Ammonia conversion and NOx formation in laminar coflowing nonpremixed methane-air flames

Ammonia Conversion and NOx Formation in Laminar Coflowing Nonpremixed Methane-Air Flames NEAL SULLIVAN†, ANKER JENSEN, and PETER GLARBORG

Department of Chemical Engineering, Technical University of Denmark, DK-2800, Lyngby, Denmark

MARCUS S. DAY, JOSEPH F. GRCAR*, and JOHN B. BELL

Center for Computational Sciences and Engineering, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA

CHRISTOPHER J. POPE

Combustion Research Facility, Sandia National Laboratories, Livermore, CA 94551, USA

and ROBERT J. KEE

Engineering Division, Colorado School of Mines, Golden, CO 80401, USA This paper reports on a combined experimental and modeling investigation of NOx formation in laminar, ammonia-seeded, nitrogen-diluted, methane diffusion flames. The methane-ammonia mixture is a surrogate for biomass fuels, which contain significant fuel-bound nitrogen. The experiments use flue-gas sampling to measure the concentration of stable species in the exhaust gas. The computations use adaptive mesh refinement (AMR) to capture fine-scale features of the flame. The model includes a detailed chemical mechanism, differential diffusion, buoyancy, and radiative losses. The model shows good agreement with the measurements over the full range of experimental NH3 seeding amounts. As more NH3 is added, a greater percentage is converted to N2 rather than to NO. The simulation results are analyzed to trace the changes in NO formation mechanisms with increasing amounts of ammonia in the fuel. © 2002 by The Combustion Institute

INTRODUCTION The dependence on combustion for meeting world energy demands has given rise to many harmful side effects, such as photochemical smog and “acid rain,” due in part to the emissions of the nitrogen oxides NO and NO2, collectively termed NOx . With increasing governmental regulation of pollutant emissions, the control and reduction of NOx is not only an environmental matter, but a financial one as well [1]. The problem is exacerbated by a desire to use a combination of national coal reserves and biomass power supplies to alleviate oil dependence. Coal and biomass fuels may contain significant amounts of chemically bound nitrogen—as much as 2% by mass. The nitrogenous gases that are released from these solid fuels during pyrolysis are converted in the flame *Corresponding author. E-mail: [email protected] † Currently at ITN Energy Systems, 8130 Shaffer Parkway, Littleton, CO 80127 USA. COMBUSTION AND FLAME 131:285–298 (2002) © 2002 by The Combustion Institute Published by Elsevier Science Inc.

to either N2 or NOx , depending on the local combustion conditions. The chemical form taken by volatile fuelnitrogen has little influence on the overall conversion rate to NO [2–7]. Typically, the nitrogen-containing compounds form either HCN or NH3. Although the fraction of each species remains the subject of research, ammonia is typically present in higher concentrations than other fuel-nitrogen species in volatiles from biomass feedstocks [7, 8]. Thus, we primarily use an ammonia-seeded fuel in the present study, but for a few experiments we use hydrogen cyanide seeding for comparison. The main chemical formation routes of NOx in flames are well established [9]. At temperatures exceeding 1800 K, atmospheric nitrogen oxidizes to NO through the thermal (or Zeldovich) mechanism. In the prompt (or Fenimore) mechanism, NO forms in the flame zone through reactions between molecular nitrogen and hydrocarbon radicals. In the ammonia mechanism (shown in Fig. 1, adapted from 0010-2180/02/$–see front matter PII S0010-2180(02)00413-3

286

N. SULLIVAN ET AL.

Fig. 1. Ammonia oxidation mechanism [50].

Miller and Bowman [9]), fuel-nitrogen species present as NH3 undergo hydrogen abstraction reactions. Each resulting NHi radical then participates in one of two subsequent reaction mechanisms: oxidation leading to NO formation, or conversion to N2 through reactions that additionally consume NO. Like most solid fuels, volatiles from biomass are burned in non-premixed flames. Many previous studies of NOx formation in diffusion flames focus on the relative roles of thermal and prompt formation routes under a variety of conditions [10 –13]. However, NOx from the combustion of biomass and other solid fuels is dominated by the conversion of significant amounts of fuel-bound nitrogen; thermal and prompt mechanisms are comparatively insignificant. Despite its importance, the formation of NOx from fuel-nitrogen in non-premixed flames has seen comparatively little investigation. In the present work we consider an axisymmetric, laminar, coflowing, non-premixed flame depicted in Fig. 2. This represents a reasonably complete model system for studying NOx formation during the combustion of volatiles from solid fuels. It affords evaluation of production pathways in a realistic flow field without introducing models for turbulence or compromising the fidelity of the chemical kinetics representation. Early models of this system assumed negligible axial diffusion (i.e., the “boundary layer” assumption). Burke and Schumann [14] demonstrated that such a model could accurately pre-

Fig. 2. Schematic of the laminar, coflowing, non-premixed flame.

dict diffusion flame heights. Roper et al. [15, 16] and Gordon et al. [17, 18] generalized this approach into a well-established flame analysis tool. Miller and Kee [19] combined the boundary layer assumption with detailed reaction kinetics to simulate a laminar, hydrogen-air, nonpremixed flame. Interestingly, they found that flame-zone radical concentrations may exceed equilibrium values by more than an order of magnitude. Super-equilibrium atomic O concentrations may strongly affect NOx formation [13, 20]. A non-premixed flame forms where there is a balance between chemical reactions and diffusive and convective transport. For lower-speed flows, where axial diffusion is comparable to the advective transport, Chung and Law [21] demonstrated that the boundary layer assumption leads to poor predictions for flame height and shape of diffusion flames. The experimental work of Ishizuka and Sakai [22] and numerical studies by Takagi and Xu [23] indicated that axial diffusion plays a key role in the diffusion flame’s structure. The work of Smooke et al. [24, 25] was apparently the first to successfully model the coflowing laminar non-premixed flame without making use of the boundary layer approximation. The 2-D model has served well as a tool to provide significant insight into the structure of diffusion flames [24 –31], including NO forma-

