Crystal systems, Bravais lattices, extinction groups, space groups, and intervals of unit cell volume are classified using a convolutional neural network based on the deep learning of model full-profile powder XRD patterns computed from ICSD data. A new method is proposed to significantly increase the classification accuracy by normalizing unit cell volumes to the total fixed value when computing the model XRD patterns. The classification accuracy determined from an independent set of normalized model XRD patterns is 97.4% for crystal systems and 88.0% for space groups. Due to its high accuracy, this neural network can be utilized to perform a crystal symmetry analysis using experimental unit cell volumes and to determine space groups without using reflection extinction rules after finding the unit cell volume by commonly utilized indexing programs.
Zaloga et al. (Sun,) studied this question.