mmc2 PDF
mmc2 PDF
mmc2 PDF
HIGHLIGHTS
Development of a model for
extraction of espresso
Experimental observation of
inhomogeneous extraction
Optimization of espresso
parameters to minimize coffee
waste
Article
Systematically Improving
Espresso: Insights from Mathematical
Modeling and Experiment
Michael I. Cameron,1,10 Dechen Morisco,1,10 Daniel Hofstetter,2 Erol Uman,3 Justin Wilkinson,4
Zachary C. Kennedy,5 Sean A. Fontenot,8 William T. Lee,6,7 Christopher H. Hendon,8,11,*
and Jamie M. Foster9,*
Of all of the coffee formats, espresso is by far the most complicated and susceptible
to fluctuations in beverage quality. As historically defined by the Specialty Coffee
Association, an espresso is a 25–35 mL (ca. 20–30 g) beverage prepared from
7–9 g of ground coffee made with water heated to 92 C–95 C, forced through the
granular bed under 9–10 bar of static water pressure and a total flow time of
20–30 s. These metrics have been grandfathered into the industry and are signifi-
cantly detached from the recipes used in most cafes today. Coffee shops routinely
favor higher dry coffee mass (15–22 g), resulting in larger volume beverages
(30–60 g beverage mass), produced on machines that dynamically control both
water pressure and temperature. The variables of tamp force, flow rate or time, dry
mass of coffee, and beverage volume are all determined by the machine’s operator.
There are other variables that have an impact on the beverage quality prior to the
ground coffee being exposed to water. The grind setting determines the particle
size distribution of the coffee grounds (and therefore the surface area).13 Once com-
pacted into a granular bed, the particle size distribution plays a role in controlling the
permeability of the bed and consequently the flow rate. A decreased flow rate can
be achieved in a number of ways: by decreasing the water pressure, grinding finer,
packing the bed more tightly, using more coffee, or some combination of these. A
further source of variability is that roasted coffee ages through off-gassing, losing
roast-generated volatiles thereby altering the resultant beverage density and
flavor.14,15
r
z
Outlet
P|z=L = 0
R0
Fines Boulders
The solubles in the liquid phase are transported by a combination of diffusion and
convection due to the flow of the liquid through the bed. The concentration of sol-
vated coffee is therefore governed by an advection-diffusion equation:
vcl
= V,ðDl Vcl ucl Þ in Ul ; (Equation 1)
vt
where t, Dl , and u are time, the diffusivity of solubles within the liquid, and the veloc-
ity of the liquid, respectively. The liquid flow is solved for via the Navier-Stokes
equations
vu 1 m
+ ðu , VÞu = VP + V2 u; (Equation 2)
vt r r
V,u=0 in Ul ; (Equation 3)
where m, r, and P are liquid viscosity, density, and overpressure, respectively.
Work by Spiro and colleagues demonstrated that the transport of coffee solubles
through the interior of the grounds can be described by a diffusion process.22–28
Hence,
vcs
= V,ðDs Vcs Þ in Us ; (Equation 4)
vt
where Ds is the diffusivity of solubles within the grains. Here, we treat coffee grounds as
spherical dense particles, but we note that the coffee grains themselves may be irregu-
larly shaped and feature intragranular macropores, as previously observed in scanning
electron micrographs.29 As discussed in the next section, most particles in ground cof-
fee are smaller the macropore diameter observed in the micrographs. Nitrogen physi-
sorption was used to assess the microporosity of the coffee grounds; the data suggest
that ground coffee does not feature microporosity (see Supplemental Information).
Thus, we expect our description to hold for most espresso grind settings.
u,bt jz = 0 = 0; (Equation 6)
Pjz = L = 0; (Equation 8)
u,bt jz = L = 0; (Equation 9)
ð Dl Vcl Þ,b
n jz = L = 0: (Equation 10)
At the vertical edges of the cylinder, R = R0 , no flux conditions are applied to the
liquid coffee concentration, because the liquid cannot exit in these directions:
On the boundaries between the grains and inter-granular pore space, Gint , there is a
flux of solubles per unit area, which we denote by G. Appropriate boundary condi-
tions are
ð Ds Vcs Þ , b
n = G; (Equation 13)
ð Dl Vcl + ucl Þ , b
n = G; (Equation 14)
Determining the form of the reaction rate, G, is non-trivial, and it is something that is
not readily measured experimentally. However, it can be reasonably assumed that
the rate of transfer of solubles between the phases should depend on the local con-
centrations of solubles near the interface. Furthermore, the rate of extraction is zero
when (1) the liquid immediately outside the grain is saturated (i.e., at a concentration
csat ) or (2) when the liquid outside the grain is at the same concentration as the grain
(i.e., in equilibrium) or (3) when the grain is depleted of solubles (the experimental
upper limit of extraction is approximately 30% by mass). We therefore postulate a
rate that satisfies all of the above conditions, namely
Coffee particulates remain dry until they are connected to the extraction apparatus,
at which point water is rapidly introduced to the bed, serving to wet the entire puck
and stabilize the particle temperature. Modeling this initial wetting (i.e., pre-infu-
sion) stage poses another series of interesting problems; the model presented
here is only valid once liquid infiltration has taken place, and we refer the interested
reader to a discussion on pre-infusion.30 We avoid explicitly modeling this stage by
assuming that at t = 0, when extraction begins, the bed is filled with liquid water that
is free from solubles. We therefore have
A
contribution
total
Relative
surface area
counts
B
fines boulders
20 GS
2.5
boulders large particles are
exit grinder further ground
Volume percent (%)
2.0
15
1.5
10 1.0
fines exit
grinder
5
0
1 10 100 1000
Particle diameter (µm)
Figure 2. Particle Size Distributions Collected Using the Method Described in the Experimental
Procedures
(A) Surface area and number of coffee particulates produced with a grind setting G S = 2.5. Here,
99% of the particles are <100 mm in diameter and account for 80% of the surface area.
