THERMOFLUIDODYNAMIC SIMULATION OF PRACTICAL COMBUSTION SYSTEMS AND PREDICTION OF NOX BY REACTOR NETWORK ANALYSIS

THERMOFLUIDODYNAMIC SIMULATION OF PRACTICALCOMBUSTION SYSTEMS AND PREDICTION OF NOX BY REACTOR

NETWORK ANALYSIS

M. Falcitelli(a), S. Pasini(b), L. Tognotti(c)

(a) Consorzio Pisa Ricerchec/o ENEL S.p.A. Produzione RicercaVia A. Pisano, 120 - 56122 Pisa, Italy (Fax n. +39 (0) 50 535521)E-mail: [email protected]

(b) ENEL S.p.A. Produzione RicercaVia A. Pisano, 120 - 56122 Pisa, Italy (Fax n. +39 (0) 50 535531)E-mail: [email protected]

(c) Università degli Studi di Pisa - Dipartimento di Ingegneria Chimica, Chimica Industriale eScienza dei Materiali – Via Diotisalvi 2, 56100, Pisa, Italy (Fax n. +39 (0) 50 511266)E-mail: [email protected]

RADIATION - 200117-22 June, 2001 Antalya, TURKEY

Corresponding author:Dr. Mariano Falcitellic/o ENEL S.p.A. Produzione RicercaVia A. Pisano, 120 - 56122 Pisa, Italy.Tel: +39 50 535537FAX: +39 50 535521E-mail: [email protected]

THERMOFLUIDODYNAMIC SIMULATION OF PRACTICAL COMBUSTION SYSTEMSAND PREDICTION OF NOX BY REACTOR NETWORK ANALYSIS

M. Falcitelli *, S. Pasini**, L. Tognotti****Consorzio Pisa Ricerche. P.zza A. D’Ancona, 1. - Pisa, Italy

**ENEL S.p.A. Produzione Ricerca. Via A. Pisano,120. - Pisa, Italy***Università degli Studi di Pisa - Dip. di Ingegneria Chimica,

Chimica Industriale e Scienza dei Materiali - Pisa, Italy

ABSTRACT. An integrated methodology for the simulation of practical combustion systems ispresented. A detailed 3D simulation of flow is performed adopting adequate CFD models for heattransfer, chemical reactions and turbulence. Then flow information is extracted from resulting CFDfields to create simplified reactor networks, for utilising detailed reaction chemistry and predictingthe interconversion of the species involved in combustion. The study of two glass melting furnacesare shown as example. The furnaces were experimentally characterised, then CFD simulations wereperformed. From each CFD simulation, a chemical reactor network was extracted, to perform thecomputation of the secondary product combustion species. An evaluation of the models was givencomparing the measurements with of both the temperature CFD field and the NOx prediction. Adiscussion of the effect of the various sub-models is presented with a special emphasis on the heat-transfer boundary conditions and combustion chemistry interaction.

INTRODUCTION

Simulation of industrial combustion system using Computational Fluid Dynamics (CFD) modellingis still a challenging domain. Up to now, the event to incorporate a detailed reaction kineticsdirectly in a 3D CFD code is still unfeasible, because of the exorbitant computational demands(both in terms of memory and CPU speed) it would request. Therefore approximations at someappropriate level have to be made. For the procedure showed in the present work, the choice is toperform on the first a CFD simulation on a fine grid using a kinetic mechanism describing theoxidation of the main species involved in combustion; then, on the basis of resulting flow fields, an“equivalent” network of ideal chemical reactors is extracted, as simplified flow model, and theconcentration of minor species is calculated, using a detailed combustion chemical reactionschemes, including those of nitrogen species. The separation in two steps is possible as minorspecies have a neglecting influence on the flow field and heat exchange. In the present work thismodelling methodology has been applied to the study of two glass melting furnaces. The furnaceswere experimentally characterised, then CFD simulations were performed, setting carefully theboundary conditions for the radiative heat exchange and the sub-model for the chemistry. Fromeach CFD simulation, a chemical reactor network was extracted, to perform the computation of thesecondary product combustion species by means of a complex kinetics mechanism. An evaluationof the models was given comparing the measurements with both the temperature CFD field and theNOx prediction.

