MIT Open Access Articles

A decentralized biomass torrefaction reactor concept. Part

I: Multi-scale analysis and initial experimental validation

The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.

Citation: Kung, Kevin S. et al. "A decentralized biomass torrefaction reactor concept. Part I:
Multi-scale analysis and initial experimental validation." Biomass and Bioenergy 125 (June
2019): 196-203 © 2018 Elsevier Ltd

As Published: http://dx.doi.org/10.1016/j.biombioe.2018.11.004

Publisher: Elsevier BV

Persistent URL: https://hdl.handle.net/1721.1/127993

Version: Author's final manuscript: final author's manuscript post peer review, without
publisher's formatting or copy editing

Terms of use: Creative Commons Attribution-NonCommercial-NoDerivs License

 1 A Decentralized Biomass Torrefaction Reactor Concept. Part I: Multi-Scale Analysis and

2 Initial Experimental Validation

3 Kevin S. Kung1,2,3, Santosh Shanbhogue1,2, Alexander H. Slocum1,2, and Ahmed F. Ghoniem1,2

5 1 Department of Mechanical Engineering, MIT, Cambridge, MA

6 2 Tata Center for Technology and Design, MIT, Cambridge, MA

7 3 Department of Biological Engineering, MIT, Cambridge, MA

9 ABSTRACT:

10 A new, simplified biomass torrefaction reactor concept that operates under oxygen-lean

11 conditions is proposed as a potential way to downscale torrefaction reactors for small- and

12 medium-scale applications. To verify the feasibility of the concept, a multi-scale analysis

13 was conducted to understand the design requirements, underlying chemistry, intra-particle

14 effects, and overall reactor-scale heat transfer. We demonstrate that the heat transfer

15 within the reactor and the appropriate reactor height is largely determined by gas-phase

16 advection. Finally, by implementing a laboratory-scale reactor and operating it under

17 diverse conditions, we show that such a design can indeed satisfy the requirements for

18 torrefaction. This lays the basis for the second part of this two-part paper, where we

19 develop a detailed mathematical model for this concept. In future studies, we will also

20 systematically define and map the performance metrics and reaction conditions in order to

21 understand the scaling laws for potential commercialization of this concept.

22
23 Highlights

24 • A small to medium scale, decentralized biomass torrefaction reactor operating

25 authothermally was proposed.

26 • Multi-scale analysis incorporating reaction thermochemistry and reactor heat

27 transfer was conducted.

28 • An experimental set-up encapsulating this reactor concept was proposed and

29 implemented.

30 • The experimental measurements on the temperature and solid outputs demonstrate

31 initial data consistent with torrefaction.

32 • Detailed mathematical modeling and systematic quantification of the reactor

33 performance are the next steps.

34

35 Keywords: Biomass; torrefaction; reactor design; multi-scale modeling.

36 Corresponding Author: Kevin S. Kung

37 Address: Room 3-339, 77 Massachusetts Avenue, Cambridge, MA 02139

38 Email: [email protected]

39 Telephone: +1.617.324.4449

40 1. Introduction

41 Most biomass residues, in their raw form, are not useful or desirable: they can be wet,

42 loose, and bulky. This makes them difficult and expensive to collect, transport, process, and

43 utilize in energy applications. Furthermore, specific biomass types such as agricultural

44 wastes are seasonal, making it a challenge to store for long-term consumption during the
45 off-seasons. In order to address these challenges, many studies have considered biomass

46 torrefaction, which is a thermochemical process whereby biomass is heated to a

47 temperature of around 200-320°C. Under such conditions, it releases low-energy volatile

48 molecules and densifies into a carbon-rich, energy-dense solid called torrefied biomass,

49 with improved characteristics. For example, torrefied biomass is mostly hydrophobic,

50 making it resistant to moisture attack and extending its shelf life [1]. Torrefied biomass

51 also has improved energy density, which effectively lowers its transportation cost [2].

52 Finally, it takes less energy to grind torrefied biomass into smaller pieces than it does to

53 grind raw biomass. For many thermal applications where the input fuel must first be

54 pulverized, this saves on the pre-processing costs [3].

55

56 To date, various torrefaction reactors have been developed on a laboratory or commercial

57 scale. These have been described elsewhere [4-6]. Table S1 in the supplementary section

58 summarizes some of these reactors as well as their intended production capacity. Most of

59 the torrefaction reactors developed or proposed to date have considerable production

60 capacities, typically on the order of 104-105 kg day-1. While these installations may be

61 effectively used when co-located with large, centralized biomass waste generation sources

62 (such as a lumber mill or an agricultural processing mill), their current designed scale is

63 incompatible for deployment in rural and decentralized areas, where much of the biomass

64 waste is actually being produced.

65

66 In order to understand the scale requirements of decentralized biomass torrefaction, let us

67 consider a potential use case in India using rural agricultural residues. While there is a
68 wide distribution in the sizes, the typical farm size in South Asia is around 1.3 × 104 m2 (1.3

69 hectares) [7]. While the yield of agricultural residues per hectare is highly dependent upon

70 various factors, we assume that a representative value is 0.225 kg m-2 on harvest [8]. Thus,

71 a representative farm in rural India may have about 3,000 kg of post-agricultural residues

72 to be processed, and these farms are spread geographically amongst a large rural region.

73 For a portable, low-cost, low-maintenance torrefaction reactor that can move from farm to

74 farm on a daily basis, it must have a biomass processing capacity of on the order of 103 kg

75 day-1 [9]. This represents the need to downscale current torrefaction reactor designs by a

76 factor of at least 10.

77

78 Svanberg et al. [10], in characterizing the capital cost of current torrefaction reactor

79 designs, noted that almost all existing designs enforce a near-inert condition inside the

80 biomass reactor, and this is one main reason why the capital cost and operational

81 complexity has remained prohibitively high in small- to medium-scale torrefaction

82 applications. That study further commented that if this inert requirement can be relaxed,

83 then both the capital and operational costs associated with the reactor operation are likely

84 to be reduced. In the search of a simplified torrefaction reactor design for decentralized

85 deployment, this low-oxygen torrefaction appears to be an attractive option for further

86 experimentation.

