Sei sulla pagina 1di 9

Proceedings of the Combustion Institute, Volume 28, 2000/pp.

783791

A COMPUTATIONAL STUDY OF COMBUSTION INSTABILITIES DUE


TO VORTEX SHEDDING
C. FUREBY
Department of Weapons and Protection
Division of Warheads and Propulsion
FOA Defence Research Establishment
S-172 90 Stockholm, Sweden

The influence of unsteady vortex shedding on combustion instabilities is examined using simulation
databases for two configurations: a dump combustor and a model afterburner equipped with a bluff-body
flame holder for different operating conditions. These databases have been generated using a recently
developed flame-wrinkling large eddy simulation (LES) model, described elsewhere. As a first necessary
step, a detailed comparison is made between predicted and experimentally obtained statistics to quantify
the accuracy of the LES and determine whether LES reproduces the observed statistical trends in the
experiments. The objective of the present study is to examine these databases using visualization techniques
in order to investigate the mechanisms responsible for combustion instabilities in relation to unsteady
vortex phenomena. Combustion instabilities are found, experimentally as well as computationally, in both
configurations and under different operating conditions. Based on the LES databases, the origin of these
instabilities can be identified and attributed to key events such as vortex shedding, excitation of pressure
fluctuations due to exothermicity, extinction by strain, and operating conditions. The unsteady behavior of
the dump combustor is dominated by extinction due to the high strain rate, while the dynamics of the
model afterburner was associated mainly with excitation of pressure fluctuations due to exothermicity and
operating conditions.

Introduction
The development of future combustors, including
lean premixed prevaporized (LPP) combustors and
ramjets, in which the flame is stabilized in the wake
of a flame holder or a rearward facing step, requires
better understanding of combustion instabilities, unsteady vortex dynamics, ignition, flashback, and
flame-holding capacity. The operation of such devices is often impaired by potentially harmful combustion instabilities, which at the lean limit may
cause blow off, and at the rich limit may lead to
flashback. In dump combustors and combustors with
bluff-body flame holders, vortices are formed in the
shear layer, separating regions of high-speed flow
from regions of low speed flow due to KelvinHelmholtz instabilities. In general, the high-speed flow is
composed of an unburned mixture of fuel and air
while the low-speed stream is composed of hot combustion products, forming the flame holder recirculation region behind the dump plane or the bluffbody. Turbulent mixing after the trailing edges of a
bluff-body flame holder or behind the step in a
dump combustor together with the effects of recirculation are known to be important to the stability
of the flame and have been successfully examined in
different experimental facilities (e.g., Refs. [13]).
Computational studies are often restricted by the
783

limitations of the Reynolds average simulation (RAS)


models [4] and the turbulent combustion models
used to represent the mean flame and its effects on
the flowfield. The concept of large eddy simulation
(LES) [5] thus provides a more natural approach to
unsteady flows with ranges of parameters similar to
those of practical or engineering-like systems [6].
The present study is concerned with the application of LES to examine the key steps of combustioninduced instabilities and the associated unsteady
vortex dynamics. To this end, two laboratory devices,
a dump combustor [7] and a model afterburner with
a triangular-shaped bluff-body flame-holder [8],
have been selected as well-characterized and wellbehaved representatives of real combustors. For
both facilities, extensive experimental [7,8] and LES
[9,10], databases for non-reacting and reacting conditions exists. The LES databases are obtained using
a recently developed flame-wrinkling LES combustion model [9,10], carefully validated against firstand second-order statistical moments of velocity,
temperature, and species concentrations using different experimental sources. The main objective of
the present study is to use the LES databases to
obtain a more complete understanding of the physics
behind combustion-induced instabilities due to vortex shedding. It should be pointed out that the discussion of the results draws on a deep analysis of the

784

COMBUSTION INSTABILITY AND PULSATIONS

extensive material in the LES databases, and selected figures are used only to illustrate some general features of the combustion dynamics. A further
objective is to establish how well the LES flamewrinkling model can represent different modes of
operation found in the laboratory experiments. In
this work, we discuss flowfield parameters such as
vorticity, density, and rate-of-strain and how they are
related to key phenomena such as ignition, flashback, and flame stabilization.

