Classification of Rock Types Using Machine Learning

Abstract

Determining the petrophysical rock type often excludes measured multiphase flow properties, such as relative permeability curves. This is due to limitations in SCAL experiments or difficulties in correlating relative permeability characteristics with standard rock types. However, with a significant number of relative permeability curves, Machine Learning methods can be applied to automatically and objectively classify rock types based on the shape of these curves. This approach combines principal component analysis with unsupervised clustering schemes and preprocesses relative permeability curve data by integrating irreducible water saturation and residual oil. The methodology was tested on real data from carbonate reservoirs with a substantial number of relative permeability curves, demonstrating successful clustering based on fractional flow curves. The results indicate that this clustering can classify rocks from poor to optimal displacement efficiency. Furthermore, the study highlights the importance of high-quality SCAL experiments for normalizing curves and ensuring consistency between capillary pressure measurements and relative permeability. This Machine Learning approach is also compared with capillary pressure analysis, showing that relative permeability data provides additional information in rock typing studies, affirming the feasibility of Machine Learning for automatic rock type classification.