87

88 It is only in recent years, that torrefaction in an oxygen-lean environment has been the

89 subject of laboratory studies. Some studies [11] have found negligible effect of the presence

90 of oxygen on the overall yield and efficiency, while others [12-14] have found that under
 91 highly oxidative environments, there are sacrifices in output characteristics. None of these

92 studies, to the best of our knowledge, have moved beyond an idealized batch-type furnace

93 towards a more realistic reactor design, and it is currently unknown how these effects play

94 out on a continuous reactor prototype.

95

96 This two-part paper has been written with the above-mentioned goal in mind, of designing

97 a biomass torrefaction reactor for small- to medium-scale operations in rural,

98 decentralized regions, and validating its functionality. In Part I, we describe the concept

99 and analysis of an oxygen-lean torrefaction reactor concept that greatly simplifies the

100 reactor design. We then built and tested a laboratory-scale prototype to demonstrate that

101 our concept does indeed meet the requirements for biomass torrefaction. In Part II, we will

102 develop a more detailed and versatile mathematical model that fits with our experimental

103 observations, in order to better understand the design and scaling laws of this reactor

104 concept.

105 2. Basis and Derivation of an Oxygen-Lean Torrefaction Reactor Concept

106 We begin with a simple moving bed, counterflow continuous reactor design, inspired by

107 similar designs in biomass gasifiers. A schematic of the design is shown in Figure 1.

108

109 In this design, raw biomass is continuously fed from the top into a moving bed. At the

110 bottom, a turning auger continuously removes the torrefied biomass from the moving bed,

111 allowing the incoming biomass column to migrate downwards by gravity. Air (at room

112 temperature) is introduced near the bottom of the reactor. With an initial heat source to
113 start the reaction, we hypothesize that the incoming air will continuously combust with a

114 limited amount of biomass in the reaction zone at a reaction temperature that is at or

115 above the auto-ignition point of the specific type of biomass. We further hypothesize that

116 by varying the amount of air injecting at this location, we can increase or decrease the

117 extent of biomass combustion, thereby regulating the steady-state temperature profile in

118 this reaction zone. As the incoming air is consumed by the combustion, this results in a low-

119 oxygen environment that supports torrefaction at the bottom of the moving bed. As the

120 reacting air, volatiles, and flue gas mixture travels upwards through the moving bed in a

121 counter-flow manner, it cools and is exhausted from the top of the reactor. Therefore, while

122 the biomass is traveling downwards, it is also being heated until it reaches torrefaction

123 temperature at the bottom of the moving bed. To the side of the moving bed, there is an

124 extension with a length of the auger conveying biomass away from the moving bed. This

125 length serves two functions. Firstly, as the biomass at the bottom of the moving bed is hot,

126 it needs to be cooled before emerging from the reactor, or else there may be a spontaneous

127 combustion. Therefore, the primary function of the auger extension is to cool the hot,

128 torrefied biomass. Therefore, this section is also referred to as the “char-cooling segment”.

129 Secondly, the auger flights and the inner diameter of the char-cooling segment form quite a

130 tight seal, such that it prevents most of the injected air, volatiles, and post-combustion flue

131 gas mixture from escaping sideways with the cooling char, but rather directs almost all of it

132 upwards through the moving bed. In order for this to happen, the pressure drop for the gas

133 across the char-cooling segment should be higher than that for the gas across the moving

134 bed. This means that for the preliminary design, the char-cooling segment should be at

135 least longer than the height of the moving bed.

136 3. Multiscale Analysis of the Reactor Concept

137 It is prudent to carry out an initial analysis of our proposed reactor concept, in order to

138 verify that it actually meets the various chemistry and heat transfer requirements for

139 torrefaction. We need to consider processes at different scales: at a molecular scale,

140 chemical kinetics of solid devolatilization dictates the temperature and timescale

141 requirements for the torrefaction reaction. At the particle scale, thermal thickness in the

142 biomass particles may affect the homogeneity of the output. Finally, at the reactor scale, the

143 heat transfer mechanisms need to be reconciled with the temperature and timescale

144 requirements identified earlier. A more detailed mathematical model of the reactor is

145 beyond the scope of the present paper, but will be presented in a subsequent study to fit

146 the experimental data obtained in the present paper.

147 3.1 Micro-Scale Kinetics and Thermochemistry

148 In order to describe the solid devolatilization process, we utilized an existing kinetic model

149 proposed by Bates and Ghoniem [15], developed for the case of willow (Figure 2). In this

150 model, the solid-phase devolatilization following a two-step lump-sum process. First, the

151 raw biomass (A) can either decompose into volatile gas (V1) or become solid intermediate

152 (B). Then, the solid intermediate (B) becomes either volatile gas (V2) or char (C). Each of

153 these reactions are assumed to be first-order Arrhenius in nature. In addition, for the

154 drying of biomass, we utilized a simplified model proposed by Peters and Bruch [16],

155 where water bound to the biomass (M(b)) in the solid phase becomes unbound (M(ub))in

156 the gas phase as a first-order Arrhenius-type process.

157
158 This gives us the following rate equations:

−75976
159 𝐴(𝑠) → 𝐵(𝑠) , 𝑘1 = 2.48 × 104 exp ( ),
𝑅𝑇

−114214
160 𝐴(𝑠) → 𝑉1(𝑔) , 𝑘V1 = 3.23 × 107 exp ( ),
𝑅𝑇

−151711
161 𝐵(𝑠) → 𝐶(𝑠) , 𝑘2 = 1.10 × 1010 exp ( ),
𝑅𝑇

−151711
162 𝐵(𝑠) → 𝑉2(𝑔) , 𝑘V2 = 1.59 × 1010 exp ( ),
𝑅𝑇

−87900
163 𝑀(b) → 𝑀(ub) , 𝑘M = 5.56 × 106 exp ( ).
𝑅𝑇

164 In order to understand the timescales (~𝑘𝑖−1) necessary to undertaken each of these ith

165 reactions, in Figure 3(a) we plot these various timescales as a function of temperature. As

166 we can see, for the 200-300°C range, which is typical for torrefaction, the characteristic

167 timescale τres for solid devolatilization ranges from 10 minutes to 10 days. Moreover,

168 drying (dashed purple line) is comparatively fast, ranging from 10 seconds to 10 minutes.

