地铁地下结构抗震分析并行计算显式与隐式算法比较

摘要

The computing platform is based on the EM64T hardware framework, dual-path Intel Xeon processor, and Gigabit Ethernet subsystem which contains 32 CPUs and configures 64-bit ABAQUS applications and Linux operating system. The calculation precisions and efficiency of the explicit finite element method which u-ses the central difference method and the implicit finite element method which uses the Hilber-Hughes-Taylor method are verified, the precisions with the viscous-spring artificial boundaries are verified simultaneously. The results show that the calculation precisions of the explicit finite element method and the implicit finite element method are basically the same, but the efficiency of the explicit finite element method is much higher than that of the implicit finite element method. In the research, the comparisons between the explicit and implicit finite element methods using multiple processors in the 2D and 3D nonlinear seismic responses of metro station structures were achieved. The results show as follows: For the case of the 3D large-scale model with DOF being 387 426, the solution time of the explicit finite element method using 8 CPU, 16 CPU and 32 CPU are respectively 21. 89%, 23. 10% and 4. 32% of that of the implicit finite element method; for the case of the 2D small-scale model with DOF being 10 516, the solution time of the implicit finite element method using 1 CPU, 2 CPU and 4 CPU are respectively 41. 4%, 45. 3% and 51.8% of that of the explicit finite element method; so the explicit finite element method with multiple processors parallelization is fit for large-scale numerical computing problems while the implicit finite element method with multiple processors parallelization is fit for small-scale numerical computing problems.%基于EM64T硬件构架、双路Intel Xeon处理器、Linux操作系统、64位ABAQUS软件、千兆以太网络为集群子网络构建的32 CPU并行计算集群平台,对有限元并行计算中心差分显式算法与Hilber-Hughes-Taylor隐式算法的计算精度和效率以及ABAQUS软件中设置黏弹性人工边界计算精度进行比较验证.结果表明:显式与隐式算法计算精度基本相当,但显式算法计算效率远高于隐式算法.对地铁地下结构三维和二维非线性地震反应分析有限元并行计算显式和隐式算法进行对比研究,结果表明:对于自由度数为387 426的地铁车站结构三维非线性地震反应分析,8、16、32 CPU并行计算显式和隐式算法耗时比依次为21.89％、23.10％、4.32％；对于自由度数为10 516的地铁车站结构二维非线性地震反应分析,1、2、4 CPU并行计算隐式和显式算法计算耗时比依次为41.4％、45.3％、51.8％；多CPU并行计算显式算法适合求解大规模数值计算问题,多CPU并行计算隐式算法适合求解小规模数值计算问题.