Energy & Fuels 2000, 14, 947-952
947
Articles Comparisons of the Validity of Different Simplified NH3-Oxidation Mechanisms for Combustion of Biomass Tommy Norstro¨m, Pia Kilpinen,* Anders Brink, Esa Vakkilainen, and Mikko Hupa Combustion Chemistry Research Group, Åbo Akademi University, Lemminka¨ isenkatu 14-18 B, 20520 Turku, Finland Received July 9, 1999. Revised Manuscript Received March 29, 2000
In this work comparisons and validations of several simplified ammonia oxidation mechanisms at conditions relevant for combustion of biomass have been performed. The aim of the investigation was to clarify the weakness and the strength of the available simplified mechanisms for computational fluid dynamics (CFD) based prediction of NOx emissions. The mechanisms have been tested at typical conditions for biomass combustion, considering temperature and gas composition. All the simplified mechanisms have been compared against a validated comprehensive elementary reaction mechanism containing 340 reversible reactions between 56 chemical species. In comparing the simplified mechanisms, the mixing pattern between the species was described by a continuous stirred tank reactor, with typical residence times in eddy dissipation concept fine structure. The results show that, for substoichiometric conditions at low temperatures (1000 °C) and at oxidizing conditions the mechanisms worked better. The deviation from the results obtained with the comprehensive mechanism are in the range of 20% in nitric oxide levels and ammonia consumption at longer residence times, whereas the deviations increases up to 90% at shorter residence times.
Introduction Nitric oxide (NO) is a harmful emission from combustion. The main part of the NO emission from combustion of biomass originates in the fuel-bound nitrogen. The nitrogen in the fuel is released during the drying and pyrolysis stage of biomass particles in the combustion zone. The released organic nitrogen is converted mainly to ammonia and molecular nitrogen. The ammonia is then oxidized to nitric oxide or to molecular nitrogen depending on the conditions, such as local air ratio and temperature, in the furnace. This study is of interest for biomass combustion on grate, or in the bubbling fluidized bed combustion, in which the oxidation of ammonia takes place, mostly as homogeneous gas-phase reactions. This study is also relevant for burning of the pulping process spent liquor, black liquor, in the black liquor recovery furnace. For recovery furnaces the NO emission is, compared with other types of furnaces, * Corresponding author. Phone: +358-2-215-4681. Fax: +358-2215-4780. E-mail:
[email protected].
relatively low. Still a reduction of NO emissions is desired. Since the formation of nitric oxide inside a furnace is dependent on the chemical environment in combination with temperature in the furnace, primary actions inside the furnace are one efficient controlling tool for NOx minimization. With adjustments of the local air ratio and adjustments of the air port locations, changes in temperature field, flow pattern, and chemical concentration of key species are achieved. Computer simulations can easily test the effect of such manipulations of a furnace. Computational fluid dynamics (CFD) has become a frequently used tool in furnace analysis and development work.1,2 In CFD a number of such models are used to describe the physical and chemical phenomena, for example the radiation, turbulencechemistry interaction, and the combustion itself. The accuracy of the predictions depends strongly on the (1) Adams, T.; Stewart, R.; Jones, A. Tappi engineering conference; 1993. (2) Vakkilainen, E.; Kja¨ldman, L.; Taivassalo, V.; Kilpinen, P.; Norstro¨m, T. International Chemical Recovery Conference, Tampa, FL, June 1-4, 1998.
