Energy Fuels 2010, 24, 1274–1282 Published on Web 12/01/2009
: DOI:10.1021/ef900933k
Kinetics of the Thermal Decomposition of Biomass A. Saddawi,† J. M. Jones,*,† A. Williams,† and M. A. W ojtowicz‡ †
Energy and Resources Research Institute, School of Process, Environmental and Materials Engineering, University of Leeds, Leeds, United Kingdom, LS2 9JT and ‡Advanced Fuel Research, 87 Church Street, East Hartford, Connecticut 06108 Received August 27, 2009. Revised Manuscript Received November 5, 2009
This paper is concerned with the kinetics of the thermal decomposition of a woody biomass, willow. It addresses two questions. First, what method of data analysis is appropriate for extracting reliable kinetic data from thermogravimetric analysis (TGA) experiments? Second, what kinetics are most suitable for high heating rate situations such as those present in pulverized fuel power stations? It contains kinetic analysis of willow TGA data using a variety of approaches. A review of previously published work on biomass and its polymeric components helps ascertain the variation in kinetics, reasons for differences, and extrapolation to flame temperatures. The data falls into two main categories: (1) very high E and A values (>100 and up to 270 kJ/mol, and up to 1017 s-1) derived when model biomass components are studied, for example, cellulose; or the data is interpreted as the sum of a number of individual first-order reactions, for example, FG-BioMass; (2) intermediate and low E and A values (50-100 kJ/mol and 103 K/s) high E kinetics predict conversion well, and this can be rationalized since primary cracking reactions will dominate under these conditions. However, at heating rates of 105 K/s and temperatures of 1500 C (i.e., flame conditions), a compensation on the rates is seen and the choice of rate parameters is less critical. Two sets of kinetic data, E = 178.7 kJ/ mol, A = 2.2 1013 s-1 and E = 48.7 kJ/mol, A = 6.84 103 s-1, both predict conversions in keeping with the available experimental data.
mass transfer processes may play an important role, often complicating the interpretation of kinetic analyses.4,5 Most studies of thermal decomposition have been made using thermogravimetric analysis (TGA) at relatively slow heating rates and at temperatures up to 900 C. Consequently, TGA is often criticized that the data are not directly applicable to most industrial applications where the heating rates are greater, and, in the case of combustion, the final temperatures can be much higher. Weber has pointed out the difficulty of using TGAderived kinetic data for high temperature situations.5 A number of studies (using TGA) have examined the nature of the products and their rates of formation, and the rate of reaction of the parent biomass. Various authors have postulated the chemical steps that are involved in the decomposition of the constituents of biomass, cellulose, hemicellulose, and lignin. Cellulose pyrolysis, in particular, has attracted considerable attention, and the rate of reaction has been described as first-order.6-10 Biomass decomposition models have been
1. Introduction The decomposition of biomass has attracted considerable attention because it is a major step in its combustion1,2 and is also the key step during thermal processing methods, such as fast pyrolysis, to produce chemicals and bio-oil.3 Knowledge of the kinetics of this step is important in the modeling of both combustion processes and reactor kinetics. The rate of release, quantity, and composition of the volatiles influence flame ignition, stability, and the temperature profile in the radiant part of the furnace. Thus, these factors are important in burner design in pulverized fuel power stations, where they impact on NOx reduction mechanisms. Likewise, in fast pyrolysis the relative rates of decomposition, cracking, and repolymerization/condensation reactions influence the quantity and quality of bio-oil produced as well as the long-term stability of the oil. Biomass is an organic solid, and the thermal decomposition involves a number of simultaneous parallel reactions. The fact that the reaction involves a solid matrix means that heat and
(4) Klass, D. L. Biomass for Renewable Energy, Fuels, and Chemicals; Academic Press: San Diego, 1998. (5) Weber, R. J. Energy Inst. 2008, 81, 226. (6) Conesa, J. A.; Marcilla, A.; Caballero, J. A.; Font, R. J. Anal. Appl. Pyrolysis 2001, 58-59, 617. (7) Fitzpatrick, E. M.; Jones, J. M.; Pourkashanian, M.; Ross, A. B.; Williams, A.; Bartle, K. D. Energy Fuels 2008, Vol. 22, 3771. (8) Milosavljevic, I.; Suuberg, E. M. I&EC Res. 1995, Vol. 34, 1081. (9) Antal, M. J.; Varhegyi, G.; Jakab, E. I&EC Res. 1998, Vol. 37, 1267. (10) Gronli, M.; Antal, M. J.; Varhegyi, G. I&EC Res. 1999, Vol. 38, 2238.
