![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
getting faster, when there aren't issues
Mar. 13th, 2023 11:15 pmWhen I first started doing data processing for the LA-ICP-MS trace element composition maps I made from samples from a variety of soapstone quarries, I didn't really know what the chemistry of the minerals likely to be present should look like in the results. The laser shoots at the sample, ablating a small bit of the surface into very tiny particles, which get carried up the tube on a stream of He gas, till it comes tot the ICP-MS, where it gets sent through the plasma, which is hot enough to ionize the particles, which continue rushing sorta forward, each ion deflected, just a little, depending on its mass, till they hit the counting wall, where, if the door is open for the right spot for the mass of a given ion, it will pass through the door and get counted. The door opening location cycles back and forth across the various points that correspond to the the masses that the operator (that would be me) told it were interesting for the experiment. The ICP-MS happily counts all the ions of each mass that it was looking for, cycling rapidly through the mass numbers, till the experiment is done. Then it spits out a report saying how many ions of each mass it saw every second.
That is when the data processing starts--I take that list, drop it into a program designed to handle large amounts of time-stamped data, follow that with the "laser log file", which explains when the laser was and was not firing, for how long, at what strength, and what spot size, and then help the program decide from the names in both files which bits of the signal correspond to the standard reference materials, which bits correspond to the maps of minerals in soapstone, and which bits are "blanks" (the ICP-MS was counting, but nothing but He was coming up the line.
Then one applies a "data reduction scheme", wherein the program compares the counts per second results for each element of the standard reference materials I analysed with the reported composition, then uses that information to calculate the amounts of those elements in my sample, reporting it in parts per million (ppm).
Because I didn't ask the ICP-MS to count every possible element (due to how long it would take--one wants it to have more than one time through the list before the laser moves on to the next spot), and there are differences between how deeply the laser digs into each mineral and into each reference material with each shot, the reported ppm doesn't tend to add up to exactly 1 million, and my results are considered "semi-quantitative", even though they are reported to several decimal places. Even so, the information is still useful, but it needs more steps...
The next step is to figure out how many phases there are in today's sample, and what they might be. For this step I look at the maps showing the relative amounts of each element in each location across the map, like this set, which shows that the S-shaped bright white crystal has more Fe, Ti, and Mn than the grey background matrix minerals (each small photo has a scale on the side, showing that the dark reds are the lowest concentration of that element, the brighter reds indicate a bit more, the dark blues a bit more, and if it gets all the way to the bright blueish green there is really quite a lot of that element). It also shows that there is more than one type of grey matrix minerals, all of which are richer in Al and Cr, but two small areas are higher in Y, Sr, and P than the rest of the map.
Now that I have some idea of how many different minerals there are in the map I need to figure out how to define their differences. One tool for this is a three-channel image, like the one labeled a. on this set of figures, which assigns P to the colour red, Al to the colour green, and Ti to the colour blue, and lets the ppm values for each of these elements in each spot contribute to the colour of the corresponding pixel. This three-channel map makes it obvious that there are three different matrix minerals, and that the large crystal is a single mineral.
Then it is a simple, if tedious, matter to apply filters to the data, in steps. At first I decided that the big crystal is all data points that are more than 6,500 ppm Ti. Indeed stopping that filter there shows that one has just selected every data point that corresponds to that phase. However, along the edges of that grain one is going to get "mixed data", where the laser ablated both the crystal of interest, and the surrounding matrix grains. To filter out those mixed data analyses I added in additional filters, to whit: P < 500; Mg < 2,000; Al < 1,000; and Si < 5,000. Once I am satisfied that the remaining grains, which cover a slightly smaller area than the crystal itself, as in figure c. in the above linked second figure are a good representation of the composition of that mineral, I am ready to do the next set of calculations on it: converting ppm into the mineral chemical formula.
For that step I prepared a spreadsheet containing the entire periodic table, showing each element by name, its atomic number, and its atomic mass. Below that I have a table set up into which I can drop the "average composition" for this mineral (which I get by using Excel's "calculate average" formula for all the data points corresponding to a single element for that mineral). Those amounts are then multiplied by appropriate atomic weight for each mineral, and then I have additional columns set up to show how many atoms of each element is present in the mineral if I set the element of interest to be exactly 1, or 2, or 3, or....
Since the main crystal in this map contains both Fe and Ti, and all of the other ingredients it contains are measured in very low amounts, the obvious choice for possible mineral is ilmenite, which contains one each atom of Fe and Ti for every three atoms of oxygen. The ICP-MS doesn't measure oxygen in the sample, so I look at the table to see how many Ti atoms are present when the total number of atoms is 2. In this case the spreadsheet replied that we get closest to a total of two atoms when there is exactly 1.1 atoms of Ti, at which point there is 0.7 atoms of Fe, and the remaining number of atoms is made up by very small amounts of Mg, Mn, Si, and Al, so my hypothesis that it is ilmenite is confirmed, those numbers being well within the reasonable range of ilmenite composition.
