Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Modeling of Complex Premixed Burner Systems by Using

Flamelet-Generated Manifolds
J. A. VAN OIJEN,* F. A. LAMMERS, and L. P. H. DE GOEY
Eindhoven University of Technology, Department of Mechanical Engineering, P.O. Box 513, 600 MB Eindhoven,
The Netherlands

The numerical modeling of realistic burner systems puts a very high demand on computational resources. The
computational cost of combustion simulations can be reduced by techniques that simplify the chemical kinetics.
In this paper, the recently introduced flamelet-generated manifold method for premixed combustion systems is
applied to laminar flames. In this method, the reduced mechanism is created by using solutions of
one-dimensional flamelet equations as steady-state relations. For a methane/air mixture a manifold is
constructed with two controlling variables: one progress variable and the enthalpy to account for energy losses.
This manifold is used for the computation of a two-dimensional burner-stabilized flame and the results are
compared with results of detailed computations. The results show that these two controlling variables are
sufficient to reproduce the results of detailed computations. The influence of flame stretch on the accuracy of
the method is investigated by simulating strained flames in stagnation-point flows. The computation time can
be reduced by a factor of 20 when a flamelet-generated manifold is applied.
The reduction in computation time enables us to perform simulations of combustion in more complex
combustion systems. To show that the method can be used to give accurate predictions, a semi-practical furnace
is modeled and the results are compared with temperature measurements. The experimental setup consists of
a cylindrical radiating furnace with a ceramic-foam surface burner in the top disc. Radial profiles of
temperature have been measured at two different heights in the furnace. The measurements agree quite well
with the results of the numerical simulation using a flamelet-generated manifold. © 2001 by The Combustion
Institute

INTRODUCTION chemical processes with a much smaller time


scale than the flow time scales. These fast
Due to the still increasing power of modern chemical processes may be decoupled and as-
supercomputers, detailed numerical simulations sumed to be in steady state. The most promi-
of flames have become within reach. Numerical nent reduction techniques based on these as-
studies of two-dimensional (2D) unsteady sumptions are the systematic reduction
flames have been reported by several authors technique [4], the intrinsic low-dimensional
during the last several years (e.g., [1–3]). Al- manifold (ILDM) approach [5], and the com-
though most of the simulations in these studies putational singular perturbation method [6].
have been performed by using highly optimized Recently, the so-called flamelet-generated
codes and consider only academic problems manifold (FGM) method has been introduced
with simple geometries, the computation time [7], which is not only based on “chemical”
prohibits extensive parameter studies. The mod- assumptions, but also takes the most important
eling of more realistic burner systems puts an even transport processes into account. It shares the
higher demand on computational resources be- idea with the so-called flamelet approaches [8,
cause radiation and three-dimensional unsteady 9] that a more-dimensional flame can be con-
phenomena may play an important role. sidered as an ensemble of one-dimensional
In order to reduce the computational cost, (1D) flames. In the FGM method a manifold is
several methods have been developed during constructed by using 1D laminar flamelets,
the last decade, which simplify the description which can be used in subsequent flame simula-
of the chemical kinetics. Most of these reduc- tions. A similar reduction technique called
tion methods are based on the observation that flame prolongation of ILDM (FPI) has been
a typical combustion system contains many introduced by Gicquel et al. [10].
In this paper, the FGM method will be ap-
* Corresponding author. E-mail: j.a.v.oijen@tue.nl plied to the modeling of a ceramic-foam surface
COMBUSTION AND FLAME 127:2124 –2134 (2001)
0010-2180/01/$–see front matter © 2001 by The Combustion Institute
PII 0010-2180(01)00316-9 Published by Elsevier Science Inc.
MODELING OF PREMIXED SYSTEMS WITH FGM 2125

