You are on page 1of 12

Chemical Engineering Science 58 (2003) 4871 – 4882

www.elsevier.com/locate/ces

Combustion characteristics and #ame stability at the microscale:


a CFD study of premixed methane/air mixtures
D. G. Norton, D. G. Vlachos∗
Department of Chemical Engineering and Center for Catalytic Science and Technology (CCST), University of Delaware,
Newark, DE 19716-3110, USA
Received 28 June 2002; received in revised form 19 December 2002; accepted 20 December 2002

Abstract

A two-dimensional elliptic, computational #uid dynamics (CFD) model of a microburner is solved to study the e5ects of microburner
dimensions, conductivity and thickness of wall materials, external heat losses, and operating conditions on combustion characteristics and
#ame stability. We have found that the wall conductivity and thickness are very important as they determine the upstream heat transfer,
which is necessary for #ame ignition and stability, and the material’s integrity by controlling the existence of hot spots. Two modes of
#ame extinction occur: a spatially global type for large wall thermal conductivities and/or low #ow velocities and blowout. It is shown that
there exists a narrow range of #ow velocities that permit sustained combustion within a microburner. Large transverse and axial gradients
are observed even at these small scales under certain conditions. Periodic oscillations are observed near extinction in cases of high heat
loss. Engineering maps that delineate #ame stability, extinction, and blowout are constructed. Design recommendations are =nally made.
? 2003 Elsevier Ltd. All rights reserved.

Keywords: Methane; Microburners; Fluid mechanics; Heat conduction; Reaction engineering; Simulation

1. Introduction geometry, composition, and #ow rate, methane/air #ames are


typically quenched when con=ned within spaces with criti-
Microburners are emerging as a powerful tool for cal dimensions of ¡ 1–2 mm (Davy, 1817; Maekawa, 1975;
portable production of energy. Due to the large energy Ono & Wakuri, 1977; Fukuda, Koji, & Sakamoto, 1981;
density of hydrocarbons (40 vs. 0:5 MJ=kg), microcombus- Lewis & von Elbe, 1987; Linan & Williams, 1993). Flames
tion may eventually replace expensive and environmentally are quenched in these small dimensions because of two
non-benign Li batteries in laptops, cellular phones, and other primary mechanisms, namely thermal and radical quench-
communication devices (Sitzki, Borer, Schuster, Ronney, & ing (Vlachos, Schmidt, & Aris, 1994; Raimondeau, Norton,
Wussow, 2001). Furthermore, due to the inherently higher Vlachos, & Masel, 2002). Thermal quenching occurs when
heat-transfer coeCcients of microscale systems, lower suCcient heat is removed through the walls, that combus-
combustion temperatures may be envisioned, which can tion cannot be self-sustained. Radical quenching occurs via
considerably minimize thermal NOx formation (Aghalayam adsorption of radicals on the system walls and subsequent
& Vlachos, 1998). For the same reason, microburners can recombination, which results in lack of homogeneous chem-
be eCcient heat sources for endothermic reactions, such as istry. The small scales of these systems make them signi=-
steam reforming and ammonia decomposition, in integrated cantly more prone to both quenching mechanisms because
microchemical systems for the production of hydrogen for of the high surface area to volume ratios, i.e., enhanced heat
fuel cell applications (Ganley, Seebauer, & Masel, 2003). transfer from the #ame to the walls and increased radical
Experiments performed in 1817 by Davy, and followed mass transfer. In addition to #ame quenching, blowout can
by many groups, suggest that it is impossible to propa- occur when the burner exit velocity exceeds the #ame burn-
gate #ames in gaps of sub-millimeter scale. Depending on ing velocity (Linan & Williams, 1993). In this mechanism,
the reaction shifts downstream until it exits the microburner.
∗ Corresponding author. Tel.: +1-302-831-2830; The interplay between blowout and quenching has been ob-
fax: +1-302-831-1048. served in elliptic models with detailed catalytic chemistry
E-mail address: vlachos@udel.edu (D. G. Vlachos). (Maruta et al., 2002). Recent experiments have achieved

0009-2509/$ - see front matter ? 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2002.12.005
4872 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

