Sei sulla pagina 1di 163

Linear Modeling and Analysis of Thermoacoustic

Instabilities in a Gas Turbine Combustor


by

Christopher A. Fannin

Dissertation submitted to the Faculty of the


Virginia Polytechnic Institute and State University
in partial ful¯llment of the requirements for the degree of

Doctor of Philosophy

in

Mechanical Engineering

Approved:

Dr. William R. Saunders, Chair


Dr. William T. Baumann
Dr. Uri Vandsburger
Dr. Harry H. Robertshaw
Dr. Harsha Chelliah

July 2000

Blacksburg, Virginia

Keywords: acoustic, heat release, time delay, stability


Linear Modeling and Analysis of Thermoacoustic
Instabilities in a Gas Turbine Combustor

Christopher A. Fannin, Ph.D.


Virginia Polytechnic Institute and State University, 2000
Advisor: Dr. William R. Saunders, Chair

Abstract

A dynamic model is developed for the purpose of predicting stability characteristics of an


industrial-scale, swirl-stabilized premixed combustor located at the National Energy Technology
Laboratory (NETL) in Morgantown, WV. The model consists of modular blocks that assemble
into an open-loop transfer function depicting the frequency response of the thermoacoustic system.
These blocks include the system acoustic response to unsteady heat release forcing, the air-side
coupling of acoustic particle velocity to inlet fuel mass fraction, transport delays present in the
mixing nozzle and combustion chamber, and dynamic heat release excitation from unsteady inlet
fuel mass fraction. By examing the frequency response with linear stability techniques, the existence
of limit cycles due to linear instabilities is predicted. Further, the frequency response analysis is
used to predict limit cycle frequencies in the case of predicted instability. The analysis predictions
are compared with the results of tests performed at NETL, demonstrating a capability of replicating
many of the observed stability characteristics.

ii
To Susan
Acknowledgments

This work has been funded under the Department of Energy AGTSR subcontract # 98-01-SR065.
I owe special thanks to the Virginia Space Grant Consortium for their additional support. I would
like to thank Dr. George Richards and Mr. Douglas Straub from the National Energy Technology
Laboratory for their cooperation and assistance with this research.

Dr. William Saunders, my tireless and good spirited advisor, deserves my deepest gratitude.
Through all the time we've worked together, he has helped me to stay the course to achieve my
goals, even when the path was not at all well lit. He always made me feel that I was a valued asset
to his research program, o®ering me responsibility and seeking my opinion on a variety of matters.
As an advisor, he gave me the freedom to persue my own academic interests, often putting my
needs above his own. Working for Dr. Saunders has been a great pleasure. I would like to wish him
well for the future and congratulate him on his gaining tenure this year.

My other committee members have also helped me a great deal. Dr. Bill Baumann's enthu-
siasm for his job can only be described as inspiring. He has forced me to be critical of my work by
never taking anything for granted. Dr. Harry Robertshaw has been my mentor for all the teaching
that I have done as a graduate student. Whenever I had a problem as an instructor, he always
had the right insight to help me deal with it. He made me realize that being a good teacher is not
about having all the answers all the time, but about guiding students and allowing them to struggle
with the material so that they can really \own it." I will miss working with him very much. Dr.
Vandsburger has instructed me in the many mysteries of combustion science, and he has always
been around for a nice political discussion to break the monotony of a long work day.

iv
There are several students in the department who have been extremely helpful over the
course of my completing this research. Particularly, Scott Liljenberg, who has provided me with
many valuable insights with his acoustics research, not to mention being a ¯ne o±ce mate. Vivek
Khanna was always been available for me to consult on °uid mechanics and combustion as well as
life in general. I also want to thank all the other folks in the lab and around the department for
being around to bounce around ideas and to share successes and failures.

I am extremely grateful for the love and support of my family and friends. I especially want
to thank my dear Susan for her love and friendship and for being my balance these last few years.
I can't wait to begin our lives together as husband and wife.

Christopher A. Fannin

Virginia Tech
July 2000

v
Contents

Abstract iii

Acknowledgments iv

List of Tables ix

List of Figures x

Nomenclature xiv

Chapter 1 Introduction 1
1.1 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Combustor Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Heat Release Variation Caused by Air-Side Coupling . . . . . . . . . . . . . 6
1.1.3 Multiple and Single Transport Delays . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Brief Description of Thermoacoustic Instability . . . . . . . . . . . . . . . . 8
1.2.2 Modeling of Thermoacoustic Instabilities . . . . . . . . . . . . . . . . . . . . 11
1.2.2.1 Dynamic Flame Models . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.3 Acoustic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.3.1 Acoustic Finite Element Methods . . . . . . . . . . . . . . . . . . . 16
1.3 Organization of the Dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Chapter 2 Acoustic Finite Element Model 19


2.1 Modeling Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

vi
2.1.1 Source Strength and Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1.1.1 Scaling of Pulsating Cylinder to Unsteady Heat Release Rate . . . 24
2.1.2 Fluid Properties From CFD . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Model Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.1 Fluid Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.3 Geometry Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.4 Distributed Forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

Chapter 3 Acoustic Model Validation 37


3.1 Acoustic Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.1 FEM Mesh Re¯nement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.1.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.2 Sensitivity to Geometric Tolerances . . . . . . . . . . . . . . . . . . . . . . . 45
3.2.3 Speaker Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Comparison of Model to Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.1 Geometry E®ects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.2 Boundary Condition E®ects . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Chapter 4 Unsteady Well-Stirred Reactor 64


4.1 Model Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1.1 Conservation Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1.2 Mean Conditions Using CFD Results . . . . . . . . . . . . . . . . . . . . . . 71
4.1.3 Linearization and Conversion to State Space . . . . . . . . . . . . . . . . . . 73
4.1.4 Inlet Fuel Mass Fraction Variation . . . . . . . . . . . . . . . . . . . . . . . 76
4.2 Modeling Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2.1 E®ects of Mean Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Chapter 5 Time Delay and Multi-Port Fuel Injection 87


5.1 Linear Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

vii
5.1.1 ¿ f Plotting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.2 Multiple Time Delays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.1 Delay Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.2.2 Multi-Port Injector Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Chapter 6 Stability Predictions and Conclusions 104


6.1 Closed-Loop Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1.1 Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.1.2 Time Delay and Inlet Fuel Mass Fraction Scaling . . . . . . . . . . . . . . . 107
6.1.3 Flame Dynamics and Acoustic Input Scaling . . . . . . . . . . . . . . . . . . 108
6.1.4 Composite Loop Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 Stability Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2.1 Predicted and Observed Frequencies . . . . . . . . . . . . . . . . . . . . . . 113
6.2.1.1 Spectrum Comparison . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

Bibliography 133

Appendix A NETL Testing Power Spectra 137

Appendix B NETL CFD Model Results 141

Vita 147

viii
List of Tables

2.1 Combustion chamber temperature distribution based on CFD estimates . . . . . . . 27

4.1 Mean Operating Conditions for Flame Model . . . . . . . . . . . . . . . . . . . . . . 72

6.1 Time delay values used for analysis (values shown in ms). . . . . . . . . . . . . . . . 114
6.2 Stability Frequency Data Taken at NETL. . . . . . . . . . . . . . . . . . . . . . . . 116
6.3 Stability Frequency Data Predicted by Stability Model. . . . . . . . . . . . . . . . . 117
6.4 Di®erences Between Data and Stability Prediction Frequencies. . . . . . . . . . . . . 118

B.1 Combustor inlet and exit results from CFD runs . . . . . . . . . . . . . . . . . . . . 141

ix
List of Figures

1.1 Stability analysis °owchart. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2


1.2 Block diagram of closed-loop thermoacoustic system . . . . . . . . . . . . . . . . . . 4
1.3 Graphical depiction of ¯rst mode instability in a Rijke tube. (a) First mode pressure
and velocity mode shape. (b) Rayleigh Index. . . . . . . . . . . . . . . . . . . . . . 10

2.1 Isometric diagram of NETL combustion rig. . . . . . . . . . . . . . . . . . . . . . . 23


2.2 Axisymmetric FEM mesh for NETL combustion rig. . . . . . . . . . . . . . . . . . . 23
2.3 Plenum and nozzle detail of NETL rig FEM mesh. . . . . . . . . . . . . . . . . . . . 23
2.4 Fluid property distribution in combustor region. . . . . . . . . . . . . . . . . . . . . 26
2.5 FRF comparison for hot and cold °uid temperatures in NETL combustion rig. . . . 28
2.6 FRF comparison for equivalence ratio e®ects on acoustic response. . . . . . . . . . . 29
2.7 FRF comparison for exhaust gas temperature e®ects on acoustic response. . . . . . 30
2.8 FRF comparison for inlet air temperature e®ects on acoustic response. . . . . . . . 31
2.9 FRF comparison for various boundary conditions on exhaust line. . . . . . . . . . . 32
2.10 FRF comparison for tolerances on refractory plug size and placement. . . . . . . . . 34
2.11 FRF comparison for a single phase and distributed phase forcing function. NOTE:
the distributed force has a 180 degree phase distribution at 400 Hz . . . . . . . . . . 35

3.1 NETL combustion rig. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39


3.2 FRF's of long and short NETL combustor rig at room temperature. . . . . . . . . . 40
3.3 Plenum side of NETL rig showing speaker location for acoustic tests. . . . . . . . . 41
3.4 Mesh of long rig with microphone node locations indicated. . . . . . . . . . . . . . . 43
3.5 Mesh of long rig with combustor plug removed. . . . . . . . . . . . . . . . . . . . . 43

x
3.6 Mesh of long rig with combustor plug and mixing nozzle removed. . . . . . . . . . . 43
3.7 Mesh re¯nement test mesh #1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.8 Mesh re¯nement test mesh #2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.9 Mesh re¯nement test mesh #3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.10 FRF's of mesh convergence meshes 1, 2, and 3 at measurement location 1. . . . . . 46
3.11 Ratios of mesh convergence FRF's at location 1. . . . . . . . . . . . . . . . . . . . . 46
3.12 FRF's of mesh convergence meshes 1, 2, and 3 at measurement location 2. . . . . . 47
3.13 Ratios of mesh convergence FRF's at location 2. . . . . . . . . . . . . . . . . . . . . 47
3.14 FRF's of mesh convergence meshes 1, 2, and 3 at measurement location 3. . . . . . 48
3.15 Ratios of mesh convergence FRF's at location 3. . . . . . . . . . . . . . . . . . . . . 48
3.16 FRF's of opposite extreme tolerance limits for the NETL long rig at location 3. . . 49
3.17 FRF's of opposite extreme tolerance limits for the NETL long rig at location 4. . . 49
3.18 FRF's of opposite extreme tolerance limits for the NETL long rig at location 5. . . 50
3.19 FRF's of opposite extreme tolerance limits for the NETL long rig at location 6. . . 50
3.20 Experimental and theoretical FRF's of NETL long rig measured at location 4. . . . 56
3.21 Experimental and theoretical FRF's of NETL long rig with no plug measured at
location 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.22 Experimental and theoretical FRF's of NETL long rig with no plug and no nozzle
measured at location 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.23 Experimental and theoretical FRF's of NETL long rig measured at location 3. . . . 59
3.24 Experimental and theoretical FRF's of NETL long rig measured at location 5. . . . 60
3.25 Experimental and theoretical FRF's of NETL long rig measured at location 6. . . . 61
3.26 Experimental and theoretical FRF's showing boundary condition e®ects at location 3. 62
3.27 Experimental and theoretical FRF's showing boundary condition e®ects at location 5. 63

4.1 FETC combustion rig mixing nozzle (not to scale) . . . . . . . . . . . . . . . . . . . 77


4.2 Input/Output relationships for each of the three inputs to the °ame model . . . . . 82
4.3 Input/Output relationships with all inputs related back to acoustic velocities . . . . 82
q0
4.4 E®ect of equivalence ratio variation on Yfin
°ame transfer function for a an inlet
mean velocity of 50 m/s. Arrows indicate increasing Á. . . . . . . . . . . . . . . . . 86

xi
q0
4.5 E®ect of mean °ow variation on Yfin
°ame transfer function for an equivalence ratio
of 0.67. Arrow indicates increasing mean °ow rate. . . . . . . . . . . . . . . . . . . 86

5.1 Single mode example of resonant system. . . . . . . . . . . . . . . . . . . . . . . . . 91


5.2 Graphical interpretation of ¿ f product. . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3 Frequency response functions of multiple time delays. . . . . . . . . . . . . . . . . . 94
5.4 Equally spaced linear array. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.5 Frequency response amplitude of 5 time delays showing e®ect of amplitude shading
using constant side lobe level Dolph-Chebychev array design. . . . . . . . . . . . . . 99
5.6 Frequency response amplitude of 5 time delays showing optimization results. . . . . 101
5.7 Frequency response amplitude of 5 time delays showing optimization results with
objective function augmented by acoustic gain penalty. . . . . . . . . . . . . . . . . 103
5.8 Composite amplitude response showing the product of the acoustic gain and time
delay array gains for two optimizations. . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.1 Block diagram of open loop transfer function. . . . . . . . . . . . . . . . . . . . . . 105


6.2 Nozzle particle velocity response to pulsating cylinder velocity input. . . . . . . . . 106
6.3 Time delay and velocity to inlet fuel mass fraction gain for injector location B, mean
velocity 50 m/s (¿ = 5.1 ms). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.4 Flame dynamics and acoustic input scaling transfer function. . . . . . . . . . . . . . 108
6.5 Open-loop transfer function for thermoacoustic plant. . . . . . . . . . . . . . . . . . 109
6.6 ¿ f plot of experimental (circles) and theoretical (stars) instabilities. . . . . . . . . . 111
6.7 Nyquist diagram of zero-delay loop gain. . . . . . . . . . . . . . . . . . . . . . . . . 113
6.8 Histogram showing occurances of instability frequencies in NETL data. . . . . . . . 114
6.9 Histogram showing occurances of instability frequency predictions. . . . . . . . . . . 115
6.10 Comparison of PSD data and loop gain stability prediction for injector B, 30 m/s. . 120
6.11 Comparison of PSD data and loop gain stability prediction for injector B, 40 m/s. . 121
6.12 Comparison of PSD data and loop gain stability prediction for injector B, 50 m/s. . 122
6.13 Comparison of PSD data and loop gain stability prediction for injector B, 60 m/s. . 123
6.14 Comparison of PSD data and closed-loop tranfer function for injector B, 30 m/s. . . 124
6.15 Comparison of PSD data and closed-loop transfer function for injector B, 40 m/s. . 125

xii
6.16 Comparison of PSD data and closed-loop transfer function for injector B, 50 m/s. . 126
6.17 Comparison of PSD data and closed-loop transfer function for injector B, 60 m/s. . 127
6.18 Injector array amplitude responses for a 5 injectors spaced 6 cm apart. . . . . . . . 132

A.1 Power spectral densities for injector location A (8.3 cm upstream of combustor inlet). 138
A.2 Power spectral densities for injector location B (14.4 cm upstream of combustor inlet).138
A.3 Power spectral densities for injector location C (19.4 cm upstream of combustor inlet).139
A.4 Power spectral densities for injector locations A and B. . . . . . . . . . . . . . . . . 139
A.5 Power spectral densities for injector locations A and C. . . . . . . . . . . . . . . . . 140

B.1 Test condition \a" temperature distribution (K) . . . . . . . . . . . . . . . . . . . . 142


B.2 Test condition \a" axial velocity distribution (m/s) . . . . . . . . . . . . . . . . . . 142
B.3 Test condition \b" temperature distribution (K) . . . . . . . . . . . . . . . . . . . . 143
B.4 Test condition \b" axial velocity distribution (m/s) . . . . . . . . . . . . . . . . . . 143
B.5 Test condition \c" temperature distribution (K) . . . . . . . . . . . . . . . . . . . . 144
B.6 Test condition \c" axial velocity distribution (m/s) . . . . . . . . . . . . . . . . . . 144
B.7 Test condition \d" temperature distribution (K) . . . . . . . . . . . . . . . . . . . . 145
B.8 Test condition \d" axial velocity distribution (m/s) . . . . . . . . . . . . . . . . . . 145
B.9 Test condition \e" temperature distribution (K) . . . . . . . . . . . . . . . . . . . . 146
B.10 Test condition \e" axial velocity distribution (m/s) . . . . . . . . . . . . . . . . . . 146

xiii
Nomenclature

a Angle of incoming plane wave to a linear array


A Cross-sectional area or amplitude as in an array
AF Array factor of a linear antenna array
A=F Air to fuel (mass) ratio
Bl Electromechanical coupling
c Characteristic acoustic speed
C Speaker damping coe±cient
Cp Constant pressure speci¯c heat
d Spacing of array elements
Dx Matrix of partial derivatives with respect to states in x
ex Speci¯c internal energy of species x
f Frequency in Hz
F=A Fuel to air (mass) ratio
hx Speci¯c enthalpy of species x
i Complex variable or Current
k Wavenumber (de¯ned as !c )
K Speaker sti®ness
L Length or Voice coil inductance
m Mass
M Speaker mass

xiv
MW Molecular weight
P Acoustic pressure
Q Total heat release rate (Watts)
^s
Q Acoustic source strength amplitude
r Radius
R Rayleigh index or Voice coil resistance
t Time variable
T Temperature or period of oscillation
u Velocity
U Laplace tranform of velocity
V Total volume of a control volume or Voltage applied to speaker
x Spatial Variable
[X] Molar concentration of species X
Yx Mass fraction of species x ( M Wx½ [Xx ] )
° Ratio of speci¯c heats
±(¢) Dirac delta function
r2 Laplacian operator
½ Density
F=A
Á Equivalence Ratio ( F=Astoichiometric )
à Angular coordinate variable for linear arrays
¿ Time delay
¿s Time delay spacing in an equally spaced delay array
! Frequency in radians per second
!_ Molar production rate term

xv
Subscripts & Superscripts

(x)0 Perturbation of variable x (as in an acoustic perturbation)


¹
(x) Mean variable x
_
(x) Time rate of change of variable x
(x)f Referring to a fuel species
(x)o Referring to an oxidizer species
(x)p Referring to a product species
(x)in At the inlet (of a control volume or chamber)
(x)out At the exit (of a control volume or chamber)
(x)s At the location of the (acoustic) source
x Boldface indicates a vector quantity
fxgT Transpose of matrix or vector x

xvi
Chapter 1

Introduction

The problem of thermoacoustic instabilities in gas turbine combustors has become a topic of concern
in recent years. Included in the research e®ort has been a considerable amount of work directed
at understanding and properly modeling the underlying phenomena that lead to the instabilities.
Many models have been developed, but they vary widely in complexity and accuracy. In part, the
variety of models is due to the many forms of thermoacoustic instability phenomena that have been
identi¯ed. Since all of these involve reacting °ows in one way or other, attempts to model the exact
phenomena can become mired in the complexity of the °uid mechanical equations. Attempts at
predicting behavior while preserving a degree of modeling simplicity have led to empirical studies
and a wide array of simplifying assumptions.

For this research, a systems level approach was taken with the modeling. The unsteady rate
of heat release due to combustion acts as a forcing term on the acoustics of the combustor. The
acoustic velocity °uctuations then feed back, augmenting the unsteadiness in heat release rate. The
system was modeled as a feedback network with the components broken down into individual blocks
which interact in a closed-loop sense. The stability of this closed loop was used to indicate whether
or not the thermoacoustic system would be self-excited. The following list, and the °owchart in
Figure 1.1 outline a complete modeling approach that can be used to assess the dynamic behavior of

1
Acoustic Finite Experimental Verification
Element Model
Temperatures

Input Scaling Multiple Time Delays Steady-State CFD

Temperatures &
Array Design Mean Velocities

STANJAN
WSR Dynamic Species
Flame Model Concentrations

Loop G ain Stability Analysis

Figure 1.1: Stability analysis °owchart.

a thermoacoustic system, and determine limit cycle frequencies for self-excited oscillations exhibited
by that system:

² Combustor acoustics were modeled using ¯nite element methods, and veri¯ed with cold
acoustic rig measurements.

² Flame dynamics were represented using an unsteady, well-stirred reactor model with steady-
state CFD used in determining mean variables. The unsteady fuel/air ratio, caused by acoustic
particle velocity perturbations inside mixing nozzle was the key input to the °ame model.

² Time delays were chosen based on mean °ow velocities and fuel injector locations. Length
corrections were added to account for travel time inside combustor.

² Appropriate gains were chosen to couple all components into an open-loop transfer function
of the system.

² Frequency response stability analysis of the open-loop transfer function was used to predict
self-excited oscillation frequencies.

2
Though the use of a feedback network is not new, a detailed model for predicting thermoa-
coustic instability frequencies in an industrial scale combustor based on linear, loop gain predictions
has not been seen. Analyses include the integration of multi-port fuel injection, a technique that
has empirically demonstrated a degree of success for abating thermoacoustic instabilities passively.
A design methodology is proposed which would use an array of fuel injectors to decrease the gain in
the system, reducing the tendency toward self-excited oscillations. The closed-loop stability model,
with the use of frequency response analysis techniques, facilitated capturing the true loop gain in
the feedback system, and was shown to predict limit cycle frequencies similar to those observed
experimentally.

In the following sections, the motivation for this research e®ort and background information
are presented. The remainder of the dissertation is organized into chapters discussing individual
components of the analysis scheme including the ¯nite element model, the °ame dynamic model,
and time delay model, particularly for the case of multi-port fuel injection. The work concludes
with the ¯nal assembly of these components into an open-loop transfer function and comparison of
frequency response stability predictions with experimental data. A discussion of topics raised for
future research is also included.

1.1 Research Objectives

The purpose of the present work is to develop feedback models which provide accurate and insightful
dynamic representations of thermoacoustic phenomena in a full-scale laboratory combustor located
at the National Energy Technology Laboratory (NETL) in Morgantown, West Virginia. The com-
bustor features a swirl-stabilized, lean premixed °ame in a can-shaped combustion chamber with
a long afterburner-like exhaust section. It was built to study, among other things, the e®ects of
transport delays in the mixing nozzle on combustion stability.

The analytical tools developed in this research should be applicable to a variety of geometries

3
Unozzle
Acoustics Uinlet
(FEM) Mixture
Uexit Ratio

Q’ Flame Dynamics
τ
(Unsteady WSR) Yfuel,in

Figure 1.2: Block diagram of closed-loop thermoacoustic system

and serve as a means for incorporating thermoacoustic models into the design process. They should
facilitate development of combustion systems that are less susceptible to thermoacoustic instability.
Each of the items in the following list represents a building block for a systems-level approach
to modeling thermoacoustic instabilities. Without addressing all of these items, a reliable model
cannot be developed. Figure 1.2 shows how the components of the system interact. The sections
that follow will elaborate on the details of each of these modeling tasks.

² Accurate modeling of the system acoustics in the presence of a °ame.

² Appropriate representation of the acoustic forcing provided by the °ame.

² Reliable depiction of heat release variation caused by acoustic perturbations.

² E®ects of time delay(s) on phase and magnitude response.

1.1.1 Combustor Acoustics

The acoustics of the combustion chamber and its surrounding connections are important to cap-
turing the feedback coupling to the unsteady heat release rate. Unsteady heat release forcing of

4
the acoustic wave equation is derived and discussed in detail in Section 1.2.3. Though acoustic
modeling appears to be the most straightforward task at ¯rst glance, geometry of the combustor
and temperature distributions in the presence of a °ame pose nontrivial modeling challenges.

Most thermoacoustic instability researchers have modeled the acoustics the same way; they
use acoustic transmission lines similar to those used for designing mu²ers and ventilation ducts
[21]. One of the problems with using this method is it is a one-dimensional technique. Detailed
geometric variations require appropriate estimates for impedances. Extreme care must be exercised
when making such estimates with one-dimensional foundations, not to misrepresent the physics.

An improved means of discretizing a system is with the use of ¯nite element techniques.
Such models can provide modal and frequency response information on the system. Because of the
geometric complexities of both the combustor, °uid property distribution, and the acoustic forcing
function, ¯nite element techniques can provide a more °exible alternative to using transmission
elements.

The distribution of heat release rate to the acoustics is an issue that previous researchers
have not fully considered. It is generally accepted that thermoacoustic instabilities occur chie°y
when the °ame is spatially compact. Based on arguments for air-side coupling, which is discussed
in Section 1.1.2 below, there is reason to believe that the °ame may not act as a monopole type
source, rather as an undulating, and spatially distributed forcing function. The appropriate forcing
function must be included in the analysis, as it can a®ect the frequency response, particularly the
phase response, of the acoustic block shown in Figure 1.2. In this case, it will be shown that a force
distribution behaves much like an additional time delay in the system.

In addition to distribution of heat release rate along the °ame, the physical size of the °ame
itself is also signi¯cant to the stability characteristics of the combustor. The °uid mechanics which
dictate the shape of the °ame are based on swirling, reacting °ows. Due to the complexity of such a
°ow ¯eld, a CFD task to determine mean °ow characteristics will signi¯cantly enhance the integrity
of a combustor stability model by providing much needed information about the distribution of

5
temperature in the combustor. Using Fluent, steady-state, steady °ow CFD runs provide mean
temperature and velocity information for integration into acoustic ¯nite element models.