burner in a radiating furnace. Surface burners The diffusive transport of the species is de-
are of particular interest because they have a scribed by a Fick-like approximation, where the
very low NOx emission due to the cooling of the Lewis numbers— defined as Lei ⫽ ␭/␳Di cp with
gases before they burn [11]. Bouma et al. [12] Di the diffusion coefficient of species i—are
studied the stability, performance and radiative assumed to be constant [14]. The chemical
output of ceramic-foam surface burners in a source term has been divided in a production
cold environment. The performance and stabil- part ẇ i⫹ and a consumption part ẇ i⫺.
ity of ceramic-foam burners in high-tempera- To reduce the computational cost, a large
ture systems, where the risk of flashback is number of partial differential equations is
enlarged, has been investigated by Lammers et replaced by algebraic ones in reduction meth-
al. [13]. The main goal of this paper is not to ods. This, effectively, restricts the chemical
study the behavior of this semi-practical burner
state to a low-dimensional manifold of the
system, but to show that efficient and accurate
composition space. Most reduction methods
numerical simulations of such a system can be
are based on the assumption that a large
performed by using the FGM method.
number of species (or linear combinations of
The outline of this paper is as follows. In the
next section, the basic idea behind the FGM them) is in steady state. This means that the
method is explained. Starting from the three- left-hand side of Eq. (1) can be neglected,
dimensional (3D) unsteady conservation equa- leaving a balance between chemical produc-
tions, a set of flamelet equations is derived and tion and consumption:
it is shown how a manifold can be constructed
by using solutions of these flamelet equations. ẇ i⫹ ⫺ ẇ i⫺ ⫽ 0, i ⫽ 1, . . . , M, (2)
The application of the method to premixed
laminar methane/air flames is described in the assuming the first M species to be in steady
following section. A manifold will be generated state. In high-temperature regions of a flame
and applied to a 2D Bunsen-type flame. In the chemical processes are dominant and,
order to validate the FGM method the results therefore, the steady-state relations give ac-
will be compared with results of detailed com- curate results. In cold regions the influence of
putations. The influence of flame stretch on the chemical reactions is negligible and, there-
accuracy of the method is investigated. In the fore, accurate modeling of the chemistry is
subsequent section, the method will be applied not needed; bad manifold data will probably
to a ceramic-foam surface burner in a radiating not hurt. However, there is a large region in
furnace and the results will be compared with between where chemical reactions might be
thermocouple measurements. Some conclusions balanced by convection and/or diffusion pro-
are drawn in the final section. cesses. If accurate results are wanted in this
region, only a small number of steady-state
FLAMELET-GENERATED MANIFOLDS relations can be applied. This diminishes the
efficiency of the reduction methods based on
In this section, we outline the FGM method; for
such steady-state relations.
more details, the reader is referred to [7].
Premixed laminar flames are governed by the A better approximation of the mixture com-
Navier–Stokes equations, the energy equation, position in the convection-diffusion-reaction re-
and conservation equations for the species mass gion of a premixed flame can be found if the
fractions, Yi: most important transport processes are also
taken into account. Consider a curve x(s)

⭸t
共␳Yi兲 ⫹ ⵜ 䡠 共␳vY i兲 ⫺ ⵜ 䡠冉 ␭
c pLei

ⵜY i ⫽ (1)
through a premixed flame, locally perpendicular
to isosurfaces of a certain species mass fraction
Yj, and parametrized by the arc length s. The
ẇ i⫹ ⫺ ẇ i⫺, i ⫽ 1, . . . , N,
evolution of each species mass fraction Yi along
with ␳ the mass density, v the flow velocity, ␭ the this curve may be described approximately by a
thermal conductivity, and cp the specific heat. 1D equivalent of Eq. (1):
2126 J. A. VAN OIJEN ET AL.