a stable #ame in microburners through appropriate surface tinuity, momentum, energy and species conservation equa-
modi=cation of ceramic materials to limit radical loss and tions in the #uid and the 2D energy equation in the wall,
insulation to limit heat losses, for methane/air mixtures com- listed in Table 1. Fluent 6.0 was used to perform these sim-
busted between parallel plates with gaps ¡ 300 m (Masel ulations (Fluent, 2002). Steady-steady simulations are per-
& Shannon, 2001; Jensen, Masel, Moore, & Shannon, 2003). formed, unless otherwise stated. Preliminary simulations we
While #ame propagation at the microscale is feasible, the have performed with surface and gas radiation show that the
interplay of kinetics and transport in #ame stability and com- wall temperatures are reduced, and as a result the ignition
bustion characteristics of these systems is poorly understood. distance is slightly increased. However, these changes are
The inability of conducting spatially resolved measurements not dramatic and radiation is left out from the simulations
(Eckbreth, 1988; Linan & Williams, 1993), inherent to the below in order to isolate the e5ect of conduction of walls
microscale, underscores the need for detailed mathematical on #ame stability.
modeling. We have recently performed 2D simulations uti- The #uid density is calculated using the ideal gas law.
lizing the boundary layer approximation in cylindrical mi- The #uid viscosity, speci=c heat, and thermal conductivity
crochannels using detailed gas-phase chemistry and wall are calculated from a mass fraction weighted average of
radical quenching chemistry to create engineering maps of species properties, as indicated in Table 1, and the species
wall-sticking coeCcients and external heat loss coeCcients speci=c heat is calculated using a piecewise polynomial =t
that permit #ame propagation (Raimondeau et al., 2002). of temperature (Fluent, 2002).
The boundary layer approximation ignores axial species and The primary focus of this research is on understanding
energy di5usion. In those simulations, ignition was caused the e5ect of wall heat transfer within the system. As shown
through suCcient preheating of premixed methane/air mix- below, heat transfer along the wall strongly a5ects the #ame
tures and the issue of self-sustained combustion could not be stability and makes this distributive reactor very di5erent
addressed. Furthermore, comparison of elliptic and parabolic from the classic CSTR and PFR. Since recent experimen-
simulations even for fast #ows demonstrated that the slow tal work has overcome the radical quenching mechanism,
gaseous heat conduction is important for quantitative pre- through a surface treatment that produced a “quenchless”
diction of ignition distance and #ame stability. material (Masel & Shannon, 2001; Jensen et al., 2003), and
Aside from the stability of stationary solutions, an ex- simulations with complex gaseous chemistry have been con-
perimental and theoretical phenomenon that can occur in ducted for the e5ect of radical quenching (Raimondeau et
non-adiabatic combustion is oscillatory behavior (Uppal, al., 2002), our focus here is on heat transfer as a means of un-
Ray, & Poore, 1974; Uppal, Ray, & Poore, 1976; Gray & derstanding energy management at small scales. Therefore,
Scott, 1990). In these cases, stable periodic swings in pro- a reduced methane combustion mechanism was used for
cess variables are observed for high heat loss. These os- these simulations, as thermal e5ects are relatively well cap-
cillations may lead to undesired properties such as thermal tured by a few steps of chemistry (Williams, 1991; Mauss &
shock, vibrations, or reduced conversion. Due to the large Peters, 1993). Speci=cally, the mechanism used is a one-step
surface area to volume ratio, microburners may su5er from irreversible reaction determined by Westbrook and Dryer
oscillations and, thus, the potential for oscillatory instabili- (1981),
ties needs to be exploited for robust design.
In this work, the reaction and transport of methane/air rCH4 (kgmol=m3 =s)
mixtures in microburners is studied through 2D fully ellip-
= 2:119 × 1011 exp[(−2:027 × 108 J=kgmol)=RT ]
tic simulations by treating explicitly heat transfer through
the wall. The e5ects of microburner dimensions, materials ×[CH4 ]0:2 [O2 ]1:3 ; (1)
of construction, external heat loss, and operating conditions
on #ame stability and combustion characteristics are delin- where the concentrations are in units of kgmol=m3 . This
eated. Oscillatory instabilities are also explored. Finally, us- one-step approximation is a useful tool for describing #ame
ing the results of these studies, regions of #ame stability for dynamics as well as #ame responses to external perturba-
di5erent design parameters are designated, and design rec- tions (Williams, 1991).
ommendations are made. The boundary conditions are as follows. At the inlet, a
=xed, #at velocity pro=le is used. This boundary condi-
tion =xes the convective component of the #ux of species
2. Model and energy, but the di5usive component depends on the
gradient of the computed temperature or species =elds. Sym-
Premixed, stoichiometric methane/air mixtures are fed to metry boundary conditions are applied at the centerline be-
a microburner. The microburner consists of two parallel, in- tween the two plates. At the exit, a =xed pressure is speci-
=nitely wide plates of distance L apart and length of 1 cm. =ed and far-=eld conditions are imposed for the rest of the
The walls have =nite thickness Lw . Due to the aspect ratio, variables. At the interface between the wall and the #uid,
the burner is modeled as a 2D system, as shown in Fig. 1a. no-slip boundary condition and no-species #ux normal to
A =nite-di5erence method is used to discretize the 2D con- the wall surface are employed. The heat #ux at the wall/#uid
D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882 4873

Norton and Vlachos


Heat Flux = h(Tw,ext –Text)
No Heat No Heat
Wall thickness, Lw
Flux Flux

Velocity, Mass Pressure


½ Gap thickness, L/2
Fractions and Outlet
Temperature
Symmetry Boundary Condition
(a)

(b)
2500
Centerline Temperature [K]

2000

1500
600 Nodes
2400 Nodes
1000 5400 Nodes
9600 Nodes

500

0
0 2 4 6 8 10
(c) Axial Displacement [mm]
Fig. 1. (a) Schematic of computational domain along with a description of some of the boundary conditions used in the model. (b) Mesh used in these
simulations. More nodes are placed around the reaction zone. (c) Example of centerline temperature pro=les for meshes with di5erent nodal densities.
The parameters are V0 = 0:5 m=s, kw = 7:5 W=m=K, h = 0 W=m2 =K, L = 600 m, and Lw = 200 m.

interface is computed using Fourier’s law and continuity in Comparison to published modeling work is worthy. Previ-
temperature and heat #ux links the #uid and solid phases. ous CFD modeling was focused mainly on catalytic combus-
The left and right edges of the wall are assumed to be insu- tion of wide channels. Generally, most work has been based
lated. Newton’s law of cooling is used at the outer edge of on the parabolic (boundary layer) approximation, one-step
the wall catalytic chemistry, without considering heat transfer within
the walls. Noteworthy, Groppi et al. have performed studies
q = h(Tw; ext − Ta; ext ); (2)
for the catalytic combustion of methane in channels using a
where h is the exterior convective heat transfer coeCcient, 2D elliptic treatment of the walls but a lumped approxima-
Tw; ext is the temperature at the exterior wall surface, and tion for the #uid equations, based on a semi-theoretical cor-
Ta; ext is the ambient temperature (taken to be 300 K). Note relation for the Nusselt and Sherwood numbers (Groppi &
that all internal heat transfer between the #uid and the wall Tronconi, 2000). They also performed studies with 2D and
is calculated by accounting explicitly for the convective and 3D boundary layer approximation for the #uid elliptic treat-
conductive heat transport in the 2D elliptic model within ment of heat transfer in the wall (Groppi, Tronconi, Berg,
the #uid and within the wall. The wall thermal conductivity & Forzatti, 2000). Hayes et al. have also performed simula-
and exterior convective heat loss coeCcient are taken as in- tions of transient catalytic combustion in channels. The gas
dependent parameters to understand how important thermal phase was assumed to be in quasi-steady-state. The walls
management is. were transiently modeled using two-dimensional energy
4874 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

Table 1
R Other symbols have the conventional meaning
Modeling equations. Here  is in kg/m/s, T is in K, and  is in A.