This approach is repeated for each of the other mineral phases, until I am happy that I know which phases are present. For this sample the defining limits I used were:
ilmenite: Ti > 6,500; P < 500; Mg < 2,000; Al < 1,000; Si < 5,000
apatite: Ti < 500; P > 2,500; Mg < 2,000; Al < 1,000; Si < 5,000
talc: Ti < 500; P < 100; Mg > 2,500; Al < 1,000; Si > 5,000
chlorite: Ti < 100; P < 100; Mg > 3,500; Al > 2,500; Si > 2,000
Once I have the list of phases, what is in them, and where they are located, I am finally ready to start writing up the sample, and finalize the set of figures for the sample.
When I first started doing this I often spent a week or more on the sample (depending on how complicated it was--some of the maps I did turned out to have more than seven different minerals). However, with my funding running out rapidly I need to get faster at this part of the project Luckily, this turns out to be possible. This morning I managed to complete several half-done other samples, and then did a whole new sample before my big lunch break.
After lunch turned out to not be so easy. The next sample on my list is one for which I failed to press the "make laser log" button on the laser before running the experiment (which, oddly enough one would have thought that making the logs should be the default setting, but that laser manufacturer doesn't agree).
Luckily, the program is still able to do a map from just the ICP-MS data, since that data comes in the form of a single row of the map at a time, with gaps of no counting in between, so it just presents the results sequentially, one row under the next. But the proportions are wrong, since it had no information of the X-Y data to work with. No problem: export the map as a jpg, and then stretch it to be the same height and width as the photo of the sample area that was mapped.
But what about the nice colour-coded maps that match the colours used for each mineral, like that in the second link above? How is that possible if one doesn't have the X-Y location data? Step one: look at the original notes from when the experiment was run, to determine that there are 17 laser lines that comprise this map. Step two look at the total number of data points exported from the program, and divide by 17. Of course it didn't come out to a whole number: 123.4 so try it anyway--in the spreadsheet number the rows 1 to 123, and then start over from 1, and so on, till you get to the bottom of the list. Call that column East. Then fill in the north column by counting each group of 123, then try plotting the data, setting the colours based on the mineral name for each row, and look at the result.
Decide that it is a mess, but it is only a bit of a mess--one can clearly see the small grain of ilmenite that looks like a cherry stem over the spinel crystal, but the two larger ilmenite grains that should be on each side of the ilmenite grain are instead between two separate halves of that grain. After staring at it and the original photo for a while, decide that he problem is that everything is offset by a few data points, which means a few the first row, then twice that many the second row, and so on till the offset is quite a bit for the last row.
But, being tired by then, my first attempt to fix it moved it even further the wrong direction.
I tried again, moving things the right way, but over compensating a little. At that point it was closing in on 22:00, and I had put in a 12 hour day. So I put it down and spent a little time with Keldor, who had just finished up his work for the day working on the sword in progress before doing my yoga and my 100 sword blows for the day, and then deciding to type up this. Now it is closing in on 01:00, and I had better get to sleep. I think I will be able to finish this sample tomorrow, and, with luck, be able to get two others done as well.
That is when the data processing starts--I take that list, drop it into a program designed to handle large amounts of time-stamped data, follow that with the "laser log file", which explains when the laser was and was not firing, for how long, at what strength, and what spot size, and then help the program decide from the names in both files which bits of the signal correspond to the standard reference materials, which bits correspond to the maps of minerals in soapstone, and which bits are "blanks" (the ICP-MS was counting, but nothing but He was coming up the line.
Then one applies a "data reduction scheme", wherein the program compares the counts per second results for each element of the standard reference materials I analysed with the reported composition, then uses that information to calculate the amounts of those elements in my sample, reporting it in parts per million (ppm).
Because I didn't ask the ICP-MS to count every possible element (due to how long it would take--one wants it to have more than one time through the list before the laser moves on to the next spot), and there are differences between how deeply the laser digs into each mineral and into each reference material with each shot, the reported ppm doesn't tend to add up to exactly 1 million, and my results are considered "semi-quantitative", even though they are reported to several decimal places. Even so, the information is still useful, but it needs more steps...
The next step is to figure out how many phases there are in today's sample, and what they might be. For this step I look at the maps showing the relative amounts of each element in each location across the map, like this set, which shows that the S-shaped bright white crystal has more Fe, Ti, and Mn than the grey background matrix minerals (each small photo has a scale on the side, showing that the dark reds are the lowest concentration of that element, the brighter reds indicate a bit more, the dark blues a bit more, and if it gets all the way to the bright blueish green there is really quite a lot of that element). It also shows that there is more than one type of grey matrix minerals, all of which are richer in Al and Cr, but two small areas are higher in Y, Sr, and P than the rest of the map.
Now that I have some idea of how many different minerals there are in the map I need to figure out how to define their differences. One tool for this is a three-channel image, like the one labeled a. on this set of figures, which assigns P to the colour red, Al to the colour green, and Ti to the colour blue, and lets the ppm values for each of these elements in each spot contribute to the colour of the corresponding pixel. This three-channel map makes it obvious that there are three different matrix minerals, and that the large crystal is a single mineral.
