Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Proceedings of ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition

GT2017
June 26-30, 2017, Charlotte, NC, USA

GT2017-64233

EXTRACTION OF LINEAR GROWTH AND DAMPING RATES OF HIGH-FREQUENCY


THERMOACOUSTIC OSCILLATIONS FROM TIME DOMAIN DATA

Tobias Hummel∗1,2 , Frederik Berger1 , Nicolai Stadlmair1 , Bruno Schuermans2,3 , Thomas Sattelmayer1
1 Lehrstuhl für Thermodynamik, Technische Universität München, Garching, Germany
2 Institute for Advanced Study, Technische Universität München, Garching, Germany
3 GE Power, Baden, Switzerland

ABSTRACT results. Numerical test cases given by time domain formulations


This paper presents a set of methodologies for the extraction of the acoustic conservation equations including high-frequency
of linear growth and damping rates associated with transversal flame models as well as acoustic damping terms are set up and
eigenmodes at screech level frequencies in thermoacoustically solved. The resulting unsteady pressure and heat release data
non-compact gas turbine combustion systems from time domain are then subjected to the proposed identification methodologies
data. Knowledge of these quantities is of high technical rele- to present corresponding proof of principles and grant suitability
vance as an required input for the design of damping devices for employment on real systems.
for high frequency oscillations. In addition, validation of pre-
diction tools and flame models as well as the thermoacoustic
characterization of a given unstable/stable operation point in NOMENCLATURE
terms of their distance from the Hopf bifurcation point occurs Abbreviations
via the system growth/damping rates. The methodologies solely CW clockwise
rely on dynamic measurement data (i.e. unsteady heat release CCW counter-clockwise
and/or pressure recordings) while avoiding the need of any ex- FEM Finite Element Method
ternal excitation (e.g. via sirens), and are thus in principle suit- MIMO Multi-Input-Multi-Output
able for the employment on operational engine data. Specifically, OP Operation Point
the following methodologies are presented: 1) The extraction of ROM Reduced Order Model
pure acoustic damping rates (i.e. without any flame contribution) SDE Stochastic Differential Equation
from oscillatory chemiluminescence and pressure recordings. 2) T1 first transversal
The obtainment of net growth rates of linearly stable operation
points from oscillatory pressure signals. 3) The identification
of net growth rates of linearly unstable operation points from Latin Symbols
noisy pressure envelope data. The fundamental basis of these A system matrix
procedures is the derivation of appropriate stochastic differen- B input matrix/vector
tial equations, which admit analytical solutions that depend on C output matrix/vector
the global system parameters. These analytical expressions serve c speed of sound
as objective functions against which measured data are fitted to
D absorption function
yield the desired growth or damping rates. Bayesian methods
dt time step
are employed to optimize precision and confidence of the fitting
F slowly varying amplitude of CW mode
F non-compact transfer function
∗ Address all correspondence to this author(email:hummel@td.mw.tum.de).

1 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


f Fourier coefficient of CW mode Subscripts
G slowly varying amplitude of CCW mode a acoustic
g Fourier coefficient of CCW mode f flame
He Helmholtz number F CW mode
H transfer function G CCW mode
i imaginary number n mode number
K transfer function gain parameter L linear
k autocorrelation function NL nonlinear
n normal vector s segment
p pressure
q volumetric heat release rate
INTRODUCTION
T final time
Engineering combustion systems that are based on lean, pre-
t time
mixed technologies – as implemented in modern gas turbine
V volume power systems – exhibit a sensitive susceptibility to develop
x state vector thermoacoustic instabilities [1]. Physically, these instabilities
x coordinate vector emerge as self-sustained pressure oscillations due to construc-
x, y, z cartesian coordinates tive coupling mechanisms between the unsteady flame and natu-
x, r, θ cylindrical coordinates ral acoustics (i.e. eigenmodes) of the chamber. If the phase re-
lation between heat release oscillations and eigenoscillations are
(un-)favorable (cf. Rayleigh criterion [2]), the former presents
a source of acoustic energy. Additionally, if this energy addition
Greek Symbols
exceeds the acoustic dissipation of the particular eigenmode (e.g.
α acoustic damping rate
due to vortex shedding [3–5]), the associated oscillation ampli-
β flame driving rate tudes amplify exponentially, and the respective mode is labeled
δ flame length scale linearly unstable. Eventually, the flame’s energy provision at-
η acoustic oscillation variable tenuates due to nonlinear effects, which saturates the amplitude
Γ noise intensity growth into a constant amplitude limit cycle oscillation [6]. In
κ nonlinearity coefficient the vice versa scenario of a stable system, there is no growth of
Λ air excess ratio oscillation amplitude, the evolution of which is instead stochas-
λ wave length tically governed by turbulent combustion noise. The linear sta-
bility of a thermoacoustic system is mathematically quantified by
ν net growth rate
the complex eigenfrequency – expanding into growth rate and os-
φ slowly varying phase
cillation frequency – of the individual eigenmodes. This growth
ψ mode shape rate can be interpreted as an integral measure of net acoustic
ρ density energy due to flame driving and acoustic dissipation processes.
σ axial/radial mode shape Hence, a positive and negative growth rate value indicates that
τ time delay the corresponding eigenmode is linearly unstable and stable, re-
ω angular frequency spectively. Obtaining these modal growth rates – as well as indi-
Ξ non-compact white noise source vidual contributions due to flame driving and acoustic damping
ξ white noise source mechanisms – from measurement data is of high practical rele-
χ white noise source vance due to the following reasons:
First of all, the absolute value of the growth rate characterizes
a mode in terms of its distance from the stability boundary. For
unstable cases, this knowledge presents a crucial input for proper
Superscripts
0 acoustic fluctuation in time domain design of damping devices (e.g. Helmholtz resonator) to sup-
press the respective mode’s instability [7, 8]. Conversely, a sta-
¯ mean quantity ble (negative) growth rate’s magnitude informs about the modes
ˆ acoustic fluctuation in frequency domain sensitivity to supercritical bifurcations e.g. due to a change in
˙ time derivative operation conditions. Second of all, the physical understanding
∗ complex conjugate of flame driving and acoustic damping behavior as a function of

2 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


different operation parameters can be gained by knowledge of CONVENTIONS
the growth rates. Third of all, experimentally retrieved growth In general, the quantities damping/growth rate can be physically
rates serve usefully for validation of modal thermoacoustic anal- interpreted as an integral measure of dissipated/generated
ysis tools for stability prediction as well as underlying models acoustic energy associated with a distinct acoustic mode. In
for aeroacoustic damping and flame driving mechanisms. order to avoid ambiguity regarding these terms throughout this
paper, the following conventions are established:
This work seeks to present three methodologies to retrieve
Damping Rate
modal growth rates of linearly (1) stable and (2) unstable sys-
The term damping rate – symbolized by αn – implies loss
tems as well as (3) the pure damping rate due to aeroacoustic
of energy due to pure acoustic dissipation mechanisms, and
dissipation mechanisms only from dynamic measurement data.
remains positive (αn > 0) throughout this work.
Hereby, specific focus is placed on non-compact thermoacoustic
systems governed by transversal modes at screech level high-
Growth Rate
frequencies. Previous work on system identification method-
The term growth rate – symbolized by νn – refers to the net
ologies introduced for low-frequency, compact thermoacoustic
value of converted acoustic energy due to flame driving (βn )
systems served as motivation and theoretical starting point (cf.
and acoustic dissipation (αn ) mechanisms acting in the system.
[8–11]). The basic idea of these methodologies is to model the
Thus, νn = βn − αn . A positive and negative value of the growth
thermoacoustic system dynamics via stochastic differential equa-
rate implies a linearly unstable (νn > 0) and stable (νn < 0)
tions. Then, respective solution expressions to these equations
configuration, respectively.
are derived, which include the growth rates as model constants.
The desired growth rates are extracted by fitting corresponding
measurement data against these expressions. Specifically, the
determination of the stable growth and pure acoustic damping HIGH VS. LOW FREQUENCY THERMOACOUSTICS
rates rely on a second order coupled oscillator system stochasti- The thermoacoustic configuration used in this work is a per-
cally forced by white noise. The mathematical linearity of these fectly premixed, swirl-stabilized combustor. A schematic, illus-
equations allows to derive an analytical solution (for both cases) trating the basic operation principle and design is presented in
in form of autocorrelation functions for the fit against time do- Fig. 1. Moreover, this paper specifically aims at high-frequency
main data and growth/damping rate retrieval. In order to identify (HF) thermoacoustics oscillations that are governed by transver-
unstable growth rates, the linear oscillator system is expanded sal acoustic modes. Such systems exhibit distinct differences
by an analytical nonlinear flame transfer function. The resulting compared to the ”classical” low-frequency (LF) regime, which
second order system of equations is transformed into a first order are outlined in the following:
system that governs the amplitude-phase dynamics of the under-
lying limit cycle oscillation. The required solution expression
is produced by linearizing the equations around the limit cycle 1. Multidimensional Modes
amplitude, which is formulated as an autocorrelation relation for The term HF implies that oscillations occur at frequencies
the subsequent fit, too. In order to optimize the robustness and beyond the geometrical cut-on value of the subjected cham-
reliability of the identification procedures, Bayesian methodolo- ber. Thus, multidimensionally distributed oscillations mani-
gies are employed. The identification methodologies are output fest via transversal, radial, longitudinal (and any combinations
only, which implies that the measured data is retrieved from au- thereof) modes, whereas the LF regime only features planar, one-
tonomously operating combustors without the need of external dimensional acoustic modes. Typical mode shapes of LF and HF
intrusions such as through actuated damping and excitation de- (here: first transversal mode) systems are comparatively shown
vices. in Figs. 1 (a) and (b).

The paper is structured as follows. First, relevant conventions 2. Non-Compact Flames


for this work are established, followed by the provision of some Figures 1 display characteristic length scale relations between
practical background on specific features of transversal, non- flame and acoustic modes. For LF system, this ratio is small,
compact thermoacoustic systems. Then, governing equations which allows to neglect any local thermoacoustic interactions
and solution approaches are presented, accompanied by outlin- across the flame volume, rendering the flame as compact. Op-
ing the corresponding identification procedures. Verification test positely, the length scales in high-frequency system are of equal
cases are defined and carried out next utilizing synthetically gen- order of magnitude so that local variation of flame-acoustic cou-
erated data in order to establish a proof of principle of the pro- pling effects require consideration, labeling the flame as ther-
posed methods’ applicability. moacoustically non-compact. Mathematically, this (non-) com-

3 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


r in accordance with the Rayleigh criterion. Specific details on
(a) Longitudinal pressure mode x
λa thermoacoustic driving mechanism in HF system can be found
in previous publication of this work’s authors in [12, 13].

δf GOVERNING EQUATIONS AND SOLUTIONS


This section presents the governing equations on which the sys-
tem identification methods presented in this work are based. The
starting point is given by the inhomogeneous wave equation that

describes transversal thermoacoustic systems
-1 1 Flame contour

∂ 2 p0 ∂ p0 ∂ q0
 
(b) First transversal pressure mode 2 1 0
2
+ D − ρ̄ c̄ ∇ · ∇p = , (1)
∂t ∂t ρ̄ ∂t
δf λa /2
where the solution variable p0 = p0 (x,t) and source term
q0 = q0 (x,t) denote spatio-temporal acoustic pressure and
Pressure heat release oscillations, respectively. The spatial variable is
probes a vector x that can describe either Cartesian x = [x, y, z]T or
Swirl burner Combustion chamber
cylindrical x = [x, r, θ ]T coordinate systems. The mean density
ρ̄ = ρ̄(x) and speed of sound c̄ = c̄(x) is spatially varying as a
FIGURE 1. MODEL COMBUSTION SYSTEM WITH (a) LON-
consequence of the temperature distribution induced by the com-
GITUDINAL PRESSURE MODE AND COMPACT FLAME, (b)
bustion process. Any physical mechanisms leading to damping
TRANSVERSAL PRESSURE MODE AND NON-COMPACT
of acoustic oscillations – e.g. due to vortex shedding [4] – are
FLAME
modeled by the absorption function D = D(x). This function is
non-zero only in regions where this damping physically occurs
pactness is quantified by the Helmholtz number via (in accordance with [4]) as indicated in Fig. 2. Notice that first
transversal modes – which are of primary concern in this work
( – are independent of any in- and outlet dissipation due to the
δf  0.1 → compact
He = = longitudinal attenuating of the mode shape. The reason for this
λa ≥ 0.1 → non-compact attenuation is the increasing cut-on frequency value in up- and
downstream direction due to the smaller cross-section of the
where δ f and λa refer to the characteristic flame length and the mixing tube before, and the elevated temperature after the flame
acoustic wavelength at the considered mode as indicated in Figs. zone, respectively (cf. details in [13]).
1(a)-(b). Thus, the terms HF, transversal and non-compact ther-
moacoustic systems imply the same features, and are used inter- As the first step to enable analytical access to Eqn. 1, the spatio-
changeably within this work. temporal acoustic pressure is assumed to be solely governed by
the first transversal mode. In practice, this single mode gover-
nance is achieved by bandpass-filtering any time signals around
3. Driving Mechanisms the respective frequency. Due to azimuthal periodicity in cyclin-
The physical interaction mechanisms between acoustic and heat drical chambers, the pressure variable is then expanded by a two-
release oscillations differ between LF and HF systems, too. The fold Fourier series given by [14]
former is governed by convective transport of perturbations (e.g.
of equivalence ratio) that are converted into heat release oscilla-
p0 (x,t) = f (t)ψ f (x) + g(t)ψg (x) + c.c.1 , (2)
tions at the flame. These mechanisms cause a predominant sen-
sitivity of heat release oscillations to acoustic velocity perturba-
tions. For elevating frequencies, the convective transport expe- where the shape functions ψ f (x) and ψg (x) are associated with a
riences increasing dissipation until it is negligible (i.e. low pass clockwise (CW) and counterclockwise (CCW) rotating transver-
filter behavior) in the HF regime. Then, thermoacoustic driv- sal mode, respectively. The complex Fourier coefficients f (t)
ing is induced locally due to periodic flame shape deformations and g(t) are time-dependent signals associated with each rotating
and displacement effects. These local modulations result in an
overall in-phase relation between heat release and pressure os-
cillations, and thus, always resemble a source of acoustic energy 1 complex conjugate