1.1.2 Heat Release Variation Caused by Air-Side Coupling

The means by which the acoustics a®ect unsteady heat release rate serves as the feedback mech-
anism for self-excitation. There are many types of thermoacoustic coupling. However, based on
experiments performed at NETL and elsewhere, one that is believed to be signi¯cant for many
applications is known as air-side coupling. The basic premise is that the unsteady velocity due to
acoustic perturbations leads to a variation in stoichiometry in the mixing nozzle of the combustor.
This variation is convected by the mean °ow to the °ame where it reacts.

The other important consideration with regard to heat release rate is the e®ect of reaction
kinetics on the dynamics of the °ame itself. If the reaction occurs quickly, then the variation of
heat release (it can be argued) can be attributed to variations of the mixture strength that are
taking place on the acoustic scale of time. Typically, this is not a bad assumption provided the
mixture is not too lean or rich. Lieuwen et. al. [13] discussed this issue, along with the e®ects of
partial mixing, using a well-stirred reactor model. They showed that the bandwidth of the unsteady
combustion process was related to the characteristic ignition time, which increases signi¯cantly close
to the °amability limit. The high gradient of ignition time at lean conditions suggests that slight
perturbations in Á can result in large variations in reaction rate. This observation led the authors
to conclude that unmixedness could play a key role in combustion instabilities as the reaction rate
would vary greatly under partial mixing conditions.

In this research, a °ame model based on an unsteady well-stirred reactor similar to that used
by Lieuwen et. al. [13] is used to determine the dynamic response of heat release rate to perturbations
in the reactant fuel/air ratio. This model will provide the proper gain scaling to enable unsteady
acoustic velocities to drive heat release variations which, in turn, feed back, forcing the acoustic
¯eld. It serves as a simpli¯ed °ame model, but is not intended to predict °ame behavior. Its

6
purpose is to allow a closed-loop, linear analysis of thermoacoustic stability by accounting for gain
through an unsteady heat release mechanism.

1.1.3 Multiple and Single Transport Delays

As mentioned in the previous section, there is a convective transport delay (of stoichiometry vari-
ations) involved in the coupling process. The researchers at NETL have designed a mixing nozzle
which allows them to change that delay by changing the distance of the fuel spokes from the com-
bustor entrance. It also allows them to split the fuel injection between two or three locations in the
nozzle, for multiple time lags. Experiments have demonstrated that the time delays associated with
these various fuel injection arrangements can a®ect the dynamical response of the combustor. It is
also possible that the time delays can a®ect the mean °ame distribution due to partial premixing
e®ects, again raising issues that may be best dealt with using CFD, but which are not to be explored
in this research e®ort.

A speci¯c objective of the research is to develop analyses which allow designers to choose
transport delays that minimize the possibility of instability, so a thorough understanding of the
e®ects of time lag will be needed. In order to properly predict the trends that are observed in the
NETL data, time lag must be included since that was the only variable in many of their experiments.

Interestingly, transport delays in the mixing nozzle are not necessary for self-excited oscilla-
tions to occur. Because of the importance of phasing between the unsteady heat release rate and
acoustic pressure, the acoustic mode shapes and dynamic behavior of the °ame itself dictate the
stability characteristics of the system. Time delays will add phase to a system, and when more
than one delay exists, the amplitude can be attenuated over a range of frequencies as well. So, they
can cause a system that was otherwise stable to become unstable and vice versa. From a design
standpoint, changing a system delay to enhance stability is an attractive alternative.

7
1.2 Literature Review

This section describes previous research about the history, nature, and physics of thermoacoustic
instability as well as provide information on basic system theory, time delay systems, and acoustic
modeling techniques. The volume of research that surrounds the topic of thermoacoustic instability
extends back over a century in time and could be characterized as daunting. However, it appears
that systems level, block diagram types of analyses have gained considerable attention for modeling
these phenomena only recently. This is possibly due to the recently developed interest in using sys-
tems theory for active control of thermoacoustic instabilities. Despite the advent of computational
techniques for modeling °uid and combustion systems, a reduced-order, top-down approach remains
an attractive analysis solution due to its relative simplicity and use of well-established stability the-
ory. The following sections review a variety of viewpoints used in the analysis of thermoacoustic
instabilities.

1.2.1 Brief Description of Thermoacoustic Instability

Thermoacoustic instabilities have been the focus of extensive research not only as they relate to lean
premixed combustion in gas turbines, but in rockets, ramjets, and as phenomena unto themselves.
In 1878, Lord Rayleigh stated his famous criterion describing the mechanism by which heat release
rate excites acoustic vibration: [28]

\If heat be periodically communicated to, and abstracted from, a mass of air vibrating
(for example) in a cylinder bounded by a piston, the e®ect produced will depend upon
the phase of the vibration at which the transfer of heat takes place. If heat be given
to the air at the moment of greatest condensation, or be taken from it at the moment
of greatest rarefaction, the vibration is encouraged. On the other hand, if heat be
given at the moment of greatest rarefaction, or abstracted at the moment of greatest
condensation, the vibration is discouraged."

8
Since then, researchers extending into the modern day have referred to Rayleigh's criterion
and used it to determine the tendency of acoustical systems toward instability. In a short commu-
nication, Culick [6] showed the relationship between the normal modes of a forced acoustic plant
and the energy added or abstracted during a cycle. In other words, Rayleigh's criterion can be cast
in many ways including the input-output relationship of an acoustic system. A similar argument
has been posed by Annaswamy et al. [1], which shows that the Rayleigh criterion relates directly to
stable or unstable closed-loop system poles. The most common form of Rayleigh's criterion is the
integral form shown below. If the Rayleigh index, R, is evaluated to be positive, there is energy
being added during a cycle and the vibration is encouraged. If the index is negative, the vibration
is damped.

Z T
R= P 0(t)Q0 (t) dt (1.1)
0

R > 0 ) unstable; R < 0 ) stable

The phenomenon identi¯ed by Rayleigh has shown its presence in many types of systems. The
well-known Rijke tube is an example of self-excited acoustic oscillation that is so basic in form, one
could be set up within minutes using only a glass tube and a blowtorch. In fact, many experimental
test rigs for the study of thermoacoustic instability have made use of this basic geometry. A
survey paper by Raun et. al. [27] summarizes the work on Rijke tubes quite well. It is well
understood that placing the heat source in the lower half, particularly at the quarter-tube location,
of a Rijke tube results in self-excited oscillation, while placing the source in the upper half will lead
to attenuation of oscillations. Figure 1.3 shows the distribution of pressure, velocity, and Rayleigh
index, a quantitative realization of Equation 1.1, with the assumption that Q0(t) will vary directly
with acoustic velocity, along the length of a Rijke tube. This is a graphical illustration of the relation
between acoustic mode shape and Rayleigh's criterion as discussed by Culick [6]. The ¯gure shows
how the location and therefore the phasing of the heat source relative to the acoustic mode shape
determines whether energy is added or removed during a cycle.

9
P'

u'

- + - +
R

u
Source

(a) (b)

Figure 1.3: Graphical depiction of ¯rst mode instability in a Rijke tube. (a) First
mode pressure and velocity mode shape. (b) Rayleigh Index.

Raun et. al. [27] discuss the physical mechanisms for heat driven oscillations but a brief
summary will provide the background necessary for a basic understanding of thermoacoustic phe-
nomena. First, there are several types of heat sources discussed including hot wire gauze, wall
temperature di®erences in the tube itself, and of course, combustion. Each type of source transfers
heat to the acoustic ¯eld by a di®erent mechanism, but all of them (it can be argued) get their
unsteadiness from a coupling with the (unsteady) velocity ¯eld.

Any heat source will lead to mean convection velocity through the tube itself if the tube is
oriented vertically such that buoyancy will cause an upward °ow. If an acoustic ¯eld is present,
there will be an unsteady velocity associated with the acoustic oscillation which superposes on the
mean °ow already established. The unsteady velocity leads to unsteady heat transfer to the °uid.
For a wire gauze, heat convection is enhanced by higher velocity, leading to more heat transfer to
the °uid when unsteady velocity is greatest. Similarly, combustion is enhanced by higher velocity
as the °ame must consume the incoming mixture at a greater rate. Even a bed of burning coals can
couple by the oscillation of the velocity causing a varying supply of oxidizer, thereby changing the

10
burning rate, and therefore the heat added to the °uid, during each cycle. These are good examples
of how unsteady velocity forms a direct coupling to unsteady heat release rate.

As stated by Rayleigh, if the heat release is maximum at the time the pressure is maximum,
the vibration is encouraged. In a Rijke tube (open at both ends), the ¯rst pressure mode shape has
nodes at each end and an antinode in the center. The acoustic velocity for the same mode has a
node at the center and is positive in the lower half and negative in the upper half when the pressure
is positive. Thus, the total velocity is greatest at the lower open end of the Rijke tube when the
pressure is maximum. Assuming the unsteady heat release rate varies in phase with the unsteady
velocity, the Rayleigh index is maximum at the quarter-tube location as shown in Figure 1.3. The
Rijke tube \sings" when the heat source is placed in the lower half of the tube because the Rayleigh
Index is positive. Placing the heat source in the upper half would result in damping of the acoustic
¯eld by a similar argument.

This is exactly the same type of behavior that occurs in a gas turbine combustor which
is su®ering from thermoacoustic instabilities. The acoustic ¯eld in the combustor is such that
Rayleigh's criterion is satis¯ed for some or all of the gas turbine operating conditions. In addition
to the phasing of the normal modes of the combustion chamber, transport lags in the system (e: g:
mixing of fuel and oxidizer and convection of the mixture to the °ame) can in°uence the relative
phasing of the acoustic pressure and unsteady heat release rate. The following section will outline
some of the methods currently employed for modeling these phenomena in a gas turbine combustor.

1.2.2 Modeling of Thermoacoustic Instabilities

Despite a large volume of literature and research on the subject, there is no consensus on the
correct method for modeling thermoacoustic instabilities in gas turbines. The most likely reason
for this is that there are a host of dynamic mechanisms that can potentially contribute. Since
the acoustic source often consists of a swirl-stabilized °ame, dynamics associated with the °uid
mechanics and chemical kinetic reactions can easily multiply into an intractable modeling task.

11
Unsteady °ow ¯elds provide coupling between acoustics and heat release in many ways, providing a
multitude of potential driving mechanisms for thermoacoustic self-excitation. Candel [3] discusses
a variety of combustion stability mechanisms including pulsating instabilities, periodic extinction,
hydrodynamic instabilities including vortex rollup, and parametric °ame instabilities, to name a
few. Reasonable simpli¯cations for these complicated dynamics have been the subject of much of
the work in thermoacoustic modeling.

There appear to be two main schools of thought on the root causes and acoustic coupling of
unsteady heat release in modern lean-premix, gas turbine combustors. The instabilities are either
caused by a pulsation in rate of the °ame due to some °uctuation in mixture strength or mass °ux,
or they are strictly connected to a °ow instability, such as vortex rollup. Because of the complexity
of each of these mechanisms, most research in the literature tends to concentrate on one or the
other individually. The di®erences between these two mechanisms may likely be geometry driven,
but there are cases where the geometry does not plainly dictate which would be expected to play a
dominant role.

For modeling combustors driven by pulsation type instabilities, it is generally assumed that
the oscillation in heat release rate comes about as a result of stoichiometric variations in incoming
mixture from variations in air mass °ux upstream in the mixing nozzle of the combustor. Assuming
the fuel addition is choked for simpli¯cation, the mixture of fuel and air is varied by the oscillating
acoustic velocity, thereby varying the equivalence ratio, Á. This type of e®ect of the acoustic ¯eld
on the heat release is often referred to as air side coupling. Lieuwen et al. [14] note that, especially
for ultra lean conditions, small variations in Á can result in large swings in the reaction rate and
therefore in the amount of heat liberated per unit mass. There is considerable experimental support
in the literature for the air side coupling assumption. Bloxsidge et al. [2] noted that for low Mach
number °ows, the fractional change in velocity is much larger than the fractional change of other
°ow variables. As such, the velocity °uctuations due to acoustic pressure variations can have a
signi¯cant e®ect on the unsteady mass °ux of air in a mixing nozzle leading to oscillations in
the equivalence ratio and heat release rate. Many investigations have shown data to support this
phenomenon including work by Lieuwen and Zinn [15], Murray et al. [22], Peracchio and Proscia

12
[24], and Richards et al. [29], [30].

Also included in many of these studies is time delay. There is a transport lag from the mixing
nozzle to the reaction zone. By intentionally adjusting this time lag, researchers have investigated
the possibility of changing the stability mapping of the combustor by adjusting the critical phase
relation between unsteady heat release rate and acoustic pressure (i: e: the Rayleigh criterion).
KrÄ
uger et al. [12] were interested in similar types of coupling, but considered the unsteady heat
release rate to be dependent more on unsteady mass °ux than stoichiometric variations. Using a
CFD model of their combustor to compute the response to a step change in mass °ow rate through
the main swirler, they constructed frequency response characteristics for heat release reacting to an
acoustic disturbance at the inlet, then used a Nyquist criterion to predict stability for the closed
loop system. They concluded that the model showed a pronounced tendency toward instability
near 180 Hz, a frequency that was observed in experiments.

It is evident that acoustic velocities can in°uence the incoming mixture strength depending
on the nozzle geometry and operation. However, velocity °uctuations can also a®ect shear layers in
a °ow leading to a coupling with vortices shedding from the lip of a °ame holder. This type of hy-
drodynamic instability, sometimes called vortex rollup, has been observed in several con¯gurations
from swirled °ows, to blu®-body stabilized °ames in a duct, such as those used by Keller [10] and
Macquisten et al. [16]. The unstable °uid °ow convects downstream at a frequency dictated mostly
by the Strouhal number and modulates the heat release rate as it burns. In turn, the acoustic ¯eld
which results, feeds back on the °ow instability at the point of shear layer separation leading to a
more ampli¯ed °ow perturbation and self-excitation. These analyses require much closer attention
to °uid mechanical phenomena as the root cause for such instabilities. Reduced-order °uid insta-
bility models that have been developed have been mostly empirical as in Bloxsidge et al. [2], and
Gutmark and Paschereit [9] or extensions of simpler, laminar °ame models as in Dowling [7].

13
1.2.2.1 Dynamic Flame Models

Much of the work surrounding the study of thermoacoustic instability has been concentrated on the
dynamics of the °ame itself. Most of the literature on °ame dynamics deals with °at, laminar °ames.
These are not the types of °ames that would be observed in a lean premix gas turbine combustor,
but a brief summary is appropriate as some of these models have formed the basis for simpli¯ed
analyses of the more complex, turbulent °ames. Perhaps the most comprehensive treatment of
°ame stability analyses is by Markstein [17] where °ame dynamic models are constructed from
perturbation analyses starting with the basic conservation equations. That early work describes all
sorts of °ame stability phenomena and was, no doubt, the springboard for most modern analytical
studies of °ame stability behavior.

Among the models discussed by Markstein was a parametric °ame instability. In this case,
the acceleration of the column of °uid surrounding the wrinkled °ame, modeled as a second-order
oscillator, could force it to respond at a subharmonic frequency. This parametric instability was
described using the Mathieu equation (which is described in Nayfeh [23]) and was subsequently
studied by Searby and Rochwerger [33]. It is believed that such a subharmonic resonance was
observed in the unstable thermoacoustic signature measured in the Virginia Tech tube combustor
as discussed by Saunders et al. [31].

Much of the work on °ame stability has centered on the cellular structure of laminar °ames
as in Markstein [18] and McIntosh [19]. Essentially, it is practically impossible to realize a truly
°at °ame sheet because of nonuniformities in the mean °ow stream. The °ame sheet will always
bubble with cells, though sometimes very small in magnitude. When there is an oscillation in the
velocity ¯eld in which a °ame is anchored, the °ame surface will naturally oscillate as well. The
stretching and contorting of the °ame sheet changes its characteristic propagation speed which
is directly related to the rate of combustion and therefore the rate of heat release. Such °ame
surface area variation due to velocity perturbations provides an excellent mechanism for self-excited
thermoacoustic oscillation.

14
Flei¯l et. al. [8] formulated a kinematic model of the °ame which causes the surface area
of the °ame to respond to °uctuations in °ow velocity. Heat release rate is then modeled as
proportional to total °ame area while the °ame speed is assumed constant. Though there are some
considerable simpli¯cations made, this model shows that there is a low-pass dynamic nature to the
°ame. The normalized °ame area perturbation, the parameter that determines heat release rate,
goes down with increasing frequency due to decreasing cell size at high frequencies. This causes the
°ame to respond well to low frequency velocity perturbations, and not to respond higher frequencies.
The model also demonstrates that the unsteady heat release can have phase relative to the acoustic
velocity perturbation due to the kinematic area relationship. Though the model was designed to
describe a simple, laminar °ame, it has been applied to several models of turbulent °ames in lean
premix combustors as in Dowling [7] and Peracchio and Proscia [24].

1.2.3 Acoustic Modeling

Because of the important role that combustor acoustics play in thermoacoustic oscillations, stability
modeling requires a thorough treatment of the acoustic dynamics of the plant. The gain and phase
of the acoustic velocities, in relation to the unsteady heat release rate, are critical to the tendency
of any particular system toward or away from self-excitation as demonstrated in the discussion
on Rijke tubes. Starting with the conservation of mass, momentum, and energy, and neglecting
transport processes, the following equation can be derived.

@2P 0 2 2 0 @Q0
¡ c r P = (° ¡ 1) (1.2)
@t2 @t

Here, the unsteady heat release rate appears as a forcing term on the familiar acoustic wave
equation. Culick [6] notes that, according to this equation, generation of acoustic waves by heat
addition is due to P v work. He goes on to state that heat addition acts as a monopole source in
a fashion similar to mass addition. Note the similarity to the point mass source wave equation

15
(shown in Equation 1.3) from Pierce [26].

1 @ 2P 0
r2 P 0 ¡ = ¡m
Ä S (t)±(x ¡ xS ) (1.3)
c2 @t2

The heat release forcing can be treated just like any other type of acoustic source. Assuming
that Equation 1.2 is accompanied by a set of boundary conditions for the combustion system,
a Galerkin procedure can be used to determine the acoustic transfer function using the normal
modes of the system and any arbitrary forcing function distribution. However, as mentioned in
Section 1.1.1, a ¯nite element analysis can also provide direct frequency response information for
any modeled geometry and °uid temperature distribution.

1.2.3.1 Acoustic Finite Element Methods

Like any partial di®erential equation, the acoustic wave equation has exact solutions for only a
small number of very speci¯c boundary value problems. To ¯nd solutions to problems exhibiting
more complex geometry or distributions of acoustic speed, numerical approximations must be used.
A popular means for performing such analyses is with the use of ¯nite element methods. Craggs
[5] notes that there exist several means of discretizing the acoustic wave equation including ¯nite
di®erences or residual methods such as Galerkin's. However, he chooses to show formulations
of acoustic ¯nite elements based on variational principles much like the ones used in structural
vibration literature [4], [20].

Craggs discusses several types of elements and their uses and limitations. It is important
to consider the problem being modeled when choosing elements. If the geometry is such that one-
dimensional elements will su±ce, they are much more computationally e±cient. However, they are
unable to resolve acoustic pressure distributions that are not uniform over a cross section. Using
¯nite elements, systems with area discontinuities cannot be modeled using equivalent impedances
as with tranmission line elements [21]. Thus, multi-dimensional elements must be used in such

16
instances.

Finite elements are °exible enough to be used for an incredible variety of problems where
the use of transmission elements would quickly become cumbersome. Because every element in the
mesh is mathematically equivalent, the designer need not worry about what kind of element to use
at any given location in the geometry. Only the number and distribution of elements is important.
Munjal [21] notes that a general guideline for the maximum typical dimension of an element is
that it should not exceed twenty percent of the minimum acoustic wavelength. Typically, the ¯nite
element method involves iteration until the solution doesn't change signi¯cantly with further mesh
re¯nement. Since the acoustic wave equation is reduced to a set of ordinary di®erential equations
with the ¯nite element method, frequency response can be computed directly for incorporation into
a stability analysis.

1.3 Organization of the Dissertation

The following chapters are used to describe the details of the linear, closed-loop stability analysis.
Each chapter is dedicated to a di®erent block diagram component in the system. The ¯rst is the
acoustic ¯nite element modeling. In that chapter appear details of the model development including
proper scaling of unsteady heat release forcing of the ¯nite element acoustic mesh. The relative
importance of system parameters such as °uid temperatures, boundary conditions, and geometry
to the acoustic frequency response are also presented.

Immediately following the chapter on acoustic ¯nite element model development is a separate
chapter on acoustic model veri¯cation. Frequency response measurements were made at room
temperature, in the NETL rig using a speaker and microphones for the purpose of comparison with
¯nite element frequency response predictions. Geometry and boundary conditions were adjusted
and the results compared between model and experiment, demonstrating the ¯delity of the ¯nite
element technique for predicting the dynamics of the rig.

17
The fourth chapter discusses the unsteady well-stirred reactor °ame model. It includes details
of the model development including conservation and kinetic reaction relations, linearization to state
space, and proper scaling of acoustic velocity perturbations to input stoichiometry unsteadiness.
The e®ects of various simplifying assumptions are discussed in terms of their e®ect on the predicted
frequency response of the unsteady heat release rate to particle velocity perturbations in the mixing
nozzle.

The last major element in the loop gain is the time delay. Again, it is this parameter which
most strongly a®ects the phase response of the open-loop system, and therefore the stability of the
closed-loop, self-excited system. The ¯fth chapter opens with a brief discussion of linear stability
theory and how the frequency response relates to a common form of data presentation, the ¿ f plot.
Then, the multiple delay case is discussed and a design procedure is presented for reducing the loop
gain over selected frequency bands using multi-port fuel injector arrays.

The ¯nal chapter presents the assembly of all the components into an open-loop transfer
function which is used for stability prediction. By looking at the frequency response of the open-
loop transfer function, a limit cycle frequency can be predicted based on the frequency of 180 degree
phase crossings which indicate instability in the closed-loop plant. The results of the analysis
are compared with experimental data generated at NETL, and show the model to be capable
of predicting proper trends and similar instability frequencies. The ¯nal chapter concludes with
a discussion of areas of the modeling procedure which require further re¯nement including the
complexity of the °ame model itself, determination of exact values of time delay values based on
°uid mechanical modeling or carefully designed experiments, and °ow ¯eld e®ects in the coupling
between the unsteady acoustic velocity in the nozzle and unsteady equivalence ration entering the
reaction zone.

18
Chapter 2

Acoustic Finite Element Model

Acoustic modeling is vitally important to the prediction of thermoacoustic instability in a combus-


tion system. It is the gain and phase of the acoustic velocity perturbations that cause the °ame to
couple and feed back, forcing the acoustics as shown in Equation 1.2. Because the phase response
changes most drastically at the acoustic resonances, and the amplitude response is highest at those
same resonances, the resonant acoustic characteristics of the combustor have a profound in°uence
on the closed loop stability behavior.

The ¯nite element method (FEM) was chosen as the best acoustic modeling format for
this research for a variety of reasons. Because the combustor, when operating, would contain
°uid at a range of temperatures, the modeling would have to accommodate non-uniform property
distributions in the acoustic domain. Boundary element methods, for example, can only be used
when the °uid inside the volume is homogeneous because the analysis is based on solving the series
of free ¯eld Green's functions within the volume de¯ned by the boundary elements. If the acoustic
properties in the volume aren't constant, the solution to the free ¯eld Green's functions are not
known.

Transmission elements are another common method of acoustic modeling. The acoustic

19
response is based on the interaction of a series of traveling waves moving back and forth within the
acoustic domain. Here, a distribution of acoustic properties can be accommodated by discretizing
the domain su±ciently. However, this method is based on a one-dimensional acoustic ¯eld and can
be limited by its ability to capture acoustic characteristics in the so-called near ¯eld where large
area constrictions exist. They are commonly used for duct acoustic analysis, but for a complex
geometry, ¯nite element methods show a clear advantage.

The ¯nite element method is a discrete analysis, just as the others mentioned. The elements
themselves represent the order of the system. Pre-processing allows a user to construct a relatively
complex geometry in the form of small acoustic elements in a relatively short period of time.
Boundary conditions and forcing functions can be applied to the mesh and the acoustic frequency
response computed directly using the coordinates of the element pressure and velocities.

The FEM software package that was chosen for the analyses presented here is SYSNOISE
Version 5.4 for Windows NT and is written by LMS International. Mesh generation was performed
on SDRC's IDEAS Version 7.00.000 also on a Windows NT platform. The results are based on
frequency analysis with the complex pressure and particle velocity response to a known forcing
function. These frequency responses are generated by SYSNOISE, then exported to MATLAB
for purposes of display and eventually, closed loop stability analysis. The following sections will
describe the modeling process in detail as well as the veri¯cation with experimental measurements.

2.1 Modeling Details

The NETL combustion rig that is being modeled for this research consists of a long, cylindrical
shell. Air is introduced through a plenum before proceeding through a swirler where it is mixed with
gaseous fuel and then expanded and combusted in a chamber. The combustion products proceed
through an acoustic tuning plug before being quenched by a water spray and proceeding out the
exhaust end of the rig. The rig operates at 7.5 atmospheres and mean °ows which accelerate air

20
through the mixing nozzle at speeds of up to 60 m=s. Given this information, when deciding how
to model the acoustics of this rig, several considerations were made:

² What is the frequency band of interest for the stability analysis?

