Applied Thermal Engineering 31 (2011) 3885e3896

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Numerical investigation of burner positioning effects in a multi-burner flameless

combustion furnace
B. Danon a, b, *, E.-S. Cho a, W. de Jong a, D.J.E.M. Roekaerts a, b
a
Energy Technology, 3ME Faculty, Delft University of Technology, The Netherlands
b
Multi-Scale Physics, Faculty of Applied Sciences, Delft University of Technology, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: In this paper results are presented of a numerical study performed for four different burner configura-
Received 29 May 2011 tions in a furnace equipped with three pairs of flameless combustion burners firing Dutch natural gas.
Accepted 19 July 2011 The simulations have been validated against previously published results of an experimental study [1].
Available online 26 July 2011
The commercial Computational Fluid Dynamics (CFD) code Fluent 6.3 was used for the calculations.
Using the Eddy Dissipation Concept (EDC) model for turbulenceechemistry interaction in combination
Keywords:
with the realizable keε model for turbulence and a skeletal chemistry mechanism, the main furnace
Flameless combustion
performance was consistently reproduced for all the investigated burner configurations. Moreover, it was
Multi-burner furnace
CFD simulation
found that due to relatively low Reynolds numbers in the cooling air flow in the annulus of the cooling
Burner positioning tubes, predictions of the heat extraction rates of these cooling tubes were improved by treating the flow
in the cooling tubes as laminar. Furthermore, the applied error tolerance of the ISAT procedure was
insufficient for accurate species concentration predictions, however, based on analysis of the main
species concentrations in the flue gas, this inaccuracy did not influence the overall predictions.
The most important experimental results have been investigated using the CFD simulations. Firstly,
a longer path length from the firing burners to the stack, compared to the path length to the regenerating
burners, explained the lower CO emissions in the flue gas in the stack. Secondly, it was found that
a recirculation zone between the upper firing burners and the stack in configurations C4 and C5 resulted
in a smaller fraction of the flue gases leaving the furnace via the stack compared to the other configu-
rations. Thus, a larger fraction left the furnace via the regenerating burners and this resulted in higher
preheat temperatures of the combustion air. Furthermore, more pronounced recirculation zones in
configurations C3 and C4 led to higher temperature uniformities in the furnace. Finally, it was confirmed
that the jets of the burners in configurations C1 and C3 showed similar merging behavior, leading to
similar NO emissions, as observed in the experiments.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction high momentum injection, large quantities of hot flue gases are
entrained into the jets before they mix with each other, lowering
The industrial wish for higher energy efficiency of large-scale the oxygen availability in the reaction zone, thus, also lowering the
furnaces, and its associated fuel savings, has demanded for new local reaction rates and the peak temperatures. These low peak
combustion technologies, combining heat recirculation from the temperatures reduce the thermal NO emissions [5].
flue gas with low pollutant emissions. Flameless combustion (also Since the introduction of flameless combustion in the early
known as flameless oxidation (FLOX) [2], HiTAC [3] or MILD nineties of the last century, many universities and research
combustion [4]) is such a novel combustion technique. In this departments of industry have made efforts in experimentally
technique the combustion air can be highly preheated, without investigating this new technology. These studies have been per-
increasing the pollutant emissions, in particular NOx. The air and formed on many different scales, from single open flames in jet-in-
fuel are injected at high velocity and spatially separated. Due to this hot-coflow setups [6,7] to (semi-)industrial scale test furnaces
equipped with multiple burners [8e11]. A focus is here on furnaces
equipped with multiple flameless combustion burners.
* Corresponding author. Energy Technology, 3ME Faculty, Delft University of
Technology, The Netherlands. Many of these experimental studies have been complemented
E-mail address: [email protected] (B. Danon). with Computational Fluid Dynamics (CFD) simulations. In the

1359-4311/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2011.07.036
3886 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

Fig. 1. Furnace sketch, representing burner configuration C5 firing in parallel mode. The boxed numbers 1 and 2 indicate the two sample positions for the flue gas. Sampling point 2
is after the regenerators. The vertical symmetry plane is indicated by the (green) shaded plane. All dimensions are in mm. The two inserts show enlarged front views of the mesh
around a burner (left hand side) and a cooling tube (right hand side). The total mesh contains approximately 1.5 million hexahedral cells. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)

following the quality of these simulations of furnaces with multiple In 2004 Hekkens et al. simulated the experiments performed
flameless combustion burners and the choice of physical models in with the 1000 kW_th IFRF furnace firing natural gas using the
these simulations are discussed. Fluent CFD package [15,16]. Again, the standard keε turbulence
The 200 kW_th semi-industrial furnace at the Royal Institute of model was used. Three different combustion models were applied;
Technology (KTH) has been simulated extensively using the STAR- two Probability Density Function (PDF) methods, assuming chem-
CD CFD package [12]. Besides the standard keε turbulence model, ical equilibrium and laminar flamelets, and the ED/FR model. In this
several combustion models have been investigated, of which the last model a two-step chemistry for the combustion of methane
combination of the Eddy Dissipation model and Finite Rate chem- was used. Radiation was incorporated using the Discrete Ordinates
istry (ED/FR) turned out to give better results. A two-step chemical method and the absorption coefficient was calculated with the
mechanism for the combustion of the fuel (LPG, i.e., propane), with domain-based WSGG model. After detailed comparison of the
CO as an intermediate, was used. Thermal radiation was calculated numerical and experimental results, it turned out that the ED/FR
with the Discrete Transfer method and the absorption coefficient of model performed best regarding the species concentration
the gases using the Weighted Sum of Gray Gases (WSGG) model. predictions inside the furnace. The PDF methods were unable to
The cooling tubes were not incorporated in the simulation, but correctly predict the temperatures inside the furnace, which can be
a temperature, based on measured temperatures, was set as explained by the fact that they assume (too) fast chemistry.
a boundary condition. For the validation of the simulations However, also the ED/FR model overpredicted the peak tempera-
measured heat fluxes, wall temperatures and in-furnace species tures inside the furnace, even though two model constants have
concentrations were used [13]. The validation results were been adjusted for flameless combustion purposes using previous
reasonable, however, the in-furnace temperatures, which this set of IFRF measurements.
models is known to overpredict, were not compared. The three The 200 kW_th natural gas fired flameless combustion furnace
experimentally investigated firing modes, i.e., which combination at the Faculté Polytechnique de Mons has been simulated by Lupant
of burners form a burner pair, were also investigated numerically. et al. using the Fluent CFD package [10,17,18]. The standard keε
The main differences observed in the experiments were repro- turbulence model is applied, with a model constant adjusted for
duced by the simulations. Initially, the analysis of the results was improved prediction of the spreading rate of the jets. Combustion
focused on the fundamental properties of flameless combustion. In modeling is done using both the PDF method assuming chemical
a later study (firing natural gas) also the observed and reproduced equilibrium and the combined ED/FR model. Again, as in the
differences between the firing modes were investigated numeri- simulations for the IFRF furnace, two model constants in the ED/FR
cally in more detail [14]. model were changed in order to improve the results. A one-step