(B) The volume percent particle size distribution at G S = 2.5, 2.0, 1.5, and 1.0. Grinding finer reduces
the average boulder size and increases the number of fines. Intruders are boulders that are larger
than the aperture of the burr set and hence further fractured until they can exit the burrs.
Particle size distributions were measured using our described experimental proced-
ure; these data are shown in Figure 2A. We observe that, similar to the formation of
two families of particle sizes found in exploding volcanic rock,31 there are two groups
of particle sizes in ground coffee. Namely, boulders (which we define as larger than
100 mm) and fines (smaller than 100 mm in diameter). This bimodal distribution is
caused by large particles fracturing until they are sufficiently small to exit through
the grinder burr aperture.13 The size of the boulders are determined by the burr sep-
aration, whereas the fines (much smaller than the burr aperture) are thought to be
produced at the fracture interface. One piece of evidence supporting this idea is
that as the grind setting, GS, is reduced, the relative proportion of fines increases,
but their size remains constant (Figure 2B).
systematically reduce the system using the asymptotic technique of multiple scales
homogenization. Such techniques have been applied to problems with a similar
structure in electrochemistry,32 and rather than present this very involved calculation
in full here, we provide a summary in the Supplemental Information and refer the
interested reader to Richardson and co-workers33 where the details of an analogous
calculation are presented.
The macroscopic system of equations, valid on the larger macroscopic length scale
of the entire bed, systematically follow from the microscopic Equations 1, 2, 3, 4, 5,
6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, and 17. The application of the multiple scales
technique significantly reduces the model complexity, but the requisite information
about the microscale variations is retained. For example, because the dissolution
rates depend on the concentration of solubles on the microscopic particle surfaces,
the multi-scale system contains a series of microscale transport problems that must
be solved inside representative grains. It is crucial that this microscopic information
is preserved in the multi-scale model, because it will allow us to study the effects of
different grind settings on the overall macroscopic behavior of the extraction.
Motivated by the bimodal distribution of particle sizes in the model, it may be assumed
that the bed is composed of two families of spherical particles with radii a1 (fines) and a2
(boulders). Further, we denote the Brunauer-Emmett-Teller (BET) surface area of the
different families of particles by bet1 and bet2 , respectively. The BET surface area char-
acterizes the amount of interfacial surface area between two intermingled phases per
unit volume of the mixture, and therefore has units of 1=m. We also introduce cs1 and
cs2 to denote the concentrations of solubles in the two particle families. The resulting
macroscopic equation for the concentration of solubles in the liquid is
vc v vc
ð1 fs Þ l = Deff l qc l + bet;1 G1 + bet;2 G2 : (Equation 18)
vt vz vz
Here, the quantity c l is the concentration of solubles in the liquid as predicted by the
multi-scale modeling approach; whereas cl appearing in the equations in the Sup-
plemental Information is the concentration of solubles in the liquid as predicted
by the original microscopic model. The upscaled and reduced versions of Equations
7 and 10 are
vc
Deff l + qc l = 0; (Equation 19)
vz z=0
vc l
Deff = 0: (Equation 20)
vz z = L
These assert that there should be no flux of solubles across the inlet and no diffusive
contribution to the flux at the outlet. In the next section, we show that parameter es-
timates indicate that diffusive fluxes are negligible compared with those due to
advection in typical espresso brewing conditions. Hence, it is the flow of the liquid
through the pores that is primarily responsible for moving solubles through the
bed once they have been dissolved. Hence, even though the physical relevance of
the latter condition in Equation 19 is not completely clear, it has negligible impact
on the model solution.
and the symmetry and dissolution rate boundary conditions, which arise from Equa-
tion 13, and which act to couple the micro- and macroscale transport problems, are
9
vcsi
Ds =0 > >
vr r = 0 =
for i = 1; 2: (Equation 22)
vcsi >
>
;
Ds = G
vr r = ai
i
A formula for EY in terms of the model variables can be derived by first noting that it
follows from Equations 18 and 20 that an expression for the rate at which soluble
mass enters the cup is given by
dMcup
= pR 20 qcl z = L : (Equation 23)
dt
On integrating this equation along with the initial condition that there is no solvated
mass at t = 0 and dividing by the dry mass of coffee initially placed in the basket, Min ,
we obtain
Rt
pR 20 q 0shot cl jz = L dt
Extraction yield ðEYÞ = ; (Equation 24)
Min
where EY is described as the fraction of solvated mass compared with the total mass
of available coffee. Here, tshot is the flow time. Equation 24 is used in the following
sections as a means to compare model predictions of EY with experimental
measurements.
