TGP II: Particle Distribution Analysis of Grinder Adjustments - Interim Results
- another_jim
- Team HB
{Added in 2012: This thread shows that laser diffractometers are not quite sensitive enough to tell the difference between the same grinder making espresso at slightly different doses, or different grinders set to make the shots with the same dose, time and brew ratio. Laser diffractomeerts size particles by measuring the average dimaeter of the shadow they cast. This is too approximate to work for the small size differences in the working range of espresso grinders. However, there's alot of cool graphs for those who are interested}
My apologies for the delay. Kendall finished the laser sizing very quickly; but the results have been head scratchers, so they took a while to analyze.
I am going to put up two posts.
And here it is -- the most complete set of espresso grind distributions ever collected:
Baratza Vario, Single, Basaric
Versalab M3, Single, Basaric
Ascaso I2, Hopper, Campbell
Lelit PL53, Hopper, Cunha
Macap MXK, Single, Leighton
Compak K10, Single, Schulman
Macap M7K, Single & Hopper, Campbell
Compak K10, Single & Hopper, Cobo
Mazzer Kony-E, Single & Hopper, Ivanitzky
Mazzer Robur-E, Hopper two pair, Lampert
Compak K-10, Single & Hopper, Lundgaard
Baratza Vario, Single & Hopper, Lundgaard
Cimbali Junior, Single & Hopper, Wookie
My apologies for the delay. Kendall finished the laser sizing very quickly; but the results have been head scratchers, so they took a while to analyze.
I am going to put up two posts.
- This first one contains 12 graphs for the 12 grinders submitting samples. Each graph shows the distributions of the two or four samples submitted for that grinder. The title of each graph shows what these are. Glancing at these graphs, you'll see where the hitch lies -- the 18 gram samples aren't always coarser than the 15 gram samples
- The second shows a single analytical graphic displaying the 18 versus 15 gram dose differences for all grinders. Based on this, I discuss the problems with the data and how we intend to proceed.
And here it is -- the most complete set of espresso grind distributions ever collected:
Baratza Vario, Single, Basaric
Versalab M3, Single, Basaric
Ascaso I2, Hopper, Campbell
Lelit PL53, Hopper, Cunha
Macap MXK, Single, Leighton
Compak K10, Single, Schulman
Macap M7K, Single & Hopper, Campbell
Compak K10, Single & Hopper, Cobo
Mazzer Kony-E, Single & Hopper, Ivanitzky
Mazzer Robur-E, Hopper two pair, Lampert
Compak K-10, Single & Hopper, Lundgaard
Baratza Vario, Single & Hopper, Lundgaard
Cimbali Junior, Single & Hopper, Wookie
Jim Schulman
- another_jim (original poster)
- Team HB
As you can see from the set of graphs, sometimes there appear to be more coarse particles in the finer ground 15 gram samples. What is the source of the problem? Kendall did two sample test/retests and we had two pairs of identical samples submitted by Lampert (thanks!). These test/retest of identical samples allow us to establish a no difference zone. All the anomalous samples fall int this no difference zone. So do some of the samples that are not anomalous. In the end, only five pairs have enough difference between the 15 gram and 18 gram set to be significant, and on all these the 18 gram sample is coarser.
The following graphic shows all the grinder results in one result. To do this, I summarised the distribution data into two parameters - the proportion of coarse particles to fine particles (by volume) and the average size of the coarse particles. The graphic displays the difference between the 18 and 15 gram sample of each pair as a letter a through t.
There are also four red # showing the test/retest difference of identical samples (they have all been put in the upper right quadrant for convenience sake. The circle whose center lies at the intersecting axes and whose diameter passes through the outer points is the no difference zone. !5 of the twenty samples fall in this zone. The five significant samples all show various degrees of either larger coarse particles or a higher proportion of coarse particles. I have placed the legend showing which letter belongs to which grinder below the graphic.
Discussion: Why do fifteen of the samples show no real difference in grind size? Kendall tested and retested samples, and this shows that both he and the machine are operating properly. People submitting no difference samples followed the proper procedure, and set the grinders coarser, there is no change of massive errors on their part, i.e fifteen mislabels or forgetting to set the grinders etc.
There is also no gap in the analysis. The fines do not change in size when the grind is adjusted, and looking at the raw data as counts or areas does not yield anything different either -- the same five pairs show up as different, and the other fifteen as the same, no matter how one looks at them.
