Permeability Prediction of 3-D Binary Segmented Images using Neural Networks

3-D digital rock image have recently become an important piece of information about the hydrocarbon reservoir properties. The goal of this project is to explore and employ machine learning as a tool to better understand the 3-D digital rocks from geometric measurement and features extracted from the image. Introduction This project applies machine learning to 3-D binary segmented images of the Fontainebleau and Berea sandstone, with the intention of finding a robust alternative way to estimate transport properties such as permeability. The permeability is one of the key to understand the nature of hydrocarbon reservoir and estimate its production capability. Conventionally, permeability is obtained from laboratory measurement of a rock core, which can take up to months to be completed. Recent technology in high resolution x-ray tomography have led to the increase in digital rock database. 3-D binary segmentation is applied after scanning. For single-phase fluid flow, the LatticeBoltzmann method is the established method for solving absolute permeability. The LatticeBoltzmann method approximates the Navier-Stokes equations at the pore scale. The calculation can be computationally expensive for large digital rock images. The calculation assumes no flow boundary conditions, and that the permeability depends only on pore geometry. In our calculations, the pressure gradient is only along the x-axis so that dP