The Ignition, Combustion and Flame Structure of Carbon Monoxide-hydrogen Mixtures. Note 2 Fluid Dynamics and Kinetic Aspects of Syngas Combustion

7/27/2019 The Ignition, Combustion and Flame Structure of Carbon Monoxide-hydrogen Mixtures. Note 2 Fluid Dynamics an

1/15

International Journal of Hydrogen Energy 32 (2007) 3486 3500www.elsevier.com/locate/ijhydene

The ignition, combustion and flame structure of carbon monoxide/hydrogenmixtures. Note 2: Fluid dynamics and kinetic aspects of syngas combustion

A. Cuoci, A. Frassoldati, G. Buzzi Ferraris, T. Faravelli, E. Ranzi

CMIC Dipartimento di Chimica, Materiali e Ingegneria Chimica, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milano, Italy

Received 28 November 2006; received in revised form 20 February 2007; accepted 21 February 2007

Available online 5 April 2007

Abstract

The kinetic characterization of the H2/CO system in presence of nitrogen components was systematically revised in the first note of this

work [Frassoldati A, Faravelli T, Ranzi E. The ignition, combustion and flame structure of carbon monoxide/hydrogen mixtures. Note 1:

detailed kinetic modeling of syngas combustion also in presence of nitrogen compounds. Int J Hydrogen Energy; 2007, in press]. This second

note analyses three different turbulent non-premixed syngas flames by using different approaches such as the Eddy dissipation (ED) the Eddy

dissipation concept (EDC) and steady laminar flamelets (SLF) model.

Detailed kinetic schemes are too large and computationally expensive to be directly applied to CFD codes. Pollutants marginally affect

the main combustion process and consequently it is feasible to post-process the CFD results with large detailed kinetic schemes, capable of

accurately predicting the formation of pollutants, such as NOx , CO, PAH and soot. Thus, in order to predict NO x formation in these flames,

a detailed kinetic scheme is applied by means of a newly conceived numerical tool: the kinetic post-processor (KPP).

The successful prediction of flame structures and NOx formation supports the proposed approach and makes the KPP code a useful tool for

optimizing the design of new burners.

2007 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.

Keywords: Syngas kinetics; Syngas flames; Turbulent combustion modeling; NOx formation

1. Introduction

Syngas is a mixture of hydrogen and carbon monoxide and

can be obtained from natural gas, coal, petroleum, biomass

and organic waste [1]. Methanol synthesis and FischerTropsch

synthesis remain the largest use of syngas. However, syngas

has also become a significant source of environmentally clean

fuels of late and this is why an accurate study of the structureof syngas flames with a special attention on pollutant formation

is of such interest.

The syngas-fuelled combustion system designs can utilize

CFD to optimize efficiency. At the same time, however, com-

bustion systems have to respect increasingly stringent pollutant

emission limits. Therefore, pollutant formation must be one

of the main focuses of new burner designs: this explains the

Corresponding author. Tel.: +39 02 33993286; fax: +39 02 70638173.

E-mail address: [email protected] (A. Frassoldati).

0360-3199/$- see front matter 2007 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.

doi:10.1016/j.ijhydene.2007.02.026

increasing demand for computational tools capable of charac-

terizing the combustion systems in terms of pollutant species

also. The direct coupling of detailed kinetics and complex

CFD is a very difficult task, especially when considering the

typical dimensions of the computational grids used for com-

plex geometries and industrial applications. The computational

cost significantly increases with the number of computational

grids (NC) and also with the second or third power of thenumber of reacting species (NS). Moreover, the turbulent

flow of the practical combustion devices leads to and in-

volves strong interactions between fluid mixing and chemical

reactions.

The general concept of reactor network analysis has al-

ready been employed by various authors to post-process CFD

results and evaluate the formation of pollutants, using detailed

kinetic mechanisms for various applications by utilizing a dif-

ferent level of description and various numerical methodologies

[25].
http://www.elsevier.com/locate/ijhydenemailto:[email protected]://-/?-http://-/?-http://-/?-mailto:[email protected]://www.elsevier.com/locate/ijhydene


2/15

A. Cuoci et al. / International Journal of Hydrogen Energy 32 (2007) 3486 3500 3487

Nomenclature

Roman symbols

A pre-exponential factor, kmol, m3, s

CC correction coefficient due to temperature fluctua-

tionsCP specific heat, J/kg/K

C1 C2 turbulence model constants

E activation energy, kJ/kmol

H sensible enthalpy, kJ/kg

J diffusion flux, kg/m2/s

k reaction rate, kmol, m3, s

M molecular weight, kg/kmol

NC total number of computational cells

NP total number of reactors

NR total number of chemical reactions

NS total number of chemical species

NV total number unknown variables

Pr Prandtl number

r rate of formation, kg/m3/s

R universal gas constant, kJ/kmol/K

S surface, m2

Sc Schmidt number

SH heat source, J/m3/s

T temperature, K

Tk kinetic equivalent mean temperature, K

U velocity, m/s

V reactor volume, m3

V effective volume, m3

W total convective flux, kg/s

x axial distance from fuel inlet, m

Greek symbols

temperature exponent

kinetic energy dissipation rate, m2/s3

reactor volume fraction

stoichiometric coefficient

turbulent kinetic energy, m2/s2

thermal conductivity, J/m/s/K

dynamic viscosity, kg/m/s

kinematic viscosity, m2/s

mixture fraction

mixture fraction variance

density, kg/m3

T

temperature standard deviation, K

reactor residence time, s

mass fraction

x gradient of x

Subscripts

i species

j reaction

n reactor surface

p reactor

t turbulent value

In this paper we analyze and discuss different approaches to

tackling this problem. Pollutant species only marginally affect

the main combustion process and consequently do not signifi-

cantly influence the overall temperature and flow fields. Conse-

quently it is feasible to evaluate the structure of the flame with

simplified kinetic schemes first and then post-process the CFD

results with this newly conceived numerical tool, the kinetic

post-processor (KPP) [6,7]. This KPP model, which has already

been applied to evaluating industrial burner performance, is

able to accurately predict the formation of different pollutants,

such as NOx

, CO and unburned hydrocarbons as well as poly-

cyclic aromatic hydrocarbons and soot. In order to demonstrate

the validity of this approach, three different syngas turbulent

jet flames are used as typical test cases.

