Contents lists available atScienceDirect
Applied Energy
journal homepage:www.elsevier.com/locate/apenergy
Fast pyrolysis of lignins with different molecular weight: Experiments and
modelling
P.S. Marathe, R.J.M. Westerhof, S.R.A. Kersten
⁎Sustainable Process Technology (SPT), Department of Science and Technology (TNW), University Twente, 7522 NB Enschede, the Netherlands
H I G H L I G H T S
•
Pyrolysis of 14 lignins with molecular weight between 588 Da and 3596 Da.•
A population balance model describing lignin fast pyrolysis.•
Limited influence of temperature on lumped product yield.•
Molecular weights of the oils are function of pressure. A R T I C L E I N F O Keywords: Pyrolysis Lignin Molecular weight Model Mass transport A B S T R A C TLignins with number average molecular weights between 350 Da and 1900 Da were characterised and subse-quently pyrolysed in a screen-heater at pressures of 500 Pa and 105Pa between 425 °C and 793 °C. Upwards of
530 °C, the temperature turned out to have only a minor influence on the yields and composition of the oils produced. Clear trends were observed as a function of the molecular weight and pressure – (1) at increasing molecular weight of the lignin, the oil yield decreases while yields of char and gas increase, (2) the molecular weight of the oil is lower for oils produced at 105Pa as compared to the ones obtained at 500 Pa, (3) above a
certain molecular weight of the lignins, ∼400 Da for 105Pa and ∼800 Da for 500 Pa, the molecular weight of
the oil becomes independent of the molecular weight of the lignin. A mathematical model has been developed, which includes three concurrently occurring processes, viz. cracking and polymerisation reactions and removal, hence mass transport, of unconverted lignin and reaction products from the reaction zone. This model can describe all the trends observed experimentally and provides, after parametrisation, reasonable qualitative predictions of the yield and molecular weight of the oils. Knowledge of the role of the interplay between mass transport and chemistry in the pyrolysis process is further accumulating, and from this the development of lignin valorisation can avail. For instance, it has become clear that in the pyrolysis process the molecular weight of lignin oil, which is an important characteristic for the upgrading of the oil to chemicals and/or fuels, can be steered with the pressure.
1. Introduction
Lignin is an amorphous, highly cross-linked, hetero-polymer that provides structural integrity and stiffness to plant cell walls[1]. It is the most abundant naturally occurring aromatic polymer and it accounts for ∼30% of all organic carbon in the bio-sphere[2]. Typically, it is burned as a low-value fuel, however, being aromatic in nature, it could be a potential source of mono-aromatics and other value-added che-micals[3–5].
Lignin is isolated from biomass via several methods such as kraft, organosolv, ball milling, and cold-water precipitation of pyrolytic lignin
[6–8]. The extent to which the structure of native lignin is preserved is controlled by the severity of the isolation method[6,9,10]. Milled wood lignin (MWL), obtained by Björkman’s method, is considered to re-present native lignin most accurately[11].
Starting in the 1970s, the fast pyrolysis process has been studied extensively to liquefy and depolymerise lignin in various experimental devices[12–19]. Aligning the results obtained in the different reactor systems used is difficult. For example, with respect to the production of mono-aromatics yields of up to 20% were reported in micro-pyrolysers [20,21], whereas in screen-heaters and fluidised beds typical yields were only 1 to 4%[16,22,23]. Oil yields as high as 80% were obtained
https://doi.org/10.1016/j.apenergy.2018.12.058
Received 21 September 2018; Received in revised form 10 December 2018; Accepted 11 December 2018
⁎Corresponding author.
E-mail address:s.r.a.kersten@utwente.nl(S.R.A. Kersten).
Applied Energy 236 (2019) 1125–1137
0306-2619/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
in screen-heaters under vacuum conditions[22,24]while in the other reactors the maximal yield was limited to 40%[20,23,25,26]. There is also no consensus on the chemical mechanisms leading to monomers and oligomers. Patwardhan et al. claimed that monomeric compounds are the primary products of lignin pyrolysis and that they polymerise during condensation of the vapours (to oil) to form oligomers [20]. Contrary, Piskorz et al.[27]postulated that oligomers eject from pyr-olysing biomass particles, which was later verified experimentally by Small-Angle Neutron Scattering (SANS) analysis by Fratini et al. and by Montoya et al. for lignin, and by Teixeira et al. for cellulose[28–30].
Models for lignin pyrolysis range from weight loss kinetic models [31–35], lumped component class models [36–38], via population balance calculations over the molecular weight distribution of the re-acting lignin[39,40], to models combining a structural description of lignin combined with the reactivity of linkages between the lignin building blocks[41,42]. Weight loss kinetics models are always based on low heating rate data and use experimentally observed yields of volatile species to estimate kinetic parameters to simulate slow pyr-olysis of lignin. These models can predict the non-isothermal weight loss of the reacting lignin and yields of volatile species (e.g. gases, methanol). However, they cannot predict the oil yield nor the mole-cular weight distribution of the oil, which limits their applicability. The lumped class models of Faravelli et al. [36], Hough et al.[37], and Dussan et al. [38] use experimentally obtained and/or theoretically (group contribution method and/or density functional theory) calcu-lated kinetic parameters to predict the yields of oil (light and heavy), char, and gas as well as weight loss kinetics. The experimentally ob-tained kinetic parameters were derived under slow pyrolysis conditions. These models use complex reaction networks, composed of more than a hundred series/parallel primary and/or secondary reactions. As input, the lumped class models use a simplified lignin structure, while the
models of Klein et al. and Yanez et al. use the spruce lignin structure proposed by Freudenberg[41,43]and an artificially generated struc-ture of wheat straw lignin[42,44]. In the latter, probabilities of reac-tions pathways are assigned based on experimental observareac-tions as well as theoretical calculations. The model of Klein et al. uses empirically obtained kinetic parameters[41], and their values are used as an initial estimate in the model of Yanez et al.[42]. These models predict the temporal evolution of the yields of major product fractions and the molecular weight of the liquid and gaseous products. However, they do not capture the influence of mass transfer. The population balance model of Solomon et al.[40], originally developed for coal pyrolysis, assumes the cracking of weak bonds and transport of volatile products away from the reacting lignin. Interestingly, their model captures the effect of the external pressure on the product yields. Solomon et al. have described the transport of species by using Fick’s law of diffusion, where the diffusion coefficient is a function of pressure, temperature and molecular weight[39,40].
This paper aims at advancing the understanding of lignin pyrolysis in order to increase the valorisation of this abundant natural aromatic resource. We assume that, like for cellulose[45], the pyrolysis of lignin is controlled by the interplay of chemical reactions taking place in the liquid/solid state and the removal rate of compounds from the reaction zone. If this hypothesis is true, then the product yields and molecular weight distribution of the produced oil depend on the molecular weight (distribution) of the lignin feedstock and the bond type interconnecting the aromatic units. For lignin pyrolysis, the dominant reactions with respect to molecular weight distribution of the oil produced can be generalised into cracking and polymerisation reactions. To investigate this hypothesis, 14 lignins with a mass average molecular weight ran-ging from 588 to 3596 Da were characterised. Secondly, the lignins were pyrolysed in a screen-heater designed to minimise
non-Nomenclature
A,B parameters used to describe the transport rate ofDPi
CiL molar concentrations ofDPiin reacting lignin at t t= ,molem
L 3
Ci feedL, molar concentrations ofDPiin lignin feedstock,molem
L 3
Ci tL,=0 molar concentrations ofDPiin lignin att=0,molem
L 3 CiO molar concentrations ofDP iin oil at t t,= molem L 3
Ci tO,=0 molar concentrations ofDPiin oil att=0,molem
L 3
Ci tO, = molar concentrations ofDPiin oil at t= ,molem
L 3
Ci tL, = molar concentrations ofDPiin lignin at t = ,molem
L 3
DPi degree of polymerisation withinumber of monomers, –
Ei transport rate ofDPifrom the reacting lignin to cold glass
wall at t t,= mol m tL3
Ki overall cracking rate ofDPiat t t,= m tmol
L 3
kT i, evaporation/sublimation/ejection rate constant ofDPi,1t
kK first order cracking reaction rate constant,1t
kP second order polymerisation reaction rate constant,moltmL
3
m M( )q mass corresponding to each molecular weight slice (q), g
mi mass corresponding to eachDPiobtained after
discretiza-tion, g
mLignin total mass of lignin pressed between the two screens, g
Mw DP, i molecular weight of eachDPi, Da
Mw molecular weight, Da
〈Mn〉 number average molecular weight, Da
〈Mw〉 mass average molecular weight, Da
MwD molecular weight distribution, –
MnL 〈Mn〉 of lignin calculated from GPC chromatogram, Da
Mn tO, = calculated 〈Mn〉 of oil at t = , Da
Mw tO, = calculated 〈Mw〉 of oil at t = , Da
i mole fraction of eachDPi, –
iL mole fraction of eachDPipresent in lignin, –
i cal tO, , = calculated mole fraction of each DPi present in oil at
t = , –
Ntotal initial number of moles of lignin, mol
Pi overall polymerisation rate ofDPiat t t= ,molm t
L 3
RK i, cracking reaction rate of eachDPi,m tmol
L 3
RP i, polymerisation reaction rate for eachDPi,m tmol
L 3
VL volume of lignin pressed between the two screens,mL3 w logM( )q molecular weight distribution corresponding to each
mo-lecular weight slice (q), –
wi mass fraction of eachDPiin lignin or pyrolysis oils, –
wi expO, mass fraction of each DPi present in oil obtained from
experiments, – wi cal tO
, , = calculated mass fraction of eachDPipresent in oil at t = ,
–
YChar exp, experimentally obtained char yield,kgkg
Lignin
YChar cal t, , = calculated char yield at t = ,kgkg
Lignin
YOil cal t, , = calculated oil yield at t = ,kgkg
Lignin Superscripts L lignin O oil c consumption f formation
isothermality and to maximise the quenching rate of products (and unconverted lignin) after leaving the reaction zone [45–47]. Tem-peratures (425–793 °C) and pressures (500 Pa and 105Pa) were varied
in order to map the operating window and to further analyse the aforementioned interplay. Finally, the experimental results were com-pared with a mathematical model including the aforementioned che-mical reactions and mass transfer.
In the literature, numerous studies [9,18,21,25,48–57]compared the pyrolysis of various lignins which were chosen based on either their source and/or isolation methods, while we are the firsts, to the best of our knowledge, to interpret the results on the basis of the molecular weights of the lignins. The developed population balance model is re-latively simple compared to lumped class and linkage reactivity based models and has the easily measurable molecular weight distribution of lignin as the main input.
2. Experimental
2.1. Materials
Spruce organosolv lignin (SL) and wheat straw organosolv lignin (WSL) were kindly provided by the Energy Research Centre of the Netherlands (ECN), of which the isolation process is described else-where [10]. Another organosolv lignin (SOL) was purchased from Sigma Aldrich. SOL was prepared from a mixture of three hardwoods (50% Mapel, 35% Birch, 15% Poplar). Two batches of pyrolytic lignins (PL) were prepared, from pine wood derived fast pyrolysis oil, by cold water precipitation method described in the literature[7]. Milled wood lignin (MWL) was obtained from pine wood using the Björkman method [6]. For the fractionation experiments, Dichloromethane (DCM, Li-Chrosolv®, Purity ≥ 99.8%) was used. To recover the oils, the glass vessels were rinsed with tetrahydrofuran (THF, Merck LC LiChrosolv®, Purity ≥ 99.8%), which was also used for GPC analysis.
2.2. Screen-heater setup
A schematic representation of the screen-heater setup is shown in Fig. 1. The screen-heater setup characteristics and its operating proce-dure were described in our previous work[45–47]. Nevertheless, the main characteristics of the screen-heater setup are described in short here. An evenly distributed sample (∼50 mg) was pressed between two screens (thickness: ∼45 µm) to ensure uniform heating of the sample. Note, the screens were catalytically inactive[45]. The screens, with lignin sample, were heated to the final screen temperature (TFS) at
∼5000 °C s−1approaching as close as possible to isothermal
measure-ments. The holding time of the screens at TFSwas 5 s, and after that,
they were cooled at a rate of ∼60 °C s−1. The temperature of the
screens was measured and regulated by a pyrometer and a PID control routine programmed in LabVIEW, respectively. The screens holding sample were placed in a glass vessel, which was equilibrated at −180 °C during experiment by using liquid nitrogen (−196 °C). In the present contribution, tests were performed at 500 Pa and 105Pa. It was
estimated that for both pressures the quenching rate of the formed products was very fast (hot vapour residence time in order of 20 ms) which ruled out a significant effect of reactions outside the reacting lignin (minimised vapour phase reactions)[45]. At 500 Pa the escape rate of molecules away from the reacting lignin is higher than at 105Pa.
Lumped product yields (oil, char, and gas) and the mass balance closure was determined at the end of each experiment. In a series of experi-ments, pyrolysis of SOL was carried out in a wide temperature range (425–793 °C). It was found that upward of 530 °C the oil yield became constant, seeFig. 6andTable S3 in the supplementary information (SI). Therefore, all experiments at constant temperature were carried out at a TFSof 530 °C.
2.2.1. Product recovery
At the end of each experiment, the reactor was allowed to reach room temperature. In 500 Pa experiments, a gas sample was taken from
P 2X12 Volt DC High ampere 1 2 3 4 10 5 5 6 6 7 8 9 11 1 4 2 13 12
1 Vessel (glass)
2 Mesh/Lignin sample
3 Pyrometer spot
4 Liquid nitrogen bath (glas)
5 Copper electrode/clamp
6 Tape
7 Thermocouple connection
8 Syringe connection (gas sample)
9 Pressure sensor
10 Vacuum pump
11 Nitrogen gas
12 Pyrometer
13 Tube (glass)
14 Silicon sealing
14the glass vessel after filling the reactor with dry nitrogen gas, and for 105Pa experiments, it was taken from the gas bag. The composition of
gases was determined using gas chromatography, which was used to calculate the gas yield on lignin basis (kg gas kg−1lignin). The char
yield (kg char kg−1lignin) was estimated by the difference between the
mass of the screens before (with lignin), and after (with char) the ex-periment. Oil was condensed primarily on the interiors of the glass vessel and to a minor extent on the electrodes, clamps and bolts. The mass of the tape, wrapped around the electrodes, clamps and bolts, and the glass vessel were recorded before and after the experiment and used to calculate the oil yield (kg oil kg−1lignin). Independently determined
oil, char, and gas yields were summed together to determine the overall mass balance closure of the experiment. The formulas used for the calculation of oil, char, and gas yields are shown in theSection 1 of the SI.
2.2.2. Mass balance closures
The mass balance closure for individual pyrolysis experiments car-ried out at 500 Pa was between 0.87 kg kg−1and 0.97 kg kg−1, while
for 105 Pa experiments a slight decrease in mass closure
(0.8–0.88 kg kg−1) was observed. This decrease can be accounted for by
the observation that at 105Pa more pyrolysis products condensed on
the gas sampling line, thermocouple, and pressure sensor line instead of on the wall. Light organics (e.g. formic acid, formaldehyde) and water were most likely (partly) lost during gas sampling and during the dis-mantling of the setup, which explains that the mass balance closure was always below 100%. The number of experiments performed for each lignin, under identical conditions, were at least 3. Except, in the case of MWL only two experiments were performed at both pressures because of its limited availability. Reproducibility of the experiments was sa-tisfactory as shown in theTable S2 in the SI. The error bars on yields (Fig. 4andFig. 6) and 〈Mn〉 of the oils (Fig. 5) represent the standard
deviations on the mean.
2.3. Methods and analytical techniques 2.3.1. Fractionation
Technical lignins are a mixture of compounds with a varied number of aromatic rings, which can be fractionated into low and high mole-cular weight fractions by solvent extraction[58,59]. A short recap of the method used – a lignin sample was added to dichloromethane (DCM) in a weight ratio of 1 to 10. The DCM insoluble fraction, referred to as a heavy fraction (H), was separated from the soluble fraction by filtration. The DCM soluble fraction, referred to as a light fraction (L), was recovered by evaporating DCM. SeeFig. S1in the SI for the sche-matics.
2.3.2. Ash content determination
The ash content of lignin samples was determined by using the dry oxidation method. Summary – A cleaned and oven dried crucible was weighed three times with and without sample on a weighing balance (Mettler Toledo AX205, max 220 g, readability 0.01 mg). Next, the crucible containing the sample was kept in a muffle furnace at 575 °C for 24 h, after that it was weighed three times, after cooling, using the same balance. The ash content of the lignin was estimated by sub-tracting the weight of the empty crucible from the crucible with the ash divided by the weight of the lignin sample.
2.3.3. Analytical techniques
The composition of the gas sample was determined by gas chro-matography (Varian MicroGC CP4900). The molecular weight dis-tribution (MWD) of lignins and their pyrolysis oils was determined by
using GPC (Agilent Technologies 1200) equipped with three columns (7.5 mm × 300 mm, particle size 3 µm) placed in series packed with highly cross-linked polystyrene-divinylbenzene copolymer gel (Varian PLgelMIXED E). The column temperature was maintained at 40 °C
during analysis. Samples were dissolved in tetrahydrofuran (THF) in a 1:100 mass ratio and were filtered through a 0.45 µm Whatman RC Agilent filter. THF was used a mobile phase at a flowrate of 1 mL min−1. A variable wavelength detector (VWD, λ = 254 nm) was
installed. The calibration line was made with ten polystyrene standards with molecular weights ranging from 162 Da to 29510 Da. The ele-mental composition of lignins was determined by Eleele-mental Analyzer Inter Science Flash 2000. The abundance of β-O-4 linkages in different lignins were measured and quantified by using a 2D-CAHSQC NMR technique reported in the literature (Bruker Avance II 600 MHz spec-trometer)[60,61].
3. Characterisation of lignins
Physical characteristics such as elemental composition, aromaticity, mass average molecular weight (〈Mw〉), polydispersity values, and
content of β-O-4 linkages present in lignins are presented inTable 1. Note, all lignin samples were dried for 12 h in a vacuum oven (< 500 Pa, 20 °C) before any analysis and pyrolysis experiment.
As can be seen, MWL has an oxygen content of 33% which is con-siderably higher compared to the other lignins. The oxygen content of the other lignins (i.e. SL, SOL, WSL, PL) is quite comparable. SL has the highest carbon content. Furthermore, all lignins have similar hydrogen (∼6%), and nitrogen (< 1%) content. The ash content of all lignins is lower than 5 mg kg−1, and therefore it has assumed that it does not
affect the pyrolysis outcome.
In the case of WSL, the ratio of light fraction (L) to a heavy fraction (H) is 1:1, whereas for both pyrolytic lignins it is in the range of 0.65:0.35 (SeeTable S1in the SI). The molecular weights of light and heavy fractions are presented inTable 1. Note, the fractionation ex-periments were performed twice to check the reproducibility. The standard deviation in the yields of light fractions and heavy fractions, obtained from two separate fractionation experiments, was below 1% for all fractioned lignins.
As can be seen, the MWL is the heaviest lignin (3596 Da) studied in this work. The 〈Mw〉 of both pyrolytic lignins (PL1 and PL2) is ∼9.5
times lower than the MWL. Three organosolv lignins (SL, SOL and WSL) had their 〈Mw〉 between 2000 Da and 2500 Da. The 〈Mw〉 of light
fractions (2,5,8,11) are notably lower than their parent lignins (1,4,7,10) and much lower than the heavy fractions (3,6,9,12).
The poly-dispersity of 2.5 and high 〈Mw〉 of MWL, indicates MWL is
mainly composed of high molecular weight compounds originating from native lignin as well as low molecular weight compounds. Organosolv lignins have polydispersity around ∼2.1. The poly-dispersity values of light fractions (2,5,8,11) are much smaller than the corresponding heavy fractions (1,4,7,10). The MwD of all lignins (VWD
response) are presented inSection 6 of the SI.
The MWL lignin has the highest abundance of β-O-4 linkages namely 34.5 per 100 aromatic units. This value is close to the values typically reported for the softwood lignins, which are in the range of 45 to 50[62,63]. In three organosolv lignins (SL, SOL, WSL), the abun-dance of β-O-4 linkage is in the range from 1.8 to 8.6 per 100 aromatic units. Expectedly, in pyrolytic lignin (10) no traces of β-O-4 linkages were observed, which is in line with the literature[64].
In summary, these lignins have the same aromaticity as indicated by their equal H/C ratio (∼1.1), despite originating from three different isolation processes and differ uniquely in their molecular weights (588 Da to 3596 Da, 〈Mw〉). Only MWL stands out for having the
highest amount of β-O-4 linkages, while pyrolytic lignin represents the other end of that spectrum, i.e. no β-O-4 linkages. This set of lignins is envisaged to be a powerful tool to test the validity of the hypothesis discussed in the introduction with respect to the influence of the mo-lecular weight (distribution) on the pyrolysis process.
4. Modelling
The model consists of population balances for the reacting lignin and the produced oil. InFig. 2, the model is schematically presented. The population balances are made over segments calledDPi, wherei
refers to the number of monomer units out of which the lignin mole-cules in a segmentDPiare build-up. In a segmentDPithe lignin
mole-cules are identical linear polymers. The molecular weight of a mono-meric unit is set to 200 Da corresponding to the molecular weight of a typical building block alcohols of lignin (i.e. sinapyl alcohol of 210 Da). The following processes are considered for each segment of reacting lignin: cracking reactions, polymerisation reactions and removal (transport) from the reaction zone of unreacted lignin and reaction products. It is assumed that this is the minimal set of reactions and mass transport needed for predicting yields of oil and char and the molecular weight distribution of oil. For these processes experimental proofs are available. Cracking reactions are evident from the yield of lowerDPi
fractions in the oil as compared to their presence in the feedstock (see Fig. S2 in the SI). Washing the char with tetrahydrofuran showed a liquid fraction that contained larger molecules than present in the feed (seeFig. S3in the SI), hence polymerisation reactions must have taken place. Reactions in the vapour phase outside the particle are excluded by the low hot vapour residence time (20 ms). It could be that vapours react while still in/on the hot sample. Therefore, cracking and poly-merisation reactions used in the model must be considered as a lumped description of these reactions in the liquid/vapour/solid state of the hot reacting sample. Transport of molecules from the reaction zone is ob-vious as oil was collected at the cold vessel wall. It is assumed, based on experimental verification[45], that there are no reactions in the vapour phase and in the oil phase which is condensed on the cold vessel wall. Consequently, for the oil product, only the transport of molecules from the reacting lignin sample is considered. Cracking and mass transport were also included in the population balance model of Solomon et al. [39], though mass transport was described differently.
TheDPigrid ranges from i 1= to i=2imaxwithimaxbeing 50 in the
standard simulations (DPmax=10, 000Da). In the model, the bins be-tween DP( max+1) to DP2 max represent char. The characteristics of the
molecules in the char bins are – (1) they do not react (2) they cannot leave the hot reacting lignin. Without these assumptions i.e. without char bins, zero char production is predicted by the model. Summarizing, char is set to be equivalent to the >DP50molecules in the calculation domain. This is a rather a blunt assumption, however, it can be argued that molecules with DP>25will eventually react to char due to their very low likelihood of leaving the reaction zone, especially for 105Pa experiments. It should be noted that the molecules in the char
bins ( DP( max+1) to DP2 max) have same structure as that of molecules
present in the reacting lignin (DP1to DPmax). The cross-linking reactions
of the molecules are not included in the model. 4.1. Mole balances
The mole balances for the segmentDPiof the reacting lignin(CiL)
and oil C( iO)are given by Eq. (4-1) and Eq. (4-2): dC dt E K P iL i i i = + + (4-1) dC dt E iO i = (4-2) whereEiis the rate of transport ofDPifrom the reacting lignin to oil,Ki
is the net rate at which differentDPiare formed during cracking
reac-tions, andPiis the net rate at which different DPiare formed during
polymerisation reactions. The mathematical description of these three processes will be discussed in following paragraphs.
4.2. Transport from the reaction zone
It is assumed that a single first order equation can describe these processes:
Ei=k CT i i, L (4-3)
kT i, =e Ai+B (4-4)
The exponential term describes the evaporation and/or sublimation rate which is, as the Mwand the pressure affect the diffusion coefficient
[39,40], higher for lighter molecules and at lower pressure.B is in-cluded to allow a fixed,DPiindependent, ejection rate. In simulations,
the value ofBwas set to 0.02 and 0 for 500 Pa and 105Pa experiments,
respectively. Transport away from the reaction front is expected to be faster at lower pressure. Instead of using a flux equation for the escape rate, like the film model used by Solomon[39], a purely mathematical approach is used (see Eq. (4-3) and Eq. (4-4)) that does include the effects described above. This is rationalized by the expectedly very vigorous mass transport under the conditions applied, which is not captured by simplified models like the film model. In literature, the transport of species has been described by Darcy’s law[65]or modified Clausius–Clapeyron equation[66]. Formation of bubbles was observed on the reacting lignin (at 105Pa), which points towards vigorous mass
transport and to the generation of aerosols (seeFig. S4in the SI). For DP
( max+1)to DP2 maxbins, i.e. for char, the transport rate is set to zero.
The green arrows inFig. 2representsDPi dependent transport rates of
molecules, highest forDP1and lowest for DPmax.
4.3. Cracking
The following first order reaction describes cracking:
DPi DPj+DPi j(i>j) (i>1) (j>0) (4-5)
RK i, =k CK iL (4-6)
It is assumed that only the bonds connecting the monomer units present in theDPisegment can be cracked. Therefore, the segmentDPi
can crack in i (i 1)
i
( 1) !
( 2) ! = different ways. It is also assumed that all possibilities of cracking have the same likelihood, i.e. the rate constant kK, is the same for all cracking reactions. Under these assumptions the
cracking rateKicofDPibecomes,
Kic=(i 1)RK i, =(i 1)k CK iL (4-7)
DPiis also formed by cracking reactions. Obviously,DPican only be
formed from a segmentDPjwith j i> . The rate of formation ofDPiby
cracking is given by: Kif 2 k C j i i k jL 1 max = = + (4-8)
It is set in the model that char ( DP( max+1)to DP2 max) does not crack
and therefore the sum in Eq. (4-8) runs untilimax. Eq. (4-7) and Eq. (4-8)
are also used by Soloman et al.[39]. Combing Eq. (4-7) and Eq. (4-8) gives the overall reaction rate ofDPidue to cracking:
K K K k C i k C j K K 2 ( 1) 2 1 1 i if ic j i i K jL K iL j i i j c ic 1 1 max max = = = = + = + (4-9) For example,DP1andDP7are formed by consumption of DP8in a cracking reaction, seeFig. 2.
4.4. Polymerisation
The following second order reaction describes polymerisation: DPj+DPi j DP ii( >j) (i>1) (j>0) (4-10)
RP i, =k C CP jL i jL (4-11)
It is assumed that all molecules in segments betweenDP1to DPmax
will polymerize with each other, in all possible combinations, to form polymerisation products which are not bigger than DP2 max. As
pre-viously mentioned, the DP( max+1)to DP2 maxsegments do not take part
in polymerisation reactions but will be formed as a product, which represents char. Like in case of cracking, all polymerisation possibilities have the same likelihood i.e. the reaction rate constant,kP, is the same
for all polymerisation reactions. Thus, the rate of consumption ofDPiin
polymerisation reactions can be written as follows.
Pic k C C k C C 1 i i j i P iL jL P iL j iL max 1 max = + = = (4-12) For example,DP3andDP5are consumed in a polymerisation reac-tion to form DP8, seeFig. 2. Also,DPiwill be formed by polymerisation,
e.g. DP20is formed as a result of polymerisation reaction between DP8 andDP12(Fig. 2). The rate at whichDPiis formed by polymerisation is
described by the following equation, in which k C CP jL 0
i jL = for all
combinations with i j( )>DPmax, i.e. products larger than DP2 max
cannot be formed. P k C C P k C C i i i i i i (2 1) 1 3 (2 1) 2 2 2 if j P jL i jL if j P jL i jL max max 1 1 i i 1 2 2 = = + = = (4-13) The net reaction rate ofDPiby polymerisation is given by combining
Eq. (4-12) and Eq. (4-13),
Pi=Pif Pic (4-14)
4.5. Discretization of GPC chromatogram
The way in which MwD of lignin or pyrolysis oil, in terms of
w logM( ), is calculated has been described elsewhere[67]. m M w logM
ln
( ) ( )
10
q= q (4-15)
Mass corresponding to each molecular weight slice (q) between 1 Da and 10,000 Da is calculated and then subsequently discretised into 50 segments of 200 Da. The mass fraction ofDPiis calculated,
w m m i i i 1 i 50 = = (4-16)
Note, the MwDs, in terms of mass fractions (wi), of all lignins and
experimentally obtained pyrolysis oils presented in this work (Fig. 3 and all figures inSection 6 of the SI) are calculated by using this pro-cedure. These discretised mass fractions of each DPi segments were
converted to corresponding mole fractions by Eq. (4-17),
i w M i 1Mw 50 i w DPi i w DPi , , = = (4-17) 4.6. Initial conditions
The mole fraction of eachDPi is used to calculate the molar
con-centrations of theDPi’s in the lignin feedstock.
C N V V i feedL i L total L iL mM L , Lignin nL = = < > (4-18) The molar concentration distribution of the lignin is the initial condition for CiLwith1 i imax.
Ci tL,=0=Ci feedL, 1 i imax
For the char bins, i.e. the segments between DP( max+1)to2DPmax
Ci t,L=0=0 (imax+1) i 2imax
Note, the MwD of a lignin sample obtained from the GPC depends on
– (1) packing material of the column, (2) geometry of column, (3) eluent and its flow rate, and (4) detector type[67–69].
Ci tO,=0=0 1 i imax
4.7. Numerical method
The model equations were implemented in the coding environment of Matlab® 2017a. The numerical integration was carried out with the built-in ode23tb solver which is especially suited for stiff systems and is based on the trapezoidal rule with a second order backward difference approximation.
4.8. Characteristics of the model
The model has two important characteristics:
(1) The predictions of CiLandCiOat t = ∞ are invariant to the
multi-plication of the constants kT i,,kPandkKwith the same factor (for
details seeSection 7.2.1 of the SI).
(2) Changing the computational grid by changing the starting point of the char bins (remaining the final grid points) does not influence the predictions after fittingk
k
T avg P
, andk
kKP as long as the startingDPiof the char bins corresponds with a fraction of zero of thatDPiin the
lignin feed (for details seeSection 7.2.2 of the SI). 4.9. Post-processing
The mass fraction and mole fraction of theDPisegments in the oil
are given by:
C M C M wi cal tO i t O w DP i i i tO w DP , , , , 1 , , i max i = = = = = (4-19) i cal tO M i i M , , w 1 w i cal t O w DPi max i cal tO w DPi , , , , , , = = = = = (4-20) The 〈Mw〉 and 〈Mn〉 of the oil is calculated as follows:
Mw tO i w M i i cal t O w DP ,= = max=1 , ,= , i (4-21) Mn tO i M i i cal tO w DP ,= = max=1 , ,= , i (4-22)
The oil and char yields are calculated with the following equations:
Y C M C M C M Char cal t i i i i tL w DP i i i tO w DP i i i i tL w DP , , 1 2 , , 1 , , 1 2 , , max max i max i max max i = + = = + = = = = + = (4-23)
YOil cal t, ,= =1 YChar cal t, ,= (4-24)
4.10. Parameter estimation
fj= {(wi expO, wi cal tO, ,= )2+(YChar exp, YChar cal t, ,= ) }2 (4-25) As mentioned before, the predictions of CiLandCiO at t = ∞ are
invariant to multiplication the constants kT i,,kPandkKwith the same
factor. This in combination with the fact that the experimental method does not allow the determination of the temporal evolution of oil and char yield, instead final yields (t = ∞) are obtained, leads to the re-striction that it is not possible to determine all three constants by fitting them to the experimental results. In fact, it is only possible to determine the ratio of the constants, here expressed ask
k
T avg P
, andk kKP.
For each lignin, the experimental results were fitted to the model usingk k T avg P , andk k K
P as parameters. This is called the individual fit pro-cedure. The objective function for this (Eq. (4-25)) is comprised of – (1) the sum of squared difference between the MwD of experimentally
ob-tained oil and calculated, and (2) the squared difference between the experimentally obtained and calculated char yield. Note, as mentioned in theSection 2.2, some of the volatile products (potential oil) could not be recovered, and therefore, the char yield is used in the minimization routine since it was measured more accurately compared to oil yield. Parameters estimation is carried out with the Matlab in-built optimi-sation function lsqnonlin. 95% confidence intervals for each parameter were determined by the built-in Matlab function nlparci.
g min f j n j 1 = = (4-26)
In order to estimate a single set of values ofkT avgk P
, andk k
K
P per pres-sure, describing pyrolysis of all lignins studied in this work, the ob-jective function shown in Eq. (4-26) is used, where n represents the number of lignins. This is what is called the total fit procedure.
5. Experimental results
5.1. Molecular weight distribution of oils at 530 °C
InFig. 3, the MwD of two lignins (11 and 14) and their oils produced
at 500 Pa and 105Pa are plotted. Also, model predictions are shown.
Because these two lignins are entirely different from each other re-garding their 〈Mw〉, 〈Mn〉, MwD and β-O-4 linkage content, they were
selected to illustrate the observed differences in pyrolysis behaviour. In the case of lignin-11, the MwD of the oils nearly overlapped with the
MwD of the parent lignin (Fig. 3-A). On the contrary, a significant shift
towards the lower molecular weight region was observed in the MwD
for the of lignin-14 oils as compared to the feedstock (Fig. 3-B). For lignin-14, the oil obtained at 500 Pa was heavier than the oil obtained at 105Pa.
5.2. Lumped product yields at 530 °C
InFig. 4, gas, oil and char yields are plotted as a function of the 〈Mn〉 of the lignin. At both pressures, a decrease in oil yield was
ob-served with an increase in the 〈Mn〉 of lignin. At 105Pa lower oil yields
were achieved than at 500 Pa. This difference in oil yield became wider with an increase in 〈Mn〉 of the lignins. At both pressures, the oil yields
obtained for the light fractions (2,5,11) were higher than the yields achieved for their parent lignins (1,4,10) and their respective heavy fractions (3,6,12). For low 〈Mn〉 lignins (< 500 Da), high oil yields of
up to 0.96 kg kg−1 were achieved at 500 Pa and decreased to
0.78 kg kg−1 at 105 Pa. In case of high 〈M
n〉 lignins (> 500 Da), a
significant drop in oil yield was observed (0.32 kg kg−1) at 105Pa
compared to the 500 Pa experiments. In the literature, oil yields of ∼0.12 kg kg−1 (at 105Pa) [70] and ∼0.8 kg kg−1 (∼500 Pa)
[22,24,31] in screen-heater reactors, ∼0.55 kg kg−1in a tubular
re-actor[16], and ∼0.40 kg kg−1in a fluidised bed reactor[23,25]were
reported.
Also at both pressures, the char yield increased with the increase in 〈Mn〉 of the lignins. For low molecular weight lignins (< 500 Da), the
char yields were below 0.1 kg kg−1, while the difference between the
char yields obtained at 500 Pa and 105Pa was small. In case of lignins
heavier than 500 Da, char yields increased significantly especially for experiments performed at 105Pa. At both pressures, the char yields
obtained for the light fractions (2,5,11) were lower than the yields achieved for their parent lignins (1,4,10) and their respective heavy fractions (3,6,12). In the literature, char yields of ∼0.75 kg kg−1in
TGA[23,33,71,72], ∼0.45 kg kg−1(at 105Pa)[70]and ∼0.12 kg kg−1
(∼500 Pa)[22,24]in screen-heater reactors, 0.22 kg kg−1(500 Pa) to
0.35 kg kg−1 (105Pa) in a modified pyroprobe reactor [73],
∼0.35 kg kg−1in a tubular reactor[16], and ∼0.35 kg kg−1in a
flui-dised bed reactor[23,25]were reported.
Only a small amount of gases (∼0.025 kg kg−1) were produced for
all lignins during pyrolysis at 500 Pa. A marginal increase, if at all, in gas yield was observed at 105Pa. Gases produced during pyrolysis were
mainly CO2and CH4, while CO was only present in trace quantities. In
literature, at ∼530 °C similar gas yields were obtained by Iatridis et al. and Zhou et al. in screen-heater reactors and by Lou et al. in a tubular reactor [16,22,70]. Contrary, gas yields up to ∼0.25 kg kg−1 and
∼0.30 kg kg−1 were reported in micro-pyrolysers [20,74] and TGA
[75], respectively.
5.3. The average molecular weight of oils at 530 °C
Fig. 5shows the 〈Mn〉 of the oils as a function of the 〈Mn〉 of the
lignins. At both pressures, the oils obtained from pyrolysis of low 〈Mn〉
lignins (< 500 Da) showed no significant reduction in 〈Mn〉 as
com-pared to their parent lignins (data close to parity line). From ∼500 Da and above the oils seemed to have a constant 〈Mn〉 which was
in-dependent of the 〈Mn〉 of the lignin. This trend was most clearly
ob-served at 105Pa. At 500 Pa, larger deviations in molecular weights of
oils were observed, which might be explained by the random ejection of molecules from the reacting lignin. It is worth noticing that the yields and the 〈Mn〉 of the oils obtained from lignin 14 (richest in β-O-4
lin-kages) do not deviate from the trend as observed for other lignins. 5.4. Effect of temperature
Fig. 6shows the oil, char and gas yields obtained from the pyrolysis of lignin-13 versus the final screen temperature (TFS), at two different
pressures. The experiments were carried out in a temperature range between 425 °C and 793 °C.Fig. 6shows that char yield decreased with an increase in TFS, at both pressures. It can be seen that at 500 Pa, the
maximal amount of gases were produced above the TFSof 530 °C and
their yield remained independent of TFSafter that. At 105Pa, the gas
yield was below 0.06 kg kg−1 until 530 °C and thereafter increased
significantly to 0.17 kg kg−1(at 793 °C). It can be seen that at 500 Pa,
0.62 kg kg−1 of oil yield was achieved at 425 °C and above 530 °C it
became constant at a value of ∼0.8 kg kg−1. At 105Pa, the oil yield was
independent of TFS,i.e. it remained constant at 0.46–0.49 kg kg−1.
Si-milarly, Lou et al. observed a constant yield of oil (∼0.5 kg kg−1) in a
wide temperature range [76]. The effect of TFSon the 〈Mn〉 of oils
collected from the pyrolysis of lignin 13 is presented inFig. 7, showing independence at 105 Pa and a slight increase with increasing T
FSat
500 Pa. A similar trend was observed for pyrolysis of kraft lignin by Ben et al.[77]at 105Pa.
6. Discussion
6.1. Chemical reactions and mass transport
As mentioned in the introduction, we hypothesize that, like for cellulose[45], lignin pyrolysis cannot be described solely by chemical reactions. Instead, there is an interplay of chemistry and mass transport. As previously mentioned, the fact that the oil is collected indicates that mass transport of compounds away from the reaction front is taking place. Due to the influence of pressure on the escape rate – the higher rate at lower pressure – heavier oils are expected at lower pressure. This behaviour was observed in the lignin pyrolysis experiments – lighter oils were collected at 105Pa (∼400 Da) than at 500 Pa (∼800 Da), see
Fig. 5. At both pressures, the oil yields measured for the low 〈Mn〉
lignins (< 500 Da) were high (Fig. 4), and no significant change in their 〈Mn〉 was observed compared to the 〈Mn〉 of the parent lignins (Fig. 5).
This indicates that these low 〈Mn〉 lignins are mainly evaporating/
sublimating/ejecting while the extent of reactions, in particular, cracking reactions, is limited. On the contrary, for high 〈Mn〉 lignins,
the obtained oil yields were (much) higher than the mass fraction of lights in the feed (Fig. S2in the SI), and a noticeable decrease in 〈Mn〉
of oils compared to the feedstock (Fig. 5), affirms the presence of cracking reactions. Because the reactions take place on the reacting lignin (in the liquid and/or solid phase), it is not expected that the pressure influences their rates. At lower pressures, the likelihood of char production decreases (as observed experimentally, seeFig. 4) be-cause of the rate of mass transport away from the reacting lignin in-creases leaving less time for polymerisation. More char and less oil (see Fig. 4) at increasing 〈Mn〉 can be explained by the fact that heavier
molecules have more possibilities (more reactions) to form char in combination with a lower escape rate. Heavier oils at a higher tem-perature (seeFig. 7) is not expected when only considering chemical reactions, but can be explained by a higher escape rate of heavy mo-lecules from the reacting sample at higher temperatures.