AMMONIA CONVERSION IN NON-PREMIXED FLAMES tion [12]. However, no study of fuel-nitrogen effects has yet been performed. Nishioka et al. [32] correlated flame structure from simulations of axisymmetric, coflowing, non-premixed flames with that of one-dimensional, opposed-flow, non-premixed flames through a “representative diffusion time.” The ability to capture the physics of this 2-D problem through a series of one-dimensional calculations represented a significant breakthrough and decrease in complexity. However, in a subsequent study, Zhu et al. [33] found that this correlation breaks down when considering NOx formation, indicating that emission characteristics from 2-D flames are different from those of 1-D flames. In this study, we investigate a laminar, ammonia-enriched, methane-air non-premixed flame through both experiment and computation. The focus of the work is to understand how the fuel-nitrogen routes for NO formation differ from those of flames without fuel-nitrogen. The simulations provide steady, axisymmetric spatially resolved species and thermal profiles for detailed analysis of the NO production pathways. Experimental data are used to validate global predictions of the model, and to provide a basis for comparing discrepancies between different detailed chemical mechanisms. EXPERIMENTAL SETUP The flame depicted in Fig. 2 is established in a quartz reactor of dimensions ␦ ⫽ 1 mm, rf ⫽ 6 mm, rox ⫽ 14 mm, and height 760 mm. Long entrance lengths are provided for the inlet gases so that fully developed flow is established in the fuel and oxidizer tubes upstream of the fuel nozzle. The two streams are mixed through entrainment and diffusion, and for suitable fuel and oxidizer flow rates, a stable flame is produced. The fuel is a mixture of methane and nitrogen at flow rates of 150 and 220 mL/min, respectively. Ammonia is added to the fuel stream in amounts ranging from 0 to 1000 ppm of the total fuel-oxidizer inflow. In selected experiments with fuel-N doping levels below 150 ppm, the NH3 is replaced with HCN to observe the effects of fuel-N speciation on NO formation. The oxidizer is a mixture of research grade

287

oxygen and nitrogen at flow rates of 840 mL/min and 3160 mL/min, respectively (21% oxygen). The nitrogen dilution in the fuel stream results in a relatively cool flame, where peak temperatures rise only slightly above the 1800 K required for significant thermal NO production. Flue gases from the reactor are dehumidified through a water trap upstream of Hartmann & Braun gas analyzers for measurement of O2, NO, CO2, and CO concentration. THE COMPUTATIONAL MODEL The flame is assumed to be axisymmetric so the model has two spatial dimensions, axial z and radial r. The computational domain extends 11 cm downstream from z ⫽ 0 at the exit edge of the fuel tube. This is shorter than the 76-cm quartz reactor, but it is much longer than the flame as indicated by peak temperature. Inflow boundary conditions specify fully developed laminar pipe flow, or co-annular flow, for the fuel and oxidizer, respectively (Refuel ⫽ 41, Reox ⫽ 127). No-slip conditions apply along the outer wall and at the fuel tube edge. A non-reflecting outflow boundary is enforced at z ⫽ 11 cm. The outer wall of the domain has a fixed piecewiselinear temperature profile obtained by experimental measurement (rising from 500 K at the inlet to 800 K at 6 cm, then dropping to 300 K at 50 cm). The fuel tube edge is assumed to be at 300 K. All gases enter the system at standard temperature and pressure. We use the computational approach presented by Day and Bell [34]. This consists of a time-dependent, low Mach number model that includes buoyancy, mixture-averaged differential diffusion [35] in both spatial coordinates, and optically thin (emission-only) radiation [36]. The numerical discretization is based on a robust projection formulation that accommodates large density contrasts. The algorithm uses an operator-split treatment of stiff reaction terms. The basic computational approach is embedded in an adaptive mesh refinement (AMR) framework that uses structured hierarchical grids to concentrate mesh points near the flame sheet. The integration algorithm subcycles in time and preserves the discrete conservation properties of the underlying, single-grid

288

Fig. 3. Measured and computed flue-gas NO concentrations versus NH3 seeding as a fraction of the total fuel-air inflow (ammonia is added only to the fuel stream). Error bars on the experimental data represent ⫾6␴ (␴ is the SD). The reduced-model curve and the GRI data are discussed later.

algorithm. Details of the discretization and implementation are discussed in Day and Bell [34]. The flow is “ignited” by choosing initial, t ⫽ 0, temperature conditions that have a hot zone (T ⫽ 2000 K) across the fuel and oxidizer interface at z ⫽ 0. The solution then evolves in time until steady-state conditions are achieved. Convergence is monitored by the axial profile of the radially integrated NO profile. The simulations are performed using two different chemical mechanisms. The first is a mechanism proposed by Glarborg et al. consisting of 66 species and 447 reactions [37]. It contains the oxidation of C1 and C2 hydrocarbons, HCN, and NH3, with a subset describing interactions between hydrocarbons and nitrogenous species. The second is the GRI 3.0 mechanism containing 53 species and 325 chemical reactions [38]. COMPARISON OF DATA AND SIMULATION RESULTS In Fig. 3, we present experimental and computed values for the flue-gas NO concentration for various amounts of ammonia in the fuel stream. The experimental values have been corrected for the water removal upstream of the gas analyzers by assuming stoichiometric H2O production in the flame zone. With no ammonia added to the fuel stream,

N. SULLIVAN ET AL. the methane-air non-premixed flame is found to produce 25 ppm of NO. This level increases to 297 ppm NO in flue gases when the inflow contains 1000 ppm NH3. While the flue gas NO concentration increases with ammonia addition, the conversion rate of NH3 to NO decreases from over 50% at [NH3]in ⱕ 100 ppm to less than 30% at [NH3]in ⱖ 800 ppm. Similar behavior has been observed in previous fuel-nitrogen studies [3]. Experiments with NH3 replaced by HCN show no difference in NO formation. This is in agreement with results from the literature, which indicate that the speciation of the volatile nitrogen compounds does not have a significant effect on the NO yield, both in lean premixed flames and in diffusion flames [7]. Results from the simulation are shown for three amounts of fuel ammonia: 0, 500 ppm, and 1000 ppm NH3 of the combined fuel-oxidizer inflow of 4.37 liter/min. As a convenience for subsequent discussion, Fig. 3 also includes data derived from a reduced model and from simulations incorporating alternative chemistry mechanisms that are discussed later. Generally, the agreement between experiment and modeling is good at all three ammonia concentrations simulated. The Glarborg et al. mechanism overall provides the more accurate and consistent NO predictions. The model results using the GRI 3.0 mechanism overpredict the flue gas NO concentration for the ammonia-enriched flames by as much as 30% in the 1000 ppm case. We examine the source of these differences later. ANALYSIS OF COMPUTATIONAL RESULTS Flame Structure The results in this section apply to all the flames investigated. The NH3 content of the fuel stream is found to have no significant effect on the overall flame structure in the computational model. General agreement with previous studies (particularly [12] and others cited below) and the comparison in the previous section with experimentally determined flue gas NO concentration provide a measure of model validation. The following discussion is based on the model results using the Glarborg et al. mechanism.