2. Experimental data

Three different turbulent non-premixed syngas flames were

studied in this paper, all flames consist of a central fuel jet

surrounded by a co-flowing air stream. The geometry of the

nozzle and the composition of the fuels are shown in Table 1.

The first two flames (Flame A and B) are described by Bar-

low et al. [8,9] and were experimentally investigated in the

framework of the International Workshop on Measurements and

Computation of Turbulent Non-premixed Flames. The com-

position measurements were made at Sandia National Labo-

ratories, Livermore, California; velocity measurements were

obtained at ETH Zurich, Switzerland [10]. The flames are

unconfined and the fuel composition is 40% CO, 30% H 2, 30%

N2 (%Vol).

The burner has a central duct constructed from straight tubing

with squared-off ends with an internal diameter of 4.58 mm

for Flame A and 7.72 mm for Flame B. The thick wall of the

tubing (0.88mm) creates a small recirculation zone that aids

the flame stabilization. The computational grid was refined inthis zone to better resolve the details of the near-nozzle flow.

The central fuel jet mixes with the co-flow air stream, re-

sulting in a turbulent unconfined diffusion flame. The jet fuel

velocity is 76 m/s for Flame A and 45 m/s for Flame B, theco-flow air is 0.70 m/s velocity and both the streams areat a temperature of 292 K; the resulting Reynolds number is

16 700. Experimental results include axial and radial profilesof mean and root mean square (rms) values of temperatures

and major species concentrations as well as velocity statistics

and Reynolds stresses. Radial profiles of nitric oxide and OH

radical concentration are also available at different locations.

The third flame (Flame C) was experimentally investi-

gated by Drake et al. [11]. The fuel is fed in a central tube
http://-/?-http://-/?-http://-/?-http://-/?-


3/15

3488 A. Cuoci et al./ International Journal of Hydrogen Energy 32 (2007) 3486 3500

Table 1

Experimental syngas flames studied in this paper

Flame Nozzle ID (mm) Ujet (m/s) Rejet Ref. Fuel composition (% vol)

CO H2 N2 CH4 NH3

A 4.58 76.0 16700 [8,9] 40 30 30

B 7.72 45.0 16700 [8,9] 40 30 30 C 3.20 54.6 8500 [11] 39.7 29.9 29.7 0.7 01.64

(3.2 mm internal diameter and 1.6 mm wall thickness), cen-

tered in a 15 cm 15 cm square test section 1 m long, with flatpyrex windows on the four sides. The fuel molar composition,

very similar to the previous one, is 39.7% CO, 29.9 H2, 29.7

N2 and 0.70 CH4. Ammonia was added in different amounts

up to 1.64%; in the absence of ammonia, methane was not

included in the fuel mixture. The average fuel flow velocity

was 54.6 m/s with a resulting Reynolds number of8500; theinlet flow air velocity was 2.4 m/s. The inlet temperature of

both the streams is 300 K. Several radial profiles of velocity,temperature and species concentrations are available at differ-

ent distances from the fuel inlet. The NO concentration was

experimentally analyzed only at a distance of 100 diametersdownstream of the nozzle.

3. Flame modeling

The flames were simulated with the commercial CFD code

FLUENT 6.2. A 2D steady-state simulation of the physical do-

main was considered due to the axial symmetry of the system.

The adopted 2D grid for the Flame C was structured and non-uniform, with high resolution in the flame region close to the

inlets. The domain was resolved by 320 130 control volumesin a cylindrical coordinate system. The grid used for the Flames

A and B, on the other hand, was unstructured and consisted of

30 000 triangular computational cells. For the spatial resolu-tion the second-order upwind scheme was adopted. The segre-

gated implicit solver was used with the SIMPLE procedure for

pressurevelocity coupling. PRESTO! (PREssure Staggering

Options) algorithm was used for pressure interpolation [12].

Turbulence was modeled via the RANS approach, using the

standard model. However, it is well recognized that the

model poorly predicts the velocity field in round jets: in partic-ular it tends to overestimate the spreading rate and the decay

rate [13]. Also in these cases the model over-predicts the

diffusion of the central jet and predicts the axial velocity on

the centerline lower than the one actually measured. The mod-

ification suggested by McGuirk and Rodi was adopted to solve

this problem: the parameter C1 in the k model was corrected

from 1.44 to 1.60 [14,15]. The modified k model gives sat-

isfactory results using this correction and can be used to avoid

the additional computational expense of the Reynolds stress

model (RSM) [16]. The radiative heat transfer was calculated

with the DO model [12].

The description of the turbulencechemistry interactions rep-

resents one of the most difficult tasks in turbulent combustion.

Three different models were adopted: the Eddy dissipation (ED)

model [17], the Eddy dissipation concept (EDC) model [18,19]

and the Steady laminar flamelets (SLF) model [2022]. A brief

discussion of these approaches is reported in Appendix A.

4. The kinetic post-processor

As previously mentioned the KPP operates by assuming the

temperature and flow fields to be those predicted by the CFDcodes and solves the overall system of mass balance equations

in a complex reaction network with detailed kinetic schemes.

Even with new generation computers, the direct coupling of

detailed kinetics and complex CFD remains a very difficult and

expensive task, especially when considering the usual number

of grid points used in industrial applications. When referring

to 105106 grid cells and 100200 reacting species, the dimen-

sions of the overall system of mass balance equations become

higher than 107108.

Two major simplifications are applied and they make this

numerical approach feasible and advantageous over the direct

coupling of a detailed kinetic scheme inside the CFD code.The first feature is the transformation of the original compu-

tational grid into a reactor network. Knowledge of the thermo

fluid dynamic field, as evaluated by the CFD code, allows sev-

eral adjacent and very similar cells to be lumped or grouped

into single equivalent reactors. A second way of making the

numerical computations more stable and viable is to define an

average and fixed temperature inside the different reactors.

