Stochastic characterization of textile reinforcements in composites based on X-ray microtomographic scans

A stochastic method is introduced to characterize the dual-scale geometry of textile reinforcements in composites. The fiber tows are identified automatically from X-ray microtomographic scans with a machine learning algorithm, quantifying the error of the procedure. The tow geometry is then used to construct a stochastic model as a Gaussian Random Process which permits quantification of the uncertainty in the measurements of micro-scale fiber volume fraction. The hyperparameters of the model are calibrated with a custom-built multi-objective evolutionary algorithm. The approach is illustrated by the analysis of a vinyl-ester composite reinforced with a glass fiber non-crimp fabric.