​International Journal of Rock Mechanics and Mining Sciences | Quantifying lamina-controlled variations in rock physical properties from AI-reconstructed 3D multimineral digital rocks

By  李勇    2026-10-15    Visited 10 times

Abstract

Laminated fine-grained porous media commonly consist of micrometer-scale, multimineral laminae, including felsic-mineral, clay-mineral, carbonate-mineral, and mixed-mineral laminae. Because these laminae and associated lamina-parallel fractures occur at small scales and involve complex mineral mixing, conventional laboratory and imaging workflows often struggle to quantify how lamina architecture and lamina fractures influence bulk physical properties. Digital rock analysis provides an effective pathway to address these limitations, yet robust reconstruction of 3D, large-scale, and multimineral laminated digital rocks from 2D images remains challenging. In this study, we propose a hybrid workflow (2D-3DGAN-MPS) that integrates a 2D-to-3D Generative Adversarial Network (2D-3DGAN) with Multiple-Point Statistics (MPS) to reconstruct large-scale 3D multimineral laminated digital rocks from 2D imagery. The 2D-3DGAN is used to generate 3D digital models for individual lamina types, and MPS is then applied to upscale the model size. Two or more lamina types with prescribed thicknesses are subsequently fused to form laminated digital rocks with and without lamina fractures. Using these reconstructions, we systematically evaluate the impacts of lamina combination style and lamina fractures on porosity, permeability, and brittleness. For systems containing two lamina types, the felsic-mineral plus clay-mineral configuration yields the highest porosity, permeability, and elastic moduli, and these properties improve with increasing felsic-lamina thickness. Across the reconstructed samples, porosity and permeability follow a power-law relationship, whereas the brittleness index varies linearly with brittle-mineral content. Particularly, the fitted relationships differ substantially between fractured and unfractured laminations. Lamina fractures reduce the energy threshold for fracture propagation, resulting in higher apparent brittleness. The proposed workflow enables efficient predictive models for permeability and brittleness in laminated fine-grained reservoirs, with practical implications for identifying sweet spots and improving digital-rock-based property evaluation.

Paper Information:

Wu, Y., Ao Y., Yang J, et al. Quantifying lamina-controlled variations in rock physical properties from AI-reconstructed 3D multimineral digital rocks. International Journal of Rock Mechanics and Mining Sciences, 2026, 207: 106691. https://doi.org/10.1016/j.ijrmms.2026.106691