You are on page 1of 14

Combustion and Flame 150 (2007) 263276

www.elsevier.com/locate/combustflame

Flame acceleration in the early stages of burning in tubes


Vitaly Bychkov a, , Vyacheslav Akkerman a,b , Gordon Fru a ,
Arkady Petchenko a , Lars-Erik Eriksson c
a Institute of Physics, Ume University, S-901 87 Ume, Sweden
b Nuclear Safety Institute (IBRAE) of Russian Academy of Sciences, B. Tulskaya 52, 115191 Moscow, Russia
c Department of Applied Mechanics, Chalmers University of Technology, 412 96 Gteborg, Sweden

Received 20 March 2006; received in revised form 13 December 2006; accepted 18 January 2007

Abstract
Acceleration of premixed laminar flames in the early stages of burning in long tubes is considered. The accel-
eration mechanism was suggested earlier by Clanet and Searby [Combust. Flame 105 (1996) 225]. Acceleration
happens due to the initial ignition geometry at the tube axis when a flame develops to a finger-shaped front, with
surface area growing exponentially in time. Flame surface area grows quite fast but only for a short time. The
analytical theory of flame acceleration is developed, which determines the growth rate, the total acceleration time,
and the maximal increase of the flame surface area. Direct numerical simulations of the process are performed for
the complete set of combustion equations. The simulations results and the theory are in good agreement with the
previous experiments. The numerical simulations also demonstrate flame deceleration, which follows acceleration,
and the so-called tulip flames.
2007 Published by Elsevier Inc. on behalf of The Combustion Institute.

Keywords: Premixed flames; Flame acceleration; Tulip flames; Direct numerical simulations

1. Introduction turbulent burning [714] and on flame instabilities


[1519].
Another interesting example of flame acceleration
When speaking about flame acceleration, resear-
has been suggested and studied experimentally by
chers typically mean detonation-to-deflagration tran-
Clanet and Searby [20]. Let us consider a flame prop-
sition (DDT) according to the Shelkin scenario; see
agating in a cylindrical tube of radius R with ideally
[16]. However, flame may accelerate not only in the
slip adiabatic walls as shown in Fig. 1. One end of the
scope of the DDT problem. For example, flame ve-
tube is closed; the flame is ignited at the symmetry
locity increases due to external turbulence or intrin-
axis at the closed end. In that case the flame front de-
sic flame instabilities. We do not discuss these ef- velops from a hemispherical shape at the beginning to
fects in the present paper because of limited space; a finger-shape; see Fig. 1. In the process of burning,
one can see the following reviews and papers on the volume of the burnt gas increases as
dV
= Sw Uf , (1)
* Corresponding author. Fax: +46 90 786 66 73. dt
E-mail address: vitaliy.bychkov@physics.umu.se where Uf is the planar flame velocity, Sw is the to-
(V. Bychkov). tal surface area of the flame front, and = f /b
0010-2180/$ see front matter 2007 Published by Elsevier Inc. on behalf of The Combustion Institute.
doi:10.1016/j.combustflame.2007.01.004
264 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

analytical and numerical results agree with the ex-


perimental data [20]. The simulations also clarify the
effect of the so-called tulip flame.
The paper is organized as follows. In Section 2
we develop the analytical theory of flame acceleration
in the early stages of burning. Details of the direct
Fig. 1. Geometry of a flame accelerating due to the initial numerical simulations are presented in Section 3. In
conditions. Section 4 we compare the theory and the simulation
results to the experimental data [20]. The paper is con-
is the expansion factor, defined as the density ratio cluded by a brief summary.
of the fuel mixture and the burnt matter. The major
contribution to the flame surface area Sw in Fig. 1
comes from the flame skirt, Sw 2 RZtip , where 2. The analytical theory of flame acceleration
Ztip is the coordinate of the flame tip, and the vol-
ume of the burnt matter may be roughly evaluated as We consider a flame propagating in a cylindrical
V R 2 Ztip . Then Eq. (1) reduces to tube of radius R with ideally slip adiabatic walls and
dZtip U with one end closed. The flame is ignited at the sym-
= 2 f Ztip , (2) metry axis at the closed end. In the analytical theory
dt R
we will use the dimensionless coordinates (; ) =
which leads to the acceleration of the flame tip as (r; z)/R, velocities (w; v) = (ur ; uz )/Uf , and time
Ztip exp(2Uf t/R). (3) = Uf t/R. We employ the standard model of an in-
finitesimally thin flame front, which propagates nor-
It is interesting to compare this effect to the Shelkin mally with the velocity Uf (or unity in the dimension-
acceleration mechanism. The growth rate 2Uf /R less variables). At the beginning, the flame is ignited
is rather large in comparison with that obtained ac- at the tube axis at the closed end (; ) = (0; 0). Ini-
cording to the Shelkin scenario for laminar flames tially, the front is hemispherical, but the flame shape
[5,6]. However, the Shelkin mechanism is not limited changes as the flame skirt f moves along the tube
in time; it takes place until the detonation is trig- end wall ( = 0) from the axis = 0 to the side wall
gered. In contrast, the acceleration mechanism [20] = 1 as shown in Fig. 2. We stress that Fig. 2 and
works in a short time interval, and it is unlikely to the calculations below consider only the flow infini-
produce a DDT under normal conditions. The accel- tesimally close to the wall, at 0. In that limit the
eration starts when the flame evolves from a hemi- flame front touching the wall may be treated as locally
spherical kernel to the finger-shaped front of Fig. 1 cylindrical not only in the case of a finger shape, but
(t > tsph ); the acceleration stops when the flame skirt even for the hemispherical front. The flame separates
touches the wall (t < twall ). According to the experi- the flow into two regions of the fresh fuel mixture and
mental measurements [20], one has tsph 0.1R/Uf , the burnt matter. The incompressible flow is described
twall 0.26R/Uf , which leaves only a short time in-
terval twall tsph 0.16R/Uf for the acceleration.
It is interesting how strongly the flame surface area
may increase during such a short time interval, but
Ref. [20] did not address this question. The purpose
of the present paper is to answer this question as well
as to clarify other aspects of the flame acceleration
obtained experimentally by Clanet and Searby. The
current study is also conceptually close to the work of
Zeldovich [21].
In the present paper we consider flame accelera-
tion in the early stages of burning in tubes with slip
walls, as proposed by Clanet and Searby [20]. We de-
velop the analytical theory of flame acceleration. We
derive the formulas for the acceleration time interval,
tsph < t < twall , and for the growth rate. We show that
the flame surface area increases because of accelera-
tion approximately by a factor of 2 in comparison
with the tube cross section. We also perform direct
numerical simulations of the flame acceleration. The Fig. 2. Flow close to the tube end wall.
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 265

