Professional Documents
Culture Documents
Docslide - Us Combustionj PDF
Docslide - Us Combustionj PDF
Turbine Engines:
Operational Experience,
Fundamental Mechanisms,
and Modeling
Edited by
Timothy C. Lieuwen
School of Aerospace Engineering
Georgia Institute of Technology
Atlanta, Georgia
Vigor Yang
Department of Mechanical Engineering
The Pennsylvania State University
University Park, Pennsylvania
Volume 210
PROGRESS IN
ASTRONAUTICS AND AERONAUTICS
in the United States of America. All rights reserved. Reproduction or translation of any part
of this work beyond that permitted by subsections 107 and 108 of the U.S. Copyright Law
without the permission of the copyright owner is unlawful. The code following this statement
indicates the copyright owner’s consent that copies of articles in this volume may be made
for personal or internal use, on condition that the copier pay the per-copy fee ($2.00) plus the
per-page fee ($0.50) through the Copyright Clearance Center, Inc., 222 Rosewood Drive,
Danvers, Massachusetts 01923. This consent does not extend to other kinds of copying,
for which permission requests should be addressed to the publisher. Users should employ
the following code when reporting copying from this volume to the Copyright Clearance
Center:
Data and information appearing in this book are for informational purposes only. AIAA is
not responsible for any injury or damage resulting from use or reliance, nor does AIAA
warrant that use or reliance will be free from privately owned rights.
ISBN 1-56347-669-X
Progress in Astronautics and Aeronautics
Editor-in-Chief
Frank K. Lu
University of Texas at Arlington
Editorial Board
David A. Bearden Richard C. Lind
The Aerospace Corporation University of Florida
Gas turbines have made substantial gains in performance since their initial demon-
stration in jet powered aircraft and power turbines. The performance, noise char-
acteristics, and pollutant emissions of gas turbines for propulsive applications
continue to improve. On the ground, contemporary gas turbines produce higher
operating efficiencies and emit fewer pollutants than other major chemical-energy
conversion devices. In addition, the low capital investment, ease of permitting, and
quick installation have made them attractive to investors. As a result, gas turbines
have become a dominant technology for new power generating capacity in the
United States and worldwide.
A variety of factors have contributed to the popularity of gas turbine technology.
Financing considerations are the key high-level driver. Pollutant emissions play
another important role, particularly in motivating the specific technology improve-
ments and innovations over the last decade. For example, in the United States, the
Clean Air Act Amendments of 1990 imposed strict guidelines on the control of
nitrogen oxides, NOx, which, along with SO2 , is a major contributor to acid rain
This book focuses on a particularly serious difficulty in low emissions gas tur-
bines: combustion-driven oscillations. These instabilities routinely constrain the
operating envelope and power output of fielded machines and, in some cases, lead to
serious damage of hot section components. Gas turbine users have found that com-
ponents such as combustor liners, transition pieces, and fuel nozzles need routine
examination for part cracking or excessive wearing because of vibration-induced
fretting. At a minimum, this requires downtime for inspections and part repair,
thereby reducing machine availability. At the worst, a cracked piece may be liber-
ated into the hot gas path, potentially requiring replacement of expensive turbine
components. In addition, users in certain geographic areas have found that engines
must be seasonally retuned to eliminate oscillations due to ambient temperature
changes. The cost for the repair and replacement of hot section components, much
of which is directly attributable to the combustion instability problem, exceeds
$1 billion annually and constitutes up to 70% of the nonfuel costs of F-class gas
turbines. Major power generating companies have suffered losses in the hundreds
of millions of dollars because of lost revenue from forced outages, resulting in a
number of lawsuits.
Although instabilities have not been nearly as severe a problem in nonpremixed
aero engine combustors, they have appeared in a few cases and posed serious
challenges in the development stage. Military engines, however, have experienced
major problems with low-frequency instabilities in augmentors. A large-scale ef-
fort is currently underway at several gas turbine manufacturers in the United States,
in cooperation with the U.S. Air Force, to overcome such difficulties.
Over the last decade, substantial efforts have been expended in the industrial,
government, and academic communities to understand the unique issues associated
with combustion instabilities in low-emissions gas turbines. The objective of this
book is to compile these results into a series of chapters that address various
xiii
xiv PREFACE
facets of the problem. In planning this volume, it was decided to include a few
comprehensive chapters, rather than a large number of more narrowly focused
contributions. As such, it was not possible to solicit articles from every contributor
to the field, although it is certainly our hope that all relevant works are appropriately
represented in the book.
Following the overview, the book is organized into four basic sections: The
Case Studies section compiles chapters from gas turbine manufacturers and users
that detail specific experiences with combustion instabilities in the development
stage and in fielded turbine engines. These chapters describe the basic instability
mitigation approaches that were developed and the tradeoffs encountered between
instabilities and other performance metrics, such as NOx emissions. The Funda-
mental Processes and Mechanisms section addresses the basic phenomenology
of combustion instabilities in premixed and nonpremixed combustors, the mech-
anisms through which unsteady heat release processes may become self-excited,
and measurement techniques for characterizing them. Next, the Modeling and Di-
agnostics section describes analytical and computational approaches to model the
complex acoustic characteristics of combustor geometries and the interactions be-
tween flames and acoustic waves. Finally, the Combustion Instability and Control
section addresses active and passive control of combustion instabilities, including
an industry perspective into approaches for incorporating instability considerations
into the design process.
Publication of this volume was made possible through the substantial contri-
butions of a number of individuals. We would like to first thank the authors for
sharing their time and talent in preparing their manuscripts and carefully revising
them. The invaluable assistance of Rodger Williams, Heather Brennan, and Janice
Saylor of the AIAA in the preparation of the volume for publication is gratefully
acknowledged. Last, but by no means least, we wish to thank Danning You and
Yanxing Wang for providing the technical drawing services.
Timothy C. Lieuwen
Vigor Yang
July 2005
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
I. Overview
vii
viii
I. Introduction
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ David S. Lewis Jr. Chair and Regents’ Professor, School of Aerospace Engineering.
† Associate Professor, School of Aerospace Engineering.
3
4 B. T. ZINN AND T. C. LIEUWEN
be shown that the combustion process adds (removes) energy from the acoustic
oscillations locally if the integral in Fig. 1.1, which is often referred to as Rayleigh’s
integral,1 is positive (negative). The sign of this integral depends on the phase dif-
ference between the heat-release and pressure oscillations and is positive (negative)
when this phase difference is smaller (larger) than 90 deg. As shown in Fig. 1.1,
combustion instability spontaneously occurs only if the energy supplied to the
acoustic mode by the combustion process exceeds the energy losses of the mode
caused by, for example, radiation and convection of acoustic energy out of the
combustor (e.g., through the nozzle), viscous dissipation, and heat transfer. Thus,
as long as the magnitude of the driving exceeds the magnitude of the damping
process, the energy of the mode will increase with time. In such a case, the ampli-
tude of oscillations initially increases exponentially with time until it saturates at
some limit-cycle amplitude. When this occurs, the time averages of the driving and
damping processes are equal and no net energy is added to the oscillating mode.
In general, combustion instabilities occur at frequencies associated with natural
acoustic modes of the combustor. These include, for example, bulk (i.e., Helmholtz-
type oscillations), axial, and transverse (i.e., tangential and/or radial) modes (see
Fig. 1.2). On occasion, however, the oscillations are not associated with a purely
acoustic mode and are excited by a coupled convective-acoustic mode similar to
cavity tones,2 which often occur at frequencies lower than those of purely acoustic
modes. Such oscillations occur when an entropy wave (i.e., a hot-gas packet) or
a vortex generated in the flame region is convected toward (and impinges on) the
nozzle, at which it excites an acoustic wave that propagates back to the flame,3,4
exciting another convected wave and thus repeating the process. These types of
modes are often encountered in systems that are operating at conditions close to
flame blowoff.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 5
Longitudinal
Because the initial amplitudes of most instabilities are generally quite small,
their characteristics are described by the linear-wave equation.5 The frequencies
and mode shapes of these oscillations and the conditions under which they sponta-
neously occur are determined by the solutions of these equations. Linear analyses
cannot, however, predict the magnitude of the limit-cycle amplitude attained by the
instability because it is controlled by nonlinear processes. Furthermore, nonlinear
processes may allow a large-amplitude disturbance whose amplitude exceeds a
certain threshold value, A T , to trigger instability in a system that is linearly stable;
that is, one in which low-amplitude oscillations are not spontaneously self-excited.
Consequently, both the characteristics of the limit-cycle oscillations and conditions
under which finite amplitude disturbances trigger instabilities can only be deter-
mined by solving the nonlinear equations that describe the system’s dynamics.6,7
To prevent the onset of detrimental combustion instabilities, the processes re-
sponsible for their driving and damping must be understood. This chapter presents
an overview of the current understanding of these processes with the objective of
providing the reader with the background needed for the more detailed discussions
of these and related subjects in the remaining chapters of this volume. To attain this
goal, the following topics are discussed in this chapter: the history of this problem,
driving and damping of combustion instabilities, common instability mechanisms
in gas turbines, the initial growth and saturation of the oscillations, and some basic
characteristics of limit-cycle oscillations.
p' – pressure
oscillation
u' – velocity
oscillation
Fuel+Air
Fig. 1.3 An example of a flame-driven instability in a tube open at both ends.
Fig. 1.5 Photograph of rocket injector with welded baffle plates that was tested to
prevent transverse mode instabilities during F-1 engine development.26
8 B. T. ZINN AND T. C. LIEUWEN
where p (x, t), q (x, t), V, T , and L i are the combustor pressure oscillations, heat-
addition oscillations, combustor volume, period of the oscillations, and the ith
acoustic energy loss process (e.g., viscous dissipation, radiation of acoustic en-
ergy out of the combustor through its boundaries), respectively. The equal sign in
Eq. (1.1) describes conditions when limit-cycle oscillations are attained and the
time average of the energies added and removed from the oscillations are equal.
The integral on the left side of Eq. (1.1) is referred to as the Rayleigh integral and
is often used in experimental or numerical studies to quantify the energy transfer
from the combustion process to the acoustic field. Note that the inner integral on
the left side of Eq. (1.1) is the integral shown in Fig. 1.1, which describes the
local driving/damping of the acoustic oscillations by the heat-addition process
q (x, t).
B. Instability-Driving Mechanisms
As discussed earlier, combustion instabilities are excited by feedback between
the combustion process and acoustic oscillations that depends on the system char-
acteristics and operating conditions. This section provides a brief overview of
common instability mechanisms and the conditions under which they are self-
exciting.
Figure 1.6 describes the generic feedback loop responsible for combustion in-
stabilities. It consists of the following sequence of events: 1) Fluctuations in the
velocity and/or thermodynamic-state variables induces a fluctuation in the heat-
release rate, 2) the heat-release fluctuation excites acoustic oscillations, and 3) the
acoustic oscillations generate the velocity and thermodynamic-state variable fluc-
tuations that are described in step 1, thus closing the feedback loop. Depending
on the relative magnitudes of the energy added and removed from the acoustic
oscillations, the amplitude of oscillations may decrease, remain constant, or grow
during each cycle of this loop.
Several mechanisms capable of driving combustion instabilities in gas turbines
have been identified, as indicated in Fig. 1.7. Because most of these mechanisms
10 B. T. ZINN AND T. C. LIEUWEN
Heat Release
Oscillations
Flow and
Mixture Acoustic
Perturbations Oscillations
Fig. 1.6 Illustration of the feedback processes responsible for combustion instability.
are discussed in more detail in Chapter 9, the remainder of this section only pro-
vides a brief description of each.
Fuel/air ratio
Fuel flow rate oscillations
oscillations
Flame area and
reaction rate
oscillations
Flow rate
oscillations Combustion
Products
Unsteady mixing,
vaporization,
atomization Vortex/flame
interactions
Fig. 1.7 Flow and flame processes that can cause combustion instabilities in
gas turbines.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 11
Fig. 1.8 Computed image of swirling flame distorted by vortical structures. Courtesy
of Y. Huang and V. Yang.40
T
a. Pressure
T
at flame
τ1
b. Process 1
τ2
Heat Release
Perturbation c. Process 2
due too: τ3
d. Process 3
τ4
e. Process 4
time
0.08
0.04 430 Hz
630 Hz
0.02
0
15 20 25 30 35 40
Premixer Velocity (m/s)
combustion instability must have magnitudes that are of the order of some integer
multiple of the acoustic period of the modes.
Note also that combustor pressure oscillations generally vary harmonically with
time and do not exhibit the pulselike behavior used in the examples in Fig. 1.9.
Consequently, the time delays and acoustic periods of the various modes need
not be exactly equal for the process to add energy to the acoustic field; instead,
these characteristic times must satisfy a relationship of the form T − T /4 < nτ <
T + T /4, where n is an integer; that is, n = 1, 2, . . . 17
Some of these points are illustrated by the results in Fig. 1.10,17 which describe
the dependence of the amplitudes of a 430- and 630-Hz combustor mode on the
mean velocity of the reactants in the combustor premixer. The instabilities in this
combustor were driven by a mechanism that depends on the time required to trans-
port the reactants from the fuel-injection point to the combustor,17 a time delay that
is inversely proportional to the mean velocity of the gases in the premixer. Con-
sequently, this time delay decreases as the gas velocity in the premixer increases.
Figure 1.10 shows that, as the velocity in the combustor premixer increased and
the convective time delay decreased, the amplitude of the lower-frequency mode
(i.e., 430 Hz with the longer period T ) decreased and the amplitude of the higher-
frequency mode (i.e., 630 Hz with the shorter period T ) increased. This finding
indicates that when the premixer velocity was low the driving process coupled
with the 430-Hz mode and this coupling “switched” to the 630-Hz mode as the
premixer velocity increased.
C. Damping Processes
As noted earlier, acoustic damping processes play an important role in the
determination of the conditions under which combustion instability occurs. This
section summarizes important damping mechanisms in combustors and the key
parameters that influence their magnitude.
Energy can be dissipated/removed from an unstable mode via the following three
processes: 1) transfer of acoustic energy to vortical or entropy disturbances through
viscous and heat-transfer processes, respectively; 2) convection and/or radiation
of acoustic energy out of the system; and 3) transfer of energy between acoustic
14 B. T. ZINN AND T. C. LIEUWEN
modes. Note that in all these cases, the “dissipation process” refers to the transfer
of acoustic energy out of the combustor, frequency regions, or modes of oscillation
in a manner that reduces the acoustic energy of the unstable mode.For example,
in the second mechanism, the acoustic energy is simply radiated or convected out
of the combustor through its boundaries and represents part of the noise heard
outside the system. The rest of this section describes each of these mechanisms in
further detail.
Reflection Coefficient
0.8
0.6
0.4
0.2
0
0 0.1 0.2 0.3 0.4
Mach Number
Fig. 1.11 Dependence of the reflection coefficient from an open-ended pipe on the
flow Mach number. Taken from Ingard and Singhal.41
εH
Driving/Damping
1 H(A)
D(A)
εD
1
A ALC
Fig. 1.12 Hypothetical dependence of the acoustic driving, H( A), and damping, D( A),
processes on the instability amplitude A.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 17
D(A), resulting in further growth of the disturbance. Because these two curves
diverge near the origin, their difference increases with amplitude, implying that
the amplitude growth rate increases with amplitude. When the amplitudes of the
oscillations are small, the driving and damping processes can be approximated
by linearized expressions; that is, H(A) = εH A and D(A) = εD A. Thus, linear
combustor processes (i.e., processes whose magnitude is directly proportional to
amplitude A), in general, control the balance between driving and damping pro-
cesses when the amplitude of oscillations is small and thus determine the frequency
and growth rate A ∼ eαt of inherent combustor disturbances. It can be shown that
the initial growth rate of the instability α is proportional to the difference between
the driving and damping processes in the linear regime; that is, α ∼ εH − εD .
For this reason, linear combustor stability models (such as those described in
Chap. 13) are routinely used to determine whether a given operating point is stable
or unstable.
Nonlinear combustor processes control the dynamics of the oscillations as the
driving and damping processes become amplitude dependent. Figure 1.12 de-
scribes a situation in which H(A) saturates and D(A) increases linearly with the
amplitude A, thus resulting in an intersection of the two curves at the limit-cycle
amplitude A LC . Note that this limit-cycle amplitude is stable; that is, a perturbation
of the amplitude to the left (right) of this intersection point causes H(A) to become
larger (smaller) than D(A), thus causing the amplitude to increase (decrease) and
return to its limit-cycle value A LC .
Next, consider a situation in which some combustor parameter is systematically
varied in such a way that εh increases while εd remains constant (see Fig. 1.12).
For εh < εd , A = 0 is the stable solution; that is, the system is linearly stable be-
cause all perturbations imposed on the system will decay. However, when εh > εd ,
the solution A = 0 becomes unstable, causing the amplitude of the disturbance to
increase toward a new stable, limit-cycle equilibrium. The εh = εd condition sep-
arates two regions of fundamentally different dynamics and is referred to as a
supercritical bifurcation point. These ideas are illustrated in the bifurcation dia-
gram in Fig. 1.13, which shows the dependence of the amplitude A LC on εh − εd .
Figure 1.13 shows that as εh − εd becomes positive, the system becomes linearly
Stable
Unstable
Amplitude
εh - εd
0.02
0.01
0.005
0
18 21 24 27 30
Mean Inlet Velocity (m/s)
H(A)
Driving/Damping
1
εD
D(A)
1 εH
A AT ALC
Fig. 1.16 Hypothetical dependence of the acoustic driving, H( A), and damping, D( A),
processes on amplitude A, that produce the triggering of instabilities.
Stable
Unstable
Amplitude
AT
εh-εεd
Fig. 1.17 An example of a subcritical bifurcation.
before. For εh > εd , the A = 0 solution becomes unstable (as indicated by the solid
line becoming dashed), and only a single stable solution is present. In this case, if
a system parameter is monotonically increased to change the sign of εh − εd from
a negative to a positive value, the system’s amplitude will jump discontinuously
from A = 0 to A = A LC at εh − εd = 0. Hysteresis is also present in the system,
because if the system parameter is subsequently decreased, the system’s amplitude
decreases as it follows the stable branch on top, even for a range of εh < εd
values, before it discontinuously “jumps” to the A = 0 solution. Experimental
data exhibiting such behavior are presented in Fig. 1.18.25
0.015
Normalized Pressure (p'/p)
0.0125
0.01
0.0075
0.005
0.0025
0
13 13.5 14 14.5 15 15.5
Mean Inlet Velocity (m/s)
Nonlinear processes are also responsible for two other phenomena observed in
unstable combustors: generation of harmonics and changes in the mean value of
certain system properties. These phenomena can be understood by considering
the response of a nonlinear system to harmonic forcing. Consider, for example,
the gas-dynamic nonlinearity generated by the isentropic relationship between
pressure and density; that is,
γ
p(t) ρ(t)
= (1.3)
p̄ ρ̄
Assume that the density oscillates harmonically with an amplitude A and angular
frequency ω, that is,
ρ (t)
= A sin ωt (1.4)
ρ̄
Substitute this expression into Eq. (1.3) and expand the resulting expression in
a Taylor series about A = 0 to obtain the following expression for the pressure
oscillations:
p (t) (γ − 1) 2 (γ − 1)(γ − 2) 3 (γ − 1) 2
= A + A+ A sin ωt − A cos 2ωt
γ p̄ 4 8 4
(γ − 1)(γ − 2) 3
− A sin 3ωt + · · · (1.5)
24
Equation (1.5) shows that the expansion for the pressure consists of a linear term
that is proportional to A (the second term on the right), and a collection of nonlinear
terms that are proportional to higher powers of the amplitude A. The latter includes
a time-independent term (first term on the right) that is proportional to A2 and
represents the change in the mean pressure caused by the purely oscillatory density
disturbance. Furthermore, the expansion includes harmonics that are proportional
to A2 , oscillating at twice the disturbance frequency, and terms proportional to
A3 , oscillating with frequencies of 3ω and ω. Such harmonics of the unstable
combustor mode are routinely observed.
Changes in mean pressure of up to several hundred pounds per square inch have
been observed during very-large-amplitude instabilities in solid rockets. On the
other hand, instabilities in gas turbines generally have much smaller magnitudes,
and no reports have been published of significant shifts in mean pressure because
of these instabilities. However, changes in mean flame location and length are
routinely observed.28,29 These changes, in turn, can have an impact on the static
pressure distribution in the combustor because it depends on the distribution of the
combustion-process heat release.
C. Causes of Nonlinearities
Numerous processes that drive and damp oscillations in unstable combustors de-
pend nonlinearly on the instability amplitude. These nonlinearities may be caused
22 B. T. ZINN AND T. C. LIEUWEN
by flow and combustion processes that occur either within the combustor volume
or at its boundaries.
Gas-dynamic nonlinearities within the combustor volume are introduced by
processes described by nonlinear terms in the Navier–Stokes and energy equations,
for example, convective terms such as ρ u · ∇ u in the Navier–Stokes equation
or the previously discussed nonlinear pressure–density relationship. Such terms,
in general, become significant when the amplitudes of the fluctuating pressure,
density, or velocity become on the order of the mean pressure, density, or speed of
sound, respectively. Consequently, these terms are generally not important when
the relative amplitudes of the acoustic disturbances are low (e.g., p / p̄ < 10%).
On the other hand, when very-large-amplitude oscillations are encountered, these
nonlinear processes strongly affect the characteristics of the instabilities in these
systems. Examples of the treatments of these nonlinear processes can be found in,
for example, the works of Zinn and coworkers,30 Culick and coworkers,31,32 and
Yang.33
Combustion-process nonlinearities are introduced by the nonlinear dependence
of the heat-release oscillations on the disturbance amplitude. Additionally, flow
oscillations may control the response of the combustion process, resulting (in
these cases) in nonlinearities that become important when the ratio u /ū ∼ O(1).25
Consequently, in these cases, the relevant velocity scale that determines when
nonlinearities are important is the mean velocity and not the sound speed (as in
the gas-dynamic nonlinearities). In fact, it has been shown that such combustion-
process nonlinearities play a key role in the stability of lean premixed combus-
tion systems.34 Further discussion of these nonlinearities can be found in Chap-
ters 12 and 13.
The nonlinearities in processes that occur at or near the combustor bound-
aries also affect the combustor dynamics as they are introduced into the anal-
ysis of the problem through nonlinear boundary conditions. Such nonlinearities
are caused by, for example, flow separation at sharp edges or rapid expansions,
which, as discussed earlier, cause stagnation-pressure losses and a corresponding
transfer of acoustic energy into vorticity. The resulting nonlinear damping is pro-
portional to ρ| u |2 , resulting in an unsteady damping process that is proportional
u |2 + 2|
to ρ̄(| u | · |u¯ |) + ρ (|u¯ |2 + |
u |2 + 2|
u | · |u¯ |).35 This expression indicates
that the presence of mean flow introduces linear and nonlinear damping terms and
that the damping is proportional to the square of the magnitude of the velocity am-
plitude | u |2 in the absence of mean flow (i.e., |u¯ | = 0). Such nonlinear damping
has been well characterized in experiments (see Ref. 35, for example).
Also, wave reflection and transmission processes through choked and unchoked
nozzles become amplitude dependent at large amplitudes. The effect of these
processes on the instabilities is described by a nonlinear boundary condition.36
V. Conclusion
This chapter concludes with a summary of the state of the art of current model-
ing capabilities of combustion instabilities. From a practical point of view, three
basic instability characteristics must be predicted and understood: 1) frequency of
oscillations, 2) conditions under which the oscillations occur, and 3) their final,
limit-cycle amplitude, which are listed in increasing order of predictive difficulty.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 23
Specifically, the frequency and mode shape are easier to predict because they gen-
erally only require an understanding of the system’s linear dynamics, whereas
determination of the limit-cycle amplitude requires an understanding of nonlinear
system characteristics.
Consider first the prediction of instability frequencies. Although some fun-
damental problems remain, such as analytical descriptions of the combustor’s
acoustic boundary conditions, prediction of instability frequencies is a relatively
mature area. In fact, capabilities for modeling the acoustics of combustors are rea-
sonably well developed, as described in Chapter 13, in this volume and also, for
example, in recent proceedings of the American Society of Mechanical Engineers
Turbo Expo conferences or in Munjal’s book.37 These descriptions indicate that
the frequencies and mode shapes of the excited instabilities can often be predicted
with good accuracy. In general, accurate predictions of these properties simply
require knowledge of the geometric characteristics of the system and average tem-
perature distributions. Although unsteady heat-release effects, which are much
harder to predict, have some impact on instability frequencies, this impact is often
small.
Predicting the conditions under which instabilities occur (the second issue) is
considerably more difficult than predicting acoustic mode shapes and frequencies,
because it requires knowledge and modeling of the interactions of flow and mixture
disturbances with flames and damping processes. In recent years, the focus of
much of the research in this area has been on developing these understanding
and modeling capabilities. Examples of the findings of some of these studies that
investigated, for example, acoustic wave–flame interactions and other mechanisms
that can drive these instabilities are discussed Chapters 9 and 12. Much progress
has been made in these areas to date and many gas-turbine manufacturers have
reported success in predicting instability frequencies and mode shapes, and the
conditions under which they occur, as discussed in the subsequent case-study
chapters in this book. Furthermore, even in cases in which predictive capabilities
do not exist, reasonable understanding of qualitative combustor stability trends
can often be obtained through analysis of combustor data after they have been
obtained; for example, rational correlations of the data can be developed.
Note that direct experimental verification of linear combustor dynamics models
and mechanisms of instability is difficult, because instabilities, in general, are
experimentally studied under limit-cycle conditions (with the notable exception
of Ref. 38). For example, consider the role of vortex shedding, which is thought
to be an important instability mechanism in gas turbines. The mere observation
of vortex shedding at the instability frequency under limit-cycle conditions does
not necessarily imply that it is also responsible for initiating the instability. Vortex
shedding can be forced at the frequency of an external excitation when the forcing
occurs at sufficiently large amplitude, even if this frequency does not coincide with
the natural shedding frequency.39 Consequently, flow oscillations may be excited
by some other instability mechanism, which subsequently forces the shedding of
vortices at the instability frequency.
In contrast to the prediction of the instability characteristics and the condi-
tions under which it spontaneously occurs, which can be determined with linear
combustor models, the prediction of the limit-cycle amplitude of the instability
and the conditions under which large-amplitude disturbances destabilize a linearly
24 B. T. ZINN AND T. C. LIEUWEN
References
1
Rayleigh, J. S. W., The Theory of Sound, Vol. 2, Dover, New York, 1945.
2
Crighton, D., “Airframe Noise,” Aeroacoustics of Flight Vehicles, edited by H. H.
Hubbard, Acoustical Society of America, New York, 1995.
3
Yu, K., Trouve, A., and Daily, J., “Low-Frequency Pressure Oscillations in a Model
Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, 1991, pp. 47–72.
4
Marble, F., and Candel, S., “Acoustic Disturbance from Gas Non-uniformity Convected
Through a Nozzle,” Journal of Sound Vibrations, Vol. 55, 1977, pp. 225–243.
5
Crocco, L., and Cheng, S., Theory of Combustion Instability in Liquid Propellant Rocket
Motors, Butterworths Scientific Publications, London, 1956.
6
Yang, V., Kim, S. I., and Culick, F. E. C., “Triggering of Longitudinal Pressure Os-
cillations in Combustion Chambers: I: Nonlinear Gasdynamics,” Combustion Science and
Technology, Vol. 72, 1990, pp. 183–214.
7
Wicker, J. M., Greene, W. D., Kim, S. I., and Yang, V., “Triggering of Longitudinal
Combustion Instabilities in Rocket Motors: Nonlinear Combustion Response,” Journal of
Propulsion and Power, Vol. 12, 1996, pp. 1148–1158.
8
Swift, G., Thermoacoustics, Acoustical Society of America, New York, 2002.
9
Rott, N., “Damped and Thermally Driven Acoustic Oscillations,” Zeitschrift fuer Ange-
waxdte Mathematik und Physik, Vol. 20, 1969, p. 230.
10
Jones, A. T., “Singing Flames,” Journal of the Acoustical Society of America, Vol. 16,
No. 4, 1945, pp. 254–266.
11
Le Conte, J., Philosophical Magazine HP, 235.
12
Putnam, A., Combustion Driven Oscillations in Industry, American Elsevier Publishers,
New York, 1971.
13
Pun, W. Ph.D. Thesis, California Inst. of Technology, Pasadena, CA, 1991.
14
Blomshield, F. S., “Historical Perspective of Combustion Instability in Motors: Case
Studies,” AIAA Paper 2001-3875, 2001.
15
Culick, F., “Combustion Instabilities in Liquid-Fueled Propulsion Systems-An
Overview,” AGARD, 1977.
16
Kendrick, D. W., Anderson, T. J., and Sowa, W. A., “Acoustic Sensitivities of Lean-
Premixed Fuel Injectors in a Single Nozzle Rig,” American Society of Mechanical Engi-
neers, Paper 98-GT-382, 1998.
17
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182–190.
COMBUSTION INSTABILITIES: BASIC CONCEPTS 25
18
Candel, S., “Combustion Dynamics and Control: Progress and Challenges,” Proceed-
ings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
19
Hegde, U. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Instabilities in Dump Type Ramjet Combustors,” Combustion Science and Technology,
Vol. 55, 1987, pp. 125–138.
20
Schadow, K., and Gutmark, E., “Combustion Instability Related to Vortex Shedding
in Dump Combustors and Their Passive Control,” Progress in Energy and Combustion
Science, Vol. 18, pp. 117–132, 1992.
21
Temkin, S., Elements of Acoustics, Wiley, New York, 1981.
22
Roberson, J., and Crowe, C., Engineering Fluid Mechanics, Houghton Mifflin, New
York, 1993.
23
Kim, J. S., “Effects of Turbulence on Linear Acoustic Instability: Spatial Inhomogene-
ity,” Liquid Rocket Engine Combustion Instability, edited by V. Yang and W. Anderson,
AIAA, Washington, DC, 1994, Chap. 16.
24
Lieuwen, T., Neumeier, Y., and Rajaram, R., “Measurements of Incoherent Acoustic
Wave Scattering from Turbulent Premixed Flames,” Proceedings of the Combustion Insti-
tute, Pittsburgh, PA, Vol. 29, 2002.
25
Lieuwen, T., “Experimental Investigation of Limit Cycle Oscillations in an Unsta-
ble Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, No. 1, 2002,
pp. 61–67.
26
Oefelein, J. C., and Yang, V., “Comprehensive Review of Liquid-Propellant Combustion
Instabilities in F-1Engines,” Journal of Propulsion and Power, Vol. 9, 1993, pp. 657–
677.
27
Drazin, P. G., and Reid, W. H., Hydrodynamic Stability, Cambridge Univ. Press,
Cambridge, England, U.K., 1981.
28
Polifke, W., Fischer, A., and Sattelmayer, T., “Instability of a Premix Burner with Non-
Monotonic Pressure Drop Characteristics,” Journal of Engineering for Gas Turbines and
Power, Vol. 125, No. 1, 2003, pp. 20–27.
29
Broda, J. C., Seo, S., Santoro, R. J., Shirhattikar, G., and Yang, V., “An Experimental
Investigation of Combustion Dynamics of a Lean, Premixed Swirl Injector,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1849–1856.
30
Zinn, B. T., and Powell, E. A., “Nonlinear Combustion Instability in Liquid- Propellant
Rocket Engines,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 13, 1970.
31
Culick, F. E. C., Burnley, V., and Swenson, G., “Pulsed Instabilities in Solid-Propellant
Rockets,” Journal of Propulsion and Power, Vol. 11, No. 4, 1995, pp. 657–665.
32
Culick, F. E. C., “Nonlinear Growth and Limiting Amplitude of Acoustic Oscillations
in Combustion Chambers,” Combustion Science and Technology, Vol. 3, No. 1, 1971.
33
Wicker, J. M., Greene, W. D., Kim, S.-I., and Yang, V., “Triggering of Longitudinal
Combustion Instabilities in Rocket Motors: Nonlinear Combustion Response,” Journal of
Propulsion and Power, Vol. 12, No. 6, 1996.
34
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat Release/Acoustic Model for
Thermo-Acoustic Instability in Lean Premixed Combustors,” Journal of Engineering for
Gas Turbines and Power, Vol. 121, 1999.
35
Zinn, B. T., “A Theoretical Study of Nonlinear Damping by Helmholtz Resonators,”
Journal of Sound Vibrations, Vol. 13, No. 3, pp. 347–356.
36
Zinn, B. T., and Crocco, L, “Periodic Finite Amplitude Oscillations in Slowly Con-
verging Nozzles,” Astronautica Acta, Vol. 13, Nos. 5 and 6, Aug. 1968, pp. 481–488.
37
Munjal, M., Acoustics of Ducts and Mufflers, Wiley, New York, 1987.
26 B. T. ZINN AND T. C. LIEUWEN
38
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., Esposito, E., and Surget, J., “Ini-
tiation and Suppression of Combustion Instabilities by Active Control,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 22, pp. 1363–1370, 1988.
39
Blevins, “The Effect of Sound on Vortex Shedding from Cylinders,” Journal of Fluid
Mechanics, Vol. 161, 1985, pp. 217–237.
40
Huang, Y., and Yang, V., “Effect of Swirl on Combustion Dynamics in a Lean-Premixed
Swirl-Stabilized Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 30, 2004, pp. 1771–1778.
41
Ingard and Singhal, “Effect of Flow on the Acoustic Resonances of an Open Ended
Duct,” Journal of the Acoustical Society of America, Vol. 58, No. 4, 1975, pp. 788–793.
II. Case Studies
Chapter 2
I. Introduction
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
∗ Manager, Advanced Combustion.
† Senior Principal Combustion Engineer.
29
30 K. O. SMITH AND J. BLUST
range in size (outer diameter) from approximately 30 in. (Centaur) to 38 in. (Titan).
Table 2.1 presents a summary of Solar’s turbine product line.
At the time of LP combustor development, the Centaur engine full-load com-
bustor inlet conditions were 100 psia and 600◦ F (690 kPa and 590 K). The corre-
sponding Mars conditions were 220 psia and 800◦ F (1520 kPa and 700 K). The
Centaur combustor has outer and inner diameters of 30 and 24 in. (76 and 61 cm),
respectively; 12 fuel injectors are used. The outer and inner diameters of the Mars
liner are 36 and 28 in. (91 and 71 cm), respectively; the Mars uses 14 fuel injectors.
Fuel Injector Combustor Housing
Combustor Liner
Turbine Nozzle
AIR
INLET
MARS
INJECTOR
PREMIX
DUCT
Fig. 2.3 Cross section of typical lean premixed gaseous fuel injector.
COMBUSTION INSTABILITIES IN INDUSTRIAL GAS TURBINES 33
into the combustor. The pilot flame, therefore, has stability characteristics that
are more like those of a diffusion flame. The pilot injector was originally used to
provide a stable flame for engine startup, low-load operation, and off-load tran-
sients. The pilot has since proved invaluable for overcoming CPOs. In fact, Solar’s
experience indicates that the most productive means of achieving reductions in
CPOs is through optimization of the pilot specifically and the fuel injector more
generally.
Solar’s initial LP combustor liner designs used traditional film cooling to keep
liner temperatures low. Since then, liner cooling has evolved to either effusion
cooling or convective backside cooling, in which no cooling air is injected into
the combustion zone. During this period of development, the number and size of
holes in the combustor liners have varied widely. To date, however, no signifi-
cant relationship has been identified between the method of liner cooling and the
occurrence of high CPOs.
Fig. 2.4 Liner internal splash plate damaged by high-amplitude combustor pressure
oscillations.
5
T6102 S/N 001 BLD 3 3/24/92
4021 KW 100 PCT LOAD
Pressure Oscillation (arbitrary units)
LOG
5E-4
0 1000 2000
Fig. 2.5 Typical experimental pressure frequency spectrum showing amplitude peak
at 440 Hz.
COMBUSTION INSTABILITIES IN INDUSTRIAL GAS TURBINES 35
4 50
30
NOx Target
2
20
1
Dynamic Pressure 10
New Design
0 0
0 2 4 6 8 10 12 14 16
PILOT FUEL, %
Fig. 2.6 Pilot fuel effect on combustor pressure oscillations and NOx emissions.
could most easily be varied were rig tested to quantify their effect on CPOs. These
elements included injector premixing barrel inner diameter and the distance from
the fuel-injection spokes to the injector exit plane. These parameters affect the
injector exit velocity, the overall fuel–air mixedness, and the fuel transport time
from the spokes to the flame.
A reduction in exit velocity was expected to have a stabilizing effect as long
as flashback into the premixing channel was avoided. Similarly, moving the fuel
spokes downstream would reduce premixedness, which would enhance flame sta-
bility but at the cost of increased NOx emissions. Also, changes in the transport
time might reduce any fuel system–liner coupling that was driving CPOs, as dis-
cussed subsequently.
Testing of modified injectors for both the Mars and Centaur engines demon-
strated that instabilities strongly depended on these design features. Small shifts
in either the fuel-spoke location or exit velocity were found to impact CPOs
significantly.
Injector screening for CPO characteristics was conducted in two ways. For rela-
tively minor changes in injector configuration, full sets of injectors were modified
and then tested in engines or in high-pressure rigs that attempted to simulate the
acoustics of the engine environment. In some cases, injector modifications were
substantial. In those cases, a single injector was fabricated for testing in a single-
injector test rig that contained a can combustor. The can combustors had the same
lengths as their annular counterparts and had diameters equal to the annulus height
of their corresponding annular liner. Testing duplicated the engine-operating con-
ditions, but flow rates were scaled for single-injector operation.
Special provision was made in an attempt to duplicate the instability frequency
of the full annular combustor in the single-injector rig. Following the approach of
Richards et al.,4 the combustor was fitted with a refractory plug similar to the one
shown in Fig. 2.7. The combustion zone and the plug form a Helmholtz resonator.
The plug was sized to match the resonance frequency experienced in the engine
environment.
Comb
Zone
Fig. 2.7 Can combustor with refractory plug to simulate a Helmholtz resonator.
COMBUSTION INSTABILITIES IN INDUSTRIAL GAS TURBINES 37
Clearly the single-injector rig was unable to duplicate the acoustic characteristics
of the annular combustor, in which circumferential and mixed-mode oscillations
could appear. The rig proved valuable as a screening tool, however. Injectors that
tended to drive oscillations in the rig environment almost always showed similar
tendencies in an engine. Injectors that operated without high CPOs in the single-
injector rig were frequently stable in the engine, although this was not universally
the case. Thus, the single-injector rig helped identify potentially better injector
configurations without fabricating complete sets of injectors and conducting costly
engine tests.
Subsequent testing of the modified injector indicated that the reduction in Uax
was an extremely robust solution to CPOs in the Centaur combustor. No significant
CPOs occurred at any point within the engine’s operating envelope. NOx levels
below 25 ppmv were readily achieved without high levels of pilot fueling. Because
of the minimal effort required to reduce CPOs, no further effort was made to
assess whether any of the other injector design parameters were actually of greater
importance than Uax in terms of CPO levels.
Fig. 2.8 Cross section of Mars injector, showing provision for air bleed.
COMBUSTION INSTABILITIES IN INDUSTRIAL GAS TURBINES 39
Although the modification to the Mars injector did reduce CPOs during part-
load operation, further reductions were sought. Work was continued to define the
role of fuel spoke location on CPOs.7 Development injectors were fabricated with
different axial distances from the fuel spokes to the exit plane of the injector.
Spokes were located at seven different distances from the exit, ranging from 1.38
inches (3.85 cm) to 3.0 inches (7.6 cm). The first observation was that small axial
changes in spoke position had a significant effect on the amplitude of any CPOs.
Certain spoke locations did not lead to coupling of the flame and the combustion
system. According to the simplistic model described, changing the axial location
of the spoke altered the relationship between τ and f C such that coupling was
avoided. Data from these Mars injector tests are presented in Table 2.2.
The data indicate that discrete regions of instability exist and that these regions
can be defined by the magnitude of the nondimensional number τ f C . As seen in
Fig. 2.9, for these Mars injectors, a region of instability was bounded by values
of τ f C of 0.30 and 0.48. These values suggest that significant coupling occurred
for approximately one-third of the duration of one pressure cycle. Whether this
finding is of general validity or specific to the Mars combustion system remains
to be determined.
On the basis of the preceding tests, a modified Mars injector was developed
that included both a change in flow cross-sectional area and a new spoke location.
Subsequent tests on production Mars engines confirmed that these changes, in
L = 1.38 in = 3.0
0.30 0.48
τ fc
Fig. 2.9 Mars CPO amplitudes as a function of fuel-spoke location and τ fC .
40 K. O. SMITH AND J. BLUST
conjunction with the pilot injector, were effective in reducing the severity of CPOs
without compromising the NOx emissions goal of 25 ppmv.
this need is probably an unrealistic hope in the near term, given the complexity
of CPOs. More realistic might be the development of models that can predict the
onset of CPOs (rather than limit cycle behavior). Similarly, a more rigorous means
of extrapolating CPO data from test rigs to the engine environment would help
reduce the time and costs associated with CPO mitigation.
References
1
Smith, K. O., Angello, L. C., and Kurzynske, F. R., “Design and Testing of an Ultra-
Low NOx Gas Turbine Combustor,” American Society of Mechanical Engineers, New York,
Paper 86-GT-263, 1986.
2
Etheridge, C. J., “Mars SoLoNOx: Lean Premix Combustion Technology in Production,”
American Society of Mechanical Engineers, New York, Paper 94-GT-255, 1994.
3
Rawlins, D. C., “Dry Low Emissions: Improvements to the SoLoNOx Combustion
System,” 11th Symposium on Industrial Applications of Gas Turbines, Canadian Gas As-
sociation, Banff, Alberta, Canada, Oct., 1995.
4
Richards, G. A., Gemmen, R. S., and Yip, M. J., “A Test Device for Premixed Gas
Turbine Combustion Oscillations,” Journal of Engineering for Gas Turbines and Power,
Vol. 119, 1997, pp. 776–782.
5
Putnam, A., Combustion Driven Oscillations in Industry, Elsevier, New York, 1971.
6
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premix
Gas Turbine Combustion,” American Society of Mechanical Engineers, New York, Paper
97-GT-244, 1997.
7
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., “Passive Control of
Combustion Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
Turbines and Power, Vol. 122, 2000, pp. 412–419.
Chapter 3
I. Introduction
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Section Manager, Advanced Combustion Engineering.
† Subsection Manager, Advanced Industrial Aeroderivative Combustor Design.
‡ Lead Engineer.
§ Staff Engineer.
43
44 H. C. MONGIA ET AL.
Various acoustic-control strategies, both passive and active, are then applied to
deal with the unacceptable levels of dynamics. However, the application of these
strategies is a largely empirical process with little assurance of success.
The conventional approach to this problem is mostly based on empirical cor-
relations and design experience. Attempts are made to predict frequencies of
combustion–acoustic waves, such that none of the subsystems (fuel nozzle, heat
shield, and combustor liners, etc.) of the total combustion system have natural fre-
quencies that can couple with the combustion–acoustic frequency. Although good
success has been obtained in predicted acoustic mode shapes and frequencies, the
amplitude of the oscillation(s) is not easy to predict.
To reduce the risk of uncontrollable combustion-acoustic behavior, detection
and abatement of combustion–acoustic susceptibility is required in the early
stages of a design. A comprehensive strategy to predict, avoid, and/or improve
the combustion-acoustic performance of a combustion system includes both ana-
lytical and experimental determination of system and component properties and
interactions. The basic framework of such a strategy is shown in Fig. 3.1. In brief,
a semianalytical model is used to link the acoustic characteristics of the sub-
component parts of the combustion system and boundaries. The characteristics of
these subcomponents can be derived either from analytical models or from well-
characterized empirical testing. A more detailed description will be given in a later
section.
This chapter is intended to provide an overview of the combustion-dynamics
problems observed in aeroderivative industrial gas-turbine engines and flight-
propulsion engines. The fundamental issues driving combustion dynamics in prac-
tical gas-turbine combustors are reviewed as a means to interpret the observed
combustion behavior. Several methods used in the laboratory and in production
engines for controlling combustion dynamics are described, along with two exam-
ples of combustion-dynamics control in production gas-turbine engines. Finally,
a framework for analysis and design of gas-turbine combustors to mitigate the
occurrence and impact of combustion dynamics is presented.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 45
system coupling as the most probable source. Altering the fueling strategy during
the engine start transient eliminated the objectionable tone.
Two distinct frequency peaks (signals) were also observed from pressure trans-
ducers mounted on the combustor near the flame zone for the higher-frequency
dynamics. The two primary frequencies observed were instability at 580 Hz and
a second tone at approximately 700 Hz. These acoustic modes were collectively
called organ tone because of their proximity in frequency to the quarter-wave axial
mode. This instability occurred in an operating mode in which only the outer (pi-
lot) dome of the DAC combustor was fired near its maximum fuel flow. Because
reliable analytical design tools were not available, an iterative empirical approach
was used to fix the instability problem. Several candidate approaches were identi-
fied and screened for ease of implementation, including alternate fueling modes,
Helmholtz resonators, fuel-nozzle cavity modifications, and spray-angle changes.
Variation in the fueling mode was selected as the prime approach. This approach
involved controlling certain main burner fuel nozzles in various circumferential
patterns. The strategy was to reduce the local pilot stoichiometry and to introduce
circumferential nonuniformity in the main stage to reduce dynamics.
Two separate engine tests were conducted to document the engines combustor-
dynamics characteristics. The first ground engine test was used to establish the
organ-noise threshold limits as a function of combustor-operating conditions and
for a range of core engine speeds. Analysis clearly showed the influence of pilot
fuel–air ratio on organ tone amplitude. Data from the ground engine tests repli-
cated earlier engine results, where two pure tones, one tone at 600 Hz and a second
tone at 680 Hz, were again observed when operating in the pilot-only fueling
mode at greater than 2000 rpm engine fan speed. A second engine test on a flying
test bed was subsequently conducted to get additional data from an altitude en-
gine environment. Data from the flying test bed engine were also used to map
the organ-noise envelope in the pilot-only fueling-mode. Extensive testing of the
fueling-mode changes identified a new fueling mode as the optimal configuration
48 H. C. MONGIA ET AL.
for minimizing the instability. This fueling mode incorporates two main-stage fuel
nozzles operating adjacent to each other at two circumferential locations in the
operating range where the instability was previously observed. Subsequent engine
tests validated the fueling-mode change as the most practical and effective way to
eliminate the dynamics problem in DAC combustors.
Fig. 3.4 Schematic illustration of the fuel-delivery system of the LM2500/6000 DLE
combustion system.
increase the air available for combustion, the combustor liners are convectively
cooled by using turbine-cooling air.
The 75 premixers are arranged on 15 two-cup and 15 three-cup assemblies. The
two-cup assemblies do not have the innermost premixer. Each “cup” consists of
a double annular counterrotating swirler (DACRS) premixer, in which two axial
counterrotating coaxial swirlers are mounted with a hub separating them, followed
by a mixing duct.5 The middle and inner dome premixers are identical whereas
the outer premixers are somewhat larger, such that the dome reference velocities
are all similar.
The fuel-delivery system consists of individual fuel controls for each dome of
the combustion system as described by Joshi et al.5 (Fig. 3.4). Independent control
of the fuel flow to each of the three domes is used to operate within the combined
constraints of emissions, combustion acoustics, and demanded total fuel flow rate
(shaft power). In addition, compressor air is bled from the engine at part-power
operation to extend the range over which the flame temperatures can be maintained.
As power is further reduced, radial and circumferential staging modes are also used
to bring flame temperatures within their operating limits (Fig. 3.5.)
Combustion dynamics can create substantial limitation to the operation of a lean
premixed combustor. Three design features of the system are used to expand the
range over which dynamics are within acceptable limits. First, a small fraction of
the fuel is injected into the combustor from holes in the walls of the mixing duct
(Fig. 3.6). This method of fuel injection has been shown in several instances to
provide improved performance of combustion dynamics in premixed combustion
devices.
Second, a set of quarter-wave tubes is provided on the premixers outside the en-
gine to absorb combustion-generated noise. These damper tubes, of three different
50 H. C. MONGIA ET AL.
Fig. 3.5 Schematic illustration of the fuel-staging strategy for the LM2500/6000 DLE
combustion system.
lengths, are installed on the fuel nozzles on the engine as shown in Fig. 3.3. Each
damper tube length is designed for critical operating band dynamic frequencies.
The damper tubes open into the diffuser cavity and communicate with the com-
bustor through the premixers.
Finally, the combustor-dome flame temperatures are empirically scheduled
to achieve low emissions and stable operation of the gas-turbine engine over its
entire operating range. The stable values of dome flame temperatures are affected
by variation in fuel property, ambient conditions, and load changes; thus, unstable
operation of the combustor can develop from changes in these parameters.
Fig. 3.6 The LM6000 DLE premixer showing ELBO fuel injection for controlling
dynamics.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 51
V. Combustion–Acoustic Modeling
A framework for eventually developing this predictive capability is shown in
Fig. 3.1. The basic philosophy behind this approach is the combination of analytical
modeling with experimentally determined boundary and submodel behavior. The
fundamentals of acoustic-wave propagation are well known; thus, determination of
available acoustic modes is straightforward once a temperature field is established
through the use of either empirical means or by computational fluid dynamics. The
remainder of the modeling problem consists of determining boundary conditions
and defining the submodel behavior. In particular, the accurate modeling of the
interaction of the combustion process with pressure waves is critical to the success
of the overall approach.
a) b) c)
Fig. 3.7 Contour plots: a) axial velocity, b) temperature, and c) reaction progress
variables from a steady CFD simulation of the LM6000 DLE combustor. The views
shown are on the plane that intersects the three-cup premixer.
flame speed determine the location of the flame sheet. Given the complexities of
the flowfield aft of the swirl-stabilized premixers, there can be little expectation
that convective times of individual fuel parcels originating within the same pre-
mixer will be confined to a narrow range. In addition, variation between premixers,
whether by design or chance, can result in further disparity in convective times.
A steady-state computational fluid dynamics (CFD) simulation17 of a 24-deg
sector of the LM6000 DLE combustor provides an illustrative example of the
variation in convective timescales. Plots of axial-flow velocity, temperature, and
reaction progress variables are shown in Fig. 3.7 for a typical high-power operating
point. In each annulus, corner recirculation zones anchor the flame front. The
extent of vortex breakdown and subsequent recirculation zone aft of the premixer
strongly depends on the premixer swirl number and expansion ratio. The lower
swirl number in the outer premixers results in a flame surface that is elongated
relative to those found in the middle and inner premixers. The swirl numbers of the
latter two premixers are equivalent; however, the inner premixer expansion ratio
is much larger because of increased premixer spacing.
To computationally determine the location of the flame front, a threshold value
of unity is chosen for the reaction progress variable G(x, t). This choice defines the
flame surface as the locus of points where the local mixture is in the burned state
50% of the time. The resulting distribution of calculated convective timescales
is provided in Fig. 3.8. Fuel–air mixing is not modeled; instead, experimentally
derived profiles of velocity and mixture fraction are used as inlet boundary condi-
tions. As such, the convective delay times represent only the time of flight from the
premixer exit planes to the flame sheet. The time of flight from the injection points
to the premixer exits can be determined by using through-the-vane CFD simula-
tions but, for the purposes of this example, convective times within the combustor
are sufficient to illustrate the complications of choosing a single timescale.
As seen in Fig. 3.8, convective times between the premixer exit and the flame
front range from 0.1 to more than 0.7 ms. The median, the mass-weighted mean, and
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 53
Fig. 3.8 Distribution of convective times from the premixer exit plane to the flame
front. Also shown is a log normal distribution.
the standard deviation are 0.40, 0.46, and 0.18 ms, respectively. For comparison
proposes, the average residence time of the fuel–air mixture in the premixer is
approximately 0.45 ms. Statistical differences between individual premixers are
significant because median times for individual premixers range from 0.25 to
0.47 ms. The overall distribution is not well fit by using either normal or log normal
distributions (for the log normal distribution, see Fig. 3.14). These time lags can
also be expressed as phase lags in the frequency domain. At a frequency of 500 Hz,
the convective time-delay range of 0.1–0.7 ms is equivalent to a 18- to 126-deg
phase lag, which covers both positive and negative Rayleigh index values. To
account for the multiple convective timescales associated with the mixer, the single-
time-lag-modeling approach is oversimplistic and may need to be expanded to
reproduce observed behavior. An approach for deriving complex premixer transfer
functions from steady-state CFD simulations has recently been proposed.18
“mode-locking” is evident. However, these levels are beyond the levels sustainable
in an engine without damage and are thus of little practical interest.
Although simple in concept, several issues make the interpretation of test results
less simple than one might wish. Because of overall pressure-drop limitations and
mechanical durability issues, the flow through the pistons is not choked. How-
ever, by measuring dynamic pressures on both sides of the pistons, the complex
impedance of the pistons can be determined analytically. The geometry of the
combustion chamber is quite simple, only supporting axial acoustic modes of
well-defined frequencies. Although the axial mode shape of the TCA is desirable
from an analytical perspective, the measured and calculated mode shapes in an an-
nular combustor often contain a significant circumferential component. Provided
that the primary acoustic coupling mechanism is through pressure-wave-induced
fluctuations in equivalence ratio, lack of circumferential modes developed in the
test rig will be unimportant. The typical length scale of a pressure wave is on the
order of 1 m, whereas the length scale of the swirler is typically an order of mag-
nitude less. From the perspective of the pressure fluctuation, the mode shape will
not significantly alter the response of the fuel nozzle–swirler. However, the flame
dynamic response will also depend on the velocity fluctuation, which is sensitive
to orientation and thus mode shape. This limitation cannot be easily overcome
through this type of test and would require a very different approach.
A typical result from the TCA test rig is shown in Fig. 3.10. The vertical axis is
clock time, during which the downstream piston is moved from its maximum to
minimum length position, and then back to its original position. The piston velocity
is constant, and thus the axis can also be interpreted as effective combustor length.
The horizontal axis is acoustic frequency, and the color scale indicates amplitude
at the corresponding frequency. An envelope of the maximum response is shown
at the bottom of the figure. As is evident, the system response strongly depends
on the combined natural resonant frequency of the combustion chamber and the
response of the fuel nozzle–swirler. Because of the relative simplicity of the phys-
ical experiment, computational modeling of the test vehicle can be accomplished
with confidence. The combustor–acoustic interaction is still treated as a transfer
function; in this case, the experimental data are used to determine the parame-
ters of the transfer function. Under the assumption of the mode shape–combustor
response independence described earlier, this transfer function is expected to be
applicable to the full annular model of the combustor as well.
A significant advantage of this approach is that it permits evaluation of the impact
of swirler and fuel nozzle design variables on combustion acoustics, without the
expense and delay associated with construction of a full-engine set of hardware and
test. It also avoids the circumstance under which a design change only moves the
maximum response frequency of the fuel nozzle to a nonresonant mode of a fixed-
geometry single-cup test rig. As stated earlier, a large-scale annular combustor
has numerous mode shapes covering a wide range of frequencies available to it,
whereas a fixed-geometry single-cup rig generally has only one or two natural
acoustic frequencies. The design strategy described is still being validated and has
not yet been demonstrated to completion. Given the risk to combustor and engine
development schedules represented by combustion acoustics, the value of such a
strategy is clear.
Fig. 3.11 Comparison of measured and predicted combustion acoustic response from
LM6000 DLE.
58 H. C. MONGIA ET AL.
a)
b)
c)
d)
of the flame front (which is an input critical to estimating the characteristic convec-
tion time distribution) by examining the steady-state heat-release contours could
be a factor contributing to the discrepancy in frequency calculation. The uncer-
tainty in calculating transport-time distribution associated with the unsteady heat
release will induce a different phase lag into the model and thus predicts a dif-
ferent resonant frequency. Based on the current analysis, the dynamic pressures
response in the vicinity of 500 Hz is mainly contributed by the plane waves from
the middle and inner domes. On the other hand, the source of the 600-Hz acoustic
mode appears to be the first-order circumferential mode arising in the outer ring.
The acoustic analyses of the baseline and uniform-temperature-distribution
case were conducted to investigate the effect of radial temperature nonunifor-
mity on combustion instability. The measured and predicted responses are shown
in Figs. 3.12a and 3.12b, respectively, for baseline operation, and in Figs. 3.12c and
3.12d, respectively, for an operating condition with uniform temperature in all three
domes. The measured spectra shown in these figures were based on dynamic pres-
sure measurements from pressure sensors installed just downstream of the flame
front in the DLE combustor. The predicted response is obtained by taking the max-
imum values of the individual responses from each of the three rings. Two distinct
frequency peaks near 400 and 600 Hz were found from the analysis, similar to those
observed from engine tests. These figures also show that the 400-Hz mode is more
active than the 600-Hz dynamics for both operating conditions. In addition, the
measured amplitudes of the two acoustic modes at the baseline condition showed
relatively low levels. The predicted stability indices for these two peaks were less
than 1 and are both categorized as acoustically inactive. In contrast, an acousti-
cally active mode near 400 Hz with a stability index of 3 is found for the operating
condition with uniform flame temperatures in all three rings. This predicted trend
agrees well with test data and further demonstrates the capability of the current
model to distinguish between acoustically active and inactive regimes of operation.
The methodology required to incorporate passive damping devices was a prime
requirement in this modeling activity.22,23 As mentioned in Sec. III, a set of
22 damper tubes is provided upstream of the premixers, outside the engine to
absorb combustion-generated noise. Because the mean flow Mach number in the
air column of these devices is very small (typically less than 0.05) and the diameter
of these quarter-wave tubes is much smaller than the wavelength of the resonant
frequency of oscillation, the dynamic pressure and mass flow within the damper
tubes can be represented by a one-dimensional analysis.
These devices are quarter-wave resonators installed in the cold section of the
combustor just upstream of the premixers. The installation schematic of the damper
tubes used in the formulation of this analysis is shown in Fig. 3.13. These devices
act to detune the combustor by providing a finite number of discontinuities at the
locations where they are installed. The incident and reflected acoustic waves in the
diffuser cavity are significantly altered to the extent that certain discrete oscillations
are attenuated and therefore become less destructive to the combustor. The acoustic
damping devices in the current analysis are modeled by treating each quarter-wave
tube as a monopole acoustic source characterized by its acoustic impedance. The
effect of damper tubes (tuned for 510 Hz) on combustion dynamics in the pilot
(middle) ring is shown qualitatively in Fig. 3.14. A 7-fold decrease in stability
index at the tuning frequency of the damper tubes is predicted based on this analysis.
60 H. C. MONGIA ET AL.
Although analytical techniques have improved markedly during the past few
years, the difficulty in representing the combustion–pressure wave interaction will
likely remain a barrier to full predictive capability. Thus, it is likely that a combined
approach, including both experimental and high-fidelity unsteady CFD modeling,
will be required to develop models of the full annular combustor response that will
have utility in defining design direction.
VII. Conclusion
Combustion instability in conventional diffusion-flame gas-turbine combustors,
if encountered, can be eliminated, in general, with simple modifications to the
design or operating parameters. The design guidelines based on empirical know-
how are generally good enough to avoid the unforeseen occurrence of dynamics
during the engine-certification phase. If encountered, simple analysis coupled with
empirical guidelines and a systematic testing and development process are used to
find engineering solutions without adversely affecting the key combustion-system
design requirements.
On the other hand, controlling combustion dynamics in industrial DLE combus-
tion systems remains a substantial challenge for designers. The combustion process
is pushed close to the limits of lean flame stability and/or heat-release rates that
lead to strongly coupled nonlinear interaction between the flame exothermicity,
acoustic behavior of the system, and components. The approach for controlling
dynamics in the LM engines (bleed air, fuel staging, dome flame temperatures
differences, ELBO, acoustic damper tubes, and ABAL control logic) has worked
very well. However, to further improve on the design of lean premixed combustors,
we need to make significant advances in this area.
More recent directions in propulsion engine technology are driving designs that
have more in common with the lean premixed DLE combustors than with the
rich-dome combustors of the past. As these designs progress toward maturity, it
is likely that combustion dynamics will become a limiting factor in the attainable
performance of these combustors. Thus, improved predictive capability, funda-
mental understanding, and control technologies for combustion dynamics need to
be developed. The current state of the art is represented by a combined-systems
approach in which empirically derived submodels are linked with a physics-based
system model. In the future, we anticipate an increased role in developing LES
capabilities as a tool to develop these submodels, rather than relying on component
test or engine data. In the longer term, more direct linkage of the physical coupling
mechanisms between the components of the combustion system (fuel system, fuel
injector, combustor cavity, and combustion–pressure wave interactions) will make
it possible to accurately describe, predict, and control the physical phenomena
associated with combustion dynamics.
References
1
Lieuwen, T. “Analytical Modeling of Combustion-Acoustic Wave Interactions: A Re-
view,” Journal of Propulsion and Power, Vol. 11, No. 22, 2003, pp. 222–444.
2
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., “Passive Control of
Combustion Instability in Lean Premixed Combustors,” American Society of Mechanical
Engineers, Paper 99-GT-52, 1999.
62 H. C. MONGIA ET AL.
3
Straub, D. L., and Richards, G. A., “Effect of Axial Swirl Vane Location on Combustion
Dynamics,” American Society of Mechanical Engineers, Paper 99-GT-109, 1999.
4
Paschereit, C. O., and Gutmark, E., “Passive Combustion Control for Enhanced Stability
and Reduced Emissions in a Swirl-Stabilized Burner,” AIAA Paper 2003-1011, 2003.
5
Joshi, N., Epstein, M., Durlak, S., Marakovits, S., and Sabla, P., “Development of a Fuel
Air Premixer for Aero-Derivative Dry Low Emissions Combustors,” American Society of
Mechanical Engineers, Paper 94-GT-253, 1994.
6
Scarinci, T., and Halpin, J. L., “Industrial Trent Combustor – Combustion Noise Char-
acteristics,” American Society of Mechanical Engineers, Paper 99-GT-9, 1999.
7
James, D., “A Solution for Noise Associated with a Series Staged DLE Combustion
System,” Proceedings of the 4th International Pipeline Conference, 29 Sept. to 3 Oct. 2002.
8
Schlein, B. C., Anderson, D. A, Beukenberg, M., Mohr, K. D., Leiner, H. L., and
Träptau, W., “Development History and Field Experiences of the First F T8 Gas Turbine
with Dry Low NOX Combustion System,” American Society of Mechanical Engineers,
Paper 99-GT-241, 1999.
9
Johnson, C. E., Neumeier, Y., Lubarsky, E., Lee, J. Y., Neumaier, M., and Zinn, B.
T., “Suppression of Combustion Instabilities Using a Fast Adaptive Control Algorithm,”
AIAA Paper 2000-16365, 2000.
10
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Structure and Control of
Thermoacoustic Instabilities in a Gas-Turbine Combustor,” Combustion Science and
Technology, Vol. 138, 1998, pp. 213–232.
11
Magill, J., Bachmann, M., and McManus, K., “Combustion Dynamics and Control in
Liquid-Fueled Direct Injection Systems,” AIAA Paper 2000-1022, 2000.
12
Jones, C. M., Lee, J. G., and Santavicca, D. A., “Closed-Loop Active Control of
Combustion Instabilities Using Subharmonic Secondary Fuel Injection,” Journal of
Propulsion and Power, Vol. 15, No. 2, 1999, pp. 1–7.
13
Leonard, G., and Stegmaier, J., “Development of An Aero-Derivative Gas Turbine Dry
Low Emissions Combustion System,” American Society of Mechanical Engineers, Paper
93-GT-288, 1993.
14
Joshi, N. D., Mongia, H. C., Leonard, G., Stegmaier, J. W., and Vickers, E. C., “Dry
Low Emissions Combustor Development,” American Society of Mechanical Engineers,
Paper 98-GT-310, 1998.
15
Patt, R., “Development and Operating Experience of DLE Combustion Systems,” 12th
Symposium on Industrial Applications of Gas Turbines, 15–17 Oct. 1997.
16
Mongia, H. C., “TAPS: A 4th Generation Low Emissions Propulsion Engine Com-
bustion System,” The AIAA/ICAS International Air and Space Symposium, 2003, AIAA,
Reston, VA, 2003.
17
Held, T. J., Mueller, M. A., Li, S.-C., and Mongia, H. C., “A Data-Driven Model for
NOX , CO and UHC Emissions for a Dry Low Emissions Gas Turbine Combustor,” AIAA
Paper 2001-3425, 2001.
18
Flohr, P., Paschereit, C. O., and Belluci, V., “Steady CFD Analysis for Gas Turbine
Burner Transfer Functions,” AIAA Paper 2003-1346, 2003.
19
Huang, Y., Sung, H. G., Hsieh, S. Y., and Yang, V., “An LES Study of Combustion
Instabilities in Lean-Premixed Gas Turbine Combustors,” Journal of Propulsion and
Power, Vol. 19, 2003, pp. 782–794.
20
Stone, C., and Menon, S., “Open-Loop Control of Combustion Instabilities in a Model
Combustor,” Journal of Turbulence, Vol. 4, 2003, p. 20.
COMBUSTION INSTABILITY ISSUES INTO DESIGN PROCESS 63
21
Schlüter , J. U., “Static Control of Combustion Oscillations by Coaxial Flows: An
LES Investigation,” Journal of Propulsion and Power, Vol. 20, No. 3, 2004, pp. 460–467.
22
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Investigation of Combus-
tion Dynamics in Dry Low Emission Gas Turbine Engines,” AIAA Paper 98-3381, 1998.
23
Pandalai, R. P., and Mongia, H. C., “Combustion Instability Characteristics of
Industrial Engine Dry Low Emission Combustion System,” AIAA Paper 98-3379, 1998.
Chapter 4
Tomas Scarinci∗
Rolls-Royce Canada, Quebec, Canada
Nomenclature
c = speed of sound
d = location of a source of fuel injection
dk = location of the kth fuel-injection source
p = limit-cycle pressure oscillation amplitude
f = frequency
H ( f ) = transfer function
K = regression constant
ki = regression constant
L c = length of combustion chamber
L m = length of fuel–air-mixing region
M = Mach number
p = instantaneous static pressure
P = average static pressure
p = pressure fluctuation
P3 = compressor discharge pressure
Re = real part of a complex number
s = spacing between two points of fuel injection
sk = strength of the kth fuel-injection source
T 3 = compressor discharge temperature
Ti = flame temperature of the ith combustion zone
u = instantaneous velocity
U = average velocity
u = velocity fluctuation
Copyright c 2005 by Rolls Royce Plc. Published by the American Institute of Aeronautics and
Astronautics, Inc., with permission.
∗ Director of Gas Turbine Engineering, Rolls-Royce Energy Business.
65
66 T. SCARINCI
Fig. 4.1 Cross-section of Trent 60 DLE aeroderivative with details of the original
three-stage dry, low-NOx combustion system.
terms of premixer (or zone) temperatures. For instance, once a primary temperature
is chosen (say, 1750 K), a secondary temperature (say, from 1200 K to 2000 K) can
be chosen. The amount of fuel to be allocated to the tertiary stage falls out from
the total amount of fuel required by the engine. Note that allocating no fuel at all
to the tertiary stage is also (sometimes) an option. Fig. 4.2 shows typical possible
operating envelopes for the Trent combustor at different power levels. The temper-
atures of the y axis and x axis are the primary- and secondary-flame temperatures,
respectively, from which a reference temperature was subtracted. Any point inside
68 T. SCARINCI
250
primary premixer temperature, K
200
80% power
150
100% power
100
50
0
-50
-100
-150
-200
-300 -200 -100 0 100 200 300 400
secondary premixer temperature, K
Fig. 4.2 Typical operating envelopes for the Trent 60 combustor (ISO Day).
the operating envelope defines a unique fuel split between primary, secondary,
and tertiary stages. One can easily imagine that the NOx level at the combustor
exit will depend on the chosen fuel split. By appropriately staging the fuel, NOx
emissions in the 10- to 20-ppm range can be achieved from 50–100% power.
The amplitude of these limit cycles is typically on the order of 1–2% of the
static pressure inside the combustion chamber, and typical frequencies seem to
range from approximately 100 Hz to roughly 1–2 kHz, although these values
depend on the engine geometry and size. These amplitudes and frequencies can be
highly problematic for the mechanical components of most engines because they
are a source of high-alternating mechanical stresses. In the absence of control or
avoidance procedures (or designs), mechanical components can abruptly fail by
high-cycle fatigue under these conditions.
1.4
1.2
1.0
P/ Pref (%)
0.8
0.6
0.4
0.2
0.0
0.00 0.25 0.50 0.75 1.00
L/λ
Fig. 4.4 Measured combustor pressure-oscillation amplitudes and frequencies for
the longitudinal acoustic modes of the combustion chamber.
This contour map, and others at 50% and 80% load,1 shows that it is always
possible to find regions of high- and low-pressure-oscillation amplitude at any
power condition. In other words, the rms level of pressure fluctuations is clearly
affected by the distribution of heat release inside the combustor. In general, the
pressure-oscillation amplitude scales linearly with the combustor pressure. Thus,
the potential for structural damage is much higher at high engine pressure ratios.
The potential for structural damage is a difficult challenge for the Trent, which
operates at pressures up to 40 MPa.
Depending on the power level and the ambient conditions, the regions of large-
pressure-oscillation amplitude will be located in a different region of the operating
envelope. In Fig. 4.3, there are two regions of large-pressure-oscillation ampli-
tude; both are located in the region of low-primary temperatures. However, as the
secondary temperature is varied, it becomes possible to find an optimum condition
that will minimize the pressure-oscillation amplitude. The two distinct regions of
large-pressure-oscillation amplitude in Fig. 4.3 correspond to two different acous-
tic modes of the combustor. In all, a total of three different longitudinal acoustic
modes were seen under different operating conditions. These three modes are il-
lustrated in Fig. 4.4, which condenses the results from more than 300 tests by
plotting the pressure amplitude against a nondimensional frequency. Note that
all results are clearly grouped into three well-defined acoustic modes. At a given
condition, there may no large-amplitude oscillations, that is, there is no clearly ob-
servable limit-cycle oscillation. Nonetheless, one can always identify a dominant
frequency in the pressure spectrum, and this dominant frequency always seems to
correspond to one of the normal acoustic modes of the combustion chamber. In
this case, that is, when there is no clear limit-cycle oscillation, the rms level of the
pressure fluctuations is on the order of 0.1–0.2% of the reference pressure. When
oscillatory combustion sets in, it always appears at the frequency of one of the
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 71
100%
primary bias = 0 K
3rd m ode
(% of maximum)
noise amplitude
60%
40%
2nd m ode
20%
0%
-50 0 50 100 150 200 250
secondary tem perature bias, K
Fig. 4.5 Change in acoustic mode and pressure oscillation amplitude as the
combustor fuel split is varied at a steady-engine-operating condition near base load
(100% power).
first three natural acoustic modes. Depending on the operating conditions, the fuel
split at which the limit cycle appears and its frequency and amplitude will vary.
We discuss this relationship in more detail next.
Figure 4.5 is effectively a slice through the pressure-oscillation amplitude map
of Fig. 4.3. As the secondary bias is increased, the pressure-oscillation amplitude
decreases until it reaches a minimum. At the amplitude minimum, however, a
mode-switching phenomenon occurs and the frequency of the limit-cycle oscilla-
tion switches from the second to the third acoustic mode. It can also be seen that the
overall pressure oscillation levels can be reduced if the primary bias is increased.
This figure provides a good illustration of the flexibility offered by the three-stage
design. At a fixed-engine-running condition, a small change in the primary and/or
secondary temperature can reduce the combustion pressure-oscillation amplitude
by 50% and/or select the frequency of the limit cycle.
Figure 4.6 plots the dependence of the amplitude of the second longitudinal
mode (i.e., L c /λ = 0.5) on combustor-fueling conditions and engine power. The
amplitude of this mode always decreases as the primary bias is increased. What
appears to be large scatter at a given primary bias is actually the effect of the
secondary bias being changed. Even though the pressure-oscillation amplitude is
nondimensionalized by the combustor reference pressure, all the curves do not
collapse into one. Thus, the second-mode amplitude depends to a certain degree
on the total energy being released inside the combustor.
The third unstable mode (L c /λ = 0.75) has amplitude characteristics that are
completely different from those observed with the second mode. As seen from
Fig. 4.7, it is possible to collapse all the data onto a single curve. For the third
mode to appear, the secondary-zone temperature must exceed the primary-zone
temperature.
72 T. SCARINCI
140%
80%
60%
40%
20%
0%
-100 -50 0 50 100 150 200 250
primary temperature bias, K
120%
100%
80%
(% of maximum)
noise amplitude
60%
40%
20%
0%
-100 -80 -60 -40 -20 0 20 40 60 80 100
Secondary minus Primary Zone Temperature (K)
Fig. 4.7 Amplitude characteristics of the third unstable mode (Lc /λ = 0.75). The
data in this figure are obtained at engine power levels ranging from 30% to 100%
power.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 73
F A R u
=− (4.1)
F AR ū
Lieuwen and Zinn3 further noted that, because of the low Mach number of the
flow in the passage (typically, M ∼ 0.05), small pressure fluctuations can result
74 T. SCARINCI
u 1 p
= (4.2)
U γM P
Thus, a 1% fluctuation in static pressure inside the combustor can typically result
in a 15% FAR fluctuation from the premixers. In other words, a pressure oscillation
inside the combustor would normally lead to a “FAR-wave” emanating from the
premixer exit and this can in turn modulate the heat release inside the combustion
zone. If the heat release associated with the FAR oscillations is in phase with the
original pressure waves, the oscillating heat release will augment the original pres-
sure waves in the combustion chamber. We will refer to this particular feedback
mechanism of combustion instability in DLE systems as the “Richards–Lieuwen
mechanism.” This expression seems to be useful shorthand and it recognizes the
empirical explanation first proposed by Richards in 1997 and the theoretical foun-
dations subsequently proposed by Lieuwen in 1998.
Although, as indicated in Chapter 9, many other potential mechanisms can lead
to combustion instability in a premixed system, even in the absence of fuel–air ratio
oscillations,5,6 none seem as important, or at least as powerful, as the Richards–
Lieuwen mechanism. At least, the Richards–Lieuwen mechanism seems to be the
most powerful in DLE combustion systems as they apply to most industrial gas
turbines.
In Fig. 4.8, a simplified account (compared with the more fundamental discus-
sion found in Chapters 9 to 12 (i.e., Sec. III of the book) is presented of the key
instability mechanisms believed to be present in most practical lean premixed sys-
tems. In simple terms, large-amplitude pressure waves can be generated either from
heat-release oscillations or the interaction of density oscillations with the pressure
gradient associated with the choked turbine (entropy noise). Note, in Fig. 4.8, that
the presence of fuel–air ratio oscillations emanating from the premixers can lead
to the generation of pressure waves in several ways. First, the oscillating FAR is a
direct source of heat-release oscillations (the Richards–Lieuwen mechanism). The
time-varying FAR from the premixers will also lead to a time-varying density field
inside the combustor, a time-varying flame-surface area, and a time-varying flame-
burning speed. All of these can lead to the generation of pressure waves either by
way of entropy noise or heat-release oscillations. A significant effort has been made
during the past few years, in particular, using active-control methodologies, to find
ways of eliminating FAR oscillations in DLE systems. The approach taken at Rolls-
Royce has been, somewhat analogous to acoustic-damping devices, to damp the
presence of FAR oscillations from the premixers. This is referred to as “FAR-Wave
damping” in Fig. 4.8, where it is quite clear that this sort of damping has the po-
tential to weaken numerous feedback loops leading to oscillatory combustion. The
efforts to develop FAR-wave damping technologies are described in Sec. II.B.2.
Note, in Fig. 4.8, that most instability mechanisms involve a coupling by means
of pressure waves. The aim of any passive-design solution is to reduce the strength
of any of the feedback loops. Irrespective of the mechanism, acoustic-attenuation
devices, such as resonators, always achieve this goal because the strength of pres-
sure waves is reduced. A brief description of our efforts to implement such acoustic
resonators is described in Sec. II.B.3.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 75
Pressure Waves
in Premixer Aerodynamic
Aerodynamic
Damping
FAR-Wave
Damping
Damping
FAR Oscillations
FAR (t)
Propagation of Pressure
Pressure Waves Wave
in Combustor Damping
Damping
Fig. 4.8 Driving mechanisms for combustion instability and means for suppression
of instabilities. The thicker line refers to mixing-induced oscillation (R-L mechanism).
Also noted in Fig. 4.8 are aerodynamic fluctuations (i.e., broadband turbulent
fluctuations) which disturb the flame and cause a general broadband level of in-
creased combustor pressure fluctuations. Efforts to minimize these aerodynamic
sources are described briefly in Sec. II.B.4.
1. Fuel Staging
The variable distribution of heat release along the length of the combustor
provides an effective mechanism to control the amplitude of longitudinal resonance
modes of the combustor. Having obtained some indications of the limit-cycle
behavior, an attempt was made to develop empirical correlations that could be
used by a controller to predict the amplitude of each of the unstable modes under
a particular condition.
In developing this correlation, it was assumed that the nondimensional amplitude
of the pressure oscillation is predominantly a function of the temperature ratio
76 T. SCARINCI
across the region of heat release. In a three-stage (axially staged) combustor there
are three regions of heat release, and hence there are at least three important
temperature ratios that will affect pressure-oscillation amplitude. It might be argued
that the overall heat release in the combustor (and the corresponding temperature
ratio from combustor inlet to combustor exit) might also be a controlling parameter.
For these reasons, the functional form of the empirical correlations that were used
for each of the unstable modes was:
N
p (Ti − T3 ) αi
= K 1 + K 2 P3A
(4.3)
P i=1
ki
where K 1 , K 2 , A, ki , and αi are all constants to be fitted to the data. The summation
index i, which runs from 1 to N , refers to each of the combustion zones of the
combustor, including the overall heat release inside the combustor. The arbitrary
constants were obtained from linear regression of engine data for each of the
unstable modes.
Three independent empirical correlations were obtained for each of the longi-
tudinal resonant modes. The assumption that was then made was that the mode
observed would be the one having the highest amplitude of the three, given the
combustor-operating conditions. Note that this approach also permits prediction of
not only the amplitude, but also the frequency of the combustion resonance. These
correlations are able to capture relatively well the effects of engine cycle, ambient
conditions, and changes in fuel schedule (or equivalently, variations in combustor
fuel splits). The correlations were able to reproduce the pressure-oscillation ampli-
tude mapping results within approximately 10% accuracy. These developed corre-
lations were used to design fuel schedules to minimize high-combustion pressure-
oscillation amplitude across the operating range of the engine. Of course, having a
three-stage premix combustor was almost a prerequisite for the implementation of
this type of approach. A three-stage premix combustor offered sufficient fuel-split
options to control pressure oscillations whilst meeting emissions requirements.
Air Air
A B
s
x=-d x=0 x=- d x=0
fuel-1 fuel-N
fuel inlet
air jets
Air
C D
Lm Lm
x=-d x=0 x=-d x=0
in the system. Longer time delays provide more multiples of each of the fundamen-
tal unstable modes of the system. However, and in a DLE combustor, if there was a
sufficiently long mixing time (ignoring autoignition concerns for a moment) then
no FAR fluctuations would be coming out of the exit of the premixer, because they
would be eliminated by the mixing process. Hence, there are competing issues: as
the residence time in the premixer increases, there is a large number of possible
unstable modes, but there is no longer any fuel–air ratio (temporal) fluctuations
coming out of the premixer to drive the instability.
Although we refer the reader to our original publications for the details, con-
siderable effort was spent to develop and experimentally calibrate models that
captured the effects of convection and turbulent dispersion upon FAR perturba-
tions. A variety of candidate-injector configurations were examined experimen-
tally, analytically, and computationally. These configurations are shown in Fig. 4.9.
The laboratory tests made use of simple small-scale atmospheric rigs. Hot wires
and pressure transducers were used to monitor unsteady-flow conditions (u and
p ) and a proprietary fast flame-ionization detector (FID)8 was used to measure
instantaneous changes in fuel concentration (FAR ). Bottled ethylene was used
as fuel. The air supply was fan driven, and a downstream variable speed rotary
valve produced premixer airflow oscillations with peak amplitudes (u /U ) of 30%.
The mean airflow rate and the total fuel flow rate were held constant for all test
configurations.
Some of the key considerations associated with the effect of premixer design
variables on combustor stability characteristics can be understood by applying the
characteristic time analysis, as described in Chapter 17. As shown, instabilities
from the Richards–Lieuwen mechanism are excited when the time for the mixture
to travel from the injection point(s) to the flame lies within some range of integer
multiple of the acoustic period. Consider N sources of fuel-injection locations,
78 T. SCARINCI
such as shown in configuration C in Fig. 4.9. Each location has a strength sk and is
located at a position dk . Neglecting turbulent-dispersion effects for the time being,
it can be shown9 that the Lieuwen–Richards mechanism causes amplification of
acoustic waves when the following inequality is satisfied:
N
−i (ωdk ( U1 + 1c )− π2 )
Re sk e ≥0 (4.4)
k=1
This equation assumes for simplicity that the reactive mixture is immediately
consumed at x = 0. Referring back to Eq. (4.1), we also define a parameter α as
the proportional change of FAR oscillations divided by the proportional change in
velocity fluctuation, that is:
F A Rr ms u r ms
α= (4.5)
F AR ū
Fig. 4.10 Regions of positive amplification of pressure waves as the parameter d/U
(premixer residence time) is varied.
Potentially in a combustor with only a single unstable mode could such a location be
identified (e.g., see Chap. 2); however, it is much more difficult in the Trent 60 DLE
in which one is dealing with three distinct modes (and frequencies). The only way
that such a configuration could be made sufficiently robust then, would be to posi-
tion the fuel-injection location far enough upstream that the FAR perturbation was
washed out by turbulence. To determine the level of turbulent dispersion of a FAR
disturbance with a delay-time representative of that of the Trent 60 DLE, measure-
ments were made in a representative configuration (configuration A in Fig. 4.9).
Typical raw data obtained for u (t) and F A R (t) are shown in Fig. 4.11. The dashed
horizontal lines in Fig. 4.11 correspond to the peak-to-peak amplitude of u which
would give α = 1.0, that is, the level of FAR fluctuation expected to occur from
the airflow fluctuation u , according to Eq. (4.1). In this case, α was approximately
0.95. As such, very little attenuation of FAR fluctuations occus in the premixer.
Configuration B can be seen as a logical attempt to deal with both problems
described here: time delay and turbulent dispersion. In principle, this approach
could eliminate a FAR wave of a given frequency by using a “wave-cancellation”
approach. The concept consists of adding a second fuel-injection point and dividing
the fuel flow equally between the two points. Such a system can be designed with
the aid of Eq. (4.4) to determine an appropriate d1 and d2 with a wider stable range.
A typical result that includes turbulent-dispersion effects, reproduced from Ref. 9
is shown in Fig. 4.12. The injector spacing is given by s = d2 − d1 . Although this
solution is more robust than the single-injector configuration, it still has problems.
First, it is still difficult to determine optimum injector location that stabilizes
all relevant combustor modes over all operating conditions. Second, the effect
of turbulent dispersion, which does not substantially alter the result shown in
Fig. 4.10 for N = 1, actually broadens amplification regions. Thus, the criterion
summarized in Eq. (4.4) is overly optimistic.
If such a configuration cannot be used for a suitably robust determination of
convective time delays to ensure stable combustion, what impact would it have on
80 T. SCARINCI
12
10
u (m/s), FAR (%)
alpha = 1
4
2
FID signal (% fuel)
0
0 10 20 30 40 50
time (ms)
Fig. 4.11 Results obtained for configuration A showing time dependence of air-
velocity oscillations in premixer and FAR oscillation at premixer exit.
Fig. 4.12 The regions of positive-pressure-wave amplification with two fuel sources
in the premixer in the presence of turbulent mixing (d/U = 4 ms).
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 81
1
Configuration A
0.8
0.6
alpha
0.4
Configuration B
Configuration D
0.2
0
0 1 2 3 4 5
fL/U
12
10
u (m/s), FAR (%)
6
hot wire signal (m/s)
alpha = 1
4
2
FID signal (% fuel)
0
0 10 20 30 40 50
time (ms)
Fig. 4.14 Results obtained for configuration D showing time dependence of air-
velocity oscillations in premixer and FAR oscillation at premixer exit.
confirmed in the lab, see Fig. 4.14. These data show that configuration D appeared
nearly free of any FAR oscillations at the exit of the premixer, that is, α < 0.1.
This sort of behavior, in which airflow oscillations become decoupled from the
FAR oscillations emanating from the mixing duct, is exactly what was desired
and this appeared (from Fig. 4.13) to have been achieved over a wide frequency
range.
The process by which the FAR and airflow oscillations are decoupled is not quite
as simple as what was portrayed in the case of progressively adding more points of
fuel injection. As will be seen, the fuel–air mixing process and its dynamics were
fundamentally changed with the new design.
First, consider the case in which the airflow in the premixer is oscillating, for
example, in response to a pressure oscillation in the combustion chamber. The
mixing arrangement may be thought of as a “phased array of air jets,” that is, there
are N rows of air-injection jets, not necessarily equally spaced, over a distance L m .
The upstream end of the mixing duct contains only pure fuel. At the location of the
first row of air jets, because the air jets are modulated, a local FAR oscillation is
generated. This FAR oscillation is convected to the next row of air jets. Depending
on the frequency of the oscillation, the air from the second row of air jets will
either weaken or sustain the FAR oscillation received from the first row of air
jets. This process is repeated again for each of the remaining N − 2 rows of air
jets. With numerous rows, there are numerous convective delays between any
two given rows, and hence a multiplicity of phasing relationships is generated as
the fuel makes its way toward the exit of the mixing duct. Thus, a progressive
smoothing of FAR oscillations occurs along the length of the premixer. As an
order-of-magnitude estimate one would expect that the FAR oscillations will be
successfully “integrated out” if a sufficient number of cycles are experienced in
the mixing process; that is, if the wavelength of the velocity fluctuation is smaller
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 83
than the distance L m over which the air jets are distributed:
f Lm
≥1 (4.6)
U
1
|H ( f )| =
2 (4.7)
1+ 4π 2 f Lm
U
Substituting values into this equation shows that when f L m /D = 1, the FAR
oscillations are 85% attenuated. When f L m /D = 2 the FAR oscillations are 92%
attenuated. These levels of FAR attenuation are close to an order of magnitude
more than anything that could be achieved by using passive turbulent diffusion
alone.
This mixing-duct technology was implemented and tested on the Trent 60 DLE
aeroderivative. For the new mixing ducts to work effectively, it is necessary to
ensure that the FAR-damping process becomes effective at frequencies below the
frequencies of the lowest acoustic mode of the combustor. For effective damping,
the FAR-wave cutoff frequency of the premixer must be lower than the natural
frequency of the lowest acoustic mode of the combustor, that is,
U c
< (4.9)
Lm 4L c
Lm
c
>4 (4.10)
Lc U
Equation (4.10) relates the design parameters of the new mixing ducts to the overall
geometry of the combustion system.
Figure 4.15 shows a comparison between the pressure-oscillation levels with
the original Trent premixer (i.e., a combustor cross section as in Fig. 4.1 and
using a fuel–air concept analogous to configuration A of Fig. 4.9) and a mod-
ified premixer (i.e., one with the design implementation of configuration D of
Fig. 4.9). The pressure-oscillation measurements from the original Trent premixer
were obtained during stable-engine operation, that is, the fuel split between the
84 T. SCARINCI
100%
Noise Amplitude (arb. units) .
80%
Original Trent Premixers
60%
40%
Modified Trent Premixers
20%
0%
0% 20% 40% 60% 80% 100%
Power (% of full load)
primary, secondary, and tertiary premixers was chosen to avoid any serious in-
stability in the combustor. This line thus represents the “background” operating
pressure fluctuations of the original design, in the absence of any observable limit-
cycle combustion oscillation. The pressure-fluctuation levels from the new Trent
60 DLE, which implements the style of the premixer as in configuration D of
Fig. 4.9, are considerably lower even though the fuel split was not purposefully
chosen. If the acoustic properties of the two combustors are relatively similar, then
the gap between the two lines can be seen as a measure of the stability offered by
the new premixers. The new premixers thus offer lower-pressure-fluctuation levels
and improved stability margin.
A more important characteristic of the pressure-fluctuation levels is their ro-
bustness against fuel-split variations. Even though one could design a nominal or
optimized fuel schedule, the engine in the field is exposed to many sources of un-
certainty (e.g., sensor- and fuel-valve accuracy, gas-composition variations, engine
deterioration over time, manufacturing tolerances, etc.) that cause the nominal (or
desired) fuel schedule to never be exactly achieved. Thus, there will be natural
fuel-split (or flame-temperature) variations on the engine. To explore the margins
against combustion instability, a large variety of fuel splits at various engine-power
levels were explored. In the industrial Trent, being a three-stage, premixed com-
bustor, it is possible to vary the flame temperature (i.e., the FAR) of two of the
three premixers independently while keeping the engine at a fixed-power condition.
In Fig. 4.16, the results of base-load, pressure-fluctuation amplitude-mapping ex-
periments with the new premixers are compared with the behavior previously
observed with the original premixer. In both cases the engine is operating at
steady-state base load, and the fuel splits to the primary and secondary pre-
mixers are explored systematically. As before, fuel splits are expressed in terms
of a flame-temperature bias from the nominal (or reference) flame temperature.
COMBUSTION INSTABILITY AND ITS PASSIVE CONTROL 85
10.00
RMS Noise Amplitude (% of Pref) .
1.00
- 50K
REF
+ 60K
+ 75K- REF 75K
0.10
0.01
-150 -100 -50 0 50 100 150
secondary temperature bias, K
Fig. 4.16 Comparison of the original1 and modified Trent premixers,10 the showing
influence of fuel splits on pressure-fluctuation amplitude, as obtained from a mapping
procedure at engine base load.
Again, Fig. 4.16 clearly shows that the overall pressure-fluctuation levels from
the new premixers (that have FAR-wave damping) is roughly one order of magni-
tude less than those from the original combustor. Also note that the dependence of
pressure-fluctuation amplitude on fuel splits is significantly reduced. Thus, it seems
reasonable to assume that the presence of FAR-wave damping has, in the Trent
combustor, fundamentally addressed what might have been the dominant mecha-
nism of the instability, that is, the Richards–Lieuwen mechanism.
Having established a satisfactory configuration that was validated on develop-
ment engines, the new mixing-duct concept was tested on a real power-generating
site. Over a field trial of 3500 h, the engine showed satisfactory compliance at all
times against the 25-vppm legislation requirement for the site, while keeping the
same low-pressure-fluctuation levels as seen on development engines. Details of
the field-trial results are in Ref. 10.
3. Acoustic Resonators
Because the topic of resonators is covered in detail in several other chapters (e.g.,
see Chaps. 13 or 17), our discussion of their implementation in the Trent engine will
be brief. A suitably designed and well-located pressure-wave attenuator clearly can
significantly reduce the strength of feedback coupling for most if not all feedback
mechanisms that can lead to combustion instabilities. Despite the great care taken
to remove FAR oscillations from the premixers, we nonetheless included several
pressure-wave attenuators as part of the combustor structure. We included these
attenuators as a potential protection against instability mechanisms other than the
Richards–Lieuwen mechanism (see Fig. 4.8).
86 T. SCARINCI
Pressure Wave
Damping
FAR-Wave Damping
Aerodynamic
Damping
Fig. 4.17 Cross section of the modified two-stage Trent 60 DLE combustion system.
was to damp out these pressure fluctuations as much as possible within the combus-
tion system. Several airflow-damping devices were tested and their configuration
was optimized. These devices are located upstream of the premixers (see Fig. 4.17).
Some of the configurations tested included a number of reticulated materials,
such as honeycomb and metal foams, all of which were tested for various thick-
ness, porosity, and pressure-loss characteristics. The device reduces aerodynamic
plenum fluctuations by more than 20 dB over a wide-frequency range.11 Thus, the
premixers and, hence, the flame experience a reduced level of velocity fluctuation,
resulting in a quieter combustion process. Note that the pressure loss introduced
by the damping device is typically less than 0.2%.
Acknowledgments
The investigative work on combustion instability, or combustion oscillations,
summarized here, and the physical and practical understanding that came with
it as a result, was carried over a number of years and by quite a large number
of people. Nonetheless, a few individuals must be singled out. Chris Freeman,
now retired from Rolls-Royce (Derby, United Kingdom) and Ivor Day, from the
Whittle Laboratory of Cambridge University, have been long-standing soul mates
on our quest to find passive design solutions to combustion instability in DLE
systems. Chris was key in formulating the right penetrating questions at times
when only confusion seemed to be the correlating parameter. Ivor, through his
heart-of-the-matter $1 experiments, has a unique ability to redefine what common
sense actually means. I have learned much from these two close friends and this
chapter is really dedicated to the fun we had throughout this project.
My colleagues, past and present, mainly from Montreal, but also Derby, Ansty,
and Indianapolis have often made it possible to translate research ideas into real
88 T. SCARINCI
engine hardware and results and have been a constant source of support. It’s been
a privilege working with them.
Some key senior members of Rolls-Royce Engineering, Chris Barkey, Vic
Szewczyk, Mike Howse, and Phil Ruffles (now retired) have been crucial sup-
porters of the ideas put forward in this chapter. They all offered and created an
appropriate environment for technical innovation to happen.
References
1
Scarinci, T., and Halpin, J. H., “Industrial Trent Combustor–Combustion Noise Char-
acteristics,” Journal of Engineering for Gas Turbines and Power, Vol. 122, No. 2, 2000,
pp. 280–286.
2
Richards, G., and Janus, M. C., “Characterization of Oscillations During Premix Gas
Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 120, No. 2,
1998, pp. 294–302.
3
Lieuwen, T., and Zinn, B. T., “A Mechanism for Combustion Instabilities in Premixed
Gas Turbine Engines,” Journal of Engineering for Gas Turbines and Power, Vol. 242, No. 5,
2001, pp. 893–905.
4
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1809–1816.
5
Poinsot, T., Trouvé, A., Veynante, D., Candel, S., and Esposito, E., “Vortex Driven
Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics, Vol. 177,
1987, pp. 265–292.
6
McManus, K. R., Poinsot, T., and Candel, S. M., “A Review of Active Control of
Combustion Instabilities,” Progress in Energy and Combustion Science, Vol. 19, No. 1,
1993, pp. 1–30.
7
Manneville, P., “Dissipative Structures and Weak Turbulence,” Perspective in Physics,
edited by H. Araki, A. Libchaber, and G. Parisi, Academic Press, San Diego, 1990, Chap. 1.
8
Cheng, W. K., Summers, T., and Collings, N., “The Fast-Response Flame Ionization
Detector,” Progress in Energy Combustion Science, Vol. 24, 1998, pp. 89–124.
9
Scarinci, T., and Freeman, C., “The Propagation of a Fuel-Air Ratio Disturbance in a
Simple Premixer and its Influence on Pressure Wave Amplification,” American Society of
Mechanical Engineers, Paper 2000-GT-0106, May 2000.
10
Scarinci, T., Freeman, C., and Day, I., “Passive Control of Combustion Instability in a
Low Emissions Aeroderivative Gas Turbine,” American Society of Mechanical Engineers,
Paper 2004-53767, June 2004.
11
Scarinci, T., and Barkey, C., “Dry Low Emissions Technology for the Trent 50 Gas
Turbine,” Proceedings of PowerGen Europe, Pennwell, U.K. 2004.
Chapter 5
I. Introduction
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Power Generation.
89
90 W. KREBS ET AL.
Table 5.1 Performance data for selected Siemens gas turbines for
large-scale applications
Gas-turbine frame
Values SGT-1000F SGT5-4000F SGT6-5000F SGT6-6000G
General/boundary conditions
Grid frequency 50/60 50 60 60
Rotor speed, rpm 5400 3000 3600 3600
Performance data, single cycle
Gross power output, MW 68 278 198 266
Gross efficiency, % 35.1 39.0 38.0 39.3
Performance data, combined cycle-single shaft
Net power output, MW 101 407 293 391
Net efficiency, % 52.6 57.7 57.0 58.4
with a gross power output of 266 MW is the largest engine in the 60-Hz range. The
cross section of the engine is shown in Fig. 5.2. It is fired by 16 Can-type combus-
tion systems with steam-cooled transitions. The SGT6-5000F is also fired by 16
can-type combustors that are air cooled with a gross power output of 198 MW. All
combustors operate in premix mode at base load to provide low NOx emissions.
III. Phenomenological Description
Combustion-driven oscillations or thermoacoustically induced oscillations (also
called combustion dynamics) are characterized by a feedback cycle that converts
chemical energy to acoustic energy at a rate of about 10−4 . Typical for thermo-
acoustic oscillations is the observation of pronounced peaks at the resonance fre-
quencies of the combustion system.
1.0
HFD
LFD
IFD
0.9
0.8
0.7
Amplitude
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0 500 1000 1500 2000 2500 3000 3500
Frequency (Hz)
Air
Combustor
Plenum
properties of the burner plenum have an impact on the impedance of the burner
exit, thus affecting the magnitude of the pressure-induced fluctuations in volume
flow. The acoustic properties of the fuel lines determine the magnitude of fuel flow
fluctuations at the fuel nozzles. The interaction between the fuel flow and airflow
results in equivalence-ratio oscillations, which in turn produce oscillations of the
heat release. The acoustic geometry of the combustor determines the shape of the
acoustic modes, which are also influenced by the acoustic-boundary conditions
at the combustor exit. The source of the instability is the perturbed flame. In that
context it is important to understand and quantify the heat-release fluctuations
induced by dynamic-pressure perturbations.
The technical objectives of thermoacoustic design are 1) Determine the nature
of the thermoacoustic-feedback cycle and investigate the interaction of different
components. 2) Optimize the thermoacoustic-design process: Identify thermoa-
coustically relevant design parameters and evaluate the thermoacoustic impact of
design changes. 3) Optimize the prediction capability of test rigs and develop com-
putational models to predict engine performance based on rig results. For acoustic
energy balance, note that the acoustic energy generated by the flame is mainly lost
at the inlets and outlets of the combustion system under consideration. Because
the test rigs differ especially at their inlets and outlets from the engine design,
the knowledge of these losses is of crucial importance for successful gas-turbine
combustion design. 4) Develop active and passive means for the suppression of
thermoacoustic oscillations.
Prediction methods
Acoustic properties Transmission line method Resonator design
of combustor parts
One-dimensional Transfer matrix network Evaluate acoustic properties
acoustic analysis like impedances for
gas-turbine components
Three-dimensional Finite element methods, Evaluate acoustic-pressure
acoustic analysis Sysnoise distributions for
a) monitoring
b) developing resonators etc.
Flame response Time-lag models Evaluate impact of design
changes on flame response
Flame response Unsteady computational Evaluate impact of design
fluid dynamics changes on Flame Response
Full-stability analysis Transfer-matrix approach Evaluate impact of design
Galerkin method changes on stability
Interaction of components
Component Testing
Tunable rig with Evaluate impact of design
variable-exhaust- changes on stability
passage impedance Interaction of components
The main advantage of the method is that the acoustic properties of each acousti-
cally relevant component (e.g., a duct) are represented by a separate transfer matrix.
The acoustic properties of the complete system are obtained by connecting these
transfer matrices in a transfer-matrix network. Applications of the transfer-matrix
approach to gas-turbine combustion systems have been described by Krüger et al.4
The transfer-matrix method is successfully applied to identify the properties of cer-
tain components of the gas-turbine combustion system, like fuel lines or exhaust
passages of test rigs, for which the one-dimensional sound propagation is valid.
1) Simplified models of only the combustion chamber that include only one
burner (Fig. 5.7, right-hand side). Impedance-boundary conditions are set on the
boundary faces.
2) The impedance boundary conditions are important for the final stability and
the acoustic-pressure level of the system because the impedance at the acoustic
boundary face determines the acoustic energy loss over that surface.7 Hence, to
Can-combustion-chamber
Transition-piece
Simulated ambience Plenum
Burner
Vane-simulation-section
Diffuser-inlet
Exhaust passage
Fig. 5.7 Typical computational domain for finite element acoustic analyses ranging
from diffuser inlet–compressor exit to exhaust.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 97
improve the accuracy of the prediction the computational domain can be extended
as shown in Fig. 5.7 (left-hand side). The computational domain now includes the
incoming flow path upstream of the burner (plenum), the combustion chamber,
turbine vanes or the vane-simulation section, and the exhaust passage.
3) The most complex (and most costly) model covers the whole annular or
can-annular combustion system, including all the burners in the entire engine.
Because the acoustic environment of a test rig and an engine differ, the three-
dimensional acoustic analysis using finite element codes is essential for the evalu-
ation of test-rig results. In addition the codes are used to point out the differences
in acoustic properties between engine and rig.
C. Flame Response
The analysis of the flame response is crucial for the evaluation of the thermoa-
coustic stability of gas-turbine combustion systems because, as shown by Poinsot
and Veynante,7 it determines the source term in the transport equation for acoustic
energy. In general, the flame-response function expresses the heat-release fluctua-
tions induced by acoustic-pressure waves. The instantaneous response of a flame
caused by an acoustic perturbation is given by the pressure-coupled and velocity-
coupled response functions given in Eqs. (5.1) and (5.2), respectively:
q (t)/q̄
F1 (t) = (5.1)
p (t)/ p̄
q (t)/q̄
F2 (t) = (5.2)
u (t)| burner exit face /ū burner exit face
q describes the integral of heat-release fluctuation over the flame surface, defined
as
q (t) = Q (r , t) dV [W ]
heat release zone
Technical aspects and the impact of design parameter on the flame response are
discussed by Krebs et al.10
In conclusion, the dynamic properties of the flame are mainly represented by
the time lag and different combustion designs can be compared by looking at their
time-lag distribution. The time-lag distribution can be obtained in a postprocessing
step to steady computational fluid dynamics (CFD) analysis.
