大规模并行结构动力分析分层计算方法

A hierarchical computing approach for the massively parallel structural dynamic analysis

多核分布式存储超级计算机的兴起为大规模并行结构动力分析提供了强有力的计算工具。根据多核分布式计算环境的特点,提出了一种大规模并行结构动力分析分层计算方法。该方法在传统隐式动力分析的区域分解法的基础上,利用两级分区和两次缩聚策略进行求解。不但通过进一步缩减求解问题规模有效提高了界面方程的收敛速度,而且通过三层并行计算有效提高了通信效率。该方法并不对有限元模型引入近似,属于精确的动力子区域分层计算方法。典型数值算例表明,该方法计算精度与商业软件ANSYS完全法求解精度相当;同传统区域分解法相比,该方法能够获得较高的并行计算性能。

It provides a powerful computational tool for the massively parallel structural dynamic analysis with the rise of the multi-core distributed memory supercomputer.According to the characteristics of the distributed computing environment,a hierarchical computing approach for the massively parallel struc-tural dynamic analysis is proposed.The proposed approach is based on the conventional domain decompo-sition method for implicit dynamic analysis and utilizes the strategies of two-level partitioning and twice condensation to complete the solution.It not only efficiently improves the convergence rate of interface equations by further reducing the problem size,but also significantly improves the communication effi-ciency with three-layer parallel computing.The proposed approach does not introduce any approximation for the finite element model.Therefore,it belongs to the exact hierarchical computing method based on domain decomposition method for dynamic analysis.Typical numerical experiments show that the com-putational accuracy of the proposed approach is equivalent with the full method of the commercial soft-ware ANSYS.Compared with the conventional domain decomposition method,the proposed approach is superior in the performance of parallel computing.