ARTICLE IN PRESS CNF:6965

JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.1 (1-15)

Combustion and Flame ••• (••••) •••–•••

www.elsevier.com/locate/combustflame

Scaling of NOx emissions from a laboratory-scale mild

combustion furnace
G.G. Szegö ∗ , B.B. Dally, G.J. Nathan
School of Mechanical Engineering, The University of Adelaide, South Australia 5005, Australia
Received 10 September 2007; received in revised form 2 February 2008; accepted 7 February 2008

Abstract
A systematic experimental campaign has been carried out to investigate the scaling of NOx emissions from a
moderate or intense low-oxygen dilution (MILD) combustion furnace operating with a parallel jet burner system
in which the reactants and the exhaust ports are all mounted on the same wall. Its maximum capacity was 20 kW
from the fuel and 3.3 kW from air preheat, with a turndown ratio of 1:3. The burner system was configured to
achieve high dilution of the incoming reactants. A comprehensive data set comprising 191 global measurements of
temperature and exhaust gas emissions is presented, together with temperature contours on the furnace centerline
plane. It was found that, although heat extraction, air preheat, excess air, firing rate, dilution, and fuel type all
affect global NOx emissions, they do not control NOx scaling. The combined effects of these global parameters
can be ultimately characterized by a furnace temperature and a global residence time. A temperature–time scaling
approach, previously reported for open jet diffusion flames, proved to be a useful tool for comparison of NOx
emissions from highly diluted furnace environments regardless of the furnace/burner geometries. Regression-
based predictions found the characteristic temperature to correlate with 85% of the data with an accuracy of only
±50%. The leading-order approach also showed that the jet exit Froude number is of limited value for NOx scaling
in the MILD regime. Because of the weak dependence on temperature observed in the data and the moderate
magnitude of the measured temperatures, it is deduced that the prompt-NO and/or N2 O-intermediate pathways
are of significance comparable to that of the thermal-NO pathway. The analysis also suggests that NOx formation
is controlled neither by kinetics nor by mixing, and hence the conditions inside this furnace approach or span the
range in which Damköhler numbers are of order unity, Da = O(1).
© 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Keywords: MILD furnace; Moderate temperatures; NOx scaling

1. Introduction change, have intensified discussions about the com-

bustion of fossil and alternative fuels. Combustion-
Rising concentrations of atmospheric greenhouse generated CO2 emissions account for about 80% of
gases, which are widely accepted to lead to climate anthropogenic greenhouse gases [1], so their mitiga-
tion is vital, and an aim of policies around the world.
* Corresponding author. Fax: +61 8 8303 4367. Improving energy efficiency is often the most cost-
E-mail address: george.szego@adelaide.edu.au effective way to reduce CO2 emissions [2]. A wide
(G.G. Szegö). range of low- or zero-carbon-emission technologies
0010-2180/$ – see front matter © 2008 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2008.02.001

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS CNF:6965
JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.2 (1-15)
2 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••

are already available [3]. However, the abatement

of pollutants often comes at the price of efficiency
losses. Moderate or intense low-oxygen dilution
(MILD) combustion [4] is a credible candidate to si-
multaneously meet thermal efficiency needs and pol-
lutant emission restrictions. Under MILD conditions,
reactants are highly diluted with hot combustion prod-
ucts, causing reactions to occur in a distributed reac-
tion zone with reduced peak temperatures [5]. As a
consequence, the temperature distribution is nearly
uniform, the net radiation flux can be enhanced, and
pollutant emissions, NOx in particular, are lower than
those from conventional flames.
Over the past few decades, MILD combustion
technology has been implemented at full scale in var-
ious industrial sectors [6] and tested at pilot scale in
other applications [7–10] and for other fuel types [11].
Nevertheless, despite considerable industrial success,
important aspects of MILD combustion remain un-
known. For example, little is known about the scaling
relations of NOx emissions in complex recirculating
flows with low oxygen concentrations at moderate
temperature ranges (800 ◦ C < T < 1400 ◦ C), typi-
cally found in MILD conditions.
The scaling of NOx has been comprehensively in-
vestigated in unconfined nonpremixed turbulent jet Fig. 1. Schematic diagram of MILD combustion furnace and
flames [12], and also given considerable attention parallel jet burner system.
in enclosed furnace environments with conventional
combustion systems [13–16]. These studies have examine the NOx emissions from a laboratory-scale
found a strong correlation of normalized NOx emis- MILD combustion furnace (MCF), operating with a
sion indices with fuel jet exit Froude number and of parallel jet burner system in which the reactants and
NOx formation rates with global temperatures and the exhaust ports are all mounted on the same wall.
residence times. These relationships are useful, not The maximum furnace capacity is about 20 kW from
only for understanding which geometrical or opera- the fuel and 3.3 kW from air preheat. First, results of
tional parameters control pollutant formation, but also thermal field measurements are presented to provide
for assessing the relative importance of the alternative some insight into the major NOx formation and/or de-
NOx formation mechanisms. struction regions. Then, the effects of fuel dilution on
More specifically related to the MILD regime, So- global NOx emissions are discussed. Finally, a com-
biesiak et al. [17] obtained an empirical correlation prehensive experimental data set, which covers a va-
for NOx emissions as a function of excess air, air riety of operating conditions, is used to investigate the
preheat temperature, burner turndown ratio, and ex- scaling of NOx emissions with global parameters.
haust gas temperature. Although relatively few exper-
imental data points were available for the analysis, a
power-law type expression was used to show that the 2. Experimental arrangement and methodology
exhaust temperature term had the strongest effect on
NOx emissions. Other researchers [18,19] attempted 2.1. MILD combustion furnace and burner
to relate NOx emission trends to burner parameters,
such as fuel injection angle and fuel-to-air velocity ra- Fig. 1 shows a schematic diagram of the MILD
tios. The aforementioned MILD combustion studies combustion furnace and the parallel jet burner config-
focused on minimizing NOx emissions of a particu- uration built for this project. The combustion cham-
lar burner design and led to the patenting [20–22] and ber has a square cross section of 280 × 280 mm2
commercialization of the technology. and a height of 585 mm. It is well insulated, with
To the best of our knowledge, the results of NOx four layers of 38-mm-thick high-temperature ceramic
scaling relations have not been previously reported fiberboards that allow only about 20% of heat to be
for MILD combustion conditions, either in enclosed lost through the walls. This assists in the establish-
or in open flame systems. The current work aims to ment and stability of the MILD regime and results in