m
⭸Yi ⭸
⭸s
⫺ 冉␭ ⭸Y i
⭸s cp Lei ⭸s

⫽ ẇ i⫹ ⫺ ẇ i⫺ ⫹ P i共s, t兲,
composition space: the point corresponding to
the unburnt mixture and the equilibrium point.
A 2D manifold can be constructed from a set of
(3)
flamelets starting at different points on a 1D
with m taken to be a constant mass flow rate. curve in composition space with a constant
Transient and multi-dimensional effects, e.g., enthalpy, pressure, and element composition. A
flame stretch and curvature, are gathered in the unique choice for this starting curve cannot be
perturbation term, Pi(s, t). Because the curve is given, although it should include the unburnt
perpendicular to the isosurfaces of Yj, the main mixture and must be chosen in such way that the
part of diffusive transport is represented by the resulting manifold is as large as possible. The
second term in the left-hand side of Eq. (3). It is exact choice of the starting curve, however, has
expected that in most situations in premixed only an effect on the manifold close to this
laminar flames Pi is small compared to the other starting curve and because chemistry is negligi-
terms in Eq. (3), although this might not be ble in this region, the influence can be ne-
justified under extreme circumstances, such as glected. This method can be extended to a
near local flame quenching. The perturbation Pi multi-dimensional manifold of d dimensions
can be neglected if the curvature radius of the generated by a (d ⫺ 1)D starting “plane.”
flame front is much larger than the flame thick- Before a manifold can be used, it should be
ness ␦, the transient time scales are longer than parametrized by controlling variables that result
the chemical time scales, and if the Karlovitz in a unique mapping Yi ⫽ Yi(Ycv,1, Ycv,2, . . . ).
number is small: K␦/sL ⬍ 1, with K the stretch In general, the conserved quantities might
rate and sL the burning velocity. What remains change due to processes other than chemical
is a balance between reaction, convection, and reaction, e.g., mixing and cooling. If variations
diffusion, which can be considered as a steady- in a conserved variable are expected to be
state relation. Together with a similar 1D equa- important in the application, then this variable
tion for the enthalpy, this set of equations is should be added as an extra controlling variable.
called flamelet equations. The set of flamelet If, e.g., enthalpy variations are expected, a series
equations can be solved by treating the system of manifolds is made for different values of the
as a 1D adiabatic premixed flame. Its solution is enthalpy and the enthalpy is added as extra
called a flamelet and forms a 1D curve in controlling variable to the manifold: Yi ⫽
composition space parametrized by s. This curve Yi(Ycv,1, Ycv,2, . . . , h).
can be considered as a 1D manifold.
As in the ILDM method a distinction is made APPLICATION TO PREMIXED LAMINAR
between variables that are conserved by chem- METHANE/AIR FLAMES
ical reactions such as the element mass frac-
tions, pressure, and enthalpy, and variables that In this paper we consider premixed methane/air
are changed by reactions: the species mass flames at atmospheric pressure with an equiva-
fractions. Note that the chemical composition of lence ratio of ␾ ⫽ 0.9. The mass fraction of
the burnt mixture is determined by the con- oxygen Y O2 is continuously decreasing during
served quantities, whereas the combustion pro- the combustion process in these flames. There-
cess from unburnt to burnt state is parametrized fore, Y O2 is an appropriate progress variable.
by the reactive controlling variables. These re- For now, we assume that one progress variable
active controlling variables are, therefore, often is enough to represent the combustion process.
called progress variables. Following existing This will be validated by a comparison between
manifold techniques, we first consider the case results of detailed and reduced computations.
of constant and given conserved variables. Because the pressure is constant and because
The 1D manifold in composition space is there is no mixing of the inlet flow with sur-
simply the flamelet starting at the point that rounding air or other gas flows with a different
represents the unburnt mixture for which the stoichiometry, the pressure and the element
manifold is created. In this way, the manifold mass fractions are not used as extra controlling
connects the two most distinguished points in variables. The enthalpy, however, is not con-
MODELING OF PREMIXED SYSTEMS WITH FGM 2127

served in the flames studied here, because the


stabilization of the flame is caused by energy
losses to the burner. Moreover, radiation and
cooling due to walls and heat exchangers may
change the enthalpy. To account for these non-
adiabatic effects the enthalpy h is introduced as
an extra controlling variable. This means that
the flamelet equations [Eq. (3)] have to be
solved for different values of the enthalpy. The
enthalpy of the flamelet can be changed in many
ways. Here we have chosen a combination of
two ways. First, the enthalpy of the flamelet is
varied simply by changing the temperature of
the unburnt mixture Tu. Here the temperature
of the unburnt mixture is varied between 400
Fig. 1. Projection of the 2D FGM data set onto the Y O2 ⫺
and 240 K, corresponding to h ⫽ ⫺0.122 and h plane for a methane/air mixture with an equivalence ratio
⫺0.295 kJ/g, respectively. For Tu ⫽ 240 K this of ␾ ⫽ 0.9.
results in a flame temperature of Tb ⫽ 2105 K,
which is still quite high. In order to reach lower
values of Tb without using unrealistic values of

Tu, we calculate burner-stabilized flamelets. By 共 ␳ h兲 ⫹ ⵜ 䡠 共 ␳ vh兲
decreasing the mass flow rate the flame loses ⭸t

冉 冊
more energy to the burner and the enthalpy ␭
decreases. This procedure is continued until the ⫺ⵜ䡠 ⵜh ⫽ ⵜ 䡠 共ᏰⵜY O2), (5)
cp
flame extinguishes. In Ref. [7] the enthalpy is
decreased by converting fuel and oxygen into where an effective diffusion coefficient Ᏸ (Y O2,
products keeping the temperature constant. Al- h) has been introduced. Following Ref. [7] the
though two different methods have been used, coefficient Ᏸ can be determined by considering
the simulation results are as good as undistin- the enthalpy flux due to preferential diffusion
guishable. This confirms that the exact choice of
Ᏺ:
the starting curve has a negligible effect on the

