Porous Media Classification Using Multivariate Statistical Methods

The Earth’s subsurface consists of porous media (i.e., rocks, soils) with vastly varied internal structures and properties. Characterizing the physical properties of porous media is an important activity for geologists, geotechnical, structural, environmental, and petroleum engineers. For instance, porous media's fluid flow properties are essential information for many natural and industrial processes such as groundwater movement, oil extraction, and geologic CO₂ sequestration. While porous media characterization is a complex process that involves laborious lab experiments or computationally expensive computer simulations, classifying the type of the porous media (e.g., sandstone, carbonate) often provides a preliminary estimate of the physical properties of interest. Here, we apply principal component analysis (PCA), partial least squares (PLS), and orthogonal partial least squares (OPLS) methods in conjunction with discriminant analysis to categorize porous media samples based on pore features extracted from micro-CT scans. We find that OPLS is the most efficient method by providing a more reduced form of data while having higher predictability to the porous media sample type when used with discriminant analysis. Specifically, OPLS reaches a classification accuracy of 97.17% on the testing datasets. It also provided a surrogate tool to study the key characteristics defining the porous media sample and to analyze the samples' homogeneity, which is one of the key characteristics that drive a porous media sample physical properties, including, but not limited to, its permeability of a fluid flow.