This paper introduces a generalized deep learning model for classifying materials' crystal systems and space groups using X-ray diffraction (XRD) data. The model addresses the limitations of current automated XRD analysis, which often lack performance and adaptability. The approach uses a holistically represented training dataset accounting for various experimental conditions and crystal properties, incorporates an expedited learning technique, and optimizes model architecture based on Bragg's Law. Evaluation on experimental data and unseen materials demonstrates state-of-the-art performance, particularly in space group classification.

Jerardo E. Salgado, Samuel Lerman, Zhaotong Du, Chenliang Xu, Niaz Abdolrahim