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
a warm-up time of about 3 h from a cold state un-
til steady-state operation. The inner surfaces of the
combustion chamber were coated with a commercial
product, “Kaowool J-Coat” rigidizer, to improve the
resistance to erosion caused by high gas velocities.
The MCF has full optical access through five open-
ings (120 × 76 mm2 ) equally spaced along the ver-
tical direction on three sides of the furnace. These
openings can accommodate interchangeable insulat-
ing window plugs or UV-grade fused silica windows.
The chamber is slightly pressurized to prevent air
ingress, and high-temperature gaskets were used to
seal all surface joints for safety. A U-shaped cool-
ing loop with variable heat exchange area was used
to control the heat load. The heat exchanger can be
inserted through any of the five window openings (po-
sitions A1–A5).
The configuration of reactants and exhaust ports
was optimized using a computational fluid dynam-
ics (CFD) modeling study [23]. The furnace was de-
signed for a maximum capacity of 20 kW from the
fuel and 3.3 kW from air preheat with a turndown ra-
tio of 1:3. The air was preheated with a Leister LE
3000 electrical air heater that has a built-in poten-
tiometer for air temperature control. The burner con-
sists of a single air nozzle on the axis of the furnace,
and four exhaust ports and four fuel ports arranged
symmetrically in ring patterns (see Fig. 1). The dis-
tance between the center of the air nozzle and the
centers of the exhaust ports and fuel nozzles are 55
and 110 mm, respectively. In this arrangement the in-
coming air stream is separated from the fuel streams
by outgoing exhaust gases, much as in the fuel direct
injection (FDI) concept introduced by Nakamachi et
al. [24] and in the IFRF burner design [25]. This
ensures that the reactants are diluted with products
before any combustion reaction can occur. The final
straight sections of the air and fuel supply lines have
length-to-diameter ratios of L/D ≈ 10 and L/d ≈ 18
(id = 2 mm), respectively.
2.2. Temperature and global emission measurements
Time-averaged temperatures were measured with
bare, fine-wire type R (Pt/Pt–13% Rh) thermocou-
ples of 254-µm-diameter wires and 1.2-mm-diameter
beads under steady-state conditions. This is sufficient
for reliable measurements of the mean temperature
after correction for radiation, given the low temper-
ature fluctuations in the MILD regime, and provides
satisfactory spatial resolution while being sufficiently
robust to avoid probe breakage. For accurate posi-
tioning, the thermocouple wires were supported by
a 6-mm-diameter rigid ceramic sheath, with the final
5 mm left exposed. The differential voltage signal was
continuously sampled at 1 Hz and automatically con-
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.3 (1-15)
3
verted to temperature units with a USB-TC data log-
ger. According to the manufacturer’s specifications,
the uncertainties associated with linearization, cold-
junction compensation, and system noise produce a
maximum error of ±1.07 ◦ C in a measurement range
from 250 to 1768 ◦ C. This represents only 0.13% er-
ror in the worst case.
Eighteen thermocouple ports (see Fig. 1) were
equally spaced in the vertical direction, through
which a thermocouple could be traversed. Typically,
temperatures were monitored simultaneously by fixed
thermocouples in two regions of interest, one at the
top section and one at the bottom section of the com-
bustion chamber. The center of the air nozzle exit
plane is defined to be the origin of a Cartesian co-
ordinate system (x, y, z). The top temperature was
measured on the furnace centerline (0, 0, 542.5) and
was denoted as the furnace reference temperature.
The bottom temperature was measured flush with
the inner surface (0, 140, 42.5) and was denoted as
the reference wall temperature. Temperature profiles
were also measured for a selected baseline case. The
thermocouple probes were traversed in the radial di-
rection (y-direction) across different planes at each
hole to provide a total of 72 points.
The correction to the temperature measurements
for radiation was determined from a steady-state en-
ergy balance on the thermocouple bead surface. The
actual gas temperature (Tg ) was determined from the
measured temperature (Ttcb ) and net radiation flux
between diffuse-gray surfaces in an enclosure accord-
ing to

Atcb εtcb N 4 4
w=1 σ G1w (Ttcb − Tw )
Tg = Ttcb + , (1)
h
where Atcb is the thermocouple bead surface area,
εtcb is the thermocouple bead surface emissivity, N
is the total number of surfaces (i.e., walls) in the en-
closure, σ is the Stefan–Boltzmann constant, G1w is
the fraction of radiation emitted by the thermocouple
bead that is incident on a particular wall surface w
in the furnace enclosure and is absorbed [26], Tw is
the surface temperature of wall w in the enclosure,
and h is the convection coefficient calculated for a
sphere [27]. Catalytic and oxidation effects were ne-
glected, as well as conductive heat transfer along the
ceramic sheath. The G1w correction factor was cal-
culated for each probe position. The method led to a
maximum correction of 9% and a typical correction
of less than 2% between raw and actual gas tempera-
tures, so that the total uncertainty is within ±1.2%.
Exhaust gas composition was measured inside
the exhaust pipe with a TESTO 350XL portable
gas analyzer. The gas analyzer was equipped with
electrochemical sensors with nominal accuracies,
respectively, for CO (±5%: 0–2000 ppm, ±10%:
 ARTICLE IN PRESS CNF:6965
JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.4 (1-15)
4 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••

2001–10000 ppm), NO (±5 ppm: 0–99 ppm), NO2

(±5 ppm: 0–99.9 ppm), O2 (±0.8%: 0–25%), and
UHC (<400 ppm: 100–4000 ppm CH4 ) and with
a nondispersive infrared (NDIR) sensor for CO2
(±0.3% m.v. + 0.44: 0–25%). The analyzer was
checked with a calibration gas to yield total NOx
emission accuracies better than ±4 ppm for the range
from 0 to 70 ppm, in which all present data lie.
The combustion products were cooled and filtered
to remove moisture and particles before reaching the
sensing cells. Total NOx emission (NO + NO2 ) con-
centrations are reported by volume on a dry basis
corrected to 3% O2 concentration. Emission indices
are expressed as the ratio of pollutant formed to the
mass of fuel input [28]. The emission indices were
determined from measured species concentrations as
XNO MWNO + XNO2 MWNO2
EINOx [g/kg fuel] =
(XCO + XCO2 )MWCx Hy
× 1000, (2)
XCO MWCO
EICO [g/kg fuel] =
(XCO + XCO2 )MWCx Hy
× 1000, (3)
where EINOx is the NOx emission index, Xi is the
mole fraction of species i, and MW is the molec-
ular weight of species i or of the hydrocarbon fuel
(Cx Hy ). For most operating conditions, unburned hy-
drocarbons (UHC) were either below the detection
limit of 100 ppm or below the minimum O2 concen-
tration (≈2%) required for a reliable measurement.
Therefore, for consistency, UHC was neglected as a
possible source of carbon in the formulation. Due
to the short length of the flue gas sampling probe
and hose (≈2.5 m), the manufacturer reports that the
amount of NO2 absorbed in the condensate trap is in-
significant. Consequently, both measured components
of NOx can be considered to be reliable.

2.3. Operating conditions

The parametric study investigated a total of 191
measurement points under various test conditions. In
all cases, thermal equilibrium was ensured and sta-
ble MILD combustion was achieved, based on the
requirement of no visible flame and NOx emission
levels being below 70 ppm throughout the experi-
ment. The oxidant was air, whose temperature was
varied from ambient to 780 ◦ C. Either commercial
natural gas (NG) or liquefied petroleum gas (LPG)
was used as the fuel to provide thermal inputs over
the range 7.5–20 kW. Table 1 shows typical compo-
sition and properties of both fuels. Two alternative-
sized fuel nozzles were employed, 2.0 or 3.0 mm
in diameter. The global equivalence ratio (φ) ranged
Fig. 2. Schematic diagram of heat exchanger insertion posi-
tions.
from 0.67 to 0.98, which is equivalent to an excess air
range of 33–2%. The location of the heat exchanger
was varied from the top window A1 (centerline at
z = 524.5 mm) to the bottom window A4 (centerline
at z = 176.5 mm) and its exposed surface area varied
from 0.015 m2 for 25% insertion to 0.06 m2 for 100%
insertion (Fig. 2). The dilution of the fuel stream was
varied, using either N2 or CO2 , over the range 0–76%
by mass. Table 2 presents a summary of all test con-
ditions.

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS CNF:6965
JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.5 (1-15)
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–••• 5

Table 1
Typical fuel properties and composition
Fuel LHVb (A/F)st c MWd N2 e CO2 e CH4 e C2 H6 e C3 H8 e C4 H10 e C5 H12 e C6 H14 e
NGa 51.154 17.11 17.71 1.278 2.084 91.36 4.364 0.62 0.20 0.055 0.04
LPGa 45.993 15.66 44.07 0.73 0.05 – 1.12 96.32 1.78 – –
a Gas analysis provided by Origin Energy Australia.
b Lower heating value (MJ/kg).
c Stoichiometric air-to-fuel ratio (kg/kg).
d Molecular weight (kg/kmol).
e Mole fraction (Xi ) (%).

Table 2
Summary of all test conditions
Fuel type Data d Ydil Q̇fuel φ Q̇hx Tair Tfurnace τG
points (mm) (%) (kW) (%) (◦ C) (◦ C) (s)
NG 145 2.0 0 7.6–20.4 0.67–0.98 17–51 20–780 768–1368 2.8–11.4
2 3.0 0 15.6 0.80 18 450 1295 3.9
NG/CO2 8 3.0 38–76 15.6 0.80 16–34 454 1053–1309 1.8–3.1
NG/N2 3 2.0 41–70 7.9–10.5 0.80 25–29 23–445 951–1129 3.7–4.5
5 3.0 33–71 15.6 0.80 18 450 1285 1.6–3.0
LPG 1 3.0 0 15.0 0.80 21 450 1290 3.9
LPG/CO2 10 2.0 31–71 9.9–14.4 0.75–0.81 24–41 458 1031–1261 1.3–3.5
4 3.0 50–74 15.0 0.80 20 450 1268–1306 1.1–2.0
LPG/N2 9 2.0 19–65 14.7 0.80 18–46 447 1046–1298 1.1–3.0
4 3.0 45–70 15.0 0.79 20 450 1285–1326 0.9–1.8

The baseline case of this project was defined as the
2.0-mm burner with a 15-kW thermal input (Q̇fuel ) of
natural gas, an equivalence ratio of 0.80, and 450 ◦ C
air preheat (Tair ). For this condition the inlet jet ve-
locities for air and fuel are U = 29.1 m/s and u =
17.3 m/s, respectively. The reference position for the
heat exchanger was position A3 (centerline at z =
292.5 mm) with 25% insertion.
The furnace was started from a cold state with a
nonpremixed flame stabilized by a bluff body. When
the temperature at the lower corner of the furnace,
which was identified as safety TC in Fig. 1, exceeded
autoignition (≈800 ◦ C) and the exhaust temperature
exceeded a set threshold (≈600 ◦ C), the furnace was
switched to the MILD combustion mode.
3. Results
3.1. Temperature and pollutant emissions during
transition
Fig. 3 shows the temperature and emissions mon-
itored as a function of time for the baseline case. The
dashed line represents the time when MILD com-
bustion was activated, the bluff-body was retracted,
and no flame front was visible [29]. The successful
transition between the two regimes is evidenced by
the sudden decrease in furnace reference temperature
and by the fourfold drop in NOx emissions. The dif-
ferences seen in the levels of O2 , and consequently
of CO2 , during the transition to the MILD regime
were caused by a slight increase in pressure drop
in the fuel supply system. However, it is clear that
these changes did not significantly influence perfor-
mance. Fine-tuning of the fuel flow rate was typically
needed to compensate for small pressure variations
in the supply line. For excess air levels of 25% or
more (φ  0.8), CO and UHC emissions were be-
low the detection limit. From the time the furnace
was switched to the MILD regime, about 1.5 h was
needed to reach steady-state MILD combustion con-
ditions. Under steady-state conditions, the tempera-
ture difference between the top and the bottom sec-
tions of the furnace was around 110 ◦ C for the base-
line case. The low temperature gradient throughout
the furnace is an important characteristic of this com-
bustion regime. The performance characteristics of
the furnace/burner have been discussed previously in
more detail [29]. Together they demonstrate that the
present furnace/burner configuration is able to sus-
tain stable MILD combustion, while maintaining high
modulation ratios for various thermal input, heat ex-
traction, and air preheat conditions. The investigation
also identified cases in which the MILD regime was
established without air preheat.