Min
pR 20 L = f: (Equation 25)
rgrounds s
While it is likely that bed depth varies slightly across the range of grind settings (as
the volume fraction changes), we assume that the bed depth is constant for a given
dry mass of coffee, Min . Values for both the radii and BET surface area for the two
differently sizes families of particles in the grounds can be derived from the data
shown in Figure 3 provided that both families are distributed homogeneously
throughout the bed. The Darcy flux, q, determines the flow rate of the liquid through
the bed and varies with the grind setting. They are estimated using the shot times
presented in Figure 4 from the equation
Mout
q= ; (Equation 26)
pR 20 rout tshot
A 28
impact of dose on EY
27 6 bar / 40 g out
26
16 g
Extraction yield (mass %)
25
18 g
24
23 20 g
22 22 g
21 24 g
20
19
18
1.1 1.3 1.5 1.7 1.9 2.1 2.3
EK 43 grind setting, GS
B
26
impact of pressure on EY
25 20 g in / 40 g out
4 bar
24
Extraction yield (mass %)
6 bar
23
22 8 bar
21 10 bar
20
19
18
1.1 1.3 1.5 1.7 1.9 2.1 2.3
EK 43 grind setting, GS
Figure 3. Extraction Yield as a Function of Grind Size, with Varying Coffee Dose and Water
Pressure
(A) The effect of changing the coffee dose Min with constant water pressure shows that reducing the
initial coffee mass but keeping the beverage volume constant results in higher extractions.
(B) The effect of changing the pump overpressure, P, with a constant brew ratio shows an increase in
extraction yield with decrease in water pressure.
where Mout is the mass of the beverage (40 g), and we make the assumption that the
density of the drink, rout , is the same as that of water, but we note that this is an area
that could be improved in future model developments. We emphasize the difference
between Mout and Mcup ; the former is the total mass of the beverage, whereas Mcup
(used in Equation 23) is the total mass of solubles in the beverage. The parameter
values discussed above are tabulated in the tables presented in the Supplemental
Information.
A 45
R2 = 0.995
Shot time (tshot, s)
40
35
30
25
P = 6 bar
20
B
Regime 2: Partially clogged flow
23
Extraction yield (mass %)
22
21
Deff csat
qcsat : (Equation 27)
L
We note that this same conclusion was also reached previously.30 Henceforth, we
assign a small value to the macroscopic diffusivity of solubles in the liquid so that
diffusive transport is negligible compared with convection due to liquid flow. The
final two parameters, Ds and k, are fitted to the experiment. The results of this fitting
are shown in Figure 4 and lead to values of
2
Ds = 6:25 3 1010 m2 s; k = 6 3 107 m7 kg s1 : (Equation 28)
BREWING ESPRESSO
The model predicts that EY can be increased by grinding finer, using lower
pressure water, and/or using less coffee. The model was then compared with
experimental coffee brewing, performed in a cafe setting. Using an espresso ma-
chine set to P = 9 bar of water pressure resulted in clogging at fine grind settings
(see the Supplemental Information). To circumvent this difficulty, the pressure was
reduced to P = 6 bar. Coffee must be tamped to level the granular bed. We explored
a range of tamp pressures but did not observe an appreciable variation in shot time
or EY, and so we standardized our tamping procedure using an automated device
that pressed the bed at 98 N (see the Supplemental Information). Flavor differences,
however, were noted but not quantified. The barista needs to taste the coffee in the
cafe setting to ensure the beverage has the qualities that they desire. In summary,
however, lower water pressure and tamp force allowed for access to a wider range
of grind settings, thereby allowing for systematic sampling of shot time and EY
over all relevant espresso grind sizes (Figure 4). These results indicate that the rela-
tionship between shot time is linear (Figure 4A); with a coarser grind setting resulting
in shorter shot times.
25
experimental mean
±1 std. deviation
Standard flow
24
Extraction yield (mass %)
23
22
21
Our results provide an avenue to tackle three highly relevant issues in the coffee in-
dustry: first, how can one improve espresso reproducibility given the non-linear
dependence of EY on the grind setting? Second, what should one do to reduce
shot time or EY variability? And third, can we systematically improve espresso repro-
ducibility while minimizing coffee waste? We address these questions later.
of partial clogging in which some regions of the coffee bed have zero flow (and are
therefore non-extracted), while the remainder is subject to homogeneous flow.
Based on a comparison between the model prediction and the experimental mea-
surement, and assuming this extreme case of partial clogging (i.e., there are regions
in the coffee bed that are entirely dry), we find that there is a 13.1%, 6.1%, and 2.6%
difference in predicted EY versus experimental values for grind settings of 1.1, 1.3,
and 1.5, respectively. Of course, it is likely that the clogging results in a spread of
EYs, with a portion of the mass extracted to 0%, 1%, 2%, etc. Thus, estimating the
level of inefficiency in a given extraction is not possible using the refractive index
measurement alone, and we hope that follow-up studies will provide molecular han-
dles on the disparities in flavor comparing homogeneous and partially clogged ex-
tractions with identical EYs.
