Titan Grinder Project: Particle size distributions of ground coffee
-
RapidCoffee
- Team HB
Here at Team H-B, we pride ourselves on our equipment reviews, and go to great (some might say bizarre) lengths to evaluate new espresso gear. A prime example is the Titan Grinder Project. Not only did we plan on pulling and tasting thousands of espresso shots, but we also considered techniques for quantitatively evaluating grinder performance. Two ideas that became reality were grind size and shape measurements.
Grind particle size is undoubtedly one of the most critical factors in a good grinder. The distribution of particle sizes and (to some extent) shapes may be determined by measuring the scattering of a beam of coherent light as it passes through the sample. This measurement is done in an instrument known as a laser diffraction particle size analyzer. Beam scattering measurements may be done wet (with the sample particles suspended in water) or dry (in air).
With the assistance of Russ Lingenfelter, a colleague at SDSM&T, I recently ran particle size analyses on three of the Titan grinders: the Macap MXK conical burr grinder, the Mazzer Robur conical burr grinder, and the Mazzer Super Jolly flat burr grinder.
To eliminate as many variables as possible, I prepared all the samples from one batch of roasted coffee in the same session. The coffee was a Kenya, homeroasted on an AeroRost electric drum roaster:
The samples were prepared by adjusting each grinder to yield a specific type of shot on one espresso machine. In this case, a ridgeless double basket was dosed with 19g coffee, and the grind adjusted to yield a 60ml shot in 30 seconds on a QuickMill Vetrano. After adjustment, three samples were ground on each grinder. Sample size ranges from 10g (1/8 cup) for wet analysis to 40g (1/2 cup) for dry analysis. Samples were placed in airtight containers (plastic ziplock baggies) for analysis later that day:
Yours truly in the lab:
Here's a full frontal view of the MicroTrac S3000 laser diffractometer:
You have to add the samples and flush the sample chamber manually, but the analyses are computer controlled:
On this day the machine was set up for wet analyses. For a wet analysis, you drop a couple of pinches (pardon the technical jargon) of grinds into a small chamber filled with water, and allow the grinds to circulate through the system for particle sizing. Somewhat surprisingly, the particle distribution shifts to slightly smaller sizes over time. My explanation: the ground coffee particles absorb water and swell when they hit the water. Circulation in water gradually breaks up clumps, resulting in smaller sized particles. This was a minor effect, but I decided to give each sample two minutes of circulation before running the analysis.
The MicroTrac software allows you to extract the number of particles, particle surface area, and particle volume distributions as a function of particle diameter in microns. Here are the results from my samples:
As you may recall, I ground three samples from each grinder. The results were very consistent, indicating a low degree of inter-shot variability. The Mazzers in particular showed virtually no differences between samples. I used the average of the three samples in the plots presented above.
Volume distribution appears to be the standard for most particle size analyses. These grinders all produce bimodal volume distributions. Presumably the larger particle size peak (around 400-500um) is indicative of the grinder setting, and the smaller particle size peak (around 40um) represents fines. If that's true, these results indicate that flat burr grinders tend to produce more uniform grinds with fewer fines than conical burr grinders. The flat burr Super Jolly peak is the tallest and narrowest, followed by the conical burr Robur. The conical burr MXK has lowest and widest peak, showing the greatest spread of grind sizes. The MXK also has the largest fines peak.
According to Jim Schulman, the volume peaks are quite close to the values quoted by Illy (although surface area and number of particles are not):
It's interesting to note that the volume curves correlate with my taste impressions. I've consistently placed the MXK at one end of the taste spectrum, the Super Jolly at the other, and the Robur in between. More on this later.
Two days later I was finally able to run a dry particle size analysis on the same samples. The particle size distribution results are as follows:
You'll notice that I added another flat burr grinder: a Bunn ES-1G (actually a rebadged Cunill Columbia, with the same 60mm flat burrs and 1/3hp motor as the Cunill Tranquilo). Why bother? I needed another 2 cups of grinds to calibrate the dry analysis setup, and thought it would be interesting to compare another flat burr grinder. Indeed, the Bunn curves follow the Super Jolly quite closely, even though the Bunn sample came from another batch of Kenya (roasted just the day before).
