PAGFEM: Parallel Adaptive Generalized Finite Element Methods for Large Scale Fractures
Carlos Armando Duarte
Award year: 2006-2007
The prediction of crack growth is of great importance in several areas of application, such as the analysis of fatigue failure of structural components and the assessment of the performance of structures subjected to extreme conditions. Several catastrophic accidents, like the loss of part of the fuselage of a Boeing 737-200 during the Aloha Airlines flight 243, were caused by structural failures due to crack growth. The finite element method with remeshing after every crack propagation step, is currently one of the few tools available for the simulation of evolving three-imensional cracks. This approach, however, quickly leads to extremely large and computationally intensive models when applied to realistic three-dimensional problems. The situation gets even more challenging when several cracks are involved, such as in the simulation of corrosion assisted cracks, multi-site damage analysis of lap joints used in aircrafts, high velocity impact problems and modeling of polycrystalline grains. The computing power required to solve this class of problems using existing methodologies and tools is formidable. Previous efforts aimed at parallelizing the simulation of propagating cracks have also exposed the low scalability of existing methodologies. The use of massive computational power by itself is not sufficient. Instead, advances in existing techniques and the development of scalable algorithms are needed. The proposer and collaborators have recently developed a generalized finite element method (GFEM) that overcomes many of the limitations of existing methods. The meshes used the GFEM, for example, do not have to fit crack surfaces and thus the method eliminates the need of remeshing the computational domain after each crack propagation step. The method partitions each crack in a structure into a number of smaller ones that can be efficiently solved using available computational resources. It is therefore ideal for multi-core/multi-processor computers. Preliminary numerical studies show that even on a single-processor computer the method is very effective and allows the solution of larger fracture mechanics problems than would otherwise be possible using existing computational methods. This project seeks the collaboration of NCSA researchers on the parallelization of this emerging simulation tool developed in the Department of Civil and Environmental Engineering (CEE). A strong collaboration between NCSA and CEE is envisioned. In particular, we will work with the NCSA research group of Nahil Sobh, head of the Performance Engineering Team in the Persistent Infrastructure Directorate. This group will provide guidance on porting the proposer's GFEM code to NCSA's SGI Altix supercomputer, parallelization of the code using OpenMP, selection of optimization, debugging and visualization tools. The proposed project leverages ongoing research projects of the proposer and NCSA expertise and extends the frontiers of computational fracture mechanics enabling the solution of problems that are currently not amenable to computer simulation. We expect that the interactions with NCSA forged during this project continue to grow and strengthen in coming years through external support.