Conservation equations
  
@ @ @ @Vx @Vy
Continuity: = − Vx + Vy + +
@t @x @y @x @y
 
@ @(Vx Vx ) @(Vx Vy ) @p @xx @yx
Momentum: (Vx ) = − + − + +
@t @x @y @x @x @y
 
@ @(Vx Vy ) @(Vy Vy ) @p @xy @yy
(Vy ) = − + − + +
@t @x @y @y @x @y
   
@Yi @Yi
 
@ Di; m @ Di; m
@ @(Yi Vx ) @(Yi Vy ) @x @y
Species: (Yi ) = − + + + + Ri
@t @x @y @x @y
        
@T @T @Yi @Yi
 @ kf  @ kf @ h i D i; m @ hi Di; m
  @y 
Fluid energy:
@
(h) = −
@(hVx )
+
@(hVy )
+
@x
+
@y
+  @x
+  −  hi R i
@t @x @y @x @y  @x @y 
i i

 
@ @2 T @2 T
Wall energy: (h) = kw +
@t @x2 @y2

Properties
Mass-averaged viscosity, speci=c heat, and thermal conductivity:
  
 = Yi i ; Cp; f = Yi Cp; i ; k f = Yi ki
i i i
  1=2
1 1
T3 +
Mw; i Mw; j
Binary di5usivities: Dij = 0:0188
pij2 D
 
15 R 4 Cp; i Mw; i 1
Thermal conductivities: ki = i +
4 Mw; i 15 R 3

MW T
Viscosities: i = 2:67 × 10−6
2 

di5usion and radiation. The gas phase was modeled using presented in this work were achieved using a mesh consist-
2D forms of the momentum, energy, and species equations. ing of 5400 nodes.
The chemistry was modeled as a one-step heterogeneous To solve the conservation equations, a segregated solu-
reaction (Hayes, Kolaczkowski, Thomas, Titiloye, 1996). tion solver with an under-relaxation method is used. The
All these studies have focused on conventional reactor segregated solver =rst solves the momentum equations, then
dimensions. solves the continuity equation, and updates the pressure and
In our simulations, a non-uniform mesh is used with more mass #ow rate. The energy and species equations are sub-
nodes accumulated around the reaction zone, as shown in sequently solved and convergence is checked. The latter is
Fig. 1b. Computations were performed using meshes with monitored through both the values of the residuals of the
varying nodal densities to determine the optimal node spac- conservation equations and the di5erence between subse-
ing and density that would give the desired accuracy and quent iterations of the solution.
minimize computation time. Fig. 1c shows the centerline Simulations were performed on a Beowulf cluster consist-
temperature pro=les for some of the meshes used. As the ing of 17 nodes. Each node contains a 1:9 GHz PentiumJ
mesh density increases, there is a convergence of the solu- IV processor with 1 GB of RAM. Numerical convergence
tion. The coarsest mesh, consisting of 600 nodes in total, is in general diCcult because of the inherent sti5ness of
fails to accurately capture the in#ection point at the igni- the chemistry as well as the disparity between the wall
tion point and the maximum temperature. Solutions obtained and the #uid heat conductivities. In order to assist con-
with meshes consisting of a few thousand nodes are reason- vergence and compute extinction points, natural parameter
ably accurate. Larger mesh densities, up to 21,600 nodes continuation was implemented. The calculation time var-
(not shown), yielded no obvious advantage. All solutions ied for these problems, depending on the initial guess
D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882 4875

as well as the parameter set, from ∼30 min to several the #uid temperature, so energy transfers from the wall to the
hours. #uid. The wall thermal conductivity is signi=cantly larger
than that of the #uid mixture, so most of the upstream con-
ductive heat #ux occurs within the walls of the reactor. This
3. Combustion characteristics at the microscale energy is brought upstream from the post-combustion region
where walls are considerably hot. Since the mixture warms
For most cases studied, microburners exhibit similar com- up from the wall towards the centerline, ignition occurs near
bustion characteristics that are summarized in this section. the wall and the #ame stabilizes at the centerline, consis-
Fig. 2 shows temperature, reaction rate, and conversion con- tent with ignition and extinction studies conducted in 1D
tours for a typical case and Fig. 3 depicts the reaction rate at stagnation #ows (Vlachos, Schmidt, & Aris, 1993, 1994).
the centerline and the temperature along the centerline and This ignition mode is di5erent from the case where outside
the wall interior for the same conditions. preheating is used to ignite the mixture. In the latter case,
Three regions can be seen in these microburners, namely ignition occurs at the centerline (Daou, 2002; Raimondeau
preheating, combustion, and post-combustion or cooling, as et al., 2002).
sketched in Fig. 3. The width of these regions changes as a Once the #uid reaches the ignition temperature of approx-
function of operating conditions, and their distinction is not imately 1000◦ C, there is an in#ection point in the temper-
always as sharp, as discussed further below. In the preheat- ature pro=le. The mixture combusts rapidly, releasing heat,
ing region I, the wall temperature is signi=cantly higher than which causes a sharp rise in the #uid temperature in the
combustion region II. This “take-o5 ” point is consistent in
2320
temperature with the experimental =ndings of Dagaut, Boet-
tner, and Cathonnet (1991). Because the walls preheat the
incoming #uid, in zone I, it is important that the wall tem-
300 (a)
perature in zone I reaches ∼1000◦ C to allow #uid ignition
in zone II. The combustion zone is relatively narrow (lo-
11.8
calized), a characteristic of highly activated reactions. Even
at these relatively small scales, the transverse heat transfer
0.0 (b)
within the #uid is much slower than the rate of heat re-
0 lease so that the #uid centerline temperature in this zone
approaches approximately the adiabatic #ame temperature.
In the post-combustion (cooling) region III, after the reac-
1 (c) tants have been consumed, the reaction stops, the #uid cools
down to the wall temperature, and the walls are cooled by
Fig. 2. Contours of (a) temperature (K), (b) reaction rate (kgmol=m3 =s), exterior heat losses. There are no signi=cant axial or trans-
and (c) methane conversion. These =gures are not to scale. The parameters verse gradients within this zone. In non-adiabatic cases, both
are the same as in Fig. 1c.
the #uid and the wall would eventually reach room temper-
ature in suCciently long burners.
In some cases (e.g., Figs. 1c and 3), the maximum #uid
2500 12 temperature exceeds the adiabatic #ame temperature of
I II III
methane/air mixture computed for an inlet temperature of
Reaction Rate [kgmoles/m3/s]

Fluid
2000
10 300 K. Traditional CSTR or PFR analysis suggests that
this is impossible to achieve with one-step chemistry. How-
Temperature [K]

