Accepted Manuscript

Time-Resolved Extraction of Caffeine and Trigonelline from Finely-Ground

Espresso Coffee with Varying Particle Sizes and Tamping Pressures

Michael Kuhn, Sandra Lang, Franziska Bezold, Mirjana Minceva, Heiko Briesen

PII: S0260-8774(17)30083-3

DOI: 10.1016/j.jfoodeng.2017.03.002

Reference: JFOE 8806

To appear in: Journal of Food Engineering

Received Date: 26 October 2016

Revised Date: 02 March 2017

Accepted Date: 03 March 2017

Please cite this article as: Michael Kuhn, Sandra Lang, Franziska Bezold, Mirjana Minceva, Heiko
Briesen, Time-Resolved Extraction of Caffeine and Trigonelline from Finely-Ground Espresso
Coffee with Varying Particle Sizes and Tamping Pressures, Journal of Food Engineering (2017),
doi: 10.1016/j.jfoodeng.2017.03.002

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form.
Please note that during the production process errors may be discovered which could affect the
content, and all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Highlights

 Extraction kinetics of caffeine and trigonelline from espresso coffee are studied.
 Particle size significantly affects the extraction kinetics.
 Tamping pressure has no detectable effect on extraction kinetics.
 Total extracted coffee mass influences the coffee composition.
 A mechanistic model with physically meaningful parameters is introduced.
 ACCEPTED MANUSCRIPT

1 Time-Resolved Extraction of Caffeine and Trigonelline from Finely-Ground Espresso Coffee with

2 Varying Particle Sizes and Tamping Pressures

4 Michael Kuhna, Sandra Langa, Franziska Bezoldb, Mirjana Mincevab, Heiko Briesena

6 a TUM School of Life Sciences Weihenstephan, Chair of Process Systems Engineering, Technical

7 University of Munich, Gregor-Mendel-Strasse 4, 85354 Freising, Germany

8 b TUM School of Life Sciences Weihenstephan, Chair of Biothermodynamics, Technical University of

9 Munich, Gregor-Mendel-Strasse 4, 85354 Freising, Germany

10

11 Corresponding author: Heiko Briesen, [email protected]

12

13 Abstract

14 We investigated the extraction kinetics of caffeine and trigonelline from espresso coffee prepared in

15 a commercial machine under realistic process conditions with varying particle sizes and tamping

16 pressures of the coffee powder. On the one hand, it was found that the particle size significantly

17 affects the extraction kinetics with smaller particles leading to a higher extracted amount of caffeine

18 and trigonelline per collected coffee mass. Tamping pressure, on the other hand, has no detectable

19 effect. Furthermore, the total extracted coffee mass was found to influence coffee composition as

20 measured by the trigonelline/caffeine ratio. All data were sampled in triplicates with a high time

21 resolution using a newly constructed sampling device. Finally, we introduced a new reduced model

22 that describes the measured data well and contains only physically meaningful parameters. This work

23 provides detailed data for better understanding the extraction of nonvolatile water-soluble

24 components from espresso coffee, thereby aiding model development and validation. The presented

25 simplified model may also prove useful for other related applications.

26

27 Keywords: caffeine, coffee, extraction kinetics, modeling, trigonelline

 ACCEPTED MANUSCRIPT

28

29

30 1. Introduction

31 Coffee is an important trading commodity (International Coffee Organization). Different beverages

32 prepared from roasted coffee beans are widely consumed all over the world, and coffee

33 consumption is still increasing (International Coffee Organization). Coffee produced from more than

34 9 million tons of raw beans was consumed in the year 2015. Among the different coffee brewing

35 styles, espresso is one of the most popular. It is a highly concentrated drink obtained using a high

36 water pressure a high water pressure of 8-11 bars applied during extraction and a short percolation

37 time of 15-30 s for about 15-30 ml of espresso (Illy and Viani, 2005; Petracco, 2008); however, it must

38 be mentioned that the values in the literature differ to some degree.

39

40 There has been much coffee research addressing the roasting process, sensory analysis, and

41 physiological aspects of coffee consumption (Viani and Petracco, 2005; Illy and Viani, 2005; Eggers

42 and Pietsch, 2008; Schilter et al., 2008). In contrast, this work focusses on coffee brewing alone, i.e.

43 on percolation and the thereby achieved extraction of coffee components. Although many baristas

44 have a lot of tacit knowledge about the effects of different extraction variables on the resulting

45 coffee taste, reproducible quantitative data is still sparse, especially when it comes to time-resolved

46 measurements. For the most part, the early studies investigating coffee extraction did not consider

47 realistic process conditions. Instead, results were presented for batch conditions; i.e., coffee

48 extraction was done in a stirred reactor. Under such conditions, the effect of particle size was

49 examined and a significantly faster caffeine extraction for smaller particles was found (Spiro and

50 Selwood, 1984), the extraction of caffeine under different degrees of roasting was studied (Spiro and

51 Hunter, 1985), and the effect of intra-bean diffusion on caffeine extraction was evaluated (Spiro et

52 al., 1989). Zanoni et al. (1992) observed different extraction phases while measuring concentrations

53 of soluble substances, and Jaganyi and Madlala (2000) investigated the extraction kinetics of mineral

54 ions and caffeine.

 ACCEPTED MANUSCRIPT

55

56 An early work studying the effects of realistic coffee brewing conditions was reported by Bell et al.

57 (1996). They found that both finer grains and greater amounts of coffee used lead to more

58 cumulative caffeine in the cup. Hinz et al. (1997) presented data on the amount of total extracted

59 solids over time for filter coffee and provided a simple mechanistic model for explaining this data.

60 In recent years, there have been renewed efforts to understand aspects directly related to the

61 brewing process. Andueza et al. (2002) evaluated the effect of different extraction pressures on the

62 quality of espresso coffee as reflected in physicochemical and sensory characteristics, whereas

63 Mateus et al. (2007) investigated the wetting dynamics of coffee particles. Albanese et al. (2009)

64 focused on the extraction temperature and found that extraction can be considered as an isothermal

65 process with a true extraction temperature lower than the water reservoir temperature. Gloess et al.

66 (2013) compared nine different extraction methods and found that the quality of coffee depends on

67 the extraction method. Booth et al. (2013) show some evidence on the varying extraction kinetics of

68 different coffee compounds; however, these data are not linked to the prevailing process conditions

69 such as flow rate or particle size. Caprioli et al. (2014) quantified the extraction of caffeine,

70 trigonelline, and nicotinic acid in espresso coffee under varying water temperature and pressure.

71 They also did a preliminary investigation of the extraction kinetics at a low time resolution and found

72 that after 25 s of extraction, further extraction merely dilutes the coffee.

73 Parenti et al. (2014) investigated the effect of different brewing techniques and found that capsule

74 systems provide the best product reproducibility. Corrochano et al. (2015) studied the steady-state

75 permeability of coffee beds and provided a corresponding modification of the Kozeny–Carman

76 equation. Moroney et al. (2015) presented and validated a multiscale model for the extraction

77 dynamics of filter coffee, but considered only the total solid content instead of single components.

78 The same model is further analyzed in Moroney et al. (2016) and limiting solutions are derived.

79 Moroney et al. (2016) also consider only the total solid content of coffee, even though the benefit of

80 addressing the extraction of different coffee components is mentioned in the outlook. Sánchez-López

81 et al. (2014) and Sánchez-López et al. (2016) conducted an online analysis of the extraction of volatile
 ACCEPTED MANUSCRIPT

82 organic compounds from espresso coffee by proton-transfer-reaction time-of-flight mass

83 spectrometry. The first study found different extraction kinectics for different coffee compounds; the

84 latter investigates the influence of temperature and pressure and found an increased extraction of

85 volatile organic compounds for higher values of both variables.

