[asme asme turbo expo 2010: power for land, sea, and air - glasgow, uk (june 14–18, 2010)] volume...
TRANSCRIPT
Proceedings of ASME Turbo Expo 2010: Power for Land, Sea and AirGT2010
June 14–18, 2010, Glasgow, UK
GT2010-23577
THERMOACOUSTIC INSTABILITIES IN AN ANNULAR RIJKE TUBE
Jonas P. Moeck∗
Markus Paul
Christian Oliver PaschereitInstitut fur Stromungsmechanik und Technische Akustik
Technische Universitat Berlin
10623 Berlin, Germany
Email: [email protected]
ABSTRACT
Thermoacoustic instabilities are a major concern in the de-
sign of gas turbine combustors. Most modern combustion cham-
bers have an annular shape with multiple circumferentially ar-
ranged burners and, accordingly, suffer often from azimuthal in-
stability modes. However, due to the complexity of a full annu-
lar system with a large number of burners, most experimental
and numerical studies focus on single burner systems with es-
sentially purely longitudinal acoustics. In the present work, we
therefore introduce a thermoacoustic surrogate system – an an-
nular Rijke tube – which, albeit its simplicity, possesses the ba-
sic mechanisms to feature unstable azimuthal modes. As in a
conventional Rijke tube, the sources of mean and unsteady heat
release in our set-up are electrically driven heating grids. Differ-
ent azimuthal instability modes are observed in the experiment,
and the effect of two types of circumferential variations of the
power input is investigated. A full suppression of the unstable
modes is achieved by the application of an elementary feedback
controller. The experimental investigations are accompanied by
corresponding calculations with a low-order system model. The-
oretical and experimental results are found to agree well.
[Keywords: combustion instability, thermoacoustic,
azimuthal modes, annular Rijke tube]
∗Address all correspondence to this author.
NOMENCLATURE
A cross-sectional area
a modal coefficient
c speed of sound
ck, cτ low-pass parameters
D duct prolongation matrix
F heat release transfer function
f , g down-/upstream traveling wave amplitude
hm,c|s circumferential basis function
I identity matrix
Im imaginary part
i√−1
K feedback gain matrix
k wave number, eigenfunction index
m azimuthal mode order
N number of heat release elements
P electric power
p acoustic pressure
q heat release rate
Re real part
r duct radius
T transfer matrix
T temperature
u axial particle velocity
X characteristic impedance ratio matrix
x coordinate vector
.
...
1 Copyright c© 2010 by ASME
Proceedings of ASME Turbo Expo 2010: Power for Land, Sea and Air GT2010
June 14-18, 2010, Glasgow, UK
GT2010-23577
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
greek and calligraphic letters
α acoustic mode index
δi j Kronecker delta
φ angular coordinate
γ ratio of specific heats
ω angular frequency
ψ acoustic eigenfunction
ρ fluid density
σ asymmetry parameter
ξ characteristic impedance ratio
ζ damping coefficient
A admittance matrix
F scaled heat release transfer function matrix
R reflection coefficient matrix
S system matrix
Z impedance matrix
..
sub- and superscripts
(·)d downstream of heat source
(·)u upstream of heat source
(·)n in tube n¯(·) mean quantityˆ(·) Fourier transform, amplitude
INTRODUCTION
For reasons of compactness and of having a uniform tem-
perature distribution at the turbine inlet, modern gas turbine com-
bustion chambers mostly have an annular shape with multiple cir-
cumferentially arranged burners. This geometrical set-up hosts
azimuthal acoustic modes, which may become unstable under
certain conditions. If the heat release provided by the flame
responds dynamically to acoustic perturbations, energy can be
added to the acoustic field, and the pressure fluctuations grow
until limited by nonlinear effects [1]. Although investigations on
annular thermoacoustic systems are significantly more complex
than in the purely longitudinal single burner case, a number of
experimental, numerical, and modeling studies focusing on az-
imuthal modes has been performed over the last decade. We give
a brief overview in the following.
Seume et al. [2] applied a feedback control scheme using
modulated pilot fuel to suppress unstable azimuthal modes in a
full scale annular combustion chamber with 24 burners. They
observed 2nd and 4th order circumferential modes. The un-
stable acoustic modes in this type of combustor have been fur-
ther analyzed by Krebs et al., Berenbrink and Hoffmann, and
Kruger et al. [3–6] based on test rig measurements, finite ele-
ment computations, and low-order Galerkin and network type
models. From the experimental data [4], first and second order
azimuthal modes, dominantly standing as well as rotating, were
identified from the magnitude and phase patterns of the acous-
tic pressure along the circumference of the annular combustion
chamber. Introducing a circumferential variation by means of
cylindrical and asymmetric burner outlets applied to part of the
burners was found to have a stabilizing influence [5, 6].
Evesque et al. [7, 8] developed a low order network model
for annular combustion chambers allowing for asymmetries in-
troduced by the burner transfer matrices. They considered an
annular model combustion chamber geometry with no sources
of unsteady heat release and compared eigenvalue computations
with a finite element solver; good agreement was found. A simi-
lar approach was used by Kopitz et al. [9] to model longitudinal
and circumferential instabilities that were observed in an annular
model combustion chamber.
The Rolls-Royce/Cambridge network code was used in a
number of modeling and control studies considering annular
combustor configurations [10–13]. Stow and Dowling [13] im-
plemented the effect of Helmholtz resonators attached to the cir-
cumference of the combustion chamber and studied the associ-
ated modal coupling. Optimal resonator placement for maximum
damping of unstable azimuthal modes was deduced. In [12], the
same authors extended the approach to account for nonlinearities
in the flame model. By using describing function techniques, the
limit cycle amplitude of thermoacoustic oscillations in an annu-
lar configuration could be calculated. Stow and Dowling [11]
extended the approach to time domain simulations based on the
impulse response of heat release perturbations to those in ap-
proach flow velocity and a nonlinear flame model. The final limit
cycle solution of unstable azimuthal modes were always of rotat-
ing type in accordance with the theoretical considerations in [14]
(see below). Morgans and Stow [10] used the same network code
to develop model-based control strategies for thermoacoustic in-
stabilities in an annular model combustor. In the simulations,
they were able to suppress decoupled azimuthal modes in a rota-
tionally symmetric set-up but also coupled modes resulting from
non-identical burners.
Based on a time domain network model, Schuermans et
al. [14] investigated thermoacoustic instabilities in a rotationally
symmetric annular combustion chamber. They observed that,
while growing, the unstable mode was of standing type (resulting
from axisymmetric white noise excitation). However, the final
limit cycle solution was always of dominantly rotating type with
no preference of either spinning direction. It was argued that
the saturation nonlinearity in the flame destabilizes the standing
wave mode at finite amplitude and therefore promotes the travel-
ing wave solution. For a simpler model system consisting of only
two first order azimuthal wave components, this mechanism was
investigated in more detail. Depending on the initial conditions,
the growing wave could be arbitrarily made up of azimuthally
traveling or standing components, but the final periodic solution
was again always of rotating type. It was shown that there ex-
ist two types of equilibrium solutions at finite amplitude in wave
amplitude phase-space, one with equal clockwise and counter-
clockwise traveling wave components and the other purely spin-
2 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
ning. The former, corresponding to a standing wave, was associ-
ated with a saddle point and therefore unstable. The two purely
rotating solutions were found to be stable fixed points.