This leaves just one hypothesis. The laser sizer isn't all it is cracked up to be. In fact, it usually misses the effect of grind adjustments in the espresso range. How can this be? Here's the theory. Several lasers are used so that each particle casts several shadows. From this, a particle size is computed that is the diameter of a sphere that would cast shadows of equal area. This conversion may not be right when investigating particle packing, since if the particles are not at all spherical, but say flakes, they would pack based on their thinnest dimension, not their average one.
To check this theory, Kendall will test one of the big difference and one or two of the no difference samples on more exotic and far less available equipment (I got the mass production SEM, but there is other stuff beyond my ken). These are capable of giving shape sensitive distribution data. These may show whether the laser sizer muffed the no difference pairs, or whether we are all deluded when we think adjusting our grinders makes a difference.
The following graphic shows all the grinder results in one result. To do this, I summarised the distribution data into two parameters - the proportion of coarse particles to fine particles (by volume) and the average size of the coarse particles. The graphic displays the difference between the 18 and 15 gram sample of each pair as a letter a through t.
There are also four red # showing the test/retest difference of identical samples (they have all been put in the upper right quadrant for convenience sake. The circle whose center lies at the intersecting axes and whose diameter passes through the outer points is the no difference zone. !5 of the twenty samples fall in this zone. The five significant samples all show various degrees of either larger coarse particles or a higher proportion of coarse particles. I have placed the legend showing which letter belongs to which grinder below the graphic.
a K10 Hopper Cobo
b M7K Single Campbell
c M7K Hopper Campbell
d MXK Single Leighton
e K10 Hopper Lundgaard
f K10 Single Lundgaard
g Var Single Lundgaard
h Var Hopper Lundgaard
i CJR Single Wookie
j CJR Hopper Wookie
k K10 Single Schulman
l K10 Single Cobo
m VM3 Single Basaric
n Var Single Basaric
o Lel Hopper Cunha
p Kon Hopper Ivanitzky
q Kon Single Ivanitzky
r Rob Hopper Lampert
s Rob Hopper Lampert
t AI2 Hopper Campbell
# Test/retest of same sample
There is also no gap in the analysis. The fines do not change in size when the grind is adjusted, and looking at the raw data as counts or areas does not yield anything different either -- the same five pairs show up as different, and the other fifteen as the same, no matter how one looks at them.
This leaves just one hypothesis. The laser sizer isn't all it is cracked up to be. In fact, it usually misses the effect of grind adjustments in the espresso range. How can this be? Here's the theory. Several lasers are used so that each particle casts several shadows. From this, a particle size is computed that is the diameter of a sphere that would cast shadows of equal area. This conversion may not be right when investigating particle packing, since if the particles are not at all spherical, but say flakes, they would pack based on their thinnest dimension, not their average one.
To check this theory, Kendall will test one of the big difference and one or two of the no difference samples on more exotic and far less available equipment (I got the mass production SEM, but there is other stuff beyond my ken). These are capable of giving shape sensitive distribution data. These may show whether the laser sizer muffed the no difference pairs, or whether we are all deluded when we think adjusting our grinders makes a difference.
Jim Schulman
- cafeIKE
"Legal disclaimer and copyright: This content and all its parts are copyrighted, all rights reserved. Content including all text and images in part or in whole may not be reproduced in any electronic or printed medium without prior permission."
The Zip is by permission of another_jim [Jim Schulman] and H-B and is for personal non-commercial use only.
For convenience, the grinder graphs are zipped here for easy download to print.
The Zip is by permission of another_jim [Jim Schulman] and H-B and is for personal non-commercial use only.
For convenience, the grinder graphs are zipped here for easy download to print.
Ian's Coffee Stuff
http://www.ieLogical.com/coffee
http://www.ieLogical.com/coffee
- SandBaggerOne
Nice job, guys. If nothing else we have a nice comparison of size distributions analyzed on the same instrument, even if it missed the espresso range adjustments. I like the look of all of the Vario profiles, that's for sure. It appears to compare favorably to the M3 and K10 in terms of fines production.
Cheers,
Colin
Cheers,
Colin
LMWDP #310
My guess is that flakes tumbling in the fluid stream impart a fuzzy distribution and that what is important to espresso extraction is more fine than the resolving power of a top of the line machine. If I had a mono-disburse flake slurry where every flake was absolutely identical, I would still see a relatively broad peak due to the differing ways the light can encounter a rotating flake. I use 100nm latex spheres to verify the calibration of the machine and that produces a sharp single peak, but thats because the machine is treating spheres like spheres.
BET is an interesting technique that will tell me the surface area of a material which increases as an object is split into smaller particles. This is a useful parameter in my lab where we mill materials to reduce their size. It uses nitrogen adsorption/desorption to compute the surface area. I'm concerned that there will be more internal surface area in the cellular matrix than will be exposed due to finer grinding, we shall see. The N2 molecule is pretty small, hard to guess how far it will penetrate a bean. Usually I measure nano ceramics and metals with no internal void space.