The Computational Approach


With LES, we attempt to resolve most of the entrainment-dominating large-scale coherent structures (CSs) by choosing the cutoff wavelength within
the inertial subrange. The challenge is to correctly
mimic the fluid dynamics near the cutoff and to ensure proper interactions between resolved and unresolved eddy scales. The LES equations describe
the evolution of the large-scale features of the flow
and are derived by applying a low-pass filter to the
balance equations of mass, momentum, and energy,
given appropriate constitutive equations. The effects
of the unresolved eddies result in additional unknown subgrid terms in the LES equations that must
be separately modeled. Subgrid models for non-reacting LES have been developed in the past [11],
but few extensions to reacting flows have been made
[9,10,1214], since additional closure problems arise
from the combustion-related terms which are difficult to model. However, the unresolved transport
terms are not unique to reacting flows and can be
modeled by conventional subgrid models. A recent
alternative is monotone integrated LES (MILES)
[15], in which the reacting NavierStokes equations
are solved by high-resolution locally monotonic algorithms with built-in low-pass filtering and subgrid
turbulence models. In this study, we use a flamewrinkling LES combustion model [9,10] to examine
the reacting flow in the previously described combustor configurations.
In the flame-wrinkling LES combustion model,
the reaction is measured with a reaction coordinate
b satisfying
t(qb) div (qvb) div (D grad b) m

where v is the velocity; q, density; D, effective diffusivity, and m


, mass supply term. If the diffusion
reaction terms are combined and represented by
qSu|grad b|
where Su is the laminar flame speed, a Hamilton
Jacobi equation is obtained. If the governing equations, that is, the balance equations of mass, momentum, and energy, and the HamiltonJacobi
equation for b are low-pass filtered using the kernel

G, the LES equations result. The unresolved transport terms are not unique to reacting LES, and ordinary subgrid models, such as the one-equation
eddyviscosity model [16], can therefore be used to
close the LES equations and to model the effects of
the subgrid eddy scales on the resolved flow. The
definition of the Favre-filtered reaction coordinate
b provides a suitable measure of the large-scale geometry of the flame, but to accurately simulate flame
propagation we must also consider the subgrid flame
wrinkling. It can be shown [9] that the filtered
source term in the HamiltonJacobi equation can be
decomposed as

qSu|grad b| qSu R
where   denotes surface filtering, and
|grad b|
R
is the amount of flame surface within the support of
the filter kernel G. By introducing the flame-wrinkling density

N R/|grad
b|
where

|grad b|
is the area of the resolved flame surface, we may take
the decomposition one step further, so that

qSu|grad b| qSuN|grad b|
From the definition of N, it is evident that it represents the amount of surface per unit filtered flame
surface. Based on the true balance equation for N, a
modeled balance equation has been proposed [9].
By decomposition into total and surface strain rates
rT and rI, respectively, and by introducing the subgrid generation and removal rates G and R, this
model becomes
t(N) vI N GN R(N 1) N(rI rT)
where vI is modeled as
slip Sun/N
vI v (1 b)v
and n is the unit normal vector of b. Following [9
10], v is the LES velocity, vslip is the velocity difference over the flame due to the density ratio and the
subgrid turbulence, and Sun/N represents the effects of differential propagation and cusp formation.
The turbulenceflame interaction model uses the
flame speed model of Gulder [17], resulting in expressions for G and R [9]. To account for strain and
curvature, a transport equation for the flame speed
Su is introduced [9], under the assumption that Su is
advected with the flame velocity and affected only
by the resolved strain rate and chemical time scale.
The LES model finally consists of low-pass filtered
equations of mass, momentum, and energy; the

COMBUSTION INSTABILITIES DUE TO VORTEX SHEDDING

HamiltonJacobi equation for b; and model equations for N and Su. To close the equations and to
model the subgrid turbulence, a one-equation eddy
viscosity model [16], is used. The laminar flame
speed and the associated strain response data are
derived from databases to correspond to the freestream flow conditions.
The governing equations are discretized using an
unstructured finite volume method. Second-order
schemes are used in space and time, central differencing for velocity, a bounded normalized variable
diagram (NVD) scheme for scalars [18], and a threepoint backward differencing in time. To decouple
the pressurevelocity system, a Poisson equation,
obtained from the discretized continuity and momentum equations, is constructed for the pressure.
The set of scalar equations are solved sequentially
with iteration over the explicit coupling terms to obtain rapid convergence. The segregated approach results in a Courant number restriction; a maximum
Courant number of 0.5 gives satisfactory numerical
stability and temporal accuracy, but a value of 0.2 is
preferable for temporal accuracy.

