Deep Learning–Based 3D Multi-Mineral Digital Rock Modeling and Diagenetic Simulation

The construction of three-dimensional multi-mineral digital rock cores is essential for the fine characterization of reservoir pore–throat structures and for the quantitative evaluation of reservoir electrical, acoustic, and mechanical properties.Existing digital rock core modeling approaches can be broadly classified into physical experimental methods and numerical reconstruction methods.Physical experimental methods can produce relatively realistic three-dimensional digital rock cores; however, they are costly and struggle to achieve fine mineral discrimination.Numerical reconstruction methods offer advantages such as low cost and high efficiency; however, most high-fidelity approaches remain limited to single-mineral digital rock cores, whereas multi-mineral modeling methods often rely on idealized assumptions.To address these limitations, this study proposes a 2D–3DGAN-based deep learning algorithm capable of generating three-dimensional multi-mineral digital rock cores from a single AMICS image.Using the “AMICS + 2D–3DGAN” modeling framework, three-dimensional multi-mineral digital rock cores are constructed with high accuracy, efficiency, and realistic multi-mineral representation.The accuracy of the generated results is systematically evaluated by analyzing diagenetic characteristics and by comparing the generated cores with training images in terms of mineral content, pore size distribution, and two-point correlation functions.The results demonstrate that the proposed method significantly enhances reconstruction accuracy and generation efficiency while maintaining economic feasibility, thereby providing a solid foundation for subsequent simulations of multi-mineral diagenetic evolution and reservoir property analysis.Previous studies on diagenesis have largely relied on qualitative approaches, such as identifying diagenetic evolution sequences and constructing two-dimensional schematic representations.Based on the generated three-dimensional multi-mineral digital rock cores, this study proposes a “nucleation–growth” algorithm for numerically simulating diagenetic processes, enabling quantitative modeling of diverse cementation morphologies that closely reflect geological conditions.Meanwhile, the four-parameter structure method is improved to quantitatively simulate dissolution and replacement processes with varying morphologies and to control the precipitation location of dissolution by-products, either inside or outside dissolution pores.Ultimately, a three-dimensional digital rock core diagenetic evolution model based on diagenetic sequences is established, enabling analysis of the evolution of reservoir properties—such as porosity, permeability, tortuosity, and coordination number—through diagenetic processes.