Parametric modeling of 2.5D woven composites based on computer vision feature extraction

Abstract

The aim of this paper is to develop a comprehensive modeling strategy for creating a realistic representative volume element (RVE) of 2.5D woven composites. The strategy consists of two main parts: the extraction of geometric feature parameters and the establishment of a parametric voxel-mesh full-cell model (VFM). Firstly, a neural network model is constructed to achieve an accurate segmentation of yarn cross-sections from X-ray computed tomography (XCT) images. Secondly, geometric feature parameters are then extracted from the segmentation results using image algorithms. Finally, a parametric modeling method is proposed to establish the VFM of the material. To evaluate the performance of the VFM, its structural sizes, overall fiber volume fraction (FVF), and stiffness prediction accuracy are assessed. The comparison results indicate that the VFM achieves a fine mesoscale characterization and a high stiffness prediction accuracy.