Abstract:

Digital rock technology enables the precise digital characterization of core samples and facilitates the study of microscale rock physical properties through numerical simulations. Unconventional reservoir rocks display distinct features across various scales, and multiscale imaging technology can capture the rock’s microstructure at resolutions ranging from sub-nanometer to millimeter levels. However, single-resolution scanning methods fail to resolve cross-scale structural information, making the development of multiscale, multiresolution, and multicomponent digital rock models crucial to overcoming this limitation. Existing multiscale digital rock modeling methods can be broadly categorized into two main approaches: image fusion modeling, which relies on mixed overlays, template matching and deep learning, and pore network integration modeling, which incorporates explicit micropore networks, additional throat networks, and fracture systems. The image fusion approach accurately represents the three-dimensional distribution of pores and minerals across various scales and supports multiphysics simulations. However, its computational efficiency constrains its ability to manage large-scale discrepancies in hybrid modeling. Conversely, the pore network integration approach allows for modeling across multiple contiguous scales, requires less storage space, and offers high numerical simulation efficiency, although it is limited to certain physical properties. Moreover, digital rock workflows still face challenges, such as the precise extraction of minerals and the determination of suitable representative elementary volumes. Future research should focus on optimizing models using experimental data, studying physical properties as needed, and integrating homogenization and equivalent theory modeling to develop specific application systems that enhance well-logging interpretation and hydrocarbon reservoir development.

Key words: digital rock, multiscale, 3D stochastic reconstruction, image fusion, pore network model