The wet and dry volume-based size distributions are reasonably consistent. The wet large peak and the dry small peak have shifted to the right (towards larger particle sizes), presumably due to swelling and clumping, respectively. There are some interesting changes in curve shapes as well. But these effects are relatively minor, and could be partially attributed to the two-day lag between wet and dry sample analyses. I will probably opt for wet analyses in the future, unless someone can make a convincing argument in favor of dry analyses.
Grind particle size is undoubtedly one of the most critical factors in a good grinder. The distribution of particle sizes and (to some extent) shapes may be determined by measuring the scattering of a beam of coherent light as it passes through the sample. This measurement is done in an instrument known as a laser diffraction particle size analyzer. Beam scattering measurements may be done wet (with the sample particles suspended in water) or dry (in air).
With the assistance of Russ Lingenfelter, a colleague at SDSM&T, I recently ran particle size analyses on three of the Titan grinders: the Macap MXK conical burr grinder, the Mazzer Robur conical burr grinder, and the Mazzer Super Jolly flat burr grinder.
To eliminate as many variables as possible, I prepared all the samples from one batch of roasted coffee in the same session. The coffee was a Kenya, homeroasted on an AeroRost electric drum roaster:
The samples were prepared by adjusting each grinder to yield a specific type of shot on one espresso machine. In this case, a ridgeless double basket was dosed with 19g coffee, and the grind adjusted to yield a 60ml shot in 30 seconds on a QuickMill Vetrano. After adjustment, three samples were ground on each grinder. Sample size ranges from 10g (1/8 cup) for wet analysis to 40g (1/2 cup) for dry analysis. Samples were placed in airtight containers (plastic ziplock baggies) for analysis later that day:
Yours truly in the lab:
Here's a full frontal view of the MicroTrac S3000 laser diffractometer:
You have to add the samples and flush the sample chamber manually, but the analyses are computer controlled:
On this day the machine was set up for wet analyses. For a wet analysis, you drop a couple of pinches (pardon the technical jargon) of grinds into a small chamber filled with water, and allow the grinds to circulate through the system for particle sizing. Somewhat surprisingly, the particle distribution shifts to slightly smaller sizes over time. My explanation: the ground coffee particles absorb water and swell when they hit the water. Circulation in water gradually breaks up clumps, resulting in smaller sized particles. This was a minor effect, but I decided to give each sample two minutes of circulation before running the analysis.
The MicroTrac software allows you to extract the number of particles, particle surface area, and particle volume distributions as a function of particle diameter in microns. Here are the results from my samples:
As you may recall, I ground three samples from each grinder. The results were very consistent, indicating a low degree of inter-shot variability. The Mazzers in particular showed virtually no differences between samples. I used the average of the three samples in the plots presented above.
Volume distribution appears to be the standard for most particle size analyses. These grinders all produce bimodal volume distributions. Presumably the larger particle size peak (around 400-500um) is indicative of the grinder setting, and the smaller particle size peak (around 40um) represents fines. If that's true, these results indicate that flat burr grinders tend to produce more uniform grinds with fewer fines than conical burr grinders. The flat burr Super Jolly peak is the tallest and narrowest, followed by the conical burr Robur. The conical burr MXK has lowest and widest peak, showing the greatest spread of grind sizes. The MXK also has the largest fines peak.
According to Jim Schulman, the volume peaks are quite close to the values quoted by Illy (although surface area and number of particles are not):
The volume curves also resemble Teme's particle size distribution plots.another_jim wrote:The Illy book gives three graphs. The number of particle one is unimodal and peaks at 0.2 microns. The surface area and volume ones are bimodal peaking at 1 and 20, and 40 and 500 respectively.
It's interesting to note that the volume curves correlate with my taste impressions. I've consistently placed the MXK at one end of the taste spectrum, the Super Jolly at the other, and the Robur in between. More on this later.
