Deep-Learning Detection of Cracks in In-Situ Computed Tomograms of Nano-Engineered Composites

Abstract

The deformation and damage development of nano-engineered composites have not yet been investigated in 3D, although it can provide a deeper insight into their damage behavior. To fill this gap, we perform a tensile test on a nano-engineered composite with in-situ X-ray micro-Computed Tomography (micro-CT). The composite is made from woven alumina fibers with grafted carbon nanotubes (CNTs) and epoxy. More diffuse damage seems to exist for the materials with CNTs compared to the baseline material. However, at such resolution where individual fibers are vaguely visible, grayscale thresholding does not accurately characterize the matrix cracks due to their small opening and low contrast with the material itself. Thus, we employ a deep-learning tool, called RootPainter, for segmentation of cracks with small opening in relation to the voxel size, in the 3D images. The results show that RootPainter can reliably identify these small cracks. In addition to the investigation of the mechanical performance of the nano-engineered composite, this study provides a novel and reliable method for the characterization of micro-cracks in in-situ tomograms of these composites.