Coffee Data Science
Coffee Grind Differences with Variable RPM
A first look at RPM using pattern recognition
I recently borrowed a Molar Z with variable RPM. I wanted to understand how lower RPM affects the grind, so I collected some data and applied some pattern recognition.
Pattern recognition is a powerful tool for image processing. This tool comes in handy when talking about coffee grinders. Often, we have these particle distribution curves for coffee, but there’s not a great connection between particle distributions and taste/extraction. Two grinders could produce very similar particle distributions but have greatly different tastes.
Particle shape analysis hasn’t been widely used to help distinguish grinders, and I think it is a useful tool.
In general, I found the Molar outperforms the Niche in taste and extraction yield at 60 RPM. I also found that 400 RPM lowered taste scores. I looked at one coffee grind, and I ground enough coffee for each sample to make sure there wasn’t overlap due to retention. The molar is not quite a zero-retention grinder, so extra care was taken to purge the grinder.
From a particle distribution standpoint, the same grind setting at different RPM’s was pretty similar with a shift. The Niche also has the same distribution for the Molar at 120 RPM.
Looking at the cumulative distribution, the grind settings are close aside from 60 RPM.
Pattern Recognition
Using k-means clustering on Linear Binary Patterns (LBP), we can look at the actual particle shapes. I split this by particle size so the sizes are roughly the same. 120 RPM seems closer to the Niche than anything else.
We can look at two cases for baseline, the Niche and the Molar at 400 RPM. The Molar at 400RPM is more similar to 60 and 120 RPM than the Niche. So there is something interesting to the particle shapes. I thought I would see some more variety across the bin sizes, but this data doesn’t show it.
If we sort the feature vectors of all the particles for each bin, we can see that even though the Niche is close to the Molar 120 RPM, there are some large differences.
We can compare grinders and sizes in one more way because maybe certain size particles would compare better to each other.
I didn’t know what this work would bring, but at least in this sample of coffee and grind, the differences in particle distribution don’t quite show the differences seen in the particle shape analysis. The difference at 120 RPM is particularly interesting, and I think more analysis could be done on grinder RPM.
If you like, follow me on Twitter, YouTube, and Instagram where I post videos of espresso shots on different machines and espresso related stuff. You can also find me on LinkedIn. You can also follow me on Medium and Subscribe.
