Quadrilateral Mesh Generation Method Based on Convolutional Neural Network

Abstract

The frame field distributed inside the model region characterizes the singular structure features inside the model. These singular structures can be used to decompose the model region into multiple quadrilateral structures, thereby generating a block-structured quadrilateral mesh. For the generation of block-structured quadrilateral mesh for two-dimensional geometric models, a convolutional neural network model is proposed to identify the singular structure inside the model contained in the frame field. By training the network model with a large number of model region decomposition data obtained in advance, the model can identify the vectors of the frame field in the region located in the segmentation field. Then, the segmentation streamline is constructed from the annotation. Based on this, the geometric region is decomposed into several small regions, regions which are then discretized with quadrilateral mesh elements. Finally, through two geometric models, it is verified that the convolutional neural network model proposed in this study can effectively identify the singular structure inside the model to realize the model region decomposition and block-structured mesh generation.