From X-ray CT to finite element models: A fully automated pipeline for mesoscale modelling of as-manufactured textile composites

Artikel i vetenskaplig tidskrift, 2026

Structural parts incorporating 3D-textile reinforced composites show great promise in high performance lightweight applications. For widespread industrial use, accurate predictions of mechanical properties are required. Experimental testing campaigns to generate these properties can be prohibitively expensive. To overcome this, meso scale models of the yarn architecture can be derived from X-ray Computed Tomography (XRCT), and computational homogenisation can be performed in the material’s as-manufactured configuration. In this work a fully automated pipeline for the prediction of the full 3D elastic properties of 3D-reinforced textile composites from XRCT scans is presented. The proposed methodology enables the study of variations in the as-manufactured material properties from a single large field of view XRCT scan encompassing multiple unit cells, such that finite element homogenisation can yield statistical information. The pipeline includes a machine learning based segmentation model, finite element meshing and boundary condition assignment, and a material mapping procedure. For segmentation, a for textile reinforced composites completely novel 3D U-net architecture can be utilised, owing to the use of a fully synthetic automatically labelled training data set. An application of the pipeline on a 3D-reinforced material sample results in accurately predicted homogenised elastic stiffnesses, with a deviation from experiments of less than 6.5%.