² Is a three dimensional model necessary, or can a two dimensional model su±ce?

² How will the acoustics be forced, and how will that forcing be represented?

² What boundary conditions should be used, particularly in the exhaust piping?

² What are the °uid properties in the acoustic domain, especially for hot operating conditions?

Many of these concerns were easily addressed. For example, the NETL stability data showed
limit cycle frequencies ranging up to about 250 Hz. It was assumed that any acoustic behavior
above that frequency would not be relevant to the stability characteristics of the combustor. Thus,
the rig acoustics were modeled up to 400 Hz. The dimensionality of the model is related to the
bandwidth. According to duct acoustics, as discussed in Kinsler and Frey [11], for example, the
lowest cuto® frequency for a circular duct is the (0,1) mode. This mode will cut on for wavenumbers
above 2:40=a where a is the radius of the duct. For the NETL rig, this corresponds to wavenumbers
greater than 17:58, which at an acoustic speed of 343 m=s means frequencies above 960 Hz. In other
words, no transverse modes will be present in the system below 960 Hz. This number will be even
higher for the case of the hot combustor as the acoustic propagation speed will correspondingly
increase.

This means that there will be no signi¯cant transverse acoustics in the frequency band of
interest. It is therefore not necessary to model the system three dimensionally as there will be no
pressure variation along the cross section of the rig in the low bandwidth selected. There will be a
cross sectional distribution in the local regions of area changes, but these will be evanescent as none
of the cross sectional modes are traveling in this bandwidth. From an acoustic modeling standpoint,
a two dimensional model should be su±cient, with the provision that cross section changes should

21
include enough elements to resolve the near ¯eld e®ects. SYSNOISE includes an axisymmetric
option which proved ideal for use with the circular cross sectional geometry of the NETL rig.

A diagram of the NETL combustion rig is shown in Figure 2.1. The ¯gure shows the location
of the mixing nozzle between the plenum and combustion chamber, but the nozzle itself is not
shown. Downstream of the quench spray the rig continues until ¯nally connecting to exhaust piping
which leads out of the room. The combustion chamber can be seen just upstream of a refractory
plug that is installed for acoustic tuning. The rig was divided down the centerline and modeled
axisymmetrically with the mesh shown in Figures 2.2 and 2.3. Note the perforated plates located in
the plenum of the rig. These features were not shown in Figure 2.1, but are used to dissipate large
scale turbulence in the head of the plenum before entering the nozzle. Because of the area restriction
of the swirl nozzle, geometric details inside the plenum of the combustor had little consequence with
regard to the system acoustic frequency response. Capturing the volume of the plenum was the only
really important detail. Also indicated in Figure 2.3 is the location of the °ame forcing function
used in the acoustic frequency response analysis.

2.1.1 Source Strength and Forcing

The FEM software o®ers several choices of forcing functions and boundary conditions to excite the
acoustic ¯eld de¯ned by the mesh. For this research, the panel vibration option was chosen. This
option is designed to allow for an excitation of some portion of the boundary to excite the acoustic
¯eld. User-supplied data can be used as well. So, a vibration could be measured experimentally
and then supplied to SYSNOISE as an excitation source. In this case, panel vibration was chosen
to allow for proper scaling of the source strength, as will be discussed next.

Looking over the FEM mesh shown in Figure 2.2, the axis of symmetry is along the top edge
of the mesh. This is the centerline of the combustion rig. However, what is not obvious is that the
top edge of the combustion chamber and exhaust sections is o®set from that of the plenum in this
mesh. In fact, the top edge of the combustion chamber section is a half millimeter o®set from the

22
Figure 2.1: Isometric diagram of NETL combustion rig.

Figure 2.2: Axisymmetric FEM mesh for NETL combustion rig.

Figure 2.3: Plenum and nozzle detail of NETL rig FEM mesh.

23
true centerline of the rig. It is as if there is a one millimeter rod located at the centerline of the rig.
This feature allowed for panel excitation near the centerline of the rig, presumably where a °ame
would reside. By forcing the acoustics in this manner, the source strength could be computed and
properly scaled for the unsteady heat release forcing.

Source strength is de¯ned as the rate of volume displacement of the source [26]. This equates
to the velocity of the surface of the source object multiplied by its surface area. In this case, the
vibrating object is a very small pulsating cylinder located as shown in Figure 2.3. The source
strength is therefore computed as 2¼rL u0panel . Here, the pulsating cylinder has a radius of 0:5mm
and a total length of 0:06m. For a panel vibration in SYSNOISE, the user speci¯es a (complex)
velocity value along an element face which is on the edge of the mesh. The sign convention is based
on the unit normal of the face pointing away from the element, in this case, toward the centerline
of the combustor.

With the forcing function established as described, the analysis in SYSNOISE consisted of a
frequency by frequency computation of the complex acoustic potentials at each node in the mesh.
The phase of these potential functions (pressure and acoustic particle velocity) is thus referenced
to the phase of the forcing function. By de¯ning the panel vibration as having a value of -1, the
phase response is relative to the expansion of the source into the acoustic ¯eld.

2.1.1.1 Scaling of Pulsating Cylinder to Unsteady Heat Release Rate

Because the actual combustion system acoustics are forced by the unsteady heat release rate, the
source strength of the pulsating cylinder must be properly scaled for the ¯nal closed-loop analysis.
The following derivation will generate a block which represents the gain from the unsteady heat
release rate, Q0 to panel vibration surface velocity, u0panel . The derivation begins with a slightly
di®erent form of Equation 1.2.

24
(° ¡ 1) dQ0
(r2 + k 2)P = ¡ ±(x ¡ xS ) (2.1)
c2 dt

From Pierce [26], the wave equation forced by a point mass source, an idealization of a small
source with a time varying volume, is as follows:

dQ^s
(r2 + k 2 )P = ¡½ ±(x ¡ xS ) (2.2)
dt

By comparing Equation 2.2 with Equation 2.1 above, the following equality is obtained.

(° ¡ 1) dQ0 dQ^s
¡ = ¡½ (2.3)
c2 dt dt
^
where Qs = 2¼rL u0panel

Finally, the expression relating heat release rate to panel velocity.

u0panel (° ¡ 1) i! (° ¡ 1)
0
= 2
= (2.4)
Q i! ½ 2¼rL c ½ 2¼rL c2

2.1.2 Fluid Properties From CFD

The purpose of the acoustic ¯nite element model was to replicate the acoustic behavior of the
combustor at operating temperatures. This means that the °uid properties in the acoustic domain
must be commensurate with those observed in the hot combustion rig. Very little data was taken
at NETL regarding temperatures inside the combustor, making it di±cult to determine what °uid

25
Figure 2.4: Fluid property distribution in combustor region.

properties should be used. However, NETL generated several steady-state CFD runs of the com-
bustion rig using Fluent. The results of these models are summarized in Appendix B, and include
temperature information in the combustion chamber.

The °uid temperature in the combustion rig has a nonuniform distribution which will a®ect
the frequency response results. It is known that the preheated air comes into the plenum at ap-
proximately 590K, and a thermocouple downstream of the spray quench indicates exhaust gases
at approximately 500K. The exhaust thermocouple is not corrected for radiation, but its reading
is assumed to be reasonably accurate. The only regions where temperatures are not known at all
is in the combustion chamber itself. The CFD results were invaluable for determining reasonable
temperature values to apply to the °uid in the combustion region. Based on the distributed tem-
peratures and on the mean combustor pressure, acoustic speeds and °uid densities could be de¯ned
for the entire acoustic domain.

An example of the distribution of °uid properties in the combustion chamber can be seen
in Figure 2.4. Here, the incoming °uid (region 1) is much cooler than the surroundings. Similarly,
the outer wall region of the combustion chamber (region 2) is cooler than the °uid in the area
exiting the combustor, region 4. Upstream of the combustor, the nozzle is assumed to be at the
plenum temperature, and downstream of the spray quench is assumed to be uniform at the exhaust
temperature. The values chosen for each of these regions are summarized in Table 2.1 organized by
mean equivalence ratio.

26
Table 2.1: Combustion chamber temperature distribution based on CFD estimates

Á T1 (K) T2(K) T3(K) T4 (K)


0.59 959 1419 1878 1823
0.63 982 1465 1948 1882
0.67 1004 1507 2011 1941
0.71 1026 1551 2077 2000
0.77 1056 1613 2169 2088

2.2 Model Sensitivity

The purpose of the sensitivity study is to fortify the assumptions made during the modeling process.
By associating the modeling assumptions with their relative importance to the frequency response
behavior of the acoustic plant, con¯dence is inspired in the accuracy of the ¯nal results. The sensi-
tivity also provides a means for determining which variables may require more detailed investigation,
and which ones may simply be ignored as insigni¯cant. Accurate acoustic frequency response be-
havior is the desired product for this component of the modeling e®ort. As such, sensitivity will be
analyzed as it pertains to resonant frequencies observed in the FRF's, as well as gain e®ects, if any.

The sensitivity analysis focuses on the parameters that are least well known in the combustion
rig. These include the distribution and values of °uid properties within the combustion rig, acoustic
boundary conditions, particularly at the tailpipe, and the refractory plug location and size.

2.2.1 Fluid Properties

As discussed in Section 2.1.2, the temperature distribution in the NETL rig, particularly in the
combustion section, is not well known. The characteristic impedance of a °uid is determined by its
acoustic speed. The temperature of a °uid determines its acoustic propagation speed. Therefore, the
temperature in the combustion rig is quite important to determining its acoustic behavior. Figure

27
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

200
hot
100 cold
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.5: FRF comparison for hot and cold °uid temperatures in NETL combustion
rig.

2.5 shows the frequency response of acoustic velocity inside the mixing nozzle of the NETL rig using
a uniform distribution of room temperature air compared to the frequency response obtained using
a temperature distribution similar to that seen during operating conditions. Clearly, the resonant
behavior of the rig is signi¯cantly changed from cold to hot conditions, but how accurately must the
temperatures be known to have con¯dence in the frequency response results generated with ¯nite
element analysis?

First, the sensitivity to the °uid in the combustion chamber was studied. During operation,
the combustion process can raise the temperature in the combustor to levels in excess of 2000 K.
The temperature in the combustor is most closely related to the mean equivalence ratio. Table 2.1
summarizes the temperatures in the combustion chamber by region (see Figure 2.4) and equivalence
ratio.

The model was run with each of these ¯ve sets of temperature data applied in the combustion
region. The inlet plenum was maintained at 580 K and the exhaust region (downstream of the
refractory plug) at 500 K. A pressure release boundary condition was used on the tailpipe end.

28
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

φ = .59
0 φ = .63
φ = .67
-100 φ = .71
Phase (deg)

φ = .77
-200

-300

-400
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.6: FRF comparison for equivalence ratio e®ects on acoustic response.

These tests were designed to demonstrate the e®ects of temperature distribution in the combustion
region on the acoustic frequency response of the rig. Figure 2.6 shows that even with a total range
of over 200 K, the temperature distribution in the combustion chamber and plug have a minimal
e®ect on the resonant behavior of the rig. As before, these frequency response functions are acoustic
particle velocity responses taken inside the mixing nozzle where fuel is added to the swirling air.

One reason that the combustion chamber temperature has so little e®ect is that it is physically
compact. When compared to the wavelength for the low frequency band of interest, the combustion
chamber length is very short indeed. It is therefore anticipated that changes in such a short region
would be relatively unimportant to the acoustic resonant behavior of the rig. Conversely, the
exhaust section, which extends from the spray quench at the exit of the plug to the exhaust piping,
accounts for most of the length of the rig. How will uncertainties in this downstream section a®ect
the accuracy of the acoustic predictions? Figure 2.7 clearly shows that even a 50 K uncertainty in
the downstream temperature will have much more profound e®ects on the acoustic resonances of
the rig.

The downstream temperature was measured during the stability experiments using a single,

29
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

200 500 K
550 K
600 K
100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.7: FRF comparison for exhaust gas temperature e®ects on acoustic response.

uncompensated thermocouple. For future studies, it would be wise to instrument the longer sections
of the test rig for °uid temperature measurements as temperature is most critical in these regions.
Since the addition of the long, afterburner section of the test rig, NETL has added several thermo-
couples downstream of the combustion chamber to give a better estimate of the °uid temperatures
in the longer sections of the rig.

The only other region of the NETL rig is the plenum. The inlet air temperature, however,
is known quite well and doesn't vary from test to test. At any rate, any variation of plenum
temperature would be relatively insigni¯cant as that section of the combustion rig is somewhat
acoustically isolated by the constriction of the mixing nozzle. Figure 2.8 clearly shows that even a
100 K variation in inlet air temperature hardly a®ects the acoustic response in the nozzle at all.

30
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

200
500 K
100 600 K
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.8: FRF comparison for inlet air temperature e®ects on acoustic response.

2.2.2 Boundary Conditions

The NETL rig features a large diameter chamber extending downstream of the water spray, even-
tually constricting down to a 4" exhaust pipe that leads out of the test chamber. The exhaust line
is not explicitly modeled with the FEM code, so a boundary condition must be applied to that end
of the model to account for the exhaust piping. The uncertainty lies in how to treat the exhaust
line boundary condition to most closely simulate the actual scenario. The exhaust piping is very
long. It is not known exactly how long the piping is, but an estimate of 8 meters was used to verify
modeling techniques.

The sensitivity analysis here is aimed at making a reasonable determination of what the
boundary condition should be. For purposes of comparison, four di®erent tests were performed.
First, the termination was treated as a rigid boundary, as if there were no exhaust piping at all.
This type of boundary condition results in pressure antinodes at the tailpipe termination. The
opposite e®ect can be obtained by imposing a zero pressure or pressure release boundary condition.
An 8 meter piping impedance was also tested. Finally, the simple characteristic impedance (½c)

31
0
10
Ampl

-2
10

-4
10
0 50 100 150 200 250 300 350 400

200
rigid
100 P=0
Phase (deg)

8 m tailpipe
0 ρc

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.9: FRF comparison for various boundary conditions on exhaust line.

boundary, which would equate to attaching an in¯nitely long exhaust pipe to the rig. Frequency
response results for these four boundary conditions are shown in Figure 2.9.

Logic dictates that the rigid termination be eliminated immediately, as it is most unlike the
other three and least true to the physics of the situation. The pressure release condition, too, is
unlike the physics. These two cases were included to bound the results, as they represent the two
extremes. It is interesting to note that at higher frequencies, the characteristic impedance and pipe
impedance models follow the pressure release behavior, while at low frequencies they tend toward
the rigid boundary result. It is also noteworthy that the 8 meter pipe and in¯nite line results lie
almost on top of one another. Thus, they are capturing more or less the same physics. The resonant
behavior of the 8 meter pipe can be seen in the response with the addition of many small peaks,

32
but overall its main peaks are in the same place as the in¯nite line results.

As can be seen by the lower peaks and less abrupt phase transitions of the in¯nite line case,
the characteristic impedance boundary condition added damping to the system. Supposing that
there were damping mechanisms along the length of the exhaust piping of the actual rig, it would
behave like a semi-in¯nite line. Acoustic waves would dissipate as they traveled down the pipe
and back from the end re°ection so that by the time they returned to the expansion in the rig
itself, they would be greatly attenuated. If the termination on the end of the exhaust piping were
not rigid (as is certainly the case) even more energy would be dissipated by transmission at that
boundary. Based on these justi¯cations and in the absence of any experimental veri¯cation, the
boundary condition of choice was the characteristic impedance or semi-in¯nite exhaust line.

2.2.3 Geometry Variation

The rig itself, particularly the refractory plug, was built to speci¯cation, but design tolerances
are ubiquitous. The tolerances of the plug were investigated as possible sources of uncertainty for
frequency response characterization. The plug was manufactured by pouring a refractory aggregate
into a mold. The ¯nished plug was inserted into the rig by hand. Thus, the tolerances on its
location and size were the most relaxed in the entire rig. The actual tolerances, though unknown,
were assumed to be § 1 cm on the length and location of the plug. These tolerances would result
in the possibility of not only a shorter or longer plug, but a shorter or longer combustion chamber
as well. Figure 2.10 clearly shows that, for the band of frequencies chosen, these tolerances had
negligible e®ect on the acoustic resonances of the rig. The four FRF's are labeled according to
the length of the combustion chamber and plug respectively, and bound the range of tolerances
accordingly.

33
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

short/short
100 short/long
long/short
long/long
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.10: FRF comparison for tolerances on refractory plug size and placement.

2.2.4 Distributed Forcing

Another topic of study for this work was dealing with the potential for a distributed forcing function
in the combustor. In other words, since fuel mass fraction enters the combustion chamber with
a spatio-temporal variation due to oscillatory velocity ¯eld, it could potentially cause a spatio-
temporal variation in the reaction zone itself. Should this occur, the forcing due to unsteady heat
release rate would contain a phase variation across the length of the °ame.

Figure 2.11 shows a comparison of the acoustic response with and without a phase distribution
in the forcing function. For the case shown, the force distribution has a 180 degree phase di®erence
across the 6 cm length of the pulsating cylinder source at 400 Hz. This would appear to be a
relatively extreme case as the °ame would be oscillating as a dipole at the higher frequencies
shown. Still, the e®ect of this distribution was not profound with respect to the amplitude response
and the phase delay could easily be modeled with the addition of a transport lag such as the one
that already exists in this system.

34
0
10

Ampl
-2
10

-4
10
0 50 100 150 200 250 300 350 400

-100 in phase
distributed

-200
Phase (deg)

-300

-400

-500
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 2.11: FRF comparison for a single phase and distributed phase forcing function.
NOTE: the distributed force has a 180 degree phase distribution at 400 Hz

The main problem with distributed forcing analysis is that there is so much unknown about
the shape of the unsteady °ame itself. Looking over the axial velocity distributions shown in
Appendix B, it is clear that the °ow stagnates and turns back into a recirculation zone near the
centerline of the combustion rig. Here, the temperature is very high, indicating the primary reaction
zone is located in the vicinity. For a swirl-stabilized °ame, the °ow ¯eld is very complex. It is not
obvious looking at a steady-state °ow ¯eld what the exact path reactants may follow to reach the
reaction zone may be. Along the path from the combustion chamber inlet to the actual reaction
zone, the °ow reverses, meaning the wavelength of the fuel distribution will become very small due
to the low °ow velocities. It is not clear how the distribution of fuel will look by the time it gets
to the reaction zone itself. It may even be dissipated by the turbulence or distributed to various
regions forming a large, monopole type liberation of heat release.

There are many questions that arise out of consideration of the swirling °ow ¯eld. Has the
assumption of a \frozen mixture" become faulty due to turbulence in the combustion chamber?
Where is the reaction zone located, and what path do the reactants take to arrive there? These
are questions that are perhaps best answered using CFD. An unsteady analysis would be able to

35
indicate whether there is a strong distribution of phase in the °ame zone due to unsteadiness in the
reactant mixture ratio as well as predict the degree of unsteadiness in the reactants upon reaching
the °ame zone.

It certainly appears, based on the results shown in Figure 2.11, that a distributed force could
certainly support an instability, only at a di®erent frequency than would be observed in the absence
of the distribution. It is clear that the distribution hasn't signi¯cantly changed the amplitude of
the frequency response. In other words, the forcing doesn't cancel itself out. Thus, a non-compact
reaction zone would not tend to be stable based on the force distribution alone, rather on the degree
of unsteadiness possible in the reactants when they arrive at the °ame. The longer the reactants
are in the turbulent °ow ¯eld before reacting, the less unsteady the mixture will become, and the
smaller the amplitude of unsteady heat release forcing.

In summary, the temperature of the °uid in the acoustic domain is most important for the
long exhaust section. The tailpipe boundary condition is modeled as a semi-in¯nite line impedance,
and the acoustic forcing is generated by a small, pulsating cylinder whose source strength is known
by the dimensions and amplitude of the surface velocity. The following chapter outlines acoustic
modeling and testing of the NETL rig at room temperature for purposes of acoustic FEM model
veri¯cation.

36
Chapter 3

Acoustic Model Validation

Because the acoustic characterization of the NETL rig was being performed exclusively by com-
puter modeling, it seemed necessary to provide some means of assuring that the FEM models were
accurate. If models could be veri¯ed with experimental data from a room temperature test, that
would lend support to the validity of hot temperature models. By taking acoustic frequency re-
sponse measurements in the cold rig using a speaker and microphones, a direct comparison could
be drawn to acoustic models of the same geometry.

What was being sought with this investigation was the veri¯cation that ¯nite element models
will capture the correct acoustic behavior for the geometry modeled. In other words, will changing
geometric features and boundary conditions in the ¯nite element model show the same e®ects in
frequency response characteristics as the data taken for the corresponding changes? Capturing the
exact frequency response behavior for any single test at any single location was not the purpose
of this investigation. As will be shown, uncertainties exist that make such a goal unreachable.
However, it was important to demonstrate the accuracy of the ¯nite element models for correctly
predicting changes in resonance frequencies and overall frequency response with changes in geometry
and boundary conditions. The reliability of the ¯nite elements for predicting the acoustic behavior
of the hot temperature rig will then be extrapolated from the success achieved in this task.

37
3.1 Acoustic Experiments

The stability data collected at NETL and presented by Straub and Richards [34] in 1998 came from
testing performed on a slightly di®erent rig than the one used for these acoustic measurments. A
photograph in Figure 3.1 shows the present day NETL rig with key features labeled. The original
stability data were collected on the rig before the afterburner section was added (see Figure 2.1).
The afterburner signi¯cantly increased the length, and therefore changed the acoustic character of
the test rig from its previous con¯guration. The acoustic ¯nite element frequency response of the
two con¯gurations modeled at room temperature can be seen superposed in Figure 3.2 for purposes
of comparison. Note that the long rig exhibits many more resonances in the 0-400 Hz bandwidth.
The new afterburner section included a constriction near the tailpipe made of refractory material
that channeled the °ow down to a water quench. All of the acoustic testing discussed in this
chapter was performed on this longer rig, hereafter called the long rig. The geometry used during
the stability testing of 1998 and for stability analysis work will hereafter be called the short rig to
avoid confusion.

The acoustic measurements were made with a three inch Radio Shack speaker and six mi-
crophones placed in sealed penetrations along the length of the rig. The speaker was suspended in
the combustion chamber just downstream of the mixing nozzle. This location was chosen simply
because the °ame, which is the thermoacoustic source, resides in approximately the same location
during operation. Figure 3.3 shows the speaker located just inside the combustion chamber. The
view seen in this photograph is from the plenum side with the plenum head and mixing nozzle re-
moved. The nozzle assembly is mounted °ush with the wall shown using three stud bolts which can
be seen in the photograph. After installing the speaker, the nozzle and plenum head were replaced
for testing.

Speci¯cations for the equipment used to generate the frequency response data for these
experiments are summarized in the following list.

38
Figure 3.1: NETL combustion rig.

² Frequency response data were generated using a four channel HP digital signal analyzer with
a frequency range of 0 to 400 Hz and 1 Hz bin spacing. The FRF's were computed with 50
ensemble averages using a Hanning window with 50% overlap.

² The speaker was excited using white noise from the source generator built into the HP.

² The excitation signal was ampli¯ed using a Radio Shack audio ampli¯er with a measured
nominal gain of 22dB.

² The acoustic measurements were made with Knowles model BL microphones with a reported
nominal sensitivity of -69dB re 1.0 volt/microbar at 1000 Hz, and a 3V preampli¯er.

The experiments were designed to illustrate the e®ects of geometry and boundary conditions
on the acoustic frequency response of the NETL rig. For this reason, three major geometries were

39
Frequency responses for long and short rig (room temperature)
20

(dB) 0
panel
/U
nozzle

-20
U

-40
0 50 100 150 200 250 300 350 400

200 short rig


long rig

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.2: FRF's of long and short NETL combustor rig at room temperature.

tested with two di®erent boundary conditions on the tailpipe. The exhaust piping shown in Figure
3.1 was not connected to the rig for the acoustic tests. The long rig terminated at a °anged pipe
which could be covered with a °ange cap or left open to the room, forming either a pressure antinode
at the end, or a pressure release boundary condition. Each of the following geometries were tested
with both boundary conditions.

² Nozzle and combustor plug installed.

² The combustor plug removed.

² The mixing nozzle and combustor plug removed.

3.2 Model Development

The long combustor was modeled in a similar manner to the short combustor described in the
previous chapter. The combustor was modeled with a two-dimensional axisymmetric mesh. Several

40
Figure 3.3: Plenum side of NETL rig showing speaker location for acoustic tests.

slightly di®ering models were made to study e®ects of geometric tolerances on the two refractory
components, the combustor plug and spray constriction in the afterburner. The plug was assumed
to have tolerances of §1 cm on its size and location in the rig. The tolerances on the spray restriction
were §1 cm for size and §3 cm for its location in the rig. These tolerance values were chosen based
on discussions with NETL personnel regarding the construction and installation of these features
in the rig. Because the tolerances on these features are so loose, the acoustic response can vary
within the tolerance range given. Sensitivity to these tolerances is discussed in Section 3.2.2.

