响应.将适用于单个初始荷载向量的Lanczos方法推广到由多个初始荷载向量线性组合的一般动力荷载情况.由于不仅Lanczos块内的向量之间相互正交,而且Lanczos块之间也相互正交,屋盖结构的运动方程变换为块三对角矩阵的带状形式.这个途径不仅便于有效地应用时间域逐步积分解法,而且便于应用频率域解法.对脉动风荷载作用下的屋盖结构,多个初始荷载向量可应用本征正交分解得到.它将风压场分解为主坐标与协方差模态的组合,主坐标仅依赖时间而协方差模态仅依赖空间位置.根据圆拱顶屋盖模型风洞试验得到的风荷载,采用块Lanczos法、一般的Lanczos法以及传统的模态叠加法计算了屋盖竖向位移响应.对块Lanczos法的精度及效率作了讨论.     

基于射线穿透法的GPU并行阶梯型有限差分网格生成算法

三维大规模有限差分网格生成技术是三维有限差分计算的基础，网格生成效率是三维有限差分网格生成的研究热点。传统的阶梯型有限差分网格生成方法主要有射线穿透法和切片法。本文在传统串行射线穿透法的基础上，提出了基于GPU （graphic processing unit）并行计算技术的并行阶梯型有限差分网格生成算法。并行算法应用基于分批次的数据传输策略，使得算法能够处理的数据规模不依赖于GPU内存大小，平衡了数据传输效率和网格生成规模之间的关系。为了减少数据传输量，本文提出的并行算法可以在GPU线程内部相互独立的生成射线起点坐标，进一步提高了并行算法的执行效率和并行化程度。通过数值试验的对比可以看出，并行算法的执行效率远远高于传统射线穿透法。最后，通过有限差分计算实例可以证实并行算法能够满足复杂模型大规模数值模拟的需求。     

轴对称薄壁结构自由振动的边界元分析

采用双互易法分析薄壁轴对称结构自由振动的特征频率以及特征模态.首先,采用径向基函数插值域积分里的位移,利用双互易法将域积分转化为子午面边界的积分.然后,将边界物理量、基本解和特解展开为傅里叶级数,沿环向积分后得到的边界积分方程可用于轴对称结构带体积力问题和受非对称载荷的动力学分析,其积分域为轴对称结构子午面边界上的线积分,进一步降低了问题的维度和离散的难度.文章详细探讨了源点处于对称轴的特殊情况,根据基本解和特解的退化形式,针对无体积力和有体积力分别给出了处理奇异矩阵的方案.对于薄壁结构,采用双曲正弦变换处理近奇异积分有效提高积分精度.最后将双互易法和双曲正弦变化应用于薄壁轴对称结构带体积力的静力学和自由振动分析.数值结果表明,文章提出的处理奇异矩阵的方法能够有效处理源点处于对称轴的情况;当圆筒厚高比为$10^{-3}$,边界元计算的特征频率的相对误差为$10^{-3}$,且优于有限元的结果.      

基于Newton/Gauss-Seidel迭代的DGM隐式方法

在Newton迭代方法的基础上,对高阶精度间断Galerkin有限元方法(DGM)的时间隐式格式进行了研究. Newton迭代 法的优势在于收敛效率高效,并且定常和非定常问题能够统一处理,对于非定常问题无需引入双时间步策略. 为了避免大型矩阵的求逆,采用一步Gauss-Seidel迭代和Matrix-free技术消去残值Jacobi矩阵的上、下三角矩阵,从而只需计算和存储对角(块)矩阵. 对角(块)矩阵采用数值方法计算. 空间离散采用Taylor基,其优势在于对于任意形状的网格,基函数的形式是一致的,有利于在混合网格上推广. 利用该方法,数值模拟了Bump绕流和NACA0012翼型绕流. 计算结果表明,与显式的Runge-Kutta时间格式相比,隐式格式所需的迭代步数和CPU时间均在很大程度上得到减少,计算效率能够提高1～ 2个量级.     

DUAL RECIPROCITY HYBRID BOUNDARY NODE METHOD FOR THREE-DIMENSIONAL ELASTICITY WITH BODY FORCE

Combining Dual Reciprocity Method （DRM） with Hybrid Boundary Node Method （HBNM）, the Dual Reciprocity Hybrid Boundary Node Method （DRHBNM） is developed for three-dimensional linear elasticity problems with body force. This method can be used to solve the elasticity problems with body force without domain integral, which is inevitable by HBNM. To demonstrate the versatility and the fast convergence of this method, some numerical examples of 3-D elasticity problems with body forces are examined. The computational results show that the present method is effective and can be widely applied in solving practical engineering problems.     

Extension de la DRBEM à la propagation guidée en écoulement cisaillé

The present study aims to extend the Dual Reciprocity Boundary Element Method in order to solve acoustic wave propagation equations in the frequency domain for a parallel shear flow. The Linearized Euler Equations are written as a coupled pair of equations, which are second-order in terms of acoustic pressure and first-order in terms of normal acoustic velocity. Good agreement between numerical results and analytical solutions for a low Mach number shear flow (M<0.1) shows the interest of the method.     