Two days later I was finally able to run a dry particle size analysis on the same samples. The particle size distribution results are as follows:
You'll notice that I added another flat burr grinder: a Bunn ES-1G (actually a rebadged Cunill Columbia, with the same 60mm flat burrs and 1/3hp motor as the Cunill Tranquilo). Why bother? I needed another 2 cups of grinds to calibrate the dry analysis setup, and thought it would be interesting to compare another flat burr grinder. Indeed, the Bunn curves follow the Super Jolly quite closely, even though the Bunn sample came from another batch of Kenya (roasted just the day before).
The wet and dry volume-based size distributions are reasonably consistent. The wet large peak and the dry small peak have shifted to the right (towards larger particle sizes), presumably due to swelling and clumping, respectively. There are some interesting changes in curve shapes as well. But these effects are relatively minor, and could be partially attributed to the two-day lag between wet and dry sample analyses. I will probably opt for wet analyses in the future, unless someone can make a convincing argument in favor of dry analyses.
John
-
another_jim
- Team HB
Great work.
Based on your, Cannonfodder's, and Greg Scace's observation that the small conicals taste brighter, the planars mellower, and the Robur in between (or just right for some many folks); and based on the way the grind samples distribute, I'm getting a germ of a hypothesis:
One may be able to equalize the tastes by changing dose (in the same basket). The greater number of fines and the higher volume peaks of the MXK to get the same flow indicates that it may be underextracting relative to the Robur, while the SJ is over-extracting relative to it. If this is true, one can compensate by lowering the dose and grinding finer on the MXK, and vice versa on the SJ. I'm fearlessly speculating that the dosages at which the volume peaks are the same would yield similar or identical taste results.
I can do my part in this by baking some pucks when I get the grinders. The hypothesis implies I'll find that the MXK is extracting less, and the SJ more (the mini in my case), than the Robur. This should show up in the weight loss of the pucks.
This assumes that the Kony is producing a size distribution similar to the MXK's.
Based on your, Cannonfodder's, and Greg Scace's observation that the small conicals taste brighter, the planars mellower, and the Robur in between (or just right for some many folks); and based on the way the grind samples distribute, I'm getting a germ of a hypothesis:
One may be able to equalize the tastes by changing dose (in the same basket). The greater number of fines and the higher volume peaks of the MXK to get the same flow indicates that it may be underextracting relative to the Robur, while the SJ is over-extracting relative to it. If this is true, one can compensate by lowering the dose and grinding finer on the MXK, and vice versa on the SJ. I'm fearlessly speculating that the dosages at which the volume peaks are the same would yield similar or identical taste results.
I can do my part in this by baking some pucks when I get the grinders. The hypothesis implies I'll find that the MXK is extracting less, and the SJ more (the mini in my case), than the Robur. This should show up in the weight loss of the pucks.
This assumes that the Kony is producing a size distribution similar to the MXK's.
Jim Schulman
-
Ken Fox
I'll second Jim's comment on GREAT WORK!
This is all very interesting and I'm sure it will create some lively discussions. I am now going to do my best to throw a monkey wrench into the gears of these evolving discussions.
All of this stuff is extremely complex, as complex as all the variables that go into espresso preparation. One of the peculiarities of North American high end espresso preparation is that we use a LOT of coffee in our baskets to make our shots. This is a topic that has been addressed by Jim Schulman and Andy Schecter recently in their work on extraction ratios and dosing. As a result of my conversations with these people, and with others recently (yesterday with Aaron De Lazzer in Vancouver, Canada) I'm coming around to the view that more is not necessarily better when it comes to the amount of coffee used to make an espresso. And, I acknowledge that in the past I have been a proponent of the idea of massive basket updosing. One of my first orders of business when I get back home from my current trip, is to do some experimentation with dosing; I doubt that 18 - 19g is going to remain my defacto standard. It will probably drop down to 16-ish and I'll play around with even smaller doses. This may result in a need to change my roast levels, to back off on them, even though right now I roast only to the very initiation of 2nd crack.
Beans are going to fracture differently in different grinders based on factors other than whether they are conicals or planars (or mixed, as in the case of my Cimbali Maxs). As far as taste of the grinds is concerned, I think it will NOT turn out that every commercial conical tastes like every other commercial conical, nor will the planars taste the same. Throw in some other variables such as roast levels, and the beans won't fracture the same and the grinds will differ from what they were when the coffee was roasted to a different level.
