Correlated electron diffraction and energy-dispersive X-ray for automated microstructure analysis

• Published in: Computational materials science 2023-09, Vol.228, p.112336, Article 112336
• Main Authors: Duran, E.C., Kho, Z., Einsle, J.F., Azaceta, I., Cavill, S.A., Kerrigan, A., Lazarov, V.K., Eggeman, A.S.
• Format: Article
• Language: English
• Subjects: 4D-STEM; Correlated STEM; Fuzzy clustering; Heusler alloys; STEM-EDS; Unsupervised machine learning
• ISSN: 0927-0256; 1879-0801
• DOI: 10.1016/j.commatsci.2023.112336

[Display omitted] •A workflow is presented to successfully merge and analyse simultaneously collected signals.•Appropriate pre-processing is shown to improve the segmentation of the data.•Unsupervised clustering of merged signals allows microstructural features to be identified in complex coherent materials. In this study the effect of merging correlated energy dispersive X-ray (EDS) spectra and electron diffraction data on unsupervised machine learning (clustering) is explored. The combination of data allows second phase coherent precipitates to be identified, that could not be determined from either the individual EDS or diffraction data alone. In order to successfully combine these two distinct data types we leveraged a data fusion method where both data sets were normalised and combined using a robust scaler followed by variance equalisation. A machine learning pipeline was implemented which performs dimensional reduction with PCA and followed by fuzzy C-means clustering, as this allows signals from overlapping regions of the microstructure to be partitioned between different clusters. User control of this partition is used to confirm a change in the stoichiometry of the embedded second phase regions.