DESCRIPTION OF THE GLASS FURNACES MODELLED

The application of the modelling approach to a “end-port” regenerative glass melting furnace isshown. This kind of furnace, sketched in Fig. 1, is essentially a large tank, covered by an arch

ceiling. The walls and the ceiling are refractory lined, in order to ensure a high thermal inertia andto allow an uniform irradiation of the melting glass. The batch enters from the “dog-house” and themelted glass flows towards the end-wall, which is opposite to the inlet-outlet front-wall. On thisside, there are two firing ports, each one equipped with three under port barrel burners, fed bynatural gas. Regenerative heat exchangers are located before each port, and the furnace is firedalternately from either port (with a cycle time of about 20 minutes); so that, in a cycle, preheated airof 1200 ºC is fed through the inlet port, while exhaust from the outlet port allows the other heatexchanger to regenerate.

Two furnaces of different size (5 and 10 MW of thermal power) were fully characterised by meansof temperatures and chemical species measurements [1]. For in-flame gas temperaturemeasurements and gas sampling, a suction pyrometer equipped with water cooled lances ofdifferent lengths was used, coupled to the following gas analysers: Ultramat–Siemens NDIR, forNOx, CO, CO2 and Oximat–Siemens Paramagnetic analyser, for O2. Refractory temperatures werealso measured, using a monochromatic optical pyrometer, to allow a right evaluation of theboundary conditions for modelling. Further data, such flow and composition of secondary streams(infiltrating air and evaporation from melted bath) have been extrapolated with a mass balance onthe measured chemical composition of the exhaust and the operating conditions of the feeds. Thecharacteristic parameters of the two furnaces are listed in Table 1.

Table 1Furnace parameters

5 MWth furnace 10 MWth furnaceDimensions (L,W,H) 8.4 m, 5.9 m, 2.3 m 11 m, 7 m, 2.4 mNumber of burners 3 2Natural gas inlet(LCV= 50.165 MJ/kg)

136 m/s, 0.1195 kg/s, 25 °C 150 m/s, 0.2020 kg/s, 25 °C

Pre-heated combustion air flow 12.34 m/s, 2.06 kg/s, 1200 °C 11.44 m/s, 3.602 kg/s, 1300 °CInfiltrating air from “dog-house” 0.0847 kg/s, 50 °C -Infiltrating air from tank border 0.0396 kg/s, 225 °C 0.2020 kg/s, 225 °CRelease from melting glassflow rate, temperaturecomposition (% wt)

0.1756 kg/s, 1460 °C17.33% H2O, 82.67% CO2

0.375 kg/s, 1500 °C17.33% H2O, 82.67% CO2

Inlet Air

Burners

Outlet Glass Melted LoadDog-House

INLET/ OUTLET DUCTS

VAULT

GAS/ GUN - BURNERS

GLASS MELTED LOAD“DOG HOUSE”

Figure 1. Schematic draw of an “end-port” glass melting furnace (left) and lateral view (right).

INLET/OUTLET DUCTS

VAULT

GAS/GUN - BURNERS

GLASS MELTED LOAD“DOG HOUSE”

CFD MODELING

For the 3D CFD calculation the IPSE code was used. It is a finite-volume code, belonging to ENEL,developed for the numerical modeling of reacting flows, with a special emphasis on 3D simulationof combustion systems. The code solves the Favre-averaged Navier-Stokes equations for a dilatablefluid, together with mass and enthalpy conservation equations, transport equations for chemicalspecies and equations of state for ideal gases in the well established form [2]. In the case studied ofglass melting furnaces, the particular models adopted for turbulence, combustion chemistry and heattransfer are described as follows.

Turbulence model A simple zero equation turbulence model is used in order to leave greatercomputational resources to the representation of combustion chemistry. The turbulence is modeledadopting the Boussinesq hypothesis, for which the Reynolds Stresses are taken to be proportional tothe mean rates of deformation. A constant value of the cinematic turbulent viscosity t is assignedto the whole domain evaluating the following relation for the feeding jets which transport themaximum momentum flow,

where d is the amplitude or diameter of the nozzle, U0 is the inlet velocity and C is an adimensionalconstant which is worth 0.0128 for axisymmetric jets and 0.0185 for plane jets [3].