10.1021/ef990150z CCC: $19.00 © 2000 American Chemical Society Published on Web 08/03/2000
948
Energy & Fuels, Vol. 14, No. 5, 2000
accuracy of the submodels used, and therefore improvement of submodels is of great interest among CFD users. One of the most frequently used ways in CFD simulations for implementation of the effect of nitrogen chemistry is to use simplified, global kinetic expressions, which account for oxidation of ammonia (hydrogen cyanide in the case of burning a fossil fuel) to nitrogen oxide, and to molecular nitrogen. This work has been focused on validation of models for NH3 oxidation, often used in computer simulations. Five mechanisms from the literature,3-5 modifications of Duo by Jensen and Johnsson10 and Brouwer et al.6 have been tested together with a four-step mechanism for the main combustion chemistry, according to Jones and Lindstedt.7 Also a regression-based model by Brink et al.12 has been tested for NH3 oxidation. The mechanisms have been compared with a comprehensive reaction mechanism, consisting of 340 reversible elementary reactions between 56 species, as compiled by Kilpinen et al.9 It has been validated during the years against a large amount for experimental data, and it has been shown to give good prediction of NH3 oxidation at wide temperature and stoichiometric ratio ranges at relevant residence times. The mixing conditions have been chosen to correspond with the description of the fine structure in the eddy dissipation concept (EDC),13 i.e., continuous stirred tank reactor (CSTR), with residence times relevant for CFD use. The aim of this work was to point out the weaknesses or strengths of simplified mechanisms often used. Validation Approach and Initial Conditions. The conditions in furnaces with staged combustion change as a function of height in the furnace. In the lower part the conditions are strongly reducing, while the upper parts are slightly oxidizing. Also the temperature varies relatively much inside the furnaces. The warmer regions are often located in the lower part, whereas the cooler regions can be found in the upper regions. Since many of the today used simplified mechanisms have been developed for totally different conditions than can be found at the combustion zones in biomass furnaces, could the performance and applicability of them be questionable. To get an answer to the question of how well they work for these different conditions, one-dimensional modeling studies of six simplified schemes,3-5 and modifications of Duo by Jensen and (3) DeSoete, G. G. 15th Symposium International on Combustion (1093); 1974. (4) Duo, W. Ph.D. Thesis, Technical University of Denmark, 1990. (5) Mitchell, J. W.; Tarbell, J. M. AIChE J. 1982, 28 (2), 302. (6) Brouwer, J.; Heap, M. P.; Persing, D. W.; Smith, P. J. 26th Symposium International on Combustion, Naples, FL, 1996. (7) Jones, W. P.; Lindstedt, R. P. Combust. Flame 1988, 73, 233. (8) Forsse´n, M.; Hupa, M.; Pettersson, R.; Martin, D. J. Pulp Pap. Sci. 1997, 23 (9), J439-J446. (9) Kilpinen, P.; Hupa, M.; Aho, M.; Ha¨ma¨la¨inen, J. 7th International Workshop on Nitrous Oxide Emissions, Cologne, Germany, Apr. 21-23, 1997; Bergische Universita¨t Gesamthochschule Wuppertal: 1997. (10) Jensen, A.; Johnsson, J.-E. The Finn.-Swed. Flame Days 1996 (Sep). (11) Aho, K. Nitrogen. Licentiate thesis; Åbo Akademi University, 1994. (12) Brink, A.; Kilpinen, P.; Hupa, M. Report 95-17; Åbo Akademi University: Turku, Finland, 1995. (13) Magnussen, B. F. Eighteenth International Congress on Combustion Engines, Tianjin, China; 1989.
