calibrating a coffee grinder
when dialling in a coffee recipe, baristas will intuitively change the values of their input parameters to get their desired response values. those input parameters and response values are not standard across the community and there’s so many input parameters and subjectivity surrounding response values that we may never find a communal agreement on what those standards should be.
input parameters don’t necessarily differ between filter and espresso extractions at the fundamental level, although the parameters that can be controlled by the barista might take a slightly different form depending on the equipment. if we consider espresso extraction, wherein we don’t change the equipment’s hardware and the workflow (puck prep, etc.) is established, the input parameters that can be controlled are (i) the mass of coffee in the filter basket, (ii) the particle size of the coffee in the filter basket, (iii) the temperature of the water used for extraction, (iv) the flow rate of the water used for extraction and (v) the time of the extraction. Water flow rate might take the form of boiler pressure setting, or particle size might be grinder setting, for example, but on a fundamental level, as they relate to extraction dynamics, those are the input parameters that matter.
the response values that most baristas can consider include (i) the extraction yield (mass in grams), (ii) the extraction yield (% solids), (iii) the extraction efficiency (% — mass output relative to mass input), (iv) acidity of the brew, (v) the bitterness of the brew, (vi) the fruitiness of the brew and well, way more than i should list here.
each of these input parameters has a significant impact on extraction. one of the input parameters that has the greatest effect on extraction is the particle size — resulting from grinding the coffee. the action that baristas take is therefore to change the grinder setting and in a world with seemingly little access to particle size analysers, it’s good to know that baristas can use the grinder setting as a proxy for particle size.
this long intro was supposed to set the scene for a short post regarding calibrating a grinder and showing how powerful understanding grinder calibration data can be to the barista.
layout
i’m setting this one out like this:
• calibration methodology
• calibration of a DF64 with SSP multi-purpose burrs
• calibration of a Fellow ODE with mazzer super jolly burrs
• calibration of an EK43 with standard burrs
• calibration of an EK43 with turkish-style burrs
• calibration of an EG1 with base burrs
• calibration of a Niche with standard burrs
• calibration of a Timemore Chestnut with standard burrs
• conclusions & outlook
calibration methodology
coffee
for calibrations, i used a washed-processed pacamara from jose ovidio flores out of finca el trompillo, el salvador – imported by primavera and roasted by dialect to a mass loss of 12.5% in 04:12 mm:ss.
grinding
for the calibration, coffee was ground at several grinder settings that are equally distributed. three grinder settings can be enough for a narrow range of particle sizes — five settings is more than enough in most cases. so for example, 0, 3, 6 for espresso range or 0, 3, 6, 9, 12 for espresso and filter range. for grinders with a secondary input parameter, such as rotation speed on the EG1, the selected range consisted of a 3x3 matrix that yielded a response surface – this was to capture the non-linearity of the response — the experimental structure is outlined below.
here, 40 g of coffee were ground at different grinder settings with the order of settings randomised. 10 g of coffee were used to purge the grinder at the new setting before sampling.
particle size analysis
particle size analysis of 20 g samples were performed in duplicate using laser diffraction (HELOS/KF + RODOS/M + VIBRI/R + R7, Sympatec).
data analysis
data from the particle size analyser namely consists of a particle size distribution that looks something like this:
in the particle size distribution, there are two distinct peaks that correspond to two modes: one in the range of 0–100 μm and the other in the range of 100–1500 μm — this is therefore a bimodal distribution of polydisperse, granular media.
when changing grinder setting, the two peaks vary in distinct ways. the fine peak tends to move vertically in the y-axis, but not often horizontally in the x-axis. the coarse peak tends to move both vertically in the y-axis and horizontally in the x-axis. in this way, for a coffee with a finer particle size, the fine peak shifts up in the y-axis and the coarse peak shifts left in the x-axis and down in the y-axis.
the secondary data which is more practical for the barista are the correlations of values computed from the particle size distribution. the two most useful responses are the median particle size (x₅₀) and the percentage fines (%fines). the x₅₀ is the median value of the particle size distribution, where 50% of all particles are this size or smaller. the % fines is the percentage of particles whose size are smaller than 100 μm.
the median particle size tends to be linearly correlated (y=a+bx) with the grinder setting in the range of espresso and filter extractions. whilst the %fines tends to be related to the median particle size via a single-term power law (y=axᵇ). these are shown in the figure below.
in practice, as you change grinder setting, the median particle size will shift according to the gradient of the line in the plot of median particle size versus grinder setting:
median particle size = 205 + 13.4 * grinder settingso in this example, for this coffee, the gradient of these data explains that moving one integer in the grinder setting will yield a median particle size shift of 13.4 μm, whilst the intercept of these data explains that at point 0 (or at chirping-point if you’ve set the grinder up in this way), the grinder will yield particles with a median size of 205 μm.
as for the %fines versus median particle size plot, the shift in grinder setting is also predictable but you’ll probably need a computer/phone to do the calculation.