The solution of the CFD code provides the detailed flow,

composition and temperature fields, and this information al-

lows critical and non-critical zones in the overall reacting sys-

tem to be identified. The description detail can be reduced in

several regions without affecting significantly the results. The

grouping or clustering of several kinetically similar cells intoa single lumped reactor reduces the dimensions of the overall

system. The fixed temperature inside these reactors reduces the

extreme non-linearity of the system which is mainly related to

the reaction rates and to the coupling between mass and energy

balances.

4.1. Grouping of cells and grid sensitivity

The temperature, composition and fluid dynamic fields ob-

tained through the CFD code allow the identification of the

critical zones in the combustion chamber, i.e. the specific re-

gions where large temperature and/or composition gradients are
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


4/15


CO

NOx (x300)

number of reactors

massfra

ction

1.0E-02

9.0E-03

8.0E-03

7.0E-03

6.0E-03

5.0E-03

4.0E-03

3.0E-03

2.0E-03

1.0E-03

0.0E+000 2000 4000 6000 8000 10000120001400016000

Fig. 1. Predicted CO and NOx emissions in the exit flue gases as a function

of the number of reactors used in the simulation (Flame A).

present. It is convenient to retain the original detail of the CFD

grid in these zones. However, large volumes of the system are

less critical from a kinetic point of view, e.g. cold and/or non-

reactive zones. This fact suggests that the detail of the grid can

be locally reduced by clustering and combining several cells

into a single equivalent reactor. Of course, the lumped cell vol-

ume is simply the sum of the volumes of the grouped cells. The

original grid size is thus transformed into a network of several

reactors where the links between the different reactors simply

combine and reflect the original flow field as evaluated by the

CFD code. This allows the total number of equivalent reactorsto be reduced and makes it feasible to handle the mass balance

equations by using detailed kinetic schemes with a large num-

ber of species. The original 105106 cells can be conveniently

grouped into 103104 equivalent reactors thus maintaining a

more than reasonable description of the flame structure and the

reacting system.

The mesh-coarsening algorithm was designed in order to

prevent possible dangerous situations such as the creation of ge-

ometrical irregularities and/or non-smooth transition between

zones with very different volumes. The interlinking flows are

evaluated on the basis of the convective rates exchanged be-

tween the cells belonging to the different reactors. The massdiffusion coefficients for the coarse mesh are calculated in

agreement with the original diffusive flow rates. Temperature

and initial compositions in the equivalent reactors are the

volume averaged values of the combined cells. Different clus-

tering levels are sequentially adopted and calculations are iter-

atively performed by increasing the number of cells up to the

final convergence, i.e. up to the point where a further increase

in the reactor network dimensions makes no significant differ-

ence to the final solution. The accuracy and convergence of the

solution together with the effect of the coarsening of the mesh

need to be monitored and these points are analyzed later in this

paper when numerical procedure is discussed. Fig. 1 shows

the typical effect of clustering on NOx and CO predictions.

4.2. Average temperature, temperature fluctuations and rate

constant evaluations

As already mentioned, the KPP uses the temperature field

as obtained by the CFD computations. A fixed average tem-

perature is assumed in each equivalent reactor and the rates of

all the reactions involved in the kinetic scheme are evaluated.In turbulent combustion conditions, these reaction rates cannot

simply be calculated as a function of the mean temperature and

composition, mainly due to the highly non-linear dependence

of reaction rates on temperature. Temperature dependence of

rate constants is usually described via the modified Arrhenius

equation:

k(T ) = A T exp

E

RT

. (1)

Consequently, during turbulent combustions, temperature fluc-

tuations in particular have a significant effect on the average

rates of reactions with high activation energy. This effect is

very important for the reactions involved in NOx formation and

needs to be taken into account [18]. The average fluid dynamic

temperature T is different from the equivalent average temper-

ature from a kinetic point of view Tk. In other words, the av-

erage rate value (which accounts for temperature fluctuations

over the time) is very different from the reaction rate calculated

at the mean temperature T

k =

0 k(T(t)) dt

= k(T). (2)

This difference obviously increases for high temperature fluctu-ations and for reactions with high activation energies. To tackle

this problem with reasonable computational efforts, the Taylor

expansion of the reaction rate around T is used in the post-

processing procedure:

k(T ) = k(T ) +

n=1

1

n!

jnk

jTn

T

Tn

(3)

a few mathematical arrangements allow the following to be

deduced:

k = k(T)

1 +

2R2 R2 + 2ERT1( 1) + E2T2

4R2

T

T

2+

= k(T) CC = k(Tk), (4)

where CC is a correction coefficient due to the temperature

fluctuations. Because of the high fluctuations and slow con-

vergence, the series expansion needs to account for up to the

eighth order terms.

Of course, CC value changes for the various reactions due to

the different activation energies.


5/15


T=2000 KCorrection

Coefficient(Cc)

T=2000 K

T=1500 KT=1500 K

0 0.2 0.4 0.60

5

10

15

E/R=15000

0 0.2 0.4 0.60

500

1000

1500

E/R=35000

'T0 /T'T0 /T

Fig. 2. Correction coefficient of the rate constants versus the intensity of temperature fluctuations at average temperatures of 1500 and 2000K. Continuous

lines refer to series truncated at the eighth order whilst pointed lines are limited to the sixth order: (a) activation energy = 30000 cal/mol; (b) activationenergy = 70000cal/mol.

Fig. 2a and b show the values of the correction coefficient

as a function of temperature fluctuations, respectively, for twodifferent activation energies (30 000 and 70 000cal/mol) at

average temperatures of 1500 and 2000 K, and assuming a

sinusoidal fluctuation T(t ) = T0 sin(t ). As expected, this co-efficient is higher at 1500 K and increases with the activation

energy. These figures also show the results obtained with dif-

ferent truncation orders; CC coefficient estimation converge

when accounting the first 34 terms of the series (up to the

eighth order). These results have been proved to be fully

consistent with those obtained through rigorous computation

k(T ) CC =

0 k(T + T

0 sin(t)) dt

. (5)

The correction coefficients calculated using this approach also

agree with those predicted by the more accurate but computa-

tionally more expensive -pdf model [22]. To further clarify the

physical meaning of these corrections, we should point out that

the equivalent average kinetic temperature Tk becomes 2630K

instead of the average temperature T = 2000 K, when assum-ing the higher activation energy and T0/T = 0.5. Similarly, Tkwould become 2030K when the average temperature is 1500 K.