Kinetic model of hydrocarbon combustion The solution of a conservation equation includingconvective and diffusive terms is computed for each chemical species present in the system. Thereaction rates, that appear as source terms, are computed from the Arrhenius kinetic rate expressionwhich considers both the forward and backward rate constant for reactions. The turbulent Schmidtnumber is assumed 0.7.

Heat Transfer Radiation is the predominant mechanism of heat transfer in glass melting furnaces.The Discrete Ordinates method in the S4-approximation is used to solve the radiation transportequation. The angular quadrature schemes for the S4 approximations with N=24 directions andwm=/6 weight, such as the numerical solution procedure adopted, are the ones proposed byTruelove (1987) [4].

Numerical Methods The time discretization is formulated as explicit for of all the transportequations, with the exception of species transport equations, where the convective and diffusiveterms of transport over the cells are treated explicitly, while the source term inside each cell, due tothe finite rate reaction chemistry, is solved implicitly, after being linearized respect to the massfraction increments. The solution scheme is transient SIMPLE like [5], with the difference that ateach time step the use of a direct matrix inversion algorithm yields the exact solution of pressureequation.

REACTOR NETWORK ANALYSIS: THE METHODOLOGY

A basic organisation of the procedure to create simplified reactor networks, for utilising detailedreaction chemistry and predicting the interconversion of the secondary species is described below.1. A CFD simulation is performed on a fine mesh to compute the flow, temperature and the massfraction fields of the major species involved in combustion. Then the data are post-processed inorder to calculate are the local stoichiometry (defined as the inverse of equivalence ratio).2. Analysing the distribution of the local values of the computed fields, the cells of the mesh areclustered by ranges of values of temperature. Clustering is made considering the correlationdisplayed for each system. The result of clustering is that every cell belongs to a homogeneous zoneand so each zone is modelled as an ideal reactor.

0dUCt =

T

3. The operating parameters are assigned to the reactors. The volume of each reactor is taken as thesum of the belonging cells volume. The reactors are considered isothermal at the operativetemperature computed from the CFD field using the enthalpy conservation expression:

where mi and Ti are the mass and temperature of the ith cell and the sum extends over all the cellsbelonging to the reactor. The clustered zones are modelled as perfectly stirred (PSR) or plug flowreactors (PFR) on the basis of velocity vector distribution.4. The mass exchange between all the reactors and the feeding streams are computed using the CFDspecific flow field as the sum of the mass flows between cells belonging to different reactors. In thisway all the mass exchanges are considered and the network is designed including recycling streams.5. The kinetics computation is performed on the reactor network using a detailed kinetic model forchemical species involved in combustion. The main oxidation mechanism, for the simulation ofhydrocarbon mixtures containing up to twelve carbon atoms, is the core of the overall kineticscheme [6]. Within this comprehensive box, a series of hierarchical reaction blocks are disposed,starting from the simplest and basic CO/H2 oxidation mechanisms. External reaction blocks, eachone separately describing the oxidation chemistry involved in other classes of problems, likenitrogen oxides, soot, PAH (Poly-Aromatic Hydrocarbons) and chlorinated compounds, can belinked to the main scheme. The hydrocarbon combustion mechanism involves about 200 speciesand more then 3000 reactions, the nitrogen sub-mechanism involves about 200 reactions and 40species [7].

RESULTS AND DISCUSSION

The goal of the investigation was to tune the computational tools, previously tested on utility boilers[8], to this kind of industrial furnace, in order to assist design modifications. Many CFD simulationswere performed, adopting different boundary conditions for the radiative heat exchange. The effectsof the models and conditions were evaluated comparing both the flow and temperature CFD fieldsand NOx concentration calculated by RNA, with the measurements.