169 Therefore, the solid residence time inside the reactor is imposed by the solid

170 devolatilization kinetics. To understand what this means for the reactor dimensions, we

171 utilize the desired reactor throughput of 𝑚̇BM =5,000 kg day-1 as derived previously. The

172 reactor volume Vr necessary to process 𝑚̇BM of biomass is given by 𝑉𝑟 ~ 𝑚̇BM 𝜏res ⁄𝜌BM ,

173 where the bulk density of biomass ρBM ~ 100 kg m-3. Figure 3(b) plots the characteristic

⁄
174 dimension 𝐿𝑟 ~𝑅𝑟 ~𝑉𝑟−1 3 , for the different reaction steps. We can see that a realistic

175 dimension on the order of one or several meters can be achieved when the biomass solid

176 devolatilization is carried out at the higher temperatures (above 250°C). At a lower

177 temperature, torrefaction reaction proceeds very slowly, and the required solid residence

178 time is long, such that the required reactor dimension is prohibitively large. Therefore, in
179 our reactor design, a higher-temperature, shorter-residence-time treatment is preferred

180 over a lower-temperature, longer-residence-time one. This is consistent with the

181 observations by Prins [17]. Continuing the analysis above, by assuming a characteristic

182 reactor dimension of 1 m, and a maximum residence time of 104 s, we found the

183 characteristic velocity of the biomass through the reactor is vs ~ 10-4 m s-1. This implies

184 that the characteristic mass flux of the biomass through the reactor is Φs ~ ρBM vs ~ 0.01 kg

185 m-2 s-1.

186

187 One major caveat is that the model above was proposed by fitting experimental data in an

188 idealized reactor set-up in inert conditions. The chemical kinetics could be very different in

189 the oxygen-lean torrefaction regime. However, to our knowledge there is currently no

190 detailed chemical kinetics study carried out under such regime, and it is outside the scope

191 of this work to quantify the chemical kinetics in detail. Therefore, for the purpose of

192 general reactor sizing, we chose to utilize the Bates and Ghoniem mechanism as the best

193 alternative substitution.

194 3.2 Particle-Scale Effects

195 Depending on the thermal thickness of the biomass particles, the timescale for heat to

196 penetrate into the biomass particle could also be significant, leading to thermal gradient

197 over the particle and inhomogeneous torrefaction. This problem has been observed

198 previously in torrefaction reactor designs such as fluidized bed [18] and microwave [19].

199 To understand whether or not this particle-scale effect could cause problems for our

200 reactor concept, we select three diverse types of commonly available biomass in our
201 vicinity to analyze and later experimentally test: pine shavings, hay, and rice husks. All

202 these biomass types are also fairly common or resemble common crop and forest residues

203 in India, our case study country. As shown in Table S2 in the Supplementary Section, these

204 types of biomass generally have equivalent spherical diameters of less than 10 mm. The

205 particle-scale effects in the torrefaction regime have been considered earlier by Bates and

206 Ghoniem [18]. This study showed that, as long as the particle thickness is less than ~10

207 mm, the overall heating timescale is slow compared to intraparticle effects. If processing of

208 larger biomass particles is desired, then either such particles need to be first reduced in

209 size, or the reactor configuration (which, as will be explained below, affects the heat

210 transfer timescale) will need to be adjusted accordingly.

211 3.3 Reactor-Scale Heat Transfer Description

212 3.3.1 Mechanisms of Heat Transfer

213 How do we describe the heat transfer characteristics within the biomass moving bed? To

214 inform the modeling process, we examined the literature for modeling counterflow moving

215 bed biomass gasifiers (downward-moving biomass and upward-moving gas). Even though

216 the fundamental thermochemistry of gasification and torrefaction is different, we assume

217 that the physical heat transfer properties through the moving bed should be similar.

218

219 We first note that the moving bed itself actually consists of two inhomogeneous phases: a

220 solid phase (biomass) and a gas phase (a mixture of air, volatile gases, and post-combustion
221 flue gases). The two phases transfer heat differently in the axial direction, and the two

222 phases can also exchange heat with each other. We will analyze each phase separately.

223

224 Solid-phase conduction and radiation. The actual heat transfer coefficient ks in the solid

225 phase is a complex interplay of (i) heat conduction within a single biomass particle, (ii)

226 biomass particle-to-particle contact conduction, (iii) particle-to-particle radiation via the

227 porous void, and so forth. In simplifying these effects, we searched for the closest possible

228 lump-sum description in the moving-bed biomass gasifier literature, and utilized the

229 empirical correlations given by Hobbs [20]. In the thermochemical regime of interest, ks ~

230 1 W m-1 K-1. Working from this value, the characteristic heat transfer timescale due to solid-

231 phase conduction and radiation can be approximated as [18] 𝜌𝑠 𝑐𝑝𝑠 𝐿2𝑟 ⁄𝑘𝑠 ~105 s, where Lr ~

232 1 m is the axial length of the reactor, ρs ~100 kg m-3 the bulk solid density in the moving

233 bed, and cps ~ 1,000 J kg-1 K-1 the specific heat capacity of the gas phase.