The approaches based on a steady-state analysis have a main drawback; that
is, the impact of pressure waves on the strain rate and the impact of the flame–
vortex interaction cannot be represented. In addition these approaches assume that
the amplitude of the fluctuation once generated does not change its value while
approaching the flame front. Finally, the steady-state approach assumes that the
flame is stationary; in some cases, the mean flame position shifts when combustion
instability occurs. More insight into the detailed processes can be obtained by
unsteady CFD methods like large-eddy simulation (LES), in which the large-scale
vortices are resolved. Siemens is currently developing unsteady methods to treat
this issue.
D. Full-Stability Analysis
The goal of a full-stability analysis is to predict the excitation or damping
of an acoustic mode. This process involves the representation of the acoustic-
feedback cycle between the flame response and the acoustic environment. The
thermoacoustic stability is the result of the thermoacoustic flame response and
the acoustic properties of all components between compressor and turbine exit. In
contrast to CFD methods the computational domain must be extended to encompass
all components. A solution method that has been extended to meet this requirement
is the transfer-matrix network. This method is explained subsequently, with the
analysis of the annular combustor rig as an example. The cross section of the
annular combustor rig is shown in Fig. 5.8.
Fig. 5.9 Acoustic transfer-matrix network representing the annular test rig.
The airflow enters at the test rig inlet (left), flows through the compressor outlet
diffuser, and is discharged through 24 burners. Combustion takes place in an an-
nular combustor. The hot gases are finally discharged through the exhaust-diffuser
passage. The annular combustor is operated at ambient conditions, and hence
acoustically soft conditions (i.e., low impedance) have been assumed at the rig
inlet and outlet.
The acoustic network representing the test rig is shown in Fig. 5.9. In this ex-
ample the annular combustor rig is represented by two interconnected rings of
one-dimensional duct elements. The outer ring represents the plenum, whereas
the inner ring represents the annular combustor. The length of the duct elements
representing the annulus are selected to cover the average acoustic-passage length.
The average acoustic-passage length is determined as the circumference having
the same azimuthal eigenfrequencies as the annulus of finite width. The eigenfre-
quencies of the combustor are obtained by three-dimensional finite element modal
analysis. This approach is valid because in the annular combustion systems of
the Siemens product family (SGT-1000F, SGT6-4000F, and SGT5-4000F) pure
azimuthal mode shapes are predominant. The elements shown are submodels con-
sisting of further elements representing the actual geometry as shown in Fig. 5.5.
Altogether the whole system may contain more than 1000 different transfer-
matrix elements. The transfer-matrix representation is more-or-less a mathematical
framework in which all the information on acoustic properties of the different
components can be lumped together. The quality of the model depends on the
quality of these elements. In addition to models of annular ducts, Siemens AG has
specially developed models for diffusers and nozzles capable of generating the
100 W. KREBS ET AL.
Table 5.3 Complex eigenfrequencies found for the annular combustor rig
transfer matrix, even for high-Mach-number flows that may occur in fuel-supply
systems. The network formulation results in a matrix equation described by, for
example, Hubbard and Dowling,11 which can be solved to obtain the complex
eigenfrequencies (ωn = ω f + i ∗ α) of the system.
To determine the stability of the system, the determinant of the transfer ma-
trix is calculated by using the appropriate boundary conditions. According to the
decomposition defined earlier, the amplitude of a pressure oscillation grows if
the imaginary part of the complex eigenfrequency becomes negative. This part
is called the damping coefficient. The complex eigenfrequencies found for the
annular combustor rig operated at nominal conditions are listed in Table 5.3.
Table 5.3 indicates an unstable eigenmode at 196 Hz, which is in agreement with
experiments in which a single unstable mode at about 200 Hz has been found.
V. Application
Because thermoacoustic stability results from the interaction of several different
components, numerous design options are available to increase the thermoacoustic-
stability range. As described in the preceding section, solution methods and design
tools are needed to identify promising design modifications and to investigate
their impact quantitatively. The goal of each of these options is to extend the
operating range of the engine to improve performance or emissions and to extend
the operating life of the engine components.
Several design options are listed in Table 5.4. Most design options considered
involve changes of the burner design and they primarily affect the flame response.
One approach is to reduce the interaction of the thermoacoustic source with the
pressure field by changing the time-lag distribution. Another option is the ap-
plication of different types of resonators to increase the damping of the system
by absorbing acoustic waves amplified in a certain frequency range. Their effi-
ciency greatly depends on the width of the frequency range at which damping
is added. A third approach is to make use of control methods. Active instability
control (AIC) systems achieve stability by perturbing the combustion process at or
near the frequency of the combustion instability (normally 100–200 Hz) to damp
the cycle-to-cycle pressure oscillations that occur on the order of milliseconds.
Hoffmann and Hermann (Chap. 19) describe such a system that was used in
Siemens products SGT6-4000F and SGT5-4000F. Another control strategy is to
use low-bandwidth control (<2 Hz; e.g., industrial fuel-control valves) to make
adjustments to the mean operating conditions (airflow, fuel flow, and fuel distri-
bution). Such methods are called active combustion control or automatic tuning
systems. Both the high-bandwidth and low-bandwidth control approaches require
reliable real-time measurement of the combustion-process oscillations; dynamic
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 101
Table 5.4 Design options for extending the safe operation range of
gas-turbine combustion systems
Burner
1 Change flame-front location by modifiying Flame response
burner exit geomentry (cylindrical burner, outlet)
2 Changing of fuel-concentration profiles Flame response
3 Fuel staging technology Flame response
4 Adaptive fuel-nozzle impedance Flame response
5 Modify heat-release distribution Flame response
Combustor
6 Resonators Damping
Control
7 Active instability control Flame response
8 Active combustion control Flame response
Engine:
24-burner
configuration
Fig. 5.10 Modulus of the dynamic-pressure distribution: first azimuthal mode shape
in annular combustion chamber.
102 W. KREBS ET AL.
Acoustic excitation
at burners
in-phase alternating out-of-phase
Phase ‘Can-Can
- Phase Interaction’!
+
Fig. 5.11 Axial mode shape in can-annular combustion chamber.
grey indicates a pressure antinode, whereas dark grey indicates a pressure node. In
the intermediate range of frequencies, azimuthal mode shapes are predominant in
annular combustion systems of large-scale engines like Siemens products SGT6-
4000F and SGT5-4000F. Azimuthal mode shapes are characterized by a dominant
dependence of the dynamic-pressure amplitude on the azimuthal coordinate as
visualized in Fig. 5.10. The presence of two pressure antinodes and two pressure
nodes indicates a first azimuthal mode corresponding to a frequency of about
100 Hz. The thermoacoustics of can-annular combustion systems is driven by
axially oriented mode shapes as shown in Fig. 5.11. In the engine, basically two
types of mode shapes are possible. On the left-hand side the dynamic pressures of
all single can-combustion chambers are acoustically in phase. On the right-hand
side the pressures in neighboring cans are acoustically out of phase. In this case
the different cans are acoustically connected by the passage in front of the turbine.
The impact of the can–can interactions through the annular manifold upstream
of the turbine inlet on the acoustical behavior of can-annular combustion systems is
shown in Fig. 5.12. In addition to the acoustic modes that are present in a single-can
configuration (compare, solid line), there are further modes in between (compare,
dotted line). These modes are characterized by relatively high-acoustic velocities
in the annular manifold between the can-combustion chambers. To include all
phenomena during the testing phase the exhaust passages of the test rigs have to be
modified accordingly to get the desired mode shape. In addition the analysis reveals
that acoustic phenomena in can-annular combustion systems are distributed over
neighboring cans rather than limited to a single can.
B. Flame Response
A design method that makes use of changing the flame response has been suc-
cessfully applied in the V-Frame. This design approach is sketched in Fig. 5.13.
Mounting cylinders of different lengths on the burner exit results in an elongation
of the flame front.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 103
The temperature distributions for the different design configurations are shown
in Fig. 5.14. The zero-velocity contours are also plotted to indicate the size of the
recirculation zones. The burner installed in the SGT-1000F is the so-called hybrid
burner shown in Fig. 5.5. Swirl induces a central-recirculation zone that provides
flame stabilization over the whole gas-turbine operation envelope. In addition,
outer-recirculation zones are formed that also help to stabilize the flame. As typical
Modifications
NBO
CBO short
CBO long
Fig. 5.13 Modification of the burner outlet by cylinders of different lengths (NBO =
normal burner outlet; CBO = cylindrical burner outlet).
104 W. KREBS ET AL.
NBO CBO-Short
CBO-Long
1.00
0.80
Distribution [ - ]
NBO
0.60
Short CBO
0.40 Long CBO
0.20
0.00
0.0000 0.1000 0.2000 0.3000 0.4000
t / t_0 [ - ]
travel between the burner-outlet surface and the flame front has been statistically
evaluated. In this procedure the flame front is characterized by the maximum heat-
release rate encountered by the particle track during its trip through the combustor.
The statistical evaluation results in a time-lag distribution that is generated by the
finite extension of the flame. The time lag corresponding to the modification with
no CBO is the smallest and is characterized by the narrowest distribution. The
longer CBO leads to significantly increased time-lag values and more widespread
distribution. Application of the Rayleigh criterion showed that the increased time
lag improves the thermoacoustic stability; this result was verified in field tests.5
The concept works well because the reacting flowfield is only shifted by the CBO
and the main parameter that has been changed is the time lag.
Other design options for affecting the flame response–time-lag distribution are
changing the fuel-concentration profile at the burner exit, changing the fuel-nozzle
impedance, and changing residence time in the combustor by adjusting the burner
diameter or swirl number. These design options may change other stability para-
meters, mainly arising from the flame–vortex interaction, which is not understood
at presents hence, the simple time-lag concept cannot be used to evaluate these
design changes. Siemens is currently investigating the flame–vortex interaction to
improve the design methodology in this aspect. Another design option widely used
is the implementation of fuel-staging concepts, which allows modification of the
flame response during engine operation in response to combustion instability.
C. Resonators
1. High-Frequency Resonators
Relatively few analytical or numerical analyses of high-frequency dynamics
have been performed so far because of the high-temporal resolution that is needed
to investigate the phenomenon. Feedback cycles supporting the dynamic excitation
have time constants in the range of 0.2 to 0.6 ms, which are the time constants
of the combustion process itself. Hence, the origin of high-frequency dynamics
should be found in the reacting shear layer. Although the origin of the phenomenon
remains unclear, high-frequency dynamics can be well damped by resonators.
With Helmholtz resonators, a cavity is acoustically connected to the inside of the
combustion chamber by an orifice as shown in Fig. 5.16.
The eigenfrequency of a typical Helmholtz resonator configuration is given by
c Sneck
f Helmholtz = (5.3)
2π V (l + l)
The eigenfrequency is determined by the ratio of the area of the orifice of the
resonator tube S and resonator volume V times the length of the resonator tube l.
l expresses the elongation of the resonator tube to take into account radiation
effects. l is computed approximately as 0.85Dtube . The speed of sound is denoted
by c.
To increase the damping performance of high-frequency resonators, the
acoustic-absorption area of the resonators has to be high. In practical configu-
rations of resonators, several resonator tubes are connected in parallel to one large
106 W. KREBS ET AL.
Holes for
entering purge
air
Helmholtz-
Volume V
Combustor
L
Resonator Tube
Dtube
2. Intermediate-Frequency Resonators
Helmholtz resonators may also be used for suppressing intermediate-frequency
dynamics. To design intermediate-frequency Helmholtz resonators for annular
combustors, tests in an annular combustor rig have been performed. The annular
combustor rig is shown in Fig. 5.18.
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 107
A cross section of the rig has already been shown in Fig. 5.8. The annular
combustor rig is equipped with an exit diffuser connected to a large exhaust volume.
When operating the annular combustor rig at atmospheric conditions, a strong
200-Hz combined azimuthal and axial mode is observed. In the tests 14 Helmholtz
resonators were mounted on the outer shell of the combustor to suppress this
mode. The arrangement of resonators is depicted in Fig. 5.19. The nonsymmetric
1.5 m
Fig. 5.19 Combustor test rig with 14 Helmholtz resonators placed circumferentially.
In Fig. 5.21 the rms value of the dynamic pressure is plotted vs air–fuel ratio.
The air mass flow through the combustor is held constant at 8 kg/s and the pilot
mass flow is set to 48 g/s. For the baseline case without resonators the fuel mass
flow can increase toward an airflow ratio (AFR) of 35. If the fuel mass flow is
further increased a sudden excitation of the 200-Hz mode occurrs. This mode can
only be damped if the fuel mass flow is considerably reduced toward an AFR of 38.
Fig. 5.23 Eigenfrequencies of test rig with (circles) and without (squares) resonators.
Such a hysteresis has also been observed at different operating points characterized
by different air mass flows and pilot mass flows. With the large resonator, the fuel
can be increased up to an AFR of 30.5 before a sudden excitation occurs. Again
the mode can only be completely damped if the fuel mass flow is decreased toward
an AFR of 38; however, the amplitude is reduced quickly when AFR is increased
above 31.0. The implementation of the resonators results in a significantly wider
oscillation-free operation envelope.
To assess the damping achieved with this configuration before its realization, a
thermoacoustic model was set up using the network method described previously.
The model shown in Fig. 5.22 corresponds to the model of the combustor rig in
Fig. 5.9 equipped with resonators at the locations where they are planned to be
installed in the test rig.12 The transfer-matrix model of the Helmholtz resonators has
been qualified in single-impedance tests. It has been connected between the outer
annular ring of the plenum and the inner ring representing the annular combustor.
Computational analyses have shown that sufficient damping can be achieved with
this configuration. In Fig. 5.23, eigenfrequencies of the test rig without and with
resonators are compared. It can be seen that the originally critical eigenfrequency
is now replaced by a high number of strongly damped eigenfrequencies. It can
therefore be expected that the test rig can be operated at design conditions in
stable mode.
VI. Conclusion
Thermoacoustically induced combustion oscillations are a complex phe-
nomenon arising from the interaction of the flame with the acoustic environ-
ment that is present in all types of combustion systems. The complexity of the
engineering treatment originates from the long wavelengths associated with the
THERMOACOUSTIC DESIGN TOOLS AND PASSIVE CONTROL 111
References
1
Szabo, T. L., “Lumped-Element Transmission-Line Analog of Sound in a Viscous
Medium,” Journal of the Acoustical Society of America, Vol. 45, 1969, pp. 124–130.
2
Munjal, M. L., Acoustics of Ducts and Mufflers with Application to Exhaust and Venti-
lation System Design, Wiley, New York, 1987.
3
Prade, B., Gruschka, U., Hermsmeyer, H., Hoffmann, S., Krebs, W., and Schmitz, U.,
“V64.3A Gas Turbine Natural Gas Burner Development,” American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
4
Krüger, U., Hüren, J., Hoffmann, S., Krebs, W., Flohr, P., and Bohn, D., “Prediction and
Measurment of Thermoacoustic Improvements in Gas Turbines with Annular Combustion
Systems,” Journal of Engineering for Gas Turbines and Power, Vol. 123, 2001, pp. 557.
5
Bethke, S., Krebs, W., Flohr P., and Prade, B., “Thermoacoustic Properties of Can
Annular Combustors,” AIAA Paper 2002-2570, May 2002.
6
Morse, P. M., and Ingard, K. U., Theoretical Acoustics, McGraw–Hill, New York, 1968.
7
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Flourtown, PA, 2001.
8
Büchnér, H., Lohrmann, M., Zarzalis, N., and Krebs, W., “Flame Transfer Func-
tion Characteristics of Swirl Flames for Gas Turbine Applications,” American Society of
Mechanical Engineers, Paper GT-2003-38113, June 2003.
9
Cho, J. H., and Lieuwen, T., “Laminar Premixed Flame Response to Equivalence Ratio
Oscillations,” Combustion and Flame., Vol. 140, No. 1-2, pp. 116–129, Jan. 2005.
112 W. KREBS ET AL.
10
Krebs, W., Flohr, P., Prade, B., and Hoffmann, S., “Thermoacoustic Stability Chart for
High Intense Gas Turbine Combustion Systems,” Combustion, Science and Technology,
Vol. 174, 2003, pp. 99–128.
11
Hubbard, S., and Dowling, A. P., “Acoustic Instabilities in Premix Burners,” AIAA
Paper 98–2272, 1998.
12
Lepers, J., Krebs, W., Prade, B., Flohr, P., Pollarolo, G., and Ferrante, A. “Investigation
of Thermoacoustic Stability Limits of an Annular Gas Turbine Combustor Test Rig with
and without Helmholtz Resonators,” American Society of Mechanical Engineers, Paper
GT-2005-68246, 2005.
Chapter 6
I. Introduction
Sections 1–3 and 5–7 are works of the U.S. Government and are not subject to copyright protection
in the United States. Section 4, Copyright
c 2005 by the United Technologies Corporation. Published
by the American Institute of Aeronautics and Astronautics, Inc., with permission.
∗ Aerodynamics Manager, Combustor and Augmentor Technology.
† Fellow, Instability Modeling & Analysis.
‡ Controls Lead, Active Combustion Control Technology.
113
114 J. M. COHEN ET AL.
The demonstration of the utility of this process will lend credence to its a priori
use in the combustor design and development process.
70
60
50
Elapsed Time, sec
40
Mid-power
point chosen for
30 analysis
20
10
0
0 100 200 300 400 500 600 700 800 900 100
Frequency, Hz
Fig. 6.1 “Waterfall” plot of combustor pressure spectra from an engine test, showing
evolution of the instability during an acceleration event.
116 J. M. COHEN ET AL.
Although fast-response combustor pressure data were acquired during the engine
tests, they were acquired at a limited number of locations. For this reason, it was
difficult to draw any significant conclusions about the nature of the instability
purely from the engine data. The analyses described in Sec. III were used to
augment and interpret the engine data. The combination of the engine data, the
acoustic analyses, and the nonreacting swirler–injector characterization provided
the basis for replicating the problem in a single-nozzle combustor rig.
Engine
Rig
Downstream
Upstream Plenum
Plenum Air Swirler
Prediffuser Combustor
Choked
Exit
Fig. 6.2 Cross-sectional area vs axial position for quasi-one-dimensional Euler model.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 117
The solver was first used to compute the steady-state results, and then unsteady
results were obtained by forcing the system. Forcing was accomplished by adding
an unsteady component to the heat release. Two types of forcing signals were
used: 1) broadband white-noise distributed forcing, and 2) swept-sine forcing
over a range of suitable frequencies. The resulting pressure response indicated the
frequency dependence of the combustion system.
Quasi-one-dimensional Euler calculations were conducted for the engine
configuration at an intermediate operating condition: 771◦ F (684 K) and 200 psia
(1.2 MPa). The engine geometry was converted into a one-dimensional description
of area vs axial position, as shown by the dashed line in Fig. 6.2, which shows the
distributions for both the engine and the rig (to be discussed later). The geometry
used included an inlet plenum, the engine prediffuser, diffuser plenum, the cowl
or hood, the swirler, the combustor liner, and turbine vanes. The combustor lies
between x = 0 and 9.25 in. (23.5 cm). Beyond the turbine vane choked exit the
area was expanded rapidly to create a plenum dump. The boundary conditions
used were constant total pressure at the inlet plenum and constant static pressure
at the exit plenum.
The acoustic response of the system was obtained by swept-sine forcing of
the entire heat-release distribution. The fluctuation levels imposed on the heat
release were 10% of the mean. The unsteady pressure amplitude at the x = 3 in.
(7.6 cm) location in the combustor (3 in. downstream of the combustor dump
plane) was used to determine the pressure response. The swept-sine response is
shown in Fig. 6.3 and indicates a resonance at approximately 575 Hz. The width
log(Unsteady Pressure (psi))
Fig. 6.3 Computed power spectrum of combustor pressure for engine configuration
with sine-sweep forcing.
118 J. M. COHEN ET AL.
0.02
Unsteady Pressure Amplitude @575 Hz(psi)
0.015
0.01
0.005
−0.005
−0.01
−0.015
−0.02
−10 −5 0 5 10 15 20
X (in.)
Fig. 6.4 Computed pressure mode shape for 575-Hz mode for the engine configura-
tion at evaluation point conditions.
In the fuel injector–air swirler under study, air is passed through inlets into
a multipassage swirler. At the exit of the swirler, the highly swirled air flows
sandwich a thin sheet of fuel between them, shearing it into small droplets. The
first study, discussed in Sec. IV.A, investigated the pure acoustic (no throughflow)
of the fuel injector–air swirler to identify potential acoustic resonances in the
frequency range of interest. The second study, discussed in Sec. IV. B, measured
the effects of throughflow on the swirler impedance and unsteady fluid mechanics
to identify potential unsteady fluid mechanic coupling. The third study, described
in Sec. IV.C, used a new measurement technique to assess how the fuel spray was
modulated by acoustic pressure fluctuations.
A. Acoustic-Impedance Measurements
The objectives in making air swirler acoustic-impedance measurements were
2-fold. First, acoustic-impedance measurements will reveal any interesting atten-
uation or amplification characteristics of the air swirlers in the frequency range of
interest, 300 Hz to 600 Hz. Any resonances or antiresonances observed will aid in
interpretation of the combustor test data and guide design of an efficient strategy for
acoustic–fluid dynamic interaction tests. Second, experimental air swirler acoustic-
impedance measurements are needed inputs into combustor acoustic models.
Acoustic impedance is defined as the complex quotient of acoustic pressure di-
vided by acoustic volume velocity at a surface.4 It can be thought of as a frequency-
dependent transfer function that describes the response of the device to an incident
acoustic wave. The acoustic impedance includes resistive effects associated with
pressure losses and reactive effects associated with the inertance and compliance
of the device.
An experimental impedance-tube rig was used to measure the air swirler acoustic
impedances. Important design considerations for the rig were the overall tube
length, tube inner diameter, and the axial locations of the acoustic transducers.5
Overall tube length is suggested at least 5 L/D’s (length/diameters) to ensure that
the waves incident on the duct exit plane are planar. Tube lengths of 0.3 m (1 ft) and
0.91 m (3 ft) were used for these measurements. The duct had roughly the same
cross-sectional shape and diameter as that of the swirler for which impedance was
measured. The two-transducer, transfer-function method was used to measure the
acoustic impedance of the air swirler.5 To validate the measurement technique, the
open-tube acoustic impedance was measured and compared with analytical theory
for a long, open radiating tube.6 The measured impedance was in good agreement
with theory in the band of frequencies from 100 Hz to almost 1000 Hz.
Experimental measurements of air swirler acoustic impedance are shown with
the bounding cases of open-tube and rigid-wall boundary conditions in Fig. 6.5.
The impedance curves of different swirlers were essentially grouped based on the
effective airflow area of the separate devices. These different swirlers represented
a design matrix of swirl angles, effective areas, and flow splits. The frequency
response of the air swirlers is similar to that of an open tube with about 10 dB higher
impedance. The air swirlers did not exhibit significant low- or high-impedance
spikes in the frequency range of interest, indicating the absence of resonances or
antiresonances. As a result, the no-flow acoustics-impedance measurements did
not identify any particular resonances in the frequency range of interest.
120 J. M. COHEN ET AL.
60 200
Open tube
Open tube 150 Closed tube
40 Closed tube Swirler
Swirler 100
Magnitude (dB)
Phase (deg)
20 50
0 0
-50
-20
-100
-40 -150
-60 -200
200 400 600 800 1000 200 400 600 800 1000
Frequency (Hz) Frequency (Hz)
Fig. 6.5 Comparison of measured swirler impedance with open and closed tubes.
fD
St = (6.1)
U
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 121
-40
-42
-44
Velocity (dB)
-46
-48
-50
-52
-54
200 400 600 800 1000
Frequency (Hz)
Fig. 6.6 Spectra of plenum pressure, swirler exit velocity for 500-Hz forcing. Coher-
ence = 0.85.
122 J. M. COHEN ET AL.
-10 200
150
-20
100
Magnitude (dB)
-30
Phase (deg)
50
-40 0
-50
-50
-100
-60
-150
-70 -200
100 200 300 400 500 600 700 800 100 200 300 400 500 600 700 800
Frequency (Hz) Frequency (Hz)
Fig. 6.7 Swirler exit velocity/plenum pressure transfer function using swept-sine
technique.
Microphone Acoustic
driver
Spray
nozzle
PMT
Spray
pattern Receiver
Amplifier
Light sheet
Signal
forming
Analyzer
system
Optical Laser
Fiber
Fig. 6.8 Test setup for fluorescence measurements of fuel mass flow fluctuations. The
airflow system includes an acoustic driver that perturbs the flow. The microphone
measures the acoustic forcing of the spray.
the spray excited fluorescence proportional to the mass of the fuel within the
sheet. Continuous monitoring of the fluorescence intensity with a single detector
provided a temporal record of the mass and its fluctuations. Correlations of mass
fluctuations with measurements of the acoustic pressure signal provided a means
of examining the level of coupling. The laser sheet could be moved along the spray
axis to determine the spatial nature of the coupling process. A detailed discussion
of the optical technique is contained in Ref. 2.
The optical technique was used to characterize the dynamical signatures of four
fuel injector–swirler combinations. They will be referred as injectors A through D,
respectively, with injector A being the injector tested in the engine. The injectors
utilize the relative motion9 between a low-velocity sheet of liquid fuel and sur-
rounding high-velocity air streams to effectively disrupt and break apart the liquid
sheet into unstable ligaments and large droplets. The liquid fuel (Jet A) is injected
onto a filming surface and is then atomized through the combined influences of
inner and outer airflows having identical swirl directions.
An initial assessment of the technique was done in the form of a calibration.
This entailed measuring the photomultiplier tube (PMT) signal for a variety of
fuel flow rates. The airflow rate was regulated by a choked venturi and adjusted
to yield a pressure drop of 17.2 kPa (2.5 psi) across the air swirler, whereas the
mass flow rate of fuel was varied between 0 and 113 kg/h (250 lbm/h). Both
quantities encompassed the normal (scaled to atmospheric pressure) operating
range of the device. For this series of tests, the laser sheet was located 100 mm (4 in.)
from the exit plane of the injector. The calibration for this technique was roughly
linear.2
The measured PMT signal was nondimensionalized by the incident laser in-
tensity Io . Two data sets are shown with a linear curve fit to demonstrate the
repeatability of the technique. The linearity of the curves is apparent: increasing
the mass flow rate of fuel increased fluorescence intensity and consequently the
PMT signal, as expected.
To quantify the degree of interaction between the pulsed airflow (forcing func-
tion) and the fuel spray (output), the coherence between the acoustic pressure
signal and PMT signal signals was measured for a variety of flow conditions. A
Hewlett–Packard signal analyzer provided an input to drive the valve and recorded
both the acoustic pressure measured in the air-supply plenum and the PMT signal,
representing the spray response. By continuously varying the frequency of the
forcing function, a map of the output’s response was obtained, detailing possible
frequency bands in which interactions between the two signals could be seen. This
measurement, therefore, yielded essential information on the dynamic behavior of
prospective injectors. Figure 6.9 depicts the coherence measurements from injec-
tor A at several axial stations as the driver was forced at a variety of frequencies
between 0 and 900 Hz. The figure shows a location at which the acoustic fluc-
tuations have the greatest effect on the spray. This finding is consistent with the
understanding of the evolution of the atomization process. As the liquid fuel is-
sues from the injector, it is first atomized through the influences of the momentum
flux ratio existing between the high-velocity airflow and the slower moving liquid
flow. After a primary atomization process in which large, unstable ligaments and
droplets are formed, a secondary atomization commences that further reduces the
particles’ size. This process is typically Weber number dependent.10 Only when
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 125
1.0
1 "
0.8
2 "
3 "
Coherence
0.6 4 "
5 "
0.4
0.2
0.0
0 300 600 900
Forcing Frequency (Hz)
Fig. 6.9 Coherence measurements for the injector A at several axial stations.
the droplets have attained a unique size can they be more easily influenced by
the surrounding airflow; if they are too large, prevailing forces are insufficient to
accelerate them. This process, in general, relies on the governing Stokes number
to be much less than unity. The Stokes number (St ) is typically defined as the
particle’s response time (τ p ) divided by the eddy-turnover time (τe ) or
τp
St = (6.2)
τe
If particles are too large, they become centrifuged inside turbulent or large-scale
structures so that their response to flowfield changes is negligible. As the particles
are reduced in size with increasing axial distance, creation of an optimal axial
location for forcing should occur. The reason for the drop in droplet response
for greater axial positions (>4 in.) is unclear, however. One possible explanation
could be the attenuation of the acoustic energy over an increasingly larger area
with subsequent downstream locations, thereby diminishing its influence on the
liquid droplets. Another reason could simply be the damping of the spray’s motion
because of spreading and drag effects.
Also apparent is the strong coupling in the 300- to 700-Hz region at all axial
distances. There are even tails on either side of the main peak, hinting at other
frequency bands of interaction. The increasing coherence near-zero frequency is
typical of zero-frequency functions, which have very strong linear relationships,
varying only by a gain factor.3
Figure 6.10 summarizes the coherence measurements for all four injectors taken
at 4 in. from the exit plane. Except for injector D, all the injectors exhibited spray–
acoustic coupling within the frequency range of interest (350–700 Hz). This is
important because it reflects the ability of the droplets to respond to the exter-
nal excitation, thereby allowing for the possibility of a corresponding fluctuating
heat-release trace that could lead to unstable burning. Injector C exhibited the
highest coherence over the frequency range of interest. Between 400 and 650 Hz,
the coherence remained at a fairly constant and high value of about 0.85 for this
swirler. Even at lower frequencies (150–350 Hz), the coherence remained at fairly
126 J. M. COHEN ET AL.
1.0
injector A
injector B
0.8
injector C
injector D
Coherence
0.6
0.4
0.2
0.0
0 300 600 900
Forcing Frequency (Hz)
Fig. 6.10 Summary of coherence measurements for all injector configurations, 4 in.
from the exit plane.
0.10 180
120
0.08
60
Phase (deg.)
Ma gn itu de
0.06
0
0.04
-60
0.02 -120
0.00 -180
350 400 450 500 550 600 650 350 400 450 500 550 600 650
Frequency (Hz) Frequency (Hz)
Fig. 6.11 Magnitude and phase of the mass fluctuation/ p transfer function at a lo-
cation 4 in. downstream of nozzle A. The mass fluctution transfer function is derived
from the mass/ p transfer function by multiplying by frequency.
of 1.38 in. (3.5 cm) downstream of the venturi throat. The upstream boundary was
largely established by the normal shock and the sudden expansion of the flow at
the prediffuser dump. The downstream boundary was defined by using a choked
exhaust nozzle at the station occupied by the first turbine inlet vane.
number and size of the combustion and dilution air holes was adjusted to provide
proper penetration of these air jets.
The single-nozzle combustor rig design was established to preserve the axial
lengths and cross-sectional areas of the engine configuration relative to a single
fuel nozzle. The approximately the same area vs axial position distribution was
maintained (Fig. 6.2), but some variation existed because of differences in engine
hardware and the axisymmetric hardware to be used in the single-nozzle combustor.
Airflow splits and pressure losses (swirler, bulkhead, liner, primary and dilution
jets) were also preserved by design. As mentioned, the inlet and exit of the rig were
choked to acoustically isolate the system. The configuration of the test section is
illustrated in Fig. 6.12.
Provisions for high-response pressure transducers and for gas sampling (not
reported herein) were incorporated into the test section. Three transducers, equally
spaced around the circumference, were located in the primary combustion zone.
One transducer was located in the secondary zone, and one was in the dilution
zone.
Unsteady shroud-flow pressure measurements were provided at the location of
the liner primary and dilution holes. Bosses for diffuser unsteady-pressure mea-
surements, upstream of the combustor, were also provided.
Quasi-one-dimensional Euler acoustic analyses were conducted for the baseline
rig configuration at the evaluation-point operating conditions: 771◦ F (684 K), 200
psia (1.2 MPa). The acoustic analyses included inlet and exit plenums upstream
and downstream of the choke points to allow constant total pressure and constant
static pressure to be specified, respectively, as boundary conditions to be applied
to the Euler code domain. Swept-sine forcing over the frequency range from 100
to 800 Hz was applied to the heat release. The resulting power spectrum of the
pressure response is shown in Fig. 6.13, indicating the presence of resonances at
115 Hz and 550 Hz. The level of forcing used in the analysis is arbitrary, and
within the linear-response range, so that no significance should be attached to the
absolute levels of the ordinates in Figs. 6.13 and 6.14.
Further analysis of the mode shapes associated with these resonances revealed
that the low-frequency 115-Hz mode was a first-order longitudinal mode in which
130 J. M. COHEN ET AL.
00
10-1
100 200 300 400 500 600 700 800
freq (Hz)
Fig. 6.13 Computed power spectrum of combustor pressure at x = 3.0 in. for the rig
configuration. Quasi-one-dimensional Euler code results for 100 to 800-Hz swept-sine
forcing of heat release.
the diffuser and combustor were in phase. The primary mode of interest was the
550-Hz mode because the observed instability frequency in the engine was 525
Hz. The pressure-mode shape is shown in Fig. 6.14. The 550-Hz mode was essen-
tially a half-wave longitudinal mode considering closed/closed acoustic boundary
conditions from diffuser inlet to combustor exit. A pressure node appeared to occur
at the air swirler–fuel injector location. The pressure in the diffuser was 180 deg
out of phase from the pressure in the combustor. Note that some activity occurred
downstream of the combustor exit, but calculations performed with varying exit
Unsteady Pressure Amplitude @ 575 Hz (psi)
0.08
0.06
0.04
0.02
−0.02
−0.04
−0.06
−0.08
−15 −10 −5 0 5 10 15 20
X (in.)
Fig. 6.14 Computed pressure mode shape for the 550-Hz mode. Rig configuration at
evaluation-point conditions.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 131
0.4
0.35
Unsteady Combustor Pressure Amplitude (psi)
0.3
0.25
0.2
0.15
0.1
0.05
0
0 500 1000 1500 2000 2500
Frequency (Hz)
Fig. 6.15 Measured power spectrum of unsteady combustor pressure at x = 1.9 in.
for evaluation-point operating conditions, showing resonance at 566 Hz, with an am-
plitude of 0.39 psi (0.78 psi p- p).
plenum lengths did not indicate significant changes in the resonant frequency.
Comparison of the engine and rig acoustic analyses (Fig. 6.3 vs. Fig. 6.13 and
Fig. 6.4 vs. Fig. 6.14) showed good agreement between the acoustic response of
the engine and the rig as designed.
Based on the results of these analyses and the stated design principles, the
experimental test rig design was finalized. It was fabricated and installed in a high-
pressure, high-temperature combustion test cell. The operating conditions of the
combustor could be completely described by the following parameters: diffuser
air pressure (P3), diffuser air temperature (T3), and combustor fuel–air ratio (f/a).
Values for each of these parameters were chosen to correspond to three different
engine-operating conditions. These are shown in Table 6.2. The fuel–air ratio
referred to is that estimated at the exit of the combustor and accounted for all of
the air flowing into the combustor through the air swirler, primary and dilution
holes, and liner- and bulkhead-cooling passages. It was not possible to vary the
test parameters independently because of the choked, fixed-area combustor exit.
PCB piezoelectric pressure transducers (P/N 124A21) were selected for this
experiment. They were capable of measuring pressure fluctuations at frequencies
between 0.5 Hz and 10 kHz at high-mean pressures. Satisfactory durability was
achieved by use of an integral water-cooled mounting fixture that maintained an ac-
ceptable temperature around the sensor. The liner-pressure sensors communicated
with the combustor through a 0.062-in. (1.6-mm)-diam, 0.83-in. (2.1-cm)-long
sensing tube. Nitrogen was used to purge the tube; the amount of purge flow rate
was negligible. The quarter-wave resonant frequency of the cavity within this tube
was far above the frequency range of interest in this experiment. Analog data were
low-pass filtered at 2 kHz and digitally sampled at 5 kHz by using a simultaneous
sample-hold data-acquisition system.
For the evaluation-point operating conditions, an instability was observed at a
frequency of 566 Hz (Fig. 6.15). The amplitude of this mode at these conditions
was ±0.39 psi (2.7 kPa). The unsteady pressure results presented here are from
the transducer located at 1.9 in. (4.7 cm) downstream of the combustor bulkhead.
The amplitude of the instability increased with increasing fuel–air ratio for fixed
P3 and T3. At higher fuel–air ratios, the overall rms pressure fluctuations were
dominated by this single tone. There was significant noise generated in the 100-
to 300-Hz range, although none of it was particularly coherent.
Figure 6.16 shows the spatial distribution of the unsteady pressure at three
locations within the combustor and one location in the diffuser region upstream.
3.5
Magnitude Phase 150
3
100
Amplitude/ Amplitude PLA1C1
Bulkhead
2.5
50
2
0
1.5
-50
Combustor
1 Exit
-100
0.5
-150
0
-2 0 2 4 6 8 10
X (in.)
Fig. 6.16 Measured distribution of 566-Hz mode, showing magnitude and phase ref-
erenced to pressure measurement PLA1C1 at x = 1.9 inches downstream of the com-
bustor bulkhead.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 133
0.45
0.3
0.25
0.2
0.15
0.1
0.05
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 6.17 Power spectra of unsteady combustor pressure at three operating conditions
corresponding to the conditions in Table 6.2, showing decreasing amplitude of the
566-Hz mode with decreasing power level.
Engine
Rig
0. 3
Amplitude (psi)
0. 2
0. 1
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 6.18 Comparison of engine and combustor rig pressure spectra for evaluation-
point operation.
unsteady pressure in the diffuser upstream of the combustor would be 180 deg out
of phase with the combustor pressure. The experimental results indicated a signif-
icant phase shift in the diffuser section, lagging the combustor pressure by about
90 deg at 566 Hz. This discrepancy is likely associated with the one-dimensional
limitations of the model. For example, it is expected that some level of coupling to
the outer-shroud passage would occur, which is also coupled to the combustor via
the air-mixing holes. Therefore, some transition of the phase from in phase with
the combustor outside the mixing holes to out of phase in the diffuser section is
expected in the three-dimensional problem. The result could be a phase relation
in the diffuser section between 0 and 180 deg.
Note that, because the Euler code is essentially an acoustic calculation, it is
fundamentally limited in its ability to calculate the amplitude of the pressure
oscillations without the addition of a combustion–acoustic coupling model. In
calculations for the engine using a constant relative forcing level, the Euler code
indicated that both the frequency and amplitude of the instability should increase
with increasing engine-power level. This trend was validated with engine data and
was also reproduced in the single-nozzle experiment (Figs. 6.1 and 6.17). It is also
encouraging that the damping mechanisms present in the calculations produced a
broad peak at 575 Hz, much like the peak seen in the experiment (Fig. 6.17).
Figure 6.18 shows a comparison between the fluctuating-pressure spectrum in
the engine and the single-nozzle combustor at comparable operating conditions.
Both data sets were acquired over 10 and were processed by using the same
techniques. The frequency of the target mode was reproduced within 12%. The
amplitude of this mode was matched within 3%. The spectral peak was signif-
icantly narrower in the engine data, indicating a more coherent instability. The
single-nozzle combustor also exhibited a higher overall level of noise in the sig-
nal, especially at frequencies below 350 Hz.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 135
A. Actuator Characterization
To demonstrate instability control, a suitable fuel actuator was necessary. The
specifications for the fuel valve, shown in Table 6.3, were derived from the rig
fuel flow requirements and prior experience. Several fuel actuator concepts were
investigated, and two were chosen for further development. A high-frequency fuel
valve built by the Georgia Institute of Technology was selected for experimental
testing because of the maturity of the concept (Fig. 6.19) The valve included both
a high-frequency flow modulation component and a mean flow control component
in a single device.
To provide a way to conduct steady-state and dynamic characterization of the
capabilities of the fuel valve, a characterization rig was developed and fabricated.
The rig was able to deliver up to 2 gal/min continuous water flow at up to 600 psia.
It was designed to provide an isolated test section for the valve to simulate the
valve/feed-line/injector (VFI) environment encountered in combustor rig testing.
An accumulator at the valve inlet provided isolation from the supply dynamics.
Downstream from the valve, an orifice simulated a fuel injector and a pressurized,
air-filled volume emulated the combustor.
Steady valve flow characterization was conducted first. This procedure consisted
primarily of mapping fluid flow vs valve displacement to quantify the valve mean
flow control authority. Steady valve flow characterization was also used later to op-
timize the valve position to maximize high-frequency fuel-modulation amplitude.
136 J. M. COHEN ET AL.
Fuel
Outlet
Auxiliary Fuel
Cooling Inlet
Air
1î
The valve exhibited a well-behaved, monotonic increase in flow as the valve open-
ing was increased. Also, once the valve reached a displacement of approximately
0.015 in. it was fully open and was no longer able to modulate the flow. To control
the mean flow and also to modulate the dynamic flow, the valve position had to be
maintained in the range between 0.005 and 0.015 in.
For the dynamic characterization of the valve, dynamic pressure transducers
were placed upstream and downstream of the valve, and downstream of the fuel
injector. Initially, a minimum feed-line length (just long enough to incorporate
the transducer) was used between the valve and the injector. Use of the minimum
feed-line length between the valve and injector allowed direct measurement of the
valve P frequency response while minimizing the interaction with the feed line.
A sinusoidal input signal of ±1 V was sent to the valve, and the pressure drop
across the valve was analyzed with respect to the input signal. Line lengths of 1 ft
and then 2 ft were inserted between the valve and the simulated injector orifice
to simulate the effect of realistic line lengths as would be encountered when lines
were installed on a combustor rig or on an engine.
Figure 6.20 shows the transfer function between valve command and valve
pressure drop. As can be seen from the transfer function, adding line length between
the valve and the fuel injector decreased the resonant frequency of the fuel system.
As line length was increased from 0 to 2 ft, the resonant frequency decreased
toward the 500- to 600-Hz combustor resonance frequency. Having this response
singularity at or near the controller frequency of interest should be avoided because
this proximity of frequency values can frustrate attempts to control the unsteady
fuel mass flow as required for active instability control. Thus, there is a maximum
installation fuel-line length between the fuel valve and the injector above which
the interaction between the fuel system and the combustor instability will become
extremely complicated.
Thus, the valve authority, that is, the level at which the valve is able to perturb
the fuel flow and thus the combustor pressure, was ultimately determined experi-
mentally in the combustor rig. The valve was set to a nominal steady opening based
on the characterization rig results. Open-loop, sinusoidal valve command voltage
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 137
0.2
500Hz resonance
0.15
DP23a/input, psi/volt
0.1
0.05
700Hz resonance
0
360
180
-180
Phase, deg
-360
-540
-720 0ft-1V
1ft-1V
-900
2ft-1V
-1080
0 500 1000 1500 2000 2500
Frequency, Hz
Fig. 6.20 Dynamic-valve response showing the transfer function between valve-
commanded voltage (input) and valve Delta-P (DP23) for three different feed-line
lengths.
variations were provided to the valve. The single-nozzle combustor rig was oper-
ated at conditions close to those in Table 6.2 that gave a ∼530-Hz combustion in-
stability. The frequency and amplitude of the valve command voltage were varied,
and the combustor pressure was monitored by using the combustor pressure sensor
1.9 in. downstream of the bulkhead. Representative results are shown in Fig. 6.21.
For a 300-Hz, ±2.5-V (maximum allowed) valve command, the combustor rig
dynamic pressure was shown to have a sharp response to the valve perturbations.
The pressure response was imposed on top of the combustion-instability-pressure
variations. Similar results are shown for a 600-Hz valve command. There was
some initial concern that, even if the valve was able to impose large fuel mass
flow variations, the prefilming features of the fuel injector would reduce actuator
138 J. M. COHEN ET AL.
a) 300 Hz 2.5
Valve Command Voltage
3
2 2
Amplitude, volts
Amplitude, volts
1
1.5
0
1
-1
0.5
-2
0 -3
Combustor Pressure
0.4 4
0.3 2
Amplitude, psi
Amplitude, psi
0.2 0
0.1 -2
0 -4
0 200 400 600 800 1000 0 0.01 0.02 0.03 0.04
Frequency, Hz Time, sec
b) 600 Hz 2.5
Valve Command Voltage
3
2 2
Amplitude, volts
Amplitude, volts
1
1.5
0
1
-1
0.5
-2
0 -3
Combustor Pressure
0.4 4
0.3 2
Amplitude, psi
Amplitude, psi
0.2 0
0.1 -2
0 -4
0 200 400 600 800 00 0 0.01 0.02 0.03 0.04
Frequency, Hz Time, sec
Instability Pressure
Acoustics
Pressure from +
+
Σ
Fuel Fuel lines, Injector Fuel Modulation Combustor Pressure
Flame
Valve & Combustion
Phase Shift
Controller
Filter
authority. However, these tests confirmed that sufficient authority (on the order of
the instability amplitude) was available.
Pressure from
instability
Boundary of effective
stability region
Boundary of restricted
control region
Fig. 6.23 ASPAC control approach finds a phase region that provides reduction in
the overall combustor pressure.
a slower, more gradual adaptation of the controller gain. Further details on the
ASPAC method can be found in Refs. 6.16 and 6.17.
The second control method was a model-based approach. The multiscale ex-
tended Kalman (MSEK) approach, like the first method, also sensed combus-
tor pressure. The MSEK method, shown in Fig. 6.24 combined a multi scale
(waveletlike) analysis and an extended Kalman filter observer to predict (model)
the time-delayed states of the thermoacoustic combustion pressure oscillations.
The commanded fuel modulation was calculated from a predictive (damper) ac-
tion based on the predicted states, and an adaptive, tone-suppression action based
on the multiscale estimation of the pressure oscillations and other transient dis-
turbances. The controller attempted to automatically adjust the gain and phase of
these actions to minimize timescale-averaged variances of the combustor pres-
sure. The controller operated at a sample frequency of 5 kHz. Further information
on the MSEK control approach is in Ref. 18. Both control methods were ini-
tially evaluated against reduced-order oscillator models of the combustor pressure
Upstream
injector
sensed
Sensed Time-Scale pressure
combustion Averaged
pressure Pressure
Multi-Scale Variance Fuel
Tones Analysis modulation
command
Parameter Upstream
Tuning Compensation
Phase Drift
Estimation
EK States Damper
Phase-Adjusted Predictor
Reconstruction
Suppression
ACC sim
0.5
0.4
Open Loop
PSI / Hz
0.3
Closed Loop
0.2
0.1
0
0 200 400 600 800 1000
Frequency, Hz
Fig. 6.25 Pressure amplitude spectra showing approximately 60% reduction in in-
stability amplitude predicted for model-based control method.
0.3 0.3
Amplitude, psi
0.25
Amplitude, psi
0.25
0.2 0.2
0.15 0.15
142
0.1 0.1
0.05 0.05
0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
Frequency,Hz Frequency,Hz
Fig. 6.26 Combustor pressure amplitude spectra for initial active combustion instability control test showing closed-loop instability
suppression for both control methods.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 143
authority. The closed-loop controller was then engaged. Two sets of tests were
run with both controllers being evaluated during each test. During the first test, a
reduction in instability amplitude was observed for both control methods. However,
for both control methods, low-frequency (<30 Hz) oscillations were seen in the
combustor pressure frequency spectra. It was suspected that these low-frequency
oscillations were caused by interactions between the instability controller and the
valve mean flow control or some other low-frequency phenomena.
For the second set of tests, additional filtering was added to the controllers to
reduce this interaction. The results, shown in Fig. 6.26, demonstrated that this
approach was effective in greatly reducing the low-frequency oscillations. As
can be seen, both control methods were able to identify the instability frequency
and reduce the amplitude by about 30%. This reduction was accomplished with-
out inducing secondary peaks at adjacent frequencies as has been seen in other
combustion-instability-control studies.12,20,21,22 Chapter 23 of this collection in-
cludes a detailed discussion of several factors that will affect the controller’s ability
to attenuate an instability.
VII. Conclusion
In summary, aggressive performance and emission goals for aircraft gas-turbine
combustors are leading to the development of combustors with increased suscep-
tibility to combustion instabilities. In this effort, a process has been developed
that addresses the difficulty in predicting and mitigating these combustion insta-
bilities. This process has been applied to an example problem consisting of an
engine-observed combustion instability.
Acoustic analysis of the engine combustor geometry using quasi-one-
dimensional Euler calculations was conducted to help explain the engine-instability
behavior. This analysis showed a susceptibility of the combustion chamber to
acoustic resonance at the observed 525-Hz instability frequency. Based on the
limited engine-dynamic data available, the analysis results seem to indicate a lon-
gitudinal acoustic mode frequency and mode shape which are believed to be that
exhibited by the engine.
Dynamic-response experiments were conducted with the engine fuel injector–
air swirler. Acoustic impedance tests showed that no specific acoustic parameters
relating to the geometry of this swirler would cause preferential amplification (and
instability coupling) of pressure fluctuations in the 400- to 600-Hz range. Fluid
dynamic forcing tests showed, however, that the air velocity near the filmer of
the swirler was highly sensitive over this range. The swirler exhibited preferen-
tial amplification ranging from 300 to 500 Hz, peaking at approximately 350 Hz.
The frequency of the observed fluid mechanic unsteadiness scaled directly with
velocity, indicating that this unsteadiness may be related to periodic vortex shed-
ding or other phenomena that are velocity sensitive. A novel optical technique
was devised to monitor the fuel flow fluctuations in an acoustically excited spray,
using the fluorescence of additives and aromatic constituents in the fuel. The tech-
nique was applied to four aeroengine fuel injectors and identified strong acoustic
coupling between the pulsed airflow and the fuel spray mass flow in the 350- to
650-Hz range. Modifications to the fuel injector–air swirler design based on the
dynamic spray and fluid mechanic response measurements resulted in significant
144 J. M. COHEN ET AL.
Acknowledgments
Numerous outstanding engineers and technicians made very significant contri-
butions to the concept and execution of the projects we have described. At NASA,
they are Clarence Chang, Joseph Saus, Daniel Paxson, George Kopasakis, Dzu Le,
Daniel Vrnak, Kevin Breisacher, and James May. At Pratt and Whitney they are
Jeffery Lovett, Michael Ondas, Saumil Shah, Donald Kendrick, and Saadat Syed.
At United Technologies Research Center they are Brian Wake, Thomas rosfjord,
John McVey, Torger Anderson, Randy Hibshman, Karen Teerlinck, Michael Carey,
Jeffrey Walker, and Walter Borst. At Georgia Institute of Technology they are
Yedidia Neumeier, Eugene Lubarsky, Ben Zinn.
References
1
Wake, B. E., Choi, D., and Hendricks, G. J., “Numerical Investigation of Pre-Mixed
Step-Combustor Instabilities,” AIAA Paper 96-0816, Jan. 1996.
2
Anderson, T. J., Kendrick, D. W., and Cohen, J. M., “Measurement of Spray/Acoustic
Coupling in Gas Turbine Fuel Injectors,” AIAA Paper 98-0718, Jan. 1998.
3
Cohen, J. M., and Hibshman, J. R., “An Experimental Study of Combustor Air Swirler
Acoustic and Fluid Dynamic Sensitivities,” Propulsion Engineering Research Center 9th
Annual Symposium on Propulsion, 1–2 Oct. 1997.
4
Kinsler, L. E., Frey, A. R., Coppens, A. B., and Sanders, J. V., Fundamentals of Acoustics,
3rd ed., Wiley, New York, 1982, pp. 230–243.
5
Munjal, M. L., Acoustics of Ducts and Mufflers with Application to Exhaust and Venti-
lation System Design, Wiley, New York, 1987, pp. 201–207.
6
Beranek, L. L., Acoustics, Acoustical Society of America, Melville, NY, 1993, pp. 123–
128.
7
Bendat, J. S., and Piersol, A. G., Random Data, 2nd ed., Wiley, New York, 1986,
p. 172.
CHARACTERIZATION OF AEROENGINE COMBUSTION INSTABILITY 145
8
Arnold, A., Dinkelacker, F., Heitzmann, T., Monkhouse, P., Schafer, M., Sick, V., and
Wolfrum, J., “DI Diesel Engine Combustion Visualized by Combined Laser Techniques,”
Twenty-fourth Symposium (International) on Combustion, The Combustion Institute, Pitts-
burgh PA, 1992, pp. 1605–1609.
9
Castleman, R. A., “The Mechanism of the Atomisation of Liquids,” Journal of Research
of the National Bureau of Standards, Vol 6, No. 281, 1931, pp. 369–376.
10
Hopfinger, E., and Lasheras, J., “Breakup of a Water Jet in High Velocity Co-flowing
Air,” 6th International Symposium on Liquid Atomization, Spray Systems, 1994, Institute
for Liquid Atomization and Spray Systems, Irvine, CA, pp. 110–117.
11
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., “Active Control of
Combustion Instability in a Liquid-Fueled Low-NOx Combustor,” Journal of Engineering
for Gas Turbines and Power, Vol. 121, No. 2, April 1999, pp. 281–284.
12
Hibshman, J. R., Cohen, J. M., Banaszuk, A., Anderson, T. J., and Alholm, H. A., “Active
Control of Combustion Instability in a Liquid-Fueled Sector Combustor,” American Society
of Mechanical Engineers, Paper 99-GT-215, June 1999.
13
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Combustion Driven
Oscillations by Equivalence Ratio Modulations,” American Society of Mechanical Engi-
neers, Paper 99-GT-118, June 1999.
14
Peracchio, A. A., Rosfjord, T., McVey, J., Anderson, T., Banaszuk, A., Cohen, J.,
Hibshman, J., Jacobson, C., Khibnik, A., Proscia, W., and Rey, N., “Active Control for
Marine and Land-Based Aeroderivative Gas Turbine Engines,” Vol. 1, Defense Advanced
Research Projects Agency Final Contractor Rept., United Technologies Research Center
Rept. 98-16, December 1998.
15
Hoffman, S., Weber, G., Judith, H., Herrmann, J., and Orthmann, J., “Application of
Active Combustion Instability Control To Siemens Heavy Duty Gas Turbines,” Symposium
of the AVT Panel on Gas Turbine Engine Combustion, Emissions and Alternative Fuels,
RTO-MP-14, AGARD, NATO Research & Technology Organization, Oct. 1998.
16
Kopasakis, G., and DeLaat, J., “Adaptive Instability Suppression Controls in a Liquid-
Fueled Combustor,” 38th Joint Propulsion Conference and Exhibit, AIAA Paper 2002-4075,
NASA TM-2002-21805, July 2002.
17
Kopasakis, G., “High-Frequency Instability Suppression Controls in a Liquid-Fueled
Combustor,” 39th Joint Propulsion Conference and Exhibit, AIAA Paper 2003-1458, July
2003.
18
Le, D., DeLaat, J., and Chang, C., “Control of Thermo-Acoustic Instabilities: The
Multi-Scale Extended Kalman Approach,” 39th Joint Propulsion Conference and Exhibit,
AIAA Paper 2003–4934, July 2003.
19
Paxson, D., “A Sectored-One-Dimenstional Model for Simulating Combustion Insta-
bilities in Premix Combustors,” 38th Aerospace Sciences Meeting & Exhibit, AIAA Paper
2000-0313, NASA TM-1999-209771, Jan. 2000.
20
Murugappan, S., Acharya, S., Gutmark, E., and Messina, T., “Characteristics and Con-
trol of Combustion Instabilities in a Swirl-Stabilized Spray Combustor,” 35th Joint Propul-
sion Conference and Exhibit, AIAA Paper 99-31259, June 1999.
21
McManus, K. R., Magill, J. C., Miller, M. F., and Allen, M. G., “Closed-Loop System
for Stability Control In Gas Turbine Combustors,” AIAA Paper 97-0463, Jan. 1997.
22
Barooah, P., Anderson, T. J., and Cohen, J. M., “Active Combustion Instability Control
with Spinning Valve Actuator,” American Society of Mechanical Engineers, Paper GT-
2002-30042, June 2002.
Chapter 7
I. Introduction
Copyright 2005
c by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Turbine Technology Engineer.
† Director of Turbine Maintenance.
147
148 J. B. SEWELL AND P. A. SOBIESKI
failures occur. Cracking of combustion parts and other component failures have
been known to liberate pieces and introduce significant downstream damage. Sec-
tion III describes part damage encountered in Calpine power plants because of
combustion-instability problems. Second, CDM can be used as a general com-
bustor health monitoring tool; part failures, flow blockages, fuel–air distribution
problems, etc., can often be detected from the CDM system by recognizing changes
in the dynamic-spectrum signatures. Section IV describes several case studies in
which the CDM system is used to detect combustor problems in this way.
A. Pressure Instrumentation
In general, the instrumentation configuration is driven by the temperature limita-
tions of the transducers. Because of the lack of suitable high-temperature pressure
instrumentation, the sensor is always physically placed away from the flame. This
necessarily introduces some transfer function between the actual pressure in the
combustion chamber and that measured at the transducer. Depending on the stand-
off tube configuration and sensor location, amplitude roll off or attenuation can
become significant.
Either of the following configurations were used for the data discussed in the
following sections. In Siemens–Westinghouse engines, a 3/8-in. waveguide or tube
is mounted flush to the combustion liner, providing acoustic access to a sampling
point in the primary zone of the combustor (Fig. 7.1). The waveguide passes
through a pressure seal in the engine casing at the combustor “top hat,” through
an isolation valve, and to the dynamic-pressure sensor. For this configuration the
dynamic-pressure sensor is located about 12–14 in. from the combustion zone. The
waveguide further extends to an attenuation tube that consists of a 3/4-in. tube filled
with 1/16-in. capillary tubes that dampens the signal and minimizes reflections.
Because of condensation in the tubing, it is periodically purged with dry nitrogen.
Figure 7.2 shows a general description and layout of the system. The sensor is a
fast-response piezoelectric crystal-type dynamic-pressure transducer. The sensor is
hermetically sealed, electrically isolated, and vibration compensated. In multi-can
combustors, one of these sensors has acoustic access to each combustor can.
In GE engines, a 1/4-in. tubing is run from an access point on the combustion
liner wall to an instrumentation/purge panel located outside the turbine enclosure
(Fig. 7.3). The typical distance from the acoustic-access point and the dynamic-
pressure sensor is not more than 27 ft. Damping or attenuation is done by traditional
infinite coils, that consist of 125 ft of 1/4-in. tubing. The system also uses dry
COMBUSTION INSTABILITIES MONITORING: CALPINE EXPERIENCE 149
Waveguide
Acoustic Access Point
Combustor
Top Hat
Fig. 7.1 Acoustic access to primary zone of the combustor for Siemens–Westinghouse
F class engines.
Attenuation Tube
Piezoelectric transducer
Charge
Converter
Nitrogen
Combustor Purge
Top Hat
Analyzer
Fig. 7.3 Acoustic access to primary zone of the combustor for General Electric
engines.
alarm thresholds are determined empirically; all that is really known from this type
of data is that large amplitudes are damaging and low amplitudes are less damaging.
engines. A key indicator of a tuning issue in the dynamic spectrum is that LFD is
present in all combustors. If it were a localized issue, such as a part failure with one
combustor, the symptom would be observed in the localized area. In the spring,
when ambient temperatures are on the rise, the opposite symptoms occur.
Midfrequency dynamics (MFD) are generally observed in the 100- to 250-Hz
frequency range. For example, tones occur at 130–150 Hz in the GE 7FA and at
120–140 Hz in the S/W 501F. In the industry, these oscillations are often referred to
as “hot tones,” because their amplitude often increases with firing temperature and
engine power output. For S/W engines the 135- to 145-Hz hot tone is dominant
at higher loads, whereas a 315-Hz mode can become excited depending on the
degree of gas heating and fuel splits. There are situations in which both hot tones
and cold tones can be present at base load depending on the fuel splits. At part
loads of < 85% of base load these engines are relatively stable as long as a high
degree of pilot fuel is contributing in the splits. Both hot tones and cold tones
are rarely present at the same time unless a hardware issue exists. In contrast,
GE7FA engines burn quite a bit leaner throughout the load range. These engines
can see both hot tones and cold tones throughout the load range depending on the
fuel splits. In addition, it is possible to excite multiple dominant frequencies by
adjusting fuel splits.
Finally, high-frequency dynamics (HFD), sometimes called screech, is occa-
sionally observed in industrial gas turbines. Although not very common, HFD is
particularly destructive to engine hardware; parts have been observed to fail in as
little as a few minutes during high-amplitude HFD. HFD at ∼1600 Hz and above
are observed in the S/W 501F and 501G and at ∼260 Hz or above in the GE F class.
Dynamics monitoring has traditionally been used for engine tuning. Tuning DLN
combustors involves changing the amount of fuel delivery between the various
fuel stages and in some cases controlling the amount of air by adjusting the inlet
guide vane (IGV) angles. If the engine has a closed-loop exhaust temperature
control, the IGVs are modulated by the control system to either maintain exhaust
temperature based on a defined curve or to maintain a specified margin lower than
the actual exhaust temperature limit or the so-called firing curve. Open-loop control
modulates the IGVs based on the load in either power output or a percentage of
full load. For either control scheme, the tuner can make adjustments to the IGV
schedule throughout the load range. Care must be taken not to run into a stall or
surge, which are not typically published or provided by some OEMs.
Changes are made on line and are done in small increments. Good practice will
include mapping the engine out to optimize emissions and stability. For example,
at a given load, the tuner will start with one fuel stage and increase the fraction by
enough of a percentage to make a noticeable change in dynamics and emissions.
If the changes improve the current condition, this process is continued. Next, the
original fraction is restored and compared with the original base line. Then, the
fuel fraction of that same stage is decreased until noticeable changes occur in
emissions and dynamics. This procedure will expose the optimum envelope for
that stage and load.
This procedure is done stage by stage, sometimes adjusting airflow throughout
the load range until the stability has been mapped out and an acceptable set of
fuel fractions or splits is obtained. Some balancing or iterations will be required
because of coupling of the stages.
COMBUSTION INSTABILITIES MONITORING: CALPINE EXPERIENCE 153
Many combinations of fuel splits for the same load range can meet emissions
and stability requirements. The fuel-split changes are typically used to control
NOx and stability. The IGVs have a significant impact on part-load CO emissions,
whereas a 0.5- to 1-deg change can change CO emissions by as much as 100+ ppm.
Compared with DLN combustors, conventional systems are much simpler to
tune. Conventional systems have only one fuel stage, and water or steam injection is
used for NOx abatement. For water injection, the water is injected at the fuel nozzle
directly into the primary zone. For steam injection, the steam can be introduced
at the nozzle in the primary zone, downstream of the primary zone, or in the
combustor shell before the combustor. They usually run at a steam-to-fuel ratio of
1.2 to 1.5.
In general, it is not possible to tune in anticipation of upcoming ambient changes
such as temperature, humidity, or barometric pressure. In some cases, a unit can
experience extremes in temperature and humidity through various load ranges
several times a day. For example, increases by 50–100% in dynamics amplitude
have been observed as ambient-temperature drops and the engine output increases.
Without continuous monitoring, the only way operators know if there is a tuning
issue, is when dynamics become audible and the engine begins to rumble or trips.4
Other indications of tuning troubles include excessive emissions and unstable
operation such as load swings. In such instances the engine will be load limited
to prevent harmful operation in the range of instability until the engine can be
retuned. Keeping the engine tuned to meet the changes in ambient and special
operations extends parts life and can significantly improve maintenance cycles.2,5
positions. In contrast, a tuning issue will be visible in most, if not all, combustors
and can be further isolated by using plant supervisory data.
However, it is not always possible to determine which part has failed until after
an inspection. By using combustor-dynamics data along with plant supervisory
data, the objective for operations is to quickly determine one of the following; run
the engine normally, shut it down, or operate at reduced load until parts, manpower,
and equipment can be brought to the site.
A. Flow Obstructions
In the situation described here, an engine experienced LFD at about 25 Hz as it
approached base load, which would periodically exceed the alarm level. Figure 7.7
shows the combustor exhibiting LFD (0.837 psi at 23.75 Hz). The solid lines de-
pict the amplitude threshold value programmed in the software. Alarm threshold
in the low-frequency zone (10–50 Hz) was set for 0.5 psi (0 peak). Note that there
is actually a higher amplitude MFD in the 120- and 140-Hz range. However, as
can be seen, this mode’s amplitude is below the alarm threshold. Examination
of the data from all positions indicated that the problem was only experienced
in a single combustor can; the amplitude in other cans always remained below
the threshold value. Further investigation of the incidents with the plant historian
showed at one of the same instants that the dynamics spiked, a flashback thermo-
couple in the same combustor increased in temperature by 20 to 60◦ F, as shown
in Fig. 7.8. This figure shows combustor temperatures and pressure amplitude as
the engine is loaded, holding base load, and subsequently unloaded on operator
control.
The flashback thermocouple (Tfb ) is located in a high-velocity region of the
combustor, downstream of the premixers, and before the primary zone of the
combustor, which is in the same plane as the combustor-dynamics access point.
156 J. B. SEWELL AND P. A. SOBIESKI
Alarm
Threshold
Fig. 7.7 Dynamic spectrum with 23.75-Hz spike in alarm, amplitude is pounds per
square inch (0 peak).
Fig. 7.9 Borescope pictures of combustor premixer showing flow obstructions in swirl
vanes.
This temperature rise by itself would not have triggered an alarm, because typical
alarm limit set points for these engines are about 200◦ F. Also the indication has to
stay above threshold for 30 s before the engine will auto unload. In this example,
there were no changes in emissions, exhaust, or blade-path temperatures. Without
the combustor dynamics alarm, there would have been no control indication that
something was wrong.
This signature has been seen several times in our fleet and is an indication
of either lean blowout or a flow disturbance in the combustor. Findings from
inspections include lock wire and other debris lodged in the swirlers, broken swirler
pins, and holes burned in the sides of the combustors. Two examples are shown
in Fig. 7.9. In one case a linkage assembly for a combustor bypass valve had
been installed one spline off, roughly 15% more open than the other valves. In
this particular case the airflow in the subject combustor was reduced, slowing
down the air velocity, and the conditions were favorable for true flashback. Note
that an increased-temperature indication is not always a real flashback. In the
preceding case, it was caused by a localized disturbance or recirculation caused
by an obstruction in the airflow at the premixers.
B. Pilot-Nozzle Cracking
In this example, the first indication that the operator received was an alarm
in the 100- to-500-Hz band on combustor 16. This alarm indication can be seen
in Fig. 7.10, which plots the pressure spectra for combustors 11–16. The alarm
indication showed a distinct spike at 156 Hz, which is slightly outside the nor-
mal 125- to 135-Hz frequency range of the system. Looking closer, there is
some 156-Hz contribution in combustor 15 and, although not shown, in com-
bustor 1 (which is adjacent to position 16). However, the 156 Hz is not visible in
combustors 11 through 14. The significance of this detail is that combustors are
acoustically coupled by means of cross-flame tubes and noise can be heard and
detected in the adjoining baskets. This observation is a very important in any part-
failure signature, rather than instrumentation or tuning, to distinguish a localized
event.
158 J. B. SEWELL AND P. A. SOBIESKI
Fig. 7.10 Initial pilot-nozzle failure spectrum without peak hold, amplitude in
position 16 is above threshold, and the adjacent combustor exhibits the same
frequency.
The nozzle failed on the hot side at which the flange bolts onto the engine.
This area is subject to a cantilever motion and stresses. The crack initiated at the
base of the hot-side cantilever. Some studies have shown that the failure mode
was at 125 Hz, which is very close to the dominant operational range of this
combustor. Such pilot-nozzle failures are caused by high-vibratory motion during
engine operation. They have occurred on two separate styles of pilot nozzles, one
that has a fillet weld at hot-side–cold-side flange interface and one that has a butt
weld at the hot-side–cold-side flange interface. Both designs have had similar
failures, which were caused for the same reason.
Fig. 7.12 Borescope of a pilot nozzle lodged between row 1 turbine vanes.
160 J. B. SEWELL AND P. A. SOBIESKI
Total outage time for this case was about 12 h. The pilot nozzle was replaced
and the engine was returned to service.
C. Transition-Piece Failure
In this combustor, the typical dominant frequency is about 135 to 145 Hz, which
can be seen in all the combustors (Fig. 7.13). Note, however, an additional spike at
225 Hz in combustors 4, 5, and 6. This frequency has higher amplitudes in position
5 than the adjoining combustors 4 and 6, but it is clearly more obvious than the
pilot-nozzle-failure example.
Fig. 7.13 Early indication of transition piece failure (combustor 5). Lines indicate
peak hold and live trace: Combustors 1 through 4. (Continued)
COMBUSTION INSTABILITIES MONITORING: CALPINE EXPERIENCE 161
225 Hz signature of
combustor with cracked
transition piece
Fig. 7.13 (Continued) Early indication of transition piece failure (combustor 5). Lines
indicate peak hold and live trace: Combustors 5 through 8.
At this point, it is difficult to say how big the crack in the transition piece was. As
time progressed, this tone grew in amplitude, even when the load was reduced from
100% power. Shortly after this spectral snapshot the engine tripped on high blade-
path variance, this time because of a very-low-temperature reading. As the crack
propagates on the transition panel, it allows compressor discharge air to bypass the
head end of the combustor, thus cooling the exit of the damaged combustor and
showing up in the exhaust as a cold thermocouple or high-blade-path variance.
Figure 7.14 shows the as-found condition of the transition. A typical transition
failure of this type is initiated by low-cycle fatigue followed by high-cycle-fatigue
propagation. The combination of high steady-state stresses coupled with high-
dynamic stresses can cause quick and damaging failures.
162 J. B. SEWELL AND P. A. SOBIESKI
References
1
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, 2001, pp. 182–189.
2
Stuttaford, P., Martling, V., Green, A., and Lieuwen, T., “Combustion Noise Measure-
ment System for Low Emissions Combustor Performance Optimization and Health Moni-
toring,” American Society of Mechanical Engineers, Paper GT2003-38255, 2003.
3
Richards, G., and Straub, D., “Passive Control of Combustion Dynamics in Stationary
Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 795–810.
4
Mongia, C., Held, T., Hsiao, G., and Pandalai, R., “Challenges and Progress in Con-
trolling Dynamics in Gas Turbine Combustors,” Journal of Propulsion and Power, Vol. 19,
No. 5, 2003, pp. 822–829.
5
Hobson, D., Fackrell, J., and Hewitt, G., “Combustion Instabilities in Industrial Gas
Turbines-Measurements on Operating Plant and Thermoacoustic Modeling,” Journal of
Engineering for Gas Turbines and Power, Vol. 122, 2000, pp. 420–428.
6
McManus, K., and Lieuwen, T., “That Elusive Hum,” Mechanical Engineering Power,
June 2002.
Chapter 8
I. Introduction
Copyright c 2005 by E.ON UK. Published by the American Institute of Aeronautics and Astronau-
tics, Inc., with permission.
∗ Power Technology.
163
164 C. J. GOY ET AL.
continuously monitor the health of the gas turbine and its combustion system to
obtain the earliest possible indication of deterioration or failure. Early detection
of abnormal operation enables commercially appropriate action to be planned,
whether this involves a machine inspection or the reoptimization of the combus-
tion system.
If just one major combustion failure can be prevented by regular health mon-
itoring, the resulting financial savings will far exceed the costs of installing the
required equipment.
Power Station
Power Technology
Specialist Advice
On Line
Combustion
Dynamic data
On Line
Vibration
Data exported by
Wide Area
On Line Network
EDMS Specialist PC
Evaluation
and Optimisation
Plant data On Line
Performance
On Line
Pyrometry
alarms can be configured to provide greater protection during the period of oper-
ation until the shutdown can be arranged.
Combustion components were inspected during overhauls and detailed infor-
mation about their condition has been recorded. On the basis of these inspections,
correlations have been developed between individual modes of instability and dam-
age to specific combustion components. This information has been fed back into
the daily screening activity to focus on the most damaging modes and to fine-tune
alarm criteria.
V. Case Studies
Once a machine’s operating behavior is characterized, deviations from the nor-
mal dynamic response can quickly be identified. This characterization must take
place over an extended period to take account of variations caused by ambient
conditions and operating regimes. Subsequently discussed are examples of this
sensitivity in a heavy industrial gas turbine with can-annular combustion system
that is operating in a combined cycle application for power generation.
Combustion-dynamics monitoring also enables investigations of complete fail-
ures to be undertaken. The examples given here illustrate how this experience has
been used to avoid the recurrence of failure modes.
12.00 400
Low f level Mid f level O Ambient T X Unit Load
Dynamic Pressure / Temperature
X XX X X XX X X X X X X X X X XX X 350
X X XXX XX O X X OX X XX
10.00
X X X X X X O X X XO X O
X X O O O OO X
XX X X O O O XX X X X
X XX O 300
X XX O O O X
X X
8.00
O O OO
O
Unit Load
250
O O O
O O
6.00 O 200
O O
O
150
4.00
O O
0.00 0
00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00 03:00 06:00 09:00
Time
Fig. 8.4 Combustion dynamics as a function of load and ambient temperature.
to the range of conditions during the 33 low periods shown. However, the higher-
frequency mode (dashed line) is excited during part-load operation, and especially
at times of low ambient temperature. The amplitude of the peak instability occur-
ring between 0300 and 0600 hrs on the first day, as shown in Fig. 8.4, is more than
three times greater than the peak level recorded between 0300 and 0600 hrs on the
second day. The operating load is identical in the two cases; the only difference is
that the ambient temperature is significantly lower on the first day. The difference
in response of the two dynamic modes is a result of the impact of varying ambient
conditions on the coupling between the premixed flame and the combustion cham-
ber acoustics, as has been investigated elsewhere.9 This sensitivity must, therefore,
be kept in mind if combustion optimization is carried out on a warm day because
recorded dynamics levels would be minimal throughout an engine’s operating
range under warm ambient conditions. A safety margin must be maintained from
damaging levels of all dynamic modes and over all anticipated ambient conditions
in which the machine is expected to operate. Continuous combustion-dynamics
monitoring throughout the year provides the necessary understanding of the re-
sponse of each individual instability mode to a change in ambient temperature,
such that the combustion system can be optimized for dynamically quiet operation
at all times.
210 3.0
Dynamic Pressure
190 2.5
GT Load
170 2.0
150 1.5
130 1.0
110 0.5
90 0.0
load to modify the engine speed; therefore, the operation of the machine is never
steady. When operating in this much more transient, frequency-sensitive (FS)
mode (described as “FS mode = 1 ACTIVE” in Fig. 8.5), the dynamics levels
increase significantly, then gradually decay each time the machine returns to
steady-state operation.
Fig. 8.6 Damaged combustion liner (view looking downstream, burners removed).
Dynamic pressure
Hot tone level NOx 140
6
120
Dynamic Pressure
5
100
4
NOx
80
3
60
2
40
1 20
0 0
1-Dec 2-Dec 3-Dec 4-Dec 5-Dec 6-Dec 7-Dec 8-Dec 9-Dec 10-Dec
Fig. 8.7 NOx and dynamics levels in the week leading up to the combustion liner
failure. (Note that the periodic spikes of NOx emissions correspond to monitoring
equipment calibrations.)
170 C. J. GOY ET AL.
625 250
GT Load
615 150
610 100
605 50
600 0
Fig. 8.8 Combustor-related exhaust temperature during the combustion liner failure.
from about 20 mg/m3 to 180 mg/m3 . This change led to the decision to shut down
the unit. When the relevant liner was removed from the engine, it was found that a
crack had initially developed at a weld at the aft end, propagated around the weld,
then moved toward the forward end to form a flap of material that had become
partially detached (Fig. 8.6).
The ability to confirm, via reference to exhaust gas temperature data, the com-
bustor in which components are deteriorating is a valuable check. Mapping of
temperature variations back to the individual combustor at the root of the problem
Failure
(High Frequency Mode)
Dynamic Pressure
0.00
0 50 100 150
Days
Fig. 8.10 Dynamics data from failed combustor in the 6 months before failure. (The
high-frequency, radial acoustic mode is shown here.)
helps to reduce repair downtime. Further indications of liner crack growth can be
found in changes in emissions. Low NOx depends on good mixing and distribu-
tion of fuel and air, both of which can be affected by liner damage, resulting in
significant variations in NOx emissions.
Step Decrease
(Lower Frequency Mode) .
In Dynamics At
Dynamic Pressure
Time Of Failure
0.00
Fig. 8.11 Dynamics data from affected combustor in the month before failure.
EGT Spread
NOx
0 0
Fig. 8.13 NOx levels and EGT spread in the six months before failure.
Ambient Temperature
Dynamic Pressure
Time
Fig. 8.15 Impact of ambient temperature and gas turbine load on high-frequency
dynamics.
MONITORING COMBUSTION INSTABILITIES 175
X. Conclusion
On-line combustion-dynamics monitoring offers significant commercial advan-
tages to power generators by providing a tool with which greater gas-turbine
reliability and availability can be achieved. The principal purpose of this activity
is to make operators aware of occasions when their gas turbines are operating
in a potentially damaging way. This aim has been realized. Although the cost of
installing monitoring equipment may be high, the potential savings resulting from
the increased flexibility of operation and the avoidance of catastrophic failure far
outweigh the cost. As a result, combustion monitoring has become a valuable
component of the integrated health-screening activity.
In addition, dynamics data have been used retrospectively to aid in failure in-
vestigations. An understanding of the effects of specific combustion modes on
individual components has also been developed. This is continuously revisited to
provide the most up-to-date understanding possible.
References
1
Lefebvre, A. H., “Gas Turbine Combustion,” Taylor & Francis, Washington, DC, 1998.
2
Lieuwen, T., and Yang, V., “Combustion Instability in LPP Combustors,” Journal of
Propulsion and Power, Vol. 19, No. 5, 2003.
3
Willis, J. D., and Moran, A. J., “Industrial RB211 DLE Gas Turbine Combustion Up-
date,” American Society of Mechanical Engineers, Paper 2000-GT-109, May 2000.
4
Scarinci, T., and Halpin, J. L., “Industrial Trent Combustor–Combustion Noise Char-
acteristics,” Journal of Engineering of Gas Turbines and Power, Vol. 122, April 2000,
pp. 280–286.
5
Berenbrink, P., and Hoffmann, S., “Suppression of Dynamic Combustion Instabilities
by Passive and Active Means,” American Society of Mechanical Engineers, Paper 2000-
GT-079, May 2000.
6
Pandalai, R., and Mongia, H., “Combustion Instability Characteristics of Industrial En-
gine Dry Low Emissions Combustion Systems,” AIAA Joint Propulsion Conference, AIAA,
Cleveland, OH, Paper AIAA 98-3379, July 1998.
7
Colechin, M., Rea, S., Goy, C., and James, S., “On-line Combustion Monitoring on Dry
Low NOx Industrial Gas Turbines,” Institution of Mechanical Engineers Seminar, London,
Jan. 2003.
8
Rea, S., James, S., Goy, C., and Colechin, M., “On-line Combustion Monitoring on Dry
Low NOx Industrial Gas Turbines,” Measurement Science and Technology, Vol. 14, 2003,
pp. 1123–1130.
9
Janus, M., Richards, G., Yip, M., and Robey, E., “Effects of Ambient Conditions and
Fuel Composition on Combustion Stability,” American Society of Mechanical Engineers,
Paper 97-GT-266, June 1997.
III. Fundamental Processes and Mechanisms
Chapter 9
I. Introduction
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Research Scientist, Laboratoire EM2C.
† Assistant Professor, ECP, Laboratoire EM2C.
‡ Senior Research Engineer, Laboratoire EM2C.
§ Professor, ECP and Institut Universitaire de France, Laboratoire EM2C, Fellow AIAA.
179
180 S. DUCRUIX ET AL.
Acoustics
Atomization/ Flame
vaporization/ wall
Upstream mixing interactions Down-
dynamics stream
dynamics
Feed line Injection Stabilization Heat
dynamics Entropy Exhaust
release waves impedance
Impedance conditions
conditions Mixing Flame/
Organized
vortex vortex
structures interactions
Acoustics
Fig. 9.1 Basic interactions leading to combustion instabilities (from Ref. 7).
d2 x dx
+ 2ζ ω0 + ω02 x(t − τ ) = 0 (9.1)
dt 2 dt
Expanding Eq. (9.1) in a Taylor series to first order yields
d2 x dx
2
+ ω0 (2ζ − ω0 τ ) + ω02 x(t) = 0 (9.2)
dt dt
The damping coefficient is negative if ω0 τ > 2ζ . If the delay τ is long enough
with respect to the period T = 2π/ω0 , the amplitude of any perturbation will grow
exponentially. More generally, combustion instability occurs when the natural
COMBUSTION INSTABILITY MECHANISMS 181
F+O Flame
P Plate
Flame Flame
d) P
Flame
Equivalence f) g)
ratio e) F+O Plane
perturbation acoustic
Plane Plane waves
acoustic acoustic
waves waves
Fig. 9.2 Elementary processes: a) Unsteady strained diffusion flame, b) flame roll up
in a vortex, c) premixed flame/vortex interaction, d) equivalence ratio perturbation
interacting with a premixed flame, e) acoustically modulated conical flame, f) acous-
tically modulated V-flame, and g) perturbed flame interacting with a plate (adapted
from Ref. 7).
process is exemplified in Fig. 9.3 (from Ref. 20), which gives an illustration of
Rayleigh’s criterion in a vortex-driven instability. Characteristic time delays consti-
tute a generic feature of combustion instability, and are introduced in the following
illustrations.
It is convenient to begin with a wave equation for reacting flows (Sec. II). This
wave equation for the pressure fluctuations in the system features an unsteady
heat-release source term. At this point, the problem is not completely solved,
because no simple expression exists that relates the heat-release fluctuations to
the acoustic variables (pressure, velocity). A classical representation based on the
(n − τ ) model is introduced, showing once again the importance of characteristic
time delays. This development indicates that an accurate description of the driving
path linking heat release to pressure fluctuations is crucial.
Heat release is then considered as a pressure source in Sec. III, in which three
different situations are analyzed. The first is that of vortex-driven fluctuations, in
which vortices interact with a flame producing a heat-release pulse. This mech-
anism has been studied in many laminar and turbulent configurations, because it
constitutes a powerful driving process. In the second case studied in this section,
self-sustained oscillations of a flame impinging on a plate produce an intense radi-
ation of sound. This elementary process typifies heat-release fluctuations resulting
from flame–wall interactions. One may also infer from this example that similar
processes may take place as a result of mutual flame interactions in the core of
the flow. This third process effectively produces rapid changes of flame surface
area and correspondingly large fluctuations in heat release. These processes are
demonstrated by well-controlled model-scale experiments, but these mechanisms
are generic and probably drive many of the instabilities observed in larger-scale
combustors.
Section IV deals with heat-release fluctuations driven by waves or flow pertur-
bations. Three situations are envisaged. In the first situation, a conical flame is
modulated by acoustic waves. It is shown that the response of the flame may be
represented by a transfer function, which can be used to describe the stability map
of the burner. Experimental measurements of this transfer function are compared
with analytical estimates and numerical results. In the second situation, inhomo-
geneities formed in the upstream flow impinge on a flame producing a fluctuation
in heat release. The time delay between injection and combustion is the key pa-
rameter in the process and it defines conditions of oscillation. In the third situation,
heat-release perturbations result from a time-variable strain rate. A low-pass filter
behavior of the flame is found in this configuration. These three examples typify
interactions that may take place in practical systems. Many other cases are treated
in this book and in the references listed at the end of the chapter.
mixture of N species24 and using various simplifications, one can derive a wave
equation for the logarithm of the pressure16 :
c2 d 1 d
∇ · ∇ln p − ln p = ∇ · (ρ −1 ∇ · τ )
γ dt γ dt
d 1 N N
− ∇ · λ∇T + − h k ẇk − ρYk c pk vk · ∇T
D
dt ρc p T k=1 k=1
d2
− (ln R) − ∇v : ∇v (9.3)
dt 2
where c designates the speed of sound; ρ, p, T , Yk , v, and vkD are the density,
pressure, temperature, species mass fractions, velocity, and diffusion velocity, re-
spectively. , c pk , γ , λ, and R designate the viscous dissipation function, specific
heats, specific heat ratio, heat conductivity, and gas constant. h k and ẇk are, re-
spectively, the specific enthalpies and rates of reaction.
In expressions similar to Eq. (9.3), the splitting of terms between the left- and
right-hand sides is somewhat arbitrary, because some of the terms in the right-
hand side describe features of the propagation of sound in the medium and should
then be included in the left-hand side. This point is discussed by Doak25 in the
context of aerodynamic sound and by Kotake26 in a study of combustion noise.
Nevertheless, it is useful to regard the terms appearing in the right-hand side of
Eq. (9.3) as the source terms generating the pressure waves in the reactive mixture.
In a turbulent reacting mixture, an order-of-magnitude analysis indicates that, in
low-speed combustors, the dominant source terms are associated with the chemical
heat-release fluctuations.26 Neglecting all other terms, one obtains
c2 d 1 d d 1 N
∇ · ∇ln p − ln p = dt ρc T h k ẇk (9.4)
γ dt γ dt p k=1
Considering low-speed reactive flows, the convective term in the material derivative
may be neglected d/dt ∼ ∂/∂t. Assuming, in addition, that the specific heat ratio
is constant, Eq. (9.4) becomes
∂ 2
∂ 1 N
∇ · c2 ∇ln p − 2 ln p = h k ẇk (9.5)
∂t ∂t ρcv T k=1
This equation is not linearized, and it can be used to describe finite amplitude waves.
However, in many circumstances, the wave amplitude is relatively weak, and lin-
earization is appropriate. The pressure is then expressed as a sum of a mean and
fluctuating components: p = p0 + p1 with p1 / p0 1. Then, ln p p1 / p0 and
Eq. (9.5) becomes
p1 ∂ 2 p1 ∂ 1 N
∇·c ∇2
− 2 = h k ẇk (9.6)
p0 ∂t p0 ∂t ρcv T k=1
COMBUSTION INSTABILITY MECHANISMS 185
In addition to Eq. (9.7), an expression is needed for the acoustic velocity. This
expression can be obtained by linearizing the momentum equation and neglecting
the viscous stresses. This yields
∂v1 1
= − ∇ p1 (9.8)
∂t ρ0
where v1 represents the velocity fluctuations. Equations (9.7) and (9.8) describe
the propagation and generation of small perturbations in the reactive mixture. As
already mentioned, the problem is not completely solved, because a third relation
between heat release ( h k ẇk ) and acoustic fluctuations ( p1 , v1 ) is necessary to
close the system. This last relation can be deduced from experiments, theories,
or simulations. Section II.B gives an example of an analytical model using the
so-called (n − τ ) model.
Considering again the source term corresponding to the nonsteady heat release,
one may assume for simplicity that the chemical change occurs by a single-step
reaction. Then, if
h ◦f designates the change of formation enthalpy per unit mass
of the mixture, and if ẇ represents the rate of reaction, the chemical source term
becomes (∂/∂t)(γ − 1)(−
h ◦f )ẇ. In most cases, the only time dependence in this
expression is a result of the nonsteady rate of reaction, and, as a consequence, the
acoustic source term associated with chemical reaction may be written in the form:
∂ Q 1m
(γ − 1) (9.9)
∂t
where Q 1m represents the nonsteady rate of heat release per unit mass of mixture.
The wave equation (9.7) and the source term (9.9) indicate that the pressure field
is driven by the nonsteady release of heat. A coupled motion can take place if this
last quantity is influenced by acoustic variables, pressure, or velocity.
Fig. 9.4 Sketch of the model compact flame geometry. The flame zone is thin
compared to the acoustic wavelength. Arrows A, B, C and D indicate acoustic
waves propagating in the system. Flame is assumed to be located at x = a (adapted
from Ref. 4).
where Q 1a represents the instantaneous heat-release rate per unit area. When the
flame is compact, the nonsteady release of heat determines the jump in acoustic
velocities.
The determination of Q 1a as a function of the perturbed motion is by no means
trivial. One has to relate the time-varying flow variables and the dynamic response
of the flame. One approach27–29 uses a time-lag hypothesis to express Q 1a in terms
of the time-delayed upstream velocity perturbation,
Q 1a
(γ − 1) = nv f (a− , t − τ ) (9.11)
ρ f c2f
where n is an interaction index, and τ represents a time lag. The heat-release term
is modeled as a function of an acoustic-wave variable alone. This approach is
clearly a simplified representation of more complex processes involving the flow,
turbulence dynamics and large-scale motions, flame interactions with neighboring
flames and walls, heat transfer at the boundaries, etc. Some of these processes are
described subsequently.
COMBUSTION INSTABILITY MECHANISMS 187
The value of the time lag τ relative to the frequency often defines ranges of
instability.30, 31 This value is recognized from the early work on rocket engine
instability29 (see also the review in Ref. 22) and from many recent studies. One
possible use of expressions like Eq. (9.11) is reduced modeling of active control.
As shown, for example, in Ref. 4, this modeling yields simple time-lag conditions
for instability development and control. In what follows, time lags will be analyzed
in various laboratory-scale situations.
A. Flame–Vortex Interactions
Flame–vortex interactions have been observed in many unstable combustion
systems. Two distinct mechanisms are usually involved. In the first, the flame
area is rapidly changing because of vortex roll up.20, 32 In the second, the vortex
interacts with a wall or another structure, which induces a sudden ignition of fresh
material.21
Vortex roll up often controls the mixing of fresh gases into the burning regions.
This roll up determines the nonsteady rate of conversion of reactants in the flow
and the amplitude of the pressure pulse resulting from the vortex burnout. When
the flame is rolled up, the surface area increases rapidly. The growth is limited
by flame shortening, which results from interactions of neighboring elements, and
consumption of the reactants entrained by the vortex. Such rapid variations of
flame surface correspond to the first mechanism. Flame–vortex dynamics have
been studied extensively (see Ref. 33 for a review). Much of the experimental
work has concerned toroidal or pairs of counter-rotating vortices running into a
traveling premixed flame34 or an established strained diffusion flame,35 which do
not quite correspond to situations of interest in combustion instability.
Observations of combustion oscillations indicate that vortex roll up takes place
while the flame develops. The vortex entrains fresh materials and hot products and
ignites at a later time, producing a pulse, which feeds energy in one of the resonant
modes of the combustor.36, 37 This process is more difficult to study experimentally
and is less well documented. Interactions between adjacent reactive vortices may
also take place, leading to formation of fine-grain turbulence.
188 S. DUCRUIX ET AL.
Many studies have also focused on the natural instabilities of wakes and jets.38–41
Indeed, the vortical structures that are involved in the flame–vortex interactions
are often naturally generated and shed at dump planes. When one of the natural
frequencies of the jet matches one of the acoustic resonance frequencies, that
is, when the characteristic times are close, the coupling between acoustics and
combustion is made easier. An alternative way to study these situations is to force
the flow by using driver units or pistons and to analyze the response sensitivity of
the jet or wake to this forcing.42–46 The conversion of energy between vortices and
acoustics is also of great importance and has been studied in nonreactive47, 48 and
reactive situations.33
One example of self-sustained oscillations controlled by vortices is reported in
Ref. 36. A multiple-inlet combustor is fed with a mixture of air and propane, and it
features a dump plane (Fig. 9.5). The low-frequency instability observed in this case
Fig. 9.5 Geometry of the multiple-flame-holder dump combustor studied in Ref. 36.
Spark-schlieren photograph of the central jet for the 530 Hz unstable regime.
COMBUSTION INSTABILITY MECHANISMS 189
is acoustically coupled and occurs at one eigenfrequency of the system. The flame
visualization of Fig. 9.5 clearly shows that the largest-amplitude oscillations are
vortex driven. The following processes are involved: 1) A vortex is shed at the dump
plane when the velocity perturbation is maximum (v1 −→ Ω1 in a driving path,
where Ω1 represents vorticity fluctuations). 2) The vortex is convected, accelerated,
and entrains hot gases from its surroundings. A combustion pulse is produced when
two adjacent vortices interact, creating a large amount of small-scale turbulence
and flame surface area. 3) The sudden heat release constitutes a source that feeds
energy into the perturbed acoustic motion. Self-sustained oscillations can only
occur when the processes are correctly phased, that is, when the convective time
lag is in a suitable range with respect to the period of the motion as shown in Fig. 9.6.
The self-sustained oscillations of a laminar V-flame interacting with vortices are
analyzed in Refs. 20, 36, and 49.
Collisions of reacting vortices with boundaries are less well covered but are
often observed in premixed devices.21, 50, 51 A mechanism of this type is featured
in Ref. 50. A vortex is shed from a single inlet into a dump combustor (Fig. 9.7,
from Ref. 51). The vortices are synchronized by one of the longitudinal modes of
the system. Figure 9.7 shows a typical vortex-shedding event and the heat-release
distribution at a later time. Reducing the height of the combustor enhances the
interaction between the vortex and the lateral boundary, which produces longer
axial burning regions and augments the overall straining of the vortex. Fast burning
of the fresh reactants entrained by the structure takes place when the vortex collides
with the wall. The general process of flame interactions with boundaries is detailed
in the next section, because it can constitute a source of heat-release fluctuations
even in the absence of vortex shedding.
In the first situation described in this section, the mechanism involves flow
perturbations producing vorticity, which results in rapid changes of flame area,
inducing a heat-release pulse. In the second case, the vorticity directly causes a
volumetric expansion, leading to the heat-release pulse. This may be represented
globally by the following expression:
1 −→ Q 1 −→ p1
Many other studies21, 37 have revealed the key role of vortex structures. Coming
back to the theoretical expressions of Sec. II, one clearly sees that a model giving Q 1
as a function of 1 would make it possible to solve the complete set of equations.
Because the process involves convection, ignition, and combustion delays, one
may try an (n − τ ) formulation. This, however, requires further analysis of the
elementary steps, leading from vortex shedding to vortex burning.
Fig. 9.7 Schlieren photograph of a vortical structure entering a dump plane combus-
tor and chemiluminescence image representing the heat release rate distribution at a
later time during the instability cycle (from Ref. 51).
In the driving path, surface-area fluctuations produce nonsteady heat release, which
induces acoustic pressure radiation. It may be represented schematically by
A1 −→ Q 1 −→ p1
zone 2
CH*
filter
zone 1
mixture of
gases
M0
loudspeaker
Fig. 9.8 Schematic view of the experimental setup used to study interactions of a
perturbed flame with a cooled wall. This configuration radiates an intense acoustic
field (from Ref. 53).
system is 10–20 dB higher than that emitted by a free flame submitted to the
same modulation, without the plate. The interaction of the flame with the plate
leads to rapid changes of the flame surface, which constitutes a major source of
sound in this situation. This well-controlled experiment typifies more complicated
situations in which the flame spreads in a chamber (as in a gas-turbine combustor)
and can produce pressure oscillations when impinging on the walls or on adjacent
flame sheets. When the phase is suitable, and when the gain exceeds the losses,
the oscillation may reach large amplitudes leading to instability.
It is also possible to observe self-sustained oscillations of a flame impinging
on a plate. The experimental setup is similar to that used in Ref. 52 but the driver
unit is removed (Fig. 9.8).53 A 10-mm-thick water-cooled plate, which can move
vertically, is placed above the cylindrical burner. An oscillation develops naturally
if both the driving and coupling paths are present in the system. For certain plate-
to-burner distances, intense emission of sound is observed. Figure 9.9 shows the
steady flame (a), when no sound emission is observed, and a complete cycle of
oscillation (b–e), when the instability is triggered. The flame front is undulated
by the perturbation, which is convected from the burner rim to the plate. These
visualizations are close to those obtained in the external modulation case described
previously. The sound emitted features many harmonics, with a fundamental fre-
quency at about 200 Hz.
The burner behaves like a Helmholtz resonator with a resonance frequency of
200 Hz. The resonant behavior of the system may be described analytically by
combining a model for the flame interaction with the plate and a representation of
the burner acoustics. The acoustic velocity v1 and pressure p1 at the burner exit
may be related by a second-order equation53 :
d2 v1 dv1 d p1
M 2
+R + kv1 = −S1 (9.12)
dt dt dt
COMBUSTION INSTABILITY MECHANISMS 193
Fig. 9.9 Different views of a flame interacting with a wall: a) Steady state; b–e)
Instantaneous images of the flame during an instability cycle (from Ref. 53).
194 S. DUCRUIX ET AL.
3 3
CH*
LDV 1
1
Micro M 3 d(I(CH*))/dt
0 0
-1 -1
0 5 10 15 20
Time (ms)
where R is the system damping, and k is the stiffness of the gas volume acting
as a restoring force on the effective mass of air M. According to Eq. (9.12), the
resonator is driven by external pressure fluctuations p1 at the burner outlet. To
pursue the analysis, it is necessary to model the driving process appearing in the
right-hand side of Eq. (9.12).
The formulation is based on the following considerations. First, the source term
on the right-hand side of Eq. (9.12) originates from rapid changes of the flame
surface and subsequent noise radiation. Noise is generated when large portions of
the flame collapse because of interaction with and quenching by the plate. The
pressure field radiated by a compact source of nonsteady heat release takes the
form55–57
ρ∞ ρ f dQ
p∞ (r, t) = −1 (9.13)
4πr ρb dt t−τa
In this equation, ρ∞ , ρ f , ρb are the densities in the far-field air, the fresh gas, and
the burned gas, respectively; τa is the time required by sound propagation over
a distance r from the sources to the detector. In gaseous premixed flames, the
far-field radiated pressure p∞ can be related to the time-retarded rate of change of
the flame-surface area A:
ρ∞ ρ f dA
p∞ (r, t) = − 1 SL (9.14)
4πr ρb dt t−τa
where SL is the laminar burning velocity. The fast rate of extinction of the flame
area at the cold boundary induces a significant acoustic pressure radiation, which
shown in Fig. 9.10, where p∞ is measured by a microphone. The time derivative
of the heat-release signal is shown at the bottom of the graph. This signal nearly
coincides with the pressure signal detected by the microphone.
COMBUSTION INSTABILITY MECHANISMS 195
where n characterizes the coupling between the surface fluctuations and the veloc-
ity perturbations, and τc is the time required by convection from the burner lip to
the plate. This (n − τ )-like formulation is supported by the detailed experiments
carried out in Ref. 52. Expressions (9.12) and (9.15) may be combined yielding a
second-order equation for the velocity fluctuations:
2
d2 v1 dv1 d v1
+ 2δ + ω0 v1 = −N
2
(9.16)
dt 2 dt dt 2 t−τ
a) b)
c) d)
(Fig. 9.11b). As the deformation travels along the flame front, the two branches
of the “M” are stretched in the vertical direction and get closer (Fig. 9.11c), up
to an instant in the cycle where the flame-surface area is maximum, and two
flame elements interact (Fig. 9.11d). The outcome of this mutual annihilation de-
pends on the spatial position of the first interaction. In some cases, pockets of
fresh reactants may be trapped in a torus, but in other cases this will not oc-
cur.59 For some operating conditions not shown here, up to two flame tori can be
produced.
During interaction of these flame elements, the shape of the reactive front under-
goes a strong alteration. As in the flame–plate situation, after the mutual interaction,
the flame quickly retrieves its initial shape at the beginning of the following cycle
(Fig. 9.11a). In this cycle, the short phase of flame-surface destruction produces a
faster rate of change of the flame-surface area than the longer phase of flame-surface
production by stretch. The same mechanism operates as in the flame–plate
COMBUSTION INSTABILITY MECHANISMS 197
A1 −→ Q 1 −→ p1
p1 −→ v1 −→ A1 −→ Q 1
Transfer functions proposed in Refs. 68–71 indicated that the flame usually
behaves like a low-pass filter, providing a qualitative representation of the flame
response. Laminar conical flames are considered by Blackshear,66 De Sœte,67 and,
more recently, Baillot et al.72 Further theoretical efforts by Fleifil et al.73 and a
combination of theoretical analysis and detailed measurements by Ducruix et al.74
have advanced the status of this basic problem. More recent work by Schuller et al.75
provides additional clues on flame response in the high-frequency range. Dowling76
uses a similar approach to derive a model for the low-frequency nonlinear response
of a ducted V-flame in a geometry close to that considered in an earlier work by
Marble and Candel.77
The modulation of a conical flame is now considered in further detail. Our
objective is to describe the unsteady rate of heat release as a function of acoustic
variables. A laminar premixed flame is anchored on a cylindrical burner, and it is
submitted to acoustic waves generated by a loudspeaker placed at the bottom of the
burner. The flame response is driven by the acoustic velocity, and the aim is to find
the transfer function between heat-release fluctuations and velocity modulations:
Q 1 (ω)/Q 0
F(ω) = (9.17)
v1 (ω)/v0
where ω is the angular frequency of the modulation. In the linear range, the transfer
function is a good representation of the relation leading from acoustic variables to
heat-release fluctuations. Nonlinear effects will not be considered here, but they
are examined in many references.76, 78, 79 The modulus of F gives the amplitude of
heat-release fluctuations as a function of velocity modulations, whereas its phase
characterizes the time lag existing between velocity and heat-release fluctuations.
A complete analysis of this problem can be found in Refs. 74 and 80. Selected
results are highlighted subsequently.
The burner consists of a converging nozzle, which is water cooled, and a 120-
mm-long cylindrical tube, placed upstream from the nozzle and containing various
grids and honeycombs to produce a laminar flow. The conical flame is stabilized on
a 22-mm-diam burner rim. A driver unit placed at the base of the burner generates
perturbations, which wrinkle the flame front. The shape of the perturbed flame
depends on the frequency and amplitude of modulation. The typical flame shapes
displayed in Fig. 9.12 are visualized with a four-color schlieren technique. The
use of modern diagnostic techniques [particle-imaging velocimetry (PIV), instan-
taneous visualizations using intensified cameras, etc.] has provided new informa-
tion concerning the geometry of the flame front, the local and global heat-release
rates,74 and the velocity field at the burner exhaust and in the flowfield.80 This has
allowed direct measurements of the flame-transfer function defined by Eq. (9.17).
These measurements can be compared with theoretical and numerical predictions.
An analytical transfer function can be derived by decomposing the flow in mean
and perturbed components. The geometry of the problem is sketched in Fig. 9.13.
