Assessing Parameterized Geometric Models of Woven Composites using Image-Based Simulations

TL;DR

This paper compares idealized analytical models and real-world as-manufactured woven composites using image-based simulations, revealing how mesoscale geometry affects effective thermal properties and the accuracy of predictive models.

Contribution

It introduces a methodology combining X-ray CT, deep learning, and finite element simulations to evaluate the impact of manufacturing heterogeneity on composite properties.

Findings

Analytical and sub-unit cell geometries predict thermal properties well.

Manufacturing heterogeneity causes local property variations.

Image-based simulations improve understanding of real composite behavior.

Abstract

Mesoscale simulations of woven composites using parameterized analytical geometries offer a way to connect constituent material properties and their geometric arrangement to effective composite properties and performance. However, the reality of as-manufactured materials often differs from the ideal, both in terms of tow geometry and manufacturing heterogeneity. As such, resultant composite properties may differ from analytical predictions and exhibit significant local variations within a material. We employ mesoscale finite element method simulations to compare idealized analytical and as-manufactured woven composite materials and study the sensitivity of their effective properties to the mesoscale geometry. Three-dimensional geometries are reconstructed from X-ray computed tomography, image segmentation is performed using deep learning methods, and local fiber orientation is obtained using the structure tensor calculated from image scans. Suitable approximations to composite properties, using analytical unit cell calculations and effective media theory, are assessed. Our findings show that an analytical geometry and sub-unit cell geometry provide reasonable predictions for the effective thermal properties of a multi-layer production composite.