AMMONIA CONVERSION IN NON-PREMIXED FLAMES

289

Fig. 5. Carbon reaction paths (Glarborg et al. mechanism). The thickness of an arrow indicates the quantity (mol/s) of atomic carbon moving through the path; only paths at least 4% of the greatest are shown. Table 1 gives the percent of each path as a result of various reactions.

Fig. 4. a) Temperature (K) and b) heat release (ergs/cm3 sec) with superimposed advection streamlines. A non-linear scale is used for the heat release contours to highlight the triple flame structure and the endothermic zone.

Figure 4a shows temperature in the lower part of the computational domain. The flame is lifted ⬃0.1 cm above the edge of the fuel tube, and is anchored at an ignition point located at z ⬇ 0.1 cm, r ⬇ 0.7 cm. Based on peak temperature, the flame length extends to z ⫽ 3.6 cm (Tmax ⫽ 1847 K). Figure 4b shows heat release and superimposed convection streamlines that reveal entrainment of oxidizer fluid into the fuel stream. From the ignition zone, a rich premixed flame extends on the fuel side, while a lean premixed flame burns on the oxidizer side. A non-premixed flame lies between the two, located roughly at the stoichiometric line. This triple-flame structure has been noted in previous studies [26, 27]. A weakly endothermic region lies between the rich premixed flame and the non-premixed flame. This is the result of two hydrocarbon dissociation reactions, C2Hn⫹1(⫹M) º C2Hn ⫹ H(⫹M),

n ⫽ 2, 4.

Endothermicity at the flame tip because of acetylene formation has been observed in the flames simulated by Ern et al. [27]. The chemical mechanism used in that work did not include ethylene, which increases the area and magnitude of the endothermic zone in this simulation. The formation of these species has been experimentally confirmed by Gordon et al. [39]. Reaction path analysis reveals the carbon oxidation system shown in Fig. 5. The Appendix describes how the diagram is generated from the simulation. The overall chemical pathways are similar to those observed in premixed flame simulations [40, 41]. Although there is significant activity involving C2 hydrocarbons, the bulk of the methane is converted through the C1 path. Ammonia Oxidation and NOx Formation Ammonia-Free Flame Nitrogen reaction pathways for the ammoniafree flame are shown in Fig. 6. The most important mechanism for NO formation in this flame is prompt NO, initiated by the reaction CH⫹N2 3 HCN ⫹ N. Thermal NO, initiated by N2 ⫹ O 3 N ⫹ NO, is much less important, as the flame’s peak temperature is barely high enough to support this mechanism. The relative contributions (77% vs. 17%) of these two reactions to the path N2 3 N indicate the relative importance of the prompt and thermal mechanisms in this flame. In addition to prompt NO and thermal NO, mechanisms involving N2O yield a considerable amount of NO. The nitrous oxide is formed mostly from the reaction N2 ⫹ O ⫹ M 3 N2O ⫹ M, but the sequence N2 ⫹ H 3 NNH and NNH ⫹ O 3 N2O ⫹ H also contributes.

290

N. SULLIVAN ET AL. TABLE 1 Composition of the Carbon Reaction Paths Depicted in Fig. 5

(100) CH4 3 CH3 68% ⫹ H 26% ⫹OH (77) CO 3 CO2 98% ⫹OH (50) HCO 3 CO 78% ⫹M (47) CH2O 3 HCO 52% ⫹OH 34% ⫹H 10% ⫹CH3 (33) C2H4 3 C2H3 80% ⫹H 19% ⫹OH

(29) C2H3 3 C2H2 65% ⫹M 15% ⫹H 13% ⫹CH3 (28) C2H5 3 C2H4 99% ⫹M (20) CH2OH 3 CH2O 83% ⫹M 13% ⫹ O2 (20) CH3 3 CH4 53% ⫹H⫹M 23% ⫹CH2O (18) CH3 3 CH2OH 100% ⫹OH

(17) C2H2 3 HCCO 100% ⫹O (16) CH3 3 CH2O 98% ⫹O (15) CH3 3 C2H6 100% ⫹CH3 (15) C2H6 3 C2H5 64% ⫹H 20% ⫹OH 11% ⫹ CH3 (13) CH3 3 C2H5 100% ⫹CH3

(9) HCCO 3 CO 57% ⫹H 26% ⫹O2 (8) CH2 (S) 3 CH2 60% ⫹ N2 29% ⫹H2O (7) CH3 3 CH2(S) 79% ⫹OH 20% ⫹H (5) HCCO 3 CH2(S) 100% ⫹H (5) HCCO 3 C2O 100% ⫹OH

(4) CH2 3 C2H4 100% ⫹CH3 (4) CH2 3 CH2O 74% ⫹OH 24% ⫹CO2 (4) C2O 3 CO 58% ⫹O2 26% ⫹OH (4) CH3 3 C2H4 98% ⫹CH2 (4) CO2 3 CO 32% ⫹H 31% ⫹CH2(S) 25% ⫹CH2

Paths are listed by decreasing amount of atomic carbon (mol/s) moving through them; only paths at least 4% of the greatest appear here and in the figure. The percent of each path because of various reactions is shown; only contributions of at least 10% are included.

Subsequently, N2O is largely recycled to N2, but a minor fraction reacts with H or O to form NO. As can be seen from Fig. 6, the bulk of the NO is formed from oxidation of atomic nitrogen, with significant contribution from oxidation

of HNO and NH. Interestingly, most of the N is not formed directly from N2, but rather is derived from NH that in turn is formed from carbon-bearing species originating in HCN. Figure 7 shows the mole fraction and the net production of NO in the lower part of the computational domain. The right side reveals a two-layer structure which extends from the ignition zone to nearly the flame tip and remains roughly parallel to the bulk convection. That NO is produced on the lean side of the flame and is consumed on the rich side has been observed in previous non-premixed flame studies [13, 20, 42]. Transverse to the flow, nitrogen atoms in the species NO, HCN, NCO, NH, N, and HNO, move back and forth between the layers of NO production and consumption. This “recycling” can be observed as the closed loop in Fig. 6. The consumption of NO diffusing toward the fuel stream occurs through two reactions: NO ⫹ HCCO º HCN ⫹ CO2, NO ⫹ CH2 º HCN ⫹ OH.