As with the short rig, the centerline edge of the mesh didn't end on the exact centerline;
rather 0.5 mm o® center. The forcing function consisted of two vibrating panels on the slightly
o® center edge of the mesh, forming two small pulsating cylinders. The two cylinders were phased
180 degrees apart forming a dipole source, designed to emulate a speaker radiating freely into the
rig while hanging suspended in the center of the combustion chamber. The amplitude of the panel

41
vibrations were determined by scaling the source strength of the pulsating cylinder to match that
of the three inch speaker.

2¼rcyl Lcyl u0panel = ¼rspkr


2
u0spkr
2
rspkr (:0318m)2
u0panel = u0spkr = u0 = 50:4 u0spkr (3.1)
2rcyl Lcyl 2(:0005m)(:02m) spkr

Steady-state acoustic potentials were computed in physical coordinates at frequencies ranging


from 2 to 400 Hz in 2 Hz increments. Since the excitation was a vibrating panel, and all the potentials
were computed relative to the magnitude and phase of that input, the results represent frequency
response functions for the system. FRF's were computed at nodal locations corresponding to the
locations of the six microphones in the experiment. Meshes for the three main geometries are shown
in Figures 3.4 - 3.6, with microphone locations indicated in Figure 3.4.

3.2.1 FEM Mesh Re¯nement

Several meshes were generated with increasingly more detailed element con¯gurations. The fre-
quency response to a known input was measured in each case and the results compared for similar-
ity. The three most re¯ned meshes are shown in Figures 3.7 - 3.9. There was very little di®erence
in the frequency response characteristics of these models in the bandwidth from 0-400 Hz. The fact
that the frequencies are low allows the coarser mesh to match up well since the wavelengths are
long compared to the element size. Using the results obtained with this comparison, FEM analysis
models were developed with element sizes similar to those found in the meshes shown in Figures
3.7 and 3.9.

42
Figure 3.4: Mesh of long rig with microphone node locations indicated.

Figure 3.5: Mesh of long rig with combustor plug removed.

Figure 3.6: Mesh of long rig with combustor plug and mixing nozzle removed.

43
Figure 3.7: Mesh re¯nement test mesh #1.

Figure 3.8: Mesh re¯nement test mesh #2.

Figure 3.9: Mesh re¯nement test mesh #3.

44
3.2.1.1 Results

Shown in Figure 3.7 is a description of the forcing and response measurement locations for the mesh
re¯nement procedure. The same vibrating panel location and three response points indicated are
used on all three meshes for comparison purposes. In the ¯gures that follow, the FRF's from each
mesh are shown at each measurement location. Also shown are the ratios of the FRF's for successive
mesh re¯nements. These plots demonstrate how close the FRF's are compared to each other. The
ratio of FRF's from meshes 2 and 3 are very small indeed, both in terms of magnitude and phase
di®erences. It can thus be concluded that further re¯nement of the mesh will not signi¯cantly a®ect
the frequency response.

3.2.2 Sensitivity to Geometric Tolerances

This aspect of the investigation showed some interesting results. It was found that the tolerances
on the combustor plug and spray constriction had a signi¯cant e®ect on the frequency response
characteristics of the rig. Several combinations of geometric tolerance limits were simulated in this
study. For the sake of brevity, only two results are presented here. The two frequency responses
shown in Figures 3.16 - 3.19 show the most extreme cases based on the tolerances given for the plug
and spray restriction. The locations correspond to microphone placements as indicated in Figure 3.4.
These two responses bound all of the results of this geometry study. It is interesting to see that some
resonances are much more profoundly impacted by geometry changes than others. These results
demonstrate that, in the range of frequencies between 200 and 250 Hz, there is a large variability in
the observed resonances as a result of changing the length and position of the combustor plug and
spray constriction within the geometric tolerance limits. The frequency response is very sensitive
to these tolerances in that frequency band.

These tests also serve to bound the achievable accuracy of the models compared to experimen-
tal measurements. This is perhaps the most signi¯cant result of the sensitivity study. Uncertainty

45
40

20
Mag

-20

-40
0 50 100 150 200 250 300 350 400
freq (Hz)
300
mesh 2
200 mesh 3
Phase (deg)

mesh 1
100

-100
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.10: FRF's of mesh convergence meshes 1, 2, and 3 at measurement location


1.

4
Magnitude difference (dB)

-2

-4
0 50 100 150 200 250 300 350 400
freq (Hz)
20
Phase difference (deg)

10

-10
mesh 2 / mesh 1
-20 mesh 3 / mesh 2

-30
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.11: Ratios of mesh convergence FRF's at location 1.

46
60

40
Mag

20

-20
0 50 100 150 200 250 300 350 400
freq (Hz)
300

250 mesh 2
mesh 3
Phase (deg)

200 mesh 1

150

100

50
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.12: FRF's of mesh convergence meshes 1, 2, and 3 at measurement location


2.

4
Magnitude difference (dB)

-2

-4
0 50 100 150 200 250 300 350 400
freq (Hz)

40
Phase difference (deg)

20

mesh 2 / mesh 1
-20 mesh 3 / mesh 2

-40
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.13: Ratios of mesh convergence FRF's at location 2.

47
50

40
Mag

30

20

10
0 50 100 150 200 250 300 350 400
freq (Hz)
250

200 mesh 2
Phase (deg)

mesh 3
mesh 1
150

100

50
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.14: FRF's of mesh convergence meshes 1, 2, and 3 at measurement location


3.

0.6
Magnitude difference (dB)

0.4

0.2

-0.2

-0.4
0 50 100 150 200 250 300 350 400
freq (Hz)
2
Phase difference (deg)

-2
mesh 2 / mesh 1
mesh 3 / mesh 2
-4
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.15: Ratios of mesh convergence FRF's at location 3.

48
102

101
|A|

100

10-1
0 50 100 150 200 250 300 350 400

200 long combustor & plug, short afterburner & constric.


short combustor & plug, long afterburner & constric.
150

100
Phase (deg)

50

-50

-100
-150

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.16: FRF's of opposite extreme tolerance limits for the NETL long rig at
location 3.

101

100
|A|

10-1

0 50 100 150 200 250 300 350 400

long combustor & plug, short afterburner & constric.


200 short combustor & plug, long afterburner & constric.

150

100
Phase (deg)

50

-50

-100

-150

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.17: FRF's of opposite extreme tolerance limits for the NETL long rig at
location 4.

49
101

100
|A|

10-1

0 50 100 150 200 250 300 350 400

long combustor & plug, short afterburner & constric.


short combustor & plug, long afterburner & constric.
200

150

100
Phase (deg)

50

-50

-100

-150

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.18: FRF's of opposite extreme tolerance limits for the NETL long rig at
location 5.

101

100
|A|

10-1

10-2
0 50 100 150 200 250 300 350 400

long combustor & plug, short afterburner & constric.


short combustor & plug, long afterburner & constric.
200

150

100
Phase (deg)

50

-50

-100

-150

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.19: FRF's of opposite extreme tolerance limits for the NETL long rig at
location 6.

50
in the acoustic measurements is di±cult to quantify, especially in terms of frequency response and
observed frequencies. For practical purposes, the measurements were assumed to be °awless. How
then, could a model's accuracy be evaluated? The only means for comparing experimental and ¯nite
element results was by attempting to determine the amplitude of uncertainties, particularly those
based on geometry. Here, the geometric uncertainties in°uenced the values of certain frequencies
by § 7 Hz or more in the FEM analysis. Thus, it is not practical to expect a model to match
experimental measurements to within less than 7 Hz in those regions.

3.2.3 Speaker Model

The speaker needed to be modeled in order to account for its dynamics in the comparison of
measured and modeled results. Typically, a speaker is modeled as a single degree of freedom
oscillator forced by a ¯rst order electrical circuit. The system is coupled electromechanically as the
motion of the cone induces a voltage in the voice coil and current applied to the coil, in turn, drives
the motion of the cone. In order to determine the values for mass, sti®ness, and electromechanical
coupling coe±cients, a series of simple experiments were assembled. The following description will
not be detailed, rather an overview of the technique used.

First, the mass and sti®ness of the mechanical part of the system had to be determined. To
accomplish this, the speaker was forced with a sine wave and the frequency swept, all the time
monitoring the voltage across a small resistor that was connected in series with the voice coil. As
the forcing frequency approached the mechanical resonance frequency, the back EMF would increase
as the velocity of the cone approached a maximum. The voltage drop across the test resistor would
decrease due to the increased value of the back EMF which acts as a negative source voltage. In this
way, the frequency at which the measured voltage was minimum was assumed to be the resonance
frequency of the lightly damped mechanical system. Repeating the same experiment with a known
mass added to the speaker cone in the form of a small piece of clay resulted in a di®erent resonance
frequency. Based on the two resonance frequencies measured and the known mass added to the
speaker, two equations were formed for the two unknowns, the mass and sti®ness of the speaker.

51
Once the mass and sti®ness were solved, the electromechanical coupling could then be determined.

The electromechanical coupling consists of the magnetic °ux density, B multiplied by the
length of the voice coil wire. This product, multiplied by the current in the coil produces a force
that causes the motion of the voice coil which is attached to the speaker cone. It was not necessary,
however, to have a keen knowledge of either the magnetic ¯eld strength of the permanent magnet
or the length of the voice coil windings, as the product of these two quantities is what produces
the force. Looking at Equation 3.2, it is evident that for a static current, the speaker cone will
be displaced by a value determined by the product, Bl and the current in the coil, i. By careful
measurement of the cone displacement and coil current and by knowing the value of the mechanical
sti®ness, the product Bl, now referred to simply as the electromechanical coupling constant, could
be solved. After measuring the electrical resistance and inductance values of the voice coil with an
LCR meter, the only remaining coe±cient was the mechanical damping. This was determined by
matching the frequency response of the speaker model with a near ¯eld pressure measurement. The
resulting speaker transfer function from input voltage to speaker cone velocity is shown in Equation
3.3 below.

M xÄ + C x_ + Kx = Bli (3.2)

U Bls
= (3.3)
V MLs + (MR + CL)s + (KL + RC + Bl2 )s + KR
3 2

It was determined that the dynamics of the speaker should simply multiply with the acoustic
response of the ¯nite element model for comparison with measurements. Because the speaker is
not °ush mounted in the rig, seeing the acoustic pressure of the rig on one face and atmospheric
pressure on the other, there is no acoustic structure interaction present in this scenario. From the
standpoint of the dynamics of the speaker itself, hanging freely in the combustor rig is similar to
radiating into free space. Thus, there is no coupled problem to solve and the dynamics of the
speaker and acoustic ¯eld are simply multiplied together.

52
3.3 Comparison of Model to Experiment

For comparison of experiment and ¯nite element model, the frequency response functions considered
included only those taken at locations 3-6 shown in Figure 3.4. The plenum (location 1) was found to
be too isolated to get good coherence in the measurement. The measurement using the microphone
located in the combustion chamber was subject to poor coherence as well, probably due to its
proximity to the speaker itself. As a dipole source, the speaker's near ¯eld, when observed from the
side, would have a very small amplitude due to its directivity. The microphone in the plenum was
so oriented relative to the speaker. For these reasons, the measurements taken at locations 1 and 2
were disregarded.

Frequency responses were generated for comparison with experiments by multiplying the
FEM velocity response by the speaker model frequency response, the appropriate source scaling
V oltmic
factor, and the gains of the ampli¯er and microphones to achieve a true, V oltsource
frequency response
function. Interestingly, the speaker model dynamics always appeared to disagree with experimental
data. When compared, the speaker resonance appeared to occur much too high in the model
compared with the measurements. The origin of this disagreement is not known as the speaker was
modeled very meticulously. Nonetheless, the model results shown here depict the speaker model as
a simple 90 degree lag with the gain determined by the speaker model presented in Section 3.2.3.
The model behaves as if the entire bandwidth depicted is above the speaker resonance frequency. It
appears that this assumption is good above about 50 Hz. Data below that mark are, for the most
part, unreliable based on the coherence measurements so the disagreement is insigni¯cant. The
speaker model discrepancy was never resolved, nor was it vigorously pursued, mostly because the
speaker dynamics themselves were not signi¯cant to the purpose of this study, which was validation
of the ¯nite element models. As long as the speaker dynamic behavior could be accounted for in
some way, it was not worthwhile to determine the exact error in the modeling process.

The important aspects for comparison are the consistency of behavior between model and
experiment, not necessarily exact matching of each frequency response function. In the measure-

53
ments, frequencies below 30 Hz are not excited well by a small, three inch speaker. As a result,
the coherence was characteristically low in the experimental results for the low frequency range.
Due to modeling sensitivity issues discussed earlier, the range of frequencies from about 200 to 250
Hz tended to be uncertain in the modeled results. The frequency range above 250 Hz was not of
strong interest in light of the range of frequencies observed in the stability data and the temperature
scaling issues discussed previously. Thus, the range of frequencies that are most closely scrutinized
in this study are those between about 30 and 200 Hz. For most of the measurements and ¯nite
element results, there tend to be between four and six resonances observed in this frequency range.

3.3.1 Geometry E®ects

Looking over the results, the model did tend to exhibit similar behavior to the data taken for all
test conditions. Figures 3.20 - 3.22 show how the model results compare for the three geometric
variations of the NETL long rig; the full version with plug and nozzle in place, the long rig with the
combustor plug removed, and the long rig with combustor plug and mixing nozzle both removed.
Up to about 150 Hz, there was very little di®erence between the ¯rst two cases, both data and
model. The resonance that appeared around 160 Hz, moved signi¯cantly between these two cases.
Both the data and the model showed a 12 Hz shift in this resonance. Both the model and data
also showed an upward shift in all the resonances observed when the mixing nozzle was removed.
The low frequency behavior appears to show the worst agreement. This was most likely due to
the e®ect of several modeling assumptions applied to the plenum area, particularly the perforated
plate. However, as the plenum is isolated when the nozzle is in place, its dynamics are not strongly
emphasized, and detailed modeling of its acoustics is therefore not a key issue.

Comparing the result of the measurements and ¯nite element model at each location along
the combustion rig also showed that the model exhibited similar trends and reasonable agreement
overall. Figures 3.23 - 3.25 all show results for the full length rig with nozzle and plug in place.
Paying particular attention to the phase, the results showed very good agreement overall. Thus,
a global picture clearly demonstrates the ¯delity of the model compared with the data for the

54
geometric variations discussed.

3.3.2 Boundary Condition E®ects

The e®ect of the tailpipe boundary condition was also investigated. It was found that this only
in°uenced the behavior at frequencies below about 80 Hz. Based on the temperature scaling, this
could translate to frequencies of 120 Hz or more in the hot rig. Therefore, the e®ects could be
signi¯cant. Two tests were conducted for this comparison; the rig was tested with and without the
tailpipe section blocked o® by a °ange cover. In the ¯nite element model, these boundary conditions
were emulated using rigid wall and pressure release (P = 0) boundary conditions respectively. There
were two resonances between 30 and 80 Hz. In both the model and the test measurements, the rigid
wall boundary causes these two resonances to occur at lower frequencies than the pressure release
boundary condition. The frequencies match between the model and the experimental results to
within less than 5 Hz in all cases. Figures 3.26 - 3.27 show the two cases for the model and
experiments up to 100 Hz. Note that above 80 Hz, there is no di®erence between the two boundary
conditions.

It has been shown that the ¯nite element modeling technique could be validated with acoustic
measurements in the NETL combustion stability testbed, lending credence to the results presented
for the hot combustor case. Having thus completed the discussion of the acoustic modeling task
of this research, the model used for describing the gain and phase relationship of the rate of heat
release to acoustic velocity perturbations is presented in the following chapter.

55
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz) experiment
180 FEM
Phase (deg)

90

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.20: Experimental and theoretical FRF's of NETL long rig measured at loca-
tion 4.

56
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz)
experiment
180 FEM

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.21: Experimental and theoretical FRF's of NETL long rig with no plug mea-
sured at location 4.

57
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz)
experiment
FEM
180

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.22: Experimental and theoretical FRF's of NETL long rig with no plug and
no nozzle measured at location 4.

58
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz)
180
experiment
90 FEM
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1
Coherence

0.5

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.23: Experimental and theoretical FRF's of NETL long rig measured at loca-
tion 3.

59
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz)
experiment
180 FEM

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.24: Experimental and theoretical FRF's of NETL long rig measured at loca-
tion 5.

60
0
10
|A|

-5
10
0 50 100 150 200 250 300 350 400
freq (Hz)
experiment
180 FEM

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 3.25: Experimental and theoretical FRF's of NETL long rig measured at loca-
tion 6.

61
-1
10

-2
10
|A|

-3
10
test - closed BC
-4 test - open BC
10 FEM - closed BC
0 10 20 30 40 50 60 70 80 90 100
FEM - open BC
freq (Hz)
180

90
Phase (deg)

-90

-180
0 10 20 30 40 50 60 70 80 90 100
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80 90 100
freq (Hz)

Figure 3.26: Experimental and theoretical FRF's showing boundary condition e®ects
at location 3.

62
-1
10

-2
10
|A|

test - closed BC
-3
10 test - open BC
FEM - closed BC
FEM - open BC
-4
10
0 10 20 30 40 50 60 70 80 90 100
freq (Hz)
180

90
Phase (deg)

-90

-180
0 10 20 30 40 50 60 70 80 90 100
freq (Hz)
1

0.8
Coherence

0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80 90 100
freq (Hz)

Figure 3.27: Experimental and theoretical FRF's showing boundary condition e®ects
at location 5.

63
Chapter 4

Unsteady Well-Stirred Reactor

In order to determine how velocity perturbations in the acoustic ¯eld would a®ect the overall heat
release rate, a simple control volume analysis was performed. The unsteadiness is described by
conservation of species and energy in the control volume and the heat release rate computed as the
change in sensible enthalpy from the reactant °ow entering to products exiting the control volume.
For simplicity, the species consist of three components: fuel, oxidizer, which is air in this case, and
products. The volume is assumed to be steady, but the mass fraction of species and temperature of
the control volume are allowed to vary. Single-step, Ahrenius kinetics describe the rate of change
of reactant and product species in the control volume. The control volume is termed a well-stirred
reactor (WSR) since the species are assumed to be perfectly and instantaneously mixed. This type
of analysis is quite common for steady-state analysis of gas turbine combustion systems as shown
in Turns [36] and has been used for unsteady analysis by Lieuwen et al. [13] as well.

The conservation equations are linearized about an equilibrium determined by the mean
stoichiometry and °ow variables. The linearization is used to formulate a state model that relates
the rate of heat release from the control volume to acoustic velocity perturbations. Since the
overall analysis merely involves stability prediction, the linearization need only be valid for small
perturbations in the mean variables.

64
4.1 Model Description

The NETL combustor is operated at a mean pressure of 7.5 atmospheres. It was assumed that
acoustic perturbations in pressure would be too small to a®ect the temperature of the control
volume directly through the perfect gas relation and also too small to have any e®ect in a pressure
term of the energy equation. Thus, the control volume analysis considers the e®ects of acoustics
only in terms dealing with species mass °ux which is determined by acoustic velocities, not acoustic
pressures.

Unsteady acoustic velocity can couple into the unsteady heat release rate by at least two
means. First, the velocity perturbations cause unsteady mass °uxes over control surfaces. This
type of coupling alone could potentially cause thermoacoustic instability as proposed by Flei¯l
[8], Bloxidge [2] and others and is typically associated with a °ame surface area variation which
modulates the amount of heat release rate. For a steady state, steady °ow condition in a control
volume, heat release rate is expressed in terms of mass °ux and change in sensible enthalpy through
the control volume. By imposing a perturbation on the mass °ux term, the heat release rate will
become unsteady.

_ pout (Tout ¡ Tref ) ¡ Cpin (Tin ¡ Tref ))


Q = m(C (4.1)

On the other hand, supposing that the acoustic velocity could a®ect the mixture strength,
the change in sensible enthalpy through the control volume could be made unsteady by changes
in reaction rate caused by the unsteady mixture ratio. This is the case for the many time delay
models where the acoustic velocity a®ects the air/fuel ratio of the reactants. It can be shown that
small acoustic velocity perturbations can lead to signi¯cant variations in equivalence ratio for low
Mach °ows. The equivalence ratio, Á, is equal to the fuel/air ratio normalized by the stoichiometric
value. The stoichiometric air/fuel ratio of methane is 17.123, which appears explicitly in Equations
4.2 - 4.4. The equations are written in terms of mean and unsteady mass °uxes of oxidizer and fuel

65
inside a mixing nozzle.

¹ 0 ¹_ f + m
m _ 0f
(Á + Á ) = 17:123 (4.2)
m¹_ o + m
_ 0o

Dividing through by Á¹ and subtracting 1 from both sides,

Á0 ¹_ f + m_ 0 f
17:123 m
= ¡1 (4.3)
Á¹ Á¹ m ¹_ o + m_ 0 o

which can be rewritten as,

m_ 0 f
Á0 17:123 m ¹_ f m¹_ f
+1
¹ = ¹ ¡1 (4.4)
Á Ám¹_ o m_ 0 o
+1
m¹_ o

reducing ¯nally to the form shown below:

m_ 0 f _ 0o
Á0 m¹_ f
¡m¹_ o
m
= (4.5)
Á¹ m_ 0 o
+1
m¹_ o

For simplicity, it is assumed that the fuel °ow is choked, which means that m0f = 0. Because
oxidizer mass °ux can be expressed in terms of velocity as: m_ o = ½Au, variation in the equivalence
ratio is determined by the acoustic particle velocity at the fuel injection point. This equivalence
ratio variation is then convected down the mixer at the mean °ow velocity, u¹, to the °ame location
where it reacts. Thus, the equivalence ratio can be assumed to behave according to the following
relation where L is the convection distance:

66
u0 (t ¡ Lu¹ ) 0 L
¹ u (t ¡ u¹ )
Á0 (t) = ¡Á¹ = ¡Á (4.6)
u¹ + u0 (t ¡ Lu¹ ) u¹

Equation 4.6 shows that equivalence ratio variations occur at the same order as acoustic
velocity perturbations. Especially for lean mixtures, this variation can cause large °uctuations in
reaction rate [13].

Two mechanisms have been discussed for acoustic velocity coupling to heat release rate. Both
have appeared in the literature, yet seldom are both considered simultaneously. The model devel-
oped for this research was to include both of the mechanisms described here. However, because the
¯nal model resulted in a ¯xed control volume analysis, it was di±cult to determine the appropriate
unsteady velocities to include. Early attempts were made to account for °ame area variations,
but with the swirling °ow ¯eld, it proved exceedingly di±cult to determine a °ame location and
values for the reactant temperatures and speeds. Thus, the model consisted of a steady volume
with control surfaces determined by the inlet and exit planes of the NETL combustion chamber
rather than by a moving °ame surface.

4.1.1 Conservation Relations

The dynamic °ame model is based on the set of ¯rst order conservation equations for the mass of
each species and energy in the control volume. Though there are three species, one of the species
conservation equations says that the sum of species mass fractions is equal to one. Therefore,
the concentration of each of the three species is described by two mass conservation di®erential
equations and the sum of mass fractions relation. The mass conservation relations include terms
describing the production and consumption of species due to chemical kinetics. The relations used
in this research are based on those presented in Turns [36].

According to Turns, and with a few manipulations of the variables, the rate of change of

67
methane concentration due to the combustion reaction can be described by the following relation
kmol
which has units of m3 s
. This is the Arrhenius rate term for consumption of species due to com-
bustion which appears explicitly in the mass conservation relations shown in Equations 4.8 and
4.10.

kmol ¡15098
!_ CH4 = (¡24100 ) ½ YCH4 (t)¡:3 (0:233 Yair (t))1:3 exp T (t) (4.7)
kg s

Having thus described the rate of consumption of fuel in the reaction, the continuity relation
for fuel mass fraction is shown in Equation 4.8. This mass conservation equation includes mass
°uxes accross inlet and outlet control surfaces, mass consumption due to combustion, and mass
storage in the control volume. As fuel is consumed in a reaction, so is the oxidizing species, but at a
di®erent rate depending on the stoichiometry. The rate of consumption of oxidizer can be described
using Equation 4.7 as well. The gain on the consumption term for the oxidizer conservation equation
is dependent on the relative concentration of oxidizer to fuel in the control volume, and is derived
and shown in Equation 4.9 below, followed by the conservation of mass equation for the oxidizer.

dYf (t)
½ V = ½in Yfin (t)Ain uin (t) ¡ ½Yf (t)Aout uout (t) + MWf !_ f (t)V (4.8)
dt

½
= [F uelin ] + [Oxin ]
MWmix
M Wo [Oxin ] ( M W½mix ¡ [F uelin ])MWo [Oxin ]M Wo
A=F = = = ½
MWf [F uelin ] [F uelin ]M Wf ( M Wmix ¡ [Oxin ])M Wf
8 ½
> A=F M Wf
< [Oxin ] = MWmix
M Wo +A=F M Wf
) ½
M Wo
>
: [F uelin ] = M Wmix
M Wo +A=F M Wf
M Wf
T hus; [Oxin ] = A=F [F uelin ]
MWo
d[Oxin ] MWf d[F uelin ]
) = A=F
dt M Wo dt

68
MWf
¡or¡ !_ o = A=F !_ f (4.9)
MWo

dYo (t) Yoin


½ V = ½in Yoin Ain uin (t) ¡ ½Yo (t)Aout uout (t) + MWf !_ f (t)V (4.10)
dt Yfin (t)

The product species mass fraction is determined by the following expression. Thus, the
conservation of mass equations are complete. There are two di®erential equations: one for fuel
(Equation 4.8), and one for oxidizer (Equation 4.10).

