New method to measure grinder fines percentage?

Grinders are one of the keys to exceptional espresso. Discuss them here.
crwper
Posts: 131
Joined: 3 years ago

#1: Post by crwper »

For the last couple of months, I've been measuring the density of the tamped puck for my morning espresso by using a caliper to measure the depth of the puck's top surface. By varying grind size, I got a plot of tamped density vs. grind size like this:



I was able to come up with a simple model which fits this data beautifully. I've described the model in detail here:

https://quantitativecafe.com/2021/10/23 ... -grinding/

The end result is a fit that looks like this:



This fit basically predicts a transition between a puck dominated by the volume of the boulders, vs. a puck dominated by the volume of the fines (the transition occurs at the little bump in the plot).

What's really neat about this, I think, is that it gives a way to infer the percentage of fines/boulders using simple density measurements. This is described at the end of the blog post linked above. There are a couple of follow-up experiments I'd like to do next:

First, I'd love to see what these plots look like when a low-fines burr set is used. The initial experiments were done on a Eureka Mignon Specialita. I would think the little peak in the plot should move left, and the predicted "thickness of the brittle layer" should get smaller.

Second, I'm planning to split a single bag of beans into two parts, freeze one, and keep the other on the counter. Then I can perform the same experiments with both parts. It seems like common wisdom is that grinding frozen beans makes them shatter more, so I would expect the fines percentage to go up compared to beans stored on the counter.

I've got a few different experiments on the go, so I figured I might as well set up a blog to organize them. As much as anything, I find its useful to formalize ideas like this, as it forces me to go into more detail than I otherwise might. I hope it's also useful for someone else, and of course I welcome any constructive criticism.

I'd love to hear any thoughts you all might have on measuring fines percentage this way. If you have a low-fines burr set and access to a caliper with a depth gauge, I would be very interested to see the results of similar tests on those burrs.

jpender
Posts: 3863
Joined: 11 years ago

#2: Post by jpender »

My thought is that it doesn't seem to me that you have enough data to show that there really is the transition elbow you have drawn in. I know you would like the data to match your theory but I doubt the accuracy of your density measurement would allow for that without many more samples.

User avatar
Jeff
Team HB
Posts: 6808
Joined: 19 years ago

#3: Post by Jeff »

Definitely an interesting hypothesis.

"Boulders", to me, are those significantly larger than the primary mode of the distribution. It seems as though you might be using that term to describe the target grind size, rather than the outliers.

crwper (original poster)
Posts: 131
Joined: 3 years ago

#4: Post by crwper (original poster) »

jpender wrote:My thought is that it doesn't seem to me that you have enough data to show that there really is the transition elbow you have drawn in. I know you would like the data to match your theory but I doubt the accuracy of your density measurement would allow for that without many more samples.
It's certainly possible. I've got similar fits for a handful of different beans/roasts.

One thing I'd really like to do is to walk through the entire grind range in steps of, say, 0.5, and do 3-5 repetitions at each grind setting. That would give a much better indication if the data is following the model. I've had a bit of a hard time justifying grinding 2 lb of beans straight into the compost, so I'm looking for other ways to test the model.

That's part of the reason I decided to reach out here. If someone else is able to replicate the result with a different grinder/measurement equipment, it would help firm things up a bit.
Jeff wrote:"Boulders", to me, are those significantly larger than the primary mode of the distribution. It seems as though you might be using that term to describe the target grind size, rather than the outliers.
It's possible I'm using the term incorrectly. My understanding has been that boulders are the larger particles which conform to the grind setting. How would particles significantly larger than the spacing of the burrs make it through?

Is there a different term you would use to describe particles which conform to the primary mode of the size distribution?

User avatar
another_jim
Team HB
Posts: 13871
Joined: 19 years ago

#5: Post by another_jim »

This is good work, but I don't think there's enough data to justify the degrees of freedom taken up by the notch in your model. It also doesn't accord with the basic geometry of grinding.

Fines amount to only 2% to 3% of the total mass, rising only slightly, and continuously, as the grind is made finer. But each fine particle has roughly 1000th to 100th the mass of the regular particles, so there's lots of them to fill in the gaps between the coarse particles.

The mental picture of a puck is a stone wall with irregularly shaped stones of coarse particles bound by a cement of fines. As the irregular stones get smaller, the gaps between them get smaller too, and the fines get more efficient at filling them. But since there are more small stones, there are also more gaps. The equation for the total gap volume, since we are in three dimensions, would be cubic polynomial without turning points. The density is the inverse of the total gap volume. So if this mental picture is correct, the correct model would be derived using a GLM with a cubic polynomial and an inverse link.

The slight curve in your data (without the hitch) is completely compatible with that geometry derived model.
Jim Schulman

crwper (original poster)
Posts: 131
Joined: 3 years ago

#6: Post by crwper (original poster) »

another_jim wrote:The slight curve in your data (without the hitch) is completely compatible with that geometry derived model.
That was actually an intermediate model I experimented with--basically, the fines disappear into the gaps, and contribute to the mass but not the volume.

At that point, the data in the "bend" was mostly from my morning coffee, and the data outside the bend was obtained using similar methods, but not as part of my morning routine, so I wondered if it might just be some subtle difference in my measurement. However, more consistent measurements (i.e., grinding coffee just to measure the density) hasn't eliminated the "bend", so at this point I'm inclined to say it's real.

That said, one of the most beautiful things about science, for me, is the opportunity to be wrong. I think it would be fair to say that if I ground a couple pounds of coffee straight into the compost, obtaining several points for each grind setting, and if we still see the "bend" in that case (with a deviation larger than experimental error), then either there is an issue with the measurement apparatus, or the simpler model can't be right.