by the continuity equation and the equation for the flame front
1 v df  
(w) + = 0. (4) ( 1) 1 f2 = 1. (13)
d
According to Eq. (13), we can separate two opposite
The boundary condition at the end-wall = 0 is
regimes of flame propagation: when the flame skirt
v = 0. We are interested in the flow along the wall,
is close to the axis f  1, and when it is close to the
in the limit 0. In the fuel mixture (labeled 1)
wall 1f  1. In the limit of f  1, the flame prop-
the flow is potential; we assume
agates with the velocity df /d = (or R f = Uf ).
The same velocity takes place for an expanding hemi-
v1 = A1 , (5)
spherical flame front. In the other case of 1 f 
where the factor A1 may depend on time. The as- 1, the flame propagation velocity is df /d = 1 (or
sumption is consistent with the porous piston effect R f = Uf ). In that limit a locally cylindrical flame skirt
as the flame approaches the tube wall. In the hemi- approaches the wall; the radial velocity of the fresh
spherical regime of flame propagation it works only fuel mixture tends to zero, and the flame skirt prop-
sufficiently close to the wall as the leading term in agates with the planar flame velocity with respect to
for 0. Then the radial velocity in the fuel mixture the tube end wall. Integrating Eq. (13) with the initial
is calculated from Eq. (4) as condition f = 0 at = 0, we find
   
A 1 1 + f
w1 = 1 . (6) = ln , (14)
2 2 f
or
In Eq. (6) we have taken into account the bound-
ary condition at the side-wall of the tube w = 0 at exp(2 ) 1
f = = tanh( ), (15)
= 1. The velocity distribution in the burnt matter exp(2 ) + 1
(labeled 2) takes the form where

v2 = A2 , (7) = ( 1). (16)
A
w2 = 2 . (8) According to Eq. (15), the flame velocity is equal
2 to the velocity of a hemispherical front close to the
In general, the flow in the burnt matter of Fig. 1 is ro- axis, when 2  1 and f = (or Rf = Uf t).
tational because of the curved flame shape. However, Respectively, one should expect transition to the
close to the wall, the flame front is locally cylindri- finger-shape at the characteristic time
cal and the flow (7) and (8) is potentialy similar to (5)
and (6). In Eq. (8) we have also taken into account the sph 1/2, (17)
boundary condition at the tube axis w = 0 at = 0. when the flame skirt is at f 0.46/. Of course,
To complete the solution we consider the matching there is no exact mathematical definition of the tran-
conditions at the flame front = f , sition time. Still, the characteristic time, Eq. (17),
comes as a parameter into Eq. (15) and may be com-
df
w1 = 1, (9) pared to the experiments [20]. For the expansion fac-
d tors = 68, typical for propane flames, we find
w1 w2 = 1, (10) 5.57.5 and 0.06 < sph < 0.09. The theoretical
v1 = v2 . (11) evaluation (17) agrees quite well with the experimen-
tal estimates sph 0.1 [20] taking into account a
The condition (9) specifies the fixed propagation ve- rather vague definition of the value. Transition to the
locity Uf of the flame front with respect to the fuel finger shape takes place approximately when the
mixture (which is unity in scaled variables). The con- flame skirt has moved halfway to the tube side wall. It
ditions (10) and (11) describe the jump of the normal is much easier to determine the time instant when the
velocity and continuity of the tangential velocity at flame skirt touches the tube wall. Substituting f = 1
the front. We stress that Eq. (11) applies only at the into Eq. (14), we obtain
flame skirt close to the wall and it follows from the  
irrotational assumption combined with the cylindrical 1 +
wall = ln . (18)
flame shape, i.e., no shear across the flame front. Sub- 2
stituting Eqs. (5)(8) into Eqs. (9)(11), we obtain For the expansion factors = 68 we find 0.23 <
wall < 0.28, which is in very good agreement with
A1 = A2 = 2( 1)f , (12) the experimental evaluation wall 0.26 [20]. Thus,
266 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

Fig. 4. Flow close to the tube axis.

or Zwall = R in the dimensional units. By the end of


the acceleration we can neglect the decaying term in
Fig. 3. The time limits of flame acceleration, sph and wall , Eq. (21). Then the flame tip accelerates exponentially,
versus the expansion factor . tip exp( ), with the growth rate

the flame surface area grows almost exponentially in = 2 = 2 ( 1), (23)
time during the interval sph < < wall . The time in- which is a little different from the model estimate 2
stant sph defined by Eq. (17) and the time wall when of Eq. (3); see [20]. This difference is small, since
the flame skirt touches the wall, Eq. (18), are shown / = 0.910.94 for propane burning with = 68.
in Fig. 3 versus the expansion factor . It is also interesting to evaluate total increase of
We can also find evolution of the flame tip. If we the flame surface area during the flame acceleration.
consider the flow along the axis = 0, as shown in By the end of the acceleration, the flame shape is al-
Fig. 4, then the equation for tip becomes most self-similar, with f 1, tip  1, and tip
dtip exp(2 ). We look for the flame shape f = f (, )
v2 = . (19) in the form
d
Equation (19) is the condition of a fixed propagation f (, ) = () exp(2 ). (24)
velocity of a planar flame front written with respect to
the burnt matter. Similarly to Fig. 2, the flame shape is The accuracy of such an approximation is 1  1,
locally planar close to the axis, at 0. In that limit which is acceptable for the propane flames used in the
the flow may be described as a potential one, with the experiments [20]. Then the equation of flame evolu-
axial velocity component v determined by a function tion may be written with respect to the burnt matter as
similar to Eq. (7). Besides, the solution for v along  2
1/2
the axis has to coincide with Eq. (7) at 0, 0. f f f
+ w2 v2 = 1 +
Thus we obtain the same formula for v2 along the axis
as in Eqs. (7) and (12). We stress that such reasoning
f
does not hold away from the axis in the burnt gas, . (25)
where the flow is rotational. Still, in the present cal-
culations we have to know only the gas velocity along Of course, the approximation |f/|  1 holds only
the axis. As a result, we come to the differential equa- near the wall, certainly not near the axis. Still, the
tion for the flame tip, region near the wall contributes most to the flame sur-
dtip face area. Taking into account the exponential regime
2( 1)f ( )tip = , (20) of flame acceleration (24) and the velocity distribu-
d
tion (7) and (8) we reduce Eq. (25) to
with the initial condition tip (0) = 0 and with the so-
lution   d
2( + 1) = ( 1) . (26)
  d
tip = exp(2 ) exp(2 )
4 Integrating Eq. (26), we find
 