8 ever, in this distributed model, the wall acts as a conduit


1500
for heat transfer from the hot exiting products to the cold
Wall
6 entering reactants. This link results in a heat recycle within
1000 the system, which increases the temperatures near the inlet
Fluid 4 and allows a maximum temperature that is greater than
the adiabatic #ame temperature. Because of the overall en-
500
2 ergy conservation, the exiting #uid temperature is in these
cases lower than the adiabatic #ame temperature, as shown
0 0 for example in Fig. 1c for adiabatic walls. In the section
0 2 4 6 8 10 of velocity e5ects, we show an example of higher inlet
Axial Displacement [mm] temperatures caused via heat transfer from the wall.
Despite the small scales, there are still signi=cant trans-
Fig. 3. Temperature and reaction pro=les along the wall and #uid cen-
terline. Three regions can be distinguished: I. preheating, II. combustion,
verse gradients in the #uid temperature and reaction rate
and III. post-combustion (Cooling). The parameters are the same as in until nearly the end of the reaction zone, as shown in Fig.
Fig. 1c. 2 (in the reaction region axial and transverse temperature
4876 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

gradients of the order of ∼5 × 106 K=m are observed, which 2500


necessitates a 2D simulation of this system). Calculation of k = 0.6 W/m/K
W
an approximate reaction time scale (methane concentration 2000
at the maximum temperature over the reaction rate from

Wall Temperature [K]


Eq. (1)) is ∼10 s, whereas an average radial conduction 1500
7.5 W/m/K
time for a 600 m gap distance is ∼100 s. This di5erence 100 W/m/K
in time scales explains the radial gradients seen in Figs. 2
1000
and 3 for these kinetics and length scales. Despite the #uid
transverse gradients, there are no signi=cant transverse gra-
500
dients within the walls themselves in all cases studied due
to the short timescale for conduction in the wall and their (a)
large aspect ratio. The e5ect of operating conditions on spa- 0
tial temperature and species uniformity will be discussed
further below. 1000

[K]
500
4. Role of wall thermal conductivity and external heat

CL
T -T
loss in materials integrity and ame stability 0

W
h = 0 W/m 2/K
Our simulations indicate that the wall thermal conductiv- -500 8 W/m 2/K
ity plays a vital role in the #ame stability and materials in- 14 W/m 2/K
tegrity of microburners. The wall plays a dual, competing -1000
(b)
role in the overall heat transfer. On one hand, it provides a 0 2 4 6 8 10
Axial Displacement [mm]
route for heat transfer from the post-combustion region to
upstream for preheating that is necessary for ignition and Fig. 4. (a) Wall temperature vs. axial displacement for several wall ther-
#ame stability. On the other hand, it allows exterior heat mal conductivities. As the wall thermal conductivity increases, the wall
losses, which can delay ignition and cause extinction. The temperature “evens” out, reducing possibilities for hot spots. The param-
dominant route is delineated next. Note that the importance eters are V0 = 0:5 m=s, h = 8 W=m2 =K, L = 600 m and Lw = 200 m.
(b) Radial gradients between the wall interior and the center line (CL)
of wall conduction on #ame stability has been illustrated in-
gas-phase temperatures vs. axial displacement for increasing exterior heat
dependently for Swiss roll type microburners using an ana- losses. With increasing heat losses, the radial gradients increase. The pa-
lytical treatment of the governing equations (Ronney, 2003). rameters are V0 =0:5 m=s, kw =7:5 W=m=K, L=600 m, and Lw =200 m.
Hot spot formation and high wall temperatures can be se-
rious problems causing the materials to melt or degrade. The
wall thermal conductivity has a large e5ect on the wall tem- To quantitatively di5erentiate between the di5erent re-
perature pro=le. Fig. 4a shows the wall axial temperature gions and monitor the possible #ame blowout, the #ame lo-
pro=le for di5erent wall thermal conductivities. Very low cation was computed as the axial position (node) with the
wall thermal conductivity results in much steeper axial tem- greatest reaction rate. In all cases, this node occurred on the
perature gradients within the wall, with lower upstream wall centerline. Fig. 5a shows the #ame location vs. wall conduc-
temperatures and higher downstream wall temperatures. For tivity for three exterior heat transfer coeCcients indicated.
walls with intermediate thermal conductivities, on the other Figs. 5b–d show temperature contours for selected wall ther-
hand, axial temperature pro=les are more uniform, and ma- mal conductivities.
terials with metallic-like thermal conductivity are practically For low wall thermal conductivity, the #ame location is
isothermal. considerably downstream and blowout may occur depend-
Although the wall is essentially isothermal in the trans- ing on burner length and #ow velocity. For cases of low
verse direction for all cases, a large wall thermal conduc- wall thermal conductivities, the upstream heat #ux through
tivity greatly reduces the axial temperature gradients and the wall is limited, causing low upstream wall temperatures.
spreads the energy, regulating the maximum wall tem- This in turn results in slower preheating of the #uid to the
perature observed. For example, Fig. 4a shows that for ignition temperature, which shifts the #ame location down-
very small wall thermal conductivities, resembling asbestos stream. In contrast to common thinking and previous mod-
(0:6 W=m=K), the wall temperature approaches the adiabatic els (Daou, 2002), very insulating materials are not better ei-
#ame temperature of ∼2100 K. When the wall thermal con- ther from the hot spot point of view discussed above or the
ductivity is increased to that of a high-temperature ceramic #ame stability. Lack of upstream heating is one mechanism
(7:5 W=m=K) or a low-conductivity metal (100 W=m=K), of loss of #ame stability.
the maximum wall temperatures are approximately 1600 As the conductivity increases slightly to moderate values,
and 1400 K, respectively. These temperatures can be real- typical of ceramics, a drastic reduction in #ame location
izable experimentally. from the entrance is observed, as shown in Fig. 5a. Our
D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882 4877

5 60
2 Extinction
h = 14 W/m /K
50
4 k = 7.5 W/m/K
Flame Location [mm]

