A two-stage statistical reconstruction method for characterization of anisotropic Cu–W composites from 2D cross-sections

This paper introduces a novel two-stage statistical reconstruction technique for generating realistic, anisotropic, three-dimensional Representative Volume Elements (RVEs) of multiphase, heterogeneous materials from perpendicular two-dimensional microscopy data. The proposed three-dimensional reconstruction method can generate 3D representations of multi-phase heterogeneous microstructures (including systems with more than three phases) featuring complex anisotropic patterns, while simultaneously accurately accounting for both phase clustering and agglomeration. Moreover, the present method can be employed for 3D reconstruction based on any microstructural statistical descriptor. The technique, employing Cellular Automata and Monte Carlo methods, accurately preserves critical microstructural features, including two-point correlation functions, phase distribution, morphology, and volume fractions, minimizing errors in the full two-point correlation function. This approach effectively bridges the gap between 2 and 3D analyses by enabling accurate meshing and high-resolution finite-element simulations (using tools such as Avizo, Ansa, and Abaqus) to predict anisotropic electrical and thermal conductivity. To ensure reliability, a detailed investigation of RVE size effects was carried out, with particular focus on anisotropic microstructures, verifying both numerical convergence and statistical representativeness within the finite element framework. The proposed workflow was experimentally validated using two copper–tungsten (Cu–W) composite samples, showing consistency between the 3D simulation results and measured electrical and thermal conductivity data. While AI-based reconstruction methods are powerful, they typically require large training datasets and intensive computation. In contrast, the technique presented here delivers high accuracy with minimal data input, making it especially suitable for applications where experimental data is limited or difficult to obtain.