86

87 This work focusses on caffeine and trigonelline. Both have been analyzed in many previous works as

88 key components and important indicators of coffee quality. The extraction of caffeine has been

89 addressed in several reports (Spiro and Selwood, 1984; Spiro and Hunter, 1985; Spiro et al., 1989;

90 Zanoni et al., 1992; Bell et al., 1996; Jaganyi and Madlala 2000a; Albanese et al. 2009; Gloess et al.

91 2013; Caprioli et al. 2014; Parenti et al. 2014); trigonelline has been also investigated previously

92 (Farah et al., 2006; Caprioli et al., 2014; Parenti et al., 2014).

93 Trigonelline and caffeine are among the components with the highest mass fraction in coffee (Viani

94 and Petracco, 2000; Illy and Viani, 2005). Both are nonvolatile, water soluble, and bioactive (Buffo

95 and Cardelli-Freire, 2004; Caprioli et al., 2014). Typical values of caffeine content in dried green

96 Arabica and Robusta coffee beans are 1.2 and 2.4 wt%, respectively. Trigonelline is present at about

97 1.0 wt% in green Arabica and 0.7 wt% in green Robusta beans. Caffeine content is unaffected by

98 roasting whereas trigonelline decomposes to other substances and is thereby reduced by 30–80% in

99 mass depending on the degree of roasting. Typical values in a coffee cup extracted from 7.5 g of

100 roasted ground coffee at an extraction yield of 22% lie in the range of 50–150 mg caffeine and 30–

101 60 mg trigonelline (Caprioli et al. 2014).

102

103 As can be seen from the literature review, there is gap in current knowledge with respect to the

104 detailed extraction kinetics of nonvolatile espresso components under realistic process conditions.

105 We, therefore, study in this paper the extraction of caffeine and trigonelline from espresso coffee

106 brewed in a commercial machine. Our aim is to provide reproducible data sampled with a high time

107 resolution. Furthermore, a simplified model, containing only physically meaningful parameters, is
 ACCEPTED MANUSCRIPT

108 derived and applied. The model aids a mechanistic understanding of coffee extraction and paves the

109 path for more complex modeling approaches.

110

111 2. Materials and Methods

112 2.1 Chemicals

113 Caffeine (analytical standard) and trigonelline (analytical standard), both with a purity >99%, were

114 obtained from Sigma Aldrich (Taufkirchen, Germany). Methanol for HPLC analysis (HiPerSolv

115 CHROMANORM) was purchased from VWR Chemicals (Ismaning, Germany).

116

117 2.2 Milling and Sieving

118 Coffee (Espresso Nicaragua, 100% Arabica, variety Caturra) was obtained in 250 g packages from a

119 local roaster (Caffé Fausto, Munich, Germany). All beans were freshly roasted with the same

120 temperature profile, and the sealed packages were stored for no longer than four weeks before the

121 experiments. Packages once opened were not stored again to guarantee fresh ground coffee quality

122 for every set of experiments. A precision balance (FB6CCE-H, Sartorius AG, Göttingen, Germany) was

123 used to weigh 14 g of freshly milled beans for each experiment that was conducted; this is

124 appropriate as the typical amount of ground coffee beans for a double espresso lies in the range of

125 10 to 16 g (Illy and Viani, 2005). Milling was done with a professional espresso coffee mill (Combo

126 Coffee Grinder and Grater FMC6, Fama Industrie, Rimini, Italy) for which the reproducibility of the

127 milling results was assured by preliminary experiments. Particles sizes were chosen in order to

128 achieve manageable flow rates during coffee extraction under the process conditions described

129 subsequently. For this reason, a comparatively fine ground was used. For some experiments,

130 particles were sieved (vibrational sieving tower AS 200, Retsch, Haan, Germany) to achieve narrower

131 particle size distributions. Sieves with mesh sizes of 250, 280, 315, and 355 µm were used, as also

132 reflected in the values shown in Table 1. A sieving time of 25 min and sieving intervals of 10 s with an

133 amplitude of 1.2 mm were used. The ground coffee was sieved together with sieving aids (three

134 rubber balls with a diameter of 20 mm) to reduce agglomeration of coffee particles. Mass
 ACCEPTED MANUSCRIPT

135 conservation on all sieves was assured and a brush was used to clean sieves of adhering particles

136 which were also added to the corresponding sieve fractions. In preliminary experiments, the sieving

137 strategy was validated by repeated sieving runs and subsequent particle size analyses, assuring stable

138 results. In all cases,All particle sizes were measured by laser diffraction under dry-dispersed

139 conditions (HELOS, Sympatec GmbH, Clausthal-Zellerfeld, Germany). For one basic experimental

140 scenario, particle shapes were quantified by quasi-static image analysis (QICPIC, Sympatec GmbH,

141 Clausthal-Zellerfeld, Germany).

142

143 2.3 Brewing and Sampling

144 Coffee samples were prepared in a conventional espresso machine (Boxer, Rocket Espresso Milano,

145 Italy). The machine provides a constant water pressure of 11 bars at the fluid outlet, which was

146 measured by an empty portafilter equipped with a manometer. This value is in the range that is used

147 in practice and it was also used in other academic works (Andueza et al., 2002; Sánchez-López et al.,

148 2016).

149 A mass flow rate of 10.4 ± 0.4 g s−1 was achieved with the given machine and an empty portafilter,

150 i.e., when no coffee powder was added. The mass flow rate was reproducible and stayed constant

151 even if water masses of over 90 g were collected, i.e. significantly more water than used for a single

152 or double espresso. It is important to mention that the portafilter’s resistance, which can be

153 computed from the viscosity of water, pressure, and flow rate using Darcy’s law, cannot be simply

154 subtracted from the total resistance to obtain the resistance of the coffee bed (referred to as puck).

155 This is due to the increased resistance resulting from the interaction of coffee particles and the

156 portafilter, referred to as interference resistance (VDI, 2010). For this reason, we do not provide

157 permeability results of coffee pucks here.

158 Before coffee brewing, the machine was subjected to a preflushing of 5 s; i.e., hot water was

159 extracted without the mounted portafilter to heat up the machine. Additionally, the espresso

160 machine automatically provides a preinfusion phase of 3 s in which the coffee bed is wetted before

161 the actual coffee extraction takes place.

 ACCEPTED MANUSCRIPT

162

163 For all experiments, tamping is performed with a conventional 58-mm-diameter handheld espresso

164 tamper (CA Tamper TEB, concept-art GmbH, Nürnberg, Germany) using a tamping station on a table

165 scale (Soehnle Professional, Backnang, Germany) with which the tamping force was determined.

166 Weights of 1 and 30 kg (equivalent to a force of 9.81 and 294.3 N, respectively) were applied.

167

168 The temperature at the outlet of the coffee-filled portafilter was measured using a type-K

169 thermocouple and a handhold digital analyzer (GMH 3230, Greisinger electronic, Regenstauf,

170 Germany). The temperature was monitored for all tested cases in a set of preliminary experiments to

171 prevent disturbing the flow in the experiments where samples were taken for further analysis. The

172 temperature was 88 ± 2 °C in all cases.