4 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


jections [15, 16]. Finally, one obtains the second order system
Mean heat release distribution q̄(x)
f¨ + g¨∗ + 2αn ( f˙ + g˙∗ ) + ωn2 ( f + g∗ ) = q̇a , (7)
∇ψ · n = 0
f¨∗ + g̈ + 2αn ( f˙∗ + ġ) + ωn2 ( f ∗ + g) = q̇b , (8)
Damping region D y
x
1 where the time derivatives were replaced by dot symbols, i.e.
∇ψ = 0
d/dt → (˙) for simplicity of notation. The parameter αn and
ψ =0 ωn denote acoustic damping rate and oscillation frequency, re-
0 spectively. The damping rate is directly related to the absorption
function D as a consequence of the employed Galerkin projec-
T1 mode shape |ψ(x)| tion:

1 Dσ 2 dV
R
FIGURE 2. COMPUTATIONAL DOMAIN WITH DAMPING RE- αn = RV 2 (9)
GION D, BOUNDARY CONDITIONS, MEAN HEAT RELEASE DIS- 2 V σ dV
TRIBUTION AND T1 MODE SHAPE
In practical swirl-stablized combustion systems, the damping
rates and frequencies of the CCW and CW rotating modes dif-
mode. The mode shapes expand to fer due to mean flow effects [17]. However, these differences
are typically small, and can be neglected for this work. Equa-
ψ f (x) = ψg∗ (x) = σ (x, r) exp(−iθ ) (3) tions 7-8 govern the dynamics of the Fourier coefficients of the
counter-rotating modes f (t) and g(t) depending on the unsteady
ψg (x) = ψ ∗f (x) = σ (x, r) exp(iθ ) (4) heat release sources terms qa and qb . These sources expand to

where exp(iθ ) and exp(−iθ ) imply CCW and CW direction of


Z
qa = q0 ψg dV, (10)
rotation, respectively. Any axial and radial variability of the V
mode shapes is absorbed in σ (x, r), which is a real expression.
Z
qb = q0 ψ f dV, (11)
Notice that higher transversal modes can be treated analogeously V
as shown herein for the T1 mode. In Eqns. 3-4, the asterisk
(∗ ) indicates the complex conjugate of the concerned function. and are complex quantities, too. Next, analytical solutions – in-
Mode shape solutions are typically obtained by solving the ho- cluding the growth and damping rates as constant parameters –
mogeneous Helmholtz equation [15] e.g. by the Finite Element are derived, and the respective usage for parameter identification
Method (FEM) from time domain data is outlined.
 
1 (1) Damping Rate Identification
ωn2 ψn + ρ̄ c̄2 ∇ · ∇ψn = 0, (5)
ρ̄ First, the identification of the pure acoustic damping rate αn of
the T1 mode in a non-compact combustor is subjected, following
where ωn denotes the modal eigenfrequency. Boundary condi- the work of [11] for LF systems. For this purpose, the complex
tions are set to fully reflecting as shown in Fig. 2 along with the valued dynamical system in Eqn. 7-8 is expanded into real and
computational mesh for the FEM simulations. Furthermore, the imaginary components, which are then summed up to transform
modes feature orthonormality, i.e. the multi-variable system into a single-variable equation:

( η̈ + 2αn η̇ + ωn2 η = q̇, (12)


1→n=m
Z
ψn ψm∗ dV = (6)
V 0 → n 6= m
Herein, η = fR + gR + fI − gI and q = qa,R + qb,R + qa,I − qb,I ,
where the indeces R and I denote real and imaginary part, re-
where V is the combustor volume. Utilizing Eqns. 2-6 allows spectively. The resulting equation represents a forced linear os-
to transform the wave equation in Eqn. 1 into a system of ordi- cillator, where the acoustic oscillations η are the response to the
nary differential equations by employing complex Galerkin pro- forcing due to heat release oscillations q. The latter includes