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
6 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
Fig. 3. Temperature and emissions recorded during the tran-
sition from conventional to MILD combustion for the base-
line operating conditions (d = 2.0 mm, Q̇fuel = 15 kW NG,
φ = 0.80, Tair = 450 ◦ C, 25% insertion at position A3).
3.2. Thermal field measurements
Measurement of the thermal field is a necessary
step in identifying and analyzing potential regions of
pollutant formation and/or destruction. Fig. 4 presents
the measured temperature distribution for the base-
line case in the x–z plane through the furnace cen-
terline. The solid blue arrow represents the incoming
air stream (x = 0 mm), the solid green arrows the
fuel streams (x = −110 mm and x = 110 mm), and
the dotted red arrows the outgoing exhaust stream
(x = −55 mm and x = 55 mm). The dashed line
shows the heat exchanger location (window A3) and
insertion length. The temperature contours were gen-
erated using a Kriging-type interpolation algorithm
[30] from the original measurement grid, which con-
sisted of 3 × 4 × 6 points in each Cartesian direction.
A nonsymmetrical temperature field is noticeable
inside the furnace and is attributed to the presence
of the heat exchanger inserted 25% through win-
dow A3. The highest temperature region was mea-
sured to be 1300 ◦ C in the region 80 < x < 100 mm
and 200 < z < 300 mm. The temperature differen-
tial is only about 100 ◦ C within about two-thirds of
the furnace (z  180 mm). This is consistent with the
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.6 (1-15)
Fig. 4. Measured temperature distribution (◦ C) for the base-
line case (d = 2.0 mm, Q̇fuel = 15 kW NG, φ = 0.80,
Tair = 450 ◦ C, 25% insertion at position A3) at plane y = 0.
The arrows represent the incoming air (solid blue) and fuel
(solid green) streams and the outgoing exhaust (dotted red)
stream. The dashed line shows the heat exchanger (HX) lo-
cation and insertion length. (For interpretation of the refer-
ences to color in this figure legend, the reader is referred to
the web version of this article.)
temperature measurements presented in Fig. 3. The
uniformity suggests that a nonadiabatic global fur-
nace temperature can reliably represent the reaction
zone temperature. Nonetheless, the most representa-
tive location for such a measurement will depend on
the heat exchanger location. The region in the vicinity
of the heat exchanger is about 100 ◦ C cooler than that
on the opposite side of the furnace. From the knowl-
edge of heat extracted (Q̇hx ) and material properties,
the average heat exchanger surface temperature (T̄hx )
was calculated assuming one-dimensional conduction
to be T̄hx ≈ 70 ◦ C. Hence, local temperatures around
the heat exchanger are expected to be even lower than
1175 ◦ C. This will result in local extinction and par-
tial oxidation near to the cold surface.
3.3. Effect of fuel dilution
A high degree of dilution of the reactants is one of
the necessary conditions for establishing MILD com-
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
Fig. 5. Effect of fuel dilution on temperature and NOx emis-
sions with fuel type and burner size as parameters for the
baseline operating conditions (Q̇fuel = 15 kW, φ = 0.80,
Tair = 450 ◦ C, 25% insertion at position A3). Solid sym-
bols: N2 dilution, open symbols: CO2 dilution, solid lines
with small symbols: d = 2.0 mm, and dashed lines with
large symbols: d = 3.0 mm. Red colored lines and symbols
represent PSR calculations for methane (GRI 3.0). (For in-
terpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)
bustion. An interesting approach to increasing vitia-
tion and delaying the reaction between air and fuel
is the addition of an inert gas to the fuel stream.
Fig. 5 shows the effects of fuel dilution on temper-
ature and NOx emissions for two different fuels and
burner sizes under the baseline operating conditions.
It also shows predictions from a perfectly stirred re-
actor (PSR) model for the NG cases. The fuel stream
was diluted using either N2 or CO2 and the corre-
sponding diluent mass fraction (Ydil ) is displayed in
the abscissa. The fuel and the air mass flow rates were
kept constant while the mass flow rate of diluent was
varied.
It is clear from Fig. 5 that, although dilution can be
expected to cool the reaction zone, this effect is small.
For each case, up to 76% fuel dilution causes a reduc-
tion in furnace temperature of only 2–4%, while the
decrease in NOx emissions can be as large as 48%.
The figure also shows that the furnace operates at
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.7 (1-15)
7
slightly different temperatures in each case, despite
the nominal thermal input being held constant. For
the same diluent mass fraction, the difference between
each operating condition is at most 70 ◦ C, but can be
as low as 5 ◦ C. This small difference does not explain
the variations observed in NOx emissions. Hence, it
is evident that fuel type, diluent, and burner size have
a significant effect on NOx emissions. Each of these
effects is examined in turn below and discussed in de-
tail later.
First considered are the cases in which only the
fuel type was varied, for instance LPG/CO2 (3 mm)
versus NG/CO2 (3 mm). The general trend observed
in the graph is that NOx emissions from LPG combus-
tion are higher than those from NG. Dally et al. [7] re-
ported similar trends for diluted methane and propane
flames.
Next examined are the cases in which the diluent
type is varied. Again in agreement with the results
reported by Dally et al. [7], Fig. 5 shows that fuel
dilution with CO2 (open symbols) produces lower
NOx emission levels than does the equivalent fuel di-
lution with N2 (solid symbols) for the same burner
size. Further, the CO2 dilution causes a decrease in
NOx emissions of up to 48% over the operating range,
whereas the N2 dilution causes only about 2–10%
reduction, even though the N2 content inside the fur-
nace increases up to 19% for the highest diluent mass
fraction cases (Ydil > 68%). The ratio of the specific
heats of CO2 to that of N2 is 1.076 at 1280 ◦ C. Since
the furnace temperature is virtually constant, the dif-
ferences in specific heats and gas radiation character-
istics are deduced to be unable to explain the different
behaviors.
Next assessed are the results from the PSR model
for the combustion of NG. The PSR calculations were
carried out using the steady-state solver from the AU-
RORA code of the Chemkin 4.1 package with the
GRI 3.0 [31] detailed kinetic mechanism and are rep-
resented in Fig. 5 by the red colored lines and sym-
bols. The combustion chamber, which has a volume
of 0.0459 m3 , was modeled as a nonadiabatic reac-
tor at atmospheric pressure. The energy equation was
solved for a fixed heat loss of 5.5 kW, which is con-
sistent with the energy balance shown in a previous
work [29]. Four inlet streams were considered: an air
stream with a mass flow rate of 6.243 g/s at 450 ◦ C,
a fuel stream with a mass flow rate of 0.305 g/s at
20 ◦ C, a diluent stream with mass flow rates varying
according to the NG fuel dilution experiments for the
3.0 mm burner, and a recirculation stream with 10%
of the total mass flow rate of reactants, 0.65 g/s, at
1000 ◦ C. The latter was considered to be a mixture
of inert gases (N2 /H2 O/CO2 = 0.771/0.150/0.079 by
volume) and was included to represent the diluted
conditions of the MCF. The predictions from the
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
8 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
Fig. 6. Relationship between NOx emissions and furnace
temperature, Tfurnace , for all tested operating conditions.
The solid line represents a linear curve fit for all data. The ex-
perimental data points are sorted according to fuel type, dilu-
ent, and burner size. Diamonds: NG, circles: LPG, thin line
symbols: no fuel dilution, solid symbols: N2 dilution, thick
line symbols: CO2 dilution, small symbols: d = 2.0 mm, and
large symbols: d = 3.0 mm.
PSR model show the same general trend observed in
the measurements of decreasing gas temperature and
NOx emission levels with fuel dilution. However, the
maximum predicted NOx emissions are only 6 ppm
for both diluents. A sensitivity analysis (not shown)
showed that the recirculation stream has little effect
on NOx formation at 1280 ◦ C. It also revealed that the
reactor temperature would need to be at least 200 ◦ C
higher to reproduce the NOx emission levels observed
in the experiments, which is when the thermal-NO
mechanism starts to dominate.
Finally, we assess the cases in which the burner di-
ameter is varied. It is noticeable from Fig. 5 that NOx
emissions increase when the burner size is increased.
For a constant diluent mass fraction and thermal input
from the fuel, an increase in nozzle diameter causes
a decrease in the fuel jet velocity by 44% and a de-
crease in the jet momentum flux by the same amount
for the same diluent. A lower fuel jet momentum will
reduce the entrainment of hot furnace gases and hence
change the mixing characteristics close to the burner
exit. In an attempt to confirm this trend, a third fuel
nozzle of diameter 4.6 mm was also tested. However,
stable MILD combustion could not be sustained for
the baseline operating conditions, and a flame was
visible. A discussion of the stability of the MILD
regime is beyond the scope of the present paper.
3.4. Global nitrogen oxides emission correlations
Fig. 6 presents the NOx emissions as a function
of the furnace reference temperature (see also Fig. 1).
The experimental data points are sorted according to
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.8 (1-15)
Fig. 7. Relationship between NOx emissions and exhaust
temperature, Texhaust , for all tested operating conditions.
The experimental data points are grouped according to fuel
input power and fuel type. No distinction is made with re-
spect to fuel dilution and burner size. The thick solid black
line represents a linear curve fit for all data. The other thin
colored lines are linear curve fits for each fuel input power
bin. Open symbols: all cases burning NG, and solid symbols:
all cases burning LPG.
fuel type, diluent, and burner size. The solid line rep-
resents a linear fit to all of the data. Although there
is a general trend of increasing NOx emissions with
furnace temperature, the wide scatter and low corre-
lation coefficient (R 2 = 0.50) show that the depen-
dence is weak. It is worth highlighting that very low
NOx levels, <10 ppm, are achieved for furnace tem-
peratures below 900 ◦ C. However, at temperatures
around 1400 ◦ C, NOx emissions exceeding 60 ppm
were measured in the MILD regime.
Fig. 7 shows NOx emissions data plotted against
the exhaust temperature, which, while lower than the
furnace temperature, avoids the bias in some data due
to the location of the cooling loop relative to the ther-
mocouple. It can be seen that there is a stronger in-
fluence of temperature for all data (R 2 = 0.56) and
for individual fuel input ranges. Even though this
general correlation is somewhat stronger, it is suffi-
ciently poor to demonstrate that other parameters are
also important. This result shows that a representative
furnace temperature alone does not adequately scale
NOx emissions.
Fig. 8 shows NOx emissions plotted against heat
extraction. The heat extraction, Q̇hx , is expressed as a
percentage of the total thermal input. As expected, an
increase in heat extraction, which conversely means a
decrease in temperature, leads to a decrease in NOx
emissions. Because the heat extraction is only part of
the net heat output, it is not surprising that the correla-
tion with heat extraction is even weaker (R 2 = 0.39)
than in Fig. 7.
There is no visible correlation with fuel type, dilu-
ent, or burner size, either in Fig. 6 or in Fig. 8. No
 ARTICLE IN PRESS CNF:6965
JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.9 (1-15)
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–••• 9

Fig. 8. Relationship between NOx emissions and heat ex-

tracted, Q̇hx , for all tested operating conditions. The solid
line represents a linear curve fit for all data. The experimen-
tal data points are sorted according to fuel type, diluent, and
burner size. Diamonds: NG, circles: LPG, thin line symbols:
no fuel dilution, solid symbols: N2 dilution, thick line sym-
bols: CO2 dilution, small symbols: d = 2.0 mm, and large
symbols: d = 3.0 mm.

clear trend is found in Fig. 7 with regard to fuel type

either.

3.5. Froude number scaling

To further investigate the parameters control-

ling NOx yields, the leading-order scaling approach
of Røkke et al. [32] is explored, despite its lim-
itations [12]. Although developed for buoyancy-
controlled hydrocarbon diffusion flames, it has been
applied more broadly in the past [13,16]. Based on
a simplified finite-rate chemistry that specifically in-
cludes thermal and prompt NO formation, on the
concept of laminar flamelet structure [33] and other
several assumptions, the authors derived a correla-
tion to predict the NOx emission index, EINOx . In
their scaling of open flames, the flame volume is the Fig. 9. Leading-order NOx scaling correlation,
leading-order term, and the simple correlation was EINOx (ρfuel u/d), versus fuel jet exit Froude number,
Fre , for (a) all data points, (b) all cases burning NG,
found to be equal to EINOx (ρfuel u/d) ≈ 44 Fr0.6 e , and (c) all cases burning LPG. The experimental data
where Fre is the jet exit Froude number. Weber [16]
points are sorted according to fuel type, diluent, and
proposed an extension of Røkke’s relationship to ac- burner size. Diamonds: NG, circles: LPG, thin line sym-
count for heat extraction under furnace conditions bols: no fuel dilution, solid symbols: N2 dilution, thick
for the Scaling 400 burner series operating with con- line symbols: CO2 dilution, small symbols: d = 2.0 mm,
ventional natural gas flames. He varied the propor- and large symbols: d = 3.0 mm. Blue colored line and
tionality constant according to the levels of heat ex- symbols in part (b) represent data for a reference con-
traction, whilst conforming to the same slope of 0.6. dition (17% < Q̇hx < 30%, 12 kW < Q̇fuel < 16 kW,
For example, the scaling correlation was changed to 0.78 < φ < 0.82, 400 ◦ C < Tair < 500 ◦ C). (For interpre-
EINOx (ρfuel u/d) ≈ 23.2 Fr0.6
e for 40% heat extrac-
tation of the references to color in this figure legend, the
tion. Weber’s approach resulted in a series of parallel reader is referred to the web version of this article.)
lines.
Fig. 9 shows a log–log graph of the variation of LPG. In all figures, the experimental data points are
EINOx (ρfuel u/d) with fuel jet exit Froude number. sorted according to fuel type, diluent, and burner size.
Part (a) presents all experimental data, part (b) those The theoretical correlation of Røkke et al. [32] and
cases burning NG, and part (c) those cases burning the empirical curve fit for 40% heat extraction from

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
10 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
Fig. 10. Relationship between the NOx emission in-
dex, EINOx , and the momentum flux ratio of air-to-fuel,
Gair /Gfuel , for all tested operating conditions. The experi-
mental data points are sorted according to fuel type, diluent,
and burner size. Diamonds: NG, circles: LPG, thin line sym-
bols: no fuel dilution, solid symbols: N2 dilution, thick line
symbols: CO2 dilution, small symbols: d = 2.0 mm, and
large symbols: d = 3.0 mm.
previous work by Weber [16] are also displayed. The
fuel jet exit Froude number (Fre ) is the ratio of jet
momentum flux to buoyant forces and is calculated
from u2 /gd, where g is the gravitational acceleration.
From Fig. 9a, it is evident that some data depart
from the correlation for the present study (NG+LPG)
by almost an order of magnitude. Indeed, several data
sets are essentially independent of the Froude number,
indicating that other parameters affect NOx emissions
and the role of buoyancy is negligible. This is consis-
tent with the Froude number exceeding unity by three
orders of magnitude (Fre > 103 ), which suggests a
momentum-controlled system. Even those data sub-
sets that do scale with Fre , such as the correlation
(NG reference) represented in blue in Fig. 9b, exhibit
a slope slightly different (∝Fr0.68
e ) from those found
earlier [32]. Here, ∝ is used to indicate proportional-
ity. Costa et al. [34] found similar departures for their
momentum-controlled open jet methane flames.
Figs. 9b and 9c show that the slope for NG as the
fuel (∝Fr0.75
e ) is significantly different from that of
LPG as the fuel (∝Fr0.38 e ). The dependence on fuel
type is expected from the model of Røkke et al. [32].
The effects of fuel dilution can also be observed in
Fig. 9c, when the N2 dilution cases (∝Fr0.41 e ) and
the CO2 dilution cases (∝Fr0.30e ) are separately cor-
related.
Fig. 10 shows the variation of the NOx emission
index, EINOx , with the momentum flux ratio of air,
Gair , to fuel, Gfuel . Once again, the experimental
data points are sorted according to fuel type, diluent,
and burner size. Here Gair /Gfuel = nṁair U/ṁfuel u,
where ṁair is the air mass flow rate and ṁfuel is the
fuel mass flow rate. It is evident that no obvious cor-
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.10 (1-15)
relation exists between EINOx and Gair /Gfuel . Other
attempts to correlate EINOx with the total momen-
tum flux (Gair + Gfuel ), momentum flux ratio of fuel
to exhaust (Gfuel /Gexhaust ), and momentum flux ra-
tio of air to exhaust (Gair /Gexhaust ) have also failed
to collapse the data.
3.6. Temperature–time NOx scaling
To further assess the role of the thermal-NO
mechanism, the methodology proposed by Turns
and Myhr [35] and Turns et al. [36] is adopted.
They developed a successful global temperature–time
NOx scaling relationship that accounted for radiation
losses from open turbulent jet flames. The essence of
the temperature–time scaling is the definition of an
NOx production rate, which is calculated by divid-
ing an average NOx concentration by a characteristic
mixing time scale. The experimental NOx concentra-
tion is determined from mass-based emission indices
and stoichiometry as
[NOx ] [mol/m3 ]
fρrct EINOx
=  XNO   XNO  . (4)
MWNO X + MWNO2 X 2
NOx NOx
Adapting the definition presented by Turns and
Myhr [35] for open flames to enclosed multiple-jet
systems, the global residence time (τG ) is charac-
terized by the combustion chamber volume and the
volumetric flow rate of hot reactants as
4VMCF fρrct
τG [s] = , (5)
π d 2 unρfuel
where VMCF is the MCF volume, f is the fuel mixture
fraction, ρrct is the density of the mixture of reactants
at furnace temperature, d is the fuel nozzle diameter,
u is the fuel inlet velocity for a single nozzle, n is
the number of round fuel nozzles, and ρfuel is the fuel
density at ambient temperature. The use of the furnace
volume to represent the reaction zone is a reasonable
approximation for MILD combustion conditions and
is consistent with the global treatment used in other
studies involving furnaces [14]. The fuel mixture frac-
tion is calculated from f = φ/φ + (A/F )stoic and
implicitly incorporates the air mass flow rate. Here,
(A/F )stoic is the value of stoichiometric air-to-fuel
ratio. The denominator in Eq. (5) is directly related to
the fuel mass flow rate, which in turn determines the
fuel input power, also referred to as the firing rate.
A convective time scale, d/u, is commonly used as
the scaling parameter in momentum-controlled open
jet flames. For those flames, d/u is directly propor-
tional to residence time because it controls flame vol-
ume. The convective time scale strongly influences
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
Fig. 11. NOx production rates calculated with the global res-
idence time, τG , as a function of the reciprocal of the furnace
temperature. The experimental data points are grouped ac-
cording to global residence time. +, CH4 [42]; ∗, NG [5];
a, CH4 [41]; ⊕, NG [17]; ×, NG [25]; ⊗, CH4 [38,40]; ∅,
LPG [39]; ", CH4 and C3 H8 [7].
the flame radiant fraction, and thus also the tempera-
ture and NOx emissions [35,37]. In contrast, when the
flame volume is fixed and where exit strain does not
control flame stabilization, d/u and τG are no longer
directly proportional. For this reason, d/u is not a
relevant parameter in enclosed systems, and does not
collapse the NOx data.
Fig. 11 shows an Arrhenius-type plot of NOx
production rates calculated with the global residence
time, τG , as a function of the reciprocal of the fur-
nace temperature. The experimental data points are
grouped according to global residence time. The var-
ious blue symbols refer to NOx production rates esti-
mated from the work of various other researchers [5,
7,17,25,38–42]. Three different lines are also shown
in Fig. 11. The dashed line represents the theoretical
NO production rate via the Zel’dovich thermal-NO
mechanism. The maximum NO formation rates were
calculated for φ = 0.80 considering a quasi-steady-
state approximation for N atoms, assuming equilib-
rium concentrations for O atoms, and applying the
forward reaction coefficient, kf1 , for the rate-limiting
step in the thermal-NO formation process, O + N2 →
NO + N. The value for kf1 was computed from the
expression recommended by Davidson and Hanson
[43], kf1 [m3 s/mol] = 1.95 × 108 exp(−38,660/T ).
The concentrations for the O and N atoms were calcu-
lated from the NASA chemical equilibrium code [44].
The dash-dotted line represents the two-parameter re-
gression fit
 