Staffelbach et al. [15] performed large eddy simulations of
a full annular helicopter combustion chamber. They observed a
dominantly rotating first order azimuthal wave at 740 Hz with a
clockwise/counterclockwise amplitude ratio of 3. The dominant
effect of the acoustic waves on the heat release rate was found
to result from a modulation of the axial mass flow through the
burners. The same configuration was investigated by Sensiau et
al. [16]. Finite element computations based on the Helmholtz
equation with feedback from the heat release (see also [17]) and
non-trivial boundary conditions were used with mean fields and
flame response data obtained from corresponding LES solutions.
Studying thermoacoustic instabilities in annular combustion
chambers is a complex task (from an experimental as well as
from a computational point of view). Therefore, in contrast to
thermoacoustic instabilities in purely longitudinal geometries,
less information regarding azimuthally unstable modes is avail-
able. In this work, we present a mock-up system which is of sig-
nificantly lower complexity than an annular combustion chamber
but has all essential ingredients so that thermoacoustic instabili-
ties associated with circumferential acoustic modes can be stud-
ied.
The generic system set-up we consider is as follows. N
straight tubes (the “burners”) are connected to an annular duct
(the “combustion chamber”) with their downstream end. In each
tube, there is an element that provides a mean heat release and
responds dynamically to fluctuations in axial velocity. A princi-
pal difference to an annular combustion chamber in a gas turbine
is that, in the set-up considered here, the sources of heat release
do not respond to azimuthal velocity fluctuations in the down-
stream annulus but only to an axial perturbation in the tubes. We
assume the latter effect to be generally dominant, as was found
in Ref. [15].
EXPERIMENTAL SET-UP
The annular Rijke tube has 12 tubes connected to an annular
duct at their downstream end. Figure 1 shows a schematic of the
set-up with the basic geometrical dimensions; a photograph of
the experimental arrangement is presented in Fig. 2. The tubes
have an inner diameter of 60 mm with an associated cut-on fre-
quency for the first azimuthal mode beyond 3 kHz at room tem-
perature. The annular duct is 400 mm in length and has a mean
diameter of 720 mm with a “hub–tip ratio” of 0.8. All parts are
made of aluminum with a wall thickness of 10 mm.
As in a conventional Rijke tube (see, e.g., Refs. [18] or [19]),
the sources of mean and unsteady heat release are electrically
driven heating grids. Obviously this type of heat source is much
simpler than a premixed flame, however, both have a qualita-
tively similar frequency response to axial flow perturbations –
800O640O
60O
720O
265
400
Figure 1. SCHEMATIC OF THE ANNULAR RIJKE TUBE WITH THE
BASIC DIMENSIONS
low-pass characteristics with an associated phase-lag. The axial
extent of the grids is 15 mm, thus clearly complying with the as-
sumption of a compact heat source. The heating grids are pow-
ered by three independent DC sources, each with a maximum
output of 1.5 kW. In the nominal set-up, 4 heating grids are con-
nected in parallel. It was taken special care that all heating grids
are identical so that the nominal system would be as symmetric
as possible. The relative difference in the cold electrical resis-
tance of the 12 heating grids was less than 0.5%. Unless men-
tioned otherwise, all heating grids are supplied with equal elec-
trical power. A photograph of one of the heating grids is shown
in Fig. 3.
No external source is used to drive a mean flow. As in the
original Rijke tube, the mean flow is solely convection induced.
This restricts the parameter space (power input and mean flow
velocity cannot be varied independently) but allows for essen-
tially noise-free measurements. Moreover, the acoustic bound-
ary conditions, which have a significant effect on stability and
microphonesheating grids
speaker
Figure 2. PHOTOGRAPH OF THE ANNULAR RIJKE TUBE
3 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
Figure 3. PHOTOGRAPH OF A HEATING GRID (TOP VIEW)
oscillation amplitudes [20], are well defined.
Microphones and speakers can be mounted to the annular
Rijke tube via 48 ports in total, 24 each at the tubes and at the
annular duct. In this study, only 12 microphones, mounted in the
tubes, upstream of the heating grids, were used. For the appli-
cation of feedback control, three speakers were attached to the
tubes downstream of the heat sources (see Fig. 2).
LOW-ORDER SYSTEM MODEL
In this section, an analytical system model for the annular
Rijke tube is presented. The model is based on a frequency do-
main representation of the acoustic field coupled to the heat re-
lease transfer functions of the unsteady heat sources. Although
the heat sources considered in this work are electrically driven
heating grids, the principal methodology can be applied to annu-
lar combustors as well.
The basic strategy in setting up the system model will be
to represent the acoustic response up- and downstream of the
elements of heat release separately and then combine the two
by making use of the appropriate coupling relations across the
heat sources. Figure 4 displays the notation that will be used
for the acoustic pressures and velocities immediately up- and
downstream of the heat sources. pu/d.n and uu/d.n denote acoustic
pressure and axial particle velocity up-/downstream of the heat
source in duct n. The ducts are counted in anti-clockwise di-
rection when looking downstream. The acoustic pressure is as-
sumed to be scaled by the (local) characteristic impedance ρc
(the product of mean density and speed of sound). We assume
that the axial extent of the elements of heat release as well as the
diameters of the ducts are much shorter than the relevant acoustic
wavelengths. In this case, we can consider the zone of heat re-
lease as compact, and the acoustic field in the ducts can be treated
as one-dimensional.
The model is completely linear. Accordingly, we only intend
to identify unstable modes and do not attempt to quantify the
limit cycling amplitude. For the latter, detailed knowledge on the
nonlinear response of the heat sources would be required [12, 21]
which was, however, not available.
xxx x(p, u)u.2
(p, u)d.2
(p, u)u.1
(p, u)d.1
(p, u)u.N
(p, u)u.N
flow
direction
Figure 4. NOTATION FOR ACOUSTIC PRESSURES AND VELOCI-
TIES UP- AND DOWNSTREAM OF THE HEAT SOURCE
Up- and downstream acoustic response
To model the acoustics up- and downstream of the heat
sources, we first consider each part separately. Since we as-
sume plane wave propagation in the tubes, the whole action of
the acoustic field downstream of all heat sources can be repre-
sented as a generalized impedance in the sense that it maps the
velocity fluctuations downstream of the heat sources to the pres-
sures, viz.
pd = Z ud, (1)
where Z is an impedance matrix such that element (i, j) is the
pressure response downstream of the heat source in duct i to a
unit excitation in acoustic velocity in duct j while all other acous-
tic velocities are set to zero. Hence, the elements of Z are given
by Zi j = pd.i/ud. j, with ud.k = 0 for all k , j.
