Multiscale modeling of woven composites using inhomogeneous finite elements

Abstract Simulating the behavior of woven composites is challenging due to their complex microstructure, which often leads to discrepancies between predicted and actual performance. Traditional modeling approaches require labor-intensive preprocessing and a large number of elements to capture the detailed interactions between fibers and the matrix. This study introduces a novel multiscale modeling method that uses inhomogeneous finite elements to overcome these limitations. The approach leverages analytical surface equations to classify each element’s integration points as either matrix or fiber bundle, effectively representing the microstructure without the need for explicit meshing of the intricate yarn geometry. For the fiber bundles, a mean-field multiscale homogenization approach is applied to accurately model their behavior from the microscale. The proposed method offers significant advantages in both efficiency and accuracy. By using a structured mesh with a minimal number of hexahedral elements, it substantially reduces computational time and resources. Furthermore, it eliminates time-consuming preprocessing tasks, such as explicit CAD modeling and meshing, which can take hours and require specialized expertise. The method was validated by applying six independent loading cases (three uniaxial tension and three pure shear) to obtain the full orthotropic homogenized response. Results show high accuracy, with predicted axial moduli and Poisson’s ratios deviating by only about 5% from conventional methods, and shear moduli predictions within 10%. By concealing the microstructure in the preprocessing stage, this approach allows for efficient analysis, with the inhomogeneous nature of the material becoming apparent only after the solution is obtained. This makes the proposed method a highly efficient and accurate alternative for the analysis of woven composites.