Correction coefficients for the reverse reactions are calculated

with the same procedure but only using the parameters of the

reverse reactions.

If not directly available from the CFD simulation, the tem-

perature variance calculation is based on an approximate form

of the variance transport equation obtained assuming equal pro-

duction and dissipation of variance [12,23]

2T =C1 t (T )

2

C2 (/k), C1 = 2.86, C2 = 2.0. (6)

4.3. Mass balance equations

CFD results are used to define the overall system by describ-

ing the mass balance equations of all the chemical species in-

volved in the detailed kinetic scheme as well as providing the

initial composition guess.

For all the equivalent reactors, the steady mass balance of

each species (i ) accounts for convection, diffusion and chem-ical reaction terms:

Wp inp,i Wp

outp,i +

NFn=1

[ Jp,n,i Sp,n]

+ Vp Mi

NRj=1

ij rp,j = 0, i = 1, . . . , N C,

p = 1, . . . , N P, (7)

where Wp is the total convective flow pertaining to the reactor.

The mass diffusion term is the sum of all the contributions

pertaining to the adjacent reactors and is computed in the fol-

lowing form:

Ji = t

Sc t i , (8)

where Sct is the turbulent Schmidt number and t the turbulent

viscosity. Laminar diffusion is neglected because it is usually

overwhelmed by turbulent transport, at least for high Reynolds

numbers. Sn is the surface between the adjacent reactors. The

reaction term contains the molecular weight Mi and the alge-

braic sum of all the reaction rates evaluated at the equivalent

kinetic temperature:

rj = kj(Tkj) f (c ). (9)

The mean reaction term in each computational cell is calcu-

lated according to the EDC model [18] and by referring to the

effective volume V.

4.4. Numerical method and control of convergence

As already mentioned, the dimension of the overall system,

which is conveniently reduced using the grouping procedure,

becomes NP NS (NP is the total number of lumped reactors)to ensure that the KPP can handle this overall system.

As an example, Fig. 3 shows a typical Boolean structure of

the whole matrix system for a simple structured 2D grid as well

as the structure of the single block.
http://-/?-http://-/?-


6/15


0500

520

540

560

580

600

500 520 540 560 580 600 620

500

1000

1500

2000

0 500 1000 1500 2000

reactorindex

reactorindex

reactorindex

reactor index reactor index reactor index

0

5

10

15

20

25

3035

40

45

50

0 5 10 15 20 25 30 35 4540 50

Fig. 3. Panel a: example of a Boolean structure of the whole matrix system for a simple structured 2D computational mesh. Panel b: zoom of the diagonal

region (square in panel a). Panel c: zoom of the single block structure (square in panel b).

The global Newton or modified Newton methods are notrobust enough to solve the system using CFD results as a first-

guess. It is therefore convenient to approach a better estimate

of the solution by iteratively solving the sequence of individual

reactors with successive substitutions. Each reactor is solved by

using a local Newton method with the possible use of a false

transient method (time stepping) to improve the initial guess

or to approach the solution.

Only when the residuals of all the equations reach sufficiently

low values, can a modified global Newton method be applied to

the whole system. Otherwise the previous procedure is iterated

to further improve the residuals.

The Newton method involves the solution of a linear system

of the Jacobian coefficient matrix. In order to increase the com-putational efficiency, special attention is devoted to the evalu-

ation of the sparse Jacobian coefficients. The derivates of rate

equations are evaluated analytically rather than numerically.

The bottleneck of this very large system comes both in mem-

ory allocation and in CPU time when a Gauss factorization

method is applied to the whole system. Thus, Gauss factoriza-

tion is applied only to the main diagonal blocks, while an iter-

ative method is applied to the other terms. This approach saves

the memory allocation and makes the solution of this overall

system viable.

In this case too, if the global Newton method does not con-

verge, a false transient method is applied to ensure a betterapproach to the solution of the whole system.

The global Newton method not only increases efficiency but,

more importantly still, ensures the complete convergence to the

solution. In fact, it is necessary to speed up the convergence

procedure, very slow in the case of direct substitutions. More-

over, it has to be clearly underlined that high attention is re-

quired in the convergence check. In fact, in the case of direct

substitution, convergence is generally controlled by the typical

normalized error sum of squares:

err =Nv

i=1

(n)i

(n1)

i

(n)i

2

, (10)

CFD Results

First guess solution

Local solution in each reactor

Newtons method

OK

No

Time integration

(ODE)

Time integration

(ODE)

Yes

OK

No

Yes

Low residuals in all equations? Yes

Global solution for all reactors

Newtons method

SolutionLow pollutant concentrations(ppm)

Needof very low residuals

No

Fig. 4. Numerical procedure to solve the global system.

where NV = NS NP is the total number of variables (massfractions) and the suffix (n) refers to the iteration. The request

that err has to be less than a fixed minimum () is a neces-

sary but not sufficient condition. A small err value may just

be the result of convergence difficulties rather than the numeri-

cal solution. The KPP complete numerical procedure is shown

schematically in Fig. 4. Additional details regarding the KPP

are reported in [7,24].

5. Kinetic schemes

The reactions adopted for the ED simulation are very

simple and correspond to the complete oxidation of syngas
http://-/?-http://-/?-http://-/?-


7/15


8/15


x/d=40x/d=20

x/d=20x/d=40

x/d=20 x/d=40

x/d=20x/d=40

Fig. 7. Temperature and composition profiles (Flame A) [9] along the axis and at different distances from the burner surface in radial direction ( d is the

internal diameter of the nozzle, x is the axial distance from the nozzle outlet). EDC model: black line, SLF model: gray line, ED model: dashed line.

major species for the three different turbulent combustion mod-

els. The ED model coupled with a simplified kinetic mechanism

gives unsatisfactory predictions, both for the temperature and

compositional fields. A better and satisfactory agreement with

experimental measurements is obtained with the EDC model.

SLF model overestimates the axial temperature profile, espe-

cially in the post-flame zone (x/d > 40).

