The availability of high-resolution 3D digital rocks in ever increasing quantities calls for intelligent Machine Learning (ML) techniques to classify them according to diverse characteristics of their pore structures. If stable classes could be identified, they would aid us to develop better models for rock typing, to gain sounder understanding of the links between the pore structures and the fluid flow behaviours and to develop predictive models of effective flow properties with many potential applications in the petroleum industry and beyond. We reported an approach that the authors developed for classifying digital samples. There, the pore structure is characterised by topological and geometrical attributes obtained from topology-preserved pore networks for each sample. Each attribute is then represented as a lst-order tensor and normalised so that it is comparable for images sampled at different scales and resolutions. Machine learning techniques are then used to carry out actual classification from a training dataset containing labelled and unlabelled samples. The viability and extendibility of this approach are discussed. We show that this approach can be implemented to classify samples in progressive, recursive and regressive manners, and can be extended to develop correlation between the classes of samples and their fluid flow properties.
展开▼
