Crystal symmetry classification from powder X-ray diffraction patterns using a convolutional neural network

• Published in: Materials today communications Vol. 25; p. 101662
• Main Authors: Zaloga, Alexander N., Stanovov, Vladimir V., Bezrukova, Oksana E., Dubinin, Petr S., Yakimov, Igor S.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2020
• Subjects: Artificial neural networks; Crystal systems; Powder diffraction; Space groups; Powder diffraction; Space groups; Crystal systems; Artificial neural networks
• ISSN: 2352-4928; 2352-4928
• DOI: 10.1016/j.mtcomm.2020.101662

A convolutional artificial neural network was applied to identify crystal systems and symmetry space groups by full-profile X-ray diffraction patterns calculated from crystal structures of the ICSD 2017 database. The database contains 192 004 crystal structures; 80 % of them were used as a training dataset, and the other 20 % were used as a test dataset to establish the accuracy of classification. The neural network identified crystal systems correctly for 90.02 % of structures and space groups for 79.82 % of structures from the test dataset. Factors affecting the classification accuracy were established. The first, nonlinear normalization of intensities of diffraction peaks increases the accuracy, and the second, the accuracy depends on the number of structures represented in each space group.