A G equation is used to describe the flame position:
∂G
+ v · ∇G = −S D |∇G| (9.18)
∂t
COMBUSTION INSTABILITY MECHANISMS 199
40 40
30 30
y (mm)
y (mm)
20 20
10 10
0 0
-20 -10 0 10 20 -20 -10 0 10 20
x (mm) x (mm)
Fig. 9.12 Methane air conical flame modulated by longitudinal acoustic perturba-
tions. fe = 150.5 Hz, ω∗ 28, v̄ = 1.44 ms−1 , v /v̄ = 0.13, Φ = 1.05. Top: schlieren
images for two different instants. Bottom: corresponding numerical simulations
(adapted from Ref. 80).
y
v
burnt
u gases
αo
ds
n r fresh
L gases
ηo (r) η (r,t)
r
-R (a) (b) R
Fig. 9.13 Geometry of a) the conical flame in the steady situation and b) in the
perturbed case; from Ref. 74.
200 S. DUCRUIX ET AL.
∂η1 ∂η1
= SL cos α0 + v1 (9.19)
∂t ∂r
Q1 v1 2
= [(1 − cos ω∗ ) cos (ωt) + (ω∗ − sin ω∗ ) sin (ωt)] (9.21)
Q0 v0 ω∗2
and the transfer function is easily deduced therefrom. The resulting expression
depends on a reduced frequency ω∗ = ω R/(SL cos α0 ), where R is the burner
radius. Expression (9.21) may be used as a source term in wave equation (9.7),
providing a complete dynamical description of a system featuring an initially
conical flame.
The analytical flame response to acoustic modulations obtained in this way relies
on many simplifying assumptions. It was assumed that the perturbed velocity is
axial and uniform. Data obtained with PIV80 show that this assumption may be
acceptable for weakly wrinkled flames, with a small velocity radial component, that
is, in the low-frequency range (ω∗ < 2). In this case, the flame responds as if it were
globally stretched and compressed by the modulation while keeping an essentially
conical shape. In contrast, these assumptions are too strong for larger frequencies
to correctly represent the acoustic–flame interactions. In that range, the velocity
field convects structures with important gradients and a radial component exists
near the burner exhaust, which clearly shows that simplified low-order models
have a limited range of validity.
COMBUSTION INSTABILITY MECHANISMS 201
1.0
(Qrms/Q)/(vrms/v)
0.8
0.6
0.4
0.2
0
v=0.97m/s
v=1.22m/s
phase difference (rad)
v=1.70m/s
model A
4 model B
0
1 10 20 30
ω*
Fig. 9.14 Comparisons between calculations (solid line), analytical results (dashed
and solid line) and measurements (symbols) for the transfer function of a conical flame
(from Ref. 80).
p1 −→ φ1 −→ convection −→ Q 1
This mechanism is illustrated in Fig. 9.15. In a first step, a pressure oscillation arises
in the system, which will modify the fuel flow rate and change the equivalence
a)
b)
c)
d)
e)
f)
Fig. 9.15 Instability driven by equivalence ratio perturbations. Time traces of pres-
sures, equivalence ratios, and heat release in the flame (from Ref. 82).
COMBUSTION INSTABILITY MECHANISMS 203
ratio. Three time delays define the process. The first τi corresponds to a phase shift
between the pressure at the injector and fuel mass flow rate ṁ F1 . Oscillations in this
flow rate induce fluctuations in the equivalence ratio φ1 . An inhomogenous mixture
is then convected to the reaction zone with a delay τconv . The response of the flame
to the impinging fluctuations φ1 comes after a combustion delay τchem . Oscillations
will be sustained by this process if the pressure and heat-release fluctuations are
in phase (Rayleigh’s criterion), which is the case if the total delay is such that
T
τi + τconv + τchem = (2n − 1) (9.22)
2
where T is the period of the combustion instability, and n is an integer (n > 0).
In many cases, the dominant delay is associated with convection, and the last
condition becomes τconv (2n − 1)T /2.
One fundamental aspect of this process is the response of the flame to incom-
ing equivalence ratio perturbations.86 Another aspect that will also influence this
mechanism is the level of mixing taking place between the injector and the flame.87
If this mixing is efficient, the initial level of fluctuations will be diminished to a
great extent by reducing the fluctuation in heat release. Effects of inhomogeneities
are also examined in more detail in other parts of this book (see also Ref. 8).
d
= − β 2 (9.23)
dt
1 1
= 2 (0 cos ωt + ω sin ωt) (9.25)
0 0 + ω 2
204 S. DUCRUIX ET AL.
1 1
= cos ωt (9.26)
0 0
1 1
= sin ωt (9.27)
0 ω
This mechanism applies equally well to premixed and nonpremixed flames. This
type of interaction modulates the flame-surface density and can be represented
schematically by
p1 −→ v1 −→ flow −→ 1 −→ A1
The second type of interaction involves a direct effect on the reaction rate per unit
flame surface and is represented by
p1 −→ v1 −→ flow −→ 1 −→ ω̇1
This second effect is effective in the nonpremixed case, because the reaction rate
is directly related to the species gradients at the flame, which are fixed by the strain
rate. In the premixed case, the consumption rate is weakly influenced by the strain
rate, except near extinction conditions.
The flame response to strain rate has been extensively studied in turbulent com-
bustion.88, 89 Other studies deal with the response of flames to external strain-rate
modulations. The problem is envisaged experimentally90 and often treated by di-
rect calculations using time-dependent solutions of strained flames with complex
chemistry.91, 92 Analytical expressions of the flame response have also been de-
termined by using asymptotics.93 It was found that flames behave like low-pass
filters, when the perturbed strain-rate fluctuations do not exceed the extinction
value. Considering nonpremixed flames and assuming the infinitely fast chemistry
limit, the flame-transfer function is defined in the frequency domain as the ratio of
the relative reaction rate modulation to the relative strain-rate perturbation7 :
ṁ(ω) − ṁ 0 (ω) − 0
F(ω) = (9.28)
ṁ 0 0
1 1
F(ω) = (9.29)
2 1 + i(ω/20 )
COMBUSTION INSTABILITY MECHANISMS 205
a) b)
c) d)
e) f)
Fig. 9.16 DNS computations of the mutual flame annihilation as a limitation mecha-
nism for flame surface production. The different lines represent the peak consumption
rates of CH4 , O2 , H2 and CO at a) 0.61t f , b) 0.72t f , c) 0.75t f , d) 0.78t f , e) 0.81t f , and
f) 1.1t f , where t f is the flame time (from Ref. 96).
The effect of unsteady strain on premixed flames cannot be described in such simple
terms. Numerical calculations by Im and Chen94 indicate that the reponse of the
flame to modulated strain rates takes the form of cycles around the steady-state
line. The size of the cycle diminishes as the frequency increases.
The flame-surface area is augmented when the strain rate acting on the reactive
elements is lower than the extinction value. Conversely, the flame area is limited by
a mechanism of mutual interactions of adjacent reactive elements. This mechanism
has been identified as a fundamental process reducing the flame-surface area (see,
for example, Echekki et al.95 or Chen et al.96 and Fig. 9.16). This mechanism may
also influence the dynamics of turbulent flames. The mutual interaction of strained
flames is now well understood in cases in which the flame elements tend to propa-
gate away from each other. The case of strained elements approaching each other
and leading to a shortening of the flame is less easy to study experimentally. The
possibility of having synchronized interactions leading to instability is not gener-
ally considered. This type of coupling process has been observed experimentally
at least by Schuller et al.59 The corresponding driving process is briefly evoked in
Sec. III.
206 S. DUCRUIX ET AL.
Effect Induced
Initial on flow field
perturbation and flame Main result Consequences changes
V. Conclusion
The development of predictive methods for combustion instabilities is an impor-
tant technological objective. This prediction is now essential to the development
of advanced combustors for gas turbines. Considerable progress has been made
in this direction. Experiments and detailed analysis have generated a wealth of
information on the basic processes involved. This chapter illustrates some of these
processes and focuses on the driving and coupling relations that exist between
heat-release fluctuations and acoustic variables. Tables 9.1 and 9.2 summarize the
different paths examined in this review. By using well-controlled experiments, it
is shown that rapid changes of the flame surface generate an intense radiation of
sound. In practical situations, there are many possible mechanisms that may pro-
duce or destroy flame surface at a fast rate, such as flame–wall interactions and
collisions between adjacent flames or between neighboring flow structures like
vortices or reactant jets. These processes may feed energy into a resonant mode if
Effect
Initial on flow
perturbation and flame Main result Consequences
they are properly phased with respect to the pressure. Fast changes in flame-surface
area constitute an important driving process of combustion instabilities.
The coupling (or feedback) mechanism between the pressure field and the com-
bustion process may take many different forms. It is illustrated here with a set of
experiments with laminar flames, but previous experiments on turbulent ducted
configurations have indicated that premixed flames are quite susceptible to such
modulations. The flame is highly wrinkled by the external field, giving rise to
surface and heat-release fluctuations. In simple cases, it is possible to define a
transfer function between the relative velocity and heat-release fluctuations. Com-
parisons between analytical models, numerical simulations, and experiments are
reviewed. It is shown that simple filter models do not provide a suitable description
of the phase when the modulation frequency is high and that more refined methods
must be used to get a better description of this quantity. In practical systems, the
flame may also be modulated by many other means. Equivalence ratio perturba-
tions caused by the differential response of the injection system may also induce
heat-release fluctuations when these perturbations convected downstream reach
the flame, which has been identified as a possible driving process for some types
of gas-turbine instabilities. Flame modulation may result from the field of variable
strain rate, which can be induced by the nonsteady motion in the combustor. The
variable strain rate can produce or diminish the flame-surface area, and modify the
local rate of reaction per unit surface. If the fluctuations are suitably phased, they
will feed energy back into the acoustic motion.
Although the information accumulated over many years of research is quite sub-
stantial, additional fundamental experiments and intermediate scale investigations
are still needed. Further modeling, with a focus on coupling and driving processes
is required together with detailed simulations. The results gathered recently could
be used to check numerical tools and validate simulations of combustion dynamics.
The interactions examined in this chapter only portray some of the mechanisms in-
volved in the more complex dynamics of gas-turbine combustors, a subject covered
in further detail in this book.
References
1
Mallard, E., and Le Châtelier, H., “Recherches Expérimentales et Théoriques sur la
Combustion de Mélanges Gazeux Explosifs,” Annales des Mines, Paris Series, Vol. 8,
1883, pp. 274–377.
2
Lord Rayleigh, “The Explanation of Certain Acoustic Phenomena,” Nature, Vol. 18,
1878, pp. 319–321.
3
Candel, S., “Combustion Instabilities Coupled by Pressure Waves and Their Active Con-
trol,” Proceedings of the Combustion Institute, Vol. 24, The Combustion Inst., Pittsburgh,
PA, 1992, pp. 1277–1296.
4
McManus, K., Poinsot, T., and Candel, S., “A Review of Active Control of Combustion
Instabilities,” Progress in Energy and Combustion Science, Vol. 19, No. 1, 1993, pp. 1–29.
5
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engines Combustion Instabilities,
Vol. 169, Progress in Astronautics and Aeronautics, AIAA, Reston, VA, 1995.
6
De Luca, D., Price, E. W., and Summerfield, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 1992.
208 S. DUCRUIX ET AL.
7
Candel, S., “Combustion Dynamics and Control: Progress and Challenges,” Proceed-
ings of the Combustion Institute, Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002,
pp. 1–28.
8
Lieuwen, T., and McManus, K. (eds.), “Combustion Dynamics in Lean-Premixed Pre-
vaporized (LPP) Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003,
pp. 721–829.
9
Poinsot, T. (ed.), “Large Eddy Simulation of Reacting Flows,” Flow Turbulence and
Combustion, Vol. 65, No. 2, 2000, pp. 111–244.
10
Desjardins, P. E., and Frankel, S. H., “Two-dimensional Large Eddy Simulation of Soot
Formation in the Near-field of a Strongly Radiating Nonpremixed Acetylene-Air Turbulent
Jet Flame,” Combustion and Flame, Vol. 119, No. 1–2, 1999, pp. 121–133.
11
Kim, W. W., Menon, S., and Mongia, H. C., “Large Eddy Simulation of a Gas Turbine
Combustor Flow,” Combustion Science and Technology, Vol. 143, No. 1–6, 1999, pp. 25–62.
12
Fureby, C., “A Computational Study of Combustion Instabilities due to Vortex Shed-
ding,” Proceedings of the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh,
PA, 2000, pp. 783–791.
13
Pitsch, H., and Duchamp de Lageneste, L., “Large-Eddy Simulation of Premixed Tur-
bulent Combustion Using a Level-Set Approach,” Proceedings of the Combustion Institute,
Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002, pp. 2001–2008.
14
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., “Large-Eddy Simulation of Com-
bustion Dynamics of Lean-Premixed Swirl-Stabilized Combustor,” Journal of Propulsion
and Power, Vol. 19, No. 5, 2003, pp. 782–794.
15
Selle, L., Lartigue, G., Poinsot, T., Koch, R., Schildmacher, K.-U., Krebs, W., Prade,
B., Kaufmann, P., and Veynante, D., “Compressible Large Eddy Simulation of Turbulent
Combustion in Complex Geometry on Unstructured Meshes,” Combustion and Flame,
Vol. 137, No. 3, 2004, pp. 489–505.
16
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, Edwards,
Philadelphia, 2001, 473 pp.
17
Ducruix, S., Schuller, T., Durox, D., and Candel, S., “Combustion Dynamics and Insta-
bilities: Elementary Coupling and Driving Mechanisms,” Journal of Propulsion and Power,
Vol. 19, No. 5, 2003, pp. 722–734.
18
Lieuwen, T., “Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
view,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 765–781.
19
Dowling, A. P., and Stow, S. R., “Modal Analysis of Gas Turbine Combustor Acoustics,”
Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 751–764.
20
Yu, K. H., Trouvé, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in
a Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, Nov. 1991, pp. 47–72.
21
Smith, D. A., and Zukoski, E. E., “Combustion Instability Sustained by Unsteady Vortex
Combustion,” AIAA-SAE-ASME-ASEE Twenty-First Joint Propulsion Conference, AIAA
Paper 85-1248, July 1985.
22
Culick, F. E. C., and Yang, V., “Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines,” Liquid Rocket Engines Combustion Instability, Vol. 169,
Progress in Astronautics and Aeronautics, AIAA, Reston, VA, 1995, pp. 3–37.
23
Strahle, W. C., “Duality, Dilatation, Diffusion, and Dissipation in Reacting Turbulent
Flows,” Proceedings of the Combustion Institute, Vol. 19, The Combustion Inst., Pittsburgh,
PA, 1982, pp. 337–347.
24
Williams, F. A., Combustion Theory, Benjamin Cummings, Menlo Park, CA, 1985.
25
Doak, P. E., “Fundamentals of Aerodynamic Sound Theory and Flow Duct Acoustics,”
Journal of Sound and Vibration, Vol. 28, No. 3, 1973, pp. 527–561.
COMBUSTION INSTABILITY MECHANISMS 209
26
Kotake, S., “On Combustion Noise Related to Chemical Reactions,” Journal of Sound
and Vibration, Vol. 42, No. 3, 1975, pp. 399–410.
27
Crocco, L., “Aspects of Combustion Instability in Liquid Propellant Rocket Motors,”
Journal of the Aeronautical Research Society, Vols. 21 and 22, 1952.
28
Tsien, H. S., “Servo-stabilization of Combustion in Rocket Motors,” American Rocket
Society Journal, Vol. 22, 1952, pp. 256–263.
29
Crocco, L., and Cheng, S. L., Theory of Combustion Instability in Liquid Propel-
lant Rocket Motors, Agardograph No. 8, Butterworths Science Publication, Butterworths,
London, 1956.
30
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,”
Combustion and Flame, Vol. 70, No. 3, 1987, pp. 281–289.
31
Gulati, A., and Mani, R., “Active Control of Unsteady Combustion-Induced Oscilla-
tions,” Journal of Propulsion and Power, Vol. 8, No. 5, 1992, pp. 1109–1115.
32
Hedge, U. G., Reuter, D., and Zinn, B. T., “Sound Generation by Ducted Flames,” AIAA
Journal, Vol. 26, No. 5, 1988, pp. 532–537.
33
Renard, P. H., Thévenin, D., Rolon, J. C., and Candel, S., “Dynamics of Flame-Vortex
Interactions,” Progress in Energy and Combustion Science, Vol. 26, No. 3, 2000, pp. 225–
282.
34
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., “Local Flame Propagation Speeds
Along Wrinkled, Unsteady, Stretched Flames,” Proceedings of the Combustion Institute,
Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 827–832.
35
Renard, P. H., Rolon, J. C., Thévenin, D., and Candel, S., “Investigation of Heat Release,
Extinction and Time Evolution of the Flame Surface, for a Nonpremixed Flame Interacting
with a Vortex,” Combustion and Flame, Vol. 117, No. 1, 1999, pp. 189–205.
36
Poinsot, T., Trouvé, A., Veynante, D., Candel, S., and Esposito, E., “Vortex-Driven
Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics, Vol. 177,
April 1987, pp. 265–292.
37
Schadow, K. C., Gutmark, E. J., Parr, T. P., Parr, D. M., Wilson, K. J., and Crump,
J. E., “Large-Scale Coherent Structures as Drivers of Combustion Instability,” Combustion
Science and Technology, Vol. 64, No. 4–6, 1989, pp. 167–186.
38
Crow, S. C., and Champagne, F. H., “Orderly Structure in Jet Turbulence,” Journal of
Fluid Mechanics, Vol. 48, No. 3, 1971, pp. 547–591.
39
Ho, C. M., and Nosseir, N. S., “Dynamics of an Impinging Jet. Part 1: The Feedback
Phenomenon,” Journal of Fluid Mechanics, Vol. 105, April 1981, pp. 119–142.
40
Ho, C.-M., and Huerre, P., “Perturbed Free Shear Layers,” Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365–424.
41
Kaiktsis, L., Karniadakis, G. E., and Orszag, S., “Unsteadiness and Convective Instabil-
ities in Two-Dimensional Flow over a Backward-Facing Step,” Journal of Fluid Mechanics,
Vol. 321, August 1996, pp. 157–187.
42
Hussain, A. K., and Zaman, K. B., “Vortex Pairing in a Circular Jet under Controlled
Excitation. Part 1. General Jet Response,” Journal of Fluid Mechanics, Vol. 101, Dec. 1980,
pp. 449–491.
43
Hussain, A. K., and Zaman, K. B., “Vortex Pairing in a Circular Jet under Controlled
Excitation. Part 2. Coherent Structure Dynamics,” Journal of Fluid Mechanics, Vol. 101,
Dec. 1980, pp. 493–544.
44
Oster, D., and Wygnanski, I., “The Forced Mixing Layer Between Parallel Streams,”
Journal of Fluid Mechanics, Vol. 123, Oct. 1982, pp. 91–130.
45
Gaster, M., Kit, E., and Wygnanski, I., “Large-Scale Structures in a Forced Turbulent
Mixing Layer,” Journal of Fluid Mechanics, Vol. 150, Jan. 1985, pp. 23–39.
210 S. DUCRUIX ET AL.
46
Ghoniem, A. F., and Ng, K. K., “Numerical Study of the Dynamics of a Forced Shear
Layer,” Physics of Fluids, Vol. 30, No. 3, 1987, pp. 706–723.
47
Mitchell, B. E., Lele, S. K., and Moin, P., “Direct Computation of the Sound Generated
by Vortex Pairing in an Axisymmetric Jet,” Journal of Fluid Mechanics, Vol. 383, March
1999, pp. 113–142.
48
Bogey, C., Bailly, C., and Juvé, D., “Numerical Simulation of Sound Generated by
Vortex Pairing in a Mixing Layer,” AIAA Journal, Vol. 38, No. 12, 1999, pp. 2210–2218.
49
Durox, D., Schuller, T., and Candel, S., “Combustion Dynamics of Inverted Conical
Flames,” Proceedings of the Combustion Institute, Vol. 30, The Combustion Inst., Pittsburgh,
PA, 2004, pp. 1717–1724.
50
Kendrick, D. W., Zsak, T. W., and Zukoski, E. E., “An Experimental and Numerical
Investigation of Premixed Combustion in a Vortex in a Laboratory Dump Combustor,”
Unsteady Combustion, NATO ASI Series, Vol. 306, Kluwer, Dordrecht 1996, pp. 33–69.
51
Zsak, T. W., “An Investigation of the Reacting Vortex Structures Associated with Pulse
Combustion,” Ph.D. Dissertation, California Inst. of Technology, Pasadena, CA, April
1993.
52
Schuller, T., Durox, D., and Candel, S., “Dynamics of and Noise Radiated by a Perturbed
Impinging Premixed Jet Flame,” Combustion and Flame, Vol. 128, No. 1–2, 2002, pp. 88–
110.
53
Durox, D., Schuller, T., and Candel, S., “Self-induced Instability of a Premixed Jet
Flame Impinging on a Plate,” Proceedings of the Combustion Institute, Vol. 29, The Com-
bustion Inst., Pittsburgh, PA, 2002, pp. 69–75.
54
Schäfer, O., Koch, R., and Wittig, S., “Measurement of the Periodic Flow of an En-
closed Lean Premixed Prevaporized Stagnation Flame,” Tenth International Symposium on
Applications of Laser Techniques to Fluid Mechanics, Lisbon, 2000.
55
Bragg, S. L., “Combustion Noise,” Journal of the Institute of Fuel, Vol. 36, 1963,
pp. 12–16.
56
Abugov, D. I., and Obrezkov, O. I., “Acoustic Noise in Turbulent Flames,” Combustion,
Explosions and Shock Waves, Vol. 14, 1978, pp. 606–612.
57
Clavin, P., and Siggia, E., “Turbulent Premixed Flames and Sound Generation,” Com-
bustion Science and Technology, Vol. 78, No. 1–3, 1991, pp. 147–155.
58
Kidin, N., Librovich, V., Roberts, J., and Vuillermoz, M., “On Sound Sources in Tur-
bulent Combustion,” Dynamics of Flames and Reactive Systems, Vol. 95, Progress in As-
tronautics and Aeronautics, AIAA, New York, 1984, pp. 343–355.
59
Schuller, T., Durox, D., and Candel, S., “Self-induced Combustion Oscillations of
Laminar Premixed Flames Stabilized on Annular Burners,” Combustion and Flame,
Vol. 135, No. 4, 2003, pp. 525–538.
60
Candel, S., Durox, D., and Schuller, T., “Flame Interactions as a Source of Noise and
Combustion Instabilities,” AIAA Paper 2004-2928, AIAA/CEAS Aeroacoustics Confer-
ence, Manchester, England, U.K., 2004.
61
McIntosh, A. C., “On Flame Resonance in Tubes,” Combustion Science and Technology,
Vol. 69, No. 4–6, 1990, pp. 147–152.
62
McIntosh, A. C., “The Linearised Response of the Mass Burning Rate of a Premixed
Flame to Rapid Pressure Changes,” Combustion Science and Technology, Vol. 91, No. 4–6,
1993, pp. 329–346.
63
Edwards, N. R., McIntosh, A. C., and Brindley, J., “The Development of Pressure
Induced Instabilities in Premixed Flames,” Combustion Science and Technology, Vol. 99,
No. 1–3, 1996, pp. 373–386.
COMBUSTION INSTABILITY MECHANISMS 211
64
Park, S., Annaswamy, A. M., and Ghoniem, A. F., “Heat Release Dynamics Modeling
of Kinetically Controlled Burning,” Combustion and Flame, Vol. 128, No. 3, 2002, pp. 217–
231.
65
Markstein, G. H., Non Steady Flame Propagation, Pergamon Press, Elmsford, NY,
1964.
66
Blackshear, P. L., “Driving Standing Waves by Heat Addition,” Proceedings of the
Combustion Institute, Vol. 4, The Combustion Inst., Pittsburgh, PA, 1953, pp. 553–566.
67
De Sœte, G., “Etude des Flammes Vibrantes. Application à la Combustion Turbulente,”
Revue de l’Institut Français du Pétrole et Annales des Combustibles Liquides, Vol. 19, No. 6,
1964, pp. 766–785.
68
Merk, H. J., “An Analysis of Unstable Combustion of Premixed Gases,” Proceedings
of the Combustion Institute, Vol. 6, The Combustion Inst., Pittsburgh, PA, 1956, pp. 500–
512.
69
Becker, R., and Günther, R., “The Transfer Function of Premixed Turbulent Jet Flames,”
Proceedings of the Combustion Institute, Vol. 13, The Combustion Inst., Pittsburgh, PA,
1971, pp. 517–526.
70
Baade, P. K., “Design Criteria and Modes for Preventing Combustion Oscillations,”
ASHRAE Transactions, Vol. 1, 1978, pp. 449–465.
71
Matsui, Y., “An Experimental Study on Pyro-acoustic Amplification of Premixed
Laminar Flames,” Combustion and Flame, Vol. 43, No. 2, 1981, pp. 199–209.
72
Baillot, F., Durox, D., and Prud’homme, R., “Experimental and Theoretical Study of
a Premixed Vibrating Flame,” Combustion and Flame, Vol. 88, No. 2, 1992, pp. 149–
168.
73
Fleifil, M., Annaswamy, A. M., Ghoneim, Z. A., and Ghoniem, A. F., “Response of a
Laminar Premixed Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic
Instability Results,” Combustion and Flame, Vol. 106, No. 4, 1996, pp. 487–510.
74
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determinations
of the Transfer Function of a Laminar Premixed Flame,” Proceedings of the Combustion
Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 765–773.
75
Schuller, T., Durox, D., and Candel, S., “A Unified Model for the Prediction of Flame
Transfer Functions: Comparison Between Conical and V-flame Dynamics,” Combustion
and Flame, Vol. 134, No. 1–2, 2003, pp. 21–34.
76
Dowling, A. P., “A Kinematic Model of a Ducted Flame,” Journal of Fluid Mechanics,
Vol. 394, September 1999, pp. 51--72.
77
Marble, F. E., and Candel, S., “An Analytical Study of the Non-steady Behavior of
Large Combustors,” Proceedings of the Combustion Institute, Vol. 17, The Combustion
Inst., Pittsburgh, PA, 1978, pp. 761–769.
78
Bourehla, A., and Baillot, F., “Appearance and Stability of a Laminar Conical Premixed
Flame Subjected to an Acoustic Perturbation,” Combustion and Flame, Vol. 114, No. 3–4,
1998, pp. 303–318.
79
Lieuwen, T., and Neumeier, Y., “Nonlinear Pressure-Heat Release Transfer Function
Measurements in a Premixed Combustor,” Proceedings of the Combustion Institute, Vol. 29,
The Combustion Inst., Pittsburgh, PA, 2002, pp. 99–105.
80
Schuller, T., Ducruix, S., Durox, D., and Candel S., “Modeling Tools for the Prediction
of Premixed Flame Transfer Functions,” Proceedings of the Combustion Institute, Vol. 29,
The Combustion Inst., Pittsburgh, PA, 2002, pp. 107–113.
81
Boyer, L., and Quinard, J., “On the Dynamics of Anchored Flames,” Combustion and
Flame, Vol. 82, No. 1, 1990, pp. 51–65.
212 S. DUCRUIX ET AL.
82
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driving
Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion Insti-
tute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 1809–1816.
83
Lee, J. G., Kwanwoo, K., and Santavicca, D. A., “Measurement of Equivalence Ratio
Fluctuation and Its Effect on Heat Release during Unstable Combustion,” Proceedings of
the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 415–421.
84
Hathout, J. P., Fleifil, M., Annaswamy, A. M., and Ghoniem, A. F., “Heat-Release Actu-
ation for Control of Mixture-Inhomogeneity-Driven Combustion Instability,” Proceedings
of the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 721–
730.
85
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182–190.
86
Marzouk, Y. M., Ghoniem, A. F., and Najm, H. N., “Dynamic Response of Strained
Premixed Flames to Equivalence Ratio Gradients,” Proceedings of the Combustion Institute,
Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 1859–1866.
87
Lieuwen, T., Neumeier, Y., and Zinn, B. T., “The Role of Unmixedness and Chemical
Kinetics in Driving Combustion Instabilities in Lean Premixed Combustors,” Combustion
Science and Technology, Vol. 135, No. 1–6, 1998, pp. 193–211.
88
Law, C. K., “Dynamics of Stretched Flames,” Proceedings of the Combustion Institute,
Vol. 22, The Combustion Inst., Pittsburgh, PA, 1988, pp. 1381–1402.
89
Peters, N., Turbulent Combustion, Cambridge University Press, Cambridge, U.K., 2000.
90
Welle, E. J., Roberts, W. L., Decroix, M. E., Carter, C. D., and Donbar, J. M., “Simulta-
neous Particle-Imaging Velocimetry and OH Planar Laser Induced Fluorescence Measure-
ments in an Unsteady Counterflow Propane Diffusion Flame,” Proceedings of the Combus-
tion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 2021–2027.
91
Darabiha, N., “Transient Behaviour of Laminar Counter Flow Hydrogen-Air Flames
with Complex Chemistry,” Combustion Science and Technology, Vol. 86, No. 1–6, 1992,
pp. 163–181.
92
Egolfopoulos, F. N., and Campbell, C. S., “Unsteady Counterflowing Strained Diffusion
Flames: Diffusion-Limited Frequency Response,” Journal of Fluid Mechanics, Vol. 318,
July 1996, pp. 1–29.
93
Joulin, G., “On the Response of Premixed Flames to Time-Dependent Stretch and
Curvature,” Combustion Science and Technology, Vol. 97, No. 1–3, 1994, pp. 219–229.
94
Im, H. G., and Chen, J. H., “Effects of Flow Transients on the Burning Velocity of
Laminar Hydrogen-Air Premixed Flames,” Proceedings of the Combustion Institute, Vol. 28,
The Combustion Institute, Pittsburgh, 2000, pp. 1833–1840.
95
Echekki, T., Chen, J., and Gran, I., “The Mechanism of Mutual Annihilation of
Stoichiometric Premixed Methane-Air Flames,” Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 855–863.
96
Chen, J., Echekki, T., and Kollman, W., “The Mechanism of Two-Dimensional Pocket
Formation in Lean Premixed Methane-Air Flames with Implication to Turbulent Combus-
tion,” Combustion and Flame, Vol. 116, No. 1–2, 1999, pp. 15–48.
Chapter 10
I. Introduction
F UEL injection and mixing are critical to achieving efficient and clean com-
bustion in modern gas-turbine engines, whether they are powered by gaseous
or liquid fuels. For gaseous fuels, the major concern is to obtain an optimal level of
mixing between air, fuel, and combustion products in the combustion zone. When
liquid fuels are employed, they must be atomized into small droplets and then dis-
tributed in an airstream before entering the combustion zone.1 Most gas-turbine
injectors employ swirl configurations that produce central toroidal recirculation
zones (CTRZs) to provide the dominant flame-stabilization mechanism. Flows in
this region are generally associated with high shear rates and strong turbulence
intensities resulting from vortex breakdown.
Many experimental studies have been conducted to investigate the flow and flame
dynamics of swirl injectors. An overview of the use of these injectors in liquid-
propellant rocket engines was recently published by Bazarov et al.2 For fuel in-
jectors typical of gas-turbine engines, excellent descriptions and their applications
in modern dry low-emission (DLE) combustors were given by Lefebrve.3 Wang
et al.4, 5 conducted an experimental study of a 3 × scale model of the CFM56 coaxial
swirl cup. The droplet dynamics were characterized by means of phase Doppler in-
terferometry. Jeng and colleagues6–8 examined the counter-rotating flow structures
produced by the CFM56 swirl cup. The effects of air temperature, fluid property,
and the equivalence ratio on spray characteristics were studied under both nonre-
acting and reacting conditions. Cowell and Smith9 tested a liquid-fueled injector
in a bench-scale can combustor to evaluate critical design and operating parame-
ters for emission characteristics. Both axial and radial swirlers were explored. The
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and
Astronautics, Inc., with permission.
∗ Postdoctoral Research Associate. Member AIAA.
† Research Associate. Member AIAA.
‡ Distinguished Professor. Fellow AIAA.
213
214 Y. HUANG ET AL.
size and shape of the recirculation zone associated with each swirler were found
to exert a strong influence on combustor performance. Snyder et al.10 carried out
an experimental program to develop a liquid-fueled injector with tangential entry.
The combustion performance was evaluated in a high-pressure, single-nozzle test
facility. A similar injector configuration was later explored by Cohen et al.11 in
the development of an active combustion control system at realistic engine (i.e.,
FT-8) operating conditions.
For laboratory research swirl injectors, a considerable amount of experimental
work was conducted. Richards et al.12, 13 investigated the flame dynamics of a pre-
mixed fuel injector using natural gas. A simple time-lag model was proposed to
characterize experimentally observed combustion oscillations. Broda et al.,14 Lee
et al.,15 Venkataraman et al.,16 and Lee et al.17, 18 studied the combustion dynamics
of gaseous-fueled single-element swirl injectors. The effects of the equivalence
ratio, inlet velocity, temperature, fuel distribution, swirl number, and the pres-
ence of a centerbody recess on combustion stability were examined. Mordaunt
et al.19 conducted a series of experiments to investigate combustion dynamics
in a single-element injector using various fuels, including gaseous ethylene and
three different liquid hydrocarbon fuels: n-heptane, JP-8, and a coal-based fuel.
Cohen and Rosfjord,20, 21 Chin et al.,22, 23 and Hardalupas et al.24, 25 measured the
spray characteristics downstream of swirl injectors to improve the understanding
of the liquid fuel atomization process. Presser et al.26, 27 examined the aerodynamic
characteristics of a swirling spray flame using a pressure-jet atomizer. The effects
of swirl on droplet transport, as well as the interactions between droplets and air
flowfield, were studied under both nonreacting and reacting conditions. Paschereit,
et al.,28, 29 Acharya et al.,30, 31 Murugappan et al.,32 Lee et al.,33 Richards, et al.,34, 35
and Zinn and colleagues36–39 investigated the characteristics and control of com-
bustion instabilities in swirl-stabilized combustors. Bernier et al.40 analyzed the
combustion dynamics in a liquid-fueled premixed prevaporized burner using co-
and counter-rotating swirl injectors. Li and Gutmark41 examined the effects of
swirler orientation and exhaust-nozzle geometry on the flow and flame character-
istics of a dual-fueled multiple swirler combustion system.
Extensive efforts were applied to numerical studies of gas turbine combustion.
Brewster et al.42 conducted a comprehensive review of numerical simulations for
stationary gas turbines that were based on the Reynolds-averaged Navier–Stokes
(RANS) equations. The RANS simulation may be appropriate for time-mean tur-
bulent flow properties, but its validity for unsteady flow evolution has yet to be
established, especially for problems involving such complicated configurations
as swirl injectors. Recent advances in large-eddy simulations (LES) have shown
promise for studying the dynamics of swirl injectors. The technique computes ex-
plicitly the contributions of large energy-carrying structures to mass, momentum,
and energy transfer in the flowfield, with the effects of unresolved small-scale tur-
bulence modeled either analytically or empirically. Although the RANS method
remains the main workhorse for combustor design analysis, LES is considered to be
the next-generation analytical design tool for gas turbine combustors. In an effort
to improve the understanding of the flow and flame phenomena within the CFM56
aero-engine swirl cup, a series of numerical investigations were performed by GE
Aircraft Engines using both RANS and LES techniques.43–51 The LES method was
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 215
25.4
unit: mm 600
Sec. III, the flame dynamics of an axial-entry swirl injector operating over a wide
range of flow conditions is studied. Finally, a summary is given in Sec. IV.
S = 0.3
0 5 10 x/H
S = 0.5
0 5 10 x/H
Fig. 10.2 Streamlines based on mean axial and radial velocity components; swirl
numbers S = 0.3 and 0.5.
where Rh and Rn are the radii of the centerbody and the inlet duct, respectively. If
we assume that the axial and azimuthal velocities are uniform and that the vanes
are thin, the swirl number can be written as
2 1 − (Rh /Rn )3
S= tan ϕ (10.2)
3 1 − (Rh /Rn )2
where ϕ is the swirler vane angle. The present discussion focuses on the high
swirl-number case of S = 0.5 because of the occurrence of vortex breakdown.
The situation with S = 0.3, in which vortex breakdown is not observed, is also
treated for comparison.
∂p ρUθ2
= (10.3)
∂r r
The pressure tends to be minimized in regions where strong swirling motions
occur, that is, in the wake of the centerbody. As the flow expands and the azimuthal
velocity decays with the axial distance, the pressure is recovered in the downstream
218 Y. HUANG ET AL.
Fig. 10.3 Time evolution of streamlines based on mean axial and radial velocity
components spatially averaged in the azimuthal direction (time increment of 0.6 ms),
swirl number S = 0.5.
region. A positive pressure gradient is consequently generated along the axial axis,
which may lead to the formation of a recirculation zone in a high swirling flow, a
phenomenon commonly termed vortex breakdown. In the present configuration and
flow condition, vortex breakdown occurs only at high swirl numbers, as evidenced
in Fig. 10.2. A CTRZ is formed in the central region for S = 0.5, reaching from
x/H ≈ 0.36 to 7.8.
The temporal evolution of the flowfield is examined to explore the phenomenon
of vortex breakdown. Figure 10.3 shows the instantaneous streamlines on an x-
r plane, spatially averaged in the azimuthal direction, at various times during a
typical flow evolution period. The time increment between the snapshots is 0.6 ms,
and t = 0 corresponds to the instant at which data collection begins after the flow
reaches its stationary state. At t = 37.7 ms, a new vortical bubble is generated in
front of a braid of vortical bubbles. These bubbles then coalesce at t = 39.0 ms.
The bubble located in the downstream side of the vortical braid is separated into
two structures at t = 39.6 ms; one stays at basically the same location, and the
other is convected downstream and finally disappears because of turbulent dif-
fusion and viscous dissipation at t = 41.4 ms. During this period, the coalesced
vortical bubble separates, and another new bubble appears in the upstream region at
t = 42.0 ms. These snapshots exhibit a very complicated vortex evolution in the
central region.
Figure 10.4 shows snapshots of the axial velocity fields for the swirl numbers of
0.3 and 0.5. Shear layers are produced at the trailing edges of the centerbody and
the backward-facing step because of the Kelvin–Helmholtz instabilities in both
the axial and azimuthal directions. The flow evolution in the azimuthal direction
is presented on the two crosssections at x/H = 0.36 and 1.54. The shear layer
originating from the backward-facing step remains almost symmetric immediately
downstream of the dump plane (x/H = 0.36). It then rolls up and forms large
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 219
Fig. 10.4 Instantaneous axial velocity fields on x−r plane and two cross-sections,
x/H = 0.36 and 1.54. Swirl numbers S = 0.3 and 0.5. Contour levels between –12
and 33 m/s with increment of 3 m/s. Solid lines represent positive values and dashed
lines represent negative values.
asymmetric structures at x/H = 1.54, because of the strong shear force in the
azimuthal direction.
The large velocity difference in the azimuthal direction at a high swirl number
(S = 0.5) significantly increases the strength of the shear layer, especially around
the boundary of the CTRZ. As a consequence of flow reversal, the effective flow
passage area in the chamber is reduced, which increases the axial velocity differ-
ence and further enhances the shear layer in the axial direction. The large-scale
structures are eventually dissipated by turbulent diffusion and viscous damping
when the flow convects downstream.
Figure 10.5 shows snapshots of the isosurfaces of vorticity magnitude at || =
1.5U /H (i.e., 1133 1/s) for S = 0.3 and 0.5. Helical vortex tubes develop from
Fig. 10.5 Snapshots of isosurfaces of vorticity magnitude at 1.5 × U/H. Dark lines
represent streamlines; swirl numbers S = 0.3 and 0.5.
220 Y. HUANG ET AL.
Fig. 10.6 Instantaneous fluctuating pressure field on x−r plane and cross-sections
at x/H = 1.94, 5.87, 9.81, and 13.75. Contour levels between –600 and 600 Pa with
increment of 50 Pa. Solid lines represent positive values and dashed lines represent
negative values; swirl number S = 0.5.
the inlet and travel in a direction opposite to the main swirling flow, although the
whole structure follows the motion of the main flow. The swirl number plays an
important role in dictating the flow evolution and its underlying mechanisms. The
helical structure at S = 0.5 arises from the vortex breakdown and expands in the
downstream region. The situation is, however, different for a low swirl number, in
which the helical structure of the vortex tube shrinks in the downstream region.
This phenomenon may be attributed to the precession of the vortex core around the
centerline. The resultant intermittent occurrence of vortex breakdown causes the
helical structure issuing from the centerbody to vanish rapidly as the flow evolves
downstream. A high swirl number apparently helps maintain flow coherence and
leads to strong flow reversal.
2. Vortico-Acoustic Interaction
The strong vortical motion in the chamber often produces acoustic waves prop-
agating throughout the entire field. The shear layers, however, are susceptible to
acoustic excitations if such disturbances occur at appropriate locations and frequen-
cies. A feedback loop can thus be established, depending on the mutual coupling
between the vortical and acoustic fields. To explore vortico-acoustic interaction,
the fluctuating pressure field is obtained by subtracting the longtime averaged
pressure from its instantaneous quantity.
Figure 10.6 shows snapshots of the fluctuating pressure field on several cross
sections for the high swirl-number case with S = 0.5. Considerable pressure fluc-
tuations take place immediately downstream of the centerbody, where strong vor-
ticity is present. The entire field exhibits a wide range of length scales, with
broadband turbulent motion in the upstream region. This motion develops into
large-scale coherent acoustic motion in the downstream region. To help identify
the wave characteristics, especially those associated with longitudinal waves, the
complicated three-dimensional field shown in Fig. 10.6 is spatially averaged in
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 221
Fig. 10.7 Time evolution of fluctuating pressure field spatially averaged in the
azimuthal direction. Contour levels between –600 and 600 Pa with increment of
50 Pa (time increment of 0.3 ms). Solid lines represent positive values and dashed
lines represent negative values; swirl number S = 0.5.
the azimuthal direction. Figure 10.7 presents the time sequence of the resultant
quasi-two-dimensional fields. The data are further reduced by spatial averaging
over each cross section along the axial axis. The averaged quasi-one-dimensional
fields shown in Fig. 10.8 indicate that a negative pressure peak, followed by a
positive one, forms periodically and travels downstream at the speed of sound.
Two negative pressure peaks are observed in the chamber at t = 35.9 ms. Because
the evolution pattern of these two fluctuations is almost identical, the frequency
of the pressure wave is approximately 655 Hz, based on the distance between the
two pressure peaks at t = 35.9 ms and the wave propagation speed.
The frequency spectra of pressure fluctuations are obtained to quantitatively
characterize the acoustic flow evolution. Figure 10.9 shows the result at y/H =
0.06 and z/H = 0.07 with three different axial positions, x/H = 0.56, 2.19, and
7.20, which are located in the upstream, center, and downstream regions, re-
spectively, of the central recirculation zone. The most dominant mode is 1380
Hz, and other characteristic frequencies include 660, 2040, and 3420 Hz. The
222 Y. HUANG ET AL.
50 t = 34.4 ms t = 35.6 ms
0
p' , Pa
–50
–100
50 t = 34.7 ms t = 35.9 ms
0
p' , Pa
–50
–100
50 t = 35.0 ms t = 36.2 ms
0
p' , Pa
–50
–100
50 t = 35.3 ms t = 36.5 ms
0
p' , Pa
–50
–100
0 5 10 15 20 0 5 10 15 20
x/H x/H
Fig. 10.8 Time evolution of fluctuating pressure field spatially averaged over r –θ
cross section; swirl number S = 0.5.
a) b) c)
60
p' , Pa
40
20
0
0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000
frequency, Hz frequency, Hz frequency, Hz
Fig. 10.9 Frequency spectra of pressure fluctuations at different axial locations near
the centerline: a) x/H = 0.38; b) x/H = 2.19; and c) x/H = 7.20, y/H = 0.06, and
z/H = 0.07. Swirl number S = 0.5.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 223
fSδ 1
St = and Ū = (U1 + U2 ) (10.4)
Ū 2
where δ is the initial momentum thickness of the shear layer, and U1 and U2
are the freestream velocities on the two sides of the shear layer. The mean averaged
axial velocity Ū is approximately 10 m/s near the trailing edge of the centerbody.
The most unstable mode of an unforced planar shear layer occurs at St ≈ 0.044–
0.048 for turbulent flows.78 The momentum thickness δ is roughly one-fourth of
the vorticity thickness,79 which can be calculated from the axial velocity profile in
the radial direction. On the basis of Eq. (10.4), the frequency of the most unstable
mode of shear-layer instability, f S0 , is approximately on the order of 103 Hz near
the downstream region of the centerbody.
For the case with a swirl number of 0.5, the most prevalent acoustic mode
shown in Fig. 10.9 has a frequency of 1380 Hz, which is consistent with the
frequency of the most unstable shear-instability mode. Consequently, the shear
layer originating from the centerbody can be easily locked to the first-tangential
mode of the acoustic oscillation in the chamber. At this response frequency, f SR ,
the shear layer rolls up into discrete vortices and reinforces the acoustic oscillation
in the chamber. A feedback loop between the acoustic fluctuation and shear-layer
instability is thus established and leads to a large excursion of flow motions. When
the discrete vortices are convected downstream, they pair with the adjacent ones
to form larger structures with a characteristic frequency of f SR /2 = 690 Hz. This
subharmonic frequency is also close to the frequency of the longitudinal acoustic
mode in the chamber, 655 Hz, as determined from the reduced one-dimensional
pressure profiles shown in Fig. 10.8.
The effects of swirl number on the acoustic field were examined by considering a
low swirl number of S = 0.3. The dominant frequency over the entire field becomes
224 Y. HUANG ET AL.
3900 Hz, which corresponds to the mixed first tangential (1T) and first radial (1R)
acoustic mode in the chamber. A proper orthogonal decomposition (POD)81 anal-
ysis of the fluctuating pressure flowfield also confirms the prevalence of the mixed
1T/1R acoustic mode. As previously mentioned, the two shear layers originating
from the trailing edges of the centerbody and the backward-facing step may exert
significant influence on the oscillatory flow characteristics. Their specific effects
depend on the swirl number and the chamber geometry. At the high swirl number of
S = 0.5, the large vortical structure associated with the central recirculating flow
overshadows the shear layer originating from the corner region and dominates the
flow development in the chamber. The resultant acoustic wave thus has a character-
istic frequency of 1380 Hz, which matches the frequency of the centerbody shear-
layer instability. At a low swirl number of S = 0.3, no vortex breakdown occurs and
the importance of the flowfield on the downstream side of the centerbody in excit-
ing acoustic oscillations diminishes. In contrast, the shear layer in the corner region
plays a crucial role in dictating the acoustic flow evolution, whose characteristic fre-
quency of 3900 Hz matches that of the mixed 1T/1R acoustic mode in the chamber.
In short, the dominant acoustic mode in the chamber is sensitive to unsteady
vorticity evolution, which in turn strongly depends on the swirl number.
S1
S2 S3
S2 S3
S1
Fig. 10.10 Schematic of gas turbine swirl injector with radial entry.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 225
mixing duct includes a center cylindrical passage and two annular passages, which
are spaced radially outward from the axial axis. Three radial-entry swirlers—
denoted S1 , S2 , and S3 and counter-rotating with each other—are located at the
entrance. This injector is also referred to as a high-shear nozzle/swirler (HSNS)
and has four major advantages. First, it reduces smoke by introducing high swirl
from the first passage. Second, the middle swirler is implemented to generate
strong shear layers in both the axial and azimuthal directions and to reduce the
overall swirl angle. Consequently, the fuel–air mixing is improved. The inclusion
of the second passage makes it easy to control the initial swirl number of the flow.
Third, a stronger CTRZ is generated, which increases the relight stability. Finally,
the relight stability and the total flow rate can be decoupled by shifting the airflow
through the third passage.82
The mixing duct in the injector, shown in Fig. 10.10, has a diameter of D0 =
32 mm at the exit. Two different sets of swirl vanes are explored herein.62, 63 The
low swirl-number (LSN) case has swirl vane angles of S1 = 30 deg, S2 = −45 deg,
and S3 = 50 deg; and the high swirl-number (HSN) case has S1 = 45 deg, S2 =
−60 deg, and S3 = 70 deg. The corresponding swirl numbers are 0.35 and 0.49,
respectively, based on the flow properties at the injector exit. The baseline flow
condition includes an ambient pressure of 1 atm, an inlet temperature of 293 K,
and a mass flow rate of 0.077 kg/s. The Reynolds number that is based on the
diameter and the bulk axial velocity at the exit is 2 × 105 .
Fig. 10.11 Snapshots of vorticity magnitude contours: a) low swirl number and
b) high swirl number.
When the swirl number changes, the dominant instability mode may switch cor-
respondingly. A detailed analysis of these phenomena is given in the following
sections.
a. Vortex Breakdown. Much insight into the vortex breakdown in the core
flow region can be obtained from the isosurfaces of the azimuthal velocity shown in
Fig. 10.12. In the low swirl-number case, a stable bubble type of vortex breakdown
is clearly observed in the downstream region of the centerbody, whereas a much
more complex structure prevails at the high swirl number. The streamlines of
the mean flowfields given in Fig. 10.13 quantitatively reveal the formation of a
central toroidal recirculation zone in this region. As the swirl number increases,
the size of the recirculation zone accordingly becomes greater. The stagnation
point of the vortex breakdown moves upstream for an equilibrium position and
finally reaches the centerbody. The local flow development depends on the relative
magnitudes of the downward momentum inertia of the incoming flow and the
outward flow motion arising from the centrifugal force. Although the downward
momentum inertia remains almost the same because of the fixed inlet mass flow
rate employed, the weaker centrifugal force in the low swirl-number case causes the
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 227
a)
θ
−
b)
θ
−
Fig. 10.13 Streamlines of mean flowfields for swirl numbers of S = 0.35 and 0.49.
incoming flow to penetrate all the way to the core region, as evidenced in Fig. 10.13.
The ensuing flow structure bears a close resemblance to a tornado near the ground
where a large accumulation of vorticity in the center region takes place, a kind of
collapse of the swirling flow.83
The temporal evolution of the flowfield permits insight into the vortex break-
down phenomenon. Figure 10.14 shows instantaneous streamlines on a longitudi-
nal plane, spatially averaged in the azimuthal direction, at various times during a
typical flow evolution period for the high swirl-number case. Uneven time inter-
vals between frames were chosen to show the important phases of the oscillation.
Obviously, the spatially averaged flow structures are more distinguishable than
Fig. 10.14 Close-up views of streamlines downstream of centerbody for high swirl-
number case of S = 0.49. Flowfields spatially averaged in azimuthal direction. The
time interval between pictures is not constant.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 229
a)
b)
Fig. 10.15 Snapshots of azimuthal velocity fields on four transverse cross sections,
contour levels between –70 and 120 m/s with increment of 10 m/s. Solid lines represent
positive values and dashed lines represent negative values: a) low swirl number and
b) high swirl number.
swirl-number case. The center recirculating flow even intersects the outer shear
layer, causing a complex flowfield near the injector exit.
The dominant frequency of the vortex shedding because of the Kelvin–
Helmholtz instability in the streamwise direction can be estimated by using Eq.
(10.4). In the present configuration, the mean velocity, Ū , is 50 m/s, and the mo-
mentum thickness of the shear layer, θ , is around 0.2 mm for both swirl numbers.
The frequency of the most unstable mode, f n , is estimated to be 1 × 104 Hz.
This value is comparable with the numerically calculated instability frequency
of 13,000 Hz, further demonstrating that the outer shear flow dynamics is dic-
tated by the Kelvin–Helmholtz instability in the streamwise direction in the low
swirl-number case.
The situation is vastly different in the high swirl-number case. As a result of the
strong shear force and the associated Kelvin–Helmholtz, helical, and centrifugal
instabilities in the azimuthal direction, the flow becomes highly disordered soon
after the incoming streams merge in the region downstream of the guide vanes.
The interaction between the outer shear layer and the central toroidal recirculating
flow also contributes to the eddy breakup and mixing processes.
c. Interaction and Competition of Instability Modes. As previously men-
tioned, three major flow mechanisms (i.e., vortex breakdown, Kelvin–Helmholtz
instability, and helical instability) exist and interact with one another within the
injector. The specific type of coupling depends on the swirl number and can be
classified in two categories. First, the outer shear layer may interact with the large
disorganized structures arising from evolution of the central recirculating flow
when the swirl number exceeds a threshold value, as evidenced in Fig. 10.11. The
interaction usually increases with increasing swirl number and varies within each
flow evolution period. The vortex shedding tends to be more organized when the
center recirculation zone shrinks and less organized when it grows. The turbulent
kinetic energy in the central recirculation zone and in the wake of the guide vanes
is much greater than that in the rest of the domain because of vigorous vortical
motions in these regions. The two shear layers are distinctly separate in the low
swirl-number case but merge in the high swirl-number case. Because liquid fuel is
delivered into the injector from the centerbody, the high turbulence intensity in this
region can significantly enhance the atomization of the injected fuel. At the same
time, the strong shear stress in the downstream region of the second guide vane
promotes rapid mixing between the air and the fuel impinging and accumulating
on the second guide vane.
In the second type of flow coupling, the instability waves in the axial and az-
imuthal directions in the outer shear layer compete with each other. In the low
swirl-number case, the streamwise instability dominates the shear-layer evolution;
therefore, the billow structures and subsequent hairpin vortices prevail in the flow-
field. In the high swirl-number case, the development of the billows is suppressed
and flow structures are severely distorted by the azimuthal flow instabilities.
Several other competing mechanisms may also exist in the flowfield, such as the
one involving the Kelvin–Helmholtz and centrifugal instabilities. Swirling flows
usually result in an unstable radial stratification, thereby leading to centrifugal
instability,83 which is enhanced by a higher azimuthal velocity gradient and which
further influences the streamwise Kelvin–Helmholtz instability in the outer shear
layer.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 231
2. Spectral Characteristics
The injector dynamics involve an array of intricate flow processes characterized
by a wide range of time and length scales. Quantitative information can be obtained
by using spectral and proper-orthogonal-decomposition analyses for the low swirl-
number case, the dominant frequency of pressure oscillation along the main flow
passage is 13,000 Hz, corresponding to the most amplified mode of the shear-
layer instability downstream of the first guide vane. In the outer region of the
central recirculation zone, the prevalent frequency of 5783 Hz corresponds to the
precession of the vortex core (PVC). The phenomenon is confirmed by visual
inspection of the flow evolution data.
The situation is qualitatively different for the high swirl-number case, as shown
in Fig. 10.16. As a consequence of the strong interactions between the outer shear
layer and the central recirculation zone, the spectral content of the flowfield be-
comes very rich and is characterized by several different frequencies in various
regions. A low-frequency mode around 500 Hz dominates the flow oscillations
Fig. 10.16 Frequency spectra of pressure oscillations along main flow passage; high
swirl flow (S = 0.49).
232 Y. HUANG ET AL.
p: −4 −3 −2 −1 0 1 2 3 4
Fig. 10.17 POD mode shape of pressure field showing existence of precessing vortex;
f = 4.0 kHz.
near the inlet (probes 1-1 and 1-2), whereas high-frequency modes around 4000 Hz
prevail in the downstream region (probe 1-4). The former may be attributed to the
flow displacement effect of the central recirculation zone. The occurrence of the
4000 Hz oscillation at the injector exit can be explained by considering the flow
development along the boundary of the central recirculation zone in Fig. 10.17,
which shows the three-dimensional POD mode shape corresponding to the fre-
quency of 4000 Hz. The existence of precessing vortex motion in the outer region
of the central recirculating flow is revealed. Figure 10.18 shows the characteris-
tic frequencies in different regions of the flowfield. The prevalence of distinct
frequencies in different regions suggests that the flow instability mechanisms
vary in different regions, a phenomenon consistent with Martin and Meiburg’s
expectation.84
where ṁ 0 and f F denote the mean mass flow rate and the forcing frequency,
respectively. The amplitude of the oscillation, α, is fixed at 10%. The forcing
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 233
fsh ~
ear 13k fshear~13k Hz
Hz
fpassage~1.2-1.8k Hz
fPVC ~ 4.0k Hz
recirculation zone
facoustic~11k Hz
fvortical~1.7k Hz
frequency covers a range from 400 through 13,000 Hz, commensurate with the
broadband nature of the injector flow dynamics. Only the higher swirl-number
case with S1 = 45 deg, S2 = −60 deg, and S3 = 70 deg is considered herein.
The vortical and acoustic fields in the injector can be globally characterized
by two frequencies, f v and f a , measuring the convective and acoustic motions,
respectively. The former can be estimated by the mean flow residence time and
has a value of 1.7 kHz. The latter is obtained on the basis of the time required
for a downstream acoustic wave to travel through the injector and has a value of
11 kHz. The phase difference of the traveling acoustic wave between the entrance
and the exit of the injector, θ , is
θ ≈ 2π L/l F = 2π f F / f a (10.6)
where L is the length of the main flow passage and l F is the acoustic wavelength
at the forcing frequency.
a. Instantaneous Flow Structures. Figure 10.19 shows snapshots of the fluc-
tuating vorticity magnitude fields, | |, obtained by subtracting the longtime aver-
aged quantity from its instantaneous value, at various forcing frequencies. When
the frequency is higher than f v , well-defined vortical structures are observed in
the forward section of the injector. These waves, generated by the flow oscilla-
tions at the entrance, are convected downstream with the local flow velocity. The
wavelength is inversely proportional to the forcing frequency and shortens in the
234 Y. HUANG ET AL.
middle region of the injector because of the flow turning effect, that is, the flow
direction turns in this region and the velocity component perpendicular to the
wave front decreases. The intensive turbulent fluctuations downstream of the cen-
terbody overshadow the organized vortical waves, which are eventually damped
out by turbulent diffusion and viscous dissipation. When the forcing frequency is
less than f v , it is difficult to clearly observe organized vortical waves inside the in-
jector because of the long vortical wavelengths associated with the low-frequency
oscillations.
Figure 10.20 shows snapshots of the fluctuating velocity and pressure fields
under external forcing with a frequency of 13 kHz. This case was chosen be-
cause of the presence of a well-established vortical wave, which helps identify
the disturbance propagation mechanisms. The vortical wave is mainly aligned
with the fluctuating azimuthal velocity, whereas the acoustic wave is most closely
related to the pressure oscillation. The imposed excitation at the injector entrance
can be decomposed into two components in the azimuthal and radial directions.
The former generates a vortical wave because of the shear stress resulting from
the flow oscillation in the azimuthal direction, and its dynamics are governed by the
conservation of angular momentum. The latter produces an irrotational, traveling
acoustic wave and can be characterized by the pressure and streamwise-velocity
fluctuations through mass conservation.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 235
10
Distance, mm
5
vortical wave acoustic wave
~40 m/s ~400 m/s
0
vortical wavelength is smaller than its acoustic counterpart by almost one order
of magnitude. Considering the injector dimension and forcing frequency under
consideration, the wavelength of the organized vortical motion is closer to the
large scales in various regions of the injector and are less than the characteristic
length of the main flow passage. Since interactions between flow motions with
similar scales are generally stronger than those with highly disparate scales, a
vortical wave with a frequency higher than f v (i.e., wavelength less than the flow-
passage length) exerts more significant influence on the energy transfer process
involving different scales.
The impressed periodic forcing provides an additional channel to transfer energy
between the mean and turbulent flowfields through organized motions.63, 66, 86, 87
This energy redistribution process is manifested by the presence of vorticity pock-
ets in the flowfield, in which the fluctuating vorticity is greater than a prespecified
threshold value, ||T . Figure 10.19 shows that those pockets with intensive vor-
ticity fluctuation are enhanced at low forcing frequencies (e.g., 500 and 1500 Hz),
but suppressed at high forcing frequencies (e.g., 4000 and 13,000 Hz).
Figure 10.22 shows the evolution of the instantaneous axial velocity field, which
is spatially averaged in the azimuthal direction, within one cycle of oscillation at
a forcing frequency of 1500 Hz. Also included is the time trace of the mass flow
rate at the injector exit, obtained by filtering out turbulent fluctuations. When the
mass flow rate achieves the maximum at t f F = 38.43, a ring structure with strong
positive velocity appears in the downstream region of the second guide vane, where
the mean axial-flow velocity also reaches its maximum. The ring structure then
sheds downstream while the mass flow rate decreases. A new one is produced
when the mass flow rate increases in a new cycle. The strong flow oscillation in
this region can potentially influence the atomization process of the liquid film
accumulated on the surface of the second guide vane. This evolution pattern,
however, cannot be observed at the other forcing frequencies. The discrepancy may
be attributed to the fact that 1500 Hz is closer to the characteristic frequency of flow
Fig. 10.22 Time evolution of axial velocity field within one cycle of oscillation with
forcing frequency of 1500 Hz, spatially averaged in azimuthal direction. Contour levels
between –50 and 100 m/s with increment of 6 m/s. Solid lines represent positive values
and dashed lines represent negative values.
238 Y. HUANG ET AL.
fF = 4000 Hz fF = 13,000 Hz
Fig. 10.23 Effect of forcing frequency on longtime averaged azimuthal velocity field.
Contour levels between –90 and 150 m/s with increment of 10 m/s. Solid lines represent
positive values and dashed lines represent negative values.
convection, f v , than the others studied under this flow configuration. The flow
tends to resonate with the external excitation at this frequency in the streamwise
direction.
b. Mean Flow Properties. Figure 10.23 shows the longtime averaged az-
imuthal velocity fields at f F = 1500, 4000, and 13,000 Hz. No discernible differ-
ence is observed between the flows with and without external excitations except in
the region where the counter-rotating flows through the S1 and S2 swirlers merge
at f F = 13, 000 Hz. The mixing region can be characterized by the line of zero
azimuthal velocity, which shrinks by almost half at this forcing frequency. The
impressed oscillation resonates with the local shear-layer instability (i.e., 13,000
Hz) when the two frequencies match each other and even causes the reversal of the
azimuthal flow direction. The effect of external forcing on flow development can
be further examined in Fig. 10.24, which shows snapshots of the isosurfaces of
the azimuthal velocities at u θ = −2 and 2 m/s in the azimuthal phase space (θ =
0 to 360 deg). The flowfield exhibits a helical structure that originates from the
trailing edge of the first guide vane under conditions without external forcing. The
coherent structure, however, is destroyed by the impressed axisymmetric distur-
bance at the injector entrance and breaks into small bulbs. As discussed previously,
in addition to the external forcing, two mechanisms contribute to this phenomenon:
the Kelvin–Helmholtz instability in the azimuthal direction and centrifugal insta-
bility. Both of them strongly depend on the swirl number. The ensuing enhancement
of flow instability in the azimuthal direction considerably enhances local turbulent
mixing.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 239
a)
b)
It should be noted that in an operational injector, liquid fuel injected from the
centerbody impinges onto the inner surface of the second guide vane and forms a
liquid film, which is then atomized to a spray of fine droplets by the local shear
flow near the rim of the second guide vane. The potential influence of external
forcing on the breakup of the liquid film appears in the two conflicting areas.
On the one hand, the strong fluctuation in the azimuthal direction promotes the
development of an instability wave on the fuel filming surface and the subsequent
atomization process.88 On the other hand, as shown in Fig. 10.23, the external
forcing may significantly modify the mean azimuthal velocity field near the fuel
filming surface, especially when the forcing frequency approaches the shear-layer
characteristic frequency (i.e., 13 kHz). The flow near the downstream part of
the second guide vane even changes its direction from counter-rotating to co-
rotating with the flows in the main and the third (S3 ) passages. This qualitative
switch of flow pattern represents an undesired feature from the perspective of fuel
atomization.23
In spite of the modification of the flowfield between the first and second guide
vanes at f F = 13 kHz, the distribution of the turbulent kinetic energy appears to be
insensitive to external forcing in the bulk of the flowfield. This may be attributed
to the weakness of the excitation as compared with the intrinsic high-intensity
flow motion. The kinetic energy of the periodic motion is considerably smaller
than that of the turbulent motion at the injector outlet. The broadband nature of
the injector flow also discourages the modulation of the mean flow by a single-
harmonic excitation unless the forcing resonates with the local flow structure at
appropriate frequencies.89
c. Acoustic Admittance at Injector Exit. The global response of the injector
can be described by the acoustic admittance at the exit. The information obtained
can be effectively used to serve as the upstream boundary condition for analyzing
the unsteady flow motion in a combustion chamber.90 The admittance function,
also the reciprocal of the impedance function, measures the velocity fluctuation in
response to incident pressure fluctuation. Following common practice, the acoustic
admittance function, Ad , is defined as
û a /ā
Ad ( f ) = (10.7)
p̂a /γ p̄
where p̄ and ā denote the mean pressure and the speed of sound, respectively.
The overhat ( )a represents the Fourier component of the oscillation at the forcing
frequency. Because the background noise in the free-forcing case is too strong
to obtain meaningful results, an external excitation is required to determine the
acoustic admittance at the frequency of concern.
Figure 10.25 shows the radial distributions of the admittance functions at the
injector exit for four different forcing frequencies: 500, 900, 1500, and 4000Hz.
The maximum response occurs at 500 Hz, especially near the rim of the second
guide vane. Excitations at 500, 900, and 1500Hz exhibit the same trend; and the
admittances achieve their maxima when the outer boundary r = R0 is approached.
This outcome may be attributed to the relatively low pressure oscillation and high
velocity fluctuations near the upper boundary. In this region, the pressure response
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 241
1.0
r/R0
0.5 fF (Hz)
500
900
1500
4000
0.0
0 5 10
Magnitude of Ad
1.0
fF (Hz)
500
900
r/R0
0.5 1500
4000
0.0
−1.0π −0.5π 0.0 0.5π 1.0π
Phase of Ad , radian
Fig. 10.25 Radial distributions of acoustic admittance function at injector exit for
different forcing frequencies.
at 500 Hz forcing is less than 300 Pa, which is smaller than its counterparts at other
excitation frequencies (> 1000 Pa). When the oscillation is impressed at 4000 Hz,
the velocity response in the outer region (0.8 < r/R0 < 1.0) becomes very small.
Since the liquid film breaks up in the trailing edge of the second guide vane,
the flow response in this region plays an important role in dictating the dynamic
behavior of the liquid fuel.85 A small pressure oscillation at 500 Hz may result in
a large velocity fluctuation, which consequently exerts a strong influence on spray
formation at that location.
The phase distribution of the admittance function indicates a lag between
the velocity and pressure fluctuations of around 90 deg in the main flow passage
242 Y. HUANG ET AL.
(0.3 < r/R0 < 0.8). The situation is consistent with the behavior of a simple trav-
eling acoustic wave without much influence from shear layers. The phase behavior
for the 4000 Hz case exhibits a trend distinct from that of the other cases, especially
in the central recirculation zone. A major factor contributing to this phenomenon
is the proximity of this forcing frequency to the characteristic frequency of the
central recirculating flow. The imposed axisymmetric excitation in the streamwise
and azimuthal directions does not promote the evolution of the precessing vortex
along the boundary of the central recirculation zone. The pressure and velocity
coupling at 4000 Hz differs from that at other frequencies because of the phase
difference between the oscillations induced by external forcing and intrinsic flow
instabilities.
d. Mass Transfer Function. Another important measure of the injectors dy-
namic response is the transfer function of the total mass flow rate between the
injector entrance and exit, defined as
ṁˆ aex
m ( f ) = (10.8)
ṁˆ in
a
Here ṁˆ a is the Fourier component of the mass flow rate at the forcing frequency,
which is obtained by integrating the mass flux over the entire surface of concern.
Figure 10.26 shows the magnitude and phase of
m as a function of the forcing
frequency. The magnitude reaches its maximum at f F = 1500 Hz, as expected
from the previous results. A large disparity in the fluctuation of the mass flow rate
between the entrance and exit is clearly found. At first glance, this observation
seems to violate the law of mass conservation for such an acoustically compact
2
Magnitude of Πm
0
0
Phase of Πm , radian
−π
−2π
0 5000 10000
Forcing frequency, Hz
Fig. 10.26 Effect of forcing frequency on transfer function of total mass flow rate.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 243
injector. The forcing frequency is much lower than the acoustic characteristic fre-
quency of the injector, f a . Under this condition, the flowfield in the injector can
be treated as incompressible, and the instantaneous total mass flow rate at the en-
trance and exit should be identical. To explore the underlying physical mechanisms
responsible for the phenomenon shown in Fig. 10.26, and to ensure numerical ac-
curacy, the time-averaged mass flow rates at the injector entrance and exit are
calculated. The result confirms the conservation of the overall mass flow rate for
all the forcing frequencies considered herein. The 1500 Hz forcing indeed excites
the flowfield at the expense of suppressing fluctuations at other frequencies. The
mass-flow transfer function for the 4000 Hz forcing is less than unity. In addition to
channeling mass flow among different Fourier components, flow compressibility
takes effect at high-frequency forcing, allowing temporarily for a relatively large
mass variation inside the injector. In short, the forcing frequency affects not only
the spatial distribution of the mass flux fluctuation but also the temporal variation
of the overall mass flow rate.
The phase shift in Fig. 10.26 exhibits a linear distribution with the forcing
frequency because of compressibility effects. This phenomenon can be examined
by using the acoustic characteristic frequency, f a , and the phase difference, θ ,
in Eq. (10.6). The good agreement between the analytic estimation, Eq.(10.6),
and the numerical result further verifies that the oscillation of the mass flow rate
propagates in the form of an acoustic wave.
Fig. 10.27 Schematic of CFM56 gas-turbine swirl cup assembly with radial jet entry.
Fig. 10.28 Contours of normalized mean axial velocity, counter levels between
−0.6 and 2.2, with increment of 0.2. Solid lines represent positive values and dashed
lines represent negative values.
region. The resultant strong shear stress near the wall, especially in the vicinity of
the venturi surface, facilitates atomization of the injected fuel. It is worth noting
that in an operational injector, liquid fuel injected from the nozzle impinges onto
the venturi surface and forms a thin film, which needs to be atomized to a spray
of fine droplets by the local shear flow. The flowfield near the injector wall plays a
crucial role in dictating the liquid-sheet-breakup and droplet-formation processes.
The turbulent kinetic energy field shown in Fig. 10.29 exhibits three different re-
gions with high turbulence intensity. Region 1, formed by the merging of the eight
swirl jets, includes the head-end region near the centerline, where the liquid fuel is
discharged. The vigorous flow motion promotes liquid breakup and atomization.
Region 2 covers the stagnation point of the central recirculating flow. A precessing
vortex core is also observed originating from this region, and it is addressed in
detail in the following section. Both the vortex precession and flow recirculation
enhance local flow oscillations. Region 3 consists of the field surrounding the cen-
tral recirculation zone in the main flow passage. Large coherent structures evolve,
especially when the primary flow merges with the counter-rotating flow through
the secondary swirlers. The strong radial motion accelerates the development of
the Kelvin–Helmholtz instability in the liquid film issuing from the trailing edge
of the venturi. At the same time, the shear force associated with the counter-rotating
flow in the azimuthal direction also enhances the atomization process. Thus, the
flow structure in the present injector provides an effective capability to atomize
the liquid film.
a) b)
c) d)
Fig. 10.31 Time evolution of streamlines and pressure field (time increment of
0.1 ms); the thick dark line indicates the contour of zero axial velocity.
with p = 99.5 kPa, indicating the existence of a PVC structure. The low-pressure
core initially aligns with the axial axis in the region downstream of the fuel nozzle.
It is then driven away from the centerline at the stagnation point of the recirculat-
ing flow and extends downstream spirally against the direction of the main flow
rotation, although the whole structure follows the main flow. Figure 10.31 shows
snapshots of the instantaneous streamlines and pressure fields on a longitudinal
plane at various times. The clustered streamlines indicate that large vortices around
the low-pressure core are pushed outward. Furthermore, the induced low-pressure
core is located outside the region defined by zero axial velocity. The situation is
consistent with the observation of Syred and Beer,93 in that the PVC is located
between the zero axial-velocity and dividing surfaces in the outer region of the
recirculation zone. When the vortex core rotates in the injector, large structures
are peeled off from the spiral core periodically and are convected downstream by
the local flow.
Novak and Sarpkaya94 found that the direction of the spiral winding changed
randomly in their experimental study of turbulent swirling flows at high Reynolds
numbers. Lucca-Negro and O’Doherty92 reviewed the twist direction for the spiral
form of vortex breakdown in detail but drew no definite conclusions. Although the
248 Y. HUANG ET AL.
Fig. 10.32 First two POD modes of normalized fluctuating pressure field.
vortex precession at low Reynolds numbers is not identical to the spiral form of
vortex breakdown, the precessing process can be considered as periodical spiral
motion in the conical form of turbulent vortex breakdown observed by Novak and
Sarpkaya.94
Figure 10.32 shows the first two POD mode shapes of the fluctuating pressure
field, which account for more than 30% of the total energy of the field. The mode
shapes and associated time-varying coefficients demonstrate the presence of spiral
rotating motion (i.e., vortex breakdown) at a frequency of 1266 Hz. This result
further confirms that vortex precession is a dominant mechanism in the cold flow
evolution.
U(Ux)
1
y/D0
−1
1 2 3 4 5 6
x/D0
V(Ur)
1
y/D0
−1
1 2 3 4 5 6
x/D0
W(Uθ)
1
y/D0
−1
1 2 3 4 5 6
x/D0
Fig. 10.33 Radial distributions of mean-flow axial, radial, and azimuthal velocity
components (symbol: co-rotation; line: counter-rotation).
250 Y. HUANG ET AL.
235 mm
38 mm
Fig. 10.34 Schematic of a model swirl-stabilized gas-turbine combustor (after
Broda et al., 1998, Ref. 14).
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 251
ϕ
Dn Dh
c
A'
Fig. 10.35 Schematic of top and cross-section views of a flat vane swirler
(after Seo, 1999, Ref. 95).
a) 15.0
p'rms /Pc
0.0
600 620 640 660 680 700
inlet air temperature, T in (K)
15.0
(b) 15.0
Pc = 0.286 MPa
12.5 Pc = 0.463 MPa
Pc = 0.638 MPa
10.0
p'rms/Pc (%)
7.5
7.5
(%)
c
/P p'
rms
5.0
2.5
0.0
0.0
0.50
0.50 0.55 0.60
0.60 0.65 0.7
0.70 0.75 0.80
0.80
equivalence
Overall ratio,
Equivalence φφο
Ratio,
Fig. 10.36 Stability maps as function of inlet air temperature and equivalence ratio:
a) Pc = 0.45 MPa, φ = 0.573; b) Tin = 669 K (after Seo, 1999, Ref. 95).
was then investigated by increasing the inlet temperature from 600 to 660 K. The
mean temperature contours and pseudo streamlines on the x–r plane, shown in
Fig. 10.38, are based on the mean axial and radial velocity components for a stable
flame. A CRTZ is established in the wake of the centerbody under the effects
of the swirling flow. The CTRZ, a form of vortex breakdown, serves as a flame
stabilization region, where hot products mix with the incoming mixture of air and
fuel. In addition, as a result of the sudden increase in combustor area, a CRZ is
formed downstream of the backward-facing step.
The calculated pressure and velocity fields exhibit small-amplitude fluctuations,
with a dominant harmonic mode at 3214 Hz, corresponding to the frequency of the
vortex shedding from the centerbody. Figure 10.39 presents the flame evolution
and vortex shedding process in the upstream region of the chamber over one cycle
of oscillation. The pressure and velocity are measured at the middle point of the
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 253
T in = 570 K
T in = 660 K
20
15
10
p' (kPa)
−5
−10
−15
−20
0 10 20
time (ms)
Fig. 10.37 Top: photographic images of stable and unstable flames; bottom: pressure-
time trace, Pc = 0.483 MPa, φ = 0.573 (after Seo, 1999).
254 Y. HUANG ET AL.
Fig. 10.38 Mean temperature contours and streamlines of stable flame for S = 0.76.
θ
t
θ
t
θ
t
θ
t
140 470
u p
u (m/s)
p (kpa)
120 460
100 450
inlet annulus exit. The phase angle θ is referenced with respect to the acous-
tic velocity at the interface between the inlet and combustor. The entire process
is dictated by the temporal evolution and spatial distribution of the flame front,
which moves back and forth under the influences of the vortical motion (indicated
by the concentrated streamlines) in the chamber. A new vortex begins to shed
from the center body at θ = 90 deg, accompanied by a higher local flow velocity.
As the vortex moves downstream (θ = 180 deg −270 deg), it distorts the flame
front or even produces a separated flame pocket. At the same time, the higher-
speed mixture pushes the flame downstream. When the vortex moves away from
the flame (θ = 360 deg) and dissipates into small-scale structures, the flame front
propagates upstream (since the higher-speed mixture is convected downstream)
and interacts with another incoming vortex. During this process, a new vortex
appears at the corner of the centerbody and the cycle repeats.
a)
b)
c)
d)
e)
f)
g)
h)
Fig. 10.40 Transition from stable to unstable flame with increased inlet temperature
from 600 to 660 K.
In the vortex-flashback process, as shown in Figs. 10.40f–10.40h, the flame
propagates upstream under the influence of the vortical motion. A counterclock-
wise rotating vortex originally shed from the edge of the backward-facing step
approaches the flame front in the corner recirculation zone and then pushes it
toward the dump plane. At the same time, a small flame pocket is produced
and separated from the main stream. After this vortex is convected downstream
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 257
and passes through the flame, another vortex approaches and interacts with the
flame. The process continues, and eventually the fresh reactants in the corner
recirculation zone are completely burnt. The flame is stabilized by both the corner-
recirculating flow and the center-recirculating flow, and its overall length is sub-
stantially reduced. This situation renders the combustor more prone to instabil-
ities according to the Rayleigh criterion,96 since considerable heat is released
within a short distance close to the chamber head end (i.e., the acoustic antinode
point).
Once the flame becomes unstable (oscillatory) when the inlet flow temperature
exceeds the critical value Tin∗ , it becomes difficult to reestablish stable operation
unless the inlet temperature is reduced to a level significantly lower than Tin∗ .
This phenomenon is commonly referred to as hysteresis and has been experi-
mentally observed by many researchers.14, 64 The occurrence of hysteresis under
the current circumstance may be explained as follows. During unstable com-
bustion, the corner recirculation zone is filled with high-temperature products,
and the chamber wall in this region is heated to reach the local flame tempera-
ture. To recover stable operation, the cold flow needs not only to extinguish the
flame stabilized by the corner recirculating flow through entrainment or flame
liftoff but also to offset the effects of the high-temperature wall, which tends
to increase the local gas temperature and inhibit extinction and near-wall flash-
back. Consequently, a much lower inlet temperature is required to regain stable
operation.
In light of the preceding observations, we conclude that the flashback phe-
nomenon dictates the flame bifurcation process. Flashback in premixed combus-
tion has been the subject of a number of experimental, analytical, and numerical
studies in the past. Its occurrence is usually attributed to two mechanisms. The
first involves flame propagation in the boundary layer along a solid wall, where
the local velocity diminishes toward the surface. The second mechanism is asso-
ciated with flow reversal, which is usually caused by vortical motions or acoustic
oscillations. Both mechanisms are observed in the present case.
For lean-premixed combustion, the laminar flame speed SL increases with an
increase in the equivalence ratio φ. Thus, increases in the equivalence ratio and in-
let temperature exert similar effects on the flame evolution. However, the chemical
reaction rate and heat release are much more sensitive to variations in the equiv-
alence ratio under lean conditions than under stoichiometric conditions. More-
over, near the lean blowout limit, perturbations in the equivalence ratio φ can
cause periodic extinction of the flame. As a result, the equivalence ratio oscil-
lation under lean conditions is prone to inducing flow oscillation97 and subse-
quently increases turbulent velocity fluctuation v . This outcome suggests that a
lean premixed turbulent flame is more susceptible to flashback, since the turbu-
lent flame speed ST increases not only with the laminar flame speed SL , but also
with turbulent velocity fluctuation v .98 The result helps explain why the tran-
sition from a stable to an unstable state, as described in Refs. 14 and 95, only
occurs when the equivalence ratio falls in the range between 0.5 and 0.7. Since
the flame bifurcation is largely determined by the flashback phenomenon in the
corner recirculation zone in the present case, one effective way to avoid its oc-
currence is to inject cold flow into that region. This procedure suppresses the
upstream propagation of local flame and consequently leads to a much more stable
system.
258 Y. HUANG ET AL.
Fig. 10.41 Mean temperature fields and streamline patterns for S = 0.44, 0.76,
and 1.10.
the centerbody and the backward-facing step, where large velocity fluctuations
are produced because of the strong turbulent mixing in the shear layers between
the incoming flow and the recirculation flows. The evolution of turbulent kinetic
energy is governed by the following equation:
D u i2 /2 /Dt = ∂ − u j p/ρ0 − u i2 u j /2 + 2vu i ei j /∂ x j
− u i u j ∂Ui /∂ x j − 2vei j ei j (10.9)
x = 60 mm
x = 40 mm
x = 80 mm
x = 25 mm
x = 30 mm
x = 50 mm
2
r (cm)
1
0
0 100 0 100 0 100 0 100 0 100 0 100
u x (m/s)
x = 25 mm
x = 40 mm
x = 80 mm
x = 60 mm
x = 50 mm
x = 30 mm
2
r (cm)
0
0 100 0 100 0 100 0 100 0 100 0 100
u r (m/s)
x = 40 mm
x = 30 mm
x = 50 mm
x = 80 mm
x = 60 mm
x = 25 mm
2
r (cm)
0
0 100 0 100 0 100 0 100 0 100 0 100
u θ (m/s)
x = 80 mm
x = 30 mm
x = 60 mm
x = 40 mm
x = 25 mm
x = 50 mm
2
r (cm)
0
0 50 0 50 0 50 0 50 0 50 0 50
√ k (m/s)
Fig. 10.42 Radial distributions of mean velocity components and turbulent kinetic
energy at various axial locations for S = 0.44 and 1.10.
Consequently, much stronger turbulent kinetic energy is observed for the high
swirl-number case with S = 1.10.
2. Instantaneous Flowfield
Vorticity is of concern because of its dominant influence in determining the
flow entrainment in the reaction zone and the subsequent flame evolution. Fig-
ure 10.43 shows snapshots of the vorticity magnitude field on an x–r plane. For
the low swirl-number case with S = 0.44, large vortical structures, arising from
the shear layers downstream of the dump plane and centerbody, are convected
downstream and then dissipated into small-scale eddies. The same phenomenon is
observed for the high swirl-number case, in which well-organized vortices are shed
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 261
Fig. 10.43 Snapshots of vorticity magnitude field on x–r and x–θ planes for S = 0.44
and 1.10.
from the edge of the backward-facing step. The vortex motions downstream of
the centerbody, however, become quite disordered, because of the presence of
the strong central recirculating flow. In both cases, the vortex shedding frequen-
cies are close to that of the first tangential mode of the acoustic wave in the
chamber.
Figure 10.44 shows snapshots of the isovorticity surface at ω =75,000 1/s.
The flowfield in the region r > 2 cm is blanked to provide a clear picture of the
vortex structures. For the low swirl-number case with S = 0.44, a vortex spiral
evolves from the shear layer originating at the backward-facing step, because of
the Kelvin–Helmholtz instabilities in both the axial and azimuthal directions. This
vortical structure gyrates around the centerline and persists for several turns before
breaking up into small fragments. For the high swirl-number case with S =1.10, a
spiral vortex structure can also be observed. The structure, however, is much more
complex because of the high centrifugal force. It spreads outward rapidly and soon
breaks up into small-scale structures.
The evolution of these spiral vortex structures can be regarded as a kind of
vortex shedding process with well-defined frequencies, as previously described.
262 Y. HUANG ET AL.
Fig. 10.44 Snapshots of isovorticity surface at ω = 75, 000 s−1 (left: r > 0.02 m is
blanked, right: r > 0.01 m is blanked) for S = 0.44 and 1.10.
One may conjecture55 that the vortical motions in the shear layers resonate with
acoustic oscillations in the chamber. In the present configuration, two shear layers
exist downstream of the rear-facing step and the centerbody. The axial momen-
tum thickness, θ0 , of each shear layer is estimated to be around 0.1 mm for the
low swirl-number case based on the calculated mean velocity distribution. For
the high swirl-number case, the momentum thickness of the inner shear layer
(≈0.25 mm) differs from its outer counterpart (≈0.05 mm) by a factor of five, be-
cause of its stronger swirl strength. A linear stability analysis100 has been carried
out to provide more insight into the shear-layer instability phenomena for annular
swirling flows in an open atmosphere. The geometric parameters were selected
to match the current physical model. The effects of momentum thickness, swirl
strength, and density and velocity ratios were studied systematically with different
azimuthal wave numbers. The predicted most-amplified frequencies are different
for the two swirl numbers considered herein, mainly because of the disparity of
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 263
the axial momentum thickness between the two cases. In addition, the predicted
values are much higher than the vortex-shedding frequency, which corresponds to
the frequency of the first tangential (1T) mode of acoustic motion in the present
chamber. This observation indicates that the acoustic oscillation acts as a forced
excitation to the system. The shear layers respond to the excitation by locking their
shedding frequencies close to the forcing frequency.
Q̇ = ρu h 0f ST A (10.10)
where ρu is the unburnt gas density, ST is the subgrid turbulent flame speed,
h 0f
is the heat of reaction, and A is the total filtered flame surface area.
Figure 10.45 shows the power spectral densities of the total filtered flame
surface-area and heat-release fluctuations. At S = 0.44, a dominant mode exists
400 S = 0.44 flame surface area 400 S = 1.10 flame surface area
Amplitude, mm2/Hz
Amplitude, mm2/Hz
1761
300 300
100 100
11,712
0 0
0 10 20 0 10 20
Amplitude, kJ/s/Hz
1761
6 6
4 3320 4
11,712
2 20,532 2
0 0
0 10 20 0 10 20
Frequency, kHz Frequency, kHz
Fig. 10.45 Power spectral densities of total flame surface-area and heat-release fluc-
tuations for S = 0.44 and 1.10.
264 Y. HUANG ET AL.
Fig. 10.46 Temporal evolution of temperature field over one cycle of 1L mode of
oscillation for S = 0.44.
resemblance to that of flame surface-area variations. A small spike near the fre-
quency of 20,532 Hz, however, is observed for S = 0.44, which arises from the
fluctuations in the subgrid turbulent flame speed ST .55 In light of the preceding
observations, one can conclude that low-frequency acoustic perturbations exert a
strong influence on the fluctuations of the total flame surface area and heat release.
In contrast, high-frequency acoustic oscillations travel through the flame zone
without significantly affecting the flame surface area and heat-release variations,
although they may impose a significant impact on the local flame propagation. The
results qualitatively agree well with the prediction from a companion analytical
analysis of flame response.101 The calculated mean flame surface area and the root
mean square of the fluctuating quantity for the high swirl-number case are much
smaller than those of the low swirl-number case. However, owing to the increased
turbulence intensity and the ensuing enhancement of the flame speed in the high
swirl-number case, the mean heat-release rate and the associated fluctuation are
very close in these two cases.
Fig. 10.47 Time histories of pressure immediately downstream of the dump plane
(top), flame surface area (middle), and heat release rate (bottom) for S = 0.44; the
thick black lines represent the extracted 1L oscillations.
266 Y. HUANG ET AL.
pressure
250
500 4000
heat release, kJ
400 flame surface area
p, kpa
3000
200
100
23 24 25
Time, ms
Fig. 10.48 Time histories of pressure immediately downstream of the edge of the
centerbody (top), flame surface-area (middle), and heat release rate (bottom) for
S = 1.10.
surface-area variation can be elucidated by considering its interaction with the local
oscillatory flowfield. It lags behind the pressure oscillation by 76 deg. During the
period from θ = −166 deg (t = 24.09 ms) to 14 deg (t = 24.38 ms), a relatively
lower pressure field exists near the dump plane, facilitating the delivery of the
fresh reactants into the chamber. Intensive heat release then occurs after a short
fluid-mixing and chemical-induction time. The resultant flow expansion pushes
the flame outward and causes the flame surface area to increase from a trough
to a crest. Unburned mixture fragments may be shattered away from the main
stream and may generate local hot spots when convected downstream. During the
period from θ = 14 deg (t = 24.38 ms) to 194 deg (t = 24.66 ms), the relatively
higher pressure near the dump plane prevents the fresh reactants from traveling
downstream into the chamber. The flame zone is thus reduced and becomes a lit-
tle more compact. The same process then repeats for another cycle of oscillation.
Figure 10.47 also indicates that the heat-release and flame-surface-area fluctuations
are nearly in phase. The former only lags behind the flame surface-area oscilla-
tion by 4 degrees. For the high swirl-number case with S = 1.10, no obvious 1L
oscillation can be observed, as shown in Fig. 10.48.
Figure 10.49 presents the temporal evolution of the temperature field in the
upstream section of the chamber on the x–r plane over one cycle of the 1T mode
of acoustic oscillation. As the swirl number increases, the flame anchored by
the center recirculating flow may propagate upstream periodically and cause flame
flashback. Two mechanisms, as previously stated, have been identified for the
occurrence of flame flashback. In this case, the flashback is closely linked to
the strong reverse flow in the center recirculation zone. The swirl strength is so
strong that it sometimes causes the center recirculating flow to enter the inlet annu-
lus. As a consequence, the flame attached to the centerbody travels upstream and
flashback occurs.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 267
Fig. 10.49 Temporal evolution of temperature field on an x–r plane over one cycle of
1T mode of oscillation for S = 0.44 and 1.10.
Fig. 10.50 Vortex and flame front interaction over one cycle of 1T mode of oscillation
for the case S = 0.76.
IV. Conclusion
In this chapter, the flow and flame dynamics of several different types of swirl-
injectors for contemporary gas turbine engines were studied. Both axial- and radial-
entry configurations were explored over a wide range of flow conditions. The
effects of single vs multiple swirlers and co-rotating vs counter-rotating arrange-
ments were also examined systematically. Most of the results presented here were
270 Y. HUANG ET AL.
obtained from large eddy simulations of detailed flow structures under conditions
with and without external forcing. Where appropriate, experimental observations
are also given for the sake of completeness.
For cold-flow characterization of injector dynamics, various fundamental mech-
anisms dictating the flow evolution in swirl injectors were identified. These include
vortex breakdown; precession of vortex core; and Kelvin–Helmholtz, helical and
centrifugal instabilities. The injector response to external forcing was studied in
terms of the acoustic admittance and mass transfer functions over a broad range of
frequencies. Low-frequency excitations generally promote flow fluctuations, but
the trend is reversed for high-frequency excitations. The fluctuation of the mass
flow rate of a given frequency component at the injector exit may reach a magnitude
substantially greater than that at the entrance when the forcing resonates with the
injector flow. Results of this kind can be effectively used in analyzing combustion
instabilities in gas-turbine engines. The influence of swirler orientation on injector
flow development was also examined by considering both counter- and co-rotating
inflow conditions. The counter-rotating configuration is more desirable because
of its reduced central recirculation zone and stronger shear stress and turbulence
intensity in regions where liquid fuel atomization occurs.
The second part of the chapter deals with the flame dynamics of swirl injectors.
Detailed flow structures and flame evolution were investigated under various oper-
ating conditions. The inlet air temperature and equivalence ratio were found to be
the two key parameters determining the stability characteristics of a lean-premixed
swirl-stabilized combustor. A slight increase in the inlet airflow temperature across
the stability boundary leads to a sudden increase in the chamber flow oscillation. Fi-
nally, the underlying mechanisms responsible for driving combustion oscillations
and the transition from a stable to an unstable flame were identified.
Acknowledgments
The work reported in this chapter was sponsored in part by the Office of Naval
Research under Grant No. N00014-96-1-0405, in part by the NASA Glenn Re-
search Center under Grant NAG 3-2151, and in part by the Air Force Office of
Scientific Research under Grant No. F49620-99-0290. The support and encourage-
ment provided by Gabriel Roy and Kevin Breisacher are gratefully acknowledged.
References
1
Lefebvre, A. H., “The Role of Fuel Preparation in Low-emission Combustion,” Journal
of Engineering for Gas Turbines and Power, Vol. 117, 1995, pp. 617–654.
2
Bazarov, V., Yang, V., and Puri, P., “Design and Dynamics of Jet and Swirl Injectors,”
Chapter 2, Liquid Rocket Thrust Chambers: Aspects of Modeling, Analysis, and Design,
edited by V., Yang, M. Habiballah, J. Hulka, and M., Popp, Progress in Astronautics and
Aeronautics, Vol. 200, AIAA, Reston VA, 2004, pp. 19–103.
3
Lefebvre, A. H., Gas Turbine Combustion, Taylor & Francis, Philadelphia, 1998.
4
Wang, H., McDonell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study of
a Model Gas Turbine Combustor Swirl Cup, Part I: Two-Phase Characterization,” Journal
of Propulsion and Power, Vol. 10, No. 4, 1994, pp. 441–445.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 271
5
Wang, H., McDonell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study
of a Model Gas Turbine Combustor Swirl Cup, Part II: Droplet Dynamics,” Journal of
Propulsion and Power, Vol. 10, No. 4, 1994, pp. 446–452.
6
Fu, Y., Cai, J., Elkady, M., Jeng, S. M., and Mongia, H. C., “Fuel and Equivalence Ratio
Effects on Spray Combustion of a Counter-Rotating Swirler,” AIAA Paper 2005-0354,
2005.
7
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Air Temperature Effects on Non-Reacting
Spray Characteristics Issued from a Counter-Rotating Swirler,” AIAA Paper 2005-0355,
2005.
8
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Fluid Property Effects on Non-Reacting
Spray Characteristics Issued from a Counter-Rotating Swirler,” AIAA Paper 2005-0356,
2005.
9
Cowell, L. H., and Smith, K. O., “Development of a Liquid-Fueled, Lean-Premixed Gas
Turbine Combustor,” Journal of Engineering for Gas Turbines and Power, Vol. 115, 1993,
pp. 554–562.
10
Snyder, T. S., Rosfjord, T. J., McVey, J. B., and Chiappetta, L. M., “Comparison of
Liquid/Air Mixing and NOx Emission for a Tangential Entry Nozzle,” American Society
of Mechanical Engineers, Paper 94-GT-283, 1994.
11
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., “Active Control of
Combustion Instability in a Liquid-Fueled Low-NOx Combustor,” Journal of Engineering
for Gas Turbines and Power, Vol. 121, 1999, pp. 281–284.
12
Richards, G. A., Gemmen, R. S., and Yip, M. J., “A Test Device for Premixed Gas
Turbine Combustion Oscillations,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, 1998, pp. 294–302.
13
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premixed
Gas Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 119,
1997, pp. 776–782.
14
Broda, J. C., Seo, S., Santoro, R. J., Shirhattikar, G., and Yang, V., “An Experimental
Study of Combustion Dynamics of a Premixed Swirl Injector,” Proceedings of the Com-
bustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1849–1856.
15
Lee, S. Y., Seo, S., Broda, J. C., Pal, S., and Santoro, R. J., “An Experimental Estimation
of Mean Reaction Rate and Flame Structure During Combustion Instability in a Lean
Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute, Pittsburgh,
PA, Vol. 28, 2000, pp. 775–782.
16
Venkataraman, K. K., Preston, L. H., Simons, D. W., Lee, B. J., Lee, J. G., and
Santavicca, D. A., “Mechanism of Combustion Instability in a Lean Premixed Dump Com-
bustor,” Journal of Propulsion and Power, Vol. 15, No. 6, 1999, pp. 909–918.
17
Lee, J. G., Kim, K., and Santavicca, D. A., “Measurement of Equivalence Ratio Fluc-
tuation and Its Effect on Heat Release During Unstable Combustion,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 415–421.
18
Lee, J. G., and Santavicca, D. A., “Experimental Diagnostics for the Study of Combus-
tion Instabilities in Lean-Premixed Combustors,” Journal of Propulsion and Power, Vol.
19, No. 5, 2003, pp. 735–749.
19
Mordaunt, C., Brossard, C., Lee, S. Y., and Santoro, R. J., “Combustion Instability
Studies in a High-Pressure Lean-Premixed Model Combustor Under Liquid Fuel Opera-
tion,” Proceedings of International Joint Power Generation Conference, JPG C2001/FACT-
19089, June 2001.
272 Y. HUANG ET AL.
20
Cohen, J. M., and Rosfjord, T. J., “Spray Pattern at High Pressure,” Journal of Propul-
sion and Power, Vol. 7, No. 4, 1991, pp. 1481–1487.
21
Cohen, J. M., and Rosfjord, T. J., “Influence on the Sprays Formed by High-Shear
Fuel-Nozzle/Swirler Assemblies,” Journal of Propulsion and Power, Vol. 9, No. 1, 1993,
pp. 16–27.
22
Chin, J. S., Rizk, N. K., Razdan, M. K., “Study on Hybrid Airblast Atomization,”
Journal of Propulsion and Power, Vol. 15, No. 2, 1999, pp. 241–247.
23
Chin, J., Rizk, N., and Razdan, M., “Effect of Inner and Outer Airflow Characteristics on
High Liquid Pressure Prefilming Airblast Atomization,” Journal of Propulsion and Power,
Vol. 16, 2000, pp. 297–301.
24
Hardalupas, Y., and Whitelaw, J. H., “Interaction Between Sprays from Multiple Coax-
ial Airblast Atomizers,” Journal of Engineering for Gas Turbines and Power, Vol. 118,
1996, pp. 762–771.
25
Hardalupas, Y., Pantelides, K., and Whitelaw, J. H., “Particle Tracking Velocimetry in
Swirl-Stabilized Burners,” Experiments in Fluids, [Suppl.], Vol. 29, 2000, pp. S220–S226.
26
Presser, C., Gupta, A. K., and Semerjian, H. G., “Aerodynamic Characteristics of
Swirling Spray Flame: Pressure-Jet Atomizer,” Combustion and Flame, Vol. 92, 1993,
pp. 25–44.
27
Presser, C., Gupta, A. K., Semerjian, H. G., and Avedisian, C. T., “Droplet Transport in
a Swirl-Stabilized Spray Flames,” Journal of Propulsion and Power, Vol. 10, No. 5, 1994,
pp. 631–638.
28
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Thermoacoustic In-
stabilities and Emissions in an Industrial-Type Gas Turbine Combustor,” Proceedings of
the Combustion Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1817–1824.
29
Paschereit, C. O., and Gutmark, E., “Proportional Control of Combustion Instabilities
in a Simulated Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, 2002,
pp. 1298–1304.
30
Acharya, S., Murugappan, S., O’Donnel, M., and Gutmark, E. J., “Characteristics and
Control of Combustion Instabilities in a Swirl-Stabilized Spray Combustor,” Journal of
Propulsion and Power, Vol. 19, No. 3, 2003, pp. 484–496.
31
Uhm, J. H., Acharya, S., “Control of Combustion Instability with a High-Momentum
Air-Jet,” Combustion and Flame, Vol. 139, 2004, pp. 106–125.
32
Murugappan, S., Acharya, S., Allgood, D. C., Park, S., Anaswamy, A. M., and Ghoniem,
A. F., “Optimal Control of a Swirl-Stabilized Spray Combustion Using System Identification
Approach,” Combustion Science and Technology, Vol. 175, 2003, pp. 55–81.
33
Lee, J. G., Kim, K., Santavicca, D. A., “Effect of Injection Location on the Effective-
ness of an Active Control System Using Secondary Fuel Injection,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 739–746.
34
Richards, G. A., Janus, M., and Robey, E. H., “Control of Flame Oscillation with
Equivalence Ratio Modulation,” Journal of Propulsion and Power, Vol. 15, No. 2, 1999,
pp. 232–240.
35
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive Control of Combustion
Dynamics in Stationary Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5,
2003, pp. 795–809.
36
Sattinger, S. S., Neumeier Y., Nabi, A., Zinn, B. T., Amos, D. J., and Darling,
D. D., “Sub-Scale Demonstration of the Active Feedback Control of Gas-Turbine Com-
bustion Instabilities,” Journal of Engineering for Gas Turbines and Power, Vol. 122, 2000,
pp. 262–268.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 273
37
Johnson, C. E., Neumeier, Y., Neumaier, M., Zinn, B. T., Darling, D. D., and Sattinger,
S. S., “Demonstration of Active Control of Combustion Instabilities on a Full-Scale Gas
Turbine Combustor,” ASME 2001-GT-519, 2001.
38
Conrad, T., Bibik, A., Shcherbik, D., Lubarsky, E., and Zinn, B. T., “Slow Control of
Combustion Instabilities by Fuel Spray Modification Using Smart Fuel Injector,” AIAA
Paper 2004-1034, 2004.
39
Conrad, T., Bibik, A., Shcherbik, D., Lubarsky, E., and Zinn, B. T., “Control of In-
stabilities in Liquid Fueled Combustor by Modification of the Reaction Zone Using Smart
Fuel Injector,” AIAA Paper 2004-4029, 2004.
40
Bernier, D., Lacas, F., and Candel, S., “Instability Mechanisms in a Premixed Pre-
vaporized Combustor,” Journal of Propulsion and Power, Vol. 20, No. 4, 2004, pp. 648–
656.
41
Li, G., and Gutmark, E. J., “Effect of Exhaust Nozzle Geometry on Combustor Flow
Field and Combustion Characteristics,” Proceedings of the Combustion Institute, Pittsburgh,
PA, Vol. 30, 2004.
42
Brewster, B. S., Cannon, S. M., Farmer, J. R., and Meng, F., “Modeling of Lean Pre-
mixed Combustion in Stationary Gas Turbines,” Progress in Energy and Combustion Sci-
ence, Vol. 25, 1999, pp. 353–385.
43
Mongia, H. C., Al-Roub, M., Danis, A., Elliott-Lewis, D., Jeng, S. M., John, A.,
McDonell, V. G., Samuelsen, G. S., and Vise, S., “Swirl Cup Modeling, Part I,” AIAA
Paper 2001-3576, 2001.
44
Hsiao, G., Mongia, H. C., Vij, A., “Swirl Cup Modeling Part II: Inlet Boundary Con-
ditions,” AIAA Paper 2001-1350, 2001.
45
Hsiao, G., and Mongia, H. C., “Swirl Cup Modeling Part III: Grid Independent Solution
with Different Turbulence Models,” AIAA Paper 2001-1349, 2001.
46
Cai, J., Fu, Y., Jeng, S. and Mongia, H. C., “Swirl Cup Modeling Part IV: Effect of
Confinement on Flow Characteristics,” AIAA Paper 2003-0486, 2003.
47
Wang, S., Yang, V., and Mongia, H. C., “Swirl Cup Modeling Part V: Large Eddy
Simulation of Cold Flow,” AIAA Paper 2001-0485, 2003.
48
Giridharan, M. G., Mongia, H. C., and Singh, G., “Swirl Cup Modeling Part VI: Dilution
Jet Modeling,” AIAA Paper 2003-1203, 2003.
49
Stevens, E. J., Held, T. J., and Mongia, H. C., “Swirl Cup Modeling Part VII: Pre-
mixed Laminar Flamelet Model Validation and Simulation of a Single-Cup Combustor
with Gaseous N-Heptane,” AIAA Paper 2003-0488, 2003.
50
Giridharan, M. G., Mongia, H. C., and Jeng, S. M., “Swirl Cup Modeling Part VIII:
Spary Combustion in CFM56 Single Cup Flame Tube,” AIAA Paper 2003-0319, 2003.
51
Jeng, S. M., Flohre, N. M., and Mongia, H. C., “Swirl Cup Modeling Part IX: Atom-
ization,” AIAA Paper 2004-0137, 2004.
52
James, S., Zhu, J., and Anand, M. S., “Large-Eddy Simulation of Gas Turbine Com-
bustors,” AIAA Paper 2004-0552, 2004.
53
Kim, W. W., and Syed, S., “Large-Eddy Simulation Needs for Gas Turbine Combustor
Design,” AIAA Paper 2004-0331, 2004.
54
Kim, W. W., Lienau, J. J., Van Slooten, P. R., Colket, M. B., Malecki, R. E., and Syed, S.,
“Towards Modeling Lean Blow Out in Gas Turbine Flameholder Applications,” American
Society of Mechanical Engineers, 2004-GT-53967, 2004.
55
Huang, Y., Sung, H. G., Hsieh, S. Y. and Yang, V., “Large Eddy Simulation of Com-
bustion Dynamics of Lean-Premixed Swirl-Stabilized Combustor,” Journal of Propulsion
and Power, Vol. 19, 2003, pp. 782–794.
274 Y. HUANG ET AL.
56
Stone, C., and Menon, S., “Swirl Control of Combustion Instabilities in a Gas Tur-
bine Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002,
pp. 155–160.
57
Stone, C., and Menon, S., “Open Loop Control Combustion Instabilities in a Model
Gas Turbine Combustor,” Journal of Turbulence, Vol. 4, 020, 2003.
58
Pierce, C. D., and Moin, P., “Progress-Variable Approach for Large-Eddy Simulation
of Non-Premixed Turbulent Combustion,” Journal of Fluid Mechanics, Vol. 504, 2004,
pp. 73–97.
59
Selle, L., Lartigue, G., Poinsot, T., Koch, R., Schildmacher, K. U., Kerbs, W., Kaufmann,
P., and Veynante, D., “Compressible Large Eddy Simulation of Turbulent Combustion in
Complex Geometry on Unstructured Meshes,” Combustion and Flame, Vol. 137, 2004,
pp. 489–505.
60
Grinstein, F. F., and Fureby, C., “LES Studies of the Flow in a Swirl Combustor,”
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 30, 2005, pp. 1791–1798.
61
Sommerer, Y., Galley, D., Poinsot, T., Ducruix, S. Lacas, F., and Veynante, D., “Large
Eddy Simulation and Experimental Study of Flashback and Blow-Off in a Lean Partially
Premixed Swirled Burner,” Journal of Turbulence, Vol. 5, 037, 2004.
62
Wang, S. W., Hsieh, S. Y., and Yang, V., “Unsteady Flow Evolution in Swirl Injector
with Radial Entry, Part 1: Stationary Conditions,” Physics of Fluids, Vol. 17, No. 4, 2005,
p. 045106.
63
Wang, S. W., and Yang, V., “Unsteady Flow Evolution in Swirl Injector with Radial
Entry, Part 2: External Excitations,” Physics of Fluids, Vol. 17, No. 4, 2005, p. 045107.
64
Huang, Y., and Yang, V., “Bifurcation of Flame Structure in a Lean-Premixed Swirl-
Stabilized Combustor: Transition from Stable to Unstable Flame,” Combustion and Flame,
Vol. 136, 2004, pp. 383–389.
65
Huang, Y., and Yang, V., “Effect of Swirl on Combustion Dynamics in a Lean-Premixed
Swirl-Stabilized Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 30, 2005, pp. 1775–1782.
66
Huang, Y., Wang, S., and Yang, V., “A Systematic Analysis of Combustion Dynamics
in a Lean-Premixed Swirl-Stabilized Combustor,” AIAA Journal (accepted for publication),
2005.
67
Oefelein, J. C., and Yang, V., “Simulation of High-Pressure Spray Field Dynamics,”
Chapter 11, Recent Advances in Spray Combustion, Vol. 2, edited by K. K. Kuo Progress
in Astronautics and Aeronautics, Vol. 171, AIAA, Reston, VA, 1996, pp. 263–304.
68
Chiu, H. H., and Oefelein, J. C., “Modeling Liquid-Propellant Spray Combustion Pro-
cesses,” Chapter 6, Liquid Rocket Thrust Chambers: Aspects of Modeling, Analysis, and
Design, Vol. 200, edited by V. Yang, M., Habiballah, J., Hulka, and M., Popp, Progress in
Astronautics and Aeronautics, AIAA, Reston, VA, 2004, pp. 251–293.
69
Caraeni, D., Bergstrom, C., and Fuchs, L., “Modeling of Liquid Fuel Injection, Evapo-
ration and Mixing in a Gas Turbine Burner Using Large Eddy Simulation,” Flow, Turbulence
and Combustion, Vol. 65, 2000, pp. 223–244.
70
Sankaran, V., and Menon, S., “LES of Spray Combustion in Swirling Flows,” Journal
of Turbulence, Vol. 3, 011, 2001.
71
Apte, S. V., Mahesh, K., Moin P., and Oefelein, J. C., “Large-Eddy Simulation of
Swirling Particle-Laden Flows in a Coaxial-Jet Combustor,” International Journal of Mul-
tiphase Flow, Vol. 29, 2003, pp. 1311–1333.
72
Wang, S., Yang, V. and Koo, J. Y., “Large-Eddy Simulation of Spray-Field Dynamics
in Cross Flows,” AIAA Paper 2005-0729, 2005.
FLOW AND FLAME DYNAMICS OF SWIRL INJECTORS 275
73
Menon, S., “Acoustic-Vortex-Flame Interactions in Gas Turbines,” Combustion Insta-
bilities in Gas Turbine Engines: Operational Experience, Fundamental Mechanisms, and
Modeling, edited by T. Lieuwen and V. Yang, Progress in Astronautics and Aeronautics,
AIAA, Reston, VA, 2005, pp. 277–314.
74
Lu, X. Y., Wang, S., Sung, H. G., Hsieh, S. Y., and Yang, V., “Large Eddy Simulation
of Turbulent Swirling Flows Injected into a Dump Chamber,” Journal of Fluid Mechanics,
Vol. 527, 2005, pp. 171–195.
75
Wang, S., Yang, V., Hsiao G., Hsieh, S. Y., and Mongia, H., “Large Eddy Simulation of
Gas-Turbine Swirl Injector Flow Dynamics,” to be submitted to Journal of Fluid Mechanics,
2005.
76
Wang, S., “Vortical Flow Dynamics and Acoustic Response of Gas-Turbine Swirl-
Stabilized Injectors,” Ph.D. Thesis, Pennsylvania State Univ., University Park, PA, 2002.
77
Favaloro, S. C., Nejad, A. S., Ahmed, S. A., Vanka, S. P., and Miller, T. J., “An Exper-
imental and Computational Investigation of Isothermal Swirling Flow in an Axisymmetric
Dump Combustor,” AIAA Paper 89-0620, 1989.
78
Wu, J. Z., Lu, X. Y., Denny, A. G., Fan, M., and Wu, J. M., “Post-Stall Flow Control
on an Airfoil by Local Unsteady Forcing,” Journal of Fluid Mechanics, Vol. 371, 1998,
pp. 21–58.
79
Ho, C.-M., and Huerre, P., “Perturbed Free Shear Layers,” Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365–424.
80
Panda, J., and McLaughlin, D. K., “Experiments on the Instabilities of a Swirling Jet,”
Physics of Fluids, Vol. 6, 1994, pp. 263–276.
81
Berkooz, G., Holmes P., and Lumley J. L., “The Proper Orthogonal Decomposition
in the Analysis of Turbulent Flows,” Annual Review of Fluid Mechanics, Vol. 25, 1993,
pp. 539–575.
82
Graves, C. B., “Outer Shear Layer Swirl Mixer for a combustor,” U.S. Patent 5-603-211,
1997.
83
Shtern, V., and Hussain, F., “Collapse, Symmetry, Breaking, and Hysteresis in Swirling
Flows,” Annual Review of Fluid Mechanics, Vol. 31, 1999, pp. 537–566.
84
Martin, J. E., and Meiburg, E., “Nonlinear Axisymmetric and Three-Dimensional Vor-
ticity Dynamics in a Swirling Jet Model,” Physics of Fluids, Vol. 8, 1996, pp. 1917–1928.
85
Cohen, J., and Hibshman, J., “An Experimental Study of Combustor Air Swirler Acous-
tic and Fluid Dynamic Sensitivities,” Proceedings of Propulsion Engineering Research
Center, Pennsylvania State Univ., University Park, PA, 1997.
86
Hussain, A., and Reynolds, W., “The Mechanics of Organized Wave in Turbulent Shear
Flow,” Journal of Fluid Mechanics, Vol. 41, 1970, pp. 241–258.
87
Huang, Y., “Modeling and Simulation of Combustion Dynamics in Lean-Premixed
Swirl-Stabilized Gas Turbine Engines,” Ph.D. Thesis, Dept. of Mechanical Engineering,
Pennsylvania State Univ., University Park, PA, 2003.
88
Lasheras, J., and Hopfinger, E., “Liquid Jet Instability and Atomization in a Coaxial
Gas Stream,” Annual Review of Fluid Mechanics, Vol. 32, 2000, pp. 275–308.
89
Brereton, G., Reynolds, W., and Jayaraman, R., “Response of a Turbulent Boundary
Layer to Sinusoidal Free-Stream Unsteadiness,” Journal of Fluid Mechanics, Vol. 221,
1990, pp. 131–159.
90
Culick, F. E. C., and Yang, V., “Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines,” Chapter 1, Liquid Rocket Engine Combustion Instability, edited
by V. Yang and W. E. Anderson, Progress in Astronautics and Aeronautics, Vol. 169, AIAA,
Washington, DC, 1995, pp. 3–37.
276 Y. HUANG ET AL.
91
Gupta, A. K., Lilley, D. G., and Syred, N., Swirl Flows, Abacus Press, London, 1984.
92
Lucca-Negro, O., and O’Doherty, T., “Vortex Breakdown: A Review,” Progress in
Energy and Combustion Science, Vol. 27, 2001, pp. 431–481.
93
Syred, N., and Beer, J., “The Damping of Precessing Vortex Cores by Combustion in
Swirl Generators,” Astronautica Acta, Vol. 17, 1972, pp. 783–801.
94
Novak, F., and Sarpkaya, T., “Turbulent Vortex Breakdown at High Reynolds Number,”
AIAA Journal, Vol. 38, No. 5, 2000, pp. 287–296.
95
Seo, S., “Parametric Study of Lean Premixed Combustion Instability in a Pressured
Model Gas Turbine Combustor,” Ph.D. Thesis, Dept. of Mechanical Engineering, Pennsyl-
vania State Univ., University Park, PA, 1999.
96
Rayleigh, J. W. S., The Theory of Sound, Vol. 2, Dover, New York, 1945.
97
Lieuwen, T., Torres, H., Johnson, C. and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean-Premixed Gas-Turbine Combustor,” Journal of Engineering for Gas
Turbine and Power, Vol. 123, 2001, pp. 182–189.
98
Lipatnikov, A. N., and Chomiak, J., “Turbulent Flame Speed and Thickness: Phe-
nomenology, Evaluation and Application in Multi-Dimensional Simulations,” Progress in
Energy and Combustion Science, Vol. 28, 2000, pp. 1–74.
99
Chao, Y. C., Leu, J. H., and Huang, Y.F., “Downstream Boundary Effects on the Spectral
Characteristics of a Swirling Flowfield,” Experiments in Fluids, Vol. 10, 1991, pp. 341–348.
100
Liu, T., Huang, Y., and Yang, V., “Linear Stability Analysis of Annular Swirling Jets,”
Physics of Fluids (submitted for publication).
101
You, D. N., Huang, Y., and Yang, V., “A Generalized Model of Acoustic Response
of Turbulent Premixed Flame and Its Application to Gas-Turbine Combustion Instability
Analysis,” Combustion Science and Technology, Vol. 177, No. 5–6, 2005, pp. 1109–1150.
Chapter 11
Suresh Menon∗
Georgia Institute of Technology, Atlanta, Georgia
I. Introduction
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Professor, School of Aerospace Engineering.
277
278 S. MENON
For the preceding length scales, the turbulent Reynolds number, Rel = u l/ν,
where u is the turbulence intensity, is estimated to be in the 102 –104 range. The
higher value reflects the high level of turbulence in the regions of high shear (i.e.,
shear layer) in gas turbine engines. On using inertial range scaling, l/η ≈ Rel 3/4 ,
the Kolmogorov scale η can be estimated to be in the range 10−4 –10−5 m. Thus,
fluid dynamic length scales that are characteristic of vortex motion range from
10−5 –10−1 m. This is a four-order-of-magnitude (O(4)) range in length scales of
interest.
For reacting flows, additional length scales have to be considered. For example,
in two-phase systems, droplets are in the 10−4 –10−6 m range, whereas molecular
mixing and combustion occur in the 10−8 –10−9 m range. The latter estimate is
based on the observation that the typical CH molecule size is around 1.09 Angstrom
(10−10 m) and that the reaction-zone thickness, δ R R is at least 10–100 CH molecules
wide. This figure is just an estimate, since mixing and flame regions vary over
a wide range. Furthermore, the reaction-zone thickness is substantially smaller
than the effective flame thickness, δ f , especially in premixed systems where the
flame thickness can be substantially larger, particularly in the thin-reaction-zone
regime.21,28−30 Nevertheless, it is clear that species vaporization (in the case of
liquid fuel), mixing, and combustion occur in a range from 10−4 –10−9 m, an O(5)
range of scales.
Acoustic timescales can also be estimated on the basis of the range of frequencies
known to be excited in gas turbine engines. Typical frequencies are in the range
of 100–1000 Hz, and under standard conditions the wavelength range is 0.03–
0.3 m31,32 (under “hot” conditions, these values will be even higher). Thus, the
typical acoustic length scales of interest are in the 10−2 –100 m range.
Thus, there is an O(7–9) range of scale that has to interact in a turbulent reacting
flow under realistic conditions. These interactions have to occur in a time-accurate
manner (in nature, there is no other way!); and hence, the range of timescales is
equally large.
These estimates suggest that although a significant disparity exists between
the characteristic length scales in which vortex motion, acoustic fluctuations, and
heat release each dominate, there are also some regions of overlap. For example,
coupling between energy-containing eddies in a turbulent shear flow with acoustic
wavelengths in the 100–10,000 Hz range is feasible; however, at the other extreme,
eddies in the inertial-dissipation range are more likely to interact with unsteady
heat release at the molecular level.
The preceding argument is a rather simplistic view, since in reality, eddies of all
scales coexist and interact in a highly nonlinear manner in a turbulent flow. Regard-
less, these estimates suggest that the mechanism that couples acoustic fluctuations
with unsteady heat release in a turbulent flow is the dynamic range of scales inher-
ent in high-Re turbulent motion. In laminar flows, acoustic-vortex interactions can
still occur32−34 because, in addition to length scales, the timescales of interaction
are of the same order. For example, the flame-response timescale, τ f = δ f /SL ,
where δ f and SL are the laminar flame thickness and speed, respectively, for a
premixed system, is in the range of 10−3 –10−2 s, which is of the same order as
acoustic timescales for frequencies in the 100–1000 Hz range. Thus, it is possible
for an acoustic field to couple with heat release even in the absence of a turbulent
cascade of length scales.
280 S. MENON
This wide range of scales offers a serious challenge to both experimentalists and
modelers. Experimental diagnostic tools and simulation models both have to be
refined well enough to capture this wide range of scales accurately. That is easier
said than done.
The preceding discussion used order-of-magnitude estimates of time and length
scales independent of the problem of interest and the nature of the physics associ-
ated with these scales. As discussed in previous studies,28 the characteristic length
scales of interest can also be defined on the basis of the physics. For example, in
premixed combustion, the well-known Borghi diagram35 has been used in the past
to identify the type of flames (and hence, the associated length scales) in premixed
systems. Other diagrams have also been proposed on the basis of experimental and
numerical results in premixed and nonpremixed systems.31,36 These diagrams can
also be used to identify characteristic time and length scales of interest.
A. Governing Equations
For the purpose of the current discussion, we assume no external forces and
an inviscid, non–heat conducting, multispecies, nondiffusive reacting mixture.
The more general viscous equations and the linearization of the inviscid form for
acoustic analysis are reported in Chapter 13. The governing equations are
1 Dρ
∇ ·v= − (11.1)
ρ Dt
Dv ∇p
=− (11.2)
Dt ρ
De p
= − (∇ · v) (11.3)
Dt ρ
DYk ω˙k
= , k = 1, . . . N (11.4)
Dt ρ
N
DS 1 µk
=− ω˙k (11.5)
Dt ρT k=1 Mk
Here, S is the entropy and µk is the chemical potential.39 This equation shows
that entropy is convected because of fluid motion and can be generated by heat
release.
The mass conservation equation, Eq. (11.1) can be rewritten in terms of pres-
sure to identify the acoustic mode. Many forms of pressure (or wave) equations
have been derived in the past, including an equation that is often called the wave
equation.31 Here, a form that can be obtained by combining Eqs. (11.1), (11.5),
and the state equation is39
1 Dp 1 N
∇ ·v=− − σk ω˙k (11.6)
γ p Dt ρ k=1
where σk = h k /(C p T ) for an ideal gas mixture with constant average molecular
weight and where C p is the specific heat at constant pressure for a calorically
perfect gas mixture.
Equation (11.6) shows that the volumetric dilatation (∇ · v) can be affected by
pressure-wave motion and heat release. Additional manipulation of this equation
can result in a generalized equation for the logarithmic of the pressure31,40,41 that
can be used to model the acoustic wave motion.
By taking the curl of the momentum equation, Eq. (11.2), one can derive an
equation for vorticity, Ω:
DΩ ∇ρ × ∇ p
= (Ω · ∇)v − Ω(∇ · v) + (11.7)
Dt ρ2
For analysis purposes, Eq. (11.7) is often combined with Eq. (11.1) to obtain
D Ω Ω 1
= · ∇v + ∇ρ × ∇ p (11.8)
Dt ρ ρ ρ3
Both forms of the vorticity equation show important features that are relevant
to the current discussion. The first term on the right-hand-side of Eq. (11.7) and
Eq. (11.8) is the well-known vortex-stretching term. The other terms are unique
to compressible and reacting flow. The quantity Ω(∇ · v) is the thermal expansion
term, which is nonzero only in compressible flow and in reacting flow. In Eq. (11.8),
this term is absorbed into the convective term on the left-hand side. Therefore, the
left-hand side of Eq. (11.8) contains convection by both the hydrodynamic and
the acoustic velocity. As shown subsequently, this form can be used to identify
how acoustic and hydrodynamic disturbances can interact and contribute to vortex
motion.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 283
The last term, ∇ρ × ∇ p/ρ 3 , often called the baroclinic torque, is a consequence
of the relative orientation of the pressure and density gradients. This term can be
significant in both compressible and reacting flows. For example, baroclinic torque
has been shown to affect flame wrinkling indirectly because of the changes induced
in the vorticity field.42 Pressure fluctuations normal to the density gradient caused
by the flame also contribute to vorticity generation. In gas turbine combustors,
the flame (and hence, the density gradient) is highly three-dimensional; and there-
fore, both longitudinal and transverse acoustic waves in the combustor can cause
changes to the local vortical field. Furthermore, this term can cause enhancement
or suppression of vorticity in the combustor depending on local conditions. This
behavior is discussed in the final section.
Equations (11.5), (11.6), and (11.7) (and Eq. (11.4) for species conservation) can
be considered to be an alternative set of governing equations in terms of entropy,
pressure, and vorticity instead of the conventional conservation equations. Except
for neglecting transport properties (e.g., dissipative effects) and external forces (and
these assumptions can be easily relaxed), these equations are exact, nonlinear, and
define the flowfield.
Although these equations are informative, their solution is not easy because of
the nonlinearity. These equations are, however, useful for linearized analysis and
for interpreting results. [See Chaps. 13 and 10]. Additionally, the direct numerical
simulation (DNS) or large-eddy simulation (LES) database can be used to extract
some of these terms explicitly.
=
p +
+
S (11.9)
(x, t) =
(x)
¯ +
(x,
˜ t; τ ) +
(x,
ˆ t) (11.10)
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 285
Here,
¯ is the mean value,
On using Eq. (11.13) in Eq. (11.12) and noting that (∂ρ/∂ p) S = a S 2 is the
isentropic (frozen) acoustic speed, we obtain
D 1 Dp 1 2 1 D 1 ∂ρ DS
− ∇ p = ∇v : ∇v − 2 ∇ρ · ∇ p −
Dt ρa S 2 Dt ρ ρ Dt ρ ∂ S p Dt
(11.14)
The first term in the RHS of this equation is the classical Lighthill’s Reynolds
stress sound-generation term.31 All other terms on the RHS are nonzero only for
reacting and compressible flow.
Thus, the acoustic equation, Eq. (11.14) contains explicit terms for hydrodynam-
ics and combustion heat release that can either enhance or suppress the pressure
disturbance. The analysis of these terms can shed insight into AVF interactions.
Such an analysis can only be done by using numerical simulation data and sug-
gests a possible avenue by which the DNS or LES database could be used to help
interpret experimental observations, which are likely to be limited in spatial and
temporal resolution.
The sources and sinks in the vorticity equation can also be analyzed. If we
consider that the velocity field is decomposed into a hydrodynamic and an acoustic
component, as v = v + v p , then the left-hand side of Eq. (11.8) can be written
as
D D D D 1 1
= + + −
Dt ρ Dt ρ0 Dt p ρ Dt ρ ρ0
(11.16)
The implication of this decomposition is that the first term on the RHS is the
Orr–Sommerfield operator (in the linearized limit). The second term can be rewrit-
ten as
D
≈ vp · ∇ (11.17)
Dt p ρ ρ
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 287
A. Swirl
Swirl is a key element in all gas-turbine engines and is used to create a region
of high entrainment and mixing for the fuel–air mixture. Swirl also provides an
efficient mechanism to stabilize the flame in a compact region without requiring
a physical flame holder. In general, the extent of swirl is typically defined by the
swirl number Si , which is the ratio of the axial flux of angular momentum to the
product of the inlet radius and axial flux of axial momentum. Other parameters—
such as inlet swirl-vane geometry, Reynolds number, confinement geometry, and
inlet velocity (both mean and fluctuation) profiles—can all affect swirl effects.
With and without confinement, the nature of the flow downstream changes
significantly as a function of the swirl number Si . As long Si is below a critical
value, typically 0.6 for dump combustors,46 the shear layer from the inlet separates
from the dump plane and rolls up into vortices that eventually coalesce into large-
scale coherent vortex structures. These structures propagate downstream at a phase
velocity, which is of the order of the mean velocity, and maintains coherence for
some distance before breaking down into more irregular, three-dimensional vortical
structures.
Results from classical linear stability theory47,48 can be used to obtain insight
into the vortex motion for low swirl numbers. For example, stability analysis sug-
gests that the characteristic Strouhal number (Stθ = f θ θ/U ) for the most unstable
mode in a shear layer is 0.032. Here, θ is the shear layer momentum thickness, U
is the characteristic inflow mean velocity, and f θ is the characteristic frequency
288 S. MENON
Fig. 11.1 Flame and vortex structures in the combustor for inlet swirl number
of Si = 0.56.
Previous studies20,21 have compared the LES predictions against measured data
for a similar single-element rig, and very good agreement was obtained. Here,
we discuss more qualitative features extracted from these simulations to highlight
the nature of AVF interactions. Figure 11.1 shows the typical instantaneous flame–
vortex structure for a low-swirl, (Si = 0.56) case. Here, the reference swirl number
Si is defined at the inlet boundary (it drops to 0.42 at the dump plane). The flame in
this figure is represented by an isoscalar surface, and the coherent structure is shown
as the λ2 isosurface.51 It can be seen that low swirl results in large-scale, coherent
ring-like structures that are similar to structures seen in a nonswirling forced jet
shear layer. These structures undergo rotation and exhibit azimuthal perturbation
that rotates and grows as the structures move downstream. The flame moves in
phase with these structures until the vortical structures break down because of
instability. At this stage, the flame decouples from the vortex ring and retreats
toward the inlet. It is then pulled forward by the next large-scale structure. Thus,
the vortex-flame structure undergoes a periodic pulsation.
The shape and pulsation of the flame in the low-swirl case is reminiscent of the
“tulip” flames seen in acoustic coupled laminar flames.40 Here, although the con-
ditions are highly turbulent, the modulation of the flame by the coherent structures
seems to result in the observed shape.
In Chapter 10 additional flame-vortex interaction effects are discussed, except
that in their case, the combustor consists of an annular swirling flow in the inlet with
a centerbody. Although some differences are attributable to the geometric nature
of the two devices, significant similarities suggest that many observed features are
fundamental components of swirling combustion.
When the swirl is increased beyond the critical value, a major physical change in
the flow is observed. Around the critical value, the adverse axial pressure gradient
290 S. MENON
caused by swirl exceeds the forward momentum force of the inflow, and the flow
reverses. This reversal typically occurs first in the centerline region for a single
swirling jet inlet. In the coaxial swirl combustor or in combustors with a centerbody,
the location of flow reversal is off-center but still axially located in the region where
the swirling flow enters the combustor.
High swirl in the inflow results in the formation of a vortex breakdown bubble
(VBB), or a recirculation bubble. This VBB can be a single bubble at the centerline
or a toroidal structure depending on the inlet conditions, the inlet, and the dump
combustor design. In multielement, multiswirler sector rigs, the shape of VBBs can
be even more complex and can consist of both core recirculation zones and offset
toroidal structures.52,53 Depending on the geometry and the operating condition,
the leading edge of the VBB can move into the inlet as well.
In any event, the VBB acts as an aerodynamic blockage that inhibits the classical
growth and rollup–pairing processes in the shear layer. The initial rollup–merging
process and the formation of large-scale structures still occur. However, since the
shear layer is diverted radially away from the centerline because of the presence
of the VBB, the large-scale structures undergo rapid stretching in both the radial
and azimuthal directions. This stretching causes these structures to break down
quickly into three-dimensional relatively small-scale, but still vortically coher-
ent, structures.54 This breakdown leads to the formation of more axially oriented
streamwise vortices that form the precessing vortex core (PVC) that rotates about
the central VBB46,55 while the shear flow propagates downstream.
The effect of high swirl and of the formation of the recirculation bubble on
the flame structure is dramatic. Figure 11.2 shows an instantaneous flame–vortex
Fig. 11.2 Flame and vortex structures in the combustor for inlet swirl number of
Si = 1.12.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 291
structure for a high-inlet-swirl (Si = 1.12, which reduces to around 0.74 at the
dump plane) case for the same combustor shown previously (both simulations
employed identical conditions except for an increase in Si ). With an increase in
swirl, the vortical ring structures break down quickly, and the coupling between
the flame and the vortex is broken. The result is a compact flame that stabilizes
very close to the dump plane and upstream of the VBB.
Figures 11.3a and 11.3b show representative particle paths in the combustor
for the low and high swirl cases, respectively. The streamlines are computed by
using the time-averaged velocity field and therefore do not represent the actual
instantaneous motion of the fluid element in the flow. Nevertheless, these particle
paths provide some insight into the complex flow in the combustor. For each case,
two seed particles are injected at the inlet; one near the centerline and the other
in the boundary layer on the inlet wall. In Fig. 11.3a (low-swirl case), the particle
injected at the centerline is propelled almost without interruption through the entire
combustor. The particle injected near the inlet wall shows a much stronger effect
of the swirl. Inside the combustor, the fluid particle initially slows down and then
moves upstream on entering the separated flow region at the base of the dump plane.
Figure 11.3b shows that for the high-swirl case, the particle injected along the
centerline can get entrained into the VBB and undergoes multiple revolutions in
it before being ejected. The outer particle does not enter the VBB; rather it is
quickly accelerated around the VBB toward the rear of the recirculation zone.
From there, the particle velocity rapidly decreases, and a slow rotating motion is
initiated before the particle is finally ejected into the main flow and accelerated
out of the combustor. An analysis of the instantaneous particle paths indicates that
particles in the shear layer tend to be present more in the PVC than in the VBB.
The flow inside the VBB contains primarily hot burned products, is highly
turbulent with strong three-dimensional variation of turbulent kinetic energy, and
with significant anisotropy in the Reynolds stress components.46 Studies with a
central fuel jet surrounded by a swirling coaxial air56 have shown that the flame
essentially surrounds the toroidal VBB. Furthermore, they showed that the VBB
size, the mixing rate, and the flame length depend on the degree of recirculation and
the ratio of the momentum of the fuel jet to the momentum inside the recirculation
region. These results suggest that the VBB acts as a large eddy that is not just a
passive structure in the flame-stabilization process.
Stability analysis of highly swirling jets, with and without confinement, have
shown that many modes of instability can occur in swirling shear layers. Studies57
in an unconfined swirling water jet in a low-to-moderate Re (∼ 1200) showed
that, in addition to the classical VBB, a conical vortex sheet structure is formed.
Both these structures became asymmetric with an increase in Reynolds number,
suggesting a spiral mode of instability.
In much higher Re swirling free jet and wake flows,54,58 more-complex insta-
bilities were found. In particular, it was shown that both axisymmetric and helical
instability waves exist in an St = 0.75–1.5 range. The vortex structure around the
VBB was less coherent and irregular for this case; but with even weak acoustic
excitation, the coherence and periodicity of the structures improved dramatically.
This observation is an important one for gas-turbine combustors, since confinement
results in the excitation of the acoustic modes in the combustor. Thus, it is possible
that the observed coherence of large-scale structures in these combustors may be
292 S. MENON
a)
b)
Fig. 11.3 Particle paths in the combustor as a function of inlet swirl number:
a) low swirl, Si = 0.56 and b) high swirl, Si = 1.12.
a)
b)
Fig. 11.4 Axial velocity contours and flame structure as a function of swirl number.
The recirculation zones in the combustor are identified by black solid lines: a) low
swirl, Si = 0.56 and b) high swirl, Si = 1.12.
294 S. MENON
solid line. The typical flame structure is also identified is an isocontour. In the
low-swirl case, there is no VBB. However, for the high-swirl case, a large VBB is
clearly visible in the combustor. The dump-plane base recirculation bubble (BRB)
is also visible in both these figures. For low swirl, the BRB is relatively large;
whereas for high swirl, the BRB is very small (but with more intense flow motion)
and trapped up in the step corner.
In gas-turbine combustors, the confinement geometry can introduce new features
that can affect shear flow and flame stabilization. In the configuration discussed
previously, the rearward-facing step at the dump plane creates a base recircu-
lation bubble (BRB) containing hot combustion products that provides another
mechanism for flame reignition and stabilization. Flame stabilization using the
rearward-facing step is well known and is employed in dump combustors as in a
ramjet in the absence of swirl. The current results suggest that the size of the BRB
depends on both the swirl intensity and the shape of the device. Low swirl creates a
well-defined BRB, since no VBB is in the core of the flow [Fig. 11.4a]. However,
when the swirl is very high, the VBB is very close to the dump plane, and the
lateral divergence of the shear layer compresses this BRB into a small region that
is very close to the step corner [Fig. 11.4a]. Thus, it appears that the BRB may not
play a major role in flame stabilization in the high-swirl case.
The preceding observations are only valid in a dump combustor with a single
inlet. In combustors with coaxial flow or with centerbodies,53,59 multiple BRBs can
form. However, they are much smaller than the primary VBB and may only play a
secondary role in flame stabilization. In most operational gas turbine combustors,
the VBB is the primary aerodynamic flame-stabilizing mechanism.
The behavior of the PVC is also significantly altered by combustion heat release
and by the system parameters (e.g., swirl number, axial velocity and geometry).
Previous studies46,60 suggest that the type of combustion (i.e., premixed or non-
premixed) can affect the PVC frequency and its intensity. Geometrical and system
features, such as secondary air injection for cooling and nonaxial injection of
fuel, can also lead to significant changes to the PVC and to the flame-stabilization
process.
The unsteady motion of the VBB, the PVC, and the flame structure all can lead to
unsteady pressure fluctuation in the combustor. As previously noted, in-phase fluc-
tuations can lead to instability. However, under certain conditions, since an increase
in swirl can actually lead to rapid breakdown of the coherent vortices in the shear
layer, it can contribute to a de-coupling of the vortex motion from acoustic fluctu-
ations and unsteady heat release. As a result, instability may actually be averted.
The stability of the combustion process can be determined by evaluating the
Rayleigh criterion. This criterion says that when unsteady heat release is in phase
with the unsteady pressure fluctuation, the heat release adds energy to the oscilla-
tion. This energy addition can lead to combustion instability if it exceed the losses
(caused by viscous dissipation and by outflow) from the system. In spatially evolv-
ing unsteady reacting flow, as in a combustor, the phase between unsteady heat
release and unsteady pressure can vary locally both in space and in time. Thus,
in some regions in the combustor, the instability could be suppressed whereas
in other locations it could be enhanced. To quantify
the state of combustion, a
volume-averaged Rayleigh parameter R(t) = V p q dV can be defined. Here,
the integral is over the entire combustor domain and q and p are the unsteady
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 295
a)
6
Unstable ( prms = 2)
4
2
*
p' and R
-2
-4
Rayleigh parameter
Pressure fluctuation
-6
1.0 1.2 1.4 1.6 1.8 2.0
Time (ms)
b)
6
4 Stable ( prms = 1)
2
*
p' and R
-2
-4
Rayleigh parameter
Pressure fluctuation
-6
1.0 1.2 1.4 1.6 1.8 2.0
Time (ms)
Fig. 11.5 Rayleigh parameter and pressure fluctuation in the combustor: a) low swirl,
Si = 0.56 and b) high swirl, Si = 1.12.
zone is higher than the mean. Because of this high-pressure zone, the axial velocity
is low and the flame is able to propagate further upstream, consuming more fuel
along the way. Under this condition, unsteady heat release occurs in phase with
pressure fluctuation, leading to R ∗ > 0.
For the-high swirl case, Fig. 11.5b, the flame is very compact and the vortical
structures are no longer coherent rings, as in the low-swirl case. Flow is accelerated
around the VBB; and near the dump plane, heat release is no longer coupled to the
pressure fluctuation. As a result, the Rayleigh parameter and pressure fluctuation
are no longer in phase and combustion is more stable. An estimate for the pressure
root-mean-square fluctuation intensity shows that pr ms is decreased by nearly 100
percent when the combustion process becomes more stable.
In the preceding discussion, the effect of swirl was discussed primarily for
premixed combustion. However, swirl is also very important for all gas-turbine
engines used for propulsion where liquid fuel is employed. The fuel-injection
system creates droplets over a wide range of sizes by the atomization process.
Droplet transport, vaporization, fuel–air mixing and combustion in the combustor
depend in part on how these droplets are entrained and dispersed within the swirling
air stream. These issues are discussed in the following two sections.
B. Droplet-Vortex Interactions
Before discussing spray dispersion and combustion in gas-turbine combus-
tors, summarizing observations from fundamental studies of droplet dispersion
in shear layers is worthwhile. Previous studies22,61 using direct numerical simula-
tions (DNSs) of droplet-laden temporal mixing layers provide insight into droplet–
vortex interaction, and some results are discussed subsequently.
Simulations of two-phase flows are carried out by using the previously noted
finite-volume code with a Eulerian–Lagrangian approach. In this method, the gas
phase is simulated by using the usual finite-volume approach, whereas the particle
motion is simulated by using a Lagrangian scheme. Full two-way coupling is
included in this approach. Droplet vaporization, gaseous fuel mixing with air, and
subsequent combustion can also be simulated with this solver, as reported in the
following section.
Here, we show some results for droplet motion in a temporal mixing layer. The
test conditions used here are identical to those used in an earlier DNS study using
a pseudospectral DNS code.61 A cubic domain is discretized by using a uniform
grid of 64 × 64 × 64. A temporal mixing layer, initialized by a tangent hyperbolic
mean profile and perturbed by the first two most unstable two-dimensional modes
is simulated here. The computational domain is chosen such that one vortex pairing
can be simulated.
We study the effect of Stokes number St0 on particle dispersion. Here, St0 is
defined as St0 = τd /τ f , the ratio of a particle response time τd and a characteristic
flow time τ f . Here, τd = (ρd D 2 )/18µ is the particle response time, ρd is the
particle density, D is the particle diameter, and µ is the molecular viscosity of
the gas phase. The characteristic flow time is τ f = L/U0 where L and U0 are the
characteristic length and velocity scale for the flow.
As previously defined, St0 1 implies that the particle can easily respond to
the changes in flow, and therefore the particle and the flow velocity will reach
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 297
an equilibrium. However, for St0 1 particles will not have time to respond to
the flow velocity change and particle motion will not be significantly affected.
For a given flow time, variation in St0 is directly related to the particle diameter.
In the following, we compare the behavior of particles with St0 of 0.1, 1, 4, and
100 in the mixing layer. The initial condition for the droplets is such that they are
uniformly distributed throughout the domain and are in dynamic equilibrium with
the gas phase. One particle per cell is placed at the start of the simulation. For
each simulation, droplets of same size are used (fixed St0 ) and no vaporization is
included here.
Figure 11.6 shows the distribution of droplets at a nondimensional time of
T = 28. At very small St0 , droplets are dragged along with the fluid and can
a) b)
c) d)
therefore even be convected into the core of the spanwise vortical structures. As
the droplet Stokes number approaches unity, droplets begin to accumulate near the
circumference of the large vortical structures, which leads to an increase in lateral
droplet dispersion with droplets marking the lateral boundaries of the rolled-up
vortical structures. With a further increase in Stokes number, droplet response time
further increases and the influence of flow on the droplets decreases. Some droplets
are then observed to pass through the core of these structures and accumulate in
the braid regions between the large vortical structures. This trend is in very good
agreement with observations in the earlier DNS study.61
Dispersion behavior of droplets of St0 = 4 is shown in Fig. 11.7. Droplet dis-
tribution is shown at two nondimensional times. Figure 11.8 shows the span-
wise vorticity and the droplet distribution at the same instants. Droplets tend
to accumulate around the circumference of the large vortical structures that are
formed by rollup. The primary spanwise vortical structure is essentially devoid of
droplets because the strong vorticity present at the core of the spanwise vorticity
centrifuges the droplets away from the center. These droplets accumulate near
the high-strain regions of the flow, such as the braid regions between the larger
spanwise rollers. These results agree with past observations61−63 that droplets
tend to accumulate in regions of low vorticity and high strain rate. Such prefer-
ential concentration of droplets can have serious implications in liquid-fueled
combustors, where preferential accumulation may lead to large spatial varia-
tion in the mixture-equivalence ratio, which in turn, may lead to incomplete
combustion.
Figure 11.9 shows the root mean square (RMS) of droplet number per cell
(Nrms ) over the whole field. This number can be used to characterize the overall
accumulation tendency of the droplets in the flowfield. The Nrms is defined as61
Nc
N 2
Nrms = i
(11.18)
i=1
N C
a)
b)
Fig. 11.7 Droplet dispersion in mixing layer for St0 = 4: a) T = 12 and b) T = 20.
300 S. MENON
a) b)
Fig. 11.8 Droplet distribution and spanwise vorticity in mixing layer for St0 = 4:
a) T = 12 and b) T = 20.
defined as61
1 Np
D y (t) = [Yi (t) − Ym (t)]2 (11.19)
N p i=1
where N p is the total number of droplets, Ym (t) is the mean value of the droplet
displacement in the vertical direction at time t, and Yi (t) is the displacement of an
ith droplet in the vertical direction at time t.
T = 24 (Present DNS )
RMS of particle numbers per cell
T = 28
T = 36
T = 24 (Spectral-DNS)
T = 28
2 T = 36
1.5
4
St O = 0.1
St O = 1
St O = 4
3 St O = 10
St O = 100
Dy (t)
0
0 10 20 30
Nondimensional time
Fig. 11.10 Stokes number effect on transverse droplet dispersion.
Figure 11.10 shows this dispersion function as a function of time for the droplets
for various St0 . At very small times, droplets with St0 = 0.1 show more dispersion.
However, at later times, droplets with St0 of order unity show more dispersion.
Droplets with very large Stokes numbers have much lower dispersion compared
with lighter droplets.
Vortex–droplet coupling has also been observed in liquid-fueled com-
bustors.22,64 Figure 11.11 shows a GE Dual-Annular Counter-Rotating Swirling
(GE-DACRS) combustor that was used in a recent study to understand droplet dis-
persion in swirling flow. The computational grid used 141 × 75 × 81 grid points
in the axial, radial, and azimuthal directions, respectively, with clustering in high-
shear regions. A large number of particles, typically more than 100,000 com-
putational parcels are tracked. Each of these parcels contains an average of 150
individual droplets. Thus, an average of 15 million droplets are present in the
computation. A monodisperse distribution with spherical droplets of size 40 µm
(which is approximately the Sauter mean diameter of the droplets in DACRS)
is introduced in the inlet through the central shaft at half-angle of 10 deg to the
flow. For the simulated case, the inlet pressure was 13.8 atm and the bulk Re was
260,000.
Figures 11.12a and 11.12b show a low swirl (Si = 0.75) case and a high swirl
(Si = 1.5) case, respectively. Because on the shape of the inlet, the actual swirl
numbers at the dump plane are approximately 0.5 and 0.8, respectively, attributable
to flow relaxation and decay in the inlet. The vortical structures undergo helical
instability and break down rapidly with an increase in swirl, as seen in the pre-
mixed combustor. Closer examination shows that sheetlike structures associated
with the spanwise vorticity z form tubular rings, whereas structures associated
with the streamwise vorticity x form tubular braidlike connections between the
rings. However, farther downstream as the rings break down, more randomly ori-
ented tubular structures appear. Analysis described elsewhere22 has shown that the
magnitude of the strain field controls this breakdown.
With an increase in swirl, droplet dispersion increases significantly. As observed
in temporal mixing layers, droplets tend to collect in the low-vorticity, high-strain
regions; and their dispersion is significantly enhanced once the large-scale ring
structures have lost their coherence.
C. Droplet–Vortex–Flame Interactions
In spray combustion systems, additional complexity can be created by the in-
teraction between droplets and flame. Droplet vaporization is accompanied by
fuel–air mixing and occurs in a swirling flow. However, in most situations, mix-
ing is not perfect everywhere; and in general, premixed, partially premixed, and
nonpremixed regions can coexist in the mixing region. Thus, the flame structure
can be quite complex in these regions. Both isolated flames around droplets and
group combustion of droplets (where droplets are clustered close together so that
only the region at the edge of the cloud contains vaporized fuel) have been seen in
spray combustion simulations.22
These observations suggest that the location and the structure of the unsteady
heat-release zone in the combustor can vary significantly depending on the ability of
the inlet swirl to mix and then redistribute the mixed fluid. Very little experimental
data on operational gas turbines describe how the flame structure actually correlates
with the vortical structures and with the acoustic fluctuation as the operational point
is systematically changed.
However, some insight into the flame structure has been possible from sim-
ulation data. It appears that both spray and premixed combustion systems have
physical characteristics that are remarkably similar.22,65 These results also agree
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 303
a)
b)
Fig. 11.12 Spray dispersion and vortex structure in the GE-DACRS combustor:
a) low swirl, Si = 0.75 and b) high swirl, Si = 1.5.
304 S. MENON
X
Z
with classical DNS studies in simpler flows at relatively low Re, suggesting some
sort of universality in these features. The most likely strain state is appears to be
axisymmetric extension; and the vorticity tends to align itself with the intermediate
strain rate, whereas the scalar gradient aligns with the most compressive strain rate.
The magnitude of these alignments is found to decrease in the presence of droplets
and with heat release or an increase in swirl. Analysis shows that both tubelike
and sheetlike structures can coexist in the combustor and that their relative abun-
dance (or lack thereof) is a function of spatial location in the combustor and swirl
number. For example, tubelike structures are more likely in regions with intense
vorticity gradients, whereas regions of increased scalar gradients form sheetlike
structures that in turn wrap around the tubular vortical structures. The scalar gra-
dients in these sheets are amplified by the interaction between the strain rate and
the vorticity fields, thus increasing mixing and reaction in these regions.
As a final example of spray combustion, Fig. 11.13 shows a sector rig consisting
of two cups, each with multiple annular swirlers that surround the primary fuel
injector.52 This combustor is hereafter identified as GE-1.53 Although the two cups
are identical, the sector shape is not. Therefore, the regions downstream of the two
cups are not the same, and significant three-dimensional interaction between the
two cups occurs. Fuel (kerosene) is injected from the primary injector in the center
of each cup and from 20 injectors placed equally far apart on the rim surrounding
the primary cups. The total fuel flow rate is split equally between the two cups.
Spray is injected by using parcels to represent groups of droplets with similar
properties. Typically, an average of approximately 350,000 droplet parcels (with
10–15 droplets per group) are present when the simulation reaches the stationary
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 305
a)
b)
Fig. 11.14 Spray and reaction rate contours in the center x–y plane: a) spray and
CO2 mass fraction and b) reaction rate.
a)
b)
Fig. 11.15 Axial velocity contours and streamline pattern in the center x–y plane:
a) axial velocity contours and b) streamlines and temperature contours.
308 S. MENON
Fig. 11.16 Instantaneous z−component of the baroclinic torque in the center x–y
plane.
the burned regions are trapped between the swirling outer streamlines. Analysis
shows that the classical PVC from each cup persists only for a short region around
the VBB. Farther downstream, the structures from both cups break down and
merge into a single stream with very little local coherence. This result is probably
more realistic (and part of the design strategy), since the outflow from this sector
is actually more uniform than in a single injector case.
To understand how vortex–flame interactions occur in this device and how this
process affects the acoustic fluctuations, additional postprocessing of the results is
required. Analysis of the time-evolving flow data can also be used to understand
how AVF interactions occur. For example, Fig. 11.16 shows a center-plane view
of the instantaneous z component of the previously discussed baroclinic torque
term. Both positive and negative values of this quantity are present in the flow. As
noted, positive values indicate enhancement of vorticity, whereas negative values
indicate suppression. Overall, the pattern closely follows the flame front, but this
quantity is nonzero in the regions where the droplets are vaporizing and mixing as
well.
Figures 11.17a and 17b show the mean pressure and the unsteady dilatation,
respectively, at the center plane. Although very little three-dimensional variation
seems to occur in the mean pressure in this plane, local variations still occur,
especially near the dump plane. The mean dilatation field (not shown) also shows
a similar behavior, with three-dimensional variations near the dump plane that
quickly smooth out as the exit plane is approached.
On the other hand, the unsteady dilatation shows significant three-dimensional
structure and a periodic wavelike structure propagating from the inlet. The unsteady
a)
b)
Fig. 11.17 Mean pressure and unsteady dilatation fields in the center x–y plane:
a) mean pressure and b) unsteady dilatation.
310 S. MENON
pressure troughs (not shown) are closely related to the unsteady dilatation crests.
Time-series analysis indicates that these fields exhibit axially moving, hemispher-
ical wavelike structures that grow from the two inlets. However, halfway into the
combustor, the disturbances from the two cups start to interfere and loose their
coherence. The radial crossflow caused by the geometrical convergence introduces
transverse disturbances that interact with the axially moving waves, so that further
downstream, both unsteady pressure and dilatation fields exhibit disturbances that
are more axially oriented.
Analysis of these fields shows that for the test conditions employed here, the
unsteady dilatation and the unsteady pressure fields are only weakly correlated with
the vortex–flame structures in the entire combustor. As a result, no enhancement
of the acoustic fluctuations occurs and the combustion process is stable. Since the
test conditions were chosen for a stable condition, this observation is encouraging.
V. Conclusion
This chapter summarizes some observations of AVF interaction in gas-turbine
engines. Since swirl is a key feature in all operational gas-turbine combustors, all
observed AVF phenomena in these combustors are somehow affected by the type
and intensity of swirl in the inlet. In addition, the fuel-injection method, fuel–air
mixing (especially in liquid-fueled systems), flame stabilization, and combustor
geometry (multi-injector) are also important parameters in gas-turbine systems.
Understanding the sensitivity of the combustor’s performance to these parameters
requires a comprehensive and integrated experimental and simulation strategy that
is only now becoming a reality.
Acknowledgments
The results reported in this chapter have been obtained with support from Army
Research Office, General Electric Power Systems, and General Electric Aircraft
Engine Company. The simulations reported here were carried out by C. Stone, V.
Sankaran, and N. Patel.
References
1
Mongia, H. C., Held, T. J., Hsiao, G. C., and Pandalai, R. P., “Challenges and Progress
in Controlling Dynamics in Gas Turbine Combustors,” Proceedings of the Combustion
Institute, Vol. 19, The Combustion Inst., Pittsburgh, PA, 2003, pp. 822–829.
2
Correa, S. M., “Power Generation and Aeropropulsion Gas Turbines: From Combustion
Science to Combustion Technology,” Proceedings of the Combustion Institute, Vol. 27, The
Combustion Inst., Pittsburgh, PA, 1998, pp. 1793–1807.
3
Chu, B. T., and Kovasznay, L. S. G., “Non-Linear Interactions in a Viscous Heat-
Conducting Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, 1958, pp. 494–
514.
4
Culick, F. E. C., and Yang, V., “Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines,” Liquid Rocket Engine Combustion Instability, edited by V. Yang
and W. E. Anderson, Progress in Astronautics and Aeronautics, AIAA, Washington, DC
1995, pp. 3–37.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 311
5
Vuillot, F., “Vortex-Shedding Phenomena in Solid Rocket Motors,” Journal of Propul-
sion and Power, Vol. 11, No. 4, 1995, pp. 626–639.
6
Apte, S., and Yang, V., “Unsteady Flow Evolution and Combustion Dynamics of
Homogeneous Solid Propellant in a Rocket Motor,” Combustion and Flame, Vol. 131,
No. 1–2, 2002, pp. 110–146.
7
Schadow, K. C., and Gutmark, E., “Combustion Instability Related to Vortex Shedding in
Dump Combustors and their Passive Control,” Progress in Energy and Combustion Science,
Vol. 18, No. 2, 1992, pp. 117–132.
8
Menon, S., and Jou, W.-H., “Large-Eddy Simulations of Combustion Instability in an
Axisymetric Ramjet Combustor,” Combustion Science and Technology, Vol. 75, No. 1,
1991, pp. 53–72.
9
Yu, K. H., Trouvé, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in a
Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, No. 11, 1991, pp. 47–72.
10
Poinsot, T., Trouve, A. C., Veynante, D., Candel, S. M., and Esposito, E. J., “Vortex-
Driven Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics,
Vol. 177, No. 4, 1987, pp. 265–292.
11
Jou, W.-H., and Menon, S., “Modes of Oscillation in a Nonreacting Ramjet Combustor
Flow,” Journal of Propulsion and Power, Vol. 6, No. 5, 1990, pp. 535–543.
12
Takahashi, F., Schmoll, W. J., Trump, D. D., and Goss, L. P., “Vortex-Flame Interactions
and Extinction in Turbulent Jet Diffusion Flames,” Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 145–152.
13
Santoro, V. S., Kyritsis, D. C., and Gomez, A., “An Experimental Study of Vortex-
Flame Interactions in Counterflow Spray Diffusion Flames,” Proceedings of the Combustion
Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 1023–1030.
14
Renard, P.-H., Thevenin, D., Rolon, J. C., and Candel, S., “Dynamics of Flame/Vortex
Interactions,” Progress in Energy and Combustion Science, Vol. 26, No. 3, 2000, pp. 225–
282.
15
Dowling, A. P., “Vortex, Sound and Flames—A Damaging Combination,” Aeronautical
Journal, Vol. 104, No. 1033, 2000, pp. 105–116.
16
Paschereit, O. C., Gutmark, E., and Weisenstein, W., “Excitation of Thermoacoustic
Instabilities by Interaction of Acoustics and Unstable Swirling Flow,” AIAA Journal, Vol. 38,
No. 6, 2000, pp. 1025–1034.
17
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NO Gas Turbines,” Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 1809–1816.
18
Lee, T. W., and Santavicca, D. A., “Experimental Diagnostics for the Study of Combus-
tion Instabilities in Lean Premixed Combustors,” Journal of Propulsion and Power, Vol. 19,
No. 5, 2003, pp. 735–750.
19
Cohen, J. M., Wake, B. E., and Choi, D., “Investigation of Instabilities in a Lean
Premixed Step Combustor,” Journal of Propulsion and Power, Vol. 19, No. 11, 2003,
pp. 81–88.
20
Kim, W.-W., Menon, S., and Mongia, H. C., “Large-Eddy Simulation of a Gas Turbine
Combustor Flow,” Combustion Science and Technology, Vol. 143, No. 1, 1999, pp. 25–62.
21
Kim, W.-W., and Menon, S., “Numerical Simulations of Turbulent Premixed Flames in
the Thin-Reaction-Zones Regime,” Combustion Science and Technology, Vol. 160, No. 1,
2000, pp. 119–150.
22
Sankaran, V., and Menon, S., “LES of Spray Combustion in Swirling Flows,” Journal
of Turbulence, Vol. 3, No. 11, 2002.
312 S. MENON
23
Stone, C., and Menon, S., “Open Loop Control of Combustion Instabilities in a Model
Gas Turbine Combustor,” Journal of Turbulence, Vol. 4, No. 1, 2003.
24
Huang, Y., Sung, H.-G., Hsieh, S.-Y., and Yang, V., “Large Eddy Simulation of Com-
bustion Dynamics of Lean Premixed Swirl-Stabilized Combustors,” Journal of Propulsion
and Power, Vol. 19, No. 5, 2003, pp. 782–794.
25
Sommerer, Y., Galley, D., Poinsot, T., Ducruix, S., Lacas, F., and Veynante, D., “Large-
Eddy Simulation and Experimental Study on Flashback and Blow-Off in a Lean Partially
Premixed Swirled Burner,” Journal of Turbulence, Vol. 5, No. 1, 2004.
26
Rockwell, D., and Naudascher, E., “Self-Sustained Oscillations of Impinging Shear
Layers,” Annual Review of Fluid Mechanics, Vol. 11, 1979, pp. 67–94.
27
Stow, S. R., Dowling, A. P., and Hynes, T. P., “Reflection of Circumferential Modes in
a Choked Nozzle,” Journal of Fluid Mechanics, Vol. 467, 2002, pp. 215–239.
28
Peters, N., Turbulent Combustion, Cambridge Monographs on Mechanics, Cambridge
Univ. Press, U.K., 2000.
29
Sankaran, V., and Menon, S., “Structure of Premixed Turbulent Flames in the Thin-
Reaction-Zones Regime,” Proceedings of the Combustion Institute, Vol. 28, The Combus-
tion Inst., Pittsburgh, PA, 2000, pp. 203–209.
30
Sankaran, V., and Menon, S., “Subgrid Combustion Modeling of 3-D Premixed Flames
in the Thin-Reaction-Zone Regime,” Proceedings of the Combustion Institute, Vol. 30, The
Combustion Inst., Pittsburgh, PA, 2005, pp. 575–582.
31
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, R. T. Edwards,
Philadelphia, 2001.
32
Lieuwen, T., “Modeling Premixed Combustion-Acoustic Wave Interactions: A Re-
view,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 765–776.
33
Clanet, C., Searby, G., and Clavin, P., “Primary Acoustic Instability of Flames Propa-
gation in Tubes: Cases of Spray and Premixed Combustion,” Journal of Fluid Mechanics,
Vol. 385, No. 4, 1999, pp. 157–197.
34
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determination
of the Transfer Function of a Laminar Premixed Flame,” Proceedings of the Combustion
Institute, Vol. 28,The Combustion Inst., Pittsburgh, PA, 2000, pp. 765–772.
35
Borghi, R., “Turbulent Combustion Modeling,” Progress in Energy and Combustion
Science, Vol. 14, No. 4, 1988, pp. 245–292.
36
Poinsot, T., Veynante, D., and Candel, S., “Quenching Processes and Premixed Turbu-
lent Combustion Diagrams,” Journal of Fluid Mechanics, Vol. 228, No. 8, 1991, pp. 561–
606.
37
Dunlap, R., and Brown, R. S., “Exploratory Experiments on Acoustic Oscillations
Driven by Periodic Vortex Shedding,” AIAA Journal, Vol. 19, No. 3, 1981, pp. 408–409.
38
McMurtry, P. A., Riley, J. J., and Metcalfe, R. W., “Effects of Heat Release on the
Large-Scale Structure in Turbulent Mixing Layers,” Journal of Fluid Mechanics, Vol. 199,
No. 2, 1989, pp. 297–332.
39
Williams, F. A., Combustion Theory, 2nd ed., Benjamin/Cummings Publishing Com-
pany, Metro Park, CA 1985.
40
Candel, S. M., “Combustion Dynamics and Control: Progress and Challenges,” Pro-
ceedings of the Combustion Institute, Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002,
pp. 1–28.
41
Ducruix, S., Schuller, T., Durox, D., and Candel, S., “Combustion Dynamics and Insta-
bilities: Elementary Coupling and Driving Mechanisms,” Journal of Propulsion and Power,
Vol. 19, No. 5, 2003, pp. 722–734.
ACOUSTIC-VORTEX-FLAME INTERACTIONS IN GAS TURBINES 313
42
Sinibaldi, J. O., Mueller, C. J., and Driscoll, J. F., “Local Flame Propagation Speeds
Along Wrinkled Unsteady Stretched Premixed Flames,” Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 827–832.
43
Hussain, A. K. M. F., “Coherent Structures—Reality and Myth,” Physics of Fluids,
Vol. 26, No. 10, 1983, pp. 2816–2850.
44
Metcalfe, R. W., Hussain, A. K. M. F., Menon, S., and Hayakawa, M., “Coherent Struc-
tures in a Turbulent Mixing Layer: A Comparison Between Direct Numerical Simulations
and Experiments,” Turbulent Shear Flows, Vol. 5, 1987, pp. 110–123.
45
Rivero, A., Ferre, J. A., and Giralt, F., “Organized Motions in a Jet in Crossflow,”
Journal of Fluid Mechanics, Vol. 444, No. 10, 2001, pp. 117–149.
46
Lilley, D. G., “Swirl Flows in Combustion: A Review,” AIAA Journal, Vol. 15, No. 8,
1977, pp. 1063–1078.
47
Michalke, A., “On the Inviscid Instability of the Hyperbolic Tangent Profile,” Journal
of Fluid Mechanics, Vol. 19, No. 4, 1964, pp. 543–556.
48
Ho, C.-M., and Huerre, P., “Perturbed Free Shear Layers,” Annual Review of Fluid
Mechanics, Vol. 16, 1984, pp. 365–424.
49
Gutmark, E., and Ho, C.-M., Preferred Modes and the Spreading Rate of Jets,” Physics
of Fluids, Vol. 26, No. 10, 1983, pp. 2932–2938.
50
Stone, C., and Menon, S., “Adaptive Swirl Control of Combustion Instability in Gas
Turbine Combustors,” Proceedings of the Combustion Institute, Vol. 29, The Combustion
Inst., Pittsburgh, PA, 2002, pp. 155–160.
51
Jeong, J., and Hussian, F., “On the Identification of a Vortex,” Journal of Fluid
Mechanics, Vol. 285, 1995, pp. 69–94.
52
Mongia, H. C., “TAPS - A 4th Generation Propulsion Combustor Technology for Low
Emissions,” AIAA Paper 03-2657, Jan. 2003.
53
Menon, S., “CO Emission and Combustion Dynamics Near Lean Blow-Out in Gas
Turbine Engines,” ASME GT2004-53290, June 2004.
54
Panda, J., and McLaughlin, D. K., “Experiments on the Instabilities of a Swirling Jet,”
Physics of Fluids, Vol. 6, No. 1, 1994, pp. 263–276.
55
Lucca-Negro, O. and O’doherty, T., “Vortex Breakdown: A Review,” Progress in Energy
and Combustion Science, Vol. 27, No. 4, 2001, pp. 431–481.
56
Chen, R.-H., and Driscoll, J. F., “The Role of the Recirculation Vortex in Improving
Fuel-Air Mixing within Swirling Flames,” Proceedings of the Combustion Institute, Vol. 22,
The Combustion Inst., Pittsburgh, PA, 1988, pp. 531–540.
57
Billant, P., Chomaz, J.-M., and Huerre, P., “Experimental Study of Vortex Break-
down in Swirling Jets,” Journal of Fluid Mechanics, Vol. 376, No. 12, 1998, pp. 183–
219.
58
Ruith, M. R., Chen, P., Meiburg, E., and Maxworthy, T., “Three-Dimensional Vortex
Breakdown in Swirling Jets and Wakes: Direct Numerical Simulation,” Journal of Fluid
Mechanics, Vol. 486, No. 8, 2003, pp. 331–378.
59
Eggenspieler, G. and Menon, S., “LES of Premixed Combustion and Pollutant Emis-
sion in a DOE-HAT Combustor,” Journal of Propulsion and Power, Vol. 20, No. 6, 2004,
pp. 1076–1086.
60
Syred, N., and Beer, J. M., “Combustion in Swirling Flows: A Review,” Combustion
and Flame, Vol. 23, No. 2, 1974, pp. 143–201.
61
Ling, W., Troutt, J. N., and Crowe, C. T., “Direct Numerical Simulation of a Three-
Dimensional Temporal Mixing Layer with Particle Dispersion,” Journal of Fluid Mechanics,
Vol. 358, No. 3, 1998, pp. 61–85.
314 S. MENON
62
Lazaro, B. J., and Lasheras, J. C., “Particle Dispersion in the Developing Free Shear
Layer, Part 1: Unforced Flow Turbulent Channel Flow,” Journal of Fluid Mechanics,
Vol. 235, 1992, pp. 143–178.
63
Eaton, J. K., and Fessler, J. R., “Preferential Concentration of Particles by Turbulence,”
International Journal of Multiphase Flow, Vol. 20, No. 1, 1994, pp. 169–209.
64
Sankaran, V., and Menon, S., “Vorticity-Scalar Alignments and Small-Scale Structures
in Swirling Spray Combustion,” Proceedings of the Combustion Institute, Vol. 29, The
Combustion Inst., Pittsburg, PA, 2002, pp. 577–584.
65
Smith, T. M., and Menon, S., “The Structure of Premixed Flames in a Spatially Evolving
Turbulent Flow,” Combustion Science and Technology, Vol. 119, No. 1–6, 1996, pp. 77–106.
66
Menon, S., Stone, C., and Patel, N., “Multi-Scale Modeling for LES of Engineering
Designs of Large-Scale Combustors,” AIAA Paper 2004-0157, Jan. 2004.
67
Menon, S., McMurtry, P., and Kerstein, A. R., “A Linear Eddy Mixing Model for Large
Eddy Simulation of Turbulent Combustion,” LES of Complex Engineering and Geophysical
Flows, edited by B. Galperin and S. Orszag, Cambridge Univ. Press, Cambridge, U.K., 1993.
68
Menon, S., and Calhoon, W., “Subgrid Mixing and Molecular Transport Modeling
for Large-Eddy Simulations of Turbulent Reacting Flows,” Proceedings of the Combustion
Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 59–66.
69
Menon, S., and Pannala, S., “Subgrid Combustion Simulations of Reacting Two-Phase
Shear Layers,” AIAA Paper No. 98-3318, July 1998.
70
Chakravarthy, V., and Menon, S., “Large-Eddy Simulations of Turbulent Premixed
Flames in the Flamelet Regime,” Combustion Science and Technology, Vol. 162, No. 1,
2001, pp. 1–48.
Chapter 12
Timothy C. Lieuwen∗
Georgia Institute of Technology, Atlanta, Georgia
Nomenclature
A FL = flame-surface area
A LC = limit-cycle amplitude
c = speed of sound
d = flame thickness
E = energy
E a = overall activation energy
f = frequency
F = flame-transfer function
G = flame-area-transfer function
h = enthalpy
h R = heat of reaction per unit mass of reactant
Ia = net acoustic energy flux out of the flame
k = wave number
K a = Karlovitz number
Le = Lewis number
L F = flame length
ṁ = mass flow rate
M = Mach number
M = total mass
Ma = Markstein number
N = dimensionless length scale defined in Eq. (12.34)
p = pressure
Q = heat-release rate
r = radial coordinate
Copyright c 2005 by the author. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Associate Professor, School of Aerospace Engineering.
315
316 T. C. LIEUWEN
Greek
β = ratio of the flame length to radius, β = L f /R
ε = normalized amplitude of velocity disturbance, ε = u/u o
εf = disturbance amplitude for flashback defined in Eq. (12.33)
φ = equivalence ratio oscillation
γ = specific heats ratio
κ = stretch rate
η = nondimensional disturbance convection velocity defined in Eq. (12.24)
λ = wavelength
Superscripts
( ) = fluctuating quantities
(¯ ) = mean quantities
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 317
Subscripts
1 = upstream side of the flame
2 = downstream side of the flame
a = acoustic disturbances
b = flame base, burned gas
c = conical flames
F = flame, fuel
in = inlet value
v = vortical disturbances
s = entropy disturbances
o = mean quantities
ox = oxidizer
w = wedge flames
I. Introduction
Acoustic disturbances
LF
Fluid mechanic
disturbances
II. Background
A. Combustion Regimes
Acoustic wave–flame interactions involve unsteady kinetic, fluid mechanic, and
acoustic processes over a large range of timescales. Fundamentally different phys-
ical processes may dominate in different regions of the relevant parameter space,
depending on the relative magnitudes of various temporal–spatial scales. The
different regimes of interaction between acoustic waves, the combustion process,
and broadband turbulent fluctuations can be readily visualized with the combustion
diagram in Fig. 12.2.4 The regions denoted by wrinkled and corrugated flamelets
correspond to situations in which the reactions occur in thin sheets that retain their
laminar structure. These sheets become increasingly wrinkled and multiconnected
with increasing values of u /SL , where u is the fluctuating velocity and SL is the
laminar flame speed. Acoustic or vortical flow disturbances push these reaction
sheets around, causing additional wrinkling of the flame over well-defined spatial
and temporal scales. In addition, the local propagation velocity of these sheets into
the reactants is modulated by local strain rate, pressure, and temperature fluctua-
tions in the wave.
In the distributed reaction zone, the smallest turbulent length scales are of the
same order as the flame thickness and thus alter the laminar flame structure. The
well-stirred reactor regime corresponds to the limit in which mixing occurs much
more rapidly than chemical kinetics, and reaction occurs homogeneously over a
distributed volume. Some debate exists about the characteristics of the combustion
process in the regions noted by well-stirred reactor and distributed reaction zone.5
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 319
100 Re =10
4
RMS velocity/ flame speed (u'/SL)
Da=1
Well Stirred Reactor
10 Distributed
Re =102 Ka=1
Reaction Zone
Corrugated
Flamelets
1
Wrinkled
Re =1 Flamelets
0.1
1 10 100 1000
Integral length scale/flame thickness ( /d)
Fig. 12.3 Instantaneous pressure contours (solid lines) and velocity vectors in a dump-
stabilized combustor geometry for which the flame was excited from upstream. Aver-
age flame location given by heavy inclined line. Adapted from Lee and Lieuwen.10
instabilities. The phase velocity and growth rate of the flow instabilities is strongly
affected by the amplitude of forcing and the relationship between the acoustic-
forcing frequency and the intrinsic flow instability. Acoustic excitation often causes
their shedding rate to lock in to the forcing frequency or one of its harmonics.
When the forcing frequency is much lower than the natural shedding frequency,
a collective-interaction phenomenon occurs in which the flow instabilities form
at their natural rate but subsequently coalesce to form a vortex whose forma-
tion frequency coincides with the forcing frequency.15 For example, if the forcing
Flame
Vorticity
iso-countours
Fig. 12.5 Dependence of shear-wave convection velocity and growth rate in a jet flow
on Strouhal number and ratio of boundary-layer thickness to jet radius.
frequency is 10 times lower than that of the intrinsic instability, 10 vortices would
discretely form but subsequently merge into a single larger vortex.
The characteristics of the instability waves that grow and merge to form these
large-scale structures are a function of the specific characteristics of the burner-
exit shear layer, such as coflow velocity, and specifically of the receptivity of this
shear layer to external disturbances. For example, the phase speed of the convected
vortical instability waves is not necessarily equal to the flow velocity but varies
with frequency and shear-layer characteristics. The instability wave-growth rate
similarly varies with frequency and the shear-layer characteristics.
To illustrate, Fig. 12.5 plots Michalke’s16 theoretical curves of the dependence of
the phase speed u c of shear-layer instability waves in a jet flow on Strouhal number,
Sθ = f θ/u o , for several values of the momentum thickness θ, jet radius R, and
ratio R/θ . The figure shows that, for all R/θ values, the ratio u c /u o equals unity
and 0.5 for low and high Strouhal numbers. For thin boundary layers, for example,
R/θ = 100, the phase velocity actually exceeds the maximum axial-flow velocity
in a certain Sθ range. The dispersive character of the instability-wave convection
velocity has been confirmed by a variety of measurements. For example, Baillot
et al.17 measured u c /u o values of 0.88 and 0.98 at 35 and 70 Hz, respectively, on
a conical Bunsen flame. Similarly, Durox et al.18 measured u c /u o = 0.5 values at
150 Hz in an axisymmetric wedge flame.
In general, the disturbance field may have both acoustic and vortical components,
whose relative magnitudes depend strongly on the vortex-shedding dynamics at
the burner-shear layer. For example, Richards and coworkers11 found that the
disturbance field transitioned from a convected character to one with an acoustic
character at low and high frequencies ( f > 100 Hz), respectively.
Even in the absence of convected vorticity waves, the impact of the fluctuating
flame position on the acoustic field causes the acoustic disturbance field to have
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 323
0.30
0.25
CH*'/CH*o 0.20
0.15
0.10
0.05
0.00
0.00 0.05 0.10 0.15 0.20 0.25 0.30
u'/uo
A. Flame Sheets
1. Basic Concepts and Analytical Framework
We begin this section with a derivation of the fundamental equations describing
the dynamics of flame sheets. This approach for treating unsteady flame problems
was apparently first introduced by Markstein24 and Marble and Candel,25 and
subsequently developed by many other authors.17,26−33 More detailed treatments
can be found in Refs. 24 and 34.
Consider a flame front of arbitrary shape whose instantaneous surface is de-
scribed by the parametric equation f (x , t) = 0. It is assumed that the surface is
continuous with a uniquely defined normal at each point. Markstein derives the
following kinematic equations, which relate the flame-surface position to the local
flow and flame-burning velocities.24
∂f
+ u 1 · ∇ f − S1 |∇ f | = 0 (12.1)
∂t
∂f
+ u 2 · ∇ f − S2 |∇ f | = 0 (12.2)
∂t
where S and u denote the flame speed relative to the gases and flow velocity,
respectively. Subscripts 1 and 2 denote the value of each quantity on the up- and
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 325
downstream side of the flame, respectively. Either of the preceding two expres-
sions are often referred to as the G-Equation in the flame-dynamics literature. The
flowfields up- and downstream of the flame are coupled across the front by the
relations24 :
Mass: ρ1 S1 = ρ2 S2 (12.3)
Normal momentum: p1 + ρ1 S12 = p2 + ρ2 S22 (12.4)
Tangential momentum: u 1 − u 2 ) × ∇ f = 0
( (12.5)
u 1 · u 1 u 2 · u 2
Energy: ρ1 S1 h 1 + = ρ2 S2 h 2 + (12.6)
2 2
where ρ and h denote density and enthalpy, respectively. The dynamics of the
thermodynamic and flow variables up- and downstream of the flame are described
by the mass, momentum, and energy-conservation equations.
In many cases, the quantity of primary interest is the overall heat release. (For
acoustically compact flames, only the spatially integrated heat release is impor-
tant. However, for high-frequency oscillations, where the flame is not acoustically
compact, its spatial distribution is also important.) The global heat-release rate of
the flame is given by
Q(t) = ρ1 S1 h R dAFL (12.7)
S
where the integral is performed over the flame surface AFL and h R is the heat
release per unit mass of reactant. Equation (12.7) shows the four fundamentally
different ways of generating heat-release disturbances in a premixed flame: fluctu-
ations in density, flame speed, heat of reaction, or flame area. As noted by Clanet
et al.,35 they can be classified based on either their modification of the local internal
structure of the flame (such as the local burning rate) or its global geometry (such
as its area).
Fluctuations in the mass flow rate of reactive mixtures into the flame, cor-
responding to ρ1 S1 in Eq. (12.7), is the most basic mechanism of heat-release
oscillation. These density fluctuations could be caused by both acoustic and en-
tropy fluctuations. The flame’s burning rate S1 is sensitive to the perturbations
in pressure, temperature, strain rate, or mixture composition that accompany the
acoustic wave. These pressure and temperature fluctuations are usually generated
by acoustic perturbations, whereas the strain-rate fluctuations are associated with
acoustic or vortical velocity fluctuations.
Flame-area fluctuations are associated with disturbances in the flame’s position
and orientation that, in turn, are generated by fluctuations in either the local burning
rate or flow velocity. To illustrate the disturbance of a flame by an acoustic velocity
disturbance, Fig. 12.7 shows a photograph from Ducruix et al.36 of a simple Bunsen
flame disturbed by acoustic-flow oscillations generated by a loudspeaker placed
upstream of the flame. The figure clearly shows the large distortion of the flame
front that is evidenced by the pronounced cusp in the center of the flame. This
flame disturbance is convected downstream by the mean flow, so that it varies
spatially over a convective wavelength.37
326 T. C. LIEUWEN
Instantaneous
flame position
Cold Reactants x=ζ(y,t)
Transmitted
Reflected Wave Wave
Θ
Hot Products
Incident Acoustic
Wave Convected vortical
and entropy disturbances
As shown in Ref. 22, this leads to the following approximate expressions coupling
the axial velocity u and pressure p across the flame:
u 2 u 1 S1 γ − 1 p1
− = (
− 1)Ms − (12.9)
c̄1 c̄1 S̄ 1 γ p̄1
p2 = p1 (12.10)
where Ms , γ , and
refer to the laminar-flame-speed Mach number, Ms = S̄ 1 /c̄1 ,
specific heat ratio, and mean temperature ratio across the flame. This equation
neglects variations in γ and terms of O(Ms2 ). To this order, the unsteady pressure
is continuous across the flame. However, there is a jump in unsteady velocity
across the flame; that is, the flame looks like an acoustic volume source or a
monopole. The terms on the right side of Eq. (12.9) quantifying this jump are
related to the flame’s unsteady rate of heat release and result in the amplification
of acoustic waves. This jump is directly proportional to the temperature jump
(or, more fundamentally, the gas-expansion ratio) across the flame and the
flame-speed Mach number Ms which typically has quite low values (∼0.001 for
a stoichiometric methane–air flame). Assuming a typical acoustic scaling, that
is, p ∼ ρcu , it can be seen that the second fluctuating term on the right side
results in a velocity increment across the flame that is on the order of Ms and,
thus, quite small. The relative magnitudes of the S1 term on the right side of
Eq. (12.9) and the fluctuating velocity-perturbation quantities on the left depend
on the specific processes causing the flame-speed perturbation. More detailed
analyses in Sec. III.A.4 suggest that flame-speed perturbations caused by pressure
and/or temperature fluctuations are of similar magnitude S1 / S̄ 1 ∼ O( p / p̄). Thus,
acoustic-wave amplification induced by the pressure or temperature sensitivity of
the flame speed is nonzero, but it is of O(M S ) and therefore quite weak.
Assuming that the source terms on the right side of Eq. (12.9) are small, the
acoustic field can be accurately calculated by ignoring them. As such, the leading-
order calculation of the acoustic field is equivalent to replacing the flame front
with a passive-temperature discontinuity, as assumed by Mason.38 Ignoring these
terms does not allow one to calculate the slight amplification or damping of sound
waves at the flame, but it does allow for an accurate calculation of the acoustic
field that is disturbing the flame. We will proceed in a sequential fashion, first
considering the leading-order problem, then considering the higher-order effects
needed to calculate the acoustic amplification and damping by the flame.
Consider first the problem of an acoustic wave of pressure amplitude PI im-
pinging normally (i.e.,
= 0) on the flame. The acoustic pressure and velocity
328 T. C. LIEUWEN
where R and T are the reflection and transmission coefficients, denoting the am-
plitude of the acoustic waves reflected from and transmitted through the flame.
Matching the pressure and axial velocity at the flame by using the leading-order
approximation of Eqs. (12.9) and (12.10), p2 = p1 and u 2 = u 1 , leads to the fol-
lowing solution for R and T:
p̄1 c̄1 p̄ p̄ S̄ 1 p̄ c2
(12.14)
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 329
R
r
R
Lf
ξ (r , t ) ξ (r , t )
Fig. 12.9 Illustration of conical (left) and wedge-shaped (right) flame geometries.
Depending on the temperature ratio across the flame, the magnitude and phase of
the flame-burning velocity response, and the angle of incidence between the wave
and the flame, the acoustic disturbance can be damped or amplified. Plots showing
typical results can be found in Ref. 22.
Equation (12.14) only describes the acoustic-field energy balance. Energy
is also added to the vortical and entropy fields. Although not shown here, Ref.
22 includes expressions for these convected waves. Although the vortical wave
couples with the acoustic field at O(Ms ) (resulting in acoustic damping and
vorticity amplification), at this order entropy waves are forced disturbances and
do not have an impact on the acoustic or vortical fields. Full coupling between all
three disturbances occurs at O(Ms2 ).
Note that all energy amplification and damping processes in this equation are
relatively small, being of O(Ms ). Flame-area fluctuations, which are discussed in
the next section, are usually a much stronger source of acoustic energy.
where u and v are axial and radial velocity, respectively. Note that a differential
element of flame surface is related to the flame position through the relation
2
∂ξ
dAFL = 1+ dr (12.16)
∂r
We assume that the mean velocity is uniform and purely axial (i.e., v̄ = 0), that
v = 0, and that the mean flame speed is constant. Although these assumptions are
not necessary to proceed with the analysis, they do yield more transparent results
that retain many of the basic phenomena of interest.
We next focus on the flame area’s linear dynamics. Nonlinear dynamics are
considered later. As shown in the equation below, the linear solution to the equation
for ∂ξ /∂r , or the flame-surface area, can be decomposed into two canonical
components: the homogeneous solution (second term on the right side) containing
the influence of boundary conditions and the particular solution caused by spatial
nonuniformities in flow forcing (or flame speed). To simplify the equation, a flame
coordinate along the nominal flame surface position s is introduced:
s
∂ξ (s, t) 1 ∂ s − x 1 s
= u x ,t− dx + u s = 0, t − − u base
∂s uo ∂s uo uo uo
0
(12.17)
where u base denotes the velocity of the end of the flame sheet at the attachment point.
A spatially uniform velocity disturbance ∂u /∂s = 0 excites only the homogeneous
solution. This disturbance can be understood by first assuming that the flame edge
moves exactly in step with the particle velocity u (s = 0) = u base . In this case,
the entire flame moves up and down in a bulk motion without change in flame
orientation or area. However, if a flame-anchoring boundary condition is imposed,
for example, u base = 0, such that the flame remains fixed at a point, the flow
disturbance excites a flame-front disturbance that originates at the boundary and
propagates along the flame front.
If the disturbance flowfield is spatially nonuniform, ∂u /∂s = 0, the particular
solution is excited. This results in waves originating at the spatial location(s) of
flow nonuniformity that also propagate along the flame at roughly the mean flow
velocity.
As will be shown subsequently, the flame area acts as a low-pass filter to flow
disturbances, so that the amplitudes of the two canonical solutions individually
decay with frequency as roughly 1/f, but, in general, do not become identically
zero. As such, the transfer function relating the response of the flame area to a
spatially uniform velocity disturbance (where only the homogeneous solution is
excited), (A /Ao )/(u /u o ) has a value of unity at zero frequency and decays with
frequency. In contrast, when the flame is perturbed by a spatially nonuniform dis-
turbance (so that both the homogeneous and particular solutions are excited), the
flame area consists of a superposition of the two solutions. As such, although
each solution decreases with frequency, their sum has oscillatory behavior in
cases where they constructively interfere and even causes the transfer function
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 331
(A /Ao )/(u /u o ) to exceed unity. This result was first predicted and then experi-
mentally confirmed by Schuller et al.40 and Durox et al.,18 respectively. In addition,
the two solutions can destructively interfere, and in certain cases, exactly cancel
each other so that the resulting transfer function (A /Ao )/(u /u o ) identically equals
zero.
We now consider linear solutions to the nondimensionalized form of Eq. (12.15),
that is,
1 + β 2 ∂ξ 2
∂ξ ∂r
+ = u(ξ, t) (12.18)
∂t 1 + β2
ξ (r = 1, t) = 0 (12.20)
where subscripts c and w are used to denote axisymmetric conical and wedge
flames, respectively. It is shown in Ref. 41 that the solution for the conical flame
area transfer function, G c = (Ac /Ac,o )/(u /u o ) is
where
St (1 + β 2 )
St2 = (12.23)
β2
uo β 2
η= (12.24)
uc 1 + β 2
Note that the contribution of the flow nonuniformity and boundary conditions are
explicitly separated, following the discussion of Eq. (12.17).
The solution for the wedge-flame-transfer function, G w = (Aw /Aw,o )/(u /u o ),
is
The magnitude of this ratio is identical for both wedge and conical flames (see
Fig. 12.10). The phase of this ratio is different for conical and wedge flames and
plotted in Fig. 12.11.
It is instructive to analyze the characteristics of this ratio for limiting values of the
parameters η and St2 . First, note that in the η → 0 limit (i.e., a spatially uniform dis-
turbance), the flame dynamics for both the wedge and conical flames is controlled
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 333
Fig. 12.10 Strouhal number dependence of the magnitude of the ratio of the transfer
functions caused by the flow-forcing and boundary-condition terms for different values
of η.
Fig. 12.11 Strouhal number dependence of the phase of the ratio of the transfer
functions caused by the flow-forcing and boundary-condition terms for wedge flames
for different values of η. Shaded regions indicate points where boundary-condition
and flow-forcing terms are in phase.
334 T. C. LIEUWEN
The boundary-condition and flow-forcing terms dominate when η <1 and η >1,
respectively. For long flames (β 1), this physically corresponds to situations
in which the disturbance-phase velocity is greater than and less than the mean
flow velocity, respectively. The two terms tend toward equal magnitudes when
η = 1. These points can be clearly observed in Fig. 12.10. Note also that the flow-
disturbance and boundary-condition terms are 180 deg out of phase for low St2
values (see Fig. 12.11).
In the St2 1 limit, the contribution from both the boundary conditions and
flow-forcing term are equal, as shown in Fig. 12.10 and in Eq. (12.30):
G c,Flow
Lim = −1
St2 → ∞ G c,BC
G w,Flow
Lim = − exp [i(η − 1)St2 ] (12.30)
St2 →∞ G w,BC
Equation (12.30) also shows that, in this limit, the relative magnitude contribu-
tion of these two terms is independent of η (assuming that the ηSt2 product does
not simultaneously go to zero). Moreover, the two terms are always out of phase for
conical flames irrespective of the Strouhal number and η; typical relative phases
range between 140 and 220 deg. In contrast, for wedge flames the phase differ-
ence between the two contributions monotonically increases with St2 , as shown in
Fig. 12.11 (the shaded bands in the figure indicate regions of constructive inter-
ference).
The dependence of the magnitude and phase of the conical flame-transfer func-
tion G c (St2 , η) on St2 is plotted in Figs. 12.12 and 12.13, respectively. Consider
the magnitude results first. Note that the transfer-function gain is identical in the
cases in which η = 0 or 1. Physically, this corresponds to cases in which the dis-
turbance velocity is uniform (η = 0) or its phase speed matches the flame-front
disturbance velocity (η = 1). The gain-transfer function differs for all other
disturbance-phase velocity cases. Note also that the gain value is always less than
one and generally decreases monotonically with St2 , although some ripple occurs
at higher St2 values because of constructive and destructive interference between
G c,Flow and G c,BC . The transfer-function phase starts at zero degrees at low St2
and initially increases monotonically with St2 .
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 335
Fig. 12.12 Axisymmetric conical linear transfer function G c (St2 , η) amplitude de-
pendence on the reduced Strouhal number (St2 ) for different values of η.
For wedge flames, the transfer-function gain G w (St2 , η) is plotted in Fig. 12.14.
Note that all gain values tend toward values of unity at low St2 . However, only
in the uniform velocity case, η = 0, does the gain monotonically decrease with
increases in St2 . In all other cases, the gain increases to values of greater than unity
because of constructive interference between G w,Flow and G w,BC .
Fig. 12.13 Axisymmetric conical linear transfer function G c (St2 , η) phase depen-
dence on the reduced Strouhal number (St2 ) for different values of η.
336 T. C. LIEUWEN
Fig. 12.14 Axisymmetric wedge linear transfer function Gw(St2 , η) amplitude depen-
dence on the reduced Strouhal number (St2 ) for different values of η.
A (t)/Ao = n u u (t − τ ) (12.31)
where n u = 1/ S̄ 1 , τ conical = [(η + 1)L F ]/3u o , and τ wedge = [2(η + 1)L F ]/3u o .
Eq. (12.31) indicates that the time response of the flame area to perturbations in
acoustic velocity is delayed by a retarded time τ . This retarded time equals the
time taken for the mean flow to convect some fractional distance of the flame
length, which is equivalent to replacing the distributed flame by a concentrated
source at this location; for example, for a conical flame this effective position of
concentrated heat release is L eff ≈ (η + 1)L F /3.
We next turn to the response of the flame area in the general, nonlinear case. Note
that in the linear case the transfer function is described by only two parameters;
that is, G Lin = G(St2 , η). For the general nonlinear case, however, the gain G also
depends on ε and β; that is, G = G(St2 , η, β, ε). Before considering specific
results, several general conclusions that can be obtained from analysis of the
equations should be considered. The key mechanism of nonlinearity is illustrated
in Fig. 12.15. In this illustration, a flame is perturbed into a corrugated front but
then allowed to relax back to its steady-state, planar position. Flame propagation
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 337
Fig. 12.15 Sketch of a flame that is initially wrinkled (top), showing the destruction
of flame area by kinematic restoration processes (bottom).
normal to itself smoothes out the wrinkle, so that its area eventually returns to being
constant in time. As shown by the dashed lines in the bottom sketch, opposed flame
branches merge to form a sharp cusp and propagate forward to destroy flame area.
The cusp-formation time τ cusp of a front with a corrugation of length scale λ f is
proportional to λ f /S1 .
The rate of flame-area destruction depends nonlinearly on the amplitude of the
flame-front disturbance. Large-amplitude corrugations are smoothed out at a rela-
tively faster rate than small-amplitude perturbations. In the same way, short-length-
scale corrugations are smoothed out faster than long-length scales. As discussed
further below, corrugation smoothing is the reason that nonlinearity is enhanced
at higher disturbance frequencies, which generate shorter-length-scale flame cor-
rugations.
Consider the effects of these nonlinearities on the flame disturbances generated
at the boundaries and regions of flow nonuniformity, as discussed after Eq. (12.17).
If only the homogeneous solution of Eq. (12.17) is excited, as in a spatially uniform
velocity-perturbation field, nonlinear effects always cause the nonlinear transfer
function relating flame area (at the disturbance frequency) and velocity perturba-
tions, (A /Ao )/(u /u o ), to monotonically decrease with disturbance amplitude. In
other words, the linear transfer function is always larger than the nonlinear transfer
function.
If the velocity field is nonuniform, the effects of nonlinearity on both the par-
ticular and homogeneous solutions causes the overall solution characteristics to
depend on whether the two solutions lie in a region of constructive or destructive
interference. If they lie in a region of constructive interference, a conclusion simi-
lar to the previous one holds: the transfer function (A /Ao )/(u /u o ) decreases with
disturbance amplitude. Opposite behavior may occur if the two solutions destruc-
tively interfere, because they are affected unequally by nonlinearity. Nonlinearities
have a longer time to destroy flame area for the boundary-condition term that prop-
agates the entire flame length, as opposed to the flow nonuniformity terms that are
excited at each point along the flame. As will be shown subsequently, the result is
that the nonlinear transfer function can actually exceed its linear value.
338 T. C. LIEUWEN
Next, consider the dominant factors that affect flame-area nonlinearity. Note
that these nonlinearities arise fromthree sources. The first factor is the nonlinear
flame dynamics through the term 1 + β 2 (∂ξ/∂r )2 in Eq. (12.18). The second
factor is the static nonlinearity introduced through the dependence of the flame area
on flame-position gradient through a term with the same form, 1 + β 2 (∂ξ/∂r )2
[see Eq. (12.21)]. In both of these cases, the nonlinearity is purely geometric in
origin and is introduced by the relationship between the instantaneous flame-front
normal and flame-position gradient. The third nonlinearity is caused by the flow
forcing itself and the dependence of the disturbance velocity at the flame front on
the flame position u(ξ, t).
The fact that the first two sources of nonlinearity are identical can be used to write
the final expressions for the flame area, Eq. (12.21), in a revealing form. By substi-
tuting Eq. (12.18) into Eq. (12.21), note that the term (1 + β 2 (∂ξ/∂r )2 )/(1 + β 2 )
which appears in both the area integrals can be written as
1 + β 2 (∂ξ /∂r )2 ∂ξ
= u(ξ, t) − (12.32)
1 + β2 ∂t
d. Flame Aspect Ratio. When β 1, that is, when the flame is very long,
the flame dynamics are approximately described by the equation
∂ξ ∂ξ
± = u(ξ, t)
∂t ∂r
In this case, the flame dynamics are linear, although the flow-forcing term does not
need to be linear. Thus, β is an important nonlinearity parameter for this problem;
that is, the flame’s area response can be anticipated to exhibit a linear dependence
on the perturbation velocity for much larger ε values at large β values.
The rest of this section presents typical results comparing the linear and nonlin-
ear flame-transfer function. These results were obtained by numerically integrating
the governing equation (12.18).41 The nonlinear flame-transfer function was deter-
mined by computing the flame area only at the forcing frequency (because higher
harmonics are also excited) via the Fourier transform.
The boundary condition, Eq. (12.20), cannot be used for disturbance-velocity
magnitudes in which the instantaneous flow velocity is lower than the flame speed.
In this case, the flame flashes back and Eq. (12.20) must be replaced by a different
condition (see Sec. III.A.7). Results are shown in the following text for disturbance
amplitudes up to the point of flashback, given by the disturbance amplitude ε = εf ,
where
1
εf = 1 − (12.33)
1 + β2
Figure 12.16 plots the St2 dependence of the nonlinear transfer-function gain for
a wedge flame. The gain-transfer functions are normalized by their linear values,
Fig. 12.16 Strouhal number dependence of the ratio of the magnitude of the flame
area–velocity transfer function to its linear value for the axisymmetric wedge flame,
β = 1, η = 0.
340 T. C. LIEUWEN
Fig. 12.17 Strouhal number dependence of the ratio of the magnitude of the flame-
area–velocity-transfer function to its linear value for an axisymmetric wedge and
conical flame, β = 1, ε/ε f = 0.99.
G/G Lin . Results are shown for a uniform-velocity field η = 0 and a value of β = 1.
As predicted previously, the response tends to its linear value in all cases at low
St2 . Note the substantial reduction in the flame area relative to its linear value; that
is, there is a substantial degree of gain saturation. In agreement with the Strouhal
number argument (Sec. III.A.3.a), the degree of nonlinearity increases with St2 .
For the present case, the gain for the wedge flames decreases by about 75% at
ε = εf . Although the phase of the area response is not shown here it exhibits little
amplitude dependence, varying by a total of about 8 deg at ε = εf .
Figure 12.17 shows the St2 dependence of the gain and phase of the nonlinear
transfer function for a wedge and conical flame at a given velocity amplitude for a
range of η values. The results are shown for a velocity amplitude of ε/εf = 0.99.
Note that, consistent with the boundary conditions argument (Sec. III.A.3.c), the
wedge exhibits a far more nonlinear response than a conical flame.
In the η = 0 case, nonlinearity causes a monotonic decrease in transfer function
with disturbance amplitude. This result is not true, in general, because of the inter-
actions between the boundary-condition and flow-forcing nonuniformity solutions
noted previously. To illustrate, Fig. 12.18 shows the flame response for a wedge
flame when η = 2 (i.e., disturbances are traveling at approximately half the mean
flow speed). Note that the gain results are not normalized by their linear value here.
The gain result indicates that, in the 6 < St2 < 8 range, the nonlinear transfer func-
tion actually exceeds its linear value. This result can be understood by noting that
this behavior occurs in the vicinity of the regions where the linear-transfer func-
tion achieves a minimum. At these St2 values, the contributions attributable to the
boundary conditions and the flow-forcing terms exactly cancel each other, leading
to zero gain. As the velocity amplitude is increased, nonlinearities cause the gain
attributable to both the boundary conditions and the flow-forcing terms to decrease.
Because the individual gain decreases by different amounts, the total gain does not
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 341
go to zero at the St2 value at which the linear gain is zero but actually shifts to a
higher St2 value in the ε = 0.2ε f case. At higher disturbance levels, the two terms
never exactly cancel and the gain does not go to zero. Rather, a monotonic decrease
occurs in the gain of the transfer function with the increase in velocity amplitude.
These results are consistent with the related measurements of Durox et al.18 Analo-
gous behavior also occurs in conical flames, although less dramatically. In addition,
unlike the η = 0 case, the phase exhibits a stronger amplitude dependence.
Although these results have focused on theoretical predictions, they are generally
found to be in good agreement with experiments, assuming that the correct velocity
characteristics are used in the model.
These results have implications on the type of bifurcations that may be ob-
served in unstable combustors in cases in which heat-release nonlinearities are
the dominant source of nonlinearity (see the discussion in Chap. 1). In situations
in which the gain curves resemble those qualitatively shown in Fig. 12.16, only
supercritical bifurcations will occur and only a single stable-limit-cycle amplitude
ALC is possible. In situations in which the gain exceeds, then is less than, the
linear gain, multiple stable solutions for the instability amplitude may exist, and
subcritical bifurcations are possible. Depending on the operating conditions and
frequency, both types of gain curves can be obtained. This can be seen in Fig. 12.19,
which plots the dependence of A /Ao vs ε at the two conditions, St2 = 2.5 and
St2 = 6.25. Note the similarity in shape of these curves with those plotted in
Chapter 1 for the sub- and supercritical bifurcations. This similarity implies that
unstable combustors driven by these flame-area-fluctuation mechanisms may or
may not exhibit hysteresis and triggering, depending on the operating condition
and frequency.
342 T. C. LIEUWEN
A’/Ao
Another point of interest is that the linear gain and nonlinear saturation ampli-
tude are not related; for example, it is not possible to draw definitive conclusions
about the flame’s saturation amplitude based on how strongly it responds to low-
amplitude fluctuations. It is possible to find regions in which change in a particular
parameter causes the saturation amplitude to either increase, not change, or de-
crease with variations in the linear gain. For example, Fig. 12.20 illustrates an
example in which the linear gain and nonlinear saturation amplitude have opposite
trends. This case corresponds to a situation in which the flame length is doubled
at a constant frequency. For example, such a trend could be associated with a
decrease in flame speed. This example has clear implications on the applications
of linear-stability analyses to inferring instability amplitude trends; increases in
instability-growth rate do not necessarily imply increases in combustion-instability
amplitude. The amplitude could also stay the same or, as just discussed, could ac-
tually decrease.
the acoustic wave and flame preheat and reaction zone. Following McIntosh,47
define the following ratios of these length and timescales:
These ratios are related by the Mach number of the flame-burning velocity:
SL 1
Ms = = (12.35)
cu τM N
Ea
θE = (12.36)
Rg Tb
where E a is the overall activation energy, Rg is the gas constant, and Tb is the
burned-gas temperature. Four different regimes exist whose characteristics depend
on the relative magnitudes of these parameters:
10 0
-10
-20
phase (ν)
1 -30
|ν|
-40
-50
0.1 -60
0.01 0.1 1 10
Flame response time/Acoustic Period, τ
where we assume an exp (−iωt) time dependence, Le is the Lewis number, and
s = 1 − 4iτ M /Le and r = 1 − 4iτ M (12.38)
or decrease the flame speed; this sensitivity is quantified by the Markstein number
Ma. For weakly stretched flames, a linearized expression for this dependence is:
Sc /SL = 1 − Ma K a (12.39)
where Sc and SL denote the stretched and unstretched flame speeds, K a = κ/SL d
is the Karlovitz number, d is the flame thickness, and κ is the stretch rate, given by:
κ = ∇t · u − Sc / (12.40)
where ∇t and denote the tangential divergence of the velocity field at the flame
surface and its radius of curvature, respectively.
Equation (12.40) shows that, in the steady case, the burning-velocity depen-
dence on curvature (second term) and hydrodynamic stretch (first term) combines
into a single term.49 Quasi–steady-strain fluctuations cause flame-speed oscilla-
tions about its nominal value with a Markstein number-dependent magnitude.
However, in the general unsteady case, Joulin’s50 analysis predicts that the flame-
speed sensitivity to the two terms in Eq. (12.40) have different frequency-response
characteristics. The unsteady strain effect diminishes with frequency, whereas the
unsteady-curvature term is independent of frequency. The latter prediction has
apparently not been assessed experimentally or computationally. The former pre-
diction is consistent with Im and Chen’s51 calculations, which predicted that the
flame-speed response to strain-rate fluctuations attenuates as the frequency in-
creases (see Fig. 12.22). This figure plots the instantaneous flame-consumption
speed as a function of the instantaneous stretch rate. The unclosed line that spans
the entire range of y-axis values corresponds to the steady-state result and shows the
consumption-speed augmentation by stretch. The filled circles correspond to the in-
stantaneous correspondence between consumption speed and stretch rate when the
stretch rate is oscillated sinusoidally. The 10-Hz case closely follows the steady
line. With increasing frequency, however, the amplitude of consumption-speed
oscillations monotonically decreases and becomes quite small at 1000 Hz. This
result emphasizes the importance of dynamic effects and reinforces a point from
Sec. II.A that the flame response is not quasi-steady, even at relatively low fre-
quencies.
Although unsteady stretch effects have not been systematically evaluated in
acoustically forced flames, they may be responsible for the filtering phenomenon
experimentally observed by Baillot and coworkers.52−54 When the Bunsen flame is
forced with high-frequency, low-amplitude disturbances, they observed that flame
wrinkles are only evident at the flame base and quickly decay with at axial locations
farther downstream. This behavior may be caused by the increased importance of
the flame’s curvature-dependent burning velocity and the very short convective
wavelengths of the imposed disturbances at these higher frequencies.
Finally, consider the effect of mixture-composition (i.e., equivalence ratio) os-
cillations on the flame speed. Insight into this sensitivity can be gained from the
steady-state dependence of flame speed on equivalence ratio. First, the flame speed
has a maximum value (i.e., ∂S1 /∂φ = 0) under near-stoichiometric conditions,
implying that the flame speed under such conditions is insensitive to equivalence
ratio oscillations. Second, because the flame-speed sensitivity to equivalence ratio
346 T. C. LIEUWEN
Sc/SL
∂ S1 /∂φ generally grows as φ decreases from unity, the amplitude of the flame-
speed oscillations generated by a fixed φ disturbance grows with decreases in the
mean equivalence ratio.
Similar to the fluctuating strain case, the flame’s response to dynamic fluctua-
tions in mixture composition decreases with increases in frequency. Sankaran and
Im studied the dynamic response of lean methane–air-premixed flames to such
fluctuations and also found substantial dynamic effects in phase.55 For example,
at 400 Hz, their analysis predicts a large-phase shift between the instantaneous
equivalence ratio and flame speed, with the effect that the flame speed actually
increases or decreases with decreases or increases in the equivalence ratio.
The overall response of the flame’s heat release Q(t) to flame-speed perturba-
tions is complex, because the flame’s position, and therefore surface area, is also
affected. As will be shown subsequently, the contributions of the flame-speed and
flame-area perturbations constructively and destructively interfere, respectively,
depending on the frequency of oscillations. We describe next an analysis56 of the
response of a conical flame to a convected φ disturbance that illustrates these cou-
pled dynamics. The analysis is similar to that described in Sec. III.A.3, except
now the flame-speed fluctuation terms are retained and the velocity disturbances
are neglected. To simplify the expressions from Ref. 56, it will be assumed that
β 1, so that (1 + β 2 )/β 2 ≈ 1 and St2 ≈ St.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 347
Assume that the equivalence-ratio perturbation is convected with the mean flow
velocity and, thus, has an axial distribution given by
φ (x, t) = φb exp [−iω(t − x/ū)] = φb exp (−iωt) exp [i St(1 − r/R)]
(12.41)
φb in Eq. (12.41) denotes the perturbation in equivalence ratio at the flame base. The
flame-speed perturbation is related to the perturbation in mixture stoichiometry by
dS1
S1 = · φ (12.42)
dφ φ
Assume that the equivalence-ratio disturbance occurs at constant density; that is,
ρ1 = 0. Define the following flame-transfer functions to perturbations in equiva-
lence ratio, Fφ
Q φ / Q̄
Fφ = = FH + FS = FH + (FS,dir + FA ) (12.44)
φb /φ̄
where
d(h R /h̄ R ) 2
FH = {1 + i St − exp (i St)}
d(φ/φ̄) φ̄ St 2
d(S1 / S̄ 1 ) 2
FS,dir = {1 + i St − exp (i St)} (12.45)
d(φ/φ̄) φ̄ St 2
d(S1 / S̄ 1 ) 2
FA = {1 − (1 − i St) exp (i St)}
d(φ/φ̄) φ̄ St 2
dφb (t)
Q φ (t)/ Q̄ = n H φb (t − τ H ) + n S (12.46)
dt
where
d(h R /h̄ R ) LF 1 L F d(S1 / S̄1 )
nH = , τH = , ns =
dφ φ̄ 3ū 3 ū dφ φ̄
pattern caused by the alternating phase relationship between FS,dir and FA . The
total heat-release response Fφ increases until St ∼ 4 and decreases in an oscillatory
manner.