We are able to adapt the model to recover this downward trend after reaching the
critical grind setting. In essence, this is achieved by reducing the accessible surface
area of a fraction of the granular bed, simply modeled by reducing the surface area
of the planar faces of the cylindrical bed that is exposed to incident water. This pro-
cedure was performed empirically until our predicted EY matched the experimental
data. More important, however, are the implications of the need to reduce the pro-
portion of coffee that is accessible in the model. First, this suggests that in many cir-
cumstances where flow is inhomogeneous, there are regions of the granular bed that
have been extracted far higher than measured with an average EY. Second, the grind
setting plays a major role in determining how much dry coffee mass is wasted in the
brewing process. Finally, marriage of the model and the experiments provide us with
one clear avenue to optimize espresso extraction.
A
Improved espresso reproducibility Locate maximum extraction for a Decrease shot volume (i.e. ratio)
Increased beverage concentration fixed brew ratio and water pressure to obtain similar tasty point EY
Extraction yield
Extraction yield
Barista determined
tasty point
Extraction yield
15 g in / 40 g out 15 g in / 40 g out
Grind setting
20 g in / 40 g out
Tasty point Optimized tasty point 20 g in / 40 g out
B
Decreased beverage concentration Reduce dry coffee mass Arrive at similar tasty point EY
Reduction in coffee dry mass used in extraction by grinding much finer
The Specialty Coffee Association espresso parameters mandate that the extraction
should take 20–30 s; we speculate that this might be partially responsible for the pre-
vailing empirical truth that most coffee is brewed using grind settings that cause
partially clogged/inhomogeneous flow. Remembering that the initial tasty point
may lie in the clogged flow regime, some of the bed is extracted much more than
the refractive index measurement suggests. By lowering the dry coffee mass and
grinding to maximize EY, the operator may notice that they are able to push their ex-
tractions much higher than before, while achieving highly reproducible espresso.
Indeed, the two approaches presented in Figure 6 are complimentary, because
the former increases the shot concentration and the latter decreases the dry mass.
There are circumstances where businesses make decisions on the minimum concen-
tration and volume of espresso that is acceptable to present to customers. When
iterated, these approaches result in optimization of beverage volume, concentra-
tion, and other economic implications.
25
combined espresso shots to obtain
complexity and high reproducibility
24 expected EY 15 g in / 40 g out
Extraction yield (mass %)
23
partially clogged
flow tasty point
20 g in / 40 g out
22 low EY, high TDS
21
+ to obtain more consistently
Blending Shots
Beyond sensory science studies, a persistent difficulty is that there is no rapid route
to assessing the quality of two identical EYs made with different grind settings or
brew parameters. It is clear that espresso made at 22% EY in the partially clogged
regime tastes more ‘‘complex’’ than a fast 22% EY obtained using the optimization
routine presented in Figure 6. In an attempt to recover the same flavor profile as
the partially clogged flow regime, a shot must contain a mixture of higher and lower
extractions. Consider the tasty point in Figure 7: One can approximate its flavor pro-
file by blending two shots: (1) a low extraction/high dose (purple point) and (2) a high
extraction/low dose (green point). This procedure can more economically yield a
shot with a flavor profile that should approximate that which was previously only
obtainable in an economically inefficient partially clogged shot. Blending shots
does double the total volume of the beverage, and the procedure comes with the
added combinatorial complexity associated with calibrating two shots that, when
mixed together, yield superior flavor. We expect only the most enthusiastic practi-
tioners would consider this approach, but it may well be actionable in an industrial
setting where extraction is carried out in bulk.
increase of $3,620 per year. In addition to the monetary saving, the shot times were
routinely reduced to 14 s, significantly reducing the order-to-delivery time. From this
proof of concept, we can speculate on the larger economic benefit gained by the
procedure detailed herein. Encompassing both specialty- and commodity-grade
coffee beverage products, we estimate that the average coffee beverage is pro-
duced using ca. $0.10 of coffee. Considering a 25% reduction in coffee mass (i.e.,
$0.025 saving), and considering the daily coffee consumption in the United States
(124,000,000 espresso-based beverages per day),21 our protocol yields $3.1 million
savings per day, or $1.1 billion per year. Of course, this poses significant problems
for the entire supply chain, because being more efficient with ground coffee does
yield less revenue for roasters, importers, and producers and further highlights
that there is still much work to be done to improve efficiency in the industry, while
also uniformly increasing profits.
EXPERIMENTAL PROCEDURES
Espresso was prepared using standard equipment at Frisky Goat Espresso. Twenty
gram ridge-less baskets were fitted into the porta-filters of a San Remo Opera three
group espresso machine. The Opera allows for precise control of shot time, water
pressure (PW), and temperature. Coffee was ground on a Mahlkönig EK 43 grinder
fitted with coffee burrs. Espresso is typically ground at a grind setting (GS, a.u.) =
1.3–2.3, depending on the coffee. Tamp force (tF) was controlled using the Barista
Technology BV Puqpress, an automated tamper accurate to within G3 N. We elec-
ted to use an espresso profile specialty coffee, roasted by Supreme Roasters (Bris-
bane, Australia). The mass of coffee in the basket and the mass of the outgoing liquid
coffee were measured on an Acaia Lunar espresso balance.