Yp (t) = 1 ¡ Yf (t) ¡ Yo (t) (4.11)

Having described the conservation of species with Equations 4.8, 4.10, and 4.11, the only
remaining equation needed to describe the dynamics of the WSR is energy conservation, which is
shown in Equation 4.12 below. Again, there are °ux terms describing the amount of energy entering
and leaving the control volume, and there is a storage term to balance them.

3
d X
½V [ Yi (t)ei (t)] = ½in Ain uin (t)(Yoin hoin + Yfin (t)hfin )
dt i=1
3
X
¡ ½ Aout uout (t) Yi (t)hiout (t) (4.12)
i=1

The speci¯c enthalpy, h, is de¯ned for each species using a linearization about the mean
°ame temperature, and computed using polynomial ¯ts from Turns [36]. The speci¯c enthalpy of
the product species was determined based on the molar concentration of all the products in the
equilibrium chemistry solution. Using STANJAN, molar concentrations of CO, CO2, OH, H2 O,
O, O2 , N, N2, NO, and N O2 were obtained for the mean equivalence ratio and used along with
individual polynomial ¯ts of each species to ¯nd the speci¯c enthalpy of the product mixture.

69
Once the enthalpies, hf , ho , and hp were all computed, they were linearized about the adiabatic
°ame temperature which was assumed to be close to the equilibrium temperature of the well-stirred
reactor. Using the procedure described, the speci¯c enthalpies in Equation 4.12 were replaced with
the expression that appears in Equation 4.13 below.

hi (t) = h¹i + Cpi T (t) (4.13)

The conservation equations are therefore functions of the species mass fractions and reactor
temperature, all of which are time varying. Additionally, the time varying inlet and exit velocities,
as well as the time varying mixture fraction of incoming fuel, are parameters that appear in these
equations as well. Equations 4.8, 4.10, and 4.12 form a set of ¯rst order di®erential equations
in Yf (t); Yo (t), and T (t). These state equations can be linearized about an equilibrium solution,
allowing for incorporation into a linear stability model of the thermoacoustic plant. All that remains
is to solve for the unsteady heat release rate which forces the acoustics. As mentioned previously,
the heat release rate is computed as the change in sensible enthalpy of the °uid as it passes through
the reaction zone. For an adiabatic reactor, Q is computed as follows:

3
X
¹
Q+Q0
= ½ Aout uout (t) Yi (t)Cpi (T (t)¡Tref )¡½in Ain uin (t)(Yoin Cpo +Yfin (t)Cpf )(Tin ¡Tref ) (4.14)
i=1

Only the unsteady part of the heat release rate is needed as it is the forcing term to the
acoustic model (see Equation 1.2). The expression in Equation 4.14 will be linearized to form the
output equation in a linear state space model which describes the dynamics of the heat release
response to °ow velocity perturbations. In order to linearize the state equations properly, equi-
librium solutions to the nonlinear equations must be found. The next section will describe the
equilibrium solutions and mean conditions for each of the twenty test conditions modeled in the
NETL combustion rig.

70
4.1.2 Mean Conditions Using CFD Results

The tests conditions that were modeled for this research span a range of equivalence ratios and mean
°ow conditions. Speci¯cally, the equivalence ratios tested were 0:59, 0:63, 0:67, 0:71, and 0:77. The
values shown represent the overall equivalence ratio based on the total fuel and air entering the
combustion chamber. At each of these equivalence ratios, tests were performed at mean °ow speeds
(in the annular nozzle) of 30, 40, 50, and 60 ms . Thus, there are a total of twenty di®erent mean
conditions for which the °ame dynamic model can be derived.

At each of these operating conditions, there are several mean variables that are used in the
°ame model. These include the mean temperature, mean inlet and exit velocity, and mean inlet
mass fractions of air and fuel. These variables are determined using equilibrium chemistry estimates
of adiabatic °ame temperature and using mean °ow CFD data provided by engineers at NETL (see
Appendix B). Inlet and exit velocities and temperatures were computed for each of the ¯ve CFD
cases. These data were then used to extrapolate corresponding values for the range of operating
conditions represented in the NETL stability data.

It was observed in the CFD data that exit temperature was dependent mostly on equivalence
ratio. Therefore, the exit temperatures for the NETL data were extrapolated using Equation 4.15
which was regressed using the CFD results. This in turn became the target equilibrium temperature
for the WSR model. The inlet velocities were all assumed to be about 95% of the nominal value,
while the exit speeds were computed based on continuity using the mean temperature and area of
the exit plane.

T¹ = 953 + 1474Á (4.15)

For each test condition, the WSR volume was adjusted until the equilibrium temperature
equaled the value obtained using Equation 4.15. Thus, the mean exit temperature was made the

71
Table 4.1: Mean Operating Conditions for Flame Model

Á Y¹fin Vnom ( ms ) Vin ( ms ) Tad (K) T¹ (K) V ol ¤ 103 (m3 ) ¹


Q(kW )
0.59 0.0333 30 28.5 1878 1823 2.95 406
0.59 0.0333 40 38.0 1878 1823 3.95 541
0.59 0.0333 50 47.5 1878 1823 4.90 676
0.59 0.0333 60 57.0 1878 1823 5.90 812
0.63 0.0355 30 28.5 1948 1882 2.24 430
0.63 0.0355 40 38.0 1948 1882 3.01 574
0.63 0.0355 50 47.5 1948 1882 3.78 717
0.63 0.0355 60 57.0 1948 1882 4.54 861
0.67 0.0376 30 28.5 2011 1941 1.95 449
0.67 0.0376 40 38.0 2011 1941 2.60 599
0.67 0.0376 50 47.5 2011 1941 3.24 749
0.67 0.0376 60 57.0 2011 1941 3.89 898
0.71 0.0398 30 28.5 2077 2000 1.65 479
0.71 0.0398 40 38.0 2077 2000 2.24 640
0.71 0.0398 50 47.5 2077 2000 2.77 799
0.71 0.0398 60 57.0 2077 2000 3.30 958
0.77 0.0430 30 28.5 2169 2088 1.42 517
0.77 0.0430 40 38.0 2169 2088 1.89 689
0.77 0.0430 50 47.5 2169 2088 2.36 861
0.77 0.0430 60 57.0 2169 2088 2.83 1033

same for each of the four velocities at every equivalence ratio modeled. Smaller velocities required
smaller volumes to maintain the equilibrium temperature. As the volume is increased or the inlet
velocity decreased, the mean temperature would approach the adiabatic °ame temperature since
the product concentrations are based on that equilibrium solution. Smaller volumes simply provide
less residence time for the species resulting in lower equilibrium temperatures. The largest volume
required to obtain the desired equilibrium temperature was 0:0059m3 which happens to be exactly
equal to the total volume of the NETL combustion chamber. Equilibrium data including mean heat
release rate are provided for each of the twenty test conditions in Table 4.1.

In the original formulation, the °ame surface was intended to contain the control volume
for this analysis. However, after developmental di±culties with determining the exact location of
the °ame surface within the combustion chamber as well as the inability to know inlet properties

72
reliably, the control surfaces were chosen to be the inlet and exit to the combustion chamber
itself. As mentioned previously, the CFD results facilitated accurate estimates of the inlet and exit
conditions. Were it not possible to estimate these conditions accurately, the energy balance within
the control volume, and therefore the estimate of the total heat release rate would be in error.

Using the conservation equations shown, the equilibrium solution for the nonlinear di®erential
equations were determined by setting the derivative terms equal to zero and solving the resulting
set of algebraic equations for Y¹f ; Y¹o ; Y¹p and T¹ . These mean values, together with the mean values
of inlet and exit velocity, inlet species mass fractions, and temperature were assembled and used
for linearization of the conservation relations. The following section describes in detail the method
used for linearizing these equations about the equilibrium conditions.

4.1.3 Linearization and Conversion to State Space

Once the equilibrium solution is found, the system can be linearized about that equilibrium. The
assumption here is that for small perturbations about the equilibrium solution, the dynamics of the
system can be approximated by a linear system of equations. Because this research is concerned
with linear stability behavior, a linearized °ame model is ideally suited.

The nonlinear system of conservation equations can be written in assembled form as follows
where B is a nonlinear vector ¯eld. The system can be transformed into state variable form by
multiplying both sides by the matrix inverse of A. The resulting nonlinear vector ¯eld describes
the coupled, nonlinear dynamics of the state variables, fYo ; Yf ; T gT , subject to the time varying
parameters, uin ; uout ; and Yfin .

73
8 9
> >
>
>
>
>
Y_ o >
>
>
>
< =
A(uin ; uout ; Yfin ) > Y_ f > = B(fYo ; Yf ; T gT ; uin ; uout ; Yfin )
>
> >
>
>
> >
: T_ > ;
8 9
> >
>
>
>
>
Y_ o >
>
>
>
< =
Y_f = F (fYo ; Yf ; T gT ; uin ; uout ; Yfin ) (4.16)
>
> >
>
>
> >
>
>
: T_ >
;

The nonlinear vector ¯eld in Equation 4.16 can be linearized by assuming that each of the
three state variables can be expressed as a sum of a mean value and a perturbation as in Equation
4.17 below. These, along with the mean and perturbation values for the parameters (shown in
Equation 4.18) are then plugged into the nonlinear vector ¯eld equation and expanded in a Taylor
series, retaining only the linear terms as shown in Equation 4.19.

Yo (t) = Y¹o + Yo0(t)


Yf (t) = Y¹f + Yf0 (t) (4.17)
T (t) = T¹ + T 0(t)

uout (t) = u¹out + u0out (t)


uin (t) = u¹in + u0in (t) (4.18)
Yfin (t) = Y¹fin + Yf0in (t)

8 9 8 9 8 9
> > > > > >
>
>
>
>
Y_ 0o >
>
>
>
>
>
>
>
Yo0 >
>
>
>
>
>
>
>
u0in >
>
>
>
< = < = < =
> Y_ 0 f >
= F (x0 ; u0 ) + Dx F (x0; u0 ) > Yf0 > + Du F (x0 ; u0) > u0out >
(4.19)
>
> >
> >
> >
> >
> >
>
>
> >
> >
> > > >
: T_ 0 ; : T0 >; >
: Yf0 >
;
in

where x0 = fY¹o ; Y¹f ; T¹ gT

and uin ; u¹out ; Y¹fin gT


u0 = f¹

74
The ¯rst term in Equation 4.19 is the equilibrium solution which is equal to zero. The
notation D is a matrix operator denoting the Jacobian matrix with partial derivatives taken with
respect to the states (x) or parameters (u) depending on the subscript. The resulting linear equation
is a linear state model with three states, and three inputs. The output equation results from a similar
linearization of Equation 4.14 where G represents the scalar function description of the right hand
side.

8 9 8 9
>
> >
> >
> >
>
>
>
>
Yo0 >
>
>
>
>
>
u0in >
>
>
< = < =
¹ + Q0 = G(x0; u0 ) + Dx G(x0 ; u0) Y 0 + Du G(x0 ; u0) u0
Q (4.20)
>
> f >
> >
> out >
>
>
> >
> >
> >
>
> 0
: T ; > >
: Y0 >
;
fin

The constant term on the right side of Equation 4.20 represents the mean rate of heat
release and balances with the mean term on the left. Because the unsteady term is what forces
the acoustics, the mean term is dropped. This output equation shows that the linear system
features feedthrough dynamics as the output equation includes forcing terms. Though the notion of
this system exhibiting feedthrough dynamics initially de¯es intuition, a simple explanation exists.
Looking again at the expression for heat release rate shown in Equation 4.14, it is obvious that each
of the inputs appears explicitly. The physical interpretation is that unsteady velocities a®ect the
mass °ux which contributes directly to the change in sensible enthalpy. The unsteady mass fraction
of inlet fuel will instantaneously change the heat capacity of the inlet mixture. In other words, the
sensible enthalpy of the reactants changes as the mass fraction °uctuates. This °uctuation therefore
results in a direct °uctuation in the heat release rate. As it turns out, the feedthrough e®ects of the
unsteady mass fraction of inlet fuel are nearly insigni¯cant, yet it is an interesting thing to note.
The feedthrough terms on the unsteady inlet and exit velocities are much more signi¯cant to their
dynamics, and must therefore not be neglected.

As shown, the forcing consists of the unsteady inlet and exit velocity as well as the unsteady
inlet fuel mass fraction. Since the forcing for the overall feedback model results from the acoustic
perturbations in the combustor alone, the mass fraction variation must be related back to an acoustic

75
particle velocity to show the proper interaction of the acoustics with the °ame model. The next
section illustrates how acoustic particle velocity translates into variations in inlet fuel mass fraction
and also explains why the unsteadiness in the mass fraction of inlet air is neglected in this model.

4.1.4 Inlet Fuel Mass Fraction Variation

In the NETL combustion rig, air is swirled and fuel added in a mixing nozzle before entering an
expansion chamber where the °ame is stabilized by the swirling °ow ¯eld. Thus, the incoming °uid
contains both oxidizer and fuel in time varying proportions that are determined by the unsteadiness
in the nozzle °ow ¯eld. Since the combustor is run lean for all operating conditions, the mass fraction
of air is always very large compared to that of fuel. Over the range of equivalence ratios the mass
fraction of air in the incoming mixture is never less than 0:957. Though the incoming mixture
will be unsteady, the perturbation in the air mass fraction will be negligible compared to the mean,
while the unsteadiness in the fuel mass fraction will be comparable to the much smaller mean values.
Thus, the mass fraction of incoming oxidizer is assumed constant, while the mass fraction of the
incoming fuel will be unsteady. This concept is illustrated in the following derivation.

u0
m u + u0 ) = m
_ o = ½Au = ½A(¹ ¹_ o (1 + ) (4.21)

Since the fuel spoke is choked, the mass °ux of fuel is constant, and will be denoted m
_ f . The
mass ratio of oxidizer in the nozzle is determined as follows.

m _o
Yo =
m_ o+m _f
0
m¹_ o (1 + uu¹ )
= 0
¹_ f + m
m ¹_ o (1 + uu¹ )
0
1 + uu¹
= u0
F=A + 1 + u
¹

76
B A

Combustion Air

Pilot
Fuel

Swirler Ring Spacer Rings

Figure 4.1: FETC combustion rig mixing nozzle (not to scale)

1 0
1+F=A
(1 + uu¹ )
= 1 u0
(4.22)
1 + 1+F=A u¹

Retaining only ¯rst order terms, Equation 4.22 reduces as follows.

1 1 u0 1 u0
Yo = ¡ +
1 + F=A (1 + F=A)2 u¹ 1 + F=A u¹
1
' = Y¹o (4.23)
1 + F=A

The mass fraction of fuel can be determined by a continuity balance of the unsteady °ow ¯eld
in the nozzle and the steady stream of fuel entering that °ow ¯eld through fuel spokes which are
assumed choked for simplicity. A schematic of the nozzle can be seen in Figure 4.1. To determine
the mixture ratio of fuel entering the combustion chamber, the mass ratio of fuel from each fuel
spoke must be determined. Supposing there were a control volume drawn around a small slug of
°uid as it passed by a fuel spoke, the mass concentration of fuel in that control volume could be
derived using a continuity relation and linearized relative to the acoustic velocity perturbations as
follows.

77
m_f
Yf =
m_ f +m _o
m¹_ f
= u0
¹_ f + m
m ¹_ o (1 + u
¹
)
Y¹f
=
1 + Y¹o u
0
u
¹
0
u
' Y¹f (1 ¡ Y¹o ) (4.24)

A multiple fuel injector case would proceed in a similar manner, and the unsteady mass
fractions from each injector would simply sum to determine the overall variation in mass fraction.
The mass fraction which is derived at each fuel injection point is assumed to remain \frozen" as it
convects down the nozzle at the mean °ow velocity. When the mixture arrives at the reaction zone,
the mixture variation is based on acoustic response at some time before, giving rise to a time lag
term in the incoming fuel mass ratio. The multiple fuel injector analysis appears below for the ith
injector location.

¹_ fi
m
Y fi = u0
m ¹_ o + m_ 0 o + m
¹_ fi + (m ¹_ fi¡1 + m_ 0 fi¡1 + :::)(1 + u
¹
)
Y¹fi
=
1 + (Y¹oi + Y¹fi + Y¹fi¡1 + :::) u
0
u
¹
0
u
' Y¹fi (1 ¡ (Y¹oi + Y¹fi + Y¹fi¡1 + :::) ) (4.25)

m_
¹o
where Y¹oi =
m¹_ o + m
¹_ fi + m ¹_ fi¡1 + :::
m¹_ fi
Y¹fi =
m¹_ o + m
¹_ fi + m ¹_ fi¡1 + :::

A recent paper by Scarinci and Freeman [32] discusses the validity of the frozen fuel-air ratio
assumption and proposes a probabilistic modeling scheme for dealing with amplitude degradation
of the unsteady fuel-air ratio as it traverses a mixing nozzle. The model is based on principles of
atmospheric physics dealing with mixing of non-homogeneous species such as air pollution. The

78
net result is that the longer the unsteady mixture has to travel before reacting, the more mixed
it becomes and the less pronounced the oscillation in fuel-air ratio. In addition, the time delay
corresponding to the travel distance becomes less certain as that distance increases. No provisions
are made in this model for accounting for such degradation, but the e®ect is here noted as having
the potential to impact the ¯nal results.

In the ¯rst step leading to Equation 4.25, the unsteady mass terms will sum to approximately
zero due to mass conservation. They represent higher order terms for this equation and therefore
disappear when the equations are divided through by the sum of the mean mass °ows. Note that
the sum of Y 's multiplying the unsteady velocity term will approach unity as i increases. The
expressions in Equation 4.24 and the more general expression in 4.25 describe the in°uence of the
acoustic velocity on fuel mass ratio.

For the case of multiple fuel spokes, the mass ratio shown in Equation 4.25 represents the
mass ratio for the ith fuel spoke at ith location along the injector. Again, this unsteady fuel ratio is
assumed to remain \frozen" as it travels downstream to the °ame location. The total fuel ratio at
the °ame would then be equal to the sum of each individual mass ratios with the transport delay
included. Equation 4.26 shows how the total mass ratio of incoming fuel relates the acoustic particle
velocity in the nozzle and transport delay times from each fuel spoke location along the injector
nozzle to the °ame zone. Note that the product of the mean mass ratio of inlet fuel and the inverse
of the mean velocity in the nozzle attenuate the unsteady velocity to obtain an unsteady fuel ratio.
The time delay is determined by the transport time along the distance from the ith fuel spoke to
the °ame, denoted Li . Here, the length includes the total length along the injector nozzle as well
as an end correction to account for transport inside the combustion chamber before entering the
reaction zone. Determination of more exact values for Li will be discussed later.

N i 0 Li
X X u (t ¡ )
Yfin (t) = Y¹fi (1 ¡ (Y¹o + Y¹fj ) u
¹
) (4.26)
i=1 j=1 u¹

79
4.2 Modeling Results

In the previous section, frequency response characteristics were computed at each of the operating
conditions shown in Table 4.1 using the state space description in Equations 4.19 and 4.20. For
each operating condition, the frequency responses from each of the three inputs, Yf0in ; u0in , and u0out,
are combined to compute the overall unsteady heat release rate of the thermoacoustic system as
shown in Figure 1.2. Because the system has been linearized, this addition is a legitimate analysis
technique.

Looking over the frequency response from each of the three inputs in Figure 4.2, it initially
appears that the contributions from the inlet and exit velocity perturbations are negligibly small
compared to the °uctuating mass ratio of inlet fuel. The inlet fuel mass fraction °uctuation,
however, when converted back to units of acoustic velocity °uctuation in the mixing nozzle, is much
smaller and is then on a similar order to the inlet and exit velocity °uctuations as shown in Figure
4.3.

The inlet control surface, the mixing nozzle exit plane, was clearly not the same as the location
where reactants burned. However, the inlet variables of fuel mass fraction and total incoming
enthalpy would not be expected to change from the time they entered the control volume to the
actual combustion region. The unsteady velocities, on the other hand, would clearly be di®erent
between the analytical control surface and the actual reaction site. Similarly, the exit control surface
for the present model is located downstream of the actual °ame zone. Product species concentrations
and temperatures vary little from °ame to combustion chamber exit, but the unsteady velocity at
the exit surface would clearly not be similar to that of the surface of the °ame itself. De¯ning
the control surfaces away from the actual °ame surface thus provides a di±culty in determining
unsteady velocity °uctuations which may a®ect the dynamic behavior. The unsteady velocities are
modeled to directly impact the mass °ux into the reaction zone itself. Since the control surfaces
could not be located directly on the actual °ame surfaces, it would not be possible to determine
the unsteady velocities that e®ect bulk mass °ux terms directly.

80
The linearization showed the inlet and exit velocity perturbations to be on the same order
as the velocity perturbation that determined the equivalence ratio variations. However, the heat
release responses to the unsteady inlet and exit velocities, as seen in Figure 4.3 were nearly the
same amplitude and opposite phase. Since there were no modeled °ame area variations in response
to the unsteady velocities, the inlet and exit velocity contributions to unsteady heat release rate
would contribute mass addition and subtraction alone, not augmented reaction rate. Because of
the phase di®erence and amplitude similarity, they would approximately cancel out when added
together across a ¯xed area. It was therfore assumed that inlet and exit velocity perturbations, in
the absence of modeled °ame area variations, would not signi¯cantly impact the linear dynamic
behavior of the thermoacoustic system compared with the e®ects of equivalence ratio variations,
and were therefore neglected. The data bore out this conclusion.

The frequency responses of the well-stirred reactor system are proper. That is, they are
bounded as the frequency goes to in¯nity (Wolovich [37]). However, they are not strictly proper,
and therefore have as many zeros as poles. The gain of the °ame model was not signi¯cantly changed
over the frequency band of interest (0-400 Hz) for the range of operating conditions represented.
The intent of developing this model was not so much to capture the exact °ame behavior, as it could
easily be argued a simple control volume analysis would likely fall short. Rather, it was developed
to o®er a meaningful and reasonably accurate representation of the dynamics that a °ame may be
expected to contribute to the closed loop thermoacoustic system and to demonstrate the e®ects of
mean combustor conditions on the gain (and phase) characteristics.

4.2.1 E®ects of Mean Variables

The °ame response results presented in this section only appear for frequencies ranging from 0 to
400 Hz. The reason for this is that the closed loop analysis includes only acoustic modeling in
that same range of frequencies. It was anticipated that the °ame response model would roll o® at
frequencies in the range of 500 to 1000 Hz. In fact, the model showed that the transfer function
relating the unsteady inlet mass fraction to unsteady heat release rate began to roll o® at 10 Hz. In

81
150

130

110

90

70
0 1 2
10 10 10
freq (Hz) Q/Yfuel,in
270 Q/uin
Q/uout
180

90

-90
0 1 2
10 10 10
freq (Hz)

Figure 4.2: Input/Output relationships for each of the three inputs to the °ame model

90

85

80

75

70
0 1 2
10 10 10
freq (Hz) Q/unozzle
270 Q/uin
Q/uout
180

90

-90
0 1 2
10 10 10
freq (Hz)

Figure 4.3: Input/Output relationships with all inputs related back to acoustic veloc-
ities

82
similar models, the response of a propane system had much higher frequency response. However,
propane has a much higher pre-exponential factor for consumption rate than methane, meaning a
lower frequency response is expected.

When linearized about the mean temperature and species concentrations, the gain of the
system was somewhat lower than expected based on mean variable calculations. For example,
looking at Table 4.1, a DC gain can be estimated for the response of heat release rate to inlet mass
fraction °uctuations. At a mean velocity of 30 m=s and inlet mass fraction of 0.0333, the mean heat
release rate was 406 kW while at the same speed for an inlet mass fraction of 0.0355, the mean heat
release rate was 430 kW . Thus, for a perturbation of 0.0022 in inlet mass fraction, the resulting
perturbation in heat release rate was 24 kW . This corresponds to a gain of (24000/0.0022) or 141
dB. The linear model predicts only a 128 dB DC gain for the same operating conditions. The
nonlinear state equations can be integrated to show that the linearization is good for this range.
Thus, there appears to be a 13 dB bias error in the model response. This bias error was discovered
to exist over the range of operating conditions modeled. Similarly, a bias error of 16 dB was found
on the transfer function relating unsteady heat release rate to unsteady inlet velocity.

The only parameter that changes between the two successive runs mentioned above is the
equivalence ratio. There is a change in mean variables, but the linear model also underestimates
the degree by which they change as well. In the analysis, there are three species involved. This is
a common analysis technique in the literature, for both steady and unsteady well-stirred reactor
models. The composition of the product species was determined by examining the equilibrium
composition for each of the equivalence ratios modeled in this study. Thus, the only thing that
could prevent the linearized system from accurately capturing the gain is the di®erence in mole
fractions of the various species in the products. Even though the changes in equivalence ratio are
very small, the change in product species concentrations is signi¯cant to the behavior of the model.
This is because of the vast range of enthalpies of formation among the various product species.
Even slight changes in the mole fraction of carbon dioxide, for example can drastically a®ect the
sensible enthalpy of the products because the enthalpy of formation is so large.

