PRIOR SHAPE ANALYSIS FOR YARN SEGMENTATION IN XCT IMAGES

This study addresses the automatic segmentation of textile reinforcements in fan blades using X-ray CT imaging at resolutions above 120 µm. The proposed approach is a scalable analysis orthogonal to the main yarn orientations (warp and weft), involving the identification of yarn centers and crosssections on a 2D plane.

The method statistically characterizes three properties: the standard shape of yarn cross-sections using Principal Component Analysis (PCA), the continuity of each yarn path, and their arrangement relative to neighbouring yarns. These properties are exploited in a variational formulation for optimal yarn positions, displacement from previous planes, and neighbour arrangements.

Yarn center positions are determined over successive planes perpendicular to the mean yarn orientation, successfully tracking around 1,500 yarns over 900 slices with a 95\% success rate for annotated yarns. Then the process iteratively refines segmentation by erasing and re-segmenting yarns. This method shows potential for large-scale textile annotation.