GSA Connects 2022 meeting in Denver, Colorado

Paper No. 215-8
Presentation Time: 11:25 AM

A NOVEL STOCHASTIC PORE-SCALE SIMULATION APPROACH FOR PERMEABILITY PREDICTION IN HETEROGENOUS POROUS MEDIA FROM MICP AND/OR MICRO-CT DATA


ISHOLA, Olubukola, Boone Pickens School of Geology, Oklahoma State University, Stillwater, OK 74078 and VILCAEZ, Javier, Boone Pickens School of Geology, Oklahoma State University, 105 Noble Research Center, Stillwater, OK 74078

Estimating rock permeability is vital in resource exploration, resource production, and environmental management. Due to the time and/or resource required to measure permeability, quick permeability estimates is usually obtained from model equations that relate permeability to other pore microstructural features such as pore-throat size distribution (PTSD), pore size distribution (PSD), and/or porosity. While these approaches might perform excellently for homogenous media (often siliciclastic rocks and soil materials), it might not always suit heterogenous porous media such as carbonate rocks. For instance, a single porosity value with a fixed PSD can be associated with permeability values that could range several orders of magnitude. This is because these approaches do not consider other pore microstructural parameters, especially pore connectivity. In this study, we used a novel stochastic pore-scale simulation approach that requires generation of hundreds of 3D pore microstructures of the same PSD and porosity but different stochastic pore connectivity. PSD and porosity information of rock samples to implement the proposed pore-scale simulation approach were obtained from mercury injection capillary pressure (MICP) and Micro-CT data while the pore shape was simplified to spherical geometry. We tested the approach on four carbonate and five siliciclastic rock cores. Preliminary results show that permeability estimation from our approach is more accurate than well-known model equations of permeability examined in this study. In a second phase of the study, machine learning was used to reduce the computational power needed for pore-scale numerical simulations by a factor of 157 times while having a mean absolute percentage error (MAPE) of only 10%; making it easy to routinely implement our approach in practice.