Figure 12.25 plots the effect of the mean equivalence ratio on the flame-transfer
function. It shows that mixture stoichiometry has little effect on the transfer-
function magnitude for St 1 and at subsequent minima. In most cases, however,
the flame response increases with a decreasing equivalence ratio because of the
increased sensitivity of the flame speed to the equivalence ratio for lean mixtures
referred to previously. Although not shown, the heat-release response can either
lead or lag the φ perturbation, depending on the mean φ value.
An important conclusion to be drawn from these results is the importance of
both the local and global effect of a perturbation on the overall flame response. For
example, a flame-speed perturbation causes not only a local change in heat-release
rate per unit area, but also the overall flame area. The transfer-function results
illustrated previously show that inclusion of both effects is crucial in modeling the
overall flame response.
The nonlinear heat-release response to high-amplitude flame-speed disturbances
has not been analyzed to date. A partial accounting for the nonlinear dependence
of the φ disturbance on the acoustic field was modeled in a quasi-steady sense
by Peracchio and Proscia.58 They assumed the following relationship for the re-
sponse of the instantaneous mixture composition leaving the nozzle exit to velocity
perturbations:
φ̄
φ(t) = (12.47)
1 + ku (t)/u
where k is a constant with a value near unity. They also utilized a nonlinear
relationship relating the heat release per unit mass of mixture to the instantaneous
equivalence ratio, similar to the correlations used to derive the preceding results.
-2
10
-3
10 f i=15 kHz
PSD (arb. units)
-4
10
-5
10 f i=7.5 kHz
-6
10
-20 -10 0 10 20
f-f i
Fig. 12.26 Measured spectra of scattered acoustic field excited by 7.5- and 15-kHz
incident waves. Adapted from Lieuwen.61
wave frequency f i . The broadened spectrum of the scattered field is caused by the
randomly moving flame front, resulting in Doppler-shifted scattered waves. These
characteristics are clearly illustrated by data plotted in Fig. 12.26, which shows
the spectra of 7.5- and 15-kHz sound waves scattered from a turbulent flame. The
narrow band, coherent peak at the incident wave frequency, and distributed side
bands are clearly evident in the figure.61 Note the broader bandwidth of the in-
coherent sidebands for the 15-kHz sound waves. This result can be understood
by noting that a harmonically oscillating acoustic wave incident on a reflecting
surface moving with a Mach number, M̄, generates reflected waves oscillating at
the Doppler-shifted frequency
· n)
(1 + M
f refl = f drive (12.48)
(1 − M · n)
where n denotes the unit normal direction of the incident wave. Noting that the
Mach number of flame-front motion is very small, this expression can be written
as
· n )2 1/2
( f − f drive )2 1/2 = ( f )2 1/2 ≈ 2 f drive ( M (12.49)
This equation shows that the bandwidth of the scattered waves f grows with rms
flame-front velocity and incident-wave frequency.
If the flame does not add energy to the acoustic field, the energy in the incoherent
sidebands is derived from the coherent wave. Thus, the wrinkled characteristics
of the flame act as a potential source of damping of coherent acoustic energy. In
cases in which the flame amplifies sound waves, the overall energy balance of the
352 T. C. LIEUWEN
Acoustic
Disturbance
Flame
Fig. 12.27 Image of instantaneous pressure field and flame front. Reproduced with
permission from A. Laverdant and D. Thevenin.63
0
10
-1
10
-2
10
0.1 0.2 0.3 0.4 0.5
σ/λ
question lies in the value of the ratio of the disturbance wavelength, whether acous-
tic or convective, to the turbulent flame-brush thickness. For typical longitudinal
mode instabilities, the ratio of σ/λ is often very small, implying that wrinkled
flame effects provide only a small correction from laminar-flame analyses. At
these same frequencies, however, the convective wavelength could potentially be
of the same magnitude as σ . This potential similarity implies that the conclusions
of Sec. III.A.3 or III.A.4, where convective disturbances or flame wrinkles vary
over a convective scale, could be modified in the turbulent case. In the same way,
the response of flames to high-frequency acoustic waves, such as during screeching
instabilities, could also be substantially different than the response anticipated from
laminar-flame analyses.
oscillate at half the period of the acoustic oscillations. This parametric acoustic
instability is caused by the periodic acceleration of the flame front by the unsteady
velocity field, which separates two regions of differing densities. With increased
amplitudes, these organized cellular structures break down into a highly disordered,
turbulent front. In the case in which the ambient flowfield is highly turbulent, Vaezi
and Aldredge70 found that the parametric instability still appeared. In addition, they
noted that for sufficiently high-turbulence levels, the appearance of the parametric
instability did not result in additional acceleration of the flame front. This finding
is in contrast with the case in which the ambient flowfield is quiescent, where the
parametric instability results in substantial flame acceleration.
Markstein first recognized that the period-doubling behavior occurring during
the parametric instability was indicative of a parametrically pumped oscillator, in
which the parametric excitation is caused by the oscillatory acceleration field. The
flame-front dynamics can be described by a parametric oscillator equation of the
356 T. C. LIEUWEN
form66
d2 y(k, t) dy(k, t)
A +B + [Co − C1 cos(wt)]y(k, t) = 0 (12.50)
dt 2 dt
where A, B, and C are coefficients defined in Ref. 66, k is the wave number
of the perturbation, and ω is the frequency of imposed oscillation. The damping
coefficient, B, is always positive, whereas the coefficient Co is negative if the
planar flame front is nominally unstable. In the case in which Co is negative,
this equation has the properties that the solution is unstable in the absence of
imposed oscillations (i.e., C1 = 0), is stabilized in the presence of small but finite
amplitude perturbations, and is destabilized in the presence of large-amplitude
parametric oscillations.
Flow oscillations and the resulting oscillatory strain rates also cause local or
global flame extinction. Such extinction events in unsteady flows have been demon-
strated in several studies in fundamental geometries, such as counterflow flames.72
In addition, the flame may extinguish at locations of sharp curvature, such as in
cusps. Such extinction phenomena are routinely observed in turbulent flames.
Unsteady extinction and reignition of local or global regions of the flame intro-
duces nonlinearity in the acoustic–heat-release relationship. This point was em-
phasized in the measurements of the growth of the instability amplitude by Poinsot
et al.,73 who observed that the point of saturation of its amplitude corresponded
to the point where the instantaneous chemiluminescence emissions reached zero
during part of a cycle. Modeling extinction requires treating the internal flame
structure, and is considerably more involved than the simple front-tracking ap-
proach detailed in this chapter. For this reason, existing models have handled
extinction in a phenomenological manner. For example, Dowling74 observed that
the instantaneous heat release cannot go negative, thus limiting its magnitude to
100% of its mean value. She incorporated these observations into a phenomeno-
logical model for the finite amplitude response of a flame to velocity perturbations
in which heat-release saturation occurred at amplitudes at which the instantaneous
heat-release value went to zero.
B. Well-Stirred Reactors
1. Basic Analytical Framework
Returning to the combustion regime diagram in Fig. 12.2, consider next the
opposite extreme to the flamelet regime, the “well-stirred reactor” (WSR) regime.
It has been suggested that flame–acoustic interactions in this regime can be modeled
by generalizing the steady WSR equations to include nonsteady effects. Unsteady
reactor models are also routinely used to study kinetically driven instabilities
in multistep chemical mechanisms and extinction and ignition phenomenon.75,76
These unsteady reactor equations can be derived from a straightforward spatial
integration of the conservation equations over the WSR region by assuming that
all spatial quantities are uniform77 :
dM
= ṁ in − ṁ (12.51)
dt
dE
= ṁ in h in − ṁh (12.52)
dt
dMk
= ṁ in Yk,in − ṁYk − Ẇk (12.53)
dt
where ṁ and Ẇk are the mass flow rate and consumption rate of the kth species,
respectively. M, E, and Mk denote the total mass, total energy, and total mass
of the kth species in the reactor, respectively. The subscript in denotes the inlet
value. The steady-state characteristics of the well-stirred reactor are controlled by
the ratio of the chemical kinetic time to the reactor residence time, given by the
ratio of the mass flow rate and reactor volume, τres = ṁ/V . The reactor volume or
residence time cannot, in general, be specified by simplified analysis, because it
358 T. C. LIEUWEN
is determined by reaction and mixing rates; prior studies have used experimental
and computational analysis to determine these quantities, which are then used as
inputs to simplified models.78
Note that the preceding equations assume that the perturbations are spatially
uniform. If the flame zone is acoustically compact, such an approximation may
be adequate to describe acoustic perturbations. Entropy and vorticity fluctuations
could potentially be of much shorter length scales than the reactor size, how-
ever, indicating that the perturbation variables and flow-strain field are spatially
distributed in the reactor.
The total heat release from the reactor is given by the volume integral
Q(t) = ω̇h r dV (12.54)
v
where ω̇ and V denote the volumetric reactant consumption rate and reactor vol-
ume. This expression is analogous to that of Eq. (12.7), except it is on a volumetric,
rather than surface-area basis. The terms ρ 1 S1 and d A are replaced by ω̇ and d V ,
respectively. Note that heat-release fluctuations are generated by reaction rate,
reactor volume, and heat of reaction fluctuations.
where A is a pre-exponential factor, Y F and Yox denote fuel and oxidizer mass
fractions, and a, b, and c denote sensitivity coefficients. Whereas sensitivity of ω̇
to disturbances in any of the quantities in Eq. (12.55) can be determined from the
value of the exponential coefficients, it must be emphasized that any disturbance
does not occur in isolation. For example, fluctuations in temperature have an impact
on the residence time or reactor fuel and oxidizer concentration. The sensitivities
of ω̇ to perturbations in pressure, mass flow rate, inlet temperature, and equivalence
ratio have been discussed in Refs. 79, 80, and 81, whose results are summarized
subsequently.
Because the sensitivity of the Arrhenius term e−Eu /RT to temperature varia-
tions grows with temperature, the effect of temperature perturbations in the inlet-
reactant stream on ω̇ grow with increases in mean reactor temperature, such as
with increases in mean equivalence ratio. Opposite sensitivity is obtained with
equivalence-ratio oscillations, whose effect on ω̇ oscillations grows with decreas-
ing equivalence ratio. This point is analogous to the dependence of the flame speed
on the equivalence ratio discussed in Sec. III.A.4 and can be understood by noting
that ω̇ reaches a maximum near φ = 1 (i.e., it has no sensitivity to φ perturba-
tions) and decreases as φ becomes leaner. Both of these sensitivities were deter-
mined assuming that the reactor residence time was fixed; that is, oscillations in
reactive-mixture composition or temperature did not affect the residence time.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 359
Blowout
Decreasing
Residence Time
Reactor Temperature
Fig. 12.30 Dependence of rate of reaction (solid line) and convection (dashed line) on
reactor temperature. Steady-state reactor solution occurs at high-temperature inter-
section of two points. After Park et al.82
To extend the reactor approach to situations where the flame was convectively
noncompact and, thus, flow disturbances varied substantially over the flame region,
Lieuwen et al.84 treated the combustion zone as a distribution of infinitesimal, inde-
pendent reactors whose input conditions were given by that of the local flow at the
associated location. Although this heuristic treatment allowed for a consideration
of important noncompactness effects, its basic assumption of reactor independence
is questionable (e.g., as noted in Sec. III.A.3, disturbances generated at one point of
a flamelet convect downstream and affect its dynamics at other points). However,
as is the more general problem with these reactor-based approaches, it is not clear
how to incorporate these interaction effects in a rational manner.
Incorporating finite amplitude effects into unsteady, well-stirred reactor calcula-
tions is straightforward, although it may require numerics for time stepping through
Eqs (12.51–12.53).79,81 The WSR equations are often used for model problems in
nonlinear dynamics studies because of the complex, even chaotic dynamics85 they
exhibit.
Thus, any fluctuation in heat release that does occur is a dynamic effect; that is, the
pressure perturbations referred to previously could potentially cause heat-release
oscillations at sufficiently large frequency ω. In this case, it is then necessary to
model the dynamics of the global reaction-zone response to the perturbation. This
situation is analogous to that encountered in flamelet studies (see Sec. III.A.4), in
which flame speed and area fluctuations are coupled; for example, quasi-steady
fluctuations in flame speed do not cause the global heat release to oscillate be-
cause of the accompanying oscillations in flame area. The difference is that these
coupled dynamics can be reasonably modeled in the flamelet case from first prin-
ciples. It is not clear how to couple these dynamics in WSR models, given their
phenomenological nature.
IV. Conclusion
The ultimate goal of this work is to develop models that can predict the qualita-
tive, and preferably quantitative, dependence of flame response in realistic com-
bustors on geometric and fuel composition parameters. Reasonable, quantitative
predictions of flame–acoustic interaction phenomena have been demonstrated
for a few simple configurations, such as the laminar Bunsen burner of Ducruix
et al.36 or Baillot et al.,53 or the nominally flat flame of Searby and Rochwerger.66
These successes illustrate the rapid progress being made in modeling kinematic
processes in flame–acoustic interactions. In addition, progress is being made in
developing hybrid models that use computational simulations to determine various
components of the combustor system–flame interactions.86 The development of
accurate, predictive combustion-response models for realistic (i.e., turbulent)
configurations has not been achieved, however, and remains a key challenge.
The subsequent discussion suggests some requirements needed to achieve this
capability.
First and most generally, it seems critical that better coordination between mod-
els and experiments be achieved. At present, a significant part of the relevant liter-
ature consists of essentially decoupled theoretical models or experimental studies,
even in rather fundamental configurations. For example, a substantial number of
fundamental studies have theoretically investigated the response of flat, laminar
flames to pressure perturbations.42−47 No serious effort appears to have been ini-
tiated to subject these predictions to experimental scrutiny. Similarly, although
equivalence-ratio oscillations are known to play an important role in exciting heat-
release oscillations, no experimental work has been performed to examine the
accuracy of models that relate them to the resultant heat-release oscillations. Even
though these highly fundamental studies may be far removed from practical flames,
they are prerequisite building blocks toward modeling realistic systems.
Second, work is needed to develop simplified models of vortex–flame interac-
tions. The existing theoretical work on this subject is largely numeric. Analytical
methodologies for modeling unstable, reacting shear flows have been developed87
and need to be extended to incorporate the unsteady flow effects on the flame.
Third, predicting the response of flames to finite amplitude waves is immature.
Substantial progress could be made by a set of experiments that isolate the key non-
linear processes that modelers need to focused on. Interpretive guidance of these
results can be achieved by parallel systematic studies of potential nonlinearities.
362 T. C. LIEUWEN
Acknowledgments
This work was supported by the National Science Foundation, General Electric,
the U.S. Department of Energy, and Georgia Institute of Technology.
References
1
De Luca, L., Price, E., and Summerfield, M., Nonsteady Burning and Combustion Sta-
bility of Solid Propellants, Progress in Aeronautics and Astronautics, Vol. 143, AIAA,
Washington, DC, 1992.
2
Sirignano, W., Fluid Dynamics and Transport of Droplets and Sprays, Cambridge Univ.
Press, Cambridge, England, U.K., 1999.
3
Harrje, D., Reardon, F. (eds.) Liquid Propellant Rocket Instability, SP-194, NASA, 1972.
4
Peters, N., “Laminar Flamelet Concepts in Turbulent Combustion,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 21, 1986, pp. 1231–120.
5
Poinsot, T., and Veynante, D., Theoretical and Numerical Combustion, Edwards,
Philadelphia, PA, 2001.
6
Turns, S., An Introduction to Combustion, McGraw–Hill, New York, 2000.
7
Chu, B. T., and Kovasnay, L. S. G., “Nonlinear Interactions in a Viscous, Heat Conduct-
ing; Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, 1958, pp. 494–514.
8
Kovaszay, L. S. G., “Turbulence in Supersonic Flow,” Journal of the Aeronautical Sci-
ences, Vol. 20, No. 10, 1953, pp. 657–674.
9
Jou, W. H., and Menon, S., “Modes of Oscillation in a Nonreacting Ramjet Combustor
Flow,” Journal of Propulsion and Power, Vol. 6, No. 5, 1990, pp. 535–543.
10
Lee, D. H., and Lieuwen, T., “Acoustic Nearfield Characteristics of a Conical, Premixed
Flame,” Journal of the Acoustical Society of America, Vol. 113, No. 1, 2003, pp. 167–177.
11
Ferguson, D., Richards, G., Woodruff, S., Bernal, S., and Gautam, M., Proceedings of
the 2nd Joint Meeting U.S. Sections of the Combustion Institute, Pittsburgh, PA, 2001.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 363
12
Schadow, K., and Gutmark, E., “Combustion Instability Related to Vortex Shedding
in Dump Combustors and Their Passive Control,” Progress in Energy and Combustion
Science, Vol. 18, 1992, pp. 117–132.
13
Coats, C., “Coherent Structures in Combustion,” Progress in Energy and Combustion
Science, Vol. 22, 1996, pp. 427–509.
14
Stone, C., and Menon, C., “Swirl Control of Combustion Instabilities in a Gas Turbine
Combustor,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
15
Ho, C. H., Nosseir, N. S., “Dynamics of an Impinging Jet. Part 1. The Feedback
Phenomenon,” Journal of Fluid Mechanics, Vol. 105, 1981, pp. 119–142.
16
Michalke, A., Zeitschrift für Flugwissenschaften (in German), Vol. 19, 1971.
17
Baillot, F., Durox, D., and Prud’homme, R., “Experimental and Theoretical Study of a
Premixed Vibrating Flame,” Combustion and Flame, Vol. 88, 1992, pp. 149–168.
18
Durox, D., Schuller, T., and Candel, S., “Combustion Dynamics of Inverted Conical
Flames,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 30, 2004.
19
Strahle, W., “On Combustion Generated Noise,” Journal of Fluid Mechanics, Vol. 49,
No. 2, 1971.
20
Putnam, A. A, “Combustion Roar of Seven Industrial Burners,” Journal of the Institute
of Fuel, Vol. 49, 1976, pp. 135–138.
21
Marble, F., and Candel, S., “Acoustic Disturbance from Gas Non-uniformity Convected
Through a Nozzle,” Journal of Sound and Vibration, Vol. 55, 1977, pp. 225–243.
22
Lieuwen, T., “Theoretical Investigation of Unsteady Flow Interactions With a Premixed
Planar Flame,” Journal of Fluid Mechanics, Vol. 435, 2001, pp. 289–303.
23
Emmons, H. W., “Flow Discontinuities Associated with Combustion,” Fundamentals
of Gas Dynamics, Vol. III. High Speed Aerodynamics and Jet Propulsion, edited by H. W.,
Emmons, Princeton Univ. Press, Princeton, NJ, 1958, p. 584.
24
Markstein, G. H., Nonsteady Flame Propagation, Pergamon Press, New York, 1964.
25
Marble, F. E., and Candel, S. M., “An Analytical Study of the Non-Steady Behavior
of Large Combustors,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 17,
1978, pp. 761–769.
26
Subbaiah, M. V., “Nonsteady Flame Spreading in Two Dimensional Ducts,” AIAA
Journal, Vol., 21, No. 11, 1983, pp. 1557–1564.
27
Poinsot, T., and Candel, S. M., “A Nonlinear Model for Ducted Flame Combustion
Instabilities,” Combustion Science and Technology, Vol. 61, 1988.
28
Yang, V., and Culick F. E. C., “Analysis of Low Frequency Combustion Instabilities in
a Laboratory Ramjet Combustor,” Combustion Science and Technology, Vol. 45, pp. 1–25.
29
Boyer, L., and Quinard, J., “On the Dynamics of Anchored Flames,” Combustion and
Flame, Vol. 82, 1990, pp. 51–65.
30
Fleifel, M., Annaswamy, A. M., Ghoniem, Z. A., and Ghoniem, A. F., “Response of a
Laminar Premixed Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic
Instability Results,” Combustion and Flame, Vol. 106, 1996, pp. 487–510.
31
Ashurst, W., and Sivashinsky, G., “On Flame Propagation through Periodic Flow
Fields,” Combustion Science and Technology, Vol. 80, 1991, pp. 159–164.
32
Joulin, G., Sivashinsky, G., Pockets in Premixed Flames and Combustion Rate, Com-
bustion Science and Technology, Vol. 77, 1991, pp. 329–335.
33
Aldredge, R., “The Propagation of Wrinkled, Premixed Flames in Spatially Periodic
Shear Flow,” Combustion and Flame, Vol. 90, No. 2, 1992, pp. 121–133.
34
Matalon, M., and Matkowsky, B., “Flames as Gas Dynamic Discontinuities,” Journal
of Fluid Mechanics, Vol. 124, 1982, pp. 239–260.
364 T. C. LIEUWEN
35
Clanet, C., Searby, G., and Clavin, P., “Primary Acoustic Instability of Flames Propa-
gating in Tubes: Cases of Spray and Premixed Combustion,” Journal of Fluid Mechanics,
Vol. 385, 1999, pp. 157–197.
36
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determinations
of the Transfer Function of a Laminar Premixed Flame,” Proceeding of the Combustion
Institute, Vol. 28, 2000.
37
Blackshear, P., “Driving Standing Waves by Heat Addition,” Proceedings of Combus-
tion Symposium, Vol. 4, 1953, pp. 553–566.
38
Mason, N., “Contribution to the Hydrodynamical Theory of Flame Vibration,” Pro-
ceedings of the Seventh International Congress on Applied Mechanics, Vol. 2, 1948,
pp. 187–199.
39
Chu, B. T., “On the Generation of Pressure Waves at a Plane Flame Front,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 4, 1953, pp. 603–612.
40
Schuller, T., Durox, D., and Candel, S., “A Unified Model for the Prediction of Lam-
inar Flame Transfer Functions: Comparisons between Conical and V-Flame Dynamics,”
Combustion and Flame, Vol. 134, 2003, pp. 21–34.
41
Preetham, and Lieuwen, T., “Nonlinear Flame-Flow Transfer Function Calculations:
Flow Disturbance Celerity Effects,” AIAA Paper 2004–4035.
42
McIntosh, A. C., “Pressure Disturbances of Different Length Scales Interacting with
Conventional Flames,” Combustion Science and Technology, Vol. 75, 1991, pp. 287–309.
43
Peters, N., and Ludford, G. S. S., “The Effect of Pressure Variations on Premixed
Flames,” Combustion Science and Technology, Vol. 34, 1983, pp. 331–344.
44
Van Harten, A., Kapila, A., and Matkowsky, B. J., “Acoustic Coupling of Flames,”
SIAM Journal of Applied Mathematics, Vol. 44, No. 5, 1984, pp. 982–995.
45
Keller, D., and Peters, N., “Transient Pressure Effects in the Evolution Equation for
Premixed Flame Fronts,” Theoretical and Computational Fluid Dynamics, Vol. 6, 1994,
pp. 141–159.
46
Ledder, G., and Kapila, A. K., “The Response of Premixed Flames to Pressure Pertur-
bations,” Combustion Science and Technology, Vol. 76, 1991, pp. 21–44.
47
McIntosh, A. C., “Deflagration Fronts and Compressibility,” Philosophical Trans-
actions of the Royal Society of London, Series A: Mathematical and Physical Sciences,
Vol. 357, 1999, pp. 3523–3538.
48
Huang, Z., Bechtold, J., and Matalon, M., “Weakly Stretched Premixed Flames in
Oscillating Flows,” Combustion Theory Modeling, Vol. 2, 1998, pp. 115–133.
49
Clavin, P., “Dynamic Behavior of Premixed Flame Fronts in Laminar and Turbulent
Flows,” Progress in Energy and Combustion Science, Vol. 11, 1985.
50
Joulin, G., “On the Response of Premixed Flames to Time Dependent Stretch and
Curvature,” Combustion Science and Technology, Vol. 97, 1994, pp. 219–229.
51
Im, H. G., and Chen, J. H., “Effects of Flow Transients on the Burning Velocity
of Laminar Hydrogen/Air Premixed Flames,” Proceedings of the Combustion Institute,
Pittsburgh, PA, Vol. 28, 2000, pp. 1833–1840.
52
Bourehla, A., and Baillot, F., “Appearance and Stability of a Laminar Conical Premixed
Flame Subjected to an Acoustic Perturbation,” Combustion and Flame, Vol. 114, 1998,
pp. 303–318.
53
Baillot, F., Bourehla, A., and Durox, D., “The Characteristic Method and Cusped Flame
Fronts,” Combustion Science and Technology, Vol. 112, 1996, pp. 327–350.
54
Durox, D., Baillot, F., Searby, G., and Boyer, L., “On the Shape of Flames Under
Strong Acoustic Forcing: A Mean Flow Controlled by an Oscillating Flow,” Journal of
Fluid Mechanics, Vol. 350, 1997, pp. 295–310.
PREMIXED COMBUSTION-ACOUSTIC WAVE INTERACTIONS 365
55
Sankaran, R., and Im, H., “Dynamic Flammability Limits of Methane/air Premixed
Flames with Mixture Composition Fluctuations,” Proceedings of the Combustion Institute,
Pittsburgh, PA, Vol. 29, 2002, pp. 77–84.
56
Cho, J. H., and Lieuwen, T., “Laminar Premixed Flame Response to Equivalence Ratio
Oscillations,” Combustion and Flame, Vol. 140, No. 1–2, pp. 116–129.
57
Abu-Off, G. M., and Cant, R. S., “Reaction Rate Modeling for Premixed Turbulent
Methane-air Flames,” Proceedings of the Joint Meeting of Spanish, Portuguese, Swedish
and British Sections of the Combustion Institute, Madeira, 1996.
58
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat Release/Acoustic Model
for Thermo-Acoustic Instability in Lean Premixed Combustors,” American Society of
Mechanical Engineers, Paper 98-GT-269.
59
Lieuwen, T., “Theory of High Frequency Acoustic Wave Scattering by Turbulent
Flames,” Combustion and Flame, Vol. 126, No. 1–2, 2001, pp. 1489–1505.
60
Lieuwen, T., “Analysis of Acoustic Wave Interactions with Turbulent Premixed
Flames,” Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 29, 2002.
61
Lieuwen, T., Neumeier, Y., and Rajaram, R., “Measurements of Incoherent Acoustic
Wave Scattering from Turbulent Premixed Flames,” Proceedings of the Combustion Insti-
tute, Pittsburgh, PA, Vol. 29, 2002.
62
Lieuwen T., and Cho., J. H., “Coherent Acoustic Wave Amplification/Damping by
Wrinkled Flames,” Journal of Sound and Vibration (to be published).
63
Laverdant, A., and Thevenin, D., “Interaction of a Gaussian Acoustic Wave with a
Turbulent Premixed Flame,” Combustion and Flame, Vol. 134, 2003, pp. 11–19.
64
Searby, G., “Acoustic Instability in Premixed Flames,” Combustion Science and Tech-
nology, Vol. 81, 1992, pp. 221–231.
65
Markstein, G., “Flames as Amplifiers of Fluid Mechanical Disturbances,” Proceedings
of the Sixth National Congress on Applied Mechanics, 1970, pp. 11–33.
66
Searby, G., and Rochwerger, D., “A Parametric Acoustic Instability in Premixed
Flames,” Journal of Fluid Mechanics, Vol. 231, 1991, pp. 529–543.
67
Pelce, P., and Rochwerger, D., “Vibratory Instability of Cellular Flames Propagating
in Tubes,” Journal of Fluid Mechanics, Vol. 239, 1992, pp. 293–307.
68
Vaezi, V., and Aldredge, R., “Laminar Flame Instabilities in a Taylor-Couette Com-
bustor,” Combustion and Flame, Vol. 121, 2000, pp. 356–366.
69
Bychkov, V., “Analytical Scalings for Flame Interaction with Sound Waves,” Physics
of Fluids, Vol. 11, No. 10, 1999, pp. 3168–3173.
70
Vaezi, V., and Aldredge, R. C., “Influences of Acoustic Instabilities on Turbulent-Flame
Propagation,” Experimental Thermal and Fluid Science, Vol. 20, 2000, pp. 162–169.
71
Dowling, A. P., “A Kinematic Model of a Ducted Flame,” Journal of Fluid Mechanics,
Vol. 394, 1999, pp. 51–72.
72
Sung, C. J., and Law, C. K., Combustion and Flame, Vol. 123, 2000, pp. 375–388.
73
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., Esposito, E., and Surget, J., Pro-
ceedings of the Combustion Institute, Pittsburgh, PA, Vol. 22, 1988.
74
Dowling, A. P., “Nonlinear Self-Excited Oscillations of a Ducted Flame,” Journal of
Fluid Mechanics, Vol. 346, 1997, pp. 271–290.
75
Park, Y., and Vlachos, D., “Isothermal Chain-Branching, Reaction Exothermicity, and
Transport Interactions in the Stability of Methane/Air Mixtures,” Combustion and Flame,
Vol. 114, 1998, pp. 214–230.
76
Kalamatianos, S., Park, Y., and Vlachos, D., “Two-Parameter Continuation Algorithms
for Sensitivity Analysis, Parametric Dependence, Reduced Mechanisms, and Stability Cri-
teria of Ignition and Extinction,” Combustion and Flame, Vol. 112, 1998, pp. 45–61.
366 T. C. LIEUWEN
77
Williams, F., Combustion Theory, Addison Wesley, Redwood City, CA, 1985.
78
Sturgess, G., Hedman, P., Sloan, D., and Shouse, D., “Aspects of Flame Stability in a
Research Dump Combustor,” American Society of Mechanical Engineers, Paper 94-GT-49.
79
Richards, G. A., Morris, G. J., Shaw, D. W., Keely, S. A., and Welter, M. J., “Thermal
Pulse Combustion,” Combustion Science and Technology, Vol. 94, 1993, pp. 75–85.
80
Janus, M. C., and Richards, G., “Results of a Model for Premixed Combustion Oscil-
lations,” Proceedings of the 1996 AFRC International Symposium, 1996.
81
Lieuwen, T., Neumeier, Y., and Zinn, B. T., “The Role of Unmixedness and Chemical
Kinetics in Driving Combustion Instabilities in Lean Premixed Combustors,” Combustion
Science and Technology, Vol. 135, 1998, pp. 193–211.
82
Park, S., Annaswamy, A., and Ghoniem, A., “Heat Release Dynamics Modeling of
Kinetically Controlled Burning,” Combustion and Flame, Vol. 128, 2002, pp. 217–231.
83
Park, S., Annaswamy, A., and Ghoniem, A., “Dynamic Characteristics of Kinetically
Controlled Combustion and their Impact on Thermoacoustic Instability,” Proceedings of
ICDERS 2001.
84
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182–190.
85
Rhode, M. A., Rollins, R. W., Markworth, A. J., Edwards, K. D., Nguyen, K., Daw, C. S.,
Thomas, J. F., “Controlling Chaos in a Model of Thermal Pulse Combustion,” Applied
Physics, Vol. 78, No. 4, 1995, p. 2224.
86
Polifke, W., Poncet, A., Paschereit, C. O., Dobbeling, K., “Reconstruction of Acous-
tic Transfer Matrices by Instationary Computational Fluid Dynamics,” Journal of Sound
Vibration, Vol. 245, No. 3, 2001, pp. 483–510.
87
Wee, D., Park, S., Annaswamy, A., and Ghoniem, A., “Reduced Order Modeling of
Reacting Shear Flow,” AIAA Paper 2002-0478, 2002.
88
Bellows, B., and Lieuwen T., “Nonlinear Response of a Premixed Combustor to Forced
Acoustic Oscillations,” AIAA Paper 2004-0455.
IV. Modeling and Diagnostics
Chapter 13
I. Introduction
C OMBUSTION instability has become a major issue for gas turbine manufac-
turers. Stricter emission regulations, in particular, on nitrogen oxides, have
led to the development of new combustion methods such as lean premixed, prevap-
orized (LPP) combustion to replace the traditional diffusion flame. However, LPP
combustion is much more liable to generate strong oscillations that can damage
equipment and limit operating conditions. In this chapter, methods to investigate
combustion instabilities are reviewed (see also Dowling and Stow1 ). The emphasis
is on gas-turbine applications and LPP combustion. The flow is modeled as a one-
dimensional mean with linear perturbations. Calculations are typically done in the
frequency domain. The techniques described lead to predictions for the frequencies
of oscillations and the susceptibility to instabilities in which linear disturbances
grow exponentially in time. Appropriate boundary conditions are discussed, as is
the change in the linearized flow across zones of heat addition and/or area change.
Many of the key concepts are first introduced by considering one-dimensional
perturbations. Later, higher-order modes, in particular, circumferential waves, are
introduced and modal coupling is discussed. The modeling of a simplified combus-
tion system, from compressor outlet to turbine inlet, is described, as is the potential
for acoustic absorbers to control the instability. The approaches are simple and fast
enough to be used at the design stage. The effect of nonlinearity is discussed along
with techniques for predicting the amplitude of the resulting limit cycles.
LPP gas-turbine combustors have the great advantage of very low nitrous oxide
(NOx) emissions, but they are susceptible to instability. These instabilities involve
coupling between the rate of combustion and acoustic waves in the combustor. Un-
steady combustion generates acoustic waves that alter the inlet flow rates of fuel
and air. At lean premixed conditions, this changed fuel–air ratio leads to significant
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Professor of Mechanical Engineering, Department of Engineering. Senior Member AIAA.
† Research Associate, Department of Engineering.
369
370 A. P. DOWLING AND S. R. STOW
where p is the pressure, ρ is the density, u is the velocity, and σi, j is the viscous
stress tensor. Here D/Dt is the material derivative ∂/∂t + u · ∇ and ei represents
the unit vector in the direction of coordinate i. For a perfect gas, we have the gas
law p = Rgas ρT , where T is the temperature, Rgas = c p − cv is the gas constant,
and c p and cv are the specific heats at constant pressure and volume, respectively.
The internal energy per unit mass e is equal to cv T , and the enthalpy h is c p T =
e + p/ρ. Conservation of energy gives the energy equation,
D 1 ∂
ρ e + u2 = −∇ ∇ · ( pu) + q + ∇ · (k∇∇T ) + (σi, j u i ) (13.2)
Dt 2 ∂x j
where k is the conductivity and q is the rate of heat added to the fluid per unit
volume. By using Eq. (13.1b), this can be written as
Dh Dp ∂u i
ρ = ∇ T ) + σi, j
+ q + ∇ · (k∇ (13.3)
Dt Dt ∂x j
showing that it is heat release, heat transfer, and viscous effects that lead to an
entropy increase for a material particle. Taking the curl of Eq. (13.1b) and using
Eq. (13.1a) gives an equation for the development of the vorticity, ξ = ∇ × u,
D ξ ξ 1 1 1 ∂σi, j
= · ∇ u + 3∇ρ × ∇ p + ∇ × ei (13.5)
Dt ρ ρ ρ ρ ρ ∂x j
The first term on the right-hand side describes how the stretching of vortex lines
intensifies the local vorticity, and the last term clearly represents generation of
372 A. P. DOWLING AND S. R. STOW
vorticity by viscous effects. The second term shows that vorticity can be created
when the pressure gradient and density gradient are not aligned. An example of
this would be an acoustic pressure oscillation with a component normal to a flame
front (density gradient), so that, for instance, circumferential waves will generate
vorticity at combustion zones.
We will now assume inviscid flow (σi, j ≡ 0). We will also assume the fluid is an
ideal gas (i.e., in addition to being a perfect gas, it does not conduct heat), and we
take c p and cv to be constant. From the preceding definition of entropy, we find that
S = cv log( p/ρ γ ) (plus an arbitrary constant that we set to zero), where γ = c p /cv
is the ratio of specific heats. We take the flow to be composed of a steady uniform
mean flow (denoted by overbars) and a small perturbation (denoted by primes),
and similarly for the other flow variables. From Eqs. (13.1), (13.4), and (13.5), the
linearized equations for these perturbations are
D̄ρ
∇ · u = 0
+ ρ̄∇ (13.7a)
Dt
D̄u 1
+ ∇ p = 0 (13.7b)
Dt ρ̄
D̄S
ρ̄ T̄ = q (13.7c)
Dt
Dξξ
=0 (13.7d)
Dt
where D̄/Dt = ∂/∂t + ū · ∇ and we have used ξ̄ξ = 0. Combining Eqs. (13.7a–
13.7c) and using S = cv p / p̄ − c p ρ /ρ̄ = 0 leads to the inhomogeneous wave
equation,
1 D̄2 p 2 γ − 1 D̄q
− ∇ p = (13.8)
c̄2 Dt 2 c̄2 Dt
where c is the speed of sound. We see that the vorticity equation (13.7d) is not
coupled to either the pressure or the entropy. For no unsteady heat release, the
pressure equation (13.8) and entropy equation (13.7c) are also uncoupled. Any
perturbation can then be thought of as the sum of three types of disturbances11 : 1)
an acoustic disturbance that is isentropic and irrotational; 2) an entropy distur-
bance that is incompressible and irrotational; and 3) a vorticity disturbance that is
incompressible and isentropic. These three types of disturbances are independent
and can be considered separately. For the pressure (acoustic) disturbance, we have
S = 0 and ξ = 0; hence, ρ = p /c̄2 . Since q = 0, Eq. (13.8) becomes the wave
equation with convection for p ,
1 D̄2
− ∇2 p = 0 (13.9)
c̄2 Dt 2
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 373
and the corresponding u is given by Eq. (13.7b). Disturbances of this type are
acoustic waves; relative to the fluid, they propagate at the speed of sound. For
the entropic disturbance, p = 0 and u = 0. From Eq. (13.7c) we see that the
disturbance is stationary relative to the fluid, that is, it is convected with the mean
flow. This disturbance can be thought of as an entropy wave, and is sometimes
referred to as a convected hot spot. For the vortical disturbance, p = ρ = 0 and
∇ · u = 0, and Eq. (13.7d) shows that this type of disturbance (a vorticity wave)
is also convected with the mean flow. If the mean flow is zero, only acoustic
disturbances propagate.
ρ2 u 2 = ρ1 u 1 (13.10a)
p2 + ρ2 u 22 = p1 + ρ1 u 21 (13.10b)
ρ2 u 2 H2 = ρ1 u 1 H1 + Q A (13.10c)
B. Boundary Conditions
At the inlet and outlet of the combustion system, there are boundary conditions
that the perturbations must satisfy. If the outlet discharges into the atmosphere or a
large plenum chamber (as is often the case for combustor test rigs), we may model
this as an open end, taking p (r, θ, t) = 0. If the inlet is supplied by a plenum
chamber we may treat this also as an open end [ p (r, θ, t) = 0] and, in addition,
assume that no entropy or vorticity disturbances are present. The compressor exit
and turbine inlet of a gas turbine can be modeled as a choked inlet and choked
outlet, respectively, to the combustion system. The nozzle guide vanes at the entry
to the turbine are choked, that is, the mean flow velocity accelerates to the local
speed of sound. At the compressor exit of a gas-turbine combustor, the flow is
nearly choked (meaning that the mass and energy flow rates are nearly constant
irrespective of downstream pressure perturbations) and so a choked inlet boundary
condition provides an approximation. At a choked outlet, the nondimensional mass
flow rate (defined as the mass flow rate multiplied by the square root of stagnation
temperature and divided by the stagnation pressure) is constant and, for one-
dimensional perturbations, Marble and Candel12 showed that at a compact choked
outlet this condition reduces to
u ρ p
2 + − =0 (13.11)
ū ρ̄ p̄
374 A. P. DOWLING AND S. R. STOW
Stow et al.13 have shown that this condition still applies for circumferential-varying
disturbances in a narrow annular gap (disturbances in narrow annular gap geome-
tries are discussed in Sec. IV.C).
For a compact choked inlet, Stow et al.13 considered the interaction of the shock
position and the flow perturbations (see also Yang and Culick14 and Culick and
Rogers15 ), finding that, for one-dimensional disturbances, the perturbations in mass
flux and energy flux are zero just after the shock and that, for circumferential-
varying disturbances in a narrow annular gap, the angular-velocity perturbation
is also zero. From conservation of mass, energy, and angular momentum, these
quantities are also zero at the start of a straight duct with a low-Mach-number
mean flow M̄ 1 just downstream of the choking plane. This gives the inlet boundary
conditions:
ρ u p ρ u
+ = − + (γ − 1) M̄ 1 = w = 0 (13.12)
ρ̄ ū p̄ ρ̄ ū
For a weak shock, one would expect that there is negligible entropy production.
However, the equations imply that the (usually ignored) entropy perturbation down-
stream of the inlet is in fact comparable with the acoustic oscillations. In a frame
of reference moving with the shock, the acoustic perturbations are indeed, much
larger than the entropy disturbance, but viewed in a stationary frame close to the
shock, the discrepancy is not as great. After an area increase to a low-Mach-number
region, the acoustic perturbations are smaller still and are then of the same order as
the entropy perturbations. For circumferential-varying disturbances, a significant
vorticity perturbation is also produced.
Other analytical inlet and outlet boundary conditions, such as acoustically closed
ends (u = 0) or semi-infinite (nonreflecting) pipes, can also be used. Alternatively,
the acoustic impedance of the inlet or outlet can be measuring experimentally by
using microphones and an acoustic source driven over a range of frequencies.
This approach is similar to the measurement of the transfer matrix for a premixer
discussed in Sec. V.B.
where the functions f (t) and g(t) are arbitrary. From the one-dimensional form of
the linearized momentum equation (13.7b), the particle velocity in the x direction
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 375
is given by
∂u 1 ∂ p 1 ∂ 1 ∂
=− = f (t − x/c̄) − g(t + x/c̄) (13.14a)
∂t ρ̄ ∂ x ρ̄ c̄ ∂t ρ̄ c̄ ∂t
that is,
1
u (x, t) = ( f (t − x/c̄) − g(t + x/c̄)) (13.14b)
ρ̄ c̄
Equation (13.15a) then leads to ĝ = − f̂ , and it follows directly from Eq. (13.15b)
that
ωl
cos =0 (13.17a)
c̄
1 2
x=b
with solutions
1 π c̄
ω = ωn = n − (13.17b)
2 l
for integer n 1. These are the resonant frequencies ωn of the duct, describing
the oscillations in which the pressure oscillates without decay. The corresponding
mode shapes are
(2n − 1)π x
p̂(x) = An sin (13.18a)
2l
iAn (2n − 1)π x
û(x) = cos (13.18b)
ρ̄ c̄ 2l
Example 1
Suppose that the rate of heat-release perturbation q (x, t) is influenced by the
local pressure but lags it by a time delay τ . It is convenient to write the constant
of proportionality as 2α/(γ − 1), that is,
2α
q (x, t) = p (x, t − τ ) (13.20)
γ −1
The form of the pressure perturbation can be determined by substituting for q (x, t)
in Eq. (13.19) and seeking a separable solution, p (x, t) = Re( p̂(x)eiωt ). This
substitution leads, after application of the boundary conditions, to p̂(x) of the
form given in Eq. (13.18), and the equation for the resonant frequency ω is
When α = 0, the roots of Eq. (13.21) are the undamped resonant organ-pipe
frequencies ωn .
When α = 0, τ = 0, the quadratic equation (13.21) for ω can be readily solved
to give
1/2
ω = −iα ∓ ωn2 − α 2 (13.22)
ω is now complex. Because the time dependence is eiωt , − Im(ω) is the growth rate
of the disturbances. Here eiωt = exp[αt ∓ i(ωn2 − α 2 )1/2 t], showing that the oscil-
lations grow exponentially in time if α is positive. We have recovered Rayleigh’s
criterion2 from this particular example. Unsteady heat release in phase with the
pressure perturbation has a destabilizing effect and tends to increase the amplitude
of the perturbations. In contrast, for negative α, that is, heat release in antiphase
with the pressure, the oscillations are damped.
When α = 0, τ = 0, Eq. (13.21) would, in general, need a numerical solution,
and some results are shown in Fig. 13.2. In Fig. 13.2 and subsequently, a normalized
frequency f N = Re(ω)/ω1 and a normalized growth rate g N = − Im(ω)/ω1 are
used. However, the general characteristics of the solution can be investigated by
considering small α, and determining the roots iteratively. We have already noted
1.03
normalized frequency, fN
1.02
1.01
0.99
0.98
0.97
0 0.2 0 .4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
0.02
normalized growth rate, gN
0.01
0.01
0.02
0.03
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
Fig. 13.2 Variation with τ of the root of Eq. (13.21) near ω1 : ———, α/ω1 = 0.01;
– – –, α/ω1 = 0.02; and –··–, α/ω1 =−0.01. a) Frequency. b) Growth rate.
378 A. P. DOWLING AND S. R. STOW
that, for α = 0, a root of Eq. (13.21) is at ω = ωn . For small α, this root moves to
ω = ωn + ε, where ε is small; substitution into Eq. (13.21) shows that
Example 2
We now consider that the unsteady heat release is concentrated at a single axial
plane x = b and is related to the oncoming air velocity there with a time delay τ ,
where Q (t) is the rate of heat release per unit area and subscript 1 denotes condi-
tions just upstream of this region of heat release, that is, u 1 (t) = u(b− , t). Chap-
ter 12 discusses forms of the unsteady heat release. In this chapter, we note that the
nondimensional number β can be expected to range from 0 to about 10 and that
in a LPP system τ is typically the convection time from fuel injection to its com-
bustion. For simplicity, u 1 has been taken to be the velocity just upstream of the
flame. However, for consistency, with τ being the fuel-convection time, the flame
model should really be referenced to the perturbations at the fuel-injection point,
as is done in example 5. However, the distance between these points is typically
short compared with the wavelengths, so the phase difference between them will
be small, although they may differ in magnitude by the area ratios.
With the rate of heat release q (x, t) as given in Eq. (13.24a), Eq. (13.19) reduces
to the homogeneous wave equation in the regions x < b and x > b. Integration of
Eqs. (13.7b) and (13.19) across x = b gives
x=b+
p x=b− = 0 (13.25a)
x=b+
∂p γ − 1 dQ
=− 2 (13.25b)
∂ x x=b− c̄ dt
x=b+ γ −1
u x=b− = Q (t) (13.25c)
ρ̄ c̄2
relating the volumetric expansion to the instantaneous rate of heat release. After
substitution for the particular Q (t) in Eq. (13.24b), we obtain
We will consider solutions with time dependence eiωt and want to find the resonant
frequencies ω and the mode shapes.
In x < b, the solution of the homogeneous wave equation that satisfies the inlet
boundary condition p̂(0) = 0 is
where k is the wave number ω/c̄ and the complex constant A has yet to be deter-
mined. Similarly, in x > b, the boundary condition û(l) = 0 leads to
1
normalized frequency, fN
08
06
04
02
0
0 01 02 03 04 05 06 07 08 09 1
β
Fig. 13.3 Variation of frequency with β for the root of Eq. (13.30) near ω1 , taking τ =
0, b=l/10: ——, exact solution; and – – –, one-term Galerkin approximation (13.39).
whereas the velocity jump condition (13.26), on division by Eq. (13.29), gives
The resonant frequencies follow from a numerical solution of Eq. (13.30). Their
dependence on β and τ is shown in Figs. 13.3 and 13.4.
For β = 0, the roots are at ω = ωn . As β varies, for τ = 0, the rate of heat release
is in quadrature with the pressure perturbation [note the 90-deg phase difference
between p and u in Eq. (13.27)] and so shifts only the frequency of oscillation.
A time lag is required for the unsteady heat release to destabilize the system. For
0.1
normalized growth rate, gN
0.05
0.05
0.1
0.15
0.2
0 0.2 0 .4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω1 τ / (2π)
Fig. 13.4 Variation of growth rate with τ for the root of Eq. (13.30) near ω1 , taking
b=l/10: ——, β= 0.2; – – –, β= 0.4; –··–, β= 0.6; and · · · ·, β= 0.8.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 381
τ = 0, the unsteady heat release affects both the growth rate and the frequency
of oscillation. Perturbations grow in time if, in this undamped system, the rate of
heat release has a component in phase with pressure perturbation. It is clear from
the form of the heat release in Eq. (13.24b) and the mode shape in Eq. (13.28) that
this requires
C. Galerkin Series
Another way of solving the inhomogeneous wave equation Eq. (13.19) is through
a Galerkin expansion. The Galerkin expansion involves, the expansion of the pres-
sure perturbation as a Galerkin series:
∞
p (x, t) = ηm (t)ψm (x) (13.33)
m=1
where the functions ψm (x) are the eigensolutions or normal modes of the homoge-
neous wave equation that satisfy the same boundary conditions as p . In general,
these functions are orthogonal, and we will denote their eigenfrequencies by ωm .
Substitution for the pressure perturbation from Eq. (13.33) into Eq. (13.19) then
leads to
∞ 2
d ηm ∂q
+ ω 2
η m ψ m (x) = (γ − 1) (13.34)
m=1
dt 2 m
∂t
Culick and Yang.17 When ∂q /∂t = 0, the nth mode is p̂(x) = ηn (t)ψn (x) with fre-
quency ωn . This acoustic approximation is used when evaluating ∂q /∂t, replacing
the pressure and velocity perturbations by ηn (t)ψn (x) and (η̇n (t)/ρ̄ωn2 ) dψn /dx,
respectively, where the dot denotes a time derivative. If the second derivatives
of the amplitudes arise, they are replaced by the zeroth-order approximation,
η̈n (t) ≈ −ωn2 ηn (t). The errors introduced by these approximations can be checked
by applying the method to find the lowest frequency of oscillation in example 2.
∂q β c̄2 dψ1
(x, t) = η1 (t − τ ) (b)δ(x − b) (13.36)
∂t γ −1 dx
d2 η1 β c̄2 dψ1
2
+ ω 2
1 η 1 = η1 (t − τ ) (b)ψ1 (b) (13.37)
dt E1 dx
The solutions ψn of the homogeneous wave equation are given in Eq. (13.18) and
ψ1 (x) = sin(π x/2l), leading to E 1 = 12 l. Equation (13.37), therefore, simplifies
to
d2 η1 β c̄2 π πb πb
2
+ ω1 η1 = 2 η1 (t − τ ) cos
2
sin (13.38)
dt l 2l 2l
The root of this equation is shown as a dashed line in Fig. 13.3 for the particular
case τ = 0. Comparison with the exact solution given in Eq. (13.30) shows that
the one-term Galerkin expansion gives the correct frequency and gradient at β = 0
but that it rapidly diverges from the exact solution as β increases. The divergence
is not really surprising; this method treats the shift in frequency as small, but it can
be substantial for the type of combustion response typical of LPP systems. The
inadequacy of the one-term Galerkin for a more complicated model problem was
discussed by Dowling.18 Annaswamy et al.19 noted that three terms in the Galerkin
series were needed to model the system dynamics for feedback control.
D. Temperature Gradients
So far, our examples have been somewhat artificial; they have had an unsteady
heat release q (x, t) and yet the mean temperature has been uniform. In practice,
of course, heat release is associated with temperature gradients and the mean
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 383
temperature and density are functions of position. We will introduce these effects
through discussion of the zero-mean-flow case. Then the momentum equation
(13.1b) ensures that the mean pressure is uniform and for linearized perturbations,
∂u
ρ̄ ∇ p
= −∇ (13.40)
∂t
in an inviscid flow. We show in the Appendix that the mass conservation equation
(13.1a) and the entropy equation (13.4) can be combined to give
1 ∂ p γ −1
= ∇ · u + q (13.41)
ρ̄ c̄2 ∂t ρ̄ c̄2
when heat conduction and viscous effects are neglected. Eliminating u from Eqs.
(13.40) and (13.41), we obtain
1 ∂ 2 p 1 γ − 1 ∂q
∇·
− ρ̄∇ ∇p = (13.42)
c̄2 ∂t 2 ρ̄ c̄2 ∂t
In this equation, ρ̄ and c̄ vary with position, but ρ̄ c̄2 = γ p̄ is uniform if the small
dependence of γ on temperature is neglected. We can illustrate the influence of
temperature variation by extending example 2 to the case in which the mean
temperature rises from T̄1 to T̄2 across the zone of heat release at x = b, with
corresponding changes in sound speed and mean density.
Example 3
Consider one-dimensional linear disturbances of frequency ω in the system
illustrated in Fig. 13.5. Just as in example 2, we again apply the boundary conditions
(13.16) and the flame model (13.24).
T1 T2
ρ1 ρ2
c1 c2
x=b
Outside the flame zone x = b, the solutions of the homogeneous wave equa-
tion (13.42), satisfying the appropriate boundary conditions, have the same form
as in example 2 provided the local mean flow variables are used. Hence, using
Eq. (13.27) and Eq. (13.28), we can write in x < b
and in x > b
x=b+ γ −1
u x=b− = Q (t) (13.44c)
ρ̄ 1 c̄21
After substituting for the particular Q (t) in Eq. (13.24b) and using Eq. (13.43),
we obtain
ρ̄ 2 c¯2
tan(k1 b) tan(k2 (l − b)) = (1 − βe−iωτ ) (13.45)
ρ̄ 1 c¯1
A comparison with Eq. (13.30) shows that the varying temperature effects appear
not only in the wave numbers k1 and k2 , which account for propagation effects,
but also in the factor ρ̄ 2 c¯2 /(ρ̄ 1 c¯1 ), which describes the impedance change across
the flame zone. The solid line in Fig. 13.6 shows how the temperature variations
affect the resonant frequency. A typical LPP gas-turbine combustor operates with
a temperature ratio of about 3 (T̄1 ∼ 700 K, T̄2 ∼ 2000 K).
So far, we have assumed that the flame is compact, that is, axially short com-
pared with the wavelengths of the perturbations. If the flame is not compact, we
may approximate the axial heat-release distribution by discretizing into a series of
compact flames, each having the form described earlier. Between these series, there
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 385
2
normalized frequency, fN
1.8
1.6
1.4
1.2
1
1 1.5 2 2.5 3 3.5 4 4.5 5
T2 / T1
Fig. 13.6 Variation of frequency with mean temperature ratio for solution near ω1
taking β = τ = 0, b = l/10: ———, root of Eq. (13.45); – – –, uniformly distributed
heat release between x = 0 and x = 2 b approximated by 10 temperature jumps;
and –··–, the same but using 5 jumps.
is assumed to be no heat release and we use the usual wave propagation (13.15).
The dashed and dashed–dotted lines in Fig. 13.6 show results for applying this
approximation when, instead of a compact flame at x = b, we have a uniformly
distributed heat release between x = 0 and x = 2b. It leads to a 6% shift in the
frequency at a temperature ratio 3. An alternative approach is to seek a continuous
analytical solution. Exact solutions can be found for particular temperature distri-
butions (such as linear variations and power laws)20–25 and also for certain area
variations.26, 27
E. Mean Flow
Most combustion systems involve a mean flow that brings fresh reactants into
the combustion zone. The Mach number of the oncoming flow is so small (typically
less than 0.1) that it is tempting to neglect this mean velocity. The errors introduced
by such an approximation are investigated in this section.
A mean flow has two main consequences. Trivially, it affects the speed of prop-
agation of the acoustic waves, with one-dimensional disturbances then traveling
downstream with speed c̄ + ū and upstream at c̄ − ū. In addition, the mean flow ad-
mits the possibility of convected entropy and vorticity disturbances. These modes
are coupled by the requirement of conservation of mass, momentum, and energy
across zones of heat release.
Example 4
These effects may be illustrated by extending example 3 to include a mean flow.
For definiteness we again apply an open-end inlet boundary condition p (0) = 0.
At the downstream end, we assume an area restriction in which the flow becomes
choked, and so Eq. (13.11) is the appropriate boundary condition. Note that the
hard-end boundary condition u = 0 is recovered from Eq. (13.11) as ū tends to
zero. The heat release will again be considered as concentrated at the fixed plane
386 A. P. DOWLING AND S. R. STOW
x = b, the rate of heat release per unit of cross-sectional area being denoted by
Q (t) with Q (t) given by the particular flame model in Eq. (13.24).
Upstream of the zone of heat release, acoustic waves are propagating in both
directions, and the flow is isentropic. The pressure perturbation is the general
solution of the wave equation with convection (13.9); this gives
p (x, t) = Aeiωt e−iωx/(c̄1 (1+ M̄ 1 )) − eiωx/(c̄1 (1− M̄ 1 )) (13.46)
ρ̄ 1 c̄1 u (x, t) = Aeiωt e−iωx/(c̄1 (1+ M̄ 1 )) + eiωx/(c̄1 (1− M̄ 1 )) (13.47)
The fluxes of mass, momentum, and stagnation enthalpy into the combustion zones
[defined in Eq. (13.10)] can be expressed in terms of the unknown complex A
through Eqs. (13.46) and (13.47).
Downstream of the region of heat release, there might be a convected hot spot
in addition to plane sound waves, and so
p (x, t) = eiωt Ce−iωx/(c̄2 (1+ M̄ 2 )) + Deiωx/(c̄2 (1− M̄ 2 )) (13.48a)
ρ̄ 2 c̄2 u (x, t) = eiωt Ce−iωx/(c̄2 (1+ M̄ 2 )) − Deiωx/(c̄2 (1− M̄ 2 )) (13.48b)
p (x, t) S ρ̄ 2 iω(t−x/ū 2 )
ρ (x, t) = 2
− e (13.48c)
c̄2 cp
p (x, t) S c̄22
c p T (x, t) = + eiω(t−x/ū 2 ) (13.48d)
ρ̄ 2 (γ − 1)c p
for b x l and M̄ 2 = ū 2 /c̄2 . C and D are the amplitudes of the acoustic waves,
S is the amplitude of the entropy wave or convected hot spot, and no vorticity
waves occur in this one-dimensional example. The wave amplitudes C, D, and S
can be found in terms of A through Eqs. (13.10a–13.10c).
Care needs to be taken to recover the jump conditions for zero mean flow from
Eq. (13.10). In the limit ū 1 , ū 2 → 0, Eq. (13.10b) clearly simplifies to p2 = p1 ,
the zero-mean-flow jump condition [Eqs. (13.25a) and (13.44a)]. At first sight
one might assume that Eq. (13.10a) gives ρ̄ 2 u 2 = ρ̄ 1 u 1 as ū 1 , ū 2 tend to zero.
That is wrong. Note it is incompatible with Eq. (13.44c). The resolution of this
apparent inconsistency is that the strength of the entropy wave S enters the jump
conditions (13.10) only in the product ū 2 S. In the limit ū 2 → 0, S tends to infinity,
in such a way as to keep the product ū 2 S and, hence, ū 2 ρ2 and ū 2 T2 , finite. For
low-Mach-number mean flows, very large entropy fluctuations occur downstream
of the flame zone. To see these fluctuations mathematically it is convenient to first
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 387
ρ̄ 2 ū 2 (c p T2 + ū 2 u 2 ) = Q + ρ̄ 1 ū 1 (c p T1 + ū 1 u 1 )
(13.49)
− ( H̄2 − H̄1 )(ρ̄ 1 u 1 + ρ1 ū 1 )
for linear perturbations. After using Eq. (13.48d) to expand c p T2 and taking the
limit ū → 0, this equation simplifies to
ρ̄ 2 ū 2 c̄22
Seiω(t−b/ū 2 ) = Q − c p (T̄2 − T̄1 )ρ̄ 1 u 1 (13.50)
c p (γ − 1)
Physically, Eq. (13.50) shows that entropy is generated unsteadily at the combus-
tion zone whenever Q = c p (T̄2 − T̄1 )ρ̄ 1 u 1 , that is, whenever there is unsteadiness
in the rate of heat addition per unit mass. In particular, the preceding assertion that
ū S remains finite for small ū is confirmed. Equation (13.48c) clearly shows that,
in this limit, the left-hand side of Eq. (13.50) is equal to −ū 2 c̄22 ρ2 /(γ − 1), and
hence, the equation can be rearranged to give
γ −1
ū 2 ρ2 = − Q + (ρ̄ 1 − ρ̄ 2 )u 1 (13.51)
c̄22
where we have used the perfect gas relationships to rewrite c p (γ − 1) × (T̄2 − T̄1 )
ρ̄ 1 /c̄22 as ρ̄ 1 − ρ̄ 2 . Finally, substitution for ū 2 ρ2 in the equation of mass conserva-
tion leads to
γ −1
ρ̄ 2 u 1 = ρ̄ 2 u 2 − Q (13.52)
c̄22
1
normalized frequency, fN
0.8
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
M1
Fig. 13.7 Variation of frequency with Mach number for lowest-frequency mode,
taking Q̄= 0: ———, acoustic mode when diffusion attenuates the entropy waves by
the combustor exit; hence, mode is near ω1 and – – –, including entropy waves; hence,
mode is a low-frequency convection mode.
(and often the most unstable) modes are associated with circumferential waves.
We discuss these more general modes in the next section.
A. Cylindrical Duct
Using cylindrical polar coordinates x, r , and θ, we are interested in a straight
cylindrical duct 0 r b. Because we are assuming that the mean flow is uniform,
we must have v̄ = w̄ = 0. We look for separable solutions for the three types of
disturbance mentioned earlier. The general solution is a sum of such separable
solutions.
We first consider a pressure disturbance. We seek a separable solution by sub-
stituting p = F(t)X (x)B(r )(θ) into Eq. (13.9) to give
F F X 2X
2 X (r B ) −2
+ 2ū + ū − c̄ + +r =0 (13.53)
F FX X X rB
where the prime denotes a derivative with respect to the argument. We see that
solutions take the form F(t) = eiωt , X (x) = eikx , and (θ ) = einθ , with
(r B ) + (λ2 − n 2r −2 )r B = 0 (13.54)
dJn
(λb) = 0 (13.55)
dr
For a given n, this gives an infinite number of discrete solutions for λ. The solutions
are all real,29 and without loss of generality we may take λ 0. We define λn,m to
be the (m + 1)th solution. The full solution can be expressed as an acoustic wave
390 A. P. DOWLING AND S. R. STOW
of the form30
with Bn,m (r ) = Jn (λn,m r ). [Note that the perturbations as given in Eq. (13.56) will
be complex, but it is assumed that we take the real part.] Here α± = ω + ūk± ,
and M̄ is the mean Mach number (which is assumed to be less than unity). Also,
ωc = c̄λn,m (1 − M̄ 2 )1/2 is the complex cutoff frequency of the duct, and A± , which
may be complex, are the wave amplitudes. For real ω > ωc , A+ represents a
downstream-propagating wave and A− represents an upstream-propagating wave.
For real ω < ωc the waves are cut off. Defining the square root in Eq. (13.57)
to be a negative imaginary number, A+ now represents a downstream-decaying
disturbance and A− represents an upstream-decaying disturbance. For complex ω,
a combination of these behaviors is seen.
The separable solutions for an entropy disturbance are entropy waves of the
form
1
ρ = − A E eiωt+inθ+ik0 x E(r ) (13.58)
c̄2
n
u = A V eiωt+inθ+ik0 x V (r ) (13.59a)
ρ̄ c̄
k0 r
w = − A V eiωt+inθ+ik0 x V (r ) (13.59b)
ρ̄ c̄
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 391
B. Annular Duct
Many gas turbines, particularly aeroengines, have an annular geometry. Hence,
we will now consider the form of perturbations that can occur in the gap between
two rigid-walled concentric cylinders a r b. The acoustic waves are the same
as for a cylindrical duct except that now31
dYn dJn
Bn,m (r ) = (λn,m b)Jn (λn,m r ) − (λn,m b)Yn (λn,m r ) (13.61)
dr dr
and λn,m 0 is now the (m + 1)th solution of
dJn dYn dJn dYn
(λn,m a) (λn,m b) = (λn,m b) (λn,m a) (13.62)
dr dr dr dr
from the rigid-wall boundary conditions on r = a and r = b. [By using an approach
similar to that given by Watson29 to prove that Jn only has real zeros, it can be
shown that the solutions of Eq. (13.61) are again all real.] The entropy waves are
unchanged. For the vorticity waves, there is now no restriction on the function
V (r ), whereas for W (r ) we have W (a) = W (b) = 0.
a) 0
normalized growth rate, gN
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6
normalized frequency, fN
b) 0
normalized growth rate, gN
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6
normalized frequency, fN
Fig. 13.8 Frequencies and growth rates of resonant modes of a duct with choked
outlet: ×, choked inlet with entropy and vorticity waves included; ◦, choked inlet with
convected waves dissipated; +, open inlet; and – – –, cutoff frequency of the duct for
n = 1. a) n = 0. b) n = 1.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 393
where ωc = n c̄/R is the cutoff frequency of the duct, with the mode shapes also
approximated by Eqs. (13.63b) and (13.63c). In particular, for a given n the lowest
frequency mode is close to the cutoff frequency and has a pressure perturbation
that is roughly uniform axially. The frequencies [= Re(ω)/(2π )] and growth rates
[= − Im(ω)] of the modes for n = 0 and 1 are shown as circles in Fig. 13.8. The
pressure distribution for the second n = 1 mode (m = 1) at a sequence of times in
its oscillation period (T = 1/frequency) is shown in Fig. 13.9. Axially the mode
is a standing half-wave, whereas circumferentially it is a spinning whole wave.
All the modes have a negative growth rate, because the choked inlet and choked
outlet boundary conditions do not give a perfect reflection of acoustic waves and
are, therefore, sources of damping. If entropy and vorticity wave propagation is
included, many more modes are found, as denoted by crosses in Fig. 13.8. The
modes are roughly ū/(2l) Hz apart, that is, Re(ω) ≈ π ū/l. The least stable
modes, that is, those with the largest growth rates, are found to be close to the
modes when entropy and vorticity waves are ignored. For comparison with the
preceding examples, results for an open inlet/choked outlet are shown as pluses
in the figure. As we would expect the frequencies lie midway between the choked
inlet/choked outlet frequencies. In this case, neither entropy nor vorticity waves
are generated by the open inlet, so neither are present in the duct. Also, the growth
rates are less negative here because the open inlet gives no damping.
a) spin direction
spin direction 1
p′ (arbitrary scale)
1
1
p′ (arbitrary scale)
1
z /R
0
0
z /R
0
0
−1
−1 1
1 −1
1 −1 0
1 0.5
0 0.5 y /R −1 0 x/l
y / R −1 0 x/l
p′ (arbitrary scale)
1 1
z /R
z /R
0 0
0 0
−1 −1
1 −1 1 −1
1 1
0 0.5 0 0.5
y/R −1 0 x/ l y/ R −1 0 x/ l
Fig. 13.9 Time sequence of pressure distribution in thin annular duct for second
mode in Fig. 13.8b (choked inlet with convected waves dissipated). a) t = 0. b) t = T/4.
c) t = T/2. d) t = 3T/4.
394 A. P. DOWLING AND S. R. STOW
plenum combustor
A. Plenum
We investigate the form of linear disturbances in the geometry of a plenum
section, premixing ducts, and combustor. In this example, we will assume that the
frequency of oscillation is sufficiently low that only plane waves carry acoustic
energy, with all higher-order modes decaying exponentially with axial distance.
At the inlet, representing compressor exit, the flow is nearly choked, which leads
to inlet boundary conditions for the linear waves of frequency ω (as discussed in
Sec. II.B). The relative wave strengths at A–A (Fig. 13.11) are then completely
determined.
Equations (13.7c), (13.7d), and (13.9) describe how those waves develop along
the plenum, hence, determining the unsteady flow at entrance to the premix duct.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 395
plenum combustor
A’ x1 x2
fuel injection
A schematic diagram is shown in Fig. 13.12. The 2×2 matrix N depends on the
details of the geometry and the flow between x1 and x2 . For example, for the duct
with a uniform cross-sectional area, and a flow with negligible mean flow and the
premix duct
1 3 2
combustion zone
Fig. 13.12 Schematic diagram of premix duct and combustion zone (for definition of
transfer matrix).
396 A. P. DOWLING AND S. R. STOW
flame model in Eq. (13.24b), we found [see Eqs. (13.25a) and (13.26)]
1 0
N= (13.65)
0 (1 − βe−iωτ )−1
In the case in which the premix duct is short and has a small cross-sectional area and
no combustion occurs, the flow in the premix duct is effectively incompressible and
the pressure difference across it can be related to the rate of change of momentum
in the premix duct. For negligible mean flow, the relationship is A3 ( p̂ 1 − p̂ 2 ) =
(∂/∂t)(ALρu)3 = iωρ̄ A3 L 3 û 3 , where A represents the cross-sectional area, L is
the effective axial length, and subscript 3 denotes flow within the premix duct.
From conservation of mass, A1 û 1 = A2 û 2 = A3 û 3 . Hence, we have
1 iωρ̄ A2 L 3 /A3
N= (13.66)
0 A2 /A1
A2 ρ2 u 2 = A1 ρ1 u 1 (13.67a)
H2 = H1 (13.67b)
A2 p2 + A2 ρ2 u 22 = A2 p1 + A1 ρ1 u 21 (13.67c)
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 397
where subscripts 1 and 2 denote the flow parameters and areas before and after
the area change, respectively. Here the pressure on the abrupt expansion has been
taken to be p1 ; however, some pressure recovery could be included through the use
of a loss coefficient. To find the perturbations after the area increase, Eq. (13.67) is
linearized in the usual way to give a transfer matrix relating the downstream and
upstream flow.
An area decrease can be assumed to be isentropic, hence,
γ γ
p2 /ρ2 = p1 /ρ1 (13.68)
and conservation of mass and energy give Eqs. (13.67a) and (13.67b) as before.
For no mean flow, the jump conditions at any area change simplify to
The flame is also treated as compact, and so Eq. (13.10) applies across it.
However, this approach needs a flame model relating the instantaneous rate of heat
release to the oncoming flow. Flame models are discussed in Chapter 12, but here
we note that they can be determined either by analytical descriptions of the flame
dynamics9, 39 or through numerical36 or experimental investigations40–42 of the un-
steady combustion response to inlet flow disturbances. Measurements carried out
at low and high pressure have remarkably similar forms42 but different amplitudes,
supporting the idea that the flame-transfer function can be investigated by suitably
scaled experiments or through local computational fluid dynamics (CFD) solutions.
C. Combustor
Once the fluxes of mass, momentum, and energy are known in the combustor
just downstream of the zone of combustion, the strengths of the linear waves can
be calculated. Equations (13.7c), (13.7d), and (13.9) describe how those waves
develop along the combustor, thus determining the flow at exit. For a general
value of frequency ω, this will not satisfy the downstream boundary condition.
The resonant frequencies are the values of ω at which the downstream boundary
condition is satisfied.
Example 5
We now consider an example of a complete system consisting of a plenum,
premix system, and combustor, similar to that shown in Fig. 13.11 except that the
combustor has an open end. Details of the geometry are given in Table 13.1. A
simple flame model,
Q̂ m̂ i
= −k e−iωτ (13.70)
Q̄ m̄ i
is used at the start of the combustor, where m i is the air-mass flow at the fuel-
injection point (taken to be at the start of premixer). The circles in Fig. 13.13
denote the resonant modes of the geometry for k = 0. Several modes are seen, all
of which are stable as we would expect because there is no unsteady heat release.
The premix duct provides sufficient blockage that it acts approximately like a hard
398 A. P. DOWLING AND S. R. STOW
Description Value
150
100
50
0
Im ω (s 1 )
50
100
150
growth rate,
200
250
300
350
400
450
0 50 100 150 200 250 300 350 400 450 500 550 600
frequency (Hz)
D. Annular Combustors
We now consider an annular gas turbine for which the plenum and combustor
have a narrow annular gap cross section, as discussed earlier. Hence, we take the
perturbations to have the form of a circumferential mode. Wave propagation in
the plenum and combustor is given by Eqs. (13.7c), (13.7d), and (13.9) as before
(see also Sec. IV). When joining annular ducts of different inner and outer radii,
considering conservation laws in a thin sector of the transition leads to the same
flux relationships as for plane waves, with the addition that the angular-momentum
flux is unchanged.10 Hence, if the premix region also had an annular geometry,
the perturbations for a circumferential mode could be found in much the same
way as described earlier for plane waves. Typically, however, the premix region
consists of a large number of identical premix ducts that are evenly distributed
around the circumference. Hence, there is a loss of axisymmetry, and we might
expect that this would interact with the circumferential wave in the plenum to
produce circumferential waves of other orders, that is, modal coupling would
occur. In fact, any additional modes will be high order and decay rapidly with
axial distance (see Sec. VI). Thus, it is valid to consider a single circumferential
wave of a selected order in the plenum. The premix ducts will usually also be
annular; however, they will have a much smaller cross section than the plenum
and combustor and so, for frequencies of interest, the perturbations in them will
be one dimensional. The circumferential wave in the plenum produces identical
perturbations in the ducts, except that each perturbation is phase shifted. The
equations relating the perturbations in the plenum to those in the premix ducts
are similar to those for a simple area decrease, with adjustments due to the change
from a circumferential disturbance to a set of one-dimensional perturbations. The
propagation of these one-dimensional disturbances along the premix ducts can
400 A. P. DOWLING AND S. R. STOW
p′ (arbitrary scale)
p′ (arbitrary scale)
a) b)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
c) d)
p′ (arbitrary scale)
p′ (arbitrary scale)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
e) f)
p′ (arbitrary scale)
p′ (arbitrary scale)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
p′ (arbitrary scale)
p′ (arbitrary scale)
g) h)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
j)
p′ (arbitrary scale)
p′ (arbitrary scale)
i)
1 1
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
x (m) x (m)
Fig. 13.14 Mode shapes for simple combustor, k= 1. a) 30-Hz mode. b) 104-Hz mode.
c) 168-Hz mode. d) 203-Hz mode. e) 300-Hz mode. f) 312-Hz mode. g) 396-Hz mode.
h) 415-Hz mode. i) 495-Hz mode. j) 514-Hz mode.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 401
be found as before. At the inlet to the combustor, the ring of phase-shifted one-
dimensional disturbances creates a circumferential wave of an order identical with
that in the plenum. The resonant modes for circumferential waves of this selected
order can then be found by investigating the propagation of this circumferential
mode through the combustor and determining the resonant frequencies at which
the downstream boundary condition is satisfied.
VI. Modal Coupling
In uniform cylindrical and annular ducts, the solutions in Eq. (13.56) for different
values of n and m are independent and can be considered separately. However,
nonuniformities can lead to a coupling of these modes. For instance, if the duct
has an area change, but remains axisymmetric, the circumferential modes, that is,
different values of n, are still independent but the radial modes, that is, different
values of m, become coupled. Consider, for example, a circular duct that has an
abrupt area increase at x = 0. We denote conditions in x < 0 by superscript (1)
and in x > 0 by superscript (2). The duct is then r b(1) for x < 0 and r b(2)
for x > 0, with b(2) > b(1) . For no mean flow, only acoustic waves are present, and
so from Eq. (13.56) for a given n, we may write for x < 0
∞
+ −
p = eiωt+inθ A+(1)
n,m e
ikn,m x
+ A−(1)
n,m e
ikn,m x
Bn,m (r ) (13.71a)
m=1
∞
+ −
p = eiωt+inθ A+(2)
n,m e
ikn,m x
+ A−(2)
n,m e
ikn,m x
Bn,m (r ) (13.71b)
m=1
with similar expressions for the other flow variables. Miles43 and Alfredson44 con-
sidered this problem for plane waves; however, the extension to n = 0 is straight-
forward (as is the extension to annular ducts). At x = 0, we must have continuity of
p and u for 0 r b(1) (continuity of ρ , v , and w follow from continuity of p ),
and on the rigid wall b(1) r b(2) we require u = 0. This continuity leads to a
linear system of equations relating A±(1) ±(2)
n,m and An,m . The amplitudes for one value
m are found to depend on those for all other values of m, meaning that the radial
modes are coupled. In Eq. (13.71) we included all the radial modes; however, in
practice, for m that is sufficiently large the waves will be highly cut off and so
can be ignored. Hence, we can approximate using a finite number of radial modes,
for example, 0 < m < M. Some examples of results for n = 0 modes in a duct in
which the area doubles are shown in Fig. 13.15.
The radial variations of the magnitudes of the pressure and axial velocity on
either side of the area change are shown; the solid and dashed lines denote the
values in the larger and smaller area regions, respectively. These results are for
M = 5; as more radial modes are included, the matching becomes better and the
solutions more accurate.
A similar approach was used by Akamatsu and Dowling45 to consider three-
dimensional combustion instabilities in a cylindrical combustor with a ring of
402 A. P. DOWLING AND S. R. STOW
a) 1.25 b) 1.25
1 1
p′ (arbitrary scale)
u′ (arbitrary scale)
0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
r / b(2 ) r / b(2 )
premix ducts. Oscillations in the premix ducts were assumed to be one dimen-
sional, and these were treated as point sources when joining to the combustion
chamber. The loss of radial symmetry here led to a coupling of the radial modes
in the combustor. Perhaps surprisingly, because the premixers were identical and
evenly distributed circumferentially, the circumferential modes remained uncou-
pled. Similarly, Evesque and Polifke46 found that circumferential modes became
coupled only when their premix ducts were nonidentical. In fact, it can be shown
that a ring of identical premix ducts does not introduce coupling of circumferential
modes provided that N is less than half the number of ducts. In other words, any
coupling occurs in high-order modes that decay rapidly with axial distance and
are not of practical interest.
Coupling of circumferential modes in a narrow annular gap has been considered
by Stow and Dowling.47 The presence of Helmholtz resonators in the geometry
destroys the axisymmetry causing modal coupling. We now describe their method
of solution because the approach, in general, should be applicable to finding linear
resonances in problems
with modal coupling. We write p (t, x, θ) = p̂(x, θ )eiωt
with p̂(x, θ) = ∞ n=−∞ p̂ n (x)e , and similarly for the other variables. For |n|
inθ
large (for example, |n| > N ), the mode will be highly cut off; hence, in a way sim-
ilar to the radial
N modes earlier, we approximate circumferential modes by taking
p̂(x, θ) = n=−N p̂ n (x)einθ . At the inlet of the geometry, there are boundary con-
ditions that apply to each mode independently. These define the perturbations for
each circumferential mode n except for an unknown parameter λn . For instance, if
it is an open end, p̂ = 0 for all θ , implying that p̂ n = 0 for all n, and so we may set
A+ −
n = −An = λn (with no entropy or vorticity waves). Here, λ = [λ−N , . . . , λ N ]
T
describes the relative amplitude and phase of the modes at the inlet and must be
found as part of the solution. Similarly, at the outlet, there is a boundary condi-
tion that applies to each mode independently. We define µn to be the error in this
boundary condition for circumferential mode n; for example, for an open end we
may take µn = p̂ n . Given ω and λ , all the circumferential components at the inlet
are known. We can step through the geometry calculating all the circumferential
modes at each sec. before continuing to the next. For the solution thus found, each
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 403
mode will have an error at the outlet µn . We must find ω and λ to satisfy µn = 0,
thus, giving a resonance of the geometry. For a given ω, we define the matrix M
to be such that Mn,m is the value of µn for the solution with λi = δi,m . For a gen-
eral λ , µn = Mn,m λm because the perturbations are linear. Hence, for the correct
values of ω and λ , Mλλ = 0. For a solution to exist, λ = 0, and so this implies that
det M = 0. Thus the procedure to find a complex resonant frequency ω is to first
guess the value of ω and calculate the matrix M and then iterate the value of ω to
achieve det M = 0. For this value of ω, a λ will exist giving Mλ λ = 0. Finally, this
correct λ is calculated by using an inverse iteration method (Mλ λnew = λ old ). The
mode shape for the resonance can then be calculated using this λ . As before, the
resonant frequency and growth rate are given by ω.
170
sound pressure level (dB)
160
150
140
130
120
110
0 100 200 300 400 500 600 700 800 900 1000
frequency (Hz)
will cause large velocity oscillations into and out of the resonator. These velocity
oscillations dissipate energy leading to a damping of the acoustic perturbations
in the duct. This source of damping is a nonlinear effect (see Ref. 51), relying
on the velocity oscillations in the neck to have a large enough amplitude so that
significant kinetic energy is dissipated in the unsteady jets that form. Specifically,
the acoustic energy of the velocity oscillations at the neck is converted to vortical
energy and ultimately dissipated as heat. If the pressure perturbations in the duct
are low amplitude, this mechanism gives negligible damping. Hence, it can reduce
the amplitude of an existing instability but cannot stabilize the mode.
In gas-turbine applications, there will be a mean flow through the combustor
and hence across the neck of the Helmholtz resonator, and the requirement to cool
the resonator may lead to an additional flow through the neck; these flows lead to
additional sources of damping. In this situation, the acoustic waves modulate the
vortex shedding at the neck and lead to a linear source of damping in the sense that
the proportion of acoustic energy absorbed is independent of the sound pressure.
This mechanism therefore has the potential to stabilize a mode. However, there
is a danger with this configuration that, in some frequency ranges, generation of
sound instead of absorption can occur because of vortices being shed from the
upstream lip of the neck and impinging on the downstream lip. This problem
can be overcome by using a downstream lip that is rounded, not sharp-edged.
Alternatively, introducing a sufficiently strong cooling flow through the neck into
the combustor can remove the problem, because the vortices are then driven away
from the downstream lip.
To include a Helmholtz resonator in an acoustic calculation of the type described
earlier, one can consider conservation of unsteady mass, momentum, and energy
between the point in the duct just upstream of the resonator and the point just down-
stream. However, one needs to account for the mass flow perturbation m̂ entering
the duct through the neck of the resonator. In no mean flow, as stated previously,
nonlinear effects are important and so m̂ is not simply linearly proportional to
the amplitude of the oscillations. Hence, a nonlinear calculation is now required,
such as the describing-function approach considered in the next section. With a
mean crossflow and/or neckflow, the system remains linear and the calculation
techniques described previously can be applied directly. In this case, writing p̂ 1
and p̂ 2 for the pressure perturbations in the combustor and inside the resonator,
respectively, we may define the Rayleigh conductivity κ of the neck by
iωm̂
κ= (13.72)
p̂ 2 − p̂ 1
The rate of decrease of mass inside the resonator must equal m̂ and so, assuming
isentropic conditions there, we have m̂ = −iωV p̂ 2 /c̄2 , where V and c̄ are the
volume of the resonator and the speed of sound inside, respectively. If the mean
flow through the neck is much larger than the crossflow, the Rayleigh conductivity
for a circle aperture found by Howe et al.52 can be used, with a correction to account
for the length of the neck. Conversely, if the only mean flow is across the neck,
the result by Howe53 is applicable. The case in which the cross- and throughflows
are comparable has been considered by Dupère and Dowling.54 Details on the
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 405
modeling of Helmholtz resonators with and without mean flows and a discussion of
practical issues for their use in real combustors are given by Dupère and Dowling.55
The placement of Helmholtz resonators can be an important consideration. For
maximum effect, they should be placed where the amplitude of pressure oscillation
is high, whereas at a pressure node they would have no effect. In relation to this
effect, circumferential modes in annular combustors using only a single resonator
give no damping. A circumferential standing wave is produced (the clockwise and
anticlockwise spinning circumferential modes becoming coupled) which aligns
itself to have a pressure node at the resonator neck. Hence, at least two resonators
are required to damp such an oscillation. The azimuthal placement of resonators
to achieve the best damping of circumferential waves has been investigated by
Stow and Dowling.47 A drawback of using Helmholtz resonators is that they give
good damping only over a relatively small-frequency band. If there are several
modes of instability, several resonators may be required. However, in a situation
in which at different operating conditions a single frequency dominates, but in
which the frequency varies (either continuously or suddenly) as the operating con-
ditions are changed, an interesting alternative is the use of a Helmholtz resonator
that can retune itself to damp the current instability. This form of actively tuned
passive damping or semiactive control has been investigated by Wang.56 Figure
13.16 shows experimental results for an atmospheric rig similar to that shown in
Fig. 13.11, with an adjustable-volume Helmholtz resonator attached to the com-
bustion chamber. A feedback algorithm was used to tune the resonator, leading
to a more than 15-dB reduction in peak amplitude. [We thank Dr. Chuan-Han
Wang (Cambridge University Engineering Department) for permission to show
this figure.]
Liners with bias flow have the potential to damp oscillations over a much
broader range of frequencies than Helmholtz resonators have. This type of acous-
tic absorber was investigated for plane waves by Eldredge and Dowling,57 who
found that more than 80% absorption can be acheived. Eldredge58 extended their
analysis to higher-order modes. The configuration is as follows. A section of the
duct in which we wish to damp oscillations (for example, the combustion chamber)
is replaced by a liner consisting of an array of holes, through which a mean flow
passes into the duct. This flow can form part of the cooling of the duct. On the
other side of the liner we could simply have a duct, a large chamber, or one or more
additional liners supplying the flow. This setup changes the performance of the
liner but not the underlying principles. The mechanism of absorption is very simi-
lar to the case of a Helmholtz resonator with a mean flow through the neck, namely
the conversion of acoustic energy to vortical energy in the shed vortices, which is
then dispersed. However, the liner does not rely on matching a resonant frequency
and can absorb over a large frequency range. As with Helmholtz resonators, the
liner is most effective if located at a region of large pressure oscillations. To include
such a liner in the linear models described earlier, the liner must be discretized
axially. For instance, one may represent the liner as a series of compact regions
containing the holes separated by straight ducts. The perturbations at these hole
regions can be calculated by using the Rayleigh conductivity of the holes in much
the same way as for a Helmholtz resonator with neck flow; the main difference is
the treatment of the perturbations on the other side of the liner. If there is a large
chamber on the other side, we may assume that the pressure oscillation is zero;
406 A. P. DOWLING AND S. R. STOW
150
Im ω (s 1 )
100
50
0
growth rate,
50
100
150
0 50 100 150 200 250 300 350 400 450 500 550 600
frequency (Hz)
Fig. 13.17 Resonant modes of simple combustor: ◦, with Helmholtz resonator; and
×, without Helmholtz resonator.
if a duct or secondary liner is present, this must be modeled as part of the linear
calculations.
As an example of the use of acoustic dampers in linear acoustic models, we now
consider adding a Helmholtz resonator to example 5 (with k = 1). The resonator is
placed halfway along the combustor and is assumed to have a mean neck outflow
of 10 ms−1 . (The crossflow in the combustor is negligible compared with this.) The
neck is taken to have a radius of 7 mm and length of 30 mm, and the temperature
in the volume is set to be 1000 K. We seek to damp the most unstable mode, which
is at 168 Hz; hence, we set this to be the resonant frequency of the Helmholtz
resonant by taking its volume to be 1.24 × 10−3 m3 . The resulting resonant modes
are plotted as circles in Fig. 13.17, with the modes without the resonator shown
as crosses (see also Fig. 13.13). We find that the growth rate of the targeted mode
is reduced, which indicates damping, and there is a small shift in the frequency to
166 Hz. Also an additional mode has appeared at 159 Hz; this mode is associated
with the resonator and is highly damped. The resonator has only a minor effect
on the frequency of the other modes, however, many have a significant increase
or decrease in their growth rate. This change is perhaps surprising, in particular,
where the growth rate is increased because a Helmholtz resonator with a mean
flow through the neck never generates acoustic energy. The effect is caused by the
fact that the resonator, independent of any damping effects, alters the acoustics of
the combustor because of the inertia of the mass of air in the neck. Although this
inertia has only a small effect on the frequency, it is enough to change the difference
between acoustic energy gained from and lost at the combustor boundaries, which
is much more sensitive. This has a direct effect on the growth rate.
mode usually provides a good approximation to that of the resulting limit cycle.
Damage is often the result of a oscillation frequency being close to the structural
resonant frequency of a component of the gas turbine, so knowledge of potential
frequencies can be very useful.
However, to obtain predictions of the amplitudes of oscillation, the nonlinear
effects that limit the size of the perturbations must be modeled. The limit cycle usu-
ally involves oscillations in which the pressure perturbation is small in comparison
with the mean pressure, and the fluctuating velocity is much less than the speed of
sound. These conditions ensure that the acoustic waves are still linear. The main
nonlinearity is usually in the combustion response to velocity and equivalence ratio
fluctuations which can be of the order of their mean. The time-domain Galerkin
method can be extended to include nonlinearity and hence give limit-cycle solu-
tions.59 The transfer matrix and conservation approaches can be converted to the
time domain (at least for plane waves), and so, similarly, once nonlinear effects
are included, these can be used to predict limit cycles. However, a faster and sim-
pler method is to remain in the frequency domain and use a describing-function
approach as follows (see also Dowling16 ). The main effect determining the limit-
cycle amplitude is likely to be a saturation of the heat-release oscillation from the
flame. Consider a flame being forced by a time-varying input (such as the air-mass
flow at the fuel-injection point m i ) at a single frequency. The heat release from
the flame will be periodic with the same frequency, but at a high forcing am-
plitude the response may contain multiple frequencies, because nonlinearity can
generate the harmonics of the forcing frequency. However, we expect the flame
will respond less to high-frequency disturbances, suggesting that these harmonics
are not important in the feedback loop between the heat release and the acoustics
of the geometry. Hence, the flame can be characterized by a nonlinear flame-
transfer function relating the heat-release component at the forcing frequency to
the flow perturbation as a function of both frequency and amplitude of forcing. For
example,
Q̂ m̂ i
= T (ω, A) (13.73)
Q̄ m̄ i
where T is the nonlinear transfer function and A = |m̂ i |/m̄ i is the forcing ampli-
tude. Typically, increasing A will decrease the magnitude of T because of satura-
tion effects, whereas the effect on the phase of T is often less significant. We have
already seen in example 5 that decreasing the gain of the flame-transfer function
has a stabilizing effect. For a mode that is linearly unstable, the amplitude will
initially increase with time, thus reducing the gain and hence reducing the growth
rate. Eventually we reach a point where the growth rate is zero. This mode is a
stable limit cycle; at lower amplitudes the oscillations are still growing, whereas at
higher amplitudes they will decay. One may assume that elsewhere in the geometry
nonlinear effects are less important and that the linear models are still applicable
there. Instead of solving for complex ω in the linear problem, the solution now
amounts to finding real ω and A such that the exit boundary condition is satisfied.
This solution gives the limit-cycle frequency, amplitude, and dimensional mode
shape.
408 A. P. DOWLING AND S. R. STOW
m i (t − τ )
Q L (t) = −k Q̄ (13.74)
m̄ i
The subscript L denotes that this a linearized result. One could consider nonlinear
effects on the equivalence ratio, convection time, and flame response (see Stow
and Dowling60 ); however, here we will assume a simple saturation in Q(t):
Q L (t) for |Q (t)| α Q̄
Q (t) = (13.75)
α Q̄ sign Q L (t) for |Q (t)| > α Q̄
This nonlinear flame model is very similar to the one used in a time-domain
approach by Bellucci et al.61 To obtain the nonlinear transfer function we need
to convert to the frequency domain. Setting m i (t) = A cos(ωt)m̄ i , the transfer
function is found by calculating the component of Q (t) at frequency ω,
ω 2π/ω
T (ω, A) = Q (t)e−iωt dt (13.76)
π A Q̄ 0
In this model, for A α/|k| no saturation occurs and so the transfer function
is the same as the linearized version, that is, T (ω, A) = TL (ω) = −ke−iωτ . For
A > α/|k|, it can be shown that
φ/ω (π −φ)/ω
2ωe−iωτ
T (ω, A) = − αe−iωt dt + |k|A cos(ωt)e−iωt dt
πA −φ/ω φ/ω
(13.77)
Figure 13.18 shows the variation of T /TL with β. We see that for A > α/|k|
the flame response decays monotonically, tending to zero for large amplitudes.
Note that in this model the phase of the heat release is unaffected by the forcing
amplitude.
As an illustration, we now consider a limit-cycle calculation for the geometry
in example 5. In the linear calculations, we found that there were several unstable
modes for k = 1 in the flame model. The describing-function approach is not
strictly applicable unless there is a single dominant frequency in the limit cycle,
so it is questionable to use the technique for this case. However, if instead we take
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 409
12
08
T TL
06
04
02
0
0 1 2 3 4 5 6 7 8 9 10
β
k = 0.4 there is only one linearly unstable mode in the frequency range considered.
This is at 290 Hz and has a growth rate of 30 s−1 . Using the nonlinear flame model
in Eq. (13.78) with α set to be 0.1, we find that the corresponding limit cycle has
a frequency of 288 Hz and amplitude A = 0.65. The dimensional mode shape is
shown in Fig. 13.19.
IX. Conclusion
A series of model problems with very simple geometries has been considered
to demonstrate an acoustic analysis of the various components that make up a
gas-turbine combustion system.
The form of the coupling between the heat release and the unsteady flow has
been demonstrated to have a crucial effect on the frequency of oscillation. A one-
term Galerkin series expansion is not adequate to determine this frequency shift
for the sorts of unsteady combustion response typical of gas-turbine combustors.
The effect of the mean temperature ratio across the combustion zone can be signif-
icant. Mean flow effects are not significant for Mach numbers less than about 0.2;
4000
3000
p′ (Pa)
2000
1000
0
0 0 25 05 0 75 1 1 25 1 5 1 75 2 2 25 25 2 75
x (m)
however, a mean flow does introduce the possibility of a new mode of oscillation
at a much lower frequency where the period of oscillation is set primarily by the
time taken for the convection of entropy waves, or hot spots. Higher-order modes
in the annular and cylindrical ducts bring in the possibility that the modes are cut
off. We have described how a typical LPP combustion system can be built up and
analyzed through connection of a series of cylindrical and/or annular ducts. In
many geometries the premix ducts provide sufficient blockage that these modes
of oscillation are close to separate modes of the plenum and combustor with a
hard or approximately constant velocity boundary condition at the premixer. We
have also noted that modal coupling may occur when the geometry is no longer
axisymmetric, and we have seen how to include acoustic absorbers in the models.
The linear models discussed in this chapter relate to small oscillations only, and
hence they give predictions of the stability of modes but not the amplitude of the
resulting limit cycles. However, these models can still provide important infor-
mation to gas-turbine designers and operators. Furthermore, we have seen that by
using describing-function analysis these models can be extended to give amplitude
predictions. The great advantage of the approaches presented here is their speed.
Many geometry configurations and operating conditions can be investigated in a
relatively short time.
DS
ρT =q (13.A1)
Dt
ρT cv D p Dρ
= −c p T +q (13.A2)
p Dt Dt
After substitution for Dρ/Dt from the equation of mass conservation, we obtain
cv D p
∇ ·u+q
= c p Tρ∇ (13.A3)
Rgas Dt
which is equivalent to
Dp
∇ · u + (γ − 1)q
= c2 ρ∇ (13.A4)
Dt
since c2 = γ Rgas T and Rgas /cv = γ − 1. Equation (13.41) is the linearized form
of Eq. (13.A4).
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 411
References
1
Dowling, A. P., and Stow, S. R., “Acoustic Analysis of Gas Turbine Combustors,” Journal
of Propulsion and Power, Vol. 19, No. 5, 2003, pp. 751–764.
2
Rayleigh, L., The Theory of Sound, 2nd ed., Vol. 2, Macmillan, London, 1896, pp. 226–
227.
3
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engine Combustion Instability,
Vol. 169, Progress in Astronautics and Aeronautics, AIAA, New York, 1995.
4
De Luca, D., Price, E. W., and Summerfield, M. (eds.), Nonsteady Burning and Com-
bustion Stability of Solid Propellants, Vol. 143, Progress in Astronautics and Aeronautics,
AIAA, New York, 1992.
5
Keller, J. J., “Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,”
AIAA Journal, Vol. 33, No. 12, 1995, pp. 2280–2287.
6
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Combustion Dynamic
Modeling for Gas Turbine Engines,” AIAA Paper 98-3380, July 1998.
7
Hsiao, G. C., Pandalai, R. P., Hura, H. S., and Mongia, H. C., “Investigation of Combus-
tion Dynamics in Dry-Low-Emission (DLE) Gas Turbine Engines,” AIAA Paper 98-3381,
July 1998.
8
Lovett, J. A., Chu, W.-W., and Shah, S. N., “Modeling of Combustion Chamber Acoustics
and Control of Combustion Instabilities in Gas Turbines,” 6th International Congress on
Sound and Vibration, July 1999.
9
Dowling, A. P., and Hubbard, S., “Instability in Lean Premixed Combustors,” Journal
of Power and Energy, Vol. 214, No. 4, 2000, pp. 317–332.
10
Stow, S. R., and Dowling, A. P., “Thermoacoustic Oscillations in an Annular Combus-
tor,” American Society of Mechanical Engineers, Paper 2001-GT-0037, June 2001.
11
Chu, B.-T., and Kovasznay, L. S. G., “Non-linear Interactions in a Viscous Heat-
Conducting Compressible Gas,” Journal of Fluid Mechanics, Vol. 3, Feb. 1958, pp. 494–
514.
12
Marble, F. E., and Candel, S. M., “Acoustic Disturbance from Gas Non-uniformities
Convected Through a Nozzle,” Journal of Sound and Vibration, Vol. 55, No. 2, 1977,
pp. 225–243.
13
Stow, S. R., Dowling, A. P., and Hynes, T. P., “Reflection of Circumferential Modes in
a Choked Nozzle,” Journal of Fluid Mechanics, Vol. 467, Sept. 2002, pp. 215–239.
14
Yang, V., and Culick, F. E. C., “Analysis of Unsteady Inviscid Diffuser Flow with a
Shock Wave,” Journal of Propulsion and Power, Vol. 1, No. 3, 1985, pp. 222–228.
15
Culick, F. E. C., and Rogers, T., “The Response of Normal Shocks in Diffusers,” AIAA
Journal, Vol. 21, No. 10, 1983, pp. 1382–1390.
16
Dowling, A. P., “A Kinematic Model of a Ducted Flame,” Journal of Fluid Mechanics,
Vol. 394, Sept. 1999, pp. 51–72.
17
Culick, F. E. C., and Yang, V., “Overview of Combustion Instabilities in Liquid-
Propellant Rocket Engines,” Liquid Rocket Engine Combustion Instability, edited by
V. Yang, and W. E. Anderson, Vol. 169, Progress in Astronautics and Aeronautics, AIAA,
New York, 1995, pp. 3–37.
18
Dowling, A. P., “The Calculation of Thermoacoustic Oscillations,” Journal of Sound
and Vibration, Vol. 180, No. 4, 1995, pp. 557–581.
19
Annaswamy, A. M., Fleifil, M., Hathout, J. P., and Ghoniem, A. F., “Impact of Linear
Coupling on Design of Active Controllers for Thermoacoustic Instability,” Combustion
Science and Technology, Vol. 128, No. 1–6, 1997, pp. 131–160.
412 A. P. DOWLING AND S. R. STOW
20
Cummings, A., “Ducts with Axial Temperature Gradients: An Approximate Solution
for Sound Transmission and Generation,” Journal of Sound and Vibration, Vol. 51, 1977,
pp. 55–67.
21
Jones, H., “The Mechanics of Vibrating Flames in Tubes,” Proceedings of the Royal
Society of London Series A, Vol. 353, No. 1675, 1977, pp. 459–473.
22
Sujith, R. I., Waldherr, G. A., and Zinn, B., “An Exact Solution for One-dimensional
Acoustic Fields in Ducts with an Axial Temperature Gradient,” Journal of Sound and
Vibration, Vol. 184, No. 3, 1995, pp. 389–402.
23
Kumar, B, M., and Sujith, R. I., “Exact Solution for One-dimensional Acoustic Fields
in Ducts with Polynomial Mean Temperature Profiles,” Journal of Vibration and Acoustics,
Vol. 120, No. 4, 1998, pp. 965–969.
24
Karthik, B., Kumar, B. M., and Sujith, R. I., “Exact Solutions to One-dimensional
Acoustic Fields with Temperature Gradient and Mean Flow,” Journal of the Acoustical
Society of America, Vol. 108, No. 1, 2000, pp. 38–43.
25
Sujith, R. I., “Exact Solutions for Modeling Sound Propagation Through a Combustion
Zone,” Journal of the Acoustical Society of America, Vol. 110, No. 4, 2001, pp. 1839–
1844.
26
Eisenberg, N. A., and Kao, T. W., “Propagation of Sound Through a Variable-Area Duct
with Steady Compressible Flow,” Journal of the Acoustical Society of America, Vol. 49,
No. 1, 1971, pp. 169–175.
27
Subrahmanyam, P. B., Sujith, R. I., and Lieuwen, T., “A Family of Exact Transient
Solutions for Acoustic Wave Propagation in Inhomogeneous, Non-uniform Area Ducts,”
Journal of Sound and Vibration, Vol. 240, No. 4, 2001, pp. 705–715.
28
Zhu, M., Dowling, A. P., and Bray, K. N. C., “Self-excited Oscillations in Combustors
with Spray Atomisers,” Journal of Engineering for Gas Turbines and Power, Vol. 123,
No. 4, 2001, pp. 779–786.
29
Watson, G. N., A Treatise on the Theory of Bessel Functions, 2nd ed., Cambridge
University Press, Cambridge, England, UK, 1944, p. 482.
30
Eversman, W., “Theoretical Models for Duct Acoustic Propagation and Radiation,”
Aeroacoustics of Flight Vehicles: Theory and Practice, edited by H. H. Hubbard, Vol. 2,
Acoustical Society of America, New York, 1994, pp. 101–163.
31
Tyler, J. M., and Sofrin, T. G., “Axial Compressor Noise Studies,” SAE Transactions,
Vol. 70, No. 31, 1962, pp. 309–332.
