Furfural Production by Reactive Stripping: Process Optimization by a

To this end, a predictive activity-based first-principle reactor model was developed for process optimization purposes. This model considers several ...
0 downloads 0 Views 6MB Size
Subscriber access provided by UNIV OF LOUISIANA

Kinetics, Catalysis, and Reaction Engineering

Furfural Production by Reactive Stripping: Process Optimization by a Combined Modelling and Experimental Approach Vladan Krzelj, Jasper van Kampen, John Van Der Schaaf, and Maria Fernanda Neira D'Angelo Ind. Eng. Chem. Res., Just Accepted Manuscript • DOI: 10.1021/acs.iecr.9b00445 • Publication Date (Web): 09 Apr 2019 Downloaded from http://pubs.acs.org on April 15, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Furfural Production by Reactive Stripping: Process Optimization by a Combined Modeling and Experimental Approach

V. Krzelj, J. van Kampen, J. van der Schaaf, and M. F. Neira d'Angelo



Chemical Reactor Engineering Laboratory, Chemical Engineering and Chemistry Department., Eindhoven University of Technology, The Netherlands E-mail: *[email protected]

Abstract In this paper, we describe the continuous production of furfural coupled with insitu stripping using hydrogen gas. With respect to the conventional semi-batch process that requires excessive steam as stripping agent and results in highly diluted furfural, this new process conguration reduces the net energy input, increases the eciency of the downstream hydrogenation of furfural, and proposes a shift towards a continuous operation. Based on well-established thermodynamic data and previously reported kinetics, we have developed a rst-principle reactor model that successfully describes the experimental observations without the use of any tting parameters. This robust predictive model is used to further optimize the continuous production of furfural by this route.

1

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Introduction Following concerns about the environment and the long-term availability of resources, the scientic community is increasingly devoted to ease the transition to a bio-based economy. 14 Biomass, as the only renewable source of carbon, is perceived as the ideal feedstock for the production of chemicals, and bio-derived furanics are foreseen to play a major role as biobased building blocks. For example, furfural is a key intermediate in the valorisation of the hemicellulose fraction of biomass. 46 Around 60-70% of the furfural produce worldwide is further hydrogenated to furfuryl alcohol for the production of sand binders in foundry industry and the remaining part is used as extractant for aromatics from lubricating oils, purication solvent for C4 and C5 hydrocarbons, reactive solvent, wetting agent, chemical feedstock for other furan derivatives and others. 7 Further, furfuryl alcohol oers alternative leads such as the possibility to convert to levulinate esters or ethylfurfuryl ether. 8 Other uses of furfural include its hydrogenation to methylfuran and methyltetrahydrofuran (i.e., two promising gasoline components) or decarbonylation to furan and subsequently hydrogenation to tehrahydrofuran. 8 Recently, it was also shown that furfural can be further hydrogenated to cyclopentanone, a potentially attractive monomer in the polymer industry. 9 Currently, the production of furfural is carried out via hemicellulose hydrolysis and dehydration in acid media, usually with sulphuric acid, operated in a semi-batch mode. 10,11 Corn cobs or bagasse are usually the starting material due to their high hemicellulose content, while the remaining biomass constituents (i.e., cellulose and lignin) are not being utilized. A large amount of steam is used to strip the furfural produced from the reaction medium, thus minimizing its subsequent degradation via condensation and resinication reactions. 10 While this process uses an excessive amount of steam, the usual production yield remains rather low. As a consequence, several industrial plants based on this process (e.g. Agrifurane, Escher Wyss, Rosenew) were discontinued due to the high costs, 10 while newer pilot processes are being developed with the aim to improve the overall eciency and yield (e.g. Bione, Vedernikovs, Suprayield, CIMV, Lignol, MTC.) 11 In the case of Huaxia/Westpro 2

ACS Paragon Plus Environment

Page 2 of 33

Page 3 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

plants, currently operating in China, 25-35 tons of steam is consumed per ton of furfural at about a 50% yield of theoretical pentosan content. 12 Further increasing furfural stripping renders product dilution, which eventually results in expensive downstream separation. Even more, recycling pure steam back to the reactor without additional purication steps is not possible since water and furfural form a minimum boiling point azeotrope. From this point of view, the use of a "non-condensable" gas (e.g., CO2 or hydrogen) as stripping agent may be an attractive alternative as they can be easily separated from furfural by a ash step, and subsequently recycled back to the reactor. Agirrezabal and co-workers compared the nitrogen stripping process and stream stripping process, and showed that the former can lead up to 60% reduction in overall utility costs. 13 Interestingly, if hydrogen is used as stripping agent, the mixtures furfural/hydrogen may be directly fed to the downstream hydrogenation reactors with no additional separation steps. Despite the promising results anticipated by Agirrezabal and co-authors, the thermodynamic and kinetic implications of such process conguration during the synthesis of furfural, and thus on the overall eciency of the process have not been thoroughly investigated. 14,15 Furthermore, a shift from semibatch to a continuous operation mode would be extremely advantageous from an industrial implementation perspective. This work aims to contribute to the understanding and eventual development of a more ecient and continuous furfural production process using a non-condensable gas (e.g., hydrogen) as a stripping agent. We have performed the synthesis of furfural using xylose as model compound, representative of other feedstocks like hemicellulose hydrolysates, which will become of crucial importance in the future bioreneries where biomass is fractionated prior to valorization of the individual streams, or even more complex biomass feeds. Xylose yields similar amounts of furfural as other biomass feedstocks like corncob, 13 while it still provides a well-dened reaction system that allowed us to study the eect of thermodynamics and reactor conguration on the overall performance. To this end, a predictive activitybased rst-principle reactor model was developed for process optimization purposes. This 3

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

model considers several thermodynamic and kinetic models, 1618 and successfully describes the experimental observations from semi-batch and continuous reactions using nitrogen as stripping gas. Further, the model can be extrapolated to other non-condensable gases, and is hereby used to evaluate the inuence of various process parameters (i.e., stripping gas ow rate, temperature, feed and catalyst concentration) on xylose conversion and selectivity towards furfural.

Methods Experimental setup and procedure The experiments were performed in a 300 ml mechanically agitated titanium autoclave reactor as sketched in Scheme 1. The reactor may be operated in semi-batch mode with continuous stripping (see Scheme 2a) or in a continuous/fed-batch mode where both gas and liquid are continuously fed but only gas exits the reactor (see Scheme 2b). Both gas and liquid ow rates are controlled by a Bronkhorst mass ow controllers (MFC) and a Gilson HPLC pump, respectively, and the reactor pressure is regulated by a Bronkhorst back pressure regulator (PC) located after a condensing vessel at the reactor outlet. Note that the gas ow rates are reported at normal conditions (i.e., 20◦ C and 1 atm). Two thirds of the reactor were lled with a solution of sulfuric acid (0.5wt%) before the start of the each experiment. Then, the reactor was purged, pressurized and heated up to the desired reaction conditions (i.e., 10 bar, 170◦ C). Both the reactor wall and the reactor lid were heated to avoid cold surfaces and reux of the vapors. Absence of temperature gradients inside the reactor were ensured by checking that the water collected in the condensing vessel perfectly matches the model predictions assuming isothermal operation. After stable temperature conditions were reached, initial concentration of xylose was achieved by feeding the concentrated xylose solution over a short period of time (1-2 min). The end of this feeding time marks start of the reaction (i.e., t=0). At this moment, a preheated nitrogen 4

ACS Paragon Plus Environment

Page 4 of 33

Page 5 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Scheme 1: The experimental setup

Scheme 2: a) Semi-batch operation; 2b) fed-batch/continuous operation

5

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 33

stream was introduced from the bottom of the reactor. Two dierent modes of operation were tested: a) semi-batch with continuous nitrogen stripping; b) fed-batch with continuous nitrogen stripping, as sketched in Scheme 2. From this point onward, the semi-batch operation will be referred as batch, and the fed-batch will be referred as continuous.

Product analysis The samples were analyzed by HPLC Shimadzu LC 20AD. The column used was Agilent Metacarb 67C. Xylose was detected with Refractive index detector (RI), while the furfural was detected with PDA detector. The mobile phase was HPLC grade water with a ow rate ml . The column oven was set to temperature of 85◦ C. Conversion of xylose (Xxyl ), of 0.5 min

total furfural yield (YT ) and furfural yield in condensate (YC ) were calculated using the equations:

Xxyl

YT =

 = 1−

moles of xylose at time t initial moles of xylose+moles of xylose fed

 · 100

(1)

moles of furfural at time t in reactor+moles of furfural at time t in condensate · 100 initial moles of xylose+moles of xylose fed (2) moles of furfural at time t in condensate YC = · 100 (3) initial moles of xylose+moles of xylose fed

Model Assumptions The reactor model developed in this study assumes that the reactor presents and ideally mixed behavior, and that the mass transfer between the phases is suciently fast to guarantee equilibrium conditions, which is in line with the experimental observations. Further, the vapor-liquid equilibrium is described by the modied Raoult's law, and the gas phase is assumed to obey the ideal gas law. Xylose, sulfuric acid and side products are considered 6

ACS Paragon Plus Environment

Page 7 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

to be non-volatile under the reaction conditions. All the relevant equations are presented in the following sections.

Kinetic models Several kinetic models have been reported for this reaction. The various reaction schemes and kinetic data are listed in Table 1. Table 1: Kinetic models Scheme

Expression

ln A[s−1 ]

MODEL

X ⇒ F + 3H2 Oa

31.60

133.30

A

X ⇒ DP1 F ⇒ DP2

Ea ln k ∗ = ln A − RT k = k ∗ [H3 O+ ] γH3 O+

30.39 26.64

133.30 125.10

MODEL

X ⇒ F + 3H2 O

28.53

123.91

B

X + F ⇒ DP1 F ⇒ DP2

15.78 10.73 Ar [min−1 ] 0.3

72.47 67.58

41

132.9

Ea RT k = k ∗ [H3 O+ ]

ln k ∗ = ln A −

X ⇒ Xu MODEL

a



h

−Ea R



1 T

1 Tr

i

− Xu ⇒ F k = Ar exp C X⇒F X ⇒ DP1 k = k ∗ [H3 O+ ] Xu ⇒ DP2 F ⇒ DP3 X ≡ Xylose; F ≡ F urf ural; Xu ≡ Xylulose; DP1 , DP2 , DP3

Ea

 kJ  mol

Ref 16

17

67.4 18

4.5 166.2 2.3 162.9 28 134.5 0.2 72.0 ≡ Degradation products

As can be seen in Table 1, Model A considers the direct dehydration of xylose to furfural, and two degradation routes, one from xylose and one from furfural. In addition, this model assumes the same temperature dependence for the rate of the two xylose-consuming reactions. On the other hand, model B assumes that there is no direct decomposition of xylose, but together with furfural xylose undergoes condensation reactions to form degradation products. Lastly, model C includes an additional route for furfural formation via dehydration of xylulose, a xylose isomer. In this model it is assumed that both xylose and xylulose, as well as the product furfural are susceptible to degradation reactions. In this work, kinetic 7

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 33

models A, B and C have been considered for the prediction of the activity data without using any tting parameter. Model A includes the dependency of the kinetic constant on the activity of the acid. 16 In this work, the activity coecient of H2SO4 was obtained from eNRTL Aspen database, and tted as a function of temperature and acid concentration. It should be noted that the values originally reported for the kinetic constants (k) in models B and C were estimated for a xed acid concentration. In this work, however, the concentration of acid during the batch mode of operation will not remain constant as water evaporates together with furfural during the stripping process. To incorporate the eect of increasing acid concentration during the reactive stripping process, a rst order kinetics with respect to the concentration of protons has been assumed, and a new kinetic constant (k') has been dened accordingly. Notice that the values of k' have been reported in Table 1. Finally, it should also be mentioned that model B has been originally proposed by Weingarten 17 to describe the dehydration of xylose with HCl instead of H2SO4, a diprotic acid. However, it is well known that the second dissociation of sulfuric acid largely depends on temperature, 19 and that the bisulfate ion becomes a very weak acid at the temperatures of interest of this work. Therefore, the second dissociation step is neglected.

Thermodynamic models Several thermodynamic models have been considered for the prediction of activity coecients. These are listed in Table 2.

Mole balances The ordinary dierential equations that describe the system are derived from the mole balances:

dNf ur,L dt

dNxyl = Fxyl − (k1 + k2 ) Cxyl Cacid γacid VL dt   γf ur χf ur psat f ur VV = k1 Cxyl Cacid γacid VL − KL a − Nf ur,V RT 8

ACS Paragon Plus Environment

(4) (5)

Page 9 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Table 2: Thermodynamic models Model Van Laar

NRTL

Equations

2



A21 x2  A12 x1 + A21 x2 2 A12 x1 ln γ2 = A21 x  A12 x1 + A21  2 2 G21 τ12 G12 2 ln γ1 = x2 τ21 x1 +x2 G21 + x2 +x1 G12 ln γ1 = A12

G21 = exp −α21 τ21 ln γ2 =

x21

 τ12



G12 x2 +x1 G12

2

+

τ21 G21 x1 +x2 G21

G12 = exp −α12 τ12 Wilson

 τ12 = a12 +

b12 T

+ e12 ln T

α12 = c12 + d12 (T ) b12 ln λ12 = α12 + T α12 = −2.34 α21 = 21.10 b12 = −346.94

ln γ1 = − ln (x1 + x2 λ12 )   λ12 λ21 +x2 x1 +x − x2 +x1 λ21 2 λ12 ln γ2 = − ln (x2 + x1 λ21 )

λ21 − x2 +x 1 λ21     UNIQUAC ln γ1 = − ln xΦ1 + Z2 q1 ln φθ11 +   +φ2 l1 − rr12 l2 − q1 ln (θ1 + θ2 τ21 )+   τ21 τ 12 +θ2 q1 θ1 +θ2 τ21 − θ2 +θ1 τ12 x1 r1 φ1 = x1 r1 + x2 r2 x1 q 1 θ1 = x1 q 1 + x2 q 2 l1 = Z2 (r1 − q1 ) − (r1 − 1)

+x1

Parameters 205.5 A12 = + 0.9017 Tsat 909.1 A21 = − 0.6089 Tsat b12 τ12 = a12 + T α12 = 0.2

λ12 x1 +x2 λ12

9

b21 = −8627.31 b12 τ12 = α12 + T α12 = 2.134 α21 = 1.143 b12 = 826.83 b21 = 244.67 r1 = 3.168 r2 = 0.92 q1 = 2.484 q2 = 1.4 Z=5

ACS Paragon Plus Environment

Ref 20

21

22

22

21

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 33

 γf ur χf ur psat FN2 yf ur f ur VV − Nf ur,V − (6) RT 1 − yf ur − ywater   dNwater,L γwater χwater psat water VV = Fwater + 3 k1 Cxyl Cacid γacid VL − KL a − Nwater,V (7) dt RT   dNwater,V γwater χwater psat FN2 ywater water VV = KL a − Nwater,V − (8) dt RT 1 − yf ur − ywater dNf ur,V = KL a dt



dNwater,C FN2 ywater = dt 1 − yf ur − ywater

(9)

dNf ur,C FN2 yf ur = dt 1 − yf ur − ywater

(10)

1 dND1 = k2 Cxyl Cacid γacid VL dt 3

(11)

dND2 = k3 Cxyl Cacid γacid VL dt

(12)

The system of ordinary dierential equations was solved numerically by ODE15s in MAT-

®.

LAB

Results and discussion Batch operation: experimental performance In Figure 1, yield of furfural at dierent stripping rates is shown. Furfural was the only product detected in the condensate samples. The batch-wise dehydration of xylose without in-situ stripping gives nearly 30% of furfural yield under the explored conditions. Introducing a continuous gas stream at a ow rate ml renders a sharp increase in the furfural yield up to ca. 70%. Rapid removal of of 150 Nmin

furfural from the reaction medium hinders its consequent degradation, thus increasing the nal furfural selectivity. Yet, under these conditions, around 30% of the furfural produced remains in the reaction medium, which indicates that the rate of furfural stripping is still lower than that of its production. Ideally, one would aim at these two processes (i.e., fur-

10

ACS Paragon Plus Environment

Page 11 of 33

100 Yield in condensate Yield total

90 80 70

Yield [%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

60 50 40 30 20 10 0

0

150

300

500

Stripping rate [ml/min]

°

Figure 1: Maximum yield of furfural for dierent stripping ows. Conditions: T=170 C, p=10bar, Cxylin =1wt%, CH2 SO4 =0.5wt% fural production and stripping) to occur at the same rate in order to achieve 100% furfural selectivity. However, while the amount of furfural stripped out of the reactor increases upon increasing the gas ow rate, the total furfural yield remains nearly unaected. Even worse, further increasing the gas ow has a negative eect on the nal furfural yield. This can be attributed to the unavoidable evaporation of water under these reaction conditions and the consequent increase in the concentration of all soluble species (e.g., xylose, sulfuric acid, furfural), which leads to intensive furfural degradation. Hence, excessive stripping has a negative impact on the furfural yield. Figure 2 displays the eect of the nitrogen volumetric ow rate on the experimental and predicted mass balance of xylose and furfural (distributed between reactor and condensate) as a function of time. Accordingly, the temporal evolution of the xylose conversion and furfural yields under those conditions are calculated and reported in Figure 3. Note that the model predictions presented in these gures make use of the best tting kinetics and thermodynamic models, which will be discussed in the following section. As expected, these trends describe the typical prole of reactions in series, which in this case is the conversion of xylose to furfural in the reactor, followed by the transfer of furfural in the reaction solution to the stripping stream, which is eventually recovered in the condenser. In this study, we

11

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research

100

Conversion/Yield [%]

Flow =150 ml/min

80

60

40

Data: Conversion Data: Yield total Data: Yield in condensate Model: Conversion Model: Yield total Model: Yield in condensate

20

0 0

50

100

150

Residence time [min] (a)

100

Conversion/Yield [%]

Flow =300 ml/min

80

60

40

Data: Conversion Data: Yield total Data: Yield in condensate Model: Conversion Model: Yield total Model: Yield in condensate

20

0 0

50

100

150

Residence time [min] (b)

100 Flow =500 ml/min

Conversion/Yield [%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

80

60

40

Data: Conversion Data: Yield total Data: Yield in condensate Model: Conversion Model: Yield total Model: Yield in condensate

20

0 0

50

100

150

Residence time [min] (c)

Figure 2: Experimental data and model predictions of conversion, total yield of furfural and yield of furfural in condensate. Conditions: T=170 C, p=10bar, Cxylin =1wt%, ml CH2 SO4 =0.5wt%; Kinetic model A; Thermodynamic model NRTL ASPEN;(a) FN2 =150 Nmin ml ml (b) FN2 = 300 Nmin (c) FN2 = 500 Nmin 12

°

ACS Paragon Plus Environment

Page 12 of 33

Page 13 of 33

2.0

Flow =150 ml/min

Weight [g]

1.5

Data: Xylose in reactor Data: Furfural in reactor Data: Furfural in condensate Model: Xylose in reactor Model: Furfural in reactor Model: Furfural in condensate

1.0

0.5

0.0

0

50

100

150

Residence time [min] (a)

2.0

Flow =300 ml/min

Weight [g]

1.5

Data: Xylose in reactor Data: Furfural in reactor Data: Furfural in condensate Model: Xylose in reactor Model: Furfural in reactor Model: Furfural in condensate

1.0

0.5

0.0

0

50

100

150

Residence time [min] (b)

2.0

Flow =500 ml/min

1.5

Weight [g]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Data: Xylose in reactor Data: Furfural in reactor Data: Furfural in condensate Model: Xylose in reactor Model: Furfural in reactor Model: Furfural in condensate

1.0

0.5

0.0

0

50

100

150

Residence time [min] (c)

Figure 3: Experimental data and model predictions of the amounts of xylose and furfural in reactor and condensate. Conditions: T=170 C, p=10bar, Cxylin =1wt%, CH2 SO4 =0.5wt%; ml ml Kinetic model A; Thermodynamic model NRTL ASPEN; (a) FN2 =150 Nmin (b) FN2 =300 Nmin ml (c) FN2 = 500 Nmin 13

°

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research

report both the yield of furfural recovered in the condenser, and the total yield including the remaining furfural in the reactor. In line with previous observations, the experimental results plotted in Figure 3 reveal that the amount of furfural stripped out of the reactor increases as the nitrogen ow rate increases. Nevertheless, the benecial eect of the stripping on the nal furfural yield diminishes at very large gas ow rates (i.e., 500

N ml min

in this study), due to

excessive water co-evaporation and a visible increase in the extent of degradation reactions. These results also show that xylose conversion is nearly unaected by the nitrogen ow rate, and thus by the furfural stripping rate. To a certain extent, it is not surprising that the changes in furfural concentration caused by in-situ removal will not aect the rate of xylose consumption. Nevertheless, the largest gas ow rate explored in this study leads to a noticeable increase in the xylose conversion. Signicant water evaporation under such conditions results in an increase of the concentration of all soluble species, including that of sulfuric acid, which will certainly aect the rate of xylose conversion to furfural. Furthermore, the direct consumption of xylose in the degradation reactions with other soluble humin precursors, might be accelerated under such conditions. 3.0

Furfural in condensate [wt%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 33

Data: 150ml/min Data: 300ml/min Data: 500ml/min Model: 150ml/min Model: 300ml/min Model: 500ml/min

2.5

2.0

1.5

1.0

0.5

0.0 0

20

40

60

80

100

120

140

160

Residence time [min]

Figure 4: Experimental data and model predictions for weight percentage of furfural in condensate. Conditions: T=170 C, p=10bar, Cxylin =1wt%, CH2 SO4 =0.5wt%; Kinetic model A; Thermodynamic model NRTL ASPEN

°

In view of process protability, besides increasing the total product yield, achieving high furfural concentrations in the condenser is also desired. Figure 4 shows the experimental and 14

ACS Paragon Plus Environment

Page 15 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

predicted concentration of furfural in the condensate as a function of residence time and gas ow rate. Regardless of the gas ow rate, the temporal evolution of furfural concentration follows a volcano curve, with a maximum that shifts to lower residence times for the larger gas ow rates. The concentration of furfural in the condensate is determined by the furfuralwater stripped out of the reaction mixture. On the one hand, the molar ow of water evaporation for a given inlet gas ow rate remains constant throughout the process, assuming thermodynamic equilibrium and a negligible change in the concentration of water in the reactor. Logically, water evaporation increases with the nitrogen ow rate. On the other hand, the extent of furfural evaporation depends on its concentration in the reaction mixture, which is a function of the residence time. At the start of the reaction, the total stripping rate exceeds that of the furfural production, which results in a large water-to-furfural ratio in the outlet gas stream. As the reaction proceeds, the furfural production peaks and furfural is stripped out in high concentrations. Note that the peak in furfural production rate is achieved at earlier residence times when operating under fast stripping rates. This is attributed to an increase in acid concentration under these conditions. At high xylose conversion, furfural production rates begin to fade while the rate of water stripping remains constant. Accordingly, the processes can be further optimized by operating the reactor at maximum furfural concentration. For example, a well-mixed continuous system may be suitable for this purpose. Despite the undesired co-evaporation of water, the thermodynamic equilibrium of furfuralwater mixtures favors furfural evaporation over that of water (Figure 5). The model predictions of Vapor Liquid Equilibrium (VLE) shown in Figure 5 are in good agreement with the experimental data found in the literature. 23 Note that the concentration of furfural in the condenser reaches higher values than the initial xylose concentration (i.e., 1 wt%). Thus, besides preventing furfural degradation by in-situ product removal, this process conguration manages to reduce the energy demand in the downstream processing of furfural, which is produced, isolated and concentrated in one single step. Nevertheless, operating at very high 15

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research

gas ow rates leads to unnecessary large amounts of water evaporation (i.e., to low furfural concentrations in the condensate, as shown in Figure 4), and to limited or no benets in terms of furfural yield. Thus, the optimum ow rate of the stripping gas can be determined, for example, from an economic evaluation that takes into account the gains in product yield, and the losses in water evaporation and product dilution. Process optimization may be attained by tuning the vapor-liquid equilibrium of this system. For example, the addition of salts like sodium sulfate can minimize the evaporation of water by increasing its boiling point. In light of these ndings, low temperature operations may also become attractive, although an obvious sacrice in terms of reaction rate is envisioned. 260

Vapor composition: model Liquid composition: model

250

Boiling point [° C]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 33

240

170

230

169 168

220

167

210 166

200

165 0

5

10

15

20

25

30

190 180 170 160

0

10

20

30

40

50

60

70

80

90

100

Furfural composition [wt%]

Figure 5: Water Furfural VLE predictions from NRTL ASPEN model at 7.7 bar

Batch operation: model predictions Figures 2 and 3 show that the model is able to accurately predict the experimental data obtained at lower stripping rates when using NRTL and kinetic model A to describe the thermodynamic equilibrium and the reaction mechanism, respectively. However, at higher 16

ACS Paragon Plus Environment

Page 17 of 33

stripping rates (i.e., 500

N ml ), min

this model overestimates the amount of xylose in the reactor

and that of furfural in the condensate at longer residence times. In other words, the extent of degradation reactions involving the consumption of furfural and likely xylose, is greater than the model predicts under these experimental conditions. Note that the eect of water evaporation on the concentration of xylose and sulfuric acid, and the impact of these on the xylose consumption rate are accounted for in this model. 5

H2SO 4 in reactor [wt%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Data: 150 ml/min Data: 300 ml/min Data: 500 ml/min Model: 150 ml/min Model: 300 ml/min Model: 500 ml/min

4

3

2

1

0 0

50

100

150

Residence time [min]

Figure 6: Experimental data and model predictions of sulfuric acid concentration. Conditions: T=170 C, p=10bar, Cxylin =1wt%, CH2 SO4 =0.5wt%; Kinetic model A; Thermodynamic model NRTL ASPEN

°

Figure 4 also indicates that the model successfully predicts the concentration of furfural in the condensate at lower stripping rates, strongly suggesting that indeed the NRTL thermodynamic model is suitable to describe the system under consideration. Thus, the discrepancies between the experiments and the model predictions at large gas ow rates are attributed to an oversimplication of the degradation reactions in the kinetic model A. In particular, the participation in degradation reactions of furfural, and perhaps xylose and other soluble species (i.e., intermediates or soluble degradation products) under increasingly acidic conditions is insuciently described. In addition, the reversible oligomerization of xylose under acid conditions is plausible, and it may potentially explain the observed increase in xylose conversion at high stripping rates. Figure 6 shows the increase of sulfuric acid concentration for each of the stripping rates. Further investigation on the reaction mechanism is needed 17

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 33

for an accurate description of the system at high acid concentration. Figure 7 shows the model predictions of the experimental data under 150 N ml min

(b) and 500

N ml min

N ml min

(a), 300

(c) gas ow rates using several thermodynamic models (i.e., NRTL,

Van Laar, Wilson and UNIQUAC). Further, two dierent values for the NRTL parameters were considered ( 21 and 22 ). Clearly, the NRTL model using parameters from ASPEN plus database is the most suitable thermodynamic model to describe this system. Van Laar model also describes reasonably well the thermodynamic equilibrium of the reaction mixture, while Wilson and UNIQUAC are not suitable models for this purpose. Figure 8 illustrates the model prediction of the experimental data considering three possible kinetic mechanisms reported in literature (see Table 1). The main dierence between these models concern the possible routes toward degradation products. While all of them consider the direct degradation of furfural, they also oer additional degradation routes from xylose (mechanism A and C), xylose with furfural (mechanism B), and intermediates (mechanism C). When considering the experimental data obtained at 150 and 300

N ml min

gas ow

rates, our results seem to follow the trends predicted by kinetic mechanism A. Thus, we may conclude that in addition to the main dehydration path towards furfural, xylose converts through (at least) one parallel pathway that does not render furfural, but likely degradation products. Notice that, among all possible products, only furfural has been experimentally quantied in this study. As previously discussed, the extent of degradation reactions at higher acid concentrations (present in the reaction medium at very fast stripping rates) seem to be poorly described by this mechanism. In that case, both xylose and furfural seem to react away faster than model A predicts. In this line, the kinetic model proposed by Weingarten 17 (i.e., kinetic model B) considers the reaction between xylose and furfural together to form degradation products. Nevertheless, as shown in Figure 8, this model predicts a lower xylose conversion and a higher furfural selectivity that those observed in the experiments. This is in agreement with the ndings of Weingarten and co-authors, who reported discrepancies

°

between model and experiments at temperatures higher than 160 C. Notice that this study 18

ACS Paragon Plus Environment

Page 19 of 33

2.0

Data: Furfural in reactor Data: Furfural in condensate Model: Furfural in reactor Model: Furfural in condesate NRTL ASPEN NRTL Fele Van Laar Wilson UNIQUAC

Flow =150 ml/min 1.8 1.6

Weight [g]

1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0

0

20

40

60

80

100

120

140

160

Residence time [min]

(a)

2.0

Data: Furfural in reactor Data: Furfural in condensate Model: Furfural in reactor Model: Furfural in condesate NRTL ASPEN NRTL Fele Van Laar Wilson UNIQUAC

Flow =300 ml/min 1.8 1.6

Weight [g]

1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0

0

20

40

60

80

100

120

140

160

Residence time [min]

(b)

2.0

Data: Furfural in reactor Data: Furfural in condensate Model: Furfural in reactor Model: Furfural in condesate NRTL ASPEN NRTL Fele Van Laar Wilson UNIQUAC

Flow =500 ml/min 1.8 1.6 1.4

Weight [g]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

1.2 1.0 0.8 0.6 0.4 0.2 0.0

0

20

40

60

80

100

120

140

160

Residence time [min]

(c)

Figure 7: Experimental data and model predictions of the amounts of furfural in reactor and condensate for various thermodynamic models. Conditions: T=170 C, p=10bar, ml ml Cxylin =1wt%, CH2 SO4 =0.5wt%; Kinetic model A;(a) FN2 =150 Nmin (b) FN2 =300 Nmin (c) N ml FN2 =500 min 19

°

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research

100

Conversion/Yield [%]

Flow =150 ml/min 80

60 data Model [A] Model [B] Model [C] Conversion Yield total Yield in condensate

40

20

0 0

50

100

150

Residence time [min] (a)

100

Conversion/Yield [%]

Flow =300 ml/min 80

60 data Model [A] Model [B] Model [C] Conversion Yield total Yield in condensate

40

20

0 0

50

100

150

Residence time [min] (b)

100 Flow =500 ml/min

Conversion/Yield [%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 33

80

60 data Model [A] Model [B] Model [C] Conversion Yield total Yield in condensate

40

20

0 0

50

100

150

Residence time [min] (c)

Figure 8: Experimental data and model predictions of the xylose conversion and furfural yield in reactor and condensate for various kinetic models. Conditions: T=170 C, p=10bar, ml Cxylin =1wt%, CH2 SO4 =0.5wt%; Thermodynamic model NRTL ASPEN ;(a) FN2 = 150 Nmin ml ml (b) FN2 = 300 Nmin (c) FN2 = 500 Nmin 20

°

ACS Paragon Plus Environment

Page 21 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

°

is conducted at 170 C. The authors proposed an additional reaction term which would only become important at elevated temperatures (i.e., with a large activation energy), although no kinetic tting was reported for this specic reaction. Finally, the Ershova 18 proposed that not only xylose and furfural are subject to independent degradation reactions, but also a reaction intermediate (i.e., xylulose) converts into degradation products (i.e., kinetic model C). This model fails to predict the experimental trends recorded in the present study. More precisely, while the consumption of xylose is described accurately by the model, the formation of furfural is underpredicted. In fact, the presence of xylulose in the reaction mixture was not conrmed in the current study. These ndings suggest that indeed the formation of xylulose and its further degradation is not a signicant reaction path under these experimental conditions. This is in line with the current understanding that the isomerization of xylose to xylulose is not signicant in the presence

°

of Brönsted acids at around 180-220 C. 18 24

Continuous operation: model predictions & experimental validation Figure 9a shows the observed and predicted values of xylose conversion and furfural yield as a function of time on stream for a continuous operation. Unlike in the previous sections, we now incorporate a continuous liquid feed of xylose in water. The concentration and ow of this stream is adjusted to compensate for water evaporation and to match reaction rate of xylose at those conditions. As observed in Figure 9a, the model successfully predicts the experimental results, even with the largest ow rate of stripping gas. This proves that indeed the excessive evaporation of water and the consequent increase in the concentration of soluble species caused the discrepancies between model and experiments during the batch operation under high stripping rates. As observed in Figure 9a, the system did not reach steady state conditions during the initial 150 min tested in the experiments. Since there is no liquid outlet, it should be noted that the only possible steady state conversion is 100%. Nevertheless, with the experimental 21

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

conditions tested in this study (e.g., inlet ow rate and concentrations), such degree of conversion would only be reached with a reaction volume greater that that of the reactor used herein. Thus, it was essentially impossible to reach steady state conditions under the experimental conditions under evaluation. Prior to the steady state conditions, the unreacted xylose and non-evaporated water accumulate in the reactor, leading to an obvious volume increase, as shown in Figure 9b. Notably, the model was able to successfully predict the overall mass balance of the reactor and condensate in the transient state, and is therefore considered reliable to predict the nal steady state.