0)( =∑ ∫i

T

Tipi dTTcm

Table 2Boundary conditions and parameter used in the CFD simulations

5 MWth furnace 10 MWth furnaceComputational grid 13104 cells (26x12x42) 21525 (35x15x41)Turbulent kinematic viscosity 0.212 m2/s 0.371 m2/sAbsorption coefficient for gas 0.269 m-1 0.269 m-1

Scattering coefficient for gas 0 0Melted glass surface emissivity 0.5 0.5Melted glass surfacetemperature

1460 °CFrom inlet/outlet side to bottom1480 °C, 1520 °C, 1560 °C

Refractory walls emissivity 0.7 0.7Thermal conductivity k W/(mK)vaultvault/side walls borderside wallstank border

0.9970.811

0.9970.811

External temperature 57 °C 57 °C

( ) ( )exti TTkTq −=− int4

int

CFD: computational grid and boundary conditions The finest grid adopted was a Cartesianmesh with 21525 nodes (35x41 for the base, 15 for the height). The best agreement with themeasurements was obtained using the conditions summarised in Table 2.

The boundary conditions for the calculation of the radiative heat transfer were set: for the glassmelted surface specifying the temperature (as read from the optical pyrometer) and the emissivity,for the side walls and the vault by the following wall equation.

The emissivity , the conductivity k and the external temperature Text were specified on the basis ofmanufacture data and temperature measurements, thus, during the CFD computation run time, theinternal temperature Tint of the walls was recalculated at each iteration using the incident radiationqi. The combustion model adopted is based on a “quasi-global” scheme [9], combining a singleirreversible reaction of the fuel (as CH4) and oxygen to form CO and H2, together with a detailedCO/H2 oxidation mechanism with 8 species and 9 reactions, as shown in Table 3. The kineticscheme adopted for the CO/H2 system was a sub-set of the mechanism proposed by Westbrook andDryer (1984) [9]. It was derived selecting the most important elementary reactions involving theCO, CO2, H2, H2O, O2, molecules and the OH, O, H radicals. The kinetic constants adopted are theresult of both extensive literature review and comparison with predictions produced by morecomplex mechanisms.

For each simulation a chemical reactor network was extracted by RNA, as simplified flow model,and the chemical species concentrations were recalculated by means of the kinetic mechanism.

RNA: building the reactor network model The chemical engineering model for both the furnaceswere generated in a similar way. In the following the network obtained for the 5 MW furnace isdiscussed in detail. Figure 2 shows the distribution of the CFD Stoichiometry and Temperaturelocal values, related to all the mesh cells. The total stoichiometry of the feeds is almost unitary

Table 3Reaction mechanism rate coefficients in cm3 sec cal mol K units,

Reaction A E Reference1) CH4 + 1/2 O2 ⇒ CO + 2H2 7.e19 0.00 46700 [9]2) CO + OH ⇔ CO2 + H 1.50E+07 1.30 -760 [9]3) H+ O2 ⇔ O + OH 5.10E+16 -0.82 16510 [10]4) O + H2 ⇔ H + OH 1.80E+10 1.00 8830. [11]5) H2 + OH ⇔ H2O + H 1.20E+09 1.30 3630 [12]6) OH + OH ⇔ O + H2O 6.00E+08 1.30 0 [13]7) O2 + M ⇔ O + O + M 1.85E+11 0.50 95560 [14]8) H + H + M ⇔ H2 + M 1.00E+18 -1.00 0Enhanced third-body efficiencies:

H2 = 0, CO2 = 0, H2O = 0[15]

9) O + H + M ⇔ OH + M 4.00E+18 -1.00 0Enhanced third-body efficiencies:

H2 = 2, O2 = 0.4, H2O = 6.5[16]

10) H + OH + M ⇔ H2O + M 7.5E+23 -2.60 0Enhanced third-body efficiencies:

H2O = 20[17]

( )RTEATk /exp −=

(slightly in excess of oxygen), but the local values of stoichiometry and temperature are arrangedover a wide range, since the combustion is turbulent non-premixed. Nevertheless these twoquantities show some clear correlation: the points are arranged below a curve which is similar to thegraph of adiabatic flame temperature vs stoichiometry. In fact the peak temperature is obtained forstoichiometry less than one, due to the strong influence of radiation on the local temperature field.Cells at the exhaust draw a tail around the stoichiometric value of the feeds with decreasingtemperatures; moreover some scattered cells at greater stoichiometry are shown, due to the presenceof some infiltrating air streams near the outlet.

Clustering cells by temperature and stoichiometry ranges, 12 zones of the furnace were modelled as

Figure 3. Left side: reactor network produced by the procedure for the 5 MW. The numbers neareach reactor show the concentration of NOx (ppm vol.) resulting by the kinetics computation.Right side: comparison of measured and predicted NOx concentration, as NO2, in the exhaust, forthe 10 MW furnace. In the plant the variation of the Oxygen concentration in the exhaust is dueto the intrinsic instability of the inlet conditions; for the Reactor Network Model the variation isreproduced with slight variations of the ratio fuel/air.

R1302

R4909

R32900R123524

R2218

R91143

R6951

R8927

R5264

R10344

R7774

R11752

1.2 1.4 1.6 1.8 2.0800

900

1000

1100

1200

1300

1400

1500

1600

1700

1800 measured values predicted by RNA (CFD with kinetics) predicted by RNA (CFD with equilibrium)

NO

2 [ m

g/N

m3 @

8%

O2 ]

O2 (% dry vol.)

0.0 0.5 1.0 1.5 2.0 2.5150016001700180019002000210022002300240025002600

R12

R1

R11R10 R8R7

R4R2 R9R6R5

R3

Tem

pera

ture

[K]

Stoichiometry

0 1 2 3 4 5 6

0

1

2

3

4

5

6

7

8

9

Z (m

)

X (m)0 1 2

0

1

2

3

4

5

6

7

8

9

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12

Z (m

)

Y (m)0 1 2

0

1

2

3

4

5

6

7

8

9

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12

Z (m

)

Y (m)

Figure 2. Distribution of the CFD Stoichiometry and Temperature local values, related to all themesh cells (left) and subdivision of the modelled volume resulting by clustering for two sections:horizontal at burner level y = 0.25 m (middle), vertical at inlet port level x = 1.85 m (right). Thedifferent colours together with the dashed lines show the clustered zones assigned to reactors.

ideal chemical reactors. Figure 2 shows also the resulting subdivision of the modelled volume fortwo sections. The reactor network "extracted" by the postprocessing procedure is shown in Fig. 3: itconsists of two reactors receiving the main feeds (R1, R2), two reactors, one reducing, the otheroxidising, for the zone with highest temperature of flame (R3 , R12), one series of reactors withexcess of oxygen (R4, R9, R8), one series of sub-stoichiometric reactors (R5, R10), one series ofreactors with unitary stoichiometry (R6, R7, R11), as interface between the other two, whichfollows the formation of exhausts. The scheme includes the main streams and feeds; infiltrating airstream, evaporation from the melted bath and minor recycling streams are not drawn to notcomplicate the scheme itself. Anyhow, all the feeds and mass flows of the system were consideredwhen the complex kinetic calculation was performed. Reaction progress within the individualreactors was calculated using DSMOKE, a CHEMKIN like software with a special way for solvingthe reactions with abstracting radicals [18], on a Personal Computer with a Pentium II processor.CPU time demands for a plug flow reactor was about 25 seconds, while for a perfectly stirredreactor it was about 5 seconds and thus 1 hour and quarter for the present network looped 25 times.

Simulation results For both the furnaces the resulting CFD flow field was “U” shaped with twomain recirculating zones: one central, the other placed near the end corner opposite to the outletport. For the 5 MW furnace the CFD temperature field has been compared with 6 measurementpoints. The resulted agreement was a mean shifting (|∆T|/T) of 0.03, and standard deviation (sd)0.01. For the 10 MW furnace the comparison with measurements (34 probing points), shown in Fig.4, gave the following result for the temperature field: mean |∆T|/T = 0.03, sd = 0.02. The final NOxconcentration calculated using reactor network models network was in good agreement withmeasurements. For the 5 MW furnace, a concentration of 1459 mg/Nm3@8%O2 dry of nitrogenoxides, as NO2, was measured, while the simulated value was 1500 (3 % greater). For the 10 MWfurnaces Figure 3 shows a graph with the measured NOx concentration in the exhaust versus theoxygen concentration and with the response of the network respect to the variation of the sameparameter, obtained with slight changes in the fuel/air ratio of the feeds. The agreement betweenpredicted and measured NOx concentration was again very good less than 2%. A sensitivityanalysis performed on the reactor networks showed that the increase of number and the change oftypology of the reactors affected the result by 3%, while the oxygen concentration and thetemperatures are crucial parameters.

Figure 4. 10 MW furnace: comparison of measured and predicted temperatures. The origin of thereference frame is placed at the outlet side corner.

Meas.Points

Coordinates(x,y,z) m.

Meas.Points

Coordinates(x,y,z) m.

1 6.00, 0.43, 3.60 18 5.00, 1.03, 8.852 5.00, 0.43, 3.60 19 6.00, 1.03, 8.853 6.50, 0.43, 6.35 20 1.00, 0.43, 3.604 6.00, 0.43, 6.35 21 1.00, 0.43, 6.355 5.00, 0.43, 6.35 22 2.00, 0.43, 6.356 6.00, 0.43, 8.85 23 3.00, 0.43, 6.357 5.00, 0.43, 8.85 24 1.00, 0.43, 8.858 4.00, 0.43, 8.85 25 2.00, 0.43, 8.859 5.65, 0.84, 3.60 26 4.00, 0.43, 8.8510 5.00, 0.84, 3.60 27 1.00, 0.84, 6.3511 6.00, 0.84, 6.35 28 2.00, 0.84, 6.3512 5.00, 0.84, 6.35 29 1.00, 0.84, 8.8513 6.00, 0.84, 8.85 30 2.00, 0.84, 8.8514 5.00, 0.84, 8.85 31 1.35, 1.03, 3.6015 5.65, 1.03, 3.60 32 2.00, 1.03, 6.3516 5.00, 1.03, 6.35 33 1.00, 1.03, 8.8517 6.00, 1.03, 6.35 34 1.00, 1.03, 8.85

0 5 10 15 20 25 30 351550

1600

1650

1700

1750

1800

1850

1900

1950

200010 MW furnace

measurements kinetic model equilibrium chemistry

Tem

pera

ture

(°C

)

measurement positions

Effect of the boundary condition for the radiative heat exchange The sensitivity of wallconditions and chemical reaction sub-model adopted for CFD simulation was investigated. OtherCFD simulations have shown that the local radiation and temperature fields are fairly dependent onboundary conditions. With the lack of some data to define these conditions, it is common to adoptthe adiabatic assumption. Hence, it is interesting to show the results of an adiabatic CFD+RNA.The adiabatic condition was achieved cancelling the black body radiative term for the walls andceiling (it remained for the glass surface) and considering all the incident radiation re-emitted withdiffused reflection. In this case the resulting flow field had the same general shape as the previousone. On the contrary the temperature field was in worse agreement with the measurements (|∆T|/T =0.04, sd = 0.02), but above all, in comparison with the foregoing case, it was globally 30 °C colder,with a peak temperature of 400 °C lower. The NOx concentration produced by the kineticcalculation on the reactor network was strongly influenced by these differences, it produced 345mg/Nm3@8%O2 (as NO2), about a fourth of the measured value.

Effect of the chemical reactions sub-model in the CFD model Modelling the 10 MW furnaceoffered the opportunity to compare the effects on the result of two chemical sub-models, morecommonly used in CFD codes, for the calculation of the heat release by chemical reactions. Inaddition to the case, already discussed, with the reduced chemical reaction scheme for methaneoxidation, a simulation with local equilibrium of chemical species was performed. This modelspeeds up the calculation run time, because the number of chemical species are reduced (theradicals are missing) and no rate equation has to be solved. Nevertheless, the resulting CFD fieldsof chemical species concentration strongly differ from those obtained with the simplified kinetics.On the one hand, the comparison of the temperature field with measurements resulted in a worseagreement respect to the case with kinetic model, as shown in Fig 4. On the other hand, the largestdifference was for the oxygen mass fraction field: in the case with kinetics the fraction of themodelled volume in which sub-stoichiometric conditions were present was 29%, while in the casewith equilibrium it was 55%. These differences between CFD fields have affected the RNA resultsas shown in Fig. 3, the case of RNA extracted from CFD with chemical equilibrium disagree withmeasurement: the predicted NOx are 846 (46% lower).

