基于液态食品冷冻浓缩冰晶生长机制并行计算

发布时间：2019-01-20 14:47

【摘要】：本课题研究了基于在液态食品冷冻浓缩冰晶生长机制研究中建立起来的相场理论模型前提下,利用并行数值计算技术实现冰晶生长过程的数值模拟。模拟出在二维方向上冰晶的结构图,并进行了结果分析。在本论文中重点讨论了偏微分方程的数值分析方法以及并行计算机系统体系结构原理和理论以及其与并行应用程序设计有关的编程原则和方法。具体内容如下：1)针对液态冷冻浓缩冰晶生长机制研究中利用相场理论建立的相场微分方程、溶质场微分方程和扰动微分方程数学模型,经过深入的理解和推导微分方程等效形式,分析了适合数值计算的形式并找出在时间和空间上数值离散的方法。2)详细地讨论了并行计算机系统的结构一般原理,特别是与并行计算性能相关的硬件特征。3)结合上海超算中心曙光5000A(230Tflops)超级计算机,讨论了的并行计算机的体系结构,特别是计算结点的资源情况,将数值分析的结果映射到曙光5000A计算结点上,抽象出并行计算的求解模型。4)讨论了MPI、OpenMP并行设计方法的一般原理,结合数值分析,讨论了粒度匹配加速比模型、研究了并行粒度划分与并行性能的关系以及计算与通信重叠的并行编程技术。5)设计出了适合并行计算数据处理的算法和并行程序。6)讨论了程序的数据结果分析、测试、并行性能评价方法。
[Abstract]:In this paper, the numerical simulation of the ice crystal growth process is realized based on the phase field theory model established in the research of the ice crystal growth mechanism of liquid food freezing and concentration. The structure of ice crystals in the two-dimensional direction is simulated, and the result analysis is carried out. In this paper, the numerical analysis method of partial differential equation and the principle and theory of parallel computer system architecture and its programming principles and methods are discussed. The specific content is as follows: 1) The phase field differential equation, the solute field differential equation and the mathematical model of the disturbance differential equation are established by using the phase field theory in the research of the liquid freezing and concentration ice crystal growth mechanism, and the equivalent form of the differential equation is further understood and derived, this paper analyzes the form of suitable numerical calculation and finds out the method of numerical dispersion in time and space. 2) The general principle of the parallel computer system is discussed in detail, in particular the hardware features related to the performance of parallel computation. In this paper, the architecture of the parallel computer is discussed, in particular, the resource of the node is calculated. The result of the numerical analysis is mapped to the node of the Dawning 5000A and the solution model of the parallel computation is abstracted. 4) The general principle of MPI and OpenMP parallel design method is discussed, and the numerical analysis is combined. In this paper, the model of particle size matching acceleration ratio is discussed, the relation between the parallel granularity partition and the parallel performance, and the parallel programming technique for computing and communication overlap are studied. 5) The algorithm and the parallel program for parallel computing data processing are designed. The data result analysis and test of the program are discussed. Parallel performance evaluation method.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TS205.7;TP338.6

【共引文献】