173

174 Sampling at the portafilter outlet was done with a newly constructed, carrousel-like device shown in

175 Figure 1. Sixteen glass flasks were screwed directly to a rotating wheel driven by an electric motor

176 with a constant angular velocity. The inlets to the flasks were built in a funnel-like manner to

177 properly separate the coffee fractions without any loss in sample volume. The design allows a

178 complete detachment of the flasks for cleaning between the different experiments. In all

179 experimental cases, samples were taken over a time period of 26 s, typical for a double espresso (Illy

180 and Viani, 2005). Sampling was recorded for all experiments with a digital camera (EOS 1100 D,

181 Canon, Krefeld, Germany). In order to avoid artifacts due to the initial acceleration of the motor or

182 speed variations, the actual sampling times were determined from the videos. The masses in the

183 glass flasks were determined using an analytical balance (Entris 224I-1S, Sartorius AG, Göttingen,

184 Germany).

185

186 Figure 1: Sampling device (a) and procedure of sampling (b).

187

188 2.4 Density Measurements

 ACCEPTED MANUSCRIPT

189 The density of the coffee fractions was determined using an oscillating U-tube density meter (DM40,

190 Mettler Toledo, USA). Before density measurements, the coffee fractions were heated to 80 °C in a

191 water bath to avoid gas bubbles inside the tube. The chosen temperature is close to the brewing

192 temperature but does not lead to significant degassing, according to optical inspection, and

193 therefore artifacts are avoided.

194

195 2.5 pH Measurements

196 The collected coffee fractions were heated to 80 ± 5 °C in a water bath for pH measurements. The

197 temperature was chosen for the same reason as in the determination of density. Measurements

198 were conducted with a pH probe (S220, Mettler Toledo, USA). It was assured experimentally that the

199 reheating of the coffee samples did not influence the measured pH value.

200

201 2.6 HPLC Analysis

202 Caffeine and trigonelline contents were determined by HPLC. As a pretreatment for HPLC analysis,

203 the samples were diluted with filtered water (MilliQ, 18 MΩ, Merck KGaA, Germany). The dilution

204 ratio was adjusted for each sample. The first samples with a higher concentration of analytes and

205 impurities had to be stronger diluted than the latter fractions. Dilution ratios between 1:2 and 1:80

206 were used, where the ratios denote coffee volume to total volume. After dilution, samples were

207 centrifuged for 10 min at 5000 rpm in a laboratory centrifuge (Sigma 3-16KL, Sigma Laborzentrifugen

208 GmbH, Osterode am Harz, Germany). The supernatant was subsequently filtered with syringe filters

209 (PES, pore size 0.22 µm, Berrytec GmbH, Grünwald, Germany). Preliminary experiments were

210 conducted to confirm that there is no loss of caffeine or trigonelline due to adsorption on the filter

211 material.

212

213 An HPLC system with a liquid handling unit (Gilson, USA) and a UV-detector with two variable

214 detection wavelengths (Gilson, USA) were used with an RP-C18 column (VDSpher OptiAqua C18, 150

215 × 4.6 mm, 100 Å, particle size 5 µm). The flow rate was 1 mL min−1, the injection volume 10 µL.
 ACCEPTED MANUSCRIPT

216 Gradient elution was used to resolve trigonelline and caffeine from the coffee matrix. Gradient

217 conditions were set as follows: 5% methanol from 0 to 3 min, a linear gradient of 5% to 27%

218 methanol from 3 to 4 min, 27% methanol from 4 to 14.5 min, followed by reconditioning of the

219 column to 5% methanol. Calibration was performed in the range of 10 to 575 mg L−1 for caffeine and

220 5 to 350 mg L−1 for trigonelline, both in the linear concentration range. Caffeine was detected at a

221 wavelength of 272 nm, trigonelline at 264 nm. The software Trilution LC (Gilson, USA) was used for

222 HPLC data analysis.

223

224 2.7 Data Analysis

225 Data analysis was performed with MATLAB (version 2015a, MathWorks, Natick, Massachusetts, USA).

226 Linear equations were fitted using MATLAB’s fit function; in case of nonlinear models, the lsqcurvefit

227 function was used. Symbolic computations and Laplace transformations were performed using the

228 Symbolic Math Toolbox.

229

230 3. Results and Discussion

231 Four experimental scenarios were considered, with their labels listed in Table 1. The scenario Basic,

232 as the name indicates, is the basis for comparison with the others. It is also characterized in a more

233 detailed way than the other scenarios.

234

235 Table 1: Nomenclature and parameters for experimental scenarios.

236

237 All milled and sieved particles showed the characteristic bimodal size distribution also reported in

238 literature, e.g. in Corrochano et al. (2015) and Moroney et al. (2015). Figure 2 shows the measured

239 particle diameter distribution and Table 2 lists the median values of particle diameter for the

240 different scenarios. It can be seen that sieving does not lead to a sharp cut in the particle size

241 distribution but rather has a low separation efficiency; this is because coffee particles show strong

242 interparticle adhesion. As explained in the materials-and-methods section, comparatively small

 ACCEPTED MANUSCRIPT

243 coffee particles were used in order to achieve manageable flow rates at our extraction pressure of

244 11 bars.

245

246 Figure 2: Density distribution (a) and cumulative distribution (b) of coffee particle sizes as measured

247 by laser diffraction. The unsieved particle size distributions (•), as used in the scenarios Basic and

248 Strongly pressed, are compared with the distributions for Sieved small (▼) and Sieved large (▲);

249 different symbol fillings mark different tested samples.

250

251 Table 2: Median particle diameter for different scenarios.

252

253 Particle shape was analyzed only for the Basic scenario and was found to be dependent on particle

254 size, as shown in Figure 3. This result had a good reproducibility and is in agreement with the recent

255 study of Corrochano et al. (2015). The appearance of the size and shape distributions can be

256 attributed to the coffee’s cellular microstructure that leads to the observed characteristic fracture

257 behavior (Illy and Viani, 2005; Moroney et al., 2015).

258

259 Figure 3: Particle shapes for Basic scenario as measured by quasi-static image analysis; different

260 symbol fillings mark different tested samples.

261

262 Figure 4 depicts the cumulative-mass-versus-time curves for the different scenarios; the slope of the

263 curves corresponds to the mass flow rate of coffee, which is reported in Table 3. A good linearity can

264 be observed in all cases. The lack of nonlinearities at the beginning of the percolation, sometimes

265 reported in the literature (Illy and Viani, 2005), can be attributed to the automatic preinfusion phase

266 of the espresso machine that was described in the Materials and Methods section. It can also be seen

267 that the flow rate of water through the coffee bed depends on the particle size distribution. Tamping

268 pressure was found to have nearly no influence on the flow rate. It can be hypothesized that the

269 negligible effect of tamping is possibly also caused by the preinfusion phase, during which the coffee
 ACCEPTED MANUSCRIPT

270 bed expands and, therefore, the effect of the previously applied compression is reversed. The effects

271 of both particle size and tamping pressure on extraction will be discussed below.

272

273 Figure 4: Total collected coffee mass as a function of time, along with linear fits. Scenarios Basic (•),

274 Strongly pressed (♦), Sieved small (▼), and Sieved large (▲) are shown; different symbol fillings mark

275 different experimental runs.