785

Dump Combustors

Fig.
1.
Typical
profiles
of
(a)
v1,

(b) vrms
1 , (c) T, and (d) YCO2 for the dump combustor at
x1/h 5.0. , LES of non-reacting case; , LES of
reacting case at 170 80 40; - -, LES of reacting case
at 255 120 60; , experiments of non-reacting case;
and experiments of reacting case.

The dump combustor chosen has been used by


Ganji and Sawyer [1], Keller et al. [2], and Pitz and
Daily [7], and experimental data exist for various operating conditions. It consists of a rectilinear premixing section followed by a contraction to one half
of its height h continued by a step expansion into the
combustor, while in the simulations only the expansion and the combustor section were considered. In
the LES, appropriate boundary conditions are given
at the inlet (propane/air mixture with an equivalence
ratio of 0.57, p0 101 kPa, u0 13.3 m/s,
and T0 293 K), outlet, top and bottom walls, while
periodic conditions are enforced in the spanwise direction (see Ref. [9]. All simulations are initialized
with quiescent conditions, and the unsteady flow
evolves naturally. The effective computational domain size is 16 h 2 h 4 h in the streamwise,
spanwise, and cross-stream directions, and the grids
typically use between 170 80 40 and 255
120 60 cells, refined in the vicinity of the top and
bottom walls and stretched in other parts of the domain. Non-reacting and reacting LES have been
presented separately [9,19], and here we use these
LES databases to examine the coupling between unsteady coherent vortical structures, exothermicity,
and acoustics.
Figure 1 presents typical profiles of the time-averaged streamwise velocity v1, its (rms) root mean
square fluctuations v rms
1 , the time-averaged temperature T and the time-averaged CO2 mass fraction Y CO2, for the non-reacting and reacting cases,

respectively. The Y CO2 is determined from b and the


initial mass fractions corresponding to the equivalence ratio. The agreement between the predictions
[9], and the experiments [7] is sufficiently good to
warrant this study, although some differences can be
observed in the v rms
1 profile for the reacting case. The
length of the recirculation region is shorter in the
reacting case (l/h 4.6) as compared to the nonreacting case (l/h 7.2) and in good agreement
with the experimental data of l/h 4.5 and 7.3,
respectively. Regions of high-velocity fluctuations
are confined by the shear layer which widens with
downstream distance from the step. In particular,
the peak values of vrms
increase initially due to the
1
formation of large CSs, to finally stabilise around
20% of u0, being in good agreement with experiments [7]. Profiles of T and Y CO2  show good qualitative and quantitative agreement between simulations and experiments, although minor differences in the profile shape are observed for Y CO2.
Global quantities, such as the vorticity thickness,
Strouhal number, entrainment, and shear layer
growth rate, are all very well predicted by the LES
model. Moreover, the actual growth of the shear
layer is delayed in the reacting case as compared to
the non-reacting case due to the temperature-dependent viscosity.
Figure 2 shows schlieren images from Pitz and

786

COMBUSTION INSTABILITY AND PULSATIONS

Fig. 2. Experimental and computational visualization of the dump combustor using (a) schlieren [7], (c) an isosurface
of k2 to illustrate the vorticity, (b) semitranslucent isosurfaces of b to emulate the schlieren image, and (d) contours of
the effective reaction rate combined with iso-surfaces of the peak values. The computational results are selected from a
typical cycle in an animation to match the experimental schlieren image.

vorticity,
Daily [7], semitranslucent iso-surfaces of b,
and the effective reaction rate. Following Jeong and
Hussain [20], the vorticity is represented as an iso
surface of k2 (for values smaller than zero), where
k2 is the intermediate eigenvalue of the tensor L2
D2 W2, where D is the rate-of-strain tensor and
W the rate-of-rotation tensor. This interpretation of
a vortex is based on the fact that L2 determines
the existence of a local pressure minimum due to
vortical motion. By comparing schlieren images with
or the density q, we see that the
animations of b,
LES model is successful in capturing the flow, in
particular the shape of the shear layer and the development of large CS. The spanwise vortices that
shed of the step develop as a result of the rollup of
the boundary layer. Due to concentration of vorticity,
longitudinal vortices develop together with undulations of a newly shed spanwise vortex in areas of

high-strain between subsequent vortices of the same


orientation. As spanwise and longitudinal vortices
are advected downstream, under the influence of
vortex stretching they reconnect and merge in a specific manner [19], forming a mixed pattern of X and
big K vortices that impinge on the lower wall and
are either carried away downstream and transformed
into arches or become trapped in the recirculation
bubble. As result of the vortical flow, the growth of
CSs with downstream distance from the step affects
the recirculation region and the rate of spread of the
upper boundary of the shear layer into the freestream. Volumetric expansion due to exothermicity,
baroclinic torque effects, and temperature-dependent viscosity combine to form thicker vortical structures, as compared to the non-reacting flow, thus reducing the strain of the flame and the mass
entrainment rate. The effective reaction rate

