Deep learning-based 3D reconstruction of SiCf/SiC-W-ZrB2 composites during chemical vapor infiltration and slurry infiltration

LI H. ¹, WEI C. ¹, LI X. ¹

¹ Northwestern Polytechnical University, Xi'an, China

This study investigates the microstructural evolution of SiC_f/SiC-W-ZrB₂ composites manufactured using multi-step chemical vapor infiltration (CVI) and slurry infiltration (SI). Here, X-ray computed tomography (CT) imaging was performed after each CVI step to capture the progression of the evolving microstructure. And the evolution of the internal microstructure in composites was visualized and quantified by carrying out based on the combination of multiple segmentation methods and deep learning. The feasibility and effectiveness of deep learning-based image segmentation for SiC_f/SiC composites with high-density modified components were studied. Meanwhile, the accuracy of deep learning segmentation results was verified. The quantitative reconstruction results reveal that the mean relative errors of SiC_f/SiC and Metal(W/ZrB₂) reconstruction are 2.68% and 6.99%, respectively, and the errors of each component are less than 5%. In each process stage, the SiC_f/SiC reconstruction results have the highest matching degree due to the distribution rule, and the average relative error is 3.74%. Overall, this study provides a basis for high-precision 3D image reconstruction of SiC_f/SiC-W-ZrB₂ composites, which is of great significance for understanding and optimizing the preparation process. And this will facilitate the performance simulation, preparation, and development of SiC_f/SiC composites with high-density components. In the future, we will use this characterization method to explore more interesting directions of functional structure integrated composite materials independently developed.

Keywords: Microstructural characterization, 3D reconstruction, SiCf/SiC, Deep learning, CT