Where Laser Particle Analysis Went Wrong for Coffee | by Robert McKeon Aloe | May, 2023
Coffee Data Science
Let’s do better
A few weeks ago, there was a discussion about fines in coffee where another coffee enthusiast noticed an anomaly for grounds less than 100um in diameter. I started looking more at some of my data and also found the trend he was seeing. However, I found another issue: binning.
Particle distributions are made by binning particles and forming a histogram. How many particles are within 90um and 100um in size? What is the percentage of the total or what is the percentage of the total volume for those particles? Those are the types of questions answered by particle distributions.
This is a sample particle distribution using the same binning used in a laser particle analyzer. When measuring particles, you count how many particles like within a certain bin, and each bin has a range. This forms a histogram, and typically, histograms have equal bin sizes. In my field, they are prevalent in discussing the variations of color values in an image.
However, it seems laser particle analyzers use a log scale for bin size. This is very problematic because it changes how the curve looks and how we interpret the data. This graph shows the bin size at different particle sizes.
I took a measurement from previous work, and I put it into 1, 2, 5, 10, and 20 um bins to see how it affects the signal.
This is hard to compare, so I normalized all the curves. At 1um, there is a lot of noise because there is a wide variation in the number of samples. I can test this and improve the noise using multiple measurements or for laser diffraction, more particles could be put through.
Let’s zoom in to less than 200 um.
There is a pattern that is lost by the time you get to 20 um bins. It is still there with 10 um bins.
If we go even lower to 100um, the bin sizes become a big deal:
We can see three peaks below 100 um. Typically, we see a second peak below 100 um.
Another way to view the data is to normalize by bin size. Where in the original image, there seemed to be two peaks below 100um and a tiny bump, once normalized, the third peak is more apparent.
Looking back at actual laser PSD data, we can see visually the information is clearer when normalizing by bin size (width):
Originally, it was hard to see what was going on around 40 um, and this second image shows those details without having to zoom in on the graph on the left.
Coffee Data Science
Let’s do better
A few weeks ago, there was a discussion about fines in coffee where another coffee enthusiast noticed an anomaly for grounds less than 100um in diameter. I started looking more at some of my data and also found the trend he was seeing. However, I found another issue: binning.
Particle distributions are made by binning particles and forming a histogram. How many particles are within 90um and 100um in size? What is the percentage of the total or what is the percentage of the total volume for those particles? Those are the types of questions answered by particle distributions.
This is a sample particle distribution using the same binning used in a laser particle analyzer. When measuring particles, you count how many particles like within a certain bin, and each bin has a range. This forms a histogram, and typically, histograms have equal bin sizes. In my field, they are prevalent in discussing the variations of color values in an image.
However, it seems laser particle analyzers use a log scale for bin size. This is very problematic because it changes how the curve looks and how we interpret the data. This graph shows the bin size at different particle sizes.
I took a measurement from previous work, and I put it into 1, 2, 5, 10, and 20 um bins to see how it affects the signal.
This is hard to compare, so I normalized all the curves. At 1um, there is a lot of noise because there is a wide variation in the number of samples. I can test this and improve the noise using multiple measurements or for laser diffraction, more particles could be put through.
Let’s zoom in to less than 200 um.
There is a pattern that is lost by the time you get to 20 um bins. It is still there with 10 um bins.
If we go even lower to 100um, the bin sizes become a big deal:
We can see three peaks below 100 um. Typically, we see a second peak below 100 um.
Another way to view the data is to normalize by bin size. Where in the original image, there seemed to be two peaks below 100um and a tiny bump, once normalized, the third peak is more apparent.
Looking back at actual laser PSD data, we can see visually the information is clearer when normalizing by bin size (width):
Originally, it was hard to see what was going on around 40 um, and this second image shows those details without having to zoom in on the graph on the left.
Denial of responsibility! Techno Blender is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.