Meso-scale permeability prediction from the multi-block 3D reconstruction of fiber reinforced polymer composite with experimental validation

Estimating the permeability of fiber-reinforced polymer (FRP) composites for liquid composite molding is particularly challenging due to the multi-scale porous nature of yarn layups. Methods using X-ray computed tomography (XCT) and geometrical reconstruction for numerical prediction of permeability particularly experience challenges in creating a realistic 3D porous computational mesh that accurately represents the yarn tortuosity while ensuring acceptable computational times. This work presents a methodology for developing a realistic dual meso/micro-scale yarn permeability model, addressing traditional limitations through segmentation, pre-processing, flow mesh creation, and permeability prediction. Structure-tensor analysis determines the yarn orientation, allowing for segmentation and generation of finite element meshes. The estimated fiber volume fraction (V_f) allows integration of intra-yarn permeability without resolving individual fibers and thus enables large-domain flow simulations. Dual-scale flow simulations, incorporating intra- and inter-yarn porosity and directional permeability, predict the three principal permeability components (K11, K22, K33) of FRP composites. The study further evaluates the domain dependence selected segmentation methods, the effects of mesh density reduction through voxel skipping and the impact of wall boundary conditions on unit cells. Computational time is reduced to 15% and 2% respectively for 8-fold and 64-fold mesh reductions while the corresponding deviations in permeability predictions range from 20%–30% and 45%–55% respectively. Despite these deviations, changes in V_f remain below 1.5%, and the model predictions align well with the experimental measurements of K11, K22, and K33. A robust framework is thus demonstrated, balancing prediction accuracy and computational efficiency, and practical guidelines are offered for XCT-based modeling of large-domain FRP composites.