Computational strategies for iterative solutions of large FEM applications employing voxel data

FE-models for structural solid mechanics analyses can be readily generated from computer images via a `voxel conversion' method, whereby voxels in a twoor three-dimensional computer image are directly translated to elements in a FE-model. The fact that all elements thus generated are the same creates the possibilities for fast solution algorithms that can compensate for a large number of elements. The solving methods described in this paper are based on an iterative solving algorithm in combination with a unique-element element-by-element (EBE) or with a newly developed row-by-row (RBR) matrix-vector multiplication strategy. With these methods it is possible to solve FE-models on the order of 10/sup 5/ 3-D brick elements on a workstation and on the order of 10/sup 6/ elements on a Cray computer. The methods are demonstrated for the Boussinesq problem and for FE-models that represent a porous trabecular bone structure. The results show that the RBR method can be 3.2 times faster than the EBE method. It was concluded that the voxel conversion method in combination with these solving methods not only provides a powerful tool to analyse structures that can not be analysed in another way, but also that this approach can be competitive with traditional meshing and solving techniques