Although the acoustic fields in the tubes upstream of the heat
sources are not directly connected, we construct the model to al-
low for such a coupling, too (in the presence of a plenum, for in-
stance). Analogous to the downstream part, we define an admit-
tance matrix A, which relates the acoustic velocities upstream of
the heat sources to the pressures by
uu =A pu. (2)
Now, as mentioned above, in the present case considered, the
upstream ends of the tubes are not connected to each other so
that A is a diagonal matrix. (This holds true only approximately,
see below.) Moreover, since the tubes are all of the same length,
the diagonal entries are all identical and simply correspond to
the admittance of a duct with an unflanged open end. Hence, we
have
Ai j = −1 − i ZLS tan kLu
ZLS − i tan kLu
δi j, (3)
where k = ω/c is the wave number (ω and c denoting angular
frequency and speed of sound, respectively), δi j represents the
Kronecker delta, and ZLS = (kr)2/4+ i 0.6133 kr is the long-wave
4 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
Levine–Schwinger open end impedance [22]; r denotes the tube
radius and Lu the length upstream of the heat source. In principal,
Lu could be different for each tube, which corresponds to the case
where the heating grids are at different axial locations. Then A
would still be diagonal but not isotropic. Modeling the upstream
admittance matrix by Eq. (3) is only an approximation for two
reasons. First, the upstream ends of the ducts are not ideally un-
flanged due to a supporting ring (see Fig. 2). Second, since there
is some sound radiation out of the tube inlets, A is not strictly
diagonal. Both effects can, however, be assumed negligible in
the case of small tube Helmholtz numbers.
The acoustic response matrix of an annular duct.
The impedance matrix of the downstream part, Z, can be ob-
tained by making use of an eigenfunction expansion of the
Greens function of the Helmholtz equation. Since the eigenfunc-
tions for an annular duct are known analytically, this method is
flexible and fast. The approach is described in detail in Refs. [23]
and [24]; we only briefly reproduce the essentials below.
In terms of the eigenfunctions of the Helmholtz equation,
the pressure response to a velocity excitation on a surface S at
the boundary can be written as
p(x) = −ik∑
α
ψα(x)
k2 − iζα − k2α
∫
S
ψ∗α(x0)uν(x0)dA0, (4)
where α = {l,m, n} is a multiindex ordered such that kα ≤ kα+1, l,
m, and n are longitudinal, azimuthal, and radial mode indices, re-
spectively, and ψα is the eigenfunction associated with the eigen-
value kα. ζα denotes a modal damping coefficient. With a con-
stant forcing amplitude at the inlet section on an area correspond-
ing to one of the tubes, the impedance matrix can be computed
from (4) if the series is truncated at sufficiently large values of α.
Since the annular Rijke tube has a hub–tip ratio close to one
(as is the case for most gas turbine combustion chambers), we do
not include any radial modes when computing (4). We simply
use the mean radius for the eigenvalues, which is a good approx-
imation for hub–tip ratios close to one [25].
Prolongation of the annular duct response along
an array of straight tubes. The acoustic response matrix
Z needs to be known at the reference locations immediately
downstream of the heat source. The eigenfunction expansion
presented above, however, only allows to calculate Z directly
at the annular duct. A transformation of the reference locations
along an array of straight tubes (which do not necessarily need to
have identical lengths), such as in Fig. 4, can be achieved in the
following way.
Consider the impedance matrix of an annular duct (no tubes
attached). We denote the up- and downstream traveling plane
wave components at coupling area i as gi and fi. Then the ele-
ments Ri j of the reflection coefficient matrix R can be defined
as Ri j = gi/ f j, with fk = 0 if k , j. Since, for plane waves,
we have pi = fi + gi and ui = fi − gi, the impedance matrix re-
lates incident and reflected plane wave components according to
f + g = Z( f − g). Hence, the reflection coefficient matrix can be
computed from the impedance matrix via
R = (Z + I)−1(Z − I), (5)
where I is the N×N identity matrix. This is a generalization of
the case of a plane wave in a single tube.
Denoting the length of the tubes as ∆xi, the up- and down-
stream traveling waves at the heat source locations, gq.i and fq.i,
say, can be related to those at the inlet of the annular duct as
gq.i = e−ik∆xi gi and fq.i = eik∆xi fi. Accordingly, we have for
the reflection coefficient matrix with reference planes at the heat
sources
Rq.i j = e−ik(∆xi+∆x j) Ri j. (6)
Finally, the impedance matrix at the desired reference location,
Zq, can be calculated using the equivalent of (5)
Zq = (I −Rq)−1(I +Rq). (7)
With (6) and (7), Eq. (5) can be reformulated to directly relate
the impedance matrix at the heat sources to that at the inlet of the
annular duct, viz.
Zq =[
(Z + I)D−1 − (Z − I)D]−1 [
(Z + I)D−1 + (Z − I)D]
.
(8)
In (8), the matrix D has elements Di j = e−ik∆xi δi j.
Heat release dynamics
In each tube, the relations between the up- and downstream
acoustic variables are determined by the dynamic response of
the heat source (in that duct). In a general linear framework, this
relation can be expressed by a 2×2 transfer matrix in frequency
domain
[
pd.n
ud.n
]
= Tn
[
pu.n
uu.n
]
. (9)
For the type of compact heat source we consider here, and for
a vanishing mean flow Mach number, it is reasonable to assume
5 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
that the pressure loss is negligible. Furthermore, the heat source
induces a jump in acoustic velocity via [26]
A(ud.n − uu.n) =γ − 1
γP0
qn, (10)
where qn and γ denote the heat release rate and the ratio of spe-
cific heats, respectively, A is the cross-sectional area, and P0 rep-
resents the mean pressure.
We introduce the heat release transfer function as the nor-
malized response of heat release fluctuations to normalized per-
turbations in the upstream velocity
Fn =qn
uu.n
uu.n
qn
. (11)
Then the transfer matrix of the heat source in duct n can be writ-
ten as
Tn =
[
ξn 0
0 1 +(
Td.n
Tu.n− 1)
Fn
]
, (12)
where ξn =√
Td.n/Tu.n is the ratio of characteristic impedances
up- and downstream of heating element n. As proposed by
Lighthill [27], we model the heat release transfer function as a
first order low-pass
Fn =ck.n
1 + iωcτ.n, (13)
with gain and phase parameters ck.n and cτ.n. Note that we ex-
plicitly allow for different temperature jumps and heat release
transfer functions in each tube. For later use, we define diag-
onal matrices X and F with elements Xi j = ξiδi j and Fi j =
(Td.i/Tu.i − 1)Fiδi j.
Submodel coupling and dispersion relationTo couple the acoustic models of the up- and downstream
parts, we use the dynamic heat release characteristics given by
Eq. (12). Hence, we have pd = X pu, i.e., the pressure is contin-
uous across the heat source, and the jump in acoustic velocity is
given by
ud =(
I +F)
uu. (14)
Combining Eqs. (1), (2), (14), and the pressure continuity con-
dition, and dropping the subscript u of the vector of pressures
upstream of the heat source, we obtain
X p = Z(
I +F)
A p. (15)
From Eq. (15), it follows that for non-trivial solutions to exist,
the dispersion relation
det S = 0 (16)
must be satisfied, where the system matrix S is given by
S = Z(
I +F)
A − X. (17)
Solutions ωk of the dispersion relation are the system eigenfre-
quencies, Re(ωk) representing the (angular) oscillation frequency
and −Im(ωk) the growth rate. The purely acoustic eigenfrequen-
cies can be obtained for the case with no dynamic heat release as
the solutions of |ZA − X| = 0.
It is interesting to note here that Eq. (16) with S given in
(17) is again a generalization of the simple 1D case of a sin-
gle Rijke tube. In this case, the impedance and admittance are
scalar functions which relate acoustic velocity to pressure and
vice versa. The dispersion relation for this configuration can be
written as Z(1+F )A−ξ = 0 [24], where A and Z are admittance
and impedance immediately up- and downstream of the source of
heat release.
