著者
Naoki MORITA Kazuo YONEKURA Ichiro YASUZUMI Mitsuyoshi TSUNORI Gaku HASHIMOTO Hiroshi OKUDA
出版者
The Japan Society of Mechanical Engineers
雑誌
Mechanical Engineering Letters (ISSN:21895236)
巻号頁・発行日
vol.2, pp.16-00082, 2016 (Released:2016-03-03)
参考文献数
9

Structural elements which have 6 degrees of freedom (DOFs) at each node, such as shell and beam elements, are widely used in structural analysis. A matrix originating from finite element method (FEM) is stored in some sparse matrix storage format with a suitable block size for the number of DOFs at each node. When both 6 DOF structural elements and general 3 DOF solid elements are employed, the sparse matrix storage format becomes complicated due to combination of different block sizes. High computational efficiency of finite element analysis has become more important in large-scale structural problems. The complicated storage format leads to deterioration of computing performance in solving linear equations by an iterative procedure, conjugate gradient (CG) iterations for example. A computational technique is required that allows us to use existing parallel linear solvers without deteriorating the performance for solving linear equation systems originating from combination of solid and structural elements. This research aims to develop 3×3 DOF blocking structural elements to enhance the computational intensity of iterative linear solver, such as the CG method. As numerical results, the proposed 3×3 DOF blocking elements have shown better performance for each CG iteration than the conventional structural elements. The computational efficiencies are 95.0% with single thread execution and 76.6% with 8-thread execution of a theoretical peak performance based on the STREAM benchmark.