冘 h 冉 Le1 ⫺ 1冊 c␭ ⵜY
final result. N
The manifold data set is shown in Fig. 1. This Ᏺ⫽ i i (6)
data set is stored in a database that can be i⫽1 i p

linked to a flame simulation code. Together


with the momentum and continuity equations, Substituting the manifold relations Yi ⫽ Yi(Y O2,
the CFD code has to solve conservation equa- h) yields:
tions for the controlling variables. In manifold
冘h
N
methods these equations are derived by a pro-
Ᏺ⫽
jection of the full system on the manifold. In the i
i⫽1
FGM method the perturbation P is the only
term in Eq. (3), which is changed by the projec-
tion. Because this term can be neglected com-
pared to the other terms, ordinary conservation
冉 1
Lei
⫺1 冊 冉
␭ ⭸Y i
c p ⭸Y O2
ⵜYO2 ⫹
⭸Y i
⭸h
ⵜh . 冊
equations for the controlling variable Y O2 and h (7)
are found [7]:
To avoid unrealistic energy fluxes through inert

⭸t 冉
(␳Y O2) ⫹ ⵜ 䡠 共 ␳ vY O2兲 ⫺ ⵜ 䡠
1 ␭
LeO2 c p
ⵜY O2 冊 walls, the gradient of the enthalpy is replaced by
using the lower-dimensional manifold relation
h ⫽ Ᏼ(Y O2). Finally, this gives us the following
⫽ ẇ O2, (4) expression for Ᏺ:
2128 J. A. VAN OIJEN ET AL.

Fig. 2. Isocontours of (a) T, (b) Y O2, (c) YCO, and (d) YO computed by using (left) a detailed mechanism and (right) a FGM.
The same isolevels are used for detailed and FGM computations. The spatial coordinates are given in cm.

Ᏺ ⫽ ᏰⵜY O2, with velocity profile at the inlet is parabolic with a


maximum velocity of vmax ⫽ 1 m/s at the center.

冘 h 冉 Le1 ⫺ 1冊 c␭ 冉 ⭸Y⭸Y 冊 Isocontours of temperature and several mass


N
i ⭸Y i ⭸Ᏼ
Ᏸ⫽ i ⫹ . fractions are displayed in Fig. 2 for both de-
i⫽1 i p O2 ⭸h ⭸Y O2
tailed and reduced computations. Note that not
(8) the complete computational domain is shown,
but only the most interesting region around the
This coefficient is stored in the FGM database
flame front. From the contour plots in Fig. 2 we
as a function of the controlling variables to-
gether with all other variables needed to solve may conclude that the results of the FGM
the conservation equations, i.e., ␳, ␭, cp, T, and method using only one progress variable and
ẇ O2. the enthalpy are in excellent agreement with the
The performance of the FGM method using detailed computations: not only the position of
one progress variable and h is demonstrated by the flame front is predicted very well, but the
a simulation of a premixed methane/air that absolute values of the mass fractions as well.
which stabilizes on a 2D slot burner mounted in Flame cooling governing the stabilization of the
a cooled box. The burner slot is 6 mm wide, flame on the burner is reproduced very well by
whereas the box is 24 mm wide. The tempera- the FGM, although one can hardly speak of
ture of both the burner and the box walls are flamelets in this cold region. Also, in the flame
kept at a temperature of Twall ⫽ 300 K. The tip, where stretch and curvature are very impor-
MODELING OF PREMIXED SYSTEMS WITH FGM 2129

Fig. 4. Scaled mass-burning rate of stagnation flames as


Fig. 3. Temperature and mass fraction profiles of HCO and function of the Karlovitz integral Ka for Lei ⫽ 1. Solid:
CH2O in a stretched flame for unit Lewis numbers and a detailed reaction mechanism. Dashed: FGM with one
strain rate of 1000/s. The symmetry plane is located at x ⫽ progress variable (Y O2). Dotted: theory according to Eq. (4).
0. Solid: detailed reaction mechanism. Dashed: FGM with
one progress variable (Y O2).
sb

tant, the reduced computations appear to coin-


cide with the detailed calculations. Close to the
Ka ⫽
1
␳ us L 冕
⫺⬁
␳ KỸ ds (9)