*Author to whom correspondence should be addressed. Telephone: þ44 113 3432498. Fax: þ44 113246 7310. E-mail:
[email protected]. uk. (1) Backreedy, R. I.; Fletcher, L. M.; Jones, J. M.; Ma, L.; Pourkashanian, M.; Williams, A. Proc. Combust. Inst. 2005, 30, 2955. (2) Ma, L.; Jones, J. M.; Pourkashanian, M.; Williams, A. Fuel 2007, 86, 1959. (3) Bridgewater, A. V.; Imeche, I. I. Thermal Processing of Biomass for Fuels and Chemicals. International Conference on Renewable Bioenergy Technologies, Risks and Rewards, London, England, 2002. r 2009 American Chemical Society
1274
pubs.acs.org/EF
Energy Fuels 2010, 24, 1274–1282
: DOI:10.1021/ef900933k
Saddawi et al.
suggested in which reactions for the three constituents occur simultaneously,11-13 but other evidence has been provided showing that under certain conditions there can be interaction between components.14 However, in the case of TGA little interaction of components has been observed. More recently a number of studies have been made at moderately high temperatures and with high heating rates to simulate the conditions pertaining to flash pyrolysis.15,16 Few kinetic studies have been made at high temperatures similar to flame conditions, but there are some examples.17,18 So, the issue arises whether it is possible to extrapolate the mechanisms and kinetic data derived at low heating rates and temperatures to high heating rate/temperature situations. Reaction schemes consisting of multiple reactions are difficult to handle for moderately large organic molecules in the gas phase, and usually reaction scheme reduction systems are used to condense the large reaction mechanisms into more easily manageable systems. Alternatively, experimentally obtained data is used to generate empirical equations. In the case of biomass pyrolysis this latter approach is utilized with schemes that use one, two, and up to six first-order reactions. Commonly, one first-order reaction is used for computational fluid dynamic (CFD) models. If the experimental data is from TGA, or similar studies, there appears to be no consensus or general agreement on which region of experimental data should be considered for kinetic calculations. Also, in the case of multiple first-order reaction regions, the selection of the final mass (i.e., mass of reagent), utilized in some first-order reaction kinetics calculation models, can have a significant impact on the determined activation energy and pre-exponential factor values. Although nonlinear calculations eliminate the need for final mass allocation, the question remains whether the kinetics should be determined on a global basis or divided into multiple regions. These are a few of the difficulties that have arisen in the way experimental data is interpreted. In addition, such data is frequently obtained using TGA experiments at 300-900 C and then applied in high temperature combustion situations. This paper addresses two questions. First, what method of data analysis is appropriate for extracting reliable kinetic data from TGA experiments? Second, what kinetics are most suitable for high heating rate situations such as those present in pulverized fuel power stations? It contains kinetic analysis of willow TGA data using a variety of approaches. A review of previously published work on biomass and its polymeric components helps ascertain the variation in kinetics, reasons for differences, and extrapolation to flame temperatures.
