0% found this document useful (0 votes)

42 views28 pages

Poinsot 2016

Uploaded by

zkqb123

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

42 views28 pages

Poinsot 2016

Uploaded by

zkqb123

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

JID: PROCI

ARTICLE IN PRESS [m;June 25, 2016;16:33]

Available online at [Link]

Proceedings of the Combustion Institute 000 (2016) 1–28

[Link]/locate/proci

Prediction and control of combustion instabilities in real

engines
T. Poinsot a,b,c,∗
a IMF Toulouse, INP de Toulouse and CNRS, Toulouse 31400, France
b CERFACS, Toulouse 31057, France
c Center for Turbulence Research, Stanford, CA 94305, USA

Received 24 November 2015; accepted 19 May 2016

Available online xxx

Abstract

This paper presents recent progress in the field of thermoacoustic combustion instabilities in propulsion en-
gines such as rockets or gas turbines. Combustion instabilities have been studied for more than a century
in simple laminar configurations as well as in laboratory-scale turbulent flames. These instabilities are also
encountered in real engines but new mechanisms appear in these systems because of obvious differences with
academic burners: larger Reynolds numbers, higher pressures and power densities, multiple inlet systems,
complex fuels. Other differences are more subtle: real engines often feature specific unstable modes such as
azimuthal instabilities in gas turbines or transverse modes in rocket chambers. Hydrodynamic instability
modes can also differ as well as the combustion regimes, which can require very different simulation models.
The integration of chambers in real engines implies that compressor and turbine impedances control instabil-
ities directly so that the determination of the impedances of turbomachinery elements becomes a key issue.
Gathering experimental data on combustion instabilities is difficult in real engines and large Eddy simulation
(LES) has become a major tool in this field. Recent examples, however, show that LES is not sufficient and
that theory, even in these complex systems, plays a major role to understand both experimental and LES
results and to identify mitigation techniques.
© 2016 by The Combustion Institute. Published by Elsevier Inc.

Keywords: Instabilities; LES; Thermoacoustics; Gas turbines

1. Introduction working on steady combustion chambers (furnaces,

gas turbines, power plants) know that ’sometimes’,
Most combustion systems are designed to oper- a chamber starts exhibiting unexpected oscillations.
ate in stable regimes. However, all experimentalists These combustion instabilities (often called ther-
moacoustic) lead to additional noise and vibration,
which can be ignored or tolerated if the level of
∗ Correspondence to: IMF Toulouse, INP de Toulouse oscillations remains small (less than a few mbars).
and CNRS, 31400 Toulouse, France In other cases, however, pressure fluctuations can
E-mail address: [Link]@[Link] reach values of the order of a large fraction of

[Link]
1540-7489 © 2016 by The Combustion Institute. Published by Elsevier Inc.

Please cite this article as: T. Poinsot, Prediction and control of combustion instabilities in real engines,
Proceedings of the Combustion Institute (2016), [Link]
JID: PROCI
ARTICLE IN PRESS [m;June 25, 2016;16:33]

2 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

the mean pressure and lead to more serious con- modes in rocket engines [19,20], which ap-
sequences: the chamber can quench, the flame can pear on real systems but are difficult to trig-
flashback and burn part of the injection system. ger in laboratory set-ups, which are scaled
The pressure oscillations can also become large down in terms of pressure and power density
enough to damage the combustor structure or lead [21–23].
to the explosion of the engine. • Most laboratories study instabilities in
Combustion instabilities (CIs), also called flame gaseous flames. Even if research on CIs
dynamics, are an important field of combustion actually started in the 50s for liquid-fueled
research. It combines all usual sciences involved systems [24–26], studying instabilities in
in reacting flows (kinetics, transport, fluid me- liquid fueled combustors is much more
chanics, thermodynamics) but also requires the complicated [27,28].
introduction of acoustics, hydrodynamic stabil- • The geometry of the chamber is a first-order
ity, dynamical systems and control theory. Even parameter for CIs: in gas turbines, combus-
though CIs can appear in almost any combustion tion chambers have annular shapes where
system, industry is not especially keen on studying azimuthal instability modes can develop,
them or in recognizing that they can be a problem something that had never been studied in lab-
in their company’s engines. The main reason for oratories until very recent times [29–31] for
this is that CIs are the cause of major problems, obvious cost reasons. Therefore the physics
which are difficult to master because they occur of azimuthal modes was known mainly on
during the last stages of development and are still the basis of limited experimental observa-
difficult to predict today: having an unstable engine tions performed on real full turbines [32–34].
is still similar to catching some kind of disease for • Laboratory burners use single injection sys-
companies. Since the adventure of the F1 engine tems, which create isolated flames. Gas tur-
during the Apollo program in the 60s, which cost bines chambers can have twenty injectors and
billions of dollars before a solution was found to flames. Rocket chambers feature hundreds of
mitigate CIs [1], companies know that CIs are a injectors: therefore flame/flame interactions
major industrial risk, for which communication are dominant in real engines and absent in
may not be the best solution. CIs have been the most laboratory burners.
hidden and feared problem of many combustion • Most instability modes studied in labora-
programs, starting with solid and liquid fuel rocket tories involve longitudinal acoustic modes,
engines in the 50s and continuing more recently in which propagate along the flow direction.
gas turbines, industrial furnaces or even domestic In real engines, modes can also be trans-
heaters. verse, radial or azimuthal. More importantly,
Laboratories, on the other hand, have no diffi- modes can be much more numerous than in
culty studying instabilities in canonical cases such a laboratory experiment because real config-
as laminar premixed flames and the literature con- urations are geometrically complex. An in-
tains a very large amount of research work dedi- dustrial gas turbine can exhibit 30 acoustic
cated to CIs in simple flames [2–9]. When it comes eigenmodes between 0 and 250 Hz, many
to real engines,1 the situation is different. Here, CI of them having the capacity to become self-
mechanisms involve not only those found in aca- excited [34]. This never happens in labora-
demic experiments, but also introduce new physics tory systems, which are much simpler. Being
that is usually not studied in laboratories: able to recognize which mode appears in such
a system becomes significantly more compli-
• Real flames are turbulent and many of them
cated than in a laboratory scale experiment,
are swirled and confined in complex shape
where only a few modes can be found and are
chambers. CIs in turbulent swirled flames
easily identified by their frequencies.
are now commonly studied in laboratories
• Many unstable modes studied in laborato-
[10–18] but usually in simplified chambers
ries correspond to situations where entropy
that do not contain the complexities found in
waves play a limited role. In chambers termi-
real engines.
nated by a nozzle or by a stator/rotor stage,
• Power density is known to be an impor-
entropy waves can be reflected back as acous-
tant parameter for instabilities: when com-
tic waves into the chamber, creating a new
bustion chambers become smaller or when
class of CI, called entropy - acoustic modes.
their power increases, CIs are more prone to
Fluctuations of equivalence ratio are also en-
appear. This usually occurs for high-pressure
countered in many real engines, where they
systems, which are more difficult to study in
can induce specific CI modes.
laboratories. The best example is transverse
• Finally, simple, well-defined inlet and out-
let conditions (imposed pressure or im-
1 The term ’real engine’ will be used here to designate en- posed velocity) can be imposed to control
gines used in real industrial systems, as opposed to cham- acoustic reflections in laboratory combustors
bers studied in laboratory environments. (Section 4.3). This is different in real engines.

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 3

Fig. 1. Rocket engine destroyed by instability during the early years of the US rocket program (left) and a laboratory
burner exhibiting both stable and unstable regimes (right) [38].

In gas turbines, for example, the chamber is chemistry and unsteady two-phase flow in complex
fed by a compressor and blows into a tur- geometries. The scales to capture vary from the
bine. Determining the impedances of these laminar flame thickness (less than 0.1 mm) to the
elements is a daunting task in itself. acoustic wavelengths (a few meters) and the speeds
from flame speeds (less than 1 m/s) to the sound
This review discusses combustion instabilities speed (more than 600 m/s in the burnt gases). Com-
appearing in engines. Its objective is to describe puting CIs is more difficult than computing steady
modern simulation methods combined with new combustion and requires more sophisticated tools
experiments and theoretical developments for such because they must capture unsteady phenomena
instabilities. Readers are referred to other reviews and unstable mechanisms. The intensity of the
for complete descriptions of instability mecha- acoustic field generated by a flame in the absence
nisms [20,35–37]. Of course, the presentation builds of CI is small: the acoustic power created by a
on fundamental results obtained for CIs in labora- combustor is typically less than 10−8 times the
tories over the last hundred years but its main goal combustor power. Predicting precisely the acoustic
is to discuss what must be added to these elements field created by a flame is much more difficult than
when real engines are considered. simulating the flame itself [39,40]. Even if such a
Thermoacoustic CIs are due to coupling mech- small conversion factor from mechanical energy
anisms between unsteady combustion and acoustic to sound leads to a relatively small level of energy
waves propagating in the chamber and reflecting contained in the acoustic field, these acoustic
on its walls, inlets and outlets.2 The left image fluctuations can have a strong effect on the flames
of Fig. 1 shows a NASA rocket engine partially themselves, closing a resonant feedback loop which
destroyed after the engine encountered CI while is difficult to predict: capturing flame / acoustic
the right picture displays high-speed views of the coupling to predict self-sustained instabilities is
flow in a laboratory premixed burner for a stable one of the overarching simulation problems in the
and an unstable regime [38]: the instability changes combustion community.
the flow drastically, creating mushroom-shaped The basic mechanisms leading to combustion
vortices at a frequency of 530 Hz and a very short instabilities were identified 150 years ago by Lord
and intense turbulent flame that can destroy the Rayleigh [41] but they have become real research
chamber in a few minutes.3 topics as well as practical dangers for many in-
From a fundamental point of view, CIs consti- dustrial programs when the power density of com-
tute one of the most challenging problems in fluid bustion chambers has increased sufficiently, first in
mechanics: they combine turbulence, acoustics, rocket [24] and later in jet engines [19,42,43]. Stud-
ies of combustion instabilities and noise are numer-
2 Most unconfined flames do not exhibit strong com- ous [24,26,44–53] and started long ago [41]. A flame
bustion instabilities. is not needed to produce such coupling: as shown
3 In a real combustor, this can be much faster: a fighter- by Rijke [54,55], a heated gauze placed in a tube is
aircraft engine submitted to screech, a strong CI mode, enough to produce a “singing” tube caused by the
or a rocket engine where a transverse mode grows can ex- coupling between the acoustic modes of the duct
plode in a few seconds. and the unsteady heat released by the gauze [56,57].

4 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

However, when the heat release is due to a flame, are relevant for annular chambers but to other
more energy can be transmitted into the acoustic engines as well (Section 4): the coupling between
field and the effects of combustion instabilities are wall heat transfer and CIs (Section 4.1), the recent
much more dangerous. discovery of intrinsic acoustic modes (Section 4.2),
CIs are also known under other names such as the need to consider impedances to analyze CIs
thermoacoustics in flames or combustion dynam- (Section 4.3, which is critical for gas turbines),
ics: the Hottel lecture of S. Candel in 2002 [35] pro- recent results on hydrodynamic stability of swirled
vides a precise history of CI research and of meth- flows (Section 4.4) and the existence of flame
ods used to control them. The books of Lieuwen bifurcations in swirled combustors that can be
and Yang [36] and Culick [19] describe CI physics triggered by CIs (Section 4.5). Finally, the need to
in multiple real combustion systems. The objective introduce UQ (Uncertainty Quantification) for CI
of the present lecture is not to repeat these excel- studies is the topic of Section 4.6.
lent reviews but to complement them by discussing
important progress achieved in the last 10 years in
the field of CI modeling and simulation. A major 2. Modeling and computing combustion
revolution in this domain has been the introduc- instabilities
tion of LES methods for CI computations. LES has
become one essential tool to analyze CIs but, as 2.1. A model problem illustrating key features of
shown in the next sections, it is not sufficient to fully combustion instabilities
analyze CI: like experiments, LES can tell whether
a given combustor will be unstable but it does not Most CIs are due to a resonance between un-
tell why this is so and how we can control this insta- steady combustion processes and acoustic waves
bility. To understand why instabilities appear and propagating in the combustion chamber. The
how to control them, other approaches are needed mechanisms leading to an amplification of com-
such as theory and simplified simulation tools (lin- bustion/acoustic processes are best explained by be-
earized Euler equations, for example), which will be ginning with the simple model problem described
discussed in the next sections. in Fig. 2. Consider a constant cross-section duct
The present paper focuses on gas turbines: they where a flame is stabilized at the dump plane sepa-
provide excellent examples of CIs that require rating the injection tube (length l1 , section S1 ) and
more studies than laboratory systems because they the combustion chamber (length l2 , section S2 ).
lead to completely new physics. First, a simple For low-frequency longitudinal modes, planar
pedagogical model of dump combustor is de- acoustic waves propagate both in the injection tube
scribed to recall classical coupling mechanisms and the chamber so that the fluctuating acoustic
between flames and acoustics and introduce stabil- pressure ( pi ) and velocity (ui ) signals in these two
ity criteria (Section 2.1). Simulation methods for ducts (numbered i = 1 to 2, x = 0 corresponds to
CIs are presented in Section 2.2, before applying the flame position) are:
them to the specific case of gas turbine engines
in Section 3, which discusses recent LES and 1
experimental results on annular chambers. The ui (x, t) = Re A+ i e
jki x
− A− i e
− jki x − jωt
e (1)
ρi c i
configurations that will be presented, are typical
of gas turbine engines (Section 3.1) and they pi (x, t) = Re A+ i e
jki x
+ A− i e
− jki x − jωt
e . (2)
exhibit azimuthal modes that cannot be observed
in usual laboratory set-ups. Recently, theory has where ki = ω/ci is the wave number in duct i, ω the
been introduced with significant success to analyze pulsation and ci the sound speed in duct i. To de-
these modes (Section 3.2). CI control methods termine the acoustic wave amplitudes A− + −
1 , A1 , A2
+
in annular chambers have also become a topic in and A2 , boundary conditions are imposed at the
itself, that is discussed in Section 3.3. After this inlet (usually u1 = 0 because velocity is imposed)
section, the presentation opens to new topics that and at the outlet (usually p = 0 for chambers open