burner wall there are some differences in the


where ␳usL is the stretchless mass-burning rate,
YCO contours: the contours don’t attach to the
K is the local stretch rate, and Ỹ is the normal-
wall in the FGM case. Because YCO follows
ized progress variable. By using this definition
directly from the manifold database, which does
for Ka and the ideas of integral analysis of
not include diffusion along the flame front, the Chung and Law [16], De Goey et al. [9] showed
CO concentration is underpredicted by the that the scaled mass-burning rate decreases
FGM in this diffusion-dominated area. To take linearly with Ka:
this into account the manifold should be ex-
tended with a second progress variable. ␳ u/ ␳ us L ⫽ 1 ⫺ Ka. (10)
To get more insight in the influence of flame
The mass-burning rate is determined at sb, the
stretch on the accuracy of the method, strained point where the chemical source term of the
flames are simulated. Planar stoichiometric progress variable has been decreased by a factor
methane/air flames are computed in the back- of 10 from its maximum value. When unit Lewis
to-back configuration [15] for different strain numbers are used for all species, the element
rates using both detailed and reduced reaction mass fractions and the enthalpy are constant in
mechanisms. Temperature field and mass frac- these flames. The results displayed in Fig. 4
tion profiles of HCO and CH2O in a strained show that in this case a FGM with only one
flame are shown in Fig. 3 for unit Lewis num- progress variable is sufficient to reproduce de-
bers and a strain rate of 103/s. The results tailed computations. This indicates that the
computed with a 1D FGM agree very well with main effects of flame stretch on the mass-
the detailed results. Not only the absolute con- burning rate can be reproduced by a 1D FGM.
centrations of the radicals, but also the position When non-unit Lewis numbers are used, vari-
of the flame front compares well, indicating that ations in the element mass fractions and the
the mass-burning rate is well predicted. Follow- enthalpy occur due to preferential diffusion
ing De Goey and Ten Thije Boonkkamp [9] we effects, which result in an extra decrease of the
study the mass-burning rate ␳u, as a function of mass-burning rate as function of Ka compared
the Karlovitz integral Ka, given by: to Eq. (10). To study the influence of these
2130 J. A. VAN OIJEN ET AL.

high strain rates (Ka ⬎ 0.1). To tackle this


problem a higher dimensional manifold should
be constructed by treating the variations in
enthalpy and element mass fractions indepen-
dently.
Besides the accuracy, the efficiency is another
important aspect of reduction methods. In or-
der to give an indication of the efficiency of the
FGM method the computation times of detailed
and reduced simulations are compared. For
both models we determined the time needed to
perform a time-dependent 1D flame simulation
of 10⫺3 s under the same conditions. When a
relatively simple detailed reaction mechanism
Fig. 5. Scaled mass-burning rate of stagnation flames as (16 species, 25 reactions) is used, a single time
function of the Karlovitz integral Ka for Lei ⫽ 1. Solid: step takes 247 ms central processing unit time
detailed reaction mechanism. Dashed: FGM with one on a Silicon Graphics workstation, whereas each
progress variable. Dotted: FGM with two controlling
variables.
time step lasts only 32 ms when the FGM is
applied. This speed up is caused by the reduc-
tion of the number of differential equations to
variations two FGMs are used: 1) a manifold be solved and a faster evaluation of the chemi-
with one progress variable; and 2) a manifold cal source terms. Another advantage of the
with one progress variable and one extra con- reduced model is that larger time steps can be
trolling variable to represent the variations in taken because the smallest time scales have
element mass fractions and enthalpy. In the case been eliminated. Therefore, the total CPU time
of weak stretch, one can show that the enthalpy for the reduced computation was 20 times lower
and element fractions change linearly with the than for the detailed simulation. An even higher
strain rate [9]. This coupling between the differ- efficiency is reached if an explicit solver is used
ent elements and enthalpy is used in the con- for the reduced computations. In that case the
struction of the second manifold. To extend the computation time reduces with a factor of 40.
1D manifold with an extra controlling variable, For more complex reaction mechanisms the
a series of flamelets is computed, where both speed up will be even larger.
the enthalpy and element composition of the The computation of the FGM database in-
unburnt mixture is varied simultaneously using volved ⬃30 min, which seems quite long. How-
this coupling. This set of flamelets is used to ever, the computation time, which can be gained
contruct a 2D manifold, which is parametrized in the modeling of series of complex burner
by one progress variable and one additional systems, is orders of magnitude larger than the
controlling variable (element mass fraction of H time needed to construct the database.
for numerical reasons).
The mass-burning rates of the flames com-
puted by using these manifolds and a detailed RESULTS OF A CERAMIC-FOAM
reaction mechanism are shown in Fig. 5. It can SURFACE BURNER IN A FURNACE
be seen that variations in the enthalpy and the
element mass fractions influence the mass-burn- Due to the enormous reduction in computation
ing rate significantly. The results of the manifold time, the FGM method allows us to model more
with the extra controlling variable are, there- complex burner systems as well. In this paper,
fore, better than when only one controlling results of simulations of the steady laminar
variable is used. Because the coupling between combustion on a ceramic-foam burner in a
the variations in enthalpy and elements de- radiating furnace are presented and compared
creases for increasing strain rates, deviations with measurements. The experimental setup has
from the detailed results occur at moderate and been designed to investigate the stability and
MODELING OF PREMIXED SYSTEMS WITH FGM 2131

TABLE 1

Properties of the ceramic-foam surface burner.

Property Symbol Value

Porosity ␹ 0.8 —
Tortuosity ␶ 0.4 —
Conductivity ␭s 1.0 䡠 10⫺1 W/mK
Heat transfer coefficient ␣v 4.0 䡠 106 W/m3K

⫺ⵜ 䡠 共共1 ⫺ ␹ 兲 ␶␭ sⵜT s兲 ⫽ ⫺␣ v共T s ⫺ T兲 ⫹ Q rs,


(11)
Fig. 6. Cross-section of the experimental setup. with ␹ the porosity, ␶ the tortuosity, and ␭s the
heat-conduction coefficient of the solid phase.
The values used for these coefficients and ␣v are
performance of a ceramic-foam burner in a hot shown in Table 1.
radiating furnace [13] and is shown in Fig. 6. The radiation term inside and outside the
The setup consists of a cylindrical furnace made ceramic-foam (Qrs and Qrg, respectively) are
of fire-resistant bricks. In the top of the furnace determined by using the discrete transfer
a round ceramic-foam burner with a diameter of method [17] for a gray gas both inside and
20 cm is fitted. At the bottom a water-cooled outside the ceramic foam. The discrete transfer
heat exchanger is located. method solves the radiative transfer equation:
Wall temperatures are measured by using
K-thermocouples at 5 and 15 cm from the top at
the inside and outside of the brick wall. Gas
di⬘
ds
冉 ␴
⫽ ⫺k i⬘ ⫺ T 4 ,

冊 (12)

temperatures can be measured by inserting S- which relates the local and directional radiative
thermocouples into the furnace through holes intensity i⬘ to the path length s and the local
in the brick wall at 2.5 and 8 cm below the absorption coefficient k and the temperature. In
burner. Measurements have been performed for Eq. (12), ␴ denotes the Stefan–Boltzmann con-
a methane/air mixture with ␾ ⫽ 0.9 and a stant. The absorption coefficient used for the
thermal load of 200 kW/m2 corresponding to a ceramic foam is 104/m, whereas in the hot
uniform inlet velocity of v ⫽ 7 cm/s and the flue-gas region an absorption coefficient of
results are corrected for radiative losses. 0.46/m is used. In the ceramic-foam region the
To include the gas-solid heat transfer in the radiation originates from the solid phase and,
porous burner and radiative heat exchange out- therefore, in this region the solid temperature is
side the foam, the enthalpy Eq. (5) has been used in Eq. (12) instead of the gas temperature.
extended with two extra terms. The first extra The boundary surfaces are modeled as diffuse
term accounts for the heat transfer between gas gray surfaces with an emissivity ⑀. The radiative
and solid in the ceramic-foam burner and is heat transfer for these surfaces is given by:

冉 冕 冊
given by ␣v (Ts ⫺ T), with ␣v the volumetric
heat-transfer coefficient and Ts the temperature
qr ⫽ ⑀ ␴T 4 ⫺ i⬘i cos ␪ d ␻ , (13)
of the solid phase. Outside the ceramic-foam
2␲
burner this term is zero. The second extra term
is a radiation term, Qrg, which is only present where ␪ denotes the angle with the normal
outside the foam burner and accounts for the vector on the surface and ␻ a solid angle. The
radiative heat transfer due to gas radiation. intensity i⬘i of the radiation impinging on the
Inside the porous region that forms the burner, surface is found by tracing rays back to their
an equation for the temperature of the solid origin and solving Eq. (12) over this path. The
phase is solved: boundary condition for Eq. (12) at the origin is
2132 J. A. VAN OIJEN ET AL.

found from the intensity i⬘o of the radiation


leaving the originating surface:

i⬘o ⫽
1⫺⑀
␲ 冕2␲
i⬘i cos ␪ d ␻ ⫹
⑀␴ 4

T . (14)

