Watertight conforming mesh generation of continuous fiber-reinforced composites from μ-CT images via deep learning and boolean repair

Abstract

Although significant progress has been made in reconstructing high-fidelity FE models of continuous fiber reinforced ceramic matrix composites (CFCMC), many existing workflows still involve complex procedures, or require specialized programming expertise. These factors reduce accessibility and reproducibility, restricting broader adoption in engineering applications. To address these limitations, this study develops a workflow built entirely on open-source tools for generating finite element (FE) models of carbon fiber reinforced silicon carbide composite (C_f/SiC) from μ-CT data. The workflow integrates U-Net segmentation, Alpha-Wrap reconstruction with volume correction, PyMeshLab Boolean operations with watertight repair, and tetrahedral meshing using Gmsh. Validation demonstrates close agreement with voxel-based models, with fiber bundle volume fraction differences below 2%. Generalization tests further demonstrate robustness across different regions of interest. The proposed approach prioritizes ease of use and reproducibility, offering a practical and accessible alternative that broadens access to realistic CFCMC modeling for both research and engineering applications.