C1 C3 C4 C5
Fig. 2. Overview of simulated burner configurations. The two rectangles represent the two side walls of the furnace with the burner flanges. The large circles represent the burner
flanges, if filled black it is occupied by a firing burner, if filled gray it is occupied by a regenerating burner. The black and gray circles switch after a period of 30 s. The small circles
denote the position of the cooling tubes.
 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896 3887

Table 1 this standard set of models, a wide variety of models for turbulence,
Composition of actual and simulation fuel for Dutch natural gas [26]. wall functions and PDF shapes were compared [21]. The simula-
Component DNG tions are validated against in-furnace temperature measurements
Actual Simulation
[11]. The agreement is reasonable outside the combustion zone.
Inside the combustion zone the temperatures are fairly over-
%-mass %-mass
predicted, as was observed also in the other simulations using a PDF
CH4 69.97 75.98
method. After the validation of the simulation, several numerical
C2H6 4.63 e
C3H8 0.90 e
investigations were performed [19,20].
C4H10 0.47 e Generally, it can be concluded that combustion models
C5H12 0.16 e assuming fast chemistry (e.g. PDF methods assuming chemical
C6H14 0.23 e equilibrium) are not applicable for the simulation of flameless
O2 0.02 0.02
combustion. Also, a careful choice of the chemical mechanism is
N2 21.52 21.87
CO2 2.10 2.13 important.
At Delft University of Technology (DUT) a 300 kW_th furnace
equipped with three pairs of flameless combustion burners, with
the unique possibility to vary the positions of the burners in the
1450
furnace, has been investigated experimentally [1,22]. In this paper
the results of detailed CFD simulations of this furnace are pre-
x
COOLING sented. The objective of the simulations was to investigate and
150

100

z
AIR explain the observed trends in the furnace performance for the
different burner configurations in the furnace.
1282

Fig. 3. Schematic representation of a single cooling tube. All dimensions are in mm. 2. Experimental setup

The furnace is equipped with three pairs of regenerative

chemistry mechanism has been used for the combustion of flameless burners. The burner pairs have a thermal power of
methane. Radiative heat transfer is modeled using the Discrete 100 kW_th each, thus 300 kW_th in total. The furnace has inner
Ordinates method and the WSGG model was used for the calcula- dimensions of 1500
1500
1850 mm (length
width
height).
tion of the absorption coefficient. Validation of the simulation is The insulation consists of three layers of ceramic fire bricks,
performed by comparison of the numerical results with in-furnace together 300 mm thick. In total 18 flanges for the burners are
radiative heat flux, temperature and species measurements. divided over two opposite sides of the furnace (nine each). In this
Reasonable agreement was achieved for the heat fluxes, however, way, it is possible to investigate different burner configurations in
all simulations overpredicted the temperature inside the furnace. the furnace. The heat sink is made of eight single ended cooling
This difference was smaller in the simulation where the model tubes, four placed above the burners and four beneath, see also
constants in the ED/FR model were adjusted. Although the Fig. 1 and Section 3.2.
maximum concentrations of O2 were predicted rather well, the The burners were manufactured by Wärmeprozesstechnik
concentration dropped too fast after its maximum. This indicates GmbH and are of the REGEMAT CD 200 type. Each burner has four
that the combustion reactions are too fast, which also explains the combustion air/flue gas nozzles around a central fuel nozzle, with
prediction of the temperatures being too high. Finally, it is noted diameters of 20 mm and 12 mm, respectively. The burners have
that the modification of the model constant in the turbulence ceramic honeycomb heat exchangers (regenerators) incorporated.
model improved the prediction of the delay in the reaction. Eighty percent on mass basis of the flue gas is sucked by a fan via
Several regenerative reheating furnaces at the NKK steel the air nozzles over these honeycombs (thus heating them) for
company firing Low Calorific Value gases were simulated by Ishii regeneration of the heat, while the remaining twenty percent is
et al. using the Fluent CFD package [11,19,20]. Again, the standard leaving the furnace via the central stack at the roof. Three burners
keε model for turbulence is used. For the modeling of the are firing simultaneously, while the other three burners are
combustion the PDF method assuming chemical equilibrium is regenerating. After a preset time interval of 10e30 s they switch
applied. Radiation is incorporated using the P-1 model and again and the firing burners start regenerating, and vice versa. The cycle
the WSGG model was used for the absorption coefficient. Besides time (tcycle) is defined as one periodic cycle, i.e., the sum of the firing

a 8 b 8
z = 0.32 z = 0.64 z = 0.96 z = 0.32 z = 0.64 z = 0.96
Velocity magnitude (m/s)

Velocity magnitude (m/s)

6 6

4 4

2 2

0 0
−0.1 −0.05 0 0.05 0.1 −0.1 −0.05 0 0.05 0.1
Radial distance x (m) Radial distance x (m)

Fig. 4. Predicted velocity magnitude profiles (m/s) in the cooling tube at three axial distances z (m) versus the radial distance x with x ¼ 0 at the axis. (a) Realizable keε turbulence
model and (b) laminar zone.
3888 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

Table 2 The simulations were performed with the commercial code

Heat extraction rate (Qtube) and temperature of the cooling air at the outlet (Tcoolout) Fluent 6.3.26 [23]. The second order implicit steady solver was used
of a cooling tube from configuration C1.
with Green-Gauss cell based interpolation. A SIMPLE algorithm is
Variable Experiment Turbulent CFD Laminar CFD used for pressure-velocity coupling.
Qtube (kW) 12.3 18.9 14.6 The calculations were performed on a cluster with 4 dual core
Tcoolout ( C) 496 702 550 AMD Opteron 8222 (3.0 GHz) processors with 8 GB DRAM per
processor, thus 32 GB in total. The simulations were calculated
parallel on all eight nodes.
1200
T T
furn preheat
1100 3.1. Furnace
Temperature (°C)

