Data Processing

Introduction

After you scan a few cores, you'll probably be eager to look at the results. If you have some experience working with XRF data, you may be satisfied by just looking at the raw spectra. The data is available in the form of ".spe" files, which are simply text files containing metadata and a list of counts per channel. You can open these files and plot the data yourself using a variety of methods. However, if you're like most people, you're going to want to see numbers for the elements. That will require you to process the spectra. 

When you convert your raw spectrum files to peak intensities, you're performing an interpretation. This point cannot be emphasized enough, especially for users who are new to XRF core scanning. In order to extract the intensities of different peaks and elements, you have to first tell the software what peaks and elements it should find. While it might seem like a good idea to simply tell the software to "find everything", doing so will lead to a number of problems, not the least of which is that the computer will, in fact, give you numbers for "everything". That does NOT mean, however, that all of the numbers that you get back are statistically significant or even real. In addition, you'll need to tell the software how to handle the ratios of characteristic lines, how model the background (so it can be removed from peak areas), how to calibrate the relationship between x-ray energy and detector channel, and how to model the shapes of the peaks. Each step involves judgement, trial-and-error, and a thorough understanding of the compromises that must be made when modelling an EDS spectrum. These issues will be discussed in much more depth in the model building section.

Fortunately, we have a few default processing models available to get you started. If you are looking for specific elements, please make sure to speak to our staff so we can customize some files for your specific needs. In addition, we highly recommend that all users spend some time working with the Lab Manager to build a processing model from the ground up. 

One final note: YOU ARE ACCOUNTABLE FOR YOUR INTERPRETATIONS OF THE DATA, INCLUDING THE PROCESSED INTENSITIES. We strongly recommend that all users get access to spectrum processing software and learn how to process data themselves. We will make sure you leave the lab with a set of numbers, but the initial processing results should only be a start for you. Our Lab Manager will be happy to answer any questions that you have during and after your visit.

Batch Processing

In our lab, we offer two different data processing programs for you to use: bAxil (recommended), and the older WinAxil. This document will describe how to use bAxil and its various component modules.

As the name implies, bAxilBatch is used when you want to process multiple files at once. Find the program in the start menu, and open it. The first thing that you'll want to do is to select the outputs that you'd like to receive in your data report. The elements and peaks of interest are designated in the model file, but here you'll have a chance to decide if you just want peak area or if you'd like additional information such as background counts, model standard deviations, etc. (Figure 1). If you've already loaded a template (model) file, you'll need to reload it to see the updated outputs in the column headings.

Next, you'll want to select the files that you wish to process. Click the "+" button in the top left part of the window (Figure 2), and when the file window pops up, highlight all of the files that you want (hold down the shift key) and click "Open". That will add the files to the list on the left side of the window. 

Now you'll need to select your template (model) file. Click the "..." button in the top right part of the window (Figure 2), select the desired model, and click "Open". You should see the column headers change in the empty data table.

Once you've loaded the files and you've selected a model, you can start the batch processing job. Click the start button at the top of the window (Figure 3) and you should see the data table fill up, row by row. At the bottom of the window, you'll updates as each file gets processed (Figure 3). If you need to stop the job before it finishes on its own, just click the stop button.

You'll most likely want to save the contents of the data table, so to do so, click the disk icon at the top of the window (Figure 4). Most people choose to save their results as a ".csv" file, but you have a few different options. Once you're finished, you'll want to clear the results and empty the list of files (Figure 4). Then you'll be able to set up your next batch processing job.

Viewing the Processed Spectra

It's a very good idea to spot-check your models by looking at how well they fit some of your spectra. If you only rely on batch processing, you won't be able spot systematic errors or regions of poor fit. Many scientists have spent significant amounts of time interpreting and publishing meaningless intensity data. In all cases that I am aware of, the users would have been able to avoid wasting their time if they had compared their raw spectra to their model fits.

I recommend that you choose a handful of spectra that cover different lithologies and different excitation conditions. Open a file using bAxil, and then select the template (model) that you used for your batch processing (Figure 5).

Click on the gears icon in the top right part of the window, and your spectrum will be fit using the model parameters (Figure 6). You should see your raw data as the black line, the modeled peaks as the yellow fill, and the modeled background as the green fill. Spectral artifacts (if included in the model and turned on in the display) will appear as red and brown areas. At the bottom of the window, you'll see the numerical results for the different peaks.

You can adjust the display to highlight different parts of the spectrum, switch the x-axis back and forth between channel and energy, modify the scales, etc. (Figure 7). You'll want to make sure that you zoom in to different regions and look at how well the model is fitting different peaks. 

Take note of any places where the peak labels might be off-center relative to the maximum peak heights. Look for regions where the background doesn't fit very well. All of these features point to problems with the model fits. Then, you should take note of which peaks are in those regions. Those are the peaks that you should be cautious about during your interpretation. 

If you find any problems and / or want to make modifications to your model, please see the model building section or talk to the Lab Manager.

Basic Data QA / QC

Once you have your data in a spreadsheet, you're going to want to do some basic QA / QC before you take it any further. We'll talk about this topic a bit more in the quality assurance section, so we're just going to cover the basics here.

First, you should make sure that you have, at a minimum, the total spectrum throughput, the peak area, and the peak area standard deviation. If you don't have one or more of these values, you should modify the bAxilBatch outputs (Figure 1) and reprocess your data.

Second, look at the spectrum throughput data for all of your points. Throughput is a measure of the total counts per second that reach your detector. This number should be in the range of ~150-300k for a typical spot size (~1cm by 1cm). If you use a smaller spot size, make sure you run the standards at the same spot size so you have a benchmark for a good throughput at each of your excitation conditions. If you have analyses with much smaller throughput values, it likely means that the prism didn't fully land on the sample surface or the prism landed on cracks. You may want to discard those points or at least be wary of them.

Next, look through the peak area values. Look for numbers that seem too high or too low. While peak intensity is only partly related to element concentration, you should still expect larger peaks for major elements and smaller peaks for minor and trace elements. If you see a peak with millions of counts, and you know the element should only be present in trace concentrations, you should open the spectrum files in bAxil and investigate the model. You may also see negative peak values. This can happen in areas with strong peak interference, and you should ignore those elements. The only peak that we hope is negative is the argon peak. If argon is close to zero or is positive, you either had air in the prism (hole in film) or you measured some air instead of sample. Consider discarding those points as well.

Finally, you'll want to note that any peak values that are too low to be meaningful. This includes other peaks that are negative and peaks that are less than one or two standard deviations above zero (see quality assurance).