Stochastic multi-scale modelling of composite materials based on experimental data

ariability in textile composites is substantial. A full characterisation of the internal geometry supports material design and increases the reliability of numerical simulations. An approach is presented and tested to obtain a full statistical description of the internal structure of a carbon-epoxy 2/2 twill woven composite, produced using resin transfer moulding. In a first step experimental data are collected using micro-CT to characterise the short range, i.e. within unit cell dimensions, and optical imaging for the long range, spanning several unit cells. The statistical information of the tow reinforcement is quantified per tow type, warp and weft direction, in terms of standard deviation and correlation length. The largest variation is found for the in-plane tow centroid coordinate with a significantly different behaviour in warp and weft direction. Tow cross-sectional properties such as area and aspect ratio vary within the unit cell size with limited scatter. Cross-correlation between neighbouring tows is only observed for the in-plane position. In a second step the statistical data are the input for a multi-scale modelling strategy to generate virtual specimens with random tow reinforcement. While the mean trends of each tow property are directly taken from the experiments, the deviations are generated by two different geometry generators. The properties without cross-correlation are produced using a Markov Chain algorithm, while the cross-correlated in-plane coordinate is generated using the Karhunen-Loève series expansion. Virtual specimens are obtained which possess the same statistical information as present in the experimental samples.status: publishe