BET is an interesting technique that will tell me the surface area of a material which increases as an object is split into smaller particles. This is a useful parameter in my lab where we mill materials to reduce their size. It uses nitrogen adsorption/desorption to compute the surface area. I'm concerned that there will be more internal surface area in the cellular matrix than will be exposed due to finer grinding, we shall see. The N2 molecule is pretty small, hard to guess how far it will penetrate a bean. Usually I measure nano ceramics and metals with no internal void space.
- TrlstanC
After looking over all the graphs, I've got a few thoughts, in no particular order at this point:
1. A lot of the grinder plots looked pretty much like I would've expected, a larger proportion of slightly coarser particles in the 18g sample. And there were some patterns to be found between single dosing and hopper grinding. Although given the small numbers and amount of noise it's tough to draw conclusions.
2. There seemed to be two types of grinders, the ones that had a lot of coarse particles and the ones that had a similar ratio of coarse to fine particles. For example, if you showed me a bunch of new samples and asked if each one "might be a Vario" or "definitely wasn't a Vario" I'm pretty confident I could pick correctly, same thing goes for the K10s.
3. Even though a lot of the samples fall within the range of essentially being identical, I'm surprised to see how many fell in the lower left quadrant, which is if I'm reading the graph correctly, basically means it was a finer grind for the 18g sample. There also seemed to be more single dose samples in that area (although with such a small sample size it's certainly not statistically significant).
A couple questions:
1. Is the strong positive correlation in the summary graph a function of particles in general, or could we see a different pattern with different ways of creating particles. It appears to say that as coarse particles get bigger, their number increases as well, right? Is that just a function of grinding in general, or is it a result assumptions made when measuring?
2. The "no difference in size" zone shown by the red # shows the variation in measured sizes for identical samples, so basically we assume that a sample that's actually at (0,0) on the graph could show up anywhere within the radius of the furthest #, right? Why aren't there any samples in the top left quadrant? It seems like with the number of samples, and the possible variance in measurement, at least one would end up over there by chance. Again, is this a function of the way the particles were ground, or is it a result of the way they were measured?
1. A lot of the grinder plots looked pretty much like I would've expected, a larger proportion of slightly coarser particles in the 18g sample. And there were some patterns to be found between single dosing and hopper grinding. Although given the small numbers and amount of noise it's tough to draw conclusions.
2. There seemed to be two types of grinders, the ones that had a lot of coarse particles and the ones that had a similar ratio of coarse to fine particles. For example, if you showed me a bunch of new samples and asked if each one "might be a Vario" or "definitely wasn't a Vario" I'm pretty confident I could pick correctly, same thing goes for the K10s.
3. Even though a lot of the samples fall within the range of essentially being identical, I'm surprised to see how many fell in the lower left quadrant, which is if I'm reading the graph correctly, basically means it was a finer grind for the 18g sample. There also seemed to be more single dose samples in that area (although with such a small sample size it's certainly not statistically significant).
A couple questions:
1. Is the strong positive correlation in the summary graph a function of particles in general, or could we see a different pattern with different ways of creating particles. It appears to say that as coarse particles get bigger, their number increases as well, right? Is that just a function of grinding in general, or is it a result assumptions made when measuring?
2. The "no difference in size" zone shown by the red # shows the variation in measured sizes for identical samples, so basically we assume that a sample that's actually at (0,0) on the graph could show up anywhere within the radius of the furthest #, right? Why aren't there any samples in the top left quadrant? It seems like with the number of samples, and the possible variance in measurement, at least one would end up over there by chance. Again, is this a function of the way the particles were ground, or is it a result of the way they were measured?
- another_jim (original poster)
- Team HB
Good question.
There is a strong positive correlation between mean coarse particle size and the proportion of coarse to fine particles. This is expected: The fines are like bread crumbs, the same size no matter how thin or often you slice the loaf. So if the coarse particles are larger, they were created by fewer breakages of the bean, and therefore also mean less fines. This is why you don't see any grinder points in the off quadrants.
Since all the anomalous sets in the bottom left quadrant are in the zero zone (as are all the hopper/single differences), we can only say that five sample pairs from four grinders (two K10s, one Kony-E and one Cimbali Junior) produced definitively different sized samples for 15 and 18 grams.
But this is impossible -- all the samples pairs have to be different, otherwise the grinders they came from don't work. This means the laser sizers don't have enough resolution for the job, and that the data that has been previously published or posted has questionable relevance to grind or grinder quality.