276

277 Table 3: Mass flow rate of coffee for different scenarios as determined by linear fitting.

278

279 The extraction kinetics are the main focus of this work. First, all results are shown as a function of

280 total coffee mass instead of time to eliminate pseudo effects due to variations in the flow rate

281 between the different scenarios, as shown in Figure 4 and Table 3. The two independent variables

282 are linked by

283

𝑡
𝑚tot(𝑡) = ∫0𝑚(𝑡 ∗ ) 𝑑𝑡 ∗ , (1)

284

285 where 𝑚tot is total coffee mass, 𝑚(𝑡) is the coffee’s mass flow rate, and 𝑡 is time. It should be

286 mentioned that Eq. 1 is the theoretical relationship. In practice, 𝑚tot was obtained by directly adding

287 all the masses 𝑚𝑖 in the different glass flasks of the sampling wheel:

288

𝑁
𝑚tot,N = ∑ 𝑚𝑖 , (2)
𝑖=1

289
 ACCEPTED MANUSCRIPT

290 where 𝑁 denotes the flask under consideration and 𝑖 is the index of all flasks up to the one under

291 consideration. The results are then shown as 𝑐𝑁 = 𝑓(𝑚(𝑁)), where 𝑐𝑁 is the component

292 concentration in the current flask as analyzed by HPLC. Results of the temporal development are

293 shown later in the text when the fitting model is applied.

294

295 First, the Basic scenario is focused upon. It was analyzed in greatest detail and only the most relevant

296 measurements were performed for the remaining three scenarios to keep the experimental effort

297 reasonable. Figure 5 shows the extraction of both caffeine and trigonelline as a function of total

298 collected coffee mass for the Basic scenario.

299

300 Figure 5: Extraction of caffeine (black) and trigonelline (blue) as a function of total coffee mass;

301 different symbol fillings mark different experimental runs.

302

303 It can be seen that the two substances have different initial concentrations and follow different

304 extraction kinetics with trigonelline being extracted faster with collected coffee mass, i.e., with a

305 steeper initial slope. For coffee, the composition in the cup is the most important factor. Therefore,

306 the concentration results for the basic scenario were converted into cumulative component masses

307 𝑚c,tot using the relation

𝑡 𝑐c(𝑡 ∗ ) 𝑚
𝑚c,tot(𝑡) = ∫0 ∗
𝜌(𝑡 )
𝑑𝑡 ∗ , (3)

308

309 where 𝑐c is the components’ mass concentration, and 𝑚 and 𝜌 are the coffee’s mass flow rate and

310 density, respectively. Variable indices for the two different components, caffeine and trigonelline,

311 are not introduced for reasons of readability. The results are plotted as functions of total extracted

312 coffee volume 𝑉tot since this is the quantity most often found in the literature and also used in

313 practice. 𝑉tot was computed from

 ACCEPTED MANUSCRIPT

𝑡 𝑚
𝑉tot(𝑡) = ∫0𝜌(𝑡 ∗ ) 𝑑𝑡 ∗ . (4)

314

315 Both integrals were evaluated numerically using the trapezium rule from the discrete values given by

316 the measurements. The mass flow rate 𝑚 used in Eqs. (3) and (4) is the one determined by linear

317 fitting (Table 3); it is therefore considered constant here. The cumulative component masses can be

318 also easily compared with literature data; the numerical values are in the typical range (Viani and

319 Petracco, 2000). It is important to mention that we used the amount of coffee typical for a double

320 espresso whereas some literature data on cup concentrations refer to a single espresso. Note that

321 the locally different slopes of the cumulative component masses shown in Figure 6 indicate that the

322 coffee composition changes with the total coffee volume that is extracted, a point that will be further

323 discussed later on in this article.

324

325 Figure 6: Cumulative component masses for caffeine (black) and trigonelline (blue); different symbol

326 fillings mark different experimental runs.

327

328 For computing the data shown in Figure 6, the variation of coffee density with collected total coffee

329 mass needs to be known. The trajectory of coffee density for the Basic scenario is given in Figure 7.

330 The variations in density are small overall and approach a limiting value, which is the density of pure

331 water at the respective temperature. The small deviations in density suggest that using the mean

332 coffee density offers a good approximation for all practical purposes; therefore, the mean coffee

333 density is used in the model fitting discussed below.

334

335 Figure 7: Development of coffee density as a function of total coffee mass for the Basic scenario;

336 different symbol fillings mark different experimental runs.

337
 ACCEPTED MANUSCRIPT

338 Additionally, the pH change for the Basic scenario was measured (Figure 8). The beverage

339 approached the neutral regime starting from the acidic regime. These findings are in agreement with

340 the results of Severini et al. (2016).

341

342 Figure 8: Development of coffee’s pH value as a function of total coffee mass for the Basic scenario;

343 different symbol fillings mark different experimental runs.

344

345 Next, we compare the different scenarios and discuss the effects of particle size and tamping

346 pressure on extraction. In Figure 9, the corresponding results are shown as functions of total coffee

347 mass. The most obvious difference between the scenarios lies in the development of the collected

348 coffee mass and therefore in the extracted mass after 26 s, i.e., at the end of the brewing process.

349 This behavior reflects the different mass flow rates shown in Figure 4 and Table 3. Particle size has a

350 significant effect on the extraction kinetics, as seen in Figures 9 (a) and (b). Both caffeine and

351 trigonelline are extracted faster from smaller particles than from larger particles with respect to total

352 coffee mass, as indicated by the steeper initial slope. As in the basic scenario, trigonelline is extracted

353 faster than caffeine with respect to total coffee mass. Note that even though data were obtained

354 with a high resolution, the concentration curves do not show a plateau at the beginning of

355 extraction, as reported by Moroney et al. (2015). This could be caused by the preinfusion phase of

356 the espresso machine. Analogous to the cumulative mass depicted in Figure 4, differing tamping

357 pressure had no measurable effect on the extraction behavior, shown in detail in Figure 9 (d) where

358 the scenarios Basic and Strongly pressed are juxtaposed directly. This result corroborates the findings

359 of Severini et al. (2016), who also reported that tamping made no difference.

360

361 Figure 9: Influence of particle size distribution and tamping pressure on extraction kinetics of caffeine

362 (black) and trigonelline (blue); the scenarios Sieved small (a), Sieved large (b), and Strongly pressed

363 (c) are shown; in (d) the scenarios Basic (•) and Strongly pressed (♦) are juxtaposed directly; different

364 symbol fillings mark different experimental runs.

 ACCEPTED MANUSCRIPT

365

366 To further elucidate the effect of the extracted amount of coffee on the beverage composition, the

367 trigonelline/caffeine concentration ratio was evaluated. Figure 10 shows this ratio as a function of

368 collected total coffee mass. The varying lengths of the curves are caused by the difference in total

369 extracted masses at the end of the brewing process between the scenarios. Again, tamping had no

370 detectable effect on coffee composition (scenarios Basic and Strongly pressed). Note that the

371 scenario Sieved small differs in its qualitative behavior from the others because the

372 trigonelline/caffeine concentration ratio increases again after an extracted coffee mass of about 25 g

373 for two out of three experimental runs. The reason for this behavior is not yet clear. All results in

374 Figure 10 clearly show that the relative amount of both components, and thus the composition of

375 the espresso, significantly changes with the total coffee mass that is extracted.

376

377 Figure 10: Influence of particle size and tamping pressure on coffee composition as measured by the

378 trigonelline/caffeine concentration ratio; Scenarios Basic (•), Sieved small (▼), Sieved large (▲), and

379 Strongly pressed (♦) are shown; different symbol fillings mark different experimental runs.

380

381 The experimental data were further analyzed by fitting a simplified analytical model which is derived

382 in detail in the Appendix. According to our model, component concentrations in the liquid phase as a

383 function of time are given by the structural form:

384

𝑐𝑙(𝑡) = 𝐴1 ⋅ 𝑒𝑥𝑝 (𝐴2 ⋅ 𝑡) ⋅ (𝑐𝑜𝑠ℎ (𝐴3 ⋅ 𝑡) + 𝐴4) . (5)

385 The constants 𝐴1 to 𝐴4 are not only fitting parameters but reflect aggregated physical parameters as

386 given in Eqs. (6) to (9).