Fig. 6. Nitrogen reaction paths for the flame without ammonia seeding (Glarborg et al. mechanism). Note the continuous recycling of nitrogen: starting in the form of NO, it passes through carbon-bearing species to NH and finally back to NO. The thickness of an arrow indicates the quantity (mol/s) of atomic nitrogen moving through the path; only paths at least 0.8% of the greatest are shown. Table 2 gives the percent of each path as a result of various reactions.

Some of the HCN produced in these reactions accumulates in the fuel stream and flows upward toward the flame tip. Above z ⫽ 3 cm, diffusion from the production zone to the consumption layer must compete with the upward fluid advection. The remaining HCN is converted to NO, which then peaks in concentration along the centerline just above the flame

AMMONIA CONVERSION IN NON-PREMIXED FLAMES

Fig. 7. NO a) mole fraction and b) net production rate (mole/cm3 sec) for the ammonia-free methane-air flame.

tip. Beyond this location, NO diffuses across the domain, and is carried out the flue. Comparison of this mechanism with the prompt mechanism from Miller and Bowman [9] reveals interesting differences. While the initiation reaction is identical, production of NO by the HNO route is absent in [9]. Virtually all of the atomic nitrogen is converted to NO, with negligible N2 formation. This is likely because of the low NO concentration in this ammonia-free flame, minimizing the effect of the N ⫹ NO 3 N2 ⫹ O reaction. Finally, extensive recycling reactions through cyano-species are captured in the Glarborg et al. mechanism. In view of Fig. 6, these reactions play a significant role and lead to a more complex collection of reactions paths than the prompt mechanism shown in [9]. NO Formation in the Ammonia-Containing Flame Figure 8 displays the NO concentration and net production rate with 1000 ppm NH3 added to the fuel stream. Generally, nitrogen species

291

Fig. 8. NO a) mole fraction and b) net production rate (mole/cm3s) for the 1000 ppm ammonia-seeded flame.

concentrations and production rates increase by an order of magnitude, as does the final, flue gas NO concentration (from 25 ppm to 297 ppm). In comparison with Fig. 7a, a much different NO field is observed. The peak in concentration shifts from the area around the flame tip to the ignition region at the flame base. (Planar laserinduced fluorescence images appear in a subsequent paper [43].) NO formation in the ignition region is an order of magnitude higher than at axial positions z ⱖ 1 cm. From Figs. 7b and 8b, whether or not ammonia is added, it is clear that net NO production is on the lean side of the flame zone and net NO consumption is on the rich side. Diffusion to the fuel stream causes the NO concentration to reach nearly a third of its peak value along the centerline. Notice that here, the reactions with HCCO and CH2 are not strong enough to significantly consume the fuel-side NO profile; the pool of hydrocarbons for these reactions has not changed from that of the ammonia-free flame. In Fig. 8a, a mild NO reduction zone is evident in the interior of the flame just below the flame tip from reactions with HCCO.

292

N. SULLIVAN ET AL. TABLE 2 Composition of the Nitrogen Reaction Paths Depicted in Fig. 6 for the Flame without Ammonia Seeding

(100) N2 3 NNH 51% ⫹H⫹O2 48% ⫹H (96) NNH 3 N2 91% ⫹O2 (26)NO2 3 NO 77% ⫹H 18% ⫹O (24) NO 3 NO2 66% ⫹O⫹M 33% ⫹HO2 (15) HNO 3 NO 49% ⫹H 45% ⫹OH (14) N2 3 N2O 100% ⫹O⫹M (14) N2O 3 N2 96% ⫹H (12) NO 3 HNO 80% ⫹H⫹M 11% ⫹H⫹N2

(9) N 3 NO 80% ⫹OH 11% ⫹O2 (8) NO 3 HCN 57% ⫹HCCO 22% ⫹CH2 (6) NH 3 N 68% ⫹H 32% ⫹OH (5) NCO 3 NH 100% ⫹H (5) HCN 3 NCO 100% ⫹O (4) HNCO 3 NH2 100% ⫹H (3) NH2 3 NH 55% ⫹H 40% ⫹OH (3) NCO 3 HNCO 70% ⫹H2O 21% ⫹H2

(3) CN 3 NCO 66% ⫹OH 31% ⫹O2 (3) NO 3 HCNO 68% ⫹CO 32% ⫹H (3) NH 3 HNO 96% ⫹OH (2) NH 3 NO 98% ⫹O (2) HCN 3 CN 49% ⫹H 44% ⫹OH (2) HCNO 3 NO 79% ⫹OH 21% ⫹H (2) HOCN 3 NCO 96% ⫹H (2) NO 3 HONO 100% ⫹OH⫹M (2) HONO 3 NO2 91% ⫹OH

(2) N2 3 N 77% ⫹CH 17% ⫹O (2) HCN 3 HOCN 100% ⫹OH (2) H2CN 3 HCN 100% ⫹M (2) N2 3 HCN 100% ⫹CH (1) N 3 H2CN 100% ⫹CH3 (1) NNH 3 N2O 100% ⫹O (1) NCO 3 NO 99% ⫹O (1) N2O 3 NO 76% ⫹H 23% ⫹O (1) HCN 3 NH 100% ⫹O (1) CH2CN 3 CN

(1) HCN 3 CH3CN 100% ⫹CH3 (1) CH3CN 3 CH2CN 87% ⫹H 12% ⫹OH (1) N2O 3 NH 99% ⫹H (1) CN 3 HCN 39% ⫹CH4 33% ⫹C2H4 (0.9) NO 3 N 61% ⫹C 38% ⫹CH (0.8) NO 3 CN 55% ⫹C2H 44% ⫹C (0.8) NH2 3 NH3 42% ⫹H⫹M 41% ⫹H2O 15% ⫹H2

98% ⫹O

Paths are listed by decreasing amount of atomic nitrogen (mol/s) moving through them; only paths at least 0.8% of the greatest appear here and in the figure. The strongest paths (above the line in the table) are opposed and net to weaker paths. The percent of each path because of various reactions is shown; only contributions of at least 10% are listed.