COMBUSTION INSTABILITIES DUE TO VORTEX SHEDDING

787

w
quSuN |grad b|
is primarily confined to topological arrangements of
sheetlike structures, which fold into the cores of the
vortices where rapid burning takes place. The longitudinal vortices mainly wrinkle the reaction sheet,
and hence this surface develops regions of high curvature. As the flame is advected downstream, it
propagates normal to itself at the flame speed SuN,
causing negatively curved wrinkles to contract and
positively curved wrinkles to expand. This also increases the possibility of pockets of reactants breaking away from the rest of the reactants. As spanwise
vortices shed of the step rollup, cold premixed reactants and hot combustion products become entrained and mix macroscopically. Reaction is, however, suppressed by the high strain in the shear layer
after which the reaction continues, causing further
volumetric expansion and intense burning, forming
the characteristic large-scale pattern of burning CSs.

Combustors with Bluff Body Flameholders


The bluff-body flame holder case consists of a rectilinear duct, with rectangular cross-section, that is
divided into an inlet and a combustor section
equipped with a two-dimensional triangular-shaped
flame holder [8]. Results will be discussed for a nonreacting case characterized by u0 17 m/s and T0
293 K (case I) and two reacting cases using a lean
propane/air mixture ( 0.65 at p0 101 kPa),
characterized by u0 17 m/s and T0 288 K (case
II) and u0 34 m/s and T0 600 K (case III),
respectively. In the simulations, appropriate boundary conditions are specified at the inlet, outlet, upper
and lower walls, while periodic conditions are enforced in the spanwise direction. All three simulations are initialized with quiescent conditions, allowing the unsteady flow to evolve by itself. The size of
the domain is 20 h 3 h 3 h (where h is the
bluff-body height) in the streamwise, spanwise, and
cross-stream directions, and different grids, employing between 200 30 80 and 300 45 120
cells, have been used. The grids are refined in the
vicinity the walls and around the flame holder and
stretched elsewhere. Further details from these simulations are presented elsewhere [10].
Figure 3 shows a typical comparison between predicted and experimental data for the non-reacting
case I, as well as for the reacting cases II and III.
The general agreement for v1 and v rms
is good, al1
though it is better for the non-reacting case than for
the reacting cases. Comparing profiles of v1, with
and without combustion, given the same airflow,
shows that the reacting cases have a longer and wider
recirculation region and, hence, a more gradual dissipation of momentum in the wake region as compared to the non-reacting cases. The simulations

Fig.
3.
Typical
profiles
of
(a)
v1,

(b) vrms
1 , (c) T, and (d) YCO2 for the bluff-body flame
holder case at x1/h 11.5. - - , LES of case I;
, LES of case II; , LES of case III; , experiments of
case I, , experiments of case II, and , experiments of
case III.

successfully capture the differences in v1 between


the non-reacting and reacting flows, and more importantly, the differences between cases II and III.
Case III results in the widest and longest recirculation region; the reason for this is mainly due to the
higher flame speed and different density ratios, resulting from the preheated inflow, but is also due to
different levels of flame-generated turbulence. Far
downstream, the velocity profiles parallel those of
fully developed turbulent channel flow but having a
bulk velocity about twice the inlet velocity due to the
volumetric expansion caused by exothermicity. The
streamwise rms-velocity fluctuations vrms
are prop1
erly predicted, and the differences between case II
and III are here very distinct. The predicted profiles
of T are in reasonable agreement with the experimental data from gas analysis [8] and coherent antiRaman scattering (CARS) [21], and hence the predictions are within the overall uncertainty levels of
the measurements, determined by the difference
between measurement data from gas analysis and
CARS. The difference between cases II and III is
evident also from the T profiles, and further emphasized in the probability density functions (PDFs)
of T, where case III shows bimodal PDFs in the
shear layers [10]. Concerning profiles of
Y CO2, the agreement with the measured data is reasonable for both cases, but with some differences far
downstream. Global quantities, such as the length of

