Mei Li (1), Aly Abdelaziz (1), Earl Magsipoc (1), Johnson Ha, Karl Peterson (1), Giovanni Grasselli (1)
SSRN. (23 March 2023). DOI: http://dx.doi.org/10.2139/ssrn.4397765

Keywords

True-triaxial hydraulic fracturing, Montney shale, fracture network geometry, machine learning, high-resolution imaging

Abstract

Fractures are the main paths for fluid flowing in low matrix permeability rock formations and the fracture geometry determines the volume and rate of fluid extraction/injection. With the current work, we contribute to understanding how subsurface fracture complexity is generated through the quantitative characterization of the geometry of a shale fracture network artificially induced by laboratory hydraulic fracturing testing. The identified fracture regions (reactivated natural fractures, parted beddings, and hydraulic fractures) of the high-resolution digital shale fracture network were segmented into individual regions of interest using supervised learning algorithm with a U-Net architecture convolutional neural network model, which enables the geometrical characterization at the scale of both fracture network and fracture regions. The quantification of the fracture connectivity, volume, orientation, surface area, and aperture demonstrate that approximately 90% of the fracture network is associated to the reactivation/propagation of pre-existing discontinuities and over 70% are low-dip angle fractures. The results suggest that the pre-existing discontinuities, especially the beddings, greatly impact the orientation and propagation of the fluid-driven fractures and significantly contribute to the fracture network volume, thus resulting in favored lateral propagation of the fracture network rather than the expected vertical/across-bedding propagation. This observation could greatly impact the well completion strategies in the field and help guide the operations of subsurface activities.

How Our Software Was Used

Fracture regions of high-resolution digital shale fracture networks were segmented into individual regions of interest using Dragonfly’s Deep Learning tool. The quantification of the fracture connectivity, volume, orientation, surface area, and aperture were also computed in Dragonfly.

Author Affiliation

(1) University of Toronto