If an eigenvalue ωk of (16) has been obtained, the corre-
sponding pressure pattern in the N tubes, at the heat source po-
sitions, can be computed as the null space of the system matrix
evaluated at ωk, hence
pk = ker S(ωk). (18)
Note here that for a distinct eigenvalue, we have dim ker S(ωk) =
1, whereas for a degenerate one, such as in the case of a fully
symmetric configuration, where two mode shapes share one char-
acteristic value, the eigenspace is two-dimensional.
RESULTS AND DISCUSSION
Identifying the modal structure of the pressure fieldTo assess the azimuthal structure of the experimental and the
model data, we determine the coefficients am,c|s of the m-th order
cosine and sine modes by means of a discrete projection of the N
pressures on an angular Fourier basis hm,c|s(φ), viz.
am,c|s =
N∑
n=1
pnhm,c|s(φn), (19)
where the φn correspond to the circumferential locations of the
tube center axes. In the case of even N, we use
hm,c = bm,c cos(mφ), m = 0 . . .N/2, (20a)
hm,s = bm,s sin(mφ), m = 1 . . . (N/2 − 1), (20b)
6 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
0 1 1 2 2 3 3 4 4 5 56 0 1 1 2 2 3 3 4 4 01 1 2 2 3 3 5 5 4 4 6200
300
400
500
600
700
800
900
1000
azimuthal mode order m
freq
uen
cyin
Hz
low-order model
FEM computation
Figure 5. ACOUSTIC EIGENFREQUENCIES AND CORRESPONDING
AZIMUTHAL MODE ORDER m; MODEL AND FEM RESULTS ARE
COMPARED.
were the coefficients bm,c and bm,s are chosen such that the basis
has unit norm in the discretized N-space. Note here that, with this
normalization, the amplitudes of the modal coefficients are larger
than the actual pressure amplitudes by about a factor of√
N. The
azimuthal mode order of the eigenfunctions computed from the
model (Eq. (18)) can be determined in the same way. Using pk in
(19) yields the Fourier transformed modal coefficients am,c|s (the
modal amplitudes) of the k-th eigenfunction.
Eigenfrequencies without dynamic heat releaseThe eigenfrequencies of the annular Rijke tube, computed
with the model presented in the previous section, for a temper-
ature jump of 50 K across the heat source, are shown in Fig. 5.
No transfer function for the dynamic heat release was included
here. To assess the accuracy of the low-order model, the results
obtained with a commercial finite element code are also plot-
ted. The latter computes the eigenfrequencies and corresponding
eigenfunctions from the Helmholtz equation
∇·(
c2∇p)
+ ω2 p = 0. (21)
In the finite element model, we used pressure node boundary
conditions at the inlets of the tubes and at the downstream end
of the annular duct. To account for the end correction of the
open tubes, we extended the part upstream of the heat source ac-
cordingly. Radiation from the open end as well as from the outlet
of the annular duct was not included.
Since the boundary conditions at the up- and downstream
end of our configuration are open, all modes have at least a half-
wave structure in axial direction. Moreover, only the axisymmet-
ric mode and those with azimuthal orders being an integer mul-
tiple of 6 are not degenerate, since we have a 12-fold rotational
Figure 6. PRESSURE PATTERNS CORRESPONDING TO THE
ACOUSTIC MODES OF THE ANNULAR RIJKE TUBE AT EIGENFRE-
QUENCIES 541, 554, AND 557 Hz (FROM LEFT TO RIGHT)
symmetry (with additional mirror symmetries). The fact that cer-
tain azimuthal modes are distinct is a direct result of that symme-
try structure [28, 29]. By deliberately introducing asymmetries,
we can also split modes which are degenerate in the nominal,
symmetric case. This is demonstrated in a later section.
We also note that the modal density between 500 and 600 Hz
is particularly high (Fig. 5). Azimuthal modes of order 4, 5, and
6 in this frequency range were found to be unstable in the exper-
iment (see below). The pressure patterns in the annular duct cor-
responding to these modes are shown in Fig. 6 (computed with
the Helmholtz solver). It should be noted, however, that the an-
gular location of the pressure nodes and antinodes is arbitrary for
azimuthal orders 4 and 5, since these modes are degenerate.
Linear stability
The linear stability characteristics for the experimental sys-
tem were computed with the low-order model including the heat
release transfer function (13). The complex eigenvalues, deter-
mined from Eq. (16), are shown in Fig. 7. Negative imaginary
parts correspond to instability. All eigenvalues with frequencies
larger than 600 Hz were strongly damped and are thus not shown.
The dominant feature is a group of eigenvalues around 550 Hz
with zero or small negative imaginary part. These eigenvalues
correspond to azimuthal mode orders m = 4, 5, and 6 (with in-
creasing frequency) and can be expected to be essential in the
experiment.
Instabilities in the experiment
Consistent with the linear stability analysis based on the
low-order model, azimuthal modes of orders 5 and 6 were ob-
served in the experiment for a certain range of input powers.
The modal amplitudes were computed from the 12 pressure sig-
nals using Eqs. (19) and (20). Figure 8 shows the average
amplitude of the modal coefficients for a variation in the in-
put power. For azimuthal orders 1–5, the effective amplitude
(a2m,c+a2
m,s)1/2 is plotted. All heating grids were driven with iden-
tical power. No instability is observed for input powers smaller
than 1500 W. With increasing heating power, the first 6th order
azimuthal mode becomes unstable. The amplitude grows with
7 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
200 300 400 500 600−0.5
0
0.5
1
1.5
2
2.5
3
3.5
Re (ωk)/(2π) in Hz
Im(ω
k)/
(2π
)in
s−1
0
1
2
3
4
5 6
0 1
2
Figure 7. EIGENVALUES OF THE ANNULAR RIJKE TUBE COM-
PUTED WITH THE LOW-ORDER MODEL. NUMBERS INDICATE AZI-
MUTHAL MODE ORDER.
an increase in the electrical power up to about 1700 W. However,
the first 5th order azimuthal mode also starts to grow and dom-
inates for higher power values. At an input power of 2000 W,
the m = 6 mode is almost completely suppressed. For larger
powers, however, this mode remains significant, although the 5th
order mode is clearly dominant. The strong suppression of the
m = 6 mode around 2000 W was a reproducible feature. Also
note that the power dependence of the amplitude of the 5th order
azimuthal mode clearly corresponds to a supercritical bifurca-
tion. The modes with azimuthal order lower than 5 only have a
small contribution to the measured pressure signals. When the
heating grid power is switched on, the m = 6 mode is actually
growing much faster and is limited and then reduced in ampli-
tude through the 5th order azimuthal mode (see also Fig. 17 in
the section on feedback control).
For exemplary purposes, the amplitude spectrum of one of
1000 1500 2000 2500 3000 3500 4000 45000
250
500
750
1000
1250
1500
1750
2000
total power in W
mo
dal
amp
litu
de
inP
a
m = 0 . . .4
m = 5
m = 6
Figure 8. MEASURED MODAL AMPLITUDES FOR VARYING TOTAL
POWER INPUT
0 250 500 750 1000 1250 1500 1750 200040
60
80
100
120
140
frequency in Hz
spec
tral
amp
litu
de
ind
B
Figure 9. PRESSURE AMPLITUDE SPECTRUM FOR A TOTAL INPUT
POWER OF 3200 W
the pressure signals is shown in Fig. 9. The data corresponds to a
total input power of 3200 W. A strong peak, extending more than
four orders of magnitude above the background noise, is visible
around 560 Hz. At this scale, the frequencies corresponding to
the 5th and 6th and the stable 4th order azimuthal modes cannot
be distinguished. A few harmonics are also present.