Norstro¨ m et al. Table 1. Pyrolysis Gas Composition Used at 800 °C and at a Stoichiometric Ratio 0.9 species
mole fraction
species
mole fraction
O2 H2 CO2
0.0000 0.0209 0.1420
CO H2O CH4
0.0211 0.2120 0.01/0.00
Table 2. Matrix of the Simulation Conditions temp pyrolysis gas at stoichiometric ratio methane residence times
800-1500 °C, stepping 100 °C 0.95 and 1.3 0.0 and 1.0 vol % 0.1-100 ms, in ∼200 steps
Johnsson,10 Brouwer et al.,6 and Brink et al.14 have been performed. Brink et al. differs from the others by giving the conversion factors of NH3 to NO and N2 explicitly instead of giving reaction rates. The simulation matrix has been divided into two main areas. The first corresponds to the substoichiometric areas to be found in the lower part of the furnaces, and the second, to the upper part of the furnaces where the conditions are more of an oxidizing nature. The second most interesting parameter for nitrogen studies, temperature, has stepwise been varied between 800 and 1500 °C. The residence time has been varied between 0.1 and 100 ms, which corresponds to the typical turbulent time scales in a furnace. The shorter residence times correspond to regions with higher turbulence intensity, i.e., near burner regions or air inlets, whereas the longer residence times correspond more to the regions far away from the burners or air ports. A typical composition of a slightly oxidized biomass pyrolysis gas was used at several stoichiometric ratios and at different temperatures. In Table 1, an example of a pyrolysis gas at 800 °C and stoichiometric ratio of 0.95 is given. The pyrolysis gas composition was adjusted to correspond to the conditions at biomass combustion at different temperatures and air ratios. The initial amount of NH3 was chosen to 100 ppmv. However, a reasonable variation of inlet NH3 concentration showed that the conclusions in this study were not sensible to this assumption of NH3 concentration. The effect of methane on the nitrogen chemistry is important, so all the simulations were performed both with and without a methane concentration of 1 vol %. A summary of the simulation matrix is given in Table 2. Mechanisms. Comprehensive. The comprehensive elementary reaction mechanism used included both the reactions for the main combustion chemistry (hydrocarbons up to C2) as well as the reactions for the nitrogen chemistry, as compiled by Kilpinen et al.9 The mechanism is based on the mechanism of Glarborg and Hadvig16 but has, however, been revised several times and grown consequently to a mechanism consisting of over 340 reactions between 57 species. The mechanism (14) Brink, A.; Kilpinen, P.; Hupa, M.; Kja¨ldman, L.; Ja¨a¨skela¨inen K. EPRI/EPA 1995 Joint Symposium on Stationary Combustion NOx Control, Book 3; 1995. (15) Glarborg, P.; Alzueta, M.; Dam-Johansen, K.; Miller, J. Combust. Flame 1998, 115, 1-27. (16) Glarborg, P.; Hadvig, S. NGC report NGC89/FM/1-01; Nordic Gas Technology Centre: Hørsholm, Denmark, 1989. (17) Glarborg, P.; Lilleheie, N. I.; Byggstøyl, S.; Magnussen, B. F.; Kilpinen, P.; Hupa, M. 24th Symposium (International) on Combustion; 1992.
NH3-Oxidation Mechanisms for Biomass Combustion
Energy & Fuels, Vol. 14, No. 5, 2000 949
was compared with several others by Kilpinen et al.18 The mechanism appeared to correspond well with the latest mechanism of Glarborg et al.15 Four-Step. The simplified four-step main chemistry reaction mechanism of Jones and Lindstedt (1988) was used as a base for testing the simplified NH3 mechanisms. The mechanism consists of four reactions. The third reaction was assumed to be irreversible, since the dissociation of water vapor is not of significance at the temperatures studied. The fourth reaction was the reversible water gas shift reaction. The reaction mechanism used for the main chemistry is shown as follows:
Duo. Duo4 has developed his mechanism for SNCR conditions and at constant oxygen concentrations of 4%, based on laboratory flow reactor studies in the temperature range 800-1050 °C.
CH4 + 1/2O2 f CO + 2H2
(1)
NH3 + NO + 1/4O2 f N2 + 3/2H2O
(9)
CH4 + H2O f CO + 3H2
(2)
NH3 + 5/4O2 f NO + 3/2H2O
(10)
H2 + /2O2 f H2O
(3)
CO + H2O ) CO2 + H2
(4)
Brouwer et al. In Brouwer et al.6 the rate expressions have been fitted via sensitivity analysis performed with comprehensive mechanism to an Arrhenius form with a preexponential factor, temperature exponent, and activation energy. The effect of CO on the radical chemistry, which highly effects the nitrogen chemistry, is integrated into the Arrhenius expressions for the reaction rates, as an additional term added to the temperature. The additional term is a function of the CO concentration.
