Computational Intelligence in Stochastic Reconstruction of Porous Microstructures for Image-Based...

Understanding microstructure-property relationships (MPRs) in random porous media is a fundamental challenge across numerous scientific and engineering disciplines. Image-based poro/micro-mechanical modeling offers a powerful noninvasive technique to investigate MPRs via numerical simulations. However, the stochastic nature and inherent randomness of porous media necessitate extensive datasets of 3D digital microstructures for reliable statistical analysis. Stochastic microstructure reconstruction provides an efficient and cost-effective approach to generate large numbers of virtual microstructures using limited statistical information from real porous materials, establishing it as a critical tool for advancing research in this field. This review presents a comprehensive examination of stochastic reconstruction methodologies, spanning traditional algorithm-based methods and emerging computational intelligence-based approaches. Particular emphasis is placed on computational intelligence-based approaches, such as generative adversarial networks, while also discussing the foundational contributions and limitations of traditional methods. These advancements have significantly enhanced the fidelity, efficiency and scalability of microstructure reconstruction, enabling robust statistical investigations of MPRs, such as Monte Carlo analysis. Despite substantial progress, challenges such as data scarcity, high computational costs, limited interpretability, and the need for physically realistic reconstruction remain, especially for complex pore network systems. Emerging trends, including physics-aware machine learning and hybrid AI frameworks integrating domain-specific knowledge, offer promising avenues to overcome these limitations. By bridging disciplinary gaps, this review provides a roadmap for future research in stochastic microstructure reconstruction, facilitating deeper insights into MPRs and broadening applications across various scientific and engineering domains.