Rock Reconstruction with Deep Generative Network
Qinglong Cao, Yuntian Chen · 2024
The reconstruction of Digital Rock is a crucial challenge in understanding the microstructure of rocks and its impact on pore-scale flow through numerical modeling. This is particularly significant due to the typically large samples required to address inherent uncertainties. Despite notable advancements in traditional process-based techniques, statistical methods, and recent popular deep learning models, there is a limited focus on deep learning approaches specifically tailored for reconstructing rocks with predefined properties, such as porosity. To address this gap, our research employs Artificial Intelligence Generative Component (AIGC) technologies to precisely generate rock structures with specified properties. Our experimental results demonstrate the successful application of our method in reconstructing rock images based on target properties. The generation of randomly reconstructed samples with distinct rock properties holds promise for advancing research in pore-scale multiphase flow and uncertainty quantification in subsequent studies.