1000 For turbulence closure the realizable keε model was used. This
model is shown to give improved results, compared to the standard
900
keε model, for the prediction of round jets [24], as is the case in
800 these burners.
For the combustion of the fuel the mechanism of Smooke et al.
700
for the combustion of methane was applied [25]. This so-called
600 skeletal mechanism consists of 16 species (resulting in 16 trans-
0 50 100 150 200 250 300 port equations to be solved) and incorporates 46 reactions. It is
Samlping time (s) noted here that a mechanism for the reaction of methane only is
used. This implies that the Dutch natural gas (DNG) is replaced by
Fig. 5. Furnace temperature (Tfurn,  C) and preheat temperature of the combustion air a simulation fuel. This simulation fuel contains, as a fuel compo-
(Tpreheat,  C) over a period of 5 entire cycles from the C3 burner configuration
nent, only methane, also replacing the higher hydrocarbons present
experiment.
in DNG, in such way that the heat release per unit mass stays equal.
In Table 1 the composition of the actual DNG and its simulation fuel
and regenerating period for one burner. The experimental data is presented.
with the highest cycle time (60 s) were taken for the present steady The turbulenceechemistry interaction is taken into account by
RANS simulations, since these experiments are considered to be using the Eddy Dissipation Concept (EDC) model [27]. The EDC
closest to steady state operation. The data with an excess air ratio model has been successfully applied for various turbulent and
(l) of 1.25 were used for the simulations. Dutch natural gas (DNG) flameless combustion applications [28,29]. Parente et al. compared
was used as the fuel, which has a net (lower) calorific value of results for flameless combustion with different turbu-
31.669 MJ/m3. The furnace operates at atmospheric pressure. Using lenceechemistry interaction models and chemistry mechanisms
an NDIR gas analyzer, the NO and CO concentrations in the flue gas [30]. They concluded that the EDC model combined with a skeletal
from the stack and from the regenerators (boxed numbers 1 and 2 or full chemistry mechanism performed best. Additionally, Stefa-
in Fig. 1, respectively) were measured on-line. In the same positions nidis et al. also compared the EDC model with the combined ED/FR
the O2 concentration was determined on-line paramagnetically for approach [31]. They report that the main disadvantage of the ED/FR
normalization purposes. More details on the experimental setup model is too fast reaction rates for the oxidization of the fuel. This
and results can be found in previous publications [1,22]. disadvantage is especially undesired for flameless combustion
Four different burner configurations were investigated numeri- applications, since in flameless combustion the reaction rates are
cally. For all of these configurations the firing mode was parallel, relatively low. Finally, Kim et al. performed a study on the use of
indicating that all the firing burners are on one side of the furnace, different chemistry mechanisms for the simulation of flameless
while all the regenerating burners are on the other side of the combustion. The EDC model was used in all the test cases, since it
furnace. In Fig. 2 the simulated burner configurations are presented. was considered by the authors as a very suitable model for the
simulation of flameless combustion [32].
3. Numerical setup The in situ adaptive tabulation (ISAT) procedure, for the efficient
computation of homogeneous reactions in chemically reacting
A three-dimensional mesh of the furnace was generated using flows, has been applied [33]. The error tolerance of this procedure
Gambit 2.4, see also Fig. 1. The mesh contains approximately 1.5 was stepwise decreased until a value of 104. For one case, reported
million hexahedral cells. Exploiting the symmetry of the furnace below, the error tolerance was decreased further to a value of 105.
with respect to its vertical midplane, to reduce the computational Radiative heat transfer is accounted for using the Discrete
time, only half of the furnace is meshed. The cooling tubes are Ordinates method and the domain-based Weighted Sum of Gray
entirely incorporated in the simulations. Gases model for the absorption coefficient of the gas mixture. The

Table 3
Comparison of main variables between experiments (EXP) and simulations (CFD) for the four different burner configurations (C1, C3, C4 and C5).

Variable C1 C3 C4 C5

EXP CFD EXP CFD EXP CFD EXP CFD

Tfurn ( C) 1036 1121 1051 1113 1074 1142 1052 1128
Tstack ( C) e 1182 e 1192 e 1216 e 1195
Qcool (kW) 99 104 97 101 98 102 90 99
Qloss (kW) 145 110 122 108 141 110 150 110
*
Qloss (kW) 137 e 112 e 134 e 144 e
O2 (%-vol, dry) 3.48 3.22 3.95 4.76 3.22 3.47 3.79 3.43
Oth
2 (%-vol, dry) 3.66 3.64 5.58 5.56 4.16 4.14 4.03 4.01
 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896 3889

a 1500 b 160 c 5
Q Q

Heat extraction rate (kW)

T T cool loss
furn stack
4
1400

O (%−vol, dry)
140
Temperature (°C)

1300 3
120
1200 2

2
100
1100 1
O2
1000 80 0
1st−ip3 1st−ip4 1st−ip5 2nd−ip4 1st−ip3 1st−ip4 1st−ip5 2nd−ip4 1st−ip3 1st−ip4 1st−ip5 2nd−ip4

Fig. 6. Main predicted results for configuration C5 parallel versus ISAT error tolerance (IET) and discretization scheme; ip3 = IET of 103, ip4 ¼ IET of 104, ip5 ¼ IET of 105,
1st ¼ first order discretization and 2nd ¼ second order discretization. (a) Furnace temperature (Tfurn,  C) and flue gas temperature in the stack (Tstack,  C). (b) Heat extraction rate of
the cooling tubes (Qcool, kW) and other heat losses (Qcool, kW). (c) Oxygen concentration in the flue gas from the stack (%-vol, dry).

mean beam length was based on the dimensions of the furnace 3.2. Cooling tubes
since the flue gas composition inside the furnace is reasonably
uniform and with this approach improved predictions of the radi- In preliminary simulations of the furnace the heat extraction
ation fluxes between the walls and cooling tubes are expected. The rate of the cooling tubes was fairly overpredicted. In order to
emissivity of steel was set to 0.8 and that of the firebrick insulation investigate this difference a separate simulation was performed on
to 0.68. a single cooling tube. The mesh of this single cooling tube is kept
The density of the mixture is calculated by the multi-component exactly the same as the mesh of a cooling tube in the mesh of the
ideal gas law using the low Mach number approximation p ¼ pref. furnace. As a boundary condition the furnace temperature was set
The molecular viscosity and thermal conductivity of the gas mixture on the outer tube in this separate simulation.
are taken constant at 1.72
105 kg/ms and 0.0241 W/mK, respec- Each cooling tube consists of two concentric annular tubes. Air
tively. The mass diffusivity is calculated by Fick’s law with a constant was used as the cooling medium and enters the inner tube, turns at
and identical mass diffusion coefficient of 2.88
105 m2/s for all the end and flows back through the annulus between the inner and
species. Finally, the heat capacities are determined per species with outer tubes. This setup was chosen to minimize the temperature
a temperature dependent piece-wise polynomial provided by gradients along the length of the outer tube, thus, creating an as
Fluent, and for the mixture with the mixing law. uniform as possible heat extraction distribution. The inner tube has
The heat losses through the walls were incorporated by calcu- an internal diameter of 100 mm and penetrates 1282 mm into the
lating the convective and radiative heat flux to the environment. outer tube, which in turn penetrates 1450 mm into the furnace and
For a realistic value of the effective thermal conductivity of the has a inner diameter of 150 mm. In Fig. 3 a schematic representa-
walls, the predicted wall temperatures were compared with the tion of a single cooling tube is presented.
experimental values. In order to characterize the flow regime inside the cooling tube
In previous studies this set of models proved to be suitable for the the Reynolds number was calculated. The Reynolds number (Re) is
simulation of flameless combustion in furnace environments defined in Equation (1).
[34,14]. Additionally, De et al. performed a detailed comparison
study of these and comparable models for experimental results of rudh
Re ¼ (1)
a jet-in-hot-coflow setup, which emulates flameless combustion m
[35]. Overall, the set of models was well capable of predicting
where r is the mean density (kg/m3), u the mean velocity magni-
flameless combustion. Moreover, it was shown that the realizable
tude (m/s), dh the hydraulic diameter (m) and m the molecular
keε performed better than other two-equation turbulence models.
viscosity (kg/ms). The hydraulic diameter dh is 0.1 m and 0.05 m for
Also, the use of the above-mentioned constant values for the
the inner tube and the annulus, respectively. The Reynolds number
viscosity and the species mass diffusivities showed no significant
differences with simulations with a temperature dependent
viscosity or using multi-component diffusion. The largest difference,
Regenerators (ppmv@3%O2, dry)

