Accurate characterization of fibrous microstructures is essential for the design and optimization of nonwoven materials and filter media, but conventional methods are often time-consuming and labor-intensive. This work presents a non-destructive approach that combines X-ray computed tomography (µCT) with AI-driven analysis within a material Digital Twin framework to enable efficient and reproducible characterization.

Three-dimensional µCT data are processed in the GeoDict software, including preprocessing (cropping, rotation, optional image enhancement) and fiber segmentation and quantification based on the approach of Griesser et al.[1]. A novel 3D U-Net neural network is used to detect fiber centerlines. Trained on synthetically generated fiber structures with known ground-truth, the network provides an analytical description of individual fibers, including diameter, orientation, and curvature. These parameters are used to generate a fully parameterized Digital Twin of the nonwoven microstructure using GeoDict’s FiberGeo module.

The Digital Twin is subsequently used to predict key performance metrics such as filtration efficiency and air permeability. The simulation results show strong agreement with predictions based directly on the original µCT-derived structures, demonstrating the accuracy and reliability of the proposed approach. The method enables rapid, quantitative, and non-destructive microstructural analysis and supports efficient material optimization and performance prediction.