The source terms in the energy equation for the


gas phase and the solid phase (Qrg and Qrs,
respectively) are determined from:

Qr ⫽ k 冉冕
4␲

i⬘d ␻ ⫺ 4 ␴ T 4 . (15)

For the angular integral in Eq. (15), the inten-


Fig. 7. Numerical results for temperature (gray scales) and
sities of the rays traced for the surface calcula-
streamlines (lines).
tions are used.
The inlet and outlet of the furnace are treated
as black surfaces (⑀ ⫽ 1) radiating at a temper- streamlines and the temperature field is repre-
ature of 300 K. The temperature of the furnace sented with gray scales. In the figure it is shown
wall, Twall, is computed by using a relation for that the burnt gases with a temperature of
nearly 1700 K are cooled by the walls and the
the heat flux through the wall q ⫽ ⫺␣s(Twall ⫺
heat exchanger and leave the combustion cham-
Tair), with Tair the temperature of the surround-
ber with a temperature of ⬃1000 K. The flame
ing air and ␣s the heat-transfer coefficient con-
itself is hardly visible because it is only a milli-
taining contributions from conduction through
meter thick and is located very close to and
the wall and free convection and radiation at the
partly within the burner. To show the flame
outside of the wall. The temperature of the
structure in more detail, temperature and spe-
water-cooled heat exchanger is assumed to be
cies profiles along the center axis are shown in
300 K. The furnace walls and the heat ex-
Fig. 8. It shows clearly that the small scales are
changer are treated as diffuse gray surfaces with
resolved by the computation and that the com-
an emissivity of 0.4. More details about the bustion process starts within the porous me-
radiation model can be found in Ref. [18]. dium. In the same figure we have also shown the
The computation has been performed by
using the manifold presented in the previous
section. This manifold is based on a skeletal
mechanism consisting of 16 species and 25
reactions [14] and does not include C2 and NO
chemistry. However, we want to emphasize that
the FGM method can be applied to any reaction
mechanism. Therefore, if NO chemistry is in-
cluded in the starting mechanism, it will be
possible to predict NO emissions by using FGM.
The flow grid is a rectangular mesh of 106 ⫻
50 cells, with refinements at the flame front and
near the walls. The radiation grid is constructed
from the flow grid by combining multiple flow
cells to a single radiation cell. The computing
time for this simulation was ⬃15 h on a Silicon
Fig. 8. Temperature and YO profiles along the center axis.
Graphics R10000 workstation. Symbols and lines are used to represent FGM and 1D
The results of the numerical computation are detailed results, respectively. The detailed results have been
shown in Fig. 7. The flow pattern is indicated by computed on a much finer grid.
MODELING OF PREMIXED SYSTEMS WITH FGM 2133

Fig. 9. Radial profiles of the temperature at 2.5 and 8 cm Fig. 10. Vertical temperature profiles along the inside and
from the burner surface. Symbols and lines are used to outside of the furnace wall. Symbols and lines are used to
represent measurements and simulations, respectively. represent measurements and simulations, respectively.

CONCLUSIONS
results of a 1D simulation [12], because the
flame at the center of the burner can be consid- A semi-practical burner system has been mod-
ered to be 1D. This 1D computation is per- eled by using a FGM with two controlling
formed by using detailed chemical kinetics on a variables and the results have been compared
much finer grid. It demonstrates that the results with measurements. The small differences be-
of the temperature and species profiles com- tween computed and measured temperatures
puted by using the FGM are in good agreement are within the experimental error. Measure-
with detailed computations. The small differ- ments of species concentrations and flow veloc-
ence in temperature profile (and therefore in ities are planned for the near future. The results
YO) is caused by the different treatment of of detailed simulations of a 2D Bunsen flame
radiation and conduction in the ceramic foam. are also reproduced very well by the reduced
However, the temperature of the burnt gas is computations.
well predicted, which indicates that the stabili- Because in the FGM method a manifold is
zation of the flame on the burner can be mod- generated by using 1D flamelets, it can be
eled by using a FGM. considered as a combination of a manifold and
The radial profiles of the temperature have a flamelet approach. The method shares the
been measured at 2.5 and 8 cm from the burner basic idea with flamelet approaches that multi-
surface and the results are shown in Fig. 9. It dimensional flames can be considered as a set of
can be seen that the predicted temperatures 1D flames. The implementation, however, is
coincide well with the experimental values. The typical for a manifold method (ILDM), which
main problem in the measurements is the un- means that ordinary conservation equations are
certainty in the location of the thermocouple. solved for the controlling variables. This may be
Due to this, the error in the measured temper- considered as an advantage of the FGM method
atures is about 50 K. In Fig. 10 the measured compared to flamelet methods. Moreover, the
wall temperatures are compared with the com- number of progress variables is not limited to
puted results. Both the temperatures measured one as in existing flamelet approaches. In the
at the inside and at the outside coincide well regions of a premixed flame where reaction is
with the numerical results. The temperatures balanced by convection and diffusion the FGM
measured at z ⫽ 15 cm are outside the compu- method is more accurate than reduction tech-
tational domain. However, when the numerical niques based on local chemical equilibria, be-
results are extrapolated, they seem to agree with cause the most important transport processes
the measurements. are also taken into account. The accuracy of the
2134 J. A. VAN OIJEN ET AL.

