Data-driven analysis of synchrotron Laue microdiffraction

Grain morphology, textures and crystal structures are the most important structure factors, which directly determine the properties of solid materials. By state-of-art synchrotron X-ray microdiffraction, all these three characteristics can be measured simultaneously. However, the analysis of tens of thousands of Laue patterns is quite time consuming, sometime is impossible for unknown crystals. We developed a machine learning approach that can generate an orientation map from a set of Laue scans without knowing the crystal structure and indexing information.

X-ray microdiffraction is a powerful experimental tool to study the microstructure-property relation of solid materials, however the data analysis is quite complicated and sometimes is time consuming. The machine learning model we developed provides an analysis method independent of indexing procedure. It doesn't require crystallographic information as the input. The model is derived by CNN using 50,000 training patterns from different solid materials with various textures. It immensely speeds up the analysis of Laue microdiffraction by a large margin.

CNN Auto Encorder

The examples of various materials analyzed by our machine learning approach without knowing the crystal structure.