Fig. 2. A simple pedagogical model for combustion instabilities: a laminar flame stabilized at the dump plane separating
the injection duct and the combustion chamber. The color field is the velocity modulus [58]. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 5

to the atmosphere where pressure is constant). At the mode starts growing, but the previous demon-
the dump plane where the flame is stabilized, jump stration shows that the essence of CI can be de-
conditions from one side of the flame to the other scribed in only a few lines.
allow to relate pressure and velocity perturbations The general solution of Eq. (6) is difficult to
on both sides of the flame front, assuming that the express but an analytical expression is easily ob-
flame is compact compared to the acoustic wave- tained in a simplified case where the tubes have the
length. Through such a compact flame, pressure same sections and lengths (S2 = S1 and l2 = l1 = a)
perturbations are conserved while the unsteady vol- and the flame induces a negligible heat release so
ume flow rate increases because of the total un- that ρ2 = ρ1 and c2 = c1 = c. In this case, = 1
˙ [49,59,60]:
steady heat release in the flame and the dispersion relation becomes cos2 (ωa/c ) −
sin2 (ωa/c )(1 + ne jωτ ) = 0. In the absence of flame
p2 (x = 0, t) = p1 (x = 0, t) and (n = 0) the solution of the dispersion equation sim-
γ −1 ply corresponds to the acoustic eigenmodes of a
S2 u2 (x = 0, t) = S1 u1 (x = 0, t) +
˙ (3) duct of length 2a: the first mode is such that kol =
ρ1 c21
π /4. It has a zero growth rate ((ko ) = 0) and a
where ρ j is the mean density in section j and γ the wave length λ0 = 2π /ko = 8a (four times the total
˙ length of the duct 2a) explaining why this mode
ratio of specific heats. A convenient scaling for
is called the quarter-wave mode. Its period is To =
is to express it as a function of the chamber in-
2π /ωo = 8a/c.
let velocity fluctuations u1 , as proposed by Crocco
If the flame is active and n is non zero but still
[24] who introduced an interaction indeed n (mea-
small, the solution for k can be written as a Taylor
suring the strength of the flame response) and a
expansion around ko so that k = ko + k with:
time delay τ (measuring the time required by the
flame to react to forcing): n
Re(k) = π /(4a ) − cos(2π τ /To ) and
γ −1 4a
˙ = nu1 (x = 0, t − τ )
(4) n
ρ1 c21 S1 (k) = (k ) = − sin(2π τ /To ) (7)
4a
so that, assuming harmonic variations for all per-
Since n is small in this approach, the pulsation of
turbations f = fˆe− jωt , jump conditions become: the mode (Re(k)/c) is only weakly affected by the
flame effect: the unstable mode frequency remains
pˆ 2 (x = 0, t) = pˆ 1 (x = 0, t) and close to the quarter-wave frequency without ac-
S2 uˆ2 (x = 0, t) = S1 uˆ1 (x = 0, t)(1 + ne jωτ ) (5) tive flame. The active flame (n = 0), however, con-
trols the growth rate of the mode: the combustor
Eq. (4) is the heart of most CI models: it allows will be unstable if (k) > 0, which implies here
linking heat release fluctuations (due to convective sin (2π τ /To ) < 0 or (p + 1/2)To < τ < (p + 1)To
and chemical effects) to a single acoustic velocity where p is an integer. This instability criterion in-
at the chamber inlet (x = 0). Once it is accepted, dicates that certain values of the flame delay τ will
there is no need to solve for any other mechanisms lead the flame to instability. For the first mode ( p =
except than acoustics. The whole problem of CI be- 0), the delay τ must be larger than the half-period
comes an acoustic problem only, and Eq. (3) to- To of the first acoustic mode and less than To :
gether with boundary conditions at the inlet (con-
stant velocity at x = −l1 , which imposes A+ jki l1
− 1
1e To > τ > To (8)
− − jki l1 2
A1 e = 0) and at the outlet (constant pressure
at x = l2 , which imposes A+ 2e
jki l2
+ A− 2e
− jki l2
= 0) Even if the assumptions used to derive this sta-
leads to an homogeneous equation for the wave am- bility criterion are crude, this analysis contains all
plitudes A− + − +
1 , A1 , A2 and A2 , which has a non-zero
the ingredients of many low-order models used for
solution only if: thermoacoustics:
• It requires all convective and chemistry ef-
l2 l1 l2 l1
cos ω cos ω − sin ω sin ω fects to be modeled as a function of a purely
c2 c1 c2 c1 acoustic quantity (which must be either pres-
ρ c
2 2 1S sure or acoustic velocity). Here the Crocco
× (1 + ne jωτ ) = 0 with = (6)
ρ1 c 1 S 2 model was used where the unsteady reac-
tion rate ˙ is expressed as a function of
Eq. (6) is a dispersion relation for ω: the real part of the acoustic velocity at the chamber inlet
ω is the pulsation of the mode that will occur in a CI u1 (x = 0, t). More sophisticated models may
oscillation; its imaginary part provides the growth be found in the literature [6,61,62]. Most of
rate. If it is positive, this model predicts that the them assume that ˙ depends on previous
mode will be linearly amplified, leading to CI. Of values of the reference velocity u1 (x = 0).
course, this linear approach cannot predict the am- This dependence is usually expressed through
plitude of the limit cycle that might be reached after a flame transfer function (FTF) F, depending

6 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 3. The structure of the 1/4 and 3/4 wave modes in the model of Fig. 2.

on the pulsation ω: approach in a one-dimensional case. More sophis-

ticated methods can be developed in three dimen-
¯˙ = F (u (x = 0, t∗; t∗ < t)/ū(x = 0) )
˙ / sions, in time or Fourier space but they all share the
1
(9) same basic idea: avoiding the complexity of flow
and chemistry by lumping all their effects into some
The FTF can also depend on the ampli- form of FDF.
tude of the oscillation A in which case it is FTFs and FDFs required for TA modes can be
called an FDF (Flame Describing Function) obtained analytically in simple cases [9,40,73], or
[63–65]. In many combustors, the fresh gases experimentally [64,74]. To complement these ap-
velocity may not be the only quantity affect- proaches, a second class of methods was intro-
ing unsteady combustion. The fluctuations duced around 2000 to compute explicitly the flame
of equivalence ratio φ at the chamber in- dynamics, using full LES of the forced reacting flow
let have been identified as another important [14,75,76]. This is more expensive and it raises vari-
control parameter [66–68] so that present ex- ous difficulties to handle acoustic boundary condi-
pressions for FDF are often written as: tions, chemistry, turbulence... LES can be used for
CI studies in two modes: brute force LES consists
˙¯ = F (u1 (x = 0, t)/ū(x = 0), φ /φ̄, A )
˙ / in setting a computational domain as large as pos-
(10) sible (e.g. accounting for all geometrical parts of
the engine), matching all boundary conditions (in-
• The previous derivation was performed for
cluding impedances at exits and inlets) and letting
the first acoustic mode (1/4 wave) but other
the LES solver compute the self-excited instabili-
higher-order modes can be amplified too. In
ties of the combustor. The second method (called
the model of Fig. 2, those are the 3/4, 5/4...
forced open-loop LES) uses LES only to compute
modes. In most chambers, only the lowest or-
the FTFs of a given flame and provide this informa-
der acoustic modes are expected because the
tion as input data for TA codes. Both approaches
acoustic dissipation increases rapidly with
will be used in the present review.
mode order and frequency.
• It leads to a stability criterion that depends
on τ (and weakly on n) in most cases. When 3. Azimuthal modes in gas turbines
k is determined, the mode structures (i.e. the
dependence on p and u on spatial coordi- While laboratory studies have been mostly lim-
nates) can be obtained too. As an example, ited to cylindrical or parallepipedic combustion
Fig. 3 displays the structure of the first two chambers fed by a single burner, real systems such
modes (1/4 and 3/4 wave). as gas turbines use annular geometries for combus-
tion and feed them with multiple burners (N = 10
2.2. Classification of simulation methods for
to 24). From a thermoacoustic point of view, this
combustion instabilities
introduces two new types of physics:
Two main classes of techniques are used to sim- • Since the combustion chamber is an annu-
ulate combustion instabilities (Fig. 4): the first cat- lus, azimuthal acoustic modes due to acoustic
egory; and the only one until the 2000s, is thermoa- waves traveling along the two azimuthal di-
coustic codes (called TA here) in which flames are rections (clockwise (CW) and anticlockwise
not simulated but replaced by their equivalent FTF (ACW)) can become unstable.
or FDF [69–72]. The mean flow is frozen and so- • Instead of considering the response of a sin-
lutions are sought for the linearized perturbations. gle burner to longitudinal acoustic modes,
The toy model of Fig. 2 is an example of such an annular chambers require to understand how

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 7

Fig. 4. Simulation methods for combustion instabilities.

N burners can couple with the acoustic in this direction. These efforts include LES but also
field. new experiments and theory.

3.1. Azimuthal instability modes in annular

Since the perimeter of most annular chambers chambers
is of the same order as their length, azimuthal and
longitudinal modes develop in similar frequency Five years ago, the development of powerful
ranges [34,36,59,71]: being able to distinguish these LES techniques for reacting flows [14,76,82,83] ap-
modes by looking at their frequencies only is im- plied to full annular combustors [84,85] confirmed
possible. A second specificity of azimuthal modes that azimuthal modes could change nature ran-
is their nature: the acoustic waves developing in domly, evolving from spinning to standing struc-
the annular chamber can be standing (with certain ture at random instants. Fig. 5 displays an exam-
burners being submitted to zero pressure fluctua- ple of annular geometry (only one sector with one
tions at nodes while others located near antinodes burner is shown for clarity) and instantaneous pres-
experience large pressure oscillations) or turning (in sure and velocity fields for an unstable helicopter
which case, all burners in the chamber see the same engine configuration. The pressure field is turning,
pressure fluctuation levels). This has been recog- modifying combustion in each sector and feeding
nized and discussed in many past studies since the the instability mechanism.
works of companies like Siemens [34] or Alstom Interestingly, LES showed right away that the
[77], who showed that both spinning and standing mode nature was changing with time: in the same
azimuthal modes were observed in an annular gas LES, without any parameter change, the mode
turbine. In 2002, Krebs et al. [34] showed experi- would alternate between standing and turning
mentally that the modes identified in a real gas tur- characteristics. After LES revealed that azimuthal
bine were sometimes turning, sometimes standing modes could be captured numerically, new experi-
and could transition from one state to another for ments were also developed [30,86], confirming LES
the same operating point. results and showing that azimuthal modes could be
Even if the question of the mode structure reproduced in a laboratory environment and that
(standing vs turning) is an exciting one from the they were indeed intermittent and switching from
point of view of stability and chaos theories, the one type to another. Fig. 6 shows the experimen-
practical question is more to know how to elimi- tal configuration of Cambridge [30,87] (left im-
nate these modes rather than to understand them. age) and an example of structure analysis (right
In the 1990s, active control was shown to be an ef- image). This analysis uses multiple microphones
fective way to control unstable modes in combus- located along the azimuthal direction to measure
tion chambers [78–80], including annular chambers the acoustic waves amplitudes turning clock wise
[81]. However, the cost and the certification difficul- (called A−) and anticlockwise (A+). When one of
ties have shown that it was more interesting to build these two waves dominate the other, the mode is
combustors that would be stable by design rather turning. When the two waves have similar ampli-
than trying to control them with active systems. The tudes, the mode is standing. The scatter plot of
next sections describe some of the recent theories Fig. 6 reveals a wide distribution of the proba-

8 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 5. LES of the first azimuthal unstable mode in an helicopter engine [84,85] . Left: geometry of a single sector. Right:
two snapshots of pressure on a cylinder passing through the burner axis and isosurfaces of temperature colored by axial
velocity.

Fig. 6. Cambridge annular chamber [87,92]. Left and center: configuration. Right: joint pdf of mode state in terms of
clockwise and anticlockwise wave amplitudes.

bility density function of the chamber state de- No analytical method or simulation has been able
fined by (A−, A+) doublets. Various theories have to predict this mode for the moment.
been proposed to understand this type of results Studying azimuthal modes in annular chambers
[77,88–91] but the issue is still open. requires to investigate a new generic problem:
Experiments rapidly raised additional ques- the response of flames to transverse velocity fluc-
tions: for example, triggering azimuthal modes tuations. Indeed, these modes create oscillating
proved to be difficult because longitudinal modes velocities that are normal to the flow (unlike usual
would often grow faster than azimuthal modes. longitudinal modes). Such a transverse forcing can
Furthermore, unexpected factors such as the re- actually be created in laboratory experiments that
spective lengths of the chamber inner and outer are similar to the ones used for FTF of flames
tubes seemed to control the existence of azimuthal submitted to longitudinal waves, except that, for
modes. transverse forcing, waves are produced by lateral
Spinning and standing modes were not the loudspeakers to induce velocity fluctuations nor-
only modes that were discovered experimentally. A mally to the flow axis [20,94–97]. These set-ups
’slanted’ mode was reported by the EM2C group demonstrate that the flame response depends
[31] in an annular chamber fed by matrix grids. This on the nature of the flame position in terms of
mode at 450 Hz is a combination of two modes with acoustic field. If the flame is located near a pres-
coinciding frequencies, the first one being a stand- sure antinode, it will sense mainly axial flow rate
ing azimuthal mode and the second one, an axial oscillations and its response will be similar to the
mode. Half of the flames (foreground of Fig. 7) one obtained though axial forcing. If the flame
are oscillating in a limited way and remain stuck sits at a velocity antinode, it will be submitted to
to the matrix injection systems while the other side strong transverse movements, which have a limited
(background of Fig. 7) are more unstable and lift effect on the unsteady heat release when flames
from the matrix: burners like number 5 or 13, for do not interact (since the flame is only oscillating
example, exhibit very different mean flame shapes. around it mean position). Neighboring flames

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 9

Fig. 7. EM2C annular chamber [31,93]. Left: configuration. Right: slanted mode visualization. In the foreground (burner
13), flames are stuck on the grids. In the back (burner 5), the flames are blown off at the periphery of the grids.