= sinh(2 ). (21)
2 = C ( 1) , (27)
Just after ignition, 2  1, the flame tip moves in where C is the constant of integration and
the same way as the flame skirt, tip = f = ; see
 
Eqs. (15) and (21), which are similar to an expand-
ing hemispherical flame front. When the flame skirt =2 1 =2 1 . (28)
1 1
touches the wall, = wall , the position of the flame
tip is In the solution (27) and (28) we have to keep only
the terms of the principal order in 1  1, and take
wall = tip (wall ) = sinh(2wall ) = , (22) into account that f (0) = wall = , Eq. (22). Then
2
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 267


Eqs. (24), (27), and (28) reduce to (ur ) + (uz ur zr )
t z
f = (1 )1/ . (29) 1   2  P 1
+ r ur rr = , (34)
r r r r
We stress that the solution (29) does not satisfy the  
boundary condition at the axis f/ = 0 at = 0. + ( + P )uz zz uz zr ur + qz
t z
This is, of course, a consequence of the approxima-
1   
tion + r ( + P )ur rr ur zr uz + qr
r r
 2
1/2
f f = 0, (35)
1+ (30)  
1 Y
(Y ) + rur Y r
t r r Sc r
used in Eq. (25) in the limit of 1  1. Still, the  
Y
principal contribution to the flame surface area comes + uz Y
z Sc z
from the flame skirt. Another shortcoming of the eval-
Y
uation (29) is that we adopted the potential flow (7) = exp(Ea /Rp T ), (36)
and (8) everywhere in the burnt matter, while in re- R
ality it holds only close to the tube end wall and to where
the tube axis. Outside of these regions, the flow may
 2 
be rotational because of the curved flame shape. Us- = (QY + CV T ) + uz + u2r (37)
ing the evaluation (29), the scaled surface area of the 2
curved flame front is calculated as is the total energy per unit volume, Y is the mass frac-
 tion of the fuel mixture, Q is the energy release in the
1  2
2 f reaction, and CV is the heat capacity per unit mass
Sw / R = 2 1 + d at constant volume. We consider a single irreversible

0 reaction of the first order. Temperature dependence
1 of the reaction rate obeys the Arrhenius law with the
f 2 activation energy Ea and the constant of time dimen-
2 d = 2 . (31)
+1 sion R . The stress tensor ij is
0
 
Thus, for stoichiometric propane flames with = 8, 4 uz 2 ur 2 ur
zz = , (38)
we should expect a maximal flame surface area 3 z 3 r 3 r
 
as large as Sw / R 2 14.2. Of course, order-of- 4 ur 2 uz 2 ur
magnitude evaluations may be obtained within an rr = , (39)
3 r 3 z 3 r
even simpler and cruder approximation of a cylindri-  
4 ur 2 uz 2 ur
cal shape of radius R and height R with Sw / R 2 = , (40)
2. Relative difference between this approximation 3 r 3 z 3 r
 
and Eq. (31) is about 1/. uz ur
zr = + , (41)
r z
and the heat diffusion vector qi is given by
3. Details of the numerical simulations
 
CP T Q Y
To assess accuracy of the analytical theory de- qz = + , (42)
Pr z Sc z
veloped in the previous section, we performed direct  
CP T Q Y
numerical simulations of the hydrodynamic and com- qr = + . (43)
Pr r Sc r
bustion equations including chemical kinetics and
transport processes. In this section we are using di- Here = is the dynamic viscosity, CP is the heat
mensional variables. We investigated the case of an capacity per unit mass at constant pressure, and Pr and
axisymmetric flow described by the equations Sc are the Prandtl and Schmidt numbers, respectively.
The fuel mixture is assumed to be a perfect gas of
1
+ (rur ) + (uz ) = 0, (32) constant molecular weight m = 2.9 102 kg/mol,
t r r z with CP = 7Rp /2m and CV = 5Rp /2m, where Rp =
8.31 J/(mol K) is the perfect gas constant. The equa-
 2 
(uz ) + uz zz tion of state is
t z