Samples in the range of 5 mg were heated in a purge of nitrogen at a rate of 25 C/min to a final temperature of 550 C. The results have been interpreted in a number of ways as described in Section 3. 2.1. Materials and Sample Preparation. The SRC willow was obtained from Rural Generation Ltd. (Londonderry, Northern Ireland). The sample was ground and sieved. 0.15-0.18 mm samples were used in the TGA studies and some were further demineralized according to the method described previously by Nowakowski et al.19 2.2. Overview of Mathematical Methods Used to Extract Kinetic Parameters of Pyrolysis. 2.2.1. Reaction Rate Constant Method. A simple and widely used mathematical method to derive the pre-exponential factor and activation energy based on TGA experiments relies on the reaction rate constant, which is assumed to follow the Arrhenius function: E ð1Þ k ¼ A exp RT where k is the reaction rate constant, A is the pre-exponential factor, E is the activation energy, R is the gas constant, and T is the temperature. If the weight loss with time curve is assumed to be the result of one or more first-order reactions, then each reaction can be described by the following relation: 1 dm ð2Þ kt ¼ ðm -m¥ Þ dt It is worth noting that the calculated values of k rely upon a chosen terminal mass, and can deviate greatly depending on what mass value is chosen as terminal (M¥). Evaluation of A and E is straightforward using the following relationship: E1 ð3Þ ln k ¼ ln A RT A and E are found by the intercept and slope of a plot of ln k vs 1/T. 2.2.2. Temperature Integral Approximation by Murray and White. This method, along with the two that follow in Sections 2.2.3 and 2.2.4, provide different approximations to the temperature or Arrhenius integral.5 The reactions are assumed to be first-order. The Murray and White20 approximation of the temperature integral yields the following relation: -lnð1 -RÞ AR E ð4Þ ¼ ln ln 2 T EB RT where R = 1 - m/mo, m and mo representing current and original sample weight, respectively; and B is a set heating rate. A plot of ln[-ln(1 - R)/T2] versus 1/T will yield the values of A and E from the intercept and the slope. 2.2.3. Temperature Integral Approximation by Doyle. Doyle21 suggested a linear approximation to the temperature integral, which yielded the following relation: AE E -1:0518 ln½ -lnð1 -RÞ ¼ ln -5:33 ð5Þ BR RT
2. Biomass Pyrolysis Studies: Experimental Data The pyrolysis of SRC willow was made using a TGA analyzer (Stanton Redcroft Analyzer STA-780 Series). (11) Koufopanos, C. A. M., G.; Lucchesi, A. Can. J. Chem. Eng. 1989, 67. (12) Miller, R. S.; Bellan, J. Combust. Sci. Technol. 1997, Vol. 126, 97. (13) Ranzi, E.; Cuoci, A.; Faravelli, T.; Frassoldati, A.; Migliavacca, G.; Pierucci, S.; Sommariva, S. Energy Fuels 2008, Vol. 22, 4292. (14) Hosoya, T.; Kawamoto, H.; Saka, S. J. Anal. Appl. Pyrolysis 2007, 80, 118. (15) Shuangning, X.; Zhihe, L.; Baoming, L.; Weiming, Y.; Xueyuan, B. Fuel 2006, 85, 664. (16) Biagini, E.; Barontini, F.; Tognotti, L. I&EC Res. 2006, Vol. 45, 4486. (17) Dupont, C.; Chen, L.; Cances, J.; Commandre, J.-M.; Cuoci, A.; Pierucci, S.; Ranzi, E. J. Anal. Appl. Pyrolysis 2009, 85, 260. (18) Darvell, L. I.; P., H.; Jones, J. M.; Nowakowski, D. J.; Pourkashanian, M.; Williams, A. World Renewable Energy Congress (WREC 2005); Elsevier Ltd., 2005.
Again, with a predetermined heating rate B, A, and E can easily be found from the intercept and slope of a plot of the left-hand side of eq 5 versus 1/T. (19) Nowakowski, D. J.; Jones, J. M.; Brydson, R. M. D.; Ross, A. B. Fuel 2007, 86, 2389. (20) Murray, P.; W, J. Trans. Br. Ceram. Soc. 1955, 54, 204. (21) Doyle, C. D. J. Appl. Polym. Sci. 1962, 6, 639.
1275
Energy Fuels 2010, 24, 1274–1282
: DOI:10.1021/ef900933k
Saddawi et al.
Figure 1. (a) Pyrolysis weight loss curve for demineralized willow. Heating rate = 25 C/min. (b) Linear regression curves according to the reaction rate constant method (demineralized willow).