8
compared to the experiments, was that the ignition was predicted to
EXP
occur too early. Finally, a limitation of the use of the EDC model with
CFD
a skeletal chemistry mechanism was demonstrated for flows with 6
(x 0.1) (x 0.01)

low turbulence Reynolds numbers. In the present furnace the Rey-

nolds numbers were verified to be higher than the Reynolds
numbers where these limitations become apparent. 4

a O2 11%
b O 12%
2
2
CO < 1% CO < 1%

H2O 46% H2O 45% 0

0 2 4 6 8
Stack (ppmv@3%O2, dry)
CO 43%
2 CO2 42%
(x 0.1) (x 0.01)
CH4 < 1% CH4 < 1% Fig. 8. Comparison of experimental values for CO concentration (0.1
ppmv @3%O2,
dry, EXP) and predicted values from the simulations (0.01
ppmv @3%O2, dry, CFD) in
Fig. 7. Main species mass fractions () in the flue gas in the stack with burner the flue gas from the stack and from the regenerators for all four burner configurations
configuration C5 for (a) IET ¼ 104 and (b) IET ¼ 105. firing in parallel mode.
3890 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

a b
8000
Stack Regenerating burner

CO (ppmv@3%O2 dry)
1000
6000
500

4000 0
1.6 2 2.4

2000

0
0 1 2 3 4
Distance along streampath (m)

Fig. 9. (a) Sideview of the furnace showing the pathlines from the nozzles of a firing burner in configuration C5. Pathlines are colored by the velocity magnitude (m/s). (b) CO
concentration (ppmv @3%O2, dry) versus distance along the paths from the firing burner to the opposing regenerating burner (solid line) and to the stack (dashed line). Results from
CFD simulations.

of the flow in the inner tube and annulus are around 20,000 and 3.3. Data comparison
10,000, respectively. These values indicate that the flow in the inner
tube is turbulent, however, in the annulus it is in the range of For comparison of the transient experimental data (due to the
Reynolds numbers that indicate the transition regime between switching of the burners) and the steady state numerical results,
laminar and turbulent flow. averaging is required. In Fig. 5 the furnace temperature, charac-
In order to further investigate the flow regime in the annulus, the terized by a thermocouple in the furnace side wall, and the preheat
predicted velocity magnitude profiles, using the turbulence model, temperature of the combustion air in one burner are presented over
in the cooling tube at three axial positions (z) are presented in a period of 5 entire cycles. These data are from the C3 burner
Fig. 4(a). Indeed, the velocity profile in the inner tube has the shape configuration experiment.
typical for turbulent pipe flow. However, the velocity profile in the It is noted that the transient behavior is well represented in the
annulus tend to a parabolic form, which indicates laminar flow. preheat temperature, while the furnace temperature is not
In a subsequent simulation the entire cooling tube has been showing any correlation with the cycle time. Actually, all the
treated as a laminar zone. In Table 2 the numerical results for the measurements used as boundary conditions or for validation of
single cooling tube (with boundary condition values taken from the the simulations (wall temperatures, heat extraction rates of the
C1 configuration experiment), treating the flow in the cooling tube cooling tubes, composition of the flue gas) show similar steady
either as turbulent or as laminar, are compared with the experi- signals.
mental values. It is observed that the predictions for both the total The preheat temperature of the combustion air is also used as
heat extraction rate (Qtube) and for the temperature of the cooling air a boundary condition of the simulations, since the regenerators are
at the outlet (Tcoolout) were closer to the experimental values when not incorporated in the simulations. Therefore, the data of the
the cooling tubes were treated as laminar. This can be explained by preheat temperature of an entire firing period is averaged and set as
the flow in the annulus to rather resemble laminar flow behavior the boundary condition.
and thus creating a sort of thermal resistance, limiting the total heat Furthermore, in order to have a similar spatial heat distribution
extraction rate of the cooling tube. In Fig. 4(b) the velocity profiles in the simulations as in the experiments, two separate steady state
using the laminar model are presented. It is observed that the simulations were performed for each configuration. For both parts
velocity profiles in the annulus are more pronounced parabolic. of an entire cycle a steady simulation was performed, i.e., one
To conclude, in the furnace simulations, the flow in the cooling simulation with one set of burners firing and one simulation with
tubes is treated separately from the rest of the furnace and as the other set of burners firing was performed, see also Fig. 2. For the
laminar flow. The flow within the furnace itself is still treated as final comparison of the experimental and numerical results, the
turbulent flow. results of these two simulations were averaged.

a b
1000
Air preheat temperature (°C)

c1par c3par c4par c5par

950 T
regenerator
900

850
regenerator
800

750
1 1.1 1.2 1.3 1.4
λ (−)

Fig. 10. (a) Experimental values of the combustion air preheat temperatures ( C) for all configurations, parallel firing mode, all values of l and cycle time [1]. (b) Position of
thermocouple in regenerators.
 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896 3891

