Parallel Computing of Large Eigenvalue Problems for Engineering Structures

摘要

Aimed at large, sparse eigenvalue problems in modal analysis of engineering structures, some predominant algorithms and their corresponding software packages are summarized. A parallel solving system was constructed via integrating these software packages into the finite-element parallel computing framework-PANDA. The finite element model (FEM) of engineering structures was built in pre-processing commercial software-MSC. Patran. Based on the interface between PANDA and MSC. Patran, the model information was translated in PANDA to generate stiffness and mass matrices in a parallel way. Utilizing these matrices, a large-scale parallel computing of eigenvalues was carried out via calling software packages in PANDA. The numerical results show that PANDA frame is competent for carrying out large-scale parallel computing of eigenvalue problems; in virtue of supercomputer, the computing scale attains millions of freedom degrees; and the parallel efficiency is favorable.