107.4 kJ/mol. In both cases, the final mass used in the analysis was that remaining at 550 C (the end of pyrolysis). It should be noted that this method is most valid when a very small region of the weight loss curve is examined. Also, biomass is composed of three polymers that degrade over different temperature regions. Thus, this method is not well suited for global calculations as the rate of mass loss does not remain linear. 3.2. Interpretation of Data According to the Temperature Integral Approximations. The three approximations of the temperature integral methods, and the reaction rate constant method, yielded varying values for A and E in both the demineralized and raw willow samples (calculated for various parts of the TGA weight-loss curve), detailed in Tables 1 and 2. It is clear that the models used to describe the demineralized sample kinetics can predict activation energies ranging from 44 kJ/mol (low) to over 107 kJ/mol (high), and from 58 kJ/mol (low) to 112 kJ/mol (high) for raw willow. Although there is variation in kinetic parameters predicted by the different methods, there is little variation in the parameters each method predicts for the two different samples (demineralized vs the raw willow samples). The A and E values obtained by each method are used in a back-calculation of the weight loss for the particular region considered, and is compared to the experimental data, to see how well the methods predict the actual weight loss. The variance of each method from experimental values (for the three TGA weight loss regions examined) is listed in Tables 1 and 2 and is further illustrated in Figures 2 and 3. As with the reaction rate constant method, the approximations by Murray and White and by Doyle rely on a linear regression and are commonly used for smaller initial ranges of weight loss data. Although the two approximations eliminate the risk of error introduced in the need for a selection of the appropriate terminal mass, the variance obtained by comparing the weight loss values derived from the predicted A and E values and the experimental weight loss data, listed in Tables 1 and 2, clearly shows that the reaction rate constant method is reliable if care is taken in choosing the appropriate terminal mass for the weight loss region considered. In this work the terminal mass was taken as the mass remaining at the end of the pyrolysis process (mass remaining at 550 C).
2.2.4. Temperature Integral Approximation by Senum and Yang. Senum and Yang approximated the temperature integral as a ratio of two polynomials.5,22 Once this ratio approximation is inserted in place of the temperature integral, a nonlinear regression is employed to determine the preexponential factor and the activation energy. The temperature integral is as follows: expð -xÞ x4 þ 18x3 þ 86x2 þ 96x ð6Þ pðxÞ = 2 4 x x þ 20x3 þ 120x2 þ 240x þ 120 where x = E/RT. 3. Biomass Pyrolysis Studies: Interpretation of Data Pyrolysis studies of SRC Willow (both raw and demineralized) using a TGA have been conducted and the results have been interpreted in a number of ways according to the techniques discussed in Section 2. The method of analysis is vital. Both high and low activation energies can be derived depending on the analytical method used and on the range of data selected. 3.1. Interpretation of Data According to the Reaction Rate Constant Method. The apparent first-order reaction rate constant method is typically applied to the initial data generated by TGA pyrolysis, where the conversion is low and the approximation described by eq 2 is valid. Conversion is defined as the fractional mass loss where, at 1, all mass remains and at 0, no mass remains. Two regions of the pyrolysis weight loss curve, highlighted in Figure 1a, were chosen to evaluate the initial kinetics and to compare the effect of deviation from this initial region on the values of the pre-exponential factor and the activation energy. The region labeled as “part 1” represents the traditionally chosen data range for this method, and the second region, “part 2,” represents a range of data slightly further along in the pyrolysis. Linear regression plots, described in Section 2.2.1, were calculated as seen in Figure 1b. As expected, the two regions yielded different pyrolysis kinetic values. For the data range represented in part 1, A and E were 3.51 103 s-1 and 68.8 KJ/mol, respectively. For part 2, the pre-exponential factor was 4.35 106 s-1, and the value activation energy rose to (22) Senum, G. I.; Y, R. T. J. Therm. Anal. Calorim. 1977, 11, 445.
1276
Energy Fuels 2010, 24, 1274–1282
: DOI:10.1021/ef900933k
Saddawi et al.