method can be enhanced by adding extra 6. Massias, A., Diamantis, D., Mastorakos, E., and Gous-
progress variables. Results of strained methane/ sis, D. A., Combust. Flame 117:685–708 (1999).
7. van Oijen, J. A., and de Goey, L. P. H., Combust. Sci.
air flames in stagnation-point flow have shown Technol. 161:113–138 (2000).
that the main effects of flame stretch on the 8. Peters, N. Twenty-First Symposium (International) on
mass burning rate can be modeled by using a Combustion, The Combustion Institute, Pittsburgh,
FGM with only one progress variable. To cap- PA, 1986, pp. 1231–1250.
ture the small remaining effects of flame 9. de Goey, L. P. H., and Ten Thije Boonkkamp,
J. H. M., Combust. Flame 119:253–271 (1999).
stretch, extra controlling variables can be added. 10. Gicquel, O., Darabiha, N., and Thévenin, D. Twenty-
The computation time of flame simulations Eighth Symposium (International) on Combustion, The
can be reduced by at least a factor of 20 by Combustion Institute, Pittsburgh, PA, 2000, pp. 1901–
applying a FGM. Finally, we may conclude that 1908.
the FGM method can be used to perform 11. Viskanta, R. Eighth Symposium (International) on
Transport Phenomena in Combustion, San Francisco,
accurate and efficient simulations of premixed 1995, pp. 64 – 87.
laminar flames in complex burner systems. 12. Bouma, P. H., and de Goey, L. P. H., Combust. Flame
119:133–143 (1999).
The financial support of the Dutch Technology 13. Lammers, F. A., Bouma, P. H., Althuizen, T. A. M.,
Foundation (STW) and NOVEM is gratefully and de Goey, L. P. H. Fifth International Conference on
Technologies and Combustion for a Clean Environment,
acknowledged.
Lisbon, Portugal, 1999, pp. 533–539.
14. Smooke, M. D., and Giovangigli, V., in Reduced Ki-
netic Mechanisms and Asymptotic Approximations for
REFERENCES Methane-Air Flames (M. D. Smooke, Ed.), Springer-
1. Maas, U., and Thévenin, D. Twenty-Seventh Sympo- Verlag, Berlin, 1991, pp. 1–28.
sium (International) on Combustion, The Combustion 15. Dixon–Lewis, G. Twenty-Third International Sympo-
Institute, Pittsburgh, PA, 1998, p. 1183–1189. sium on Combustion, The Combustion Institute, Pitts-
2. Chen, J. H., Echekki, T., and Kollmann, W., Combust. burgh, PA, 1990, pp. 305–324.
Flame 116:15– 48 (1998). 16. Chung, S., and Law, C. K., Combust. Flame 72:325–336
3. Najm, H. N., Paul, P. H., Mueller, C. J., and Wyckoff, (1988).
P. S., Combust. Flame 113:312–332 (1998). 17. Shah, N. G. (1979). Ph.D. thesis, Dept. of Mechanical
4. Peters, N., in Reduced Kinetic Mechanisms and Asymp- Engineering, University of London, London, UK.
totic Approximations for Methane-Air Flames (M. D. 18. Lammers, F. A. (2001). Ph.D. thesis, Dept. of Mechan-
Smooke, Ed.), Springer-Verlag, Berlin, 1991, pp. 48 – ical Engineering, Eindhoven University of Technology,
67. Eindhoven, The Netherlands.
5. Maas, U., and Pope, S. B. Twenty-Fifth Symposium
(International) on Combustion, The Combustion Insti- Received 24 February 2001; revised 25 July 2001; accepted
tute, Pittsburgh, PA, 1994, pp. 1349 –1356. 20 August 2001

You might also like