Grinding Coffee with a Sifter. Diving deep into how a sifting grinds… | by Robert McKeon Aloe | Dec, 2022

Coffee Data Science

Diving deep into how a sifting grinds coffee

In my many explorations into sifting, I did not consider that the sifter itself could be modifying the coffee grounds. The act of sifting could modify a coffee particle as the grounds are rubbing back and forth. To better understand these theoretical implications, I collected some data.

My aim was two folds:

1. Does sifting grind coffee?
2. Does sifting fundamentally change the particle shape?

First, I used the Fellow Shimmy to remove less than 300um. Then I sifted the coarser particles using a Kruve sifter with a 500um screen. I did this a few times and collected some samples. I sifted initially, then for 2 minutes more, and then 4 minutes more, hence the 2m and 4m labels.

These distributions show more finer particles being removed from the coarser particles.

At the same time, the finer particles <500um start much smaller and then start to get coarser after more sifting.

I classified the particles using K-means clustering on top of Linear Binary Patterns (LBP). I applied this method to selected particle bins to better understand how the shapes are changing.

On the top are similarity matrices to show how closely different samples match each others based on the K-means clustering. On the bottom are the K-means clustering of the particles so one can see how they develop differences.

One interesting example is >500um 4m which is the last sample. Those particles cluster in a smaller number of bins thus the particle shapes are more homogenous. For 500um particles, the number of clusters and distributions remain roughly the same across samples.

We can focus on the 400um particles, and it is interesting to see how the shapes start to change over the samples.

I have other data on sifting from when I applied the salami technique to sifting. This means I sifted and took multiple samples during the sift to better understand the particles. There was definitely a shift in distributions, but I wanted to know how the particle shapes were affected. So I applied the same technique.

There was a shift over time that was minor for 300um, but for the 400um and 500um, the shapes were definitely changing over the coarse of the sift.

I focused on just 400um here, and you can see in the similarity matrix a noticeable change.

This study highlights another variable of sifting coffee. It also means that when using a sifter to measure particle distributions, the sifter is causing a modification to the particle shape and size. I’m not sure how this should be taken into account when basing conclusions compared to laser PSD’s.

This study also points to the possibility of using a sifter as a grinder. If you sifted fast enough, you might be able to have a more unimodal particle distribution.

Coffee Data Science

Diving deep into how a sifting grinds coffee

In my many explorations into sifting, I did not consider that the sifter itself could be modifying the coffee grounds. The act of sifting could modify a coffee particle as the grounds are rubbing back and forth. To better understand these theoretical implications, I collected some data.

My aim was two folds:

1. Does sifting grind coffee?
2. Does sifting fundamentally change the particle shape?

First, I used the Fellow Shimmy to remove less than 300um. Then I sifted the coarser particles using a Kruve sifter with a 500um screen. I did this a few times and collected some samples. I sifted initially, then for 2 minutes more, and then 4 minutes more, hence the 2m and 4m labels.

These distributions show more finer particles being removed from the coarser particles.

At the same time, the finer particles <500um start much smaller and then start to get coarser after more sifting.

I classified the particles using K-means clustering on top of Linear Binary Patterns (LBP). I applied this method to selected particle bins to better understand how the shapes are changing.

On the top are similarity matrices to show how closely different samples match each others based on the K-means clustering. On the bottom are the K-means clustering of the particles so one can see how they develop differences.

One interesting example is >500um 4m which is the last sample. Those particles cluster in a smaller number of bins thus the particle shapes are more homogenous. For 500um particles, the number of clusters and distributions remain roughly the same across samples.

We can focus on the 400um particles, and it is interesting to see how the shapes start to change over the samples.

I have other data on sifting from when I applied the salami technique to sifting. This means I sifted and took multiple samples during the sift to better understand the particles. There was definitely a shift in distributions, but I wanted to know how the particle shapes were affected. So I applied the same technique.

There was a shift over time that was minor for 300um, but for the 400um and 500um, the shapes were definitely changing over the coarse of the sift.

I focused on just 400um here, and you can see in the similarity matrix a noticeable change.

This study highlights another variable of sifting coffee. It also means that when using a sifter to measure particle distributions, the sifter is causing a modification to the particle shape and size. I’m not sure how this should be taken into account when basing conclusions compared to laser PSD’s.

This study also points to the possibility of using a sifter as a grinder. If you sifted fast enough, you might be able to have a more unimodal particle distribution.