IF one had a preference for bean "A" roasted to 442F ground in grinder "X" and used in a 14g double, would that bean still be preferred in that grinder if roasted to 436F and used in a 16g double? I have my doubts.
I'll throw another monkey wrench in the gears; we are doing this sort of research in a bassackwards way. What would make more sense would be to try to establish which grinders produced the tastiest shots, then to try to find out why by studying such things as particle sizes and distributions. At this point there is this underlying assumption (in the TGP thread, not in this one) that conical grinders make better shots. As much as I would like to believe that this were true, the skeptic in me will not allow me to accept this without some sort of proof, more proof in fact than simply knowing that Andy Schecter has put his MM into the closet
ken
This is all very interesting and I'm sure it will create some lively discussions. I am now going to do my best to throw a monkey wrench into the gears of these evolving discussions.
All of this stuff is extremely complex, as complex as all the variables that go into espresso preparation. One of the peculiarities of North American high end espresso preparation is that we use a LOT of coffee in our baskets to make our shots. This is a topic that has been addressed by Jim Schulman and Andy Schecter recently in their work on extraction ratios and dosing. As a result of my conversations with these people, and with others recently (yesterday with Aaron De Lazzer in Vancouver, Canada) I'm coming around to the view that more is not necessarily better when it comes to the amount of coffee used to make an espresso. And, I acknowledge that in the past I have been a proponent of the idea of massive basket updosing. One of my first orders of business when I get back home from my current trip, is to do some experimentation with dosing; I doubt that 18 - 19g is going to remain my defacto standard. It will probably drop down to 16-ish and I'll play around with even smaller doses. This may result in a need to change my roast levels, to back off on them, even though right now I roast only to the very initiation of 2nd crack.
Beans are going to fracture differently in different grinders based on factors other than whether they are conicals or planars (or mixed, as in the case of my Cimbali Maxs). As far as taste of the grinds is concerned, I think it will NOT turn out that every commercial conical tastes like every other commercial conical, nor will the planars taste the same. Throw in some other variables such as roast levels, and the beans won't fracture the same and the grinds will differ from what they were when the coffee was roasted to a different level.
IF one had a preference for bean "A" roasted to 442F ground in grinder "X" and used in a 14g double, would that bean still be preferred in that grinder if roasted to 436F and used in a 16g double? I have my doubts.
I'll throw another monkey wrench in the gears; we are doing this sort of research in a bassackwards way. What would make more sense would be to try to establish which grinders produced the tastiest shots, then to try to find out why by studying such things as particle sizes and distributions. At this point there is this underlying assumption (in the TGP thread, not in this one) that conical grinders make better shots. As much as I would like to believe that this were true, the skeptic in me will not allow me to accept this without some sort of proof, more proof in fact than simply knowing that Andy Schecter has put his MM into the closet
ken
What, me worry?
Alfred E. Neuman, 1955
Alfred E. Neuman, 1955
-
RapidCoffee (original poster)
- Team HB
Agreed on all counts. Every grinder is unique, and we can expect different results based on type of bean(s), roast level, roast age, grind fineness, etc. Please consider these to be preliminary results, to be extended to a more comprehensive study as time and resources permit.Ken Fox wrote:Beans are going to fracture differently in different grinders based on factors other than whether they are conicals or planars (or mixed, as in the case of my Cimbali Maxs). As far as taste of the grinds is concerned, I think it will NOT turn out that every commercial conical tastes like every other commercial conical, nor will the planars taste the same. Throw in some other variables such as roast levels, and the beans won't fracture the same and the grinds will differ from what they were when the coffee was roasted to a different level.