% fines = 23500 * median particle size ^ (-1.22)the value of b in the equation above (-1.22) will always be negative, as % fines increase as median particle size decreases and will often be in the range of -5.0 to -0.1, whilst a in the equation above (23500) will often be in the range of 500 to 100000. the values of these coefficients depend on both the grinder and the coffee and can vary significantly due to the power law relation.
the important learning from these plots is that, in most cases, as the grinder setting tends to zero (i.e., burr gap decreases until chirping-point), the increase in % fines becomes greater with each integer setting. in practice, this means a fine control over grinder setting in the espresso range is needed for fine tuning of brew time during extraction.
some examples
calibration of a DF64 with SSP multi-purpose burrs
in this example, a DF64 with SSP multi-purpose burrs was calibrated using the washed-processed pacamara at grinder settings 5, 15, 25, 35, 45, 55 & 65.
starting with the particle size distributions, the effect of grinder setting is as expected — as grinder setting increases and the burr gap increases, the fine peak decreases in share, whilst the coarse peak increases in share and shifts to a larger particle size.
as for the correlation of median particle size with grinder setting, the linear fit is sensible at lower grinder settings, although as median particle size approaches 1000 μm, particle size becomes more variable (i.e., less consistent) and the data begins to diverge from the linear expectation.
the correlation of % fines with median particle size shows that a power law model is suitable for these data.
the performance of the DF64 with SSP multi-purpose burrs is predictable based on the following two calculations.
median particle size = 205 + 13.4 * grinder setting% fines = 23500 * median particle size ^ (-1.22)as the intercept of the median particle size-grinder setting correlation is 205 μm and the corresponding share of fines in this region is > 20% the grinder is capable of espresso extractions.
as the share of fines corresponding to median particle sizes of 600–1000 μm are < 10%, the grinder that is also capable of high clarity filter extractions.
calibration of a Fellow ODE with mazzer super jolly burrs
in this example, a fellow ODE with mazzer super jolly burrs was calibrated using the washed-processed pacamara at grinder settings 1, 3, 5, 7 & 9.
there are no surprises with these particle size distributions, although the differences between grinder setting 7 and 9 are small.
median particle size versus grinder setting shows linearity for grinder settings 1–7, although differences between 7 and 9, corresponding to median particles sizes 900-1000 μm are not significant. in this way, the linear fit has been applied to settings 1–7 only.
correlation of % fines and median particle size yields a clear, power law trend, wherein all data are predictable using the following equations.
median particle size = 486 + 59.6 * grinder setting% fines = 30342 * median particle size ^ (-1.25)it’s worth noting that although the grinder is still not capable of espresso extractions with these burrs, the % fines for median particle sizes in the range of 500–800 μm is relatively large and will yield similar extraction times at similar median particle sizes as those from the DF64.
calibration of an EK43 with standard burrs
in this example, an EK43 with standard burrs was calibrated using the washed-processed pacamara at grinder settings 0, 3, 6, 9 & 12.
the particle size distributions are sensible and the grinder’s behaviour is as expected with regards to the shifting fine and coarse peak as grinder setting is varied.
whilst the correlation of median particle size with grinder setting is linear within the tested range, as the median particle size approaches 800 μm, the variability of particle size is apparent — starting at a lower median particle size than the DF64.
here, the correlation of % fines with median particle size was more difficult to fit to a power law model. the figure provides two power law fits that correspond to fitting with two sets of grinder setting: fit 1 — settings 0, 3, 6, 9 and fit 2 — settings 3, 6, 9, 12. this demonstrates that the power law model is appropriate within a certain and range and does not fully capture the real-system physics.
the grinder’s performance can be predicted using the following equations.
