Reconstruction of a SiC-SiC CMC Microstructure Using Deep Learning and Advanced Image Processing Technique

Abstract

The paper presents a workflow for the reconstruction of a SiC-SiC ceramic matrix composite (CMC) microstructure using advanced image processing techniques and deep learning. The objective of this research is to develop highly accurate physics-based computational models for CMCs by gaining a comprehensive understanding of the microstructural features and their impact on material properties. A workflow is presented to classify voxels into individual components and extract stochastic data for establishing microstructure-property correlations. X-ray computed tomography (CT) data of the SiC/SiC CMC with a plain weave architecture and 00/900 fiber orientation are reconstructed using advanced image processing techniques. The resulting CT data is successfully segmented into three material constituents: tows, matrix, and pores. Pore segmentation is accomplished using the Otsu segmentation algorithm, while a deep learning-based U-net model is employed for accurate segmentation between SiC tows and the SiC matrix. Furthermore, an anisotropic segmentation algorithm is utilized to classify tow voxels along different directions, capturing the intricate variations within the microstructure. Geometrical and morphological attributes are extracted from the segmented data for further analysis.