News:

:) You cannot see "members only" boards if you are not a member, so please join the forum!

Main Menu

mineral identification

Started by DirkMueller, December 11, 2023, 05:28:11 AM

Previous topic - Next topic

DirkMueller

I am performing a lot of EPMA analyses on "unknown" mineral phases. This can be any kind of feldpars (K-fsp, plg), pyroxenes, olivines, epidotes, zeolites, etc..   Sorting and identification is quite time consuming.

Is there a software which can "automatically" do the job?

Sorting should be easy by defining a bunch of elements and I get clusters of this or that composition.
Identification should also be easy (at least for the major rock forming minerals) by using a database, that contains typical ranges of characteristic elements.
In the optimal case that software would also calculate the respective stoichiometry so that I directly know which kind of e.g. plagioclase it is. (This last point can be done by the Cameca PeakSight software, but I need to tell the software which analysis represents which kind of mineral.)

Thanks for any kind of answer :)
Dirk

Anette von der Handt

Hi Dirk,

Probe for EPMA has a "Match" function where you can match the composition against a database. The ones provided with PFE include the Deer/Howie/Zussman and Dana. However, it looks like you look for more of an automated matching.

Have a look at this recent publication here: da Silva et al. (2021) Qmin: A machine learning-based application for mineral chemistry data processing and analysis. Computers and Geosciences, 157, doi.org/10.1016/j.cageo.2021.104949.

There is also:
Brandelik, A. (2009). CALCMIN–an EXCEL™ Visual Basic application for calculating mineral structural formulae from electron microprobe analyses. Computers & Geosciences, 35(7), 1540-1551.
Walters, J. B. (2021). MinPlot: A mineral formula recalculation and plotting program for electron probe microanalysis. Mineralogia, 53(1), 51-66
Against the dark, a tall white fountain played.

Probeman

The Match feature in Probe for EPMA is discussed here:

https://smf.probesoftware.com/index.php?topic=1000.0

And one can match phases identified in CalcImage described here:

https://smf.probesoftware.com/index.php?topic=1071.msg7095#msg7095

You can also do a compositional match in the Standard application that comes with the free CalcZAF software:



The only stupid question is the one not asked!

DirkMueller

Thank you for your answers!

Since we (still) don't have PfEPMA (mea culpa) I need open source solutions.

The machine learning stuff is exactly what I was looking for. I expected that something like this is already out there. I need to find out how the tool works.

My "wish list" would be: throw in my data and the software is intelligent enough to identify the column assignment, find clusters, assign them to mineral phases, calculate the stoichiometry, gives back a table with my analyses and the corresponding mineral name together with a hit probability for each analysis.

... Christmas is the time for wishes :)


sem-geologist

Dirk,
You had forgot to list the other following steps in the wish list: ..., drafting a paper, choosing the right journal for biggest impact, and dealing with intelligent-differently reviewers.

Jokes away, I as human being with decade of experience run often into trouble of identifying with 100% certainty the new to me minerals from only the chemical composition. It often needs additional external information by other method to get closer to what the thing could actually be. Yes, ML will help you and can increase throughput if You know what minerals you have in the sample, You need first to train ML to recognize those. I.e. If You know that your samples have only Quartz, Plagioclase, biotite, amphibole and pyroxene and nothing else, and train the ML to recognize those - that will work exceptionally well. But throw something what ML was not trained to recognize and this will derail spectacularly. But wait - actually in some cases ML will have problem to distinguish pyroxene from amphibole when amphibole is K-poor and Fe rich. There is such enormous compositional overlap between some groups, that slight miss-calibration in measurement would result in wrong classification.

DirkMueller

Some updates to Qmin:

a) The link to the webpage has changed to the one mentioned in their paper. The actual one is: https://apps.sgb.gov.br/qmin/

b) Qmin works quite well for my purpose. Of course, glass is been identified as amphibole and zeolites as micas... But for minerals, which are in the database it works pretty good.

c) Unfortunately, it does not write the paper ;)

Nicholas Ritchie

As an alternative, you can search the WikiData (>3700 minerals) and RRUFF (>5700 minerals) mineral databases using the Julia language (https://julialang.org/) and the NeXLCore X-ray microanalysis library (https://github.com/usnistgov/NeXLCore.jl)

I've added a Jupyter notebook that shows how to: https://github.com/usnistgov/NeXLCore.jl/blob/v0.3.15/jupyter/mineral_db.ipynb
"Do what you can, with what you have, where you are"
  - Teddy Roosevelt

Probeman

#7
Note also that one can perform compositional matching in the Standard application that comes with CalcZAF (and in Probe for EPMA and CalcImage as well) as described here:

https://smf.probesoftware.com/index.php?topic=1588.msg12253#msg12253

using the provided AMCSD.MDB (American Mineralogist Crystal Structure Database) mineral database which contains over 9000 mineral compositions.
The only stupid question is the one not asked!