Combustion and Flame 169 (2016) 287–296

Contents lists available at ScienceDirect

Combustion and Flame

journal homepage: www.elsevier.com/locate/combustflame

On the experimental determination of growth and damping rates for

combustion instabilities
D. Mejia∗, M. Miguel-Brebion, L. Selle
Institut de Mécanique des Fluides de Toulouse (IMFT) - Université de Toulouse, CNRS-INPT-UPS, Toulouse, France

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents four experimental methods for the evaluation of growth rates of combustion insta-
Received 1 February 2016 bilities. A systematic investigation is conducted on a laminar slot burner with five operating points (two
Revised 3 May 2016
stable and three unstable). The accuracy of the methods is assessed by cross comparison and the use of
Accepted 4 May 2016
three different flow variables as input: velocity, pressure and heat release rate fluctuations. Finally, the
Available online 24 May 2016
experimental determinations of the growth rates are compared to the prediction of a low-order acoustic
Keywords: model fed with a Flame Transfer Function.
Combustion instabilities
© 2016 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
Flame Transfer Function
Growth rate
Damping rate

1. Introduction puted with high-fidelity unsteady 3D numerical simulations [12–

14].
Combustion instabilities (CI) are a major problem for the de- For the design of stable combustion devices, a trial-and-error
sign and operation of power-generation systems such as gas tur- experimental approach is very costly so that alternatives using
bines, aeronautical engines and rocket engines [1]. The construc- numerical simulation are sought. The first option, which may be
tive coupling between acoustic waves and unsteady heat release called a brute-force approach is to solve the reacting Navier–Stokes
rate constitutive of CI, is responsible for loss of performance, re- equations over the whole configuration. Large-Eddy Simulation, for
striction under operating conditions and sometimes catastrophic example, has shown its potential for this task [15,16]. The compu-
failures. The challenge for understanding and predicting CI lies in tational cost of a single simulation is however so large that the
the multiplicity of physical phenomena involved in the unstable exploration of the whole range of operating conditions or design
loop [2]: acoustics, vortex dynamics, mixing, turbulence, chemistry, variations is impractical. An alternative is to solve only for acous-
two-phase flows, etc. One of the canonical configurations for the tic perturbations and model the flame via the FTF: the resulting
study of CI is the laminar premixed flame, which has been exten- tool may be called a thermoacoustic code. Whether the equations
sively scrutinized [3–6]. Despite its apparent simplicity, this flame be Linearized Euler or Helmholtz, in complex 3D geometries or
contains one of the main physical mechanisms driving CI: the dy- 1D network models, the methodology is usually computationally
namic interaction between acoustics, vortical structures in the flow lean and has shown its ability to predict stability maps [10,17–20].
field and the flame front. Comparative studies of conical, V-shaped The typical output of a thermoacoustic solver is threefold: (1) the
and triangular flames gives insight into the mechanisms driving frequency of the instability, (2) the shape of the associated pres-
the overall response of the flame to acoustic perturbations [7]. The sure and velocity fields and (3) the linear growth rate of individual
crux of the matter for the prediction of CI is to model this re- eigenmodes.
sponse, which is often achieved through a so-called Flame Transfer Because the impact of the flame as an active acoustic source is
Function (FTF), relating the unsteady heat release rate of the flame, usually small, the prediction of eigenfrequencies and mode shapes
q˙  , to the incident acoustic velocity perturbation, v . In many of is not a difficult task. The determination of the growth rates is
the above mentioned configurations, reasonably accurate analytical more subtle as it requires a precise evaluation of the flame re-
models for the FTF can be derived [6,8]. For realistic combustion sponse (with a FTF for example), the fluxes through the boundaries
devices, the FTF must be measured experimentally [9–11] or com- (by specifying the impedances) and the sources of internal dissipa-
tion (through friction at the walls and transfer from acoustics to
vorticity). And yet, a precise determination of the linear growth
∗
Corresponding author. rate is crucial for the design and safe operation of an engine. In-
E-mail address: [email protected], [email protected] (D. Mejia). deed, even if a given operating point is observed as stable, it is

http://dx.doi.org/10.1016/j.combustflame.2016.05.004
0010-2180/© 2016 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
288 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296

