基于异构平台的高层钢筋混凝土结构非线性响应分析方法

发布时间：2018-03-13 21:58

  本文选题：高层钢筋混凝土结构　切入点：精细化有限元模型　出处：《哈尔滨工业大学》2015年博士论文　论文类型：学位论文

【摘要】：随着高层建筑结构向精细化建模发展,精细化的高层三维结构分析模型自由度数量巨大,计算工作量庞大,对软件分析平台的计算速度要求更高。然而目前国内外分析软件多基于单一CPU串行平台的构建方式,计算耗时成为突出问题,如何实现高性能低价格的有限元分析平台,成为迫切需要解决的问题。采用宏观有限元模型进行非线性分析能有效减小计算工作量、缩短计算时间,然而宏观有限元模型无法完整描述三维结构非线性特性,可能导致分析结果存在较大误差。为寻求能够兼顾计算精度与计算效率二者平衡的精细化模型分析软件平台,本文通过CPU串行计算与GPU高性能并行计算相结合,建立CPU-GPU异构平台,开发适用于GPU的非线性有限元并行算法程序和软件,实现高层结构弹塑性时程分析的高精度、高效率。论文主要研究工作如下:(1)CPU-GPU异构平台建立与验证。基于CPU和GPU的混合编程思路,让CPU处理数据的读写和逻辑控制等顺序型任务,GPU则处理计算密集的大规模数据操作,建立了与有限元“分而治之”并行思想统一的CPU-GPU异构平台。基于CUDATM并行编程模型分别开发了适用于GPU加速运算的共轭梯度法(CG)和预处理共轭梯度法(PCG)并行求解器,实现有限元大型稀疏线性方程组在CPU-GPU异构平台上的高效并行求解,采用多个适用于GPU计算的并行优化策略,进一步提高了CPU-GPU异构平台的计算效率,并通过框架结构算例初步验证了异构平台的可靠性和有效性。(2)基于CPU-GPU异构平台的纤维模型非线性分析方法研究。在CPU-GPU异构平台上建立了考虑轴力-弯矩-剪力-扭矩共同作用的纤维模型,对箍筋约束混凝土补充了材料损伤试验研究,以此建立了约束混凝土的材料模型。提出了适用于GPU计算,且能有效解决结构负刚度问题的位移增量迭代算法,从有限元离散化角度出发,建立了全局自由度计算数据与GPU线程一一对应映射关系,结合纤维模型、材料模型和位移控制算法开发了基于GPU的有限元并行程序,实现了高精度高效率的梁柱结构非线性模拟。(3)基于CPU-GPU异构平台的空间壳元非线性分析方法研究。通过分层式的平面应力单元和考虑剪切变形的Mindlin板单元组合,构造了基于CPU-GPU异构平台的高精度空间壳元模型,开发了适用于模拟剪力墙和楼板的二维混凝土滞回本构模型,给出了应力状态更新的混凝土本构积分方法,提出了适用于GPU计算的Newton-Raphson迭代算法,实现了高效的板壳结构非线性有限元分析。通过与试验结果对比,表明CPU-GPU异构平台建立的空间壳元能够较为准确地描述剪力墙复杂受力状态下的非线性行为,并且计算效率较高。(4)基于CPU-GPU异构平台的复杂高层钢筋混凝土结构非线性动力时程分析方法研究。在CPU-GPU异构平台上建立了高层建筑结构非线性地震反应分析的空间结构模型,提出了包含复杂材料本构计算的精细化空间纤维模型和空间分层式壳元模型的地震响应分析GPU并行计算方法。开发了基于GPU的弹塑性地震响应分析程序,结合EBE技术将结构的“整体”计算分解到“单元”上,有效减小系统方程组求解对内存空间的需求。通过与振动台试验结果进行对比分析,验证了程序的有效性。选取2个实际高层钢筋混凝土框筒结构工程进行大震弹塑性时程分析,实例分析表明本文所建立的CPU-GPU异构平台及所开发并行程序具有求解规模大、计算精度高、计算效率高、可移植性好、适应性强等优势,具有良好的工程应用前景。
[Abstract]:With the high-rise building structure to the development of fine modeling, 3D structure refinement analysis of high degree of freedom model of huge amount, huge workload of computing, software analysis and calculation speed of the platform more demanding. However, both at home and abroad based on the construction of a single multi way analysis software CPU serial platform, computing time become a prominent issue, how to achieve high performance finite element analysis of the low price of the platform, become an urgent problem to be solved. The macro finite element model for nonlinear analysis can effectively reduce the computational workload, shorten the calculation time, but the macro finite element model to describe the nonlinear characteristics of complete three-dimensional structure, may cause the analysis result has larger errors. In order to seek to take into account the refined model calculation accuracy and the calculation efficiency of the balance of the two analysis software platform, through CPU serial calculation and GPU high performance parallel computing The combination of the establishment of CPU-GPU platform, nonlinear finite element algorithm and software development for GPU, to achieve high accuracy, high structure elastoplastic analysis of high efficiency. The main research work is as follows: (1) establishment and verification of CPU-GPU platform. CPU and GPU mixed programming method based on CPU read and write data processing and control logic of sequential tasks, GPU will deal with the large-scale data intensive computing operation, CPU-GPU platform is established with the finite element of "divide and rule" idea of parallel unification. CUDATM parallel programming model are developed for GPU acceleration algorithm based on conjugate gradient method (CG) and the preconditioned conjugate gradient method (PCG) solver, efficient parallel finite element solution of large sparse linear equations in the CPU-GPU platform, the more suitable for GPU parallel computing optimization strategy, a To further improve the computational efficiency of CPU-GPU heterogeneous platform, and through the frame structure example verified the reliability and validity of the heterogeneous platform. (2) analysis method of nonlinear fiber model based on CPU-GPU heterogeneous platform. At CPU-GPU platform fiber model considering the axial force and bending moment - Shear Torque interaction is established to supplement the confined concrete, experimental study on material damage, in order to establish the material model of confined concrete is proposed. For GPU computation, and can effectively solve the structure of negative displacement increment iterative algorithm the stiffness problem of finite element discretization, starting from the point of view, to build a global freedom calculation data and GPU thread one-to-one mapping relationship with fiber model, material model and the displacement control algorithm is developed for parallel program based on GPU of finite element, nonlinear beam column structure to achieve high precision and high efficiency Quasi. (3) analysis method of space shell element nonlinear CPU-GPU based on heterogeneous platforms. The plane layered stress element and considering the shear deformation of Mindlin plate element combination, structure of the high precision space shell element model based on CPU-GPU heterogeneous platform, developed for the simulation of shear wall and slab two-dimensional hysteretic the constitutive model is given to update the stress state of concrete constitutive integration method, proposed Newton-Raphson iterative algorithm for GPU computation, the shell structure nonlinear finite element analysis, and test results. Through comparison, indicates that the space shell element CPU-GPU heterogeneous platform can accurately describe the complex nonlinear shear wall the behavior under stress, and high computational efficiency. (4) CPU-GPU heterogeneous platform of complex nonlinear high-rise reinforced concrete structure dynamic analysis method based on research In the CPU-GPU platform. The model of spatial structure analysis of nonlinear seismic response of tall building structures is established, puts forward the calculation method of parallel analysis of GPU constitutive model and the fine spatial fiber space layered shell element model containing complex material seismic response. The development of GPU elastoplastic seismic response analysis program based on combined with the technology of EBE structure of the "whole" calculation "unit", to effectively reduce the demand of the system equations of the memory space. Through a comparative analysis and shaking table test results, verify the validity of the program. Select 2 actual high-rise reinforced concrete frame tube structure for engineering seismic elastic-plastic the time history analysis, example analysis shows that the CPU-GPU platform is established in this paper and the development of parallel programs is solved in a large scale, high accuracy, high computation efficiency, good portability, suitable It has the advantages of strong stress and so on, and has a good prospect of engineering application.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TU973.12
，

本文编号：1608292