4. Results and discussion 6

4.1. Validation 5.5

In Table 3 the main results of the experiments are compared

τ (−)
CFD
with the numerical results for all the configurations. It is noted here 5
EXP
that all the four configurations are simulated with the exact same
set of physical models and settings, as described above. 4.5
The furnace temperature is overpredicted in the simulations
with around 70 C, however, these differences are considered to be 4
acceptable. C1 C3 C4 C5
The heat extraction by the cooling tubes (Qcool) is predicted
Fig. 11. Comparison of the maximum value of s () at the hot side of the regenerators
reasonably accurately. The predicted values are slightly higher for in the experiments (EXP) and the predicted s () of the flue gas entering the regen-
all configurations, but the values are close. The other heat losses erating burners (CFD).
(Qloss), mainly through the walls, are underpredicted in the simu-
lations. This is partly due to the calculation method of this value in
the experiments. In the analysis of the experimental results, the Finally, in Fig. 7 the mass fractions of the main species in the flue
heat flux through the stack is calculated assuming that the flue gas gas in the stack are compared for the simulations of configuration
leaving the furnace via the stack is at a temperature equal to the C5 with IET 104 and 105. It is noted that in the simulation with IET
furnace temperature. However, comparing the furnace tempera- 104 lower O2 concentrations and higher CO2 and H2O concentra-
ture (Tfurn) and the flue gas temperature in the stack (Tstack) in the tions are predicted. In the case of IET 105 the O2 content is slightly
simulations, the latter is consistently higher. This indicates that by higher and the CO2 and H2O contents are lower. Thus, by decreasing
taking the furnace temperature as the stack flue gas temperature, the IET the oxygen concentration is tending toward the experi-
the experimental heat flux through the stack is underestimated. mental and theoretical values. For both cases, the concentrations of
The values for the experimental heat losses calculated with the CO and CH4 are very low; they are in the order of 10 and 104 ppm,
predicted temperature of the gas in the stack (Qloss * ) show that this respectively. It can be concluded that in the simulation with the
error explains only part of the difference. higher value of IET (104) the combustion is complete, i.e., there is
Finally, the oxygen concentration (O2) in the flue gas is consid- hardly any CO and CH4 left in the flue gas. Therefore, no large errors
ered. In Table 3 the measured and predicted values of the O2 in the total heat distribution in the furnace due to integration errors
concentration in the flue gas from the stack are presented. Addi- are to be expected.
tionally, the theoretical value (Oth 2 ) calculated from the measured
Although no in-furnace measurements were performed for
flows of comburants for complete combustion, both using the a detailed validation, the consistency of the predictions for all the
actual fuel (DNG) and using the simulation fuel (only containing four investigated configurations confirmed the suitability of the set
CH4 as a hydrocarbon, see above), are reported. In the first place, of physical models for a numerical investigation of this furnace.
the measured values are compared with the theoretical values for
the experiments. All the measured O2 values are lower than the 4.2. Carbon monoxide emissions
theoretical values. These differences are, besides the error in the
measurement of the oxygen concentration itself, due to errors in In Fig. 8 the concentrations of CO in the flue gas from the stack
the measurement of the comburant mass flows. Secondly, and from the regenerators are presented for the experiments and
comparing the theoretical values for the actual fuel (DNG) and the the simulations. For these two flue gas sampling locations, see the
simulation fuel, it is noted that they are very similar. Finally, boxed numbers in Fig. 1.
comparing the predicted values in the simulations with the theo- All the predicted values of the CO concentration are much higher
retical values for the simulation fuel, it is noted that the predicted than the measured values. This is a deficiency of the used chemistry
values are systematically lower. This is due to the integration of the mechanism and was also mentioned as such by the developers of the
chemistry in the simulations. As stated before, the ISAT procedure mechanism [25]. Additionally, in the simulations the CO concen-
was used for shorter computational times. Gordon et al. investi- trations in the flue gas from the regenerators are taken before the
gated the maximal value for the ISAT error tolerance (IET) for regenerators (at the hot side), whereas in the experiments this CO
several mechanisms [36], concluding a maximal value of 106 for concentration is measured after the regenerators (at the cold side).
the Smooke mechanism is required. Due to too long computational Thus, part of the excess CO observed in the numerical results would
times, this value was not supportable for the current mesh in have been converted to CO2 in the regenerators if simulated.
combination with the current cluster. However, the trend in the difference between the values in the
To verify the impact of the error due to this deficiency, one case flue gas from the stack and from the regenerators is reproduced
(configuration C5 with the black burners firing, see Fig. 2) was run correctly; in the stack flue gas, relatively smaller amounts of CO are
with IET 105. In Fig. 6 the main results, as presented in Table 3, are predicted, whereas in the regenerator flue gas larger amounts of CO
plotted for the simulations with the different values of the IET. It is are predicted.
noted that a large difference in the main results is observed between The difference in the CO concentrations in the two exits is
the simulations with IET 103 and 104. In case of IET 105 the related to the path length of the mean flow to these exits. In
results do improve, however, relatively marginally. It is concluded Fig. 9(a) the pathlines along the streamlines of the mean velocity
here, that these simulations would improve by additional calculation
with IET 105, however, considering the time investment versus the
expected improvement, this was not performed. Table 4
Additionally, in Fig. 6, the results of the simulation with IET 104 Measured mass percentage (%-mass) of flue gas leaving the furnace via the stack.
with both first order and second order discretization schemes are C1 C3 C4 C5
compared. It is noted that the differences in main results between Mass percentage stack (%-mass) 27.3 31.2 22.1 22.1
these two simulations are marginal.
3892 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

(taken from the CFD simulations) of the gases flowing from the were converted to residence times by division with the local
middle firing burner in configuration C5, to the opposing regener- velocity. The residence time of the gases from this burner to the
ating burners and to the stack are presented. The burners and the regenerating burner was 0.2 s. From the firing burner to the stack
stack are indicated by the arrows. It is observed that the pathlines the residence time was 0.5 s. Again, these values include the direct
to the stack are longer than those to the regenerating burner. Also, paths only, excluding recirculation.
it is noted that part of the gases not immediately leaves the furnace, In Fig. 9(b) the CO concentration along the considered paths
but is recirculated to the root of the comburants jet. from the firing burner to the opposing regenerating burner and to
Based on the observed pathlines from CFD, the approximate the stack are presented from the simulations. The initial 1.5 m is
path lengths of the mean flow were calculated exploiting only identical for both paths, since in both cases the gases cross the
straight lines and right angles. The recirculating gases were not furnace width first. The peak in CO concentration, which can be
included in this calculation since they join the jet again at its root. regarded as the center of the combustion zone, is at approximately
Furthermore, it can be assumed that the gases that immediately 1.2 m from the burner nozzles. After this point, the CO concentra-
leave the furnace, either via a regenerating burner or via the stack, tion decreases steadily. It is observed that the CO in the gases to the
account for the largest part of the observed CO concentrations. regenerating burner is not fully converted before reaching the exit
The path lengths between all firing burners and all regenerating point. On the other hand, in the gases traveling toward the stack,
burners were calculated and subsequently averaged. The same was the CO concentration further decreases (see insert with enlarge-
done for the path lengths between the firing burners and the stack. ment in Fig. 9(b)).
The averaged path lengths from the firing burners to the regener- Thus, it is concluded that the flue gas leaving the furnace via the
ating burners and to the stack are approximately 2.0 m and 3.0 m, stack had a long enough residence time (due to its longer path
respectively. For the firing burner in Fig. 9(a) these path lengths length) to have most of the CO converted to CO2.

