Variational segmentation of textile composite preforms from X-ray computed tomography

Abstract

Prediction of the thermo-mechanical behavior of woven composites necessitates a reliable knowledge of their inner structure. A sufficiently accurate description of the fabric geometry could be obtained using X-ray computed microtomography (µCT) at the mesoscopic scale. However, systematic construction of numerical models from µCT remains a difficult task. To address this challenge, we propose a variational segmentation approach which combines µCT with a prior geometric model that is iteratively improved thanks to a heuristic optimization process. The fidelity of the models with respect to the input µCT is evaluated using a measure of similarity including both gray levels and local directions. Our method allowed to build realistic numerical models of woven fabrics that preserve the prescribed weaving pattern, and that are free of interpenetration, which makes them compatible with further numerical simulations. Using our approach, models of complex woven fabrics, but also of woven composites, could be consistently generated from µCT and can serve as reference models, e.g. to analyze in situ tests by pro-*

A review of applications of CT imaging on fiber reinforced composites

“…A global overlapping stack filtering step was used followed by a local fiber tracking step. The new algorithm was found capable of efficiently defining fiber centerlines for the generation of micro-scale finite element 130 models with high fidelity. Figure 14 showed micro-scale finite element models from synchrotron XCT images for multidirectional carbon fiber reinforced composites.…”

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

“…The idea of matching an idealized geometry with the real image or its semantic segmentation can be seen as a variation of the 3D active contour model [23]. In [24], individual tows are segmented by exploiting the result of an idealized geometric model optimization, which fitted the voxelization of the tows' surface to the real image. This approach guarantees a clean and topologically correct result, which can then be used for a finite element mesh generation.…”

Descriptive Modeling of Textiles using FE Simulations and Deep Learning

“…Other novel descriptive approaches based on textile models have recently been proposed [22,23]. In both cases an a priori 3D model of the textile is progressively deformed (optimized) so as to conform to the observed sample.…”