Machine learning automates identification of crystal structures in new materials

Providing a method for eliminating some of the guesswork from crystal structure determination, a machine learning-based approach to determining crystal symmetry and structure from unknown samples may greatly improve the speed and accuracy of this process. The new method brings crystallography into the high-throughput world of artificial intelligence (AI). From geology to biology to materials science, identifying crystal structure is critical to understanding its general characteristics and properties. Electron backscatter diffraction (EBSD) is the standard technique for identifying crystal structure. However, while powerful, EBSD requires user input concerning critical elements of structure, such as crystal phase, which can be both time-consuming and prone to error. A fully autonomous approach to more hands-on crystallography would open the door to a high-throughput evaluation of a material’s properties, according to the authors. Kevin Kaufmann and colleagues developed an autonomous, machine learning-based method for rapidly determining crystal structure from EBSD data with high accuracy. The authors used a convoluted neural network to identify unique crystal symmetries in EBSD pattern images using the same symmetry features a crystallographer would use. According to the results, the trained algorithm was capable of accurately identifying and classifying various aspects of crystal structure from diffraction patterns of materials that it was not trained on, and with almost no human input. The platform opens the door to high-throughput determination of structures for multiple fields.

###