Fig. 12. Pathlines from all grid cells of the firing burner inlets of all configurations colored by the local temperature (K) viewed toward the vertical midplane of the furnace.
 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896 3893

Cooling tube heat extraction (kW)

1180
120 c4par c5par
EXP CFD
1170

Temperature (°C)
110
1160

100 1150

1140
90
1130

80 1120
C1 C3 C4 C5 0 4 8 12 16 20
Fig. 13. Cooling tube heat extraction rates (kW) comparison for the experiments (EXP) Thermocouple #
and simulations (CFD).

4.3. Combustion air preheat temperature 1 3 5 7 10 12 14 16

C4
2 4 6 8 17
In the experiments a difference in the preheat temperature of
9 11 13 15 18
the combustion air was observed for the different burner configu-
rations. In Fig. 10 the experimental values of the preheat temper-
ature of the combustion air are presented. It was observed that the
preheat temperatures for the C4 and C5 configurations are higher
than those for configurations C1 and C3. 10 12 14
1 3 5 7 16
In Fig. 11 the maximum value of the normalized temperature (s) C5
2 6 8 13 17
at the hot side of the regenerator in the experiments is compared
with the predicted flue gas temperature entering the regenerating 4 9 11 15 18
burners. The normalized temperature s is defined in Equation (2).

T
s ¼ (2) Fig. 15. Predicted temperatures (K) for all thermocouples in burner configurations C4
Tref and C5. The four rectangles represent the four side walls of the furnace with the
numbers indicating the positions of the thermocouples.
where T is the temperature (K) and Tref the reference temperature
(273.15 K). due to the fact that in these configurations the three higher burners
The numerical values are higher than the experimental values. are positioned in the upper level, whereas in C4 and C5 they are
This is due to two reasons. In the first place, the thermocouples positioned in the middle level, see also Fig. 2.
inserted in the regenerators only penetrate until the end of the In Fig. 12 pathlines along the streamlines of the mean velocity
regenerators, while in the simulations the flue gas temperature is are presented from all grid cells of the firing burner inlets of all the
taken at the nozzle, see also Fig. 10(b). In the second place, the configurations, colored by the local temperature. It is observed that
measured temperatures will be slightly lower than the actual gas in configuration C4 and C5, as expected, less of these pathlines
temperatures due to radiative heat losses of the thermocouples. leave the furnace via the stack in the roof. This is due to an addi-
The reported experimental measurements were not corrected for tional recirculation zone above the firing burners compared to
these radiative losses. configuration C1 and C3.
However, in both the experiments and the simulations the It is concluded that in configurations C1 and C3, due to the firing
values of s are higher for configurations C4 and C5. This is explained burners positioned closer to the stack and the absence of a recir-
by the amount of flue gas leaving the furnace via the stack. In culation zone between these firing burners and the stack, signifi-
Table 4 the experimental values of the mass percentage of flue gas cantly more flue gas leaves the furnace via the stack and relatively
leaving the furnace via the stack is presented for the four config- less hot flue gas is available for regeneration, leading to lower
urations. It is noted here that the mass percentage of flue gas to the preheat temperatures of the combustion air.
stack is set as a boundary condition in the simulations.
It is observed that for both C1 and C3 relatively larger
Cooling tube heat extraction (kW)

percentages of the flue gases leave the furnace via the stack. This is 120
c1par c3par c4par c5par

110
0.996
Temperature uniformity (−)

0.994 100

0.992
90
0.99
80
0.988
0.986 0.988 0.99 0.992 0.994 0.996
EXP CFD
0.986 Temperature uniformity (−)
C1 C3 C4 C5
Fig. 16. Cooling tube heat extraction rate (kW) versus temperature uniformity () for
Fig. 14. Temperature uniformity () comparison for experiments (EXP) and simula- all burner configurations in parallel firing mode. Black symbols are the experimental
tions (CFD). values, gray symbols represent the numerical values.
3894 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

C1 Parallel C3 Parallel

120 120
z/d=4
z/d=6
z/d=8
z−velocity (m/s)

z−velocity (m/s)
80 80

40 40

0 0
−20 −20
0 0.2 0.4 0.6 0 0.2 0.4 0.6
x−distance (m) x−distance (m)

1300 1300

1100 1100
Temperature (°C)

Temperature (°C)

900 900

700 700
z/d=4
500 z/d=6 500
z/d=8
300 300
0 0.2 0.4 0.6 0 0.2 0.4 0.6
x−distance (m) x−distance (m)

0.7 0.7
z/d=4
z/d=6
z/d=8
0.5 ξstoich 0.5
ξ (−)

ξ (−)

0.3 0.3

0.1 0.1
0 0.2 0.4 0.6 0 0.2 0.4 0.6
x−distance (m) x−distance (m)

Fig. 17. Development of the velocity in the z -direction (m/s), temperature ( C) and mixture fraction (x, ) for the lower burners in configurations C1 and C3 parallel. The data are
plotted over the three gray lines depicted in the sketch of the numerical domain of configuration C3. The (green) shaded plane is the vertical symmetry plane. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)

4.4. Cooling tube heat extraction However, the trend between the different cases is well captured by
the simulations.
In Fig. 13 the experimental and numerical values of the cooling The cooling tube heat extraction is directly related to the total
tube heat extraction rate of the four configurations are compared. mass flow of cooling air to the cooling tubes. In the total mass flow
The predicted values are slightly higher than the experimental. of cooling air the same trend is observed as in Fig. 13. Since the
 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896 3895