°

Figure 10 shows the model predictions at 170 C for dierent nitrogen ows. Under these reaction conditions, it takes approximately 100, 200 and 400 minutes to reach steady state ml nitrogen ow rate, respectively (see Figure 10b). As shown in Figure for 150, 300 and 500 Nmin

10a, the steady state conversion of xylose reaches indeed 100% in all cases, whereas the yield of furfural is around 70%, with a slight positive eect of stripping rate on nal yield. Here, we can observe that most of the furfural is collected in the condensate, particularly when the stripping rate increases, although an important dilution of the collected furfural is observed for the highest furfural stripping rate. An important technical challenge of the continuous system operated as a fed-batch (i.e., without a liquid outlet) is the long-term accumulation of degradation products (i.e., humins) inside the reactor. A possible solutions would be to introduce a liquid outlet and recirculate the stream after ltration of the reaction mixture externally. Table 3 presents the dierences in furfural production rate, furfural concentration in the condensate and furfural yield for continuous and batch mode of operation. For the batch system, the residence time considered to calculate the production rate was the moment when 99% of the furfural was in the condensate. A remarkable 46 fold increase in the furfural production rate is obtained in the continuous reactor with respect to the batch system. This result may sound counter-intuitive as it is well-known that a CSTR (i.e., similar to the continuous reactor in this study) operates at lower reactant concentrations, and therefore 22

ACS Paragon Plus Environment

Page 22 of 33

Page 23 of 33

100 Flow =500 ml/min

Conversion/Yield [%]

80

60

40

Data: Conversion Data: Yield total Data: Yield in condensate Model: Conversion Model: Yield total Model: Yield in condensate

20

0 0

50

100

150

200

250

300

Time [min] (a)

400 350

Mass [g]

300 250 200 150 Data: Mass in reactor Data: Mass in condensate Model: Mass in reactor Model: Mass in condensate

100 50 0

0

50

100

150

200

250

300

Time [min] (b)

5.0

Furfural in condensate [wt%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

4.0

3.0

2.0

1.0 Data: Furfural wt% in condensate Model: Furfural wt% in condensate 0.0 0

100

200

300

400

500

600

Time [min] (c)

Figure 9: (a) Experimental and model data of conversion and yield for continuous system; (b)Experimental and model data for mass in the reactor and mass of condensate (c)Weight percentage of furfural in condensate. Conditions: T=168 C, p=10bar, Cxylin = 1wt%, ml CH2 SO4 = 0.5wt% ;FH2 O = 1.3 · 10−3 mol ;Fxyl 23 = 8.14 · 10−6 mol ;FN2 = 500 Nmin ; Kinetic model s s Paragon Plus Environment A; Thermodynamic model NRTLACS ASPEN.

°

Industrial & Engineering Chemistry Research

Conversion/Yield [%]

100

80

60

40

N2 Flow=150 ml/min N2 Flow=300 ml/min N2 Flow=500 ml/min Model: Conversion Model: Yield total Model: Yield in condensate

20

0 0

500

1000

1500

2000

2500

3000

Time [min] (a)

Weight percent [%]

2.0 N2 flow=150 ml/min N2 flow=300 ml/min N2 flow=500 ml/min Model: Xylose in reactor Model: Furfural in reactor

1.5

1.0

0.5

0.0

0

100

200

300

400

500

Time [min] (b)

6.0

Furfural in condensate [wt%]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

5.0 4.0 3.0 2.0 N2 flow=150 ml/min N2 flow=300 ml/min N2 flow=500 ml/min

1.0 0.0 0

500

1000

1500

2000

2500

3000

Time [min] (c)

Figure 10: (a) Model predictions of conversion and yield for continuous system (b)Model predictions of xylose and furfural concentrations in the reactor (c)Weight percentage of furfural in condensate. Conditions: T=170 C, p=10bar, Cxylin = 1wt%, CH2 SO4 = 0.5wt% ml ;FH2 O = 0.4 · 10−3 ; 0.81 · 10−3 ; 1.35 · 10−3 mol ;F = 6.47 · 10−6 mol ;FN2 = 100; 300; 500 Nmin ; s 24xyl s ACS Paragon Plus Environment Kinetic model A; Thermodynamic model NRTL ASPEN

°

Page 24 of 33

Page 25 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

lower reaction rates, than a batch reactor operated under similar conditions (e.g., same starting concentrations and residence time). In this case, however, the continuous reactor is in fact a fed-batch reactor (i.e., with no liquid outlet) where the concentration of xylose is maintained at initial conditions (i.e., 1wt%) by constantly feeding a xylose solution with the exact concentration to balance the rates of xylose conversion and that of water evaporation. On the other hand, the concentration of xylose in the batch reactor decreases over time, thus leading to an overall lower reaction rate. As shown in Table 3, it is possible to increase the Table 3: Comparison of batch and continuous operation  ml  g M ode N2 f low Nmin YC [%] P roduction rate hr FC [wt%] Batch 150 68.19 0.258 0.527 300 72.24 0.364 0.686 500 73.45(56.64a ) 0.524(0.423a ) 0.966(0.760a ) Continuous 150 68.78 1.537 5.366 300 72.64 1.625 2.950 500 74.30 1.663 1.844 a

Experimental values are displayed in brackets for reference. Notice that in the specic case of large stripping rates, the model predictions dier signicantly from the experimental values obtained during batch operation. furfural productivity and the furfural concentration in the condensate when the reactor is operated continuously with respect to the reference batch process. Increasing the nitrogen ow rate in the continuous system leads to a slightly better yield but at the expense of a more diluted product in condensate.

Comparison of the industrial and the proposed process A comparison between the industrial process (i.e., using steam stripping) and the process under investigation (i.e., using non-condensable gases like H2 as stripping agent) on the basis of the net energy consumption is shown in Table 4. The benchmark industrial process is based on the batch HUAXEI plant. 12 The latter is described by the model predictions of the batch operation under optimized conditions. An important dierence between these

25

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 26 of 33

two examples is the feed used. While the industrial process uses biomass (e.g., corn cob or bagasse) as a starting material, our results are obtained using xylose as model compound. One may argue that, with respect to the conversion of xylose, processing biomass would lead to additional degradation reactions and thus lower furfural yield. However, as shown by Agirrezabal-Telleira et. al, 13 processing corncobs and xylose actually yield similar amounts of furfural when using nitrogen as stripping agent. This may be well understood by the fact that the most critical part of the conversion of biomass to furfural, both in terms of reaction rate and selectivity, concerns the dehydration of sugars, while the initial hydrolysis step is faster and does not present major selectivity issues. 25 Thus, for the sake of this study, we trust that this is a valid comparison and that the conclusions derived from it with respect to energy savings will remain applicable when using more complex feedstocks like hemicellulose or biomass. In our analysis, two specic scenarios have been taken into account: 1) a scenario in which the batch reactor is optimized to reach a yield of 50% of furfural, similar to the benchmark case; and 2) a scenario in which the furfural yield is maximized. In all cases, the xylose concentration equals to that in the HUAXEI plant (i.e., feed of 12wt% corn cobs with 35% pentose content 26 ). Table 4: Comparison of the industrial and the proposed process

Yield in condensate [%] Xylose feed [wt%] Steam feed [ton/ton fur] H2 feed [ton/ton fur] Energy for feed steam generation [GJ/ton fur] Energy for preheating H2 feed [GJ/ton fur] Energy for in-situ steam generation [GJ/ton fur] Energy for preheating reaction mixture [GJ/ton fur] SUM [GJ/ton fur]

HUAXEI plant 50 4.2 25-35 n/a 82.7-115.8 n/a n/a 40.9 123.6-156.7

H2 Stripping Scenario 1 Scenario 2 50 72.2 4.2 4.2 n/a n/a 0.68 0.97 n/a n/a 1.4 2.0 65.7 93.9 40.9 40.9 108.1 136.8

As shown in Table 4, the use of non-condensable gases leads to an important reduction (ca.30%) in the energy usage when the process is optimized to reach 50% furfural yield. Greater energy savings may be achieved if the hot furfural/H2 outlet stream is directly fed 26

ACS Paragon Plus Environment

Page 27 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

to the downstream hydrogenation unit (e.g., for cyclopentatone synthesis), and later on, the remaining unreacted H2 is recirculated back to the dehydration reactor. This strategy would eliminate the need to H2 pre-heating before the dehydration unit, and would further reduce the energy requirements of the hydrogenation unit. The comparison shown in Table 4 also reveals that further increasing the furfural yields from 50 to over 72 % (i.e., by extending the residence time and therefore increasing the amount of both furfural and water stripping) has an energy penalty due to excessive water evaporation. Note that in Scenario 2 the energy costs are 26% higher since at this conditions the rate of water evaporation exceeds that of water. Here, process optimization may be accomplished by trading-o the feed and energy costs. Inexpensive biomass is obviously an attractive feedstock that can make this process feasible even when a large part of the feed (e.g., cellulose and lignin) is miss-utilized and furfural yields are relatively low with respect to the theoretical maximum, as long as the energy demands are maintained low. Although the exact threshold will depend on energy prices, we envision that the energy savings achieved by the proposed process, particularly in Scenario 1, will be instrumental in the valorization of low cost biomass feeds. On the other hand, processing more expensive hemicellulose or even sugar-rich hydrolyzate streams will become of increased relevance in the context of future bio-reneries, where raw biomass is rst fractionated prior to valorization of the individual fractions to valuable chemicals. Since hemicellulose hydrolysis and sugar dehydration take place under similar conditions, decoupling the processes in dierent steps does not make a signicant dierence in terms of energy demands. 5,6 However, the pressure on product yield will become more important to compensate for the additional biomass processing costs. Thus, the proposed process, particularly in Scenario 2, is expected to play an important role when processing more costly feedstocks. Here, additional revenues from the valorization of other stream (e.g., cellulose and lignin) should be also taken into account.

27

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Conclusions In this paper, we demonstrate by experiments and rst-principle modeling that using noncondensable gases instead of steam as a stripping agent results in important energy savings in the furfural production process. We further suggest that using hydrogen can lead to additional energy integration routes, and thus greater energy savings, if furfural is later on hydrogenated to added value chemicals. Using batch experiments with continuous nitrogen stripping we show that there is an optimal stripping ow rate that results in a maximum furfural yield of around 72%. Larger stripping ow rates render excessive water evaporation, leading to increased furfural degradation rates. Further, the continuous synthesis of furfural was demonstrated experimentally by coupling the batch process with a continuous feed of xylose in water. Simultaneously with the experimental study, a rst-principle reactor model was developed using various thermodynamic and kinetic models available in the literature. This model successfully predicted the experimental observations of the batch and continuous operations in a large range of experimental conditions without the use of any tting parameters. The kinetics by Marcotullio et al. 16 and the NRTL thermodynamic model (using Aspen database) gave the most accurate prediction of the experiments results. Yet, signicant discrepancies between the model predictions and the experiments were observed when using ml ) in batch conditions. This was attributed to the the highest stripping ow rate (i.e., 500 Nmin

limitations of the kinetic model with respect to side and degradation reactions under highly acidic conditions resulting from the excessive evaporation of water under such conditions. Hence, this work highlighted the need to further investigate the kinetic mechanism of this reaction.

Acknowledgment This work was performed under the framework of Chemelot InSciTe and is supported by contributions from the European Regional Development Fund (ERDF) within the framework 28

ACS Paragon Plus Environment

Page 28 of 33

Page 29 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

of OP-Zuid and with contributions from the province of Brabant and Limburg and the Dutch Ministry of Economy.

