Pore-scale machine-learning modeling of flow and transport properties of carbonate rocks

Carbonate rocks constitute one of the most common type of groundwater and fossil fuel reservoirs in the USA. Design of optimum strategies for groundwater and fossil fuel resources development is generally done through the application of computer simulation programs where users need to input a set of flow and transport properties of the reservoir rock.

The traditional approach to estimate flow and transport properties of reservoir rocks consists of using deterministic model equations. This approach frequently results in large disagreement between measured and simulated multiphase flow and reactive transport processes in carbonate reservoirs. The complex heterogeneous pore microstructure of carbonate rocks is reflected by stochastic (random) relationships between fundamental petrophysical properties (e.g., porosity) and flow and transport properties (e.g., permeability and tortuosity) of carbonate rocks that are practically impossible to capture by deterministic model equations.

The goals of this research are (1) to develop a predictive understanding of the stochastic relationship between the pore microstructure and flow and transport properties of carbonate rocks, and (2) to establish machine learning models that capture the stochastic relationship between fundamental petrophysical properties and flow and transport properties of carbonate rocks. This research will advance national welfare and prosperity by enabling the design of environmentally responsible and optimum strategies to develop groundwater and fossil fuel resources from carbonate reservoirs, and it will contribute to the education of students on the use of artificial intelligence (e.g., machine learning) technologies to develop groundwater and fossil fuel resources.

The goals of this research will be achieved by using a novel approach that overcomes limitations regarding the relatively small number of pore microstructures and associated flow and transport properties of carbonate rocks that can be experimentally determined by routine and special analyses. This novel approach consists of the construction of thousands of 3D pore microstructures of stochastic pore connectivity honoring pore size distribution curves obtained from nuclear magnetic resonance (NMR) measurements and pore geometries obtained from 2D scanning electron microscopy (SEM) imaging.

Direct pore-scale simulations of flow and transport properties for thousands of 3D pore microstructures of the same pore size distribution and dominant pore geometry, but different pore connectivity, as it happens in real carbonate rocks, will make possible achieving the goals of this research by applying statistical/stochastic principles and machine learning technologies. The focus will be on carbonate rocks collected from the Mississippian Lime Play and Arbuckle Group in Oklahoma and Kansas.

This project is jointly funded by the Hydrologic Sciences Program (HS) and the Established Program to Stimulate Competitive Research (EPSCoR).

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.