基于精细时程法的结构动力响应并行分析系统

PARALLELANALYSIS SYSTEM FOR STRUCTURAL DYNAMIC RESPONSES BASED ON HIGH ACCURACY DIRECT INTEGRATION METHOD

结构动力学问题在岩土、船舶、汽车等工程领域有着广泛的应用,同时也对大规模数值计算提出了越来越高的要求。我国在超级计算机硬件技术方面已达到国际先进水平,但由于在技术上超级计算机一般是源代码兼容,国内超级计算机不能直接安装商业性有限元软件,大大限制了国产超级计算机的工程应用。从分析有限元结构的基本过程入手,提出将超级计算机并行求解程序和商业性有限元软件集成于一体的解决方案。将占动力响应分析过程总耗时70%以上的方程组求解并行化,从而缩短总时间,前后处理仍在商业性软件中进行。该方案能把商业性软件强大的前后处理功能和超级计算机的高速计算能力结合起来。并行程序采用精细时程积分法的并行算法来实现,并给出了节省存储量的一些途径。算例验证的结果表明,并行程序具有明显的加速比和并行效率。

The structural dynamics is applied widely to many fields including geo-engineering, ship building, car industry, etc.. It also demands large-scale numerical calculations. Hardware technologies of supercomputers have achieved internationally advanced level in China. Generally, supercomputers demand compatibility at the source-code level, so the commercial FEM software usually cannot be adopted directly to domestic Chinese supercomputers, thus puts limitations on applications of supercomputers for engineering fields. To overcome this difficulty, a scheme is presented based on analysis of basic processes of the FEM, using integrated parallel calculation programs for supercomputers and commercial FEM software. During a dynamic response analysis, 70 percent of the total time or more is spent on the solution of equations. Therefore, the total time of the dynamic response analysis can be effectively saved on the parallel solution of the equations. The other parts of the dynamic response analysis are serially performed by commercial FEM software. Obviously, the scheme combines strong pre-processing of commercial FEM software with high-speed calculation of supercomputers. The high accuracy direct integration method uses a mass-based matrix-multiplication algorithm and can be easily parallelized. Based on parallel calculation of me high accuracy direct integration method, parallel programs are implemented and some approaches are given in order to save memory. Some existing examples have been tested with the developed technique and the results prove that the algorithm has high speedup and parallel efficiency.