Machine Learning predicts Fracture Growth in Sandstone Analog
The prediction of earthquake timing and magnitude is of fundamental interest to geoscientists. For decades, researchers have tried to understand the physics behind earthquakes through laboratory deformation experiments, where fractures grow through the coalescence of microcracks. I simulate laboratory deformation experiments on a cohesive sandstone analog to document four independent precursors of rock fracture over a range of stress conditions. These precursors can be employed with machine learning to predict the time and stress required to initiate a fracture. My machine learning algorithm further reveals that failure prediction is improved by analyzing multiple precursors. My findings suggest that to earthquake forecasting techniques may be improved by employing catalogs of individual precursors measuring the abundance, size, mechanism and spatial distribution of microcracks.