Geometrical and deep learning approaches for instance segmentation of CFRP fiber bundles in textile composites

Automated conformal mesh generation chain for woven composites based on CT-scan images with low contrasts

“…In the case of small contact zones, the tows can be separated by identifying isolated tow sections and by joining these sections across the contact zones [29]. Another possibility is to determine the planes between the plies by analyzing the matrix distribution in the REV [40].…”

Validation of Deep Learning Segmentation of CT Images of Fiber-Reinforced Composites

“…Creveling et al [19] recently validated their template-matching approach for segmenting fibers in graphite-epoxy composites by comparing with training using synthetic images that mimic the quality and resolution of CT images, including artifacts such as beam hardening and noise. The use of synthetic data avoids the effort of manual labeling of fibers and the human errors that might result [20]. Emerson et al validated their machine learning method for fiber segmentation using high-resolution optical and SEM images [21].…”

Instance segmentation of 3D woven fabric from tomography images by Deep Learning and morphological pseudo-labeling

“…This more accurate segmentation is known as instance segmentation, since it does not only seek which pixels belong to a yarn, but also to identify that there are several instances of yarns in each slice of the tomography. This kind of segmentation could be perform either directly from the neural network [44] or by combining semantic segmentation with a suitable post-processing [42,45].…”

“…Finally, very lately, Ali et al [42] and Sinchuk et al [45] proposed instance segmentation frameworks combining different DCNN for a first step of semantic segmentation followed by the watershed technique [47,48] during the post-processing, separating connected yarns.…”

“…Note that although some aforementioned works are devoted to perform instance segmentation only on dry fabrics [42,44,45], they are very useful and suitable to a better analysis of textile composites. Indeed, the composite reinforcement modeling at the mesoscale (in both predictive and descriptive approaches) improves the overall knowledge of textile composite manufacturing in assessing the mechanical response of the textile under various external loads [14,19,49,50] as well as the preform behavior during the forming [15] process.…”