particular solutions to the corresponding four damping/growth rate

Nomenclature identification methods. Then, the experimental setup and diagnos-
tics are presented in Section 3. The experimental results for each
Greek symbols of the four identification methods is then discussed in Section 4.
α Natural damping of the system under non-reacting Finally, a low-order model is derived in Section 5 and compared to
conditions [s −1 ] the results of Section 4 for the validation of its ability to predict
β Growth/damping rate induced by the flame [s −1 ] the linear growth rate of CIs.
δ Impulse signal in the IR test [Volts]
η Any state variable 2. Theoretical model
κ Coefficient that accounts for the non-linearity of the
flame response The objective of this work is the evaluation of the growth rate
ν Growth/damping rate of the system [s −1 ] for a specific mode and as explained in [19,21], around a given
ω Angular frequency [Rad/s] eigenmode of angular frequency ω0 , it is a valid assumption to
ω0 Response/resonance angular frequency [Rad/s] consider that the thermoacoustic system obeys a second-order dif-
ω0, nr Response angular frequency under non-reacting ferential equation of the form:
configuration [Rad/s]
η̈ − f (η, η˙ ) + ω02 η = ξ (1)
Equivalence ratio [-]
φ Phase [Rad] This formulations follows that of [21], where η represents any
ϕ Phase of the FTF [Rad] state variable and f is a non-linear function of η and its first deriva-
τ Time delay between the velocity and heat release tive η˙ , which accounts for internal dissipation, fluxes of acoustic
rate signals [s] energy at the boundaries and the contribution of unsteady com-
ξ Forcing term bustion, i.e. the flame. ξ , corresponds to a forcing term due to an
ζ Uncorrelated with noise external device such as a loudspeaker. In the present study, η can
be the acoustic pressure p or acoustic velocity v at a given lo-
Roman letters cation in the experimental rig, or the heat release rate integrated
A Amplitude of the limit cycle over the entire combustion chamber, q˙  . Following [21], the func-
a Acceleration [G] tion f may be modeled as:
A0 Amplitude of the limit cycle without stochastic forc-
ing f (η, η˙ ) = η˙ (β − α − κη2 ) (2)
AIR Amplitude of the impulse signal in the IR test where α corresponds to the contribution of linear acoustic absorp-
[Volts] tion in the volume and fluxes at the boundaries, meaning that it
C Combustion noise factor [-] is positive. β is the linear contribution of the feedback induced by
c0 Normalized stochastic forcing level [-] the flame. It is positive when the flame drives the instability and
E Fresh to burnt-gases density ratio [-] negative when unsteady combustion damps acoustic fluctuations.
F Flame Transfer Function [-] The coefficient κ accounts for the non-linearity of flame response
f Frequency [Hz] to acoustics and controls the amplitude of the limit cycle reached
f0 Response frequency [Hz] when the system is linearly unstable. The growth/damping rate, ν ,
f0, nr Response frequency under no-reacting configuration of the system is defined as:
[Hz]
G β −α
Gain of the FTF [-] ν= . (3)
h Effective slot height [m] 2
hs Slot height [mm] Without combustion, β = 0, then ν = −α /2, so that the system is
I Intensity of CH∗ radical [Volts] always stable. In a reacting flow, if β > α then ν > 0, meaning that
ls Slot length [mm] the driving contribution of the flame overcomes the linear acous-
N Combustion interaction index [-] tic losses and the system becomes linearly unstable. On the other
p Acoustic pressure [Pa] hand, if β < α the system is linearly stable.
q˙ Heat release rate [∝Js −1 ] Eqs. (1)–3 provide the proper background for the extraction of
r Pinching distance [m] the deterministic quantities ω0 and ν from experimental data. Four
S Power spectral density identifications methods are now presented: Harmonic Response
Ss Slot cross section [m2 ] (HR), Impulse Response (IR), Active Control (AC) and White Noise
t Time [s] (WN).
Tg Fresh gases temperature [°C]
Ts Slot temperature [°C] 2.1. Harmonic response method (HR)
v velocity [m/s]
vb Bulk velocity [ms −1 ] The harmonic response has been extensively used in the study
ws Slot width [mm] of combustion instabilities for non-reacting [22] and reacting
[20,23] flows. However, this technique is limited to systems featur-
ing negative growth rates, i.e. ν < 0. For a linearly stable system,
useful to know how far from stability it is so that small changes
the non-linear part of Eq. (2) can be neglected as long as pertur-
(manufacturing variations, aging through mechanical and thermal
bations to the equilibrium state are small so that Eq. (1) reduces
stress, variations in fuel properties, etc.) do not trigger a CI.
to:
The objective of the present work is to present various experi-
mental techniques that can be used to measure the growth rate of η̈ − 2ν η˙ + ω02 η = ξ (4)
stable or unstable modes with the intent to serve as a validation Taking the power spectral density of Eq. (4) yields:
for thermoacoustic solvers. The manuscript is organized as follows:
First, in Section 2 a general model for a second-order dynamic sys-
Sξ
Sη = (5)
tem is presented. This model is the starting point to derive the (ω02 − ω2 )2 + 4ν 2 ω2
 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296 289

where Sη (respectively Sξ ) is the power spectral density of η (re- Table 1

spectively ξ ). Eq. (5) indicates that the system
 has a maximum
Summary of the different methods and the configuration to which they apply.

response at the angular frequency ω = ω0 1 − (ν /ω0 )2 with a Model equation Configuration

width at the half maximum of 2ν . Consequently, the principle of Sξ
HR Sη (ω ) = (ω2 −ω2 )2 +4ν 2 ω2 Stable
the HR method is to determine ω0 and ν by applying an optimal 0
IR η (t ) = η0 eνt/2 cos(ω0 t + φ ) Stable
fit of Eq. (5) to the power spectral density of experimentally mea- AC ν
η (t ) = η0 e t/2 cos(ω0 t + φ ) Unstable
sured p , v or q˙  . WNa
S
Sη (ω ) = (ω2 −ω2 )2ξ +4ν 2 ω2 Stable
0
A2
WNb SA (ω ) = π0 ω2 +(2cν0 +c )2 Unstable
0
2.2. Impulse response method (IR)

