The images shown here are an electron back-scatter image,
and corresponding mineral map of each pixel in the image analyzed with the new
technique.
|
|
Figure 1: Automated nanomin analysis of mineral sample (no user intervention)
This approach does not constrain the number of minerals per
pixel, unlike a standard QEMSCAN analysis. Rather, it allows multiple minerals
to be reported within a single pixel. This goes beyond the EDS limit, revealing
sub-pixel detail for the first time. It also allows each mineral to consist of
multiple mineral variety end-members that are linked into a solid solution. For
example, the plagioclase feldspar solid solution links the albite and anorthite
end-member mineral spectra into a solid solution as a single mineral. Any
mineral detected on that solid solution is automatically reported as a
plagioclase, regardless of the percentage of sodium and calcium in the
spectrum. It automatically computes the interchange of sodium and calcium, balancing out the other elements, without any interaction required from the user.
This removes the burden of managing complex mineral
chemistry within the analytical software, and frees up time for analysis of the
data.
In SEM-EDS mineral analysis, the interaction volume of the
EDS signal is approximately 1-5um for typical measurement conditions. That is,
each pixel represents the signal from a 5um volume. And when that volume has
more than 1 mineral present, the signal measured is a sum of the several
minerals, which may or may not have different elements.
The implication here is apparent - any pixel that does not fall wholly
within a pure mineral grain should be reported as a mix of minerals. A few commercial
SEM-EDS techniques attempt to avoid this problem by trying to avoid measuring pixels
that aren’t clearly in the center of a mineral grain. Other try to remediate it
by using a statistical approach based on the surrounding mineral classifications
to touch-up the classification of the center pixel. Both of these work-arounds will
work for a limited set of situations, but they fail where it’s impossible to avoid
a fine texture.
|
|
|
Figure 2: Automated QEMSCAN analysis (no user intervention)
The centroid method risks runs a risk that the simplified
analysis will miss the important features within the sample. And it also introduces
a bias into the result if portions of the sample are skipped or left unclassified.
In the image above, an automated QEMSCAN analysis shows large swathes of
unclassified pixels, requiring further input from the user. The user must
interactively decide whether to skip the unclassified pixels, try to recover
them from surrounding data, or manually add mineral entries for the missing
minerals. In other words, there is a significant effort required to improve the
accuracy from the initial classification. While all these problems can be
solved, they also add significant time to the process.
In contrast, the new technique shows a result where the
pixels from multiple minerals have been reported, saving a significant amount of
time to an operator. The operator has more time for interpreting the results,
or classifying other samples.
At the current time, this technique is only offered by FEI (http://www.fei.com) within the Maps-Mineralogy
software package.
The only link currently available is https://www.linkedin.com/pulse/finer-side-automated-spatial-mineralogy-lucy-plant
The only link currently available is https://www.linkedin.com/pulse/finer-side-automated-spatial-mineralogy-lucy-plant