Laser diffraction particle size analysis was performed on a Beckman Coulter LS13 320
MW. The instrument has a built-in dark-field reticule, which is used to ensure correct
optical alignment. An alignment check was carried out prior to every run to ensure
the optimum accuracy of the particle size distribution.
SUPPLEMENTAL INFORMATION
Supplemental Information can be found online at https://doi.org/10.1016/j.matt.
2019.12.019.
ACKNOWLEDGMENTS
This work was enabled by charitable donations of equipment from Barista Technol-
ogy BV (Puqpress), Acaia Corp. (balance), Frisky Goat Espresso (personnel and cof-
fee), and our continued collaboration with Meritics Ltd (particle size analyses). The
authors are grateful to M. and L. Colonna-Dashwood, A. Thomas Murray, and R.
Woodcock for insightful discussions. We thank Tailored Coffee Roasters, Eugene,
OR, for their implementation of the procedures detailed herein, and B. Sung and
M. Pierson for providing their revenue data. This work used the Extreme Science
and Engineering Discovery Environment (XSEDE), which is supported by National
Science Foundation grant number ACI-1053575.
AUTHOR CONTRIBUTIONS
The study was conceived by C.H.H., M.I.C., J.M.F, and W.T.L. M.I.C. and D.M.
brewed the espresso. D.H. and E.U. performed the grinding experiments. Z.C.K
and S.A.F. performed the gas sorption measurements. J.W., W.T.L, and J.M.F.
formulated the mathematical model. J.M.F carried out the model analysis and wrote
the code to solve the reduced model. C.H.H., W.T.L., and J.M.F. wrote the manu-
script, and all authors contributed to the final version.
DECLARATION OF INTERESTS
The authors declare no competing financial interests.
REFERENCES
1. Sunarharum, W.B., Williams, D.J., and Smyth, S., Miller, B., Butler, K.T., Melot, B.C., Speirs, 27. Spiro, M., and Chong, Y.Y. (1997). The kinetics
H.E. (2014). Complexity of coffee flavor: a R.W., and Hendon, C.H. (2016). The effect of and mechanism of caffeine infusion from
compositional and sensory perspective. Food bean origin and temperature on grinding coffee: the temperature variation of the
Res. Int. 62, 315–325. roasted coffee. Sci. Rep. 6, 24483. hindrance factor. J. Sci. Food Agric. 74,
416–420.
2. Spencer, M., Sage, E., Velez, M., and Guinard, 14. Glöss, A.N., Schönbächler, B., Rast, M.,
J.-X. (2016). Using single free sorting and Deuber, L., and Yeretzian, C. (2014). Freshness 28. Corrochano, B.R., Melrose, J.R., Bentley, A.C.,
multivariate exploratory methods to design a indices of roasted coffee: monitoring the loss Fryer, P.J., and Bakalis, S. (2015). A new
new coffee taster’s flavor wheel: design of of freshness for single serve capsules and methodology to estimate the steady-state
coffee taster’s flavor wheel. J. Food Sci. 81, roasted whole beans in different packaging. permeability of roast and ground coffee in
S2997–S3005. Chimia (Aarau) 68, 179–182. packed beds. J. Food Eng. 150, 106–116.
3. Schenker, S., Heinemann, C., Huber, M., 15. Ross, C.F., Pecka, K., and Weller, K. (2006). 29. Coffee beans. www.optics.rochester.edu/
Pompizzi, R., Perren, R., and Escher, R. (2002). Effect of storage conditions on the sensory workgroups/cml/opt307/spr16/.
Impact of roasting conditions on the formation quality of ground Arabica coffee. J. Food Qual.
of aroma compounds in coffee beans. J. Food 29, 596–606. 30. Moroney, K.M. (2016). Heat and mass transfer
Sci. 67, 60–66. in dispersed two-phase flows, PhD thesis
16. Farah, A. (2012). Coffee constituents. In Coffee, (University of Limerick).
4. Baggenstoss, J., Poisson, L., Kaegi, R., Perren, Y.-F. Chu, ed. (John Wiley & Sons), pp. 21–58.
R., and Escher, F. (2008). Coffee roasting and 31. Fowler, A.C., and Scheu, B. (2016). A theoretical
aroma formation: application of different time- 17. Folmer, B. (2016). The Craft and Science of
Coffee (Academic Press). explanation of grain size distributions in
temperature conditions. J. Agric. Food Chem. explosive rock fragmentation. Proc. R. Soc. A
56, 5836–5846. 472, 20150843.