A more detailed view of the dominant frequency content of
the pressure oscillations is given in Fig. 10. The data corre-
sponds to total input powers of 1700 and 2100 W (cf. Fig. 8).
A peak with small magnitude can be observed between 520 and
530 Hz. This corresponds to the m = 4 mode that was found to
be marginally stable in the low-order model. Two stronger fre-
quency components are found around 540 and 550 Hz. The lo-
cation of the peaks slightly increases with input power due to the
larger speed of sound downstream of the heat source. The spac-
ing of the two dominant peaks is only 3–4 Hz, which also com-
pares well with the model results. For a power input of 1700 W,
the peak with the larger frequency is about an order of magnitude
stronger. On comparison with Fig. 8, we find this spectral com-
ponent to be associated with the m = 6 mode. At 2100 W, the 5th
order azimuthal mode clearly dominates.
The simultaneous oscillation of two instability modes is not
often observed in thermoacoustic instability investigations, the
common explanation being that the oscillation of the “more un-
stable” mode suppresses the growth of the weaker one. In our
case, the presence of two unstable modes with significant ampli-
tudes clearly is related to the close proximity of their associated
oscillation frequencies (see also Fig. 10). A closer inspection of
the temporal evolution of the modal amplitudes showed, how-
ever, that the weaker m = 6 mode did not have a constant am-
plitude. It was growing and decaying, seemingly randomly, on
a much larger time scale than the actual oscillation. A growth
in amplitude was always associated with a decay of the m = 5
mode. Thus, there was a continuous competition between the
8 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
500 520 540 560 580 60040
60
80
100
120
140
frequency in Hz
spec
tral
amp
litu
de
ind
B
P = 1700 W
P = 2100 W
Figure 10. PRESSURE AMPLITUDE SPECTRA FOR INPUT POWERS
OF 1700 AND 2100 W WITH THE m = 6 AND m = 5 MODE BEING
DOMINANT, RESPECTIVELY
two modes rather than two coexisting oscillations with constant
amplitudes.
The azimuthal instability modes observed in annular com-
bustion chambers can be either standing or rotating in angular
direction or be a mixture of both [4, 14, 15]. In the present case,
this is more difficult to asses due to the two coexisting oscillation
modes. However, a careful investigation of the time traces of the
12 pressure sensors showed that the m = 6 mode was always of
standing type. This was expected, since this mode is not degen-
erate and, therefore, rotating wave solutions do not exist in the
linear regime. In Ref. [14], it was shown that the saturation in
the heat release response at high amplitudes favors rotating wave
solutions. We conjecture, however, that this does not apply to
azimuthal modes which are not degenerate.
The 5th order azimuthal mode is degenerate so that spin-
ning and standing oscillation patterns are possible. No informa-
tion can be gained from the stability analysis, since all linear
combinations of the two basis functions spanning the degener-
ate eigenspace are equivalent. The postprocessing of the exper-
imental data showed that the m = 5 mode was not of definite
standing or rotating type. In terms of the cos(mφ), sin(mφ) basis
(Eq. (20)), azimuthally standing components are associated with
in-phase modal coefficients am,c and am,s. Conversely, a purely
rotating solution has modal coefficients which are in quadrature
but have identical magnitudes [8]. Thus, the nature of the 5th or-
der azimuthal mode can best be visualized by means of a phase
plane with the coefficients of the cosine and sine modes (am,c and
am,s) as coordinates. Then a circumferentially standing wave is
represented by a straight line whose slope determines the angular
location of the pressure nodes, and a purely spinning wave traces
out a circle.
Figure 11 shows four exemplary results, which correspond
to the same operating conditions (total power of 4500 W) but dif-
−2
−1
0
1
2
a5,s
(t)
ink
Pa
(a) (b)
−2 −1 0 1 2
−2
−1
0
1
2
a5,c(t) in kPa
a5,s
(t)
ink
Pa
(c)
−2 −1 0 1 2
a5,c(t) in kPa
(d)
Figure 11. PHASE PLANE TRAJECTORIES IN a5,c–a5,s SPACE. COL-
ORS FROM BLUE TO RED VIA WHITE INDICATE INCREASING TIME.
THE LENGTH OF EACH TRAJECTORY CORRESPONDS TO A TIME
INTERVAL OF 49 MILLISECONDS, AND THE DATA IS SAMPLED AT
213 HZ. TOTAL ELECTRIC POWER 4500 W.
ferent periods of time. Apparently, the 5th order azimuthal mode
is neither distinctly spinning nor standing. In fact, it does not
exhibit any stationary characteristic at all. Figure 11 (b), for in-
stance, clearly corresponds to a circumferentially standing wave,
whereas Fig. 11 (c) represents an essentially rotating mode. Tran-
sitions between different oscillation patterns can be observed in
frames (a) and (d). Qualitatively similar characteristics were also
present at other input powers, for which the m = 5 mode was un-
stable.
The effect of an asymmetric power distribution on ther-moacoustic stability
An azimuthal variation of burner and flame properties in
an annular combustion chamber, either through geometrical
changes [5, 6] or burner group staging [30], can have a stabi-
lizing influence on the pressure oscillations. We mimic this ef-
fect in the annular Rijke tube through an azimuthal variation of
the power supplied to each heating grid. Changing the geome-
try of a burner or the equivalence ratio with which it is supplied
typically changes the flame response. An analogous effect is ob-
tained when varying the power of a heating grid. In general, such
a variation has an impact on the gain as well as on the phase of
the heat release transfer function. For the heating grid, a larger
power leads to an increase in the downstream temperature and in
9 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
540 550 560 570 580 590 600−1
0
1
2
3
C12 C4 C3
Re (ωk)/(2π) in Hz
Im(ω
k)/
(2π
)in
s−1
Figure 12. VISUALIZATION OF THE NOMINAL POWER DISTRI-
BUTION AND THE TWO STAGING PATTERNS WITH ASSOCIATED
EIGENVALUES; DARKER CIRCLES INDICATE HIGHER ELECTRIC
POWER. THE EIGENVALUES OF THE C4 AND THE C3 CASE COR-
RESPOND TO σ = 1 AND σ = 1.5, RESPECTIVELY.
the mean velocity (because it is convection induced). Since the
heat release response of the grid is essentially a function of the
Strouhal number, the effect of a change in the supplied power can
be estimated based on measured temperatures and mean veloci-
ties for different heating powers.