interaction (something that cannot be studied with conditions for velocities. It is a direct extension of
the experiments of Lespinasse et al. [94,98] or the model of Section 2.1: here, N burners (instead
of O’Connor et al. [95]) might lead to stronger of a single one for Section 2.1) feed a 1D annular
pulsations of heat release and can be studied only chamber (Fig. 8). In the chamber, between two
in full 360 degree combustors [30]. burners, simple acoustic propagation takes place
with two co- and counter- rotating waves. At the
3.2. Analytical methods for azimuthal instability junctions between chamber and burner, jump
modes conditions can be written. The length and section
of the ith burner are noted Li and Si while the
A major limitation of both experimental and perimeter and the cross-section of the annular
LES studies in this field is cost. A second one is chamber are 2Lc = 2π Rc and Sc respectively.
that even if they allow us to capture azimuthal Points in the burners are located using the axial
modes, they do not provide information on mech- coordinate z where z = 0 designates the upstream
anisms and on possible control strategies. There- end and z = Li the burner/chamber junction.
fore, simpler tools (TA class of Fig. 4) are needed The ith compact flame location is given by the
to explore azimuthal CIs basic nature and this has normalized abscissa α = z f ,i /Li . An impedance
to be done in idealized configurations. Such tools Z is imposed at the upstream end of each burner
can be built using network approaches and fully (z = 0). Subscript c corresponds to the chambers
analytical methods [20,53,99,100]. Recently, ana- and subscript u to the unburnt gases upstream of
lytical studies have progressed in two directions: the flame in the burners. Unsteady combustion
1) Linear theories are based on network models is modeled using an FTF for each flame: in each
[99–102]. The acoustic-flame behavior is assumed burner, the unsteady heat release depends linearly
linear and modeled by a Flame Transfer Function on the acoustic velocity upstream of the flame in
(FTF) while major features of the configuration are the corresponding burner (Eq. (4)).
retained such as complex burners, including an an- For small values of n, a fully analytical solution
nular plenum and a chamber, taking into account can be obtained for the eigenmodes by a Taylor ex-
a mean azimuthal flow etc. These studies can de- pansion around the mode that exists in the absence
termine the stability of the configuration but also of active flame (ni = 0). Theory shows first that the
predict linear effects on mode structure. 2) Non- single most important parameter controlling stabil-
linear approaches usually based on Galerkin meth- ity is the set of the N coupling parameters i given
ods [77,91], where the configuration is reduced to by:
a simple annulus with zero or an infinite num-
ber of burners and no plenum, but the acoustic- 1 Si ρ 0 c0
i = tan(ku Li ) 1 + ni e jωτi (11)
flame behavior can be more complicated and inte- 2 Sc ρu0 c0u
grate non-linear effects using a Flame Describing
Function (FDF), allowing the investigation of limit where ku = cωu and (ni , ˜τ i ) are the interaction index
cycles. and the time-delay of the FTF for the ith flame. Eq.
As an example of the power of analytical tools, (11) corresponds to a case where Z = ∞ (the inlet
the ATACAMAC approach [99,101] is described velocity of the burners is fixed). If all burners are
here. ATACAMAC describes acoustic waves identical, all i ’s are equal.
propagation in an annular chamber as a network The expression of the frequencies of the two
of one-dimensional ducts where flames create jump first azimuthal modes (clockwise and anticlock-

10 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 8. Analytical model [99,101] to study unstable modes in annular chambers with a steady and uniform azimuthal flow
(constant Mach number Mθ ). Left: modeling a full turbine by a network of 1D elements. Right: replacing burners by a
translated impedance on the chamber walls.

wise) is: next section provides an example of application of

cc cc these equations to passive control using symmetry
fCW = − ( + S0 ) and breaking.
2Lc 4π Lc
cc cc
fACW = − ( − S0 ) (12)
2Lc 4π Lc
3.3. Passive control of azimuthal instability modes
where two quantities only control the frequencies using symmetry breaking
((f CW ) and (f ACW )) and growth rates ((f CW ) and
(f ACW )) of the two azimuthal modes: One attractive method to mitigate azimuthal
combustion instabilities is to avoid using burners
• the coupling strength , which is the sum of that are identical [33]. This is also called symme-
all coupling parameters: try breaking. It is a well known method to avoid
ρ 0 c0 instabilities in many systems (one of the reasons
= i=1 i
N
= for not having army walking over a bridge at the
2Sc ρu0 c0u
same pace in the field of structural mechanics).
× i=1 Si tan(ku Li ) 1 + ni e jωτi
N
(13) For an annular chamber, this means using burners
The coupling strength does not depend on that have different ni ’s and τ i ’s. There is clearly a
the pattern used to distribute burners along wide range of choices if one tries to have ’differ-
the azimuthal direction. ent’ burner types in the same engine. For the mo-
• the splitting strength S0 , which determines ment, most tests have been performed using only
the frequency split between the two modes two different burner types: Siemens engines, for ex-
f CW and f ACW . It is a function of the coupling ample, sometimes have two types of burners in the
parameters i of the N burners and (unlike same machine. Moeck et al [57] demonstrated ac-
) of the pattern used to distribute burners tive control in an annular chamber where flames
along the azimuthal direction: were replaced by electrically driven heating grids
and showed that they could damp azimuthal modes
N by breaking symmetries (in this case, this meant us-
4 pπ
S02 = 02 − A = i0 0j cos ( j − i) ing different electrical power in each sector).
N
i, j=1 To elucidate how symmetry breaking affects
= γ (2 p) × γ (−2 p) (14) azimuthal modes, LES or 3D TA codes are not
the best tools: a guide is needed to understand the
where γ (k) = N
i=1 i0 e− j2kπ i/N
is the kth physics before trying to simulate these mechanisms
Fourier coefficient of the coupling factor as- in details. Here Eqs. (12)–(14) provide a good exam-
sociated with the azimuthal distribution 0 ple of the power of analytical approaches. The fre-
and p is the mode order (usually only the quency of the clockwise mode (Eq. (12)) is fCW =
first azimuthal mode p = 1 is observed). If all cc
− 4πccLc ( + S0 ). The first contribution to f CW
2Lc
burners are identical, the splitting strength S0 is 2Lc , which is the frequency of the first azimuthal
cc
is 0, and the two azimuthal modes have iden-
mode in the chamber without active flames. The
tical frequencies and growth rates.
effect of the active flames is explicitly revealed in
Eqs.(12) –((14) ) can then be used to analyze the the following correction term + S0 : active flames
stability of any annular chamber as soon as its di- act collectively to increase the coupling strength
mensions, temperatures and FTFs are known. The independently of their positions. However, if burn-

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 11

Fig. 9. Analytical analysis of symmetry breaking in a 24 burners chamber. Evolution of the splitting pattern factor K as a
function of the angle separating groups of twin type 2 burners when 4 type 2 burners are mixed with 20 type 1 burners.

ers are different, their distribution along θ matters two pairs of burners are located on opposite sides
and is measured by the splitting strength S0 . of the chamber, maximize the splitting strength and
Eq. (12) also provides a remarkable result: it will promote instability.
shows that splitting modes (increasing the splitting
strength S0 ) is usually detrimental for stability. The
imaginary parts of f CW and f ACW (the growth rates 4. Recent progress on mechanisms controlling
of the two modes) are modified by the imaginary instabilities
part of the splitting strength S0 in opposite ways:
if one of them becomes more stable, the other one The last 10 years have revealed that multiple
becomes more unstable. This has been observed mechanisms had to be taken into account to ana-
in other studies for annular chambers [57,103] but lyze instabilities in real engines. The following sec-
also in other fields of physics [104,105]. Therefore, tions present recent results on the effects on CI of
it may be safer to try to mitigate azimuthal modes wall temperatures (Section 4.1), on the existence
by changing the coupling strength rather than of intrinsic CI mode that are not controlled by
the splitting strength [103]. the eigenmode of the chamber (Section 4.2), on
If only two burner types are used (with coupling the importance and determination of impedances
factors 1 and 2 ), this analysis can be extended (Section 4.3), on hydrodynamic stability results
because the expression of S0 becomes: (Section 4.4) and on the links between bifurcations
Splitting pattern
and CIs (Section 4.5).

S0 = 2K . ( 1 − 2 ) (15) 4.1. The effects of wall temperatures

Burner difference
The temperatures of a combustor’s wall can
where the reduced splitting strength S0 depends modify thermoacoustic instabilities in different
only on K, called a splitting pattern factor and on ways:
(1 − 2 ), which is fixed by the difference between
the two burner types coupling factors. • Walls cool down the burnt gases, decreas-
Eq. (15) allows us to predict whether modes will ing their temperature and therefore the local
split or not, and whether the resulting modes will be sound speed. As a consequence, the eigen-
stable or not. It also provides a guide to place burn- modes of the chamber can change and their
ers along the circumferential direction, in order to stability too. Adiabatic and non- adiabatic
maximize damping or avoid splitting. For exam- configurations exhibit different stability re-
ple, the distribution of four Type 2 burners with 20 gions: this is easily observed in simulations
Type 1 burners in a 24 sector machine can be done where changing the wall heat transfer condi-
in many ways (Fig. 9): if the four burners of type tion from adiabatic to isothermal is sufficient
2 are grouped two by two, the only parameter con- to trigger or damp modes [106,107]. This is a
trolling the pattern is the angle between the two rather obvious effect due to changes in sound
groups of type 2 burners. Eq. (15) gives the value of speeds and flame shapes which will not be
the splitting pattern factor K and shows that cer- discussed here.
tain patterns (like those obtained for = 75◦ or • Heat losses in the zones that are critical
255°) lead to a system where symmetry is broken for flame stabilization play a more interest-
but the mode still is degenerate (K = 0). On the ing role. Since these regions (flame holders
contrary, some patterns like = 165◦ , where the for example) control the flame roots, they

12 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 10. Demonstrating the effects of flame holder temperature on the stability of a laminar premixed Bunsen flame [113].
Left: a Bunsen flame stabilized on a water-cooled slot. Right: evolution of pressure oscillations and slot wall temperature
versus time. The cooling system is started at t =490 s.

also affect its response to perturbations [73]. [114] and by others [115]. In these first approaches,
This point is discussed here because it is of- the flame front was supposed to be perfectly an-
ten ignored in simulation tools (Fig. 4) even chored to the flame holder and unable to move.
though recent studies prove that it can be a The first authors who mentioned that the point
critical issue. where the flame is stabilized (the flame root) also
moves and can affect the flame FTF, were Lee and
It is well-known that heat losses introduced by Lieuwen [116] who proposed to separate the dy-
flame-holders or by walls close to the stabiliza- namics and therefore the FTF of an anchored flame
tion zone of flames induce strong changes on the into two contributions:
flame topology even in the absence of any insta-
bility [37,108]. For porous burners, the whole sta- • flame front contribution (the movements of
bilization process and the flame response to un- the flame when it is perturbed: this is an es-
stable perturbations is controlled by heat losses to sentially kinematic mechanism that can be
the porous plate [109–111]. Even for usual Bun- predicted with tools such as the G-equation)
sen burners, the temperature of the lateral walls • flame root contribution (the movements of
[8,112] plays a major role on flame stabilization. the point where the flame anchors when it is
Therefore, it is not surprising that instabilities are perturbed, which requires to solve the near-
also affected by the temperature field of the solid wall region where the flame touches the wall).
on which a flame is anchored: an example of wall
temperature effects on combustion instabilities was While the first contribution has been studied
given by Mejia et al. [113] who showed that the self- by many authors, the second contribution due to
excited mode of a laminar premixed flame stabi- flame root movements remains the weak part of
lized on a slot was directly controlled by the slot this approach because it required solving for the
wall temperature. This metal temperature was con- temperature field in both gas and solid. Following
trolled by liquid cooling and measured by a ther- the analysis of Rook et al. [117] (for flat flames),
mocouple. When the experiment is ignited, walls Cuquel et al. [9] derived a full model for anchored
are cold and the instability begins right away at a flames accounting for both flame root and flame
high level (110 dB). The wall temperature increases front dynamics. Fig. 11 illustrates these two mech-
slowly and when it does, the pressure oscillations anisms and shows how the stand-off distance be-
decrease. After 300 s, the walls are warm (close to tween flame holder and flame root was estimated
120 °C) and the instability has completely disap- experimentally by Mejia et al. [113] from a direct
peared. At 490 s, the liquid cooling system is acti- view of light emission in a slot stabilized premixed
vated: the wall temperature goes down again and flame.
the instability goes back to its initial level. This When the flame is submitted to acoustic fluctu-
demonstrates that the temperature of the wall plays ations (for example to estimate its FTF), perturba-
a strong role in the determination of the stability tions propagate along the flame front (flame front
characteristics of this flame. contribution) but the flame root moves too (flame
Why the flame-holder temperature changes the root contribution). The movement of the flame
stability of a combustor is not discussed often. In during an oscillation cycle is displayed in Fig. 12
most models, the flame-holders are supposed to be (left) while the movement of the flame root (marked
adiabatic and the flame is anchored on the flame by a cross) is displayed in Fig. 12 right.
holder itself. This allows a theoretical analysis of Mejia et al. [113] showed that accounting for the
the flame response to forcing using G-equation flame root dynamics allowed to explain the effects
formulations as proposed by Boyer and Quinard of the wall temperature on the flame stability: it

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 13

Fig. 11. Left: the two mechanisms contributing to the FTF of an inverted V-flame stabilized on a slot (from [9,113]). Right:
visualization of the stand-off distance between flame holder and flame root [113].

Fig. 12. Left: visualization of the flame movements for an inverted V-flame submitted to a 60 Hz forcing with a flame
holder temperature of 50 °C. The + symbol marks the flame root and the trajectories of the flame root are displayed for
three different temperatures of the flame holder (50, 90 and 150 °C) [113].

