Michael Zaiser

Abstract:

Materials failure has for decades been considered one of the paradigmatic multiscale phenomena, involving processes from the atomic to the systems scale. On the continuum level, a well established approach is provided by the laws of fracture mechanics established by Griffith's seminal work exactly a century ago. However, the relationship between key concepts of fracture mechanics such as fracture toughness on the one hand, and parameters characterizing the microstructure of materials from the atomic to the grain scale on the other hand, remains poorly understood. The same is true for the transition from diffuse accumulation of damage to the formation and propagation of a macroscopic crack. Data analytic approaches may offer new pathways towards closing the gap between discrete and continuous descriptions of material microstructures undergoing failure under load. We illustrate this on a range of examples from the atomic to the geo-scale. On the atomic level, we show how machine learning methods can be used to identify local atomic configurations prone to irreversible change under load, and how continuum mechanics concepts can provide essential 'domain knowledge' in approaching this task. On the mesoscale, we demonstrate how network theoretical concepts can be used to identify potential failure locations in load-carrying structures that can be mapped onto networks transmitting linear momentum. Finally, on the macroscale, we discuss how macroscopic monitoring data can be used to predict imminent failure under load.