5 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


both, thermoacoustic and stochastic effects due to acoustics- at the chamber’s faceplate as indicated in Fig. 1. The autocor-
flame coupling mechanisms and broadband combustion noise, relation is a time-domain function, and can thus be subjected
respectively. The oscillator model in Eqn. 12 is equally valid for to Bayesian fitting methods, which enables robust and reliable
linearly stable and unstable cases, the respective acoustic time extraction of desired parameters as is explicated in more detail
traces of which are governed by white noise and coherent limit in [20].
cycle oscillations. This is somewhat unintuitive (especially for
the unstable case) as any underlying physics of unsteady flame
(2) Growth Rate Identification – Stable Cases
behavior – e.g. causing the saturation of heat relase – are cer-
This section outlines the identification of the growth rate for lin-
tainly nonlinear. However, the reason for this general applica-
early stable cases of the T1 mode. First, the non-compact ther-
bility is due to the overall acoustic linearity that persists in both
moacoustic source term in Eqn. 1 is written as
scenarios. As oscillation amplitudes remain small compared to
the mean flow, the signal associated with the heat release os-
cillations q stays harmonic, too. The overall linearity allows to q0 = FL p0 + Ξ, (16)
Fourier transform Eqn. 12 and formulate the transfer function
between heat release (input) and pressure oscillations (output), where FL = FL (x) is the spatially variable (across the flame vol-
which reads ume) linear transfer function, which is real as explained in [12].
White noise sources Ξ = Ξ(x,t) represent broadband combus-
η̂(ω) −iω tion noise emitted by the flame as described in [9]. Note that
Hη/q (ω) = = 2 , (13)
q̂(ω) ω − 2iωαn − ωn2 the white noise assumption for combustion noise does hold true
for practical system in general. However, for the present pur-
where the independent variable ω is the angular frequency. Uti- poses it is justified as the noise sprectrum can be considered con-
lizing the Wiener-Khinchin theorem – which states that a time stant over the peak width of concerned modal oscillations’ power
signal’s the power spectral density is given by the Fourier trans- spectrum. Substitution of Eqn. 2 into Eqn. 16 along with sub-
form of the autocorrelation [10, 18] – allows to derive the auto- sequent Galerkin projections using the shape functions given by
correlation function associated with the transfer function in Eqn. Eqn. 4 and Eqn. 3 respectively yields
13, i.e. Z
Z ∞ qa = q0 ψg dV = 2βn ( f + g∗ ) + ξ (17)
V
kH (τ) = SH (ω) exp(iωτ)dω. (14) Z
−∞ qb = q0 ψ f dV = 2βn ( f ∗ + g) + ξ (18)
V

Herein, SH = |Hη/q |2 is the power spectral density, while τ is


the time delay vector. The integral in Eqn. 14 is analytically for the integral source terms, where βn represents the pure con-
evaluated by employing the Residue Theorem [19] as well as tribution to the growth rate due to flame driving, and ξ = ξ (t) is
assuming αn  ωn to give a white noise signal. Note that the driving rate Rβ can be related
F σ 2 dV
to the linear transfer function FL , i.e. βn = 21 VR Lσ 2 dV , which
V

kH (τ) = exp(−αn τ) cos(ωn τ), (15) is a consequence of the Galerkin projections as for the relation
between damping rate and absorption function in Eqn. 9. The re-
sulting coupled oscillator of Eqns. 7-8 is analogously processed
which is normalized such that kH (0) = 1. The acoustic damp- as in the previous section, i.e. separating the complex Fourier
ing rate αn along with oscillation eigenfrequency ωn of the con- coefficients into real and imaginary part, followed by the sum-
cerned transversal mode is retrieved by fitting Eqn. 15 against mation into a single-variable equation:
the autocorrelation computed from time domain signals. Notice
that Eqn. 15 represents an overdetermined system of equations
with Nτ (i.e. length of the autocorrelation vector) single equa- η̈ − 2νn η̇ + ωn2 η = ξ˙ (19)
tions available to determine two unknowns, i.e. αn and ωn . Pro-
cedurally, this autocorrelation from the time domain data is ob- This equation can be interpreted as a stochastic linear oscilla-
tained by computing the transfer function between the acoustic tor where the solution variable is associated to the Fourier co-
oscillations η and integral heat release oscillations from the non- efficients by η = fR + gR + fI − gI , which is randomly driven
compact flame utilizing Eqn. 14. The acoustic signal η is ulti- by white noise ξ˙ . Notice that the derivative of an uncorrelated,
mately obtained by solving Eqn. 2 in a least-square sense from white noise signal is assumed to be an uncorrelated, white noise
pressure traces recorded e.g. at several circumferential positions signal, too. The net growth rate given by νn = βn − αn in Eqn.

6 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


19 is required to be negative – i.e. βn < αn – for the presumed sociated with limit cycle oscillations. Therefore, the Fourier co-
linearly stable situation in this section. In order to retrieve the efficients of the counter-rotating T1 modes are decomposed into
growth rate, Eqn. 19 is – identically as in the previous section
and in [10] for compact systems – transformed into frequency f = F(t) exp[iωt + iφF (t)], (24)
domain to derive the autocorrelation function of the signal η uti-
g = G(t) exp[iωt + iφG (t)], (25)
lizing Eqn. 14. Finally, one obtains

where F(t), G(t) and φF (t), φG (t) respectively denote time de-
kηη (τ) = exp(νn τ) cos(ωn τ), (20) pendent amplitude and phase of the corresponding oscillatory
quantity. The time-scale of variation of these amplitudes/phases
which is normalized such that kηη (0) = 1. The autocorrelation is assumed as much slower than the oscillatory time scale.
of the acoustic oscillations η is then fitted against this analyti- Hence, the former can be considered constant during one oscilla-
cal expression (using Bayesian methods) to extract the underly- tion period. Substituting Eqns. 24-25 into the oscillator system,
ing growth rate and oscillation frequency of a linearly stable T1 and employing a complex temporal averaging procedure (cf. de-
mode. tails in [16, 17]) yields the following system of SDE
 
1 3 Γ
(3) Growth Rate Identification – Unstable Cases Ḟ = νn F − 3κ F + FG2 + + χF , (26)
2 4ωn2 F
A linearly unstable system is characterized by a positive growth  
1 3 Γ
rate (νn = βn − αn > 0) leading to self-sustained limit cycle os- Ġ = νn G − 3κ G + GF 2 + + χG , (27)
cillations. The latter are physically caused by saturation of heat 2 4ωn2 G
 