Fig. 13. Top: the classical paradigm for combustion instabilities (a resonance between the flame and the acoustic eigen-
modes of the chamber reflecting on its inlet and outlet). Bottom: (ITA) intrinsic thermoacoustic modes (a resonant mode
that does not involve any acoustic eigenmode of the chamber).

modifies the FTF sufficiently to transform a stable model of Fig. 2. These eigenmodes are due (for
into an unstable flame (and vice versa) and explains longitudinal oscillations) to acoustic reflections at
the observations of Fig. 10. the inlet and outlet characterized by their respec-
tive reflection coefficients R1 and R2 . This view
has many direct implications for the analysis of
4.2. Intrinsic acoustic modes instabilities:
The general picture used to study and mitigate
combustion instabilities today (Fig. 13) links os- • When a combustor is unstable, the usual pro-
cillations of heat release with acoustic eigenmodes cedure is to compute the acoustic modes
of the combustion chamber as introduced for the of the chamber and check whether the

14 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

frequency of the instability matches one of where q (q ≥ 1) is an integer giving the mode or-
the eigenmodes frequencies. der. The first ITA mode (q = 1) has a real pulsation
• To stabilize the mode, increasing acoustic ωr = πτ and a period T = 2τ . It is amplified if ωi is
losses at inlet and outlet is the usual route: di- positive which is the case when n ≥ nc = 1+ or, in

minishing R1 and R2 is supposed to diminish terms of the modulus of the FTF F(ω): F ≥ Fc =
the growth rate of the modes by increasing 1+
.
θ
losses. This first ITA mode is very different from usual
In 2014, the TU Munich and the Eindhoven thermoacoustic modes:
groups [118,119] indicated that another mechanism • Its stability is not controlled by the time delay
may be found in flames: intrinsic thermoacous- τ (as it was for the toy model: see Eq. (8)) but
tic (ITA) modes. The theory for ITA modes is rather by the FTF gain n, i.e. by the strength
simple and starts from the following question: of the flame response to acoustic perturba-
what would happen in the flame of Fig. 2 if both tion.
extremities would be perfectly anechoic (Fig. 13)? • Its period T is not linked to any acoustic pe-
According to the classical paradigm for CI, such riod of the combustor (that has no eigenfre-
a system would have no acoustic eigenmode and a quency in any case because it is terminated by
maximum level of acoustic loss: any perturbation anechoic sections on both sides). T is simply
created by the flame and propagating toward inlet equal to two times the flame delay τ .
or outlet would simply leave, thereby eliminating
possible resonances with the flame. Therefore the ITA modes have two additional properties: (1)
two acoustic waves A+ − they react to changes in boundary conditions dif-
1 and A2 would be zero.
In practice, this is not exactly what theory says. ferently from usual thermoacoustic modes and (2)
Starting from the equations of the toy model they can interact with usual thermoacoustic modes.
(Eq. (3)) and using A+ − For example, adding acoustic dissipation at in-
1 = A2 = 0 does not lead to
an impossible solution but to : let and outlet in a burner can make ITA modes
more unstable, a property that is totally unexpected
for classical acoustic modes. Hoeijmakers et al.
p2 (x = 0, t) = p1 (x = 0, t) so that [118] show for example a map of the modes location
A+ − + − in the (ωi , ωr ) plane for a toy model similar to Fig. 2.
2 + A2 = A1 + A1 (16)
They use two cases: on the left of Fig. 14, for a case
and where the ITA mode is stable (F ≤ Fc ), there is an
γ −1 unstable standing mode when R1 = 1 and R2 = −1.
S2 u2 (x = 0, t) = S1 u1 (x = 0, t) +
˙ (17) When the reflection coefficients of inlet and/or out-
ρ1 c21
let decrease, this mode becomes more stable and
so that: when R1 = R2 = 0, the system reaches the condi-
+ tion where the ITA mode may appear. Since the
A+ − −
2 − A2 = A1 − A1 (1 + θ F (ω) ) (18) mode is stable, it does not appear and the system
ρ2 c 2 S 1
behaves as expected: making the inlet and outlet
where = and the general expression θ F(ω)
ρ1 c 1 S 2 anechoic drives the system to stability. On the other
has been used to replace nejωτ in the Crocco equa- hand, if the ITA mode is unstable (F ≥ Fc , right
tion. The θ factor (θ = T2 /T1 − 1) corresponds to image in Fig. 14), the standing mode that is unsta-
the low-frequency value of the FTF and provides ble when R1 = 1 and R2 = −1 becomes an unsta-
a proper scaling for F(ω). Eq. (18) has a solution ble ITA mode when R1 and R2 vanish. In this sit-
when: uation, making inlet and outlet anechoic does not
1+ stabilize the system: it transforms the initially un-
1 + [1 + θ F (ω)] = 0 or θ F(ω) = − (19) stable standing mode into an unstable ITA mode.

The ITA instability criterion F ≥ Fc = 1+ θ
can
where θ = TT2 − 1. The solutions of Eq. (19) are a be explicited for a perfect gas withconstant molec-
1
set of modes of pulsation ω that must satisfy: ular weight where = ρ2 c 2 S 1
= T1 S1
, knowing
⎧ ∗
ρ1 c 1 S 2 T2 S2
⎨arg (F (ω) )=(2q − 1)π (q ∈ N ) that θ = T2
− 1:
T1
⎩ +1 (20)
|F (ω)|= 1+ 1 S2 T1
θ F ≥ Fc = = 1+ (22)
When the usual Crocco expression is used for the θ T2 /T1 − 1 S1 T2
FTF F(ω): θ F (ω ) = ne jωτ , Eq. (20) has an explicit In most flames the maximum values of the FTF
solution: gain F are known and are of order unity. ITA
⎧
⎪ (2q − 1)π modes will appear if the critical threshold Fc be-
⎨ωr =
τ (21) comes less than F. Eq. (22) shows that the ITA crit-
⎪ 1 n ical threshold Fc goes down when the section ratio
⎩ ωi = ln
τ 1+ between inlet duct and combustion chamber goes

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 15

Fig. 14. Example of trajectories of ITA modes for a toy model similar to Fig. 2 when the reflection coefficients of the inlet
and outlet vary (from Hoeijmakers et al.[118]). Left: stable ITA mode (F ≤ Fc ). Right : unstable ITA mode (F ≥ Fc ). The
color scale corresponds to the value of R21 + R22 and measures the separation from a perfectly anechoic system.

Fig. 15. First ITA mode structure for the configuration of Fig. 2 [120]. Solid line: theory (Eq. (23)). Symbols: DNS.

down (S2 /S1 decreasing) or when the temperature velocity fluctuations (modulus and phase) obtained
ratio goes up (T2 /T1 increasing): intense flames in by theory (solid lines) and by DNS. The structure
chambers with small section changes (strong con- of the first ITA mode can be obtained by injecting
finement) should be more prone to intrinsic insta- the ω expression (Eq. (21)) into Eq. (1) leading to:
bilities. This may explain why ITA instabilities have
not been observed very often up to now: they are | pˆ 2 | |uˆ2 | S1
=1 and = (θ|F | − 1 ) (23)
triggered when the power per unit volume goes up | pˆ 1 | |uˆ1 | S2
(high values of T2 /T1 ) or when the chambers vol- for modulus and:
ume diminishes. Since both effects are sought in π π
future engines, ITA might appear in real engines arg[ pˆ 1 ] = − x arg[ pˆ 2 ] = x
soon. Their study and control will require to think c1 τ c2 τ
π π (24)
differently compared to today’s state of the art: for arg[uˆ1 ] = − x arg[uˆ2 ] = x+π
example, ITA modes will respond in unexpected c1 τ c2 τ
ways to increased acoustic losses that will make for phases.
ITA modes even more unstable. A whole field of The agreement between theory (Eq. (23)) and
research is probably opening up in this domain. DNS is very good and confirms the expected nature
The exact mechanisms that trigger the unstable of the mode. Only acoustic propagation is observed
loop of ITA modes without feeding the acoustic downstream or upstream of the flame zone: the
chamber modes are not fully clear yet. Courtine phase unwraps at the sound speeds on both sides
et al. [58] used DNS of ITA modes in a laminar of the flame and the unsteady pressure is the same
flame similar to Fig. 2. They studied various con- everywhere, showing that the flame is the acous-
finements (S2 /S1 from 1.5 to 6) and showed that, tic source and that waves propagate from the flame
as expected, the smallest confinement ratios lead to zone without any reflection. No node is observed
unstable ITA modes. Fig. 15 shows the pressure and anywhere. The jump in unsteady velocity between

16 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 16. Unstable loop driving the first ITA mode of Fig. 2 [58]. .

the cold and the burnt gases SS1 (θ |F | − 1 ) is also At this time the velocity in the inlet duct is mini-
2
well captured even if hydrodynamic mechanisms in mum and the flame has to retract very rapidly to-
the flame zone induce unsteady velocities that can- ward the dump plane by the usual flame restoration
not be captured by theory. mechanism. This creates an acoustic wave propa-
The mechanisms controlling ITA modes are gating upstream and impinging on the corner. At
obviously present in the DNS but they also are this instant (time t4 ), mode conversion takes place
contained in the FTF formulation used for theory. at the corner, transforming the acoustic wave into
The FTF assumes velocity-sensitive flames: the a new vortical wave and closing the cycle. Mode
flame is modified when the inlet velocity of the conversion [122] is an important part of the unsta-
chamber is changing. For ITA modes, this concept ble loop: it transforms acoustic waves into vorticity
becomes a little bit more difficult to understand at the corners of the dump plane. All mechanisms
because there is no downstream acoustic wave take place between the dump plane and the extreme
entering the burner of Fig. 2 through the inlet position of the flame: downstream convection of
which is anechoic. Still, the flame oscillate. This the vortical wave created by mode conversion at the
point has been discussed in the literature and is dump plane followed by a fast acoustic propagation
still an open question [58,119,120]. leading to a new mode conversion. No acoustic re-
Independently of the FTF formulation neces- flection from the chamber inlet or outlet is involved:
sary to capture ITA modes, the mechanisms con- this was also verified by Courtine [58] by perform-
trolling the mode can be isolated from the DNS. ing the same simulation in a chamber where the
Fig. 16 displays four snapshots of the flow dur- lengths of inlet and combustion chamber were mul-
ing one unstable cycle (left) and the time evolution tiplied by two, leading to exactly the same mode.
of chamber pressure, reference point velocity (in Most studies on ITA modes have been theoret-
inlet duct) and total heat release. All time signals ical [119] or numerical [58] but up to now, exper-
are strongly non linear, even pressure, something imental work has been limited to the PhD work
unusual in most usual thermoacoustic instability. of Hoeijmakers in Eindhoven. To construct a set-
The chamber pressure and the reference point ve- up exhibiting an ITA mode, the difficulty is that
locity are perfectly out of phase as expected from inlet and outlet must both be as anechoic as pos-
Crocco’s relation Eq. (4) when the period of the sible to ensure that R1 = R2 = 0. This can be ob-
mode is twice the flame delay τ . Since the cham- tained by installing horns on inlet and outlet ducts
ber pressure and the reference velocity are also out but doing so perfectly is arduous especially on the
of phase (Fig. 15), the heat release and the chamber exhaust side where hot gases leave the chamber
pressure are in phase as expected from the Rayleigh and a heat exchanger is required to protect the ex-
criterion. The left part of Fig. 16 shows that the cy- haust duct. Fig. 17 shows the experiment of Hoei-
cle begins when a vortical perturbation (visualized jmakers, including a large horn at the flow inlet
by the Q criterion [121]) is initiated at the corners of and a set of laminar premixed flames in the cham-
the dump plane (instant t1 ). This vortical perturba- ber. ITA modes renew the classical view of ther-
tion travels along the flame front and increases its moacoustic modes in a combustion chamber when
surface (time t2 ). At instant t3 , the flame reaches its only one loop was present : the flame creates noise
maximum length and heat release is maximum too. which is reflected back to the flame. This classical

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 17

Fig. 17. Experimental configuration to study ITA modes for laminar flames (PhD of P.G.M. Hoeijmakers, 2014).

interpretation ignores ITA modes which are due measures the ratio of the acoustic wave entering the
to a resonant loop within the flame zone itself: in domain A− to the wave leaving it, A+ . Reflection
practice, thermoacoustics in a combustion cham- coefficients depend on a given axis orientation (to
ber involves two different loops, one associated to decide which wave is the incident and which one
the flame itself, controlled by the FTF, and an- is the reflected one): for example, if R is defined as
other one controlled by the geometry of the com- A− /A+ , Z is simply given by: R = (Z + 1)/(Z − 1).
bustor, in particular its inlet/outlet reflection coef- Knowing Z or R (as a function of frequency) at the
ficients. ITA modes (observed for zero reflection inlet or outlet of a combustion chamber is suffi-
coefficients) and cavity modes (observed without cient to study its stability in the linear domain. For
active flame) are decoupled. However, as soon as longitudinal waves, Z fully describes all relevant
these two loops start interacting (with non-null re- mechanisms taking place outside of the chamber.
flection coefficients and an active flame), they per- The need to integrate impedances into any
turb both cavity modes and ITA modes, making analysis of combustion instabilities has many
them potentially unstable. Perturbed cavity modes implications:
are the classical thermoacoustic modes, while per-
turbed ITA modes are still to be studied. How these • The stability of an isolated combustion
two types of modes interact in a real configuration chamber has no real meaning: the stability
is an open topic today. of a chamber depends on its own character-
istics but also on the impedances imposed on
4.3. Impedances all its inlets and outlets. The same chamber
tested in a given bench will behave in a differ-
As suggested by the analysis of Section 2.1, ent way when installed on the real engine for
all combustion instabilities are controlled by the example. This explains why studying insta-
acoustic behavior of the inlet and outlet of the bilities can become complicated: extracting a
combustion chamber: the impedances (or the re- chamber from an unstable engine to study it
flection coefficients) at the chamber extremities af- in the laboratory (with different impedances)
fect directly the frequency and the growth rates of will lead to different results. Vice versa, strug-
all modes. The notion of impedance is the simplest gling to stabilize a chamber on a laboratory
approach (in the linear regime) to characterize wave bench might be detrimental to the stability of
transmission and reflection at a given section (usu- the real engine.
ally at the inlet and outlet of the combustion cham- • The only meaningful approach to combus-
ber). The effect of all parts located downstream of tion instabilities is therefore to integrate the
this section are lumped in an impedance Z defined impedances of inlets and outlets into the
by: analysis, considering them as input data con-
1 pˆ trolling stability like equivalence ratio, total
Z= (25) flow rate or geometry. Here combustion sys-
ρ̄c uˆ tems can be split into two categories: (1) sys-
Impedances are complex numbers varying with tems where impedances are known or can be
frequency. They can also be expressed in terms determined reasonably well and (2) systems
of reflection coefficients R: at an outlet, using the where impedances are very difficult to evalu-
notations of Eq. (1), the reflection coefficient R ate, for example gas turbines where the inlet

18 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Table 1
Impedances (Z) of one-dimensional ducts.
Case Configuration Boundary Impedance
1 p
condition Z= ρ̄c u