40 k = 3.0 W/m/K

QL [W/m]
3 k = 1.0 W/m/K
30

2 20
8 W/m2/K Blowout
10
2
1 0 W/m /K
Asbestos SiO Al O
2 2 3
Carbon Pt 0
Steel 0 5 10 15 20
(a)
0 h [W/m2/K]
10-1 100 101 102
Fig. 6. Integrated heat loss from the wall to the surroundings from the inlet
Wall Thermal Conductivity [W/m/K] to the #ame location as a function of exterior heat transfer coeCcient for
2380 three di5erent wall thermal conductivities. Extinction occurs at ∼50 W=m.
The parameters are V0 = 0:5 m=s, L = 600 m, and Lw = 200 m.
(b)

(c)
transfer is in the exterior #uid, even in the limit of low wall
300 (d)
thermal conductivity and high exterior heat losses.
The above order of magnitude estimates indicate that ex-
Fig. 5. (a) Flame location vs. wall thermal conductivity for three di5er- tinction for high conductivities and large exterior heat trans-
ent exterior convective heat loss coeCcients. Temperature contours for fer coeCcients is not caused because of enhanced transverse
h = 14 W=m2 =K with kw = 1:1 (b), 3.0 (c), and 14:0 W=m=K (d), re- heat transfer through the wall. For these cases, the wall ax-
spectively. For very low wall thermal conductivities, the limited axial
ial temperature is nearly uniform as shown in Fig. 5d. As a
conduction through the wall forces the #ame location downstream. For
high wall thermal conductivities with low exterior heat losses, conductiv- result, larger external surface area of the burner is at high
ity has little e5ect on #ame location. For high wall thermal conductivities temperatures, compared to the lower conductivities cases,
and high exterior heat losses, increasing wall thermal conductivity shifts especially near the burner entrance. This e5ect increases the
the #ame location downstream. The other parameters are V0 = 0:5 m=s, external heat losses, rendering the #ame less stable. With
L = 600 m, and Lw = 200 m.
such increased heat losses, the average burner temperature
also decreases, as shown in Fig. 5d, and the reaction is not
simulations indicate that some moderate wall conductivity as localized (not shown) as compared to the example de-
is essential for transferring heat upstream to cause ignition picted in Fig. 2. Under these conditions, the burner is more
and stabilize the #ame near the microburner entrance. isothermal-like and extinction is a more global type of phe-
For systems with a high wall thermal conductivity and nomenon (delocalized extinction mode).
low heat losses, the wall thermal conductivity plays a min- Aside from simulations where the wall conductivity was
imal role in #ame location. In contrast, for higher exterior the primary continuation parameter, we have carried simu-
heat losses, high wall thermal conductivities cause the #ame lations for =xed wall conductivity by varying the heat loss
location to be shifted downstream and may cause extinc- (data not shown). Monotonic behavior has been observed;
tion. In order to understand the primary resistance to heat as the exterior heat transfer coeCcient is increased, the wall
loss, the overall transverse heat transfer coeCcient, U , is temperature near the inlet is decreased, and the #ame lo-
computed from cation is shifted downstream because the feed is preheated
more slowly. Extinction or blowout eventually occurs when
1 Lw 1 the external convective heat loss coeCcient is within the
= + : (3)
U kw h range of 14 –20 W=m2 =K. Typical values for convective heat
loss coeCcients for gases in free convection are in the range
Typical ranges for the parameters we consider are of 2–25 W=m2 =K (Incropera & DeWitt, 1996). As the ex-
Lw ∼100–1000 m, kw ∼0:1–100 W=m=K and h∼1– ternal heat transfer coeCcient increases, larger transverse
20 W=m2 =K. The typical resistance through a wall of thick- temperature gradients are observed, as seen in Fig. 4b.
ness 200 m ranges from 2 × 10−3 to 2 × 10−6 Km2 =W, To better understand the underlying cause for the di5er-
whereas the typical resistance of the exterior #uid ranges ence between extinction and blowout, the heat lost to the
from 1 to 0:05 Km2 =W. Thus, for the relatively thin walls surroundings from the wall was integrated from the inlet to
considered here, the primary resistance to transverse heat the #ame location, indicated as QL . Fig. 6 shows QL vs. the
4878 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

20 lowed while allowing combustion within the microburner.


Unignited Interestingly, such materials were found also experimentally
to be more robust (Masel & Shannon, 2001; Jensen et al.,
15 2003).
The results we obtain here demonstrate that self-sustained
h [W/m /K]

