Michael A. Puso

Accomplishments and Challenges in Code Development for Parallel Multimechanics Simulations


The Methods Development Group at Lawrence Livermore National Laboratory has historically developed and supported software for engineering simulations, with a focus on nonlinear structural mechanics and heat transfer. The quality, quantity and complexity of engineering analyses have continued to increase over time as advances in chip speed and multiprocessing computers have empowered this simulation software. As such, the evolution of simulation software has seen a greater focus on multi-mechanics and the incorporation of more sophisticated algorithms to improve accuracy, robustness and usability. This talk will give an overview of the latest code technologies developed by the Methods Development group in the areas of large deformation transient analysis and implicit coupled codes. Applications were run on the state of the art hardware available at the national laboratories.

First, the algorithms related to explicit transient dynamic analyses and their parallelization will be presented. This includes dynamically defined contact (i.e., evolving contact surfaces, dynamic partitioning) and coupled finite element/meshless simulations. Applications here include penetration and automobile crash analyses.

Second, recent accomplishments in coupled code development will be discussed. The main focus here will be on implicitly integrated dynamics and statics and multiple continuum field theories including solid mechanics, heat transfer and electro-magnetics. Whereas solid and thermal discretizations typically rely on nodal degrees of freedom, electro-magnetic Nedelec elements employ edge-based degrees of freedom and finite volume methods employ face and cell based degrees of freedom. A hierarchical data model based on object oriented FORTRAN 95 was implemented to accommodate these node-edge-face-cell type objects to simplify code structure. Distributed memory parallelism is accomplished through standard MPI functionalities. Classical node-onsegment contact algorithms are typically not robust, particularly for implicit structural mechanics and electro-magnetics. Consequently, research in the area of mortar segmentto-segment contact has produced state-of-the-art algorithms that exploit this hierarchical data model. An overview of the coupled non-linear solution techniques used for the coupled systems will be given. Results exploiting adaptive meshing and applications in the area of rail gun and reactor technology will be presented.

Finally, some open challenges that reflect current research efforts within the Methods Development Group will be outlined. In particular, portable intercode coupling methodologies, accurate error estimation techniques, linear equation solvers, multi-scale material models, embedded mesh, and multi-core computer architectures all represent opportunities for innovative strategies in the future.

Toulouse | France | 2008 | June | 24  25  26  27