release oscillations as a function of the amplitude level. Mathe- 1 1
Φ̇ = (φ̇F − φ̇G ) = + χΦ , (28)
matically, this saturation is typically described by a cubic relation F G
of the acoustic pressure [21], yielding a nonlinear thermoacous-
tic source term formulation for non-compact flames where χF , χG , χΦ denote white noise signals of intensity Γ/2ωn
[22]. Notice that the phase SDE in Eqn. 28 is decoupled from
q0 = FL p0 − FNL p03 + Ξ, (21) the amplitude SDE in Eqns. 26-27, and is thus not needed for the
following derivations. Analytical access towards a solution of the
SDE is established by assuming small stochastic perturbations of
where FL and Ξ are spatially dependent linear and stochastic parts the amplitude due to noise as in [9] for LF systems, i.e.
of flame dynamics as in Eqn. 16, whereas FNL denotes a con-
stant nonlinearity parameter used to desribe the flame’s satura- F = F̄ + F 0 (t) → F 0  F̄ (29)
tion strength. As for the stable case in the previous section, the
G = Ḡ + G0 (t) → G0  Ḡ (30)
pressure terms in Eqn. 21 are expanded via Eqn. 2 to derive the
integral heat release source term of a non-compact flame gov-
erned by a linearly unstable transversal mode where F̄/Ḡ and F 0 /G0 represent mean and perturbation quan-
tities, respectively. The mean amplitudes are obtained from a
Z fixed point analysis of the deterministic constituents (i.e. for
qa = q0 ψg dV = 2β ( f + g∗ ) + 3κ( f + g∗ )2 ( f ∗ + g) + ξ , Γ = 0 & χF = χG = 0 of the SDE. As is shown in [17, 23], rotat-
V
ing modes are the only two physically feasible, i.e. stable, fixed
(22)
Z point solutions 2 , which unfold into
qb = q0 ψ f dV = 2β ( f ∗ + g) + 3κ( f ∗ + g)2 ( f + g∗ ) + ξ ,
V p
(23) FP #1: F̄ = 2νn /3κ, Ḡ = 0 → CW-rot. mode, (31)
p
FP #2: F̄ = 0, Ḡ = 2νn /3κ → CCW-rot. mode. (32)
where the constant κ resembles an integral value of the nonlin-
earity parameter FNL . The corresponding coupled oscillator sys- For practical swirl-stabilized combustion systems, the direction
tem is of nonlinear type, which prohibits the employment of fre- of rotation (of the T1 modes) follows the direction of the swirling
quency domain approaches for the identification procedures as
in the two previous sections. Instead, the second order system
is transformed into a first order system of stochastic differential 2 if the system is autonomous, that is, unequipped with dampers and rotation-

equations (SDE) that govern the amplitude-phase dynamics as- ally uniform flame shape

7 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


mean flow as is explicitly discussed in [17]. For this work, CCW Non-Compact Reduced Order Model
rotation prevails, although the procedure is analogous if a CW A Reduced Order Model (ROM) capable of handling non-
rotating mode constitutes the limit cycle. Next, the SDE are lin- compact thermoacoustic systems is utilized to generate the re-
earized by substitution of Eqns. 29-30 into Eqns. 26-27. Ad- quired synthetic time series.
ditionally exploiting Eqn. 32 allows one to obtain a linear first Mathematically, this ROM is formulated as a state-space system
order equation for the CCW mode’s amplitude perturbation, i.e. given by

  dx
Γ = Ax + Bu, (35)
Ġ0 = −2νn + G0 + χG , (33) dt
4ωn2 Ḡ2
y = Cx. (36)

where the presumed low noise intensity leads to Γ/4ω02 Ḡ2  Herein, x is the state vector, and A is the system matrix carrying
1 → Γ ≈ 0. The CW mode’s counterpart equation for F 0 is not information about the systems physics, which can be interpreted
shown as it results in a trivial expression due to the zero mean as a discrete, low order description of Eqn. 1 with a passive
amplitude of the fixed point, and is thus, not required for the flame, i.e. q0 = 0.
desired purpose of identifying the growth rate. Finally, Eqn. 33 The in- and output matrices B and C allow the insertion and ex-
is transformed into a normalized autocorrelation function, which traction of multiple signals (e.g. heat release and pressure os-
reads cillations) from and into the system. These signals are used to
model the non-compact flame dynamics at the first transversal
mode with a flame segmentation approach. This segmentation
kGG = exp(−2νn τ). (34) implicates the division of the non-compact flame volume into
compact subregions. A feedback loop is then established for each
subregion utilizing the nonlinear, stochastic transfer function in
This relation is used to extract the growth rate of an unstable T1 Eqn. 21 via multiple in- and output signal routes as illustrated in
mode by fitting the right hand side against the autocorrelation Fig. 3.
obtained from time traces of the CCW mode’s amplitude pertur- This Multi-Input-Multi-Output (MIMO) feedback loop essen-
bation G0 , which is obtained from the respective Fourier coeffi- tially models the non-compact thermoacoustic system. The
cient g(t) via Eqn. 25 (Hilbert transform) and Eqn. 30. Notice flame volume on which the flame segmentation is carried out
that the first order nature of the underlying differential equation is based on an experimentally obtained, time averaged OH∗ -
yields exponential decay only, without any oscillatory compo- chemiluminescence image (cf. Fig. 3). The ROM is based on the
nent. Using standard least-square techniques for the fitting tasks 3D wave equation in Eqn. 1, and a corresponding FEM descrip-
would certainly suffice here, although the Bayesian framework is tion of the test case combustor. The nature of the ROM allows to
employed for consistency reasons. setup stable and unstable operation points with definite knowl-
Finally, the identification of both, damping and growth rates al- edge of the latter values as desired, while providing associated
lows the determination of the pure thermoacoustic driving rate time series upon simulations. As stochastic forcing effects are
induced by the unsteady flame via βn = νn + αn as is shown in included by the white noise source terms, the generated data sets
Fig. 7. Consequently, all individual energetic contributions of can be considered comparable to real, experimentally obtained
flame driving and acoustic dissipation can be quantified, and as- measurements. Specific details regarding the non-compact ROM
sessed independently from each other, e.g. for model validation along with underlying theory, derivations and associated Finite-
or system characterization. Element discretization techniques can be found in [24].

Operation Points
TEST CASE COMBUSTION SYSTEM Six test case operation points are established, which feature three
The lab-scale swirl-stabilized combustion rig in Fig. 1 is used linearly stable and three unstable T1 modes at different levels
as a test case system for this work. Numerical simulation of this of growth and damping rates. A nominal operation point fea-
rig’s thermoacoustic performance are carried out to produce time turing a thermal power value of 200kW at an air excess ratio
domain data with known values of growth and damping rates on of Λ = 1.4 is selected as base case for the numerical demon-
which the above-introduced system identification techniques are stration studies. The required flame volume, mean heat release
tested and verified. Details on the numerical setup, operation and density distributions are retrieved from time-averaged OH∗ -
points and generated time domain data are provided in the fol- chemiluminescence images as presented in detail in [12]. Differ-
lowing. ent levels of growth and damping rates are achieved by varying