83
The resolution to this inconsistency was to simply add the gain de¯cit into the analysis,
presuming the dynamics would not change signi¯cantly. Figures 4.2 and 4.3 re°ect this adjustment.
In order to capture these e®ects, the model would have to be expanded to include conservation
equations for each of the species in the reaction, thereby increasing the order of the dynamic
system. That way, the concentrations of all the product species would be allowed to vary with
the variable inlet mass fraction of fuel. The present three-species model is simply too restrictive
in regard to the product species concentrations to be able to accurately predict the gain from a
perturbation in inlet fuel mass fraction to heat release rate.

The e®ects of mean °ow and mean equivalence ratio on the unsteady inlet fuel mass fraction
to unsteady heat release rate frequency responses are shown in Figures 4.4 and 4.5. Mean velocity
has a greater e®ect than equivalence ratio. This may be due to the higher mean heat release rate
that resulted from higher °ow rates (see Table 4.1). A given percentage change in heat release rate
will be greater for a larger mean heat release rate value. Overall, other than the gain, the °ame
dynamics didn't change much over the range of conditions modeled.

Looking at the conservation equations, a parameter that is important to the dynamic be-
V
havior, that is equivalent to a time constant is the ratio Au
. Here, the volume, and control surface
area and velocity °ux terms combine to form a time constant which controls the bandwidth of the
system. Thus, if the control volume is made bigger, it has an increased capacity to store mass and
energy, and the bandwidth will decrease. Likewise, a decrease in the °ux velocity will decrease
the bandwidth by increasing the time constant. It was found for this model that the dominant
dynamics (two of the three poles) were associated with this time constant.

The chemical kinetics can in°uence the system response as well, in how they a®ect the time
constant ratio discussed in the previous paragraph. Supposing the Arrhenius pre-exponential factor
were greater, for example, the result would be faster kinetics. Using the procedure discussed in
Section 4.1.2, the WSR volume would have to be reduced to achieve the desired mean temperature
because the residence time required for completion of the kinetic reaction would be shorter. Since
the time constant is directly a®ected by the volume, the frequency response bandwidth will go up.

84
Similarly, increasing the equivalence ratio will a®ect the speed of the reaction, resulting in smaller
volumes and therefore higher bandwidth.

This chapter described how the acoustic particle velocity in the mixing nozzle causes the
inlet mass fraction to become unsteady. The unsteady rate of heat release response to unsteady
inlet mass fraction was then derived and used to represent the °ame dynamic contribution to the
system loop gain. The ¯nal dynamic component involved in the frequency domain stability analysis
is time delay of the unsteady inlet fuel ratio traveling down the nozzle to the combustion zone. The
following chapter presents the dynamics of multiple time delays and discusses the advantages of
multi-port fuel injection for enhancing stability characteristics of the thermoacoustic system.

85
Const Velocity, φ increasing
140

130

120

110
0 1 2
10 10 10
freq (Hz)

180

160

140

120

0 1 2
10 10 10
freq (Hz)

0
Figure 4.4: E®ect of equivalence ratio variation on Yqf °ame transfer function for a an
in
inlet mean velocity of 50 m/s. Arrows indicate increasing Á.

Const φ, Velocity increasing


140

130

120

110
0 1 2
10 10 10
freq (Hz)

180

160

140

120

0 1 2
10 10 10
freq (Hz)

0
Figure 4.5: E®ect of mean °ow variation on Yqf °ame transfer function for an equiva-
in
lence ratio of 0.67. Arrow indicates increasing mean °ow rate.

86
Chapter 5

Time Delay and Multi-Port Fuel


Injection

The experiments conducted at NETL and presented by Straub and Richards [34] were speci¯cally
designed to isolate and observe the e®ects of time delay processes to combustion stability in their
rig. They clearly demonstrated that varying con¯gurations, where the only change would have been
a time delay, resulted in changing stability behavior. However, a good explanation was lacking to
explain the root cause of the observations. Because of the crucial role that time delay played in
the manifestation of the type of thermoacoustic instability studied here, a thorough understanding
of the dynamics of time delays and how they interact with the plant was required. This chapter
discusses the e®ects of time delay in the context of closed loop stability. The e®ects of multiple
time delays are also discussed and presented, while developing a basis for a design procedure based
on carefully chosen time delay values, for stabilization of the self-excited thermoacoustic system.

87
5.1 Linear Stability Analysis

It is well known that time delay will destabilize a linear control system. The reason for this is that
the time delay causes a lag in the response of the controller to the motions of the plant that is being
controlled. The controller is unable to e®ect a response as the transients in the system are observed
because the control is subject to some kind of delay. A classical example of a time delay plant is a
chemical plant where there are transport delays caused by moving substances through piping.

Looking at the mathematical representation of a time delay in the frequency domain, e¡i!¿ ,
it is clear that the amplitude is constant, while phase delay increases with increasing frequency.
The tendency of higher values of time delay to cause a closed-loop plant to become unstable can be
presented in terms of frequency response analysis. If the gain of the open-loop system is above unity
as the phase crosses 180 degrees, the system is going to be unstable. As the time delay increases,
phase lag becomes greater. In this way, further increasing time delay will eventually destabilize a
controller or any other closed-loop system, such as a thermoacoustic system.

The stability of the linearized thermoacoustic system will be determined using the frequency
response methods. The Nyquist criterion is a frequency domain technique which uses a conformal
mapping of the Nyquist path trajectory from the complex plane with the open loop transfer function
as the mapping to indicate the number of closed loop roots that appear in the unstable, right side
of the complex plane. The Nyquist path is de¯ned as a trajectory which encircles the entire right
half plane. Thus, the Nyquist plot contains two symmetric loops, one corresponding to positive
frequencies and one to negative, from the Nyquist trajectory traversing the entire imaginary axis,
positive and negative, as it encircles the right half plane.

Since the roots of the closed-loop system characteristic equation, 1 + G(i!)H(i!), are being
examined, any encirclements of roots inside the Nyquist path will correspond to an encirclement of
the origin through the mapping 1 + G(i!)H(i!). The analysis is simpli¯ed so that the mapping
is simply the open-loop transfer function, G(i!)H(i!), and thus encirclements of the -1 point will

88
indicate unstable closed-loop poles. The procedure just discussed applies only to negative feedback
systems. For positive feedback, the closed-loop characteristic equation changes to 1 ¡ G(i!)H(i!),
which implies that encirclements of the +1 point using the open-loop transfer function as the
mapping would indicate unstable closed loop poles. It is therefore very important to know the sign
of the feedback mechanism to determine the stability characteristics of the closed-loop system.

The Nyquist criterion states that the number of unstable poles in the closed-loop system is
equal to the number of clockwise encirclements of the -1 point plus the number of open-loop poles
in the right half plane. In the present system model, the °ame dynamics and the acoustics are
stable, meaning the system contains no unstable open-loop poles. Any clockwise encirclements of
the -1 point would therefore indicate that the closed-loop system would be unstable. In terms of
the magnitude and phase of the open-loop plant, a gain of greater than one with a phase of 180
degrees translates exactly to an encirclement of the -1 point in the complex plane. A more thorough
explanation of the Nyquist stability criterion can be found in control theory texts such as Phillips
and Harbor [25], or Wolovich [37].

5.1.1 ¿ f Plotting

A common means of displaying combustion instability data is by the use of ¿ f plots. The value
of the relevant system time delay is multiplied by the observed frequency of instability to form a
dimensionless ordinate axis against which the amplitude of the limit cycle is plotted. Instability
data plotted in this manner will typically collapse into bands of ¿ f values based on the system and
its stability characteristics.

The Nyquist plot enables a graphical interpretation of these groupings relative to the open
loop frequency response. Nyquist plots depict the real and imaginary terms of a system parameter-
ized by frequency. Thus, the amplitude is equal to the distance from the origin and the phase as the
angle measured counterclockwise from the positive real axis. For a frequency, !o , that has a gain
greater than unity, the system would become unstable if the phase at that frequency were changed

89
by an angle, µ, between its current phase and -180 degrees, producing an encirclement of the -1
point. The time delay necessary to produce this encirclement is given by the following expression:

µ = !o ¿ = 2¼f¿ (5.1)

Rearranging Equation 5.1, the ¿ f product can be expressed as a normalized angle in the
complex plane.

µ
¿f = (5.2)

Bands of ¿ f data that cause instability can therefore be interpreted as corresponding to the
range of normalized phase angles which result in an encirclement of the -1 point. Using a describing
function linear analysis as described in Phillips and Harbor [25], it is possible to determine the
frequency of a limit cycle oscillation from the Nyquist plot as well. Supposing that the describing
function were plotted on the complex plane with the Nyquist plot, the frequency of the observed
limit cycle would be the frequency where the Nyquist plot and describing function plot intersect.
Assuming further, that the describing function would be real valued, this crossing would always
occur on the negative arm of the real axis. Thus, the frequencies used to generate the ¿ f plots
shown later always correspond to the 180 degree crossing point.

A one-mode example (shown in Figure 5.1) is used to illustrate the ¿ f graphical interpre-
tation. The range of frequencies for which the loop gain is greater than unity is 363 rad=s to 384
rad=s. The loop gain is high enough to cause closed-loop instabilities only within this range of fre-
quencies. Using the graphical interpretation described above, encirclements will occur for rotations
of the Nyquist plot ranging from 109 degrees to 251 degrees. The corresponding range of scatter in
¿ f data would therefore be equal to the normalized rotation angles, 0.303 to 0.697. The single time
delay required to cause instability can be computed based on the graphical interpretation described
above.

90
Bode Diagram

10

-10
Phase (deg); Magnitude (dB)

-20

-30

-40

100

50
To: Y(1)

-50

-100
10 2 10 3
Frequency (rad/sec)

Figure 5.1: Single mode example of resonant system.

2 4

1.5 384 rad/s 3

1 2 384 rad/s
0.5 1

Im Axis 71o Imag Axis


0 0

-0.5
-1

-1 -2

-1.5

A -2
B
-3

-4
-3 -2 -1 0 1 2 3
-2 -1 0 1 2 3
Real Axis
Real Axis
2 2.5

1.5 2
374 rad/s
1.5
1
1
0.5
o 0.5
Im Axis 18 Imag Axis 374 rad/s
0 0

-0.5 -0.5

-1
-1
-1.5
-1.5

C -2
-3 -2 -1 0 1 2 3
D
-2

-2.5
-3 -2 -1 0 1 2 3
Real Axis Real Axis

Figure 5.2: Graphical interpretation of ¿ f product.

91
Figure 5.2 is used to illustrate how the Nyquist plot can be used to determine stability bounds
for a time delay system. The Nyquist plot of the example system is shown along with the unit circle
for clari¯cation of relative gains. The minimum amount of phase required to drive the example
system unstable is determined by computing the minimum angle through which the Nyquist loop
must be rotated to cause an encirclement. Thus, the point where the Nyquist plot initially crosses
the unit circle is the ¯rst point with enough gain to cause an encirclement when rotated around the
origin by a time delay. In this case, a rotation of 109 degrees is required, which corresponds to a
normalized angle of 0.3028. The point chosen on the Nyquist plot in Figure 5.2A corresponds to a
frequency of 384 rad/s. The required time delay is computed as follows:

2¼(:303)
¿= = 0:005s (5.3)
384rad=s

When the time delay of 0.005 seconds is added to the system, the resulting Nyquist plot
(Figure 5.2B) shows that the system is on the verge of instability as the -1 point is almost being
encircled. In a similar manner, a time delay of 0.012 seconds results in incipient instability with
the Nyquist plot crossing the negative real axis at a frequency of 363 rad/s. Figures 5.2C and D
demonstrate this concept for an arbitrary time delay between 0.005 and 0.012 seconds. Here the
delay required to move the indicated point, which corresponds to a frequency of 374 rad/s, around
to the negative real axis is computed in the same manner as before. When plotted in Figure 5.2D,
the time delay of 0.0076 seconds, does, in fact cause the Nyquist plot to cross the negative real axis
resulting in an encirclement as predicted.

The reason it is interesting to view ¿ f in this manner is it equates the product with a
phase angle in the complex plane. As shown, for this simple open-loop system, instabilities will
be expected in the corresponding closed-loop system for time delays ranging between 0.005 and
0.012 seconds assuming negative feedback. This translated to a range of ¿f from 0.303 to 0.697.
Therefore, the ¿f plot for this system would be expected to contain data only in this band. Thus,
a frequency domain stability analysis can be used to determine the characteristic bands of ¿f for a
given system.

92
5.2 Multiple Time Delays

The key di®erence between single and multiple time delays is the e®ect that multiple delays can
have on amplitude. Recall that a single delay is unity magnitude with phase delay increasing with
frequency. In the case of multiple delays, the phase response is a®ected, but so is the amplitude
response as can be seen in the derivation for two equal amplitude delays shown in Equation 5.4.
Whereas in the case of a single delay, instability was determined strictly by the phase added by the
time lag, multiple delays can cause attenuation over the bandwidth in question, possibly causing a
stabilizing e®ect.

1 ¡i!¿1 1 ¡i!¿2 1 ¡i ! (¿1 ¡¿2 ) ¡i ! (¿1 +¿2 ) 1 ¡i ! (¿2 ¡¿1 ) ¡i ! (¿1 +¿2 )
e + e = e 2 e 2 + e 2 e 2
2 2 2 2
1 ³ ¡i ! (¿1 ¡¿2 ) !
´ !
= e 2 + e¡i 2 (¿2 ¡¿1 ) e¡i 2 (¿1 +¿2 )
2 µ ¶
! !
= cos (¿2 ¡ ¿1 ) e¡i 2 (¿1 +¿2 ) (5.4)
2

The expression for the sum of two equal amplitude time delays reduces to the product of a
frequency dependent amplitude variation and a single exponential with an argument that includes
the mean value of the two delays. Thus, when plotting results from two time delays in a ¿ f plot,
the mean value of the two time delays should be used for the abscissa. Similarly, a combination of
three or more equally spaced time delays will reduce to a variable amplitude multiplied by a single
exponential term with the overall mean time delay as the argument.

µ ¶
1 ¡i!¿1 1 ¡i!¿2 1 ¡i!¿3 2 1 ¡i ! (¿1 +¿2 +¿3 )
e + e + e = cos(!(¿1 ¡ ¿2 )) + e 3 (5.5)
3 3 3 3 3

µ ¶ µ ¶
1 ³ ¡i!¿1 ´ ! ! !
e + e¡i!¿2 + e¡i!¿3 + e¡i!¿4 = cos (¿1 ¡ ¿2 ) cos (¿1 ¡ ¿3 ) e¡i 2 (¿2 +¿3 ) (5.6)
4 2 2

93
0
10
|A|
-1
10

-2
10
0 50 100 150 200 250 300 350 400
freq (Hz)
200
2 delays
Phase (deg)

100 3 delays
4 delays
0

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 5.3: Frequency response functions of multiple time delays.

The frequency responses of the multiple delay cases are shown together in Figure 5.3. It is
interesting to note that, although the mean value of the time delay is the same for all three plots,
(0.008 s) the frequency responses look quite di®erent. As more delays are added, the amplitudes of
lobes in the response tend to go lower, and the ¯rst null point tends to move to higher frequency.
The attenuation patterns always repeat, but as more delays are added, the frequency range of
repetition increases. Note that the two delay case repeats after every null point. The three delay
case repeats after every two nulls, the four delay case repeats after every three nulls and so on. The
phase responses are relatively similar as they all have the same exponential term, only sign changes
occur at each null in the frequency response, causing a 180 degree phase shift.

This simple example illustrates the advantage that multiple time delays can o®er over single
delay systems. The amplitude e®ect can reduce the gain in the loop over a range of frequencies
allowing for enhanced stability according to the Nyquist criterion. If the multiple time delay am-
plitude response lowered the gain enough near instability frequencies, the system may not cause an
encirclement in the Nyquist plane, and would therefore remain stable. This simple concept o®ers a

94
partial explanation for the enhanced stability seen in the NETL data with multiple fuel injection
points installed in the mixing nozzle. The idea that multiple time delays could add a stabilizing
in°uence could be extended to a design procedure based on the amplitude response of a series of
time delays in the loop.

5.2.1 Delay Arrays

The formulation of multiple time delays as a sum of exponential functions is mathematically identical
to that of a linear antenna array. The array factor is de¯ned as the sum of antenna responses which
have relative phase shifts determined by the angle of an incoming plane wave relative to the line
formed by the antenna elements, and amplitudes (and phase shifts) from the transmission line
connected to each antenna element. The array factor, or ampli¯cation factor in a certain direction
at an angle a from the axis of the linear array as shown in Figure 5.4, is determined by the angle
of incidence of incoming plane waves and is computed as shown in Equation 5.7.

AF = A0 + A1 eikd cos(a) + A2 e2ikd cos(a) + A3 e3ikd cos(a) + A4e4ikd cos(a) (5.7)

As shown in Stutzman and Thiele [35], the array factor can be transformed to a new coordi-
nate system which is periodic over the interval 2¼.

AF = A0 + A1 eià + A2 e2ià + A3 e3ià + A4 e4ià (5.8)

Now, the à coordinate represents the angular variation of the array factor. Looking at
Equations 5.4 - 5.6, the form is nearly identical to that shown in Equation 5.8. Thus, the product
of the time delay and angular frequency in the above equations is equal to à and the formulation
proceeds similar to that of an equally spaced linear array. As such, the sum of time delays will be

95
λ = 2π k
)
(a
os
dc

a
0 1 2 3 4
d d
Figure 5.4: Equally spaced linear array.

called a \delay array," denoting its similarity with the linear array formulation.

The key di®erence between the linear array formulation and the delay array formulation is
the spacing of the array elements relative to the origin. In the case of the linear array, the origin is
always chosen as the ¯rst array element. In the delay array, this can be done as well, but it must
be understood that the ¯rst time delay value would thus correspond to the origin.

For example, suppose the ¯rst delay in Equation 5.4 was chosen as the origin. The second
delay, would then be chosen relative to the ¯rst. In the case shown in Figure 5.3, the two delay
values were 0.005 s and 0.011 s. Thus, the second delay is 0.006 s greater than the ¯rst, meaning
the time delay spacing, ¿s , is 0.006 s. This means that the product of !¿s = Ã implies that the
frequency response would repeat over an interval of 2¼=0:006s or 1047 rad/s (167 Hz). Looking at
Figure 5.3, the two delay case is clearly periodic over the interval of 167 Hz.

In array theory, the phase response is not of great concern for a single antenna array, so it
is payed little attention. The main focus is on beam patterns in space. The phasing of the array
elements can be changed electronically to allow a beam pattern to be steered around in space. This
concept is known as a phased array. The spatial coordinate is the angle, a measured from the
axis of the linear array and the beam pattern determines the directivity of the array based on the

96
coordinate transformation from à to a. For the time delay system, the analysis in à coordinates
proceeds exactly the same way, but the coordinate transformation is slightly di®erent and clearly
much simpler than that of the antenna arrays. Rather than a beam pattern in space, the time
delay analysis is concerned with a \beam pattern" in frequency obtained through the coordinate
transformation !¿s = Ã. By analogy, the amplitude response of the delay array is dependent only
on the spacing of the time delays in the array, not on their speci¯c values. Only the phase response
will depend on the actual value of the delays, speci¯cally the mean value. This e®ect can be seen
in Equations 5.4 - 5.6 as well. The amplitude terms in these equations are clearly dependent only
on the spacing of the delays.

The analogies of array to the frequency response can be used for designing attenuation bands
in the frequency response. In array theory, the array elements are typically arranged close together
so that the beam pattern will feature a single main lobe and several smaller, side lobes. The design
objective is to reduce the levels of the side lobes relative to the main lobe. By analogy, the main
lobe in the frequency response of the time delay system is always centered on a frequency of zero.
This is because the transformation from à coordinates doesn't include any phasing terms as the
case of a steered array would in array theory. The amplitude response of the delay array cannot be
\steered" with time delays, so the DC gain is always equal to unity.

From Stutzman and Thiele [35] the following trends will be observed in à coordinates as the
number of array elements, N , of a uniformly excited array increases.

² As N increases, the main lobe narrows.

² The number of lobes including one main lobe and side lobes equals N ¡ 1 over the interval of
2¼.

² The minor lobes are width 2¼=N and main lobe is twice this width.

² The side lobe peaks decrease in amplitude with increasing N . As N increases, the side lobes
approach a level of -13.3 dB relative to the main lobe.

97
² The array factor function is symmetric about ¼.

These same trends can be observed in the amplitude responses of Figure 5.3. The main lobe
is always centered at 0 Hz, and as the number of time delays increased, the main lobe narrowed.
Recall that the lobe narrows in à coordinates. For the three cases shown, in order to keep the same
mean value for time delay and same minimum and maximum value of time delay, the spacing of the
delays was decreased. This meant that the frequency interval corresponding to 2¼ in à coordinates
increased with the addition of time delays. Thus, the main lobe appears to be getting wider. In
order to see a reduction in the width of the main lobe in frequency coordinates, the time delay
spacing would have to be kept constant, making the delay array much longer with the addition of
array elements and increasing the overall mean time delay value, adding considerable phase to the
system.

In accordance with the trends mentioned above, side lobe levels were observed to decrease
with additional delays. For the case of a uniform array of 5 delays, the reduction would be -12
dB, quite close to the -13.3 dB theoretical maximum. Adding more delays beyond ¯ve will not
help signi¯cantly unless the amplitudes of the array elements are made nonuniform. \Shading" the
amplitudes of the array elements allows for a greater side lobe reduction which, again is a typical
goal for array design. Using a design procedure known as Dolph-Chebychev array synthesis, all of
the side lobe levels can be reduced to a uniform level chosen by the designer. The design tradeo®
is that, as the side lobes are made smaller, the main lobe will become wider. A simple, ¯ve delay
example shows the di®erences between amplitude shading for 2 Dolph-Chebychev designs and the
uniform distribution. Here, the delays are kept the same in all cases, only the amplitudes of the
delay elements are changed. Figure 5.5 illustrates that as the side lobes are reduced, the main
lobe becomes wider. It also shows the shape of the amplitude distributions used for each case. To
increase the reduction in side lobe levels, the amplitudes are tapered with the center element having
the greatest amplitude, and outer elements having the smallest.

98
0

|A| (dB) -10

-20

-30

-40
0 50 100 150 200 250 300 350 400
freq (Hz)
Array element amplitudes equal amplitudes
-20 dB SLL
0.3 -30 dB SLL

0.2
|A|

0.1

0
0.002 0.004 0.006 0.008 0.010 0.012 0.014
τ (s)

Figure 5.5: Frequency response amplitude of 5 time delays showing e®ect of amplitude
shading using constant side lobe level Dolph-Chebychev array design.

5.2.2 Multi-Port Injector Design

The preceding discussion o®ers an opportunity for developing a design procedure for attenuation
of thermoacoustic instabilities using a fuel injector delay array. By adding multiple fuel injection
ports in the mixing nozzle, the resulting time delays will behave as a delay array which can be used
to reduce gain in the loop. By shading the amplitude of the injection points, the rig can be designed
to have signi¯cant amplitude reduction over some bandwidth determined by the time delay spacing
of the injectors. Amplitude shading could be realized by adjusting the fuel split to the various
injectors. The injectors located at the center of the array should account for a greater percentage
of the total fuel than the ones at the edges in order to get the side lobe reductions discussed in the
previous section.

In creating a design procedure for multi-port fuel injection it soon becomes apparent that
the design goals, though similar to those of linear array designs, may not require strict adherence to

99
the linear array design procedures. For example, in linear array synthesis, it is typically desirable to
have a beam pattern with only a single main lobe; none of the so-called grating lobes. A grating lobe
is a high amplitude lobe other than the main lobe. Thus, additional array elements tend to reduce
the main lobe width and add only low amplitude side lobes. In the thermoacoustic system, it may
not be practical to continue to add more and more equally spaced fuel injector ports due to space
limitations. By adding more injectors at smaller intervals, the main lobe still gets wider, though
attenuation will extend out to higher frequencies. It is possible that this broadband attenuation
may be a waste, in that the system may not require attenuation over such a vast frequency range.
What may be needed for the thermoacoustic system design is larger reductions over small bands
of frequencies. Such localized reductions could be realized if the spacing of the injectors were not
¯xed. There is no practical reason that fuel injectors could not be installed on irregular intervals
with a variety of fuel split amplitudes.

For designing nonuniform, unequally spaced arrays, a simple optimization routine was used
with a cost function that could include penalties at the frequencies of high acoustic gain. The
FMINS command in Matlab is a multivariable function minimization routine which uses a Nelder-
Meade type simplex search method. Here, the objective function is de¯ned as the largest singular
value of delay array gain de¯ned over a prescribed frequency interval. The optimization allows the
amplitudes and/or time delays of the delay array to adapt, with the bounding high and low time
delays ¯xed, in this case at 0.005 s and 0.011 s to compare with data shown previously.

