Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 33: Topical session (Symposium MM): Big Data Analytics in Materials Science

MM 33.6: Vortrag

Donnerstag, 4. April 2019, 12:15–12:30, H43

Crystal-structure identification in polycrystals via Bayesian deep learning — •Angelo Ziletti, Andreas Leitherer, Matthias Scheffler, and Luca Ghiringhelli — Fritz Haber Institute of the Max Planck Society Faradayweg 4-6 14195 Berlin, Germany

Thanks to open-access online computational repositories (e.g. http://nomad-coe.eu) and experiments (e.g. atom probe tomography), researchers have now access to a vast amount of three-dimensional structural data. To extract valuable information for materials characterization and analytics, computational methods that automatically and efficiently detect long-range order are needed. Current methods are either not stable with respect to defects, or base their representation on local atomic neighbourhoods, which in turn makes it difficult to detect "average" longe-range order. In the proposed approach, for a given crystal structure, we simulate its (three-dimensional) diffraction pattern, and by means of a spherical-harmonics expansion, we compute a rotationally and translationally invariant representation. A convolutional neural network is then used to identify the correct crystal structure; in particular, we use a Bayesian neural network in order to obtain statistically-principled classification probabilities and model uncertainty. This methodology is used to classify grains in polycrystals, find coherent regions in amorphous solids, but also detect crystallographic defects such as twin boundaries, stacking faults, and edge dislocations in heavily defected crystal structures.