Benchmarking Virtual Permeability Predictions of Real Fibrous Microstructure

Abstract

ABSTRACT For fast and complete impregnation in Liquid Composite Molding, knowledge about the permeability of the fibrous reinforcement is required. While development of experimental methods continues, a parallel benchmark effort to numerically characterize permeability is being pursued. The approach was to send out the images of a real fibrous microstructure to a number of participants, in order for them to apply their methods for virtual permeability prediction. Via resin transfer molding a plate was manufactured, using the glass woven fabric Hexcel 01102 (295 g/m²) at a fiber volume content of 54% and a thermoset resin. From this plate, a specimen was scanned using a 3D x-ray microscope at a scan size of 1000 x 1000 x 1000 μm³ and a resolution of 0.521 μm³ per voxel. The sample extracted for the simulations with a size of 523 x 65 x 507 μm³ contains about 400 fibers of a single tow. It revealed a variation of filament diameters between 7.5-9.3 μm and a fiber volume content in average of 56.46% with a variation of 54 - 59% in the individual 2D-slices transverse to the fiber direction. The image segmentation was performed by 2D-slices, to which a Hough transform was applied to detect fiber centers and cross-sections. Then fiber paths were tracked through-out the slices by the closest neighbor algorithm. Finally, fiber paths were smoothened by means of the local regression using weighted linear least squares and a 1st degree polynomial model. The participants received a stack of 973 segmented (binary) 2D-images and a corresponding segmented 3D volume raw-file. They were asked to calculate the full permeability tensor components and fill out a detailed questionnaire including questions e.g. on applied flow models and conditions, numerical discretization and approximation methods, fluid properties etc. The received results scatter considerably over two orders of magnitude, although the participants were provided an already segmented image structure, thus eliminating from the beginning a significant source of variation that could have come from image processing. Model size, meshing and many other sources of variation were identified, allowing further specification of the guidelines for the next step.