For combustion application, the impulse response method has

The case of a linearly unstable system featuring a positive
been less used [20,24] than the HR method. Moreover, it is only
growth rate (ν > 0) is now considered. As lucidly explained in
suitable for linearly stable systems, i.e. when ν < 0. The system is
[21] the amplitude, A, of the limit cycle fluctuates around a mean
pushed away from its steady state by a very short perturbation and
value, Ā, because of the stochastic perturbations from ξ . These
it subsequently behaves as a damped harmonic oscillator, which
fluctuations are denoted A and follow the first-order differential
obeys the homogeneous form of Eq. (1):
equation:
η̈ − 2ν η˙ + ω02 η = 0 (6)
A˙  = −(2ν + c0 )A + ζ (8)
Because non-linear terms have been neglected, Eq. (6) is only valid
where c0 = π Sξ /2ω02 A20 ,
A0 is the amplitude of the limit cycle
for small amplitudes. In the limit case where (ν /ω0 )2 < <1, which
without stochastic forcing1 and ζ is an uncorrelated white noise
has been verified for all cases presented in this paper, the solution
with an intensity related to that of ξ through Sζ = Sξ /2ω02 . Taking
of Eq. (6) is:
the PSD of Eq. (8) yields:
η = η0 eνt cos(ω0t + φ ) (7)
A20 c0
SA = (9)
Thus, the principle of the IR method is to determine ω0 and ν by π ω 2 + ( 2 ν + c0 )2
fitting Eq. (7) to the time traces of the flow variables p , v or q˙  .
The principle of the WN method for an unstable flame consists in
fitting Eq. (9) to experimental data. As noted in [21] this method is
2.3. Active control method (AC)
valid only if the limit cycle is weakly perturbed by additive white
noise, i.e. ν  c0 .
Active control of combustion instabilities has been used in the
The various methods proposed for the determination of
combustion community for a while [25–29] and more recently
the growth rate of combustion instabilities are summarized in
[30]. It was first introduced in the 1980s on a Rijke tube [31],
Table 1. It should be pointed out that the methods described in
demonstrating that using a very small fraction of the system en-
Section 2.1 and Section 2.4 for stable systems assume that there
ergy, the acoustic pressure amplitude can be reduced by several
is no linear modal interaction, i.e. the eigenfrequencies are not too
orders of magnitude. The theory behind active control of combus-
close from each other. Otherwise, the responses related to two ad-
tion instabilities is out of the scope of this study, here we simply
jacent modes may overlap. In the present work, this is not a prob-
use this technique with the intent to measuring ω0 and ν .
lem since the closest mode is several hundred Hertz away. How-
Once a system with a positive growth rate (ν > 0), is stabi-
ever, in annular systems for example, the modal density is usually
lized by an active control method, it is possible to stop the con-
quite high, questioning the applicability of the methods.
trol and study the temporal evolution of the system in the linear
regime. Although the growth rate is positive, the assumptions of
Section 2.2 still hold so that the system follows Eq. (7). 3. Experimental set-up
The principle of the AC method is then the same as the IR
method: recovering ω0 and ν by fitting Eq. (7) to the time traces The experiment is a slot burner [20,33] with a laminar pre-
of p , v or q˙  . Care needs to be taken regarding the assumption of mixed methane/air flame stabilized at the burner outlet (cf. Fig. 1).
linearity of the flame response meaning that only small amplitudes The four main components are: a mixing chamber, a plenum, a
should be considered. convergent nozzle and a rectangular-cross-section slot (width ws =
10 mm and length ls = 100 mm) with a height hs = 70 mm. The
flow rates of air and methane are controlled and measured via in-
2.4. White noise method (WN)
dividual mass flow meters (Bronkhorst F-201 AV-AAD-33-V). Reac-
tants are injected at the bottom of the mixing chamber in a cross-
Stochastic white noise was used in [21,32] to retrieve the de-
flow configuration to maximize mixing. The premixed flow is then
terministic parameters of a second-order dynamic system and in
laminarized in the plenum by an array of small glass balls and
[13] to calculate the FTF of a swirl burner in a LES simulation. In
three honeycomb panels. Finally, it passes through the converg-
[21], the forcing originated from turbulent fluctuations or from ac-
ing nozzle and the slot. The level of premixing is excellent as the
tuators excited by a signal generator. Unlike other methods, WN
flame shape is regular and steady in stable operating points. The
is adequate for both stable and unstable cases. However, as it is
temperature of the burner rim is controlled by four water cooled
based on the processing of stochastic data, long temporal samples
cylindrical passages. The water temperature in the upper chan-
are needed to obtain accurate results.
nels can be set between 1 and 99 °C, with a mass flow rate up
When the system features a negative growth rate (ν < 0), the
to 1 kg/min. In the lower channels, water at ambient temperature
assumption of Section 2.1 still hold. The difference between the HR
flows at 5 kg/min. The purpose of this dual-channel cooling system
and WN methods lies in the nature of the forcing, which is har-
is to be able to control Ts , while maintaining the lower part of the
monic for the former and non-coherent stochastic white noise for
slot and the burner under ambient conditions.
the latter. The procedure for the determination of ω0 and ν is then
the same as for the HR method: fitting Eq. (5) to the power spec-
tral density of experimentally measured p , v or q˙  . 1
In the limit of small perturbations, one has Ā ≈ A0 .
290 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296

Table 2
Summary of the operating points chosen for the evaluation of the characteristic
parameters of the system. The identification methods applicable to each operation
points are highlighted in grey.

Operating point Name [-] Ts [°C] HR IR AC WN

Non Reacting NR 0.00 20 N/A N/A N/A

Reacting Very Stable RVS 0.90 120 N/A N/A N/A
Reacting Stable RS 0.85 120 N/A N/A N/A
Reacting Slightly Unstable RSU 0.80 120 N/A
Reacting Unstable RU 0.80 60 N/A N/A
Reacting Very Unstable RVU 0.80 20 N/A N/A

Fig. 1. Sketch of the burner: transverse cut and experimental diagnostics. M0 and Fig. 2. Results for the HR method for the three stable operating points, NR (left),
M1 are 1/4  microphones. PM + CH∗ is a photomultiplier equipped with CH∗ fil- RVS (center) and RS (right). The rows correspond to the three different state vari-
ter. HW corresponds to a hotwire. Ts is K-type thermocouple. S1 and S2 are two ables: internal pressure p0 (top row), velocity v (middle row) and heat release rate
loudspeakers. AC is an accelerometer. q˙  (bottom row).