234

235 Gas-phase conduction, natural convection, and radiation. We first consider conduction,

236 natural convection and radiation in the gas phase (independently from the axial advective

237 effects). Similar to the solid effective bulk thermal conductivity ks, the combined effects of

238 conduction, convection, and radiation in the gas phase can be described by an effective bulk

239 thermal conductivity kg inside the biomass moving bed, with the empirical correlations

240 given in Hobbs [20]. In our regime of interest, kg ~0.01 W m-1 K-1. The characteristic heat

241 transfer timescale due to gas-phase conduction, convection, and radiation can be

242 approximated as 𝜌𝑔 𝑐𝑝𝑔 𝐿2𝑟 ⁄𝑘𝑔 ~105 s, where ρg ~1 kg m-3 the gas density in the moving bed,

243 and cpg ~ 1,000 J kg-1 K-1 the specific heat capacity of the solid phase.
244

245 Axial advection: Both the solid and gas phases are continuously moving axially (in

246 counterflow) inside the reactor, and therefore they carry enthalpies with them. In the solid

247 phase, we have already determined above that, in order to satisfy the residence time

248 requirement, its downward velocity in the reactor, its superficial flow velocity vs is on the

249 order of 10-4 m s-1. Therefore, its characteristic advective timescale within the reactor (of

250 dimension 1 m) is on the order of 104 s. To determine the characteristic advective

251 timescale of the gas phase, we note that the stoichiometric air-biomass ratio for complete

252 combustion of biomass is typically in the range of 4-7 on a mass basis [21]. From this, we

253 estimate that in the oxygen-lean combustion regime, the amount of air flux to supply into

254 the reactor is likely on the same order as the amount of biomass flux (or an effective excess

255 air ratio of ~0.2):

256 Φair ~ΦBM ~ 0.01 kg m−2 s−1 .

257 Therefore, the gas-phase characteristic advective timescale is on the order of 𝐿𝑟 ρ𝑔 ⁄Φair

258 ~100 s, and its characteristic velocity, vg ~ Φair ⁄ρ𝑔 ~ 0.01 m s-1, which is about 100 times

259 faster than the solid-phase movement.

260

261 Solid-gas heat transfer. In the biomass moving bed, the solid and the surrounding gas can

262 exist at different temperatures (Ts and Tg, respectively), and there may be heat transfer

263 between the two phases governed by the lump-sum heat transfer coefficient hsg. According

264 to prior studies [22-27], for biomass moving bed, hsg ~ 10 W m-2 K-1. It can be shown that

265 the timescale of the solid-gas heat transfer can be approximated by 𝜌𝑠 𝑐𝑝,𝑠 𝑑𝑝 ⁄ℎ𝑠𝑔 ~100 s,

266 assuming that dp ~ 0.01 m from Table S2.

267

268 Heat losses: Both the solid and gas phases can interact with the insulated reactor wall and

269 transfer heat through the wall to the ambient (with a lower temperature Tamb). The heat

270 transfer from the moving bed (gas and solid phases) to the reactor side wall is the first step

271 of the heat loss, and this is characterized by two heat transfer coefficients defined in [28-

272 30], the gas-to-wall heat transfer coefficient hgw ~ 1 W m-2 K-1, and the solid-to-wall heat

273 transfer coefficient hsw ~ 0.01 W m-2 K-1. Therefore, the timescale of gas-to-wall heat

274 transfer is given by 𝜌𝑔 𝑐𝑝,𝑔 𝑅𝑟 ⁄ℎgw which depends on the reactor diameter. For a small

275 laboratory-scale equipment, Rr ~ 0.05 m which means that the gas-to-wall timescale is on

276 the order of 100 s, but as we scale up the reactor for commercial deployment at Rr ~ 1 m,

277 the timescale is lengthened to 1,000 s. On the other hand, for the solid-to-wall heat transfer,

278 the timescale, expressed by the term 𝜌𝑠 𝑐𝑝𝑠 𝑅𝑟 ⁄ℎsw , is 107 s at Rr ~ 0.05 m, and 108 s at Rr ~

279 1 m.

280 3.3.2 Analysis of Heat Transfer Timescales

281 Having described the various types of heat transfer mechanisms above, we are now poised

282 to answer the following questions: Which of these is the rate-limiting step in the heat

283 transfer process? Which steps of heat transfer occur so rapidly that we can approximate

284 them as almost instantaneous? Figure 3

285 summarizes the different heat transfer timescales calculated in the previous section.

286

287 We can see that, in the regime of the torrefaction reactor dimensions (height ~ 40 cm), heat

288 transfer through advection (both solid and gas phases) occurs much more rapidly than
289 bulk effective conduction in the respective phases. Therefore, axially, heat transfer is

290 carried out primarily by the gas-phase advective pathway. Furthermore, gas advection

291 occurs on a timescale 100 times faster than solid advection, suggesting that the solid phase

292 migrates downward so slowly that, in the timescale of the upward gas migration, the solid

293 phase may be treated as pseudo-static. However, as the heat exchange between the solid

294 and gas phases also occur quickly, this suggests that as soon as the heat is carried up axially

295 in the gas phase, the surrounding solid is rapidly heated up without advecting too much of

296 this heat back down towards the bottom of the reactor. While it is tempting to state that,

297 due to the high rate of thermal exchange between the solid and gas phases, Tg ≈ Ts, prior

298 studies have shown that this approximation is not always valid in moving bed biomass

299 gasifiers [31-32]. Finally, at the laboratory scale (Rr ~ 0.05 m), the rate of heat loss from the

300 reactor in the gas phase occurs comparatively rapidly with respect to the residence time

301 (advection timescale). Therefore, heat losses to the reactor wall also cannot be ignored

302 when the test reactor is small. This perspective is contrary to many recent approaches to

303 modeling biomass gasifier, which sometimes ignore heat losses from the side wall [33]. As

304 the reactor is scaled up, we expect the magnitude of the heat loss rate to be reduced

305 inversely to the reactor diameter.

306

307 In Part II of the study, a more detailed mathematical modeling based on the above-

308 mentioned analysis will shed further light into the actual dimensions of the reaction zone,

309 the relative magnitudes of the various heat transfer pathways, as well as how these

310 quantities change as the reactor is scaled up. For now, from the above-mentioned analysis,

311 we will size and design a preliminary experimental reactor and operate it at about ΦBM ~
312 Φair ~ 0.01 kg m-2 s-1 (calculated from Figure 3 above) to show that the reactor can indeed

313 support biomass torrefaction.