median particle size = 221 + 58.4 * grinder settingFit 1 (settings 0 to 9)
% fines = 6040 * median particle size ^ (-0.95)
Fit 2 (settings 3-12)
% fines = 37378 * median particle size ^ (-1.24)as the intercept of the median particle size-grinder setting correlation is 221 μm and the corresponding share of fines in this region is > 30% the grinder is capable of espresso extractions but this is expected to yield high extraction efficiency — this is likely why EK43 users tend toward turbo shots.
the share of fines corresponding to median particle sizes of 600–1000 μm are in the range of 8–15%, the grinder that is also capable of filter extractions but note that the extraction efficiency at similar median particle size as the DF64 with SSP multi-purpose burrs (whose share of fines in this range are much lower) is expected to be higher and therefore impact flavour accordingly (i.e., higher coffee intensity, more chocolatey, etc.).
calibration of an EK43 with turkish-style burrs
in this example, an EK43 with turkish-style burrs was calibrated using the washed-processed pacamara at grinder settings 0 (just above chirping-point), 3, 6, 9 & 12.
now, the particle size distributions for the turkish-style burrs are beyond expectation! as the grinder setting tends to zero and the burr gap closes in, the shift in coarse peak to a smaller particle size is so significant that it begins to merge with the fine peak. in this way, the fine peak is still < 100 μm but the share of fines is elevated significantly.
to understand what is happening at grinder settings approaching zero, the median particle size versus grinder setting correlation demonstrates that a linear fit is sensible for grinder settings 6–12 and the gradient of the fitted line is similar to that of the standard burrs. note that for grinder settings 0–3, corresponding to median particle sizes < 200 μm, these data diverge from the linear fit.
in a similar approach to the fitting of the share of fines versus median particle size taken with the standard burrs, the physics of grinding with turkish-style burrs are also not fully-captured by a power law model.
the grinder’s performance can be predicted using the following equations.
median particle size = -11 + 56.9 * grinder settingFit 1 (settings 0 to 9)
% fines = 624 * median particle size ^ (-0.55)
Fit 2 (settings 6 to 12)
% fines = 18440 * median particle size ^ (-1.12)a comparison of the EK43 burr-sets
interestingly, let’s compare the two types of burrs (installed in different grinders).
whilst the median particle size at a given grinder setting is different for the two burr-sets (higher for the standard burrs), they align on the share of fines versus median particle size. therefore, the EK43 with turkish-style burrs has the same grinding performance as that with the standard burrs in the median particle size range of 200–1000 μm but is also capable of grinding much finer.
a greater range, particularly on the finer side, doesn’t necessarily yield better performance for espresso extractions, though. the thing to realise here is that the turkish-style burrs can produce an 18 g in, 36 g out, 30 s shot (on a linea PB with a boiler pressure of 7 bar) at grinder setting 4.2 — this corresponds to a median particle size of approximately 220 μm with the turkish-style burrs and corresponds to setting 0 with the standard burrs. so, in the context of the current-wave café, turkish-style burrs might be more suited to less developed coffees, whilst the capability of the standard burrs might be on the tipping-point of capability, depending on the flow capabilities of the espresso machine and the physical properties of the coffee.
calibration of an EG1 with base burrs
in this example, an EG1 with base burrs was calibrated using the washed-processed pacamara. grinder settings (GS) ranged 1(just above chirping-point), 4 & 7, with rotation speeds (RS) that ranged 500, 1000 and 1500 rpm, such that the following combinations were used: GS(1)-RS(500), GS(1)-RS(1000), GS(1)-RS(1500), GS(4)-RS(500), GS(4)-RS(1000), GS(4)-RS(1500), GS(7)-RS(500), GS(7)-RS(1000) & GS(7)-RS(1500).
the impact of grinder setting
the impact of grinder setting at each rotation speed is not dissimilar to the other grinders shown here, with changes in the particle size distributions aligning with expectation.
the impact of rotation speed
here’s where it gets more interesting. as rotation speed increases, the relative share of the coarse peak does not significantly change, particularly at grinder settings 3 and 6. although, at grinder setting 0, just beyond chirping-point, the difference between 500 and 1000 rpm is more significant than the difference between 1000 and 1500 rpm. the correlations of median particle size and grinder setting, as well as the share of fines versus median particle size, also show that the differences between 500 and 1000 rpm are greater than those between 1000 and 1500 rpm.
in this way, lower rotation speeds yield lower median particle sizes and correspond to a greater % fines — attributed to longer residence times within the burrs. the correlation of % fines and median particle size outlines that in the range of 600–900 μm the effect of rotation speed is small but present, whereas in the range of 350–350 μm, the effect is much more significant, with 500 rpm yielding an increase in fines by more than 7% compared with 1500 rpm.