1
De Soete. The first in the row of simplified ammonia oxidation mechanisms tested, and perhaps also the most frequently used in current CFD simulations, is the mechanism by DeSoete.3 This mechanism has been developed on the basis of measurements on flat premixed hydrocarbon/oxygen argon (or helium) flames. Overall NO and N2 reaction rates where determined, at flame temperatures varied between approximately 1500 and 2100 °C. For the NO formation reaction, the reaction order with respect to O2 is a function of the measured O2 concentration and varies between zero and one. The reactions used from the mechanism are
NH3 + O2 f NO + H2O + 1/2H2 1
NH3 + NO f N2 + H2O + /2H2
(5) (6)
Mitchell and Tarbell. The second mechanism has been developed for the conditions of pulverized coal combustion.5 The reaction rates have been derived via regression analysis of data, obtained from experiments, and are fitted for a temperature range between 1200 and 1600 °C. The mechanism of Mitchell and Tarbell consists totally of 12 reactions between 12 species. For this application only four of these reactions were relevant. The reactions used are the following:
NH3 + O2 f NO + H2O + 1/2H2
(5)
NH3 + NO f N2 + H2O + 1/2H2
(6)
HCN + H2O f NH3 + CO
(7)
NO + CH4 f HCN + H2O + 1/2H2
(8)
Reaction 8 is in the original version made for any hydrocarbon, but in this work the reaction has been specified for methane only. In the mechanism also reactions involving HCN have been taken into account. This is interesting due to the fact that we have small amounts of methane in the biomass pyrolysis gases. (18) Kilpinen, P.; Kallio, S.; Hupa, M. Proceedings of the 15th International Conference on Fluidized Bed Combustion, Savannah, GA; 1999.
NH3 + NO f N2 + H2O + 1/2H2
(6)
NH3 + O2 f NO + H2O + 1/2H2
(5)
Jensen and Johnsson10 have modified the mechanism of Duo for nonconstant oxygen concentration conditions. With their modifications the mechanism can be written as
S(CO) ) 17.5 log(XCO × 106) - 68 K The total model consists of six species and seven irreversible reactions. In this work only two reactions have been used.
NH3 + NO f N2 + H2O + 1/2H2
(6)
NH3 + O2 f NO + H2O + 1/2H2
(5)
Brink et al. Brink et al.12 have developed a regressionbased model for description of the nitrogen chemistry in the near-burner region in pulverized coal combustion. The model explicitly gives the conversion of the most important nitrogen containing species. In CSTR simulations with a comprehensive mechanism they noticed an abrupt change in the concentration which was due to the ignition. Much of the conversion of the nitrogen containing species took place during this stage, where the radicals play an important role. Since the fuel oxidation process usually is described in a very simplified manner, they suggested that a regression-based model for the NOx chemistry would be a way to overcome this problem. Brink et al.12 used the results from calculations of a CSTR with a comprehensive mechanism as reference data for the regression model. In their model they assumed that the conversion grade for the nitrogen containing species included in the model, i.e., NO, HCN, and NH3, are roughly independent of the incoming concentrations. They used this assumption to construct expressions for the conversion of NO to NH3, HCN, and N2 and comparable expression for the conversion of HCN and NH3 separately. This procedure was repeated for five discrete residence times, i.e., 0.1 ms, 1 ms, 10 ms, 100 ms, and 1 s. For intermediate residence times the conversions were
950
Energy & Fuels, Vol. 14, No. 5, 2000
Norstro¨ m et al.