314 4. Experimental Set-Up and Design

315 In the previous section, we have gained a coarse-grained understanding of the significant

316 chemical and thermal processes occurring within the reactor. In this section, we describe

317 how we designed and built an experimental apparatus to test our design concept further.

318

319 In order to control the torrefaction conditions under the proposed direct oxygen lean

320 conditions, the test reactor needs to be able to control two independent variables that

321 dominate the chemistry of torrefaction: the solid-phase residence time, and the reactor

322 temperature, which itself depends on how much air is injected relative to how much

323 biomass passes through the reactor. In the next section, we describe our strategies for

324 controlling these two independent parameters.

325 4.1 Control of Solid Residence Time

326 A lab-scale, continuous-flow biomass torrefaction reactor design for a feedstock capacity of

327 about 1 kg/h and a cross-sectional diameter of 10.2 cm, was fabricated and tested in this

328 study. This test reactor utilizes a moving-bed reactor concept (labeled in Figure 4a), with

329 an auger at the bottom with the function of continuously removing the torrefied biomass.

330 The auger was sized to be 61 cm (2 feet) in length in order to allow sufficient time for the

331 hot, torrefied biomass to cool off before it exits the reactor outlet. The auger is connected to
332 a motor drive with an adjustable speed, so that we are able to control the solid-phase

333 residence time inside the reactor for the torrefaction reaction.

334

335 In order to calculate the solid residence time, the density ρBM and input mass flow rate

336 of biomass were measured for each type of biomass and auger rotational speed. Given the

337 radius Rr and height Hr of the fixed-bed reactor, the effective residence time τres of biomass

338 is obtained from:

𝜌BM 𝜋𝑅𝑟2 𝐻𝑟
339 𝜏res = .
𝑚̇BM

340 Given our specific lab-scale reactor dimensions (diameter = 10.2 cm, depth of torrefaction

341 zone ~ 10 cm), in Table 2, we provide the typical range of possible input feed rates

342 corresponding to the normal ranges of torrefaction residence times.

343

344 It can be seen that for our lab-scale reactor, by fixing the torrefaction residence time, the

345 biomass feeding rate is generally on the order of 0.1-2 kg/h. While this is sufficient for lab-

346 scale tests for biomass samples in small quantities, at scale, a commercial torrefaction

347 reactor capable of processing about 500 kg/h of biomass will need to be increased by a

348 factor of 250-5,000. In terms of the reactor diameter, this will be an increase of a factor of

349 16-71. These numbers are, once again, consistent with the analysis above in Figure 3.

350 4.2 Control of Torrefaction Severity

351 At a given solid residence time (i.e. feed rate and reactor volume), in order to control the

352 torrefaction severity, we need to introduce air in different proportions to the solid feedrate.
353 As shown in Figure 4(c), in order to achieve this, we introduced two air flow inlets near the

354 base of the fixed bed reactor. These inlets are connected to a compressed air cylinder

355 (Airgas AI D300). A mass flow controller (Omega Engineering FMA-5528A) was used to

356 adjust the input air flow rate, with an adjustable flow rate range between 0.1 to 50.0

357 standard L/min. We divide the mass flow rate of air by the mass flow rate of the incoming

358 biomass to obtain a dimensionless air-biomass ratio (ABR). For normal operations, with an

359 ABR ~ 1.0, the typical air flow rate is set to ~10 L/min, depending on the specific type of

360 biomass and its specific solid mass flow rate.

361 5. Demonstration of Stable Operating Conditions

362 Once we have defined the independent parameters (solid residence time τres and the air-

363 biomass ratio), we can select a few combinations of (τres , ABR) to demonstrate that the

364 torrefaction reactor can indeed operate continuously and stably.

365

366 In order to monitor the reactor operation, we inserted 10 thermocouples, in approximately

367 2.5-cm spacing, axially throughout the length of the reactor (Figure 5a). As a sample proof

368 of steady-state operation, we flowed air into the reactor at the normalized air-biomass

369 ratio of ABR = 1.0, and the screw auger turning rate was adjusted such as the solid

370 residence time τres = 11 min. We then ignited some initial pine shavings at the bottom of

371 the reactor, and continuously fed pine shavings from the top such that the level of pine

372 shavings at the top was maintained constant. After an initial start-up period, where the

373 entire reactor assembly and the fixed bed were heated up, a steady-state condition was

374 reached at each axial location, and we demonstrated that the steady-state condition could
375 be sustained for at least 100 minutes before turning the reactor off by quenching the

376 torrefaction reaction with inert nitrogen in place of air. The mean steady-state temperature

377 readouts at different axial locations were compiled into an axial temperature profile of the

378 reactor (Figure 5b). We note that the fixed bed temperature increases axially towards the

379 bottom of the reactor, and reaches a maximum near where the air inlets are located. The

380 maximum temperature measured, reaching above 200°C, is within the torrefaction regime.

381

382 We have demonstrated a stable reactor steady state for one particular reactor operating

383 condition for pine shavings. In order to show that this is not a one-off coincidence, the next

384 question that we address is: Can we show steady states for a wide range of reactor

385 conditions? We operated the reactor under additional conditions, by first changing the air-

386 biomass ratio (τres = 11 min, ABR = 2.0), then changing the residence time also (τres = 37

387 min, ABR = 2.4), and finally changing the biomass type from pine shavings to rice husks.

388 Figure 6a-d show that each of these conditions also result in stable steady-state

389 temperature profiles. Notably, in (b), compared to (a) as we increase the torrefaction

390 severity by introducing more air, we observe a higher measured temperature, as expected.

391 In (c), compared to (a), we observe a gentler increase in the measured temperature profile

392 throughout a broader region of the reactor, because as we increase the solid residence time

393 given the same reactor dimension, we are also reducing the downward solid flux.

394 Therefore, the upward countercurrent gas flux will carry the heat from the bottom of the

395 reactor further up. In (d) compared to (a), we observe a sharper increase in the measured

396 temperature profile as we switch from pine shavings to rice husks, because rice husks have
397 a higher bulk bed density, which means a reduced effective axial effective thermal

398 conductivity.