combustion at the microscale is possible. The heat recircula-


2

10 Extinction tion along the walls is key in predicting this behavior. Note
Blowout also that old experiments were done mostly with cold walls
(Lewis & von Elbe, 1987) as transient #ames passed through
5
a narrow channel (reminiscent of our previous modeling
work (Raimondeau et al., 2002)). So the #ame stability ana-
Ignited lyzed here numerically is more appropriate for self-sustained
0
10
-1
10
0
10
1
10
2
devices used for energy generation such as the one of Masel
Wall Thermal Conductivity [W/m/K] and Shannon (2001) and Jensen et al. (2003).
Fig. 7. Stability diagram for methane combustion in a microburner. The
plotted points indicate blowout or extinction and the line is a smooth
interpolation of the points. In the low wall thermal conductivity regime, 5. E#ects of ow velocity
limited heat transfer within the wall retards preheating, whereas for high
wall conductivity, heat loss from the entire burner surface dominates. The The e5ect of inlet velocity on the #ame location in mi-
parameters are V0 = 0:5 m=s, L = 600 m, and Lw = 200 m. croburners is shown in Fig. 8a. As the inlet velocity increases
from moderate values, the convective timescale decreases,
and a shift in the #ame location downstream is observed, as
expected. At suCciently high #ow velocities, blowout oc-
external convective heat transfer coeCcient. As the exter- curs. This behavior is consistent with parabolic (boundary
nal convective heat transfer coeCcient increases, there is a layer approximation) #ow simulations (Raimondeau et al.,
monotonic increase in QL . It appears that extinction occurs 2002). However, in contrast to parabolic (boundary layer
when a constant amount of heat per unit time is removed approximation) simulations, a non-linear relationship is ob-
from the system in this region, which, for these conditions, served for relatively slow #ows. In particular, for suCciently
is ∼50 W=m. If #ame conditions are unfavorable, yet the low inlet velocities, the #ame location increases again sig-
QL has not reached the critical value, blowout occurs. This ni=cantly as the inlet velocity decreases, resulting in extinc-
critical value of QL may not hold constant for di5erent #ow tion.
velocities. Based on a simple CSTR or 1D reactor model, For low inlet velocities (high residence times), the con-
in dimensionless quantities, extinction is expected when the vective time scale becomes long. In order to understand this
ratio of heat generated and removed (inverse of Zeldovich extinction mechanism, Fig. 8b shows #uid centerline reac-
number) reaches a constant value (Ronney, 2003). Reduced tion rate pro=les at selected #ow velocities. For slow #ows,
models are obviously desirable to quantitatively test such axial conduction through the gas and the solid dominate,
criteria using CFD output, and this will be the subject of spreading the heat, leading to nearly isothermal conditions.
future work. This behavior is similar to the high conductive walls case
Fig. 7 shows the locus of extinction or blowout points analyzed above in Figs. 5 and 7. Under such conditions, the
from the one-parameter continuation runs shown in Fig. 5 maximum temperature in the reactor decreases, the reaction
plus simulations of varying heat transfer coeCcient men- occurs over a larger area of the burner (delocalized reac-
tioned above. There are two regimes in determining the tion zone), and heat transfer near the microburner entrance
#ame stability within these microburners. For materials of is enhanced. If the axial conduction draws too much heat
low wall thermal conductivity, while exterior heat losses away from the #ame, the exterior heat loss is enhanced, and
are unimportant, the limited upstream heat hinders ignition, extinction occurs. For lower inlet velocities, the wall tem-
causing blowout. In contrast, for suCciently high wall ther- perature remains high while the power generated by the re-
mal conductivities, exterior heat losses play a dominant role action decreases, resulting in an increase in heat lost relative
in #ame stability and cause a more spatially global type of to heat generated, causing extinction.
#ame stability loss when QL becomes signi=cant. For fast inlet velocities (low residence times), the con-
Fig. 7 indicates that there appears to be an optimum vective time scale becomes short compared to the axial con-
range of wall thermal conductivities in the range of 3– ductive time scale. This has a twofold e5ect. First, the #ame
5 W=m=K (obviously, this range may depend on operat- loses little heat to the wall and more power is generated, re-
ing conditions). According to Incropera and DeWitt (1996), sulting in a high temperature, and a short, intense reaction
these thermal conductivities correspond to ceramics such as zone. Second, the wall takes a longer distance to preheat the
high-temperature silica and alumina. Within this conductiv- feed to the ignition temperature increasing the #ame loca-
ity range, maximum exterior heat transfer coeCcient is al- tion, leading to blowout.
D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882 4879

5 2360 (a)
(b)
4 (c)
Flame Location [mm]

1330
3
(d)

2 300
Contours of Temperature [K]
9.70 (a)
1 (b)
(c)

0
0 0.5 1 1.5 2 4.85
(d)
(a) Inlet Velocity (m/s)
30 1500 0.00
Contours of Reaction Rate [kgmole/m3/s]
Reaction Rate [kgmoles/m /s]

25 1000
0 (a)
3

(b)
To [K]

0.5 (c)
20 500 1.1
1.9 m/s 0.5
1.40 m/s (d)
0
15 0 0.1 0.2 0.3
y [mm]

10 1
Contours of Conversion
V = 0.55 m/s
5 Fig. 9. Contours of temperature, reaction rate, and conversion for plate
0.38 m/s
distance L of (a) 400 m, (b) 600 m, (c) 1000 m, and (d) 4000 m.
0 The parameters are V0 = 0:5 m=s, kw = 7:5 W=m=K, h = 0 W=m2 =K, and
0 1 2 3 4 5 Lw = 200 m.
(b) Axial Displacement [mm]

Fig. 8. (a) Flame location vs. inlet velocity. A minimum #ame location
exists. For fast #ows, blowout occurs, whereas for slow #ows, global
type extinction occurs due to slow convective heat transfer. (b) Reaction Fig. 9 shows contours of temperature, reaction rate, and
rate pro=les on the #uid centerline. High velocities exhibit a thin intense conversion for several plate separations. For large separa-
reaction zone and slow velocities exhibit broader, less intense reaction tion distances, such as the 4 mm case, the reaction rate is
zone. The inset shows transverse inlet temperature pro=les for three inlet
velocities indicated. The parameters are kw =7:5 W=m=K, h=8 W=m2 =K,
low and reaction spreads out over approximately 5 –6 mm
L = 600 m, and Lw = 200 m. of the burner due to slow heat transfer from the wall where
ignition starts towards the centerline where uncombusted
mixture exists. For smaller separation distances, such as the
The inset in Fig. 8b shows the transverse temperature 600 m case, reaction is more localized, with greater inten-
pro=les at the inlet for three di5erent inlet velocities. Sharp sity, causing steeper temperature and composition gradients.
temperature gradients near the inlet caused by energy con- When the separation distance is small (e.g., 400 m),
duction from the hot walls result in increased inlet temper- the transverse length scale of the #uid is small enough that
atures, which in turn cause ignition and higher than normal the transverse heat transfer a5ects the centerline tempera-
adiabatic #ame temperatures. Large inlet velocities decrease ture, and thus, the maximum #uid temperature. In this case,
the temperature gradients, causing inlet temperatures to be the maximum #uid temperature is signi=cantly lower than
closer to 300 K. For large-enough inlet velocities, the inlet that for larger separations and incomplete conversion is ob-
solution approaches a Dirichlet-like situation. served. It appears that for channels that are 400 m and be-
low, combustion becomes diCcult because of the increased
transverse heat transfer and there exists a point where com-
6. E#ects of microburner dimensions bustion is no longer possible. Jensen et al. experimentally
determined this distance to be on the order of 100 m for
The above simulations have been conducted for a few materials that do not adsorb radicals (Jensen et al., 2003).
microburner dimensions. Here we compare the combustion There appears to be an optimum plate separation, of approx-
characteristics in microburners when varying the separation imately 600 m, that exhibits the shortest #ame location.
distance between the two plates as well as the wall thickness. This distance permits the wall to preheat the inlet #uid fast
4880 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

2000 2800
LW = 50 µm
2600
1800
Wall Temperature [K]

2400

Temperature [K]
200 µm 2200
1600
2000
600 µm
1400 1800
1600
1200
1400

1000 1200
0 2 4 6 8 10
0 0.2 0.4 0.6 0.8 1
Axial Displacement [mm] (a) Conversion