788

COMBUSTION INSTABILITY AND PULSATIONS

Fig. 4. Computational visualisations of the bluff-body flame holder cases II (a) and III (b), by means of an isosurface
and contours of the Rayleigh parameter R in copperplate
of k2 to illustrate the vorticity, semitranslucent isosurfaces of b,
style cov[p, Q] at the centerplane. Gray contour lines denote regions with R in copperplate style  0, and black
contour lines denote regions with R in copperplate style  0, representative of driving.

the recirculation region and the Strouhal (St) number, are very well predicted (see Ref. [10]), for the
non-reacting as well as for the two reacting cases.
Figure 4a and b shows iso-surfaces of b and k2 and
contours of the covariance between the pressure and
the heat release for cases II and III, respectively. In
both cases, the flame is essentially confined to topological arrangements of sheetlike structure, emanating in the shear layers, and stabilized through recirculation of hot combustion products in the near
wake. The general flow features include spanwise
vortices being shed of the edges of the prism, either
simultaneously as in case II or alternating as in case
III, and rollup while advected downstream, under
the influence of vortex stretching, allowing reactants
and combustion products to mix macroscopically before burning. For case II, LES predicts a varicoselike behavior, with symmetric shedding of spanwise
vortices, and characterized by longitudinal oscillations. For case III, a sinuous-like behavior is obtained that renders large-scale antisymmetric shedding with every other spanwise vortex being shed
from the upper (or lower) edge of the prism. Depending on if the spanwise vortices are shed simultaneously (as in case II) or alternating (as in case III),
the resulting vortex interaction processes will be different, before being further modified by the exothermicity. For case II, the simultaneous shedding and
rollup results in longitudinal vortices stretched between sequential spanwise vortices on either side of
the centerline of the wake. For case III, the alternate

shedding and rollup results in vortex dynamics similar to that of the non-reacting case with longitudinal
vortices stretched between successive spanwise vortices of alternate sign. In both cases, however, following the first rollup and the subsequent growth of
longitudinal vortices, combustion-related effects,
such as volumetric expansion, baroclinic torque effects, and increased molecular viscosity (due to exothermicity), combine to modify the vorticity, resulting in the development of multiple small-scale
vortices with reduced vorticity magnitude in the core
region, as apparent in the k2 distributions of Fig. 4.
Following Putnam [22], it can be argued that the
dynamics observed are due to the combined effects
of vortex shedding and excitation of acoustic oscillations due to exothermicity. According to Rayleigh
[23], this occurs when a proper phase relationship
between the (periodic) heat release and pressure oscillations exists. To examine this we compare, in Fig.
4a and b, the Rayleigh parameter R in copperplate
style cov[p, Q], where p is the pressure and Q the
heat release, between cases II and III. For case II,
driving occurs frequently in the shear layers between
the spanwise vortical structures and in regions with
positive curvature, while for case III, driving occurs
less frequently in selected regions with positive curvature. In addition to this, other mechanisms, such
as blow off, may be relevant to consider. By varying
u0 and , with reference to case II, we find that the
varicose mode dominates at low-flow velocities,
while for higher-flow velocities, near the blow off

COMBUSTION INSTABILITIES DUE TO VORTEX SHEDDING

limit, the sinuous mode is the preferred configuration of the flame. However, for the asymmetric
flame, a symmetric region is observed near the
prism, where the recirculation region exists.

