Carbon fibre reinforced polymers (CFRPs) have a well-established role as structural engineering materials, especially in the aerospace industry. High-resolution micro-CT can be used as a non-destructive tool to study the 3D textile architecture of composites at the meso-level without the use of contrast enhancement agents (i.e. uncontaminated). In addition, correct sample preparation (e.g., casting the sample in resin to get cylindrical shape to avoid reconstruction artefact) and advance image segmentation is needed to get the best results from the image. It was found that the Histogram of Oriented Gradients (HOG) gave the best segmentation outcome when the specimen was sized to fit at least two voxels within a fibre width.
The application of these results will assist researchers in better understanding the evolution of microcracks and damage in textile composites while enabling physics-based multiscale modelling approaches to be validated with realistic textile architectures.
Figure 1. Flowchart summarising the key steps when implementing HOG to CFRPs. A small (3-by-3-pixel window is shown for demonstration of the problem at each step)
Figure 2. Ross plot showcases the gradient at different angle computed by the HOG algorithm
The Histograms of Oriented Gradients (HOG) algorithm was initially developed as a feature extraction algorithm used for human detection. It computes the magnitude of gradient at specified angular intervals between 0 and 180 over a window based on the fibre texture. The image is then classified into different tows based on the result of HOG algorithm (Figure 3). The CT images that were processed in 2D planes are separated by one pixel from the next, and this allowed all the 2D segmented images to be stacked and converted to 3D (Figure 4).
Figure 3. Example of segmentation of the tomograms for specimen 4 using HOG. Note; the layers presented starting from the back are: a raw image from the CT, 2D segmentation of a slice using HOG, an image of the same slice after conversion to a 3D
Figure 4. Reconstructed and segmented images of a CFRP specimen (T300/3900-2) at 0% failure strain