measured total cooling air mass flows are set as boundary condi- In Fig. 17 the velocity in the (axial) z -direction, the temperature
tions in the simulations, the trends in the total cooling tube heat and the mixture fraction are plotted on lines at three distances from
extraction rates are also reproduced in the simulations. the nozzles of these burners. The distance from the burner nozzles z
is normalized by the diameter of the air nozzles d (0.02 m). The fuel
4.5. Temperature uniformity nozzles are positioned at x ¼ 0.295 m and 0.75 m and are sur-
rounded by two high momentum air nozzles. The mixture fraction
In Fig. 14 the experimental and numerical values of the temper- was calculated by applying Bilger’s formula [37].
ature uniformity of the four configurations are compared. The The point of confluence is defined as the point where the low-
temperature uniformity (Tu) is defined as in the following equation. momentum jet is absorbed by the high momentum jet(s) [38].
The further the point of confluence from the nozzles, the more
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N 
u 1 X ðTi  TÞ 2 internal flue gas recirculation is present before the fuel and oxidizer
Tu ¼ 1  t (3) jets merge and lower NO emissions are expected.
N i¼1 T For the burner at x ¼ 0.295 m no clear differences can be
observed between configuration C3 and C1, both seem to have
where N is the total number of temperature measurement posi-
a point of confluence at z/d ¼ 6. The burner at x ¼ 0.75 m has a point
tions, Ti is the temperature in the ith position and T is the mean of
of confluence at z/d ¼ 8 in both configurations (only configuration
all the temperature measurements. The value of the Tu is between
C3 is shown in Fig. 17, but the burner of configuration C1 showed
0 and 1, where the value 1 indicates a perfectly uniform furnace.
the same value). Concluding, it is confirmed that the jets of the
The trend in the temperature uniformities agrees well between
burners in configurations C1 and C3 show similar merging
the experiments and simulations. Especially, the observation that
behavior, which is in agreement with the proposed explanation for
configurations C3 and C4 achieve higher temperature uniformities,
the similar NO concentrations in their flue gases. Also, no distinct
compared to configurations C1 and C5, is reproduced. However,
differences in the maximum temperatures could be identified
there is an opposing trend in the temperature uniformities of C3
between the two configurations. Finally, it can be expected that the
versus C4 and C1 versus C5.
middle burner (at x ¼ 0.75 m) produces less NO than the two outer
In Fig. 15 the predicted values for the wall temperatures are
burners.
presented for burner configuration C4 and C5. The exact positions
of the thermocouples are denoted in the drawings of the four side
5. Conclusions
walls of the furnace.
It can be observed, comparing the thermocouples in the highest
A numerical study has been performed for four different burner
level (numbers 1, 3 and 5) with those in the middle or lower level
configurations in a furnace equipped with three pairs of flameless
(numbers 2, 4 and 6), that the difference in the temperature is
combustion burners. A carefully selected set of physical models
larger for configuration C5, which results in a lower temperature
proved to be adequate to reproduce the main results observed in
uniformity. Actually, in configuration C4, thermocouples 1, 3 and 5
the furnace experiments. The simulations have been validated
have relatively higher values. This is due to the fact that a stronger
against the main results of the furnace experiments and by
recirculation zone in this configuration results in lower tempera-
consistently capturing the trends in these results for the different
tures in the upper zone of the furnace, see also Fig. 12. This also
burner configurations. Moreover, it was found that due to relatively
explains the lower temperature uniformity in configuration C1,
low Reynolds numbers in the cooling air flow in the annulus of the
where the least pronounced recirculation zones can be identified.
cooling tubes, predictions of the heat extraction rates of these
For the values of the thermocouples in the opposite wall (numbers
cooling tubes were improved by treating the flow in the cooling
10 to 15) the same behavior can be observed.
tubes as laminar. Furthermore, the applied error tolerance of the
ISAT procedure was shown to be insufficient for accurate species
4.6. Temperature uniformity versus cooling tube heat extraction concentration predictions, however, based on analysis of the main
species concentrations in the flue gas, this inaccuracy did not
In the experimental results, a linear correlation was observed influence the overall predictions.
between the cooling tube heat extraction and temperature The most important experimental results have been investi-
uniformity [1]. In Fig. 16 these values are plotted for both the gated using the simulations. In the first place, a longer path length
experiments and the simulations. It is observed that the decreasing from the firing burners to the stack, compared to the path length to
trend in the cooling tube heat extraction rate with increasing value the regenerating burners, explained the lower CO emissions in the
of the temperature uniformity is not reproduced. Due to the choice flue gas in the stack. Secondly, it was found that a recirculation zone
of experiments simulated it is not possible to either confirm or between the upper firing burners and the stack in configurations C4
contradict the observations of the experimental results. In order to and C5 resulted in a smaller fraction of the flue gases leaving the
be able to do this, additional simulations of cases exploiting more furnace via the stack compared to the other configurations. Thus,
variation in the values of the cooling tube heat extraction and the a larger fraction left the furnace via the regenerating burners and
temperature uniformity are required. Due to time limitations, these this resulted in higher preheat temperatures of the combustion air.
additional simulations are currently not performed. Furthermore, the total cooling tube heat extraction was directly
related to the total amount of cooling air. In the fourth place, more
4.7. Point of confluence pronounced recirculation zones in configurations C3 and C4 led to
higher temperature uniformities in the furnace. Finally, it was
In a previous study on a comparable furnace [14] the point of confirmed that the jets of the burners in configurations C1 and C3
confluence of the comburant jets was indicated as an important showed similar merging behavior, leading to similar NO emissions,
factor for the NO formation. In this study the NO emissions were as have indeed been observed in the experiments.
similar for all configurations and operating conditions [1]. To verify It can be concluded that by the numerical study more insight is
whether the jets show similar merging behavior in this furnace the attained regarding the effects of burner positioning in a multi-
firing burners in the lowest level of configurations C1 and C3 are burner flameless combustion furnace. In the future, the simula-
compared. tions can be improved by further decreasing the error tolerance of
3896 B. Danon et al. / Applied Thermal Engineering 31 (2011) 3885e3896