8 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


# OP 1 2 3 4 5 6
Ξs −FNL,s
νn (rad/s) −25 −15 −5 5 15 25
u = q0s αn (rad/s) 35 25 15 25 25 25
s-th SUB- βn (rad/s) 10 10 10 30 40 50
FL,s ( )3
REGION K/KOP5 0.25 0.25 0.25 0.75 1 1.25
D/DOP5 1.4 1 0.6 1 1 1
y= p0s TABLE 1. TEST CASE OPERATION POINTS, GROWTH RATES,
A
DAMPING RATES, FLAME DRIVING RATES, VALUES OF
TRANSFER FUNCTION GAIN AND DAMPING ABSORPTION PA-
RAMETER

r r
x θ
(OP#2) and unstable (OP#5) operation point are shown in Fig. 4.
The values for nonlinearity coefficient κ and noise intensity of
q̄(x) Ξ are arbitrarily prescribed (and held constant for all operation
0 max points) such that the time series produced by the simulations re-
semble experimentally measured counterpart data of real systems
A A-A (cf. [25]).
Before proceeding to the test case results, relevant remarks re-
FIGURE 3. NON-COMPACT FLAME SEGMENTATION, MEAN
garding the determination of input time series for the respective
HEAT RELEASE DISTRIBUTION, AND MULTI-INPUT-MULTI-
methods are made to ensure procedural clarity: The Fourier coef-
OUTPUT (MIMO) FEEDBACK CONNECTIONS FOR REDUCED
ficient time series f (t) and g(t) – required for method (1) and (2)
ORDER MODELING
– are obtained by solving Eqn. 2 in a least square sense. There-
fore, acoustic pressure traces are retrieved at N p = 6 probes that
the absorption parameter D in Eqn. 1 as well as the gain of the are located at several azimuthal positions θ p and fixed radial as
linear transfer function in Eqn. 16. The latter reads well as axial positions in the vicinity of the chamber faceplate
(cf. Fig. 1). The integral heat release time series qa and qb –
FL (x) = K q̄(x) (37) additionally necessary for method (1) – are determined by nu-
merically evaluating Eqns. 17-18. This requires the time series
where q̄(x) is the mean volumetric heat release rate. The term of heat release oscillations at the individual flame segments cen-
K is a tuning parameter to alter the gain of the transfer func- ters, i.e. q0s for s = 1, 2, ...90 (cf. Fig. 3), along with the re-
tion, which can be physically interpreted as controlling the global spective segment’s volume Vs as well as the local mode shape
thermal power value. For each case, a ROM delievers reference values ψg,s /ψ f ,s . The latter mode shapes result from FEM simu-
values of T1 mode’s growth and damping rates by computing the lations of the Helmholtz equation in Eqn. 5. Finally, the slowly
complex eigenvalues of the system matrices of the closed and varying amplitudes F(t) and G(t) – required for method (3) –
open loop MIMO feedback model, respectively. The six oper- are obtained by employing the Hilbert transform to the Fourier
ation points are summarized in Tab. 1 in terms of pre-defined coefficient time series (cf. Eqns. 24-25).
growth and damping rates as well as associated values of absorp-
tion and gain parameter. Note that the oscillation frequency is
equal for all cases at wn /2π = 3001Hz.
SYSTEM IDENTIFCATION VERIFICATION TEST CASES
Synthetic Time Series Data This section utilizes the previously obtained synthetic time se-
Time domain simulations of all considered configurations in Tab. ries data – of which the underlying growth and damping rates
1 are performed using a 5th order Dormant-Price Runge Kutta in- are known – to test and verify system identification techniques in
tegration scheme. Time step and final time for the integration are successive order as above-introduced. Notice that the extraction
selected to dt = 1.5 · 10−5 sec. and T f = 30 sec. Representative of the oscillation frequency from time domain data is straight-
time traces of modal Fourier coefficients f (t) and g(t) and cor- forwardly achieved by Fast Fourier Transforms, and is thus, not
responding slowly varying amplitudes F(t) and G(t) for a stable discussed in the following.

9 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


(a) (a)
1.5 1.5
gR (t) fR (t) qa,R (t) P(qa,R )
Fourier coefficients (-)

Heat release signal (-)


1 1

0.5 P(gR ) 0.5

0 0

-0.5 P( fR ) -0.5

-1 -1
F(t)/G(t)
-1.5 -1.5
10 10.5 11 11.5 12 12.5 13 13.5 14 Probability 10 10.5 11 11.5 12 12.5 13 13.5 14 Probability
t(s) Density t(s) Density
(b) (b)
1.5 1.5
gR (t) fR (t) qa,R (t) P(qa,R )
Fourier coefficients (-)

1 1

Heat release signal (-)


0.5 0.5
P(gR )
0 0
P( fR )
-0.5 -0.5

-1 -1
F(t)/G(t)
-1.5 -1.5
10 10.5 11 11.5 12 12.5 13 13.5 14 Probability 10 10.5 11 11.5 12 12.5 13 13.5 14 Probability
t(s) Density t(s) Density

FIGURE 4. REPRESENTATIVE SIGNALS OF FOURIER COEF- FIGURE 5. INTEGRAL HEAT RELEASE SIGNAL OF (a) STABLE
FICENTS OF THE CCW ROTATING T1 MODE FOR ONE (a) (OP#2) AND (b) UNSTABLE (OP#5) OPERATION POINTS
LINEARILY STABLE AND (b) UNSTABLE OPERATION POINT.
PROBABILTY DENSITY FUNCTIONS P REVEAL THE NOISY
#5, along with the reconstructed counterpart is shown in Fig. 6,
AND LIMIT CYCLIC NATURE OF THE RESPECTIVE SIGNALS
which reveals satisfactory agreement.
AS PER NORMAL AND BIMODAL DISTRIBUTIONS.
Figure 7 displays the identified damping rates against the pre-
defined reference values. An overall accurate reproduction is
achieved, while relative errors remain below 4%.
(1) Damping Rate
The pure acoustic damping rate αn can be extracted for both,
linearly stable and unstable configurations (i.e. operation points (2) Growth Rate – Stable Case
OP #1 − #6 in Tab. 1). Procedurally, the transfer function be- The identification of growth rates for stable cases (OP#1 − #3
tween acoustic oscillation and integral heat release signal needs in Tab. 1) follows the same procedure as for the damping rates
to be computed using Eqn. 13 at first. Notice that for experimen- in the previous section – except the need of heat release signals.
tal setups, the distributed heat release signals can be retrieved Therefore, the autocorrelation is simply computed from acoustic
from dynamic OH∗ -recordings. The transfer function is used oscillation signals. Associated plots of the autocorrelation and fit
to determine the autocorrelation function of Eqn. 14. Required results look similar to Fig. 6. The comparison between reference
time traces of the integral heat release oscillation are displayed in and extracted growth rates in Fig. 7 reveal an acceptably correct
Figs. 5, representativly showing one stable (OP#2) and unstable reproduction with relative errors remaining below 6%.
(OP#5) operation point.
The autocorrelation is then fitted against the analytical expres- (3) Growth Rate – Unstable Case
sion in Eqn. 15 using a Bayesian method. The outcomes of this Finally, the identification of growth rates from linearly unstable
fit are the desired damping rate αn and oscillation frequency ωn . configurations (OP#4 − #6 in Tab. 1) is carried out. The ex-
Details on the advantages of applying Bayesian methods for pa- tracted and reference growth rates are shown in Fig. 7 revealing
rameter identification of thermoacoustic systems can be found an accurate reproduction of the former with relative errors re-
in [20]. The autocorrelation obtained from the time series of OP maining below 6%. Also, the growth rates due to flame driving

