Automated segmentation of computed tomography images of fiber-reinforced composites by deep learning

Abstract

A deep learning procedure has been examined for automatic segmentation of 3D tomography images from fiber-reinforced ceramic composites consisting of fibers and matrix of the same material (SiC), and thus identical image intensities. The analysis uses a neural network to distinguish phases from shape and edge information rather than intensity differences. It was used successfully to segment phases in a unidirectional composite that also had a coating with similar image intensity. It was also used to segment matrix cracks generated during in situ tensile loading of the composite and thereby demonstrate the influence of nonuniform fiber distribution on the nature of matrix cracking. By avoiding the need for manual segmentation of thousands of image slices, the procedure overcomes a major impediment to the extraction of quantitative information from such images. The analysis was performed using recently developed software that provides a general framework for executing both training and inference.