An azimuthal variation of the heat release transfer functions
does not necessarily have a positive effect on thermoacoustic sta-
bility. Nominally degenerate azimuthal modes can be split if the
system’s symmetry group is reduced in a particular way through
the introduced circumferential variation [28]. To illustrate this
effect, we investigate two patterns of azimuthal “staging”: one
has a 4-fold rotational symmetry (denoted by C4) with reduced
power at every third heating grid, and the other one has a 3-fold
rotational symmetry (denoted by C3) with groups of two heating
grids alternating in lower and higher electric power. The stag-
ing patterns are visualized in Fig. 12. In all results shown in this
section, the total electric power is held constant at 2900 W. To
quantify the strength of the circumferential modulation, we de-
fine the asymmetry parameter σ as
σ =∆P
Pmean
, (22)
where ∆P is the difference in electric power of the two types of
heating grids in one staging pattern, and Pmean is the mean power
(which is held constant).
We discuss the model results for the two staging patterns
first. The system eigenvalues for the reduced symmetry groups
C4 and C3 and the corresponding staging patterns are shown in
Fig. 12. For comparison, the eigenvalues for the original fully
symmetric case (C12) are also plotted. Only the frequency range
between 540 and 600 Hz is shown in Fig. 12 for clearness. Con-
sider the effect of the C4 staging pattern first. Here, σ = 1 has
been used for the computations. The m = 4 mode, degenerate
in the baseline case with a frequency of 546 Hz and marginally
stable, splits into two distinct modes as a result of the reduction
in symmetry. As it turns out, this is an essentially destabilizing
effect, because one of the split modes has a larger imaginary part,
whereas the other moves down into the unstable half plane. The
modes of azimuthal order 5 and 6, dominant in the fully sym-
metric case, are slightly stabilized. The mode with a frequency
of 593 Hz in the baseline case also splits, because it is of 2nd
azimuthal order. However, both m = 2 split modes remain sta-
ble. The removal of the degeneracy of the 4th and 2nd order
azimuthal modes is consistent with the splitting rule stated by
Perrin [28]. Thus, based on the linear analysis, we summarize
the effect of the C4 staging pattern as a slight stabilization of the
nominally unstable m = 5 and m = 6 modes at the cost of a
significant destabilization of one of the m = 4 split modes.
The results of the C3 staging pattern are univocal; all modes
are either stabilized or affected only insignificantly (also those
not shown in Fig. 12). However, the damping of the 5th or-
der azimuthal mode is only slightly increased so that it remains
marginally stable. It is important to note here that no destabi-
lizing splitting of the 6th order azimuthal mode occurs, because
this mode is already distinct in the baseline case due to the 12-
fold rotational symmetry.
To assess the effect of the reduced circumferential symme-
try in the experiment, measurements were performed with a con-
stant total power of 2900 W while varying the distribution to the
12 heating grids according to the C3 and C4 staging patterns.
The measured modal amplitudes as a function of the asymme-
try parameter σ are shown in Figs. 13 and 15. At σ = 0, only
the 5th and 6th order azimuthal modes are oscillating, with the
former dominating. This corresponds to the case already con-
sidered before. Increasing the asymmetry according to the C4
pattern (Fig. 13) leads to a continuous decrease in the amplitudes
of the 5th and 6th order azimuthal modes. At σ = 0.75, both of
these modes are stabilized. However, at the same time, the m = 4
mode is destabilized and grows in amplitude with a further in-
crease of σ. These results correspond well with the eigenvalue
analysis for σ = 1 (Fig. 12).
Since, for the C4 staging pattern, the m = 4 modes are not
degenerate any more, they have a distinct pressure pattern along
the circumference. Hence, we can compare the modeled circum-
ferential pressure distribution (Eq. 18) corresponding to the un-
stable m = 4 mode with the measured one. This comparison is
presented in Fig. 14. For completeness, the pressure pattern of
the stable split mode is also shown. The measured and modeled
pressure patterns both correspond to σ = 1. The experimen-
10 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
0 0.25 0.5 0.75 1 1.25 1.50
250
500
750
1000
1250
1500
1750
2000
asymmetry σ
mo
dal
amp
litu
de
inP
a
m = 0 . . .3
m = 4
m = 5
m = 6
Figure 13. MEASURED MODAL AMPLITUDES AS A FUNCTION OF
THE ASYMMETRY PARAMETER σ; TOTAL POWER INPUT 2900 W.
tal observations confirm the model results. Clearly, the selected
pressure pattern identified in the measured data corresponds to
the computed unstable mode. The reason for this pattern be-
ing more unstable is that the pressure nodes are arranged such
that they coincide with the locations of the heating grids with
reduced power. The measured pressure values represent an arbi-
trary snapshot. Note, however, that the amplitude of the modeled
pressure distribution is undetermined, because it is an eigenfunc-
tion. In Fig. 14, the amplitude of the computed pressure pattern
has been scaled such that it matches the experimentally deter-
mined magnitudes.
An increase in σ according to the C3 pattern (Fig. 15) results
in a decrease in the amplitude of the 5th order azimuthal mode.
In contrast, the m = 6 mode is not damped by the asymmetry. In
fact, it increases in amplitude, quite significantly aroundσ ≈ 0.6,
0 2 4 6 8 10 12−500
−250
0
250
500
tube number n
pin
Pa
pexp Re ( pmod,u) Re ( pmod,s)
Figure 14. MODELED AND MEASURED PRESSURE DISTRIBU-
TIONS IN THE TUBES CORRESPONDING TO THE 4TH ORDER AZI-
MUTHAL MODE FOR σ = 1. HEATING GRIDS IN TUBES 2, 5, 8, AND
11 HAVE A REDUCED POWER INPUT. THE STABLE (pmod,u) AND THE
UNSTABLE (pmod,s) m = 4 SPLIT MODES ARE SHOWN.
0 0.25 0.5 0.75 1 1.25 1.50
250
500
750
1000
1250
1500
1750
2000
asymmetry σ
mo
dal
amp
litu
de
inP
a
m = 0 . . .4
m = 5
m = 6
Figure 15. MEASURED MODAL AMPLITUDES AS A FUNCTION OF
THE ASYMMETRY PARAMETER σ; TOTAL POWER INPUT 2900 W.
and settles on a constant level for σ = 0.7–1. Increasing the
asymmetry parameter further, both modes are eventually stabi-
lized at σ = 1.3. The fact that both modes become stable at the
same value of the asymmetry parameter indicates significant in-
teraction even at low amplitudes. The eigenvalue analysis for this
staging pattern with σ = 1.5 (Fig. 12) showed the m = 6 mode to
be stabilized and the m = 5 mode to be marginally stable. This
does not fully correspond to the experimental data but points in
the same direction. (See the discussion on model accuracy at the
end of the results section.)
Feedback control
In this section, we show that, in addition to studying self-
excited instabilities, the annular Rijke tube can also be used to
test active control schemes. Devising an elaborate model-based
control algorithm, such as, e.g., in Refs. [10, 23], is beyond the
scope of the present paper. Instead we present the effect of sim-
ple proportional control on the unstable modes. We used three
speakers, mounted to the tubes at the pressure ports immediately
downstream of the heat sources (see Figs. 1 and 2) at equidistant
azimuthal angles.
To account for the effect of control in the model, we add a
linear feedback term in the jump conditions (14), viz.
ud =(
I +F)
uu +K pu, (23)
where K is a constant N×N gain matrix that basically maps the
input to the output locations. Equation (23) models the loud-
speaker action as an additional volume source at the heating grid
location. This does not represent the experimental feedback con-
figuration exactly, since the actuators are located 65 millimeters
downstream of the heat source. Also, the feedback signal is not
11 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
the pressure directly at the heat source but 45 millimeters up-
stream. On the other hand, the pressure field does not change
significantly over these short distances, and, in addition to that,
we are interested in the resonance frequencies only, at which the
complete pressure field in one tube is essentially in phase. There-
fore, we can consider (23) as a reasonably accurate representa-
tion of the feedback configuration in the experiment.
