Year
2019
Abstract
Synchrotron radiation-based analysis of particulate samples, including swipe samples, provides an avenue to obtain spatially resolved elemental maps and spectroscopic data without sample destruction or pretreatment. With third generation synchrotron radiation facilities, such as the National Synchrotron Light Source II (NSLS-II) at Brookhaven National Laboratory, Upton, NY, USA, it is possible to rapidly collect X-ray Fluorescence (XRF) spectra at micron-scale resolution. The XRF spectra are the basis for elemental maps. At the Submicron Resolution X-ray Spectroscopy beamline at NSLS-II, it is possible to collect data in a single scan that can be converted into maps for elements from Ca through the actinides at concentrations down to trace-elemental levels. Elemental maps are used to identify areas of interest for follow-on measurements, which may include more detailed elemental maps at micron-to-submicron spatial resolution and submicron X-ray Absorption Near-Edge Structure spectroscopy to determine the oxidation state and binding environment of elements of interest (e.g., U). The synchrotron radiation-based XRF analysis produces a significant amount of data (up to 1 TB) per scan. Analyzing this volume of data by standard approaches involving visual inspection of element maps and overlays of element maps is time consuming. Here we report on a method using k-means cluster analysis to analyze elemental maps faster and more rigorously. The cluster analysis—a form of Machine Learning—allows one to evaluate the homogeneity of the sample, find and quantify multiple locations of populations of chemically distinct particles, and guide the identification of areas of interest for follow-on analysis either with synchrotron radiation-based techniques or laboratory-based destructive analysis. Combined with methods of registration; standard sample holders; and robotics, the cluster analysis could be incorporated into an automated workflow that produces elemental maps, identifies chemically distinct particles, and guides follow-on analysis. This envisioned automated workflow has the potential to generate a wealth of contextual information complementary to destructive analysis on the same sample as well as form a basis for a prioritization of samples for labor-intensive, follow-on, destructive analyses. The development of a data analytics approach, such as the cluster analysis reported here, is central to an automated workflow that enables high sample throughput.