Uncertainty quantification of mechanical properties for three-dimensional orthogonal woven composites. Part I: Stochastic reinforcement geometry reconstruction

For woven composites, the stochasticity of mechanical properties is mainly dependent on the reinforcement variability. Represent Volume Elements with realistic reinforcement architecture can obtain accurate predictions and identify the variability of mechanical properties. This work presents a data-driven modeling framework to generate statistically equivalent RVEs for three-dimensional orthogonal woven composites, in which retaining the knowledge of reinforcement variability from experimental observation. The reinforcement geometry is characterized by Micro CT in terms of fiber tow centroid coordinates and cross-sectional dimensions. A comprehensive slicing and data correction method, which transforms the intact 3D geometry into a four-dimensional dataset, is established for all tow genera. A quantitative understanding of the variability of reinforcement architecture is presented via various statistical descriptors. In order to preserve the mainly statistical characteristics of tow feature parameters, D-vine copula functions are adopted to address their irregular marginal distributions and joint dependence. The reconstructed data is verified by the marginal probability density functions, the bivariate distributions and the correlation matrices of feature parameters. An inverse design from data to textile geometry is achieved by dedicated codes in TexGen. The whole framework is driven by acquired data automatically and can generate any number of statistically equivalent RVEs for further simulations.