32
Paschereit, C. O., Schuermans, B., Polifke, W., and Mattson, O., “Measurement of
Transfer Matrices and Source Terms of Premixed Flames,” Journal of Engineering for Gas
Turbines and Power, Vol. 124, No. 2, 2002, pp. 239–247.
33
Ohtsuka, M., Yoshida, S., Inage, S., and Kobayashi, N., “Combustion Oscillation Anal-
ysis of Premixed Flames at Elevated Pressures,” American Society of Mechanical Engineers,
Paper 98-GT-581, June 1998.
34
Hobson, D. E., Fackrell, J. E., and Hewitt, G., “Combustion Instabilities in Industrial
Gas Turbines—Measurements on Operating Plant and Thermoacoustic Modeling,” Journal
of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420–428.
35
Dowling, A. P., “Thermoacoustic Instability,” 6th International Congress on Sound and
Vibration, July 1999, pp. 3277–3292, http://icsv6.dat.dtu.dk.
36
Krüger, U., Hürens, J., Hoffmann, S., Krebs, W., and Bohn, D., “Prediction of Ther-
moacoustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU,” American Society of Mechanical Engineers, Paper 99-GT-111,
June 1999.
ACOUSTIC ANALYSIS OF GAS-TURBINE COMBUSTORS 413
37
Krüger, U., Hürens, J., Hoffmann, S., Krebs, W., Flohr, P., and Bohn, D., “Prediction and
Measurement of Thermoacoustic Improvements in Gas Turbines with Annular Combustion
Systems,” American Society of Mechanical Engineers, Paper 2000-GT-0095, May 2000.
38
Lovett, J. A., and Uznanski, K. T., “Prediction of Combustion Dynamics in a Staged
Premixed Combustor,” American Society of Mechanical Engineers, Paper GT-2002-30646,
June 2002.
39
Ni, A., Polifke, W., and Joos, F., “Ignition Delay Time Modulation as a Contribution to
Thermo-acoustic Instability in Sequential Combustion,” American Society of Mechanical
Engineers, Paper 2000-GT-0103, May 2000.
40
Lawn, C. J., “Interaction of the Acoustic Properties of a Combustion Chamber with
Those of Premixture Supply,” Journal of Sound and Vibration, Vol. 224, No. 5, 1999,
pp. 785–808.
41
Krebs, W., Hoffmann, S., Prade, B., Lohrman, M., and Büchner, H., “Thermoacoustic
Flame Response of Swirl Flames,” American Society of Mechanical Engineers, Paper GT-
2002-30065, June 2002.
42
Cheung, W. S., Sims, G. J. M., Copplestone, R. W., Tilston, J. R., Wilson, C. W.,
Stow, S. R., and Dowling, A. P., “Measurement and Analysis of Flame Transfer Function
in a Sector Combustor under High Pressure Conditions,” American Society of Mechanical
Engineers, Paper GT-2003-38219, June 2003.
43
Miles, J., “The Reflection of Sound Due to a Change in Cross Section of a Circular
Tube,” Journal of the Acoustical Society of America, Vol. 16, No. 1, 1944, pp. 14–19.
44
Alfredson, R. J., “The Propagation of Sound in a Circular Duct of Continuously Varying
Cross-sectional Area,” Journal of Sound and Vibration, Vol. 23, No. 4, 1972, pp. 433–442.
45
Akamatsu, S., and Dowling, A. P., “Three Dimensional Thermoacoustic Oscillation in
an Premix Combustor,” American Society of Mechanical Engineers, Paper GT-2001-0034,
June 2001.
46
Evesque, S., and Polifke, W., “Low-Order Acoustic Modelling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” American Society of Mechanical Engineers,
Paper GT-2002-30064, June 2002.
47
Stow, S. R., and Dowling, A. P., “Modelling of Circumferential Modal Coupling Due
to Helmholtz Resonators,” American Society of Mechanical Engineers, Paper GT-2003-
38168, June 2003.
48
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combustion
System Damping Augmentation with Helmholtz Resonators,” Journal of Engineering for
Gas Turbines and Power, Vol. 122, No. 2, April 2000, pp. 269–274.
49
Bellucci, V., Flohr, P., Paschereit, C. O., and Magni, F., “On the Use of Helmholtz
Resonators for Damping Acoustic Pulsations in Industrial Gas Turbines,” Journal of Engi-
neering for Gas Turbines and Power, Vol. 126, No. 2, April 2004, pp. 271–275.
50
Dowling, A. P., and Ffowcs Williams, J. E., Sound and Sources of Sound, Ellis Horwood,
London, 1983.
51
Cummings, A., “Acoustic Nonlinearities and Power Losses at Orifices,” AIAA Journal,
Vol. 22, No. 6, 1984, pp. 786–792.
52
Howe, M. S., Scott, M. I., and Sipcic, S. R., “The Influence of Tangential Mean Flow
on the Rayleigh Conductivity of an Aperture,” Proceedings of the Royal Society of London
Series A, Vol. 452, 1996, pp. 2303–2317.
53
Howe, M. S., “On the Theory of Unsteady High Reynolds Number Flow Through a
Circular Aperture,” Proceedings of the Royal Society of London Series A, Vol. 366, 1979,
pp. 205–223.
414 A. P. DOWLING AND S. R. STOW
54
Dupère, I. D. J., and Dowling, A. P., “The Absorption of Sound by Helmholtz Resonators
with and Without Flow,” AIAA Paper 2002-2590, June 2002.
55
Dupère, I. D. J., and Dowling, A. P., “The Use of Helmholtz Resonators in a Practical
Combustor,” Journal of Engineering for Gas Turbines and Power, Vol. 127, No. 2, April
2005, pp. 268–275.
56
Wang, C.-H., “Actively-Tuned Passive Control of Combustion Instabilities,” Ph.D.
Thesis, Cambridge Univ. Cambridge, England, U.K., 2004.
57
Eldredge, J. D., and Dowling, A. P., “The Absorption of Axial Acoustic Waves by
a Perforated Liner with Bias Flow,” Journal of Fluid Mechanics, Vol. 485, June 2003,
pp. 307–335.
58
Eldredge, J. D., “On the Interaction of Higher Duct Modes with a Perforated Liner
System with Bias Flow,” Journal of Fluid Mechanics, Vol. 510, July 2004, pp. 303–331.
59
Culick, F. E. C., “Some Recent Results for Nonlinear Acoustics in Combustion Cham-
bers,” AIAA Journal, Vol. 32, No. 11, Jan. 1994, pp. 146–169.
60
Stow, S. R., and Dowling, A. P., “Low-Order Modelling of Thermoacoustic Limit
Cycles,” American Society of Mechanical Engineers, Paper GT-2004-54245, June 2004.
61
Bellucci, V., Schuermans, B., Nowak, D., Flohr, P., and Paschereit, C. O., “Thermoa-
coustic Modeling of a Gas Turbine Combustor Equipped with Acoustic Dampers,” American
Society of Mechanical Engineers, Paper GT-2004-53977, June 2004.
Chapter 14
Nomenclature
a = speed of sound in mixture
C = constant-volume specific heat for liquid phase
Cf = coefficient of source term model, Equation (14.33)
Ch = coefficient of source term model, Equation (14.33)
CJ = coefficient of Bessel Function of the first kind, Equation (14.20) or
(14.25)
C p = constant-pressure specific heat for two-phase mixture
CY = coefficient of Bessel Function of the second kind, Equation (14.21)
f = Equation (14.12), or frequency
h = Source term in wave equation
I = acoustic intensity, Equations (14.62)
Jm = mth-order Bessel Function of the first kind
kmn = eigenvalue of mnth mode
L = chamber length
M = number of modes in azimuthal direction, or Mach number
N = number of modes in radial direction
N T = total number of cells
p = pressure
q = heat flux vector
Q̇ = rate of heat release
R = Gas constant for two-phase mixture
Rc = radius of circular chamber
Copyright c 2005 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Postdoctoral Research Associate. Member AIAA.
† Distinguished Professor. Fellow AIAA.
‡ Professor. Member AIAA. Present address: Beijing University of Aeronautics and Astronautics,
Beijing, China.
415
416 D. YOU AND V. YANG
Greek Symbols
α = axial wave number
βa = reflection coefficient, Eq. (14.66)
βs = reflection coefficient, Eq. (14.68)
ρ = density of two-phase mixture
δ = Kronecker delta
ψ = normal mode function
θ = circumferential coordinate
η = series coefficient of Fourier-type expansion, Eq. (14.35)
γ = specific heat ratio for mixture
= frequency
σ̇ = rate of entropy generation
Subscripts
i = imaginary part
j = cell index
l = axial direction mode
m = circumferential direction mode
n = radial direction mode
r = real part
T = transverse plane
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 417
I. Introduction
acoustic damper
diffuser shroud
turbine
flow vane
fuel unsteady oscillation
compressor nozzle combustion s
diffuser
swirler
primary air
cooling air
for the effects of geometric and flow variations, the chamber is discretized into a
number of circular or annular cells along the axial direction. The cross-sectional
area and axial distributions of mean flow properties are assumed to be uniform
within each cell. A combined modal-expansion and spatial averaging technique
is then applied to solve for unsteady motions in one cell. The next step involves
matching the oscillatory flowfields in adjacent cells at the interface according to
the conservation laws. Finally, a set of equations is established by combining all
the interface and boundary conditions. The procedure eventually leads to deter-
mining the stability characteristics of the entire system of concern. The analysis
is validated against several well-defined problems for which either closed-form or
numerical solutions are available. A parametric study is also conducted to inves-
tigate the underlying mechanisms for driving instabilities in a model combustor.
A. Wave Equation
A generalized wave equation that characterizes the acoustic motions of a two-
phase mixture in the combustion chamber is derived. Following the approach by
Culick and Yang,4,5 the conservation equations of mass, momentum, and energy
can be expressed in the following forms, written for the behavior of the gas phase:
∂ρ
+ u · ∇ρ = W (14.1)
∂t
∂u
ρ + ρu · ∇u = −∇ p + F (14.2)
∂t
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 419
∂p
+ u · ∇ p = −γ p∇ · u + P (14.3)
∂t
where u and p are the velocity and pressure of the gas phase, respectively but
where ρ and γ are the mass-averaged values of the condensed and gas phases. The
source terms W, F, and P include the effects of viscous dissipation, heat release,
and two-phase interactions. Their explicit expressions are given in Refs. 4 and 5.
To derive the wave equation, each dependent variable is decomposed into a
time-mean and a fluctuating quantity:
u(r , t) = ū(r ) + u (r , t)
ρ(r , t) = ρ̄(r ) + ρ (r , t) (14.4)
p(r , t) = p̄(r ) + p (r , t)
The density fluctuation contains two components: an isentropic part that propagates
in the form of an acoustic wave, and a nonisentropic part that results from entropy
oscillation and is convected with the local mean flow. Thus, from the equation of
state, we have
∂ρ ∂ρ
dρ = dp + ds (14.5)
∂p s ∂s p
where C p is the constant-pressure specific heat of the mixture. Similarly, the tem-
perature fluctuation can be written as
1 ā 2
T = p + s (14.7)
ρ̄C p (γ − 1) C 2p
Substituting the decomposed variables into Eqs. (14.2) and (14.3) and linearizing
the result yields
∂u
∇ p = −ρ̄ − ρ̄(ū · ∇)u − ρ̄(u · ∇)ū − ρ (ū · ∇)ū + F (14.8)
∂t
∂ p
+ ū · ∇ p + u · ∇ p̄ = −γ̄ ( p̄ · ∇u + p · ∇ ū) + P (14.9)
∂t
We differentiate Eq. (14.9) with respect to time and substitute Eq. (14.8) for ∂u ∂t
to find the wave equation governing the oscillatory field in a two-phase mixture:
1 ∂ 2 p
∇ 2 p − =h (14.10)
ā 2 ∂t 2
420 D. YOU AND V. YANG
h = hI + hI I + hI I I (14.11)
And where
∂u
h I = − ∇ ρ̄ − ∇[ρ̄(ū · ∇)u ] − ∇[ρ̄(u · ∇)ū]
∂t
p 1 ∂ 1 ∂
− ∇ 2 (ū · ∇)ū + 2 (ū · ∇ p ) + 2 (u · ∇ p̄) (14.11a)
ā ā ∂t ā ∂t
γ̄ ∂ 1 ∂
+ 2 ( p̄∇ · u ) + 2 ( p ∇ · ū)
ā ∂t ā ∂t
ρ̄s
hI I = ∇ (ū · ∇)ū (14.11b)
C̄ p
1 ∂
h I I I = − 2 P + ∇F (14.11c)
ā ∂t
The first term, h I , represents the linear gasdynamic effect. The second term, h I I ,
arises from entropy fluctuations. The third term, h I I I , is related to processes,
such as two-phase interactions, combustion heat release, and viscous dissipation.
Equation (14.10) suggests that combustion instabilities can be viewed as classical
acoustic motions perturbed by various source terms in the chamber. The major
driving source inevitably arises from unsteady heat release because of its exceed-
ingly large contribution to the energy of the oscillatory field as compared with
other source terms.
B. Boundary Conditions
The boundary condition for Eq. (14.10) is set on the gradient of p , obtained
by taking the scalar product of the outward normal vector with the perturbed
momentum equation, Eq. (14.8).
n · ∇ p = − f
∂u
= n · − ρ̄ − ρ̄(ū · ∇)u − ρ̄(u · ∇)ū − ρ (ū · ∇)ū + F (14.12)
∂t
u /ā
Ad = n · (14.13)
p /γ p̄
1 2 3 j j +1
∞
∞
p (r , t) = [ψmn (θ, r )ηmn (x, t)] (14.14)
n=0 m=−∞
where subscripts m and n stand for the spatial variations in the circumferential and
radial directions, respectively. The problem then becomes solving for the series
coefficients, ηmn (x, t).
The eigenfunction, also called the normal mode, satisfies the Helmholtz equation
in the transverse plane,
subject to the following boundary condition for a rigid surface along the combustor
wall.
n · ∇T ψmn = 0 (14.16)
422 D. YOU AND V. YANG
where kmn is the wave number. The transverse Laplacian operator ∇T2 in the cylin-
drical coordinates is defined as
1 ∂ ∂ 1 ∂2
∇T =
2
r + 2 2 (14.17)
r ∂r ∂r r ∂θ
where the integral is performed throughout the entire cross section and where δ
denotes the Kronecker delta function.
For an annular duct with the inner and outer radii of Rt and Rh , respectively,
the eigenfunction can be expressed as
1
ψmn (θ, r ) = √ exp(imθ )[C J mn Jm (kmn r ) + CY mn Ym (kmn r )] (14.19)
2π
where
−1/2
Rt2 m2 Rh2 m2
C J mn = 1 − 2 2 Bm,t −
2
1− 2 2
Bm,h (14.20)
2 kmn Rt 2 kmn Rh
and where
dJm (kmn r ) dr
CY mn = − C J mn (14.21)
dYm (kmn r ) dr
r =Rt
For a circular duct with radius Rc , the transverse eigenvalue takes the form
1
ψmn (θ, r ) = √ exp(imθ )C J mn Jm (kmn r ) (14.24)
2π
where
−1/2
Rc2 m2
CJ mn = 1 − 2 2 Jm (kmn Rc )
2
(14.25)
2 kmn Rc
The eigenvalue kmn can be obtained from the following wall condition:
d
Jm (kmn r ) = 0 at r = Rc (14.26)
dr
Applying Green’s theorem and substituting boundary conditions (12) and (16) into
Eq. (14.27) yields
∂ 2 p 1 ∂ 2 p
p kmn
2
ψmn − ψmn + ψ mn 2 ds = − ψmn hds − ψmn f T dl
∂x2 ā ∂t 2
(14.28)
where f T = −n · ∇T p at the wall. The line integral dl is performed along the
surface of the cross section. Substitute Eq. (14.14) into Eq. (14.28) and rearrange
the result to yield
∂ 2 ηmn 1 ∂ 2 ηmn
2
kmn ηmn − + ψ 2
ds = − ψmn hds − ψmn f T dl
∂x2 ā 2 ∂t 2 mn
(14.29)
For linear stability analysis, each fluctuating quantity can be decomposed into a
spatial and a time-harmonic temporal part, i.e.,
ηmn (x, t) = η̂mn (x) · exp(i t), h (r , t) = ĥ (r ) · exp(i t), etc. (14.30)
424 D. YOU AND V. YANG
= r + i i (14.31)
The real part r represents the radian frequency of oscillation, and the imaginary
part i is called the damping coefficient, because its value determines the decay
rate of a particular acoustic mode. Consequently, the equation governing the axial
variation is derived as follows:
d2 η̂mn 2
+ − k 2
mn η̂mn = ψ mn ĥds + ψ fˆ
mn T dl (14.32)
dx 2 ā 2
The source terms ĥ and fˆT are functions of both the mean and oscillatory flow
properties. The latter consists of a series of transverse acoustic modes. For a reason-
able approximation, the acoustic mode coupling in evaluating the source terms in
Eq. (14.32) can be ignored because of the disparity of the length scales associated
with these modes. It can easily be shown that the cross-coupling terms are much
smaller and that only the specific mode of concern dominates. Thus, to facilitate
formulation, the surface and line integrals on the right-hand side of Eq. (14.32)
can be modeled as the products of the axial variation η̂mn and coefficients C h,mn
and C f,mn , which incorporate all the distributed and surface effects at a given cross
section through spatial averaging:
ψmn ĥds = C h,mn η̂mn (x)
ψmn fˆT dl = C f,mn η̂mn (x) (14.33)
+
+ −
−
η̂mn (x) = pmn exp iαmn x + pmn exp iαmn x (14.34)
− +
where pmn and pmn are the complex amplitudes of the upstream and downstream
±
traveling waves, respectively. The axial wave number αmn is related to the fre-
quency , eigenvalue kmn , and source-term coefficients:
±
2 2 ± ±
αmn = 2
− kmn
2
− C h,mn αmn , · · · − C f,mn αmn , ··· (14.35)
ā
∞
∞
+ −
p (r , t) = ei t + iαmn
ψmn (θ, r ) pmn − iαmn
e x + pmn e x (14.36)
n=0 m=−∞
Similarly, the radial and circumferential velocity fluctuations are derived by sub-
stituting Eq. (14.36) into the linearized momentum equations in the radial and
circumferential directions, respectively:
ei t
∞ ∞ + −
+ iαmn − iαmn
∂ψmn (θ, r ) pmn e x pmn e x
v (r , t) = i · + + −
ρ̄ n=0 m=−∞ ∂r + ūαmn + ūαmn
(14.38)
ei t ∞ ∞ + −
+ iαmn x − iαmn x
pmn e pmn e
w (r , t) = − mψmn (θ, r ) · + + −
ρ̄r n=0 m=−∞ + ūαmn + ūαmn
(14.39)
In addition to the acoustic field, entropy fluctuation arising in the flame zone
must be treated with care. The following transport equation is used:
Ds ∇ ·q
=− + σ̇ (14.40)
Dt ρT
The first term on the right-hand side represents the rate of entropy change caused by
the heat flux, q. The second term stands for irreversibilities, which can be modeled
as follows if we only consider the entropy generated from heat release, Q̇:
Q̇
σ̇ = (14.41)
ρT
Following the procedure described in Ref. 7, we obtain the equation governing the
transport of entropy oscillation:
∞
∞
s = exp(i t) exp(−i x/ū) ŝmn + Cs,mn ψmn (r, θ ) (14.42)
m=−∞ n=0
426 D. YOU AND V. YANG
where ŝmn is the amplitude of the entropy disturbance and where Cs,mn is a co-
efficient associated with the unsteady heat release represented by pressure and
velocity perturbations
+
Q̄ x Rp αmn Ru +
−
Cs,mn (x, r, θ ) = − + eiαmn x pmn
ρ̄ T̄ ū p̄ ρ̄ ā( + ūαmn )
−
Rp αmn Ru −
x −
+ − − e iαmn
p mn (14.43)
p̄ ρ̄ ā( + ūαmn )
2ū ∂ 2 p ū 2 ∂ 2 p
h= + (14.44)
ā 2 ∂ x∂t ā 2 ∂ x 2
± 2 M̄ ± ± 2
C h,mn = ψmn
2
− αmn − M̄ αmn
2
ds (14.45)
ā
h = − (γ − 1) i Q̇ /ā 2 (14.46)
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 427
where
i(γ − 1) Q̄
G±
hmn = −
ρ̄ ā 2
±
R p ρ̄ Ru αmn ψmn iRv ∂ψmn m Rw ψmn
× ψmn − ± + ± − ±
p̄ ā( + ūαmn ) ā( + ūαmn ) ∂r ār ( + ūαmn )
(14.49)
The formulation is thus closed after those combustion response functions are estab-
lished. Several empirical and analytical models, including the time-lag9 and flame
response10,11 models, were developed and employed to represent the combustion
responses in gas-turbine combustors.
∂v
f T = ρ̄ = ρ̄(i )v̂ (14.50)
∂t
where v is the radial velocity fluctuation at the resonator entrance. It can be related
to the local pressure fluctuation by means of the acoustic admittance function
defined in Eq. (14.13). Thus, we have
f T = i Ad p̂/ā (14.51)
428 D. YOU AND V. YANG
sˆ j −1
Rc,j
Rc,j-1 sˆ j
p +j −1 p+j
j-1 p−j −1 j
p−j
x0
0 0+
-
Fig. 14.3 Schematic of two adjacent cells with different cross-sectional areas.
Substituting of Eq. (14.51) into Eq. (14.33) gives rise to the formula for coefficient
C f,mn :
C f,mn = ψmn
2
i Ad /ā dl (14.52)
E. Matching Conditions
The oscillatory field in each cell must be matched with its counterpart in ad-
jacent cells by enforcing conservation laws at the interfaces. Figure 14.3 shows
schematically the fluctuating quantities on both sides of the interface at x0 between
two neighboring cells having different cross-sectional areas. The matching condi-
tions require continuities of mass, momentum, and energy fluxes over the region
0 ≤ r ≤ Rc, j−1 . It is assumed that Rc, j−1 ≤ Rc, j without loss of generality.
Mass flux:
Momentum flux:
p + ρu 2 j−1, x − = p + ρu 2 j, x + (14.54)
0 0
Energy flux:
1 1
C p T + u 2 = C p T + u 2 (14.55)
2 j−1, x0− 2 j, x0+
Because no mass flow is allowed to pass through the solid region Rc, j−1 ≤ r ≤ Rc, j
at the interface,
Equation (14.56) must be combined with the mass continuity equation (14.53) to
complete the mass-balance condition.
A general form of these matching conditions can be derived in terms of the
+ −
wave amplitudes pmn , pmn , and ŝmn and corresponding coefficients. The deriva-
tion is achieved by 1) decomposing the variables in Eqs. (14.53)–(14.56) into the
mean and fluctuating parts; 2) linearizing the results; 3) substituting the fluctuating
quantities given in Eqs. (14.6), (14.7), (14.36), (14.37), and (14.42) into the lin-
earized equations; and 4) combining terms with the wave amplitudes of the same
kind:
∞
∞
A+
m n pm+ n + A−
m n pm− n + Asm n ŝm n
m =−∞ n =0 j−1
∞
∞
+ A+pq p +pq + A−pq p −pq + Aspq ŝ pq =0 (14.57)
p=−∞ q=0 j
The explicit expressions of the coefficients in the preceding equation can be found
in Ref. 7.
Equation (14.57) indicates that the interfacial matching conditions contain in-
+ −
finite summations of unknown acoustic and entropy wave amplitudes, pmn , pmn ,
and ŝmn . In practice, a finite number of modes is sufficient because higher modes
will be either cut off or damped out. Thus, the total numbers of modes in the cir-
cumferential direction, M, and in the radial direction, N , can be selected to provide
a faithful solution. The number of unknowns in terms of wave amplitudes within
each cell is 3MN . For this reason, 3MN equations are required for each interface.
However, only three equations, Eqs. (14.53)–(14.55), have been formulated so far.
The additional equations can be constructed by employing orthonormal proper-
ties of eigenfunctions, according to the procedure given subsequently. A detailed
derivation is given in Ref. 7.
The general form for the momentum and energy interfacial conditions, Eqs.
(14.54) and (14.55), given in Eq. (14.57), is further manipulated by multiplying
the eigenfunction ψmn, j−1 and integrating the result over the cross-sectional area
S j−1 . This operation yields
M−1
N −1
A+
m n pm+ n + A−
m n pm− n + Asm n ŝm n ψmn, j−1 ds
m =−∞ n =0 j−1
S j−1
M−1
N −1
+ A+pq p +pq + A−pq p −pq + Aspq ŝ pq ψmn, j−1 ds = 0 (14.58)
p=−∞ q=0 j
S j−1
m = 0, 1, . . . , M − 1 (14.59)
n = 0, 1, . . . , N − 1
Equation (14.59) can be rearranged in the following form:
+ + − −
Bmn pmn + Bmn pmn + Bsmn ŝmn j−1
M−1 N −1
+
+ B pq,mn p +pq + B − −
pq,mn p pq + Bs, pq,mn ŝ pq =0
p=−∞ q=0 j
m = 0, 1, . . . , M − 1 (14.60)
n = 0, 1, . . . , N − 1
Similarly, the general form of Eq. (14.57) for the mass interfacial condition,
Eqs. (14.53) and (14.56), can be rearranged by multiplying eigenfunction ψmn, j
and integrating the result over the cross-sectional area S j , yielding
M−1 N −1
+
+ − −
Bm n ,mn pm n + Bm n ,mn pm n + Bs,m n ,mn ŝm n
m =−∞ n=0 j−1
+ + − −
+ Bmn pmn + Bmn pmn + Bsmn ŝmn j = 0,
m = 0, 1, . . . , M − 1
n = 0, 1, . . . , N − 1 (14.61)
Consequently, 3MN matching conditions are obtained from the mass, momentum,
and energy balances and are given by Eqs. (14.60) and (14.61).
F. Boundary Conditions
The acoustic boundary conditions at the combustor inlet and outlet play an
important role in determining the stability behavior of the entire system and must
be carefully treated. The effect of the boundary can be effectively measured with
an acoustic admittance function Ad , which quantifies the magnitude and direction
of the energy flow across the boundary, as evidenced in the expression for the
acoustic intensity, I :
I = p u = Ad · p 2 /(ρ̄ ā) (14.62)
It is apparent that energy is delivered into the system if the pressure and velocity
fluctuations are in phase, thereby exerting a destabilizing influence. A more rig-
orous theory of linear combustion stability clearly shows that the growth rate and
frequency of oscillation depend on the real and imaginary parts of the admittance
function, respectively.4
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 431
ŝmn = 0 (14.64)
At the combustor exit, the boundary conditions can be more conveniently spec-
ified by the reflection coefficients. The acoustic wave reflected from the boundary
consists of contributions from the incident acoustic and entropy disturbances. Thus,
the pressure of the reflected wave takes the form
p − = pa− + ps− = βa p + + βs ŝ· (γ p̄) C p (14.65)
βa = pa− / p + (14.66)
By applying Eq. (14.65) at the combustor exit, we can express the outlet boundary
equation as follows, in terms of the amplitudes of the incident and reflected acoustic
waves, as well as the entropy fluctuation:
+ + − − γ p̄
βa exp(iαmn d x) pmn − exp(iαmn d x) pmn + βs ŝmn exp (−i d x/ū) = 0
C̄ p
(14.69)
relation.8
2u ρ p
+ − =0 (14.70)
ū ρ̄ p̄
1 − (γ − 1) M̄/2 − M̄/2
βa = and βs = (14.71)
1 + (γ − 1) M̄/2 1 + (γ − 1) M̄/2
In the limit of zero Mach number, βa and βs approach unity and zero, respectively,
rendering an acoustically closed boundary.
G. System Equations
It has previously been shown that the number of unknowns in terms of the wave
+ −
amplitudes pmn , pmn , and ŝmn (n = 0, 1, · · · , N − 1 and m = 0, 1, · · · , M − 1)
for the oscillatory flowfield in each cell is 3MN , where M and N are the numbers
of the tangential and radial acoustic modes, respectively. If the combustor is divided
into NT cells along the length of the chamber, then in addition to the frequency ,
the total number of unknowns is 3 · N · M · NT unknowns. The number of equa-
tions is also 3 · N · M · NT , obtained by combining the (3 · N · M) · (NT − 1)
interfacial matching conditions, in the form of Eq. (14.60), and the 3MN inlet and
exit boundary conditions given in Eqs. (14.63), (14.64), and (14.69). As a result,
a set of equations governing the acoustic characteristics of the entire system is
established as follows:
C1 ··· ··· ··· ··· ··· D1
. .. .. .. .. .. .
.. . . ..
. . .
··· ··· C j−1
··· ··· ··· D j−1
=0 (14.72)
··· ··· ··· Cj ··· ··· Dj
. .
. .. .. .. .. .. .
. . . . . . .
··· ··· ··· ··· ··· C NT
DNT
R2
R1
L1
L2
Mode 1T 1T/1R
thereby serving as a challenging test problem. The duct is discretized into two
cells, and the cross-sectional area of each cell is uniform. Three different acoustic
modes are studied: the first longitudinal (1L), the first tangential (1T), and the
mixed first tangential/first radial (1T/1R) mode. If all source terms are absent and
if the mean-flow Mach number is ignored, the frequency of oscillation depends
only on the sound speed and chamber configuration. In other words, the acoustic
characteristics in each cell can be determined by the following Helmholtz equation,
subject to appropriate boundary conditions:
d 2 η̂mn 2
+ − k mn η̂mn = 0
2
(14.73)
dx2 ā 2
If the duct is acoustically closed, the frequency of the 1L mode can be analytically
determined from the following equation:
where L 1 , L 2 , R1 , and R2 denote the lengths and radii of the small and large
cells, respectively. The present analysis predicts the longitudinal-mode frequencies
identical to the analytical solutions of Eq. (14.74).
Results for transverse oscillations are validated against calculations by the finite
element software ANSYS.13 Table 14.1 compares the frequencies of the 1T and
1T/1R modes obtained from the present analysis and from ANSYS. The discrep-
ancy is less than 1%. Two azimuthal modes (i.e., M = 3) and two radial modes
(i.e., N = 3) were used to described the wave motion in each cell. Fig. 14.5 shows
the calculated acoustic pressure fields of the two modes. For pure transverse oscil-
lations in a straight duct with uniform temperature, the axial wave number in Eq.
(14.35) is zero. The oscillation frequencies normalized by ā/R2 for the 1T mode
of the simple small and large cells are 3.68 and 1.84, respectively. For the 1T/1R
mode, the normalized frequency of the large cell is 5.33. However, because of the
geometric change the axial wave number αmn for a duct with an area variation can-
not be zero. For the 1T mode, the acoustic field in the large cell dictates the wave
motion in the entire chamber. The frequency of 1.96 is close to that of a simple cell
with a diameter R2 (i.e., 1.84). This frequency yields a nontrivial complex axial
wave number in the small cell to satisfy the boundary condition, thereby leading to
an axially attenuated 1T wave. In other words, the 1T mode is cut off in the small
cell, in which the oscillation has a much lower amplitude, as shown in Fig. 14.5a.
Similarly, for the 1T/1R mode in the entire chamber, the prevalence of the wave
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 435
a) b)
1.00
0.60
0.20
c) d)
−0.20
−0.60
−1.00
motion in the large cell results in a cutoff phenomenon in the small cell, as shown
in Fig. 14.5b. The normalized frequency of 4.40 for the whole duct falls between
3.68 (i.e., the 1T mode for a simple small cell) and 5.33 (i.e., the 1T/1R mode for
a simple large cell).
The coexistence of different modes with structures corresponding to the 1T
mode in the small cell and the 1T/1R mode in the large cell at a single frequency
(4.40) represents a phenomenon called modal coupling. In the present case, the
coupling arises from the abrupt change in the cross-sectional area, which leads to
mode transition in the axial direction, as evidenced in the acoustic pressure field
on a longitudinal plane along the centerline shown in Fig. 14.6. The transition
is influenced by the cell length. For example, when the lengths of both cells are
reduced by one-half, the frequencies of the entire duct become 2.13 for the 1T
mode and 4.75 for the 1T/1R mode. This increase in frequency is attributed to the
stronger transition caused by the shorter length. However, when the duct length
is increased, the frequencies of transverse oscillations decrease as predicted. In
any case, the frequency of the entire duct always lies between those of the two
individual cells.
a) b) 1.00
0.60
0.20
−0.20
−0.60
−1.00
The prediction from the present analysis exactly matches the analytical solution
of Eq. (14.75). The normalized 1T frequencies for different cell lengths L 1 are
given in Fig. 14.8. Two azimuthal and two radial modes are used to represent the
wave motion in each cell. If L 1 = 0, the situation corresponds to an acoustic wave
propagating at the speed ā2 in a straight duct. Thus, the normalized frequency
/ (ā2 kmn ) becomes unity. As L 1 increases, the effect of the lower sound speed
ā1 becomes stronger. The frequency decreases and finally reaches its minimum
of ā1 /ā2 as L 2 approaches zero. Similar to the step-duct case, mode transition
phenomena are observed in the acoustic pressure field shown in Fig. 14.9. The 1T
mode is attenuated in the high-temperature section.
R
T1 T2
L1 L2
1
a1 / a2 = 0.9
0.9
Ω/(a2k mn )
0.8
a1 / a2 = 0.7
0.7
0.6
a1 / a2 = 0.5
0.5
0 0.25 0.5 0.75 1
L 1/(L 1 + L2)
Fig. 14.8 Normalized frequency of first tangential mode in a straight duct with tem-
perature jump.
Z
0.2
Y X
1 −0.2
0
−1
Fig. 14.9 Acoustic pressure field of first tangential mode in a straight duct with tem-
perature jump (ā1 /ā2 = 0.5, L1 = L2 = R); cross-sections at x = 0(z > 0 is blanked),
x = L, and slice z = 0.
235 mm
38 mm
Fig. 14.10 Schematic of a swirl-stabilized combustor.
438 D. YOU AND V. YANG
an inlet annulus and downstream with a choked nozzle. Broda et al.14 performed
extensive experiments on this combustor to obtain a stability map for the range
of operating conditions that is conducive to the occurrence of instabilities. When
the inlet air temperature exceeded a threshold value and the equivalence ratio fell
into a certain range, substantial pressure oscillations occurred, with their limiting
amplitudes being about 20% of the mean quantity. The underlying mechanisms for
driving instabilities are discussed in detail in the chapter by Huang et al. (Chap. 10)
in this volume.
Two cases are investigated herein. Case 1 deals with a stable operating con-
dition with an inlet temperature of 600 K. Case 2 corresponds to an unstable
situation with an inlet temperature of 660 K. The equivalence ratio remains at
0.573 for both cases. Figure 14.11 shows the physical domain of concern. The
chamber length is selected for convenient specification of the boundary condition.
At the inlet, the admittance function can be obtained from an impedance-tube
experiment for the swirler.14 At the outlet, the boundary condition for a choked
compact nozzle is applied, as given in Eq. (14.71). The mean flow properties
can be acquired from the numerical simulation of the conservation equations by
using either the Reynolds-averaged Navier–Stokes (RANS) or large-eddy simu-
lation(LES) approach.15 Figure 14.12 shows the mean-temperature contours and
streamlines on a longitudinal plane for two different inlet temperatures. A cen-
tral toroidal zone and a corner recirculation zone exist in both cases because of
the swirling effect and the geometrical configuration. In case 1, the flame spreads
from the corner of the centerbody to the chamber wall. In case 2, the flame is
anchored by both the corner-recirculating flow and the center-recirculating flow
and forms a compact enveloped shape, which is in sharp contrast with the shape of
case 1.
The combustion responses of these two flames to acoustic perturbations were
comprehensively analyzed by You et al.11 All known factors affecting the un-
steady heat release were examined, including the heat of the reaction, density,
flame speed, and flame-surface area. Briefly, the fluctuation of the heat of the re-
action is attributed to changes in the mixture-equivalence ratio resulting from flow
disturbances. The density fluctuation, mainly arising from pressure perturbation,
has a negligible effect on unsteady heat release, as compared with the other
20 mm
20.3 mm
45 mm
9.53 mm
224 mm
a)
b)
2.5
♦ pressure (experiment)
pressure
2.0
Normalized amplitude
velocity
1.5
♦
1.0 ♦
0.5
♦
0.0
0 0.1 0.2
Axial coordinate (m)
Fig. 14.13 Distributions of acoustic pressure and velocity oscillations in swirl-
stabilized combustor; first longitudinal mode.
440 D. YOU AND V. YANG
a) b) c)
d) e) f)
g)
a) b) c)
d) e) f)
g)
three factors. The oscillation of flame speed, similar to that of the heat of reaction,
is also caused by the equivalence-ratio fluctuation for a given chamber and flow
condition. The mechanisms of flame surface-area fluctuation are relatively com-
plicated and are primarily dictated by local velocity perturbations. The analytical
forms of the response functions derived in Ref. 11 accommodate the effects of
spatial variations in chamber geometry and mean flowfield and can be effectively
incorporated in the present stability analysis.
The computational domain was discretized into one and six axial cells in the
inlet and combustor, respectively. Since oscillations with frequencies greater than
the 1R mode can be ignored, only two azimuthal (M = 3) and two radial (N = 3)
modes are considered in each cell. Another calculation involving a total of 14 cells
442 D. YOU AND V. YANG
was also performed for comparison. The slight discrepancy of 0.6% in predicted
frequencies between the two cases confirms the validity of the use of seven cells in
determining the system stability behavior. Table 14.2 summarizes the calculated
oscillation frequencies and damping coefficients of the first longitudinal, first tan-
gential, and first radial modes for two different inlet temperatures. The acoustic
motion in case 1 is stable, whereas that in case 2 is unstable. The phenomena
are consistent with experimental observations14 and numerical simulations.16 The
calculated frequency of 1753 Hz in Case 2 matches closely the experimental value
of 1750 Hz. The spatial distribution of the 1L mode shown in Fig. 14.13 further
demonstrates the validity of the present analysis. The spatial distributions of the
1T and 1R modes are shown in Figs. 14.14 and 14.15. A detailed discussion of
the underlying mechanisms of driving combustion instabilities in the chamber is
given in the chapter by Huang et al. (Chap. 10) in this volume.
V. Conclusion
In this chapter, a three-dimensional linear acoustic analysis of gas-turbine com-
bustion instability was established. The work provides an effective means of un-
derstanding, analyzing, and predicting the stability characteristics of gas-turbine
combustors with geometric and mean flow variations. The purpose is to construct
a general framework within which the effects of all known quantities, including
both distributed and boundary source terms, can be assessed quantitatively. The
approach was extensively validated against problems for which either analytical
or numerical solutions, augmented by experimental measurements, are available.
Sample calculations were also conducted to investigate the effects of chamber
geometry, mean-flow distributions, and unsteady heat release on the behavior of
oscillatory flowfields.
VI. Acknowledgments
The work reported in this chapter was sponsored in part by Rolls-Royce plc,
in part by the Air Force Office of Scientific Research under Grant No. F49620-
99-0290, in part by the NASA Marshall Space Flight Center under the Grant
NAG8-187, and in part by Pennsylvania State University. The encouragement and
support from M. S. Anand and B. Bullard are gratefully acknowledged.
References
1
Culick, F. E. C., “Stability of High-Frequency Pressure Oscillations in Rocket Combus-
tion Chambers,” AIAA Journal, Vol. 1, No. 5, 1963, pp. 1097–1104.
2
Mitchell, C. E., “Analytical Models for Combustion Instability,”Liquid Rocket Engine
Combustion Instability, edited by V. Yang and W. E. Anderson, Progress in Astronautics
and Aeronautics, Vol. 169, AIAA, Washington, DC, 1995, pp. 403–430.
3
Zinn, B. T., and Powell, E. A., “Nonlinear Combustion Instability in Liquid Propel-
lant Rocket Engines,” Proceedings of the 13th Symposium (International) on Combustion,
Combustion Institute, Pittsburgh, PA, 1971, pp. 491–503.
4
Culick, F. E. C., and Yang, V., “Prediction of the Stability of Unsteady Motions in Solid
Propellant Rocket Motors,” Chapter 18, Nonsteady Burning and Combustion Stability of
ANALYSIS OF GAS TURBINE COMBUSTION DYNAMICS 443
I. Introduction
F OR stationary gas turbines, the drive for lower emissions of oxides of nitrogen
during the past decade has lead to the widespread use of lean premix burners
and convectively cooled combustion chambers. These technological changes have
resulted in a reduced stability of flame anchoring and lower acoustic damping.
Consequently, modern gas turbines are more susceptible to combustion-driven
oscillations and the importance of thermoacoustic phenomena in gas turbine com-
bustors has increased sharply.
To prevent acoustic instabilities, accurate models are needed to describe the dy-
namic properties of the combustion process and the propagation of acoustic waves.
This information can then even be used in the design process of a combustor to op-
timize combustor design. Knowing the stability borders of a combustor allows us to
define the operational concept for the gas turbine without exceeding instability am-
plitudes that would harm the combustor. An additional advantage is the reduction
of commissioning time and expensive and time-consuming engine testing.
The thermoacoustic system can then be modeled as a network of acoustic el-
ements, in a similar way to that described by Munjal1 and Polifke et al.2 To de-
termine the stability borders of an acoustic network containing measured transfer
functions, no transient properties of the transfer functions are required; thus, the
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved.
∗ Chaired Professor.
† Group Leader.
‡ Senior Scientist.
§ Department ad interim Leader.
445
446 C. O. PASCHEREIT ET AL.
transfer matrices only have to be measured as a function of the real part of the fre-
quency. A detailed description is given in Chapter 12 by Ann Dowling. The whole
system can then be assessed either in the frequency domain or in the time domain.
For gas-turbine manufacturers, stability borders are of crucial interest. Because
operating conditions such as power, flame temperature, and pilot to premix fuel
ratio affect the pulsation behavior of the combustor, knowledge of the combustor
stability map is one of the most important points to allow smooth operation of the
engine. The modeling of pressure pulsation amplitudes as well as the mode distri-
bution is the load input for the mechanical integrity assessment of the combustor
and thus is the basis for lifetime calculations. Modern gas turbines are required to
work with increasing operation hours between between major overhauls. The use of
advanced simulation tools in the design process allows the delivery of combustors
that exceed customer’s requirements.
A general method to predict stability of complex thermoacoustic systems in the
frequency domain was presented by Schuermans et al.3 The method used to solve
for the stability borders was based on a method described by Lang et al.4 and
relies on solving the complex eigenvalues of the resulting system; however, the
equations are solved by using a numerical and graphical approach that enables one
to find the stability borders of networks of any complexity. The stability borders
of an atmospheric combustion test facility with variable length and variable exit
conditions were predicted and compared with experimental results. The influence
of the thermal power of the combustion process on the transfer function is measured
and corresponds with basic physical understanding of the thermoacoustic process.
Although the network modeling of the combustor acoustics and the methods
that assess combustor stability require fast and efficient models, the main task
in combustor stability modeling remains properly describing the burner and the
flame. Without this knowledge, confidence in these tools remains low. A proper
description of the flame as a function of its operational parameters is needed to
increase confidence about the simulation.
Thermoacoustic flame models have been derived that describe the interaction
between the acoustic waves and the combustion process; they are discussed in de-
tail in Chapter 12 by Lieuwen. However, because of the highly three-dimensional
flowfield of swirl-stabilized combustion and the interaction between heat release
and the flow-field, these flame models often are unrealistic. An experimental de-
termination of the transfer matrix is still preferred. The measured transfer matrix
can be used directly in the network model, or it can be used to validate analytically
derived transfer functions.
Paschereit et al.5,6 and Schuermans et al.7 developed a method to experimentally
determine transfer matrices of swirl-stabilized flames. The method is based on a
technique described by Cremer,8 Bodén and Åbom,9 and Lavrentjev and Åbom,10
who applied these techniques to describe the acoustic properties of fans in flow
ducts. The element “burner and flame” is considered to be a black box that takes
into account the complex interaction between turbulent flow, flow instabilities,
and unsteady heat release. The property of the burner to produce sound by the
interaction of burner flow instabilities and unsteady heat release was taken into
account by measuring its source term.
This chapter describes how these approaches were extended by using computa-
tional fluid dynamics to determine the main parameters of the analytical transfer
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 447
+ + +
function. The integration into a fast and efficient simulation of combustor pulsation
properties is described, and application to real engines is demonstrated.
p̂d T11 T12 p̂u ps
= + (15.1)
û d T21 T22 û u us
fd SC11 SC12 fu fs
= + (15.2)
gu SC21 SC22 gd gs
448 C. O. PASCHEREIT ET AL.
Here, the subscripts u and d indicate upstream and downstream properties. The
elements of the transfer matrix [Eq. (15.1)] and the scattering matrix [Eq. (15.2)]
are complex-valued and a function of the angular frequency ω. Both descriptions
are equivalent and can be transformed from one description to the other. It has been
assumed that the acoustic waves are longitudinal and propagate one-dimensionally.
This assumption is valid because wave lengths corresponding to the frequency
range of interest are much larger than the non-axial dimensions of the test rig. The
hat in the equations denotes the complex-valued amplitude of acoustic pressure
and velocity.
The transfer matrix of the burner with flame is of crucial importance because
an interaction between acoustic fluctuation and heat release by the flame (which
may be a driver for thermoacoustic instabilities) takes place in this element. This
interaction could possibly be the result of coherent vortices building up and break-
ing down, resulting in fluctuating flame surface area and causing fluctuating heat
release. This coupling can also be caused by pressure and velocity fluctuations at
the burner location, resulting in fluctuations in the fuel–air ratio, and therefore in
oscillating heat release.
The transfer matrix and the scattering matrix in Eqs. (15.1) and (15.2) describe
a passive element, that is, an element that does not generate sound itself but that
amplifies and reflects or transmits incoming signals. If an acoustic element contains
an independent source of acoustic energy, these relations are not valid any more,
since the sound generated by the element has to be added to the outgoing waves
of the element.
In the case of gas-turbine burners, the source term contains noise generated by
the turbulent flow in the burner and flame, possibly involving large-scale hydrody-
namic structures. The source term consists of “colored” noise caused by turbulent
flow; it is expected to have a preferred frequency that depends on the Strouhal
number and a magnitude that depends on the mean flow velocity. If relations such
as Eqs. (15.1) and (15.2) can be found for all elements in an acoustic network,
either analytically or experimentally, then these relations can be combined to-
gether with the appropriate boundary conditions in one linear system of equations
[Eq. (15.3)]:
Sr = q (15.3)
Here, S represents the system matrix: a square matrix containing the coefficients
of all the transfer and scattering matrices of the individual network elements.
The vector of unknowns, r , contains the unknown quantities: Riemann invariants
( f and g) or acoustic pressure and velocity ( p and u) at every node of the network.
The forcing vector q contains the excitation signals or inputs of the system. They
can be the source terms but can also be the excitation signal of, for example, a
loudspeaker or fuel flow actuator. The matrix S and vectors r and q are functions of
the angular frequency, ω. If the transfer matrices and source terms of all network
elements are known, Eq. (15.3) can be solved for r to find the response of the
system to acoustic excitation or to perform a stability analysis.
The stability of the system can be determined from the homogeneous system of
equations derived from network analysis or, equivalently, by evaluating the Nyquist
diagram of the open-loop transfer function and using the method developed by
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 449
Polifke et al.2 If the system is unstable or close to the instability border, extremely
high amplitudes can be expected. However, even if the criterion of stability is ful-
filled, the magnitude of the oscillations may be unacceptably high if the forcing by
inhomogeneities goes into resonance with one of the otherwise stable eigenmodes
of the system. To determine the pressure amplitudes in this case, the response
of the system to the source term has to be calculated, which is equivalent to solv-
ing the nonhomogeneous system of equations in which the source terms appear
on the right-hand side. If one of the preferred frequencies of the source term is in
proximity to one of the resonance frequencies of the system, high amplitudes will
result.
combustion air
cooling air
4 loudspeakers (downstream)
water-cooled
water-cooled
sensor holders
sensor holders
(downstream)
(upstream)
air-cooled
swirl-stabilized combustion
burner glass tubes chamber
4 loudspeakers (upstream)
adjustable end
the microphone positions. The pressure signal can thus be considered as the sum
of three different contributions, as shown in Eq. 15.4:
P = Pe + Ps + Pr (15.4)
The contribution p e represents sound that is the response of the system to the
excitation by the loudspeaker. This part of the sound is used when determining
transfer matrices. The system is forced with a sequence of pure tones at distinct
frequencies, so p e also consists of pure tones. The amplitude, p e , can be obtained
from the averaged cross power spectrum between the excitation signal and the
microphone signal.
The contribution p s represents sound that is the response of the system to the
source term. The source term consists of noise generated by the turbulent flow
through the burner and of noise generated in the swirl-stabilized flame. This part
of the noise is independent of and thus uncorrelated with the speaker signal. How-
ever, this part is not random; it is coherent in the sense that there is a high correlation
of this part of the signal at different axially spaced microphone positions. This part
of the sound is needed to determine the source term. The term p s can be obtained
from the averaged cross power spectrum between the microphone signals and a
reference microphone signal but only in the case in which no excitation by the
loudspeakers occurs.
The contribution pr represents random noise caused, for example by local tur-
bulence generating pressure fluctuations at the microphone location. This part is
uncorrelated with the speaker signal and the source term. Because of the local char-
acter of the turbulence, these signals have a high correlation only for very small
axial spacing of the microphones. By taking the averaged cross spectra between
microphone signals and a reference signal, this part of the noise will be averaged
out if the axial spacing between the microphones is large enough.
In Fig. 15.3, the inputs of the system are the sources of sound, e, s and r .
The microphone signals p1 and p2 are the response of the system to all three
inputs. The one-dimensional transfer function of the burner is H , and G 1 and G 2
describe how acoustic waves are propagated and reflected by the geometry. The
transfer function H and the source-term s have to be determined now from the two
microphone signals. Since s and r are not correlated to the excitation signal e, these
s r
p p
e 1 2
G1 H
G2
Fig. 15.3 Simplified representation of the combustion system. Three different contri-
butions to the pressure signal are shown: p e , sound caused by external excitation; p s ,
sound generated by the burner itself, source term; pr , random noise caused by local
turbulence, for example.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 451
contributions will vanish when averaging the cross spectra between the excitation
signal and the microphone signal, and thus p e can be obtained from this signal.
Using Eq. (15.5), the transfer function H can easily be found:
P̂ e2
H= (15.5)
P̂ e1
By calculating the cross spectra between the microphone signals and a signal of a
third reference microphone, only the random noise will be averaged out. In this way
pe,s can be obtained: it is the part of the microphone signal that is the response of the
source term and of the speaker signal. As shown in Eq. (15.6), the source term can
now be found by using the transfer matrix H , which was determined previously:
s = P̂ e,s e,s
2 − H p̂ 1 (15.6)
Once the transfer matrix H has been found, the source term can also be
determined from a second measurement, one in which the loudspeakers are shut
off. Equation (15.6) will still be valid for this case; the only difference is that e = 0.
The approach described above can be extended to systems with four pole-
transfer or scattering matrices. The multimicrophone method was used to obtain the
Riemann invariants from multiple axially distributed microphones; more details
about this technique are given in Paschereit.5 The scattering matrix can be calcu-
lated from the signals that are cross-correlated to the speaker signal by using Eq.
(15.2). Because the signals are cross-correlated to the speaker signal, the signal
does not contain any response of the source terms, and Eq. (15.2) is therefore
reduced by the source term.
Because four elements of the scattering matrix have to be found and since
Eq. (15.2) without the source term only provides two equations, at least two in-
dependent test states are needed to solve the system of equations. These two
independent test states are generated by forcing with speakers downstream and
upstream from the burner. The result is a system of equations [Eq. (15.7)], which
has to be solved for the four complex-valued elements of the scattering matrix:
e e
f d A f deB SC11 SC12 f u A f ueB
= (15.7)
gue A gue B SC21 SC22 gde A gde B
Here, subscripts A and B refer to test states A (upstream forcing) and B (down-
stream forcing). As previously indicated, the elements of the scattering matrix
are functions of the angular frequency, ω. After the scattering matrix has been
found, the source term can be found from the signals that were cross-correlated to
a reference microphone by using Eq. (15.8):
e,s
fs fd SC11 SC12 f ue,s
= − (15.8)
gs gde,s SC21 SC22 gde,s
This equation is also valid if there is no forcing from the speakers (e = 0).
452 C. O. PASCHEREIT ET AL.
These three independent test states can be generated by forcing upstream, down-
stream, and on both sides of the burner at the same time.
A. Experimental Validation
The previously described method has been used to measure the source terms and
scattering matrix of a swirl-stabilized premix burner at certain operating conditions.
A first validation of the method can be done by modeling the combustion test rig as
a network of acoustic elements and comparing the results obtained in this way with
measured results. The acoustic network of the test facility consists of a measured
reflection coefficient, duct with flow, measured scattering matrix of the burner and
flame, measured source term, duct with flow, and a measured reflection coefficient.
The reflection coefficients were determined by calculating the ratios of the
Riemann invariants at the entrance and exit of the combustion system. The network
was simulated with a computer program. This program combines all the elements
(obtained analytically or experimentally) of an acoustic network into one system
of equations. The nonhomogeneous system of equations was then solved to obtain
the response of the system to excitation by the source term. Because of the linear
approach, the oscillation amplitudes are proportional to the magnitude of the source
term if the system is stable. If the system is not stable, the absolute values of the
oscillations cannot be predicted by using this linear approach.
To validate the measurement method and the network modeling, the system
of equations is solved to find the spectrum of the fluctuations at a certain po-
sition in the combustion chamber. The spectrum of one of the microphone sig-
nals has been plotted together with the result of this simulation in Fig. 15.4. The
frequency scale is normalized by dividing the actual frequency by the highest
frequency measured; the highest frequency corresponds to a Strouhal number
of St = 2.68. The pressure scale is normalized by dividing the pressure spectra
obtained from simulation and experiment by the highest value in the measured
spectrum. Two cases were considered: 1) nonreflecting boundary conditions, and
2) reflecting boundary conditions. The nonreflecting boundary conditions in the
experiment were achieved by an orifice at the exit of the test rig.11 The simulation
of the nonreflecting boundary conditions was almost identical to the measured data
(Fig. 15.4a). This outcome is not surprising, since almost all components of the
network model consist of measured elements. However, it shows that no errors
have been made when processing the raw experimental data to obtain transfer ma-
trices, source terms, and reflection coefficients; and it shows that modeling acoustic
networks with measured transfer matrices and source terms yields valid results.
A real test for checking the predictive capabilities of the method is to change
one of the boundary conditions of the test rig and then compare measured pressure
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 453
a) 1.0
0.8
Normalized amplitude
0.6
0.4
0.2
Fig. 15.4 The predicted spectrum of the pressure fluctuation (dashed lines) and the
measured fluctuations (dotted lines): a) in the combustion test rig with a nonreflecting
exit; b) in the combustion test rig with a reflecting exit.
spectra with computational results obtained with the adjusted reflection coeffi-
cient and the previously determined burner transfer matrix and source terms. The
boundary at the exit has been changed to an almost fully reflecting end, and the
pressure spectrum in the test rig has been measured at the same operating condi-
tions. The comparison between the predicted and measured spectrum is shown in
Fig. 15.4b.
Again, good agreement exists between the values predicted by the model and
the experimental data. By comparing the spectra measured in the test rig using
a reflecting end and a nonreflecting end, not only are higher overall amplitudes
observed but a shift in the resonance frequencies can also be seen. This outcome
occurs because the reflection coefficient is a complex quantity; by changing the
geometry of the end, the absolute values as well as the argument of the reflection
coefficient differ. This change in phase of the reflection coefficient can cause a
shift of the resonance frequency.
The pressure amplitudes in Figs. 15.4a and Fig. 15.4b are normalized, but the
scaling factor for all graphs is the same. The two dominant peaks in the spectra
roughly correspond to the quarter wave and three-quarter wave resonance mode
of the combustion chamber. Around the peak frequency, the predicted absolute
454 C. O. PASCHEREIT ET AL.
1.0 3
abs(T22)
arg(T22)
0.8 2
Normalized abs(T22)
1
0.6
arg(T22)
0
0.4
–1
0.2 –2
0.0 –3
Fig. 15.5 Comparison of T22 element of the transfer function, measured at two dif-
ferent forcing amplitudes.
values do not match the experimental data well. An explanation of this mismatch
is that the assumption of linearity is no longer adequate, specifically for the source
term or the transfer matrix when the acoustic state of the system is at very high
amplitudes. In the case of a system driven by a source term, the response, that
is, the observed pressure spectrum, follows such nonlinear changes in the source
term. In the case of self-excited instability, modifications of the transfer matrix
caused by high amplitude will lead to nonlinear cycle limitation.
The linearity of the system has been assessed by forcing the system with two
different forcing levels. Figure 15.5 shows the T22 element of the transfer function
and proves its linearity.
0 1 2
Fig. 15.6 Sketch of the premix burner. Reference points 0, 1 upstream- and down-
stream of the burner element are indicated, as well as reference points 1, 2, upstream-
and downstream of the flame front.
where
1 1
A0 u(s)
L red = ds = ds (15.11)
0 A(s) 0 u0
is a virtual length of the oscillating air column inside the burner. The derivation of
Eq. (15.10) is based on the assumption of compactness, that is, no physical length
of the burner element. The transfer function of the cold burner was measured and
compared against the previously described model with good agreement.
B. Flame Model
A real flame does not have a steady position in the combustor, but varies in
position. This fluctuation is taken into account in the model. The flame is fed with
a premixed fuel–air stream, where the fuel injection takes place at the location i
inside the burner, before the flame front with upstream and downstream states
1 and 2; see Fig. 15.6. The common approach to model the acoustic behavior
of such a flame is based on the assumption that the acoustic and heat-release
fluctuations at the flame front are coupled with fluctuations in the fuel–air mixture
that are attributable to acoustic disturbances at the fuel injectors. This implies
the existence of a characteristic time lag τ , after which the fuel particles reach the
456 C. O. PASCHEREIT ET AL.
flame; thus, for the fuel–air mixture φ and its fluctuation in time φ (t),
φ1 (t) φ (t − τ )
= i (15.12)
φ1 φi
Q = φ1 ρ1 S f h fuel (15.13)
Q φ ρ
=2 1 + 1 (15.14)
Q φ1 ρ1
By using the Rankine–Hugoniot jump conditions across the flame and the pre-
ceding equations, we can formulate the flame model as
p̂ 2 1 ρ1 c1 TT21 − 1 M1 1 − 2e−iωτ p̂ 1
=
(15.15)
û 2 0 1 − TT21 − 1 2e−iωτ û 1
In particular, the amplitude of the velocity fluctuations across the flame has been
changed, which is important because no free parameter (such as the interaction
parameter n in the n–τ model) exists to adjust this condition.
The model is still based on a time-delay τ . It is an oversimplification for a
realistic flame, in which fuel injectors are spread over an axial distance inside the
burner, the flame is nonplanar, or both. As will be shown subsequently, the effect of
the time-delay spread can in fact significantly influence the stability characteristics
of a burner system. This effect can be incorporated in the flame model by dividing
the fuel inlet points of the burner into p submodels (an alternative that is not
pursued here is to divide the model at the flame front itself). The fuel from each of
the inlet points reaches the flame after a certain time delay τ j , and the following
relation for the velocity fluctuations across the flame can be derived:
p
2 T2 −iωτ j
û 2 = 1 − aj −1 e û 1 (15.16)
j=1
p T1
Here, only a burner configuration in which the fuel injectors are homogeneously
distributed is considered, and thus a j = 1 for the weight factors of each inlet point.
To illustrate the effect of the model, a situation with linearly distributed time
delays such that τ j [τmax − τ ; τmax ] is considered. In this case, the flame speed
model reads
T2 1 −iω(τmax −τ )
û 2 = 1 − 2 −1 e − e−iωτmax û 1 (15.17)
T1 iωτ
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 457
5 3
4 2
1
abs(T22)
arg(T22)
3
0
2
–1
1 –2
0 –3
4 3
2
3
1
abs(T22)
2 0 arg(T22)
–1
1
–2
0 –3
Fig. 15.7 T22 element of the flame transfer function, with measured (solid) and mod-
eled (dashed) values: a) best fit of the constant time-delay model; b) best fit for the
two-parameter model with linear distribution of time delays.
The improvement of this two-parameter model over the constant time-delay ap-
proach is shown in Fig. 15.7, in which both models have been used to fit experi-
mental results.
1. Numerical Setup
The CFD calculations of the experimental premix burner are based on the steady-
state three-dimensional Navier–Stokes equations with a second-order accurate fi-
nite volume solver. The geometry of the burner is schematically shown in Fig. 15.8.
The preheated and compressed air enters the swirler through the two inlet slots on
the cone shell in a circumferential direction. The resulting swirl flow breaks down
near the exit, leading to a recirculation region. Gaseous fuel is injected into the
passing airstream along the burner slots. Flame stabilization takes place in this inner
recirculation region formed by the vortex breakdown and in the outer recirculation
region formed by the dump plane. The combustor and the upstream plenum have
been modeled by an unstructured hybrid grid with approximately 600,000 cells.
The fuel injectors are not resolved explicitly in the CFD simulations. Instead, the
fuel injection is modeled by using numerical source terms that are placed shortly
downstream of the injection plane (see Fig. 15.9). It is an approximation of the real
configuration, in which the fuel emanates from an injection hole into the passing air
cross-stream. The fuel is injected into a grid cell through a numerical source term.
The expected fuel penetration at this location has been derived from the appropriate
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 459
injection pathline
injection cell
gas channel
Fig. 15.9 Cut through the cone shell at the inlet slot.
correlations and reference experiments. In this way, local variations in the mixture
fraction can be considered while keeping the computational cost reasonable.
The turbulent flow is modeled by the standard k − model and combustion is
incorporated by using a turbulent flame-speed closure model.14 In this model, the
combustion process is described by a reaction progress variable c, which takes
values between zero (fresh mixture) and one (burnt products); and the temper-
ature field is linked to the reaction progress variable and the local (mean) fuel
concentration.
The time delays τ j have been derived from the converged solution by injecting
particles into the flow at the location of the fuel injection and then tracking their
trajectories and measuring the travel times until they hit the flame front. The flame
front is defined somewhat arbitrarily where c = 0.5 for the reaction progress. We
assume complete burning at this location for the fuel tracer particles, and we do
not include effects of local variations in heat-release rates because of uncertainties
that are associated with the flame model itself.
To include the effect of turbulent diffusion on the displacement of the fuel par-
ticles for each of the fuel ports, a random velocity is added to the mean convective
velocity u, which is obtained from the CFD computation. The turbulent displace-
ment velocity is modeled by a Gaussian white-noise process dW(t) that is scaled
by the local turbulent kinetic energy k√in the flow, such that the total displacement
dx of the fuel particle is dx = udt + 2k/3dW(t). The time-step size dt has been
chosen such that the travel distance of the fuel particles, for each time step, was
small compared with the local grid resolution.
Each of the distributions presented subsequently has been obtained from tracking
q = 128 particle injection points distributed along the fuel-injection ports. Where
we added the effect of turbulent dispersion, n = 10 realizations have been used
for each injection location (n = 100 has also been used to check for convergence
in the statistics).
0.5
A
B
0.4 C
0.3
f
τ
0.2
0.1
2 4 6 8 10 14
0.5 1 1.5 2 2.5 3 3.5
τ/ τ ref
0.5
B
0.4
0.3
f
τ
0.2 D
0.1
2 8 10 14
0.5 1 1.5 2 2.5 3 3.5
τ/ τ ref
Fig. 15.10 Top: normalized distribution of time delays, as obtained from particle
trajectories. Bottom: For configuration B, the effect of turbulent diffusion on redis-
tributing particle trajectories is shown (D).
location). The distributions A–C (top part of Fig. 15.10) explicitly study the effect
of variations in fuel penetration. The distribution B denotes the reference simulation
that is expected to match best with the experiment; A and C are simulations with
lower and higher fuel-penetration depths, respectively, and could be associated
with, for example, smaller and larger fuel ports. A change in fuel penetration will
change the convective times τ j because that different streamlines inside the burner
will be fed by the fuel if the penetration changes, and as a consequence, the flame
shape changes, as well.
All three variants show the same characteristic behavior, namely, a distributed
time delay with two distinct peaks, one at τ/τref ≈ 1, and one at τ/τref ≈ 2 to 2.5.
Only the peak at large times is affected by the modification in fuel penetration,
and this effect can in fact be correlated with changes in the shape of the flame,
which are not shown here. Unfortunately, when incorporating these distributions
into the flame model [Eq. (15.16)], the agreement with the experiment is found
to be unacceptable. The damping of higher-frequency modes is not captured ap-
propriately. Also, the main mode of instability, where the phase of the measured
transfer function cuts the frequency axis, at a normalized frequency of St = 0.26,
is not captured appropriately; As subsequently explained, CFD generally has a
tendency to overpredict time delays.
This situation changes, at least partially, if the effect of turbulent mixing is
included on the fuel particle trajectories (bottom part of Fig. 15.10). Again, the
peak on the right of the distribution density function f τ is primarily affected. It is
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 461
4 3
absolute 2
3 phase
1
arg(T22)
abs(T22)
2 0
–1
1
–2
–3
0
0.2 0.4 0.6 0.8 1.0
Normalized frequency
Fig. 15.11 The T22 flame transfer matrix element, derived from the time-delay dis-
tributions in Fig. 15.10; measured (solid) and modeled (dashed) values.
Fig. 15.12 Normalized axial velocity along the burner axis; 0 corresponds to the
burner exit position.
annular duct is derived. A comparison with results obtained from finite element
analysis is made. Interconnecting of several systems into a network of acoustic
systems is done by using linear fractional transforms. A stability analysis is then
made by evaluating the eigenfrequencies of the interconnected system. Then in
a second step, the validity of this approach is demonstrated on a very simple,
one-dimensional thermoacoustic system. The eigenfrequencies are solved for ana-
lytically and compared with the results obtained from modal expansion and linear
fractional transforms. A second validation is performed on a system consisting of
two annular ducts interconnected by one-dimensional tubes; the eigenfrequencies
are compared with results obtained from finite element analysis. A network model
of an annular, multiburner, gas-turbine combustion chamber is then derived.
A. Network Interconnections
To obtain a model of the acoustic behavior of a gas-turbine combustion system,
acoustic transfer functions need to be combined in a network of acoustic elements.
The resulting system can then be analyzed to assess its stability to calculate sta-
bility borders or to calculate frequency spectra. Two different methods for system
interconnection and subsequent analysis will be discussed here. The first one is
a typical frequency-domain approach. The second method yields a state–space
representation and can be analyzed either in the frequency domain or in the time
domain. As an example for both methods, the lumped-element representation of a
combustion system shown in Fig. 15.13, is discussed.
p1 p2 p3 p4
P B F C E
u1 u2 u3 u4
Plenum Burner Flame us Combustion Exit
chamber
Fig. 15.13 Interconnection of subsystems of the combustion system; note that all
arrows represents vectors of input or output signals.
in this example. The elements in Fig. 15.13 are then all 2 × 2 transfer matrices,
except for P and E, which are 1 × 1 transfer functions. We also assume that all
transfer matrices or functions are stable. This assumption is generally a safe one;
the problem of thermoacoustic instabilities is not that one of the transfer functions
is unstable but that the interconnected system can become unstable under certain
conditions. A linear, stable system is completely characterized by its impulse
response or by its frequency response, which is the Fourier transform of the impulse
response. This property is exploited here to determine the stabiliy and stability
borders of the interconnected system.
All the transfer functions can be combined into one system of equations, as
shown in Eq. (15.18). The left-hand side of this equation consist of a large matrix
S(ω) that contains the transfer matrices and of a vector P̂(ω) that contains the
unknown pressures and velocities. The right-hand side of this equation contains
fˆ(ω) the source signals. In this example, the only nonzero entry is u s :
P −1 p1 0
B11 B12 −1 0 u1 0
B −1 p 0
21 B22 0 2
F11 F11 −1 0 u2 0
= (15.18)
F11 F11 0 −1 p3 0
C11 C11 −1 0 u3 us
C11 C11 0 −1 p4 0
E −1 u4 0
imaginary parts of all roots ωn are larger than zero, then the system is stable. If one
or more roots have negative imaginary parts, then the system is unstable. Hence,
finding all roots of the equation is crucial. As previously mentioned, the measured
transfer functions are only known for real-valued frequencies. Two ways of cir-
cumventing this problem are possible. The first possibility is to fit a function to the
measured transfer function. Once the function is obtained, complex values can be
substituted for the frequency. Great care has to be taken with the choice of function
to be used for fitting the experimental data. To make physical and mathematical
sense, the function should be analytic for imaginary frequencies smaller than zero.
For practical applications, fulfilling these conditions can be very difficult; and for
that reason, this method is not pursued here. The second possibility is to look for
stability borders of the system as a function of the operation parameters, such as
flame temperature and power, instead of solving for parametres. The results is
a map indicating the value of the parameter (such as a temperature, combustor
length, or time delay) for which the system changes from stable to unstable. It will
be demonstrated here that to find stability borders, the transfer functions only need
to be known as a function of the real part of the frequency. A system is said to
be on a stability border if the imaginary part of the frequency equals zero. Thus,
solving Eq. (15.19) yields the stability border for an arbitrary parameter τ at R(ω);
clearly, no information on the dependence on the imaginary part of the frequency
is needed:
2.5
Unstable
2.4
T2
2.3
T1
2.2
Stable
2.1
0.5 1.0 1.5 2.0 2.5
v /v
ref
Fig. 15.14 System stability for a typical test rig configuration; stability borders
for variations in burner velocity (i.e., power) vs flame temperature, are shown. The
simulations are based on the linear time-delay model.
466 C. O. PASCHEREIT ET AL.
v = v ref
Unstable
time
max in sec
v = 1.5 v ref
Stable
Convection
v = 2 v ref
Unstable
Mixing spread
n
Fig. 15.15 System stability for a typical test-rig configuration; stability borders for
variations in time delay spread vs convection time (i.e., 1/power) are shown.
stability border; the thickness of this border could be reduced for significantly
larger computing times. It is evident from this figure that both stable and unstable
regions can be expected for this system. This result also agrees with the exper-
imental observation. It also opens ways to stabilize and destabilize a system by
varying power levels or flame temperatures accordingly and can guide the operator
of a machine to avoid regions of unstable combustion.
D. Time-Domain Approach
Although analysis of systems in the frequency domain seems very straight-
forward, assessing the stability of the system is, strictly speaking, impossible if
no information on the dependence of the functions on the imaginary part of the
frequency is available. Even if all transfer functions have been defined in the en-
tire complex-frequency plane, stability analysis is not straightforward, because it
requires a numerical search for all possible eigenvalues. Especially if thermoa-
coustic systems modeling is extended to multi-input multioutput systems, such as
multiburner systems, the numerical root-finding procedure may be forbiddingly
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 467
expensive. The problems with determining the stability of the frequency domain
thermoacoustic models have led to a time-domain formulation of the problem. The
idea is to represent all transfer functions as differential equations and to combine
them all in one system of equations. This interconnection can be done in a very
elegant and straightforward manner by making use of state-space representations
and Redheffer star products. A state-space representation of a dynamic system is
a system of first-order differential equations that is equivalent to one higher-order
differential equation. The advantage of using state-space systems is that they are
numerically very robust, and that extension to multi-input multioutput systems is
straightforward. A linear, time-invariant state-space system has the general struc-
ture give in Eq. (15.20):
where u and y are vectors of input and output signals, x is an internal state vector
and the matrices A, B, C, and D represent the system. Any interconnection of such
systems again yields a system with the same general structure but with different
A, B, C, and D matrices. These matrices are real-valued and are independent of
time or frequency. The stability analysis of such a state-space system is thus very
straightforward: The system is stable if the matrix eigenvalues of A all have a
negative real part. The eigenvalues of a real-valued matrix are computed by using
standard methods available in linear algebra.
The entire system of Fig. 15.13 can now be modeled by interconnecting all the
outputs of the subsystems to the inputs of their “neighbors.” These subsystems
can be interconnected in a very convenient way by making use of the Redheffer
star product. The Redheffer star product is a matrix operation based on a linear
fractional transform.19 It is often used in control theory to model uncertainty in
systems but can be used to interconnect any network of state-space systems. The
interconnection of two ducts (or any other systems) H and G is then simply given
by H G, in which denotes the Redheffer star product, and is defined as
F1 (H, G 11 ) H12 (I − G 11 H22 )−1 G 12
H G =
G 21 (I − H22 G 11 )−1 H21 Fu (G, H22 )
in which Fu () and Fl () denote the upper and lower linear fractional transform,
defined as: Fl (M, g): = M11 + M12 g(I − M22 g)−1 M21 .
Thus, the system of Fig. 15.13 can easily be represented by the matrix
S = P B F C E. This system has no inputs or outputs; they could of course
be added, but they are not required for a stability analysis. The stability require-
ment of the system is then satisfied if the real parts of all eigenvalues of S are
negative.
Not only the stability analysis is very straightforward in this approach. Time-
domain simulations, even those including nonlinear elements, and frequency
responses can very easily be performed.
468 C. O. PASCHEREIT ET AL.
E. Modal Expansion
In this section, acoustic transfer functions are derived for geometries with mul-
tiple inputs and multiple outputs (MIMO). Starting from the wave equation with
sources on the surface f , but without sources in the volume,
∂ 2 p
c2 ∇ 2 p − =0 (15.21)
∂t 2
n̂ · ∇ p = − f (15.22)
∞
c2 ψn (x)
p̂(x) = ψ(xs ) fˆ(xs )dS (15.23)
n=0
(ω2 − ωn2 ) s
where = ψ 2 dV . The eigenfunctions ψ and eigenfrequencies ωn can be ob-
tained analytically for simple geometries or numerically (e.g., finite element
method) for more-complex systems. If the source function is a source of acoustic
velocity on the boundary, it can be written as: fˆ (ω, xs ) = iωρ û s (ω, xs )
An acoustic transfer function can be defined as the ratio between the acoustic
pressure at a certain position x in the volume to the acoustic velocity acting as an
input on an area As centered on the boundary at xs . If the extent of the area As is
small compared with the wavelength, then Eq. (15.23) can be rewritten to obtain
the transfer function H (ω) between p̂(x) and û s (x0 ):
p̂(x) ∞
ψn (x)ψn (xs )
H (ω) = = iωρ As c2 (15.24)
û s (xs ) n=0
(ω2 − ωn2 )
Equation (15.24). relates the acoustic pressure at one location to the acoustic
velocity at one other location. This single input single output (SISO) representation,
which corresponds to the acoustic impedance, can easily be extended to the general
MIMO case. The K velocities at xin are then related to the J pressures at x out by
a J × K transfer matrix H : p(xout ) = H u(x in ), in which the elements of H are
given by
∞
ψn (xj )ψn (xk )
H jk = −iωρ Ak c2 (15.25)
(n=0)
(ω2 − ωn2 )
1. State-Space Representation
Because all elements of the transfer matrix H have the same eigenvalues (ωn ) the
transfer matrix can be expressed more conveniently by a state-space representation.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 469
Note that α, the modal damping, has been introduced here. The value of α is
assumed to be small compared with ωn . Many different state-space representations
of a system are possible. The representation chosen here has the advantage that it
can easily be extended to the MIMO case. The state of the system is represented
by the 2 × 1 vector xn (this notation is chosen to be consistent with notation used
in control theory and should not be confused with geometrical position x). The
structure of the equations for the general case with N modes, J inputs, and K
outputs is the same as Eq. (15.26). However, the matrix A becomes a 2N × 2N
block-diagonal matrix and the matrices B and C become 2N × J and K × 2N
matrices, respectively:
ẋ1 A1 x1 B11 . . . B1J u1
.. . . . . .. ...
.
. = .. .. + .. ..
ẋ2N AN x2N BN 1 . . . BN J uj
p1 C11 . . . C1N x1 0 ... 0 u1
1 . .
.. = .. ..
. .. .. + .. . . . .. ...
. . . .
(15.27)
ρc
pK CK 1 . . . CK N x2N 0 ... 0 uJ
need to be known. For practical (often very complicated) systems, the eigen-
frequencies and vectors can be obtained from a finite element analysis. In finite
element analysis, only a modal analysis, which is very computationally efficient, is
required. Moreover, only the modal values at one position on the interface locations
are required; thus, very little output is needed. For more simple geometries, the
eigenfrequencies and vectors can be obtained analytically. As an example, the
transfer function of an annular duct is derived. A combustion chamber of a gas
turbine can be represented as an annular duct with J input and output ports, J
being the number of burners.
The input–output relation is given by a transfer matrix relating J inputs to J
outputs. The required eigenvalues and vectors for a thin annular duct of length L,
mean diameter D, and height h are given by
!
2cm 2
πcn 2
ωn,m = + (15.28)
D L
πnx cos(mφ)
ψn,m = cos
L sin(mφ)
L Dπ h
n,m = (2 − δkr on (n))(2 − δkr on (m))
L Dπ h
2(1 + δkr on (n))δkr on (m)
in which n and m are the numbers of the longitudinal and azimuthal modes,
respectively. Thus, the mode is notated here as (n, m).
Because of the rotational symmetry of the annular duct, all eigenvalues are
two fold degenerate except for n = 0 and have two orthogonal eigenmodes. Sub-
stituting of Eq. (15.28) into Eq. (15.26) and Eq. (15.27) yields the state-space
representation of a thin annular duct with colocated inputs and outputs at one side
of the duct:
−α ωn,m
ωn,m −α
=
ωn,m
An,m (15.29)
−α
ωn,m
−α
0 ... 0
cos(mφ1 ) . . . cos(mφ j )
Bn,m =
0 ... 0
sin(mφ1 ) . . . sin(mφ j )
(2 − δkr on (n))(2 − δkr on (m)) T
Cn,m = B n,m (15.30)
JL
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18
ω/c
Fig. 15.16 Frequency response of annular duct, calculated with Sysnoise (solid),
analytic solution (dotted), and modal expansion (dashed).
solution and with a solution obtained from the commercial finite element method
package Sysnoise (Fig. 15.16). The modal-expansion representation is mathemat-
ically equal to the analytic solution for N = M = ∞. However, for the result
presented in Fig. 15.16, values of N = 2 and M = 4 have been used. In Sysnoise,
the zero Hertz mode is not calculated because it causes numerical difficulties.22
As a consequence, the frequency response calculated by Sysnoise is incorrect for
the very low frequency regime.
This representation can easily be extended to the more general case with inputs
and outputs on both sides of the duct (at x = 0 and x = L). When doing so, it is
helpful to apply the following partioning of the B, C, and D matrices:
ẋ = Ax + BI u1 + Br ur (15.31)
pI
= C1 x + Dll u1 + D1r ur
ρc
pr
= Cr x + Dr1 u1 + Drr ur
ρc
in which l and r refer to the left- or right-hand side inputs and outputs. The matrix A
is the same as for the single-sided duct. The matrices B1 and C1 contain the values
of the eigenvector on the left side of the duct and are identical to the matrices B and
C in Eq. (15.29). The matrices Br and Cr contain the values of the eigenvectors
at x = L. Because cos (πnx/L) = (−1)n if x = L, the following expressions are
obtained for Br and Cr in annular ducts:
Although the D matrices are empty again, they are shown here to be consistent
with a more general notation of partitioned state-space systems:
H11 H11 A Bl Br
H= = Cl Dll Dlr
H21 H22
Cr Drl Drr
G. Examples
1. Can-Type Combustor
As a first example, the one-dimensional thermoacoustic system described in
Lang et al4 is analyzed. This system consists of a straight duct, closed on one side,
open on the other side, with a flame stabilized in the middle of the duct. The pressure
drop across the flame sheet is assumed to be negligible. The acoustic velocity jump
is modeled by the so-called n–τ model: u 2 (t) = u 1 (t) + nu 1 (t − τ ), in which τ is
a delay time and n is referred to as the interaction coefficient. The impedance of the
open end is simply Z 3 = 0. The n–τ model contains a delay and is thus of infinite
order. To avoid systems of infinite order, the time delay is approximated by a Padé
approximation, a technique commonly used in control theory. The upstream duct is
represented by a transfer function similar to Eq. (15.24) but becomes more simple
because m = 0 in the one-dimensional case. By using the Redheffer star product,
the system can be represented as: S = P ∗ F ∗ C ∗ Z. The eigenfrequencies or
poles of the system are the complex eigenvalues of the matrix S. The eigenvalues
of S corresponding to the first resonant mode have been calculated for several
values of τ . According to Lang et al.,4 the eigenvalues of the system are shown to
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 473
1.04 0.04
1.02 0.02
Real (ω)/ω0
Im (ω)/ω0
1.00 0.00
0.98 –0.02
0.96 –0.04
0 π 2π 3π 4π 0 π 2π 3π 4π
ωτ ωτ
Fig. 15.17 Eigenfrequencies of the one-dimensional combustion system as a funtion
of normalized time delay, τ , left: real part of frequency; right: imaginary part. Solid
line: analytic solution, dotted line: modal expansion.
be the roots of
ωL ωL
cos 2 − sin 2
ne−iωτ = 0 (15.32)
c c
which requires a numerical search. The results obtained through modal expansion
and the roots of Eq. (15.31) are both plotted in Fig. 15.17, in which ω0 is the
resonance frequency in the case n = 0. For large values of ω0 τ both curves deviate
because the order of the Padé approximation was relatively low (six).
6
5
4
3
ω/c
2
1
0
0 5 10 15 20 25 30 35
Eigenvalue number
Fig. 15.18 Eigenfrequencies of the coupled duct with temperature jump obtained
directly by Sysnoise (o) compared with the modal expansion method based on
numerically obtained eigenvalues and vectors (x) and using analytically obtained
eigenvectors (+).
√
easily obtained from: ωn = (c/cref )ωn,ref . The scaled eigenvectors: ψ/ remain
unchanged. Although the system interconnected in this way approximates the
numerical solution, a perfect match is not obtained—even for a system of very high
order because the interconnection at each interface position is one-dimensional, and
some important three-dimensional effects close to the interface are neglected. This
phenomenon is well known from Helmholz resonator theory: a length-correction
factor (virtual length) has to be applied to compensate for the local deformation
of the potential field. We can easily do this by adding a one-dimensional duct
element at the interconnection. The virtual length is very small; thus, a zero- or
first-order expansion is sufficient. The value of the length correction has been set
to (As /n)1/2 , as described in Rienstra and Hirschberg.24 The eigenfrequencies
of the coupled annular ducts calculated in three different ways are plotted in
Fig. 15.18, the values on the x-axis correspond to the numbering of the modes.
p1 p2 p2 p3
- ρ 3 c3
ρ 2 c2
Lζ(s)
u1
A1
A2
u2 u2
( − 1) e
T3
T2
− iωτ
e
1 ω 2σ 2
2
-
u3
Fig. 15.19 Block diagrams of the burner model (left), and the flame model (right).
rather than having one delay, a distribution of time delays is assumed here. The
block diagram is shown in Fig. 15.19. The relation between ∞ the acoustic veloc-
ities accross the flame is then given by û 2 (s) = (1 − n 0 ξ (τ ) e−sτ dr ) û 1 (s), in
which τ is the convective time delay between fuel injection and consumption
and ξ (τ ) is the probability density distribution function of time delays. The in-
teraction coefficient is given by n = 1 − T2 /T1 . The distribution of time delays
can either be obtained numerically13,26 or from experimental fits.27 If a Gaussian
distribution of time delays is assumed with mean value τ and standard deviation
στ , then, after carrying out the integration, the flame model can be written as
û 2 (s) = (1 − nes2 σ τ e−sτ )û 1 (s). In this case, the values of τ and στ have been
1/2 2
-1
Fig. 15.21 Hood and combustor modes.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 477
Hood
Burner Burner
pˆ j , uˆ j Flame Flame
Resonator
Combustor
channels and hood and by the large-flow Mach number at the combustor exit. For
acoustic wavelengths that are much larger than burner and resonator dimensions,
the acoustic pressure p̂ j and normal acoustic velocity û j are assumed to be uniform
on the opening area A j centered on the boundary at x j (see Fig. 15.22).
Cooling air for the Helmholtz resonator is supplied by the hood and enters the
resonator through an opening located on the resonator volume. The neck mouth
communicates directly with the combustion chamber. Both the resonator neck and
volume are modeled as ducts in which plane acoustic wave propagation occurs.
The relation between acoustic impedances at duct extremities 1 and 2 is given by
Tijdeman28 :
i
ek − e−k + Z 2 ek + e−k
Z 1 = k (15.33)
e + e−k + Z 2 ek − e−k
i +1 γ −1
=i+√ 1+ √ (15.34)
2Sh Pr
and where Sh and Pr are the shear number and Prandtl number, respectively. The
area jump between neck and resonator volume is modeled by forcing the continuity
of p̂ and û A. Furthermore, at the neck ends an additional transfer function must
be considered to account for end resistance and end reactance. The end resistance
is attributable to the area change pressure drop. The end reactance accounts for the
fluid mass inside the combustor that is involved in fluctuations by the air fluctuating
inside the neck. The Helmholtz dampers were also tested in the impedance lab in
atmospheric conditions and showed excellent agreement with theory.29
478 C. O. PASCHEREIT ET AL.
Normalized p̂ Normalized p̂
simulation simulation
engine data engine data
Two different setups were considered in the modeling and were compared with
engine data:
1) Helmholtz dampers designed to suppress two unstable modes. A strong
instability at St ≈ 0.7 was observed.
2) Additional Helmholtz dampers were designed to suppress the third unstable
mode.
Only a limited volume was available to mount the resonators. The additional
damping power was thus limited as well. The design task therefore had the extended
goal of suppressing the instability by the most efficient placement of the damper
elements. The results are displayed in Fig. 15.23. Good agreement was found
between modeling and measurement. The third unstable mode was effectively
suppressed by the additional Helmholtz damper.
VII. Conclusion
A thermoacoustic network analysis method was presented. The network uses
both measured and analytically derived transfer functions of the components in
the combustion system. Transfer functions of burners and flames were obtained
experimentally by forcing the combustion system with loudspeakers. The transfer
function is then obtained from microphone signals by using a cross-correlation
technique. The system is assumed to be linear and time-invariant. The linearity
assumption of the burner and flame acoustics is validated by determining ex-
perimentally the transfer functions at several acoustic pressure levels. The flame
properties were show to be linear in the amplitude range of interest.
The network modeling approach was tested by modeling an atmospheric com-
bustion test facility with measured burner and flame transfer functions. The in-
fluence of changing boundary conditions was then predicted by using this model.
The predicted and measured spectra corresponded very well.
IMPLEMENTATION OF INSTABILITY PREDICTION IN DESIGN 479
References
1
Munjal, M. L., Acoustics of Ducts and Mufflers, John Wiley & Sons, New York, 1986.
2
Polifke, W., Paschereit, C. O., and Sattelmayer, T., “A Universally Applicable Sta-
bility Criterion for Complex Thermo-Acoustic Systems,” VDI-Berichte, 1997, pp. 455–
460.
3
Schuermans, B., Polifke, W., and Paschereit, C. O., “Prediction of Acoustic Pressure
Spectra in Gas Turbines Based on Measured Transfer Matrices,” ASME Turbo Expo ’00,
Munich, Germany, May 2000.
4
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,” Com-
bustion and Flame, Vol. 70, 1987, pp. 281–289.
5
Paschereit, C. O., Schuermans, B., Polifke, W., and Mattson, O., “Measurement of Trans-
fer Matrices and Source Terms of Premixed Flames,” ASME Turbo Expo ’99, Indianapolis,
IN, June 1999.
6
Paschereit, C. O., and Polifke, W., “Investigation of the Thermoacoustic Characteristics
of a Lean Premixed Gas Turbine Burner,” ASME Turbo Expo ’98, Paper 98-GT-582, June
1998.
7
Schuermans, B. B. H., Polifke, W., Paschereit, C. O., and van der Linden, J., “Prediction
of Acoustic Pressure Spectra in Combustion Systems Using Swirl Stabilized Gas Turbine
Burners,” ASME Turbo Expo ’00, Munich, Germany, May 2000.
8
Cremer, L., “The Second Annual Fairy Lecture: The Treatment of Fans as Black Boxes,”
Journal of Sound and Vibration, Vol. 16, 1971, pp. 1–15.
9
Bodén, H. and Åbom, M., “Modelling of Fluid Machines as Sources of Sound in Duct
and Pipe Systems,” Acta Acustica, 1995, pp. 549–560.
10
Lavrentjev, J., and Åbom, M., “Characterization of Fluid Machines as Acoustic Mul-
tiport Sources,” Journal of Sound and Vibration, Vol. 197, 1996, pp. 1–16.
11
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Excitation of Thermoacoustic
Instabilities by the Interaction of Acoustics and Unstable Swirling Flow,” AIAA Journal,
Vol. 38, 2000, pp. 1025–1034.
480 C. O. PASCHEREIT ET AL.
12
Polifke, W., Poncet, A., Paschereit, C. O., and Döbbeling, K., “Reconstruction of Acous-
tic Transfer Matrices by Instationary Computational Fluid Dynamics,” Journal of Sound
and Vibration, Vol. 245, 2001, pp. 483–510.
13
Flohr, P., Paschereit, C. O., and van Roon, B., “Using CFD for Time-Delay Modeling
of Premix Flames,” ASME Turbo Expo ’01, New Orleans, LA, June 2001.
14
Polifke, W., Flohr, P., and Brandt, M., “Modeling of Inhomogeneously Premixed Com-
bustion with an Extended TFC Model,” ASME Turbo Expo ’00, Munich, Germany, May
2000.
15
Flohr, P., and Paschereit, C. O., “Mixing Prediction in Premix Burners Using Industrial
LES Tools,” Symposium on Computational Modeling of Industrial Combustion Systems,
ASME International Mechanical Engineering Congress and Exposition, New Orleans, LA,
Nov. 2002.
16
Dowling, A. P., “The Calculation of Thermoacoustic Oscillations,” Journal of Sound
and Vibration, Vol. 180, 1995, pp. 557–581.
17
Pankiewitz, C., and Sattelmayer, T., “Time Domain Simulation of Combustion Instabil-
ities in Annular Combustors,” ASME Turbo Expo ’02, GT-2002-30063, Amsterdam, The
Netherlands, June 2002.
18
Schuermans, B., and Paschereit, C. O., “Investigation of Thermoacoustic Oscillations
in Combustion Systems using an Acoustic Network Model,” Acoustics of Combustion;
EU-ROTHERM Seminar No. 67, Univ of Twente, Enschede, The Netherlands, July 2000.
19
Zhou, K., and Doyle, C., Essentials of Robust Control, Prentice-Hall, Upper Saddle
River, NJ, 1998.
20
Culick, F., “Combustor Dynamics: Fundamentals, Acoustics and Control,” Active Con-
trol of Engine Dynamics, Von Karman Institute for Fluid Dynamics, Rhode-Saint-Genese,
Belgium, RTO-EN-20, May 2001.
21
Annaswamy, A., Fleifil, M., Rumsey, J., Prasanth, R., Hathout, J., and Ghoniem, A.,
“Thermoacoustic Instability: Modelbased Optimal Control Design and Experimental Vali-
dation,” IEEE Transactions Control Systems Technology, Vol. 8, No. 6, 2000.
22
LMS International, “SYSNOISE Revision 5.4 documentation, Version 1.0,” LMS In-
ternational, Leuven, Belgium, May 1999.
23
Evesque, S., and Polifke, W., “Low-Order Acoustic Modelling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” ASME Turbo Expo ’02, No. GT-2002-30064,
Amsterdam, The Netherlands, June 2002.
24
Rienstra, S., and Hirschberg, A., “An Introduction to Acoustics,” Report IWDE99–02,
TU Eindhoven, 1999.
25
Schuermans, B., Paschereit, C. O., and Polifke, W., “Modeling Transfer Matrices of
Premixed Flames,” ASME Turbo Expo ’99, Indianapolis, IN, June 1999.
26
Polifke, W., Kopitz, J., and Serbanovic, A., “Impact of the Fuel Time Lag Distribution
in Elliptical Premix Nozzles on Combustion Stability,” 7th AIAA/CEAS Aeroacoustics
Conference, Maastricht, The Netherlands, May 2001.
27
Bellucci, V., Paschereit, C. O., Flohr, P., and Schuermans, B., “Thermoacoustic Simu-
lation of Lean Premixed Flames Using an Enhanced Time-Lag Model,” 31th AIAA Fluid
Dynamics Conference (Aeroacoustics Section), Anaheim, CA, June 2001.
28
Tijdeman, H., “On the Propagation of Sound Waves in Cylindrical Ducts,” Journal of
Sound and Vibration, Vol. 39, 1975, pp. 1–33.
29
Bellucci, V., Schuermans, B., Nowak, D., Flohr, P., and Paschereit, C. O., “Thermoa-
coustic Modeling of a Gas Turbine Combustor Equipped with Acoustic Dampers,” ASME
Turbo Expo ’04, No. 2004-GT-53977, Vienna, Austria, June 2004.
Chapter 16
I. Introduction
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights re-
served.
∗ Senior Research Associate.
† Professor of Mechanical Engineering.
481
482 J. G. LEE AND D. A. SANTAVICCA
produces a fluctuating rate of heat release. If the fuel fluctuation arrives at the flame
front in-phase with the pressure fluctuation, the resulting heat-release fluctuation
amplifies the oscillations, whereas, if the fuel fluctuation arrives at the flame front
out-of-phase with the pressure fluctuation, the resulting heat-release fluctuation
damps the oscillations.
Spray–flow interactions refer to several phenomena that act to drive unstable
combustion, including oscillations in droplet atomization and droplet vaporization
and spray–vortex interaction. Any of these phenomena can cause modulation of
the fuel concentration and/or fuel distribution and thereby produce fluctuations in
the rate of heat release.
To understand the role and relative importance of flame–vortex interaction, feed-
system coupling, and spray–flow interactions during unstable combustion, mea-
surements must be made that characterize the mechanisms, the resulting instability,
and the relationship between the two. Of particular importance are measurements
of the fluctuations in pressure, heat release, fuel concentration, and flame structure.
This paper discusses several diagnostic techniques that have been used to make
such measurements. Most of the techniques require optical access and therefore
are limited to use in laboratory-scale combustors or in full-scale single-nozzle
combustor test rigs where optical access is available. The purpose of this paper is
to discuss the application of these techniques to the study of combustion instabil-
ities and, in particular, to demonstrate how these techniques can be used to gain
an improved understanding of the mechanisms of unstable combustion in gas- and
liquid-fueled combustors. To date, these techniques have primarily been used to
study the mechanisms of unstable combustion under limit-cycle conditions, and
very few studies have been done of the mechanisms involved in the transition from
stable to unstable combustion.
a)
2
1
p' (psi)
-1
-2
0.000 0.025 0.050 0.075 0.100
Time (sec)
b)
0
Power spectrum (dBVRMS)
-10
-20
-30
-40
-50
-60
-70
0 200 400 600 800 1000 1200 1400 1600
Frequency (Hz)
Fig. 16.1 Pressure fluctuation during unstable combustion: a) typical time trace
and b) power spectrum.
-200
Phase angle of Pf with respect to Pc
-240
-280
-320
-360
0.55 0.60 0.65 0.70 0.75
Equivalence ratio
Fig. 16.2 Phase difference between fuel-line pressure (P f ) and combustor pressure
(Pc ) fluctuations vs equivalence ratio ( , original length fuel line; , extended length
fuel line).
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 485
0.6
0.5
0.4
Pc,rms (psi)
0.3
0.2
0.1
0.0
0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80
Equivalence ratio
Fig. 16.3 The rms fluctuation in combustor pressure vs equivalence ratio
( , original length fuel line; , extended length fuel line).
10.0
CH*
OH*
6.0
CO2*
4.0
2.0
0.0
300 350 400 450 500 550 600
Wavelength (nm)
this technique has been widely used for measuring both local and overall rates of
heat release in lean premixed combustors under both stable and unstable operating
conditions.
The strongest chemiluminescence emission from lean hydrocarbon flames
comes from CH∗ , OH∗ , and CO∗2 (the asterisk indicates an excited species), whereas
in rich hydrocarbon flames strong chemiluminescence emission also comes from
C∗2 . Fig. 16.4 shows a chemiluminescence emission spectrum measured in the
laboratory-scale optically accessible lean premixed combustor that is illustrated
19mm dia
exit
110mm dia x 375mm
fused silica
combustor section
Fig. 16.5 Schematic drawing of optically accessible lean premixed combustor.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 487
2.0 0.70
Tinlet = 650 K
1.5
0.65
(a.u.)
1.0
0.60
0.55
0.5
0.50
φ = 0.45
0.0
3 4 5 6 7 8 9
8
Vinlet = 67 m/s Tinlet = 673 K
Intensity/ Fuel Flow Rate (a.u.)
0
0.4 0.6 0.8 1
Equivalence Ratio
Fig. 16.7 Overall CO2 chemiluminescence emission divided by fuel flow rate vs equiv-
alence ratio.
50
30 φ = 0.60
φ = 0.55
20
φ = 0.45
10
0
45 65 85 105 125
Combustor Inlet Velocity (m/s)
Fig. 16.8 Overall CO2 chemiluminescence intensity divided by fuel flow rate vs
combustor velocity.
exponential dependence on equivalence ratio.23 For these measurements, the OH∗
chemiluminescence was detected by using a bandpass filter centered at 307 nm with
a full width at half-maximum (FWHM) of 10 nm, and the CH∗ chemiluminescence
was detected with a bandpass filter centered at 430 nm and a FWHM of 10 nm.
In addition to the effects of fuel flow rate and equivalence ratio, some studies
have shown that turbulence reduces the intensity of the overall chemiluminescence
emission.24,27 This effect was not observed in tests conducted in the same lean
premixed combustor discussed previously (Fig. 16.5). These results are shown
in Fig. 16.8, which is a plot of the overall CO∗2 chemiluminescence intensity
divided by the fuel flow rate vs the combustor inlet velocity for constant values of
equivalence ratio at an inlet temperature of 650 K and a pressure of 100 kPa. As
shown, the inlet velocity was increased by a factor of 2, corresponding to a change
in the Reynolds number from 9000 to 18,000, with no apparent decrease in the
overall chemiluminescence intensity.
Several studies involving detailed chemical kinetic calculations of lean premixed
laminar methane–air flames have been conducted to investigate the relationship
between the local rate of heat release, that is, the rate of heat release per unit flame
area, and the local chemiluminescence emission, that is, the rate of chemilumi-
nescence emission per unit flame area.41−44 These studies have shown that CH∗ ,
OH∗ , and CO∗2 occur within the reaction zone, which indicates that the location
of the chemiluminescence emission can be used as an indicator of the location
of the reaction zone. They have also shown that a correlation exists between the
chemiluminescence emission from both OH∗ and CO∗2 and the local rate of heat
release. An exception to both of these results occurs in extreme local strain or
flame curvature, for example, at cusps, where the calculations show that the local
chemiluminescence emission can effectively go to zero without local extinction of
the flame. These studies also indicate that most of the fuel goes through a reaction
path that includes the formation of CO∗2 , suggesting that CO∗2 chemiluminescence
should be a good indicator of the rate of heat release. Last, the studies show that
the local rate of heat release (HRlocal ) and the local chemiluminescence emission
(Ilocal ) are affected by unsteady strain and flame curvature and that they increase
490 J. G. LEE AND D. A. SANTAVICCA
where the exponent α is a positive number and depends on the flame temperature
(as determined by the equivalence ratio, unburned gas temperature, dilution, and
radiation losses) and the effects of unsteady strain and flame curvature.41
To determine the relationship between the overall chemiluminescence emission
(Ioverall ) and the overall rate of heat release (HRoverall ), one must integrate the local
values over the flame area, that is,
Ioverall = Ilocal dAflame and HRoverall = HRlocal dAflame (16.2)
A A
If the flame temperature, that is, the equivalence ratio, unburned gas temperature,
dilution, and radiation losses, is constant and the effects of strain and flame cur-
vature are negligible or constant, then Ilocal , HRlocal , and α are constant over the
flame. These constant values result in the proportionality of the overall chemilu-
minescence emission and the overall rate of heat release, that is,
where the constant C depends on the flame temperature (i.e., equivalence ratio,
unburned gas temperature, dilution, and radiation losses) and the effects of strain
and curvature. This result is consistent with the experimental results presented in
Fig. 16.6, which show that for a fixed equivalence ratio and inlet temperature the
overall chemiluminescence emission increases linearly with fuel flow rate, that
is, the overall rate of heat release, and that the slope depends on the equivalence
ratio. Similarly, the results presented in Fig. 16.7, which show that the overall
chemiluminescence emission divided by the fuel flow rate increases exponentially
with the equivalence ratio, are also predicted by the detailed chemical kinetic
calculations. And last, the fact that the local chemiluminescence emission can be
affected by unsteady strain and flame curvature is consistent with the observations
that turbulence can reduce the overall chemiluminescence emission.
The relationship between the overall chemiluminescence emission and the over-
all rate of heat release is more complicated if the equivalence ratio and/or the effects
of strain and curvature vary over the flame surface. For example, a more compli-
cated relationship would occur in a partially premixed turbulent flame in which the
equivalence ratio is not constant over the flame surface. In this case, the exponent
α, in the equation relating the local chemiluminescence emission to the local rate
of heat release, varies with location on the flame surface, which in turn affects the
relationship between the overall chemiluminescence emission and the overall rate
of heat release. To some extent such variations are likely to average out such that
the relationship between the overall chemiluminescence and the overall rate of heat
release can be expressed in terms of the average equivalence ratio. Data supporting
this are shown in Fig. 16.9, which is a plot of the overall CO∗2 chemiluminescence
emission, for fixed overall equivalence ratio vs a parameter that is referred to as
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 491
φ = 0.64
1.0
φ = 0.58
0.5 φ = 0.51
φ = 0.45
0.0
0 20 40 60 80 100
% Premixed
Fig. 16.9 The effect of incomplete fuel–air mixing on the overall CO∗2 chemilumines-
cence intensity.
release in lean premixed flames, but the studies also clearly indicate that such
measurements should be interpreted with caution. (Another technique that has
been proposed for making quantitative measurements of the rate of heat release is
HCO fluorescence.43,44 This technique is discussed in the Sec. V.)