The goal of these studies is to establish some relationship between a measurable parameter (such as particle size distribution) and taste in the cup. I don't think it really matters whether we measure particle sizes before or after cupping, as long as the cupping is double blinded. Unfortunately it's much more difficult to cup espresso, and coarse cupping grind size distributions may not equate to the finer grinds needed for espresso. I'm open to suggestions on how best to handle this.Ken Fox wrote:I'll throw another monkey wrench in the gears; we are doing this sort of research in a bassackwards way. What would make more sense would be to try to establish which grinders produced the tastiest shots, then to try to find out why by studying such things as particle sizes and distributions. At this point there is this underlying assumption (in the TGP thread, not in this one) that conical grinders make better shots.
BTW, several recent posts in the TGP thread show that any perceived bias towards conicals is rapidly fading away.
John
-
gscace
Hi there:
I'm having problems with the graphs. Can you tell me the correct units for the X axes and proper scaling? I'd think area ought to be in microns ^2 and volume ought to be in microns^3, and I wouldn't think the magnitudes should be similar for all three, would they be? For example, a square shape with side length of 100 microns oughtta be 10000 microns^2 in area. So there is something wrong with the graph scaling, although the relative distributions may be properly shown.
-Greg
I'm having problems with the graphs. Can you tell me the correct units for the X axes and proper scaling? I'd think area ought to be in microns ^2 and volume ought to be in microns^3, and I wouldn't think the magnitudes should be similar for all three, would they be? For example, a square shape with side length of 100 microns oughtta be 10000 microns^2 in area. So there is something wrong with the graph scaling, although the relative distributions may be properly shown.
-Greg
-
RapidCoffee (original poster)
- Team HB
I tried to keep everything as standard as possible in Excel, so the X units are particle "diameter" and the Y units are the percentage of the total particles. As I understand it (and please be aware that my understanding of the machine is limited), the raw data measurements are associated with diffraction of a laser beam that passes through the particles. Assumptions are entered for the refractive index of the liquid (for wet analyses) and the particle characteristics (spherical vs. irregular shape, transparent vs. absorbing, etc.). Then special formulas (aka "magic") are used to translate these light measurements into particle number, surface area, and volume.gscace wrote:I'm having problems with the graphs. Can you tell me the correct units for the X axes and proper scaling? I'd think area ought to be in microns ^2 and volume ought to be in microns^3, and I wouldn't think the magnitudes should be similar for all three, would they be? For example, a square shape with side length of 100 microns oughtta be 10000 microns^2 in area. So there is something wrong with the graph scaling, although the relative distributions may be properly shown.
For example, here's a plot generated by the MicroTrac software for a Robur particle volume analysis:
This plot shows the particle sizes as a frequency distribution or histogram (% channel) and a cumulative distribution (% passing). You can think of the latter in terms of sieves of different mesh size. The larger the mesh, the greater the percentage of particles that pass through. My plots show the frequency distribution, with the Y axis labeled as % total particles rather than % channel.
EDIT:
The MicroTrac manual indicates that the particle volume distribution is generated first, and surface area and particle number are derived from that. The X axis label represents the particle "diameter" in microns, in this case also derived from the volume distribution. I'm putting diameter in quotes because it's not a well-defined quantity unless the particles are assumed to be spherical.
Hope this helps.
John
-
another_jim
- Team HB
I'm not sure "tastiest" is necessarily the thing to go for here. More analytical terms like brightness, bitterness, sweetness, body, aroma, etc may be better.Ken Fox wrote:What would make more sense would be to try to establish which grinders produced the tastiest shots, then to try to find out why by studying such things as particle sizes and distributions.
That being said, we are all going to be doing some blind tasting of the grinders, and I'll be doing a great deal. I'll also be measuring how the grinders differ in extraction properties. I'm hoping we'll be able to correlate the tasting, extraction, grind size/shapes, and grinder specifications in a fairly straightforward way.
The idea is not just to crown the best grinder. There may be grinders that are worse under all circumstances and in every aspect, but my feel is that such models get taken off the professional market fairly quickly, although they can hang in forever in the consumer market. Instead, I think you are probably right that there's grinders that suit some styles of coffee, dosing or taste better than others. Our goal is to find out a lot more about this kind of thing.
We'd all like to give you a a clear program on all the things we are going to do; but it's too early for that. I have a feeling that the first people to test these grinders will come up with a bunch of issues and questions that have never been considered before, and the latter people will be trying to get some answers to these.