3.1. Diagnostics perature, which can be used in a reliable and repeatable way to
adjust the level of instability [20]. Changing the cooling-water tem-
A K-type thermocouple is placed 1 mm below the burner out- perature allows to cross the stability limit so that the accuracy
let. It gives the temperature, Ts , of the material as close as possi- of each identification method can be tested. Five different com-
ble to the flame base. A hot-wire probe (Dantec Dynamics, Mini binations of equivalence ratio, , and burner rim temperature, Ts ,
CTA equipped with a 55P16 probe), labeled HW in Fig. 1, is used were chosen as operating points. The bulk velocity, vb = 1.8 ms −1 ,
to measure the unsteady axial velocity v(t), located 55 mm up- and fresh gases temperature, Tg = 27 °C, are kept constant. Table 2
stream of the slot outlet. A photomultiplier (Thorlabs, Bialkali PMT, presents the operating conditions: one non reacting and five re-
PMM01), labeled PM in Fig. 1, measures the unsteady intensity, acting cases. Two stable2 and three unstable conditions are con-
I(t), of spontaneous emission of CH∗ radicals. The PM is equipped sidered. The unstable points have different limit-cycle amplitudes
with a narrow-band filter centered on a wavelength λ = 430 nm of heat release rate fluctuations, which are 3%, 7% and 12% of
and is placed 400 mm away from the flame, aligned with the z the mean, for the Reacting Slightly Unstable (RSU), Reacting Un-
axis, facing the longitudinal axis of the flame, which is fully in- stable (RU) and Reacting Very Unstable (RVU) cases, respectively.
cluded in its field of view. It has been shown that for lean pre- Table 2 also shows, in grey, which identification method is appli-
mixed flames, I(t) is proportional to the heat release rate [34]. Two cable for each operating point.
microphones, M0 and M1 , (B&K 1/4” Prepolarized Free-field 4954-
B) are used to record the acoustic pressure fluctuation, p0 (t ) and
4.1. Harmonic response
p1 (t ), respectively. M0 is placed in a waveguide connected to a
pressure plughole in the middle of the plenum [35], while M1 is
The HR method can be applied to the three stable cases: NR,
located 300 mm away from the burner. Two loudspeakers (Focal,
RVS and RS. The external loudspeaker S1 (cf. Fig. 1) delivers an
ISN 100), one in the interior, S0 and the other at the exterior of
harmonic excitation ξ = (ξ0 e−iωt+φ ) towards the flame. The fluc-
the burner, S1 , are used to generate acoustic waves prescribed by
tuations of the flow variables, were characterized for frequencies,
a signal generator (National Instruments, PCI-6052E). Moreover, an
between f = 30 Hz and 80 Hz by steps of  f = 0.5 Hz. The am-
accelerometer (AllianTech, ultra miniature 3224A3 accelerometer)
plitude of the input signal is kept constant for all frequencies. The
is placed on the membrane of the loudspeaker S0 in order to mon-
signal is 2 s long and the result, for each frequency, is the average
itor its displacement.
of 4 samples.
Figure 2 shows the normalized power spectral density of pres-
4. Experimental results
sure (S p ), velocity (Sv ) and heat release rate (Sq˙  ) fluctuations for
0

The four identification methods described in Section 2 are now the three cases.
applied to the burner of Fig. 1 in order to estimate the frequency,
ω0 and growth/damping rate, ν . This slot burner is operated close 2
Heat release rate fluctuations are less than 1 % of the mean during stable oper-
to its stability limit. This limit is controlled by the burner-rim tem- ation.
 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296 291

Fig. 3. PSD of the velocity signal, Sv : + RS operating point, fit with Eq. (5).

Table 3
Results for the HR method applied to the stable cases, NR, RVS and RS.

NR RVS RS
Fig. 4. Typical response to an impulse delivered by S0 for the RVS case. Top: elec-
f0, nr [Hz] α [s −1
] f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
] trical impulse, δ , sent to the loud speaker; Middle: velocity signal, v; Bottom: ac-
celerometer signal, a.
p0 53.50 −23.57 58.87 −10.55 58.52 −7.47
v 52.35 −17.86 58.58 −9.06 58.36 −7.09
q˙  – – 58.63 −9.50 58.39 −7.38 the bottom of the plenum (cf. Fig. 1):
Mean 52.99 −20.71 58.69 −9.70 58.43 −7.31
std. 0.8 4.04 0.16 0.77 0.08 0.20 1
δ (t ) = √ e−(t/τIR )
2
(10)
AIR π
where AIR = 0.94 V is the amplitude of the electric signal sent to
All variables exhibit a peak in the magnitude of the response, the amplifier and τIR = 0.2 ms. The time traces recorded for the
typical of a resonance, as expected from Eq. (5). The presence of RVS case are shown in Fig. 4. At t < 0 the system is stable, then at
the flame has a three consequences: t = 0 the impulse is generated. The membrane of the loudspeaker
1. The peak shifts to the right meaning that unsteady combustion experiences a strong acceleration and returns to equilibrium in less
slightly increases ω0 . than 20 ms. The flow velocity is also perturbed by the impulse and
2. The width of the resonance peak decreases with combustion, slowly returns to equilibrium, in about 400 ms, with oscillations
indicating that the flame has a destabilizing effect on the typical of an under-damped system. The fact that the loudspeaker
system. However, its contribution is not sufficient to overcome membrane returns to equilibrium faster than the acoustic velocity
linear acoustic losses so that the burner remains stable. is a good indication that fluctuations observed in the flow vari-
3. The absolute amplitude of the acoustic signal at the response ables, correspond to the response of the burner, rather than resid-
frequency increases with combustion, which confirms item (2). ual forcing. The accuracy of this method is limited to high values
The rms of pressure fluctuations for the NR, RVS and RS cases of signal to noise ratio. Significant errors occur when the signal to
are 3.7 Pa, 9.7 Pa and 12.1 Pa, respectively. noise ratio falls below 10 with respect to the initial amplitude [24].
In order to eliminate the high frequency noise seen in the velocity
It should also be pointed out that the PSD of pressure in the signal of Fig. 4, the impulse is repeated 400 times spaced of 10 s
NR case is more noisy. No exact explanation for this is available at for each operating point. Finally, the IR is averaged over the 400
the moment. However, taking into account that the same excitation samples, increasing the signal to noise ratio from 11 to 31.
amplitude was used in non-reacting and reacting cases, the signal The averaged time traces for p0 , v and q˙  are presented in
to noise ratio is lower in the NR case due to its smaller growth Fig. 5, for the three linearly stable cases NR, RVS and RS. While ve-
rate. Nevertheless, this does not elucidate why the velocity spec- locity shows a monotonous decay, pressure fluctuations initially in-
trum is less noisy than that of the pressure. crease after the impulse for about two cycles. Given the amplitude
In order to illustrate the post-processing procedure, the PSD of the excitation, the initial heat release rate fluctuations are fairly
of velocity in the RS operating point is presented in Fig. 3, to- non-linear, which is to be expected [22]. Like for the HR method
gether with the corresponding fit from Eq. (5). For this specific (cf. Fig. 2), the pressure signal for the non-reacting case is not as
case, f0 = ω0 /2π = 58.36 ± 0.016 Hz and ν = 7.1 ± 0.14 s −1 , with clean as for the other operating conditions.
a 95% confidence interval. For the fit to Eq. (7), the experimental data is restricted be-
The results for the HR method are summarized in Table 3 and tween 80% and 10% of its peak amplitude. A typical example of
they quantify the qualitative observations obtained from Fig. 2. The this fit for the RVS case is presented in Fig. 6.
peak in the spectrum of the response is shifted by 10% between Table 4 summarizes the results of the IR method, which are
non-reacting and reacting cases and it is almost equal for the RVS consistent with the HR results of Table 3.
and RS cases. With the flame, ν increases but remains negative.
The system is still stable but closer to the stability limit. Likewise, 4.3. Active control
|ν | is larger for the RVS than RS case, which is consistent with the
results of [36]. The AC method can in principle be applied to all unstable
operating conditions. However, because the amplitude of the fluc-
4.2. Impulse response tuations are very small in the RSU case, the active control does not
bring significant reduction for this regime. It is therefore only ap-
The IR method is now applied to the NR, RVS and RS operating plied to the RU and RVU cases. The active control technique used
points. A short impulse, δ , is delivered by the loudspeaker S0 at in this study is a fixed-coefficients closed-loop controller [26].
292 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296