Hewson and Kerstein [26] studied flame A using a RANS

approach and overpredicted the temperature in the flame tail by

50150 K. According to their work there are two possible rea-

sons responsible for temperature overprediction in this flame:

neglect of radiative heat losses and underprediction of the dis-

sipation rate. They estimated that radiation is not expected to

play a major role in this flame, because the time scales for ra-

diative heat losses are long relative to the flame evolution time.

Thermal radiation, which is taken into account in this work us-

ing the discrete ordinates model [12], affects the peak temper-

ature only by about 3040 K.

We performed a sensitivity analysis on the SLF simula-

tions, which confirmed that the predicted temperature profile

is nearly insensitive to the grid and the numerical schemes

or to the kinetic mechanism used to generate the flamelet li-

brary, but is mostly affected by the turbulence model used.

In fact, different turbulence models affect the jet penetration

but also the scalar dissipation rate and thus turbulent mixing.

Higher mixing rates noticeably shorten the flame, as already

discussed elsewhere [26]. In fact, better temperature profiles

in the flame tail can be obtained adopting SLF with RSM or

the standard k turbulence models, but the consequence is

the overestimation of the temperature close to the nozzle. A

further discussion on SLF modeling goes beyond the scope of

this work which will focus on the NOx chemistry in syngas

flames.

It is evident that any model overestimations of the flame

temperature affect the prediction of pollutant species with

the KPP.


9/15


2.59e-05

2.46e-05

2.33e-05

2.20e-05

2.07e-05

1.94e-051.81e-05

1.68e-05

1.55e-05

1.42e-05

1.29e-05

1.16e-05

1.04e-05

9.08e-06

7.76e-06

6.47e-06

5.18e-06

3.88e-06

2.59e-06

1.29e-06

0.00e+00

1.00e-05

9.50e-06

9.00e-06

8.50e-06

8.00e-06

7.50e-06

7.00e-06

6.50e-06

6.00e-06

5.50e-06

5.00e-06

4.50e-06

4.00e-06

3.50e-06

3.00e-06

2.50e-06

2.00e-06

1.50e-06

1.00e-06

5.00e-070.00e+00

5.29e-06

5.03e-06

4.76e-06

4.50e-06

4.23e-06

3.97e-063.70e-06

3.44e-06

3.17e-06

2.91e-06

2.64e-06

2.38e-06

2.12e-06

1.85e-06

1.59e-06

1.32e-06

1.06e-06

7.93e-07

5.29e-07

2.64e-07

0.00e+00

Fig. 8. Predicted maps of NOx mass fractions for Flame A. Panel (a) NO, Panel (b) N 2O, Panel (c) NO2. (a) NO Mass fraction; (b) N2O mass fraction;

(c) NO2 mass fraction.

Moving from these fields obtained with the EDC model for

Flames A and B, the KPP is applied with the detailed kinetic

scheme to predict NOx formation in the flame also. The pre-

dicted NOx species maps are reported in Fig. 8 for flame A.

The significant role played by N2O in the flame front and for-

mation of NO2 in the post-flame zone can be observed.

Two main NOx -forming reaction paths are relevant in these

syngas flames: thermal NO and the nitrous oxide mechanism

(N2O). The NO formation through N2O is initiated by the third

order reaction N2 + O + M = N2O + M which is followed byseveral N2O reactions with O, OH and H radicals, ultimately

leading to the formation of NO and N2. The selectivity of thisprocess is ruled by the local temperature and composition of

the flame. The NOx is formed mostly via the N2O mecha-

nism and, to a limited extent, through the thermal mechanism

(about 25% for Flame A). The significant role played by the

N2O mechanism in syngas combustion is a consequence of the

significantly enhanced production of O radicals [27].

The thermal mechanism is initiated and controlled by the so-

called Zeldovich mechanism through O + N2 = NO + N, whichis followed by N + O2 = NO + O and N + OH = NO + H.

Fig. 9 shows a comparison of NO measurements and predic-

tions along the axis of the flame and the effect of temperature

fluctuations on NO formation. The effect of temperature fluc-

tuations is relevant especially for flame B where the thermal

mechanism accounts for about half of the NO formed.

The agreement on these absolute values is satisfactory, even

though there are some discrepancies. The shape of the radial NO

profile in flame is correctly reproduced at the various distances

from the burner surface and is in very good agreement with

measurement results at x/d > 30. NO concentration is, how-

ever, overestimated close to the nozzle (x/d < 30) as shown in

Fig. 10 which compares NOx instantaneous measurements and

predictions at different axial locations. The measurements of

Fig. 10 are single-shot NO and OH measurements [9] and are

shown in scatter plot as a function of the mixture fraction at dif-ferent axial locations of Flame A. The mixture fraction is calcu-

lated here from the local composition using the Bilger formula

[28]. NO predictions, obtained using the KPP, are compared

with the scatter plot measurements of NO using a continuous

line. It is quite evident that the predicted NO mass fraction is in

good agreement with the average NO at high x/d while close

to the nozzle NO tends to be overestimated. It is interesting to

note that the predicted NO profile obtained when suppressing

the effect of temperature fluctuations (CC = 1, dashed line) liesat the lower boundary of the scatter plot.

Panel b shows the comparison between single-shot OH mea-

surements [9] and predictions obtained directly in the CFD


10/15


0 100

Flame B 2.46e-052.33e-05

2.20e-05

2.07e-05

1.94e-051.81e-05

1.68e-05

1.55e-05

1.42e-05

1.29e-05

1.16e-05

1.04e-05

9.06e-06

7.76e-06

6.47e-06

5.18e-06

3.88e-06

2.59e-06

1.29e-06

0.00e+00

2.59e-05

NO [m.f.]5.00E-05

4.00E-05

3.00E-05

2.00E-05

1.00E-05

0.00E+00

NOMassFractio

n

Flame A

Axial Position [mm]

200 300 400 500 600 700

Fig. 9. Panel a: NO mass fraction profiles along the centerline for Flame A (continuous line and symbols) and Flame B (dashed line and open symbols).Gray lines indicate the prediction of NO obtained without the proper correction for temperature fluctuations. Panel b: map of predicted NO (mass fraction)

for Flame A. The map on the right side is obtained neglecting the effect of temperature fluctuations.

