Concerning the modeling problem, a damage meso-model [1,2,3] for laminates developed over the past twenty years at LMT-Cachan was used. This meso-model takes into account the intra-laminar damage mechanisms as well as the inter-laminar deterioration. The internal variables characterizing the damage are considered constant over the thickness of the elementary layer and the inter-laminar interface behavior law is coupled to the internal variables of the adjacent layers. These two assumptions make the model highly non-local, and therefore not adaptable for an implementation within commercial software. Because the model is non-local, the information of the neighbouring elements are required for the correct integration of the behaviour law.

After finite element discretization, nonlocality implies that one element requires information from its neighboring elements in order to integrate the behavior law correctly, which makes the preprocessing and partitioning stage very delicate. Thus special constraints have to be imposed on the mesh partitioning procedure. This is done with a new approach that consists of a graph reduction technique, coupled with a third party partitioning tool [4]. This technique enables us to perform fast, hight quality and automatic partitions of the mesh.

A enhanced version of a classic Newton-Krylov-Shur (NKS) method is introduced. A second Newton for the localization step of the Balancing Domain Decomposition (BDD) [5] method was used to improve the performance. This local Newton method enables to treat the nonlinearities at the correct scale. Therefore, in the case when the nonlinearity is localized in a reduced zone and has little impact on the rest of the structure, the nonlinearity is treated almost completely at the subdomain scale, so a few or no iterations are required for the global Newton problem.

To be able take advantage of the latest supercomputer a parallel implementation of the resolution technique was required. The software developed was written in C++ using MPI and multithreating techniques. This software can take advantage of the architecture of the latest SMP (Symmetric Multi-Processor) clusters.

Finally, we showed some preliminary results and the level of performance which can be expected from such an approach.

1

P. Ladevèze, "About a damage mechanics approach", Mechanics and Mechanisms of Damage in Composite and Multimaterials, 119-142, 1998.

2

P. Ladevèze, G. Lubineau, "An enhanced mesomodel for laminates based on micromechanics", Composites Science and Technology, 62, 533-541, 2002. doi:10.1016/S0266-3538(01)00145-2

3

G. Lubineau, P. Ladevèze, "Construction for a micromechanics-based intralaminar mesomodel, and illustrations in ABAQUS/Standard", Computational Materials Science, 43(1), 137-145, 2008. doi:10.1016/j.commatsci.2007.07.050

4

G. Karypis, V. Kumar, "A fast and high quality multilevel scheme for partitioning irregular graphs", Journal on Scientific Computing, 20, 359-392, 1998. doi:10.1137/S1064827595287997

5

J. Mandel, "Balancing domain decomposition", Communications in Numerical Methods in Engineering, 9, 233-241, 1993. doi:10.1002/cnm.1640090307

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £72 +P&P)