Research On The Parallel Finite Element Method For Large Structures Based On Linux Cluster

With the construction developing in our country, more and more large-scale, super large-scale complex structures are constructing. These engineering not only have huge dimension but also have very complicated condition, and relate to intricate nonlinear constitutive relation, time-consuming. Traditional finite element method running on single PC can't satisfy the computation of these problems. The technology of parallel computation is introduced into the structure finite element analysis to increase the size of the problems, accelerate the analysis and facilitate the application of the finite element method on large-scale structures.With the developing of computer and network technology, PC clusters have gained an increasing importance as platforms for parallel high-performance simulation problems. On this background, a high-performance computing cluster using PC is set up based on Linux. The detailed method of configuration and use is presented. The scientific compute tool PETSc is configured and used to solve PDE in the distributing environment. This dissertation introduces the method of using and the process of executing.The domain decomposition is a numerically approximating solution to partial differential equations on parallel computers. It divides the domain into subdomains, and uses parallel method to solve the problems in the subdomains. Based on the trait of domain decomposition and the parallel precondition conjugate gradient method, using C++ and MPI, This dissertation develops a parallel finite element program ADVPFEM based on Linux cluster.Using the parallel program make a parallel finite element analysis for the cube, and gain the accelerated ratio and parallel efficiency. The results gained by parallel program are the same as the results gained by the ABAQUS software. The 3D parallel finite element analysis for the simple model of Longma soil-stone Dam with million elements is completed on the Linux cluster. With the same CPU, the required computing time for the different subdomain is presented. The results show that the parallel computing time will not decrease at all times when the CPU and subdomain increase.