Outlet 1/ Infinite duct Non reflecting 1

Inlet 2/ Infinite duct Non reflecting −1

Outlet 3/ Constant pressure p = 0 0

Inlet 4/ Wall u = 0 ∞

Inlet 5/ Choked nozzle u = 0 ∞

Outlet 6/ Choked nozzle u = 0 ∞

of the chamber corresponds to the outlet of can also be added, for example at the chamber inlet,
a compressor and the chamber outlet is the to control its impedance [130,131] either to damp a
turbine inlet. Determining the impedances of given CI mode or to reinforce it (Fig. 18). These
turbomachinery systems is still an open re- studies confirm the importance of inlet and outlet
search question today. impedances to predict CIs. This is a major difficulty
in real systems as shown in the next section.
4.3.1. The impedances of laboratory rigs
Inlets and outlets ducts in laboratory experi- 4.3.2. The impedances of compressors and turbines
ments used for CIs studies are normally designed In gas turbines, the presence of a compressor
to provide simple impedances (Tab. 1). Outlets and a turbine raises a new and unexpected difficulty
for example often correspond to a duct termi- to predict CIs. The impedances of these turboma-
nating into open atmosphere (case 3 in Tab. 1). chinery systems control the acoustic modes of the
A few other cases are relevant for combustion whole system and are required to predict CIs but
chambers installed in laboratories. Motheau et al. unfortunately they are usually unknown and diffi-
[123,124] showed that a choked nozzle at the inlet cult to measure. Only turbine companies and a few
of a chamber (Case 5) imposes a constant inlet ve- laboratories in the world can build benches where
locity (u = 0). When the chamber is terminated by the impedance of a compressor or of a turbine can
a choked nozzle, the simplest acoustic approxima- be measured with precision. In addition, compres-
tion [125] is to replace it by a wall (Case 6). This sors and turbines are not passive acoustic elements:
low-frequency evaluation can be replaced by more they can inject unsteady energy into the chamber
sophisticated approaches when the nozzle cannot on a number of frequencies, thereby exciting the
be considered as compact compared to the acous- combustion process itself.
tic wavelength [126,127]. The question of turbomachinery impedances
In real engines, Tab. 1 is rarely useful and needed to study CIs has many common as-
more complex impedances are required. In a gas pects with the problem of combustion noise
turbine, the only simple case is a choked chamber [132,133] and especially of noise transmission and
outlet which can be approximated by u = 0 to generation through turbine stages [126,134–140].
first order (Case 6). In a rocket engine, terminated This problem has been studied in details over the
by a large nozzle, this may also be an acceptable last 10 years as combustion noise has progressively
approximation. become a significant part of the overall noise of
For CIs, knowing impedances is critical but ma- aircraft and helicopters because the other sources
nipulating inlet and outlet impedances can also be of noise (jet noise, fan noise) have decreased. To
useful: modifying impedances on any side of com- predict combustion noise (Fig. 19), it is necessary
bustion chamber is a well-known method to mit- to build a model describing how much acoustic
igate combustion instabilities in academic systems. energy is transported from the chamber to the
Active control as developed in the 80s can be viewed atmosphere through the turbine stages. This task
as such a technique [78,128,129]. Passive systems is almost the same as predicting how much of this

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 19

Fig. 18. Examples of systems to adjust inlet impedances: left, a variable length inlet duct used by Cosic et al. [131] to force
transverse modes; right: a perforated plate with adjustable bias flow by Tran et al. [130] to inhibit modes.

Fig. 19. A combustion chamber in a gas turbine with impedances on inlet (compressor) and outlet (turbine).

Fig. 20. Transmission of acoustic (A+ +

1 ) and entropy (As ) waves through the turbine: the transmitted acoustic wave (A2 ) is
the combustion noise; the reflected acoustic wave (A−1 ) can excite mixed modes in the chamber.

energy is actually reflected back into the chamber, noise (A+2 ) when the entropy waves are acceler-
an information directly linked to the impedance ated within the turbine stages. To describe how
needed for CI studies. entropy and acoustic waves interact and propagate
The calculation of impedances of turbine (or within turbo machinery stages, the disk actuator
of compressors using the same methodology) can theory assumes that the stage is compact: jump
be performed with various levels of complexity. conditions, rigorously valid at zero frequencies, are
The impedance of a nozzle can be computed using used to link incident and transmitted waves. By
compact theories [138,141] (where the wavelength assembling jump conditions for rotor and stator
of the acoustic waves is supposed to be large stages, the impedance of a full turbine or of a
compared to the nozzle axial dimension) or, more compressor can be obtained [143,144].
recently, new analytical theories [126,127] that
provide nozzle impedances at all frequencies cor- 4.3.3. Entropy–acoustic modes
responding to longitudinal modes. A simple area The transmission of entropy waves through tur-
contraction however is a poor model for what is bine stages creates indirect noise that is propagated
taking place in a real turbine stage where strong downstream. During the same process, the entropy
flow deviations are created by vanes. Moreover, all waves also induce acoustic waves that are reflected
rotor stages also introduce enthalpy jumps. Cump- back into the chamber and create a new class of
sty and Marble [134,142] were the first to propose CI: entropy–acoustic modes. For these modes,
models to describe the impedances of stator stages reflected acoustic waves (A− 1 in Fig. 20) propagate
in the low-frequency limit (‘disk actuator’ theory). back into the combustion chamber and generate
These studies were motivated by indirect noise, a CIs that are not captured by usual thermoacoustic
mechanism that transforms hot spots (generated analysis because the acoustic field is not fed by
within the chamber: As in Fig. 20) into acoustic unsteady reaction rate (as supposed in Crocco’s

20 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 21. A mixed mode cycle (four instants separated by π /2) in an aircraft chamber terminated by a nozzle [124]. Left:
pressure fluctuations. Right: temperature fluctuations.

model, Eq. (4)) but by the acoustic reflection of mixed modes efficiently but this is not a general
entropy waves hitting the turbine. This mechanism, rule: for short chambers where the turbine is close
sometimes called mixed entropy-acoustic mode to the injector, or for chambers where dilution
[145–147], is specific to chambers terminated by jets can induce unsteady temperature fluctuations
area restrictions: academic chambers terminated when they mix with burnt gases, mixed modes
by a nozzle as well as real gas turbines chambers can be observed. For example, Motheau et al.
feeding turbine stages. Precise models are difficult [124] showed that a mixed mode was responsible
to construct for mixed modes because entropy for a strong CI at a frequency that does not match
waves are dissipated by turbulent mixing much any acoustic mode of the chamber in an aircraft
faster than acoustic waves in their travel from the configuration. An unstable cycle is displayed in
combustion zone to the turbine. Evaluating this Fig. 21: the temperature field shows how hot pock-
dissipation is complicated [148] because it depends ets of burnt gases are released behind the dilution
on the flow details within the combustor [145]. jets and propagate toward the outlet nozzle where
Entropy waves are often dissipated too fast to feed they create an upstream acoustic wave (Fig. 22).

Fig. 22. Emerson et al. [149] experiment: a turbulent premixed flame is stabilized behind a bluff body for two values of
the density ratio between fresh and burnt gases. Streamlines (left) and flame front position (right) for ρu /ρb = 1.7 (a and
c) and ρu /ρb = 2.4 (b and d).

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 21

4.4. Hydrodynamic stability swirled flows, the most well known is the PVC
(precessing vortex core) and it plays a role in
Even though most of this review has focused on many CIs. Experiments [15,151–153] as well as LES
acoustic mechanisms controlling CI, purely hydro- [12,14,154,155] showed that PVCs can trigger cer-
dynamic phenomena play an essential role too be- tain CI modes, disappear when the flame is ignited
cause they determine the flow sensitivity to acous- or appear only for certain flame positions. PVCs
tic forcing. Two aspects are especially relevant when can even appear intermittently during CI.
discussing instabilities observed in gas turbine en- The exact causality link between PVC and CI re-
gines: (1) the effects of the temperature (or density) mains mysterious in most cases, however. Are PVCs
differences between fresh and burnt gases and of a source of CIs or just a second-order phenomenon
confinement in real engines and (2) the specific in- ? To investigate this question, an important tool is
stabilities due to swirl. Both issues are critical in real the hydrodynamic stability analysis of swirled flows
engines. with density changes. Like Emerson [149], Terhaar
et al. [16] proved that density variations play a
4.4.1. Density ratio effects strong role on the linear stability of a swirled flow.
An important difference between laboratory ex- By forcing swirled flames using both experiments
periments and real engines is the temperature of the and linear stability analysis based on the mean flow
inlet gases: in laboratory systems, it is usually of the profiles, Oberleithner et al. [153,156] described how
order of 300 K while it is closer to 700 K or more in the flame reacts to forcing and saturates at large
real gas turbines. This increase of inlet temperature amplitudes, explaining the mechanisms that lead to
does not lead to a comparable increase on the adia- the variations of Flame Describing Functions with
batic flame temperature because of dissociations in amplitude. None of these experiments corresponds
the burnt gases so that the burnt gas temperature to self-excited CI but results demonstrate that lin-
remains in the range 1500–2500 K. Moreover, en- ear stability analysis will play a strong role in the
gines often run leaner than laboratory test rigs so future for CI investigations.
that the burnt gas temperature decreases. This im-
plies that the density ratio between fresh and burnt 4.5. Hysteresis and combustion instability induced
gases ρ u /ρ b which is of the order or 6–8 in labora- bifurcations
tory flames, decreases significantly in a real engine
to values smaller than 3. This observation would Most approaches for CIs view them as small per-
be of limited importance if the hydrodynamic sta- turbations imposed on a fixed mean flow. In reality,
bility of shear layers with density gradients would CIs can induce perturbations that are so large that
not vary when ρ u /ρ b decreases. In practice, it does the mean flow itself changes. This can occur in two
change significantly and shear layers with small val- ways:
ues of ρ u /ρ b exhibit additional sinuous modes that
• The level of perturbations can be large
are not observed in laboratory flames. A good ex-
enough to create a pulsating flame which, if
ample of such mechanisms was given by Emerson
averaged over time, is quite different from
et al. [149] using experimental data and linear sta-
the stable flame but returns to its stable state
bility analysis of a flame stabilized behind a bluff
if the instability is controlled. A simple ex-
body where the inlet temperature could be changed
ample was given in the bottom right im-
(using a vitiation system) to reach values of ρ u /ρ b
ages of Fig. 1 where mushroom-shaped vor-
as low as 2.
tices appear during CI but the flame returns
Emerson’s results suggest that real engines may
to its stable position when the instability
exhibit new hydrodynamic sinuous modes com-
disappears.
pared to usual laboratory flames. Once again, this
• A more complex situation can be observed
also demonstrates the difficulty of studying CIs of
where a CI mode will trigger a full change of
real engines in laboratory set-ups where the condi-
the mean flow state. This is observed in com-
tions are very different: using large values of the
bustion systems that are prone to hysteresis
density ratio ρ u /ρ b may simply lead us to ignore an
mechanisms and exhibit multiple states for
unstable hydrodynamic mode that is important in a
a fixed regime: CI induced oscillations can
real engine.
force the flame to transition from one state
A second specificity of real engines is the strong
to another. Flame flashback [157,158] can be
confinement effects compared with many labora-
one manifestation of such an interaction be-
tory experiments: this leads to flows that are more
tween CI and mean flow.
unstable than unconfined flows and where absolute
instabilities appear sooner [150]. This section discusses the second case, observed,
for example, in the complex swirled flames found in
4.4.2. Global stability of swirling flows gas turbines. It is well known that swirled flames
Many natural hydrodynamic modes appear can exhibit multiple stabilization states depend-
in swirled flows, with and without combustion. ing on small variations of geometry or of oper-
Among all helical, shear later disturbances of ating conditions [159]. However, even for a fixed

22 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 23. State change in a swirled laboratory flame due to a variation of inlet temperature Tin [163]. Left (Tin = 570 K):
the ORZ (outer recirculation zone) contains cold gases. Right (Tin = 660 K): the ORZ contains burnt gases.

Fig. 24. Burner configuration (left) and combustion chamber (right) [164].

geometry and for fixed operating conditions, Depending on the flame initialization strategies,
swirled flames can sometimes exhibit different the LES of Hermeth et al. [164] leads to two states.
shapes depending on the flame history. Hystere- The topology of both states (Detached and At-
sis is commonly observed and one of the phe- tached) can be visualized by plotting an isosurface
nomena that can control the transition between of temperature (T /Tmean = 1.3, Fig. 25). The iso-
the various states is CI: swirling flames are espe- surface is colored by the normalized axial velocity.
cially sensitive to hysteresis [160] because the recir- Velocities are non-dimensionalized by the bulk ve-
culation zones (either the central recirculation zone locity. In the Attached state (Fig. 25 left), the CRZ is
(CRZ) or the outer recirculation zones (ORZ)) on ignited, it contains burnt gases and the flame is sta-
which the flame stabilizes can contain cold or burnt bilized on the central hub of the burner. In the De-
gases, leading to (at least) two different stable states. tached state (Fig. 25 right), the CRZ is not ignited
Huang and Yang [161,162] performed an LES of and the flame is stabilized only by the ORZ, lead-
the dump-stabilized flame of Seo [163] where ORZ ing to a longer flame, weakly stabilized. Of course,
contain fresh or burnt gases depending on the in- temperature profiles for both states are totally dif-
let air temperature. They showed that the transition ferent as shown in Fig. 26 right. Obviously, NOx
from a flame where the ORZ is cold (left image in emissions would also be vastly different.
Fig. 23) to a flame where the ORZ is hot (right im- In practice, experimentalists know that these
age in Fig. 23) was due to the increased flame speed two states exist, even though it is quite difficult
induced by the higher inlet temperature. to perform measurements in these large cham-
No hysteresis is observed here (the transi- bers. This transition can be triggered by any small
tion takes place because the inlet temperature is change in operating conditions, by turbulent fluc-
changed) but similar state changes can also occur tuations or as tested by Hermeth et al. [164], by
for the same regime, leading to hysteresis if CIs oc- acoustic waves. This acoustic forcing may be due to
cur: Hermeth et al. [164] used LES to demonstrate a self-excited CI mode or to external forcing: Her-
that a turbine burner (typical of large power gas meth et al. [164] forced the air inlet of the combus-
turbine systems, Fig. 24) installed in an octogonal tor at various frequencies and amplitudes to investi-
laboratory chamber at Ansaldo Energia S.p.A can gate the response of the combustor state to acoustic
exhibit two stable states for one fixed regime and waves.
that CI can trigger a transition from one to the In the absence of forcing, both states can be
other. maintained for very long times. When acoustic

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 23

Fig. 25. Temperature iso-surface (T /Tmean = 1.3) colored by normalized axial velocity for the Attached state (left) and the
Detached state (right) [164].