The main experimental trends can be summarised as – (1) the lower the pressure, the higher (lower) the oil (char) yield, (2) the lower the pressure, the heavier the collected oil, (3) the heavier the lignin, the higher the char yield, (4) heavier oil at higher temperature all support that pyrolysis of lignin is an interplay of chemistry and mass transport. It is also clear that the molecular weight distribution of the lignin is not the only parameter controlling the pyrolysis yields and oil MwD. For
instance, lignin 5 and 13 having similar 〈Mn〉 but different oil, char and
gas yields, which may arise from their structural differences. 6.2. Modelling
Fitting each lignin individually to the model results ink k
T avg P
, andk kKP values per lignin that predict the char yield very accurately (Fig. 8-A). For 105Pa experiments, the predicted 〈M
n〉 of oils also match very well
with the experimental values, while they are slightly under-predicted for 500 Pa experiments (Fig. 8-B). This mismatch in the predictions can be a result of the random ejection of heavier molecules which is difficult to describe. Also, the MwD is predicted fairly well as can be seen in
Fig. 3. The range of fittedk k
T avg P
, andk
kKP are listed inTable 2(individual Fig. 5. Experimentally obtained 〈Mn〉 of oils obtained at 500 Pa and 105Pa as a
function of the 〈Mn〉 of the lignins.
Fig. 6. Oil, char, and gas yields as a function of the final mesh temperature
fit), and inTable S4in the SI they are listed per lignin with their 95% confidence intervals. As expectedk
k
T avg P
, is significantly higher at 500 Pa
due to the higher escape rate at lower pressure. The variation in k k
K P between the pressures is small, which should be the case as this is a ratio of pressure independent reactions. As the range of the parameters is not that large for the individual fits, it has been attempted to fit a single kT avgk
P
, andk k
K
P per pressure for all the lignins together, which is referred to as the total fit procedure (seeTable 2for the values). This results, on average, in a poor prediction of 〈Mn〉 and char yield (seeFig.
S9in the SI). However, the macro trends are still predicted. From the individual fits, it is found that the ratiok
kKPis a function of 〈Mn〉 of the lignins (Fig. S10in the SI).kkKPdecreases with an increase in
〈Mn〉, i.e. largerDPimolecules present in the lignin tend to polymerise
faster than described in the original model. No clear relation was ob-served between the values of k
k
T avg P
, and 〈M
n〉 for 500 Pa experiments,
while for the 105Pa experiments the values are independent of 〈M n〉 of
lignins. The fittedk
kKP as a function of 〈Mn〉 of the lignins together with
k k
T avg P
, (seeTable 2, total fit) are used as input for the model. The
pre-dicted char yields and 〈Mn〉 of oils are monotonous in nature and follow
the experimentally observed trend (Fig. 8-C andFig. 8-D). However, it
can be seen that 〈Mn〉 of oils are under-predicted and char yields are
over-predicted at 500 Pa.
After parameterisation, predicted char yields and 〈Mn〉 of the oils as
a function of temperature match the experimentally obtained values (Fig. S11in the SI). It can be seen from Fig. S 12-A in the SI that the values of kT avgk
P
, increase as a function of temperature (especially at
500 Pa) indicating a stronger increase of the mass transport rate with temperature as compared to the polymerisation rate. With the increase in temperature the values ofk
kKP increased, while they were similar at both pressures for any given temperature (Fig. S12-Bin the SI). This indicates that the ratio of reaction rate constants (k
kKP) is a function of temperature and is not affected by the pressure, as was also assumed. Concluding, after parameterisation of each lignin individually the model describes the experiments very well. However, it is not able to provide good quantitative predictions for all the lignins tested based on a single set ofk
k
T avg P
, (T and P dependent) andk k
K
P (T dependent). This could obviously be explained by the lignins tested being different in other aspects than the MwD, e.g. chemical structure. However,
in-cluding ak k
K
P relation accounting for a relatively higher polymerisation rate for larger molecules results in a reasonable qualitative and quan-titative description. The model including this relation can be used to predict yields and oil MwD based on the MwD of the lignin feedstock. It
must, however, be noted that the measured MwD of the feedstock
de-pends on the GPC/SEC machine and method used. As a result,k k T avg P , and k k K
P which are obtained by fitting will also depend on the GPC/SEC machine and method.
6.3. Process considerations
Depolymerisation of lignin is proposed for the production of fuels and platform chemicals[14,15,62,63,78,79]. The results obtained in this paper show that under (very) fast heating conditions and minimal chemical activity in the vapour/gas phase, pressure can steer the mo-lecular weight of the oil. All technical lignins with 〈Mn〉 > 800 Da
result in an 〈Mn〉 of the oil that depends on the pressure, but is invariant
to the MwD of the feedstock. At atmospheric pressure the oil is rather
light, dominated by monomers to trimers, which is beneficial for the production of chemicals. Potentially, high pressure lignin pyrolysis may favour the production of monomers, however, experimental investiga-tion is needed to gain further insights. Light oil comes, however, at the expense of lower oil yields and higher char yields. Nevertheless, also light oil can also be produced with a yield exceeding 50% at the tested conditions. There is a clear effect of the 〈Mn〉 of the feedstock and the
Table 1
Characterization of lignin.
Lignin Code C H O* N H/C 〈M
w〉** Đ β-O-4 linkages
(–) (–) (% on mass basis, dry) (mole mole−1) (Da) (–) (per 100 Ar units)
SL 1 66.9 6 27 0.1 1.1 2515 2.1 1.8 L-SL 2 – – – – – 1591 1.7 – H-SL 3 – – – – – 3462 1.8 – WSL 4 64.8 5.8 28.6 0.8 1.1 2043 2.0 8.6 L-WSL 5 – – – – – 1449 1.7 – H-WSL 6 – – – – – 2601 2.0 – PL1 7 68.1 6.3 25.5 0.1 1.1 725 1.5 – L-PL1 8 – – – – – 670 1.5 – H-PL1 9 – – – – – 1047 1.6 – PL2 10 64.8 6.5 28.6 0.1 1.2 616 1.6 0 L-PL2 11 – – – – – 588 1.6 – H-PL2 12 – – – – – 1241 2.0 – SOL 13 63.9 5.7 30.3 0.1 1.1 1858 2.2 7.8 MWL 14 60.7 6.3 33 < 0.1 1.2 3596 2.5 34.5 – Not measured.
* Oxygen content by difference: (100 – C – H – N). ** 〈Mw〉 is calculated from UV detector response.
Fig. 7. 〈Mn〉 of oil obtained from lignin 13 as a function of TFSat 500 Pa and
oil and char yield. If possible, a light feedstock should be selected as this will increase the oil yield. At low pressures, in our case 500 Pa, the oil is heavier, but its yield is considerably higher. These conditions may be interesting for fuel production. Also a process, in which first as high as possible oil yield is achieved at low pressure followed by further tuning of the product distribution in the vapour/gas phase can be considered.
7. Conclusions
In this work, lignins having number average molecular weight be-tween 350 Da and 1900 Da were pyrolysed bebe-tween 425 °C and 793 °C and pressures of 500 Pa and 105Pa. It was found that – (1) With an
increase in number average molecular weight, oil yield decreased, and char and gas yields increased, (2) At higher pressures (105Pa) lighter
oils were collected than for 500 Pa, (3) Above 500 Da (at 105Pa) and
1000 Da (at 500 Pa) the molecular weights of oils became independent of the number average molecular weight of the parent lignins, and (4) Heavier oils were collected with an increase in temperature. Based on the gathered experimental evidence of cracking and polymerisation reactions, and mass transport of species from the reacting lignin, a population balance model, which includes these processes, was devel-oped. After parameterisation and by using the molecular weight dis-tribution of lignin as an input, the model predicts all experimentally observed trends. It is noteworthy that the model also predicts the char yields and the number average molecular weight of oils reasonably for
lignins, for which the β-O-4 linkage content varies significantly. The present contribution highlights that during fast pyrolysis of lignin the temperature alone has a limited impact on the oil yield and the number average molecular weight of the oils. The molecular weight distribution is one of the most important characteristics of the lignin, which has a significant influence on the pyrolysis product distribution. Modelling results show that heavier molecules tend to polymerise faster than lighter molecules, which results in higher char yields. The Table 2
The values ofkT avg kP
, andkK
kP at 500 Pa and 10
5Pa.
Fit procedure kT avg
kP , kK kP (mole m−3) (mole m−3) P = 500 Pa Individual fit 4–16 27–96 Total fit 10*( ± 0.6)** 39 ( ± 0.01)** P = 105Pa Individual fit 2–5 24–81 Total fit 3*( ± 0.2)** 42 ( ± 0.3)** * Values ofkT avg kP
, used with correlation for model predictions inFig. 8-C and
Fig. 8-D.