In every case, the amplitude response results are compared with those of the -30 dB Dolph-
Chebychev design. In Figure 5.6, the optimizations are shown for objective functions evaluated in
the narrow band of frequencies from 150 to 250 Hz. Note that the main lobe is narrower for the
optimized arrays than the Dolph-Chebychev solution but the band of attenuation much smaller.
Supposing that the gain in the thermoacoustic system is large in this narrow band, good attenuation
can be achieved only where it is needed, illustrating the advantage of using optimization for multi-
port fuel injector design.

It is also interesting to note the similarities in the amplitude response for the two cases

100
0

|A| (dB) -10

-20

-30

-40
0 50 100 150 200 250 300 350 400
freq (Hz) amplitude optimized
Array element amplitudes
delay optimized
both optimized
0.3 -30 dB SLL

0.2
|A|

0.1

0
2 4 6 8 10 12 14
τ x 10
-3

Figure 5.6: Frequency response amplitude of 5 time delays showing optimization re-
sults.

of amplitude only optimization and amplitude and delay (both) optimization. The contributions
of gain and delay are distributed di®erently in each case, yet the overall amplitude response is
approximately equivalent in the frequency band shown. These results illustrate how concentrating
the gain toward the center of the array tends to cause the low side lobe level.

A second example illustrates how the open loop acoustic gain of the combustor can be used in
the cost function to penalize bands of high acoustic gain. Here, the objective function is de¯ned as
before, but the delay array gain is multiplied by the acoustic gain of the plant. Also, the objective
function is evaluated over the band of frequencies from 0 to 400 Hz in this example. The results
shown in Figure 5.7 illustrate optimization results with and without the acoustic gain penalty
present. In each case, the amplitudes and middle three time delays are allowed to adapt. The
acoustic gain plot is shown superposed to illustrate where the gain is highest. Figure 5.8 shows the
acoustic gain with the time delay results of the two optimizations multiplied by the acoustics. This
plot shows how the optimizations a®ect the overall loop gain. Note that the optimization which
includes the gain penalty results in a °atter overall gain, bringing some peaks down more than

101
others.

In summary, a design method has been proposed which uses the natural gain e®ects of an
array of time delays in the system to attenuate certain bands of frequencies, thereby enhancing
the stability characteristics of the system. This design method could prove useful in the quest
to quell thermoacoustic instability as it employs the same types of hardware already installed in
the combustion system, only arranged in a scheme that takes advantage of the natural dynamics
of the system itself. This design procedure didn't include any discussion of phase, but that too
could be adjusted by moving the entire array up or downstream, changing the mean time delay
value if necessary. The reader is referred back to the discussion in Section 5.1.1 describing how the
Nyquist plot can be used to determine the amount of time lag to add to further enhance stability.
The following chapter will discuss assembly of all system components including time delay for the
purpose of thermoacoustic stability predictions.

102
0

-10
|A| (dB)

-20

-30

-40
0 50 100 150 200 250 300 350 400
freq (Hz)
Array element amplitudes

0.3

0.2
|A|

0.1 no acoustic gain penalty


with acoustic gain penalty

0
2 4 6 8 10 12 14
τ -3
x 10

Figure 5.7: Frequency response amplitude of 5 time delays showing optimization results
with objective function augmented by acoustic gain penalty.

-20

-40
|A| (dB)

-60

-80
acoustic gain
-100 composite gain (no gain penalty)
composite gain (gain penalty)
-120
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 5.8: Composite amplitude response showing the product of the acoustic gain
and time delay array gains for two optimizations.

103
Chapter 6

Stability Predictions and Conclusions

The stability of the system is determined by the assembly of all the components and whether
a perturbation will tend to be ampli¯ed or attenuated. Using frequency response methods, the
determination of stability can be made by examining the amplitude of the loop gain at the point
where the open-loop phase is equal to 180 degrees (for a negative feedback plant). If the gain
is greater than unity at a frequency where the phase is 180, an ampli¯cation will occur at that
frequency in the closed-loop system. The acoustics will, in general, provide the frequency dependent
gain necessary to allow instabilities to occur. The °ame dynamics serve to low-pass ¯lter the plant,
disallowing high frequency instabilities. Time delay serves to add phase to the system to change
the frequencies where instabilities will be observed.

For this system, it was indeed possible to show that this linear analysis can be validated
with actual test data from NETL. The various injector port locations result in di®erent amounts
of phase from time delays, so the instability frequencies observed in the data change with di®erent
fuel injection locations. Mean °ow velocity, like the injector location, has a great in°uence on the
stability behavior due to its e®ect on transport delays in the system. Changes in equivalence ratio,
however did not cause signi¯cant changes in the acoustic response of the testbed, nor did they cause
great changes in the °ame dynamic behavior shown in prior chapters. Thus, the mean equivalence

104
Q’ γ −1 U’panel Acoustic U’nozzle Y fuel,in Y’fuel,in Transport Y’fuel,in(t-τ) Flame Dynamic Q’
ρ c 2 2 π rL Transfer Function − Delay Transfer Function
(-) U nozzle

Figure 6.1: Block diagram of open loop transfer function.

ratio wouldn't be expected to signi¯cantly a®ect the stability behavior of the system.

6.1 Closed-Loop Stability Analysis

In this section, the procedure for assembling the individual components of the system into a closed-
loop system will be described. Simply stated, the frequency responses of each block were multiplied
together to obtain the appropriate gains and unit conversions. Once all the components were
multiplied together into an open-loop transfer function, the magnitude and phase could be used to
determine stability. The resulting open-loop transfer function has an input and output with the
same physical units. When the loop is closed, therefore, the output can feed back into the input
forming a closed-loop transfer function. Figure 6.1 shows once again how all of these components
¯t into the open-loop transfer function. In order to be unstable, the gain of the open-loop transfer
function (loop gain) would have to exceed unity for at least some frequencies as a gain less than
that would be stable regardless of phase.

The following sections will illustrate a step-by-step process for assembling the various com-
ponents into an open-loop transfer function, using the operating condition of injector B (located
14.4 cm upstream of the combustor inlet) and 50 m=s mean °ow.

105
0

unozzle/upanel (dB)
-20

-40

-60

-80
0 50 100 150 200 250 300 350 400
freq (Hz)
180

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.2: Nozzle particle velocity response to pulsating cylinder velocity input.

6.1.1 Acoustics

The acoustics are based on the FEM mesh with forcing provided by a small cylindrical source
pulsating at a velocity of 1 m=s. Thus, the particle velocity response of the plant is relative to
a unity input. The tailpipe boundary condition was equal to the characteristic impedance of the
exhaust °uid ½c. The exhaust temperature was reported by NETL to be approximately 500 K
downstream of the spray quench at the exit of the combustor plug. The acoustics were assumed to
remain constant for all operating conditions, since the temperature distribution in the combustor
itself wasn't a large factor, and the exhaust temperature was not known for each condition. The axial
velocity perturbation in the mixing nozzle is what determines the unsteady inlet fuel mass fraction
as described in Section 4.1.4. Thus, the acoustic particle velocity in the nozzle is what couples the
acoustics to the °ame dynamics. The frequency response shown in Figure 6.2 corresponds to the
mixing nozzle particle velocity response to a unit input surface velocity of the pulsating cylinder
source.

106
20

0
(dB) -20
nozzle

-40
/u
f,in

-60
Y

-80
0 50 100 150 200 250 300 350 400
freq (Hz)
-Y/u * time delay
180 time delay alone

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.3: Time delay and velocity to inlet fuel mass fraction gain for injector location
B, mean velocity 50 m/s (¿ = 5.1 ms).

6.1.2 Time Delay and Inlet Fuel Mass Fraction Scaling

The acoustic velocity response is multiplied by the time delay dynamics to include the e®ect of the
transport delay on the open-loop phase response. The gain of the acoustic velocity response must
also be adjusted to scale the output properly for input to the °ame dynamics. This is achieved by
fin Y¹
multiplying the particle velocity response by the ratio ¡ u¹nozzle which was derived in Section 4.1.4
as the proper scaling for acoustic particle velocity in the nozzle to inlet fuel mass fraction variation.
The time delay response for the example case, along with the scaling gain mentioned, is shown in
Figure 6.3. The time delay response alone is also shown as a dashed line. Note that, as is the case
for a single delay, the gain is constant and the phase decreases linearly with frequency.

107
140

(dB) 120
f,in
/Y
panel

100
u

80
0 50 100 150 200 250 300 350 400
freq (Hz) scaled flame dynamics
flame dynamics alone
180
Phase (deg)

90

-90
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.4: Flame dynamics and acoustic input scaling transfer function.

6.1.3 Flame Dynamics and Acoustic Input Scaling

The next component is the °ame dynamic response. Driven by unsteady inlet fuel mass fraction
caused by acoustic velocity perturbations in the mixing nozzle, the °ame dynamic model output is
the unsteady heat release rate which, in turn drives the acoustic plant. However, since the acoustic
plant is forced by a pulsating cylinder, the heat release rate must be properly converted into units of
panel velocity. This scaling was demonstrated in Section 2.1.1.1. The values used to compute this
scaling factor are: the length (0.06 m) and radius (0.0005 m) of the pulsating cylinder, the density
kg m
of the °uid at the °ame source (1.31 m3
), the average acoustic speed of the acoustic domain (488 s
)
and the ratio of speci¯c heats of air (1.40). The reason the mean acoustic speed is used as opposed
to the acoustic speed of the °ame is that it came from dividing the acoustic wave equation by c2.
Since the acoustic wave equation describes the acoustics throughout the combustor, the acoustic
speed used here must represent some type of averaged value.

It is also important to note that the sign of the transfer function shown in Figure 6.4 is
inverted. The loop gain therefore represents a negative feedback plant. If positive feedback was the

108
10
0
(dB) -10
-20
-30
-40
-50
0 50 100 150 200 250 300 350 400
freq (Hz)
180

90
Phase (deg)

-90

-180
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.5: Open-loop transfer function for thermoacoustic plant.

assumption for analysis, this inversion would not be included, and 0 degree phase crossings in the
frequency response would indicate instabilities, as opposed to 180 degree phase crossings. The sign
convention chosen for the feedback plant is simply a matter of preference; the stability analysis will
give the same results whether positive or negative feedback is assumed as long as the loop gain and
analysis is appropriate to the assumption.

6.1.4 Composite Loop Gain

Finally, all the components shown can be multiplied together to render the properly scaled composite
loop gain. This is the frequency response that is used to determine the stability characteristics of
the closed-loop thermoacoustic plant. Since the loop gain was developed for negative feedback,
instabilities will be indicated by 180 degree phase crossings with corresponding gains greater than 0
dB. In Figure 6.5, the phase response crosses 180 degrees at 36, 60, 72, 234, and 378 Hz. Of these,
the only one with enough gain to cause an instability is the one at 36 Hz. The gain is not quite
equal to 0 dB at the 234 Hz crossing, but clearly it is high relative to the other nearby gains. It isn't

109
di±cult to imagine that the loop gain shown here could easily be underestimating the true system
loop gain due to uncertainties in the various components. What this plot shows, however, is that
the system will go unstable and the instability will be expected to occur at a frequency of 36 Hz.
In fact, the NETL combustor was observed to oscillate in a 234 Hz limit cycle for this operating
condition. The loop gain plot clearly shows a propensity toward instability near that frequency as
well.

That fact that the gain appeared to be so close to 0 dB in general is quite satisfying based on
the stability behavior of the NETL plant. Sometimes the combustor is stable, and sometimes it is
not, depending on the operating condition. This implies that the loop gain is high enough to cause
instability at some frequencies, but not high enough to cause instability at any frequency where the
open-loop phase response crosses 180 degrees. Therefore, having the loop gain appear near 0 dB,
and not way above it, is suggested by the stability behavior observed in the real combustor.

6.2 Stability Prediction

The previous section demonstrated how the components of the thermoacoustic loop gain are assem-
bled for stability analysis. The result proved to have multiple potential instabilities. This will be
even more pronounced in a case with greater time delay and more 180 degree phase crossings. The
question then arises as to how the stability prediction should proceed. What will be the frequency
that will display self-excited growth to a limit cycle in the case of multiple instabilities? Looking
over the data, there is very rarely a case where multiple limit cycles are observed coexisting. For
purposes of this analysis, the instability frequency with the highest gain shall be chosen as the
preferred instability of the system. This appeals to some intuition about the system. Supposing the
gain were raised gradually, the highest gain instability would become unstable ¯rst. However, this
criterion should be thought of more as a guideline than a strict physical truth. Further research
into the mechanisms governing this issue may be warrented.

110
1

0.9

0.8

STABLE
0.7
Normalized Amplitude

0.6

0.5

STABLE
0.4

0.3

0.2

0.1

0
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5
τ*f

Figure 6.6: ¿f plot of experimental (circles) and theoretical (stars) instabilities.

The NETL data show the pressure limit cycle frequencies along with their amplitudes as
a percentage of the mean combustor pressure. The limit cycles range from 120 to 260 Hz. The
equivalence ratio, in most cases, had little impact on the frequencies observed, while mean °ow
velocity and fuel injector position had a great in°uence. Appendix A shows power spectral densities
of the pressures measured in the NETL rig during the stability tests. In each plot, three equivalence
ratios are represented, and clearly, the power spectral densities are similar accross equivalence ratios
in these plots. Thus, the transport delay, which is dictated by the fuel injector position and mean
°ow velocity, was the dominant control variable for the observed instabilities. The data are therefore
presented in a typical format of a ¿ f plot shown in Figure 6.6. This plot shows the amplitude of
the limit cycle plotted against the product of the time delay and observed instability frequency.
The data tend to collapse in bands which indicate the stable and unstable regimes of the system.
As discussed in Section 5.1.1, these bands can also be thought of as the normalized angles that will
cause 180 degree phase crossings in areas of high loop gain.

The only problem with presenting data in this manner, in general, is that it presumes that
the values for the time delays are well known. In fact, this is not necessarily the case. The time
delay values used for the NETL data were based on a transport distance of the length of the

111
mixing nozzle from the fuel spoke to the combustor plus one inch, and on the mean °ow velocity.
The one inch correction was an attempt to account for the transport inside the combustor before
reacting. However, looking at the steady-state CFD runs in Appendix B, the reactants travel a
great deal further than one inch before reacting. The blue plume of cool reactants extends far into
the combustion can in every case. The reactants can't combust until they preheat and decelerate
to allow °ame propagation. This occurs in a distributed sense, but it seems very clear qualitatively,
that the one inch assumption is probably too conservative.

The proposed solution to this dilemma was to determine the time delay that would match
the observed instability using the analytical °ame and acoustic results. This seemed as good a
method as any, and it used the data to guide some of the analysis. The question that remained
was, once a value was picked for one operating condition, would it be consistent across the range,
and would it make physical sense? Using the method discussed, values of time delay were chosen
that changed the end correction from 1 inch (or 2.5 cm) to about 11.5 - 13 cm. Using these longer
delay times, the analytical results corresponded quite well with the available data. Also, looking
again at the steady state CFD models, it appears that the reactant plume is about 0.1 m in length,
further supporting the correction length used. The end correction used here is quite di®erent from
that used by Straub and Richards [34], illustrating the need for further investigation of methods for
choosing appropriate values of time delay.

A further testament to the validity of this correction comes from the stability analysis itself.
Figure 6.7 shows the positive frequency half of a Nyquist plot of the loop gain for a system with
zero time delay in it. Notice the positions of the higher frequency Nyquist loops relative to the
horizontal. Given a su±cient amount of gain, in order to cause an encirclement of the -1 point,
these loops could be rotated anywhere from a little under zero to around 145 degrees in a clockwise
direction. This range of angles means a normalized range of about 0 to 0.4 which would repeat at
whole numbers, meaning ranges of 0 - 0.4, 1 - 1.4, 2 - 2.4, etc. Given the variations in open-loop
phase over the range of operating conditions, the limits of these bands may be expected to vary a
little, but this provides a good estimate of where the ¿ f data should be expected to fall. Again,
because the time delays in this research were based on a di®erent end correction than Straub and

112
0.5

-0.5

122 Hz
-1
236 Hz

34 Hz
-1.5

-2

-2.5
-1 -0.5 0 0.5 1 1.5 2

Figure 6.7: Nyquist diagram of zero-delay loop gain.

Richards, the stability bands predicted here will be distinct from those presented in their 1998
paper.

6.2.1 Predicted and Observed Frequencies

For the stability analysis that follows, since the loop gain was typically a little low, 8 dB of gain was
added to each case to increase the range of frequencies with a gain greater than unity. Since the true
gain of the system may have been underestimated, this extra gain was needed to demonstrate the
stability prediction results. The loop gain showed a very strong peak at about 30 Hz, and the range
of frequencies up to about 100 Hz had relatively high gain in general. However, no frequency lower
than 125 Hz was ever observed in the instability data. Thus, even though a low frequency phase
crossing may have been the highest gain of multiple crossings, it was never chosen as the predicted
frequency. These two ¯xes were not based on any modeled phenomena, rather an observation that
the predictions could be very close to the data provided these adjustments were made.

113
Occurances of Instability Frequencies Observed in NETL Data
10
9
8
7
6
5
4
3
2
1
0
115

125

135

145

155

165

175

185

195

205

215

225

235

245

255

265
freq (Hz)

Figure 6.8: Histogram showing occurances of instability frequencies in NETL data.

Table 6.1: Time delay values used for analysis (values shown in ms).

Injector u¹ = 30m=s u¹ = 40m=s u¹ = 50m=s u¹ = 60m=s


A 6.6 4.9 4.0 3.3
B 8.9 6.7 5.2 4.4
C 10.7 8.1 6.4 5.3

By adjusting the time delays according to the methods described, reasonable agreement with
the experimental data was achieved. The resulting end correction appeared to make physical sense
given the steady state CFD data as well. Based on the modeled dynamics in the loop, stability
bands were established for the ¿f data. Most of the data and predictions appear to fall in these
bands. Thus, the time delays were adjusted based on a number of criteria, and the ¯nal choices can
be justi¯ed by several means.

The histogram shown in Figure 6.8 shows the number of occurances of instability freuquencies
over the range of tests performed. The graph shows that the instabilities clustered into a couple
ranges of frequencies. The range centered on 230 Hz corresponds with one of the acoustic resonances

114
Occurances of Instability Frequencies Predicted by Analysis
16
14
12
10
8
6
4
2
0
100
115
130
145
160
175
190
205
220
235
250
265
280
295
310
325
340
355
freq (Hz)

Figure 6.9: Histogram showing occurances of instability frequency predictions.

in the model. However, the cluster of data around 165 Hz doesn't correspond to a predicted acoustic
resonance. These instability data exist because of time delays in the system. Instabilities need not
occur at acoustic resonances only. There was a predicted acoustic resonance at 130 Hz, but very few
tests showed instabilities at that frequency. Figure 6.9 shows a similar histogram for the stability
predictions. In this case, there are many more predicted instabilities near the acoustic resonance at
130 Hz, and fewer near 160 Hz. The predictions didn't include many instances near 160 Hz because
the loop gain was very low in that frequency band. Had the gain been higher, however, there would
have been many more predicted instabilities near 160 Hz because 180 degree phase crossings were
often seen in the open-loop transfer functions near that frequency.

In general, the stability model appeared to exhibit similar behavior to the observed data.
Tables 6.2 and 6.3 show the results of the NETL tests and the stability predictions of the loop gain
frequency response analysis. Comparing the tabular data, the A and B data are quite similar. The
C data were harder to match, but it is clear that the types of trends seen in the data can also be seen
in the stability predictions resulting from this analysis. Figure 6.6 shows a ¿ f plot of the NETL
data (circles) and stability prediction data (stars). The stability bands motivated by the Nyquist

115
Table 6.2: Stability Frequency Data Taken at NETL.

Injector u¹(m=s) Á = 0:59 Á = 0:63 Á = 0:67 Á = 0:71 Á = 0:77


A 30 161 161 164 173 174
40 215 220 223 229 236
50 220 229 243 STABLE STABLE
60 264 STABLE STABLE NO DATA NO DATA
B 30 126 126 129 229 223
40 161 164 167 170 STABLE
50 214 229 234 STABLE STABLE
60 223 234 240 NO DATA NO DATA
C 30 176 179 179 179 179
40 231 234 237 237 237
50 161 164 164 STABLE STABLE
60 STABLE STABLE STABLE NO DATA NO DATA
A&B 30 155 155 STABLE STABLE STABLE
40 176 STABLE 223 229 234
50 229 STABLE STABLE STABLE STABLE
60 STABLE STABLE STABLE STABLE STABLE
A&C 30 158 161 STABLE STABLE 226
40 223 223 231 229 234
50 237 243 STABLE STABLE STABLE
60 STABLE STABLE STABLE STABLE STABLE

plot of Figure 6.7 are shown as dashed lines. Most of the data seem to fall within predicted unstable
bands, supporting the validity of the time delay values chosen for the various injector locations.

Table 6.4 shows the di®erence between observed and predicted instability frequencies and
also where stable operation was both observed and predicted. Blank ¯elds indicate disagreement
between prediction and observation of stable conditions. In other words, if the combustor ran stable,
but an instability was predicted at a given operating condition, the ¯eld is left blank. Likewise,
if stability is predicted, but not observed, the ¯eld is left blank. Clearly, the predictions are not
perfect, but there is good agreement for injector locations A and B in particular.

116
Table 6.3: Stability Frequency Data Predicted by Stability Model.

Injector u¹(m=s) Á = 0:59 Á = 0:63 Á = 0:67 Á = 0:71 Á = 0:77


A 30 STABLE 145 146 146 147
40 235 235 235 235 235
50 243 243 243 243 244
60 101 101 102 103 104
B 30 123 123 123 123 123
40 STABLE 144 144 145 146
50 233 233 233 233 234
60 239 238 239 239 239
C 30 115 115 116 116 116
40 127 127 127 127 128
50 STABLE STABLE 350 350 351
60 232 232 232 233 233
A&B 30 130 129 130 130 130
40 STABLE STABLE STABLE STABLE 224
50 237 237 237 237 238
60 STABLE 246 246 247 247
A&C 30 STABLE STABLE 211 212 213
40 239 239 239 239 239
50 STABLE STABLE STABLE STABLE STABLE
60 STABLE STABLE 359 359 359

6.2.1.1 Spectrum Comparison

Power spectral densities (PSD's) were provided for the instability data on the NETL rig. They
are summarized in Appendix A. These plots proved very useful for determining the ¯delity of the
linear stability model in several ways. First, they demonstrate that equivalence ratio has a very
small e®ect on the behavior. This was true of the model as well. More importantly, however, the
PSD's show more information than can be derived by looking at tabulated instability data. For
example, the acoustic gain can be observed in the power spectral density plots. They indicated
that the acoustic resonances of the operating combustor appear to be well captured by the acoustic
FEM modeling. Acoustic gain bands can sometimes be seen in the PSD's at about 230 Hz and at
350 Hz. These correspond quite well with the acoustic frequency response peaks predicted by the
FEM model.

117
Table 6.4: Di®erences Between Data and Stability Prediction Frequencies.

Injector u¹(m=s) Á = 0:59 Á = 0:63 Á = 0:67 Á = 0:71 Á = 0:77


A 30 16 18 27 27
40 -20 -15 -12 -6 1
50 -23 -14 0
60 163
B 30 3 3 6 106 100
40 20 23 25
50 -19 -4 1
60 -16 -4 1
C 30 61 64 63 63 63
40 104 107 110 110 109
50 -186 STABLE¤ STABLE¤
60
A&B 30 15 16
40 STABLE 10
50 -8
60 STABLE STABLE¤ STABLE¤ STABLE¤ STABLE¤
A&C 30 STABLE¤ STABLE¤ 13
40 -16 -16 -8 -10 -5
50 STABLE STABLE STABLE
60 STABLE STABLE STABLE¤ STABLE¤ STABLE¤

* - stability was observed in data and analysis showed very low loop gain instability.

A sample case was taken from the analysis to demonstrate how the stability predictions
related to the actual stability data. Here, the PSD's for all 4 mean velocities of the injector B data
are shown with the analytical loop gain response. Note that the PSD's for 3 of the 5 equivalence
ratios are shown together, demonstrating that the equivalence ratio variation has little e®ect on
the observed instability behavior. Again, stability is predicted based on the 180 degree phase
crossings and the corresponding gain for the open-loop system. The loop gains shown here appear
as calculated, which means the gain appears to be too low for some of the observed instabilities.

Figures 6.10 - 6.13 show how the 180 degree crossings line up relative to the instability
frequencies observed in the data. Clearly the magnitude of the loop gain is too low to have these
exact instabilities (see Figure 6.11 in particular), but the proper trends exist for these time delays
to render the observed stability behavior provided the loop gain were slightly higher. Figures 6.14 -

118
6.17 show the same specrta with the predicted closed-loop transfer functions (input, Q0 to output,
0
Unozzle ). Note that the closed-loop peaks are sharpened where the open-loop transfer function had
a phase crossing. The closed-loop peak was observed to move by about 5 Hz between the 50 m/s
and 60 m/s cases. This was as expected based on the open-loop phase crossings shown. Again, the
time delay end correction used for these runs was 12 cm at the mean nozzle velocities indicated.

