A deep learning-assisted Micro-CT fusion approach for high-fidelity braided composites RVE modeling and mechanical performance analysis with geometric-dispersion considerations

This study combines Micro-CT scan data for high-resolution 3D data acquisition with deep learning-based image preprocessing to achieve a high-precision representative volume element (RVE) of 3D four-directional braided composites (3D4DCs). By integrating Cycle Generative Adversarial Network (CycleGAN) for image enhancement and YOLOv8 for pixel-level segmentation, this method enables comprehensive and automated analysis of Micro-CT scan data, outperforming traditional thresholding, thereby laying the groundwork for high-fidelity numerical simulation. A statistical mean–based reconstruction method is proposed to reconstruct a high-fidelity 3D4DCs RVE model. Numerical results demonstrate significantly improved predictive accuracy, with simulated stress-strain curves showing close alignment with experimental data. It is found that the pressurization process during curing significantly alters the fiber orientation distribution within the composite material, transforming the material from the originally designed approximately transverse isotropy to orthotropic anisotropic composite material. Furthermore, geometric uncertainty systematically reshape the mechanical response landscape: each deformation mode activates distinct micro-structural feature, causing the same spatial variability to impact elastic constants with notably different intensity. As a result, certain properties exhibit high sensitivity while others remain largely unaffected, implying that the degree of dispersion is inherently property-specific. This work not only establishes a “geometry-property” relationship for braided composites but also introduces a novel deep learning-assisted modeling method. The proposed methodology significantly enhances the accuracy and reliability of braided composites simulations, providing considerable potential for high-precision assessment of braided composites in engineering applications.