Studying crystallization using X-ray radiography and machine learning
An X-ray radiographic study of the crystallization behaviour of liquid alloys is described by Enzo Liotti, Andrew Lui and Patrick Grant in Crystal nucleation in metallic alloys using X-ray radiography and machine learning in Science Advances. Working with colleagues from Andrew Zisserman’s computer vision group in the Department of Engineering Science, Liotti et al uses machine learning techniques to teach a computer to automatically detect the nucleation of crystals in terra-bytes of X-ray radiographic videos obtained during solidification experiments at the European Synchrotron Radiation Facility (ESRF). The quality of the videos combined with computer vision techniques allows the alloy composition at the point and instant of nucleation to be determined automatically, which in turn allows an estimate of the temperature and nucleation undercooling for every crystallization event. Studying thousands of nucleation events, they show how undercooling varies with solidification conditions, and explain how sudden bursts of crystallization are linked to the thermal-solute conditions in the liquid. Machine learning computer vision allowed enormous volumes of data that were unanalysable by hand to be converted robustly into distributions of nucleation undercoolings.