Fig. 10. Axial wall temperature pro=les for three di5erent wall thick- 2800
nesses. Thicker walls show higher temperature uniformity and lower tem- 2600
perature hotspots. The parameters are V0 = 0:5 m=s, kw = 7:5 W=m=K,
h = 6 W=m2 =K, and L = 600 m. 2400

Temperature [K]
2200

enough, yet not to reduce signi=cantly the centerline tem- 2000


perature. 1800
The wall thickness has also a large e5ect on the tem-
1600
perature pro=le in the microburner walls. Fig. 10 contrasts
the wall axial temperature pro=les for 50, 200, and 600 m 1400
walls. For thin walls, the wall temperature pro=le is similar 1200
to that of the #uid. There is a peak temperature of approx- 0.005 0.01 0.015
imately 1900 K near the #ame location, and the inlet wall (b) Time [s]
is ∼1000 K. Thicker walls exhibit more uniform temper- Fig. 11. (a) Limit cycle for a point on the centerline, indicating
atures, as they have larger area for axial heat conduction. self-sustained oscillations in cases of high heat losses near extinction.
As a result of temperature uniformity, lower maximum tem- (b) The temperature of this point vs. time. The oscillations occur with
peratures (∼1500–1600 K) and higher wall temperatures a period of ∼1 ms. The parameters are V0 = 0:5 m=s, kw = 7:5 m=s,
near the inlet (∼1350–1450 K) are observed. These higher h = 12 W=m2 =K, L = 600 m, and Lw = 200 m.
inlet temperatures tend to enhance #ame stability because
of their increased preheating capability but promote overall In these oscillatory cases, the reaction quickly uses up the
heat loss, especially from the crucial preheating region. In methane and oxygen, causing a rapid increase in #uid tem-
previous experimental work, #ames migrated to the areas perature. Once the conversion is almost complete, the reac-
of the burners with thicker walls (Jensen et al., 2003). It is tion dies, and the gaseous temperature drops, as fresh fuel
possible that this is due to the increased axial heat transfer. and air enter the reactor. The thermal mass of the reactor
walls is large enough that the wall temperature is practically
constant. The wall then serves as a heat source for re-ignition
7. Periodic oscillatory behavior near extinction of the cold incoming reactants. While the di5erences in os-
cillatory behavior observed here from that in a CSTR with
The previous solutions are stationary. Near extinction, the heat loss need further investigation, the presence of the wall
stationary solution can become unstable in cases of high heat and its nearly constant temperature may play an interesting
losses, giving rise to periodic behavior. Under these condi- and possibly unique role in oscillations of the distributed
tions, the real part of at least one of the eigenvalues of the system.
Jacobian becomes positive. Transient simulations are then Since the wall temperature changes less than 1 K in these
necessary to observe the behavior of the system, especially oscillations, thermal shock is not a consideration. A primary
far from Hopf bifurcations. For these transient cases, a tran- concern with these oscillations is pressure waves associated
sient solver was used, using a slightly perturbed steady state with the rapid density changes in the #uid. These oscillations
solution as the initial condition. As an example, Fig. 11a cause #uctuations in the local pressure with swings of up to
shows a limit cycle for a point on the centerline in one of 150 Pa for the depicted conditions that may vibrate the sys-
these unstable cases. Fig. 11b shows the temperature of the tem apart. This is then a consideration for small-scale com-
point used in Fig. 11a as a function of time. The period of bustion devices where heat losses are signi=cant. (Norton
these cycles is on the order of 1 ms. and Vlachos, 2003).
D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882 4881