** Values in parentheses are the 95% confidence intervals on the ratios.
Fig. 8. Parity plots in which the experimental char yields (A) and 〈Mn〉 of oils (B) are plotted against their predicted values using individual fit procedure; Char yields
pressure, at which the pyrolysis is carried out, is certainly the most influencing process parameter, which alters the transport rate of mo-lecules away from the reaction front, thereby, changing their residence time in the hot reaction zone. It can be considered as the main steering wheel to manipulate the product yields and the number average mo-lecular weight of the oils.
Acknowledgements
The authors would like to thank Ivar Stokvis, Tom van der Meer, Peter Heerspink and Miranti Budi Kusumawati for their contribution to a part of the experimental work, to Dion Smink for providing the milled wood lignin sample, and to Alberto Juan for NMR analysis. Additionally, the authors would like to acknowledge the technical staff of the SPT group (Benno Knaken, Karst van Bree and Johan Agterhorst) for their excellent technical support. This work is financially supported by NWO (Project number - 717-014-006), The Netherlands to which authors are grateful.
Appendix A. Supplementary material
Supplementary data to this article can be found online athttps:// doi.org/10.1016/j.apenergy.2018.12.058.
References
[1] Rowell R. Handbook of Wood Chemistry and Wood Composites; 2005. [2] Boerjan W, Ralph J, Baucher M. Lignin biosynthesis. Annu Rev Plant Biol
2003;54:519–46.
[3] Stewart D. Lignin as a base material for materials applications: chemistry, appli-cation and economics. Ind Crops Prod 2008;27:202–7.
[4] Upton BM, Kasko AM. Strategies for the conversion of lignin to high-value poly-meric materials: review and perspective. Chem Rev (Washington, DC, US) 2016;116:2275–306.
[5] Collett JR, Billing JM, Meyer PA, Schmidt AJ, Remington AB, Hawley ER, et al. Renewable diesel via hydrothermal liquefaction of oleaginous yeast and residual lignin from bioconversion of corn stover. Appl Energy 2019;233–234:840–53. [6] Bjorkman A. Studies on finely divided wood. Part 1. Extraction of lignin with
neutral solvents. Sven Papperstidn 1956;59:477–85.
[7] Scholze B, Meier D. Characterization of the water-insoluble fraction from pyrolysis oil (pyrolytic lignin). Part I. PY-GC/MS, FTIR, and functional groups. J Anal Appl Pyrolysis 2001;60:41–54.
[8] Brauns FE, Brauns DA. Chem Lignin 1952.
[9] Wang S, Ru B, Lin H, Sun W, Luo Z. Pyrolysis behaviors of four lignin polymers isolated from the same pine wood. Bioresour Technol 2015;182:120–7. [10] Wildschut J, Smit AT, Reith JH, Huijgen WJJ. Ethanol-based organosolv
fractio-nation of wheat straw for the production of lignin and enzymatically digestible cellulose. Bioresour Technol 2013;135:58–66.
[11] Fasching M, Schröder P, Wollboldt P, Weber H, Sixta H. A new and facile method for isolation of lignin from wood based on complete wood dissolution. Holzforschung 2008.
[12] Adler E. Lignin chemistry—past, present and future. Wood Sci Technol 1977;11:169–218.
[13] Antal Jr. M. Biomass pyrolysis: a review of the literature Part 2—lignocellulose pyrolysis. In: Böer K, Duffie J, editors. Advances in solar energy. US: Springer; 1985. p. 175–255.
[14] Amen-Chen C, Pakdel H, Roy C. Production of monomeric phenols by thermo-chemical conversion of biomass: a review. Bioresour Technol 2001;79:277–99. [15] Pandey MP, Kim CS. Lignin depolymerization and conversion: a review of
ther-mochemical methods. Chem Eng Technol 2011;34:29–41.
[16] Lou R, Wu S, Lyu G. Quantified monophenols in the bio-oil derived from lignin fast pyrolysis. J Anal Appl Pyrolysis 2015;111:27–32.
[17] Kawamoto H. Lignin pyrolysis reactions. J Wood Sci 2017;63:117–32. [18] McClelland DJ, Motagamwala AH, Li Y, Rover MR, Wittrig AM, Wu C, et al.
Functionality and molecular weight distribution of red oak lignin before and after pyrolysis and hydrogenation. Green Chem 2017;19:1378–89.
[19] Guo D-L, Wu S-B, Liu B, Yin X-L, Yang Q. Catalytic effects of NaOH and Na2CO3 additives on alkali lignin pyrolysis and gasification. Appl Energy 2012;95:22–30. [20] Patwardhan PR, Brown RC, Shanks BH. Understanding the fast pyrolysis of lignin.
ChemSusChem 2011;4:1629–36.
[21] Jiang G, Nowakowski DJ, Bridgwater AV. Effect of the temperature on the com-position of lignin pyrolysis products. Energy Fuel 2010;24:4470–5.
[22] Zhou S, Garcia-Perez M, Pecha B, Kersten SRA, McDonald AG, Westerhof RJM. Effect of the fast pyrolysis temperature on the primary and secondary products of lignin. Energy Fuel 2013;27:5867–77.
[23] de Wild PJ, Huijgen WJJ, Heeres HJ. Pyrolysis of wheat straw-derived organosolv lignin. J Anal Appl Pyrolysis 2012;93:95–103.
[24] Hoekstra E, Van Swaaij WPM, Kersten SRA, Hogendoorn KJA. Fast pyrolysis in a
novel wire-mesh reactor: decomposition of pine wood and model compounds. Chem Eng J (Lausanne) 2012;187:172–84.
[25] Nowakowski DJ, Bridgwater AV, Elliott DC, Meier D, de Wild P. Lignin fast pyr-olysis: results from an international collaboration. J Anal Appl Pyrolysis 2010;88:53–72.
[26] Li D, Briens C, Berruti F. Improved lignin pyrolysis for phenolics production in a bubbling bed reactor – effect of bed materials. Bioresour Technol 2015;189:7–14. [27] Piskorz J, Majerski P, Radlein D. Pyrolysis of biomass – aerosol generation: prop-erties, applications, and significance for process engineers. 4th, Biomass conference of the Americas. Oakland, CA: Elsevier Science; 1999. p. 1153–60.
[28] Fratini E, Bonini M, Oasmaa A, Solantausta Y, Teixeira J, Baglioni P. SANS analysis of the microstructural evolution during the aging of pyrolysis oils from biomass. Langmuir 2005;22:306–12.
[29] Montoya J, Pecha B, Janna FC, Garcia-Perez M. Micro-explosion of liquid inter-mediates during the fast pyrolysis of sucrose and organosolv lignin. J Anal Appl Pyrolysis 2016;122:106–21.
[30] Teixeira AR, Mooney KG, Kruger JS, Williams CL, Suszynski WJ, Schmidt LD, et al. Aerosol generation by reactive boiling ejection of molten cellulose. Energy Environ Sci 2011;4.
[31] Nunn TR, Howard JB, Longwell JP, Peters WA. Product compositions and kinetics in the rapid pyrolysis of milled wood lignin. Ind Eng Chem Process Des Dev 1985;24:844–52.
[32] Grønli MG, Várhegyi G, Di Blasi C. Thermogravimetric analysis and devolatilization kinetics of wood. Ind Eng Chem Res 2002;41:4201–8.
[33] Jiang G, Nowakowski DJ, Bridgwater AV. A systematic study of the kinetics of lignin pyrolysis. Thermochim Acta 2010;498:61–6.
[34] Burhenne L, Messmer J, Aicher T, Laborie M-P. The effect of the biomass compo-nents lignin, cellulose and hemicellulose on TGA and fixed bed pyrolysis. J Anal Appl Pyrolysis 2013;101:177–84.
[35] Lotfi S, Mollaabbasi R, Patience GS. Kinetics of softwood kraft lignin inert and oxidative thermolysis. Biomass Bioenergy 2018;109:239–48.
[36] Faravelli T, Frassoldati A, Migliavacca G, Ranzi E. Detailed kinetic modeling of the thermal degradation of lignins. Biomass Bioenergy 2010;34:290–301.
[37] Hough BR, Schwartz DT, Pfaendtner J. Detailed kinetic modeling of lignin pyrolysis for process optimization. Ind Eng Chem Res 2016;55:9147–53.