Fig. 26. Left: bifurcation diagram for the Initially Attached and Initially Detached flames as a function of the forcing
amplitude. Right: mean temperature profiles for the Attached and Detached states.

forcing is applied, the Attached state is the most to manufacturing tolerances but also to wear-out
stable: even high levels of forcing cannot lead to a phenomena during operation), air temperature,
quenching of the ORZ and a transition to the De- fuel spray characteristics, wall temperatures. The
tached state as shown by the bifurcation diagram swirling flows used in most gas turbines are very
in Fig. 26 left. The Detached state, however, is sen- sensitive [142,165,166]: a minute change of ge-
sitive to acoustic waves and if the burner is forced ometry in a swirler is sufficient for combustion to
acoustically (at a frequency of the order of a few bifurcate from a stable quiet regime to an unstable
hundred Hz), the flame starts oscillating, invading one destroying the combustor in a few minutes.
the CRZ and finally stabilizing in the Attached state For solid rocket engines, out of ten (supposedly)
as shown by Fig. 27. This is obtained only for suf- identical engines, eight can be stable during tests
ficiently high pulsation amplitudes (at least 15% of and two unstable: identifying the source of this
the mean velocity) (Fig. 26 left). At low forcing am- variation is a critical and challenging question.
plitudes, the Detached state only oscillates around In the context of the introduction of alternative
its mean position. fuels (bio fuels for example or mixtures of gases) in
combustors, UQ becomes mandatory: is it possible
4.6. UQ (Uncertainty Quantification) for that by changing slightly the fuel composition or by
combustion instabilities mixing two fuels, a stable combustor might become
unstable?
An additional difficulty to predict CIs is the The UQ problem also extends to the simula-
effect of uncertain parameters: most CI codes tion tools themselves and puts a new constraint
provide a bi-modal answer (yes or no) to the on them: it is not enough to predict the stabil-
question ’is this burner stable or not?’ A major ity map of a given combustor (the domain where
question is then to know how robust this answer this combustor can be operated safely) any more,
is to uncertainties, in other words to determine it is also necessary to determine the precision as-
the probability that a mode will be stable or not, sociated to this prediction. Uncertainty sources are
taking into account the uncertainties on input linked not only to physical parameters (geometry,
parameters. CIs are sensitive to many parameters regimes, impedances) but also to modeling issues
that have unknown values or are even not identi- (mesh size, numerical scheme accuracy, multiplic-
fied: fuel composition, geometrical changes (due ity of sub-models).

24 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Fig. 27. Time sequence of normalized temperature field on the middle cut plane during the attachment sequence of the
Detached flame at a pulsation amplitude of 45% [164].

The first difficulty to apply UQ tools to ther- τ i of the 19 burners. All other parameters were sup-
moacoustics is the large number of uncertain posed to be fixed. Even for this reduced set of un-
parameters that must be included: numerical pa- certain input parameters, a 38 dimension space is
rameters (mesh, physical submodels, boundary still a very large one and an active subspace method
conditions) as well as physical parameters (ge- [168] was used to reduce this dimension before us-
ometry, chemistry, impedances). The well-known ing UQ analysis. A typical result is displayed in
’curse of dimensionality’ (i.e. the fact that too Fig. 28: all burners are supposed to be submitted
many parameters are uncertain) hits the field of to independent fluctuations of 5% on ni and 10%
thermoacoustics directly because most simulation on τ i which are typical of experimental uncertain-
tools (Fig. 4) are expensive and cannot be run when ties on FTFs (Dr D. Durox, private communica-
too many input parameters have to be changed. tion). A Monte Carlo method using 10000 samples
For LES, this will probably be impossible for a was used to build the pdf of the mode growth rate
long time. For TA codes, this is easier to imagine and, from this value, measure the probability that
and certain tests show the benefits of this exercise: the mode would be stable or not. The determinis-
a proper approach to demonstrate the expected tic value (white square in Fig. 28) predicts stabil-
results of UQ for CI is to replace 3D solvers by a ity. The UQ analysis, on the other hand, shows a
surrogate model. wide pdf with unstable samples. Changing ni ’s and
For azimuthal modes discussed in Section 3, a τ i ’s by only 5% and 10% leads to a wide range of
good surrogate model for TA codes is the theoret- growth rates, from −35˜s−1 to +15˜s−1 and an over-
ical approach presented in Section 3.2. Since this all probability of instability of 39%. It is a sober-
model provides a fully analytical expression of the ing observation indicating that small uncertainties
mode frequencies (Eq. (12)), it can be run for multi- on input data of the stability analysis (the 38 val-
ple input parameters at low cost. This was done re- ues used for ni and on τ i ’s) can lead to a result that
cently for an annular chamber, with a single plenum is almost useless: the deterministic result predicts
and 19 burners [167]. The parameters that were sup- stability but this result has a 40% probability to be
posed to be uncertain are the 38 parameters ni and wrong.
Please cite this article as: T. Poinsot, Prediction and control of combustion instabilities in real engines,
Proceedings of the Combustion Institute (2016), [Link]
JID: PROCI
ARTICLE IN PRESS [m;June 25, 2016;16:33]

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 25

Fig. 28. UQ analysis of the growth rate of the first azimuthal mode vs real frequency in an annular chamber taking into
account a 10% uncertainty on the 38 input parameters (ni and τ i for each burner).

5. Conclusions [4] T. Lieuwen, J. Fluid Mech. 435 (2001) 289–303.

[5] F. Boudy, D. Durox, T. Schuller, S. Candel, Proc.
Real engines offer an immense field of investiga- Combust. Inst. 33 (2011) 1121–1128.
tion for combustion instabilities because their com- [6] C.O. Paschereit, W. Polifke, B. Schuermans,
O. Mattson, J. Eng. Gas Turbine Power 124 (2002)
plexity reveals mechanisms that cannot be captured
239–247.
in academic laboratory set-ups. This review has fo- [7] V. Kornilov, R. Rook, J. ten Thije Boonkkamp,
cused on a few questions that are specific to gas tur- L. de Goey, Combust. Flame 156 (2009) 1957–1970.
bine engines such as the effects of heat losses, of [8] K. Kedia, A. Ghoniem, Proc. Combust. Inst. 34 (1)
flame bifurcations or of azimuthal modes that can (2013) 921–928.
be found only in annular chambers. Paradoxically, [9] A. Cuquel, D. Durox, T. Schuller, C. R. Acad. Sci.
as the complexity of the configuration increases Mecanique 341 (2013) 171–180.
when real engines are considered, this review shows [10] Y. Huang, H.G. Sung, S.Y. Hsieh, V. Yang,
that the importance of theory increases: linear J. Prop. Power 19 (5) (2003) 782–794.
[11] P. Weigand, W. Meier, X. Duan, W. Stricker,
stability analysis of swirled flows, acoustic net-
M. Aigner, Combust. Flame 144 (1–2) (2006)
work descriptions of azimuthal modes, uncertainty 205–224.
quantifications and symmetry breaking are manda- [12] S. Wang, V. Yang, G. Hsiao, S.-Y. Hsieh, H.C. Mon-
tory notions to tackle instabilities in real engines. gia, J. Fluid Mech. 583 (2007) 99–122.
Of course, LES has changed the field by allowing [13] W. Meier, P. Weigand, X. Duan, R. Giezendan-
high-fidelity computations of real engines but LES ner-Thoben, Combust. Flame 150 (1–2) (2007) 2–26.
alone cannot solve the overarching problem in this [14] L.Y.M. Gicquel, G. Staffelbach, T. Poinsot, Prog.
domain: understand CI to control them. This can Energy Combust. Sci. 38 (6) (2012) 782–817.
be achieved only by combining LES results with [15] M. Stohr, I. Boxx, C.D. Carter, W. Meier, Combust.
Flame 159 (8) (2012) 2636–2649.
careful experiments, theory and field data.
[16] S. Terhaar, K. Oberleithner, C.O. Paschereit, Proc.
Combust. Inst. 35 (2014).
[17] P.L. Therkelsen, J.E. Portillo, D. Littlejohn,
Acknowledgments S.M. Martin, R.K. Cheng, Combust. Flame 160 (2)
(2013) 307–321.
The research leading to these results has re- [18] M. Bauerheim, G. Staffelbach, N. Worth, J. Daw-
ceived funding from the European Research son, L. Gicquel, T. Poinsot, Proc. Combust. Inst. 35
Council under the European Union’s Seventh (3) (2015) 3355–3363.
[19] F.E.C. Culick, P. Kuentzmann, Unsteady Motions
Framework Programme (FP/2007-2013)/ERC in Combustion Chambers for Propulsion Systems,
grant agreement ERC-AdG 319067-INTECOCIS. NATO Research and Technology Organization,
The support of the Center for Turbulence Re- 2006.
search and of INCITE and PRACE programs, [20] J. O’Connor, V. Acharya, T. Lieuwen, Prog. Energy
which provided CPU resources, is also gratefully Combust. Sci. 49 (C) (2015) 1–39.
acknowledged. [21] Y. Mery, S. Ducruix, P. Scouflaire, S. Candel,
Comptes Rendus Mécanique 337 (6) (2009) 426–437.
[22] Y. Méry, L. Hakim, P. Scouflaire, L. Vingert,
References S. Ducruix, S. Candel, Comptes Rendus Mécanique
341 (1) (2013) 100–109.
[1] J.C. Oefelein, V. Yang, J. Prop. Power 9 (5) (1993) [23] L. Hakim, T. Schmitt, S. Ducruix, S. Candel, Com-
657–677. bust. Flame 162 (10) (2015) 3482–3502.
[2] S. Ducruix, T. Schuller, D. Durox, S. Candel, [24] L. Crocco, J. Am. Rocket Soc. 21 (1951) 163–178.
J. Prop. Power 19 (5) (2003) 722–734. [25] L. Crocco, S.I. Cheng, Theory of Combustion Insta-
[3] S. Ducruix, T. Schuller, D. Durox, S. Candel, Prog. bility in Liquid Propellant Rocket Motors, vol. Agar-
Energy Comb. Sci. 210 (2005) 179. dograph No 8, Butterworths Science, 1956.

26 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

[26] D.J. Harrje, F.H. Reardon, Technical Report Report [59] T. Poinsot, D. Veynante, Theoretical and
SP-194, NASA, 1972. Numerical Combustion, third ed., 2011.
[27] S. Hochgreb, D.J.C. Dennis, I. Ayranci, W. Bain- [Link]/elearning
bridge, S. Cant, in: Proceedings of ASME Turbo [60] M. Bauerheim, F. Nicoud, T. Poinsot, Combust.
Expo 2013—GT2013-9531, 2013. Flame 162 (1) (2015) 60–67.
[28] M.d.l.C. García, E. Mastorakos, A.P. Dowling, [61] K. Truffin, T. Poinsot, Combust. Flame 142 (4)
Combust. Flame 156 (2) (2009) 374–384. (2005) 388–400.
[29] J. Kopitz, A. Huber, T. Sattelmayer, W. Polifke, in: [62] T. Sayadi, P.J. Schmid, F. Richecoeur, D. Durox,
ASME GT2005-68797, Reno, NV, U.S.A., 2005. Phys. Fluids 27 (3) (2015) 037102.
[30] N. Worth, J. Dawson, Proc. Combust. Inst. 34 (2012) [63] N. Noiray, D. Durox, T. Schuller, S. Candel, J. Fluid
1–8. Mech. 615 (2008) 139–167.
[31] J.F. Bourgouin, D. Durox, J.P. Moeck, T. Schuller, [64] D. Durox, T. Schuller, N. Noiray, S. Candel, Proc.
S. Candel, Proc. Combust. Inst. 35 (3) (2015) Combust. Inst. 32 (1) (2009) 1391–1398.
3237–3244. [65] P. Palies, D. Durox, T. Schuller, S. Candel, Combust.
[32] W. Krebs, G. Walz, S. Hoffmann, in: 5th AIAA Flame 158 (10) (2011) 1980–1991.
Aeroacoustics Conference AIAA 99-1971, 1999. [66] T. Lieuwen, B.T. Zinn, Proc. Combust. Inst. 27
[33] P. Berenbrink, S. Hoffmann, in: ASME Turbo Expo (1998) 1809–1816.
2001. Paper 2001-GT-42, 2001. [67] T. Sattelmayer, J. Eng. Gas Turbine Power 125 (2003)
[34] W. Krebs, P. Flohr, B. Prade, S. Hoffmann, Com- 11–19.
bust. Sci. Technol. 174 (2002) 99–128. [68] A. Birbaud, S. Ducruix, D. Durox, S. Candel, Com-
[35] S. Candel, Proc. Combust. Inst. 29 (1) (2002) 1–28. bust. Flame 154 (3) (2008) 356–367.
[36] T. Lieuwen, V. Yang, Prog. in Astronautics and [69] S. Camporeale, B. Fortunato, G. Campa, J. Eng.
Aeronautics AIAA, vol. 210, 2005. Gas Turbine Power 133 (1) (2011).
[37] L.P.H. De Goey, J.A. van Oijen, V.N. Kornilov, [70] F. Nicoud, L. Benoit, C. Sensiau, T. Poinsot, AIAA
J.H.M. ten Thije Boonkkamp, Proc. Combust. Inst. J. 45 (2007) 426–441.
33 (1) (2011) 863–886. [71] C. Pankiewitz, T. Sattelmayer, J. Eng. Gas Turbine
[38] T. Poinsot, A. Trouvé, D. Veynante, S. Candel, Power 125 (3) (2003) 677–685.
E. Esposito, J. Fluid Mech. 177 (1987) 265–292. [72] A.P. Dowling, S.R. Stow, J. Prop. Power 19 (5)
[39] M. Ihme, H. Pitsch, H. Bodony, Proc. Combust. (2003) 751–764.
Inst. 32 (2009) 1545–1554. [73] K. Kedia, H. Altay, A. Ghoniem, Proc. Combust.
[40] S. Candel, D. Durox, S. Ducruix, A. Birbaud, Inst. 33 (2011) 1113–1120.
N. Noiray, T. Schuller, Int. J. Acoust. 8 (2009) 1–56. [74] S. Ducruix, D. Durox, S. Candel, Proc. Combust.
[41] L. Rayleigh, Nature July 18 (1878) 319–321. Inst. 28 (1) (2000) 765–773.
[42] A.A. Putnam, J.M. Beer, Fuel and Energy Science [75] P. Schmitt, T. Poinsot, B. Schuermans, K.P. Geigle,
Series, American Elsevier, 1971. J. Fluid Mech. 570 (2007) 17–46.
[43] F.A. Williams, Combustion Theory, Benjamin Cum- [76] C. Fureby, Flow Turbine Combust. 84 (2010)
mings, Menlo Park, CA, 1985. 543–564.
[44] E.W. Price, in: 12th Symp. (Int.) on Combus- [77] N. Noiray, M. Bothien, B. Schuermans, Combust.
tion, The Combustion Institute, Pittsburgh, 1969, Theory Model. 15 (2011) 585–606.
pp. 101–113. [78] G. Bloxsidge, A. Dowling, N. Hooper,
[45] M. Barrère, F.A. Williams, Proc. Combust. Inst. 12 P. Langhorne, AIAA J. 26 (1988) 783–790.
(1968) 169–181. [79] T. Poinsot, W. Lang, F. Bourienne, S. Candel, E. Es-
[46] W. Strahle, Prog. Energy Combust. Sci. 4 (1978) posito, J. Prop. Power 5 (1989) 14–20.
157–176. [80] G. Billoud, M. Galland, C. Huynh, S. Candel, Com-
[47] B.D. Mugridge, J. Sound Vib. 70 (1980) 437–452. bust. Sci. Technol. 81 (1992) 257–283.
[48] V. Yang, F.E.C. Culick, Combust. Sci. Technol. 45 [81] J. Seume, N. Vortmeyer, W. Krause, J. Hermann,
(1986) 1–25. C. Hantschk, P. Zangle, S. Gleis, D. Vortmeyer,
[49] D.G. Crighton, A.P. Dowling, J.E.F. Williams, A. Orthmann, J. Eng. Gas Turbines Power 120
M. Heckl, F. Leppington, Lecture Notes, Springer (1998) 721–726.
Verlag, New-York, 1992. [82] N. Gourdain, L. Gicquel, M. Montagnac, O. Ver-
[50] S. Candel, C. Huynh., T. Poinsot, in: NATO ASI Se- morel, M. Gazaix, G. Staffelbach, M. Garcia,
ries, Kluwer Academic Publishers, Dordrecht, 1996, J. Boussuge, T. Poinsot, Comput. Sci. Discovery 2
pp. 83–112. (2009a) 015003.
[51] A.P. Dowling, J. Sound Vib. 180 (4) (1995) 557–581. [83] N. Gourdain, L. Gicquel, G. Staffelbach, O. Ver-
[52] C. Martin, L. Benoit, Y. Sommerer, F. Nicoud, morel, F. Duchaine, J.-F. Boussuge, T. Poinsot,
T. Poinsot, AIAA J. 44 (4) (2006) 741–750. Comput. Sci. Discovery 2 (1) (2009b) 28pp.
[53] A.S. Morgans, S.R. Stow, Combust. Flame 150 (4) [84] G. Staffelbach, L. Gicquel, G. Boudier, T. Poinsot,
(2007) 380–399. Proc. Combust. Inst. 32 (2009) 2909–2916.
[54] P.L. Rijke, Phil. Mag. 17 (1859a) 419–422. [85] P. Wolf, G. Staffelbach, L. Gicquel, J.-D. Muller,
[55] P.L. Rijke, Ann. Phys. 107 (1859b) 339. T. Poinsot, Combust. Flame 159 (11) (2012)
[56] J. Moeck, M. Oevermann, R. Klein, C. Paschereit, 3398–3413.
H. Schmidt, Proc. Combust. Inst. 32 (2009) [86] J. Bourgouin, D. Durox, J.P. Moeck, T. Schuller,
1199–1207. S. Candel, in: ASME Turbo Expo 2013
[57] J. Moeck, M. Paul, C. Paschereit, in: ASME Turbo GT2013-95010, 2013.
Expo 2010 GT2010-23577, 2010. [87] N. Worth, J. Dawson, in: Proc. Combust. Inst., vol.
[58] E. Courtine, L. Selle, T. Poinsot, Combust. Flame 34, 2013, pp. 3127–3134.
162 (2015) 4331–4341.