Table 3. Simplified Mechanisms Testeda author Jones and Lindstedt7
formal reactions
kinetic expression
rate unit kmol/(m3 s)
CH4 + 1/2O2 f CO + 2H2
-rCH4 ) 0.44 × 1012 exp(-300000/(RT))[CH4]0.5[O2]1.2
CH4 + H2O f CO + 3H2
-rCH4 ) 0.3 × 109 exp(-30000/(RT))[CH4][H2O] - rH2 ) [H2]0.5[O2]2.25 0.25 × 1017 exp(-40000/(RT)) T [H2O]
H2 + 1/2O2 f H2O
-rCO ) 0.275 × 1010 exp(-20000/(RT))[CO][H2] -
CO + H2O S CO2 + H2
0.275 × 1010 exp(-20000/(RT))[CO2][H2]/K(T) DeSoete3
Mitchell and Tarbell5
b -rNH3 ) 4.0 × 106XNH3X O exp(-32000/(RT)) 2
NH3 + O2 f NO + H2O + 1/2H2
8
NH3 + NO f N2 + H2O + 1/2H2
-rNH3 ) 1.8 × 10 XNH3XNO exp
NH3 + O2 f NO + H2O + 1/2H2
-rNH3 ) 3.48 × 1020 exp(-100000/(RT))
NH3 + NO f N2 + H2O + 1/2H2
NO + CH4 f HCN + H2O + 1/2H2
P -rNH3 ) 6.22 × 1014 exp(-55000/(RT)) XNH3XNO R′T P -rHCN ) 1.94 × 1015 exp(-78400/(RT)) XNOXCH4 R′T P -rNO ) 1.00 × 104XNOXCH4 R′T
NH3 + NO f N2 + H2O + 1/2H2
-rNH3 ) 2.45 × 1014 exp(-29400/(RT))[NH3][NO]
NH3 + O2 f NO + H2O + 1/2H2
-rNH3 ) 2.212 × 1014 exp(-38160/(RT))[NH3]
NH3 + NO + 1/4O2 f N2 + 3/2H2O
-rNH3 ) 5.07 × 1014 exp(-35230/(RT))[NH3]0.5[NO]0.5[O2]0.5
NH3 + 5/4O2 f NO + 3/2H2O
-rNH3 ) 1.11 × 1012 exp(-276800/(RT))[NH3][O2]
HCN + H2O f NH3 + CO
Duo4
Jensen and Johnsson10
s-1
(-27000/(RT))
XNH3XO2
-rNH3 ) 2.24 × 108(T + S(CO))5.3[NH3][NO] exp(-83600/R(T+S(CO)))
Brouwer et al.6 NH3 + NO f N2 + H2O + 1/2H2
s-1
kmol/(m3 s)
kmol/(m3 s)
kmol/(m3 s)
-rNH3 ) 3.50 × 105(T + S(CO))7.65[NH3][O2] exp(-125300/R(T+S(CO)))
NH3 + O2 f NO + H2O + 1/2H2
S(CO) ) 17.5 log(XCO × 106) - 68 Brink et al.12
P
(1 + 6.90 × 10-6 exp(-42000/(RT))XO2) R′T
K
f (T,λ) ) b0 + b1T 2 + b3λ + b4λ2 + b5λT 2 + b6λ2T
NH3 f N2, NO HCN f N2, NO NO f N2
a
b in the expressions of DeSoete is a function of measured O2 level in the flame. The value of b gives the reaction order with respect to O2 and varies between 0 and 1. X is molar fraction. R ) 8.314 J/(mol K). R′ ) 1.987 kcal/(mol K). If not specified the units follow the SI system.
obtained by means of interpolation. Brink et al.12 reported that best results were obtained when the point related to nonignited systems were excluded from the reference set. They suggested that a separate regression equation should be used for estimating the ignition time during the prediction step. The form of the expression they used is
f (T,λ) ) b0 + b1T + b2T 2 + b3λ + b4λ2 + b5λT 2 + b6λ2T The validity range was 1000-1550 °C and the stoichiometric rate ratio from 0.5 to 1.5. The fuel was assumed to consist of a mixture of CO, H2, and methane. In a study by Brink et al.,14 where the regression-based model was compared to the reduced mechanism for the nitrogen chemistry of Glarborg et al.,15 they pointed out that when using the regression-based NOx model, it is important that the conditions respond to those the model was set up for. This restricts the use of the regression-based NOx model to systems where the main part of the NOx is formed from fuel-N, and the contribution from thermal-NO is only minor. A complete list of the simplified mechanisms tested in this work is given in Table 3.
Results and Discussion The concentrations of nitric oxide and ammonia obtained from the simulations were plotted versus residence time in CSTR. The interesting residence times in CFD modeling are in the range of 0.1-100 ms. The time scales in the fine structure correspond to the intensity of the turbulence. As the turbulence intensity increases, the time scales decrease and residence times in the CSTR decrease as well according to the EDC model. In a furnace, regions with high turbulence intensity are located at the burners or in the burner neighborhood, whereas regions with low turbulence intensity are located far from the burners, and typically at the upper part of the boilers. The diagrams show comparisons of the results of all tested simplified mechanisms and the comprehensive mechanism. By this technique, the best working mechanism for each set of conditions can be pointed out. By comparing the results obtained with the simplified mechanisms and the comprehensive mechanism, it becomes evident that no one of the simplified mechanisms tested well manages the complex nitrogen chemistry at the conditions tested. The deviations from the results obtained with the comprehensive mechanism were largest at low temper-
NH3-Oxidation Mechanisms for Biomass Combustion
Energy & Fuels, Vol. 14, No. 5, 2000 951
Figure 1. NO mole fraction as function of residence time at 900 °C. The pyrolysis gas is derived at a stoichiometric ratio of 0.95.