Does that sound fair? If it's providing a meaningful separation between two models, maybe I can justify grinding away a couple pounds of coffee. :-)

User avatar
cafeIKE
Posts: 4704
Joined: 18 years ago

#7: Post by cafeIKE »

Very interesting.

Cupla things:

Does the basket have vertical walls and right angle join at bottom and how tight is the tamper?

The sphere model maybe a bit suspect as coffee isn't.

Is it possible to derive your graph from particle distribution data from the Titan Grinder Project?

To what extent does massaging the coffee affect compression? Think of a scoop of fresh river rock in the back of a pickup. Drive 50 miles over rough roads and all sand will be on the bottom. AKA Cashew Tin Shake. Anyone submitting data would need to follow a protocol.

Regarding 'morning' vs 'not morning' grind. Used to be pretty normal for me to adjust a grinder through out the day due to environmental changes.

If grinding a couple of pounds on the go, it might be well to have a protocol. Say 5 shots in the bin and then grind a shot every N minutes so as to not introduce thermal effects...

jpender
Posts: 3863
Joined: 11 years ago

#8: Post by jpender »

crwper wrote:That said, one of the most beautiful things about science, for me, is the opportunity to be wrong. I think it would be fair to say that if I ground a couple pounds of coffee straight into the compost, obtaining several points for each grind setting, and if we still see the "bend" in that case (with a deviation larger than experimental error), then either there is an issue with the measurement apparatus, or the simpler model can't be right.
What is the experimental error? On your page about measuring density you have this graph:



Does that represent the spread (for each coffee) in a handful of identically ground samples?


Also, what about puck prep? Is it a manual process?

User avatar
Jake_G
Team HB
Posts: 4295
Joined: 6 years ago

#9: Post by Jake_G »

another_jim wrote:Fines amount to only 2% to 3% of the total mass, rising only slightly, and continuously, as the grind is made finer. But each fine particle has roughly 1000th to 100th the mass of the regular particles, so there's lots of them to fill in the gaps between the coarse particles.
I'm not sure there is data to support that fines production is this low when grinding at espresso settings.

The image below is two (technically 3, but I digress) particle density plots from the same grinder, shifting from a "target" grind size of 300 um to roughly 150 um.


Note the portion of fines.

I like the cumulative distribution plots for 'sizing up' the fines portions. If you look at where the red lines cross over the 100 um particle size, you'll see that roughly 30% of the total dose is already accounted for (this is 30% by volume, which is analogous to mass, not particle count).

Swinging over to the grey lines, you can see that when the target size is reduced, the fines portion skyrockets to around 65%.

Where are you seeing only 2-3% fines?
LMWDP #704

crwper (original poster)
Posts: 131
Joined: 3 years ago

#10: Post by crwper (original poster) »

cafeIKE wrote:Does the basket have vertical walls and right angle join at bottom and how tight is the tamper?
The walls appear to be quite vertical, but there is a small fillet on the bottom of the basket. One of the first tests I did was to add varying amounts of ground coffee to the basket, then measure the depth below the rim. Here's the plot:



On the far left, you have a measurement of the depth of the basket with no coffee in it. On the right, varying amounts of ground coffee have been placed in the basket and tamped. What struck me here is that the linear fit through the points on the right (omitting the points at the far left) passes nearly through the points on the left. This leads me to believe that the assumption of a cylindrical basket is justified.
cafeIKE wrote:The sphere model maybe a bit suspect as coffee isn't.
There's an old joke about physicists assuming "spherical cows in a vacuum". You're of course right that coffee grains aren't spherical. From a modeling standpoint, it's a classic assumption because it makes the math easy. I think this does come up, e.g., in the fact that the packing density of the grains is higher than one would expect from spherical particles.
cafeIKE wrote:Is it possible to derive your graph from particle distribution data from the Titan Grinder Project?
I'll definitely have to take a closer look at that thread. Perhaps there will also be some ideas for an improved protocol for this kind of testing.
cafeIKE wrote:To what extent does massaging the coffee affect compression? Think of a scoop of fresh river rock in the back of a pickup. Drive 50 miles over rough roads and all sand will be on the bottom. AKA Cashew Tin Shake. Anyone submitting data would need to follow a protocol.

Regarding 'morning' vs 'not morning' grind. Used to be pretty normal for me to adjust a grinder through out the day due to environmental changes.

If grinding a couple of pounds on the go, it might be well to have a protocol. Say 5 shots in the bin and then grind a shot every N minutes so as to not introduce thermal effects...
Agreed. I think one of the biggest reasons for the spread within the "bump" is that those points are collected over a couple of weeks (although points collected within this range of grind settings as part of testing also show the same bump). This is another reason it might be good to collect a volume of data with this particular experiment in mind.

I'm pretty consistent about puck preparation. Roughly speaking, I would say the protocol here has been:
  • Change the grind setting.
  • Use bellows to blow out grounds until nothing is coming out.
  • Grind at the specified setting.
  • Blow out remaining grounds into the portafilter. I think this ensures that we're collecting all the products of grinding.
  • Use WDT in a "spirograph" pattern, slowly raising the tool so that a level bed remains.
  • Tamp. I don't use a calibrated tamper, as my feeling is that the grounds will compact until the bed is packed, then produce a much greater force against further tamping.
  • Zero the caliper against the counter.
  • Measure depth from the rim of the basket to the center of the top surface of the tamped puck.
I'd love to hear any thoughts on how this protocol could be improved.

Post Reply