1   P P= Rp T . (44)
+ r(uz ur zr ) = , (33) m
r r z
268 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

Similarly to our previous simulations [5,6,15,16], we than the tube radius, 20 < L/R < 100. We employed
used the initial pressure and temperature of the fuel a square grid with the grid walls parallel to the radial
mixture Pf = 105 Pa and Tf = 300 K. We have cho- and axial directions. The grid was uniform with the
sen the chemical parameters of the fuel mixture Ea , cell size 0.2Lf . It was demonstrated recently in [22]
Q, R , which provide strongly subsonic flame prop- that such a cell size is quite sufficient to simulate dy-
agation with the initial Mach number Ma Uf /cs = namics of a strongly curved flame front. Indeed, the
103 corresponding to the planar flame velocity Uf = characteristic length scale of the temperature profile
34.7 cm/s. In that case the flow is almost isobaric and inside the flame (the burning zone) is (45)Lf , while
thermal expansion is coupled to the energy release in the characteristic width of the active reaction zone is
the burning process as comparable to Lf [22]. Thus, taking a grid size of
0.2Lf , we obtain about 5 grid points inside the ac-
T Q
= b =1+ . (45) tive reaction zone and 2025 grid points inside the
Tf C P Tf
flame. For this reason, the grid used in the present
To imitate typical propane burning we chose the ex- simulations is sufficiently fine to resolve the inner
pansion factor = 8, the dynamic viscosity = flame structure even for a strongly elongated flame.
2.38 105 Ns/m2 , and the Prandtl number Pr = The simulation time step was much smaller than the
0.7. The flame thickness has been defined in a con- characteristic time of flame dynamics R/Uf .
ventional way as Similarly to the theory, we consider a tube with
ideally slip adiabatic walls and with one end closed.
Lf . (46) Boundary conditions at the side walls are
Pr f Uf
To avoid the Zeldovich (thermal-diffusion) instability T
ur = 0, uz = 0, =0
we assumed the coefficients of thermal diffusivity and r
fuel diffusion to be equal, i.e., the Lewis number equal at r = 0, R. (49)
to unity, Le Pr/Sc = 1 [3,23]. We chose the value
of the activation energy Ea = 7Rp Tb , which allows To avoid the influence of sound waves and weak
smoothing the reaction zone over a few computational shocks reflected from the open tube end, we applied
cells. Still, in the case of Lewis number Le = 1, nu- nonreflecting boundary conditions at the end. These
merical results do not depend on the magnitude of Ea conditions were tested in detail in [22]; they were
[3]. used also in our numerical works on flame acceler-
Similarly to [6,16], we used the axisymmetric hy- ation due to the Shelkin mechanism [5,6]. The initial
drodynamic Eulerian code developed in Volvo Aero. conditions were chosen in the form
The solver is robust and it was utilized quite suc-  
cessfully in studies of laminar burning, hydrodynamic T /Tf = 1 + ( 1) exp ( R0 )/Lf ,
flame instabilities, and flame acceleration in tubes
if  R0 , (50)
with nonslip at the walls according to the Shelkin
mechanism. The numerical scheme of the code and and T /Tf = , if < R0 , with
the computational method were described in detail in
[16]. We considered different tube radii in the domain T /Tf
Y= , (51)
5 < R/Lf < 30, which corresponds to relatively low 1
values of the Reynolds number related to the flame
propagation, where 2 = r 2 + z2 and R0  R is the initial radius
of a hemispherical flame front. The smaller R0 we
Uf R R use, the better we reproduce point ignition of a flame
Reflame = = = 1050. (47)
Pr Lf front; see Fig. 1. In the present simulations we chose
The Reynolds number related to the flow of the fuel R0 /Lf = 23. Equations (50) and (51) imitate the
mixture is much larger, hemispherical counterpart of the ZeldovichFrank
Kamenentsky solution [3]. The initial conditions at
2R uz
the open tube end are
Reflow = , (48)

and it may reach 103 104 at the end of the flame ac- = f , T = Tf ,
celeration, when the flame skirt touches the wall. For uz = 0, ur = 0, Y = 1. (52)
comparison, steady flows in tubes are usually laminar
at Reflow < 2 103 . Still, the values of Eq. (52) were not fixed during the
In order to reproduce an infinitely long tube, in the simulation run, and they may change because of the
present simulations we used a tube length much larger flame propagation.
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 269

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 5. Flame shape for R = 20Lf at the time instants tUf /R = 0, 0.12, 0.21, 0.28, 0.32, 0.43, 0.64, 1.03. The colors designate
the temperature from the cold gas (blue) to the burnt matter (red). For interpretation of the references to color, in this figure
legend, the reader is referred to the web version of this article.

4. Results and discussion Transition from the hemispherical shape to the


finger-shape and the flame acceleration reproduce
In the present section we discuss the simulations the model [20], Fig. 1, and our theory of Section 2.
results and compare them to the analytical theory of Still, one has to remember that the simulations are
Section 2 and to the experimental measurements [20]. performed for a flame of finite thickness, and the
Typical evolution of a flame ignited at the tube axis limit of an infinitesimally thin flame front may be
at the closed end wall is shown in Fig. 5 for the achieved only asymptotically. We start with a hemi-
tube of radius R = 20Lf for different time instants spherical flame front of small radius 3Lf , shown in
Uf t/R = 0, 0.12, 0.21, 0.28, 0.32, 0.43, 0.64, 1.03 Fig. 5a. At the initial stage the flame expands more
(Figs. 5a5h, respectively). The temperature distrib- or less uniformly outward with velocity Uf ; the fac-
ution is described by the color (from blue in the fresh tor comes from the expansion of the burnt matter.
gas to red in the burnt matter). Mark that Figs. 5a5e The regime of hemispherical expansion goes on un-
are shown in the domain 0 < z/R < 8, but the domain til the flow becomes affected noticeably by the tube
is shifted along the z-axis in the other three figures to sidewall. According to the theory of Section 2, for an
illustrate the flame structure better. Fig. 5 illustrates expansion factor as large as = 8, it happens approx-
different stages of flame dynamics. We observe ex- imately half way from the axis to the wall. Fig. 5b
pansion of the hemispherical flame in the initial stage corresponds roughly to the end of the hemispheri-
(Figs. 5a and 5b), acceleration of the finger-shaped cal stage, though there is no exact definition of this
flame in Figs. 5c5e, fast deceleration of the flame time instant. The respective flow pattern is presented
in Fig. 5f, and the tulip flame in Fig. 5g, which col- in Fig. 6a. It is well known that in the expansion of
lapses again to an almost planar front in Fig. 5h. Later, a spherical (or hemispherical) flame in the opening,
the flame will acquire a curved shape again because of the burnt matter is quiescent, and the fuel mixture is
the DarrieusLandau (DL) instability [3]. This effect pushed in the radial direction. The flow in Fig. 6a is
was well investigated in the previous numerical simu- already different from the ideal hemispherical expan-
lations of laminar burning in cylindrical tubes [16,18], sion. The walls do not allow radial expansion, and the
and we do not consider it here. flow is pushed along the axis both in the fuel mixture
270 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

(a)

(a)

(b)

Fig. 6. Flame shape and the flow for R = 20Lf at the time
instants tUf /R = 0.12 (a), 0.28 (b).

