Image Analysis using Convolutional Neural Networks for Modeling 2D Fracture Propagation

The primary failure mechanism in brittle materials such as ceramics, granite and some metal alloys is through the presence of defects which result in crack formation and propagation under the application of load. We are interested in studying this process of crack propagation, interaction and coalescence, which degrades the strength of the specimen. Traditionally, engineering applications that study these materials employ finite element mesh-based methods that require hundreds of hours of processing time on multi-core high performance clusters. We have developed a graph-based reduced order model that captures key geometric and topological features of the dynamic fracture propagation network. We report here the early stages of our study in which deep neural networks will be applied to dynamic directed weighted graphs capturing various metrics of crack-pair interaction strength with the aim of predicting crack lengths, dynamic crack growth/coalescence properties, distributions of these properties over the entire material through time, failure paths and time to failure. Our graph-based representations allow us to consider detailed topology in conjunction with metric geometry to gain insights into the dominant mechanisms that drive the physics in these systems.