Jon Estrada, University of Michigan
Krishna Garikipati, University of Michigan
Samantha Daly, UCSB
Experimental studies of the mechanics of materials often require the analysis of large amounts of high-resolution, multi-modal datasets. Significant gains in materials synthesis capabilities (e.g. additive manufacturing) have led to material systems with increasing structural complexity and hierarchy, often entailing disparate scale and time dependent deformation and failure processes. At the same time, in the last two decades the output of new and powerful experimental methods place data science techniques at a tipping point for the exploration and analysis of material behavior. This data can be highly multi-modal, containing images and physical quantities obtained from different probes, all measured at a range of spatio-temporal resolutions. This wealth of experimental data presents unparalleled opportunities for exploration, yet these data are also challenging to utilize effectively in their entirety. This symposium addresses approaches inspired by data sciences, machine learning, and artificial intelligence communities to decipher correlations between synthesis methods, microstructure, and properties, including interfacing with multiscale modeling efforts. Also of interest are system inference methods to identify optimal models, encompassing algebraic and differential equations, that govern the data, while balancing accuracy and parsimony, and quantifying uncertainties in the models and predictions. Machine learning applied to the experimental mechanics domain is in its early stages; this symposium aims to advance the development of adaptable frameworks for the handling and exploration of these high-dimensional data spaces. This includes consideration of the intersection of machine learning and physical underpinnings, for example encoding physical structures into machine learning models, or using machine learning to identify structural relationships in large, multi-modal experimental datasets.