and in the burnt matter in agreement with the theoret-


ical formulas (5)(8).
As soon as the walls influence the flow consider-
(b)
ably, the velocity pattern changes. The flame becomes
elongated, and it was described as a finger-shaped Fig. 7. The flame surface area versus time for tubes of the
in the experiments [20]. The skirt of the flame front radii R/Lf = 1030: (a) usual time scale; (b) logarithmic
is almost cylindrical at that stage (which looks almost time scale.
planar in the axial cross-section shown in Fig. 5d).
The respective flow is shown in Fig. 6b. In that case a certain sense they are related to a tulip-flame phe-
there is almost no flow in the radial direction in the nomenon [2427]. However, the concave shape of the
fuel mixture, and the flame skirt propagates to the wall flame front in Fig. 5g is not stationary; it is much
with the planar flame velocity Uf . Strong expansion more elongated than the respective stationary concave
of the burning matter leads to a strong flow in the ax- flames developing because of the DL instability [16].
ial direction in the burnt gas. This flow drifts the tip For this reason, the tulip shape collapses, producing
of the finger-shaped flame in the axial direction and an almost planar flame front in Fig. 5h. At this point
produces a flame acceleration in agreement with the we typically stopped the simulation runs.
experiments [20] and the theory of Section 2. By the It is also interesting to compare the theory, the
end of the acceleration process, the height of the flame simulations, and the experiments quantitatively. The
tip, Ztip , is much larger than the tube radius. Because most important characteristic of the described process
of the elongated shape, the surface area of the flame is the flame surface area scaled by the tube cross-
front, Sw , is much larger than the tube cross-section, section, Sw / R 2 . In the case of an infinitely thin
R 2 (which is the surface area of a planar flame prop- flame, the surface area characterizes the total burn-
agating in the same tube). ing rate. When the flame thickness is finite, these
When the flame skirt touches the walls at the time two values (scaled) in general do not coincide, but
instant twall the acceleration stops. Because of the the correlation is strong. The relative increase of the
small angle between the flame and the wall, the prop- surface area is shown in Fig. 7 for the tube widths
agation velocity of the angle point considerably ex- R/Lf = 1030. In all plots of Fig. 7a one can eas-
ceeds the planar flame velocity; the skirt catches up ily recognize the interval of the flame acceleration.
with the flame tip rather fast. The surface area of the The same plots presented in the logarithmic scale
flame front decreases drastically, and the flame decel- in Fig. 7b demonstrate clearly that the flame accel-
erates as shown in Figs. 5e and 5f. The deceleration erates exponentially in time as Sw exp( Uf t/R).
is accompanied by one more interesting effect, which The scaled growth rate is evaluated as = 2 in
may be observed in Figs. 5e5g: the convex flame tip the model [20], Eq. (3), and as = 2 ( 1),
in Fig. 5e becomes flat in Fig. 5f and inverted (con- Eq. (23), in the analytical theory of Section 2. Here we
cave) in Fig. 5g. Similar modifications of the flame would like to comment that agreement of the plots for
shape have been observed experimentally in [20]. In R/Lf = 15, 20, 25 in Fig. 7 is even too good, taking
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 271

Fig. 8. The scaled growth rate versus the tube radius. The Fig. 9. The growth rate versus the equivalence ratio for
markers show the results of numerical simulations. The solid propane flames for tube radii 2.5 and 5 cm. The filled mark-
line corresponds to the theory, Eq. (23); the dashed line ers show the experimental results [20]; the empty markers
present the model [20]. and the lines show the theoretical result, Eq. (23) calculated
using the data of Table 1.

into account inevitable numerical errors of the simu-


lations; we believe this is a coincidence. On the other
hand, difference of the curves with R/Lf = 30 and
R/Lf = 15, 20, 25 should not be treated as disagree-
ment. Particularly, the difference in wall obtained for
R/Lf = 30 and, say, R/Lf = 25 is about 10%, which
is a reasonable accuracy for direct numerical simu-
lations. Fig. 8 compares the theoretical formulas for
the growth rate with the numerical results obtained
for different tube radii. We can see that the numerical
growth rate increases for wider tubes and approaches
the theoretical evaluations for R/Lf > 20. For tubes
Fig. 10. The scaled flame surface area, Sw /R 2 , the tip po-
of smaller radii R/Lf < 20 the numerical growth rate
sition, Ztip /R, and the skirt position, Zskirt /R, versus time
is noticeably smaller than the theoretical predictions
for R = 20Lf .
because of the smoothing effect of finite flame thick-
ness. Fig. 9 compares the experimental results on the
growth rate to the theoretical formulas for different
equivalence ratios of propane flames for tubes of two
radii, R = 2.5 and 5 cm. In order to perform the com-
parison, we have taken the values of the planar flame
velocity Uf and the expansion factor for propane
flames versus the equivalence ratio from [28]; see
Table 1. The filled markers in Fig. 9 show the ex-
perimental results for the growth rate [20], and the
curves present the theoretical results; the empty mark-
ers on the curves correspond to the data for Uf and
available in Table 1. We can see that the theoretical Fig. 11. Maximal surface area of the flame front versus the
predictions agree quite well with the experiments. tube radius. The markers show the results of numerical sim-
The acceleration in Fig. 7a ends with a sharp peak ulations. The solid line corresponds to the theory, Eq. (31).
corresponding to the maximal surface area Sw when
the flame skirt touches the wall. Fig. 10 presents to- presented by the dashed line. We can see that the
gether the flame surface area, the tip position, and the maximal surface area increases for wider tubes, and it
position of the flame skirt when it moves along the approaches the theoretical prediction. This tendency
tube side wall for R/Lf = 20. We can see clearly is related to the influence of finite flame thickness,
that the abrupt drop of the surface area happens at which becomes smaller for wider tubes. Besides, one
the time instant twall , which is the starting point for has to remember that Eq. (31) was obtained with ac-
the skirt-plot. The maximal surface area obtained curacy 1  1. Keeping in mind the limitations of
in numerical simulations is shown in Fig. 11 versus the theory, we can say that the theory and the sim-
the tube width R/Lf (markers); the theoretical for- ulations agree rather well. Unfortunately, the experi-
mula (31) for an infinitesimally thin flame front is ments [20] did not address the maximal surface area,
272 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