the ISAT procedure. Also, in order to verify the observation in the a pilot-scale furnace. In: Proceedings of the 6th High Temperature Air and
Gasification Conference (HTACG). Essen, Germany; 2005.
experiments of a decreasing cooling tube heat extraction with
[18] D. Lupant, B. Pesenti, P. Lybaert, Assessment of combustion models of a self-
increasing temperature uniformity, additional simulations are regenerative flameless oxidation burner Mons, Belgium. in: P. Lybaert (Ed.),
required. Ninetheenth Journees d’etudes of the Belgian Section of the Combustion
Institute (2006).
[19] T. Ishii, C. Zhang, S. Sugiyama, Numerical simulations of highly preheated air
Acknowledgements combustion in an industrial furnace, J. Energy Res. Technol. 120 (1998)
276e284.
[20] T. Ishii, C. Zhang, Y. Hino, Numerical study of the performance of a regenera-
This project has been financially supported by the Dutch Tech- tive furnace, Heat Transfer Eng. 23 (2002) 23e33.
nology Foundation (STW) and the Dutch Flame Foundation (NVV). [21] N. Stockwell, C. Zhang, T. Ishii, Y. Hino, Numerical simulations of turbulent
non-premixed combustion in a regenerative furnace, ISIJ Int. 41 (10) (2001)
1272e1281.
References [22] E.S. Cho, B. Danon, W. de Jong, D. Roekaerts, Behavior of a 300 kwth regen-
erative multi-burner flameless oxidation furnace. Appl. Energy (in press)
[1] B. Danon, E.S. Cho, W. de Jong, D. Roekaerts, Parametric optimization study of 2011;doi10.1016/j.apenergy.2011.06.039.
a multi-burner flameless combustion furnace, Appl. Therm. Eng. 31 (2011) [23] Fluent 6.3 User’s Guide. Fluent, Lebanon e New Hampshire, 2006.
3000e3008. [24] T.H. Shih, W. Liou, A. Shabbir, Z. Yang, J. Zhu, A new keε eddy viscosity model
[2] J. Wünning, J. Wünning, Flameless oxidation to reduce thermal NO-formation, for high Reynolds number turbulent flows, Comput. Fluids 24 (1995)
Prog. Energy Combust. Sci. 23 (1997) 81e94. 227e238.
[3] H. Tsuji, A. Gupta, T. Hasegawa, M. Katsuki, K. Kishimoto, M. Morita, High [25] M. Smooke, I. Puri, K. Seshadri, A comparison between numerical calculations
Temperature Air Combustion: from Energy Conservation to Pollution Reduc- and experimental measurements of the structure of a counterflow diffusion
tion. CRC Press, 2003. flame burning diluted methane in diluted air, in: Proceedings of the Twenty-
[4] A. Cavaliere, M. De Joannon, Mild combustion, Prog. Energy Combust. Sci. 30 First Symposium (International) on Combustion, 21, 1986, pp. 1783e1792.
(2004) 329e366. [26] Physical Properties of Natural Gases. Nederlandse Gasunie N.V., 1980.
[5] J. Miller, C. Bowman, Mechanism and modeling of nitrogen chemistry in [27] I. Ertesvåg, B. Magnussen, The eddy dissipation turbulence energy cascade
combustion, Prog. Energy Combust. Sci. 15 (1989) 287e338. model, Combust. Sci. Technol. 159 (2000) 213e235.
[6] P. Medwell, P. Kalt, B. Dally, Simultaneous imaging of OH, formaldehyde, and [28] F. Christo, B. Dally, Modeling turbulent reacting jets issuing into a hot and
temperature of turbulent nonpremixed jet flames in a heated and diluted diluted coflow, Combust. Flame 142 (2005) 117e129.
coflow, Combust. Flame 148 (2007) 48e61. [29] J.P. Kim, U. Schnell, G. Scheffknecht, A. Benim, Numerical modelling of mild
[7] E. Oldenhof, M. Tummers, E. van Veen, D. Roekaerts, Ignition kernel formation combustion for coal, Prog. Comput. Fluid Dyn. 7 (2007) 337e346.
and lift-off behaviour of jet-in-hot-coflow flames, Combust. Flame 157 (6) [30] A. Parente, C. Galletti, L. Tognotti, Effect of the combustion model and kinetic
(2010) 1167e1178. mechanism on the MILD combustion in an industrial burner fed with
[8] N. Rafidi, W. Blasiak, Heat transfer characteristics of HiTAC heating furnace hydrogen enriched fuels, Int. J. Hydrogen Energy 33 (2008) 7553e7564.
using regenerative burners, Appl. Therm. Eng. 26 (2006) 2027e2034. [31] G. Stefanidis, B. Merci, G. Heynderickx, G. Marin, CFD simulations of steam
[9] J. Adolfi, M. Boss, S. Santos, Commissioning Report HECeEEC Furnace and cracking furnaces using detailed combustion mechanisms, Comput. Chem.
Burners (Part 2 of 2). Tech. Rep. C108/y/2. IFRF Research Station, 2004. Eng. 30 (2006) 635e649.
[10] D. Lupant, B. Pesenti, P. Evrard, P. Lybaert, Numerical and experimental [32] J. Kim, U. Schnell, G. Scheffknecht, Comparison of different global reaction
characterization of a self-regenerative flameless oxidation burner operation in mechanisms for mild combustion of natural gas, Combust. Sci. Technol. 180
a pilot-scale furnace, Combust. Sci. Technol. 179 (2007) 437e453. (2008) 565e592.
[11] C. Zhang, T. Ishii, Y. Hino, S. Sugiyama, The numerical and experimental study [33] S. Pope, Computationally efficient implementation of combustion chemistry
of non-premixed combustion flames in regenerative furnaces, J. Heat Transfer using in situ adaptive tabulation, Combust. Theor. Model. 1 (1997) 41e63.
122 (2000) 287e293. [34] B. Danon, W. de Jong, D. Roekaerts, Experimental and numerical investigation
[12] W. Yang, W. Blasiak, CFD as applied to high temperature air combustion in of a flox combustor firing low calorific value gases, Combust. Sci. Technol. 182
industrial furnaces, IFRF Combust. J. (2006) 200603. (9) (2010) 1261e1278.
[13] W. Yang, M. Mörtberg, W. Blasiak, Influence of flame configurations on flame [35] A. De, E. Oldenhof, P. Sathiah, D. Roekaerts, Numerical simulation of Delft-Jet-
properties and NO emissions in combustion with high-temperature air, Scand. in-Hot-Coflow (DJHC) flames using the Eddy Dissipation Concept model for
J. Metall. 34 (2005) 7e15. turbulence-chemistry interaction. Flow Turbul. Combust. (in press) 2011;
[14] B. Danon, A. Swiderski, W. de Jong, W. Yang, D. Roekaerts, Emission and doi10.1007/s10494-011-9337-0.
efficiency comparison of different firing modes in a furnace with four HiTAC [36] R. Gordon, A. Masri, S. Pope, G. Goldin, Transport budgets in turbulent lifted
burners, Combust. Sci. Technol. 183 (7) (2011b) 686e703. flames of methane autoignition in a vitiated co-flow, Combust. Flame 151
[15] R. Hekkens, M. Mancini, Non-isothermal CFD Model of the HEC Burner and (2007) 495e511.
Furnace. Tech. Rep. G108/y/2. IFRF Research Station, 2004. [37] R. Bilger, S. Stårner, R. Kee, On reduced mechanisms for methane-air
[16] R. Hekkens, Non-isothermal CFD Model of the HEC Burner and Furnace combustion in nonpremixed flames, Combust. Flame 80 (1990) 135e149.
(Additional Calculation). Tech. Rep. G108/y/3. IFRF Research Station, 2004. [38] E. Grandmaison, I. Yimer, H. Becker, A. Sobiesiak, The strong-jet/weak-jet
[17] D. Lupant, B. Pesenti, P. Evrard, P. Lybaert, Numerical and experimental problem and aerodynamic modeling of the CGRI burner, Combust. Flame
characterization of a self-regenerative flameless oxidation burner operation in 114 (1998) 381e396.