We will consider two types of proportional feedback: (i) a
single sensor signal is fed back to all three actuators and (ii) the
actuators feed back the pressure signal from their respective tube.
For case (i), the gain matrix takes the form
K(i) = [c 0], (24)
where cT = [κ 0 0 0 κ 0 0 0 κ 0 0 0] and κ is a negative constant.
In the experiment, κ has to be tuned to compensate for input and
output amplifiers and the speaker transfer function. For multiple
inputs (case (ii)), the gain matrix K(ii) is diagonal with
K(ii)nn =
κ, n ∈ {1, 5, 9},0, else.
(25)
The effect of these types of control on linear stability can
be assessed by computing the system eigenvalues based on the
modified jump conditions (23). In this case, the system matrix
accounting for the effect of feedback control takes the form
S = Z(
I +F)
A − X +ZK(i,ii). (26)
The influence of control on the stability of the dominant
modes is shown in Fig. 16. The results from type (i) and type
(ii) control are compared to the uncontrolled case; a feedback
gain κ = −0.03 has been used for the computations. In case of
type (i) control, when only a single pressure signal is fed back to
all three actuators, only the 6th order azimuthal mode at 564 Hz
is stabilized. (The damping rate of the stable m = 0 mode at
572 Hz is also increased.) In contrast, type (ii) control stabilizes
all modes; the damping rates of the m = 6 and m = 0 modes co-
incide with those for type (i) control. This is because for axisym-
metric and 6th order azimuthal modes, it makes no difference if
only one signal is fed back; the pressure in all of the actuator
tubes is identical for these modes. However, since the m = 5
mode also shows strong oscillations in the uncontrolled case, we
expect type (ii) control to be superior in suppressing the instabil-
ity.
Results from the experimental application of the two types
of proportional control are presented in Fig. 17. The amplitudes
of the 5th and the 6th order azimuthal modes are plotted versus
time. Here, the temporal evolution of the modal amplitudes was
520 540 560 580 600−1
0
1
2
3
Re (ωk)/(2π) in Hz
Im(ω
k)/
(2π
)in
s−1
no control
type (i) control
type (ii) control
Figure 16. THE EFFECT OF TWO TYPES OF CONTROL ON THE
SYSTEM EIGENVALUES
computed as the magnitudes of the filtered analytic signals of the
modal coefficients am. A few seconds after the power is turned
on, the amplitudes of the two unstable modes start to grow ex-
ponentially, the m = 6 mode increasing distinctly faster. As
soon as the m = 5 mode has grown to a significant amplitude,
the 6th order azimuthal mode actually decreases, and both settle
on approximately constant levels. The strong overshoot of a6 is
obviously a result of the nonlinear interaction of the two unsta-
ble modes, probably associated with the close proximity of their
oscillation frequencies. A detailed investigation of this nonlin-
ear interaction is, however, beyond the scope of the present pa-
per. Type (i) control is activated from t = 20 s to approximately
t = 30 s. Only a small effect can be observed. There is no signif-
icant reduction of the oscillations, but the amplitude a6 seems to
settle on a somewhat smaller level and shows less variations. The
linear analysis based on the model predicted a stabilization for
the 6th order azimuthal mode using this type of control (Fig. 16).
On the other hand, since the m = 5 mode is still oscillating with
0 5 10 15 20 25 30 35 40 45 500
500
1000
1500
2000
2500
time in s
mo
dal
amp
litu
de
inP
a
m = 5 m = 6
poweron
type (i)control on
type (i)control off
type (ii)control on
Figure 17. TEMPORAL EVOLUTION OF MEASURED MODAL AM-
PLITUDES WITH THE EFFECT OF FEEDBACK CONTROL; TOTAL
POWER INPUT 3200 W.
12 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
finite amplitude, the linear analysis may not be applicable. When
activating type (ii) control (based on three pressure sensors) at
approximately t = 40 s, the oscillation diminishes immediately.
This is fully consistent with the results from the linear analysis
(Fig. 16) where all modes in the unstable frequency range were
predicted to be stabilized by the control action.
The amplitude spectra of one of the measured pressure sig-
nals (Fig. 18) shows that type (ii) control completely stabilizes
the system. In the uncontrolled case, the spectral peak corre-
sponding to the instability has a magnitude of almost 140 dB.
Applying type (ii) control, this component is reduced to below
60 dB, resulting in a peak reduction of four orders of magnitude.
The pressure signal used for the spectra in Fig. 18 does not cor-
respond to one of the tubes with actuator.
0 250 500 750 1000 1250 1500 1750 200040
60
80
100
120
140
frequency in Hz
spec
tral
amp
litu
de
ind
B control off
type (ii) control
Figure 18. PRESSURE AMPLITUDE SPECTRA FOR THE BASELINE
AND THE CONTROLLED CASE. TOTAL POWER INPUT 3200 W.
Discussion on model accuracy
The low-order model used in the present work provided
good results that were in accordance with the experimental data.
Unstable modes could be identified, and the effects of asymme-
tries in the circumferential heat release distribution and propor-
tional feedback were fairly well predicted. On the other hand,
we do not claim that the model can be considered quantitatively
correct in every aspect. An exact prediction of the growth rates,
e.g., and, along the same lines, an accurate determination of sta-
bility boundaries, is likely beyond its capabilities. Such quan-
titative predictions are inherently difficult, even with a model
built exclusively from experimental data [31]. One potentially
important aspect was not included in the low-order model. We
assumed a constant temperature downstream of the heat source
for simplicity. Pointwise temperature measurements in the ex-
perimental set-up showed that this was not true. Noticeable vari-
ations in axial as well as in angular direction could be found.
A non-uniform temperature distribution may well have an effect
on thermoacoustic mode stability [32, 33]. Assessing the effect
of an inhomogeneous temperature field in the annular duct on
the basis of the eigenfunction expansion approach used in the
present work is not straightforward, but two potential remedies
have been identified and are subject of ongoing work.
SUMMARY AND OUTLOOK
We presented an experimental thermoacoustic surrogate sys-
tem that facilitates an in-depth study of instabilities associated
with circumferential acoustic modes. For a range of input pow-
ers, we observed thermoacoustic oscillations corresponding to
the 5th and 6th azimuthal mode order. A low-order model repre-
senting the experimental arrangement was set up and gave results
fully consistent with the measured data. The effect of a symme-
try reduction through circumferential modulations of the heating
power was investigated. With increasing asymmetry, the 5th and
6th order azimuthal modes were stabilized, but, for one of the
staging patterns, a mode of 4th circumferential order became un-
stable at the same time. Also in this case, the model was capable
of predicting the effect of the introduced asymmetry.
The control approach presented in this work was, though
effective, rather elementary. A more advanced control scheme,
based on a time-domain model and utilizing a modal decom-
position of all 12 sensor signals in combination with feedback
through 6 speakers, has been successfully applied and is subject
of a forthcoming publication. Extension of the model and an
investigation of different damper configurations is the focus of
ongoing and future work.