The nitrogen reaction paths are shown in Fig. 9. Despite the complexity of this diagram, comparison of the NO formation routes with the ammonia oxidation mechanism in Fig. 1 shows excellent agreement. It is noteworthy, however, that the NNH route shown in Fig. 1 is insignificant in this flame. As seen in the reaction pathways, atomic nitrogen both produces and consumes NO. In fuel rich zones, atomic nitrogen consumes NO through the reaction N ⫹ NO 3 N2 ⫹ O, while in fuel lean zones, atomic nitrogen is oxidized to form NO through reactions with OH. In the ignition region, the majority of the atomic nitrogen is oxidized to NO while less than 20% forms N2, consuming only a fraction of NO in the process. Comparison of the NOx formation pathways with and without ammonia present in the fuel stream (Figs. 9 and 6, respectively) reveals some similarities. In both cases, the bulk of the NO is formed either from HNO reactions with H and OH, or from N reactions with OH. The NO recycling reactions through cyano species are identical. The prompt-NOx chemistry of Fig. 6 is

also active in the ammonia-containing flame. However, the production of HCN and N from fuel-nitrogen routes in the ignition region are two orders of magnitude higher than that from the prompt route, limiting the significance of prompt NOx chemistry in the ammonia-containing flame. Finally, the conversion of atomic nitrogen to N2 through reactions with NO is insignificant in the flame without ammonia, while it is quite significant in the ammonia-containing flame. This is because of the comparatively low NO concentration in the pure methane-air flame. The activity of the N ⫹ NO 3 N2 ⫹ O reaction increases with increasing ammonia concentration resulting in the non-linear ammonia conversion rate observed in Fig. 3. Reduced Model Analysis of the Declining Efficiency of NO Production The results presented here and in previous studies [3] show that NO production increases with declining efficiency as NH3 is added to the

AMMONIA CONVERSION IN NON-PREMIXED FLAMES

293

be useful in elucidating the behavior of trace species in other reacting flows. The mass conservation equation for the k-th species is ⭸␳Yk ⫹ ⵜ 䡠 ␯ជ ␳ Y k ⫽ ⵜ 䡠 ␳ D kⵜY k ⫹ M k␻ ˙k ⭸t in which ␳ is the mass density, Yk is the mass fraction, ␯ជ is the velocity, Dk is a mixture-averaged diffusion coefficient, Mk is the molar mass, and ␻ ˙k is the molar rate of formation. The latter is given by

␻ ˙k ⫽

fuel stream. It is worth noting that this is a second-order effect, so it cannot be explained by sensitivity analysis, which relies on first-order derivatives. Also, except at very high fuel-N levels [44] it is not observed in premixed flames [3], so it cannot be explained by studying the chemical mechanism alone. Evidently the declining efficiency of NO production depends on chemistryfluid interactions in non-premixed flames. A reduced model that more simply describes the conversion of NH3 to NO can identify the reactions involved. The following analysis, which simplifies the entire chemical-fluid simulation in a way that preserves second-order information, may

out

册 冋冕

共nជ 䡠 ␯ជ 兲 ␳ Y k ␲ k ⫹

in

册 冘冋

共nជ 䡠 ␯ជ 兲 ␳ Y k ␲ k ⫽

i

⫺

冕

冉 写C

共f兲 vi,k 兲 ki共 f 兲

冕

共nជ 䡠 ␯ជ 兲␳Yk⫹

out

共f 兲

vi,l l

l

⫺ ki共r兲

写C 冊 共r兲

l

vi,l l

冕

共nជ 䡠 ␯ជ 兲␳Yk⫽

in

Mk␻ ˙k

vol

where nជ is the exterior unit normal and the orifices have been segregated into outflow and inflow surfaces. If each species in some group of species is subject to a multiplicative perturbation, ␲k ⬇ 1, which is uniform throughout the chamber, and if other quantities remain fixed, then the multiplicative perturbations can be pulled outside of the concentrations, densities, and integrals to leave the following equation. 共r兲 共f兲 M k共v i,k ⫺ v i,k 兲

冕

i

冋

册

k i共 f 兲

写C 写␲

冕

k i共r兲

vol

冘

The quantities in square brackets are fixed and can be determined from the simulation. Upon replacing these quantities by specific numbers,

i

共r兲 i,k ⫺

in which ki is the rate constant of the i-th reaction, ␯i,k is the stoichiometric coefficient of the k-th species in the i-th reaction, Cl is the molar concentration of the l-th species, and the superscripts (f) and (r) indicate forward or reverse. Suppose the reacting fluid is in steady-state so that the time derivatives vanish, and suppose further the fluid is in a chamber with vanishing composition gradients normal to any orifices. Then the conservation equations can be integrated over the volume of the chamber to leave

Fig. 9. Nitrogen reaction paths for the 1000 ppm ammoniaseeded flame (Glarborg et al. mechanism). The thickness of an arrow indicates the quantity (mol/s) of atomic nitrogen moving through the path; only paths at least 4% of the greatest are shown. Table 3 gives the percent of each path as a result of various reactions.

冋冕

冘共v

共r兲 共f兲 M k共v i,k ⫺ v i,k 兲

vol

共f兲

l

v i,l l

l

册

共f兲

v i,l l

写C 写␲ 共r兲

l

v i,l l

l

共r兲

v i,l l

(1)

there results a system of algebraic equations that can be solved for the multiplicative perturbations, ␲k. In the present situation the species

294

N. SULLIVAN ET AL. TABLE 3 Composition of the Nitrogen Reaction Paths Depicted in Fig. 9 for the Flame with Ammonia Seeding