In studies of unstable combustion in lean premixed combustors, chemilumines-
cence emission has been used by numerous researchers to characterize temporal
fluctuations in both the overall heat release12,20,30,31,35,38−40 and the spatial distri-
bution of the local heat release.7,12,20,30−32,34,35,37,39 As the preceding discussion
indicates, care must be taken when interpreting such measurements. For example,
when making overall chemiluminescence measurements, it is important to realize
that changes in the fuel flow rate and changes in the equivalence ratio indepen-
dently affect the overall chemiluminescence emission, whereas only changes in
the fuel flow rate affect the overall rate of heat release. This point can be illustrated
by considering two combustors in which the equivalence ratio at the inlet to the
combustor is fluctuating but for different reasons. In the first case, the equivalence-
ratio fluctuations are the result of fluctuations in the airflow rate, whereas the fuel
flow rate is constant. Under these conditions, fluctuations will occur in the overall
chemiluminescence emission; however, the overall rate of heat release will be con-
stant. (This is not to be confused with the fact that the local rate of heat release, i.e.,
the local flame speed, changes with the local equivalence ratio. This discrepancy
is compensated for by changes in the flame area such that the overall rate of heat
release remains constant.) In the second case, the equivalence ratio fluctuations are
the result of fluctuations in the fuel flow rate, whereas the airflow rate is constant.
Under these conditions, fluctuations will occur in the overall chemiluminescence
emission, that, in part, will be caused by equivalence ratio fluctuations and, in
part, by fuel flow rate fluctuations. In this situation, the chemiluminescence fluc-
tuations overestimate the fluctuations in the overall rate of heat release. The only
situation where the fluctuation in the overall chemiluminescence can be attributed
solely to fluctuations in the overall rate of heat release is when the equivalence
ratio is constant. In general, such conditions can only be achieved in a labora-
tory combustor, whereas in an actual combustor one would expect some degree of
feed-system coupling and, as a result, fluctuations in the equivalence ratio. Under
such conditions, measurements of the overall chemiluminescence-emission fluc-
tuations without simultaneous measurements of the equivalence ratio fluctuations
can potentially give misleading information about both the amplitude and phase
of the overall heat-release fluctuations.
There are also considerations when using chemiluminescence emission as a
measure of the local rate of heat release. The most obvious consideration is that
the chemiluminescence-emission measurement is a line-of-sight measurement,
that is, one measures the total emission integrated along the line of sight. This
effect can be significantly reduced by using an optical arrangement with a very
short depth of field; however, this reduction is at the expense of significantly re-
duced signal strength.37 Another approach, if the flame is axisymmetric, is to
use a deconvolution technique to reconstruct the two-dimensional emission field
from line-of-sight chemiluminescence images. (This approach is discussed and
illustrated later in this section) It is important to realize that the two-dimensional
chemiluminescence images obtained in this manner do not actually represent the
local chemiluminescence intensity, that is, on the scale of the flame thickness.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 493
where Ālocal is the flame area within a volume defined by the resolution of the
line-of-sight measurement.
Similarly, one can define a two-dimensional rate of heat release (HR2D ), which
represents the product of the local rate of heat release and the local flame area,
that is, HR2D = HRlocal · Ālocal . Of interest is the relationship between the two-
dimensional chemiluminescence emission and the two-dimensional rate of heat
release. Use of the preceding equations gives I2D = (Ilocal /HRlocal ) HR2D . Using
the power-law relationship between the local chemiluminescence emission and the
local rate of heat release, discussed previously, gives the following relationship:
where C2D depends on the local flame temperature and any factors that affect
the flame temperature. Therefore, the intensity of the chemiluminescence emis-
sion shown in the two-dimensional chemiluminescence images is indicative of the
two-dimensional rate of heat release; however, it can also change independently
of the rate of heat release as a result of changes in the flame temperature and
any factors that affect the flame temperature. As with overall chemiluminescence
measurements, the most likely concern would be in a partially premixed flame in
which variations in the local equivalence ratio could lead to an inaccurate estimate
of the local rate of heat release.
An example of a measurement of the overall heat-release fluctuations during un-
stable combustion made in the optically accessible lean premixed combustor illus-
trated in Fig. 16.5 with CO∗2 chemiluminescence emission is shown in Fig. 16.10a.
The simultaneously recorded pressure fluctuation is shown in Fig. 16.10b. The
overall CO∗2 chemiluminescence emission plotted in Fig. 16.10a was measured
by imaging the entire flame onto a photomultiplier tube through an appropriate
bandpass filter as described previously. Care must be taken when making such
measurements to collect the chemiluminescence emission from the entire flame
so as to obtain an accurate indication of the total heat-release rate and to avoid
erroneous fluctuations caused by the flame moving in and out of the field of view.
Simultaneous measurements of the overall heat-release rate and the pressure, such
as shown in Fig. 16.10a and b, can be used to determine the phase difference be-
tween the heat-release and pressure fluctuations that is related to the overall system
damping and gain characteristics.3 Such measurements also provide information
on how the flame’s heat release responds to pressure fluctuations. For example,
Fig. 16.11 shows a plot of the rms overall heat-release fluctuation normalized
by the mean overall heat release (measured by using CO∗2 chemiluminescence)
vs the rms combustor pressure fluctuation during unstable combustion. These
494 J. G. LEE AND D. A. SANTAVICCA
a)
0.8
Overall Heat Release (a.u.)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0.000 0.025 0.050 0.075
Time (sec)
b)
2
1
p´ (psi)
-1
-2
0.000 0.025 0.050 0.075
Time (sec)
Fig. 16.10 Simultaneous measurement of a) the overall heat-release fluctuations and
b) the combustor pressure fluctuations during unstable combustion.
35
30
25
qrms/qrms (%)
20
15
10
0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Prms(psi)
Fig. 16.11 The rms heat-release fluctuation normalized by the average heat release vs
rms pressure fluctuation at various unstable operating conditions in a lean premixed
combustor.
Low High
a) b)
30 individual images at that same phase angle are averaged to obtain the phase-
averaged image shown in Fig. 16.12a. This image is a record of the line-of-sight
integrated chemiluminescence intensity and therefore does not reveal the cross-
sectional structure of the flame. If the flame is assumed to be axisymmetric, one
can use a deconvolution procedure to reconstruct the two-dimensional flame struc-
ture, including “onion-peeling,” Abel transformation, and filtered backprojection
methods.47 The line-of-sight image shown in Fig. 16.12a was processed with
an Abel deconvolution procedure. The resulting image, which is shown in Fig.
16.12b, reveals the two-dimensional structure of the flame that was not apparent
in the original line-of-sight image.
A basic assumption of the deconvolution procedure is that the image is ax-
isymmetric. Because the line-of-sight image in Fig. 16.12a is not perfectly ax-
isymmetric, the upper and lower halves of the image were averaged to create an
axisymmetric image before applying the Abel inversion. This procedure of cre-
ating an axisymmetric line-of-sight image is usually necessary, and care must be
taken when interpreting the resulting reconstructed images. If the line-of-sight
images are reasonably axisymmetric, the insights gained from the reconstructed
two-dimensional images usually outweigh the uncertainty associated with the ax-
isymmetric approximation. Unfortunately there is no way to quantify this tradeoff;
therefore, the reconstructed two-dimensional images must always be interpreted
with care.
Figure 16.13 shows a sequence of 12 phase-averaged two-dimensional CO∗2
chemiluminescence images recorded in increments of 30 phase-angle degrees dur-
ing one period of a 235-Hz instability in the same laboratory-scale lean premixed
dump combustor mentioned previously. In this case the combustor was operating
at 100 kPa, with an inlet temperature of 673 K, an inlet velocity of 45 m/s, and an
equivalence ratio of 0.45. In addition, the exit of the combustor was not restricted,
Low High
#1 #2 #3 #4
Image Number
1 3 5 7 9 11 13
Pressure (psi)
1.0
0.5
#5 0.0 #6
-0.5
-1.0
0.0 1.0 2.0 3.0 4.0
#7 Time (msec) #8
Fig. 16.13 Flame-structure evolution during one period of unstable combustion with
a frequency of 235 Hz.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 497
as it is in Fig. 16.5; hence, a marked change occurred in the frequency of the insta-
bility compared with the previously presented results. Note that only the upper half
of the flame is shown, because the reconstructed images are axisymmetric. Note
also, to the left of each image, a line drawing illustrates the location of the image
relative to the combustor. Figure 16.13 also contains a plot of the combustor pres-
sure vs time, measured at the dump plane, over one period of the instability with
markers indicating when each of the images was recorded. The two-dimensional
flame-structure image sequence reveals the temporal evolution of the flame struc-
ture during the instability and provides insight regarding the phenomenology of
the instability. For example, the images in Fig. 16.13 show a flame that is anchored
on the centerbody and extends outward into the recirculation zone and all the way
to the wall of the combustor. The overall flame shape remains very nearly the same
during the instability but there is a noticeable change in the overall intensity of the
flame’s heat release, indicating that minimum heat release, that is, images 9, 10
and 11, occurs when the pressure is minimal. There is also a periodic break in the
flame between where it is attached to the centerbody and the recirculation zone,
which also occurs when the pressure and overall heat release are at their minimum
levels.
Two flame-structure image sequences are shown in Fig. 16.14a and 16.14b,
which correspond to instabilities in the same lean premixed combustor (Fig. 16.5),
at the same operating conditions (Tinlet = 623 K, Vinlet = 59 m/s, p = 100 kPa, and
φ = 0.58), but with different inlet fuel distributions. In Fig. 16.14a, the fuel and
air are completely mixed before entering the combustor, whereas in Fig. 16.14b,
although the overall equivalence ratio is the same, the equivalence ratio increases
with increasing radius across the annular outlet of the mixing section. Again,
only the upper half of the image is shown because the reconstructed images are
axisymmetric. In both cases the instability frequency is approximately 350 Hz. The
images are phase synchronized relative to the pressure oscillation at the combustor
entrance and are acquired in increments of 24 phase-angle degrees, giving a total
of 15 images within one period, in which each image is an average of 30 individual
images acquired at a given phase angle. Both cases show evidence of flame–vortex
interaction, but the details of the interaction are noticeably different in the two
cases. The flame in Fig. 16.14a appears to be wrapped around the vortex, which
results in stretching and contraction of the flame zone, whereas the flame in Fig.
16.14b appears to be contained within the vortex, exhibiting periodic extinction
and reignition of the entire reaction zone.
Two-dimensional flame-structure images, such as those shown in Figs. 16.13 and
16.14, reveal the location and intensity of the flame’s heat release and its temporal
evolution during one period of the instability. Combining this information with the
measured pressure fluctuation, one can calculate the Rayleigh index distribution,
R(x, y), which is given by the following equation:
1
R(x, y) = p (x, y, t) q (x, y, t) dt (16.6)
T π
a)
#1 #2 #3 #4 #5
#6 #7 #8 #9 #10
b)
#1 #2 #3 #4 #5
#6 #7 #8 #9 #10
Fig. 16.14 Flame-structure evolution during one period of unstable combustion with
a frequency of 350 Hz at two different operating conditions in the same lean premixed
combustor.
a)
400
0
-200 200
400
0 600
b)
0 1000
800
600
400
200
Fig. 16.15 Rayleigh index distributions corresponding to the unstable flames shown
in Fig. 16.14a and 16.14b, respectively (white, positive Rayleigh index; gray, negative
Rayleigh index).
behind the centerbody and the other in the recirculation zone downstream of the
dump plane; whereas a larger region of positive Rayleigh index is located along
the shear layer between the dump plane and the centerbody recirculation zones.
Figure 16.15b shows a significantly different Rayleigh index distribution with a
large region of positive Rayleigh index centered in the dump-plane recirculation
zone and a smaller damping region immediately downstream of the dump plane. A
comparison of the locations of minimum and maximum Rayleigh index, that is, the
locations of damping and gain, with the flame-image sequence can provide insight
as to the phenomenology of the instability and the role of the instability-driving
mechanisms.
Chemiluminescence-emission measurements have also been shown to provide
information and insights that can be used to optimize active combustion-control
systems employing modulated secondary fuel injection for the suppression of un-
stable combustion. For example, it has been shown that effective suppression of an
instability can be achieved with less secondary fuel if the fuel is injected into the
region of maximum damping indicated in the Rayleigh index distribution.48 Chemi-
luminescence imaging can also be used to determine the optimum phase delay in
active combustion-control systems employing secondary fuel flow modulation.48
When a pulse of secondary fuel is injected into a combustor it produces a detectable
500 J. G. LEE AND D. A. SANTAVICCA
where T is the period of the secondary fuel flow modulation, to is the time delay
between the pressure signal zero crossing and the secondary fuel valve trigger
signal, p is the measured pressure signal, and qsecondary is the flame-response
function. An example of the flame-response function is shown in Fig. 16.16a.48
In this case, subharmonic secondary injection is used where the frequency of
secondary fuel injection is one-fourth that of the instability. This is illustrated in
Fig. 16.16b, in which the flame-response function is shown along with the pressure
oscillation. Also shown in Fig. 16.16b is the control signal to the secondary fuel-
control valve and the time delay between that signal and the zero crossing of the
pressure signal. The flame-response Rayleigh index is a measure of the effect of the
heat release caused by the secondary fuel on the instability. If its value is positive,
the secondary fuel acts to amplify the instability, whereas if it has a negative value,
it acts to damp the instability, where the optimum phase delay corresponds to the
case of maximum damping. The flame-response Rayleigh index as a function of
the time delay to , is plotted in Fig. 16.16c for the flame-response function and
pressure oscillation shown in Fig. 16.16b. According to this result, the time delay
between the zero crossing of the pressure oscillation and the valve trigger signal
for maximum damping is approximately 1.25 ms. This compares reasonably well
with the experimentally determined delay time for maximum suppression of this
instability, which is approximately 1 ms.
I l
= 10− o ε cdx (16.8)
Io
a) 0.040
chemiluminescence)
0.035
0.030
q´ (CO2*
0.025
0.020
0.015
0.010
0.005
0.000
0 1 2 3 4 5 6 7 8 9 10 11
Time (msec)
b) 0.03
time delay (to) q´secondary
Heat release (a.u.)
0.01
0
Period (T)
-0.01
-0.02
0 2 4 6 8 10 12
Time (msec)
c) 0.010
0.005
Global Response
Rayleigh Index
0.000
-0.005
-0.010
-0.015
-0.020
0 1 2 3 4
time delay (msec)
Fig. 16.16 a) Flame-response function, b) flame-response function superimposed
on the unstable pressure trace, and c) flame-response Rayleigh index vs time-delay
prediction.
502 J. G. LEE AND D. A. SANTAVICCA
a) 0.50
0.40
0.30
I/Io
0.20
0.10
0.00
250 300 350 400 450 500 550 600 650 700
Temperature (K)
b) 1
0.1
I/Io
0.01
T = 295 K
T = 473 K
T = 573 K
0.001 T = 673 K
that is, the measured attenuation is the result of the integrated absorption over the
entire beam path and, therefore, is a measure of the average flowfield properties
along the beam path.
The laser-absorption measurement that has proven most valuable in the study of
combustion dynamics is an infrared-absorption measurement of hydrocarbon fuel
concentration based on the fortuitous matchup between the 3.39-µm wavelength
of the infrared helium-neon laser and a vibrational-rotational energy level transi-
tion in hydrocarbon molecules.49,51 In this case, the absorbing molecule is a stable
species, therefore, the simplest procedure for making quantitative measurements of
concentration is to empirically determine the pressure and temperature dependence
of the absorption coefficient. An example of this empirical determination for
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 503
methane is shown in Fig. 16.17a and 16.17b, in which plots of the normalized
transmittance (I /Io ) vs temperature at a fixed pressure and vs pressure for values
of constant temperature, respectively, are shown.12 These measurements were
made in a flow cell with a homogeneous methane–air mixture at known conditions.
The 3.39-µm output from a He-Ne laser (3 mW) was used as a light source and
the transmitted light was detected by using a thermoelectrically cooled indium-
arsenide (InAs) detector. Once the normalized transmittance is known, the decadic
molar absorption coefficient ε can be obtained from Beer–Lambert’s Law, that is,
1 I
ε = − log10 (16.9)
cl Io
By using the data shown in Fig. 16.17, the following expression for the decadic
molar absorption coefficient for methane as a function of pressure and temperature
is obtained12 :
Po T P
ε = 84737 C1 + C2 −1 (16.10)
P 293 K Po
with
C1 = −0.1131 + 1.1875 (293 K/T )
and
a) Data Adapters
Acquisition InAs Detector
System
InAs Detector
HeNe Laser
(3.39 µm) Data Acquisition
System
Center bluff-body
b) 7 mm
Diffuser
Measurement Location
Preheated Air
at a pressure of 110 kPa, with an inlet temperature of 658 K and an inlet velocity of
100 m/s. It is assumed that the temperature of the mixture in the nozzle is constant
and, therefore, only fluctuations in the pressure are accounted for when converting
the measured transmittance to equivalence ratio. This result clearly shows that
feed-system coupling is playing a significant role in this instability, resulting in
peak-to-peak fluctuations in the equivalence ratio of approximately ±0.05 about a
mean of 0.65. The frequency spectrum corresponding to these fluctuations is shown
in Fig. 16.19d. The equivalence-ratio fluctuations show a dominant frequency
at 465 Hz, with weaker oscillations appearing at the higher harmonics. Shown
in Fig. 16.19b and 16.19c are the phase-synchronized pressure and heat-release
measurements for this instability, and shown in Fig. 16.19e and 16.19f are the
corresponding frequency spectra. The dominant frequencies of the equivalence-
ratio, pressure, and heat-release fluctuations are clearly the same, that is, 465 Hz,
although there are significant differences in the relative magnitude of the second
harmonic oscillations. Most noticeable is the 10-dB difference between the first and
second harmonics of the equivalence-ratio fluctuation vs 23-dB difference between
the first and second harmonics of the heat-release fluctuations. This suggests that
the flame, which is spatially distributed, acts to average out the higher-frequency
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 505
a) d)
0.75 -10
Equivalence
0.70 -20
φ)
dBVrms (φ
Ratio
0.65 -30
-40
0.60
-50
0.55 -60
b) e)
1 -10
dBVrms (Pc)
Pc (psi)
-30
0 -50
-70
-1 -90
c) f)
0.40 -10
dBVrms (q)
0.35
0.30 -30
q
0.25 -50
0.20 -70
0.15
0.0 0.2 0.4 0.6 0.8 1.0 0 400 800 1200 1600
t/T Frequency (Hz)
Fig. 16.19 a–c) Time traces and d–f) power spectra of equivalence ratio φ, combustor
pressure Pc , and heat release q during one period T of unstable combustion.
fluctuations in the equivalence ratio. Also note that the relative magnitudes of the
pressure, equivalence-ratio, and heat-release fluctuations for this instability are
3%, 5%, and 23% of the mean, respectively.
Simultaneous pressure, heat-release, and equivalence-ratio fluctuation measure-
ments can be used to determine the phase delay or time lag between these processes.
Of particular interest is the time lag between the equivalence-ratio fluctuation and
the heat-release fluctuation, because it is important in assessing the role of feed-
system coupling. To estimate whether the equivalence-ratio fluctuation produced
by feed-system coupling arrives at the flame front in-phase with the heat-release
fluctuation, one must estimate the convection time between the fuel-injection loca-
tion and the flame front. The most difficult part of the convection time to estimate
is the time required for the fuel to travel from the entrance of the combustor to
the flame front where the fuel burns. The phase delay or the time lag between the
equivalence ratio and the heat-release fluctuation shown in Fig. 16.19a and 16.19c
is a direct measurement of that quantity.
The 3.39-µm helium-neon laser-absorption technique has also been imple-
mented in a fiber-optic probe52,57 and in a fast-response extraction probe.59 The
main advantage of this approach is that spatially resolved measurements are pos-
sible, that is, with a spatial resolution on the order of 1 mm. Such probes have
been successfully used to measure spatial fuel distributions and equivalence-ratio
506 J. G. LEE AND D. A. SANTAVICCA
where E is the laser irradiance (J/cm2 ), h is Planck’s constant, ν is the laser fre-
quency, χabs is the molar concentration of the fluorescence species, σ (λ, T ) is the
molecular absorption cross section of the fluorescence species, and φ(λ, P, T ) is
the fluorescence quantum yield. In this equation, the effect of gas composition on
the fluorescence yield is assumed to be constant. For a fixed excitation wavelength
and under isothermal and isobaric conditions, the fluorescence signal is only pro-
portional to the mole fraction of the fluorescence seed. However, in flows in which
the temperature and pressure are changing, the effect of the energy-level popula-
tion distribution and of collisional quenching and redistribution on the absorption
cross section and the fluorescence quantum yield, and therefore the fluorescence
signal, must be accounted for, as discussed before.
Laser-induced fluorescence can be used to make point measurements with sub-
millimeter spatial resolution or to make two-dimensional measurements, also re-
ferred to as planar laser-induced fluorescence (PLIF) measurements.67,68 In making
point measurements, the laser beam is focused with a spherical lens to a small-
diameter beam waist and the fluorescence signal is detected by imaging the beam
waist through an aperture onto a photomultiplier tube. In making two-dimensional
measurements, the laser beam passes through a combination of cylindrical and
spherical lenses to produce a thin laser sheet and the resulting fluorescence signal
is detected by imaging a portion of the laser sheet onto an intensified CCD camera.
In both cases, an appropriate interference filter is used to isolate and selectively
detect the desired fluorescence wavelength. The strength of the fluorescence sig-
nal for gas-phase fluorescence measurements is typically very low, requiring the
use of high-power pulsed lasers, where depending on the fluorescence species and
its concentration, single-pulse measurements are often possible. Unfortunately,
high-power pulsed lasers operate at relatively low pulse rates, that is, typically
10–20 Hz; therefore, this technique does not provide a continuous measure of the
fluorescence species concentration. Although periodic phenomena, such as unsta-
ble combustion, in which the measurements can be phase synchronized with the
instability and the periodic behavior reconstructed, present a special case.
A useful application of laser-induced fluorescence in the study of combustion
dynamics is in the characterization of fuel–air mixing. Measurement of fuel–air
mixing is important because both the temporal and spatial fuel distribution can have
a significant effect on the stability characteristics of the combustor.32 Many fuels
of interest, however, are not well suited for fluorescence measurements, and so a
common approach is to seed the fuel with a small amount of a fluorescence seed, in
which the concentration of the fluorescence seed, as measured by the fluorescence
technique, is assumed to be an indicator of the fuel concentration. An advantage
of using a fluorescence seed is that its concentration can be controlled. In addition,
a fluorescence seed can be selected that has optimum physical and spectroscopic
properties, making quantitative equivalence ratio measurements possible. Various
508 J. G. LEE AND D. A. SANTAVICCA
species have been used as fluorescence seeds for fuel–air-mixing studies and de-
tailed information on their fluorescence characteristics can be found in Refs. 69–74.
Several factors must be considered when selecting a fluorescence seed to charac-
terize mixing, including boiling point, autoignition temperature, absorption and
fluorescence characteristics, mass diffusion coefficient, cost, and toxicity.
For gaseous fuels, acetone is commonly used as a fluorescence seed because
of its low-boiling point (50◦ C at 1 atm) and high-vapor pressure (184 torr at
20◦ C), which allows for easy seeding and high-seed density. The spectroscopic
characteristics of acetone are also well known. It absorbs over a broad range of
wavelengths (225–320 nm) with maximum absorption between 270 and 280 nm.
The fluorescence emission is broadband in the blue (350–550 nm) and short lived
(τ ≈ 4 ns) with a fluorescence efficiency of 0.2%.71−74
Because acetone is a stable species, the effects of pressure and temperature
on the absorption cross section and the fluorescence yield are best accounted
for empirically. Figures 16.20a and 16.20b show the effect of temperature and
pressure, respectively, on the acetone fluorescence signal with 266-nm excita-
tion, that is, the fourth harmonic output of a Nd:YAG laser. Each panel shows
two curves. The filled circles are the actual measurements, and the filled squares
have been corrected to constant number density. The measurements were made in
a flow cell with the volume fraction of acetone fixed at 1%; therefore, changes
in pressure and temperature also affect the fluorescence signal as a result of
changes in the density. The actual measurements show that the fluorescence sig-
nal decreases with increasing temperature and increases with increasing pressure.
After correcting for changes in density, however, the fluorescence signal is in-
dependent of pressure, but it decreases with increasing temperature. Knowledge
of the effect of pressure and temperature on the fluorescence signal, indepen-
dent of their effect on density, is critical to the proper interpretation of fluores-
cence measurements. The temperature and pressure dependence of acetone-laser-
induced fluorescence with different excitation wavelengths can be found in the
Refs. 71–74.
Acetone PLIF was used to measure the fuel distribution in the natural-gas-fueled
laboratory-scale dump combustor that is shown schematically in Fig. 16.21a.75
This particular combustor was used in a study of the effect of combustor-inlet
fuel distribution on combustion stability and emissions and, hence, has the unique
capability of allowing for systematic variation of the fuel distribution. Systematic
variation is accomplished by injecting the fuel at one or more of three injection
locations, labeled (1), (2), and (3) in Fig. 16.21a. For the acetone-fluorescence
measurements, the fuel is replaced with air to which 0.5%, by volume, of acetone
has been added. (The measurements are made without combustion.) To ensure
complete vaporization of the acetone, the acetone is injected into the air by using a
spray nozzle and the air is preheated to 100◦ C. Note also that in injection locations
(2) and (3), the flow rate of the simulated fuel was set to match the momentum
flux of the actual fuel jet to properly simulate the mixing characteristics.
The excitation source is the fourth harmonic (266 nm) output of a pulsed
Nd:YAG laser, with a laser pulse energy and duration of 40 mJ/pulse and 7 ns,
respectively. The laser beam is formed into a 0.5-mm-thick by 40-mm-high sheet
that is positioned approximately 1 mm downstream of the dump plane across the
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 509
a) 2.5 800
2.0
600
Sfl • T
1.5
Sfl (•• )
400
1.0
200
0.5
0.0 0
300 400 500 600 700
Temperature (K)
b) 1.5 0.3
1.0 0.2
)
Sfl (•• )
Sfl/P (
0.5 0.1
0.0 0.0
0 2 4 6 8 10 12
P (atm)
Fig. 16.20 Acetone LIF signal with 266-nm excitation: a) temperature dependence
and b) pressure dependence.
exit of the annular mixing section. The fluorescence signal is recorded by using
an intensified CCD camera positioned downstream of the combustor and perpen-
dicular to the laser sheet. Subtraction of background noise and a uniform field
correction are applied to each of the acetone PLIF images. In addition, the images
are corrected for pulse-to-pulse fluctuations in laser energy. Figure 16.21b shows
the processed images for four different fuel distributions, in which only the fuel
distribution across the annular mixing section is shown and the equivalence-ratio
values are indicated by the accompanying gray scale. These results are averages
of 30 individual images and therefore represent the average fuel distribution. In
all four cases the overall equivalence ratio is the same, that is, 0.7, as are the
510 J. G. LEE AND D. A. SANTAVICCA
Center bluff-body
(2)
Laser sheet
Air
(1) Premixed injection (PM)
(1) (2) Center body injection (CB)
(3) Downstream injection (DS)
b)
Equivalence Ratio Scale
PM CB 50% CB/50% DS DS
c) 1.05
0.95 PM
Equivalence Ratio
0.85 DS
0.75
0.65
50CB/50DS
0.55
0.45
0.35 CB
Vinlet = 50 m/s, T = 373K, φavg = 0.70
0.25
0.00 0.04 0.08 0.12 0.16 0.20 0.24 0.28 0.32 0.36
Radial Distance (in)
Fig. 16.21 a) Side view of the optically accessible axial dump combustor and
schematic diagram of PLIF setup; b) processed acetone PLIF images; and
c) equivalence-ratio distribution over the annular mixing section at the inlet of the
combustor.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 511
combustor-inlet velocity (50 m/s), inlet temperature (373 K), and pressure (100
kPa). In the case labeled PM, the fuel and air are premixed well upstream of the
combustor. In this case the fuel distribution is expected to be perfectly uniform.
In the case labeled CB, all the fuel is injected through holes in the centerbody
at a location approximately 25 mm upstream of the dump plane, that is, location
(2). In this case the fuel penetrates to the outer wall of the mixing section, which
results in fuel-rich conditions along the outer wall and fuel-lean conditions along
the centerbody. In the case labeled DS, all the fuel is injected through holes in
the outer wall of the mixing section at a location approximately 25 mm upstream
of the dump plane, that is, location (3). In this case the fuel penetrates to the
centerbody, which results in fuel-rich conditions along the centerbody and fuel-
lean conditions along the outer wall of the mixing section. Note also the evidence
of the effect of the six-vane swirler on the fuel distribution in the “DS” image
shown in Fig. 16.21b. Last, in the case labeled 50%CB/50%DS, half of the fuel
is injected through the holes located in the centerbody, location (2), and half is
injected through the holes in the outer wall of the mixing section, location (3). The
resulting fuel distribution is very uniform. These results are further quantified by
calculating an average radial fuel distribution for each case. This calculation is an
average of the radial fuel distribution over 12 radial profiles spaced 30 deg apart
around the axis of the combustor. These results are presented in Fig. 16.21c.
Figure 16.22 shows the CO∗2 chemiluminescence flame-structure images cor-
responding to the PM, CB, and DS fuel distributions shown in Fig. 16.21. Note
that the velocity, temperature, and overall equivalence ratio were the same for all
three cases and the combustor was stable at these conditions. In the PM case the
flame is anchored on the centerbody and extends outward all the way to the wall of
combustor. For the CB case, a noticeable shift occurs in the most intense region of
the flame toward the outer wall of the combustor, which is consistent with the fact
that the fuel concentration is greatest away from the centerbody. For the DS case
the most intense region of the flame has moved closer to the centerbody, where
the fuel concentration is greatest. The fuel distribution also had an effect on the
stability characteristics. With premixed injection (PM) combustion was stable at
low-inlet velocities but became unstable as the inlet velocity increased, whereas
with centerbody (CB) injection the reverse was true, that is, combustion became
unstable as the velocity was decreased. Because velocity has little effect on the fuel
PM CB DS
Fig. 16.22 The effect of inlet fuel distribution on flame structure for the PM, CB, and
DS fuel distributions shown in Fig. 16.21.
512 J. G. LEE AND D. A. SANTAVICCA
a)
b)
laminar-flame regime.82 Last, there is clear evidence of the interaction between the
flame front and the vortex that is shed from the dump-plane shoulder. The effect of
this interaction on the area of the flame, because the flame area is directly related
to the flame’s rate of heat release, is of particular interest.
The first step in calculating the flame area is to determine the location of the
flame front. This determination involves correcting the images for shot-to-shot
laser-energy variations, background noise, and laser-sheet nonuniformity. Because
there is a marked increase in OH concentration at the leading edge of the flame
front, its location can be readily determined by applying a threshold to the corrected
image. In this case, the threshold was not based on the magnitude of the OH
fluorescence intensity but rather on the magnitude of the local gradient of the OH
intensity.79 This magnitude was used to avoid mistakenly identifying as a flame
front the boundary between combustion products and unburned reactants that result
from the mixing of products and reactants in the recirculation zone of the dump
combustor. The flame front determined by this threshold procedure from the OH
PLIF image shown in Fig. 16.23a is shown in Fig. 16.23b. Once the flame front is
determined, the total flame area is calculated by revolving the flame front around
the centerline of the combustor. This calculation assumes that the flame front
determined from the OH PLIF image is representative of the flame front at other
cross sections of the flame. In addition, because the OH PLIF image is a single-shot
image, one must calculate the area for many such images from which an average
flame area can be determined.
To understand the role of flame-area changes during unstable combustion it
is necessary to obtain OH PLIF images at various times during one period of
the instability. The image-acquisition rate of the OH PLIF technique, however,
is considerably less than typical instability frequencies. Therefore, it is necessary
to reconstruct the image sequence by obtaining images at different phase angles
from different cycles. Figure 16.24 shows a sequence of flame fronts determined
from single-shot OH PLIF images over one period of a 378-Hz instability. These
measurements were made in the laboratory combustor referred to previously, op-
erating on natural gas at an equivalence ratio of 0.9, an inlet velocity of 59 m/s,
an inlet temperature of 623 K, and a pressure of 100 kPa. These results clearly
show the evolution of the interaction between the flame front and the vortex and
the resulting changes in the flame length. Five single-shot OH PLIF images were
acquired at each phase angle, the flame area was then calculated for each image
by using the procedure described earlier, and an average flame area at each phase
angle was calculated. The resulting flame area vs phase angle over one period of
the instability is plotted in Fig. 16.25, along with the measured overall heat-release
fluctuation, that is, the overall CO∗2 chemiluminescence-intensity fluctuation. In
this case the area and heat-release fluctuations are very nearly in phase, indicat-
ing that flame-area changes caused by flame-vortex interactions play an important
role in this instability. Results such as these provide valuable insight regarding the
phenomenology of unstable combustion and can be used to provide guidance for
the development of reduced-order models of unstable combustion.46
Detailed two-dimensional OH flame-structure measurements can also be used
to calculate the local flame-surface density, which is a measure of the local reac-
tion rate in turbulent flames. Such measurements have been made in a lean pre-
mixed combustor under stable and unstable operating conditions and have been in
516 J. G. LEE AND D. A. SANTAVICCA
7 Heat release
24
ea Area
Ar 22
Flame area (sq. inches)
6
se
20
Heat release (a.u.)
ea
He
5
el
18
tr
at
ea
re
16
H
le
ea
4
as
14
Ar
e
12
3
10
8
2
0 0.5 1.0 1.5 2.0 2.5
Time (msec)
Fig. 16.25 Variation of flame area and heat release with time during one period of
unstable combustion.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 517
t=8.0 ms t=9.0 ms
Fig. 16.28 Phase-averaged sequence of Mie-scattering image from a modulated jet
in crossflow (period of modulation, 10 ms).
520 J. G. LEE AND D. A. SANTAVICCA
in a spray that is modulated by fuel flow rate modulation, the system input might
be the mass flow rate of fuel at the exit of the injector as a function of time. In either
case, the system output could be the amount of fuel at the inlet to the combustor
as a function of time, that is, the time of arrival of the fuel at the combustor inlet.88
An example of a measurement of the fuel flow rate at the exit of an injector as a
function of time is shown in Fig. 16.29. This is a plot of the volume fraction of liquid
fuel exiting the injector as a function of time during one modulation period.88 In
this case the fuel flow rate was modulated at a frequency of 80 Hz with a duty cycle
of 50% by using an automotive fuel injector located upstream of the 0.27-mm-diam
injector. This measurement was made by using a rotating patternator technique.88
The output of the system in response to this input is the amount of fuel entering the
combustor as a function of time, which is measured by using a time-of-arrival Mie-
scattering technique, which is illustrated schematically in Fig. 16.30. As shown, a
laser sheet is positioned perpendicular to the crossflow, at a position corresponding
to the entrance to the combustor, and the Mie scattering from drops, as they pass
through the laser sheet, is detected by a photomultiplier tube (PMT) positioned 30
deg from the plane of the laser sheet. Time-of-arrival Mie-scattering measurements
corresponding to the system input shown in Fig. 16.29 are shown in Fig. 16.31.
The only difference between the operating conditions for the results shown in Fig.
16.31 is in the crossflow velocity, which is 50 m/s in the lower plot and 90 m/s
in the upper plot. The pronounced difference between these two results can be
explained in terms of improved atomization, that is, smaller droplet size, which
occurs with increased crossflow velocity, that is, an increased Weber number. For
example, smaller droplet size corresponds to a narrower droplet-size distribution,
as a result of which there is less dispersion of the droplets in the 90 m/s case, that
is, the output (Fig. 16.31) retains the shape of the input (Fig. 16.29). A detailed
0.14
0.12
0.10
Vliq 0.08
∑ V liq 0.06
0.04
0.02
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
t / Tinj
Fig. 16.29 The volume fraction of liquid exiting the injector as a function of time
during one modulation period for an 80-Hz modulation with a 50% duty cycle.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 521
TEST SECTION
CROSSFLOW
LASER
PMT
0.70
Mie Intensity (Volts)
0.60 Vcrossflow= 90 m/s
0.50
0.40
0.30
0.20
0.10
0.00
0.70
Mie Intensity (Volts)
0.50
0.40
0.30
0.20
0.10
0.00
0.000 0.005 0.010 0.015 0.020 0.025 0.030
Time (s)
Fig. 16.31 Time-of-arrival Mie-scattering measurements for a crossflow velocity of
90 m/s (upper) and 50 m/s (lower).
point measurement, it is very time consuming to map an entire spray. There have
been many applications of PDPA to gas-turbine fuel sprays operating at steady
conditions, wherein the detailed spray characteristics as a function of operating
conditions have been determined.91 Such information is useful in studies of com-
bustion instabilities in that the spray characteristics determine the fuel distribution
in the combustor, which has a significant effect on the stability characteristics of
the combustor. However, no PDPA measurements of droplet size in fuel sprays
that are modulated have been reported, either for the purpose of active combus-
tion control or because of interaction with an acoustic field, for example, during
unstable combustion.
VIII. Conclusion
Chemiluminescence emission, infrared absorption, laser-induced fluorescence,
laser Mie scattering, and PDPA measurement techniques, in particular, when com-
bined with phase-synchronized pressure-fluctuation measurements, can be used
to obtain a detailed characterization of unstable combustion and the underlying
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 523
Acknowledgments
We thank the many graduate and postdoctoral students who contributed to the
work presented in this chapter, including R. Bandaru, S. Berksoy, J. M. Deepe, E.
Gonzalez, K. Kim, S. Miller, L. Preston, J. Samperio, D. Simons, J. Stenzler, and
K. K. Venkataraman. We are also grateful for the financial support provided by the
Advanced Gas Turbine Systems Research Program of the Department of En-
ergy, the Air Force Office of Scientific Research, the Office of Naval Research,
NASA Glenn Research Center, General Electric, Pratt & Whitney, Siemens-
Westinghouse, Solar Turbines, and United Technologies Research Center.
References
1
Lieuwen, T., and McManus, K. (eds.), “Combustion Dynamics in Lean-premixed Pre-
vaporized (LPP) Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5, 2003,
pp. 721–829.
2
Yang, V., and Anderson, W. E. (eds.), Liquid Rocket Engine Combustion Instability,
Progress in Astronautics and Aeronautics, Vol. 169, AIAA, Washington, DC, 1992.
3
Oran, E. S., and Gardner, J. H., “Chemical-Acoustic Interactions in Combustion
Systems,” Progress in Energy and Combustion Science, Vol. 11, No. 4, 1985, pp. 253–
276.
4
Candel, S. M., and Poinsot, T. J., “Interactions Between Acoustics and Combustion,”
Proceedings of the Institute of Acoustics, Vol. 10, Inst. of Acoustics, Hertfordshire, U.K.,
1988, pp. 103–153.
5
Keller, J. J., “Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,”
AIAA Journal, Vol. 33, No. 12, 1995, pp. 2280–2287.
6
Dowling, A. P., “The Calculation of Thermoacoustic Oscillations,” Journal of Sound
and Vibration, Vol. 180, No. 4, 1995, pp. 557–581.
7
Poinsot, T., Trouve, A., Veynante, D., Candel, S., and Esposito, E., “Vortex-Driven
Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics, Vol. 177,
1987, pp. 265–292.
8
Schadow, K. C., Gutmark, E., Parr, T. P., Parr, D. M., Wilson, K. J., and Crump,
J. E., “Large-Scale Coherent Structures as Drivers of Combustion Instability,” Combus-
tion Science and Technology, Vol. 64, 1989, pp. 167–186.
9
Straub, D., Richards, G., Yip, M. J., Rogers, W. A., and Robey, E. H., “Importance of
Axial Swirl Vane Location on Combustion Dynamics For Lean Premix Fuel Injectors,”
AIAA Paper 98-3909, July 1998.
10
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premixed
Gas Turbine Combustion,” ASME Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 2, 1998, pp. 294–302.
11
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 1809–1816.
12
Lee, J. G., Kim, K., and Santavicca, D. A., “Measurement of Equivalence Ratio Fluc-
tuation and Its Effect in Heat Release During Unstable Combustion,” Proceedings of the
Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA, 2000, pp. 415–421.
13
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean Premixed Gas Turbine Combustors,” ASME Transactions, Vol. 123,
No. 1, 2001, pp. 182–189.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 525
14
Lieuwen, T. C., “Experimental Investigation of Limit-Cycle Oscillations in an Unstable
Gas Turbine Combustor,” Journal of Propulsion and Power, Vol.18, No. 1, 2002, pp. 61–67.
15
Lieuwen, T. C., “Statistical Characteristics of Pressure Oscillations in a Premixed Com-
bustor,” Journal of Sound and Vibration, Vol. 260, No. 1, 2003, pp. 3–17.
16
Harrje, D. T., and Reardon, F. H. (ed.), “Liquid Propellant Rocket Combustion Insta-
bility,” NASA SP-194, 1972, pp. 467–468.
17
Lieuwen, T., and Zinn, B. T., “On the Experimental Determination of Combus-
tion Process Driving in an Unstable Combustor,” Combustion Science and Technology,
Vol. 157, 2000, pp. 111–127.
18
Lee, D. H., and Lieuwen, T., “Acoustic Nearfield Characteristics of a Conical, Pre-
mixed Flame,” Journal of the Acoustic Society of America, Vol. 113, No. 1, 2003, pp. 167–
177.
19
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive-Active Control of Combustion
Oscillations,” Spring Meeting of the Central States Section of the Combustion Institute,
Knoxville, TN, April 2002.
20
Lee, J. G., Kim, K., and Santavicca, D. A., “A Study of the Role of Equivalence Ratio
Fluctuation During Unstable Combustion in a Lean Premixed Combustor,” AIAA Paper
2002-4015, July 2002.
21
Gaydon, A. G., Spectroscopy of Flames, Chapman and Hall, London, 1974.
22
Gaydon, A. G., and Wolfhard, H. G., Flames, Their Structure, Radiation and Temper-
ature, Chapman and Hall, London, 1979.
23
Miller, S. A., “Development of a Flame Chemiluminescence Probe for Determination of
Primary Zone Equivalence Ratio in Gas Turbine Combustors,” M. S. Thesis, Pennsylvania
State Univ., University Park, PA, 1999.
24
John, R., and Summerfield, M., “ Effect of Turbulence on Radiation Intensity From
Propane-Air Flames,” Jet Propulsion, Vol. 27, 1957, pp. 169–178.
25
Clark, T., “Studies of OH, CO, CH and C2 Radiation From Laminar and Turbulent
Propane-Air and Ethylene-Air Flames,” NACA Technical Note 4266, 1958.
26
Diederichsen, J., and Gould, R. D., “Combustion Instability: Radiation From Premixed
Flames of Variable Burning Velocity,” Combustion and Flame, Vol. 9, 1965, pp. 25–31.
27
Hurle, I. R., Price, R. B., Sugden, T. M., Thomas, R. R. S., and Thomas, A., “Sound
Emission From Open Turbulent Premixed Flames,” Proceedings of the Royal Society of
London, Series A: mathematical and Physical Sciences, Vol. 303, 1968, pp. 409–427.
28
Price, R., Hurle, I., and Sugden, T., “Optical Studies of the Generation of Noise in
Turbulent Flames,” Proceedings of the Combustion Institute, Vol. 12, The Combustion
Inst., Pittsburgh, PA, 1968, pp. 1093–1102.
29
Hedge, E. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Instabilities in Dump Type Ramjet Combustors,” Combustion Science and Technology,
Vol. 55, 1987, pp. 125–138.
30
Langhorne, P. J., “Reheat Buzz: An Acoustically Coupled Combustion Instability. Part
1. Experiment,” Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
31
Samaniego, J. M., Yip, B., Poinsot, T., and Candel, S., “Low-Frequency Combustion
Instability Mechanisms in a Side-Dump Combustor,” Combustion and Flame, Vol. 94, 1993,
pp. 363–380.
32
Shih, W.-P., Lee, J. G., and Santavicca, D. A., “Stability and Emissions Characteristics
of a Lean Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 2771–2778.
526 J. G. LEE AND D. A. SANTAVICCA
33
Bandaru, R. V., Miller, S., Lee, J. G., and Santavicca, D. A., “Sensors for Measur-
ing Primary Zone Equivalence Ratio in Gas Turbine Combustors,” Proceedings of SPIE -
The International Symposium on Industrial and Environmental Monitors and Biosensors,
Vol. 3535, Society of Photo-Optical Instrumentation Engineers, Bellingham, WA, Nov.
1998, pp. 104–114.
34
Broda, J. C., Seo, S., Santoro, Shirhattakar, G., and Yang, V., “An Experimental Study
of Combustion Dynamics of a Premixed Swirl Injector,” Proceedings of the Combustion
Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998, pp. 1849–1856.
35
Venkataraman, K. K., Preston, L. H., Simons, D. W., Lee, B. J., Lee, J. G., and
Santavicca, D. A., “Mechanism of Combustion Instability in a Lean Premixed Dump Com-
bustor,” Journal of Propulsion and Power, Vol. 15, No. 6, 1999, pp. 909–918.
36
Haber, L., Vandsberger, U., Saunders, W., and Khanna, V., “An Examination of the
Relationship Between Chemiluminescence Light Emissions and Heat Release Rate Under
Non-adiabatic Conditions,” Proceedings of the International Gas Turbine Institute, 2000-
GT-0121, 2000.
37
Ikeda, Y., Kojima, J., Nakajima, T., Akamatsu, F., and Katsuki, M., “Measurement of the
Local Flame-Front Structure of Turbulent Premixed Flames by Local Chemiluminescence,”
Proceedings of the Combustion Institute, Vol. 28, The Combustion Inst., Pittsburgh, PA,
2000, pp. 343–350.
38
Lieuwen, T., and Neumeier, Y., “Nonlinear Pressure-Heat Release Transfer Func-
tion Measurements in a Premixed Combustor,” Proceedings of the Combustion Institute,
Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002, pp. 99–105.
39
Paschereit, C. O., and Gutmark, E. J., “Enhanced Performance of a Gas-Turbine
Combustor Using Miniature Vortex Generators,”Proceedings of the Combustion Institute,
Vol. 29, The Combustion Inst., Pittsburgh, PA, 2002, pp. 123–129.
40
Bellows, B. D., Zhang, Q., Neumeier, Y., Lieuwen, T., and Zinn, B. T., “Forced Response
Studies of a Premixed Flame to Flow Disturbances in a Gas Turbine Combustor,” AIAA
Paper 2003-824, Jan. 2003.
41
Samaniego, J. M., Egolfopoulos, F. N., and Bowman, C. T., “CO∗2 Chemilumines-
cence in Premixed Flames,” Combustion Science and Technology, Vol. 109, No. 1, 1995,
pp. 183–203.
42
Dandy, D., and Vosen, S., “Numerical and Experimental Studies of Hydroxyl Rad-
ical Chemiluminescence in Methane-Air Flames,” Combustion Science and Technology,
Vol. 82, No. 1, 1992, pp. 131–150.
43
Najm, H. N., Paul P. H., Mueller, C. J., and Wyckoff, P. S., “On the Adequacy of
Certain Experimental Observables as Measurements of Flame Burning Rate,” Combustion
and Flame, Vol. 113, No. 3, 1998, pp. 312–332.
44
Najm, H. M., Knio, O. M., Paul, P. H., and Wyckoff, P. S., “A Study of Flame Observ-
ables in Premixed Methane-Air Flames,” Combustion Science and Technology, Vol. 140,
No. 3, 1998, pp. 369–403.
45
Dowling, A. P., “Nonlinear Self-Excited Oscillations of a Ducted Flame,” Journal of
Fluid Mechanics, Vol. 346, 1997, pp. 271–290.
46
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat-Release/Acoustic Model for
Thermoacoustic Instability in Lean Premixed Combustors,” American Society of Mechan-
ical Engineers, Paper 98-GT-269, 1998.
47
Dasch, C. J., “One-Dimensional Tomography: a Comparison of Abel, Onion-Peeling,
and Filtered Backprojection Methods,” Applied Optics, Vol. 31, No. 8, 1992, pp. 1146–1152.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 527
48
Kim, K., Lee, J. G., and Santavicca, D. A., “Optimization of the Spatial and Tem-
poral Fuel Distribution for Active Control of Combustion Dynamics in Lean Premixed
Combustors,” AIAA Paper 2002-4024, July 2002.
49
Mallard, W. G., and Gardiner, W. C. Jr., “Absorption of the 3.39 µm He-Ne Laser
Line by Methane from 300 to 2400K,” Journal of Quantitative Spectroscopy and Radiative
Transfer, Vol. 20, No. 2, 1978, pp. 135–149.
50
Philippe, L. C., and Hanson, R. K., “Laser Diode Wavelength-Modulation Spectroscopy
for Simultaneous Measurement of Temperature, Pressure and Velocity in Shock-Heated
Oxygen Flows,” Applied Optics, Vol. 32, No. 30, 1993, pp. 6090–6103.
51
Yoshiyama, S., Hamamoto, Y., Tomita, E., and Minami, K., “Measurement of Hy-
drocarbon Fuel Concentration by Means of Infrared Absorption Technique with 3.39 µm
He-Ne Laser,” JSAE Review, Vol. 17, No. 4, 1996, pp. 339–345.
52
Mongia, R. K., Tomita, E., Hsu, F. K., Talbot, L., and Dibble, R. W., “Use of an Optical
Probe for Time-Resolved In situ Measurement of Local Air-to-Fuel Ratio and Extent of Fuel
Mixing with Applications to Low NOx Emissions in Premixed Gas Turbines,” Proceedings
of the Combustion Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 2749–
2755.
53
Koenig, M., and Hall, M. J., “Measurements of Local In-Cylinder Fuel Concentration
Fluctuations in a Firing SI Engine,” Society of Automotive Engineers, Paper 971644, 1997.
54
Wehe, S. D., Baer, D. S., and Hanson, R. K., “Tunable Diode-Laser Absorption Mea-
surements of Temperature, Velocity and H2 O in Hypersonic Flows,” AIAA Paper 97-3267,
July 1997.
55
Mihalcea, R. M., Baer, D. S., and Hanson, R. K., “Diode-Laser Sensor for Measurements
of CO, CO2 and CH4 in Combustion Flows,” Applied Optics, Vol. 36, No. 33, 1997,
pp. 8745–8752.
56
Seitzman, J. M., Tamma, R., and Vijayan, R., “Infrared Absorption Based Sensor
Approaches for High Pressure Combustion,” AIAA Paper 97–0318, Jan. 1997.
57
Mongia, R. K., Dibble, R. W., and Lovett, J., “Measurement of Air-Fuel Ratio Fluctu-
ations Caused by Combustor Driven Oscillations,” American Society of Mechanical Engi-
neers, Paper 98-GT-304, 1998.
58
Mihalcea, R. M., Baer, D. S., and Hanson, R. K., “Advanced Diode Laser Absorption
Sensor for In-Situ Combustion Measurements of CO2 , H2 O and Gas Temperature,” Pro-
ceedings of the Combustion Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA, 1998,
pp. 95–101.
59
Mongia, R., Torres, J., Dibble, R., Lee, D., Anderson, T., and Sowa, W., “Fast Re-
sponse Extraction Probe for Measurement of Air-Fuel Ratio Fluctuations in Lean Premixed
Combustors,” American Society of Mechanical Engineers, Paper 99-GT-277, 1999.
60
Ebert, V., Fernholz, T., Giesemann, C., Pitz, H., Teichert, H., and Wolfrum, J.,
“Simultaneous Diode-Laser-Based In-Situ-Detection of Multiple Species and Tempera-
ture in a Gas-Fired Power-Plant,” Proceedings of the Combustion Institute, Vol. 28, The
Combustion Inst., Pittsburgh, PA, 2000, pp. 423–430.
61
Webber, M. E., Wang, J., Sanders, S. T., Baer, D. S., and Hanson, R. K., “In-situ
Combustion Measurements of CO, CO2, H2O and Temperature Using Diode Laser Ab-
sorption Sensors,” Proceedings of the Combustion Institute, Vol. 28, The Combustion Inst.,
Pittsburgh, PA, 2000, pp. 407–413.
62
Drallmeier, J., “Hydrocarbon-Vapor Measurements in Pulsed Fuel Sprays,” Applied
Optics, Vol. 33, No. 33, Nov. 1994, pp. 7781–7788.
528 J. G. LEE AND D. A. SANTAVICCA
63
Edwards, J. L., Gouldin, F. C., and MacDonald, M. A., “High Speed Absorption
Tomography with Advanced Reconstruction Algorithms,” AIAA Paper 2003-1013, Jan.
2003.
64
Eckbreth, A. C., Laser Diagnostics for Combustion Temperature and Species, 2nd ed.
Gordon and Breach, New York, 1996.
65
Nguyen, Q.-V., and Paul, P. H., “The Time Evolution of a Vortex-Flame Interaction
Observed via Planar Imaging of CH and OH,” Proceedings of the Combustion Institute,
Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 357–364.
66
Seitzman, J. M., Kychakoff, G., and Hanson, R. K., “Instantaneous Temperature Field
Measurements Using Planar Laser-Induced Fluorescence,” Optics Letters, Vol. 10, No. 9,
1985, pp. 439–441.
67
Hanson, R. K., “Combustion Diagnostics: Planar Flowfield Imaging,” Proceedings of
the Combustion Institute, Vol. 21, The Combustion Inst., Pittsburgh, PA, 1986, pp. 1677–
1691.
68
Hanson, R. K., “Planar Laser-Induced Fluorescence Imaging,” Journal of Quantitative
Spectroscopy and Radiative Transfer, Vol. 40, No. 3, 1988, pp. 343–362.
69
Pringsheim, P., Fluorescence and Phosphorescence, Interscience Publishers, New York,
1949.
70
Berlman, I. B., Handbook of Fluorescence Spectra of Aromatic Molecules, Academic
Press, New York, 1971.
71
Lozano, A., Yip., B., and Hanson, R. K., “Acetone: A Tracer for Concentration Mea-
surements in Gaseous Flows by Planar Laser-Induced Fluorescence,” Experiments in Fluids,
Vol. 13, 1992, pp. 369–376.
72
Thurber, M. C., and Hanson, R. K., “Simultaneous Imaging of Temperature and Mole
Fraction Using Acetone Planar Laser Induced Fluorescence,” Experiments in Fluids, Vol.
30, 2001, pp. 93–101.
73
Koch, J. D., and Hanson, R. K., “Ketone Photophysics for Quantitative PLIF Imaging,”
AIAA Paper 2000-0413, Jan. 2001.
74
Thurber, M. C., and Hanson, R. K., “Pressure and Composition Dependence of Acetone
Laser-Induced Fluorescence with Excitation at 248, 266 and 308 nm,” Applied Physics B:
Lasers and Optics, Vol. 69, 1999, pp. 229–240.
75
Samperio, J. L, Lee, J. G., and Santavicca, D. A., “Characterization of the Effect of
Inlet Operating Conditions on the Performance of Lean Premixed Gas Turbine Combustors,”
AIAA Paper 2003-0825, Jan. 2003.
76
Lee, J. G., and Santavicca, D. A. “Fiber-Optic Probe for Laser-Induced Fluorescence
Measurements of the Fuel-Air Distribution in Gas-Turbine Combustors,” Journal of Propul-
sion and Power, Vol. 13, No.3, 1997, pp. 384–387.
77
Melton, L. A., “Exciplex-Based Vapor/Liquid Visualization Systems Appropriate for
Automotive Gasolines,” Applied Spectroscopy, Vol. 47, No. 6, 1993, pp. 782–786.
78
Tong, K., Quay, B. D., Zello, J. V., and Santavicca, D. A., “Fuel Volatility Effects
on Mixture Preparatin and Performance in a GDI Engine During Cold Start,” Society of
Automotive Engineers, Paper 2001-01-3650, Sept. 2001.
79
Venkataraman, K. K., “An Investigation of the Instability Mechanism in Lean Premixed
Dump Combustors, Ph.D. Thesis, The Pennsylvania State University, University Park, PA,
2000.
80
Lee, S.-Y., Seo, S., Broda, J. C., Pal, S., and Santoro, R. J., “An Experimental Estimation
of Mean Reaction Rate and Flame Structure During Combustion Instability in a Lean
Premixed Gas Turbine Combustor,” Proceedings of the Combustion Institute, Vol. 28, The
Combustion Inst., Pittsburgh, PA, 2000, pp. 775–782.
EXPERIMENTAL DIAGNOSTICS OF COMBUSTION INSTABILITIES 529
81
Santhanam, V., Knopf, F. C., Acharya, S., and Gutmark, E., “Fluorescence and Tem-
perature Measurements in an Actively Forced Swirl-Stabilized Spray Combustor,” Journal
of Propulsion and Power, Vol. 18, No. 4, 2002, pp. 855–865.
82
Peters, N., “Laminar Flamelet Concepts in Turbulent Combustion,” Proceedings of the
Combustion Institute, Vol. 21, The Combustion Inst., Pittsburgh, PA, 1986, pp. 1232–1250.
83
Buschmann, A., Dinkelacker, F., Schäfer, T., and Wolfrum, J., “Measurement of the
Instantaneous Detailed Flame Structure in Turbulent Premixed Combustion,” Proceedings
of the Combustion Institute, Vol. 26, The Combustion Inst., Pittsburgh, PA, 1996, pp. 437–
445.
84
Dunkelacker, F., Soika, A., Most, D., Hofman, D., Leipertz, A., Polifke, W., and
Döbbeling, K., “Structure of Locally Quenched Highly Turbulent Lean Premixed Flames,”
Proceedings of the Combustion Institute, Vol. 27, The Combustion Inst., Pittsburgh, PA,
1998, pp. 857–865.
85
Grehan, G., Gouesbet, G., and Rabasse, C., “Monotonic Relationships Between Scat-
tered Powers and Diameters in Lorenz-Mie Theory for Simultaneous Velocimetry and Sizing
of Particles,” Applied Optics, Vol. 20, No. 5, March 1981, pp. 796–799.
86
Stenzler, J. N, Lee, J. G., and Santavicca, D. A., “Penetration and Dispersion of Liquid
Jets in a Heated Crossflow,” AIAA Paper 2003-1327, 2003.
87
Lin, K. C., Kennedy, P. J., and Jackson, T. A., “Penetration Heights of Liquid Jets in
High-Speed Crossflows,” AIAA Paper 2002-0873, Jan. 2002.
88
Stenzler, J. N, Lee, J. G., Deepe, J. M., Santavicca, D. A., and Lee, W., “Fuel Transfer
Function Measurements in Modulated Liquid Jets,” ASME International Congress, IMECE
2004-60673, 2004.
89
Yu, K. H., Wilson, K. J., Parr, T. P., and Schadow, K. C., “Active Combustion Control
Using Multiple Vortex Shedding,” AIAA Paper 96–2760, July 1996.
90
Bachalo, W. D., and Houser, M. J., “Development of the Phase/Doppler Spray Analyzer
for Liquid Drop Size and Velocity Characterizations,” AIAA Paper 84–1199, June 1984.
91
Wang, H., McDonnell, V. G., Sowa, W. A., and Samuelsen, S., “Experimental Study of
a Model Gas Turbine Combustor Swirl Cup. I - Two-Phase Characterization. II - Droplet
Dynamics,” Journal of Propulsion and Power, Vol. 10, No. 4, 1994, pp. 441–452.
V. Combustion Instability Control
Chapter 17
Nomenclature
G = transfer function relating relative heat release to relative
acoustic pressure (-)
H = transfer function relating relative acoustic pressure to relative
heat release (-)
L = length, m
M = Mach number, µ/c (-)
P = time-average pressure, Pa
Q = time-average heat-release rate, W
R = acoustic transfer matrix for a cylindrical element
S = acoustic transfer matrix for a step expansion
T = gas temperature (K), or the acoustic transfer matrix for a damper
Z = acoustic impedance, p/ν, (m s)−1
c = speed of sound (m/s)
f = frequency, Hz
k = stagnation pressure loss coefficient (-)
p = complex acoustic pressure, Pa
q = complex amplitude of heat-release variation, W
s = cross-sectional area, m2
Copyright c 2005 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
∗ Energy Systems Dynamics Focus Area Leader, National Energy Technology Laboratory.
† Mechanical Engineer.
‡ Scientist.
533
534 G. A. RICHARDS ET AL.
I. Introduction
Combustion
+ Σ
Processes
G
-
q'
p'
Acoustic
Processes
H
will establish standing waves that can produce a periodic disturbance in the heat-
release rate, q . The system will be unstable if the timing (phase) and the amplitude
(gain) of these variations in pressure and heat-release rate produce constructive
feedback. This feedback process is analogous to conventional feedback-control
systems, where the processes shown in Fig. 17.1 would correspond to control-
system components. The G and H nomenclature and the summation circle shown
on the left side of Fig. 17.1 follow directly from the control-system literature and
will be discussed in more detail later.
Because of the feedback analogy, control-system models have become popu-
lar tools to both represent and diagnose combustion instabilities. Various levels
of detail can be included in control-system models, ranging from reduced-order
models,6 to computational fluid mechanics,7 to complete engine models, by us-
ing a combination of approaches.8 Practical application of these models has been
demonstrated by many authors.1,6,8,9,10 Even though a full model is not sought
to solve a particular problem, it is helpful to understand the concepts, because
many experimental efforts to develop passive control can be explained by control-
system ideas. Thus, in what follows, a simple representation of a control model
for combustion dynamics is presented as a framework for subsequent discussions.
Sections II.A and II.B provide background information, so that no prior training
in control theory is necessary. In Sec. II.C an example calculation is discussed
to demonstrate the use of feedback models. Section II.D reviews the physical
processes that contribute to the flame response, and Sec. II.E reviews various
computational approaches used to predict combustion-system stability.
τ
Air Flame
produced in the flame region, as well as the pressure produced by the acoustic
characteristics of the system.
In Fig. 17.1, the system element G represents the conversion of a pressure vari-
ation to a variation in the heat-release rate. Many mechanisms can contribute to a
variable heat-release rate. These mechanisms may include periodic changes in the
flame-surface area,11 changes in equivalence ratio,1,12−17 vortex shedding,18−23
changes in the bulk flow,24 and changes in flame anchoring.25−27 Which of these
mechanisms contributes to oscillations in a given problem is an important prac-
tical question and is discussed later. However, attention is often focused on the
equivalence-ratio variation, because it will usually accompany all the other mech-
anisms. The pressure drop across the premixer air passage is typically a few percent
of the operating pressure. Therefore, modest perturbations in the combustor pres-
sure will create significant variations in premixer airflow, and subsequent variations
in fuel–air ratio in the premixer. These variations in fuel–air mixture are transported
to the flame after a convection time lag τ , creating a heat-release perturbation that
may add to perturbations produced by other mechanisms, such as a variable flame
area.
A cos(ωt) B cos(ωt+φ)
Transfer
jω t function jω t+φ jω t
Ae Be = B' e
a) Transfer function nomenclature
connect
B'/A D'/C
equivalent to
B' D'
AC
B'/A
+
Σ B'/A +D'/C
+
D'/C
equivalent to
Sequential processes, in which the output of one transfer function supplies the
input to a second process, are analyzed by multiplying the transfer functions in or-
der. As shown in Fig. 17.3b, if the B /A process described before connects to a D /C
process, the net transfer function is (B /A)∗ (D /C). Signals can also be added alge-
braically because attention is restricted to a linear system. For example, a given in-
put may supply both the A-B transfer function and the C-D transfer function, with
the outputs combined (see Fig. 17.3c). The combined system response is the com-
plex sum (B /A) + (D /C). By adding or multiplying individual transfer functions,
it is possible to reduce more complex physical processes into a forward transfer
function (usually denoted G) and a feedback transfer function (usually denoted H).
Flame Position
10 cm
summing point. In this paper, the feedback that emerges from H is subtracted from
the disturbance. This operation is defined as negative feedback in control theory.
With this nomenclature, disturbances that originate at the summing point will pass
through G and H, with modifications to both amplitude and phase. Note that a phase
of 180 deg(π ) corresponds to multiplication by −1 [i.e., cos(π )]. Intuitively, if the
disturbance is returned from H with a larger absolute magnitude and a negative
sign, the original disturbance will have a larger amplitude after passing through
the summing point. Note that as the signal from the feedback block passes through
the negative branch of the summing point, the signal is multiplied by −1. Under
this idealized condition, the disturbance will grow in amplitude each time it passes
around the loop, and the system will become unstable.
This intuitive understanding can be matched by formal analysis that leads to
a criterion for stability. The output of a signal passed through G and H but not
returned through the summation point is known as the open-loop frequency re-
sponse. The open-loop frequency response can be used to evaluate stability from
both Bode and Nyquist plots described subsequently.
Two combustor examples shown in Fig. 17.4 will be analyzed. The fuel–air
premixer at the left supplies a step expansion into the region where the flame is
stabilized. The flame is treated as a thin disk located just downstream of the step
expansion. The remainder of the combustor is a long tube, including a second step
expansion, and is then terminated at a closed acoustic boundary. These examples
approximate the conditions typically encountered in combustion test rigs, in which
the downstream boundary may represent a backpressure-control valve. For the cal-
culated results presented next, parameters such as pressure, temperature, flow rate,
and fuel–air ratio are all selected to be representative of gas-turbine combustors.
Treating the flame as a discontinuity that interacts with acoustic waves, a one-
dimensional acoustic analysis is used to determine the acoustic pressure produced
by imposed heat-release perturbations. For the geometry shown, the flame–acoustic
relations presented by Chu5 are used, and the acoustics are modeled using a transfer
matrix method.28 The transfer matrix method can account for mean flow effects
and acoustic losses at abrupt area changes. An outline of these calculations is
presented in the Appendix.
a) p′ / P
=H
q′ / Q
0.3
Phase (deg)
200
Magnitude 100
0.2
0
0.1 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
Frequency (Hz) Frequency (Hz)
b) q ′ / Q
= G = 6e−jωτ
p′ / P
200
Phase (deg)
Magnitude
6.0
100
4.0 0
2.0 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
Frequency (Hz) Frequency (Hz)
c) GH
2.0 200
Phase (deg)
Magnitude
1.5 100
1.0 0
0.5 -100
0.0 -200
0 100 200 300 400 0 100 200 300 400
Frequency (Hz) Frequency (Hz)
Fig. 17.5 Frequency response (Bode plots) of transfer functions used in the first
example problem with a single acoustic mode (Fig. 17.4, top).
to the acoustic losses or damping in the system. In this example, acoustic losses
arise from the step expansion, mean flow, and mean heat release. For this example,
Fig. 17.5a represents the H transfer function depicted in Fig. 17.1.
The combustion response to acoustic pressure perturbation must also be ana-
lyzed. In actual applications, this response must account for the various mech-
anisms that will be described later (i.e., variable fuel–air ratio, variable flame
area). For the purposes of this example, a simple flame-transfer function will be
considered. Again referring to the normalized perturbations ( p /P and q /Q), the
flame will be treated as having a constant gain of magnitude 6.0, but with a time
delay τ = 2 ms relative to the acoustic pressure at the flame. This transfer function
is 6.0e− j2πτ f . Thus, a normalized pressure perturbation produces a normalized
heat-release rate perturbation six times larger and 2 ms later, which is easily re-
alized in practical systems. This transfer function for the combustion response is
shown in Fig. 17.5b. The phase plot in Fig. 17.5b is representative of all time-delay
systems. The phase angle decreases in a linear fashion with frequency because the
phase angle θ = −2π τ f . In this plot, the phase angle is wrapped into the range
−180 to +180 deg. The same information can be plotted from 0 to 360 deg as
well, avoiding the abrupt discontinuity at ±180 deg.
The open-loop response of this example is the series connection of both the G
and H transfer functions. As explained earlier (Fig. 17.3), this series connection
is computed as the product of the individual gain functions and the sum of the
individual phase angles. The resulting frequency response is shown in Fig. 17.5c.
Note that the magnitude is greater than unity at ∼240 Hz and the phase angle is
±180 deg. If this GH output is connected to the summation point in Fig. 17.1
(i.e., closed loop), the system would be unstable for the reason explained earlier,
that is, the disturbance would grow in amplitude each time around the loop. If
the open-loop gain is less than unity at a phase angle of ±180 deg, the system
would be stable because the signal returning from the summing junction would be
smaller than the original disturbance each time around the loop. This reasoning
is not entirely complete because of other complications such as whether the gain
plot crosses magnitude 1.0 more than once. This complication can be addressed by
using a Nyquist analysis, which is discussed next. The presentation that follows is
an adaptation of analysis discussed by Fannin et al.29 Although a brief description
of the Nyquist stability criterion is presented in the following paragraphs, it is
not intended to be a complete tutorial. A more complete description of Nyquist
analysis is found in control textbooks.30
a) Nyquist Plot
2
1 258 Hz
Imaginary
τ = 1.6 ms
0
-1
-2
-2 -1 0 1 2
Real
b) Nyquist Plot
2
251 Hz 263 Hz
1
Imaginary
τ = 2 ms
0
-1
241 Hz
-2
-2 -1 0 1 2
Real
c) Nyquist Plot
2
1
Imaginary
-1 241 Hz τ = 2.6 ms
-2
-2 -1 0 1 2
Real
Fig. 17.6 Nyquist plots for the single acoustic mode example (Fig. 17.4, top) at three
different time lags (τ = 1.6, 2.0, and 2.6 ms).
542 G. A. RICHARDS ET AL.
corresponding to Fig. 17.5c, where the time lag is 2 ms. The circular lobe corre-
sponds to frequencies between 200 and 300 Hz, in which appreciable magnitude
exists from the open-loop response. For clarity, three of the frequencies are indi-
cated on the lobe. The corresponding points (phase, magnitude, and frequency)
can be found from a close inspection of the Bode plot (Fig. 17.5c). Note that the
lobe represents a small range of frequencies in this example problem.
As explained in control-theory textbooks, the system stability can be evaluated
by counting how many times the Nyquist plot encircles the point −1 on the x axis.
The definition of what constitutes encirclement is fairly involved, and one must
refer to control textbooks for complete details.30 In brief, encirclement direction
(clockwise or counterclockwise) must be counted as positive or negative encir-
clement, and the sum of all the positive and negative encirclements are added to
arrive at a net number of encirclements. The plot also requires considering infor-
mation at negative frequencies, essentially a reflection of the Bode plot into the
negative-frequency axis, and the open-loop system must itself be stable. In this ex-
ample problem, these details do not enter the discussion but should be considered
before using Nyquist analysis on more complex problems. The complete Nyquist
analysis predicts that the system will be unstable if the net number of encirclements
is greater than zero. Figure 17.6b shows that the Nyquist plot does indeed encircle
−1 on the real axis and would therefore be unstable.
The benefit of the Nyquist analysis becomes very apparent when assessing how
different time lags affect system stability. Figure 17.6 shows the open-loop system,
GH, at three different values of the time lag: τ = 1.6, 2.0, and 2.6 ms. Notice that
increasing the time lag rotates the lobe clockwise. This behavior can be understood
by noting that over the small-frequency range in which the amplitude is significant
(230–270 Hz), changes to the phase angle θ = −2πτ f are dominated by changes
in τ . Changes in τ appear to rotate each point on the lobe approximately the same
angle, producing a rotation of the lobe.
The Nyquist plots can also be used to investigate stability boundaries for the
system. For example, both the short and the long time lags shown in Fig. 21.6
almost encircle the −1 point on the real axis. The values of τ that almost produce
encirclement of the −1 point are stability boundaries for the system. Each of these
stability boundaries, or values of τ , has a corresponding frequency at which the lobe
crosses the negative real axis. In this example, the system would be unstable for
frequencies of 258 and 241 Hz at τ = 1.6 and 2.6 ms, respectively. The frequency
range and the size of the lobe depend on the rolloff in the phase-angle function.
For problems that have larger acoustic losses, the phase rolloff near the resonant
frequency may cover a larger frequency range, and the lobe in the Nyquist plot
would also cover a wider range of frequencies.
If the system in this example were actually operated in the closed-loop mode, the
limit-cycle frequencies could be estimated from the so-called describing-function
theory.30 Under the assumption of real-valued describing functions, the limit-cycle
frequency would correspond to the frequency at which the Nyquist plot crosses
the negative real axis. Considering the sequence of time lags in Fig. 17.6, the
implication is that as the time lag increases from 1.6 to 2.6 ms, the frequency
would change from 258 to 241 Hz. This frequency shift as a function of time lag
is a general feature of the Nyquist analysis, and the range of frequencies depends
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 543
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
200
Phase (degrees)
150
100
50
0
-50
-100
-150
-200
0 100 200 300 400 500
Frequency (Hz)
Fig. 17.7 The open-loop frequency response (Bode plot) for the example with two
resonant acoustic modes (Fig. 17.4, bottom), τ = 1.5 ms.
544 G. A. RICHARDS ET AL.
a) Nyquist Plot
3
2 τ = 0.9 ms
411 Hz
Imaginary 1
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
b) Nyquist Plot
3
2 τ = 1.5 ms
408 Hz
1
Imaginary
0
-1 186 Hz
-2
-3
-3 -2 -1 0 1 2 3
Real
c) Nyquist Plot
3
2 τ = 2.0 ms
184
1
Imaginary
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
Fig. 17.8 Nyquist plots for the example with two resonant acoustic modes (Fig. 17.4,
bottom) at three different time lags (τ = 0.9, 1.5, and 2.0 ms).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 545
the time lag to τ = 1.5 ms makes the system unstable to the high-frequency mode
(408 Hz). Note that the high-frequency lobe encircles the −1 point on the real axis,
but the low-frequency mode does not. A further increase to τ = 2.0 ms causes the
instability to shift frequency modes and become unstable at 184 Hz.
The rotation of these lobes underscores the fundamental problem with achieving
stability from changes to the combustion time lag (i.e., by changing the phase).
Changes in time lag can simply change the oscillating frequency rather than pro-
duce stability. Even when a combustor has just a single dominant acoustic mode,
the width of the phase roll can produce oscillations over a range of frequencies
corresponding to the range of selected time lags as shown in Fig. 17.6. In summary,
careful consideration must be given to the acoustic modes before attempting to
solve a dynamics problem by adjusting the time lag.
The preceding example assumed that the flame response is a constant magnitude.
In real applications, the transfer-function magnitude and phase are governed by
the flame response to acoustic perturbations. As noted earlier, this response can
involve multiple physical processes that are often difficult to differentiate. The
next section reviews the physical processes associated with the flame response and
various approaches used to describe a combustion-transfer function.
Q = ρY f SA f H (17.1)
where ρ is the density of the reactants, Y f is the mass fraction of fuel in the
premixed gases, S is the flame speed, A f is the area of the flame, and H is the
heat of reaction per unit mass of fuel. Based on this equation, the heat release
clearly can vary with perturbations in density, fuel mass fraction, flame speed,
and flame area. In gas-turbine combustion, the density perturbations arising from
acoustic pressure are typically much smaller than the other terms and are often
neglected. On the other hand, factors such as unsteady aerodynamics may produce
a significant modulation in flame area and may not be neglected. Likewise, changes
in the flow of either fuel and/or air will change the fuel mass fraction and the flame
speed.
In short, numerous mechanisms can generate perturbations in the heat release
at the flame. In most practical applications, it is difficult to separate and con-
trol these mechanisms to achieve stable combustion. Nevertheless, this section
is aimed at discussing some of these physical processes and attempts to model
the combustion-transfer function. In the following paragraphs, previous efforts to
understand simple premixed flames with constant fuel–air ratio will be discussed.
Following this discussion, the combined problem of fuel–air variation and flame-
area response will be considered.
premixed burner to acoustic perturbations. This work showed that variations in the
flame area were responsible for the driving, or damping, acoustic waves imposed on
the burner. Companion experiments demonstrated that the flame response depends
significantly on the mean flow velocity and the fuel–air ratio. Merk32 presented an
improved analysis of a premixed burner flame and was able to derive an explicit
expression for the flame-transfer function [see Eq. (17.2)].
1
A = u (17.2)
1 + jωτ1
of the transfer function was poorly predicted for frequencies beyond ∼30 Hz. For
frequencies higher than ∼30 Hz, the phase of the response depended significantly
on the mean flow velocity and the burner dimensions. Additional experiments (not
reported) showed that the spatial velocity distribution at the base of the flame must
be accounted for to improve model predictions.
Although the preceding discussion is limited to relatively simple premixed Bun-
sen or jet flames, the difficulty of accurately predicting a flame-transfer function
is clear. In bluff or step-stabilized flames (e.g., afterburners or dump combustors),
flame-area variations may originate from oscillations in the shear layer and from
vortex merging.18−24,39 In swirl-stabilized flames, these aerodynamic phenomenon
are even more complicated. The swirl angle, the size of the combustor-step expan-
sion, and the length of the combustor can all affect the flow dynamics. Thus, for
swirl-stabilized flames, there is no general approach to estimate the contribution
of the flame-area variation to the flame response. Some recent attempts to measure
or predict transfer functions for swirl flames are discussed below.
-15
(dB)
Magnitude (dB)
-25
Magnitude
-35
Φ = 0.55
Φ = 0.60
Φ = 0.65
-45
10 100 1000
Frequency (Hz)
Frequency (Hz)
0
-200
(degres)
Phase (degrees)
-400
Phase
-600
-800 Φ = 0.55
Φ = 0.60
Φ = 0.65
-1000
10 100 1000
Frequency (Hz)
Frequency (Hz)
Nyquist plots discussed earlier and is an example of how the flame dynamics can
change appreciably with relatively minor changes in operating conditions.
In summary, Sec. II has shown how control models can be used to evaluate
the stability of combustion systems. The combustion process can be treated as
the forward transfer function G and the system acoustics can be represented in the
feedback path H. The physical processes that contribute to the combustion response
have been reviewed along with measurements and models that have been used to
describe the transfer functions. Nyquist analysis demonstrates both the potential
benefits and potential problems associated with adjusting the combustion time lag
to produce stability. In particular, frequency shifts and mode changes can frustrate
attempts to solve a combustion-dynamics problem. This will be discussed in more
detail with reference to experimental data in the next section.
τ/T = τ f = 1, 2, 3, . . . (17.3)
The criterion for instability expressed by Eq. (17.3) is an analog to the Nyquist
stability criterion described in Sec. II. That is to say, if the time delay (phase
angle) between the richer pockets of fuel–air ratio produced in the premixer and a
subsequent increase in the heat release is such that the combustor becomes unstable,
then the Nyquist plot should produce an encirclement of the −1 point on the real
axis. The concept introduced by Eq. (17.3) expresses the same idea. For example,
if p produces a heat-release rate variation q that will amplify the next cycle of p
(i.e., τ f = 1), or subsequent cycles (i.e., τ f = 2, 3, . . . ), then an instability will
occur. The sequence 1, 2, 3, . . . is specific to the simple example described. As
noted by Putnam,41 it is not necessary that the p and q perturbations be precisely
in phase to meet the Rayleigh criterion. In principle, oscillations can occur for a
range of τ f (i.e., ± 0.25) centered on the indices shown before.
Depending on the acoustic response of both the fuel system and the air passage
in the premixer, rich and lean pockets of mixture may be produced at various phase
angles relative to the pressure p in the combustor. For example, if the premixer
response produces richer pockets at the minimum of the combustor pressure, then
only 1/2 acoustic period is needed to align q with the next maxima in pressure.
Thus, the sequence would be
Lacking details of the premixer response, it is not known a priori what sequence of
numbers will describe oscillating regions in a given application. To further com-
plicate matters, τ is an average representation of the time from fuel injection to the
time of combustion. As discussed by Lieuwen et al.,65 this time lag depends on
the flame location and the flame shape. In practical systems, neither flame location
nor flame shape is easily measured or predicted. For these reasons, the sequences
described in Eq. (17.3) and Eq. (17.4) are typically determined experimentally, al-
though some applications of computational fluid dynamics have shown promising
predictions of both the time lag and the associated stability regions.8,12,50
Experimental evaluation of the time-lag model described before has been
demonstrated by Richards and Janus13 and Straub and Richards.63 A can-style
combustor test rig (Fig. 17.10) has been used to record the pressure dynamics from
552 G. A. RICHARDS ET AL.
Optical Access
(OH* Signal)
Fig. 17.10 Schematic illustration of experimental combustion test rig from Ref. 13.
a premixed gas-turbine fuel injector. The modular premixer design (Fig. 17.11)
allows the position of fuel injection to be changed from three positions A, B, C,
or simultaneously from two of the three positions. Thus, changes to the time lag
could be studied by changing the bulk flow velocity and the physical distance L.
Data are collected over a range of operating conditions. The time lag for each
condition has been estimated by using a fixed-flame standoff.
Figure 17.12 shows the observed rms pressure as a function of τ f . These data
indicate that oscillations are confined to a band 0.45 < τ f < 0.7. The edges of
this region are referred to as the stability boundaries. This plot can be used to
understand how proposed changes in the nozzle geometry, such as increasing
the premixer length, would affect stability. For a stable combustor operating at
τ f = 0.4, Fig. 17.12 indicates that proposals to move the fuel injector upstream
at a fixed velocity will increase τ f such that oscillations will occur. This increase
has been demonstrated experimentally by Richards and Janus.13 It would also be
possible (but not necessarily advisable) to increase the fuel time lag enough to
get to the upper-stability boundary. This increase in fuel time lag has been shown
experimentally by Straub and Richards63 and will be discussed in more detail later.
It is useful to understand how the τ f plot and the control-model analysis are
related. In the Nyquist analysis, the lobes of the plot rotate as the time lag is
changed. As previously discussed (Fig. 17.6a–c), if encirclement occurs at a par-
ticular frequency (i.e., f 1 ), small increases in the time lag will continue the rotation
of the Nyquist plot, and the frequency of the instability will decrease until the lobe
does not encircle −1 at a lower frequency (i.e., f 2 ). Attempts to solve instability
problems by adjusting the time lag must recognize that the frequency will change
in response to changes in the time lag.
Although the stability boundary in the τ f plot is clearly recognized (Fig. 17.12),
increasing the time lag will result in lower-frequency oscillations, depending on
the bandwidth of the resonant frequency. The bandwidth is often overlooked, or
misunderstood, when using plots similar to Fig. 17.12, but the Nyquist analysis
indicates this very clearly (Fig. 17.6a–c). Note also that the left stability boundary
shown in Fig. 17.12 is typically accompanied by the higher-frequency oscillations,
assuming the combustion system has a single resonant frequency.
This discussion suggests that the bandwidth of the acoustic response is an impor-
tant consideration when applying passive-control techniques to control combustion
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 553
Geometry 1
Geometry 2
5.0
RMS Pressure (%)
4.0
3.0
2.0
1.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
(Time Lag)*(Frequency)
Fig. 17.12 Data showing the experimentally determined stability boundaries using a
time-lag model from Ref. 13.
554 G. A. RICHARDS ET AL.
τf = τf = τf =
0.50-0.75 1.50-1.75 2.50-2.75
1000
Natural
800 freq's
Frequency (Hz)
f1 f1 =
600 640 Hz
f1 or f2?
f1 or f3?
f2 f2 =
400
410 HZ
f2 or f3?
200 f3 f3 =
190 HZ
0
0 1 2 3 4 5
Fig. 17.13 Stability boundaries of the time-lag model plotted as frequency verses time
lag. Also shown are three different frequency modes that can complicate approaches
to control combustion oscillations by changing the time lag.
dynamics. If the bandwidth of the acoustic response is large, the lobe of the Nyquist
plot covers a wide frequency range, and significant changes in the value of τ are re-
quired to cross the stability boundaries. Fannin et al.29 showed a more detailed com-
parison between the Nyquist stability boundaries and the simple time-lag model.
Multiple acoustic modes introduce another complication to the time-lag model.
When the system acoustics produce strong response at several frequencies, at-
tempts to modify the time lag could result in a jump between frequencies rather
than produce stability. This point has been demonstrated in the example Nyquist
plots shown in Fig. 17.8, in which subsequent lobes of the Nyquist diagram rotate
and encircle −1 as the time lag is increased (or decreased). The concept can also
be described less formally by a plot like the one shown in Fig. 17.13. In Fig. 17.13,
frequency is plotted on the vertical axis, and the time lag is plotted on the horizon-
tal axis. The indices at which instability occurs are plotted as regions having an
assumed width (i.e., 0.5 < τ f < 0.75, etc.). Horizontal lines are drawn through
combustor natural acoustic frequencies as shown. Combinations of τ f within these
shaded regions have the correct phase to meet the Rayleigh criterion and may pro-
duce an unstable condition. If a given combustor exhibits oscillations at frequency
f 1 with τ = 1 ms, a proposed solution might be to increase the time lag. However,
because of the multiple acoustic modes, this solution has the potential to get to
the edge of the stability boundary τ f 1 = 0.75, and then drop to τ f 2 = 0.50. It is
conceivable that continued increases in time lag to ∼2 ms could produce stability,
but slightly greater time lag could then jump back to τ f 1 = 1.50, or τ f 3 = 0.50.
Thus, adjustment of the time lag must be pursued with a full understanding of the
frequency spacing of acoustic modes. A similar conclusion can be drawn from the
formal Nyquist analysis, in which multiple natural frequencies produce multiple
lobes on the Nyquist diagram.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 555
8.0%
RMS Pressure
6.0%
4.0%
2.0%
0.0%
0 100 200 300 400 500
Frequency (Hz)
Fig. 17.14 Experimental data showing multiple frequency modes.
8.0%
RMS Pressure
6.0%
4.0%
2.0%
0.0%
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80
(Time Lag) * (Frequency)
Fig. 17.15 Data showing experimentally determined stability boundaries for
multiple-frequency mode combustor by using the time-lag model.
This discussion has shown that solving dynamics problems by adjusting the time
lag may be complicated by the multiple τ f bands and multiple acoustic modes. If
the frequency spacing f between adjacent acoustic modes is such that mode tran-
sitions are possible (as shown in Fig. 17.13), there is a good chance that changing
the time lag will simply produce a frequency shift rather than stable combustion.
This conclusion depends on the gain of the various modes, but it does suggest the
following rule of thumb: where the mode spacing f is such that f τ ∼ < 0.5,
200
100
0
0 2 4 6 8 10
Time Lag (msec)
Fig. 17.16 Experimental data showing transition between τ f bands and different
frequencies.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 557
changes in the time lag can lead to mode switching. In these instances, rather than
modifying τ to produce stability, consideration should be given to increasing the
time-lag distribution or modifying the acoustic response of the combustor.
Location A Location A
10%
10%
RMS Pressure
8%
RMS Pressure
8%
6%
6%
4%
4%
2%
2%
0%
0.59
0% Equiv.
0.59
30 40
0.67
Equiv.
30 40
0.67
50 60 Ratio
0.77
50 60 Ratio Velocity
0.77
Velocity
(m/s)
(m/s)
Location B Location C
10% 10%
RMS Pressure
RMS Pressure
8% 8%
6% 6%
4%
4%
2%
2%
0%
0.59
Equiv. 0%
0.59
Velocity 30 40 50
0.67
Ratio 30 40 Equiv.
0.67
0.77
(m/s) 60
10% 10%
RMS Pressure
RMS Pressure
8% 8%
6% 6%
4% 4%
2% 2%
0% 0%
0.59
Equiv.
0.59
30
Equiv.
0.67
40
30 40
0.67
50
Velocity Ratio
60
50
0.77
Velocity Ratio
0.77
(m/s) 60
(m/s)
Fig. 17.17 Experimental data showing the effect of changing the distribution of time
lag by using fuel injection from two axial locations, A and B and A and C in Fig. 17.11.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 559
different fuel injectors at different angles relative to the combustor flow axis. The
resulting asymmetry in the flame shape prevents uniform coupling to the acoustics
and also has the effect of shifting the flame position on different oriented injectors.
The net effect is again a distribution of time lags among the different injectors.
Lovett and Uznanski66 used a different concept to produce multiple time lags
in a single fuel injector. Their approach allowed a split between fuel supplied
to a main and a secondary premixer passage. A different time lag is associated
with each of these passages. Therefore, a composite response could be achieved
by adjusting the flow splits. This approach produced reduced dynamics at select
values of the secondary fuel flow.
As described in the preceding paragraphs, deliberate introduction of multiple
time lags is one approach to improve combustion stability. Another approach is
to simply take advantage of the mixing processes that occur during fuel injec-
tion. Because the fuel–air perturbations are mixed by turbulent processes in the
premixer, a single time lag does not completely describe the premixer response
to flow perturbations. To account for turbulent mixing, a distribution of time lags
will better represent the response of even a single point of fuel injection. Scarinci
and Freeman45 showed that turbulence in the premixer can significantly disperse
fuel–air perturbations and suggested that dispersion may reduce the oscillating
amplitude of combustors using longer premixer barrels.
Sattelmayer46 has also shown that the time-lag distribution can play a key role
in reducing the overall magnitude of the heat-release oscillation. Thus, attempts
to improve combustion stability should recognize the contribution of time-lag
distribution in the premixing process and in the flame geometry. These flame-
geometry effects will be described next.
C. Flame-Geometry Effects
Another consideration that has an impact on the time-lag distribution is the
geometry of the flame. This effect can be understood by noting that the arrival of
fuel–air perturbations is distributed over the surface of what is typically a conical-
flame geometry. The conical geometry means that the delivery of a fuel perturbation
will produce a combustion response that is distributed over the surface of the flame
and subsequently over a range of time lags. Putnam41 recognized the importance
of these geometric features and developed a correction to a time lag by integrating
over the surface of the flame. More recently, Lieuwen et al.65 developed a similar
integral analysis and applied it directly to flame geometries that are of interest to
gas-turbine combustors. This analysis showed that correction factors to the time
lag can be as large as a factor of 1.5, and the correction is sensitive to the shape of
the flame.
Because of the uncertainties connected with describing the position and shape
of turbine combustion flames, rig measurements of the time lag may be necessary
to describe the flame response. Straub et al.71 have made preliminary attempts to
measure the flame time lag in a practical scale combustor, but more work is needed
on this topic. Krebs et al.50 measured the time lag in an atmospheric-pressure model
of a gas-turbine combustor and showed that a time-lag distribution was needed to
describe the measured response. The time-lag distribution is needed to account for
the flame shape.
560 G. A. RICHARDS ET AL.
stationary engines, the disparity of damper use may be related to the relatively low
frequencies encountered in most turbine applications (hundreds of hertz) vs those
typically encountered in rocket engines (kilohertz range). The lower frequencies
require physically larger dampers, which complicates engine packaging, and may
be a potential drawback. Nevertheless, dampers should not be overlooked in a
strategy to stabilize combustion.
Given the difficulty of proposing changes to the combustion response (the G
transfer function, in Fig. 17.1), a damper design can be proposed that will very
likely reduce acoustic feedback (the H transfer function, in Fig. 17.1). This proposal
is not to imply that damper design is easy or without uncertainties. However, given
the cost of reengineering a combustion system to produce a desired combustion
response, addition of dampers may be worth consideration. These acoustic dampers
are described next.
A. Damper Description
The simplest damper of all is a hole, releasing acoustic energy from the com-
bustion chamber that would otherwise return to the feedback loop. The efficacy
of this method is well known to practicing combustion engineers. Putnam41 noted
the following advice for practitioners faced with stubborn oscillation problems in
industrial burners, “To solve an oscillating combustion problem, drill a hole. If that
doesn’t work, drill two holes.” Although this anecdote is a humorous, it represents
genuine experience that reducing the acoustic gain can stabilize oscillating systems.
Conversely, eliminating holes can lead to combustion instabilities. Modern pre-
mixed combustors are designed specifically to avoid dilution holes, removing a
source of acoustic damping. The avoidance of holes is yet another reason why
premixed combustors tend to have problems with dynamics. Earlier diffusion-
flame combustors used numerous dilution holes around the perimeter of the liner,
providing a source of acoustic damping that is absent in premixed combustors.
Although drilling holes may be an acceptable control strategy in industrial burn-
ers, it is not an option for gas-turbine combustors in which flow splits must be ac-
curately controlled to meet performance targets. Alternatively, closed resonators
could be used to absorb acoustic energy. Because the resonators are closed, they
do not compromise the designed flow splits. These types of resonators have been
used extensively in rockets9,72 and afterburners.73 Figure 17.18 shows two com-
mon resonator geometries that have been used in gas-turbine applications. Both
types (i.e., the Helmholtz resonator and the quarter-wave resonator) are shown in
Fig. 17.18 for convenience. These two geometries are not necessarily combined in
practice, however. Figure 17.18 also shows the formulas and nomenclature used
to calculate the natural frequency f 0 of the resonator.
When pressure oscillations occur, flow enters and exits the resonator mouth.
The energy dissipated at the entrance–exit provides damping to the system. It can
be shown9 that the greatest losses are generated by maximizing the magnitude of
the oscillating velocity; this maximization is achieved by tuning the resonator so
that the natural frequency f 0 is close to the frequency that is to be damped in the
combustor. For a given f 0 , the Helmholtz neck area S can be optimized with the
length L and volume V to meet packaging space requirements. In the quarter-wave
design, f 0 can only be established by the length.
562 G. A. RICHARDS ET AL.
Helmholtz
c S
Resonator f0 ⋅
2⋅ π ( L + 1.7⋅ a ) ⋅ V
Fuel Volume, V
Radius, a
Area, S
Combustion
Air Flame products
Quarter L
wave 1 c
f0 ⋅
resonator 4 L
Fig. 17.18 Schematic illustration of resonator concepts to dampen the acoustic feed-
back of the combustion system.
The actual performance of a resonator depends on the resonator geometry and the
operating conditions. Because the acoustic dissipation occurs at the entrance/exit
flow near the mouth, the geometry of this region is very important. Laudien et al.72
show that rounded versus square corners at the resonator–combustor connection
produce a significant difference in the acoustic response. Furthermore, because the
resonator gas may include combustor products and purge cooling gas, the speed of
sound is uncertain. This uncertainty makes the design the resonator for a specific
natural frequency more complex. The correct resonator design and tuning is not
usually achieved from analysis alone. In most rocket applications and in the few
turbine applications cited subsequently, the resonator properties are finalized by
experimental testing.
Selecting the position and number of resonators is also an important considera-
tion. It does little good to place the resonator at an acoustic node. For example, in
annular combustors, circumferential acoustic modes (with waves traveling around
the annulus) are often characterized by standing nodes at specific positions. Res-
onators added at these node positions will not provide damping, because there is
no acoustic pressure to drive the oscillator. Attempts to position resonators at the
pressure antinodes may be frustrated by a repositioning of the node to the new res-
onator location. Thus, it may be essential to position multiple resonators by using
the number of resonators that are guaranteed to produce an assymetry relative to
the acoustic mode shape. For example, three resonators cannot all be aligned with
a wave structure having only two nodes. Decisions about the number of resonators
are made based on an analysis of acoustic waveform and the required damping in
specific situations. Some examples are described in the next section.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 563
By using the approximate volume ratio described previously and the natural fre-
quency, only the neck length L and cross section S must be chosen. Assuming the
neck radius a is a direct function of S, a range of values for S can be studied. Note
that S must be a significant fraction of the combustor cross section to allow the
transfer of acoustic energy between the combustor and resonator. In this example,
S is chosen as 10% of the combustor cross section. This choice allows the length
L to be calculated directly from Eq. (17.5). The resulting resonator geometry is
shown to scale relative to the overall combustion system in Fig. 17.19. Note that
space and packaging considerations may lead to variations on this initial design.
566 G. A. RICHARDS ET AL.
Resonator
Flame
D
C 10 cm
B
A
Fig. 17.19 Geometry of Helmholtz resonator design, shown approximately to scale,
on the example combustor of Fig. 17.4 (bottom). Positions A to D represent different
resonator locations.
For example, the same natural frequency could be achieved with a shorter neck
but larger volume.
After the resonator geometry is defined, the resonator location can be assessed
by using the system model described in the example problem. Experimental ver-
ification is also warranted for actual applications, but, for the purposes of this
example, the effects of resonator location will be pursued by using the model
outlined in the Appendix. Four resonator locations A thru D will be analyzed to
demonstrate the effect of axial position on the system response.
As noted in the Appendix, the resonator can be included in the transfer-matrix
analysis by the addition of a transfer matrix containing the resonator impedance
Z r . Laudien et al.72 provide an expression for the resonator impedance:
2π f 2 4 L µvis π f 2π f (L + 1.7a) f 02
Zr = + ε+ +j· 1− 2 (17.6)
c S 2a ρ S f
In this expression, f is the frequency and f 0 is the natural frequency of the resonator
calculated from Eq. (17.5). The speed of sound c and the dynamic viscosity µvis
are evaluated at the (assumed) gas conditions in the resonator. The dimensionless
term ε is a resistance factor that depends on the quality of the orifice and nonlinear
processes at the mouth of the resonator. This term is difficult to evaluate analytically
and depends on the magnitude of the oscillating pressure. The resistance factor ε
has been measured, and the value typically varies from 10 to more than 30.72 A
value of 20 is selected for this study. Again, a parametric study of this parameter on
resonator performance is suggested to define the range of possible behavior before
conducting experimental testing. For this example, investigation of the effect of ε
(not reported) showed only a modest effect on resonator attenuation.
Given the geometry and parameter selections described in preceding text,
Fig. 17.20 presents the open-loop frequency response of the example combus-
tor without the resonator. Note that the baseline case without the resonator is
shown in the top of Fig. 17.20. Recall that the baseline corresponds to the spec-
tra of Fig. 17.7. With the resonator tuned to 410 Hz, the amplitude of the higher
combustor mode is replaced by two much smaller adjacent modes. Gysling et al.74
noted that the addition of a resonator produced two new eigenvalues, representing
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 567
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
5.0
4.0
Magnitude
3.0
2.0
1.0
0.0
0 100 200 300 400 500
Frequency (Hz)
the interaction of the resonator with the existing acoustic modes. The appearance
of additional modes is not a complication in this example, because their amplitude
is very low. However, the possibility of creating oscillations at different frequen-
cies should not be overlooked when adding a resonator. If the combustion gain
were much larger at the new frequencies, oscillations might occur. Note that the
resonator has little effect on the lower-frequency combustor mode, as expected.
Figure 17.21 shows the Nyquist plot for the combustor and resonator. Fig-
ure 17.21 can be directly compared with Fig. 17.8a. Note that the resonator has
eliminated the higher-frequency lobe that encircled the −1 point, and so, accord-
ing to this analysis, the combustor would be stable. However, note that the lower
frequency would still be unstable at longer time lags, because the low-frequency
lobe would rotate, as in Fig. 17.8, at longer time lags.
As noted, several parameters are assumed in the resonator design. If the gas
temperature is different than assumed temperature, the resonator natural frequency
will differ from the planned value. To explore this issue, the same calculations are
carried out to compare resonators with natural frequencies of 390, 410, and 430
Hz. The resulting Bode plots are shown in Fig. 17.22. For clarity, the scale is
increased relative to Fig. 17.20, and only the high-frequency range is shown. Note
that all cases have a magnitude less than unity, meaning that the combustion would
be stable. However, if the assumed combustor gain is larger, some new frequencies
568 G. A. RICHARDS ET AL.
Nyquist Plot
3
2
1
Imaginary
0
-1
-2
-3
-3 -2 -1 0 1 2 3
Real
Fig. 17.21 Nyquist plot of open-loop frequency response with the resonator. These
results compare with plots in Fig. 17.8a.
could become unstable. For example, adding a resonator tuned to 390 Hz produces
a peak response of 0.5 at 420 Hz, implying that if the combustor gain rises by a
factor of 2, the system could oscillate at 420 Hz.
Figure. 17.23 shows the effect of moving the axial position of the 410-Hz res-
onator to the axial positions A thru D in Fig. 17.19. Because the resonator ideally
must interact with the pressure antinode near the flame, moving the damper along
the axis reduces the effectiveness. All the gains are still less than unity, implying
stability, but this stability again depends on the assumed combustion gain and other
parameters.
In summary, these example calculations have shown that the design of a damp-
ing resonator is possible by selecting a resonator with a natural frequency equal to
the mode to be damped. The system response is analyzed by inserting the resonator
impedance Z r into the acoustic analysis presented in the Appendix. These calcula-
tions can be repeated to assess the effect of various design parameters, such as the
natural frequency, loss mechanisms, and resonator position. Although predictions
in this example show that the system is stabilized by the damper, care must be
1.0
0.8
Magnitude
0.6
390
0.4
410
0.2 430
0.0
300 400 500
Frequency (Hz)
Fig. 17.22 Effect of resonator tuning on the Bode plot for the higher-frequency
range. Resonators with natural frequencies 390 and 430 Hz are compared with
the baseline (410 Hz).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 569
1.0
0.8
Magnitude
D
C
0.6
B
0.4
A
0.2
0.0
300 400 500
Frequency (Hz)
Fig. 17.23 Effect of resonator position on the Bode plot for the higher-frequency
range. Axial positions A to D are shown in Fig. 17.19.
V. Conclusion
This chapter described various passive-control methods for stabilizing premixed
combustion in gas turbines. Feedback-control models reviewed in Sec. II showed
how both the flame-transfer function and the feedback acoustics are linked in a
dynamics problem. Nyquist analysis illustrated how multiple acoustic modes can
confound approaches to eliminate combustion instabilities by changing the con-
vective time lag. The physical processes affecting the flame response have been
reviewed, and some examples of the complex details of practical flame-transfer
functions were provided. In Sec. III, applications of time-lag modifications were
presented, and the limitations of these approaches were discussed. Experimen-
tal results demonstrating frequency shifts predicted by the Nyquist analysis were
shown. Furthermore, the stabilizing effect of time-lag distribution was discussed,
along with some field applications that confirmed the beneficial effect of increas-
ing the time-lag distributions. Improvements to stability from flame-anchoring
modifications, and pilot flames were also reviewed. In Sec. IV, a review of acous-
tic damping suggested that dampers can be used effectively to stabilize premixed
combustion. A few laboratory-scale and fielded-engine studies were reviewed,
showing excellent attenuation from the dampers. Example calculations showed
the various factors to consider in planning a damper addition and demonstrated
how damper performance can be predicted from Nyquist analysis.
