MPM parallel computing is implemented using the Structured Adaptive Mesh Refinement Application Infrastructure (SAMRAI) developed by Lawrence Livermore National Laboratories. SAMRAI allows the use of serial MPM code for multi-processor distributed memory computation without any change on the MPM algorithm. Multi-level grid refinement and temporal refinement are adopted to refine regions where high accuracy is needed. With multi-level refinement, the computational cost can be reduced significantly while still maintaining high accuracy. Generally, the computational domain is composed of multiple grid levels and each level is composed of multiple rectangular patches. The patches in the same level are overlapped through ghost cells for data communication. Parallel communication schemes, which include the communication between patches in the same level and communication between two neighboring levels, are described based on the SAMRAI framework. Material point refinement and coarsening algorithms are developed for the communication between two neighboring grid levels. Several indentation problems are simulated using parallel computation. In order to handle the contact between the indenter and work piece better, the multi-mesh MPM algorithm is used. The area below the indenter, where high stress gradient is expected, is refined initially and kept throughout the simulation. With static load balance, good speed-ups are achieved. The parallel MPM results, i.e., the stress contour plots and the indentation load-displacement curves, in general agree with FEM results. This MPM parallel computation scheme has the potential to be used for multi-scale simulations, in which large amount of computation, as well as several length scales, are involved.
Back to Graduate Student Competition
Back to SES Abstracts
Back to The 41st Annual SES Technical Meeting