Modeling of 3D-printed composite filaments using X-ray computer tomography images segmented by neural networks

Abstract

This study introduces a novel semi-analytical two-scale approach to model 3D-printed composite filament where both fibers and voids are present. The proposed approach utilizes micro computer tomography (micro-CT) images to visualize fibers and voids. Then, a novel segmentation process based on neural network algorithms (specifically the YOLO v7 algorithm) is employed to segment the distinct fiber and void phases. The volume fractions and the geometries of the phases are then used in the first scale (the micro-scale) of the approach and the effective elastic properties of internal regions of the filament are obtained using the Mori–Tanaka effective field method. These regions are then joined together to form the second scale (the macro-scale), for which the elastic properties are evaluated using the finite element method. The two-scale approach is applied for 3D-printed nylon reinforced with continuous carbon fibers and the predictions of the elastic properties are compared to tensile tests conducted on the single filament. A great agreement is observed for Young’s modulus along the fiber direction of the single printed filament.