关键词: 数值流形法, 覆盖系统, 并行计算, 区域分解算法, 流形单元, 子区域, 界面单元

Abstract: Computational time of numerical manifold method is a bottleneck problem in simulating large-scale practical engineering problems. Parallel computing is an effective accelerating technique. Based on the serial algorithm, the parallel cover system and the manifold element generation algorithm based on the distributed memory programming mode are studied. Compared with the serial algorithm, differences and difficulties of parallel algorithm are firstly described. The domain decomposition algorithm is used to assign parallel tasks, and the main steps of parallel algorithm of cover system are proposed. The generation of sub-domain is formed by using a temporary coarse grid taking load balancing into consideration, and the corresponding sub-domain generation algorithm is proposed. In subsequence, based on the analysis of the relationship among mathematical covers, physical covers and manifold elements of numerical manifold method, the essential techniques in parallel algorithm, as well as physical coverage and manifold element numbering rules are discussed. Then the algorithm for establishing the interface information is proposed for the parallel algorithm, which provides necessary information for data exchange in the parallel computing process of the numerical manifold method. Finally, based on the parallel generation algorithm of cover system, a code was developed using C++ and object-oriented programming technology. Two numerical examples verify the parallel algorithm, which provides a preprocessor for subsequent parallel simulation.

Key words: numerical manifold method, cover system, parallel computing, domain decomposition algorithm, manifold element, sub-domain, interface element