3.00 E-05 6.00 E-03

5.00 E-03

4.00 E-03

3.00 E-03

2.00 E-03

1.00 E-03

0.00 E+00

6.00 E-05

5.00 E-05

4.00 E-05

3.00 E-05

2.00 E-05

1.00 E-05

0.00 E+00

2.50 E-05

2.00 E-05

1.50 E-05

1.00 E-05

5.00 E-06

0.00 E-000

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

x/d=20 x/d=20

x/d=50x/d=50

0.9 1

NOMassFraction

NO

MassFra

ction

NO

MassFra

ction

OHMa

ssFraction

Mixture fraction

Mixture fraction

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Mixture fraction

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Mixture fraction

5.00 E-03

4.00 E-03

3.00 E-03

2.00 E-03

1.00 E-03

0.00 E+00

Fig. 10. NO and OH mass fraction at different distances from the nozzle for Flame A. Comparison between single shot measurements (symbols) [9] and

average model results (lines). Panel a: continuous lines are the result of the KPP, dashed lines show the effect of neglecting temperature fluctuations on NO

predictions. Panel b: OH radicals calculated using the EDC model in FLUENT (dotted line) and the KPP (continuous line).


11/15


12/15


x/d=50

0 1 2 3 4 6 7 8

ED

SLF

x/d=50x/d=10

x/d=50x/d=10

0 5 10 15 20 25 30 35

0 5 10 15 20 25 30 35

0 5 10 15 20 25 30 35

0 5 10 15 20 25 30 35

x/d=50x/d=10

Temp

erature[K]

1.80E+03

1.60E+03

1.40E+03

1.20E+03

1.00E+03

8.00E+02

6.00E+02

4.00E+02

2.00E+025

Radial position [mm]

0 1 2 3 4 6 7 85




0 1 2 3 4 6 7 85

Radial position [mm] Radial position [mm]

0 1 2 3 4 6 7 85


2.00E+03

1.80E+03

1.60E+03

1.40E+03

Temp

erature[K]

1.20E+03

1.00E+03

8.00E+02

6.00E+02

4.00E+02

2.00E+02

Radial position[mm]

CO

2ma

ssfraction

1.40E-01

1.20E-01

1.00E-01

8.00E-02

6.00E-02

4.00E-02

2.00E-02

0.00E+00

CO

2ma

ssfraction

1.40E-01

1.20E-01

1.00E-01

8.00E-02

6.00E-02

4.00E-02

2.00E-02

0.00E+00

H2

Omassfraction

1.60E-01

1.40E-01

1.20E-01

1.00E-01

8.00E-026.00E-02

4.00E-02

2.00E-02

0.00E+00

1.20E-01

1.00E-01

8.00E-02

6.00E-02

4.00E-02

2.00E-02

0.00E+00

H2

O

massfraction

4.00E-01

3.50E-01

CO

massfraction3.00E-01

2.50E-01

2.00E-01

1.50E-01

1.00E-01

5.00E-02

0.00E+00

7.00E-02

6.00E-02

5.00E-02

CO

massfraction

4.00E-02

3.00E-02

2.00E-02

1.00E-02

0.00E+00

x/d=10exp

EDC

Fig. 13. Temperature and composition profiles (in terms of mass fractions) along the axis and at different distances from the burner surface in radial direction

(d is the internal diameter of the nozzle, x is the axial distance from the nozzle outlet) doped with 0.80% (volume) of NH 3. EDC model: black line, SLF

model: gray line, ED model: dashed line.


13/15


3.18e-03

3.05e-03

2.96e-03

2.86e-032.76e-03

2.67e-03

2.57e-03

2.48e-03

2.38e-03

2.29e-03

2.19e-03

2.10e-03

2.00e-03

1.91e-03

1.81e-03

1.72e-03

1.62e-03

1.53e-03

1.43e-03

1.33e-03

1.24e-03

1.14e-031.05e-03

9.53e-04

8.58e-04

7.63e-04

6.67e-04

5.72e-04

4.77e-04

3.81e-04

2.86e-04

1.91e-04

9.53e-05

0.00e-00

4.95e-05

4.75e-05

4.60e-05

4.45e-054.30e-05

4.16e-05

4.01e-05

3.86e-05

3.71e-05

3.56e-05

3.41e-05

3.27e-05

3.12e-05

2.97e-05

2.82e-05

2.67e-05

2.52e-05

2.38e-05

2.23e-05

2.08e-05

1.93e-05

1.78e-051.63e-05

1.48e-05

1.34e-05

1.19e-05

1.04e-05

8.91e-06

7.42e-06

5.94e-06

4.45e-06

2.97e-06

1.48e-06

0.00e+00

1.96e+03

1.89e+03

1.83e+031.76e+03

1.69e+03

1.63e+03

1.56e+03

1.49e+03

1.43e+03

1.36e+03

1.29e+03

1.23e+03

1.16e+03

1.10e+03

1.03e+03

9.63e+02

8.97e+028.30e+02

7.64e+02

6.98e+02

6.32e+02

5.65e+02

4.99e+02

4.33e+02

3.66e+02

3.00e+02

Fig. 14. Temperature and NOx mass fraction fields with 0.80% of total amount of added NH3. (a) NO mass fraction; (b) NO2 mass fraction; (c) Temperature (K).

N

Experimenal

Simulation

NO - Experimenal

NO - Simulation

NOx - Experimental

NOx - SimulationN

O

[ppmb

yvolumedry]

150

140

130

120

110

100

90

80

700 0.5 1 1.5 2 2

Yie

ld

[molesNO/molesNH3]

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0

0 0.5 1 1.5Total amount of added NH3 [%] Total amount of added NH3 [%]

Fig. 15. Panel a: comparison of measured [11] and predicted peak axial values of NO at x/d = 100. Panel b: total yield of NO and NOx formation atx/d = 100 as a function of added NH3.

various amounts of added NH3 [11]. The agreement is very sat-

isfactory, even though the predicted results tend to slightly un-

derestimate NO formation with larger amounts of added NH 3.

