X-ray fluorescence microscopy (µ-EDXRF, µXRF)

BGR has three energy-dispersive X-ray fluorescence microscopes of the M4 Tornado and M4 Tornado Plus types from Bruker Nano, and an AttoMap from Sigray (Fig. 1). µXRF is primarily used due to its ability to fast and non-destructively provide qualitative and semi-quantitative geochemical and mineralogical data in 2D. Sample preparation is easy, as it only requires a flat sample surface. Consequently, a variety of sample types such as drill cores, hand specimens, rock fragments, sediments, lacquer profiles, etc., can be analyzed. This method is particularly suitable for obtaining a quick overview of a larger quantity and size of samples and selecting regions of interest for further detailed analysis.

Figure 1: µ-XRF Tornado Plus with open sample chamber (left) and Sigray Attomap (right) Source: BGR

µ-XRF technique

Figure 2: µXRF-data of a hand specimen (ca. 9 x 7 cm) with a sample photo (top), distribution of selected elements (middle) and mineral interpretation with modal mineralogy (bottom) from Petrograpic Analyst Source: BGR

µ-XRF uses a focused X-ray beam to scan a sample surface and record the resulting fluorescence signals at a micrometer resolution. For each scanned pixel, an energy dispersive spectrum is recorded with intensities of all measurable elements, which can be displayed as element distribution maps and regions of interest can be selected for a semi-quantitative chemical quantification of selected areas.
The Bruker devices employ a rhodium tube that can operate with up to 50 kV and 600 µA or alternatively a Cr tube with up to 50 kV and 700 µA. A poly-capillary focuses the polychromatic X-ray beam to a spot size of approximately 20 µm (Nikonow & Rammlmair, 2016). The resulting fluorescence is measured by two detectors (silicon drift detectors, SDD) facing each other at 180°. Two detectors are necessary to remove diffraction signals that overlap real element peaks in the spectrum. The sample chamber can fit samples up to 16 cm x 20 cm and can be evacuated to 2 mbar, enabling the detection of signals from carbon to uranium for Tornado Plus. Sample preparation is reduced to creating a flat surface. Measurement conditions such as step size and measurement time per point are adjustable, starting from 4 µm and 0.01 milliseconds, respectively. Point measurements, point grids, profiles, or area measurements can be conducted. An automated script can be used to measure and save up to 24 individual samples in series. Measurement times vary depending on sample size and measurement conditions, and can range from 30 minutes to a few hours for area measurements.
The fluorescence spectra are stored individually for point measurements and for area measurements in a data set from which the corresponding information can be displayed and extracted after the measurement, e.g. element distribution maps for single or multiple elements in gray scale or false color images (Fig. 2b). The chemical quantification of spectra is performed either through calibration using samples of known chemical composition or by standardless fundamental parameter (FP) methods. Specific areas from the measurement can be selected and quantified separately.

Data analysis

Spatially-resolved µXRF deals with the visualization of chemical, mineralogical, and textural information based on identified element-specific fluorescence signals from C to U. Characteristic areas of the entire sample, such as layers or individual phases, can be selected, and chemical analyses can be calculated from the mean spectra of all pixels within a selected region of interest. This provides a way to characterize areas chemically that would require significant preparatory effort for bulk chemical analysis.

Figure 3: Section of a quartz thin section under crossed polarizers (a) and evaluation of the diffraction from the µXRF (b). The µXRF shows a silicon signal in the quartz and, depending on the orientation of the quartz grains, a series of diffraction peaks that can be used to separate individual grains. Diagram c) shows that the grain size distribution (in 2D) from the thin section correlates with the grain size distribution from the µXRF data. Source: BGR

The spectra of selected areas reflecting chemical characteristics (spectral fingerprints) can be used to differentiate individual mineral phases. This information is further utilized to create a mineral database for identifying individual phases through hyperspectral analysis methods (similar to MLA). In this way, mineral distribution images can be created, which can be quantitatively evaluated for individual mineral phases based on textural criteria (Fig. 2c). The distinction of minerals is limited to minerals with different spectral fingerprints; therefore, polymorphs (such as calcite and aragonite, or rutile, anatase and brookite) cannot be distinguished with this method alone. By identifying and extracting diffraction-related signals in the spectrum (diffraction peaks or Bragg peaks) and combining the two detectors, monomineralic areas can be separated into individual grains (Fig. 3, Nikonow & Rammlmair, 2016).

Petrographic Analyst is an in-house software tool based on the IDL programming language. It includes automation of data processing and analysis including the graphical representation of the mineral distribution (Fig. 2c), the rock classification and the grain size distribution of selected mineral phases (Nikonow & Rammlmair, 2017). Petrographic Analyst is not limited to µXRF data, but can also process image information from other sources. Thus, data from different methods such as hyperspectral imaging (HSI) oder laser induced plasma spectroscopy (LIBS) can be combined and analyzed (Fig. 4; Nikonow et al., 2019).

Figure 4: Element intensities from µXRF (left, Si, Fe and K in white, green and blue) and LIBS (middle, Li in red, Fe in green and K in blue) and classification (right) from a syenogranite sample. While μXRF can only identify plagioclase, LIBS can measure light elements like lithium and distinguish between Li-rich and Li-poor plagioclase and biotite. Some of the upper right plagioclases show Li-rich cores, while plagioclase in the lower part contains only sodium and some calcium besides aluminum and silicon. Source: Nikonow et al. 2019

Keywords
µXRF, XRF imaging, element mapping, automated mineralogy, hyperspectral imaging, Bruker M4 Tornado, Sigray Attomap

Literature

Nikonow, W., Rammlmair, D., 2016. Risk and benefit of diffraction in Energy Dispersive X-ray fluorescence mapping. Spectrochimica Acta Part B: Atomic Spectroscopy 125, 120-126. https://doi.org/10.1016/j.sab.2016.09.018

Nikonow, W., Rammlmair, D., 2017. Automated mineralogy based on micro-energy-dispersive X-ray fluorescence microscopy (µ-EDXRF) applied to plutonic rock thin sections in comparison to a mineral liberation analyzer. Geosci. Instrum. Method. Data Syst. 6, 429-437. https://doi.org/10.5194/gi-6-429-2017

Nikonow, W., Rammlmair, D., Meima, J.A., Schodlok, M.C., 2019. Advanced mineral characterization and petrographic analysis by μ-EDXRF, LIBS, HSI and hyperspectral data merging. Mineralogy and Petrology 113, 417-431. https://doi.org/10.1007/s00710-019-00657-z