RKPM形状函数的矩式显式表述及快速计算

给出了计算再生核质点法(RKPM)形状函数及其导数的矩式显式处理方法。其特点是在计算形状函数及其导数时不涉及矩阵的求逆或者线性方程组的求解，从而减少计算误差的产生并提高了计算速度。二维及三维形状函数计算算例表明该方法是提高RKPM计算效率的一种有效途径。     

基于改进降维法的可靠度分析

单变量维数缩减法可以高效、准确地进行结构响应矩的分析.与传统的一阶可靠度算法FORM(First Order Reliability Method),二阶可靠度算法SORM(Second Order Reliability Method)相比,该方法不需要响应的导数,也不需要迭代搜索最可能失效点.然而近期的研究发现,该...     

A fast 3D dual boundary element method based on hierarchical matrices

In this paper a fast solver for three-dimensional BEM and DBEM is developed. The technique is based on the use of hierarchical matrices for the representation of the collocation matrix and uses a preconditioned GMRES for the solution of the algebraic system of equations. The preconditioner is built exploiting the hierarchical arithmetic and taking full advantage of the hierarchical format. Special algorithms are developed to deal with crack problems within the context of DBEM. The structure of DBEM matrices has been efficiently exploited and it has been demonstrated that, since the cracks form only small parts of the whole structure, the use of hierarchical matrices can be particularly advantageous. Test examples presented show that, with the proposed technique, substantial increase in number of elements over the crack surfaces leads only to moderate increases in memory storage and solution time.     

大型结构特征值问题的Lanczos子结构并行算法

本文利用子结构和Lanczos方法,提出了大型结构固有频率与模态的并行解法。该方法在Lanczos方法的求解过程中,仅利用子结构刚度阵和质量阵并行进行凝聚,进而求得新的迭代矢量,最终求得三对角阵对应的特征值和特征向量。该算法在西安交通大学ELXSI-6400并行计算机上程序实现,计算结果表明能有效地节省计算时间和计算机的内存,为一种有效的大型工程结构动力问题的求解方法。     

The numerical method for analysing the transient temperature field and transient phase transformation of metal materials in heat treating

In this paper,the Kirchhoffs transformation is popularized to the nonlinear heat conduction problem which the heat conductivity can be expressd as a multinomial of temperature firstly,the boundary condition of heat conduction problem is determined by analytics.Secondly,the incubation peroid superposition and the linear combination law is employed to simulate the transient phasses transformation in the process of heat treatment of materials.That the begin time of phase transformation,the type of phase transformation and the amount of phase constitution is determined simply.Finally,the three-dimension Dual Reciprocity Boundary Element Method is usedto analysis the total process of various heat treatment of component,the results of numerical calculation of examples show that the method provided in this paper is effectivce.     

The fast multipole method on parallel clusters,multicore processors,and graphics processing units

In this article, we discuss how the fast multipole method (FMM) can be implemented on modern parallel computers, ranging from computer clusters to multicore processors and graphics cards (GPU). The FMM is a somewhat difficult application for parallel computing because of its tree structure and the fact that it requires many complex operations which are not regularly structured. Computational linear algebra with dense matrices for example allows many optimizations that leverage the regular computation pattern. FMM can be similarly optimized but we will see that the complexity of the optimization steps is greater. The discussion will start with a general presentation of FMMs. We briefly discuss parallel methods for the FMM, such as building the FMM tree in parallel, and reducing communication during the FMM procedure. Finally, we will focus on porting and optimizing the FMM on GPUs.     

Coupling the finite element method and molecular dynamics in the framework of the heterogeneous multiscale method for quasi-static isothermal problems

Multiscale models are designed to handle problems with different length scales and time scales in a suitable and efficient manner. Such problems include inelastic deformation or failure of materials. In particular, hierarchical multiscale methods are computationally powerful as no direct coupling between the scales is given. This paper proposes a hierarchical two-scale setting appropriate for isothermal quasi-static problems: a macroscale treated by continuum mechanics and the finite element method and a microscale modelled by a canonical ensemble of statistical mechanics solved with molecular dynamics. This model will be implemented into the framework of the heterogeneous multiscale method. The focus is laid on an efficient coupling of the macro- and micro-solvers. An iterative solution algorithm presents the macroscopic solver, which invokes for each iteration an atomistic computation. As the microscopic computation is considered to be very time consuming, two optimisation strategies are proposed. Firstly, the macroscopic solver is chosen to reduce the number of required iterations to a minimum. Secondly, the number of time steps used for the time average on the microscale will be increased with each iteration. As a result, the molecular dynamics cell will be allowed to reach its state of thermodynamic equilibrium only in the last macroscopic iteration step. In the preceding iteration steps, the molecular dynamics cell will reach a state close to equilibrium by using considerably fewer microscopic time steps. This adapted number of microsteps will result in an accelerated algorithm (aFE-MD-HMM) obtaining the same accuracy of results at significantly reduced computational cost. Numerical examples demonstrate the performance of the proposed scheme.