(100) NO2 3 NO 79% ⫹H 15% ⫹O (89) NO 3 NO2 63% ⫹M 36% ⫹OH

(40) NO 3 HNO 74% ⫹H⫹M 11% ⫹H⫹N2 10% ⫹HCO (33) N 3 NO 78% ⫹OH

(17) NO 3 N2 58% ⫹N 23% ⫹NH2 16% ⫹NH (16) NH 3 NO 96% ⫹O

(12) NO 3 HCNO 61% ⫹CO 38% ⫹H (11) CN 3 NCO 61% ⫹OH 36% ⫹O2

(9) N 3 H2CN 100% ⫹CH3 (9) HOCN 3 NCO 96% ⫹H (8) HCN 3 HOCN 100% ⫹OH

(74) NH3 3 NH2 65% ⫹OH 28% ⫹H (72) NH2 3 NH 60% ⫹H 35% ⫹OH (70) HNO 3 NO 49% ⫹H 39% ⫹OH (48) NH 3 N 72% ⫹H 27% ⫹OH

13% ⫹O2 (32) NO 3 HCN 51% ⫹HCCO 26% ⫹CH2 (21) NH 3 HNO 94% ⫹OH (20) NCO 3 NH 100% ⫹H (20) NNH 3 N2 71% ⫹O2 18% ⫹ (18) HCN 3 NCO 100% ⫹O

(16) N2O 3 N2 86% ⫹H 10% ⫹M (14) HNCO 3 NH2 100% ⫹H (13) NCO 3 HNCO 68% ⫹H2O 19% ⫹H2 (12) N2 3 NNH 53% ⫹H⫹O2 46% ⫹H (12) H2CN 3 HCN 100% ⫹M

(11) NO 3 HONO 100% ⫹OH⫹M (11) HONO 3 NO2 89% ⫹OH (11) HCNO 3 NO 74% ⫹OH 25% ⫹O (10) HCN 3 CN 48% ⫹H 44% ⫹OH (10) N 3 N2 96% ⫹NO (10) NO 3 N2O 97% ⫹NH (10) NH 3 N2O 99% ⫹NO

(7) NH2 3 HNO 100% ⫹O (5) NCO 3 NO 99% ⫹O (5) CH2CN 3 CN 99% ⫹O (5) HCN 3 CH3CN 100% ⫹CH3 (4) HCN 3 NH 100% ⫹O (4) CH3CN 3 CH2CN 86% ⫹H 13% ⫹OH (4) NH2 3 N2 93% ⫹NO

The strongest paths (above the line in the table) are opposed and net to weaker paths. Paths are listed by decreasing amount of atomic nitrogen (mol/s) moving through them; only paths at least 4% of the greatest appear here and in the figure. The percent of each path because of various reactions is shown; only contributions of at least 10% are listed.

of interest are those except N2 that contain nitrogen. They are present in such small quantities that fluctuations in them would not be expected to alter the velocity and temperature fields. That is, the rate constants ki(f) and ki(r) are insensitive to the nitrogen species, other than N2, so the model reduction described above applies to them. To apply the reduced model’s equations, (1), we evaluate the quantities in square brackets from the simulation with 1000 ppm of NH3 flowing in. The equation for ammonia is set aside, and the variable ␲NH3 is treated as a free parameter. Altogether there are 23 equations and variables for the nitrogen species besides N2 and NH3. The equations are solved using Mathematica [46] to express ␲k for the remaining nitrogen species as a function of ␲NH3. The exact analytic formulas found by Mathematica are quite complicated. However, the following low-order Pade´ approximation for the NO dependent variable is almost indistinguishable from the analytic solution.

␲ NO ⫽

2.902 ␲2NH3 ⫹ 4.084 ␲ NH3 ⫺ 0.017

␲2NH3 ⫹ 4.125 ␲ NH3 ⫹ 1.844

(2)

The reduced model can be used in a variety of ways to obtain information about the chemicalfluid system. The expression

冋冕

out

册

共nជ 䡠 ␯ជ 兲 ␳ Y NO ␲ NO

is the reduced model’s prediction for the amount of NO flowing out of the reaction chamber. The independent variable, ␲NH3, can be similarly scaled to the desired physical units. A graph of this prediction is shown in Fig. 3 along with the experimental data. Note that the reduced model matches the predictions of the reacting flow simulation at 1000 ppm NH3, as it should, since this is where the reduced model was derived. Moreover, the good agreement with experimental data down past 500 ppm NH3 is remarkable given the extent of consolidation in creating the reduced model. This indicates

AMMONIA CONVERSION IN NON-PREMIXED FLAMES

295

Fig. 10. (left) The reduced model’s predictions of NO consumption by individual (forward or reverse) reactions in the Glarborg et al. mechanism. The 15 strongest consumers are identified in the legend (center). Five reactions (solid curves) accelerate their consumption of NO with increased NH3 seeding. (right) When the amount of NO consumed by these reactions is restored to the net NO production, the total has a linear variation with the amount of NH3 seeding.

that the assumptions underlying the reduced model remain valid for relatively large changes in NH3. Near 0 ppm NH3 the ammonia chemistry is insignificant, so the reduced model of NH3 oxidation fails to predict NO production there. The reduced model identifies the reactions responsible for the declining efficiency of NO production. Information about individual reactions is readily available from the reduced model’s formulas. For the 75 reactions in the Glarborg et al. mechanism that consume NO (forward and reverse directions are considered separate), Fig. 10 (left) plots their consumption of NO as a function of NH3 seeding. Five reactions can be seen to accelerate their consumption of NO with increased seeding; they are the reactions in which NO combines with another nitrogen species. In order of declining strength they are: N ⫹ NO 3 N2 ⫹ O, NH ⫹ NO 3 N2O ⫹ H, NH ⫹ NO 3 N2 ⫹ OH,

progress. Figure 10 (right) shows that the amount of NO consumed by these five reactions is significant because it is comparable to the flame’s net production. If the consumption by these five reactions is added back to the net NO production, then the figure shows the resulting quantity varies linearly with NH3 seeding. Thus these reactions account for the non-linearity of the NH3 to NO conversion. Comparison of NOx Formation Between Mechanisms Reaction pathway analysis of simulations using the GRI 3.0 mechanism provide similar results to those shown in Figs. 6 and 9 for the Glarborg et al. mechanism, although some differences are evident. The most significant difference between the mechanisms is the level of conversion of NH to HNO. In the Glarborg et al. mechanism, this occurs through reaction with hydroxyl radical, with about 20% of the NH being converted to HNO. Using GRI 3.0, two additional NH 3 HNO conversion routes exist: Rx. 195 NH ⫹ H2O º HNO ⫹ H2

NH2 ⫹ NO 3 N2 ⫹ H2O,

k ⫽ 2 䡠 1013 exp(⫺13850/RT)

NH2 ⫹ NO 3 NNH ⫹ OH.