Fig. 7. PSD of velocity fluctuations: with active control, without

active control.

Fig. 5. Results for the impulse response method for the three stable operating
points (NR, RVS and RS) and the three measured variables (pressure p0 , velocity
v and heat release rate q˙  . All signals are normalized by their maximum value.

Fig. 8. Internal acoustic pressure, p0 (t ), for a RVU cycle of operation where the
active control is switched OFF and ON.

Table 5
Results of the AC method for the RU and RVU operating points.

Fig. 6. Velocity signal for the IR method: + RVS operating point, fit with RU RVU
Eq. (7). f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
]

p0 56.02 1.80 55.00 4.91

Table 4
v 56.01 1.84 54.99 4.93
Results for the IR tests applied to the stable cases, NR, RVS and RS.
q˙  56.01 1.61 55.00 4.38
NR RVS RS Mean 56.01 1.75 55.00 4.74
std. 0.00 0.12 0.00 0.31
f0, nr [Hz] α [s −1
] f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
]

p0 51.60 −15.20 58.67 −11.53 58.34 −6.67

v 51.68 −15.61 59.01 −11.50 58.35 −6.64 at t = 0 and the instability grows. A typical time trace of plenum
q˙  – – 58.91 −11.41 58.38 −6.63 pressure fluctuations, p0 is shown in Fig. 8.
Mean 51.64 −15.41 58.90 −11.48 58.44 −6.65
std. 0.08 0.29 0.12 0.06 0.02 0.02
This procedure is repeated 40 times and the measurements are
phase averaged in order to reduce noise. Figure 9 shows the cor-
responding traces for RU and RVU, restricted to the interval where
the amplitude is between 20% and 60% of the limit cycle, which is
The signal from the plenum microphone, M0 is phase delayed, considered for post-processing. It is obvious that the RVU case is
amplified and sent to the speaker S0 mounted on the bottom of the most unstable as it takes less cycles to reach the upper bound.
the plenum (cf. Fig. 1). The gain and phase lag of this controller The quantitative results of the AC method are summarized in
are constant and identical for the RU and RVU cases. Table 5. The growth rate is larger for the RVU condition, as ex-
Figure 7 shows the spectra of the velocity fluctuations for the pected from Fig. 9. Results also show very little dispersion regard-
RVU case, with and without active control. When activated, AC less of the variable used for the analysis.
reduces the amplitude of the fundamental frequency by almost
50 dB. 4.4. White noise
The procedure for the AC measurements is based on that of
[26] and is as follows: the controller is turned on and the sys- For the WN method, the stochastic non-coherent noise is im-
tem is kept stable for a few seconds; the control is then turned off posed by the plenum loudspeaker S0 (see Fig. 1). The amplitude of
 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296 293

Fig. 11. PSD of velocity fluctuations with the WN method: + RVS operating point,
fit with Eq. (5).

Table 6
Results of the WN method for the stable operating points: NR, RVS and RS.

NR RVS RS

f0, nr [Hz] α [s −1
] f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
]
Fig. 9. Averaged time traces of the AC test for RU and RVU conditions for the three
flow variables, p0 , v and q˙  .
p
0 53.81 −27.12 60.33 −14.81 58.80 −7.91
v 51.83 −17.34 59.48 −13.23 58.56 −7.19
q˙  – – 59.50 −13.48 58.53 −6.98
Mean 52.82 −22.23 59.77 −13.84 58.63 −7.63
std. 1.40 6.92 0.49 0.85 0.15 0.49

Fig. 11 shows the fit with Eq. (5) of the acoustic velocity spectrum
for the RVS case.
The quantitative results of the WN method for the cases with
negative growth rates are summarized in Table 6. These results are
consistent with that of the HR methods (cf. Table 3): the predic-
tion of the frequency are within 1 Hz but the damping is system-
atically overestimated by the WN method, especially for the RVS
case where it is 50% larger.

