Assessing the Systematic and Random Error From a Single Snapshot in Stochastic Homogenization of Industrial Scale Composites

Abstract

ABSTRACT We estimate effective linear elastic properties from a single microstructure image using a sliding‐window subsampling strategy applied to fields obtained by computational micromechanics. The method yields both mean properties and variance estimates without requiring multiple realizations. We compare against a classical RVE study with 7500 microstructures and demonstrate comparable accuracy at significantly reduced computational cost. Variance reduction based on volume fraction matching is shown to significantly reduce both random and systematic errors, despite strong correlations between samples.