Estimating Fault Slip Potential with CNN-Based Deep Learning: Integrating Mohr-Coulomb and Non-Linear Failure Criteria for Advanced Seismic Risk Assessment

Estimating fault slip potential due to changes in in-situ pore pressure and stress is crucial in subsurface operations, such as Enhanced Geothermal Systems (EGS) development. A key factor in this estimation is selecting appropriate rock failure criteria. While the Mohr-Coulomb failure criterion is widely used, non-linear failure criteria, such as the Hoek-Brown criterion, can offer more accurate predictions. However, non-linear failure criteria require parameters that are often difficult to obtain directly. We proposed two ways to derive the equivalent Hoek-Brown parameters by friction coefficient and cohesion used in Mohr-Coulomb criterion. One approach used nonlinear least-squares regression to fit the Hoek-Brown and Mohr-Coulomb failure envelopes in terms of principal stresses. The other employed a convolutional neural network (CNN) for direct estimation of the Hoek-Brown parameters. We applied both the Mohr-Coulomb and equivalent Hoek-Brown criteria to assess fault slip probability in mapped faults within the DOE Utah FORGE project area during reservoir stimulation. By incorporating non-linear failure criteria, our method enhances the accuracy of seismic risk assessments and provides an operational tool for earthquake hazard management in EGS operations.