This agreement is quite clearly confirmed by the comparisons

reported in Fig. 15b in which total NO (and NOx = NO+ NO2)formation is related to the NH3 feed. It is clear that the pre-

dicted results are very close to the experimental measurements

and therefore not only is the adopted kinetic scheme capable


14/15


of correctly describing NOx formation and NH3 consumption,

but the CFD simulation of the flame was properly grasped. The

difference between NOx and NO is mainly due to the succes-

sive formation of NO2 when the temperature is decreasing.

7. Conclusions

The detailed kinetic scheme of oxidation and combustion of

syngas, already discussed and validated in the previous note of

this paper [25], is here successfully applied to model turbulent

diffusion flames and to predict NOx formation mechanisms.

Detailed kinetic schemes are usually too large and computa-

tionally expensive for their direct application in the CFD code,

especially in the case of large 3D grids. For this reason three

different flames have been here analyzed with a new and effec-

tive numerical tool: the kinetic post-processor (KPP). Pollutant

species only marginally affect the main structure of the flame

(i.e. temperature and flow field). Thus, the CFD results obtained

with simpler kinetics are post-processed by using large detailed

kinetic schemes, able to accurately predict also the formation of

different pollutants, such as NOx , CO, PAH and soot. A global

solution method is discussed and applied in order to overcome

the difficulties and uncertainties in the convergence attainment.

The successful prediction of flame structures and NOx for-

mation in these flames not only contributes to a further val-

idation of the kinetic scheme but also supports the proposed

approach for the KPP. The KPP code already is a very useful

tool for the optimal design of new burners with a particular at-

tention to pollutants formation. Prediction of soot formation in

turbulent diffusion flames will be the natural extension and ap-

plication of this tool. A good prediction of the flame structures

is obviously a necessary condition for the correct applicationof the KPP. In fact, the reliability of the KPP results in terms

of pollutant predictions is strongly dependent on the complete-

ness and consistency of the original CFD simulation.

Acknowledgments

The authors acknowledge the financial support of Technip

BV and of ENEA (FISR project).

Appendix A.

A.1. Eddy dissipation model

The idea behind this model is that chemistry does not play

any explicit role while turbulence controls the reaction rate. In

fact most fuels are fast burning, and the overall reaction rate is

controlled by turbulent mixing. In non-premixed flames, turbu-

lence slowly convects/mixes fuel and oxidizer into the reaction

zones where they burn quickly. In such cases, the combustion

is mixing-limited, and the complex and often unknown, chem-

ical kinetic rates can be safely ignored. For the simple reaction

F + qO (1 + q)P, the fuel burning rate is estimated frommean mass fractions of fuel F, oxidizer O and products P,

and from a turbulent mixing time tmix. If the turbulence

model is adopted for turbulence modeling, the mixing time is

inversely proportional to the specific turbulent dissipation rate

/ and the burning rate is

RF = A1

tmixmin

F,

O

q, B

P

1 + q

= A

minF, Oq

, BP

1 + q , (11)

where A and B are two model constants. In this equation the

reaction rate is limited by the deficient mean species. This is

acceptable if the reactions are very fast compared to the turbu-

lent time scales. Generally speaking, this model is too simple

to correctly predict the thermal and compositional fields for

turbulent non-premixed flames, but can be useful as the first-

guess solution for the application of different and more detailed

combustion models.

A.2. Eddy dissipation concept model

In a turbulent environment, combustion takes place wherethere is a molecular mixing, i.e. at small turbulence scales. Ac-

cording to the EDC model, the chemical reactions occur only

in small scale micro-mixed turbulent structures known as fine

structures. These fine structures are treated as a perfectly stirred

reactors (PSR) with a residence time and mass fraction .

Their volume fraction is a function of turbulent properties;

the reactions proceed in the fine structures, according to a de-

tailed kinetic scheme, for a time equal to a residence time

and an effective volume V

= 0.41

1/2

, = 2.13

2 1/4

, V = V. (12)

Based on the mass transfer between the fine structures and their

surroundings, the mean reaction term becomes

Ri =2

(1 3

)(i

0i ), (13)

where is the density and the laminar kinematic viscosity.

The basic assumption is that chemical reactions are quenched if

the characteristic chemical times for limiting species are longer

than .

The EDC model can incorporate detailed chemical mecha-

nisms into turbulent reacting flows and can be used when the

assumption of fast chemistry is invalid. However, typical mech-

anisms are invariably stiff and their numerical integration is

computationally costly.

A.3. Steady laminar flamelets model

The basic assumption is that instantaneous thermo-chemical

state of the fluid is related to a conserved scalar quantity known

as the mixture fraction. In this way the species transport equa-

tions can be reduced to a transport equation for the mixture

fraction and one for its variance 2:

j

jt() +

j

jxi(Ui) =

j

jxi t

Sct

j

jxi , (14)


15/15


j

jt

2

+j

jxi(Ui

2)

=j

jxi

t

Sct

j2

jxi

+ 2

tSc t

j

2

jxi

2 C

2. (15)

The constant C appearing in the dissipation term can be de-rived by turbulent spectral analysis and is usually set at 2. If

the system is not adiabatic, the enthalpy balance equation must

be solved

j

jt(H) +

j

jxi(Ui H) =

j

jxi

kt

Cp

jH

jxi

+ SH, (16)

where kt and Cp are thermal conductivity and specific heat of

the mixture, and SH is a generic source term which accounts

for the non-adiabatic behavior of the system.

The temperature and thermo-chemical variables are extracted

from a flamelets library, in which the temperature and compo-

sition corresponding to each value of mean mixture fraction ,

mixture fraction variance 2 and enthalpy H are stored. The

average value of the generic scalar stored in the library is

evaluated by the following integral:

=

10

f() (, H ) d, (17)

where f() is the mixture fraction probability distribution

function (PDF) and (, H ) is the relationship that links mix-

ture fraction and enthalpy to the scalar . In general a -PDF

is employed. The function (, H ), which takes into account

the treatment of complex chemistry, can be modeled following

different approach. In the case of the SLF model the turbulent

flame is considered as an ensemble of discrete, steady laminar

flames, called flamelets. The individual flamelets are assumed

to have the same structure as laminar flames in simple config-

urations.

References

[1] Wender I. Reactions of synthesis gas. Fuel processing technology, vol.