387
 ACCEPTED MANUSCRIPT

4 ⋅ 𝐵1
𝐴1 = (6)
𝐵2

388

𝐵3
𝐴2 = ‒ (7)
𝐵2

389

𝐵4
𝐴3 = (8)
𝐿 ⋅ 𝜀 ⋅ (𝜀 ‒ 1)

390

(1 ‒ 𝜀) ⋅ 𝑐𝑠,0 ⋅ 𝐾 ⋅ 𝐿 ⋅ 𝑘 ∗ + 𝑐𝑙,0 ⋅ 𝐾 ⋅ 𝐿 ⋅ 𝜀 ⋅ 𝑘 ∗
𝐿 ⋅ 𝜀 ⋅ sinh (𝐵4) ⋅ (1 ‒ 𝜀) ⋅
( 𝐵1
‒
)
𝐵3
𝐵2 (9)
𝐴4 =‒
𝐵4

391

392 𝐵1 to 𝐵4 are introduced to simplify the equations. They are defined in Eqs. (10) to (13).

393

(10)
𝐵1 =‒ 𝑐𝑙,0 ⋅ L ⋅ ε2 + 𝑐𝑙,0 ⋅ L ⋅ ε

394

(11)
𝐵2 =‒ 2 ⋅ L ⋅ ε2 + 2 ⋅ L ⋅ ε

395

(12)
𝐵3 = 2 ⋅ q + L ⋅ k ∗ ‒ 2 ⋅ ε ⋅ q ‒ L ⋅ ε ⋅ k ∗ + K ⋅ L ⋅ ε ⋅ k ∗

396

𝐾2 ⋅ 𝐿2 ⋅ 𝜀2 ⋅ 𝑘 ∗ 2 𝐾 ⋅ 𝐿2 ⋅ 𝜀2 ⋅ 𝑘 ∗ 2 𝐾 ⋅ 𝐿2 ⋅ 𝜀 ⋅ 𝑘 ∗ 2 𝐿2 ⋅ 𝜀2 ⋅ 𝑘 ∗ 2 𝐿2 ⋅ 𝜀 ⋅ 𝑘 ∗ 2 𝐿2 ⋅ 𝑘 ∗ 2
𝐵4 = ( ‒ + + ‒ +
4 2 2 4 2 4
(13)
1

+ 𝐾 ⋅ 𝐿 ⋅ 𝜀2 ⋅ 𝑘 ∗ ⋅ 𝑞 ‒ 𝐾 ⋅ 𝐿 ⋅ 𝜀 ⋅ 𝑘 ∗ ⋅ 𝑞 + 𝐿 ⋅ 𝜀2 ⋅ 𝑘 ∗ ⋅ 𝑞 ‒ 2 ⋅ 𝐿 ⋅ 𝜀 ⋅ 𝑘 ∗ ⋅ 𝑞 + 𝐿 ⋅ 𝑘 ∗ ⋅ 𝑞 + 𝜀2 ⋅ 𝑞2 ‒ 2 ⋅ 𝜀 ⋅ 𝑞2 + 𝑞2 )2
 ACCEPTED MANUSCRIPT

397

398 In the above equations, 𝑡 is the independent variable time, 𝑐𝑙(𝑡) is the time-dependent component

399 concentration in the liquid phase, 𝑐𝑙,0 is the initial concentration in the liquid phase, 𝑐𝑠,0 is the initial

400 concentration in the solid phase, ε is the porosity of the coffee bed, 𝑘 ∗ is the mass transfer

401 coefficient weighted by the specific surface of the coffee bed, 𝐾 is the distribution coefficient

402 between solid and liquid phase, 𝐿 is the coffee bed height, and 𝑞 is coffee’s superficial velocity. 𝑞 and

𝑚⋅4
403 𝐿 are clearly known from the experimental conditions. 𝑞 is computed from 2
𝜌 ⋅ 𝑑pf ⋅𝜋
, where 𝑚 is the

404 coffee mass flow rate (Table 3), 𝜌, mean coffee density, is assumed to be 980 g dm−3 (Figure 7), and

405 𝑑pf is the portafilter diameter (58 mm). The coffee bed height 𝐿 is set to 11 mm. 𝜀 is assumed to have

406 a constant value of 0.30, following Corrochano et al. (2015). 𝜀 can also be determined by fitting and

407 the results lie in a realistic range; however, the overall fitting results become more stable if the

408 porosity is provided as a constant. This leaves four constants that are determined by fitting: 𝑐𝑙,0, 𝑐𝑠,0,

409 𝐾, and 𝑘 ∗ . Note that all fitting constants as well as 𝑐𝑙(𝑡) refer to one component only, i.e. caffeine or

410 trigonelline. However, we have introduced no additional indices for improving readability.

411

412 As can be seen, all model equations contain only physically meaningful parameters and can be easily

413 fitted to experimental data due to their analytical nature. Even though the model contains many

414 simplifications, open to scrutiny in the Appendix, it still describes the experimental results quite well,

415 as seen in Figure 11. The parameter values determined by fitting are given in Tables 4 and 5. In

416 addition to the visual evaluation of the model fit, these parameters allow for further assessment.

417 Intuitively, one would expect that the initial concentration in the solid phase 𝑐𝑠,0 as well as the

418 distribution coefficient 𝐾 are identical in all scenarios because both are material constants, which is

419 approximately the case, given the strong assumptions in the model derivation. However, some

420 variations in 𝑐𝑠,0 and 𝐾 are imaginable, because both variables as used here are averaged quantities,

421 as explained in detail in the Appendix. Therefore, we do not claim universality of the presented

422 simple model. Instead we provide a qualified fitting equation, with parameters that can to some
 ACCEPTED MANUSCRIPT

423 extent be physically interpreted. The numerical values of the component concentrations 𝑐𝑠,0 for

424 caffeine and trigonelline in the solid phase are in the range reported for espresso coffee (Viani and

425 Petracco, 2000; Illy and Viani, 2005; Petracco, 2008). The parameter 𝑘 ∗ , product of the mass transfer

426 coefficient and the specific surface of the coffee bed, reflects the combined influence of particle size

427 and flow rate. It appears that the dominant factor affecting extraction is the flow rate and, therefore,

428 the varying residence time of fluid in the coffee bed. This is because larger particles result in larger

429 flow rates at a given constant pressure, according to the Kozeny–Carman model. Therefore, more

430 fluid passes through the coffee bed per unit time leading to a steeper decrease in the concentration

431 curves plotted over time. It is important to mention that the dynamical extraction behavior plotted

432 against time, shown in Figure 11, differs from the behavior plotted as a function of total coffee mass,

433 shown in Figure 9. This is because the influence of the varying flow rate between the different

434 scenarios is already accounted for when total coffee mass is used as the independent variable,

435 whereas the influence is only implicitly given when time is the independent variable. In practice, the

436 amount of extracted coffee mass or coffee volume is the determining factor. For this reason, the

437 main qualitative interpretations in this paper are given in terms of total coffee mass. However, the

438 extraction behavior as a function of time is shown because it is more easily accessible for the

439 presented model.

440

441 Figure 11: Extraction of caffeine (black) and trigonelline (blue) for the scenarios Basic (a), Sieved

442 small (b), Sieved large (c), and Strongly pressed (d); different symbols fillings mark different

443 experimental runs; solid lines represent the model fits.

