Active yarn meshes for segmentation on X-ray computed tomography of textile composite materials at the mesoscopic scale

Abstract

For the past decade, the composite material community has been struggling to bridge the gap between the X-ray tomography of fiber composites and the realistic modelling of their internal structure at the mesoscopic scale. This paper proposes a new yarn segmentation method relying on a variational approach with deformable 3D surfacic meshes representing the yarn envelopes. It requires the previous localisation of yarn centroids. We apply five forces directly applied on the mesh vertices to drive the mesh fitting towards the visible yarn edges on the tomographic volume. The five forces enable a compromise between the intrinsic aspect of the envelope and the attachment to the yarn contours and regions, while limiting inter-yarn overlap. Results are presented on a X-ray tomography of a non-crimp fabric composite with unidimensional fiber bundles and on a woven composite tomography.