48; 1996. p. 189207.

[2] Falcitelli M, Pasini S, Tognotti L. An algorithm for extracting chemical

reactor network models from CFD simulation of industrial combustion

systems. Combust Sci Technol 2002;174:22.

[3] Falcitelli M, Pasini S, Rossi N, Tognotti L. CFD+reactor network

analysis: an integrated methodology for the modelling and optimisation

of industrial systems for energy saving and pollution reduction. Appl

Therm Eng 2002;22:971.

[4] SkjZth-Rasmussen MS, Holm-Christensen O, ]stberg M, Christensen

TS, Johannessen T, Jensen A. et al. Post-processing of detailed

chemical kinetic mechanism onto CFD simulations. Comput Chem Eng

2004;28:235161.

[5] Novosselov IV, Malte PC, Yuan S, Srinivasan R, Lee JCY. Chemical

reactor network application to emissions prediction for industrial DLE

gas turbine. Presented at Proceedings of GT2006 ASME Turbo Expo

2006: Power for Land, Sea and Air, Barcelona, Spain; 2006.

[6] Faravelli T, Bua L, Frassoldati A, Antifora A, Tognotti L, Ranzi E. A

new procedure for predicting NOx emissions from furnaces. Comput

Chem Eng 2001;25:6138.

[7] Frassoldati A, Buzzi-Ferraris G, Faravelli T, Ranzi E. Post-processing

of CFD simulations with detailed kinetics. Presented at 28th meeting of

Italian Section of The Combustion Institute, Naples, Italy; 2005.

[8] Barlow RS, Fiechtner GJ, Carter CD, Chen JY. Experiments on the

scalar structure of turbulent CO/H2/N2 jet flames. Combust Flame

2000;120:54969.

[9] Barlow RS, Fiechtner GJ, Carter CD, Chen JY. Sandia/ETH-Zurich

CO/H2/N2 Flame Data - Release 1.1. www.ca.sandia.gov/TNF , SandiaNational Laboratories, 2002.

[10] Flury M. Experimentelle Analyze der Mischungsstruktur in turbulenten

nicht vorgemischten Flammen. PhD thesis, ETH Zurich, Switzerland,

1999.

[11] Drake MC, Correa SM, Pitz RW, Lapp M. Nitric oxide formation from

thermal and fuel-bound nitrogen sources in a turbulent nonpremixed

syngas flame. Proceedings of the Combustion Institute, vol. 20; 1984.

p. 198390.

[12] FLUENT 6.2Users Guide.

[13] Wilcox DC. Turbulence modeling for CFD2nd ed. DCW Industries,

Inc.; 2004.

[14] Hossain M, Jones JC, Malalasekera W. Modelling of a bluff-body

nonpremixed flame using a coupled radiation/flamelet combustion model.

Turb Combust 2001;67:217.

[15] Dally BB, Fletcher DF, Masri AR. Modelling of turbulent flamesstabilised on a bluff-body. Combust Theory and Modell 1998;2:193.

[16] Zucca A. Modelling of turbulence-chemistry interaction and soot

formation in turbulent flames. PhD thesis, Department of Chemical

Engineering, Politecnico di Torino, Italy; 2004.

[17] Magnussen BF, Hiertager BH. On mathematical modeling of turbulent

combustion. 16th symposium (Int.) on combustion, The Combustion

Institute, Pittsburgh; 1976. p. 71927.

[18] Magnussen BF. On the structure of turbulence and a generalized Eddy

dissipation concept for chemical reactions in turbulent flows. 19th AIAA

aerospace science meeting, St. Louis, Missouri, 1981.

[19] Magnussen BF. Modeling of pollutant formation in gas turbine

combustor with special emphasis on soot formation and combustion.

18th international congress on combustion engines, international council

on combustion engines, Tianjin, China; 1989.

[20] Peters N. Turbulent combustion. Cambridge: Cambridge UniversityPress; 2000.

[21] Poinsot T, Veynante D. Theoretical and numerical combustion.

Philadelphia: Edwards; 2001.

[22] Fox RO. Computational models for turbulent reacting flows. Cambridge:

Cambridge University Press; 2003.

[23] Beretta A, Mancini N, Podenzani F, Vigevano L. The influence of the

temperature fluctuations variance on NO predictions for a gas flame.

Combust Sci Technol 1996;121:193216.

[24] Frassoldati A, Cuoci A, Faravelli T, Ranzi E. Optimal design of burners

and furnaces with CFD simulations and detailed kinetics. Presented at

6th international symposium on high temperature air combustion and

gasification, Essen, Germany; 2005.

[25] Frassoldati A, Faravelli T, Ranzi E. The ignition, combustion and

flame structure of carbon monoxide/hydrogen mixtures. Note 1:

detailed kinetic modeling of syngas combustion also in presence ofnitrogen compounds. Int J Hydrogen Energy; 2007, in press, doi:

10.1016/j.ijhydene.2007.01.011 .

[26] Hewson JC, Kerstein AR. Stochastic simulation of transport and chemical

kinetics in turbulent CO/H2/N2 flames. Combust Theory Modell

2001;5:669897.

[27] Steele RC, Malte PC, Nicol DG, Kramlich JC. NOx and N2 O in lean-

premixed jet-stirred flames. Combust Flame 1995;100:4409.

[28] Bilger RW, Starner SH, Kee RJ. On reduced mechanisms for methaneair

combustion in nonpremixed flames. Combust Flame 1990;80:13549.

[29] Barlow RS, Smith NSA, Chen JY, Bilger RW. Nitric oxide formation

in dilute hydrogen jet flames: isolation of the effects of radiation and

turbulence-chemistry submodels. Combust and Flame 1999;117:431.
http://www.ca.sandia.gov/TNFhttp://www.ca.sandia.gov/TNFhttp://10.0.3.248/j.ijhydene.2007.01.011http://10.0.3.248/j.ijhydene.2007.01.011http://10.0.3.248/j.ijhydene.2007.01.011http://www.ca.sandia.gov/TNF

Documents

The Ignition, Combustion and Flame Structure of Carbon Monoxide-hydrogen Mixtures. Note 2 Fluid Dynamics and Kinetic Aspects of Syngas Combustion