[38] Dussan K, Dooley S, Monaghan RFD. A model of the chemical composition and pyrolysis kinetics of lignin. Proceedings of the combustion institute. 2018. [39] Solomon PR, King H-H. Tar evolution from coal and model polymers: theory and
experiments. Fuel 1984;63:1302–11.
[40] Avni E, Coughlin RW, Solomon PR, King HH. Mathematical modelling of lignin pyrolysis. Fuel 1985;64:1495–501.
[41] Klein MT, Virk PS. Modeling of lignin thermolysis. Energy Fuel 2008;22:2175–82. [42] Yanez AJ, Natarajan P, Li W, Mabon R, Broadbelt LJ. Coupled structural and kinetic
model of lignin fast pyrolysis. Energy Fuel 2018;32:1822–30.
[43] Karl Freudenberg ACN. Constitution and biosynthesis of lignin. Springer-Verlag, Berlin Heidelberg; 1968.
[44] Yanez AJ, Li W, Mabon R, Broadbelt LJ. A stochastic method to generate libraries of structural representations of lignin. Energy Fuel 2016;30:5835–45.
[45] Westerhof RJM, Oudenhoven SRG, Marathe PS, Engelen M, Garcia-Perez M, Wang Z, et al. The interplay between chemistry and heat/mass transfer during the fast pyrolysis of cellulose. React Chem Eng 2016;1:555–66.
[46] Marathe PS, Oudenhoven SRG, Heerspink PW, Kersten SRA, Westerhof RJM. Fast pyrolysis of cellulose in vacuum: the effect of potassium salts on the primary re-actions. Chem Eng J (Lausanne) 2017;329:187–97.
[47] Hoekstra E, van Swaaij WPM, Kersten SRA, Hogendoorn KJA. Fast pyrolysis in a novel wire-mesh reactor: design and initial results. Chem Eng J (Lausanne) 2012;191:45–58.
[48] Saiz-Jimenez C, de Leeuw JW. Pyrolysis-gas chromatography-mass spectrometry of isolated, synthetic and degraded lignins. Org Geochem 1984;6:417–22. [49] Faix O, Meier D, Grobe I. Studies on isolated lignins and lignins in woody materials
by pyrolysis-gas chromatography-mass spectrometry and off-line pyrolysis-gas chromatography with flame ionization detection. J Anal Appl Pyrolysis 1987;11:403–16.
[50] Jakab E, Faix O, Till F, Székely T. Thermogravimetry/mass spectrometry study of six lignins within the scope of an international round robin test. J Anal Appl Pyrolysis 1995;35:167–79.
[51] Terrón MC, Fidalgo ML, Galletti GC, González AE. Pyrolysis-gas chromatography/ mass spectrometry of milled wood lignin of two Chilean woods naturally decayed by Ganoderma australe, Phlebia chrysocrea and a brown-rot fungus. J Anal Appl Pyrolysis 1995;33:61–75.
[52] Ferdous D, Dalai AK, Bej SK, Thring RW. Pyrolysis of Lignins: experimental and kinetics studies. Energy Fuel 2002;16:1405–12.
[53] Beis S, Mukkamala S, Hill N, Joseph J, Baker C, Jensen B, et al. Fast Pyrolysis of Lignins 2010.
[54] Wang S, Lin H, Ru B, Sun W, Wang Y, Luo Z. Comparison of the pyrolysis behavior of pyrolytic lignin and milled wood lignin by using TG–FTIR analysis. J Anal Appl Pyrolysis 2014;108:78–85.
[55] Zhang J, Kim KH, Choi YS, Motagamwala AH, Dumesic JA, Brown RC, et al. Comparison of fast pyrolysis behavior of cornstover lignins isolated by different methods. ACS Sustain Chem Eng 2017;5:5657–61.
[56] Chu J, Jiang W, Wu S. Depolymerization characteristics during the pyrolysis of two industrial lignins. BioResources 2017.
[57] Zhang L, Chen K, Peng L. Comparative research about wheat straw lignin from the black liquor after soda-oxygen and soda-AQ pulping: structural changes and pyr-olysis behavior. Energy Fuel 2017;31:10916–23.
heterogeneity of glycol lignin. Ind Eng Chem Res 1991;30:232–40.
[59] Thring RW, Griffin SL. The heterogeneity of two Canadian kraft lignins. Can J Chem 1995;73:629–34.
[60] Koskela H, Kilpeläinen I, Heikkinen S. Some aspects of quantitative 2D NMR. J Magn Reson 2005;174:237–44.
[61] Constant S, Wienk HLJ, Frissen AE, Peinder Pd, Boelens R, van Es DS, Grisel RJH, Weckhuysen BM, Huijgen WJJ, Gosselink RJA, Bruijnincx PCA. New insights into the structure and composition of technical lignins: a comparative characterisation study. Green Chem 2016;18:2651–65.
[62] Zakzeski J, Bruijnincx PCA, Jongerius AL, Weckhuysen BM. The catalytic valor-ization of lignin for the production of renewable chemicals. Chem Rev (Washington, DC, US) 2010;110:3552–99.
[63] Rinaldi R, Jastrzebski R, Clough MT, Ralph J, Kennema M, Bruijnincx PCA, et al. Paving the way for lignin valorisation: recent advances in bioengineering, bior-efining and catalysis. Angew Chem Int Ed 2016;55:8164–215.
[64] Scholze B, Hanser C, Meier D. Characterization of the water-insoluble fraction from fast pyrolysis liquids (pyrolytic lignin): Part II. GPC, carbonyl goups, and 13C-NMR. J Anal Appl Pyrolysis 2001;58–59:387–400.
[65] Di Blasi C. Heat, momentum and mass transport through a shrinking biomass particle exposed to thermal radiation. Chem Eng Sci 1996;51:1121–32. [66] Dufour A, Ouartassi B, Bounaceur R, Zoulalian A. Modelling intra-particle
phe-nomena of biomass pyrolysis. Chem Eng Res Des 2011;89:2136–46. [67] Hoekstra E, Kersten SRA, Tudos A, Meier D, Hogendoorn KJA. Possibilities and
pitfalls in analyzing (upgraded) pyrolysis oil by size exclusion chromatography (SEC). J Anal Appl Pyrolysis 2011;91:76–88.
[68] Gosselink RJA, Abächerli A, Semke H, Malherbe R, Käuper P, Nadif A, et al.
Analytical protocols for characterisation of sulphur-free lignin. Ind Crops Prod 2004;19:271–81.
[69] Baumberger S, Abaecherli A, Fasching M, Gellerstedt G, Gosselink R, Hortling B, et al. Molar mass determination of lignins by size-exclusion chromatography: to-wards standardisation of the method. Holzforschung 2007:459.
[70] Iatridis B, Gavalas GR. Pyrolysis of a precipitated kraft lignin. Ind Eng Chem Prod Res Dev 1979;18:127–30.
[71] Di Blasi C. Comparison of semi-global mechanisms for primary pyrolysis of lig-nocellulosic fuels. J Anal Appl Pyrolysis 1998;47:43–64.
[72] Yang H, Yan R, Chen H, Lee DH, Zheng C. Characteristics of hemicellulose, cellulose and lignin pyrolysis. Fuel 2007;86:1781–8.
[73] Pecha MB, Terrell E, Montoya JI, Stankovikj F, Broadbelt LJ, Chejne F, et al. Effect of pressure on pyrolysis of milled wood lignin and acid-washed hybrid poplar wood. Ind Eng Chem Res 2017;56:9079–89.
[74] Caballero JA, Font R, Marcilla A. Pyrolysis of Kraft lignin: yields and correlations. J Anal Appl Pyrolysis 1997;39:161–83.
[75] Raveendran K, Ganesh A, Khilar KC. Pyrolysis characteristics of biomass and bio-mass components. Fuel 1996;75:987–98.
[76] Lou R, Wu S-B. Products properties from fast pyrolysis of enzymatic/mild acidolysis lignin. Appl Energy 2011;88:316–22.
[77] Ben H, Ragauskas AJ. NMR characterization of pyrolysis oils from kraft lignin. Energy Fuel 2011;25:2322–32.
[78] Ragauskas AJ, Beckham GT, Biddy MJ, Chandra R, Chen F, Davis MF, et al. Lignin valorization: improving lignin processing in the biorefinery. Science 2014;344. [79] Demirbas A. Biomass resource facilities and biomass conversion processing for fuels