399

400 To further demonstrate that the reactor outputs are indeed torrefied biomass instead of

401 just burned ash, we subject the output from each of the experiments to proximate analysis

402 (TA Instruments Q50 TGA using the ASTM D-3175 standard), ultimate (elemental) analysis

403 (Thermo Finnigan Flash EA 1112 CHON Analyzer using the ASTM-D5373-16 standard), and

404 bomb calorimetry (Parr Instrument Model 6200 isoperibol). These results are detailed in

405 Table 4, in comparison with the same analysis for raw (untreated) biomass in Table 3.

406

407 In interpreting the various results given in Table 4, we also obtained the proximate

408 analysis, ultimate analysis, and bomb calorimetry measurements for raw pine shavings and

409 raw rice husks (presented in Table 3). Firstly, in terms of the higher heating value (HHV),

410 that of raw pine shavings is around 20.5 ± 1.0 MJ kg-1, and that of rice husks is around 15.8

411 ± 1.0 MJ kg-1, on a dry basis. Therefore, all the experiments above have raised the HHV

412 values of the raw biomass by different amounts. This is consistent with the energy

413 densification effect of biomass torrefaction. Secondly, in terms of the fixed carbon, that of

414 raw pine shavings is 8% ± 1% and that of rice husks, 12% ± 2%, on a dry basis. In

415 comparison, we also see elevated fixed carbon readouts in all the outputs from the different

416 experiments, consistent with biomass torrefaction. Finally, we note that for raw pine

417 shavings, the elemental composition is such that (C, H, O) = (49.6%, 5.7%, 44.8%), and for

418 raw rice husks, (C, H, O) = (37.6%, 3.9%, 58.3%), on a dry basis. By comparing these

419 baseline values with those obtained in the experimental outputs, we observe that our
420 reactor conditions cause the raw biomass to gain carbon and lose oxygen. These chemical

421 changes are also consistent with biomass torrefaction. As we make the torrefaction

422 conditions more severe such as by increasing the air-biomass ratio, we notice that HHV,

423 fixed carbon content, elemental carbon content are also increased, while elemental

424 hydrogen and oxygen contents are decreased. These are consistent with the expected

425 principles and trends of biomass torrefaction, which increases the energy content and

426 carbon content in the raw biomass. Another interesting point to note is that, as ash does

427 not volatilize in biomass torrefaction, any torrefaction reaction tends to concentrate the

428 amount of ash in the torrefied biomass. This is especially remarkable in the case of rice

429 husks, which already have a high ash content to start with, and in our torrefied rice husk

430 sample, more than a third of the mass constitutes ash. We expect that the presence of this

431 inert solid will make our torrefaction reactor operation less efficient with respect to high-

432 ash biomass feedstock varieties.

433

434 There is much more that can be quantified regarding the reactor operations, and the scope

435 of this present paper is to demonstrate the initial validation data for our experimental set-

436 up. In a subsequent study, we will develop a fine-grained model description of these

437 experimental data in order to better understand the scaling law of this reactor concept. We

438 will also define more concrete performance metrics, and conduct a more systematic

439 mapping of the reactor conditions to these performance metrics under a wider range of

440 biomass.
441 6. Conclusion and Discussions

442 6.1 Conclusion

443 In Part I of this two-part study, we presented a case for the design and development of a

444 small- to medium-scale torrefaction reactor for decentralized deployment in rural areas.

445 Based on the concept of oxygen-lean torrefaction, we developed a countercurrent moving

446 bed reactor. By performing a multi-scale analysis and subsequently implementing a

447 laboratory-scale reactor set-up, we showed that the temperature profiles, output solid

448 higher-heating value, proximate analysis, and elemental analysis values are generally

449 consistent with generally accepted biomass torrefaction conditions. This validates our

450 reactor concept, and shows that it can indeed satisfy the requirements for biomass

451 torrefaction.

452 6.2 Discussion

453 In Part II of the study, we will develop a more detailed mathematical model description,

454 based on the multi-scale analysis performed above. In subsequent studies, we will also

455 introduce a framework where we can systematically map various performance metrics of

456 this oxygen-lean torrefaction reactor. Of course, the validation of our oxygen-lean

457 torrefaction reactor concept will not be complete until we compare its performance

458 metrics with current torrefaction reactors that operate under inert conditions. This

459 quantitative comparison will also be performed as a part of a future study.

460

461
462 Acknowledgements

463 The materials and equipment of work were funded by the MIT Tata Center. In addition, KSK

464 would like to acknowledge the MIT Tata Center Fellowship, the Dolores Zohrab Liebmann

465 Fellowship, as well as the Legatum Fellowship for supporting his tuition and stipend as a

466 doctoral student. The authors declare no competing interest.