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 27

[88] N. Noiray, B. Schuermans, Proc. R. Soc. A: Math. [117] R. Rook, L. de Goey, L. Somers, K. Schreel,
Phys. Eng. Sci. 469 (2151) (2012) 20120535. R. Parchen, Combust. Theory Model. 6 (2) (2002)
[89] B. Schuermans, C. Paschereit, P. Monkewitz, in: 223–242.
44th AIAA Aerospace Sciences Meeting and Ex- [118] M. Hoeijmakers, V. Kornilov, I.L. Arteaga, P. de
hibit, vol. AIAA paper 2006-0549, 2006. Goey, H. Nijmeijer, Combust. Flame 161 (2014)
[90] C. Sensiau, F. Nicoud, T. Poinsot, Int. J. Aeroacoust. 2860–2867.
8 (1) (2009) 57–68. [119] T. Emmert, S. Bomberg, W. Polifke, Combust.
[91] G. Ghirardo, M. Juniper, in: 2013-0232, 2013. Flame 162 (1) (2015) 75–85.
[92] N. Worth, J. Dawson, Combust. Flame 160 (11) [120] E. Courtine, L. Selle, F. Nicoud, W. Polifke,
(2013) 2476–2489. C.F. Silva, M. Bauerheim, T. Poinsot., in: Proc. of
[93] J.-F. Bourgouin, D. Durox, J.P. Moeck, T. Schuller, the Summer Program, Center of Turbulence Re-
S. Candel, J. Eng. Gas Turbine Power 137 (2) (2015) search, Stanford University/NASA, Ames, USA,
021503-11. 2014, pp. 169–178.
[94] F. Baillot, F. Lespinasse, Combust. Flame 161 (5) [121] F. Hussain, J. Jeong, J. Fluid Mech. 285 (1995)
(2014) 1247–1267. 69–94.
[95] J. O’Connor, T. Lieuwen, J. Eng. Gas Turbine Power [122] A.L. Birbaud, D. Durox, S. Ducruix, S. Candel,
134 (011501-9) (2012) 134. Phys. Fluids 19 (2007) 013602.
[96] J. Blimbaum, M. Zanchetta, T. Akin, V. Acharya, [123] E. Motheau, L. Selle, F. Nicoud, J. Sound Vib. 333
J. O’Connor, D. Noble, T. Lieuwen, Int. J. Spray (1) (2013) 246–262.
Combust. Dyn. 4 (4) (2012) 275–298. [124] E. Motheau, F. Nicoud, T. Poinsot, J. Fluid Mech.
[97] V. Acharya, M. Malanoski, M. Aguilar, T. Lieuwen, 749 (2014) 542–576.
J. Eng. Gas Turbine Power 136 (5) (2014) 051503-10. [125] F.E. Marble, S. Candel, J. Sound Vib. 55 (1977)
[98] F. Lespinasse, F. Baillot, T. Boushaki, C. R. Acad. 225–243.
Sci. Mec. 341 (1–2) (2013) 110–120. [126] M. Huet, A. Giauque, J. Fluid Mech. 733 (2013)
[99] J.-F. Parmentier, P. Salas, P. Wolf, G. Staffelbach, 268–301.
F. Nicoud, T. Poinsot, Combust. Flame 159 (7) [127] I. Duran, S. Moreau, J. Fluid Mech. 723 (2013) 190–
(2012) 2374–2387. 231, doi:10.1017/jfm.2013.118.
[100] S. Evesque, W. Polifke, C. Pankiewitz, in: 9th [128] W. Lang, T. Poinsot, S. Candel, Combust. Flame 70
AIAA/CEAS Aeroacoustics Conference, vol. (1987) 281–289.
AIAA Paper 2003–3182, 2003. [129] K. McManus, T. Poinsot, S. Candel, Prog. Energy
[101] M. Bauerheim, J.-F. Parmentier, P. Salas, Combust. Sci. 19 (1993) 1–29.
F. Nicoud, T. Poinsot, Combust. Flame 161 (5) [130] N. Tran, S. Ducruix, T. Schuller, Proc. Combust.
(2014) 1374–1389. Inst. 32 (2) (2009) 2917–2924.
[102] L. Li, X. Sun, Combust. Flame 162 (3) (2015) [131] B. Cosic, B. Bobusch, J. Moeck, C. Paschereit, J.
628–641. Eng. Gas Turbine Power 134 (2012) 061502-1.
[103] M. Bauerheim, P. Salas, F. Nicoud, T. Poinsot, [132] W.C. Strahle, J. Fluid Mech. 49 (1971) 399–414.
J. Fluid Mech. 760 (2014) 431–465, doi:10.1017/jfm. [133] A.P. Dowling, Y. Mahmoudi, Proc. Combust. Inst.
2014.578. 35 (1) (2015) 65–100.
[104] A. Davey, H. Salwen, J. Fluid Mech. 281 (1994) [134] N.A. Cumpsty, F.E. Marble, Proc. R. Soc. Lond. A
357–369. 357 (1977) 323–344.
[105] J. Guckenheimer, A. Mahalov, Phys. Rev. Lett. 68 [135] F. Bake, N. Kings, A. Fischer, R. I., Int. J. Aeroa-
(1992) 2257. coust. 8 (1–2) (2008) 125–142.
[106] A. Sengissen, J.F.V. Kampen, R. Huls, G. Stoffels, [136] F. Bake, C. Richter, B. Muhlbauer, N. Kings,
J.B.W. Kok, T. Poinsot, Combust. Flame 150 (2007) I. Rohle, F. Thiele, B. Noll, J. Sound Vib. 326 (2009)
40–53. 574–598.
[107] R. Garby, L. Selle, T. Poinsot, C. R. Acad. Sci. Mec. [137] M. Leyko, F. Nicoud, T. Poinsot, AIAA J. 47 (11)
341 (1–2) (2013) 220–229. (2009) 2709–2716.
[108] K. Kedia, A. Ghoniem, Combust. Flame 162 (4) [138] I. Duran, S. Moreau, T. Poinsot, AIAA J. 51 (1)
(2015) 1304–1315. (2013) 42–52.
[109] R. Raun, M. Beckstead, Combust. Flame 94 (1993) [139] M.S. Howe, J. Fluid Mech. 659 (2010) 267–288.
1–24. [140] W. Ullrich, T. Sattelmayer, in: 21st AIAA CEAS
[110] K.J. Bosschaart, L. de Goey, Combust. Flame 132 Aeroacoustics Conference, 2015, pp. 1–19.
(1–2) (2003) 170–180. [141] F.E. Marble, S. Candel, J. Sound Vib. 55 (1977)
[111] K.J. Bosschaart, L.P.H. De Goey, Combust. Flame 225–243.
136 (3) (2004) 261–269. [142] S. Candel, D. Durox, T. Schuller, J.-F. Bourgouin,
[112] F. Duchaine, F. Boudy, D. Durox, T. Poinsot, Com- J.P. Moeck, Ann. Rev. Fluid Mech. 46 (1) (2014)
bust. Flame 158 (12) (2011) 2384–2394. 147–173.
[113] D. Mejia, L. Selle, R. Bazile, T. Poinsot, Proc. Com- [143] I. Duran, M. Leyko, S. Moreau, F. Nicoud,
bust. Inst. 35 (3201–3208) (2014) 3. T. Poinsot, C. R. Acad. Sci. Mec. 341 (1–2) (2013)
[114] L. Boyer, J. Quinard, Combust. Flame 82 (1990) 131–140.
51–65. [144] M. Leyko, I. Duran, S. Moreau, F. Nicoud,
[115] M. Fleifil, A.M. Annaswamy, Z.A. Ghoneim, T. Poinsot, J. Sound Vib. 333 (23) (2014) 6090–6106.
A.F. Ghoniem, Combust. Flame 106 (4) (1996) [145] J. Eckstein, E. Freitag, C. Hirsch, T. Sattelmayer, J.
487–510. Eng. Gas Turbine Power 128 (2) (2006) 264–270.
[116] S.H. Lee, J.G. Ih, K.S. Peat, J. Sound Vib. 303 (3–5) [146] W. Polifke, C. Paschereit, K. Döbbeling, Int. J.
(2007) 741–752. Acoust. Vib. 6 (3) (2001) 135–146.

28 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

[147] C.S. Goh, A.S. Morgans, Combust. Sci. Technol. S. Menon, J.M. Seitzman, T.C. Lieuwen, Combust.
185 (2) (2013) 249–268. Sci. Technol. 186 (8) (2014) 1041–1074.
[148] A.S. Morgans, C.S. Goh, J.A. Dahan, J. Fluid Mech. [160] M.J. Tummers, A.W. Hübner, E.H.v. Veen, K. Han-
733 (R2) (2013), doi:10.1017/jfm.2013.448. jalić, T.H.v.d. Meer, Combust. Flame 156 (2) (2009)
[149] B. Emerson, J. O’Connor, M. Juniper, T. Lieuwen, 447–459.
J. Fluid Mech. 706 (2012) 219–250. [161] Y. Huang, V. Yang, Combust. Flame 136 (2004)
[150] M. Juniper, J. Fluid Mech. 565 (2006) 171–195. 383–389.
[151] A. Steinberg, I. Boxx, M. Stöhr, C. Carter, [162] Y. Huang, V. Yang, Prog. Energy Combust Sci. 35
W. Meier, Combust. Flame 157 (2010) 2250–2266. (4) (2009) 293–364.
[152] S. Terhaar, K. Oberleithner, C.O. Paschereit, Com- [163] S. Seo, Parametric Study of Lean Premixed Com-
bust. Sci. Technol. 186 (7) (2014). bustion Instability in a Pressurized Model Gas Tur-
[153] K. Oberleithner, M. Stöhr, S.H. Im, C.M. Arndt, bine Combustor, Department of Mechanical Engi-
A.M. Steinberg, Combust. Flame 162 (8) (2015) neering, Pennsylvania State University, 1999 (Ph.D.
3100–3114. thesis).
[154] S. Roux, G. Lartigue, T. Poinsot, U. Meier, C. Bérat, [164] S. Hermeth, G. Staffelbach, L.Y. Gicquel, V. Anisi-
Combust. Flame 141 (2005) 40–54. mov, C. Cirigliano, T. Poinsot, Combust. Flame 161
[155] A. Ghani, T. Poinsot, L. Gicquel, J.-D. Muller, (1) (2014) 184–196.
Flow, Turbine Combust. 96 (2016) 207–226, doi:10. [165] M. Vanierschot, E.V. den Bulck, Exp. Thermal Fluid
1007/s10494- 015- 9654- 9. Sci. 31 (6) (2007) 513–524.
[156] K. Oberleithner, S. Schimek, C.O. Paschereit, Com- [166] M. Falese, L.Y. Gicquel, T. Poinsot, Comput. Fluids
bust. Flame 162 (1) (2015) 86–99. 89 (2014) 167–178.
[157] J.O. Keller, L. Vaneveld, D. Korschelt, G.L. Hub- [167] M. Bauerheim, A. Ndiaye, P. Constantine, G. Iac-
bard, A.F. Ghoniem, J.W. Daily, A.K. Oppenheim, carino, S. Moreau, F. Nicoud, in: Proceedings of the
AIAA J. 20 (1981) 254–262. CTR Summer Program, 2014, pp. 209–210.
[158] Y. Sommerer, D. Galley, T. Poinsot, S. Ducruix, [168] T. Chantrasmi, G. Iaccarino, Int. J. Uncertainty
F. Lacas, D. Veynante, J. Turbine 5 (2004) 037. Quantif. 2 (2) (2012) 125–143.
[159] I. Chterev, C.W. Foley, D. Foti, S. Kostka,
A.W. Caswell, N. Jiang, A. Lynch, D.R. Noble,