Table 1
Parameters of the propaneair frame versus the equivalence ratio according to [28]
0.63 0.65 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
6.04 6.21 6.56 7.15 7.66 8.02 8.08 8.0 7.88 7.74 7.6 7.48
Uf (cm/s) 14.7 17.0 21.7 30.3 37.4 41.8 42.9 39.9 32.2 22.6 13.8 9.8

Fig. 12. Position of the flame tip by the end of the acceler- Fig. 13. Position of the flame tip versus time according
ation versus the tube radius. The markers show the results to the theory (dashed line), Eq. (21), experiments [20]
of numerical simulations; the solid line corresponds to the for the propaneair mixture with = 0.7, and modeling.
theory, Eq. (22). The marker indicates the time instant when the flame skirt
touches the wall.
and we cannot compare the respective numerical and
theoretical results to the experimental ones. Fig. 12
shows the height of the flame tip till the end of the
flame acceleration obtained in the numerical simula-
tions and predicted by the theory for = 8, Eq. (22).
This plot looks qualitatively similar to the previous
one: the numerical results for the tip height increase
with the tube radius (with decreasing influence of
flame thickness) and approach the theoretical predic-
tions. Fig. 13 presents the tip position for a propane
air flame of equivalence ratio = 0.7 obtained ex-
perimentally [20]. The dashed line shows theoretical
predictions of Section 2, Eq. (21), for = 6.56 cor- Fig. 14. Characteristic time instants of the flame evolution,
responding to = 0.7. We can see that the theory tsph , twall , and ttulip , versus the ratio R/Uf (s). The markers
is in excellent agreement with the experiments. The show the experimental results [20]; the lines correspond to
marker indicates the time instant when the flame skirt the theoretical formulas, Eq. (17), (18), and (59), with = 1.
touches the wall, with the tip position Zwall = R, takes the average value 7.28.
Eq. (22). Strictly speaking, the theory of Section 2
holds only up to this time instant. Still, as we can theoretical evaluations of these two time instants are
see from Fig. 13, the theory describes the tip posi- given by Eqs. (17) and (18). As for the numerical re-
tion quite well even beyond that limit until a tulip sults, it is difficult to extract tsph from the numerical
flame is formed with temporal retreat of the axial simulations because of the vague definition of that in-
point at the flame front. For comparison we show also stant. In the experiments [20] this time has not been
the results of numerical simulations in Fig. 13. The measured either; instead, it was calculated on the ba-
numerical simulations provide qualitatively the same sis of the model Eqs. (1)(3) using the formula
picture. However, one should not expect quantitative
R
agreement in that case because of the inevitable nar- tsph = twall ln(Zwall /R). (53)
row tubes in the simulations. Besides, the simulations 2Uf
assumed = 8, which is different from the experi- In contrast, the instant twall is defined quite well, and
mental data of Fig. 13. it may be compared to the theory and the experiments.
Another interesting question concerns time limits Fig. 14 compares the theoretical predictions and the
of the acceleration: it starts when the spherical flame numerical results to the experimental data of [20].
develops into a finger-shaped front at tsph , and it At that point we have to say that different equiva-
stops when the flame touches the wall at twall . The lence ratios of the propaneair flame correspond to
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 273

Fig. 15. Position of the flame skirt versus time according to Fig. 16. Correlation of the skirt velocity with the value
Eq. (54), experiments [20] for the propaneair mixture with Zwall /R. The markers show the experimental results [20];
= 0.7 (markers), and modeling. the line corresponds to Eq. (54) with = 1.25.

the different expansion factors = 68 and, there- The characteristic experimental evaluations for these
fore, lead to somewhat different values for tsph , twall . values were twall 0.26R/Uf , ttulip 0.33R/Uf ,
To plot the theoretical lines we took an intermediate tinv 0.07R/Uf . We point out that the last value
value = 7.23 calculated as an average for all exper- correlates rather well with the inverse growth rate
imental points. Still, even with this inevitable reason 1 R/Uf . The reason for such a correlation may
for scattering of the experimental points, the theory be demonstrated within the following semiqualitative
agrees well with the experiments. model. As the skirt position moves along the wall, the
As soon as the flame touches the side wall of the burnt gas velocity, Eq. (7), modifies roughly as
tube, the skirt of the flame front sweeps along the wall
at high speed because of the small angle between the v2 = A2 ( skirt ) 2( 1)( skirt )
flame front and the wall. Fig. 15 shows the skirt posi- ( skirt ), (55)
tion versus time obtained experimentally for = 0.7
in [20] and in the numerical simulations. Similarly to and Eq. (20) reduces to
Fig. 13 for the flame tip, the numerical results under- dtip
estimate the skirt velocity, presumably because of the tip 2 ( wall ) + , (56)
d
influence of the finite flame thickness. It was pointed
out in [20] that during the interval of flame decelera- taking into account Eq. (54). Because of the initial
tion the skirt velocity is almost constant, and it corre- conditions tip (wall ) = , Eq. (22), we can neglect
lates rather well with the value Uf Zwall /R. The last the last term in Eq. (56) in the limit of 1  1.
value may be calculated on the basis of our theory in Then, integrating (56), we find that

Section 2 as 2 ( 1)Uf , which gives the skirt  
velocity tip = ( 1) exp ( wall )
 
