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

MM: Fachverband Metall- und Materialphysik

MM 31: Data Driven Materials Science: Big Data and Work Flows – Machine Learning

MM 31.1: Vortrag

Mittwoch, 29. März 2023, 15:45–16:00, SCH A 251

Neural networks trained on synthetically generated crystals can classify space groups of ICSD powder X-ray diffractograms — •Henrik Schopmans, Patrick Reiser, and Pascal Friederich — Institute of Theoretical Informatics, KIT, Karlsruhe, Germany

Machine learning techniques have successfully been used to extract structural information such as the crystal space group from powder X-ray diffraction (XRD) patterns. However, training directly on simulated patterns from databases like the ICSD is problematic due to its limited size, class-inhomogeneity, and bias toward certain structure types. We propose an alternative approach of generating random crystals with random coordinates by using the symmetry operations of each space group. Based on this approach, we present a high-performance distributed python framework to simultaneously generate structures, simulate patterns, and perform online learning. This allows training on millions of unique patterns per hour. For our chosen task of space group classification, we achieve a test accuracy of 77.4% on new ICSD structure types not included in the statistics dataset guiding the random generation. Instead of space group classification, the developed framework can also be used for other common tasks, such as augmentation and mixing of patterns for phase fraction determination. Our results demonstrate, using the domain of X-ray diffraction, how state-of-the-art models trained on large, fully synthetic datasets can be used to guide the analysis of physical experiments.