Characterization of heterogeneous reservoir rocks using multi-scale image reconstruction and octree structures
Loading...
Date issued
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Reuse License
Abstract
Researchers commonly use multi-scale modeling to characterize heterogeneous rock samples. Despite their utility, these models face significant challenges. One major issue is the tradeoff between resolution and field of view (FoV) during imaging. Although various novel approaches attempt to address this issue, they often produce unrealistic results or incur high computational costs. In this study, we present methods for the multi-scale reconstruction of pore networks and images for three heterogeneous rocks using octree structures. We performed multi-scale image reconstruction on Berea Sandstone (BS) and Edward Brown Carbonate (EBC) rocks, and multi-scale reconstruction of pore networks on Indiana Limestone (ILS) rock. We scanned all samples at low (LR) and high (HR) resolutions using X-ray microtomography. In our novel multi-scale direct numerical simulation (DNS) reconstruction approach, we reconstructed multi-scale images using the Octree structure to reduce computational cost. In addition, we also presented a novel method for multi-scale pore network modeling (PNM) reconstruction using an artificial neural network to connect the PNMs at different scales in octree structure. The results showed good agreement with the HR images and experimental properties. We concluded that using the Octree structure decreased runtime up to three times for the multi-scale DNS reconstruction. For multi-scale PNM reconstruction, the runtime was nearly equal to the normal method. Additionally, the memory used for the multi-scale DNS reconstruction was reduced by 20 to 130 times, and for multi-scale PNM reconstruction, by up to three times. To further reduce computational costs, we also introduced new multi-scale upscaling methods. The first method, which utilized the Octree structure and unresolved clusters, was applied to the BS rock images for multi-scale DNS upscaling. The second method was applied to the ILS rock images for multi-scale PNM upscaling. We concluded that the multi-scale DNS upscaling method, while maintaining accuracy, reduced runtime up to eight times and memory consumption up to two times. Similarly, the multi-scale PNM upscaling method reduced runtime up to two times and memory consumption about three times, without reducing accuracy.