Comparing Laser Diffraction to Imaging of Coffee Particles | by Robert McKeon Aloe | Apr, 2023

Coffee Data Science

Splitting beans

For the past two years, I have been measuring coffee ground distributions using imaging. This has been so useful in diving deeper into what is going on with a coffee grinder, and I really enjoyed the challenge of the image processing required. However, I was always curious how these measurements compared to laser diffraction. Laser diffraction is typically very expensive, but I had an opportunity to run a few samples and compare to my method.

Laser diffraction for particle distribution uses a laser and a diffraction grating to more accurately measure small particles. It outputs the average diameter of the particle, and a typical laser particle analyzer uses a feed tube to measure particles sequentially. A probability distribution is then made based on the volume of the particles. These machines cost around $100,000, which is not economical for coffee hobby explorations.

Coffee particles can also be measured using imaging. This involves putting a sample of coffee on a piece of paper, spreading out / declumping the coffee grounds, and taking a calibrated image. A calibrated image means the image plane can translate to a real measurement of size.

I really like being able to do shape analysis, and I think that is more informative than particle diameter. However, I would love to have both the high accuracy of laser imaging and the particle shape information from camera imaging.

Let’s look at some data. I have both ground coffee and spent coffee. Spent coffee clumps much less, but it is not always representative of the grinder.

In looking at all the bins, it is clear that the cumulative distribution can not be looked at without removing the lower bins for the laser technique. In this way there is more alignment, but a better test could have also sifted the coffee grounds to be able to insure the measurement between laser and imaging was the same particle type.

In my usual imaging measurement, I use the minimum diameter as that is what a sifter will measure. However, to get closer to the lasered measurement, I looked at minimum and average diameter. The average diameter seems to fit a bit better for larger particles over 100um, but the minimum diameter fits better for particles less than 100um.

I will stick with average diameter because that is closer to what the laser measurement is providing. I can look at a grind before and after the shot. The imaged data shows the post-shot particles get a little finer while the lasered shows the opposite. I wonder if this is tied to the accuracy at the lower readings.

I plotted the cumulative percent of particles against each other, and this would be a flat line if they were the same.

These results still show a gap in performance, but imaging is seemingly within range of lasered data. I don’t usually compare these distributions to laser diffraction measurements, so within the same method, the variables are more controlled than comparing to another measurement type.

I still prefer to have the precision of the laser measurement for the 3D shape as well as the actual 3D shape to better understand a grinder, and that may one day happen if someone made a desktop laser diffraction particle analyzer.

Coffee Data Science

Splitting beans

For the past two years, I have been measuring coffee ground distributions using imaging. This has been so useful in diving deeper into what is going on with a coffee grinder, and I really enjoyed the challenge of the image processing required. However, I was always curious how these measurements compared to laser diffraction. Laser diffraction is typically very expensive, but I had an opportunity to run a few samples and compare to my method.

Laser diffraction for particle distribution uses a laser and a diffraction grating to more accurately measure small particles. It outputs the average diameter of the particle, and a typical laser particle analyzer uses a feed tube to measure particles sequentially. A probability distribution is then made based on the volume of the particles. These machines cost around $100,000, which is not economical for coffee hobby explorations.

Coffee particles can also be measured using imaging. This involves putting a sample of coffee on a piece of paper, spreading out / declumping the coffee grounds, and taking a calibrated image. A calibrated image means the image plane can translate to a real measurement of size.

I really like being able to do shape analysis, and I think that is more informative than particle diameter. However, I would love to have both the high accuracy of laser imaging and the particle shape information from camera imaging.

Let’s look at some data. I have both ground coffee and spent coffee. Spent coffee clumps much less, but it is not always representative of the grinder.

In looking at all the bins, it is clear that the cumulative distribution can not be looked at without removing the lower bins for the laser technique. In this way there is more alignment, but a better test could have also sifted the coffee grounds to be able to insure the measurement between laser and imaging was the same particle type.

In my usual imaging measurement, I use the minimum diameter as that is what a sifter will measure. However, to get closer to the lasered measurement, I looked at minimum and average diameter. The average diameter seems to fit a bit better for larger particles over 100um, but the minimum diameter fits better for particles less than 100um.

I will stick with average diameter because that is closer to what the laser measurement is providing. I can look at a grind before and after the shot. The imaged data shows the post-shot particles get a little finer while the lasered shows the opposite. I wonder if this is tied to the accuracy at the lower readings.

I plotted the cumulative percent of particles against each other, and this would be a flat line if they were the same.

These results still show a gap in performance, but imaging is seemingly within range of lasered data. I don’t usually compare these distributions to laser diffraction measurements, so within the same method, the variables are more controlled than comparing to another measurement type.

I still prefer to have the precision of the laser measurement for the 3D shape as well as the actual 3D shape to better understand a grinder, and that may one day happen if someone made a desktop laser diffraction particle analyzer.