A Comprehensive Study on The Kinetics of Defect Formation and Interaction Coupling In-Situ TEM with Deep Neural Networks

Simulating reactor conditions inside a transmission electron microscope (TEM) gives insight into the rate of defect formation and survivability of materials in the environments in which they were designed to perform. Efforts to fully encapsulate dynamic processes occurring in a material have been challenging, owing to the large volume of information collected from an individual experiment and multiple interactions occurring simultaneously. Hereby, deep neural networks emerge as a suitable method to extract complicated information from input images and output useful analytics that help understand the physics of a reaction. This talk discusses strategies for handling a range of datasets from <i>in situ</i> TEM experiments that span different material systems, imaging conditions, and irradiation conditions to predict the stability of materials. Emphasis will be placed on developing end-to-end process flows equipped with generating synthetic data to learn feature representations of images, pattern recognition to highlight hidden trends in the data, and overall implications for the advancement of nuclear materials characterization.