Figure 4. NH3 mole fraction as function of residence time at 1100 °C. The pyrolysis gas is derived at a stoichiometric ratio of 0.95.
Figure 2. NH3 mole fraction as function of residence time at 900 °C. The pyrolysis gas is derived at a stoichiometric ratio of 0.95.
Figure 5. NO mole fraction as function of residence time at 1300 °C. The pyrolysis gas is derived at stoichiometric ratio of 1.3. The methane concentration in the inlet gas is 1 vol %.
Figure 3. NO mole fraction as function of residence time at 1100 °C. The pyrolysis gas is derived at a stoichiometric ratio of 0.95.
Figure 6. NH3 mole fraction as function of residence time at 1300 °C. The pyrolysis gas is derived at stoichiometric ratio of 1.3. The methane concentration in inlet gas is 1 vol %.
atures (800-900 °C) combined with reducing conditions, especially at short residence times, Figures 1 and 2. When the temperatures were increased, the correspondence with the comprehensive mechanism became better, but still not at all of the same magnitude. An example of this is given in Figures 3 and 4, where the NO concentration and NH3 concentration are plotted at 1100 °C for all the mechanisms tested, including the regression-based model and the comprehensive mechanism. The correspondence with the comprehensive mechanism is very poor for all the mechanisms considering the NO concentration, whereas the Brouwer et al. model seems to predict the NH3 oxidation relatively well. The other simplified mechanisms shows slow
oxidation of NH3 and also much slower formation of NO. At conditions slightly oxidizing, the mechanisms worked better. This is due to the fact that many of the mechanisms were developed for oxidizing conditions, even though the temperatures and the gas compositions differed. In Figures 5 and 6 typical results from slightly oxidizing conditions at 1300 °C are shown. The correspondence with the comprehensive mechanism is much better for all of the simplified mechanisms, even though the mechanism by DeSoete still was too slow for this temperature region. The mechanism did not succeed in predicting the reduction of nitric oxides to molecular nitrogen, nor did it succeed in predicting the final concentrations of ammonia and nitric oxide.
952
Energy & Fuels, Vol. 14, No. 5, 2000
The regression-based model worked relatively well at those conditions that were inside the limitations set at the development. Since the model incorporates terms of up to the fifth grade, the model is strongly restricted to the limits by means of stoichiometric ratio and temperature. Conclusions Six simplified mechanisms for ammonia oxidation to nitric oxide and to molecular nitrogen have been tested against a comprehensive elementary reaction mechanism, as one-dimensional gas-phase simulations. The validation was made for testing the suitability of the mechanism to be used in computational fluid dynamics (CFD) modeling. The mechanisms were validated for conditions relevant for biomass combustion. The mechanisms failed to predict, especially in the reducing parts, the concentrations of ammonia and nitric oxide. This work shows that none of the existing mechanisms have
Norstro¨ m et al.
the potential to correctly predict the nitrogen chemistry at biomass combustion conditions. The applicability of them in those kinds of conditions could strongly be questionable. These results further emphasize the need for a simplified description of the nitrogen chemistry, which would be applicable in CFD use. One solution might be a set of simplified mechanisms, which are optimized for different conditions reflecting different regions in the furnaces. Acknowledgment. This work was a part of the Finnish national research program CODE that was initiated by the National Technology agency in Finland (Tekes). The financial support by Tekes, the Nordic Energy Research Program, the Academy of Finland, Kvaerner Pulping Oy, Fortum Power and Heat Oy, and Ahlstro¨m Machinery is gratefully acknowledged. EF990150Z