8. Conclusions Acknowledgements

The combustion characteristics in microburners were This work was supported by the Army Research OCce
studied using an elliptic two-dimensional computational under contract DAAD19-01-1-0582. Any opinions, =ndings,
#uid dynamics model. Transverse gradients in tempera- and conclusions or recommendations expressed are those of
ture and reaction rate were observed for many conditions, the authors and do not necessarily re#ect the views of the
despite the small scale of the burner. Thick walls and/or Army Research OCce. The authors also like to thank Prof.
larger thermal conductivities tend to make the burner more Paul Ronney of USC for useful comments on the submitted
isothermal-like. manuscript.
The wall thermal conductivity plays a vital role in #ame
stability. Two mechanisms of losing #ame stability have
been observed. Low wall thermal conductivity limits the
upstream heat transfer through the wall, which limits the
preheating of the feed, inhibiting the onset of combustion, References
and causes blowout. Low wall thermal conductivity also
causes hot spots of high temperatures within the wall, which Aghalayam, P., & Vlachos, D. G. (1998). The roles of thermal and
can lead to mechanical failure. High thermal conductivity chemical quenching in NOx and fuel emissions: Combustion of
surface-stabilized hydrogen/air mixtures. A.I.Ch.E. Journal, 44(9),
walls are essentially isothermal and have lower tempera-
2025–2034.
tures. However, they o5er a larger hot area for external heat
Dagaut, P., Boettner, J. C., & Cathonnet, M. (1991). Methane oxidation:
transfer and become susceptible to spatially global-like ex- Experimental and kinetic modeling study. Combustion Science and
tinction. An optimum wall thermal conductivity of approxi- Technology, 77, 127–148.
mately 3–5 W=m=K, typical of ceramics, allows the largest Daou, J. (2002). In#uence of conductive heat-losses on the propagation
external heat transfer coeCcient for the conditions studied. of premixed #ames in channels. Combustustion and Flame, 128,
The #ow rate has a strong e5ect on #ame stability. Fast 321–339.
#ows decrease the rate of heat transfer upstream and can Davy, H. (1817). Some researches on #ame. Transactions of the Royal
Society of London, 107, 45–76.
cause blowout. Slow #ows can also cause loss of #ame
Eckbreth, A. C. (1988). Laser diagnostics for combustion temperature
stability by making conductive heat transfer so dominant
and species. Kent, United Kingdom: Abacus Press.
that the reaction zone is spread, and the external heat losses
Fluent (2002). Fluent 6.0. Lebanon, N.H.
become too large compared to heat generation to sustain
Fukuda, M., Koji, K., & Sakamoto, M. (1981). On quenching distance
combustion. An optimum #ow velocity for greatest #ame of mixtures of methane and hydrogen with air. Bulletin of the JSME,
stability exists (∼0:5 m=s for the conditions studied here). 24(193), 1192–1197.
Thin walls (∼50 m) exhibit large axial temperature gra- Ganley, J. C., Seebauer, E. G., & Masel, R. I. (2003). Porous anodic
dients, resulting in hot spots. Thicker walls (∼200–600 m) alumina posts as a catalyst support in microreactors for production of
have a large cross-sectional area, which allows for greater hydrogen from ammonia. A.I.Ch.E. Journal. (accepted).
heat transfer and more uniform, lower temperatures. Burn- Gray, P., & Scott, S. K. (1990). Chemical oscillations and instabilities,
nonlinear chemical kinetics. Oxford: Clarendon Press.
ers with small plate-separation distances (∼400 m) su5er
Groppi, G., & Tronconi, E. (2000). Design of novel monolith catalyst
from axial heat transfer through the walls of the burner, re- supports for gas/solid reactions with heat exchange. Chemical
ducing the maximum #uid temperature and #ame stability. Engineering Science, 55, 2161–2171.
Burners with large gaps (∼4 mm) slowly transport heat ax- Groppi, G., Tronconi, E., Berg, M., & Forzatti, P. (2000). Development
ially to preheat the feed, resulting in a late #ame ignition and application of mathematical models of pilot scale catalytic
and possible blowout. combustors fueled by gasi=ed biomasses. Industrial and Engineering
Combustion at the microscale can o5er advantages. Aside Chemistry Research, 39, 4106–4113.
from the larger radial heat transfers, faster ignition can hap- Hayes, R. E., Kolaczkowski, S. T., Thomas, W. J., & Titiloye, J. (1996).
Transient Experiments and Modeling of the Catalytic Combustion of
pen, and low temperatures could be envisioned (mild com- Methane in a Monolith Reactor. Industrial and Engineering Chemistry
bustion especially near extinction) that could be bene=cial Research, 35, 406–414.
for NOx reduction. However, loss of #ame stability due to Incropera, F. P., & DeWitt, D. P. (1996). Fundamentals of heat and
blowout, extinction, or oscillations and high wall tempera- mass transfer. New York: Wiley.
tures are main issues that require careful thermal manage- Jensen, C., Masel, R. I., Moore, G. V., & Shannon, M. (2003). Burner
ment. Ideally, burner walls should have anisotropic thermal designs for microcombustion. Combustion Science and Technology,
conductivity; they should allow axial but inhibit transverse In Press.
conduction. Such anisotropic walls would allow upstream Lewis, B., & von Elbe, G. (1987). Combustion, >ames and explosions
of gases. Orlando, FL: Academic Press.
heat #ux to preheat the feed, yet not allow transverse heat
Linan, A., & Williams, F. A. (1993). Fundamental aspects of combustion.
loss to the surroundings. This can be accomplished exper- New York: Oxford University Press.
imentally through either composite materials or by manu- Maekawa, M. (1975). Flame quenching by rectangular channels as
facturing small volumes of vacuum within the burner walls a function of channel length for methane-air mixture. Combustion
with high aspect ratios. Science and Technology, 11, 141–145.
4882 D. G. Norton, D. G. Vlachos / Chemical Engineering Science 58 (2003) 4871 – 4882

Maruta, K., Koichi, T., Ahn, J., Borer, K., Sitzki, L., Ronney, P., & Sitzki, L., Borer, K., Schuster, E., Ronney, P. D., & Wussow, S.
Deutschmann, O. (2002). Extinction limits of catalytic combustion (2001). Combustion in microscale heat-recirculating burners. The
in microchannels. The twenty ninth international symposium on Third Asia-Paci?c Conference on Combustion, Seoul, Korea.
combustion, Sapporo, Japan: The Combustion Institute. Uppal, A., Ray, W. H., & Poore, A. B. (1974). On the dynamic behavior
Masel, R. I., & Shannon, M. (2001). Microcombustor having of continuous stirred tank reactors. Chemical Engineering Science, 29,
submillimeter critical dimensions, US06193501. USA, The Board of 967–985.
Trustees of the University of Illinois, Urbana, IL. Uppal, A., Ray, W. H., & Poore, A. B. (1976). The classi=cation of
Mauss, F., & Peters, N. (1993). Reduced Kinetic Mechanisms for the dynamic behavior of continuous stirred tank reactors-In#uence of
Premixed Methane-Air Flames. In N. Peters, & B. Rogg (Eds.), reactor residence time. Chemical Engineering Science, 31, 205–214.
Reduced kinetic mechanisms for applications in combustion systems, Vlachos, D. G., Schmidt, L. D., & Aris, R. (1993). Ignition and extinction
Vol. 15 (pp. 58–75). New York: Springer. of #ames near surfaces: Combustion of H2 in air. Combustion Flame,
Norton, D. G., & Vlachos, D. G. (2003). Oscillations in microburners: 95, 313–335.
Premixed methane=air mixtures, on CD of the Third Joint Meeting Vlachos, D. G., Schmidt, L. D., & Aris, R. (1994). Ignition and extinction
of the V.S. Sections of The Combustion Institute. The Combustion of #ames near surfaces: Combustion of CH4 in air. A.I.Ch.E. Journal,
Institute, Chicago, IL. 40(6), 1005–1017.
Ono, S., & Wakuri, Y. (1977). An experimental study on the quenching of Westbrook, C. K., & Dryer, F. L. (1981). Simpli=ed reaction mechanisms
#ame by narrow cylindrical passage. Bulletin of the JSME, 20(147), for the oxidation of hydrocarbon fuels in #ames. Combustion Science
1191–1198. and Technology, 27, 31–43.
Raimondeau, S., Norton, D., Vlachos, D. G., & Masel, R. I. (2002). Williams, F. A. (1991). Overview of Asymptotics for Methane Flames.
Modeling of high temperature microburners. Proceedings of the In M. D. Smooke (Ed.), Reduced kinetic mechanisms and asymptotic
Combustion Institute 29: 901–907 (2003). approximations for methane-air >ames, Vol. 384 (pp. 68–85). Berlin:
Ronney, P. D. (2003). Analysis of non-adiabatic heat-recirculating Springer-Verlag.
combustors. Combustion and Flame, (accepted).

You might also like