Table 1. Pre-exponential Factor, Activation Energy, and Variance Values for Various Methods (Demineralized Willow) Senum and Yang
Murray and White
Doyle
reaction rate constant
T range (C): E (KJ/mol) A variance
235-285 65.0 1.28 103 1.92 10-5
235-285 44.3 7.49 9.88 10-7
235-285 50.5 9.97 101 7.98 10-3
235-285 68.8 3.51 103 1.85 10-5
T range (C): E (KJ/mol) A variance
336-370 67.1 1.16 103 1.55 10-5
336-370 63.8 5.08 102 7.63 10-6
336-370 70.5 4.65 103 3.16 10-1
336-370 107.4 4.35 106 6.92 10-4
T range (C): E (KJ/mol) A variance
203-390 66.9 1.18 103 2.38 10-4
203-390 46.1 1.25 101 5.67 10-4
203-390 51.7 1.70 103 2.00 10-1
203-390 74.1 9.24 103 4.88 10-4
Table 2. Pre-exponential Factor, Activation Energy, and Variance Values for Various Methods (Raw Willow) Senum and Yang
Murray and White
Doyle
reaction rate constant
T range (C): E (KJ/mol) A variance
235-285 64.6 1.00 103 2.83 10-6
235-285 61.7 4.44 102 2.93 10-6
235-285 67.0 3.16 103 5.04 10-3
235-285 74.6 1.50 104 5.44 10-5
T range (C): E (KJ/mol) A variance
336-370 64.6 1.00 103 4.46 10-5
336-370 57.7 2.04 102 1.35 10-5
336-370 64.6 2.21 103 4.59 10-1
336-370 112.0 1.61 107 1.36 10-3
T range (C): E (KJ/mol) A variance
203-390 64.5 1.01 103 3.36 10-5
203-390 60.7 3.83 102 4.40 10-5
203-390 65. 6 2.61 103 3.43 10-1
203-390 73.1 1.03 104 3.66 10-4
On the basis of the fit of modeled versus actual weight loss, it is evident from Figure 2 that the Senum and Yang approximation is the most accurate of the considered methods on a more global basis, predicting the activation energy to be in the region of 65 kJ/mol (low) with a preexponential factor on the order of 1 103 s-1 (raw willow). The reaction rate constant method gives a respectable accuracy compared to the Senum and Yang approximation, predicting slightly higher activation energies of 75 kJ/mol (part 1), 112 kJ/mol (part 2) and 73 kJ/mol (global) and corresponding pre-exponential factors of 1.50 104, 1.61 107, and 1.03 104 s-1, respectively. From this work it is clear that, of the methods studied, the Senum and Yang, Murray and White, and the reaction rate constant method are yielding kinetics that describe the initial pyrolysis of willow more accurately, but which suggest a low activation energy is appropriate (less than 80 kJ/mol in this case). As discussed in the next section, both high and low activation energies have been reported for biomass and model compound pyrolysis. It is interesting that the same experimental data can yield different kinetic parameters depending upon the method used to extract them, and this may partly explain the wide variation of values reported in the literature. Of interest to industrial design applications is what kinetics are appropriate under flame conditions? A similar debate regarding coal pyrolysis kinetics unfolded in the latter part of the last century and consequently, similarities can be drawn, as discussed below.
Figure 2. Weight loss data comparing various temperature integral approximation methods to experimental data for the 203-390 C region (demineralized willow).
The Senum and Yang approximation is unique in that it employs nonlinear regression and is thus generally much more accurate than the other methods, and it is also capable of application to wider ranges of the TGA data while maintaining this accuracy. The choice of weight loss region incorporated in the calculations, however, remains critical for meaningful kinetics, since TGA experiments can become mass transfer limited at high degrees of conversion.
4. Previous Kinetic Studies of Solid Fuels Coal pyrolysis has been extensively studied and debated, particularly during the period of 1970-2000. Since the 1990s, there has also been a large research activity concerning biomass pyrolysis. It is noticeable that there is a wide scatter 1277
Energy Fuels 2010, 24, 1274–1282
: DOI:10.1021/ef900933k
Saddawi et al.
Figure 3. (a) Weight loss data comparing the reaction rate constant method (part 1) to experimental data (demineralized willow). (b) Weight loss data comparing the reaction rate constant method (part 2) to experimental data (demineralized willow).