4.4.2. Positive growth rate

For the linearly unstable conditions (RSU, RU and RVU), the
damping rate of the system is obtained with the WN method
Fig. 10. Power spectral density obtained with the WN method for the stable oper- [21] by fitting the PSD of limit-cycle amplitude variations with
ating points, NR, RVS and RS.
Eq. (9). Similar to the linearly stable cases analyzed in Section 4.4.1,
each 100 s trace is split in 5 equal parts, then the PSD of the
magnitude of the analytic signal (i.e. the original signal + i times
the additive white Gaussian noise was chosen so that, in the case
its Hilbert transform) is calculated for each interval. The resulting
of a system with positive growth rate, ν is at least one order of
50 PSDs are averaged, which reduces the high frequency noise by
magnitude larger than the amplitude of the normalized stochastic
10 dB and are presented in Fig. 12 for the for the three linearly
forcing level, c0 , as required for the analysis (cf. Section 2.4). This
unstable operating points. Qualitatively, Fig. 12 suggests that the
amplitude was kept constant for all stable and unstable cases. The
cut-off frequency in the PSD of the limit-cycle amplitude increases
processed signals are 100 s long and the acquisitions are repeated
from the RSU to the RU and RVU cases. Using Eq. (9), this indi-
ten times for each operating point, this is equivalent to having sig-
cates that the growth rate increases. The sharp peak around 8 Hz
nals containing approximately 60,0 0 0 cycles for unstable cases.
in the spectra of heat release rate corresponds to the flickering of
the flame due to buoyancy. This perturbation is restricted to the
4.4.1. Negative growth rate burnt gases and therefore not measured in the pressure and veloc-
When the system is linearly stable, it is possible to extract the ity signals.
damping rate by fitting the power spectral density of the flow vari- The post processing procedure is illustrated in Fig. 13, showing
ables signals with the second order model given in Eq. (5). Each how the experimental PSD of the velocity signal for RSU is fit to
100 s long trace is split in 20 equal parts for which the PSD is cal- Eq. (9).
culated. The resulting 200 PSDs are averaged to lower stochastic Finally, the results of the WN method applied to the linearly
noise and the results of this procedure are shown in Fig. 10 for the unstable cases are summarized in Table 7. Comparing with the AC
three linearly stable cases NR, RVS and RS. The PSDs are very simi- method presented in Table 5, the predictions of ω0 are virtually
lar to the results of the HR method (cf. Fig. 2) except that they are identical. The growth rates determinations are similar within a 15%
not as smooth. In order to illustrate the post processing procedure, range.
294 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296

Fig. 12. PSD of the limit-cycle amplitude variations obtained with the WN method
for the unstable operating points: RSU, RU and RVU. Fig. 14. Summary of all measurements for the determination of the frequency and
growth rate of the first eigenmode of the slot burner.

5. Validation

At this point it is interesting to try and predict the stability map

of Fig. 14. For this purpose, a low-order model of the present slot
burner is now given, based on previous studies [22,35,37]. Because
the combustion instability is here locked on the Helmholtz mode
of the plenum, one can derive an analytical model valid for this
specific mode. For harmonic velocity perturbations, v = v˜ e−iωt , the
system obeys a second-order equation for the pulsation ω, which
may be written in the form [20,36]:
  2
1 + N eiωτ ω + 2iα ω − ω02,nr = 0 (11)

where ω0, nr is the resonance frequency and α the damping rate

of the system, both under non-reacting conditions. The fact that
Fig. 13. Spectrum of limit-cycle amplitude variations for the velocity: + RVU oper-
ating point, fit with Eq. (9).
ω0, nr and α correspond to non-reacting conditions stems from the
derivation of the model [35–37]. The influence of the flame is mod-
eled by an interaction index, N , and a time delay, τ , which are
Table 7 linked to the Flame Transfer Function (FTF). The FTF, F (ω ), is a
Results of the WN method for the linearly unstable operating points: RSU, RU and complex number measuring the response of the flame to acous-
RVU. tics, and is expressed as the ratio of the relative heat release rate
RSU RU RVU fluctuations, q˙  /q,
¯˙ to the incoming velocity fluctuations, v /v̄:

f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
] f0 [Hz] ν [s −1
] q˙  /q¯˙
F ( ω ) =  = G ( ω ) e iϕ (12)
p0 56.22 0.50 55.94 1.70 54.97 5.62 v /v̄
v 56.22 0.50 55.94 1.63 54.97 5.21
where G is the gain of the FTF and ϕ its phase, which is linked to
q˙  56.22 0.55 55.94 1.46 54.97 5.22
Mean 56.22 0.51 55.94 1.60 54.97 5.35 the time delay, τ , of the flame through ϕ = ωτ . Then, the interac-
std. 0.00 0.03 0.00 0.12 0.00 0.23 tion index, N , is defined as:

N = CG (ω ) (13)

where C is a factor accounting for sound radiation from the flame,

4.5. Summary depending on the burner geometry and the operating conditions.
Following the work of Price et al. [34]:
The measurements of f0 and ν obtained with the four tech- 1 Ss E − 1
niques on the five operating points are summarized in Fig. 14, C= (14)
4π h r
where each symbol corresponds to the mean value of all three
flow variables for each operating point. For the NR case, because where Ss = 10−3 m2 is the slot cross section, h = 0.106 m the ef-
the pressure PSD are noisy, the evaluation with the HR and WN fective height of the slot3 and E is the ratio of the fresh to the
methods were discarded. All identification methods show consis- burnt-gases density. E changes because the equivalence ratio is not
tent results for the resonance frequency, with a dispersion usually the same for all cases. It is computed by an equilibrium calculation
smaller than 1 Hz. The growth / damping rates predictions are also with the Cantera software. The distance between the noise source
surprisingly gathered. The largest dispersion between identification
methods is found for the RVS case but no specific reason for this 3
Two corrections must be applied to h in order to account for the converging
larger scatter is foreseen. section upstream of the slot and the classical end-correction at the exit [36].
 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296 295

Table 8
Combustion noise parameters for the various operating conditions.

Test E [-] r [mm] C [-]

RVS 7.10 24.35 0.19

RS 6.89 26.33 0.17
RUS 6.65 29.04 0.15
RU
RVU

Fig. 16. Validation of Eq. (11) for the prediction of the frequency, f0 and growth
rate, ν of the eigenmode of the slot burner.

