Parallel quadrilateral subdomain generation for finite element analysis

In this article the Sub-domain Generation Method (SGM) previously developed for finite element meshes composed of triangular elements is generalised t o account for meshes composed of arbitrary quadrilateral elements. A neural network of enhanced accuracy was developed to predict the number of quadri laterals generated within each coarse quadrilateral element after refinemen t using an adaptive re-meshing procedure. Partitioning of the meshes was un dertaken on the coarse mesh using a genetic algorithm to optimise the parti tions considering both load balancing and interprocessor communication of t he subsequent finite element analysis. A series of examples of increasing r efinement were decomposed by considering a single coarse mesh of a fixed nu mber of elements. (C) 1999 Elsevier Science Ltd and Civil-Comp Ltd. All rig hts reserved.