This is the reason we are doing the innards of the review in full public -- we don't want to pretend we're comparing these grinders on the basis of an authoritative checklist. Such a checklist does not exist. Instead, we're hoping these discussions will help show us what's important and what isn't.
Finally, I'd like to add my personal thanks to Jim Piccinich of 1st-Line for sponsoring and sitting still for this process -- it goes waaay beyond anything any company has done when sponsoring product reviews.
Jim Schulman
-
jesawdy
Ken-Ken Fox wrote:This is all very interesting and I'm sure it will create some lively discussions. I am now going to do my best to throw a monkey wrench into the gears of these evolving discussions.
...
I'll throw another monkey wrench in the gears; we are doing this sort of research in a bassackwards way. What would make more sense would be to try to establish which grinders produced the tastiest shots, then to try to find out why by studying such things as particle sizes and distributions. At this point there is this underlying assumption (in the TGP thread, not in this one) that conical grinders make better shots. As much as I would like to believe that this were true, the skeptic in me will not allow me to accept this without some sort of proof, more proof in fact than simply knowing that Andy Schecter has put his MM into the closet
You make many valid points... but this work (and the grinder review) has only just begun.
John had an opportunity to test running particle size analysis samples just this week, and it was mostly to see how well it worked, how much time it would take, whether we should run the wet and/or dry samples, and to help develop a sampling and testing protocol for the Titan Grinder Project (TGP). Indeed, we also wanted to see if the idea of measuring particle size and distribution was even worthwhile pursuing further.
Once John posted his (very) preliminary results to Team HB, Dan suggested that we make this information public now, so that we can get more "eyes" on the information, the wheels turning, and the monkey wrenches thrown! The readership of HB is varied enough that you never know who might have some very useful ideas to contribute to the study.
Jeff Sawdy
-
gscace
That clears things up a lot. Thanks!RapidCoffee wrote:I tried to keep everything as standard as possible in Excel, so the X units are particle "diameter" and the Y units are the percentage of the total particles. As I understand it (and please be aware that my understanding of the machine is limited), the raw data measurements are associated with diffraction of a laser beam that passes through the particles. Assumptions are entered for the refractive index of the liquid (for wet analyses) and the particle characteristics (spherical vs. irregular shape, transparent vs. absorbing, etc.). Then special formulas (aka "magic") are used to translate these light measurements into particle number, surface area, and volume.
For example, here's a plot generated by the MicroTrac software for a Robur particle volume analysis:
<image>
This plot shows the particle sizes as a frequency distribution or histogram (% channel) and a cumulative distribution (% passing). You can think of the latter in terms of sieves of different mesh size. The larger the mesh, the greater the percentage of particles that pass through. My plots show the frequency distribution, with the Y axis labeled as % total particles rather than % channel.
EDIT:
The MicroTrac manual indicates that the particle volume distribution is generated first, and surface area and particle number are derived from that. The X axis label represents the particle "diameter" in microns, in this case also derived from the volume distribution. I'm putting diameter in quotes because it's not a well-defined quantity unless the particles are assumed to be spherical.
Hope this helps.
-
rasqual
I'll resist the temptation to lavish praise on everyone working hard to provide very cool information from reams of data in these parts of late, and merely relate an anecdote.
When I was tinkering with polyester felts for filtering conventionally brewed coffee last year (or was it the year before? I'm getting old), I settled on 5 micron (nominal) as best eliminating fines while providing extraordinary throughput. Just here, in these graphs, I believe I'm seeing something to make me nod with familiarity.
Next step: I want to see someone slice and dice an individual bean into 500 micron cubes using lasers. A small kerf will help.
When I was tinkering with polyester felts for filtering conventionally brewed coffee last year (or was it the year before? I'm getting old), I settled on 5 micron (nominal) as best eliminating fines while providing extraordinary throughput. Just here, in these graphs, I believe I'm seeing something to make me nod with familiarity.
Next step: I want to see someone slice and dice an individual bean into 500 micron cubes using lasers. A small kerf will help.