Rx. 278 NH ⫹ CO2 º HNO ⫹ CO

For each of these reactions, increased NH3 seeding increases the concentrations of both reactants, which by mass action kinetics, quadratically increases the reaction’s rate of

k ⫽ 1 䡠 1013 exp(⫺14350/RT) Reaction numbers are given in the order listed in the GRI 3.0 mechanism. These two addi-

296 tional reactions increase the NH 3 HNO conversion by a factor of two, which comes at the cost of the NH 3 N conversion. For both mechanisms, virtually all HNO is converted to NO, while atomic nitrogen may produce or consume NO depending upon local combustion conditions. This favoring of HNO over N leads to the higher NO concentrations predicted by the GRI 3.0 mechanism. The validity of the rate expressions for the two reactions, above, has been called into question in previous studies [47]. To assess the claims suggested in the reference, three simulations were performed using the GRI 3.0 mechanism with reactions 195 and 278 removed; results are included in Fig. 3 as the modified GRI mechanism. With 1000 ppm of ammonia in the fuel stream, the agreement between simulation and experiment improves significantly. This dramatic effect is not observed in the ammoniafree flame; the removal leads to a slight increase (5 ppm) in flue gas NO concentration. Nearly all atomic nitrogen and nitroxyl are oxidized to NO (Fig. 6) so that conversion of NH to either species in ammonia-free flames yields the same NO production. Further comparison of mechanisms using the PREMIX flame code [48] reveals that the removal of these reactions results in only a 3% decrease in NO concentration. This implies that the strong effect of these reactions may be limited to non-premixed flames.

CONCLUSIONS In this study, the conversion of volatile fuelnitrogen species is investigated through a series of experiments and computations involving a laminar, coflowing, non-premixed flame. In one series of experiments, increasing concentrations of ammonia are added to the fuel stream of a methane-air flame. The conversion efficiency of ammonia to NO is found to decrease from over 50% at low ammonia concentration to less than 30% at higher ammonia concentration. The measured data are compared with simulation results for three different ammonia concentrations using two different chemical mechanisms. Better agreement between experiment and

N. SULLIVAN ET AL. model is found using the Glarborg et al. mechanism. A two-zone NOx formation structure is observed in the ammonia-free flame. At low axial positions, NOx forms primarily via the prompt mechanism, with significant production through the HNO intermediate. NO is also formed in minor amounts through N2O and by the thermal mechanism. Thermal NO formation is limited because of the low temperatures (1800 K) found in this flame. Significant NO consumption occurs on the fuel side of the flame through reactions with hydrocarbon radicals. The resulting HCN is advected upward along the axis, and converted back to NO as it crosses at the flame tip. In the ammonia-seeded case, much more NO is produced than can be consumed by the available CH2 and HCCO. Moreover, for the larger ammonia seeding rates, a greater fraction of the NO produced is converted to N2. Simulations using both the Glarborg et al. and the GRI 3.0 chemical mechanisms match experimental flue gas measurements quite well. The Glarborg et al. mechanism produces more accurate and consistent results; its performance is attributed to the differences in the HNOformation reactions and the more complete NO-recycling chemistry. The work of Sullivan, Glarborg and Jensen was supported by the research program CHEC (Combustion and Harmful Emission Control), which is co-funded by the Danish power companies Elsam and Energi E2, and by the Danish Ministry of Energy. The work of Day, Grcar and Bell was supported by the Applied Mathematical Sciences Program of the DOE Office of Mathematics, Information, and Computational Sciences, under contract DEAC03-76SF00098.

REFERENCES Bowman, C. T., Proc. Comb. Inst. 24:859 – 878 (1992). Fenimore, C. P., Combust. Flame 19:289 –296 (1972). Sarofim, A. F., Williams, G. C., Modell, M., and Slater, S. M., AlChE Symp. Ser. 71(148):51– 61 (1975). 4. Pfefferle, L. D., and Churchill, S. W., Combust. Sci. Techn. 49:235–249 (1986). 5. Sarv, H., and Cernansky, N. P., Combust. Flame 76: 265–283 (1989).

1. 2. 3.

AMMONIA CONVERSION IN NON-PREMIXED FLAMES 6. 7. 8. 9. 10. 11. 12.

13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.

29.

30. 31.

32. 33. 34. 35.

Martin, R. J., and Brown, N. J., Combust. Flame 80:238 –255 (1990). Glarborg, P., Jensen, A., and Johnsson, J. E., Prog. Energy Combust. Sci., (2002) in press. Johnsson, J. E., Fuel 73:1398 –1415 (1994). Miller, J. A., and Bowman, C. T., Prog. Energy Combust. Sci. 15(4):287–338 (1989). Drake, M. C., and Blint, R. J., Combust. Flame 76:151– 167 (1989). Drake, M. C., and Blint, R. J., Combust. Flame 83:185– 203 (1991). Smooke, M. D., Ern, A., Tanoff, M. A., Valdati, B. A., Mohammed, R. K., Marran, D. F., Long, M. B. Proc. Comb. Inst. 26:2161–2170 (1996). Smyth, K. C., Combust. Sci. Tech. 115:151-176 (1996). Burke, S. P., and Schumann, T. E. W., Ind. Engin. Chem. 20(10):998 (1928). Roper, F. G., Combust. Flame 29:219 –226 (1977). Roper, F. G., Smith, C., and Cunningham, A. C., Combust. Flame 29:227–234 (1977). Gordon, A. S., Li, S. C., Libby, P. A., and Williams, F. A., Combust. Sci. Tech. 100:395–399 (1994). Gordon, A. S., Li, S. C., and Williams, F. A., Combust. Sci. Tech. 141:1–18 (1999). Miller, J. A., and Kee, R. J., J. Phys. Chem. 81(25): 2534 –2541 (1977). Turns, S. R., Prog. Energy Combust. Sci. 21:361–385 (1995). Chung, S. H., and Law, C. K., Combust. Sci. Tech. 37:21– 46 (1984). Ishizuka, S., and Sakai, Y., Proc. Comb. Inst. 21:1821– 1828 (1986). Takagi, T., and Xu, Z., Combust. Flame 96:50 –59 (1994). Smooke, M. D., Mitchell, R. E., and Keyes, D. E., Combust. Sci. Tech. 67:85 (1989). Smooke, M. D., Lin, P., Lam, J. K., and Long, M. B., Proc. Comb. Inst. 23:575–582 (1990). Xu, Y., Smooke, M. D., Lin, P., and Long, M. B., Combust. Sci. Tech. 90:289 (1993). Ern, A., Douglas, C. C., and Smooke, M. D., Int. J. Sup. Appl. 9:167–186 (1995). Smooke, M. D., McEnally, C. S., Pfefferle, L. D., Hall, R. J., and Colket, M. B., Combust. Flame 117:117–139 (1999). McEnally, C. S., Pfefferle, L. D., Schaffer, A. M., Long, M. B., Mohammed, R. K., Smooke, M. D., and Colket, M. B., Proc. Comb. Inst. 28:2063–2070 (2000). Bennett, B. A. V., McEnally, C. S., Pfefferle, L. D., and Smooke, M. D., Combust. Flame 123:522–546 (2000). Bennett, B. A. V., McEnally, C. S., Pfefferle, L. D., Smooke, M. D., and Colket, M. B., Combust. Flame 127:2004 –2022 (2001). Nishioka, M., Takemoto, Y., Yamashita, H., and Takeno, T., Proc. Comb. Inst. 26:1071–1077 (1996). Zhu, X. L., Nishioka, M., and Takeno, T., Proc. Comb. Inst. 27:1369 –1376 (1998). Day, M. S., and Bell, J. B., Combust. Theory Modelling 4:535–556 (2000). Kee, R. J., Dixon-Lewis, G., Warnatz, J., Coltrin, M. E., and Miller, J. A. “A FORTRAN computer code

