TWO DIMENSIONAL ADAPTIVE MESH GENERATION USING NEURAL NETWORKS

Abstract

In the field of Engineering design, the Finite Element Method (FEM) provides a powerful dicretized numerical technique for analysis. The process to decompose a whole domain into many meshes is indespensable in FEM. The accuracy of finite element solution depends on the discretization which is characterized by FEM mesh and geometrical quality of meshes. There are four stages for finite element adaptive mesh generation; to generate background mesh, finite element analysis, posteriori error estimation and adaptive refinement. The quality of meshes in the finite element analysis to achieve better performance can be improved using Back propagation (BP) Neural Network which have adaptive pattern recognition capabilities to predict pattern of meshes and their nodal parameter for refinement (decomposition). To check the performance of mesh generation program in two dimensional plane using triangular and quadrilateral element a cantilever elastic beam and a low darn with a seepage gallary has been analysed. It is found that ANN recognise the geometrical quality of element very well and as per the stress concentration remeshing of element is done