444

445 Table 4: Model parameters for caffeine as determined by fitting.

446

447 Table 5: Model parameters for trigonelline as determined by fitting.

448
 ACCEPTED MANUSCRIPT

449 4. Conclusions

450 We presented high-time-resolution data of the extraction of caffeine and trigonelline, two key

451 espresso components, obtained under realistic process conditions. Three different particle size

452 distributions and two different tamping pressures were investigated. For each scenario, particle sizes

453 were characterized, and for one basic scenario, particle shapes and the time evolution of density and

454 pH were also monitored. All measurements were conducted in triplicates. It was shown that particle

455 size influences extraction kinetics, with smaller particles leading to faster extraction per collected

456 coffee mass. Additionally, particle size influences the coffee composition quantified by the

457 trigonelline/caffeine ratio. Tamping pressure had no measurable influence on the flow rate,

458 extraction dynamics, and coffee composition. In all scenarios, the total amount of extracted liquid

459 affects the coffee composition. Furthermore, an analytical fitting model was derived and its

460 applicability was shown. All parameters used in the model have a physical meaning, and their

461 identified values are in a realistic range. The fitting model is also expected to prove useful for similar

462 processes, such as decaffeination.

463 To our knowledge, this is the first report on highly-time-resolution resolved extraction data of water-

464 soluble key espresso components under realistic process conditions. Our data confirm quantitatively

465 and reproducibly what most coffee drinkers know from experience, namely that particle size and the

466 total amount of coffee extracted influence the coffee’s composition and therefore its taste. The

467 latter is reflected in the coffee classification based on extracted volume: Ristretto, Espresso, and

468 Lungo. Furthermore, the data are valuable for model development and validation of existing models.

469 In future work, data for more key coffee components will be measured and linked more closely with

470 mechanistic modeling approaches.

471

472 Acknowledgements

473 We thank Florian Schauer and Christoph Kaesbauer of Caffe Werkstatt in Freising, Germany, for

474 providing the espresso machine and coffee mill used in the experiments and their patience in

475 supporting this work. We also thank Caffé Fausto in Munich, Germany, for supplying the roasted
 ACCEPTED MANUSCRIPT

476 beans and its manager Klaus Wildmoser for his advice in selecting a reproducible roast. Further

477 thanks go to the workshop of the Chair for Process Systems Engineering of Technical University of

478 Munich for manufacturing the sampling device.

479

480 Appendix: Derivation of Fitting Model

481 In this appendix, an analytical fitting model is derived from a multiscale mass transport model using a

482 twofold averaging procedure. First, averaging is conducted over the particle diameter of the coffee

483 particles; second, the solution is averaged over the coffee bed height. The resulting model is then

484 only time dependent and can be solved analytically by Laplace transformation. A similar strategy is

485 successfully used e.g. by Espinoza-Perez et al. (2007) where decaffeination from whole beans is

486 addressed and averaging is conducted over bean volume. We are aware that our approach contains

487 strong simplifications and that some of the physics behind the process is eliminated by the two

488 averaging steps. However, the goal was to arrive at a somewhat mechanistic model comprising only

489 physically meaningful parameters, with the model still being easy to fit to experimental results. For a

490 more rigorous modeling approach, we refer to the recent works of Moroney et al. (2015) and

491 Moroney et al. (2016).

492

493 We start from a coupled model of diffusion in coffee particles and advective transport through the

494 porous particle bed. This model contains the following simplifications: coffee particles are assumed

495 to be spherical; the coffee bed is treated as homogeneous and incompressible; the flow is stationary

496 and initial wetting is not considered; dispersion is neglected.

497 Diffusion is modeled according to Fick’s second law:

498

∂2𝑐𝑠 2 ∂𝑐𝑠
∂𝑐𝑠
∂𝑡
1 ∂
𝑟 ∂𝑟
2
= 2 ( ‒ 𝑟 ⋅ 𝐽) = 𝐷 ⋅
( + ⋅
∂𝑟2 𝑟 ∂𝑟
,
) (A.1)

499
 ACCEPTED MANUSCRIPT

500 where 𝑐𝑠 is the component concentration in the solid phase, 𝐽 is the diffusive flux, and 𝐷 is the

501 diffusion coefficient, which here is an effective value because coffee particles are not one-phase

502 systems. 𝑟 and 𝑡 are the independent variables in the radial and temporal direction, respectively. It is

503 important to note that all concentration variables 𝑐 refer to one component only, caffeine or

504 trigonelline. We have refrained from introducing additional indices for improving readability.

505

506 The following initial and boundary conditions hold: 𝑐𝑠(𝑡 = 0) = 𝑐𝑠,0, 𝐽(𝑟 = 0) = 0, 𝐽(𝑟 = 𝑅) = ‒ 𝑘𝑙 ⋅ (

507 𝑐 ∗𝑙 ‒ 𝑐𝑙), where 𝑐 ∗𝑙 is the concentration at the solid–liquid-phase boundary and 𝑘𝑙 is the mass

508 transfer coefficient in the liquid phase. The minus sign at the outer boundary reflects the fact that

509 the flux leaves the coffee particles. Solid and liquid phase are coupled via the distribution coefficient:

510

(A.2)
𝑐 ∗𝑙 = 𝐾 ⋅ 𝑐 ∗𝑠 .

511

512 Also 𝐾, analogous to 𝐷, must be interpreted as an effective value due to the multicomponent

513 composition of coffee particles. Macroscopically, transport through the coffee bed is modeled by the

514 advection equation for porous media:

∂𝑐𝑙 ∂𝑐𝑙
𝜀⋅ =‒ 𝑞 ⋅ +𝑠, (A.3)
∂𝑡 ∂𝑧

515

516 where 𝜀 is the porosity, 𝑞 the fluid’s superficial velocity, 𝑠 a volumetric source term, and 𝑧 is the

517 spatial coordinate along the coffee bed height. 𝑠 is derived as follows:

518

1
𝑠=
𝑉
⋅ ∫𝐴 𝐽(𝑟 = 𝑅) 𝑑𝐴 = 𝑆𝑣 ⋅ 𝑘𝑙 ⋅ (𝑐 ∗𝑙 ‒ 𝑐𝑙) , (A.4)
int
 ACCEPTED MANUSCRIPT

519 where 𝐴int is the interfacial area between solid and liquid phases, and 𝑆𝑣 is the specific surface, i.e.,

520 interfacial area per total volume (Vafai, 2005). The first averaging step consists of an averaging over

521 the particle volume, leading to the particle-averaged concentration

1 𝑉 4𝜋 𝑅
〈𝑐〉𝑝 =
𝑉 ∫0 𝑐 𝑑𝑣 =
𝑉 ∫0 𝑟2𝑐 𝑑𝑟 , (A.5)

522

523 assuming symmetry along both angular directions in the spherical coordinate system. According to

524 Gauss’ divergence theorem, the flux out of a vector field passing through a closed surface is equal to

525 the volume integral of the divergence of the field inside the region. In our case, this allows the

526 substitution of the integral resulting from the averaging by the flux. Thus, applying Eq. (A.5) to Eq.