Then it is a simple, if tedious, matter to apply filters to the data, in steps. At first I decided that the big crystal is all data points that are more than 6,500 ppm Ti. Indeed stopping that filter there shows that one has just selected every data point that corresponds to that phase. However, along the edges of that grain one is going to get "mixed data", where the laser ablated both the crystal of interest, and the surrounding matrix grains. To filter out those mixed data analyses I added in additional filters, to whit: P < 500; Mg < 2,000; Al < 1,000; and Si < 5,000. Once I am satisfied that the remaining grains, which cover a slightly smaller area than the crystal itself, as in figure c. in the above linked second figure are a good representation of the composition of that mineral, I am ready to do the next set of calculations on it: converting ppm into the mineral chemical formula.
For that step I prepared a spreadsheet containing the entire periodic table, showing each element by name, its atomic number, and its atomic mass. Below that I have a table set up into which I can drop the "average composition" for this mineral (which I get by using Excel's "calculate average" formula for all the data points corresponding to a single element for that mineral). Those amounts are then multiplied by appropriate atomic weight for each mineral, and then I have additional columns set up to show how many atoms of each element is present in the mineral if I set the element of interest to be exactly 1, or 2, or 3, or....
Since the main crystal in this map contains both Fe and Ti, and all of the other ingredients it contains are measured in very low amounts, the obvious choice for possible mineral is ilmenite, which contains one each atom of Fe and Ti for every three atoms of oxygen. The ICP-MS doesn't measure oxygen in the sample, so I look at the table to see how many Ti atoms are present when the total number of atoms is 2. In this case the spreadsheet replied that we get closest to a total of two atoms when there is exactly 1.1 atoms of Ti, at which point there is 0.7 atoms of Fe, and the remaining number of atoms is made up by very small amounts of Mg, Mn, Si, and Al, so my hypothesis that it is ilmenite is confirmed, those numbers being well within the reasonable range of ilmenite composition.
This approach is repeated for each of the other mineral phases, until I am happy that I know which phases are present. For this sample the defining limits I used were:
ilmenite: Ti > 6,500; P < 500; Mg < 2,000; Al < 1,000; Si < 5,000
apatite: Ti < 500; P > 2,500; Mg < 2,000; Al < 1,000; Si < 5,000
talc: Ti < 500; P < 100; Mg > 2,500; Al < 1,000; Si > 5,000
chlorite: Ti < 100; P < 100; Mg > 3,500; Al > 2,500; Si > 2,000
Once I have the list of phases, what is in them, and where they are located, I am finally ready to start writing up the sample, and finalize the set of figures for the sample.
When I first started doing this I often spent a week or more on the sample (depending on how complicated it was--some of the maps I did turned out to have more than seven different minerals). However, with my funding running out rapidly I need to get faster at this part of the project Luckily, this turns out to be possible. This morning I managed to complete several half-done other samples, and then did a whole new sample before my big lunch break.
After lunch turned out to not be so easy. The next sample on my list is one for which I failed to press the "make laser log" button on the laser before running the experiment (which, oddly enough one would have thought that making the logs should be the default setting, but that laser manufacturer doesn't agree).
Luckily, the program is still able to do a map from just the ICP-MS data, since that data comes in the form of a single row of the map at a time, with gaps of no counting in between, so it just presents the results sequentially, one row under the next. But the proportions are wrong, since it had no information of the X-Y data to work with. No problem: export the map as a jpg, and then stretch it to be the same height and width as the photo of the sample area that was mapped.
But what about the nice colour-coded maps that match the colours used for each mineral, like that in the second link above? How is that possible if one doesn't have the X-Y location data? Step one: look at the original notes from when the experiment was run, to determine that there are 17 laser lines that comprise this map. Step two look at the total number of data points exported from the program, and divide by 17. Of course it didn't come out to a whole number: 123.4 so try it anyway--in the spreadsheet number the rows 1 to 123, and then start over from 1, and so on, till you get to the bottom of the list. Call that column East. Then fill in the north column by counting each group of 123, then try plotting the data, setting the colours based on the mineral name for each row, and look at the result.
Decide that it is a mess, but it is only a bit of a mess--one can clearly see the small grain of ilmenite that looks like a cherry stem over the spinel crystal, but the two larger ilmenite grains that should be on each side of the ilmenite grain are instead between two separate halves of that grain. After staring at it and the original photo for a while, decide that he problem is that everything is offset by a few data points, which means a few the first row, then twice that many the second row, and so on till the offset is quite a bit for the last row.
But, being tired by then, my first attempt to fix it moved it even further the wrong direction.
I tried again, moving things the right way, but over compensating a little. At that point it was closing in on 22:00, and I had put in a 12 hour day. So I put it down and spent a little time with Keldor, who had just finished up his work for the day working on the sword in progress before doing my yoga and my 100 sword blows for the day, and then deciding to type up this. Now it is closing in on 01:00, and I had better get to sleep. I think I will be able to finish this sample tomorrow, and, with luck, be able to get two others done as well.