467
468 Bibliography
469
470 [1] B. Acharya, A. Dutta, Fuel property enhancement of lignocellulosic and
471 nonlignocellulosic biomass through torrefaction, Biomass Conversion and Biorefinery 6
472 (2015) 139-149.
473
474 [2] J.S. Tumuluru, S. Sokhansanj, J.R. Hess, C.T. Wright, and R.D. Boardman, A review on
475 biomass torrefaction process and produt properties for energy applications, Industrial
476 Biotechnology 7 (2011) 384-401.
477
478 [3] B. Arias, C. Pevida, J. Fermoso, M.G. Plaza, F. Rubiera, J.J. Pis, Influence of torrefaction on
479 the grindability and reactivity of woody biomass, Fuel Processing Technology 89 (2008)
480 169-175.
481
482 [4] D. Thrän, J. Witt, K. Schaubach, J. Kiel, M. Carbo, J. Maier, C. Ndibe, J. Koppejan, E.
483 Alakangas, S. Majer, F. Schipfer, Moving torrefaction towards market introduction –
484 Technical improvements and economic-environmental assessment along the overall
485 torrefaction supply chain through the SECTOR project, Biomass and Bioenergy 89 (2016)
486 184-200.
487
488 [5] J. Koppejan, Status overview of torrefaction technologies. IEA Bioenergy Task 32 report,
489 2012.
490
491 [6] Kung’s book chapter on green charcoal production
492
493 [7] S.K. Lowder, J. Skoet, T. Raney, The number, size, and distribution of farms, smallholder
494 farms, and family farms worldwide, World Development 87 (2016) 16-29.
495
496 [8] M. Laurin, A. Chamberland, Gasification of agricultural residues for energy production,
497 Energy Sources 5(4) (1981) 361-80.
498
499 [9] K.S. Kung, Design and validation of a decentralized biomass torrefaction system, PhD
500 thesis, Massachusetts Institute of Technology, 2017.
501
502 [10] M. Svanberg, I. Olofsson, J. Floden, A. Nordin, Analysing biomass torrefaction supply
503 chain costs, Bioresource Technology 142 (2013) 287-96.
504
505 [11] C. Wang, J. Peng, H. Li, X.T. Bi, R. Legros, C.J. Lim, S. Sokjansanj, Oxidative torrefaction of
506 biomass residues and densification of torrefied sawdust to pellets, Bioresource Technology
507 127 (2013) 318-25.
508
509 [12] K.M. Lu, W.J. Lee, W.H. Chen, S.H. Liu, T.C. Lin, Torrefaction and low temperature
510 carbonization of oil palm fiber and eucalyptus under nitrogen and air atmospheres,
511 Bioresource Technology 123 (2010) 98-105.
512
513 [13] W.H. Chen, K.M. Lu, S.H. Liu, C.M. Tsai, W.J. Lee, T.C. Lin, Biomass torrefaction
514 characteristics in inert and oxidative atmospheres at various superficial velocities,
515 Bioresource Technology 146 (2013) 152-60.
516
517 [14] W.H. Chen, K.M. Lu, W.J. Lee, S.H. Liu, T.C. Lin, Non-oxidative and oxidative torrefaction
518 characterization and SEM observation of fibrous and ligneous biomass, Applied Energy 114
519 (2014) 104-13.
520
521 [15] R. Bates, A.F. Ghoniem, Biomass torrefaction: Modeling of volatile and solid product
522 evolution kinetics, Bioresource Technology 124 (2012) 460-9.
523
524 [16] B. Peters, C. Bruch, Drying and pyrolysis of wood particles: experiments and
525 simulation, J. Analy. Appl. Pyrolysis 70 (2003) 233-50.
526
527 [17] M.J. Prins, Thermodynamic analysis of biomass gasification and torrefaction, PhD
528 thesis, Eindhoven University of Technology, 2005.
529
530 [18] R. Bates, A.F. Ghoniem, Modeling kinetics-transport interactions during biomass
531 torrefaction: The effects of temperature, particle size, and moisture content, Fuel 137
532 (2014): 216-29.
533
534 [19] A. Dhungana, P. Basu, A. Dutta, Effects of reactor design on the torrefaction of biomass,
535 Journal of Energy Resources Technology 134 (2012).
536
537 [20] M.L. Hobbs, P.T. Radulovic, L.D. Smoot, Modeling fixed-bed coal gasifiers, AIChE 38(5)
538 (1992) 681-702.
539
540 [21] B.M. Jenkins, L.L. Baxter, T.R. Miles Jr., T.R. Miles, Combustion properties of biomass,
541 Fuel Processing Technology 54 (1998) 17-46.
542
543 [22] T.H. Chilton, A.P. Colburn AP, Ind. Eng. Chem. 26 (1934) 1183.
544
545 [23] B.W. Gamson, G. Thodos, O.A. Hougen, Trans. Am. Inst. Chem. Engrs. 39 (1943) 1.
546
547 [24] J. De Acetis J, G. Thodos, Ind. Eng. Chem. 52 (1960) 1003.
548
549 [25] A.S. Gupta, G. Thodos, Direct analogy between mass and heat transfer to beds of
550 spheres, AIChE 9(6) (1963) 751.
551
552 [26] C. Di Blasi C, Modeling wood gasification in a countercurrent fixed-bed reactor, AIChE
553 50(9) (2004) 2306-19.
554
555 [27] P.T. Radulovic, M.U. Ghani, L.D. Smoot, An improved model for fixed bed coal
556 combustion and gasification, Fuel 74(4) (1995) 582-94.
557
558 [28] C. Di Blasi C, Dynamic behavior of stratified downdraft gasifiers. Chemical Engineering
559 Science 55 (2000) 2931-44.
560
561 [29] G.F. Froment, K.B. Bischoff, Chemical Reactor Design. Wily, New York, 1979.
562
563 [30] A.P. DeWasch, G.F. Froment GF, A two-dimensional heterogeneous model for fixed bed
564 catalytic reactors, Chem Eng Sci 26 (1971) 629.
565
566 [31] J. Cooper, W.L.H. Hallett, A numerical model for packed-bed combustion of char,
567 Chemical Engineering Science 55 (2000) 4451-60.
568
569 [32] L. Rosa, R. Tosat, Modelling and testing a cogeneration plant based on wood
570 gasification, 7th European Conference on Industrial Furnaces and Boilers, 2006.
571
572 [33] M. Puig-Arnavat, J.C. Bruno, A. Cornoas, Review and analysis of biomass gasification
573 models, Renewable and Sustainable Energy Review 14 (2010) 2841-2851.
574
575 FIGURES

576

577 Figure 1 – (a) Schematic of an oxygen-lean moving bed torrefaction reactor design. (b)

578 Guessed axial temperature profile along the reactor.

579
580
581
582

583

584 Figure 2 - A simplified kinetics model for torrefaction and drying. (a) A kinetic description

585 of the solid devolatilization kinetics during biomass torrefaction, as proposed by Bates and

586 Ghoniem [10]. (b) A one-step simplified model describing drying kinetics in biomass, as

587 adopted from Peters and Bruch [11].

588
589

590 Figure 3. (a) Characteristic timescales and (b) characteristic reactor dimensions related to

591 the various solid devolatilization and drying steps in biomass torrefaction.

