Fast X-ray Micro-Tomography of Multiphase Flow in Berea Sandstone: A Sensitivity Study on Image Processing

Synchrotron-based fast micro-tomography is the method of choice to observe in situ multiphase flow and displacement dynamics on the pore scale. However, the image processing workflow is sensitive to a suite of manually selected parameters which can lead to ambiguous results. In this work, the relationship between porosity and permeability in response to systematically varied gray-scale threshold values was studied for different segmentation approaches on a dataset of Berea sandstone at a voxel length of 3  \(\upmu \) m. For validation of the image processing workflow, porosity, permeability, and capillary pressure were compared to laboratory measurements on a larger-sized core plug of the same material. It was found that for global thresholding, minor variations in the visually permissive range lead to large variations in porosity and even larger variations in permeability. The latter is caused by changes in the pore-scale flow paths. Pore throats were found to be open for flow at large thresholds but closed for smaller thresholds. Watershed-based segmentation was found to be significantly more robust to manually chosen input parameters. Permeability and capillary pressure closely match experimental values; for capillary pressure measurements, the plateau of calculated capillary pressure curves was similar to experimental curves. Modeling on structures segmented with hysteresis thresholding was found to overpredict experimental capillary pressure values, while calculated permeability showed reasonable agreement to experimental data. This demonstrates that a good representation of permeability or capillary pressure alone is not a sufficient quality criterion for appropriate segmentation, but the data should be validated with both parameters. However, porosity is the least reliable quality criterion. In the segmented images, always a lower porosity was found compared to experimental values due to micro-porosity below the imaging resolution. As a result, it is recommended to base the validation of image processing workflows on permeability and capillary pressure and not on porosity. Decane-brine distributions from a multiphase flow experiment were modeled in a thus validated \(\upmu \) -CT pore space using a morphological approach which captures only capillary forces. A good overall correspondence was found when comparing (capillary-controlled) equilibrium fluid distributions before and after pore-scale displacement events.