Dresden 2017 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 42: Topical session: Data driven materials design - high throughput
MM 42.4: Vortrag
Mittwoch, 22. März 2017, 13:00–13:15, BAR 205
Atom-mining: Improving the spatial resolution of Field Ion Microscopy using atomistic simulations and data mining — •Gh Ali Nematollahi1, Shyam S. Katnagallu1, Michal Dagan2, Baptiste Gault1, Blazej Grabowski1, Paul Bagot2, Michael Moody2, Dierk Raabe1, and Jörg Neugebauer1 — 1Max-Planck Institut für Eisenforschung, D-40237 Düsseldorf, Germany — 2Department of Materials, University of Oxford, Parks Road, Oxford, OX1 3PH UK
Field Ion Microscopy (FIM) relies on the ionization of inert gas atoms from the specimen surface subjected to an intense electric field. Any FIM image represents a "snapshot" of individual surface atoms and 3D information of the bulk can be obtained by removing the surface atoms using field evaporation. However, the analysis of the atomistic information is not straightforward due to strong deformations and intensity variations caused by the electric field, and so far automated techniques to reconstruct the 3D atomistic structure are lacking. In this work, building on recent efforts by Dagan et al.[Microsc. Microanal.], we developed a new framework for automated reconstruction, using data mining and atomistic simulation techniques. In particular, different unsupervised and semi-supervised machine learning techniques are used to detect atoms, crystallographic planes and defects in FIM image. The results show that machine learning can be successfully employed to correct for the deformation of the sample caused by the strong electric fields. In a last step, atomistic relaxation based on empirical potentials is used to further improve the 3D-reconstructed data.