It is interesting to speculate on the potential situations in which dynamics may
be a concern in future gas-turbine engine applications. It is often assumed that
dynamics problems are limited to stationary-gas turbines using premixed combus-
tion. However, two different integrated gasification combined-cycle power plants
recently suffered dynamics problems well into engine commissioning.77 The trend
for most advanced-power generators and aeroengine applications is to raise the op-
erating pressure to enhance efficiency. At the same time, it is desirable to reduce
the cooling and dilution flows to enhance performance or to reduce emissions. The
higher operating pressures release more heat in the same volume, increasing the
magnitude of heat-release rate perturbations. The reduction in dilution or cooling
570 G. A. RICHARDS ET AL.
flows also reduces the acoustic losses from the combustion liner. These combined
features raise the potential for dynamics. Thus, it seems likely that the prominence
of this problem will continue to be an issue. A combination of passive-control
strategies outlined here and emerging active-control strategies offer an opportu-
nity to mitigate these problems as new systems are designed.
Acknowledgments
This work was supported by the U.S. Department of Energy National En-
ergy Technology Laboratory. We acknowledge many helpful discussions with
colleagues at Virginia Polytechnic Institute and State University (Will Saunders,
Bill Baumann, and Uri Vandsburger) who encouraged the use of feedback-control
modeling.
Appendix
This Appendix documents the calculations used in the generation of Figs.
17.5–17.8. The acoustic computations are described first, followed by a deriva-
tion of the response of the system to a velocity source located at the flame and
incorporation of a flame model to yield the H transfer function. Note that
Fuel Flame
A B D E G
Combustion
Air products
0 12 34 56 9
Station #'s
Fig. 17.A1 Generic combustor geometry (not to scale).
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 571
Table 17.A1 Geometry and gas conditions used in the two-mode combustor model
Geometric region
Dual-mode combustor
simulation parameters A+B D E G
Fig. 17.A1. To simplify the calculations, “hard” acoustic boundaries were assumed
to exist at stations 0 and 9 (i.e., the acoustic velocity is zero). The specific geometry
and other conditions used in the calculations for the two-mode system are listed
in Table 17.A1.
The acoustic properties of the combustor model are described by the acoustic
impedances. Impedance is the complex ratio of acoustic pressure to mass velocity
at each station. The impedance at each station is computed through the use of
acoustic transfer matrices. These matrices of complex numbers relate the acoustic
pressure and velocity at one station to the next station. These relations are a function
of the geometry and the local-gas conditions. For example,
p1 p
=C 2 (17.A1)
ν1 ν2
p0 p1
= AB (17.A2)
ν0 ν1
and
p2 p3
=D (17.A3)
ν2 ν3
respective regions. The matrices C and F do not correspond to specific regions but
represent the acoustic transfer matrices for the step expansions that occur between
regions AB and D and regions E and G, respectively. The acoustic pressure and
velocity at a particular station are a linear function of the acoustic pressure and
velocity at the next station downstream. A similar linear relationship does not exist
between stations 3 and 4 because of the presence of the flame, which acts as an
acoustic source.
The 2 × 2 acoustic transfer matrices for various geometries are defined in
Munjal.28 For simple cylindrical elements:
cos 2π f L jc
· sin 2πcf L
c s µ 2π f
R(s, L , µ, f, c) = · exp − j L
js
· sin 2π f L
cos 2πcf L c c
c c
(17.A4)
For step expansions such as the region between stations 1 and 2, the transfer matrix
is
−1 k Ma2 k Ma ζa
1 M b ζb 1− 1−Ma2 1−Ma2
S(ζa , ζb , Ma , Mb , k, γ ) := Mb
1 (γ −1)k Ma3 (γ −1)k Ma2
ζb
(1−Ma2 )ζa
1− 1−Ma2
1 Ma ζa
× Ma (17.A5)
ζa
1
The acoustic impedances at each station in the model are calculated starting from
known boundary conditions. At station 0, ν0 = 0 such that
Z 0 = p0 /ν0 = ∞ (17.A7)
At station 2,
1
p1 C0,0 p2 + C0,1 ν2 ν2 C0,0 Z 2 + C0,1
Z1 = = · = (17.A9)
ν1 C1,0 p2 + C1,1 ν2 1
ν2
C1,0 Z 2 + C1,1
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 573
therefore
−(Z 1 C1,1 − C0,1 )
Z2 = (17.A10)
(Z 1 C1,0 − C0,0 )
Z 9 = p9 /ν9 = ∞ (17.A12)
At station 6,
At station 5,
And at station 4,
E0,0 Z 5 + E0,1
Z4 = (17.A15)
E1,0 Z 5 + E1,1
The flame is modeled as an acoustic velocity source located in the plane be-
tween stations 3 and 4. The boundary conditions at this interface are p3 = p4 and
ν4 = ν3 ∗ (ρ4 /ρ3 ) + νs . The transfer function relating the acoustic pressure to the
velocity source, that is,
p3
ψ= (17.A16)
νs
Therefore,
Z3 Z4
ψ= (17.A18)
Z 3 − ρρ43 · Z 4
574 G. A. RICHARDS ET AL.
Equation 17.A19 provides the last link needed to derive H, the relative pressure
response to oscillations in heat release. Derivation of the H transfer function started
with equation A19 by separating terms involving p3 from ν3 and dividing through
by P3 to yield
p3 1 1 T4 q T4 µ3 s3
− ρ4 − 1 µ3 s 3 = ρ4 −1 (17.A20)
P3 ψ P3 T3 Q T3 P3
Therefore,
A simple form for transfer function G, the relative heat-release response to pres-
sure fluctuations, was assumed to compute the open-loop response function. The
function G is assumed to be a simple time lag. The time-lag model is shown here
with a gain of one half:
q /Q 1
G= = e−1 jwτ (17.A22)
p3 /P3 2
The gain and time lag of function G can be varied according to the situation being
modeled. The equations for G and H are used to compute the open-loop response
functions, G∗H, for the example Bode and Nyquist plots discussed in the body of
the paper.
Finally, for use in the discussion of acoustic dampers, the addition of a side-
branch resonator can be handled28 by adding one more acoustic matrix T that
includes the impedance of the resonator mouth Z r :
1 0
T = (17.A23)
1/Z r 1
This matrix is simply inserted at the interface between the cylindrical matrices
(A4) that comprise the geometry of the combustor. In this manner, the damper can
be easily added, or repositioned by including the matrix T in the multiplication of
matrices that define any of the secs. A, B, D, E, or G in Fig. 17.A1. In the example
calculations presented in the text, the resonator is added to the matrix defining
sec. E.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 575
References
1
Hobson, D. E., Fackrell, J. E., and Hewitt, G., “Combustion Instabilities in Indus-
trial Gas Turbines - Measurements on Operating Plant and Thermoacoustic Modeling,”
Journal of Engineering for Gas Turbines and Power, Vol. 122, No. 3, 2000, pp. 420–
428.
2
Sholz, M. H., and Depietro, S. M., “Field Experience on DLN Typhoon Industrial Gas
Turbines,” American Society of Mechanical Engineers, Paper 97-GT-61, 1997.
3
Pandalai, R. P., and Mongia, H. C., “Combustion Instability Characteristics of Industrial
Engine Dry Low Emission Combustion Systems,” AIAA Paper 98-3379, 1998.
4
Richards, G. A., Straub, D. L., and Robey, E. H., “Passive Control of Combustion
Dynamics in Stationary Gas Turbines,” Journal of Propulsion and Power, Vol. 19, No. 5,
2003, pp. 795–810.
5
Chu, B. T. “On the Generation of Pressure Waves at a Plane Flame Front,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 4., 1953, pp. 603–612.
6
Peracchio, A. A., and Proscia, W. M., “Nonlinear Heat-Release/Acoustic Model for
Thermoacoustic Instability in Lean Premixed Combustors,” Journal of Engineering for
Gas Turbines and Power, Vol. 121, No. 3, 1999, pp. 415–421.
7
Brookes, S. J., Cant, R. S., Dupere, I. D., and Dowling, A. P., “Computational Modeling
of Self-Excited Combustion Instabilities,” American Society of Mechanical Engineers,
Paper 2000-GT-0104, 2000.
8
Kruger, U., Huren, J., Hoffinan, S., Krebs, W., and Bohn. D., “Prediction of Thermoa-
coustic Instabilities with Focus on the Dynamic Flame Behavior for the 3A-Series Gas
Turbine of Siemens KWU,” American Society of Mechanical Engineers, Paper 99-GT-11,
1999.
9
Harrje, D. T., and Reardon, F. H., “Liquid Propellant Rocket Combustion Instability,”
NASA SP-194, 1971.
10
Arana, C. A., Sekar, B., and Mawid, M. A., “Determination of Thermoacoustic Re-
sponse in a Demonstrator Gas Turbine Engine,” Journal of Engineering for Gas Turbines
and Power, Vol. 124, No. 1, 2000, pp. 46–57.
11
Fleifil, M., Annaswamy, A. M., Ghoneim, Z. A., and Ghoniem, A. F., “Response of a
Laminar Flame to Flow Oscillations: A Kinematic Model and Thermoacoustic Instability
Results,” Combustion and Flame, Vol. 106, 1996, pp. 487–510.
12
Steele, R. C., Cowell, L. H., Cannon, S. M., and Smith, C. E., “Passive Control of
Combustion Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
Turbines and Power, Vol. 122, No. 3, 2000, pp. 412–419.
13
Richards, G. A., and Janus, M. C., “Characterization of Oscillations During Premix
Gas Turbine Combustion,” Journal of Engineering for Gas Turbines and Power, Vol. 120,
No. 2, 1998, pp. 294–302.
14
Keller, J. J., “Thermoacoustic Oscillations in Combustion Chambers of Gas Turbines,”
AIAA Journal, Vol. 33, No. 12, 1995, pp. 2280–2287.
15
Lieuwen, T., and Zinn, B. T., “The Role of Equivalence Ratio Oscillations in Driv-
ing Combustion Instabilities in Low NOx Gas Turbines,” Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 27, 1998, pp. 1809–1816.
16
Lee, J. G., Kim, K., and Santavicca, D. A., “Measurement of Equivalence Ratio Fluc-
tuation and Its Effect on Heat-release During Unstable Combustion,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 415–421.
17
Mongia, R. K., Tomita, E., Hsu, F. K., Talbot, L., and Dibble, R. W., “Use of an Optical
Probe for Time-Resolved In Situ Measurement of Local Air-to-Fuel Ratio and Extent of Fuel
576 G. A. RICHARDS ET AL.
Mixing with Applications to Low NOx Emission in Premixed Gas Turbines,” Proceedings
of the Combustion Institute, Pittsburgh, PA, Vol. 26, 1996, pp. 2749–2755.
18
Schadow, K. C., and Gutmark, E., “Combustion Instabilities Related to Vortex Shedding
in Dump Combustors and Their Passive Control,” Progress in Energy and Combustion
Science, Vol. 18, 1992, pp. 117–132.
19
Smith, D. A., “An Experimental Study of Acoustically Excited, Vortex Driven Combus-
tion Instability Within a Rearward Facing Step Combustor,” Ph.D. Dissertation, California
Inst. of Technology, Pasadena, CA, 1985.
20
Sterling, J. D., and Zukoski, E. E., “Longitudinal Mode Combustion Instabilities in a
Dump Combustor,” AIAA Paper 87-0220, 1987.
21
Gutmark, E., Schadow, K. C., Sivasegaram, S., and Whitelaw, J. H., “Interaction Be-
tween Fluid-Dynamic and Acoustic Instabilities in Combusting Flows Within Ducts,” Com-
bustion Science and Technology, Vol. 79, 1991, pp. 161–166.
22
Poinsot, T. J., Trouve, A. C., Veynante, D. P., Candel, S. M., and Esposito, E. J., “Vortex-
Driven Acoustically Coupled Combustion Instabilities,” Journal of Fluid Mechanics,
Vol. 177, 1987, pp. 265–292.
23
Langhorne, P. J., “Reheat Buzz: An Acoustically Coupled Combustion Instability,
Part 1, Experiment,” Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
24
Yu, K. H., Trouve, A., and Daily, J. W., “Low-Frequency Pressure Oscillations in a
Model Ramjet Combustor,” Journal of Fluid Mechanics, Vol. 232, 1991, pp. 47–72.
25
Schuermans, B. H., and Polifke, W., “Modeling Transfer Matrices of Premixed Flames
and Comparison with Experimental Results,” American Society of Mechanical Engineers,
Paper 99-GT-132, 1999.
26
Kendrick, D. W., Anderson, T. J., Sowa, W. A., and Snyder, T. S., “Acoustic Sensitivities
of Lean-Premixed Fuel Injectors in a Single Nozzle Rig,” Journal of Engineering for Gas
Turbines and Power, Vol. 121, No. 3, 1999, pp. 429–436.
27
Pascheriet, C. O., Flohr, P., Knopfel, H., Geng, W., Steinbach, C., Stuber, P., Bengtsson,
K., and Gutmark, E., “Combustion Control by Extended EV Burner Fuel Lance,” American
Society of Mechanical Engineers, Paper GT-2002-30462, 2002.
28
Munjal, M. L., Acoustics of Ducts and Mufflers, Wiley, New York, 1987.
29
Fannin, C. A., Baumann, W. T., Saunders, W. R., Richards, G. A., and Straub, D. L.,
“Thermoacoustic Stability Analysis for Mutli-Port Fuel Injection in a Lean Premixed Com-
bustor,” AIAA Paper 2000-0711, 2000.
30
Phillips, C. L., and Harbor, R. D., Feedback Control Systems, 4th ed., Prentice–Hall,
Upper Saddle River, NJ, 2000.
31
Blackshear, P. L. “Driving Standing Waves by Heat Addition,” Proceedings of the
Combustion Institute, Pittsburgh, PA, Vol. 4, 1953, pp. 553–556.
32
Merk, H. J. “An Analysis of Unstable Combustion of Premixed Gases,” Proceedings
of the Combustion Institute, Vol. 6, 1956, pp. 501–512.
33
Becker, R., and Gunther, R., “The Transfer Function of Premixed Turbulent Jet Flames,”
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 13, 1971, pp. 517–526.
34
Baade, P. K., “Design Criteria and Models for Preventing Combustion Oscillations,”
Transactions of the American Society of Heating, Refrigerating and Air-Conditioning En-
gineers, ASHRAE 84, Part 1, 1978.
35
Mugridge, B. D., “Combustion Driven Oscillations,” Journal of Sound and Vibrations,
Vol. 70, No. 3, 1980, pp. 437–452.
36
Matsui, Y., “An Experimental Study on Pyro-Acoustic Amplification of Premixed
Laminar Flames,” Combustion and Flame, Vol. 43, 1981, pp. 199–209.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 577
37
Ducruix, S., Durox, D., and Candel, S., “Theoretical and Experimental Determination
of the Transfer Function of a Laminar Premixed Flame,” Proceedings of the Combustion
Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 765–773.
38
Hadvig, S., “Combustion Instability: System Analysis,” Journal of the Institute of Fuel,
Vol. 44, 1971, pp. 550–558.
39
Hegde, U. G., Reuter, D., Daniel, B. R., and Zinn, B. T., “Flame Driving of Longitudinal
Instabilities in Dump Type Ramjet Combustors,” Combustion Science and Technology,
Vol. 55, 1987, pp. 125–138.
40
Fieler, C. E., and Heidmann, M. F., “Dynamic Response of Gaseous Hydrogen
Flow Systems and Its Application to High-Frequency Combustion Instability,” NASA TN
D-4040, 1967.
41
Putnam, A. A., Combustion Driven Oscillations in Industry, Elsevier, New York, 1971.
42
Janardan, B. A., Daniel, B. R., and Zinn, B. T., “Driving of Combustion Oscillations
by Gaseous Propellant Injectors,” Proceedings of the Combustion Institute, Pittsburgh, PA,
Vol. 17, 1977, pp. 1353–1361.
43
Lieuwen, T., and Zinn, B. T., “Theoretical Investigation of Premixed Combustion In-
stability Mechanisms,” AIAA Paper 98-0641, 1998.
44
Richards, G. A., Straub, D. L., and Robey, E. H., “Dynamic Response of A Premix
Fuel Injector,” American Society of Mechanical Engineers, Paper 2001-GT-036, 2001.
45
Scarinci, T., and Freeman, C., “The Propagation of a Fuel-Air Ratio Disturbance in a
Simple Premixer and Its Influence on Pressure Wave Amplification,” American Society of
Mechanical Engineers, Paper 2000-GT-0106, 2000.
46
Sattlelmayer, T., “Influence of the Combustor Aerodynamics on Combustion Instabil-
ities From Equivalence Ratio Fluctuations,” American Society of Mechanical Engineers,
Paper 2000-GT-0082, 2000.
47
Anderson, T. J., Sowa, W. A., and Morford, S. A., “Dynamic Flame Structure in a Low
NOx Premix Combustor,” American Society of Mechanical Engineers, Paper 98-GT-568,
1998.
48
Paschereit, C. O., Polifke, W., Schuermns, B., and Mattson, O., “Measurement of
Transfer Matrices and Source Terms of Premixed Flames,” Journal of Engineering for Gas
Turbines and Power, Vol. 124, No. 2, 2002, pp. 239–247.
49
Lawn, C. J., “Thermo-Acoustic Frequency Selection by Swirled Premixed Flames,”
Proceedings of the Combustion Institute, Pittsburgh, PA, Vol. 28, 2000, pp. 823–830.
50
Krebs, W., Hoffman, S., Prade, B., Lohrmann, M., and Buchner, H., “Thermoacoustic
Flame Response of Swirl-Stabilized Flames,” American Society of Mechanical Engineers,
Paper GT-2002-30065, 2002.
51
Khanna, V. K., “A Study of the Dynamics of Laminar and Turbulent Fully and Par-
tially Premixed Flames,” Ph.D. Dissertation, Virginia Polytechnic Inst. and State Univ.,
Blacksburg VA, 2001.
52
Khanna, V., Vandsburger, U., Saunders, W. R., and Baumann, W. T., “Dynamic Analysis
of Swirl Stabilized Turbulent Gaseous Flame,” American Society of Mechanical Engineers,
Paper GT-2002-30061, 2002.
53
Bohn, D., Deutsch, G., and Kruger, U., “Numerical Prediction of the Dynamic Behav-
ior of Turbulent Diffusion Flames,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 4, 1998, pp. 713–720.
54
Bohn, D., Li, Y., Matouschek, G., and Kruger, W., “Numerical Prediction of the Dy-
namic Behavior of Premixed Flames Using Systematically Reduced Multi-Step Reaction
Mechanisms,” American Society of Mechanical Engineers, Paper 97-GT-265, 1997.
578 G. A. RICHARDS ET AL.
55
Kruger, U., Hoffman, S., Krebs, W., Judith, H., Bohn, D., and Matouschek, G., “Influ-
ence of Turbulence on the Dynamic Behavior of Premixed Flames,” American Society of
Mechanical Engineers, Paper 98-GT-323, 1998.
56
Smith, C. E., and Leonard, A. D., “CFD Modeling of Combustion Instability in Pre-
mixed Axisymmetric Combustors,” American Society of Mechanical Engineers, Paper 97-
GT-305, 1997.
57
Zhu, M., Dowling, A. P., and Bray, K. N. C., “Flame Transfer Calculations for Com-
bustion Oscillations,” American Society of Mechanical Engineers, Paper 2001-GT-0374,
2001.
58
Walz, G., Krebs, W., Hoffmann, S., and Judith, H., “Detailed Analysis of the Acous-
tic Mode Shapes of an Annular Combustion Chamber,” American Society of Mechanical
Engineers, Paper 99-GT-113, 1999.
59
Cronemyr, P. J. M., Hulme, C. J., and Troger, C., “Coupled Acoustic-Structure Analysis
of an Annular DLE Combustor,” American Society of Mechanical Engineers, Paper 98-GT-
502, 1998.
60
Pankiewitz, C., and Sattelmayer, T., “Time Domain Simulation of Combustion Instabil-
ities in Annular Combustor,” American Society of Mechanical Engineers, Paper GT-2002-
30063, 2002.
61
Evesque, S., and Polifke, W., “Low-Order Acoustic Modeling for Annular Combustors:
Validation and Inclusion of Modal Coupling,” American Society of Mechanical Engineers,
Paper GT-2002-30064, 2002.
62
Krebs, W., Walz, G., Flohr, P., and Hoffmann, S., “Modal Analysis of Annular Com-
bustors: Effect of Burner Impedance,” American Society of Mechanical Engineers, Paper
2001-GT-0042, 2001.
63
Straub, D. L., and Richards, G. A., “Effect of Fuel Nozzle Configuration on Premix
Combustion Dynamics,” American Society of Mechanical Engineers, Paper 98-GT-492,
1998.
64
Hermsmeyer, H., Prade, B., Gruschka, U., Schmitz, U., Hoffmann, S., and Krebs, W.,
“V64.3A Gas Turbine Natural Gas Burner Development,” American Society of Mechanical
Engineers, Paper GT-2002-30106, 2002.
65
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism of Combustion
Instability in Lean Premixed Gas Turbine Combustors,” Journal of Engineering for Gas
Turbines and Power, Vol. 123, No. 1, 2001, pp. 182–189.
66
Lovett, J. A., and Uznanski, K. T., “Prediction of Combustion Dynamics in a Staged
Premixed Combustor,” American Society of Mechanical Engineers, Paper GT-2002-30646,
2002.
67
Berenbrink, P., and Hoffmann, S. “Suppression of Dynamics Combustion Instabilities
by Passive and Active Means,” American Society of Mechanical Engineers, Paper 2000-
GT-0079, 2000.
68
Richards, G. A., Straub, D. L., and Robey, E. H. “Control of Combustion Dynamics
Using Fuel System Impedance,” American Society of Mechanical Engineers, Paper GT-
2003-38521, 2003.
69
Lee, J. G., Kim, K., and Santavicca, D. A., “A Study of the Role of Equivalence Ratio
Fluctuations During Unstable Combustion in a Lean Premixed Combustor,” AIAA Paper
2002-4015, 2002.
70
Lovett, J. A. , “Bluffbody Flameholders for Low-Emission Gas Turbine Combustors,”
U.S. Patent No. 5,471,840, 1995.
PASSIVE CONTROL OF COMBUSTION IN GAS TURBINES 579
71
Straub, D. L., Richards, G. A., Baumann, W. T., and Saunders, W. R., “Measurement of
Dynamics Flame Response in a Lean Premixed Single-Can Combustor,” American Society
of Mechanical Engineers, Paper 2001-GT-0038, 2001.
72
Laudien, E., Pongratz, R., Pierro, R., and Preclik, D., “Experimental Procedures Aiding
the Design of Acoustic Cavities,” Liquid Rocket Engine Combustion Instability, Progress
in Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 377–399.
73
Lewis, G. D., and Garrison, G. D., “The Role of Acoustic Absorbers in Preventing
Combustion Instability,” AIAA Paper 71-699, 1971.
74
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combus-
tion System Damping Augmentation with Helmholtz Resonators,” American Society of
Mechanical Engineers, Paper 98-GT-268, 1998.
75
Bellucci, V., Paschereit, C. O., Flohr, P., and Magni, F., “On the Use of Helmholtz
Resonators for Damping Acoustic Pulsations in Industrial Gas Turbines,” American Society
of Mechanical Engineers, Paper 2001-0039, 2001.
76
Mitchell, C. E., “Analytical Models for Combustion Instability in Liquid Rocket En-
gine Combustion Instability,” Liquid Rocket Engine Combustion Instability, Progress in
Astronautics and Aeronautics, AIAA, Washington, DC, Vol. 169, 1995, pp. 403–430.
77
DeBiasi, V., “Gasification on Track to Turn Problem Fuels into Electricity and Power,”
Gas Turbine World, Nov–Dec, 1999, pp. 12–20.
Chapter 18
Jeffrey M. Cohen∗
Pratt & Whitney, East Hartford, Connecticut
and
Andrzej Banaszuk†
United Technologies Research Center, East Hartford, Connecticut
Nomenclature
G0 = transfer function representing combustor, fuel line, and valve
dynamics
Gc = controller transfer function
N = random-input describing function
S = sensitivity function
b = “on” level for solenoid valve
k = exponent determining rolloff of open-loop transfer function
outside control bandwidth
ω1 = performance bandwidth
ω2 = control bandwidth
ii = power-spectral density function of the input disturbance
pp = power-spectral density function of the combustor pressure
ε = required attenuation level for sensitivity function over
performance bandwidth
φ = fuel–air equivalence ratio
µ = mean fuel–air ratio
ω = frequency
ωb = lower-boundary control bandwidth
ωc = higher-boundary control bandwidth
σ = standard deviation of valve command
σr = real part of unstable pole
τ = delay
581
582 J. M. COHEN AND A. BANASZUK
I. Introduction
A. Combustor Dynamics
The character of the pressure oscillations to be controlled is important to the
effectiveness of the control system. This effectiveness includes the fundamental
issue of whether the system is dynamically unstable (i.e., in a limit cycle) or whether
it is linearly stable and driven by noise. Large, coherent pressure oscillations are
possible in either case. Another important factor is the mechanism (or mechanisms)
through which the unsteady heat release and pressure couple with each other. To
arrive at an effective control system, it is important to understand and prioritize
these mechanisms to be able to devise actuation schemes to interfere with them. If
multiple mechanisms are at play, it may be difficult to deal with more than one at
once. Some mechanisms may not lend themselves to practical actuation schemes or
may be better suited to passive-control approaches, such as acoustic resonators.12
Multiple-instability modes and the interaction between those modes (especially
nonlinear interactions) may also compromise the performance of a control system.
Further issues with how the combustor pressure oscillations behave as a function
of engine-operating conditions and engine transients will also affect controller
performance.
B. Actuation System
For the actuation system to be truly effective, it must interfere with the root-
cause physics that lie behind the instability. The most important consideration in
actuation-system design is developing an understanding of what these physics are.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 583
task if the combustor’s operating conditions were fixed; the transport delay were
small (much less than the acoustic period); the actuator authority and bandwidth
were adequate; and a model of pressure response to the fuel valve command
were available. In such case, the control algorithm would need to provide an
appropriate phase shift of the pressure signal. An appropriate phase shift could be
provided by using several approaches,2,3,9,11,14−16 including the time delay, lead
lag, linear quadratic regulator, H infinity (minimizing gain from disturbance energy
to pressure energy), H 2 (minimizing gain from disturbance energy to maximum
value of pressure), and observer-based controllers.
The first obstacle to model-based control design is lack of accurate physics-
based predictive models for combustion-system response. Therefore, the control
design typically utilizes either a model obtained from or calibrated with experimen-
tal data2,3,9,11,14,15 or an adaptive scheme17−20 to automatically tune the controller
parameters in a way that reduces pressure oscillations. In particular, an adaptive
scheme is needed as a preliminary control algorithm in the case in which obtaining
a data-based model is not practical. This situation would be true of an unstable in-
dustrial combustor operating in a regime where hardware damage is likely or when
operating conditions vary (as in power transients for industrial gas turbines) and
are subject to unknown disturbances (external temperature, power load changes).
Both fixed-parameter and adaptive-control approaches have their own limitations,
and will be discussed next.
As mentioned earlier, the level of suppression achieved with a fixed-parameter
controller varied between various experiments. We will explain, using methods of
control theory, how large transport delay (comparable with the acoustic period)
and limited actuator bandwidth reduce the achievable attenuation level of pressure
oscillations. In essence, in the presence of a large delay, attenuation of pressure os-
cillations at certain bands of frequencies is accompanied by the excitation of oscil-
lations in adjacent bands.15,21,22 Limited actuator bandwidth prevents the possibil-
ity of compensation for this problem in the control algorithm. The tradeoff between
the attenuation of oscillations in certain frequency bands and excitation in adjacent
bands is expressed in terms of a controller-independent lower bound15,21,23,24
on the function that shows maximum-pressure-magnitude magnification over the
excitation band caused by the controller. The lower bound is an increasing function
of the transport delay and a decreasing function of the actuator bandwidth.15,21
The limitations of adaptive-control algorithm performance include the limita-
tions of the fixed-parameter controllers with few extra limitations introduced by the
adaptation. First, unless the combustor transient timescale is an order of magnitude
slower than the adaptation timescale, the stability of any adaptive scheme cannot
be guaranteed.25−27 (For some ad hoc adaptive schemes, no stability guarantees
exist even under the assumption of the timescale separation.) Unfortunately, one
cannot arbitrarily decrease the control-parameter adaptation timescale to achieve
the timescale separation. One factor that limits the speed of adaptation, especially
in industrial applications, is the noise present in the pressure time traces,17 which
can be attributed to the response of the acoustic modes to random disturbances
(such as turbulence). This noise needs to be filtered out so that the control algo-
rithm can distinguish the pressure reaction to the control input from a response to
random disturbances. The presence of the noise filters necessarily slows down the
speed of adaptation because the time required to average out the effect of noise
is proportional to the noise-to-signal ratio.17,28 The reduced speed explains why
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 585
Burst disk
Bypass leg
Choked orifice
Choked plate
Heated, high venturi
pressure air
Emissions Probe
Field of View
Air Supply
Premixer (3) Combustion
Products
Fiber Optic
Optical Emissions
Probe (3)
Fig. 18.2 Cross section of a three-premixer sector combustor test facility with an
instrumentation and actuation system.
586 J. M. COHEN AND A. BANASZUK
Fig. 18.3 Schematic illustration of premixing fuel nozzle. Top picture shows tangen-
tial air scrolls and gaseous-fuel-injection scheme. Bottom picture shows nozzle cross
section with a liquid-fuel-injection scheme.
conditions (nominally 710 K and 1.5 MPa, respectively). The experiments used
similar embodiments of the same engine-scale premixing nozzles. The three-nozzle
sector rig used a 60-deg-arc sector of the engine combustor liner with convec-
tively cooled sidewalls. The combustor rigs are discussed in detail in Refs. 1, 2,
and 29.
The premixing fuel injector used in these combustors has been described in
detail by Stufflebeam et al.30 The fuel nozzle is shown schematically in Fig. 18.3.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 587
Air was delivered into the premixing chamber through two tangentially oriented air
slots that ran the entire axial length of the chamber. Natural gas fuel was injected
through a row of orifices in the inlet section to each of these air slots. Fuel–air
mixing was measured and optimized for a low-emissions operation, as described
by Stufflebeam et al.30 The liquid-fuel version of the injector used a series of six
axial spokes to atomize and inject fuel in the interior of the premixer.
The instability mode (∼200 Hz) to which control systems were applied was a
Helmholtz mode (n = 0), in which the fluctuating pressure was uniform within
the combustor. Pressure fluctuations were coupled with the heat-release process
through their effect on the flow rate of air delivered through the premixer. The
time-varying air-flow rate produced a time-varying equivalence ratio at the fuel
nozzle exit and, therefore, a time-varying heat-release rate. This conceptual model
of the instability was discussed in more detail by Peracchio and Proscia31 and the
phenomenon has been described by many other authors.5,6,32 Fuel-concentration
measurements performed by Lee and Anderson33 in this combustor confirmed this
link between equivalence-ratio fluctuations and pressure fluctuations. Figure 18.4
shows the spectrum of the fluctuating pressure for this instability, as observed in
the single-nozzle (flame-tube) version of the combustor.
The fundamental problem of control-system design for low-emissions combus-
tors is to maximize the system’s authority over the relevant dynamic processes
while minimizing the combustor’s emissions and ensuring this performance over a
range of operating conditions. The active-control system consisted of three parts: a
pressure sensor, a control algorithm, and an actuator. Because of the uniform spatial
distribution of the fluctuating pressure within the combustor, only one combus-
tor pressure measurement was required to describe the unsteady pressure field as
input to the control system. In both experiments, on/off actuation of a portion of
the fuel flow to the premixer was performed. A closed-loop control algorithm was
developed to use the actuators’ authority to damp combustor pressure oscillations.
14
12
10
Amplitude (psi)
0
0 200 400 600 800 1000
Frequency (Hz)
Uncontrolled Controlled
Combustor
pressure
Premixed
equivalence
ratio
Time
Fig. 18.5 Illustration of the intent of the actuation technique, utilizing the reduction
of equivalence-ratio fluctuations by pulsed fuel injection.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 589
likely require on-line adaptation of the amplitude of the fuel flow modulation. Only
fixed-amplitude (on/off) actuation was used in this study.
The gas–fuel premixer was modified to incorporate four different actuated fuel-
injection configurations, as shown in Figs. 18.6 and 18.7. These injection config-
urations were designed specifically to provide different levels of mixing between
the actuated fuel flow and the remainder of the premixed reactants. Three of the
configurations used an axially oriented spoke mounted on the premixer centerbody.
The length of the spoke and the number of injection sites were varied to modify
mixing. The fourth configuration modulated the flow through two of the injection
orifices in the main fuel-injection array at the inlet to the air scroll. In all cases,
the level of actuated fuel flow was held constant.
Fig. 18.6 Cross section schematic illustration of the premixing fuel nozzle with an
actuated fuel-injection spoke (configurations 1–3, top) and actuated portion of the
main injection array (configuration 4, bottom) for introduction of controlled fuel flow.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 591
the temporal character of the fuel–air ratio at the exit of the premixer without
excessively degrading the spatial fuel–air mixing at any point in time.
Fig. 18.8 Steady-state fuel–air concentration profiles at premixer exit for four fuel-
injection concepts. Concentration values have been normalized by the mean value for
each case.
592 J. M. COHEN AND A. BANASZUK
flow rate was necessary to truly isolate the effect of the unsteady aspect of the fuel
injection.
Three of the poorer-mixing injection concepts showed increased NOx and CO
levels (relative to baseline levels) during operation with steady, nonmodulated flow
through the control fuel system. These levels did not change appreciably when
control was applied (for the same mean flow through the control fuel system). In
general, the level of NOx increase that was observed correlated with the degree
of unmixedness that was observed. Emissions were measured by using a ganged
array of six water-cooled sampling probes near the exit of the combustor.
For each of these injection configurations (1–3), a piloting effect on the instabil-
ity was also observed. When a constant fuel flow was delivered through the control
fuel system, the local enriching resulted in an attenuation of the pressure fluctua-
tions. This behavior is similar to that observed when a diffusion-flame pilot was
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 593
Fig. 18.10 Variation of spatially averaged fuel–air ratio at a premixer exit over one
period (5 ms) of 200-Hz fuel flow modulation for configuration 4. Bars indicate spatial
variance with about average value at each time.
applied to the system. When piloted, the pressure-fluctuation levels were reduced
but at the expense of higher NOx emissions.
Configuration 4, in which the control fuel was delivered through a portion of
the main fuel-injection array, showed no difference in either pressure fluctuations
or emissions between the baseline case and the case in which steady fuel flow was
delivered through the control fuel system (as expected).
Figure 18.11 shows that the level of reduction in the amplitude of the dominant
instability mode correlated with the mixing performance when amplitude was
18
Reduction in amplitude of instability (dB)
16 Configuration 4
14
12
Configuration 2
10
Configuration 3
8
4
Configuration 1
2
0 0.05 0.1 0.15 0.2
Unmixedness
Fig. 18.11 The effect of steady-state fuel–air mixing on actuator authority in com-
bustion tests for four different actuated fuel-injection configurations.
594 J. M. COHEN AND A. BANASZUK
A. Experimental Observations
Speculating that the effectiveness of the controller could be increased with the
additional actuator authority produced by actuating more nozzles, the control sys-
tem was tested with multiple, simultaneously actuated fuel nozzles. Figure 18.12
shows power-spectral density (PSD) plots for phase-optimized (minimum pressure
oscillations) single, dual, and triple closed-loop, controlled fuel nozzles. Com-
bustor pressure oscillations were reduced by going from single- to dual-nozzle
actuation, but no further reduction was obtained by actuating all three fuel noz-
zles, despite the open-loop forcing results. The best control was achieved with
dual-nozzle actuation, yielding a 6.5 dB (2.1 times or 53%) reduction in the bulk
mode pressures and a 25% reduction in broadband rms pressure. These reductions
in combustor pressure oscillations via active control were accompanied with no
penalties to emissions compared with the uncontrolled operation. The magnitude
of the reduction was limited by the splitting of the spectral peak into two smaller
peaks. This splitting behavior was evident for both two- and three-nozzle actu-
ation, but the amplitude of the secondary peaks was larger for the three-nozzle
case.
In later experiments with the gas-fueled, single-nozzle flame-tube combustor,
the length of tubing between the actuation fuel valve and the fuel-injection location
was changed from 2.8 cm (1.1 in.) to 15.2 cm (6 in.) and 45.7 cm (18 in.). The ef-
fect of these variations on the control-system performance is shown in Fig. 18.13.
The negative effects of peak splitting increased as the fuel line was made longer.
Although no direct measurements of time delay were made for the different con-
figurations, it is reasonable to assume that increasing the fuel-line length increased
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 595
1.6
uncontrolled
1.4 single nozzle
no control
dual nozzle
Combustor Pressure (psi)
1.2
triple nozzle
1
single nozzle triple nozzle
0.8 dual nozzle
0.6 triple nozzle
dual nozzle
0.4
0.2
Frequency, Hz
Fig. 18.12 Multiple-nozzle, closed-loop actuation led to relatively small incremental
reductions in pressure-fluctuation levels because of the peak-splitting phenomenon.
the time delay of the actuation system. This phenomenon has been experimentally
observed by other investigators, as well.7
B. Dynamic Interpretation
The explanation for this phenomenon traces back to the fundamental nature
of the pressure oscillations observed in the combustor. Pressure oscillations in a
combustor dominated by a narrow-frequency band can be interpreted by using
a limit-cycling model or a stable, noise-driven model. Most references attribute
pressure oscillations in combustors to self-excitation of coupled acoustics and
heat-release systems, resulting in a limit-cycling behavior. However, it will be
shown that the uncontrolled-sector combustor behavior and the splitting of the
bulk-mode peak observed during controlled operation can be better explained
with a model of the combustor as a lightly damped, linearly stable system driven
by noise attributed to turbulence. Note that important differences exist between
behavior observed in small laboratory combustion-control experiments and full-
scale industrial combustors. Laboratory combustors typically have no liner and
thus have lower damping than industrial combustors. At the same time, laboratory
combustors may have lower turbulence levels than larger, more complex devices.
With low-damping and low-noise levels, it is likely that significant pressure oscil-
lations will only occur because of self-excited limit-cycle oscillations.14 Industrial
combustors can exhibit noticeable pressure oscillations in a stable, noise-driven
regime; hence, a self-excited model is, in many cases, not necessary. In this sense
the term “combustion instability” is less appropriate for this analysis, because it
will use a stable model of the sector rig and add a driving broadband stochastic
2.5
1-in.; 3.6x
in.;3.6x
1.11-in.; 3.6X
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
2.5
6 6-in.;
in.; 2.6X
6-in.; 2.6x
2.6x
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
2.5
1818-in.;
in.; 1.7X
18-in.; 1.7x
1.7x
Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 18.13 Combustor pressure spectra showing the reductions in peak amplitude of
optimum-phase control for three different lengths of fuel manifold. Results indicate
a reduction in control system authority because of the appearance of dual peaks at
longer tubing lengths. Solid lines represent controlled cases. Dashed lines represent
uncontrolled cases.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 597
nd
0.015 ì 2nd
“2 Order”
Orderî system
system
0.01
0.005
0
100 120 140 160 180 200 220 240 260 280 300
Frequency, Hz Measured Response:
Response:
500
time delay
time delay factored
factoredout
out
nd
ì“22nd Order”
Orderî fit
Phase
-500
Phase Response
dominated by delay
-1000
100 120 140 160 180 200 220 240 260 280 300
Frequency, Hz
Fig. 18.14 Bode plot of combustor pressure over valve-command signal with no
control.
800
700
Number of samples
600
500
400
300
200
100
0
-6 -4 -2 0 2 4 6
Pressure (psi)
Fig. 18.15 Distribution of 20,000 samples of uncontrolled, unsteady combustor pres-
sure (thin line) and a fit with a Gaussian distribution (thick line).
lag caused by delay was subtracted from the experimental phase lag, yielding a
nearly classical second-order response with the phase dropping 180 deg through
the magnitude-response peak. A stable second-order transfer function with two
poles and one zero was fitted numerically.
A schematic of the closed-loop simulation block diagram is shown in Fig. 18.16.
The plant G 0 ( jω) is the empirical second-order system with delay representing
the combustor dynamics. With the controller off, the standard deviation of the
white Gaussian noise was adjusted in the simulation to match the PSD of the
experimental pressure data. A likely physical source of the driving disturbance
was the turbulent flow fluctuations driving the acoustic mode directly or through
the heat-release process.
The effect of multiple-nozzle actuation was simulated by linearly scaling the
controller output by the number of nozzles, which was consistent with experimental
open-loop forcing results. Figure 18.17 shows that the simulation exhibited a peak-
splitting phenomenon similar to the phenomena observed in the experiments. The
amplitude at the dominant oscillation frequency was attenuated, whereas secondary
Plant
Noise + Pressure
Go((jω)
jω)
Fuel flowl
-
modulation
b Gc((jω)
jω)
-b
peaks were amplified by actuating more nozzles and therefore more authority. The
slight asymmetry of peaks after the third valve was turned on in the experiment
can be attributed to a different phase lag of the third actuation system relative to
the first two systems. In fact, this asymmetry was reproduced in the model by
assigning a larger delay to valve 3 than to valves 1 and 2 in the model.
Even though a linear, stable system driven by a white Gaussian disturbance
was a good model of the sector combustor during the experiments, the use of
on/off valves for control made the closed-loop system strongly nonlinear. Thus,
an analysis of peak splitting by using linear control-theory tools14 may not seem
immediately relevant. However, Banaszuk and colleagues21,22 have argued that
a quasi-linear analysis using random input-describing functions is appropriate to
study the nonlinear dynamics of the combustion model with on/off valves in the
presence of large-amplitude noise, as with this closed-loop model. In this tech-
nique, the signals in the model were approximated as sums of constant, sinusoidal,
and random components with Gaussian distribution. The static nonlinear elements
were replaced with equivalent gains called random input-describing functions.36
The values of constant components, amplitudes, and frequencies of sinusoidal
components and standard deviations of Gaussian components in the system can
be found by solving a system of nonlinear equations.
It can be shown that the system of Fig. 18.16 with the identified standard devia-
tion of Gaussian input has low effective gain of the on/off valve for the sinusoidal
signal so that a limit-cycle oscillation cannot be sustained. Therefore, only the
600 J. M. COHEN AND A. BANASZUK
which can be solved numerically. Once the value of σ is known, the pressure PSD
pp ( jω) can be obtained from the formula
Fig. 18.18 Nyquist diagram for single-nozzle, closed-loop control near the optimum-
control phase, showing that the controller excited secondary peaks (B and C) and
attenuated the primary peak (A).
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 601
the controller G C ( jω) used in the experiments and in the simulation was a nonlinear
phase-shifting controller based on a frequency-tracking extended kalman filter,
which does not have a simple linear-transfer function, even though for a fixed
central frequency of oscillations it can be closely approximated by a linear-transfer
function.]
By analyzing the Nyquist diagram shown in Fig. 18.18, we conclude that the
controller amplifies certain frequency bands while attenuating the pressure oscil-
lations at the frequency band centered at about 208 Hz (peak A in Fig. 18.17). The
two nearly symmetric branches of the Nyquist plot that cross into the unit-radius
circle, centered at (−1, 0), for frequencies greater than 225 Hz and less than 195
Hz were the root cause of the secondary peaks B and C in Fig. 18.17. They arise
because of presence of large delay in the combustor-transfer function causes signif-
icant rolloff of the phase of the open-loop transfer function G 0 ( jω)G C ( jω)N (σ ).
Although adding more nozzles and therefore more actuator authority increased
the control gain in the attenuation band, it also increased the control gain in the
excitation band, imposing a limit on the phase-shifting controller’s effectiveness.
Note, however, that increasing the number of actuated nozzles does not correspond
to a proportional gain increase, because the standard deviation σ of the Gaussian
process at the input of the valve is a function of b.
where
1
S( jω) := (18.5)
1 + G 0 ( jω)G c ( jω)
is called the sensitivity function. Note that the square of the sentitivity function is
the factor by which the pressure PSD in the combustor is reduced (or amplified)
at any given frequency. The objective for active control of combustion is to shape
the sensitivity function so that it is small at and near the resonant frequency ωr of
602 J. M. COHEN AND A. BANASZUK
where ω1 is the so-called performance bandwidth, that is, the interval containing
the resonant frequency ωr over which reduction of pressure oscillations by the fac-
tor of ε relative to uncontrolled level is enforced. The fundamental limitations23,24
yield controller-independent lower bounds on the maximum of the sensitivity func-
tion. Assume that the combustor response-transfer function G 0 ( jω) has, at most,
one unstable complex conjugate pole pair with the real part denoted by σr . If
the combustor model is stable, we define σr = 0. An example of fundamental
limitations is the Bode integral formula for the sensitivity function:
∞
ln |S( jω)|dω = 2πσr (18.7)
0
This equation shows that the negative area under the logarithm of the abso-
lute value of the sensitivity function (corresponding to attenuation of pressure
oscillations relative to an uncontrolled combustor) in one frequency band must be
accompanied by a positive area (amplification of pressure oscillations) in some
other band. If the control bandwidth is infinite, the positive area may be distributed
over a wide frequency range so that amplification at any given frequency may be
designed to be arbitrarily small. However, if the control bandwidth is finite be-
cause of factors such as actuator bandwidth [so that G 0 ( jω)G C ( jω) is close to
zero beyond certain low and high frequencies], the positive area would have to be
accommodated in a smaller band (where loop gain is high), and this accomoda-
tion would necessarily result in peaking of the sensitivity function. If the peaking
occurs in the region in which the combustor response-transfer function has a non-
vanishing gain, the peaking in the sensitivity function will result in a peak splitting
in the closed-loop response. Figure 18.19 and 18.20 illustrate this phenomenon.
2.5
Combustor Pressure (psi)
1.5
0.5
0
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
8
6
4
2
+ +
dB 0
-
-2
-4
-6
-8
-10
0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz)
Fig. 18.20 A typical sensitivity function, showing the sensitivity tradeoffs caused by
finite-controller performance bandwidth.
ω 1+k
c
|G 0 ( jω)G c ( jω)| ≤ δ for ω > ωc (18.8)
ω
Here, it is assumed that δ < 1/2 and k > 0 (relative degree of at least two). We
impose a similar constraint on the loop gain:
1+k
ω
|G 0 ( jω)G c ( jω)| ≤ δ for ω < ωb (18.9)
ωb
(Control bandwidth)
|S(j
|S(jω)| ∆ω2
1
Loop gain small at Loop gain rolls off
low frequencies at high frequencies
ε
ωb ωr ωc Frequency
0
∆ω 1
(Performance bandwidth)
1 1 3δωc 1
2π σr + ω1 log − ωb log − 1− (18.10)
ε 1−δ 2k (1 + π/τ ωc )
The preceding formula shows that the factors that bound the supremum of the
sensitivity function from below are as follows:
5
4.5 ∆ω1 = 40 Η z
||S|| 4 δ = 0.1 (20 dB)
8
3.5 σr= 0
3 20 dB
2.5
2
15 dB
1.5
1 10 dB
2 3 4 5 6 7 8 9 10
∆ω 2 /∆ω1
Fig. 18.22 Lower bounds on sensitivity-function norm as function of control band-
width for three values of ε.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 605
2
∆ω1 = 40Η z
1.8
ε = .3 (10 d B)
1.6
δ = 0.1 (20 dB)
1.4 τ = 43 ms
||S|| σr= 0
τ = 4.3 ms
8
1.2
1 τ = 0.6 ms
0.8 τ = 0 (no delay)
0.6
2 3 4 5 6 7 8 9 10
∆ω 2/ ∆ω1
Fig. 18.23 Lower bounds on sensitivity-function norm as function of control band-
width for four values of τ .
4) The combustion response delay τ . One can verify that the lower bound on the
sensitivity peak is an increasing function of the delay (assuming other parameters
are fixed).
Using the inequality on the sensitivity peaking, Figs. 18.22, 18.23, and 18.24
plot lower bounds on sensitivity-function norms. The plots show that as the ratio
of the control bandwidth to the performance bandwidth decreases, the sensitivity
peaking becomes more and more severe. Further, the peaking is accentuated by an
increase in the delay τ , an increase in the performance requirements (lower ε), or
an increase in the real part of the unstable pole of the open-loop plant σr .
Extension of fundamental limitations to the case in which either plant or con-
troller has nonlinear characteristics is possible by using the concept of random
input-describing functions introduced in the preceding section. For example, in
the sector combustor controlled with on/off valves the fundamental limitations in
terms of lower bounds on the logarithm of sensitivity function as presented in this
section apply to G C ( jω) replaced with G C ( jω)N (σ ), where N (σ ) is the random
7
6 ∆ω1 = 40Η z
ζ = −4 % ε = .3 (10 d B)
5 δ = 0.1 (20 dB)
||S|| 4 ζ = − 2% τ = 4.3ms
8
3 ζ = − 0.4%
2
1
2 3 4 5 6 7 8 9 10
stablel
∆ω 2 / ∆ω1
Fig. 18.24 Lower bounds on sensitivity function norm as function of control band-
width for four values of combustor pole-damping ratio.
606 J. M. COHEN AND A. BANASZUK
input-describing function of the on/off valve and σ is the standard deviation of the
Gaussian component of the valve command. Fundamental limitation analysis also
extends to the cases in which more general nonlinearities are present in the model
of combustor or controller, Gaussian noise sources are present, and the combustor
feedback loop operates at a limit cycle. In each particular case one has to prove
existence of the Gaussian and periodic signals in the feedback loop that provide
balance and examine stability of solution. More details are presented in Ref. 21.
VI. Conclusion
We have discussed some of the factors that may limit the performance of active
control systems for the attenuation of combustion instabilities. A broad range of
factors was identified and discussed. This list is not complete, by any measure.
Many effects can be system dependent; they are critical to one system but irrelevant
for another. System dependence may vary depending on the nature of the control-
system architecture and on the combustion dynamics that are being controlled.
Three of the critical factors limiting the control of the lean, premixed combustor
design considered in this paper have been examined in more detail.
The ability of the fuel actuator to affect the root-cause physics behind pressure–
heat-release coupling was found to be tied strongly to the mixing of the actuated
fuel flow with the remainder of the premixed reactants. In effect, the actuated fuel
flow must act to achieve a high degree of premixedness, both in time and space.
Another limiting factor relates to actuation time delay, as represented by the time
between movement of the fuel valve and realization of that fuel flow modulation
in the unsteady heat release or combustor pressure. Large values of time delay
were found to shrink the frequency band over which the control system could
attenuate pressure oscillations and to provide a mechanism through which pressure
oscillations outside this bandwidth could be amplified. This finding led to the peak-
splitting phenomenon that limited the degree to which pressure oscillations could
be suppressed.
Attempts to deal with these issues led to a sensitivity-function analysis of the
fundamental limits of combustor pressure oscillation control. It was shown that one
cannot arbitrarily decrease the level of pressure oscillations using linear controllers.
This limit is strongly influenced by system time delays, control bandwidth, and
performance bandwidth.
Although these issues represent current limitations, they are certainly not the
only factors that will affect performance of instability-control systems. Other sys-
tems (not of this type) will face different issues and different factors may control
their performance. Active combustion-instability control has been well demon-
strated as a technology with significant potential. For the technology to mature to
the point of being practically applicable, future efforts must focus on these limiting
factors, quantify them, and devise methods for dealing with them.
Acknowledgments
The work presented in this paper was supported by Contract MDA972-95-C-
0009 from Defense Advanced Research Projects Agency and Contract F49620-
01-C-0021 from the Air Force Office of Scientific Research and by United
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 607
References
1
Hibshman, J. R., Cohen, J. M., Banaszuk, A., Anderson, T. J., and Alholm, H. A., “Active
Control of Combustion Instability in a Liquid-Fueled Sector Combustor,” American Society
of Mechanical Engineers, Paper 99-GT-215, June 1999.
2
Cohen, J. M., Rey, N. M., Jacobson, C. A., and Anderson, T. J., “Active Control of
Combustion Instability in a Liquid-Fueled Low-NOx Combustor,” Journal of Engineering
for Gas Turbines and Power, Vol. 121, No. 2, 1999, pp. 281–284.
3
Seume, J. R., Vortmeyer, N., Krause, W., Hermann, J., Hantschk, C.-C., Zangl, P.,
Gleis, S., Vortmeyer, D., and Orthmann, A., “Application of Active Combustion Instability
Control to a Heavy Duty Gas Turbine,” Journal of Engineering for Gas Turbines and Power,
Vol. 120, No. 4, 1998, pp. 721–726.
4
Hoffmann, S., Weber, G., Judith, H., Hermann, J., and Orthmann, A., “Application
of Active Combustion Instability Control to Siemens Heavy Duty Gas Turbines,” Sym-
posium of the AVT Panel on Gas Turbine Engine Combustion, Emissions and Alternative
Fuels, NATO Research and Technology Organization, Neuilly-Sur-Seine Cedex, France,
Oct. 1998.
5
Paschereit, C. O., Gutmark, E., and Weisenstein, W., “Control of Combustion Driven
Oscillations by Equivalence Ratio Modulations,” American Society of Mechanical Engi-
neers, Paper 99-GT-118, June 1999.
6
Richards, G. A., Yip, M. J., Robey, E., Cowell, L., and Rawlins, D., “Combustion
Oscillation Control by Cyclic Fuel Injection,” American Society of Mechanical Engineers,
Paper 95-GT-224, June 1995.
7
Anson, B., Critchley, I., Schumacher, J., and Scott, M., “Active Control of Combustion
Dynamics For Lean Premixed Gas Fired Systems,” American Society of Mechanical Engi-
neers, Paper GT-2002-30068, June 2002.
8
Sattinger, S. S., Neumeier, Y., Nabi, A., Zinn, B. T., Amos, D. J., and Darling, D. D.,
“Sub-scale Demonstration of the Active Feedback Control of Gas-Turbine Combustion
Instabilities,” Journal of Engineering for Gas Turbines and Power, Vol. 122, April 2000,
pp. 262–270.
9
Bloxsidge, G. J., Dowling, A. P., Hooper, N., and Langhorne, P. J., “Active Control of
Reheat Buzz,” AIAA Journal, Vol. 26, 1988, pp. 783–790.
10
Langhorne, P. J., Dowling, A. P., and Hooper, N., “Practical Active Control Sys-
tem of Combustion Oscillations,” Journal of Propulsion and Power, Vol. 6, pp. 324–333,
1981.
11
Fleifil, M., Annaswamy, A. M., Hathout, J. P., and Ghoniem, A. F., “The Origin of
Secondary Peaks with Active Control of Thermoacoustic Instability,” Combustion Science
and Technology, Vol. 133, June 1998, pp. 227–260.
12
Gysling, D. L., Copeland, G. S., McCormick, D. C., and Proscia, W. M., “Combus-
tion System Damping Augmentation With Helmholtz Resonators,” American Society of
Mechanical Engineers, Paper 98-GT-268, June 1998.
608 J. M. COHEN AND A. BANASZUK
13
Hermann, J., Gleis, S., and Vortmeyer, D., “Active Instability Control (AIC) of Spray
Combustors by Modulation of the Liquid Fuel Flow Rate,” Combustion Science and
Technology, Vol. 118, Taylor & Francis, Philadelphia, 1996, pp. 1–25.
14
Saunders, W. R., Vaudrey, M. A., Eisenhower, B. A., Vandsburger, U., and Fannin, C.
A., “Perspectives on Linear Compensator Designs for Active Combustion Control,” AIAA
Paper 99-0717, Jan. 1999.
15
Banaszuk, A., Jacobson, C. A., Khibnik, A. I., and Mehta, P. G., “Linear and Nonlinear
Analysis of Controlled Combustion Processes. Part I: Linear Analysis,” Proceedings of
Conference on Control Applications, IEEE Publications, Piscataway, NJ, Aug. 1999.
16
Hathout, J. P., Annaswamy, A. M., Fleifil, M., and Ghoniem, A. F., “A Model-Based Ac-
tive Control Design for Thermoacoustic Instability,” Combustion Science and Technology,
Vol. 132, 1998, pp. 99–105.
17
Banaszuk, A., Aryiur, K. B., Krstic, M., and Jacobson, C. A., “An Adaptive Algo-
rithm for Control of Combustion Instability,” Automatica Vol. 40, No. 11, Nov. 2004,
pp. 1965–1972.
18
Johnson, C. E., Neumeier, Y., Lubarsky, E., Lee, Y. J., Neumaier, M., and Zinn, B. T.,
“Suppression of Combustion Instabilities in a Liquid Fuel Combustor Using a Fast Adaptive
Algorithm,” AIAA Paper 2000-0476, Jan. 2000.
19
Murugappan, S., Gutmark, E. J., and Acharya, S., “Application of Extremum-Seeking
Controller to Suppression of Combustion Instabilities in Spray Combustion,” AIAA Paper
2000-1025, Jan. 2000.
20
Evesque, S., “Adaptive Control of Combustion Oscillations,” Ph.D. Dissertation,
Cambridge Univ., Cambridge, England, U.K., 2000.
21
Banaszuk, A., Mehta, P. G., Jacobson, C. A., and Khibnik, A. I., “Limits of Achievable
Performance of Controlled Combustion Processes,” submitted to IEEE Transaction on
Automatic Control (submitted for publication, 2005).
22
Banaszuk, A., Jacobson, C. A., Khibnik, A. I., and Mehta, P. G., “Linear and Nonlinear
Analysis of Controlled Combustion Processes. Part II: Nonlinear Analysis,” Proceedings
of Conference on Control Applications, IEEE Publications, Piscataway, NJ, Aug. 1999.
23
Freudenberg, J. S., and Iooze, D. P., “A Sensitivity Tradeoff for Plants With Time
Delay,” IEEE Transactions on Automatic Control, Vol. AC-32, Feb. 1987, pp. 99–104.
24
Seron, M. M., Braslavsky, J. H., and Goodwin, G. C., Fundamental Limitations in
Filtering and Control, Springer, New York, 1997.
25
Ariyur, K. B., “Multivariable Extremum-Seeking Adaptive Control,” Ph.D. Disserta-
tion, University of California, San Diego, 2002.
26
Krstic, M., and Wang, H. H., “Stability of Extremum Seeking Feedback for General
Nonlinear Dynamic Systems,” Automatica, Vol. 36, April 2000, pp. 595–601.
27
Krstic, M., “Performance Improvement and Limitations in Extremum Seeking Control,”
Systems and Control Letters, Vol. 39, April 2000, pp. 313–326.
28
Zhang, Y., “Stability and Performance Tradeoff with Discrete Time Triangular Search
Minimum Seeking,” Proceedings of the American Control Conference, IEEE Publications,
Piscataway, NJ, June 2000.
29
Cohen, J. M., Stufflebeam, J. H., and Proscia, W., “The Effect of Fuel/Air Mixing
On Actuation Authority In An Active Combustion Instability Control System,” Journal of
Engineering for Gas Turbines and Power, Vol. 123, No. 3, 2001, pp. 537–542.
30
Stufflebeam, J. H., Kendrick, D. W., Sowa, W. A., and Snyder, T. S., “Quantifying
Fuel/Air Unmixedness in Premixing Nozzles Using an Acetone Fluorescence Technique,”
American Society of Mechanical Engineers, Paper 99-GT-399, June 1999.
FACTORS AFFECTING THE CONTROL OF UNSTABLE COMBUSTORS 609
31
Peracchio, A. A., and Proscia, W., “Nonlinear Heat-Release/Acoustic Model for Ther-
moacoustic Instability in Lean Premixed Combustors,” Journal of Engineering for Gas
Turbines and Power, Vol. 121, No. 3, 1999, pp. 415–421.
32
Lieuwen, T., Torres, H., Johnson, C., and Zinn, B., “A Mechanism for Combustion
Instabilities in Premixed Gas Turbine Combustors,” Journal of Engineering for Gas Turbines
and Power, Vol. 123, No. 1, 2001, pp. 182–190.
33
Lee, D. S., and Anderson, T. J., “Measurements of Fuel/Air – Acoustic Coupling in
Lean Premixed Combustion Systems,” AIAA Paper 99-0450, Jan. 1999.
34
Lieuwen, T. C., “Experimental Investigation of Limit Cycle Oscillations in an Unstable
Gas Turbine Combustor,” Journal of Propulsion and Power, Vol. 18, No. 1, 2002, pp. 61–67.
35
Mezic, I., and Banaszuk, A., “Comparison of Systems with Complex Behavior: Spectral
Methods,” Physica D, Nonlinear Phenomena, Vol. 197, No. 1–2, October 2004, pp. 101–
133.
36
Gelb, A., and Vander Velde, W. E., Multiple-Input Describing Functions and Nonlinear
System Design, McGraw-Hill, New York, 1968.
Chapter 19
Jakob Hermann∗
If TA GmbH, Groebenzell, Germany
and
Stefan Hoffmann†
Siemens AG, Mülheim, Germany
I. Introduction
I N 1995, Siemens started its first tests with the new gas-turbine family Vx4.3A in
its test facility in Berlin. In contrast to the former gas-turbine family with silo-
combustion chambers, annular combustion chambers were implemented to achieve
an increased power density and efficiency and lower emission values. However,
during the tests, combustion instabilities detrimental to a reliable operation of the
gas turbines were observed. Already very early in the process of working out
countermeasures against this phenomenon, it was decided that besides passive
measures, such as operational modifications or an acoustic detuning by design
modifications of the burner,1 also active measures for avoiding these instabilities
should be developed and tested.
At that time, active combustion control was still in a research stage, as sum-
marized in several review papers.2−6 Various authors described successful tests
based on laboratory-scale burners with a thermal power of between 1 kW and
250 kW.7−9 In all these publications, attenuation of combustion oscillations was
achieved by phase-shifted sound signals generated via loudspeakers. In addition
to this method, other types of intervention and control strategies were researched
for various combustion systems. However, all tests were performed on labora-
tory scale.10−17 Based on these experiences a system with an adequate sensor,
controller, and actuator had to be developed to make it possible to damp com-
bustion instabilities in a heavy-duty gas turbine. The first full-scale tests with this
Copyright c 2004 by the authors. Published by the American Institute of Aeronautics and Astro-
nautics, Inc., with permission.
∗ Managing and Research Director.
† Director, Gas Turbine Component Engineering, Power Generation.
611
612 J. HERMANN AND S. HOFFMANN
technique, called active instability control (AIC), were made in 1996 in the test
facility of Siemens on the V84.3A gas-turbine delivering 160 MW of electrical
power.18 Because of the positive results, this technique was also applied to the
largest type of this family of gas turbines, the Siemens V94.3A with an electric
power output of 267 MW.19,20
The implementation of the AIC technology in Siemens Vx4.3A gas turbines and
the problems encountered are the subjects of Sec. II. Results achieved with AIC
are discussed in Sec. III. Sections IV and V present long-term experiences and
a short evaluation of the advantages offered by AIC in comparison with passive
measures.
from this mixed mode to a purely premixed operation. To stabilize the premix
flame, every Siemens hybrid burner features an additional pilot burner, which is a
diffusion burner delivering approximately 10% of the thermal power provided by
the complete burner unit.
A. Instability Problem
During the test phase of the prototype and the commissioning of the type of
gas turbine described in preceding text, problems occurred with self-excited com-
bustion oscillations in various power ranges, depending on the modifications per-
formed on the hybrid burner used. Because of the low frequency of the instabilities,
this phenomenon is sometimes called “humming.”
Research on the instabilities by Seume et al.18 showed that standing sound waves
are generated within the annular combustion chamber by self-excited combustion
oscillations. According to the direction of propagation of these waves along the
circumference of the annular combustor, they are designated “azimuthal modes.”
For example, Fig. 19.4 shows the azimuthal modes of the second harmonic, char-
acterized by a total of four nodes and four antinodes. For the largest version of
this gas-turbine type, V94.3A, this eigenmode is excited at a frequency of approx-
imately 170 Hz.
The equation f = n · c/(π · d) provides a theoretical estimate for the charac-
teristic acoustical frequencies of any annular combustion chamber. At an average
combustion chamber diameter of d = 3 m and a speed of sound of c = 844 m/s
(assuming an average combustion chamber temperature of 1500◦ C), the resulting
frequency for a V94.3A is 179 Hz for the second harmonic (n = 2), a value agree-
ing well with experimentally determined characteristic frequencies. For smaller
Fig. 19.4 Excited azimuthal modes for the second harmonic in the annular combus-
tion chamber of the V94.3A.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 615
versions of this turbine family, this equation returns higher frequencies for the
occurrence of combustion oscillations, because of the reduced diameter of their
combustion chambers (see Ref. 18).
B. AIC Installation
To avoid the oscillation problem, an AIC system for this type of gas turbine was
developed in addition to passive measures such as burner-design modifications.1
Figure 19.5 provides a simplified schematic diagram of the basic design of this
AIC.
1. Sensor
As an input quantity, this AIC system uses the sound pressure measured at the
burner flanges. The wall temperatures are substantially lower at the burner flanges
than within the combustion chamber so that high-temperature piezo-pressure trans-
ducers can be used without requiring any additional cooling. Several tests verified
that the sound-pressure signals measured at the burner flanges correlate suffi-
ciently well in amplitude and phase with the pressure signals within the combustion
chamber.
During some initial tests, potential uses of optical probes for measuring AIC
input signals were researched in addition to sound-pressure measurements at the
burner flanges. However, because of limited viewing angles and thermal problems
with probe installations, this method was soon given up (see Ref. 18 for some more
detail).
Fig. 19.5 Schematic AIC diagram of the Siemens model Vx4.3A heavy-duty gas
turbine.
616 J. HERMANN AND S. HOFFMANN
2. Actuator
The crucial problem for implementing an AIC system is actively influenc-
ing the combustion. Owing to the elevated air and fuel volume flows through
Vx4.3A gas turbines, these flows cannot be sufficiently modulated in full, nei-
ther by means of acoustical actuators nor by valves. Detailed research showed
that the main premix flame of the Siemens hybrid burner, controlled by much
smaller pilot-diffusion flames comprising no more than approximately 10% of
the entire mass flow, will respond very precisely to fluctuations in the conversion
rates of those pilot flames. Therefore, modulating the pilot-gas mass flow con-
trols not only the pilot flames themselves but also the premixed main flame to a
significant extent. Pilot flames are supplied via their own fuel line. The actuator,
a direct-drive valve (DDV) (shown in Fig. 19.6) developed specifically for this
application by Moog Germany, is integrated in this fuel line. To obtain maximum
control of the main flames, and thus of the combustion oscillations arising within
the gas turbine, every burner of the turbine was equipped with its own valve.
Thus, a total of 24 valves was installed around the annular combustor of the gas
turbine.
In the absence of any AIC signal, DDV valves will be 50% open, so that the
pilot-gas mass flows required for normal premixed operation will reach the burners.
To modulate individual pilot-gas mass flows and thus the pilot flames themselves,
valve spools will be moved around their static opening value of 50%, that is,
opened further or closed down at the frequency of the combustion oscillation to be
damped. The level of fuel flow modulation achieved can be determined by valve
spool stroke. According to the frequency response of the used valve type (shown in
Fig. 19.7) the valves allow the control of combustion oscillations with frequencies
of up to 400 Hz.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 617
By moving the valve spools at frequencies within this range, sound waves will
be generated within the pilot-gas systems in addition to the modulation of the
gas mass flow. The resulting effects must be taken into account and the pilot-gas
system must be tuned as described in Hermann et al.16 and Hantschk et al.17 to
optimize the modulation amplitude.
3. Controller
As described in Seume et al.18 and shown in Fig. 19.4, azimuthal modes are ex-
cited within annular combustion chambers by combustion oscillations. Excitation
by combustion oscillation means that the pressure fluctuations excited along the
combustion-chamber circumference are characterized by both different amplitudes
and different phases and that the burners placed evenly along the circumference of
the annular combustion chamber are accordingly located at positions characterized
by differing pressure amplitude and phase values. Fluctuations of the heat-release
rate are coupled characteristically with pressure oscillations (self-excitation), thus
heat released at the various burners will fluctuate at differing amplitudes and phases.
Because AIC requires both antiphase and in-amplitude influencing of the unsteady
flame, an individual control of each burner is necessary, which means that it must
be possible to control every burner individually. The least complicated case would
therefore require one sensor and one feedback loop for every burner, so that a
24-channel AIC system would have to be installed.
To minimize the number of feedback loops, the symmetry of azimuthal modes
arising within the combustion chamber was exploited. As shown by the az-
imuthal mode corresponding to the first harmonic in Fig. 19.8, the sound-pressure
fluctuations for two burners lying precisely opposite each other are characterized
by identical amplitudes p̂ and a phase shift ϕ of 180 deg (Fig. 19.9). As indicated in
618 J. HERMANN AND S. HOFFMANN
Fig. 19.8 Exploiting the symmetry of azimuthal modes, for example, for the first
harmonic. One sensor and one controller provide the input signals for two DDVs.
Fig. 19.9, for an AIC system, burners placed precisely opposite each other means
that a signal measured at a specific position by one sensor can be used to control not
only the actuator for this position, but also the one located at the precisely opposite
point of the combustion chamber by merely inverting the controller output signal
for the second actuator.
In second or even higher harmonics, the number of feedback loops can be
reduced even further, as shown in Fig. 19.10. Here, four actuators are con-
trolled by one input signal and one feedback loop. For the control system of
V94.3A burners, an installation as in Fig. 19.8 was chosen, because for this
type of gas turbine, the first harmonic will likewise appear due to self-excited
Fig. 19.9 Sound pressure at the two valve positions of Fig. 19.8.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 619
Fig. 19.10 Exploiting the symmetry of azimuthal modes, for example, for the second
harmonic. One sensor and one controller provide the input signals for four DDVs.
oscillations. In total, this type of gas turbine was fitted with 12 sensors and 12
feedback loops.
Those 12 loops were realized by means of six signal processors, every processor
handling two input signals and generating four output signals for a total of 24 ac-
tuators. The algorithm used works within the frequency range and can, simplified,
be characterized as a phase shifter and amplifier. Because of its implementation
in the frequency domain the controller works extremely rapidly. In addition, this
algorithm allows the specific control of individual frequencies and is suitable for
controlling two frequencies, a restriction entailed by the limited computing capac-
ity of the type of signal processor used. The necessary control parameters depend
on the operation parameters of the gas turbine and are set automatically during con-
tinuous operation. Hermann et al.20 provides a detailed description of this control
system.
4. Technical Realization
The hardware and software setup is optimized with respect to time-saving im-
plementation and commissioning as well as facilitated maintenance. Figure 19.11
shows a complete AIC setup with a front view of the electronic hardware con-
sisting of a control and a power cabinet. The cabinets are self-contained and only
need a minimum number of data lines to interface with the gas-turbine control
system. Thus they can easily be integrated into already existing control systems.
The following hardware and software features have been included: modular ar-
chitecture (easy to extend and service), hardware monitoring for automatic error
diagnosis, actuator monitoring (valve seizing, deviation from the desired set point,
marginal check), sensor monitoring (cable and sensor defects), redundant power
supply (interchangeability during operation), integrated liquid-crystal displays and
620 J. HERMANN AND S. HOFFMANN
Fig. 19.11 The complete AIC setup for the V94.3A gas turbine.
keyboard for monitoring and user access, data and control interface to the gas-
turbine control system, failure- and data-recording system, and special tools for
fast commissioning.
Fig. 19.12 Operating map of the gas turbine with active control of combustion in-
stabilities at several operating points observed in combination with different burner
types.
the premixed flame, the pilot burner is switched on while switchover is in progress.
Figure 19.13 shows this process without an activated AIC system and Fig. 19.14
shows it with an activated AIC system. From top to bottom, the following quanti-
ties are plotted: maximum sound-pressure spectrum calculated based on the signals
delivered by the 12 AIC sensors at three different points of time; maximum rms
value of the sound pressure versus time, calculated based on the signals delivered
by the 12 AIC sensors; opening cross section vs time for the gas valves actuated
during switchover, as a percentage of the maximum opening cross section for the
corresponding valve.
As shown in Fig. 19.13 strong combustion oscillations occur approximately 3.5 s
after opening the pilot-gas main valve (start of switchover, marked in the bot-
tom diagram) as the pilot-gas flows into the combustion chamber. After another
2 s, the diffusion-gas valve is closed and the premix-gas valve opens, with the
thermal power within the combustion chamber being kept constant during that
process. About 10.5 s after initiating the switchover (t = 15 s), that is, as the
diffusion-gas valve closes, the combustion-oscillation amplitude slightly weakens
before it returns, after another 5 s, to the value characterizing constant opera-
tion (see rms signal, Fig. 19.13). At t = 20 s, the gas turbine is running in its
fully stabilized premix mode. Considering the frequency spectrum of the sound
pressure measured at various times during switchover, it becomes obvious that,
in the beginning, the first harmonic dominates at about 90 Hz (t = 9.7 s). At
t = 13.7 s, this frequency will have dropped to approximately 80 Hz. Because of a
622 J. HERMANN AND S. HOFFMANN
Fig. 19.13 Switchover from mixed to premixed operation without AIC. The inlet
guide vane remained permanently closed.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 623
Fig. 19.14 Switchover from mixed to premixed operation with AIC. The inlet guide
vane remained permanently closed.
624 J. HERMANN AND S. HOFFMANN
nonlinear effect, high amplitudes of the first harmonic will, produce higher-order
harmonics.
Figure 19.14 shows the same switchover process with activated AIC. As the
sound-pressure history for the rms value indicates, AIC achieves an almost com-
plete attenuation of combustion oscillations. Only one specific position of the
diffusion-gas valve (t = 15.2 s) produces a short peak lasting approximately 0.5 s,
its rms value already substantially reduced. The corresponding frequency spectrum
reveals that the harmonic excited is no longer the first but the third one, at approx-
imately 250 Hz. To achieve even further damping of this oscillation by means of
AIC, the system was preset for both frequencies (first and third harmonics). As
shown in the spectrum of another switchover recorded 10 min later, the amplitude
of this oscillation was further attenuated with the first harmonic being damped at
the same time. It must be stated that the damping achieved by AIC was more than
sufficient for safe gas-turbine operation.
Fig. 19.15 Load increase to base load in premixed operation with activated AIC
system. While running up the gas turbine, the system must be activated earlier than
when the gas turbine is being run down (see Fig. 19.16). The AIC system was also
always activated at base load to suppress combustion instabilities.
626 J. HERMANN AND S. HOFFMANN
Fig. 19.16 Load decrease to part load in premixed operation with activated AIC
system.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 627
the second to the third harmonic did not produce any significant amplitudes. The
diagrams in Fig. 19.15 demonstrate this AIC operation.
Fig. 19.17 Base-load operation without AIC during the long-term field demonstration
in a V94.3A.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 629
Fig. 19.18 Base-load operation with AIC during the long-term field demonstration in
a V94.3A. The system has been operating continuously since January 1999, enhancing
the base-load limit by 5% points.
630 J. HERMANN AND S. HOFFMANN
Fig. 19.19 Suppression of two frequency peaks by AIC during part-load operation.
(maximum values for the 12 sensors) versus time. Dark areas indicate high am-
plitudes; lighter areas indicate the general noise level. The two horizontal bars
at completely deactivated AIC mark the two eigenmodes excited, in this case,
at 145 Hz and 290 Hz. Starting at t = 70 s, by slowly reducing the AIC output
signal to suppress the fourth harmonic at 290 Hz within the 70 s< t < 85 s range,
the amplitudes of this harmonic increase. During the next 23 s, AIC only damps
oscillations at 145 Hz. By completely switching off the AIC system at t = 108
s, the second harmonic is likewise excited at 145 Hz. By reactivating the AIC
for both frequencies at t = 127 s, the two oscillations are damped out completely
once again. This example demonstrates that both oscillations can be self-excited
independently of each other and that, in general, if combustion oscillations are to
be avoided completely, both will have to be damped by using AIC.
Fig. 19.21 AIC operating hours for 14 commercially operating V94.3A gas turbines
named A through N.
632 J. HERMANN AND S. HOFFMANN
VI. Conclusion
AIC has made it possible to successfully suppress self-excited combustion oscil-
lations in a Siemens V94.3A heavy-duty gas turbine with an electric power output
of 267 MW. This suppression was achieved by developing a multichannel AIC
system with a maximum of 12 control loops, each with an individual sensor and
a total number of 24 actuators in the pilot-gas system. Counteracting the oscilla-
tions is realized by a modulation of the pilot-gas mass flow out of phase to the
heat-release fluctuation of the flame. Because self-excited oscillations are typically
caused by a coupling between the heat-release rate of the flame and the acoustics,
eliminating the heat-release fluctuations also eliminates the acoustic oscillations.
ACTIVE CONTROL IN A FULL-SCALE GAS-TURBINE COMBUSTOR 633
The results presented of long-term tests also show the high reliability of the system
even under demanding operating conditions. The fail-safe and the condition-based
maintenance features worked well over time. Results of test series at steady and un-
steady operation of the gas turbine demonstrate sustainable damping performance
for longer operation periods. Even after 18,000 h of operation, the actuators and
controllers demonstrated excellent performance, proving that AIC has reached
maturity.
The modular and compact design of the stand-alone system with a minimized
number of interfaces to the controller allows for rapid implementation into existing
systems and makes the overall arrangement very service friendly. By foreseeing
the interfaces in the design of a new engine, retrofitting an AIC system can be
performed within the time frame of a regular engine outage.
Despite the costs connected with the use of an AIC system the features men-
tioned earlier make AIC a very powerful risk-mitigation feature for the market
introduction of new or revised combustion systems, thus allowing for a smoother
market introduction of new products or power upgrades. The market-introduction
delays which every original equipment manufacturer (OEM) encountered in the
mid-1990s with their highly efficient frames could have been significantly reduced
by the use of AIC.
References
1
Berenbrink, P., and Hoffmann, S., “Suppression of Dynamic Combustion Instabilities
by Passive and Active Means,” American Society of Mechanical Engineers, Paper 2000-
GT0079, May 2000.
2
McManus, K. R., Poinsot, T., and Candel, S. M., “A Review of Active Control of Com-
bustion Instability,” Progress in Energy and Combustion Science, Vol. 19, 1993, pp. 1–29.
3
Yang, V., and Schadwo, K. C., “AGARD Workshop on Active Combustion Control
for Propulsion Systems,” Proceedings of NATO-RTO Symposium on Gas Turbine Engines
Combustion, Emissions and Alternative Fuels, RTO-MP-14, 1998, pp. 36/1–36/20.
4
Dowling, A. P., “Active Control of Instabilities in Gas Turbines,” RTO NATO
Conference, Braunschweig, Germany, May 2000.
5
Annaswamy, A. M. and Ghoniem, A. F., “Active Control of Combustion Instability:
Theory and Practice,” IEEE Control Systems Magazine, Vol. 22, No. 6, Dec. 2002, pp. 37–54.
634 J. HERMANN AND S. HOFFMANN
6
Candel, S., “Combustion Dynamics and Control: Progress and Challenges. (Hottel
Lecture).” 29th Proceedings of the Combustion Institute, Vol. 29, Sapporo, Japan, 2002,
pp. 1–28.
7
Lang, W., Poinsot, T., and Candel, S., “Active Control of Combustion Instability,”
Combustion and Flame, Vol. 70, 1987, pp. 281–289.
8
Gulati, A., and Mani, R., “Active Control of Unsteady Combustion-Induced Oscilla-
tions,” Journal of Propulsion and Power, Vol. 8, No. 5, 1992, pp. 1109–1115.
9
Poinsot, T., Veynante, D., Bourienne, F., Candel, S., and Esposito, E., “Initiation and
Suppression of Combustion Instabilities by Active Control,” 22nd Symposium (Interna-
tional) on Combustion, The Combustion Inst., Pittsburgh, PA, 1988, pp. 1363–1370.
10
Bloxsidge, G., Dowling, A., Hooper, N., and Langhorne, P., “Active Control of Reheat
Buzz,” AIAA Journal, Vol. 26, No. 7, 1988, pp. 783–790.
11
Langhorn, P. J., “Reheat Buzz an Acoustically Coupled Combustion Instability,”
Journal of Fluid Mechanics, Vol. 193, 1988, pp. 417–443.
12
Poinsot, T., Candel, S., Esposito, E., Lang, W., and Bourienne, F., “Suppression of
Combustion Instabilities by Active Control,” Journal of Propulsion and Power, Vol. 5,
No. 1, Jan.–Feb. 1989, pp. 14–20.
13
Wilson, K. J., Gutmark, E., and Schadwo, K. C., “Flame-Kernel Pulse Actuator
for Active Combustion Control,” ASME 1992, Active Control of Noise and Vibration,
DSC-Vol. 38, 1992, pp. 75–81.
14
Yu, K. H., Parr, T. P., Wilson, K. J., Schadow, K. C., and Gutmark, E. J., “Active
Control of Liquid-Fueled Combustion Using Periodic Vortex-Droplet Interaction,” 26th
Symposium (International) on Combustion, Naples, Italy, 1996, pp. 2843–2850.
15
Yu, K., Wilson, K. J., and Schadow, K. C., “Scale-Up Experiments on Liquid-Fueled
Active Combustion Control,” AIAA Paper 98-3211, July 12–15 1998.
16
Hermann, J., Gleis, S., and Vortmeyer, D., “Active Instability Control (AIC) of Spray
Combustors by Modulation of the Liquid Fuel Flow Rate,” Combustion Science and
Technology, Vol. 118, 1996, pp. 1–25.
17
Hantschk, C., Hermann, J., and Vortmeyer, D., “Active Instability Control with
Direct Drive Servo Valves in Liquid-Fuelled Combustion Systems,” 26th Symposium
(International) on Combustion, The Combustion Inst., Pittsburgh, PA, 1996.
18
Seume, J. R., Vortmeyer, N., Krause, W., Hermann, J., Hantschk, C.-C., Zangl, P.,
Gleis, S., Vortmeyer, D., and Orthmann, A., “Application of Active Combustion Instability
Control to a Heavy Duty Gas Turbine,” American Society of Mechanical Engineers, Paper
97-AA-119, September 1997.
19
Hoffmann, S., Weber, G., Judith, H., Hermann, J., and Orthmann, A., “Application
of Active Combustion Control to Siemens Heavy Duty Gas Turbines,” Presented at
the Symposium of the AVT Panel on Gas Turbine Engine Combustion, Emissions and
Alternative Fuels, Lisbon 12–16 Oct., RTO Meeting Proceedings 14, 1998, pp. 40-1–40-13.
20
Hermann, J., Orthmann, A., and Hoffmann, S., “Application of Active Instability Con-
trol to a Heavy Duty Gas Turbine,“ XIV ISABE, 5–10 Sept., Florence, Italy, A99-34186,
1999.
SUBJECT INDEX
Acoustic modeling
combustion and 51
damping processes and 59
deviations in accuracy of 57 59
difficulty of 54
passive damping devices in 59
results of 57
Acoustic modes energy transfer 8
Acoustic nodes 565
Acoustic oscillations 5 433
Acoustic perturbation 317
Acoustic phenomena
of can-annular combustion system 102
Acoustic pressure 439
in swirl-stabilized combustor 439
Acoustic pressure perturbations 344
Acoustic reflection coefficient 431
Acoustic resonators 85 404
Acoustic transfer-matrix 98
Acoustic velocity 186
Acoustic wave frequency 185
Acoustic wave generation 223
Acoustic wave interactions 315
Acoustic waves 78
Acoustic-flame coupling 181
Acoustic-flame interactions
spatial and temporal scales’ relation with 319
Index Terms Links
Acoustic-impedance
definition of 119
experimental rig for 119
measurements of 119
Acoustic-pressure distribution 101
Acoustics and blowout avoidance logic (ABAL) 51
Acoustics for reacting flows 183
compact flame in duct for 185
heat-release fluctuation for 183
Acoustic-vortex-flame (AVF)
field composition interpretation using 283
gas-turbine combustion systems and 277
interactions
droplet-vortex interactions and 296
droplet-vortex-flame interactions and 302
factors affecting 287
in gas turbines 277
occurrence of 278
swirl and 287
time-evolving flow data for understanding
of 308
length and time scales of 278
swirl and 287
theoretical considerations for 280
Active combustion control (ACC) 100 135
Active control
at base-load operation 627
at load change during premixed operation 624
Index Terms Links
Combustion (Cont.)
semiempirical 53
spontaneous oscillatory 68
spray 302 304
time-lag 51
turbulent 319
unstable 187
as velocity source term 186
waves/flow perturbations’ relation to 181
Combustion dynamics
in aircraft propulsion 43
CFD modeling of 56
concerns with 534
control of in gas-turbine combustors 47
control strategies for 45
controlling difficulty of 61
coupling mechanics 45
detrimental impacts of 153
engine tests for 47
examples of 46
framework for modeling of 44
frequency response and 168
as function of load and ambient temperature 167
as function of load and frequency response 168
fundamental causes of 45
gas-turbine 415
laboratory scale experiments of 127
Index Terms Links
Combustor (Cont.)
description of 585
double annular 46
dry low-emissions 48 68 213
dump 284
dynamics 582
engines’ rig pressure compared to 134
exit 431
fuel-air ratio 131
gas-turbine 465 475 611
geometry 570
health monitoring 154
heat release in 76
high mach-number 469
inlet 431
lean-premixed 49 487
linear processes of 17
linear waves and 397
liquid-fueled three-nozzle sector 594
low-emissions 587
LPP gas-turbine 369 370 378
modes 476
multiple-flame-holder dump 188
multiple-frequency mode 556
nonlinear processes of 15 17 21
operating conditions of, in single-nozzle test rig 131
physical domain of model 438
Index Terms Links
Combustor (Cont.)
premixed 179
premixer 13
pressure spectra of 596
second-order 599
silo 475
simple 400 406 409
geometry and flow conditions for 398
simple quasi-one-dimensional 395
single-nozzle research 485
stability map of 446
stability modeling of 446
subscale, experiment of 127
swirl-stabilized 251 436
three-stage premix 76
of Trent 60 68
two-mode 571
unstable 15 581
unsteady pressure of 132
Combustor pressure 130
ASPAC and 139
of Bode plot 597
experimentally determined frequency
responses of 597
fluctuations 483 494
measurements 484
spectra of 596
Index Terms Links
Effusion cooling 33
Index Terms Links
Engines (Cont.)
combustor’s rig pressure compared to 134
design for testing operating conditions of 128
F-1 6
GE Aircraft 243
Mars 33 38
tuning issues with 152
Entropy
equation of 281
unsteady generation of 387
Entropy disturbances 319 323
propagation of 320
separable solutions for 390
Entropy fluctuations 387
of combustor inlet 431
mean-flow gradients’ relation to 426
Entropy generation 323
Entropy perturbations 280
Entropy waves 388 392
Equivalence ratio
flame-speed sensitivity to 345
increasing 488
mean 349 350
overall Chemiluminescence emission effected
by 488 492
Equivalence-ratio
oscillations 10
Index Terms Links
F-1 engine 6
Facility air piping 127
FAR. See Fuel Concentration; Fuel-air ratio
Far-field radiated pressure 194
Fast flame-ionization detector (FID) 77
Fast Fourier transform (FFT) 150
Fast-response extraction probe 505
Favre-filtered conversation equations 216
FEA. See Finite element analysis
Feedback acoustics 569
Feedback element 535
Index Terms Links
Feedback loop 9
Feedback transfer function 537
Feed-system coupling 481 504
FEM. See Finite element method
FFT. See Fast Fourier transform
Fiber-optic laser fluorescence equivalence ratio
probe 512
FID. See Fast flame-ionization detector
Field decomposition
interpretation using 283
Field equations
sources and sinks in 285
Filter
bandpass 122 126
frequency-tracking extended kalman 601
low-pass 198
Filtered backprojection methods 496
Finite element analysis (FEA) 549
Finite element method (FEM) 475
Finite-controller performance bandwidth 603
First radial (1R) 224
Flame 181 183 185 189
319 488
See also Heat release
acoustic field effected by 350
acoustic interactions with 265
acoustic/vortical perturbation responded to by 317
Index Terms Links
Flame (Cont.)
bifurcation of structure of 255
brush thickness of 352 353
computation of, disturbed by vortical
structure 321
conical 197 334
corrugations 337
counterflow 357
disturbance field effected by 326
geometries 329 559
global-heat release rate of 325
heat release monitored by chemiluminescence
and 523
heat release response to flame speed
perturbations and 346
hysteresis 257
impinging on plate by 192
incoming equivalence-ratio perturbations
responded to 203
inherent stabilities of 353
instabilities of 317
acoustic interactions with 353
during instability cycle 193
interacting with plate 194
interacting with wall 193
interactions of, with heat release boundaries 189
intrinsic premixed instabilities 353
Index Terms Links
Flame (Cont.)
laminar 317
lean hydrocarbon 486
as low-pass filter 198 204
modulated conical 197
mutual annihilations of 181 195
as nonlinear flame-transfer function 407 408 409
nonlinearities’ explicit form disappearing in 338
as not compact 384
overall shape of 497
perturbed, interactions with boundaries of 181
perturbed oblique 201
photographic images of stable/unstable 253
premixed 318 395
premixed swirl stabilized 548
response of, in general, nonlinear case 336
simple premixed 545
speed of, response to perturbations 342
stabilization 294
stable 253 254
step-stabilized 547
stoichiometric methane-air 344
strained 205
swirl-stabilized 547
transfer matrices of 446
turbulent 317
unstable 253
unsteady velocity jump across 327
Index Terms Links
Flame (Cont.)
vortex interactions with 267
wedge 329 334 335
wrinkled 352
wrinkled effects of 350
zero response of area of 349
Flame anchoring 356 536 560
Flame area 515 516
Flame aspect ratio 339
Flame bifurcation 255
vortex-flashback dictating 257
Flame blowoff 4
Flame disturbances 326
Flame dynamics 215
of axial-entry swirl injector 250
flame sheets and 324
of lean premixed swirl injectors 213
oscillatory 258
stable 251
Flame front 515
Flame geometries 559
conical 329
wedge-shaped 329
Flame instabilities 317
Flame interactions 183 265 277
Flame location 557
Flame model 455
time-delay spread on 456
Index Terms Links
Flame-front motion
Mach number of 351
Flame-geometry effects 559
Flame-response function
pressure trace superimposed on 501
Flame-response Rayleigh index 500 501
Flame-speed
sensitivity of, to equivalence ratio 345
Flame-speed perturbations 342
flame’s heat release’s responded to by 346
Flame-speed response 342
Flame-structure evolution 496
Flame-structure image sequences
two-dimensional 497
Flame-surface
heat-release fluctuations evaluations by 200
mutual flame annihilations control of
area of 197
Flame-trapping process 255
Flame-vortex dynamics 187
Flame-vortex interactions 181 481
in unstable combustions systems 187
Flashback 356
unsteady phenomena of 356
Flashback thermocouple 155
Flight propulsion engines 44
Index Terms Links
Fluctuations (Cont.)
temporal 492
velocity 185
vortex-driven 183
Fluid mechanic response measurements 120
Fluorescence 123
OH planar laser-induced 513
Fluorescence seeds 508
Fluorescence signal 506 508
Fluorescence tracers 513
Fluorescence yield 508
Forcing frequency 241 321
Forcing signals 117
Forward transfer function 537
Fourier transform 83
Frequencies 106 174 321 392
393 440 539 556
568
See also
High-frequency dynamics;
Intermediate-frequency dynamics;
Low-frequency dynamics; Midfrequency
dynamics
acoustic wave 185
annular duct and 451 471
baseline open-loop 567
characteristic 233
combustion dynamics and 168
Index Terms Links
Frequencies (Cont.)
of combustion instabilities 4 91
combustor pressure and 597
complex thermoacoustic systems and 462
of DDV 617
discrete sinusoidal 232
duct-cutoff 185
forcing 241 321
high, resonators 105
instability 23
intermediate 106
multiple resonant 543
natural acoustic 15
natural shedding 321
oscillation 22 440 552
pressure fluctuations and 221 222
resonant 539 543
sinusoidal 232
time lag’s relation to 555
transfer functions and 539
Frequency domain stability
analysis of gas-turbine combustor and 465
operating point effect 465
time-lag spread effect 466
Frequency dynamics
high 91
intermediate 91
low 91
Index Terms Links
Frequency modes
experimental data for 555
Frequency response 597
of annular duct 451
combustion dynamics and 168
Frequency-domain approach
for complex thermoacoustic systems 462
Frequency-switching behavior 555
Frequency-tracking extended kalman filter 601
Fuel actuator 135 606
Fuel component stimulants 513
Fuel composition 534
Fuel concentration 77
Fuel concentration measurements 512
Fuel feed line-acoustic coupling 10
Fuel flow actuator 448
Fuel flow distribution 45
Fuel flow rate
measurement of, at injector exit 520
overall chemiluminescence emission
related to 490
Fuel injection 213
Fuel injector
CFD and 458
importance of 118
LP gaseous 32
multiple time lags in 557
Index Terms Links
Gas (Cont.)
sampling 129
Gas flow 521
Gas turbine combustion dynamics
three-dimensional linear stability analysis 415
Gas turbine swirl injector 224
Gas-dynamic nonlinearities 22
Gas-phase fluorescence measurements 507
Gas-turbine combustion dynamics
three-dimensional linear stability analysis for 415
boundary conditions for 430
Greek symbols for 415
matching conditions 428
modal expansion and 421
overscripts and superscripts for 415
spatial averaging and 421
subscripts for 415
symbols for 415
system equations for 432
theoretical formulation for 418
treatment of inhomogeneous terms for 426
Gas-turbine combustion instability
three dimensional acoustic analysis
solution procedure for 433
Gas-turbine combustion systems 135 174
acoustic analysis of 369
aeroengine 113 143
AIC in 135 620
Index Terms Links
Inhomogeneous terms
combustion heat release and 426
mean flow’s effect on 426
surface condition effect of 427
treatment of 426
Inhomogeneous wave equation 381
Injector exit 240 520
Injector response
to external excitation 232
acoustic admittance at injector exit 240
instantaneous flow structures and 233
mass transfer function and 242
mean flow properties and 238
Injectors 520 589
See also Axial-entry
swirl injector; Coaxial swirl injector;
Fuel injector; Fuel injector-air swirler;
Radial-entry swirl injector; Swirl
injectors
air-blast swirl 224
fuel 118 458
gas turbine swirl 224
laboratory swirl 214
Lean premixed (LP) gaseous fuel 32
Lean-premixed swirl 213 215
liquid-fueled swirl 215
Mars 38
Index Terms Links
Injectors (Cont.)
operational 240
VFI 135
Inlet boundary conditions 379
Inlet combustor 431
Inlet guide vane (IGV) 152
Inlet swirl-vane geometry 287
Inlet velocity 287
Input-output relation 470
Instability characteristics 151
Instability driving mechanism 9
Instability frequencies 23
Instability prediction 445
instability-amplitude prediction 24
Instantaneous axial velocity fields 219 236
Instantaneous flow structures 233
Instantaneous flowfield 260
Instantaneous fluctuation pressure field 220
Instantaneous Karlovitz number 346
Instantaneous pressure field 352
Instantaneous total mass flow rate 243
Integrated liquid-crystal displays 619
Intermediate-frequency dynamics (IFD) 91
Intermediate-frequency resonators 106
International Organization for Standardization
(ISO) 89
Index Terms Links
Jet fuel
fluorescence of 123
Jet A 123
Jets
instabilities of 188
Jump condition 380 387 456
Oscillations (Cont.)
conditions of 22
equivalence-ratio 10
FAR 74 78 82
frequency of 22 440
heat-release 11 325
large-amplitude 189
large-velocity 354
limit-cycle 20 407
low-amplitude 5
lower-frequency 552
pressure 154 584
thermoacoustically induced 90 110
transverse 434
in unstable combustor 15
velocity 439
Oscillatory atomization 10
Oscillatory combustion 66
basic characteristics of 69
FAR-wave in 74
fuel splits changed in 73
instability control for 73
in Trent 60 DLE 68
Oscillatory flame dynamics 258
acoustic and flame interaction and 265
flame surface/heat release evolution and 263
instantaneous flowfield of 260
mean flow structures of 258
Index Terms Links
Pressure 12 70 84 126
130 132 134 220
247 352 393 482
494 501 508
See also Combustor pressure; Combustor
pressure oscillations
acoustic 439
acoustic, distribution 101
acoustic, perturbations 344
combustor spectra of 596
diffuser air 131
disturbance interaction 45
drops 128
dynamic spectrum of 92
dynamic, transducers 164
far-field radiated 194
field 352
fluctuating field of 248
heat-release as source for 187
high-response, transducers 129
instrumentation 148
mean 308 309
measured signal of 484
measurements 482
oscillations 154 584
perturbations 377 387
signal zero crossing 500
Index Terms Links
Pressure (Cont.)
simultaneous 505
source 187
spray mass flow-acoustic, transfer functions 126
transducer 483
unsteady 132
waves 195
Pressure fluctuations
frequency fluctuations of 221 222
in fuel line 484
in nozzle 484
predicted spectrum of, in combustion test rig
with a nonreflecting/reflecting exit 453
random noise generating 450
Pressure oscillations
analysis of 154
transport delay/limited actuator bandwidth’s
effect 584
Pressure-oscillation amplitude 70
Pressure-wave attenuator 86
Primary-zone temperature 72
Probability density distribution function 475
Probability density function (PDF) 251
Proper orthogonal decomposition (POD) 224
Propulsion 3
PSD. See Power-spectral density
Pulsed airflow 124
PVC. See Precessing vortex core
Index Terms Links
Rayleigh’s integral 4
Reacting flows 183
Reacting vortices 189
Recirculation bubble 290
Redheffer star products 467 472
Reduced order models 535
Redundant power supply 619
Reflection coefficients 452
Relative velocity 207
Remote data analysis 165
Resonant frequency 539 543
Resonators 105
See also Helmholtz
resonator
acoustic 85
AFR increased by 110
Bode plot effected by 568
configurations of 563
damper 564
for damping 564
geometry of 562
high-frequency 105 107
installing 563
intermediate-frequency 106
modeling of 109
parameters assumed in 567
placement of 106
positioning of 562 565
Index Terms Links
Space Shuttle 8
Spanwise vorticity 300
Spatial averaging 421 423
Spatially uniform velocity disturbance 330
Spectral analysis 150
Spectral characteristics 231
Spiral vortex 261
Spontaneous oscillatory combustion 68
Spray 122
combustion 304
combustions simulations 302
dispersion 303
fuel 124
liquid 215
liquid-fuel 517
modulated 519
Spray and reaction rate contours 306
Spray mass flow-acoustic pressure transfer
function 126
Spray-flow interactions 481 482
Spray-vortex interactions 482
Stability 463
borders of 446 464 466
of complex thermoacoustic systems 463
linear analysis of 262
linear/nonlinear 16
nonlinear un- 18
Index Terms Links
Stability (Cont.)
thermoacoustic 100
of transfer matrix 463
Stability analysis. See also Linear stability
analysis; Three-dimensional linear
stability analysis
Bode 538
CFD/FEA used by 549
Full 98
of linearized one-dimensional conservation 283
Nyquist 540
Stability boundaries 552
experimentally determined 553
experimentally determined for
multiple-frequency mode combustor 556
Stable flame 253 254
Stable flame dynamics 251
Stagnation enthalpy 386
State-space representation 462 467
of an annular duct 469
modal expansion and 468
Stationary gas turbines 533
Step duct 433
Step-stabilized flames 547
Stoichiometric methane-air flame 344
Stokes number 125 297 299 300
Straight chamber 436
Straight duct 435
Index Terms Links
Turbulent dispersion 79
Turbulent flame 317
Turbulent flow 459
Turbulent kinetic energy 259
Turbulent mixing 460
Turbulent spreading 461
Two-dimensional Mie-scattering 518
Two-mode combustor model 571
Two-phase interactions 419 420
Velocity (Cont.)
perturbations 195
radial 249
relative 207
shear-wave convection 322
unsteady 327
unsteady heat release’s relation to 336
Venturi 127 250
VFI. See Valve/feed-line/injector
Viscous damping mechanisms 14
Viscous dissipation 419 420
Volumetric dilation 282
Vortex
breakdown 226 270
burning 189
flame interactions with 267
merging 547
motion 277
rollup 187
shedding 11 23 189 536
periodic 288
spiral 261
structure 303
structure of 303
Vortex breakdown bubble (VBB) 290
Vortex-driven fluctuations 183
Vortex-droplet coupling 301
Vortex-flashback 257
Index Terms Links
Banaszuk, A. 581
Bellucci, V. 445
Bethke, S. 89
Blust, J. 29
Candel, S. 179
Cohen, J. M. 113 581
DeLaat J. 113
Dowling, A. P. 369
Ducruix, S. 179
Durox, D. 179
Flohr, P. . 89
Flohr, P. 445
Index Terms Links
Goy, C. J. 163
Held, T. J. 43
Hermann, J. 611
Hoffmann, S. 611
Hsiao, G. C. 43
Huang, Y. 213
James, S. R. 163
Johnson, C. 89
Krebs, W. 89
Lee, J. G. 481
Lepers, J. 89
Lieuwen T. C. 3 315
Menon, S. 277
Mongia, H. C. 43
Index Terms Links
Pandalai, R. P. 43
Paschereit, C. O. 445
Prade, B. 89
Proscia, W. 113
Rea, S. 163
Richards, G. A. 533
Robey, E. H. 533
Santavicca, D. A. 481
Sattinger, S. 89
Scarinci, T. 65
Schuermans, B. 445
Schuller, T. 179
Sewell, J. B. 147
Smith, K. O. 29
Sobieski, P. A. 147
Stow, S. R. 369
Straub, D. L. 533
Wang, S. 213
Index Terms Links