527 (A.1) and utilizing the divergence theorem leads to

528

∂〈𝑐𝑠〉𝑝
= ‒ 𝑆𝑣,𝑝 ⋅ 𝑘𝑙 ⋅ (𝑐 ∗𝑙 ‒ 𝑐𝑙) = ‒ 𝑆𝑣,𝑝 ⋅ 𝑘𝑙 ⋅ (𝐾 ⋅ 𝑐 ∗𝑠 ‒ 𝑐𝑙) , (A.6)
∂𝑡

529

530 where 𝑆𝑣,𝑝 is the particle specific surface, i.e., particle surface per particle volume. 𝑐 ∗𝑠 is subsequently

531 assumed to be equal to 〈𝑐𝑠〉𝑝. This is a strong assumption and implies that the coffee particles have a

532 homogenous component concentration over their radius. We are aware that more physically correct

533 modeling approaches are possible; however, this simple model was chosen to keep the number of

534 model parameters small. 𝑆𝑣,𝑝 is substituted using the following relationship (Bear, 1988):

𝑆𝑣
𝑆𝑣,𝑝 = , (A.7)
(1 ‒ 𝜀 )

535

536 leading to

537
 ACCEPTED MANUSCRIPT

∂〈𝑐𝑠〉𝑝 1
(A.8)
(1 ‒ 𝜀 ) 𝑣 𝑙 (
= ‒ ⋅ 𝑆 ⋅ 𝑘 ⋅ 𝐾 ⋅ 𝑐 ∗𝑠 ‒ 𝑐𝑙) .
∂𝑡

538

539 The second averaging step consists of an averaging over the coffee bed height 𝐿 with

1 𝐿
〈𝑐〉𝑏 =
𝐿 ∫0𝑐 𝑑𝑧, (A.9)

540

541 leading to

∂〈𝑐𝑙〉𝑏 𝑐𝑙(𝐿) ‒ 𝑐𝑙(0)

𝜀⋅ = 𝑞⋅ + 𝑆𝑣 ⋅ 𝑘𝑙 ⋅ (𝐾 ⋅ 〈〈𝑐𝑠〉𝑝〉𝑏 ‒ 〈𝑐𝑙〉𝑏). (A.10)
∂𝑡 𝐿

542

543 Note that the prefactor of the source term (1 ‒ 𝜀) cancels because now the source is formulated

544 volumetrically with respect to the solid phase and must therefore be weighted with (1 ‒ 𝜀) to be

545 consistent with the balance equation, which refers to the total bed volume. However, Eq. (A.10) still

546 contains two unknowns, 𝑐𝑙(𝐿) and 𝑐𝑙(0). To eliminate them, a linear profile is assumed as a closure

547 condition, leading to

548

𝑐𝑙(𝐿) ‒ 𝑐𝑙(0) = 2 〈𝑐𝑙〉. (A.11)

549

550 Thus, the final two equations are

∂〈𝑐𝑠〉𝑝 1
=‒ ⋅ 𝑆 ⋅ 𝑘 ⋅ (𝐾 ⋅ 〈𝑐𝑠〉𝑝 ‒ 𝑐𝑙) , (A.12)
∂𝑡 (1 ‒ 𝜀 ) 𝑣 𝑙

551
 ACCEPTED MANUSCRIPT

∂〈𝑐𝑙〉𝑏 2〈𝑐𝑙〉𝑏
𝜀⋅ = 𝑞⋅ + 𝑆𝑣 ⋅ 𝑘𝑙 ⋅ (𝐾 ⋅ 〈〈𝑐𝑠〉𝑝〉𝑏 ‒ 〈𝑐𝑙〉𝑏) , (A.13)
∂𝑡 𝐿

552 which are coupled ordinary differential equations dependent only on time. They can be solved

553 analytically by Laplace transformation. It must be noted that 𝑆𝑣 and 𝑘𝑙 are only present in both

554 equations as a product and are therefore substituted by (𝑆𝑣 ⋅ 𝑘𝑙) = 𝑘 ∗ . Furthermore, only the

555 concentration in the liquid phase at the outlet 𝑐𝑙(𝐿) is assessable experimentally; therefore, only this

556 analytical solution is shown in the main text. For this purpose, the closure condition of Eq. (A.11) is

557 applied:

(A.14)
𝑐𝑙(𝐿) = 2 〈𝑐𝑙〉 + 𝑐𝑙(0) = 2 〈𝑐𝑙〉 ,

558

559 leading to a multiplication by 2 of the solution of the Laplace transformation before it is used for

560 fitting. The results are shown as Eqs. (5)–(13) in the main text. Further, it can be shown that a purely

561 exponential model results if the advective part is neglected, i.e., if 𝑞 = 0. Such a modeling approach

562 was used by several researches who considered extraction under batch conditions (Spiro and

563 Selwood, 1984; Spiro et al., 1989; Zanoni et al., 1992; Jaganyi and Madlala, 2000) or from a fixed bed

564 (Severini et al., 2016), for which it is a good approximation as long as the superficial velocities are

565 small.

566

567

568 References

569 Albanese, D., Di Matteo, M., Poiana, M., Spagnamusso, S., 2009. Espresso coffee (EC) by POD: Study

570 of thermal profile during extraction process and influence of water temperature on chemical-physical

571 and sensorial properties. Food Res. Int. 42 (5-6), 727-732.

572
 ACCEPTED MANUSCRIPT

573 Andueza, S., Maeztu, L., Dean, B., de Peña, M.P., Bello, J., Cid, C., 2002. Influence of water pressure

574 on the final quality of arabica espresso coffee. Application of multivariate analysis. J. Agric. Food

575 Chem. 50 (25), 7426-7431.

576

577 Bear, J., 1988. Dynamics of fluids in porous media. Dover publications, New York, USA.

578

579 Bell, L. N., Wetzel, C. R., Grand, A. N., 1996. Caffeine content in coffee as influenced by grinding and

580 brewing techniques. Food Res. Int. 29 (8), 785-789.

581

582 Booth, C., Cummins, C., Dalwadi, M., Dellar, P., Devereux, M., Dewynne, J., Donohue, J., Duncan, A.,

583 Fitzmaurice, F., Gordon, A., Hennessy, M., Hinch, J., Hickey, C., Hjorth, P., Kyrke-Smith, T., Leahy, D.,

584 Lee, W., Lynch, E., Mercier,O., Miklavcic,S., Russell, S., Schwartz, L., Shozi, B., Swierczynski, P.,

585 Timoney, C., Tomczyk, J., Warneford, E., 2012. Brewing of filter coffee. In: Technical Report from

586 MACSI's 2012 Problem-Solving Workshop with Industry, Report MACSI/ESGI/0033. URL:

587 http://www.maths-in-industry.org/miis/605/1/Philips_Research_ESGI_87_Report.pdf, accessed:

588 21.10.2016.

589

590 Buffo, R. A., Cardelli-Freire, C., 2004. Coffee flavour: an overview. Flavour Fragrance J. 19 (2), 99-104.

591

592 Caprioli, G., Cortese, M., Maggi, F., Minnetti, C., Odello, L., Sagratini, G., Vittori, S., 2014.

593 Quantification of caffeine, trigonelline and nicotinic acid in espresso coffee: the influence of espresso

594 machines and coffee cultivars. Int. J. Food Sci. Nutr. 65 (4), 465-469.

595

596 Corrochano, B. R., Melrose, J. R., Bentley, A. C., Fryer, P. J., Bakalis, S., 2015. A new methodology to

597 estimate the steady-state permeability of roast and ground coffee in packed beds. J. Food Eng. 150,

598 106-116.

599
 ACCEPTED MANUSCRIPT

600 Eggers, R., Pietsch, A., 2008. Technology I: Roasting. In: Clarke, R.J., Vitzthum, O.G. (Eds.), Coffee –

601 Recent Developments, Blackwell Science, pp. 90-107.

602

603 Espinoza-Perez, J. D., Vargas, A., Robles-Olvera, V. J., Rodriguez-Jimenes, G. C., Garcia-Alvarado, M. A,

604 2007. Mathematical modeling of caffeine kinetic during solid-liquid extraction of coffee beans. J.