the grinder’s performance can be predicted using the following equations.
median particle size = 334 + 75.5 * grinder setting + 0.07 * rotation speed500 rpm
% fines = 37094 * median particle size ^ (-1.26)
1000 rpm
% fines = 38777 * median particle size ^ (-1.28)
1500 rpm
% fines = 17666 * median particle size ^ (-1.16)in practice, rotation speed modulates particle size in a significant way that can be useful for extractions if the baseline for your daily routine is 1000 rpm. therein, if you dial in a new coffee and yield a faster or slower extraction than the previous coffee, there was once only the option (with regards to grinding) to grind finer, whereas now, with this additional input parameter, the rotation speed could instead be decreased or increased to yield a similar response to the previous coffee.
calibration of a Niche with standard burrs
in this example, a Niche Zero with standard burrs was calibrated using the washed-processed pacamara at grinder settings 0, 12, 24, 36 & 48.
similar to the EK43 with turkish burrs, the particle size distributions for the Niche exhibit very high %fines! as the grinder setting tends to zero and the burr gap closes in, the shift in coarse peak to a smaller particle size is so significant that it begins to merge with the fine peak and for the Niche, whilst the fine peak is still < 100 μm, the share of fines is elevated so significantly that it’s much greater in share than the coarse peak!
the median particle size versus grinder setting correlation demonstrates that a linear fit is sensible for grinder settings 12–48, yet grinder settings 0–12, corresponding to median particle sizes < 100 μm, data diverge from the linear fit and a power law curve is more appropriate to cover the entire range of grinder settings (0–48). this is an example of how linear steps in burr gap yields a non-linear change in median particle size that seems to correspond with the limit of the grinder’s capabilities (i.e., the grinder cannot achieve particle sizes < 50 μm).
in a similar approach to the fitting of the share of fines versus median particle size taken with the EK43 with turkish-style burrs, the data is not accurately-captured by a power law model over the entire range of grinder settings. in this way, fit 1 corresponds with data over the entire range, whilst fit 2 corresponds with grinder settings (24–48) which are more appropriate for espresso extractions. for coffee properties suitable for filter extractions, grinding beyond the specified scale is necessary but to mitigate error in the calibration, this range was not examined here.
the grinder’s performance can be predicted using the following equations.
Fit 1 (settings 0 to 48)
median particle size = 39.4 + 1.6 * grinder setting ^ (1.46)
Fit 2(settings 12 to 48)
median particle size = -60.2 + 11.1 * grinder settingFit 1 (settings 0 to 48)
% fines = 535 * median particle size ^ (-0.52)
Fit 2 (settings 12 to 48)
% fines = 2580 * median particle size ^ (-0.79)calibration of a Timemore Chestnut with standard burrs
in this example, a Timemore Chestnut C2 with standard burrs was calibrated using the washed-processed pacamara at grinder settings 10, 14, 18, 22, & 26.
the particle size distributions are sensible and the grinder’s behaviour is as expected with regards to the shifting fine and coarse peak as grinder setting is varied.
the correlation of median particle size with grinder setting shows that the linear fit is sensible across all grinder settings.
the correlation of % fines with median particle size shows that a power law model is also suitable for these data.
the grinder’s performance can be predicted using the following equations.
median particle size = 3.7 + 38.5 * grinder setting% fines = 246000 * median particle size ^ (-1.55)conclusions & outlook
let’s be real for a moment…most baristas will not need to know the majority of what i’ve written here and only a few will ever have access to particle size analysers that can guide their approach to extraction, but, the general idea of grinder calibrations is extremely useful for baristas. that is: knowing that most grinders are designed in such a way that each notch of the grinder setting has a predictable effect on the median particle size and % fines. with this, baristas can then map the effect of grinder setting on extraction responses, such as time and extraction efficiency, safely knowing that these correlations will relate back to the fundamental properties of the polydisperse granular media (the coffee)… more on this soon.
acknowledgements
with thanks to Julian van de Sande at Bankai (Utrecht, NL) for access to the EK43 with turkish-style burrs and a space to test theories on grinding and extraction. thanks also to Alex Wallace & Mat North for pushing concepts, coffee and myself to get these data into the world.
data availability
for those that want to do something else with the data, have at it…find it here.
version history
v1.1 — updated to include Niche Zero and Timore Chenstnut C2