This is an interesting, albeit unilluminating result; since it contradicts the implicit claims in the literature that laser sizers are adequate for analyzing ground coffee and coffee grinders.
There is a strong positive correlation between mean coarse particle size and the proportion of coarse to fine particles. This is expected: The fines are like bread crumbs, the same size no matter how thin or often you slice the loaf. So if the coarse particles are larger, they were created by fewer breakages of the bean, and therefore also mean less fines. This is why you don't see any grinder points in the off quadrants.
Since all the anomalous sets in the bottom left quadrant are in the zero zone (as are all the hopper/single differences), we can only say that five sample pairs from four grinders (two K10s, one Kony-E and one Cimbali Junior) produced definitively different sized samples for 15 and 18 grams.
But this is impossible -- all the samples pairs have to be different, otherwise the grinders they came from don't work. This means the laser sizers don't have enough resolution for the job, and that the data that has been previously published or posted has questionable relevance to grind or grinder quality.
This is an interesting, albeit unilluminating result; since it contradicts the implicit claims in the literature that laser sizers are adequate for analyzing ground coffee and coffee grinders.
Jim Schulman
- shadowfax
Shoot, now I'm aching to redo all this with a basket that takes to a 15 g dose a little better than the LM double basket. That's certainly the other possible confounding explanation I can think of (beyond the conjecture that the instrumentation isn't accurate enough to consistently read the changes), and maybe both are at work in the noise of this data.
What next? I'm reading some interesting thoughts, though I must admit I am pretty horrible at statistics, but I'm getting the feeling that this data is maybe not terribly interesting as it stands. If we feel the adjustment is too small for the instruments available to reliably differentiate, should we refine the experiment parameters? Maybe broaden the grind step and use baskets that might be kinder to a low dose and high dose (e.g., 15 g normale and 19g ristretto in a Synesso 14g ridgeless double basket)?
Just my off the cuff thoughts. Thank you to all of you who helped with this, especially Kendall. This is seriously interesting even if it's not as immediately informative as I'd hoped. and by hoped, I mean fantastically dreamed.
What next? I'm reading some interesting thoughts, though I must admit I am pretty horrible at statistics, but I'm getting the feeling that this data is maybe not terribly interesting as it stands. If we feel the adjustment is too small for the instruments available to reliably differentiate, should we refine the experiment parameters? Maybe broaden the grind step and use baskets that might be kinder to a low dose and high dose (e.g., 15 g normale and 19g ristretto in a Synesso 14g ridgeless double basket)?
Just my off the cuff thoughts. Thank you to all of you who helped with this, especially Kendall. This is seriously interesting even if it's not as immediately informative as I'd hoped. and by hoped, I mean fantastically dreamed.
Nicholas Lundgaard
- TrlstanC
I think a second round would make sense, and try changing one or more of our constants to see if different parameters would yield more distinct samples. I'd certainly order another bag or two of beans if everyone else (especially Kendall) thought it was worth the effort.
Edit: I also just went back and looked at the 5 samples with large enough changes to be meaningful, and did some quick "holding a ruler up to the screen" math with the idea of comparing not the ratio of coarse to fine particles, but the total number in each sample. If we were graphing the total number of particles it would move the black lines up by about 20% (since there are 20% more particles in an 18g sample than a 15g one). I'd estimate that the number of fines is about the same in both samples for each grinder, but the number of coarse particles increases dramatically.
So, basically it looks like you get approximately the same number of fines in both a 15g and 18g dose, get slightly larger coarse particles in the larger dose, but get a lot more of them, probably about 3 grams worth.
Edit: I also just went back and looked at the 5 samples with large enough changes to be meaningful, and did some quick "holding a ruler up to the screen" math with the idea of comparing not the ratio of coarse to fine particles, but the total number in each sample. If we were graphing the total number of particles it would move the black lines up by about 20% (since there are 20% more particles in an 18g sample than a 15g one). I'd estimate that the number of fines is about the same in both samples for each grinder, but the number of coarse particles increases dramatically.
So, basically it looks like you get approximately the same number of fines in both a 15g and 18g dose, get slightly larger coarse particles in the larger dose, but get a lot more of them, probably about 3 grams worth.
It makes intuitive sense that the consistency of fines at 15g and 18g might be produced merely as a function of fracturing the particles, regardless of the grinder setting (to a point)... It does not make sense the coarse particles would be so consistent. Anyway, if we decide to repeat with different parameters, I am also happy to be on board and will order enough coffee next time to do the full hopper tests also.
Dana Leighton
LMWDP #269
LMWDP #269