Dresden 2020 – scientific programme
The DPG Spring Meeting in Dresden had to be cancelled! Read more ...
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 6: Topical Session: Data Driven Materials Science - Materials Design II (joint session MM/CPP)
MM 6.1: Talk
Monday, March 16, 2020, 11:45–12:00, BAR 205
Versatile Bayesian deep-learning framework for crystal-structure recognition in single- and polycrystalline materials — •Andreas Leitherer, Angelo Ziletti, Matthias Scheffler, and Luca M. Ghiringhelli — Fritz Haber Institute of the Max Planck Society, Berlin, Germany
Physical properties of a given material are directly related to its structure. In particular, in polycrystalline materials the location and nature of grain boundaries are crucial features determining their properties. For instance, mechanical characteristics of steels are strongly influenced by grain boundaries. In this work, we propose strided pattern matching which is a framework using single-crystal classification to investigate polycrystals. Accessible crystal-structure identification methods are either very robust – but can treat only few classes – or include a large number of classes – but are not very robust. We use a Bayesian neural network in combination with the smooth-overlap-of-atomic-positions (SOAP) descriptor, allowing us to classify, robustly and without any predefined threshold, more than 100 prototypes including not only bulk but also two- and one-dimensional materials (e.g., fullerenes). Furthermore, we are able to quantify the uncertainty in the model predictions. As an example for polycrystal investigation, we apply our model to recognize an ordered L12 phase in a disordered fcc matrix. This serves as a model system for precipitate detection in Ni-based superalloys, which are materials used in aircraft engines.