This project develops an integrated platform that combines computational modeling and machine learning to provide a coherent and integrated physical picture to gain mechanistic insights from multi-modal in situ characterization techniques to accelerate material discovery in the energy arena. BNL is developing a diverse array of characterization techniques at both NSLS-II and the CFN to provide the quantitative information needed to rationally design functional materials for energy applications. These techniques exploit both in-situ and operando approaches, which each generate high rate data streams. Thus, automatic, real time analyses and data driven discovery have been identified as critical Laboratory goals. Other science domains have seen the application of machine learning, which has led to multiple breakthrough discoveries. However, these approaches are in their infancy when applied to challenges in materials science.
We develop methods that allow near real time analysis of transmission electron microscopy images from a 3GB/s image stream, solving a general problem that will soon be seen at multiple beamlines at NSLS-II. We will also use first principles computational modeling to gain physical insights into the atomic-scale changes that occur in complex dynamic processes in operando conditions. These two activities will be combined in a robust, integrated manner to describe the dynamic structural changes that occur in a functioning model nanocatalyst.