Identification of microstructures and damages in silicon carbide ceramic matrix composites by deep learning.

Continuous silicon carbide fibre reinforced silicon carbide ceramic matrix composite (SiC/SiC) is a multiphase non-homogeneous anisotropic material used in new generation aero-engines. However, it is difficult to identify its microstructural features and related complex spatial distributions, which determine its mechanical properties. Herein, shallow cross-linked (2.5D) SiC/SiC, fibre bundle SiC/SiC and filament SiC/SiC were prepared by chemical vapor infiltration and their tensile properties were tested. Microstructural features and damage characteristics were studied using micro and nano computed tomography (CT). Deep learning was applied to process the CT images to obtain complex 3D structural features of the composite, especially with the help of segmentation using ORS Dragonfly software. Structural units such as fibre, interphase and matrix, as well as damage features such as matrix cracks, pull-out holes and pull-out fibres were accurately identified at macro-scale, meso-scale and micro-scale (2.5D architecture, bundle and filament, respectively). Important characteristics such as porosity, fibre pull-out length distribution and periodic crack distribution were obtained. This information would be helpful to understand the macro/microscopic mechanical behaviours of SiC/SiC composites and optimize their preparation process. • Micro and nano CT were applied to investigate the microstructural features of SiC/SiC composite. • Constituent units and damage patterns of SiC/SiC were accurately identified from CT images at macro/ meso /micro material scales. • The periodical matrix cracking with minor crack opening distances was precisely distinguished by deep learning. [ABSTRACT FROM AUTHOR]