ACKNOWLEDGMENT
Financial support from the German Research Foundation
through the Collaborative Research Center 557 is gratefully ac-
knowledged.
REFERENCES
[1] Lieuwen, T. C., and Yang, V., eds., 2005. Combustion In-
stabilities in Gas Turbine Engines, Vol. 210 of Progress in
Astronautics and Aeronautics. AIAA, Inc.
[2] Seume, J. R., Vortmeyer, N., Krause, W., Hermann, J.,
Hantschk, C.-C., Zangl, P., Gleis, S., Vortmeyer, D., and
Orthmann, A., 1998. “Application of active combustion in-
stability control to a heavy duty gas turbine”. Journal of
Engineering for Gas Turbines and Power, 120, pp. 721–
726.
[3] Krebs, W., Walz, G., Flohr, P., and Hoffmann, S., 2001.
“Modal analyis of annular combustor: Effect of burner
impedance”. ASME paper 2001-GT-0042.
[4] Krebs, W., Flohr, P., Prade, B., and Hoffmann, S., 2002.
“Thermoacoustic stability chart for high-intensity gas tur-
13 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms
bine combustion systems”. Combustion Science and Tech-
nology, 174(7), pp. 99–128.
[5] Berenbrink, P., and Hoffmann, S., 2000. “Supression
of dynamic combustion instabilities by passive and active
means”. ASME paper 2000-GT-0079.
[6] Kruger, U., Huren, J., Hoffmann, S., Krebs, W., Flohr,
P., and Bohn, D., 2001. “Prediction and measurement of
thermoacoustic improvements in gas turbines with annular
combustion systems”. Journal of Engineering for Gas Tur-
bines and Power, 123, pp. 557–566.
[7] Evesque, S., and Polifke, W., 2002. “Low-order acoustic
modelling for annular combustors: Validation and inclusion
of modal coupling”. ASME paper GT-2002-30064.
[8] Evesque, S., Polifke, W., and Pankiewitz, C., 2003. “Spin-
ning and azimuthally standing acoustic modes in annular
combustors”. AIAA paper 2003-3182.
[9] Kopitz, J., Huber, A., Sattelmayer, T., and Polifke, W.,
2005. “Thermoacoustic stability analysis of an annular
combustion chamber with acoustic low order modeling and
validation against experiment”. ASME paper GT2005-
68797.
[10] Morgans, A. S., and Stow, S. R., 2007. “Model-based
control of combustion instabilities in annular combustors”.
Combustion and Flame, 150, pp. 380–399.
[11] Stow, S. R., and Dowling, A. P., 2009. “A time-domain
network model for nonlinear thermoacoustic oscillations”.
Journal of Engineering for Gas Turbines and Power, 131,
pp. 031502–1 (9 pages).
[12] Stow, S., and Dowling, A. P., 2004. “Low-order modelling
of thermoacoustic limit cycles”. ASME paper GT2004-
54245.
[13] Stow, S. R., and Dowling, A. P., 2003. “Modelling of cir-
cumferential modal coupling due to Helmholtz resonators”.
ASME paper GT2003-38168.
[14] Schuermans, B., Paschereit, C. O., and Monkewitz, P.,
2006. “Non-linear combustion instabilities in annular gas-
turbine combustors”. AIAA paper 2006-0549.
[15] Staffelbach, G., Gicquel, L. Y. M., and Poinsot, T., 2009.
“Large eddy simlation of self-excited azimuthal modes in
annular combustors”. Proceedings of the Combustion Insti-
tute, 32(2), pp. 2909–2916.
[16] Sensiau, C., Nicoud, F., and Poinsot, T., 2008. “Computa-
tion of azimuthal combustion instabilities in an helicopter
combustion chamber”. AIAA paper 2008-2947.
[17] Nicoud, F., Benoit, L., Sensiau, C., and Poinsot, T., 2007.
“Acoustic modes in combustors with complex impedances
and multidimensional active flames”. AIAA Journal, 45(2),
pp. 426–441.
[18] Heckl, M. A., 1990. “Non-linear acoustic effects in the
Rijke tube”. Acustica, 72, pp. 63–71.
[19] Raun, R. L., Beckstead, M. W., Finlinson, J. C., and
Brooks, K. P., 1993. “A review of Rijke tubes, Rijke burners
and related devices”. Progress in Energy and Combustion
Science, 19(4), pp. 313–364.
[20] Bothien, M. R., Moeck, J. P., and Paschereit, C. O., 2008.
“Active control of the acoustic boundary conditions of com-
bustion test rigs”. Journal of Sound and Vibration, 318,
pp. 678–701.
[21] Noiray, N., Durox, D., Schuller, T., and Candel, S., 2008.
“A unified framework for nonlinear combustion instability
analysis based on the flame describing function”. Journal
of Fluid Mechanics, 615, pp. 139–167.
[22] Levine, H., and Schwinger, J., 1948. “On the radiation of
sound from an unflanged circular pipe”. Physical Review,
73(4), pp. 383–406.
[23] Schuermans, B., Bellucci, V., and Paschereit, C. O., 2003.
“Thermoacoustic modeling and control of multi burner
combustion systems”. ASME paper 2003-GT-38688.
[24] Schuermans, B., 2003. “Modeling and control of thermoa-
coustic instabilities”. PhD thesis, EPF Lausanne, Switzer-
land. http://library.epfl.ch/theses/?nr=2800.
[25] Stow, S. R., Dowling, A. P., and Hynes, T. P., 2002.
“Reflection of circumferential modes in a choked nozzle”.
Journal of Fluid Mechanics, 467, pp. 215–239.
[26] Lieuwen, T., 2003. “Modeling premixed combustion–
acoustic wave interactions: A review”. Journal of Propul-
sion and Power, 19(5), pp. 765–781.
[27] Lighthill, M. J., 1954. “The response of laminar skin fric-
tion and heat transfer to fluctutations in the stream veloc-
ity”. Proceedings of the Royal Society of London, A 224,
pp. 1–23.
[28] Perrin, R., 1977. “A group theoretical approach to warble in
ornamented bells”. Journal of Sound and Vibration, 52(3),
pp. 307–313.
[29] Hamermesh, M., 1989. Group Theory and its Application
to Physical Problems. Dover Pubs.
[30] Kokanovic, S., Guidati, G., Torchalla, S., and Schuermans,
B., 2006. “Active combustion control system for reduction
of NOx and pulsation levels in gas turbines”. ASME paper
GT2006-90895.
[31] Bothien, M. R., Moeck, J. P., and Paschereit, C. O., 2010.
“Comparison of linear stability analysis with experiments
by actively tuning the acoustic boundary conditions of a
premixed combustor”. Journal of Engineering for Gas Tur-
bines and Power, 132.
[32] Eisenhower, B., Hagen, G., Banaszuk, A., and Mezic, I.,
2008. “Passive control of limit cycle oscillations in a ther-
moacoustic system using asymmetry”. Journal of Applied
Mechanics, 75, pp. 011021–1 (9 pages).
[33] Matveev, K. I., and Culick, F. E. C., 2003. “A study of
transition to instability in a Rijke tube with axial temper-
ature gradient”. Journal of Sound and Vibration, 264(3),
pp. 689–706.
14 Copyright c© 2010 by ASME
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 05/07/2014 Terms of Use: http://asme.org/terms