Please cite this article as: T. Poinsot, Prediction and control of combustion instabilities in real engines,
Proceedings of the Combustion Institute (2016), [Link]

1 s2.0 S1540748916300074 Main
No ratings yet
1 s2.0 S1540748916300074 Main
28 pages
1 s2.0 S0010218016300372 Main
No ratings yet
1 s2.0 S0010218016300372 Main
12 pages
Large Eddy Simulation and Low-Order Modelling of A
No ratings yet
Large Eddy Simulation and Low-Order Modelling of A
13 pages
Simulation Methodologies and Open Questions For Acoustic Combustion Instability Studies
No ratings yet
Simulation Methodologies and Open Questions For Acoustic Combustion Instability Studies
10 pages
Combustion System Damping Augmentation With Helmholtz Resonators
No ratings yet
Combustion System Damping Augmentation With Helmholtz Resonators
6 pages
GT2002 30063
No ratings yet
GT2002 30063
12 pages
Applied Acoustics: Rogério Corá, Cristiane Aparecida Martins, Pedro Teixeira Lacava
No ratings yet
Applied Acoustics: Rogério Corá, Cristiane Aparecida Martins, Pedro Teixeira Lacava
10 pages
Onset of Entropy-Driven Combustion Instability in Turbulent Combustors
No ratings yet
Onset of Entropy-Driven Combustion Instability in Turbulent Combustors
25 pages
1 s2.0 S0010218016300797 Main
No ratings yet
1 s2.0 S0010218016300797 Main
10 pages
Combustion Instabilities PDF
No ratings yet
Combustion Instabilities PDF
33 pages
Experimental Study of Premixed Flame Oscillations in Pressures Above Atmospheric
No ratings yet
Experimental Study of Premixed Flame Oscillations in Pressures Above Atmospheric
11 pages
Instability in Rocket Engines
No ratings yet
Instability in Rocket Engines
13 pages
Thermo Acoustic Instabilities in The PRECCINSTA Combustor Investigated Using A Compressible LES Approach
No ratings yet
Thermo Acoustic Instabilities in The PRECCINSTA Combustor Investigated Using A Compressible LES Approach
17 pages
2019 Chen Stoehr CF Author Final
No ratings yet
2019 Chen Stoehr CF Author Final
32 pages
Annurev Fluid 121021 032828
No ratings yet
Annurev Fluid 121021 032828
25 pages
Pratt & Whitney PW1921G Geared Turbo Fan Engine Possible Resonance
No ratings yet
Pratt & Whitney PW1921G Geared Turbo Fan Engine Possible Resonance
10 pages
Harun Yilmaz, Omer Cam, Ilker Yilmaz: Full Length Article
No ratings yet
Harun Yilmaz, Omer Cam, Ilker Yilmaz: Full Length Article
14 pages
Damping Combustion Instabilities With Perforates at The Premixer Inlet of A Swirled Burner
No ratings yet
Damping Combustion Instabilities With Perforates at The Premixer Inlet of A Swirled Burner
8 pages
Combustion Instabilities Explained
No ratings yet
Combustion Instabilities Explained
16 pages
Azimuthal Instabilities in Annular Combustion Chambers
No ratings yet
Azimuthal Instabilities in Annular Combustion Chambers
11 pages
Thermo-Acoustic Analysis of Gas Turbine Burner
100% (5)
Thermo-Acoustic Analysis of Gas Turbine Burner
30 pages
Gas Turbine Combustor Thermoacoustics MARCO ZEDDA ROLLS ROYCE gtc15
No ratings yet
Gas Turbine Combustor Thermoacoustics MARCO ZEDDA ROLLS ROYCE gtc15
28 pages
Combustion Instabilities in Rocket Engines
No ratings yet
Combustion Instabilities in Rocket Engines
6 pages
1 s2.0 S0094576515300400 Main
No ratings yet
1 s2.0 S0094576515300400 Main
10 pages
Combustion Noise and Combustion Instabilities in Propulsion Systems
No ratings yet
Combustion Noise and Combustion Instabilities in Propulsion Systems
29 pages
Instability Prediction of GO /GCH Flames in A Single Recessed Coaxial Injector Using 1D Lumped Network Model
No ratings yet
Instability Prediction of GO /GCH Flames in A Single Recessed Coaxial Injector Using 1D Lumped Network Model
10 pages
Combustion Instability in Gas Turbines
No ratings yet
Combustion Instability in Gas Turbines
7 pages
Thermo-Acoustic System Modelling and Stability Analysis
100% (1)
Thermo-Acoustic System Modelling and Stability Analysis
49 pages
Combustion Dynamics in A Single-Element Lean Direct Injection Gas Turbine Combustor
No ratings yet
Combustion Dynamics in A Single-Element Lean Direct Injection Gas Turbine Combustor
29 pages
Final Version
No ratings yet
Final Version
22 pages
Thermoacoustic Response of Swirling Flames To The Acoustic Disturbance in A
No ratings yet
Thermoacoustic Response of Swirling Flames To The Acoustic Disturbance in A
15 pages
Report Thermoacoustic Instabilities of Remeha Boiler Koen Pot 03-2022
No ratings yet
Report Thermoacoustic Instabilities of Remeha Boiler Koen Pot 03-2022
30 pages
Aim Paper: Iiii/Sie Joinl Propulsion Speeialisi Conierenee
100% (1)
Aim Paper: Iiii/Sie Joinl Propulsion Speeialisi Conierenee
24 pages
Progress and Challenges in Swirling Flame Dynamics
No ratings yet
Progress and Challenges in Swirling Flame Dynamics
11 pages
Combustion Instability Vortex Shedding
No ratings yet
Combustion Instability Vortex Shedding
16 pages
环形燃气轮机燃烧室的方位热声模态分析
No ratings yet
环形燃气轮机燃烧室的方位热声模态分析
8 pages
Impact of Swirl Fluctuations On The Flame Response of A Perfectly Premixed Swirl Burner
No ratings yet
Impact of Swirl Fluctuations On The Flame Response of A Perfectly Premixed Swirl Burner
7 pages
Large-Eddy Simulations of Combustion Instability in An Axisymmetric Ramjet Combustor
No ratings yet
Large-Eddy Simulations of Combustion Instability in An Axisymmetric Ramjet Combustor
21 pages
Bifurcation of Flame Structure in A Lean-Premixed Swirl-Stabilized Combustor: Transition From Stable To Unstable Flame
No ratings yet
Bifurcation of Flame Structure in A Lean-Premixed Swirl-Stabilized Combustor: Transition From Stable To Unstable Flame
7 pages
Palestra Ben Zinn
No ratings yet
Palestra Ben Zinn
56 pages
Low Order Modelling of Thermoacoustic Instabilities and Intermittency - Flame Response Delay and Nonlinearity
No ratings yet
Low Order Modelling of Thermoacoustic Instabilities and Intermittency - Flame Response Delay and Nonlinearity
16 pages
Ko Pitz 2008
No ratings yet
Ko Pitz 2008
25 pages
GT2017 64348
No ratings yet
GT2017 64348
10 pages
ControlofCombustionInstability MIT PDF
No ratings yet
ControlofCombustionInstability MIT PDF
43 pages
Combustion Instability Research Paper Analysis
No ratings yet
Combustion Instability Research Paper Analysis
30 pages
Numerical Investigation On Combustion Instabilities and Its Passive Control - Zhong
No ratings yet
Numerical Investigation On Combustion Instabilities and Its Passive Control - Zhong
103 pages
Practical Active Control System For Combustion Oscillations: University of Otago, Dunedin, New Zealand
No ratings yet
Practical Active Control System For Combustion Oscillations: University of Otago, Dunedin, New Zealand
10 pages
Afterburners 1 3 3
No ratings yet
Afterburners 1 3 3
5 pages
Aircraft Noise Generation and Assessment
No ratings yet
Aircraft Noise Generation and Assessment
22 pages
Combustion Instability in Gas Turbines
100% (1)
Combustion Instability in Gas Turbines
9 pages
CFX Exercise
No ratings yet
CFX Exercise
163 pages
IB Physics Extended Essay Sample 18/34
No ratings yet
IB Physics Extended Essay Sample 18/34
20 pages
Non-Normality and Nonlinearity in Combustion-Acoustic Interactions in Diffusion Flames
No ratings yet
Non-Normality and Nonlinearity in Combustion-Acoustic Interactions in Diffusion Flames
23 pages
Combustion Instabilities in Gas Turbine Engines PDF
100% (1)
Combustion Instabilities in Gas Turbine Engines PDF
2 pages
V002t02a015 99 GT 110
No ratings yet
V002t02a015 99 GT 110
17 pages
GTC 1
No ratings yet
GTC 1
60 pages
Combustion Dynamics in Gas Turbines
100% (2)
Combustion Dynamics in Gas Turbines
3 pages
Jet Engine Combustion Chamber Dynamics
No ratings yet
Jet Engine Combustion Chamber Dynamics
6 pages
The Premixing Method Prinicple Design and Construction
No ratings yet
The Premixing Method Prinicple Design and Construction
153 pages
Discrete Time Crystal Phase As A Resource For Quantum Enhanced Sensing - 2405.00328v4
No ratings yet
Discrete Time Crystal Phase As A Resource For Quantum Enhanced Sensing - 2405.00328v4
11 pages
Applications of Thermodynamics in Daily Life
No ratings yet
Applications of Thermodynamics in Daily Life
5 pages
Bulletin of Earthquake Engineering (BEEJ)
No ratings yet
Bulletin of Earthquake Engineering (BEEJ)
39 pages
Shear Strength of Reinforced Concrete
No ratings yet
Shear Strength of Reinforced Concrete
19 pages
Ophthalmic Lens Materialand Design
No ratings yet
Ophthalmic Lens Materialand Design
42 pages
Exercises On Quadratic Functions
No ratings yet
Exercises On Quadratic Functions
4 pages
Topic 8 Angular Kinetics
No ratings yet
Topic 8 Angular Kinetics
31 pages
Magnet Answer
No ratings yet
Magnet Answer
2 pages
High Voltage Safety Course Certificate
No ratings yet
High Voltage Safety Course Certificate
1 page
Operating and Servicing Instructions: For Co-Rotating Twin Screw Extruder Gearbox
No ratings yet
Operating and Servicing Instructions: For Co-Rotating Twin Screw Extruder Gearbox
33 pages
Kerala University BSc Exam Timetable 2022
No ratings yet
Kerala University BSc Exam Timetable 2022
8 pages
Large Scale Geometry Second Edition Piotr Nowak Newest Edition 2025
No ratings yet
Large Scale Geometry Second Edition Piotr Nowak Newest Edition 2025
133 pages
Elasticity, Heat and Thermodynamics CJM21
No ratings yet
Elasticity, Heat and Thermodynamics CJM21
2 pages
Physics Problems for IIT Aspirants
No ratings yet
Physics Problems for IIT Aspirants
5 pages
(ECE 401) Electrostatics and Magnetostatics - Week 6
No ratings yet
(ECE 401) Electrostatics and Magnetostatics - Week 6
17 pages
Electricity
No ratings yet
Electricity
59 pages
NSEP WORKSHOP Standardized TEST 1 Solutions - 250526 - 113655
No ratings yet
NSEP WORKSHOP Standardized TEST 1 Solutions - 250526 - 113655
20 pages
Atom v-2100G Infant Incubator - User Manual
No ratings yet
Atom v-2100G Infant Incubator - User Manual
74 pages
134 Celsius To Fahrenheit - What Is 134 °C in °F
No ratings yet
134 Celsius To Fahrenheit - What Is 134 °C in °F
18 pages
Topic 8L - Space Checklist Answers
No ratings yet
Topic 8L - Space Checklist Answers
4 pages
HOQ Project
No ratings yet
HOQ Project
1 page
Formulas For Calculating Losses in Pipes
No ratings yet
Formulas For Calculating Losses in Pipes
8 pages
Artigo Luben
No ratings yet
Artigo Luben
33 pages
Managing Track Stiffness in Transition Zones en
No ratings yet
Managing Track Stiffness in Transition Zones en
5 pages
Understanding Volume in Physics
No ratings yet
Understanding Volume in Physics
19 pages
Composite Leaf Spring Fatigue Analysis
No ratings yet
Composite Leaf Spring Fatigue Analysis
22 pages
BSM by Pros
No ratings yet
BSM by Pros
16 pages
Aghamohammadi, Eslami-Farsani, ... - 2020 - The Effect of Multi-Walled Carbon Nanotubes On The Mechanical Behavior of Basalt Fibers Meta
No ratings yet
Aghamohammadi, Eslami-Farsani, ... - 2020 - The Effect of Multi-Walled Carbon Nanotubes On The Mechanical Behavior of Basalt Fibers Meta
9 pages
Carbopol Rheology
No ratings yet
Carbopol Rheology
18 pages

Poinsot 2016

Uploaded by

Poinsot 2016

Uploaded by

JID: PROCI

ARTICLE IN PRESS [m;June 25, 2016;16:33]

Available online at [Link]

Proceedings of the Combustion Institute 000 (2016) 1–28

Prediction and control of combustion instabilities in real

Received 24 November 2015; accepted 19 May 2016

Keywords: Instabilities; LES; Thermoacoustics; Gas turbines

1. Introduction working on steady combustion chambers (furnaces,

2 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 3

4 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 5

6 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

on the pulsation ω: approach in a one-dimensional case. More sophis-

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 7

Fig. 4. Simulation methods for combustion instabilities.

3.1. Azimuthal instability modes in annular

8 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 9

10 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

wise) is: next section provides an example of application of

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 11

S0 = 2K . ( 1 − 2 ) (15) 4.1. The effects of wall temperatures

12 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 13

14 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 15

16 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 17

18 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

Outlet 1/ Infinite duct Non reflecting 1

Inlet 2/ Infinite duct Non reflecting −1

Outlet 3/ Constant pressure p = 0 0

Inlet 5/ Choked nozzle u = 0 ∞

Outlet 6/ Choked nozzle u = 0 ∞

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 19

Fig. 20. Transmission of acoustic (A+ +

20 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 21

22 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 23

24 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 25

5. Conclusions [4] T. Lieuwen, J. Fluid Mech. 435 (2001) 289–303.

26 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28 27

28 T. Poinsot / Proceedings of the Combustion Institute 000 (2016) 1–28

You might also like

S0 = 2K . ( 1 − 2 ) (15) 4.1. The effects of wall temperatures