i.e. [57,59] Hz. Thus, the necessary condition for an instability to

develop, resulting from the Rayleigh criterion is fulfilled [22]. Nev-
ertheless, only RSU RU RVS are unstable. The reason is that this
condition is necessary but no sufficient: in order to trigger an in-
stability the gain of the FTF must be large enough to overcome the
various acoustic losses, which is not the case for the RVS an RS
Fig. 15. FTF of the five reacting cases of Table 2. Gain, G (top) and phase ϕ (bot-
conditions.
tom).
The solutions of Eq. (11) are presented in Fig. 16. It is clear that
the predictions of Eq. (11) closely follow the experimental data and
and the burner outlet is denoted r and it is obtained from direct always recover the proper stability of the eigenmode. While the
visualization as the location where the flame pinches [22]. The nu- frequencies typically match by less than 5%, the growth rates are
merical values of E, r and C and are recalled in Table 8. systematically overestimated by a few Hz for the reacting cases.
The stability of the system is determined by the complex roots,
ω = ω0 + iν, of Eq. (11), which can be solved once the values of 6. Conclusion
N , τ , ω0, nr and α are known. This is the classical paradigm of
stability map prediction with an acoustic solver: one must know a Four methods were presented for the experimental evaluation
priori the response of the flame and the acoustic damping rate of of the growth (or damping) rate associated with an acoustic eigen-
the setup. The requirement to know ω0, nr is specific to the present mode. The objective is the detailed analysis of combustion insta-
model, which is valid for a flame locked on a Helmholtz mode. A bilities, either for the mitigation of unstable modes, or the as-
more generic methodology, using for example a Helmholtz solver sessment of the distance to the stability limit for a stable mode.
does not require the a priori knowledge of the eigenmodes. Using These methods were tested on five operating points of a laminar
the results of Section 4 in the NR case, the values f0,nr = 52.50 Hz slot burner: two stable and three unstable. The evaluation of the
and α = 16.94 s −1 are used. frequency and growth rate were consistent between the methods,
Finally, the FTF are obtained by the classical procedure of send- giving confidence for their application.
ing a small amplitude harmonic signal through the speaker S0 . Then a low-order model of the slot burner was derived for
The flame must be stable in order to measure the FTF and as ex- the prediction of its stability. The model was fed by Flame Trans-
plained in [36], a simple way to stabilize this setup is to increase fer Functions obtained experimentally and its solutions were com-
the height of the slot. This modification does not alter the flame pared to the experiment. Both the frequency and the growth rate
but shifts down the frequency of the Helmholtz mode. For the of the eigenmode were predicted with good accuracy for stable
measurement of the FTF in the RSU, RU and RVU conditions, the and unstable operating conditions. These results give a clear in-
height is increased to 150 mm, resulting in stable combustion. It dication that the overall methodology of feeding low-order models
was checked for the stable operating points that the FTFs measured or acoustic solvers by FTF is sound and accurate, even close to the
with a 150 mm slot and 70 mm slot are identical. Only the fre- stability limit.
quency range [48, 62] Hz is needed for this study. Heat release rate It is suggested that these methods be used more systematically
and velocity fluctuations are recorded simultaneously by the pho- on laboratory and industrial burners because they provide the dis-
tomultiplier and the hot-wire probe, respectively (cf. Fig. 1). The tance to the stability limit, which is very useful for the assessment
level of rms velocity fluctuation is kept constant at 10% of the bulk of the robustness of a given design. Nevertheless, not all methods
velocity (vb = 1.8 m/s) for all frequencies. Figure 15 shows the gain are suitable for all systems and some practical limitations should
and the phase of the FTF for the five reacting operating points, RVS, be discussed. For example, installing actuators such as loudspeak-
RS, RSU, RU, RVU. Because in the range of interest both the gain ers inside combustion chambers is not always possible, especially
and the phase of the FTF vary linearly with ω, for the numerical on industrial burners with compact geometries. In such situation,
resolution of Eq. (11), they were interpolated with linear functions. if the flow is turbulent, the WN method presented in Section 2.4 is
It should be pointed out that, for all operating points, the phase a good option. Also, the HR and WNa methods assume that there
of the FTF is between [π , 2π ] at the frequency of the eigenmode, is no linear modal interaction, that is, the eigenfrequencies need
296 D. Mejia et al. / Combustion and Flame 169 (2016) 287–296