 + ( wall ) . (57)
Uskirt = 2 ( 1)Uf , (54)
The instant ttulip corresponds to Ztip = Zskirt or
with the factor comparable to unity. The physical dtip /d = 0, and the interval of inversion tinv =
meaning of is that it compares the skirt velocity ttulip twall is calculated from (57) as
during the deceleration to the characteristic parameter
 
of velocity dimension Uf Zwall /R during the flame
acceleration. In Fig. 15 we took = 1.25 to obtain Uf tinv /R = inv = 1 ln . (58)
1
the best agreement of Eq. (54) with the experiments;
Uskirt is calculated from the slope of Zskirt . However, The evaluation (58) for tinv is simpler than the respec-
in general may depend on the expansion factor . tive formulas in [20], and it demonstrates clearly the
Fig. 16 compares the skirt velocity measured experi- correlation tinv 1 R/Uf . Still, one has to remem-
mentally to the theoretical value for Uf Zwall /R; the ber that calculations (55)(58) are only a qualitative
solid line presents Eq. (54) for = 1.25. Indeed, we model based on the empirical formula (54). There-
can see a very good correlation between the experi- fore, in reality, one should interpret (58) as another
mental data and Eq. (54). Eventually the skirt catches correlation of the form
up with the tip, and the flame front becomes first pla-
tinv = 1 R/Uf (59)
nar and then inverted (or tulip-shaped). The time in-
stant of inversion was called ttulip in [20], and the time with the coefficient comparable to unity. The eval-
interval needed for the inversion is tinv = ttulip twall . uation (59) is compared to the experimental data in
274 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

process is inversion of a flame shape, when a con-


vex flame becomes concave or vice versa. From that
point of view, the notion of tulip flames captures
only one possibility, namely, transition from a con-
vex flame shape to a concave one, as in experiments
[20] and in the present numerical simulations. Clanet
and Searby related inversion of the flame shape to
the flame deceleration and the RichtmeyerMeshkov
mechanism of shock interaction with a surface of den-
sity jump [29,30]. The first theoretical discussion of
flameshock interaction in that context goes back to
Markstein [31], who considered linear dynamics of a
Fig. 17. A curved flame shape in an open 2D channel with slightly curved flame under the action of a pulse ac-
adiabatic and slip walls [15]. celeration representing a shock wave. Unfortunately,
Markstein adopted rather peculiar initial conditions at
Fig. 14 for ttulip = twall + tinv with = 1. In the nu- the flame front corresponding to zero time derivative
merical simulations of the present paper the flame of the flame perturbations. This may be achieved by
inversion happens somewhat later, about ttulip a unique combination of the growing and decaying
0.45R/Uf , which is also related to the stabilizing role modes of the DL instability, which should happen ex-
of finite flame thickness together with lower values of actly at the time when a shock arrives. At present,
the skirt velocities in Fig. 15. it is difficult to say what the reason for such a pe-
Finally, we would like to discuss the so-called culiar choice was. Probably, Markstein tried to imi-
tulip flame observed in the experiments [20] and in tate a curved stationary flame typical to the nonlinear
the present simulations. The notion of a tulip flame stage of the DL instability, or to achieve initial con-
is poorly defined, which allows rather loose specula- ditions similar to the RichtmeyerMeshkov instabil-
tions about its origin; see, for example, [2427] and ity [29,30]. Anyway, the specific choice of the initial
other references in [20]. The problem is reviewed conditions led to the conclusion of limited validity
quite well in [20]. From the geometrical point of that a curved flame front is always inverted by a
view, the tulip shape is simply a concave shape of shock irrespective of the shock intensity. More phys-
a curved flame front with a sharp cusp pointing to ical initial conditions for the linear problem corre-
the burnt gas. Such a shape may be produced for spond to a dominating growing DL mode considered
a large number of different reasons with little rela- in [32]. It was obtained in [32] that a flameshock
tion between each other or with no relation at all. collision (face-to-face) may lead to different possi-
The shape may be stationary or nonstationary. For ex- bilities depending on the shock intensity: (1) a flame
ample, a concave stationary flame develops because front may be temporarily stabilized without inversion;
of the DL instability, as demonstrated in the sim- (2) it may be stabilized with inversion; or (3) it may
ulations for 2D channels and cylindrical tubes [15, be destabilized with inversion. The last two possibil-
16,19]. Even more, in a 2D channel with adiabatic ities may be interpreted as formation of a tulip-
slip walls the concave (tulip) flame is identical to flame. It was also demonstrated in direct numerical
a convex (non-tulip) flame. For example, Fig. 17 simulations [33] that the critical values for the flame
shows a characteristic shape of a curved stationary inversion obtained in the linear analysis [32] work
flame in a 2D channel shown by isotherms (the flame quite well even for the nonlinear problem of flame
propagates downward, R = 20Lf [15]). The adia- shock collision. A shock of sufficient strength may
batic slip walls of the channel at x/R = 0, 1 may even turbulize the flame, but such a case was not con-
be treated as symmetry axes. Then, reflecting the sidered in [33]. In the numerical simulations of the
curved flame with respect to the axis x/R = 1, we present paper we used nonreflecting boundary condi-
obtain a concave (tulip) flame, but taking a reflec- tions at the open end of the tube, and there were no
tion with respect to the axis x/R = 0, we come to real shocks coming back to the flame. However, flame
a convex (non-tulip) flame with identical proper- deceleration in the time interval twall < t < ttulip has
ties. The difference between the convex and concave the same influence on flame dynamics as an effec-
stationary flames appears only in a 3D flow (e.g., tive coming shock. The flame propagation velocity
in cylindrical tubes), where the convex flames have Uf Sw / R 2 drops from 2Uf to Uf . This veloc-
higher propagation velocity [16]. Flame interaction ity drop is much larger than the critical values needed
with nonslip walls may lead to the oscillating con- to invert the flame shape by a shock [32,33]. For this
cave shape of the flame front, which may be also reason, in the present simulations we observe flame
interpreted as tulip flames [22]. Another interesting inversion leading to a strongly concave (tulip) flame
V. Bychkov et al. / Combustion and Flame 150 (2007) 263276 275

shape with the surface area Sw / R 2 = 2.4 exceed- the most interesting stages of flame dynamics: hemi-
ing noticeably the respective surface area of the sta- spherical expansion, acceleration of the finger-shaped
tionary concave flames Sw / R 2 = 1.11.2; see [16]. flame, rapid deceleration when the flame skirt sweeps
The maximal flame curvature in the tulip flame cor- along the tube side wall, and inversion of the flame
responds to the local (in time) maximum of the flame shape, which may be also interpreted as a tulip
surface area in Fig. 7a achieved at Uf t/R 0.6. Later, flame.
the strongly concave flame collapses again to the pla- In the present paper we considered only the case of
nar flame shape, which may eventually develop to a adiabatic tube walls. Losses to the walls may reduce
stationary DL curved flame like that obtained in [16]. the effective thermal expansion and make the flame
To conclude the discussion, we believe that the notion acceleration noticeably weaker. This is a subject for
of tulip flame is misleading, since it embraces too future work.
many combustion phenomena of different origin. In-
stead, it would be much better to talk about curved
flames, indicating the reason for the curved shape. Acknowledgments
A particular example is inversion of the curved flame
shape because of a shock as in [32,33] or because of The authors thank Geoffrey Searby for the experi-
flame deceleration as in [20] and in the present pa- mental data of Figs. 9 and 1316 and Mikhail Ivanov
per. for useful discussions. This work has been supported
in part by the Swedish Research Council (VR) and by
the Kempe Foundation.
5. Summary

