Non-Intrusive Parallel Implementation of GFEM with Global-Local Enrichment for Multi-Domain Simulations

Resumo

This paper expands the non-intrusive implementation of the Generalized Finite Element Method with Global-Local enrichment (IGL-GFEMgl) for multi-domain analyses simulated under a parallel algorithm approach. In IGL-GFEMgl, the global problem is initially discretized using a coarse mesh without considering localized phenomena. This problem is solved by the Abaqus commercial software using the Finite Element Method. The localized phenomena of interest, such as cracks, are effectively represented on a local scale, defining problems associated with one or more local domains. Mesoscale problems are created as intermediate scales between the global and local domains. The association between each mesoscale and its respective local problem is carried out through the global-local enrichment of GFEMgl. The coupling of the mesoscales to the global problem is established through the transfer of displacements and generalized forces, characterizing the non-intrusive strategy known as Iterative Global-Local (IGL). Numerical simulations using GFEMgl are performed in the computational system INSANE (Interactive Structural ANalysis Environment – www.insane.dees.ufmg.br), a free software project developed at the Department of Structural Engineering at the Federal University of Minas Gerais. Numerical examples are presented to demonstrate the simulation's performance and to investigate the influence of the main parameters related to the proposed strategy. The quality of the results is evaluated based on the state of stresses and strains at critical points of the models and Linear Elastic Fracture Mechanics parameters. The investigations indicate that the convergence process is not affected by the consideration of multiple local domains, as long as these do not result in a significant increase in the stiffness difference between the coupled scales. Additionally, the computational performance of the algorithm was evaluated, comparing simulations in which the mesoscale processing was executed sequentially and in parallel.