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

KFM: Fachverband Kristalline Festkörper und deren Mikrostruktur

KFM 19: Topical Session: In Situ and Multimodal Microscopy in Materials Physics (joint session MM/KFM)

KFM 19.8: Topical Talk

Mittwoch, 20. März 2024, 17:30–18:00, C 130

Unsupervised Machine Learning Analysis for Electron Microscopy Datasets — •Mary Scott — Department of Materials Science and Engineering, University of California, Berkeley, Berkeley, CA 94720, USA — National Center for Electron Microscopy, Molecular Foundry, Lawrence Berkeley National Lab, Berkeley, CA 94720, USA

Electron microscopy is the characterization method of choice to observe local atomic-scale and microstructural features within materials that play a critical role in material performance. Recently developed high frame rate electron detectors acquire diffraction patterns from nanoscale regions at frame rates of 100 kHz, an approach that enables multimodal analysis from the same dataset to create maps of crystal orientation, strain, and more. This method, termed 4D-STEM, creates datasets can contain tens of thousands of diffraction patterns from heterogeneous structural regions. The large datasets cannot be analyzed manually, and lack of prior knowledge of the crystal structure of diverse samples limits the application of supervised automated approaches, motivating the development of unsupervised analysis. Here I will overview implementation of an automated, unsupervised clustering pipeline for 4D-STEM data, emphasizing the importance of input data representation. Futhermore, I will describe an ensemble approach to generate more stable clustering results. This type of unsupervised data analysis pipeline is an important step towards incorporating rapid 4D-STEM analysis into material discovery and design efforts, particularly when evaluating defect-rich and disordered materials.

Keywords: Electron Microscopy; Machine learning