18. Tan, C.-Y., and Huang, Y.-X. (2015).
5. Rao, S. (2014). The Roaster’s Companion (Scott Dependence of refractive index on
Rao). concentration and temperature in electrolyte 32. Foster, J.M., Chapman, S.J., Richardson, G.,
solution, polar solution, nonpolar solution, and and Protas, B. (2017). A mathematical model for
6. Rao, S. (2008). The Professional Barista’s protein solution. J. Chem. Eng. Data 60, 2827– mechanically-induced deterioration of the
Handbook (Scott Rao). 2833. binder in lithium-ion electrodes. SIAM J. Appl.
Math. 77, 2172–2198.
7. Rao, S. (2010). Everything but Espresso: 19. V. Fedele, Universal refractometer apparatus
Professional Coffee Brewing Techniques (Scott and method, (2012), US Patent 8239144, filed 33. Richardson, G., Denuault, G., and Please, C.P.
Rao). Mar 31, 2010, and published Aug 7, 2012. (2012). Multiscale modelling and analysis of
lithium-ion battery charge and discharge.
8. National Coffee Association of USA. 20. Kanouté, P., Boso, D.P., Chaboche, J.L., and J. Eng. Math. 72, 41–72.
Understanding the economic impact of the Schrefler, B.A. (2009). Multiscale methods for
United States coffee industry. http://www. composites: a review. Arch. Comput. Methods 34. Foster, J.M., Gully, A., Liu, H., Krachkovskiy, S.,
ncausa.org/Industry-Resources/Economic- Eng. 16, 31–75. Wu, Y., Schougaard, S.B., Jiang, M., Goward,
Impact/Economic-Impact-Infographic. G., Botton, G.A., and Protas, B. (2015).
21. E-Imports. Coffee statistics. http://www.e- Homogenization study of the effects of cycling
9. Gay, C., Estrada, F., Conde, C., Eakin, H., and importz.com/coffee-statistics.php. on the electronic conductivity of commercial
Villers, L. (2006). Potential impacts of climate lithium-ion battery cathodes. J. Phys. Chem. C
change on agriculture: a case of study of coffee 22. Spiro, M., and Selwood, R.M. (1984). The 119, 12199–12208.
production in Veracruz, Mexico. Clim. Change kinetics and mechanism of caffeine infusion
79, 259–288. from coffee: the effect of particle size. J. Sci. 35. von Bruggeman, D.A.G. (1935). Berechnung
Food Agric. 35, 915–924. verschiedener physikalischer konstanten von
10. Jaramillo, J., Muchugu, E., Vega, F.E., Davis, A.,
23. Spiro, M., and Page, C.M. (1984). The kinetics heterogenen substanzen. i.
Borgemeister, C., and Chabi-Olaye, A. (2011).
and mechanism of caffeine infusion from dielektrizitätskonstanten und leitfähigkeiten
Some like it hot: the influence and implications
coffee: hydrodynamic aspects. J. Sci. Food der mischkörper aus isotropen substanzen.
of climate change on coffee berry borer
Agric. 35, 925–930. Ann. Phys. 416, 636–664.
(Hypothenemus hampei) and coffee
production in East Africa. PLoS One 6, e24528.
24. Spiro, M., and Hunter, J.E. (1985). The kinetics 36. Robinson, J.O. (1970). The misuse of taste
11. Davis, A.P., Gole, T.W., Baena, S., and Moat, J. and mechanism of caffeine infusion from names by untrained observers. Br. J. Psychiatry
(2012). The impact of climate change on coffee: the effect of roasting. J. Sci. Food Agric. 61, 375–378.
indigenous Arabica coffee (Coffea arabica): 36, 871–876.
predicting future trends and identifying 37. Cruz, A., and Green, B.G. (2000). Thermal
priorities. PLoS One 7, e47981. 25. Spiro, M., Toumi, R., and Kandiah, M. (1989). stimulation of taste. Nature 403, 889–892.
The kinetics and mechanism of caffeine
12. Bunn, C., Läderach, P., Rivera, O.O., and infusion from coffee: the hindrance factor in 38. Clará, R.A., Marigliano, A.C.G., and Sólimo,
Kirschke, D. (2015). A bitter cup: climate intra-bean diffusion. J. Sci. Food Agric. 46, H.N. (2009). Density, viscosity, and refractive
change profile of global production of Arabica 349–356. index in the range 283.15 to 353.15 k and vapor
and Robusta coffee. Clim. Change 129, 89–101. pressure of a-pinene, d-limonene, G-linalool,
26. Spiro, M. (1993). Modelling the aqueous and citral over the pressure range 1.0 kPa
13. Uman, E., Colonna-Dashwood, M., Colonna- extraction of soluble substances from ground atmospheric pressure. J. Chem. Eng. Data 54,
Dashwood, L., Perger, M., Klatt, C., Leighton, roast coffee. J. Sci. Food Agric. 61, 371–372. 1087–1090.
Supplemental Information
Systematically Improving
Espresso: Insights from Mathematical
Modeling and Experiment
Michael I. Cameron, Dechen Morisco, Daniel Hofstetter, Erol Uman, Justin
Wilkinson, Zachary C. Kennedy, Sean A. Fontenot, William T. Lee, Christopher H.