592
593

594 Figure 4 - A lab-scale biomass torrefaction reactor was designed and built. (a) Cross-

595 section rendering of the reactor design in SolidWorks, with biomass feeding inlet (1), fixed-

596 bed reactor zone (2), screw auger (3), and torrefied product outlet (4) labeled. (b) An

597 actual lab-scale test reactor assembly in operation. (c) We utilized adjustable air flow inlets

598 to control torrefaction severity. As shown in the schematic, the air is injected at the bottom

599 of the reactor and its flow rate is controlled via a mass flow controller.
600

601 Figure 5 - Sample steady-state temperature validation for the torrefaction reactor

602 operating on pine shavings, with the following reaction conditions τres = 11 min, ABR = 1.0.

603 (a) Throughout the axial length of the reactor, various thermocouples are placed about 2.5

604 cm apart to measure the temperature at each point in the biomass fixed bed. This

605 schematic is juxtaposed next to the distance axis in (b) to illustrate the locations of each

606 temperature measurement at the center of the moving bed reactor in the context of the

607 larger reactor design. (b) A stable steady-state axial temperature profile consistent with

608 torrefaction conditions, with error bars indicated, is achieved for at least 100 minutes.

609
610

611 Figure 6 - Steady-state temperature profiles were achieved under a variety of reactor

612 operating conditions, for pine shavings for the torrefaction reactor under various operating

613 conditions: (a) τres = 11 min, ABR = 1.0, and (b) τres = 11 min, ABR = 2.0, and (c) τres = 37

614 min, ABR = 2.4. We also operated the reactor using (d) rice husks, with τres = 27 min, ABR =

615 1.0.

616
617 TABLES
618
619 Table 1 – Characteristic timescales of the various heat-transfer mechanisms, with a lab-

620 scale reactor Rr = 0.05 m.

Heat transfer mechanism Characteristic timescale

Solid-phase bulk effective conduction 𝜌𝑠 𝑐𝑝𝑠 𝐿2𝑟

~105 s
𝑘𝑠

Gas-phase bulk effective conduction 𝜌𝑔 𝑐𝑝,𝑔 𝐿2𝑟

~105 s
𝑘𝑔

Gas-phase advection 𝐿𝑟
~102 s
𝑣𝑔

Solid-phase advection 𝐿𝑟
~104 s
𝑣𝑠

Gas-to-wall heat loss 𝜌𝑔 𝑐𝑝,𝑔 𝑅𝑟

~102 − 103 s
2ℎgw

Solid-to-wall heat loss 𝜌𝑠 𝑐𝑝,𝑠 𝑅𝑟

~107 − 108 s
2ℎsw

Solid-gas exchange 𝜌𝑠 𝑐𝑝,𝑠 𝑑𝑝

~102 s
ℎ𝑠𝑔

621

622
623 Table 2 - Typical ranges of possible input feeding rates in our lab-scale reactor for different

624 solid residence times. The possible ranges of feeding rates depend significantly on the type

625 of biomass.

Residence time 5 min 20 min 40 min

Pine shavings / hay 0.58 kg/h 0.15 kg/h 0.073 kg/h

Rice husk 2.0 kg/h 0.51 kg/h 0.26 kg/h

626

627 Table 3 - Proximate analysis, ultimate (elemental) analysis, and higher heating value

628 measured on the raw (untreated) pine shavings and rice husks used for our experiments,

629 on a dry basis. The error bars are based on triplicate measurements.

Biomass type Pine shavings Rice husks

Bulk density 30 kg m-3 100 kg m-3

HHV (dry basis) 20.5 ± 1.0 MJ kg-1 15.8 ± 1.0 MJ kg-1

Fixed carbon 8 ± 1% 12 ± 2%

Ash content 1.6 ± 1% 17.0 ± 2%

Elemental C 49.6% 37.6%

Elemental H 5.7% 3.9%

Elemental O 44.8% 58.3%

630
631 Table 4 – Proximate analysis, ultimate (elemental) analysis, higher heating value, and yield

632 values measured on the solid fuel outputs from the different experiments, on a dry basis.

Parameter Experiment (a) Experiment (b) Experiment (c) Experiment (d)

Biomass type Pine shavings Pine shavings Pine shavings Rice husks

Residence time 11 min 11 min 37 min 27 min

ABR 1.0 2.0 2.4 1.0

Output HHV 25.1 MJ kg-1 27.3 MJ kg-1 23.8 MJ kg-1 17.6 MJ kg-1

Fixed carbon 31% 41% 32% 48%

Ash 0.8% 0.9% 1.0% 37%

Elemental C 58% 66% 61% 44%

Elemental H 5% 4% 5% 2%

Elemental O 37% 30% 35% 54%

Mass yield 51% 36% 52% 62%

Energy yield 63% 47% 60% 69%

633
634
635 SUPPLEMENTAL MATAERIALS
636
637 Supplemental Table S1 – Design production capacities of different torrefaction
638 technologies [1]

Torrefaction Technology Design Production Capacity

Torr-Coal (the Netherlands) Rotary drum 110 tons day-1
BioEndev (Sweden) Screw conveyor 50 tons day-1
Solvay (USA) Screw conveyor 800 tons day-1
Topell (the Netherlands) Fluidized bed 190 tons day-1
River Basin Energy (USA) Fluidized bed 140 tons day-1
ECN (NL) / Andritz (Denmark) Moving bed 20 tons day-1
Thermya/Areva (France) Moving bed 60 tons day-1
639

640
641 Supplemental Table S2 – Characteristic dimensions of sample biomass particles, as

642 computed by averaging 100 sample biomass particles from each type. The sample photo

643 illustrates how the lengths of major and minor axes are measured.

Biomass type Pine shavings Hay Rice husks

Sample photo

Major axis 11.1 mm 6.4 mm 0.2 mm

Minor axis 6.4 mm 4.4 mm 1.8 mm

dp 3.0 mm 9.4 mm 2.9 mm

644