In the present paper we investigated acceleration References


of premixed laminar flames in tubes according to the
mechanism suggested in [20]. Acceleration happens [1] K.I. Shelkin, Zh. Exp. Teor. Fiz. 10 (1940) 823.
due to the initial geometry of flame ignition, when [2] J.H.S. Lee, Handbook of Shock Waves, Academic
a flame develops from a spherical kernel to a finger- Press/Elsevier, New York, 2001.
shaped front with a tip accelerating exponentially in [3] Ya.B. Zeldovich, G.I. Barenblatt, V.B. Librovich, G.M.
time. This effect should not be confused with the Makhviladze, Mathematical Theory of Combustion and
Explosion, Consultants Bureau, New York, 1985.
classical Shelkin mechanism of flame acceleration be-
[4] G. Roy, S. Frolov, A. Borisov, D. Netzer, Prog. Energy
cause of the nonslip at the walls. Acceleration stud- Combust. Sci. 30 (2004) 545.
ied in the present paper happens with a much higher [5] V. Bychkov, A. Petchenko, V. Akkerman, L.E. Eriks-
growth rate, but only during a limited time, and it is son, Phys. Rev. E 72 (2005) 046307.
unlikely to lead to the DDT under normal conditions. [6] V. Akkerman, V. Bychkov, A. Petchenko, L.E. Eriks-
As an option, the present effect may act as a precursor son, Combust. Flame 145 (2006) 206.
to the Shelkin mechanism. Another option, which yet [7] G. Searby, P. Clavin, Combust. Sci. Technol. 46 (1986)
has to be checked, is that the present mechanism can 167.
trigger detonation in a strongly preheated fuel mix- [8] R. Aldredge, F. Williams, J. Fluid Mech. 228 (1991)
ture similarly to the modelling [34] and the analytical 487.
[9] B. Denet, Phys. Rev. E 55 (1997) 6911.
theory [35].
[10] V. Bychkov, Phys. Rev. E 68 (2003) 066304.
We have developed the analytical theory of flame [11] V. Akkerman, V. Bychkov, Combust. Theory Model. 9
acceleration. We have obtained the growth rate, (2005) 323.
which is slightly different from the previous estimates [12] D. Bradley, P. Gaskell, X. Gu, A. Sedaghat, Combust.
of [20]. We have also derived formulas for the time Flame 143 (2005) 227.
limits of the acceleration process: the acceleration [13] V. Bychkov, A. Petchenko, V. Akkerman, Combust.
starts when the hemispherical flame develops to the Sci. Technol. 179 (2007) 137.
finger-shaped flame; the acceleration stops when the [14] T. Poinsot, D. Veynante, Theoretical and Numerical
flame skirt touches the walls. The growth rate and Combustion, Edwards, Ann Arbor, MI, 2005.
the time limits of the process agree well with experi- [15] V. Bychkov, S. Golberg, M. Liberman, L.E. Eriksson,
Phys. Rev. E 54 (1996) 3713.
ments [20]. We find the maximal increase of the flame
[16] V. Bychkov, S. Golberg, M. Liberman, A. Kleev, L.E.
surface area by the end of the acceleration, which
Eriksson, Combust. Sci. Technol. 129 (1997) 217.
is roughly Sw / R 2 2, Eq. (31). We performed [17] S. Kadowaki, Phys. Fluids 11 (1999) 3426.
direct numerical simulations of the flame accelera- [18] V. Bychkov, M. Liberman, Phys. Rep. 325 (2000)
tion for the complete set of combustion equations. 115.
The simulations results and the theory are in agree- [19] S. Kadowaki, T. Hasegawa, Prog. Energy Combust.
ment with the previous experiments, demonstrating Sci. 31 (2005) 193.
276 V. Bychkov et al. / Combustion and Flame 150 (2007) 263276

[20] C. Clanet, G. Searby, Combust. Flame 105 (1996) 225. [28] S.G. Davis, J. Quinard, G. Searby, Combust. Flame 130
[21] Ya.B. Zeldovich, Air Mat. Commands Technol. (2002) 123.
Reports, 1947, F-TS-1226-1A. [29] R.D. Richtmeyer, Comm. Pure Appl. Math. 13 (1960)
[22] V. Akkerman, V. Bychkov, A. Petchenko, L.E. Eriks- 297.
son, Combust. Flame 145 (2006) 675. [30] E.E. Meshkov, Fluid. Dynam. (USSR) 4 (1969) 101.
[23] G.I. Barenblatt, Ya.B. Zeldovich, A.G. Istratov, Prikl. [31] G. Markstein, in: G. Markstein (Ed.), Non-Steady
Mech. Techn. Fiz. 2 (1962) 21. Flame Propagation, Pergamon, London, 1964, p. 15.
[24] H. Genoche, in: G. Markstein (Ed.), Non-Steady Flame [32] V. Bychkov, Phys. Fluids 10 (1998) 2669.
Propagation, Pergamon, London, 1964, p. 107. [33] O. Travnikov, V. Bychkov, M. Liberman, Combust.
[25] M. Gonzalez, R. Borghi, A. Saouab, Combust. Flame Sci. Technol. 142 (1999) 1.
88 (1992) 201. [34] L. Kagan, G. Sivashinsky, Combust. Flame 134 (2003)
[26] B. Nkonga, G. Fernandez, H. Guillard, B. Larroutrou, 389.
Combust. Sci. Technol. 87 (1992) 69. [35] V. Bychkov, V. Akkerman, Phys. Rev. E 73 (2006)
[27] M. Gonzalez, Combust. Flame 107 (1996) 245. 066305.

You might also like