Generating 3D digital material twins for woven ceramic-matrix composites from μCT images

This paper proposes a novel method for generating 3D digital material twins (DMTs) from μCT images for woven ceramic-matrix composites. The key points to generating DMTs are the efficient and high-precision identification and segmentation of fiber yarns, matrix, and defects in μCT images. Due to the low gray contrast among fiber yarns, traditional threshold segmentation methods cannot effectively obtain the corresponding woven structures. Therefore, a novel deep convolution neural network (DCNN) model with the U-Net architecture is developed to overcome these difficulties. Based on this approach, cross sections and centerlines of fiber yarns are identified and used to reconstruct their 3D architectures. Defects are introduced through the spatial mapping between their locations in μCT images and the 3D meshes of the matrix. The DMTs of a C/SiC woven composite with quantitative matrix porosity defects are established, and a simulation of the composite under tension is presented. The DMTs approach provides new insights into damage and failure analyses of ceramic-matrix composites.