Seminar : Samantha Daly

Machine Learning and Experimental Mechanics

This talk will discuss some of the core opportunities and challenges at the intersection of machine learning and experimental mechanics. The application of data-driven and machine learning approaches to the mechanics of structural materials under a range of representative applications will be discussed, largely in the context of enabling high throughput experimentation and analysis, and in enabling new modes of structural health monitoring. Examples will include i) the use of spectral clustering to identify damage mechanisms from their acoustic emission spectra in ceramic matrix composites for the first time; ii) addressing the pervasive data scarcity problem by using generative adversarial networks in the synthesis of realistic morphologies for high throughput predictive simulations; (iii) addressing the spatial and temporal resolutions of limited experimental data that is vital to our understanding of material response, by adapting physics-informed superresolution approaches to materials data structures; and (iv) the high-throughput segmentation, identification, and analysis of the relationships between microstructure and deformation mechanisms. This talk will also incorporate discussion of the need for, and requirements of, trustworthy machine learning approaches in the physical sciences.