[NOx ] C
ln = A + B ln(τG ) + (6)
τG T is performed for the complete data set using a number
for hydrocarbon flames for a residence time of 100 ms
extrapolated for a lower temperature range from pre-
vious work by Turns et al. [36]. Finally, the solid line
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.11 (1-15)
11
represents the data from the present study regressed
in the same form as proposed by Turns and Myhr [35]
and Turns et al. [36], for a residence time of 4 s. Here,
[NOx ]/τG is the characteristic NOx production rate
and the regression coefficients A, B, and C are shown
in Table 3.
Several observations arise from the results shown
in Fig. 11. The use of the global residence time to
determine NOx production rates provides a general
collapse of the experimental data. However, there is
still considerable scatter in the data, which is slightly
masked by the logarithmic scales on the axis. This
scatter around a general correlation is also evident
in the data of other researchers in the MILD regime.
For example, the data from Plessing et al. [41] and
Dally et al. [7] lie well within an acceptable error
range. Interestingly, their measurements were car-
ried out in a furnace with dimensions and operating
conditions similar to those for the present facility,
but with a completely different burner design, the
FLOX® burner. On the other hand, all other data ex-
cept for those of Özdemir and Peters [38,40], the
rest of the data were measured in furnaces of differ-
ent sizes that were fitted with unique burner arrange-
ments. The temperature–time scaling demonstrates a
dependence of NOx production rates on furnace tem-
perature considerably weaker than that predicted by
the thermal NOx route alone. This trend applies for
highly diluted furnace environments regardless of the
furnace/burner geometry. Importantly, although the
magnitude of [NOx ]/τG depends on the combustion
chamber volume, the slope is independent of this de-
finition. Sobiesiak et al. [17] also found a departure
from the thermal-NO mechanism based on a well-
stirred reactor model, especially at low characteristic
temperatures.
The slope of the present data is also lower than that
for the open methane, ethylene, and propane flames of
Turns et al. [36]. This shows that the dependence of
NOx production rate on temperature is stronger for
open turbulent jet flames, than for the present MILD
combustion conditions. This finding is further sup-
ported by the much larger value of the temperature
coefficient, C, presented in Table 3 for open flame
conditions. Nevertheless, despite this weaker depen-
dence, it appears that the temperature–time scaling
correlation provides the single best description of
NOx formation in the MILD regime.
To evaluate the quality of the regression fit for the
present study, the percentage difference between the
regression predictions of NOx production rates and
the experiments is presented in Fig. 12. The analysis
of alternative parameters substituted in Eq. (6) (cf. Ta-
ble 3). From top to bottom in Fig. 12, the regression fit
assesses, respectively, furnace temperature and global
 ARTICLE IN PRESS CNF:6965
JID:CNF AID:6965 /FLA [mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.12 (1-15)
12 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••

Table 3
Regression coefficients for Eq. (6)
Database Parameters A B C Temperature range
(ln(mol/cm3 s)) (1/s) (K) (K)
Turns et al. [36] Tflame , τG 1.1146 −0.7410 −16,347 1900–2350
Tfurnace , τG −6.141 −0.387 −3,634 1050–1650
Present study τG −8.106 −0.733 0 N/A
Tehxaust −2.249 0 −6,681 760–1070

levels of up to 400%, an almost identical fit is found

when only residence time is used, 77% of the data
are within an error band of ±50% (or 87% within
an error band ±100%). A slightly better fit is found
when the regression is based on the exhaust tempera-
ture alone, 85% of the data are within an error band
of ±50% (or 97% within an error band of ±100%).
The higher quality of the (Texhaust ) regression than of
the (Tfurnace , τG ) regression is consistent with the ex-
haust temperature being less susceptible to bias due to
the proximity to a heat sink, as previously noted. The
largest errors were found in the predictions of CO2
dilution cases. However, excluding those points did
not improve the regression significantly. Likewise, a
regression using exhaust temperature and global resi-
dence time (Texhaust , τG ) was not significantly differ-
ent.

4. Discussion

The Zel’dovich thermal mechanism is usually the

Fig. 12. Percentage difference between the regression pre-
dictions (cf. Table 3) of NOx production rates and the exper-
iments for all 191 data points.
residence time (Tfurnace , τG ), global residence time
(τG ) only, and exhaust temperature (Texhaust ) only as
parameters. For the (Tfurnace , τG ) case, 77% of the
data are within an error band of ±50% (or 92% within
an error band of ±100%). Although the errors reach
major contributor to NOx formation in most high
temperature (T > 1500 ◦ C) combustion systems. As
already seen in Figs. 4 and 6, the temperatures in
the present furnace are considerably lower than this
(T < 1360 ◦ C) and its distribution is quite uniform.
Because of the low fluctuations of MILD combustion
[45], it is unlikely that instantaneous temperatures
will significantly exceed the measured mean values.
Hence, consistent with the well-established under-
standing of the operation of MILD combustion sys-
tems, NOx formation via the thermal-NO mechanism
is inhibited relative to conventional combustion [42].
Thermal NOx not only is suppressed, but also does
not control NOx emissions, as was previously noted
in Fig. 11. The results presented in Figs. 7 and 8 re-
inforce that the thermal-NO route is not dominant.
Therefore, Fenimore’s prompt-NO [46] and/or N2 O-
intermediate [47] mechanisms are also expected to be
important.
The fuel dilution results from Fig. 5 suggest that
NOx formation may be dependent on the ratio of
carbon to hydrogen atoms (C/H) in the fuel mole-
cule. Since the temperatures are so similar for the
dilution cases, the difference cannot be attributed to

Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
the thermal-NO route. Rather, the fuel type depen-
dence is further evidence that the prompt-NO mech-
anism is significant. The dependence on fuel type
has also been observed in Fig. 9. The relevance of
the prompt-NO pathway in the MILD regime has
been speculated on by a few researchers [5,17,25,
48]. In the predictions of Mancini et al. [48], only
5% of global NOx emissions were calculated to be
of prompt-NO origin. On the other hand, it has been
observed that the N2 O-intermediate mechanism can
represent up to 65% of the total NOx formation in
lean-premixed methane combustion in a jet-stirred re-
actor at atmospheric pressure in a temperature range
from 1140 to 1570 ◦ C [49]. Further assessments are
required to quantify the contributions of each path-
way to the overall NOx emissions.
Several parameters seem to alter NOx formation
in MILD combustion conditions and hence add to the
complexity of the scaling problem. The previous sec-
tions have outlined the influence of heat extraction, air
preheat, oxygen concentration, dilution, and fuel type
on global NOx emissions. While all these parameters
are not independent from each other, the combined
effect can be ultimately associated with a character-
istic nonadiabatic temperature and residence time, as
shown in Fig. 11. However, the fairly poor fit found
in the regression analysis presented in Fig. 12 demon-
strates that temperature and residence time influence,
but do not control, the scaling of NOx . It is not pos-
sible to isolate one or two controlling parameters, be-
cause many parameters seem to have an influence of
comparable magnitude.
Although the Froude number has been used as a
scaling parameter in furnace conditions with conven-
tional flames [16], Fig. 9 suggests that it is not the
most relevant parameter to scale NOx emissions from
furnaces operating in the MILD regime. This is be-
cause buoyancy forces are particularly weak in an
environment with low temperature gradients. Never-
theless, some dependence on Fre is evident, which
implies that fuel jet momentum is still an important
parameter. The data for the two different burner sizes
shown in Fig. 5 give further evidence of the influ-
ence of momentum. A lower fuel jet momentum will
lead to a change in the jet trajectory, much like that in
the strong-jet/weak-jet model of Grandmaison et al.
[50], and hence reduce the amount of dilution with
hot combustion products prior to mixing with the ox-
idant stream. These changes will affect local species
concentrations and temperatures in the reaction zone,
altering the ratio between mixing and reaction times,
i.e., the local Damköhler number (Da = τmix /τchem ).
A low local Damköhler number can favor nonequilib-
rium effects, such as superequilibrium O atoms [51].
Both entrainment levels of hot combustion products
and nonequilibrium chemistry may account for the
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.13 (1-15)
13
differences between CO2 and N2 dilution. However,
the results from Fig. 10 demonstrate that, although
mixing has some influence, it does not control global
NOx yields. Therefore other effects, such as those due
to chemical kinetics, also influence the NOx forma-
tion in this furnace.
The PSR calculations from Fig. 5 indicate that the
assumption of a well-stirred reactor (Da 1) used
by some researchers [52,53] does not appear to be
reasonable in the present MCF system. This is con-
sistent with the recent modeling study of Mancini et
al. [54]. They also found that the combustion chamber
cannot be considered simply as a well-stirred reac-
tor. They proposed a reactor network model, which
consisted of 22 PSRs, to predict NOx emissions from
a semi-industrial-scale furnace. Their results empha-
sized the importance of the mixing characteristics
close to the burner exit. This implies that the NOx for-
mation process is not controlled by chemical kinetics.
In summary, the analysis described above high-
lights the fact that neither momentum nor kinetics
has a controlling influence on NOx emissions, but
that rather both are important. This suggests that NOx
formation in the present furnace may be within the
regime where mixing time scales are comparable to
chemical time scales, and hence Damköhler numbers
are of order unity, Da = O(1). Therefore, in order to
accurately predict NOx emissions from MILD com-
bustion systems, any model will need to include the
effects of both mixing and chemistry.
5. Summary and conclusions
Global measurements of temperature and exhaust
gas emissions obtained for 191 conditions in a 20-kW
MILD combustion furnace operating with a parallel
jet burner system have been presented. The main find-
ings of the analysis are as follows:
1. The presence of a water-cooled heat exchanger
dominated the temperature gradients and caused
a nonsymmetrical temperature differential of
about 100 ◦ C across the furnace.
2. Overall NOx emission values as low as 5 ppmv
dry at 3% O2 and as high as 66 ppmv dry at 3%
O2 were recorded in the MILD regime for fur-
nace temperatures of up to 1370 ◦ C.
3. Fuel dilution of up to 76% by mass, with either
CO2 or N2 , reduced NOx emissions by up to
48% and 10%, respectively. These levels of di-
lution did not affect furnace temperatures signifi-
cantly, with a maximum reduction of only 56 ◦ C.
The fuel type, diluent type, and fuel jet mo-
mentum were all found to influence global NOx
yields. However, neither mixing nor chemistry
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
14 G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
effects were found to be dominant, suggesting
that the characteristic Damköhler number spans,
or is on the order of, unity. The failure of PSR
calculations to predict the NOx emissions further
supports this hypothesis, as does the finding that
the Froude number scaling revealed a limited de-
pendence of NOx emissions on jet momentum.
4. A global NOx production rate, calculated based
on a global residence time and a characteris-
tic furnace temperature, was found to give the
best scaling of the data across all conditions as-
sessed. However, the dependence on temperature
was found to be much weaker than that expected
based on the thermal-NO mechanism alone. This
trend also applied to other MILD combustion
systems, irrespective of the furnace/burner geom-
etry. Furthermore, the regression analysis found
that neither temperature nor global residence
