![]() | Only 14 pages are availabe for public view |
Abstract Finite element modeling is an established part of engineering design and analysis. it is based on a methodology in which a global representation of a system is formed as an assemblages of simple local forms. Finite element analysis exploits the capabilities of existing serial hardware but is still limited by them. Parallel architectures offer the prospect of significant performance enhancement. there are difficulties to be overcome before useable software can be made available on such machines that benefits from their power. the essence of the finite element method is to partition the domain of the problem into non-overlapping elements and to provide an approximate solution that has a simple form within each element. these local representations are to be patched together to form a global solution with certain desired degree of smoothness. As the local form of the solution is to be kept simple, Accuracy is achieved by making the elements as small as possible. This in turn means that the approximate problem is defined by means of a large number of equations. A piecewise function defined in terms of the values at the nodes determined by the element geometry, leads to a sparse coefficient matrix and this makes it possible to solve larger problems. In our thesis we try to parallelize the finite element |