10 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


40
α n,ref 25 ν n,ref 50 ν n,ref
35 α n,fit ν n,fit ν n,fit
30 15
40
αn (rad/s)

βn (rad/s)
νn (rad/s)
25 5
30
20 −5
15 20
−15
10
−25 10
5
0 0
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
Operation Point (#) Operation Point (#) Operation Point (#)

FIGURE 7. SYSTEM IDENTIFICATION RESULTS – LEFT: DAMPING RATES , MIDDLE: NET GROWTH RATES, RIGHT: PURE FLAME
DRIVING VIA βn = νn + αn

1 G(t)
F(t)
kH,data (τ)
Slowly varying amplitudes (-)
0.8 1
kH, f it (τ)
0.6
0.8
0.4 OP#4 OP#5 OP#6
Autocorrelation (-)

0.2
0.6
Deterministic
0 amplitude
evolutions
-0.2 0.4

-0.4
0.2
-0.6

-0.8 0
0 5 10 15 20 25 30 35
3 4 5
-1 t(s)
0 2 4 6 8 10 12 14 16 18 20
τ(s) FIGURE 8. TIME TRACES OF SLOWLY VARYING AMPLI-
TUDES
FIGURE 6. RECONSTRUCTED AUTOCORRELATION FUNC-
TION FOR IDENTIFICATION OF DAMPING RATE SHOWN FOR
UNSTABLE OPERATION POINT OP #5 resentatively for OP#5 in Fig. 9, which supports the accuracy of
the extraction results.
If one seeks to employ the presented methodologies on measured
only are displayed in Fig. 7, which completes the energetic char- data of a real combustor, the following aspects should be consid-
acterization of the transversal thermoacoustic mode performance ered. First of all, it is certainly helpful to assess a given time
in terms of contributing components. The method of unstable series of the concerned combustor to identify whether it resem-
growth rate extraction relies on time traces of the slowly varying bles a linearly unstable (e.g. bimodal probability density) or sta-
amplitudes, which are shown in Fig. 8 for all three cases. Note ble (e.g. normal probability density), and then apply the corre-
that due the constant noise intensity, the perturbation of the am- sponding methodology. For this, inspecting the probability den-
plitude trace is stronger and weaker for smaller and larger growth sity distribution is a useful and reliable tool. However, such an
rates, respectively, although the linearization condition of Eqns. assessment is not per se necessary. The first method (damping
29-30 is not violated. rate) is equally applicable to both, stable and unstable operation
Finally, the autocorrelation of the amplitude perturbation G0 is points, where the stability state does not affect the damping rate.
compared against the reproduced counterpart (cf. Eqn. 34) rep- Applying the second method (stable growth rate) to a linearly

11 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


unstable case would yield zero valued growth rates. This zero such systems, three distinct identification methods were pre-
value is due to the general feature of a limit cycle, i.e. that en- sented:
ergy addition (flame driving) equals energy dissipation (acoustic
1. Extraction of the damping rate due to pure acoustic dissi-
damping). Thus, the second method automatically delivers infor-
pation mechanisms of linearly stable and unstable systems
mation about the stability state, where a negative and zero valued
from dynamic pressure and heat release data.
growth rate imply linearly stable and unstable scenarios, respec-
2. Extraction of the growth rate due to flame driving and acous-
tively. Finally, applying the third method (unstable growth rate)
tic damping mechanisms of linearly stable systems from dy-
to stable operation points should in principle work, and yield
namic pressure data.
negative growth rate values as is demonstrated in [8] for LF sys-
3. Extraction of the growth rate due to flame driving and acous-
tems. Consequently, the third method could also yield informa-
tic damping mechanisms of linearly unstable systems from
tion about the stability state of the concerned operation points as
dynamic pressure data.
per the sign of the identified growth rate (negative=stable, pos-
itive=unstable). However, employing the third method to stable In order to demonstrate these methods’ applicability, several
cases of HF systems has yet to be carried out in the future. verification test cases were conducted using a lab-scale swirl-
stabilized model gas turbine combustor as test case configuration.
The 3D wave equation including models for acoustic dissipation
1
and flame driving served as mathematical basis. This frame-
kGG,data (τ) kGG, f it (τ) work allowed to define several operation points featuring both,
0.9 linearly stable and unstable stability states at different levels of
growth and damping rates. Each of these operation points were
0.8
then simulated by means of non-compact Reduced Order Models
Autocorrelation (-)

0.7 to produce the desired synthetic times series to employ the sys-
tem identification routines. The reference growth and damping
0.6 rate values could be accurately extracted from the respective time
0.5
series. Consequently, the presented methods could be rendered
applicable to real thermoacoustic systems governed by high fre-
0.4 quency, transversal modes.
Future work comprises the application of the identification meth-
0.3
ods to time traces from measurements in order to extract growth
0.2 and damping rates of transversal thermoacoustic modes from ex-
0 2 4 6 8 10 12 14 16
perimental setups. These values will then be compared against
τ(s) growth and damping rates computed via thermoacoustic analysis
FIGURE 9. FIT RESULTS – UNSTABLE CASE OP #5 tools for reciprocal validation. Furthermore, parameter studies
using the isolated values of pure acoustic dissipation and flame
driving will be carried out to generate more physical understand-
ing of transversal combustion dynamics.

