Reducing User Bias in X-ray Computed Tomography-Derived Rock Parameters through Image Filtering

Autor: Ellen P. Thompson, Brian R. Ellis, Kira Tomenchok, Tyler Olson
Rok vydání: 2021
Předmět:
Zdroj: Transport in Porous Media. 140:493-509
ISSN: 1573-1634
0169-3913
DOI: 10.1007/s11242-021-01690-3
Popis: Accurate representation of pore space is essential for predicting fluid flow through subsurface porous media. Pore volume fraction, geometry, and topology determine transport characteristics at the pore scale and are used to make upscaled projections about reservoir behavior. X-ray computed tomography (XCT) allows for nondestructive 3D imaging of rock core samples and can therefore provide valuable information about the pore network in situ, but segmentation of XCT datasets into pore and mineral space is not trivial. In this study, three filters (contrast enhancement, noise reduction, and beam hardening correction) were applied to XCT datasets of rock core samples prior to training class definition for machine learning-based segmentation. Porosities derived from segmented datasets with and without filtering were compared and were validated with experimental values. XCT-derived porosity had reduced variance and was closer to experimental data when all three filters were applied. A case study of one rock core sample compared pore size distribution and simulated permeability to experimental data. Computational fluid dynamics simulations of flow through the pore network using OpenFOAM showed improved consistency in permeability values when all three filters had been applied. This suggests that the application of these filters prior to machine learning training class definition can improve the reproducibility of the segmentation results and reduce user bias, thereby increasing confidence in digitally derived rock parameters. Reliable initial porosity and permeability data are critical for improving fluid transport and fate projections in a broad range of subsurface systems. .
Databáze: OpenAIRE
Externí odkaz: