Using machine learning to reveal rock physics
Machine learning (ML) methods are providing new insights into strain localization and fracture development leading to failure. ML analyses that use data from X-ray tomography triaxial compression deformation experiments provide unparalleled access to unique observations and efficient means of analyzing such big data.
Recent analyses indicate that ML methods can predict the likelihood of fracture growth and the timing of macroscopic failure with high accuracy. These successful predictions allow an analysis of the factors that control fracture growth and the timing of macroscopic failure. In this project, the student will test a suite of ML algorithms in order to identify the models that produce the most accurate predictions of fracture growth and failure.
The student will then compare differences between the algorithms, and in particular, which features are considered to have the most predictive power. Identifying the characteristics of fractures and the localizing strain field that provide the most predictive power may help the community better predict the onset of failure in engineered structures or earthquakes.