Parallelizing Of Finite Element Analysis In JIFEX Software

With the development of modern mechanics, computing mathematics and computer technology, as numerical analysis tool which have solid theoretical foundation and wide application power, Finite Element Analysis (FEA) software help a lot in solving complex engineering problems and play an important role in the national economy building and the development of science. One of them is the JIFEX software, which is jointly developed by Department Mechanics, Dalian University of Technology and the State Key Laboratory of Structural Analysis for Industrial Equipment. However, the calculation speed of CPU and the limited storage space restrict the usage of traditional FEA software for the analysis of ultra-large-scale engineering problems. It often results in too long calculation time and sometimes it may become impossible. This has become the bottleneck of applying traditional FEA software to deeper engineering practices.In recent years, due to the rapid development of high speed networks, cluster parallel computing has been a major international parallelism direction. Cluster has enormous potential calculation, a good performance/cost ratio and the flexible architecture. Combination of cluster and FEA can increase analysis speed and decrease the time of the analysis of ultra-large-scale engineering problems. It is one of the most notable forward topics in FEA all over the world.In this paper, parallel computing and MPI are introduced and distributed FEA parallel algorithms based on distributed networks are studied. After a series of analysis, equation parallel direct algorithm, which has good stability, high precise and the advantage of easily modifying the existed serial programs, is implemented to FEA part of JIFEX. The original interfaces of JIFEX are preserved on the base of adding parallelism modules.Parallel LDL~T decomposition algorithm and triangular equation system solution on one dimension variable bandwidth matrix are designed and implemented. Because the stiffness matrix of large scale structure is huge, the matrix is divided into small blocks. The efficient combination of parallelism and block method make the usage of memory reduced.Finally, the parallelism JIFEX software is tested on PC clusters. The numerical results show that the algorithms have good speedup radio and efficiency and the FEA speed of JIFEX is obviously expedited.