Hendon, and Jamie M. Foster
Supplemental Information
Systematically improving espresso: insights from mathematical modeling and
experiment
Michael I. Cameron and Dechen Morisco
Frisky Goat Espresso, 171 George St., Brisbane City, QLD, 4000, AUS and
Current affiliation: ST. ALi Coffee Roasters, 12-18 Yarra Place, South Melbourne, VIC, 3205, AUS
Daniel Hofstetter
Daniel Hofstetter Performance, Laenggenstrasse 18, CH-8184 Bachenbuelach, CHE
Erol Uman
Meritics Ltd., 1 Kensworth Gate, Dunstable, LU6 3HS, UK
Justin Wilkinson
Faculty of Mathematics, University of Cambridge, Cambridge, CB3 0WA, UK
Zachary C. Kennedy
National Security Directorate, Pacific Northwest National Laboratory, Richland, WA, 99352, USA
Sean A. Fontenot
Department of Chemistry and Biochemistry, University of Oregon, Eugene, OR, 97403, USA.
William T. Lee
Department of Computer Science, University of Huddersfield, Huddersfield, HD1 3DH, UK and
MACSI, Department of Mathematics and Statistics, University of Limerick, Limerick, Ireland
Christopher H. Hendon
Materials Science Institute and Department of Chemistry and Biochemistry,
University of Oregon, Eugene, OR, 97403, USA
Jamie M. Foster
School of Mathematics & Physics, University of Portsmouth, Portsmouth, PO1 2UP, UK
(Dated: January 7, 2020)
I. SUPPLEMENTAL EXPERIMENTAL lutions into/out of the liquid arising from the boundary
PROCEDURE conditions (Equation 14) results in a reaction-advection-
diffusion equation. The calculation can be adapted
A. Multiple scales homogenization appropriately1 to show that
κ
In this section we carry out upscaling (or homogeniza- q = − ∇P, (1)
µ
tion) of the system of equations formulated in the Supple-
∂c∗
mental Information. Rather than resorting to a lengthy (1 − φs ) l
multiple scales analysis we use the previous results from ∂t (2)
the rigorous analysis1 to formulate the multi-scale sys- = ∇ · (Def f ∇c∗l − qc∗l ) + bet,1 G1 + bet,2 G2 ,
tem of equations. Before proceeding we note the abuse
of notation used in this section and henceforth; namely, where the star has been appended to cl to emphasize that
that the dependent variables here, and in the multi-scale it has undergone the homogenization procedure. Here, κ
homogenization (main text) are not strictly the same as is the permeability of the packed bed and q is the Darcy
those appearing in model development (main text) but flux (i.e., the discharge per unit area with units of m/s)
are instead their homogenized counterparts. Despite this, which is related to the average fluid velocity within the
in the interests of brevity we opt not to embed this dis- pore space, ν, via q = (1 − φs )ν. One might conjec-
tinction within the notation. ture that in espresso making applications pressures are
Upscaling the Navier-Stokes equations at suitably sufficiently high, and the pores sufficiently small, that
small Reynolds numbers (so that flow is laminar) on a turbulent flow could be present. However, the experi-
porous media results in Darcy’s law2,3 , whilst upscaling mental evidence strongly indicates that a model based
(Equation 1) and accounting for the source/sink of so- on Darcy’s law (rather than Ergun or Forchheimer4,5 )
2
TABLE S1. Parameter values above the horizontal score were taken from the literature, and those below were chosen to mimic
the experimental extraction protocol used here.
TABLE S2. Parameter values above the horizontal score were extracted directly from the experimental data shown in Figure
2 and those below were subsequently inferred using Supplemental Equation 3. Throughout, we take the radius of fines to be
fixed with a1 = 12µm.
perfect conservation of coffee mass, which can be hard in r and at these locations we denote the value of the cof-
to ensure using standard finite differences owing to the fee concentration in small and large particles using the
singularity at the origin of the radial coordinate in the following shorthands cs1 (z, r, t)|z=zj ,r=rk = cs1,j,k (t) and
spherical diffusion equation. Secondly, in contrast to cs2 (z, r, t)|z=zj ,r=rk = cs2,j,k (t), respectively. Thus, the
many other control volume methods, it provides direct index j indicates the representative particle’s position in
access to the concentration on the surface of the parti- z whereas k labels the radial position within a particu-
cles thereby avoiding the need for extrapolation which lar representative particle. The (2M + 1) × N unknown
would inevitably introduce additional errors. This sur- functions of time were converted into one large column
face concentration determines the reaction rate across the vector u(t) as follows
solid grain boundaries and so accurate evaluation of this
quantity is crucial for reliable simulation. After applying u(t) = [cl (t)∗T cs1,1 (t)T cs1,2 (t)T
these treatments for the spatial dependencies, the system (6)
of PDEs (Equation 17–Equation 22) are reduced to sys- · · · cs1,N (t)T cs2,1 (t)T cs2,2 (t)T · · · cs2,N (t)T ]T .
