Speaker
Description
Accurate modelling of fluid flow in multi-scale porous media, such as carbonate rocks, is hindered by the inherent trade-off between the field of view and the resolution in imaging technologies, complicating the characterization of pore structures across multiple length scales. Microporosity phases or unresolved regions on 3D X-ray computed tomography (micro-CT) images contain nanometer-scale pore throat structures that can be fully resolved in scanning electron microscopy (SEM) images where only 2D information is available. To address this multi-scale imaging challenge, deep learning models have been developed to enhance the image resolution of 3D micro-CT images using information from SEM images. However, it remains unclear whether statistics derived from 2D rock cross-sections are sufficient to enable high-fidelity 3D digital rock modelling of heterogeneous and anisotropic samples. Furthermore, there is no established methodology for selecting and preparing rock samples for high-resolution imaging that ensures representative and uncertainty-aware digital rock models.
In this study, we utilize a data assimilation technique to develop a powerful image-based digital rock modelling framework for heterogeneous carbonate rocks and to guide optimal sample preparation for subsequent high-resolution imaging. Permeability and porosity for two 6 mm mini-plugs from different carbonate rock types were experimentally measured and imaged under X-ray micro-CT (voxel size 3 μm) and SEM (voxel size 0.5 μm). First, we implemented a deep learning super-resolution algorithm to build a high-resolution 3D digital rock model using the acquired images. Subsequently, we utilized the ensemble smoother with multiple data assimilation (ESMDA) algorithm to constrain and assess the uncertainty of each microporosity phase property. Specifically, a conditional GAN (cGAN) model coupled with our open-source eXtensive Pore Modeling XPM (https://github.com/dp-69/xpm) simulator enables efficient memory management during ESMDA regression. Compared to pure image-based deep learning algorithms, the developed ESMDA-assisted digital rock modelling achieves better accuracy when validated against experimental measurements and unseen SEM images. More importantly, the uncertainty estimates of each microporosity phase properties obtained during ESMDA regression can be leveraged to identify phases requiring further data acquisition, thereby optimizing the subsequent sample preparation strategies. Herein, our proposed workflow provides a viable option for high-fidelity digital rock modelling of multi-scale carbonate rocks.
| Country | United Kingdom |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
