上海交通大学学报 ›› 2019, Vol. 53 ›› Issue (9): 1100-1106.doi: 10.16183/j.cnki.jsjtu.2019.09.012

• 学报(中文) • 上一篇    下一篇

基于多重节点的结构动力学显式异步长并行计算方法

马志强,楼云锋,李俊杰,金先龙   

  1. 上海交通大学 机械与动力工程学院; 机械系统与振动国家重点实验室, 上海 200240
  • 发布日期:2019-10-11
  • 通讯作者: 金先龙,男,教授,博士生导师,E-mail: jxlong@sjtu.edu.cn.
  • 作者简介:马志强(1990-),男,安徽省蚌埠市人,博士生,研究方向为大规模结构动力学分析与并行计算.
  • 基金资助:
    国家重点研发计划(2016YFB0201800),国家自然科学基金(51475287,11772192)资助项目

Explicit Asynchronous Time Steps Parallel Computational Method for Structural Dynamics Based on Multiple Overlapping Nodes

MA Zhiqiang,LOU Yunfeng,LI Junjie,JIN Xianlong   

  1. School of Mechanical Engineering; State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai 200240, China
  • Published:2019-10-11

摘要: 局部精细化网格会导致结构有限元动力学分析的计算时间大幅增加.为了提高计算效率,结合域分解法与子循环方法,提出一种基于多重节点的结构动力学显式异步长并行计算方法.该方法采用节点分割将模型划分为若干子分区,子分区采用显式Newmark时间积分格式并根据分区单元特性选用时间步长;相邻分区通过多重节点构成耦合区域,小步长分区子循环过程中不涉及边界数据的插值过程.“天河二号”超算平台上的算列表明:采用显式异步长并行方法计算结构动力学问题,在提高计算效率的同时可以保持较高的计算精度.

关键词: 显式积分; 有限元分析; 多重节点; 异步并行; 精细网格

Abstract: The local refined mesh leads to a large increase of calculation time in the structural dynamic finite element analysis. In order to improve the calculation efficiency, an explicit asynchronous parallel computational method based on multiple overlapping nodes is proposed. The procedure combines domain decomposition with subcycling method. The whole structure is divided into several subdomains by node partition method and the integral step size is selected according to the internal mesh sizes of the subdomain. An explicit Newmark time integral scheme is adopted in the subdomain. Multiple overlapping nodes between different subdomains constitute the coupling region without the interpolation process in the subcycling process. Simulation results of two cases implemented on Tianhe 2 supercomputer platform show that the proposed method can improve the computational efficiency with a high computational accuracy.

Key words: explicit integration; finite element analysis; multiple overlapping nodes; asynchronous parallel; refined mesh

中图分类号: