Image-based 3D mesoscopic modeling and thermo-mechanical properties prediction of SiC/SiC composites with different preforms

Ceramic matrix composites (CMCs) woven by textile technology are considered to be excellent materials for high-temperature components in new-generation aircraft engines. Therefore, it is necessary to develop models reflecting the real structure of CMCs and systematically investigate the effect of the microstructure on their macroscopic properties. In this work, based on micro-computed tomography (CT) scanning images, 3D models of the internal structure of 2D plain-woven and 2.5D SiC/SiC composites are developed using a deep learning (DL) method, and the related properties are predicted. First, CT scanning images of 3D models are segmented using an efficient and accurate DL neural network. The fabric structures of the yarns are reconstructed and real 3D descriptions are generated. Second, 3D finite element models are developed using voxel meshing and element classification programs. Finally, the stress and heat flux distributions of the two composites are calculated and analyzed, and the elastic modulus and thermal conductivity are predicted. The simulation results show that the out-of-plane properties of the plain-woven SiC/SiC composites were much lower than the in-plane ones, and the in-plane modulus and thermal conductivity were 5 and 2 times the out-of-plane ones, respectively. The difference between the out-of-plane and in-plane properties of the 2.5D composites was small, while the out-of-plane thermal properties of the 2.5D SiC/SiC composites were much larger than those of the plain-woven SiC/SiC composites. And the error between the predicted value and the experimental value is all less than 20%, making them more suitable for predicting the heat transfer and elasticity response of CMCs, while further used to predict their strength and damage behavior.