36. 37. 38.

39. 40.

41. 42. 43.

44. 45.

46.

47.

48.

49. 50.

297

package for the evaluation of gas-phase multicomponent transport properties.” Sandia National Laboratories Report No. SAND86-8246. Baxlow, R. S., http://www.ca.sandia.gov/tdf/Workshop/ Submodels.htmi. Glarborg, P., Alzueta, M. U., Dam-Johansen, K., and Miller, J. A., Combust. Flame 115:1–27 (1998). Smith, G. P., Golden, D. M., Frenklach, M., Moriarty, N. W., Eiteneer, B., Goldenberg, M., Bowman, C. T., Hanson, R. K., Song, S., Gardiner Jr., W. C., Lissianski, V. V., and Qin, Z., http://www.me.berkeley.edu/ gri-mech/. Gordon, A. S., Smith, S. R., and McNesby, J. R., Combust. Sci. Tech. 7:317 (1958). Warnatz, J., in Combustion Chemistry (W. C. Gardiner Jr., editor). Springer-Verlag, New York, 1984, pp. 197–360. Najm, H. N., Paul, P. P., Mueller, C. J., and Wyckoff, P. S., Combust. Flame 113(3):312–332 (1998). Mereb, J. B., and Wendt, J. O. L., Fuel 73(7):1020 – 1026 (1994). Bell, J. B., Day, M. S., Grcar, J. F., Bessler, W. G., Schultz, C., Glarborg, P., and Jensen, A. D., Proc. Comb. Inst. 29 (2002), in press. Wendt, J. O. L., and Sternling, C. V., J. Air Poll. Control Assoc. 24:1055–1058 (1974). Kee, R. J., Ruply, R. M., Meeks, E., and Miller, J. A., “Chemkin-Ill: A FORTRAN chemical kinetics package for the analysis of gas-phase chemical and plasma kinetics,” Sandia National Laboratories Report No. SAND96-8216. Wolfram, S., The Mathematica Book (4th ed.). Wolfram Media/Cambridge University Press, Champaign, IL, 1999. Glarborg, P., Kristensen, P. G., Dam-Johansen, K., Alzueta, M. U., Millera, A., and Bilbao, R., Energy Fuels 14:828 – 838 (2000). Kee, R. J., Grcar, J. F., Smooke, M. D., and Miller, J. A. “A fortran program for modeling steady, laminar, one-dimensional premixed flames.” Sandia National Laboratories Report No. SAND85-8240. Warnatz, J., Maas, U., and Dibble, R.W., Combustion (3rd ed.). Springer-Verlag, Berlin, 2001. Miller, J. A., Smooke, M. D., Green, R. M., and Kee, R. J., Combust. Sci. Tech., 34:149 –176 (1983).

Received 19 December 2001; revised 4 June 2002; accepted 7 June 2002

APPENDIX Reaction Path Diagrams The reaction path diagrams appearing in this paper are generated using the steady 2D solutions directly. Automatically generated reaction path diagrams have appeared elsewhere. For example, Warnatz et al. [49] discuss “integral

298 reaction flow analysis” with edge weights integrated over a region of space, as in this paper, or over an interval of time. We wish to acknowledge the help of Prof. D. G. Goodwin in formulating the conserved-scalar approach used here. Only conserved scalars provide a consistent measure of the exchange of material among species because of chemical reactions. For two species, s1 and s2, let ni,j(s1, s2) ⱖ 0 be the rate (mol/cm3s) at which atoms of type j are transferred from s1 to s2 as a result of reaction i. The total transfer of j from s1 to s2 is then ⌺i 兰vol ni,j(s1, s2) (mol/s), where the sum is overall reactions. In our diagrams the integral is over the whole computational domain; however, it may be useful to restrict the integration to some part of the flame determined by auxiliary field quantities, such as mixture fraction, temperature, or characteristics of the flow field. The resulting sum represents the weight of the path s1 3 s2, and its value determines the direction and relative thickness the arrow between the

N. SULLIVAN ET AL. species in the reaction path diagram for element j. The data needed to evaluate these weights is readily available from the simulation and from the chemical mechanism’s CHEMKIN database. A minor difficulty arises when the CHEMKIN description of chemical reactions is inadequate to infer ni,j(s1, s2) uniquely. For example, the CHEMKIN input “N2H2 ⫹ NH ⫽ NNH ⫹ NH2” may mean that an H atom leaves N2H2 to form NNH, or that an N atom leaves to form NH2. Most ambiguities in the present study’s CHEMKIN mechanisms are resolved by supposing that the reactions shift the fewest possible atoms. The remaining cases (four, for the nitrogen chemistry) are resolved by shifting the species fragment of lower atomic weight. Reactions whose species contain many atoms of the same kind account for the ambiguities; they make relatively small contributions to the path diagram’s arrows, so resolving the ambiguities differently leaves the figures unchanged.