605 Food Eng. 81 (1), 72-78.

606

607 Farah, A., Monteiro, M., Calado, V., Franca, A., Trugo, L., 2006. Correlation between cup quality and

608 chemical attributes of Brazilian coffee. Food Chem. 98 (2), 373-380.

609

610 Gloess, A. N., Schönbächler, B., Klopprogge, B., D`Ambrosio, L., Chatelain, K., Bongartz, A.,

611 Strittmatter, A., Rast, M., Yeretzian, C., 2013. Comparison of nine common coffee extraction

612 methods: instrumental and sensory analysis. Eur. Food Res. Technol. 236 (4), 607-627.

613

614 Hinz, T., Steer, A., Waldmann, C., Cammenga, H. K., Eggers, R., 1997. Röstkaffee-Extraktion:

615 Einflußparameter und Modellierung. Chem. Ing. Tech. 69 (5), 685-690.

616

617 Illy, A., Viani, R. (Eds.), 2005. Espresso Coffee – The Science of Quality, Elsevier Academic Press.

618

619 International Coffee Organization, URL: http://www.ico.org/monthly_coffee_trade_stats.asp, last

620 accessed: 21.10.2016.

621

622 Jaganyi, D., Madlala, S.P., 2000. Kinetics of coffee infusion: a comparative study on the extraction

623 kinetics of mineral ions and caffeine from several types of medium roasted coffees. J. Sci. Food Agric.

624 80 (1), 85-90.

625
 ACCEPTED MANUSCRIPT

626 Mateus, M.-L., Rouvet, M., Gumy, J.-C., Liardon, R., 2007. Interactions of water with roasted and

627 ground coffee in the wetting process investigated by a combination of physical determinations. J.

628 Agric. Food Chem. 55 (8), 2979-2984.

629

630 Moroney, K.M., Lee, W.T., O’Brien, S.G.B., Suijver, F., Marra, J., 2015. Modelling of coffee extraction

631 during brewing using multiscale methods: An experimentally validated model. Chem. Eng. Sci. 137

632 (0), 216-234.

633

634 Moroney, K. M., Lee, W. T., O'Brien, S. B., Suijver, F., Marra, J., 2016. Coffee extraction kinetics in a

635 well mixed system. J. Math. Ind. 7 (1), 1-19.

636

637 Parenti, A., Guerrini, L., Masella, P., Spinelli, S., Calamai, L., Spugnoli, P., 2014. Comparison of

638 espresso coffee brewing techniques. J. Food Eng. 121, 112-117.

639

640 Petracco, M., 2008. Technology IV: Beverage Preparation: Brewing Trends for the New Millennium.

641 In: Clarke, R.J., Vitzthum, O.G. (Eds.), Coffee – Recent Developments, Blackwell Science, pp. 140-164.

642

643 Sánchez-López, J. A., Wellinger, M., Gloess, A. N., Zimmermann, R., Yeretzian, C., 2016. Extraction

644 kinetics of coffee aroma compounds using a semi-automatic machine: on-line analysis by ptr-tof-ms.

645 Int. J. Mass Spectrom. 401, 22-30.

646

647 Sánchez-López, J. A., Zimmermann, R., Yeretzian, C., 2014. Insight into the time-resolved extraction

648 of aroma compounds during espresso coffee preparation: online monitoring by ptr-tof-ms. Anal.

649 Chem. 86 (23), 11696-11704.

650

651 Schilter, B., Cavin, C, Tritscher, A., Constable, A., 2008. Health Effects and Safety Considerations, O.G.

652 (Eds.), Coffee – Recent Developments, Blackwell Science, pp. 165-183.

 ACCEPTED MANUSCRIPT

653

654 Severini, C., Derossi, A., Fiore, A. G., De Pilli, T., Alessandrino, O., Del Mastro, A., 2016. How the

655 variance of some extraction variables may affect the quality of espresso coffees served in coffee

656 shops. J. Sci. Food Agric. 96 (9), 3023-3031.

657

658 Spiro, M., Selwood, R. M., 1984. The kinetics and mechanism of caffeine infusion from coffee: the

659 effect of particle size. J. Sci. Food Agric. 35, 915-924.

660

661 Spiro, M., Hunter, J. E., 1985. The kinetics and mechanism of caffeine infusion from coffee: the effect

662 of roasting. J. Sci. Food Agric. 36 (9), 871-876.

663

664 Spiro, M., Toumi, R., Kandiah, M., 1989. The kinetics and mechanism of caffeine infusion from coffee:

665 the hindrance factor in intra-bean diffusion. J. Sci. Food Agric. 46 (3), 349-356.

666

667 Vafai, K. (Ed.), 2005. Handbook of porous media. CRC Press, Boca Raton, Florida, USA.

668

669 VDI, 2010. VDI 2702: Mechanical solid-liquid-separation by cake filtration, part 2: determination of

670 filter cake resistance, Beuth Verlag, Berlin, Germany.

671

672 Viani, R. and Petracco, M., 2000. Coffee. In: Ullmann's Encyclopedia of Industrial Chemistry, Vol. 9,

673 Wiley-VCH, pp. 467-498.

674

675 Zanoni, B., Pagliarini, E., Peri, C., 1992. Modelling the aqueous extraction of soluble substances from

676 ground roasted coffee. J. Sci. Food Agric. 58 (2), 275-279.

 ACCEPTED MANUSCRIPT

Table 1: Nomenclature and parameters for experimental scenarios.

Label Tamping force (N) Sieving range (µm)

Basic 9.81 None
Sieved small 9.81 250–280
Sieved large 9.81 315–355
Strongly pressed 294.3 None
 ACCEPTED MANUSCRIPT

Table 2: Median particle diameter for different scenarios.

Scenario Median (µm)

Basic/Strongly pressed 238.29
Sieved small 107.62
Sieved large 282.78
 ACCEPTED MANUSCRIPT

Table 3: Mass flow rate of coffee for different scenarios as determined by linear fitting.

Scenario Mass flow rate (g s−1) R² value (-)

Basic 4.02 0.98
Sieved small 1.66 0.83
Sieved large 9.18 0.98
Strongly pressed 3.65 0.91
 ACCEPTED MANUSCRIPT

Table 4: Model parameters for caffeine as determined by fitting.

𝒄𝒍,𝟎 (g L−1) 𝒄𝒔,𝟎 (g L−1) 𝑲 (-) 𝒌 ∗ (s−1)

Basic 6.62 7.26 1.24 × 103 9.27 × 10−5
Sieved small 9.56 6.58 1.61 × 103 4.63 × 10−5
Sieved large 5.08 6.61 1.73 × 103 1.30 × 10−4
Strongly pressed 6.15 7.44 1.25 × 103 8.55 × 10−5
 ACCEPTED MANUSCRIPT

Table 5: Model parameters for trigonelline as determined by fitting.

𝒄𝒍,𝟎 (g L−1) 𝒄𝒔,𝟎 (g L−1) 𝑲 (-) 𝒌 ∗ (s−1)

Basic 2.70 1.66 2.00 × 103 7.35 × 10−5
Sieved small 4.89 2.05 2.37 × 103 3.68 × 10−5
Sieved large 2.90 2.33 2.80 × 103 1.03 × 10−4
Strongly pressed 2.93 1.81 1.38 × 103 9.08 × 10−5