Nomenclature Latin symbols

Greek symbols α

Non-randomness factor in NRTL model

γ

Activity coecient

λ

Wilson model parameter

F

Pre-exponential factor [s−1 ]   Concentration mol L  J  Activation energy mol   Flow mol s

τ

Parameter for interaction energy in NRTL model

G

NRTL parameter

χ

Molar fraction in the liquid phase [mol%]

k

Kinetic constant [s−1 ]

A C Ea

KL a

Subscripts

Overall mass transfer [s−1 ]

acid

Sulfuric acid

N

Moles [mol]

C

Condensate

p

Pressure [P a]

D1

Degradation product 1

D2

Degradation product 2

R

Saturation pressure [P a]  J  Gas constant Kmol

f ur

Furfural

T

Temperature [K]

L

Liquid

Saturation temperature [K]

N2

Nitrogen

t

Time [s]

V

Vapor

V

Volume [L]

xyl

Xylose

X

Conversion [%]

Y

Yield [%]

y

Molar fraction vapor [mol%]

psat

Tsat

References (1) National Research Council, Biobased Industrial Products: Research and Commercial-

ization Priorities ; National Academies Press, 2000. 29

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

(2) de Jong, E.; Higson, A.; Walsh, P.; Wellisch, M. Bio-based Chemicals Value Added Products from Bioreneries. Report prepared for IEA Bioenergy  Task 42 Biorenery. International Energy Technology IEA, 2012. (3) Philp, J. C.; Ritchie, R. J.; Allan, J. E. Biobased Chemicals: The Convergence of Green Chemistry with Industrial Biotechnology. Trends in Biotechnology 2013, 31, 219222. (4) Top Value Added Chemicals from Biomass Volume IResults of Screening for Potential Candidates from Sugars and Synthesis Gas. Produced by the Sta at Pacic Northwest National Laboratory (PNNL) National Renewable Energy Laboratory (NREL). Oce of Biomass Program (EERE) For the Oce of the Biomass Program Werpy, T; Petersen, G. Eds. U.S. Department of Energy (DOE), 2004. (5) Luterbacher, J.; Alonso, D. M.; Dumesic, J. Targeted Chemical Upgrading of Lignocellulosic Biomass to Platform Molecules. Green Chemistry 2014, 16, 48164838. (6) Mamman, A. S.; Lee, J.-M.; Kim, Y.-C.; Hwang, I. T.; Park, N.-J.; Hwang, Y. K.; Chang, J.-S.; Hwang, J.-S. Furfural: Hemicellulose/Xylose-derived Biochemical. Biofu-

els, Bioproducts and Biorening 2008, 2, 438454. (7) Hoydonckx, H.; Van Rhijn, W.; Van Rhijn, W.; De Vos, D.; Jacobs, P. Furfural and Derivatives. In Ullmann's Encyclopedia of Industrial Chemistry ; Elvers, B., Ed.; Wiley Online Library, 2007; pp 285-296. (8) Lange, J.-P.; Van Der Heide, E.; van Buijtenen, J.; Price, R. FurfuralA Promising Platform for Lignocellulosic Biofuels. ChemSusChem 2012, 5, 150166. (9) Hronec, M.; Fulajtarová, K. Selective Transformation of Furfural to Cyclopentanone.

Catalysis Communications 2012, 24, 100104. (10) Zeitsch, K. J. The Chemistry and Technology of Furfural and its many By-products ; Elsevier, 2000; Vol. 13. 30

ACS Paragon Plus Environment

Page 30 of 33

Page 31 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

(11) De Jong, W.; Marcotullio, G. Overview of Bioreneries Based on Co-production of Furfural, Existing Concepts and Novel Developments. International journal of chemical

reactor engineering 2010, 8 . (12) Win, D. T. Furfural-Gold from Garbage. AU J. Technol. 2005, 8, 185190. (13) Agirrezabal-Telleria, I.; Gandarias, I.; Arias, P. Production of Furfural from Pentosanrich Biomass: Analysis of Process Parameters during Simultaneous Furfural Stripping.

Bioresource technology 2013, 143, 258264. (14) Agirrezabal-Telleria, I.; Larreategui, A.; Requies, J.; Güemez, M.; Arias, P. Furfural Production from Xylose using Sulfonic Ion-exchange Resins (Amberlyst) and Simultaneous Stripping with Nitrogen. Bioresource technology 2011, 102, 74787485. (15) Agirrezabal-Telleria, I.; Requies, J.; Güemez, M.; Arias, P. Furfural Production from Xylose + Glucose Feedings and Simultaneous N2 -stripping. Green Chemistry 2012, 14, 31323140. (16) Marcotullio, G. The Chemistry and Technology of Furfural Production in Modern Lignocellulose-Feedstock Bioreneries. Ph.D. thesis, TU Delft Faculty of Mechanical, Maritime and Materials Engineering, Delft, Netherlands, 2011. (17) Weingarten, R.; Cho, J.; Conner Jr, W. C.; Huber, G. W. Kinetics of Furfural Production by Dehydration of Xylose in a Biphasic Reactor with Microwave Heating. Green

Chemistry 2010, 12, 14231429. (18) Ershova, O.; Kanervo, J.; Hellsten, S.; Sixta, H. The Role of Xylulose as an Intermediate in Xylose Conversion to Furfural: Insights via Experiments and Kinetic Modelling. RSC

Advances 2015, 5, 6672766737. (19) Dickson, A. G.; Wesolowski, D. J.; Palmer, D. A.; Mesmer, R. E. Dissociation Constant

31

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 32 of 33

°

of Bisulfate Ion in Aqueous Sodium Chloride Solutions to 250 C. Journal of Physical

Chemistry 1990, 94, 79787985. (20) Smuk, J.; Harris, J. F. Engineering Calculations for the Distillation of the Furfural-

Water System ; Madison, Wis. : U.S. Dept. of Agriculture, Forest Service, Forest Products Laboratory, 1959. (21) Fele, L.; Grilc, V. Separation of Furfural from Ternary Mixtures. Journal of Chemical

& Engineering Data 2003, 48, 564570. (22) Aspen Plus User Guide. version 10.2, Aspen Technology, Inc. (23) Curtis, R.; Hatt, H. Equilibria in Furfural-Water Systems under Increased Pressure and the Inuence of Added Salts upon the Mutual Solubilities of Furfural and Water. Aust.

J. Sci. Res. 1948, 1, 213235. (24) Choudhary, V.; Sandler, S. I.; Vlachos, D. G. Conversion of Xylose to Furfural using Lewis and Brønsted Acid Catalysts in Aqueous Media. ACS Catalysis 2012, 2, 2022 2028. (25) ResearchGate

publication

page.

https://www.researchgate.net/publication/

228501806_CHEMICALS_AND_BIOFUELS_FROM_HARDWOODS_FUEL_CROPS_AND_ AGRICULTURAL_WASTES, (accessed February 2019). (26) Yang, S.-T.; El-Ensashy, H.; Thongchul, N. Bioprocessing Technologies in Biorenery

for Sustainable Production of Fuels, Chemicals, and Polymers ; John Wiley & Sons, 2013.

32

ACS Paragon Plus Environment

Page 33 of 33 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

For Table of Contents Only

33

ACS Paragon Plus Environment