Discussion and Concluding Remarks


In this study, LES databases of two different laboratory combustors have been examined in order to
gain additional insight into combustion instabilities
and unsteady vortex dynamics. The simulations are
performed with a flame-wrinkling LES combustion
model [9,10], and first a quantitative comparison is
made with experimental data from both combustors
in order to verify the accuracy of the LES model.
These comparisons give satisfactory agreement with
the experimental data for first- and second-order statistical moments of velocity, temperature, and species concentrations, as well as for probability density
functions of the temperature [10]. A qualitative comparison also shows that the LES model is capable of
separating the statistics from two operating conditions being tested on the model afterburner. The
LES suggests that the dynamics of these cases are
different, which is confirmed by high-speed video
photography [8]. Hence, the quality of the simulations is sufficient to use the databases for more indepth studies of basic physical phenomena. This
study does not provide combustor design criteria,
per se, but insight into the underlying physics which
can guide the designer in the development of combustors and passive control methods.
The shed vortices behind the dump plane in a
dump combustor or behind a bluff-body flame
holder grow and merge as they are convected downstream under the influence of vortex stretching, exothermicity, and volumetric expansion. Due to
merger and entrainment, the shear layer widens and
may interact with other shear layers, being either in
or out of phase, depending on the confinement and
the operating conditions. Non-reacting simulations
of the dump combustor and the bluff-body flame
holder combustor supports the notion that Kelvin
Helmholtz instabilities are responsible for rollup of
vortices in the shear layers, having a frequency that
scales with the characteristic length scale and the
flow speed. The predicted pressure spectrum of the
dump combustor displays several coexisting instabilities at frequencies ranging from 20 to 400 Hz. For
the bluff-body flame holder the situation is more
complex; case II shows longitudinal pressure oscillations at 100 Hz, causing the shear layers to rollup
in a symmetric manner, resulting in periodic heat
release that feeds energy back into the pressure. In
case III, the time history of the cross-stream velocity
component indicates vortex shedding, due to vortices being shed from alternate sides of the prism at
about 120 Hz. For this case, the pressure spectrum

789

of shows peaks at 120 and 950 Hz, suggesting that


this mode is a superposition of a low-frequency longitudinal and a high-frequency transverse mode.
The development of large-scale coherent structures enhances macroscopic mixing (eddy scales
larger than the typical grid size D) in the shear
layer(s), although it hinders microscopic mixing
(eddy scales smaller than D), necessary to initiate
combustion. A more detailed analysis of the databases shows that in the presence of large coherent
structures transition to microscopic mixing is initiated in the braid region of the spanwise vortices
where the high strain rates between the two streams
prevail. However, production of fine-scale turbulence appears restricted to the interface between the
two streams. Moreover, the turbulent production is
enhanced during the rollup of the vortices and by
the merging of vortices. In combustors with flame
holders, multiple shear-layer interaction results in an
increase in the flame wrinkling N, which in turn, increases the effective reaction rate, which is followed
by rapid and local exothermicity, as is the case for
the bluff-body flame holder configuration II. Flapping of the flame branches may also cause the flame
to interact with the confinement, as is the case for
the bluff-body flame holder configuration III, which
also leads to intermittent and local exothermicity
often harmful for the combustor wall.
Acknowledgment
The author wishes to thank H. G. Weller for many fruitful discussions and for developing the multipurpose CFD
code FOAM used in the present study.

REFERENCES
1. Ganji, A. R., and Sawyer, R. F., AIAA J. 18:817824
(1980).
2. Keller, J. O., Vaneveld, L., Korschelt, D., Hubbard,
G. L., Ghoniem, A. F., Daily, J. W., and Oppenheim,
A. K., AIAA J. 20:254262 (1982).
3. Smith, D. A., and Zukoski, E. E., Combustion Instability Sustained by Unsteady Vortex Combustion,
AIAA paper 85-1248, 1985.
4. Speziale, C. G., Modeling of Turbulent Transport
Equations, in Simulation and Modeling of Turbulent
Flows, (T. B. Gatski, M. Y. Hussaini, and J. L. Lumley,
eds.), Oxford University Press, New York, 1996.
5. Reynolds, W. C., Turbulence at the Crossroads, (J. L.
Lumley ed.), Springer Verlag, 1992, p. 313.
6. Bray, K. N. C., Proc. Combust. Inst. 26:126 (1996).
7. Pitz, R. W., and Daily, J. W., AIAA J. 21:15651570
(1983).
8. Sjunneson, A., Olovsson, S., and Sjoblom, B., Validation RigA Tool for Flame Studies, Report No.
9370-308 VOLVO Aero AB, S-461 81, Trollhattan,
Sweden, 1991.

790

COMBUSTION INSTABILITY AND PULSATIONS

9. Weller, H. G., Tabor, G., Gosman, A. D., and Fureby,


C., Proc. Combust. Inst. 27:899907 (1998).
10. Fureby, C., Combust. Sci. Technol., in press (2000).
11. Lesieur, M., and Metais, O., Annu. Rev. Fluid Mech.
28:4582 (1996).
12. Grinstein, F. F., and Kailasanath, K. K., Combust.
Flame 100:210 (1994).
13. Menon, S., and Kerstein, A. R., Proc. Combust. Inst.
24:443450 (1992).
14. Piana, J., Ducros, F., and Veynante, D., Turbulent
Shear Flows 11, Grenoble, France, Sept. 810, 1997,
p. 2113.
15. Boris, J. P., Whither Turbulence? Turbulence at the
Crossroads, (J. L. Lumly, ed.), Springer Verlag, Berlin,
1990, pp. 344265.

