Large-Area X-Ray Micro-Fluorescence Imaging OF Heterogeneous Materials

Abstract

An X-ray Micro-Fluorescence (XRMF) spectrometer, with an analysis area of about 100 by 150 microns, has been used to collect 2-dimensional X-ray intensity maps over large-area (5 to 50 mm in X and Y) samples. These intensity maps were collected by scanning the sample on an XY stage, and converting X-ray Energy-Dispersive spectra to peak intensities for the elements of interest. The maps, when displayed using false-color or pseudogray scales, show the distribution of individual chemical elements over the analysis area. These maps can be collected at speeds from about 1 minute per frame (analysing 25 elements simultaneously). Greater precision of chemical intensities, or larger area maps, may require several hours, particularly if extensive data processing is performed at each point. XRMF has advantages over more conventional SEM-EDS X-ray mapping, including sample preparation and presentation, as well as improved signal-to-noise ratios.A technique is described which assists in analyzing the large amount of data which is collected in each map. Principal Component Analysis (PCA) is performed on all of the elemental maps simultaneously. This technique compresses the many elemental intensity maps into a few principal components, resulting in many fewer maps to evaluate. The intensity maps of these principal components display the most pertinent information. They can also be plotted as scatter plots which can help with the partitioning of the data into individual phases. This procedure can potentially be automated as a method for phase analysis. The selected pixels from the scatter plots can be averaged and converted into phase compositions, and the phase information re-displayed on the original elemental or principal component maps.This technique has been applied to a thin section of rock, and to a synthetic multiphase alloy sample.

Automated segmentation of μ-XRF image sets

“…by threshholding) and will not be addressed here in order to simplify this Ðrst data set somewhat (see later). This type of material is similar to that described by Cross et al 16 In Fig. 6, the principal component images resulting from PCA analysis of the data in Fig.…”

“…Techniques for doing this have been reviewed recently by Bonnet8,9 and illustrated with data sets taken from electron probe microanalysis,11,12 Auger13 and EELS mapping14 and secondary ion microscopy. 15 For the speciÐc case of l-XRF data sets, Cross et al 16 have described the use of principal component Step size, 20 lm ; image size, 50 Ã 50 pixels or 1.00 Ã 1.00 mm ; spectrum collection time per pixel, 60 s. analysis (PCA) for collinearity removal and dimensionality reduction. By manually grouping pixels in the space of the resulting principle components (see below), semi-automated or supervised image segmentation was shown to be feasible for data sets in which a limited number of three principal components were present.…”

“…This approach to segmentation is termed interactive correlation partitioning (ICP). 12 In this paper, two alternative approaches to image segmentation are compared which can be automated to a higher degree than the procedure described by Cross et al 16 and which (as a result) can also be easily applied to data sets in which more than three principal components are present. With this type of procedure, automated correlation partitioning (ACP) then becomes possible.9 Also, the usefulness of these segmentation procedures for extracting from a l-XRF image data set a small number of region-speciÐc sum spectra is evaluated.…”

Highlights of x‐ray spectrometry for microanalysis

“…Samples could be viewed on the stage by means of an optical microscope coupled to a video camera. Mathematical methods using principle component analysis were developed by Cross et al 13 for displaying multi-element 2D maps. A beneÐt of the method was that the time of measurement could be signiÐcantly decreased and that was demonstrated in the application to the analysis of heterogeneous materials.…”

The Use of a Scanning X-ray Microprobe for Simultaneous XRF/XRD Studies of Fly-Ash Particles

“…The enormous amount of analytical data generated by such a scan cannot be handled in a conventional way and new algorithms and methods have to be developed and applied in order to obtain reasonable data reduction and to present relevant analytical information as twodimensional images corresponding to the scanned sample area. For fluorescence data, many interesting developments are already occurring within the field of principal component analysis (Cross, Lamb, Ma & Paque, 1992) and other types of software post processing (Rindby & Voglis, 1992). However, for diffraction data the problem will become quite different as the data from each individual pixel in an image presents itself as an image.…”