Figure 4. Comparison of kinetic rates for coal pyrolysis.
among the values derived for apparent first-order kinetics, with similar fuels giving both high and low activation energies. The variation in coal pyrolysis rates was highlighted by Solomon and Serio23 and can be demonstrated on an Arrhenius plot (Figure 4), where the curve labels are in keeping with the original source. Figure 5 gives a similar plot for apparent first-order rates for pyrolysis of woody biomass, detailed in Table 3. It can be seen for both solid fuels that “low” or “high” activation energies have been reported. If low activation energies are used in predictive modeling, then the decomposition rate becomes significant at lower temperatures compared to if a high activation energy is used. The major applications have been for modeling coal gasification at about 1000 C and coal combustion at higher temperatures, where the coal particle goes through the process of devolatilization at temperatures of 500-1000 C. A variety of values have been used, and since the key temperature range is about 600-700 C the choice of value has not been critical
Figure 5. Comparison of kinetic rates for biomass, where solid lines represent pure species and the oval highlights the curves (10, 11) generated by this present work (demineralised willow).
since most curves cross at about that temperature (refer to Figure 4). In the CFD modeling of coal combustion the value is rather more critical since it influences the flame position and the length of the flame (determined by the period of time for the devolatilization to take place). Several studies have indicated that the high values should be used.30 In the case of coal pyrolysis substantial research has been conducted using TGA, wire mesh reactors, and drop tube furnaces, and it has been suggested that wide variations in rate can arise because of the uncertainty in particle temperature in the latter two cases.23 During coal pyrolysis, many simultaneous reactions are occurring involving different kinetics. As well as fragmentation and depolymerization of the coal structure; cross-linking to generate char is also an important pathway. If only the fragmentation reactions are considered, then higher activation energies would be expected. However, if all these are lumped together in an apparent
(23) Solomon, P. R.; Serio, M. A. Fundametals of the PhysicalChemistry of Pulverized Coal combustion; Lahaye, J., P., G. Ed.; Martinus Nijhoff Publishers: 1987; pp 126.
1278
Energy Fuels 2010, 24, 1274–1282
: DOI:10.1021/ef900933k
Saddawi et al.
Table 3. Details Accompanying Curves in Figure 5 reference
curve number
25 (Table 3) 25 (Table 3) 25 (Table 3) 25 (Table 3) 25 (Table 3) 25 (Table 3) 25 (Table 3) 26 (Table 4 26 (Table 4) 27 (Table 4)
1.1 1.2 1.3 1.4 1.5 1.6 1.7 2.1 2.2 3.2
27 (Table 4)
3.3
28 (Table 5)
4.1
28 (Table 5)
4.2
5.1 5.2 7.1 7.2 8.1 9.1 9.2 9.3 9.4 10
xylan xylan xylan xylan xylan xylan xylan cellulose hemicellulose water washed rice hulls (2nd lump) water washed rice hulls (3rd lump) Scots pine (hemicellulose 1) Scots pine (hemicellulose 2) Scots pine (Lignin) Scots pine (cellulose 1) Scots pine (cellulose 2) raw willow demineralized willow raw willow demineralized willow pine wood sawdust wheat straw coconut shell rice husk cotton stalk demineralized willow
11
demineralized willow
28 (Table 5)
4.3
28 (Table 5)
4.4
28 (Table 5) 19 (Table 4) 19 (Table 4) 18 (Table 1) 18 (Table 1) 29 15 (Table 4) 15 (Table 4) 15 (Table 4) 15 (Table 4) present work (Senum model) present work (rxn rate const model)
biomass
4.5
particle size
heating rate
Pre-exponential factor, A (1/sec)
activation energy, E (kJ/mol)
∼40 μm ∼40 μm ∼40 μm ∼40 μm ∼40 μm ∼40 μm ∼40 μm N.A. N.A. 1-2 mg piece
20 C/min 20 C/min 20 C/min 20 C/min 20 C/min 20 C/min 20 C/min variable variable 3,10,30,60,100 K/min
3.18 1011 8.38 1011 5.95 1013 2.93 1017 3.95 1016 7.16 1016 2.09 1016 6.53 107 9.71 108 8.81 107
188.0 218.0 206.0 267.0 246.0 238.0 234.0 154.0 199.0 165.0
1-2 mg piece
3,10,30,60,100 K/min
2.92 109
216.0
