Characterizing Fracture Asperity Distribution in the Laboratory with Internal Seismic Sources and Triplet Neural Networks

ABSTRACT: Reliable energy production and anthropogenic waste isolation depend on the ability to accurately gauge fluid flow through fractures in rock. Direct measurements of fracture properties within deep subsurface reservoirs are not possible. However, remote sensing of natural and induced seismicity represents a promising approach because seismic waves contain information on the complex media through which they propagate. Here, we study the ability to extract information on the asperity distribution in a single fracture using passive monitoring of waves from sources internal to the fracture. A synthetic parallel-plate fracture was created from acrylic blocks separated by acrylic rods that represented asperities. Chattering dust was used to generate numerous acoustic emissions from within the fracture. A triplet neural network (TNN) was applied to the recorded signals for dimensionality reduction. The TNN identified features of the spatial frequency of the periodic asperity distributions that occurred at the arrival time of a Scholte wave. The work presented here demonstrates the potential for identifying changes in fracture geometry by evaluating the scattered fields from signals generated internally in a fracture.