119
Mean Velocity = 30 m/s Power Spectral Density
0
10

φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Loop Gain
20

0
|A| (dB)

-20

-40
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.10: Comparison of PSD data and loop gain stability prediction for injector
B, 30 m/s.

120
Mean Velocity = 40 m/s Power Spectral Density
0
10
φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Loop Gain
20

0
|A| (dB)

-20

-40
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.11: Comparison of PSD data and loop gain stability prediction for injector
B, 40 m/s.

121
Mean Velocity = 50 m/s Power Spectral Density
0
10
φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Loop Gain
20

0
|A| (dB)

-20

-40
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.12: Comparison of PSD data and loop gain stability prediction for injector
B, 50 m/s.

122
Mean Velocity = 60 m/s Power Spectral Density
0
10

φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Loop Gain
20

0
|A| (dB)

-20

-40
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.13: Comparison of PSD data and loop gain stability prediction for injector
B, 60 m/s.

123
Mean Velocity = 30 m/s Power Spectral Density
0
10

φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Closed loop transfer function


0

-20
|A| (dB)

-40

-60

-80
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.14: Comparison of PSD data and closed-loop tranfer function for injector B,
30 m/s.

124
Mean Velocity = 40 m/s Power Spectral Density
0
10
φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Closed loop transfer function


0

-20
|A| (dB)

-40

-60

-80
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.15: Comparison of PSD data and closed-loop transfer function for injector B,
40 m/s.

125
Mean Velocity = 50 m/s Power Spectral Density
0
10
φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Closed loop transfer function


0

-20
|A| (dB)

-40

-60

-80
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.16: Comparison of PSD data and closed-loop transfer function for injector B,
50 m/s.

126
Mean Velocity = 60 m/s Power Spectral Density
0
10

φ = .59
φ = .63
φ = .67
-2
10

-4
10

-6
10

-8
10
0 50 100 150 200 250 300 350 400
freq (Hz)

Closed loop transfer function


0

-20
|A| (dB)

-40

-60

-80
0 50 100 150 200 250 300 350 400

200

100
Phase (deg)

-100

-200
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.17: Comparison of PSD data and closed-loop transfer function for injector B,
60 m/s.

127
6.3 Summary

This research has shown that a linear stability analysis can be used to characterize the stability
behavior of a thermoacoustic plant. By properly scaling each of the components, a true loop
gain response was generated, though it seems the predicted gain was still slightly lower than that
needed for many of the observed instabilities to be predicted. The exact loop gain is dependent on
several factors including damping and scaling parameters with unceratain values. The predicted
gain was raised arti¯cially, but having the loop gain prediction come so close to unity was indeed
an encouraging result.

The model was used to capture the gain and phase characteristics of the open-loop transfer
function and frequency response analysis was used to determine stability of the corresponding closed-
loop system. The stability predictions of this linear analysis exhibited similar behavior to the test
data for the same plant. The frequencies of instability were observed to jump across resonances
depending on the time delay and a broad range of instability frequencies were observed despite
no change in the actual acoustic resonance frequencies because of phase e®ects from the di®erent
time delays at di®erent operating conditions. Instabilities are caused at the frequencies where the
open-loop transfer function phase crosses 180 degrees, which can change when di®erent time delays
are introduced. This clearly demonstrates that observed instabilities need not occur at the acoustic
resonances themselves.

The dynamics of the °ame show a low-pass frequency response characteristic. Thus, acoustic
resonances at 350 Hz and higher are attenuated in the loop gain response which could explain why
they are not observed in the experimental stability data. It was shown that a three species dynamic
model such as the one used for this research will likely result in underestimating the gain. This
problem was not due to the linearization of the vector ¯eld, rather the simpli¯cation of the product
species concentration to be ¯xed by the mean equivalence ratio value. Even small perturbations in
the mixture strength will result in di®erent product species concentrations, thereby changing the
total enthalpy of the products. A model which fails to account for the changing concentrations

128
of all the product species will not accurately predict the amplitude of the unsteady heat release
rate. The underestimate was consistent across the operating range, so additional gain was simply
added to the predicted gain, and the dynamics were assumed to be similar to those that would be
predicted by a higher order model.

6.4 Conclusions

Time delay is a critical parameter in the stability of combustion systems such as the one in this
study. Because it was di±cult to estimate the time delay values accurately, experimental data
were used to help ¯t the stability predictions by changing the time delay values. What resulted
was a nozzle length end-correction of approximately 12 cm based on the delay values chosen. It
is important to note that the exact value of time delay cannot be known accurately for a variety
of reasons, including the fact that the oscillation in mixture strength probably will not stay frozen
through the combustor. Mixing tends to smear the oscillation spatially, resulting in a smaller gain
as well as a less certain value for the time delay itself. Detailed study of this phenomenon was
outside the scope of this particular e®ort.

The loop gain showed a high amplitude resonance at approximately 35 Hz. Though the
stability analysis showed 180 degree phase crossings near this resonance and other frequencies
nearby, instabilities are never observed at low frequency. In fact, looking at power spectral densities
of the unstable system, there doesn't even appear to be much energy below 100 Hz for the unstable
system. This discrepancy is hypothesised to be related to unmodeled heat transfer e®ects in the
combustor itself. The combustor walls are water cooled which would provide a near constant
temperature boundary. In this case, heat transfer to the walls could account for a large amount of
heat loss at frequencies below 50 Hz. Since the e®ects of unsteady heat transfer are not modeled,
there is an excessive amount of heat being kept in the °uid that would have otherwise been transfered
out of the control volume at lower frequencies. If heat transfer e®ects were included, the °ame model
would likely show reduced gain in the bandwidth below 50 Hz or so, lowering the loop gain in that

129
region, possibly enough to disallow low frequency instabilities as was observed in the data.

The linear dynamic model for the NETL thermoacoustic testbed combustor proved capable of
replicating many characteristics of the stability behavior observed in the testbed itself by predicting
the true loop gain for the thermoacoustic plant. Using test data along with the dynamic model, time
delays were determined that would account for an e®ective inlet mixing length and predict stability
maps similar to those seen in the lab. The dynamic °ame model used here was not su±cient for
determining the exact gain of heat release rate, nor for capturing relevant heat transfer e®ects in
the system. However, the error in gain was easily documented and corrected and thus the model
did provide a means for scaling the feedback relationship between the acoustic particle velocity and
the unsteady heat release forcing of the acoustics themselves.

6.5 Future Work

Future e®orts should include more comprehensive modeling of the °ame dynamic behavior with
higher order models including multiple species and heat transfer e®ects. These models could then
be linearized as in this study and used for stability predictions with a loop gain frequency response
analysis. Higher order models could also be used to generate data to which a lower order model
could be ¯t. The resulting low order model would then contain all the relevant gain e®ects that the
low order model in this study did not.

A more comprehensive °ame dynamic model should also include heat transfer e®ects as they
would likely a®ect the loop gain in the low frequency band, enabling accurate prediction without
the need for including arti¯cial ¯xes as was done here. Understanding the reaction of the °ame to
mass °ux variations with the inclusion of a °ame area variation model could further improve the
integrity of the °ame model.

Another area that needs further investigation is the time delay itself. Analytical and experi-

130
mental evidence was presented for the validity of the time delay values chosen, but a more rigorous
and detailed investigation of this critical parameter is needed. Experiments could be designed to
attempt to measure time delays in the system under controlled conditions. The use of CFD, again,
would be invaluable for exploring this subject by creating models that could indicate the tempo-
ral response and spatial distribution of heat release relative to imposed perturbations in mixture
strength.

Details of how the unsteady mixture travels and reacts to the mean °ow ¯eld is not well
understood. It is clear that this concentration distribution cannot remain frozen throughout the
journey from injector to reaction zone, but accurate models of the relevant phenomena should be
explored. This information would be invaluable when considering the prospect of using a multi-port
injector array for the purposes of stabilizing a thermoacoustic plant. As was shown, the longer the
array, the lower the bandwidth of attenuation. If low frequency (around 125 Hz) oscillations are
targeted for suppression, the injector array must be quite long spatially. Thus, understanding how
the unsteady mixture dissipates in space is important to knowing whether such a control scheme is
feasible.

Based on the loop gain predicted by the linear model, injector array ¯lter designs proposed
in Section 5.2.2 would be su±cient for lowering the loop gain over a bandwidth to promote stable
operation. The only limitation would be the bandwidth of the ¯lter determined by the main, 0 Hz
lobe. Many of the array designs showed that the main lobe extended out to 150 Hz or so. The
bandwidth of the main lobe is limited only by the physical spacing limitations of the fuel injectors
themselves, along with the mean °ow velocity, both of which would determine the time delay spacing
of the fuel injector spokes.

It would be interesting to investigate the practical e®ectiveness of an injector array control


scheme with some experimental test runs. Based on the multi-port injector designs in Section 5.2.2,
the spacing of the injectors would have to be between 4.5 cm and 9.0 cm, depending on the mean
°ow, to get the desired frequency response. Figure 6.18 shows how the width of the main beam will
increase with mean °ow velocity for an injector spacing of 6 cm. As the time delay between injectors

131
0
Vmean = 30 m/s
-2 Vmean = 60 m/s

-4

-6

-8
|A| (dB)

-10

-12

-14

-16

-18

-20
0 50 100 150 200 250 300 350 400
freq (Hz)

Figure 6.18: Injector array amplitude responses for a 5 injectors spaced 6 cm apart.

decreases, the main lobe will move to higher frequencies, so spacing the injectors out is desirable.
However, there may be practical limitations to the physical size of the array due to mixing e®ects
in the nozzle smearing out the unsteady mixture. These are the types of issues that can best be
addressed with laboratory experiments.

Should a multi-port fuel injector design enhance the stability characteristics of the system as
predicted in this analytical study, a control scheme would exist that requires very little hardware,
and perhaps very little knowledge of the combustion dynamics themselves. Should similar systems
behave as this one does, having a loop gain which is close to unity over the bandwidth, it is
conceivable that a simple injector array design tailored for suppression of frequencies commonly
observed in unstable operation of a combustor could be su±ciently e®ective. One way to test this
hypothesis would be to change the dynamics of the combustor and attempt to design the injector
array around the observed instabilities. Successful demonstration of such a generalized, multi-port
injector control scheme could be an excellent eventuality for this research e®ort.

132
Bibliography

[1] A. M. Annaswamy, M. Flei¯l, J. P. Hathout, and A. F. Ghoniem. An input-output model of


thermoacoustic instability and active control design. Report 9705, Dept. of Mech Engineering,
MIT, November 1997. 38 pages.

[2] G. J. Bloxsidge, A. P. Dowling, and P. J. Langhorne. Reheat buzz: an acoustically coupled


combustion instability. part 2. theory. Journal of Fluid Mechanics, 193:445{473, 1988.

[3] S¶ebastien M. Candel. Combustion instabilities coupled by pressure waves and their active
control. In Proc. of the Twenty-Forth Symposium (International) on Combustion, pages 1277{
1296. The Combustion Institute, 1992.

[4] Ray W. Clough and Joseph Penzien. Dynamics of Structures. McGraw-Hill, Inc., New York,
NY, 1975.

[5] A. Craggs. Acoustic modeling: ¯nite element method. Chapter 14 of Encyclopedia of Acoustics,
Malcom J. Crocker, editor. John Wiley & Sons, Inc., New York, NY, 1997.

[6] F. E. C. Culick. Short communication: A note on Rayleigh's criterion. Combustion Science


and Technology, 56:159{166, 1987.

[7] A. P. Dowling. Thermoacoustic instability. In Sixth International Congress on Sound and


Vibration, Copenhagen, Denmark, July 1999.

[8] M. Flei¯l, A. M. Annaswamy, Z. A. Ghoniem, and A. F. Ghoniem. Response of a laminar


premixed °ame to °ow oscillations: A kinematic model and thermoacoustic instability results.
Combustion and Flame, 106:487{510, 1996.

133
[9] Ephraim J. Gutmark and Christian Oliver Paschereit. Sources and control of thermoacoustic
instabilities in gas turbines. In Sixth International Congress on Sound and Vibration, Copen-
hagen, Denmark, July 1999.

[10] Jakob J. Keller. Thermoacoustic oscillations in combustion chambers of gas turbines. AIAA
Journal, 33(12):2280{2287, 1995.

[11] Lawrence E. Kinsler, Austin R. Frey, Alan B. Coppens, and James V. Sanders. Fundamentals
of Acoustics. John Wiley and Sons, 3 edition, 1982.

[12] Uwe KrÄ


uger, Jens HÄ
uren, Sefan Ho®mann, Werner Krebs, and Dieter Bohn. Prediction of
thermoacoustic instabilities with focus on the dynamic °ame behavior for the 3A-Series gas
turbine of Siemens KWU. In Proc. of the International Gas Turbine & Aeroengine Congress
& Exhibition, Indianapolis, IN, June 1999. 99-GT-111.

[13] T. Lieuwen, Y. Neumeier, and B. T. Zinn. The role of unmixedness and chemical kinetics
in driving combustion instabilities in lean premixed combustors. Combustion Science and
Technology, 135:193{211, 1998.

[14] Tim Lieuwen, Hector Torres, Cli®ord Johnson, and Ben T. Zinn. A mechanism of combustion
instability in lean premixed gas turbine combustors. In Proc. of the International Gas Turbine
& Aeroengine Congress & Exhibition, Indianapolis, IN, June 1999.

[15] Tim Lieuwen and Ben T. Zinn. The role of equivalence ratio oscillations in driving combustion
instabilities in low NO x gas turbines. In Proc. of the 1998 27th International Symposium on
Combustion, pages 1809{1816, Boulder, CO, 1998. The Combustion Institute.

[16] M. A. Macquisten and A. P. Dowling. Low-frequency combustion oscillations in a model


afterburner. Combustion and Flame, 94:253{264, 1993.

[17] G. H. Markstein. Nonsteady Flame Propagation. The Macmillann Company, Pergamon Press,
Oxford, England, 1964.

[18] George H. Markstein. Instability phenomena in combustion waves. In Proc. of the 1952 4th In-
ternational Symposium on Combustion, pages 44{59, Cambridge, Mass, 1952. The Combustion
Institute.

134
[19] A. C. McIntosh. On the cellular instability of °ames near porous-plug burners. Journal of
Fluid Mechanics, 161:43{75, 1985.

[20] Leonard Meirovitch. Elements of Vibration Analysis. McGraw-Hill, Inc., New York, NY, 1986.

[21] M. L. Munjal. Acoustics of Ducts and Mu²ers with application to exhaust and ventilation
system design. John Wiley & Sons, New York, NY, 1987.

[22] R. M. Murray, C. A. Jacobson, R. Casas, A. I. Khibnik, C. R. Johnson Jr., R. Bitmead, A. A.


Peracchio, and W. M. Proscia. System identi¯cation for limit cycling systems: A case study for
combustion instabilities. In Proceedings of the American Control Conference, pages 2004{2008,
Philadelphia, PA, June 1998.

[23] Ali Hasan Nayfeh. Introduction to Perturbation Techniques. John Wiley and Sons, 1993.

[24] A. A. Peracchio and W. M. Proscia. Nonlinear heat-release/acoustic model for thermoacoustic


instability in lean premixed combustors. In 1998 ASME Gas Turbine and Aerospace Congress,
1998.

[25] Charles L. Phillips and Royce D. Harbor. Feedback Control Systems. Prentice-Hall, Inc., Upper
Saddle River, NJ, third edition, 1996.

[26] Allan D. Pierce. Acoustics: An Introduction to its Physical Principles and Applications.
Acoustical Society of America, 1991.

[27] R. L. Raun, M. W. Beckstead, J. C. Finlinson, and K. P. Brooks. A review of Rijke tubes,


Rijke burners and related devices. Progress in Energy and Combustion Science, 19:313{364,
1993.

[28] Lord Rayleigh. The explanation of certain acoustical phenomena. Royal Institution Proceedings,
8:536{542, 1878.

[29] Geo A. Richards and Michael C. Janus. Conrol of °ame oscillations with equivalence ratio
modulation. Journal of Propulsion and Power, 15(2):232{240, 1999.

135
[30] George A. Richards, Randall S. Gemmen, and M. Joseph Yip. A test device fo premixed gas
turbine combustion oscillations. Technical note DOE/METC-96/1027, US DOE Morgantown
Energy Technology Center, Morgantown, WV, March 1996.

[31] William R. Saunders, Lars Nord, Christopher A. Fannin, Ximing Huang, William T. Baumann,
Uri Vandsburger, Vivek Khanna, Ludwig Haber, Bryan Eisenhower, and Scott Liljenberg.
Diagnostics and modeling of acoustic signatures in a tube combustor. In Sixth International
Congress on Sound and Vibration, Copenhagen, Denmark, July 1999.

[32] Thomas Scarinci and Christopher Freeman. The propagation of a fuel-air ratio disturbance
in a simple premixer and its in°uence on pressure wave ampli¯cation. In Proceedings of the
ASME Turbo Expo 2000, Munich, Germany, May 2000. ASME 2000-GT-0106.

[33] G. Searby and D. Rochwerger. A parametric acoustic instability in premixed °ames. Journal
of Fluid Mechanics, 231:529{543, 1991.

[34] Douglas L. Straub and Geo A. Richards. E®ect of fuel nozzle con¯guration on premix combus-
tion dynamics. In Proceedings of the 1998 ASME Turbo Expo. ASME 98-GT-492, 1998.

[35] Warren L. Stutzman and Gary A. Thiele. Antenna Theory and Design. John Wiley & Sons,
Inc., New York, NY, 1981.

[36] Stephen R. Turns. An Intoduction to Combustion: Concepts and Applications. McGraw-Hill,


1996.

[37] William A. Wolovich. Automatic Control Systems, Basic Analysis and Design. Saunders
College Publishing, Fort Worth, TX, 1994.

136
Appendix A

NETL Testing Power Spectra

For each of the stability tests performed at NETL, the pressure was measured in the combustor
section. The time data were averaged on a spectrum analyzer and are shown here for each of the
¯ve injector con¯gurations tested. With few exceptions, the equivalence ratio showed very little
in°uence in the stability behavior and so the data are sorted by mean velocity and injector position.
Three of the ¯ve equivalence ratios tested appear in each plot. Stable conditions are di®erentiated
from unstable by the tendency for a limit cycle to cause a very sharp peak in the power spectral
density and evidence of harmonic frequencies associated with the limit cycle oscillation. Several of
the operating conditions show no such peak. These are assumed to be stable operating conditions.

137
0
Location A: Mean Velocity = 30 m/s Power Spectral Density Location A: Mean Velocity = 40 m/s Power Spectral Density
0
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2
-2
10 10

-4
10 10
-4

-6
10 10
-6

-8
-8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)
Location A: Mean Velocity = 50 m/s Power Spectral Density
0 Location A: Mean Velocity = 60 m/s Power Spectral Density
10 0
10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2
-2
10 10

-4
-4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)

Figure A.1: Power spectral densities for injector location A (8.3 cm upstream of com-
bustor inlet).

0
Location B: Mean Velocity = 30 m/s Power Spectral Density Location B: Mean Velocity = 40 m/s Power Spectral Density
0
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)
Location B: Mean Velocity = 50 m/s Power Spectral Density 0
Location B: Mean Velocity = 60 m/s Power Spectral Density
0
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)

Figure A.2: Power spectral densities for injector location B (14.4 cm upstream of
combustor inlet).

138
Location C: Mean Velocity = 30 m/s Power Spectral Density Location C: Mean Velocity = 40 m/s Power Spectral Density
0 0
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)
Location C: Mean Velocity = 50 m/s Power Spectral Density 0
Location C: Mean Velocity = 60 m/s Power Spectral Density
0 10
10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67
φ = .67
-2
-2
10 10

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)

Figure A.3: Power spectral densities for injector location C (19.4 cm upstream of
combustor inlet).

0
Location AB: Mean Velocity = 30 m/s Power Spectral Density Location AB: Mean Velocity = 40 m/s Power Spectral Density
0
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)
Location AB: Mean Velocity = 50 m/s Power Spectral Density 0
Location AB: Mean Velocity = 60 m/s Power Spectral Density
0
10 10

φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8
-8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)

Figure A.4: Power spectral densities for injector locations A and B.

139
0
Location AC: Mean Velocity = 30 m/s Power Spectral Density 0
Location AC: Mean Velocity = 40 m/s Power Spectral Density
10 10
φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2 -2
10 10

-4 -4
10 10

-6 -6
10 10

-8
-8 10
10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)
Location AC: Mean Velocity = 50 m/s Power Spectral Density 0
Location AC: Mean Velocity = 60 m/s Power Spectral Density
0
10 10

φ = .59 φ = .59
φ = .63 φ = .63
φ = .67 φ = .67
-2
-2
10 10

-4
-4
10 10

-6 -6
10 10

-8 -8
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
freq (Hz) freq (Hz)

Figure A.5: Power spectral densities for injector locations A and C.

140
Appendix B

NETL CFD Model Results

This appendix summarizes the CFD results generated by Mr. Bill Rogers at NETL using Fluent.
They are mean °ow solutions and show the distributions of temperature and velocity for ¯ve di®erent
test conditions in the combustion chamber of the NETL combustion rig. The data are used to
determine values for inlet and exit velocities and temperatures. The results of these computations
are summarized along with the test condition descriptions in Table B.1. On the pages that follow,
there are ¯gures showing the distribution of temperature and axial velocity in the combustor. The
plots are axisymmetric, so the bottom of each plot represents the centerline of the combustor. On
the left is the injection nozzle, and on the right is the plug entrance.

Table B.1: Combustor inlet and exit results from CFD runs

Label Á Vnominal ( ms ) Vinlet ( ms ) Tinlet (K) Vexit( ms ) Texit (K)


a 0.6 50 44.9 608 35.3 1843
b 0.8 50 45.1 605 39.8 2128
c 0.55 50 43.8 606 34.4 1754
d 0.6 30 28.6 613 24.6 1858
e 0.6 60 58.0 604 44.3 1860

141
2200
0.09
2000
0.08
1800
0.07
Radial location (m)

1600
0.06
1400
0.05
1200
0.04

0.03 1000

0.02 800

0.01 600

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.1: Test condition \a" temperature distribution (K)

60
0.09

0.08 50

0.07 40
Radial location (m)

0.06
30
0.05
20
0.04
10
0.03

0.02 0

0.01 -10

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.2: Test condition \a" axial velocity distribution (m/s)

142
2400
0.09 2200
0.08 2000

0.07 1800
Radial location (m)

0.06 1600

0.05 1400

0.04 1200

0.03 1000

0.02 800

0.01 600

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.3: Test condition \b" temperature distribution (K)

60
0.09

0.08 50

0.07 40
Radial location (m)

0.06
30

0.05
20
0.04
10
0.03

0.02 0

0.01 -10

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.4: Test condition \b" axial velocity distribution (m/s)

143
0.09 1800

0.08
1600
0.07
Radial location (m)

1400
0.06

0.05 1200

0.04
1000
0.03
800
0.02

0.01 600

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.5: Test condition \c" temperature distribution (K)

60
0.09
50
0.08
40
0.07
30
Radial location (m)

0.06
20
0.05
10
0.04
0
0.03

-10
0.02

-20
0.01

-30
0.2 0.25 0.3 0.35
Axial location (m)

Figure B.6: Test condition \c" axial velocity distribution (m/s)

144
2200
0.09
2000
0.08
1800
0.07
Radial location (m)

1600
0.06
1400
0.05
1200
0.04

1000
0.03

0.02 800

0.01 600

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.7: Test condition \d" temperature distribution (K)

40
0.09

0.08 30

0.07
Radial location (m)

20
0.06

0.05
10
0.04

0.03 0

0.02
-10
0.01

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.8: Test condition \d" axial velocity distribution (m/s)

145
2200
0.09
2000
0.08
1800
0.07
Radial location (m)

1600
0.06
1400
0.05
1200
0.04

0.03 1000

0.02 800

0.01 600

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.9: Test condition \e" temperature distribution (K)

80
0.09
70
0.08
60
0.07
50
Radial location (m)

0.06
40
0.05
30
0.04
20

0.03
10

0.02 0

0.01 -10

0.2 0.25 0.3 0.35


Axial location (m)

Figure B.10: Test condition \e" axial velocity distribution (m/s)

146
Vita

Christopher Fannin was born in Washington, D.C. in 1972 and grew up in Fairfax County, Virginia.
He graduated from Thomas Je®erson High School for Science and Technology in June, 1990. In
the fall of 1990, he started his undergraduate work at the College of Engineering at Virginia Tech.
During his undergraduate program, he worked as a Co-op engineer for Framatome Technologies
in Lynchburg, Virginia. He worked in stress analysis, and tooling design and participated in an
intern exchange program, working one summer in Chalons-sur-Saone, France. He completed his
bachelor of science, graduating Magna Cum Laude in May, 1995, and went on to get a master's
degree which he completed in February, 1997. He immediately started his Ph.D. work in the area of
active combustion control with Dr. Will Saunders. While a graduate student, he has been a GTA
and a course instructor for a senior level undergraduate class. Upon completion of his degree, he
will begin working for the Architechtual Hardware division of Ingersoll-Rand in Indianapolis, IN.
He is engaged to be married with Dr. Susan Holland next spring.

147

Potrebbero piacerti anche