CONCLUSIONS

A procedure, called RNA, have been introduced. It allows to extract from the CFD fields a ReactorNetwork model, on which detailed kinetic mechanisms can be employed for the calculation of thepollutant formation/destruction. The concept is not restricted to the NOx calculation, it can be usedalso for any other species for which a detailed reaction scheme is available. This approach would bea reasonable trade-off between the complexities of the phenomena occurring in reactive flow fieldsand the engineering demands for addressing the design of practical combustion systems. Thisprocedure well performed for glass melting furnaces: a 12 reactor network suitably describes thecomplex flow in these systems, resulting in a very good agreement between measured and predictedNOx concentration in the exhaust. The importance of setting the right boundary conditions andchoosing a suitable chemical reaction sub-model in the CFD for heat release and stoichiometricfield calculation was demonstrated.

REFERENCES

1. Benedetto, D., Gelmini, A., Mola, A., Pasini, S., Santero, A., Thermofluidodynamic 3Dmodelling of glass melting furnaces for combustion optimisation and emission reduction. XVA.T.I.V. Meeting -Glass Industry Towards 2000 - Parma, Italy, 15-17 September, 1999.

2. Anderson, D.A., Tannehill, J.C. and Pletcher, R.H., Computational Fluid Mechanics and HeatTransfer, Hemisphere Publishing Corporation, Taylor & Francis Group, New York., 1984.

3. Schlichting, H., Boundary Layer Theory. McGraw-Hill, 1960.

4. J.S. Truelove, Discrete-Ordinate Solutions of the Radiation Transport Equation, Journal of HeatTransfer, Vol. 109, pp 1048-1051, 1987.

5. Versteg, H.K., Malalasekera, W., An introduction to computational fluid dynamics. The finitevolume method. John Wiley & Sons Inc., 1995.

6. Ranzi, E., Gaffuri, P., Faravelli, T., Dagaut, P., A Wide Range Modeling Study of n-HeptaneOxidation, Combust. Flame Vol. 103, pp 91-106, 1995.

7. Faravelli, T., Antichi, A., Callierotti, C., Ranzi, E., Benedetto, D., A Kinetic Study of AdvancedReburning Process, Combustion Theory and Modelling, Vol. 1, pp 377-393, 1997.

8. Benedetto, D., Pasini, S., Falcitelli, M., La Marca, C. and Tognotti, L., NOx emissionprediction from 3-D complete modelling to reactor network analysis, Combust. Sci. and Tech.,Vol. 153, pp 279-294, 2000.

9. Westbrook, C.K. and Dryer, F.L., Chemical kinetic modeling of hydrocarbon combustion,Prog. Energy Combust. Sci. Vol. 10, pp 1-57, 1984.

10. Miller, J.A., J. Chem. Phys. Vol. 74, p 5120, 1981.

11. Baulch, D.L., Drysdale, D.D., Horne, D.G. and Lloyd, A.C., Evaluated Kinetic Data for HighTemperature Reactions, vols 1 and 2, Butterworths, London, 1973.

12. Cohen, N., Westberg, K., Evaluation and Compilation of Chemical Kinetic Data, J. Phys. Chem.83, 46 (1979).

13. Ernst, J., Wagner, H. Gg., Zellner, R., Direct Rate Measurements for OH + OH H2O + O in theRange 1200-1800 K, Ber. Bunsenges. Phys. Chem., Vol. 81, p 1270, 1977.

14. Watt,W.S., Myerson, A.L., Atom formation rates behind shock waves in oxygen, J. Chem.Phys., Vol. 51, pp 1638-1643, 1969.

15. Warnatz, J., Rate coefficients in the C/H/O system. In "Combustion Chemistry", ed. W.C.Gardiner, Jr., pub. Springer-Verlag, NY, p 197, 1984.

16. Tsang, W., Hampson, R.F., Chemical kinetic data base for combustion chemistry. Part I.Methane and related compounds. J. Phys. Chem. Ref. Data, Vol. 15, p1087, 1986.

17. Wilson,W.E., Jr., A critical review of the gas-phase reaction kinetics of the hydroxyl radical, J.Phys. Chem. Ref. Data, Vol. 1, pp 535-573, 1972.

18. Manca, D., Faravelli, T., Pennati, G., Buzzi Ferraris, G., Ranzi, E., Numerical Integration ofLarge Kinetic Systems. AIDIC Conference Series, Vol. 1, p 115, 1995.