Compositional Community Detection: Automated Identification of Chemical Segregation in Atom Probe Tomography Data

Abstract

We introduce a fully unsupervised clustering method we call Compositional Community Detection (CCD) to identify chemical motifs in atom probe tomography (APT) reconstructions. In the CCD approach, APT point clouds are broken into overlapping spherical neighborhoods, and repeated k-means clustering coupled with Louvain community detection is used to group neighborhoods based on their ion composition. Kolmogorov-Smirnov statistics for present ion types provide interpretable descriptors of each community that indicate the relative level of enrichment or depletion of ions within a community. We demonstrate our technique on a set of APT reconstructions of irradiated 316 stainless steel. Our method detected chromium carbide and Ni-Si-rich precipitates and located a grain boundary based on Ni and Si enrichment. Spatial correlations between communities indicated that chromium carbide precipitates were flanked by regions of Fe depletion. Our results highlight the potential of CCD in the analysis of chemical segregation in broader classes of materials, in terms of both varying synthesis methods and exposure to extreme environments.

Keywords: atom probe tomography; data analytics; radiation-induced segmentation; unsupervised clustering.