Voids are unavoidable during the manufacturing of 3D braided composites. This study proposes an unsupervised machine learning method combined with micro-computed tomography (micro-CT) scanning and a progressive damage analysis to analyze defects in these composites at a -scale level. The method enables the creation of real multiscale models and the determination of the porosity in both the intra-yarn (1.52 %) and inter-yarn (5.04 %) planes. Here, the unsupervised machine learning method is introduced in a trans-scale damage analysis to reduce calculation dimensions and to visualize the clustering data. A user-defined material subroutine (UMAT) is also developed to implement the trans-scale damage model. The experimental validation of the simulation results demonstrates the effective trans-scale damage analysis, showing the predominant pull-shear damage in the yarns, which is primarily located at the interfaces both between the yarns and between the yarns and the matrix. Finally, based on the scanned geometric data the degradation in modulus and strength of 3D braided composites with porosity is studied.

中文翻译：

---

基于显微 CT 成像和无监督机器学习的带空隙 3D 编织复合材料的跨尺度分析

在 3D 编织复合材料的制造过程中，空隙是不可避免的。这项研究提出了一种无监督机器学习方法，结合微计算机断层扫描 (micro-CT) 扫描和渐进式损伤分析，以在规模水平上分析这些复合材料中的缺陷。该方法能够创建真实的多尺度模型并确定纱线内（1.52％）和纱线间（5.04％）平面的孔隙率。这里，在跨尺度损伤分析中引入无监督机器学习方法，以减少计算维度并使聚类数据可视化。还开发了用户定义的材料子程序（UMAT）来实现跨尺度损伤模型。模拟结果的实验​​验证证明了有效的跨尺度损伤分析，显示了纱线中主要的拉剪损伤，该损伤主要位于纱线之间以及纱线与基体之间的界面处。最后，基于扫描的几何数据，研究了具有孔隙的 3D 编织复合材料的模量和强度的退化。