This contribution introduces an enhanced cell selection strategy for accurately determining the effective heat conductivity of composites with stochastic microstructures, leveraging analysis of micro-computerized tomography ( \(\mu \) CT) images. The strategy employs statistical features and porosity analysis to optimize cell (or RVE realization) selection, ensuring high accuracy and reliability of results with a reduced number of realizations. The approach utilizes asymptotic homogenization (AH) for upscaling the heat transfer problem, employing systems of partial differential equations (PDEs) known as cell problems. We use finite element (FE) method to solve the cell problem within selected cell domains to calculate the effective heat conductivity, characterizing the homogenized system. The employed AH-based multiscale methodology is suitable for stochastic microstructures as it avoids the need for imposing boundary conditions (BCs) at interfaces. The workflow encompasses image preprocessing, segmentation to identify pores (voids) and solid matrix, periodicity establishment, and mesh generation to create a computational domain suitable for the upscaling process. Statistical descriptions of \(\mu \) CT images provide crucial microstructural characteristics that inform the RVE selection process. A representativity analysis demonstrates significant improvements over standard random or equidistant cell selection techniques.
