An Investigation On Automatic Partitioning Algorithm Of Unstructured Meshes

Based on the domain decomposition of parallel computing in Fluid Dynamics, this thesis is dedicated to the research of automatic partitioning algorithm on unstructured meshes, covering computing theories and their practical application.In this thesis, grid partitioning is converted to an unoriented and weighted graph, on which partitioning algorithm is studied. Firstly, the size of the graph is reduced by collapsing vertices and edges. Then, each smaller graph is divided into a certain number of subdomains by means of initial partitioning algorithm. Finally, these subdomains are refined back to the original graph through mapping and multi-partitioning refinement strategy. Combining multilevel scheme, spectral bisection method, KL/FM refinement, as well as multi-partitioning refinement strategy, the computing method is characterized by high efficiency, flexibility, and simplicity because it both maintains load balancing and minimizes the complexity of communication among processors.Several partitioning algorithmic examples in unstructured meshes, ranging from simple 2D meshes to complicated 3D ones, are also given in the thesis. And the partitioning results are followed by an analysis of the superiority of this computing method. Experiments prove that it is more effective and reasonable than traditional methods in terms of time and quality in partitioning. The method is also tested by finite volume scheme and partitioning algorithm of Euler Equation in parallel computing in Fluid Dynamics. The testing results demonstrate that the suggested method in this method is of practical and scientific significance in parallel computing in Fluid Dynamics.