to be sufficiently far from each other. This may limit their use in [17] N. Noiray, D. Durox, T. Schuller, S. Candel, A unified framework for nonlinear
systems where the modal density is high. In these cases the IR, AC combustion instability analysis based on the flame describing function, J. Fluid
Mech. 615 (2008) 139.
or WNb methods may be tried. [18] J. Pieringer, T. Sattelmayer, F. Fassl, Simulation of combustion instabilities in
liquid rocket engines with acoustic perturbation equations, J. Propuls. Power
Acknowledgments 25 (5) (2009) 1020–1031.
[19] C.F. Silva, F. Nicoud, T. Schuller, D. Durox, S. Candel, Combining a Helmholtz
solver with the flame describing function to assess combustion instability in a
The research leading to these results has received funding from premixed swirled combustor, Combust. Flame 160 (9) (2013) 1743–1754.
the European Research Council under the European Union’s Sev- [20] D. Mejia, L. Selle, R. Bazile, T. Poinsot, Wall-temperature effects on flame re-
sponse to acoustic oscillations, Proc. Combust. Inst. 35 (3) (2015) 3201–3208.
enth Framework Programme (FP/2007-2013) / ERC Grant Agree-
[21] N. Noiray, B. Schuermans, Deterministic quantities characterizing noise driven
ment ERC-AdG 319067-INTECOCIS. Hopf bifurcations in gas turbine combustors, Int. J. Non. Linear. Mech. 50
(2013) 152–163.
References [22] T. Schuller, D. Durox, S. Candel, Self-induced combustion oscillations of laminar
premixed flames stabilized on annular burners, Combust. Flame 135 (4) (2003)
525–537.
[1] T.C. Lieuwen, Unsteady combustor physics, Cambridge University Press, 2012.
[23] P. Palies, D. Durox, T. Schuller, S. Candel, Nonlinear combustion instability anal-
[2] F. Culick, Unsteady motions in combustion chambers for propulsion systems,
ysis based on the flame describing function applied to turbulent premixed
NATO/RTO-AG-AVT-039, 2006.
swirling flames, Combust. Flame 158 (10) (2011) 1980–1991.
[3] L. Boyer, J. Quinard, On the dynamics of anchored flames, Combust. Flame 82
[24] M. Wagner, C. Jörg, T. Sattelmayer, Experimentally determining the acoustic
(1) (1990) 51–65.
damping rates of a combustor with a swirl stabilized lean premixed flame,
[4] T. Schuller, D. Durox, S. Candel, Dynamics of and noise radiated by a perturbed
Proceedings of the ASME Turbo Expo 2013 (2013), pp. 1–11. GT2013-94844.
impinging premixed jet flame, Combust. Flame 128 (1-2) (2002) 88–110.
[25] G.J. Bloxsidge, A.P. Dowling, N. Hooper, P.J. Langhorne, Active control of an
[5] H.Y. Wang, C.K. Law, T. Lieuwen, Linear response of stretch-affected premixed
acoustically driven combustion instability, J. Mec. Theor. Appl. Suppl. 6 (1987)
flames to flow oscillations, Combust. Flame 156 (4) (2009) 889–895.
161–175.
[6] R.S. Blumenthal, P. Subramanian, R. Sujith, W. Polifke, Novel perspectives
[26] W. Lang, T. Poinsot, S. Candel, Active control of combustion instability, Com-
on the dynamics of premixed flames, Combust. Flame 160 (7) (2013) 1215–
bust. Flame 70 (3) (1987) 281–289.
1224.
[27] T. Poinsot, B. Yip, D. Veynante, A. Trouve, J.M. Samaiego, S. Candel, Active
[7] T. Schuller, D. Durox, S. Candel, A unified model for the prediction of laminar
control: an investigation method for combustion instabilities, J. Phys. III 2 (7)
flame transfer functions comparisons between conical and V-flame dynamics,
(1992) 1331–1357.
Combust. Flame 134 (1-2) (2003) 21–34.
[28] K.R. McManus, T. Poinsot, S. Candel, A review of active control of combustion
[8] A. Cuquel, D. Durox, T. Schuller, Impact of flame base dynamics on the non–
instabilities, Prog. Energy Combust. Sci. 19 (1) (1993) 1–29.
linear frequency response of conical flames, Comptes Rendus Mécanique 341
[29] A.M. Annaswamy, M. Fleifil, J.W. Rumsey, R. Prasanth, J.P. Hathout,
(1-2) (2013) 171–180.
A.F. Ghoniem, Thermoacoustic instability: model-based optimal control designs
[9] K. Kim, J. Lee, B. Quay, D. Santavicca, Response of partially premixed flames to
and experimental validation, IEEE Trans. Control Syst. Technol. 8 (6) (20 0 0)
acoustic velocity and equivalence ratio perturbations, Combust. Flame 157 (9)
905–918.
(2010) 1731–1744.
[30] M.R. Bothien, J.P. Moeck, C.O. Paschereit, Comparison of linear stability analy-
[10] D. Durox, J.P. Moeck, J.-F. Bourgouin, P. Morenton, M. Viallon, T. Schuller,
sis with experiments by actively tuning the acoustic boundary conditions of a
S. Candel, Flame dynamics of a variable swirl number system and instability
premixed combustor, J. Eng. Gas Turb. Power 132 (12) (2010) 121502.
control, Combust. Flame 160 (9) (2013) 1729–1742.
[31] M.A. Heckl, Active control of the noise from a rijke tube, J. Sound Vib. 124 (1)
[11] B. Ćosić, S. Terhaar, J.P. Moeck, C.O. Paschereit, Response of a swirl-stabilized
(1988) 117–133.
flame to simultaneous perturbations in equivalence ratio and velocity at high
[32] N. Stadlmair, M. Wagner, C. Hirsch, T. Sattelmayer, Experimentally determining
oscillation amplitudes, Combust. Flame 162 (2014) 1046–1062.
the acoustic damping rates of a combustor with a swirl stabilized lean pre-
[12] A. Giauque, L. Selle, L. Gicquel, T. Poinsot, H. Buechner, P. Kaufmann, W. Krebs,
mixed flame, Proceedings of ASME Turbo Expo 2015 (2015), pp. 1–10. GT2015-
System identification of a large-scale swirled partially premixed combustor us-
42683.
ing LES and measurements, J. Turbul. 6 (February 2012) (2005) N21.
[33] L. Selle, T. Poinsot, B. Ferret, Experimental and numerical study of the accuracy
[13] L. Chong, T. Komarek, R. Kaess, S. Föller, W. Polifke, Identification of flame
of flame-speed measurements for methane/air combustion in a slot burner,
transfer functions from LES of a premixed swirl burner, Proceedings of the
Combust. Flame 158 (1) (2011) 146–154.
ASME Turbo Expo 2010 Power Land, Sea Air, ASME, Glasgow (2010).
[34] R. Price, I. Hurle, T. Sugden, Optical studies of the generation of noise in tur-
[14] X. Han, J. Li, A.S. Morgans, Prediction of combustion instability limit cycle os-
bulent flames, Symp. Combust. 12 (1) (1969) 1093–1102.
cillations by combining flame describing function simulations with a thermoa-
[35] D. Durox, T. Schuller, N. Noiray, A. Birbaud, S. Candel, Rayleigh criterion and
coustic network model, Combust. Flame 162 (10) (2015) 1–16.
acoustic energy balance in unconfined self-sustained oscillating flames, Com-
[15] P. Wolf, G. Staffelbach, L.Y. Gicquel, J.-D. Müller, T. Poinsot, Acoustic and large
bust. Flame 156 (1) (2009) 106–119.
Eddy simulation studies of azimuthal modes in annular combustion chambers,
[36] D. Mejia, Wall-temperature effects on flame response to acoustic oscillations,
Combust. Flame 159 (11) (2012) 3398–3413.
Université de Toulouse, 2014 Ph.D. thesis.
[16] L. Selle, R. Blouquin, M. Théron, L.-H. Dorey, M. Schmid, W. Anderson, Pre-
[37] D. Durox, T. Schuller, S. Candel, Self-induced instability of a premixed jet flame
diction and analysis of combustion instabilities in a model rocket engine, J.
impinging on a plate, Proc. Combust. Inst. 29 (2002) 69–75.
Propuls. Power 30 (4) (2014) 978–990.