tem of coupled ODEs. We select the MATLAB routine
ode15s to integrate this system forward in time because:
We now rewrite the problem in the form
(i) it is able to cope with solving a system of differential-
algebraic equations; (ii) it offers adaptive time-stepping,
and; (iii) has relatively modest computational cost. du
M = f (u), (7)
We introduce N equally spaced grid points, zj for dt
j ∈ [1, N ], thereby dividing the spatial z-domain into
N − 1 equally spaced subdomains. The grid spacing in where M is the mass matrix and f (u) is a nonlinear func-
z is therefore given by hz = 1/(N − 1). Henceforth we tion which arises from the application of finite difference
adopt the shorthand notation c∗l (z, t)|z=zj = c∗l,j (t). At approximations (to the equations in z) and the control
each station in z we must solve for the coffee concentra- volume method (for the equations in r). The system of
tion within a representative grain, i.e. , at each zj we ODEs (Equation 7) is written in the standard form ac-
must solve two equations of the form shown in Equa- cepted by MATLABs ode15s.
tion 21; one with i = 1 for cs1 and another with i = 2 Below we present the first N entries of the nonlinear
for cs2 . Each of the 2N copies of Equation 21 are dis- function f (u) arising from the discretisation of Equation
cretized by introducing M equally spaced grid points, rk 18 and its boundary conditions (Equation 19). The re-
for k ∈ [1, M ], which subdivide each r-domain into M −1 maining 2M N equations arising from applying control
subdomains. The grid spacing in r is therefore given by volumes to Equation 20 are previously detailed10 , and so
hr = 1/(M −1). In total we have N ×M different stations in the interests of brevity we do not repeat them here.
4
TABLE S3. Parameter values above the horizontal score are experimental values shown in Figure 4 whilst the values of the
Darcy flux below the horizontal score were computed via Equation 26.
TABLE S4. Parameter values that were adjusted to explore the effects of altering the dose of grounds “in”.
We have
M1,1 = 0, (8)
Def f 3 1
f1 = − − u1 + 2u2 − u3 + u1 , (9)
hz 2 2
Mi,i = 1 − φs , (10)
Def f
fi = (ui+1 − 2ui + ui−1 )
h2z
(11)
ui+1 − ui−1
− + bet,1 K,
2hz
MN,N = 0, (12)
Def f 1 3
fN =− uN −2 − 2uN −1 + uN . (13)
hz 2 2
5
TABLE S5. Parameter values that were adjusted to explore the role of altering the pump overpressure. The values of the
Darcy flux for a 5 bar overpressure are identical to those on Table S3 and the others have been computed from those via the
relationship in Equation 26.
6
20
15
N2 adsorbed (cm3/g)
10
5 desorption
adsorption
0
0.0 0.2 0.4 0.6 0.8 1.0
Partial pressure (P/P0)
55
50
9 bar, G
S = 1.3
45
Shot time (s)
40
6 bar, G =
S 1.3
35
6 bar, GS = 1.5
30 9 bar, GS = 1.
5
25
100 200 300
Tamp force, tF (N)
1
G. Richardson, G. Denuault, and C. P. Please (2012), the steady-state permeability of roast and ground coffee in
“Multiscale modelling and analysis of lithium-ion battery packed beds,” J. Food Eng. 150, 106 – 116.
7
charge and discharge,” J. Eng. Math. 72, 41–72. J. Kestin, M. Sokolov, and W. A. Wakeham (1978), “Vis-
2
G. Allaire (1989), “Homogenization of the stokes flow in a cosity of liquid water in the range - 8 c to 150 c,” J. Phys.
connected porous medium,” Asymp. Anal. 2, 203–222. Chem. Ref. Data 7, 941–948.
3 8
D. Polisevsky (1986), “On the homogenization of fluid A. S. Franca, L. S. Oliveira, J. C. F. Mendonça, and X. A.
flows through porous media,” Rend. Sem. Mat. Univ. Po- Silva (2005), “Physical and chemical attributes of defective
litecn. Torino 44, 383–393. crude and roasted coffee beans,” Food Chem. 90, 89–94.
4 9
N. Dukhan, Ö. Bağcı, and M. Özdemir (2014), “Experi- K. M. Moroney, Heat and mass transfer in dispersed two-
mental flow in various porous media and reconciliation of phase flows, Ph.D. thesis, University of Limerick (2016).
10
forchheimer and ergun relations,” Exp. Therm. Fluid Sci. Y. Zeng, P. Albertus, R. Klein, N. Chaturvedi, A. Ko-
57, 425–433. jic, M. Z. Bazant, and J. Christensen (2013), “Efficient
5
A. C. Fowler, B. Scheu, W. T. Lee, and M. J. McGuinness conservative numerical schemes for 1d nonlinear spherical
(2010), “A theoretical model of the explosive fragmenta- diffusion equations with applications in battery modeling,”
tion of vesicular magma,” Proc. Royal Soc. Lon. A 466, J. Electrochem. Soc. 160, A1565–A1571.
11
731–752. J. M. Foster, “Espresso simulation code,” https://
6
B. R. Corrochano, J. R. Melrose, A. C. Bentley, P. J. Fryer, github.com/jamiemfoster/Espresso (2019), accessed 07
and S. Bakalis (2015), “A new methodology to estimate December 2019.