16. Schumann, U., J. Comput. Phys. 18:376404 (1975).


. L., Proc. Combust. Inst. 23:743750 (1990).
17. Gulder, O
18. Jasak, H., Weller, H. G., and Gosman, A. D., Int. J.
Numer. Methods Fluids 31:431442 (1999).
19. Fureby, C., AIAA J. 37:14011410 (1999).
20. Jeong, J., and Hussain, F., J. Fluid Mech. 285:6994
(1995).
21. Sjunnesson, A., Henriksson, P., and Lofstrom, C.,
CARS Measurements and Visualization of Reacting
Flows in a Bluff Body Stabilized Flame, AIAA paper
92-3650, 1992.
22. Putnam, A. A., Combustion Driven Oscillations in Industry, Elsevier 1971.
23. Rayleigh, J. W. S., Theory of Sound, Vol. II, Dover,
New York, 1945.

COMMENTS
Thierry Baritaud, IFP, France. Did you use the law of
the wall for turbulence and combustion, or just a refined
grid? Can you comment on the influence of the used procedure?
Authors Reply. In all the simulations reported, the grid
was refined toward the combustor walls (including the
prism wall) to obtain y 895 10. The OEEVM
subgrid model is then integrated all the way to the wall.
For non-reacting cases, this approach has been discussed
and examined elsewhere [1], but for reacting flows, no
careful examination of this approach has been made. Depending on the character of the wall in the laboratory combustor, no suitable information is readily available for comparison. This is an important issue but, as yet, there is a
general lack of understanding of how to handle wall boundary conditions well in non-reacting LES, an issue even
more complicated in reacting flows where reactions may
be influenced at the wall by, for example, catalysis.

Authors Reply. In the flame-wrinkling model, the laminar-flamelet approach is adopted in which the flame propagates normal to itself at the local laminar flame speed Su.
Su is considered to be a function of the thermodynamic
state of the reactants, rate-of-strain, and curvature. To represent transport and hereditary effects, an evolution equation for Su is hypothesized in which Su is assumed to be
advected with the surface averaged interface velocity vI
influenced by chemical and strain-rate timescales, as modeled using asymptotic relaxation. The model requires the
unstrained flame speed SuO and the strain-response sect
as input, both of which may be the result of an analysis of
the reactive-diffusive structure of the premixed flame (e.g.,
via one-dimensional full kinetics laminar flame simulations). The model has been used to study several different
flames [1,2] from which it is evident that this model is superior to its equilibrium versions. The numerical results
per se are not influenced by this model. It may also be
possible to derive a similar equation from the reactive-diffusive structure, but it is not clear how to proceed.

REFERENCE
1. Fureby, C., AIAA. J. 37:1401 (1999).
REFERENCES

Arnaud Trouve, George Washington University, USA.


The flame initiation and stabilization model used in your
study is based on a transport equation for the local flame
speed. This model seems empirically based, and its domain
of validity remains unknown. Could you please comment
on the general quality of that model and on the sensitivity
of your numerical results to this particular model.

1. Weller H., Tabor G., Gosman, A. D., and Fureby, C.,


Proc. Combust. Inst. 27:899907 (1998).
2. Nwagwe, I., K., Weller, H. G., Tabor, G. R., Gosman,
A. D., Lawes, M., Sheppard, C. G. W., and Wooley, R.,
Proc. Combust. Inst. 28:5965 (2000).

COMBUSTION INSTABILITIES DUE TO VORTEX SHEDDING


Jim OConnor, Perkins Engines Company Ltd, UK. Can
you expand on why you chose the thin flame approach
for reacting flows?
Authors Reply. Previous experience with different LES
combustion models [1,2], as well as theoretical considerations, suggests that for the cases under consideration, the
FW-LES model is presently the best available candidate.

791

REFERENCES
1. Moller, S.-I., Lundgren, E., and Fureby, C., Proc. Combust. Inst. 26:241248 (1996).
2. Fureby, C., Grinstein, F. F., and Kailasanath, K., Large
Eddy Simulation of Premixed Turbulent Flow in a Rearward-Facing-Step Combustor, AIAA paper 00-0863,
2000.

Potrebbero piacerti anche