CONCLUSIONS ACKNOWLEDGMENT
In this paper, three methodologies for the identification of acous- Financial support is provided by the Technische Universität
tic damping and growth rates from time domain data was pre- München – Institute for Advanced Study funded by the Ger-
sented. Knowledge of these quantities are of high technical rel- man Excellence Initiative. The investigations were conducted
evance due the employment within the following engineering as part of the joint research programme COOREFLEX-turbo in
tasks: the frame of AG Turbo. The work was supported by the Bun-
1. Thermoacoustic quantification of combustion systems. desministerium für Wirtschaft und Technologie (BMWi) as per
2. Design of damping devices. resolution of the German Federal Parliament under grant number
3. Validation of analysis tools, flame and damping models. 03ET7021T, which is gratefully acknowledged.
4. Individual quantification of pure flame driving and acoustic
damping.
REFERENCES
Specific focus was exerted on non-compact, high-frequency ther- [1] Sattelmayer, T., 2010. Grundlagen der Verbrennung in sta-
moacoustic systems that are governed by transversal modes. For tionären Gasturbinen. Springer Verlag, ch. 9 in: Stationäre

12 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Gasturbinen, 2. neu bearbeitete Auflage, pp. 397–452. 039)TP/103 in RTO AGARDograph AG-AVT-039.
[2] Rayleigh, J., 1945. “The Theory of Sound”. Dover Publi- [16] Cveticanin, L., 1992. “Approximate analytical solutions
cations,New York, 1-2. to a class of non-linear equations with complex functions”.
[3] Bechert, D., 1980. “Sound Absorption Caused by Vortic- Journal of Sound and Vibration, 157(2), pp. 289–302.
ity Shedding, Demonstrated with a Jet Flow”. Journal of [17] Hummel, T., Berger, F., Schuermans, B., and Sattelmayer,
Sound and Vibration, 70(3), pp. 389–405. T., 2016. “Theory and Modeling of Non-Degenerate
[4] M.S.Howe, 1979. “On the Theory of Unsteady High Transversal Thermoacoustic Limit Cycle Oscillations”. In-
Reynolds Number Flow Through a Circular Aperture”. ternational Symposium on Thermoacoustic Instabilities
Proceedings of the Royal Society A, 366, pp. 205–223. in Gas Turbines and Rocket Engines: Industry meets
[5] Howe, M., 1980. “The Dissipation of Sound at an Edge”. Academia, GTRE-038.
Journal of Sound and Vibration, 70(3), pp. 407–411. [18] Gardiner, C., 2009. Stochastic Methods. Springer, Berlin.
[6] Dowling, A., 1997. “Nonlinear Self-Excited Oscillations [19] Ahlfors, L. V., 1966. Complex analysis: An introduction
of a Ducted Flame”. Journal of Fluid Mechanics, 346, to the theory of analytic functions of one complex variable.
pp. 271–290. McGraw-Hill.
[7] Noiray, N., and Schuermans, B., 2012. “Theoretical and [20] Stadlmair, N. V., Hummel, T., and Sattelmayer, T., 2017.
Experimental Investigations on Damper Performance for “Thermoacoustic Damping Rate Determination from Com-
Suppression of Thermoacoustic Oscillations”. Journal of bustion Noise using Bayesian Statistics”. In Proceedings of
Sound and Vibration, 331, pp. 2753–2763. ASME Turbo Expo 2017, no. GT2017-63338.
[8] Noiray, N., 2016. “Linear Growth Rate Estimation From [21] Noiray, N., and Schuermans, B., 2013. “On the Dynamic
Dynamics and Statistics of Acoustic Signal Envelope in Nature of Azimuthal Thermoacoustic Modes in Annular
Turbulent Combustors”. Journal of Engineering for Gas Gas Turbine Combustion Chambers”. Proceedings of the
Turbines and Power, 139(4), pp. 041503–041503–11. Royal Society A, 469.
[9] Noiray, N., and Schuermans, B., 2013. “Deterministic [22] Roberts, J. B., and Spanos, P. D., 1986. “Invited Review
Quantities Characterizing Noise Driven Hopf Bifurcations No. 1 Stochastic Averaging: An Approximate Method of
in Gas Turbine Combustion Chambers”. International Solving Random Vibration Problems”. International Jour-
Journal of Non-Linear Mechanics, 50, pp. 152–163. nal of Non-Linear Mechanics, 21(2), pp. 111–134.
[10] Lieuwen, T., 2005. “Online Combustor Stability Margin [23] Noiray, N., Bothien, M. R., and Schuermans, B., 2011.
Assessment Using Dynamic Pressure Data”. Journal of En- “Investigation of azimuthal staging concepts in annular
gineering for Gas Turbines and Power, 127(3), p. 478. gas turbines”. Combustion Theory and Modelling, 15:5,
[11] Boujo, E., A.Denisov, Schuermans, B., and Noiray, N., pp. 585–606.
2016. “Quantifying Acoustic Damping using Flame [24] Hummel, T., Temmler, C., Schuermans, B., and Sattel-
Chemiluminescence”. Journal of Fluid Mechanics, 808, mayer, T., 2015. “Reduced Order Modeling of Aeroacous-
pp. 245–257. tic Systems for Stability Analyses of Thermoacoustically
[12] Berger, F., Hummel, T., Hertweck, M., Schuermans, B., and Non-Compact Gas Turbine Combustors”. Journal of Engi-
Sattelmayer, T., 2016. “High-Frequency Thermoacoustic neering for Gas Turbines and Power, 138(5), pp. 051502–
Modulation Mechanisms in Swirl-Stabilized Gas Turbine 051502–11.
Combustors, Part I: Experimental Investigation of Local [25] Hummel, T., Hammer, K., Romero, P., Schuermans, B., and
Flame Response”. Journal of Engineering for Gas Turbines Sattelmayer, T., 2016. “Low-Order Modeling of Nonlin-
and Power, 139(7), pp. 071501–071501–9. ear Transversal Thermoacoustic Oscillations in Gas Tur-
[13] Hummel, T., Berger, F., Hertweck, M., Schuermans, B., bine Combustors”. Journal of Engineering for Gas Tur-
and Sattelmayer, T., 2016. “High-Frequency Thermoacous- bines and Power, 139(7), pp. 071503–071503–11.
tic Modulation Mechanisms in Swirl-Stabilized Gas Tur-
bine Combustors, Part II: Modeling and Analysis”. Jour-
nal of Engineering for Gas Turbines and Power, 139(7),
pp. 071502–071502–10.
[14] Bourgouin, J.-F., Durox, D., Moeck, J. P., Schuller, T., and
Candel, S., 2013. “Self-Sustained Instabilities in an An-
nular Combustor Coupled by Azimuthal and Longitudinal
Acoustic Modes”. ASME GT2013-95010, June 3-7, San
Antonio, Texas, USA.
[15] F.E.C. Culick, 2006. Unsteady Motions in Combustion
Chambers for Propulsion Systems. Number AC/323(AVT-

13 Copyright © 2017 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 08/22/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like