Fractures in subsurface are omnipresent and play a critical role both in the host rock strength and stability, and in flow through subsurface. This is especially the case in unconventional reservoirs where hydraulic fractures connect with a complex natural fracture for effective exploration. Discrete fracture networks (DFNs) are computationally efficient models that explicitly represent fracture network as a network of planar polygons in 3D. Most DFN approaches build networks as stochastic realizations of planar polygonal shapes in a large volume. The distribution of properties such as average fracture apertures, density, or fracture orientations is typically obtained from ID or 2D borehole observations. An inherent disadvantage of these methods is that they do not contain relevant detailed information of the actual three-dimensional spatial distribution and connectivity of fractures and thus cannot be used for model-experiment validation of real fracture flow in experiments imaged with x-ray computed (micro)tomography (CT or micro-CT).
Currently, there are no fully automated algorithms available to construct an image-based fracture network from a 3D image. This is principally due to the lack of algorithms capable of reading a segmented image (i.e. image with identified void and solid spaces), extracting the individual fractures from the network, and quantifying their geometric properties. The fundamental question is defining where one rough fracture starts, and another ends. We address this here by combining advanced imaging and geostatistical tools to understand and model diverse geometries from 2D and 3D micro-CT images of a Mancos shale. The segmented volumes are the starting point to isolate the fractures from the matrix. Then, unsupervised machine learning algorithms are implemented to cluster the points that belong to the same fracture. Finally, positional and geometric properties are extracted to serve as input data for the DFN software. The workflow proposed here can be a successful alternative to generate discrete fracture networks from micro-CT images.
