Learning the Geometry of Failure

Abstract

This research seeks to braid together two previously separate threads of research, in a new collaboration, to learn the geometry of failure. Previously, Dr. Jerolmack helped to discover universal behaviors in the geometry of fracture networks, like those on the surface of Earth and other planets. These fracture networks are hierarchical, and exceptions in the hierarchy may indicate the presence of water. However, the number of fracture networks analyzed is statistically insignificant due to the time-consuming nature of discovering and analyzing these networks. Additionally, using snapshots of planetary images may not be sufficient to diagnose its formative processes. This proposal seeks to address these problems by developing a machine learning model that will learn the rules of fracture network growth and to establish a rigorous link to material properties. This proposal outlines how to train Dr. Trask’s pioneering Machine Learning (ML) framework with remote imaging and Dr. Jerolmack’s laboratory findings on the suspected role of water in fracture networks. If successful, this research has implications in aiding in the discovery and interpretation of water-influenced fracture patterns on Earth and on other planets. The Scientific Advisory Board was impressed and excited with the creativity of this proposal and noted that this work is highly interdisciplinary, including geoscience, mechanical engineering, physics and computer science.

Back to New Initiative Grants