time provides a sufficiently strong correlation to
be considered dominant. The leading-order ap-
proach showed that, although the jet exit Froude
number has been used in the past for enclosed
conventional flames, it is not the appropriate pa-
rameter for NOx scaling in the MILD regime.
This is consistent with the weak role of buoyancy
in a furnace with such low temperature gradients.
5. It is clear that no single NOx production mech-
anism is dominant in the present MILD com-
bustion conditions. The nondominant role of the
Zel’dovich thermal-NO mechanism is consistent
with the moderate furnace temperatures (800–
1400 ◦ C). Hence, it is deduced that the prompt-
NO and/or N2 O-intermediate pathways are of
significance comparable to that of the thermal-
NO pathway.
Acknowledgments
The authors thank Graham Kelly and Bob Dyer for
their technical support in all stages of this project. The
financial support of the Australian Research Council
(ARC) and the School of Mechanical Engineering of
the University of Adelaide is also gratefully acknowl-
edged.
References
[1] IEA/OECD, CO2 Emissions from Fuel Combustion:
1971–2000, Organisation for Economic Cooperation
and Development and International Energy Agency,
Paris, 2002.
[2] J.M. Beer, Prog. Energy Combust. Sci. 33 (2007) 107–
134.
[3] S. Pacala, R. Socolow, Science 305 (2004) 968–972.
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.14 (1-15)
[4] A. Cavaliere, M. De Joannon, Prog. Energy Combust.
Sci. 30 (2004) 329–366.
[5] M. Katsuki, T. Hasegawa, Proc. Combust. Inst. 27
(1998) 3135–3146.
[6] H. Tsuji, M. Morita, A.K. Gupta, M. Katsuki, K. Kishi-
moto, T. Hasegawa, High Temperature Air Combus-
tion: From Energy Conservation to Pollution Reduc-
tion, CRC Press, 2003.
[7] B.B. Dally, E. Riesmeier, N. Peters, Combust. Flame
137 (2004) 418–431.
[8] M. Flamme, Appl. Therm. Eng. 24 (2004) 1551–1559.
[9] P.R. Medwell, P.A.M. Kalt, B.B. Dally, Combust.
Flame 148 (2007) 48–61.
[10] H. Zhang, G. Yue, J. Lu, Z. Jia, J. Mao, T. Fujimori, T.
Suko, T. Kiga, Proc. Combust. Inst. 31 (2007) 2779–
2785.
[11] R. Weber, J.P. Smart, W. Van der Kamp, Proc. Com-
bust. Inst. 30 (2004) 2623–2629.
[12] S.R. Turns, Prog. Energy Combust. Sci. 21 (1995) 361–
385.
[13] A.D. Al-Fawaz, L.M. Dearden, J.T. Hedley, M. Mis-
saghi, M. Pourkashanian, A. Williams, L.T. Yap, Proc.
Combust. Inst. 25 (1994) 1027–1034.
[14] T.-C.A. Hsieh, W.J.A. Dahm, J.F. Driscoll, Combust.
Flame 114 (1998) 54–80.
[15] A. Sayre, N. Lallemant, J. Dugue, R. Weber, Proc.
Combust. Inst. 25 (1994) 235–242.
[16] R. Weber, Proc. Combust. Inst. 26 (1996) 3343–3354.
[17] A. Sobiesiak, S. Rahbar, H.A. Becker, Combust.
Flame 115 (1998) 93–125.
[18] J. Newby, B. Cain, T. Robertson, in: Proceedings of
the 2nd International Seminar on High Temperature
Combustion in Industrial Furnaces, Stockholm, Swe-
den, 2000.
[19] M. Nishimura, T. Suzuki, R. Nakanishi, R. Kitamura,
Energy Convers. Manage. 38 (1997) 1353–1363.
[20] F.K. Besik, S. Rahbar, H.A. Becker, A. Sobiesiak, U.S.
Patent 5,772,421 (June 30, 1998).
[21] B.E. Cain, T.F. Robertson, J.N. Newby, U.S. Patent
6,638,061 (October 28, 2003).
[22] T. Suzuki, K. Morimoto, U.S. Patent 4,439,137 (March
27, 1984).
[23] G.G. Szegö, B.B. Dally, G.J. Nathan, F.C. Christo,
in: Australian Symposium on Combustion and the 8th
Australian Flame Days, Melbourne, Australia, 8–9 De-
cember 2003.
[24] I. Nakamachi, K. Yasuzawa, T. Miyahara, T. Nagata,
U.S. Patent 4,945,841 (August 7, 1990).
[25] R. Weber, A.L. Verlaan, S. Orsino, N. Lallemant, J.
Inst. Energy 72 (1999) 77–83.
[26] R. Siegel, J.R. Howell, Thermal Radiation Heat Trans-
fer, fourth ed., Taylor and Francis, New York, 2002.
[27] F.P. Incropera, D.P. DeWitt, Fundamentals of Heat and
Mass Transfer, fourth ed., Wiley, New York, 1996.
[28] S.R. Turns, An Introduction to Combustion: Con-
cepts and Applications, second ed., WCB/McGraw–
Hill, Boston, 2000.
[29] G.G. Szegö, B.B. Dally, G.J. Nathan, F.C. Christo,
in: The Sixth Asia–Pacific Conference on Combustion,
Nagoya, Japan, 20–23 May 2007, pp. 231–234.
[30] J.C. Davis, Statistics and Data Analysis in Geology,
Wiley, New York, 1986.
 ARTICLE IN PRESS
JID:CNF AID:6965 /FLA
G.G. Szegö et al. / Combustion and Flame ••• (••••) •••–•••
[31] Gas Research Institute, GRI-mech 3.0, available at
http://www.me.berkeley.edu/gri-mech/.
[32] N.A. Røkke, J.E. Hustad, O.K. Sonju, F.A. Williams,
Proc. Combust. Inst. 24 (1992) 385–393.
[33] N. Peters, Prog. Energy Combust. Sci. 10 (1984) 319–
339.
[34] M. Costa, C. Parente, A. Santos, Exp. Therm. Fluid
Sci. 28 (2004) 729–734.
[35] S.R. Turns, F.H. Myhr, Combust. Flame 87 (1991) 319–
335.
[36] S.R. Turns, F.H. Myhr, R.V. Bandaru, E.R. Maund,
Combust. Flame 93 (1993) 255–269.
[37] G.J.R. Newbold, G.J. Nathan, D.S. Nobes, S.R. Turns,
Proc. Combust. Inst. 28 (2000) 481–487.
[38] P.J. Coelho, N. Peters, Combust. Flame 124 (2001)
503–518.
[39] S. Kumar, P.J. Paul, H.S. Mukunda, Proc. Combust.
Inst. 29 (2002) 1131–1137.
[40] I.B. Özdemir, N. Peters, Exp. Fluids 30 (2001) 683–
695.
[41] T. Plessing, N. Peters, J.G. Wuenning, Proc. Combust.
Inst. 27 (1998) 3197–3204.
[42] J.A. Wünning, J.P. Wünning, Prog. Energy Combust.
Sci. 23 (1997) 81–94.
Please cite this article in press as: G.G. Szegö et al., Scaling of NOx emissions from a laboratory-scale mild combustion
furnace, Combust. Flame (2008), doi:10.1016/j.combustflame.2008.02.001
CNF:6965
[mNY4a; v 1.89; Prn:5/03/2008; 14:39] P.15 (1-15)
15
[43] A.M. Dean, J.W. Bozzelli, in: W.C. Gardiner Jr.
(Ed.), Gas-Phase Combustion Chemistry, second ed.,
Springer-Verlag, 2000, p. 143.
[44] S. Gordon, B.J. McBride, Report No. NASA RP-1311,
NASA Lewis Research Center, Cleveland, OH, 1994.
[45] M. Oberlack, R. Arlitt, N. Peters, Combust. Theory
Model. 4 (2000) 495–509.
[46] C.P. Fenimore, Proc. Combust. Inst. 13 (1970) 373–
380.
[47] C.T. Bowman, Proc. Combust. Inst. 24 (1992) 859–
878.
[48] M. Mancini, R. Weber, U. Bollettini, Proc. Combust.
Inst. 29 (2002) 1155–1162.
[49] R.C. Steele, P.C. Malte, D.G. Nicol, J.C. Kramlich,
Combust. Flame 100 (1995) 440–449.
[50] E.W. Grandmaison, I. Yimer, H.A. Becker, A. So-
biesiak, Combust. Flame 114 (1998) 381–396.
[51] J.F. Driscoll, R.-H. Chen, Y. Yoon, Combust. Flame 88
(1992) 37–49.
[52] M. De Joannon, A. Cavaliere, T. Faravelli, E. Ranzi,
P. Sabia, A. Tregrossi, Proc. Combust. Inst. 30 (2005)
2605–2612.
[53] M. De Joannon, A. Saponaro, A. Cavaliere, Proc. Com-
bust. Inst. 28 (2000) 1639–1645.
[54] M. Mancini, P. Schwoppe, R. Weber, S. Orsino, Com-
bust. Flame 150 (2007) 54–59.