Authors:

Alexis Nicole Hensley

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Geothermal energy production from the Earth's subsurface often relies on the flow of fluids through a network of fractures to extract heat. However, these fracture networks are sensitive to engineered and natural processes that open and close fractures in response to changes in stress, fluid pressures, and geochemical interactions. A challenge is how to link information in geophysical data to changes in the network permeability to enable updates to numerical simulators to predict and control long-term energy production. Here, we use machine learning (ML) to explore data from EGS Collab experiments at the Sanford Underground Research Facility (SURF) in Lead, South Dakota. Experiments at the 4100' level was monitored using a Continuous Active-Source Seismic Monitoring (CASSM) system. The CASSM system used 20 3- channel accelerometer sources and 72 receivers to provide near continuous observations from 2018 to 2022 during induce fracturing and shearing of known fractures. The full waveforms contained codas from multiple reflections from existing and induced fractures. A triplet loss neural network was used for signature identification of alteration of fractured rock during injection and pressurization. Shear stimulation generated a set of features that captured changes in fracture and matrix properties induced by the test. During hydraulic fracturing, the features were observed to relax over several days to a distinct new configuration reflecting the alteration of the fracture system. This analysis demonstrates that the coda signal is rich in information on the condition fracture rock and the potential for remote monitoring of subsurface geothermal fracture systems. Keywords: Shear Stimulation; Fracture Networks; Geothermal; Machine Learning; Hydraulic Fracturing

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Alexis Nicole Hensley